Re: BUG: unable to handle page fault for address, with ipv6.disable=1

From: Ido Schimmel
Date: Wed Sep 14 2022 - 11:47:28 EST

+ Eric

Original report:
https://lore.kernel.org/netdev/YyD0kMC7qIBNOE3j@riccipc/T/#u

On Tue, Sep 13, 2022 at 11:22:24PM +0200, Roberto Ricci wrote:
> Executing the `ss` command in a system with kernel 5.19.8, booted with
> the "ipv6.disable=1" parameter, causes this oops:
>
>
> [ 74.952477] BUG: unable to handle page fault for address: ffffffffffffffc8
> [ 74.952568] #PF: supervisor read access in kernel mode
> [ 74.952632] #PF: error_code(0x0000) - not-present page
> [ 74.952695] PGD 25814067 P4D 25814067 PUD 25816067 PMD 0
> [ 74.952770] Oops: 0000 [#1] PREEMPT SMP PTI
> [ 74.952816] CPU: 0 PID: 704 Comm: ss Not tainted 5.19.8_1 #1
> [ 74.952869] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
> WARNING! Modules path isn't set, but is needed to parse this symbol
> [ 74.953292] RIP: 0010:raw_diag_dump+0xea/0x1d0 raw_diag
[...]
> [ 74.954188] Call Trace:
> [ 74.954221] <TASK>
> [ 74.954248] __inet_diag_dump (net/ipv4/inet_diag.c:1179)
> [ 74.954462] netlink_dump (net/netlink/af_netlink.c:2276)
> [ 74.954549] __netlink_dump_start (net/netlink/af_netlink.c:2380)
> [ 74.954613] inet_diag_handler_cmd (net/ipv4/inet_diag.c:1347)
> [ 74.954672] ? inet_diag_dump_start_compat (net/ipv4/inet_diag.c:1244)
> [ 74.954725] ? inet_diag_dump_compat (net/ipv4/inet_diag.c:1197)
> [ 74.954768] ? inet_diag_unregister (net/ipv4/inet_diag.c:1254)
> [ 74.954811] sock_diag_rcv_msg (net/core/sock_diag.c:235 net/core/sock_diag.c:266)
> [ 74.954905] ? sock_diag_bind (net/core/sock_diag.c:247)
> [ 74.954950] netlink_rcv_skb (net/netlink/af_netlink.c:2501)
> [ 74.954993] sock_diag_rcv (net/core/sock_diag.c:278)
> [ 74.955032] netlink_unicast (net/netlink/af_netlink.c:1320 net/netlink/af_netlink.c:1345)
> [ 74.955074] netlink_sendmsg (net/netlink/af_netlink.c:1921)
> [ 74.955116] sock_sendmsg (net/socket.c:714 net/socket.c:734)
> [ 74.955199] ____sys_sendmsg (net/socket.c:2488)
> [ 74.955245] ? import_iovec (lib/iov_iter.c:2008)
> [ 74.955302] ? sendmsg_copy_msghdr (net/socket.c:2429 net/socket.c:2519)
> [ 74.955348] ___sys_sendmsg (net/socket.c:2544)
> [ 74.955447] ? __schedule (kernel/sched/core.c:6476)
> [ 74.955522] ? _raw_spin_unlock_irqrestore (./arch/x86/include/asm/preempt.h:103 ./include/linux/spinlock_api_smp.h:152 kernel/locking/spinlock.c:194)
> [ 74.955583] ? do_notify_parent_cldstop (kernel/signal.c:2191)
> [ 74.955656] ? preempt_count_add (./include/linux/ftrace.h:910 kernel/sched/core.c:5598 kernel/sched/core.c:5595 kernel/sched/core.c:5623)
> [ 74.955712] ? _raw_spin_lock_irq (./arch/x86/include/asm/atomic.h:202 ./include/linux/atomic/atomic-instrumented.h:543 ./include/asm-generic/qspinlock.h:111 ./include/linux/spinlock.h:185 ./include/linux/spinlock_api_smp.h:120 kernel/locking/spinlock.c:170)
> [ 74.955752] ? ptrace_stop.part.0 (kernel/signal.c:2331)
> [ 74.955795] __sys_sendmsg (./include/linux/file.h:31 net/socket.c:2573)
> [ 74.955835] do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
> [ 74.955914] ? syscall_exit_to_user_mode (./arch/x86/include/asm/jump_label.h:55 ./arch/x86/include/asm/nospec-branch.h:382 ./arch/x86/include/asm/entry-common.h:94 kernel/entry/common.c:133 kernel/entry/common.c:296)
> [ 74.955965] ? do_syscall_64 (arch/x86/entry/common.c:87)
> [ 74.957786] ? do_syscall_64 (arch/x86/entry/common.c:87)
> [ 74.959896] ? handle_mm_fault (mm/memory.c:5144)
> [ 74.961184] ? do_user_addr_fault (arch/x86/mm/fault.c:1422)
> [ 74.962609] ? fpregs_assert_state_consistent (arch/x86/kernel/fpu/context.h:39 arch/x86/kernel/fpu/core.c:772)
> [ 74.964171] ? exit_to_user_mode_prepare (./arch/x86/include/asm/entry-common.h:57 kernel/entry/common.c:203)
> [ 74.965968] entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)
> [ 74.967266] RIP: 0033:0x7f66aac577d3

[...]

> I reproduced this with Void Linux x86_64 in a virtual machine. The kernels are
> those provided by the distribution (Void uses vanilla kernels, I don't believe
> these very small patches make any difference
> https://github.com/void-linux/void-packages/tree/0a87c670f35e01a3ac1d850f628fe1bab5d3c433/srcpkgs/linux5.19/patches).
>
> Kernels 5.19.8 and 5.18.19 are affected, 5.16.20 is not.
> I don't know about 5.17.x because Void doesn't package it.
> The iproute2 version is 5.16.0 (but this also happens with 5.19.0).

This is most likely caused by commit 0daf07e52709 ("raw: convert raw
sockets to RCU") which is being back ported to stable kernels.

It made the initialization of 'raw_v6_hashinfo' conditional on IPv6
being enabled. Can you try the following patch (works on my end)?

diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index 19732b5dce23..d40b7d60e00e 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -1072,13 +1072,13 @@ static int __init inet6_init(void)
for (r = &inetsw6[0]; r < &inetsw6[SOCK_MAX]; ++r)
INIT_LIST_HEAD(r);

+ raw_hashinfo_init(&raw_v6_hashinfo);
+
if (disable_ipv6_mod) {
pr_info("Loaded, but administratively disabled, reboot required to enable\n");
goto out;
}

- raw_hashinfo_init(&raw_v6_hashinfo);
-
err = proto_register(&tcpv6_prot, 1);
if (err)
goto out;

Another approach is the following, but I prefer the first:

diff --git a/net/ipv4/raw_diag.c b/net/ipv4/raw_diag.c
index 999321834b94..4fbdd69a2be8 100644
--- a/net/ipv4/raw_diag.c
+++ b/net/ipv4/raw_diag.c
@@ -20,7 +20,7 @@ raw_get_hashinfo(const struct inet_diag_req_v2 *r)
if (r->sdiag_family == AF_INET) {
return &raw_v4_hashinfo;
#if IS_ENABLED(CONFIG_IPV6)
- } else if (r->sdiag_family == AF_INET6) {
+ } else if (r->sdiag_family == AF_INET6 && ipv6_mod_enabled()) {
return &raw_v6_hashinfo;
#endif
} else {