As it is, all instances of ->release() for files that have ->fasync()
need to remember to evict file from fasync lists; forgetting that
creates a hole and we actually have a bunch that *does* forget.
So let's keep our lives simple - let __fput() check FASYNC in
file->f_flags and call ->fasync() there if it's been set. And lose that
crap in ->release() instances - leaving it there is still valid, but we
don't have to bother anymore.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
skb_alloc produces linear packets (using kmalloc()). That can fail,
so should we fall back to making paged skbs.
My original version of this patch always allocate paged skbs for big
packets. But that made performance drop from 8.4 seconds to 8.8
seconds on 1G lguest->Host TCP xmit. So now we only do that as a
fallback.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a TUNGETIFF interface so that userspace can query a
tun/tap descriptor for its name and flags.
This is needed because it is common for one app to create
a tap interface, exec another app and pass it the file
descriptor for the interface. Without TUNGETIFF the spawned
app has no way of detecting wheter the interface has e.g.
IFF_VNET_HDR set.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Please see the following thread to get some context on this
http://marc.info/?l=linux-netdev&m=121564433018903&w=2
Basically the issue is that current multi-cast filtering stuff in
the TUN/TAP driver is seriously broken.
Original patch went in without proper review and ACK. It was broken and
confusing to start with and subsequent patches broke it completely.
To give you an idea of what's broken here are some of the issues:
- Very confusing comments throughout the code that imply that the
character device is a network interface in its own right, and that packets
are passed between the two nics. Which is completely wrong.
- Wrong set of ioctls is used for setting up filters. They look like
shortcuts for manipulating state of the tun/tap network interface but
in reality manipulate the state of the TX filter.
- ioctls that were originally used for setting address of the the TX filter
got "fixed" and now set the address of the network interface itself. Which
made filter totaly useless.
- Filtering is done too late. Instead of filtering early on, to avoid
unnecessary wakeups, filtering is done in the read() call.
The list goes on and on :)
So the patch cleans all that up. It introduces simple and clean interface for
setting up TX filters (TUNSETTXFILTER + tun_filter spec) and does filtering
before enqueuing the packets.
TX filtering is useful in the scenarios where TAP is part of a bridge, in
which case it gets all broadcast, multicast and potentially other packets when
the bridge is learning. So for example Ethernet tunnelling app may want to
setup TX filters to avoid tunnelling multicast traffic. QEMU and other
hypervisors can push RX filtering that is currently done in the guest into the
host context therefore saving wakeups and unnecessary data transfer.
Signed-off-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The scenario goes like this. App stops reading from tun/tap.
TX queue gets full and driver does netif_stop_queue().
App closes fd and TX queue gets flushed as part of the cleanup.
Next time the app opens tun/tap and starts reading from it but
the xoff state is not cleared. We're stuck.
Normally xoff state is cleared when netdev is brought up. But
in the case of persistent devices this happens only during
initial setup.
The fix is trivial. If device is already up when an app opens
it we clear xoff state and that gets things moving again.
Signed-off-by: Max Krasnyansky <maxk@qualcomm.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a IFF_VNET_HDR flag. This uses the same ABI as virtio_net
(ie. prepending struct virtio_net_hdr to packets) to indicate GSO and
checksum information.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ethtool is useful for setting (some) device fields, but it's
root-only. Finer feature control is available through a tun-specific
ioctl.
(Includes Mark McLoughlin <markmc@redhat.com>'s fix to hold rtnl sem).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The problem with introducing checksum offload and gso to tun is they
need to set dev->features to enable GSO and/or checksumming, which is
supposed to be done before register_netdevice(), ie. as part of
TUNSETIFF.
Unfortunately, TUNSETIFF has always just ignored flags it doesn't
understand, so there's no good way of detecting whether the kernel
supports new IFF_ flags.
This patch implements a TUNGETFEATURES ioctl which returns all the valid IFF
flags. It could be extended later to include other features.
Here's an example program which uses it:
#include <linux/if_tun.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <err.h>
#include <stdio.h>
static struct {
unsigned int flag;
const char *name;
} known_flags[] = {
{ IFF_TUN, "TUN" },
{ IFF_TAP, "TAP" },
{ IFF_NO_PI, "NO_PI" },
{ IFF_ONE_QUEUE, "ONE_QUEUE" },
};
int main()
{
unsigned int features, i;
int netfd = open("/dev/net/tun", O_RDWR);
if (netfd < 0)
err(1, "Opening /dev/net/tun");
if (ioctl(netfd, TUNGETFEATURES, &features) != 0) {
printf("Kernel does not support TUNGETFEATURES, guessing\n");
features = (IFF_TUN|IFF_TAP|IFF_NO_PI|IFF_ONE_QUEUE);
}
printf("Available features are: ");
for (i = 0; i < sizeof(known_flags)/sizeof(known_flags[0]); i++) {
if (features & known_flags[i].flag) {
features &= ~known_flags[i].flag;
printf("%s ", known_flags[i].name);
}
}
if (features)
printf("(UNKNOWN %#x)", features);
printf("\n");
return 0;
}
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
By default, tun.c running in TUN_TUN_DEV mode will set the protocol of
packet to IPv4 if TUN_NO_PI is set. My program failed to work when I
assumed that the driver will check the first nibble of packet,
determine IP version and set the appropriate protocol.
Signed-off-by: Ang Way Chuang <wcang@nav6.org>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since these operations don't go through the normal
device calls, we have to ensure we synchronize with
those paths.
Noticed by Alan Cox.
Signed-off-by: David S. Miller <davem@davemloft.net>
Noticed by Alan Cox.
The IFF_UP test is a bit racey, because other entities
outside of this driver's ioctl handler can modify that
state, even though this ioctl handler runs under
lock_kernel().
Signed-off-by: David S. Miller <davem@davemloft.net>
This is basically means that a net is set for a new device, but
actually also involves two more steps:
1. mark the tun device as "local", i.e. do not allow for it to
move across namespaces.
This is done so, since tun device is most often associated to some
file (and thus to some process) and moving the device alone is not
valid while keeping the file and the process outside. The need in
ability to move a detached persistent device is to be investigated
later.
2. get the tun device's net when tun becomes attached and put one
when it becomes detached.
This is needed to handle the case when a task owning the tun dies,
but a files lives for some more time - in this case we must not
allow for net to be freed, since its exit hook will spoil that file's
private data by unregistering the tun from under tun_chr_close.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove the static tun_dev_list and replace its occurrences in
driver with per-net one.
It is used in two places - in tun_set_iff and tun_cleanup. In
the first case it's legal to use current net_ns. In the cleanup
call - move the loop, that unregisters all devices in net exit
hook.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is the first step in making tuntap devices work in net
namespaces. The structure mentioned is pointed by generic
net pointer with tun_net_id id, and tun driver fills one on
its load. It will contain only the tun devices list.
So declare this structure and introduce net init and exit hooks.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the user gives a packet under 14 bytes, we'll end up reading off the end
of the skb (not oopsing, just reading off the end).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyanskiy <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's no reason for this to be in the header, and it just hurts
recompile time.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Max Krasnyanskiy <maxk@qualcomm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current tun/tap driver sets also net device's hw address when asked to
change character device's hw address. This is a good idea, but it
misses RTLN-locking, resulting following error message in 2.6.25-rc3's
inetdev_event() function:
RTNL: assertion failed at net/ipv4/devinet.c (1050)
Attached patch fixes this problem.
Signed-off-by: Kim B. Heino <Kim.Heino@bluegiga.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
From: "Nathaniel Filardo" <nwfilardo@gmail.com>
Taken from http://bugzilla.kernel.org/show_bug.cgi?id=9806
The TUN/TAP driver only permits one-way transitions of IFF_NO_PI or
IFF_ONE_QUEUE during the lifetime of a tap/tun interface. Note that
tun_set_iff contains
541 if (ifr->ifr_flags & IFF_NO_PI)
542 tun->flags |= TUN_NO_PI;
543
544 if (ifr->ifr_flags & IFF_ONE_QUEUE)
545 tun->flags |= TUN_ONE_QUEUE;
This is easily fixed by adding else branches which clear these bits.
Steps to reproduce:
This is easily reproduced by setting an interface persistant using tunctl then
attempting to open it as IFF_TAP or IFF_TUN, without asserting the IFF_NO_PI
flag. The ioctl() will succeed and the ifr.flags word is not modified, but the
interface remains in IFF_NO_PI mode (as it was set by tunctl).
Acked-by: Maxim Krasnyansky <maxk@qualcomm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use iov_length() instead of tun's homemade iov_total().
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a trivial fix of debug message.
When a persist flag is set, the message should say "enabled".
Signed-off-by: Toyo Abe <tabe@miraclelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We now have struct net_device_stats embedded in struct net_device,
and the default ->get_stats() hook does the obvious thing for us.
Run through drivers/net/* and remove the driver-local storage of
statistics, and driver-local ->get_stats() hook where applicable.
This was just the low-hanging fruit in drivers/net; plenty more drivers
remain to be updated.
[ Resolved conflicts with napi_struct changes and fix sunqe build
regression... -DaveM ]
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's been a useless no-op for long enough in 2.6 so I figured it's time to
remove it. The number of people that could object because they're
maintaining unified 2.4 and 2.6 drivers is probably rather small.
[ Handled drivers added by netdev tree and some missed IRDA cases... -DaveM ]
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Jeff Garzik <jeff@garzik.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>
Introduce a new syscall TUNSETGROUP for group ownership setting of tap
devices. The user now is allowed to send packages if either his euid or
his egid matches the one specified via tunctl (via -u or -g
respecitvely). If both, gid and uid, are set via tunctl, both have to
match.
Signed-off-by: Guido Guenther <agx@sigxcpu.org>
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixed tun/tap driver's handling of hw addresses. The hw address is stored
in both the net_device.dev_addr and tun.dev_addr fields. These fields were
not kept synchronized, and in fact weren't even initialized to the same
value. Now during both init and when performing SIOCSIFHWADDR on the tun
device these values are both updated. However, if SIOCSIFHWADDR is
performed on the net device directly (for instance, setting the hw address
using ifconfig), the tun device does not get updated. Perhaps the
tun.dev_addr field should be removed completely at some point, as it is
redundant and net_device.dev_addr can be used anywhere it is used.
Signed-off-by: Brian Braunstein <linuxkernel@bristyle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
To clearly state the intent of copying from 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@redhat.com>
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>
One less thing for drivers writers to worry about.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Many struct file_operations in the kernel can be "const". Marking them const
moves these to the .rodata section, which avoids false sharing with potential
dirty data. In addition it'll catch accidental writes at compile time to
these shared resources.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
File handles can be requested to send sigio and sigurg to processes. By
tracking the destination processes using struct pid instead of pid_t we make
the interface safe from all potential pid wrap around problems.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch removes readv() and writev() methods and replaces them with
aio_read()/aio_write() methods.
Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The tuntap driver allows an admin to create persistent devices and
assign ownership of them to individual users. Unfortunately, relaxing
the permissions on the /dev/net/tun device node so that they can
actually use those devices will _also_ allow those users to create
arbitrary new devices of their own. This patch corrects that, and
adjusts the recommended permissions for the device node accordingly.
Signed-off-By: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
We're leaking an skb in a failure path in this function.
Coverity #632
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently tun/tap only supports the EN10MB ARP type. For use with
wireless and other networking types it should be possible to set the
ARP type via an ioctl.
Patch v2: Included check that the tap interface is down before changing the
link type out from underneath it
Signed-off-by: Mike Kershaw <dragorn@kismetwireless.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is part of the grand scheme to eliminate the qlen
member of skb_queue_head, and subsequently remove the
'list' member of sk_buff.
Most users of skb_queue_len() want to know if the queue is
empty or not, and that's trivially done with skb_queue_empty()
which doesn't use the skb_queue_head->qlen member and instead
uses the queue list emptyness as the test.
Signed-off-by: David S. Miller <davem@davemloft.net>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
