[PATCH] sandbox: usb: Fix out-of-bounds read when fd=-1

[Sean Anderson]

sandbox_flash_bulk uses priv->read_len to determine if priv->buff contains
the response data (such as from SCSI_INQUIRY). However, if priv->fd=-1 in
handle_read, then priv->read_len is not set even though we are going to
PHASE_DATA. This causes sandbox_flash_bulk to try and read len bytes from
priv->buff, which likely goes past the end of the buffer. Fix this by always
setting priv->read_len even if we aren't going to read anything.

Fixes: f4f715360c ("dm: usb: sandbox: Add an emulator for USB flash devices")
Signed-off-by: Sean Anderson <seanga2 at gmail.com>
---
Is returning -EIO correct here? Should we return 0 (nothing read)? Or pretend to
read the whole thing and then let the caller figure it out based on the status?

 drivers/usb/emul/sandbox_flash.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/emul/sandbox_flash.c b/drivers/usb/emul/sandbox_flash.c
index edabc1b3a7..f20c6ad754 100644
--- a/drivers/usb/emul/sandbox_flash.c
+++ b/drivers/usb/emul/sandbox_flash.c
@@ -228,9 +228,9 @@ static void handle_read(struct sandbox_flash_priv *priv, ulong lba,
 			ulong transfer_len)
 {
 	debug("%s: lba=%lx, transfer_len=%lx\n", __func__, lba, transfer_len);
+	priv->read_len = transfer_len;
 	if (priv->fd != -1) {
 		os_lseek(priv->fd, lba * SANDBOX_FLASH_BLOCK_LEN, OS_SEEK_SET);
-		priv->read_len = transfer_len;
 		setup_response(priv, priv->buff,
 			       transfer_len * SANDBOX_FLASH_BLOCK_LEN);
 	} else {
@@ -336,6 +336,9 @@ static int sandbox_flash_bulk(struct udevice *dev, struct usb_device *udev,
 			if (priv->read_len) {
 				ulong bytes_read;
 
+				if (priv->fd == -1)
+					return -EIO;
+
 				bytes_read = os_read(priv->fd, buff, len);
 				if (bytes_read != len)
 					return -EIO;
-- 
2.35.1