Skip to contentSkip to conversation

net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

PATCH In review [email protected] 2 participants 18 messages Jan 12, 2024 – Feb 7, 2024 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

From: Madhuker Mythri <[email protected]>

When multiple queues configured, internally RSS will be enabled and thus TAP BPF RSS byte-code will be loaded on to the Kernel using BPF system calls.

Here, the problem is loading the existing BPF byte-code to the Kernel-5.15 and above versions throws errors, i.e: Kernel BPF verifier not accepted this existing BPF byte-code and system calls return error code "-7" as follows:
------------------------
rss_add_actions(): Failed to load BPF section l3_l4 (7): Argument list too long
------------------------

RCA: These errors started coming after from the Kernel-5.15 version, in which lots of new BPF verification restrictions were added for safe execution of byte-code on to the Kernel, due to which existing BPF program verification does not pass. Here are the major BPF verifier restrictions observed:
1) Need to use new BPF maps structure.
2) Kernel SKB data pointer access not allowed, instead use the BPF helper functions.
3) Undefined loops were not allowed(which are bounded by a variable value).
4) unreachable instructions(like: undefined array access).

After addressing all these Kernel BPF verifier restrictions able to load the BPF byte-code onto the Kernel successfully.

Verified with the latest DPDK-23.11 code, using the python extracted BPF instructions able to load sucessfully on to the Kernel tap device.

Bugzilla Id: 1329

---
 drivers/net/tap/bpf/Makefile          |    6 +-
 drivers/net/tap/bpf/bpf_api.h         |  276 ---
 drivers/net/tap/bpf/bpf_elf.h         |   53 -
 drivers/net/tap/bpf/tap_bpf_program.c |  233 +--
 drivers/net/tap/tap_bpf_api.c         |    5 +-
 drivers/net/tap/tap_bpf_insns.h       | 2300 ++++++++++++++-----------
 6 files changed, 1416 insertions(+), 1457 deletions(-)
 delete mode 100644 drivers/net/tap/bpf/bpf_api.h
 delete mode 100644 drivers/net/tap/bpf/bpf_elf.h
drivers/net/tap/bpf/Makefile +4 -2
index 9efeeb1bc7..218aeaac45 100644
@@ -6,14 +6,16 @@ CLANG=clang
 CLANG_OPTS=-O2
 TARGET=../tap_bpf_insns.h
 
+CLANG_BPF_SYS_INCLUDES ?= $(shell $(CLANG) -v -E - </dev/null 2>&1 \
+	| sed -n '/<...> search starts here:/,/End of search list./{ s| \(/.*\)|-idirafter \1|p }')
+
 all: $(TARGET)
 
 clean:
 	rm tap_bpf_program.o $(TARGET)
 
 tap_bpf_program.o: tap_bpf_program.c
-	$(CLANG) $(CLANG_OPTS) -emit-llvm -c $< -o - | \
-	llc -march=bpf -filetype=obj -o $@
+	$(CLANG) $(CLANG_OPTS) $(CLANG_BPF_SYS_INCLUDES) -target bpf -c $< -g -o $@
 
 $(TARGET): tap_bpf_program.o
 	python3 bpf_extract.py -stap_bpf_program.c -o $@ $<
drivers/net/tap/bpf/bpf_api.h deleted +0 -276

276 lines not on this page · raw message

drivers/net/tap/bpf/bpf_elf.h deleted +0 -53

53 lines not on this page · raw message

drivers/net/tap/bpf/tap_bpf_program.c +78 -155

313 lines not on this page · raw message

drivers/net/tap/tap_bpf_api.c +3 -2
index 15283f8917..7e13d4b8b6 100644
@@ -48,8 +48,9 @@ int tap_flow_bpf_cls_q(__u32 queue_idx)
  */
 int tap_flow_bpf_calc_l3_l4_hash(__u32 key_idx, int map_fd)
 {
-	l3_l4_hash_insns[4].imm = key_idx;
-	l3_l4_hash_insns[9].imm = map_fd;
+	l3_l4_hash_insns[1].imm = key_idx;
+	l3_l4_hash_insns[6].imm = map_fd;
+	l3_l4_hash_insns[6].src_reg = 1;
 
 	return bpf_load(BPF_PROG_TYPE_SCHED_ACT,
 		(struct bpf_insn *)l3_l4_hash_insns,
drivers/net/tap/tap_bpf_insns.h +1331 -969

2,329 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Signature · 1 line
2.39.1.windows.1

Part of the diff is not shown here. View the raw message.

#1 lore raw
Quoted 16 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:

From: Madhuker Mythri <[email protected]>

When multiple queues configured, internally RSS will be enabled and thus TAP BPF RSS byte-code will be loaded on to the Kernel using BPF system calls.

Here, the problem is loading the existing BPF byte-code to the Kernel-5.15 and above versions throws errors, i.e: Kernel BPF verifier not accepted this existing BPF byte-code and system calls return error code "-7" as follows:
------------------------
rss_add_actions(): Failed to load BPF section l3_l4 (7): Argument list too long
------------------------

RCA: These errors started coming after from the Kernel-5.15 version, in which lots of new BPF verification restrictions were added for safe execution of byte-code on to the Kernel, due to which existing BPF program verification does not pass. Here are the major BPF verifier restrictions observed:
1) Need to use new BPF maps structure.
2) Kernel SKB data pointer access not allowed, instead use the BPF helper functions.
3) Undefined loops were not allowed(which are bounded by a variable value).
4) unreachable instructions(like: undefined array access).

Quoted 8 lines

After addressing all these Kernel BPF verifier restrictions able to load the BPF byte-code onto the Kernel successfully.

Verified with the latest DPDK-23.11 code, using the python extracted BPF instructions able to load sucessfully on to the Kernel tap device.

Bugzilla Id: 1329

The DPDK practice is to keep patches as followons of original thread. Please use --msg-id option for reply.
Also put v2 in the mail.

The BPF program doesn't have to meet all the requirements of DPDK checkpatch since it doesn't use EAL, but it should follow conventions. For example:

$ ./devtools/checkpatches.sh ~/Downloads/tap-bpf.mbox 

### [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

WARNING:COMMIT_LOG_LONG_LINE: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#92:
When multiple queues configured, internally RSS will be enabled and thus TAP BPF RSS byte-code will be loaded on to the Kernel using BPF system calls.

WARNING:TYPO_SPELLING: 'sucessfully' may be misspelled - perhaps 'successfully'? #109:
BPF instructions able to load sucessfully on to the Kernel tap device.

                              ^^^^^^^^^^^

...

WARNING:BRACES: braces {} are not necessary for single statement blocks #659: FILE: drivers/net/tap/bpf/tap_bpf_program.c:120:
+ if (rsskey == NULL) {
return TC_ACT_OK;
}

ERROR:ELSE_AFTER_BRACE: else should follow close brace '}' #769: FILE: drivers/net/tap/bpf/tap_bpf_program.c:156:
+ }
+ else {

CHECK:BRACES: Unbalanced braces around else statement #769: FILE: drivers/net/tap/bpf/tap_bpf_program.c:156:
+ else {

ERROR:SPACING: space required before the open parenthesis '(' #790: FILE: drivers/net/tap/bpf/tap_bpf_program.c:168:
+ if(k == hash)

CHECK:BRACES: Blank lines aren't necessary before a close brace '}' #807: FILE: drivers/net/tap/bpf/tap_bpf_program.c:176:

+}

total: 2 errors, 7 warnings, 2 checks, 3314 lines checked

0/1 valid patch

#2 lore raw
In reply to [email protected]

drivers/net/tap/bpf/tap_bpf_program.c rte_softrss_be(const __u32 *input_tuple, const uint8_t *rss_key, · line 170

Quoted 11 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +6 -4
-	queue = rsskey->queues[(hash % rsskey->nb_queues) &
-				       (TAP_MAX_QUEUES - 1)];
-	skb->cb[1] = QUEUE_OFFSET + queue;
-	/* printt(">>>>> rss_l3_l4 hash=0x%x queue=%u\n", hash, queue); */
+	hash = (hash % rsskey->nb_queues) & (TAP_MAX_QUEUES - 1);
+#pragma clang loop unroll(full)
+	for (k = 0; k < TAP_MAX_QUEUES; k++) {
+		if(k == hash)
+			queue = rsskey->queues[k];
+	}

Why introduce another loop here. Original code was fine and bounded.

#3 lore raw

RE: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

With the original code BPF verifier throws error, due to undefined array access and thus could failed to load the BPF instructions onto the Kernel. For safe execution on the Kernel, the array access with such undefined value throws error.

So, came up with this loop which has Max-queues as limit.

Thanks,
Madhuker,

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: 12 January 2024 22:12
To: Madhuker Mythri <[email protected]>
Cc: [email protected]; [email protected]
Subject: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

drivers/net/tap/bpf/tap_bpf_program.c rte_softrss_be(const __u32 *input_tuple, const uint8_t *rss_key, · line 170

Quoted 11 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +6 -4
-	queue = rsskey->queues[(hash % rsskey->nb_queues) &
-				       (TAP_MAX_QUEUES - 1)];
-	skb->cb[1] = QUEUE_OFFSET + queue;
-	/* printt(">>>>> rss_l3_l4 hash=0x%x queue=%u\n", hash, queue); */
+	hash = (hash % rsskey->nb_queues) & (TAP_MAX_QUEUES - 1); #pragma 
+clang loop unroll(full)
+	for (k = 0; k < TAP_MAX_QUEUES; k++) {
+		if(k == hash)
+			queue = rsskey->queues[k];
+	}

Why introduce another loop here. Original code was fine and bounded.

#4 lore raw

Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 7 lines · On Sat, 13 Jan 2024 17:50:19 +0000 Madhuker Mythri…

With the original code BPF verifier throws error, due to undefined array access and thus could failed to load the BPF instructions onto the Kernel. For safe execution on the Kernel, the array access with such undefined value throws error.

So, came up with this loop which has Max-queues as limit.

Thanks,
Madhuker,

It does not for me.

#5 lore raw

Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

In reply to Madhuker Mythri
Quoted 7 lines · On Sat, 13 Jan 2024 17:50:19 +0000 Madhuker Mythri…

With the original code BPF verifier throws error, due to undefined array access and thus could failed to load the BPF instructions onto the Kernel. For safe execution on the Kernel, the array access with such undefined value throws error.

So, came up with this loop which has Max-queues as limit.

Thanks,
Madhuker,

The result of x & 15 is always going to be less than 16.
The Oracle BPF verifier modifications are junk if it can figure that out.

#7 lore raw

RE: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

That's true, SKB pointer access denied and throws error. So, I have to replace with this BPF helper functions.

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: 12 January 2024 22:12
To: Madhuker Mythri <[email protected]>
Cc: [email protected]; [email protected]
Subject: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 1 line · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:

2) Kernel SKB data pointer access not allowed, instead use the BPF helper functions.

This is not true, many of the current kernel BPF examples uses skb pointer.

#8 lore raw

Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 2 lines · On Sat, 13 Jan 2024 17:52:34 +0000 Madhuker Mythri…

That's true, SKB pointer access denied and throws error. So, I have to replace with this BPF helper functions.

It does not for me using current Debian kernel.
You may have enabled something else.

#9 lore raw

RE: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Hi Stephen,

Yes, I found that the Kernel we use has more Verification added for safe execution:
https://blogs.oracle.com/linux/post/bpf-in-depth-the-bpf-bytecode-and-the-bpf-verifier

Here they mentioned about SKB pointer direct access, math on pointer and array with undefined access is denied on BPF code.

Thanks,
Madhuker.

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: 13 January 2024 23:36
To: Madhuker Mythri <[email protected]>
Cc: [email protected]; [email protected]
Subject: Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 2 lines · On Sat, 13 Jan 2024 17:52:34 +0000 Madhuker Mythri…

That's true, SKB pointer access denied and throws error. So, I have to replace with this BPF helper functions.

It does not for me using current Debian kernel. You may have enabled something else.

#10 lore raw

Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 9 lines · On Tue, 16 Jan 2024 05:43:00 +0000 Madhuker Mythri…

Hi Stephen,

Yes, I found that the Kernel we use has more Verification added for safe execution:
https://blogs.oracle.com/linux/post/bpf-in-depth-the-bpf-bytecode-and-the-bpf-verifier

Here they mentioned about SKB pointer direct access, math on pointer and array with undefined access is denied on BPF code.

Thanks,
Madhuker.

Most of your changes, make sense and should be merged.
But the custom verifier changes do not need to be added.
The DPDK upstream project is not targeted at supporting custom vendor API modifications.

#11 lore raw

Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

In reply to Madhuker Mythri
Quoted 6 lines · On Tue, 16 Jan 2024 05:43:00 +0000 Madhuker Mythri…

Hi Stephen,

Yes, I found that the Kernel we use has more Verification added for safe execution:
https://blogs.oracle.com/linux/post/bpf-in-depth-the-bpf-bytecode-and-the-bpf-verifier

Here they mentioned about SKB pointer direct access, math on pointer and array with undefined access is denied on BPF code.

Have you tried running Cilium? I would expect those custom changes would break that

#12 lore raw

RE: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

No, I did not run with Cilium.

Thanks,
Madhuker.

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: 17 January 2024 08:56
To: Madhuker Mythri <[email protected]>
Cc: [email protected]; [email protected]
Subject: Re: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Quoted 6 lines · On Tue, 16 Jan 2024 05:43:00 +0000 Madhuker Mythri…

Hi Stephen,

Yes, I found that the Kernel we use has more Verification added for safe execution:
https://blogs.oracle.com/linux/post/bpf-in-depth-the-bpf-bytecode-and-the-bpf-verifier

Here they mentioned about SKB pointer direct access, math on pointer and array with undefined access is denied on BPF code.

Have you tried running Cilium? I would expect those custom changes would break that

#13 lore raw
In reply to [email protected]

drivers/net/tap/bpf/tap_bpf_program.c rte_softrss_be(const __u32 *input_tuple, const uint8_t *rss_key, · line 126

Quoted 3 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +3 -0
+	if (bpf_skb_load_bytes_relative(skb, 0, &nh, sizeof(nh), BPF_HDR_START_NET))
+		return TC_ACT_OK;
+	if (nh.iph.version == 4) {

Using IP header bytes, instead of ether header protocol (in skb->proto) is changing functionality and wrong. You don't want to be seeing some other protocol (not IPv4 or IPv6) that just happened to have the 4 in that nibble.

Original code was fine.

#14 lore raw
In reply to [email protected]

drivers/net/tap/bpf/tap_bpf_program.c rte_softrss_be(const __u32 *input_tuple, const uint8_t *rss_key, · line 129

Quoted 11 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +2 -8
 		struct ipv4_l3_l4_tuple v4_tuple = {
-			.src_addr = IPv4(*(src_dst_addr + 0),
-					*(src_dst_addr + 1),
-					*(src_dst_addr + 2),
-					*(src_dst_addr + 3)),
-			.dst_addr = IPv4(*(src_dst_addr + 4),
-					*(src_dst_addr + 5),
-					*(src_dst_addr + 6),
-					*(src_dst_addr + 7)),
+			.src_addr = bpf_ntohl(nh.iph.saddr),
+			.dst_addr = bpf_ntohl(nh.iph.daddr),

Not sure this is correct.
The RSS calculation (Toeplitz) expects that data is in network byte order. See app/test/test_thash.c for example.

#15 lore raw

RE: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

Even, in the original code the IP-address converted to host-order from network byte order. Now, I just replaced with BPF API's to convert to host-order.

-----Original Message-----
From: Stephen Hemminger <[email protected]>
Sent: 12 January 2024 22:17
To: Madhuker Mythri <[email protected]>
Cc: [email protected]; [email protected]
Subject: [External] : Re: [PATCH] net/tap: Modified TAP BPF program as per the Kernel-version upgrade requirements.

drivers/net/tap/bpf/tap_bpf_program.c rte_softrss_be(const __u32 *input_tuple, const uint8_t *rss_key, · line 129

Quoted 11 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +2 -8
 		struct ipv4_l3_l4_tuple v4_tuple = {
-			.src_addr = IPv4(*(src_dst_addr + 0),
-					*(src_dst_addr + 1),
-					*(src_dst_addr + 2),
-					*(src_dst_addr + 3)),
-			.dst_addr = IPv4(*(src_dst_addr + 4),
-					*(src_dst_addr + 5),
-					*(src_dst_addr + 6),
-					*(src_dst_addr + 7)),
+			.src_addr = bpf_ntohl(nh.iph.saddr),
+			.dst_addr = bpf_ntohl(nh.iph.daddr),

Not sure this is correct.
The RSS calculation (Toeplitz) expects that data is in network byte order. See app/test/test_thash.c for example.

#16 lore raw

RFC v3 1/1 tap: rework the BPF header parsing

In reply to [email protected]

Cleanup the header parsing.
- bpf load helpers
- split IPv4 and IPv6 into seperate functions
- drop incorrect VLAN header parsing
- simplify the check for Ipv4 fragments
- support IPv4 options
- support IPv6 extension headers
- fix the handling of L3 only hash
- allow using non-default key
- use standard network header structs
- use temporary variables to avoid so many casts
- stick to kernel data types (ie __u32)
- fix support for non-default key

- use new BPF map format based off of BTF
- use system BPF headers and helpers

This is based of the BPF map work in earlier patch from Madhuker Mythri <[email protected]>

---
This is still work in progress.
Needs more testing.
Not sure it is worth splitting into smaller patches.

 drivers/net/tap/bpf/Makefile          |   14 +-
 drivers/net/tap/bpf/bpf_api.h         |  276 ---
 drivers/net/tap/bpf/bpf_elf.h         |   53 -
 drivers/net/tap/bpf/tap_bpf_program.c |  346 ++-
 drivers/net/tap/tap_bpf_api.c         |    5 +-
 drivers/net/tap/tap_bpf_insns.h       | 3288 +++++++++++++------------
 drivers/net/tap/tap_flow.c            |   43 +
 7 files changed, 1920 insertions(+), 2105 deletions(-)
 delete mode 100644 drivers/net/tap/bpf/bpf_api.h
 delete mode 100644 drivers/net/tap/bpf/bpf_elf.h
drivers/net/tap/bpf/Makefile +9 -5
index 9efeeb1bc704..20a51f1ef4e1 100644
@@ -2,9 +2,14 @@
 # This file is not built as part of normal DPDK build.
 # It is used to generate the eBPF code for TAP RSS.
 
-CLANG=clang
-CLANG_OPTS=-O2
-TARGET=../tap_bpf_insns.h
+CLANG = clang
+CLANG_OPTS = -g -O2 -Wall -Wextra -target bpf
+
+TARGET = ../tap_bpf_insns.h
+
+# Get Clang's default includes on this system.
+CLANG_BPF_SYS_INCLUDES ?= $(shell $(CLANG) -v -E - </dev/null 2>&1 \
+	| sed -n '/<...> search starts here:/,/End of search list./{ s| \(/.*\)|-idirafter \1|p }')
 
 all: $(TARGET)
 
@@ -12,8 +17,7 @@ clean:
 	rm tap_bpf_program.o $(TARGET)
 
 tap_bpf_program.o: tap_bpf_program.c
-	$(CLANG) $(CLANG_OPTS) -emit-llvm -c $< -o - | \
-	llc -march=bpf -filetype=obj -o $@
+	$(CLANG) $(CLANG_OPTS) $(CLANG_BPF_SYS_INCLUDES) -c $< -o $@
 
 $(TARGET): tap_bpf_program.o
 	python3 bpf_extract.py -stap_bpf_program.c -o $@ $<
drivers/net/tap/bpf/bpf_api.h deleted +0 -276

276 lines not on this page · raw message

drivers/net/tap/bpf/bpf_elf.h deleted +0 -53

53 lines not on this page · raw message

drivers/net/tap/bpf/tap_bpf_program.c +167 -179

408 lines not on this page · raw message

drivers/net/tap/tap_bpf_api.c +3 -2
index 15283f8917ed..7e13d4b8b6ba 100644
@@ -48,8 +48,9 @@ int tap_flow_bpf_cls_q(__u32 queue_idx)
  */
 int tap_flow_bpf_calc_l3_l4_hash(__u32 key_idx, int map_fd)
 {
-	l3_l4_hash_insns[4].imm = key_idx;
-	l3_l4_hash_insns[9].imm = map_fd;
+	l3_l4_hash_insns[1].imm = key_idx;
+	l3_l4_hash_insns[6].imm = map_fd;
+	l3_l4_hash_insns[6].src_reg = 1;
 
 	return bpf_load(BPF_PROG_TYPE_SCHED_ACT,
 		(struct bpf_insn *)l3_l4_hash_insns,
drivers/net/tap/tap_bpf_insns.h +1698 -1590

2,226 lines not on this page · raw message

drivers/net/tap/tap_flow.c +43 -0

43 lines not on this page · raw message

The rest of this diff is not shown. View the raw message.

Signature · 1 line
2.43.0

Part of the diff is not shown here. View the raw message.

#17 lore raw
In reply to [email protected]

drivers/net/tap/bpf/tap_bpf_program.c line 32

Quoted 7 lines · On Fri, 12 Jan 2024 19:18:21 +0530 [email protected] wrote:
drivers/net/tap/bpf/tap_bpf_program.c +6 -0
+struct  {
+	__uint(type,       BPF_MAP_TYPE_HASH);
+	__type(key,  __u32);
+	__type(value,     struct rss_key);
+	__uint(max_entries,      256);
+} map_keys SEC(".maps");

Overall this patch is a big step forward in getting TAP BPF going again. But using the new BTF maps won't work with how the tap device is loading the BPF program. Getting BTF to work requires more steps and is best done by using libbpf. With this part of your version current kernels will give type mismatch in verifier since the type information for map_keys is not loaded.

See my followon RFC for what libbpf integration looks like. Ends up being a deeper rewrite.