sch_multiq: fix double free on init failure
authorNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Wed, 30 Aug 2017 09:48:58 +0000 (12:48 +0300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 15 Sep 2018 07:43:02 +0000 (09:43 +0200)
commit e89d469e3be3ed3d7124a803211a463ff83d0964 upstream.

The below commit added a call to ->destroy() on init failure, but multiq
still frees ->queues on error in init, but ->queues is also freed by
->destroy() thus we get double free and corrupted memory.

Very easy to reproduce (eth0 not multiqueue):
$ tc qdisc add dev eth0 root multiq
RTNETLINK answers: Operation not supported
$ ip l add dumdum type dummy
(crash)

Trace log:
[ 3929.467747] general protection fault: 0000 [#1] SMP
[ 3929.468083] Modules linked in:
[ 3929.468302] CPU: 3 PID: 967 Comm: ip Not tainted 4.13.0-rc6+ #56
[ 3929.468625] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 3929.469124] task: ffff88003716a700 task.stack: ffff88005872c000
[ 3929.469449] RIP: 0010:__kmalloc_track_caller+0x117/0x1be
[ 3929.469746] RSP: 0018:ffff88005872f6a0 EFLAGS: 00010246
[ 3929.470042] RAX: 00000000000002de RBX: 0000000058a59000 RCX: 00000000000002df
[ 3929.470406] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffff821f7020
[ 3929.470770] RBP: ffff88005872f6e8 R08: 000000000001f010 R09: 0000000000000000
[ 3929.471133] R10: ffff88005872f730 R11: 0000000000008cdd R12: ff006d75646d7564
[ 3929.471496] R13: 00000000014000c0 R14: ffff88005b403c00 R15: ffff88005b403c00
[ 3929.471869] FS:  00007f0b70480740(0000) GS:ffff88005d980000(0000) knlGS:0000000000000000
[ 3929.472286] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 3929.472677] CR2: 00007ffcee4f3000 CR3: 0000000059d45000 CR4: 00000000000406e0
[ 3929.473209] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 3929.474109] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 3929.474873] Call Trace:
[ 3929.475337]  ? kstrdup_const+0x23/0x25
[ 3929.475863]  kstrdup+0x2e/0x4b
[ 3929.476338]  kstrdup_const+0x23/0x25
[ 3929.478084]  __kernfs_new_node+0x28/0xbc
[ 3929.478478]  kernfs_new_node+0x35/0x55
[ 3929.478929]  kernfs_create_link+0x23/0x76
[ 3929.479478]  sysfs_do_create_link_sd.isra.2+0x85/0xd7
[ 3929.480096]  sysfs_create_link+0x33/0x35
[ 3929.480649]  device_add+0x200/0x589
[ 3929.481184]  netdev_register_kobject+0x7c/0x12f
[ 3929.481711]  register_netdevice+0x373/0x471
[ 3929.482174]  rtnl_newlink+0x614/0x729
[ 3929.482610]  ? rtnl_newlink+0x17f/0x729
[ 3929.483080]  rtnetlink_rcv_msg+0x188/0x197
[ 3929.483533]  ? rcu_read_unlock+0x3e/0x5f
[ 3929.483984]  ? rtnl_newlink+0x729/0x729
[ 3929.484420]  netlink_rcv_skb+0x6c/0xce
[ 3929.484858]  rtnetlink_rcv+0x23/0x2a
[ 3929.485291]  netlink_unicast+0x103/0x181
[ 3929.485735]  netlink_sendmsg+0x326/0x337
[ 3929.486181]  sock_sendmsg_nosec+0x14/0x3f
[ 3929.486614]  sock_sendmsg+0x29/0x2e
[ 3929.486973]  ___sys_sendmsg+0x209/0x28b
[ 3929.487340]  ? do_raw_spin_unlock+0xcd/0xf8
[ 3929.487719]  ? _raw_spin_unlock+0x27/0x31
[ 3929.488092]  ? __handle_mm_fault+0x651/0xdb1
[ 3929.488471]  ? check_chain_key+0xb0/0xfd
[ 3929.488847]  __sys_sendmsg+0x45/0x63
[ 3929.489206]  ? __sys_sendmsg+0x45/0x63
[ 3929.489576]  SyS_sendmsg+0x19/0x1b
[ 3929.489901]  entry_SYSCALL_64_fastpath+0x23/0xc2
[ 3929.490172] RIP: 0033:0x7f0b6fb93690
[ 3929.490423] RSP: 002b:00007ffcee4ed588 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
[ 3929.490881] RAX: ffffffffffffffda RBX: ffffffff810d278c RCX: 00007f0b6fb93690
[ 3929.491198] RDX: 0000000000000000 RSI: 00007ffcee4ed5d0 RDI: 0000000000000003
[ 3929.491521] RBP: ffff88005872ff98 R08: 0000000000000001 R09: 0000000000000000
[ 3929.491801] R10: 00007ffcee4ed350 R11: 0000000000000246 R12: 0000000000000002
[ 3929.492075] R13: 000000000066f1a0 R14: 00007ffcee4f5680 R15: 0000000000000000
[ 3929.492352]  ? trace_hardirqs_off_caller+0xa7/0xcf
[ 3929.492590] Code: 8b 45 c0 48 8b 45 b8 74 17 48 8b 4d c8 83 ca ff 44
89 ee 4c 89 f7 e8 83 ca ff ff 49 89 c4 eb 49 49 63 56 20 48 8d 48 01 4d
8b 06 <49> 8b 1c 14 48 89 c2 4c 89 e0 65 49 0f c7 08 0f 94 c0 83 f0 01
[ 3929.493335] RIP: __kmalloc_track_caller+0x117/0x1be RSP: ffff88005872f6a0

Fixes: 87b60cfacf9f ("net_sched: fix error recovery at qdisc creation")
Fixes: f07d1501292b ("multiq: Further multiqueue cleanup")
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[AmitP: Removed unused variable 'err' in multiq_init()]
Signed-off-by: Amit Pundir <amit.pundir@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/sched/sch_multiq.c

index 9ffbb02..66b6e80 100644 (file)
@@ -234,7 +234,7 @@ static int multiq_tune(struct Qdisc *sch, struct nlattr *opt)
 static int multiq_init(struct Qdisc *sch, struct nlattr *opt)
 {
        struct multiq_sched_data *q = qdisc_priv(sch);
-       int i, err;
+       int i;
 
        q->queues = NULL;
 
@@ -249,12 +249,7 @@ static int multiq_init(struct Qdisc *sch, struct nlattr *opt)
        for (i = 0; i < q->max_bands; i++)
                q->queues[i] = &noop_qdisc;
 
-       err = multiq_tune(sch, opt);
-
-       if (err)
-               kfree(q->queues);
-
-       return err;
+       return multiq_tune(sch, opt);
 }
 
 static int multiq_dump(struct Qdisc *sch, struct sk_buff *skb)