Ratheesh Kannoth | 6307bec | 2021-11-25 08:26:39 +0530 | [diff] [blame^] | 1 | /* |
| 2 | * sfe_ipv4_udp.c |
| 3 | * Shortcut forwarding engine - IPv4 UDP implementation |
| 4 | * |
| 5 | * Copyright (c) 2013-2016, 2019-2020, The Linux Foundation. All rights reserved. |
| 6 | * Copyright (c) 2021 Qualcomm Innovation Center, Inc. All rights reserved. |
| 7 | * |
| 8 | * Permission to use, copy, modify, and/or distribute this software for any |
| 9 | * purpose with or without fee is hereby granted, provided that the above |
| 10 | * copyright notice and this permission notice appear in all copies. |
| 11 | * |
| 12 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
| 13 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF |
| 14 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR |
| 15 | * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
| 16 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN |
| 17 | * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF |
| 18 | * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
| 19 | */ |
| 20 | |
| 21 | #include <linux/skbuff.h> |
| 22 | #include <net/udp.h> |
| 23 | #include <linux/etherdevice.h> |
| 24 | #include <linux/lockdep.h> |
| 25 | |
| 26 | #include "sfe_debug.h" |
| 27 | #include "sfe_api.h" |
| 28 | #include "sfe.h" |
| 29 | #include "sfe_flow_cookie.h" |
| 30 | #include "sfe_ipv4.h" |
| 31 | |
| 32 | /* |
| 33 | * sfe_ipv4_recv_udp() |
| 34 | * Handle UDP packet receives and forwarding. |
| 35 | */ |
| 36 | int sfe_ipv4_recv_udp(struct sfe_ipv4 *si, struct sk_buff *skb, struct net_device *dev, |
| 37 | unsigned int len, struct iphdr *iph, unsigned int ihl, bool flush_on_find) |
| 38 | { |
| 39 | struct udphdr *udph; |
| 40 | __be32 src_ip; |
| 41 | __be32 dest_ip; |
| 42 | __be16 src_port; |
| 43 | __be16 dest_port; |
| 44 | struct sfe_ipv4_connection_match *cm; |
| 45 | u8 ttl; |
| 46 | struct net_device *xmit_dev; |
| 47 | bool ret; |
| 48 | |
| 49 | /* |
| 50 | * Is our packet too short to contain a valid UDP header? |
| 51 | */ |
| 52 | if (unlikely(!pskb_may_pull(skb, (sizeof(struct udphdr) + ihl)))) { |
| 53 | sfe_ipv4_exception_stats_inc(si, SFE_IPV4_EXCEPTION_EVENT_UDP_HEADER_INCOMPLETE); |
| 54 | DEBUG_TRACE("packet too short for UDP header\n"); |
| 55 | return 0; |
| 56 | } |
| 57 | |
| 58 | /* |
| 59 | * Read the IP address and port information. Read the IP header data first |
| 60 | * because we've almost certainly got that in the cache. We may not yet have |
| 61 | * the UDP header cached though so allow more time for any prefetching. |
| 62 | */ |
| 63 | src_ip = iph->saddr; |
| 64 | dest_ip = iph->daddr; |
| 65 | |
| 66 | udph = (struct udphdr *)(skb->data + ihl); |
| 67 | src_port = udph->source; |
| 68 | dest_port = udph->dest; |
| 69 | |
| 70 | rcu_read_lock(); |
| 71 | |
| 72 | /* |
| 73 | * Look for a connection match. |
| 74 | */ |
| 75 | #ifdef CONFIG_NF_FLOW_COOKIE |
| 76 | cm = si->sfe_flow_cookie_table[skb->flow_cookie & SFE_FLOW_COOKIE_MASK].match; |
| 77 | if (unlikely(!cm)) { |
| 78 | cm = sfe_ipv4_find_connection_match_rcu(si, dev, IPPROTO_UDP, src_ip, src_port, dest_ip, dest_port); |
| 79 | } |
| 80 | #else |
| 81 | cm = sfe_ipv4_find_connection_match_rcu(si, dev, IPPROTO_UDP, src_ip, src_port, dest_ip, dest_port); |
| 82 | #endif |
| 83 | if (unlikely(!cm)) { |
| 84 | |
| 85 | rcu_read_unlock(); |
| 86 | sfe_ipv4_exception_stats_inc(si, SFE_IPV4_EXCEPTION_EVENT_UDP_NO_CONNECTION); |
| 87 | DEBUG_TRACE("no connection found\n"); |
| 88 | return 0; |
| 89 | } |
| 90 | |
| 91 | /* |
| 92 | * If our packet has beern marked as "flush on find" we can't actually |
| 93 | * forward it in the fast path, but now that we've found an associated |
| 94 | * connection we can flush that out before we process the packet. |
| 95 | */ |
| 96 | if (unlikely(flush_on_find)) { |
| 97 | struct sfe_ipv4_connection *c = cm->connection; |
| 98 | spin_lock_bh(&si->lock); |
| 99 | ret = sfe_ipv4_remove_connection(si, c); |
| 100 | spin_unlock_bh(&si->lock); |
| 101 | |
| 102 | if (ret) { |
| 103 | sfe_ipv4_flush_connection(si, c, SFE_SYNC_REASON_FLUSH); |
| 104 | } |
| 105 | rcu_read_unlock(); |
| 106 | sfe_ipv4_exception_stats_inc(si, SFE_IPV4_EXCEPTION_EVENT_UDP_IP_OPTIONS_OR_INITIAL_FRAGMENT); |
| 107 | DEBUG_TRACE("flush on find\n"); |
| 108 | return 0; |
| 109 | } |
| 110 | |
| 111 | #ifdef CONFIG_XFRM |
| 112 | /* |
| 113 | * We can't accelerate the flow on this direction, just let it go |
| 114 | * through the slow path. |
| 115 | */ |
| 116 | if (unlikely(!cm->flow_accel)) { |
| 117 | rcu_read_unlock(); |
| 118 | this_cpu_inc(si->stats_pcpu->packets_not_forwarded64); |
| 119 | return 0; |
| 120 | } |
| 121 | #endif |
| 122 | |
| 123 | /* |
| 124 | * Does our TTL allow forwarding? |
| 125 | */ |
| 126 | ttl = iph->ttl; |
| 127 | if (unlikely(ttl < 2)) { |
| 128 | struct sfe_ipv4_connection *c = cm->connection; |
| 129 | spin_lock_bh(&si->lock); |
| 130 | ret = sfe_ipv4_remove_connection(si, c); |
| 131 | spin_unlock_bh(&si->lock); |
| 132 | |
| 133 | if (ret) { |
| 134 | sfe_ipv4_flush_connection(si, c, SFE_SYNC_REASON_FLUSH); |
| 135 | } |
| 136 | rcu_read_unlock(); |
| 137 | |
| 138 | DEBUG_TRACE("ttl too low\n"); |
| 139 | sfe_ipv4_exception_stats_inc(si, SFE_IPV4_EXCEPTION_EVENT_UDP_SMALL_TTL); |
| 140 | return 0; |
| 141 | } |
| 142 | |
| 143 | /* |
| 144 | * If our packet is larger than the MTU of the transmit interface then |
| 145 | * we can't forward it easily. |
| 146 | */ |
| 147 | if (unlikely(len > cm->xmit_dev_mtu)) { |
| 148 | struct sfe_ipv4_connection *c = cm->connection; |
| 149 | spin_lock_bh(&si->lock); |
| 150 | ret = sfe_ipv4_remove_connection(si, c); |
| 151 | spin_unlock_bh(&si->lock); |
| 152 | |
| 153 | DEBUG_TRACE("larger than mtu\n"); |
| 154 | if (ret) { |
| 155 | sfe_ipv4_flush_connection(si, c, SFE_SYNC_REASON_FLUSH); |
| 156 | } |
| 157 | rcu_read_unlock(); |
| 158 | sfe_ipv4_exception_stats_inc(si, SFE_IPV4_EXCEPTION_EVENT_UDP_NEEDS_FRAGMENTATION); |
| 159 | return 0; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * From this point on we're good to modify the packet. |
| 164 | */ |
| 165 | |
| 166 | /* |
| 167 | * Check if skb was cloned. If it was, unshare it. Because |
| 168 | * the data area is going to be written in this path and we don't want to |
| 169 | * change the cloned skb's data section. |
| 170 | */ |
| 171 | if (unlikely(skb_cloned(skb))) { |
| 172 | DEBUG_TRACE("%px: skb is a cloned skb\n", skb); |
| 173 | skb = skb_unshare(skb, GFP_ATOMIC); |
| 174 | if (!skb) { |
| 175 | DEBUG_WARN("Failed to unshare the cloned skb\n"); |
| 176 | rcu_read_unlock(); |
| 177 | return 0; |
| 178 | } |
| 179 | |
| 180 | /* |
| 181 | * Update the iph and udph pointers with the unshared skb's data area. |
| 182 | */ |
| 183 | iph = (struct iphdr *)skb->data; |
| 184 | udph = (struct udphdr *)(skb->data + ihl); |
| 185 | } |
| 186 | |
| 187 | /* |
| 188 | * Update DSCP |
| 189 | */ |
| 190 | if (unlikely(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_DSCP_REMARK)) { |
| 191 | iph->tos = (iph->tos & SFE_IPV4_DSCP_MASK) | cm->dscp; |
| 192 | } |
| 193 | |
| 194 | /* |
| 195 | * Decrement our TTL. |
| 196 | */ |
| 197 | iph->ttl = ttl - 1; |
| 198 | |
| 199 | /* |
| 200 | * Do we have to perform translations of the source address/port? |
| 201 | */ |
| 202 | if (unlikely(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_XLATE_SRC)) { |
| 203 | u16 udp_csum; |
| 204 | |
| 205 | iph->saddr = cm->xlate_src_ip; |
| 206 | udph->source = cm->xlate_src_port; |
| 207 | |
| 208 | /* |
| 209 | * Do we have a non-zero UDP checksum? If we do then we need |
| 210 | * to update it. |
| 211 | */ |
| 212 | udp_csum = udph->check; |
| 213 | if (likely(udp_csum)) { |
| 214 | u32 sum; |
| 215 | |
| 216 | if (unlikely(skb->ip_summed == CHECKSUM_PARTIAL)) { |
| 217 | sum = udp_csum + cm->xlate_src_partial_csum_adjustment; |
| 218 | } else { |
| 219 | sum = udp_csum + cm->xlate_src_csum_adjustment; |
| 220 | } |
| 221 | |
| 222 | sum = (sum & 0xffff) + (sum >> 16); |
| 223 | udph->check = (u16)sum; |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * Do we have to perform translations of the destination address/port? |
| 229 | */ |
| 230 | if (unlikely(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_XLATE_DEST)) { |
| 231 | u16 udp_csum; |
| 232 | |
| 233 | iph->daddr = cm->xlate_dest_ip; |
| 234 | udph->dest = cm->xlate_dest_port; |
| 235 | |
| 236 | /* |
| 237 | * Do we have a non-zero UDP checksum? If we do then we need |
| 238 | * to update it. |
| 239 | */ |
| 240 | udp_csum = udph->check; |
| 241 | if (likely(udp_csum)) { |
| 242 | u32 sum; |
| 243 | |
| 244 | if (unlikely(skb->ip_summed == CHECKSUM_PARTIAL)) { |
| 245 | sum = udp_csum + cm->xlate_dest_partial_csum_adjustment; |
| 246 | } else { |
| 247 | sum = udp_csum + cm->xlate_dest_csum_adjustment; |
| 248 | } |
| 249 | |
| 250 | sum = (sum & 0xffff) + (sum >> 16); |
| 251 | udph->check = (u16)sum; |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | /* |
| 256 | * Replace the IP checksum. |
| 257 | */ |
| 258 | iph->check = sfe_ipv4_gen_ip_csum(iph); |
| 259 | |
| 260 | /* |
| 261 | * Update traffic stats. |
| 262 | */ |
| 263 | atomic_inc(&cm->rx_packet_count); |
| 264 | atomic_add(len, &cm->rx_byte_count); |
| 265 | |
| 266 | xmit_dev = cm->xmit_dev; |
| 267 | skb->dev = xmit_dev; |
| 268 | |
| 269 | /* |
| 270 | * Check to see if we need to write a header. |
| 271 | */ |
| 272 | if (likely(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_WRITE_L2_HDR)) { |
| 273 | if (unlikely(!(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_WRITE_FAST_ETH_HDR))) { |
| 274 | dev_hard_header(skb, xmit_dev, ETH_P_IP, |
| 275 | cm->xmit_dest_mac, cm->xmit_src_mac, len); |
| 276 | } else { |
| 277 | /* |
| 278 | * For the simple case we write this really fast. |
| 279 | */ |
| 280 | struct ethhdr *eth = (struct ethhdr *)__skb_push(skb, ETH_HLEN); |
| 281 | eth->h_proto = htons(ETH_P_IP); |
| 282 | ether_addr_copy((u8 *)eth->h_dest, (u8 *)cm->xmit_dest_mac); |
| 283 | ether_addr_copy((u8 *)eth->h_source, (u8 *)cm->xmit_src_mac); |
| 284 | } |
| 285 | } |
| 286 | |
| 287 | /* |
| 288 | * Update priority of skb. |
| 289 | */ |
| 290 | if (unlikely(cm->flags & SFE_IPV4_CONNECTION_MATCH_FLAG_PRIORITY_REMARK)) { |
| 291 | skb->priority = cm->priority; |
| 292 | } |
| 293 | |
| 294 | /* |
| 295 | * Mark outgoing packet. |
| 296 | */ |
| 297 | skb->mark = cm->connection->mark; |
| 298 | if (skb->mark) { |
| 299 | DEBUG_TRACE("SKB MARK is NON ZERO %x\n", skb->mark); |
| 300 | } |
| 301 | |
| 302 | rcu_read_unlock(); |
| 303 | |
| 304 | this_cpu_inc(si->stats_pcpu->packets_forwarded64); |
| 305 | |
| 306 | /* |
| 307 | * We're going to check for GSO flags when we transmit the packet so |
| 308 | * start fetching the necessary cache line now. |
| 309 | */ |
| 310 | prefetch(skb_shinfo(skb)); |
| 311 | |
| 312 | /* |
| 313 | * Mark that this packet has been fast forwarded. |
| 314 | */ |
| 315 | skb->fast_forwarded = 1; |
| 316 | |
| 317 | /* |
| 318 | * Send the packet on its way. |
| 319 | */ |
| 320 | dev_queue_xmit(skb); |
| 321 | |
| 322 | return 1; |
| 323 | } |