flow_dissector: reject invalid attach_flags
Subsystems (5)
PATCH 1/2 flow_dissector: reject invalid attach_flags
Using BPF_PROG_ATTACH on a flow dissector program supports neither flags nor target_fd but accepts any value. Return EINVAL if either are non-zero.
--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)
kernel/bpf/net_namespace.c +3 -0
@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret; +if (attr->attach_flags || attr->target_fd) +return -EINVAL ; + type =to_netns_bpf_attach_type (attr->attach_type);if (type <0 )return -EINVAL ;
Signature · 1 line
2.25.1
PATCH 2/2 bpf: sockmap: reject invalid attach_flags
Using BPF_PROG_ATTACH on a sockmap program currently understands no flags, but accepts any value. Return EINVAL if any flags are specified.
--- net/core/sock_map.c | 3 +++ 1 file changed, 3 insertions(+)
net/core/sock_map.c +3 -0
@@ -70,6 +70,9 @@ int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)struct fd f;int ret; +if (attr->attach_flags) +return -EINVAL ; + f =fdget (ufd); map =__bpf_map_get (f);if (IS_ERR (map))
Signature · 1 line
2.25.1
Quoted 15 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 6 lines
kernel/bpf/net_namespace.c +3
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret; +if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +
In theory it makes sense, but how did you test it? test_progs -t flow
fails 5 tests.
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;
Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.
I spent today digging through this, and the issue is actually more annoying than I thought. BPF_PROG_DETACH for sockmap and flow_dissector ignores attach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that set attach_bpf_fd for these to attach points, which tells me that this is
an easy mistake to make.
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap and flow_dissector have the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)
- Returning EINVAL from BPF_PROG_DETACH if attach_bpf_fd is specified leads to a lot of churn in selftests
Is it worth just landing these fixes on bpf or bpf-next without backporting them?
Signature · 4 lines
Lorenz Bauer | Systems Engineer 6th Floor, County Hall/The Riverside Building, SE1 7PB, UK www.cloudflare.com
Quoted 42 lines · Lorenz Bauer
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.I spent today digging through this, and the issue is actually more annoying than I thought.
BPF_PROG_DETACHfor sockmap andflow_dissectorignoresattach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that setattach_bpf_fdfor these to attach points, which tells me that this is
an easy mistake to make.
Quoted 3 lines
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap andflow_dissectorhave the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)
making flow dissector pass prog_fd as cg and lirc is certainly my preference. Especially since tests are passing fd user code is likely doing the same, so breakage is unlikely. Also it wasn't done that long ago, so
we can backport far enough.
It will remove cap_net_admin ugly check in bpf_prog_detach() which is the only exception now in cap model.
Quoted 48 lines · Alexei Starovoitov
Quoted 42 lines · Lorenz Bauer
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.I spent today digging through this, and the issue is actually more annoying than I thought.
BPF_PROG_DETACHfor sockmap andflow_dissectorignoresattach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that setattach_bpf_fdfor these to attach points, which tells me that this is
an easy mistake to make.Quoted 3 lines
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap andflow_dissectorhave the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)
Quoted 6 lines
making flow dissector pass
prog_fdas cg and lirc is certainly my preference. Especially since tests are passing fd user code is likely doing the same, so breakage is unlikely. Also it wasn't done that long ago, so
we can backport far enough.
It will removecap_net_adminugly check inbpf_prog_detach()which is the only exception now in cap model.
SGTM. What about sockmap though? The code for that has been around for ages.
Signature · 4 lines
Lorenz Bauer | Systems Engineer 6th Floor, County Hall/The Riverside Building, SE1 7PB, UK www.cloudflare.com
Quoted 51 lines · Lorenz Bauer
Quoted 48 lines · Alexei Starovoitov
Quoted 45 lines · Lorenz Bauer
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.I spent today digging through this, and the issue is actually more annoying than I thought.
BPF_PROG_DETACHfor sockmap andflow_dissectorignoresattach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that setattach_bpf_fdfor these to attach points, which tells me that this is
an easy mistake to make.Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap andflow_dissectorhave the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)
Quoted 8 lines
Quoted 6 lines
making flow dissector pass
prog_fdas cg and lirc is certainly my preference. Especially since tests are passing fd user code is likely doing the same, so breakage is unlikely. Also it wasn't done that long ago, so
we can backport far enough.
It will removecap_net_adminugly check inbpf_prog_detach()which is the only exception now in cap model.SGTM. What about sockmap though? The code for that has been around for ages.
you mean the second patch that enforces sock_map_get_from_fd doesn't use attach_flags?
I think it didn't break anything, so enforcing is fine.
or the detach part that doesn't use prog_fd ?
I'm not sure what's the best here.
At least from cap perspective it's fine because map_fd is there.
John, wdyt?
John Fastabend
maintainerQuoted 40 lines · Alexei Starovoitov
Quoted 39 lines · Lorenz Bauer
Quoted 36 lines · Alexei Starovoitov
Quoted 34 lines · Lorenz Bauer
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
I spent today digging through this, and the issue is actually more annoying than I thought.
BPF_PROG_DETACHfor sockmap andflow_dissectorignoresattach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that setattach_bpf_fdfor these to attach points, which tells me that this is
an easy mistake to make.
In sockmap case I didn't manage to think what multiple programs of the same type on the same map would look like so we can just remove whatever program is there. Is there a problem with this or is it that we just want the sanity check.
Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap andflow_dissectorhave the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)making flow dissector pass
prog_fdas cg and lirc is certainly my preference. Especially since tests are passing fd user code is likely doing the same, so breakage is unlikely. Also it wasn't done that long ago, so
we can backport far enough.
It will removecap_net_adminugly check inbpf_prog_detach()which is the only exception now in cap model.
Quoted 5 lines
Quoted 1 line
SGTM. What about sockmap though? The code for that has been around for ages.
you mean the second patch that enforces
sock_map_get_from_fddoesn't useattach_flags?
I think it didn't break anything, so enforcing is fine.
I'm ok with enforcing it.
Quoted 6 lines
or the detach part that doesn't use
prog_fd?
I'm not sure what's the best here.
At least from cap perspective it's fine becausemap_fdis there.John, wdyt?
I think we can keep the current detach without the prog_fd as-is. And then add logic so that if the prog_fd is included we check it?
Quoted 75 lines · John Fastabend
Quoted 40 lines · Alexei Starovoitov
Quoted 39 lines · Lorenz Bauer
Quoted 36 lines · Alexei Starovoitov
Quoted 34 lines · Lorenz Bauer
Quoted 20 lines · Alexei Starovoitov
Quoted 18 lines · Lorenz Bauer
Using
BPF_PROG_ATTACHon a flow dissector program supports neither flags nortarget_fdbut accepts any value. Return EINVAL if either are non-zero.Signed-off-by Lorenz Bauer <[email protected]>Fixes b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")--- kernel/bpf/net_namespace.c | 3 +++ 1 file changed, 3 insertions(+)kernel/bpf/net_namespace.c +0
-0@@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog)struct net *net;int ret;Quoted 5 lines
kernel/bpf/net_namespace.c int netns_bpf_prog_attach(const union bpf_attr *attr, struct bpf_prog *prog) · line 197
Quoted 3 lines
kernel/bpf/net_namespace.c +3
-0+if (attr->attach_flags || attr->target_fd) +return -EINVAL ; +In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
Quoted 2 lines
test_progs-t flow
fails 5 tests.Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
I spent today digging through this, and the issue is actually more annoying than I thought.
BPF_PROG_DETACHfor sockmap andflow_dissectorignoresattach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the program being detached is actually what user space expects. We actually have tests that setattach_bpf_fdfor these to attach points, which tells me that this is
an easy mistake to make.In sockmap case I didn't manage to think what multiple programs of the same type on the same map would look like so we can just remove whatever program is there. Is there a problem with this or is it that we just want the sanity check.
Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap andflow_dissectorhave the same semantics as cgroup and lirc2 requires a bunch of changes (probably a new function for sockmap)making flow dissector pass
prog_fdas cg and lirc is certainly my preference. Especially since tests are passing fd user code is likely doing the same, so breakage is unlikely. Also it wasn't done that long ago, so
we can backport far enough.
It will removecap_net_adminugly check inbpf_prog_detach()which is the only exception now in cap model.Quoted 5 lines
Quoted 1 line
SGTM. What about sockmap though? The code for that has been around for ages.
you mean the second patch that enforces
sock_map_get_from_fddoesn't useattach_flags?
I think it didn't break anything, so enforcing is fine.I'm ok with enforcing it.
Quoted 1 line
Quoted 8 lines
Quoted 5 lines
or the detach part that doesn't use
prog_fd?
I'm not sure what's the best here.
At least from cap perspective it's fine becausemap_fdis there.John, wdyt?
I think we can keep the current detach without the
prog_fdas-is. And then add logic so that if theprog_fdis included we check it?
Do you know of users that rely on this? FWIW all of the selftests actually pass attach_bpf_fd when detaching from sockmap (on a recent bpf-next at least).
It'd be nice if I could make sockmap require this to be present, just so that it's consistent with flow_dissector and other BPF_PROG_DETACH users.
OTOH I'm not sure if this is backport material after all.
Lorenz
Signature · 4 lines
Lorenz Bauer | Systems Engineer 6th Floor, County Hall/The Riverside Building, SE1 7PB, UK www.cloudflare.com