xref: /titanic_41/usr/src/uts/common/inet/ip_impl.h (revision fc724630b14603e4c1147df68b7bf45f7de7431f)
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  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #ifndef	_INET_IP_IMPL_H
27 #define	_INET_IP_IMPL_H
28 
29 /*
30  * IP implementation private declarations.  These interfaces are
31  * used to build the IP module and are not meant to be accessed
32  * by any modules except IP itself.  They are undocumented and are
33  * subject to change without notice.
34  */
35 
36 #ifdef	__cplusplus
37 extern "C" {
38 #endif
39 
40 #ifdef _KERNEL
41 
42 #include <sys/sdt.h>
43 #include <sys/dld.h>
44 
45 #define	IP_MOD_ID		5701
46 
47 #define	INET_NAME	"ip"
48 
49 #ifdef	_BIG_ENDIAN
50 #define	IP_HDR_CSUM_TTL_ADJUST	256
51 #define	IP_TCP_CSUM_COMP	IPPROTO_TCP
52 #define	IP_UDP_CSUM_COMP	IPPROTO_UDP
53 #else
54 #define	IP_HDR_CSUM_TTL_ADJUST	1
55 #define	IP_TCP_CSUM_COMP	(IPPROTO_TCP << 8)
56 #define	IP_UDP_CSUM_COMP	(IPPROTO_UDP << 8)
57 #endif
58 
59 #define	TCP_CHECKSUM_OFFSET	16
60 #define	TCP_CHECKSUM_SIZE	2
61 
62 #define	UDP_CHECKSUM_OFFSET	6
63 #define	UDP_CHECKSUM_SIZE	2
64 
65 #define	IPH_TCPH_CHECKSUMP(ipha, hlen)	\
66 	((uint16_t *)(((uchar_t *)(ipha)) + ((hlen) + TCP_CHECKSUM_OFFSET)))
67 
68 #define	IPH_UDPH_CHECKSUMP(ipha, hlen)	\
69 	((uint16_t *)(((uchar_t *)(ipha)) + ((hlen) + UDP_CHECKSUM_OFFSET)))
70 
71 #define	ILL_HCKSUM_CAPABLE(ill)		\
72 	(((ill)->ill_capabilities & ILL_CAPAB_HCKSUM) != 0)
73 /*
74  * Macro that performs software checksum calculation on the IP header.
75  */
76 #define	IP_HDR_CKSUM(ipha, sum, v_hlen_tos_len, ttl_protocol) {		\
77 	(sum) += (ttl_protocol) + (ipha)->ipha_ident +			\
78 	    ((v_hlen_tos_len) >> 16) +					\
79 	    ((v_hlen_tos_len) & 0xFFFF) +				\
80 	    (ipha)->ipha_fragment_offset_and_flags;			\
81 	(sum) = (((sum) & 0xFFFF) + ((sum) >> 16));			\
82 	(sum) = ~((sum) + ((sum) >> 16));				\
83 	(ipha)->ipha_hdr_checksum = (uint16_t)(sum);			\
84 }
85 
86 #define	IS_IP_HDR_HWCKSUM(ipsec, mp, ill)				\
87 	((!ipsec) && (DB_CKSUMFLAGS(mp) & HCK_IPV4_HDRCKSUM) &&		\
88 	ILL_HCKSUM_CAPABLE(ill) && dohwcksum)
89 
90 /*
91  * This macro acts as a wrapper around IP_CKSUM_XMIT_FAST, and it performs
92  * several checks on the IRE and ILL (among other things) in order to see
93  * whether or not hardware checksum offload is allowed for the outgoing
94  * packet.  It assumes that the caller has held a reference to the IRE.
95  */
96 #define	IP_CKSUM_XMIT(ill, ire, mp, ihp, up, proto, start, end,		\
97 	    max_frag, ipsec_len, pseudo) {				\
98 	uint32_t _hck_flags;						\
99 	/*								\
100 	 * We offload checksum calculation to hardware when IPsec isn't	\
101 	 * present and if fragmentation isn't required.  We also check	\
102 	 * if M_DATA fastpath is safe to be used on the	corresponding	\
103 	 * IRE; this check is performed without grabbing ire_lock but	\
104 	 * instead by holding a reference to it.  This is sufficient	\
105 	 * for IRE_CACHE; for IRE_BROADCAST on non-Ethernet links, the	\
106 	 * DL_NOTE_FASTPATH_FLUSH indication could come up from the	\
107 	 * driver and trigger the IRE (hence fp_mp) deletion.  This is	\
108 	 * why only IRE_CACHE type is eligible for offload.		\
109 	 *								\
110 	 * The presense of IP options also forces the network stack to	\
111 	 * calculate the checksum in software.  This is because:	\
112 	 *								\
113 	 * Wrap around: certain partial-checksum NICs (eri, ce) limit	\
114 	 * the size of "start offset" width to 6-bit.  This effectively	\
115 	 * sets the largest value of the offset to 64-bytes, starting	\
116 	 * from the MAC header.  When the cumulative MAC and IP headers	\
117 	 * exceed such limit, the offset will wrap around.  This causes	\
118 	 * the checksum to be calculated at the wrong place.		\
119 	 *								\
120 	 * IPv4 source routing: none of the full-checksum capable NICs	\
121 	 * is capable of correctly handling the	IPv4 source-routing	\
122 	 * option for purposes of calculating the pseudo-header; the	\
123 	 * actual destination is different from the destination in the	\
124 	 * header which is that of the next-hop.  (This case may not be	\
125 	 * true for NICs which can parse IPv6 extension headers, but	\
126 	 * we choose to simplify the implementation by not offloading	\
127 	 * checksum when they are present.)				\
128 	 *								\
129 	 */								\
130 	if ((ill) != NULL && ILL_HCKSUM_CAPABLE(ill) &&			\
131 	    !((ire)->ire_flags & RTF_MULTIRT) &&			\
132 	    (!((ire)->ire_type & IRE_BROADCAST) ||			\
133 	    (ill)->ill_type == IFT_ETHER) &&			\
134 	    (ipsec_len) == 0 &&						\
135 	    (((ire)->ire_ipversion == IPV4_VERSION &&			\
136 	    (start) == IP_SIMPLE_HDR_LENGTH &&				\
137 	    ((ire)->ire_nce != NULL &&					\
138 	    (ire)->ire_nce->nce_fp_mp != NULL &&	\
139 	    MBLKHEAD(mp) >= MBLKL((ire)->ire_nce->nce_fp_mp))) ||	\
140 	    ((ire)->ire_ipversion == IPV6_VERSION &&			\
141 	    (start) == IPV6_HDR_LEN &&					\
142 	    (ire)->ire_nce->nce_fp_mp != NULL &&			\
143 	    MBLKHEAD(mp) >= MBLKL((ire)->ire_nce->nce_fp_mp))) &&	\
144 	    (max_frag) >= (uint_t)((end) + (ipsec_len)) &&		\
145 	    dohwcksum) {						\
146 		_hck_flags = (ill)->ill_hcksum_capab->ill_hcksum_txflags; \
147 	} else {							\
148 		_hck_flags = 0;						\
149 	}								\
150 	IP_CKSUM_XMIT_FAST((ire)->ire_ipversion, _hck_flags, mp, ihp,	\
151 	    up, proto, start, end, pseudo);				\
152 }
153 
154 /*
155  * Based on the device capabilities, this macro either marks an outgoing
156  * packet with hardware checksum offload information or calculate the
157  * checksum in software.  If the latter is performed, the checksum field
158  * of the dblk is cleared; otherwise it will be non-zero and contain the
159  * necessary flag(s) for the driver.
160  */
161 #define	IP_CKSUM_XMIT_FAST(ipver, hck_flags, mp, ihp, up, proto, start,	\
162 	    end, pseudo) {						\
163 	uint32_t _sum;							\
164 	/*								\
165 	 * Underlying interface supports hardware checksum offload for	\
166 	 * the payload; leave the payload checksum for the hardware to	\
167 	 * calculate.  N.B: We only need to set up checksum info on the	\
168 	 * first mblk.							\
169 	 */								\
170 	DB_CKSUMFLAGS(mp) = 0;						\
171 	if (((ipver) == IPV4_VERSION &&					\
172 	    ((hck_flags) & HCKSUM_INET_FULL_V4)) ||			\
173 	    ((ipver) == IPV6_VERSION &&					\
174 	    ((hck_flags) & HCKSUM_INET_FULL_V6))) {			\
175 		/*							\
176 		 * Hardware calculates pseudo-header, header and the	\
177 		 * payload checksums, so clear the checksum field in	\
178 		 * the protocol header.					\
179 		 */							\
180 		*(up) = 0;						\
181 		DB_CKSUMFLAGS(mp) |= HCK_FULLCKSUM;			\
182 	} else if ((hck_flags) & HCKSUM_INET_PARTIAL)  {		\
183 		/*							\
184 		 * Partial checksum offload has been enabled.  Fill	\
185 		 * the checksum field in the protocl header with the	\
186 		 * pseudo-header checksum value.			\
187 		 */							\
188 		_sum = ((proto) == IPPROTO_UDP) ?			\
189 		    IP_UDP_CSUM_COMP : IP_TCP_CSUM_COMP;		\
190 		_sum += *(up) + (pseudo);				\
191 		_sum = (_sum & 0xFFFF) + (_sum >> 16);			\
192 		*(up) = (_sum & 0xFFFF) + (_sum >> 16);			\
193 		/*							\
194 		 * Offsets are relative to beginning of IP header.	\
195 		 */							\
196 		DB_CKSUMSTART(mp) = (start);				\
197 		DB_CKSUMSTUFF(mp) = ((proto) == IPPROTO_UDP) ?		\
198 		    (start) + UDP_CHECKSUM_OFFSET :			\
199 		    (start) + TCP_CHECKSUM_OFFSET;			\
200 		DB_CKSUMEND(mp) = (end);				\
201 		DB_CKSUMFLAGS(mp) |= HCK_PARTIALCKSUM;			\
202 	} else {							\
203 		/*							\
204 		 * Software checksumming.				\
205 		 */							\
206 		_sum = ((proto) == IPPROTO_UDP) ?			\
207 		    IP_UDP_CSUM_COMP : IP_TCP_CSUM_COMP;		\
208 		_sum += (pseudo);					\
209 		_sum = IP_CSUM(mp, start, _sum);			\
210 		*(up) = (uint16_t)(((proto) == IPPROTO_UDP) ?		\
211 		    (_sum ? _sum : ~_sum) : _sum);			\
212 	}								\
213 	/*								\
214 	 * Hardware supports IP header checksum offload; clear the	\
215 	 * contents of IP header checksum field as expected by NIC.	\
216 	 * Do this only if we offloaded either full or partial sum.	\
217 	 */								\
218 	if ((ipver) == IPV4_VERSION && DB_CKSUMFLAGS(mp) != 0 &&	\
219 	    ((hck_flags) & HCKSUM_IPHDRCKSUM)) {			\
220 		DB_CKSUMFLAGS(mp) |= HCK_IPV4_HDRCKSUM;			\
221 		((ipha_t *)(ihp))->ipha_hdr_checksum = 0;		\
222 	}								\
223 }
224 
225 /*
226  * Macro to inspect the checksum of a fully-reassembled incoming datagram.
227  */
228 #define	IP_CKSUM_RECV_REASS(hck_flags, off, pseudo, sum, err) {		\
229 	(err) = B_FALSE;						\
230 	if ((hck_flags) & HCK_FULLCKSUM) {				\
231 		/*							\
232 		 * The sum of all fragment checksums should		\
233 		 * result in -0 (0xFFFF) or otherwise invalid.		\
234 		 */							\
235 		if ((sum) != 0xFFFF)					\
236 			(err) = B_TRUE;					\
237 	} else if ((hck_flags) & HCK_PARTIALCKSUM) {			\
238 		(sum) += (pseudo);					\
239 		(sum) = ((sum) & 0xFFFF) + ((sum) >> 16);		\
240 		(sum) = ((sum) & 0xFFFF) + ((sum) >> 16);		\
241 		if (~(sum) & 0xFFFF)					\
242 			(err) = B_TRUE;					\
243 	} else if (((sum) = IP_CSUM(mp, off, pseudo)) != 0) {		\
244 		(err) = B_TRUE;						\
245 	}								\
246 }
247 
248 /*
249  * This macro inspects an incoming packet to see if the checksum value
250  * contained in it is valid; if the hardware has provided the information,
251  * the value is verified, otherwise it performs software checksumming.
252  * The checksum value is returned to caller.
253  */
254 #define	IP_CKSUM_RECV(hck_flags, sum, cksum_start, ulph_off, mp, mp1, err) { \
255 	int32_t _len;							\
256 									\
257 	(err) = B_FALSE;						\
258 	if ((hck_flags) & HCK_FULLCKSUM) {				\
259 		/*							\
260 		 * Full checksum has been computed by the hardware	\
261 		 * and has been attached.  If the driver wants us to	\
262 		 * verify the correctness of the attached value, in	\
263 		 * order to protect against faulty hardware, compare	\
264 		 * it against -0 (0xFFFF) to see if it's valid.		\
265 		 */							\
266 		(sum) = DB_CKSUM16(mp);					\
267 		if (!((hck_flags) & HCK_FULLCKSUM_OK) && (sum) != 0xFFFF) \
268 			(err) = B_TRUE;					\
269 	} else if (((hck_flags) & HCK_PARTIALCKSUM) &&			\
270 	    ((mp1) == NULL || (mp1)->b_cont == NULL) &&			\
271 	    (ulph_off) >= DB_CKSUMSTART(mp) &&				\
272 	    ((_len = (ulph_off) - DB_CKSUMSTART(mp)) & 1) == 0) {	\
273 		uint32_t _adj;						\
274 		/*							\
275 		 * Partial checksum has been calculated by hardware	\
276 		 * and attached to the packet; in addition, any		\
277 		 * prepended extraneous data is even byte aligned,	\
278 		 * and there are at most two mblks associated with	\
279 		 * the packet.  If any such data exists, we adjust	\
280 		 * the checksum; also take care any postpended data.	\
281 		 */							\
282 		IP_ADJCKSUM_PARTIAL(cksum_start, mp, mp1, _len, _adj);	\
283 		/*							\
284 		 * One's complement subtract extraneous checksum	\
285 		 */							\
286 		(sum) += DB_CKSUM16(mp);				\
287 		if (_adj >= (sum))					\
288 			(sum) = ~(_adj - (sum)) & 0xFFFF;		\
289 		else							\
290 			(sum) -= _adj;					\
291 		(sum) = ((sum) & 0xFFFF) + ((int)(sum) >> 16);		\
292 		(sum) = ((sum) & 0xFFFF) + ((int)(sum) >> 16);		\
293 		if (~(sum) & 0xFFFF)					\
294 			(err) = B_TRUE;					\
295 	} else if (((sum) = IP_CSUM(mp, ulph_off, sum)) != 0) {		\
296 		(err) = B_TRUE;						\
297 	}								\
298 }
299 
300 /*
301  * Macro to adjust a given checksum value depending on any prepended
302  * or postpended data on the packet.  It expects the start offset to
303  * begin at an even boundary and that the packet consists of at most
304  * two mblks.
305  */
306 #define	IP_ADJCKSUM_PARTIAL(cksum_start, mp, mp1, len, adj) {		\
307 	/*								\
308 	 * Prepended extraneous data; adjust checksum.			\
309 	 */								\
310 	if ((len) > 0)							\
311 		(adj) = IP_BCSUM_PARTIAL(cksum_start, len, 0);		\
312 	else								\
313 		(adj) = 0;						\
314 	/*								\
315 	 * len is now the total length of mblk(s)			\
316 	 */								\
317 	(len) = MBLKL(mp);						\
318 	if ((mp1) == NULL)						\
319 		(mp1) = (mp);						\
320 	else								\
321 		(len) += MBLKL(mp1);					\
322 	/*								\
323 	 * Postpended extraneous data; adjust checksum.			\
324 	 */								\
325 	if (((len) = (DB_CKSUMEND(mp) - len)) > 0) {			\
326 		uint32_t _pad;						\
327 									\
328 		_pad = IP_BCSUM_PARTIAL((mp1)->b_wptr, len, 0);		\
329 		/*							\
330 		 * If the postpended extraneous data was odd		\
331 		 * byte aligned, swap resulting checksum bytes.		\
332 		 */							\
333 		if ((uintptr_t)(mp1)->b_wptr & 1)			\
334 			(adj) += ((_pad << 8) & 0xFFFF) | (_pad >> 8);	\
335 		else							\
336 			(adj) += _pad;					\
337 		(adj) = ((adj) & 0xFFFF) + ((int)(adj) >> 16);		\
338 	}								\
339 }
340 
341 #define	ILL_MDT_CAPABLE(ill)		\
342 	(((ill)->ill_capabilities & ILL_CAPAB_MDT) != 0)
343 
344 /*
345  * ioctl identifier and structure for Multidata Transmit update
346  * private M_CTL communication from IP to ULP.
347  */
348 #define	MDT_IOC_INFO_UPDATE	(('M' << 8) + 1020)
349 
350 typedef struct ip_mdt_info_s {
351 	uint_t	mdt_info_id;	/* MDT_IOC_INFO_UPDATE */
352 	ill_mdt_capab_t	mdt_capab; /* ILL MDT capabilities */
353 } ip_mdt_info_t;
354 
355 /*
356  * Macro that determines whether or not a given ILL is allowed for MDT.
357  */
358 #define	ILL_MDT_USABLE(ill)						\
359 	(ILL_MDT_CAPABLE(ill) &&					\
360 	ill->ill_mdt_capab != NULL &&					\
361 	ill->ill_mdt_capab->ill_mdt_version == MDT_VERSION_2 &&		\
362 	ill->ill_mdt_capab->ill_mdt_on != 0)
363 
364 #define	ILL_LSO_CAPABLE(ill)		\
365 	(((ill)->ill_capabilities & ILL_CAPAB_DLD_LSO) != 0)
366 
367 /*
368  * ioctl identifier and structure for Large Segment Offload
369  * private M_CTL communication from IP to ULP.
370  */
371 #define	LSO_IOC_INFO_UPDATE	(('L' << 24) + ('S' << 16) + ('O' << 8))
372 
373 typedef struct ip_lso_info_s {
374 	uint_t	lso_info_id;	/* LSO_IOC_INFO_UPDATE */
375 	ill_lso_capab_t	lso_capab; /* ILL LSO capabilities */
376 } ip_lso_info_t;
377 
378 /*
379  * Macro that determines whether or not a given ILL is allowed for LSO.
380  */
381 #define	ILL_LSO_USABLE(ill)						\
382 	(ILL_LSO_CAPABLE(ill) &&					\
383 	ill->ill_lso_capab != NULL &&					\
384 	ill->ill_lso_capab->ill_lso_on != 0)
385 
386 #define	ILL_LSO_TCP_USABLE(ill)						\
387 	(ILL_LSO_USABLE(ill) &&						\
388 	ill->ill_lso_capab->ill_lso_flags & DLD_LSO_TX_BASIC_TCP_IPV4)
389 
390 /*
391  * Macro that determines whether or not a given CONN may be considered
392  * for fast path prior to proceeding further with LSO or Multidata.
393  */
394 #define	CONN_IS_LSO_MD_FASTPATH(connp)	\
395 	((connp)->conn_dontroute == 0 &&	/* SO_DONTROUTE */	\
396 	!((connp)->conn_nexthop_set) &&		/* IP_NEXTHOP */	\
397 	(connp)->conn_outgoing_ill == NULL)	/* IP{V6}_BOUND_IF */
398 
399 /* Definitions for fragmenting IP packets using MDT. */
400 
401 /*
402  * Smaller and private version of pdescinfo_t used specifically for IP,
403  * which allows for only a single payload span per packet.
404  */
405 typedef struct ip_pdescinfo_s PDESCINFO_STRUCT(2)	ip_pdescinfo_t;
406 
407 /*
408  * Macro version of ip_can_frag_mdt() which avoids the function call if we
409  * only examine a single message block.
410  */
411 #define	IP_CAN_FRAG_MDT(mp, hdr_len, len)			\
412 	(((mp)->b_cont == NULL) ?				\
413 	(MBLKL(mp) >= ((hdr_len) + ip_wput_frag_mdt_min)) :	\
414 	ip_can_frag_mdt((mp), (hdr_len), (len)))
415 
416 /*
417  * Macro that determines whether or not a given IPC requires
418  * outbound IPSEC processing.
419  */
420 #define	CONN_IPSEC_OUT_ENCAPSULATED(connp)	\
421 	((connp)->conn_out_enforce_policy ||	\
422 	((connp)->conn_latch != NULL &&		\
423 	(connp)->conn_latch->ipl_out_policy != NULL))
424 
425 /*
426  * These are used by the synchronous streams code in tcp and udp.
427  * When we set the flags for a wakeup from a synchronous stream we
428  * always set RSLEEP in sd_wakeq, even if we have a read thread waiting
429  * to do the io. This is in case the read thread gets interrupted
430  * before completing the io. The RSLEEP flag in sd_wakeq is used to
431  * indicate that there is data available at the synchronous barrier.
432  * The assumption is that subsequent functions calls through rwnext()
433  * will reset sd_wakeq appropriately.
434  */
435 #define	STR_WAKEUP_CLEAR(stp) {						\
436 	mutex_enter(&stp->sd_lock);					\
437 	stp->sd_wakeq &= ~RSLEEP;					\
438 	mutex_exit(&stp->sd_lock);					\
439 }
440 
441 #define	STR_WAKEUP_SET(stp) {						\
442 	mutex_enter(&stp->sd_lock);					\
443 	if (stp->sd_flag & RSLEEP) {					\
444 		stp->sd_flag &= ~RSLEEP;				\
445 		cv_broadcast(&_RD(stp->sd_wrq)->q_wait);		\
446 	}								\
447 	stp->sd_wakeq |= RSLEEP;					\
448 	mutex_exit(&stp->sd_lock);					\
449 }
450 
451 /*
452  * Combined wakeup and sendsig to avoid dropping and reacquiring the
453  * sd_lock. The list of messages waiting at the synchronous barrier is
454  * supplied in order to determine whether a wakeup needs to occur. We
455  * only send a wakeup to the application when necessary, i.e. during
456  * the first enqueue when the received messages list will be NULL.
457  */
458 #define	STR_WAKEUP_SENDSIG(stp, rcv_list) {				\
459 	int _events;							\
460 	mutex_enter(&stp->sd_lock);					\
461 	if (rcv_list == NULL) {						\
462 		if (stp->sd_flag & RSLEEP) {				\
463 			stp->sd_flag &= ~RSLEEP;			\
464 			cv_broadcast(&_RD(stp->sd_wrq)->q_wait);	\
465 		}							\
466 		stp->sd_wakeq |= RSLEEP;				\
467 	}								\
468 	if ((_events = stp->sd_sigflags & (S_INPUT | S_RDNORM)) != 0)	\
469 		strsendsig(stp->sd_siglist, _events, 0, 0);		\
470 	if (stp->sd_rput_opt & SR_POLLIN) {				\
471 		stp->sd_rput_opt &= ~SR_POLLIN;				\
472 		mutex_exit(&stp->sd_lock);				\
473 		pollwakeup(&stp->sd_pollist, POLLIN | POLLRDNORM);	\
474 	} else {							\
475 		mutex_exit(&stp->sd_lock);				\
476 	}								\
477 }
478 
479 #define	CONN_UDP_SYNCSTR(connp)						\
480 	(IPCL_IS_UDP(connp) && (connp)->conn_udp->udp_direct_sockfs)
481 
482 /*
483  * Macro that checks whether or not a particular UDP conn is
484  * flow-controlling on the read-side.  If udp module is directly
485  * above ip, check to see if the drain queue is full; note here
486  * that we check this without any lock protection because this
487  * is a coarse granularity inbound flow-control.  If the module
488  * above ip is not udp, then use canputnext to determine the
489  * flow-control.
490  *
491  * Note that these checks are done after the conn is found in
492  * the UDP fanout table.
493  * FIXME? Might be faster to check both udp_drain_qfull and canputnext.
494  */
495 #define	CONN_UDP_FLOWCTLD(connp)					\
496 	(CONN_UDP_SYNCSTR(connp) ?					\
497 	(connp)->conn_udp->udp_drain_qfull :				\
498 	!canputnext((connp)->conn_rq))
499 
500 /* Macro that follows definitions of flags for mac_tx() (see mac_client.h) */
501 #define	IP_DROP_ON_NO_DESC	0x01	/* Equivalent to MAC_DROP_ON_NO_DESC */
502 
503 #define	ILL_DIRECT_CAPABLE(ill)						\
504 	(((ill)->ill_capabilities & ILL_CAPAB_DLD_DIRECT) != 0)
505 
506 #define	ILL_SEND_TX(ill, ire, hint, mp, flag, connp) {			\
507 	if (ILL_DIRECT_CAPABLE(ill) && DB_TYPE(mp) == M_DATA) {		\
508 		ill_dld_direct_t *idd;					\
509 		uintptr_t	cookie;					\
510 		conn_t		*udp_connp = (conn_t *)connp;		\
511 									\
512 		idd = &(ill)->ill_dld_capab->idc_direct;		\
513 		/*							\
514 		 * Send the packet directly to DLD, where it		\
515 		 * may be queued depending on the availability		\
516 		 * of transmit resources at the media layer.		\
517 		 * Ignore the returned value for the time being 	\
518 		 * In future, we may want to take this into		\
519 		 * account and flow control the TCP.			\
520 		 */							\
521 		cookie = idd->idd_tx_df(idd->idd_tx_dh, mp,		\
522 		    (uintptr_t)(hint), flag);				\
523 									\
524 		/*							\
525 		 * non-NULL cookie indicates flow control situation	\
526 		 * and the cookie itself identifies this specific	\
527 		 * Tx ring that is blocked. This cookie is used to	\
528 		 * block the UDP conn that is sending packets over	\
529 		 * this specific Tx ring.				\
530 		 */							\
531 		if ((cookie != NULL) && (udp_connp != NULL) &&		\
532 		    (udp_connp->conn_ulp == IPPROTO_UDP)) {		\
533 			idl_tx_list_t *idl_txl;				\
534 			ip_stack_t *ipst;				\
535 									\
536 			/*						\
537 			 * Flow controlled.				\
538 			 */						\
539 			DTRACE_PROBE2(ill__send__tx__cookie,		\
540 			    uintptr_t, cookie, conn_t *, udp_connp);	\
541 			ipst = udp_connp->conn_netstack->netstack_ip;	\
542 			idl_txl =					\
543 			    &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];\
544 			mutex_enter(&idl_txl->txl_lock);		\
545 			if (udp_connp->conn_direct_blocked ||		\
546 			    (idd->idd_tx_fctl_df(idd->idd_tx_fctl_dh,	\
547 			    cookie) == 0)) {				\
548 				DTRACE_PROBE1(ill__tx__not__blocked,	\
549 				    boolean,				\
550 				    udp_connp->conn_direct_blocked);	\
551 			} else if (idl_txl->txl_cookie != NULL &&	\
552 			    idl_txl->txl_cookie != cookie) {		\
553 				udp_t *udp = udp_connp->conn_udp;	\
554 				udp_stack_t *us = udp->udp_us;		\
555 									\
556 				DTRACE_PROBE2(ill__send__tx__collision,	\
557 				    uintptr_t, cookie,			\
558 				    uintptr_t, idl_txl->txl_cookie);	\
559 				UDP_STAT(us, udp_cookie_coll);		\
560 			} else {					\
561 				udp_connp->conn_direct_blocked = B_TRUE;\
562 				idl_txl->txl_cookie = cookie;		\
563 				conn_drain_insert(udp_connp, idl_txl);	\
564 				DTRACE_PROBE1(ill__send__tx__insert,	\
565 				    conn_t *, udp_connp);		\
566 			}						\
567 			mutex_exit(&idl_txl->txl_lock);			\
568 		}							\
569 	} else {							\
570 		putnext((ire)->ire_stq, mp);				\
571 	}								\
572 }
573 
574 #define	MBLK_RX_FANOUT_SLOWPATH(mp, ipha)				\
575 	(DB_TYPE(mp) != M_DATA || DB_REF(mp) != 1 || !OK_32PTR(ipha) || \
576 	(((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH) >= (mp)->b_wptr))
577 
578 /*
579  * In non-global zone exclusive IP stacks, data structures such as IRE
580  * entries pretend that they're in the global zone.  The following
581  * macro evaluates to the real zoneid instead of a pretend
582  * GLOBAL_ZONEID.
583  */
584 #define	IP_REAL_ZONEID(zoneid, ipst)					\
585 	(((zoneid) == GLOBAL_ZONEID) ?					\
586 	    netstackid_to_zoneid((ipst)->ips_netstack->netstack_stackid) : \
587 	    (zoneid))
588 
589 extern int	ip_wput_frag_mdt_min;
590 extern boolean_t ip_can_frag_mdt(mblk_t *, ssize_t, ssize_t);
591 extern mblk_t   *ip_prepend_zoneid(mblk_t *, zoneid_t, ip_stack_t *);
592 extern void ill_flow_enable(void *, ip_mac_tx_cookie_t);
593 extern zoneid_t	ip_get_zoneid_v4(ipaddr_t, mblk_t *, ip_stack_t *, zoneid_t);
594 extern zoneid_t	ip_get_zoneid_v6(in6_addr_t *, mblk_t *, const ill_t *,
595     ip_stack_t *, zoneid_t);
596 
597 /*
598  * flag passed in by IP based protocols to get a private ip stream with
599  * no conn_t. Note this flag has the same value as SO_FALLBACK
600  */
601 #define	IP_HELPER_STR	SO_FALLBACK
602 
603 #define	IP_MOD_MINPSZ	1
604 #define	IP_MOD_MAXPSZ	INFPSZ
605 #define	IP_MOD_HIWAT	65536
606 #define	IP_MOD_LOWAT	1024
607 
608 #define	DEV_IP	"/devices/pseudo/ip@0:ip"
609 #define	DEV_IP6	"/devices/pseudo/ip6@0:ip6"
610 
611 extern struct kmem_cache  *ip_helper_stream_cache;
612 
613 #endif	/* _KERNEL */
614 
615 #ifdef	__cplusplus
616 }
617 #endif
618 
619 #endif	/* _INET_IP_IMPL_H */
620