Prior the state variable was introduced by Ilya, the dev member was
used to determine whether the socket was bound or not. However, when
dev was read, proper SMP barriers and READ_ONCE were missing. In order
to address the missing barriers and READ_ONCE, we start using the
state variable as a point of synchronization. The state member
read/write is paired with proper SMP barriers, and from this follows
that the members described above does not need READ_ONCE if used in
conjunction with state check.
In all syscalls and the xsk_rcv path we check if state is
XSK_BOUND. If that is the case we do a SMP read barrier, and this
implies that the dev, umem and all rings are correctly setup. Note
that no READ_ONCE are needed for these variable if used when state is
XSK_BOUND (plus the read barrier).
To summarize: The members struct xdp_sock members dev, queue_id, umem,
fq, cq, tx, rx, and state were read lock-less, with incorrect barriers
and missing {READ, WRITE}_ONCE. Now, umem, fq, cq, tx, rx, and state
are read lock-less. When these members are updated, WRITE_ONCE is
used. When read, READ_ONCE are only used when read outside the control
mutex (e.g. mmap) or, not synchronized with the state member
(XSK_BOUND plus smp_rmb())
Note that dev and queue_id do not need a WRITE_ONCE or READ_ONCE, due
to the introduce state synchronization (XSK_BOUND plus smp_rmb()).
Introducing the state check also fixes a race, found by syzcaller, in
xsk_poll() where umem could be accessed when stale.
Suggested-by: Hillf Danton <hdanton@sina.com>
Reported-by: syzbot+c82697e3043781e08802@syzkaller.appspotmail.com
Fixes: 77cd0d7b3f ("xsk: add support for need_wakeup flag in AF_XDP rings")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The umem member of struct xdp_sock is read outside of the control
mutex, in the mmap implementation, and needs a WRITE_ONCE to avoid
potential store-tearing.
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Fixes: 423f38329d ("xsk: add umem fill queue support and mmap")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Use WRITE_ONCE when doing the store of tx, rx, fq, and cq, to avoid
potential store-tearing. These members are read outside of the control
mutex in the mmap implementation.
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Fixes: 37b076933a ("xsk: add missing write- and data-dependency barrier")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Currently, addresses are chunk size aligned. This means, we are very
restricted in terms of where we can place chunk within the umem. For
example, if we have a chunk size of 2k, then our chunks can only be placed
at 0,2k,4k,6k,8k... and so on (ie. every 2k starting from 0).
This patch introduces the ability to use unaligned chunks. With these
changes, we are no longer bound to having to place chunks at a 2k (or
whatever your chunk size is) interval. Since we are no longer dealing with
aligned chunks, they can now cross page boundaries. Checks for page
contiguity have been added in order to keep track of which pages are
followed by a physically contiguous page.
Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>
Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
When an AF_XDP socket is released/closed the XSKMAP still holds a
reference to the socket in a "released" state. The socket will still
use the netdev queue resource, and block newly created sockets from
attaching to that queue, but no user application can access the
fill/complete/rx/tx queues. This results in that all applications need
to explicitly clear the map entry from the old "zombie state"
socket. This should be done automatically.
In this patch, the sockets tracks, and have a reference to, which maps
it resides in. When the socket is released, it will remove itself from
all maps.
Suggested-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This commit adds support for a new flag called need_wakeup in the
AF_XDP Tx and fill rings. When this flag is set, it means that the
application has to explicitly wake up the kernel Rx (for the bit in
the fill ring) or kernel Tx (for bit in the Tx ring) processing by
issuing a syscall. Poll() can wake up both depending on the flags
submitted and sendto() will wake up tx processing only.
The main reason for introducing this new flag is to be able to
efficiently support the case when application and driver is executing
on the same core. Previously, the driver was just busy-spinning on the
fill ring if it ran out of buffers in the HW and there were none on
the fill ring. This approach works when the application is running on
another core as it can replenish the fill ring while the driver is
busy-spinning. Though, this is a lousy approach if both of them are
running on the same core as the probability of the fill ring getting
more entries when the driver is busy-spinning is zero. With this new
feature the driver now sets the need_wakeup flag and returns to the
application. The application can then replenish the fill queue and
then explicitly wake up the Rx processing in the kernel using the
syscall poll(). For Tx, the flag is only set to one if the driver has
no outstanding Tx completion interrupts. If it has some, the flag is
zero as it will be woken up by a completion interrupt anyway.
As a nice side effect, this new flag also improves the performance of
the case where application and driver are running on two different
cores as it reduces the number of syscalls to the kernel. The kernel
tells user space if it needs to be woken up by a syscall, and this
eliminates many of the syscalls.
This flag needs some simple driver support. If the driver does not
support this, the Rx flag is always zero and the Tx flag is always
one. This makes any application relying on this feature default to the
old behaviour of not requiring any syscalls in the Rx path and always
having to call sendto() in the Tx path.
For backwards compatibility reasons, this feature has to be explicitly
turned on using a new bind flag (XDP_USE_NEED_WAKEUP). I recommend
that you always turn it on as it so far always have had a positive
performance impact.
The name and inspiration of the flag has been taken from io_uring by
Jens Axboe. Details about this feature in io_uring can be found in
http://kernel.dk/io_uring.pdf, section 8.3.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This commit replaces ndo_xsk_async_xmit with ndo_xsk_wakeup. This new
ndo provides the same functionality as before but with the addition of
a new flags field that is used to specifiy if Rx, Tx or both should be
woken up. The previous ndo only woke up Tx, as implied by the
name. The i40e and ixgbe drivers (which are all the supported ones)
are updated with this new interface.
This new ndo will be used by the new need_wakeup functionality of XDP
sockets that need to be able to wake up both Rx and Tx driver
processing.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
There are 2 call chains:
a) xsk_bind --> xdp_umem_assign_dev
b) unregister_netdevice_queue --> xsk_notifier
with the following locking order:
a) xs->mutex --> rtnl_lock
b) rtnl_lock --> xdp.lock --> xs->mutex
Different order of taking 'xs->mutex' and 'rtnl_lock' could produce a
deadlock here. Fix that by moving the 'rtnl_lock' before 'xs->lock' in
the bind call chain (a).
Reported-by: syzbot+bf64ec93de836d7f4c2c@syzkaller.appspotmail.com
Fixes: 455302d1c9 ("xdp: fix hang while unregistering device bound to xdp socket")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Completion queue address reservation could not be undone.
In case of bad 'queue_id' or skb allocation failure, reserved entry
will be leaked reducing the total capacity of completion queue.
Fix that by moving reservation to the point where failure is not
possible. Additionally, 'queue_id' checking moved out from the loop
since there is no point to check it there.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Tested-by: William Tu <u9012063@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Unlike driver mode, generic xdp receive could be triggered
by different threads on different CPU cores at the same time
leading to the fill and rx queue breakage. For example, this
could happen while sending packets from two processes to the
first interface of veth pair while the second part of it is
open with AF_XDP socket.
Need to take a lock for each generic receive to avoid race.
Fixes: c497176cb2 ("xsk: add Rx receive functions and poll support")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>
Tested-by: William Tu <u9012063@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Device that bound to XDP socket will not have zero refcount until the
userspace application will not close it. This leads to hang inside
'netdev_wait_allrefs()' if device unregistering requested:
# ip link del p1
< hang on recvmsg on netlink socket >
# ps -x | grep ip
5126 pts/0 D+ 0:00 ip link del p1
# journalctl -b
Jun 05 07:19:16 kernel:
unregister_netdevice: waiting for p1 to become free. Usage count = 1
Jun 05 07:19:27 kernel:
unregister_netdevice: waiting for p1 to become free. Usage count = 1
...
Fix that by implementing NETDEV_UNREGISTER event notification handler
to properly clean up all the resources and unref device.
This should also allow socket killing via ss(8) utility.
Fixes: 965a990984 ("xsk: add support for bind for Rx")
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Some drivers want to access the data transmitted in order to implement
acceleration features of the NICs. It is also useful in AF_XDP TX flow.
Change the xsk_umem_consume_tx API to return the whole xdp_desc, that
contains the data pointer, length and DMA address, instead of only the
latter two. Adapt the implementation of i40e and ixgbe to this change.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Cc: Björn Töpel <bjorn.topel@intel.com>
Cc: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Make it possible for the application to determine whether the AF_XDP
socket is running in zero-copy mode. To achieve this, add a new
getsockopt option XDP_OPTIONS that returns flags. The only flag
supported for now is the zero-copy mode indicator.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Add a function that checks whether the Fill Ring has the specified
amount of descriptors available. It will be useful for mlx5e that wants
to check in advance, whether it can allocate a bulk of RX descriptors,
to get the best performance.
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Passing a non-existing flag in the sxdp_flags member of struct
sockaddr_xdp was, incorrectly, silently ignored. This patch addresses
that behavior, and rejects any non-existing flags.
We have examined existing user space code, and to our best knowledge,
no one is relying on the current incorrect behavior. AF_XDP is still
in its infancy, so from our perspective, the risk of breakage is very
low, and addressing this problem now is important.
Fixes: 965a990984 ("xsk: add support for bind for Rx")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Three conflicts, one of which, for marvell10g.c is non-trivial and
requires some follow-up from Heiner or someone else.
The issue is that Heiner converted the marvell10g driver over to
use the generic c45 code as much as possible.
However, in 'net' a bug fix appeared which makes sure that a new
local mask (MDIO_AN_10GBT_CTRL_ADV_NBT_MASK) with value 0x01e0
is cleared.
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit e2ce367488.
It turns out that the sock destructor xsk_destruct was needed after
all. The cleanup simplification broke the skb transmit cleanup path,
due to that the umem was prematurely destroyed.
The umem cannot be destroyed until all outstanding skbs are freed,
which means that we cannot remove the umem until the sk_destruct has
been called.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
All the setup code in AF_XDP is protected by a mutex with the
exception of the mmap code that cannot use it. To make sure that a
process banging on the mmap call at the same time as another process
is setting up the socket, smp_wmb() calls were added in the umem
registration code and the queue creation code, so that the published
structures that xsk_mmap needs would be consistent. However, the
corresponding smp_rmb() calls were not added to the xsk_mmap
code. This patch adds these calls.
Fixes: 37b076933a ("xsk: add missing write- and data-dependency barrier")
Fixes: c0c77d8fb7 ("xsk: add user memory registration support sockopt")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch adds the sock_diag interface for querying sockets from user
space. Tools like iproute2 ss(8) can use this interface to list open
AF_XDP sockets.
The user-space ABI is defined in linux/xdp_diag.h and includes netlink
request and response structs. The request can query sockets and the
response contains socket information about the rings, umems, inode and
more.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Track each AF_XDP socket in a per-netns list. This will be used later
by the sock_diag interface for querying sockets from userspace.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Prior this commit, when the struct socket object was being released,
the UMEM did not have its reference count decreased. Instead, this was
done in the struct sock sk_destruct function.
There is no reason to keep the UMEM reference around when the socket
is being orphaned, so in this patch the xdp_put_mem is called in the
xsk_release function. This results in that the xsk_destruct function
can be removed!
Note that, it still holds that a struct xsk_sock reference might still
linger in the XSKMAP after the UMEM is released, e.g. if a user does
not clear the XSKMAP prior to closing the process. This sock will be
in a "released" zombie like state, until the XSKMAP is removed.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
net/sched/cls_api.c has overlapping changes to a call to
nlmsg_parse(), one (from 'net') added rtm_tca_policy instead of NULL
to the 5th argument, and another (from 'net-next') added cb->extack
instead of NULL to the 6th argument.
net/ipv4/ipmr_base.c is a case of a bug fix in 'net' being done to
code which moved (to mr_table_dump)) in 'net-next'. Thanks to David
Ahern for the heads up.
Signed-off-by: David S. Miller <davem@davemloft.net>
The XSKMAP update and delete functions called synchronize_net(), which
can sleep. It is not allowed to sleep during an RCU read section.
Instead we need to make sure that the sock sk_destruct (xsk_destruct)
function is asynchronously called after an RCU grace period. Setting
the SOCK_RCU_FREE flag for XDP sockets takes care of this.
Fixes: fbfc504a24 ("bpf: introduce new bpf AF_XDP map type BPF_MAP_TYPE_XSKMAP")
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The AF_XDP socket struct can exist in three different, implicit
states: setup, bound and released. Setup is prior the socket has been
bound to a device. Bound is when the socket is active for receive and
send. Released is when the process/userspace side of the socket is
released, but the sock object is still lingering, e.g. when there is a
reference to the socket in an XSKMAP after process termination.
The Rx fast-path code uses the "dev" member of struct xdp_sock to
check whether a socket is bound or relased, and the Tx code uses the
struct xdp_umem "xsk_list" member in conjunction with "dev" to
determine the state of a socket.
However, the transition from bound to released did not tear the socket
down in correct order.
On the Rx side "dev" was cleared after synchronize_net() making the
synchronization useless. On the Tx side, the internal queues were
destroyed prior removing them from the "xsk_list".
This commit corrects the cleanup order, and by doing so
xdp_del_sk_umem() can be simplified and one synchronize_net() can be
removed.
Fixes: 965a990984 ("xsk: add support for bind for Rx")
Fixes: ac98d8aab6 ("xsk: wire upp Tx zero-copy functions")
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Previously, the xsk code did not record which umem was bound to a
specific queue id. This was not required if all drivers were zero-copy
enabled as this had to be recorded in the driver anyway. So if a user
tried to bind two umems to the same queue, the driver would say
no. But if copy-mode was first enabled and then zero-copy mode (or the
reverse order), we mistakenly enabled both of them on the same umem
leading to buggy behavior. The main culprit for this is that we did
not store the association of umem to queue id in the copy case and
only relied on the driver reporting this. As this relation was not
stored in the driver for copy mode (it does not rely on the AF_XDP
NDOs), this obviously could not work.
This patch fixes the problem by always recording the umem to queue id
relationship in the netdev_queue and netdev_rx_queue structs. This way
we always know what kind of umem has been bound to a queue id and can
act appropriately at bind time.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This commit gets rid of the structure xdp_umem_props. It was there to
be able to break a dependency at one point, but this is no longer
needed. The values in the struct are instead stored directly in the
xdp_umem structure. This simplifies the xsk code as well as af_xdp
zero-copy drivers and as a bonus gets rid of one internal header file.
The i40e driver is also adapted to the new interface in this commit.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Previously, the AF_XDP (XDP_DRV/XDP_SKB copy-mode) ingress logic did
not include XDP meta data in the data buffers copied out to the user
application.
In this commit, we check if meta data is available, and if so, it is
prepended to the frame.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
xdp_return_buff() is used when frame has been successfully
handled (transmitted) or if an error occurred during delayed
processing and there is no way to report it back to
xdp_do_redirect().
In case of __xsk_rcv_zc() error is propagated all the way
back to the driver, so there is no need to call
xdp_return_buff(). Driver will recycle the frame anyway
after seeing that error happened.
Fixes: 173d3adb6f ("xsk: add zero-copy support for Rx")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This patch stops returning EMSGSIZE from sendmsg in copy mode when the
size of the packet is larger than the MTU. Just send it to the device
so that it will drop it as in zero-copy mode. This makes the error
reporting consistent between copy mode and zero-copy mode.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This patch makes sure ENOBUFS is always returned from sendmsg if there
is no TX queue configured. This was not the case for zero-copy
mode. With this patch this error reporting is consistent between copy
mode and zero-copy mode.
Fixes: ac98d8aab6 ("xsk: wire upp Tx zero-copy functions")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This patch stops returning EAGAIN in TX copy mode when the completion
queue is full as zero-copy does not do this. Instead this situation
can be detected by comparing the head and tail pointers of the
completion queue in both modes. In any case, EAGAIN was not the
correct error code here since no amount of calling sendmsg will solve
the problem. Only consuming one or more messages on the completion
queue will fix this.
With this patch, the error reporting becomes consistent between copy
mode and zero-copy mode.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This patch removes the ENXIO return code from TX copy-mode when
someone has forcefully changed the number of queues on the device so
that the queue bound to the socket is no longer available. Just
silently stop sending anything as in zero-copy mode so the error
reporting gets consistent between the two modes.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
There is a potential race in the TX completion code for the SKB
case. One process enters the sendmsg code of an AF_XDP socket in order
to send a frame. The execution eventually trickles down to the driver
that is told to send the packet. However, it decides to drop the
packet due to some error condition (e.g., rings full) and frees the
SKB. This will trigger the SKB destructor and a completion will be
sent to the AF_XDP user space through its
single-producer/single-consumer queues.
At the same time a TX interrupt has fired on another core and it
dispatches the TX completion code in the driver. It does its HW
specific things and ends up freeing the SKB associated with the
transmitted packet. This will trigger the SKB destructor and a
completion will be sent to the AF_XDP user space through its
single-producer/single-consumer queues. With a pseudo call stack, it
would look like this:
Core 1:
sendmsg() being called in the application
netdev_start_xmit()
Driver entered through ndo_start_xmit
Driver decides to free the SKB for some reason (e.g., rings full)
Destructor of SKB called
xskq_produce_addr() is called to signal completion to user space
Core 2:
TX completion irq
NAPI loop
Driver irq handler for TX completions
Frees the SKB
Destructor of SKB called
xskq_produce_addr() is called to signal completion to user space
We now have a violation of the single-producer/single-consumer
principle for our queues as there are two threads trying to produce at
the same time on the same queue.
Fixed by introducing a spin_lock in the destructor. In regards to the
performance, I get around 1.74 Mpps for txonly before and after the
introduction of the spinlock. There is of course some impact due to
the spin lock but it is in the less significant digits that are too
noisy for me to measure. But let us say that the version without the
spin lock got 1.745 Mpps in the best case and the version with 1.735
Mpps in the worst case, then that would mean a maximum drop in
performance of 0.5%.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Fixed a bug in which a frame could be completed more than once
when an error was returned from dev_direct_xmit(). The code
erroneously retried sending the message leading to multiple
calls to the SKB destructor and therefore multiple completions
of the same buffer to user space.
The error code in this case has been changed from EAGAIN to EBUSY
in order to tell user space that the sending of the packet failed
and the buffer has been return to user space through the completion
queue.
Fixes: 35fcde7f8d ("xsk: support for Tx")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Reported-by: Pavel Odintsov <pavel@fastnetmon.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The poll() changes were not well thought out, and completely
unexplained. They also caused a huge performance regression, because
"->poll()" was no longer a trivial file operation that just called down
to the underlying file operations, but instead did at least two indirect
calls.
Indirect calls are sadly slow now with the Spectre mitigation, but the
performance problem could at least be largely mitigated by changing the
"->get_poll_head()" operation to just have a per-file-descriptor pointer
to the poll head instead. That gets rid of one of the new indirections.
But that doesn't fix the new complexity that is completely unwarranted
for the regular case. The (undocumented) reason for the poll() changes
was some alleged AIO poll race fixing, but we don't make the common case
slower and more complex for some uncommon special case, so this all
really needs way more explanations and most likely a fundamental
redesign.
[ This revert is a revert of about 30 different commits, not reverted
individually because that would just be unnecessarily messy - Linus ]
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 173d3adb6f ("xsk: add zero-copy support for Rx") introduced a
regression on the XDP_SKB receive path, when the queue id checks were
removed. Now, they are back again.
Fixes: 173d3adb6f ("xsk: add zero-copy support for Rx")
Reported-by: Qi Zhang <qi.z.zhang@intel.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
With gcc-4.1.2 on 32-bit:
net/xdp/xsk.c:663: warning: integer constant is too large for ‘long’ type
net/xdp/xsk.c:665: warning: integer constant is too large for ‘long’ type
Add the missing "ULL" suffixes to the large XDP_UMEM_PGOFF_*_RING values
to fix this.
net/xdp/xsk.c:663: warning: comparison is always false due to limited range of data type
net/xdp/xsk.c:665: warning: comparison is always false due to limited range of data type
"unsigned long" is 32-bit on 32-bit systems, hence the offset is
truncated, and can never be equal to any of the XDP_UMEM_PGOFF_*_RING
values. Use loff_t (and the required cast) to fix this.
Fixes: 423f38329d ("xsk: add umem fill queue support and mmap")
Fixes: fe2308328c ("xsk: add umem completion queue support and mmap")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Pull networking updates from David Miller:
1) Add Maglev hashing scheduler to IPVS, from Inju Song.
2) Lots of new TC subsystem tests from Roman Mashak.
3) Add TCP zero copy receive and fix delayed acks and autotuning with
SO_RCVLOWAT, from Eric Dumazet.
4) Add XDP_REDIRECT support to mlx5 driver, from Jesper Dangaard
Brouer.
5) Add ttl inherit support to vxlan, from Hangbin Liu.
6) Properly separate ipv6 routes into their logically independant
components. fib6_info for the routing table, and fib6_nh for sets of
nexthops, which thus can be shared. From David Ahern.
7) Add bpf_xdp_adjust_tail helper, which can be used to generate ICMP
messages from XDP programs. From Nikita V. Shirokov.
8) Lots of long overdue cleanups to the r8169 driver, from Heiner
Kallweit.
9) Add BTF ("BPF Type Format"), from Martin KaFai Lau.
10) Add traffic condition monitoring to iwlwifi, from Luca Coelho.
11) Plumb extack down into fib_rules, from Roopa Prabhu.
12) Add Flower classifier offload support to igb, from Vinicius Costa
Gomes.
13) Add UDP GSO support, from Willem de Bruijn.
14) Add documentation for eBPF helpers, from Quentin Monnet.
15) Add TLS tx offload to mlx5, from Ilya Lesokhin.
16) Allow applications to be given the number of bytes available to read
on a socket via a control message returned from recvmsg(), from
Soheil Hassas Yeganeh.
17) Add x86_32 eBPF JIT compiler, from Wang YanQing.
18) Add AF_XDP sockets, with zerocopy support infrastructure as well.
From Björn Töpel.
19) Remove indirect load support from all of the BPF JITs and handle
these operations in the verifier by translating them into native BPF
instead. From Daniel Borkmann.
20) Add GRO support to ipv6 gre tunnels, from Eran Ben Elisha.
21) Allow XDP programs to do lookups in the main kernel routing tables
for forwarding. From David Ahern.
22) Allow drivers to store hardware state into an ELF section of kernel
dump vmcore files, and use it in cxgb4. From Rahul Lakkireddy.
23) Various RACK and loss detection improvements in TCP, from Yuchung
Cheng.
24) Add TCP SACK compression, from Eric Dumazet.
25) Add User Mode Helper support and basic bpfilter infrastructure, from
Alexei Starovoitov.
26) Support ports and protocol values in RTM_GETROUTE, from Roopa
Prabhu.
27) Support bulking in ->ndo_xdp_xmit() API, from Jesper Dangaard
Brouer.
28) Add lots of forwarding selftests, from Petr Machata.
29) Add generic network device failover driver, from Sridhar Samudrala.
* ra.kernel.org:/pub/scm/linux/kernel/git/davem/net-next: (1959 commits)
strparser: Add __strp_unpause and use it in ktls.
rxrpc: Fix terminal retransmission connection ID to include the channel
net: hns3: Optimize PF CMDQ interrupt switching process
net: hns3: Fix for VF mailbox receiving unknown message
net: hns3: Fix for VF mailbox cannot receiving PF response
bnx2x: use the right constant
Revert "net: sched: cls: Fix offloading when ingress dev is vxlan"
net: dsa: b53: Fix for brcm tag issue in Cygnus SoC
enic: fix UDP rss bits
netdev-FAQ: clarify DaveM's position for stable backports
rtnetlink: validate attributes in do_setlink()
mlxsw: Add extack messages for port_{un, }split failures
netdevsim: Add extack error message for devlink reload
devlink: Add extack to reload and port_{un, }split operations
net: metrics: add proper netlink validation
ipmr: fix error path when ipmr_new_table fails
ip6mr: only set ip6mr_table from setsockopt when ip6mr_new_table succeeds
net: hns3: remove unused hclgevf_cfg_func_mta_filter
netfilter: provide udp*_lib_lookup for nf_tproxy
qed*: Utilize FW 8.37.2.0
...
Here we add the functionality required to support zero-copy Tx, and
also exposes various zero-copy related functions for the netdevs.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Extend the xsk_rcv to support the new MEM_TYPE_ZERO_COPY memory, and
wireup ndo_bpf call in bind.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Currently, AF_XDP only supports a fixed frame-size memory scheme where
each frame is referenced via an index (idx). A user passes the frame
index to the kernel, and the kernel acts upon the data. Some NICs,
however, do not have a fixed frame-size model, instead they have a
model where a memory window is passed to the hardware and multiple
frames are filled into that window (referred to as the "type-writer"
model).
By changing the descriptor format from the current frame index
addressing scheme, AF_XDP can in the future be extended to support
these kinds of NICs.
In the index-based model, an idx refers to a frame of size
frame_size. Addressing a frame in the UMEM is done by offseting the
UMEM starting address by a global offset, idx * frame_size + offset.
Communicating via the fill- and completion-rings are done by means of
idx.
In this commit, the idx is removed in favor of an address (addr),
which is a relative address ranging over the UMEM. To convert an
idx-based address to the new addr is simply: addr = idx * frame_size +
offset.
We also stop referring to the UMEM "frame" as a frame. Instead it is
simply called a chunk.
To transfer ownership of a chunk to the kernel, the addr of the chunk
is passed in the fill-ring. Note, that the kernel will mask addr to
make it chunk aligned, so there is no need for userspace to do
that. E.g., for a chunk size of 2k, passing an addr of 2048, 2050 or
3000 to the fill-ring will refer to the same chunk.
On the completion-ring, the addr will match that of the Tx descriptor,
passed to the kernel.
Changing the descriptor format to use chunks/addr will allow for
future changes to move to a type-writer based model, where multiple
frames can reside in one chunk. In this model passing one single chunk
into the fill-ring, would potentially result in multiple Rx
descriptors.
This commit changes the uapi of AF_XDP sockets, and updates the
documentation.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Previously, rx_dropped could be updated incorrectly, e.g. if the XDP
program redirected the frame to a socket bound to a different queue
than where the XDP program was executing.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Previously the fill queue descriptor was not copied to kernel space
prior validating it, making it possible for userland to change the
descriptor post-kernel-validation.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
As suggested by Daniel Borkmann, the umem setup code was a too
defensive and complex. Here, we reduce the number of checks. Also, the
memory pinning is now folded into the umem creation, and we do correct
locking.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Here, we add a missing write-barrier, and use READ_ONCE for the
data-dependency barrier.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
In this commit we remove the explicit ring structure from the the
uapi. It is tricky for an uapi to depend on a certain L1 cache line
size, since it can differ for variants of the same architecture. Now,
we let the user application determine the offsets of the producer,
consumer and descriptors by asking the socket via getsockopt.
A typical flow would be (Rx ring):
struct xdp_mmap_offsets off;
struct xdp_desc *ring;
u32 *prod, *cons;
void *map;
...
getsockopt(fd, SOL_XDP, XDP_MMAP_OFFSETS, &off, &optlen);
map = mmap(NULL, off.rx.desc +
NUM_DESCS * sizeof(struct xdp_desc),
PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_POPULATE, sfd,
XDP_PGOFF_RX_RING);
prod = map + off.rx.producer;
cons = map + off.rx.consumer;
ring = map + off.rx.desc;
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Validate the queue id against both Rx and Tx on the netdev. Also, make
sure that the queue exists at xmit time.
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Tested-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Supporting rebind, i.e. after a successful bind the process can call
bind again without closing the socket, makes the AF_XDP setup state
machine more complex. Constrain the state space, by not supporting
rebind.
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Björn Töpel