Skip to contentSkip to conversation

mbuf: fix packet copy

PATCH In review Morten Brørup 5 participants 18 messages Nov 19, 2025 – Feb 26

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

When allocating copies from a mempool using pinned external buffers, the external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +3 -3
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags & ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of newly allocated mbuf */
+	mc->ol_flags |= m->ol_flags & ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
 
 	prev = &mc->next;
 	m_last = mc;
Signature · 1 line
2.43.0
#4 lore raw
In reply to Morten Brørup

PING for review.

Quoted 49 lines

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 19 November 2025 13.04

Requests for copying the at the end of a packet incorrectly returned NULL,
as if copying past the end of a packet.

When allocating copies from a mempool using pinned external buffers, the
external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +1 -1
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 
lib/mbuf/rte_mbuf.c +1 -2
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags &
 ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of

newly allocated mbuf */

lib/mbuf/rte_mbuf.c +1 -0
+	mc->ol_flags |= m->ol_flags &

~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);

lib/mbuf/rte_mbuf.c +0 -1
 	prev = &mc->next;
 	m_last = mc;
--

2.43.0

#5 lore raw
Quoted 31 lines

PING for review.

Quoted 26 lines

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 19 November 2025 13.04

Requests for copying the at the end of a packet incorrectly returned NULL,
as if copying past the end of a packet.

When allocating copies from a mempool using pinned external buffers, the
external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
Quoted 4 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 679

Quoted 4 lines
lib/mbuf/rte_mbuf.c +1 -1
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;

So, when off= m->pkt_len, what do we want it to return? New mbuf with pkt_len == 0?
Any point of such copying then?

Quoted 6 lines
Quoted 6 lines
lib/mbuf/rte_mbuf.c +0 -0
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
Quoted 7 lines
Quoted 7 lines
lib/mbuf/rte_mbuf.c +1 -2
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags &
 ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of

newly allocated mbuf */

lib/mbuf/rte_mbuf.c +1 -0
+	mc->ol_flags |= m->ol_flags &

~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);

Which flags in the new mbuf we want to preserve? AFAIK mbuf_alloc() doesn't set any flags.
BTW, if there are some flags that we would like to preserve, wouldn't that be a change in public API behavior?

Quoted 5 lines
Quoted 5 lines
lib/mbuf/rte_mbuf.c +0 -1
 	prev = &mc->next;
 	m_last = mc;
--

2.43.0

#6 lore raw
Quoted 34 lines

From: Konstantin Ananyev [mailto:[email protected]]

Quoted 7 lines

PING for review.

Quoted 4 lines

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 19 November 2025 13.04

Requests for copying the at the end of a packet incorrectly

returned

Quoted 4 lines
Quoted 4 lines

NULL,
as if copying past the end of a packet.

When allocating copies from a mempool using pinned external

buffers,

Quoted 10 lines
Quoted 10 lines

the
external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
Quoted 5 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m,

Quoted 5 lines
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m,

struct

Quoted 3 lines
Quoted 3 lines
lib/mbuf/rte_mbuf.c +0 -0
 rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
Quoted 8 lines
Quoted 4 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 679

Quoted 4 lines
lib/mbuf/rte_mbuf.c +1 -1
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;

So, when off= m->pkt_len, what do we want it to return? New mbuf with pkt_len == 0?
Any point of such copying then?

Empty packets are perfectly valid.
A library should not optimize them away; it is the application's choice how to handle empty packets.

Quoted 8 lines
Quoted 3 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m,

Quoted 3 lines
lib/mbuf/rte_mbuf.c +0 -0
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m,

struct

Quoted 4 lines
Quoted 4 lines
lib/mbuf/rte_mbuf.c +0 -0
 rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
lib/mbuf/rte_mbuf.c +1 -2
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags &
 ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of

newly allocated mbuf */

lib/mbuf/rte_mbuf.c +1 -0
+	mc->ol_flags |= m->ol_flags &

~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);

Which flags in the new mbuf we want to preserve? AFAIK mbuf_alloc() doesn't set any flags.

Correct: the newly allocated mbuf returned by mbuf_alloc() normally has no flags set.
However, if it is allocated from an mbuf pool holding mbufs using pinned external buffers, the RTE_MBUF_F_EXTERNAL is already set in the mbufs held in that pool, and will remain set when returning the mbuf by mbuf_alloc(). We want to preserve that flag.

Quoted 2 lines

BTW, if there are some flags that we would like to preserve, wouldn't that be a change in public API behavior?

This patch does not change which flags to preserve from the source mbuf.
The patch only adds that the flags of the newly allocated mbuf are not cleared when copying the flags from the source mbuf.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
lib/mbuf/rte_mbuf.c +0 -1
 	prev = &mc->next;
 	m_last = mc;
--

2.43.0

#7 lore raw
Quoted 69 lines
Quoted 30 lines
Quoted 5 lines
Quoted 5 lines

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 19 November 2025 13.04

Requests for copying the at the end of a packet incorrectly

returned

Quoted 4 lines
Quoted 4 lines

NULL,
as if copying past the end of a packet.

When allocating copies from a mempool using pinned external

buffers,

Quoted 10 lines
Quoted 10 lines

the
external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
Quoted 5 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m,

Quoted 5 lines
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m,

struct

Quoted 3 lines
Quoted 3 lines
lib/mbuf/rte_mbuf.c +0 -0
 rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
Quoted 8 lines
Quoted 4 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 679

Quoted 4 lines
lib/mbuf/rte_mbuf.c +1 -1
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;

So, when off= m->pkt_len, what do we want it to return? New mbuf with pkt_len == 0?
Any point of such copying then?

Empty packets are perfectly valid.
A library should not optimize them away; it is the application's choice how to handle empty packets.

Quoted 8 lines
Quoted 3 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m,

Quoted 3 lines
lib/mbuf/rte_mbuf.c +0 -0
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m,

struct

Quoted 4 lines
Quoted 4 lines
lib/mbuf/rte_mbuf.c +0 -0
 rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
lib/mbuf/rte_mbuf.c +1 -2
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags &
 ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of

newly allocated mbuf */

lib/mbuf/rte_mbuf.c +1 -0
+	mc->ol_flags |= m->ol_flags &

~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);

Which flags in the new mbuf we want to preserve? AFAIK mbuf_alloc() doesn't set any flags.

Correct: the newly allocated mbuf returned by mbuf_alloc() normally has no flags set.
However, if it is allocated from an mbuf pool holding mbufs using pinned external buffers, the RTE_MBUF_F_EXTERNAL is already set in the mbufs held in that pool, and will remain set when returning the mbuf by mbuf_alloc(). We want to preserve that flag.

Quoted 6 lines
Quoted 2 lines

BTW, if there are some flags that we would like to preserve, wouldn't that be a change in public API behavior?

This patch does not change which flags to preserve from the source mbuf. The patch only adds that the flags of the newly allocated mbuf are not cleared when copying the flags from the source mbuf.

Makes sense, thanks for clarifying.

#8 lore raw
In reply to Morten Brørup
Quoted 48 lines

PING for review.

Quoted 45 lines

From: Morten Brørup [mailto:[email protected]]
Sent: Wednesday, 19 November 2025 13.04

Requests for copying the at the end of a packet incorrectly returned NULL,
as if copying past the end of a packet.

When allocating copies from a mempool using pinned external buffers, the
external flag was not preserved in these mbufs.

Fixes c3a90c381daa ("mbuf: add a copy routine")
---
 lib/mbuf/rte_mbuf.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +1 -1
index 0d931c7a15..e639aff03e 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 
lib/mbuf/rte_mbuf.c +1 -2
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,8 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct
 rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags &
 ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/* copy flags except indirect and external, and preserve flags of

newly allocated mbuf */

lib/mbuf/rte_mbuf.c +1 -0
+	mc->ol_flags |= m->ol_flags &

~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);

Quoted 3 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,

Quoted 3 lines
lib/mbuf/rte_mbuf.c +0 -1
 	prev = &mc->next;
 	m_last = mc;
--
Quoted 1 line
Quoted 1 line

2.43.0

#9 lore raw

PATCH v2 1/1 mbuf: fix packet copy

In reply to Morten Brørup

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned external buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

---
v2:
* Improved comment about preserving flags for newly allocated mbuf potentially using pinned external buffer.
* Added missing spaces in expression. (Stephen)

---
 lib/mbuf/rte_mbuf.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +7 -3
index 0d931c7a15..a5d16e4c97 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,12 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags & ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/*
+	 * copy flags except indirect and external,
+	 * while preserving flags of newly allocated mbuf
+	 * (specifically RTE_MBUF_F_EXTERNAL if using pinned external buffer)
+	 */
+	mc->ol_flags |= m->ol_flags & ~(RTE_MBUF_F_INDIRECT | RTE_MBUF_F_EXTERNAL);
 
 	prev = &mc->next;
 	m_last = mc;
Signature · 1 line
2.43.0
#10 lore raw
Quoted 26 lines · On Fri, 16 Jan 2026 11:16:21 +0000 Morten Brørup <[email protected]>…

buf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned external buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

---
v2:
* Improved comment about preserving flags for newly allocated mbuf potentially using pinned external buffer.
* Added missing spaces in expression. (Stephen)

---
 lib/mbuf/rte_mbuf.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..a5d16e4c97 100644

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 680

Quoted 7 lines
lib/mbuf/rte_mbuf.c +1 -1
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 

I still think asking for a copy of data that isn't there should return NULL not a zero length mbuf. Kind of academic since I dont think any code uses non-zero offset now.

#11 lore raw
Quoted 37 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Friday, 16 January 2026 18.06

Quoted 3 lines · On Fri, 16 Jan 2026 11:16:21 +0000 Morten Brørup <[email protected]>…

buf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned

NULL,

Quoted 3 lines

as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned

external

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct

Quoted 20 lines

buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

---
v2:
* Improved comment about preserving flags for newly allocated mbuf potentially using pinned external buffer.
* Added missing spaces in expression. (Stephen)

---
 lib/mbuf/rte_mbuf.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..a5d16e4c97 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct

rte_mempool *mp,

Quoted 2 lines
lib/mbuf/rte_mbuf.c +0 -0
 	__rte_mbuf_sanity_check(m, 1);
 
Quoted 11 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 680

Quoted 5 lines
lib/mbuf/rte_mbuf.c +1 -1
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 

I still think asking for a copy of data that isn't there should return NULL
not a zero length mbuf. Kind of academic since I dont think any code uses
non-zero offset now.

Yes, I totally agree it's kind of academic.
But I insist that it is an off-by-one bug, so I fixed it.

Consider the function documentation:

* @param offset

 *   The number of bytes to skip before copying.
 *   If the mbuf does not have that many bytes, it is an error
 *   and NULL is returned.

An offset resulting in copying zero bytes is not an error according to this.

Also consider the comment at the comparison in the source code:
/* check for request to copy at offset past end of mbuf */

It says "past the end", not "at the end"... although I'm not confident enough in my English skills to determine if this means ">=" or ">".

#12 lore raw
Quoted 59 lines · On Fri, 16 Jan 2026 18:16:15 +0100 Morten Brørup <[email protected]>…
Quoted 37 lines

From: Stephen Hemminger [mailto:[email protected]]
Sent: Friday, 16 January 2026 18.06

Quoted 3 lines · On Fri, 16 Jan 2026 11:16:21 +0000 Morten Brørup <[email protected]>…

buf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned

NULL,

Quoted 3 lines

as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned

external

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct

Quoted 20 lines

buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

---
v2:
* Improved comment about preserving flags for newly allocated mbuf potentially using pinned external buffer.
* Added missing spaces in expression. (Stephen)

---
 lib/mbuf/rte_mbuf.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
lib/mbuf/rte_mbuf.c +0 -0
index 0d931c7a15..a5d16e4c97 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct  

rte_mempool *mp,

Quoted 2 lines
lib/mbuf/rte_mbuf.c +0 -0
 	__rte_mbuf_sanity_check(m, 1);
 
Quoted 11 lines

lib/mbuf/rte_mbuf.c rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp, · line 680

Quoted 5 lines
lib/mbuf/rte_mbuf.c +1 -1
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 

I still think asking for a copy of data that isn't there should return NULL
not a zero length mbuf. Kind of academic since I dont think any code uses
non-zero offset now.

Yes, I totally agree it's kind of academic.
But I insist that it is an off-by-one bug, so I fixed it.

Consider the function documentation:

* @param offset

 *   The number of bytes to skip before copying.
 *   If the mbuf does not have that many bytes, it is an error
 *   and NULL is returned.
Quoted 7 lines

An offset resulting in copying zero bytes is not an error according to this.

Also consider the comment at the comparison in the source code:
/* check for request to copy at offset past end of mbuf */

It says "past the end", not "at the end"... although I'm not confident enough in my English skills to determine if this means ">=" or ">".

OK, the documentation does match your change. Maybe there should be a test for that?

#13 lore raw

PATCH v3 1/1 mbuf: fix packet copy

In reply to Morten Brørup

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned external buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

---
v3:
* Improved comment about flags being copied from newly allocated mbuf.
* Added test cases for copying at and past end of packet. (Stephen) v2:
* Improved comment about preserving flags for newly allocated mbuf potentially using pinned external buffer.
* Added missing spaces in expression. (Stephen)

---
 app/test/test_mbuf.c | 20 ++++++++++++++++++++
 lib/mbuf/rte_mbuf.c  | 10 +++++++---
 2 files changed, 27 insertions(+), 3 deletions(-)
app/test/test_mbuf.c +20 -0
index 17be977f31..19b49cec12 100644
@@ -541,6 +541,26 @@ test_pktmbuf_copy(struct rte_mempool *pktmbuf_pool,
 
 	rte_pktmbuf_free(copy2);
 
+	/* test offset copy at the end */
+	copy2 = rte_pktmbuf_copy(copy, pktmbuf_pool,
+				 2 * sizeof(uint32_t), UINT32_MAX);
+	if (copy2 == NULL)
+		GOTO_FAIL("cannot copy at the end of the copy\n");
+
+	if (rte_pktmbuf_pkt_len(copy2) != 0)
+		GOTO_FAIL("copy at the end, length incorrect\n");
+
+	if (rte_pktmbuf_data_len(copy2) != 0)
+		GOTO_FAIL("copy at the end, data length incorrect\n");
+
+	rte_pktmbuf_free(copy2);
+
+	/* test offset copy past the end */
+	copy2 = rte_pktmbuf_copy(copy, pktmbuf_pool,
+				 2 * sizeof(uint32_t) + 1, UINT32_MAX);
+	if (copy2 != NULL)
+		GOTO_FAIL("can copy past the end of the copy\n");
+
 	/* test truncation copy */
 	copy2 = rte_pktmbuf_copy(copy, pktmbuf_pool,
 				 0, sizeof(uint32_t));
lib/mbuf/rte_mbuf.c +7 -3
index 0d931c7a15..a5d16e4c97 100644
@@ -675,7 +675,7 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 	__rte_mbuf_sanity_check(m, 1);
 
 	/* check for request to copy at offset past end of mbuf */
-	if (unlikely(off >= m->pkt_len))
+	if (unlikely(off > m->pkt_len))
 		return NULL;
 
 	mc = rte_pktmbuf_alloc(mp);
@@ -688,8 +688,12 @@ rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
 
 	__rte_pktmbuf_copy_hdr(mc, m);
 
-	/* copied mbuf is not indirect or external */
-	mc->ol_flags = m->ol_flags & ~(RTE_MBUF_F_INDIRECT|RTE_MBUF_F_EXTERNAL);
+	/*
+	 * copy flags except indirect and external,
+	 * while preserving flags of newly allocated mbuf
+	 * (specifically RTE_MBUF_F_EXTERNAL if using pinned external buffer)
+	 */
+	mc->ol_flags |= m->ol_flags & ~(RTE_MBUF_F_INDIRECT | RTE_MBUF_F_EXTERNAL);
 
 	prev = &mc->next;
 	m_last = mc;
Signature · 1 line
2.43.0
#14 lore raw

20/01/2026 08:24, Morten Brørup:

Quoted 5 lines

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

Quoted 7 lines

When allocating the mbuf for the copy from a mempool using pinned external buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

Applied, thanks.

#15 lore raw
Quoted 9 lines · Thomas Monjalon

20/01/2026 08:24, Morten Brørup:

Quoted 8 lines

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned NULL, as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned external buffers, the external flag in this mbuf was not preserved.

Quoted 9 lines
Quoted 4 lines
Fixes c3a90c381daa ("mbuf: add a copy routine")

Applied, thanks.

Hi All. I see this is not marked for stable release. Is it deliberate because of the flags changes ?

#16 lore raw
Quoted 20 lines

From: Kevin Traynor [mailto:[email protected]]
Sent: Wednesday, 25 February 2026 16.59

Quoted 4 lines · Thomas Monjalon

20/01/2026 08:24, Morten Brørup:

Quoted 3 lines

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned

NULL,

Quoted 3 lines
Quoted 3 lines

as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned

external

Quoted 7 lines
Quoted 6 lines

buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")
Quoted 6 lines
Quoted 4 lines

Applied, thanks.

Hi All. I see this is not marked for stable release. Is it deliberate because of the flags changes ?

IIRC, the consensus was that using pinned external buffers is considered exotic, so the bug is unlikely to occur in real applications. And if it did happen, we would have heard about it.

But, since you ask, we really should have left that decision to the stable tree maintainers.

-Morten

#17 lore raw
Quoted 24 lines · Morten Brørup
Quoted 24 lines

From: Kevin Traynor [mailto:[email protected]]
Sent: Wednesday, 25 February 2026 16.59

Quoted 4 lines · Thomas Monjalon

20/01/2026 08:24, Morten Brørup:

Quoted 3 lines

mbuf: fix packet copy

Requests for copying the at the end of a packet incorrectly returned

NULL,

Quoted 3 lines
Quoted 3 lines

as if copying past the end of a packet.

When allocating the mbuf for the copy from a mempool using pinned

external

Quoted 11 lines
Quoted 6 lines

buffers, the external flag in this mbuf was not preserved.

Fixes c3a90c381daa ("mbuf: add a copy routine")

Applied, thanks.

Quoted 7 lines
Quoted 2 lines

Hi All. I see this is not marked for stable release. Is it deliberate because of the flags changes ?

IIRC, the consensus was that using pinned external buffers is considered exotic, so the bug is unlikely to occur in real applications. And if it did happen, we would have heard about it.

But, since you ask, we really should have left that decision to the stable tree maintainers.

Thanks Morten. Background knowledge and context is important, so it's fine you made a decision on it, I just wanted to check.

Quoted 2 lines

-Morten