1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/dccp/minisocks.c 4 * 5 * An implementation of the DCCP protocol 6 * Arnaldo Carvalho de Melo <acme@conectiva.com.br> 7 */ 8 9 #include <linux/dccp.h> 10 #include <linux/gfp.h> 11 #include <linux/kernel.h> 12 #include <linux/skbuff.h> 13 #include <linux/timer.h> 14 15 #include <net/sock.h> 16 #include <net/xfrm.h> 17 #include <net/inet_timewait_sock.h> 18 19 #include "ackvec.h" 20 #include "ccid.h" 21 #include "dccp.h" 22 #include "feat.h" 23 24 struct inet_timewait_death_row dccp_death_row = { 25 .sysctl_max_tw_buckets = NR_FILE * 2, 26 .hashinfo = &dccp_hashinfo, 27 }; 28 29 EXPORT_SYMBOL_GPL(dccp_death_row); 30 31 void dccp_time_wait(struct sock *sk, int state, int timeo) 32 { 33 struct inet_timewait_sock *tw; 34 35 tw = inet_twsk_alloc(sk, &dccp_death_row, state); 36 37 if (tw != NULL) { 38 const struct inet_connection_sock *icsk = inet_csk(sk); 39 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); 40 #if IS_ENABLED(CONFIG_IPV6) 41 if (tw->tw_family == PF_INET6) { 42 tw->tw_v6_daddr = sk->sk_v6_daddr; 43 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 44 tw->tw_ipv6only = sk->sk_ipv6only; 45 } 46 #endif 47 48 /* Get the TIME_WAIT timeout firing. */ 49 if (timeo < rto) 50 timeo = rto; 51 52 if (state == DCCP_TIME_WAIT) 53 timeo = DCCP_TIMEWAIT_LEN; 54 55 /* tw_timer is pinned, so we need to make sure BH are disabled 56 * in following section, otherwise timer handler could run before 57 * we complete the initialization. 58 */ 59 local_bh_disable(); 60 inet_twsk_schedule(tw, timeo); 61 /* Linkage updates. 62 * Note that access to tw after this point is illegal. 63 */ 64 inet_twsk_hashdance(tw, sk, &dccp_hashinfo); 65 local_bh_enable(); 66 } else { 67 /* Sorry, if we're out of memory, just CLOSE this 68 * socket up. We've got bigger problems than 69 * non-graceful socket closings. 70 */ 71 DCCP_WARN("time wait bucket table overflow\n"); 72 } 73 74 dccp_done(sk); 75 } 76 77 struct sock *dccp_create_openreq_child(const struct sock *sk, 78 const struct request_sock *req, 79 const struct sk_buff *skb) 80 { 81 /* 82 * Step 3: Process LISTEN state 83 * 84 * (* Generate a new socket and switch to that socket *) 85 * Set S := new socket for this port pair 86 */ 87 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC); 88 89 if (newsk != NULL) { 90 struct dccp_request_sock *dreq = dccp_rsk(req); 91 struct inet_connection_sock *newicsk = inet_csk(newsk); 92 struct dccp_sock *newdp = dccp_sk(newsk); 93 94 newdp->dccps_role = DCCP_ROLE_SERVER; 95 newdp->dccps_hc_rx_ackvec = NULL; 96 newdp->dccps_service_list = NULL; 97 newdp->dccps_service = dreq->dreq_service; 98 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo; 99 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time; 100 newicsk->icsk_rto = DCCP_TIMEOUT_INIT; 101 102 INIT_LIST_HEAD(&newdp->dccps_featneg); 103 /* 104 * Step 3: Process LISTEN state 105 * 106 * Choose S.ISS (initial seqno) or set from Init Cookies 107 * Initialize S.GAR := S.ISS 108 * Set S.ISR, S.GSR from packet (or Init Cookies) 109 * 110 * Setting AWL/AWH and SWL/SWH happens as part of the feature 111 * activation below, as these windows all depend on the local 112 * and remote Sequence Window feature values (7.5.2). 113 */ 114 newdp->dccps_iss = dreq->dreq_iss; 115 newdp->dccps_gss = dreq->dreq_gss; 116 newdp->dccps_gar = newdp->dccps_iss; 117 newdp->dccps_isr = dreq->dreq_isr; 118 newdp->dccps_gsr = dreq->dreq_gsr; 119 120 /* 121 * Activate features: initialise CCIDs, sequence windows etc. 122 */ 123 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) { 124 sk_free_unlock_clone(newsk); 125 return NULL; 126 } 127 dccp_init_xmit_timers(newsk); 128 129 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS); 130 } 131 return newsk; 132 } 133 134 EXPORT_SYMBOL_GPL(dccp_create_openreq_child); 135 136 /* 137 * Process an incoming packet for RESPOND sockets represented 138 * as an request_sock. 139 */ 140 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb, 141 struct request_sock *req) 142 { 143 struct sock *child = NULL; 144 struct dccp_request_sock *dreq = dccp_rsk(req); 145 bool own_req; 146 147 /* TCP/DCCP listeners became lockless. 148 * DCCP stores complex state in its request_sock, so we need 149 * a protection for them, now this code runs without being protected 150 * by the parent (listener) lock. 151 */ 152 spin_lock_bh(&dreq->dreq_lock); 153 154 /* Check for retransmitted REQUEST */ 155 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) { 156 157 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) { 158 dccp_pr_debug("Retransmitted REQUEST\n"); 159 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq; 160 /* 161 * Send another RESPONSE packet 162 * To protect against Request floods, increment retrans 163 * counter (backoff, monitored by dccp_response_timer). 164 */ 165 inet_rtx_syn_ack(sk, req); 166 } 167 /* Network Duplicate, discard packet */ 168 goto out; 169 } 170 171 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR; 172 173 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK && 174 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK) 175 goto drop; 176 177 /* Invalid ACK */ 178 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 179 dreq->dreq_iss, dreq->dreq_gss)) { 180 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, " 181 "dreq_iss=%llu, dreq_gss=%llu\n", 182 (unsigned long long) 183 DCCP_SKB_CB(skb)->dccpd_ack_seq, 184 (unsigned long long) dreq->dreq_iss, 185 (unsigned long long) dreq->dreq_gss); 186 goto drop; 187 } 188 189 if (dccp_parse_options(sk, dreq, skb)) 190 goto drop; 191 192 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL, 193 req, &own_req); 194 if (child) { 195 child = inet_csk_complete_hashdance(sk, child, req, own_req); 196 goto out; 197 } 198 199 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 200 drop: 201 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET) 202 req->rsk_ops->send_reset(sk, skb); 203 204 inet_csk_reqsk_queue_drop(sk, req); 205 out: 206 spin_unlock_bh(&dreq->dreq_lock); 207 return child; 208 } 209 210 EXPORT_SYMBOL_GPL(dccp_check_req); 211 212 /* 213 * Queue segment on the new socket if the new socket is active, 214 * otherwise we just shortcircuit this and continue with 215 * the new socket. 216 */ 217 int dccp_child_process(struct sock *parent, struct sock *child, 218 struct sk_buff *skb) 219 { 220 int ret = 0; 221 const int state = child->sk_state; 222 223 if (!sock_owned_by_user(child)) { 224 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb), 225 skb->len); 226 227 /* Wakeup parent, send SIGIO */ 228 if (state == DCCP_RESPOND && child->sk_state != state) 229 parent->sk_data_ready(parent); 230 } else { 231 /* Alas, it is possible again, because we do lookup 232 * in main socket hash table and lock on listening 233 * socket does not protect us more. 234 */ 235 __sk_add_backlog(child, skb); 236 } 237 238 bh_unlock_sock(child); 239 sock_put(child); 240 return ret; 241 } 242 243 EXPORT_SYMBOL_GPL(dccp_child_process); 244 245 void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 246 struct request_sock *rsk) 247 { 248 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state"); 249 } 250 251 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack); 252 253 int dccp_reqsk_init(struct request_sock *req, 254 struct dccp_sock const *dp, struct sk_buff const *skb) 255 { 256 struct dccp_request_sock *dreq = dccp_rsk(req); 257 258 spin_lock_init(&dreq->dreq_lock); 259 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport; 260 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport); 261 inet_rsk(req)->acked = 0; 262 dreq->dreq_timestamp_echo = 0; 263 264 /* inherit feature negotiation options from listening socket */ 265 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg); 266 } 267 268 EXPORT_SYMBOL_GPL(dccp_reqsk_init); 269