Skip to contentSkip to conversation

SMP with AX.25

BUG Applied Jarek Poplawski 5 participants 21 messages Feb 6, 2008 – Mar 24, 2008 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

Here is attached a message I got with AX25 oopses.

Jarek P.

Witam Jarek,

my name is Jann Traschewski, DG8NGN and I run the AMPR-Gateway and heavy used Packet Radio Node DB0FHN on a HP Netserver Dual CPU Server. I'm in contact with Ralf, DL5RB in WW-Convers talking about the Kernel AX.25 stuff. Maybe you are also interested on my problems using SMP on my big machine (see Attachment) with Kernel 2.6.24. Do you want to have an account on the machine?

73,
Jann

--
Jann Traschewski, Drosselstr.1, D-90513 Zirndorf, Germany Tel.: +49-911-696971, Mobile: +49-170-1045937, EMail: [email protected]
Ham: DG8NGN / DB0VOX, http://www.qsl.net/db0fhn, ICQ UIN: 4130182

#1 lore raw

PATCH 1/1 SMP with AX.25

On Wed, Feb 06, 2008 at 07:45:29AM +0000, Jarek Poplawski wrote:
...

Quoted 2 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25

[AX25] ax25_timer: use mod_timer instead of add_timer

According to one of Jann's OOPS reports it looks like BUG_ON(timer_pending(timer)) triggers during add_timer() in ax25_start_t1timer(). This patch changes current use
of: init_timer(), add_timer() and del_timer() to setup_timer() with mod_timer(), which should be safer anyway.

---

 include/net/ax25.h    |    1 +
 net/ax25/af_ax25.c    |    6 +----
 net/ax25/ax25_timer.c |   60 +++++++++++++++++-------------------------------
 3 files changed, 23 insertions(+), 44 deletions(-)
include/net/ax25.h +1 -0
index 32a57e1..3f0236f 100644
@@ -416,6 +416,7 @@ extern void ax25_calculate_rtt(ax25_cb *);
 extern void ax25_disconnect(ax25_cb *, int);
 
 /* ax25_timer.c */
+extern void ax25_setup_timers(ax25_cb *);
 extern void ax25_start_heartbeat(ax25_cb *);
 extern void ax25_start_t1timer(ax25_cb *);
 extern void ax25_start_t2timer(ax25_cb *);
net/ax25/af_ax25.c +1 -5
index 8fc64e3..94b2b1b 100644
@@ -510,11 +510,7 @@ ax25_cb *ax25_create_cb(void)
 	skb_queue_head_init(&ax25->ack_queue);
 	skb_queue_head_init(&ax25->reseq_queue);
 
-	init_timer(&ax25->timer);
-	init_timer(&ax25->t1timer);
-	init_timer(&ax25->t2timer);
-	init_timer(&ax25->t3timer);
-	init_timer(&ax25->idletimer);
+	ax25_setup_timers(ax25);
 
 	ax25_fillin_cb(ax25, NULL);
 
net/ax25/ax25_timer.c +21 -39

84 lines not on this page · raw message

#2 lore raw

PATCH 1/1 ax25_ds_timer: use mod_timer instead of add_timer

Quoted 4 lines · Jarek Poplawski

On Wed, Feb 06, 2008 at 07:45:29AM +0000, Jarek Poplawski wrote:
...

Quoted 2 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25

...

[AX25] ax25_ds_timer: use mod_timer instead of add_timer

This patch changes current use of: init_timer(), add_timer() and del_timer() to setup_timer() with mod_timer(), which should be safer anyway.

---

 include/net/ax25.h       |    1 +
 net/ax25/ax25_dev.c      |    2 +-
 net/ax25/ax25_ds_timer.c |   12 ++++--------
 3 files changed, 6 insertions(+), 9 deletions(-)
include/net/ax25.h +1 -0
index 3f0236f..717e219 100644
@@ -324,6 +324,7 @@ extern void ax25_dama_on(ax25_cb *);
 extern void ax25_dama_off(ax25_cb *);
 
 /* ax25_ds_timer.c */
+extern void ax25_ds_setup_timer(ax25_dev *);
 extern void ax25_ds_set_timer(ax25_dev *);
 extern void ax25_ds_del_timer(ax25_dev *);
 extern void ax25_ds_timer(ax25_cb *);
net/ax25/ax25_dev.c +1 -1
index 528c874..a7a0e0c 100644
@@ -82,7 +82,7 @@ void ax25_dev_device_up(struct net_device *dev)
 	ax25_dev->values[AX25_VALUES_DS_TIMEOUT]= AX25_DEF_DS_TIMEOUT;
 
 #if defined(CONFIG_AX25_DAMA_SLAVE) || defined(CONFIG_AX25_DAMA_MASTER)
-	init_timer(&ax25_dev->dama.slave_timer);
+	ax25_ds_setup_timer(ax25_dev);
 #endif
 
 	spin_lock_bh(&ax25_dev_lock);
net/ax25/ax25_ds_timer.c +4 -8
index c4e3b02..2ce79df 100644
@@ -40,13 +40,10 @@ static void ax25_ds_timeout(unsigned long);
  *	1/10th of a second.
  */
 
-static void ax25_ds_add_timer(ax25_dev *ax25_dev)
+void ax25_ds_setup_timer(ax25_dev *ax25_dev)
 {
-	struct timer_list *t = &ax25_dev->dama.slave_timer;
-	t->data		= (unsigned long) ax25_dev;
-	t->function	= &ax25_ds_timeout;
-	t->expires	= jiffies + HZ;
-	add_timer(t);
+	setup_timer(&ax25_dev->dama.slave_timer, ax25_ds_timeout,
+		    (unsigned long)ax25_dev);
 }
 
 void ax25_ds_del_timer(ax25_dev *ax25_dev)
@@ -60,10 +57,9 @@ void ax25_ds_set_timer(ax25_dev *ax25_dev)
 	if (ax25_dev == NULL)		/* paranoia */
 		return;
 
-	del_timer(&ax25_dev->dama.slave_timer);
 	ax25_dev->dama.slave_timeout =
 		msecs_to_jiffies(ax25_dev->values[AX25_VALUES_DS_TIMEOUT]) / 10;
-	ax25_ds_add_timer(ax25_dev);
+	mod_timer(&ax25_dev->dama.slave_timer, jiffies + HZ);
 }
 
 /*
#3 lore raw

Re: [PATCH][AX25] ax25_ds_timer: use mod_timer instead of add_timer

Patches from Jarek applied (incl. both "testing" patches). Machine is stable since 2 days now.
Regards,
Jann

Quoted 92 lines

-----Ursprüngliche Nachricht-----
Von: Jarek Poplawski [mailto:[email protected]]
Gesendet: Mittwoch, 6. Februar 2008 10:14
An: [email protected]
Cc: Ralf Baechle; Jann Traschewski; David Miller
Betreff: [PATCH][AX25] ax25_ds_timer: use mod_timer instead of add_timer

Quoted 4 lines · Jarek Poplawski

On Wed, Feb 06, 2008 at 07:45:29AM +0000, Jarek Poplawski wrote:
...

Quoted 2 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25

...

[AX25] ax25_ds_timer: use mod_timer instead of add_timer

This patch changes current use of: init_timer(), add_timer() and del_timer() to setup_timer() with mod_timer(), which should be safer anyway.

---

 include/net/ax25.h       |    1 +
 net/ax25/ax25_dev.c      |    2 +-
 net/ax25/ax25_ds_timer.c |   12 ++++--------
 3 files changed, 6 insertions(+), 9 deletions(-)
include/net/ax25.h index renamed from include/net/ax25.h +0 -0

3f0236f..717e219 100644

include/net/ax25.h +1 -0
@@ -324,6 +324,7 @@ extern void ax25_dama_on(ax25_cb *);  
 extern void ax25_dama_off(ax25_cb *);
 
 /* ax25_ds_timer.c */
+extern void ax25_ds_setup_timer(ax25_dev *);
 extern void ax25_ds_set_timer(ax25_dev *);  extern void 

ax25_ds_del_timer(ax25_dev *); extern void ax25_ds_timer(ax25_cb *); diff --git a/net/ax25/ax25_dev.c b/net/ax25/ax25_dev.c index 528c874..a7a0e0c 100644

net/ax25/ax25_dev.c +1 -1
@@ -82,7 +82,7 @@ void ax25_dev_device_up(struct net_device *dev)
 	ax25_dev->values[AX25_VALUES_DS_TIMEOUT]= AX25_DEF_DS_TIMEOUT;
 
 #if defined(CONFIG_AX25_DAMA_SLAVE) || 
 defined(CONFIG_AX25_DAMA_MASTER)
-	init_timer(&ax25_dev->dama.slave_timer);
+	ax25_ds_setup_timer(ax25_dev);
 #endif
 
net/ax25/ax25_dev.c +0 -0
 	spin_lock_bh(&ax25_dev_lock);
(quoted diff) +0 -0

b/net/ax25/ax25_ds_timer.c index c4e3b02..2ce79df 100644

net/ax25/ax25_ds_timer.c +3 -6
@@ -40,13 +40,10 @@ static void ax25_ds_timeout(unsigned long);
  *	1/10th of a second.
  */
 
-static void ax25_ds_add_timer(ax25_dev *ax25_dev)
+void ax25_ds_setup_timer(ax25_dev *ax25_dev)
 {
-	struct timer_list *t = &ax25_dev->dama.slave_timer;
-	t->data		= (unsigned long) ax25_dev;
-	t->function	= &ax25_ds_timeout;
-	t->expires	= jiffies + HZ;
-	add_timer(t);
+	setup_timer(&ax25_dev->dama.slave_timer, ax25_ds_timeout,
+		    (unsigned long)ax25_dev);
 }
 
 void ax25_ds_del_timer(ax25_dev *ax25_dev) @@ -60,10 +57,9 

@@ void ax25_ds_set_timer(ax25_dev *ax25_dev)

net/ax25/ax25_ds_timer.c +0 -1
 	if (ax25_dev == NULL)		/* paranoia */
 		return;
 
-	del_timer(&ax25_dev->dama.slave_timer);
 	ax25_dev->dama.slave_timeout =

msecs_to_jiffies(ax25_dev->values[AX25_VALUES_DS_TIMEOUT]) / 10;

net/ax25/ax25_ds_timer.c +1 -1
-	ax25_ds_add_timer(ax25_dev);
+	mod_timer(&ax25_dev->dama.slave_timer, jiffies + HZ);
 }
 
 /*
maintainer Applied
#4 lore raw

Re: [PATCH][AX25] ax25_ds_timer: use mod_timer instead of add_timer

In reply to Jarek Poplawski

From: Jarek Poplawski <[email protected]>
Date: Wed, 6 Feb 2008 09:14:13 +0000

Quoted 9 lines

[AX25] ax25_ds_timer: use mod_timer instead of add_timer

This patch changes current use of: init_timer(), add_timer() and del_timer() to setup_timer() with mod_timer(), which should be safer anyway.

Applied, thanks.

maintainer Applied
#5 lore raw
In reply to Jarek Poplawski

From: Jarek Poplawski <[email protected]>
Date: Wed, 6 Feb 2008 08:15:09 +0000

Quoted 16 lines · Jarek Poplawski

...

Quoted 2 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25

[AX25] ax25_timer: use mod_timer instead of add_timer

According to one of Jann's OOPS reports it looks like BUG_ON(timer_pending(timer)) triggers during add_timer() in ax25_start_t1timer(). This patch changes current use
of: init_timer(), add_timer() and del_timer() to setup_timer() with mod_timer(), which should be safer anyway.

Quoted 2 lines

Applied.

#6 lore raw
In reply to Jarek Poplawski

On Wed, Feb 06, 2008 at 07:45:29AM +0000, Jarek Poplawski wrote:
...

Quoted 3 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25
To: [email protected]

According to one of OOPSes reported by Jann softirq can break while skb is prepared for netif_rx. The report isn't complete, so the real reason of the later bug could be different, but IMHO this locking in ax_bump is wrong.

I attach this patch for testing purpose only.

Jarek P.

---

 drivers/net/hamradio/mkiss.c |    5 ++---
 1 files changed, 2 insertions(+), 3 deletions(-)
drivers/net/hamradio/mkiss.c +2 -3
index cfcd15a..30c9b3b 100644
@@ -289,7 +289,6 @@ static void ax_bump(struct mkiss *ax)
 			*ax->rbuff &= ~0x20;
 		}
  	}
-	spin_unlock_bh(&ax->buflock);
 
 	count = ax->rcount;
 
@@ -297,17 +296,17 @@ static void ax_bump(struct mkiss *ax)
 		printk(KERN_ERR "mkiss: %s: memory squeeze, dropping packet.\n",
 		       ax->dev->name);
 		ax->stats.rx_dropped++;
+		spin_unlock_bh(&ax->buflock);
 		return;
 	}
 
-	spin_lock_bh(&ax->buflock);
 	memcpy(skb_put(skb,count), ax->rbuff, count);
-	spin_unlock_bh(&ax->buflock);
 	skb->protocol = ax25_type_trans(skb, ax->dev);
 	netif_rx(skb);
 	ax->dev->last_rx = jiffies;
 	ax->stats.rx_packets++;
 	ax->stats.rx_bytes += count;
+	spin_unlock_bh(&ax->buflock);
 }
 
 static void kiss_unesc(struct mkiss *ax, unsigned char s)
#7 lore raw
In reply to Jarek Poplawski

On Wed, Feb 06, 2008 at 07:45:29AM +0000, Jarek Poplawski wrote:
...

Quoted 2 lines

From: Jann Traschewski <[email protected]>
Subject: SMP with AX.25

(testing patch #2)

According to some OOPSes reported by Jann it seems ax25_kick tries to clone NULL skbs sometimes. Probably there is needed a sock lock, but until it's checked more this patch should help to avoid this. There are probably also cases where ax25 ->paclen == 0 can happen: let's check for this.

I attach this patch for testing purpose only.

Jarek P.

---

net/ax25/ax25_out.c +8 -0
diff -Nurp 2.6.24-/net/ax25/ax25_out.c 2.6.24+/net/ax25/ax25_out.c
@@ -117,6 +117,12 @@ void ax25_output(ax25_cb *ax25, int pacl
 	unsigned char *p;
 	int frontlen, len, fragno, ka9qfrag, first = 1;
 
+	if (paclen < 16) {
+		WARN_ON_ONCE(1);
+		kfree_skb(skb);
+		return;
+	}
+
 	if ((skb->len - 1) > paclen) {
 		if (*skb->data == AX25_P_TEXT) {
 			skb_pull(skb, 1); /* skip PID */
@@ -263,6 +269,8 @@ void ax25_kick(ax25_cb *ax25)
 	 * Dequeue the frame and copy it.
 	 */
 	skb  = skb_dequeue(&ax25->write_queue);
+	if (!skb)
+		return;
 
 	do {
 		if ((skbn = skb_clone(skb, GFP_ATOMIC)) == NULL) {
#8 lore raw
In reply to Jarek Poplawski
Quoted 6 lines

---------- Forwarded message ----------
From: Jann Traschewski <[email protected]>
Date: Thu, 7 Feb 2008 06:27:38 +0100
Subject: FW: smp kernel oops ax25 2008-02-07
To: Jarek Poplawski <[email protected]>

...

Quoted 42 lines
BUG: unable to handle kernel NULL pointer dereference at virtual address
00000065
printing eip: c02266c7 *pde = 00000000
Oops: 0000 [#1] SMP
Modules linked in: netconsole ppp_deflate zlib_deflate zlib_inflate bsd_comp

ppp_async ppp_generic slhc tun bitrev crc32 mkiss ax25 crc16 iptable_nat nf_nat nf_conntrack_ipv4 xt_state nf_conntrack ipt_REJECT iptable_filter iptable_mangle xt_MARK ipv6 ipip tunnel4 ide_cd cdrom aic7xxx scsi_transport_spi parport_serial parport_pc parport i2c_piix4 genrtc

Pid: 19761, comm: frmaster Not tainted (2.6.24-dg8ngn #1)
EIP: 0060:[<c02266c7>] EFLAGS: 00010297 CPU: 1
EIP is at skb_clone+0x3/0x4d
EAX: 00000000 EBX: 00000000 ECX: 00000000 EDX: 00000020
ESI: 00000008 EDI: 00000000 EBP: c0cfc8ac ESP: d3cfddc4
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
Process frmaster (pid: 19761, ti=d3cfc000 task=f7fbf570 task.ti=d3cfc000)

Stack: c0cfc800 f8a0a8f9 c0cfc800 d3cfdf6c 00000007 00000015 c0cfc8cc
e84492e0
       c0cfc800 d3cfdf64 f6e6fbc0 f8a0e9d8 d3cfde60 00000000 c1d8c340
d3cfdea0
       c0cfc800 f5d12200 d3cfdf24 c014143d 01e7a9be 00000000 d3cfdf24
00000022
Call Trace:
 [<f8a0a8f9>] ax25_kick+0xaf/0x184 [ax25]
 [<f8a0e9d8>] ax25_sendmsg+0x35f/0x49a [ax25]
 [<c014143d>] __generic_file_aio_write_nolock+0x474/0x4d3
 [<c0221214>] sock_aio_write+0xbc/0xc8
 [<c01414f7>] generic_file_aio_write+0x5b/0xb0
 [<c015a1d3>] do_sync_write+0xc7/0x10a
 [<c022b643>] dev_hard_start_xmit+0x20a/0x26a
 [<c012db6d>] autoremove_wake_function+0x0/0x35
 [<c01253d3>] lock_timer_base+0x19/0x35
 [<c015a960>] vfs_write+0x9e/0x10c
 [<c015aeb7>] sys_write+0x41/0x67
 [<c0103e4e>] syscall_call+0x7/0xb
 =======================
Code: 24 20 89 7c 24 08 89 44 24 04 03 aa 9c 00 00 00 89 2c 24 e8 cc 67 f9
ff 89 44 24 54 8b 44 24 54 83 c4 3c 5b 5e 5f 5d c3 53 89 c3 <8a> 40 65 24 18
3c 08 75 1f 8d 8b a8 00 00 00 f6 41 65 18 75 13
EIP: [<c02266c7>] skb_clone+0x3/0x4d SS:ESP 0068:d3cfddc4
---[ end trace 504af05b5c529aa1 ]---
#9 lore raw
In reply to Jarek Poplawski

Here I resend another OOPS I got from Jann:

On Thu, Feb 07, 2008 at 05:42:42PM +0100, Jann Traschewski wrote:
...

Quoted 61 lines
BUG: unable to handle kernel NULL pointer dereference at virtual address
00000065
printing eip: c02266c7 *pde = 00000000
Oops: 0000 [#1] SMP
Modules linked in: netconsole ppp_deflate zlib_deflate zlib_inflate bsd_comp

ppp_async ppp_generic slhc tun bitrev crc32 mkiss ax25 crc16 iptable_nat nf_nat nf_conntrack_ipv4 xt_state nf_conntrack ipt_REJECT iptable_filter iptable_mangle xt_MARK ipv6 ipip tunnel4 ide_cd cdrom aic7xxx scsi_transport_spi parport_serial parport_pc parport i2c_piix4 genrtc

Pid: 3035, comm: linuxnet Not tainted (2.6.24-dg8ngn #1)
EIP: 0060:[<c02266c7>] EFLAGS: 00010202 CPU: 0
EIP is at skb_clone+0x3/0x4d
EAX: 00000000 EBX: 00000000 ECX: 00000001 EDX: 00000020
ESI: 00000008 EDI: 00000000 EBP: f700a4ac ESP: f720b9ac
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
Process linuxnet (pid: 3035, ti=f720a000 task=f7c63570 task.ti=f720a000)

Stack: f700a400 f8a0a8f9 c01254db 00000000 00000000 f700a400 f700a4cc
f700a400 __wake_up_common+0x32/0x5c
__wake_up+0x32/0x42
 [<c0224e67>] sock_def_readable+0x39/0x63
 [<c02251e1>] sock_queue_rcv_skb+0xb6/0xd1
 [<c022b060>] netif_receive_skb+0x309/0x372
 [<c0117f35>] dequeue_entity+0xb/0x2a
 [<c022d42f>] process_backlog+0x5c/0xaa
 [<c022cf21>] net_rx_action+0x8d/0x173
 [<c012213a>] __do_softirq+0x5d/0xc1
 [<c01221d0>] do_softirq+0x32/0x36
 [<c01223df>] local_bh_enable_ip+0x35/0x40
 [<c0261ebe>] udp_poll+0xc1/0xd5
 [<c0220f2c>] sock_poll+0xc/0xe
 [<c016495a>] do_select+0x229/0x3c9
 [<c01650b6>] __pollwait+0x0/0xac
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c0118e73>] default_wake_function+0x0/0x8
 [<c011756c>] __wake_up_common+0x32/0x5c
 [<c011911c>] __wake_up+0x32/0x42
 [<c0164da0>] core_sys_select+0x2a6/0x2c7
 [<c011756c>] __wake_up_common+0x32/0x5c
 [<c011911c>] __wake_up+0x32/0x42
 [<c01d13c2>] tty_wakeup+0x4c/0x50
 [<c01d72c6>] pty_unthrottle+0x12/0x1d
 [<f89fc60e>] mkiss_receive_buf+0x381/0x3af [mkiss]
sys_select+0xa4/0x187
syscall_call+0x7/0xb
Kernel panic - not syncing: Fatal exception in interrupt
#10 lore raw
Quoted 24 lines · Jarek Poplawski

Here I resend another OOPS I got from Jann:

On Thu, Feb 07, 2008 at 05:42:42PM +0100, Jann Traschewski wrote:
...

Quoted 20 lines
BUG: unable to handle kernel NULL pointer dereference at virtual address
00000065
printing eip: c02266c7 *pde = 00000000
Oops: 0000 [#1] SMP
Modules linked in: netconsole ppp_deflate zlib_deflate zlib_inflate bsd_comp

ppp_async ppp_generic slhc tun bitrev crc32 mkiss ax25 crc16 iptable_nat nf_nat nf_conntrack_ipv4 xt_state nf_conntrack ipt_REJECT iptable_filter iptable_mangle xt_MARK ipv6 ipip tunnel4 ide_cd cdrom aic7xxx scsi_transport_spi parport_serial parport_pc parport i2c_piix4 genrtc

Pid: 3035, comm: linuxnet Not tainted (2.6.24-dg8ngn #1)
EIP: 0060:[<c02266c7>] EFLAGS: 00010202 CPU: 0
EIP is at skb_clone+0x3/0x4d
EAX: 00000000 EBX: 00000000 ECX: 00000001 EDX: 00000020
ESI: 00000008 EDI: 00000000 EBP: f700a4ac ESP: f720b9ac
DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
Process linuxnet (pid: 3035, ti=f720a000 task=f7c63570 task.ti=f720a000)

Stack: f700a400 f8a0a8f9 c01254db 00000000 00000000 f700a400 f700a4cc
f700a400 __wake_up_common+0x32/0x5c
__wake_up+0x32/0x42

Jann, this report is a bit damaged here, so I could be wrong, but it looks similar to your first reports, and I guess this skb_clone is called from ax25_kick too. Then my today testing patch #2 should help for this, I hope.

Jarek P.

#11 lore raw

PATCH 1/1 mkiss: ax_bump() locking fix

In reply to Jarek Poplawski

Hi,

This is unchanged patch previously sent here for testing. I think it should be applied.

Thanks,
Jarek P.

---------------->

Subject: [PATCH][AX25] mkiss: ax_bump() locking fix

According to one of OOPSes reported by Jann softirq can break while skb is prepared for netif_rx. The report isn't complete, so the real reason of the later bug could be different, but IMHO this locking break in ax_bump is unsafe and unnecessary.

---

 drivers/net/hamradio/mkiss.c |    5 ++---
 1 files changed, 2 insertions(+), 3 deletions(-)
drivers/net/hamradio/mkiss.c +2 -3
index cfcd15a..30c9b3b 100644
@@ -289,7 +289,6 @@ static void ax_bump(struct mkiss *ax)
 			*ax->rbuff &= ~0x20;
 		}
  	}
-	spin_unlock_bh(&ax->buflock);
 
 	count = ax->rcount;
 
@@ -297,17 +296,17 @@ static void ax_bump(struct mkiss *ax)
 		printk(KERN_ERR "mkiss: %s: memory squeeze, dropping packet.\n",
 		       ax->dev->name);
 		ax->stats.rx_dropped++;
+		spin_unlock_bh(&ax->buflock);
 		return;
 	}
 
-	spin_lock_bh(&ax->buflock);
 	memcpy(skb_put(skb,count), ax->rbuff, count);
-	spin_unlock_bh(&ax->buflock);
 	skb->protocol = ax25_type_trans(skb, ax->dev);
 	netif_rx(skb);
 	ax->dev->last_rx = jiffies;
 	ax->stats.rx_packets++;
 	ax->stats.rx_bytes += count;
+	spin_unlock_bh(&ax->buflock);
 }
 
 static void kiss_unesc(struct mkiss *ax, unsigned char s)
Applied
#12 lore raw

Re: [PATCH][AX25] mkiss: ax_bump() locking fix

Quoted 17 lines · Jarek Poplawski

Hi,

This is unchanged patch previously sent here for testing. I think it should be applied.

Thanks,
Jarek P.

---------------->

Subject: [PATCH][AX25] mkiss: ax_bump() locking fix

According to one of OOPSes reported by Jann softirq can break while skb is prepared for netif_rx. The report isn't complete, so the real reason of the later bug could be different, but IMHO this locking break in ax_bump is unsafe and unnecessary.

Quoted 7 lines

---

 drivers/net/hamradio/mkiss.c |    5 ++---
 1 files changed, 2 insertions(+), 3 deletions(-)

applied

#13 lore raw

PATCH 1/1 ax25_out: check skb for NULL in ax25_kick()

In reply to Jarek Poplawski

Hi,

Here is an "official" version of "testing patch #2" from this thread. The only difference: ax25->vs is changed only after checking skb is not NULL (plus a comment). IMHO it could be applied.

Thanks,
Jarek P.

---------------->

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

According to some OOPS reports ax25_kick tries to clone NULL skbs sometimes. It looks like a race with ax25_clear_queues(). Probably there is no need to add more than a simple check for this yet. Another report suggested there are probably also cases where ax25 ->paclen == 0 can happen in ax25_output(); this wasn't confirmed during testing but let's leave this debugging check for some time.

---

net/ax25/ax25_out.c +11 -2
diff -Nurp 2.6.24-mm1-/net/ax25/ax25_out.c 2.6.24-mm1+/net/ax25/ax25_out.c
@@ -117,6 +117,12 @@ void ax25_output(ax25_cb *ax25, int pacl
 	unsigned char *p;
 	int frontlen, len, fragno, ka9qfrag, first = 1;
 
+	if (paclen < 16) {
+		WARN_ON_ONCE(1);
+		kfree_skb(skb);
+		return;
+	}
+
 	if ((skb->len - 1) > paclen) {
 		if (*skb->data == AX25_P_TEXT) {
 			skb_pull(skb, 1); /* skip PID */
@@ -251,8 +257,6 @@ void ax25_kick(ax25_cb *ax25)
 	if (start == end)
 		return;
 
-	ax25->vs = start;
-
 	/*
 	 * Transmit data until either we're out of data to send or
 	 * the window is full. Send a poll on the final I frame if
@@ -261,8 +265,13 @@ void ax25_kick(ax25_cb *ax25)
 
 	/*
 	 * Dequeue the frame and copy it.
+	 * Check for race with ax25_clear_queues().
 	 */
 	skb  = skb_dequeue(&ax25->write_queue);
+	if (!skb)
+		return;
+
+	ax25->vs = start;
 
 	do {
 		if ((skbn = skb_clone(skb, GFP_ATOMIC)) == NULL) {
#14 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Applied and stable with Kernel 2.6.24.2 since 12 hours. Regards,
Jann

Quoted 77 lines

-----Ursprüngliche Nachricht-----
Von: Jarek Poplawski [mailto:[email protected]]
Gesendet: Mittwoch, 13. Februar 2008 12:56
An: David Miller
Cc: Jann Traschewski; Bernard Pidoux F6BVP; Ralf Baechle; [email protected]
Betreff: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Hi,

Here is an "official" version of "testing patch #2" from this thread. The only difference: ax25->vs is changed only after checking skb is not NULL (plus a comment). IMHO it could be applied.

Thanks,
Jarek P.

---------------->

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

According to some OOPS reports ax25_kick tries to clone NULL skbs sometimes. It looks like a race with ax25_clear_queues(). Probably there is no need to add more than a simple check for this yet.
Another report suggested there are probably also cases where ax25 ->paclen == 0 can happen in ax25_output(); this wasn't confirmed during testing but let's leave this debugging check for some time.

---

diff -Nurp 2.6.24-mm1-/net/ax25/ax25_out.c
2.6.24-mm1+/net/ax25/ax25_out.c
--- 2.6.24-mm1-/net/ax25/ax25_out.c 2008-01-24
22:58:37.000000000 +0000
+++ 2.6.24-mm1+/net/ax25/ax25_out.c 2008-02-13
10:43:50.000000000 +0000

net/ax25/ax25_out.c +6 -2
@@ -117,6 +117,12 @@ void ax25_output(ax25_cb *ax25, int pacl
 	unsigned char *p;
 	int frontlen, len, fragno, ka9qfrag, first = 1;
 
+	if (paclen < 16) {
+		WARN_ON_ONCE(1);
+		kfree_skb(skb);
+		return;
+	}
+
 	if ((skb->len - 1) > paclen) {
 		if (*skb->data == AX25_P_TEXT) {
 			skb_pull(skb, 1); /* skip PID */
@@ -251,8 +257,6 @@ void ax25_kick(ax25_cb *ax25)
 	if (start == end)
 		return;
 
-	ax25->vs = start;
-
 	/*
 	 * Transmit data until either we're out of data to send or
 	 * the window is full. Send a poll on the final I frame 

if @@ -261,8 +265,13 @@ void ax25_kick(ax25_cb *ax25)

net/ax25/ax25_out.c +5 -0
 	/*
 	 * Dequeue the frame and copy it.
+	 * Check for race with ax25_clear_queues().
 	 */
 	skb  = skb_dequeue(&ax25->write_queue);
+	if (!skb)
+		return;
+
+	ax25->vs = start;
 
 	do {
 		if ((skbn = skb_clone(skb, GFP_ATOMIC)) == NULL) {
#15 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Quoted 73 lines · Jann Traschewski

Applied and stable with Kernel 2.6.24.2 since 12 hours. Regards,
Jann

Quoted 68 lines

-----Ursprüngliche Nachricht-----
Von: Jarek Poplawski [mailto:[email protected]] Gesendet: Mittwoch, 13. Februar 2008 12:56
An: David Miller
Cc: Jann Traschewski; Bernard Pidoux F6BVP; Ralf Baechle; [email protected]
Betreff: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Hi,

Here is an "official" version of "testing patch #2" from this thread.
The only difference: ax25->vs is changed only after checking skb is not NULL (plus a comment). IMHO it could be applied.

Thanks,
Jarek P.

---------------->

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

According to some OOPS reports ax25_kick tries to clone NULL skbs sometimes. It looks like a race with ax25_clear_queues(). Probably there is no need to add more than a simple check for this yet. Another report suggested there are probably also cases where ax25
->paclen == 0 can happen in ax25_output(); this wasn't confirmed
during testing but let's leave this debugging check for some time.

---

diff -Nurp 2.6.24-mm1-/net/ax25/ax25_out.c 2.6.24-mm1+/net/ax25/ax25_out.c
--- 2.6.24-mm1-/net/ax25/ax25_out.c 2008-01-24
22:58:37.000000000 +0000
+++ 2.6.24-mm1+/net/ax25/ax25_out.c 2008-02-13
10:43:50.000000000 +0000

net/ax25/ax25_out.c +6 -2
@@ -117,6 +117,12 @@ void ax25_output(ax25_cb *ax25, int pacl
 	unsigned char *p;
 	int frontlen, len, fragno, ka9qfrag, first = 1;
 
+	if (paclen < 16) {
+		WARN_ON_ONCE(1);
+		kfree_skb(skb);
+		return;
+	}
+
 	if ((skb->len - 1) > paclen) {
 		if (*skb->data == AX25_P_TEXT) {
 			skb_pull(skb, 1); /* skip PID */
@@ -251,8 +257,6 @@ void ax25_kick(ax25_cb *ax25)
 	if (start == end)
 		return;
 
-	ax25->vs = start;
-
 	/*
 	 * Transmit data until either we're out of data to send or
 	 * the window is full. Send a poll on the final I frame 

if @@ -261,8 +265,13 @@ void ax25_kick(ax25_cb *ax25)

net/ax25/ax25_out.c +5 -0
 	/*
 	 * Dequeue the frame and copy it.
+	 * Check for race with ax25_clear_queues().
 	 */
 	skb  = skb_dequeue(&ax25->write_queue);
+	if (!skb)
+		return;
+
+	ax25->vs = start;
Quoted 7 lines

net/ax25/ax25_out.c void ax25_kick(ax25_cb *ax25) · line 277

Quoted 3 lines
net/ax25/ax25_out.c +0 -0
 	do {
 		if ((skbn = skb_clone(skb, GFP_ATOMIC)) == NULL) {

Applied too and stable with kernel 2.6.24.3 since a week.

Thanks Jarek.

Bernard Pidoux

#16 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

On Sun, Mar 09, 2008 at 10:02:39AM +0100, Pidoux wrote:
...

Quoted 1 line
Quoted 1 line
Quoted 1 line

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

...

Quoted 1 line

Applied too and stable with kernel 2.6.24.3 since a week.

Good to know! It's merged to 2.6.25-rc3, but maybe recommending it for 2.6.24 wouldn't be a bad idea...

Thanks Bernard,
Jarek P.

#17 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Indeed. I'd like to see it in 2.6.24. My heavy used machine has no problems with this patch since 13 days now.
Thank you,
Jann

Quoted 19 lines

-----Ursprüngliche Nachricht-----
Von: Jarek Poplawski [mailto:[email protected]]
Gesendet: Sonntag, 9. März 2008 15:31
An: Pidoux
Cc: Jann Traschewski; 'David Miller'; 'Ralf Baechle'; [email protected]
Betreff: Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

On Sun, Mar 09, 2008 at 10:02:39AM +0100, Pidoux wrote:
...

Quoted 1 line
Quoted 1 line
Quoted 1 line

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

...

Quoted 1 line

Applied too and stable with kernel 2.6.24.3 since a week.

Good to know! It's merged to 2.6.25-rc3, but maybe recommending it for 2.6.24 wouldn't be a bad idea...

Thanks Bernard,
Jarek P.

#18 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

Quoted 2 lines · Jann Traschewski

Indeed. I'd like to see it in 2.6.24. My heavy used machine has no problems with this patch since 13 days now.

I hope David or Ralf will consider this.

BTW, Jann, I guess we're waiting with this last AX.25 testing patch for the magic 13 too?

Thanks,
Jarek P.

maintainer
#19 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

In reply to Jarek Poplawski

From: Jarek Poplawski <[email protected]>
Date: Sun, 9 Mar 2008 15:30:52 +0100

Quoted 8 lines · Pidoux

...

Quoted 1 line
Quoted 1 line
Quoted 1 line

Subject: [AX25] ax25_out: check skb for NULL in ax25_kick()

...

Quoted 1 line

Applied too and stable with kernel 2.6.24.3 since a week.

Good to know! It's merged to 2.6.25-rc3, but maybe recommending it for 2.6.24 wouldn't be a bad idea...

I've just queued this up for the next round of fixes I will send to the -stable folks.

maintainer Applied
#20 lore raw

Re: [PATCH][AX25] ax25_out: check skb for NULL in ax25_kick()

In reply to Jarek Poplawski

From: Jarek Poplawski <[email protected]>
Date: Wed, 13 Feb 2008 11:56:07 +0000

Quoted 11 lines

[AX25] ax25_out: check skb for NULL in ax25_kick()

According to some OOPS reports ax25_kick tries to clone NULL skbs sometimes. It looks like a race with ax25_clear_queues(). Probably there is no need to add more than a simple check for this yet. Another report suggested there are probably also cases where ax25 ->paclen == 0 can happen in ax25_output(); this wasn't confirmed during testing but let's leave this debugging check for some time.

Applied, thanks Jarek.