|
 |
5e5f7c |
From 9bbb628620d4e586941344e1bdbbc166a885c0a9 Mon Sep 17 00:00:00 2001
|
|
|
5e5f7c |
From: Rob Crittenden <rcritten@redhat.com>
|
|
|
5e5f7c |
Date: Thu, 5 Sep 2019 12:45:52 -0400
|
|
|
5e5f7c |
Subject: [PATCH] Optimize closing open file descriptors
|
|
|
5e5f7c |
|
|
|
5e5f7c |
When forking, the code would close all unused file descriptors up
|
|
|
5e5f7c |
to maximum number of files. In the default case this is 1024. In
|
|
|
5e5f7c |
the container case this is 1048576. Huge delays in startup were
|
|
|
5e5f7c |
seen due to this.
|
|
|
5e5f7c |
|
|
|
5e5f7c |
Even in a default 1024 ulimit case this drastically reduces the
|
|
|
5e5f7c |
number of file descriptors to mark FD_CLOEXEC but in the container
|
|
|
5e5f7c |
default case this saves another order of magnitude of work.
|
|
|
5e5f7c |
|
|
|
5e5f7c |
This patch takes inspiration from systemd[1] and walks /proc/self/fd
|
|
|
5e5f7c |
if it is available to determine the list of open descriptors. It
|
|
|
5e5f7c |
falls back to the "close all fds we don't care about up to limit"
|
|
|
5e5f7c |
method.
|
|
|
5e5f7c |
|
|
|
5e5f7c |
https://bugzilla.redhat.com/show_bug.cgi?id=1656519
|
|
|
5e5f7c |
|
|
|
5e5f7c |
[1] https://github.com/systemd/systemd/blob/5238e9575906297608ff802a27e2ff9effa3b338/src/basic/fd-util.c#L217
|
|
|
5e5f7c |
---
|
|
|
5e5f7c |
src/subproc.c | 71 ++++++++++++++++++++++++++++++++++++++++++++-------
|
|
|
5e5f7c |
1 file changed, 62 insertions(+), 9 deletions(-)
|
|
|
5e5f7c |
|
|
|
5e5f7c |
diff --git a/src/subproc.c b/src/subproc.c
|
|
|
5e5f7c |
index e49e3762..8df836ae 100644
|
|
|
5e5f7c |
--- a/src/subproc.c
|
|
|
5e5f7c |
+++ b/src/subproc.c
|
|
|
5e5f7c |
@@ -19,6 +19,7 @@
|
|
|
5e5f7c |
|
|
|
5e5f7c |
#include <sys/types.h>
|
|
|
5e5f7c |
#include <sys/wait.h>
|
|
|
5e5f7c |
+#include <dirent.h>
|
|
|
5e5f7c |
#include <errno.h>
|
|
|
5e5f7c |
#include <fcntl.h>
|
|
|
5e5f7c |
#include <paths.h>
|
|
|
5e5f7c |
@@ -436,6 +437,25 @@ cm_subproc_parse_args(void *parent, const char *cmdline, const char **error)
|
|
|
5e5f7c |
return argv;
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
|
|
|
5e5f7c |
+/* Based heavily on systemd version */
|
|
|
5e5f7c |
+static
|
|
|
5e5f7c |
+int safe_atoi(const char *s, int *ret_i) {
|
|
|
5e5f7c |
+ char *x = NULL;
|
|
|
5e5f7c |
+ long l;
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ errno = 0;
|
|
|
5e5f7c |
+ l = strtol(s, &x, 0);
|
|
|
5e5f7c |
+ if (errno > 0)
|
|
|
5e5f7c |
+ return -1;
|
|
|
5e5f7c |
+ if (!x || x == s || *x != 0)
|
|
|
5e5f7c |
+ return -1;
|
|
|
5e5f7c |
+ if ((long) (int) l != l)
|
|
|
5e5f7c |
+ return -1;
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ *ret_i = (int) l;
|
|
|
5e5f7c |
+ return 0;
|
|
|
5e5f7c |
+}
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
/* Redirect stdio to /dev/null, and mark everything else as close-on-exec,
|
|
|
5e5f7c |
* except for perhaps one to three of them that are passed in by number. */
|
|
|
5e5f7c |
void
|
|
|
5e5f7c |
@@ -443,6 +463,9 @@ cm_subproc_mark_most_cloexec(int fd, int fd2, int fd3)
|
|
|
5e5f7c |
{
|
|
|
5e5f7c |
int i;
|
|
|
5e5f7c |
long l;
|
|
|
5e5f7c |
+ DIR *dir = NULL;
|
|
|
5e5f7c |
+ struct dirent *de;
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
if ((fd != STDIN_FILENO) &&
|
|
|
5e5f7c |
(fd2 != STDIN_FILENO) &&
|
|
|
5e5f7c |
(fd3 != STDIN_FILENO)) {
|
|
|
5e5f7c |
@@ -482,17 +505,47 @@ cm_subproc_mark_most_cloexec(int fd, int fd2, int fd3)
|
|
|
5e5f7c |
close(STDERR_FILENO);
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
- for (i = getdtablesize() - 1; i >= 3; i--) {
|
|
|
5e5f7c |
- if ((i == fd) ||
|
|
|
5e5f7c |
- (i == fd2) ||
|
|
|
5e5f7c |
- (i == fd3)) {
|
|
|
5e5f7c |
- continue;
|
|
|
5e5f7c |
+ dir = opendir("/proc/self/fd");
|
|
|
5e5f7c |
+ if (!dir) {
|
|
|
5e5f7c |
+ /* /proc isn't available, fall back to old way */
|
|
|
5e5f7c |
+ for (i = getdtablesize() - 1; i >= 3; i--) {
|
|
|
5e5f7c |
+ if ((i == fd) ||
|
|
|
5e5f7c |
+ (i == fd2) ||
|
|
|
5e5f7c |
+ (i == fd3)) {
|
|
|
5e5f7c |
+ continue;
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
+ l = fcntl(i, F_GETFD);
|
|
|
5e5f7c |
+ if (l != -1) {
|
|
|
5e5f7c |
+ if (fcntl(i, F_SETFD, l | FD_CLOEXEC) != 0) {
|
|
|
5e5f7c |
+ cm_log(0, "Potentially leaking FD %d.\n", i);
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
- l = fcntl(i, F_GETFD);
|
|
|
5e5f7c |
- if (l != -1) {
|
|
|
5e5f7c |
- if (fcntl(i, F_SETFD, l | FD_CLOEXEC) != 0) {
|
|
|
5e5f7c |
- cm_log(0, "Potentially leaking FD %d.\n", i);
|
|
|
5e5f7c |
+ } else {
|
|
|
5e5f7c |
+ while ((de = readdir(dir)) != NULL) {
|
|
|
5e5f7c |
+ int i = -1;
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ if (safe_atoi(de->d_name, &i) < 0) {
|
|
|
5e5f7c |
+ continue;
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ if ((i == fd) ||
|
|
|
5e5f7c |
+ (i == fd2) ||
|
|
|
5e5f7c |
+ (i == fd3)) {
|
|
|
5e5f7c |
+ continue;
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ if (i == dirfd(dir)) {
|
|
|
5e5f7c |
+ continue;
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
+
|
|
|
5e5f7c |
+ l = fcntl(i, F_GETFD);
|
|
|
5e5f7c |
+ if (l != -1) {
|
|
|
5e5f7c |
+ if (fcntl(i, F_SETFD, l | FD_CLOEXEC) != 0) {
|
|
|
5e5f7c |
+ cm_log(0, "Potentially leaking FD %d.\n", i);
|
|
|
5e5f7c |
+ }
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
+ closedir(dir);
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
}
|
|
|
5e5f7c |
--
|
|
|
5e5f7c |
2.21.0
|
|
|
5e5f7c |
|