From: Andrii Nakryiko <andrii@kernel.org>
Date: Tue, 14 Dec 2021 11:59:03 -0800
Subject: libbpf: Auto-bump RLIMIT_MEMLOCK if kernel needs it for BPF
Patch-mainline: v5.17-rc1
Git-commit: e542f2c4cd16d49392abf3349341d58153d3c603
References: jsc#PED-1368
The need to increase RLIMIT_MEMLOCK to do anything useful with BPF is
one of the first extremely frustrating gotchas that all new BPF users go
through and in some cases have to learn it a very hard way.
Luckily, starting with upstream Linux kernel version 5.11, BPF subsystem
dropped the dependency on memlock and uses memcg-based memory accounting
instead. Unfortunately, detecting memcg-based BPF memory accounting is
far from trivial (as can be evidenced by this patch), so in practice
most BPF applications still do unconditional RLIMIT_MEMLOCK increase.
As we move towards libbpf 1.0, it would be good to allow users to forget
about RLIMIT_MEMLOCK vs memcg and let libbpf do the sensible adjustment
automatically. This patch paves the way forward in this matter. Libbpf
will do feature detection of memcg-based accounting, and if detected,
will do nothing. But if the kernel is too old, just like BCC, libbpf
will automatically increase RLIMIT_MEMLOCK on behalf of user
application ([0]).
As this is technically a breaking change, during the transition period
applications have to opt into libbpf 1.0 mode by setting
LIBBPF_STRICT_AUTO_RLIMIT_MEMLOCK bit when calling
libbpf_set_strict_mode().
Libbpf allows to control the exact amount of set RLIMIT_MEMLOCK limit
with libbpf_set_memlock_rlim_max() API. Passing 0 will make libbpf do
nothing with RLIMIT_MEMLOCK. libbpf_set_memlock_rlim_max() has to be
called before the first bpf_prog_load(), bpf_btf_load(), or
bpf_object__load() call, otherwise it has no effect and will return
-EBUSY.
[0] Closes: https://github.com/libbpf/libbpf/issues/369
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211214195904.1785155-2-andrii@kernel.org
Acked-by: Shung-Hsi Yu <shung-hsi.yu@suse.com>
---
tools/lib/bpf/bpf.c | 81 ++++++++++++++++++++++++++++++++++++++++
tools/lib/bpf/bpf.h | 2
tools/lib/bpf/libbpf.c | 47 ++++-------------------
tools/lib/bpf/libbpf.map | 1
tools/lib/bpf/libbpf_internal.h | 39 +++++++++++++++++++
tools/lib/bpf/libbpf_legacy.h | 12 +++++
6 files changed, 143 insertions(+), 39 deletions(-)
--- a/tools/lib/bpf/bpf.c
+++ b/tools/lib/bpf/bpf.c
@@ -28,7 +28,9 @@
#include <asm/unistd.h>
#include <errno.h>
#include <linux/bpf.h>
+#include <linux/filter.h>
#include <limits.h>
+#include <sys/resource.h>
#include "bpf.h"
#include "libbpf.h"
#include "libbpf_internal.h"
@@ -94,6 +96,77 @@ static inline int sys_bpf_prog_load(unio
return fd;
}
+/* Probe whether kernel switched from memlock-based (RLIMIT_MEMLOCK) to
+ * memcg-based memory accounting for BPF maps and progs. This was done in [0].
+ * We use the support for bpf_ktime_get_coarse_ns() helper, which was added in
+ * the same 5.11 Linux release ([1]), to detect memcg-based accounting for BPF.
+ *
+ * [0] https://lore.kernel.org/bpf/20201201215900.3569844-1-guro@fb.com/
+ * [1] d05512618056 ("bpf: Add bpf_ktime_get_coarse_ns helper")
+ */
+int probe_memcg_account(void)
+{
+ const size_t prog_load_attr_sz = offsetofend(union bpf_attr, attach_btf_obj_fd);
+ struct bpf_insn insns[] = {
+ BPF_EMIT_CALL(BPF_FUNC_ktime_get_coarse_ns),
+ BPF_EXIT_INSN(),
+ };
+ size_t insn_cnt = sizeof(insns) / sizeof(insns[0]);
+ union bpf_attr attr;
+ int prog_fd;
+
+ /* attempt loading freplace trying to use custom BTF */
+ memset(&attr, 0, prog_load_attr_sz);
+ attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
+ attr.insns = ptr_to_u64(insns);
+ attr.insn_cnt = insn_cnt;
+ attr.license = ptr_to_u64("GPL");
+
+ prog_fd = sys_bpf_fd(BPF_PROG_LOAD, &attr, prog_load_attr_sz);
+ if (prog_fd >= 0) {
+ close(prog_fd);
+ return 1;
+ }
+ return 0;
+}
+
+static bool memlock_bumped;
+static rlim_t memlock_rlim = RLIM_INFINITY;
+
+int libbpf_set_memlock_rlim(size_t memlock_bytes)
+{
+ if (memlock_bumped)
+ return libbpf_err(-EBUSY);
+
+ memlock_rlim = memlock_bytes;
+ return 0;
+}
+
+int bump_rlimit_memlock(void)
+{
+ struct rlimit rlim;
+
+ /* this the default in libbpf 1.0, but for now user has to opt-in explicitly */
+ if (!(libbpf_mode & LIBBPF_STRICT_AUTO_RLIMIT_MEMLOCK))
+ return 0;
+
+ /* if kernel supports memcg-based accounting, skip bumping RLIMIT_MEMLOCK */
+ if (memlock_bumped || kernel_supports(NULL, FEAT_MEMCG_ACCOUNT))
+ return 0;
+
+ memlock_bumped = true;
+
+ /* zero memlock_rlim_max disables auto-bumping RLIMIT_MEMLOCK */
+ if (memlock_rlim == 0)
+ return 0;
+
+ rlim.rlim_cur = rlim.rlim_max = memlock_rlim;
+ if (setrlimit(RLIMIT_MEMLOCK, &rlim))
+ return -errno;
+
+ return 0;
+}
+
int bpf_map_create(enum bpf_map_type map_type,
const char *map_name,
__u32 key_size,
@@ -105,6 +178,8 @@ int bpf_map_create(enum bpf_map_type map
union bpf_attr attr;
int fd;
+ bump_rlimit_memlock();
+
memset(&attr, 0, attr_sz);
if (!OPTS_VALID(opts, bpf_map_create_opts))
@@ -251,6 +326,8 @@ int bpf_prog_load_v0_6_0(enum bpf_prog_t
union bpf_attr attr;
char *log_buf;
+ bump_rlimit_memlock();
+
if (!OPTS_VALID(opts, bpf_prog_load_opts))
return libbpf_err(-EINVAL);
@@ -456,6 +533,8 @@ int bpf_verify_program(enum bpf_prog_typ
union bpf_attr attr;
int fd;
+ bump_rlimit_memlock();
+
memset(&attr, 0, sizeof(attr));
attr.prog_type = type;
attr.insn_cnt = (__u32)insns_cnt;
@@ -1056,6 +1135,8 @@ int bpf_btf_load(const void *btf_data, s
__u32 log_level;
int fd;
+ bump_rlimit_memlock();
+
memset(&attr, 0, attr_sz);
if (!OPTS_VALID(opts, bpf_btf_load_opts))
--- a/tools/lib/bpf/bpf.h
+++ b/tools/lib/bpf/bpf.h
@@ -35,6 +35,8 @@
extern "C" {
#endif
+int libbpf_set_memlock_rlim(size_t memlock_bytes);
+
struct bpf_map_create_opts {
size_t sz; /* size of this struct for forward/backward compatibility */
--- a/tools/lib/bpf/libbpf.c
+++ b/tools/lib/bpf/libbpf.c
@@ -187,42 +187,6 @@ const char *libbpf_version_string(void)
#undef __S
}
-enum kern_feature_id {
- /* v4.14: kernel support for program & map names. */
- FEAT_PROG_NAME,
- /* v5.2: kernel support for global data sections. */
- FEAT_GLOBAL_DATA,
- /* BTF support */
- FEAT_BTF,
- /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
- FEAT_BTF_FUNC,
- /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
- FEAT_BTF_DATASEC,
- /* BTF_FUNC_GLOBAL is supported */
- FEAT_BTF_GLOBAL_FUNC,
- /* BPF_F_MMAPABLE is supported for arrays */
- FEAT_ARRAY_MMAP,
- /* kernel support for expected_attach_type in BPF_PROG_LOAD */
- FEAT_EXP_ATTACH_TYPE,
- /* bpf_probe_read_{kernel,user}[_str] helpers */
- FEAT_PROBE_READ_KERN,
- /* BPF_PROG_BIND_MAP is supported */
- FEAT_PROG_BIND_MAP,
- /* Kernel support for module BTFs */
- FEAT_MODULE_BTF,
- /* BTF_KIND_FLOAT support */
- FEAT_BTF_FLOAT,
- /* BPF perf link support */
- FEAT_PERF_LINK,
- /* BTF_KIND_DECL_TAG support */
- FEAT_BTF_DECL_TAG,
- /* BTF_KIND_TYPE_TAG support */
- FEAT_BTF_TYPE_TAG,
- __FEAT_CNT,
-};
-
-static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id);
-
enum reloc_type {
RELO_LD64,
RELO_CALL,
@@ -4352,6 +4316,10 @@ bpf_object__probe_loading(struct bpf_obj
if (obj->gen_loader)
return 0;
+ ret = bump_rlimit_memlock();
+ if (ret)
+ pr_warn("Failed to bump RLIMIT_MEMLOCK (err = %d), you might need to do it explicitly!\n", ret);
+
/* make sure basic loading works */
ret = bpf_prog_load(BPF_PROG_TYPE_SOCKET_FILTER, NULL, "GPL", insns, insn_cnt, NULL);
if (ret < 0)
@@ -4718,14 +4686,17 @@ static struct kern_feature_desc {
[FEAT_BTF_TYPE_TAG] = {
"BTF_KIND_TYPE_TAG support", probe_kern_btf_type_tag,
},
+ [FEAT_MEMCG_ACCOUNT] = {
+ "memcg-based memory accounting", probe_memcg_account,
+ },
};
-static bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
+bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id)
{
struct kern_feature_desc *feat = &feature_probes[feat_id];
int ret;
- if (obj->gen_loader)
+ if (obj && obj->gen_loader)
/* To generate loader program assume the latest kernel
* to avoid doing extra prog_load, map_create syscalls.
*/
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -427,4 +427,5 @@ LIBBPF_0.7.0 {
bpf_program__log_level;
bpf_program__set_log_buf;
bpf_program__set_log_level;
+ libbpf_set_memlock_rlim_max;
};
--- a/tools/lib/bpf/libbpf_internal.h
+++ b/tools/lib/bpf/libbpf_internal.h
@@ -291,6 +291,45 @@ static inline bool libbpf_validate_opts(
(opts)->sz - __off); \
})
+enum kern_feature_id {
+ /* v4.14: kernel support for program & map names. */
+ FEAT_PROG_NAME,
+ /* v5.2: kernel support for global data sections. */
+ FEAT_GLOBAL_DATA,
+ /* BTF support */
+ FEAT_BTF,
+ /* BTF_KIND_FUNC and BTF_KIND_FUNC_PROTO support */
+ FEAT_BTF_FUNC,
+ /* BTF_KIND_VAR and BTF_KIND_DATASEC support */
+ FEAT_BTF_DATASEC,
+ /* BTF_FUNC_GLOBAL is supported */
+ FEAT_BTF_GLOBAL_FUNC,
+ /* BPF_F_MMAPABLE is supported for arrays */
+ FEAT_ARRAY_MMAP,
+ /* kernel support for expected_attach_type in BPF_PROG_LOAD */
+ FEAT_EXP_ATTACH_TYPE,
+ /* bpf_probe_read_{kernel,user}[_str] helpers */
+ FEAT_PROBE_READ_KERN,
+ /* BPF_PROG_BIND_MAP is supported */
+ FEAT_PROG_BIND_MAP,
+ /* Kernel support for module BTFs */
+ FEAT_MODULE_BTF,
+ /* BTF_KIND_FLOAT support */
+ FEAT_BTF_FLOAT,
+ /* BPF perf link support */
+ FEAT_PERF_LINK,
+ /* BTF_KIND_DECL_TAG support */
+ FEAT_BTF_DECL_TAG,
+ /* BTF_KIND_TYPE_TAG support */
+ FEAT_BTF_TYPE_TAG,
+ /* memcg-based accounting for BPF maps and progs */
+ FEAT_MEMCG_ACCOUNT,
+ __FEAT_CNT,
+};
+
+int probe_memcg_account(void);
+bool kernel_supports(const struct bpf_object *obj, enum kern_feature_id feat_id);
+int bump_rlimit_memlock(void);
int parse_cpu_mask_str(const char *s, bool **mask, int *mask_sz);
int parse_cpu_mask_file(const char *fcpu, bool **mask, int *mask_sz);
--- a/tools/lib/bpf/libbpf_legacy.h
+++ b/tools/lib/bpf/libbpf_legacy.h
@@ -45,7 +45,6 @@ enum libbpf_strict_mode {
* (positive) error code.
*/
LIBBPF_STRICT_DIRECT_ERRS = 0x02,
-
/*
* Enforce strict BPF program section (SEC()) names.
* E.g., while prefiously SEC("xdp_whatever") or SEC("perf_event_blah") were
@@ -63,6 +62,17 @@ enum libbpf_strict_mode {
* Clients can maintain it on their own if it is valuable for them.
*/
LIBBPF_STRICT_NO_OBJECT_LIST = 0x08,
+ /*
+ * Automatically bump RLIMIT_MEMLOCK using setrlimit() before the
+ * first BPF program or map creation operation. This is done only if
+ * kernel is too old to support memcg-based memory accounting for BPF
+ * subsystem. By default, RLIMIT_MEMLOCK limit is set to RLIM_INFINITY,
+ * but it can be overriden with libbpf_set_memlock_rlim_max() API.
+ * Note that libbpf_set_memlock_rlim_max() needs to be called before
+ * the very first bpf_prog_load(), bpf_map_create() or bpf_object__load()
+ * operation.
+ */
+ LIBBPF_STRICT_AUTO_RLIMIT_MEMLOCK = 0x10,
__LIBBPF_STRICT_LAST,
};