[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20e227c0ef20063d1482fc9771f4100bda93247e.1265659937.git.jan.kiszka@web.de>
Date: Mon, 8 Feb 2010 21:12:40 +0100
From: Jan Kiszka <jan.kiszka@....de>
To: David Miller <davem@...emloft.net>,
Karsten Keil <isdn@...ux-pingi.de>
Cc: linux-kernel@...r.kernel.org, i4ldeveloper@...tserv.isdn4linux.de,
isdn4linux@...tserv.isdn4linux.de, netdev@...r.kernel.org,
Alan Cox <alan@...rguk.ukuu.org.uk>,
Marcel Holtmann <marcel@...tmann.org>
Subject: [PATCH v2 36/41] CAPI: Fix locking around capiminor's output queue and drop workaround_lock
Introduce outlock as a spin lock that protects capiminor's outqueue,
outbytes and outskb (formerly known as ttyskb). outlock can be acquired
from soft-IRQ context via capinc_write, so make it bh-safe.
This finally removes the last reason for keeping the workaround lock
around (which was incomplete and partly broken anyway). And as we no
longer call handle_recv_skb in atomic context, gen_data_b3_resp_for can
use non-atomic allocation now.
Signed-off-by: Jan Kiszka <jan.kiszka@....de>
---
drivers/isdn/capi/capi.c | 126 ++++++++++++++++++++++++---------------------
1 files changed, 67 insertions(+), 59 deletions(-)
diff --git a/drivers/isdn/capi/capi.c b/drivers/isdn/capi/capi.c
index be85c8c..074496f 100644
--- a/drivers/isdn/capi/capi.c
+++ b/drivers/isdn/capi/capi.c
@@ -95,11 +95,13 @@ struct capiminor {
struct tty_port port;
int ttyinstop;
int ttyoutstop;
- struct sk_buff *ttyskb;
- struct sk_buff_head inqueue;
- struct sk_buff_head outqueue;
- int outbytes;
+ struct sk_buff_head inqueue;
+
+ struct sk_buff_head outqueue;
+ int outbytes;
+ struct sk_buff *outskb;
+ spinlock_t outlock;
/* transmit path */
struct list_head ackqueue;
@@ -107,15 +109,6 @@ struct capiminor {
spinlock_t ackqlock;
};
-/* FIXME: The following lock is a sledgehammer-workaround to a
- * locking issue with the capiminor (and maybe other) data structure(s).
- * Access to this data is done in a racy way and crashes the machine with
- * a FritzCard DSL driver; sooner or later. This is a workaround
- * which trades scalability vs stability, so it doesn't crash the kernel anymore.
- * The correct (and scalable) fix for the issue seems to require
- * an API change to the drivers... . */
-static DEFINE_SPINLOCK(workaround_lock);
-
struct capincci {
struct list_head list;
u32 ncci;
@@ -231,6 +224,7 @@ static struct capiminor *capiminor_alloc(struct capi20_appl *ap, u32 ncci)
skb_queue_head_init(&mp->inqueue);
skb_queue_head_init(&mp->outqueue);
+ spin_lock_init(&mp->outlock);
tty_port_init(&mp->port);
mp->port.ops = &capiminor_port_ops;
@@ -271,7 +265,7 @@ static void capiminor_destroy(struct kref *kref)
{
struct capiminor *mp = container_of(kref, struct capiminor, kref);
- kfree_skb(mp->ttyskb);
+ kfree_skb(mp->outskb);
skb_queue_purge(&mp->inqueue);
skb_queue_purge(&mp->outqueue);
capiminor_del_all_ack(mp);
@@ -417,7 +411,7 @@ static struct sk_buff *
gen_data_b3_resp_for(struct capiminor *mp, struct sk_buff *skb)
{
struct sk_buff *nskb;
- nskb = alloc_skb(CAPI_DATA_B3_RESP_LEN, GFP_ATOMIC);
+ nskb = alloc_skb(CAPI_DATA_B3_RESP_LEN, GFP_KERNEL);
if (nskb) {
u16 datahandle = CAPIMSG_U16(skb->data,CAPIMSG_BASELEN+4+4+2);
unsigned char *s = skb_put(nskb, CAPI_DATA_B3_RESP_LEN);
@@ -548,9 +542,18 @@ static int handle_minor_send(struct capiminor *mp)
return 0;
}
- while ((skb = skb_dequeue(&mp->outqueue)) != NULL) {
- datahandle = atomic_inc_return(&mp->datahandle);
+ while (1) {
+ spin_lock_bh(&mp->outlock);
+ skb = __skb_dequeue(&mp->outqueue);
+ if (!skb) {
+ spin_unlock_bh(&mp->outlock);
+ break;
+ }
len = (u16)skb->len;
+ mp->outbytes -= len;
+ spin_unlock_bh(&mp->outlock);
+
+ datahandle = atomic_inc_return(&mp->datahandle);
skb_push(skb, CAPI_DATA_B3_REQ_LEN);
memset(skb->data, 0, CAPI_DATA_B3_REQ_LEN);
capimsg_setu16(skb->data, 0, CAPI_DATA_B3_REQ_LEN);
@@ -566,14 +569,18 @@ static int handle_minor_send(struct capiminor *mp)
if (capiminor_add_ack(mp, datahandle) < 0) {
skb_pull(skb, CAPI_DATA_B3_REQ_LEN);
- skb_queue_head(&mp->outqueue, skb);
+
+ spin_lock_bh(&mp->outlock);
+ __skb_queue_head(&mp->outqueue, skb);
+ mp->outbytes += len;
+ spin_unlock_bh(&mp->outlock);
+
tty_kref_put(tty);
return count;
}
errcode = capi20_put_message(mp->ap, skb);
if (errcode == CAPI_NOERROR) {
count++;
- mp->outbytes -= len;
#ifdef _DEBUG_DATAFLOW
printk(KERN_DEBUG "capi: DATA_B3_REQ %u len=%u\n",
datahandle, len);
@@ -584,13 +591,17 @@ static int handle_minor_send(struct capiminor *mp)
if (errcode == CAPI_SENDQUEUEFULL) {
skb_pull(skb, CAPI_DATA_B3_REQ_LEN);
- skb_queue_head(&mp->outqueue, skb);
+
+ spin_lock_bh(&mp->outlock);
+ __skb_queue_head(&mp->outqueue, skb);
+ mp->outbytes += len;
+ spin_unlock_bh(&mp->outlock);
+
break;
}
/* ups, drop packet */
printk(KERN_ERR "capi: put_message = %x\n", errcode);
- mp->outbytes -= len;
kfree_skb(skb);
}
tty_kref_put(tty);
@@ -609,7 +620,6 @@ static void capi_recv_message(struct capi20_appl *ap, struct sk_buff *skb)
u16 datahandle;
#endif /* CONFIG_ISDN_CAPI_MIDDLEWARE */
struct capincci *np;
- unsigned long flags;
mutex_lock(&cdev->lock);
@@ -621,7 +631,6 @@ static void capi_recv_message(struct capi20_appl *ap, struct sk_buff *skb)
if (CAPIMSG_CMD(skb->data) == CAPI_CONNECT_B3_IND)
capincci_alloc(cdev, CAPIMSG_NCCI(skb->data));
- spin_lock_irqsave(&workaround_lock, flags);
if (CAPIMSG_COMMAND(skb->data) != CAPI_DATA_B3) {
skb_queue_tail(&cdev->recvqueue, skb);
wake_up_interruptible(&cdev->recvwait);
@@ -683,7 +692,6 @@ static void capi_recv_message(struct capi20_appl *ap, struct sk_buff *skb)
#endif /* CONFIG_ISDN_CAPI_MIDDLEWARE */
unlock_out:
- spin_unlock_irqrestore(&workaround_lock, flags);
mutex_unlock(&cdev->lock);
}
@@ -1062,16 +1070,13 @@ static void capinc_tty_cleanup(struct tty_struct *tty)
static int capinc_tty_open(struct tty_struct *tty, struct file *filp)
{
struct capiminor *mp = tty->driver_data;
- unsigned long flags;
int err;
err = tty_port_open(&mp->port, tty, filp);
if (err)
return err;
- spin_lock_irqsave(&workaround_lock, flags);
handle_minor_recv(mp);
- spin_unlock_irqrestore(&workaround_lock, flags);
return 0;
}
@@ -1087,71 +1092,78 @@ static int capinc_tty_write(struct tty_struct *tty,
{
struct capiminor *mp = tty->driver_data;
struct sk_buff *skb;
- unsigned long flags;
#ifdef _DEBUG_TTYFUNCS
printk(KERN_DEBUG "capinc_tty_write(count=%d)\n", count);
#endif
- spin_lock_irqsave(&workaround_lock, flags);
- skb = mp->ttyskb;
+ spin_lock_bh(&mp->outlock);
+ skb = mp->outskb;
if (skb) {
- mp->ttyskb = NULL;
- skb_queue_tail(&mp->outqueue, skb);
+ mp->outskb = NULL;
+ __skb_queue_tail(&mp->outqueue, skb);
mp->outbytes += skb->len;
}
skb = alloc_skb(CAPI_DATA_B3_REQ_LEN+count, GFP_ATOMIC);
if (!skb) {
printk(KERN_ERR "capinc_tty_write: alloc_skb failed\n");
- spin_unlock_irqrestore(&workaround_lock, flags);
+ spin_unlock_bh(&mp->outlock);
return -ENOMEM;
}
skb_reserve(skb, CAPI_DATA_B3_REQ_LEN);
memcpy(skb_put(skb, count), buf, count);
- skb_queue_tail(&mp->outqueue, skb);
+ __skb_queue_tail(&mp->outqueue, skb);
mp->outbytes += skb->len;
+ spin_unlock_bh(&mp->outlock);
+
(void)handle_minor_send(mp);
- spin_unlock_irqrestore(&workaround_lock, flags);
+
return count;
}
static int capinc_tty_put_char(struct tty_struct *tty, unsigned char ch)
{
struct capiminor *mp = tty->driver_data;
+ bool invoke_send = false;
struct sk_buff *skb;
- unsigned long flags;
int ret = 1;
#ifdef _DEBUG_TTYFUNCS
printk(KERN_DEBUG "capinc_put_char(%u)\n", ch);
#endif
- spin_lock_irqsave(&workaround_lock, flags);
- skb = mp->ttyskb;
+ spin_lock_bh(&mp->outlock);
+ skb = mp->outskb;
if (skb) {
if (skb_tailroom(skb) > 0) {
*(skb_put(skb, 1)) = ch;
- spin_unlock_irqrestore(&workaround_lock, flags);
- return 1;
+ goto unlock_out;
}
- mp->ttyskb = NULL;
- skb_queue_tail(&mp->outqueue, skb);
+ mp->outskb = NULL;
+ __skb_queue_tail(&mp->outqueue, skb);
mp->outbytes += skb->len;
- (void)handle_minor_send(mp);
+ invoke_send = true;
}
+
skb = alloc_skb(CAPI_DATA_B3_REQ_LEN+CAPI_MAX_BLKSIZE, GFP_ATOMIC);
if (skb) {
skb_reserve(skb, CAPI_DATA_B3_REQ_LEN);
*(skb_put(skb, 1)) = ch;
- mp->ttyskb = skb;
+ mp->outskb = skb;
} else {
printk(KERN_ERR "capinc_put_char: char %u lost\n", ch);
ret = 0;
}
- spin_unlock_irqrestore(&workaround_lock, flags);
+
+unlock_out:
+ spin_unlock_bh(&mp->outlock);
+
+ if (invoke_send)
+ (void)handle_minor_send(mp);
+
return ret;
}
@@ -1159,22 +1171,24 @@ static void capinc_tty_flush_chars(struct tty_struct *tty)
{
struct capiminor *mp = tty->driver_data;
struct sk_buff *skb;
- unsigned long flags;
#ifdef _DEBUG_TTYFUNCS
printk(KERN_DEBUG "capinc_tty_flush_chars\n");
#endif
- spin_lock_irqsave(&workaround_lock, flags);
- skb = mp->ttyskb;
+ spin_lock_bh(&mp->outlock);
+ skb = mp->outskb;
if (skb) {
- mp->ttyskb = NULL;
- skb_queue_tail(&mp->outqueue, skb);
+ mp->outskb = NULL;
+ __skb_queue_tail(&mp->outqueue, skb);
mp->outbytes += skb->len;
+ spin_unlock_bh(&mp->outlock);
+
(void)handle_minor_send(mp);
- }
- (void)handle_minor_recv(mp);
- spin_unlock_irqrestore(&workaround_lock, flags);
+ } else
+ spin_unlock_bh(&mp->outlock);
+
+ handle_minor_recv(mp);
}
static int capinc_tty_write_room(struct tty_struct *tty)
@@ -1234,15 +1248,12 @@ static void capinc_tty_throttle(struct tty_struct *tty)
static void capinc_tty_unthrottle(struct tty_struct *tty)
{
struct capiminor *mp = tty->driver_data;
- unsigned long flags;
#ifdef _DEBUG_TTYFUNCS
printk(KERN_DEBUG "capinc_tty_unthrottle\n");
#endif
- spin_lock_irqsave(&workaround_lock, flags);
mp->ttyinstop = 0;
handle_minor_recv(mp);
- spin_unlock_irqrestore(&workaround_lock, flags);
}
static void capinc_tty_stop(struct tty_struct *tty)
@@ -1258,15 +1269,12 @@ static void capinc_tty_stop(struct tty_struct *tty)
static void capinc_tty_start(struct tty_struct *tty)
{
struct capiminor *mp = tty->driver_data;
- unsigned long flags;
#ifdef _DEBUG_TTYFUNCS
printk(KERN_DEBUG "capinc_tty_start\n");
#endif
- spin_lock_irqsave(&workaround_lock, flags);
mp->ttyoutstop = 0;
(void)handle_minor_send(mp);
- spin_unlock_irqrestore(&workaround_lock, flags);
}
static void capinc_tty_hangup(struct tty_struct *tty)
--
1.6.0.2
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists