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