[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5b928f0e-f3f8-4eaa-b750-e3f445d2fa46@gmail.com>
Date: Fri, 1 Nov 2024 17:41:05 +0000
From: Pavel Begunkov <asml.silence@...il.com>
To: Mina Almasry <almasrymina@...gle.com>, David Wei <dw@...idwei.uk>
Cc: io-uring@...r.kernel.org, netdev@...r.kernel.org,
Jens Axboe <axboe@...nel.dk>, Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>, "David S. Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>, Jesper Dangaard Brouer
<hawk@...nel.org>, David Ahern <dsahern@...nel.org>,
Stanislav Fomichev <stfomichev@...il.com>, Joe Damato <jdamato@...tly.com>,
Pedro Tammela <pctammela@...atatu.com>
Subject: Re: [PATCH v7 04/15] net: prepare for non devmem TCP memory providers
On 11/1/24 17:09, Mina Almasry wrote:
> On Tue, Oct 29, 2024 at 4:06 PM David Wei <dw@...idwei.uk> wrote:
...
>> +
>> static void net_devmem_dmabuf_free_chunk_owner(struct gen_pool *genpool,
>> struct gen_pool_chunk *chunk,
>> void *not_used)
>> @@ -316,10 +322,10 @@ void dev_dmabuf_uninstall(struct net_device *dev)
>> unsigned int i;
>>
>> for (i = 0; i < dev->real_num_rx_queues; i++) {
>> - binding = dev->_rx[i].mp_params.mp_priv;
>> - if (!binding)
>> + if (dev->_rx[i].mp_params.mp_ops != &dmabuf_devmem_ops)
>> continue;
>>
>
> Use the net_is_devmem_page_pool_ops helper here?
It could, but the function is there primarily for outside users to
avoid ifdefs and build problems. I don't think it worth reiteration?
I'll change if there is a next version.
...
>> @@ -244,8 +244,11 @@ page_pool_nl_fill(struct sk_buff *rsp, const struct page_pool *pool,
>> pool->user.detach_time))
>> goto err_cancel;
>>
>> - if (binding && nla_put_u32(rsp, NETDEV_A_PAGE_POOL_DMABUF, binding->id))
>> - goto err_cancel;
>> + if (net_is_devmem_page_pool_ops(pool->mp_ops)) {
>> + binding = pool->mp_priv;
>> + if (nla_put_u32(rsp, NETDEV_A_PAGE_POOL_DMABUF, binding->id))
>> + goto err_cancel;
>> + }
>
> Worthy of note is that I think Jakub asked for this introspection, and
> likely you should also add similar introspection. I.e. page_pool
I think we can patch it up after merging the series? Depends what Jakub
thinks. In any case, I can't parse io_uring ops here until a later patch
adding those ops, so it'd be a new patch if it's a part of this series.
> dumping should likely be improved to dump that it's bound to io_uring
> memory. Not sure what io_uring memory 'id' equivalent would be, if
> any.
I don't think io_uring have any id to give. What is it for in the
first place? Do you give it to netlink to communicate with devmem
TCP or something similar?
>> genlmsg_end(rsp, hdr);
>>
>> @@ -353,16 +356,16 @@ void page_pool_unlist(struct page_pool *pool)
>> int page_pool_check_memory_provider(struct net_device *dev,
>> struct netdev_rx_queue *rxq)
...
>> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
>> index e928efc22f80..31e01da61c12 100644
>> --- a/net/ipv4/tcp.c
>> +++ b/net/ipv4/tcp.c
>> @@ -277,6 +277,7 @@
>> #include <net/ip.h>
>> #include <net/sock.h>
>> #include <net/rstreason.h>
>> +#include <net/page_pool/types.h>
>>
>> #include <linux/uaccess.h>
>> #include <asm/ioctls.h>
>> @@ -2476,6 +2477,11 @@ static int tcp_recvmsg_dmabuf(struct sock *sk, const struct sk_buff *skb,
>> }
>>
>> niov = skb_frag_net_iov(frag);
>> + if (net_is_devmem_page_pool_ops(niov->pp->mp_ops)) {
>> + err = -ENODEV;
>> + goto out;
>> + }
>> +
>
> I think this check needs to go in the caller. Currently the caller
> assumes that if !skb_frags_readable(), then the frag is dma-buf, and
io_uring originated netmem that are marked unreadable as well
and so will end up in tcp_recvmsg_dmabuf(), then we reject and
fail since they should not be fed to devmem TCP. It should be
fine from correctness perspective.
We need to check frags, and that's the place where we iterate
frags. Another option is to add a loop in tcp_recvmsg_locked
walking over all frags of an skb and doing the checks, but
that's an unnecessary performance burden to devmem.
> calls tcp_recvmsg_dmabuf on it. The caller needs to check that the
> frag is specifically a dma-buf frag now.
>
> Can io_uring frags somehow end up in tcp_recvmsg_locked? You're still
> using the tcp stack with io_uring ZC right? So I suspect they might?
All of them are using the same socket rx queue, so yes, any of them
can see any type of packet non net_iov / pages
--
Pavel Begunkov
Powered by blists - more mailing lists