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