Hello,
Since 0eeb075fad736 ('net: ipv4 sysctl option to ignore routes when nexthop
link is down') fib_dump_info() and fib_sync_up() are using __in_dev_get_rcu(),
which requires (missing) RCU read side protection.
[ 60.605034] ===============================
[ 60.605035] [ INFO: suspicious RCU usage. ]
[ 60.605037] 4.1.0-next-20150626-dbg-00020-g54a6d91-dirty #244 Not tainted
[ 60.605038] -------------------------------
[ 60.605040] include/linux/inetdevice.h:205 suspicious rcu_dereference_check() usage!
[ 60.605041]
other info that might help us debug this:
[ 60.605043]
rcu_scheduler_active = 1, debug_locks = 0
[ 60.605045] 2 locks held by ip/403:
[ 60.605046] #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff81453305>] rtnl_lock+0x17/0x19
[ 60.605054] #1: ((inetaddr_chain).rwsem){.+.+.+}, at: [<ffffffff8105c327>] __blocking_notifier_call_chain+0x35/0x6a
[ 60.605060]
stack backtrace:
[ 60.605062] CPU: 2 PID: 403 Comm: ip Not tainted 4.1.0-next-20150626-dbg-00020-g54a6d91-dirty #244
[ 60.605066] 0000000000000001 ffff8800b189b728 ffffffff8150a542 ffffffff8107a8b3
[ 60.605069] ffff880037bbea40 ffff8800b189b758 ffffffff8107cb74 ffff8800379dbd00
[ 60.605071] ffff8800bec85800 ffff8800bf9e13c0 00000000000000ff ffff8800b189b7d8
[ 60.605074] Call Trace:
[ 60.605078] [<ffffffff8150a542>] dump_stack+0x4c/0x6e
[ 60.605081] [<ffffffff8107a8b3>] ? up+0x39/0x3e
[ 60.605084] [<ffffffff8107cb74>] lockdep_rcu_suspicious+0xf7/0x100
[ 60.605088] [<ffffffff814b63c3>] fib_dump_info+0x227/0x3e2
[ 60.605090] [<ffffffff814b6624>] rtmsg_fib+0xa6/0x116
[ 60.605093] [<ffffffff814b978f>] fib_table_insert+0x316/0x355
[ 60.605095] [<ffffffff814b362e>] fib_magic+0xb7/0xc7
[ 60.605098] [<ffffffff814b4803>] fib_add_ifaddr+0xb1/0x13b
[ 60.605100] [<ffffffff814b4d09>] fib_inetaddr_event+0x36/0x90
[ 60.605102] [<ffffffff8105c086>] notifier_call_chain+0x4c/0x71
[ 60.605104] [<ffffffff8105c340>] __blocking_notifier_call_chain+0x4e/0x6a
[ 60.605106] [<ffffffff8105c370>] blocking_notifier_call_chain+0x14/0x16
[ 60.605108] [<ffffffff814a7f50>] __inet_insert_ifa+0x1a5/0x1b3
[ 60.605110] [<ffffffff814a894d>] inet_rtm_newaddr+0x350/0x35f
[ 60.605113] [<ffffffff81457866>] rtnetlink_rcv_msg+0x17b/0x18a
[ 60.605115] [<ffffffff8107e7c3>] ? trace_hardirqs_on+0xd/0xf
[ 60.605118] [<ffffffff8146965f>] ? netlink_deliver_tap+0x1cb/0x1f7
[ 60.605120] [<ffffffff814576eb>] ? rtnl_newlink+0x72a/0x72a
[ 60.605122] [<ffffffff8146d299>] netlink_rcv_skb+0x48/0x8a
[ 60.605124] [<ffffffff81455724>] rtnetlink_rcv+0x26/0x2d
[ 60.605126] [<ffffffff8146cb53>] netlink_unicast+0xc6/0x145
[ 60.605128] [<ffffffff8146d144>] netlink_sendmsg+0x4c9/0x4e0
[ 60.605131] [<ffffffff81247a09>] ? import_iovec+0x67/0x89
[ 60.605134] [<ffffffff8142adaf>] sock_sendmsg+0x14/0x1f
[ 60.605136] [<ffffffff8142b38d>] ___sys_sendmsg+0x16a/0x1f1
[ 60.605140] [<ffffffff81119923>] ? handle_mm_fault+0x9dd/0xe17
[ 60.605142] [<ffffffff81119923>] ? handle_mm_fault+0x9dd/0xe17
[ 60.605144] [<ffffffff81119923>] ? handle_mm_fault+0x9dd/0xe17
[ 60.605146] [<ffffffff8107bcbc>] ? __lock_is_held+0x3c/0x57
[ 60.605149] [<ffffffff81158da8>] ? __fget_light+0x50/0x75
[ 60.605151] [<ffffffff8142c1e0>] __sys_sendmsg+0x42/0x63
[ 60.605154] [<ffffffff8142c20f>] SyS_sendmsg+0xe/0x10
[ 60.605156] [<ffffffff81510fd7>] entry_SYSCALL_64_fastpath+0x12/0x6f
-ss
On Fri, Jun 26, 2015 at 06:35:40PM +0900, Sergey Senozhatsky wrote:
> Hello,
>
> Since 0eeb075fad736 ('net: ipv4 sysctl option to ignore routes when nexthop
> link is down') fib_dump_info() and fib_sync_up() are using __in_dev_get_rcu(),
> which requires (missing) RCU read side protection.
Yep, I can reproduce this now. Thanks for reporting.
I'll take a look and should be able to post an easy fix quickly.