Re: [PATCH] selinux: use GFP_ATOMIC in convert_context()

[Ondrej Mosnacek]

On Tue, Oct 18, 2022 at 2:01 PM GONG, Ruiqi <gongruiqi1@xxxxxxxxxx> wrote:
> The following BUG_ON was triggered on a hardware environment:
>
>   SELinux: Converting 162 SID table entries...
>   BUG: sleeping function called from invalid context at __might_sleep_rtos+0x60/0x74 0x0
>   in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid: 5943, name: tar
>   CPU: 7 PID: 5943 Comm: tar Tainted: P O 5.10.0 #1
>   Call trace:
>    dump_backtrace+0x0/0x1c8
>    show_stack+0x18/0x28
>    dump_stack+0xe8/0x15c
>    ___might_sleep_rtos+0x168/0x17c
>    __might_sleep_rtos+0x60/0x74
>    __kmalloc_track_caller+0xa0/0x7dc
>    kstrdup+0x54/0xac
>    convert_context+0x48/0x2e4
>    sidtab_context_to_sid+0x1c4/0x36c
>    security_context_to_sid_core+0x168/0x238
>    security_context_to_sid_default+0x14/0x24
>    inode_doinit_use_xattr+0x164/0x1e4
>    inode_doinit_with_dentry+0x1c0/0x488
>    selinux_d_instantiate+0x20/0x34
>    security_d_instantiate+0x70/0xbc
>    d_splice_alias+0x4c/0x3c0
>    ext4_lookup+0x1d8/0x200 [ext4]
>    __lookup_slow+0x12c/0x1e4
>    walk_component+0x100/0x200
>    path_lookupat+0x88/0x118
>    filename_lookup+0x98/0x130
>    user_path_at_empty+0x48/0x60
>    vfs_statx+0x84/0x140
>    vfs_fstatat+0x20/0x30
>    __se_sys_newfstatat+0x30/0x74
>    __arm64_sys_newfstatat+0x1c/0x2c
>    el0_svc_common.constprop.0+0x100/0x184
>    do_el0_svc+0x1c/0x2c
>    el0_svc+0x20/0x34
>    el0_sync_handler+0x80/0x17c
>    el0_sync+0x13c/0x140
>   SELinux: Context system_u:object_r:pssp_rsyslog_log_t:s0:c0 is not valid (left unmapped).
>
> It was found that convert_context() (hooked by convert->func) might
> sleep in a critial section of spin_lock_irqsave in
> sidtab_context_to_sid(). Fix this problem by changing the memory
> allocation in convert_context() from GFP_KERNEL to GFP_ATOMIC.

Good catch! However, convert_context() (and
sidtab_convert_params::func) has two callers:
1. sidtab_context_to_sid(), which requires GFP_ATOMIC,
2. sidtab_convert_tree()/sidtab_convert(), where GFP_KERNEL would be okay.

So a more optimal fix would be to add a gfp_t argument to
convert_context()/sidtab_convert_params::func and pass
GFP_KERNEL/_ATOMIC as appropriate in the individual callers.

> Reported-by: Tan Ninghao <tanninghao1@xxxxxxxxxx>
> Fixes: ee1a84fdfeed ("selinux: overhaul sidtab to fix bug and improve performance")
> Signed-off-by: GONG, Ruiqi <gongruiqi1@xxxxxxxxxx>
> ---
>  security/selinux/ss/services.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/security/selinux/ss/services.c b/security/selinux/ss/services.c
> index fe5fcf571c56..523876bb7df3 100644
> --- a/security/selinux/ss/services.c
> +++ b/security/selinux/ss/services.c
> @@ -2036,7 +2036,7 @@ static int convert_context(struct context *oldc, struct context *newc, void *p)
>         args = p;
>
>         if (oldc->str) {
> -               s = kstrdup(oldc->str, GFP_KERNEL);
> +               s = kstrdup(oldc->str, GFP_ATOMIC);
>                 if (!s)
>                         return -ENOMEM;
>
> --
> 2.25.1
>

-- 
Ondrej Mosnacek
Senior Software Engineer, Linux Security - SELinux kernel
Red Hat, Inc.