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