net/sched: act_ife: fix recursive lock and idr leak
authorDavide Caratti <dcaratti@redhat.com>
Tue, 19 Jun 2018 13:39:46 +0000 (15:39 +0200)
committerDavid S. Miller <davem@davemloft.net>
Wed, 20 Jun 2018 00:03:58 +0000 (09:03 +0900)
a recursive lock warning [1] can be observed with the following script,

 # $TC actions add action ife encode allow prio pass index 42
 IFE type 0xED3E
 # $TC actions replace action ife encode allow tcindex pass index 42

in case the kernel was unable to run the last command (e.g. because of
the impossibility to load 'act_meta_skbtcindex'). For a similar reason,
the kernel can leak idr in the error path of tcf_ife_init(), because
tcf_idr_release() is not called after successful idr reservation:

 # $TC actions add action ife encode allow tcindex index 47
 IFE type 0xED3E
 RTNETLINK answers: No such file or directory
 We have an error talking to the kernel
 # $TC actions add action ife encode allow tcindex index 47
 IFE type 0xED3E
 RTNETLINK answers: No space left on device
 We have an error talking to the kernel
 # $TC actions add action ife encode use mark 7 type 0xfefe pass index 47
 IFE type 0xFEFE
 RTNETLINK answers: No space left on device
 We have an error talking to the kernel

Since tcfa_lock is already taken when the action is being edited, a call
to tcf_idr_release() wrongly makes tcf_idr_cleanup() take the same lock
again. On the other hand, tcf_idr_release() needs to be called in the
error path of tcf_ife_init(), to undo the last tcf_idr_create() invocation.
Fix both problems in tcf_ife_init().
Since the cleanup() routine can now be called when ife->params is NULL,
also add a NULL pointer check to avoid calling kfree_rcu(NULL, rcu).

 [1]
 ============================================
 WARNING: possible recursive locking detected
 4.17.0-rc4.kasan+ #417 Tainted: G            E
 --------------------------------------------
 tc/3932 is trying to acquire lock:
 000000005097c9a6 (&(&p->tcfa_lock)->rlock){+...}, at: tcf_ife_cleanup+0x19/0x80 [act_ife]

 but task is already holding lock:
 000000005097c9a6 (&(&p->tcfa_lock)->rlock){+...}, at: tcf_ife_init+0xf6d/0x13c0 [act_ife]

 other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(&(&p->tcfa_lock)->rlock);
   lock(&(&p->tcfa_lock)->rlock);

  *** DEADLOCK ***

  May be due to missing lock nesting notation

 2 locks held by tc/3932:
  #0: 000000007ca8e990 (rtnl_mutex){+.+.}, at: tcf_ife_init+0xf61/0x13c0 [act_ife]
  #1: 000000005097c9a6 (&(&p->tcfa_lock)->rlock){+...}, at: tcf_ife_init+0xf6d/0x13c0 [act_ife]

 stack backtrace:
 CPU: 3 PID: 3932 Comm: tc Tainted: G            E     4.17.0-rc4.kasan+ #417
 Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
 Call Trace:
  dump_stack+0x9a/0xeb
  __lock_acquire+0xf43/0x34a0
  ? debug_check_no_locks_freed+0x2b0/0x2b0
  ? debug_check_no_locks_freed+0x2b0/0x2b0
  ? debug_check_no_locks_freed+0x2b0/0x2b0
  ? __mutex_lock+0x62f/0x1240
  ? kvm_sched_clock_read+0x1a/0x30
  ? sched_clock+0x5/0x10
  ? sched_clock_cpu+0x18/0x170
  ? find_held_lock+0x39/0x1d0
  ? lock_acquire+0x10b/0x330
  lock_acquire+0x10b/0x330
  ? tcf_ife_cleanup+0x19/0x80 [act_ife]
  _raw_spin_lock_bh+0x38/0x70
  ? tcf_ife_cleanup+0x19/0x80 [act_ife]
  tcf_ife_cleanup+0x19/0x80 [act_ife]
  __tcf_idr_release+0xff/0x350
  tcf_ife_init+0xdde/0x13c0 [act_ife]
  ? ife_exit_net+0x290/0x290 [act_ife]
  ? __lock_is_held+0xb4/0x140
  tcf_action_init_1+0x67b/0xad0
  ? tcf_action_dump_old+0xa0/0xa0
  ? sched_clock+0x5/0x10
  ? sched_clock_cpu+0x18/0x170
  ? kvm_sched_clock_read+0x1a/0x30
  ? sched_clock+0x5/0x10
  ? sched_clock_cpu+0x18/0x170
  ? memset+0x1f/0x40
  tcf_action_init+0x30f/0x590
  ? tcf_action_init_1+0xad0/0xad0
  ? memset+0x1f/0x40
  tc_ctl_action+0x48e/0x5e0
  ? mutex_lock_io_nested+0x1160/0x1160
  ? tca_action_gd+0x990/0x990
  ? sched_clock+0x5/0x10
  ? find_held_lock+0x39/0x1d0
  rtnetlink_rcv_msg+0x4da/0x990
  ? validate_linkmsg+0x680/0x680
  ? sched_clock_cpu+0x18/0x170
  ? find_held_lock+0x39/0x1d0
  netlink_rcv_skb+0x127/0x350
  ? validate_linkmsg+0x680/0x680
  ? netlink_ack+0x970/0x970
  ? __kmalloc_node_track_caller+0x304/0x3a0
  netlink_unicast+0x40f/0x5d0
  ? netlink_attachskb+0x580/0x580
  ? _copy_from_iter_full+0x187/0x760
  ? import_iovec+0x90/0x390
  netlink_sendmsg+0x67f/0xb50
  ? netlink_unicast+0x5d0/0x5d0
  ? copy_msghdr_from_user+0x206/0x340
  ? netlink_unicast+0x5d0/0x5d0
  sock_sendmsg+0xb3/0xf0
  ___sys_sendmsg+0x60a/0x8b0
  ? copy_msghdr_from_user+0x340/0x340
  ? lock_downgrade+0x5e0/0x5e0
  ? tty_write_lock+0x18/0x50
  ? kvm_sched_clock_read+0x1a/0x30
  ? sched_clock+0x5/0x10
  ? sched_clock_cpu+0x18/0x170
  ? find_held_lock+0x39/0x1d0
  ? lock_downgrade+0x5e0/0x5e0
  ? lock_acquire+0x10b/0x330
  ? __audit_syscall_entry+0x316/0x690
  ? current_kernel_time64+0x6b/0xd0
  ? __fget_light+0x55/0x1f0
  ? __sys_sendmsg+0xd2/0x170
  __sys_sendmsg+0xd2/0x170
  ? __ia32_sys_shutdown+0x70/0x70
  ? syscall_trace_enter+0x57a/0xd60
  ? rcu_read_lock_sched_held+0xdc/0x110
  ? __bpf_trace_sys_enter+0x10/0x10
  ? do_syscall_64+0x22/0x480
  do_syscall_64+0xa5/0x480
  entry_SYSCALL_64_after_hwframe+0x49/0xbe
 RIP: 0033:0x7fd646988ba0
 RSP: 002b:00007fffc9fab3c8 EFLAGS: 00000246 ORIG_RAX: 000000000000002e
 RAX: ffffffffffffffda RBX: 00007fffc9fab4f0 RCX: 00007fd646988ba0
 RDX: 0000000000000000 RSI: 00007fffc9fab440 RDI: 0000000000000003
 RBP: 000000005b28c8b3 R08: 0000000000000002 R09: 0000000000000000
 R10: 00007fffc9faae20 R11: 0000000000000246 R12: 0000000000000000
 R13: 00007fffc9fab504 R14: 0000000000000001 R15: 000000000066c100

Fixes: 4e8c86155010 ("net sched: net sched: ife action fix late binding")
Fixes: ef6980b6becb ("introduce IFE action")
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/act_ife.c

index 8527cfdc446d9bb82e8fa9fe1364dc13249b1e03..078d52212172f18c222a6aacb29b30e63ca29223 100644 (file)
@@ -415,7 +415,8 @@ static void tcf_ife_cleanup(struct tc_action *a)
        spin_unlock_bh(&ife->tcf_lock);
 
        p = rcu_dereference_protected(ife->params, 1);
-       kfree_rcu(p, rcu);
+       if (p)
+               kfree_rcu(p, rcu);
 }
 
 /* under ife->tcf_lock for existing action */
@@ -543,10 +544,8 @@ static int tcf_ife_init(struct net *net, struct nlattr *nla,
                                       NULL, NULL);
                if (err) {
 metadata_parse_err:
-                       if (exists)
-                               tcf_idr_release(*a, bind);
                        if (ret == ACT_P_CREATED)
-                               _tcf_ife_cleanup(*a);
+                               tcf_idr_release(*a, bind);
 
                        if (exists)
                                spin_unlock_bh(&ife->tcf_lock);
@@ -567,7 +566,7 @@ metadata_parse_err:
                err = use_all_metadata(ife);
                if (err) {
                        if (ret == ACT_P_CREATED)
-                               _tcf_ife_cleanup(*a);
+                               tcf_idr_release(*a, bind);
 
                        if (exists)
                                spin_unlock_bh(&ife->tcf_lock);