| 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: <ZOjUpt16gcTSGkdl@krava>
Date: Fri, 25 Aug 2023 18:19:50 +0200
From: Jiri Olsa <olsajiri@...il.com>
To: Rong Tao <rtoax@...mail.com>
Cc: olsajiri@...il.com, andrii@...nel.org, daniel@...earbox.net,
sdf@...gle.com, Rong Tao <rongtao@...tc.cn>,
Alexei Starovoitov <ast@...nel.org>,
Martin KaFai Lau <martin.lau@...ux.dev>,
Song Liu <song@...nel.org>,
Yonghong Song <yonghong.song@...ux.dev>,
John Fastabend <john.fastabend@...il.com>,
KP Singh <kpsingh@...nel.org>, Hao Luo <haoluo@...gle.com>,
Mykola Lysenko <mykolal@...com>, Shuah Khan <shuah@...nel.org>,
"open list:BPF [GENERAL] (Safe Dynamic Programs and Tools)"
<bpf@...r.kernel.org>, open list <linux-kernel@...r.kernel.org>,
"open list:KERNEL SELFTEST FRAMEWORK"
<linux-kselftest@...r.kernel.org>
Subject: Re: [PATCH bpf-next v7] selftests/bpf: trace_helpers.c: optimize
kallsyms cache
On Fri, Aug 25, 2023 at 06:36:24PM +0800, Rong Tao wrote:
> From: Rong Tao <rongtao@...tc.cn>
>
> Static ksyms often have problems because the number of symbols exceeds the
> MAX_SYMS limit. Like changing the MAX_SYMS from 300000 to 400000 in
> commit e76a014334a6("selftests/bpf: Bump and validate MAX_SYMS") solves
> the problem somewhat, but it's not the perfect way.
>
> This commit uses dynamic memory allocation, which completely solves the
> problem caused by the limitation of the number of kallsyms.
>
> Acked-by: Stanislav Fomichev <sdf@...gle.com>
> Signed-off-by: Rong Tao <rongtao@...tc.cn>
> ---
> v7: Fix __must_check macro.
> v6: https://lore.kernel.org/lkml/tencent_4A09A36F883A06EA428A593497642AF8AF08@qq.com/
> Apply libbpf_ensure_mem()
> v5: https://lore.kernel.org/lkml/tencent_0E9E1A1C0981678D5E7EA9E4BDBA8EE2200A@qq.com/
> Release the allocated memory once the load_kallsyms_refresh() upon error
> given it's dynamically allocated.
> v4: https://lore.kernel.org/lkml/tencent_59C74613113F0C728524B2A82FE5540A5E09@qq.com/
> Make sure most cases we don't need the realloc() path to begin with,
> and check strdup() return value.
> v3: https://lore.kernel.org/lkml/tencent_50B4B2622FE7546A5FF9464310650C008509@qq.com/
> Do not use structs and judge ksyms__add_symbol function return value.
> v2: https://lore.kernel.org/lkml/tencent_B655EE5E5D463110D70CD2846AB3262EED09@qq.com/
> Do the usual len/capacity scheme here to amortize the cost of realloc, and
> don't free symbols.
> v1: https://lore.kernel.org/lkml/tencent_AB461510B10CD484E0B2F62E3754165F2909@qq.com/
> ---
> samples/bpf/Makefile | 2 +
> tools/testing/selftests/bpf/trace_helpers.c | 58 ++++++++++++++++-----
> 2 files changed, 48 insertions(+), 12 deletions(-)
>
> diff --git a/samples/bpf/Makefile b/samples/bpf/Makefile
> index 4ccf4236031c..0cd45c42af2f 100644
> --- a/samples/bpf/Makefile
> +++ b/samples/bpf/Makefile
> @@ -175,6 +175,7 @@ TPROGS_CFLAGS += -I$(srctree)/tools/testing/selftests/bpf/
> TPROGS_CFLAGS += -I$(LIBBPF_INCLUDE)
> TPROGS_CFLAGS += -I$(srctree)/tools/include
> TPROGS_CFLAGS += -I$(srctree)/tools/perf
> +TPROGS_CFLAGS += -I$(srctree)/tools/lib
> TPROGS_CFLAGS += -DHAVE_ATTR_TEST=0
>
> ifdef SYSROOT
> @@ -314,6 +315,7 @@ XDP_SAMPLE_CFLAGS += -Wall -O2 \
>
> $(obj)/$(XDP_SAMPLE): TPROGS_CFLAGS = $(XDP_SAMPLE_CFLAGS)
> $(obj)/$(XDP_SAMPLE): $(src)/xdp_sample_user.h $(src)/xdp_sample_shared.h
> +$(obj)/$(TRACE_HELPERS): TPROGS_CFLAGS := $(TPROGS_CFLAGS) -D__must_check=
please add some comment why we did that, perhaps like:
+# Override includes for trace_helpers.o because __must_check won't be
+# defined in our include path.
+$(obj)/$(TRACE_HELPERS): TPROGS_CFLAGS := $(TPROGS_CFLAGS) -D__must_check=
>
> -include $(BPF_SAMPLES_PATH)/Makefile.target
>
> diff --git a/tools/testing/selftests/bpf/trace_helpers.c b/tools/testing/selftests/bpf/trace_helpers.c
> index f83d9f65c65b..d62ab3b77153 100644
> --- a/tools/testing/selftests/bpf/trace_helpers.c
> +++ b/tools/testing/selftests/bpf/trace_helpers.c
> @@ -14,13 +14,44 @@
> #include <linux/limits.h>
> #include <libelf.h>
> #include <gelf.h>
> +#include "bpf/libbpf_internal.h"
>
> #define TRACEFS_PIPE "/sys/kernel/tracing/trace_pipe"
> #define DEBUGFS_PIPE "/sys/kernel/debug/tracing/trace_pipe"
>
> -#define MAX_SYMS 400000
> -static struct ksym syms[MAX_SYMS];
> -static int sym_cnt;
> +static struct ksym *syms;
> +static size_t sym_cap;
> +static size_t sym_cnt;
> +
> +static int ksyms__add_symbol(const char *name, unsigned long addr)
> +{
> + void *tmp;
> +
> + tmp = strdup(name);
> + if (!tmp)
> + return -ENOMEM;
> + syms[sym_cnt].addr = addr;
> + syms[sym_cnt].name = tmp;
> +
> + sym_cnt++;
> +
> + return 0;
> +}
> +
> +static void ksyms__free(void)
> +{
> + unsigned int i;
> +
> + if (!syms)
> + return;
> +
> + for (i = 0; i < sym_cnt; i++)
> + free(syms[i].name);
> + free(syms);
> + syms = NULL;
> + sym_cnt = 0;
> + sym_cap = 0;
> +}
>
> static int ksym_cmp(const void *p1, const void *p2)
> {
> @@ -33,9 +64,7 @@ int load_kallsyms_refresh(void)
> char func[256], buf[256];
> char symbol;
> void *addr;
> - int i = 0;
> -
> - sym_cnt = 0;
you need to release all symbols in here, for static version it was
enough to zero the sym_cnt value, but we need to release current
symbols now
also after discussing with Daniel we found there's a problem with
paralel run where both load_kallsyms and load_kallsyms_refresh and
other helpers like ksym_search are racy because another test can
just drop the cache
that was problem even before, but I wonder now it will expose more
because we will touch freed pointer instead of accessing same static
array
perhaps we could fix that with 2 sets of functions:
- load_kallsyms
ksym_search
ksym_get_addr
that work like now on global syms pointer, perhaps we should
initialize it before we run tests or use mutex for init
- struct ksym *load_kallsyms_local()
ksym_search_local(struct ksym *syms, ...)
ksym_get_addr_local(struct ksym *syms, ...)
that work on local ksyms cache and the test frees it at the end,
it would be used in all tests that need updated kallsyms data and
use load_kallsyms_refresh for that
jirka
> + int ret;
>
> f = fopen("/proc/kallsyms", "r");
> if (!f)
> @@ -46,17 +75,22 @@ int load_kallsyms_refresh(void)
> break;
> if (!addr)
> continue;
> - if (i >= MAX_SYMS)
> - return -EFBIG;
>
> - syms[i].addr = (long) addr;
> - syms[i].name = strdup(func);
> - i++;
> + ret = libbpf_ensure_mem((void **) &syms, &sym_cap,
> + sizeof(struct ksym), sym_cnt + 1);
> + if (ret)
> + goto error;
> + ret = ksyms__add_symbol(func, (unsigned long)addr);
> + if (ret)
> + goto error;
> }
> fclose(f);
> - sym_cnt = i;
> qsort(syms, sym_cnt, sizeof(struct ksym), ksym_cmp);
> return 0;
> +
> +error:
> + ksyms__free();
> + return ret;
> }
>
> int load_kallsyms(void)
> --
> 2.39.3
>
Powered by blists - more mailing lists