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 32 #include <sys/cdefs.h> 33 #include "opt_kern_tls.h" 34 #include "opt_param.h" 35 36 #include <sys/param.h> 37 #include <sys/aio.h> /* for aio_swake proto */ 38 #include <sys/kernel.h> 39 #include <sys/ktls.h> 40 #include <sys/lock.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/msan.h> 44 #include <sys/mutex.h> 45 #include <sys/proc.h> 46 #include <sys/protosw.h> 47 #include <sys/resourcevar.h> 48 #include <sys/signalvar.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sx.h> 52 #include <sys/sysctl.h> 53 54 #include <netinet/in.h> 55 56 /* 57 * Function pointer set by the AIO routines so that the socket buffer code 58 * can call back into the AIO module if it is loaded. 59 */ 60 void (*aio_swake)(struct socket *, struct sockbuf *); 61 62 /* 63 * Primitive routines for operating on socket buffers 64 */ 65 66 #define BUF_MAX_ADJ(_sz) (((u_quad_t)(_sz)) * MCLBYTES / (MSIZE + MCLBYTES)) 67 68 u_long sb_max = SB_MAX; 69 u_long sb_max_adj = BUF_MAX_ADJ(SB_MAX); 70 71 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 72 73 #ifdef KERN_TLS 74 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, 75 struct mbuf *n); 76 #endif 77 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); 78 static void sbunreserve_locked(struct socket *so, sb_which which); 79 80 /* 81 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY. 82 */ 83 static void 84 sbm_clrprotoflags(struct mbuf *m, int flags) 85 { 86 int mask; 87 88 mask = ~M_PROTOFLAGS; 89 if (flags & PRUS_NOTREADY) 90 mask |= M_NOTREADY; 91 while (m) { 92 m->m_flags &= mask; 93 m = m->m_next; 94 } 95 } 96 97 /* 98 * Compress M_NOTREADY mbufs after they have been readied by sbready(). 99 * 100 * sbcompress() skips M_NOTREADY mbufs since the data is not available to 101 * be copied at the time of sbcompress(). This function combines small 102 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first 103 * mbuf sbready() marked ready, and 'end' is the first mbuf still not 104 * ready. 105 */ 106 static void 107 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end) 108 { 109 struct mbuf *m, *n; 110 int ext_size; 111 112 SOCKBUF_LOCK_ASSERT(sb); 113 114 if ((sb->sb_flags & SB_NOCOALESCE) != 0) 115 return; 116 117 for (m = m0; m != end; m = m->m_next) { 118 MPASS((m->m_flags & M_NOTREADY) == 0); 119 /* 120 * NB: In sbcompress(), 'n' is the last mbuf in the 121 * socket buffer and 'm' is the new mbuf being copied 122 * into the trailing space of 'n'. Here, the roles 123 * are reversed and 'n' is the next mbuf after 'm' 124 * that is being copied into the trailing space of 125 * 'm'. 126 */ 127 n = m->m_next; 128 #ifdef KERN_TLS 129 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */ 130 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 131 (m->m_flags & M_EXTPG) && 132 (n->m_flags & M_EXTPG) && 133 !mbuf_has_tls_session(m) && 134 !mbuf_has_tls_session(n)) { 135 int hdr_len, trail_len; 136 137 hdr_len = n->m_epg_hdrlen; 138 trail_len = m->m_epg_trllen; 139 if (trail_len != 0 && hdr_len != 0 && 140 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) { 141 /* copy n's header to m's trailer */ 142 memcpy(&m->m_epg_trail[trail_len], 143 n->m_epg_hdr, hdr_len); 144 m->m_epg_trllen += hdr_len; 145 m->m_len += hdr_len; 146 n->m_epg_hdrlen = 0; 147 n->m_len -= hdr_len; 148 } 149 } 150 #endif 151 152 /* Compress small unmapped mbufs into plain mbufs. */ 153 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN && 154 !mbuf_has_tls_session(m)) { 155 ext_size = m->m_ext.ext_size; 156 if (mb_unmapped_compress(m) == 0) 157 sb->sb_mbcnt -= ext_size; 158 } 159 160 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 161 M_WRITABLE(m) && 162 (m->m_flags & M_EXTPG) == 0 && 163 !mbuf_has_tls_session(n) && 164 !mbuf_has_tls_session(m) && 165 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 166 n->m_len <= M_TRAILINGSPACE(m) && 167 m->m_type == n->m_type) { 168 KASSERT(sb->sb_lastrecord != n, 169 ("%s: merging start of record (%p) into previous mbuf (%p)", 170 __func__, n, m)); 171 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len)); 172 m->m_len += n->m_len; 173 m->m_next = n->m_next; 174 m->m_flags |= n->m_flags & M_EOR; 175 if (sb->sb_mbtail == n) 176 sb->sb_mbtail = m; 177 178 sb->sb_mbcnt -= MSIZE; 179 if (n->m_flags & M_EXT) 180 sb->sb_mbcnt -= n->m_ext.ext_size; 181 m_free(n); 182 n = m->m_next; 183 } 184 } 185 SBLASTRECORDCHK(sb); 186 SBLASTMBUFCHK(sb); 187 } 188 189 /* 190 * Mark ready "count" units of I/O starting with "m". Most mbufs 191 * count as a single unit of I/O except for M_EXTPG mbufs which 192 * are backed by multiple pages. 193 */ 194 int 195 sbready(struct sockbuf *sb, struct mbuf *m0, int count) 196 { 197 struct mbuf *m; 198 u_int blocker; 199 200 SOCKBUF_LOCK_ASSERT(sb); 201 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); 202 KASSERT(count > 0, ("%s: invalid count %d", __func__, count)); 203 204 m = m0; 205 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; 206 207 while (count > 0) { 208 KASSERT(m->m_flags & M_NOTREADY, 209 ("%s: m %p !M_NOTREADY", __func__, m)); 210 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) { 211 if (count < m->m_epg_nrdy) { 212 m->m_epg_nrdy -= count; 213 count = 0; 214 break; 215 } 216 count -= m->m_epg_nrdy; 217 m->m_epg_nrdy = 0; 218 } else 219 count--; 220 221 m->m_flags &= ~(M_NOTREADY | blocker); 222 if (blocker) 223 sb->sb_acc += m->m_len; 224 m = m->m_next; 225 } 226 227 /* 228 * If the first mbuf is still not fully ready because only 229 * some of its backing pages were readied, no further progress 230 * can be made. 231 */ 232 if (m0 == m) { 233 MPASS(m->m_flags & M_NOTREADY); 234 return (EINPROGRESS); 235 } 236 237 if (!blocker) { 238 sbready_compress(sb, m0, m); 239 return (EINPROGRESS); 240 } 241 242 /* This one was blocking all the queue. */ 243 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { 244 KASSERT(m->m_flags & M_BLOCKED, 245 ("%s: m %p !M_BLOCKED", __func__, m)); 246 m->m_flags &= ~M_BLOCKED; 247 sb->sb_acc += m->m_len; 248 } 249 250 sb->sb_fnrdy = m; 251 sbready_compress(sb, m0, m); 252 253 return (0); 254 } 255 256 /* 257 * Adjust sockbuf state reflecting allocation of m. 258 */ 259 void 260 sballoc(struct sockbuf *sb, struct mbuf *m) 261 { 262 263 SOCKBUF_LOCK_ASSERT(sb); 264 265 sb->sb_ccc += m->m_len; 266 267 if (sb->sb_fnrdy == NULL) { 268 if (m->m_flags & M_NOTREADY) 269 sb->sb_fnrdy = m; 270 else 271 sb->sb_acc += m->m_len; 272 } else 273 m->m_flags |= M_BLOCKED; 274 275 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 276 sb->sb_ctl += m->m_len; 277 278 sb->sb_mbcnt += MSIZE; 279 280 if (m->m_flags & M_EXT) 281 sb->sb_mbcnt += m->m_ext.ext_size; 282 } 283 284 /* 285 * Adjust sockbuf state reflecting freeing of m. 286 */ 287 void 288 sbfree(struct sockbuf *sb, struct mbuf *m) 289 { 290 291 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 292 SOCKBUF_LOCK_ASSERT(sb); 293 #endif 294 295 sb->sb_ccc -= m->m_len; 296 297 if (!(m->m_flags & M_NOTAVAIL)) 298 sb->sb_acc -= m->m_len; 299 300 if (m == sb->sb_fnrdy) { 301 struct mbuf *n; 302 303 KASSERT(m->m_flags & M_NOTREADY, 304 ("%s: m %p !M_NOTREADY", __func__, m)); 305 306 n = m->m_next; 307 while (n != NULL && !(n->m_flags & M_NOTREADY)) { 308 n->m_flags &= ~M_BLOCKED; 309 sb->sb_acc += n->m_len; 310 n = n->m_next; 311 } 312 sb->sb_fnrdy = n; 313 } 314 315 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 316 sb->sb_ctl -= m->m_len; 317 318 sb->sb_mbcnt -= MSIZE; 319 if (m->m_flags & M_EXT) 320 sb->sb_mbcnt -= m->m_ext.ext_size; 321 322 if (sb->sb_sndptr == m) { 323 sb->sb_sndptr = NULL; 324 sb->sb_sndptroff = 0; 325 } 326 if (sb->sb_sndptroff != 0) 327 sb->sb_sndptroff -= m->m_len; 328 } 329 330 #ifdef KERN_TLS 331 /* 332 * Similar to sballoc/sbfree but does not adjust state associated with 333 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs 334 * are not ready. 335 */ 336 void 337 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m) 338 { 339 340 SOCKBUF_LOCK_ASSERT(sb); 341 342 sb->sb_ccc += m->m_len; 343 sb->sb_tlscc += m->m_len; 344 345 sb->sb_mbcnt += MSIZE; 346 347 if (m->m_flags & M_EXT) 348 sb->sb_mbcnt += m->m_ext.ext_size; 349 } 350 351 void 352 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m) 353 { 354 355 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 356 SOCKBUF_LOCK_ASSERT(sb); 357 #endif 358 359 sb->sb_ccc -= m->m_len; 360 sb->sb_tlscc -= m->m_len; 361 362 sb->sb_mbcnt -= MSIZE; 363 364 if (m->m_flags & M_EXT) 365 sb->sb_mbcnt -= m->m_ext.ext_size; 366 } 367 #endif 368 369 /* 370 * Socantsendmore indicates that no more data will be sent on the socket; it 371 * would normally be applied to a socket when the user informs the system 372 * that no more data is to be sent, by the protocol code (in case 373 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be 374 * received, and will normally be applied to the socket by a protocol when it 375 * detects that the peer will send no more data. Data queued for reading in 376 * the socket may yet be read. 377 */ 378 void 379 socantsendmore_locked(struct socket *so) 380 { 381 382 SOCK_SENDBUF_LOCK_ASSERT(so); 383 384 so->so_snd.sb_state |= SBS_CANTSENDMORE; 385 sowwakeup_locked(so); 386 SOCK_SENDBUF_UNLOCK_ASSERT(so); 387 } 388 389 void 390 socantsendmore(struct socket *so) 391 { 392 393 SOCK_SENDBUF_LOCK(so); 394 socantsendmore_locked(so); 395 SOCK_SENDBUF_UNLOCK_ASSERT(so); 396 } 397 398 void 399 socantrcvmore_locked(struct socket *so) 400 { 401 402 SOCK_RECVBUF_LOCK_ASSERT(so); 403 404 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 405 #ifdef KERN_TLS 406 if (so->so_rcv.sb_flags & SB_TLS_RX) 407 ktls_check_rx(&so->so_rcv); 408 #endif 409 sorwakeup_locked(so); 410 SOCK_RECVBUF_UNLOCK_ASSERT(so); 411 } 412 413 void 414 socantrcvmore(struct socket *so) 415 { 416 417 SOCK_RECVBUF_LOCK(so); 418 socantrcvmore_locked(so); 419 SOCK_RECVBUF_UNLOCK_ASSERT(so); 420 } 421 422 void 423 soroverflow_locked(struct socket *so) 424 { 425 426 SOCK_RECVBUF_LOCK_ASSERT(so); 427 428 if (so->so_options & SO_RERROR) { 429 so->so_rerror = ENOBUFS; 430 sorwakeup_locked(so); 431 } else 432 SOCK_RECVBUF_UNLOCK(so); 433 434 SOCK_RECVBUF_UNLOCK_ASSERT(so); 435 } 436 437 void 438 soroverflow(struct socket *so) 439 { 440 441 SOCK_RECVBUF_LOCK(so); 442 soroverflow_locked(so); 443 SOCK_RECVBUF_UNLOCK_ASSERT(so); 444 } 445 446 /* 447 * Wait for data to arrive at/drain from a socket buffer. 448 */ 449 int 450 sbwait(struct socket *so, sb_which which) 451 { 452 struct sockbuf *sb; 453 454 SOCK_BUF_LOCK_ASSERT(so, which); 455 456 sb = sobuf(so, which); 457 sb->sb_flags |= SB_WAIT; 458 return (msleep_sbt(&sb->sb_acc, soeventmtx(so, which), 459 PSOCK | PCATCH, "sbwait", sb->sb_timeo, 0, 0)); 460 } 461 462 /* 463 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 464 * via SIGIO if the socket has the SS_ASYNC flag set. 465 * 466 * Called with the socket buffer lock held; will release the lock by the end 467 * of the function. This allows the caller to acquire the socket buffer lock 468 * while testing for the need for various sorts of wakeup and hold it through 469 * to the point where it's no longer required. We currently hold the lock 470 * through calls out to other subsystems (with the exception of kqueue), and 471 * then release it to avoid lock order issues. It's not clear that's 472 * correct. 473 */ 474 static __always_inline void 475 sowakeup(struct socket *so, const sb_which which) 476 { 477 struct sockbuf *sb; 478 int ret; 479 480 SOCK_BUF_LOCK_ASSERT(so, which); 481 482 sb = sobuf(so, which); 483 selwakeuppri(sb->sb_sel, PSOCK); 484 if (!SEL_WAITING(sb->sb_sel)) 485 sb->sb_flags &= ~SB_SEL; 486 if (sb->sb_flags & SB_WAIT) { 487 sb->sb_flags &= ~SB_WAIT; 488 wakeup(&sb->sb_acc); 489 } 490 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 491 if (sb->sb_upcall != NULL) { 492 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); 493 if (ret == SU_ISCONNECTED) { 494 KASSERT(sb == &so->so_rcv, 495 ("SO_SND upcall returned SU_ISCONNECTED")); 496 soupcall_clear(so, SO_RCV); 497 } 498 } else 499 ret = SU_OK; 500 if (sb->sb_flags & SB_AIO) 501 sowakeup_aio(so, which); 502 SOCK_BUF_UNLOCK(so, which); 503 if (ret == SU_ISCONNECTED) 504 soisconnected(so); 505 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 506 pgsigio(&so->so_sigio, SIGIO, 0); 507 SOCK_BUF_UNLOCK_ASSERT(so, which); 508 } 509 510 static void 511 splice_push(struct socket *so) 512 { 513 struct so_splice *sp; 514 515 SOCK_RECVBUF_LOCK_ASSERT(so); 516 517 sp = so->so_splice; 518 mtx_lock(&sp->mtx); 519 SOCK_RECVBUF_UNLOCK(so); 520 so_splice_dispatch(sp); 521 } 522 523 static void 524 splice_pull(struct socket *so) 525 { 526 struct so_splice *sp; 527 528 SOCK_SENDBUF_LOCK_ASSERT(so); 529 530 sp = so->so_splice_back; 531 mtx_lock(&sp->mtx); 532 SOCK_SENDBUF_UNLOCK(so); 533 so_splice_dispatch(sp); 534 } 535 536 /* 537 * Do we need to notify the other side when I/O is possible? 538 */ 539 static __always_inline bool 540 sb_notify(const struct sockbuf *sb) 541 { 542 return ((sb->sb_flags & (SB_WAIT | SB_SEL | SB_ASYNC | 543 SB_UPCALL | SB_AIO | SB_KNOTE)) != 0); 544 } 545 546 void 547 sorwakeup_locked(struct socket *so) 548 { 549 SOCK_RECVBUF_LOCK_ASSERT(so); 550 if (so->so_rcv.sb_flags & SB_SPLICED) 551 splice_push(so); 552 else if (sb_notify(&so->so_rcv)) 553 sowakeup(so, SO_RCV); 554 else 555 SOCK_RECVBUF_UNLOCK(so); 556 } 557 558 void 559 sowwakeup_locked(struct socket *so) 560 { 561 SOCK_SENDBUF_LOCK_ASSERT(so); 562 if (so->so_snd.sb_flags & SB_SPLICED) 563 splice_pull(so); 564 else if (sb_notify(&so->so_snd)) 565 sowakeup(so, SO_SND); 566 else 567 SOCK_SENDBUF_UNLOCK(so); 568 } 569 570 /* 571 * Socket buffer (struct sockbuf) utility routines. 572 * 573 * Each socket contains two socket buffers: one for sending data and one for 574 * receiving data. Each buffer contains a queue of mbufs, information about 575 * the number of mbufs and amount of data in the queue, and other fields 576 * allowing select() statements and notification on data availability to be 577 * implemented. 578 * 579 * Data stored in a socket buffer is maintained as a list of records. Each 580 * record is a list of mbufs chained together with the m_next field. Records 581 * are chained together with the m_nextpkt field. The upper level routine 582 * soreceive() expects the following conventions to be observed when placing 583 * information in the receive buffer: 584 * 585 * 1. If the protocol requires each message be preceded by the sender's name, 586 * then a record containing that name must be present before any 587 * associated data (mbuf's must be of type MT_SONAME). 588 * 2. If the protocol supports the exchange of ``access rights'' (really just 589 * additional data associated with the message), and there are ``rights'' 590 * to be received, then a record containing this data should be present 591 * (mbuf's must be of type MT_RIGHTS). 592 * 3. If a name or rights record exists, then it must be followed by a data 593 * record, perhaps of zero length. 594 * 595 * Before using a new socket structure it is first necessary to reserve 596 * buffer space to the socket, by calling sbreserve(). This should commit 597 * some of the available buffer space in the system buffer pool for the 598 * socket (currently, it does nothing but enforce limits). The space should 599 * be released by calling sbrelease() when the socket is destroyed. 600 */ 601 int 602 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 603 { 604 struct thread *td = curthread; 605 606 SOCK_SENDBUF_LOCK(so); 607 SOCK_RECVBUF_LOCK(so); 608 if (sbreserve_locked(so, SO_SND, sndcc, td) == 0) 609 goto bad; 610 if (sbreserve_locked(so, SO_RCV, rcvcc, td) == 0) 611 goto bad2; 612 if (so->so_rcv.sb_lowat == 0) 613 so->so_rcv.sb_lowat = 1; 614 if (so->so_snd.sb_lowat == 0) 615 so->so_snd.sb_lowat = MCLBYTES; 616 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 617 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 618 SOCK_RECVBUF_UNLOCK(so); 619 SOCK_SENDBUF_UNLOCK(so); 620 return (0); 621 bad2: 622 sbunreserve_locked(so, SO_SND); 623 bad: 624 SOCK_RECVBUF_UNLOCK(so); 625 SOCK_SENDBUF_UNLOCK(so); 626 return (ENOBUFS); 627 } 628 629 static int 630 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 631 { 632 int error = 0; 633 u_long tmp_sb_max = sb_max; 634 635 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 636 if (error || !req->newptr) 637 return (error); 638 if (tmp_sb_max < MSIZE + MCLBYTES) 639 return (EINVAL); 640 sb_max = tmp_sb_max; 641 sb_max_adj = BUF_MAX_ADJ(sb_max); 642 return (0); 643 } 644 645 /* 646 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 647 * become limiting if buffering efficiency is near the normal case. 648 */ 649 bool 650 sbreserve_locked_limit(struct socket *so, sb_which which, u_long cc, 651 u_long buf_max, struct thread *td) 652 { 653 struct sockbuf *sb = sobuf(so, which); 654 rlim_t sbsize_limit; 655 656 SOCK_BUF_LOCK_ASSERT(so, which); 657 658 /* 659 * When a thread is passed, we take into account the thread's socket 660 * buffer size limit. The caller will generally pass curthread, but 661 * in the TCP input path, NULL will be passed to indicate that no 662 * appropriate thread resource limits are available. In that case, 663 * we don't apply a process limit. 664 */ 665 if (cc > BUF_MAX_ADJ(buf_max)) 666 return (false); 667 if (td != NULL) { 668 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); 669 } else 670 sbsize_limit = RLIM_INFINITY; 671 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 672 sbsize_limit)) 673 return (false); 674 sb->sb_mbmax = min(cc * sb_efficiency, buf_max); 675 if (sb->sb_lowat > sb->sb_hiwat) 676 sb->sb_lowat = sb->sb_hiwat; 677 return (true); 678 } 679 680 bool 681 sbreserve_locked(struct socket *so, sb_which which, u_long cc, 682 struct thread *td) 683 { 684 return (sbreserve_locked_limit(so, which, cc, sb_max, td)); 685 } 686 687 static void 688 sbunreserve_locked(struct socket *so, sb_which which) 689 { 690 struct sockbuf *sb = sobuf(so, which); 691 692 SOCK_BUF_LOCK_ASSERT(so, which); 693 694 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 695 RLIM_INFINITY); 696 sb->sb_mbmax = 0; 697 } 698 699 int 700 sbsetopt(struct socket *so, struct sockopt *sopt) 701 { 702 struct sockbuf *sb; 703 sb_which wh; 704 short *flags; 705 u_int cc, *hiwat, *lowat; 706 int error, optval; 707 708 error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); 709 if (error != 0) 710 return (error); 711 712 /* 713 * Values < 1 make no sense for any of these options, 714 * so disallow them. 715 */ 716 if (optval < 1) 717 return (EINVAL); 718 cc = optval; 719 720 sb = NULL; 721 SOCK_LOCK(so); 722 if (SOLISTENING(so)) { 723 switch (sopt->sopt_name) { 724 case SO_SNDLOWAT: 725 case SO_SNDBUF: 726 lowat = &so->sol_sbsnd_lowat; 727 hiwat = &so->sol_sbsnd_hiwat; 728 flags = &so->sol_sbsnd_flags; 729 break; 730 case SO_RCVLOWAT: 731 case SO_RCVBUF: 732 lowat = &so->sol_sbrcv_lowat; 733 hiwat = &so->sol_sbrcv_hiwat; 734 flags = &so->sol_sbrcv_flags; 735 break; 736 } 737 } else { 738 switch (sopt->sopt_name) { 739 case SO_SNDLOWAT: 740 case SO_SNDBUF: 741 sb = &so->so_snd; 742 wh = SO_SND; 743 break; 744 case SO_RCVLOWAT: 745 case SO_RCVBUF: 746 sb = &so->so_rcv; 747 wh = SO_RCV; 748 break; 749 } 750 flags = &sb->sb_flags; 751 hiwat = &sb->sb_hiwat; 752 lowat = &sb->sb_lowat; 753 SOCK_BUF_LOCK(so, wh); 754 } 755 756 error = 0; 757 switch (sopt->sopt_name) { 758 case SO_SNDBUF: 759 case SO_RCVBUF: 760 if (SOLISTENING(so)) { 761 if (cc > sb_max_adj) { 762 error = ENOBUFS; 763 break; 764 } 765 *hiwat = cc; 766 if (*lowat > *hiwat) 767 *lowat = *hiwat; 768 } else { 769 if (!sbreserve_locked(so, wh, cc, curthread)) 770 error = ENOBUFS; 771 } 772 if (error == 0) 773 *flags &= ~SB_AUTOSIZE; 774 break; 775 case SO_SNDLOWAT: 776 case SO_RCVLOWAT: 777 /* 778 * Make sure the low-water is never greater than the 779 * high-water. 780 */ 781 *lowat = (cc > *hiwat) ? *hiwat : cc; 782 break; 783 } 784 785 if (!SOLISTENING(so)) 786 SOCK_BUF_UNLOCK(so, wh); 787 SOCK_UNLOCK(so); 788 return (error); 789 } 790 791 /* 792 * Free mbufs held by a socket, and reserved mbuf space. 793 */ 794 void 795 sbrelease_locked(struct socket *so, sb_which which) 796 { 797 struct sockbuf *sb = sobuf(so, which); 798 799 SOCK_BUF_LOCK_ASSERT(so, which); 800 801 sbflush_locked(sb); 802 sbunreserve_locked(so, which); 803 } 804 805 void 806 sbrelease(struct socket *so, sb_which which) 807 { 808 809 SOCK_BUF_LOCK(so, which); 810 sbrelease_locked(so, which); 811 SOCK_BUF_UNLOCK(so, which); 812 } 813 814 void 815 sbdestroy(struct socket *so, sb_which which) 816 { 817 #ifdef KERN_TLS 818 struct sockbuf *sb = sobuf(so, which); 819 820 if (sb->sb_tls_info != NULL) 821 ktls_free(sb->sb_tls_info); 822 sb->sb_tls_info = NULL; 823 #endif 824 sbrelease_locked(so, which); 825 } 826 827 /* 828 * Routines to add and remove data from an mbuf queue. 829 * 830 * The routines sbappend() or sbappendrecord() are normally called to append 831 * new mbufs to a socket buffer, after checking that adequate space is 832 * available, comparing the function sbspace() with the amount of data to be 833 * added. sbappendrecord() differs from sbappend() in that data supplied is 834 * treated as the beginning of a new record. To place a sender's address, 835 * optional access rights, and data in a socket receive buffer, 836 * sbappendaddr() should be used. To place access rights and data in a 837 * socket receive buffer, sbappendrights() should be used. In either case, 838 * the new data begins a new record. Note that unlike sbappend() and 839 * sbappendrecord(), these routines check for the caller that there will be 840 * enough space to store the data. Each fails if there is not enough space, 841 * or if it cannot find mbufs to store additional information in. 842 * 843 * Reliable protocols may use the socket send buffer to hold data awaiting 844 * acknowledgement. Data is normally copied from a socket send buffer in a 845 * protocol with m_copy for output to a peer, and then removing the data from 846 * the socket buffer with sbdrop() or sbdroprecord() when the data is 847 * acknowledged by the peer. 848 */ 849 #ifdef SOCKBUF_DEBUG 850 void 851 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 852 { 853 struct mbuf *m = sb->sb_mb; 854 855 SOCKBUF_LOCK_ASSERT(sb); 856 857 while (m && m->m_nextpkt) 858 m = m->m_nextpkt; 859 860 if (m != sb->sb_lastrecord) { 861 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 862 __func__, sb->sb_mb, sb->sb_lastrecord, m); 863 printf("packet chain:\n"); 864 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 865 printf("\t%p\n", m); 866 panic("%s from %s:%u", __func__, file, line); 867 } 868 } 869 870 void 871 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 872 { 873 struct mbuf *m = sb->sb_mb; 874 struct mbuf *n; 875 876 SOCKBUF_LOCK_ASSERT(sb); 877 878 while (m && m->m_nextpkt) 879 m = m->m_nextpkt; 880 881 while (m && m->m_next) 882 m = m->m_next; 883 884 if (m != sb->sb_mbtail) { 885 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 886 __func__, sb->sb_mb, sb->sb_mbtail, m); 887 printf("packet tree:\n"); 888 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 889 printf("\t"); 890 for (n = m; n != NULL; n = n->m_next) 891 printf("%p ", n); 892 printf("\n"); 893 } 894 panic("%s from %s:%u", __func__, file, line); 895 } 896 897 #ifdef KERN_TLS 898 m = sb->sb_mtls; 899 while (m && m->m_next) 900 m = m->m_next; 901 902 if (m != sb->sb_mtlstail) { 903 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n", 904 __func__, sb->sb_mtls, sb->sb_mtlstail, m); 905 printf("TLS packet tree:\n"); 906 printf("\t"); 907 for (m = sb->sb_mtls; m != NULL; m = m->m_next) { 908 printf("%p ", m); 909 } 910 printf("\n"); 911 panic("%s from %s:%u", __func__, file, line); 912 } 913 #endif 914 } 915 #endif /* SOCKBUF_DEBUG */ 916 917 #define SBLINKRECORD(sb, m0) do { \ 918 SOCKBUF_LOCK_ASSERT(sb); \ 919 if ((sb)->sb_lastrecord != NULL) \ 920 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 921 else \ 922 (sb)->sb_mb = (m0); \ 923 (sb)->sb_lastrecord = (m0); \ 924 } while (/*CONSTCOND*/0) 925 926 /* 927 * Append mbuf chain m to the last record in the socket buffer sb. The 928 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 929 * are discarded and mbufs are compacted where possible. 930 */ 931 void 932 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) 933 { 934 struct mbuf *n; 935 936 SOCKBUF_LOCK_ASSERT(sb); 937 938 if (m == NULL) 939 return; 940 kmsan_check_mbuf(m, "sbappend"); 941 sbm_clrprotoflags(m, flags); 942 SBLASTRECORDCHK(sb); 943 n = sb->sb_mb; 944 if (n) { 945 while (n->m_nextpkt) 946 n = n->m_nextpkt; 947 do { 948 if (n->m_flags & M_EOR) { 949 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 950 return; 951 } 952 } while (n->m_next && (n = n->m_next)); 953 } else { 954 /* 955 * XXX Would like to simply use sb_mbtail here, but 956 * XXX I need to verify that I won't miss an EOR that 957 * XXX way. 958 */ 959 if ((n = sb->sb_lastrecord) != NULL) { 960 do { 961 if (n->m_flags & M_EOR) { 962 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 963 return; 964 } 965 } while (n->m_next && (n = n->m_next)); 966 } else { 967 /* 968 * If this is the first record in the socket buffer, 969 * it's also the last record. 970 */ 971 sb->sb_lastrecord = m; 972 } 973 } 974 sbcompress(sb, m, n); 975 SBLASTRECORDCHK(sb); 976 } 977 978 /* 979 * Append mbuf chain m to the last record in the socket buffer sb. The 980 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 981 * are discarded and mbufs are compacted where possible. 982 */ 983 void 984 sbappend(struct sockbuf *sb, struct mbuf *m, int flags) 985 { 986 987 SOCKBUF_LOCK(sb); 988 sbappend_locked(sb, m, flags); 989 SOCKBUF_UNLOCK(sb); 990 } 991 992 #ifdef KERN_TLS 993 /* 994 * Append an mbuf containing encrypted TLS data. The data 995 * is marked M_NOTREADY until it has been decrypted and 996 * stored as a TLS record. 997 */ 998 static void 999 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m) 1000 { 1001 struct ifnet *ifp; 1002 struct mbuf *n; 1003 int flags; 1004 1005 ifp = NULL; 1006 flags = M_NOTREADY; 1007 1008 SBLASTMBUFCHK(sb); 1009 1010 /* Mbuf chain must start with a packet header. */ 1011 MPASS((m->m_flags & M_PKTHDR) != 0); 1012 1013 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 1014 for (n = m; n != NULL; n = n->m_next) { 1015 if (n->m_flags & M_PKTHDR) { 1016 ifp = m->m_pkthdr.leaf_rcvif; 1017 if ((n->m_pkthdr.csum_flags & CSUM_TLS_MASK) == 1018 CSUM_TLS_DECRYPTED) { 1019 /* Mark all mbufs in this packet decrypted. */ 1020 flags = M_NOTREADY | M_DECRYPTED; 1021 } else { 1022 flags = M_NOTREADY; 1023 } 1024 m_demote_pkthdr(n); 1025 } 1026 1027 n->m_flags &= M_DEMOTEFLAGS; 1028 n->m_flags |= flags; 1029 1030 MPASS((n->m_flags & M_NOTREADY) != 0); 1031 } 1032 1033 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail); 1034 ktls_check_rx(sb); 1035 1036 /* Check for incoming packet route changes: */ 1037 if (ifp != NULL && sb->sb_tls_info->rx_ifp != NULL && 1038 sb->sb_tls_info->rx_ifp != ifp) 1039 ktls_input_ifp_mismatch(sb, ifp); 1040 } 1041 #endif 1042 1043 /* 1044 * This version of sbappend() should only be used when the caller absolutely 1045 * knows that there will never be more than one record in the socket buffer, 1046 * that is, a stream protocol (such as TCP). 1047 */ 1048 void 1049 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) 1050 { 1051 SOCKBUF_LOCK_ASSERT(sb); 1052 1053 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 1054 1055 kmsan_check_mbuf(m, "sbappend"); 1056 1057 #ifdef KERN_TLS 1058 /* 1059 * Decrypted TLS records are appended as records via 1060 * sbappendrecord(). TCP passes encrypted TLS records to this 1061 * function which must be scheduled for decryption. 1062 */ 1063 if (sb->sb_flags & SB_TLS_RX) { 1064 sbappend_ktls_rx(sb, m); 1065 return; 1066 } 1067 #endif 1068 1069 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 1070 1071 SBLASTMBUFCHK(sb); 1072 1073 #ifdef KERN_TLS 1074 if (sb->sb_tls_info != NULL) 1075 ktls_seq(sb, m); 1076 #endif 1077 1078 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 1079 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); 1080 1081 sbcompress(sb, m, sb->sb_mbtail); 1082 1083 sb->sb_lastrecord = sb->sb_mb; 1084 SBLASTRECORDCHK(sb); 1085 } 1086 1087 /* 1088 * This version of sbappend() should only be used when the caller absolutely 1089 * knows that there will never be more than one record in the socket buffer, 1090 * that is, a stream protocol (such as TCP). 1091 */ 1092 void 1093 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) 1094 { 1095 1096 SOCKBUF_LOCK(sb); 1097 sbappendstream_locked(sb, m, flags); 1098 SOCKBUF_UNLOCK(sb); 1099 } 1100 1101 #ifdef SOCKBUF_DEBUG 1102 void 1103 sbcheck(struct sockbuf *sb, const char *file, int line) 1104 { 1105 struct mbuf *m, *n, *fnrdy; 1106 u_long acc, ccc, mbcnt; 1107 #ifdef KERN_TLS 1108 u_long tlscc; 1109 #endif 1110 1111 SOCKBUF_LOCK_ASSERT(sb); 1112 1113 acc = ccc = mbcnt = 0; 1114 fnrdy = NULL; 1115 1116 for (m = sb->sb_mb; m; m = n) { 1117 n = m->m_nextpkt; 1118 for (; m; m = m->m_next) { 1119 if (m->m_len == 0) { 1120 printf("sb %p empty mbuf %p\n", sb, m); 1121 goto fail; 1122 } 1123 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { 1124 if (m != sb->sb_fnrdy) { 1125 printf("sb %p: fnrdy %p != m %p\n", 1126 sb, sb->sb_fnrdy, m); 1127 goto fail; 1128 } 1129 fnrdy = m; 1130 } 1131 if (fnrdy) { 1132 if (!(m->m_flags & M_NOTAVAIL)) { 1133 printf("sb %p: fnrdy %p, m %p is avail\n", 1134 sb, sb->sb_fnrdy, m); 1135 goto fail; 1136 } 1137 } else 1138 acc += m->m_len; 1139 ccc += m->m_len; 1140 mbcnt += MSIZE; 1141 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1142 mbcnt += m->m_ext.ext_size; 1143 } 1144 } 1145 #ifdef KERN_TLS 1146 /* 1147 * Account for mbufs "detached" by ktls_detach_record() while 1148 * they are decrypted by ktls_decrypt(). tlsdcc gives a count 1149 * of the detached bytes that are included in ccc. The mbufs 1150 * and clusters are not included in the socket buffer 1151 * accounting. 1152 */ 1153 ccc += sb->sb_tlsdcc; 1154 1155 tlscc = 0; 1156 for (m = sb->sb_mtls; m; m = m->m_next) { 1157 if (m->m_nextpkt != NULL) { 1158 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m); 1159 goto fail; 1160 } 1161 if ((m->m_flags & M_NOTREADY) == 0) { 1162 printf("sb %p TLS mbuf %p ready\n", sb, m); 1163 goto fail; 1164 } 1165 tlscc += m->m_len; 1166 ccc += m->m_len; 1167 mbcnt += MSIZE; 1168 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1169 mbcnt += m->m_ext.ext_size; 1170 } 1171 1172 if (sb->sb_tlscc != tlscc) { 1173 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1174 sb->sb_tlsdcc); 1175 goto fail; 1176 } 1177 #endif 1178 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { 1179 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", 1180 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); 1181 #ifdef KERN_TLS 1182 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1183 sb->sb_tlsdcc); 1184 #endif 1185 goto fail; 1186 } 1187 return; 1188 fail: 1189 panic("%s from %s:%u", __func__, file, line); 1190 } 1191 #endif 1192 1193 /* 1194 * As above, except the mbuf chain begins a new record. 1195 */ 1196 void 1197 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 1198 { 1199 struct mbuf *m; 1200 1201 SOCKBUF_LOCK_ASSERT(sb); 1202 1203 if (m0 == NULL) 1204 return; 1205 1206 kmsan_check_mbuf(m0, "sbappend"); 1207 m_clrprotoflags(m0); 1208 1209 /* 1210 * Put the first mbuf on the queue. Note this permits zero length 1211 * records. 1212 */ 1213 sballoc(sb, m0); 1214 SBLASTRECORDCHK(sb); 1215 SBLINKRECORD(sb, m0); 1216 sb->sb_mbtail = m0; 1217 m = m0->m_next; 1218 m0->m_next = 0; 1219 if (m && (m0->m_flags & M_EOR)) { 1220 m0->m_flags &= ~M_EOR; 1221 m->m_flags |= M_EOR; 1222 } 1223 /* always call sbcompress() so it can do SBLASTMBUFCHK() */ 1224 sbcompress(sb, m, m0); 1225 } 1226 1227 /* 1228 * As above, except the mbuf chain begins a new record. 1229 */ 1230 void 1231 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 1232 { 1233 1234 SOCKBUF_LOCK(sb); 1235 sbappendrecord_locked(sb, m0); 1236 SOCKBUF_UNLOCK(sb); 1237 } 1238 1239 /* Helper routine that appends data, control, and address to a sockbuf. */ 1240 static int 1241 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, 1242 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) 1243 { 1244 struct mbuf *m, *n, *nlast; 1245 1246 if (m0 != NULL) 1247 kmsan_check_mbuf(m0, "sbappend"); 1248 if (control != NULL) 1249 kmsan_check_mbuf(control, "sbappend"); 1250 1251 #if MSIZE <= 256 1252 if (asa->sa_len > MLEN) 1253 return (0); 1254 #endif 1255 m = m_get(M_NOWAIT, MT_SONAME); 1256 if (m == NULL) 1257 return (0); 1258 m->m_len = asa->sa_len; 1259 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 1260 if (m0) { 1261 M_ASSERT_NO_SND_TAG(m0); 1262 m_clrprotoflags(m0); 1263 m_tag_delete_chain(m0, NULL); 1264 /* 1265 * Clear some persistent info from pkthdr. 1266 * We don't use m_demote(), because some netgraph consumers 1267 * expect M_PKTHDR presence. 1268 */ 1269 m0->m_pkthdr.rcvif = NULL; 1270 m0->m_pkthdr.flowid = 0; 1271 m0->m_pkthdr.csum_flags = 0; 1272 m0->m_pkthdr.fibnum = 0; 1273 m0->m_pkthdr.rsstype = 0; 1274 } 1275 if (ctrl_last) 1276 ctrl_last->m_next = m0; /* concatenate data to control */ 1277 else 1278 control = m0; 1279 m->m_next = control; 1280 for (n = m; n->m_next != NULL; n = n->m_next) 1281 sballoc(sb, n); 1282 sballoc(sb, n); 1283 nlast = n; 1284 SBLINKRECORD(sb, m); 1285 1286 sb->sb_mbtail = nlast; 1287 SBLASTMBUFCHK(sb); 1288 1289 SBLASTRECORDCHK(sb); 1290 return (1); 1291 } 1292 1293 /* 1294 * Append address and data, and optionally, control (ancillary) data to the 1295 * receive queue of a socket. If present, m0 must include a packet header 1296 * with total length. Returns 0 if no space in sockbuf or insufficient 1297 * mbufs. 1298 */ 1299 int 1300 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 1301 struct mbuf *m0, struct mbuf *control) 1302 { 1303 struct mbuf *ctrl_last; 1304 int space = asa->sa_len; 1305 1306 SOCKBUF_LOCK_ASSERT(sb); 1307 1308 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1309 panic("sbappendaddr_locked"); 1310 if (m0) 1311 space += m0->m_pkthdr.len; 1312 space += m_length(control, &ctrl_last); 1313 1314 if (space > sbspace(sb)) 1315 return (0); 1316 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1317 } 1318 1319 /* 1320 * Append address and data, and optionally, control (ancillary) data to the 1321 * receive queue of a socket. If present, m0 must include a packet header 1322 * with total length. Returns 0 if insufficient mbufs. Does not validate space 1323 * on the receiving sockbuf. 1324 */ 1325 int 1326 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, 1327 struct mbuf *m0, struct mbuf *control) 1328 { 1329 struct mbuf *ctrl_last; 1330 1331 SOCKBUF_LOCK_ASSERT(sb); 1332 1333 ctrl_last = (control == NULL) ? NULL : m_last(control); 1334 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1335 } 1336 1337 /* 1338 * Append address and data, and optionally, control (ancillary) data to the 1339 * receive queue of a socket. If present, m0 must include a packet header 1340 * with total length. Returns 0 if no space in sockbuf or insufficient 1341 * mbufs. 1342 */ 1343 int 1344 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 1345 struct mbuf *m0, struct mbuf *control) 1346 { 1347 int retval; 1348 1349 SOCKBUF_LOCK(sb); 1350 retval = sbappendaddr_locked(sb, asa, m0, control); 1351 SOCKBUF_UNLOCK(sb); 1352 return (retval); 1353 } 1354 1355 void 1356 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 1357 struct mbuf *control, int flags) 1358 { 1359 struct mbuf *m, *mlast; 1360 1361 if (m0 != NULL) 1362 kmsan_check_mbuf(m0, "sbappend"); 1363 kmsan_check_mbuf(control, "sbappend"); 1364 1365 sbm_clrprotoflags(m0, flags); 1366 m_last(control)->m_next = m0; 1367 1368 SBLASTRECORDCHK(sb); 1369 1370 for (m = control; m->m_next; m = m->m_next) 1371 sballoc(sb, m); 1372 sballoc(sb, m); 1373 mlast = m; 1374 SBLINKRECORD(sb, control); 1375 1376 sb->sb_mbtail = mlast; 1377 SBLASTMBUFCHK(sb); 1378 1379 SBLASTRECORDCHK(sb); 1380 } 1381 1382 void 1383 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control, 1384 int flags) 1385 { 1386 1387 SOCKBUF_LOCK(sb); 1388 sbappendcontrol_locked(sb, m0, control, flags); 1389 SOCKBUF_UNLOCK(sb); 1390 } 1391 1392 /* 1393 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 1394 * (n). If (n) is NULL, the buffer is presumed empty. 1395 * 1396 * When the data is compressed, mbufs in the chain may be handled in one of 1397 * three ways: 1398 * 1399 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 1400 * record boundary, and no change in data type). 1401 * 1402 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 1403 * an mbuf already in the socket buffer. This can occur if an 1404 * appropriate mbuf exists, there is room, both mbufs are not marked as 1405 * not ready, and no merging of data types will occur. 1406 * 1407 * (3) The mbuf may be appended to the end of the existing mbuf chain. 1408 * 1409 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 1410 * end-of-record. 1411 */ 1412 void 1413 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1414 { 1415 int eor = 0; 1416 struct mbuf *o; 1417 1418 SOCKBUF_LOCK_ASSERT(sb); 1419 1420 while (m) { 1421 eor |= m->m_flags & M_EOR; 1422 if (m->m_len == 0 && 1423 (eor == 0 || 1424 (((o = m->m_next) || (o = n)) && 1425 o->m_type == m->m_type))) { 1426 if (sb->sb_lastrecord == m) 1427 sb->sb_lastrecord = m->m_next; 1428 m = m_free(m); 1429 continue; 1430 } 1431 if (n && (n->m_flags & M_EOR) == 0 && 1432 M_WRITABLE(n) && 1433 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1434 !(m->m_flags & M_NOTREADY) && 1435 !(n->m_flags & (M_NOTREADY | M_EXTPG)) && 1436 !mbuf_has_tls_session(m) && 1437 !mbuf_has_tls_session(n) && 1438 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1439 m->m_len <= M_TRAILINGSPACE(n) && 1440 n->m_type == m->m_type) { 1441 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1442 n->m_len += m->m_len; 1443 sb->sb_ccc += m->m_len; 1444 if (sb->sb_fnrdy == NULL) 1445 sb->sb_acc += m->m_len; 1446 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1447 /* XXX: Probably don't need.*/ 1448 sb->sb_ctl += m->m_len; 1449 m = m_free(m); 1450 continue; 1451 } 1452 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) && 1453 (m->m_flags & M_NOTREADY) == 0 && 1454 !mbuf_has_tls_session(m)) 1455 (void)mb_unmapped_compress(m); 1456 if (n) 1457 n->m_next = m; 1458 else 1459 sb->sb_mb = m; 1460 sb->sb_mbtail = m; 1461 sballoc(sb, m); 1462 n = m; 1463 m->m_flags &= ~M_EOR; 1464 m = m->m_next; 1465 n->m_next = 0; 1466 } 1467 if (eor) { 1468 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 1469 n->m_flags |= eor; 1470 } 1471 SBLASTMBUFCHK(sb); 1472 } 1473 1474 #ifdef KERN_TLS 1475 /* 1476 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs 1477 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also 1478 * a bit simpler (no EOR markers, always MT_DATA, etc.). 1479 */ 1480 static void 1481 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1482 { 1483 1484 SOCKBUF_LOCK_ASSERT(sb); 1485 1486 while (m) { 1487 KASSERT((m->m_flags & M_EOR) == 0, 1488 ("TLS RX mbuf %p with EOR", m)); 1489 KASSERT(m->m_type == MT_DATA, 1490 ("TLS RX mbuf %p is not MT_DATA", m)); 1491 KASSERT((m->m_flags & M_NOTREADY) != 0, 1492 ("TLS RX mbuf %p ready", m)); 1493 KASSERT((m->m_flags & M_EXTPG) == 0, 1494 ("TLS RX mbuf %p unmapped", m)); 1495 1496 if (m->m_len == 0) { 1497 m = m_free(m); 1498 continue; 1499 } 1500 1501 /* 1502 * Even though both 'n' and 'm' are NOTREADY, it's ok 1503 * to coalesce the data. 1504 */ 1505 if (n && 1506 M_WRITABLE(n) && 1507 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1508 !((m->m_flags ^ n->m_flags) & M_DECRYPTED) && 1509 !(n->m_flags & M_EXTPG) && 1510 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1511 m->m_len <= M_TRAILINGSPACE(n)) { 1512 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1513 n->m_len += m->m_len; 1514 sb->sb_ccc += m->m_len; 1515 sb->sb_tlscc += m->m_len; 1516 m = m_free(m); 1517 continue; 1518 } 1519 if (n) 1520 n->m_next = m; 1521 else 1522 sb->sb_mtls = m; 1523 sb->sb_mtlstail = m; 1524 sballoc_ktls_rx(sb, m); 1525 n = m; 1526 m = m->m_next; 1527 n->m_next = NULL; 1528 } 1529 SBLASTMBUFCHK(sb); 1530 } 1531 #endif 1532 1533 /* 1534 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 1535 */ 1536 void 1537 sbflush_locked(struct sockbuf *sb) 1538 { 1539 1540 SOCKBUF_LOCK_ASSERT(sb); 1541 1542 while (sb->sb_mbcnt || sb->sb_tlsdcc) { 1543 /* 1544 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: 1545 * we would loop forever. Panic instead. 1546 */ 1547 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1548 break; 1549 m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); 1550 } 1551 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 1552 ("%s: ccc %u mb %p mbcnt %u", __func__, 1553 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 1554 } 1555 1556 void 1557 sbflush(struct sockbuf *sb) 1558 { 1559 1560 SOCKBUF_LOCK(sb); 1561 sbflush_locked(sb); 1562 SOCKBUF_UNLOCK(sb); 1563 } 1564 1565 /* 1566 * Cut data from (the front of) a sockbuf. 1567 */ 1568 static struct mbuf * 1569 sbcut_internal(struct sockbuf *sb, int len) 1570 { 1571 struct mbuf *m, *next, *mfree; 1572 bool is_tls; 1573 1574 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0", 1575 __func__, len)); 1576 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u", 1577 __func__, len, sb->sb_ccc)); 1578 1579 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1580 is_tls = false; 1581 mfree = NULL; 1582 1583 while (len > 0) { 1584 if (m == NULL) { 1585 #ifdef KERN_TLS 1586 if (next == NULL && !is_tls) { 1587 if (sb->sb_tlsdcc != 0) { 1588 MPASS(len >= sb->sb_tlsdcc); 1589 len -= sb->sb_tlsdcc; 1590 sb->sb_ccc -= sb->sb_tlsdcc; 1591 sb->sb_tlsdcc = 0; 1592 if (len == 0) 1593 break; 1594 } 1595 next = sb->sb_mtls; 1596 is_tls = true; 1597 } 1598 #endif 1599 KASSERT(next, ("%s: no next, len %d", __func__, len)); 1600 m = next; 1601 next = m->m_nextpkt; 1602 } 1603 if (m->m_len > len) { 1604 KASSERT(!(m->m_flags & M_NOTAVAIL), 1605 ("%s: m %p M_NOTAVAIL", __func__, m)); 1606 m->m_len -= len; 1607 m->m_data += len; 1608 sb->sb_ccc -= len; 1609 sb->sb_acc -= len; 1610 if (sb->sb_sndptroff != 0) 1611 sb->sb_sndptroff -= len; 1612 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1613 sb->sb_ctl -= len; 1614 break; 1615 } 1616 len -= m->m_len; 1617 #ifdef KERN_TLS 1618 if (is_tls) 1619 sbfree_ktls_rx(sb, m); 1620 else 1621 #endif 1622 sbfree(sb, m); 1623 /* 1624 * Do not put M_NOTREADY buffers to the free list, they 1625 * are referenced from outside. 1626 */ 1627 if (m->m_flags & M_NOTREADY && !is_tls) 1628 m = m->m_next; 1629 else { 1630 struct mbuf *n; 1631 1632 n = m->m_next; 1633 m->m_next = mfree; 1634 mfree = m; 1635 m = n; 1636 } 1637 } 1638 /* 1639 * Free any zero-length mbufs from the buffer. 1640 * For SOCK_DGRAM sockets such mbufs represent empty records. 1641 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, 1642 * when sosend_generic() needs to send only control data. 1643 */ 1644 while (m && m->m_len == 0) { 1645 struct mbuf *n; 1646 1647 sbfree(sb, m); 1648 n = m->m_next; 1649 m->m_next = mfree; 1650 mfree = m; 1651 m = n; 1652 } 1653 #ifdef KERN_TLS 1654 if (is_tls) { 1655 sb->sb_mb = NULL; 1656 sb->sb_mtls = m; 1657 if (m == NULL) 1658 sb->sb_mtlstail = NULL; 1659 } else 1660 #endif 1661 if (m) { 1662 sb->sb_mb = m; 1663 m->m_nextpkt = next; 1664 } else 1665 sb->sb_mb = next; 1666 /* 1667 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 1668 * sb_lastrecord is up-to-date if we dropped part of the last record. 1669 */ 1670 m = sb->sb_mb; 1671 if (m == NULL) { 1672 sb->sb_mbtail = NULL; 1673 sb->sb_lastrecord = NULL; 1674 } else if (m->m_nextpkt == NULL) { 1675 sb->sb_lastrecord = m; 1676 } 1677 1678 return (mfree); 1679 } 1680 1681 /* 1682 * Drop data from (the front of) a sockbuf. 1683 */ 1684 void 1685 sbdrop_locked(struct sockbuf *sb, int len) 1686 { 1687 1688 SOCKBUF_LOCK_ASSERT(sb); 1689 m_freem(sbcut_internal(sb, len)); 1690 } 1691 1692 /* 1693 * Drop data from (the front of) a sockbuf, 1694 * and return it to caller. 1695 */ 1696 struct mbuf * 1697 sbcut_locked(struct sockbuf *sb, int len) 1698 { 1699 1700 SOCKBUF_LOCK_ASSERT(sb); 1701 return (sbcut_internal(sb, len)); 1702 } 1703 1704 void 1705 sbdrop(struct sockbuf *sb, int len) 1706 { 1707 struct mbuf *mfree; 1708 1709 SOCKBUF_LOCK(sb); 1710 mfree = sbcut_internal(sb, len); 1711 SOCKBUF_UNLOCK(sb); 1712 1713 m_freem(mfree); 1714 } 1715 1716 struct mbuf * 1717 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff) 1718 { 1719 struct mbuf *m; 1720 1721 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1722 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1723 *moff = off; 1724 if (sb->sb_sndptr == NULL) { 1725 sb->sb_sndptr = sb->sb_mb; 1726 sb->sb_sndptroff = 0; 1727 } 1728 return (sb->sb_mb); 1729 } else { 1730 m = sb->sb_sndptr; 1731 off -= sb->sb_sndptroff; 1732 } 1733 *moff = off; 1734 return (m); 1735 } 1736 1737 void 1738 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len) 1739 { 1740 /* 1741 * A small copy was done, advance forward the sb_sbsndptr to cover 1742 * it. 1743 */ 1744 struct mbuf *m; 1745 1746 if (mb != sb->sb_sndptr) { 1747 /* Did not copyout at the same mbuf */ 1748 return; 1749 } 1750 m = mb; 1751 while (m && (len > 0)) { 1752 if (len >= m->m_len) { 1753 len -= m->m_len; 1754 if (m->m_next) { 1755 sb->sb_sndptroff += m->m_len; 1756 sb->sb_sndptr = m->m_next; 1757 } 1758 m = m->m_next; 1759 } else { 1760 len = 0; 1761 } 1762 } 1763 } 1764 1765 /* 1766 * Return the first mbuf and the mbuf data offset for the provided 1767 * send offset without changing the "sb_sndptroff" field. 1768 */ 1769 struct mbuf * 1770 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) 1771 { 1772 struct mbuf *m; 1773 1774 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1775 1776 /* 1777 * If the "off" is below the stored offset, which happens on 1778 * retransmits, just use "sb_mb": 1779 */ 1780 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1781 m = sb->sb_mb; 1782 } else { 1783 m = sb->sb_sndptr; 1784 off -= sb->sb_sndptroff; 1785 } 1786 while (off > 0 && m != NULL) { 1787 if (off < m->m_len) 1788 break; 1789 off -= m->m_len; 1790 m = m->m_next; 1791 } 1792 *moff = off; 1793 return (m); 1794 } 1795 1796 /* 1797 * Drop a record off the front of a sockbuf and move the next record to the 1798 * front. 1799 */ 1800 void 1801 sbdroprecord_locked(struct sockbuf *sb) 1802 { 1803 struct mbuf *m; 1804 1805 SOCKBUF_LOCK_ASSERT(sb); 1806 1807 m = sb->sb_mb; 1808 if (m) { 1809 sb->sb_mb = m->m_nextpkt; 1810 do { 1811 sbfree(sb, m); 1812 m = m_free(m); 1813 } while (m); 1814 } 1815 SB_EMPTY_FIXUP(sb); 1816 } 1817 1818 /* 1819 * Drop a record off the front of a sockbuf and move the next record to the 1820 * front. 1821 */ 1822 void 1823 sbdroprecord(struct sockbuf *sb) 1824 { 1825 1826 SOCKBUF_LOCK(sb); 1827 sbdroprecord_locked(sb); 1828 SOCKBUF_UNLOCK(sb); 1829 } 1830 1831 /* 1832 * Create a "control" mbuf containing the specified data with the specified 1833 * type for presentation on a socket buffer. 1834 */ 1835 struct mbuf * 1836 sbcreatecontrol(const void *p, u_int size, int type, int level, int wait) 1837 { 1838 struct cmsghdr *cp; 1839 struct mbuf *m; 1840 1841 MBUF_CHECKSLEEP(wait); 1842 1843 if (wait == M_NOWAIT) { 1844 if (CMSG_SPACE(size) > MCLBYTES) 1845 return (NULL); 1846 } else 1847 KASSERT(CMSG_SPACE(size) <= MCLBYTES, 1848 ("%s: passed CMSG_SPACE(%u) > MCLBYTES", __func__, size)); 1849 1850 if (CMSG_SPACE(size) > MLEN) 1851 m = m_getcl(wait, MT_CONTROL, 0); 1852 else 1853 m = m_get(wait, MT_CONTROL); 1854 if (m == NULL) 1855 return (NULL); 1856 1857 KASSERT(CMSG_SPACE(size) <= M_TRAILINGSPACE(m), 1858 ("sbcreatecontrol: short mbuf")); 1859 /* 1860 * Don't leave the padding between the msg header and the 1861 * cmsg data and the padding after the cmsg data un-initialized. 1862 */ 1863 cp = mtod(m, struct cmsghdr *); 1864 bzero(cp, CMSG_SPACE(size)); 1865 if (p != NULL) 1866 (void)memcpy(CMSG_DATA(cp), p, size); 1867 m->m_len = CMSG_SPACE(size); 1868 cp->cmsg_len = CMSG_LEN(size); 1869 cp->cmsg_level = level; 1870 cp->cmsg_type = type; 1871 return (m); 1872 } 1873 1874 /* 1875 * This does the same for socket buffers that sotoxsocket does for sockets: 1876 * generate an user-format data structure describing the socket buffer. Note 1877 * that the xsockbuf structure, since it is always embedded in a socket, does 1878 * not include a self pointer nor a length. We make this entry point public 1879 * in case some other mechanism needs it. 1880 */ 1881 void 1882 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1883 { 1884 1885 xsb->sb_cc = sb->sb_ccc; 1886 xsb->sb_hiwat = sb->sb_hiwat; 1887 xsb->sb_mbcnt = sb->sb_mbcnt; 1888 xsb->sb_mbmax = sb->sb_mbmax; 1889 xsb->sb_lowat = sb->sb_lowat; 1890 xsb->sb_flags = sb->sb_flags; 1891 xsb->sb_timeo = sb->sb_timeo; 1892 } 1893 1894 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1895 static int dummy; 1896 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, ""); 1897 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, 1898 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0, 1899 sysctl_handle_sb_max, "LU", 1900 "Maximum socket buffer size"); 1901 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1902 &sb_efficiency, 0, "Socket buffer size waste factor"); 1903