Skip to contentSkip to conversation

ath10k: htt rx fixes

PATCH New Michal Kazior 2 participants 12 messages Apr 26, 2013 – Apr 30, 2013 Qualcomm Atheros ATH10K wireless driver

lore
Share & tools
lore
#0 lore raw

cover ath10k: htt rx fixes

This patchset addresses issues with fragmented
(802.11b) frames.

Michal Kazior (7):
ath10k: zero htt_rx_info struct in frag rx handler
ath10k: do not pop more than one frame in frag rx handler
ath10k: add safety check after popping a frame in frag rx
ath10k: move ath10k_htt_rx_free_msdu_chain()
ath10k: drop all msdus in case of htt rx failure
ath10k: fix tkip fragment rx
ath10k: strip protected flag

 drivers/net/wireless/ath/ath10k/htt_rx.c |   54 ++++++++++++++++++------------
 drivers/net/wireless/ath/ath10k/txrx.c   |    7 +++-
 2 files changed, 38 insertions(+), 23 deletions(-)
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/7 ath10k: zero htt_rx_info struct in frag rx handler

The struct was filled with garbage and ended up
having fcs_err set to true. This caused mac to
drop fragmented frames.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/wireless/ath/ath10k/htt_rx.c +1 -1
index 055b88d..55c8083 100644
@@ -877,7 +877,7 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 	struct sk_buff *msdu_head, *msdu_tail;
 	struct htt_rx_desc *rxd;
 	enum rx_msdu_decap_format fmt;
-	struct htt_rx_info info;
+	struct htt_rx_info info = {};
 	int msdu_chaining;
 	bool tkip_mic_err;
 	bool decrypt_err;
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH 2/7 ath10k: do not pop more than one frame in frag rx handler

In reply to Michal Kazior

This led us to popping frames too early, before FW actually filled in frames. This was not an issue earlier. Perhaps the MSDU FW descriptor is now different.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |    4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/net/wireless/ath/ath10k/htt_rx.c +1 -3
index 55c8083..6d619e1 100644
@@ -888,7 +888,6 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 	fw_desc_len = __le16_to_cpu(frag->fw_rx_desc_bytes);
 	fw_desc = (u8 *)frag->fw_msdu_rx_desc;
 
-more:
 	msdu_head = NULL;
 	msdu_tail = NULL;
 	msdu_chaining = ath10k_htt_rx_amsdu_pop(htt, &fw_desc, &fw_desc_len,
@@ -966,8 +965,7 @@ more:
 end:
 	if (fw_desc_len > 0) {
 		ath10k_dbg(ATH10K_DBG_HTT,
-			   "expecting more fragmented rx in one indication\n");
-		goto more;
+			   "expecting more fragmented rx in one indication %d\n", fw_desc_len);
 	}
 }
 
Signature · 1 line
1.7.9.5
#3 lore raw

PATCH 3/7 ath10k: add safety check after popping a frame in frag rx

In reply to Michal Kazior

If HTT RX went haywire we dereferenced buggy
pointers that led to system crash.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |    5 +++++
 1 file changed, 5 insertions(+)
drivers/net/wireless/ath/ath10k/htt_rx.c +5 -0
index 6d619e1..fe2d27b 100644
@@ -895,6 +895,11 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 
 	ath10k_dbg(ATH10K_DBG_HTT_DUMP, "htt rx frag ahead\n");
 
+	if (!msdu_head) {
+		ath10k_warn("htt rx frag no data\n");
+		return;
+	}
+
 	if (msdu_chaining || msdu_head != msdu_tail) {
 		ath10k_warn("aggregation with fragmentation?!\n");
 		ath10k_htt_rx_free_msdu_chain(msdu_head);
Signature · 1 line
1.7.9.5
#4 lore raw

PATCH 4/7 ath10k: move ath10k_htt_rx_free_msdu_chain()

In reply to Michal Kazior

This makes it possible to call the function more
different places now.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |   22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)
drivers/net/wireless/ath/ath10k/htt_rx.c +11 -11
index fe2d27b..5afe42a 100644
@@ -235,6 +235,17 @@ static inline struct sk_buff *ath10k_htt_rx_netbuf_pop(struct ath10k_htt *htt)
 	return msdu;
 }
 
+static void ath10k_htt_rx_free_msdu_chain(struct sk_buff *skb)
+{
+	struct sk_buff *next;
+
+	while (skb) {
+		next = skb->next;
+		dev_kfree_skb_any(skb);
+		skb = next;
+	}
+}
+
 static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
 				   u8 **fw_desc, int *fw_desc_len,
 				   struct sk_buff **head_msdu,
@@ -545,17 +556,6 @@ static bool ath10k_htt_rx_hdr_is_amsdu(struct ieee80211_hdr *hdr)
 	return false;
 }
 
-static void ath10k_htt_rx_free_msdu_chain(struct sk_buff *skb)
-{
-	struct sk_buff *next;
-
-	while (skb) {
-		next = skb->next;
-		dev_kfree_skb_any(skb);
-		skb = next;
-	}
-}
-
 static int ath10k_htt_rx_amsdu(struct ath10k_htt *htt,
 			struct htt_rx_info *info)
 {
Signature · 1 line
1.7.9.5
#5 lore raw

PATCH 5/7 ath10k: drop all msdus in case of htt rx failure

In reply to Michal Kazior

It's a lot safer to simply drop everything. It's
broken anyway so why bother.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/net/wireless/ath/ath10k/htt_rx.c +3 -3
index 5afe42a..37784f6 100644
@@ -292,9 +292,9 @@ static int ath10k_htt_rx_amsdu_pop(struct ath10k_htt *htt,
 		 */
 		if (!(__le32_to_cpu(rx_desc->attention.flags)
 				& RX_ATTENTION_FLAGS_MSDU_DONE)) {
-			if (*head_msdu == msdu)
-				*head_msdu = NULL;
-			dev_kfree_skb_any(msdu);
+			ath10k_htt_rx_free_msdu_chain(*head_msdu);
+			*head_msdu = NULL;
+			msdu = NULL;
 			ath10k_err("htt rx stopped. cannot recover\n");
 			htt->rx_confused = true;
 			break;
Signature · 1 line
1.7.9.5
#6 lore raw

PATCH 6/7 ath10k: fix tkip fragment rx

In reply to Michal Kazior

TKIP fragmented frames contains different data at
the tail depending on the more flags.

---
 drivers/net/wireless/ath/ath10k/htt_rx.c |   15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
drivers/net/wireless/ath/ath10k/htt_rx.c +11 -4
index 37784f6..0b20879 100644
@@ -518,9 +518,8 @@ static int ath10k_htt_rx_crypto_tail_len(enum htt_rx_mpdu_encrypt_type type)
 	case HTT_RX_MPDU_ENCRYPT_WAPI:
 		return 0;
 	case HTT_RX_MPDU_ENCRYPT_TKIP_WITHOUT_MIC:
-		return 4;
 	case HTT_RX_MPDU_ENCRYPT_TKIP_WPA:
-		return 4 + 8;
+		return 4;
 	case HTT_RX_MPDU_ENCRYPT_AES_CCM_WPA2:
 		return 8;
 	}
@@ -878,6 +877,7 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 	struct htt_rx_desc *rxd;
 	enum rx_msdu_decap_format fmt;
 	struct htt_rx_info info = {};
+	struct ieee80211_hdr *hdr;
 	int msdu_chaining;
 	bool tkip_mic_err;
 	bool decrypt_err;
@@ -908,6 +908,7 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 
 	/* FIXME: implement signal strength */
 
+	hdr = (struct ieee80211_hdr *)msdu_head->data;
 	rxd = (void *)msdu_head->data - sizeof(*rxd);
 	tkip_mic_err = !!(__le32_to_cpu(rxd->attention.flags) &
 				RX_ATTENTION_FLAGS_TKIP_MIC_ERR);
@@ -939,7 +940,6 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 	}
 
 	if (info.encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
-		struct ieee80211_hdr *hdr = (void *)info.skb->data;
 		int hdrlen = ieee80211_hdrlen(hdr->frame_control);
 		int paramlen = ath10k_htt_rx_crypto_param_len(info.encrypt_type);
 
@@ -948,13 +948,20 @@ static void ath10k_htt_rx_frag_handler(struct ath10k_htt *htt,
 			(void *)info.skb->data,
 			hdrlen);
 		skb_pull(info.skb, paramlen);
+		hdr = (struct ieee80211_hdr *)info.skb->data;
 	}
 
 	/* remove trailing FCS */
 	trim  = 4;
-	/* remove crypto trailer; we use rx desc for mic failure */
+
+	/* remove crypto trailer */
 	trim += ath10k_htt_rx_crypto_tail_len(info.encrypt_type);
 
+	/* last fragment of TKIP frags has MIC */
+	if (!ieee80211_has_morefrags(hdr->frame_control) &&
+	    info.encrypt_type == HTT_RX_MPDU_ENCRYPT_TKIP_WPA)
+		trim += 8;
+
 	if (trim > info.skb->len) {
 		ath10k_warn("htt rx fragment: trailer longer than the frame itself? drop\n");
 		dev_kfree_skb_any(info.skb);
Signature · 1 line
1.7.9.5
#7 lore raw

PATCH 7/7 ath10k: strip protected flag

In reply to Michal Kazior

We already do decryption and all crypto
verification in HW. mac80211 tries to verify CCMP
for fragmented rx and fails without this patch.

This fixes fragmented rx on CCMP networks.

---
 drivers/net/wireless/ath/ath10k/txrx.c |    7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/net/wireless/ath/ath10k/txrx.c +6 -1
index ab5eb3b..a54ad8f 100644
@@ -238,13 +238,18 @@ void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info)
 {
 	struct ieee80211_rx_status *status;
 	struct ieee80211_channel *ch;
+	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)info->skb->data;
 
 	status = IEEE80211_SKB_RXCB(info->skb);
 	memset(status, 0, sizeof(*status));
 
-	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE)
+	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
 		status->flag |= RX_FLAG_DECRYPTED | RX_FLAG_IV_STRIPPED |
 				RX_FLAG_MMIC_STRIPPED;
+		hdr->frame_control = __cpu_to_le16(
+				__le16_to_cpu(hdr->frame_control) &
+				~IEEE80211_FCTL_PROTECTED);
+	}
 
 	if (info->status == HTT_RX_IND_MPDU_STATUS_TKIP_MIC_ERR)
 		status->flag |= RX_FLAG_MMIC_ERROR;
Signature · 1 line
1.7.9.5
#8 lore raw

PATCH 7/7 ath10k: strip protected flag

Quoted 7 lines · Michal Kazior

We already do decryption and all crypto
verification in HW. mac80211 tries to verify CCMP
for fragmented rx and fails without this patch.

This fixes fragmented rx on CCMP networks.

[...]

drivers/net/wireless/ath/ath10k/txrx.c void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info) · line 252

Quoted 8 lines
drivers/net/wireless/ath/ath10k/txrx.c +5 -1
-	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE)
+	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
 		status->flag |= RX_FLAG_DECRYPTED | RX_FLAG_IV_STRIPPED |
 				RX_FLAG_MMIC_STRIPPED;
+		hdr->frame_control = __cpu_to_le16(
+				__le16_to_cpu(hdr->frame_control) &
+				~IEEE80211_FCTL_PROTECTED);
+	}

This looks ugly. Is this a bug in mac80211 which we just workaround in ath10k? Isn't there any other way to fix this? At least this deserves a big comment in the code to explain why we are doing this.

I'll apply this anyway, but we need to discuss more about this.

Signature · 1 line
Kalle Valo
#9 lore raw

PATCH 7/7 ath10k: strip protected flag

Quoted 25 lines · Kalle Valo
Quoted 7 lines · Michal Kazior

We already do decryption and all crypto
verification in HW. mac80211 tries to verify CCMP
for fragmented rx and fails without this patch.

This fixes fragmented rx on CCMP networks.

[...]

drivers/net/wireless/ath/ath10k/txrx.c void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info) · line 252

Quoted 8 lines
drivers/net/wireless/ath/ath10k/txrx.c +5 -1
-	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE)
+	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
  		status->flag |= RX_FLAG_DECRYPTED | RX_FLAG_IV_STRIPPED |
  				RX_FLAG_MMIC_STRIPPED;
+		hdr->frame_control = __cpu_to_le16(
+				__le16_to_cpu(hdr->frame_control) &
+				~IEEE80211_FCTL_PROTECTED);
+	}

This looks ugly. Is this a bug in mac80211 which we just workaround in ath10k? Isn't there any other way to fix this? At least this deserves a big comment in the code to explain why we are doing this.

Quoted 1 line

I'll apply this anyway, but we need to discuss more about this.

I'm not quite sure whether this is an issue in mac80211 or we should be actually stripping the protected flag. b43 driver seems to strip the flag as well.

mac80211 seems to want to do some extra checks for CCMP in fragmented rx data path. Perhaps it shouldn't?

-- Pozdrawiam / Best regards, Michal Kazior.

#10 lore raw

PATCH 7/7 ath10k: strip protected flag

Quoted 26 lines · Michal Kazior
Quoted 25 lines · Kalle Valo
Quoted 7 lines · Michal Kazior

We already do decryption and all crypto
verification in HW. mac80211 tries to verify CCMP
for fragmented rx and fails without this patch.

This fixes fragmented rx on CCMP networks.

[...]

drivers/net/wireless/ath/ath10k/txrx.c void ath10k_process_rx(struct ath10k *ar, struct htt_rx_info *info) · line 252

Quoted 8 lines
drivers/net/wireless/ath/ath10k/txrx.c +5 -1
-	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE)
+	if (info->encrypt_type != HTT_RX_MPDU_ENCRYPT_NONE) {
  		status->flag |= RX_FLAG_DECRYPTED | RX_FLAG_IV_STRIPPED |
  				RX_FLAG_MMIC_STRIPPED;
+		hdr->frame_control = __cpu_to_le16(
+				__le16_to_cpu(hdr->frame_control) &
+				~IEEE80211_FCTL_PROTECTED);
+	}

This looks ugly. Is this a bug in mac80211 which we just workaround in ath10k? Isn't there any other way to fix this? At least this deserves a big comment in the code to explain why we are doing this.

Quoted 5 lines
Quoted 1 line

I'll apply this anyway, but we need to discuss more about this.

I'm not quite sure whether this is an issue in mac80211 or we should be actually stripping the protected flag. b43 driver seems to strip the flag as well.

If b43 does the same when I'm happy with this. Just send a new patch adding a comment to the code as well.

Quoted 2 lines

mac80211 seems to want to do some extra checks for CCMP in fragmented rx data path. Perhaps it shouldn't?

IMHO it shouldn't, but I'm no security expert. Maybe you should ask in the linux-wireless list?

Signature · 1 line
Kalle Valo
#11 lore raw

cover ath10k: htt rx fixes

In reply to Michal Kazior
Quoted 11 lines · Michal Kazior

This patchset addresses issues with fragmented
(802.11b) frames.

Michal Kazior (7):
ath10k: zero htt_rx_info struct in frag rx handler
ath10k: do not pop more than one frame in frag rx handler
ath10k: add safety check after popping a frame in frag rx
ath10k: move ath10k_htt_rx_free_msdu_chain()
ath10k: drop all msdus in case of htt rx failure
ath10k: fix tkip fragment rx
ath10k: strip protected flag

Thanks, all seven applied.

Signature · 1 line
Kalle Valo