Date:   Wed, 26 Aug 2020 17:06:20 -0700
From:   Yonghong Song <yhs@...com>
To:     <bpf@...r.kernel.org>, <netdev@...r.kernel.org>
CC:     Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>, <kernel-team@...com>
Subject: [PATCH bpf-next 2/5] bpf: add main_thread_only customization for task/task_file iterators

Currently, task and task_file by default iterates through
all tasks. For task_file, by default, all files from all tasks
will be traversed.

But for a user process, the file_table is shared by all threads
of that process. So traversing the main thread per process should
be enough to traverse all files and this can save a lot of cpu
time if some process has large number of threads and each thread
has lots of open files.

This patch implemented a customization for task/task_file iterator,
permitting to traverse only the kernel task where its pid equal
to tgid in the kernel. This includes some kernel threads, and
main threads of user processes. This will solve the above potential
performance issue for task_file. This customization may be useful
for task iterator too if only traversing main threads is enough.

Signed-off-by: Yonghong Song <yhs@...com>
---
 include/linux/bpf.h            |  3 ++-
 include/uapi/linux/bpf.h       |  5 ++++
 kernel/bpf/task_iter.c         | 46 +++++++++++++++++++++++-----------
 tools/include/uapi/linux/bpf.h |  5 ++++
 4 files changed, 43 insertions(+), 16 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index a6131d95e31e..058eb9b0ba78 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -1220,7 +1220,8 @@ int bpf_obj_get_user(const char __user *pathname, int flags);
 	int __init bpf_iter_ ## target(args) { return 0; }
 
 struct bpf_iter_aux_info {
-	struct bpf_map *map;
+	struct bpf_map *map;	/* for iterator traversing map elements */
+	bool main_thread_only;	/* for task/task_file iterator */
 };
 
 typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog,
diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index ef7af384f5ee..af5c600bf673 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -85,6 +85,11 @@ union bpf_iter_link_info {
 	struct {
 		__u32	map_fd;
 	} map;
+
+	struct {
+		__u32	main_thread_only:1;
+		__u32	:31;
+	} task;
 };
 
 /* BPF syscall commands, see bpf(2) man-page for details. */
diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
index 232df29793e9..362bf2dda63a 100644
--- a/kernel/bpf/task_iter.c
+++ b/kernel/bpf/task_iter.c
@@ -11,19 +11,22 @@
 
 struct bpf_iter_seq_task_common {
 	struct pid_namespace *ns;
+	bool main_thread_only;
 };
 
 struct bpf_iter_seq_task_info {
 	/* The first field must be struct bpf_iter_seq_task_common.
-	 * this is assumed by {init, fini}_seq_pidns() callback functions.
+	 * this is assumed by {init, fini}_seq_task_common() callback functions.
 	 */
 	struct bpf_iter_seq_task_common common;
 	u32 tid;
 };
 
-static struct task_struct *task_seq_get_next(struct pid_namespace *ns,
-					     u32 *tid)
+static struct task_struct *task_seq_get_next(
+	struct bpf_iter_seq_task_common *task_common, u32 *tid)
 {
+	bool main_thread_only = task_common->main_thread_only;
+	struct pid_namespace *ns = task_common->ns;
 	struct task_struct *task = NULL;
 	struct pid *pid;
 
@@ -31,7 +34,10 @@ static struct task_struct *task_seq_get_next(struct pid_namespace *ns,
 retry:
 	pid = idr_get_next(&ns->idr, tid);
 	if (pid) {
-		task = get_pid_task(pid, PIDTYPE_PID);
+		if (main_thread_only)
+			task = get_pid_task(pid, PIDTYPE_TGID);
+		else
+			task = get_pid_task(pid, PIDTYPE_PID);
 		if (!task) {
 			++*tid;
 			goto retry;
@@ -47,7 +53,7 @@ static void *task_seq_start(struct seq_file *seq, loff_t *pos)
 	struct bpf_iter_seq_task_info *info = seq->private;
 	struct task_struct *task;
 
-	task = task_seq_get_next(info->common.ns, &info->tid);
+	task = task_seq_get_next(&info->common, &info->tid);
 	if (!task)
 		return NULL;
 
@@ -64,7 +70,7 @@ static void *task_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 	++*pos;
 	++info->tid;
 	put_task_struct((struct task_struct *)v);
-	task = task_seq_get_next(info->common.ns, &info->tid);
+	task = task_seq_get_next(&info->common, &info->tid);
 	if (!task)
 		return NULL;
 
@@ -118,7 +124,7 @@ static const struct seq_operations task_seq_ops = {
 
 struct bpf_iter_seq_task_file_info {
 	/* The first field must be struct bpf_iter_seq_task_common.
-	 * this is assumed by {init, fini}_seq_pidns() callback functions.
+	 * this is assumed by {init, fini}_seq_task_common() callback functions.
 	 */
 	struct bpf_iter_seq_task_common common;
 	struct task_struct *task;
@@ -131,7 +137,6 @@ static struct file *
 task_file_seq_get_next(struct bpf_iter_seq_task_file_info *info,
 		       struct task_struct **task, struct files_struct **fstruct)
 {
-	struct pid_namespace *ns = info->common.ns;
 	u32 curr_tid = info->tid, max_fds;
 	struct files_struct *curr_files;
 	struct task_struct *curr_task;
@@ -147,7 +152,7 @@ task_file_seq_get_next(struct bpf_iter_seq_task_file_info *info,
 		curr_files = *fstruct;
 		curr_fd = info->fd;
 	} else {
-		curr_task = task_seq_get_next(ns, &curr_tid);
+		curr_task = task_seq_get_next(&info->common, &curr_tid);
 		if (!curr_task)
 			return NULL;
 
@@ -293,15 +298,16 @@ static void task_file_seq_stop(struct seq_file *seq, void *v)
 	}
 }
 
-static int init_seq_pidns(void *priv_data, struct bpf_iter_aux_info *aux)
+static int init_seq_task_common(void *priv_data, struct bpf_iter_aux_info *aux)
 {
 	struct bpf_iter_seq_task_common *common = priv_data;
 
 	common->ns = get_pid_ns(task_active_pid_ns(current));
+	common->main_thread_only = aux->main_thread_only;
 	return 0;
 }
 
-static void fini_seq_pidns(void *priv_data)
+static void fini_seq_task_common(void *priv_data)
 {
 	struct bpf_iter_seq_task_common *common = priv_data;
 
@@ -315,19 +321,28 @@ static const struct seq_operations task_file_seq_ops = {
 	.show	= task_file_seq_show,
 };
 
+static int bpf_iter_attach_task(struct bpf_prog *prog,
+				union bpf_iter_link_info *linfo,
+				struct bpf_iter_aux_info *aux)
+{
+	aux->main_thread_only = linfo->task.main_thread_only;
+	return 0;
+}
+
 BTF_ID_LIST(btf_task_file_ids)
 BTF_ID(struct, task_struct)
 BTF_ID(struct, file)
 
 static const struct bpf_iter_seq_info task_seq_info = {
 	.seq_ops		= &task_seq_ops,
-	.init_seq_private	= init_seq_pidns,
-	.fini_seq_private	= fini_seq_pidns,
+	.init_seq_private	= init_seq_task_common,
+	.fini_seq_private	= fini_seq_task_common,
 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_info),
 };
 
 static struct bpf_iter_reg task_reg_info = {
 	.target			= "task",
+	.attach_target		= bpf_iter_attach_task,
 	.ctx_arg_info_size	= 1,
 	.ctx_arg_info		= {
 		{ offsetof(struct bpf_iter__task, task),
@@ -338,13 +353,14 @@ static struct bpf_iter_reg task_reg_info = {
 
 static const struct bpf_iter_seq_info task_file_seq_info = {
 	.seq_ops		= &task_file_seq_ops,
-	.init_seq_private	= init_seq_pidns,
-	.fini_seq_private	= fini_seq_pidns,
+	.init_seq_private	= init_seq_task_common,
+	.fini_seq_private	= fini_seq_task_common,
 	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_file_info),
 };
 
 static struct bpf_iter_reg task_file_reg_info = {
 	.target			= "task_file",
+	.attach_target		= bpf_iter_attach_task,
 	.ctx_arg_info_size	= 2,
 	.ctx_arg_info		= {
 		{ offsetof(struct bpf_iter__task_file, task),
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index ef7af384f5ee..af5c600bf673 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -85,6 +85,11 @@ union bpf_iter_link_info {
 	struct {
 		__u32	map_fd;
 	} map;
+
+	struct {
+		__u32	main_thread_only:1;
+		__u32	:31;
+	} task;
 };
 
 /* BPF syscall commands, see bpf(2) man-page for details. */
-- 
2.24.1

Powered by blists - more mailing lists