A few months back a race was discused between the netpoll napi service
path, and the fast path through net_rx_action:
http://kerneltrap.org/mailarchive/linux-netdev/2007/10/16/345470
A patch was submitted for that bug, but I think we missed a case.
Consider the following scenario:
INITIAL STATE
CPU0 has one napi_struct A on its poll_list
CPU1 is calling netpoll_send_skb and needs to call poll_napi on the same
napi_struct A that CPU0 has on its list
CPU0 CPU1
net_rx_action poll_napi
!list_empty (returns true) locks poll_lock for A
poll_one_napi
napi->poll
netif_rx_complete
__napi_complete
(removes A from poll_list)
list_entry(list->next)
In the above scenario, net_rx_action assumes that the per-cpu poll_list is
exclusive to that cpu. netpoll of course violates that, and because the netpoll
path can dequeue from the poll list, its possible for CPU0 to detect a non-empty
list at the top of the while loop in net_rx_action, but have it become empty by
the time it calls list_entry. Since the poll_list isn't surrounded by any other
structure, the returned data from that list_entry call in this situation is
garbage, and any number of crashes can result based on what exactly that garbage
is.
Given that its not fasible for performance reasons to place exclusive locks
arround each cpus poll list to provide that mutal exclusion, I think the best
solution is modify the netpoll path in such a way that we continue to guarantee
that the poll_list for a cpu is in fact exclusive to that cpu. To do this I've
implemented the patch below. It adds an additional bit to the state field in
the napi_struct. When executing napi->poll from the netpoll_path, this bit will
be set. When a driver calls netif_rx_complete, if that bit is set, it will not
remove the napi_struct from the poll_list. That work will be saved for the next
iteration of net_rx_action.
I've tested this and it seems to work well. About the biggest drawback I can
see to it is the fact that it might result in an extra loop through
net_rx_action in the event that the device is actually contended for (i.e. the
netpoll path actually preforms all the needed work no the device, and the call
to net_rx_action winds up doing nothing, except removing the napi_struct from
the poll_list. However I think this is probably a small price to pay, given
that the alternative is a crash.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves neigh_setup and hard_start_xmit into the network device ops
structure. For bisection, fix all the previously converted drivers as well.
Bonding driver took the biggest hit on this.
Added a prefetch of the hard_start_xmit in the fast path to try and reduce
any impact this would have.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch changes the network device internal API to move adminstrative
operations out of the network device structure and into a separate structure.
This patch involves some hackery to maintain compatablity between the
new and old model, so all 300+ drivers don't have to be changed at once.
For drivers that aren't converted yet, the netdevice_ops virt function list
still resides in the net_device structure. For old protocols, the new
net_device_ops are copied out to the old net_device pointers.
After the transistion is completed the nag message can be changed to
an WARN_ON, and the compatiablity code can be made configurable.
Some function pointers aren't moved:
* destructor can't be in net_device_ops because
it may need to be referenced after the module is unloaded.
* neighbor setup is manipulated in a couple of places that need special
consideration
* hard_start_xmit is in the fast path for transmit.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The first argument to csum_partial is const void *
casts to char/u8 * are not necessary
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This converts pretty much everything to print_mac. There were
a few things that had conflicts which I have just dropped for
now, no harm done.
I've built an allyesconfig with this and looked at the files
that weren't built very carefully, but it's a huge patch.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
When support for multiple TX queues were added, the
netif_tx_lock() routines we converted to iterate over
all TX queues and grab each queue's spinlock.
This causes heartburn for lockdep and it's not a healthy
thing to do with lots of TX queues anyways.
So modify this to use a top-level lock and a "frozen"
state for the individual TX queues.
Signed-off-by: David S. Miller <davem@davemloft.net>
This effectively "flips the switch" by making the core networking
and multiqueue-aware drivers use the new TX multiqueue structures.
Non-multiqueue drivers need no changes. The interfaces they use such
as netif_stop_queue() degenerate into an operation on TX queue zero.
So everything "just works" for them.
Code that really wants to do "X" to all TX queues now invokes a
routine that does so, such as netif_tx_wake_all_queues(),
netif_tx_stop_all_queues(), etc.
pktgen and netpoll required a little bit more surgery than the others.
In particular the pktgen changes, whilst functional, could be largely
improved. The initial check in pktgen_xmit() will sometimes check the
wrong queue, which is mostly harmless. The thing to do is probably to
invoke fill_packet() earlier.
The bulk of the netpoll changes is to make the code operate solely on
the TX queue indicated by by the SKB queue mapping.
Setting of the SKB queue mapping is entirely confined inside of
net/core/dev.c:dev_pick_tx(). If we end up needing any kind of
special semantics (drops, for example) it will be implemented here.
Finally, we now have a "real_num_tx_queues" which is where the driver
indicates how many TX queues are actually active.
With IGB changes from Jeff Kirsher.
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds needed_headroom/needed_tailroom members to struct
net_device and updates many places that allocate sbks to use them. Not
all of them can be converted though, and I'm sure I missed some (I
mostly grepped for LL_RESERVED_SPACE)
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
zap_completion_queue() retrieves skbs from completion_queue where they have
zero skb->users counter. Before dev_kfree_skb_any() it should be non-zero
yet, so it's increased now.
Reported-and-tested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based upon a report by Andrew Morton and code analysis done
by Jarek Poplawski.
This reverts 33f807ba0d ("[NETPOLL]:
Kill NETPOLL_RX_DROP, set but never tested.") and
c7b6ea24b4 ("[NETPOLL]: Don't need
rx_flags.").
The rx_flags did get tested for zero vs. non-zero and therefore we do
need those tests and that code which sets NETPOLL_RX_DROP et al.
Signed-off-by: David S. Miller <davem@davemloft.net>
There are some place, that calculate the ARP header length. These
calculations are correct, but
a) some operate with "magic" constants,
b) enlarge the code length (sometimes at the cost of coding style),
c) are not informative from the first glance.
The proposal is to introduce a helper, that includes all the good
sides of these calculations.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The rx_flags variable is redundant. Turning rx on/off is done
via setting the rx_np pointer.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The local_mac is managed by the network device, no need to keep a
spare copy and all the management problems that could cause.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Restructure code slightly to improve readability:
* dereference device once
* change obvious while() loop
* let poll_napi() handle null list itself
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use standard routine for flushing queue.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
netpoll_poll_lock() synchronizes the ->poll() invocation
code paths, but once we have the lock we have to make
sure that NAPI_STATE_SCHED is still set. Otherwise we
get:
cpu 0 cpu 1
net_rx_action() poll_napi()
netpoll_poll_lock() ... spin on ->poll_lock
->poll()
netif_rx_complete
netpoll_poll_unlock() acquire ->poll_lock()
->poll()
netif_rx_complete()
CRASH
Based upon a bug report from Tina Yang.
Signed-off-by: David S. Miller <davem@davemloft.net>
Many places get the queue_mapping field from skb to pass it to the
netif_subqueue_stopped() which will be 0 in any case.
Make the helper that works with sk_buff
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add inline for common usage of hardware header creation, and
fix bug in IPV6 mcast where the assumption about negative return is
an errno. Negative return from hard_header means not enough space
was available,(ie -N bytes).
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes most of the generic device layer network
namespace safe. This patch makes dev_base_head a
network namespace variable, and then it picks up
a few associated variables. The functions:
dev_getbyhwaddr
dev_getfirsthwbytype
dev_get_by_flags
dev_get_by_name
__dev_get_by_name
dev_get_by_index
__dev_get_by_index
dev_ioctl
dev_ethtool
dev_load
wireless_process_ioctl
were modified to take a network namespace argument, and
deal with it.
vlan_ioctl_set and brioctl_set were modified so their
hooks will receive a network namespace argument.
So basically anthing in the core of the network stack that was
affected to by the change of dev_base was modified to handle
multiple network namespaces. The rest of the network stack was
simply modified to explicitly use &init_net the initial network
namespace. This can be fixed when those components of the network
stack are modified to handle multiple network namespaces.
For now the ifindex generator is left global.
Fundametally ifindex numbers are per namespace, or else
we will have corner case problems with migration when
we get that far.
At the same time there are assumptions in the network stack
that the ifindex of a network device won't change. Making
the ifindex number global seems a good compromise until
the network stack can cope with ifindex changes when
you change namespaces, and the like.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based upon initial work by Keiichi Kii <k-keiichi@bx.jp.nec.com>.
This patch introduces support for dynamic reconfiguration (adding, removing
and/or modifying parameters of netconsole targets at runtime) using a
userspace interface exported via configfs. Documentation is also updated
accordingly.
Issues and brief design overview:
(1) Kernel-initiated creation / destruction of kernel objects is not
possible with configfs -- the lifetimes of the "config items" is managed
exclusively from userspace. But netconsole must support boot/module
params too, and these are parsed in kernel and hence netpolls must be
setup from the kernel. Joel Becker suggested to separately manage the
lifetimes of the two kinds of netconsole_target objects -- those created
via configfs mkdir(2) from userspace and those specified from the
boot/module option string. This adds complexity and some redundancy here
and also means that boot/module param-created targets are not exposed
through the configfs namespace (and hence cannot be updated / destroyed
dynamically). However, this saves us from locking / refcounting
complexities that would need to be introduced in configfs to support
kernel-initiated item creation / destroy there.
(2) In configfs, item creation takes place in the call chain of the
mkdir(2) syscall in the driver subsystem. If we used an ioctl(2) to
create / destroy objects from userspace, the special userspace program is
able to fill out the structure to be passed into the ioctl and hence
specify attributes such as local interface that are required at the time
we set up the netpoll. For configfs, this information is not available at
the time of mkdir(2). So, we keep all newly-created targets (via
configfs) disabled by default. The user is expected to set various
attributes appropriately (including the local network interface if
required) and then write(2) "1" to the "enabled" attribute. Thus,
netpoll_setup() is then called on the set parameters in the context of
_this_ write(2) on the "enabled" attribute itself. This design enables
the user to reconfigure existing netconsole targets at runtime to be
attached to newly-come-up interfaces that may not have existed when
netconsole was loaded or when the targets were actually created. All this
effectively enables us to get rid of custom ioctls.
(3) Ultra-paranoid configfs attribute show() and store() operations, with
sanity and input range checking, using only safe string primitives, and
compliant with the recommendations in Documentation/filesystems/sysfs.txt.
(4) A new function netpoll_print_options() is created in the netpoll API,
that just prints out the configured parameters for a netpoll structure.
netpoll_parse_options() is modified to use that and it is also exported to
be used from netconsole.
Signed-off-by: Satyam Sharma <satyam@infradead.org>
Acked-by: Keiichi Kii <k-keiichi@bx.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.
In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.
The signature of the ->poll() call back goes from:
int foo_poll(struct net_device *dev, int *budget)
to
int foo_poll(struct napi_struct *napi, int budget)
The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract). The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.
The napi_struct is to be embedded in the device driver private data
structures.
Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler. Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.
With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.
Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.
[ Ported to current tree and all drivers converted. Integrated
Stephen's follow-on kerneldoc additions, and restored poll_list
handling to the old style to fix mutual exclusion issues. -DaveM ]
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 29578624e3.
Ingo Molnar reports complete breakage with his e1000 card (no
networking, card reports transmit timeouts), and bisected it down to
this commit. Let's figure out what went wrong, but not keep breaking
machines until we do.
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Olaf Kirch <olaf.kirch@oracle.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Keep netpoll/poll_napi from messing with the poll_list.
Only net_rx_action is allowed to manipulate the list.
Signed-off-by: Olaf Kirch <olaf.kirch@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
93ec2c723e applied excessive duct tape to
the netpoll beast's netpoll_cleanup(), thus substituting one leak with
another, and opening up a little buglet :-)
net_device->npinfo (netpoll_info) is a shared and refcounted object and
cannot simply be set NULL the first time netpoll_cleanup() is called.
Otherwise, further netpoll_cleanup()'s see np->dev->npinfo == NULL and
become no-ops, thus leaking. And it's a bug too: the first call to
netpoll_cleanup() would thus (annoyingly) "disable" other (still alive)
netpolls too. Maybe nobody noticed this because netconsole (only user
of netpoll) never supported multiple netpoll objects earlier.
This is a trivial and obvious one-line fixlet.
Signed-off-by: Satyam Sharma <ssatyam@cse.iitk.ac.in>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the multiqueue hardware device support API to the core network
stack. Allow drivers to allocate multiple queues and manage them at
the netdev level if they choose to do so.
Added a new field to sk_buff, namely queue_mapping, for drivers to
know which tx_ring to select based on OS classification of the flow.
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
>From my recent patch:
> > #1
> > Until kernel ver. 2.6.21 (including) cancel_rearming_delayed_work()
> > required a work function should always (unconditionally) rearm with
> > delay > 0 - otherwise it would endlessly loop. This patch replaces
> > this function with cancel_delayed_work(). Later kernel versions don't
> > require this, so here it's only for uniformity.
But Oleg Nesterov <oleg@tv-sign.ru> found:
> But 2.6.22 doesn't need this change, why it was merged?
>
> In fact, I suspect this change adds a race,
...
His description was right (thanks), so this patch reverts #1.
Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
#1
Until kernel ver. 2.6.21 (including) cancel_rearming_delayed_work()
required a work function should always (unconditionally) rearm with
delay > 0 - otherwise it would endlessly loop. This patch replaces
this function with cancel_delayed_work(). Later kernel versions don't
require this, so here it's only for uniformity.
#2
After deleting a timer in cancel_[rearming_]delayed_work() there could
stay a last skb queued in npinfo->txq causing a memory leak after
kfree(npinfo).
Initial patch & testing by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If sky2 device poll routine is called from netpoll_send_skb, it would
deadlock. The netpoll_send_skb held the netif_tx_lock, and the poll
routine could acquire it to clean up skb's. Other drivers might use
same locking model.
The driver is correct, netpoll should not introduce more locking
problems than it causes already. So change the code to drop lock
before calling poll handler.
Signed-off-by: Stephen Hemminger <shemminger@linux.foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove includes of <linux/smp_lock.h> where it is not used/needed.
Suggested by Al Viro.
Builds cleanly on x86_64, i386, alpha, ia64, powerpc, sparc,
sparc64, and arm (all 59 defconfigs).
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When a transmitted packet is looped back directly, CHECKSUM_PARTIAL
maps to the semantics of CHECKSUM_UNNECESSARY. Therefore we should
treat it as such in the stack.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
To clearly state the intent of copying to linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
For the common, open coded 'skb->h.raw = skb->data' operation, so that we can
later turn skb->h.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple cases:
skb->h.raw = skb->data;
skb->h.raw = {skb_push|[__]skb_pull}()
The next ones will handle the slightly more "complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
skb_push updates and returns skb->data, so we can just call
skb_reset_network_header after the call to skb_push.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->nh.raw = skb->data' operation, so that we can
later turn skb->nh.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->mac.raw = skb->data' operation, so that we can
later turn skb->mac.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In net poll mode, the current checksum function doesn't consider the
kind of packet which is padded to reach a specific minimum length. I
believe that's the problem causing my test case failed. The following
patch fixed this issue.
Signed-off-by: Aubrey.Li <aubreylee@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
current -git doesnt boot on my laptop due to netpoll not unlocking the
tx lock in the else branch.
booted this up on my laptop with lockdep enabled and there are no
locking complaints and it works fine.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
During boot we get:
netconsole: device eth0 not up yet, forcing it
e1000: eth0: e1000_watchdog: NIC Link is Up 100 Mbps Full Duplex
WARNING (!__warned) at kernel/softirq.c:137 local_bh_enable()
Call Trace:
[<ffffffff80235baf>] local_bh_enable+0x41/0xa3
[<ffffffff8045ab8e>] netpoll_send_skb+0x116/0x144
[<ffffffff8045b1ee>] netpoll_send_udp+0x263/0x271
[<ffffffff803d41ec>] write_msg+0x42/0x5e
[<ffffffff80230c9b>] __call_console_drivers+0x5f/0x70
[<ffffffff80230d19>] _call_console_drivers+0x6d/0x71
[<ffffffff802313f0>] release_console_sem+0x148/0x1ec
[<ffffffff802316ce>] register_console+0x1b1/0x1ba
[<ffffffff803d4178>] init_netconsole+0x54/0x68
[<ffffffff802071ae>] init+0x152/0x308
[<ffffffff804dac8b>] _spin_unlock_irq+0x14/0x30
[<ffffffff8022c15e>] schedule_tail+0x43/0x9f
[<ffffffff8020a758>] child_rip+0xa/0x12
Herbert sayeth:
Normally networking isn't invoked with interrupts turned off, but I
suppose we don't have a choice here. This is unique being a place where you
can get called with BH on, off, or IRQs off.
Given that this is only used for printk, the easiest solution is probably
just to disable local IRQs instead of BH.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
