[PATCH net 1/2] tuntap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

[PATCH net 1/2] tuntap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

[Jason Wang]

We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
network.

Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than
MAX_SKB_FRAGS.

This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify
vhost.

We can do further optimization on top.

The bug were introduced from commit 0690899b4d4501b3505be069b9a687e68ccbe15b
(tun: experimental zero copy tx support)

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/net/tun.c |   62 ++++++++++++++++++++++++++++++++--------------------
 1 files changed, 38 insertions(+), 24 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 5cdcf92..db690a3 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1035,6 +1035,29 @@ static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 	return 0;
 }
 
+static unsigned long iov_pages(const struct iovec *iv, int offset,
+			       unsigned long nr_segs)
+{
+	unsigned long seg, base;
+	int pages = 0, len, size;
+
+	while (nr_segs && (offset >= iv->iov_len)) {
+		offset -= iv->iov_len;
+		++iv;
+		--nr_segs;
+	}
+
+	for (seg = 0; seg < nr_segs; seg++) {
+		base = (unsigned long)iv[seg].iov_base + offset;
+		len = iv[seg].iov_len - offset;
+		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+		pages += size;
+		offset = 0;
+	}
+
+	return pages;
+}
+
 /* Get packet from user space buffer */
 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 			    void *msg_control, const struct iovec *iv,
@@ -1082,32 +1105,18 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 			return -EINVAL;
 	}
 
-	if (msg_control)
-		zerocopy = true;
-
-	if (zerocopy) {
-		/* Userspace may produce vectors with count greater than
-		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
-		 * to let the rest of data to be fit in the frags.
-		 */
-		if (count > MAX_SKB_FRAGS) {
-			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
-			if (copylen < offset)
-				copylen = 0;
-			else
-				copylen -= offset;
-		} else
-				copylen = 0;
-		/* There are 256 bytes to be copied in skb, so there is enough
-		 * room for skb expand head in case it is used.
+	if (msg_control) {
+		/* There are 256 bytes to be copied in skb, so there is
+		 * enough room for skb expand head in case it is used.
 		 * The rest of the buffer is mapped from userspace.
 		 */
-		if (copylen < gso.hdr_len)
-			copylen = gso.hdr_len;
-		if (!copylen)
-			copylen = GOODCOPY_LEN;
+		copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
 		linear = copylen;
-	} else {
+		if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
+			zerocopy = true;
+	}
+
+	if (!zerocopy) {
 		copylen = len;
 		linear = gso.hdr_len;
 	}
@@ -1121,8 +1130,13 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 
 	if (zerocopy)
 		err = zerocopy_sg_from_iovec(skb, iv, offset, count);
-	else
+	else {
 		err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
+		if (!err && msg_control) {
+			struct ubuf_info *uarg = msg_control;
+			uarg->callback(uarg, false);
+		}
+	}
 
 	if (err) {
 		tun->dev->stats.rx_dropped++;
-- 
1.7.1

[PATCH net 2/2] macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

[Jason Wang]

We try to linearize part of the skb when the number of iov is greater than
MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
network.

Solve this problem by calculate the pages needed for iov before trying to do
zerocopy and switch to use copy instead of zerocopy if it needs more than
MAX_SKB_FRAGS.

This is done through introducing a new helper to count the pages for iov, and
call uarg->callback() manually when switching from zerocopy to copy to notify
vhost.

We can do further optimization on top.

This bug were introduced from b92946e2919134ebe2a4083e4302236295ea2a73
(macvtap: zerocopy: validate vectors before building skb).

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/net/macvtap.c |   58 +++++++++++++++++++++++++++++++-----------------
 1 files changed, 37 insertions(+), 21 deletions(-)

diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
index a7c5654..e200f11 100644
--- a/drivers/net/macvtap.c
+++ b/drivers/net/macvtap.c
@@ -698,6 +698,28 @@ static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 	return 0;
 }
 
+static unsigned long iov_pages(const struct iovec *iv, int offset,
+			       unsigned long nr_segs)
+{
+	unsigned long seg, base;
+	int pages = 0, len, size;
+
+	while (nr_segs && (offset >= iv->iov_len)) {
+		offset -= iv->iov_len;
+		++iv;
+		--nr_segs;
+	}
+
+	for (seg = 0; seg < nr_segs; seg++) {
+		base = (unsigned long)iv[seg].iov_base + offset;
+		len = iv[seg].iov_len - offset;
+		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+		pages += size;
+		offset = 0;
+	}
+
+	return pages;
+}
 
 /* Get packet from user space buffer */
 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
@@ -744,31 +766,19 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 	if (unlikely(count > UIO_MAXIOV))
 		goto err;
 
-	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
-		zerocopy = true;
-
-	if (zerocopy) {
+	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 		/* Userspace may produce vectors with count greater than
 		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
 		 * to let the rest of data to be fit in the frags.
 		 */
-		if (count > MAX_SKB_FRAGS) {
-			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
-			if (copylen < vnet_hdr_len)
-				copylen = 0;
-			else
-				copylen -= vnet_hdr_len;
-		}
-		/* There are 256 bytes to be copied in skb, so there is enough
-		 * room for skb expand head in case it is used.
-		 * The rest buffer is mapped from userspace.
-		 */
-		if (copylen < vnet_hdr.hdr_len)
-			copylen = vnet_hdr.hdr_len;
-		if (!copylen)
-			copylen = GOODCOPY_LEN;
+		copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
 		linear = copylen;
-	} else {
+		if (iov_pages(iv, vnet_hdr_len + copylen, count)
+		    <= MAX_SKB_FRAGS)
+			zerocopy = true;
+	}
+
+	if (!zerocopy) {
 		copylen = len;
 		linear = vnet_hdr.hdr_len;
 	}
@@ -780,9 +790,15 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 
 	if (zerocopy)
 		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
-	else
+	else {
 		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 						   len);
+		if (!err && m && m->msg_control) {
+			struct ubuf_info *uarg = m->msg_control;
+			uarg->callback(uarg, false);
+		}
+	}
+
 	if (err)
 		goto err_kfree;
 
-- 
1.7.1

Re: [PATCH net 2/2] macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

[Michael S. Tsirkin]

On Wed, Jul 17, 2013 at 06:32:11PM +0800, Jason Wang wrote:
> We try to linearize part of the skb when the number of iov is greater than
> MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
> one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
> network.
> 
> Solve this problem by calculate the pages needed for iov before trying to do
> zerocopy and switch to use copy instead of zerocopy if it needs more than
> MAX_SKB_FRAGS.
> 
> This is done through introducing a new helper to count the pages for iov, and
> call uarg->callback() manually when switching from zerocopy to copy to notify
> vhost.
> 
> We can do further optimization on top.
> 
> This bug were introduced from b92946e2919134ebe2a4083e4302236295ea2a73
> (macvtap: zerocopy: validate vectors before building skb).
> 
> Cc: Michael S. Tsirkin <mst@redhat.com>
> Signed-off-by: Jason Wang <jasowang@redhat.com>
> ---
>  drivers/net/macvtap.c |   58 +++++++++++++++++++++++++++++++-----------------
>  1 files changed, 37 insertions(+), 21 deletions(-)
> 
> diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
> index a7c5654..e200f11 100644
> --- a/drivers/net/macvtap.c
> +++ b/drivers/net/macvtap.c
> @@ -698,6 +698,28 @@ static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
>  	return 0;
>  }
>  
> +static unsigned long iov_pages(const struct iovec *iv, int offset,
> +			       unsigned long nr_segs)
> +{
> +	unsigned long seg, base;
> +	int pages = 0, len, size;
> +
> +	while (nr_segs && (offset >= iv->iov_len)) {
> +		offset -= iv->iov_len;
> +		++iv;
> +		--nr_segs;
> +	}
> +
> +	for (seg = 0; seg < nr_segs; seg++) {
> +		base = (unsigned long)iv[seg].iov_base + offset;
> +		len = iv[seg].iov_len - offset;
> +		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
> +		pages += size;
> +		offset = 0;
> +	}
> +
> +	return pages;
> +}
>  
>  /* Get packet from user space buffer */
>  static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
> @@ -744,31 +766,19 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
>  	if (unlikely(count > UIO_MAXIOV))
>  		goto err;
>  
> -	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
> -		zerocopy = true;
> -
> -	if (zerocopy) {
> +	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
>  		/* Userspace may produce vectors with count greater than
>  		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
>  		 * to let the rest of data to be fit in the frags.
>  		 */

We don't linearize parts to make them fit, we simply
fall back to data copy.
So this comment no longer applies, right?

> -		if (count > MAX_SKB_FRAGS) {
> -			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
> -			if (copylen < vnet_hdr_len)
> -				copylen = 0;
> -			else
> -				copylen -= vnet_hdr_len;
> -		}
> -		/* There are 256 bytes to be copied in skb, so there is enough
> -		 * room for skb expand head in case it is used.
> -		 * The rest buffer is mapped from userspace.
> -		 */
> -		if (copylen < vnet_hdr.hdr_len)
> -			copylen = vnet_hdr.hdr_len;
> -		if (!copylen)
> -			copylen = GOODCOPY_LEN;
> +		copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
>  		linear = copylen;
> -	} else {
> +		if (iov_pages(iv, vnet_hdr_len + copylen, count)
> +		    <= MAX_SKB_FRAGS)
> +			zerocopy = true;
> +	}
> +
> +	if (!zerocopy) {
>  		copylen = len;
>  		linear = vnet_hdr.hdr_len;
>  	}
> @@ -780,9 +790,15 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
>  
>  	if (zerocopy)
>  		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
> -	else
> +	else {
>  		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
>  						   len);
> +		if (!err && m && m->msg_control) {
> +			struct ubuf_info *uarg = m->msg_control;
> +			uarg->callback(uarg, false);
> +		}
> +	}
> +
>  	if (err)
>  		goto err_kfree;
>  
> -- 
> 1.7.1

Re: [PATCH net 2/2] macvtap: do not zerocopy if iov needs more pages than MAX_SKB_FRAGS

[Jason Wang]

On 07/17/2013 07:06 PM, Michael S. Tsirkin wrote:
> On Wed, Jul 17, 2013 at 06:32:11PM +0800, Jason Wang wrote:
>> We try to linearize part of the skb when the number of iov is greater than
>> MAX_SKB_FRAGS. This is not enough since each single vector may occupy more than
>> one pages, so zerocopy_sg_fromiovec() may still fail and may break the guest
>> network.
>>
>> Solve this problem by calculate the pages needed for iov before trying to do
>> zerocopy and switch to use copy instead of zerocopy if it needs more than
>> MAX_SKB_FRAGS.
>>
>> This is done through introducing a new helper to count the pages for iov, and
>> call uarg->callback() manually when switching from zerocopy to copy to notify
>> vhost.
>>
>> We can do further optimization on top.
>>
>> This bug were introduced from b92946e2919134ebe2a4083e4302236295ea2a73
>> (macvtap: zerocopy: validate vectors before building skb).
>>
>> Cc: Michael S. Tsirkin <mst@redhat.com>
>> Signed-off-by: Jason Wang <jasowang@redhat.com>
>> ---
>>  drivers/net/macvtap.c |   58 +++++++++++++++++++++++++++++++-----------------
>>  1 files changed, 37 insertions(+), 21 deletions(-)
>>
>> diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
>> index a7c5654..e200f11 100644
>> --- a/drivers/net/macvtap.c
>> +++ b/drivers/net/macvtap.c
>> @@ -698,6 +698,28 @@ static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
>>  	return 0;
>>  }
>>  
>> +static unsigned long iov_pages(const struct iovec *iv, int offset,
>> +			       unsigned long nr_segs)
>> +{
>> +	unsigned long seg, base;
>> +	int pages = 0, len, size;
>> +
>> +	while (nr_segs && (offset >= iv->iov_len)) {
>> +		offset -= iv->iov_len;
>> +		++iv;
>> +		--nr_segs;
>> +	}
>> +
>> +	for (seg = 0; seg < nr_segs; seg++) {
>> +		base = (unsigned long)iv[seg].iov_base + offset;
>> +		len = iv[seg].iov_len - offset;
>> +		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
>> +		pages += size;
>> +		offset = 0;
>> +	}
>> +
>> +	return pages;
>> +}
>>  
>>  /* Get packet from user space buffer */
>>  static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
>> @@ -744,31 +766,19 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
>>  	if (unlikely(count > UIO_MAXIOV))
>>  		goto err;
>>  
>> -	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
>> -		zerocopy = true;
>> -
>> -	if (zerocopy) {
>> +	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
>>  		/* Userspace may produce vectors with count greater than
>>  		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
>>  		 * to let the rest of data to be fit in the frags.
>>  		 */
> We don't linearize parts to make them fit, we simply
> fall back to data copy.
> So this comment no longer applies, right?

Right, I forget to remove them. Will do it in V2.
>> -		if (count > MAX_SKB_FRAGS) {
>> -			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
>> -			if (copylen < vnet_hdr_len)
>> -				copylen = 0;
>> -			else
>> -				copylen -= vnet_hdr_len;
>> -		}
>> -		/* There are 256 bytes to be copied in skb, so there is enough
>> -		 * room for skb expand head in case it is used.
>> -		 * The rest buffer is mapped from userspace.
>> -		 */
>> -		if (copylen < vnet_hdr.hdr_len)
>> -			copylen = vnet_hdr.hdr_len;
>> -		if (!copylen)
>> -			copylen = GOODCOPY_LEN;
>> +		copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
>>  		linear = copylen;
>> -	} else {
>> +		if (iov_pages(iv, vnet_hdr_len + copylen, count)
>> +		    <= MAX_SKB_FRAGS)
>> +			zerocopy = true;
>> +	}
>> +
>> +	if (!zerocopy) {
>>  		copylen = len;
>>  		linear = vnet_hdr.hdr_len;
>>  	}
>> @@ -780,9 +790,15 @@ static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
>>  
>>  	if (zerocopy)
>>  		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
>> -	else
>> +	else {
>>  		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
>>  						   len);
>> +		if (!err && m && m->msg_control) {
>> +			struct ubuf_info *uarg = m->msg_control;
>> +			uarg->callback(uarg, false);
>> +		}
>> +	}
>> +
>>  	if (err)
>>  		goto err_kfree;
>>  
>> -- 
>> 1.7.1
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/