Skip to contentSkip to conversation

bpf: allow __sk_buff tstamp in BPF_PROG_TEST_RUN

PATCH Applied Stanislav Fomichev 5 participants 8 messages Oct 15, 2019

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 bpf: allow __sk_buff tstamp in BPF_PROG_TEST_RUN

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

---
 net/bpf/test_run.c | 9 +++++++++
 1 file changed, 9 insertions(+)
net/bpf/test_run.c +9 -0
index 1153bbcdff72..0be4497cb832 100644
@@ -218,10 +218,18 @@ static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
 
 	if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
 			   FIELD_SIZEOF(struct __sk_buff, cb),
+			   offsetof(struct __sk_buff, tstamp)))
+		return -EINVAL;
+
+	/* tstamp is allowed */
+
+	if (!range_is_zero(__skb, offsetof(struct __sk_buff, tstamp) +
+			   FIELD_SIZEOF(struct __sk_buff, tstamp),
 			   sizeof(struct __sk_buff)))
 		return -EINVAL;
 
 	skb->priority = __skb->priority;
+	skb->tstamp = __skb->tstamp;
 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
 
 	return 0;
@@ -235,6 +243,7 @@ static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
 		return;
 
 	__skb->priority = skb->priority;
+	__skb->tstamp = skb->tstamp;
 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
 }
 
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
#1 lore raw

PATCH 2/2 selftests: bpf: add selftest for __sk_buff tstamp

Make sure BPF_PROG_TEST_RUN accepts tstamp and exports any modifications that BPF program does.

---
 tools/testing/selftests/bpf/prog_tests/skb_ctx.c | 5 +++++
 tools/testing/selftests/bpf/progs/test_skb_ctx.c | 1 +
 2 files changed, 6 insertions(+)
tools/testing/selftests/bpf/prog_tests/skb_ctx.c +5 -0
index e95baa32e277..a2eb8db8dafb 100644
@@ -10,6 +10,7 @@ void test_skb_ctx(void)
 		.cb[3] = 4,
 		.cb[4] = 5,
 		.priority = 6,
+		.tstamp = 7,
 	};
 	struct bpf_prog_test_run_attr tattr = {
 		.data_in = &pkt_v4,
@@ -86,4 +87,8 @@ void test_skb_ctx(void)
 		   "ctx_out_priority",
 		   "skb->priority == %d, expected %d\n",
 		   skb.priority, 7);
+	CHECK_ATTR(skb.tstamp != 8,
+		   "ctx_out_tstamp",
+		   "skb->tstamp == %lld, expected %d\n",
+		   skb.tstamp, 8);
 }
tools/testing/selftests/bpf/progs/test_skb_ctx.c +1 -0
index 7a80960d7df1..2a9f4c736ebc 100644
@@ -16,6 +16,7 @@ int process(struct __sk_buff *skb)
 		skb->cb[i]++;
 	}
 	skb->priority++;
+	skb->tstamp++;
 
 	return 0;
 }
Signature · 1 line
2.23.0.700.g56cf767bdb-goog
Acked-by on v1 1/2
#3 lore raw
In reply to Stanislav Fomichev
Quoted 6 lines · Stanislav Fomichev

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

Please provide a cover letter next time. It makes ack-all possible.

#4 lore raw
In reply to Martin Lau
Quoted 8 lines · Martin Lau
Quoted 6 lines · Stanislav Fomichev

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

Please provide a cover letter next time. It makes ack-all possible.

SG, I'll try to add cover letter in the future if that helps.

If I remember correctly, acked-by to the cover letter was not showing up in the patchwork and people usually do it for each patch anyway. That's why I didn't bother to do it for this small change.

Quoted 1 line
#5 lore raw
In reply to Stanislav Fomichev
Quoted 18 lines · Stanislav Fomichev

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

---
 net/bpf/test_run.c | 9 +++++++++
 1 file changed, 9 insertions(+)
net/bpf/test_run.c +0 -0
index 1153bbcdff72..0be4497cb832 100644

net/bpf/test_run.c static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) · line 227

Quoted 10 lines
net/bpf/test_run.c +7 -0
@@ -218,10 +218,18 @@ static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
 
        if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
                           FIELD_SIZEOF(struct __sk_buff, cb),
+                          offsetof(struct __sk_buff, tstamp)))
+               return -EINVAL;
+
+       /* tstamp is allowed */
+
+       if (!range_is_zero(__skb, offsetof(struct __sk_buff, tstamp) +
+                          FIELD_SIZEOF(struct __sk_buff, tstamp),

with no context on this particular change whatsoever: isn't this the same as offsetofend(struct __sk_buff, tstamp)? Same above for cb.

Overall, this seems like the 4th similar check, would it make sense to add a static array of ranges we want to check for zeros and just loop over it?..

Quoted 19 lines
net/bpf/test_run.c +2 -1
                           sizeof(struct __sk_buff)))
                return -EINVAL;
 
        skb->priority = __skb->priority;
+       skb->tstamp = __skb->tstamp;
        memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
 
        return 0;
@@ -235,6 +243,7 @@ static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
                return;
 
        __skb->priority = skb->priority;
+       __skb->tstamp = skb->tstamp;
        memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
 }
 
--

2.23.0.700.g56cf767bdb-goog

#6 lore raw
Quoted 34 lines · Andrii Nakryiko
Quoted 18 lines · Stanislav Fomichev

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

---
 net/bpf/test_run.c | 9 +++++++++
 1 file changed, 9 insertions(+)
net/bpf/test_run.c +0 -0
index 1153bbcdff72..0be4497cb832 100644

net/bpf/test_run.c static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) · line 227

Quoted 10 lines
net/bpf/test_run.c +7 -0
@@ -218,10 +218,18 @@ static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
 
        if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
                           FIELD_SIZEOF(struct __sk_buff, cb),
+                          offsetof(struct __sk_buff, tstamp)))
+               return -EINVAL;
+
+       /* tstamp is allowed */
+
+       if (!range_is_zero(__skb, offsetof(struct __sk_buff, tstamp) +
+                          FIELD_SIZEOF(struct __sk_buff, tstamp),

with no context on this particular change whatsoever: isn't this the same as offsetofend(struct __sk_buff, tstamp)? Same above for cb.

Quoted 3 lines

Overall, this seems like the 4th similar check, would it make sense to add a static array of ranges we want to check for zeros and just loop over it?..

I wouldn't bother, but offsetofend() is a good suggestion that can be done in a followup.

Applied both patches. Thanks

#7 lore raw
Quoted 36 lines · Alexei Starovoitov
Quoted 34 lines · Andrii Nakryiko
Quoted 18 lines · Stanislav Fomichev

It's useful for implementing EDT related tests (set tstamp, run the test, see how the tstamp is changed or observe some other parameter).

Note that bpf_ktime_get_ns() helper is using monotonic clock, so for the BPF programs that compare tstamp against it, tstamp should be derived from clock_gettime(CLOCK_MONOTONIC, ...).

---
 net/bpf/test_run.c | 9 +++++++++
 1 file changed, 9 insertions(+)
net/bpf/test_run.c +0 -0
index 1153bbcdff72..0be4497cb832 100644

net/bpf/test_run.c static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) · line 227

Quoted 10 lines
net/bpf/test_run.c +7 -0
@@ -218,10 +218,18 @@ static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
 
        if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
                           FIELD_SIZEOF(struct __sk_buff, cb),
+                          offsetof(struct __sk_buff, tstamp)))
+               return -EINVAL;
+
+       /* tstamp is allowed */
+
+       if (!range_is_zero(__skb, offsetof(struct __sk_buff, tstamp) +
+                          FIELD_SIZEOF(struct __sk_buff, tstamp),

with no context on this particular change whatsoever: isn't this the same as offsetofend(struct __sk_buff, tstamp)? Same above for cb.

Quoted 8 lines
Quoted 3 lines

Overall, this seems like the 4th similar check, would it make sense to add a static array of ranges we want to check for zeros and just loop over it?..

I wouldn't bother, but offsetofend() is a good suggestion that can be done in a followup.

Applied both patches. Thanks

Thanks. I'll follow up with offsetofend, sounds like a good suggestion that can eliminate a bit of copy paste.