Linux-Trace-Devel Archive on lore.kernel.org
 help / color / Atom feed
* [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs
@ 2019-10-08  8:15 Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 01/13] trace-cmd: Make ports unsigned int Slavomir Kaslev
                   ` (12 more replies)
  0 siblings, 13 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

This patchset adds support for tracing guest kernels to trace-cmd.

Changes in v14 and v15:
 - rebased on top of latest master
 - no changes in behavior

Changes in v13:
 - ignore SIGINT in recording processes
 - add more checks for memory allocation failures

Changes in v12:
 - fix patches' Signed-off-by tags

Changes in v11:
 - fix flushing tracing data bug when doing VM recording

Changes in v10:
 - add support for specifying VMware Workstation guests by name on the command line
 - add docstring for tracecmd_create_recorder_virt()
 - add a comment on why we need to poll() in direct_splice_data()

Changes in v9:
 - none

Changes in v8:
 - switched protocol message encoding of ports to text
 - switched ports data type to unsigned since vsockets port numbers can overflow int and be confused with errors codes by other code
 - added automatic naming for guests with no names (for VMware Workstation guests currently)
 - added --debug option to trace-cmd agent

Changes in v7:
 - fixed potential memory leak in add_arg()
 - call msg_free() in tracecmd_msg_recv_trace_req() eventhough we're stealing
   the message buffer
 - added is_guest and is_agent macros and use throughout tracecmd_record.c
   instead of testing instance->flags bits
 - avoid memory allocation in make_dir()

Changes in v6:
 - added specialized data transfer path over FIFOs making single splice() per
   iteration
 - made tracecmd_msg_recv_trace_req/tracecmd_msg_recv_trace_resp more defensive
   to invalid messages

Changes in v5:
 - add FIFOs transport for tracing data
 - fixed a bug in tracecmd_msg_wait_close

Changes in v4:
 - detect and use splice(2) on vsock sockets if possible
 - switch port numbers to unsigned int
 - trace-cmd record --date flag is now set for all guests if provided by the user
 - removed grow_cap and exponential buffer size growth

Changes in v3:
 - addressed Steven's feedback
 - detect and disable guest tracing if <linux/vm_sockets.h> is not available
 - the --date flag is now treated as global for all guest instances
 - fixed a bug that caused --date to be ignored for host tracing data when tracing guests

Changes in v2:
 - rebased on top of protocol V3
 - fixed system clock timestamps with the --date flag


Slavomir Kaslev (12):
  trace-cmd: Make ports unsigned int
  trace-cmd: Add tracecmd_create_recorder_virt() function
  trace-cmd: Add VM tracing protocol messages
  trace-cmd: Add buffer instance flags for tracing in guest and agent
    context
  trace-cmd: Add VM kernel tracing over vsockets transport
  trace-cmd: Use splice(2) for vsockets if available
  trace-cmd: Switch stop recording signal to SIGUSR1
  trace-cmd: Add `trace-cmd setup-guest` command
  trace-cmd: Try to autodetect number of guest CPUs in setup-guest if
    not specified
  trace-cmd: Add setup-guest flag for attaching FIFOs to the guest VM
    config
  trace-cmd: Add VM kernel tracing over FIFO transport
  trace-cmd: Add support for tracing VMware Workstation VMs by name

Steven Rostedt (VMware) (1):
  trace-cmd: Detect if vsockets are available

 Makefile                       |   7 +
 include/trace-cmd/trace-cmd.h  |  34 +-
 lib/trace-cmd/trace-msg.c      | 274 ++++++++++-
 lib/trace-cmd/trace-recorder.c | 154 ++++--
 tracecmd/Makefile              |   5 +
 tracecmd/include/trace-local.h |  29 ++
 tracecmd/trace-agent.c         | 294 +++++++++++
 tracecmd/trace-cmd.c           |   4 +
 tracecmd/trace-listen.c        |  10 +-
 tracecmd/trace-record.c        | 875 +++++++++++++++++++++++++++++++--
 tracecmd/trace-setup-guest.c   | 247 ++++++++++
 tracecmd/trace-usage.c         |  22 +-
 12 files changed, 1848 insertions(+), 107 deletions(-)
 create mode 100644 tracecmd/trace-agent.c
 create mode 100644 tracecmd/trace-setup-guest.c

-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 01/13] trace-cmd: Make ports unsigned int
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 02/13] trace-cmd: Detect if vsockets are available Slavomir Kaslev
                   ` (11 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Switch ports data type to unsigned int since vsocket ports are 32 bit unsigned
integers and sometimes cause overflow when stored in int variables.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 include/trace-cmd/trace-cmd.h |  2 +-
 lib/trace-cmd/trace-msg.c     | 32 ++++++++++++++++++++++----------
 tracecmd/trace-listen.c       | 10 +++++-----
 3 files changed, 28 insertions(+), 16 deletions(-)

diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
index 4090eba..00889fb 100644
--- a/include/trace-cmd/trace-cmd.h
+++ b/include/trace-cmd/trace-cmd.h
@@ -333,7 +333,7 @@ int tracecmd_msg_wait_close(struct tracecmd_msg_handle *msg_handle);
 /* for server */
 int tracecmd_msg_initial_setting(struct tracecmd_msg_handle *msg_handle);
 int tracecmd_msg_send_port_array(struct tracecmd_msg_handle *msg_handle,
-				 int *ports);
+				 unsigned *ports);
 int tracecmd_msg_read_data(struct tracecmd_msg_handle *msg_handle, int ofd);
 int tracecmd_msg_collect_data(struct tracecmd_msg_handle *msg_handle, int ofd);
 bool tracecmd_msg_done(struct tracecmd_msg_handle *msg_handle);
diff --git a/lib/trace-cmd/trace-msg.c b/lib/trace-cmd/trace-msg.c
index 74c5254..4def7e6 100644
--- a/lib/trace-cmd/trace-msg.c
+++ b/lib/trace-cmd/trace-msg.c
@@ -161,12 +161,26 @@ static int make_tinit(struct tracecmd_msg_handle *msg_handle,
 	return 0;
 }
 
-static int write_ints(char *buf, size_t buf_len, int *arr, int arr_len)
+/* test a to u */
+static int tatou(const char *s, unsigned int *res)
+{
+        long r;
+
+        r = atol(s);
+        if (r >= 0 && r <= UINT_MAX) {
+                *res = (unsigned int)r;
+                return 0;
+        }
+        return -1;
+}
+
+static int write_uints(char *buf, size_t buf_len,
+		       unsigned int *arr, int arr_len)
 {
 	int i, ret, tot = 0;
 
 	for (i = 0; i < arr_len; i++) {
-		ret = snprintf(buf, buf_len, "%d", arr[i]);
+		ret = snprintf(buf, buf_len, "%u", arr[i]);
 		if (ret < 0)
 			return ret;
 
@@ -184,15 +198,15 @@ static int write_ints(char *buf, size_t buf_len, int *arr, int arr_len)
 	return tot;
 }
 
-static int make_rinit(struct tracecmd_msg *msg, int cpus, int *ports)
+static int make_rinit(struct tracecmd_msg *msg, int cpus, unsigned int *ports)
 {
 	int data_size;
 
-	data_size = write_ints(NULL, 0, ports, cpus);
+	data_size = write_uints(NULL, 0, ports, cpus);
 	msg->buf = malloc(data_size);
 	if (!msg->buf)
 		return -ENOMEM;
-	write_ints(msg->buf, data_size, ports, cpus);
+	write_uints(msg->buf, data_size, ports, cpus);
 
 	msg->rinit.cpus = htonl(cpus);
 	msg->hdr.size = htonl(ntohl(msg->hdr.size) + data_size);
@@ -442,7 +456,7 @@ int tracecmd_msg_send_init_data(struct tracecmd_msg_handle *msg_handle,
 	}
 
 	cpus = ntohl(msg.rinit.cpus);
-	ports = malloc_or_die(sizeof(*ports) * cpus);
+	ports = malloc(sizeof(*ports) * cpus);
 	if (!ports) {
 		ret = -ENOMEM;
 		goto out;
@@ -450,13 +464,11 @@ int tracecmd_msg_send_init_data(struct tracecmd_msg_handle *msg_handle,
 
 	buf_end = msg.buf + buf_len;
 	for (i = 0, p = msg.buf; i < cpus; i++, p++) {
-		if (p >= buf_end) {
+		if (p >= buf_end || tatou(p, &ports[i])) {
 			free(ports);
 			ret = -EINVAL;
 			goto error;
 		}
-
-		ports[i] = atoi(p);
 		p = strchr(p, '\0');
 	}
 
@@ -588,7 +600,7 @@ error:
 }
 
 int tracecmd_msg_send_port_array(struct tracecmd_msg_handle *msg_handle,
-				 int *ports)
+				 unsigned int *ports)
 {
 	struct tracecmd_msg msg;
 	int ret;
diff --git a/tracecmd/trace-listen.c b/tracecmd/trace-listen.c
index 233d661..7798fe4 100644
--- a/tracecmd/trace-listen.c
+++ b/tracecmd/trace-listen.c
@@ -465,10 +465,10 @@ static int *create_all_readers(const char *node, const char *port,
 {
 	int use_tcp = msg_handle->flags & TRACECMD_MSG_FL_USE_TCP;
 	char buf[BUFSIZ];
-	int *port_array;
+	unsigned int *port_array;
 	int *pid_array;
-	int start_port;
-	int udp_port;
+	unsigned int start_port;
+	unsigned int udp_port;
 	int cpus = msg_handle->cpu_count;
 	int cpu;
 	int pid;
@@ -476,11 +476,11 @@ static int *create_all_readers(const char *node, const char *port,
 	if (!pagesize)
 		return NULL;
 
-	port_array = malloc(sizeof(int) * cpus);
+	port_array = malloc(sizeof(*port_array) * cpus);
 	if (!port_array)
 		return NULL;
 
-	pid_array = malloc(sizeof(int) * cpus);
+	pid_array = malloc(sizeof(*pid_array) * cpus);
 	if (!pid_array) {
 		free(port_array);
 		return NULL;
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 02/13] trace-cmd: Detect if vsockets are available
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 01/13] trace-cmd: Make ports unsigned int Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 03/13] trace-cmd: Add tracecmd_create_recorder_virt() function Slavomir Kaslev
                   ` (10 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

From: "Steven Rostedt (VMware)" <rostedt@goodmis.org>

Detect and define VSOCK if vsockets are available on the system.
This macro is used to disable VM remote tracing features on older kernels.

Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 Makefile | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/Makefile b/Makefile
index bbdf15e..9723d49 100644
--- a/Makefile
+++ b/Makefile
@@ -208,6 +208,13 @@ CFLAGS ?= -g -Wall
 CPPFLAGS ?=
 LDFLAGS ?=
 
+VSOCK_DEFINED := $(shell if (echo "\#include <linux/vm_sockets.h>" | $(CC) -E - >/dev/null 2>&1) ; then echo 1; else echo 0 ; fi)
+
+export VSOCK_DEFINED
+ifeq ($(VSOCK_DEFINED), 1)
+CFLAGS += -DVSOCK
+endif
+
 export CFLAGS
 export INCLUDES
 
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 03/13] trace-cmd: Add tracecmd_create_recorder_virt() function
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 01/13] trace-cmd: Make ports unsigned int Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 02/13] trace-cmd: Detect if vsockets are available Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 04/13] trace-cmd: Add VM tracing protocol messages Slavomir Kaslev
                   ` (9 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Add tracecmd_create_recorder_virt() which creates tracecmd_recorder recording
tracing data from the trace_fd parameter instead of from the local systems
tracefs. It will be used for recording tracing data coming from VMs.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 include/trace-cmd/trace-cmd.h  |  1 +
 lib/trace-cmd/trace-recorder.c | 61 +++++++++++++++++++++++++---------
 2 files changed, 47 insertions(+), 15 deletions(-)

diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
index 00889fb..65bece5 100644
--- a/include/trace-cmd/trace-cmd.h
+++ b/include/trace-cmd/trace-cmd.h
@@ -293,6 +293,7 @@ enum {
 void tracecmd_free_recorder(struct tracecmd_recorder *recorder);
 struct tracecmd_recorder *tracecmd_create_recorder(const char *file, int cpu, unsigned flags);
 struct tracecmd_recorder *tracecmd_create_recorder_fd(int fd, int cpu, unsigned flags);
+struct tracecmd_recorder *tracecmd_create_recorder_virt(const char *file, int cpu, unsigned flags, int trace_fd);
 struct tracecmd_recorder *tracecmd_create_recorder_maxkb(const char *file, int cpu, unsigned flags, int maxkb);
 struct tracecmd_recorder *tracecmd_create_buffer_recorder_fd(int fd, int cpu, unsigned flags, const char *buffer);
 struct tracecmd_recorder *tracecmd_create_buffer_recorder(const char *file, int cpu, unsigned flags, const char *buffer);
diff --git a/lib/trace-cmd/trace-recorder.c b/lib/trace-cmd/trace-recorder.c
index 650b259..7a4728f 100644
--- a/lib/trace-cmd/trace-recorder.c
+++ b/lib/trace-cmd/trace-recorder.c
@@ -154,16 +154,22 @@ tracecmd_create_buffer_recorder_fd2(int fd, int fd2, int cpu, unsigned flags,
 	recorder->fd1 = fd;
 	recorder->fd2 = fd2;
 
-	if (flags & TRACECMD_RECORD_SNAPSHOT)
-		ret = asprintf(&path, "%s/per_cpu/cpu%d/snapshot_raw", buffer, cpu);
-	else
-		ret = asprintf(&path, "%s/per_cpu/cpu%d/trace_pipe_raw", buffer, cpu);
-	if (ret < 0)
-		goto out_free;
+	if (buffer) {
+		if (flags & TRACECMD_RECORD_SNAPSHOT)
+			ret = asprintf(&path, "%s/per_cpu/cpu%d/snapshot_raw",
+				       buffer, cpu);
+		else
+			ret = asprintf(&path, "%s/per_cpu/cpu%d/trace_pipe_raw",
+				       buffer, cpu);
+		if (ret < 0)
+			goto out_free;
+
+		recorder->trace_fd = open(path, O_RDONLY);
+		free(path);
 
-	recorder->trace_fd = open(path, O_RDONLY);
-	if (recorder->trace_fd < 0)
-		goto out_free;
+		if (recorder->trace_fd < 0)
+			goto out_free;
+	}
 
 	if ((recorder->flags & TRACECMD_RECORD_NOSPLICE) == 0) {
 		ret = pipe(recorder->brass);
@@ -183,13 +189,9 @@ tracecmd_create_buffer_recorder_fd2(int fd, int fd2, int cpu, unsigned flags,
 		recorder->pipe_size = pipe_size;
 	}
 
-	free(path);
-
 	return recorder;
 
  out_free:
-	free(path);
-
 	tracecmd_free_recorder(recorder);
 	return NULL;
 }
@@ -200,8 +202,9 @@ tracecmd_create_buffer_recorder_fd(int fd, int cpu, unsigned flags, const char *
 	return tracecmd_create_buffer_recorder_fd2(fd, -1, cpu, flags, buffer, 0);
 }
 
-struct tracecmd_recorder *
-tracecmd_create_buffer_recorder(const char *file, int cpu, unsigned flags, const char *buffer)
+static struct tracecmd_recorder *
+__tracecmd_create_buffer_recorder(const char *file, int cpu, unsigned flags,
+				  const char *buffer)
 {
 	struct tracecmd_recorder *recorder;
 	int fd;
@@ -264,6 +267,34 @@ tracecmd_create_buffer_recorder_maxkb(const char *file, int cpu, unsigned flags,
 	goto out;
 }
 
+struct tracecmd_recorder *
+tracecmd_create_buffer_recorder(const char *file, int cpu, unsigned flags,
+				const char *buffer)
+{
+	return __tracecmd_create_buffer_recorder(file, cpu, flags, buffer);
+}
+
+/**
+ * tracecmd_create_recorder_virt - Create a recorder reading tracing data
+ * from the trace_fd file descriptor instead of from the local tracefs
+ * @file: output filename where tracing data will be written
+ * @cpu: which CPU is being traced
+ * @flags: flags configuring the recorder (see TRACECMD_RECORDER_* enums)
+ * @trace_fd: file descriptor from where tracing data will be read
+ */
+struct tracecmd_recorder *
+tracecmd_create_recorder_virt(const char *file, int cpu, unsigned flags,
+			      int trace_fd)
+{
+	struct tracecmd_recorder *recorder;
+
+	recorder = __tracecmd_create_buffer_recorder(file, cpu, flags, NULL);
+	if (recorder)
+		recorder->trace_fd = trace_fd;
+
+	return recorder;
+}
+
 struct tracecmd_recorder *tracecmd_create_recorder_fd(int fd, int cpu, unsigned flags)
 {
 	const char *tracing;
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 04/13] trace-cmd: Add VM tracing protocol messages
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (2 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 03/13] trace-cmd: Add tracecmd_create_recorder_virt() function Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 05/13] trace-cmd: Add buffer instance flags for tracing in guest and agent context Slavomir Kaslev
                   ` (8 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Add TRACE_REQ and TRACE_RESP messages used for initiating guest VM tracing and
CLOSE_RESP message used for acknowledging stop recording request.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 include/trace-cmd/trace-cmd.h |  28 ++++
 lib/trace-cmd/trace-msg.c     | 234 +++++++++++++++++++++++++++++++++-
 2 files changed, 259 insertions(+), 3 deletions(-)

diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
index 65bece5..b1e45f8 100644
--- a/include/trace-cmd/trace-cmd.h
+++ b/include/trace-cmd/trace-cmd.h
@@ -329,7 +329,9 @@ int tracecmd_msg_data_send(struct tracecmd_msg_handle *msg_handle,
 			       const char *buf, int size);
 int tracecmd_msg_finish_sending_data(struct tracecmd_msg_handle *msg_handle);
 int tracecmd_msg_send_close_msg(struct tracecmd_msg_handle *msg_handle);
+int tracecmd_msg_send_close_resp_msg(struct tracecmd_msg_handle *msg_handle);
 int tracecmd_msg_wait_close(struct tracecmd_msg_handle *msg_handle);
+int tracecmd_msg_wait_close_resp(struct tracecmd_msg_handle *msg_handle);
 
 /* for server */
 int tracecmd_msg_initial_setting(struct tracecmd_msg_handle *msg_handle);
@@ -340,6 +342,32 @@ int tracecmd_msg_collect_data(struct tracecmd_msg_handle *msg_handle, int ofd);
 bool tracecmd_msg_done(struct tracecmd_msg_handle *msg_handle);
 void tracecmd_msg_set_done(struct tracecmd_msg_handle *msg_handle);
 
+int tracecmd_msg_send_trace_req(struct tracecmd_msg_handle *msg_handle,
+				int argc, char **argv);
+int tracecmd_msg_recv_trace_req(struct tracecmd_msg_handle *msg_handle,
+				int *argc, char ***argv);
+
+int tracecmd_msg_send_trace_resp(struct tracecmd_msg_handle *msg_handle,
+				 int nr_cpus, int page_size,
+				 unsigned int *ports);
+int tracecmd_msg_recv_trace_resp(struct tracecmd_msg_handle *msg_handle,
+				 int *nr_cpus, int *page_size,
+				 unsigned int **ports);
+
+/* --- Plugin handling --- */
+extern struct tep_plugin_option trace_ftrace_options[];
+
+int trace_util_add_options(const char *name, struct tep_plugin_option *options);
+void trace_util_remove_options(struct tep_plugin_option *options);
+int trace_util_add_option(const char *name, const char *val);
+int trace_util_load_plugins(struct tep_handle *pevent, const char *suffix,
+			    int (*load_plugin)(struct tep_handle *pevent,
+					       const char *path,
+					       const char *name,
+					       void *data),
+			    void *data);
+struct tep_plugin_option *trace_util_read_plugin_options(void);
+void trace_util_free_options(struct tep_plugin_option *options);
 char **trace_util_find_plugin_files(const char *suffix);
 void trace_util_free_plugin_files(char **files);
 
diff --git a/lib/trace-cmd/trace-msg.c b/lib/trace-cmd/trace-msg.c
index 4def7e6..347aee5 100644
--- a/lib/trace-cmd/trace-msg.c
+++ b/lib/trace-cmd/trace-msg.c
@@ -16,6 +16,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdarg.h>
+#include <string.h>
 #include <unistd.h>
 #include <arpa/inet.h>
 #include <sys/types.h>
@@ -59,6 +60,17 @@ struct tracecmd_msg_rinit {
 	be32 cpus;
 } __attribute__((packed));
 
+struct tracecmd_msg_trace_req {
+	be32 flags;
+	be32 argc;
+} __attribute__((packed));
+
+struct tracecmd_msg_trace_resp {
+	be32 flags;
+	be32 cpus;
+	be32 page_size;
+} __attribute__((packed));
+
 struct tracecmd_msg_header {
 	be32	size;
 	be32	cmd;
@@ -71,7 +83,10 @@ struct tracecmd_msg_header {
 	C(RINIT,	2,	sizeof(struct tracecmd_msg_rinit)),	\
 	C(SEND_DATA,	3,	0),					\
 	C(FIN_DATA,	4,	0),					\
-	C(NOT_SUPP,	5,	0),
+	C(NOT_SUPP,	5,	0),					\
+	C(TRACE_REQ,	6,	sizeof(struct tracecmd_msg_trace_req)),	\
+	C(TRACE_RESP,	7,	sizeof(struct tracecmd_msg_trace_resp)),\
+	C(CLOSE_RESP,	8,	0),
 
 #undef C
 #define C(a,b,c)	MSG_##a = b
@@ -103,6 +118,8 @@ struct tracecmd_msg {
 	union {
 		struct tracecmd_msg_tinit	tinit;
 		struct tracecmd_msg_rinit	rinit;
+		struct tracecmd_msg_trace_req	trace_req;
+		struct tracecmd_msg_trace_resp	trace_resp;
 	};
 	char					*buf;
 } __attribute__((packed));
@@ -625,6 +642,14 @@ int tracecmd_msg_send_close_msg(struct tracecmd_msg_handle *msg_handle)
 	return tracecmd_msg_send(msg_handle->fd, &msg);
 }
 
+int tracecmd_msg_send_close_resp_msg(struct tracecmd_msg_handle *msg_handle)
+{
+	struct tracecmd_msg msg;
+
+	tracecmd_msg_init(MSG_CLOSE_RESP, &msg);
+	return tracecmd_msg_send(msg_handle->fd, &msg);
+}
+
 int tracecmd_msg_data_send(struct tracecmd_msg_handle *msg_handle,
 			   const char *buf, int size)
 {
@@ -745,7 +770,7 @@ int tracecmd_msg_collect_data(struct tracecmd_msg_handle *msg_handle, int ofd)
 	return tracecmd_msg_wait_close(msg_handle);
 }
 
-int tracecmd_msg_wait_close(struct tracecmd_msg_handle *msg_handle)
+static int tracecmd_msg_wait_for_cmd(struct tracecmd_msg_handle *msg_handle, enum tracecmd_msg_cmd cmd)
 {
 	struct tracecmd_msg msg;
 	int ret = -1;
@@ -756,7 +781,7 @@ int tracecmd_msg_wait_close(struct tracecmd_msg_handle *msg_handle)
 		if (ret < 0)
 			goto error;
 
-		if (ntohl(msg.hdr.cmd) == MSG_CLOSE)
+		if (ntohl(msg.hdr.cmd) == cmd)
 			return 0;
 
 		error_operation(&msg);
@@ -771,3 +796,206 @@ error:
 	msg_free(&msg);
 	return ret;
 }
+
+int tracecmd_msg_wait_close(struct tracecmd_msg_handle *msg_handle)
+{
+	return tracecmd_msg_wait_for_cmd(msg_handle, MSG_CLOSE);
+}
+
+int tracecmd_msg_wait_close_resp(struct tracecmd_msg_handle *msg_handle)
+{
+	return tracecmd_msg_wait_for_cmd(msg_handle, MSG_CLOSE_RESP);
+}
+
+static int make_trace_req(struct tracecmd_msg *msg, int argc, char **argv)
+{
+	size_t args_size = 0;
+	char *p;
+	int i;
+
+	for (i = 0; i < argc; i++)
+		args_size += strlen(argv[i]) + 1;
+
+	msg->hdr.size = htonl(ntohl(msg->hdr.size) + args_size);
+	msg->trace_req.argc = htonl(argc);
+	msg->buf = calloc(args_size, 1);
+	if (!msg->buf)
+		return -ENOMEM;
+
+	p = msg->buf;
+	for (i = 0; i < argc; i++)
+		p = stpcpy(p, argv[i]) + 1;
+
+	return 0;
+}
+
+int tracecmd_msg_send_trace_req(struct tracecmd_msg_handle *msg_handle,
+				int argc, char **argv)
+{
+	struct tracecmd_msg msg;
+	int ret;
+
+	tracecmd_msg_init(MSG_TRACE_REQ, &msg);
+	ret = make_trace_req(&msg, argc, argv);
+	if (ret < 0)
+		return ret;
+
+	return tracecmd_msg_send(msg_handle->fd, &msg);
+}
+
+ /*
+  * NOTE: On success, the returned `argv` should be freed with:
+  *     free(argv[0]);
+  *     free(argv);
+  */
+int tracecmd_msg_recv_trace_req(struct tracecmd_msg_handle *msg_handle,
+				int *argc, char ***argv)
+{
+	struct tracecmd_msg msg;
+	char *p, *buf_end, **args;
+	int i, ret, nr_args;
+	ssize_t buf_len;
+
+	ret = tracecmd_msg_recv(msg_handle->fd, &msg);
+	if (ret < 0)
+		return ret;
+
+	if (ntohl(msg.hdr.cmd) != MSG_TRACE_REQ) {
+		ret = -ENOTSUP;
+		goto out;
+	}
+
+	nr_args = ntohl(msg.trace_req.argc);
+	if (nr_args <= 0) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	buf_len = ntohl(msg.hdr.size) - MSG_HDR_LEN - ntohl(msg.hdr.cmd_size);
+	buf_end = (char *)msg.buf + buf_len;
+	if (buf_len <= 0 && ((char *)msg.buf)[buf_len-1] != '\0') {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	args = calloc(nr_args, sizeof(*args));
+	if (!args) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	for (i = 0, p = msg.buf; i < nr_args; i++, p++) {
+		if (p >= buf_end) {
+			ret = -EINVAL;
+			goto out_args;
+		}
+		args[i] = p;
+		p = strchr(p, '\0');
+	}
+
+	*argc = nr_args;
+	*argv = args;
+
+	/*
+	 * On success we're passing msg.buf to the caller through argv[0] so we
+	 * reset it here before calling msg_free().
+	 */
+	msg.buf = NULL;
+	msg_free(&msg);
+	return 0;
+
+out_args:
+	free(args);
+out:
+	error_operation(&msg);
+	if (ret == -EOPNOTSUPP)
+		handle_unexpected_msg(msg_handle, &msg);
+	msg_free(&msg);
+	return ret;
+}
+
+static int make_trace_resp(struct tracecmd_msg *msg,
+			   int page_size, int nr_cpus, unsigned int *ports)
+{
+	int data_size;
+
+	data_size = write_uints(NULL, 0, ports, nr_cpus);
+	msg->buf = malloc(data_size);
+	if (!msg->buf)
+		return -ENOMEM;
+	write_uints(msg->buf, data_size, ports, nr_cpus);
+
+	msg->hdr.size = htonl(ntohl(msg->hdr.size) + data_size);
+	msg->trace_resp.cpus = htonl(nr_cpus);
+	msg->trace_resp.page_size = htonl(page_size);
+
+	return 0;
+}
+
+int tracecmd_msg_send_trace_resp(struct tracecmd_msg_handle *msg_handle,
+				 int nr_cpus, int page_size,
+				 unsigned int *ports)
+{
+	struct tracecmd_msg msg;
+	int ret;
+
+	tracecmd_msg_init(MSG_TRACE_RESP, &msg);
+	ret = make_trace_resp(&msg, page_size, nr_cpus, ports);
+	if (ret < 0)
+		return ret;
+
+	return tracecmd_msg_send(msg_handle->fd, &msg);
+}
+
+int tracecmd_msg_recv_trace_resp(struct tracecmd_msg_handle *msg_handle,
+				 int *nr_cpus, int *page_size,
+				 unsigned int **ports)
+{
+	struct tracecmd_msg msg;
+	char *p, *buf_end;
+	ssize_t buf_len;
+	int i, ret;
+
+	ret = tracecmd_msg_recv(msg_handle->fd, &msg);
+	if (ret < 0)
+		return ret;
+
+	if (ntohl(msg.hdr.cmd) != MSG_TRACE_RESP) {
+		ret = -ENOTSUP;
+		goto out;
+	}
+
+	buf_len = msg_buf_len(&msg);
+	if (buf_len <= 0) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	*nr_cpus = ntohl(msg.trace_resp.cpus);
+	*page_size = ntohl(msg.trace_resp.page_size);
+	*ports = calloc(*nr_cpus, sizeof(**ports));
+	if (!*ports) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	buf_end = msg.buf + buf_len;
+	for (i = 0, p = msg.buf; i < *nr_cpus; i++, p++) {
+		if (p >= buf_end || tatou(p, &(*ports)[i])) {
+			free(*ports);
+			ret = -EINVAL;
+			goto out;
+		}
+		p = strchr(p, '\0');
+	}
+
+	msg_free(&msg);
+	return 0;
+
+out:
+	error_operation(&msg);
+	if (ret == -EOPNOTSUPP)
+		handle_unexpected_msg(msg_handle, &msg);
+	msg_free(&msg);
+	return ret;
+}
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 05/13] trace-cmd: Add buffer instance flags for tracing in guest and agent context
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (3 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 04/13] trace-cmd: Add VM tracing protocol messages Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 06/13] trace-cmd: Add VM kernel tracing over vsockets transport Slavomir Kaslev
                   ` (7 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Add BUFFER_FL_GUEST and BUFFER_FL_AGENT flags to differentiate when
trace-record.c is being called to trace guest or the VM tracing agent.

Also disable functions talking to the local tracefs when called in recording
guest instances context.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/include/trace-local.h |  5 +++
 tracecmd/trace-record.c        | 58 ++++++++++++++++++++++++++++++++--
 2 files changed, 61 insertions(+), 2 deletions(-)

diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
index 05760d8..24fad8d 100644
--- a/tracecmd/include/trace-local.h
+++ b/tracecmd/include/trace-local.h
@@ -146,6 +146,8 @@ char *strstrip(char *str);
 enum buffer_instance_flags {
 	BUFFER_FL_KEEP		= 1 << 0,
 	BUFFER_FL_PROFILE	= 1 << 1,
+	BUFFER_FL_GUEST		= 1 << 2,
+	BUFFER_FL_AGENT		= 1 << 3,
 };
 
 struct func_list {
@@ -207,6 +209,9 @@ extern struct buffer_instance *first_instance;
 #define for_all_instances(i) for (i = first_instance; i; \
 				  i = i == &top_instance ? buffer_instances : (i)->next)
 
+#define is_agent(instance)	((instance)->flags & BUFFER_FL_AGENT)
+#define is_guest(instance)	((instance)->flags & BUFFER_FL_GUEST)
+
 struct buffer_instance *create_instance(const char *name);
 void add_instance(struct buffer_instance *instance, int cpu_count);
 char *get_instance_file(struct buffer_instance *instance, const char *file);
diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index 69de82a..5f0e8d3 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -828,6 +828,9 @@ static void __clear_trace(struct buffer_instance *instance)
 	FILE *fp;
 	char *path;
 
+	if (is_guest(instance))
+		return;
+
 	/* reset the trace */
 	path = get_instance_file(instance, "trace");
 	fp = fopen(path, "w");
@@ -1536,6 +1539,9 @@ set_plugin_instance(struct buffer_instance *instance, const char *name)
 	char *path;
 	char zero = '0';
 
+	if (is_guest(instance))
+		return;
+
 	path = get_instance_file(instance, "current_tracer");
 	fp = fopen(path, "w");
 	if (!fp) {
@@ -1632,6 +1638,9 @@ static void disable_func_stack_trace_instance(struct buffer_instance *instance)
 	int size;
 	int ret;
 
+	if (is_guest(instance))
+		return;
+
 	path = get_instance_file(instance, "current_tracer");
 	ret = stat(path, &st);
 	tracecmd_put_tracing_file(path);
@@ -1825,6 +1834,9 @@ reset_events_instance(struct buffer_instance *instance)
 	int i;
 	int ret;
 
+	if (is_guest(instance))
+		return;
+
 	if (use_old_event_method()) {
 		/* old way only had top instance */
 		if (!is_top_instance(instance))
@@ -2137,6 +2149,9 @@ static void write_tracing_on(struct buffer_instance *instance, int on)
 	int ret;
 	int fd;
 
+	if (is_guest(instance))
+		return;
+
 	fd = open_tracing_on(instance);
 	if (fd < 0)
 		return;
@@ -2156,6 +2171,9 @@ static int read_tracing_on(struct buffer_instance *instance)
 	char buf[10];
 	int ret;
 
+	if (is_guest(instance))
+		return -1;
+
 	fd = open_tracing_on(instance);
 	if (fd < 0)
 		return fd;
@@ -2302,6 +2320,9 @@ static void update_pid_filters(struct buffer_instance *instance)
 	int ret;
 	int fd;
 
+	if (is_guest(instance))
+		return;
+
 	fd = open_instance_fd(instance, "set_event_pid",
 			      O_WRONLY | O_CLOEXEC | O_TRUNC);
 	if (fd < 0)
@@ -2397,6 +2418,9 @@ static void set_mask(struct buffer_instance *instance)
 	int fd;
 	int ret;
 
+	if (is_guest(instance))
+		return;
+
 	if (!instance->cpumask)
 		return;
 
@@ -2428,6 +2452,9 @@ static void enable_events(struct buffer_instance *instance)
 {
 	struct event_list *event;
 
+	if (is_guest(instance))
+		return;
+
 	for (event = instance->events; event; event = event->next) {
 		if (!event->neg)
 			update_event(event, event->filter, 0, '1');
@@ -2451,6 +2478,9 @@ static void set_clock(struct buffer_instance *instance)
 	char *content;
 	char *str;
 
+	if (is_guest(instance))
+		return;
+
 	if (!instance->clock)
 		return;
 
@@ -2480,6 +2510,9 @@ static void set_max_graph_depth(struct buffer_instance *instance, char *max_grap
 	char *path;
 	int ret;
 
+	if (is_guest(instance))
+		return;
+
 	path = get_instance_file(instance, "max_graph_depth");
 	reset_save_file(path, RESET_DEFAULT_PRIO);
 	tracecmd_put_tracing_file(path);
@@ -2705,6 +2738,9 @@ static void expand_event_instance(struct buffer_instance *instance)
 	struct event_list *compressed_list = instance->events;
 	struct event_list *event;
 
+	if (is_guest(instance))
+		return;
+
 	reset_event_list(instance);
 
 	while (compressed_list) {
@@ -3631,6 +3667,9 @@ static void set_funcs(struct buffer_instance *instance)
 	int set_notrace = 0;
 	int ret;
 
+	if (is_guest(instance))
+		return;
+
 	ret = write_func_file(instance, "set_ftrace_filter", &instance->filter_funcs);
 	if (ret < 0)
 		die("set_ftrace_filter does not exist. Can not filter functions");
@@ -3927,6 +3966,9 @@ static void set_buffer_size_instance(struct buffer_instance *instance)
 	int ret;
 	int fd;
 
+	if (is_guest(instance))
+		return;
+
 	if (!buffer_size)
 		return;
 
@@ -4137,6 +4179,9 @@ static void make_instances(void)
 	int ret;
 
 	for_each_instance(instance) {
+		if (is_guest(instance))
+			continue;
+
 		path = get_instance_dir(instance);
 		ret = stat(path, &st);
 		if (ret < 0) {
@@ -4158,7 +4203,7 @@ void tracecmd_remove_instances(void)
 
 	for_each_instance(instance) {
 		/* Only delete what we created */
-		if (instance->flags & BUFFER_FL_KEEP)
+		if (is_guest(instance) || (instance->flags & BUFFER_FL_KEEP))
 			continue;
 		if (instance->tracing_on_fd > 0) {
 			close(instance->tracing_on_fd);
@@ -4240,7 +4285,7 @@ static void check_function_plugin(void)
 
 static int __check_doing_something(struct buffer_instance *instance)
 {
-	return (instance->flags & BUFFER_FL_PROFILE) ||
+	return is_guest(instance) || (instance->flags & BUFFER_FL_PROFILE) ||
 		instance->plugin || instance->events;
 }
 
@@ -4262,6 +4307,9 @@ update_plugin_instance(struct buffer_instance *instance,
 {
 	const char *plugin = instance->plugin;
 
+	if (is_guest(instance))
+		return;
+
 	if (!plugin)
 		return;
 
@@ -4361,6 +4409,9 @@ static void record_stats(void)
 	int cpu;
 
 	for_all_instances(instance) {
+		if (is_guest(instance))
+			continue;
+
 		s_save = instance->s_save;
 		s_print = instance->s_print;
 		for (cpu = 0; cpu < instance->cpu_count; cpu++) {
@@ -4387,6 +4438,9 @@ static void destroy_stats(void)
 	int cpu;
 
 	for_all_instances(instance) {
+		if (is_guest(instance))
+			continue;
+
 		for (cpu = 0; cpu < instance->cpu_count; cpu++) {
 			trace_seq_destroy(&instance->s_save[cpu]);
 			trace_seq_destroy(&instance->s_print[cpu]);
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 06/13] trace-cmd: Add VM kernel tracing over vsockets transport
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (4 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 05/13] trace-cmd: Add buffer instance flags for tracing in guest and agent context Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 07/13] trace-cmd: Use splice(2) for vsockets if available Slavomir Kaslev
                   ` (6 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

This patch adds VM tracing over vsockets. The new `trace-cmd agent` command
needs run on each guest we want to trace:

     you@guest2 # trace-cmd agent

Then `trace-cmd record` on the host can collect data from both the host and
several guests simultaneously:

     you@host $ trace-cmd record -A guest1 -e irq -e sched \
                                 -A guest2 -p function -e all

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 lib/trace-cmd/trace-recorder.c |  20 +-
 tracecmd/Makefile              |   4 +
 tracecmd/include/trace-local.h |  16 +
 tracecmd/trace-agent.c         | 263 ++++++++++++++
 tracecmd/trace-cmd.c           |   3 +
 tracecmd/trace-record.c        | 614 ++++++++++++++++++++++++++++++---
 tracecmd/trace-usage.c         |  13 +-
 7 files changed, 881 insertions(+), 52 deletions(-)
 create mode 100644 tracecmd/trace-agent.c

diff --git a/lib/trace-cmd/trace-recorder.c b/lib/trace-cmd/trace-recorder.c
index 7a4728f..d248a37 100644
--- a/lib/trace-cmd/trace-recorder.c
+++ b/lib/trace-cmd/trace-recorder.c
@@ -382,11 +382,11 @@ static long splice_data(struct tracecmd_recorder *recorder)
 	read = splice(recorder->trace_fd, NULL, recorder->brass[1], NULL,
 		      recorder->pipe_size, SPLICE_F_MOVE);
 	if (read < 0) {
-		if (errno != EAGAIN && errno != EINTR) {
-			warning("recorder error in splice input");
-			return -1;
-		}
-		return 0;
+		if (errno == EAGAIN || errno == EINTR || errno == ENOTCONN)
+			return 0;
+
+		warning("recorder error in splice input");
+		return -1;
 	} else if (read == 0)
 		return 0;
 
@@ -421,11 +421,11 @@ static long read_data(struct tracecmd_recorder *recorder)
 
 	r = read(recorder->trace_fd, buf, recorder->page_size);
 	if (r < 0) {
-		if (errno != EAGAIN && errno != EINTR) {
-			warning("recorder error in read output");
-			return -1;
-		}
-		return 0;
+		if (errno == EAGAIN || errno == EINTR || errno == ENOTCONN)
+			return 0;
+
+		warning("recorder error in read input");
+		return -1;
 	}
 
 	left = r;
diff --git a/tracecmd/Makefile b/tracecmd/Makefile
index d491aae..fb85e47 100644
--- a/tracecmd/Makefile
+++ b/tracecmd/Makefile
@@ -31,6 +31,10 @@ TRACE_CMD_OBJS += trace-show.o
 TRACE_CMD_OBJS += trace-list.o
 TRACE_CMD_OBJS += trace-usage.o
 
+ifeq ($(VSOCK_DEFINED), 1)
+TRACE_CMD_OBJS += trace-agent.o
+endif
+
 ALL_OBJS := $(TRACE_CMD_OBJS:%.o=$(bdir)/%.o)
 
 all_objs := $(sort $(ALL_OBJS))
diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
index 24fad8d..037f6b5 100644
--- a/tracecmd/include/trace-local.h
+++ b/tracecmd/include/trace-local.h
@@ -12,6 +12,8 @@
 #include "trace-cmd.h"
 #include "event-utils.h"
 
+#define TRACE_AGENT_DEFAULT_PORT	823
+
 /* fix stupid glib guint64 typecasts and printf formats */
 typedef unsigned long long u64;
 
@@ -61,6 +63,8 @@ void trace_split(int argc, char **argv);
 
 void trace_listen(int argc, char **argv);
 
+void trace_agent(int argc, char **argv);
+
 void trace_restore(int argc, char **argv);
 
 void trace_clear(int argc, char **argv);
@@ -85,6 +89,10 @@ void trace_list(int argc, char **argv);
 
 void trace_usage(int argc, char **argv);
 
+int trace_record_agent(struct tracecmd_msg_handle *msg_handle,
+		       int cpus, int *fds,
+		       int argc, char **argv);
+
 struct hook_list;
 
 void trace_init_profile(struct tracecmd_input *handle, struct hook_list *hooks,
@@ -181,6 +189,7 @@ struct buffer_instance {
 	struct func_list	*notrace_funcs;
 
 	const char		*clock;
+	unsigned int		*client_ports;
 
 	struct trace_seq	*s_save;
 	struct trace_seq	*s_print;
@@ -199,6 +208,13 @@ struct buffer_instance {
 	int			tracing_on_fd;
 	int			buffer_size;
 	int			cpu_count;
+
+	int			argc;
+	char			**argv;
+
+	unsigned int		cid;
+	unsigned int		port;
+	int			*fds;
 };
 
 extern struct buffer_instance top_instance;
diff --git a/tracecmd/trace-agent.c b/tracecmd/trace-agent.c
new file mode 100644
index 0000000..89e3568
--- /dev/null
+++ b/tracecmd/trace-agent.c
@@ -0,0 +1,263 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2018 VMware Inc, Slavomir Kaslev <kaslevs@vmware.com>
+ *
+ * based on prior implementation by Yoshihiro Yunomae
+ * Copyright (C) 2013 Hitachi, Ltd.
+ * Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com>
+ */
+
+#include <errno.h>
+#include <fcntl.h>
+#include <getopt.h>
+#include <signal.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <linux/vm_sockets.h>
+
+#include "trace-local.h"
+#include "trace-msg.h"
+
+#define GET_LOCAL_CID	0x7b9
+
+static int get_local_cid(unsigned int *cid)
+{
+	int fd, ret = 0;
+
+	fd = open("/dev/vsock", O_RDONLY);
+	if (fd < 0)
+		return -errno;
+
+	if (ioctl(fd, GET_LOCAL_CID, cid))
+		ret = -errno;
+
+	close(fd);
+	return ret;
+}
+
+static int make_vsock(unsigned int port)
+{
+	struct sockaddr_vm addr = {
+		.svm_family = AF_VSOCK,
+		.svm_cid = VMADDR_CID_ANY,
+		.svm_port = port,
+	};
+	int sd;
+
+	sd = socket(AF_VSOCK, SOCK_STREAM, 0);
+	if (sd < 0)
+		return -errno;
+
+	setsockopt(sd, SOL_SOCKET, SO_REUSEADDR, &(int){1}, sizeof(int));
+
+	if (bind(sd, (struct sockaddr *)&addr, sizeof(addr)))
+		return -errno;
+
+	if (listen(sd, SOMAXCONN))
+		return -errno;
+
+	return sd;
+}
+
+static int get_vsock_port(int sd, unsigned int *port)
+{
+	struct sockaddr_vm addr;
+	socklen_t addr_len = sizeof(addr);
+
+	if (getsockname(sd, (struct sockaddr *)&addr, &addr_len))
+		return -errno;
+
+	if (addr.svm_family != AF_VSOCK)
+		return -EINVAL;
+
+	if (port)
+		*port = addr.svm_port;
+
+	return 0;
+}
+
+static void make_vsocks(int nr, int *fds, unsigned int *ports)
+{
+	unsigned int port;
+	int i, fd, ret;
+
+	for (i = 0; i < nr; i++) {
+		fd = make_vsock(VMADDR_PORT_ANY);
+		if (fd < 0)
+			die("Failed to open vsocket");
+
+		ret = get_vsock_port(fd, &port);
+		if (ret < 0)
+			die("Failed to get vsocket address");
+
+		fds[i] = fd;
+		ports[i] = port;
+	}
+}
+
+static void agent_handle(int sd, int nr_cpus, int page_size)
+{
+	struct tracecmd_msg_handle *msg_handle;
+	unsigned int *ports;
+	char **argv = NULL;
+	int argc = 0;
+	int *fds;
+	int ret;
+
+	fds = calloc(nr_cpus, sizeof(*fds));
+	ports = calloc(nr_cpus, sizeof(*ports));
+	if (!fds || !ports)
+		die("Failed to allocate memory");
+
+	msg_handle = tracecmd_msg_handle_alloc(sd, 0);
+	if (!msg_handle)
+		die("Failed to allocate message handle");
+
+	ret = tracecmd_msg_recv_trace_req(msg_handle, &argc, &argv);
+	if (ret < 0)
+		die("Failed to receive trace request");
+
+	make_vsocks(nr_cpus, fds, ports);
+
+	ret = tracecmd_msg_send_trace_resp(msg_handle, nr_cpus, page_size, ports);
+	if (ret < 0)
+		die("Failed to send trace response");
+
+	trace_record_agent(msg_handle, nr_cpus, fds, argc, argv);
+
+	free(argv[0]);
+	free(argv);
+	free(ports);
+	free(fds);
+	tracecmd_msg_handle_close(msg_handle);
+	exit(0);
+}
+
+static volatile pid_t handler_pid;
+
+static void handle_sigchld(int sig)
+{
+	int wstatus;
+	pid_t pid;
+
+	for (;;) {
+		pid = waitpid(-1, &wstatus, WNOHANG);
+		if (pid <= 0)
+			break;
+
+		if (pid == handler_pid)
+			handler_pid = 0;
+	}
+}
+
+static pid_t do_fork()
+{
+	/* in debug mode, we do not fork off children */
+	if (tracecmd_get_debug())
+		return 0;
+
+	return fork();
+}
+
+static void agent_serve(unsigned int port)
+{
+	int sd, cd, nr_cpus;
+	unsigned int cid;
+	pid_t pid;
+
+	signal(SIGCHLD, handle_sigchld);
+
+	nr_cpus = count_cpus();
+	page_size = getpagesize();
+
+	sd = make_vsock(port);
+	if (sd < 0)
+		die("Failed to open vsocket");
+
+	if (!get_local_cid(&cid))
+		printf("listening on @%u:%u\n", cid, port);
+
+	for (;;) {
+		cd = accept(sd, NULL, NULL);
+		if (cd < 0) {
+			if (errno == EINTR)
+				continue;
+			die("accept");
+		}
+
+		if (handler_pid)
+			goto busy;
+
+		pid = do_fork();
+		if (pid == 0) {
+			close(sd);
+			signal(SIGCHLD, SIG_DFL);
+			agent_handle(cd, nr_cpus, page_size);
+		}
+		if (pid > 0)
+			handler_pid = pid;
+
+busy:
+		close(cd);
+	}
+}
+
+enum {
+	DO_DEBUG	= 255
+};
+
+void trace_agent(int argc, char **argv)
+{
+	bool do_daemon = false;
+	unsigned int port = TRACE_AGENT_DEFAULT_PORT;
+
+	if (argc < 2)
+		usage(argv);
+
+	if (strcmp(argv[1], "agent") != 0)
+		usage(argv);
+
+	for (;;) {
+		int c, option_index = 0;
+		static struct option long_options[] = {
+			{"port", required_argument, NULL, 'p'},
+			{"help", no_argument, NULL, '?'},
+			{"debug", no_argument, NULL, DO_DEBUG},
+			{NULL, 0, NULL, 0}
+		};
+
+		c = getopt_long(argc-1, argv+1, "+hp:D",
+				long_options, &option_index);
+		if (c == -1)
+			break;
+		switch (c) {
+		case 'h':
+			usage(argv);
+			break;
+		case 'p':
+			port = atoi(optarg);
+			break;
+		case 'D':
+			do_daemon = true;
+			break;
+		case DO_DEBUG:
+			tracecmd_set_debug(true);
+			break;
+		default:
+			usage(argv);
+		}
+	}
+
+	if (optind < argc-1)
+		usage(argv);
+
+	if (do_daemon && daemon(1, 0))
+		die("daemon");
+
+	agent_serve(port);
+}
diff --git a/tracecmd/trace-cmd.c b/tracecmd/trace-cmd.c
index 30691b6..3665a78 100644
--- a/tracecmd/trace-cmd.c
+++ b/tracecmd/trace-cmd.c
@@ -80,6 +80,9 @@ struct command commands[] = {
 	{"hist", trace_hist},
 	{"mem", trace_mem},
 	{"listen", trace_listen},
+#ifdef VSOCK
+	{"agent", trace_agent},
+#endif
 	{"split", trace_split},
 	{"restore", trace_restore},
 	{"stack", trace_stack},
diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index 5f0e8d3..79b39af 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -35,6 +35,9 @@
 #include <libgen.h>
 #include <pwd.h>
 #include <grp.h>
+#ifdef VSOCK
+#include <linux/vm_sockets.h>
+#endif
 
 #include "version.h"
 #include "trace-local.h"
@@ -77,8 +80,6 @@ static int buffers;
 static int clear_function_filters;
 
 static char *host;
-static unsigned int *client_ports;
-static int sfd;
 
 static bool quiet;
 
@@ -527,6 +528,25 @@ static char *get_temp_file(struct buffer_instance *instance, int cpu)
 	return file;
 }
 
+static char *get_guest_file(const char *file, const char *guest)
+{
+	const char *p;
+	char *out = NULL;
+	int ret, base_len;
+
+	p = strrchr(file, '.');
+	if (p && p != file)
+		base_len = p - file;
+	else
+		base_len = strlen(file);
+
+	ret = asprintf(&out, "%.*s-%s%s", base_len, file,
+		       guest, file + base_len);
+	if (ret < 0)
+		return NULL;
+	return out;
+}
+
 static void put_temp_file(char *file)
 {
 	free(file);
@@ -632,6 +652,25 @@ static void delete_thread_data(void)
 	}
 }
 
+static void tell_guests_to_stop(void)
+{
+	struct buffer_instance *instance;
+
+	/* Send close message to guests */
+	for_all_instances(instance) {
+		if (is_guest(instance))
+			tracecmd_msg_send_close_msg(instance->msg_handle);
+	}
+
+	/* Wait for guests to acknowledge */
+	for_all_instances(instance) {
+		if (is_guest(instance)) {
+			tracecmd_msg_wait_close_resp(instance->msg_handle);
+			tracecmd_msg_handle_close(instance->msg_handle);
+		}
+	}
+}
+
 static void stop_threads(enum trace_type type)
 {
 	int ret;
@@ -653,6 +692,11 @@ static void stop_threads(enum trace_type type)
 			ret = trace_stream_read(pids, recorder_threads, NULL);
 		} while (ret > 0);
 	}
+}
+
+static void wait_threads()
+{
+	int i;
 
 	for (i = 0; i < recorder_threads; i++) {
 		if (pids[i].pid > 0) {
@@ -2816,14 +2860,14 @@ static void flush(int sig)
 		tracecmd_stop_recording(recorder);
 }
 
-static void connect_port(int cpu)
+static int connect_port(const char *host, unsigned int port)
 {
 	struct addrinfo hints;
 	struct addrinfo *results, *rp;
-	int s;
+	int s, sfd;
 	char buf[BUFSIZ];
 
-	snprintf(buf, BUFSIZ, "%u", client_ports[cpu]);
+	snprintf(buf, BUFSIZ, "%u", port);
 
 	memset(&hints, 0, sizeof(hints));
 	hints.ai_family = AF_UNSPEC;
@@ -2850,7 +2894,189 @@ static void connect_port(int cpu)
 
 	freeaddrinfo(results);
 
-	client_ports[cpu] = sfd;
+	return sfd;
+}
+
+#ifdef VSOCK
+static int open_vsock(unsigned int cid, unsigned int port)
+{
+	struct sockaddr_vm addr = {
+		.svm_family = AF_VSOCK,
+		.svm_cid = cid,
+		.svm_port = port,
+	};
+	int sd;
+
+	sd = socket(AF_VSOCK, SOCK_STREAM, 0);
+	if (sd < 0)
+		return -errno;
+
+	if (connect(sd, (struct sockaddr *)&addr, sizeof(addr)))
+		return -errno;
+
+	return sd;
+}
+#else
+static inline int open_vsock(unsigned int cid, unsigned int port)
+{
+	die("vsock is not supported");
+	return -1;
+}
+#endif
+
+static int do_accept(int sd)
+{
+	int cd;
+
+	for (;;) {
+		cd = accept(sd, NULL, NULL);
+		if (cd < 0) {
+			if (errno == EINTR)
+				continue;
+			die("accept");
+		}
+
+		return cd;
+	}
+
+	return -1;
+}
+
+static bool is_digits(const char *s)
+{
+	for (; *s; s++)
+		if (!isdigit(*s))
+			return false;
+	return true;
+}
+
+struct guest {
+	char *name;
+	int cid;
+	int pid;
+};
+
+static struct guest *guests;
+static size_t guests_len;
+
+static char *get_qemu_guest_name(char *arg)
+{
+	char *tok, *end = arg;
+
+	while ((tok = strsep(&end, ","))) {
+		if (strncmp(tok, "guest=", 6) == 0)
+			return tok + 6;
+	}
+
+	return arg;
+}
+
+static void read_qemu_guests(void)
+{
+	static bool initialized;
+	struct dirent *entry;
+	char path[PATH_MAX];
+	DIR *dir;
+
+	if (initialized)
+		return;
+
+	initialized = true;
+	dir = opendir("/proc");
+	if (!dir)
+		die("Can not open /proc");
+
+	while ((entry = readdir(dir))) {
+		bool is_qemu = false, last_was_name = false;
+		struct guest guest = {};
+		char *p, *arg = NULL;
+		size_t arg_size = 0;
+		FILE *f;
+
+		if (!(entry->d_type == DT_DIR && is_digits(entry->d_name)))
+			continue;
+
+		guest.pid = atoi(entry->d_name);
+		snprintf(path, sizeof(path), "/proc/%s/cmdline", entry->d_name);
+		f = fopen(path, "r");
+		if (!f)
+			continue;
+
+		while (getdelim(&arg, &arg_size, 0, f) != -1) {
+			if (!is_qemu && strstr(arg, "qemu-system-")) {
+				is_qemu = true;
+				continue;
+			}
+
+			if (!is_qemu)
+				continue;
+
+			if (strcmp(arg, "-name") == 0) {
+				last_was_name = true;
+				continue;
+			}
+
+			if (last_was_name) {
+				guest.name = strdup(get_qemu_guest_name(arg));
+				if (!guest.name)
+					die("allocating guest name");
+				last_was_name = false;
+				continue;
+			}
+
+			p = strstr(arg, "guest-cid=");
+			if (p) {
+				guest.cid = atoi(p + 10);
+				continue;
+			}
+		}
+
+		if (!is_qemu)
+			goto next;
+
+		guests = realloc(guests, (guests_len + 1) * sizeof(*guests));
+		if (!guests)
+			die("Can not allocate guest buffer");
+		guests[guests_len++] = guest;
+
+next:
+		free(arg);
+		fclose(f);
+	}
+
+	closedir(dir);
+}
+
+static char *parse_guest_name(char *guest, int *cid, int *port)
+{
+	size_t i;
+	char *p;
+
+	*port = -1;
+	p = strrchr(guest, ':');
+	if (p) {
+		*p = '\0';
+		*port = atoi(p + 1);
+	}
+
+	*cid = -1;
+	p = strrchr(guest, '@');
+	if (p) {
+		*p = '\0';
+		*cid = atoi(p + 1);
+	} else if (is_digits(guest))
+		*cid = atoi(guest);
+
+	read_qemu_guests();
+	for (i = 0; i < guests_len; i++) {
+		if ((*cid > 0 && *cid == guests[i].cid) ||
+		    strcmp(guest, guests[i].name) == 0) {
+			*cid = guests[i].cid;
+			return guests[i].name;
+		}
+	}
+
+	return guest;
 }
 
 static void set_prio(int prio)
@@ -2897,6 +3123,17 @@ create_recorder_instance(struct buffer_instance *instance, const char *file, int
 	struct tracecmd_recorder *record;
 	char *path;
 
+	if (is_guest(instance)) {
+		int fd;
+
+		fd = open_vsock(instance->cid, instance->client_ports[cpu]);
+		if (fd < 0)
+			die("Failed to connect to agent");
+
+		return tracecmd_create_recorder_virt(
+			file, cpu, recorder_flags | TRACECMD_RECORD_NOSPLICE, fd);
+	}
+
 	if (brass)
 		return create_recorder_instance_pipe(instance, cpu, brass);
 
@@ -2921,7 +3158,7 @@ static int create_recorder(struct buffer_instance *instance, int cpu,
 {
 	long ret;
 	char *file;
-	int pid;
+	pid_t pid;
 
 	if (type != TRACE_TYPE_EXTRACT) {
 		signal(SIGUSR1, flush);
@@ -2940,19 +3177,23 @@ static int create_recorder(struct buffer_instance *instance, int cpu,
 		instance->cpu_count = 0;
 	}
 
-	if (client_ports) {
-		char *path;
+	if ((instance->client_ports && !is_guest(instance)) || is_agent(instance)) {
+		unsigned int flags = recorder_flags;
+		char *path = NULL;
+		int fd;
 
-		connect_port(cpu);
-		if (instance->name)
+		if (is_agent(instance))
+			fd = do_accept(instance->fds[cpu]);
+		else
+			fd = connect_port(host, instance->client_ports[cpu]);
+		if (fd < 0)
+			die("Failed connecting to client");
+		if (instance->name && !is_agent(instance))
 			path = get_instance_dir(instance);
 		else
 			path = tracecmd_find_tracing_dir();
-		recorder = tracecmd_create_buffer_recorder_fd(client_ports[cpu],
-							      cpu, recorder_flags,
-							      path);
-		if (instance->name)
-			tracecmd_put_tracing_file(path);
+		recorder = tracecmd_create_buffer_recorder_fd(fd, cpu, flags, path);
+		tracecmd_put_tracing_file(path);
 	} else {
 		file = get_temp_file(instance, cpu);
 		recorder = create_recorder_instance(instance, file, cpu, brass);
@@ -2990,7 +3231,8 @@ static void check_first_msg_from_server(struct tracecmd_msg_handle *msg_handle)
 		die("server not tracecmd server");
 }
 
-static void communicate_with_listener_v1(struct tracecmd_msg_handle *msg_handle)
+static void communicate_with_listener_v1(struct tracecmd_msg_handle *msg_handle,
+					 unsigned int **client_ports)
 {
 	char buf[BUFSIZ];
 	ssize_t n;
@@ -3033,8 +3275,8 @@ static void communicate_with_listener_v1(struct tracecmd_msg_handle *msg_handle)
 		/* No options */
 		write(msg_handle->fd, "0", 2);
 
-	client_ports = malloc(local_cpu_count * sizeof(*client_ports));
-	if (!client_ports)
+	*client_ports = malloc(local_cpu_count * sizeof(*client_ports));
+	if (!*client_ports)
 		die("Failed to allocate client ports for %d cpus", local_cpu_count);
 
 	/*
@@ -3052,13 +3294,14 @@ static void communicate_with_listener_v1(struct tracecmd_msg_handle *msg_handle)
 		if (i == BUFSIZ)
 			die("read bad port number");
 		buf[i] = 0;
-		client_ports[cpu] = atoi(buf);
+		(*client_ports)[cpu] = atoi(buf);
 	}
 }
 
-static void communicate_with_listener_v3(struct tracecmd_msg_handle *msg_handle)
+static void communicate_with_listener_v3(struct tracecmd_msg_handle *msg_handle,
+					 unsigned int **client_ports)
 {
-	if (tracecmd_msg_send_init_data(msg_handle, &client_ports) < 0)
+	if (tracecmd_msg_send_init_data(msg_handle, client_ports) < 0)
 		die("Cannot communicate with server");
 }
 
@@ -3109,7 +3352,7 @@ static void check_protocol_version(struct tracecmd_msg_handle *msg_handle)
 	}
 }
 
-static struct tracecmd_msg_handle *setup_network(void)
+static struct tracecmd_msg_handle *setup_network(struct buffer_instance *instance)
 {
 	struct tracecmd_msg_handle *msg_handle = NULL;
 	struct addrinfo hints;
@@ -3179,11 +3422,11 @@ again:
 			close(sfd);
 			goto again;
 		}
-		communicate_with_listener_v3(msg_handle);
+		communicate_with_listener_v3(msg_handle, &instance->client_ports);
 	}
 
 	if (msg_handle->version == V1_PROTOCOL)
-		communicate_with_listener_v1(msg_handle);
+		communicate_with_listener_v1(msg_handle, &instance->client_ports);
 
 	return msg_handle;
 }
@@ -3196,7 +3439,7 @@ setup_connection(struct buffer_instance *instance, struct common_record_context
 	struct tracecmd_msg_handle *msg_handle;
 	struct tracecmd_output *network_handle;
 
-	msg_handle = setup_network();
+	msg_handle = setup_network(instance);
 
 	/* Now create the handle through this socket */
 	if (msg_handle->version == V3_PROTOCOL) {
@@ -3226,28 +3469,99 @@ static void finish_network(struct tracecmd_msg_handle *msg_handle)
 	free(host);
 }
 
+static void connect_to_agent(struct buffer_instance *instance)
+{
+	struct tracecmd_msg_handle *msg_handle;
+	int sd, ret, nr_cpus, page_size;
+	unsigned int *ports;
+
+	sd = open_vsock(instance->cid, instance->port);
+	if (sd < 0)
+		die("Failed to connect to vsocket @%u:%u",
+		    instance->cid, instance->port);
+
+	msg_handle = tracecmd_msg_handle_alloc(sd, 0);
+	if (!msg_handle)
+		die("Failed to allocate message handle");
+
+	ret = tracecmd_msg_send_trace_req(msg_handle, instance->argc, instance->argv);
+	if (ret < 0)
+		die("Failed to send trace request");
+
+	ret = tracecmd_msg_recv_trace_resp(msg_handle, &nr_cpus, &page_size, &ports);
+	if (ret < 0)
+		die("Failed to receive trace response");
+
+	instance->client_ports = ports;
+	instance->cpu_count = nr_cpus;
+
+	/* the msg_handle now points to the guest fd */
+	instance->msg_handle = msg_handle;
+}
+
+static void setup_guest(struct buffer_instance *instance)
+{
+	struct tracecmd_msg_handle *msg_handle = instance->msg_handle;
+	char *file;
+	int fd;
+
+	/* Create a place to store the guest meta data */
+	file = get_guest_file(output_file, instance->name);
+	if (!file)
+		die("Failed to allocate memory");
+
+	fd = open(file, O_CREAT|O_WRONLY|O_TRUNC, 0644);
+	put_temp_file(file);
+	if (fd < 0)
+		die("Failed to open", file);
+
+	/* Start reading tracing metadata */
+	if (tracecmd_msg_read_data(msg_handle, fd))
+		die("Failed receiving metadata");
+	close(fd);
+}
+
+static void setup_agent(struct buffer_instance *instance, struct common_record_context *ctx)
+{
+	struct tracecmd_output *network_handle;
+
+	network_handle = tracecmd_create_init_fd_msg(instance->msg_handle,
+						     listed_events);
+	add_options(network_handle, ctx);
+	tracecmd_write_cpus(network_handle, instance->cpu_count);
+	tracecmd_write_options(network_handle);
+	tracecmd_msg_finish_sending_data(instance->msg_handle);
+	instance->network_handle = network_handle;
+}
+
 void start_threads(enum trace_type type, struct common_record_context *ctx)
 {
 	struct buffer_instance *instance;
-	int *brass = NULL;
 	int total_cpu_count = 0;
 	int i = 0;
 	int ret;
 
-	for_all_instances(instance)
+	for_all_instances(instance) {
+		/* Start the connection now to find out how many CPUs we need */
+		if (is_guest(instance))
+			connect_to_agent(instance);
 		total_cpu_count += instance->cpu_count;
+	}
 
 	/* make a thread for every CPU we have */
-	pids = malloc(sizeof(*pids) * total_cpu_count * (buffers + 1));
+	pids = calloc(total_cpu_count * (buffers + 1), sizeof(*pids));
 	if (!pids)
-		die("Failed to allocat pids for %d cpus", total_cpu_count);
-
-	memset(pids, 0, sizeof(*pids) * total_cpu_count * (buffers + 1));
+		die("Failed to allocate pids for %d cpus", total_cpu_count);
 
 	for_all_instances(instance) {
+		int *brass = NULL;
 		int x, pid;
 
-		if (host) {
+		if (is_agent(instance)) {
+			setup_agent(instance, ctx);
+		} else if (is_guest(instance)) {
+			setup_guest(instance);
+		} else if (host) {
 			instance->msg_handle = setup_connection(instance, ctx);
 			if (!instance->msg_handle)
 				die("Failed to make connection");
@@ -3466,6 +3780,47 @@ static void add_options(struct tracecmd_output *handle, struct common_record_con
 	add_version(handle);
 }
 
+static void write_guest_file(struct buffer_instance *instance)
+{
+	struct tracecmd_output *handle;
+	int cpu_count = instance->cpu_count;
+	char *file;
+	char **temp_files;
+	int i, fd;
+
+	file = get_guest_file(output_file, instance->name);
+	if (!file)
+		die("Failed to allocate memory");
+
+	fd = open(file, O_RDWR);
+	if (fd < 0)
+		die("error opening %s", file);
+	put_temp_file(file);
+
+	handle = tracecmd_get_output_handle_fd(fd);
+	if (!handle)
+		die("error writing to %s", file);
+
+	temp_files = malloc(sizeof(*temp_files) * cpu_count);
+	if (!temp_files)
+		die("failed to allocate temp_files for %d cpus",
+		    cpu_count);
+
+	for (i = 0; i < cpu_count; i++) {
+		temp_files[i] = get_temp_file(instance, i);
+		if (!temp_files[i])
+			die("failed to allocate memory");
+	}
+
+	if (tracecmd_write_cpu_data(handle, cpu_count, temp_files) < 0)
+		die("failed to write CPU data");
+	tracecmd_output_close(handle);
+
+	for (i = 0; i < cpu_count; i++)
+		put_temp_file(temp_files[i]);
+	free(temp_files);
+}
+
 static void record_data(struct common_record_context *ctx)
 {
 	struct tracecmd_option **buffer_options;
@@ -3477,7 +3832,9 @@ static void record_data(struct common_record_context *ctx)
 	int i;
 
 	for_all_instances(instance) {
-		if (instance->msg_handle)
+		if (is_guest(instance))
+			write_guest_file(instance);
+		else if (host && instance->msg_handle)
 			finish_network(instance->msg_handle);
 		else
 			local = true;
@@ -4717,6 +5074,7 @@ void trace_stop(int argc, char **argv)
 		c = getopt(argc-1, argv+1, "hatB:");
 		if (c == -1)
 			break;
+
 		switch (c) {
 		case 'h':
 			usage(argv);
@@ -4885,6 +5243,65 @@ static void init_common_record_context(struct common_record_context *ctx,
 #define IS_STREAM(ctx) ((ctx)->curr_cmd == CMD_stream)
 #define IS_PROFILE(ctx) ((ctx)->curr_cmd == CMD_profile)
 #define IS_RECORD(ctx) ((ctx)->curr_cmd == CMD_record)
+#define IS_RECORD_AGENT(ctx) ((ctx)->curr_cmd == CMD_record_agent)
+
+static void add_argv(struct buffer_instance *instance, char *arg, bool prepend)
+{
+	instance->argv = realloc(instance->argv,
+				 (instance->argc + 1) * sizeof(char *));
+	if (!instance->argv)
+		die("Can not allocate instance args");
+	if (prepend) {
+		memmove(instance->argv + 1, instance->argv,
+			instance->argc * sizeof(*instance->argv));
+		instance->argv[0] = arg;
+	} else {
+		instance->argv[instance->argc] = arg;
+	}
+	instance->argc++;
+}
+
+static void add_arg(struct buffer_instance *instance,
+		    int c, const char *opts,
+		    struct option *long_options, char *optarg)
+{
+	char *ptr, *arg;
+	int i, ret;
+
+	/* Short or long arg */
+	if (!(c & 0x80)) {
+		ptr = strchr(opts, c);
+		if (!ptr)
+			return; /* Not found? */
+		ret = asprintf(&arg, "-%c", c);
+		if (ret < 0)
+			die("Can not allocate argument");
+		add_argv(instance, arg, false);
+		if (ptr[1] == ':') {
+			arg = strdup(optarg);
+			if (!arg)
+				die("Can not allocate arguments");
+			add_argv(instance, arg, false);
+		}
+		return;
+	}
+	for (i = 0; long_options[i].name; i++) {
+		if (c != long_options[i].val)
+			continue;
+		ret = asprintf(&arg, "--%s", long_options[i].name);
+		if (ret < 0)
+			die("Can not allocate argument");
+		add_argv(instance, arg, false);
+		if (long_options[i].has_arg) {
+			arg = strdup(optarg);
+			if (!arg)
+				die("Can not allocate arguments");
+			add_argv(instance, arg, false);
+		}
+		return;
+	}
+	/* Not found? */
+}
 
 static void parse_record_options(int argc,
 				 char **argv,
@@ -4898,6 +5315,7 @@ static void parse_record_options(int argc,
 	char *pids;
 	char *pid;
 	char *sav;
+	int name_counter = 0;
 	int neg_event = 0;
 
 	init_common_record_context(ctx, curr_cmd);
@@ -4929,10 +5347,20 @@ static void parse_record_options(int argc,
 		if (IS_EXTRACT(ctx))
 			opts = "+haf:Fp:co:O:sr:g:l:n:P:N:tb:B:ksiT";
 		else
-			opts = "+hae:f:Fp:cC:dDGo:O:s:r:vg:l:n:P:N:tb:R:B:ksSiTm:M:H:q";
+			opts = "+hae:f:FA:p:cC:dDGo:O:s:r:vg:l:n:P:N:tb:R:B:ksSiTm:M:H:q";
 		c = getopt_long (argc-1, argv+1, opts, long_options, &option_index);
 		if (c == -1)
 			break;
+
+		/*
+		 * If the current instance is to record a guest, then save
+		 * all the arguments for this instance.
+		 */
+		if (c != 'B' && c != 'A' && is_guest(ctx->instance)) {
+			add_arg(ctx->instance, c, opts, long_options, optarg);
+			continue;
+		}
+
 		switch (c) {
 		case 'h':
 			usage(argv);
@@ -4985,6 +5413,31 @@ static void parse_record_options(int argc,
 			add_trigger(event, optarg);
 			break;
 
+		case 'A': {
+			char *name = NULL;
+			int cid = -1, port = -1;
+
+			if (!IS_RECORD(ctx))
+				die("-A is only allowed for record operations");
+
+			name = parse_guest_name(optarg, &cid, &port);
+			if (cid == -1)
+				die("guest %s not found", optarg);
+			if (port == -1)
+				port = TRACE_AGENT_DEFAULT_PORT;
+			if (!name || !*name) {
+				ret = asprintf(&name, "unnamed-%d", name_counter++);
+				if (ret < 0)
+					die("Failed to allocate guest name");
+			}
+
+			ctx->instance = create_instance(name);
+			ctx->instance->flags |= BUFFER_FL_GUEST;
+			ctx->instance->cid = cid;
+			ctx->instance->port = port;
+			add_instance(ctx->instance, 0);
+			break;
+		}
 		case 'F':
 			test_set_event_pid();
 			filter_task = 1;
@@ -5056,6 +5509,8 @@ static void parse_record_options(int argc,
 			ctx->disable = 1;
 			break;
 		case 'o':
+			if (IS_RECORD_AGENT(ctx))
+				die("-o incompatible with agent recording");
 			if (host)
 				die("-o incompatible with -N");
 			if (IS_START(ctx))
@@ -5117,6 +5572,8 @@ static void parse_record_options(int argc,
 		case 'N':
 			if (!IS_RECORD(ctx))
 				die("-N only available with record");
+			if (IS_RECORD_AGENT(ctx))
+				die("-N incompatible with agent recording");
 			if (ctx->output)
 				die("-N incompatible with -o");
 			host = optarg;
@@ -5224,6 +5681,16 @@ static void parse_record_options(int argc,
 		}
 	}
 
+	/* If --date is specified, prepend it to all guest VM flags */
+	if (ctx->date) {
+		struct buffer_instance *instance;
+
+		for_all_instances(instance) {
+			if (is_guest(instance))
+				add_argv(instance, "--date", true);
+		}
+	}
+
 	if (!ctx->filtered && ctx->instance->filter_mod)
 		add_func(&ctx->instance->filter_funcs,
 			 ctx->instance->filter_mod, "*");
@@ -5263,7 +5730,8 @@ static enum trace_type get_trace_cmd_type(enum trace_cmd cmd)
 		{CMD_stream, TRACE_TYPE_STREAM},
 		{CMD_extract, TRACE_TYPE_EXTRACT},
 		{CMD_profile, TRACE_TYPE_STREAM},
-		{CMD_start, TRACE_TYPE_START}
+		{CMD_start, TRACE_TYPE_START},
+		{CMD_record_agent, TRACE_TYPE_RECORD}
 	};
 
 	for (int i = 0; i < ARRAY_SIZE(trace_type_per_command); i++) {
@@ -5295,12 +5763,30 @@ static void finalize_record_trace(struct common_record_context *ctx)
 		if (instance->flags & BUFFER_FL_KEEP)
 			write_tracing_on(instance,
 					 instance->tracing_on_init_val);
+		if (is_agent(instance)) {
+			tracecmd_msg_send_close_resp_msg(instance->msg_handle);
+			tracecmd_output_close(instance->network_handle);
+		}
 	}
 
 	if (host)
 		tracecmd_output_close(ctx->instance->network_handle);
 }
 
+static bool has_local_instances(void)
+{
+	struct buffer_instance *instance;
+
+	for_all_instances(instance) {
+		if (is_guest(instance))
+			continue;
+		if (host && instance->msg_handle)
+			continue;
+		return true;
+	}
+	return false;
+}
+
 /*
  * This function contains common code for the following commands:
  * record, start, stream, profile.
@@ -5330,7 +5816,6 @@ static void record_trace(int argc, char **argv,
 
 	/* Save the state of tracing_on before starting */
 	for_all_instances(instance) {
-
 		if (!ctx->manual && instance->flags & BUFFER_FL_PROFILE)
 			enable_profile(instance);
 
@@ -5347,14 +5832,16 @@ static void record_trace(int argc, char **argv,
 
 	page_size = getpagesize();
 
-	fset = set_ftrace(!ctx->disable, ctx->total_disable);
+	if (!is_guest(ctx->instance))
+		fset = set_ftrace(!ctx->disable, ctx->total_disable);
 	tracecmd_disable_all_tracing(1);
 
 	for_all_instances(instance)
 		set_clock(instance);
 
 	/* Record records the date first */
-	if (IS_RECORD(ctx) && ctx->date)
+	if (ctx->date &&
+	    ((IS_RECORD(ctx) && has_local_instances()) || IS_RECORD_AGENT(ctx)))
 		ctx->date2ts = get_date_to_ts();
 
 	for_all_instances(instance) {
@@ -5391,9 +5878,13 @@ static void record_trace(int argc, char **argv,
 		exit(0);
 	}
 
-	if (ctx->run_command)
+	if (ctx->run_command) {
 		run_cmd(type, ctx->user, (argc - optind) - 1, &argv[optind + 1]);
-	else {
+	} else if (ctx->instance && is_agent(ctx->instance)) {
+		update_task_filter();
+		tracecmd_enable_tracing();
+		tracecmd_msg_wait_close(ctx->instance->msg_handle);
+	} else {
 		update_task_filter();
 		tracecmd_enable_tracing();
 		/* We don't ptrace ourself */
@@ -5409,6 +5900,7 @@ static void record_trace(int argc, char **argv,
 			trace_or_sleep(type);
 	}
 
+	tell_guests_to_stop();
 	tracecmd_disable_tracing();
 	if (!latency)
 		stop_threads(type);
@@ -5418,6 +5910,9 @@ static void record_trace(int argc, char **argv,
 	if (!keep)
 		tracecmd_disable_all_tracing(0);
 
+	if (!latency)
+		wait_threads();
+
 	if (IS_RECORD(ctx)) {
 		record_data(ctx);
 		delete_thread_data();
@@ -5554,3 +6049,40 @@ void trace_record(int argc, char **argv)
 	record_trace(argc, argv, &ctx);
 	exit(0);
 }
+
+int trace_record_agent(struct tracecmd_msg_handle *msg_handle,
+		       int cpus, int *fds,
+		       int argc, char **argv)
+{
+	struct common_record_context ctx;
+	char **argv_plus;
+
+	/* Reset optind for getopt_long */
+	optind = 1;
+	/*
+	 * argc is the number of elements in argv, but we need to convert
+	 * argc and argv into "trace-cmd", "record", argv.
+	 * where argc needs to grow by two.
+	 */
+	argv_plus = calloc(argc + 2, sizeof(char *));
+	if (!argv_plus)
+		die("Failed to allocate record arguments");
+
+	argv_plus[0] = "trace-cmd";
+	argv_plus[1] = "record";
+	memmove(argv_plus + 2, argv, argc * sizeof(char *));
+	argc += 2;
+
+	parse_record_options(argc, argv_plus, CMD_record_agent, &ctx);
+	if (ctx.run_command)
+		return -EINVAL;
+
+	ctx.instance->fds = fds;
+	ctx.instance->flags |= BUFFER_FL_AGENT;
+	ctx.instance->msg_handle = msg_handle;
+	msg_handle->version = V3_PROTOCOL;
+	record_trace(argc, argv, &ctx);
+
+	free(argv_plus);
+	return 0;
+}
diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c
index ef03eea..20273ba 100644
--- a/tracecmd/trace-usage.c
+++ b/tracecmd/trace-usage.c
@@ -235,11 +235,22 @@ static struct usage_help usage_help[] = {
 		"listen on a network socket for trace clients",
 		" %s listen -p port[-D][-o file][-d dir][-l logfile]\n"
 		"          Creates a socket to listen for clients.\n"
-		"          -D create it in daemon mode.\n"
+		"          -p port number to listen on.\n"
+		"          -D run in daemon mode.\n"
 		"          -o file name to use for clients.\n"
 		"          -d directory to store client files.\n"
 		"          -l logfile to write messages to.\n"
 	},
+#ifdef VSOCK
+	{
+		"agent",
+		"listen on a vsocket for trace clients",
+		" %s agent -p port[-D]\n"
+		"          Creates a vsocket to listen for clients.\n"
+		"          -p port number to listen on.\n"
+		"          -D run in daemon mode.\n"
+	},
+#endif
 	{
 		"list",
 		"list the available events, plugins or options",
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 07/13] trace-cmd: Use splice(2) for vsockets if available
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (5 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 06/13] trace-cmd: Add VM kernel tracing over vsockets transport Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 08/13] trace-cmd: Switch stop recording signal to SIGUSR1 Slavomir Kaslev
                   ` (5 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Detect if splice(2) reading is supported for vsockets (Linux 4.20 and
later) and use it, or fallback to read/write-ing otherwise.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/trace-record.c | 56 +++++++++++++++++++++++++++++++++++++++--
 1 file changed, 54 insertions(+), 2 deletions(-)

diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index 79b39af..b974d09 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -2916,12 +2916,61 @@ static int open_vsock(unsigned int cid, unsigned int port)
 
 	return sd;
 }
+
+static int try_splice_read_vsock(void)
+{
+	int ret, sd, brass[2];
+
+	sd = socket(AF_VSOCK, SOCK_STREAM, 0);
+	if (sd < 0)
+		return -errno;
+
+	ret = pipe(brass);
+	if (ret < 0)
+		goto out_close_sd;
+
+	/*
+	 * On kernels that don't support splice reading from vsockets
+	 * this will fail with EINVAL, or ENOTCONN otherwise.
+	 * Technically, it should never succeed but if it does, claim splice
+	 * reading is supported.
+	 */
+	ret = splice(sd, NULL, brass[1], NULL, 10, 0);
+	if (ret < 0)
+		ret = errno != EINVAL;
+	else
+		ret = 1;
+
+	close(brass[0]);
+	close(brass[1]);
+out_close_sd:
+	close(sd);
+	return ret;
+}
+
+static bool can_splice_read_vsock(void)
+{
+	static bool initialized, res;
+
+	if (initialized)
+		return res;
+
+	res = try_splice_read_vsock() > 0;
+	initialized = true;
+	return res;
+}
+
 #else
 static inline int open_vsock(unsigned int cid, unsigned int port)
 {
 	die("vsock is not supported");
 	return -1;
 }
+
+static bool can_splice_read_vsock(void)
+{
+	return false;
+}
 #endif
 
 static int do_accept(int sd)
@@ -3125,13 +3174,16 @@ create_recorder_instance(struct buffer_instance *instance, const char *file, int
 
 	if (is_guest(instance)) {
 		int fd;
+		unsigned int flags;
 
 		fd = open_vsock(instance->cid, instance->client_ports[cpu]);
 		if (fd < 0)
 			die("Failed to connect to agent");
 
-		return tracecmd_create_recorder_virt(
-			file, cpu, recorder_flags | TRACECMD_RECORD_NOSPLICE, fd);
+		flags = recorder_flags;
+		if (!can_splice_read_vsock())
+			flags |= TRACECMD_RECORD_NOSPLICE;
+		return tracecmd_create_recorder_virt(file, cpu, flags, fd);
 	}
 
 	if (brass)
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 08/13] trace-cmd: Switch stop recording signal to SIGUSR1
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (6 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 07/13] trace-cmd: Use splice(2) for vsockets if available Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 09/13] trace-cmd: Add `trace-cmd setup-guest` command Slavomir Kaslev
                   ` (4 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Using Ctrl-C to stop trace-cmd recording from the command line delivers SIGINT
to all its child processes. Switch internal recording stop notification to
SIGUSR1 so that notifying recording child processes can be delayed in order to
do proper flushing when doing remote VM tracing.

Also remove the flush() function, which is only used by trace-cmd listen, and
use finish() instead without changing listener's behavior.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/trace-record.c | 12 ++++--------
 1 file changed, 4 insertions(+), 8 deletions(-)

diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index b974d09..94180c2 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -682,7 +682,7 @@ static void stop_threads(enum trace_type type)
 	/* Tell all threads to finish up */
 	for (i = 0; i < recorder_threads; i++) {
 		if (pids[i].pid > 0) {
-			kill(pids[i].pid, SIGINT);
+			kill(pids[i].pid, SIGUSR1);
 		}
 	}
 
@@ -2854,12 +2854,6 @@ static void finish(int sig)
 	finished = 1;
 }
 
-static void flush(int sig)
-{
-	if (recorder)
-		tracecmd_stop_recording(recorder);
-}
-
 static int connect_port(const char *host, unsigned int port)
 {
 	struct addrinfo hints;
@@ -3213,7 +3207,6 @@ static int create_recorder(struct buffer_instance *instance, int cpu,
 	pid_t pid;
 
 	if (type != TRACE_TYPE_EXTRACT) {
-		signal(SIGUSR1, flush);
 
 		pid = fork();
 		if (pid < 0)
@@ -3222,6 +3215,9 @@ static int create_recorder(struct buffer_instance *instance, int cpu,
 		if (pid)
 			return pid;
 
+		signal(SIGINT, SIG_IGN);
+		signal(SIGUSR1, finish);
+
 		if (rt_prio)
 			set_prio(rt_prio);
 
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 09/13] trace-cmd: Add `trace-cmd setup-guest` command
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (7 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 08/13] trace-cmd: Switch stop recording signal to SIGUSR1 Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 10/13] trace-cmd: Try to autodetect number of guest CPUs in setup-guest if not specified Slavomir Kaslev
                   ` (3 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Add `trace-cmd setup-guest` command that creates the necessary FIFOs for tracing
a guest over FIFOs instead of vsockets.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/Makefile              |   1 +
 tracecmd/include/trace-local.h |   6 ++
 tracecmd/trace-cmd.c           |   1 +
 tracecmd/trace-setup-guest.c   | 178 +++++++++++++++++++++++++++++++++
 tracecmd/trace-usage.c         |   8 ++
 5 files changed, 194 insertions(+)
 create mode 100644 tracecmd/trace-setup-guest.c

diff --git a/tracecmd/Makefile b/tracecmd/Makefile
index fb85e47..29a623b 100644
--- a/tracecmd/Makefile
+++ b/tracecmd/Makefile
@@ -33,6 +33,7 @@ TRACE_CMD_OBJS += trace-usage.o
 
 ifeq ($(VSOCK_DEFINED), 1)
 TRACE_CMD_OBJS += trace-agent.o
+TRACE_CMD_OBJS += trace-setup-guest.o
 endif
 
 ALL_OBJS := $(TRACE_CMD_OBJS:%.o=$(bdir)/%.o)
diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
index 037f6b5..3bada82 100644
--- a/tracecmd/include/trace-local.h
+++ b/tracecmd/include/trace-local.h
@@ -14,6 +14,10 @@
 
 #define TRACE_AGENT_DEFAULT_PORT	823
 
+#define GUEST_PIPE_NAME		"trace-pipe-cpu"
+#define GUEST_DIR_FMT		"/var/lib/trace-cmd/virt/%s"
+#define GUEST_FIFO_FMT		GUEST_DIR_FMT "/" GUEST_PIPE_NAME "%d"
+
 /* fix stupid glib guint64 typecasts and printf formats */
 typedef unsigned long long u64;
 
@@ -65,6 +69,8 @@ void trace_listen(int argc, char **argv);
 
 void trace_agent(int argc, char **argv);
 
+void trace_setup_guest(int argc, char **argv);
+
 void trace_restore(int argc, char **argv);
 
 void trace_clear(int argc, char **argv);
diff --git a/tracecmd/trace-cmd.c b/tracecmd/trace-cmd.c
index 3665a78..9cbfce0 100644
--- a/tracecmd/trace-cmd.c
+++ b/tracecmd/trace-cmd.c
@@ -82,6 +82,7 @@ struct command commands[] = {
 	{"listen", trace_listen},
 #ifdef VSOCK
 	{"agent", trace_agent},
+	{"setup-guest", trace_setup_guest},
 #endif
 	{"split", trace_split},
 	{"restore", trace_restore},
diff --git a/tracecmd/trace-setup-guest.c b/tracecmd/trace-setup-guest.c
new file mode 100644
index 0000000..2f8bfeb
--- /dev/null
+++ b/tracecmd/trace-setup-guest.c
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2019 VMware Inc, Slavomir Kaslev <kaslevs@vmware.com>
+ *
+ */
+
+#include <errno.h>
+#include <fcntl.h>
+#include <getopt.h>
+#include <grp.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+#include "trace-local.h"
+#include "trace-msg.h"
+
+static int make_dir(const char *path, mode_t mode)
+{
+	char buf[PATH_MAX+2], *p;
+
+	strncpy(buf, path, sizeof(buf));
+	if (buf[PATH_MAX])
+		return -E2BIG;
+
+	for (p = buf; *p; p++) {
+		p += strspn(p, "/");
+		p += strcspn(p, "/");
+		*p = '\0';
+		if (mkdir(buf, mode) < 0 && errno != EEXIST)
+			return -errno;
+		*p = '/';
+	}
+
+	return 0;
+}
+
+static int make_fifo(const char *path, mode_t mode)
+{
+	struct stat st;
+
+	if (!stat(path, &st)) {
+		if (S_ISFIFO(st.st_mode))
+			return 0;
+		return -EEXIST;
+	}
+
+	if (mkfifo(path, mode))
+		return -errno;
+	return 0;
+}
+
+static int make_guest_dir(const char *guest)
+{
+	char path[PATH_MAX];
+
+	snprintf(path, sizeof(path), GUEST_DIR_FMT, guest);
+	return make_dir(path, 0750);
+}
+
+static int make_guest_fifo(const char *guest, int cpu, mode_t mode)
+{
+	static const char *exts[] = {".in", ".out"};
+	char path[PATH_MAX];
+	int i, ret = 0;
+
+	for (i = 0; i < ARRAY_SIZE(exts); i++) {
+		snprintf(path, sizeof(path), GUEST_FIFO_FMT "%s",
+			 guest, cpu, exts[i]);
+		ret = make_fifo(path, mode);
+		if (ret < 0)
+			break;
+	}
+
+	return ret;
+}
+
+static int make_guest_fifos(const char *guest, int nr_cpus, mode_t mode)
+{
+	int i, ret = 0;
+	mode_t mask;
+
+	mask = umask(0);
+	for (i = 0; i < nr_cpus; i++) {
+		ret = make_guest_fifo(guest, i, mode);
+		if (ret < 0)
+			break;
+	}
+	umask(mask);
+
+	return ret;
+}
+
+static void do_setup_guest(const char *guest, int nr_cpus, mode_t mode, gid_t gid)
+{
+	gid_t save_egid;
+	int ret;
+
+	if (gid != -1) {
+		save_egid = getegid();
+		ret = setegid(gid);
+		if (ret < 0)
+			die("failed to set effective group ID");
+	}
+
+	ret = make_guest_dir(guest);
+	if (ret < 0)
+		die("failed to create guest directory for %s", guest);
+
+	ret = make_guest_fifos(guest, nr_cpus, mode);
+	if (ret < 0)
+		die("failed to create FIFOs for %s", guest);
+
+	if (gid != -1) {
+		ret = setegid(save_egid);
+		if (ret < 0)
+			die("failed to restore effective group ID");
+	}
+}
+
+void trace_setup_guest(int argc, char **argv)
+{
+	struct group *group;
+	mode_t mode = 0660;
+	int nr_cpus = -1;
+	gid_t gid = -1;
+	char *guest;
+
+	if (argc < 2)
+		usage(argv);
+
+	if (strcmp(argv[1], "setup-guest") != 0)
+		usage(argv);
+
+	for (;;) {
+		int c, option_index = 0;
+		static struct option long_options[] = {
+			{"help", no_argument, NULL, '?'},
+			{NULL, 0, NULL, 0}
+		};
+
+		c = getopt_long(argc-1, argv+1, "+hc:p:g:",
+				long_options, &option_index);
+		if (c == -1)
+			break;
+		switch (c) {
+		case 'h':
+			usage(argv);
+			break;
+		case 'c':
+			nr_cpus = atoi(optarg);
+			break;
+		case 'p':
+			mode = strtol(optarg, NULL, 8);
+			break;
+		case 'g':
+			group = getgrnam(optarg);
+			if (!group)
+				die("group %s does not exist", optarg);
+			gid = group->gr_gid;
+			break;
+		default:
+			usage(argv);
+		}
+	}
+
+	if (optind != argc-2)
+		usage(argv);
+
+	guest = argv[optind+1];
+
+	if (nr_cpus <= 0)
+		die("invalid number of cpus for guest %s", guest);
+
+	do_setup_guest(guest, nr_cpus, mode, gid);
+}
diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c
index 20273ba..edc4ac6 100644
--- a/tracecmd/trace-usage.c
+++ b/tracecmd/trace-usage.c
@@ -250,6 +250,14 @@ static struct usage_help usage_help[] = {
 		"          -p port number to listen on.\n"
 		"          -D run in daemon mode.\n"
 	},
+	{
+		"setup-guest",
+		"create FIFOs for tracing guest VMs",
+		" %s setup-guest -c cpus[-p perm][-g group] guest\n"
+		"          -c number of guest virtual CPUs\n"
+		"          -p FIFOs permissions (default: 0660)\n"
+		"          -g FIFOs group owner\n"
+	},
 #endif
 	{
 		"list",
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 10/13] trace-cmd: Try to autodetect number of guest CPUs in setup-guest if not specified
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (8 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 09/13] trace-cmd: Add `trace-cmd setup-guest` command Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 11/13] trace-cmd: Add setup-guest flag for attaching FIFOs to the guest VM config Slavomir Kaslev
                   ` (2 subsequent siblings)
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

When no number of guest CPUs is provided with the -c flag to `trace-cmd
setup-guest`, try to autodetect it using virsh for libvirt managed guests.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/trace-setup-guest.c | 21 +++++++++++++++++++++
 tracecmd/trace-usage.c       |  2 +-
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/tracecmd/trace-setup-guest.c b/tracecmd/trace-setup-guest.c
index 2f8bfeb..0610104 100644
--- a/tracecmd/trace-setup-guest.c
+++ b/tracecmd/trace-setup-guest.c
@@ -93,6 +93,24 @@ static int make_guest_fifos(const char *guest, int nr_cpus, mode_t mode)
 	return ret;
 }
 
+static int get_guest_cpu_count(const char *guest)
+{
+	const char *cmd_fmt = "virsh vcpucount --maximum '%s' 2>/dev/null";
+	int nr_cpus = -1;
+	char cmd[1024];
+	FILE *f;
+
+	snprintf(cmd, sizeof(cmd), cmd_fmt, guest);
+	f = popen(cmd, "r");
+	if (!f)
+		return -errno;
+
+	fscanf(f, "%d", &nr_cpus);
+	pclose(f);
+
+	return nr_cpus;
+}
+
 static void do_setup_guest(const char *guest, int nr_cpus, mode_t mode, gid_t gid)
 {
 	gid_t save_egid;
@@ -171,6 +189,9 @@ void trace_setup_guest(int argc, char **argv)
 
 	guest = argv[optind+1];
 
+	if (nr_cpus <= 0)
+		nr_cpus = get_guest_cpu_count(guest);
+
 	if (nr_cpus <= 0)
 		die("invalid number of cpus for guest %s", guest);
 
diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c
index edc4ac6..b41d7d5 100644
--- a/tracecmd/trace-usage.c
+++ b/tracecmd/trace-usage.c
@@ -253,7 +253,7 @@ static struct usage_help usage_help[] = {
 	{
 		"setup-guest",
 		"create FIFOs for tracing guest VMs",
-		" %s setup-guest -c cpus[-p perm][-g group] guest\n"
+		" %s setup-guest [-c cpus][-p perm][-g group] guest\n"
 		"          -c number of guest virtual CPUs\n"
 		"          -p FIFOs permissions (default: 0660)\n"
 		"          -g FIFOs group owner\n"
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 11/13] trace-cmd: Add setup-guest flag for attaching FIFOs to the guest VM config
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (9 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 10/13] trace-cmd: Try to autodetect number of guest CPUs in setup-guest if not specified Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 12/13] trace-cmd: Add VM kernel tracing over FIFO transport Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 13/13] trace-cmd: Add support for tracing VMware Workstation VMs by name Slavomir Kaslev
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

This patch adds a flag for attaching the newly created FIFOs for guest tracing
as virtio serial devices to libvirt managed guests.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/trace-setup-guest.c | 54 ++++++++++++++++++++++++++++++++++--
 tracecmd/trace-usage.c       |  3 +-
 2 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/tracecmd/trace-setup-guest.c b/tracecmd/trace-setup-guest.c
index 0610104..899848c 100644
--- a/tracecmd/trace-setup-guest.c
+++ b/tracecmd/trace-setup-guest.c
@@ -111,7 +111,45 @@ static int get_guest_cpu_count(const char *guest)
 	return nr_cpus;
 }
 
-static void do_setup_guest(const char *guest, int nr_cpus, mode_t mode, gid_t gid)
+static int attach_guest_fifos(const char *guest, int nr_cpus)
+{
+	const char *cmd_fmt =
+		"virsh attach-device --config '%s' '%s' >/dev/null 2>/dev/null";
+	const char *xml_fmt =
+		"<channel type='pipe'>\n"
+		"  <source path='%s'/>\n"
+		"  <target type='virtio' name='%s%d'/>\n"
+		"</channel>";
+	char tmp_path[PATH_MAX], path[PATH_MAX];
+	char cmd[PATH_MAX], xml[PATH_MAX];
+	int i, fd, ret = 0;
+
+	strcpy(tmp_path, "/tmp/pipexmlXXXXXX");
+	fd = mkstemp(tmp_path);
+	if (fd < 0)
+		return fd;
+
+	for (i = 0; i < nr_cpus; i++) {
+		snprintf(path, sizeof(path), GUEST_FIFO_FMT, guest, i);
+		snprintf(xml, sizeof(xml), xml_fmt, path, GUEST_PIPE_NAME, i);
+		pwrite(fd, xml, strlen(xml), 0);
+
+		snprintf(cmd, sizeof(cmd), cmd_fmt, guest, tmp_path);
+		errno = 0;
+		if (system(cmd) != 0) {
+			ret = -errno;
+			break;
+		}
+	}
+
+	close(fd);
+	unlink(tmp_path);
+
+	return ret;
+}
+
+static void do_setup_guest(const char *guest, int nr_cpus,
+			   mode_t mode, gid_t gid, bool attach)
 {
 	gid_t save_egid;
 	int ret;
@@ -131,6 +169,12 @@ static void do_setup_guest(const char *guest, int nr_cpus, mode_t mode, gid_t gi
 	if (ret < 0)
 		die("failed to create FIFOs for %s", guest);
 
+	if (attach) {
+		ret = attach_guest_fifos(guest, nr_cpus);
+		if (ret < 0)
+			die("failed to attach FIFOs to %s", guest);
+	}
+
 	if (gid != -1) {
 		ret = setegid(save_egid);
 		if (ret < 0)
@@ -140,6 +184,7 @@ static void do_setup_guest(const char *guest, int nr_cpus, mode_t mode, gid_t gi
 
 void trace_setup_guest(int argc, char **argv)
 {
+	bool attach = false;
 	struct group *group;
 	mode_t mode = 0660;
 	int nr_cpus = -1;
@@ -159,7 +204,7 @@ void trace_setup_guest(int argc, char **argv)
 			{NULL, 0, NULL, 0}
 		};
 
-		c = getopt_long(argc-1, argv+1, "+hc:p:g:",
+		c = getopt_long(argc-1, argv+1, "+hc:p:g:a",
 				long_options, &option_index);
 		if (c == -1)
 			break;
@@ -179,6 +224,9 @@ void trace_setup_guest(int argc, char **argv)
 				die("group %s does not exist", optarg);
 			gid = group->gr_gid;
 			break;
+		case 'a':
+			attach = true;
+			break;
 		default:
 			usage(argv);
 		}
@@ -195,5 +243,5 @@ void trace_setup_guest(int argc, char **argv)
 	if (nr_cpus <= 0)
 		die("invalid number of cpus for guest %s", guest);
 
-	do_setup_guest(guest, nr_cpus, mode, gid);
+	do_setup_guest(guest, nr_cpus, mode, gid, attach);
 }
diff --git a/tracecmd/trace-usage.c b/tracecmd/trace-usage.c
index b41d7d5..b5788f7 100644
--- a/tracecmd/trace-usage.c
+++ b/tracecmd/trace-usage.c
@@ -253,10 +253,11 @@ static struct usage_help usage_help[] = {
 	{
 		"setup-guest",
 		"create FIFOs for tracing guest VMs",
-		" %s setup-guest [-c cpus][-p perm][-g group] guest\n"
+		" %s setup-guest [-c cpus][-p perm][-g group][-a] guest\n"
 		"          -c number of guest virtual CPUs\n"
 		"          -p FIFOs permissions (default: 0660)\n"
 		"          -g FIFOs group owner\n"
+		"          -a Attach FIFOs to guest VM config\n"
 	},
 #endif
 	{
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 12/13] trace-cmd: Add VM kernel tracing over FIFO transport
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (10 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 11/13] trace-cmd: Add setup-guest flag for attaching FIFOs to the guest VM config Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  2019-10-08  8:15 ` [PATCH v15 13/13] trace-cmd: Add support for tracing VMware Workstation VMs by name Slavomir Kaslev
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Add FIFO transport for trace-cmd VM kernel tracing if guest FIFOs are available
and attached to guest.

We can also do a direct splice from the transport FIFO to the output file
descriptor instead of two through an additional pipe buffer. This patch
implements specialized version for tracecmd_recorder data transfer for this
case.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 include/trace-cmd/trace-cmd.h  | 11 +++--
 lib/trace-cmd/trace-msg.c      | 26 ++++++----
 lib/trace-cmd/trace-recorder.c | 73 ++++++++++++++++++++++++----
 tracecmd/include/trace-local.h |  4 +-
 tracecmd/trace-agent.c         | 39 +++++++++++++--
 tracecmd/trace-record.c        | 87 ++++++++++++++++++++++++++++++----
 6 files changed, 201 insertions(+), 39 deletions(-)

diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
index b1e45f8..13afce7 100644
--- a/include/trace-cmd/trace-cmd.h
+++ b/include/trace-cmd/trace-cmd.h
@@ -286,8 +286,9 @@ struct tracecmd_output *tracecmd_get_output_handle_fd(int fd);
 
 enum {
 	TRACECMD_RECORD_NOSPLICE	= (1 << 0),	/* Use read instead of splice */
-	TRACECMD_RECORD_SNAPSHOT	= (1 << 1),	/* extract from snapshot */
+	TRACECMD_RECORD_SNAPSHOT	= (1 << 1),	/* Extract from snapshot */
 	TRACECMD_RECORD_BLOCK		= (1 << 2),	/* Block on splice write */
+	TRACECMD_RECORD_NOBRASS		= (1 << 3),	/* Splice directly without a brass pipe */
 };
 
 void tracecmd_free_recorder(struct tracecmd_recorder *recorder);
@@ -343,16 +344,16 @@ bool tracecmd_msg_done(struct tracecmd_msg_handle *msg_handle);
 void tracecmd_msg_set_done(struct tracecmd_msg_handle *msg_handle);
 
 int tracecmd_msg_send_trace_req(struct tracecmd_msg_handle *msg_handle,
-				int argc, char **argv);
+				int argc, char **argv, bool use_fifos);
 int tracecmd_msg_recv_trace_req(struct tracecmd_msg_handle *msg_handle,
-				int *argc, char ***argv);
+				int *argc, char ***argv, bool *use_fifos);
 
 int tracecmd_msg_send_trace_resp(struct tracecmd_msg_handle *msg_handle,
 				 int nr_cpus, int page_size,
-				 unsigned int *ports);
+				 unsigned int *ports, bool use_fifos);
 int tracecmd_msg_recv_trace_resp(struct tracecmd_msg_handle *msg_handle,
 				 int *nr_cpus, int *page_size,
-				 unsigned int **ports);
+				 unsigned int **ports, bool *use_fifos);
 
 /* --- Plugin handling --- */
 extern struct tep_plugin_option trace_ftrace_options[];
diff --git a/lib/trace-cmd/trace-msg.c b/lib/trace-cmd/trace-msg.c
index 347aee5..9c8a690 100644
--- a/lib/trace-cmd/trace-msg.c
+++ b/lib/trace-cmd/trace-msg.c
@@ -156,6 +156,10 @@ static int msg_write(int fd, struct tracecmd_msg *msg)
 	return __do_write_check(fd, msg->buf, data_size);
 }
 
+enum msg_trace_flags {
+	MSG_TRACE_USE_FIFOS = 1 << 0,
+};
+
 static int make_tinit(struct tracecmd_msg_handle *msg_handle,
 		      struct tracecmd_msg *msg)
 {
@@ -807,7 +811,7 @@ int tracecmd_msg_wait_close_resp(struct tracecmd_msg_handle *msg_handle)
 	return tracecmd_msg_wait_for_cmd(msg_handle, MSG_CLOSE_RESP);
 }
 
-static int make_trace_req(struct tracecmd_msg *msg, int argc, char **argv)
+static int make_trace_req(struct tracecmd_msg *msg, int argc, char **argv, bool use_fifos)
 {
 	size_t args_size = 0;
 	char *p;
@@ -817,6 +821,7 @@ static int make_trace_req(struct tracecmd_msg *msg, int argc, char **argv)
 		args_size += strlen(argv[i]) + 1;
 
 	msg->hdr.size = htonl(ntohl(msg->hdr.size) + args_size);
+	msg->trace_req.flags = use_fifos ? htonl(MSG_TRACE_USE_FIFOS) : htonl(0);
 	msg->trace_req.argc = htonl(argc);
 	msg->buf = calloc(args_size, 1);
 	if (!msg->buf)
@@ -830,13 +835,13 @@ static int make_trace_req(struct tracecmd_msg *msg, int argc, char **argv)
 }
 
 int tracecmd_msg_send_trace_req(struct tracecmd_msg_handle *msg_handle,
-				int argc, char **argv)
+				int argc, char **argv, bool use_fifos)
 {
 	struct tracecmd_msg msg;
 	int ret;
 
 	tracecmd_msg_init(MSG_TRACE_REQ, &msg);
-	ret = make_trace_req(&msg, argc, argv);
+	ret = make_trace_req(&msg, argc, argv, use_fifos);
 	if (ret < 0)
 		return ret;
 
@@ -849,7 +854,7 @@ int tracecmd_msg_send_trace_req(struct tracecmd_msg_handle *msg_handle,
   *     free(argv);
   */
 int tracecmd_msg_recv_trace_req(struct tracecmd_msg_handle *msg_handle,
-				int *argc, char ***argv)
+				int *argc, char ***argv, bool *use_fifos)
 {
 	struct tracecmd_msg msg;
 	char *p, *buf_end, **args;
@@ -895,6 +900,7 @@ int tracecmd_msg_recv_trace_req(struct tracecmd_msg_handle *msg_handle,
 
 	*argc = nr_args;
 	*argv = args;
+	*use_fifos = ntohl(msg.trace_req.flags) & MSG_TRACE_USE_FIFOS;
 
 	/*
 	 * On success we're passing msg.buf to the caller through argv[0] so we
@@ -914,8 +920,8 @@ out:
 	return ret;
 }
 
-static int make_trace_resp(struct tracecmd_msg *msg,
-			   int page_size, int nr_cpus, unsigned int *ports)
+static int make_trace_resp(struct tracecmd_msg *msg, int page_size, int nr_cpus,
+			   unsigned int *ports, bool use_fifos)
 {
 	int data_size;
 
@@ -926,6 +932,7 @@ static int make_trace_resp(struct tracecmd_msg *msg,
 	write_uints(msg->buf, data_size, ports, nr_cpus);
 
 	msg->hdr.size = htonl(ntohl(msg->hdr.size) + data_size);
+	msg->trace_resp.flags = use_fifos ? htonl(MSG_TRACE_USE_FIFOS) : htonl(0);
 	msg->trace_resp.cpus = htonl(nr_cpus);
 	msg->trace_resp.page_size = htonl(page_size);
 
@@ -934,13 +941,13 @@ static int make_trace_resp(struct tracecmd_msg *msg,
 
 int tracecmd_msg_send_trace_resp(struct tracecmd_msg_handle *msg_handle,
 				 int nr_cpus, int page_size,
-				 unsigned int *ports)
+				 unsigned int *ports, bool use_fifos)
 {
 	struct tracecmd_msg msg;
 	int ret;
 
 	tracecmd_msg_init(MSG_TRACE_RESP, &msg);
-	ret = make_trace_resp(&msg, page_size, nr_cpus, ports);
+	ret = make_trace_resp(&msg, page_size, nr_cpus, ports, use_fifos);
 	if (ret < 0)
 		return ret;
 
@@ -949,7 +956,7 @@ int tracecmd_msg_send_trace_resp(struct tracecmd_msg_handle *msg_handle,
 
 int tracecmd_msg_recv_trace_resp(struct tracecmd_msg_handle *msg_handle,
 				 int *nr_cpus, int *page_size,
-				 unsigned int **ports)
+				 unsigned int **ports, bool *use_fifos)
 {
 	struct tracecmd_msg msg;
 	char *p, *buf_end;
@@ -971,6 +978,7 @@ int tracecmd_msg_recv_trace_resp(struct tracecmd_msg_handle *msg_handle,
 		goto out;
 	}
 
+	*use_fifos = ntohl(msg.trace_resp.flags) & MSG_TRACE_USE_FIFOS;
 	*nr_cpus = ntohl(msg.trace_resp.cpus);
 	*page_size = ntohl(msg.trace_resp.page_size);
 	*ports = calloc(*nr_cpus, sizeof(**ports));
diff --git a/lib/trace-cmd/trace-recorder.c b/lib/trace-cmd/trace-recorder.c
index d248a37..36c9a96 100644
--- a/lib/trace-cmd/trace-recorder.c
+++ b/lib/trace-cmd/trace-recorder.c
@@ -8,6 +8,7 @@
 #include <stdlib.h>
 #include <fcntl.h>
 #include <time.h>
+#include <poll.h>
 #include <unistd.h>
 #include <errno.h>
 
@@ -26,6 +27,8 @@
 # define SPLICE_F_GIFT		8
 #endif
 
+#define POLL_TIMEOUT_MS		1000
+
 struct tracecmd_recorder {
 	int		fd;
 	int		fd1;
@@ -40,6 +43,7 @@ struct tracecmd_recorder {
 	int		pages;
 	int		count;
 	unsigned	fd_flags;
+	unsigned	trace_fd_flags;
 	unsigned	flags;
 };
 
@@ -127,6 +131,8 @@ tracecmd_create_buffer_recorder_fd2(int fd, int fd2, int cpu, unsigned flags,
 	if (!(recorder->flags & TRACECMD_RECORD_BLOCK))
 		recorder->fd_flags |= SPLICE_F_NONBLOCK;
 
+	recorder->trace_fd_flags = SPLICE_F_MOVE;
+
 	/* Init to know what to free and release */
 	recorder->trace_fd = -1;
 	recorder->brass[0] = -1;
@@ -171,7 +177,8 @@ tracecmd_create_buffer_recorder_fd2(int fd, int fd2, int cpu, unsigned flags,
 			goto out_free;
 	}
 
-	if ((recorder->flags & TRACECMD_RECORD_NOSPLICE) == 0) {
+	if (!(recorder->flags & (TRACECMD_RECORD_NOSPLICE |
+				 TRACECMD_RECORD_NOBRASS))) {
 		ret = pipe(recorder->brass);
 		if (ret < 0)
 			goto out_free;
@@ -380,7 +387,7 @@ static long splice_data(struct tracecmd_recorder *recorder)
 	long ret;
 
 	read = splice(recorder->trace_fd, NULL, recorder->brass[1], NULL,
-		      recorder->pipe_size, SPLICE_F_MOVE);
+		      recorder->pipe_size, recorder->trace_fd_flags);
 	if (read < 0) {
 		if (errno == EAGAIN || errno == EINTR || errno == ENOTCONN)
 			return 0;
@@ -409,6 +416,47 @@ static long splice_data(struct tracecmd_recorder *recorder)
 	return total_read;
 }
 
+/*
+ * Returns -1 on error.
+ *          or bytes of data read.
+ */
+static long direct_splice_data(struct tracecmd_recorder *recorder)
+{
+	struct pollfd pfd = {
+		.fd = recorder->trace_fd,
+		.events = POLLIN,
+	};
+	long read;
+	int ret;
+
+	/*
+	 * splice(2) in Linux used to not check O_NONBLOCK flag of pipe file
+	 * descriptors before [1]. To avoid getting blocked in the splice(2)
+	 * call below after the user had requested to stop tracing, we poll(2)
+	 * here. This poll() is not necessary on newer kernels.
+	 *
+	 * [1] https://github.com/torvalds/linux/commit/ee5e001196d1345b8fee25925ff5f1d67936081e
+	 */
+	ret = poll(&pfd, 1, POLL_TIMEOUT_MS);
+	if (ret < 0)
+		return -1;
+
+	if (!(pfd.revents | POLLIN))
+		return 0;
+
+	read = splice(recorder->trace_fd, NULL, recorder->fd, NULL,
+		      recorder->pipe_size, recorder->fd_flags);
+	if (read < 0) {
+		if (errno == EAGAIN || errno == EINTR || errno == ENOTCONN)
+			return 0;
+
+		warning("recorder error in splice input");
+		return -1;
+	}
+
+	return read;
+}
+
 /*
  * Returns -1 on error.
  *          or bytes of data read.
@@ -443,6 +491,17 @@ static long read_data(struct tracecmd_recorder *recorder)
 	return r;
 }
 
+static long move_data(struct tracecmd_recorder *recorder)
+{
+	if (recorder->flags & TRACECMD_RECORD_NOSPLICE)
+		return read_data(recorder);
+
+	if (recorder->flags & TRACECMD_RECORD_NOBRASS)
+		return direct_splice_data(recorder);
+
+	return splice_data(recorder);
+}
+
 static void set_nonblock(struct tracecmd_recorder *recorder)
 {
 	long flags;
@@ -465,10 +524,7 @@ long tracecmd_flush_recording(struct tracecmd_recorder *recorder)
 	set_nonblock(recorder);
 
 	do {
-		if (recorder->flags & TRACECMD_RECORD_NOSPLICE)
-			ret = read_data(recorder);
-		else
-			ret = splice_data(recorder);
+		ret = move_data(recorder);
 		if (ret < 0)
 			return ret;
 		total += ret;
@@ -513,10 +569,7 @@ int tracecmd_start_recording(struct tracecmd_recorder *recorder, unsigned long s
 
 		read = 0;
 		do {
-			if (recorder->flags & TRACECMD_RECORD_NOSPLICE)
-				ret = read_data(recorder);
-			else
-				ret = splice_data(recorder);
+			ret = move_data(recorder);
 			if (ret < 0)
 				return ret;
 			read += ret;
diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
index 3bada82..fedc0b7 100644
--- a/tracecmd/include/trace-local.h
+++ b/tracecmd/include/trace-local.h
@@ -17,6 +17,7 @@
 #define GUEST_PIPE_NAME		"trace-pipe-cpu"
 #define GUEST_DIR_FMT		"/var/lib/trace-cmd/virt/%s"
 #define GUEST_FIFO_FMT		GUEST_DIR_FMT "/" GUEST_PIPE_NAME "%d"
+#define VIRTIO_FIFO_FMT		"/dev/virtio-ports/" GUEST_PIPE_NAME "%d"
 
 /* fix stupid glib guint64 typecasts and printf formats */
 typedef unsigned long long u64;
@@ -97,7 +98,7 @@ void trace_usage(int argc, char **argv);
 
 int trace_record_agent(struct tracecmd_msg_handle *msg_handle,
 		       int cpus, int *fds,
-		       int argc, char **argv);
+		       int argc, char **argv, bool use_fifos);
 
 struct hook_list;
 
@@ -221,6 +222,7 @@ struct buffer_instance {
 	unsigned int		cid;
 	unsigned int		port;
 	int			*fds;
+	bool			use_fifos;
 };
 
 extern struct buffer_instance top_instance;
diff --git a/tracecmd/trace-agent.c b/tracecmd/trace-agent.c
index 89e3568..93e3edf 100644
--- a/tracecmd/trace-agent.c
+++ b/tracecmd/trace-agent.c
@@ -100,12 +100,38 @@ static void make_vsocks(int nr, int *fds, unsigned int *ports)
 	}
 }
 
+static int open_agent_fifos(int nr_cpus, int *fds)
+{
+	char path[PATH_MAX];
+	int i, fd, ret;
+
+	for (i = 0; i < nr_cpus; i++) {
+		snprintf(path, sizeof(path), VIRTIO_FIFO_FMT, i);
+		fd = open(path, O_WRONLY);
+		if (fd < 0) {
+			ret = -errno;
+			goto cleanup;
+		}
+
+		fds[i] = fd;
+	}
+
+	return 0;
+
+cleanup:
+	while (--i >= 0)
+		close(fds[i]);
+
+	return ret;
+}
+
 static void agent_handle(int sd, int nr_cpus, int page_size)
 {
 	struct tracecmd_msg_handle *msg_handle;
 	unsigned int *ports;
 	char **argv = NULL;
 	int argc = 0;
+	bool use_fifos;
 	int *fds;
 	int ret;
 
@@ -118,17 +144,22 @@ static void agent_handle(int sd, int nr_cpus, int page_size)
 	if (!msg_handle)
 		die("Failed to allocate message handle");
 
-	ret = tracecmd_msg_recv_trace_req(msg_handle, &argc, &argv);
+	ret = tracecmd_msg_recv_trace_req(msg_handle, &argc, &argv, &use_fifos);
 	if (ret < 0)
 		die("Failed to receive trace request");
 
-	make_vsocks(nr_cpus, fds, ports);
+	if (use_fifos && open_agent_fifos(nr_cpus, fds))
+		use_fifos = false;
+
+	if (!use_fifos)
+		make_vsocks(nr_cpus, fds, ports);
 
-	ret = tracecmd_msg_send_trace_resp(msg_handle, nr_cpus, page_size, ports);
+	ret = tracecmd_msg_send_trace_resp(msg_handle, nr_cpus, page_size,
+					   ports, use_fifos);
 	if (ret < 0)
 		die("Failed to send trace response");
 
-	trace_record_agent(msg_handle, nr_cpus, fds, argc, argv);
+	trace_record_agent(msg_handle, nr_cpus, fds, argc, argv, use_fifos);
 
 	free(argv[0]);
 	free(argv);
diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index 94180c2..cd16243 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -79,6 +79,8 @@ static int buffers;
 /* Clear all function filters */
 static int clear_function_filters;
 
+static bool no_fifos;
+
 static char *host;
 
 static bool quiet;
@@ -3170,12 +3172,17 @@ create_recorder_instance(struct buffer_instance *instance, const char *file, int
 		int fd;
 		unsigned int flags;
 
-		fd = open_vsock(instance->cid, instance->client_ports[cpu]);
+		if (instance->use_fifos)
+			fd = instance->fds[cpu];
+		else
+			fd = open_vsock(instance->cid, instance->client_ports[cpu]);
 		if (fd < 0)
 			die("Failed to connect to agent");
 
 		flags = recorder_flags;
-		if (!can_splice_read_vsock())
+		if (instance->use_fifos)
+			flags |= TRACECMD_RECORD_NOBRASS;
+		else if (!can_splice_read_vsock())
 			flags |= TRACECMD_RECORD_NOSPLICE;
 		return tracecmd_create_recorder_virt(file, cpu, flags, fd);
 	}
@@ -3230,10 +3237,14 @@ static int create_recorder(struct buffer_instance *instance, int cpu,
 		char *path = NULL;
 		int fd;
 
-		if (is_agent(instance))
-			fd = do_accept(instance->fds[cpu]);
-		else
+		if (is_agent(instance)) {
+			if (instance->use_fifos)
+				fd = instance->fds[cpu];
+			else
+				fd = do_accept(instance->fds[cpu]);
+		} else {
 			fd = connect_port(host, instance->client_ports[cpu]);
+		}
 		if (fd < 0)
 			die("Failed connecting to client");
 		if (instance->name && !is_agent(instance))
@@ -3517,11 +3528,42 @@ static void finish_network(struct tracecmd_msg_handle *msg_handle)
 	free(host);
 }
 
+static int open_guest_fifos(const char *guest, int **fds)
+{
+	char path[PATH_MAX];
+	int i, fd, flags;
+
+	for (i = 0; ; i++) {
+		snprintf(path, sizeof(path), GUEST_FIFO_FMT ".out", guest, i);
+
+		/* O_NONBLOCK so we don't wait for writers */
+		fd = open(path, O_RDONLY | O_NONBLOCK);
+		if (fd < 0)
+			break;
+
+		/* Success, now clear O_NONBLOCK */
+		flags = fcntl(fd, F_GETFL);
+		fcntl(fd, F_SETFL, flags & ~O_NONBLOCK);
+
+		*fds = realloc(*fds, i + 1);
+		(*fds)[i] = fd;
+	}
+
+	return i;
+}
+
 static void connect_to_agent(struct buffer_instance *instance)
 {
 	struct tracecmd_msg_handle *msg_handle;
-	int sd, ret, nr_cpus, page_size;
+	int sd, ret, nr_fifos, nr_cpus, page_size;
 	unsigned int *ports;
+	int i, *fds = NULL;
+	bool use_fifos = false;
+
+	if (!no_fifos) {
+		nr_fifos = open_guest_fifos(instance->name, &fds);
+		use_fifos = nr_fifos > 0;
+	}
 
 	sd = open_vsock(instance->cid, instance->port);
 	if (sd < 0)
@@ -3532,15 +3574,33 @@ static void connect_to_agent(struct buffer_instance *instance)
 	if (!msg_handle)
 		die("Failed to allocate message handle");
 
-	ret = tracecmd_msg_send_trace_req(msg_handle, instance->argc, instance->argv);
+	ret = tracecmd_msg_send_trace_req(msg_handle, instance->argc,
+					  instance->argv, use_fifos);
 	if (ret < 0)
 		die("Failed to send trace request");
 
-	ret = tracecmd_msg_recv_trace_resp(msg_handle, &nr_cpus, &page_size, &ports);
+	ret = tracecmd_msg_recv_trace_resp(msg_handle, &nr_cpus, &page_size,
+					   &ports, &use_fifos);
 	if (ret < 0)
 		die("Failed to receive trace response");
 
-	instance->client_ports = ports;
+	if (use_fifos) {
+		if (nr_cpus != nr_fifos) {
+			warning("number of FIFOs (%d) for guest %s differs "
+				"from number of virtual CPUs (%d)",
+				nr_fifos, instance->name, nr_cpus);
+			nr_cpus = nr_cpus < nr_fifos ? nr_cpus : nr_fifos;
+		}
+		free(ports);
+		instance->fds = fds;
+	} else {
+		for (i = 0; i < nr_fifos; i++)
+			close(fds[i]);
+		free(fds);
+		instance->client_ports = ports;
+	}
+
+	instance->use_fifos = use_fifos;
 	instance->cpu_count = nr_cpus;
 
 	/* the msg_handle now points to the guest fd */
@@ -5107,6 +5167,7 @@ enum {
 	OPT_funcstack		= 254,
 	OPT_date		= 255,
 	OPT_module		= 256,
+	OPT_nofifos		= 257,
 };
 
 void trace_stop(int argc, char **argv)
@@ -5377,6 +5438,7 @@ static void parse_record_options(int argc,
 			{"date", no_argument, NULL, OPT_date},
 			{"func-stack", no_argument, NULL, OPT_funcstack},
 			{"nosplice", no_argument, NULL, OPT_nosplice},
+			{"nofifos", no_argument, NULL, OPT_nofifos},
 			{"profile", no_argument, NULL, OPT_profile},
 			{"stderr", no_argument, NULL, OPT_stderr},
 			{"by-comm", no_argument, NULL, OPT_bycomm},
@@ -5679,6 +5741,9 @@ static void parse_record_options(int argc,
 		case OPT_nosplice:
 			recorder_flags |= TRACECMD_RECORD_NOSPLICE;
 			break;
+		case OPT_nofifos:
+			no_fifos = true;
+			break;
 		case OPT_profile:
 			handle_init = trace_init_profile;
 			ctx->instance->flags |= BUFFER_FL_PROFILE;
@@ -6100,7 +6165,8 @@ void trace_record(int argc, char **argv)
 
 int trace_record_agent(struct tracecmd_msg_handle *msg_handle,
 		       int cpus, int *fds,
-		       int argc, char **argv)
+		       int argc, char **argv,
+		       bool use_fifos)
 {
 	struct common_record_context ctx;
 	char **argv_plus;
@@ -6126,6 +6192,7 @@ int trace_record_agent(struct tracecmd_msg_handle *msg_handle,
 		return -EINVAL;
 
 	ctx.instance->fds = fds;
+	ctx.instance->use_fifos = use_fifos;
 	ctx.instance->flags |= BUFFER_FL_AGENT;
 	ctx.instance->msg_handle = msg_handle;
 	msg_handle->version = V3_PROTOCOL;
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

* [PATCH v15 13/13] trace-cmd: Add support for tracing VMware Workstation VMs by name
  2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
                   ` (11 preceding siblings ...)
  2019-10-08  8:15 ` [PATCH v15 12/13] trace-cmd: Add VM kernel tracing over FIFO transport Slavomir Kaslev
@ 2019-10-08  8:15 ` Slavomir Kaslev
  12 siblings, 0 replies; 14+ messages in thread
From: Slavomir Kaslev @ 2019-10-08  8:15 UTC (permalink / raw)
  To: rostedt; +Cc: linux-trace-devel

Read display name and CID of running VMware Workstation guests on startup so
that users can refer to them by name when recording.

Signed-off-by: Slavomir Kaslev <kaslevs@vmware.com>
---
 tracecmd/trace-record.c | 72 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 72 insertions(+)

diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index cd16243..8533b13 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -3092,6 +3092,77 @@ next:
 	closedir(dir);
 }
 
+static int load_vmw_guest(const char *vm, char **name, int *cid)
+{
+	size_t line_len = 0;
+	char *line = NULL;
+	char *p;
+	FILE *f;
+
+	f = fopen(vm, "r");
+	if (!f)
+		return -errno;
+
+	*cid = -1;
+	*name = NULL;
+	while (getline(&line, &line_len, f) != -1) {
+		if (strncmp(line, "displayName = ", 14) == 0) {
+			p = strrchr(line, '"');
+			if (p)
+				*p = '\0';
+			*name = strdup(line + 15);
+			if (!*name)
+				die("allocating guest name");
+		} else if (strncmp(line, "vmci0.id = ", 11) == 0) {
+			p = strrchr(line, '"');
+			if (p)
+				*p = '\0';
+			*cid = atoi(line + 12);
+		}
+	}
+
+	free(line);
+	fclose(f);
+	return 0;
+}
+
+static void read_vmw_guests(void)
+{
+	static bool initialized;
+	size_t line_len = 0;
+	char *line = NULL;
+	ssize_t ret;
+	FILE *f;
+
+	if (initialized)
+		return;
+
+	initialized = true;
+
+	f = popen("vmrun list", "r");
+	if (!f)
+		return;
+
+	/* Ignore the first line */
+	ret = getline(&line, &line_len, f);
+	while ((ret = getline(&line, &line_len, f)) != -1) {
+		struct guest guest = {};
+
+		if (ret > 0 && line[ret-1] == '\n')
+			line[ret-1] = '\0';
+		if (load_vmw_guest(line, &guest.name, &guest.cid))
+			continue;
+
+		guests = realloc(guests, (guests_len + 1) * sizeof(*guests));
+		if (!guests)
+			die("Can not allocate guest buffer");
+		guests[guests_len++] = guest;
+	}
+
+	free(line);
+	pclose(f);
+}
+
 static char *parse_guest_name(char *guest, int *cid, int *port)
 {
 	size_t i;
@@ -3113,6 +3184,7 @@ static char *parse_guest_name(char *guest, int *cid, int *port)
 		*cid = atoi(guest);
 
 	read_qemu_guests();
+	read_vmw_guests();
 	for (i = 0; i < guests_len; i++) {
 		if ((*cid > 0 && *cid == guests[i].cid) ||
 		    strcmp(guest, guests[i].name) == 0) {
-- 
2.20.1


^ permalink raw reply	[flat|nested] 14+ messages in thread

end of thread, back to index

Thread overview: 14+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-10-08  8:15 [PATCH v15 00/13] Add VM kernel tracing over vsockets and FIFOs Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 01/13] trace-cmd: Make ports unsigned int Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 02/13] trace-cmd: Detect if vsockets are available Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 03/13] trace-cmd: Add tracecmd_create_recorder_virt() function Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 04/13] trace-cmd: Add VM tracing protocol messages Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 05/13] trace-cmd: Add buffer instance flags for tracing in guest and agent context Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 06/13] trace-cmd: Add VM kernel tracing over vsockets transport Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 07/13] trace-cmd: Use splice(2) for vsockets if available Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 08/13] trace-cmd: Switch stop recording signal to SIGUSR1 Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 09/13] trace-cmd: Add `trace-cmd setup-guest` command Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 10/13] trace-cmd: Try to autodetect number of guest CPUs in setup-guest if not specified Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 11/13] trace-cmd: Add setup-guest flag for attaching FIFOs to the guest VM config Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 12/13] trace-cmd: Add VM kernel tracing over FIFO transport Slavomir Kaslev
2019-10-08  8:15 ` [PATCH v15 13/13] trace-cmd: Add support for tracing VMware Workstation VMs by name Slavomir Kaslev

Linux-Trace-Devel Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-trace-devel/0 linux-trace-devel/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-trace-devel linux-trace-devel/ https://lore.kernel.org/linux-trace-devel \
		linux-trace-devel@vger.kernel.org linux-trace-devel@archiver.kernel.org
	public-inbox-index linux-trace-devel

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-trace-devel


AGPL code for this site: git clone https://public-inbox.org/ public-inbox