iwlwifi: mei: add the driver to allow cooperation with CSME

CSME in two words
-----------------
CSME stands for Converged Security and Management Engine. It is
a CPU on the chipset and runs a dedicated firmware.
AMT (Active Management Technology) is one of the applications
that run on that CPU. AMT allows to control the platform remotely.
Here is a partial list of the use cases:
* View the screen of the plaform, with keyboard and mouse (KVM)
* Attach a remote IDE device
* Have a serial console to the device
* Query the state of the platform
* Reset / shut down / boot the platform

Networking in CSME
------------------
For those uses cases, CSME's firmware has an embedded network
stack and is able to use the network devices of the system: LAN
and WLAN. This is thanks to the CSME's firmware WLAN driver.

One can add a profile (SSID / key / certificate) to the CSME's OS
and CSME will connect to that profile. Then, one can use the WLAN
link to access the applications that run on CSME (AMT is one of
them). Note that CSME is active during power state and power state
transitions. For example, it is possible to have a KVM session
open to the system while the system is rebooting and actually
configure the BIOS remotely over WLAN thanks to AMT.

How all this is related to Linux
--------------------------------
In Linux, there is a driver that allows the OS to talk to the
CSME firmware, this driver is drivers/misc/mei. This driver
advertises a bus that allows other kernel drivers or even user
space) to talk to components inside the CSME firmware.
In practice, the system advertises a PCI device that allows
to send / receive data to / from the CSME firmware. The mei
bus drivers in drivers/misc/mei is an abstration on top of
this PCI device.
The driver being added here is called iwlmei and talks to the
WLAN driver inside the CSME firmware through the mei bus driver.
Note that the mei bus driver only gives bus services, it doesn't
define the content of the communication.

Why do we need this driver?
--------------------------
CSME uses the same WLAN device that the OS is expecting to see
hence we need an arbitration mechanism. This is what iwlmei is
in charge of. iwlmei maintains the communication with the CSME
firmware's WLAN driver. The language / protocol that is used
between the CSME's firmware WLAN driver and iwlmei is OS agnostic
and is called SAP which stands for Software Abritration Protocol.
With SAP, iwlmei will be able to tell the CSME firmware's WLAN
driver:
1) Please give me the device.
2) Please note that the SW/HW rfkill state change.
3) Please note that I am now associated to X.
4) Please note that I received this packet.
etc...

There are messages that go the opposite direction as well:
1) Please note that AMT is en/disable.
2) Please note that I believe the OS is broken and hence I'll take
   the device *now*, whether you like it or not, to make sure that
   connectivity is preserved.
3) Please note that I am willing to give the device if the OS
   needs it.
4) Please give me any packet that is sent on UDP / TCP on IP address
   XX.XX.XX.XX and an port ZZ.
5) Please send this packet.
etc...

Please check drivers/net/wireless/intel/iwlwifi/mei/sap.h for the
full protocol specification.

Arbitration is not the only purpose of iwlmei and SAP. SAP also
allows to maintain the AMT's functionality even when the OS owns
the device. To connect to AMT, one needs to initiate an HTTP
connection to port 16992. iwlmei will listen to the Rx path and
forward (through SAP) to the CSME firmware the data it got. Then,
the embedded HTTP server in the chipset will reply to the request
and send a SAP notification to ask iwlmei to send the reply.
This way, AMT running on the CSME can still work.

In practice this means that all the use cases quoted above (KVM,
remote IDE device, etc...) will work even when the OS uses the
WLAN device.

How to disable all this?
---------------------------
iwlmei won't be able to do anything if the CSME's networking stack
is not enabled. By default, CSME's networking stack is disabled (this
is a BIOS setting).
In case the CSME's networking stack is disabled, iwlwifi will just
get access to the device because there is no contention with any other
actor and, hence, no arbitration is needed.

In this patch, I only add the iwlmei driver. Integration with
iwlwifi will be implemented in the next one.

Co-Developed-by: Ayala Beker <ayala.beker@intel.com>
Signed-off-by: Ayala Beker <ayala.beker@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>

v2: fix a few warnings raised by the different bots
v3: rewrite the commit message
v4: put the debugfs content in a different patch
v5: fix a NULL pointer dereference upon DHCP TX if SAP is connected
    since we now have the required cfg80211 bits in wl-drv-next, add
    the RFKILL handling patch to this series.
v6: change the SAP API to inherit the values from iwl-mei.h removing
    the need to ensure the values are equal with a BUILD_BUG_ON.
    This was suggested by Arend
v7: * fix a locking issue in case of CSME firmware reset:
      When the CSME firmware resets, we need to unregister the
      netdev, first take the mutex, and only then, rely on it
      being taken.
    * Add a comment to explain why it is ok to have static variables
      (iwlmei can't have more than a single instance).
    * Add a define for 26 + 8 + 8
    * Add a define SEND_SAP_MAX_WAIT_ITERATION
    * make struct const
    * Reword a bit the Kconfig help message
    * Ayala added her Signed-off
    * fixed an RCU annotation
v8: do not require ownership upfront, use NIC_OWNER instead. This fixes
    a deadlock when CSME does not have the right WiFi FW.
    Add more documentation about the owernship transition
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Link: https://lore.kernel.org/r/20211112062814.7502-2-emmanuel.grumbach@intel.com

1 files changed, 409 insertions, 0 deletions

diff --git a/drivers/net/wireless/intel/iwlwifi/mei/net.c b/drivers/net/wireless/intel/iwlwifi/mei/net.c
new file mode 100644
index 000000000000..5f966af69720
--- /dev/null
+++ b/drivers/net/wireless/intel/iwlwifi/mei/net.c
@@ -0,0 +1,409 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2021 Intel Corporation
+ */
+
+#include <uapi/linux/if_ether.h>
+#include <uapi/linux/if_arp.h>
+#include <uapi/linux/icmp.h>
+
+#include <linux/etherdevice.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+#include <linux/ieee80211.h>
+
+#include <net/cfg80211.h>
+#include <net/ip.h>
+
+#include <linux/if_arp.h>
+#include <linux/icmp.h>
+#include <linux/udp.h>
+#include <linux/ip.h>
+#include <linux/mm.h>
+
+#include "internal.h"
+#include "sap.h"
+#include "iwl-mei.h"
+
+/*
+ * Returns true if further filtering should be stopped. Only in that case
+ * pass_to_csme and rx_handler_res are set. Otherwise, next level of filters
+ * should be checked.
+ */
+static bool iwl_mei_rx_filter_eth(const struct ethhdr *ethhdr,
+				  const struct iwl_sap_oob_filters *filters,
+				  bool *pass_to_csme,
+				  rx_handler_result_t *rx_handler_res)
+{
+	const struct iwl_sap_eth_filter *filt;
+
+	/* This filter is not relevant for UCAST packet */
+	if (!is_multicast_ether_addr(ethhdr->h_dest) ||
+	    is_broadcast_ether_addr(ethhdr->h_dest))
+		return false;
+
+	for (filt = &filters->eth_filters[0];
+	     filt < &filters->eth_filters[0] + ARRAY_SIZE(filters->eth_filters);
+	     filt++) {
+		/* Assume there are no enabled filter after a disabled one */
+		if (!(filt->flags & SAP_ETH_FILTER_ENABLED))
+			break;
+
+		if (compare_ether_header(filt->mac_address, ethhdr->h_dest))
+			continue;
+
+		/* Packet needs to reach the host's stack */
+		if (filt->flags & SAP_ETH_FILTER_COPY)
+			*rx_handler_res = RX_HANDLER_PASS;
+		else
+			*rx_handler_res = RX_HANDLER_CONSUMED;
+
+		/* We have an authoritative answer, stop filtering */
+		if (filt->flags & SAP_ETH_FILTER_STOP) {
+			*pass_to_csme = true;
+			return true;
+		}
+
+		return false;
+	}
+
+	 /* MCAST frames that don't match layer 2 filters are not sent to ME */
+	*pass_to_csme  = false;
+
+	return true;
+}
+
+/*
+ * Returns true iff the frame should be passed to CSME in which case
+ * rx_handler_res is set.
+ */
+static bool iwl_mei_rx_filter_arp(struct sk_buff *skb,
+				  const struct iwl_sap_oob_filters *filters,
+				  rx_handler_result_t *rx_handler_res)
+{
+	const struct iwl_sap_ipv4_filter *filt = &filters->ipv4_filter;
+	const struct arphdr *arp;
+	const __be32 *target_ip;
+	u32 flags = le32_to_cpu(filt->flags);
+
+	if (!pskb_may_pull(skb, arp_hdr_len(skb->dev)))
+		return false;
+
+	arp = arp_hdr(skb);
+
+	/* Handle only IPv4 over ethernet ARP frames */
+	if (arp->ar_hrd != htons(ARPHRD_ETHER) ||
+	    arp->ar_pro != htons(ETH_P_IP))
+		return false;
+
+	/*
+	 * After the ARP header, we have:
+	 * src MAC address   - 6 bytes
+	 * src IP address    - 4 bytes
+	 * target MAC addess - 6 bytes
+	 */
+	target_ip = (void *)((u8 *)(arp + 1) +
+			     ETH_ALEN + sizeof(__be32) + ETH_ALEN);
+
+	/*
+	 * ARP request is forwarded to ME only if IP address match in the
+	 * ARP request's target ip field.
+	 */
+	if (arp->ar_op == htons(ARPOP_REQUEST) &&
+	    (filt->flags & cpu_to_le32(SAP_IPV4_FILTER_ARP_REQ_PASS)) &&
+	    (filt->ipv4_addr == 0 || filt->ipv4_addr == *target_ip)) {
+		if (flags & SAP_IPV4_FILTER_ARP_REQ_COPY)
+			*rx_handler_res = RX_HANDLER_PASS;
+		else
+			*rx_handler_res = RX_HANDLER_CONSUMED;
+
+		return true;
+	}
+
+	/* ARP reply is always forwarded to ME regardless of the IP */
+	if (flags & SAP_IPV4_FILTER_ARP_RESP_PASS &&
+	    arp->ar_op == htons(ARPOP_REPLY)) {
+		if (flags & SAP_IPV4_FILTER_ARP_RESP_COPY)
+			*rx_handler_res = RX_HANDLER_PASS;
+		else
+			*rx_handler_res = RX_HANDLER_CONSUMED;
+
+		return true;
+	}
+
+	return false;
+}
+
+static bool
+iwl_mei_rx_filter_tcp_udp(struct sk_buff *skb, bool  ip_match,
+			  const struct iwl_sap_oob_filters *filters,
+			  rx_handler_result_t *rx_handler_res)
+{
+	const struct iwl_sap_flex_filter *filt;
+
+	for (filt = &filters->flex_filters[0];
+	     filt < &filters->flex_filters[0] + ARRAY_SIZE(filters->flex_filters);
+	     filt++) {
+		if (!(filt->flags & SAP_FLEX_FILTER_ENABLED))
+			break;
+
+		/*
+		 * We are required to have a match on the IP level and we didn't
+		 * have such match.
+		 */
+		if ((filt->flags &
+		     (SAP_FLEX_FILTER_IPV4 | SAP_FLEX_FILTER_IPV6)) &&
+		    !ip_match)
+			continue;
+
+		if ((filt->flags & SAP_FLEX_FILTER_UDP) &&
+		    ip_hdr(skb)->protocol != IPPROTO_UDP)
+			continue;
+
+		if ((filt->flags & SAP_FLEX_FILTER_TCP) &&
+		    ip_hdr(skb)->protocol != IPPROTO_TCP)
+			continue;
+
+		/*
+		 * We must have either a TCP header or a UDP header, both
+		 * starts with a source port and then a destination port.
+		 * Both are big endian words.
+		 * Use a UDP header and that will work for TCP as well.
+		 */
+		if ((filt->src_port && filt->src_port != udp_hdr(skb)->source) ||
+		    (filt->dst_port && filt->dst_port != udp_hdr(skb)->dest))
+			continue;
+
+		if (filt->flags & SAP_FLEX_FILTER_COPY)
+			*rx_handler_res = RX_HANDLER_PASS;
+		else
+			*rx_handler_res = RX_HANDLER_CONSUMED;
+
+		return true;
+	}
+
+	return false;
+}
+
+static bool iwl_mei_rx_filter_ipv4(struct sk_buff *skb,
+				   const struct iwl_sap_oob_filters *filters,
+				   rx_handler_result_t *rx_handler_res)
+{
+	const struct iwl_sap_ipv4_filter *filt = &filters->ipv4_filter;
+	const struct iphdr *iphdr;
+	unsigned int iphdrlen;
+	bool match;
+
+	if (!pskb_may_pull(skb, skb_network_offset(skb) + sizeof(*iphdr)) ||
+	    !pskb_may_pull(skb, skb_network_offset(skb) +
+			   sizeof(ip_hdrlen(skb) - sizeof(*iphdr))))
+		return false;
+
+	iphdrlen = ip_hdrlen(skb);
+	iphdr = ip_hdr(skb);
+	match = !filters->ipv4_filter.ipv4_addr ||
+		filters->ipv4_filter.ipv4_addr == iphdr->daddr;
+
+	skb_set_transport_header(skb, skb_network_offset(skb) + iphdrlen);
+
+	switch (ip_hdr(skb)->protocol) {
+	case IPPROTO_UDP:
+	case IPPROTO_TCP:
+		/*
+		 * UDP header is shorter than TCP header and we look at the first bytes
+		 * of the header anyway (see below).
+		 * If we have a truncated TCP packet, let CSME handle this.
+		 */
+		if (!pskb_may_pull(skb, skb_transport_offset(skb) +
+				   sizeof(struct udphdr)))
+			return false;
+
+		return iwl_mei_rx_filter_tcp_udp(skb, match,
+						 filters, rx_handler_res);
+
+	case IPPROTO_ICMP: {
+		struct icmphdr *icmp;
+
+		if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(*icmp)))
+			return false;
+
+		icmp = icmp_hdr(skb);
+
+		/*
+		 * Don't pass echo requests to ME even if it wants it as we
+		 * want the host to answer.
+		 */
+		if ((filt->flags & cpu_to_le32(SAP_IPV4_FILTER_ICMP_PASS)) &&
+		    match && (icmp->type != ICMP_ECHO || icmp->code != 0)) {
+			if (filt->flags & cpu_to_le32(SAP_IPV4_FILTER_ICMP_COPY))
+				*rx_handler_res = RX_HANDLER_PASS;
+			else
+				*rx_handler_res = RX_HANDLER_CONSUMED;
+
+			return true;
+		}
+		break;
+		}
+	case IPPROTO_ICMPV6:
+		/* TODO: Should we have the same ICMP request logic here too? */
+		if ((filters->icmpv6_flags & cpu_to_le32(SAP_ICMPV6_FILTER_ENABLED) &&
+		     match)) {
+			if (filters->icmpv6_flags &
+			    cpu_to_le32(SAP_ICMPV6_FILTER_COPY))
+				*rx_handler_res = RX_HANDLER_PASS;
+			else
+				*rx_handler_res = RX_HANDLER_CONSUMED;
+
+			return true;
+		}
+		break;
+	default:
+		return false;
+	}
+
+	return false;
+}
+
+static bool iwl_mei_rx_filter_ipv6(struct sk_buff *skb,
+				   const struct iwl_sap_oob_filters *filters,
+				   rx_handler_result_t *rx_handler_res)
+{
+	*rx_handler_res = RX_HANDLER_PASS;
+
+	/* TODO */
+
+	return false;
+}
+
+static rx_handler_result_t
+iwl_mei_rx_pass_to_csme(struct sk_buff *skb,
+			const struct iwl_sap_oob_filters *filters,
+			bool *pass_to_csme)
+{
+	const struct ethhdr *ethhdr = (void *)skb_mac_header(skb);
+	rx_handler_result_t rx_handler_res = RX_HANDLER_PASS;
+	bool (*filt_handler)(struct sk_buff *skb,
+			     const struct iwl_sap_oob_filters *filters,
+			     rx_handler_result_t *rx_handler_res);
+
+	/*
+	 * skb->data points the IP header / ARP header and the ETH header
+	 * is in the headroom.
+	 */
+	skb_reset_network_header(skb);
+
+	/*
+	 * MCAST IP packets sent by us are received again here without
+	 * an ETH header. Drop them here.
+	 */
+	if (!skb_mac_offset(skb))
+		return RX_HANDLER_PASS;
+
+	if (skb_headroom(skb) < sizeof(*ethhdr))
+		return RX_HANDLER_PASS;
+
+	if (iwl_mei_rx_filter_eth(ethhdr, filters,
+				  pass_to_csme, &rx_handler_res))
+		return rx_handler_res;
+
+	switch (skb->protocol) {
+	case htons(ETH_P_IP):
+		filt_handler = iwl_mei_rx_filter_ipv4;
+		break;
+	case htons(ETH_P_ARP):
+		filt_handler = iwl_mei_rx_filter_arp;
+		break;
+	case htons(ETH_P_IPV6):
+		filt_handler = iwl_mei_rx_filter_ipv6;
+		break;
+	default:
+		*pass_to_csme = false;
+		return rx_handler_res;
+	}
+
+	*pass_to_csme = filt_handler(skb, filters, &rx_handler_res);
+
+	return rx_handler_res;
+}
+
+rx_handler_result_t iwl_mei_rx_filter(struct sk_buff *orig_skb,
+				      const struct iwl_sap_oob_filters *filters,
+				      bool *pass_to_csme)
+{
+	rx_handler_result_t ret;
+	struct sk_buff *skb;
+
+	ret = iwl_mei_rx_pass_to_csme(orig_skb, filters, pass_to_csme);
+
+	if (!*pass_to_csme)
+		return RX_HANDLER_PASS;
+
+	if (ret == RX_HANDLER_PASS)
+		skb = skb_copy(orig_skb, GFP_ATOMIC);
+	else
+		skb = orig_skb;
+
+	/* CSME wants the MAC header as well, push it back */
+	skb_push(skb, skb->data - skb_mac_header(skb));
+
+	/*
+	 * Add the packet that CSME wants to get to the ring. Don't send the
+	 * Check Shared Area HECI message since this is not possible from the
+	 * Rx context. The caller will schedule a worker to do just that.
+	 */
+	iwl_mei_add_data_to_ring(skb, false);
+
+	/*
+	 * In case we drop the packet, don't free it, the caller will do that
+	 * for us
+	 */
+	if (ret == RX_HANDLER_PASS)
+		dev_kfree_skb(skb);
+
+	return ret;
+}
+
+#define DHCP_SERVER_PORT 67
+#define DHCP_CLIENT_PORT 68
+void iwl_mei_tx_copy_to_csme(struct sk_buff *origskb, unsigned int ivlen)
+{
+	struct ieee80211_hdr *hdr;
+	struct sk_buff *skb;
+	struct ethhdr ethhdr;
+	struct ethhdr *eth;
+
+	/* Catch DHCP packets */
+	if (origskb->protocol != htons(ETH_P_IP) ||
+	    ip_hdr(origskb)->protocol != IPPROTO_UDP ||
+	    udp_hdr(origskb)->source != htons(DHCP_CLIENT_PORT) ||
+	    udp_hdr(origskb)->dest != htons(DHCP_SERVER_PORT))
+		return;
+
+	/*
+	 * We could be a bit less aggressive here and not copy everything, but
+	 * this is very rare anyway, do don't bother much.
+	 */
+	skb = skb_copy(origskb, GFP_ATOMIC);
+	if (!skb)
+		return;
+
+	skb->protocol = origskb->protocol;
+
+	hdr = (void *)skb->data;
+
+	memcpy(ethhdr.h_dest, ieee80211_get_DA(hdr), ETH_ALEN);
+	memcpy(ethhdr.h_source, ieee80211_get_SA(hdr), ETH_ALEN);
+
+	/*
+	 * Remove the ieee80211 header + IV + SNAP but leave the ethertype
+	 * We still have enough headroom for the sap header.
+	 */
+	pskb_pull(skb, ieee80211_hdrlen(hdr->frame_control) + ivlen + 6);
+	eth = skb_push(skb, sizeof(ethhdr.h_dest) + sizeof(ethhdr.h_source));
+	memcpy(eth, &ethhdr, sizeof(ethhdr.h_dest) + sizeof(ethhdr.h_source));
+
+	iwl_mei_add_data_to_ring(skb, true);
+
+	dev_kfree_skb(skb);
+}
+EXPORT_SYMBOL_GPL(iwl_mei_tx_copy_to_csme);