1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 /* 28 * Xen network backend - mac client edition. 29 * 30 * A driver that sits above an existing GLDv3/Nemo MAC driver and 31 * relays packets to/from that driver from/to a guest domain. 32 */ 33 34 #include "xnb.h" 35 36 #include <sys/sunddi.h> 37 #include <sys/ddi.h> 38 #include <sys/modctl.h> 39 #include <sys/strsubr.h> 40 #include <sys/mac_client.h> 41 #include <sys/mac_provider.h> 42 #include <sys/mac_client_priv.h> 43 #include <sys/mac.h> 44 #include <net/if.h> 45 #include <sys/dlpi.h> 46 #include <sys/pattr.h> 47 #include <xen/sys/xenbus_impl.h> 48 #include <xen/sys/xendev.h> 49 50 typedef struct xnbo { 51 mac_handle_t o_mh; 52 mac_client_handle_t o_mch; 53 mac_unicast_handle_t o_mah; 54 mac_promisc_handle_t o_mphp; 55 boolean_t o_running; 56 boolean_t o_promiscuous; 57 uint32_t o_hcksum_capab; 58 } xnbo_t; 59 60 static void xnbo_close_mac(xnbo_t *); 61 62 /* 63 * Packets from the peer come here. We pass them to the mac device. 64 */ 65 static void 66 xnbo_to_mac(xnb_t *xnbp, mblk_t *mp) 67 { 68 xnbo_t *xnbop = xnbp->xnb_flavour_data; 69 70 ASSERT(mp != NULL); 71 72 if (!xnbop->o_running) { 73 xnbp->xnb_stat_tx_too_early++; 74 goto fail; 75 } 76 77 if (mac_tx(xnbop->o_mch, mp, 0, 78 MAC_DROP_ON_NO_DESC, NULL) != NULL) { 79 xnbp->xnb_stat_mac_full++; 80 } 81 82 return; 83 84 fail: 85 freemsgchain(mp); 86 } 87 88 static mblk_t * 89 xnbo_cksum_from_peer(xnb_t *xnbp, mblk_t *mp, uint16_t flags) 90 { 91 xnbo_t *xnbop = xnbp->xnb_flavour_data; 92 93 ASSERT(mp->b_next == NULL); 94 95 if ((flags & NETTXF_csum_blank) != 0) { 96 /* 97 * It would be nice to ASSERT that xnbp->xnb_cksum_offload 98 * is TRUE here, but some peers insist on assuming 99 * that it is available even when they have been told 100 * otherwise. 101 * 102 * The checksum in the packet is blank. Determine 103 * whether we can do hardware offload and, if so, 104 * update the flags on the mblk according. If not, 105 * calculate and insert the checksum using software. 106 */ 107 mp = xnb_process_cksum_flags(xnbp, mp, 108 xnbop->o_hcksum_capab); 109 } 110 111 return (mp); 112 } 113 114 static uint16_t 115 xnbo_cksum_to_peer(xnb_t *xnbp, mblk_t *mp) 116 { 117 uint16_t r = 0; 118 119 /* 120 * We might also check for HCK_PARTIALCKSUM here and, 121 * providing that the partial checksum covers the TCP/UDP 122 * payload, return NETRXF_data_validated. 123 * 124 * It seems that it's probably not worthwhile, as even MAC 125 * devices which advertise HCKSUM_INET_PARTIAL in their 126 * capabilities tend to use HCK_FULLCKSUM on the receive side 127 * - they are actually saying that in the output path the 128 * caller must use HCK_PARTIALCKSUM. 129 */ 130 131 if (xnbp->xnb_cksum_offload) { 132 uint32_t pflags, csum; 133 134 /* 135 * XXPV dme: Pull in improved hcksum_retrieve() from 136 * Crossbow, which gives back the csum in the seventh 137 * argument for HCK_FULLCKSUM. 138 */ 139 hcksum_retrieve(mp, NULL, NULL, NULL, NULL, 140 NULL, NULL, &pflags); 141 csum = DB_CKSUM16(mp); 142 143 /* 144 * If the MAC driver has asserted that the checksum is 145 * good, let the peer know. 146 */ 147 if (((pflags & HCK_FULLCKSUM) != 0) && 148 (((pflags & HCK_FULLCKSUM_OK) != 0) || 149 (csum == 0xffff))) 150 r |= NETRXF_data_validated; 151 } 152 153 return (r); 154 } 155 156 /* 157 * Packets from the mac device come here. We pass them to the peer. 158 */ 159 /*ARGSUSED*/ 160 static void 161 xnbo_from_mac(void *arg, mac_resource_handle_t mrh, mblk_t *mp, 162 boolean_t loopback) 163 { 164 xnb_t *xnbp = arg; 165 166 mp = xnb_copy_to_peer(xnbp, mp); 167 168 if (mp != NULL) 169 freemsgchain(mp); 170 } 171 172 /* 173 * Packets from the mac device come here. We pass them to the peer if 174 * the destination mac address matches or it's a multicast/broadcast 175 * address. 176 */ 177 /*ARGSUSED*/ 178 static void 179 xnbo_from_mac_filter(void *arg, mac_resource_handle_t mrh, mblk_t *mp, 180 boolean_t loopback) 181 { 182 xnb_t *xnbp = arg; 183 xnbo_t *xnbop = xnbp->xnb_flavour_data; 184 mblk_t *next, *keep, *keep_head, *free, *free_head; 185 186 keep = keep_head = free = free_head = NULL; 187 188 #define ADD(list, bp) \ 189 if (list != NULL) \ 190 list->b_next = bp; \ 191 else \ 192 list##_head = bp; \ 193 list = bp; 194 195 for (; mp != NULL; mp = next) { 196 mac_header_info_t hdr_info; 197 198 next = mp->b_next; 199 mp->b_next = NULL; 200 201 if (mac_header_info(xnbop->o_mh, mp, &hdr_info) != 0) { 202 ADD(free, mp); 203 continue; 204 } 205 206 if ((hdr_info.mhi_dsttype == MAC_ADDRTYPE_BROADCAST) || 207 (hdr_info.mhi_dsttype == MAC_ADDRTYPE_MULTICAST)) { 208 ADD(keep, mp); 209 continue; 210 } 211 212 if (bcmp(hdr_info.mhi_daddr, xnbp->xnb_mac_addr, 213 sizeof (xnbp->xnb_mac_addr)) == 0) { 214 ADD(keep, mp); 215 continue; 216 } 217 218 ADD(free, mp); 219 } 220 #undef ADD 221 222 if (keep_head != NULL) 223 xnbo_from_mac(xnbp, mrh, keep_head, B_FALSE); 224 225 if (free_head != NULL) 226 freemsgchain(free_head); 227 } 228 229 static boolean_t 230 xnbo_open_mac(xnb_t *xnbp, char *mac) 231 { 232 xnbo_t *xnbop = xnbp->xnb_flavour_data; 233 int err, need_rx_filter, need_setphysaddr, need_promiscuous; 234 const mac_info_t *mi; 235 char *xsname; 236 void (*rx_fn)(void *, mac_resource_handle_t, mblk_t *, boolean_t); 237 struct ether_addr ea; 238 uint_t max_sdu; 239 mac_diag_t diag; 240 241 xsname = xvdi_get_xsname(xnbp->xnb_devinfo); 242 243 if ((err = mac_open_by_linkname(mac, &xnbop->o_mh)) != 0) { 244 cmn_err(CE_WARN, "xnbo_open_mac: " 245 "cannot open mac for link %s (%d)", mac, err); 246 return (B_FALSE); 247 } 248 ASSERT(xnbop->o_mh != NULL); 249 250 mi = mac_info(xnbop->o_mh); 251 ASSERT(mi != NULL); 252 253 if (mi->mi_media != DL_ETHER) { 254 cmn_err(CE_WARN, "xnbo_open_mac: " 255 "device is not DL_ETHER (%d)", mi->mi_media); 256 xnbo_close_mac(xnbop); 257 return (B_FALSE); 258 } 259 if (mi->mi_media != mi->mi_nativemedia) { 260 cmn_err(CE_WARN, "xnbo_open_mac: " 261 "device media and native media mismatch (%d != %d)", 262 mi->mi_media, mi->mi_nativemedia); 263 xnbo_close_mac(xnbop); 264 return (B_FALSE); 265 } 266 267 mac_sdu_get(xnbop->o_mh, NULL, &max_sdu); 268 if (max_sdu > XNBMAXPKT) { 269 cmn_err(CE_WARN, "xnbo_open_mac: mac device SDU too big (%d)", 270 max_sdu); 271 xnbo_close_mac(xnbop); 272 return (B_FALSE); 273 } 274 275 if (mac_client_open(xnbop->o_mh, &xnbop->o_mch, NULL, 276 MAC_OPEN_FLAGS_USE_DATALINK_NAME) != 0) { 277 cmn_err(CE_WARN, "xnbo_open_mac: " 278 "error (%d) opening mac client", err); 279 xnbo_close_mac(xnbop); 280 return (B_FALSE); 281 } 282 283 err = mac_unicast_primary_add(xnbop->o_mch, &xnbop->o_mah, &diag); 284 if (err != 0) { 285 cmn_err(CE_WARN, "xnbo_open_mac: " 286 "failed to get the primary MAC address of " 287 "%s: %d", mac, err); 288 xnbo_close_mac(xnbop); 289 return (B_FALSE); 290 } 291 292 /* 293 * Should the receive path filter packets from the downstream 294 * NIC before passing them to the peer? The default is "no". 295 */ 296 if (xenbus_scanf(XBT_NULL, xsname, 297 "SUNW-need-rx-filter", "%d", &need_rx_filter) != 0) 298 need_rx_filter = 0; 299 if (need_rx_filter > 0) 300 rx_fn = xnbo_from_mac_filter; 301 else 302 rx_fn = xnbo_from_mac; 303 304 /* 305 * Should we set the underlying NIC into promiscuous mode? The 306 * default is "no". 307 */ 308 if (xenbus_scanf(XBT_NULL, xsname, 309 "SUNW-need-promiscuous", "%d", &need_promiscuous) != 0) 310 need_promiscuous = 0; 311 if (need_promiscuous == 0) { 312 mac_rx_set(xnbop->o_mch, rx_fn, xnbp); 313 } else { 314 err = mac_promisc_add(xnbop->o_mch, MAC_CLIENT_PROMISC_ALL, 315 rx_fn, xnbp, &xnbop->o_mphp, MAC_PROMISC_FLAGS_NO_TX_LOOP); 316 if (err != 0) { 317 cmn_err(CE_WARN, "xnbo_open_mac: " 318 "cannot enable promiscuous mode of %s: %d", 319 mac, err); 320 xnbo_close_mac(xnbop); 321 return (B_FALSE); 322 } 323 xnbop->o_promiscuous = B_TRUE; 324 } 325 326 if (!mac_capab_get(xnbop->o_mh, MAC_CAPAB_HCKSUM, 327 &xnbop->o_hcksum_capab)) 328 xnbop->o_hcksum_capab = 0; 329 330 /* 331 * Should we set the physical address of the underlying NIC 332 * to match that assigned to the peer? The default is "no". 333 */ 334 if (xenbus_scanf(XBT_NULL, xsname, 335 "SUNW-need-set-physaddr", "%d", &need_setphysaddr) != 0) 336 need_setphysaddr = 0; 337 if (need_setphysaddr > 0) { 338 err = mac_unicast_primary_set(xnbop->o_mh, xnbp->xnb_mac_addr); 339 /* Warn, but continue on. */ 340 if (err != 0) { 341 bcopy(xnbp->xnb_mac_addr, ea.ether_addr_octet, 342 ETHERADDRL); 343 cmn_err(CE_WARN, "xnbo_open_mac: " 344 "cannot set MAC address of %s to " 345 "%s: %d", mac, ether_sprintf(&ea), err); 346 } 347 } 348 349 xnbop->o_running = B_TRUE; 350 351 return (B_TRUE); 352 } 353 354 /* 355 * xnb calls back here when the user-level hotplug code reports that 356 * the hotplug has successfully completed. For this flavour that means 357 * that the underlying MAC device that we will use is ready to be 358 * opened. 359 */ 360 static boolean_t 361 xnbo_hotplug(xnb_t *xnbp) 362 { 363 char *xsname; 364 char mac[LIFNAMSIZ]; 365 366 xsname = xvdi_get_xsname(xnbp->xnb_devinfo); 367 if (xenbus_scanf(XBT_NULL, xsname, "nic", "%s", mac) != 0) { 368 cmn_err(CE_WARN, "xnbo_hotplug: " 369 "cannot read nic name from %s", xsname); 370 return (B_FALSE); 371 } 372 373 return (xnbo_open_mac(xnbp, mac)); 374 } 375 376 static void 377 xnbo_close_mac(xnbo_t *xnbop) 378 { 379 if (xnbop->o_mh == NULL) 380 return; 381 382 if (xnbop->o_running) { 383 xnbop->o_running = B_FALSE; 384 } 385 386 if (xnbop->o_promiscuous) { 387 if (xnbop->o_mphp != NULL) { 388 (void) mac_promisc_remove(xnbop->o_mphp); 389 xnbop->o_mphp = NULL; 390 } 391 xnbop->o_promiscuous = B_FALSE; 392 } else { 393 if (xnbop->o_mch != NULL) 394 mac_rx_clear(xnbop->o_mch); 395 } 396 397 if (xnbop->o_mah != NULL) { 398 (void) mac_unicast_remove(xnbop->o_mch, xnbop->o_mah); 399 xnbop->o_mah = NULL; 400 } 401 402 if (xnbop->o_mch != NULL) { 403 mac_client_close(xnbop->o_mch, 0); 404 xnbop->o_mch = NULL; 405 } 406 407 mac_close(xnbop->o_mh); 408 xnbop->o_mh = NULL; 409 } 410 411 /* 412 * xnb calls back here when we successfully synchronize with the 413 * driver in the guest domain. In this flavour there is nothing to do as 414 * we open the underlying MAC device on successful hotplug completion. 415 */ 416 /*ARGSUSED*/ 417 static void 418 xnbo_connected(xnb_t *xnbp) 419 { 420 } 421 422 /* 423 * xnb calls back here when the driver in the guest domain has closed 424 * down the inter-domain connection. We close the underlying MAC device. 425 */ 426 static void 427 xnbo_disconnected(xnb_t *xnbp) 428 { 429 xnbo_close_mac(xnbp->xnb_flavour_data); 430 } 431 432 static int 433 xnbo_attach(dev_info_t *dip, ddi_attach_cmd_t cmd) 434 { 435 static xnb_flavour_t flavour = { 436 xnbo_to_mac, xnbo_connected, xnbo_disconnected, xnbo_hotplug, 437 xnbo_cksum_from_peer, xnbo_cksum_to_peer, 438 }; 439 xnbo_t *xnbop; 440 441 switch (cmd) { 442 case DDI_ATTACH: 443 break; 444 case DDI_RESUME: 445 return (DDI_SUCCESS); 446 default: 447 return (DDI_FAILURE); 448 } 449 450 xnbop = kmem_zalloc(sizeof (*xnbop), KM_SLEEP); 451 452 xnbop->o_mh = NULL; 453 xnbop->o_mch = NULL; 454 xnbop->o_mah = NULL; 455 xnbop->o_mphp = NULL; 456 xnbop->o_running = B_FALSE; 457 xnbop->o_hcksum_capab = 0; 458 459 if (xnb_attach(dip, &flavour, xnbop) != DDI_SUCCESS) { 460 kmem_free(xnbop, sizeof (*xnbop)); 461 return (DDI_FAILURE); 462 } 463 464 return (DDI_SUCCESS); 465 } 466 467 static int 468 xnbo_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) 469 { 470 xnb_t *xnbp = ddi_get_driver_private(dip); 471 xnbo_t *xnbop = xnbp->xnb_flavour_data; 472 473 switch (cmd) { 474 case DDI_DETACH: 475 break; 476 case DDI_SUSPEND: 477 return (DDI_SUCCESS); 478 default: 479 return (DDI_FAILURE); 480 } 481 482 mutex_enter(&xnbp->xnb_tx_lock); 483 mutex_enter(&xnbp->xnb_rx_lock); 484 485 if (!xnbp->xnb_detachable || xnbp->xnb_connected || 486 (xnbp->xnb_tx_buf_count > 0)) { 487 mutex_exit(&xnbp->xnb_rx_lock); 488 mutex_exit(&xnbp->xnb_tx_lock); 489 490 return (DDI_FAILURE); 491 } 492 493 mutex_exit(&xnbp->xnb_rx_lock); 494 mutex_exit(&xnbp->xnb_tx_lock); 495 496 xnbo_close_mac(xnbop); 497 kmem_free(xnbop, sizeof (*xnbop)); 498 499 xnb_detach(dip); 500 501 return (DDI_SUCCESS); 502 } 503 504 static struct cb_ops cb_ops = { 505 nulldev, /* open */ 506 nulldev, /* close */ 507 nodev, /* strategy */ 508 nodev, /* print */ 509 nodev, /* dump */ 510 nodev, /* read */ 511 nodev, /* write */ 512 nodev, /* ioctl */ 513 nodev, /* devmap */ 514 nodev, /* mmap */ 515 nodev, /* segmap */ 516 nochpoll, /* poll */ 517 ddi_prop_op, /* cb_prop_op */ 518 0, /* streamtab */ 519 D_NEW | D_MP | D_64BIT /* Driver compatibility flag */ 520 }; 521 522 static struct dev_ops ops = { 523 DEVO_REV, /* devo_rev */ 524 0, /* devo_refcnt */ 525 nulldev, /* devo_getinfo */ 526 nulldev, /* devo_identify */ 527 nulldev, /* devo_probe */ 528 xnbo_attach, /* devo_attach */ 529 xnbo_detach, /* devo_detach */ 530 nodev, /* devo_reset */ 531 &cb_ops, /* devo_cb_ops */ 532 (struct bus_ops *)0, /* devo_bus_ops */ 533 NULL, /* devo_power */ 534 ddi_quiesce_not_needed, /* devo_quiesce */ 535 }; 536 537 static struct modldrv modldrv = { 538 &mod_driverops, "xnbo driver", &ops, 539 }; 540 541 static struct modlinkage modlinkage = { 542 MODREV_1, &modldrv, NULL 543 }; 544 545 int 546 _init(void) 547 { 548 return (mod_install(&modlinkage)); 549 } 550 551 int 552 _info(struct modinfo *modinfop) 553 { 554 return (mod_info(&modlinkage, modinfop)); 555 } 556 557 int 558 _fini(void) 559 { 560 return (mod_remove(&modlinkage)); 561 } 562