lkml.org 
[lkml]   [2015]   [Apr]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/2] livepatch: Fix the bug if the function name is larger than KSYM_NAME_LEN-1
On Tue, Apr 14, 2015 at 08:26:29AM +0800, Minfei Huang wrote:
> On 04/13/15 at 06:13P, Josh Poimboeuf wrote:
> > On Sun, Apr 12, 2015 at 09:15:54PM +0800, Minfei Huang wrote:
> > > For now, the kallsyms will only store the first (KSYM_NAME_LEN-1). The
> > > kallsyms name is same for the function which first (KSYM_NAME_LEN-1) is
> > > same, but the rest is not.
> > >
> > > Then function will never be patched, although function name and address
> > > are provided both. The reason caused this bug is livepatch cannt
> > > recognize the function name.
> > >
> > > Now, livepatch will verify the function name with first (KSYM_NAME_LEN-1)
> > > and address, if provided. Once they are matched, we can confirm that the
> > > patched function is found.
> >
> > From scripts/kallsyms.c:
> >
> > if (strlen(str) > KSYM_NAME_LEN) {
> > fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n"
> > "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
> > str, strlen(str), KSYM_NAME_LEN);
> > return -1;
> > }
> >
> > So I think such a long symbol name wouldn't be added to the kallsyms
> > database in the first place.
> >
>
> Actually, kernel allows overlength function name to be used. Following
> is my testing module.
>
> We can got the address in /proc/kallsyms.
> $ cat /proc/kallsyms | grep sysfs_print
> ffffffffa0000000 t sys_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_pri [sysfs_print]
> ffffffffa0000010 t kobj_release [sysfs_print]
> ffffffffa0000020 t sys_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_pri [sysfs_print]
> ffffffffa00004e0 b root_kobj [sysfs_print]
> ffffffffa0000200 d print_ktype [sysfs_print]
> ffffffffa00004a0 b print_kobj [sysfs_print]
> ffffffffa000004c t sys_print_exit [sysfs_print]
> ffffffffa0000144 r __func__.14514 [sysfs_print]
> ffffffffa0000230 d kobj_attrs [sysfs_print]
> ffffffffa0000240 d sys_print_kobj_attr [sysfs_print]
> ffffffffa0000260 d __this_module [sysfs_print]
> ffffffffa000004c t cleanup_module [sysfs_print]
>
> Code:
>
> static ssize_t sys_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_store(struct kobject *kobj, s
> const char *buf, size_t count)
> {
> return count;
> }
>
> static ssize_t sys_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_show(struct kobject *kobj,
> struct kobj_attribute *attr, char *buf)
> {
> return snprintf(buf, PAGE_SIZE-1, "%s\n", "This is printed by module");
> }
>
> static struct kobj_attribute sys_print_kobj_attr = __ATTR_RW(sys_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_print_p
> static struct attribute *kobj_attrs[] = {
> &sys_print_kobj_attr.attr,
> NULL
> };
>

Hm, this seems like a kallsyms bug. IMO it should either fail the build
or omit the symbol from the kallsyms db. Truncating it seems dangerous
and counterintuitive.

But regardless I really don't see a good reason to encourage this kind
of insanity in the livepatch code.

--
Josh


\
 
 \ /
  Last update: 2015-04-14 07:21    [W:0.175 / U:0.252 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site