| 1 | /* uipc_socket2.c 4.28 82/10/09 */ |
| 2 | |
| 3 | #include "../h/param.h" |
| 4 | #include "../h/systm.h" |
| 5 | #include "../h/dir.h" |
| 6 | #include "../h/user.h" |
| 7 | #include "../h/proc.h" |
| 8 | #include "../h/file.h" |
| 9 | #include "../h/inode.h" |
| 10 | #include "../h/buf.h" |
| 11 | #include "../h/mbuf.h" |
| 12 | #include "../h/protosw.h" |
| 13 | #include "../h/socket.h" |
| 14 | #include "../h/socketvar.h" |
| 15 | |
| 16 | /* |
| 17 | * Primitive routines for operating on sockets and socket buffers |
| 18 | */ |
| 19 | |
| 20 | /* |
| 21 | * Procedures to manipulate state flags of socket |
| 22 | * and do appropriate wakeups. Normal sequence from the |
| 23 | * active (originating) side is that soisconnecting() is |
| 24 | * called during processing of connect() call, |
| 25 | * resulting in an eventual call to soisconnected() if/when the |
| 26 | * connection is established. When the connection is torn down |
| 27 | * soisdisconnecting() is called during processing of disconnect() call, |
| 28 | * and soisdisconnected() is called when the connection to the peer |
| 29 | * is totally severed. The semantics of these routines are such that |
| 30 | * connectionless protocols can call soisconnected() and soisdisconnected() |
| 31 | * only, bypassing the in-progress calls when setting up a ``connection'' |
| 32 | * takes no time. |
| 33 | * |
| 34 | * From the passive side, a socket is created with SO_ACCEPTCONN |
| 35 | * creating two queues of sockets: so_q0 for connections in progress |
| 36 | * and so_q for connections already made and awaiting user acceptance. |
| 37 | * As a protocol is preparing incoming connections, it creates a socket |
| 38 | * structure queued on so_q0 by calling sonewconn(). When the connection |
| 39 | * is established, soisconnected() is called, and transfers the |
| 40 | * socket structure to so_q, making it available to accept(). |
| 41 | * |
| 42 | * If a SO_ACCEPTCONN socket is closed with sockets on either |
| 43 | * so_q0 or so_q, these sockets are dropped. |
| 44 | * |
| 45 | * If and when higher level protocols are implemented in |
| 46 | * the kernel, the wakeups done here will sometimes |
| 47 | * be implemented as software-interrupt process scheduling. |
| 48 | */ |
| 49 | |
| 50 | soisconnecting(so) |
| 51 | struct socket *so; |
| 52 | { |
| 53 | |
| 54 | so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); |
| 55 | so->so_state |= SS_ISCONNECTING; |
| 56 | wakeup((caddr_t)&so->so_timeo); |
| 57 | } |
| 58 | |
| 59 | soisconnected(so) |
| 60 | struct socket *so; |
| 61 | { |
| 62 | register struct socket *head = so->so_head; |
| 63 | |
| 64 | if (head) { |
| 65 | if (soqremque(so, 0) == 0) |
| 66 | panic("soisconnected"); |
| 67 | soqinsque(head, so, 1); |
| 68 | wakeup((caddr_t)&head->so_timeo); |
| 69 | } |
| 70 | so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING); |
| 71 | so->so_state |= SS_ISCONNECTED; |
| 72 | wakeup((caddr_t)&so->so_timeo); |
| 73 | sorwakeup(so); |
| 74 | sowwakeup(so); |
| 75 | } |
| 76 | |
| 77 | soisdisconnecting(so) |
| 78 | struct socket *so; |
| 79 | { |
| 80 | |
| 81 | so->so_state &= ~SS_ISCONNECTING; |
| 82 | so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); |
| 83 | wakeup((caddr_t)&so->so_timeo); |
| 84 | sowwakeup(so); |
| 85 | sorwakeup(so); |
| 86 | } |
| 87 | |
| 88 | soisdisconnected(so) |
| 89 | struct socket *so; |
| 90 | { |
| 91 | |
| 92 | so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); |
| 93 | so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE); |
| 94 | wakeup((caddr_t)&so->so_timeo); |
| 95 | sowwakeup(so); |
| 96 | sorwakeup(so); |
| 97 | } |
| 98 | |
| 99 | /* |
| 100 | * When an attempt at a new connection is noted on a socket |
| 101 | * which accepts connections, sonewconn is called. If the |
| 102 | * connection is possible (subject to space constraints, etc.) |
| 103 | * then we allocate a new structure, propoerly linked into the |
| 104 | * data structure of the original socket, and return this. |
| 105 | */ |
| 106 | struct socket * |
| 107 | sonewconn(head) |
| 108 | register struct socket *head; |
| 109 | { |
| 110 | register struct socket *so; |
| 111 | struct mbuf *m; |
| 112 | |
| 113 | if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) |
| 114 | goto bad; |
| 115 | m = m_getclr(M_DONTWAIT); |
| 116 | if (m == 0) |
| 117 | goto bad; |
| 118 | so = mtod(m, struct socket *); |
| 119 | so->so_type = head->so_type; |
| 120 | so->so_options = head->so_options &~ SO_ACCEPTCONN; |
| 121 | so->so_linger = head->so_linger; |
| 122 | so->so_state = head->so_state; |
| 123 | so->so_proto = head->so_proto; |
| 124 | so->so_timeo = head->so_timeo; |
| 125 | so->so_pgrp = head->so_pgrp; |
| 126 | soqinsque(head, so, 0); |
| 127 | if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH, 0, 0, 0)) { |
| 128 | (void) soqremque(so, 0); |
| 129 | m_free(m); |
| 130 | goto bad; |
| 131 | } |
| 132 | return (so); |
| 133 | bad: |
| 134 | return ((struct socket *)0); |
| 135 | } |
| 136 | |
| 137 | soqinsque(head, so, q) |
| 138 | register struct socket *head, *so; |
| 139 | int q; |
| 140 | { |
| 141 | |
| 142 | so->so_head = head; |
| 143 | if (q == 0) { |
| 144 | head->so_q0len++; |
| 145 | so->so_q0 = head->so_q0; |
| 146 | head->so_q0 = so; |
| 147 | } else { |
| 148 | head->so_qlen++; |
| 149 | so->so_q = head->so_q; |
| 150 | head->so_q = so; |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | soqremque(so, q) |
| 155 | register struct socket *so; |
| 156 | int q; |
| 157 | { |
| 158 | register struct socket *head, *prev, *next; |
| 159 | |
| 160 | head = so->so_head; |
| 161 | prev = head; |
| 162 | for (;;) { |
| 163 | next = q ? prev->so_q : prev->so_q0; |
| 164 | if (next == so) |
| 165 | break; |
| 166 | if (next == head) |
| 167 | return (0); |
| 168 | prev = next; |
| 169 | } |
| 170 | if (q == 0) { |
| 171 | prev->so_q0 = next->so_q0; |
| 172 | head->so_q0len--; |
| 173 | } else { |
| 174 | prev->so_q = next->so_q; |
| 175 | head->so_qlen--; |
| 176 | } |
| 177 | next->so_q0 = next->so_q = 0; |
| 178 | next->so_head = 0; |
| 179 | return (1); |
| 180 | } |
| 181 | |
| 182 | /* |
| 183 | * Socantsendmore indicates that no more data will be sent on the |
| 184 | * socket; it would normally be applied to a socket when the user |
| 185 | * informs the system that no more data is to be sent, by the protocol |
| 186 | * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data |
| 187 | * will be received, and will normally be applied to the socket by a |
| 188 | * protocol when it detects that the peer will send no more data. |
| 189 | * Data queued for reading in the socket may yet be read. |
| 190 | */ |
| 191 | |
| 192 | socantsendmore(so) |
| 193 | struct socket *so; |
| 194 | { |
| 195 | |
| 196 | so->so_state |= SS_CANTSENDMORE; |
| 197 | sowwakeup(so); |
| 198 | } |
| 199 | |
| 200 | socantrcvmore(so) |
| 201 | struct socket *so; |
| 202 | { |
| 203 | |
| 204 | so->so_state |= SS_CANTRCVMORE; |
| 205 | sorwakeup(so); |
| 206 | } |
| 207 | |
| 208 | /* |
| 209 | * Socket select/wakeup routines. |
| 210 | */ |
| 211 | |
| 212 | /* |
| 213 | * Interface routine to select() system |
| 214 | * call for sockets. |
| 215 | */ |
| 216 | soselect(so, rw) |
| 217 | register struct socket *so; |
| 218 | int rw; |
| 219 | { |
| 220 | int s = splnet(); |
| 221 | |
| 222 | switch (rw) { |
| 223 | |
| 224 | case FREAD: |
| 225 | if (soreadable(so)) { |
| 226 | splx(s); |
| 227 | return (1); |
| 228 | } |
| 229 | sbselqueue(&so->so_rcv); |
| 230 | break; |
| 231 | |
| 232 | case FWRITE: |
| 233 | if (sowriteable(so)) { |
| 234 | splx(s); |
| 235 | return (1); |
| 236 | } |
| 237 | sbselqueue(&so->so_snd); |
| 238 | break; |
| 239 | } |
| 240 | splx(s); |
| 241 | return (0); |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * Queue a process for a select on a socket buffer. |
| 246 | */ |
| 247 | sbselqueue(sb) |
| 248 | struct sockbuf *sb; |
| 249 | { |
| 250 | register struct proc *p; |
| 251 | |
| 252 | if ((p = sb->sb_sel) && p->p_wchan == (caddr_t)&selwait) |
| 253 | sb->sb_flags |= SB_COLL; |
| 254 | else |
| 255 | sb->sb_sel = u.u_procp; |
| 256 | } |
| 257 | |
| 258 | /* |
| 259 | * Wait for data to arrive at/drain from a socket buffer. |
| 260 | */ |
| 261 | sbwait(sb) |
| 262 | struct sockbuf *sb; |
| 263 | { |
| 264 | |
| 265 | sb->sb_flags |= SB_WAIT; |
| 266 | sleep((caddr_t)&sb->sb_cc, PZERO+1); |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * Wakeup processes waiting on a socket buffer. |
| 271 | */ |
| 272 | sbwakeup(sb) |
| 273 | struct sockbuf *sb; |
| 274 | { |
| 275 | |
| 276 | if (sb->sb_sel) { |
| 277 | selwakeup(sb->sb_sel, sb->sb_flags & SB_COLL); |
| 278 | sb->sb_sel = 0; |
| 279 | sb->sb_flags &= ~SB_COLL; |
| 280 | } |
| 281 | if (sb->sb_flags & SB_WAIT) { |
| 282 | sb->sb_flags &= ~SB_WAIT; |
| 283 | wakeup((caddr_t)&sb->sb_cc); |
| 284 | } |
| 285 | } |
| 286 | |
| 287 | /* |
| 288 | * Socket buffer (struct sockbuf) utility routines. |
| 289 | * |
| 290 | * Each socket contains two socket buffers: one for sending data and |
| 291 | * one for receiving data. Each buffer contains a queue of mbufs, |
| 292 | * information about the number of mbufs and amount of data in the |
| 293 | * queue, and other fields allowing select() statements and notification |
| 294 | * on data availability to be implemented. |
| 295 | * |
| 296 | * Before using a new socket structure it is first necessary to reserve |
| 297 | * buffer space to the socket, by calling sbreserve. This commits |
| 298 | * some of the available buffer space in the system buffer pool for the |
| 299 | * socket. The space should be released by calling sbrelease when the |
| 300 | * socket is destroyed. |
| 301 | * |
| 302 | * The routine sbappend() is normally called to append new mbufs |
| 303 | * to a socket buffer, after checking that adequate space is available |
| 304 | * comparing the function spspace() with the amount of data to be added. |
| 305 | * Data is normally removed from a socket buffer in a protocol by |
| 306 | * first calling m_copy on the socket buffer mbuf chain and sending this |
| 307 | * to a peer, and then removing the data from the socket buffer with |
| 308 | * sbdrop when the data is acknowledged by the peer (or immediately |
| 309 | * in the case of unreliable protocols.) |
| 310 | * |
| 311 | * Protocols which do not require connections place both source address |
| 312 | * and data information in socket buffer queues. The source addresses |
| 313 | * are stored in single mbufs after each data item, and are easily found |
| 314 | * as the data items are all marked with end of record markers. The |
| 315 | * sbappendaddr() routine stores a datum and associated address in |
| 316 | * a socket buffer. Note that, unlike sbappend(), this routine checks |
| 317 | * for the caller that there will be enough space to store the data. |
| 318 | * It fails if there is not enough space, or if it cannot find |
| 319 | * a mbuf to store the address in. |
| 320 | * |
| 321 | * The higher-level routines sosend and soreceive (in socket.c) |
| 322 | * also add data to, and remove data from socket buffers repectively. |
| 323 | */ |
| 324 | |
| 325 | /* |
| 326 | * Allot mbufs to a sockbuf. |
| 327 | */ |
| 328 | sbreserve(sb, cc) |
| 329 | struct sockbuf *sb; |
| 330 | { |
| 331 | |
| 332 | /* someday maybe this routine will fail... */ |
| 333 | sb->sb_hiwat = cc; |
| 334 | sb->sb_mbmax = cc*2; |
| 335 | return (1); |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Free mbufs held by a socket, and reserved mbuf space. |
| 340 | */ |
| 341 | sbrelease(sb) |
| 342 | struct sockbuf *sb; |
| 343 | { |
| 344 | |
| 345 | sbflush(sb); |
| 346 | sb->sb_hiwat = sb->sb_mbmax = 0; |
| 347 | } |
| 348 | |
| 349 | /* |
| 350 | * Routines to add (at the end) and remove (from the beginning) |
| 351 | * data from a mbuf queue. |
| 352 | */ |
| 353 | |
| 354 | /* |
| 355 | * Append mbuf queue m to sockbuf sb. |
| 356 | */ |
| 357 | sbappend(sb, m) |
| 358 | register struct mbuf *m; |
| 359 | register struct sockbuf *sb; |
| 360 | { |
| 361 | register struct mbuf *n; |
| 362 | |
| 363 | n = sb->sb_mb; |
| 364 | if (n) |
| 365 | while (n->m_next) |
| 366 | n = n->m_next; |
| 367 | while (m) { |
| 368 | if (m->m_len == 0 && (int)m->m_act == 0) { |
| 369 | m = m_free(m); |
| 370 | continue; |
| 371 | } |
| 372 | if (n && n->m_off <= MMAXOFF && m->m_off <= MMAXOFF && |
| 373 | (int)n->m_act == 0 && (int)m->m_act == 0 && |
| 374 | (n->m_off + n->m_len + m->m_len) <= MMAXOFF) { |
| 375 | bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, |
| 376 | (unsigned)m->m_len); |
| 377 | n->m_len += m->m_len; |
| 378 | sb->sb_cc += m->m_len; |
| 379 | m = m_free(m); |
| 380 | continue; |
| 381 | } |
| 382 | sballoc(sb, m); |
| 383 | if (n == 0) |
| 384 | sb->sb_mb = m; |
| 385 | else |
| 386 | n->m_next = m; |
| 387 | n = m; |
| 388 | m = m->m_next; |
| 389 | n->m_next = 0; |
| 390 | } |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * Append data and address. |
| 395 | * Return 0 if no space in sockbuf or if |
| 396 | * can't get mbuf to stuff address in. |
| 397 | */ |
| 398 | sbappendaddr(sb, asa, m0) |
| 399 | struct sockbuf *sb; |
| 400 | struct sockaddr *asa; |
| 401 | struct mbuf *m0; |
| 402 | { |
| 403 | struct sockaddr *msa; |
| 404 | register struct mbuf *m; |
| 405 | register int len = sizeof (struct sockaddr); |
| 406 | |
| 407 | m = m0; |
| 408 | if (m == 0) |
| 409 | panic("sbappendaddr"); |
| 410 | for (;;) { |
| 411 | len += m->m_len; |
| 412 | if (m->m_next == 0) { |
| 413 | m->m_act = (struct mbuf *)1; |
| 414 | break; |
| 415 | } |
| 416 | m = m->m_next; |
| 417 | } |
| 418 | if (len > sbspace(sb)) |
| 419 | return (0); |
| 420 | m = m_get(M_DONTWAIT); |
| 421 | if (m == 0) |
| 422 | return (0); |
| 423 | m->m_len = sizeof (struct sockaddr); |
| 424 | msa = mtod(m, struct sockaddr *); |
| 425 | *msa = *asa; |
| 426 | m->m_act = (struct mbuf *)1; |
| 427 | sbappend(sb, m); |
| 428 | sbappend(sb, m0); |
| 429 | return (1); |
| 430 | } |
| 431 | |
| 432 | /* |
| 433 | * Free all mbufs on a sockbuf mbuf chain. |
| 434 | * Check that resource allocations return to 0. |
| 435 | */ |
| 436 | sbflush(sb) |
| 437 | struct sockbuf *sb; |
| 438 | { |
| 439 | |
| 440 | if (sb->sb_flags & SB_LOCK) |
| 441 | panic("sbflush"); |
| 442 | if (sb->sb_cc) |
| 443 | sbdrop(sb, sb->sb_cc); |
| 444 | if (sb->sb_cc || sb->sb_mbcnt || sb->sb_mb) |
| 445 | panic("sbflush 2"); |
| 446 | } |
| 447 | |
| 448 | /* |
| 449 | * Drop data from (the front of) a sockbuf chain. |
| 450 | */ |
| 451 | sbdrop(sb, len) |
| 452 | register struct sockbuf *sb; |
| 453 | register int len; |
| 454 | { |
| 455 | register struct mbuf *m = sb->sb_mb, *mn; |
| 456 | |
| 457 | while (len > 0) { |
| 458 | if (m == 0) |
| 459 | panic("sbdrop"); |
| 460 | if (m->m_len > len) { |
| 461 | m->m_len -= len; |
| 462 | m->m_off += len; |
| 463 | sb->sb_cc -= len; |
| 464 | break; |
| 465 | } |
| 466 | len -= m->m_len; |
| 467 | sbfree(sb, m); |
| 468 | MFREE(m, mn); |
| 469 | m = mn; |
| 470 | } |
| 471 | sb->sb_mb = m; |
| 472 | } |