From: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>

The original calculation in commit 3cc70889a3 ("esp.c: prevent cmdfifo overflow
in esp_cdb_ready()") subtracted cmdfifo_cdb_offset from fifo8_num_used() to
calculate the outstanding cmdfifo length, but this is incorrect because
fifo8_num_used() can also include wraparound data.

Instead calculate the maximum offset used by scsi_cdb_length() which is just
the first byte after cmdfifo_cdb_offset, and then peek the entire content
of the cmdfifo. The fifo8_peek_bufptr() result will then return the maximum
length of remaining data up to the end of the internal cmdfifo array, which
can then be used for the overflow check.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Fixes: 3cc70889a3 ("esp.c: prevent cmdfifo overflow in esp_cdb_ready()")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3082
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Link: https://lore.kernel.org/r/20250925122846.527615-2-mark.cave-ayland@ilande.co.uk
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/esp.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/hw/scsi/esp.c b/hw/scsi/esp.c
index 1d264c40e57..2809fcdee09 100644
--- a/hw/scsi/esp.c
+++ b/hw/scsi/esp.c
@@ -447,7 +447,9 @@ static void write_response(ESPState *s)
 
 static bool esp_cdb_ready(ESPState *s)
 {
-    int len = fifo8_num_used(&s->cmdfifo) - s->cmdfifo_cdb_offset;
+    /* scsi_cdb_length() only reads the first byte */
+    int limit = s->cmdfifo_cdb_offset + 1;
+    int len = fifo8_num_used(&s->cmdfifo);
     const uint8_t *pbuf;
     uint32_t n;
     int cdblen;
@@ -457,7 +459,7 @@ static bool esp_cdb_ready(ESPState *s)
     }
 
     pbuf = fifo8_peek_bufptr(&s->cmdfifo, len, &n);
-    if (n < len) {
+    if (n < limit) {
         /*
          * In normal use the cmdfifo should never wrap, but include this check
          * to prevent a malicious guest from reading past the end of the
-- 
2.51.1

28.10.2025 20:34, Paolo Bonzini wrote:
> From: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> 
> The original calculation in commit 3cc70889a3 ("esp.c: prevent cmdfifo overflow
> in esp_cdb_ready()") subtracted cmdfifo_cdb_offset from fifo8_num_used() to
> calculate the outstanding cmdfifo length, but this is incorrect because
> fifo8_num_used() can also include wraparound data.
> 
> Instead calculate the maximum offset used by scsi_cdb_length() which is just
> the first byte after cmdfifo_cdb_offset, and then peek the entire content
> of the cmdfifo. The fifo8_peek_bufptr() result will then return the maximum
> length of remaining data up to the end of the internal cmdfifo array, which
> can then be used for the overflow check.
> 
> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
> Fixes: 3cc70889a3 ("esp.c: prevent cmdfifo overflow in esp_cdb_ready()")
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3082
> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> Link: https://lore.kernel.org/r/20250925122846.527615-2-mark.cave-ayland@ilande.co.uk
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

This smells like a qemu-stable material (this and the subsequent patch
adding the test).  Please let me know if it is not.

Thanks,

/mjt