1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 32 */ 33 34 #include <sys/cdefs.h> 35 #include "opt_kern_tls.h" 36 #include "opt_param.h" 37 38 #include <sys/param.h> 39 #include <sys/aio.h> /* for aio_swake proto */ 40 #include <sys/kernel.h> 41 #include <sys/ktls.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/msan.h> 46 #include <sys/mutex.h> 47 #include <sys/proc.h> 48 #include <sys/protosw.h> 49 #include <sys/resourcevar.h> 50 #include <sys/signalvar.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/sx.h> 54 #include <sys/sysctl.h> 55 56 #include <netinet/in.h> 57 58 /* 59 * Function pointer set by the AIO routines so that the socket buffer code 60 * can call back into the AIO module if it is loaded. 61 */ 62 void (*aio_swake)(struct socket *, struct sockbuf *); 63 64 /* 65 * Primitive routines for operating on socket buffers 66 */ 67 68 #define BUF_MAX_ADJ(_sz) (((u_quad_t)(_sz)) * MCLBYTES / (MSIZE + MCLBYTES)) 69 70 u_long sb_max = SB_MAX; 71 u_long sb_max_adj = BUF_MAX_ADJ(SB_MAX); 72 73 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 74 75 #ifdef KERN_TLS 76 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, 77 struct mbuf *n); 78 #endif 79 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len); 80 static void sbflush_internal(struct sockbuf *sb); 81 82 /* 83 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY. 84 */ 85 static void 86 sbm_clrprotoflags(struct mbuf *m, int flags) 87 { 88 int mask; 89 90 mask = ~M_PROTOFLAGS; 91 if (flags & PRUS_NOTREADY) 92 mask |= M_NOTREADY; 93 while (m) { 94 m->m_flags &= mask; 95 m = m->m_next; 96 } 97 } 98 99 /* 100 * Compress M_NOTREADY mbufs after they have been readied by sbready(). 101 * 102 * sbcompress() skips M_NOTREADY mbufs since the data is not available to 103 * be copied at the time of sbcompress(). This function combines small 104 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first 105 * mbuf sbready() marked ready, and 'end' is the first mbuf still not 106 * ready. 107 */ 108 static void 109 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end) 110 { 111 struct mbuf *m, *n; 112 int ext_size; 113 114 SOCKBUF_LOCK_ASSERT(sb); 115 116 if ((sb->sb_flags & SB_NOCOALESCE) != 0) 117 return; 118 119 for (m = m0; m != end; m = m->m_next) { 120 MPASS((m->m_flags & M_NOTREADY) == 0); 121 /* 122 * NB: In sbcompress(), 'n' is the last mbuf in the 123 * socket buffer and 'm' is the new mbuf being copied 124 * into the trailing space of 'n'. Here, the roles 125 * are reversed and 'n' is the next mbuf after 'm' 126 * that is being copied into the trailing space of 127 * 'm'. 128 */ 129 n = m->m_next; 130 #ifdef KERN_TLS 131 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */ 132 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 133 (m->m_flags & M_EXTPG) && 134 (n->m_flags & M_EXTPG) && 135 !mbuf_has_tls_session(m) && 136 !mbuf_has_tls_session(n)) { 137 int hdr_len, trail_len; 138 139 hdr_len = n->m_epg_hdrlen; 140 trail_len = m->m_epg_trllen; 141 if (trail_len != 0 && hdr_len != 0 && 142 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) { 143 /* copy n's header to m's trailer */ 144 memcpy(&m->m_epg_trail[trail_len], 145 n->m_epg_hdr, hdr_len); 146 m->m_epg_trllen += hdr_len; 147 m->m_len += hdr_len; 148 n->m_epg_hdrlen = 0; 149 n->m_len -= hdr_len; 150 } 151 } 152 #endif 153 154 /* Compress small unmapped mbufs into plain mbufs. */ 155 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN && 156 !mbuf_has_tls_session(m)) { 157 ext_size = m->m_ext.ext_size; 158 if (mb_unmapped_compress(m) == 0) 159 sb->sb_mbcnt -= ext_size; 160 } 161 162 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 && 163 M_WRITABLE(m) && 164 (m->m_flags & M_EXTPG) == 0 && 165 !mbuf_has_tls_session(n) && 166 !mbuf_has_tls_session(m) && 167 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 168 n->m_len <= M_TRAILINGSPACE(m) && 169 m->m_type == n->m_type) { 170 KASSERT(sb->sb_lastrecord != n, 171 ("%s: merging start of record (%p) into previous mbuf (%p)", 172 __func__, n, m)); 173 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len)); 174 m->m_len += n->m_len; 175 m->m_next = n->m_next; 176 m->m_flags |= n->m_flags & M_EOR; 177 if (sb->sb_mbtail == n) 178 sb->sb_mbtail = m; 179 180 sb->sb_mbcnt -= MSIZE; 181 if (n->m_flags & M_EXT) 182 sb->sb_mbcnt -= n->m_ext.ext_size; 183 m_free(n); 184 n = m->m_next; 185 } 186 } 187 SBLASTRECORDCHK(sb); 188 SBLASTMBUFCHK(sb); 189 } 190 191 /* 192 * Mark ready "count" units of I/O starting with "m". Most mbufs 193 * count as a single unit of I/O except for M_EXTPG mbufs which 194 * are backed by multiple pages. 195 */ 196 int 197 sbready(struct sockbuf *sb, struct mbuf *m0, int count) 198 { 199 struct mbuf *m; 200 u_int blocker; 201 202 SOCKBUF_LOCK_ASSERT(sb); 203 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb)); 204 KASSERT(count > 0, ("%s: invalid count %d", __func__, count)); 205 206 m = m0; 207 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0; 208 209 while (count > 0) { 210 KASSERT(m->m_flags & M_NOTREADY, 211 ("%s: m %p !M_NOTREADY", __func__, m)); 212 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) { 213 if (count < m->m_epg_nrdy) { 214 m->m_epg_nrdy -= count; 215 count = 0; 216 break; 217 } 218 count -= m->m_epg_nrdy; 219 m->m_epg_nrdy = 0; 220 } else 221 count--; 222 223 m->m_flags &= ~(M_NOTREADY | blocker); 224 if (blocker) 225 sb->sb_acc += m->m_len; 226 m = m->m_next; 227 } 228 229 /* 230 * If the first mbuf is still not fully ready because only 231 * some of its backing pages were readied, no further progress 232 * can be made. 233 */ 234 if (m0 == m) { 235 MPASS(m->m_flags & M_NOTREADY); 236 return (EINPROGRESS); 237 } 238 239 if (!blocker) { 240 sbready_compress(sb, m0, m); 241 return (EINPROGRESS); 242 } 243 244 /* This one was blocking all the queue. */ 245 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) { 246 KASSERT(m->m_flags & M_BLOCKED, 247 ("%s: m %p !M_BLOCKED", __func__, m)); 248 m->m_flags &= ~M_BLOCKED; 249 sb->sb_acc += m->m_len; 250 } 251 252 sb->sb_fnrdy = m; 253 sbready_compress(sb, m0, m); 254 255 return (0); 256 } 257 258 /* 259 * Adjust sockbuf state reflecting allocation of m. 260 */ 261 void 262 sballoc(struct sockbuf *sb, struct mbuf *m) 263 { 264 265 SOCKBUF_LOCK_ASSERT(sb); 266 267 sb->sb_ccc += m->m_len; 268 269 if (sb->sb_fnrdy == NULL) { 270 if (m->m_flags & M_NOTREADY) 271 sb->sb_fnrdy = m; 272 else 273 sb->sb_acc += m->m_len; 274 } else 275 m->m_flags |= M_BLOCKED; 276 277 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 278 sb->sb_ctl += m->m_len; 279 280 sb->sb_mbcnt += MSIZE; 281 282 if (m->m_flags & M_EXT) 283 sb->sb_mbcnt += m->m_ext.ext_size; 284 } 285 286 /* 287 * Adjust sockbuf state reflecting freeing of m. 288 */ 289 void 290 sbfree(struct sockbuf *sb, struct mbuf *m) 291 { 292 293 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 294 SOCKBUF_LOCK_ASSERT(sb); 295 #endif 296 297 sb->sb_ccc -= m->m_len; 298 299 if (!(m->m_flags & M_NOTAVAIL)) 300 sb->sb_acc -= m->m_len; 301 302 if (m == sb->sb_fnrdy) { 303 struct mbuf *n; 304 305 KASSERT(m->m_flags & M_NOTREADY, 306 ("%s: m %p !M_NOTREADY", __func__, m)); 307 308 n = m->m_next; 309 while (n != NULL && !(n->m_flags & M_NOTREADY)) { 310 n->m_flags &= ~M_BLOCKED; 311 sb->sb_acc += n->m_len; 312 n = n->m_next; 313 } 314 sb->sb_fnrdy = n; 315 } 316 317 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 318 sb->sb_ctl -= m->m_len; 319 320 sb->sb_mbcnt -= MSIZE; 321 if (m->m_flags & M_EXT) 322 sb->sb_mbcnt -= m->m_ext.ext_size; 323 324 if (sb->sb_sndptr == m) { 325 sb->sb_sndptr = NULL; 326 sb->sb_sndptroff = 0; 327 } 328 if (sb->sb_sndptroff != 0) 329 sb->sb_sndptroff -= m->m_len; 330 } 331 332 #ifdef KERN_TLS 333 /* 334 * Similar to sballoc/sbfree but does not adjust state associated with 335 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs 336 * are not ready. 337 */ 338 void 339 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m) 340 { 341 342 SOCKBUF_LOCK_ASSERT(sb); 343 344 sb->sb_ccc += m->m_len; 345 sb->sb_tlscc += m->m_len; 346 347 sb->sb_mbcnt += MSIZE; 348 349 if (m->m_flags & M_EXT) 350 sb->sb_mbcnt += m->m_ext.ext_size; 351 } 352 353 void 354 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m) 355 { 356 357 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */ 358 SOCKBUF_LOCK_ASSERT(sb); 359 #endif 360 361 sb->sb_ccc -= m->m_len; 362 sb->sb_tlscc -= m->m_len; 363 364 sb->sb_mbcnt -= MSIZE; 365 366 if (m->m_flags & M_EXT) 367 sb->sb_mbcnt -= m->m_ext.ext_size; 368 } 369 #endif 370 371 /* 372 * Socantsendmore indicates that no more data will be sent on the socket; it 373 * would normally be applied to a socket when the user informs the system 374 * that no more data is to be sent, by the protocol code (in case 375 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be 376 * received, and will normally be applied to the socket by a protocol when it 377 * detects that the peer will send no more data. Data queued for reading in 378 * the socket may yet be read. 379 */ 380 void 381 socantsendmore_locked(struct socket *so) 382 { 383 384 SOCK_SENDBUF_LOCK_ASSERT(so); 385 386 so->so_snd.sb_state |= SBS_CANTSENDMORE; 387 sowwakeup_locked(so); 388 SOCK_SENDBUF_UNLOCK_ASSERT(so); 389 } 390 391 void 392 socantsendmore(struct socket *so) 393 { 394 395 SOCK_SENDBUF_LOCK(so); 396 socantsendmore_locked(so); 397 SOCK_SENDBUF_UNLOCK_ASSERT(so); 398 } 399 400 void 401 socantrcvmore_locked(struct socket *so) 402 { 403 404 SOCK_RECVBUF_LOCK_ASSERT(so); 405 406 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 407 #ifdef KERN_TLS 408 if (so->so_rcv.sb_flags & SB_TLS_RX) 409 ktls_check_rx(&so->so_rcv); 410 #endif 411 sorwakeup_locked(so); 412 SOCK_RECVBUF_UNLOCK_ASSERT(so); 413 } 414 415 void 416 socantrcvmore(struct socket *so) 417 { 418 419 SOCK_RECVBUF_LOCK(so); 420 socantrcvmore_locked(so); 421 SOCK_RECVBUF_UNLOCK_ASSERT(so); 422 } 423 424 void 425 soroverflow_locked(struct socket *so) 426 { 427 428 SOCK_RECVBUF_LOCK_ASSERT(so); 429 430 if (so->so_options & SO_RERROR) { 431 so->so_rerror = ENOBUFS; 432 sorwakeup_locked(so); 433 } else 434 SOCK_RECVBUF_UNLOCK(so); 435 436 SOCK_RECVBUF_UNLOCK_ASSERT(so); 437 } 438 439 void 440 soroverflow(struct socket *so) 441 { 442 443 SOCK_RECVBUF_LOCK(so); 444 soroverflow_locked(so); 445 SOCK_RECVBUF_UNLOCK_ASSERT(so); 446 } 447 448 /* 449 * Wait for data to arrive at/drain from a socket buffer. 450 */ 451 int 452 sbwait(struct socket *so, sb_which which) 453 { 454 struct sockbuf *sb; 455 456 SOCK_BUF_LOCK_ASSERT(so, which); 457 458 sb = sobuf(so, which); 459 sb->sb_flags |= SB_WAIT; 460 return (msleep_sbt(&sb->sb_acc, soeventmtx(so, which), 461 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", 462 sb->sb_timeo, 0, 0)); 463 } 464 465 /* 466 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 467 * via SIGIO if the socket has the SS_ASYNC flag set. 468 * 469 * Called with the socket buffer lock held; will release the lock by the end 470 * of the function. This allows the caller to acquire the socket buffer lock 471 * while testing for the need for various sorts of wakeup and hold it through 472 * to the point where it's no longer required. We currently hold the lock 473 * through calls out to other subsystems (with the exception of kqueue), and 474 * then release it to avoid lock order issues. It's not clear that's 475 * correct. 476 */ 477 static __always_inline void 478 sowakeup(struct socket *so, const sb_which which) 479 { 480 struct sockbuf *sb; 481 int ret; 482 483 SOCK_BUF_LOCK_ASSERT(so, which); 484 485 sb = sobuf(so, which); 486 selwakeuppri(sb->sb_sel, PSOCK); 487 if (!SEL_WAITING(sb->sb_sel)) 488 sb->sb_flags &= ~SB_SEL; 489 if (sb->sb_flags & SB_WAIT) { 490 sb->sb_flags &= ~SB_WAIT; 491 wakeup(&sb->sb_acc); 492 } 493 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 494 if (sb->sb_upcall != NULL) { 495 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); 496 if (ret == SU_ISCONNECTED) { 497 KASSERT(sb == &so->so_rcv, 498 ("SO_SND upcall returned SU_ISCONNECTED")); 499 soupcall_clear(so, SO_RCV); 500 } 501 } else 502 ret = SU_OK; 503 if (sb->sb_flags & SB_AIO) 504 sowakeup_aio(so, which); 505 SOCK_BUF_UNLOCK(so, which); 506 if (ret == SU_ISCONNECTED) 507 soisconnected(so); 508 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 509 pgsigio(&so->so_sigio, SIGIO, 0); 510 SOCK_BUF_UNLOCK_ASSERT(so, which); 511 } 512 513 /* 514 * Do we need to notify the other side when I/O is possible? 515 */ 516 static __always_inline bool 517 sb_notify(const struct sockbuf *sb) 518 { 519 return ((sb->sb_flags & (SB_WAIT | SB_SEL | SB_ASYNC | 520 SB_UPCALL | SB_AIO | SB_KNOTE)) != 0); 521 } 522 523 void 524 sorwakeup_locked(struct socket *so) 525 { 526 SOCK_RECVBUF_LOCK_ASSERT(so); 527 if (sb_notify(&so->so_rcv)) 528 sowakeup(so, SO_RCV); 529 else 530 SOCK_RECVBUF_UNLOCK(so); 531 } 532 533 void 534 sowwakeup_locked(struct socket *so) 535 { 536 SOCK_SENDBUF_LOCK_ASSERT(so); 537 if (sb_notify(&so->so_snd)) 538 sowakeup(so, SO_SND); 539 else 540 SOCK_SENDBUF_UNLOCK(so); 541 } 542 543 /* 544 * Socket buffer (struct sockbuf) utility routines. 545 * 546 * Each socket contains two socket buffers: one for sending data and one for 547 * receiving data. Each buffer contains a queue of mbufs, information about 548 * the number of mbufs and amount of data in the queue, and other fields 549 * allowing select() statements and notification on data availability to be 550 * implemented. 551 * 552 * Data stored in a socket buffer is maintained as a list of records. Each 553 * record is a list of mbufs chained together with the m_next field. Records 554 * are chained together with the m_nextpkt field. The upper level routine 555 * soreceive() expects the following conventions to be observed when placing 556 * information in the receive buffer: 557 * 558 * 1. If the protocol requires each message be preceded by the sender's name, 559 * then a record containing that name must be present before any 560 * associated data (mbuf's must be of type MT_SONAME). 561 * 2. If the protocol supports the exchange of ``access rights'' (really just 562 * additional data associated with the message), and there are ``rights'' 563 * to be received, then a record containing this data should be present 564 * (mbuf's must be of type MT_RIGHTS). 565 * 3. If a name or rights record exists, then it must be followed by a data 566 * record, perhaps of zero length. 567 * 568 * Before using a new socket structure it is first necessary to reserve 569 * buffer space to the socket, by calling sbreserve(). This should commit 570 * some of the available buffer space in the system buffer pool for the 571 * socket (currently, it does nothing but enforce limits). The space should 572 * be released by calling sbrelease() when the socket is destroyed. 573 */ 574 int 575 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 576 { 577 struct thread *td = curthread; 578 579 SOCK_SENDBUF_LOCK(so); 580 SOCK_RECVBUF_LOCK(so); 581 if (sbreserve_locked(so, SO_SND, sndcc, td) == 0) 582 goto bad; 583 if (sbreserve_locked(so, SO_RCV, rcvcc, td) == 0) 584 goto bad2; 585 if (so->so_rcv.sb_lowat == 0) 586 so->so_rcv.sb_lowat = 1; 587 if (so->so_snd.sb_lowat == 0) 588 so->so_snd.sb_lowat = MCLBYTES; 589 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 590 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 591 SOCK_RECVBUF_UNLOCK(so); 592 SOCK_SENDBUF_UNLOCK(so); 593 return (0); 594 bad2: 595 sbrelease_locked(so, SO_SND); 596 bad: 597 SOCK_RECVBUF_UNLOCK(so); 598 SOCK_SENDBUF_UNLOCK(so); 599 return (ENOBUFS); 600 } 601 602 static int 603 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 604 { 605 int error = 0; 606 u_long tmp_sb_max = sb_max; 607 608 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 609 if (error || !req->newptr) 610 return (error); 611 if (tmp_sb_max < MSIZE + MCLBYTES) 612 return (EINVAL); 613 sb_max = tmp_sb_max; 614 sb_max_adj = BUF_MAX_ADJ(sb_max); 615 return (0); 616 } 617 618 /* 619 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 620 * become limiting if buffering efficiency is near the normal case. 621 */ 622 bool 623 sbreserve_locked_limit(struct socket *so, sb_which which, u_long cc, 624 u_long buf_max, struct thread *td) 625 { 626 struct sockbuf *sb = sobuf(so, which); 627 rlim_t sbsize_limit; 628 629 SOCK_BUF_LOCK_ASSERT(so, which); 630 631 /* 632 * When a thread is passed, we take into account the thread's socket 633 * buffer size limit. The caller will generally pass curthread, but 634 * in the TCP input path, NULL will be passed to indicate that no 635 * appropriate thread resource limits are available. In that case, 636 * we don't apply a process limit. 637 */ 638 if (cc > BUF_MAX_ADJ(buf_max)) 639 return (false); 640 if (td != NULL) { 641 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); 642 } else 643 sbsize_limit = RLIM_INFINITY; 644 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 645 sbsize_limit)) 646 return (false); 647 sb->sb_mbmax = min(cc * sb_efficiency, buf_max); 648 if (sb->sb_lowat > sb->sb_hiwat) 649 sb->sb_lowat = sb->sb_hiwat; 650 return (true); 651 } 652 653 bool 654 sbreserve_locked(struct socket *so, sb_which which, u_long cc, 655 struct thread *td) 656 { 657 return (sbreserve_locked_limit(so, which, cc, sb_max, td)); 658 } 659 660 int 661 sbsetopt(struct socket *so, struct sockopt *sopt) 662 { 663 struct sockbuf *sb; 664 sb_which wh; 665 short *flags; 666 u_int cc, *hiwat, *lowat; 667 int error, optval; 668 669 error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); 670 if (error != 0) 671 return (error); 672 673 /* 674 * Values < 1 make no sense for any of these options, 675 * so disallow them. 676 */ 677 if (optval < 1) 678 return (EINVAL); 679 cc = optval; 680 681 sb = NULL; 682 SOCK_LOCK(so); 683 if (SOLISTENING(so)) { 684 switch (sopt->sopt_name) { 685 case SO_SNDLOWAT: 686 case SO_SNDBUF: 687 lowat = &so->sol_sbsnd_lowat; 688 hiwat = &so->sol_sbsnd_hiwat; 689 flags = &so->sol_sbsnd_flags; 690 break; 691 case SO_RCVLOWAT: 692 case SO_RCVBUF: 693 lowat = &so->sol_sbrcv_lowat; 694 hiwat = &so->sol_sbrcv_hiwat; 695 flags = &so->sol_sbrcv_flags; 696 break; 697 } 698 } else { 699 switch (sopt->sopt_name) { 700 case SO_SNDLOWAT: 701 case SO_SNDBUF: 702 sb = &so->so_snd; 703 wh = SO_SND; 704 break; 705 case SO_RCVLOWAT: 706 case SO_RCVBUF: 707 sb = &so->so_rcv; 708 wh = SO_RCV; 709 break; 710 } 711 flags = &sb->sb_flags; 712 hiwat = &sb->sb_hiwat; 713 lowat = &sb->sb_lowat; 714 SOCK_BUF_LOCK(so, wh); 715 } 716 717 error = 0; 718 switch (sopt->sopt_name) { 719 case SO_SNDBUF: 720 case SO_RCVBUF: 721 if (SOLISTENING(so)) { 722 if (cc > sb_max_adj) { 723 error = ENOBUFS; 724 break; 725 } 726 *hiwat = cc; 727 if (*lowat > *hiwat) 728 *lowat = *hiwat; 729 } else { 730 if (!sbreserve_locked(so, wh, cc, curthread)) 731 error = ENOBUFS; 732 } 733 if (error == 0) 734 *flags &= ~SB_AUTOSIZE; 735 break; 736 case SO_SNDLOWAT: 737 case SO_RCVLOWAT: 738 /* 739 * Make sure the low-water is never greater than the 740 * high-water. 741 */ 742 *lowat = (cc > *hiwat) ? *hiwat : cc; 743 break; 744 } 745 746 if (!SOLISTENING(so)) 747 SOCK_BUF_UNLOCK(so, wh); 748 SOCK_UNLOCK(so); 749 return (error); 750 } 751 752 /* 753 * Free mbufs held by a socket, and reserved mbuf space. 754 */ 755 static void 756 sbrelease_internal(struct socket *so, sb_which which) 757 { 758 struct sockbuf *sb = sobuf(so, which); 759 760 sbflush_internal(sb); 761 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 762 RLIM_INFINITY); 763 sb->sb_mbmax = 0; 764 } 765 766 void 767 sbrelease_locked(struct socket *so, sb_which which) 768 { 769 770 SOCK_BUF_LOCK_ASSERT(so, which); 771 772 sbrelease_internal(so, which); 773 } 774 775 void 776 sbrelease(struct socket *so, sb_which which) 777 { 778 779 SOCK_BUF_LOCK(so, which); 780 sbrelease_locked(so, which); 781 SOCK_BUF_UNLOCK(so, which); 782 } 783 784 void 785 sbdestroy(struct socket *so, sb_which which) 786 { 787 #ifdef KERN_TLS 788 struct sockbuf *sb = sobuf(so, which); 789 790 if (sb->sb_tls_info != NULL) 791 ktls_free(sb->sb_tls_info); 792 sb->sb_tls_info = NULL; 793 #endif 794 sbrelease_internal(so, which); 795 } 796 797 /* 798 * Routines to add and remove data from an mbuf queue. 799 * 800 * The routines sbappend() or sbappendrecord() are normally called to append 801 * new mbufs to a socket buffer, after checking that adequate space is 802 * available, comparing the function sbspace() with the amount of data to be 803 * added. sbappendrecord() differs from sbappend() in that data supplied is 804 * treated as the beginning of a new record. To place a sender's address, 805 * optional access rights, and data in a socket receive buffer, 806 * sbappendaddr() should be used. To place access rights and data in a 807 * socket receive buffer, sbappendrights() should be used. In either case, 808 * the new data begins a new record. Note that unlike sbappend() and 809 * sbappendrecord(), these routines check for the caller that there will be 810 * enough space to store the data. Each fails if there is not enough space, 811 * or if it cannot find mbufs to store additional information in. 812 * 813 * Reliable protocols may use the socket send buffer to hold data awaiting 814 * acknowledgement. Data is normally copied from a socket send buffer in a 815 * protocol with m_copy for output to a peer, and then removing the data from 816 * the socket buffer with sbdrop() or sbdroprecord() when the data is 817 * acknowledged by the peer. 818 */ 819 #ifdef SOCKBUF_DEBUG 820 void 821 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 822 { 823 struct mbuf *m = sb->sb_mb; 824 825 SOCKBUF_LOCK_ASSERT(sb); 826 827 while (m && m->m_nextpkt) 828 m = m->m_nextpkt; 829 830 if (m != sb->sb_lastrecord) { 831 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 832 __func__, sb->sb_mb, sb->sb_lastrecord, m); 833 printf("packet chain:\n"); 834 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 835 printf("\t%p\n", m); 836 panic("%s from %s:%u", __func__, file, line); 837 } 838 } 839 840 void 841 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 842 { 843 struct mbuf *m = sb->sb_mb; 844 struct mbuf *n; 845 846 SOCKBUF_LOCK_ASSERT(sb); 847 848 while (m && m->m_nextpkt) 849 m = m->m_nextpkt; 850 851 while (m && m->m_next) 852 m = m->m_next; 853 854 if (m != sb->sb_mbtail) { 855 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 856 __func__, sb->sb_mb, sb->sb_mbtail, m); 857 printf("packet tree:\n"); 858 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 859 printf("\t"); 860 for (n = m; n != NULL; n = n->m_next) 861 printf("%p ", n); 862 printf("\n"); 863 } 864 panic("%s from %s:%u", __func__, file, line); 865 } 866 867 #ifdef KERN_TLS 868 m = sb->sb_mtls; 869 while (m && m->m_next) 870 m = m->m_next; 871 872 if (m != sb->sb_mtlstail) { 873 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n", 874 __func__, sb->sb_mtls, sb->sb_mtlstail, m); 875 printf("TLS packet tree:\n"); 876 printf("\t"); 877 for (m = sb->sb_mtls; m != NULL; m = m->m_next) { 878 printf("%p ", m); 879 } 880 printf("\n"); 881 panic("%s from %s:%u", __func__, file, line); 882 } 883 #endif 884 } 885 #endif /* SOCKBUF_DEBUG */ 886 887 #define SBLINKRECORD(sb, m0) do { \ 888 SOCKBUF_LOCK_ASSERT(sb); \ 889 if ((sb)->sb_lastrecord != NULL) \ 890 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 891 else \ 892 (sb)->sb_mb = (m0); \ 893 (sb)->sb_lastrecord = (m0); \ 894 } while (/*CONSTCOND*/0) 895 896 /* 897 * Append mbuf chain m to the last record in the socket buffer sb. The 898 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 899 * are discarded and mbufs are compacted where possible. 900 */ 901 void 902 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) 903 { 904 struct mbuf *n; 905 906 SOCKBUF_LOCK_ASSERT(sb); 907 908 if (m == NULL) 909 return; 910 kmsan_check_mbuf(m, "sbappend"); 911 sbm_clrprotoflags(m, flags); 912 SBLASTRECORDCHK(sb); 913 n = sb->sb_mb; 914 if (n) { 915 while (n->m_nextpkt) 916 n = n->m_nextpkt; 917 do { 918 if (n->m_flags & M_EOR) { 919 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 920 return; 921 } 922 } while (n->m_next && (n = n->m_next)); 923 } else { 924 /* 925 * XXX Would like to simply use sb_mbtail here, but 926 * XXX I need to verify that I won't miss an EOR that 927 * XXX way. 928 */ 929 if ((n = sb->sb_lastrecord) != NULL) { 930 do { 931 if (n->m_flags & M_EOR) { 932 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 933 return; 934 } 935 } while (n->m_next && (n = n->m_next)); 936 } else { 937 /* 938 * If this is the first record in the socket buffer, 939 * it's also the last record. 940 */ 941 sb->sb_lastrecord = m; 942 } 943 } 944 sbcompress(sb, m, n); 945 SBLASTRECORDCHK(sb); 946 } 947 948 /* 949 * Append mbuf chain m to the last record in the socket buffer sb. The 950 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 951 * are discarded and mbufs are compacted where possible. 952 */ 953 void 954 sbappend(struct sockbuf *sb, struct mbuf *m, int flags) 955 { 956 957 SOCKBUF_LOCK(sb); 958 sbappend_locked(sb, m, flags); 959 SOCKBUF_UNLOCK(sb); 960 } 961 962 #ifdef KERN_TLS 963 /* 964 * Append an mbuf containing encrypted TLS data. The data 965 * is marked M_NOTREADY until it has been decrypted and 966 * stored as a TLS record. 967 */ 968 static void 969 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m) 970 { 971 struct ifnet *ifp; 972 struct mbuf *n; 973 int flags; 974 975 ifp = NULL; 976 flags = M_NOTREADY; 977 978 SBLASTMBUFCHK(sb); 979 980 /* Mbuf chain must start with a packet header. */ 981 MPASS((m->m_flags & M_PKTHDR) != 0); 982 983 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 984 for (n = m; n != NULL; n = n->m_next) { 985 if (n->m_flags & M_PKTHDR) { 986 ifp = m->m_pkthdr.leaf_rcvif; 987 if ((n->m_pkthdr.csum_flags & CSUM_TLS_MASK) == 988 CSUM_TLS_DECRYPTED) { 989 /* Mark all mbufs in this packet decrypted. */ 990 flags = M_NOTREADY | M_DECRYPTED; 991 } else { 992 flags = M_NOTREADY; 993 } 994 m_demote_pkthdr(n); 995 } 996 997 n->m_flags &= M_DEMOTEFLAGS; 998 n->m_flags |= flags; 999 1000 MPASS((n->m_flags & M_NOTREADY) != 0); 1001 } 1002 1003 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail); 1004 ktls_check_rx(sb); 1005 1006 /* Check for incoming packet route changes: */ 1007 if (ifp != NULL && sb->sb_tls_info->rx_ifp != NULL && 1008 sb->sb_tls_info->rx_ifp != ifp) 1009 ktls_input_ifp_mismatch(sb, ifp); 1010 } 1011 #endif 1012 1013 /* 1014 * This version of sbappend() should only be used when the caller absolutely 1015 * knows that there will never be more than one record in the socket buffer, 1016 * that is, a stream protocol (such as TCP). 1017 */ 1018 void 1019 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) 1020 { 1021 SOCKBUF_LOCK_ASSERT(sb); 1022 1023 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 1024 1025 kmsan_check_mbuf(m, "sbappend"); 1026 1027 #ifdef KERN_TLS 1028 /* 1029 * Decrypted TLS records are appended as records via 1030 * sbappendrecord(). TCP passes encrypted TLS records to this 1031 * function which must be scheduled for decryption. 1032 */ 1033 if (sb->sb_flags & SB_TLS_RX) { 1034 sbappend_ktls_rx(sb, m); 1035 return; 1036 } 1037 #endif 1038 1039 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 1040 1041 SBLASTMBUFCHK(sb); 1042 1043 #ifdef KERN_TLS 1044 if (sb->sb_tls_info != NULL) 1045 ktls_seq(sb, m); 1046 #endif 1047 1048 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 1049 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); 1050 1051 sbcompress(sb, m, sb->sb_mbtail); 1052 1053 sb->sb_lastrecord = sb->sb_mb; 1054 SBLASTRECORDCHK(sb); 1055 } 1056 1057 /* 1058 * This version of sbappend() should only be used when the caller absolutely 1059 * knows that there will never be more than one record in the socket buffer, 1060 * that is, a stream protocol (such as TCP). 1061 */ 1062 void 1063 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) 1064 { 1065 1066 SOCKBUF_LOCK(sb); 1067 sbappendstream_locked(sb, m, flags); 1068 SOCKBUF_UNLOCK(sb); 1069 } 1070 1071 #ifdef SOCKBUF_DEBUG 1072 void 1073 sbcheck(struct sockbuf *sb, const char *file, int line) 1074 { 1075 struct mbuf *m, *n, *fnrdy; 1076 u_long acc, ccc, mbcnt; 1077 #ifdef KERN_TLS 1078 u_long tlscc; 1079 #endif 1080 1081 SOCKBUF_LOCK_ASSERT(sb); 1082 1083 acc = ccc = mbcnt = 0; 1084 fnrdy = NULL; 1085 1086 for (m = sb->sb_mb; m; m = n) { 1087 n = m->m_nextpkt; 1088 for (; m; m = m->m_next) { 1089 if (m->m_len == 0) { 1090 printf("sb %p empty mbuf %p\n", sb, m); 1091 goto fail; 1092 } 1093 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { 1094 if (m != sb->sb_fnrdy) { 1095 printf("sb %p: fnrdy %p != m %p\n", 1096 sb, sb->sb_fnrdy, m); 1097 goto fail; 1098 } 1099 fnrdy = m; 1100 } 1101 if (fnrdy) { 1102 if (!(m->m_flags & M_NOTAVAIL)) { 1103 printf("sb %p: fnrdy %p, m %p is avail\n", 1104 sb, sb->sb_fnrdy, m); 1105 goto fail; 1106 } 1107 } else 1108 acc += m->m_len; 1109 ccc += m->m_len; 1110 mbcnt += MSIZE; 1111 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1112 mbcnt += m->m_ext.ext_size; 1113 } 1114 } 1115 #ifdef KERN_TLS 1116 /* 1117 * Account for mbufs "detached" by ktls_detach_record() while 1118 * they are decrypted by ktls_decrypt(). tlsdcc gives a count 1119 * of the detached bytes that are included in ccc. The mbufs 1120 * and clusters are not included in the socket buffer 1121 * accounting. 1122 */ 1123 ccc += sb->sb_tlsdcc; 1124 1125 tlscc = 0; 1126 for (m = sb->sb_mtls; m; m = m->m_next) { 1127 if (m->m_nextpkt != NULL) { 1128 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m); 1129 goto fail; 1130 } 1131 if ((m->m_flags & M_NOTREADY) == 0) { 1132 printf("sb %p TLS mbuf %p ready\n", sb, m); 1133 goto fail; 1134 } 1135 tlscc += m->m_len; 1136 ccc += m->m_len; 1137 mbcnt += MSIZE; 1138 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1139 mbcnt += m->m_ext.ext_size; 1140 } 1141 1142 if (sb->sb_tlscc != tlscc) { 1143 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1144 sb->sb_tlsdcc); 1145 goto fail; 1146 } 1147 #endif 1148 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { 1149 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", 1150 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); 1151 #ifdef KERN_TLS 1152 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1153 sb->sb_tlsdcc); 1154 #endif 1155 goto fail; 1156 } 1157 return; 1158 fail: 1159 panic("%s from %s:%u", __func__, file, line); 1160 } 1161 #endif 1162 1163 /* 1164 * As above, except the mbuf chain begins a new record. 1165 */ 1166 void 1167 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 1168 { 1169 struct mbuf *m; 1170 1171 SOCKBUF_LOCK_ASSERT(sb); 1172 1173 if (m0 == NULL) 1174 return; 1175 1176 kmsan_check_mbuf(m0, "sbappend"); 1177 m_clrprotoflags(m0); 1178 1179 /* 1180 * Put the first mbuf on the queue. Note this permits zero length 1181 * records. 1182 */ 1183 sballoc(sb, m0); 1184 SBLASTRECORDCHK(sb); 1185 SBLINKRECORD(sb, m0); 1186 sb->sb_mbtail = m0; 1187 m = m0->m_next; 1188 m0->m_next = 0; 1189 if (m && (m0->m_flags & M_EOR)) { 1190 m0->m_flags &= ~M_EOR; 1191 m->m_flags |= M_EOR; 1192 } 1193 /* always call sbcompress() so it can do SBLASTMBUFCHK() */ 1194 sbcompress(sb, m, m0); 1195 } 1196 1197 /* 1198 * As above, except the mbuf chain begins a new record. 1199 */ 1200 void 1201 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 1202 { 1203 1204 SOCKBUF_LOCK(sb); 1205 sbappendrecord_locked(sb, m0); 1206 SOCKBUF_UNLOCK(sb); 1207 } 1208 1209 /* Helper routine that appends data, control, and address to a sockbuf. */ 1210 static int 1211 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, 1212 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) 1213 { 1214 struct mbuf *m, *n, *nlast; 1215 1216 if (m0 != NULL) 1217 kmsan_check_mbuf(m0, "sbappend"); 1218 if (control != NULL) 1219 kmsan_check_mbuf(control, "sbappend"); 1220 1221 #if MSIZE <= 256 1222 if (asa->sa_len > MLEN) 1223 return (0); 1224 #endif 1225 m = m_get(M_NOWAIT, MT_SONAME); 1226 if (m == NULL) 1227 return (0); 1228 m->m_len = asa->sa_len; 1229 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 1230 if (m0) { 1231 M_ASSERT_NO_SND_TAG(m0); 1232 m_clrprotoflags(m0); 1233 m_tag_delete_chain(m0, NULL); 1234 /* 1235 * Clear some persistent info from pkthdr. 1236 * We don't use m_demote(), because some netgraph consumers 1237 * expect M_PKTHDR presence. 1238 */ 1239 m0->m_pkthdr.rcvif = NULL; 1240 m0->m_pkthdr.flowid = 0; 1241 m0->m_pkthdr.csum_flags = 0; 1242 m0->m_pkthdr.fibnum = 0; 1243 m0->m_pkthdr.rsstype = 0; 1244 } 1245 if (ctrl_last) 1246 ctrl_last->m_next = m0; /* concatenate data to control */ 1247 else 1248 control = m0; 1249 m->m_next = control; 1250 for (n = m; n->m_next != NULL; n = n->m_next) 1251 sballoc(sb, n); 1252 sballoc(sb, n); 1253 nlast = n; 1254 SBLINKRECORD(sb, m); 1255 1256 sb->sb_mbtail = nlast; 1257 SBLASTMBUFCHK(sb); 1258 1259 SBLASTRECORDCHK(sb); 1260 return (1); 1261 } 1262 1263 /* 1264 * Append address and data, and optionally, control (ancillary) data to the 1265 * receive queue of a socket. If present, m0 must include a packet header 1266 * with total length. Returns 0 if no space in sockbuf or insufficient 1267 * mbufs. 1268 */ 1269 int 1270 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 1271 struct mbuf *m0, struct mbuf *control) 1272 { 1273 struct mbuf *ctrl_last; 1274 int space = asa->sa_len; 1275 1276 SOCKBUF_LOCK_ASSERT(sb); 1277 1278 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1279 panic("sbappendaddr_locked"); 1280 if (m0) 1281 space += m0->m_pkthdr.len; 1282 space += m_length(control, &ctrl_last); 1283 1284 if (space > sbspace(sb)) 1285 return (0); 1286 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1287 } 1288 1289 /* 1290 * Append address and data, and optionally, control (ancillary) data to the 1291 * receive queue of a socket. If present, m0 must include a packet header 1292 * with total length. Returns 0 if insufficient mbufs. Does not validate space 1293 * on the receiving sockbuf. 1294 */ 1295 int 1296 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, 1297 struct mbuf *m0, struct mbuf *control) 1298 { 1299 struct mbuf *ctrl_last; 1300 1301 SOCKBUF_LOCK_ASSERT(sb); 1302 1303 ctrl_last = (control == NULL) ? NULL : m_last(control); 1304 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1305 } 1306 1307 /* 1308 * Append address and data, and optionally, control (ancillary) data to the 1309 * receive queue of a socket. If present, m0 must include a packet header 1310 * with total length. Returns 0 if no space in sockbuf or insufficient 1311 * mbufs. 1312 */ 1313 int 1314 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 1315 struct mbuf *m0, struct mbuf *control) 1316 { 1317 int retval; 1318 1319 SOCKBUF_LOCK(sb); 1320 retval = sbappendaddr_locked(sb, asa, m0, control); 1321 SOCKBUF_UNLOCK(sb); 1322 return (retval); 1323 } 1324 1325 void 1326 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 1327 struct mbuf *control, int flags) 1328 { 1329 struct mbuf *m, *mlast; 1330 1331 kmsan_check_mbuf(m0, "sbappend"); 1332 kmsan_check_mbuf(control, "sbappend"); 1333 1334 sbm_clrprotoflags(m0, flags); 1335 m_last(control)->m_next = m0; 1336 1337 SBLASTRECORDCHK(sb); 1338 1339 for (m = control; m->m_next; m = m->m_next) 1340 sballoc(sb, m); 1341 sballoc(sb, m); 1342 mlast = m; 1343 SBLINKRECORD(sb, control); 1344 1345 sb->sb_mbtail = mlast; 1346 SBLASTMBUFCHK(sb); 1347 1348 SBLASTRECORDCHK(sb); 1349 } 1350 1351 void 1352 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control, 1353 int flags) 1354 { 1355 1356 SOCKBUF_LOCK(sb); 1357 sbappendcontrol_locked(sb, m0, control, flags); 1358 SOCKBUF_UNLOCK(sb); 1359 } 1360 1361 /* 1362 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 1363 * (n). If (n) is NULL, the buffer is presumed empty. 1364 * 1365 * When the data is compressed, mbufs in the chain may be handled in one of 1366 * three ways: 1367 * 1368 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 1369 * record boundary, and no change in data type). 1370 * 1371 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 1372 * an mbuf already in the socket buffer. This can occur if an 1373 * appropriate mbuf exists, there is room, both mbufs are not marked as 1374 * not ready, and no merging of data types will occur. 1375 * 1376 * (3) The mbuf may be appended to the end of the existing mbuf chain. 1377 * 1378 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 1379 * end-of-record. 1380 */ 1381 void 1382 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1383 { 1384 int eor = 0; 1385 struct mbuf *o; 1386 1387 SOCKBUF_LOCK_ASSERT(sb); 1388 1389 while (m) { 1390 eor |= m->m_flags & M_EOR; 1391 if (m->m_len == 0 && 1392 (eor == 0 || 1393 (((o = m->m_next) || (o = n)) && 1394 o->m_type == m->m_type))) { 1395 if (sb->sb_lastrecord == m) 1396 sb->sb_lastrecord = m->m_next; 1397 m = m_free(m); 1398 continue; 1399 } 1400 if (n && (n->m_flags & M_EOR) == 0 && 1401 M_WRITABLE(n) && 1402 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1403 !(m->m_flags & M_NOTREADY) && 1404 !(n->m_flags & (M_NOTREADY | M_EXTPG)) && 1405 !mbuf_has_tls_session(m) && 1406 !mbuf_has_tls_session(n) && 1407 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1408 m->m_len <= M_TRAILINGSPACE(n) && 1409 n->m_type == m->m_type) { 1410 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1411 n->m_len += m->m_len; 1412 sb->sb_ccc += m->m_len; 1413 if (sb->sb_fnrdy == NULL) 1414 sb->sb_acc += m->m_len; 1415 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1416 /* XXX: Probably don't need.*/ 1417 sb->sb_ctl += m->m_len; 1418 m = m_free(m); 1419 continue; 1420 } 1421 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) && 1422 (m->m_flags & M_NOTREADY) == 0 && 1423 !mbuf_has_tls_session(m)) 1424 (void)mb_unmapped_compress(m); 1425 if (n) 1426 n->m_next = m; 1427 else 1428 sb->sb_mb = m; 1429 sb->sb_mbtail = m; 1430 sballoc(sb, m); 1431 n = m; 1432 m->m_flags &= ~M_EOR; 1433 m = m->m_next; 1434 n->m_next = 0; 1435 } 1436 if (eor) { 1437 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 1438 n->m_flags |= eor; 1439 } 1440 SBLASTMBUFCHK(sb); 1441 } 1442 1443 #ifdef KERN_TLS 1444 /* 1445 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs 1446 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also 1447 * a bit simpler (no EOR markers, always MT_DATA, etc.). 1448 */ 1449 static void 1450 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1451 { 1452 1453 SOCKBUF_LOCK_ASSERT(sb); 1454 1455 while (m) { 1456 KASSERT((m->m_flags & M_EOR) == 0, 1457 ("TLS RX mbuf %p with EOR", m)); 1458 KASSERT(m->m_type == MT_DATA, 1459 ("TLS RX mbuf %p is not MT_DATA", m)); 1460 KASSERT((m->m_flags & M_NOTREADY) != 0, 1461 ("TLS RX mbuf %p ready", m)); 1462 KASSERT((m->m_flags & M_EXTPG) == 0, 1463 ("TLS RX mbuf %p unmapped", m)); 1464 1465 if (m->m_len == 0) { 1466 m = m_free(m); 1467 continue; 1468 } 1469 1470 /* 1471 * Even though both 'n' and 'm' are NOTREADY, it's ok 1472 * to coalesce the data. 1473 */ 1474 if (n && 1475 M_WRITABLE(n) && 1476 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1477 !((m->m_flags ^ n->m_flags) & M_DECRYPTED) && 1478 !(n->m_flags & M_EXTPG) && 1479 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1480 m->m_len <= M_TRAILINGSPACE(n)) { 1481 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1482 n->m_len += m->m_len; 1483 sb->sb_ccc += m->m_len; 1484 sb->sb_tlscc += m->m_len; 1485 m = m_free(m); 1486 continue; 1487 } 1488 if (n) 1489 n->m_next = m; 1490 else 1491 sb->sb_mtls = m; 1492 sb->sb_mtlstail = m; 1493 sballoc_ktls_rx(sb, m); 1494 n = m; 1495 m = m->m_next; 1496 n->m_next = NULL; 1497 } 1498 SBLASTMBUFCHK(sb); 1499 } 1500 #endif 1501 1502 /* 1503 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 1504 */ 1505 static void 1506 sbflush_internal(struct sockbuf *sb) 1507 { 1508 1509 while (sb->sb_mbcnt || sb->sb_tlsdcc) { 1510 /* 1511 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: 1512 * we would loop forever. Panic instead. 1513 */ 1514 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1515 break; 1516 m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); 1517 } 1518 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 1519 ("%s: ccc %u mb %p mbcnt %u", __func__, 1520 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 1521 } 1522 1523 void 1524 sbflush_locked(struct sockbuf *sb) 1525 { 1526 1527 SOCKBUF_LOCK_ASSERT(sb); 1528 sbflush_internal(sb); 1529 } 1530 1531 void 1532 sbflush(struct sockbuf *sb) 1533 { 1534 1535 SOCKBUF_LOCK(sb); 1536 sbflush_locked(sb); 1537 SOCKBUF_UNLOCK(sb); 1538 } 1539 1540 /* 1541 * Cut data from (the front of) a sockbuf. 1542 */ 1543 static struct mbuf * 1544 sbcut_internal(struct sockbuf *sb, int len) 1545 { 1546 struct mbuf *m, *next, *mfree; 1547 bool is_tls; 1548 1549 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0", 1550 __func__, len)); 1551 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u", 1552 __func__, len, sb->sb_ccc)); 1553 1554 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1555 is_tls = false; 1556 mfree = NULL; 1557 1558 while (len > 0) { 1559 if (m == NULL) { 1560 #ifdef KERN_TLS 1561 if (next == NULL && !is_tls) { 1562 if (sb->sb_tlsdcc != 0) { 1563 MPASS(len >= sb->sb_tlsdcc); 1564 len -= sb->sb_tlsdcc; 1565 sb->sb_ccc -= sb->sb_tlsdcc; 1566 sb->sb_tlsdcc = 0; 1567 if (len == 0) 1568 break; 1569 } 1570 next = sb->sb_mtls; 1571 is_tls = true; 1572 } 1573 #endif 1574 KASSERT(next, ("%s: no next, len %d", __func__, len)); 1575 m = next; 1576 next = m->m_nextpkt; 1577 } 1578 if (m->m_len > len) { 1579 KASSERT(!(m->m_flags & M_NOTAVAIL), 1580 ("%s: m %p M_NOTAVAIL", __func__, m)); 1581 m->m_len -= len; 1582 m->m_data += len; 1583 sb->sb_ccc -= len; 1584 sb->sb_acc -= len; 1585 if (sb->sb_sndptroff != 0) 1586 sb->sb_sndptroff -= len; 1587 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1588 sb->sb_ctl -= len; 1589 break; 1590 } 1591 len -= m->m_len; 1592 #ifdef KERN_TLS 1593 if (is_tls) 1594 sbfree_ktls_rx(sb, m); 1595 else 1596 #endif 1597 sbfree(sb, m); 1598 /* 1599 * Do not put M_NOTREADY buffers to the free list, they 1600 * are referenced from outside. 1601 */ 1602 if (m->m_flags & M_NOTREADY && !is_tls) 1603 m = m->m_next; 1604 else { 1605 struct mbuf *n; 1606 1607 n = m->m_next; 1608 m->m_next = mfree; 1609 mfree = m; 1610 m = n; 1611 } 1612 } 1613 /* 1614 * Free any zero-length mbufs from the buffer. 1615 * For SOCK_DGRAM sockets such mbufs represent empty records. 1616 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, 1617 * when sosend_generic() needs to send only control data. 1618 */ 1619 while (m && m->m_len == 0) { 1620 struct mbuf *n; 1621 1622 sbfree(sb, m); 1623 n = m->m_next; 1624 m->m_next = mfree; 1625 mfree = m; 1626 m = n; 1627 } 1628 #ifdef KERN_TLS 1629 if (is_tls) { 1630 sb->sb_mb = NULL; 1631 sb->sb_mtls = m; 1632 if (m == NULL) 1633 sb->sb_mtlstail = NULL; 1634 } else 1635 #endif 1636 if (m) { 1637 sb->sb_mb = m; 1638 m->m_nextpkt = next; 1639 } else 1640 sb->sb_mb = next; 1641 /* 1642 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 1643 * sb_lastrecord is up-to-date if we dropped part of the last record. 1644 */ 1645 m = sb->sb_mb; 1646 if (m == NULL) { 1647 sb->sb_mbtail = NULL; 1648 sb->sb_lastrecord = NULL; 1649 } else if (m->m_nextpkt == NULL) { 1650 sb->sb_lastrecord = m; 1651 } 1652 1653 return (mfree); 1654 } 1655 1656 /* 1657 * Drop data from (the front of) a sockbuf. 1658 */ 1659 void 1660 sbdrop_locked(struct sockbuf *sb, int len) 1661 { 1662 1663 SOCKBUF_LOCK_ASSERT(sb); 1664 m_freem(sbcut_internal(sb, len)); 1665 } 1666 1667 /* 1668 * Drop data from (the front of) a sockbuf, 1669 * and return it to caller. 1670 */ 1671 struct mbuf * 1672 sbcut_locked(struct sockbuf *sb, int len) 1673 { 1674 1675 SOCKBUF_LOCK_ASSERT(sb); 1676 return (sbcut_internal(sb, len)); 1677 } 1678 1679 void 1680 sbdrop(struct sockbuf *sb, int len) 1681 { 1682 struct mbuf *mfree; 1683 1684 SOCKBUF_LOCK(sb); 1685 mfree = sbcut_internal(sb, len); 1686 SOCKBUF_UNLOCK(sb); 1687 1688 m_freem(mfree); 1689 } 1690 1691 struct mbuf * 1692 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff) 1693 { 1694 struct mbuf *m; 1695 1696 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1697 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1698 *moff = off; 1699 if (sb->sb_sndptr == NULL) { 1700 sb->sb_sndptr = sb->sb_mb; 1701 sb->sb_sndptroff = 0; 1702 } 1703 return (sb->sb_mb); 1704 } else { 1705 m = sb->sb_sndptr; 1706 off -= sb->sb_sndptroff; 1707 } 1708 *moff = off; 1709 return (m); 1710 } 1711 1712 void 1713 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len) 1714 { 1715 /* 1716 * A small copy was done, advance forward the sb_sbsndptr to cover 1717 * it. 1718 */ 1719 struct mbuf *m; 1720 1721 if (mb != sb->sb_sndptr) { 1722 /* Did not copyout at the same mbuf */ 1723 return; 1724 } 1725 m = mb; 1726 while (m && (len > 0)) { 1727 if (len >= m->m_len) { 1728 len -= m->m_len; 1729 if (m->m_next) { 1730 sb->sb_sndptroff += m->m_len; 1731 sb->sb_sndptr = m->m_next; 1732 } 1733 m = m->m_next; 1734 } else { 1735 len = 0; 1736 } 1737 } 1738 } 1739 1740 /* 1741 * Return the first mbuf and the mbuf data offset for the provided 1742 * send offset without changing the "sb_sndptroff" field. 1743 */ 1744 struct mbuf * 1745 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) 1746 { 1747 struct mbuf *m; 1748 1749 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1750 1751 /* 1752 * If the "off" is below the stored offset, which happens on 1753 * retransmits, just use "sb_mb": 1754 */ 1755 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1756 m = sb->sb_mb; 1757 } else { 1758 m = sb->sb_sndptr; 1759 off -= sb->sb_sndptroff; 1760 } 1761 while (off > 0 && m != NULL) { 1762 if (off < m->m_len) 1763 break; 1764 off -= m->m_len; 1765 m = m->m_next; 1766 } 1767 *moff = off; 1768 return (m); 1769 } 1770 1771 /* 1772 * Drop a record off the front of a sockbuf and move the next record to the 1773 * front. 1774 */ 1775 void 1776 sbdroprecord_locked(struct sockbuf *sb) 1777 { 1778 struct mbuf *m; 1779 1780 SOCKBUF_LOCK_ASSERT(sb); 1781 1782 m = sb->sb_mb; 1783 if (m) { 1784 sb->sb_mb = m->m_nextpkt; 1785 do { 1786 sbfree(sb, m); 1787 m = m_free(m); 1788 } while (m); 1789 } 1790 SB_EMPTY_FIXUP(sb); 1791 } 1792 1793 /* 1794 * Drop a record off the front of a sockbuf and move the next record to the 1795 * front. 1796 */ 1797 void 1798 sbdroprecord(struct sockbuf *sb) 1799 { 1800 1801 SOCKBUF_LOCK(sb); 1802 sbdroprecord_locked(sb); 1803 SOCKBUF_UNLOCK(sb); 1804 } 1805 1806 /* 1807 * Create a "control" mbuf containing the specified data with the specified 1808 * type for presentation on a socket buffer. 1809 */ 1810 struct mbuf * 1811 sbcreatecontrol(const void *p, u_int size, int type, int level, int wait) 1812 { 1813 struct cmsghdr *cp; 1814 struct mbuf *m; 1815 1816 MBUF_CHECKSLEEP(wait); 1817 1818 if (wait == M_NOWAIT) { 1819 if (CMSG_SPACE(size) > MCLBYTES) 1820 return (NULL); 1821 } else 1822 KASSERT(CMSG_SPACE(size) <= MCLBYTES, 1823 ("%s: passed CMSG_SPACE(%u) > MCLBYTES", __func__, size)); 1824 1825 if (CMSG_SPACE(size) > MLEN) 1826 m = m_getcl(wait, MT_CONTROL, 0); 1827 else 1828 m = m_get(wait, MT_CONTROL); 1829 if (m == NULL) 1830 return (NULL); 1831 1832 KASSERT(CMSG_SPACE(size) <= M_TRAILINGSPACE(m), 1833 ("sbcreatecontrol: short mbuf")); 1834 /* 1835 * Don't leave the padding between the msg header and the 1836 * cmsg data and the padding after the cmsg data un-initialized. 1837 */ 1838 cp = mtod(m, struct cmsghdr *); 1839 bzero(cp, CMSG_SPACE(size)); 1840 if (p != NULL) 1841 (void)memcpy(CMSG_DATA(cp), p, size); 1842 m->m_len = CMSG_SPACE(size); 1843 cp->cmsg_len = CMSG_LEN(size); 1844 cp->cmsg_level = level; 1845 cp->cmsg_type = type; 1846 return (m); 1847 } 1848 1849 /* 1850 * This does the same for socket buffers that sotoxsocket does for sockets: 1851 * generate an user-format data structure describing the socket buffer. Note 1852 * that the xsockbuf structure, since it is always embedded in a socket, does 1853 * not include a self pointer nor a length. We make this entry point public 1854 * in case some other mechanism needs it. 1855 */ 1856 void 1857 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1858 { 1859 1860 xsb->sb_cc = sb->sb_ccc; 1861 xsb->sb_hiwat = sb->sb_hiwat; 1862 xsb->sb_mbcnt = sb->sb_mbcnt; 1863 xsb->sb_mbmax = sb->sb_mbmax; 1864 xsb->sb_lowat = sb->sb_lowat; 1865 xsb->sb_flags = sb->sb_flags; 1866 xsb->sb_timeo = sb->sb_timeo; 1867 } 1868 1869 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1870 static int dummy; 1871 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, ""); 1872 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, 1873 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0, 1874 sysctl_handle_sb_max, "LU", 1875 "Maximum socket buffer size"); 1876 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1877 &sb_efficiency, 0, "Socket buffer size waste factor"); 1878