1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (C) 2014-2018 Giuseppe Lettieri
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /* $FreeBSD$ */
30
31 #if defined(__FreeBSD__)
32 #include <sys/cdefs.h> /* prerequisite */
33
34 #include <sys/types.h>
35 #include <sys/errno.h>
36 #include <sys/param.h> /* defines used in kernel.h */
37 #include <sys/kernel.h> /* types used in module initialization */
38 #include <sys/malloc.h>
39 #include <sys/poll.h>
40 #include <sys/lock.h>
41 #include <sys/rwlock.h>
42 #include <sys/selinfo.h>
43 #include <sys/sysctl.h>
44 #include <sys/socket.h> /* sockaddrs */
45 #include <net/if.h>
46 #include <net/if_var.h>
47 #include <machine/bus.h> /* bus_dmamap_* */
48 #include <sys/refcount.h>
49
50
51 #elif defined(linux)
52
53 #include "bsd_glue.h"
54
55 #elif defined(__APPLE__)
56
57 #warning OSX support is only partial
58 #include "osx_glue.h"
59
60 #elif defined(_WIN32)
61 #include "win_glue.h"
62
63 #else
64
65 #error Unsupported platform
66
67 #endif /* unsupported */
68
69 /*
70 * common headers
71 */
72
73 #include <net/netmap.h>
74 #include <dev/netmap/netmap_kern.h>
75 #include <dev/netmap/netmap_mem2.h>
76
77 #ifdef WITH_PIPES
78
79 #define NM_PIPE_MAXSLOTS 4096
80 #define NM_PIPE_MAXRINGS 256
81
82 static int netmap_default_pipes = 0; /* ignored, kept for compatibility */
83 SYSBEGIN(vars_pipes);
84 SYSCTL_DECL(_dev_netmap);
85 SYSCTL_INT(_dev_netmap, OID_AUTO, default_pipes, CTLFLAG_RW,
86 &netmap_default_pipes, 0, "For compatibility only");
87 SYSEND;
88
89 /* allocate the pipe array in the parent adapter */
90 static int
nm_pipe_alloc(struct netmap_adapter * na,u_int npipes)91 nm_pipe_alloc(struct netmap_adapter *na, u_int npipes)
92 {
93 size_t old_len, len;
94 struct netmap_pipe_adapter **npa;
95
96 if (npipes <= na->na_max_pipes)
97 /* we already have more entries that requested */
98 return 0;
99
100 if (npipes < na->na_next_pipe || npipes > NM_MAXPIPES)
101 return EINVAL;
102
103 old_len = sizeof(struct netmap_pipe_adapter *)*na->na_max_pipes;
104 len = sizeof(struct netmap_pipe_adapter *) * npipes;
105 npa = nm_os_realloc(na->na_pipes, len, old_len);
106 if (npa == NULL)
107 return ENOMEM;
108
109 na->na_pipes = npa;
110 na->na_max_pipes = npipes;
111
112 return 0;
113 }
114
115 /* deallocate the parent array in the parent adapter */
116 void
netmap_pipe_dealloc(struct netmap_adapter * na)117 netmap_pipe_dealloc(struct netmap_adapter *na)
118 {
119 if (na->na_pipes) {
120 if (na->na_next_pipe > 0) {
121 nm_prerr("freeing not empty pipe array for %s (%d dangling pipes)!",
122 na->name, na->na_next_pipe);
123 }
124 nm_os_free(na->na_pipes);
125 na->na_pipes = NULL;
126 na->na_max_pipes = 0;
127 na->na_next_pipe = 0;
128 }
129 }
130
131 /* find a pipe endpoint with the given id among the parent's pipes */
132 static struct netmap_pipe_adapter *
netmap_pipe_find(struct netmap_adapter * parent,const char * pipe_id)133 netmap_pipe_find(struct netmap_adapter *parent, const char *pipe_id)
134 {
135 int i;
136 struct netmap_pipe_adapter *na;
137
138 for (i = 0; i < parent->na_next_pipe; i++) {
139 const char *na_pipe_id;
140 na = parent->na_pipes[i];
141 na_pipe_id = strrchr(na->up.name,
142 na->role == NM_PIPE_ROLE_MASTER ? '{' : '}');
143 KASSERT(na_pipe_id != NULL, ("Invalid pipe name"));
144 ++na_pipe_id;
145 if (!strcmp(na_pipe_id, pipe_id)) {
146 return na;
147 }
148 }
149 return NULL;
150 }
151
152 /* add a new pipe endpoint to the parent array */
153 static int
netmap_pipe_add(struct netmap_adapter * parent,struct netmap_pipe_adapter * na)154 netmap_pipe_add(struct netmap_adapter *parent, struct netmap_pipe_adapter *na)
155 {
156 if (parent->na_next_pipe >= parent->na_max_pipes) {
157 u_int npipes = parent->na_max_pipes ? 2*parent->na_max_pipes : 2;
158 int error = nm_pipe_alloc(parent, npipes);
159 if (error)
160 return error;
161 }
162
163 parent->na_pipes[parent->na_next_pipe] = na;
164 na->parent_slot = parent->na_next_pipe;
165 parent->na_next_pipe++;
166 return 0;
167 }
168
169 /* remove the given pipe endpoint from the parent array */
170 static void
netmap_pipe_remove(struct netmap_adapter * parent,struct netmap_pipe_adapter * na)171 netmap_pipe_remove(struct netmap_adapter *parent, struct netmap_pipe_adapter *na)
172 {
173 u_int n;
174 n = --parent->na_next_pipe;
175 if (n != na->parent_slot) {
176 struct netmap_pipe_adapter **p =
177 &parent->na_pipes[na->parent_slot];
178 *p = parent->na_pipes[n];
179 (*p)->parent_slot = na->parent_slot;
180 }
181 parent->na_pipes[n] = NULL;
182 }
183
184 int
netmap_pipe_txsync(struct netmap_kring * txkring,int flags)185 netmap_pipe_txsync(struct netmap_kring *txkring, int flags)
186 {
187 struct netmap_kring *rxkring = txkring->pipe;
188 u_int k, lim = txkring->nkr_num_slots - 1, nk;
189 int m; /* slots to transfer */
190 int complete; /* did we see a complete packet ? */
191 struct netmap_ring *txring = txkring->ring, *rxring = rxkring->ring;
192
193 nm_prdis("%p: %s %x -> %s", txkring, txkring->name, flags, rxkring->name);
194 nm_prdis(20, "TX before: hwcur %d hwtail %d cur %d head %d tail %d",
195 txkring->nr_hwcur, txkring->nr_hwtail,
196 txkring->rcur, txkring->rhead, txkring->rtail);
197
198 /* update the hwtail */
199 txkring->nr_hwtail = txkring->pipe_tail;
200
201 m = txkring->rhead - txkring->nr_hwcur; /* new slots */
202 if (m < 0)
203 m += txkring->nkr_num_slots;
204
205 if (m == 0) {
206 /* nothing to send */
207 return 0;
208 }
209
210 for (k = txkring->nr_hwcur, nk = lim + 1, complete = 0; m;
211 m--, k = nm_next(k, lim), nk = (complete ? k : nk)) {
212 struct netmap_slot *rs = &rxring->slot[k];
213 struct netmap_slot *ts = &txring->slot[k];
214
215 *rs = *ts;
216 if (ts->flags & NS_BUF_CHANGED) {
217 ts->flags &= ~NS_BUF_CHANGED;
218 }
219 complete = !(ts->flags & NS_MOREFRAG);
220 }
221
222 txkring->nr_hwcur = k;
223
224 nm_prdis(20, "TX after : hwcur %d hwtail %d cur %d head %d tail %d k %d",
225 txkring->nr_hwcur, txkring->nr_hwtail,
226 txkring->rcur, txkring->rhead, txkring->rtail, k);
227
228 if (likely(nk <= lim)) {
229 mb(); /* make sure the slots are updated before publishing them */
230 rxkring->pipe_tail = nk; /* only publish complete packets */
231 rxkring->nm_notify(rxkring, 0);
232 }
233
234 return 0;
235 }
236
237 int
netmap_pipe_rxsync(struct netmap_kring * rxkring,int flags)238 netmap_pipe_rxsync(struct netmap_kring *rxkring, int flags)
239 {
240 struct netmap_kring *txkring = rxkring->pipe;
241 u_int k, lim = rxkring->nkr_num_slots - 1;
242 int m; /* slots to release */
243 struct netmap_ring *txring = txkring->ring, *rxring = rxkring->ring;
244
245 nm_prdis("%p: %s %x -> %s", txkring, txkring->name, flags, rxkring->name);
246 nm_prdis(20, "RX before: hwcur %d hwtail %d cur %d head %d tail %d",
247 rxkring->nr_hwcur, rxkring->nr_hwtail,
248 rxkring->rcur, rxkring->rhead, rxkring->rtail);
249
250 /* update the hwtail */
251 rxkring->nr_hwtail = rxkring->pipe_tail;
252
253 m = rxkring->rhead - rxkring->nr_hwcur; /* released slots */
254 if (m < 0)
255 m += rxkring->nkr_num_slots;
256
257 if (m == 0) {
258 /* nothing to release */
259 return 0;
260 }
261
262 for (k = rxkring->nr_hwcur; m; m--, k = nm_next(k, lim)) {
263 struct netmap_slot *rs = &rxring->slot[k];
264 struct netmap_slot *ts = &txring->slot[k];
265
266 if (rs->flags & NS_BUF_CHANGED) {
267 /* copy the slot and report the buffer change */
268 *ts = *rs;
269 rs->flags &= ~NS_BUF_CHANGED;
270 }
271 }
272
273 mb(); /* make sure the slots are updated before publishing them */
274 txkring->pipe_tail = nm_prev(k, lim);
275 rxkring->nr_hwcur = k;
276
277 nm_prdis(20, "RX after : hwcur %d hwtail %d cur %d head %d tail %d k %d",
278 rxkring->nr_hwcur, rxkring->nr_hwtail,
279 rxkring->rcur, rxkring->rhead, rxkring->rtail, k);
280
281 txkring->nm_notify(txkring, 0);
282
283 return 0;
284 }
285
286 /* Pipe endpoints are created and destroyed together, so that endopoints do not
287 * have to check for the existence of their peer at each ?xsync.
288 *
289 * To play well with the existing netmap infrastructure (refcounts etc.), we
290 * adopt the following strategy:
291 *
292 * 1) The first endpoint that is created also creates the other endpoint and
293 * grabs a reference to it.
294 *
295 * state A) user1 --> endpoint1 --> endpoint2
296 *
297 * 2) If, starting from state A, endpoint2 is then registered, endpoint1 gives
298 * its reference to the user:
299 *
300 * state B) user1 --> endpoint1 endpoint2 <--- user2
301 *
302 * 3) Assume that, starting from state B endpoint2 is closed. In the unregister
303 * callback endpoint2 notes that endpoint1 is still active and adds a reference
304 * from endpoint1 to itself. When user2 then releases her own reference,
305 * endpoint2 is not destroyed and we are back to state A. A symmetrical state
306 * would be reached if endpoint1 were released instead.
307 *
308 * 4) If, starting from state A, endpoint1 is closed, the destructor notes that
309 * it owns a reference to endpoint2 and releases it.
310 *
311 * Something similar goes on for the creation and destruction of the krings.
312 */
313
314
netmap_pipe_krings_create_both(struct netmap_adapter * na,struct netmap_adapter * ona)315 int netmap_pipe_krings_create_both(struct netmap_adapter *na,
316 struct netmap_adapter *ona)
317 {
318 enum txrx t;
319 int error;
320 int i;
321
322 /* case 1) below */
323 nm_prdis("%p: case 1, create both ends", na);
324 error = netmap_krings_create(na, 0);
325 if (error)
326 return error;
327
328 /* create the krings of the other end */
329 error = netmap_krings_create(ona, 0);
330 if (error)
331 goto del_krings1;
332
333 /* cross link the krings and initialize the pipe_tails */
334 for_rx_tx(t) {
335 enum txrx r = nm_txrx_swap(t); /* swap NR_TX <-> NR_RX */
336 for (i = 0; i < nma_get_nrings(na, t); i++) {
337 struct netmap_kring *k1 = NMR(na, t)[i],
338 *k2 = NMR(ona, r)[i];
339 k1->pipe = k2;
340 k2->pipe = k1;
341 /* mark all peer-adapter rings as fake */
342 k2->nr_kflags |= NKR_FAKERING;
343 /* init tails */
344 k1->pipe_tail = k1->nr_hwtail;
345 k2->pipe_tail = k2->nr_hwtail;
346 }
347 }
348
349 return 0;
350
351 del_krings1:
352 netmap_krings_delete(na);
353 return error;
354 }
355
356 /* netmap_pipe_krings_create.
357 *
358 * There are two cases:
359 *
360 * 1) state is
361 *
362 * usr1 --> e1 --> e2
363 *
364 * and we are e1. We have to create both sets
365 * of krings.
366 *
367 * 2) state is
368 *
369 * usr1 --> e1 --> e2
370 *
371 * and we are e2. e1 is certainly registered and our
372 * krings already exist. Nothing to do.
373 */
374 static int
netmap_pipe_krings_create(struct netmap_adapter * na)375 netmap_pipe_krings_create(struct netmap_adapter *na)
376 {
377 struct netmap_pipe_adapter *pna =
378 (struct netmap_pipe_adapter *)na;
379 struct netmap_adapter *ona = &pna->peer->up;
380
381 if (pna->peer_ref)
382 return netmap_pipe_krings_create_both(na, ona);
383
384 return 0;
385 }
386
387 int
netmap_pipe_reg_both(struct netmap_adapter * na,struct netmap_adapter * ona)388 netmap_pipe_reg_both(struct netmap_adapter *na, struct netmap_adapter *ona)
389 {
390 int i, error = 0;
391 enum txrx t;
392
393 for_rx_tx(t) {
394 for (i = 0; i < nma_get_nrings(na, t); i++) {
395 struct netmap_kring *kring = NMR(na, t)[i];
396
397 if (nm_kring_pending_on(kring)) {
398 /* mark the peer ring as needed */
399 kring->pipe->nr_kflags |= NKR_NEEDRING;
400 }
401 }
402 }
403
404 /* create all missing needed rings on the other end.
405 * Either our end, or the other, has been marked as
406 * fake, so the allocation will not be done twice.
407 */
408 error = netmap_mem_rings_create(ona);
409 if (error)
410 return error;
411
412 /* In case of no error we put our rings in netmap mode */
413 for_rx_tx(t) {
414 for (i = 0; i < nma_get_nrings(na, t); i++) {
415 struct netmap_kring *kring = NMR(na, t)[i];
416 if (nm_kring_pending_on(kring)) {
417 struct netmap_kring *sring, *dring;
418
419 kring->nr_mode = NKR_NETMAP_ON;
420 if ((kring->nr_kflags & NKR_FAKERING) &&
421 (kring->pipe->nr_kflags & NKR_FAKERING)) {
422 /* this is a re-open of a pipe
423 * end-point kept alive by the other end.
424 * We need to leave everything as it is
425 */
426 continue;
427 }
428
429 /* copy the buffers from the non-fake ring */
430 if (kring->nr_kflags & NKR_FAKERING) {
431 sring = kring->pipe;
432 dring = kring;
433 } else {
434 sring = kring;
435 dring = kring->pipe;
436 }
437 memcpy(dring->ring->slot,
438 sring->ring->slot,
439 sizeof(struct netmap_slot) *
440 sring->nkr_num_slots);
441 /* mark both rings as fake and needed,
442 * so that buffers will not be
443 * deleted by the standard machinery
444 * (we will delete them by ourselves in
445 * netmap_pipe_krings_delete)
446 */
447 sring->nr_kflags |=
448 (NKR_FAKERING | NKR_NEEDRING);
449 dring->nr_kflags |=
450 (NKR_FAKERING | NKR_NEEDRING);
451 kring->nr_mode = NKR_NETMAP_ON;
452 }
453 }
454 }
455
456 return 0;
457 }
458
459 /* netmap_pipe_reg.
460 *
461 * There are two cases on registration (onoff==1)
462 *
463 * 1.a) state is
464 *
465 * usr1 --> e1 --> e2
466 *
467 * and we are e1. Create the needed rings of the
468 * other end.
469 *
470 * 1.b) state is
471 *
472 * usr1 --> e1 --> e2 <-- usr2
473 *
474 * and we are e2. Drop the ref e1 is holding.
475 *
476 * There are two additional cases on unregister (onoff==0)
477 *
478 * 2.a) state is
479 *
480 * usr1 --> e1 --> e2
481 *
482 * and we are e1. Nothing special to do, e2 will
483 * be cleaned up by the destructor of e1.
484 *
485 * 2.b) state is
486 *
487 * usr1 --> e1 e2 <-- usr2
488 *
489 * and we are either e1 or e2. Add a ref from the
490 * other end.
491 */
492 static int
netmap_pipe_reg(struct netmap_adapter * na,int onoff)493 netmap_pipe_reg(struct netmap_adapter *na, int onoff)
494 {
495 struct netmap_pipe_adapter *pna =
496 (struct netmap_pipe_adapter *)na;
497 struct netmap_adapter *ona = &pna->peer->up;
498 int error = 0;
499
500 nm_prdis("%p: onoff %d", na, onoff);
501 if (onoff) {
502 error = netmap_pipe_reg_both(na, ona);
503 if (error) {
504 return error;
505 }
506 if (na->active_fds == 0)
507 na->na_flags |= NAF_NETMAP_ON;
508 } else {
509 if (na->active_fds == 0)
510 na->na_flags &= ~NAF_NETMAP_ON;
511 netmap_krings_mode_commit(na, onoff);
512 }
513
514 if (na->active_fds) {
515 nm_prdis("active_fds %d", na->active_fds);
516 return 0;
517 }
518
519 if (pna->peer_ref) {
520 nm_prdis("%p: case 1.a or 2.a, nothing to do", na);
521 return 0;
522 }
523 if (onoff) {
524 nm_prdis("%p: case 1.b, drop peer", na);
525 pna->peer->peer_ref = 0;
526 netmap_adapter_put(na);
527 } else {
528 nm_prdis("%p: case 2.b, grab peer", na);
529 netmap_adapter_get(na);
530 pna->peer->peer_ref = 1;
531 }
532 return error;
533 }
534
535 void
netmap_pipe_krings_delete_both(struct netmap_adapter * na,struct netmap_adapter * ona)536 netmap_pipe_krings_delete_both(struct netmap_adapter *na,
537 struct netmap_adapter *ona)
538 {
539 struct netmap_adapter *sna;
540 enum txrx t;
541 int i;
542
543 /* case 1) below */
544 nm_prdis("%p: case 1, deleting everything", na);
545 /* To avoid double-frees we zero-out all the buffers in the kernel part
546 * of each ring. The reason is this: If the user is behaving correctly,
547 * all buffers are found in exactly one slot in the userspace part of
548 * some ring. If the user is not behaving correctly, we cannot release
549 * buffers cleanly anyway. In the latter case, the allocator will
550 * return to a clean state only when all its users will close.
551 */
552 sna = na;
553 cleanup:
554 for_rx_tx(t) {
555 for (i = 0; i < nma_get_nrings(sna, t); i++) {
556 struct netmap_kring *kring = NMR(sna, t)[i];
557 struct netmap_ring *ring = kring->ring;
558 uint32_t j, lim = kring->nkr_num_slots - 1;
559
560 nm_prdis("%s ring %p hwtail %u hwcur %u",
561 kring->name, ring, kring->nr_hwtail, kring->nr_hwcur);
562
563 if (ring == NULL)
564 continue;
565
566 if (kring->tx == NR_RX)
567 ring->slot[kring->pipe_tail].buf_idx = 0;
568
569 for (j = nm_next(kring->pipe_tail, lim);
570 j != kring->nr_hwcur;
571 j = nm_next(j, lim))
572 {
573 nm_prdis("%s[%d] %u", kring->name, j, ring->slot[j].buf_idx);
574 ring->slot[j].buf_idx = 0;
575 }
576 kring->nr_kflags &= ~(NKR_FAKERING | NKR_NEEDRING);
577 }
578
579 }
580 if (sna != ona && ona->tx_rings) {
581 sna = ona;
582 goto cleanup;
583 }
584
585 netmap_mem_rings_delete(na);
586 netmap_krings_delete(na); /* also zeroes tx_rings etc. */
587
588 if (ona->tx_rings == NULL) {
589 /* already deleted, we must be on an
590 * cleanup-after-error path */
591 return;
592 }
593 netmap_mem_rings_delete(ona);
594 netmap_krings_delete(ona);
595 }
596
597 /* netmap_pipe_krings_delete.
598 *
599 * There are two cases:
600 *
601 * 1) state is
602 *
603 * usr1 --> e1 --> e2
604 *
605 * and we are e1 (e2 is not registered, so krings_delete cannot be
606 * called on it);
607 *
608 * 2) state is
609 *
610 * usr1 --> e1 e2 <-- usr2
611 *
612 * and we are either e1 or e2.
613 *
614 * In the former case we have to also delete the krings of e2;
615 * in the latter case we do nothing.
616 */
617 static void
netmap_pipe_krings_delete(struct netmap_adapter * na)618 netmap_pipe_krings_delete(struct netmap_adapter *na)
619 {
620 struct netmap_pipe_adapter *pna =
621 (struct netmap_pipe_adapter *)na;
622 struct netmap_adapter *ona; /* na of the other end */
623
624 if (!pna->peer_ref) {
625 nm_prdis("%p: case 2, kept alive by peer", na);
626 return;
627 }
628 ona = &pna->peer->up;
629 netmap_pipe_krings_delete_both(na, ona);
630 }
631
632
633 static void
netmap_pipe_dtor(struct netmap_adapter * na)634 netmap_pipe_dtor(struct netmap_adapter *na)
635 {
636 struct netmap_pipe_adapter *pna =
637 (struct netmap_pipe_adapter *)na;
638 nm_prdis("%p %p", na, pna->parent_ifp);
639 if (pna->peer_ref) {
640 nm_prdis("%p: clean up peer", na);
641 pna->peer_ref = 0;
642 netmap_adapter_put(&pna->peer->up);
643 }
644 if (pna->role == NM_PIPE_ROLE_MASTER)
645 netmap_pipe_remove(pna->parent, pna);
646 if (pna->parent_ifp)
647 if_rele(pna->parent_ifp);
648 netmap_adapter_put(pna->parent);
649 pna->parent = NULL;
650 }
651
652 int
netmap_get_pipe_na(struct nmreq_header * hdr,struct netmap_adapter ** na,struct netmap_mem_d * nmd,int create)653 netmap_get_pipe_na(struct nmreq_header *hdr, struct netmap_adapter **na,
654 struct netmap_mem_d *nmd, int create)
655 {
656 struct nmreq_register *req = (struct nmreq_register *)(uintptr_t)hdr->nr_body;
657 struct netmap_adapter *pna; /* parent adapter */
658 struct netmap_pipe_adapter *mna, *sna, *reqna;
659 struct ifnet *ifp = NULL;
660 const char *pipe_id = NULL;
661 int role = 0;
662 int error, retries = 0;
663 char *cbra;
664
665 /* Try to parse the pipe syntax 'xx{yy' or 'xx}yy'. */
666 cbra = strrchr(hdr->nr_name, '{');
667 if (cbra != NULL) {
668 role = NM_PIPE_ROLE_MASTER;
669 } else {
670 cbra = strrchr(hdr->nr_name, '}');
671 if (cbra != NULL) {
672 role = NM_PIPE_ROLE_SLAVE;
673 } else {
674 nm_prdis("not a pipe");
675 return 0;
676 }
677 }
678 pipe_id = cbra + 1;
679 if (*pipe_id == '\0' || cbra == hdr->nr_name) {
680 /* Bracket is the last character, so pipe name is missing;
681 * or bracket is the first character, so base port name
682 * is missing. */
683 return EINVAL;
684 }
685
686 if (req->nr_mode != NR_REG_ALL_NIC && req->nr_mode != NR_REG_ONE_NIC) {
687 /* We only accept modes involving hardware rings. */
688 return EINVAL;
689 }
690
691 /* first, try to find the parent adapter */
692 for (;;) {
693 char nr_name_orig[NETMAP_REQ_IFNAMSIZ];
694 int create_error;
695
696 /* Temporarily remove the pipe suffix. */
697 strlcpy(nr_name_orig, hdr->nr_name, sizeof(nr_name_orig));
698 *cbra = '\0';
699 error = netmap_get_na(hdr, &pna, &ifp, nmd, create);
700 /* Restore the pipe suffix. */
701 strlcpy(hdr->nr_name, nr_name_orig, sizeof(hdr->nr_name));
702 if (!error)
703 break;
704 if (error != ENXIO || retries++) {
705 nm_prdis("parent lookup failed: %d", error);
706 return error;
707 }
708 nm_prdis("try to create a persistent vale port");
709 /* create a persistent vale port and try again */
710 *cbra = '\0';
711 NMG_UNLOCK();
712 create_error = netmap_vi_create(hdr, 1 /* autodelete */);
713 NMG_LOCK();
714 strlcpy(hdr->nr_name, nr_name_orig, sizeof(hdr->nr_name));
715 if (create_error && create_error != EEXIST) {
716 if (create_error != EOPNOTSUPP) {
717 nm_prerr("failed to create a persistent vale port: %d",
718 create_error);
719 }
720 return error;
721 }
722 }
723
724 if (NETMAP_OWNED_BY_KERN(pna)) {
725 nm_prdis("parent busy");
726 error = EBUSY;
727 goto put_out;
728 }
729
730 /* next, lookup the pipe id in the parent list */
731 reqna = NULL;
732 mna = netmap_pipe_find(pna, pipe_id);
733 if (mna) {
734 if (mna->role == role) {
735 nm_prdis("found %s directly at %d", pipe_id, mna->parent_slot);
736 reqna = mna;
737 } else {
738 nm_prdis("found %s indirectly at %d", pipe_id, mna->parent_slot);
739 reqna = mna->peer;
740 }
741 /* the pipe we have found already holds a ref to the parent,
742 * so we need to drop the one we got from netmap_get_na()
743 */
744 netmap_unget_na(pna, ifp);
745 goto found;
746 }
747 nm_prdis("pipe %s not found, create %d", pipe_id, create);
748 if (!create) {
749 error = ENODEV;
750 goto put_out;
751 }
752 /* we create both master and slave.
753 * The endpoint we were asked for holds a reference to
754 * the other one.
755 */
756 mna = nm_os_malloc(sizeof(*mna));
757 if (mna == NULL) {
758 error = ENOMEM;
759 goto put_out;
760 }
761 snprintf(mna->up.name, sizeof(mna->up.name), "%s{%s", pna->name, pipe_id);
762
763 mna->role = NM_PIPE_ROLE_MASTER;
764 mna->parent = pna;
765 mna->parent_ifp = ifp;
766
767 mna->up.nm_txsync = netmap_pipe_txsync;
768 mna->up.nm_rxsync = netmap_pipe_rxsync;
769 mna->up.nm_register = netmap_pipe_reg;
770 mna->up.nm_dtor = netmap_pipe_dtor;
771 mna->up.nm_krings_create = netmap_pipe_krings_create;
772 mna->up.nm_krings_delete = netmap_pipe_krings_delete;
773 mna->up.nm_mem = netmap_mem_get(pna->nm_mem);
774 mna->up.na_flags |= NAF_MEM_OWNER;
775 mna->up.na_lut = pna->na_lut;
776
777 mna->up.num_tx_rings = req->nr_tx_rings;
778 nm_bound_var(&mna->up.num_tx_rings, 1,
779 1, NM_PIPE_MAXRINGS, NULL);
780 mna->up.num_rx_rings = req->nr_rx_rings;
781 nm_bound_var(&mna->up.num_rx_rings, 1,
782 1, NM_PIPE_MAXRINGS, NULL);
783 mna->up.num_tx_desc = req->nr_tx_slots;
784 nm_bound_var(&mna->up.num_tx_desc, pna->num_tx_desc,
785 1, NM_PIPE_MAXSLOTS, NULL);
786 mna->up.num_rx_desc = req->nr_rx_slots;
787 nm_bound_var(&mna->up.num_rx_desc, pna->num_rx_desc,
788 1, NM_PIPE_MAXSLOTS, NULL);
789 error = netmap_attach_common(&mna->up);
790 if (error)
791 goto free_mna;
792 /* register the master with the parent */
793 error = netmap_pipe_add(pna, mna);
794 if (error)
795 goto free_mna;
796
797 /* create the slave */
798 sna = nm_os_malloc(sizeof(*mna));
799 if (sna == NULL) {
800 error = ENOMEM;
801 goto unregister_mna;
802 }
803 /* most fields are the same, copy from master and then fix */
804 *sna = *mna;
805 sna->up.nm_mem = netmap_mem_get(mna->up.nm_mem);
806 /* swap the number of tx/rx rings and slots */
807 sna->up.num_tx_rings = mna->up.num_rx_rings;
808 sna->up.num_tx_desc = mna->up.num_rx_desc;
809 sna->up.num_rx_rings = mna->up.num_tx_rings;
810 sna->up.num_rx_desc = mna->up.num_tx_desc;
811 snprintf(sna->up.name, sizeof(sna->up.name), "%s}%s", pna->name, pipe_id);
812 sna->role = NM_PIPE_ROLE_SLAVE;
813 error = netmap_attach_common(&sna->up);
814 if (error)
815 goto free_sna;
816
817 /* join the two endpoints */
818 mna->peer = sna;
819 sna->peer = mna;
820
821 /* we already have a reference to the parent, but we
822 * need another one for the other endpoint we created
823 */
824 netmap_adapter_get(pna);
825 /* likewise for the ifp, if any */
826 if (ifp)
827 if_ref(ifp);
828
829 if (role == NM_PIPE_ROLE_MASTER) {
830 reqna = mna;
831 mna->peer_ref = 1;
832 netmap_adapter_get(&sna->up);
833 } else {
834 reqna = sna;
835 sna->peer_ref = 1;
836 netmap_adapter_get(&mna->up);
837 }
838 nm_prdis("created master %p and slave %p", mna, sna);
839 found:
840
841 nm_prdis("pipe %s %s at %p", pipe_id,
842 (reqna->role == NM_PIPE_ROLE_MASTER ? "master" : "slave"), reqna);
843 *na = &reqna->up;
844 netmap_adapter_get(*na);
845
846 /* keep the reference to the parent.
847 * It will be released by the req destructor
848 */
849
850 return 0;
851
852 free_sna:
853 nm_os_free(sna);
854 unregister_mna:
855 netmap_pipe_remove(pna, mna);
856 free_mna:
857 nm_os_free(mna);
858 put_out:
859 netmap_unget_na(pna, ifp);
860 return error;
861 }
862
863
864 #endif /* WITH_PIPES */
865