1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lnet/lnet/lib-move.c
37  *
38  * Data movement routines
39  */
40 
41 #define DEBUG_SUBSYSTEM S_LNET
42 
43 #include "../../include/linux/lnet/lib-lnet.h"
44 
45 static int local_nid_dist_zero = 1;
46 module_param(local_nid_dist_zero, int, 0444);
47 MODULE_PARM_DESC(local_nid_dist_zero, "Reserved");
48 
49 int
lnet_fail_nid(lnet_nid_t nid,unsigned int threshold)50 lnet_fail_nid(lnet_nid_t nid, unsigned int threshold)
51 {
52 	lnet_test_peer_t  *tp;
53 	struct list_head	*el;
54 	struct list_head	*next;
55 	struct list_head	 cull;
56 
57 	LASSERT(the_lnet.ln_init);
58 
59 	/* NB: use lnet_net_lock(0) to serialize operations on test peers */
60 	if (threshold != 0) {
61 		/* Adding a new entry */
62 		LIBCFS_ALLOC(tp, sizeof(*tp));
63 		if (tp == NULL)
64 			return -ENOMEM;
65 
66 		tp->tp_nid = nid;
67 		tp->tp_threshold = threshold;
68 
69 		lnet_net_lock(0);
70 		list_add_tail(&tp->tp_list, &the_lnet.ln_test_peers);
71 		lnet_net_unlock(0);
72 		return 0;
73 	}
74 
75 	/* removing entries */
76 	INIT_LIST_HEAD(&cull);
77 
78 	lnet_net_lock(0);
79 
80 	list_for_each_safe(el, next, &the_lnet.ln_test_peers) {
81 		tp = list_entry(el, lnet_test_peer_t, tp_list);
82 
83 		if (tp->tp_threshold == 0 ||    /* needs culling anyway */
84 		    nid == LNET_NID_ANY ||       /* removing all entries */
85 		    tp->tp_nid == nid) {	  /* matched this one */
86 			list_del(&tp->tp_list);
87 			list_add(&tp->tp_list, &cull);
88 		}
89 	}
90 
91 	lnet_net_unlock(0);
92 
93 	while (!list_empty(&cull)) {
94 		tp = list_entry(cull.next, lnet_test_peer_t, tp_list);
95 
96 		list_del(&tp->tp_list);
97 		LIBCFS_FREE(tp, sizeof(*tp));
98 	}
99 	return 0;
100 }
101 
102 static int
fail_peer(lnet_nid_t nid,int outgoing)103 fail_peer(lnet_nid_t nid, int outgoing)
104 {
105 	lnet_test_peer_t *tp;
106 	struct list_head       *el;
107 	struct list_head       *next;
108 	struct list_head	cull;
109 	int	       fail = 0;
110 
111 	INIT_LIST_HEAD(&cull);
112 
113 	/* NB: use lnet_net_lock(0) to serialize operations on test peers */
114 	lnet_net_lock(0);
115 
116 	list_for_each_safe(el, next, &the_lnet.ln_test_peers) {
117 		tp = list_entry(el, lnet_test_peer_t, tp_list);
118 
119 		if (tp->tp_threshold == 0) {
120 			/* zombie entry */
121 			if (outgoing) {
122 				/* only cull zombies on outgoing tests,
123 				 * since we may be at interrupt priority on
124 				 * incoming messages. */
125 				list_del(&tp->tp_list);
126 				list_add(&tp->tp_list, &cull);
127 			}
128 			continue;
129 		}
130 
131 		if (tp->tp_nid == LNET_NID_ANY || /* fail every peer */
132 		    nid == tp->tp_nid) {	/* fail this peer */
133 			fail = 1;
134 
135 			if (tp->tp_threshold != LNET_MD_THRESH_INF) {
136 				tp->tp_threshold--;
137 				if (outgoing &&
138 				    tp->tp_threshold == 0) {
139 					/* see above */
140 					list_del(&tp->tp_list);
141 					list_add(&tp->tp_list, &cull);
142 				}
143 			}
144 			break;
145 		}
146 	}
147 
148 	lnet_net_unlock(0);
149 
150 	while (!list_empty(&cull)) {
151 		tp = list_entry(cull.next, lnet_test_peer_t, tp_list);
152 		list_del(&tp->tp_list);
153 
154 		LIBCFS_FREE(tp, sizeof(*tp));
155 	}
156 
157 	return fail;
158 }
159 
160 unsigned int
lnet_iov_nob(unsigned int niov,struct kvec * iov)161 lnet_iov_nob(unsigned int niov, struct kvec *iov)
162 {
163 	unsigned int nob = 0;
164 
165 	while (niov-- > 0)
166 		nob += (iov++)->iov_len;
167 
168 	return nob;
169 }
170 EXPORT_SYMBOL(lnet_iov_nob);
171 
172 void
lnet_copy_iov2iov(unsigned int ndiov,struct kvec * diov,unsigned int doffset,unsigned int nsiov,struct kvec * siov,unsigned int soffset,unsigned int nob)173 lnet_copy_iov2iov(unsigned int ndiov, struct kvec *diov, unsigned int doffset,
174 		   unsigned int nsiov, struct kvec *siov, unsigned int soffset,
175 		   unsigned int nob)
176 {
177 	/* NB diov, siov are READ-ONLY */
178 	unsigned int  this_nob;
179 
180 	if (nob == 0)
181 		return;
182 
183 	/* skip complete frags before 'doffset' */
184 	LASSERT(ndiov > 0);
185 	while (doffset >= diov->iov_len) {
186 		doffset -= diov->iov_len;
187 		diov++;
188 		ndiov--;
189 		LASSERT(ndiov > 0);
190 	}
191 
192 	/* skip complete frags before 'soffset' */
193 	LASSERT(nsiov > 0);
194 	while (soffset >= siov->iov_len) {
195 		soffset -= siov->iov_len;
196 		siov++;
197 		nsiov--;
198 		LASSERT(nsiov > 0);
199 	}
200 
201 	do {
202 		LASSERT(ndiov > 0);
203 		LASSERT(nsiov > 0);
204 		this_nob = min(diov->iov_len - doffset,
205 			       siov->iov_len - soffset);
206 		this_nob = min(this_nob, nob);
207 
208 		memcpy((char *)diov->iov_base + doffset,
209 			(char *)siov->iov_base + soffset, this_nob);
210 		nob -= this_nob;
211 
212 		if (diov->iov_len > doffset + this_nob) {
213 			doffset += this_nob;
214 		} else {
215 			diov++;
216 			ndiov--;
217 			doffset = 0;
218 		}
219 
220 		if (siov->iov_len > soffset + this_nob) {
221 			soffset += this_nob;
222 		} else {
223 			siov++;
224 			nsiov--;
225 			soffset = 0;
226 		}
227 	} while (nob > 0);
228 }
229 EXPORT_SYMBOL(lnet_copy_iov2iov);
230 
231 int
lnet_extract_iov(int dst_niov,struct kvec * dst,int src_niov,struct kvec * src,unsigned int offset,unsigned int len)232 lnet_extract_iov(int dst_niov, struct kvec *dst,
233 		  int src_niov, struct kvec *src,
234 		  unsigned int offset, unsigned int len)
235 {
236 	/* Initialise 'dst' to the subset of 'src' starting at 'offset',
237 	 * for exactly 'len' bytes, and return the number of entries.
238 	 * NB not destructive to 'src' */
239 	unsigned int    frag_len;
240 	unsigned int    niov;
241 
242 	if (len == 0)			   /* no data => */
243 		return 0;		     /* no frags */
244 
245 	LASSERT(src_niov > 0);
246 	while (offset >= src->iov_len) {      /* skip initial frags */
247 		offset -= src->iov_len;
248 		src_niov--;
249 		src++;
250 		LASSERT(src_niov > 0);
251 	}
252 
253 	niov = 1;
254 	for (;;) {
255 		LASSERT(src_niov > 0);
256 		LASSERT((int)niov <= dst_niov);
257 
258 		frag_len = src->iov_len - offset;
259 		dst->iov_base = ((char *)src->iov_base) + offset;
260 
261 		if (len <= frag_len) {
262 			dst->iov_len = len;
263 			return niov;
264 		}
265 
266 		dst->iov_len = frag_len;
267 
268 		len -= frag_len;
269 		dst++;
270 		src++;
271 		niov++;
272 		src_niov--;
273 		offset = 0;
274 	}
275 }
276 EXPORT_SYMBOL(lnet_extract_iov);
277 
278 
279 unsigned int
lnet_kiov_nob(unsigned int niov,lnet_kiov_t * kiov)280 lnet_kiov_nob(unsigned int niov, lnet_kiov_t *kiov)
281 {
282 	unsigned int  nob = 0;
283 
284 	while (niov-- > 0)
285 		nob += (kiov++)->kiov_len;
286 
287 	return nob;
288 }
289 EXPORT_SYMBOL(lnet_kiov_nob);
290 
291 void
lnet_copy_kiov2kiov(unsigned int ndiov,lnet_kiov_t * diov,unsigned int doffset,unsigned int nsiov,lnet_kiov_t * siov,unsigned int soffset,unsigned int nob)292 lnet_copy_kiov2kiov(unsigned int ndiov, lnet_kiov_t *diov, unsigned int doffset,
293 		    unsigned int nsiov, lnet_kiov_t *siov, unsigned int soffset,
294 		    unsigned int nob)
295 {
296 	/* NB diov, siov are READ-ONLY */
297 	unsigned int    this_nob;
298 	char	   *daddr = NULL;
299 	char	   *saddr = NULL;
300 
301 	if (nob == 0)
302 		return;
303 
304 	LASSERT(!in_interrupt());
305 
306 	LASSERT(ndiov > 0);
307 	while (doffset >= diov->kiov_len) {
308 		doffset -= diov->kiov_len;
309 		diov++;
310 		ndiov--;
311 		LASSERT(ndiov > 0);
312 	}
313 
314 	LASSERT(nsiov > 0);
315 	while (soffset >= siov->kiov_len) {
316 		soffset -= siov->kiov_len;
317 		siov++;
318 		nsiov--;
319 		LASSERT(nsiov > 0);
320 	}
321 
322 	do {
323 		LASSERT(ndiov > 0);
324 		LASSERT(nsiov > 0);
325 		this_nob = min(diov->kiov_len - doffset,
326 			       siov->kiov_len - soffset);
327 		this_nob = min(this_nob, nob);
328 
329 		if (daddr == NULL)
330 			daddr = ((char *)kmap(diov->kiov_page)) +
331 				diov->kiov_offset + doffset;
332 		if (saddr == NULL)
333 			saddr = ((char *)kmap(siov->kiov_page)) +
334 				siov->kiov_offset + soffset;
335 
336 		/* Vanishing risk of kmap deadlock when mapping 2 pages.
337 		 * However in practice at least one of the kiovs will be mapped
338 		 * kernel pages and the map/unmap will be NOOPs */
339 
340 		memcpy(daddr, saddr, this_nob);
341 		nob -= this_nob;
342 
343 		if (diov->kiov_len > doffset + this_nob) {
344 			daddr += this_nob;
345 			doffset += this_nob;
346 		} else {
347 			kunmap(diov->kiov_page);
348 			daddr = NULL;
349 			diov++;
350 			ndiov--;
351 			doffset = 0;
352 		}
353 
354 		if (siov->kiov_len > soffset + this_nob) {
355 			saddr += this_nob;
356 			soffset += this_nob;
357 		} else {
358 			kunmap(siov->kiov_page);
359 			saddr = NULL;
360 			siov++;
361 			nsiov--;
362 			soffset = 0;
363 		}
364 	} while (nob > 0);
365 
366 	if (daddr != NULL)
367 		kunmap(diov->kiov_page);
368 	if (saddr != NULL)
369 		kunmap(siov->kiov_page);
370 }
371 EXPORT_SYMBOL(lnet_copy_kiov2kiov);
372 
373 void
lnet_copy_kiov2iov(unsigned int niov,struct kvec * iov,unsigned int iovoffset,unsigned int nkiov,lnet_kiov_t * kiov,unsigned int kiovoffset,unsigned int nob)374 lnet_copy_kiov2iov(unsigned int niov, struct kvec *iov, unsigned int iovoffset,
375 		   unsigned int nkiov, lnet_kiov_t *kiov,
376 		   unsigned int kiovoffset, unsigned int nob)
377 {
378 	/* NB iov, kiov are READ-ONLY */
379 	unsigned int    this_nob;
380 	char	   *addr = NULL;
381 
382 	if (nob == 0)
383 		return;
384 
385 	LASSERT(!in_interrupt());
386 
387 	LASSERT(niov > 0);
388 	while (iovoffset >= iov->iov_len) {
389 		iovoffset -= iov->iov_len;
390 		iov++;
391 		niov--;
392 		LASSERT(niov > 0);
393 	}
394 
395 	LASSERT(nkiov > 0);
396 	while (kiovoffset >= kiov->kiov_len) {
397 		kiovoffset -= kiov->kiov_len;
398 		kiov++;
399 		nkiov--;
400 		LASSERT(nkiov > 0);
401 	}
402 
403 	do {
404 		LASSERT(niov > 0);
405 		LASSERT(nkiov > 0);
406 		this_nob = min(iov->iov_len - iovoffset,
407 			       (__kernel_size_t) kiov->kiov_len - kiovoffset);
408 		this_nob = min(this_nob, nob);
409 
410 		if (addr == NULL)
411 			addr = ((char *)kmap(kiov->kiov_page)) +
412 				kiov->kiov_offset + kiovoffset;
413 
414 		memcpy((char *)iov->iov_base + iovoffset, addr, this_nob);
415 		nob -= this_nob;
416 
417 		if (iov->iov_len > iovoffset + this_nob) {
418 			iovoffset += this_nob;
419 		} else {
420 			iov++;
421 			niov--;
422 			iovoffset = 0;
423 		}
424 
425 		if (kiov->kiov_len > kiovoffset + this_nob) {
426 			addr += this_nob;
427 			kiovoffset += this_nob;
428 		} else {
429 			kunmap(kiov->kiov_page);
430 			addr = NULL;
431 			kiov++;
432 			nkiov--;
433 			kiovoffset = 0;
434 		}
435 
436 	} while (nob > 0);
437 
438 	if (addr != NULL)
439 		kunmap(kiov->kiov_page);
440 }
441 EXPORT_SYMBOL(lnet_copy_kiov2iov);
442 
443 void
lnet_copy_iov2kiov(unsigned int nkiov,lnet_kiov_t * kiov,unsigned int kiovoffset,unsigned int niov,struct kvec * iov,unsigned int iovoffset,unsigned int nob)444 lnet_copy_iov2kiov(unsigned int nkiov, lnet_kiov_t *kiov,
445 		   unsigned int kiovoffset, unsigned int niov,
446 		   struct kvec *iov, unsigned int iovoffset,
447 		   unsigned int nob)
448 {
449 	/* NB kiov, iov are READ-ONLY */
450 	unsigned int    this_nob;
451 	char	   *addr = NULL;
452 
453 	if (nob == 0)
454 		return;
455 
456 	LASSERT(!in_interrupt());
457 
458 	LASSERT(nkiov > 0);
459 	while (kiovoffset >= kiov->kiov_len) {
460 		kiovoffset -= kiov->kiov_len;
461 		kiov++;
462 		nkiov--;
463 		LASSERT(nkiov > 0);
464 	}
465 
466 	LASSERT(niov > 0);
467 	while (iovoffset >= iov->iov_len) {
468 		iovoffset -= iov->iov_len;
469 		iov++;
470 		niov--;
471 		LASSERT(niov > 0);
472 	}
473 
474 	do {
475 		LASSERT(nkiov > 0);
476 		LASSERT(niov > 0);
477 		this_nob = min((__kernel_size_t) kiov->kiov_len - kiovoffset,
478 			       iov->iov_len - iovoffset);
479 		this_nob = min(this_nob, nob);
480 
481 		if (addr == NULL)
482 			addr = ((char *)kmap(kiov->kiov_page)) +
483 				kiov->kiov_offset + kiovoffset;
484 
485 		memcpy(addr, (char *)iov->iov_base + iovoffset, this_nob);
486 		nob -= this_nob;
487 
488 		if (kiov->kiov_len > kiovoffset + this_nob) {
489 			addr += this_nob;
490 			kiovoffset += this_nob;
491 		} else {
492 			kunmap(kiov->kiov_page);
493 			addr = NULL;
494 			kiov++;
495 			nkiov--;
496 			kiovoffset = 0;
497 		}
498 
499 		if (iov->iov_len > iovoffset + this_nob) {
500 			iovoffset += this_nob;
501 		} else {
502 			iov++;
503 			niov--;
504 			iovoffset = 0;
505 		}
506 	} while (nob > 0);
507 
508 	if (addr != NULL)
509 		kunmap(kiov->kiov_page);
510 }
511 EXPORT_SYMBOL(lnet_copy_iov2kiov);
512 
513 int
lnet_extract_kiov(int dst_niov,lnet_kiov_t * dst,int src_niov,lnet_kiov_t * src,unsigned int offset,unsigned int len)514 lnet_extract_kiov(int dst_niov, lnet_kiov_t *dst,
515 		   int src_niov, lnet_kiov_t *src,
516 		   unsigned int offset, unsigned int len)
517 {
518 	/* Initialise 'dst' to the subset of 'src' starting at 'offset',
519 	 * for exactly 'len' bytes, and return the number of entries.
520 	 * NB not destructive to 'src' */
521 	unsigned int    frag_len;
522 	unsigned int    niov;
523 
524 	if (len == 0)			   /* no data => */
525 		return 0;		     /* no frags */
526 
527 	LASSERT(src_niov > 0);
528 	while (offset >= src->kiov_len) {      /* skip initial frags */
529 		offset -= src->kiov_len;
530 		src_niov--;
531 		src++;
532 		LASSERT(src_niov > 0);
533 	}
534 
535 	niov = 1;
536 	for (;;) {
537 		LASSERT(src_niov > 0);
538 		LASSERT((int)niov <= dst_niov);
539 
540 		frag_len = src->kiov_len - offset;
541 		dst->kiov_page = src->kiov_page;
542 		dst->kiov_offset = src->kiov_offset + offset;
543 
544 		if (len <= frag_len) {
545 			dst->kiov_len = len;
546 			LASSERT(dst->kiov_offset + dst->kiov_len
547 					     <= PAGE_CACHE_SIZE);
548 			return niov;
549 		}
550 
551 		dst->kiov_len = frag_len;
552 		LASSERT(dst->kiov_offset + dst->kiov_len <= PAGE_CACHE_SIZE);
553 
554 		len -= frag_len;
555 		dst++;
556 		src++;
557 		niov++;
558 		src_niov--;
559 		offset = 0;
560 	}
561 }
562 EXPORT_SYMBOL(lnet_extract_kiov);
563 
564 static void
lnet_ni_recv(lnet_ni_t * ni,void * private,lnet_msg_t * msg,int delayed,unsigned int offset,unsigned int mlen,unsigned int rlen)565 lnet_ni_recv(lnet_ni_t *ni, void *private, lnet_msg_t *msg, int delayed,
566 	     unsigned int offset, unsigned int mlen, unsigned int rlen)
567 {
568 	unsigned int  niov = 0;
569 	struct kvec *iov = NULL;
570 	lnet_kiov_t  *kiov = NULL;
571 	int	   rc;
572 
573 	LASSERT(!in_interrupt());
574 	LASSERT(mlen == 0 || msg != NULL);
575 
576 	if (msg != NULL) {
577 		LASSERT(msg->msg_receiving);
578 		LASSERT(!msg->msg_sending);
579 		LASSERT(rlen == msg->msg_len);
580 		LASSERT(mlen <= msg->msg_len);
581 		LASSERT(msg->msg_offset == offset);
582 		LASSERT(msg->msg_wanted == mlen);
583 
584 		msg->msg_receiving = 0;
585 
586 		if (mlen != 0) {
587 			niov = msg->msg_niov;
588 			iov  = msg->msg_iov;
589 			kiov = msg->msg_kiov;
590 
591 			LASSERT(niov > 0);
592 			LASSERT((iov == NULL) != (kiov == NULL));
593 		}
594 	}
595 
596 	rc = (ni->ni_lnd->lnd_recv)(ni, private, msg, delayed,
597 				    niov, iov, kiov, offset, mlen, rlen);
598 	if (rc < 0)
599 		lnet_finalize(ni, msg, rc);
600 }
601 
602 static void
lnet_setpayloadbuffer(lnet_msg_t * msg)603 lnet_setpayloadbuffer(lnet_msg_t *msg)
604 {
605 	lnet_libmd_t *md = msg->msg_md;
606 
607 	LASSERT(msg->msg_len > 0);
608 	LASSERT(!msg->msg_routing);
609 	LASSERT(md != NULL);
610 	LASSERT(msg->msg_niov == 0);
611 	LASSERT(msg->msg_iov == NULL);
612 	LASSERT(msg->msg_kiov == NULL);
613 
614 	msg->msg_niov = md->md_niov;
615 	if ((md->md_options & LNET_MD_KIOV) != 0)
616 		msg->msg_kiov = md->md_iov.kiov;
617 	else
618 		msg->msg_iov = md->md_iov.iov;
619 }
620 
621 void
lnet_prep_send(lnet_msg_t * msg,int type,lnet_process_id_t target,unsigned int offset,unsigned int len)622 lnet_prep_send(lnet_msg_t *msg, int type, lnet_process_id_t target,
623 	       unsigned int offset, unsigned int len)
624 {
625 	msg->msg_type = type;
626 	msg->msg_target = target;
627 	msg->msg_len = len;
628 	msg->msg_offset = offset;
629 
630 	if (len != 0)
631 		lnet_setpayloadbuffer(msg);
632 
633 	memset(&msg->msg_hdr, 0, sizeof(msg->msg_hdr));
634 	msg->msg_hdr.type	   = cpu_to_le32(type);
635 	msg->msg_hdr.dest_nid       = cpu_to_le64(target.nid);
636 	msg->msg_hdr.dest_pid       = cpu_to_le32(target.pid);
637 	/* src_nid will be set later */
638 	msg->msg_hdr.src_pid	= cpu_to_le32(the_lnet.ln_pid);
639 	msg->msg_hdr.payload_length = cpu_to_le32(len);
640 }
641 
642 static void
lnet_ni_send(lnet_ni_t * ni,lnet_msg_t * msg)643 lnet_ni_send(lnet_ni_t *ni, lnet_msg_t *msg)
644 {
645 	void   *priv = msg->msg_private;
646 	int     rc;
647 
648 	LASSERT(!in_interrupt());
649 	LASSERT(LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND ||
650 		 (msg->msg_txcredit && msg->msg_peertxcredit));
651 
652 	rc = (ni->ni_lnd->lnd_send)(ni, priv, msg);
653 	if (rc < 0)
654 		lnet_finalize(ni, msg, rc);
655 }
656 
657 static int
lnet_ni_eager_recv(lnet_ni_t * ni,lnet_msg_t * msg)658 lnet_ni_eager_recv(lnet_ni_t *ni, lnet_msg_t *msg)
659 {
660 	int	rc;
661 
662 	LASSERT(!msg->msg_sending);
663 	LASSERT(msg->msg_receiving);
664 	LASSERT(!msg->msg_rx_ready_delay);
665 	LASSERT(ni->ni_lnd->lnd_eager_recv != NULL);
666 
667 	msg->msg_rx_ready_delay = 1;
668 	rc = (ni->ni_lnd->lnd_eager_recv)(ni, msg->msg_private, msg,
669 					  &msg->msg_private);
670 	if (rc != 0) {
671 		CERROR("recv from %s / send to %s aborted: eager_recv failed %d\n",
672 		       libcfs_nid2str(msg->msg_rxpeer->lp_nid),
673 		       libcfs_id2str(msg->msg_target), rc);
674 		LASSERT(rc < 0); /* required by my callers */
675 	}
676 
677 	return rc;
678 }
679 
680 /* NB: caller shall hold a ref on 'lp' as I'd drop lnet_net_lock */
681 static void
lnet_ni_query_locked(lnet_ni_t * ni,lnet_peer_t * lp)682 lnet_ni_query_locked(lnet_ni_t *ni, lnet_peer_t *lp)
683 {
684 	unsigned long last_alive = 0;
685 
686 	LASSERT(lnet_peer_aliveness_enabled(lp));
687 	LASSERT(ni->ni_lnd->lnd_query != NULL);
688 
689 	lnet_net_unlock(lp->lp_cpt);
690 	(ni->ni_lnd->lnd_query)(ni, lp->lp_nid, &last_alive);
691 	lnet_net_lock(lp->lp_cpt);
692 
693 	lp->lp_last_query = cfs_time_current();
694 
695 	if (last_alive != 0) /* NI has updated timestamp */
696 		lp->lp_last_alive = last_alive;
697 }
698 
699 /* NB: always called with lnet_net_lock held */
700 static inline int
lnet_peer_is_alive(lnet_peer_t * lp,unsigned long now)701 lnet_peer_is_alive(lnet_peer_t *lp, unsigned long now)
702 {
703 	int	alive;
704 	unsigned long deadline;
705 
706 	LASSERT(lnet_peer_aliveness_enabled(lp));
707 
708 	/* Trust lnet_notify() if it has more recent aliveness news, but
709 	 * ignore the initial assumed death (see lnet_peers_start_down()).
710 	 */
711 	if (!lp->lp_alive && lp->lp_alive_count > 0 &&
712 	    cfs_time_aftereq(lp->lp_timestamp, lp->lp_last_alive))
713 		return 0;
714 
715 	deadline = cfs_time_add(lp->lp_last_alive,
716 				cfs_time_seconds(lp->lp_ni->ni_peertimeout));
717 	alive = cfs_time_after(deadline, now);
718 
719 	/* Update obsolete lp_alive except for routers assumed to be dead
720 	 * initially, because router checker would update aliveness in this
721 	 * case, and moreover lp_last_alive at peer creation is assumed.
722 	 */
723 	if (alive && !lp->lp_alive &&
724 	    !(lnet_isrouter(lp) && lp->lp_alive_count == 0))
725 		lnet_notify_locked(lp, 0, 1, lp->lp_last_alive);
726 
727 	return alive;
728 }
729 
730 
731 /* NB: returns 1 when alive, 0 when dead, negative when error;
732  *     may drop the lnet_net_lock */
733 static int
lnet_peer_alive_locked(lnet_peer_t * lp)734 lnet_peer_alive_locked(lnet_peer_t *lp)
735 {
736 	unsigned long now = cfs_time_current();
737 
738 	if (!lnet_peer_aliveness_enabled(lp))
739 		return -ENODEV;
740 
741 	if (lnet_peer_is_alive(lp, now))
742 		return 1;
743 
744 	/* Peer appears dead, but we should avoid frequent NI queries (at
745 	 * most once per lnet_queryinterval seconds). */
746 	if (lp->lp_last_query != 0) {
747 		static const int lnet_queryinterval = 1;
748 
749 		unsigned long next_query =
750 			   cfs_time_add(lp->lp_last_query,
751 					cfs_time_seconds(lnet_queryinterval));
752 
753 		if (time_before(now, next_query)) {
754 			if (lp->lp_alive)
755 				CWARN("Unexpected aliveness of peer %s: %d < %d (%d/%d)\n",
756 				      libcfs_nid2str(lp->lp_nid),
757 				      (int)now, (int)next_query,
758 				      lnet_queryinterval,
759 				      lp->lp_ni->ni_peertimeout);
760 			return 0;
761 		}
762 	}
763 
764 	/* query NI for latest aliveness news */
765 	lnet_ni_query_locked(lp->lp_ni, lp);
766 
767 	if (lnet_peer_is_alive(lp, now))
768 		return 1;
769 
770 	lnet_notify_locked(lp, 0, 0, lp->lp_last_alive);
771 	return 0;
772 }
773 
774 /**
775  * \param msg The message to be sent.
776  * \param do_send True if lnet_ni_send() should be called in this function.
777  *	  lnet_send() is going to lnet_net_unlock immediately after this, so
778  *	  it sets do_send FALSE and I don't do the unlock/send/lock bit.
779  *
780  * \retval 0 If \a msg sent or OK to send.
781  * \retval EAGAIN If \a msg blocked for credit.
782  * \retval EHOSTUNREACH If the next hop of the message appears dead.
783  * \retval ECANCELED If the MD of the message has been unlinked.
784  */
785 static int
lnet_post_send_locked(lnet_msg_t * msg,int do_send)786 lnet_post_send_locked(lnet_msg_t *msg, int do_send)
787 {
788 	lnet_peer_t		*lp = msg->msg_txpeer;
789 	lnet_ni_t		*ni = lp->lp_ni;
790 	int			cpt = msg->msg_tx_cpt;
791 	struct lnet_tx_queue	*tq = ni->ni_tx_queues[cpt];
792 
793 	/* non-lnet_send() callers have checked before */
794 	LASSERT(!do_send || msg->msg_tx_delayed);
795 	LASSERT(!msg->msg_receiving);
796 	LASSERT(msg->msg_tx_committed);
797 
798 	/* NB 'lp' is always the next hop */
799 	if ((msg->msg_target.pid & LNET_PID_USERFLAG) == 0 &&
800 	    lnet_peer_alive_locked(lp) == 0) {
801 		the_lnet.ln_counters[cpt]->drop_count++;
802 		the_lnet.ln_counters[cpt]->drop_length += msg->msg_len;
803 		lnet_net_unlock(cpt);
804 
805 		CNETERR("Dropping message for %s: peer not alive\n",
806 			libcfs_id2str(msg->msg_target));
807 		if (do_send)
808 			lnet_finalize(ni, msg, -EHOSTUNREACH);
809 
810 		lnet_net_lock(cpt);
811 		return EHOSTUNREACH;
812 	}
813 
814 	if (msg->msg_md != NULL &&
815 	    (msg->msg_md->md_flags & LNET_MD_FLAG_ABORTED) != 0) {
816 		lnet_net_unlock(cpt);
817 
818 		CNETERR("Aborting message for %s: LNetM[DE]Unlink() already called on the MD/ME.\n",
819 			libcfs_id2str(msg->msg_target));
820 		if (do_send)
821 			lnet_finalize(ni, msg, -ECANCELED);
822 
823 		lnet_net_lock(cpt);
824 		return ECANCELED;
825 	}
826 
827 	if (!msg->msg_peertxcredit) {
828 		LASSERT((lp->lp_txcredits < 0) ==
829 			 !list_empty(&lp->lp_txq));
830 
831 		msg->msg_peertxcredit = 1;
832 		lp->lp_txqnob += msg->msg_len + sizeof(lnet_hdr_t);
833 		lp->lp_txcredits--;
834 
835 		if (lp->lp_txcredits < lp->lp_mintxcredits)
836 			lp->lp_mintxcredits = lp->lp_txcredits;
837 
838 		if (lp->lp_txcredits < 0) {
839 			msg->msg_tx_delayed = 1;
840 			list_add_tail(&msg->msg_list, &lp->lp_txq);
841 			return EAGAIN;
842 		}
843 	}
844 
845 	if (!msg->msg_txcredit) {
846 		LASSERT((tq->tq_credits < 0) ==
847 			!list_empty(&tq->tq_delayed));
848 
849 		msg->msg_txcredit = 1;
850 		tq->tq_credits--;
851 
852 		if (tq->tq_credits < tq->tq_credits_min)
853 			tq->tq_credits_min = tq->tq_credits;
854 
855 		if (tq->tq_credits < 0) {
856 			msg->msg_tx_delayed = 1;
857 			list_add_tail(&msg->msg_list, &tq->tq_delayed);
858 			return EAGAIN;
859 		}
860 	}
861 
862 	if (do_send) {
863 		lnet_net_unlock(cpt);
864 		lnet_ni_send(ni, msg);
865 		lnet_net_lock(cpt);
866 	}
867 	return 0;
868 }
869 
870 
871 static lnet_rtrbufpool_t *
lnet_msg2bufpool(lnet_msg_t * msg)872 lnet_msg2bufpool(lnet_msg_t *msg)
873 {
874 	lnet_rtrbufpool_t	*rbp;
875 	int			cpt;
876 
877 	LASSERT(msg->msg_rx_committed);
878 
879 	cpt = msg->msg_rx_cpt;
880 	rbp = &the_lnet.ln_rtrpools[cpt][0];
881 
882 	LASSERT(msg->msg_len <= LNET_MTU);
883 	while (msg->msg_len > (unsigned int)rbp->rbp_npages * PAGE_CACHE_SIZE) {
884 		rbp++;
885 		LASSERT(rbp < &the_lnet.ln_rtrpools[cpt][LNET_NRBPOOLS]);
886 	}
887 
888 	return rbp;
889 }
890 
891 static int
lnet_post_routed_recv_locked(lnet_msg_t * msg,int do_recv)892 lnet_post_routed_recv_locked(lnet_msg_t *msg, int do_recv)
893 {
894 	/* lnet_parse is going to lnet_net_unlock immediately after this, so it
895 	 * sets do_recv FALSE and I don't do the unlock/send/lock bit.  I
896 	 * return EAGAIN if msg blocked and 0 if received or OK to receive */
897 	lnet_peer_t	 *lp = msg->msg_rxpeer;
898 	lnet_rtrbufpool_t   *rbp;
899 	lnet_rtrbuf_t       *rb;
900 
901 	LASSERT(msg->msg_iov == NULL);
902 	LASSERT(msg->msg_kiov == NULL);
903 	LASSERT(msg->msg_niov == 0);
904 	LASSERT(msg->msg_routing);
905 	LASSERT(msg->msg_receiving);
906 	LASSERT(!msg->msg_sending);
907 
908 	/* non-lnet_parse callers only receive delayed messages */
909 	LASSERT(!do_recv || msg->msg_rx_delayed);
910 
911 	if (!msg->msg_peerrtrcredit) {
912 		LASSERT((lp->lp_rtrcredits < 0) ==
913 			 !list_empty(&lp->lp_rtrq));
914 
915 		msg->msg_peerrtrcredit = 1;
916 		lp->lp_rtrcredits--;
917 		if (lp->lp_rtrcredits < lp->lp_minrtrcredits)
918 			lp->lp_minrtrcredits = lp->lp_rtrcredits;
919 
920 		if (lp->lp_rtrcredits < 0) {
921 			/* must have checked eager_recv before here */
922 			LASSERT(msg->msg_rx_ready_delay);
923 			msg->msg_rx_delayed = 1;
924 			list_add_tail(&msg->msg_list, &lp->lp_rtrq);
925 			return EAGAIN;
926 		}
927 	}
928 
929 	rbp = lnet_msg2bufpool(msg);
930 
931 	if (!msg->msg_rtrcredit) {
932 		LASSERT((rbp->rbp_credits < 0) ==
933 			 !list_empty(&rbp->rbp_msgs));
934 
935 		msg->msg_rtrcredit = 1;
936 		rbp->rbp_credits--;
937 		if (rbp->rbp_credits < rbp->rbp_mincredits)
938 			rbp->rbp_mincredits = rbp->rbp_credits;
939 
940 		if (rbp->rbp_credits < 0) {
941 			/* must have checked eager_recv before here */
942 			LASSERT(msg->msg_rx_ready_delay);
943 			msg->msg_rx_delayed = 1;
944 			list_add_tail(&msg->msg_list, &rbp->rbp_msgs);
945 			return EAGAIN;
946 		}
947 	}
948 
949 	LASSERT(!list_empty(&rbp->rbp_bufs));
950 	rb = list_entry(rbp->rbp_bufs.next, lnet_rtrbuf_t, rb_list);
951 	list_del(&rb->rb_list);
952 
953 	msg->msg_niov = rbp->rbp_npages;
954 	msg->msg_kiov = &rb->rb_kiov[0];
955 
956 	if (do_recv) {
957 		int cpt = msg->msg_rx_cpt;
958 
959 		lnet_net_unlock(cpt);
960 		lnet_ni_recv(lp->lp_ni, msg->msg_private, msg, 1,
961 			     0, msg->msg_len, msg->msg_len);
962 		lnet_net_lock(cpt);
963 	}
964 	return 0;
965 }
966 
967 void
lnet_return_tx_credits_locked(lnet_msg_t * msg)968 lnet_return_tx_credits_locked(lnet_msg_t *msg)
969 {
970 	lnet_peer_t	*txpeer = msg->msg_txpeer;
971 	lnet_msg_t	*msg2;
972 
973 	if (msg->msg_txcredit) {
974 		struct lnet_ni	     *ni = txpeer->lp_ni;
975 		struct lnet_tx_queue *tq = ni->ni_tx_queues[msg->msg_tx_cpt];
976 
977 		/* give back NI txcredits */
978 		msg->msg_txcredit = 0;
979 
980 		LASSERT((tq->tq_credits < 0) ==
981 			!list_empty(&tq->tq_delayed));
982 
983 		tq->tq_credits++;
984 		if (tq->tq_credits <= 0) {
985 			msg2 = list_entry(tq->tq_delayed.next,
986 					      lnet_msg_t, msg_list);
987 			list_del(&msg2->msg_list);
988 
989 			LASSERT(msg2->msg_txpeer->lp_ni == ni);
990 			LASSERT(msg2->msg_tx_delayed);
991 
992 			(void) lnet_post_send_locked(msg2, 1);
993 		}
994 	}
995 
996 	if (msg->msg_peertxcredit) {
997 		/* give back peer txcredits */
998 		msg->msg_peertxcredit = 0;
999 
1000 		LASSERT((txpeer->lp_txcredits < 0) ==
1001 			!list_empty(&txpeer->lp_txq));
1002 
1003 		txpeer->lp_txqnob -= msg->msg_len + sizeof(lnet_hdr_t);
1004 		LASSERT(txpeer->lp_txqnob >= 0);
1005 
1006 		txpeer->lp_txcredits++;
1007 		if (txpeer->lp_txcredits <= 0) {
1008 			msg2 = list_entry(txpeer->lp_txq.next,
1009 					      lnet_msg_t, msg_list);
1010 			list_del(&msg2->msg_list);
1011 
1012 			LASSERT(msg2->msg_txpeer == txpeer);
1013 			LASSERT(msg2->msg_tx_delayed);
1014 
1015 			(void) lnet_post_send_locked(msg2, 1);
1016 		}
1017 	}
1018 
1019 	if (txpeer != NULL) {
1020 		msg->msg_txpeer = NULL;
1021 		lnet_peer_decref_locked(txpeer);
1022 	}
1023 }
1024 
1025 void
lnet_return_rx_credits_locked(lnet_msg_t * msg)1026 lnet_return_rx_credits_locked(lnet_msg_t *msg)
1027 {
1028 	lnet_peer_t	*rxpeer = msg->msg_rxpeer;
1029 	lnet_msg_t	*msg2;
1030 
1031 	if (msg->msg_rtrcredit) {
1032 		/* give back global router credits */
1033 		lnet_rtrbuf_t     *rb;
1034 		lnet_rtrbufpool_t *rbp;
1035 
1036 		/* NB If a msg ever blocks for a buffer in rbp_msgs, it stays
1037 		 * there until it gets one allocated, or aborts the wait
1038 		 * itself */
1039 		LASSERT(msg->msg_kiov != NULL);
1040 
1041 		rb = list_entry(msg->msg_kiov, lnet_rtrbuf_t, rb_kiov[0]);
1042 		rbp = rb->rb_pool;
1043 		LASSERT(rbp == lnet_msg2bufpool(msg));
1044 
1045 		msg->msg_kiov = NULL;
1046 		msg->msg_rtrcredit = 0;
1047 
1048 		LASSERT((rbp->rbp_credits < 0) ==
1049 			!list_empty(&rbp->rbp_msgs));
1050 		LASSERT((rbp->rbp_credits > 0) ==
1051 			!list_empty(&rbp->rbp_bufs));
1052 
1053 		list_add(&rb->rb_list, &rbp->rbp_bufs);
1054 		rbp->rbp_credits++;
1055 		if (rbp->rbp_credits <= 0) {
1056 			msg2 = list_entry(rbp->rbp_msgs.next,
1057 					      lnet_msg_t, msg_list);
1058 			list_del(&msg2->msg_list);
1059 
1060 			(void) lnet_post_routed_recv_locked(msg2, 1);
1061 		}
1062 	}
1063 
1064 	if (msg->msg_peerrtrcredit) {
1065 		/* give back peer router credits */
1066 		msg->msg_peerrtrcredit = 0;
1067 
1068 		LASSERT((rxpeer->lp_rtrcredits < 0) ==
1069 			!list_empty(&rxpeer->lp_rtrq));
1070 
1071 		rxpeer->lp_rtrcredits++;
1072 		if (rxpeer->lp_rtrcredits <= 0) {
1073 			msg2 = list_entry(rxpeer->lp_rtrq.next,
1074 					      lnet_msg_t, msg_list);
1075 			list_del(&msg2->msg_list);
1076 
1077 			(void) lnet_post_routed_recv_locked(msg2, 1);
1078 		}
1079 	}
1080 	if (rxpeer != NULL) {
1081 		msg->msg_rxpeer = NULL;
1082 		lnet_peer_decref_locked(rxpeer);
1083 	}
1084 }
1085 
1086 static int
lnet_compare_routes(lnet_route_t * r1,lnet_route_t * r2)1087 lnet_compare_routes(lnet_route_t *r1, lnet_route_t *r2)
1088 {
1089 	lnet_peer_t *p1 = r1->lr_gateway;
1090 	lnet_peer_t *p2 = r2->lr_gateway;
1091 
1092 	if (r1->lr_priority < r2->lr_priority)
1093 		return 1;
1094 
1095 	if (r1->lr_priority > r2->lr_priority)
1096 		return -1;
1097 
1098 	if (r1->lr_hops < r2->lr_hops)
1099 		return 1;
1100 
1101 	if (r1->lr_hops > r2->lr_hops)
1102 		return -1;
1103 
1104 	if (p1->lp_txqnob < p2->lp_txqnob)
1105 		return 1;
1106 
1107 	if (p1->lp_txqnob > p2->lp_txqnob)
1108 		return -1;
1109 
1110 	if (p1->lp_txcredits > p2->lp_txcredits)
1111 		return 1;
1112 
1113 	if (p1->lp_txcredits < p2->lp_txcredits)
1114 		return -1;
1115 
1116 	if (r1->lr_seq - r2->lr_seq <= 0)
1117 		return 1;
1118 
1119 	return -1;
1120 }
1121 
1122 static lnet_peer_t *
lnet_find_route_locked(lnet_ni_t * ni,lnet_nid_t target,lnet_nid_t rtr_nid)1123 lnet_find_route_locked(lnet_ni_t *ni, lnet_nid_t target, lnet_nid_t rtr_nid)
1124 {
1125 	lnet_remotenet_t	*rnet;
1126 	lnet_route_t		*rtr;
1127 	lnet_route_t		*rtr_best;
1128 	lnet_route_t		*rtr_last;
1129 	struct lnet_peer	*lp_best;
1130 	struct lnet_peer	*lp;
1131 	int			rc;
1132 
1133 	/* If @rtr_nid is not LNET_NID_ANY, return the gateway with
1134 	 * rtr_nid nid, otherwise find the best gateway I can use */
1135 
1136 	rnet = lnet_find_net_locked(LNET_NIDNET(target));
1137 	if (rnet == NULL)
1138 		return NULL;
1139 
1140 	lp_best = NULL;
1141 	rtr_best = rtr_last = NULL;
1142 	list_for_each_entry(rtr, &rnet->lrn_routes, lr_list) {
1143 		lp = rtr->lr_gateway;
1144 
1145 		if (!lp->lp_alive || /* gateway is down */
1146 		    ((lp->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) != 0 &&
1147 		     rtr->lr_downis != 0)) /* NI to target is down */
1148 			continue;
1149 
1150 		if (ni != NULL && lp->lp_ni != ni)
1151 			continue;
1152 
1153 		if (lp->lp_nid == rtr_nid) /* it's pre-determined router */
1154 			return lp;
1155 
1156 		if (lp_best == NULL) {
1157 			rtr_best = rtr_last = rtr;
1158 			lp_best = lp;
1159 			continue;
1160 		}
1161 
1162 		/* no protection on below fields, but it's harmless */
1163 		if (rtr_last->lr_seq - rtr->lr_seq < 0)
1164 			rtr_last = rtr;
1165 
1166 		rc = lnet_compare_routes(rtr, rtr_best);
1167 		if (rc < 0)
1168 			continue;
1169 
1170 		rtr_best = rtr;
1171 		lp_best = lp;
1172 	}
1173 
1174 	/* set sequence number on the best router to the latest sequence + 1
1175 	 * so we can round-robin all routers, it's race and inaccurate but
1176 	 * harmless and functional  */
1177 	if (rtr_best != NULL)
1178 		rtr_best->lr_seq = rtr_last->lr_seq + 1;
1179 	return lp_best;
1180 }
1181 
1182 int
lnet_send(lnet_nid_t src_nid,lnet_msg_t * msg,lnet_nid_t rtr_nid)1183 lnet_send(lnet_nid_t src_nid, lnet_msg_t *msg, lnet_nid_t rtr_nid)
1184 {
1185 	lnet_nid_t		dst_nid = msg->msg_target.nid;
1186 	struct lnet_ni		*src_ni;
1187 	struct lnet_ni		*local_ni;
1188 	struct lnet_peer	*lp;
1189 	int			cpt;
1190 	int			cpt2;
1191 	int			rc;
1192 
1193 	/* NB: rtr_nid is set to LNET_NID_ANY for all current use-cases,
1194 	 * but we might want to use pre-determined router for ACK/REPLY
1195 	 * in the future */
1196 	/* NB: ni != NULL == interface pre-determined (ACK/REPLY) */
1197 	LASSERT(msg->msg_txpeer == NULL);
1198 	LASSERT(!msg->msg_sending);
1199 	LASSERT(!msg->msg_target_is_router);
1200 	LASSERT(!msg->msg_receiving);
1201 
1202 	msg->msg_sending = 1;
1203 
1204 	LASSERT(!msg->msg_tx_committed);
1205 	cpt = lnet_cpt_of_nid(rtr_nid == LNET_NID_ANY ? dst_nid : rtr_nid);
1206  again:
1207 	lnet_net_lock(cpt);
1208 
1209 	if (the_lnet.ln_shutdown) {
1210 		lnet_net_unlock(cpt);
1211 		return -ESHUTDOWN;
1212 	}
1213 
1214 	if (src_nid == LNET_NID_ANY) {
1215 		src_ni = NULL;
1216 	} else {
1217 		src_ni = lnet_nid2ni_locked(src_nid, cpt);
1218 		if (src_ni == NULL) {
1219 			lnet_net_unlock(cpt);
1220 			LCONSOLE_WARN("Can't send to %s: src %s is not a local nid\n",
1221 				      libcfs_nid2str(dst_nid),
1222 				      libcfs_nid2str(src_nid));
1223 			return -EINVAL;
1224 		}
1225 		LASSERT(!msg->msg_routing);
1226 	}
1227 
1228 	/* Is this for someone on a local network? */
1229 	local_ni = lnet_net2ni_locked(LNET_NIDNET(dst_nid), cpt);
1230 
1231 	if (local_ni != NULL) {
1232 		if (src_ni == NULL) {
1233 			src_ni = local_ni;
1234 			src_nid = src_ni->ni_nid;
1235 		} else if (src_ni == local_ni) {
1236 			lnet_ni_decref_locked(local_ni, cpt);
1237 		} else {
1238 			lnet_ni_decref_locked(local_ni, cpt);
1239 			lnet_ni_decref_locked(src_ni, cpt);
1240 			lnet_net_unlock(cpt);
1241 			LCONSOLE_WARN("No route to %s via from %s\n",
1242 				      libcfs_nid2str(dst_nid),
1243 				      libcfs_nid2str(src_nid));
1244 			return -EINVAL;
1245 		}
1246 
1247 		LASSERT(src_nid != LNET_NID_ANY);
1248 		lnet_msg_commit(msg, cpt);
1249 
1250 		if (!msg->msg_routing)
1251 			msg->msg_hdr.src_nid = cpu_to_le64(src_nid);
1252 
1253 		if (src_ni == the_lnet.ln_loni) {
1254 			/* No send credit hassles with LOLND */
1255 			lnet_net_unlock(cpt);
1256 			lnet_ni_send(src_ni, msg);
1257 
1258 			lnet_net_lock(cpt);
1259 			lnet_ni_decref_locked(src_ni, cpt);
1260 			lnet_net_unlock(cpt);
1261 			return 0;
1262 		}
1263 
1264 		rc = lnet_nid2peer_locked(&lp, dst_nid, cpt);
1265 		/* lp has ref on src_ni; lose mine */
1266 		lnet_ni_decref_locked(src_ni, cpt);
1267 		if (rc != 0) {
1268 			lnet_net_unlock(cpt);
1269 			LCONSOLE_WARN("Error %d finding peer %s\n", rc,
1270 				      libcfs_nid2str(dst_nid));
1271 			/* ENOMEM or shutting down */
1272 			return rc;
1273 		}
1274 		LASSERT(lp->lp_ni == src_ni);
1275 	} else {
1276 		/* sending to a remote network */
1277 		lp = lnet_find_route_locked(src_ni, dst_nid, rtr_nid);
1278 		if (lp == NULL) {
1279 			if (src_ni != NULL)
1280 				lnet_ni_decref_locked(src_ni, cpt);
1281 			lnet_net_unlock(cpt);
1282 
1283 			LCONSOLE_WARN("No route to %s via %s (all routers down)\n",
1284 				      libcfs_id2str(msg->msg_target),
1285 				      libcfs_nid2str(src_nid));
1286 			return -EHOSTUNREACH;
1287 		}
1288 
1289 		/* rtr_nid is LNET_NID_ANY or NID of pre-determined router,
1290 		 * it's possible that rtr_nid isn't LNET_NID_ANY and lp isn't
1291 		 * pre-determined router, this can happen if router table
1292 		 * was changed when we release the lock */
1293 		if (rtr_nid != lp->lp_nid) {
1294 			cpt2 = lnet_cpt_of_nid_locked(lp->lp_nid);
1295 			if (cpt2 != cpt) {
1296 				if (src_ni != NULL)
1297 					lnet_ni_decref_locked(src_ni, cpt);
1298 				lnet_net_unlock(cpt);
1299 
1300 				rtr_nid = lp->lp_nid;
1301 				cpt = cpt2;
1302 				goto again;
1303 			}
1304 		}
1305 
1306 		CDEBUG(D_NET, "Best route to %s via %s for %s %d\n",
1307 		       libcfs_nid2str(dst_nid), libcfs_nid2str(lp->lp_nid),
1308 		       lnet_msgtyp2str(msg->msg_type), msg->msg_len);
1309 
1310 		if (src_ni == NULL) {
1311 			src_ni = lp->lp_ni;
1312 			src_nid = src_ni->ni_nid;
1313 		} else {
1314 			LASSERT(src_ni == lp->lp_ni);
1315 			lnet_ni_decref_locked(src_ni, cpt);
1316 		}
1317 
1318 		lnet_peer_addref_locked(lp);
1319 
1320 		LASSERT(src_nid != LNET_NID_ANY);
1321 		lnet_msg_commit(msg, cpt);
1322 
1323 		if (!msg->msg_routing) {
1324 			/* I'm the source and now I know which NI to send on */
1325 			msg->msg_hdr.src_nid = cpu_to_le64(src_nid);
1326 		}
1327 
1328 		msg->msg_target_is_router = 1;
1329 		msg->msg_target.nid = lp->lp_nid;
1330 		msg->msg_target.pid = LUSTRE_SRV_LNET_PID;
1331 	}
1332 
1333 	/* 'lp' is our best choice of peer */
1334 
1335 	LASSERT(!msg->msg_peertxcredit);
1336 	LASSERT(!msg->msg_txcredit);
1337 	LASSERT(msg->msg_txpeer == NULL);
1338 
1339 	msg->msg_txpeer = lp;		   /* msg takes my ref on lp */
1340 
1341 	rc = lnet_post_send_locked(msg, 0);
1342 	lnet_net_unlock(cpt);
1343 
1344 	if (rc == EHOSTUNREACH || rc == ECANCELED)
1345 		return -rc;
1346 
1347 	if (rc == 0)
1348 		lnet_ni_send(src_ni, msg);
1349 
1350 	return 0; /* rc == 0 or EAGAIN */
1351 }
1352 
1353 static void
lnet_drop_message(lnet_ni_t * ni,int cpt,void * private,unsigned int nob)1354 lnet_drop_message(lnet_ni_t *ni, int cpt, void *private, unsigned int nob)
1355 {
1356 	lnet_net_lock(cpt);
1357 	the_lnet.ln_counters[cpt]->drop_count++;
1358 	the_lnet.ln_counters[cpt]->drop_length += nob;
1359 	lnet_net_unlock(cpt);
1360 
1361 	lnet_ni_recv(ni, private, NULL, 0, 0, 0, nob);
1362 }
1363 
1364 static void
lnet_recv_put(lnet_ni_t * ni,lnet_msg_t * msg)1365 lnet_recv_put(lnet_ni_t *ni, lnet_msg_t *msg)
1366 {
1367 	lnet_hdr_t	*hdr = &msg->msg_hdr;
1368 
1369 	if (msg->msg_wanted != 0)
1370 		lnet_setpayloadbuffer(msg);
1371 
1372 	lnet_build_msg_event(msg, LNET_EVENT_PUT);
1373 
1374 	/* Must I ACK?  If so I'll grab the ack_wmd out of the header and put
1375 	 * it back into the ACK during lnet_finalize() */
1376 	msg->msg_ack = (!lnet_is_wire_handle_none(&hdr->msg.put.ack_wmd) &&
1377 			(msg->msg_md->md_options & LNET_MD_ACK_DISABLE) == 0);
1378 
1379 	lnet_ni_recv(ni, msg->msg_private, msg, msg->msg_rx_delayed,
1380 		     msg->msg_offset, msg->msg_wanted, hdr->payload_length);
1381 }
1382 
1383 static int
lnet_parse_put(lnet_ni_t * ni,lnet_msg_t * msg)1384 lnet_parse_put(lnet_ni_t *ni, lnet_msg_t *msg)
1385 {
1386 	lnet_hdr_t		*hdr = &msg->msg_hdr;
1387 	struct lnet_match_info	info;
1388 	int			rc;
1389 
1390 	/* Convert put fields to host byte order */
1391 	hdr->msg.put.match_bits	= le64_to_cpu(hdr->msg.put.match_bits);
1392 	hdr->msg.put.ptl_index	= le32_to_cpu(hdr->msg.put.ptl_index);
1393 	hdr->msg.put.offset	= le32_to_cpu(hdr->msg.put.offset);
1394 
1395 	info.mi_id.nid	= hdr->src_nid;
1396 	info.mi_id.pid	= hdr->src_pid;
1397 	info.mi_opc	= LNET_MD_OP_PUT;
1398 	info.mi_portal	= hdr->msg.put.ptl_index;
1399 	info.mi_rlength	= hdr->payload_length;
1400 	info.mi_roffset	= hdr->msg.put.offset;
1401 	info.mi_mbits	= hdr->msg.put.match_bits;
1402 
1403 	msg->msg_rx_ready_delay = ni->ni_lnd->lnd_eager_recv == NULL;
1404 
1405  again:
1406 	rc = lnet_ptl_match_md(&info, msg);
1407 	switch (rc) {
1408 	default:
1409 		LBUG();
1410 
1411 	case LNET_MATCHMD_OK:
1412 		lnet_recv_put(ni, msg);
1413 		return 0;
1414 
1415 	case LNET_MATCHMD_NONE:
1416 		if (msg->msg_rx_delayed) /* attached on delayed list */
1417 			return 0;
1418 
1419 		rc = lnet_ni_eager_recv(ni, msg);
1420 		if (rc == 0)
1421 			goto again;
1422 		/* fall through */
1423 
1424 	case LNET_MATCHMD_DROP:
1425 		CNETERR("Dropping PUT from %s portal %d match %llu offset %d length %d: %d\n",
1426 			libcfs_id2str(info.mi_id), info.mi_portal,
1427 			info.mi_mbits, info.mi_roffset, info.mi_rlength, rc);
1428 
1429 		return ENOENT;	/* +ve: OK but no match */
1430 	}
1431 }
1432 
1433 static int
lnet_parse_get(lnet_ni_t * ni,lnet_msg_t * msg,int rdma_get)1434 lnet_parse_get(lnet_ni_t *ni, lnet_msg_t *msg, int rdma_get)
1435 {
1436 	struct lnet_match_info	info;
1437 	lnet_hdr_t		*hdr = &msg->msg_hdr;
1438 	lnet_handle_wire_t	reply_wmd;
1439 	int			rc;
1440 
1441 	/* Convert get fields to host byte order */
1442 	hdr->msg.get.match_bits	  = le64_to_cpu(hdr->msg.get.match_bits);
1443 	hdr->msg.get.ptl_index	  = le32_to_cpu(hdr->msg.get.ptl_index);
1444 	hdr->msg.get.sink_length  = le32_to_cpu(hdr->msg.get.sink_length);
1445 	hdr->msg.get.src_offset	  = le32_to_cpu(hdr->msg.get.src_offset);
1446 
1447 	info.mi_id.nid	= hdr->src_nid;
1448 	info.mi_id.pid	= hdr->src_pid;
1449 	info.mi_opc	= LNET_MD_OP_GET;
1450 	info.mi_portal	= hdr->msg.get.ptl_index;
1451 	info.mi_rlength	= hdr->msg.get.sink_length;
1452 	info.mi_roffset	= hdr->msg.get.src_offset;
1453 	info.mi_mbits	= hdr->msg.get.match_bits;
1454 
1455 	rc = lnet_ptl_match_md(&info, msg);
1456 	if (rc == LNET_MATCHMD_DROP) {
1457 		CNETERR("Dropping GET from %s portal %d match %llu offset %d length %d\n",
1458 			libcfs_id2str(info.mi_id), info.mi_portal,
1459 			info.mi_mbits, info.mi_roffset, info.mi_rlength);
1460 		return ENOENT;	/* +ve: OK but no match */
1461 	}
1462 
1463 	LASSERT(rc == LNET_MATCHMD_OK);
1464 
1465 	lnet_build_msg_event(msg, LNET_EVENT_GET);
1466 
1467 	reply_wmd = hdr->msg.get.return_wmd;
1468 
1469 	lnet_prep_send(msg, LNET_MSG_REPLY, info.mi_id,
1470 		       msg->msg_offset, msg->msg_wanted);
1471 
1472 	msg->msg_hdr.msg.reply.dst_wmd = reply_wmd;
1473 
1474 	if (rdma_get) {
1475 		/* The LND completes the REPLY from her recv procedure */
1476 		lnet_ni_recv(ni, msg->msg_private, msg, 0,
1477 			     msg->msg_offset, msg->msg_len, msg->msg_len);
1478 		return 0;
1479 	}
1480 
1481 	lnet_ni_recv(ni, msg->msg_private, NULL, 0, 0, 0, 0);
1482 	msg->msg_receiving = 0;
1483 
1484 	rc = lnet_send(ni->ni_nid, msg, LNET_NID_ANY);
1485 	if (rc < 0) {
1486 		/* didn't get as far as lnet_ni_send() */
1487 		CERROR("%s: Unable to send REPLY for GET from %s: %d\n",
1488 		       libcfs_nid2str(ni->ni_nid),
1489 		       libcfs_id2str(info.mi_id), rc);
1490 
1491 		lnet_finalize(ni, msg, rc);
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 static int
lnet_parse_reply(lnet_ni_t * ni,lnet_msg_t * msg)1498 lnet_parse_reply(lnet_ni_t *ni, lnet_msg_t *msg)
1499 {
1500 	void	     *private = msg->msg_private;
1501 	lnet_hdr_t       *hdr = &msg->msg_hdr;
1502 	lnet_process_id_t src = {0};
1503 	lnet_libmd_t     *md;
1504 	int	       rlength;
1505 	int	       mlength;
1506 	int			cpt;
1507 
1508 	cpt = lnet_cpt_of_cookie(hdr->msg.reply.dst_wmd.wh_object_cookie);
1509 	lnet_res_lock(cpt);
1510 
1511 	src.nid = hdr->src_nid;
1512 	src.pid = hdr->src_pid;
1513 
1514 	/* NB handles only looked up by creator (no flips) */
1515 	md = lnet_wire_handle2md(&hdr->msg.reply.dst_wmd);
1516 	if (md == NULL || md->md_threshold == 0 || md->md_me != NULL) {
1517 		CNETERR("%s: Dropping REPLY from %s for %s MD %#llx.%#llx\n",
1518 			libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
1519 			(md == NULL) ? "invalid" : "inactive",
1520 			hdr->msg.reply.dst_wmd.wh_interface_cookie,
1521 			hdr->msg.reply.dst_wmd.wh_object_cookie);
1522 		if (md != NULL && md->md_me != NULL)
1523 			CERROR("REPLY MD also attached to portal %d\n",
1524 			       md->md_me->me_portal);
1525 
1526 		lnet_res_unlock(cpt);
1527 		return ENOENT;		  /* +ve: OK but no match */
1528 	}
1529 
1530 	LASSERT(md->md_offset == 0);
1531 
1532 	rlength = hdr->payload_length;
1533 	mlength = min_t(uint, rlength, md->md_length);
1534 
1535 	if (mlength < rlength &&
1536 	    (md->md_options & LNET_MD_TRUNCATE) == 0) {
1537 		CNETERR("%s: Dropping REPLY from %s length %d for MD %#llx would overflow (%d)\n",
1538 			libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
1539 			rlength, hdr->msg.reply.dst_wmd.wh_object_cookie,
1540 			mlength);
1541 		lnet_res_unlock(cpt);
1542 		return ENOENT;	  /* +ve: OK but no match */
1543 	}
1544 
1545 	CDEBUG(D_NET, "%s: Reply from %s of length %d/%d into md %#llx\n",
1546 	       libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
1547 	       mlength, rlength, hdr->msg.reply.dst_wmd.wh_object_cookie);
1548 
1549 	lnet_msg_attach_md(msg, md, 0, mlength);
1550 
1551 	if (mlength != 0)
1552 		lnet_setpayloadbuffer(msg);
1553 
1554 	lnet_res_unlock(cpt);
1555 
1556 	lnet_build_msg_event(msg, LNET_EVENT_REPLY);
1557 
1558 	lnet_ni_recv(ni, private, msg, 0, 0, mlength, rlength);
1559 	return 0;
1560 }
1561 
1562 static int
lnet_parse_ack(lnet_ni_t * ni,lnet_msg_t * msg)1563 lnet_parse_ack(lnet_ni_t *ni, lnet_msg_t *msg)
1564 {
1565 	lnet_hdr_t       *hdr = &msg->msg_hdr;
1566 	lnet_process_id_t src = {0};
1567 	lnet_libmd_t     *md;
1568 	int			cpt;
1569 
1570 	src.nid = hdr->src_nid;
1571 	src.pid = hdr->src_pid;
1572 
1573 	/* Convert ack fields to host byte order */
1574 	hdr->msg.ack.match_bits = le64_to_cpu(hdr->msg.ack.match_bits);
1575 	hdr->msg.ack.mlength = le32_to_cpu(hdr->msg.ack.mlength);
1576 
1577 	cpt = lnet_cpt_of_cookie(hdr->msg.ack.dst_wmd.wh_object_cookie);
1578 	lnet_res_lock(cpt);
1579 
1580 	/* NB handles only looked up by creator (no flips) */
1581 	md = lnet_wire_handle2md(&hdr->msg.ack.dst_wmd);
1582 	if (md == NULL || md->md_threshold == 0 || md->md_me != NULL) {
1583 		/* Don't moan; this is expected */
1584 		CDEBUG(D_NET,
1585 		       "%s: Dropping ACK from %s to %s MD %#llx.%#llx\n",
1586 		       libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
1587 		       (md == NULL) ? "invalid" : "inactive",
1588 		       hdr->msg.ack.dst_wmd.wh_interface_cookie,
1589 		       hdr->msg.ack.dst_wmd.wh_object_cookie);
1590 		if (md != NULL && md->md_me != NULL)
1591 			CERROR("Source MD also attached to portal %d\n",
1592 			       md->md_me->me_portal);
1593 
1594 		lnet_res_unlock(cpt);
1595 		return ENOENT;		  /* +ve! */
1596 	}
1597 
1598 	CDEBUG(D_NET, "%s: ACK from %s into md %#llx\n",
1599 	       libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
1600 	       hdr->msg.ack.dst_wmd.wh_object_cookie);
1601 
1602 	lnet_msg_attach_md(msg, md, 0, 0);
1603 
1604 	lnet_res_unlock(cpt);
1605 
1606 	lnet_build_msg_event(msg, LNET_EVENT_ACK);
1607 
1608 	lnet_ni_recv(ni, msg->msg_private, msg, 0, 0, 0, msg->msg_len);
1609 	return 0;
1610 }
1611 
1612 static int
lnet_parse_forward_locked(lnet_ni_t * ni,lnet_msg_t * msg)1613 lnet_parse_forward_locked(lnet_ni_t *ni, lnet_msg_t *msg)
1614 {
1615 	int	rc = 0;
1616 
1617 	if (msg->msg_rxpeer->lp_rtrcredits <= 0 ||
1618 	    lnet_msg2bufpool(msg)->rbp_credits <= 0) {
1619 		if (ni->ni_lnd->lnd_eager_recv == NULL) {
1620 			msg->msg_rx_ready_delay = 1;
1621 		} else {
1622 			lnet_net_unlock(msg->msg_rx_cpt);
1623 			rc = lnet_ni_eager_recv(ni, msg);
1624 			lnet_net_lock(msg->msg_rx_cpt);
1625 		}
1626 	}
1627 
1628 	if (rc == 0)
1629 		rc = lnet_post_routed_recv_locked(msg, 0);
1630 	return rc;
1631 }
1632 
1633 char *
lnet_msgtyp2str(int type)1634 lnet_msgtyp2str(int type)
1635 {
1636 	switch (type) {
1637 	case LNET_MSG_ACK:
1638 		return "ACK";
1639 	case LNET_MSG_PUT:
1640 		return "PUT";
1641 	case LNET_MSG_GET:
1642 		return "GET";
1643 	case LNET_MSG_REPLY:
1644 		return "REPLY";
1645 	case LNET_MSG_HELLO:
1646 		return "HELLO";
1647 	default:
1648 		return "<UNKNOWN>";
1649 	}
1650 }
1651 EXPORT_SYMBOL(lnet_msgtyp2str);
1652 
1653 void
lnet_print_hdr(lnet_hdr_t * hdr)1654 lnet_print_hdr(lnet_hdr_t *hdr)
1655 {
1656 	lnet_process_id_t src = {0};
1657 	lnet_process_id_t dst = {0};
1658 	char *type_str = lnet_msgtyp2str(hdr->type);
1659 
1660 	src.nid = hdr->src_nid;
1661 	src.pid = hdr->src_pid;
1662 
1663 	dst.nid = hdr->dest_nid;
1664 	dst.pid = hdr->dest_pid;
1665 
1666 	CWARN("P3 Header at %p of type %s\n", hdr, type_str);
1667 	CWARN("    From %s\n", libcfs_id2str(src));
1668 	CWARN("    To   %s\n", libcfs_id2str(dst));
1669 
1670 	switch (hdr->type) {
1671 	default:
1672 		break;
1673 
1674 	case LNET_MSG_PUT:
1675 		CWARN("    Ptl index %d, ack md %#llx.%#llx, match bits %llu\n",
1676 		      hdr->msg.put.ptl_index,
1677 		      hdr->msg.put.ack_wmd.wh_interface_cookie,
1678 		      hdr->msg.put.ack_wmd.wh_object_cookie,
1679 		      hdr->msg.put.match_bits);
1680 		CWARN("    Length %d, offset %d, hdr data %#llx\n",
1681 		      hdr->payload_length, hdr->msg.put.offset,
1682 		      hdr->msg.put.hdr_data);
1683 		break;
1684 
1685 	case LNET_MSG_GET:
1686 		CWARN("    Ptl index %d, return md %#llx.%#llx, match bits %llu\n",
1687 		      hdr->msg.get.ptl_index,
1688 		      hdr->msg.get.return_wmd.wh_interface_cookie,
1689 		      hdr->msg.get.return_wmd.wh_object_cookie,
1690 		      hdr->msg.get.match_bits);
1691 		CWARN("    Length %d, src offset %d\n",
1692 		      hdr->msg.get.sink_length,
1693 		      hdr->msg.get.src_offset);
1694 		break;
1695 
1696 	case LNET_MSG_ACK:
1697 		CWARN("    dst md %#llx.%#llx, manipulated length %d\n",
1698 		      hdr->msg.ack.dst_wmd.wh_interface_cookie,
1699 		      hdr->msg.ack.dst_wmd.wh_object_cookie,
1700 		      hdr->msg.ack.mlength);
1701 		break;
1702 
1703 	case LNET_MSG_REPLY:
1704 		CWARN("    dst md %#llx.%#llx, length %d\n",
1705 		      hdr->msg.reply.dst_wmd.wh_interface_cookie,
1706 		      hdr->msg.reply.dst_wmd.wh_object_cookie,
1707 		      hdr->payload_length);
1708 	}
1709 
1710 }
1711 
1712 int
lnet_parse(lnet_ni_t * ni,lnet_hdr_t * hdr,lnet_nid_t from_nid,void * private,int rdma_req)1713 lnet_parse(lnet_ni_t *ni, lnet_hdr_t *hdr, lnet_nid_t from_nid,
1714 	   void *private, int rdma_req)
1715 {
1716 	int		rc = 0;
1717 	int		cpt;
1718 	int		for_me;
1719 	struct lnet_msg	*msg;
1720 	lnet_pid_t     dest_pid;
1721 	lnet_nid_t     dest_nid;
1722 	lnet_nid_t     src_nid;
1723 	__u32	  payload_length;
1724 	__u32	  type;
1725 
1726 	LASSERT(!in_interrupt());
1727 
1728 	type = le32_to_cpu(hdr->type);
1729 	src_nid = le64_to_cpu(hdr->src_nid);
1730 	dest_nid = le64_to_cpu(hdr->dest_nid);
1731 	dest_pid = le32_to_cpu(hdr->dest_pid);
1732 	payload_length = le32_to_cpu(hdr->payload_length);
1733 
1734 	for_me = (ni->ni_nid == dest_nid);
1735 	cpt = lnet_cpt_of_nid(from_nid);
1736 
1737 	switch (type) {
1738 	case LNET_MSG_ACK:
1739 	case LNET_MSG_GET:
1740 		if (payload_length > 0) {
1741 			CERROR("%s, src %s: bad %s payload %d (0 expected)\n",
1742 			       libcfs_nid2str(from_nid),
1743 			       libcfs_nid2str(src_nid),
1744 			       lnet_msgtyp2str(type), payload_length);
1745 			return -EPROTO;
1746 		}
1747 		break;
1748 
1749 	case LNET_MSG_PUT:
1750 	case LNET_MSG_REPLY:
1751 		if (payload_length >
1752 		   (__u32)(for_me ? LNET_MAX_PAYLOAD : LNET_MTU)) {
1753 			CERROR("%s, src %s: bad %s payload %d (%d max expected)\n",
1754 			       libcfs_nid2str(from_nid),
1755 			       libcfs_nid2str(src_nid),
1756 			       lnet_msgtyp2str(type),
1757 			       payload_length,
1758 			       for_me ? LNET_MAX_PAYLOAD : LNET_MTU);
1759 			return -EPROTO;
1760 		}
1761 		break;
1762 
1763 	default:
1764 		CERROR("%s, src %s: Bad message type 0x%x\n",
1765 		       libcfs_nid2str(from_nid),
1766 		       libcfs_nid2str(src_nid), type);
1767 		return -EPROTO;
1768 	}
1769 
1770 	if (the_lnet.ln_routing &&
1771 	    ni->ni_last_alive != get_seconds()) {
1772 		lnet_ni_lock(ni);
1773 
1774 		/* NB: so far here is the only place to set NI status to "up */
1775 		ni->ni_last_alive = get_seconds();
1776 		if (ni->ni_status != NULL &&
1777 		    ni->ni_status->ns_status == LNET_NI_STATUS_DOWN)
1778 			ni->ni_status->ns_status = LNET_NI_STATUS_UP;
1779 		lnet_ni_unlock(ni);
1780 	}
1781 
1782 	/* Regard a bad destination NID as a protocol error.  Senders should
1783 	 * know what they're doing; if they don't they're misconfigured, buggy
1784 	 * or malicious so we chop them off at the knees :) */
1785 
1786 	if (!for_me) {
1787 		if (LNET_NIDNET(dest_nid) == LNET_NIDNET(ni->ni_nid)) {
1788 			/* should have gone direct */
1789 			CERROR("%s, src %s: Bad dest nid %s (should have been sent direct)\n",
1790 			       libcfs_nid2str(from_nid),
1791 			       libcfs_nid2str(src_nid),
1792 			       libcfs_nid2str(dest_nid));
1793 			return -EPROTO;
1794 		}
1795 
1796 		if (lnet_islocalnid(dest_nid)) {
1797 			/* dest is another local NI; sender should have used
1798 			 * this node's NID on its own network */
1799 			CERROR("%s, src %s: Bad dest nid %s (it's my nid but on a different network)\n",
1800 			       libcfs_nid2str(from_nid),
1801 			       libcfs_nid2str(src_nid),
1802 			       libcfs_nid2str(dest_nid));
1803 			return -EPROTO;
1804 		}
1805 
1806 		if (rdma_req && type == LNET_MSG_GET) {
1807 			CERROR("%s, src %s: Bad optimized GET for %s (final destination must be me)\n",
1808 			       libcfs_nid2str(from_nid),
1809 			       libcfs_nid2str(src_nid),
1810 			       libcfs_nid2str(dest_nid));
1811 			return -EPROTO;
1812 		}
1813 
1814 		if (!the_lnet.ln_routing) {
1815 			CERROR("%s, src %s: Dropping message for %s (routing not enabled)\n",
1816 			       libcfs_nid2str(from_nid),
1817 			       libcfs_nid2str(src_nid),
1818 			       libcfs_nid2str(dest_nid));
1819 			goto drop;
1820 		}
1821 	}
1822 
1823 	/* Message looks OK; we're not going to return an error, so we MUST
1824 	 * call back lnd_recv() come what may... */
1825 
1826 	if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
1827 	    fail_peer(src_nid, 0)) {	     /* shall we now? */
1828 		CERROR("%s, src %s: Dropping %s to simulate failure\n",
1829 		       libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
1830 		       lnet_msgtyp2str(type));
1831 		goto drop;
1832 	}
1833 
1834 	msg = lnet_msg_alloc();
1835 	if (msg == NULL) {
1836 		CERROR("%s, src %s: Dropping %s (out of memory)\n",
1837 		       libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
1838 		       lnet_msgtyp2str(type));
1839 		goto drop;
1840 	}
1841 
1842 	/* msg zeroed in lnet_msg_alloc;
1843 	 * i.e. flags all clear, pointers NULL etc
1844 	 */
1845 
1846 	msg->msg_type = type;
1847 	msg->msg_private = private;
1848 	msg->msg_receiving = 1;
1849 	msg->msg_len = msg->msg_wanted = payload_length;
1850 	msg->msg_offset = 0;
1851 	msg->msg_hdr = *hdr;
1852 	/* for building message event */
1853 	msg->msg_from = from_nid;
1854 	if (!for_me) {
1855 		msg->msg_target.pid	= dest_pid;
1856 		msg->msg_target.nid	= dest_nid;
1857 		msg->msg_routing	= 1;
1858 
1859 	} else {
1860 		/* convert common msg->hdr fields to host byteorder */
1861 		msg->msg_hdr.type	= type;
1862 		msg->msg_hdr.src_nid	= src_nid;
1863 		msg->msg_hdr.src_pid	= le32_to_cpu(msg->msg_hdr.src_pid);
1864 		msg->msg_hdr.dest_nid	= dest_nid;
1865 		msg->msg_hdr.dest_pid	= dest_pid;
1866 		msg->msg_hdr.payload_length = payload_length;
1867 	}
1868 
1869 	lnet_net_lock(cpt);
1870 	rc = lnet_nid2peer_locked(&msg->msg_rxpeer, from_nid, cpt);
1871 	if (rc != 0) {
1872 		lnet_net_unlock(cpt);
1873 		CERROR("%s, src %s: Dropping %s (error %d looking up sender)\n",
1874 		       libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
1875 		       lnet_msgtyp2str(type), rc);
1876 		lnet_msg_free(msg);
1877 		goto drop;
1878 	}
1879 
1880 	if (lnet_isrouter(msg->msg_rxpeer)) {
1881 		lnet_peer_set_alive(msg->msg_rxpeer);
1882 		if (avoid_asym_router_failure &&
1883 		    LNET_NIDNET(src_nid) != LNET_NIDNET(from_nid)) {
1884 			/* received a remote message from router, update
1885 			 * remote NI status on this router.
1886 			 * NB: multi-hop routed message will be ignored.
1887 			 */
1888 			lnet_router_ni_update_locked(msg->msg_rxpeer,
1889 						     LNET_NIDNET(src_nid));
1890 		}
1891 	}
1892 
1893 	lnet_msg_commit(msg, cpt);
1894 
1895 	if (!for_me) {
1896 		rc = lnet_parse_forward_locked(ni, msg);
1897 		lnet_net_unlock(cpt);
1898 
1899 		if (rc < 0)
1900 			goto free_drop;
1901 		if (rc == 0) {
1902 			lnet_ni_recv(ni, msg->msg_private, msg, 0,
1903 				     0, payload_length, payload_length);
1904 		}
1905 		return 0;
1906 	}
1907 
1908 	lnet_net_unlock(cpt);
1909 
1910 	switch (type) {
1911 	case LNET_MSG_ACK:
1912 		rc = lnet_parse_ack(ni, msg);
1913 		break;
1914 	case LNET_MSG_PUT:
1915 		rc = lnet_parse_put(ni, msg);
1916 		break;
1917 	case LNET_MSG_GET:
1918 		rc = lnet_parse_get(ni, msg, rdma_req);
1919 		break;
1920 	case LNET_MSG_REPLY:
1921 		rc = lnet_parse_reply(ni, msg);
1922 		break;
1923 	default:
1924 		LASSERT(0);
1925 		rc = -EPROTO;
1926 		goto free_drop;  /* prevent an unused label if !kernel */
1927 	}
1928 
1929 	if (rc == 0)
1930 		return 0;
1931 
1932 	LASSERT(rc == ENOENT);
1933 
1934  free_drop:
1935 	LASSERT(msg->msg_md == NULL);
1936 	lnet_finalize(ni, msg, rc);
1937 
1938  drop:
1939 	lnet_drop_message(ni, cpt, private, payload_length);
1940 	return 0;
1941 }
1942 EXPORT_SYMBOL(lnet_parse);
1943 
1944 void
lnet_drop_delayed_msg_list(struct list_head * head,char * reason)1945 lnet_drop_delayed_msg_list(struct list_head *head, char *reason)
1946 {
1947 	while (!list_empty(head)) {
1948 		lnet_process_id_t	id = {0};
1949 		lnet_msg_t		*msg;
1950 
1951 		msg = list_entry(head->next, lnet_msg_t, msg_list);
1952 		list_del(&msg->msg_list);
1953 
1954 		id.nid = msg->msg_hdr.src_nid;
1955 		id.pid = msg->msg_hdr.src_pid;
1956 
1957 		LASSERT(msg->msg_md == NULL);
1958 		LASSERT(msg->msg_rx_delayed);
1959 		LASSERT(msg->msg_rxpeer != NULL);
1960 		LASSERT(msg->msg_hdr.type == LNET_MSG_PUT);
1961 
1962 		CWARN("Dropping delayed PUT from %s portal %d match %llu offset %d length %d: %s\n",
1963 		      libcfs_id2str(id),
1964 		      msg->msg_hdr.msg.put.ptl_index,
1965 		      msg->msg_hdr.msg.put.match_bits,
1966 		      msg->msg_hdr.msg.put.offset,
1967 		      msg->msg_hdr.payload_length, reason);
1968 
1969 		/* NB I can't drop msg's ref on msg_rxpeer until after I've
1970 		 * called lnet_drop_message(), so I just hang onto msg as well
1971 		 * until that's done */
1972 
1973 		lnet_drop_message(msg->msg_rxpeer->lp_ni,
1974 				  msg->msg_rxpeer->lp_cpt,
1975 				  msg->msg_private, msg->msg_len);
1976 		/*
1977 		 * NB: message will not generate event because w/o attached MD,
1978 		 * but we still should give error code so lnet_msg_decommit()
1979 		 * can skip counters operations and other checks.
1980 		 */
1981 		lnet_finalize(msg->msg_rxpeer->lp_ni, msg, -ENOENT);
1982 	}
1983 }
1984 
1985 void
lnet_recv_delayed_msg_list(struct list_head * head)1986 lnet_recv_delayed_msg_list(struct list_head *head)
1987 {
1988 	while (!list_empty(head)) {
1989 		lnet_msg_t	  *msg;
1990 		lnet_process_id_t  id;
1991 
1992 		msg = list_entry(head->next, lnet_msg_t, msg_list);
1993 		list_del(&msg->msg_list);
1994 
1995 		/* md won't disappear under me, since each msg
1996 		 * holds a ref on it */
1997 
1998 		id.nid = msg->msg_hdr.src_nid;
1999 		id.pid = msg->msg_hdr.src_pid;
2000 
2001 		LASSERT(msg->msg_rx_delayed);
2002 		LASSERT(msg->msg_md != NULL);
2003 		LASSERT(msg->msg_rxpeer != NULL);
2004 		LASSERT(msg->msg_hdr.type == LNET_MSG_PUT);
2005 
2006 		CDEBUG(D_NET, "Resuming delayed PUT from %s portal %d match %llu offset %d length %d.\n",
2007 		       libcfs_id2str(id), msg->msg_hdr.msg.put.ptl_index,
2008 		       msg->msg_hdr.msg.put.match_bits,
2009 		       msg->msg_hdr.msg.put.offset,
2010 		       msg->msg_hdr.payload_length);
2011 
2012 		lnet_recv_put(msg->msg_rxpeer->lp_ni, msg);
2013 	}
2014 }
2015 
2016 /**
2017  * Initiate an asynchronous PUT operation.
2018  *
2019  * There are several events associated with a PUT: completion of the send on
2020  * the initiator node (LNET_EVENT_SEND), and when the send completes
2021  * successfully, the receipt of an acknowledgment (LNET_EVENT_ACK) indicating
2022  * that the operation was accepted by the target. The event LNET_EVENT_PUT is
2023  * used at the target node to indicate the completion of incoming data
2024  * delivery.
2025  *
2026  * The local events will be logged in the EQ associated with the MD pointed to
2027  * by \a mdh handle. Using a MD without an associated EQ results in these
2028  * events being discarded. In this case, the caller must have another
2029  * mechanism (e.g., a higher level protocol) for determining when it is safe
2030  * to modify the memory region associated with the MD.
2031  *
2032  * Note that LNet does not guarantee the order of LNET_EVENT_SEND and
2033  * LNET_EVENT_ACK, though intuitively ACK should happen after SEND.
2034  *
2035  * \param self Indicates the NID of a local interface through which to send
2036  * the PUT request. Use LNET_NID_ANY to let LNet choose one by itself.
2037  * \param mdh A handle for the MD that describes the memory to be sent. The MD
2038  * must be "free floating" (See LNetMDBind()).
2039  * \param ack Controls whether an acknowledgment is requested.
2040  * Acknowledgments are only sent when they are requested by the initiating
2041  * process and the target MD enables them.
2042  * \param target A process identifier for the target process.
2043  * \param portal The index in the \a target's portal table.
2044  * \param match_bits The match bits to use for MD selection at the target
2045  * process.
2046  * \param offset The offset into the target MD (only used when the target
2047  * MD has the LNET_MD_MANAGE_REMOTE option set).
2048  * \param hdr_data 64 bits of user data that can be included in the message
2049  * header. This data is written to an event queue entry at the target if an
2050  * EQ is present on the matching MD.
2051  *
2052  * \retval  0      Success, and only in this case events will be generated
2053  * and logged to EQ (if it exists).
2054  * \retval -EIO    Simulated failure.
2055  * \retval -ENOMEM Memory allocation failure.
2056  * \retval -ENOENT Invalid MD object.
2057  *
2058  * \see lnet_event_t::hdr_data and lnet_event_kind_t.
2059  */
2060 int
LNetPut(lnet_nid_t self,lnet_handle_md_t mdh,lnet_ack_req_t ack,lnet_process_id_t target,unsigned int portal,__u64 match_bits,unsigned int offset,__u64 hdr_data)2061 LNetPut(lnet_nid_t self, lnet_handle_md_t mdh, lnet_ack_req_t ack,
2062 	lnet_process_id_t target, unsigned int portal,
2063 	__u64 match_bits, unsigned int offset,
2064 	__u64 hdr_data)
2065 {
2066 	struct lnet_msg		*msg;
2067 	struct lnet_libmd	*md;
2068 	int			cpt;
2069 	int			rc;
2070 
2071 	LASSERT(the_lnet.ln_init);
2072 	LASSERT(the_lnet.ln_refcount > 0);
2073 
2074 	if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
2075 	    fail_peer(target.nid, 1)) { /* shall we now? */
2076 		CERROR("Dropping PUT to %s: simulated failure\n",
2077 		       libcfs_id2str(target));
2078 		return -EIO;
2079 	}
2080 
2081 	msg = lnet_msg_alloc();
2082 	if (msg == NULL) {
2083 		CERROR("Dropping PUT to %s: ENOMEM on lnet_msg_t\n",
2084 		       libcfs_id2str(target));
2085 		return -ENOMEM;
2086 	}
2087 	msg->msg_vmflush = !!memory_pressure_get();
2088 
2089 	cpt = lnet_cpt_of_cookie(mdh.cookie);
2090 	lnet_res_lock(cpt);
2091 
2092 	md = lnet_handle2md(&mdh);
2093 	if (md == NULL || md->md_threshold == 0 || md->md_me != NULL) {
2094 		CERROR("Dropping PUT (%llu:%d:%s): MD (%d) invalid\n",
2095 		       match_bits, portal, libcfs_id2str(target),
2096 		       md == NULL ? -1 : md->md_threshold);
2097 		if (md != NULL && md->md_me != NULL)
2098 			CERROR("Source MD also attached to portal %d\n",
2099 			       md->md_me->me_portal);
2100 		lnet_res_unlock(cpt);
2101 
2102 		lnet_msg_free(msg);
2103 		return -ENOENT;
2104 	}
2105 
2106 	CDEBUG(D_NET, "LNetPut -> %s\n", libcfs_id2str(target));
2107 
2108 	lnet_msg_attach_md(msg, md, 0, 0);
2109 
2110 	lnet_prep_send(msg, LNET_MSG_PUT, target, 0, md->md_length);
2111 
2112 	msg->msg_hdr.msg.put.match_bits = cpu_to_le64(match_bits);
2113 	msg->msg_hdr.msg.put.ptl_index = cpu_to_le32(portal);
2114 	msg->msg_hdr.msg.put.offset = cpu_to_le32(offset);
2115 	msg->msg_hdr.msg.put.hdr_data = hdr_data;
2116 
2117 	/* NB handles only looked up by creator (no flips) */
2118 	if (ack == LNET_ACK_REQ) {
2119 		msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
2120 			the_lnet.ln_interface_cookie;
2121 		msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
2122 			md->md_lh.lh_cookie;
2123 	} else {
2124 		msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
2125 			LNET_WIRE_HANDLE_COOKIE_NONE;
2126 		msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
2127 			LNET_WIRE_HANDLE_COOKIE_NONE;
2128 	}
2129 
2130 	lnet_res_unlock(cpt);
2131 
2132 	lnet_build_msg_event(msg, LNET_EVENT_SEND);
2133 
2134 	rc = lnet_send(self, msg, LNET_NID_ANY);
2135 	if (rc != 0) {
2136 		CNETERR("Error sending PUT to %s: %d\n",
2137 		       libcfs_id2str(target), rc);
2138 		lnet_finalize(NULL, msg, rc);
2139 	}
2140 
2141 	/* completion will be signalled by an event */
2142 	return 0;
2143 }
2144 EXPORT_SYMBOL(LNetPut);
2145 
2146 lnet_msg_t *
lnet_create_reply_msg(lnet_ni_t * ni,lnet_msg_t * getmsg)2147 lnet_create_reply_msg(lnet_ni_t *ni, lnet_msg_t *getmsg)
2148 {
2149 	/* The LND can DMA direct to the GET md (i.e. no REPLY msg).  This
2150 	 * returns a msg for the LND to pass to lnet_finalize() when the sink
2151 	 * data has been received.
2152 	 *
2153 	 * CAVEAT EMPTOR: 'getmsg' is the original GET, which is freed when
2154 	 * lnet_finalize() is called on it, so the LND must call this first */
2155 
2156 	struct lnet_msg		*msg = lnet_msg_alloc();
2157 	struct lnet_libmd	*getmd = getmsg->msg_md;
2158 	lnet_process_id_t	peer_id = getmsg->msg_target;
2159 	int			cpt;
2160 
2161 	LASSERT(!getmsg->msg_target_is_router);
2162 	LASSERT(!getmsg->msg_routing);
2163 
2164 	cpt = lnet_cpt_of_cookie(getmd->md_lh.lh_cookie);
2165 	lnet_res_lock(cpt);
2166 
2167 	LASSERT(getmd->md_refcount > 0);
2168 
2169 	if (msg == NULL) {
2170 		CERROR("%s: Dropping REPLY from %s: can't allocate msg\n",
2171 			libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id));
2172 		goto drop;
2173 	}
2174 
2175 	if (getmd->md_threshold == 0) {
2176 		CERROR("%s: Dropping REPLY from %s for inactive MD %p\n",
2177 			libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id),
2178 			getmd);
2179 		lnet_res_unlock(cpt);
2180 		goto drop;
2181 	}
2182 
2183 	LASSERT(getmd->md_offset == 0);
2184 
2185 	CDEBUG(D_NET, "%s: Reply from %s md %p\n",
2186 	       libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id), getmd);
2187 
2188 	/* setup information for lnet_build_msg_event */
2189 	msg->msg_from = peer_id.nid;
2190 	msg->msg_type = LNET_MSG_GET; /* flag this msg as an "optimized" GET */
2191 	msg->msg_hdr.src_nid = peer_id.nid;
2192 	msg->msg_hdr.payload_length = getmd->md_length;
2193 	msg->msg_receiving = 1; /* required by lnet_msg_attach_md */
2194 
2195 	lnet_msg_attach_md(msg, getmd, getmd->md_offset, getmd->md_length);
2196 	lnet_res_unlock(cpt);
2197 
2198 	cpt = lnet_cpt_of_nid(peer_id.nid);
2199 
2200 	lnet_net_lock(cpt);
2201 	lnet_msg_commit(msg, cpt);
2202 	lnet_net_unlock(cpt);
2203 
2204 	lnet_build_msg_event(msg, LNET_EVENT_REPLY);
2205 
2206 	return msg;
2207 
2208  drop:
2209 	cpt = lnet_cpt_of_nid(peer_id.nid);
2210 
2211 	lnet_net_lock(cpt);
2212 	the_lnet.ln_counters[cpt]->drop_count++;
2213 	the_lnet.ln_counters[cpt]->drop_length += getmd->md_length;
2214 	lnet_net_unlock(cpt);
2215 
2216 	if (msg != NULL)
2217 		lnet_msg_free(msg);
2218 
2219 	return NULL;
2220 }
2221 EXPORT_SYMBOL(lnet_create_reply_msg);
2222 
2223 void
lnet_set_reply_msg_len(lnet_ni_t * ni,lnet_msg_t * reply,unsigned int len)2224 lnet_set_reply_msg_len(lnet_ni_t *ni, lnet_msg_t *reply, unsigned int len)
2225 {
2226 	/* Set the REPLY length, now the RDMA that elides the REPLY message has
2227 	 * completed and I know it. */
2228 	LASSERT(reply != NULL);
2229 	LASSERT(reply->msg_type == LNET_MSG_GET);
2230 	LASSERT(reply->msg_ev.type == LNET_EVENT_REPLY);
2231 
2232 	/* NB I trusted my peer to RDMA.  If she tells me she's written beyond
2233 	 * the end of my buffer, I might as well be dead. */
2234 	LASSERT(len <= reply->msg_ev.mlength);
2235 
2236 	reply->msg_ev.mlength = len;
2237 }
2238 EXPORT_SYMBOL(lnet_set_reply_msg_len);
2239 
2240 /**
2241  * Initiate an asynchronous GET operation.
2242  *
2243  * On the initiator node, an LNET_EVENT_SEND is logged when the GET request
2244  * is sent, and an LNET_EVENT_REPLY is logged when the data returned from
2245  * the target node in the REPLY has been written to local MD.
2246  *
2247  * On the target node, an LNET_EVENT_GET is logged when the GET request
2248  * arrives and is accepted into a MD.
2249  *
2250  * \param self,target,portal,match_bits,offset See the discussion in LNetPut().
2251  * \param mdh A handle for the MD that describes the memory into which the
2252  * requested data will be received. The MD must be "free floating"
2253  * (See LNetMDBind()).
2254  *
2255  * \retval  0      Success, and only in this case events will be generated
2256  * and logged to EQ (if it exists) of the MD.
2257  * \retval -EIO    Simulated failure.
2258  * \retval -ENOMEM Memory allocation failure.
2259  * \retval -ENOENT Invalid MD object.
2260  */
2261 int
LNetGet(lnet_nid_t self,lnet_handle_md_t mdh,lnet_process_id_t target,unsigned int portal,__u64 match_bits,unsigned int offset)2262 LNetGet(lnet_nid_t self, lnet_handle_md_t mdh,
2263 	lnet_process_id_t target, unsigned int portal,
2264 	__u64 match_bits, unsigned int offset)
2265 {
2266 	struct lnet_msg		*msg;
2267 	struct lnet_libmd	*md;
2268 	int			cpt;
2269 	int			rc;
2270 
2271 	LASSERT(the_lnet.ln_init);
2272 	LASSERT(the_lnet.ln_refcount > 0);
2273 
2274 	if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
2275 	    fail_peer(target.nid, 1)) {	  /* shall we now? */
2276 		CERROR("Dropping GET to %s: simulated failure\n",
2277 		       libcfs_id2str(target));
2278 		return -EIO;
2279 	}
2280 
2281 	msg = lnet_msg_alloc();
2282 	if (msg == NULL) {
2283 		CERROR("Dropping GET to %s: ENOMEM on lnet_msg_t\n",
2284 		       libcfs_id2str(target));
2285 		return -ENOMEM;
2286 	}
2287 
2288 	cpt = lnet_cpt_of_cookie(mdh.cookie);
2289 	lnet_res_lock(cpt);
2290 
2291 	md = lnet_handle2md(&mdh);
2292 	if (md == NULL || md->md_threshold == 0 || md->md_me != NULL) {
2293 		CERROR("Dropping GET (%llu:%d:%s): MD (%d) invalid\n",
2294 		       match_bits, portal, libcfs_id2str(target),
2295 		       md == NULL ? -1 : md->md_threshold);
2296 		if (md != NULL && md->md_me != NULL)
2297 			CERROR("REPLY MD also attached to portal %d\n",
2298 			       md->md_me->me_portal);
2299 
2300 		lnet_res_unlock(cpt);
2301 
2302 		lnet_msg_free(msg);
2303 		return -ENOENT;
2304 	}
2305 
2306 	CDEBUG(D_NET, "LNetGet -> %s\n", libcfs_id2str(target));
2307 
2308 	lnet_msg_attach_md(msg, md, 0, 0);
2309 
2310 	lnet_prep_send(msg, LNET_MSG_GET, target, 0, 0);
2311 
2312 	msg->msg_hdr.msg.get.match_bits = cpu_to_le64(match_bits);
2313 	msg->msg_hdr.msg.get.ptl_index = cpu_to_le32(portal);
2314 	msg->msg_hdr.msg.get.src_offset = cpu_to_le32(offset);
2315 	msg->msg_hdr.msg.get.sink_length = cpu_to_le32(md->md_length);
2316 
2317 	/* NB handles only looked up by creator (no flips) */
2318 	msg->msg_hdr.msg.get.return_wmd.wh_interface_cookie =
2319 		the_lnet.ln_interface_cookie;
2320 	msg->msg_hdr.msg.get.return_wmd.wh_object_cookie =
2321 		md->md_lh.lh_cookie;
2322 
2323 	lnet_res_unlock(cpt);
2324 
2325 	lnet_build_msg_event(msg, LNET_EVENT_SEND);
2326 
2327 	rc = lnet_send(self, msg, LNET_NID_ANY);
2328 	if (rc < 0) {
2329 		CNETERR("Error sending GET to %s: %d\n",
2330 		       libcfs_id2str(target), rc);
2331 		lnet_finalize(NULL, msg, rc);
2332 	}
2333 
2334 	/* completion will be signalled by an event */
2335 	return 0;
2336 }
2337 EXPORT_SYMBOL(LNetGet);
2338 
2339 /**
2340  * Calculate distance to node at \a dstnid.
2341  *
2342  * \param dstnid Target NID.
2343  * \param srcnidp If not NULL, NID of the local interface to reach \a dstnid
2344  * is saved here.
2345  * \param orderp If not NULL, order of the route to reach \a dstnid is saved
2346  * here.
2347  *
2348  * \retval 0 If \a dstnid belongs to a local interface, and reserved option
2349  * local_nid_dist_zero is set, which is the default.
2350  * \retval positives Distance to target NID, i.e. number of hops plus one.
2351  * \retval -EHOSTUNREACH If \a dstnid is not reachable.
2352  */
2353 int
LNetDist(lnet_nid_t dstnid,lnet_nid_t * srcnidp,__u32 * orderp)2354 LNetDist(lnet_nid_t dstnid, lnet_nid_t *srcnidp, __u32 *orderp)
2355 {
2356 	struct list_head		*e;
2357 	struct lnet_ni		*ni;
2358 	lnet_remotenet_t	*rnet;
2359 	__u32			dstnet = LNET_NIDNET(dstnid);
2360 	int			hops;
2361 	int			cpt;
2362 	__u32			order = 2;
2363 	struct list_head		*rn_list;
2364 
2365 	/* if !local_nid_dist_zero, I don't return a distance of 0 ever
2366 	 * (when lustre sees a distance of 0, it substitutes 0@lo), so I
2367 	 * keep order 0 free for 0@lo and order 1 free for a local NID
2368 	 * match */
2369 
2370 	LASSERT(the_lnet.ln_init);
2371 	LASSERT(the_lnet.ln_refcount > 0);
2372 
2373 	cpt = lnet_net_lock_current();
2374 
2375 	list_for_each(e, &the_lnet.ln_nis) {
2376 		ni = list_entry(e, lnet_ni_t, ni_list);
2377 
2378 		if (ni->ni_nid == dstnid) {
2379 			if (srcnidp != NULL)
2380 				*srcnidp = dstnid;
2381 			if (orderp != NULL) {
2382 				if (LNET_NETTYP(LNET_NIDNET(dstnid)) == LOLND)
2383 					*orderp = 0;
2384 				else
2385 					*orderp = 1;
2386 			}
2387 			lnet_net_unlock(cpt);
2388 
2389 			return local_nid_dist_zero ? 0 : 1;
2390 		}
2391 
2392 		if (LNET_NIDNET(ni->ni_nid) == dstnet) {
2393 			if (srcnidp != NULL)
2394 				*srcnidp = ni->ni_nid;
2395 			if (orderp != NULL)
2396 				*orderp = order;
2397 			lnet_net_unlock(cpt);
2398 			return 1;
2399 		}
2400 
2401 		order++;
2402 	}
2403 
2404 	rn_list = lnet_net2rnethash(dstnet);
2405 	list_for_each(e, rn_list) {
2406 		rnet = list_entry(e, lnet_remotenet_t, lrn_list);
2407 
2408 		if (rnet->lrn_net == dstnet) {
2409 			lnet_route_t *route;
2410 			lnet_route_t *shortest = NULL;
2411 
2412 			LASSERT(!list_empty(&rnet->lrn_routes));
2413 
2414 			list_for_each_entry(route, &rnet->lrn_routes,
2415 						lr_list) {
2416 				if (shortest == NULL ||
2417 				    route->lr_hops < shortest->lr_hops)
2418 					shortest = route;
2419 			}
2420 
2421 			LASSERT(shortest != NULL);
2422 			hops = shortest->lr_hops;
2423 			if (srcnidp != NULL)
2424 				*srcnidp = shortest->lr_gateway->lp_ni->ni_nid;
2425 			if (orderp != NULL)
2426 				*orderp = order;
2427 			lnet_net_unlock(cpt);
2428 			return hops + 1;
2429 		}
2430 		order++;
2431 	}
2432 
2433 	lnet_net_unlock(cpt);
2434 	return -EHOSTUNREACH;
2435 }
2436 EXPORT_SYMBOL(LNetDist);
2437 
2438 /**
2439  * Set the number of asynchronous messages expected from a target process.
2440  *
2441  * This function is only meaningful for userspace callers. It's a no-op when
2442  * called from kernel.
2443  *
2444  * Asynchronous messages are those that can come from a target when the
2445  * userspace process is not waiting for IO to complete; e.g., AST callbacks
2446  * from Lustre servers. Specifying the expected number of such messages
2447  * allows them to be eagerly received when user process is not running in
2448  * LNet; otherwise network errors may occur.
2449  *
2450  * \param id Process ID of the target process.
2451  * \param nasync Number of asynchronous messages expected from the target.
2452  *
2453  * \return 0 on success, and an error code otherwise.
2454  */
2455 int
LNetSetAsync(lnet_process_id_t id,int nasync)2456 LNetSetAsync(lnet_process_id_t id, int nasync)
2457 {
2458 	return 0;
2459 }
2460 EXPORT_SYMBOL(LNetSetAsync);
2461