| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAEf4BzaWtO4-JE=++_64y01aWGGBnUSpw_pbQf79rFz_oVhU-Q@mail.gmail.com>
Date: Wed, 13 Aug 2025 16:52:47 -0700
From: Andrii Nakryiko <andrii.nakryiko@...il.com>
To: Jiawei Zhao <phoenix500526@....com>
Cc: ast@...nel.org, daniel@...earbox.net, andrii@...nel.org,
yonghong.song@...ux.dev, bpf@...r.kernel.org, linux-kselftest@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH bpf-next v8 1/2] libbpf: fix USDT SIB argument handling
causing unrecognized register error
On Wed, Aug 6, 2025 at 7:35 PM Jiawei Zhao <phoenix500526@....com> wrote:
>
> On x86-64, USDT arguments can be specified using Scale-Index-Base (SIB)
> addressing, e.g. "1@-96(%rbp,%rax,8)". The current USDT implementation
> in libbpf cannot parse this format, causing `bpf_program__attach_usdt()`
> to fail with -ENOENT (unrecognized register).
>
> This patch fixes this by implementing the necessary changes:
> - add correct handling for SIB-addressed arguments in `bpf_usdt_arg`.
> - add adaptive support to `__bpf_usdt_arg_type` and
> `__bpf_usdt_arg_spec` to represent SIB addressing parameters.
>
> Signed-off-by: Jiawei Zhao <phoenix500526@....com>
> ---
> tools/lib/bpf/usdt.bpf.h | 33 +++++++++++++++++++++++++++++-
> tools/lib/bpf/usdt.c | 43 ++++++++++++++++++++++++++++++++++------
> 2 files changed, 69 insertions(+), 7 deletions(-)
>
> diff --git a/tools/lib/bpf/usdt.bpf.h b/tools/lib/bpf/usdt.bpf.h
> index 2a7865c8e3fe..246513088c3a 100644
> --- a/tools/lib/bpf/usdt.bpf.h
> +++ b/tools/lib/bpf/usdt.bpf.h
> @@ -34,6 +34,7 @@ enum __bpf_usdt_arg_type {
> BPF_USDT_ARG_CONST,
> BPF_USDT_ARG_REG,
> BPF_USDT_ARG_REG_DEREF,
> + BPF_USDT_ARG_SIB,
> };
>
> struct __bpf_usdt_arg_spec {
> @@ -43,6 +44,10 @@ struct __bpf_usdt_arg_spec {
> enum __bpf_usdt_arg_type arg_type;
> /* offset of referenced register within struct pt_regs */
> short reg_off;
> + /* offset of index register in pt_regs, only used in SIB mode */
> + short idx_reg_off;
> + /* scale factor for index register, only used in SIB mode */
> + short scale;
I'd really prefer not to increase the size of __bpf_usdt_arg_spec and
not change its layout for all existing BPF_USDT_ARG_* modes just to
not have to worry about any backwards/forward compatibility issues.
Scale can be 1, 2,4, 8, is that right? Instead of using 2 bytes for
it, we should be able to use just 2 bits to represent bit shift (0, 1,
2, 3 should be enough).
We can carve out at least 3 bytes by making arg_type field into packed
single-byte enum (we'd need to be careful with big endian).
Then we can add idx_reg_off:12 and idx_scale_shift:4 somewhere between
arg_type and reg_off, taking 2 bytes in total.
We'll still be left with one byte to spare for the future (and there
are tricks we can do with arg_signed and arg_bitshift, but I'd not
touch them yet).
WDYT?
pw-bot: cr
> /* whether arg should be interpreted as signed value */
> bool arg_signed;
> /* number of bits that need to be cleared and, optionally,
> @@ -149,7 +154,7 @@ int bpf_usdt_arg(struct pt_regs *ctx, __u64 arg_num, long *res)
> {
> struct __bpf_usdt_spec *spec;
> struct __bpf_usdt_arg_spec *arg_spec;
> - unsigned long val;
> + unsigned long val, idx;
> int err, spec_id;
>
> *res = 0;
> @@ -202,6 +207,32 @@ int bpf_usdt_arg(struct pt_regs *ctx, __u64 arg_num, long *res)
> return err;
> #if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
> val >>= arg_spec->arg_bitshift;
> +#endif
> + break;
> + case BPF_USDT_ARG_SIB:
> + /* Arg is in memory addressed by SIB (Scale-Index-Base) mode
> + * (e.g., "-1@-96(%rbp,%rax,8)" in USDT arg spec). Register
> + * is identified like with BPF_USDT_ARG_SIB case, the offset
> + * is in arg_spec->val_off, the scale factor is in arg_spec->scale.
> + * Firstly, we fetch the base register contents and the index
> + * register contents from pt_regs. Secondly, we multiply the
> + * index register contents by the scale factor, then add the
> + * base address and the offset to get the final address. Finally,
> + * we do another user-space probe read to fetch argument value
> + * itself.
> + */
> + err = bpf_probe_read_kernel(&val, sizeof(val), (void *)ctx + arg_spec->reg_off);
> + if (err)
> + return err;
> + err = bpf_probe_read_kernel(&idx, sizeof(idx), (void *)ctx + arg_spec->idx_reg_off);
> + if (err)
> + return err;
> + err = bpf_probe_read_user(&val, sizeof(val),
> + (void *)val + idx * arg_spec->scale + arg_spec->val_off);
it might be just how gmail renders it, but please make sure that
wrapped argument is aligned with first argument on the previous line
> + if (err)
> + return err;
> +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
> + val >>= arg_spec->arg_bitshift;
> #endif
> break;
> default:
> diff --git a/tools/lib/bpf/usdt.c b/tools/lib/bpf/usdt.c
> index 4e4a52742b01..1f8b9e1c9819 100644
> --- a/tools/lib/bpf/usdt.c
> +++ b/tools/lib/bpf/usdt.c
> @@ -200,6 +200,7 @@ enum usdt_arg_type {
> USDT_ARG_CONST,
> USDT_ARG_REG,
> USDT_ARG_REG_DEREF,
> + USDT_ARG_SIB,
> };
>
> /* should match exactly struct __bpf_usdt_arg_spec from usdt.bpf.h */
> @@ -207,6 +208,8 @@ struct usdt_arg_spec {
> __u64 val_off;
> enum usdt_arg_type arg_type;
> short reg_off;
> + short idx_reg_off;
> + short scale;
> bool arg_signed;
> char arg_bitshift;
> };
> @@ -1283,11 +1286,39 @@ static int calc_pt_regs_off(const char *reg_name)
>
> static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec *arg, int *arg_sz)
> {
> - char reg_name[16];
> - int len, reg_off;
> - long off;
> + char reg_name[16] = {0}, idx_reg_name[16] = {0};
> + int len, reg_off, idx_reg_off, scale = 1;
> + long off = 0;
> +
> + if (sscanf(arg_str, " %d @ %ld ( %%%15[^,] , %%%15[^,] , %d ) %n",
> + arg_sz, &off, reg_name, idx_reg_name, &scale, &len) == 5 ||
see comment above about aligning wrapped argument list
> + sscanf(arg_str, " %d @ ( %%%15[^,] , %%%15[^,] , %d ) %n",
> + arg_sz, reg_name, idx_reg_name, &scale, &len) == 4 ||
> + sscanf(arg_str, " %d @ %ld ( %%%15[^,] , %%%15[^)] ) %n",
> + arg_sz, &off, reg_name, idx_reg_name, &len) == 4 ||
> + sscanf(arg_str, " %d @ ( %%%15[^,] , %%%15[^)] ) %n",
> + arg_sz, reg_name, idx_reg_name, &len) == 3
> + ) {
> + /* Scale Index Base case, e.g., 1@-96(%rbp,%rax,8)
> + * 1@(%rbp,%rax,8)
> + * 1@-96(%rbp,%rax)
> + * 1@(%rbp,%rax)
nit: let's list all variants at the same indentation level (and let's
use the more standard multi-level comment format)
/*
* Scale-Index-Base case:
* - 1@-96(%rbp,%rax,8)
* - 1@(%rbp,%rax,8)
* ...
*/
> + */
> + arg->arg_type = USDT_ARG_SIB;
> + arg->val_off = off;
> + arg->scale = scale;
> +
> + reg_off = calc_pt_regs_off(reg_name);
> + if (reg_off < 0)
> + return reg_off;
> + arg->reg_off = reg_off;
>
> - if (sscanf(arg_str, " %d @ %ld ( %%%15[^)] ) %n", arg_sz, &off, reg_name, &len) == 3) {
> + idx_reg_off = calc_pt_regs_off(idx_reg_name);
> + if (idx_reg_off < 0)
> + return idx_reg_off;
> + arg->idx_reg_off = idx_reg_off;
> + } else if (sscanf(arg_str, " %d @ %ld ( %%%15[^)] ) %n",
> + arg_sz, &off, reg_name, &len) == 3) {
> /* Memory dereference case, e.g., -4@-20(%rbp) */
> arg->arg_type = USDT_ARG_REG_DEREF;
> arg->val_off = off;
> @@ -1298,7 +1329,7 @@ static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec
> } else if (sscanf(arg_str, " %d @ ( %%%15[^)] ) %n", arg_sz, reg_name, &len) == 2) {
> /* Memory dereference case without offset, e.g., 8@(%rsp) */
> arg->arg_type = USDT_ARG_REG_DEREF;
> - arg->val_off = 0;
> + arg->val_off = off;
> reg_off = calc_pt_regs_off(reg_name);
> if (reg_off < 0)
> return reg_off;
> @@ -1306,7 +1337,7 @@ static int parse_usdt_arg(const char *arg_str, int arg_num, struct usdt_arg_spec
> } else if (sscanf(arg_str, " %d @ %%%15s %n", arg_sz, reg_name, &len) == 2) {
> /* Register read case, e.g., -4@...x */
> arg->arg_type = USDT_ARG_REG;
> - arg->val_off = 0;
> + arg->val_off = off;
why this change? it makes it seem like val_off might not be zero, for
no good reason...
>
> reg_off = calc_pt_regs_off(reg_name);
> if (reg_off < 0)
> --
> 2.43.0
>
>
Powered by blists - more mailing lists