From e04ef376c3ee89b52a5befa82f1fe768cfd08cf1 Mon Sep 17 00:00:00 2001
From: Luis Gerhorst <privat@luisgerhorst.de>
Date: Sat, 10 Jul 2021 11:08:13 +0200
Subject: [PATCH] task_find: Optimize, avoid open on regular files
Using perf record it was determined that most time in both find
task_find.bpf.c is spent in openat, therefore this has to be avoided.
---
src/task_find.bpf.c | 30 +++++++++++++++++-------------
src/task_lib.bpf.h | 11 +++++++++++
2 files changed, 28 insertions(+), 13 deletions(-)
diff --git a/src/task_find.bpf.c b/src/task_find.bpf.c
index 13352a1..cbac40d 100644
--- a/src/task_find.bpf.c
+++ b/src/task_find.bpf.c
@@ -28,7 +28,7 @@ void *u_newline;
void *u_iterdents64_buf;
size_t root_namelen;
-static long iter_dir(size_t namelen, bool print) {
+static long iter_dir(uint32_t d_type, size_t namelen, bool print) {
long err = 0;
if (namelen + 1 > PATHLEN) {
@@ -37,12 +37,6 @@ static long iter_dir(size_t namelen, bool print) {
goto end;
}
- int *prog_fd = bpf_map_lookup_elem(&state_map, &PDPF_KEY);
- if (!prog_fd) {
- bpf_printk("Error: !bpf_map_lookup_elem(&state_map, &PDPF_KEY)");
- goto end;
- }
-
int *curr_depth_p = bpf_map_lookup_elem(&state_map, &CURR_DEPTH_KEY);
if (!curr_depth_p) {
bpf_printk("Error: bpf_map_lookup_elem(&state_map, &CURR_DEPTH_KEY)");
@@ -56,6 +50,13 @@ static long iter_dir(size_t namelen, bool print) {
}
void *name_user = u_iterdents64_buf + depth * PATHLEN;
+ if (d_type != DT_DIR) {
+ if (print) {
+ write(STDOUT_FILENO, name_user, namelen);
+ write(STDOUT_FILENO, u_newline, 1);
+ }
+ goto end;
+ }
int *parent_dfd = bpf_map_lookup_elem(&dfd_map, &depth);
if (!parent_dfd) {
@@ -65,10 +66,7 @@ static long iter_dir(size_t namelen, bool print) {
int dfd = openat(*parent_dfd, name_user, O_RDONLY | O_DIRECTORY, 0777);
if (dfd < 0) {
- if (print) {
- write(STDOUT_FILENO, name_user, namelen);
- write(STDOUT_FILENO, u_newline, 1);
- }
+ bpf_printk("Warning: openat failed, is someone modifying the tree?");
/* Not a directory, no need to iterate. */
goto end;
}
@@ -90,6 +88,12 @@ static long iter_dir(size_t namelen, bool print) {
goto reset_depth;
}
+ int *prog_fd = bpf_map_lookup_elem(&state_map, &PDPF_KEY);
+ if (!prog_fd) {
+ bpf_printk("Error: !bpf_map_lookup_elem(&state_map, &PDPF_KEY)");
+ goto end;
+ }
+
err = bpf_task_iterdents64(dfd, *prog_fd, child_name_user, PATHLEN);
if (err) {
bpf_printk("Error: iterdents64(%d, %d) = %d", dfd, *prog_fd, err);
@@ -116,7 +120,7 @@ int entry(void *ctx)
{
long err = 0;
- err = iter_dir(root_namelen, false);
+ err = iter_dir(DT_DIR, root_namelen, false);
if (err) {
goto end;
}
@@ -143,7 +147,7 @@ int process_dirent(struct bpf_dirent64 *d) {
}
}
- err = iter_dir(d->namelen, match);
+ err = iter_dir(d->d_type, d->namelen, match);
if (err) {
goto end;
}
diff --git a/src/task_lib.bpf.h b/src/task_lib.bpf.h
index 84c7af2..7545e82 100644
--- a/src/task_lib.bpf.h
+++ b/src/task_lib.bpf.h
@@ -65,6 +65,17 @@ typedef off64_t off_t;
#define SEEK_DATA 3 /* Seek to next data. */
#define SEEK_HOLE 4 /* Seek to next hole. */
+/* these are defined by POSIX and also present in glibc's dirent.h */
+#define DT_UNKNOWN 0
+#define DT_FIFO 1
+#define DT_CHR 2
+#define DT_DIR 4
+#define DT_BLK 6
+#define DT_REG 8
+#define DT_LNK 10
+#define DT_SOCK 12
+#define DT_WHT 14
+
/* Copied from man getdents64(2). */
struct linux_dirent64 {
ino64_t d_ino; /* 64-bit inode number */
--
GitLab