net: initialize msg.msg_flags in recvfrom

KMSAN (KernelMemorySanitizer, a new error detection tool) reports use
of uninitialized memory in put_cmsg()):

On Tue, Mar 7, 2017 at 2:58 PM, Alexander Potapenko <glider@google.com> wrote:
> KMSAN (KernelMemorySanitizer, a new error detection tool) reports use
> of uninitialized memory in put_cmsg()):
>
> ==================================================================
> BUG: KMSAN: use of unitialized memory
> inter: 0
> CPU: 3 PID: 1086 Comm: syz-executor0 Not tainted 4.8.0-rc6+ #1920
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
>  0000000000000000 ffff88008374fa48 ffffffff82576c18 ffff88008374f988
>  ffffffff818c06d1 ffffffff85bab870 0000000000000092 ffffffff85bab550
>  0000000000000000 0000000000000092 00000000f9600a55 0000000000000002
> Call Trace:
>  [<     inline     >] __dump_stack lib/dump_stack.c:15
>  [<ffffffff82576c18>] dump_stack+0x238/0x290 lib/dump_stack.c:51
>  [<ffffffff818beef6>] kmsan_report+0x276/0x2e0 mm/kmsan/kmsan.c:1006
>  [<ffffffff818c027b>] __msan_warning+0x6b/0xc0 mm/kmsan/kmsan_instr.c:425
>  [<ffffffff842f01e7>] put_cmsg+0x6f7/0x850 net/core/scm.c:223
>  [<     inline     >] netlink_cmsg_recv_pktinfo net/netlink/af_netlink.c:1700
>  [<ffffffff844b7054>] netlink_recvmsg+0x8e4/0x1980 net/netlink/af_netlink.c:1856
>  [<     inline     >] sock_recvmsg_nosec net/socket.c:711
>  [<     inline     >] sock_recvmsg net/socket.c:718
>  [<ffffffff8426c175>] SYSC_recvfrom+0x6e5/0xa00 net/socket.c:1695
>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
> chained origin: 00000000f8c00a55
>  [<ffffffff810bd4d7>] save_stack_trace+0x27/0x50 arch/x86/kernel/stacktrace.c:67
>  [<     inline     >] kmsan_save_stack_with_flags mm/kmsan/kmsan.c:322
>  [<     inline     >] kmsan_save_stack mm/kmsan/kmsan.c:337
>  [<ffffffff818be2cb>] kmsan_internal_chain_origin+0x12b/0x1f0 mm/kmsan/kmsan.c:530
>  [<ffffffff818bfc0f>] __kmsan_store_origin+0x9f/0x120 mm/kmsan/kmsan_instr.c:190
>  [<ffffffff844b6bdd>] netlink_recvmsg+0x46d/0x1980 net/netlink/af_netlink.c:1839
>  [<     inline     >] sock_recvmsg_nosec net/socket.c:711
>  [<     inline     >] sock_recvmsg net/socket.c:718
>  [<ffffffff8426c175>] SYSC_recvfrom+0x6e5/0xa00 net/socket.c:1695
>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
> chained origin: 00000000b9a00a53
>  [<ffffffff810bd4d7>] save_stack_trace+0x27/0x50 arch/x86/kernel/stacktrace.c:67
>  [<     inline     >] kmsan_save_stack_with_flags mm/kmsan/kmsan.c:322
>  [<     inline     >] kmsan_save_stack mm/kmsan/kmsan.c:337
>  [<ffffffff818be2cb>] kmsan_internal_chain_origin+0x12b/0x1f0 mm/kmsan/kmsan.c:530
>  [<ffffffff818c01a3>] __msan_set_alloca_origin4+0xc3/0x130 mm/kmsan/kmsan_instr.c:380
>  [<ffffffff8426bc13>] SYSC_recvfrom+0x183/0xa00 net/socket.c:1671
>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
> origin description: ----msg@SYSC_recvfrom (origin=00000000b9400a53)
> ==================================================================
>
> The fields of |msg| in SYSC_recvfrom() are initialized except
> for |msg.msg_flags|, which is later updated in netlink_recvmsg():
>
>   1839                 msg->msg_flags |= MSG_TRUNC;
>
> and then used in a condition in put_cmsg():
>
>   223         if (MSG_CMSG_COMPAT & msg->msg_flags)
>
> The following program has been used to reproduce the error under KMSAN:
>
> ===================================================
>   #include <linux/netlink.h>
>   #include <stdint.h>
>   #include <stdio.h>
>   #include <sys/socket.h>
>   #include <sys/wait.h>
>
>   void child()
>   {
>     int sock = socket(PF_NETLINK, SOCK_RAW, 16);
>     uint32_t val = 0x1f;
>     setsockopt(sock, 0x10e, 0x3, &val, 4);
>     char obuf[31] = "\x1f\x00\x00\x00\x10\x00\x07\xff\xff\x08\x00\x00"
>                     "\x00\x00\x00\x00\x03\x00\x01\x0a\x56\x67\x69\x93"
>                     "\x3e\x00\x00\x01\x92\x9a\x02";
>     char ibuf[31];
>     write(sock, obuf, sizeof(obuf));
>     struct sockaddr_nl sockaddr = {AF_NETLINK, 0, 0, 0};
>     socklen_t addrlen = sizeof(sockaddr);
>     recvfrom(sock, ibuf, 0, 0, &sockaddr, &addrlen);
>   }
>
>   int main()
>   {
>     int pid = fork();
>     if (!pid) {
>       child();
>       return 0;
>     }
>     int status = 0;
>     while (waitpid(pid, &status, __WALL) != pid) {}
>     return 0;
>   }
> ===================================================
>
> Signed-off-by: Alexander Potapenko <glider@google.com>
> ---
>  net/socket.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/net/socket.c b/net/socket.c
> index 2c1e8677ff2d..e0757e648c0c 100644
> --- a/net/socket.c
> +++ b/net/socket.c
> @@ -1731,6 +1731,7 @@ SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
>         /* We assume all kernel code knows the size of sockaddr_storage */
>         msg.msg_namelen = 0;
>         msg.msg_iocb = NULL;
> +       msg.msg_flags = 0;
>         if (sock->file->f_flags & O_NONBLOCK)
>                 flags |= MSG_DONTWAIT;
>         err = sock_recvmsg(sock, &msg, flags);
> --
> 2.12.0.rc1.440.g5b76565f74-goog
>


I am not sure this is correct. The compat syscall does:

COMPAT_SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, buf,
compat_size_t, len,
      unsigned int, flags, struct sockaddr __user *, addr,
      int __user *, addrlen)
{
   return sys_recvfrom(fd, buf, len, flags | MSG_CMSG_COMPAT, addr, addrlen);
}

netlink_cmsg_recv_pktinfo expects that MSG_CMSG_COMPAT in msg.flags,
but we put it into flags.

On Tue, 2017-03-07 at 14:58 +0100, Alexander Potapenko wrote:
> KMSAN (KernelMemorySanitizer, a new error detection tool) reports use
> of uninitialized memory in put_cmsg()):

I would prefer that you do not put the stack trace in the changelog,
same for the reproducer since this has little value in understanding the
impact.

It looks like a false positive, but you do not say.

recvmsg() does not care about msg.msg_flags, only KMSAN.

(The important part is that msg.msg_control and msg.msg_controllen are
0)

Fine to avoid the false positive, but better be explicit in the
changelog and says there is no visible effect for this bug.

If there is a visible effect, please state so instead of technical
details.

We try to reduce S/N in the changelogs ;)

Thanks a lot !

On Tue, Mar 7, 2017 at 3:23 PM, Dmitry Vyukov <dvyukov@google.com> wrote:
> On Tue, Mar 7, 2017 at 2:58 PM, Alexander Potapenko <glider@google.com> wrote:
>> KMSAN (KernelMemorySanitizer, a new error detection tool) reports use
>> of uninitialized memory in put_cmsg()):
>>
>> ==================================================================
>> BUG: KMSAN: use of unitialized memory
>> inter: 0
>> CPU: 3 PID: 1086 Comm: syz-executor0 Not tainted 4.8.0-rc6+ #1920
>> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
>>  0000000000000000 ffff88008374fa48 ffffffff82576c18 ffff88008374f988
>>  ffffffff818c06d1 ffffffff85bab870 0000000000000092 ffffffff85bab550
>>  0000000000000000 0000000000000092 00000000f9600a55 0000000000000002
>> Call Trace:
>>  [<     inline     >] __dump_stack lib/dump_stack.c:15
>>  [<ffffffff82576c18>] dump_stack+0x238/0x290 lib/dump_stack.c:51
>>  [<ffffffff818beef6>] kmsan_report+0x276/0x2e0 mm/kmsan/kmsan.c:1006
>>  [<ffffffff818c027b>] __msan_warning+0x6b/0xc0 mm/kmsan/kmsan_instr.c:425
>>  [<ffffffff842f01e7>] put_cmsg+0x6f7/0x850 net/core/scm.c:223
>>  [<     inline     >] netlink_cmsg_recv_pktinfo net/netlink/af_netlink.c:1700
>>  [<ffffffff844b7054>] netlink_recvmsg+0x8e4/0x1980 net/netlink/af_netlink.c:1856
>>  [<     inline     >] sock_recvmsg_nosec net/socket.c:711
>>  [<     inline     >] sock_recvmsg net/socket.c:718
>>  [<ffffffff8426c175>] SYSC_recvfrom+0x6e5/0xa00 net/socket.c:1695
>>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
>> chained origin: 00000000f8c00a55
>>  [<ffffffff810bd4d7>] save_stack_trace+0x27/0x50 arch/x86/kernel/stacktrace.c:67
>>  [<     inline     >] kmsan_save_stack_with_flags mm/kmsan/kmsan.c:322
>>  [<     inline     >] kmsan_save_stack mm/kmsan/kmsan.c:337
>>  [<ffffffff818be2cb>] kmsan_internal_chain_origin+0x12b/0x1f0 mm/kmsan/kmsan.c:530
>>  [<ffffffff818bfc0f>] __kmsan_store_origin+0x9f/0x120 mm/kmsan/kmsan_instr.c:190
>>  [<ffffffff844b6bdd>] netlink_recvmsg+0x46d/0x1980 net/netlink/af_netlink.c:1839
>>  [<     inline     >] sock_recvmsg_nosec net/socket.c:711
>>  [<     inline     >] sock_recvmsg net/socket.c:718
>>  [<ffffffff8426c175>] SYSC_recvfrom+0x6e5/0xa00 net/socket.c:1695
>>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
>> chained origin: 00000000b9a00a53
>>  [<ffffffff810bd4d7>] save_stack_trace+0x27/0x50 arch/x86/kernel/stacktrace.c:67
>>  [<     inline     >] kmsan_save_stack_with_flags mm/kmsan/kmsan.c:322
>>  [<     inline     >] kmsan_save_stack mm/kmsan/kmsan.c:337
>>  [<ffffffff818be2cb>] kmsan_internal_chain_origin+0x12b/0x1f0 mm/kmsan/kmsan.c:530
>>  [<ffffffff818c01a3>] __msan_set_alloca_origin4+0xc3/0x130 mm/kmsan/kmsan_instr.c:380
>>  [<ffffffff8426bc13>] SYSC_recvfrom+0x183/0xa00 net/socket.c:1671
>>  [<ffffffff8426ba67>] SyS_recvfrom+0xb7/0xe0 net/socket.c:1668
>>  [<ffffffff851840db>] entry_SYSCALL_64_fastpath+0x13/0x8f arch/x86/entry/entry_64.o:?
>> origin description: ----msg@SYSC_recvfrom (origin=00000000b9400a53)
>> ==================================================================
>>
>> The fields of |msg| in SYSC_recvfrom() are initialized except
>> for |msg.msg_flags|, which is later updated in netlink_recvmsg():
>>
>>   1839                 msg->msg_flags |= MSG_TRUNC;
>>
>> and then used in a condition in put_cmsg():
>>
>>   223         if (MSG_CMSG_COMPAT & msg->msg_flags)
>>
>> The following program has been used to reproduce the error under KMSAN:
>>
>> ===================================================
>>   #include <linux/netlink.h>
>>   #include <stdint.h>
>>   #include <stdio.h>
>>   #include <sys/socket.h>
>>   #include <sys/wait.h>
>>
>>   void child()
>>   {
>>     int sock = socket(PF_NETLINK, SOCK_RAW, 16);
>>     uint32_t val = 0x1f;
>>     setsockopt(sock, 0x10e, 0x3, &val, 4);
>>     char obuf[31] = "\x1f\x00\x00\x00\x10\x00\x07\xff\xff\x08\x00\x00"
>>                     "\x00\x00\x00\x00\x03\x00\x01\x0a\x56\x67\x69\x93"
>>                     "\x3e\x00\x00\x01\x92\x9a\x02";
>>     char ibuf[31];
>>     write(sock, obuf, sizeof(obuf));
>>     struct sockaddr_nl sockaddr = {AF_NETLINK, 0, 0, 0};
>>     socklen_t addrlen = sizeof(sockaddr);
>>     recvfrom(sock, ibuf, 0, 0, &sockaddr, &addrlen);
>>   }
>>
>>   int main()
>>   {
>>     int pid = fork();
>>     if (!pid) {
>>       child();
>>       return 0;
>>     }
>>     int status = 0;
>>     while (waitpid(pid, &status, __WALL) != pid) {}
>>     return 0;
>>   }
>> ===================================================
>>
>> Signed-off-by: Alexander Potapenko <glider@google.com>
>> ---
>>  net/socket.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/net/socket.c b/net/socket.c
>> index 2c1e8677ff2d..e0757e648c0c 100644
>> --- a/net/socket.c
>> +++ b/net/socket.c
>> @@ -1731,6 +1731,7 @@ SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, ubuf, size_t, size,
>>         /* We assume all kernel code knows the size of sockaddr_storage */
>>         msg.msg_namelen = 0;
>>         msg.msg_iocb = NULL;
>> +       msg.msg_flags = 0;
>>         if (sock->file->f_flags & O_NONBLOCK)
>>                 flags |= MSG_DONTWAIT;
>>         err = sock_recvmsg(sock, &msg, flags);
>> --
>> 2.12.0.rc1.440.g5b76565f74-goog
>>
>
>
> I am not sure this is correct. The compat syscall does:
I've changed the initial value of msg.msg_flags to 0xdeadbeef and
added the following line to put_cmsg():
  pr_err("msg->msg_flags: %p\n", msg->msg_flags);
When I run the above repro, I see the following lines in the log:
  msg->msg_flags: 00000000deadbeef

Therefore I believe the bug is real.

> COMPAT_SYSCALL_DEFINE6(recvfrom, int, fd, void __user *, buf,
How is the compat syscall related to that?
> compat_size_t, len,
>       unsigned int, flags, struct sockaddr __user *, addr,
>       int __user *, addrlen)
> {
>    return sys_recvfrom(fd, buf, len, flags | MSG_CMSG_COMPAT, addr, addrlen);
> }
>
> netlink_cmsg_recv_pktinfo expects that MSG_CMSG_COMPAT in msg.flags,
> but we put it into flags.

On Tue, Mar 7, 2017 at 3:26 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Tue, 2017-03-07 at 14:58 +0100, Alexander Potapenko wrote:
>> KMSAN (KernelMemorySanitizer, a new error detection tool) reports use
>> of uninitialized memory in put_cmsg()):
>
> I would prefer that you do not put the stack trace in the changelog,
> same for the reproducer since this has little value in understanding the
> impact.
Understood.
Should be ok to put the report/reproducer below the triple dash, right?

> It looks like a false positive, but you do not say.
Ah, now I see.
Irrespective of the value of (MSG_CMSG_COMPAT & msg->msg_flags) the
code will return 0 either directly from put_cmsg(), or from
put_cmsg_compat().
I wouldn't call this a false positive, as KMSAN can't possibly know
that both branches taken depending on the uninitialized condition are
safe.
But I can imagine this to be less of an problem to the code owners who
do know that :)

> recvmsg() does not care about msg.msg_flags, only KMSAN.
>
> (The important part is that msg.msg_control and msg.msg_controllen are
> 0)
>
> Fine to avoid the false positive, but better be explicit in the
> changelog and says there is no visible effect for this bug.
Ok, I'll change the description.
> If there is a visible effect, please state so instead of technical
> details.
>
> We try to reduce S/N in the changelogs ;)
>
> Thanks a lot !
>
>