On 05/07/2018 05:51 PM, Jia-Ju Bai wrote:
So ?
On 2018/5/7 22:15, Eric Dumazet wrote:
On 05/07/2018 07:08 AM, Jia-Ju Bai wrote:Thanks for your reply :)
The write operations to "dev->stats" are protected bydev->stats is not a pointer, it is an array embedded in the
the spinlock on line 862-864, but the read operations to
this data on line 858 and 867 are not protected by the spinlock.
Thus, there may exist data races for "dev->stats".
To fix the data races, the read operations to "dev->stats" are
protected by the spinlock, and a local variable is used for return.
Signed-off-by: Jia-Ju Bai <baijiaju1990@xxxxxxxxx>
---
drivers/net/ethernet/8390/lib8390.c | 14 ++++++++++----
1 file changed, 10 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/8390/lib8390.c b/drivers/net/ethernet/8390/lib8390.c
index c9c55c9eab9f..198952247d30 100644
--- a/drivers/net/ethernet/8390/lib8390.c
+++ b/drivers/net/ethernet/8390/lib8390.c
@@ -852,19 +852,25 @@ static struct net_device_stats *__ei_get_stats(struct net_device *dev)
unsigned long ioaddr = dev->base_addr;
struct ei_device *ei_local = netdev_priv(dev);
unsigned long flags;
+ struct net_device_stats *stats;
+
+ spin_lock_irqsave(&ei_local->page_lock, flags);
/* If the card is stopped, just return the present stats. */
- if (!netif_running(dev))
- return &dev->stats;
+ if (!netif_running(dev)) {
+ stats = &dev->stats;
+ spin_unlock_irqrestore(&ei_local->page_lock, flags);
+ return stats;
+ }
- spin_lock_irqsave(&ei_local->page_lock, flags);
/* Read the counter registers, assuming we are in page 0. */
dev->stats.rx_frame_errors += ei_inb_p(ioaddr + EN0_COUNTER0);
dev->stats.rx_crc_errors += ei_inb_p(ioaddr + EN0_COUNTER1);
dev->stats.rx_missed_errors += ei_inb_p(ioaddr + EN0_COUNTER2);
+ stats = &dev->stats;
spin_unlock_irqrestore(&ei_local->page_lock, flags);
- return &dev->stats;
+ return stats;
}
/*
struct net_device
So this patch is not needed, since dev->stats can not change.
I do not understand that why "dev->stats can not change".
Its data is indeed changed by the code:
dev->stats.rx_frame_errors += ei_inb_p(ioaddr + EN0_COUNTER0);
dev->stats.rx_crc_errors += ei_inb_p(ioaddr + EN0_COUNTER1);
dev->stats.rx_missed_errors += ei_inb_p(ioaddr + EN0_COUNTER2);
So I think a data race may occur when returning "dev->stats" without lock protection.&dev->stats is a stable value.
It wont change over the lifetime of net_device object.
Adding a barrier before or after getting &dev->stats is useless, confusing and really not needed.