-----邮件原件-----
发件人: [email protected] <[email protected]> 代表 Ondrej Mosnacek
发送时间: 2021年12月6日 17:11
收件人: 赵军奎 <[email protected]>
抄送: Paul Moore <[email protected]>; Stephen Smalley <[email protected]>; Eric Paris <[email protected]>; SElinux list <[email protected]>; Linux kernel mailing list <[email protected]>
主题: Re: [PATCH v2] security/selinux: fix potential memleak in error branch
On Mon, Dec 6, 2021 at 8:19 AM Bernard Zhao <[email protected]> wrote:
> This patch try to fix potential memleak in error branch.
>
> Signed-off-by: Bernard Zhao <[email protected]>
> ---
> security/selinux/hooks.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c index
> 62d30c0a30c2..8dc140399a23 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -983,18 +983,22 @@ static int selinux_sb_clone_mnt_opts(const
> struct super_block *oldsb, static int selinux_add_opt(int token,
> const char *s, void **mnt_opts) {
> struct selinux_mnt_opts *opts = *mnt_opts;
> + bool is_alloc_opts = false;
>
> if (token == Opt_seclabel) /* eaten and completely ignored */
> return 0;
>
> + if (!s)
> + return -ENOMEM;
> +
> if (!opts) {
> opts = kzalloc(sizeof(struct selinux_mnt_opts), GFP_KERNEL);
> if (!opts)
> return -ENOMEM;
> *mnt_opts = opts;
> + is_alloc_opts = true;
> }
> - if (!s)
> - return -ENOMEM;
> +
> switch (token) {
> case Opt_context:
> if (opts->context || opts->defcontext) @@ -1019,6
> +1023,8 @@ static int selinux_add_opt(int token, const char *s, void **mnt_opts)
> }
> return 0;
> Einval:
> + if (is_alloc_opts)
> + kfree(opts);
> pr_warn(SEL_MOUNT_FAIL_MSG);
> return -EINVAL;
> }
> --
> 2.33.1
>The problem is a bit more tricky... As is, this patch will lead to double frees in some cases. Look at security_sb_eat_lsm_opts() callers, for example - some of them don't do anything when error is returned, some call
>security_free_mnt_opts() on the opts regardless, some will let it store the opts in fc->security, where
>put_fs_context() will eventually call security_free_mnt_opts() on them.
>You need to at least *mnt_opts = NULL after kfree(opts), but it would be also nice to make the LSM hook callers more consistent in what they do in the error path and document the fact that *mnt_opts will be NULL
>on error in lsm_hooks.h (in case of the sb_eat_lsm_opts hook).
Hi Ondrej Mosnacek:
Thanks for your comments!
I am not sure if there is some gap, for this part " it would be also nice to make the LSM hook callers more consistent in what they do in the error path and document the fact that *mnt_opts will be NULL
on error in lsm_hooks.h (in case of the sb_eat_lsm_opts hook)"
I am not sure if this is OK:
116 * @sb_eat_lsm_opts:
117 * Eat (scan @orig options) and save them in @mnt_opts.
118 * If error is returned, then the *mnt_opts will be NULL.
Please help to double check, thanks!
BR//Bernard
--
>Ondrej Mosnacek
>Software Engineer, Linux Security - SELinux kernel Red Hat, Inc.
On Tue, Dec 7, 2021 at 1:05 PM 赵军奎 <[email protected]> wrote:
> -----邮件原件-----
> 发件人: [email protected] <[email protected]> 代表 Ondrej Mosnacek
> 发送时间: 2021年12月6日 17:11
> 收件人: 赵军奎 <[email protected]>
> 抄送: Paul Moore <[email protected]>; Stephen Smalley <[email protected]>; Eric Paris <[email protected]>; SElinux list <[email protected]>; Linux kernel mailing list <[email protected]>
> 主题: Re: [PATCH v2] security/selinux: fix potential memleak in error branch
>
> On Mon, Dec 6, 2021 at 8:19 AM Bernard Zhao <[email protected]> wrote:
> > This patch try to fix potential memleak in error branch.
> >
> > Signed-off-by: Bernard Zhao <[email protected]>
> > ---
> > security/selinux/hooks.c | 10 ++++++++--
> > 1 file changed, 8 insertions(+), 2 deletions(-)
> >
> > diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c index
> > 62d30c0a30c2..8dc140399a23 100644
> > --- a/security/selinux/hooks.c
> > +++ b/security/selinux/hooks.c
> > @@ -983,18 +983,22 @@ static int selinux_sb_clone_mnt_opts(const
> > struct super_block *oldsb, static int selinux_add_opt(int token,
> > const char *s, void **mnt_opts) {
> > struct selinux_mnt_opts *opts = *mnt_opts;
> > + bool is_alloc_opts = false;
> >
> > if (token == Opt_seclabel) /* eaten and completely ignored */
> > return 0;
> >
> > + if (!s)
> > + return -ENOMEM;
> > +
> > if (!opts) {
> > opts = kzalloc(sizeof(struct selinux_mnt_opts), GFP_KERNEL);
> > if (!opts)
> > return -ENOMEM;
> > *mnt_opts = opts;
> > + is_alloc_opts = true;
> > }
> > - if (!s)
> > - return -ENOMEM;
> > +
> > switch (token) {
> > case Opt_context:
> > if (opts->context || opts->defcontext) @@ -1019,6
> > +1023,8 @@ static int selinux_add_opt(int token, const char *s, void **mnt_opts)
> > }
> > return 0;
> > Einval:
> > + if (is_alloc_opts)
> > + kfree(opts);
> > pr_warn(SEL_MOUNT_FAIL_MSG);
> > return -EINVAL;
> > }
> > --
> > 2.33.1
>
> >The problem is a bit more tricky... As is, this patch will lead to double frees in some cases. Look at security_sb_eat_lsm_opts() callers, for example - some of them don't do anything when error is returned, some call
> >security_free_mnt_opts() on the opts regardless, some will let it store the opts in fc->security, where
> >put_fs_context() will eventually call security_free_mnt_opts() on them.
>
> >You need to at least *mnt_opts = NULL after kfree(opts), but it would be also nice to make the LSM hook callers more consistent in what they do in the error path and document the fact that *mnt_opts will be NULL
> >on error in lsm_hooks.h (in case of the sb_eat_lsm_opts hook).
> Hi Ondrej Mosnacek:
>
> Thanks for your comments!
> I am not sure if there is some gap, for this part " it would be also nice to make the LSM hook callers more consistent in what they do in the error path and document the fact that *mnt_opts will be NULL
> on error in lsm_hooks.h (in case of the sb_eat_lsm_opts hook)"
> I am not sure if this is OK:
> 116 * @sb_eat_lsm_opts:
> 117 * Eat (scan @orig options) and save them in @mnt_opts.
> 118 * If error is returned, then the *mnt_opts will be NULL.
> Please help to double check, thanks!
I'd prefer something like:
If the hook returns 0, the caller is responsible for destroying the
returned @mnt_opts using the @sb_free_mnt_opts hook. The LSMs must not
expect the callers to destroy @mnt_opts if the hook returns an error
and should always set it to NULL in such case.
(You may want to double-check that the other implementations of this
hook (i.e. security/smack/smack_lsm.c) follow that contract and fix
them if necessary.)
Thanks for your efforts to improve this!
--
Ondrej Mosnacek
Software Engineer, Linux Security - SELinux kernel
Red Hat, Inc.