Re: [linux-yocto] [yocto-kernel-cache][yocto-5.4][PATCH] NFS: fix general protection fault in nfs_mount
In message: [yocto-kernel-cache][yocto-5.4][PATCH] NFS: fix general protection
fault in nfs_mount
on 28/06/2021 Ovidiu Panait wrote:
> While fuzzing the kernel with syzkaller, the following issue was triggered:
>
> Call Trace:
> strlen usr/src/kernel/include/linux/string.h:306 [inline]
> nfs_mount+0x316/0x740 usr/src/kernel/fs/nfs/mount_clnt.c:174
> ? __kernel_text_address+0x12/0x40 usr/src/kernel/kernel/extable.c:95
> ? mnt_xdr_dec_mountres+0x3a0/0x3a0 usr/src/kernel/fs/nfs/mount_clnt.c:371
> ? arch_stack_walk+0xa2/0xf0 usr/src/kernel/arch/x86/kernel/stacktrace.c:26
> ? mark_usage usr/src/kernel/kernel/locking/lockdep.c:3426 [inline]
> ? __lock_acquire+0x557/0x5ed0 usr/src/kernel/kernel/locking/lockdep.c:3752
> nfs_request_mount.constprop.0+0x42e/0x680 usr/src/kernel/fs/nfs/super.c:1817
> ? nfs_fs_mount+0x2870/0x2870 usr/src/kernel/fs/nfs/super.c:2080
> ? lockdep_hardirqs_on+0x580/0x580
> usr/src/kernel/kernel/locking/lockdep.c:3264
> ? lockdep_hardirqs_on+0x580/0x580
> usr/src/kernel/kernel/locking/lockdep.c:3264
> nfs_try_mount_request usr/src/kernel/fs/nfs/super.c:1839 [inline]
> nfs_try_mount+0x25b/0x93a usr/src/kernel/fs/nfs/super.c:1907
> ? lock_downgrade+0x770/0x770 usr/src/kernel/kernel/locking/lockdep.c:4043
> ? nfs_request_mount.constprop.0+0x680/0x680
> usr/src/kernel/fs/nfs/super.c:1800
> ? lock_downgrade+0x770/0x770 usr/src/kernel/kernel/locking/lockdep.c:4043
> ? rwlock_bug.part.0+0x90/0x90 usr/src/kernel/include/linux/sched.h:1250
> ? kasan_check_read+0x11/0x20 usr/src/kernel/mm/kasan/common.c:94
> ? atomic_read usr/src/kernel/include/asm-generic/atomic-instrumented.h:26
> [inline]
> ? queued_spin_is_locked usr/src/kernel/include/asm-generic/qspinlock.h:26
> [inline]
> ? debug_spin_unlock usr/src/kernel/kernel/locking/spinlock_debug.c:98
> [inline]
> ? do_raw_spin_unlock+0x59/0x260
> usr/src/kernel/kernel/locking/spinlock_debug.c:138
> ? __raw_spin_unlock usr/src/kernel/include/linux/spinlock_api_smp.h:152
> [inline]
> ? _raw_spin_unlock+0x32/0x50 usr/src/kernel/kernel/locking/spinlock.c:183
> ? find_nfs_version+0xe6/0x110 usr/src/kernel/include/linux/err.h:26
> nfs_fs_mount+0xe83/0x2870 usr/src/kernel/fs/nfs/super.c:2750
> ? nfs_remount+0x19e0/0x19e0 usr/src/kernel/include/net/ipv6.h:517
> ? trace_kfree usr/src/kernel/include/trace/events/kmem.h:137 [inline]
> ? kfree+0x194/0x2c0 usr/src/kernel/mm/slub.c:3983
> ? nfs_clone_super+0x420/0x420 usr/src/kernel/fs/nfs/super.c:2421
> ? nfs_parse_mount_options+0x2400/0x2400 usr/src/kernel/fs/nfs/super.c:1396
> ? vfs_parse_fs_string+0x11b/0x170 usr/src/kernel/fs/fs_context.c:190
> ? vfs_parse_fs_param+0x540/0x540 usr/src/kernel/fs/fs_context.c:163
> ? nfs_remount+0x19e0/0x19e0 usr/src/kernel/include/net/ipv6.h:517
> legacy_get_tree+0x10f/0x220 usr/src/kernel/fs/fs_context.c:659
> ? legacy_parse_monolithic+0x124/0x180 usr/src/kernel/fs/fs_context.c:643
> vfs_get_tree+0x98/0x3a0 usr/src/kernel/fs/super.c:1478
> do_new_mount usr/src/kernel/fs/namespace.c:2801 [inline]
> do_mount+0x1381/0x1c20 usr/src/kernel/fs/namespace.c:3121
> ? copy_mount_string+0x40/0x40 usr/src/kernel/fs/namespace.c:3024
> ? kasan_kmalloc+0x9/0x10 usr/src/kernel/mm/kasan/common.c:509
> ? kmem_cache_alloc_trace+0x147/0x310 usr/src/kernel/mm/slub.c:2810
> ? __sanitizer_cov_trace_const_cmp4+0x16/0x20 usr/src/kernel/kernel/kcov.c:189
> ? memset usr/src/kernel/include/linux/string.h:369 [inline]
> ? copy_mount_options+0x2d9/0x3d0 usr/src/kernel/fs/namespace.c:3017
> ksys_mount+0xd8/0x130 usr/src/kernel/fs/namespace.c:3330
> __do_sys_mount usr/src/kernel/fs/namespace.c:3344 [inline]
> __se_sys_mount usr/src/kernel/fs/namespace.c:3341 [inline]
> __x64_sys_mount+0xc3/0x150 usr/src/kernel/fs/namespace.c:3341
> do_syscall_64+0xc7/0x600 usr/src/kernel/arch/x86/entry/common.c:295
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> It can be reproduced using the following simplified program:
> #include
> #include
>
> int main()
> {
>mount(NULL, "./file0", "nfs", 0x400, "\x07\000\000\000");
>
>return 0;
> }
>
> It was introduced by the follwing yocto-specific commit:
> commit ad818f6b28a610b53814d8b1d0da2fadfd83a3ad
> Author: Jason Wessel
> Date: Wed Jan 7 00:59:33 2009 -0500
>
> NFS: allow nfs root mount to use alternate rpc ports
>
> Allow an nfs root mount to use alternate RPC ports for mountd and nfsd.
>
> Signed-off-by: Jason Wessel
> [forward port to 2.6.33+]
> Signed-off-by: Bruce Ashfield
>
> Since hardknott, "NFS: allow nfs root mount to use alternate rpc ports" patch
> is
> no longer applied by yocto-kernel-cache (a4283cc6be0c65 ("kver: update
> -dev to v5.6-rc5")) due to a merge conflict. Also, in later kernels (since
> v5.6
> commit f2aedb713c284 ("NFS: Add fs_context support")), the nfs parser has been
> refactored to use generic fs_context, so it is not clear how easy would be to
> reproduce this issue.
>
> The null pointer dereference takes place only when
[linux-yocto] [yocto-kernel-cache][yocto-5.4][PATCH] NFS: fix general protection fault in nfs_mount
While fuzzing the kernel with syzkaller, the following issue was triggered:
Call Trace:
strlen usr/src/kernel/include/linux/string.h:306 [inline]
nfs_mount+0x316/0x740 usr/src/kernel/fs/nfs/mount_clnt.c:174
? __kernel_text_address+0x12/0x40 usr/src/kernel/kernel/extable.c:95
? mnt_xdr_dec_mountres+0x3a0/0x3a0 usr/src/kernel/fs/nfs/mount_clnt.c:371
? arch_stack_walk+0xa2/0xf0 usr/src/kernel/arch/x86/kernel/stacktrace.c:26
? mark_usage usr/src/kernel/kernel/locking/lockdep.c:3426 [inline]
? __lock_acquire+0x557/0x5ed0 usr/src/kernel/kernel/locking/lockdep.c:3752
nfs_request_mount.constprop.0+0x42e/0x680 usr/src/kernel/fs/nfs/super.c:1817
? nfs_fs_mount+0x2870/0x2870 usr/src/kernel/fs/nfs/super.c:2080
? lockdep_hardirqs_on+0x580/0x580 usr/src/kernel/kernel/locking/lockdep.c:3264
? lockdep_hardirqs_on+0x580/0x580 usr/src/kernel/kernel/locking/lockdep.c:3264
nfs_try_mount_request usr/src/kernel/fs/nfs/super.c:1839 [inline]
nfs_try_mount+0x25b/0x93a usr/src/kernel/fs/nfs/super.c:1907
? lock_downgrade+0x770/0x770 usr/src/kernel/kernel/locking/lockdep.c:4043
? nfs_request_mount.constprop.0+0x680/0x680 usr/src/kernel/fs/nfs/super.c:1800
? lock_downgrade+0x770/0x770 usr/src/kernel/kernel/locking/lockdep.c:4043
? rwlock_bug.part.0+0x90/0x90 usr/src/kernel/include/linux/sched.h:1250
? kasan_check_read+0x11/0x20 usr/src/kernel/mm/kasan/common.c:94
? atomic_read usr/src/kernel/include/asm-generic/atomic-instrumented.h:26
[inline]
? queued_spin_is_locked usr/src/kernel/include/asm-generic/qspinlock.h:26
[inline]
? debug_spin_unlock usr/src/kernel/kernel/locking/spinlock_debug.c:98 [inline]
? do_raw_spin_unlock+0x59/0x260
usr/src/kernel/kernel/locking/spinlock_debug.c:138
? __raw_spin_unlock usr/src/kernel/include/linux/spinlock_api_smp.h:152
[inline]
? _raw_spin_unlock+0x32/0x50 usr/src/kernel/kernel/locking/spinlock.c:183
? find_nfs_version+0xe6/0x110 usr/src/kernel/include/linux/err.h:26
nfs_fs_mount+0xe83/0x2870 usr/src/kernel/fs/nfs/super.c:2750
? nfs_remount+0x19e0/0x19e0 usr/src/kernel/include/net/ipv6.h:517
? trace_kfree usr/src/kernel/include/trace/events/kmem.h:137 [inline]
? kfree+0x194/0x2c0 usr/src/kernel/mm/slub.c:3983
? nfs_clone_super+0x420/0x420 usr/src/kernel/fs/nfs/super.c:2421
? nfs_parse_mount_options+0x2400/0x2400 usr/src/kernel/fs/nfs/super.c:1396
? vfs_parse_fs_string+0x11b/0x170 usr/src/kernel/fs/fs_context.c:190
? vfs_parse_fs_param+0x540/0x540 usr/src/kernel/fs/fs_context.c:163
? nfs_remount+0x19e0/0x19e0 usr/src/kernel/include/net/ipv6.h:517
legacy_get_tree+0x10f/0x220 usr/src/kernel/fs/fs_context.c:659
? legacy_parse_monolithic+0x124/0x180 usr/src/kernel/fs/fs_context.c:643
vfs_get_tree+0x98/0x3a0 usr/src/kernel/fs/super.c:1478
do_new_mount usr/src/kernel/fs/namespace.c:2801 [inline]
do_mount+0x1381/0x1c20 usr/src/kernel/fs/namespace.c:3121
? copy_mount_string+0x40/0x40 usr/src/kernel/fs/namespace.c:3024
? kasan_kmalloc+0x9/0x10 usr/src/kernel/mm/kasan/common.c:509
? kmem_cache_alloc_trace+0x147/0x310 usr/src/kernel/mm/slub.c:2810
? __sanitizer_cov_trace_const_cmp4+0x16/0x20 usr/src/kernel/kernel/kcov.c:189
? memset usr/src/kernel/include/linux/string.h:369 [inline]
? copy_mount_options+0x2d9/0x3d0 usr/src/kernel/fs/namespace.c:3017
ksys_mount+0xd8/0x130 usr/src/kernel/fs/namespace.c:3330
__do_sys_mount usr/src/kernel/fs/namespace.c:3344 [inline]
__se_sys_mount usr/src/kernel/fs/namespace.c:3341 [inline]
__x64_sys_mount+0xc3/0x150 usr/src/kernel/fs/namespace.c:3341
do_syscall_64+0xc7/0x600 usr/src/kernel/arch/x86/entry/common.c:295
entry_SYSCALL_64_after_hwframe+0x49/0xbe
It can be reproduced using the following simplified program:
#include
#include
int main()
{
mount(NULL, "./file0", "nfs", 0x400, "\x07\000\000\000");
return 0;
}
It was introduced by the follwing yocto-specific commit:
commit ad818f6b28a610b53814d8b1d0da2fadfd83a3ad
Author: Jason Wessel
Date: Wed Jan 7 00:59:33 2009 -0500
NFS: allow nfs root mount to use alternate rpc ports
Allow an nfs root mount to use alternate RPC ports for mountd and nfsd.
Signed-off-by: Jason Wessel
[forward port to 2.6.33+]
Signed-off-by: Bruce Ashfield
Since hardknott, "NFS: allow nfs root mount to use alternate rpc ports" patch is
no longer applied by yocto-kernel-cache (a4283cc6be0c65 ("kver: update
-dev to v5.6-rc5")) due to a merge conflict. Also, in later kernels (since v5.6
commit f2aedb713c284 ("NFS: Add fs_context support")), the nfs parser has been
refactored to use generic fs_context, so it is not clear how easy would be to
reproduce this issue.
The null pointer dereference takes place only when data->version is 0x07,
because that is the only switch case that does not perform struct
nfs_parsed_mount_data initialization (case 6 in the switch). All other
switch cases fall through case 6 and perform the initializations properly.
To fix this, also run legacy version 6 binary mount logic when NFS version
is 0x07.
Signed-off-by: Ovidiu
