[PATCH bpf 1/2] bpf: fix alignment of netns_dev/netns_ino fields in bpf_{map,prog}_info

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Signed-off-by: Eugene Syromiatnikov <redacted>
---
 include/uapi/linux/bpf.h       | 8 ++++----
 tools/include/uapi/linux/bpf.h | 8 ++++----
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

-- 
2.1.4

On Sun, May 27, 2018 at 4:28 AM, Eugene Syromiatnikov [off-list ref] wrote:

Shall we add a __u32 padding variable before netns_dev? We can use it
for in the future.

Thanks,
Song

On Sun, May 27, 2018 at 01:28:42PM +0200, Eugene Syromiatnikov wrote:

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Reviewed-by: "Dmitry V. Levin" <redacted>
Cc: <redacted> # v4.16+

Thanks,


-- 
ldv

Hi,

Looks like the ABI bug in bpf_map_info and bpf_prog info introduced
in 4.16 is going to slip into 4.17, causing extra pain to 32-bit
userspace.  I'm adding Linus to this thread in hope it might help
to get a fix applied before 4.17 is released.

On Wed, May 30, 2018 at 09:18:58PM +0300, Dmitry V. Levin wrote:

On Sun, May 27, 2018 at 01:28:42PM +0200, Eugene Syromiatnikov wrote:

quoted

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Signed-off-by: Eugene Syromiatnikov <redacted>

Reviewed-by: "Dmitry V. Levin" <redacted>
Cc: <redacted> # v4.16+

Thanks,

quoted

---
 include/uapi/linux/bpf.h       | 8 ++++----
 tools/include/uapi/linux/bpf.h | 8 ++++----
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

-- 
ldv

On Fri, Jun 01, 2018 at 06:12:10AM +0300, Dmitry V. Levin wrote:

Hi,

Looks like the ABI bug in bpf_map_info and bpf_prog info introduced
in 4.16 is going to slip into 4.17, causing extra pain to 32-bit
userspace.  I'm adding Linus to this thread in hope it might help
to get a fix applied before 4.17 is released.

The issue identified in patch 1 is valid.
These two fields won't be properly accessible by 32-bit user space
on 64-bit kernel.
But the fix in patch 1 is wrong.
The patch 2 is completely unnecessary.
Everyone can also see that the patches are still marked as 'new' in patchworks
meaning that we didn't process them.
At the moment many networking folks are traveling to netconf
and we only had a chance to discuss this set last night.
We'll try to send a fix in coming days.
But regardless whether 4.17 is realesed this sunday or not
we're not going to rush wrong patch in without proper code review
and discussion.
That future patch either will land in 4.17 (if it's dealyed into next sunday)
or it will be sent to stable.

To speed up the situation next time please report the issue that you find
to public netdev mailing list instead of using proprietary distro emails.

Thanks.

On Wed, May 30, 2018 at 09:18:58PM +0300, Dmitry V. Levin wrote:

quoted

On Sun, May 27, 2018 at 01:28:42PM +0200, Eugene Syromiatnikov wrote:

quoted

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Signed-off-by: Eugene Syromiatnikov <redacted>

Reviewed-by: "Dmitry V. Levin" <redacted>
Cc: <redacted> # v4.16+

Thanks,

quoted

---
 include/uapi/linux/bpf.h       | 8 ++++----
 tools/include/uapi/linux/bpf.h | 8 ++++----
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

-- 
ldv

On Fri, Jun 01, 2018 at 04:41:16AM -0400, Alexei Starovoitov wrote:

On Fri, Jun 01, 2018 at 06:12:10AM +0300, Dmitry V. Levin wrote:

quoted

Hi,

Looks like the ABI bug in bpf_map_info and bpf_prog info introduced
in 4.16 is going to slip into 4.17, causing extra pain to 32-bit
userspace.  I'm adding Linus to this thread in hope it might help
to get a fix applied before 4.17 is released.

The issue identified in patch 1 is valid.
These two fields won't be properly accessible by 32-bit user space
on 64-bit kernel.

Yes, and currently there is no way to build a 32-bit userspace that would
work properly both with 32-bit and 64-bit kernels.

But the fix in patch 1 is wrong.

Please elaborate.

The patch 2 is completely unnecessary.

The patch 2 doesn't have to be backported to 4.16, but if it was
implemented in 4.16, there would be no bug to discuss.  That is,
the patch 2 is a good policy that would help to avoid this class of bugs
in the future.

Alexei, looks like your Acked-by on the buggy commit
52775b33bb5072fbc07b02c0cf4fe8da1f7ee7cd affects your judgement.
If this is the case, please refer patch 2 to other people who are less biased.

Everyone can also see that the patches are still marked as 'new' in patchworks
meaning that we didn't process them.
At the moment many networking folks are traveling to netconf
and we only had a chance to discuss this set last night.
We'll try to send a fix in coming days.
But regardless whether 4.17 is realesed this sunday or not
we're not going to rush wrong patch in without proper code review
and discussion.
That future patch either will land in 4.17 (if it's dealyed into next sunday)
or it will be sent to stable.

Note that the fix was submitted to netdev on 2018-05-27.
One week is surely enough to make this bug fixed, isn't it?

To speed up the situation next time please report the issue that you find
to public netdev mailing list instead of using proprietary distro emails.

Please explain to me and to the public what do you mean by making this
statement.

I do believe strace developers are free to discuss among themselves
using whatever means of communication they find appropriate.

I do believe the issue was properly reported to netdev, see
https://lkml.kernel.org/r/20180527112835.GA9118@asgard.redhat.com
https://lkml.kernel.org/r/20180527112842.GA18204@asgard.redhat.com

I'd like to use this opportunity to thank Eugene for submitting the fix
the same day the issue was identified as a kernel bug.

Alexei, please do not exclude my email address from this discussion.
Thanks.

quoted

On Wed, May 30, 2018 at 09:18:58PM +0300, Dmitry V. Levin wrote:

quoted

On Sun, May 27, 2018 at 01:28:42PM +0200, Eugene Syromiatnikov wrote:

quoted

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Signed-off-by: Eugene Syromiatnikov <redacted>

Reviewed-by: "Dmitry V. Levin" <redacted>
Cc: <redacted> # v4.16+

Thanks,

quoted

---
 include/uapi/linux/bpf.h       | 8 ++++----
 tools/include/uapi/linux/bpf.h | 8 ++++----
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
 	__aligned_u64 map_ids;
 	char name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {

@@ -1030,8 +1030,8 @@ struct bpf_map_info {
 	__u32 map_flags;
 	char  name[BPF_OBJ_NAME_LEN];
 	__u32 ifindex;
-	__u64 netns_dev;
-	__u64 netns_ino;
+	__aligned_u64 netns_dev;
+	__aligned_u64 netns_ino;
 } __attribute__((aligned(8)));
 
 /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed

-- 
ldv

On 05/29/2018 07:17 PM, Song Liu wrote:

On Sun, May 27, 2018 at 4:28 AM, Eugene Syromiatnikov [off-list ref] wrote:

quoted

Recent introduction of netns_dev/netns_ino to bpf_map_info/bpf_prog info
has broken compat, as offsets of these fields are different in 32-bit
and 64-bit ABIs.  One fix (other than implementing compat support in
syscall in order to handle this discrepancy) is to use __aligned_u64
instead of __u64 for these fields.

Reported-by: Dmitry V. Levin <redacted>
Fixes: 52775b33bb507 ("bpf: offload: report device information about
offloaded maps")
Fixes: 675fc275a3a2d ("bpf: offload: report device information for
offloaded programs")

Signed-off-by: Eugene Syromiatnikov <redacted>
---
 include/uapi/linux/bpf.h       | 8 ++++----
 tools/include/uapi/linux/bpf.h | 8 ++++----
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index c5ec897..903010a 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -1017,8 +1017,8 @@ struct bpf_prog_info {
        __aligned_u64 map_ids;
        char name[BPF_OBJ_NAME_LEN];
        __u32 ifindex;
-       __u64 netns_dev;
-       __u64 netns_ino;
+       __aligned_u64 netns_dev;
+       __aligned_u64 netns_ino;
 } __attribute__((aligned(8)));

Shall we add a __u32 padding variable before netns_dev? We can use it
for in the future.

Agree with Song, and definitely prefer that approach since we already use the hole
as a bitfield in net-next; like this https://patchwork.ozlabs.org/patch/924415/.