net: Separate the close_list and the unreg_list
Subsystems (3)
This patch from Eric may have fallen through the cracks. At the time he submitted it I did verify it in some failures I was seeing (I have not run extensive tests).
Francesco
Quoted 149 lines · Eric W. Biederman
Separate the
unreg_listand theclose_listindev_close_manypreventingdev_close_manyfrom permuting theunreg_list. The permutations of theunreg_listhave resulted in cases where the loopback device is accessed it has been freed in code such asdst_ifdown. Resulting in subtle memory corruption.This is the second bug from sharing the storage between the
close_listand theunreg_list. The issues that crop up with sharing are apparently too subtle to show up in normal testing or usage, so let's forget about being clever and use two separate lists.Cc <[email protected]>Signed-off-by "Eric W. Biederman" <[email protected]>---
Frencesco if you could test this patch and verify it fixes some of your issues that would be great. I don't expect this fixes the fundamental problem you were seeing but it should solve some related issues.
include/linux/netdevice.h | 1 + net/core/dev.c | 25 +++++++++++++------------ net/sched/sch_generic.c | 6 +++--- 3 files changed, 17 insertions(+), 15 deletions(-)include/linux/netdevice.h +1
-0@@ -1106,6 +1106,7 @@ struct net_device {struct list_head dev_list;struct list_head napi_list;struct list_head unreg_list; +struct list_head close_list;struct list_head upper_dev_list;/* List of upper devices */ net/core/dev.c +13
-12@@ -1307,7 +1307,7 @@ static int __dev_close_many(struct list_head *head)ASSERT_RTNL ();might_sleep (); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {call_netdevice_notifiers (NETDEV_GOING_DOWN , dev);clear_bit (__LINK_STATE_START, &dev->state);@@ -1323,7 +1323,7 @@ static int __dev_close_many(struct list_head *head)dev_deactivate_many (head); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {const struct net_device_ops *ops = dev->netdev_ops;/* @@ -1351,7 +1351,7 @@ static int __dev_close(struct net_device *dev)/* Temporarily disable netpoll until the interface is down */ netpoll_rx_disable (dev); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single); retval =__dev_close_many (&single);list_del (&single);@@ -1362,21 +1362,21 @@ static int __dev_close(struct net_device *dev)static int dev_close_many (struct list_head *head) {struct net_device *dev, *tmp; -LIST_HEAD (tmp_list); +LIST_HEAD (many); -list_for_each_entry_safe (dev, tmp, head, unreg_list) -if (!(dev->flags &IFF_UP )) -list_move (&dev->unreg_list, &tmp_list); +/* rollback_registered_many needs the original unmodified list */ +list_for_each_entry (dev, head, unreg_list) +if (dev->flags &IFF_UP ) +list_add_tail (&dev->close_list, &many); -__dev_close_many (head); +__dev_close_many (&many); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry_safe (dev, tmp, &many, close_list) {rtmsg_ifinfo (RTM_NEWLINK , dev,IFF_UP |IFF_RUNNING );call_netdevice_notifiers (NETDEV_DOWN , dev); +list_del_init (&dev->close_list); } -/* rollback_registered_many needs the complete original list */ -list_splice (&tmp_list, head);return 0 ; }@@ -1397,7 +1397,7 @@ int dev_close(struct net_device *dev)/* Block netpoll rx while the interface is going down */ netpoll_rx_disable (dev); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single);dev_close_many (&single);list_del (&single);@@ -5823,6 +5823,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,INIT_LIST_HEAD (&dev->napi_list);INIT_LIST_HEAD (&dev->unreg_list); +INIT_LIST_HEAD (&dev->close_list);INIT_LIST_HEAD (&dev->link_watch_list);INIT_LIST_HEAD (&dev->upper_dev_list); dev->priv_flags =IFF_XMIT_DST_RELEASE ;net/sched/sch_generic.c +3
-4@@ -818,7 +818,7 @@ void dev_deactivate_many(struct list_head *head)struct net_device *dev;bool sync_needed = false; -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {netdev_for_each_tx_queue (dev, dev_deactivate_queue, &noop_qdisc);if (dev_ingress_queue (dev))@@ -837,7 +837,7 @@ void dev_deactivate_many(struct list_head *head)synchronize_net ();/* Wait for outstanding qdisc_run calls. */ -list_for_each_entry (dev, head, unreg_list) +list_for_each_entry (dev, head, close_list)while (some_qdisc_is_busy (dev))yield (); }@@ -846,7 +846,7 @@ void dev_deactivate(struct net_device *dev) {LIST_HEAD (single); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single);dev_deactivate_many (&single);list_del (&single); } --1.7.5.4
David Miller
maintainerFrom: Francesco Ruggeri <[email protected]>
Date: Thu, 3 Oct 2013 14:51:34 -0700
Quoted 3 lines
This patch from Eric may have fallen through the cracks. At the time he submitted it I did verify it in some failures I was seeing (I have not run extensive tests).
I'd like Eric to explicitly resubmit the patch, as it probably needs to be respun to apply cleanly anyways.
PATCH 1/1 net: Separate the close_list and the unreg_list
Separate the unreg_list and the close_list in dev_close_many preventing dev_close_many from permuting the unreg_list. The permutations of the unreg_list have resulted in cases where the loopback device is accessed it has been freed in code such as dst_ifdown. Resulting in subtle memory corruption.
This is the second bug from sharing the storage between the close_list and the unreg_list. The issues that crop up with sharing are apparently too subtle to show up in normal testing or usage, so let's forget about being clever and use two separate lists.
--- include/linux/netdevice.h | 1 + net/core/dev.c | 25 +++++++++++++------------ net/sched/sch_generic.c | 6 +++--- 3 files changed, 17 insertions(+), 15 deletions(-)
include/linux/netdevice.h +1 -0
@@ -1143,6 +1143,7 @@ struct net_device {struct list_head dev_list;struct list_head napi_list;struct list_head unreg_list; +struct list_head close_list;/* directly linked devices, like slaves for bonding */ struct {
net/core/dev.c +13 -12
@@ -1307,7 +1307,7 @@ static int __dev_close_many(struct list_head *head)ASSERT_RTNL ();might_sleep (); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {call_netdevice_notifiers (NETDEV_GOING_DOWN , dev);clear_bit (__LINK_STATE_START, &dev->state);
@@ -1323,7 +1323,7 @@ static int __dev_close_many(struct list_head *head)dev_deactivate_many (head); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {const struct net_device_ops *ops = dev->netdev_ops;/*
@@ -1351,7 +1351,7 @@ static int __dev_close(struct net_device *dev)/* Temporarily disable netpoll until the interface is down */ netpoll_rx_disable (dev); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single); retval =__dev_close_many (&single);list_del (&single);
@@ -1362,21 +1362,21 @@ static int __dev_close(struct net_device *dev)static int dev_close_many (struct list_head *head) {struct net_device *dev, *tmp; -LIST_HEAD (tmp_list); +LIST_HEAD (many); -list_for_each_entry_safe (dev, tmp, head, unreg_list) -if (!(dev->flags &IFF_UP )) -list_move (&dev->unreg_list, &tmp_list); +/* rollback_registered_many needs the original unmodified list */ +list_for_each_entry (dev, head, unreg_list) +if (dev->flags &IFF_UP ) +list_add_tail (&dev->close_list, &many); -__dev_close_many (head); +__dev_close_many (&many); -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry_safe (dev, tmp, &many, close_list) {rtmsg_ifinfo (RTM_NEWLINK , dev,IFF_UP |IFF_RUNNING );call_netdevice_notifiers (NETDEV_DOWN , dev); +list_del_init (&dev->close_list); } -/* rollback_registered_many needs the complete original list */ -list_splice (&tmp_list, head);return 0 ; }
@@ -1397,7 +1397,7 @@ int dev_close(struct net_device *dev)/* Block netpoll rx while the interface is going down */ netpoll_rx_disable (dev); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single);dev_close_many (&single);list_del (&single);
@@ -6257,6 +6257,7 @@ struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,INIT_LIST_HEAD (&dev->napi_list);INIT_LIST_HEAD (&dev->unreg_list); +INIT_LIST_HEAD (&dev->close_list);INIT_LIST_HEAD (&dev->link_watch_list);INIT_LIST_HEAD (&dev->adj_list.upper);INIT_LIST_HEAD (&dev->adj_list.lower);
net/sched/sch_generic.c +3 -3
@@ -829,7 +829,7 @@ void dev_deactivate_many(struct list_head *head)struct net_device *dev;bool sync_needed = false; -list_for_each_entry (dev, head, unreg_list) { +list_for_each_entry (dev, head, close_list) {netdev_for_each_tx_queue (dev, dev_deactivate_queue, &noop_qdisc);if (dev_ingress_queue (dev))
@@ -848,7 +848,7 @@ void dev_deactivate_many(struct list_head *head)synchronize_net ();/* Wait for outstanding qdisc_run calls. */ -list_for_each_entry (dev, head, unreg_list) +list_for_each_entry (dev, head, close_list)while (some_qdisc_is_busy (dev))yield (); }
@@ -857,7 +857,7 @@ void dev_deactivate(struct net_device *dev) {LIST_HEAD (single); -list_add (&dev->unreg_list, &single); +list_add (&dev->close_list, &single);dev_deactivate_many (&single);list_del (&single); }
Signature · 1 line
1.7.5.4