Skip to contentSkip to conversation

bpf: implement bpf_send_signal() helper

PATCH Applied Yonghong Song 2 participants 12 messages May 23, 2019 – May 25, 2019

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

v5 cover bpf: implement bpf_send_signal() helper

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

Patch #1 implemented the bpf_send_helper() in the kernel. Patch #2 synced uapi header bpf.h to tools directory. Patch #3 added a self test which covers tracepoint and perf_event bpf programs.

Changelogs:
v4 => v5:
. pass the "current" task struct to irq_work as well
since the current task struct may change between nmi and subsequent irq_work_interrupt. Discovered by Daniel.
v3 => v4:
. fix one typo and declare "const char *id_path = ..."
to avoid directly use the long string in the func body in Patch #3.
v2 => v3:
. change the standalone test to be part of prog_tests.
RFC v1 => v2:
. previous version allows to send signal to an arbitrary
pid. This version just sends the signal to current task to avoid unstable pid and potential races between sending signals and task state changes for the pid.

Yonghong Song (3):
bpf: implement bpf_send_signal() helper
tools/bpf: sync bpf uapi header bpf.h to tools directory tools/bpf: add selftest in test_progs for bpf_send_signal() helper

 include/uapi/linux/bpf.h                      |  17 +-
 kernel/trace/bpf_trace.c                      |  72 +++++++
 tools/include/uapi/linux/bpf.h                |  17 +-
 tools/testing/selftests/bpf/bpf_helpers.h     |   1 +
 .../selftests/bpf/prog_tests/send_signal.c    | 198 ++++++++++++++++++
 .../bpf/progs/test_send_signal_kern.c         |  51 +++++
 6 files changed, 354 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/send_signal.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_send_signal_kern.c
Signature · 1 line
2.17.1
#1 lore raw

PATCH v5 1/3 bpf: implement bpf_send_signal() helper

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
 include/uapi/linux/bpf.h | 17 +++++++++-
 kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
  *		0 on success.
  *
  *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
 	FN(strtol),			\
 	FN(strtoul),			\
 	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
  * function eBPF program intends to call
kernel/trace/bpf_trace.c +72 -0

89 lines not on this page · raw message

Signature · 1 line
2.17.1
maintainer
#2 lore raw
Quoted 107 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
 include/uapi/linux/bpf.h | 17 +++++++++-
 kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
  *		0 on success.
  *
  *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
 	FN(strtol),			\
 	FN(strtoul),			\
 	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
  * function eBPF program intends to call
kernel/trace/bpf_trace.c +25 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}
Quoted 54 lines
kernel/trace/bpf_trace.c +37 -0
+			return -EBUSY;
+
+		/* Add the current task, which is the target of sending signal,
+		 * to the irq_work. The current task may change when queued
+		 * irq works get executed.
+		 */
+		work->task = current;
+		work->sig = sig;
+		irq_work_queue(&work->irq_work);
+		return 0;
+	}
+
+	return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
+}
+
+static const struct bpf_func_proto bpf_send_signal_proto = {
+	.func		= bpf_send_signal,
+	.gpl_only	= false,
+	.ret_type	= RET_INTEGER,
+	.arg1_type	= ARG_ANYTHING,
+};
+
 static const struct bpf_func_proto *
 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 {
@@ -617,6 +674,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 	case BPF_FUNC_get_current_cgroup_id:
 		return &bpf_get_current_cgroup_id_proto;
 #endif
+	case BPF_FUNC_send_signal:
+		return &bpf_send_signal_proto;
 	default:
 		return NULL;
 	}
@@ -1343,5 +1402,18 @@ static int __init bpf_event_init(void)
 	return 0;
 }
 
+static int __init send_signal_irq_work_init(void)
+{
+	int cpu;
+	struct send_signal_irq_work *work;
+
+	for_each_possible_cpu(cpu) {
+		work = per_cpu_ptr(&send_signal_work, cpu);
+		init_irq_work(&work->irq_work, do_bpf_send_signal);
+	}
+	return 0;
+}
+
 fs_initcall(bpf_event_init);
+subsys_initcall(send_signal_irq_work_init);
 #endif /* CONFIG_MODULES */
maintainer
#3 lore raw
Quoted 95 lines · Daniel Borkmann
Quoted 94 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
 include/uapi/linux/bpf.h | 17 +++++++++-
 kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
  *		0 on success.
  *
  *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
 	FN(strtol),			\
 	FN(strtoul),			\
 	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
  * function eBPF program intends to call
kernel/trace/bpf_trace.c +12 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
 	.arg3_type	= ARG_ANYTHING,
 };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
Quoted 8 lines

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 589

Quoted 8 lines
kernel/trace/bpf_trace.c +8 -0
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+

Oh, and one more thing:

	if (!valid_signal(sig))
		return -EINVAL;

Otherwise when deferring the work, you don't have any such feedback.

Quoted 82 lines
Quoted 5 lines
kernel/trace/bpf_trace.c +5 -0
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}
Quoted 54 lines
kernel/trace/bpf_trace.c +37 -0
+			return -EBUSY;
+
+		/* Add the current task, which is the target of sending signal,
+		 * to the irq_work. The current task may change when queued
+		 * irq works get executed.
+		 */
+		work->task = current;
+		work->sig = sig;
+		irq_work_queue(&work->irq_work);
+		return 0;
+	}
+
+	return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
+}
+
+static const struct bpf_func_proto bpf_send_signal_proto = {
+	.func		= bpf_send_signal,
+	.gpl_only	= false,
+	.ret_type	= RET_INTEGER,
+	.arg1_type	= ARG_ANYTHING,
+};
+
 static const struct bpf_func_proto *
 tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 {
@@ -617,6 +674,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 	case BPF_FUNC_get_current_cgroup_id:
 		return &bpf_get_current_cgroup_id_proto;
 #endif
+	case BPF_FUNC_send_signal:
+		return &bpf_send_signal_proto;
 	default:
 		return NULL;
 	}
@@ -1343,5 +1402,18 @@ static int __init bpf_event_init(void)
 	return 0;
 }
 
+static int __init send_signal_irq_work_init(void)
+{
+	int cpu;
+	struct send_signal_irq_work *work;
+
+	for_each_possible_cpu(cpu) {
+		work = per_cpu_ptr(&send_signal_work, cpu);
+		init_irq_work(&work->irq_work, do_bpf_send_signal);
+	}
+	return 0;
+}
+
 fs_initcall(bpf_event_init);
+subsys_initcall(send_signal_irq_work_init);
 #endif /* CONFIG_MODULES */
#4 lore raw
Quoted 107 lines · Daniel Borkmann
Quoted 95 lines · Daniel Borkmann
Quoted 94 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
  include/uapi/linux/bpf.h | 17 +++++++++-
  kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
  2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
   *		0 on success.
   *
   *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
   */
  #define __BPF_FUNC_MAPPER(FN)		\
  	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
  	FN(strtol),			\
  	FN(strtoul),			\
  	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
   * function eBPF program intends to call
kernel/trace/bpf_trace.c +12 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
  	.arg3_type	= ARG_ANYTHING,
  };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
Quoted 8 lines

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 589

Quoted 8 lines
kernel/trace/bpf_trace.c +8 -0
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+

Oh, and one more thing:

Quoted 4 lines
	if (!valid_signal(sig))
		return -EINVAL;

Otherwise when deferring the work, you don't have any such feedback.

Good advice! Do you want me send a followup patch or resend the whole series?

Quoted 84 lines
Quoted 83 lines
Quoted 6 lines
kernel/trace/bpf_trace.c +5 -0
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}
Quoted 54 lines
kernel/trace/bpf_trace.c +37 -0
+			return -EBUSY;
+
+		/* Add the current task, which is the target of sending signal,
+		 * to the irq_work. The current task may change when queued
+		 * irq works get executed.
+		 */
+		work->task = current;
+		work->sig = sig;
+		irq_work_queue(&work->irq_work);
+		return 0;
+	}
+
+	return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
+}
+
+static const struct bpf_func_proto bpf_send_signal_proto = {
+	.func		= bpf_send_signal,
+	.gpl_only	= false,
+	.ret_type	= RET_INTEGER,
+	.arg1_type	= ARG_ANYTHING,
+};
+
  static const struct bpf_func_proto *
  tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  {
@@ -617,6 +674,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  	case BPF_FUNC_get_current_cgroup_id:
  		return &bpf_get_current_cgroup_id_proto;
  #endif
+	case BPF_FUNC_send_signal:
+		return &bpf_send_signal_proto;
  	default:
  		return NULL;
  	}
@@ -1343,5 +1402,18 @@ static int __init bpf_event_init(void)
  	return 0;
  }
 
+static int __init send_signal_irq_work_init(void)
+{
+	int cpu;
+	struct send_signal_irq_work *work;
+
+	for_each_possible_cpu(cpu) {
+		work = per_cpu_ptr(&send_signal_work, cpu);
+		init_irq_work(&work->irq_work, do_bpf_send_signal);
+	}
+	return 0;
+}
+
  fs_initcall(bpf_event_init);
+subsys_initcall(send_signal_irq_work_init);
  #endif /* CONFIG_MODULES */
maintainer
#5 lore raw
Quoted 108 lines · Yonghong Song
Quoted 107 lines · Daniel Borkmann
Quoted 95 lines · Daniel Borkmann
Quoted 94 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
  include/uapi/linux/bpf.h | 17 +++++++++-
  kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
  2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
   *		0 on success.
   *
   *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
   */
  #define __BPF_FUNC_MAPPER(FN)		\
  	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
  	FN(strtol),			\
  	FN(strtoul),			\
  	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
   * function eBPF program intends to call
kernel/trace/bpf_trace.c +12 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
  	.arg3_type	= ARG_ANYTHING,
  };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
Quoted 8 lines

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 589

Quoted 8 lines
kernel/trace/bpf_trace.c +8 -0
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+

Oh, and one more thing:

Quoted 7 lines
Quoted 4 lines
	if (!valid_signal(sig))
		return -EINVAL;

Otherwise when deferring the work, you don't have any such feedback.

Good advice! Do you want me send a followup patch or resend the whole series?

Lets do follow-up.

#6 lore raw
In reply to Daniel Borkmann
Quoted 123 lines · Daniel Borkmann
Quoted 107 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
  include/uapi/linux/bpf.h | 17 +++++++++-
  kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
  2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
   *		0 on success.
   *
   *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
   */
  #define __BPF_FUNC_MAPPER(FN)		\
  	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
  	FN(strtol),			\
  	FN(strtoul),			\
  	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
   * function eBPF program intends to call
kernel/trace/bpf_trace.c +25 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
  	.arg3_type	= ARG_ANYTHING,
  };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Quoted 6 lines

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}

Not sure whether READ_ONCE is needed here or not.

The irq_work is per cpu data structure,
static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
so presumably no collision for work->flags memory reference.

If without READ_ONCE, do you still think a helper is needed?

Quoted 56 lines
Quoted 55 lines
kernel/trace/bpf_trace.c +37 -0
+			return -EBUSY;
+
+		/* Add the current task, which is the target of sending signal,
+		 * to the irq_work. The current task may change when queued
+		 * irq works get executed.
+		 */
+		work->task = current;
+		work->sig = sig;
+		irq_work_queue(&work->irq_work);
+		return 0;
+	}
+
+	return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
+}
+
+static const struct bpf_func_proto bpf_send_signal_proto = {
+	.func		= bpf_send_signal,
+	.gpl_only	= false,
+	.ret_type	= RET_INTEGER,
+	.arg1_type	= ARG_ANYTHING,
+};
+
  static const struct bpf_func_proto *
  tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  {
@@ -617,6 +674,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  	case BPF_FUNC_get_current_cgroup_id:
  		return &bpf_get_current_cgroup_id_proto;
  #endif
+	case BPF_FUNC_send_signal:
+		return &bpf_send_signal_proto;
  	default:
  		return NULL;
  	}
@@ -1343,5 +1402,18 @@ static int __init bpf_event_init(void)
  	return 0;
  }
 
+static int __init send_signal_irq_work_init(void)
+{
+	int cpu;
+	struct send_signal_irq_work *work;
+
+	for_each_possible_cpu(cpu) {
+		work = per_cpu_ptr(&send_signal_work, cpu);
+		init_irq_work(&work->irq_work, do_bpf_send_signal);
+	}
+	return 0;
+}
+
  fs_initcall(bpf_event_init);
+subsys_initcall(send_signal_irq_work_init);
  #endif /* CONFIG_MODULES */
maintainer
#7 lore raw
Quoted 131 lines · Yonghong Song
Quoted 123 lines · Daniel Borkmann
Quoted 107 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
  include/uapi/linux/bpf.h | 17 +++++++++-
  kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
  2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
   *		0 on success.
   *
   *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
   */
  #define __BPF_FUNC_MAPPER(FN)		\
  	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
  	FN(strtol),			\
  	FN(strtoul),			\
  	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
   * function eBPF program intends to call
kernel/trace/bpf_trace.c +25 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
  	.arg3_type	= ARG_ANYTHING,
  };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Quoted 6 lines

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}
Quoted 5 lines

Not sure whether READ_ONCE is needed here or not.

The irq_work is per cpu data structure,
static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
so presumably no collision for work->flags memory reference.

The busy bit you're testing is cleared via cmpxchg(), kernel/irq_work.c +169:

cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY);

#8 lore raw
Quoted 134 lines · Daniel Borkmann
Quoted 133 lines · Yonghong Song
Quoted 123 lines · Daniel Borkmann
Quoted 107 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

Such a mechanism can be implemented in the following way:
. a perf ring buffer is created between bpf program
and tracing app.
. once a particular event happens, bpf program writes
to the ring buffer and the tracing app gets notified.
. the tracing app sends a signal SIGALARM to the hhvm.

But this method could have large delays and causing profiling results skewed.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

---
   include/uapi/linux/bpf.h | 17 +++++++++-
   kernel/trace/bpf_trace.c | 72 ++++++++++++++++++++++++++++++++++++++++
   2 files changed, 88 insertions(+), 1 deletion(-)
include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
    *		0 on success.
    *
    *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
    */
   #define __BPF_FUNC_MAPPER(FN)		\
   	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
   	FN(strtol),			\
   	FN(strtoul),			\
   	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
   /* integer value in 'imm' field of BPF_CALL instruction selects which helper
    * function eBPF program intends to call
kernel/trace/bpf_trace.c +25 -0
index f92d6ad5e080..70029eafc71f 100644
@@ -567,6 +567,63 @@ static const struct bpf_func_proto bpf_probe_read_str_proto = {
   	.arg3_type	= ARG_ANYTHING,
   };
 
+struct send_signal_irq_work {
+	struct irq_work irq_work;
+	struct task_struct *task;
+	u32 sig;
+};
+
+static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
+
+static void do_bpf_send_signal(struct irq_work *entry)
+{
+	struct send_signal_irq_work *work;
+
+	work = container_of(entry, struct send_signal_irq_work, irq_work);
+	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
+}
+
+BPF_CALL_1(bpf_send_signal, u32, sig)
+{
+	struct send_signal_irq_work *work = NULL;
+
+	/* Similar to bpf_probe_write_user, task needs to be
+	 * in a sound condition and kernel memory access be
+	 * permitted in order to send signal to the current
+	 * task.
+	 */

kernel/trace/bpf_trace.c static const struct bpf_func_proto bpf_probe_read_str_proto = { · line 604

Quoted 10 lines
kernel/trace/bpf_trace.c +10 -0
+	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
+		return -EPERM;
+	if (unlikely(uaccess_kernel()))
+		return -EPERM;
+	if (unlikely(!nmi_uaccess_okay()))
+		return -EPERM;
+
+	if (in_nmi()) {
+		work = this_cpu_ptr(&send_signal_work);
+		if (work->irq_work.flags & IRQ_WORK_BUSY)

Given here and in stackmap are the only two users outside of kernel/irq_work.c, it may probably be good to add a small helper to include/linux/irq_work.h and use it for both.

Quoted 6 lines

Perhaps something like ...

static inline bool irq_work_busy(struct irq_work *work)

{
	return READ_ONCE(work->flags) & IRQ_WORK_BUSY;
}

Not sure whether READ_ONCE is needed here or not.

Quoted 7 lines
Quoted 3 lines

The irq_work is per cpu data structure,

    static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);

so presumably no collision for work->flags memory reference.

The busy bit you're testing is cleared via cmpxchg(), kernel/irq_work.c +169:

cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY);

Looks like for bpf case, we have bpf_prog_active guarding, so nested nmi won't trigger nested bpf_send_signal() call, so we should be fine without READ_ONCE here.

Also,

struct irq_work {
         unsigned long flags;
         struct llist_node llnode;
         void (*func)(struct irq_work *);
};

-bash-4.4$ egrep -r 'work->flags' irq_work.c

         flags = work->flags & ~IRQ_WORK_PENDING;
                 oflags = cmpxchg(&work->flags, flags, nflags);
         if (work->flags & IRQ_WORK_LAZY) {
                 flags = work->flags & ~IRQ_WORK_PENDING;
                 xchg(&work->flags, flags);
                 (void)cmpxchg(&work->flags, flags, flags & ~IRQ_WORK_BUSY);
         while (work->flags & IRQ_WORK_BUSY)

-bash-4.4$

From the above, flags is unsigned long and no READ_ONCE
is used, people may already assume atomic read_once is available, so not using it.

#9 lore raw

PATCH v5 2/3 tools/bpf: sync bpf uapi header bpf.h to tools directory

In reply to Yonghong Song

The bpf uapi header include/uapi/linux/bpf.h is sync'ed to tools/include/uapi/linux/bpf.h.

---
 tools/include/uapi/linux/bpf.h | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)
tools/include/uapi/linux/bpf.h +16 -1
index 63e0cf66f01a..68d4470523a0 100644
@@ -2672,6 +2672,20 @@ union bpf_attr {
  *		0 on success.
  *
  *		**-ENOENT** if the bpf-local-storage cannot be found.
+ *
+ * int bpf_send_signal(u32 sig)
+ *	Description
+ *		Send signal *sig* to the current task.
+ *	Return
+ *		0 on success or successfully queued.
+ *
+ *		**-EBUSY** if work queue under nmi is full.
+ *
+ *		**-EINVAL** if *sig* is invalid.
+ *
+ *		**-EPERM** if no permission to send the *sig*.
+ *
+ *		**-EAGAIN** if bpf program can try again.
  */
 #define __BPF_FUNC_MAPPER(FN)		\
 	FN(unspec),			\
@@ -2782,7 +2796,8 @@ union bpf_attr {
 	FN(strtol),			\
 	FN(strtoul),			\
 	FN(sk_storage_get),		\
-	FN(sk_storage_delete),
+	FN(sk_storage_delete),		\
+	FN(send_signal),
 
 /* integer value in 'imm' field of BPF_CALL instruction selects which helper
  * function eBPF program intends to call
Signature · 1 line
2.17.1
#10 lore raw

PATCH v5 3/3 tools/bpf: add selftest in test_progs for bpf_send_signal() helper

In reply to Yonghong Song

The test covered both nmi and tracepoint perf events.

  $ ./test_progs
  ...
  test_send_signal_tracepoint:PASS:tracepoint 0 nsec
  ...
  test_send_signal_common:PASS:tracepoint 0 nsec
  ...
  test_send_signal_common:PASS:perf_event 0 nsec
  ...
  test_send_signal:OK
---
 tools/testing/selftests/bpf/bpf_helpers.h     |   1 +
 .../selftests/bpf/prog_tests/send_signal.c    | 198 ++++++++++++++++++
 .../bpf/progs/test_send_signal_kern.c         |  51 +++++
 3 files changed, 250 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/send_signal.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_send_signal_kern.c
tools/testing/selftests/bpf/bpf_helpers.h +1 -0
index 5f6f9e7aba2a..cb02521b8e58 100644
@@ -216,6 +216,7 @@ static void *(*bpf_sk_storage_get)(void *map, struct bpf_sock *sk,
 	(void *) BPF_FUNC_sk_storage_get;
 static int (*bpf_sk_storage_delete)(void *map, struct bpf_sock *sk) =
 	(void *)BPF_FUNC_sk_storage_delete;
+static int (*bpf_send_signal)(unsigned sig) = (void *)BPF_FUNC_send_signal;
 
 /* llvm builtin functions that eBPF C program may use to
  * emit BPF_LD_ABS and BPF_LD_IND instructions
tools/testing/selftests/bpf/prog_tests/send_signal.c new file +198 -0

198 lines not on this page · raw message

tools/testing/selftests/bpf/progs/test_send_signal_kern.c new file +51 -0

51 lines not on this page · raw message

Signature · 1 line
2.17.1
maintainer Applied
#11 lore raw
In reply to Yonghong Song
Quoted 50 lines · Yonghong Song

This patch tries to solve the following specific use case.

Currently, bpf program can already collect stack traces through kernel function get_perf_callchain()
when certain events happens (e.g., cache miss counter or cpu clock counter overflows). But such stack traces are not enough for jitted programs, e.g., hhvm (jited php). To get real stack trace, jit engine internal data structures need to be traversed in order to get the real user functions.

bpf program itself may not be the best place to traverse the jit engine as the traversing logic could be complex and it is not a stable interface either.

Instead, hhvm implements a signal handler,
e.g. for SIGALARM, and a set of program locations which it can dump stack traces. When it receives a signal, it will dump the stack in next such program location.

This patch implements bpf_send_signal() helper to send a signal to hhvm in real time, resulting in intended stack traces.

Patch #1 implemented the bpf_send_helper() in the kernel. Patch #2 synced uapi header bpf.h to tools directory. Patch #3 added a self test which covers tracepoint and perf_event bpf programs.

Changelogs:
v4 => v5:
. pass the "current" task struct to irq_work as well
since the current task struct may change between nmi and subsequent irq_work_interrupt. Discovered by Daniel.
v3 => v4:
. fix one typo and declare "const char *id_path = ..."
to avoid directly use the long string in the func body in Patch #3.
v2 => v3:
. change the standalone test to be part of prog_tests.
RFC v1 => v2:
. previous version allows to send signal to an arbitrary
pid. This version just sends the signal to current task to avoid unstable pid and potential races between sending signals and task state changes for the pid.

Yonghong Song (3):
bpf: implement bpf_send_signal() helper
tools/bpf: sync bpf uapi header bpf.h to tools directory tools/bpf: add selftest in test_progs for bpf_send_signal() helper

Quoted 10 lines
 include/uapi/linux/bpf.h                      |  17 +-
 kernel/trace/bpf_trace.c                      |  72 +++++++
 tools/include/uapi/linux/bpf.h                |  17 +-
 tools/testing/selftests/bpf/bpf_helpers.h     |   1 +
 .../selftests/bpf/prog_tests/send_signal.c    | 198 ++++++++++++++++++
 .../bpf/progs/test_send_signal_kern.c         |  51 +++++
 6 files changed, 354 insertions(+), 2 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/send_signal.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_send_signal_kern.c

Applied, thanks. One more remark in patch 1, will reply there.