Skip to contentSkip to conversation

mISDN: hfcpci: Fix a use after free in hfcmulti_tx()

PATCH In review Dan Carpenter 2 participants 3 messages Oct 22, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mISDN: hfcpci: Fix a use after free in hfcmulti_tx()

This frees "*sp" before dereferencing it to get "len = (*sp)->len;".

Fixes af69fb3a8ffa ("Add mISDN HFC multiport driver")
---
 drivers/isdn/hardware/mISDN/hfcmulti.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
drivers/isdn/hardware/mISDN/hfcmulti.c +4 -6
index 7013a3f08429..ce6c160e0df4 100644
@@ -2152,16 +2152,14 @@ hfcmulti_tx(struct hfc_multi *hc, int ch)
 		HFC_wait_nodebug(hc);
 	}
 
+	len = (*sp)->len;
 	dev_kfree_skb(*sp);
 	/* check for next frame */
-	if (bch && get_next_bframe(bch)) {
-		len = (*sp)->len;
+	if (bch && get_next_bframe(bch))
 		goto next_frame;
-	}
-	if (dch && get_next_dframe(dch)) {
-		len = (*sp)->len;
+
+	if (dch && get_next_dframe(dch))
 		goto next_frame;
-	}
 
 	/*
 	 * now we have no more data, so in case of transparent,
Signature · 1 line
2.28.0
#1 lore raw

Hi Dan,

that looks wrong to me and never was a use after free.

sp is set either to the address containing the pointer to the actual D-channel SKB or to the actual B-channel SKB. This address is not freed and will not change in this context. The dev_kfree(*sp) will delete the old SKB and the call to get_next_[bd]frame(), if returning true, will place a new SKB into this address, so (*sp) point to this new SKB. The len of course need to be the length of the new SKB, not the old one, which would be the result of this patch.

Best regards
Karsten

Quoted 34 lines · Dan Carpenter

This frees "*sp" before dereferencing it to get "len = (*sp)->len;".

Fixes af69fb3a8ffa ("Add mISDN HFC multiport driver")
---
 drivers/isdn/hardware/mISDN/hfcmulti.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
drivers/isdn/hardware/mISDN/hfcmulti.c +4 -6
index 7013a3f08429..ce6c160e0df4 100644
@@ -2152,16 +2152,14 @@ hfcmulti_tx(struct hfc_multi *hc, int ch)
 		HFC_wait_nodebug(hc);
 	}
 
+	len = (*sp)->len;
 	dev_kfree_skb(*sp);
 	/* check for next frame */
-	if (bch && get_next_bframe(bch)) {
-		len = (*sp)->len;
+	if (bch && get_next_bframe(bch))
 		goto next_frame;
-	}
-	if (dch && get_next_dframe(dch)) {
-		len = (*sp)->len;
+
+	if (dch && get_next_dframe(dch))
 		goto next_frame;
-	}
 
 	/*
 	 * now we have no more data, so in case of transparent,
#2 lore raw
Quoted 4 lines · On Thu, Oct 22, 2020 at 04:24:00PM +0200, [email protected] wrote:

Hi Dan,

that looks wrong to me and never was a use after free.

Quoted 8 lines

sp is set either to the address containing the pointer to the actual D-channel SKB or to the actual B-channel SKB. This address is not freed and will not change in this context. The dev_kfree(*sp) will delete the old SKB and the call to get_next_[bd]frame(), if returning true, will place a new SKB into this address, so (*sp) point to this new SKB. The len of course need to be the length of the new SKB, not the old one, which would be the result of this patch.

Oh, wow. You're absolutely right. That's pretty subtle. Thanks for catching it.

regards,
dan carpenter

Quoted 38 lines

Best regards
Karsten

Quoted 34 lines · Dan Carpenter

This frees "*sp" before dereferencing it to get "len = (*sp)->len;".

Fixes af69fb3a8ffa ("Add mISDN HFC multiport driver")
---
 drivers/isdn/hardware/mISDN/hfcmulti.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
drivers/isdn/hardware/mISDN/hfcmulti.c +4 -6
index 7013a3f08429..ce6c160e0df4 100644
@@ -2152,16 +2152,14 @@ hfcmulti_tx(struct hfc_multi *hc, int ch)
 		HFC_wait_nodebug(hc);
 	}
 
+	len = (*sp)->len;
 	dev_kfree_skb(*sp);
 	/* check for next frame */
-	if (bch && get_next_bframe(bch)) {
-		len = (*sp)->len;
+	if (bch && get_next_bframe(bch))
 		goto next_frame;
-	}
-	if (dch && get_next_dframe(dch)) {
-		len = (*sp)->len;
+
+	if (dch && get_next_dframe(dch))
 		goto next_frame;
-	}
 
 	/*
 	 * now we have no more data, so in case of transparent,