[PATCH] s390: qeth driver fixes [3/6]

[PATCH 6/9] s390: qeth driver fixes [3/6]

From: Frank Pavlic <fpavlic@de.ibm.com>
       	fixed kernel panic caused by qeth driver:
        Using a bonding device qeth driver will realloc
        headroom for every skb coming from the bond device.
        Once this happens qeth frees the original skb and
        set the skb pointer to the new realloced skb.
        Under heavy transmit workload (e.g.UDP streams) through bond
        network device the qdio output queue might get full.
        In this case we return with EBUSY from qeth_send_packet.
        Returning to qeth_hard_start_xmit routine
        the skb address on the stack still points to the old address,
        which has been freed before.
        Returning from qeth_hard_start_xmit with EBUSY results in
        requeuing the skb. In this case it corrupts the qdisc queue
        and results in kernel panic.

Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

1 files changed, 99 insertions, 94 deletions

diff --git a/drivers/s390/net/qeth_main.c b/drivers/s390/net/qeth_main.c
index ffff7a12b54d..522fb9dd551e 100644
--- a/drivers/s390/net/qeth_main.c
+++ b/drivers/s390/net/qeth_main.c
@@ -3919,49 +3919,59 @@ qeth_get_ip_version(struct sk_buff *skb)
 	}
 }
 
-static inline int
-qeth_prepare_skb(struct qeth_card *card, struct sk_buff **skb,
-		 struct qeth_hdr **hdr, int ipv)
+static inline struct qeth_hdr *
+__qeth_prepare_skb(struct qeth_card *card, struct sk_buff *skb, int ipv)
 {
-	int rc = 0;
 #ifdef CONFIG_QETH_VLAN
 	u16 *tag;
-#endif
-
-	QETH_DBF_TEXT(trace, 6, "prepskb");
-	if (card->info.type == QETH_CARD_TYPE_OSN) {
-		*hdr = (struct qeth_hdr *)(*skb)->data;
-		return rc;
-	}
-        rc = qeth_realloc_headroom(card, skb, sizeof(struct qeth_hdr));
-        if (rc)
-                return rc;
-#ifdef CONFIG_QETH_VLAN
-	if (card->vlangrp && vlan_tx_tag_present(*skb) &&
+	if (card->vlangrp && vlan_tx_tag_present(skb) &&
 	    ((ipv == 6) || card->options.layer2) ) {
 		/*
 		 * Move the mac addresses (6 bytes src, 6 bytes dest)
 		 * to the beginning of the new header.  We are using three
 		 * memcpys instead of one memmove to save cycles.
 		 */
-		skb_push(*skb, VLAN_HLEN);
-		memcpy((*skb)->data, (*skb)->data + 4, 4);
-		memcpy((*skb)->data + 4, (*skb)->data + 8, 4);
-		memcpy((*skb)->data + 8, (*skb)->data + 12, 4);
-		tag = (u16 *)((*skb)->data + 12);
+		skb_push(skb, VLAN_HLEN);
+		memcpy(skb->data, skb->data + 4, 4);
+		memcpy(skb->data + 4, skb->data + 8, 4);
+		memcpy(skb->data + 8, skb->data + 12, 4);
+		tag = (u16 *)(skb->data + 12);
 		/*
 		 * first two bytes  = ETH_P_8021Q (0x8100)
 		 * second two bytes = VLANID
 		 */
 		*tag = __constant_htons(ETH_P_8021Q);
-		*(tag + 1) = htons(vlan_tx_tag_get(*skb));
+		*(tag + 1) = htons(vlan_tx_tag_get(skb));
 	}
 #endif
-	*hdr = (struct qeth_hdr *)
-		qeth_push_skb(card, skb, sizeof(struct qeth_hdr));
-	if (*hdr == NULL)
-		return -EINVAL;
-	return 0;
+	return ((struct qeth_hdr *)
+		qeth_push_skb(card, skb, sizeof(struct qeth_hdr)));
+}
+
+static inline void
+__qeth_free_new_skb(struct sk_buff *orig_skb, struct sk_buff *new_skb)
+{
+	if (orig_skb != new_skb)
+		dev_kfree_skb_any(new_skb);
+}
+
+static inline struct sk_buff *
+qeth_prepare_skb(struct qeth_card *card, struct sk_buff *skb,
+		 struct qeth_hdr **hdr, int ipv)
+{
+	struct sk_buff *new_skb;
+	
+	QETH_DBF_TEXT(trace, 6, "prepskb");
+
+        new_skb = qeth_realloc_headroom(card, skb, sizeof(struct qeth_hdr));
+       	if (new_skb == NULL)
+		return NULL;
+	*hdr = __qeth_prepare_skb(card, new_skb, ipv);
+	if (*hdr == NULL) {
+		__qeth_free_new_skb(skb, new_skb);
+		return NULL;
+	}
+	return new_skb;
 }
 
 static inline u8
@@ -4242,21 +4252,15 @@ qeth_do_send_packet_fast(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 	 * check if buffer is empty to make sure that we do not 'overtake'
 	 * ourselves and try to fill a buffer that is already primed
 	 */
-	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY) {
-		card->stats.tx_dropped++;
-		atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
-		return -EBUSY;
-	}
+	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY) 
+		goto out;
 	if (ctx == NULL)
 		queue->next_buf_to_fill = (queue->next_buf_to_fill + 1) %
 					  QDIO_MAX_BUFFERS_PER_Q;
 	else {
 		buffers_needed = qeth_eddp_check_buffers_for_context(queue,ctx);
-		if (buffers_needed < 0) {
-			card->stats.tx_dropped++;
-			atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
-			return -EBUSY;
-		}
+		if (buffers_needed < 0) 
+			goto out;
 		queue->next_buf_to_fill =
 			(queue->next_buf_to_fill + buffers_needed) %
 			QDIO_MAX_BUFFERS_PER_Q;
@@ -4271,6 +4275,9 @@ qeth_do_send_packet_fast(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 		qeth_flush_buffers(queue, 0, index, flush_cnt);
 	}
 	return 0;
+out:
+	atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
+	return -EBUSY;
 }
 
 static inline int
@@ -4296,8 +4303,7 @@ qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 	 * check if buffer is empty to make sure that we do not 'overtake'
 	 * ourselves and try to fill a buffer that is already primed
 	 */
-	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY){
-		card->stats.tx_dropped++;
+	if (atomic_read(&buffer->state) != QETH_QDIO_BUF_EMPTY) {
 		atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
 		return -EBUSY;
 	}
@@ -4320,7 +4326,6 @@ qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 				 * again */
 				if (atomic_read(&buffer->state) !=
 						QETH_QDIO_BUF_EMPTY){
-					card->stats.tx_dropped++;
 					qeth_flush_buffers(queue, 0, start_index, flush_count);
 					atomic_set(&queue->state, QETH_OUT_Q_UNLOCKED);
 					return -EBUSY;
@@ -4331,7 +4336,6 @@ qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 			 * free buffers) to handle eddp context */
 			if (qeth_eddp_check_buffers_for_context(queue,ctx) < 0){
 				printk("eddp tx_dropped 1\n");
-				card->stats.tx_dropped++;
 				rc = -EBUSY;
 				goto out;
 			}
@@ -4343,7 +4347,6 @@ qeth_do_send_packet(struct qeth_card *card, struct qeth_qdio_out_q *queue,
 		tmp = qeth_eddp_fill_buffer(queue,ctx,queue->next_buf_to_fill);
 		if (tmp < 0) {
 			printk("eddp tx_dropped 2\n");
-			card->stats.tx_dropped++;
 			rc = - EBUSY;
 			goto out;
 		}
@@ -4391,21 +4394,21 @@ qeth_get_elements_no(struct qeth_card *card, void *hdr,
 {
 	int elements_needed = 0;
 
-        if (skb_shinfo(skb)->nr_frags > 0) {
+        if (skb_shinfo(skb)->nr_frags > 0) 
                 elements_needed = (skb_shinfo(skb)->nr_frags + 1);
-	}
-        if (elements_needed == 0 )
+        if (elements_needed == 0)
                 elements_needed = 1 + (((((unsigned long) hdr) % PAGE_SIZE)
                                         + skb->len) >> PAGE_SHIFT);
 	if ((elements_needed + elems) > QETH_MAX_BUFFER_ELEMENTS(card)){
-                PRINT_ERR("qeth_do_send_packet: invalid size of "
-                          "IP packet (Number=%d / Length=%d). Discarded.\n",
+                PRINT_ERR("Invalid size of IP packet "
+			  "(Number=%d / Length=%d). Discarded.\n",
                           (elements_needed+elems), skb->len);
                 return 0;
         }
         return elements_needed;
 }
 
+
 static inline int
 qeth_send_packet(struct qeth_card *card, struct sk_buff *skb)
 {
@@ -4421,108 +4424,110 @@ qeth_send_packet(struct qeth_card *card, struct sk_buff *skb)
 	unsigned short nr_frags = skb_shinfo(skb)->nr_frags;
 	unsigned short tso_size = skb_shinfo(skb)->gso_size;
 #endif
+	struct sk_buff *new_skb, *new_skb2;
 	int rc;
 
 	QETH_DBF_TEXT(trace, 6, "sendpkt");
 
+	new_skb = skb;
+	if ((card->info.type == QETH_CARD_TYPE_OSN) &&
+	    (skb->protocol == htons(ETH_P_IPV6)))
+		return -EPERM;
+	cast_type = qeth_get_cast_type(card, skb);
+	if ((cast_type == RTN_BROADCAST) &&
+	    (card->info.broadcast_capable == 0))
+		return -EPERM;
+	queue = card->qdio.out_qs
+		[qeth_get_priority_queue(card, skb, ipv, cast_type)];
 	if (!card->options.layer2) {
 		ipv = qeth_get_ip_version(skb);
 		if ((card->dev->hard_header == qeth_fake_header) && ipv) {
-               		if ((skb = qeth_pskb_unshare(skb,GFP_ATOMIC)) == NULL) {
-                        	card->stats.tx_dropped++;
-                        	dev_kfree_skb_irq(skb);
-                        	return 0;
-                	}
+			new_skb = qeth_pskb_unshare(skb, GFP_ATOMIC);
+			if (!new_skb)
+				return -ENOMEM;
 			if(card->dev->type == ARPHRD_IEEE802_TR){
-				skb_pull(skb, QETH_FAKE_LL_LEN_TR);
+				skb_pull(new_skb, QETH_FAKE_LL_LEN_TR);
 			} else {
-                		skb_pull(skb, QETH_FAKE_LL_LEN_ETH);
+				skb_pull(new_skb, QETH_FAKE_LL_LEN_ETH);
 			}
 		}
 	}
-	if ((card->info.type == QETH_CARD_TYPE_OSN) &&
-		(skb->protocol == htons(ETH_P_IPV6))) {
-		dev_kfree_skb_any(skb);
-		return 0;
-	}
-	cast_type = qeth_get_cast_type(card, skb);
-	if ((cast_type == RTN_BROADCAST) &&
-	    (card->info.broadcast_capable == 0)){
-		card->stats.tx_dropped++;
-		card->stats.tx_errors++;
-		dev_kfree_skb_any(skb);
-		return NETDEV_TX_OK;
-	}
-	queue = card->qdio.out_qs
-		[qeth_get_priority_queue(card, skb, ipv, cast_type)];
-
 	if (skb_is_gso(skb))
 		large_send = card->options.large_send;
-
-	/*are we able to do TSO ? If so ,prepare and send it from here */
+	/* check on OSN device*/
+	if (card->info.type == QETH_CARD_TYPE_OSN)
+		hdr = (struct qeth_hdr *)new_skb->data;
+	/*are we able to do TSO ? */
 	if ((large_send == QETH_LARGE_SEND_TSO) &&
 	    (cast_type == RTN_UNSPEC)) {
-		rc = qeth_tso_prepare_packet(card, skb, ipv, cast_type);
+		rc = qeth_tso_prepare_packet(card, new_skb, ipv, cast_type);
 		if (rc) {
-			card->stats.tx_dropped++;
-			card->stats.tx_errors++;
-			dev_kfree_skb_any(skb);
-			return NETDEV_TX_OK;
+			__qeth_free_new_skb(skb, new_skb);
+			return rc;
 		}
 		elements_needed++;
-	} else {
-		if ((rc = qeth_prepare_skb(card, &skb, &hdr, ipv))) {
-			QETH_DBF_TEXT_(trace, 4, "pskbe%d", rc);
-			return rc;
+	} else if (card->info.type != QETH_CARD_TYPE_OSN) {
+		new_skb2 = qeth_prepare_skb(card, new_skb, &hdr, ipv);
+		if (!new_skb2) {
+			__qeth_free_new_skb(skb, new_skb);
+			return -EINVAL;
 		}
-		if (card->info.type != QETH_CARD_TYPE_OSN)
-			qeth_fill_header(card, hdr, skb, ipv, cast_type);
+		if (new_skb != skb)
+			__qeth_free_new_skb(new_skb2, new_skb);
+		new_skb = new_skb2;
+		qeth_fill_header(card, hdr, new_skb, ipv, cast_type);
 	}
-
 	if (large_send == QETH_LARGE_SEND_EDDP) {
-		ctx = qeth_eddp_create_context(card, skb, hdr);
+		ctx = qeth_eddp_create_context(card, new_skb, hdr);
 		if (ctx == NULL) {
+			__qeth_free_new_skb(skb, new_skb);
 			PRINT_WARN("could not create eddp context\n");
 			return -EINVAL;
 		}
 	} else {
-		int elems = qeth_get_elements_no(card,(void*) hdr, skb,
+		int elems = qeth_get_elements_no(card,(void*) hdr, new_skb,
 						 elements_needed);
-		if (!elems)
+		if (!elems) {
+			__qeth_free_new_skb(skb, new_skb);
 			return -EINVAL;
+		}
 		elements_needed += elems;
 	}
 
 	if (card->info.type != QETH_CARD_TYPE_IQD)
-		rc = qeth_do_send_packet(card, queue, skb, hdr,
+		rc = qeth_do_send_packet(card, queue, new_skb, hdr,
 					 elements_needed, ctx);
 	else
-		rc = qeth_do_send_packet_fast(card, queue, skb, hdr,
+		rc = qeth_do_send_packet_fast(card, queue, new_skb, hdr,
 					      elements_needed, ctx);
-	if (!rc){
+	if (!rc) {
 		card->stats.tx_packets++;
 		card->stats.tx_bytes += tx_bytes;
+		if (new_skb != skb)
+			dev_kfree_skb_any(skb);
 #ifdef CONFIG_QETH_PERF_STATS
 		if (tso_size &&
 		   !(large_send == QETH_LARGE_SEND_NO)) {
 			card->perf_stats.large_send_bytes += tx_bytes;
 			card->perf_stats.large_send_cnt++;
 		}
- 		if (nr_frags > 0){
+		if (nr_frags > 0) {
 			card->perf_stats.sg_skbs_sent++;
 			/* nr_frags + skb->data */
 			card->perf_stats.sg_frags_sent +=
 				nr_frags + 1;
 		}
 #endif /* CONFIG_QETH_PERF_STATS */
+	} else {
+		card->stats.tx_dropped++;
+		__qeth_free_new_skb(skb, new_skb);
 	}
 	if (ctx != NULL) {
 		/* drop creator's reference */
 		qeth_eddp_put_context(ctx);
 		/* free skb; it's not referenced by a buffer */
-		if (rc == 0)
-			dev_kfree_skb_any(skb);
-
+		if (!rc)
+		       dev_kfree_skb_any(new_skb);
 	}
 	return rc;
 }