I analyze the potential sleeping issue of the following processes:
Thread A                                Thread B
...                                     netlink_create  //ref = 1
do_mq_notify                            ...
  sock = netlink_getsockbyfilp          ...     //ref = 2
  info->notify_sock = sock;             ...
...                                     netlink_sendmsg
...                                       skb = netlink_alloc_large_skb  //skb->head is vmalloced
...                                       netlink_unicast
...                                         sk = netlink_getsockbyportid //ref = 3
...                                         netlink_sendskb
...                                           __netlink_sendskb
...                                             skb_queue_tail //put skb to sk_receive_queue
...                                         sock_put //ref = 2
...                                     ...
...                                     netlink_release
...                                       deferred_put_nlk_sk //ref = 1
mqueue_flush_file
  spin_lock
  remove_notification
    netlink_sendskb
      sock_put  //ref = 0
        sk_free
          ...
          __sk_destruct
            netlink_sock_destruct
              skb_queue_purge  //get skb from sk_receive_queue
                ...
                __skb_queue_purge_reason
                  kfree_skb_reason
                    __kfree_skb
                    ...
                    skb_release_all
                      skb_release_head_state
                        netlink_skb_destructor
                          vfree(skb->head)  //sleeping while holding spinlock

In netlink_sendmsg, if the memory pointed to by skb->head is allocated by
vmalloc, and is put to sk_receive_queue queue, also the skb is not freed.
When the mqueue executes flush, the sleeping bug will occur. Use
vfree_atomic instead of vfree in netlink_skb_destructor to solve the issue.

Fixes: c05cdb1b864f ("netlink: allow large data transfers from user-space")
Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Link: https://lore.kernel.org/r/20240122011807.2110357-1-shaozhengchao@huawei.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

As explained in commit e03781879a0d ("drop_monitor: Require
'CAP_SYS_ADMIN' when joining "events" group"), the "flags" field in the
multicast group structure reuses uAPI flags despite the field not being
exposed to user space. This makes it impossible to extend its use
without adding new uAPI flags, which is inappropriate for internal
kernel checks.

Solve this by adding internal flags (i.e., "GENL_MCAST_*") and convert
the existing users to use them instead of the uAPI flags.

Tested using the reproducers in commit 44ec98ea5ea9 ("psample: Require
'CAP_NET_ADMIN' when joining "packets" group") and commit e03781879a0d
("drop_monitor: Require 'CAP_SYS_ADMIN' when joining "events" group").

No functional changes intended.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make the code using filter function a bit nicer by consolidating the
filter function arguments using typedef.

Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Introduce an xarray for Generic netlink family to store per-socket
private. Initialize this xarray only if family uses per-socket privs.

Introduce genl_sk_priv_get() to get the socket priv pointer for a family
and initialize it in case it does not exist.
Introduce __genl_sk_priv_get() to obtain socket priv pointer for a
family under RCU read lock.

Allow family to specify the priv size, init() and destroy() callbacks.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Cross-merge networking fixes after downstream PR.

Conflicts:

drivers/net/ethernet/stmicro/stmmac/dwmac5.c
drivers/net/ethernet/stmicro/stmmac/dwmac5.h
drivers/net/ethernet/stmicro/stmmac/dwxgmac2_core.c
drivers/net/ethernet/stmicro/stmmac/hwif.h
  37e4b8df27bc ("net: stmmac: fix FPE events losing")
  c3f3b97238f6 ("net: stmmac: Refactor EST implementation")
https://lore.kernel.org/all/20231206110306.01e91114@canb.auug.org.au/

Adjacent changes:

net/ipv4/tcp_ao.c
  9396c4ee93f9 ("net/tcp: Don't store TCP-AO maclen on reqsk")
  7b0f570f879a ("tcp: Move TCP-AO bits from cookie_v[46]_check() to tcp_ao_syncookie().")

Signed-off-by: Jakub Kicinski <kuba@kernel.org>

The "NET_DM" generic netlink family notifies drop locations over the
"events" multicast group. This is problematic since by default generic
netlink allows non-root users to listen to these notifications.

Fix by adding a new field to the generic netlink multicast group
structure that when set prevents non-root users or root without the
'CAP_SYS_ADMIN' capability (in the user namespace owning the network
namespace) from joining the group. Set this field for the "events"
group. Use 'CAP_SYS_ADMIN' rather than 'CAP_NET_ADMIN' because of the
nature of the information that is shared over this group.

Note that the capability check in this case will always be performed
against the initial user namespace since the family is not netns aware
and only operates in the initial network namespace.

A new field is added to the structure rather than using the "flags"
field because the existing field uses uAPI flags and it is inappropriate
to add a new uAPI flag for an internal kernel check. In net-next we can
rework the "flags" field to use internal flags and fold the new field
into it. But for now, in order to reduce the amount of changes, add a
new field.

Since the information can only be consumed by root, mark the control
plane operations that start and stop the tracing as root-only using the
'GENL_ADMIN_PERM' flag.

Tested using [1].

Before:

 # capsh -- -c ./dm_repo
 # capsh --drop=cap_sys_admin -- -c ./dm_repo

After:

 # capsh -- -c ./dm_repo
 # capsh --drop=cap_sys_admin -- -c ./dm_repo
 Failed to join "events" multicast group

[1]
 $ cat dm.c
 #include <stdio.h>
 #include <netlink/genl/ctrl.h>
 #include <netlink/genl/genl.h>
 #include <netlink/socket.h>

 int main(int argc, char **argv)
 {
 	struct nl_sock *sk;
 	int grp, err;

 	sk = nl_socket_alloc();
 	if (!sk) {
 		fprintf(stderr, "Failed to allocate socket\n");
 		return -1;
 	}

 	err = genl_connect(sk);
 	if (err) {
 		fprintf(stderr, "Failed to connect socket\n");
 		return err;
 	}

 	grp = genl_ctrl_resolve_grp(sk, "NET_DM", "events");
 	if (grp < 0) {
 		fprintf(stderr,
 			"Failed to resolve \"events\" multicast group\n");
 		return grp;
 	}

 	err = nl_socket_add_memberships(sk, grp, NFNLGRP_NONE);
 	if (err) {
 		fprintf(stderr, "Failed to join \"events\" multicast group\n");
 		return err;
 	}

 	return 0;
 }
 $ gcc -I/usr/include/libnl3 -lnl-3 -lnl-genl-3 -o dm_repo dm.c

Fixes: 9a8afc8d3962 ("Network Drop Monitor: Adding drop monitor implementation & Netlink protocol")
Reported-by: "The UK's National Cyber Security Centre (NCSC)" <security@ncsc.gov.uk>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20231206213102.1824398-3-idosch@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

if a PF has 256 or more VFs, ip link command will allocate an order 3
memory or more, and maybe trigger OOM due to memory fragment,
the VFs needed memory size is computed in rtnl_vfinfo_size.

so introduce nlmsg_new_large which calls netlink_alloc_large_skb in
which vmalloc is used for large memory, to avoid the failure of
allocating memory

    ip invoked oom-killer: gfp_mask=0xc2cc0(GFP_KERNEL|__GFP_NOWARN|\
	__GFP_COMP|__GFP_NOMEMALLOC), order=3, oom_score_adj=0
    CPU: 74 PID: 204414 Comm: ip Kdump: loaded Tainted: P           OE
    Call Trace:
    dump_stack+0x57/0x6a
    dump_header+0x4a/0x210
    oom_kill_process+0xe4/0x140
    out_of_memory+0x3e8/0x790
    __alloc_pages_slowpath.constprop.116+0x953/0xc50
    __alloc_pages_nodemask+0x2af/0x310
    kmalloc_large_node+0x38/0xf0
    __kmalloc_node_track_caller+0x417/0x4d0
    __kmalloc_reserve.isra.61+0x2e/0x80
    __alloc_skb+0x82/0x1c0
    rtnl_getlink+0x24f/0x370
    rtnetlink_rcv_msg+0x12c/0x350
    netlink_rcv_skb+0x50/0x100
    netlink_unicast+0x1b2/0x280
    netlink_sendmsg+0x355/0x4a0
    sock_sendmsg+0x5b/0x60
    ____sys_sendmsg+0x1ea/0x250
    ___sys_sendmsg+0x88/0xd0
    __sys_sendmsg+0x5e/0xa0
    do_syscall_64+0x33/0x40
    entry_SYSCALL_64_after_hwframe+0x44/0xa9
    RIP: 0033:0x7f95a65a5b70

Cc: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Link: https://lore.kernel.org/r/20231115120108.3711-1-lirongqing@baidu.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

W=1 builds now warn if a module is built without
a MODULE_DESCRIPTION(). Fill it in for sock_diag.

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, split ops of doit and dumpit are merged into a single iter
item when they are subsequent. However, there is no guarantee that the
dumpit op is for the same cmd as doit op.

Fix this by checking if cmd is the same for both.
This problem does not occur in existing families.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20231021112711.660606-2-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

We currently push everyone to use padding to align 64b values
in netlink. Un-padded nla_put_u64() doesn't even exist any more.

The story behind this possibly start with this thread:
https://lore.kernel.org/netdev/20121204.130914.1457976839967676240.davem@davemloft.net/
where DaveM was concerned about the alignment of a structure
containing 64b stats. If user space tries to access such struct
directly:

	struct some_stats *stats = nla_data(attr);
	printf("A: %llu", stats->a);

lack of alignment may become problematic for some architectures.
These days we most often put every single member in a separate
attribute, meaning that the code above would use a helper like
nla_get_u64(), which can deal with alignment internally.
Even for arches which don't have good unaligned access - access
aligned to 4B should be pretty efficient.
Kernel and well known libraries deal with unaligned input already.

Padded 64b is quite space-inefficient (64b + pad means at worst 16B
per attr vs 32b which takes 8B). It is also more typing:

    if (nla_put_u64_pad(rsp, NETDEV_A_SOMETHING_SOMETHING,
                        value, NETDEV_A_SOMETHING_PAD))

Create a new attribute type which will use 32 bits at netlink
level if value is small enough (probably most of the time?),
and (4B-aligned) 64 bits otherwise. Kernel API is just:

    if (nla_put_uint(rsp, NETDEV_A_SOMETHING_SOMETHING, value))

Calling this new type "just" sint / uint with no specific size
will hopefully also make people more comfortable with using it.
Currently telling people "don't use u8, you may need the bits,
and netlink will round up to 4B, anyway" is the #1 comment
we give to newcomers.

In terms of netlink layout it looks like this:

         0       4       8       12      16
32b:     [nlattr][ u32  ]
64b:     [  pad ][nlattr][     u64      ]
uint(32) [nlattr][ u32  ]
uint(64) [nlattr][     u64      ]

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>