Skip to contentSkip to conversation

net, gro: fix the truesize of all fragments

RFC PATCH In review Zhi Yong Wu 3 participants 8 messages Dec 20, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 net, gro: fix the truesize of all fragments

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
 net/core/skbuff.c |   10 +++++++---
 1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +7 -3
index 2b6b863..0665bd6 100644
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 		frag->page_offset += offset;
 		skb_frag_size_sub(frag, offset);
 
-		/* all fragments truesize : remove (head size + sk_buff) */
+		/**
+		 * all fragments truesize :
+		 * remove (head size + sk_buff + offset)
+		 */
 		delta_truesize = skb->truesize -
-				 SKB_TRUESIZE(skb_end_offset(skb));
+				 SKB_TRUESIZE(skb_end_offset(skb) + offset);
 
 		skb->truesize -= skb->data_len;
 		skb->len -= skb->data_len;
@@ -3060,7 +3063,8 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 		memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
 		/* We dont need to clear skbinfo->nr_frags here */
 
-		delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
+		delta_truesize = skb->truesize -
+			SKB_DATA_ALIGN(sizeof(struct sk_buff)) - offset;
 		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
 		goto done;
 	}
Signature · 1 line
1.7.6.5
#1 lore raw

Hello.

Quoted 15 lines · Zhi Yong Wu

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
  net/core/skbuff.c |   10 +++++++---
  1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +0 -0
index 2b6b863..0665bd6 100644

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) · line 3030

Quoted 5 lines
net/core/skbuff.c +1 -1
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
  		frag->page_offset += offset;
  		skb_frag_size_sub(frag, offset);
 
-		/* all fragments truesize : remove (head size + sk_buff) */
+		/**

Why kernel-doc style comment here? Networking code uses the following
comment style:

/* bla
* bla
*/

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) · line 3038

Quoted 9 lines
net/core/skbuff.c +4 -1
+		 * all fragments truesize :
+		 * remove (head size + sk_buff + offset)
+		 */
  		delta_truesize = skb->truesize -
-				 SKB_TRUESIZE(skb_end_offset(skb));
+				 SKB_TRUESIZE(skb_end_offset(skb) + offset);
 
  		skb->truesize -= skb->data_len;
  		skb->len -= skb->data_len;

WBR, Sergei

#2 lore raw
Quoted 35 lines · Sergei Shtylyov

Hello.

Quoted 18 lines · Zhi Yong Wu

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
  net/core/skbuff.c |   10 +++++++---
  1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +0 -0
index 2b6b863..0665bd6 100644

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct · line 3032

Quoted 7 lines
net/core/skbuff.c +1 -1
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct
 sk_buff *skb)
                frag->page_offset += offset;
                skb_frag_size_sub(frag, offset);
 
-               /* all fragments truesize : remove (head size + sk_buff)
 */
+               /**

Why kernel-doc style comment here? Networking code uses the following
comment style:

Quoted 3 lines

/* bla
* bla
*/

No, If you check other comments in networking code, you will see why i did as that.

/**

 *    skb_segment - Perform protocol segmentation on skb.
 *    @skb: buffer to segment
 *    @features: features for the output path (see dev->features)
 *
 *    This function performs segmentation on the given skb.  It returns
 *    a pointer to the first in a list of new skbs for the segments.
 *    In case of error it returns ERR_PTR(err).

*/
struct sk_buff *skb_segment

Quoted 10 lines
Quoted 10 lines
net/core/skbuff.c +4 -1
+                * all fragments truesize :
+                * remove (head size + sk_buff + offset)
+                */
                delta_truesize = skb->truesize -
-                                SKB_TRUESIZE(skb_end_offset(skb));
+                                SKB_TRUESIZE(skb_end_offset(skb) +
 offset);
Quoted 5 lines
Quoted 2 lines
                skb->truesize -= skb->data_len;
                skb->len -= skb->data_len;

WBR, Sergei

Sorry, what is WBR?

Quoted 1 line
Signature · 3 lines
Regards,

Zhi Yong Wu
#3 lore raw

Hello.

Quoted 26 lines · Zhi Yong Wu
Quoted 25 lines
Quoted 15 lines

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
   net/core/skbuff.c |   10 +++++++---
   1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +0 -0
index 2b6b863..0665bd6 100644

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct · line 3032

Quoted 7 lines
net/core/skbuff.c +1 -1
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct
 sk_buff *skb)
                 frag->page_offset += offset;
                 skb_frag_size_sub(frag, offset);
 
-               /* all fragments truesize : remove (head size + sk_buff)
 */
+               /**

Why kernel-doc style comment here? Networking code uses the following
comment style:

Quoted 5 lines
Quoted 3 lines

/* bla
* bla
*/

No, If you check other comments in networking code,

I suggest you do this, and not only comments before functions but also
inside them. I also suggest that you read chapter 8 of Documentation/CodingStyle.

Quoted 2 lines

you will see why i
did as that.

I repeat, this is kernel-doc style comment, that is used to comment on a
functions/structures, and then can be automatically processed to create kernel documentation. Your case does not fit for /** comment.

Quoted 19 lines

/**

  *    skb_segment - Perform protocol segmentation on skb.
  *    @skb: buffer to segment
  *    @features: features for the output path (see dev->features)
  *
  *    This function performs segmentation on the given skb.  It returns
  *    a pointer to the first in a list of new skbs for the segments.
  *    In case of error it returns ERR_PTR(err).

*/
struct sk_buff *skb_segment

Quoted 8 lines
Quoted 8 lines
net/core/skbuff.c +4 -1
+                * all fragments truesize :
+                * remove (head size + sk_buff + offset)
+                */
                 delta_truesize = skb->truesize -
-                                SKB_TRUESIZE(skb_end_offset(skb));
+                                SKB_TRUESIZE(skb_end_offset(skb) +
 offset);
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
                 skb->truesize -= skb->data_len;
                 skb->len -= skb->data_len;

WBR, Sergei

Sorry, what is WBR?

With best regards. Nothing offensive. :-)

WBR, Sergei

#4 lore raw
Quoted 49 lines · Sergei Shtylyov

Hello.

Quoted 31 lines · Zhi Yong Wu
Quoted 29 lines
Quoted 18 lines

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
   net/core/skbuff.c |   10 +++++++---
   1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +0 -0
index 2b6b863..0665bd6 100644

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct · line 3032

Quoted 7 lines
net/core/skbuff.c +1 -1
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct
 sk_buff *skb)
                 frag->page_offset += offset;
                 skb_frag_size_sub(frag, offset);
 
-               /* all fragments truesize : remove (head size + sk_buff)
 */
+               /**

Why kernel-doc style comment here? Networking code uses the following
comment style:

Quoted 6 lines
Quoted 3 lines

/* bla
* bla
*/

No, If you check other comments in networking code,

I suggest you do this, and not only comments before functions but also
inside them. I also suggest that you read chapter 8 of Documentation/CodingStyle.

Quoted 7 lines
Quoted 2 lines

you will see why i
did as that.

I repeat, this is kernel-doc style comment, that is used to comment on a
functions/structures, and then can be automatically processed to create kernel documentation. Your case does not fit for /** comment.

Your explanation is very clear, i accept your advice, thanks.

Quoted 37 lines
Quoted 27 lines

/**

  *    skb_segment - Perform protocol segmentation on skb.
  *    @skb: buffer to segment
  *    @features: features for the output path (see dev->features)
  *
  *    This function performs segmentation on the given skb.  It returns
  *    a pointer to the first in a list of new skbs for the segments.
  *    In case of error it returns ERR_PTR(err).

*/
struct sk_buff *skb_segment

Quoted 13 lines
Quoted 10 lines
net/core/skbuff.c +4 -1
+                * all fragments truesize :
+                * remove (head size + sk_buff + offset)
+                */
                 delta_truesize = skb->truesize -
-                                SKB_TRUESIZE(skb_end_offset(skb));
+                                SKB_TRUESIZE(skb_end_offset(skb) +
 offset);
                 skb->truesize -= skb->data_len;
                 skb->len -= skb->data_len;
Quoted 4 lines
Quoted 1 line

WBR, Sergei

Sorry, what is WBR?

With best regards. Nothing offensive. :-)

WBR, Sergei

Signature · 3 lines
Regards,

Zhi Yong Wu
#5 lore raw
In reply to Zhi Yong Wu
Quoted 30 lines · Zhi Yong Wu

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
 net/core/skbuff.c |   10 +++++++---
 1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +5 -2
index 2b6b863..0665bd6 100644
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 		frag->page_offset += offset;
 		skb_frag_size_sub(frag, offset);
 
-		/* all fragments truesize : remove (head size + sk_buff) */
+		/**
+		 * all fragments truesize :
+		 * remove (head size + sk_buff + offset)
+		 */
 		delta_truesize = skb->truesize -
-				 SKB_TRUESIZE(skb_end_offset(skb));
+				 SKB_TRUESIZE(skb_end_offset(skb) + offset);
 
 		skb->truesize -= skb->data_len;
 		skb->len -= skb->data_len;

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) · line 3070

Quoted 9 lines
net/core/skbuff.c +2 -1
@@ -3060,7 +3063,8 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 		memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
 		/* We dont need to clear skbinfo->nr_frags here */
 
-		delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
+		delta_truesize = skb->truesize -
+			SKB_DATA_ALIGN(sizeof(struct sk_buff)) - offset;
 		NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
 		goto done;
 	}

Nope, this patch is not needed.

truesize is about tracking the memory we allocated, not the memory we really use.

What exact issue do you have ? I suspect a bug in a driver.

#6 lore raw
Quoted 44 lines · Eric Dumazet
Quoted 30 lines · Zhi Yong Wu

From: Zhi Yong Wu <[email protected]>

When the truesize of all fragments was calculated, it didn't take the first small piece of data into account.

---
 net/core/skbuff.c |   10 +++++++---
 1 files changed, 7 insertions(+), 3 deletions(-)
net/core/skbuff.c +5 -2
index 2b6b863..0665bd6 100644
@@ -3027,9 +3027,12 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
              frag->page_offset += offset;
              skb_frag_size_sub(frag, offset);
 
-             /* all fragments truesize : remove (head size + sk_buff) */
+             /**
+              * all fragments truesize :
+              * remove (head size + sk_buff + offset)
+              */
              delta_truesize = skb->truesize -
-                              SKB_TRUESIZE(skb_end_offset(skb));
+                              SKB_TRUESIZE(skb_end_offset(skb) + offset);
 
              skb->truesize -= skb->data_len;
              skb->len -= skb->data_len;

net/core/skbuff.c int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) · line 3070

Quoted 9 lines
net/core/skbuff.c +2 -1
@@ -3060,7 +3063,8 @@ int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
              memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
              /* We dont need to clear skbinfo->nr_frags here */
 
-             delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
+             delta_truesize = skb->truesize -
+                     SKB_DATA_ALIGN(sizeof(struct sk_buff)) - offset;
              NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
              goto done;
      }

Nope, this patch is not needed.

Quoted 2 lines

truesize is about tracking the memory we allocated, not the memory we really use.

wow, i forgot this, thanks for yor reminder.

Quoted 2 lines

What exact issue do you have ? I suspect a bug in a driver.

I didn't hit any issue, but found this when i read gro related code. a but in a driver? sorry, i don't know, look forward to seeing your fix.

Quoted 3 lines
Signature · 3 lines
Regards,

Zhi Yong Wu
#7 lore raw
Quoted 11 lines · Zhi Yong Wu
Quoted 4 lines · Eric Dumazet

Nope, this patch is not needed.

truesize is about tracking the memory we allocated, not the memory we really use.

wow, i forgot this, thanks for yor reminder.

Quoted 2 lines

What exact issue do you have ? I suspect a bug in a driver.

I didn't hit any issue, but found this when i read gro related code. a but in a driver? sorry, i don't know, look forward to seeing your fix.

No, I have no particular issue with this code, I thought you had a problem on a running machine.

If you found this by code review, then everything is fine, and no patch is needed at this point.

Thanks