Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_sockbuf.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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 /* 479 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 480 * via SIGIO if the socket has the SS_ASYNC flag set. 481 * 482 * Called with the socket buffer lock held; will release the lock by the end 483 * of the function. This allows the caller to acquire the socket buffer lock 484 * while testing for the need for various sorts of wakeup and hold it through 485 * to the point where it's no longer required. We currently hold the lock 486 * through calls out to other subsystems (with the exception of kqueue), and 487 * then release it to avoid lock order issues. It's not clear that's 488 * correct. 489 */ 490 void 491 sowakeup(struct socket *so, struct sockbuf *sb) 492 { 493 int ret; 494 495 SOCKBUF_LOCK_ASSERT(sb); 496 497 selwakeuppri(sb->sb_sel, PSOCK); 498 if (!SEL_WAITING(sb->sb_sel)) 499 sb->sb_flags &= ~SB_SEL; 500 if (sb->sb_flags & SB_WAIT) { 501 sb->sb_flags &= ~SB_WAIT; 502 wakeup(&sb->sb_acc); 503 } 504 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 505 if (sb->sb_upcall != NULL) { 506 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT); 507 if (ret == SU_ISCONNECTED) { 508 KASSERT(sb == &so->so_rcv, 509 ("SO_SND upcall returned SU_ISCONNECTED")); 510 soupcall_clear(so, SO_RCV); 511 } 512 } else 513 ret = SU_OK; 514 if (sb->sb_flags & SB_AIO) 515 sowakeup_aio(so, sb); 516 SOCKBUF_UNLOCK(sb); 517 if (ret == SU_ISCONNECTED) 518 soisconnected(so); 519 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 520 pgsigio(&so->so_sigio, SIGIO, 0); 521 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED); 522 } 523 524 /* 525 * Socket buffer (struct sockbuf) utility routines. 526 * 527 * Each socket contains two socket buffers: one for sending data and one for 528 * receiving data. Each buffer contains a queue of mbufs, information about 529 * the number of mbufs and amount of data in the queue, and other fields 530 * allowing select() statements and notification on data availability to be 531 * implemented. 532 * 533 * Data stored in a socket buffer is maintained as a list of records. Each 534 * record is a list of mbufs chained together with the m_next field. Records 535 * are chained together with the m_nextpkt field. The upper level routine 536 * soreceive() expects the following conventions to be observed when placing 537 * information in the receive buffer: 538 * 539 * 1. If the protocol requires each message be preceded by the sender's name, 540 * then a record containing that name must be present before any 541 * associated data (mbuf's must be of type MT_SONAME). 542 * 2. If the protocol supports the exchange of ``access rights'' (really just 543 * additional data associated with the message), and there are ``rights'' 544 * to be received, then a record containing this data should be present 545 * (mbuf's must be of type MT_RIGHTS). 546 * 3. If a name or rights record exists, then it must be followed by a data 547 * record, perhaps of zero length. 548 * 549 * Before using a new socket structure it is first necessary to reserve 550 * buffer space to the socket, by calling sbreserve(). This should commit 551 * some of the available buffer space in the system buffer pool for the 552 * socket (currently, it does nothing but enforce limits). The space should 553 * be released by calling sbrelease() when the socket is destroyed. 554 */ 555 int 556 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 557 { 558 struct thread *td = curthread; 559 560 SOCKBUF_LOCK(&so->so_snd); 561 SOCKBUF_LOCK(&so->so_rcv); 562 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0) 563 goto bad; 564 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0) 565 goto bad2; 566 if (so->so_rcv.sb_lowat == 0) 567 so->so_rcv.sb_lowat = 1; 568 if (so->so_snd.sb_lowat == 0) 569 so->so_snd.sb_lowat = MCLBYTES; 570 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 571 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 572 SOCKBUF_UNLOCK(&so->so_rcv); 573 SOCKBUF_UNLOCK(&so->so_snd); 574 return (0); 575 bad2: 576 sbrelease_locked(&so->so_snd, so); 577 bad: 578 SOCKBUF_UNLOCK(&so->so_rcv); 579 SOCKBUF_UNLOCK(&so->so_snd); 580 return (ENOBUFS); 581 } 582 583 static int 584 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 585 { 586 int error = 0; 587 u_long tmp_sb_max = sb_max; 588 589 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req); 590 if (error || !req->newptr) 591 return (error); 592 if (tmp_sb_max < MSIZE + MCLBYTES) 593 return (EINVAL); 594 sb_max = tmp_sb_max; 595 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 596 return (0); 597 } 598 599 /* 600 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't 601 * become limiting if buffering efficiency is near the normal case. 602 */ 603 int 604 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so, 605 struct thread *td) 606 { 607 rlim_t sbsize_limit; 608 609 SOCKBUF_LOCK_ASSERT(sb); 610 611 /* 612 * When a thread is passed, we take into account the thread's socket 613 * buffer size limit. The caller will generally pass curthread, but 614 * in the TCP input path, NULL will be passed to indicate that no 615 * appropriate thread resource limits are available. In that case, 616 * we don't apply a process limit. 617 */ 618 if (cc > sb_max_adj) 619 return (0); 620 if (td != NULL) { 621 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE); 622 } else 623 sbsize_limit = RLIM_INFINITY; 624 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, 625 sbsize_limit)) 626 return (0); 627 sb->sb_mbmax = min(cc * sb_efficiency, sb_max); 628 if (sb->sb_lowat > sb->sb_hiwat) 629 sb->sb_lowat = sb->sb_hiwat; 630 return (1); 631 } 632 633 int 634 sbsetopt(struct socket *so, int cmd, u_long cc) 635 { 636 struct sockbuf *sb; 637 short *flags; 638 u_int *hiwat, *lowat; 639 int error; 640 641 sb = NULL; 642 SOCK_LOCK(so); 643 if (SOLISTENING(so)) { 644 switch (cmd) { 645 case SO_SNDLOWAT: 646 case SO_SNDBUF: 647 lowat = &so->sol_sbsnd_lowat; 648 hiwat = &so->sol_sbsnd_hiwat; 649 flags = &so->sol_sbsnd_flags; 650 break; 651 case SO_RCVLOWAT: 652 case SO_RCVBUF: 653 lowat = &so->sol_sbrcv_lowat; 654 hiwat = &so->sol_sbrcv_hiwat; 655 flags = &so->sol_sbrcv_flags; 656 break; 657 } 658 } else { 659 switch (cmd) { 660 case SO_SNDLOWAT: 661 case SO_SNDBUF: 662 sb = &so->so_snd; 663 break; 664 case SO_RCVLOWAT: 665 case SO_RCVBUF: 666 sb = &so->so_rcv; 667 break; 668 } 669 flags = &sb->sb_flags; 670 hiwat = &sb->sb_hiwat; 671 lowat = &sb->sb_lowat; 672 SOCKBUF_LOCK(sb); 673 } 674 675 error = 0; 676 switch (cmd) { 677 case SO_SNDBUF: 678 case SO_RCVBUF: 679 if (SOLISTENING(so)) { 680 if (cc > sb_max_adj) { 681 error = ENOBUFS; 682 break; 683 } 684 *hiwat = cc; 685 if (*lowat > *hiwat) 686 *lowat = *hiwat; 687 } else { 688 if (!sbreserve_locked(sb, cc, so, curthread)) 689 error = ENOBUFS; 690 } 691 if (error == 0) 692 *flags &= ~SB_AUTOSIZE; 693 break; 694 case SO_SNDLOWAT: 695 case SO_RCVLOWAT: 696 /* 697 * Make sure the low-water is never greater than the 698 * high-water. 699 */ 700 *lowat = (cc > *hiwat) ? *hiwat : cc; 701 break; 702 } 703 704 if (!SOLISTENING(so)) 705 SOCKBUF_UNLOCK(sb); 706 SOCK_UNLOCK(so); 707 return (error); 708 } 709 710 /* 711 * Free mbufs held by a socket, and reserved mbuf space. 712 */ 713 void 714 sbrelease_internal(struct sockbuf *sb, struct socket *so) 715 { 716 717 sbflush_internal(sb); 718 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 719 RLIM_INFINITY); 720 sb->sb_mbmax = 0; 721 } 722 723 void 724 sbrelease_locked(struct sockbuf *sb, struct socket *so) 725 { 726 727 SOCKBUF_LOCK_ASSERT(sb); 728 729 sbrelease_internal(sb, so); 730 } 731 732 void 733 sbrelease(struct sockbuf *sb, struct socket *so) 734 { 735 736 SOCKBUF_LOCK(sb); 737 sbrelease_locked(sb, so); 738 SOCKBUF_UNLOCK(sb); 739 } 740 741 void 742 sbdestroy(struct sockbuf *sb, struct socket *so) 743 { 744 745 sbrelease_internal(sb, so); 746 #ifdef KERN_TLS 747 if (sb->sb_tls_info != NULL) 748 ktls_free(sb->sb_tls_info); 749 sb->sb_tls_info = NULL; 750 #endif 751 } 752 753 /* 754 * Routines to add and remove data from an mbuf queue. 755 * 756 * The routines sbappend() or sbappendrecord() are normally called to append 757 * new mbufs to a socket buffer, after checking that adequate space is 758 * available, comparing the function sbspace() with the amount of data to be 759 * added. sbappendrecord() differs from sbappend() in that data supplied is 760 * treated as the beginning of a new record. To place a sender's address, 761 * optional access rights, and data in a socket receive buffer, 762 * sbappendaddr() should be used. To place access rights and data in a 763 * socket receive buffer, sbappendrights() should be used. In either case, 764 * the new data begins a new record. Note that unlike sbappend() and 765 * sbappendrecord(), these routines check for the caller that there will be 766 * enough space to store the data. Each fails if there is not enough space, 767 * or if it cannot find mbufs to store additional information in. 768 * 769 * Reliable protocols may use the socket send buffer to hold data awaiting 770 * acknowledgement. Data is normally copied from a socket send buffer in a 771 * protocol with m_copy for output to a peer, and then removing the data from 772 * the socket buffer with sbdrop() or sbdroprecord() when the data is 773 * acknowledged by the peer. 774 */ 775 #ifdef SOCKBUF_DEBUG 776 void 777 sblastrecordchk(struct sockbuf *sb, const char *file, int line) 778 { 779 struct mbuf *m = sb->sb_mb; 780 781 SOCKBUF_LOCK_ASSERT(sb); 782 783 while (m && m->m_nextpkt) 784 m = m->m_nextpkt; 785 786 if (m != sb->sb_lastrecord) { 787 printf("%s: sb_mb %p sb_lastrecord %p last %p\n", 788 __func__, sb->sb_mb, sb->sb_lastrecord, m); 789 printf("packet chain:\n"); 790 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 791 printf("\t%p\n", m); 792 panic("%s from %s:%u", __func__, file, line); 793 } 794 } 795 796 void 797 sblastmbufchk(struct sockbuf *sb, const char *file, int line) 798 { 799 struct mbuf *m = sb->sb_mb; 800 struct mbuf *n; 801 802 SOCKBUF_LOCK_ASSERT(sb); 803 804 while (m && m->m_nextpkt) 805 m = m->m_nextpkt; 806 807 while (m && m->m_next) 808 m = m->m_next; 809 810 if (m != sb->sb_mbtail) { 811 printf("%s: sb_mb %p sb_mbtail %p last %p\n", 812 __func__, sb->sb_mb, sb->sb_mbtail, m); 813 printf("packet tree:\n"); 814 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 815 printf("\t"); 816 for (n = m; n != NULL; n = n->m_next) 817 printf("%p ", n); 818 printf("\n"); 819 } 820 panic("%s from %s:%u", __func__, file, line); 821 } 822 823 #ifdef KERN_TLS 824 m = sb->sb_mtls; 825 while (m && m->m_next) 826 m = m->m_next; 827 828 if (m != sb->sb_mtlstail) { 829 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n", 830 __func__, sb->sb_mtls, sb->sb_mtlstail, m); 831 printf("TLS packet tree:\n"); 832 printf("\t"); 833 for (m = sb->sb_mtls; m != NULL; m = m->m_next) { 834 printf("%p ", m); 835 } 836 printf("\n"); 837 panic("%s from %s:%u", __func__, file, line); 838 } 839 #endif 840 } 841 #endif /* SOCKBUF_DEBUG */ 842 843 #define SBLINKRECORD(sb, m0) do { \ 844 SOCKBUF_LOCK_ASSERT(sb); \ 845 if ((sb)->sb_lastrecord != NULL) \ 846 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 847 else \ 848 (sb)->sb_mb = (m0); \ 849 (sb)->sb_lastrecord = (m0); \ 850 } while (/*CONSTCOND*/0) 851 852 /* 853 * Append mbuf chain m to the last record in the socket buffer sb. The 854 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 855 * are discarded and mbufs are compacted where possible. 856 */ 857 void 858 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags) 859 { 860 struct mbuf *n; 861 862 SOCKBUF_LOCK_ASSERT(sb); 863 864 if (m == NULL) 865 return; 866 sbm_clrprotoflags(m, flags); 867 SBLASTRECORDCHK(sb); 868 n = sb->sb_mb; 869 if (n) { 870 while (n->m_nextpkt) 871 n = n->m_nextpkt; 872 do { 873 if (n->m_flags & M_EOR) { 874 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 875 return; 876 } 877 } while (n->m_next && (n = n->m_next)); 878 } else { 879 /* 880 * XXX Would like to simply use sb_mbtail here, but 881 * XXX I need to verify that I won't miss an EOR that 882 * XXX way. 883 */ 884 if ((n = sb->sb_lastrecord) != NULL) { 885 do { 886 if (n->m_flags & M_EOR) { 887 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ 888 return; 889 } 890 } while (n->m_next && (n = n->m_next)); 891 } else { 892 /* 893 * If this is the first record in the socket buffer, 894 * it's also the last record. 895 */ 896 sb->sb_lastrecord = m; 897 } 898 } 899 sbcompress(sb, m, n); 900 SBLASTRECORDCHK(sb); 901 } 902 903 /* 904 * Append mbuf chain m to the last record in the socket buffer sb. The 905 * additional space associated the mbuf chain is recorded in sb. Empty mbufs 906 * are discarded and mbufs are compacted where possible. 907 */ 908 void 909 sbappend(struct sockbuf *sb, struct mbuf *m, int flags) 910 { 911 912 SOCKBUF_LOCK(sb); 913 sbappend_locked(sb, m, flags); 914 SOCKBUF_UNLOCK(sb); 915 } 916 917 #ifdef KERN_TLS 918 /* 919 * Append an mbuf containing encrypted TLS data. The data 920 * is marked M_NOTREADY until it has been decrypted and 921 * stored as a TLS record. 922 */ 923 static void 924 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m) 925 { 926 struct mbuf *n; 927 928 SBLASTMBUFCHK(sb); 929 930 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 931 m_demote(m, 1, 0); 932 933 for (n = m; n != NULL; n = n->m_next) 934 n->m_flags |= M_NOTREADY; 935 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail); 936 ktls_check_rx(sb); 937 } 938 #endif 939 940 /* 941 * This version of sbappend() should only be used when the caller absolutely 942 * knows that there will never be more than one record in the socket buffer, 943 * that is, a stream protocol (such as TCP). 944 */ 945 void 946 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags) 947 { 948 SOCKBUF_LOCK_ASSERT(sb); 949 950 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); 951 952 #ifdef KERN_TLS 953 /* 954 * Decrypted TLS records are appended as records via 955 * sbappendrecord(). TCP passes encrypted TLS records to this 956 * function which must be scheduled for decryption. 957 */ 958 if (sb->sb_flags & SB_TLS_RX) { 959 sbappend_ktls_rx(sb, m); 960 return; 961 } 962 #endif 963 964 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); 965 966 SBLASTMBUFCHK(sb); 967 968 #ifdef KERN_TLS 969 if (sb->sb_tls_info != NULL) 970 ktls_seq(sb, m); 971 #endif 972 973 /* Remove all packet headers and mbuf tags to get a pure data chain. */ 974 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0); 975 976 sbcompress(sb, m, sb->sb_mbtail); 977 978 sb->sb_lastrecord = sb->sb_mb; 979 SBLASTRECORDCHK(sb); 980 } 981 982 /* 983 * This version of sbappend() should only be used when the caller absolutely 984 * knows that there will never be more than one record in the socket buffer, 985 * that is, a stream protocol (such as TCP). 986 */ 987 void 988 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags) 989 { 990 991 SOCKBUF_LOCK(sb); 992 sbappendstream_locked(sb, m, flags); 993 SOCKBUF_UNLOCK(sb); 994 } 995 996 #ifdef SOCKBUF_DEBUG 997 void 998 sbcheck(struct sockbuf *sb, const char *file, int line) 999 { 1000 struct mbuf *m, *n, *fnrdy; 1001 u_long acc, ccc, mbcnt; 1002 #ifdef KERN_TLS 1003 u_long tlscc; 1004 #endif 1005 1006 SOCKBUF_LOCK_ASSERT(sb); 1007 1008 acc = ccc = mbcnt = 0; 1009 fnrdy = NULL; 1010 1011 for (m = sb->sb_mb; m; m = n) { 1012 n = m->m_nextpkt; 1013 for (; m; m = m->m_next) { 1014 if (m->m_len == 0) { 1015 printf("sb %p empty mbuf %p\n", sb, m); 1016 goto fail; 1017 } 1018 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) { 1019 if (m != sb->sb_fnrdy) { 1020 printf("sb %p: fnrdy %p != m %p\n", 1021 sb, sb->sb_fnrdy, m); 1022 goto fail; 1023 } 1024 fnrdy = m; 1025 } 1026 if (fnrdy) { 1027 if (!(m->m_flags & M_NOTAVAIL)) { 1028 printf("sb %p: fnrdy %p, m %p is avail\n", 1029 sb, sb->sb_fnrdy, m); 1030 goto fail; 1031 } 1032 } else 1033 acc += m->m_len; 1034 ccc += m->m_len; 1035 mbcnt += MSIZE; 1036 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1037 mbcnt += m->m_ext.ext_size; 1038 } 1039 } 1040 #ifdef KERN_TLS 1041 /* 1042 * Account for mbufs "detached" by ktls_detach_record() while 1043 * they are decrypted by ktls_decrypt(). tlsdcc gives a count 1044 * of the detached bytes that are included in ccc. The mbufs 1045 * and clusters are not included in the socket buffer 1046 * accounting. 1047 */ 1048 ccc += sb->sb_tlsdcc; 1049 1050 tlscc = 0; 1051 for (m = sb->sb_mtls; m; m = m->m_next) { 1052 if (m->m_nextpkt != NULL) { 1053 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m); 1054 goto fail; 1055 } 1056 if ((m->m_flags & M_NOTREADY) == 0) { 1057 printf("sb %p TLS mbuf %p ready\n", sb, m); 1058 goto fail; 1059 } 1060 tlscc += m->m_len; 1061 ccc += m->m_len; 1062 mbcnt += MSIZE; 1063 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ 1064 mbcnt += m->m_ext.ext_size; 1065 } 1066 1067 if (sb->sb_tlscc != tlscc) { 1068 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1069 sb->sb_tlsdcc); 1070 goto fail; 1071 } 1072 #endif 1073 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) { 1074 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n", 1075 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt); 1076 #ifdef KERN_TLS 1077 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc, 1078 sb->sb_tlsdcc); 1079 #endif 1080 goto fail; 1081 } 1082 return; 1083 fail: 1084 panic("%s from %s:%u", __func__, file, line); 1085 } 1086 #endif 1087 1088 /* 1089 * As above, except the mbuf chain begins a new record. 1090 */ 1091 void 1092 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0) 1093 { 1094 struct mbuf *m; 1095 1096 SOCKBUF_LOCK_ASSERT(sb); 1097 1098 if (m0 == NULL) 1099 return; 1100 m_clrprotoflags(m0); 1101 /* 1102 * Put the first mbuf on the queue. Note this permits zero length 1103 * records. 1104 */ 1105 sballoc(sb, m0); 1106 SBLASTRECORDCHK(sb); 1107 SBLINKRECORD(sb, m0); 1108 sb->sb_mbtail = m0; 1109 m = m0->m_next; 1110 m0->m_next = 0; 1111 if (m && (m0->m_flags & M_EOR)) { 1112 m0->m_flags &= ~M_EOR; 1113 m->m_flags |= M_EOR; 1114 } 1115 /* always call sbcompress() so it can do SBLASTMBUFCHK() */ 1116 sbcompress(sb, m, m0); 1117 } 1118 1119 /* 1120 * As above, except the mbuf chain begins a new record. 1121 */ 1122 void 1123 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 1124 { 1125 1126 SOCKBUF_LOCK(sb); 1127 sbappendrecord_locked(sb, m0); 1128 SOCKBUF_UNLOCK(sb); 1129 } 1130 1131 /* Helper routine that appends data, control, and address to a sockbuf. */ 1132 static int 1133 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa, 1134 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last) 1135 { 1136 struct mbuf *m, *n, *nlast; 1137 #if MSIZE <= 256 1138 if (asa->sa_len > MLEN) 1139 return (0); 1140 #endif 1141 m = m_get(M_NOWAIT, MT_SONAME); 1142 if (m == NULL) 1143 return (0); 1144 m->m_len = asa->sa_len; 1145 bcopy(asa, mtod(m, caddr_t), asa->sa_len); 1146 if (m0) { 1147 M_ASSERT_NO_SND_TAG(m0); 1148 m_clrprotoflags(m0); 1149 m_tag_delete_chain(m0, NULL); 1150 /* 1151 * Clear some persistent info from pkthdr. 1152 * We don't use m_demote(), because some netgraph consumers 1153 * expect M_PKTHDR presence. 1154 */ 1155 m0->m_pkthdr.rcvif = NULL; 1156 m0->m_pkthdr.flowid = 0; 1157 m0->m_pkthdr.csum_flags = 0; 1158 m0->m_pkthdr.fibnum = 0; 1159 m0->m_pkthdr.rsstype = 0; 1160 } 1161 if (ctrl_last) 1162 ctrl_last->m_next = m0; /* concatenate data to control */ 1163 else 1164 control = m0; 1165 m->m_next = control; 1166 for (n = m; n->m_next != NULL; n = n->m_next) 1167 sballoc(sb, n); 1168 sballoc(sb, n); 1169 nlast = n; 1170 SBLINKRECORD(sb, m); 1171 1172 sb->sb_mbtail = nlast; 1173 SBLASTMBUFCHK(sb); 1174 1175 SBLASTRECORDCHK(sb); 1176 return (1); 1177 } 1178 1179 /* 1180 * Append address and data, and optionally, control (ancillary) data to the 1181 * receive queue of a socket. If present, m0 must include a packet header 1182 * with total length. Returns 0 if no space in sockbuf or insufficient 1183 * mbufs. 1184 */ 1185 int 1186 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa, 1187 struct mbuf *m0, struct mbuf *control) 1188 { 1189 struct mbuf *ctrl_last; 1190 int space = asa->sa_len; 1191 1192 SOCKBUF_LOCK_ASSERT(sb); 1193 1194 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1195 panic("sbappendaddr_locked"); 1196 if (m0) 1197 space += m0->m_pkthdr.len; 1198 space += m_length(control, &ctrl_last); 1199 1200 if (space > sbspace(sb)) 1201 return (0); 1202 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1203 } 1204 1205 /* 1206 * Append address and data, and optionally, control (ancillary) data to the 1207 * receive queue of a socket. If present, m0 must include a packet header 1208 * with total length. Returns 0 if insufficient mbufs. Does not validate space 1209 * on the receiving sockbuf. 1210 */ 1211 int 1212 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa, 1213 struct mbuf *m0, struct mbuf *control) 1214 { 1215 struct mbuf *ctrl_last; 1216 1217 SOCKBUF_LOCK_ASSERT(sb); 1218 1219 ctrl_last = (control == NULL) ? NULL : m_last(control); 1220 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last)); 1221 } 1222 1223 /* 1224 * Append address and data, and optionally, control (ancillary) data to the 1225 * receive queue of a socket. If present, m0 must include a packet header 1226 * with total length. Returns 0 if no space in sockbuf or insufficient 1227 * mbufs. 1228 */ 1229 int 1230 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, 1231 struct mbuf *m0, struct mbuf *control) 1232 { 1233 int retval; 1234 1235 SOCKBUF_LOCK(sb); 1236 retval = sbappendaddr_locked(sb, asa, m0, control); 1237 SOCKBUF_UNLOCK(sb); 1238 return (retval); 1239 } 1240 1241 void 1242 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0, 1243 struct mbuf *control, int flags) 1244 { 1245 struct mbuf *m, *mlast; 1246 1247 sbm_clrprotoflags(m0, flags); 1248 m_last(control)->m_next = m0; 1249 1250 SBLASTRECORDCHK(sb); 1251 1252 for (m = control; m->m_next; m = m->m_next) 1253 sballoc(sb, m); 1254 sballoc(sb, m); 1255 mlast = m; 1256 SBLINKRECORD(sb, control); 1257 1258 sb->sb_mbtail = mlast; 1259 SBLASTMBUFCHK(sb); 1260 1261 SBLASTRECORDCHK(sb); 1262 } 1263 1264 void 1265 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control, 1266 int flags) 1267 { 1268 1269 SOCKBUF_LOCK(sb); 1270 sbappendcontrol_locked(sb, m0, control, flags); 1271 SOCKBUF_UNLOCK(sb); 1272 } 1273 1274 /* 1275 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf 1276 * (n). If (n) is NULL, the buffer is presumed empty. 1277 * 1278 * When the data is compressed, mbufs in the chain may be handled in one of 1279 * three ways: 1280 * 1281 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no 1282 * record boundary, and no change in data type). 1283 * 1284 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into 1285 * an mbuf already in the socket buffer. This can occur if an 1286 * appropriate mbuf exists, there is room, both mbufs are not marked as 1287 * not ready, and no merging of data types will occur. 1288 * 1289 * (3) The mbuf may be appended to the end of the existing mbuf chain. 1290 * 1291 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as 1292 * end-of-record. 1293 */ 1294 void 1295 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1296 { 1297 int eor = 0; 1298 struct mbuf *o; 1299 1300 SOCKBUF_LOCK_ASSERT(sb); 1301 1302 while (m) { 1303 eor |= m->m_flags & M_EOR; 1304 if (m->m_len == 0 && 1305 (eor == 0 || 1306 (((o = m->m_next) || (o = n)) && 1307 o->m_type == m->m_type))) { 1308 if (sb->sb_lastrecord == m) 1309 sb->sb_lastrecord = m->m_next; 1310 m = m_free(m); 1311 continue; 1312 } 1313 if (n && (n->m_flags & M_EOR) == 0 && 1314 M_WRITABLE(n) && 1315 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1316 !(m->m_flags & M_NOTREADY) && 1317 !(n->m_flags & (M_NOTREADY | M_EXTPG)) && 1318 !mbuf_has_tls_session(m) && 1319 !mbuf_has_tls_session(n) && 1320 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1321 m->m_len <= M_TRAILINGSPACE(n) && 1322 n->m_type == m->m_type) { 1323 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1324 n->m_len += m->m_len; 1325 sb->sb_ccc += m->m_len; 1326 if (sb->sb_fnrdy == NULL) 1327 sb->sb_acc += m->m_len; 1328 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1329 /* XXX: Probably don't need.*/ 1330 sb->sb_ctl += m->m_len; 1331 m = m_free(m); 1332 continue; 1333 } 1334 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) && 1335 (m->m_flags & M_NOTREADY) == 0 && 1336 !mbuf_has_tls_session(m)) 1337 (void)mb_unmapped_compress(m); 1338 if (n) 1339 n->m_next = m; 1340 else 1341 sb->sb_mb = m; 1342 sb->sb_mbtail = m; 1343 sballoc(sb, m); 1344 n = m; 1345 m->m_flags &= ~M_EOR; 1346 m = m->m_next; 1347 n->m_next = 0; 1348 } 1349 if (eor) { 1350 KASSERT(n != NULL, ("sbcompress: eor && n == NULL")); 1351 n->m_flags |= eor; 1352 } 1353 SBLASTMBUFCHK(sb); 1354 } 1355 1356 #ifdef KERN_TLS 1357 /* 1358 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs 1359 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also 1360 * a bit simpler (no EOR markers, always MT_DATA, etc.). 1361 */ 1362 static void 1363 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1364 { 1365 1366 SOCKBUF_LOCK_ASSERT(sb); 1367 1368 while (m) { 1369 KASSERT((m->m_flags & M_EOR) == 0, 1370 ("TLS RX mbuf %p with EOR", m)); 1371 KASSERT(m->m_type == MT_DATA, 1372 ("TLS RX mbuf %p is not MT_DATA", m)); 1373 KASSERT((m->m_flags & M_NOTREADY) != 0, 1374 ("TLS RX mbuf %p ready", m)); 1375 KASSERT((m->m_flags & M_EXTPG) == 0, 1376 ("TLS RX mbuf %p unmapped", m)); 1377 1378 if (m->m_len == 0) { 1379 m = m_free(m); 1380 continue; 1381 } 1382 1383 /* 1384 * Even though both 'n' and 'm' are NOTREADY, it's ok 1385 * to coalesce the data. 1386 */ 1387 if (n && 1388 M_WRITABLE(n) && 1389 ((sb->sb_flags & SB_NOCOALESCE) == 0) && 1390 !(n->m_flags & (M_EXTPG)) && 1391 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ 1392 m->m_len <= M_TRAILINGSPACE(n)) { 1393 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len)); 1394 n->m_len += m->m_len; 1395 sb->sb_ccc += m->m_len; 1396 sb->sb_tlscc += m->m_len; 1397 m = m_free(m); 1398 continue; 1399 } 1400 if (n) 1401 n->m_next = m; 1402 else 1403 sb->sb_mtls = m; 1404 sb->sb_mtlstail = m; 1405 sballoc_ktls_rx(sb, m); 1406 n = m; 1407 m = m->m_next; 1408 n->m_next = NULL; 1409 } 1410 SBLASTMBUFCHK(sb); 1411 } 1412 #endif 1413 1414 /* 1415 * Free all mbufs in a sockbuf. Check that all resources are reclaimed. 1416 */ 1417 static void 1418 sbflush_internal(struct sockbuf *sb) 1419 { 1420 1421 while (sb->sb_mbcnt || sb->sb_tlsdcc) { 1422 /* 1423 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty: 1424 * we would loop forever. Panic instead. 1425 */ 1426 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len)) 1427 break; 1428 m_freem(sbcut_internal(sb, (int)sb->sb_ccc)); 1429 } 1430 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 1431 ("%s: ccc %u mb %p mbcnt %u", __func__, 1432 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 1433 } 1434 1435 void 1436 sbflush_locked(struct sockbuf *sb) 1437 { 1438 1439 SOCKBUF_LOCK_ASSERT(sb); 1440 sbflush_internal(sb); 1441 } 1442 1443 void 1444 sbflush(struct sockbuf *sb) 1445 { 1446 1447 SOCKBUF_LOCK(sb); 1448 sbflush_locked(sb); 1449 SOCKBUF_UNLOCK(sb); 1450 } 1451 1452 /* 1453 * Cut data from (the front of) a sockbuf. 1454 */ 1455 static struct mbuf * 1456 sbcut_internal(struct sockbuf *sb, int len) 1457 { 1458 struct mbuf *m, *next, *mfree; 1459 bool is_tls; 1460 1461 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0", 1462 __func__, len)); 1463 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u", 1464 __func__, len, sb->sb_ccc)); 1465 1466 next = (m = sb->sb_mb) ? m->m_nextpkt : 0; 1467 is_tls = false; 1468 mfree = NULL; 1469 1470 while (len > 0) { 1471 if (m == NULL) { 1472 #ifdef KERN_TLS 1473 if (next == NULL && !is_tls) { 1474 if (sb->sb_tlsdcc != 0) { 1475 MPASS(len >= sb->sb_tlsdcc); 1476 len -= sb->sb_tlsdcc; 1477 sb->sb_ccc -= sb->sb_tlsdcc; 1478 sb->sb_tlsdcc = 0; 1479 if (len == 0) 1480 break; 1481 } 1482 next = sb->sb_mtls; 1483 is_tls = true; 1484 } 1485 #endif 1486 KASSERT(next, ("%s: no next, len %d", __func__, len)); 1487 m = next; 1488 next = m->m_nextpkt; 1489 } 1490 if (m->m_len > len) { 1491 KASSERT(!(m->m_flags & M_NOTAVAIL), 1492 ("%s: m %p M_NOTAVAIL", __func__, m)); 1493 m->m_len -= len; 1494 m->m_data += len; 1495 sb->sb_ccc -= len; 1496 sb->sb_acc -= len; 1497 if (sb->sb_sndptroff != 0) 1498 sb->sb_sndptroff -= len; 1499 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA) 1500 sb->sb_ctl -= len; 1501 break; 1502 } 1503 len -= m->m_len; 1504 #ifdef KERN_TLS 1505 if (is_tls) 1506 sbfree_ktls_rx(sb, m); 1507 else 1508 #endif 1509 sbfree(sb, m); 1510 /* 1511 * Do not put M_NOTREADY buffers to the free list, they 1512 * are referenced from outside. 1513 */ 1514 if (m->m_flags & M_NOTREADY && !is_tls) 1515 m = m->m_next; 1516 else { 1517 struct mbuf *n; 1518 1519 n = m->m_next; 1520 m->m_next = mfree; 1521 mfree = m; 1522 m = n; 1523 } 1524 } 1525 /* 1526 * Free any zero-length mbufs from the buffer. 1527 * For SOCK_DGRAM sockets such mbufs represent empty records. 1528 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer, 1529 * when sosend_generic() needs to send only control data. 1530 */ 1531 while (m && m->m_len == 0) { 1532 struct mbuf *n; 1533 1534 sbfree(sb, m); 1535 n = m->m_next; 1536 m->m_next = mfree; 1537 mfree = m; 1538 m = n; 1539 } 1540 #ifdef KERN_TLS 1541 if (is_tls) { 1542 sb->sb_mb = NULL; 1543 sb->sb_mtls = m; 1544 if (m == NULL) 1545 sb->sb_mtlstail = NULL; 1546 } else 1547 #endif 1548 if (m) { 1549 sb->sb_mb = m; 1550 m->m_nextpkt = next; 1551 } else 1552 sb->sb_mb = next; 1553 /* 1554 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure 1555 * sb_lastrecord is up-to-date if we dropped part of the last record. 1556 */ 1557 m = sb->sb_mb; 1558 if (m == NULL) { 1559 sb->sb_mbtail = NULL; 1560 sb->sb_lastrecord = NULL; 1561 } else if (m->m_nextpkt == NULL) { 1562 sb->sb_lastrecord = m; 1563 } 1564 1565 return (mfree); 1566 } 1567 1568 /* 1569 * Drop data from (the front of) a sockbuf. 1570 */ 1571 void 1572 sbdrop_locked(struct sockbuf *sb, int len) 1573 { 1574 1575 SOCKBUF_LOCK_ASSERT(sb); 1576 m_freem(sbcut_internal(sb, len)); 1577 } 1578 1579 /* 1580 * Drop data from (the front of) a sockbuf, 1581 * and return it to caller. 1582 */ 1583 struct mbuf * 1584 sbcut_locked(struct sockbuf *sb, int len) 1585 { 1586 1587 SOCKBUF_LOCK_ASSERT(sb); 1588 return (sbcut_internal(sb, len)); 1589 } 1590 1591 void 1592 sbdrop(struct sockbuf *sb, int len) 1593 { 1594 struct mbuf *mfree; 1595 1596 SOCKBUF_LOCK(sb); 1597 mfree = sbcut_internal(sb, len); 1598 SOCKBUF_UNLOCK(sb); 1599 1600 m_freem(mfree); 1601 } 1602 1603 struct mbuf * 1604 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff) 1605 { 1606 struct mbuf *m; 1607 1608 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1609 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1610 *moff = off; 1611 if (sb->sb_sndptr == NULL) { 1612 sb->sb_sndptr = sb->sb_mb; 1613 sb->sb_sndptroff = 0; 1614 } 1615 return (sb->sb_mb); 1616 } else { 1617 m = sb->sb_sndptr; 1618 off -= sb->sb_sndptroff; 1619 } 1620 *moff = off; 1621 return (m); 1622 } 1623 1624 void 1625 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len) 1626 { 1627 /* 1628 * A small copy was done, advance forward the sb_sbsndptr to cover 1629 * it. 1630 */ 1631 struct mbuf *m; 1632 1633 if (mb != sb->sb_sndptr) { 1634 /* Did not copyout at the same mbuf */ 1635 return; 1636 } 1637 m = mb; 1638 while (m && (len > 0)) { 1639 if (len >= m->m_len) { 1640 len -= m->m_len; 1641 if (m->m_next) { 1642 sb->sb_sndptroff += m->m_len; 1643 sb->sb_sndptr = m->m_next; 1644 } 1645 m = m->m_next; 1646 } else { 1647 len = 0; 1648 } 1649 } 1650 } 1651 1652 /* 1653 * Return the first mbuf and the mbuf data offset for the provided 1654 * send offset without changing the "sb_sndptroff" field. 1655 */ 1656 struct mbuf * 1657 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff) 1658 { 1659 struct mbuf *m; 1660 1661 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__)); 1662 1663 /* 1664 * If the "off" is below the stored offset, which happens on 1665 * retransmits, just use "sb_mb": 1666 */ 1667 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) { 1668 m = sb->sb_mb; 1669 } else { 1670 m = sb->sb_sndptr; 1671 off -= sb->sb_sndptroff; 1672 } 1673 while (off > 0 && m != NULL) { 1674 if (off < m->m_len) 1675 break; 1676 off -= m->m_len; 1677 m = m->m_next; 1678 } 1679 *moff = off; 1680 return (m); 1681 } 1682 1683 /* 1684 * Drop a record off the front of a sockbuf and move the next record to the 1685 * front. 1686 */ 1687 void 1688 sbdroprecord_locked(struct sockbuf *sb) 1689 { 1690 struct mbuf *m; 1691 1692 SOCKBUF_LOCK_ASSERT(sb); 1693 1694 m = sb->sb_mb; 1695 if (m) { 1696 sb->sb_mb = m->m_nextpkt; 1697 do { 1698 sbfree(sb, m); 1699 m = m_free(m); 1700 } while (m); 1701 } 1702 SB_EMPTY_FIXUP(sb); 1703 } 1704 1705 /* 1706 * Drop a record off the front of a sockbuf and move the next record to the 1707 * front. 1708 */ 1709 void 1710 sbdroprecord(struct sockbuf *sb) 1711 { 1712 1713 SOCKBUF_LOCK(sb); 1714 sbdroprecord_locked(sb); 1715 SOCKBUF_UNLOCK(sb); 1716 } 1717 1718 /* 1719 * Create a "control" mbuf containing the specified data with the specified 1720 * type for presentation on a socket buffer. 1721 */ 1722 struct mbuf * 1723 sbcreatecontrol_how(void *p, int size, int type, int level, int wait) 1724 { 1725 struct cmsghdr *cp; 1726 struct mbuf *m; 1727 1728 MBUF_CHECKSLEEP(wait); 1729 if (CMSG_SPACE((u_int)size) > MCLBYTES) 1730 return ((struct mbuf *) NULL); 1731 if (CMSG_SPACE((u_int)size) > MLEN) 1732 m = m_getcl(wait, MT_CONTROL, 0); 1733 else 1734 m = m_get(wait, MT_CONTROL); 1735 if (m == NULL) 1736 return ((struct mbuf *) NULL); 1737 cp = mtod(m, struct cmsghdr *); 1738 m->m_len = 0; 1739 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), 1740 ("sbcreatecontrol: short mbuf")); 1741 /* 1742 * Don't leave the padding between the msg header and the 1743 * cmsg data and the padding after the cmsg data un-initialized. 1744 */ 1745 bzero(cp, CMSG_SPACE((u_int)size)); 1746 if (p != NULL) 1747 (void)memcpy(CMSG_DATA(cp), p, size); 1748 m->m_len = CMSG_SPACE(size); 1749 cp->cmsg_len = CMSG_LEN(size); 1750 cp->cmsg_level = level; 1751 cp->cmsg_type = type; 1752 return (m); 1753 } 1754 1755 struct mbuf * 1756 sbcreatecontrol(caddr_t p, int size, int type, int level) 1757 { 1758 1759 return (sbcreatecontrol_how(p, size, type, level, M_NOWAIT)); 1760 } 1761 1762 /* 1763 * This does the same for socket buffers that sotoxsocket does for sockets: 1764 * generate an user-format data structure describing the socket buffer. Note 1765 * that the xsockbuf structure, since it is always embedded in a socket, does 1766 * not include a self pointer nor a length. We make this entry point public 1767 * in case some other mechanism needs it. 1768 */ 1769 void 1770 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) 1771 { 1772 1773 xsb->sb_cc = sb->sb_ccc; 1774 xsb->sb_hiwat = sb->sb_hiwat; 1775 xsb->sb_mbcnt = sb->sb_mbcnt; 1776 xsb->sb_mcnt = sb->sb_mcnt; 1777 xsb->sb_ccnt = sb->sb_ccnt; 1778 xsb->sb_mbmax = sb->sb_mbmax; 1779 xsb->sb_lowat = sb->sb_lowat; 1780 xsb->sb_flags = sb->sb_flags; 1781 xsb->sb_timeo = sb->sb_timeo; 1782 } 1783 1784 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 1785 static int dummy; 1786 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, ""); 1787 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, 1788 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0, 1789 sysctl_handle_sb_max, "LU", 1790 "Maximum socket buffer size"); 1791 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 1792 &sb_efficiency, 0, "Socket buffer size waste factor");
Cache object: 19e5deb6e31e022c0bf3f2e7fbce310b
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]