[dpdk-stable] [PATCH 1/5] mbuf: fix out-of-bounds access at dyn field register

[dpdk-stable] [PATCH 1/5] mbuf: fix out-of-bounds access at dyn field register

[Xiaolong Ye]

We should make sure off + size < sizeof(struct rte_mbuf) to avoid
possible out-of-bounds access of free_space array, there is no issue
currently due to the low bits of free_flags (which is adjacent to
free_space) are always set to 0. But we shouldn't rely on it since it's
fragile and layout of struct mbuf_dyn_shm may be changed in the future.
This patch adds boundary check explicitly to avoid potential risk of
out-of-bounds access.

Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
Cc: stable@dpdk.org

Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
---
 lib/librte_mbuf/rte_mbuf_dyn.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/librte_mbuf/rte_mbuf_dyn.c b/lib/librte_mbuf/rte_mbuf_dyn.c
index 953e3ec31c..13d6da6d16 100644
--- a/lib/librte_mbuf/rte_mbuf_dyn.c
+++ b/lib/librte_mbuf/rte_mbuf_dyn.c
@@ -69,7 +69,8 @@ process_score(void)
 
 	for (off = 0; off < sizeof(struct rte_mbuf); off++) {
 		/* get the size of the free zone */
-		for (size = 0; shm->free_space[off + size]; size++)
+		for (size = 0; (off + size) < sizeof(struct rte_mbuf) &&
+			     shm->free_space[off + size]; size++)
 			;
 		if (size == 0)
 			continue;
-- 
2.17.1

[dpdk-stable] [PATCH 2/5] mbuf: fix missing errno for dyn field/flag registration

[Xiaolong Ye]

Set rte_errno as ENOMEM when allocation failure.

Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
Cc: stable@dpdk.org

Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
---
 lib/librte_mbuf/rte_mbuf_dyn.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/lib/librte_mbuf/rte_mbuf_dyn.c b/lib/librte_mbuf/rte_mbuf_dyn.c
index 13d6da6d16..de7d2eb9a5 100644
--- a/lib/librte_mbuf/rte_mbuf_dyn.c
+++ b/lib/librte_mbuf/rte_mbuf_dyn.c
@@ -278,12 +278,15 @@ __rte_mbuf_dynfield_register_offset(const struct rte_mbuf_dynfield *params,
 		mbuf_dynfield_tailq.head, mbuf_dynfield_list);
 
 	te = rte_zmalloc("MBUF_DYNFIELD_TAILQ_ENTRY", sizeof(*te), 0);
-	if (te == NULL)
+	if (te == NULL) {
+		rte_errno = ENOMEM;
 		return -1;
+	}
 
 	mbuf_dynfield = rte_zmalloc("mbuf_dynfield", sizeof(*mbuf_dynfield), 0);
 	if (mbuf_dynfield == NULL) {
 		rte_free(te);
+		rte_errno = ENOMEM;
 		return -1;
 	}
 
@@ -456,12 +459,15 @@ __rte_mbuf_dynflag_register_bitnum(const struct rte_mbuf_dynflag *params,
 		mbuf_dynflag_tailq.head, mbuf_dynflag_list);
 
 	te = rte_zmalloc("MBUF_DYNFLAG_TAILQ_ENTRY", sizeof(*te), 0);
-	if (te == NULL)
+	if (te == NULL) {
+		rte_errno = ENOMEM;
 		return -1;
+	}
 
 	mbuf_dynflag = rte_zmalloc("mbuf_dynflag", sizeof(*mbuf_dynflag), 0);
 	if (mbuf_dynflag == NULL) {
 		rte_free(te);
+		rte_errno = ENOMEM;
 		return -1;
 	}
 
-- 
2.17.1

[dpdk-stable] [PATCH 3/5] mbuf: fix free_space setting for dynamic field

[Xiaolong Ye]

The value free_space[i] is used to save the size of biggest aligned
element that can fit in the zone, current implementation has one flaw,
for example, if user registers dynfield1 (size = 4, align = 4, req = 124)
first, the free_space would be as below after registration:

  0070: 08 08 08 08 08 08 08 08
  0078: 08 08 08 08 00 00 00 00

Then if user continues to register dynfield2 (size = 4, align = 4),
free_space would become:

  0070: 00 00 00 00 04 04 04 04
  0078: 04 04 04 04 00 00 00 00

Further request dynfield3 (size = 8, align = 8) would fail to register
due to alignment requirement can't be satisfied, though there is enough
space remained in mbuf.

This patch fixes above issue by saving alignment only in aligned zone,
after the fix, above registrations order can be satisfied, free_space
would be like:

After dynfield1 registration:

  0070: 08 08 08 08 08 08 08 08
  0078: 04 04 04 04 00 00 00 00

After dynfield2 registration:

  0070: 08 08 08 08 08 08 08 08
  0078: 00 00 00 00 00 00 00 00

After dynfield3 registration:

  0070: 00 00 00 00 00 00 00 00
  0078: 00 00 00 00 00 00 00 00

This patch also reduces iterations in process_score() by jumping align
steps in each loop.

Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
Cc: stable@dpdk.org

Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
---
 lib/librte_mbuf/rte_mbuf_dyn.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/lib/librte_mbuf/rte_mbuf_dyn.c b/lib/librte_mbuf/rte_mbuf_dyn.c
index de7d2eb9a5..fd51e1b68e 100644
--- a/lib/librte_mbuf/rte_mbuf_dyn.c
+++ b/lib/librte_mbuf/rte_mbuf_dyn.c
@@ -67,13 +67,16 @@ process_score(void)
 			shm->free_space[i] = 1;
 	}
 
-	for (off = 0; off < sizeof(struct rte_mbuf); off++) {
+	off = 0;
+	while (off < sizeof(struct rte_mbuf)) {
 		/* get the size of the free zone */
 		for (size = 0; (off + size) < sizeof(struct rte_mbuf) &&
 			     shm->free_space[off + size]; size++)
 			;
-		if (size == 0)
+		if (size == 0) {
+			off++;
 			continue;
+		}
 
 		/* get the alignment of biggest object that can fit in
 		 * the zone at this offset.
@@ -84,8 +87,10 @@ process_score(void)
 			;
 
 		/* save it in free_space[] */
-		for (i = off; i < off + size; i++)
+		for (i = off; i < off + align; i++)
 			shm->free_space[i] = RTE_MAX(align, shm->free_space[i]);
+
+		off += align;
 	}
 }
 
-- 
2.17.1

[dpdk-stable] [PATCH 4/5] mbuf: fix a dynamic field dump log

[Xiaolong Ye]

For each mbuf byte, free_space[i] == 0 means the space is occupied,
free_space[i] != 0 means space is free.

Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
Cc: stable@dpdk.org

Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>
---
 lib/librte_mbuf/rte_mbuf_dyn.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/librte_mbuf/rte_mbuf_dyn.c b/lib/librte_mbuf/rte_mbuf_dyn.c
index fd51e1b68e..f071651acf 100644
--- a/lib/librte_mbuf/rte_mbuf_dyn.c
+++ b/lib/librte_mbuf/rte_mbuf_dyn.c
@@ -552,7 +552,7 @@ void rte_mbuf_dyn_dump(FILE *out)
 			dynflag->params.name, dynflag->bitnum,
 			dynflag->params.flags);
 	}
-	fprintf(out, "Free space in mbuf (0 = free, value = zone alignment):\n");
+	fprintf(out, "Free space in mbuf (0 = occupied, value = free zone alignment):\n");
 	for (i = 0; i < sizeof(struct rte_mbuf); i++) {
 		if ((i % 8) == 0)
 			fprintf(out, "  %4.4zx: ", i);
-- 
2.17.1

Re: [dpdk-stable] [PATCH 2/5] mbuf: fix missing errno for dyn field/flag registration

[Olivier Matz]

On Sat, Jun 13, 2020 at 11:49:18PM +0800, Xiaolong Ye wrote:
> Set rte_errno as ENOMEM when allocation failure.
> 
> Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>

Good catch, I was wrongly expecting that rte_malloc() functions sets
rte_errno on error.

I wonder if the malloc API shouldn't be modified, because I can see
several places where the same kind of fix would be needed: rte_efd,
rte_cuckoo_hash, rte_fbk_hash, rte_member, ...

Anatoly, what do you think?

Acked-by: Olivier Matz <olivier.matz@6wind.com>

Re: [dpdk-stable] [PATCH 3/5] mbuf: fix free_space setting for dynamic field

[Olivier Matz]

On Sat, Jun 13, 2020 at 11:49:19PM +0800, Xiaolong Ye wrote:
> The value free_space[i] is used to save the size of biggest aligned
> element that can fit in the zone, current implementation has one flaw,
> for example, if user registers dynfield1 (size = 4, align = 4, req = 124)
> first, the free_space would be as below after registration:
> 
>   0070: 08 08 08 08 08 08 08 08
>   0078: 08 08 08 08 00 00 00 00
> 
> Then if user continues to register dynfield2 (size = 4, align = 4),
> free_space would become:
> 
>   0070: 00 00 00 00 04 04 04 04
>   0078: 04 04 04 04 00 00 00 00
> 
> Further request dynfield3 (size = 8, align = 8) would fail to register
> due to alignment requirement can't be satisfied, though there is enough
> space remained in mbuf.
> 
> This patch fixes above issue by saving alignment only in aligned zone,
> after the fix, above registrations order can be satisfied, free_space
> would be like:
> 
> After dynfield1 registration:
> 
>   0070: 08 08 08 08 08 08 08 08
>   0078: 04 04 04 04 00 00 00 00
> 
> After dynfield2 registration:
> 
>   0070: 08 08 08 08 08 08 08 08
>   0078: 00 00 00 00 00 00 00 00
> 
> After dynfield3 registration:
> 
>   0070: 00 00 00 00 00 00 00 00
>   0078: 00 00 00 00 00 00 00 00
> 
> This patch also reduces iterations in process_score() by jumping align
> steps in each loop.
> 
> Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>

Acked-by: Olivier Matz <olivier.matz@6wind.com>

Thanks!

Re: [dpdk-stable] [PATCH 4/5] mbuf: fix a dynamic field dump log

[Olivier Matz]

On Sat, Jun 13, 2020 at 11:49:20PM +0800, Xiaolong Ye wrote:
> For each mbuf byte, free_space[i] == 0 means the space is occupied,
> free_space[i] != 0 means space is free.
> 
> Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com>

Acked-by: Olivier Matz <olivier.matz@6wind.com>