bpf: New approach for BPF MTU handling #370
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Master branch: ea87ae8 Pull request is NOT updated. Failed to apply https://patchwork.kernel.org/project/netdevbpf/list/?series=386911 conflict: |
|
At least one diff in series https://patchwork.kernel.org/project/netdevbpf/list/?series=386911 expired. Closing PR. |
|
Master branch: 450d060 Pull request is NOT updated. Failed to apply https://patchwork.kernel.org/project/netdevbpf/list/?series=388507 conflict: |
|
At least one diff in series https://patchwork.kernel.org/project/netdevbpf/list/?series=388507 expired. Closing PR. |
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 16, 2023
Puranjay Mohan says: ==================== arm32, bpf: add support for cpuv4 insns Changes in V2 -> V3 - Added comments at places where there could be confustion. - In the patch for DIV64, fix the if-else case that would never run. - In the same patch use a single instruction to POP caller saved regs. - Add a patch to change maintainership of ARM32 BPF JIT. Changes in V1 -> V2: - Fix coding style issues. - Don't use tmp variable for src in emit_ldsx_r() as it is redundant. - Optimize emit_ldsx_r() when offset can fit in immediate. Add the support for cpuv4 instructions for ARM32 BPF JIT. 64-bit division was not supported earlier so this series adds 64-bit DIV, SDIV, MOD, SMOD instructions as well. This series needs any one of the patches from [1] to disable zero-extension for BPF_MEMSX to support ldsx. The relevant selftests have passed expect ldsx_insn which needs fentry: Tested on BeagleBone Black (ARMv7-A): [root@alarm del]# echo 1 > /proc/sys/net/core/bpf_jit_enable [root@alarm del]# ./test_progs -a verifier_sdiv,verifier_movsx,verifier_ldsx,verifier_gotol,verifier_bswap #337/1 verifier_bswap/BSWAP, 16:OK #337/2 verifier_bswap/BSWAP, 16 @unpriv:OK #337/3 verifier_bswap/BSWAP, 32:OK #337/4 verifier_bswap/BSWAP, 32 @unpriv:OK #337/5 verifier_bswap/BSWAP, 64:OK #337/6 verifier_bswap/BSWAP, 64 @unpriv:OK #337 verifier_bswap:OK #351/1 verifier_gotol/gotol, small_imm:OK #351/2 verifier_gotol/gotol, small_imm @unpriv:OK #351 verifier_gotol:OK #359/1 verifier_ldsx/LDSX, S8:OK #359/2 verifier_ldsx/LDSX, S8 @unpriv:OK #359/3 verifier_ldsx/LDSX, S16:OK #359/4 verifier_ldsx/LDSX, S16 @unpriv:OK #359/5 verifier_ldsx/LDSX, S32:OK #359/6 verifier_ldsx/LDSX, S32 @unpriv:OK #359/7 verifier_ldsx/LDSX, S8 range checking, privileged:OK #359/8 verifier_ldsx/LDSX, S16 range checking:OK #359/9 verifier_ldsx/LDSX, S16 range checking @unpriv:OK #359/10 verifier_ldsx/LDSX, S32 range checking:OK #359/11 verifier_ldsx/LDSX, S32 range checking @unpriv:OK #359 verifier_ldsx:OK #370/1 verifier_movsx/MOV32SX, S8:OK #370/2 verifier_movsx/MOV32SX, S8 @unpriv:OK #370/3 verifier_movsx/MOV32SX, S16:OK #370/4 verifier_movsx/MOV32SX, S16 @unpriv:OK #370/5 verifier_movsx/MOV64SX, S8:OK #370/6 verifier_movsx/MOV64SX, S8 @unpriv:OK #370/7 verifier_movsx/MOV64SX, S16:OK #370/8 verifier_movsx/MOV64SX, S16 @unpriv:OK #370/9 verifier_movsx/MOV64SX, S32:OK #370/10 verifier_movsx/MOV64SX, S32 @unpriv:OK #370/11 verifier_movsx/MOV32SX, S8, range_check:OK #370/12 verifier_movsx/MOV32SX, S8, range_check @unpriv:OK #370/13 verifier_movsx/MOV32SX, S16, range_check:OK #370/14 verifier_movsx/MOV32SX, S16, range_check @unpriv:OK #370/15 verifier_movsx/MOV32SX, S16, range_check 2:OK #370/16 verifier_movsx/MOV32SX, S16, range_check 2 @unpriv:OK #370/17 verifier_movsx/MOV64SX, S8, range_check:OK #370/18 verifier_movsx/MOV64SX, S8, range_check @unpriv:OK #370/19 verifier_movsx/MOV64SX, S16, range_check:OK #370/20 verifier_movsx/MOV64SX, S16, range_check @unpriv:OK #370/21 verifier_movsx/MOV64SX, S32, range_check:OK #370/22 verifier_movsx/MOV64SX, S32, range_check @unpriv:OK #370/23 verifier_movsx/MOV64SX, S16, R10 Sign Extension:OK #370/24 verifier_movsx/MOV64SX, S16, R10 Sign Extension @unpriv:OK #370 verifier_movsx:OK #382/1 verifier_sdiv/SDIV32, non-zero imm divisor, check 1:OK #382/2 verifier_sdiv/SDIV32, non-zero imm divisor, check 1 @unpriv:OK #382/3 verifier_sdiv/SDIV32, non-zero imm divisor, check 2:OK #382/4 verifier_sdiv/SDIV32, non-zero imm divisor, check 2 @unpriv:OK #382/5 verifier_sdiv/SDIV32, non-zero imm divisor, check 3:OK #382/6 verifier_sdiv/SDIV32, non-zero imm divisor, check 3 @unpriv:OK #382/7 verifier_sdiv/SDIV32, non-zero imm divisor, check 4:OK #382/8 verifier_sdiv/SDIV32, non-zero imm divisor, check 4 @unpriv:OK #382/9 verifier_sdiv/SDIV32, non-zero imm divisor, check 5:OK #382/10 verifier_sdiv/SDIV32, non-zero imm divisor, check 5 @unpriv:OK #382/11 verifier_sdiv/SDIV32, non-zero imm divisor, check 6:OK #382/12 verifier_sdiv/SDIV32, non-zero imm divisor, check 6 @unpriv:OK #382/13 verifier_sdiv/SDIV32, non-zero imm divisor, check 7:OK #382/14 verifier_sdiv/SDIV32, non-zero imm divisor, check 7 @unpriv:OK #382/15 verifier_sdiv/SDIV32, non-zero imm divisor, check 8:OK #382/16 verifier_sdiv/SDIV32, non-zero imm divisor, check 8 @unpriv:OK #382/17 verifier_sdiv/SDIV32, non-zero reg divisor, check 1:OK #382/18 verifier_sdiv/SDIV32, non-zero reg divisor, check 1 @unpriv:OK #382/19 verifier_sdiv/SDIV32, non-zero reg divisor, check 2:OK #382/20 verifier_sdiv/SDIV32, non-zero reg divisor, check 2 @unpriv:OK #382/21 verifier_sdiv/SDIV32, non-zero reg divisor, check 3:OK #382/22 verifier_sdiv/SDIV32, non-zero reg divisor, check 3 @unpriv:OK #382/23 verifier_sdiv/SDIV32, non-zero reg divisor, check 4:OK #382/24 verifier_sdiv/SDIV32, non-zero reg divisor, check 4 @unpriv:OK #382/25 verifier_sdiv/SDIV32, non-zero reg divisor, check 5:OK #382/26 verifier_sdiv/SDIV32, non-zero reg divisor, check 5 @unpriv:OK #382/27 verifier_sdiv/SDIV32, non-zero reg divisor, check 6:OK #382/28 verifier_sdiv/SDIV32, non-zero reg divisor, check 6 @unpriv:OK #382/29 verifier_sdiv/SDIV32, non-zero reg divisor, check 7:OK #382/30 verifier_sdiv/SDIV32, non-zero reg divisor, check 7 @unpriv:OK #382/31 verifier_sdiv/SDIV32, non-zero reg divisor, check 8:OK #382/32 verifier_sdiv/SDIV32, non-zero reg divisor, check 8 @unpriv:OK #382/33 verifier_sdiv/SDIV64, non-zero imm divisor, check 1:OK #382/34 verifier_sdiv/SDIV64, non-zero imm divisor, check 1 @unpriv:OK #382/35 verifier_sdiv/SDIV64, non-zero imm divisor, check 2:OK #382/36 verifier_sdiv/SDIV64, non-zero imm divisor, check 2 @unpriv:OK #382/37 verifier_sdiv/SDIV64, non-zero imm divisor, check 3:OK #382/38 verifier_sdiv/SDIV64, non-zero imm divisor, check 3 @unpriv:OK #382/39 verifier_sdiv/SDIV64, non-zero imm divisor, check 4:OK #382/40 verifier_sdiv/SDIV64, non-zero imm divisor, check 4 @unpriv:OK #382/41 verifier_sdiv/SDIV64, non-zero imm divisor, check 5:OK #382/42 verifier_sdiv/SDIV64, non-zero imm divisor, check 5 @unpriv:OK #382/43 verifier_sdiv/SDIV64, non-zero imm divisor, check 6:OK #382/44 verifier_sdiv/SDIV64, non-zero imm divisor, check 6 @unpriv:OK #382/45 verifier_sdiv/SDIV64, non-zero reg divisor, check 1:OK #382/46 verifier_sdiv/SDIV64, non-zero reg divisor, check 1 @unpriv:OK #382/47 verifier_sdiv/SDIV64, non-zero reg divisor, check 2:OK #382/48 verifier_sdiv/SDIV64, non-zero reg divisor, check 2 @unpriv:OK #382/49 verifier_sdiv/SDIV64, non-zero reg divisor, check 3:OK #382/50 verifier_sdiv/SDIV64, non-zero reg divisor, check 3 @unpriv:OK #382/51 verifier_sdiv/SDIV64, non-zero reg divisor, check 4:OK #382/52 verifier_sdiv/SDIV64, non-zero reg divisor, check 4 @unpriv:OK #382/53 verifier_sdiv/SDIV64, non-zero reg divisor, check 5:OK #382/54 verifier_sdiv/SDIV64, non-zero reg divisor, check 5 @unpriv:OK #382/55 verifier_sdiv/SDIV64, non-zero reg divisor, check 6:OK #382/56 verifier_sdiv/SDIV64, non-zero reg divisor, check 6 @unpriv:OK #382/57 verifier_sdiv/SMOD32, non-zero imm divisor, check 1:OK #382/58 verifier_sdiv/SMOD32, non-zero imm divisor, check 1 @unpriv:OK #382/59 verifier_sdiv/SMOD32, non-zero imm divisor, check 2:OK #382/60 verifier_sdiv/SMOD32, non-zero imm divisor, check 2 @unpriv:OK #382/61 verifier_sdiv/SMOD32, non-zero imm divisor, check 3:OK #382/62 verifier_sdiv/SMOD32, non-zero imm divisor, check 3 @unpriv:OK #382/63 verifier_sdiv/SMOD32, non-zero imm divisor, check 4:OK #382/64 verifier_sdiv/SMOD32, non-zero imm divisor, check 4 @unpriv:OK #382/65 verifier_sdiv/SMOD32, non-zero imm divisor, check 5:OK #382/66 verifier_sdiv/SMOD32, non-zero imm divisor, check 5 @unpriv:OK #382/67 verifier_sdiv/SMOD32, non-zero imm divisor, check 6:OK #382/68 verifier_sdiv/SMOD32, non-zero imm divisor, check 6 @unpriv:OK #382/69 verifier_sdiv/SMOD32, non-zero reg divisor, check 1:OK #382/70 verifier_sdiv/SMOD32, non-zero reg divisor, check 1 @unpriv:OK #382/71 verifier_sdiv/SMOD32, non-zero reg divisor, check 2:OK #382/72 verifier_sdiv/SMOD32, non-zero reg divisor, check 2 @unpriv:OK #382/73 verifier_sdiv/SMOD32, non-zero reg divisor, check 3:OK #382/74 verifier_sdiv/SMOD32, non-zero reg divisor, check 3 @unpriv:OK #382/75 verifier_sdiv/SMOD32, non-zero reg divisor, check 4:OK #382/76 verifier_sdiv/SMOD32, non-zero reg divisor, check 4 @unpriv:OK #382/77 verifier_sdiv/SMOD32, non-zero reg divisor, check 5:OK #382/78 verifier_sdiv/SMOD32, non-zero reg divisor, check 5 @unpriv:OK #382/79 verifier_sdiv/SMOD32, non-zero reg divisor, check 6:OK #382/80 verifier_sdiv/SMOD32, non-zero reg divisor, check 6 @unpriv:OK #382/81 verifier_sdiv/SMOD64, non-zero imm divisor, check 1:OK #382/82 verifier_sdiv/SMOD64, non-zero imm divisor, check 1 @unpriv:OK #382/83 verifier_sdiv/SMOD64, non-zero imm divisor, check 2:OK #382/84 verifier_sdiv/SMOD64, non-zero imm divisor, check 2 @unpriv:OK #382/85 verifier_sdiv/SMOD64, non-zero imm divisor, check 3:OK #382/86 verifier_sdiv/SMOD64, non-zero imm divisor, check 3 @unpriv:OK #382/87 verifier_sdiv/SMOD64, non-zero imm divisor, check 4:OK #382/88 verifier_sdiv/SMOD64, non-zero imm divisor, check 4 @unpriv:OK #382/89 verifier_sdiv/SMOD64, non-zero imm divisor, check 5:OK #382/90 verifier_sdiv/SMOD64, non-zero imm divisor, check 5 @unpriv:OK #382/91 verifier_sdiv/SMOD64, non-zero imm divisor, check 6:OK #382/92 verifier_sdiv/SMOD64, non-zero imm divisor, check 6 @unpriv:OK #382/93 verifier_sdiv/SMOD64, non-zero imm divisor, check 7:OK #382/94 verifier_sdiv/SMOD64, non-zero imm divisor, check 7 @unpriv:OK #382/95 verifier_sdiv/SMOD64, non-zero imm divisor, check 8:OK #382/96 verifier_sdiv/SMOD64, non-zero imm divisor, check 8 @unpriv:OK #382/97 verifier_sdiv/SMOD64, non-zero reg divisor, check 1:OK #382/98 verifier_sdiv/SMOD64, non-zero reg divisor, check 1 @unpriv:OK #382/99 verifier_sdiv/SMOD64, non-zero reg divisor, check 2:OK #382/100 verifier_sdiv/SMOD64, non-zero reg divisor, check 2 @unpriv:OK #382/101 verifier_sdiv/SMOD64, non-zero reg divisor, check 3:OK #382/102 verifier_sdiv/SMOD64, non-zero reg divisor, check 3 @unpriv:OK #382/103 verifier_sdiv/SMOD64, non-zero reg divisor, check 4:OK #382/104 verifier_sdiv/SMOD64, non-zero reg divisor, check 4 @unpriv:OK #382/105 verifier_sdiv/SMOD64, non-zero reg divisor, check 5:OK #382/106 verifier_sdiv/SMOD64, non-zero reg divisor, check 5 @unpriv:OK #382/107 verifier_sdiv/SMOD64, non-zero reg divisor, check 6:OK #382/108 verifier_sdiv/SMOD64, non-zero reg divisor, check 6 @unpriv:OK #382/109 verifier_sdiv/SMOD64, non-zero reg divisor, check 7:OK #382/110 verifier_sdiv/SMOD64, non-zero reg divisor, check 7 @unpriv:OK #382/111 verifier_sdiv/SMOD64, non-zero reg divisor, check 8:OK #382/112 verifier_sdiv/SMOD64, non-zero reg divisor, check 8 @unpriv:OK #382/113 verifier_sdiv/SDIV32, zero divisor:OK #382/114 verifier_sdiv/SDIV32, zero divisor @unpriv:OK #382/115 verifier_sdiv/SDIV64, zero divisor:OK #382/116 verifier_sdiv/SDIV64, zero divisor @unpriv:OK #382/117 verifier_sdiv/SMOD32, zero divisor:OK #382/118 verifier_sdiv/SMOD32, zero divisor @unpriv:OK #382/119 verifier_sdiv/SMOD64, zero divisor:OK #382/120 verifier_sdiv/SMOD64, zero divisor @unpriv:OK #382 verifier_sdiv:OK Summary: 5/163 PASSED, 0 SKIPPED, 0 FAILED As the selftests don't compile for 32-bit architectures without modifications due to long being 32-bit, I have added new tests to lib/test_bpf.c for cpuv4 insns, all are passing: test_bpf: Summary: 1052 PASSED, 0 FAILED, [891/1040 JIT'ed] test_bpf: test_tail_calls: Summary: 10 PASSED, 0 FAILED, [10/10 JIT'ed] test_bpf: test_skb_segment: Summary: 2 PASSED, 0 FAILED [1] https://lore.kernel.org/all/[email protected]/ ==================== Link: https://lore.kernel.org/r/[email protected] Signed-off-by: Alexei Starovoitov <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Aug 29, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets with/without BPF prog. 1. Create socket pairs 2. Send a bunch of data that requires more than 256 pages 3. Read memory_allocated from the 3rd column in /proc/net/protocols 4. Check if unread data is charged to memory_allocated If BPF prog is attached, memory_allocated should not be changed, but we allow a small error (up to 10 pages) in case other processes on the host use some amounts of TCP/UDP memory. At 2., the test actually sends more than 1024 pages because the sysctl net.core.mem_pcpu_rsv is 256 is by default, which means 256 pages are buffered per cpu before reporting to sk->sk_prot->memory_allocated. BUF_SINGLE (1024) * NR_SEND (64) * NR_SOCKETS (64) / 4096 = 1024 pages When I reduced it to 512 pages, the following assertion for the non-isolated case got flaky. ASSERT_GT(memory_allocated[1], memory_allocated[0] + 256, ...) Another contributor to slowness is 150ms sleep to make sure 1 RCU grace period passes because UDP recv queue is destroyed after that. # time ./test_progs -t sk_memcg #370/1 sk_memcg/TCP :OK #370/2 sk_memcg/UDP :OK #370/3 sk_memcg/TCPv6 :OK #370/4 sk_memcg/UDPv6 :OK #370 sk_memcg:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED real 0m1.214s user 0m0.014s sys 0m0.318s Signed-off-by: Kuniyuki Iwashima <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 3, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets with/without BPF prog. 1. Create socket pairs 2. Send a bunch of data that requires more than 256 pages 3. Read memory_allocated from the 3rd column in /proc/net/protocols 4. Check if unread data is charged to memory_allocated If BPF prog is attached, memory_allocated should not be changed, but we allow a small error (up to 10 pages) in case other processes on the host use some amounts of TCP/UDP memory. At 2., the test actually sends more than 1024 pages because the sysctl net.core.mem_pcpu_rsv is 256 is by default, which means 256 pages are buffered per cpu before reporting to sk->sk_prot->memory_allocated. BUF_SINGLE (1024) * NR_SEND (128) * NR_SOCKETS (64) / 4096 = 2048 pages When it's reduced to 1024 pages, the following assertion for the non-isolated case got flaky. ASSERT_GT(memory_allocated[1], memory_allocated[0] + 256, ...) We use kern_sync_rcu() for UDP because UDP recv queue is destroyed after 1 RCU grace period. The test takes ~0.5s on QEMU w/ KVM but takes 2s w/o KVM. # time ./test_progs -t sk_memcg #370/1 sk_memcg/TCP :OK #370/2 sk_memcg/UDP :OK #370/3 sk_memcg/TCPv6 :OK #370/4 sk_memcg/UDPv6 :OK #370 sk_memcg:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED real 0m0.473s user 0m0.009s sys 0m0.201s Signed-off-by: Kuniyuki Iwashima <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 5, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets with/without BPF prog. 1. Create socket pairs 2. Send a bunch of data that requires more than 256 pages 3. Read memory_allocated from the 3rd column in /proc/net/protocols 4. Check if unread data is charged to memory_allocated If BPF prog is attached, memory_allocated should not be changed, but we allow a small error (up to 10 pages) in case other processes on the host use some amounts of TCP/UDP memory. At 2., the test actually sends more than 1024 pages because the sysctl net.core.mem_pcpu_rsv is 256 is by default, which means 256 pages are buffered per cpu before reporting to sk->sk_prot->memory_allocated. BUF_SINGLE (1024) * NR_SEND (128) * NR_SOCKETS (64) / 4096 = 2048 pages When it's reduced to 1024 pages, the following assertion for the non-isolated case got flaky. ASSERT_GT(memory_allocated[1], memory_allocated[0] + 256, ...) We use kern_sync_rcu() for UDP because UDP recv queue is destroyed after 1 RCU grace period. The test takes ~0.5s on QEMU w/ KVM but takes 2s w/o KVM. # time ./test_progs -t sk_memcg #370/1 sk_memcg/TCP :OK #370/2 sk_memcg/UDP :OK #370/3 sk_memcg/TCPv6 :OK #370/4 sk_memcg/UDPv6 :OK #370 sk_memcg:OK Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED real 0m0.473s user 0m0.009s sys 0m0.201s Signed-off-by: Kuniyuki Iwashima <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 8, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets
with/without SK_BPF_MEMCG_SOCK_ISOLATED.
1. Create socket pairs
2. Send a bunch of data that requires more than 1024 pages
3. Read memory_allocated from sk->sk_prot->memory_allocated and
sk->sk_prot->memory_per_cpu_fw_alloc
4. Check if unread data is charged to memory_allocated
If SK_BPF_MEMCG_SOCK_ISOLATED is set, memory_allocated should not
be changed, but we allow a small error (up to 10 pages) in case
other processes on the host use some amounts of TCP/UDP memory.
The amount of allocated pages are buffered to per-cpu variable
{tcp,udp}_memory_per_cpu_fw_alloc up to +/- net.core.mem_pcpu_rsv
before reported to {tcp,udp}_memory_allocated.
At 3., memory_allocated is calculated from the 2 variables twice
at fentry and fexit of socket create function to check if the per-cpu
value is drained during calculation. In that case, 3. is retried.
We use kern_sync_rcu() for UDP because UDP recv queue is destroyed
after RCU grace period.
The test takes ~2s on QEMU (64 CPUs) w/ KVM but takes 6s w/o KVM.
# time ./test_progs -t sk_memcg
#370/1 sk_memcg/TCP :OK
#370/2 sk_memcg/UDP :OK
#370/3 sk_memcg/TCPv6:OK
#370/4 sk_memcg/UDPv6:OK
#370 sk_memcg:OK
Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED
real 0m1.623s
user 0m0.165s
sys 0m0.366s
Signed-off-by: Kuniyuki Iwashima <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 9, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets
with/without SK_MEMCG_EXCLUSIVE, which can be turned on by
net.core.memcg_exclusive or bpf_setsockopt(SK_BPF_MEMCG_EXCLUSIVE).
1. Create socket pairs
2. Send a bunch of data that requires more than 1024 pages
3. Read memory_allocated from sk->sk_prot->memory_allocated and
sk->sk_prot->memory_per_cpu_fw_alloc
4. Check if unread data is charged to memory_allocated
If SK_MEMCG_EXCLUSIVE is set, memory_allocated should not be
changed, but we allow a small error (up to 10 pages) in case
other processes on the host use some amounts of TCP/UDP memory.
The amount of allocated pages are buffered to per-cpu variable
{tcp,udp}_memory_per_cpu_fw_alloc up to +/- net.core.mem_pcpu_rsv
before reported to {tcp,udp}_memory_allocated.
At 3., memory_allocated is calculated from the 2 variables twice
at fentry and fexit of socket create function to check if the per-cpu
value is drained during calculation. In that case, 3. is retried.
We use kern_sync_rcu() for UDP because UDP recv queue is destroyed
after RCU grace period.
The test takes ~2s on QEMU (64 CPUs) w/ KVM but takes 6s w/o KVM.
# time ./test_progs -t sk_memcg
#370/1 sk_memcg/TCP :OK
#370/2 sk_memcg/UDP :OK
#370/3 sk_memcg/TCPv6:OK
#370/4 sk_memcg/UDPv6:OK
#370 sk_memcg:OK
Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED
real 0m1.623s
user 0m0.165s
sys 0m0.366s
Signed-off-by: Kuniyuki Iwashima <[email protected]>
kernel-patches-daemon-bpf bot
pushed a commit
that referenced
this pull request
Sep 10, 2025
The test does the following for IPv4/IPv6 x TCP/UDP sockets
with/without SK_MEMCG_EXCLUSIVE, which can be turned on by
net.core.memcg_exclusive or bpf_setsockopt(SK_BPF_MEMCG_EXCLUSIVE).
1. Create socket pairs
2. Send a bunch of data that requires more than 1024 pages
3. Read memory_allocated from sk->sk_prot->memory_allocated and
sk->sk_prot->memory_per_cpu_fw_alloc
4. Check if unread data is charged to memory_allocated
If SK_MEMCG_EXCLUSIVE is set, memory_allocated should not be
changed, but we allow a small error (up to 10 pages) in case
other processes on the host use some amounts of TCP/UDP memory.
The amount of allocated pages are buffered to per-cpu variable
{tcp,udp}_memory_per_cpu_fw_alloc up to +/- net.core.mem_pcpu_rsv
before reported to {tcp,udp}_memory_allocated.
At 3., memory_allocated is calculated from the 2 variables twice
at fentry and fexit of socket create function to check if the per-cpu
value is drained during calculation. In that case, 3. is retried.
We use kern_sync_rcu() for UDP because UDP recv queue is destroyed
after RCU grace period.
The test takes ~2s on QEMU (64 CPUs) w/ KVM but takes 6s w/o KVM.
# time ./test_progs -t sk_memcg
#370/1 sk_memcg/TCP :OK
#370/2 sk_memcg/UDP :OK
#370/3 sk_memcg/TCPv6:OK
#370/4 sk_memcg/UDPv6:OK
#370 sk_memcg:OK
Summary: 1/4 PASSED, 0 SKIPPED, 0 FAILED
real 0m1.623s
user 0m0.165s
sys 0m0.366s
Signed-off-by: Kuniyuki Iwashima <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Pull request for series with
subject: bpf: New approach for BPF MTU handling
version: 6
url: https://patchwork.kernel.org/project/netdevbpf/list/?series=386911