1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
|
Load modules by absolute path in busybox modprobe
Our switch to busybox modprobe broke ZFS module loading where busybox
modprobe would load two modules and then fail. Limited developer time
resulted in a hack being put into place to repeat modprobe until ZFS
appeared. However, this was never a real long term solution.
Recent analysis with strace suggests that loading two modules corrupts
busybox's current working directory inside the kernel. Consequently,
subsequent tests where absolute paths were used instead of relative ones
made the problem disappear.
Modifying busybox to use full paths when loading modules makes module
loading work on all affected kernels. While the long term plan is to fix
the kernel, this workaround will be needed indefinitely for affected
kernels, even after mainline Linux is fixed.
Signed-off-by: Richard Yao <ryao@gentoo.org>
Added via commit 006a5d6d56e622b5ef82e5a066ca7af7b8c2aeed.
--- a/modutils/modprobe.c
+++ b/modutils/modprobe.c
@@ -413,7 +413,7 @@ static int do_modprobe(struct module_entry *m)
rc = 0;
while (m->deps) {
struct module_entry *m2;
- char *fn, *options;
+ char *fn, *options, *path;
rc = 0;
fn = llist_pop(&m->deps); /* we leak it */
@@ -460,7 +460,11 @@ static int do_modprobe(struct module_entry *m)
continue;
}
- rc = bb_init_module(fn, options);
+ path = xmalloc(strlen(fn) + strlen(CONFIG_DEFAULT_MODULES_DIR) + strlen(G.uts.release) + 3);
+ sprintf(path, "%s/%s/%s", CONFIG_DEFAULT_MODULES_DIR, G.uts.release, fn);
+
+ rc = bb_init_module(path, options);
+ free(path);
DBG("loaded %s '%s', rc:%d", fn, options, rc);
if (rc == EEXIST)
rc = 0;
|
GitWeb
