1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_kern_tls.h" 38 #include "opt_param.h" 39 40 #include <sys/param.h> 41 #include <sys/aio.h> /* for aio_swake proto */ 42 #include <sys/kernel.h> 43 #include <sys/ktls.h> 44 #include <sys/lock.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #include <sys/mutex.h> 48 #include <sys/proc.h> 49 #include <sys/protosw.h> 50 #include <sys/resourcevar.h> 51 #include <sys/signalvar.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/sx.h> 55 #include <sys/sysctl.h> 56 57 #include <netinet/in.h> 58 59 /* 60 * Function pointer set by the AIO routines so that the socket buffer code 61 * can call back into the AIO module if it is loaded. 62 */ 63 void (*aio_swake)(struct socket *, struct sockbuf *); 64 65 /* 66 * Primitive routines for operating on socket buffers 67 */ 68 69 u_long sb_max = SB_MAX; 70 u_long sb_max_adj = 71 (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 72 73 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 74 75 #ifdef KERN_TLS 76 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, 77 struct mbuf *n); 78 #endif 79 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); 80 static void sbflush_internal(struct sockbuf *sb); 81 82 /* 83 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY. 84 */ 85 static void 86 sbm_clrprotoflags(struct mbuf *m, int flags) 87 { 88 int mask; 89 90 mask = ~M_PROTOFLAGS; 91 if (flags & PRUS_NOTREADY) 92 mask |= M_NOTREADY; 93 while (m) { 94 m->m_flags &= mask; 95 m = m->m_next; 96 } 97 } 98 99 /* 100 * Compress M_NOTREADY mbufs after they have been readied by sbready(). 101 * 102 * sbcompress() skips M_NOTREADY mbufs since the data is not available to 103 * be copied at the time of sbcompress(). This function combines small 104 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first 105 * mbuf sbready() marked ready, and 'end' is the first mbuf still not 106 * ready. 107 */ 108 static void 109 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end) 110 { 111 struct mbuf *m, *n; 112 int ext_size; 113 114 SOCKBUF_LOCK_ASSERT(sb); 115 116 if ((sb->sb_flags & SB_NOCOALESCE) != 0) 117 return; 118 119 for (m = m0; m != end; m = m->m_next) { 120 MPASS((m->m_flags & M_NOTREADY) == 0); 121 /* 122 * NB: In sbcompress(), 'n' is the last mbuf in the 123 * socket buffer and 'm' is the new mbuf being copied 124 * into the trailing space of 'n'. Here, the roles 125 * are reversed and 'n' is the next mbuf after 'm' 126 * that is being copied into the trailing space of 127 * 'm'. 128 */ 129 n = m->m_next; 130 #ifdef KERN_TLS 131 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */ 132 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 133 (m->m_flags & M_EXTPG) && 134 (n->m_flags & M_EXTPG) && 135 !mbuf_has_tls_session(m) && 136 !mbuf_has_tls_session(n)) { 137 int hdr_len, trail_len; 138 139 hdr_len = n->m_epg_hdrlen; 140 trail_len = m->m_epg_trllen; 141 if (trail_len != 0 && hdr_len != 0 && 142 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) { 143 /* copy n's header to m's trailer */ 144 memcpy(&m->m_epg_trail[trail_len], 145 n->m_epg_hdr, hdr_len); 146 m->m_epg_trllen += hdr_len; 147 m->m_len += hdr_len; 148 n->m_epg_hdrlen = 0; 149 n->m_len -= hdr_len; 150 } 151 } 152 #endif 153 154 /* Compress small unmapped mbufs into plain mbufs. */ 155 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN && 156 !mbuf_has_tls_session(m)) { 157 ext_size = m->m_ext.ext_size; 158 if (mb_unmapped_compress(m) == 0) 159 sb->sb_mbcnt -= ext_size; 160 } 161 162 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 163 M_WRITABLE(m) && 164 (m->m_flags & M_EXTPG) == 0 && 165 !mbuf_has_tls_session(n) && 166 !mbuf_has_tls_session(m) && 167 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 168 n->m_len <= M_TRAILINGSPACE(m) && 169 m->m_type == n->m_type) { 170 KASSERT(sb->sb_lastrecord != n, 171 ("%s: merging start of record (%p) into previous mbuf (%p)", 172 __func__, n, m)); 173 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len)); 174 m->m_len += n->m_len; 175 m->m_next = n->m_next; 176 m->m_flags |= n->m_flags & M_EOR; 177 if (sb->sb_mbtail == n) 178 sb->sb_mbtail = m; 179 180 sb->sb_mbcnt -= MSIZE; 181 if (n->m_flags & M_EXT) 182 sb->sb_mbcnt -= n->m_ext.ext_size; 183 m_free(n); 184 n = m->m_next; 185 } 186 } 187 SBLASTRECORDCHK(sb); 188 SBLASTMBUFCHK(sb); 189 } 190 191 /* 192 * Mark ready "count" units of I/O starting with "m". Most mbufs 193 * count as a single unit of I/O except for M_EXTPG mbufs which 194 * are backed by multiple pages. 195 */ 196 int 197 sbready(struct sockbuf *sb, struct mbuf *m0, int count) 198 { 199 struct mbuf *m; 200 u_int blocker; 201 202 SOCKBUF_LOCK_ASSERT(sb); 203 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); 204 KASSERT(count > 0, ("%s: invalid count %d", __func__, count)); 205 206 m = m0; 207 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; 208 209 while (count > 0) { 210 KASSERT(m->m_flags & M_NOTREADY, 211 ("%s: m %p !M_NOTREADY", __func__, m)); 212 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) { 213 if (count < m->m_epg_nrdy) { 214 m->m_epg_nrdy -= count; 215 count = 0; 216 break; 217 } 218 count -= m->m_epg_nrdy; 219 m->m_epg_nrdy = 0; 220 } else 221 count--; 222 223 m->m_flags &= ~(M_NOTREADY | blocker); 224 if (blocker) 225 sb->sb_acc += m->m_len; 226 m = m->m_next; 227 } 228 229 /* 230 * If the first mbuf is still not fully ready because only 231 * some of its backing pages were readied, no further progress 232 * can be made. 233 */ 234 if (m0 == m) { 235 MPASS(m->m_flags & M_NOTREADY); 236 return (EINPROGRESS); 237 } 238 239 if (!blocker) { 240 sbready_compress(sb, m0, m); 241 return (EINPROGRESS); 242 } 243 244 /* This one was blocking all the queue. */ 245 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { 246 KASSERT(m->m_flags & M_BLOCKED, 247 ("%s: m %p !M_BLOCKED", __func__, m)); 248 m->m_flags &= ~M_BLOCKED; 249 sb->sb_acc += m->m_len; 250 } 251 252 sb->sb_fnrdy = m; 253 sbready_compress(sb, m0, m); 254 255 return (0); 256 } 257 258 /* 259 * Adjust sockbuf state reflecting allocation of m. 260 */ 261 void 262 sballoc(struct sockbuf *sb, struct mbuf *m) 263 { 264 265 SOCKBUF_LOCK_ASSERT(sb); 266 267 sb->sb_ccc += m->m_len; 268 269 if (sb->sb_fnrdy == NULL) { 270 if (m->m_flags & M_NOTREADY) 271 sb->sb_fnrdy = m; 272 else 273 sb->sb_acc += m->m_len; 274 } else 275 m->m_flags |= M_BLOCKED; 276 277 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 278 sb->sb_ctl += m->m_len; 279 280 sb->sb_mbcnt += MSIZE; 281 282 if (m->m_flags & M_EXT) 283 sb->sb_mbcnt += m->m_ext.ext_size; 284 } 285 286 /* 287 * Adjust sockbuf state reflecting freeing of m. 288 */ 289 void 290 sbfree(struct sockbuf *sb, struct mbuf *m) 291 { 292 293 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 294 SOCKBUF_LOCK_ASSERT(sb); 295 #endif 296 297 sb->sb_ccc -= m->m_len; 298 299 if (!(m->m_flags & M_NOTAVAIL)) 300 sb->sb_acc -= m->m_len; 301 302 if (m == sb->sb_fnrdy) { 303 struct mbuf *n; 304 305 KASSERT(m->m_flags & M_NOTREADY, 306 ("%s: m %p !M_NOTREADY", __func__, m)); 307 308 n = m->m_next; 309 while (n != NULL && !(n->m_flags & M_NOTREADY)) { 310 n->m_flags &= ~M_BLOCKED; 311 sb->sb_acc += n->m_len; 312 n = n->m_next; 313 } 314 sb->sb_fnrdy = n; 315 } 316 317 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 318 sb->sb_ctl -= m->m_len; 319 320 sb->sb_mbcnt -= MSIZE; 321 if (m->m_flags & M_EXT) 322 sb->sb_mbcnt -= m->m_ext.ext_size; 323 324 if (sb->sb_sndptr == m) { 325 sb->sb_sndptr = NULL; 326 sb->sb_sndptroff = 0; 327 } 328 if (sb->sb_sndptroff != 0) 329 sb->sb_sndptroff -= m->m_len; 330 } 331 332 #ifdef KERN_TLS 333 /* 334 * Similar to sballoc/sbfree but does not adjust state associated with 335 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs 336 * are not ready. 337 */ 338 void 339 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m) 340 { 341 342 SOCKBUF_LOCK_ASSERT(sb); 343 344 sb->sb_ccc += m->m_len; 345 sb->sb_tlscc += m->m_len; 346 347 sb->sb_mbcnt += MSIZE; 348 349 if (m->m_flags & M_EXT) 350 sb->sb_mbcnt += m->m_ext.ext_size; 351 } 352 353 void 354 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m) 355 { 356 357 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 358 SOCKBUF_LOCK_ASSERT(sb); 359 #endif 360 361 sb->sb_ccc -= m->m_len; 362 sb->sb_tlscc -= m->m_len; 363 364 sb->sb_mbcnt -= MSIZE; 365 366 if (m->m_flags & M_EXT) 367 sb->sb_mbcnt -= m->m_ext.ext_size; 368 } 369 #endif 370 371 /* 372 * Socantsendmore indicates that no more data will be sent on the socket; it 373 * would normally be applied to a socket when the user informs the system 374 * that no more data is to be sent, by the protocol code (in case 375 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be 376 * received, and will normally be applied to the socket by a protocol when it 377 * detects that the peer will send no more data. Data queued for reading in 378 * the socket may yet be read. 379 */ 380 void 381 socantsendmore_locked(struct socket *so) 382 { 383 384 SOCK_SENDBUF_LOCK_ASSERT(so); 385 386 so->so_snd.sb_state |= SBS_CANTSENDMORE; 387 sowwakeup_locked(so); 388 SOCK_SENDBUF_UNLOCK_ASSERT(so); 389 } 390 391 void 392 socantsendmore(struct socket *so) 393 { 394 395 SOCK_SENDBUF_LOCK(so); 396 socantsendmore_locked(so); 397 SOCK_SENDBUF_UNLOCK_ASSERT(so); 398 } 399 400 void 401 socantrcvmore_locked(struct socket *so) 402 { 403 404 SOCK_RECVBUF_LOCK_ASSERT(so); 405 406 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 407 #ifdef KERN_TLS 408 if (so->so_rcv.sb_flags & SB_TLS_RX) 409 ktls_check_rx(&so->so_rcv); 410 #endif 411 sorwakeup_locked(so); 412 SOCK_RECVBUF_UNLOCK_ASSERT(so); 413 } 414 415 void 416 socantrcvmore(struct socket *so) 417 { 418 419 SOCK_RECVBUF_LOCK(so); 420 socantrcvmore_locked(so); 421 SOCK_RECVBUF_UNLOCK_ASSERT(so); 422 } 423 424 void 425 soroverflow_locked(struct socket *so) 426 { 427 428 SOCK_RECVBUF_LOCK_ASSERT(so); 429 430 if (so->so_options & SO_RERROR) { 431 so->so_rerror = ENOBUFS; 432 sorwakeup_locked(so); 433 } else 434 SOCK_RECVBUF_UNLOCK(so); 435 436 SOCK_RECVBUF_UNLOCK_ASSERT(so); 437 } 438 439 void 440 soroverflow(struct socket *so) 441 { 442 443 SOCK_RECVBUF_LOCK(so); 444 soroverflow_locked(so); 445 SOCK_RECVBUF_UNLOCK_ASSERT(so); 446 } 447 448 /* 449 * Wait for data to arrive at/drain from a socket buffer. 450 */ 451 int 452 sbwait(struct socket *so, sb_which which) 453 { 454 struct sockbuf *sb; 455 456 SOCK_BUF_LOCK_ASSERT(so, which); 457 458 sb = sobuf(so, which); 459 sb->sb_flags |= SB_WAIT; 460 return (msleep_sbt(&sb->sb_acc, soeventmtx(so, which), 461 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", 462 sb->sb_timeo, 0, 0)); 463 } 464 465 /* 466 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 467 * via SIGIO if the socket has the SS_ASYNC flag set. 468 * 469 * Called with the socket buffer lock held; will release the lock by the end 470 * of the function. This allows the caller to acquire the socket buffer lock 471 * while testing for the need for various sorts of wakeup and hold it through 472 * to the point where it's no longer required. We currently hold the lock 473 * through calls out to other subsystems (with the exception of kqueue), and 474 * then release it to avoid lock order issues. It's not clear that's 475 * correct. 476 */ 477 static __always_inline void 478 sowakeup(struct socket *so, const sb_which which) 479 { 480 struct sockbuf *sb; 481 int ret; 482 483 SOCK_BUF_LOCK_ASSERT(so, which); 484 485 sb = sobuf(so, which); 486 selwakeuppri(sb->sb_sel, PSOCK); 487 if (!SEL_WAITING(sb->sb_sel)) 488 sb->sb_flags &= ~SB_SEL; 489 if (sb->sb_flags & SB_WAIT) { 490 sb->sb_flags &= ~SB_WAIT; 491 wakeup(&sb->sb_acc); 492 } 493 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 494 if (sb->sb_upcall != NULL) { 495 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); 496 if (ret == SU_ISCONNECTED) { 497 KASSERT(sb == &so->so_rcv, 498 ("SO_SND upcall returned SU_ISCONNECTED")); 499 soupcall_clear(so, SO_RCV); 500 } 501 } else 502 ret = SU_OK; 503 if (sb->sb_flags & SB_AIO) 504 sowakeup_aio(so, which); 505 SOCK_BUF_UNLOCK(so, which); 506 if (ret == SU_ISCONNECTED) 507 soisconnected(so); 508 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 509 pgsigio(&so->so_sigio, SIGIO, 0); 510 SOCK_BUF_UNLOCK_ASSERT(so, which); 511 } 512 513 /* 514 * Do we need to notify the other side when I/O is possible? 515 */ 516 static __always_inline bool 517 sb_notify(const struct sockbuf *sb) 518 { 519 return ((sb->sb_flags & (SB_WAIT | SB_SEL | SB_ASYNC | 520 SB_UPCALL | SB_AIO | SB_KNOTE)) != 0); 521 } 522 523 void 524 sorwakeup_locked(struct socket *so) 525 { 526 SOCK_RECVBUF_LOCK_ASSERT(so); 527 if (sb_notify(&so->so_rcv)) 528 sowakeup(so, SO_RCV); 529 else 530 SOCK_RECVBUF_UNLOCK(so); 531 } 532 533 void 534 sowwakeup_locked(struct socket *so) 535 { 536 SOCK_SENDBUF_LOCK_ASSERT(so); 537 if (sb_notify(&so->so_snd)) 538 sowakeup(so, SO_SND); 539 else 540 SOCK_SENDBUF_UNLOCK(so); 541 } 542 543 /* 544 * Socket buffer (struct sockbuf) utility routines. 545 * 546 * Each socket contains two socket buffers: one for sending data and one for 547 * receiving data. Each buffer contains a queue of mbufs, information about 548 * the number of mbufs and amount of data in the queue, and other fields 549 * allowing select() statements and notification on data availability to be 550 * implemented. 551 * 552 * Data stored in a socket buffer is maintained as a list of records. Each 553 * record is a list of mbufs chained together with the m_next field. Records 554 * are chained together with the m_nextpkt field. The upper level routine 555 * soreceive() expects the following conventions to be observed when placing 556 * information in the receive buffer: 557 * 558 * 1. If the protocol requires each message be preceded by the sender's name, 559 * then a record containing that name must be present before any 560 * associated data (mbuf's must be of type MT_SONAME). 561 * 2. If the protocol supports the exchange of ``access rights'' (really just 562 * additional data associated with the message), and there are ``rights'' 563 * to be received, then a record containing this data should be present 564 * (mbuf's must be of type MT_RIGHTS). 565 * 3. If a name or rights record exists, then it must be followed by a data 566 * record, perhaps of zero length. 567 * 568 * Before using a new socket structure it is first necessary to reserve 569 * buffer space to the socket, by calling sbreserve(). This should commit 570 * some of the available buffer space in the system buffer pool for the 571 * socket (currently, it does nothing but enforce limits). The space should 572 * be released by calling sbrelease() when the socket is destroyed. 573 */ 574 int 575 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 576 { 577 struct thread *td = curthread; 578 579 SOCK_SENDBUF_LOCK(so); 580 SOCK_RECVBUF_LOCK(so); 581 if (sbreserve_locked(so, SO_SND, sndcc, td) == 0) 582 goto bad; 583 if (sbreserve_locked(so, SO_RCV, rcvcc, td) == 0) 584 goto bad2; 585 if (so->so_rcv.sb_lowat == 0) 586 so->so_rcv.sb_lowat = 1; 587 if (so->so_snd.sb_lowat == 0) 588 so->so_snd.sb_lowat = MCLBYTES; 589 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 590 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 591 SOCK_RECVBUF_UNLOCK(so); 592 SOCK_SENDBUF_UNLOCK(so); 593 return (0); 594 bad2: 595 sbrelease_locked(so, SO_SND); 596 bad: 597 SOCK_RECVBUF_UNLOCK(so); 598 SOCK_SENDBUF_UNLOCK(so); 599 return (ENOBUFS); 600 } 601 602 static int 603 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 604 { 605 int error = 0; 606 u_long tmp_sb_max = sb_max; 607 608 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 609 if (error || !req->newptr) 610 return (error); 611 if (tmp_sb_max < MSIZE + MCLBYTES) 612 return (EINVAL); 613 sb_max = tmp_sb_max; 614 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 615 return (0); 616 } 617 618 /* 619 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 620 * become limiting if buffering efficiency is near the normal case. 621 */ 622 bool 623 sbreserve_locked(struct socket *so, sb_which which, u_long cc, 624 struct thread *td) 625 { 626 struct sockbuf *sb = sobuf(so, which); 627 rlim_t sbsize_limit; 628 629 SOCK_BUF_LOCK_ASSERT(so, which); 630 631 /* 632 * When a thread is passed, we take into account the thread's socket 633 * buffer size limit. The caller will generally pass curthread, but 634 * in the TCP input path, NULL will be passed to indicate that no 635 * appropriate thread resource limits are available. In that case, 636 * we don't apply a process limit. 637 */ 638 if (cc > sb_max_adj) 639 return (false); 640 if (td != NULL) { 641 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); 642 } else 643 sbsize_limit = RLIM_INFINITY; 644 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 645 sbsize_limit)) 646 return (false); 647 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 648 if (sb->sb_lowat > sb->sb_hiwat) 649 sb->sb_lowat = sb->sb_hiwat; 650 return (true); 651 } 652 653 int 654 sbsetopt(struct socket *so, struct sockopt *sopt) 655 { 656 struct sockbuf *sb; 657 sb_which wh; 658 short *flags; 659 u_int cc, *hiwat, *lowat; 660 int error, optval; 661 662 error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); 663 if (error != 0) 664 return (error); 665 666 /* 667 * Values < 1 make no sense for any of these options, 668 * so disallow them. 669 */ 670 if (optval < 1) 671 return (EINVAL); 672 cc = optval; 673 674 sb = NULL; 675 SOCK_LOCK(so); 676 if (SOLISTENING(so)) { 677 switch (sopt->sopt_name) { 678 case SO_SNDLOWAT: 679 case SO_SNDBUF: 680 lowat = &so->sol_sbsnd_lowat; 681 hiwat = &so->sol_sbsnd_hiwat; 682 flags = &so->sol_sbsnd_flags; 683 break; 684 case SO_RCVLOWAT: 685 case SO_RCVBUF: 686 lowat = &so->sol_sbrcv_lowat; 687 hiwat = &so->sol_sbrcv_hiwat; 688 flags = &so->sol_sbrcv_flags; 689 break; 690 } 691 } else { 692 switch (sopt->sopt_name) { 693 case SO_SNDLOWAT: 694 case SO_SNDBUF: 695 sb = &so->so_snd; 696 wh = SO_SND; 697 break; 698 case SO_RCVLOWAT: 699 case SO_RCVBUF: 700 sb = &so->so_rcv; 701 wh = SO_RCV; 702 break; 703 } 704 flags = &sb->sb_flags; 705 hiwat = &sb->sb_hiwat; 706 lowat = &sb->sb_lowat; 707 SOCK_BUF_LOCK(so, wh); 708 } 709 710 error = 0; 711 switch (sopt->sopt_name) { 712 case SO_SNDBUF: 713 case SO_RCVBUF: 714 if (SOLISTENING(so)) { 715 if (cc > sb_max_adj) { 716 error = ENOBUFS; 717 break; 718 } 719 *hiwat = cc; 720 if (*lowat > *hiwat) 721 *lowat = *hiwat; 722 } else { 723 if (!sbreserve_locked(so, wh, cc, curthread)) 724 error = ENOBUFS; 725 } 726 if (error == 0) 727 *flags &= ~SB_AUTOSIZE; 728 break; 729 case SO_SNDLOWAT: 730 case SO_RCVLOWAT: 731 /* 732 * Make sure the low-water is never greater than the 733 * high-water. 734 */ 735 *lowat = (cc > *hiwat) ? *hiwat : cc; 736 break; 737 } 738 739 if (!SOLISTENING(so)) 740 SOCK_BUF_UNLOCK(so, wh); 741 SOCK_UNLOCK(so); 742 return (error); 743 } 744 745 /* 746 * Free mbufs held by a socket, and reserved mbuf space. 747 */ 748 static void 749 sbrelease_internal(struct socket *so, sb_which which) 750 { 751 struct sockbuf *sb = sobuf(so, which); 752 753 sbflush_internal(sb); 754 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 755 RLIM_INFINITY); 756 sb->sb_mbmax = 0; 757 } 758 759 void 760 sbrelease_locked(struct socket *so, sb_which which) 761 { 762 763 SOCK_BUF_LOCK_ASSERT(so, which); 764 765 sbrelease_internal(so, which); 766 } 767 768 void 769 sbrelease(struct socket *so, sb_which which) 770 { 771 772 SOCK_BUF_LOCK(so, which); 773 sbrelease_locked(so, which); 774 SOCK_BUF_UNLOCK(so, which); 775 } 776 777 void 778 sbdestroy(struct socket *so, sb_which which) 779 { 780 #ifdef KERN_TLS 781 struct sockbuf *sb = sobuf(so, which); 782 783 if (sb->sb_tls_info != NULL) 784 ktls_free(sb->sb_tls_info); 785 sb->sb_tls_info = NULL; 786 #endif 787 sbrelease_internal(so, which); 788 } 789 790 /* 791 * Routines to add and remove data from an mbuf queue. 792 * 793 * The routines sbappend() or sbappendrecord() are normally called to append 794 * new mbufs to a socket buffer, after checking that adequate space is 795 * available, comparing the function sbspace() with the amount of data to be 796 * added. sbappendrecord() differs from sbappend() in that data supplied is 797 * treated as the beginning of a new record. To place a sender's address, 798 * optional access rights, and data in a socket receive buffer, 799 * sbappendaddr() should be used. To place access rights and data in a 800 * socket receive buffer, sbappendrights() should be used. In either case, 801 * the new data begins a new record. Note that unlike sbappend() and 802 * sbappendrecord(), these routines check for the caller that there will be 803 * enough space to store the data. Each fails if there is not enough space, 804 * or if it cannot find mbufs to store additional information in. 805 * 806 * Reliable protocols may use the socket send buffer to hold data awaiting 807 * acknowledgement. Data is normally copied from a socket send buffer in a 808 * protocol with m_copy for output to a peer, and then removing the data from 809 * the socket buffer with sbdrop() or sbdroprecord() when the data is 810 * acknowledged by the peer. 811 */ 812 #ifdef SOCKBUF_DEBUG 813 void 814 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 815 { 816 struct mbuf *m = sb->sb_mb; 817 818 SOCKBUF_LOCK_ASSERT(sb); 819 820 while (m && m->m_nextpkt) 821 m = m->m_nextpkt; 822 823 if (m != sb->sb_lastrecord) { 824 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 825 __func__, sb->sb_mb, sb->sb_lastrecord, m); 826 printf("packet chain:\n"); 827 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 828 printf("\t%p\n", m); 829 panic("%s from %s:%u", __func__, file, line); 830 } 831 } 832 833 void 834 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 835 { 836 struct mbuf *m = sb->sb_mb; 837 struct mbuf *n; 838 839 SOCKBUF_LOCK_ASSERT(sb); 840 841 while (m && m->m_nextpkt) 842 m = m->m_nextpkt; 843 844 while (m && m->m_next) 845 m = m->m_next; 846 847 if (m != sb->sb_mbtail) { 848 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 849 __func__, sb->sb_mb, sb->sb_mbtail, m); 850 printf("packet tree:\n"); 851 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 852 printf("\t"); 853 for (n = m; n != NULL; n = n->m_next) 854 printf("%p ", n); 855 printf("\n"); 856 } 857 panic("%s from %s:%u", __func__, file, line); 858 } 859 860 #ifdef KERN_TLS 861 m = sb->sb_mtls; 862 while (m && m->m_next) 863 m = m->m_next; 864 865 if (m != sb->sb_mtlstail) { 866 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n", 867 __func__, sb->sb_mtls, sb->sb_mtlstail, m); 868 printf("TLS packet tree:\n"); 869 printf("\t"); 870 for (m = sb->sb_mtls; m != NULL; m = m->m_next) { 871 printf("%p ", m); 872 } 873 printf("\n"); 874 panic("%s from %s:%u", __func__, file, line); 875 } 876 #endif 877 } 878 #endif /* SOCKBUF_DEBUG */ 879 880 #define SBLINKRECORD(sb, m0) do { \ 881 SOCKBUF_LOCK_ASSERT(sb); \ 882 if ((sb)->sb_lastrecord != NULL) \ 883 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 884 else \ 885 (sb)->sb_mb = (m0); \ 886 (sb)->sb_lastrecord = (m0); \ 887 } while (/*CONSTCOND*/0) 888 889 /* 890 * Append mbuf chain m to the last record in the socket buffer sb. The 891 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 892 * are discarded and mbufs are compacted where possible. 893 */ 894 void 895 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) 896 { 897 struct mbuf *n; 898 899 SOCKBUF_LOCK_ASSERT(sb); 900 901 if (m == NULL) 902 return; 903 sbm_clrprotoflags(m, flags); 904 SBLASTRECORDCHK(sb); 905 n = sb->sb_mb; 906 if (n) { 907 while (n->m_nextpkt) 908 n = n->m_nextpkt; 909 do { 910 if (n->m_flags & M_EOR) { 911 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 912 return; 913 } 914 } while (n->m_next && (n = n->m_next)); 915 } else { 916 /* 917 * XXX Would like to simply use sb_mbtail here, but 918 * XXX I need to verify that I won't miss an EOR that 919 * XXX way. 920 */ 921 if ((n = sb->sb_lastrecord) != NULL) { 922 do { 923 if (n->m_flags & M_EOR) { 924 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 925 return; 926 } 927 } while (n->m_next && (n = n->m_next)); 928 } else { 929 /* 930 * If this is the first record in the socket buffer, 931 * it's also the last record. 932 */ 933 sb->sb_lastrecord = m; 934 } 935 } 936 sbcompress(sb, m, n); 937 SBLASTRECORDCHK(sb); 938 } 939 940 /* 941 * Append mbuf chain m to the last record in the socket buffer sb. The 942 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 943 * are discarded and mbufs are compacted where possible. 944 */ 945 void 946 sbappend(struct sockbuf *sb, struct mbuf *m, int flags) 947 { 948 949 SOCKBUF_LOCK(sb); 950 sbappend_locked(sb, m, flags); 951 SOCKBUF_UNLOCK(sb); 952 } 953 954 #ifdef KERN_TLS 955 /* 956 * Append an mbuf containing encrypted TLS data. The data 957 * is marked M_NOTREADY until it has been decrypted and 958 * stored as a TLS record. 959 */ 960 static void 961 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m) 962 { 963 struct ifnet *ifp; 964 struct mbuf *n; 965 int flags; 966 967 ifp = NULL; 968 flags = M_NOTREADY; 969 970 SBLASTMBUFCHK(sb); 971 972 /* Mbuf chain must start with a packet header. */ 973 MPASS((m->m_flags & M_PKTHDR) != 0); 974 975 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 976 for (n = m; n != NULL; n = n->m_next) { 977 if (n->m_flags & M_PKTHDR) { 978 ifp = m->m_pkthdr.leaf_rcvif; 979 if ((n->m_pkthdr.csum_flags & CSUM_TLS_MASK) == 980 CSUM_TLS_DECRYPTED) { 981 /* Mark all mbufs in this packet decrypted. */ 982 flags = M_NOTREADY | M_DECRYPTED; 983 } else { 984 flags = M_NOTREADY; 985 } 986 m_demote_pkthdr(n); 987 } 988 989 n->m_flags &= M_DEMOTEFLAGS; 990 n->m_flags |= flags; 991 992 MPASS((n->m_flags & M_NOTREADY) != 0); 993 } 994 995 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail); 996 ktls_check_rx(sb); 997 998 /* Check for incoming packet route changes: */ 999 if (ifp != NULL && sb->sb_tls_info->rx_ifp != NULL && 1000 sb->sb_tls_info->rx_ifp != ifp) 1001 ktls_input_ifp_mismatch(sb, ifp); 1002 } 1003 #endif 1004 1005 /* 1006 * This version of sbappend() should only be used when the caller absolutely 1007 * knows that there will never be more than one record in the socket buffer, 1008 * that is, a stream protocol (such as TCP). 1009 */ 1010 void 1011 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) 1012 { 1013 SOCKBUF_LOCK_ASSERT(sb); 1014 1015 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 1016 1017 #ifdef KERN_TLS 1018 /* 1019 * Decrypted TLS records are appended as records via 1020 * sbappendrecord(). TCP passes encrypted TLS records to this 1021 * function which must be scheduled for decryption. 1022 */ 1023 if (sb->sb_flags & SB_TLS_RX) { 1024 sbappend_ktls_rx(sb, m); 1025 return; 1026 } 1027 #endif 1028 1029 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 1030 1031 SBLASTMBUFCHK(sb); 1032 1033 #ifdef KERN_TLS 1034 if (sb->sb_tls_info != NULL) 1035 ktls_seq(sb, m); 1036 #endif 1037 1038 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 1039 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); 1040 1041 sbcompress(sb, m, sb->sb_mbtail); 1042 1043 sb->sb_lastrecord = sb->sb_mb; 1044 SBLASTRECORDCHK(sb); 1045 } 1046 1047 /* 1048 * This version of sbappend() should only be used when the caller absolutely 1049 * knows that there will never be more than one record in the socket buffer, 1050 * that is, a stream protocol (such as TCP). 1051 */ 1052 void 1053 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) 1054 { 1055 1056 SOCKBUF_LOCK(sb); 1057 sbappendstream_locked(sb, m, flags); 1058 SOCKBUF_UNLOCK(sb); 1059 } 1060 1061 #ifdef SOCKBUF_DEBUG 1062 void 1063 sbcheck(struct sockbuf *sb, const char *file, int line) 1064 { 1065 struct mbuf *m, *n, *fnrdy; 1066 u_long acc, ccc, mbcnt; 1067 #ifdef KERN_TLS 1068 u_long tlscc; 1069 #endif 1070 1071 SOCKBUF_LOCK_ASSERT(sb); 1072 1073 acc = ccc = mbcnt = 0; 1074 fnrdy = NULL; 1075 1076 for (m = sb->sb_mb; m; m = n) { 1077 n = m->m_nextpkt; 1078 for (; m; m = m->m_next) { 1079 if (m->m_len == 0) { 1080 printf("sb %p empty mbuf %p\n", sb, m); 1081 goto fail; 1082 } 1083 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { 1084 if (m != sb->sb_fnrdy) { 1085 printf("sb %p: fnrdy %p != m %p\n", 1086 sb, sb->sb_fnrdy, m); 1087 goto fail; 1088 } 1089 fnrdy = m; 1090 } 1091 if (fnrdy) { 1092 if (!(m->m_flags & M_NOTAVAIL)) { 1093 printf("sb %p: fnrdy %p, m %p is avail\n", 1094 sb, sb->sb_fnrdy, m); 1095 goto fail; 1096 } 1097 } else 1098 acc += m->m_len; 1099 ccc += m->m_len; 1100 mbcnt += MSIZE; 1101 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1102 mbcnt += m->m_ext.ext_size; 1103 } 1104 } 1105 #ifdef KERN_TLS 1106 /* 1107 * Account for mbufs "detached" by ktls_detach_record() while 1108 * they are decrypted by ktls_decrypt(). tlsdcc gives a count 1109 * of the detached bytes that are included in ccc. The mbufs 1110 * and clusters are not included in the socket buffer 1111 * accounting. 1112 */ 1113 ccc += sb->sb_tlsdcc; 1114 1115 tlscc = 0; 1116 for (m = sb->sb_mtls; m; m = m->m_next) { 1117 if (m->m_nextpkt != NULL) { 1118 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m); 1119 goto fail; 1120 } 1121 if ((m->m_flags & M_NOTREADY) == 0) { 1122 printf("sb %p TLS mbuf %p ready\n", sb, m); 1123 goto fail; 1124 } 1125 tlscc += m->m_len; 1126 ccc += m->m_len; 1127 mbcnt += MSIZE; 1128 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1129 mbcnt += m->m_ext.ext_size; 1130 } 1131 1132 if (sb->sb_tlscc != tlscc) { 1133 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1134 sb->sb_tlsdcc); 1135 goto fail; 1136 } 1137 #endif 1138 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { 1139 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", 1140 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); 1141 #ifdef KERN_TLS 1142 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1143 sb->sb_tlsdcc); 1144 #endif 1145 goto fail; 1146 } 1147 return; 1148 fail: 1149 panic("%s from %s:%u", __func__, file, line); 1150 } 1151 #endif 1152 1153 /* 1154 * As above, except the mbuf chain begins a new record. 1155 */ 1156 void 1157 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 1158 { 1159 struct mbuf *m; 1160 1161 SOCKBUF_LOCK_ASSERT(sb); 1162 1163 if (m0 == NULL) 1164 return; 1165 m_clrprotoflags(m0); 1166 /* 1167 * Put the first mbuf on the queue. Note this permits zero length 1168 * records. 1169 */ 1170 sballoc(sb, m0); 1171 SBLASTRECORDCHK(sb); 1172 SBLINKRECORD(sb, m0); 1173 sb->sb_mbtail = m0; 1174 m = m0->m_next; 1175 m0->m_next = 0; 1176 if (m && (m0->m_flags & M_EOR)) { 1177 m0->m_flags &= ~M_EOR; 1178 m->m_flags |= M_EOR; 1179 } 1180 /* always call sbcompress() so it can do SBLASTMBUFCHK() */ 1181 sbcompress(sb, m, m0); 1182 } 1183 1184 /* 1185 * As above, except the mbuf chain begins a new record. 1186 */ 1187 void 1188 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 1189 { 1190 1191 SOCKBUF_LOCK(sb); 1192 sbappendrecord_locked(sb, m0); 1193 SOCKBUF_UNLOCK(sb); 1194 } 1195 1196 /* Helper routine that appends data, control, and address to a sockbuf. */ 1197 static int 1198 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, 1199 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) 1200 { 1201 struct mbuf *m, *n, *nlast; 1202 #if MSIZE <= 256 1203 if (asa->sa_len > MLEN) 1204 return (0); 1205 #endif 1206 m = m_get(M_NOWAIT, MT_SONAME); 1207 if (m == NULL) 1208 return (0); 1209 m->m_len = asa->sa_len; 1210 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 1211 if (m0) { 1212 M_ASSERT_NO_SND_TAG(m0); 1213 m_clrprotoflags(m0); 1214 m_tag_delete_chain(m0, NULL); 1215 /* 1216 * Clear some persistent info from pkthdr. 1217 * We don't use m_demote(), because some netgraph consumers 1218 * expect M_PKTHDR presence. 1219 */ 1220 m0->m_pkthdr.rcvif = NULL; 1221 m0->m_pkthdr.flowid = 0; 1222 m0->m_pkthdr.csum_flags = 0; 1223 m0->m_pkthdr.fibnum = 0; 1224 m0->m_pkthdr.rsstype = 0; 1225 } 1226 if (ctrl_last) 1227 ctrl_last->m_next = m0; /* concatenate data to control */ 1228 else 1229 control = m0; 1230 m->m_next = control; 1231 for (n = m; n->m_next != NULL; n = n->m_next) 1232 sballoc(sb, n); 1233 sballoc(sb, n); 1234 nlast = n; 1235 SBLINKRECORD(sb, m); 1236 1237 sb->sb_mbtail = nlast; 1238 SBLASTMBUFCHK(sb); 1239 1240 SBLASTRECORDCHK(sb); 1241 return (1); 1242 } 1243 1244 /* 1245 * Append address and data, and optionally, control (ancillary) data to the 1246 * receive queue of a socket. If present, m0 must include a packet header 1247 * with total length. Returns 0 if no space in sockbuf or insufficient 1248 * mbufs. 1249 */ 1250 int 1251 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 1252 struct mbuf *m0, struct mbuf *control) 1253 { 1254 struct mbuf *ctrl_last; 1255 int space = asa->sa_len; 1256 1257 SOCKBUF_LOCK_ASSERT(sb); 1258 1259 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1260 panic("sbappendaddr_locked"); 1261 if (m0) 1262 space += m0->m_pkthdr.len; 1263 space += m_length(control, &ctrl_last); 1264 1265 if (space > sbspace(sb)) 1266 return (0); 1267 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1268 } 1269 1270 /* 1271 * Append address and data, and optionally, control (ancillary) data to the 1272 * receive queue of a socket. If present, m0 must include a packet header 1273 * with total length. Returns 0 if insufficient mbufs. Does not validate space 1274 * on the receiving sockbuf. 1275 */ 1276 int 1277 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, 1278 struct mbuf *m0, struct mbuf *control) 1279 { 1280 struct mbuf *ctrl_last; 1281 1282 SOCKBUF_LOCK_ASSERT(sb); 1283 1284 ctrl_last = (control == NULL) ? NULL : m_last(control); 1285 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1286 } 1287 1288 /* 1289 * Append address and data, and optionally, control (ancillary) data to the 1290 * receive queue of a socket. If present, m0 must include a packet header 1291 * with total length. Returns 0 if no space in sockbuf or insufficient 1292 * mbufs. 1293 */ 1294 int 1295 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 1296 struct mbuf *m0, struct mbuf *control) 1297 { 1298 int retval; 1299 1300 SOCKBUF_LOCK(sb); 1301 retval = sbappendaddr_locked(sb, asa, m0, control); 1302 SOCKBUF_UNLOCK(sb); 1303 return (retval); 1304 } 1305 1306 void 1307 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 1308 struct mbuf *control, int flags) 1309 { 1310 struct mbuf *m, *mlast; 1311 1312 sbm_clrprotoflags(m0, flags); 1313 m_last(control)->m_next = m0; 1314 1315 SBLASTRECORDCHK(sb); 1316 1317 for (m = control; m->m_next; m = m->m_next) 1318 sballoc(sb, m); 1319 sballoc(sb, m); 1320 mlast = m; 1321 SBLINKRECORD(sb, control); 1322 1323 sb->sb_mbtail = mlast; 1324 SBLASTMBUFCHK(sb); 1325 1326 SBLASTRECORDCHK(sb); 1327 } 1328 1329 void 1330 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control, 1331 int flags) 1332 { 1333 1334 SOCKBUF_LOCK(sb); 1335 sbappendcontrol_locked(sb, m0, control, flags); 1336 SOCKBUF_UNLOCK(sb); 1337 } 1338 1339 /* 1340 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 1341 * (n). If (n) is NULL, the buffer is presumed empty. 1342 * 1343 * When the data is compressed, mbufs in the chain may be handled in one of 1344 * three ways: 1345 * 1346 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 1347 * record boundary, and no change in data type). 1348 * 1349 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 1350 * an mbuf already in the socket buffer. This can occur if an 1351 * appropriate mbuf exists, there is room, both mbufs are not marked as 1352 * not ready, and no merging of data types will occur. 1353 * 1354 * (3) The mbuf may be appended to the end of the existing mbuf chain. 1355 * 1356 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 1357 * end-of-record. 1358 */ 1359 void 1360 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1361 { 1362 int eor = 0; 1363 struct mbuf *o; 1364 1365 SOCKBUF_LOCK_ASSERT(sb); 1366 1367 while (m) { 1368 eor |= m->m_flags & M_EOR; 1369 if (m->m_len == 0 && 1370 (eor == 0 || 1371 (((o = m->m_next) || (o = n)) && 1372 o->m_type == m->m_type))) { 1373 if (sb->sb_lastrecord == m) 1374 sb->sb_lastrecord = m->m_next; 1375 m = m_free(m); 1376 continue; 1377 } 1378 if (n && (n->m_flags & M_EOR) == 0 && 1379 M_WRITABLE(n) && 1380 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1381 !(m->m_flags & M_NOTREADY) && 1382 !(n->m_flags & (M_NOTREADY | M_EXTPG)) && 1383 !mbuf_has_tls_session(m) && 1384 !mbuf_has_tls_session(n) && 1385 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1386 m->m_len <= M_TRAILINGSPACE(n) && 1387 n->m_type == m->m_type) { 1388 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1389 n->m_len += m->m_len; 1390 sb->sb_ccc += m->m_len; 1391 if (sb->sb_fnrdy == NULL) 1392 sb->sb_acc += m->m_len; 1393 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1394 /* XXX: Probably don't need.*/ 1395 sb->sb_ctl += m->m_len; 1396 m = m_free(m); 1397 continue; 1398 } 1399 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) && 1400 (m->m_flags & M_NOTREADY) == 0 && 1401 !mbuf_has_tls_session(m)) 1402 (void)mb_unmapped_compress(m); 1403 if (n) 1404 n->m_next = m; 1405 else 1406 sb->sb_mb = m; 1407 sb->sb_mbtail = m; 1408 sballoc(sb, m); 1409 n = m; 1410 m->m_flags &= ~M_EOR; 1411 m = m->m_next; 1412 n->m_next = 0; 1413 } 1414 if (eor) { 1415 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 1416 n->m_flags |= eor; 1417 } 1418 SBLASTMBUFCHK(sb); 1419 } 1420 1421 #ifdef KERN_TLS 1422 /* 1423 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs 1424 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also 1425 * a bit simpler (no EOR markers, always MT_DATA, etc.). 1426 */ 1427 static void 1428 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1429 { 1430 1431 SOCKBUF_LOCK_ASSERT(sb); 1432 1433 while (m) { 1434 KASSERT((m->m_flags & M_EOR) == 0, 1435 ("TLS RX mbuf %p with EOR", m)); 1436 KASSERT(m->m_type == MT_DATA, 1437 ("TLS RX mbuf %p is not MT_DATA", m)); 1438 KASSERT((m->m_flags & M_NOTREADY) != 0, 1439 ("TLS RX mbuf %p ready", m)); 1440 KASSERT((m->m_flags & M_EXTPG) == 0, 1441 ("TLS RX mbuf %p unmapped", m)); 1442 1443 if (m->m_len == 0) { 1444 m = m_free(m); 1445 continue; 1446 } 1447 1448 /* 1449 * Even though both 'n' and 'm' are NOTREADY, it's ok 1450 * to coalesce the data. 1451 */ 1452 if (n && 1453 M_WRITABLE(n) && 1454 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1455 !((m->m_flags ^ n->m_flags) & M_DECRYPTED) && 1456 !(n->m_flags & M_EXTPG) && 1457 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1458 m->m_len <= M_TRAILINGSPACE(n)) { 1459 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1460 n->m_len += m->m_len; 1461 sb->sb_ccc += m->m_len; 1462 sb->sb_tlscc += m->m_len; 1463 m = m_free(m); 1464 continue; 1465 } 1466 if (n) 1467 n->m_next = m; 1468 else 1469 sb->sb_mtls = m; 1470 sb->sb_mtlstail = m; 1471 sballoc_ktls_rx(sb, m); 1472 n = m; 1473 m = m->m_next; 1474 n->m_next = NULL; 1475 } 1476 SBLASTMBUFCHK(sb); 1477 } 1478 #endif 1479 1480 /* 1481 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 1482 */ 1483 static void 1484 sbflush_internal(struct sockbuf *sb) 1485 { 1486 1487 while (sb->sb_mbcnt || sb->sb_tlsdcc) { 1488 /* 1489 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: 1490 * we would loop forever. Panic instead. 1491 */ 1492 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1493 break; 1494 m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); 1495 } 1496 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 1497 ("%s: ccc %u mb %p mbcnt %u", __func__, 1498 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 1499 } 1500 1501 void 1502 sbflush_locked(struct sockbuf *sb) 1503 { 1504 1505 SOCKBUF_LOCK_ASSERT(sb); 1506 sbflush_internal(sb); 1507 } 1508 1509 void 1510 sbflush(struct sockbuf *sb) 1511 { 1512 1513 SOCKBUF_LOCK(sb); 1514 sbflush_locked(sb); 1515 SOCKBUF_UNLOCK(sb); 1516 } 1517 1518 /* 1519 * Cut data from (the front of) a sockbuf. 1520 */ 1521 static struct mbuf * 1522 sbcut_internal(struct sockbuf *sb, int len) 1523 { 1524 struct mbuf *m, *next, *mfree; 1525 bool is_tls; 1526 1527 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0", 1528 __func__, len)); 1529 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u", 1530 __func__, len, sb->sb_ccc)); 1531 1532 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1533 is_tls = false; 1534 mfree = NULL; 1535 1536 while (len > 0) { 1537 if (m == NULL) { 1538 #ifdef KERN_TLS 1539 if (next == NULL && !is_tls) { 1540 if (sb->sb_tlsdcc != 0) { 1541 MPASS(len >= sb->sb_tlsdcc); 1542 len -= sb->sb_tlsdcc; 1543 sb->sb_ccc -= sb->sb_tlsdcc; 1544 sb->sb_tlsdcc = 0; 1545 if (len == 0) 1546 break; 1547 } 1548 next = sb->sb_mtls; 1549 is_tls = true; 1550 } 1551 #endif 1552 KASSERT(next, ("%s: no next, len %d", __func__, len)); 1553 m = next; 1554 next = m->m_nextpkt; 1555 } 1556 if (m->m_len > len) { 1557 KASSERT(!(m->m_flags & M_NOTAVAIL), 1558 ("%s: m %p M_NOTAVAIL", __func__, m)); 1559 m->m_len -= len; 1560 m->m_data += len; 1561 sb->sb_ccc -= len; 1562 sb->sb_acc -= len; 1563 if (sb->sb_sndptroff != 0) 1564 sb->sb_sndptroff -= len; 1565 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1566 sb->sb_ctl -= len; 1567 break; 1568 } 1569 len -= m->m_len; 1570 #ifdef KERN_TLS 1571 if (is_tls) 1572 sbfree_ktls_rx(sb, m); 1573 else 1574 #endif 1575 sbfree(sb, m); 1576 /* 1577 * Do not put M_NOTREADY buffers to the free list, they 1578 * are referenced from outside. 1579 */ 1580 if (m->m_flags & M_NOTREADY && !is_tls) 1581 m = m->m_next; 1582 else { 1583 struct mbuf *n; 1584 1585 n = m->m_next; 1586 m->m_next = mfree; 1587 mfree = m; 1588 m = n; 1589 } 1590 } 1591 /* 1592 * Free any zero-length mbufs from the buffer. 1593 * For SOCK_DGRAM sockets such mbufs represent empty records. 1594 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, 1595 * when sosend_generic() needs to send only control data. 1596 */ 1597 while (m && m->m_len == 0) { 1598 struct mbuf *n; 1599 1600 sbfree(sb, m); 1601 n = m->m_next; 1602 m->m_next = mfree; 1603 mfree = m; 1604 m = n; 1605 } 1606 #ifdef KERN_TLS 1607 if (is_tls) { 1608 sb->sb_mb = NULL; 1609 sb->sb_mtls = m; 1610 if (m == NULL) 1611 sb->sb_mtlstail = NULL; 1612 } else 1613 #endif 1614 if (m) { 1615 sb->sb_mb = m; 1616 m->m_nextpkt = next; 1617 } else 1618 sb->sb_mb = next; 1619 /* 1620 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 1621 * sb_lastrecord is up-to-date if we dropped part of the last record. 1622 */ 1623 m = sb->sb_mb; 1624 if (m == NULL) { 1625 sb->sb_mbtail = NULL; 1626 sb->sb_lastrecord = NULL; 1627 } else if (m->m_nextpkt == NULL) { 1628 sb->sb_lastrecord = m; 1629 } 1630 1631 return (mfree); 1632 } 1633 1634 /* 1635 * Drop data from (the front of) a sockbuf. 1636 */ 1637 void 1638 sbdrop_locked(struct sockbuf *sb, int len) 1639 { 1640 1641 SOCKBUF_LOCK_ASSERT(sb); 1642 m_freem(sbcut_internal(sb, len)); 1643 } 1644 1645 /* 1646 * Drop data from (the front of) a sockbuf, 1647 * and return it to caller. 1648 */ 1649 struct mbuf * 1650 sbcut_locked(struct sockbuf *sb, int len) 1651 { 1652 1653 SOCKBUF_LOCK_ASSERT(sb); 1654 return (sbcut_internal(sb, len)); 1655 } 1656 1657 void 1658 sbdrop(struct sockbuf *sb, int len) 1659 { 1660 struct mbuf *mfree; 1661 1662 SOCKBUF_LOCK(sb); 1663 mfree = sbcut_internal(sb, len); 1664 SOCKBUF_UNLOCK(sb); 1665 1666 m_freem(mfree); 1667 } 1668 1669 struct mbuf * 1670 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff) 1671 { 1672 struct mbuf *m; 1673 1674 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1675 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1676 *moff = off; 1677 if (sb->sb_sndptr == NULL) { 1678 sb->sb_sndptr = sb->sb_mb; 1679 sb->sb_sndptroff = 0; 1680 } 1681 return (sb->sb_mb); 1682 } else { 1683 m = sb->sb_sndptr; 1684 off -= sb->sb_sndptroff; 1685 } 1686 *moff = off; 1687 return (m); 1688 } 1689 1690 void 1691 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len) 1692 { 1693 /* 1694 * A small copy was done, advance forward the sb_sbsndptr to cover 1695 * it. 1696 */ 1697 struct mbuf *m; 1698 1699 if (mb != sb->sb_sndptr) { 1700 /* Did not copyout at the same mbuf */ 1701 return; 1702 } 1703 m = mb; 1704 while (m && (len > 0)) { 1705 if (len >= m->m_len) { 1706 len -= m->m_len; 1707 if (m->m_next) { 1708 sb->sb_sndptroff += m->m_len; 1709 sb->sb_sndptr = m->m_next; 1710 } 1711 m = m->m_next; 1712 } else { 1713 len = 0; 1714 } 1715 } 1716 } 1717 1718 /* 1719 * Return the first mbuf and the mbuf data offset for the provided 1720 * send offset without changing the "sb_sndptroff" field. 1721 */ 1722 struct mbuf * 1723 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) 1724 { 1725 struct mbuf *m; 1726 1727 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1728 1729 /* 1730 * If the "off" is below the stored offset, which happens on 1731 * retransmits, just use "sb_mb": 1732 */ 1733 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1734 m = sb->sb_mb; 1735 } else { 1736 m = sb->sb_sndptr; 1737 off -= sb->sb_sndptroff; 1738 } 1739 while (off > 0 && m != NULL) { 1740 if (off < m->m_len) 1741 break; 1742 off -= m->m_len; 1743 m = m->m_next; 1744 } 1745 *moff = off; 1746 return (m); 1747 } 1748 1749 /* 1750 * Drop a record off the front of a sockbuf and move the next record to the 1751 * front. 1752 */ 1753 void 1754 sbdroprecord_locked(struct sockbuf *sb) 1755 { 1756 struct mbuf *m; 1757 1758 SOCKBUF_LOCK_ASSERT(sb); 1759 1760 m = sb->sb_mb; 1761 if (m) { 1762 sb->sb_mb = m->m_nextpkt; 1763 do { 1764 sbfree(sb, m); 1765 m = m_free(m); 1766 } while (m); 1767 } 1768 SB_EMPTY_FIXUP(sb); 1769 } 1770 1771 /* 1772 * Drop a record off the front of a sockbuf and move the next record to the 1773 * front. 1774 */ 1775 void 1776 sbdroprecord(struct sockbuf *sb) 1777 { 1778 1779 SOCKBUF_LOCK(sb); 1780 sbdroprecord_locked(sb); 1781 SOCKBUF_UNLOCK(sb); 1782 } 1783 1784 /* 1785 * Create a "control" mbuf containing the specified data with the specified 1786 * type for presentation on a socket buffer. 1787 */ 1788 struct mbuf * 1789 sbcreatecontrol(const void *p, u_int size, int type, int level, int wait) 1790 { 1791 struct cmsghdr *cp; 1792 struct mbuf *m; 1793 1794 MBUF_CHECKSLEEP(wait); 1795 1796 if (wait == M_NOWAIT) { 1797 if (CMSG_SPACE(size) > MCLBYTES) 1798 return (NULL); 1799 } else 1800 KASSERT(CMSG_SPACE(size) <= MCLBYTES, 1801 ("%s: passed CMSG_SPACE(%u) > MCLBYTES", __func__, size)); 1802 1803 if (CMSG_SPACE(size) > MLEN) 1804 m = m_getcl(wait, MT_CONTROL, 0); 1805 else 1806 m = m_get(wait, MT_CONTROL); 1807 if (m == NULL) 1808 return (NULL); 1809 1810 KASSERT(CMSG_SPACE(size) <= M_TRAILINGSPACE(m), 1811 ("sbcreatecontrol: short mbuf")); 1812 /* 1813 * Don't leave the padding between the msg header and the 1814 * cmsg data and the padding after the cmsg data un-initialized. 1815 */ 1816 cp = mtod(m, struct cmsghdr *); 1817 bzero(cp, CMSG_SPACE(size)); 1818 if (p != NULL) 1819 (void)memcpy(CMSG_DATA(cp), p, size); 1820 m->m_len = CMSG_SPACE(size); 1821 cp->cmsg_len = CMSG_LEN(size); 1822 cp->cmsg_level = level; 1823 cp->cmsg_type = type; 1824 return (m); 1825 } 1826 1827 /* 1828 * This does the same for socket buffers that sotoxsocket does for sockets: 1829 * generate an user-format data structure describing the socket buffer. Note 1830 * that the xsockbuf structure, since it is always embedded in a socket, does 1831 * not include a self pointer nor a length. We make this entry point public 1832 * in case some other mechanism needs it. 1833 */ 1834 void 1835 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1836 { 1837 1838 xsb->sb_cc = sb->sb_ccc; 1839 xsb->sb_hiwat = sb->sb_hiwat; 1840 xsb->sb_mbcnt = sb->sb_mbcnt; 1841 xsb->sb_mbmax = sb->sb_mbmax; 1842 xsb->sb_lowat = sb->sb_lowat; 1843 xsb->sb_flags = sb->sb_flags; 1844 xsb->sb_timeo = sb->sb_timeo; 1845 } 1846 1847 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1848 static int dummy; 1849 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, ""); 1850 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, 1851 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0, 1852 sysctl_handle_sb_max, "LU", 1853 "Maximum socket buffer size"); 1854 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1855 &sb_efficiency, 0, "Socket buffer size waste factor"); 1856