| Message ID | 20210812164557.79046-4-kuniyu@amazon.co.jp (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | BPF |
| Headers | show |
| Series | BPF iterator for UNIX domain socket. | expand |
On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > The iterator can output almost the same result compared to /proc/net/unix. > The header line is aligned, and the Inode column uses "%8lu" because "%5lu" > can be easily overflown. > > # cat /sys/fs/bpf/unix > Num RefCount Protocol Flags Type St Inode Path It's totally my OCD, but why the column name is not aligned with values? I mean the "Inode" column. It's left aligned, but values (numbers) are right-aligned? I'd fix that while applying, but I can't apply due to selftests failures, so please take a look. > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > # cat /proc/net/unix > Num RefCount Protocol Flags Type St Inode Path > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to > Yonghong Song for analysing and fixing. > > [0] https://reviews.llvm.org/D107483 > > Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > Acked-by: Yonghong Song <yhs@fb.com> > --- This selftests breaks test_progs-no_alu32 ([0], the error log is super long and can freeze browser; it looks like an infinite loop and BPF verifier just keeps reporting it until it runs out of 1mln instructions or something). Please check what's going on there, I can't land it as it is right now. [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 > tools/testing/selftests/bpf/README.rst | 38 +++++++++ > .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ > tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ > .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ > .../selftests/bpf/progs/bpf_tracing_net.h | 4 + > 5 files changed, 143 insertions(+) > create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c > [...] > + /* The name of the abstract UNIX domain socket starts > + * with '\0' and can contain '\0'. The null bytes > + * should be escaped as done in unix_seq_show(). > + */ > + int i, len; > + no_alu32 variant probably isn't happy about using int for this, it probably does << 32, >> 32 dance and loses track of actual value in the loop. You can try using u64 instead. > + len = unix_sk->addr->len - sizeof(short); > + > + BPF_SEQ_PRINTF(seq, " @"); > + > + /* unix_mkname() tests this upper bound. */ > + if (len < sizeof(struct sockaddr_un)) > + for (i = 1; i < len; i++) if you move above if inside the loop to break out of the loop, does it change how Clang generates code? for (i = 1; i < len i++) { if (i >= sizeof(struct sockaddr_un)) break; BPF_SEQ_PRINTF(...); } > + BPF_SEQ_PRINTF(seq, "%c", > + unix_sk->addr->name->sun_path[i] ?: > + '@'); > + } > + } > + > + BPF_SEQ_PRINTF(seq, "\n"); > + > + return 0; > +} > diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > index 3af0998a0623..eef5646ddb19 100644 > --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > @@ -5,6 +5,10 @@ > #define AF_INET 2 > #define AF_INET6 10 > > +#define __SO_ACCEPTCON (1 << 16) > +#define UNIX_HASH_SIZE 256 > +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) > + > #define SOL_TCP 6 > #define TCP_CONGESTION 13 > #define TCP_CA_NAME_MAX 16 > -- > 2.30.2 >
From: Andrii Nakryiko <andrii.nakryiko@gmail.com> Date: Fri, 13 Aug 2021 16:25:53 -0700 > On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > > > The iterator can output almost the same result compared to /proc/net/unix. > > The header line is aligned, and the Inode column uses "%8lu" because "%5lu" > > can be easily overflown. > > > > # cat /sys/fs/bpf/unix > > Num RefCount Protocol Flags Type St Inode Path > > It's totally my OCD, but why the column name is not aligned with > values? I mean the "Inode" column. It's left aligned, but values > (numbers) are right-aligned? I'd fix that while applying, but I can't > apply due to selftests failures, so please take a look. Ah, honestly, I've felt something strange about the column... will fix it! > > > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > # cat /proc/net/unix > > Num RefCount Protocol Flags Type St Inode Path > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to > > Yonghong Song for analysing and fixing. > > > > [0] https://reviews.llvm.org/D107483 > > > > Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > > Acked-by: Yonghong Song <yhs@fb.com> > > --- > > This selftests breaks test_progs-no_alu32 ([0], the error log is super > long and can freeze browser; it looks like an infinite loop and BPF > verifier just keeps reporting it until it runs out of 1mln > instructions or something). Please check what's going on there, I > can't land it as it is right now. > > [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 > > > > tools/testing/selftests/bpf/README.rst | 38 +++++++++ > > .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ > > tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ > > .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ > > .../selftests/bpf/progs/bpf_tracing_net.h | 4 + > > 5 files changed, 143 insertions(+) > > create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c > > > > [...] > > > + /* The name of the abstract UNIX domain socket starts > > + * with '\0' and can contain '\0'. The null bytes > > + * should be escaped as done in unix_seq_show(). > > + */ > > + int i, len; > > + > > no_alu32 variant probably isn't happy about using int for this, it > probably does << 32, >> 32 dance and loses track of actual value in > the loop. You can try using u64 instead. Sorry, I missed the no_alu32 test. Changing int to __u64 fixed the error, thanks! > > > + len = unix_sk->addr->len - sizeof(short); > > + > > + BPF_SEQ_PRINTF(seq, " @"); > > + > > + /* unix_mkname() tests this upper bound. */ > > + if (len < sizeof(struct sockaddr_un)) > > + for (i = 1; i < len; i++) > > if you move above if inside the loop to break out of the loop, does it > change how Clang generates code? > > for (i = 1; i < len i++) { > if (i >= sizeof(struct sockaddr_un)) > break; > BPF_SEQ_PRINTF(...); > } Yes, but there seems little defference. Which is preferable? ---8<--- before (for inside if) <- -> after (if inside loop) 96: 07 08 00 00 fe ff ff ff r8 += -2 | ; for (i = 1; i < len; i++) { ; if (len < sizeof(struct sockaddr_un)) | 97: bf 81 00 00 00 00 00 00 r1 = r8 97: 25 08 10 00 6d 00 00 00 if r8 > 109 goto +16 <LBB0_21> | 98: 07 01 00 00 fc ff ff ff r1 += -4 ; for (i = 1; i < len; i++) | 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> 98: a5 08 0f 00 02 00 00 00 if r8 < 2 goto +15 <LBB0_21> | 100: 07 08 00 00 fe ff ff ff r8 += -2 99: b7 09 00 00 01 00 00 00 r9 = 1 | 101: b7 09 00 00 01 00 00 00 r9 = 1 100: 05 00 16 00 00 00 00 00 goto +22 <LBB0_18> | 102: b7 06 00 00 02 00 00 00 r6 = 2 | 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> ... 111: 85 00 00 00 7e 00 00 00 call 126 | 113: b4 05 00 00 08 00 00 00 w5 = 8 ; for (i = 1; i < len; i++) | 114: 85 00 00 00 7e 00 00 00 call 126 112: 07 09 00 00 01 00 00 00 r9 += 1 | ; for (i = 1; i < len; i++) { 113: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> | 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> > 116: 07 09 00 00 01 00 00 00 r9 += 1 > ; for (i = 1; i < len; i++) { > 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> ---8<--- > > > > + BPF_SEQ_PRINTF(seq, "%c", > > + unix_sk->addr->name->sun_path[i] ?: > > + '@'); > > + } > > + } > > + > > + BPF_SEQ_PRINTF(seq, "\n"); > > + > > + return 0; > > +} > > diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > index 3af0998a0623..eef5646ddb19 100644 > > --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > @@ -5,6 +5,10 @@ > > #define AF_INET 2 > > #define AF_INET6 10 > > > > +#define __SO_ACCEPTCON (1 << 16) > > +#define UNIX_HASH_SIZE 256 > > +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) > > + > > #define SOL_TCP 6 > > #define TCP_CONGESTION 13 > > #define TCP_CA_NAME_MAX 16 > > -- > > 2.30.2 > >
On Fri, Aug 13, 2021 at 5:21 PM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > From: Andrii Nakryiko <andrii.nakryiko@gmail.com> > Date: Fri, 13 Aug 2021 16:25:53 -0700 > > On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > > > > > The iterator can output almost the same result compared to /proc/net/unix. > > > The header line is aligned, and the Inode column uses "%8lu" because "%5lu" > > > can be easily overflown. > > > > > > # cat /sys/fs/bpf/unix > > > Num RefCount Protocol Flags Type St Inode Path > > > > It's totally my OCD, but why the column name is not aligned with > > values? I mean the "Inode" column. It's left aligned, but values > > (numbers) are right-aligned? I'd fix that while applying, but I can't > > apply due to selftests failures, so please take a look. > > Ah, honestly, I've felt something strange about the column... will fix it! > > > > > > > > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > > > # cat /proc/net/unix > > > Num RefCount Protocol Flags Type St Inode Path > > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > > > Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to > > > Yonghong Song for analysing and fixing. > > > > > > [0] https://reviews.llvm.org/D107483 > > > > > > Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > > > Acked-by: Yonghong Song <yhs@fb.com> > > > --- > > > > This selftests breaks test_progs-no_alu32 ([0], the error log is super > > long and can freeze browser; it looks like an infinite loop and BPF > > verifier just keeps reporting it until it runs out of 1mln > > instructions or something). Please check what's going on there, I > > can't land it as it is right now. > > > > [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 > > > > > > > tools/testing/selftests/bpf/README.rst | 38 +++++++++ > > > .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ > > > tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ > > > .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ > > > .../selftests/bpf/progs/bpf_tracing_net.h | 4 + > > > 5 files changed, 143 insertions(+) > > > create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c > > > > > > > [...] > > > > > + /* The name of the abstract UNIX domain socket starts > > > + * with '\0' and can contain '\0'. The null bytes > > > + * should be escaped as done in unix_seq_show(). > > > + */ > > > + int i, len; > > > + > > > > no_alu32 variant probably isn't happy about using int for this, it > > probably does << 32, >> 32 dance and loses track of actual value in > > the loop. You can try using u64 instead. > > Sorry, I missed the no_alu32 test. > Changing int to __u64 fixed the error, thanks! > > > > > > > + len = unix_sk->addr->len - sizeof(short); > > > + > > > + BPF_SEQ_PRINTF(seq, " @"); > > > + > > > + /* unix_mkname() tests this upper bound. */ > > > + if (len < sizeof(struct sockaddr_un)) > > > + for (i = 1; i < len; i++) > > > > if you move above if inside the loop to break out of the loop, does it > > change how Clang generates code? > > > > for (i = 1; i < len i++) { > > if (i >= sizeof(struct sockaddr_un)) > > break; > > BPF_SEQ_PRINTF(...); > > } > > Yes, but there seems little defference. > Which is preferable? > > ---8<--- > before (for inside if) <- -> after (if inside loop) > 96: 07 08 00 00 fe ff ff ff r8 += -2 | ; for (i = 1; i < len; i++) { > ; if (len < sizeof(struct sockaddr_un)) | 97: bf 81 00 00 00 00 00 00 r1 = r8 > 97: 25 08 10 00 6d 00 00 00 if r8 > 109 goto +16 <LBB0_21> | 98: 07 01 00 00 fc ff ff ff r1 += -4 > ; for (i = 1; i < len; i++) | 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> > 98: a5 08 0f 00 02 00 00 00 if r8 < 2 goto +15 <LBB0_21> | 100: 07 08 00 00 fe ff ff ff r8 += -2 > 99: b7 09 00 00 01 00 00 00 r9 = 1 | 101: b7 09 00 00 01 00 00 00 r9 = 1 > 100: 05 00 16 00 00 00 00 00 goto +22 <LBB0_18> | 102: b7 06 00 00 02 00 00 00 r6 = 2 > | 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> > ... > 111: 85 00 00 00 7e 00 00 00 call 126 | 113: b4 05 00 00 08 00 00 00 w5 = 8 > ; for (i = 1; i < len; i++) | 114: 85 00 00 00 7e 00 00 00 call 126 > 112: 07 09 00 00 01 00 00 00 r9 += 1 | ; for (i = 1; i < len; i++) { > 113: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> | 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> > > 116: 07 09 00 00 01 00 00 00 r9 += 1 > > ; for (i = 1; i < len; i++) { > > 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> > ---8<--- > Have you tried running the variant I proposed on Clang without Yonghong's recent fix? I wonder if it works without that fix (not that there is anything wrong about the fix, but if we can avoid depending on it, it would be great). > > > > > > > > + BPF_SEQ_PRINTF(seq, "%c", > > > + unix_sk->addr->name->sun_path[i] ?: > > > + '@'); > > > + } > > > + } > > > + > > > + BPF_SEQ_PRINTF(seq, "\n"); > > > + > > > + return 0; > > > +} > > > diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > index 3af0998a0623..eef5646ddb19 100644 > > > --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > @@ -5,6 +5,10 @@ > > > #define AF_INET 2 > > > #define AF_INET6 10 > > > > > > +#define __SO_ACCEPTCON (1 << 16) > > > +#define UNIX_HASH_SIZE 256 > > > +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) > > > + > > > #define SOL_TCP 6 > > > #define TCP_CONGESTION 13 > > > #define TCP_CA_NAME_MAX 16 > > > -- > > > 2.30.2 > > >
From: Andrii Nakryiko <andrii.nakryiko@gmail.com> Date: Fri, 13 Aug 2021 17:26:05 -0700 > On Fri, Aug 13, 2021 at 5:21 PM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > > > From: Andrii Nakryiko <andrii.nakryiko@gmail.com> > > Date: Fri, 13 Aug 2021 16:25:53 -0700 > > > On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > > > > > > > > The iterator can output almost the same result compared to /proc/net/unix. > > > > The header line is aligned, and the Inode column uses "%8lu" because "%5lu" > > > > can be easily overflown. > > > > > > > > # cat /sys/fs/bpf/unix > > > > Num RefCount Protocol Flags Type St Inode Path > > > > > > It's totally my OCD, but why the column name is not aligned with > > > values? I mean the "Inode" column. It's left aligned, but values > > > (numbers) are right-aligned? I'd fix that while applying, but I can't > > > apply due to selftests failures, so please take a look. > > > > Ah, honestly, I've felt something strange about the column... will fix it! > > > > > > > > > > > > > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > > > > > # cat /proc/net/unix > > > > Num RefCount Protocol Flags Type St Inode Path > > > > ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > > > > ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > > > > > > > > Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to > > > > Yonghong Song for analysing and fixing. > > > > > > > > [0] https://reviews.llvm.org/D107483 > > > > > > > > Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > > > > Acked-by: Yonghong Song <yhs@fb.com> > > > > --- > > > > > > This selftests breaks test_progs-no_alu32 ([0], the error log is super > > > long and can freeze browser; it looks like an infinite loop and BPF > > > verifier just keeps reporting it until it runs out of 1mln > > > instructions or something). Please check what's going on there, I > > > can't land it as it is right now. > > > > > > [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 > > > > > > > > > > tools/testing/selftests/bpf/README.rst | 38 +++++++++ > > > > .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ > > > > tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ > > > > .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ > > > > .../selftests/bpf/progs/bpf_tracing_net.h | 4 + > > > > 5 files changed, 143 insertions(+) > > > > create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c > > > > > > > > > > [...] > > > > > > > + /* The name of the abstract UNIX domain socket starts > > > > + * with '\0' and can contain '\0'. The null bytes > > > > + * should be escaped as done in unix_seq_show(). > > > > + */ > > > > + int i, len; > > > > + > > > > > > no_alu32 variant probably isn't happy about using int for this, it > > > probably does << 32, >> 32 dance and loses track of actual value in > > > the loop. You can try using u64 instead. > > > > Sorry, I missed the no_alu32 test. > > Changing int to __u64 fixed the error, thanks! > > > > > > > > > > > + len = unix_sk->addr->len - sizeof(short); > > > > + > > > > + BPF_SEQ_PRINTF(seq, " @"); > > > > + > > > > + /* unix_mkname() tests this upper bound. */ > > > > + if (len < sizeof(struct sockaddr_un)) > > > > + for (i = 1; i < len; i++) > > > > > > if you move above if inside the loop to break out of the loop, does it > > > change how Clang generates code? > > > > > > for (i = 1; i < len i++) { > > > if (i >= sizeof(struct sockaddr_un)) > > > break; > > > BPF_SEQ_PRINTF(...); > > > } > > > > Yes, but there seems little defference. > > Which is preferable? > > > > ---8<--- > > before (for inside if) <- -> after (if inside loop) > > 96: 07 08 00 00 fe ff ff ff r8 += -2 | ; for (i = 1; i < len; i++) { > > ; if (len < sizeof(struct sockaddr_un)) | 97: bf 81 00 00 00 00 00 00 r1 = r8 > > 97: 25 08 10 00 6d 00 00 00 if r8 > 109 goto +16 <LBB0_21> | 98: 07 01 00 00 fc ff ff ff r1 += -4 > > ; for (i = 1; i < len; i++) | 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> > > 98: a5 08 0f 00 02 00 00 00 if r8 < 2 goto +15 <LBB0_21> | 100: 07 08 00 00 fe ff ff ff r8 += -2 > > 99: b7 09 00 00 01 00 00 00 r9 = 1 | 101: b7 09 00 00 01 00 00 00 r9 = 1 > > 100: 05 00 16 00 00 00 00 00 goto +22 <LBB0_18> | 102: b7 06 00 00 02 00 00 00 r6 = 2 > > | 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> > > ... > > 111: 85 00 00 00 7e 00 00 00 call 126 | 113: b4 05 00 00 08 00 00 00 w5 = 8 > > ; for (i = 1; i < len; i++) | 114: 85 00 00 00 7e 00 00 00 call 126 > > 112: 07 09 00 00 01 00 00 00 r9 += 1 | ; for (i = 1; i < len; i++) { > > 113: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> | 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> > > > 116: 07 09 00 00 01 00 00 00 r9 += 1 > > > ; for (i = 1; i < len; i++) { > > > 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> > > ---8<--- > > > > Have you tried running the variant I proposed on Clang without > Yonghong's recent fix? I wonder if it works without that fix (not that > there is anything wrong about the fix, but if we can avoid depending > on it, it would be great). It was with the fix. I rebuilt LLVM without the fix, then the if-inside-for code worked well :) There was no transformation from '<' to '!='. I'll drop the change in README and respin with your suggestion soon. ---8<--- ; for (i = 1; i < len; i++) { 97: bf 81 00 00 00 00 00 00 r1 = r8 98: 07 01 00 00 fc ff ff ff r1 += -4 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> 100: 07 08 00 00 fe ff ff ff r8 += -2 101: b7 09 00 00 01 00 00 00 r9 = 1 102: b7 06 00 00 02 00 00 00 r6 = 2 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> ... ; for (i = 1; i < len; i++) { 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> 116: 07 09 00 00 01 00 00 00 r9 += 1 ; for (i = 1; i < len; i++) { 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> ---8<--- > > > > > > > > > > > > > + BPF_SEQ_PRINTF(seq, "%c", > > > > + unix_sk->addr->name->sun_path[i] ?: > > > > + '@'); > > > > + } > > > > + } > > > > + > > > > + BPF_SEQ_PRINTF(seq, "\n"); > > > > + > > > > + return 0; > > > > +} > > > > diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > > index 3af0998a0623..eef5646ddb19 100644 > > > > --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > > +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > > > > @@ -5,6 +5,10 @@ > > > > #define AF_INET 2 > > > > #define AF_INET6 10 > > > > > > > > +#define __SO_ACCEPTCON (1 << 16) > > > > +#define UNIX_HASH_SIZE 256 > > > > +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) > > > > + > > > > #define SOL_TCP 6 > > > > #define TCP_CONGESTION 13 > > > > #define TCP_CA_NAME_MAX 16 > > > > -- > > > > 2.30.2
On 8/13/21 5:21 PM, Kuniyuki Iwashima wrote: > From: Andrii Nakryiko <andrii.nakryiko@gmail.com> > Date: Fri, 13 Aug 2021 16:25:53 -0700 >> On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: >>> >>> The iterator can output almost the same result compared to /proc/net/unix. >>> The header line is aligned, and the Inode column uses "%8lu" because "%5lu" >>> can be easily overflown. >>> >>> # cat /sys/fs/bpf/unix >>> Num RefCount Protocol Flags Type St Inode Path >> >> It's totally my OCD, but why the column name is not aligned with >> values? I mean the "Inode" column. It's left aligned, but values >> (numbers) are right-aligned? I'd fix that while applying, but I can't >> apply due to selftests failures, so please take a look. > > Ah, honestly, I've felt something strange about the column... will fix it! > > >> >> >>> ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer >>> ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ >>> >>> # cat /proc/net/unix >>> Num RefCount Protocol Flags Type St Inode Path >>> ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer >>> ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ >>> >>> Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to >>> Yonghong Song for analysing and fixing. >>> >>> [0] https://reviews.llvm.org/D107483 >>> >>> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> >>> Acked-by: Yonghong Song <yhs@fb.com> >>> --- >> >> This selftests breaks test_progs-no_alu32 ([0], the error log is super >> long and can freeze browser; it looks like an infinite loop and BPF >> verifier just keeps reporting it until it runs out of 1mln >> instructions or something). Please check what's going on there, I >> can't land it as it is right now. >> >> [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 >> >> >>> tools/testing/selftests/bpf/README.rst | 38 +++++++++ >>> .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ >>> tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ >>> .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ >>> .../selftests/bpf/progs/bpf_tracing_net.h | 4 + >>> 5 files changed, 143 insertions(+) >>> create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c >>> >> >> [...] >> >>> + /* The name of the abstract UNIX domain socket starts >>> + * with '\0' and can contain '\0'. The null bytes >>> + * should be escaped as done in unix_seq_show(). >>> + */ >>> + int i, len; >>> + >> >> no_alu32 variant probably isn't happy about using int for this, it >> probably does << 32, >> 32 dance and loses track of actual value in >> the loop. You can try using u64 instead. > > Sorry, I missed the no_alu32 test. > Changing int to __u64 fixed the error, thanks! Indeed for no_alu32, the index has << 32 and >> 32, which makes verifier *equivalent* register tracking not effective, see below: 96: r1 = r8 97: r1 <<= 32 98: r2 = r1 99: r2 >>= 32 100: if r2 > 109 goto +19 <LBB0_21> 101: r1 s>>= 32 102: if r1 s< 2 goto +17 <LBB0_21> 103: r9 = 1 104: r8 <<= 32 105: r8 >>= 32 Because these shifting, r1/r2/r8 equivalence cannot be easily established, so verifier ends with conservative r8 and cannot verify program successfully. Using __u64 for 'i' and 'len', the upper bound is directly tested: 98: if r8 > 109 goto +16 <LBB0_21> 99: if r8 < 2 goto +15 <LBB0_21> and verifier is very happy with this. > > >> >>> + len = unix_sk->addr->len - sizeof(short); >>> + >>> + BPF_SEQ_PRINTF(seq, " @"); >>> + >>> + /* unix_mkname() tests this upper bound. */ >>> + if (len < sizeof(struct sockaddr_un)) >>> + for (i = 1; i < len; i++) >> >> if you move above if inside the loop to break out of the loop, does it >> change how Clang generates code? >> >> for (i = 1; i < len i++) { >> if (i >= sizeof(struct sockaddr_un)) >> break; >> BPF_SEQ_PRINTF(...); >> } > > Yes, but there seems little defference. > Which is preferable? > > ---8<--- > before (for inside if) <- -> after (if inside loop) > 96: 07 08 00 00 fe ff ff ff r8 += -2 | ; for (i = 1; i < len; i++) { > ; if (len < sizeof(struct sockaddr_un)) | 97: bf 81 00 00 00 00 00 00 r1 = r8 > 97: 25 08 10 00 6d 00 00 00 if r8 > 109 goto +16 <LBB0_21> | 98: 07 01 00 00 fc ff ff ff r1 += -4 > ; for (i = 1; i < len; i++) | 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> > 98: a5 08 0f 00 02 00 00 00 if r8 < 2 goto +15 <LBB0_21> | 100: 07 08 00 00 fe ff ff ff r8 += -2 > 99: b7 09 00 00 01 00 00 00 r9 = 1 | 101: b7 09 00 00 01 00 00 00 r9 = 1 > 100: 05 00 16 00 00 00 00 00 goto +22 <LBB0_18> | 102: b7 06 00 00 02 00 00 00 r6 = 2 > | 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> > ... > 111: 85 00 00 00 7e 00 00 00 call 126 | 113: b4 05 00 00 08 00 00 00 w5 = 8 > ; for (i = 1; i < len; i++) | 114: 85 00 00 00 7e 00 00 00 call 126 > 112: 07 09 00 00 01 00 00 00 r9 += 1 | ; for (i = 1; i < len; i++) { > 113: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> | 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> > > 116: 07 09 00 00 01 00 00 00 r9 += 1 > > ; for (i = 1; i < len; i++) { > > 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> > ---8<--- > > >> >> >>> + BPF_SEQ_PRINTF(seq, "%c", >>> + unix_sk->addr->name->sun_path[i] ?: >>> + '@'); >>> + } >>> + } >>> + >>> + BPF_SEQ_PRINTF(seq, "\n"); >>> + >>> + return 0; >>> +} >>> diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h >>> index 3af0998a0623..eef5646ddb19 100644 >>> --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h >>> +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h >>> @@ -5,6 +5,10 @@ >>> #define AF_INET 2 >>> #define AF_INET6 10 >>> >>> +#define __SO_ACCEPTCON (1 << 16) >>> +#define UNIX_HASH_SIZE 256 >>> +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) >>> + >>> #define SOL_TCP 6 >>> #define TCP_CONGESTION 13 >>> #define TCP_CA_NAME_MAX 16 >>> -- >>> 2.30.2 >>>
From: Yonghong Song <yhs@fb.com> Date: Sun, 15 Aug 2021 11:10:49 -0700 > On 8/13/21 5:21 PM, Kuniyuki Iwashima wrote: > > From: Andrii Nakryiko <andrii.nakryiko@gmail.com> > > Date: Fri, 13 Aug 2021 16:25:53 -0700 > >> On Thu, Aug 12, 2021 at 9:46 AM Kuniyuki Iwashima <kuniyu@amazon.co.jp> wrote: > >>> > >>> The iterator can output almost the same result compared to /proc/net/unix. > >>> The header line is aligned, and the Inode column uses "%8lu" because "%5lu" > >>> can be easily overflown. > >>> > >>> # cat /sys/fs/bpf/unix > >>> Num RefCount Protocol Flags Type St Inode Path > >> > >> It's totally my OCD, but why the column name is not aligned with > >> values? I mean the "Inode" column. It's left aligned, but values > >> (numbers) are right-aligned? I'd fix that while applying, but I can't > >> apply due to selftests failures, so please take a look. > > > > Ah, honestly, I've felt something strange about the column... will fix it! > > > > > >> > >> > >>> ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > >>> ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > >>> > >>> # cat /proc/net/unix > >>> Num RefCount Protocol Flags Type St Inode Path > >>> ffff963c06689800: 00000002 00000000 00010000 0001 01 18697 private/defer > >>> ffff963c7c979c00: 00000002 00000000 00000000 0001 01 598245 @Hello@World@ > >>> > >>> Note that this prog requires the patch ([0]) for LLVM code gen. Thanks to > >>> Yonghong Song for analysing and fixing. > >>> > >>> [0] https://reviews.llvm.org/D107483 > >>> > >>> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp> > >>> Acked-by: Yonghong Song <yhs@fb.com> > >>> --- > >> > >> This selftests breaks test_progs-no_alu32 ([0], the error log is super > >> long and can freeze browser; it looks like an infinite loop and BPF > >> verifier just keeps reporting it until it runs out of 1mln > >> instructions or something). Please check what's going on there, I > >> can't land it as it is right now. > >> > >> [0] https://github.com/kernel-patches/bpf/runs/3326071112?check_suite_focus=true#step:6:124288 > >> > >> > >>> tools/testing/selftests/bpf/README.rst | 38 +++++++++ > >>> .../selftests/bpf/prog_tests/bpf_iter.c | 16 ++++ > >>> tools/testing/selftests/bpf/progs/bpf_iter.h | 8 ++ > >>> .../selftests/bpf/progs/bpf_iter_unix.c | 77 +++++++++++++++++++ > >>> .../selftests/bpf/progs/bpf_tracing_net.h | 4 + > >>> 5 files changed, 143 insertions(+) > >>> create mode 100644 tools/testing/selftests/bpf/progs/bpf_iter_unix.c > >>> > >> > >> [...] > >> > >>> + /* The name of the abstract UNIX domain socket starts > >>> + * with '\0' and can contain '\0'. The null bytes > >>> + * should be escaped as done in unix_seq_show(). > >>> + */ > >>> + int i, len; > >>> + > >> > >> no_alu32 variant probably isn't happy about using int for this, it > >> probably does << 32, >> 32 dance and loses track of actual value in > >> the loop. You can try using u64 instead. > > > > Sorry, I missed the no_alu32 test. > > Changing int to __u64 fixed the error, thanks! > > Indeed for no_alu32, the index has << 32 and >> 32, which makes > verifier *equivalent* register tracking not effective, see below: > > 96: r1 = r8 > > 97: r1 <<= 32 > > 98: r2 = r1 > > 99: r2 >>= 32 > > 100: if r2 > 109 goto +19 <LBB0_21> > > 101: r1 s>>= 32 > > 102: if r1 s< 2 goto +17 <LBB0_21> > > 103: r9 = 1 > > 104: r8 <<= 32 > > 105: r8 >>= 32 > > Because these shifting, r1/r2/r8 equivalence cannot be > easily established, so verifier ends with conservative > r8 and cannot verify program successfully. > > Using __u64 for 'i' and 'len', the upper bound is directly > tested: > 98: if r8 > 109 goto +16 <LBB0_21> > > 99: if r8 < 2 goto +15 <LBB0_21> > and verifier is very happy with this. Thanks for explanation! I understand that the shift dance is to mimic the overflow of int because actually 64-bit register is allocated to 'i' and 32-bit operations cannot be used in no_alu32 test, so using __64 to remove the dance resolves it. > > > > > > >> > >>> + len = unix_sk->addr->len - sizeof(short); > >>> + > >>> + BPF_SEQ_PRINTF(seq, " @"); > >>> + > >>> + /* unix_mkname() tests this upper bound. */ > >>> + if (len < sizeof(struct sockaddr_un)) > >>> + for (i = 1; i < len; i++) > >> > >> if you move above if inside the loop to break out of the loop, does it > >> change how Clang generates code? > >> > >> for (i = 1; i < len i++) { > >> if (i >= sizeof(struct sockaddr_un)) > >> break; > >> BPF_SEQ_PRINTF(...); > >> } > > > > Yes, but there seems little defference. > > Which is preferable? > > > > ---8<--- > > before (for inside if) <- -> after (if inside loop) > > 96: 07 08 00 00 fe ff ff ff r8 += -2 | ; for (i = 1; i < len; i++) { > > ; if (len < sizeof(struct sockaddr_un)) | 97: bf 81 00 00 00 00 00 00 r1 = r8 > > 97: 25 08 10 00 6d 00 00 00 if r8 > 109 goto +16 <LBB0_21> | 98: 07 01 00 00 fc ff ff ff r1 += -4 > > ; for (i = 1; i < len; i++) | 99: 25 01 12 00 6b 00 00 00 if r1 > 107 goto +18 <LBB0_21> > > 98: a5 08 0f 00 02 00 00 00 if r8 < 2 goto +15 <LBB0_21> | 100: 07 08 00 00 fe ff ff ff r8 += -2 > > 99: b7 09 00 00 01 00 00 00 r9 = 1 | 101: b7 09 00 00 01 00 00 00 r9 = 1 > > 100: 05 00 16 00 00 00 00 00 goto +22 <LBB0_18> | 102: b7 06 00 00 02 00 00 00 r6 = 2 > > | 103: 05 00 17 00 00 00 00 00 goto +23 <LBB0_17> > > ... > > 111: 85 00 00 00 7e 00 00 00 call 126 | 113: b4 05 00 00 08 00 00 00 w5 = 8 > > ; for (i = 1; i < len; i++) | 114: 85 00 00 00 7e 00 00 00 call 126 > > 112: 07 09 00 00 01 00 00 00 r9 += 1 | ; for (i = 1; i < len; i++) { > > 113: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> | 115: 25 08 02 00 6d 00 00 00 if r8 > 109 goto +2 <LBB0_21> > > > 116: 07 09 00 00 01 00 00 00 r9 += 1 > > > ; for (i = 1; i < len; i++) { > > > 117: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_17> > > ---8<--- > > > > > >> > >> > >>> + BPF_SEQ_PRINTF(seq, "%c", > >>> + unix_sk->addr->name->sun_path[i] ?: > >>> + '@'); > >>> + } > >>> + } > >>> + > >>> + BPF_SEQ_PRINTF(seq, "\n"); > >>> + > >>> + return 0; > >>> +} > >>> diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > >>> index 3af0998a0623..eef5646ddb19 100644 > >>> --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > >>> +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h > >>> @@ -5,6 +5,10 @@ > >>> #define AF_INET 2 > >>> #define AF_INET6 10 > >>> > >>> +#define __SO_ACCEPTCON (1 << 16) > >>> +#define UNIX_HASH_SIZE 256 > >>> +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) > >>> + > >>> #define SOL_TCP 6 > >>> #define TCP_CONGESTION 13 > >>> #define TCP_CA_NAME_MAX 16 > >>> -- > >>> 2.30.2 > >>>
diff --git a/tools/testing/selftests/bpf/README.rst b/tools/testing/selftests/bpf/README.rst index 9b17f2867488..314a63b69399 100644 --- a/tools/testing/selftests/bpf/README.rst +++ b/tools/testing/selftests/bpf/README.rst @@ -228,3 +228,41 @@ To fix this issue, user newer libbpf. .. Links .. _clang reloc patch: https://reviews.llvm.org/D102712 .. _kernel llvm reloc: /Documentation/bpf/llvm_reloc.rst + +bpf_iter_unix.o test failure and LLVM optimisation +================================================== + +The ``bpf_iter/unix`` subtest requires `the fix`__ to prevent the LLVM compiler +from transforming the loop exit condition. + +Without the fix, the compiler generates the following code: + +.. code-block:: c + + ; for (i = 1; i < len; i++) + 110: 07 09 00 00 01 00 00 00 r9 += 1 + 111: 5d 98 09 00 00 00 00 00 if r8 != r9 goto +9 <LBB0_18> + +The loop exit condition, where the upper bound is not a constant, is +changed from ``<`` to ``!=``. Thus, the verifier always evaluates it as true, +misleading to an infinite loop. + +.. code-block:: c + + The sequence of 8193 jumps is too complex. + processed 196506 insns (limit 1000000) max_states_per_insn 4 total_states 1830 peak_states 1830 mark_read 3 + +The fix prevents the optimisation by estimating its cost higher in such a +case: + +.. code-block:: c + + ; for (i = 1; i < len; i++) + 110: 07 09 00 00 01 00 00 00 r9 += 1 + 111: ad 89 09 00 00 00 00 00 if r9 < r8 goto +9 <LBB0_18> + +The patch is available in LLVM 14 trunk and has been `backported`__ to the LLVM +13.x release branch. + +__ https://reviews.llvm.org/D107483 +__ https://bugs.llvm.org/show_bug.cgi?id=51363 diff --git a/tools/testing/selftests/bpf/prog_tests/bpf_iter.c b/tools/testing/selftests/bpf/prog_tests/bpf_iter.c index 1f1aade56504..77ac24b191d4 100644 --- a/tools/testing/selftests/bpf/prog_tests/bpf_iter.c +++ b/tools/testing/selftests/bpf/prog_tests/bpf_iter.c @@ -13,6 +13,7 @@ #include "bpf_iter_tcp6.skel.h" #include "bpf_iter_udp4.skel.h" #include "bpf_iter_udp6.skel.h" +#include "bpf_iter_unix.skel.h" #include "bpf_iter_test_kern1.skel.h" #include "bpf_iter_test_kern2.skel.h" #include "bpf_iter_test_kern3.skel.h" @@ -313,6 +314,19 @@ static void test_udp6(void) bpf_iter_udp6__destroy(skel); } +static void test_unix(void) +{ + struct bpf_iter_unix *skel; + + skel = bpf_iter_unix__open_and_load(); + if (!ASSERT_OK_PTR(skel, "bpf_iter_unix__open_and_load")) + return; + + do_dummy_read(skel->progs.dump_unix); + + bpf_iter_unix__destroy(skel); +} + /* The expected string is less than 16 bytes */ static int do_read_with_fd(int iter_fd, const char *expected, bool read_one_char) @@ -1255,6 +1269,8 @@ void test_bpf_iter(void) test_udp4(); if (test__start_subtest("udp6")) test_udp6(); + if (test__start_subtest("unix")) + test_unix(); if (test__start_subtest("anon")) test_anon_iter(false); if (test__start_subtest("anon-read-one-char")) diff --git a/tools/testing/selftests/bpf/progs/bpf_iter.h b/tools/testing/selftests/bpf/progs/bpf_iter.h index 3d83b185c4bc..8cfaeba1ddbf 100644 --- a/tools/testing/selftests/bpf/progs/bpf_iter.h +++ b/tools/testing/selftests/bpf/progs/bpf_iter.h @@ -12,6 +12,7 @@ #define tcp6_sock tcp6_sock___not_used #define bpf_iter__udp bpf_iter__udp___not_used #define udp6_sock udp6_sock___not_used +#define bpf_iter__unix bpf_iter__unix___not_used #define bpf_iter__bpf_map_elem bpf_iter__bpf_map_elem___not_used #define bpf_iter__bpf_sk_storage_map bpf_iter__bpf_sk_storage_map___not_used #define bpf_iter__sockmap bpf_iter__sockmap___not_used @@ -32,6 +33,7 @@ #undef tcp6_sock #undef bpf_iter__udp #undef udp6_sock +#undef bpf_iter__unix #undef bpf_iter__bpf_map_elem #undef bpf_iter__bpf_sk_storage_map #undef bpf_iter__sockmap @@ -103,6 +105,12 @@ struct udp6_sock { struct ipv6_pinfo inet6; } __attribute__((preserve_access_index)); +struct bpf_iter__unix { + struct bpf_iter_meta *meta; + struct unix_sock *unix_sk; + uid_t uid; +} __attribute__((preserve_access_index)); + struct bpf_iter__bpf_map_elem { struct bpf_iter_meta *meta; struct bpf_map *map; diff --git a/tools/testing/selftests/bpf/progs/bpf_iter_unix.c b/tools/testing/selftests/bpf/progs/bpf_iter_unix.c new file mode 100644 index 000000000000..0a50b610f6f3 --- /dev/null +++ b/tools/testing/selftests/bpf/progs/bpf_iter_unix.c @@ -0,0 +1,77 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright Amazon.com Inc. or its affiliates. */ +#include "bpf_iter.h" +#include "bpf_tracing_net.h" +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_endian.h> + +char _license[] SEC("license") = "GPL"; + +static long sock_i_ino(const struct sock *sk) +{ + const struct socket *sk_socket = sk->sk_socket; + const struct inode *inode; + unsigned long ino; + + if (!sk_socket) + return 0; + + inode = &container_of(sk_socket, struct socket_alloc, socket)->vfs_inode; + bpf_probe_read_kernel(&ino, sizeof(ino), &inode->i_ino); + return ino; +} + +SEC("iter/unix") +int dump_unix(struct bpf_iter__unix *ctx) +{ + struct unix_sock *unix_sk = ctx->unix_sk; + struct sock *sk = (struct sock *)unix_sk; + struct seq_file *seq; + __u32 seq_num; + + if (!unix_sk) + return 0; + + seq = ctx->meta->seq; + seq_num = ctx->meta->seq_num; + if (seq_num == 0) + BPF_SEQ_PRINTF(seq, "Num RefCount Protocol Flags Type St Inode Path\n"); + + BPF_SEQ_PRINTF(seq, "%pK: %08X %08X %08X %04X %02X %8lu", + unix_sk, + sk->sk_refcnt.refs.counter, + 0, + sk->sk_state == TCP_LISTEN ? __SO_ACCEPTCON : 0, + sk->sk_type, + sk->sk_socket ? + (sk->sk_state == TCP_ESTABLISHED ? SS_CONNECTED : SS_UNCONNECTED) : + (sk->sk_state == TCP_ESTABLISHED ? SS_CONNECTING : SS_DISCONNECTING), + sock_i_ino(sk)); + + if (unix_sk->addr) { + if (!UNIX_ABSTRACT(unix_sk)) { + BPF_SEQ_PRINTF(seq, " %s", unix_sk->addr->name->sun_path); + } else { + /* The name of the abstract UNIX domain socket starts + * with '\0' and can contain '\0'. The null bytes + * should be escaped as done in unix_seq_show(). + */ + int i, len; + + len = unix_sk->addr->len - sizeof(short); + + BPF_SEQ_PRINTF(seq, " @"); + + /* unix_mkname() tests this upper bound. */ + if (len < sizeof(struct sockaddr_un)) + for (i = 1; i < len; i++) + BPF_SEQ_PRINTF(seq, "%c", + unix_sk->addr->name->sun_path[i] ?: + '@'); + } + } + + BPF_SEQ_PRINTF(seq, "\n"); + + return 0; +} diff --git a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h index 3af0998a0623..eef5646ddb19 100644 --- a/tools/testing/selftests/bpf/progs/bpf_tracing_net.h +++ b/tools/testing/selftests/bpf/progs/bpf_tracing_net.h @@ -5,6 +5,10 @@ #define AF_INET 2 #define AF_INET6 10 +#define __SO_ACCEPTCON (1 << 16) +#define UNIX_HASH_SIZE 256 +#define UNIX_ABSTRACT(unix_sk) (unix_sk->addr->hash < UNIX_HASH_SIZE) + #define SOL_TCP 6 #define TCP_CONGESTION 13 #define TCP_CA_NAME_MAX 16