[PATCH v1 net 0/2] net: Fix error/warning by -fstrict-flex-arrays=3

[PATCH v1 net 0/2] net: Fix error/warning by -fstrict-flex-arrays=3

[Kuniyuki Iwashima]

df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3") started applying
strict rules for standard string functions (strlen(), memcpy(), etc.) if
CONFIG_FORTIFY_SOURCE=y.

This series fixes two false positives caught by syzkaller.


Kuniyuki Iwashima (2):
  af_unix: Fix fortify_panic() in unix_bind_bsd().
  af_packet: Fix warning of fortified memcpy() in packet_getname().

 net/packet/af_packet.c |  5 ++++-
 net/unix/af_unix.c     | 16 +++++++++++++---
 2 files changed, 17 insertions(+), 4 deletions(-)

-- 
2.30.2

[PATCH v1 net 1/2] af_unix: Fix fortify_panic() in unix_bind_bsd().

[Kuniyuki Iwashima]

syzkaller found a bug in unix_bind_bsd() [0].  We can reproduce it
by bind()ing a socket on a path with length 108.

108 is the size of sun_addr of struct sockaddr_un and is the maximum
valid length for the pathname socket.  When calling bind(), we use
struct sockaddr_storage as the actual buffer size, so terminating
sun_addr[108] with null is legitimate.

However, strlen(sunaddr) for such a case causes fortify_panic() if
CONFIG_FORTIFY_SOURCE=y.  __fortify_strlen() has no idea about the
actual buffer size and takes 108 as overflow, although 108 still
fits in struct sockaddr_storage.

Let's make __fortify_strlen() recognise the actual buffer size.

[0]:
detected buffer overflow in __fortify_strlen
kernel BUG at lib/string_helpers.c:1031!
Internal error: Oops - BUG: 00000000f2000800 [#1] PREEMPT SMP
Modules linked in:
CPU: 0 PID: 255 Comm: syz-executor296 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
Hardware name: linux,dummy-virt (DT)
pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
lr : fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
sp : ffff800089817af0
x29: ffff800089817af0 x28: ffff800089817b40 x27: 1ffff00011302f68
x26: 000000000000006e x25: 0000000000000012 x24: ffff800087e60140
x23: dfff800000000000 x22: ffff800089817c20 x21: ffff800089817c8e
x20: 000000000000006c x19: ffff00000c323900 x18: ffff800086ab1630
x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000001
x14: 1ffff00011302eb8 x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : 64a26b65474d2a00
x8 : 64a26b65474d2a00 x7 : 0000000000000001 x6 : 0000000000000001
x5 : ffff800089817438 x4 : ffff800086ac99e0 x3 : ffff800080f19e8c
x2 : 0000000000000001 x1 : 0000000100000000 x0 : 000000000000002c
Call trace:
 fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
 _Z16__fortify_strlenPKcU25pass_dynamic_object_size1 include/linux/fortify-string.h:217 [inline]
 unix_bind_bsd net/unix/af_unix.c:1212 [inline]
 unix_bind+0xba8/0xc58 net/unix/af_unix.c:1326
 __sys_bind+0x1ac/0x248 net/socket.c:1792
 __do_sys_bind net/socket.c:1803 [inline]
 __se_sys_bind net/socket.c:1801 [inline]
 __arm64_sys_bind+0x7c/0x94 net/socket.c:1801
 __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
 invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
 el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
 do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
 el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
 el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
 el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591
Code: aa0003e1 d0000e80 91030000 97ffc91a (d4210000)

Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
Reported-by: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 net/unix/af_unix.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 123b35ddfd71..e1b1819b96d1 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -302,6 +302,18 @@ static void unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
 	((char *)sunaddr)[addr_len] = 0;
 }
 
+static int unix_strlen_bsd(struct sockaddr_un *sunaddr)
+{
+	/* Don't pass sunaddr->sun_path to strlen().  Otherwise, the
+	 * max valid length UNIX_PATH_MAX (108) will cause panic if
+	 * CONFIG_FORTIFY_SOURCE=y.  Let __fortify_strlen() know that
+	 * the actual buffer is struct sockaddr_storage and that 108
+	 * is within __data[].  See also: unix_mkname_bsd().
+	 */
+	return strlen(((struct sockaddr_storage *)sunaddr)->__data) +
+		offsetof(struct sockaddr_un, sun_path) + 1;
+}
+
 static void __unix_remove_socket(struct sock *sk)
 {
 	sk_del_node_init(sk);
@@ -1209,9 +1221,7 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
 	int err;
 
 	unix_mkname_bsd(sunaddr, addr_len);
-	addr_len = strlen(sunaddr->sun_path) +
-		offsetof(struct sockaddr_un, sun_path) + 1;
-
+	addr_len = unix_strlen_bsd(sunaddr);
 	addr = unix_create_addr(sunaddr, addr_len);
 	if (!addr)
 		return -ENOMEM;
-- 
2.30.2

RE: [PATCH v1 net 1/2] af_unix: Fix fortify_panic() in unix_bind_bsd().

[Willem de Bruijn]

Kuniyuki Iwashima wrote:
> syzkaller found a bug in unix_bind_bsd() [0].  We can reproduce it
> by bind()ing a socket on a path with length 108.
> 
> 108 is the size of sun_addr of struct sockaddr_un and is the maximum
> valid length for the pathname socket.  When calling bind(), we use
> struct sockaddr_storage as the actual buffer size, so terminating
> sun_addr[108] with null is legitimate.
> 
> However, strlen(sunaddr) for such a case causes fortify_panic() if
> CONFIG_FORTIFY_SOURCE=y.  __fortify_strlen() has no idea about the
> actual buffer size and takes 108 as overflow, although 108 still
> fits in struct sockaddr_storage.
> 
> Let's make __fortify_strlen() recognise the actual buffer size.
> 
> [0]:
> detected buffer overflow in __fortify_strlen
> kernel BUG at lib/string_helpers.c:1031!
> Internal error: Oops - BUG: 00000000f2000800 [#1] PREEMPT SMP
> Modules linked in:
> CPU: 0 PID: 255 Comm: syz-executor296 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
> Hardware name: linux,dummy-virt (DT)
> pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
> lr : fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
> sp : ffff800089817af0
> x29: ffff800089817af0 x28: ffff800089817b40 x27: 1ffff00011302f68
> x26: 000000000000006e x25: 0000000000000012 x24: ffff800087e60140
> x23: dfff800000000000 x22: ffff800089817c20 x21: ffff800089817c8e
> x20: 000000000000006c x19: ffff00000c323900 x18: ffff800086ab1630
> x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000001
> x14: 1ffff00011302eb8 x13: 0000000000000000 x12: 0000000000000000
> x11: 0000000000000000 x10: 0000000000000000 x9 : 64a26b65474d2a00
> x8 : 64a26b65474d2a00 x7 : 0000000000000001 x6 : 0000000000000001
> x5 : ffff800089817438 x4 : ffff800086ac99e0 x3 : ffff800080f19e8c
> x2 : 0000000000000001 x1 : 0000000100000000 x0 : 000000000000002c
> Call trace:
>  fortify_panic+0x1c/0x20 lib/string_helpers.c:1030
>  _Z16__fortify_strlenPKcU25pass_dynamic_object_size1 include/linux/fortify-string.h:217 [inline]
>  unix_bind_bsd net/unix/af_unix.c:1212 [inline]
>  unix_bind+0xba8/0xc58 net/unix/af_unix.c:1326
>  __sys_bind+0x1ac/0x248 net/socket.c:1792
>  __do_sys_bind net/socket.c:1803 [inline]
>  __se_sys_bind net/socket.c:1801 [inline]
>  __arm64_sys_bind+0x7c/0x94 net/socket.c:1801
>  __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
>  invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
>  el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
>  do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
>  el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
>  el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
>  el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591
> Code: aa0003e1 d0000e80 91030000 97ffc91a (d4210000)
> 
> Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
> Reported-by: syzkaller <syzkaller@googlegroups.com>
> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>

Reviewed-by: Willem de Bruijn <willemb@google.com>

The extensive comments are really helpful to understand what's
going on.

An alternative would be to just cast sunaddr to a struct
sockaddr_storage *ss and use that both here and in unix_mkname_bsd?
It's not immediately trivial that the caller has always actually
allocated one of those. But the rest becomes self documenting.

> ---
>  net/unix/af_unix.c | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> index 123b35ddfd71..e1b1819b96d1 100644
> --- a/net/unix/af_unix.c
> +++ b/net/unix/af_unix.c
> @@ -302,6 +302,18 @@ static void unix_mkname_bsd(struct sockaddr_un *sunaddr, int addr_len)
>  	((char *)sunaddr)[addr_len] = 0;
>  }
>  
> +static int unix_strlen_bsd(struct sockaddr_un *sunaddr)
> +{
> +	/* Don't pass sunaddr->sun_path to strlen().  Otherwise, the
> +	 * max valid length UNIX_PATH_MAX (108) will cause panic if
> +	 * CONFIG_FORTIFY_SOURCE=y.  Let __fortify_strlen() know that
> +	 * the actual buffer is struct sockaddr_storage and that 108
> +	 * is within __data[].  See also: unix_mkname_bsd().
> +	 */
> +	return strlen(((struct sockaddr_storage *)sunaddr)->__data) +
> +		offsetof(struct sockaddr_un, sun_path) + 1;
> +}
> +
>  static void __unix_remove_socket(struct sock *sk)
>  {
>  	sk_del_node_init(sk);
> @@ -1209,9 +1221,7 @@ static int unix_bind_bsd(struct sock *sk, struct sockaddr_un *sunaddr,
>  	int err;
>  
>  	unix_mkname_bsd(sunaddr, addr_len);
> -	addr_len = strlen(sunaddr->sun_path) +
> -		offsetof(struct sockaddr_un, sun_path) + 1;
> -
> +	addr_len = unix_strlen_bsd(sunaddr);
>  	addr = unix_create_addr(sunaddr, addr_len);
>  	if (!addr)
>  		return -ENOMEM;
> -- 
> 2.30.2
> 

RE: [PATCH v1 net 1/2] af_unix: Fix fortify_panic() in unix_bind_bsd().

[Kees Cook]

On July 19, 2023 3:26:35 PM PDT, Willem de Bruijn <willemdebruijn.kernel@gmail.com> wrote:
>Kuniyuki Iwashima wrote:
>The extensive comments are really helpful to understand what's
>going on.
>
>An alternative would be to just cast sunaddr to a struct
>sockaddr_storage *ss and use that both here and in unix_mkname_bsd?
>It's not immediately trivial that the caller has always actually
>allocated one of those. But the rest becomes self documenting.

I would much prefer the internal APIs actually passed around the true sockaddr_storage pointer. This is what I did recently for NFS, for example:
https://git.kernel.org/linus/cf0d7e7f4520

-Kees


-- 
Kees Cook

Re: [PATCH v1 net 1/2] af_unix: Fix fortify_panic() in unix_bind_bsd().

[Kuniyuki Iwashima]

From: Kees Cook <kees@kernel.org>
Date: Wed, 19 Jul 2023 15:34:33 -0700
> On July 19, 2023 3:26:35 PM PDT, Willem de Bruijn <willemdebruijn.kernel@gmail.com> wrote:
> >Kuniyuki Iwashima wrote:
> >The extensive comments are really helpful to understand what's
> >going on.
> >
> >An alternative would be to just cast sunaddr to a struct
> >sockaddr_storage *ss and use that both here and in unix_mkname_bsd?
> >It's not immediately trivial that the caller has always actually
> >allocated one of those. But the rest becomes self documenting.

Yeah, this is also my initial attempt, and I separted it because
unix_find_bsd() need not to call it and I tried to separate
unnecessary calls in this series (compilers might drop the unused
strlen() though).

https://lore.kernel.org/netdev/20211124021431.48956-7-kuniyu@amazon.co.jp/


> 
> I would much prefer the internal APIs actually passed around the true sockaddr_storage pointer. This is what I did recently for NFS, for example:
> https://git.kernel.org/linus/cf0d7e7f4520

We can convert struct proto_ops and struct proto later as such
if needed, but I think it's too invasive as a fix.

[PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Kuniyuki Iwashima]

syzkaller found a warning in packet_getname() [0], where we try to
copy 16 bytes to sockaddr_ll.sll_addr[8].

Some devices (ip6gre, vti6, ip6tnl) have 16 bytes address expressed
by struct in6_addr.

The write seems to overflow, but actually not since we use struct
sockaddr_storage defined in __sys_getsockname().

To avoid the warning, we need to let __fortify_memcpy_chk() know the
actual buffer size.

Another option would be to use strncpy() and limit the copied length
to sizeof(sll_addr), but it will return the partial address and might
break an application that passes sockaddr_storage to getsockname().

[0]:
memcpy: detected field-spanning write (size 16) of single field "sll->sll_addr" at net/packet/af_packet.c:3604 (size 8)
WARNING: CPU: 0 PID: 255 at net/packet/af_packet.c:3604 packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
Modules linked in:
CPU: 0 PID: 255 Comm: syz-executor750 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
Hardware name: linux,dummy-virt (DT)
pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
lr : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
sp : ffff800089887bc0
x29: ffff800089887bc0 x28: ffff000010f80f80 x27: 0000000000000003
x26: dfff800000000000 x25: ffff700011310f80 x24: ffff800087d55000
x23: dfff800000000000 x22: ffff800089887c2c x21: 0000000000000010
x20: ffff00000de08310 x19: ffff800089887c20 x18: ffff800086ab1630
x17: 20646c6569662065 x16: 6c676e697320666f x15: 0000000000000001
x14: 1fffe0000d56d7ca x13: 0000000000000000 x12: 0000000000000000
x11: 0000000000000000 x10: 0000000000000000 x9 : 3e60944c3da92b00
x8 : 3e60944c3da92b00 x7 : 0000000000000001 x6 : 0000000000000001
x5 : ffff8000898874f8 x4 : ffff800086ac99e0 x3 : ffff8000803f8808
x2 : 0000000000000001 x1 : 0000000100000000 x0 : 0000000000000000
Call trace:
 packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
 __sys_getsockname+0x168/0x24c net/socket.c:2042
 __do_sys_getsockname net/socket.c:2057 [inline]
 __se_sys_getsockname net/socket.c:2054 [inline]
 __arm64_sys_getsockname+0x7c/0x94 net/socket.c:2054
 __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
 invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
 el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
 do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
 el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
 el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
 el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591

Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
Reported-by: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
---
 net/packet/af_packet.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 85ff90a03b0c..5eef94a32a4f 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -3601,7 +3601,10 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
 	if (dev) {
 		sll->sll_hatype = dev->type;
 		sll->sll_halen = dev->addr_len;
-		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
+
+		/* Let __fortify_memcpy_chk() know the actual buffer size. */
+		memcpy(((struct sockaddr_storage *)sll)->__data +
+		       offsetof(struct sockaddr_ll, sll_addr), dev->dev_addr, dev->addr_len);
 	} else {
 		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
 		sll->sll_halen = 0;
-- 
2.30.2

Re: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Kuniyuki Iwashima]

From: Kuniyuki Iwashima <kuniyu@amazon.com>
Date: Wed, 19 Jul 2023 11:53:22 -0700
> syzkaller found a warning in packet_getname() [0], where we try to
> copy 16 bytes to sockaddr_ll.sll_addr[8].
> 
> Some devices (ip6gre, vti6, ip6tnl) have 16 bytes address expressed
> by struct in6_addr.
> 
> The write seems to overflow, but actually not since we use struct
> sockaddr_storage defined in __sys_getsockname().
> 
> To avoid the warning, we need to let __fortify_memcpy_chk() know the
> actual buffer size.
> 
> Another option would be to use strncpy() and limit the copied length
> to sizeof(sll_addr), but it will return the partial address and might
> break an application that passes sockaddr_storage to getsockname().
> 
> [0]:
> memcpy: detected field-spanning write (size 16) of single field "sll->sll_addr" at net/packet/af_packet.c:3604 (size 8)
> WARNING: CPU: 0 PID: 255 at net/packet/af_packet.c:3604 packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> Modules linked in:
> CPU: 0 PID: 255 Comm: syz-executor750 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
> Hardware name: linux,dummy-virt (DT)
> pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> lr : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> sp : ffff800089887bc0
> x29: ffff800089887bc0 x28: ffff000010f80f80 x27: 0000000000000003
> x26: dfff800000000000 x25: ffff700011310f80 x24: ffff800087d55000
> x23: dfff800000000000 x22: ffff800089887c2c x21: 0000000000000010
> x20: ffff00000de08310 x19: ffff800089887c20 x18: ffff800086ab1630
> x17: 20646c6569662065 x16: 6c676e697320666f x15: 0000000000000001
> x14: 1fffe0000d56d7ca x13: 0000000000000000 x12: 0000000000000000
> x11: 0000000000000000 x10: 0000000000000000 x9 : 3e60944c3da92b00
> x8 : 3e60944c3da92b00 x7 : 0000000000000001 x6 : 0000000000000001
> x5 : ffff8000898874f8 x4 : ffff800086ac99e0 x3 : ffff8000803f8808
> x2 : 0000000000000001 x1 : 0000000100000000 x0 : 0000000000000000
> Call trace:
>  packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
>  __sys_getsockname+0x168/0x24c net/socket.c:2042
>  __do_sys_getsockname net/socket.c:2057 [inline]
>  __se_sys_getsockname net/socket.c:2054 [inline]
>  __arm64_sys_getsockname+0x7c/0x94 net/socket.c:2054
>  __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
>  invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
>  el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
>  do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
>  el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
>  el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
>  el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591
> 
> Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
> Reported-by: syzkaller <syzkaller@googlegroups.com>
> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> ---
>  net/packet/af_packet.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index 85ff90a03b0c..5eef94a32a4f 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -3601,7 +3601,10 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
>  	if (dev) {
>  		sll->sll_hatype = dev->type;
>  		sll->sll_halen = dev->addr_len;
> -		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
> +
> +		/* Let __fortify_memcpy_chk() know the actual buffer size. */
> +		memcpy(((struct sockaddr_storage *)sll)->__data +
> +		       offsetof(struct sockaddr_ll, sll_addr), dev->dev_addr, dev->addr_len);

Sorry, this offset was wrong and needs minus
offsetof(struct sockaddr_ll, sll_family).

Will fix in v2.

pw-bot: cr


>  	} else {
>  		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
>  		sll->sll_halen = 0;
> -- 
> 2.30.2

RE: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Willem de Bruijn]

Kuniyuki Iwashima wrote:
> syzkaller found a warning in packet_getname() [0], where we try to
> copy 16 bytes to sockaddr_ll.sll_addr[8].
> 
> Some devices (ip6gre, vti6, ip6tnl) have 16 bytes address expressed
> by struct in6_addr.

Some are even larger. MAX_ADDR_LEN == 32. I think Infiniband may use
that?
 
> The write seems to overflow, but actually not since we use struct
> sockaddr_storage defined in __sys_getsockname().

Which gives _K_SS_MAXSIZE == 128, minus offsetof(struct sockaddr_ll, sll_addr).

For fun, there is another caller. getsockopt SO_PEERNAME also calls
sock->ops->getname, with a buffer hardcoded to 128. Should probably
use sizeof(sockaddr_storage) for documentation, at least.

.. and I just noticed that that was attempted, but not completed
https://lore.kernel.org/lkml/20140928135545.GA23220@type.youpi.perso.aquilenet.fr/

> 
> To avoid the warning, we need to let __fortify_memcpy_chk() know the
> actual buffer size.
> 
> Another option would be to use strncpy() and limit the copied length
> to sizeof(sll_addr), but it will return the partial address and might
> break an application that passes sockaddr_storage to getsockname().

Yeah, that would break stuff.
 
> [0]:
> memcpy: detected field-spanning write (size 16) of single field "sll->sll_addr" at net/packet/af_packet.c:3604 (size 8)
> WARNING: CPU: 0 PID: 255 at net/packet/af_packet.c:3604 packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> Modules linked in:
> CPU: 0 PID: 255 Comm: syz-executor750 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
> Hardware name: linux,dummy-virt (DT)
> pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> pc : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> lr : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> sp : ffff800089887bc0
> x29: ffff800089887bc0 x28: ffff000010f80f80 x27: 0000000000000003
> x26: dfff800000000000 x25: ffff700011310f80 x24: ffff800087d55000
> x23: dfff800000000000 x22: ffff800089887c2c x21: 0000000000000010
> x20: ffff00000de08310 x19: ffff800089887c20 x18: ffff800086ab1630
> x17: 20646c6569662065 x16: 6c676e697320666f x15: 0000000000000001
> x14: 1fffe0000d56d7ca x13: 0000000000000000 x12: 0000000000000000
> x11: 0000000000000000 x10: 0000000000000000 x9 : 3e60944c3da92b00
> x8 : 3e60944c3da92b00 x7 : 0000000000000001 x6 : 0000000000000001
> x5 : ffff8000898874f8 x4 : ffff800086ac99e0 x3 : ffff8000803f8808
> x2 : 0000000000000001 x1 : 0000000100000000 x0 : 0000000000000000
> Call trace:
>  packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
>  __sys_getsockname+0x168/0x24c net/socket.c:2042
>  __do_sys_getsockname net/socket.c:2057 [inline]
>  __se_sys_getsockname net/socket.c:2054 [inline]
>  __arm64_sys_getsockname+0x7c/0x94 net/socket.c:2054
>  __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
>  invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
>  el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
>  do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
>  el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
>  el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
>  el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591
> 
> Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
> Reported-by: syzkaller <syzkaller@googlegroups.com>
> Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> ---
>  net/packet/af_packet.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index 85ff90a03b0c..5eef94a32a4f 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -3601,7 +3601,10 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
>  	if (dev) {
>  		sll->sll_hatype = dev->type;
>  		sll->sll_halen = dev->addr_len;
> -		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
> +
> +		/* Let __fortify_memcpy_chk() know the actual buffer size. */
> +		memcpy(((struct sockaddr_storage *)sll)->__data +
> +		       offsetof(struct sockaddr_ll, sll_addr), dev->dev_addr, dev->addr_len);
>  	} else {
>  		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
>  		sll->sll_halen = 0;
> -- 
> 2.30.2
> 

RE: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Kuniyuki Iwashima]

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date: Wed, 19 Jul 2023 17:15:09 -0400
> Kuniyuki Iwashima wrote:
> > syzkaller found a warning in packet_getname() [0], where we try to
> > copy 16 bytes to sockaddr_ll.sll_addr[8].
> > 
> > Some devices (ip6gre, vti6, ip6tnl) have 16 bytes address expressed
> > by struct in6_addr.
> 
> Some are even larger. MAX_ADDR_LEN == 32. I think Infiniband may use
> that?

Exactly, I didn't know that.


>  
> > The write seems to overflow, but actually not since we use struct
> > sockaddr_storage defined in __sys_getsockname().
> 
> Which gives _K_SS_MAXSIZE == 128, minus offsetof(struct sockaddr_ll, sll_addr).
> 
> For fun, there is another caller. getsockopt SO_PEERNAME also calls
> sock->ops->getname, with a buffer hardcoded to 128. Should probably
> use sizeof(sockaddr_storage) for documentation, at least.
> 
> .. and I just noticed that that was attempted, but not completed
> https://lore.kernel.org/lkml/20140928135545.GA23220@type.youpi.perso.aquilenet.fr/

Yes, acutally my first draft had the diff below, but I dropped it
because packet_getname() does not call memcpy() for SO_PEERNAME at
least, and same for getpeername().

And interestingly there was a revival thread.
https://lore.kernel.org/netdev/20230719084415.1378696-1-leitao@debian.org/

I can include this in v2 if needed.
What do you think ?

---8<---
diff --git a/net/core/sock.c b/net/core/sock.c
index 9370fd50aa2c..f1e887c3115f 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1815,14 +1815,14 @@ int sk_getsockopt(struct sock *sk, int level, int optname,
 
 	case SO_PEERNAME:
 	{
-		char address[128];
+		struct sockaddr_storage address;
 
-		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
+		lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2);
 		if (lv < 0)
 			return -ENOTCONN;
 		if (lv < len)
 			return -EINVAL;
-		if (copy_to_sockptr(optval, address, len))
+		if (copy_to_sockptr(optval, &address, len))
 			return -EFAULT;
 		goto lenout;
 	}
---8<---

Thanks!


> 
> > 
> > To avoid the warning, we need to let __fortify_memcpy_chk() know the
> > actual buffer size.
> > 
> > Another option would be to use strncpy() and limit the copied length
> > to sizeof(sll_addr), but it will return the partial address and might
> > break an application that passes sockaddr_storage to getsockname().
> 
> Yeah, that would break stuff.
>  
> > [0]:
> > memcpy: detected field-spanning write (size 16) of single field "sll->sll_addr" at net/packet/af_packet.c:3604 (size 8)
> > WARNING: CPU: 0 PID: 255 at net/packet/af_packet.c:3604 packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> > Modules linked in:
> > CPU: 0 PID: 255 Comm: syz-executor750 Not tainted 6.5.0-rc1-00330-g60cc1f7d0605 #4
> > Hardware name: linux,dummy-virt (DT)
> > pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
> > pc : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> > lr : packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> > sp : ffff800089887bc0
> > x29: ffff800089887bc0 x28: ffff000010f80f80 x27: 0000000000000003
> > x26: dfff800000000000 x25: ffff700011310f80 x24: ffff800087d55000
> > x23: dfff800000000000 x22: ffff800089887c2c x21: 0000000000000010
> > x20: ffff00000de08310 x19: ffff800089887c20 x18: ffff800086ab1630
> > x17: 20646c6569662065 x16: 6c676e697320666f x15: 0000000000000001
> > x14: 1fffe0000d56d7ca x13: 0000000000000000 x12: 0000000000000000
> > x11: 0000000000000000 x10: 0000000000000000 x9 : 3e60944c3da92b00
> > x8 : 3e60944c3da92b00 x7 : 0000000000000001 x6 : 0000000000000001
> > x5 : ffff8000898874f8 x4 : ffff800086ac99e0 x3 : ffff8000803f8808
> > x2 : 0000000000000001 x1 : 0000000100000000 x0 : 0000000000000000
> > Call trace:
> >  packet_getname+0x25c/0x3a0 net/packet/af_packet.c:3604
> >  __sys_getsockname+0x168/0x24c net/socket.c:2042
> >  __do_sys_getsockname net/socket.c:2057 [inline]
> >  __se_sys_getsockname net/socket.c:2054 [inline]
> >  __arm64_sys_getsockname+0x7c/0x94 net/socket.c:2054
> >  __invoke_syscall arch/arm64/kernel/syscall.c:38 [inline]
> >  invoke_syscall+0x98/0x2c0 arch/arm64/kernel/syscall.c:52
> >  el0_svc_common+0x134/0x240 arch/arm64/kernel/syscall.c:139
> >  do_el0_svc+0x64/0x198 arch/arm64/kernel/syscall.c:188
> >  el0_svc+0x2c/0x7c arch/arm64/kernel/entry-common.c:647
> >  el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:665
> >  el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:591
> > 
> > Fixes: df8fc4e934c1 ("kbuild: Enable -fstrict-flex-arrays=3")
> > Reported-by: syzkaller <syzkaller@googlegroups.com>
> > Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> > ---
> >  net/packet/af_packet.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> > index 85ff90a03b0c..5eef94a32a4f 100644
> > --- a/net/packet/af_packet.c
> > +++ b/net/packet/af_packet.c
> > @@ -3601,7 +3601,10 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
> >  	if (dev) {
> >  		sll->sll_hatype = dev->type;
> >  		sll->sll_halen = dev->addr_len;
> > -		memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
> > +
> > +		/* Let __fortify_memcpy_chk() know the actual buffer size. */
> > +		memcpy(((struct sockaddr_storage *)sll)->__data +
> > +		       offsetof(struct sockaddr_ll, sll_addr), dev->dev_addr, dev->addr_len);
> >  	} else {
> >  		sll->sll_hatype = 0;	/* Bad: we have no ARPHRD_UNSPEC */
> >  		sll->sll_halen = 0;
> > -- 
> > 2.30.2

RE: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Willem de Bruijn]

> >  
> > > The write seems to overflow, but actually not since we use struct
> > > sockaddr_storage defined in __sys_getsockname().
> > 
> > Which gives _K_SS_MAXSIZE == 128, minus offsetof(struct sockaddr_ll, sll_addr).
> > 
> > For fun, there is another caller. getsockopt SO_PEERNAME also calls
> > sock->ops->getname, with a buffer hardcoded to 128. Should probably
> > use sizeof(sockaddr_storage) for documentation, at least.
> > 
> > .. and I just noticed that that was attempted, but not completed
> > https://lore.kernel.org/lkml/20140928135545.GA23220@type.youpi.perso.aquilenet.fr/
> 
> Yes, acutally my first draft had the diff below, but I dropped it
> because packet_getname() does not call memcpy() for SO_PEERNAME at
> least, and same for getpeername().
> 
> And interestingly there was a revival thread.
> https://lore.kernel.org/netdev/20230719084415.1378696-1-leitao@debian.org/

Ah interesting :) Topical.

> I can include this in v2 if needed.
> What do you think ?
> 
> ---8<---
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 9370fd50aa2c..f1e887c3115f 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -1815,14 +1815,14 @@ int sk_getsockopt(struct sock *sk, int level, int optname,
>  
>  	case SO_PEERNAME:
>  	{
> -		char address[128];
> +		struct sockaddr_storage address;
>  
> -		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
> +		lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2);
>  		if (lv < 0)
>  			return -ENOTCONN;
>  		if (lv < len)
>  			return -EINVAL;
> -		if (copy_to_sockptr(optval, address, len))
> +		if (copy_to_sockptr(optval, &address, len))
>  			return -EFAULT;
>  		goto lenout;
>  	}
> ---8<---

I agree that it's a worthwhile change. I think it should be an
independent commit. And since it does not fix a bug, target net-next.

RE: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[David Laight]

From: Willem de Bruijn
> Sent: 19 July 2023 22:34
> 
> > >
> > > > The write seems to overflow, but actually not since we use struct
> > > > sockaddr_storage defined in __sys_getsockname().
> > >
> > > Which gives _K_SS_MAXSIZE == 128, minus offsetof(struct sockaddr_ll, sll_addr).
> > >
> > > For fun, there is another caller. getsockopt SO_PEERNAME also calls
> > > sock->ops->getname, with a buffer hardcoded to 128. Should probably
> > > use sizeof(sockaddr_storage) for documentation, at least.
> > >
> > > .. and I just noticed that that was attempted, but not completed
> > > https://lore.kernel.org/lkml/20140928135545.GA23220@type.youpi.perso.aquilenet.fr/
> >
> > Yes, acutally my first draft had the diff below, but I dropped it
> > because packet_getname() does not call memcpy() for SO_PEERNAME at
> > least, and same for getpeername().
> >
> > And interestingly there was a revival thread.
> > https://lore.kernel.org/netdev/20230719084415.1378696-1-leitao@debian.org/
> 
> Ah interesting :) Topical.
> 
> > I can include this in v2 if needed.
> > What do you think ?
> >
> > ---8<---
> > diff --git a/net/core/sock.c b/net/core/sock.c
> > index 9370fd50aa2c..f1e887c3115f 100644
> > --- a/net/core/sock.c
> > +++ b/net/core/sock.c
> > @@ -1815,14 +1815,14 @@ int sk_getsockopt(struct sock *sk, int level, int optname,
> >
> >  	case SO_PEERNAME:
> >  	{
> > -		char address[128];
> > +		struct sockaddr_storage address;
> >
> > -		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
> > +		lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2);
> >  		if (lv < 0)
> >  			return -ENOTCONN;
> >  		if (lv < len)
> >  			return -EINVAL;
> > -		if (copy_to_sockptr(optval, address, len))
> > +		if (copy_to_sockptr(optval, &address, len))
> >  			return -EFAULT;
> >  		goto lenout;
> >  	}
> > ---8<---
> 
> I agree that it's a worthwhile change. I think it should be an
> independent commit. And since it does not fix a bug, target net-next.

It is potentially a bug.
There is no requirement that the compiler align 'char address[128]'.
So the accesses could fault later on.

In practise it will be aligned - unless the compiler is being
perverse.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

RE: [PATCH v1 net 2/2] af_packet: Fix warning of fortified memcpy() in packet_getname().

[Kuniyuki Iwashima]

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date: Wed, 19 Jul 2023 17:34:13 -0400
> > > > The write seems to overflow, but actually not since we use struct
> > > > sockaddr_storage defined in __sys_getsockname().
> > > 
> > > Which gives _K_SS_MAXSIZE == 128, minus offsetof(struct sockaddr_ll, sll_addr).
> > > 
> > > For fun, there is another caller. getsockopt SO_PEERNAME also calls
> > > sock->ops->getname, with a buffer hardcoded to 128. Should probably
> > > use sizeof(sockaddr_storage) for documentation, at least.
> > > 
> > > .. and I just noticed that that was attempted, but not completed
> > > https://lore.kernel.org/lkml/20140928135545.GA23220@type.youpi.perso.aquilenet.fr/
> > 
> > Yes, acutally my first draft had the diff below, but I dropped it
> > because packet_getname() does not call memcpy() for SO_PEERNAME at
> > least, and same for getpeername().
> > 
> > And interestingly there was a revival thread.
> > https://lore.kernel.org/netdev/20230719084415.1378696-1-leitao@debian.org/
> 
> Ah interesting :) Topical.
> 
> > I can include this in v2 if needed.
> > What do you think ?
> > 
> > ---8<---
> > diff --git a/net/core/sock.c b/net/core/sock.c
> > index 9370fd50aa2c..f1e887c3115f 100644
> > --- a/net/core/sock.c
> > +++ b/net/core/sock.c
> > @@ -1815,14 +1815,14 @@ int sk_getsockopt(struct sock *sk, int level, int optname,
> >  
> >  	case SO_PEERNAME:
> >  	{
> > -		char address[128];
> > +		struct sockaddr_storage address;
> >  
> > -		lv = sock->ops->getname(sock, (struct sockaddr *)address, 2);
> > +		lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2);
> >  		if (lv < 0)
> >  			return -ENOTCONN;
> >  		if (lv < len)
> >  			return -EINVAL;
> > -		if (copy_to_sockptr(optval, address, len))
> > +		if (copy_to_sockptr(optval, &address, len))
> >  			return -EFAULT;
> >  		goto lenout;
> >  	}
> > ---8<---
> 
> I agree that it's a worthwhile change. I think it should be an
> independent commit. And since it does not fix a bug, target net-next.

Sure, will post a patch to net-next later.