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 #include <net/rstreason.h> 19 20 #include "ackvec.h" 21 #include "ccid.h" 22 #include "dccp.h" 23 #include "feat.h" 24 25 struct inet_timewait_death_row dccp_death_row = { 26 .tw_refcount = REFCOUNT_INIT(1), 27 .sysctl_max_tw_buckets = NR_FILE * 2, 28 .hashinfo = &dccp_hashinfo, 29 }; 30 31 EXPORT_SYMBOL_GPL(dccp_death_row); 32 33 void dccp_time_wait(struct sock *sk, int state, int timeo) 34 { 35 struct inet_timewait_sock *tw; 36 37 tw = inet_twsk_alloc(sk, &dccp_death_row, state); 38 39 if (tw != NULL) { 40 const struct inet_connection_sock *icsk = inet_csk(sk); 41 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); 42 #if IS_ENABLED(CONFIG_IPV6) 43 if (tw->tw_family == PF_INET6) { 44 tw->tw_v6_daddr = sk->sk_v6_daddr; 45 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 46 tw->tw_ipv6only = sk->sk_ipv6only; 47 } 48 #endif 49 50 /* Get the TIME_WAIT timeout firing. */ 51 if (timeo < rto) 52 timeo = rto; 53 54 if (state == DCCP_TIME_WAIT) 55 timeo = DCCP_TIMEWAIT_LEN; 56 57 /* tw_timer is pinned, so we need to make sure BH are disabled 58 * in following section, otherwise timer handler could run before 59 * we complete the initialization. 60 */ 61 local_bh_disable(); 62 inet_twsk_schedule(tw, timeo); 63 /* Linkage updates. 64 * Note that access to tw after this point is illegal. 65 */ 66 inet_twsk_hashdance(tw, sk, &dccp_hashinfo); 67 local_bh_enable(); 68 } else { 69 /* Sorry, if we're out of memory, just CLOSE this 70 * socket up. We've got bigger problems than 71 * non-graceful socket closings. 72 */ 73 DCCP_WARN("time wait bucket table overflow\n"); 74 } 75 76 dccp_done(sk); 77 } 78 79 struct sock *dccp_create_openreq_child(const struct sock *sk, 80 const struct request_sock *req, 81 const struct sk_buff *skb) 82 { 83 /* 84 * Step 3: Process LISTEN state 85 * 86 * (* Generate a new socket and switch to that socket *) 87 * Set S := new socket for this port pair 88 */ 89 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC); 90 91 if (newsk != NULL) { 92 struct dccp_request_sock *dreq = dccp_rsk(req); 93 struct inet_connection_sock *newicsk = inet_csk(newsk); 94 struct dccp_sock *newdp = dccp_sk(newsk); 95 96 newdp->dccps_role = DCCP_ROLE_SERVER; 97 newdp->dccps_hc_rx_ackvec = NULL; 98 newdp->dccps_service_list = NULL; 99 newdp->dccps_hc_rx_ccid = NULL; 100 newdp->dccps_hc_tx_ccid = NULL; 101 newdp->dccps_service = dreq->dreq_service; 102 newdp->dccps_timestamp_echo = dreq->dreq_timestamp_echo; 103 newdp->dccps_timestamp_time = dreq->dreq_timestamp_time; 104 newicsk->icsk_rto = DCCP_TIMEOUT_INIT; 105 106 INIT_LIST_HEAD(&newdp->dccps_featneg); 107 /* 108 * Step 3: Process LISTEN state 109 * 110 * Choose S.ISS (initial seqno) or set from Init Cookies 111 * Initialize S.GAR := S.ISS 112 * Set S.ISR, S.GSR from packet (or Init Cookies) 113 * 114 * Setting AWL/AWH and SWL/SWH happens as part of the feature 115 * activation below, as these windows all depend on the local 116 * and remote Sequence Window feature values (7.5.2). 117 */ 118 newdp->dccps_iss = dreq->dreq_iss; 119 newdp->dccps_gss = dreq->dreq_gss; 120 newdp->dccps_gar = newdp->dccps_iss; 121 newdp->dccps_isr = dreq->dreq_isr; 122 newdp->dccps_gsr = dreq->dreq_gsr; 123 124 /* 125 * Activate features: initialise CCIDs, sequence windows etc. 126 */ 127 if (dccp_feat_activate_values(newsk, &dreq->dreq_featneg)) { 128 sk_free_unlock_clone(newsk); 129 return NULL; 130 } 131 dccp_init_xmit_timers(newsk); 132 133 __DCCP_INC_STATS(DCCP_MIB_PASSIVEOPENS); 134 } 135 return newsk; 136 } 137 138 EXPORT_SYMBOL_GPL(dccp_create_openreq_child); 139 140 /* 141 * Process an incoming packet for RESPOND sockets represented 142 * as an request_sock. 143 */ 144 struct sock *dccp_check_req(struct sock *sk, struct sk_buff *skb, 145 struct request_sock *req) 146 { 147 struct sock *child = NULL; 148 struct dccp_request_sock *dreq = dccp_rsk(req); 149 bool own_req; 150 151 /* TCP/DCCP listeners became lockless. 152 * DCCP stores complex state in its request_sock, so we need 153 * a protection for them, now this code runs without being protected 154 * by the parent (listener) lock. 155 */ 156 spin_lock_bh(&dreq->dreq_lock); 157 158 /* Check for retransmitted REQUEST */ 159 if (dccp_hdr(skb)->dccph_type == DCCP_PKT_REQUEST) { 160 161 if (after48(DCCP_SKB_CB(skb)->dccpd_seq, dreq->dreq_gsr)) { 162 dccp_pr_debug("Retransmitted REQUEST\n"); 163 dreq->dreq_gsr = DCCP_SKB_CB(skb)->dccpd_seq; 164 /* 165 * Send another RESPONSE packet 166 * To protect against Request floods, increment retrans 167 * counter (backoff, monitored by dccp_response_timer). 168 */ 169 inet_rtx_syn_ack(sk, req); 170 } 171 /* Network Duplicate, discard packet */ 172 goto out; 173 } 174 175 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_PACKET_ERROR; 176 177 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_ACK && 178 dccp_hdr(skb)->dccph_type != DCCP_PKT_DATAACK) 179 goto drop; 180 181 /* Invalid ACK */ 182 if (!between48(DCCP_SKB_CB(skb)->dccpd_ack_seq, 183 dreq->dreq_iss, dreq->dreq_gss)) { 184 dccp_pr_debug("Invalid ACK number: ack_seq=%llu, " 185 "dreq_iss=%llu, dreq_gss=%llu\n", 186 (unsigned long long) 187 DCCP_SKB_CB(skb)->dccpd_ack_seq, 188 (unsigned long long) dreq->dreq_iss, 189 (unsigned long long) dreq->dreq_gss); 190 goto drop; 191 } 192 193 if (dccp_parse_options(sk, dreq, skb)) 194 goto drop; 195 196 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL, 197 req, &own_req); 198 if (child) { 199 child = inet_csk_complete_hashdance(sk, child, req, own_req); 200 goto out; 201 } 202 203 DCCP_SKB_CB(skb)->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY; 204 drop: 205 if (dccp_hdr(skb)->dccph_type != DCCP_PKT_RESET) 206 req->rsk_ops->send_reset(sk, skb, SK_RST_REASON_NOT_SPECIFIED); 207 208 inet_csk_reqsk_queue_drop(sk, req); 209 out: 210 spin_unlock_bh(&dreq->dreq_lock); 211 return child; 212 } 213 214 EXPORT_SYMBOL_GPL(dccp_check_req); 215 216 /* 217 * Queue segment on the new socket if the new socket is active, 218 * otherwise we just shortcircuit this and continue with 219 * the new socket. 220 */ 221 int dccp_child_process(struct sock *parent, struct sock *child, 222 struct sk_buff *skb) 223 __releases(child) 224 { 225 int ret = 0; 226 const int state = child->sk_state; 227 228 if (!sock_owned_by_user(child)) { 229 ret = dccp_rcv_state_process(child, skb, dccp_hdr(skb), 230 skb->len); 231 232 /* Wakeup parent, send SIGIO */ 233 if (state == DCCP_RESPOND && child->sk_state != state) 234 parent->sk_data_ready(parent); 235 } else { 236 /* Alas, it is possible again, because we do lookup 237 * in main socket hash table and lock on listening 238 * socket does not protect us more. 239 */ 240 __sk_add_backlog(child, skb); 241 } 242 243 bh_unlock_sock(child); 244 sock_put(child); 245 return ret; 246 } 247 248 EXPORT_SYMBOL_GPL(dccp_child_process); 249 250 void dccp_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, 251 struct request_sock *rsk) 252 { 253 DCCP_BUG("DCCP-ACK packets are never sent in LISTEN/RESPOND state"); 254 } 255 256 EXPORT_SYMBOL_GPL(dccp_reqsk_send_ack); 257 258 int dccp_reqsk_init(struct request_sock *req, 259 struct dccp_sock const *dp, struct sk_buff const *skb) 260 { 261 struct dccp_request_sock *dreq = dccp_rsk(req); 262 263 spin_lock_init(&dreq->dreq_lock); 264 inet_rsk(req)->ir_rmt_port = dccp_hdr(skb)->dccph_sport; 265 inet_rsk(req)->ir_num = ntohs(dccp_hdr(skb)->dccph_dport); 266 inet_rsk(req)->acked = 0; 267 dreq->dreq_timestamp_echo = 0; 268 269 /* inherit feature negotiation options from listening socket */ 270 return dccp_feat_clone_list(&dp->dccps_featneg, &dreq->dreq_featneg); 271 } 272 273 EXPORT_SYMBOL_GPL(dccp_reqsk_init); 274