Luiz Capitulino reported the test_verifier test failed: "precise: ST insn causing spi > allocated_stack". And it was introduced by the following upstream commit: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Eduard's investigation [4] shows that test failure is not a bug, but a difference in BPF verifier behavior between upstream, where commits [1,2,3] by Andrii are present, and 5.10, where these commits are absent.
Backporting strategy is consistent with Eduard in kernel version 6.1 [5], but with some conflicts in patch #1, #4 and #6 due to the bpf of 5.10 doesn't support more features. Both test_verifier and test_maps have passed, while test_progs and test_progs-no_alu32 with no new failure ceses.
Commits of Andrii: [1] be2ef8161572 ("bpf: allow precision tracking for programs with subprogs") [2] f63181b6ae79 ("bpf: stop setting precise in current state") [3] 7a830b53c17b ("bpf: aggressively forget precise markings during state checkpointing")
Links: [4] https://lore.kernel.org/stable/c9b10a8a551edafdfec855fbd35757c6238ad258.came... [5] https://lore.kernel.org/all/20230724124223.1176479-2-eddyz87@gmail.com/
Andrii Nakryiko (4): bpf: allow precision tracking for programs with subprogs bpf: stop setting precise in current state bpf: aggressively forget precise markings during state checkpointing selftests/bpf: make test_align selftest more robust
Ilya Leoshkevich (1): selftests/bpf: Fix sk_assign on s390x
Yonghong Song (1): selftests/bpf: Workaround verification failure for fexit_bpf2bpf/func_replace_return_code
kernel/bpf/verifier.c | 175 ++++++++++++++++-- .../testing/selftests/bpf/prog_tests/align.c | 36 ++-- .../selftests/bpf/prog_tests/sk_assign.c | 25 ++- .../selftests/bpf/progs/connect4_prog.c | 2 +- .../selftests/bpf/progs/test_sk_assign.c | 11 ++ .../bpf/progs/test_sk_assign_libbpf.c | 3 + 6 files changed, 219 insertions(+), 33 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
Stop forcing precise=true for SCALAR registers when BPF program has any subprograms. Current restriction means that any BPF program, as soon as it uses subprograms, will end up not getting any of the precision tracking benefits in reduction of number of verified states.
This patch keeps the fallback mark_all_scalars_precise() behavior if precise marking has to cross function frames. E.g., if subprogram requires R1 (first input arg) to be marked precise, ideally we'd need to backtrack to the parent function and keep marking R1 and its dependencies as precise. But right now we give up and force all the SCALARs in any of the current and parent states to be forced to precise=true. We can lift that restriction in the future.
But this patch fixes two issues identified when trying to enable precision tracking for subprogs.
First, prevent "escaping" from top-most state in a global subprog. While with entry-level BPF program we never end up requesting precision for R1-R5 registers, because R2-R5 are not initialized (and so not readable in correct BPF program), and R1 is PTR_TO_CTX, not SCALAR, and so is implicitly precise. With global subprogs, though, it's different, as global subprog a) can have up to 5 SCALAR input arguments, which might get marked as precise=true and b) it is validated in isolation from its main entry BPF program. b) means that we can end up exhausting parent state chain and still not mark all registers in reg_mask as precise, which would lead to verifier bug warning.
To handle that, we need to consider two cases. First, if the very first state is not immediately "checkpointed" (i.e., stored in state lookup hashtable), it will get correct first_insn_idx and last_insn_idx instruction set during state checkpointing. As such, this case is already handled and __mark_chain_precision() already handles that by just doing nothing when we reach to the very first parent state. st->parent will be NULL and we'll just stop. Perhaps some extra check for reg_mask and stack_mask is due here, but this patch doesn't address that issue.
More problematic second case is when global function's initial state is immediately checkpointed before we manage to process the very first instruction. This is happening because when there is a call to global subprog from the main program the very first subprog's instruction is marked as pruning point, so before we manage to process first instruction we have to check and checkpoint state. This patch adds a special handling for such "empty" state, which is identified by having st->last_insn_idx set to -1. In such case, we check that we are indeed validating global subprog, and with some sanity checking we mark input args as precise if requested.
Note that we also initialize state->first_insn_idx with correct start insn_idx offset. For main program zero is correct value, but for any subprog it's quite confusing to not have first_insn_idx set. This doesn't have any functional impact, but helps with debugging and state printing. We also explicitly initialize state->last_insns_idx instead of relying on is_state_visited() to do this with env->prev_insns_idx, which will be -1 on the very first instruction. This concludes necessary changes to handle specifically global subprog's precision tracking.
Second identified problem was missed handling of BPF helper functions that call into subprogs (e.g., bpf_loop and few others). From precision tracking and backtracking logic's standpoint those are effectively calls into subprogs and should be called as BPF_PSEUDO_CALL calls.
This patch takes the least intrusive way and just checks against a short list of current BPF helpers that do call subprogs, encapsulated in is_callback_calling_function() function. But to prevent accidentally forgetting to add new BPF helpers to this "list", we also do a sanity check in __check_func_call, which has to be called for each such special BPF helper, to validate that BPF helper is indeed recognized as callback-calling one. This should catch any missed checks in the future. Adding some special flags to be added in function proto definitions seemed like an overkill in this case.
With the above changes, it's possible to remove forceful setting of reg->precise to true in __mark_reg_unknown, which turns on precision tracking both inside subprogs and entry progs that have subprogs. No warnings or errors were detected across all the selftests, but also when validating with veristat against internal Meta BPF objects and Cilium objects. Further, in some BPF programs there are noticeable reduction in number of states and instructions validated due to more effective precision tracking, especially benefiting syncookie test.
$ ./veristat -C -e file,prog,insns,states ~/baseline-results.csv ~/subprog-precise-results.csv | grep -v '+0' File Program Total insns (A) Total insns (B) Total insns (DIFF) Total states (A) Total states (B) Total states (DIFF) ---------------------------------------- -------------------------- --------------- --------------- ------------------ ---------------- ---------------- ------------------- pyperf600_bpf_loop.bpf.linked1.o on_event 3966 3678 -288 (-7.26%) 306 276 -30 (-9.80%) pyperf_global.bpf.linked1.o on_event 7563 7530 -33 (-0.44%) 520 517 -3 (-0.58%) pyperf_subprogs.bpf.linked1.o on_event 36358 36934 +576 (+1.58%) 2499 2531 +32 (+1.28%) setget_sockopt.bpf.linked1.o skops_sockopt 3965 4038 +73 (+1.84%) 343 347 +4 (+1.17%) test_cls_redirect_subprogs.bpf.linked1.o cls_redirect 64965 64901 -64 (-0.10%) 4619 4612 -7 (-0.15%) test_misc_tcp_hdr_options.bpf.linked1.o misc_estab 1491 1307 -184 (-12.34%) 110 100 -10 (-9.09%) test_pkt_access.bpf.linked1.o test_pkt_access 354 349 -5 (-1.41%) 25 24 -1 (-4.00%) test_sock_fields.bpf.linked1.o egress_read_sock_fields 435 375 -60 (-13.79%) 22 20 -2 (-9.09%) test_sysctl_loop2.bpf.linked1.o sysctl_tcp_mem 1508 1501 -7 (-0.46%) 29 28 -1 (-3.45%) test_tc_dtime.bpf.linked1.o egress_fwdns_prio100 468 435 -33 (-7.05%) 45 41 -4 (-8.89%) test_tc_dtime.bpf.linked1.o ingress_fwdns_prio100 398 408 +10 (+2.51%) 42 39 -3 (-7.14%) test_tc_dtime.bpf.linked1.o ingress_fwdns_prio101 1096 842 -254 (-23.18%) 97 73 -24 (-24.74%) test_tcp_hdr_options.bpf.linked1.o estab 2758 2408 -350 (-12.69%) 208 181 -27 (-12.98%) test_urandom_usdt.bpf.linked1.o urand_read_with_sema 466 448 -18 (-3.86%) 31 28 -3 (-9.68%) test_urandom_usdt.bpf.linked1.o urand_read_without_sema 466 448 -18 (-3.86%) 31 28 -3 (-9.68%) test_urandom_usdt.bpf.linked1.o urandlib_read_with_sema 466 448 -18 (-3.86%) 31 28 -3 (-9.68%) test_urandom_usdt.bpf.linked1.o urandlib_read_without_sema 466 448 -18 (-3.86%) 31 28 -3 (-9.68%) test_xdp_noinline.bpf.linked1.o balancer_ingress_v6 4302 4294 -8 (-0.19%) 257 256 -1 (-0.39%) xdp_synproxy_kern.bpf.linked1.o syncookie_tc 583722 405757 -177965 (-30.49%) 35846 25735 -10111 (-28.21%) xdp_synproxy_kern.bpf.linked1.o syncookie_xdp 609123 479055 -130068 (-21.35%) 35452 29145 -6307 (-17.79%) ---------------------------------------- -------------------------- --------------- --------------- ------------------ ---------------- ---------------- -------------------
Signed-off-by: Andrii Nakryiko andrii@kernel.org Link: https://lore.kernel.org/r/20221104163649.121784-4-andrii@kernel.org Signed-off-by: Alexei Starovoitov ast@kernel.org Stable-dep-of: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction") Conflicts: kernel/bpf/verifier.c Signed-off-by: Pu Lehui pulehui@huawei.com --- kernel/bpf/verifier.c | 35 ++++++++++++++++++++++++++++++++++- 1 file changed, 34 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index edb19ada0405..40ac67a04ab7 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -1359,7 +1359,7 @@ static void __mark_reg_unknown(const struct bpf_verifier_env *env, reg->type = SCALAR_VALUE; reg->var_off = tnum_unknown; reg->frameno = 0; - reg->precise = env->subprog_cnt > 1 || !env->bpf_capable; + reg->precise = !env->bpf_capable; __mark_reg_unbounded(reg); }
@@ -2097,12 +2097,42 @@ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int r return 0; if (!reg_mask && !stack_mask) return 0; + for (;;) { DECLARE_BITMAP(mask, 64); u32 history = st->jmp_history_cnt;
if (env->log.level & BPF_LOG_LEVEL) verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx); + + if (last_idx < 0) { + /* we are at the entry into subprog, which + * is expected for global funcs, but only if + * requested precise registers are R1-R5 + * (which are global func's input arguments) + */ + if (st->curframe == 0 && + st->frame[0]->subprogno > 0 && + st->frame[0]->callsite == BPF_MAIN_FUNC && + stack_mask == 0 && (reg_mask & ~0x3e) == 0) { + bitmap_from_u64(mask, reg_mask); + for_each_set_bit(i, mask, 32) { + reg = &st->frame[0]->regs[i]; + if (reg->type != SCALAR_VALUE) { + reg_mask &= ~(1u << i); + continue; + } + reg->precise = true; + } + return 0; + } + + verbose(env, "BUG backtracing func entry subprog %d reg_mask %x stack_mask %llx\n", + st->frame[0]->subprogno, reg_mask, stack_mask); + WARN_ONCE(1, "verifier backtracking bug"); + return -EFAULT; + } + for (i = last_idx;;) { if (skip_first) { err = 0; @@ -11846,6 +11876,9 @@ static int do_check_common(struct bpf_verifier_env *env, int subprog) 0 /* frameno */, subprog);
+ state->first_insn_idx = env->subprog_info[subprog].start; + state->last_insn_idx = -1; + regs = state->frame[state->curframe]->regs; if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { ret = btf_prepare_func_args(env, subprog, regs);
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
thanks,
greg k-h
On 2023/8/4 18:33, Greg KH wrote:
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
alright, it's fine for me. will send them soon.
thanks,
greg k-h
On Fri, 2023-08-04 at 20:15 +0800, Pu Lehui wrote:
On 2023/8/4 18:33, Greg KH wrote:
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
alright, it's fine for me. will send them soon.
Hi Pu,
I tried backporting to 5.15 and was unable to get all test_progs* green, because of necessity to pull in other conflicting commits. So, you strategy on backporting only sub-set relevant for this specific failure seems reasonable to me.
Thanks, Eduard.
thanks,
greg k-h
On 2023/8/4 20:18, Eduard Zingerman wrote:
On Fri, 2023-08-04 at 20:15 +0800, Pu Lehui wrote:
On 2023/8/4 18:33, Greg KH wrote:
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
alright, it's fine for me. will send them soon.
Hi Pu,
I tried backporting to 5.15 and was unable to get all test_progs* green, because of necessity to pull in other conflicting commits. So, you strategy on backporting only sub-set relevant for this specific failure seems reasonable to me.
yep, actually, I'd prefer to backport the first 4 patches.
Thanks, Eduard.
thanks,
greg k-h
On 04.08.23 12:33, Greg KH wrote:
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
I see you explaining this occasionally, makes me wonder if we should add something like the following to Documentation/process/stable-kernel-rules.rst (sorry, no diff/context, but I guess you'll understand things nevertheless):
""" When using option 2 or 3 you can target a specific stable series. When doing so, you have to ensure that the fix or and equivalent is present or submitted to all newer stable trees, as users updating to them otherwise might encounter regressions. Hence, if you want a patch that was merged for 5.19-rc1 included in the 5.10 stable series, you must submit it for any stable series still maintained in the v5.11..v5.18 range. """
I plan to send my second set of changes soon anyway, would be simple to include this there.
Ciao, Thorsten
On Fri, Aug 04, 2023 at 08:56:09PM +0200, Thorsten Leemhuis wrote:
On 04.08.23 12:33, Greg KH wrote:
On Tue, Aug 01, 2023 at 10:36:55PM +0800, Pu Lehui wrote:
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit be2ef8161572ec1973124ebc50f56dafc2925e07 ]
For obvious reasons, I can't take this series only for 5.10 and not for 5.15, otherwise you would update your kernel and have a regression.
So please, create a 5.15.y series also, and resend both of these, and then we will be glad to apply them. For this series, I've dropped them from my review queue now.
I see you explaining this occasionally, makes me wonder if we should add something like the following to Documentation/process/stable-kernel-rules.rst (sorry, no diff/context, but I guess you'll understand things nevertheless):
""" When using option 2 or 3 you can target a specific stable series. When doing so, you have to ensure that the fix or and equivalent is present or submitted to all newer stable trees, as users updating to them otherwise might encounter regressions. Hence, if you want a patch that was merged for 5.19-rc1 included in the 5.10 stable series, you must submit it for any stable series still maintained in the v5.11..v5.18 range. """
Sure, hopefully people read the documentation, at the very least, I can just point the at it and stop saying this over and over :)
thanks,
greg k-h
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit f63181b6ae79fd3b034cde641db774268c2c3acf ]
Setting reg->precise to true in current state is not necessary from correctness standpoint, but it does pessimise the whole precision (or rather "imprecision", because that's what we want to keep as much as possible) tracking. Why is somewhat subtle and my best attempt to explain this is recorded in an extensive comment for __mark_chain_precise() function. Some more careful thinking and code reading is probably required still to grok this completely, unfortunately. Whiteboarding and a bunch of extra handwaiving in person would be even more helpful, but is deemed impractical in Git commit.
Next patch pushes this imprecision property even further, building on top of the insights described in this patch.
End results are pretty nice, we get reduction in number of total instructions and states verified due to a better states reuse, as some of the states are now more generic and permissive due to less unnecessary precise=true requirements.
SELFTESTS RESULTS =================
$ ./veristat -C -e file,prog,insns,states ~/subprog-precise-results.csv ~/imprecise-early-results.csv | grep -v '+0' File Program Total insns (A) Total insns (B) Total insns (DIFF) Total states (A) Total states (B) Total states (DIFF) --------------------------------------- ---------------------- --------------- --------------- ------------------ ---------------- ---------------- ------------------- bpf_iter_ksym.bpf.linked1.o dump_ksym 347 285 -62 (-17.87%) 20 19 -1 (-5.00%) pyperf600_bpf_loop.bpf.linked1.o on_event 3678 3736 +58 (+1.58%) 276 285 +9 (+3.26%) setget_sockopt.bpf.linked1.o skops_sockopt 4038 3947 -91 (-2.25%) 347 343 -4 (-1.15%) test_l4lb.bpf.linked1.o balancer_ingress 4559 2611 -1948 (-42.73%) 118 105 -13 (-11.02%) test_l4lb_noinline.bpf.linked1.o balancer_ingress 6279 6268 -11 (-0.18%) 237 236 -1 (-0.42%) test_misc_tcp_hdr_options.bpf.linked1.o misc_estab 1307 1303 -4 (-0.31%) 100 99 -1 (-1.00%) test_sk_lookup.bpf.linked1.o ctx_narrow_access 456 447 -9 (-1.97%) 39 38 -1 (-2.56%) test_sysctl_loop1.bpf.linked1.o sysctl_tcp_mem 1389 1384 -5 (-0.36%) 26 25 -1 (-3.85%) test_tc_dtime.bpf.linked1.o egress_fwdns_prio101 518 485 -33 (-6.37%) 51 46 -5 (-9.80%) test_tc_dtime.bpf.linked1.o egress_host 519 468 -51 (-9.83%) 50 44 -6 (-12.00%) test_tc_dtime.bpf.linked1.o ingress_fwdns_prio101 842 1000 +158 (+18.76%) 73 88 +15 (+20.55%) xdp_synproxy_kern.bpf.linked1.o syncookie_tc 405757 373173 -32584 (-8.03%) 25735 22882 -2853 (-11.09%) xdp_synproxy_kern.bpf.linked1.o syncookie_xdp 479055 371590 -107465 (-22.43%) 29145 22207 -6938 (-23.81%) --------------------------------------- ---------------------- --------------- --------------- ------------------ ---------------- ---------------- -------------------
Slight regression in test_tc_dtime.bpf.linked1.o/ingress_fwdns_prio101 is left for a follow up, there might be some more precision-related bugs in existing BPF verifier logic.
CILIUM RESULTS ==============
$ ./veristat -C -e file,prog,insns,states ~/subprog-precise-results-cilium.csv ~/imprecise-early-results-cilium.csv | grep -v '+0' File Program Total insns (A) Total insns (B) Total insns (DIFF) Total states (A) Total states (B) Total states (DIFF) ------------- ------------------------------ --------------- --------------- ------------------ ---------------- ---------------- ------------------- bpf_host.o cil_from_host 762 556 -206 (-27.03%) 43 37 -6 (-13.95%) bpf_host.o tail_handle_nat_fwd_ipv4 23541 23426 -115 (-0.49%) 1538 1537 -1 (-0.07%) bpf_host.o tail_nodeport_nat_egress_ipv4 33592 33566 -26 (-0.08%) 2163 2161 -2 (-0.09%) bpf_lxc.o tail_handle_nat_fwd_ipv4 23541 23426 -115 (-0.49%) 1538 1537 -1 (-0.07%) bpf_overlay.o tail_nodeport_nat_egress_ipv4 33581 33543 -38 (-0.11%) 2160 2157 -3 (-0.14%) bpf_xdp.o tail_handle_nat_fwd_ipv4 21659 20920 -739 (-3.41%) 1440 1376 -64 (-4.44%) bpf_xdp.o tail_handle_nat_fwd_ipv6 17084 17039 -45 (-0.26%) 907 905 -2 (-0.22%) bpf_xdp.o tail_lb_ipv4 73442 73430 -12 (-0.02%) 4370 4369 -1 (-0.02%) bpf_xdp.o tail_lb_ipv6 152114 151895 -219 (-0.14%) 6493 6479 -14 (-0.22%) bpf_xdp.o tail_nodeport_nat_egress_ipv4 17377 17200 -177 (-1.02%) 1125 1111 -14 (-1.24%) bpf_xdp.o tail_nodeport_nat_ingress_ipv6 6405 6397 -8 (-0.12%) 309 308 -1 (-0.32%) bpf_xdp.o tail_rev_nodeport_lb4 7126 6934 -192 (-2.69%) 414 402 -12 (-2.90%) bpf_xdp.o tail_rev_nodeport_lb6 18059 17905 -154 (-0.85%) 1105 1096 -9 (-0.81%) ------------- ------------------------------ --------------- --------------- ------------------ ---------------- ---------------- -------------------
Signed-off-by: Andrii Nakryiko andrii@kernel.org Link: https://lore.kernel.org/r/20221104163649.121784-5-andrii@kernel.org Signed-off-by: Alexei Starovoitov ast@kernel.org Stable-dep-of: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction") Signed-off-by: Pu Lehui pulehui@huawei.com --- kernel/bpf/verifier.c | 103 +++++++++++++++++++++++++++++++++++++----- 1 file changed, 91 insertions(+), 12 deletions(-)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 40ac67a04ab7..dd991b39d74b 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -2023,8 +2023,11 @@ static void mark_all_scalars_precise(struct bpf_verifier_env *env,
/* big hammer: mark all scalars precise in this path. * pop_stack may still get !precise scalars. + * We also skip current state and go straight to first parent state, + * because precision markings in current non-checkpointed state are + * not needed. See why in the comment in __mark_chain_precision below. */ - for (; st; st = st->parent) + for (st = st->parent; st; st = st->parent) { for (i = 0; i <= st->curframe; i++) { func = st->frame[i]; for (j = 0; j < BPF_REG_FP; j++) { @@ -2042,8 +2045,88 @@ static void mark_all_scalars_precise(struct bpf_verifier_env *env, reg->precise = true; } } + } }
+/* + * __mark_chain_precision() backtracks BPF program instruction sequence and + * chain of verifier states making sure that register *regno* (if regno >= 0) + * and/or stack slot *spi* (if spi >= 0) are marked as precisely tracked + * SCALARS, as well as any other registers and slots that contribute to + * a tracked state of given registers/stack slots, depending on specific BPF + * assembly instructions (see backtrack_insns() for exact instruction handling + * logic). This backtracking relies on recorded jmp_history and is able to + * traverse entire chain of parent states. This process ends only when all the + * necessary registers/slots and their transitive dependencies are marked as + * precise. + * + * One important and subtle aspect is that precise marks *do not matter* in + * the currently verified state (current state). It is important to understand + * why this is the case. + * + * First, note that current state is the state that is not yet "checkpointed", + * i.e., it is not yet put into env->explored_states, and it has no children + * states as well. It's ephemeral, and can end up either a) being discarded if + * compatible explored state is found at some point or BPF_EXIT instruction is + * reached or b) checkpointed and put into env->explored_states, branching out + * into one or more children states. + * + * In the former case, precise markings in current state are completely + * ignored by state comparison code (see regsafe() for details). Only + * checkpointed ("old") state precise markings are important, and if old + * state's register/slot is precise, regsafe() assumes current state's + * register/slot as precise and checks value ranges exactly and precisely. If + * states turn out to be compatible, current state's necessary precise + * markings and any required parent states' precise markings are enforced + * after the fact with propagate_precision() logic, after the fact. But it's + * important to realize that in this case, even after marking current state + * registers/slots as precise, we immediately discard current state. So what + * actually matters is any of the precise markings propagated into current + * state's parent states, which are always checkpointed (due to b) case above). + * As such, for scenario a) it doesn't matter if current state has precise + * markings set or not. + * + * Now, for the scenario b), checkpointing and forking into child(ren) + * state(s). Note that before current state gets to checkpointing step, any + * processed instruction always assumes precise SCALAR register/slot + * knowledge: if precise value or range is useful to prune jump branch, BPF + * verifier takes this opportunity enthusiastically. Similarly, when + * register's value is used to calculate offset or memory address, exact + * knowledge of SCALAR range is assumed, checked, and enforced. So, similar to + * what we mentioned above about state comparison ignoring precise markings + * during state comparison, BPF verifier ignores and also assumes precise + * markings *at will* during instruction verification process. But as verifier + * assumes precision, it also propagates any precision dependencies across + * parent states, which are not yet finalized, so can be further restricted + * based on new knowledge gained from restrictions enforced by their children + * states. This is so that once those parent states are finalized, i.e., when + * they have no more active children state, state comparison logic in + * is_state_visited() would enforce strict and precise SCALAR ranges, if + * required for correctness. + * + * To build a bit more intuition, note also that once a state is checkpointed, + * the path we took to get to that state is not important. This is crucial + * property for state pruning. When state is checkpointed and finalized at + * some instruction index, it can be correctly and safely used to "short + * circuit" any *compatible* state that reaches exactly the same instruction + * index. I.e., if we jumped to that instruction from a completely different + * code path than original finalized state was derived from, it doesn't + * matter, current state can be discarded because from that instruction + * forward having a compatible state will ensure we will safely reach the + * exit. States describe preconditions for further exploration, but completely + * forget the history of how we got here. + * + * This also means that even if we needed precise SCALAR range to get to + * finalized state, but from that point forward *that same* SCALAR register is + * never used in a precise context (i.e., it's precise value is not needed for + * correctness), it's correct and safe to mark such register as "imprecise" + * (i.e., precise marking set to false). This is what we rely on when we do + * not set precise marking in current state. If no child state requires + * precision for any given SCALAR register, it's safe to dictate that it can + * be imprecise. If any child state does require this register to be precise, + * we'll mark it precise later retroactively during precise markings + * propagation from child state to parent states. + */ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int regno, int spi) { @@ -2061,6 +2144,10 @@ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int r if (!env->bpf_capable) return 0;
+ /* Do sanity checks against current state of register and/or stack + * slot, but don't set precise flag in current state, as precision + * tracking in the current state is unnecessary. + */ func = st->frame[frame]; if (regno >= 0) { reg = &func->regs[regno]; @@ -2068,11 +2155,7 @@ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int r WARN_ONCE(1, "backtracing misuse"); return -EFAULT; } - if (!reg->precise) - new_marks = true; - else - reg_mask = 0; - reg->precise = true; + new_marks = true; }
while (spi >= 0) { @@ -2085,11 +2168,7 @@ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int r stack_mask = 0; break; } - if (!reg->precise) - new_marks = true; - else - stack_mask = 0; - reg->precise = true; + new_marks = true; break; }
@@ -9263,7 +9342,7 @@ static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold, if (env->explore_alu_limits) return false; if (rcur->type == SCALAR_VALUE) { - if (!rold->precise && !rcur->precise) + if (!rold->precise) return true; /* new val must satisfy old val knowledge */ return range_within(rold, rcur) &&
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit 7a830b53c17bbadcf99f778f28aaaa4e6c41df5f ]
Exploit the property of about-to-be-checkpointed state to be able to forget all precise markings up to that point even more aggressively. We now clear all potentially inherited precise markings right before checkpointing and branching off into child state. If any of children states require precise knowledge of any SCALAR register, those will be propagated backwards later on before this state is finalized, preserving correctness.
There is a single selftests BPF program change, but tremendous one: 25x reduction in number of verified instructions and states in trace_virtqueue_add_sgs.
Cilium results are more modest, but happen across wider range of programs.
SELFTESTS RESULTS =================
$ ./veristat -C -e file,prog,insns,states ~/imprecise-early-results.csv ~/imprecise-aggressive-results.csv | grep -v '+0' File Program Total insns (A) Total insns (B) Total insns (DIFF) Total states (A) Total states (B) Total states (DIFF) ------------------- ----------------------- --------------- --------------- ------------------ ---------------- ---------------- ------------------- loop6.bpf.linked1.o trace_virtqueue_add_sgs 398057 15114 -382943 (-96.20%) 8717 336 -8381 (-96.15%) ------------------- ----------------------- --------------- --------------- ------------------ ---------------- ---------------- -------------------
CILIUM RESULTS ==============
$ ./veristat -C -e file,prog,insns,states ~/imprecise-early-results-cilium.csv ~/imprecise-aggressive-results-cilium.csv | grep -v '+0' File Program Total insns (A) Total insns (B) Total insns (DIFF) Total states (A) Total states (B) Total states (DIFF) ------------- -------------------------------- --------------- --------------- ------------------ ---------------- ---------------- ------------------- bpf_host.o tail_handle_nat_fwd_ipv4 23426 23221 -205 (-0.88%) 1537 1515 -22 (-1.43%) bpf_host.o tail_handle_nat_fwd_ipv6 13009 12904 -105 (-0.81%) 719 708 -11 (-1.53%) bpf_host.o tail_nodeport_nat_ingress_ipv6 5261 5196 -65 (-1.24%) 247 243 -4 (-1.62%) bpf_host.o tail_nodeport_nat_ipv6_egress 3446 3406 -40 (-1.16%) 203 198 -5 (-2.46%) bpf_lxc.o tail_handle_nat_fwd_ipv4 23426 23221 -205 (-0.88%) 1537 1515 -22 (-1.43%) bpf_lxc.o tail_handle_nat_fwd_ipv6 13009 12904 -105 (-0.81%) 719 708 -11 (-1.53%) bpf_lxc.o tail_ipv4_ct_egress 5074 4897 -177 (-3.49%) 255 248 -7 (-2.75%) bpf_lxc.o tail_ipv4_ct_ingress 5100 4923 -177 (-3.47%) 255 248 -7 (-2.75%) bpf_lxc.o tail_ipv4_ct_ingress_policy_only 5100 4923 -177 (-3.47%) 255 248 -7 (-2.75%) bpf_lxc.o tail_ipv6_ct_egress 4558 4536 -22 (-0.48%) 188 187 -1 (-0.53%) bpf_lxc.o tail_ipv6_ct_ingress 4578 4556 -22 (-0.48%) 188 187 -1 (-0.53%) bpf_lxc.o tail_ipv6_ct_ingress_policy_only 4578 4556 -22 (-0.48%) 188 187 -1 (-0.53%) bpf_lxc.o tail_nodeport_nat_ingress_ipv6 5261 5196 -65 (-1.24%) 247 243 -4 (-1.62%) bpf_overlay.o tail_nodeport_nat_ingress_ipv6 5261 5196 -65 (-1.24%) 247 243 -4 (-1.62%) bpf_overlay.o tail_nodeport_nat_ipv6_egress 3482 3442 -40 (-1.15%) 204 201 -3 (-1.47%) bpf_xdp.o tail_nodeport_nat_egress_ipv4 17200 15619 -1581 (-9.19%) 1111 1010 -101 (-9.09%) ------------- -------------------------------- --------------- --------------- ------------------ ---------------- ---------------- -------------------
Signed-off-by: Andrii Nakryiko andrii@kernel.org Link: https://lore.kernel.org/r/20221104163649.121784-6-andrii@kernel.org Signed-off-by: Alexei Starovoitov ast@kernel.org Stable-dep-of: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction") Signed-off-by: Pu Lehui pulehui@huawei.com --- kernel/bpf/verifier.c | 37 +++++++++++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+)
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index dd991b39d74b..8f1e43df8c5f 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -2048,6 +2048,31 @@ static void mark_all_scalars_precise(struct bpf_verifier_env *env, } }
+static void mark_all_scalars_imprecise(struct bpf_verifier_env *env, struct bpf_verifier_state *st) +{ + struct bpf_func_state *func; + struct bpf_reg_state *reg; + int i, j; + + for (i = 0; i <= st->curframe; i++) { + func = st->frame[i]; + for (j = 0; j < BPF_REG_FP; j++) { + reg = &func->regs[j]; + if (reg->type != SCALAR_VALUE) + continue; + reg->precise = false; + } + for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { + if (!is_spilled_reg(&func->stack[j])) + continue; + reg = &func->stack[j].spilled_ptr; + if (reg->type != SCALAR_VALUE) + continue; + reg->precise = false; + } + } +} + /* * __mark_chain_precision() backtracks BPF program instruction sequence and * chain of verifier states making sure that register *regno* (if regno >= 0) @@ -2126,6 +2151,14 @@ static void mark_all_scalars_precise(struct bpf_verifier_env *env, * be imprecise. If any child state does require this register to be precise, * we'll mark it precise later retroactively during precise markings * propagation from child state to parent states. + * + * Skipping precise marking setting in current state is a mild version of + * relying on the above observation. But we can utilize this property even + * more aggressively by proactively forgetting any precise marking in the + * current state (which we inherited from the parent state), right before we + * checkpoint it and branch off into new child state. This is done by + * mark_all_scalars_imprecise() to hopefully get more permissive and generic + * finalized states which help in short circuiting more future states. */ static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int regno, int spi) @@ -9875,6 +9908,10 @@ static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) env->prev_jmps_processed = env->jmps_processed; env->prev_insn_processed = env->insn_processed;
+ /* forget precise markings we inherited, see __mark_chain_precision */ + if (env->bpf_capable) + mark_all_scalars_imprecise(env, cur); + /* add new state to the head of linked list */ new = &new_sl->state; err = copy_verifier_state(new, cur);
From: Andrii Nakryiko andrii@kernel.org
[ Upstream commit 4f999b767769b76378c3616c624afd6f4bb0d99f ]
test_align selftest relies on BPF verifier log emitting register states for specific instructions in expected format. Unfortunately, BPF verifier precision backtracking log interferes with such expectations. And instruction on which precision propagation happens sometimes don't output full expected register states. This does indeed look like something to be improved in BPF verifier, but is beyond the scope of this patch set.
So to make test_align a bit more robust, inject few dummy R4 = R5 instructions which capture desired state of R5 and won't have precision tracking logs on them. This fixes tests until we can improve BPF verifier output in the presence of precision tracking.
Signed-off-by: Andrii Nakryiko andrii@kernel.org Link: https://lore.kernel.org/r/20221104163649.121784-7-andrii@kernel.org Signed-off-by: Alexei Starovoitov ast@kernel.org Stable-dep-of: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction") Conflicts: tools/testing/selftests/bpf/prog_tests/align.c Signed-off-by: Pu Lehui pulehui@huawei.com --- .../testing/selftests/bpf/prog_tests/align.c | 36 ++++++++++++------- 1 file changed, 23 insertions(+), 13 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/align.c b/tools/testing/selftests/bpf/prog_tests/align.c index 5861446d0777..7996ec07e0bd 100644 --- a/tools/testing/selftests/bpf/prog_tests/align.c +++ b/tools/testing/selftests/bpf/prog_tests/align.c @@ -2,7 +2,7 @@ #include <test_progs.h>
#define MAX_INSNS 512 -#define MAX_MATCHES 16 +#define MAX_MATCHES 24
struct bpf_reg_match { unsigned int line; @@ -267,6 +267,7 @@ static struct bpf_align_test tests[] = { */ BPF_MOV64_REG(BPF_REG_5, BPF_REG_2), BPF_ALU64_REG(BPF_ADD, BPF_REG_5, BPF_REG_6), + BPF_MOV64_REG(BPF_REG_4, BPF_REG_5), BPF_ALU64_IMM(BPF_ADD, BPF_REG_5, 14), BPF_MOV64_REG(BPF_REG_4, BPF_REG_5), BPF_ALU64_IMM(BPF_ADD, BPF_REG_4, 4), @@ -280,6 +281,7 @@ static struct bpf_align_test tests[] = { BPF_MOV64_REG(BPF_REG_5, BPF_REG_2), BPF_ALU64_IMM(BPF_ADD, BPF_REG_5, 14), BPF_ALU64_REG(BPF_ADD, BPF_REG_5, BPF_REG_6), + BPF_MOV64_REG(BPF_REG_4, BPF_REG_5), BPF_ALU64_IMM(BPF_ADD, BPF_REG_5, 4), BPF_ALU64_REG(BPF_ADD, BPF_REG_5, BPF_REG_6), BPF_MOV64_REG(BPF_REG_4, BPF_REG_5), @@ -311,44 +313,52 @@ static struct bpf_align_test tests[] = { {15, "R4=pkt(id=1,off=18,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, {15, "R5=pkt(id=1,off=14,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* Variable offset is added to R5 packet pointer, - * resulting in auxiliary alignment of 4. + * resulting in auxiliary alignment of 4. To avoid BPF + * verifier's precision backtracking logging + * interfering we also have a no-op R4 = R5 + * instruction to validate R5 state. We also check + * that R4 is what it should be in such case. */ - {18, "R5_w=pkt(id=2,off=0,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {19, "R4_w=pkt(id=2,off=0,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {19, "R5_w=pkt(id=2,off=0,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* Constant offset is added to R5, resulting in * reg->off of 14. */ - {19, "R5_w=pkt(id=2,off=14,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {20, "R5_w=pkt(id=2,off=14,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* At the time the word size load is performed from R5, * its total fixed offset is NET_IP_ALIGN + reg->off * (14) which is 16. Then the variable offset is 4-byte * aligned, so the total offset is 4-byte aligned and * meets the load's requirements. */ - {23, "R4=pkt(id=2,off=18,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, - {23, "R5=pkt(id=2,off=14,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {24, "R4=pkt(id=2,off=18,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {24, "R5=pkt(id=2,off=14,r=18,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* Constant offset is added to R5 packet pointer, * resulting in reg->off value of 14. */ - {26, "R5_w=pkt(id=0,off=14,r=8"}, + {27, "R5_w=pkt(id=0,off=14,r=8"}, /* Variable offset is added to R5, resulting in a - * variable offset of (4n). + * variable offset of (4n). See comment for insn #19 + * for R4 = R5 trick. */ - {27, "R5_w=pkt(id=3,off=14,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {29, "R4_w=pkt(id=3,off=14,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {29, "R5_w=pkt(id=3,off=14,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* Constant is added to R5 again, setting reg->off to 18. */ - {28, "R5_w=pkt(id=3,off=18,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, + {30, "R5_w=pkt(id=3,off=18,r=0,umax_value=1020,var_off=(0x0; 0x3fc))"}, /* And once more we add a variable; resulting var_off * is still (4n), fixed offset is not changed. * Also, we create a new reg->id. */ - {29, "R5_w=pkt(id=4,off=18,r=0,umax_value=2040,var_off=(0x0; 0x7fc)"}, + {32, "R4_w=pkt(id=4,off=18,r=0,umax_value=2040,var_off=(0x0; 0x7fc)"}, + {32, "R5_w=pkt(id=4,off=18,r=0,umax_value=2040,var_off=(0x0; 0x7fc)"}, /* At the time the word size load is performed from R5, * its total fixed offset is NET_IP_ALIGN + reg->off (18) * which is 20. Then the variable offset is (4n), so * the total offset is 4-byte aligned and meets the * load's requirements. */ - {33, "R4=pkt(id=4,off=22,r=22,umax_value=2040,var_off=(0x0; 0x7fc)"}, - {33, "R5=pkt(id=4,off=18,r=22,umax_value=2040,var_off=(0x0; 0x7fc)"}, + {35, "R4=pkt(id=4,off=22,r=22,umax_value=2040,var_off=(0x0; 0x7fc)"}, + {35, "R5=pkt(id=4,off=18,r=22,umax_value=2040,var_off=(0x0; 0x7fc)"}, }, }, {
From: Yonghong Song yhs@fb.com
[ Upstream commit 63d78b7e8ca2d0eb8c687a355fa19d01b6fcc723 ]
With latest llvm17, selftest fexit_bpf2bpf/func_replace_return_code has the following verification failure:
0: R1=ctx(off=0,imm=0) R10=fp0 ; int connect_v4_prog(struct bpf_sock_addr *ctx) 0: (bf) r7 = r1 ; R1=ctx(off=0,imm=0) R7_w=ctx(off=0,imm=0) 1: (b4) w6 = 0 ; R6_w=0 ; memset(&tuple.ipv4.saddr, 0, sizeof(tuple.ipv4.saddr)); ... ; return do_bind(ctx) ? 1 : 0; 179: (bf) r1 = r7 ; R1=ctx(off=0,imm=0) R7=ctx(off=0,imm=0) 180: (85) call pc+147 Func#3 is global and valid. Skipping. 181: R0_w=scalar() 181: (bc) w6 = w0 ; R0_w=scalar() R6_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) 182: (05) goto pc-129 ; } 54: (bc) w0 = w6 ; R0_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) R6_w=scalar(umax=4294967295,var_off=(0x0; 0xffffffff)) 55: (95) exit At program exit the register R0 has value (0x0; 0xffffffff) should have been in (0x0; 0x1) processed 281 insns (limit 1000000) max_states_per_insn 1 total_states 26 peak_states 26 mark_read 13 -- END PROG LOAD LOG -- libbpf: prog 'connect_v4_prog': failed to load: -22
The corresponding source code:
__attribute__ ((noinline)) int do_bind(struct bpf_sock_addr *ctx) { struct sockaddr_in sa = {};
sa.sin_family = AF_INET; sa.sin_port = bpf_htons(0); sa.sin_addr.s_addr = bpf_htonl(SRC_REWRITE_IP4);
if (bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa)) != 0) return 0;
return 1; } ... SEC("cgroup/connect4") int connect_v4_prog(struct bpf_sock_addr *ctx) { ... return do_bind(ctx) ? 1 : 0; }
Insn 180 is a call to 'do_bind'. The call's return value is also the return value for the program. Since do_bind() returns 0/1, so it is legitimate for compiler to optimize 'return do_bind(ctx) ? 1 : 0' to 'return do_bind(ctx)'. However, such optimization breaks verifier as the return value of 'do_bind()' is marked as any scalar which violates the requirement of prog return value 0/1.
There are two ways to fix this problem, (1) changing 'return 1' in do_bind() to e.g. 'return 10' so the compiler has to do 'do_bind(ctx) ? 1 :0', or (2) suggested by Andrii, marking do_bind() with __weak attribute so the compiler cannot make any assumption on do_bind() return value.
This patch adopted adding __weak approach which is simpler and more resistant to potential compiler optimizations.
Suggested-by: Andrii Nakryiko andrii@kernel.org Signed-off-by: Yonghong Song yhs@fb.com Signed-off-by: Andrii Nakryiko andrii@kernel.org Link: https://lore.kernel.org/bpf/20230310012410.2920570-1-yhs@fb.com Signed-off-by: Pu Lehui pulehui@huawei.com --- tools/testing/selftests/bpf/progs/connect4_prog.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tools/testing/selftests/bpf/progs/connect4_prog.c b/tools/testing/selftests/bpf/progs/connect4_prog.c index a943d394fd3a..38ab1ce32e57 100644 --- a/tools/testing/selftests/bpf/progs/connect4_prog.c +++ b/tools/testing/selftests/bpf/progs/connect4_prog.c @@ -33,7 +33,7 @@
int _version SEC("version") = 1;
-__attribute__ ((noinline)) +__attribute__ ((noinline)) __weak int do_bind(struct bpf_sock_addr *ctx) { struct sockaddr_in sa = {};
From: Ilya Leoshkevich iii@linux.ibm.com
[ Upstream commit 7ce878ca81bca7811e669db4c394b86780e0dbe4 ]
sk_assign is failing on an s390x machine running Debian "bookworm" for 2 reasons: legacy server_map definition and uninitialized addrlen in recvfrom() call.
Fix by adding a new-style server_map definition and dropping addrlen (recvfrom() allows NULL values for src_addr and addrlen).
Since the test should support tc built without libbpf, build the prog twice: with the old-style definition and with the new-style definition, then select the right one at runtime. This could be done at compile time too, but this would not be cross-compilation friendly.
Signed-off-by: Ilya Leoshkevich iii@linux.ibm.com Link: https://lore.kernel.org/r/20230129190501.1624747-2-iii@linux.ibm.com Signed-off-by: Alexei Starovoitov ast@kernel.org Conflicts: tools/testing/selftests/bpf/prog_tests/sk_assign.c Signed-off-by: Pu Lehui pulehui@huawei.com --- .../selftests/bpf/prog_tests/sk_assign.c | 25 ++++++++++++++----- .../selftests/bpf/progs/test_sk_assign.c | 11 ++++++++ .../bpf/progs/test_sk_assign_libbpf.c | 3 +++ 3 files changed, 33 insertions(+), 6 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c
diff --git a/tools/testing/selftests/bpf/prog_tests/sk_assign.c b/tools/testing/selftests/bpf/prog_tests/sk_assign.c index 3a469099f30d..e09c5239a595 100644 --- a/tools/testing/selftests/bpf/prog_tests/sk_assign.c +++ b/tools/testing/selftests/bpf/prog_tests/sk_assign.c @@ -29,7 +29,23 @@ static int stop, duration; static bool configure_stack(void) { + char tc_version[128]; char tc_cmd[BUFSIZ]; + char *prog; + FILE *tc; + + /* Check whether tc is built with libbpf. */ + tc = popen("tc -V", "r"); + if (CHECK_FAIL(!tc)) + return false; + if (CHECK_FAIL(!fgets(tc_version, sizeof(tc_version), tc))) + return false; + if (strstr(tc_version, ", libbpf ")) + prog = "test_sk_assign_libbpf.o"; + else + prog = "test_sk_assign.o"; + if (CHECK_FAIL(pclose(tc))) + return false;
/* Move to a new networking namespace */ if (CHECK_FAIL(unshare(CLONE_NEWNET))) @@ -46,8 +62,8 @@ configure_stack(void) /* Load qdisc, BPF program */ if (CHECK_FAIL(system("tc qdisc add dev lo clsact"))) return false; - sprintf(tc_cmd, "%s %s %s %s", "tc filter add dev lo ingress bpf", - "direct-action object-file ./test_sk_assign.o", + sprintf(tc_cmd, "%s %s %s %s %s", "tc filter add dev lo ingress bpf", + "direct-action object-file", prog, "section classifier/sk_assign_test", (env.verbosity < VERBOSE_VERY) ? " 2>/dev/null" : "verbose"); if (CHECK(system(tc_cmd), "BPF load failed;", @@ -129,15 +145,12 @@ get_port(int fd) static ssize_t rcv_msg(int srv_client, int type) { - struct sockaddr_storage ss; char buf[BUFSIZ]; - socklen_t slen;
if (type == SOCK_STREAM) return read(srv_client, &buf, sizeof(buf)); else - return recvfrom(srv_client, &buf, sizeof(buf), 0, - (struct sockaddr *)&ss, &slen); + return recvfrom(srv_client, &buf, sizeof(buf), 0, NULL, NULL); }
static int diff --git a/tools/testing/selftests/bpf/progs/test_sk_assign.c b/tools/testing/selftests/bpf/progs/test_sk_assign.c index 1ecd987005d2..77fd42f835fc 100644 --- a/tools/testing/selftests/bpf/progs/test_sk_assign.c +++ b/tools/testing/selftests/bpf/progs/test_sk_assign.c @@ -16,6 +16,16 @@ #include <bpf/bpf_helpers.h> #include <bpf/bpf_endian.h>
+#if defined(IPROUTE2_HAVE_LIBBPF) +/* Use a new-style map definition. */ +struct { + __uint(type, BPF_MAP_TYPE_SOCKMAP); + __type(key, int); + __type(value, __u64); + __uint(pinning, LIBBPF_PIN_BY_NAME); + __uint(max_entries, 1); +} server_map SEC(".maps"); +#else /* Pin map under /sys/fs/bpf/tc/globals/<map name> */ #define PIN_GLOBAL_NS 2
@@ -35,6 +45,7 @@ struct { .max_elem = 1, .pinning = PIN_GLOBAL_NS, }; +#endif
int _version SEC("version") = 1; char _license[] SEC("license") = "GPL"; diff --git a/tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c b/tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c new file mode 100644 index 000000000000..dcf46adfda04 --- /dev/null +++ b/tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c @@ -0,0 +1,3 @@ +// SPDX-License-Identifier: GPL-2.0 +#define IPROUTE2_HAVE_LIBBPF +#include "test_sk_assign.c"
Add CC gregkh@linuxfoundation.org I am not sure if the email address of greg@kroah.com is valid
On 2023/8/1 22:36, Pu Lehui wrote:
Luiz Capitulino reported the test_verifier test failed: "precise: ST insn causing spi > allocated_stack". And it was introduced by the following upstream commit: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Eduard's investigation [4] shows that test failure is not a bug, but a difference in BPF verifier behavior between upstream, where commits [1,2,3] by Andrii are present, and 5.10, where these commits are absent.
Backporting strategy is consistent with Eduard in kernel version 6.1 [5], but with some conflicts in patch #1, #4 and #6 due to the bpf of 5.10 doesn't support more features. Both test_verifier and test_maps have passed, while test_progs and test_progs-no_alu32 with no new failure ceses.
Commits of Andrii: [1] be2ef8161572 ("bpf: allow precision tracking for programs with subprogs") [2] f63181b6ae79 ("bpf: stop setting precise in current state") [3] 7a830b53c17b ("bpf: aggressively forget precise markings during state checkpointing")
Links: [4] https://lore.kernel.org/stable/c9b10a8a551edafdfec855fbd35757c6238ad258.came... [5] https://lore.kernel.org/all/20230724124223.1176479-2-eddyz87@gmail.com/
Andrii Nakryiko (4): bpf: allow precision tracking for programs with subprogs bpf: stop setting precise in current state bpf: aggressively forget precise markings during state checkpointing selftests/bpf: make test_align selftest more robust
Ilya Leoshkevich (1): selftests/bpf: Fix sk_assign on s390x
Yonghong Song (1): selftests/bpf: Workaround verification failure for fexit_bpf2bpf/func_replace_return_code
kernel/bpf/verifier.c | 175 ++++++++++++++++-- .../testing/selftests/bpf/prog_tests/align.c | 36 ++-- .../selftests/bpf/prog_tests/sk_assign.c | 25 ++- .../selftests/bpf/progs/connect4_prog.c | 2 +- .../selftests/bpf/progs/test_sk_assign.c | 11 ++ .../bpf/progs/test_sk_assign_libbpf.c | 3 + 6 files changed, 219 insertions(+), 33 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c
On Wed, Aug 02, 2023 at 03:23:14PM +0800, Pu Lehui wrote:
Add CC gregkh@linuxfoundation.org
Why?
I am not sure if the email address of greg@kroah.com is valid
It is.
On 2023/8/1 22:36, Pu Lehui wrote:
Luiz Capitulino reported the test_verifier test failed: "precise: ST insn causing spi > allocated_stack". And it was introduced by the following upstream commit: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Eduard's investigation [4] shows that test failure is not a bug, but a difference in BPF verifier behavior between upstream, where commits [1,2,3] by Andrii are present, and 5.10, where these commits are absent.
Backporting strategy is consistent with Eduard in kernel version 6.1 [5], but with some conflicts in patch #1, #4 and #6 due to the bpf of 5.10 doesn't support more features. Both test_verifier and test_maps have passed, while test_progs and test_progs-no_alu32 with no new failure ceses.
Commits of Andrii: [1] be2ef8161572 ("bpf: allow precision tracking for programs with subprogs") [2] f63181b6ae79 ("bpf: stop setting precise in current state") [3] 7a830b53c17b ("bpf: aggressively forget precise markings during state checkpointing")
Links: [4] https://lore.kernel.org/stable/c9b10a8a551edafdfec855fbd35757c6238ad258.came... [5] https://lore.kernel.org/all/20230724124223.1176479-2-eddyz87@gmail.com/
Andrii Nakryiko (4): bpf: allow precision tracking for programs with subprogs bpf: stop setting precise in current state bpf: aggressively forget precise markings during state checkpointing selftests/bpf: make test_align selftest more robust
Ilya Leoshkevich (1): selftests/bpf: Fix sk_assign on s390x
Yonghong Song (1): selftests/bpf: Workaround verification failure for fexit_bpf2bpf/func_replace_return_code
kernel/bpf/verifier.c | 175 ++++++++++++++++-- .../testing/selftests/bpf/prog_tests/align.c | 36 ++-- .../selftests/bpf/prog_tests/sk_assign.c | 25 ++- .../selftests/bpf/progs/connect4_prog.c | 2 +- .../selftests/bpf/progs/test_sk_assign.c | 11 ++ .../bpf/progs/test_sk_assign_libbpf.c | 3 + 6 files changed, 219 insertions(+), 33 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c
What is the question here for me?
confused,
greg k-h
On 2023-08-01 10:36, Pu Lehui wrote:
Luiz Capitulino reported the test_verifier test failed: "precise: ST insn causing spi > allocated_stack". And it was introduced by the following upstream commit: ecdf985d7615 ("bpf: track immediate values written to stack by BPF_ST instruction")
Eduard's investigation [4] shows that test failure is not a bug, but a difference in BPF verifier behavior between upstream, where commits [1,2,3] by Andrii are present, and 5.10, where these commits are absent.
Backporting strategy is consistent with Eduard in kernel version 6.1 [5], but with some conflicts in patch #1, #4 and #6 due to the bpf of 5.10 doesn't support more features. Both test_verifier and test_maps have passed, while test_progs and test_progs-no_alu32 with no new failure ceses.
I tested this one today on top of 5.10.188.
Before this series:
# #760/p precise: ST insn causing spi > allocated_stack FAIL # Summary: 1192 PASSED, 546 SKIPPED, 1 FAILED
With this series applied:
# Summary: 1193 PASSED, 546 SKIPPED, 0 FAILED
Thank you very much Pu and Eduard for your work on this. Unfortunately, I haven't been able to test the 6.1 series yet (it may take a bit before I can do it).
Tested-by: Luiz Capitulino luizcap@amazon.com
Commits of Andrii: [1] be2ef8161572 ("bpf: allow precision tracking for programs with subprogs") [2] f63181b6ae79 ("bpf: stop setting precise in current state") [3] 7a830b53c17b ("bpf: aggressively forget precise markings during state checkpointing")
Links: [4] https://lore.kernel.org/stable/c9b10a8a551edafdfec855fbd35757c6238ad258.came... [5] https://lore.kernel.org/all/20230724124223.1176479-2-eddyz87@gmail.com/
Andrii Nakryiko (4): bpf: allow precision tracking for programs with subprogs bpf: stop setting precise in current state bpf: aggressively forget precise markings during state checkpointing selftests/bpf: make test_align selftest more robust
Ilya Leoshkevich (1): selftests/bpf: Fix sk_assign on s390x
Yonghong Song (1): selftests/bpf: Workaround verification failure for fexit_bpf2bpf/func_replace_return_code
kernel/bpf/verifier.c | 175 ++++++++++++++++-- .../testing/selftests/bpf/prog_tests/align.c | 36 ++-- .../selftests/bpf/prog_tests/sk_assign.c | 25 ++- .../selftests/bpf/progs/connect4_prog.c | 2 +- .../selftests/bpf/progs/test_sk_assign.c | 11 ++ .../bpf/progs/test_sk_assign_libbpf.c | 3 + 6 files changed, 219 insertions(+), 33 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/test_sk_assign_libbpf.c
-- 2.25.1
linux-stable-mirror@lists.linaro.org
-
Eduard Zingerman -
Greg KH -
Greg KH -
Luiz Capitulino -
Pu Lehui -
Pu Lehui -
Thorsten Leemhuis