xref: /illumos-gate/usr/src/uts/intel/io/viona/viona_main.c (revision ce489a5e0a9c618cd26b66d6b1fd397d8545db9b)
1 /*
2  * Copyright (c) 2013  Chris Torek <torek @ torek net>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 /*
27  * This file and its contents are supplied under the terms of the
28  * Common Development and Distribution License ("CDDL"), version 1.0.
29  * You may only use this file in accordance with the terms of version
30  * 1.0 of the CDDL.
31  *
32  * A full copy of the text of the CDDL should have accompanied this
33  * source.  A copy of the CDDL is also available via the Internet at
34  * http://www.illumos.org/license/CDDL.
35  *
36  * Copyright 2015 Pluribus Networks Inc.
37  * Copyright 2019 Joyent, Inc.
38  * Copyright 2022 OmniOS Community Edition (OmniOSce) Association.
39  * Copyright 2024 Oxide Computer Company
40  */
41 
42 /*
43  * viona - VirtIO-Net, Accelerated
44  *
45  * The purpose of viona is to provide high performance virtio-net devices to
46  * bhyve guests.  It does so by sitting directly atop MAC, skipping all of the
47  * DLS/DLD stack.
48  *
49  * --------------------
50  * General Architecture
51  * --------------------
52  *
53  * A single viona instance is comprised of a "link" handle and two "rings".
54  * After opening the viona device, it must be associated with a MAC network
55  * interface and a bhyve (vmm) instance to form its link resource.  This is
56  * done with the VNA_IOC_CREATE ioctl, where the datalink ID and vmm fd are
57  * passed in to perform the initialization.  With the MAC client opened, and a
58  * driver handle to the vmm instance established, the device is ready to be
59  * configured by the guest.
60  *
61  * The userspace portion of bhyve, which interfaces with the PCI device
62  * emulation framework, is meant to stay out of the datapath if at all
63  * possible.  Configuration changes made via PCI are mapped to actions which
64  * will steer the operation of the in-kernel logic.
65  *
66  *
67  * -----------
68  * Ring Basics
69  * -----------
70  *
71  * Each viona link has two viona_vring_t entities, RX and TX, for handling data
72  * transfers to and from the guest.  They represent an interface to the
73  * standard virtio ring structures.  When initialized and active, each ring is
74  * backed by a kernel worker thread (parented to the bhyve process for the
75  * instance) which handles ring events.  The RX worker has the simple task of
76  * watching for ring shutdown conditions.  The TX worker does that in addition
77  * to processing all requests to transmit data.  Data destined for the guest is
78  * delivered directly by MAC to viona_rx() when the ring is active.
79  *
80  *
81  * -----------
82  * Ring States
83  * -----------
84  *
85  * The viona_vring_t instances follow a simple path through the possible state
86  * values represented in virtio_vring_t`vr_state:
87  *
88  *        +<--------------------------------------------+
89  *        |						|
90  *        V						^
91  *  +-----------+	This is the initial state when a link is created or
92  *  | VRS_RESET |	when the ring has been explicitly reset.
93  *  +-----------+
94  *        |						^
95  *        |---* ioctl(VNA_IOC_RING_INIT) issued		|
96  *        |						|
97  *        |						^
98  *        V
99  *  +-----------+	The ring parameters (size, guest physical addresses)
100  *  | VRS_SETUP |	have been set and start-up of the ring worker thread
101  *  +-----------+	has begun.
102  *        |						^
103  *        |						|
104  *        |---* ring worker thread begins execution	|
105  *        |						|
106  *        +-------------------------------------------->+
107  *        |	      |					^
108  *        |	      |
109  *        |	      *	If ring shutdown is requested (by ioctl or impending
110  *        |		bhyve process death) while the worker thread is
111  *        |		starting, the worker will transition the ring to
112  *        |		VRS_RESET and exit.
113  *        |						^
114  *        |						|
115  *        |<-------------------------------------------<+
116  *        |	      |					|
117  *        |	      |					^
118  *        |	      *	If ring is requested to pause (but not stop)from the
119  *        |             VRS_RUN state, it will return to the VRS_INIT state.
120  *        |
121  *        |						^
122  *        |						|
123  *        |						^
124  *        V
125  *  +-----------+	The worker thread associated with the ring has started
126  *  | VRS_INIT  |	executing.  It has allocated any extra resources needed
127  *  +-----------+	for the ring to operate.
128  *        |						^
129  *        |						|
130  *        +-------------------------------------------->+
131  *        |	      |					^
132  *        |	      |
133  *        |	      *	If ring shutdown is requested while the worker is
134  *        |		waiting in VRS_INIT, it will free any extra resources
135  *        |		and transition to VRS_RESET.
136  *        |						^
137  *        |						|
138  *        |--* ioctl(VNA_IOC_RING_KICK) issued		|
139  *        |						^
140  *        V
141  *  +-----------+	The worker thread associated with the ring is executing
142  *  | VRS_RUN   |	workload specific to that ring.
143  *  +-----------+
144  *        |						^
145  *        |---* ioctl(VNA_IOC_RING_RESET) issued	|
146  *        |	(or bhyve process begins exit)		^
147  *        |
148  *  +-----------+	The worker thread associated with the ring is in the
149  *  | VRS_STOP  |	process of exiting. All outstanding TX and RX
150  *  +-----------+	requests are allowed to complete, but new requests
151  *        |		must be ignored.
152  *        |						^
153  *        |						|
154  *        +-------------------------------------------->+
155  *
156  *
157  * While the worker thread is not running, changes to vr_state are only made by
158  * viona_ioc_ring_init() under vr_lock.  There, it initializes the ring, starts
159  * the worker, and sets the ring state to VRS_SETUP.  Once the worker thread
160  * has been started, only it may perform ring state transitions (still under
161  * the protection of vr_lock), when requested by outside consumers via
162  * vr_state_flags or when the containing bhyve process initiates an exit.
163  *
164  *
165  * ----------------------------
166  * Transmission mblk_t Handling
167  * ----------------------------
168  *
169  * For incoming frames destined for a bhyve guest, the data must first land in
170  * a host OS buffer from the physical NIC before it is copied into the awaiting
171  * guest buffer(s).  Outbound frames transmitted by the guest are not bound by
172  * this limitation and can avoid extra copying before the buffers are accessed
173  * directly by the NIC.  When a guest designates buffers to be transmitted,
174  * viona translates the guest-physical addresses contained in the ring
175  * descriptors to host-virtual addresses via viona_hold_page().  That pointer is
176  * wrapped in an mblk_t using a preallocated viona_desb_t for the desballoc().
177  * Doing so increments vr_xfer_outstanding, preventing the ring from being
178  * reset (allowing the link to drop its vmm handle to the guest) until all
179  * transmit mblks referencing guest memory have been processed.  Allocation of
180  * the viona_desb_t entries is done during the VRS_INIT stage of the ring
181  * worker thread.  The ring size informs that allocation as the number of
182  * concurrent transmissions is limited by the number of descriptors in the
183  * ring.  This minimizes allocation in the transmit hot-path by acquiring those
184  * fixed-size resources during initialization.
185  *
186  * This optimization depends on the underlying NIC driver freeing the mblks in
187  * a timely manner after they have been transmitted by the hardware.  Some
188  * drivers have been found to flush TX descriptors only when new transmissions
189  * are initiated.  This means that there is no upper bound to the time needed
190  * for an mblk to be flushed and can stall bhyve guests from shutting down
191  * since their memory must be free of viona TX references prior to clean-up.
192  *
193  * This expectation of deterministic mblk_t processing is likely the reason
194  * behind the notable exception to the zero-copy TX path: systems with 'bnxe'
195  * loaded will copy transmit data into fresh buffers rather than passing up
196  * zero-copy mblks.  It is a hold-over from the original viona sources provided
197  * by Pluribus and its continued necessity has not been confirmed.
198  *
199  *
200  * ----------------------------
201  * Ring Notification Fast-paths
202  * ----------------------------
203  *
204  * Device operation for viona requires that notifications flow to and from the
205  * guest to indicate certain ring conditions.  In order to minimize latency and
206  * processing overhead, the notification procedures are kept in-kernel whenever
207  * possible.
208  *
209  * Guest-to-host notifications, when new available descriptors have been placed
210  * in the ring, are posted via the 'queue notify' address in the virtio BAR.
211  * The vmm_drv_ioport_hook() interface was added to bhyve which allows viona to
212  * install a callback hook on an ioport address.  Guest exits for accesses to
213  * viona-hooked ioport addresses will result in direct calls to notify the
214  * appropriate ring worker without a trip to userland.
215  *
216  * Host-to-guest notifications in the form of interrupts enjoy similar
217  * acceleration.  Each viona ring can be configured to send MSI notifications
218  * to the guest as virtio conditions dictate.  This in-kernel interrupt
219  * configuration is kept synchronized through viona ioctls which are utilized
220  * during writes to the associated PCI config registers or MSI-X BAR.
221  *
222  * Guests which do not utilize MSI-X will result in viona falling back to the
223  * slow path for interrupts.  It will poll(2) the viona handle, receiving
224  * notification when ring events necessitate the assertion of an interrupt.
225  *
226  *
227  * ---------------
228  * Nethook Support
229  * ---------------
230  *
231  * Viona provides four nethook events that consumers (e.g. ipf) can hook into
232  * to intercept packets as they go up or down the stack.  Unfortunately,
233  * the nethook framework does not understand raw packets, so we can only
234  * generate events (in, out) for IPv4 and IPv6 packets.  At driver attach,
235  * we register callbacks with the neti (netinfo) module that will be invoked
236  * for each netstack already present, as well as for any additional netstack
237  * instances created as the system operates.  These callbacks will
238  * register/unregister the hooks with the nethook framework for each
239  * netstack instance.  This registration occurs prior to creating any
240  * viona instances for a given netstack, and the unregistration for a netstack
241  * instance occurs after all viona instances of the netstack instance have
242  * been deleted.
243  *
244  * ------------------
245  * Metrics/Statistics
246  * -----------------
247  *
248  * During operation, Viona tracks certain metrics as certain events occur.
249  *
250  * One class of metrics, known as the "error stats", refer to abnormal
251  * conditions in ring processing which are likely the fault of a misbehaving
252  * guest.  These are tracked on a per-ring basis, and are not formally exposed
253  * to any consumer besides direct memory access through mdb.
254  *
255  * The other class of metrics tracked for an instance are the "transfer stats",
256  * which are the traditional packets/bytes/errors/drops figures.  These are
257  * counted per-ring, and then aggregated into link-wide values exposed via
258  * kstats.  Atomic operations are used to increment those per-ring stats during
259  * operation, and then when a ring is stopped, the values are consolidated into
260  * the link-wide values (to prevent loss when the ring is zeroed) under the
261  * protection of viona_link`l_stats_lock.  When the kstats are being updated,
262  * l_stats_lock is held to protect against a racing consolidation, with the
263  * existing per-ring values being added in at update time to provide an accurate
264  * figure.
265  */
266 
267 #include <sys/conf.h>
268 #include <sys/file.h>
269 #include <sys/stat.h>
270 
271 #include <sys/dlpi.h>
272 #include <sys/vlan.h>
273 
274 #include "viona_impl.h"
275 
276 
277 #define	VIONA_NAME		"Virtio Network Accelerator"
278 #define	VIONA_CTL_MINOR		0
279 #define	VIONA_MODULE_NAME	"viona"
280 #define	VIONA_KSTAT_CLASS	"misc"
281 #define	VIONA_KSTAT_NAME	"viona_stat"
282 
283 
284 /*
285  * Host capabilities.
286  */
287 #define	VIONA_S_HOSTCAPS	(	\
288 	VIRTIO_NET_F_GUEST_CSUM |	\
289 	VIRTIO_NET_F_MAC |		\
290 	VIRTIO_NET_F_GUEST_TSO4 |	\
291 	VIRTIO_NET_F_MRG_RXBUF |	\
292 	VIRTIO_NET_F_STATUS |		\
293 	VIRTIO_F_RING_NOTIFY_ON_EMPTY |	\
294 	VIRTIO_F_RING_INDIRECT_DESC)
295 
296 /* MAC_CAPAB_HCKSUM specifics of interest */
297 #define	VIONA_CAP_HCKSUM_INTEREST	\
298 	(HCKSUM_INET_PARTIAL |		\
299 	HCKSUM_INET_FULL_V4 |		\
300 	HCKSUM_INET_FULL_V6)
301 
302 static void		*viona_state;
303 static dev_info_t	*viona_dip;
304 static id_space_t	*viona_minors;
305 
306 
307 static int viona_info(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg,
308     void **result);
309 static int viona_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);
310 static int viona_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);
311 static int viona_open(dev_t *devp, int flag, int otype, cred_t *credp);
312 static int viona_close(dev_t dev, int flag, int otype, cred_t *credp);
313 static int viona_ioctl(dev_t dev, int cmd, intptr_t data, int mode,
314     cred_t *credp, int *rval);
315 static int viona_chpoll(dev_t dev, short events, int anyyet, short *reventsp,
316     struct pollhead **phpp);
317 
318 static int viona_ioc_create(viona_soft_state_t *, void *, int, cred_t *);
319 static int viona_ioc_delete(viona_soft_state_t *, boolean_t);
320 
321 static int viona_ioc_set_notify_ioport(viona_link_t *, uint16_t);
322 static int viona_ioc_set_promisc(viona_link_t *, viona_promisc_t);
323 static int viona_ioc_get_params(viona_link_t *, void *, int);
324 static int viona_ioc_set_params(viona_link_t *, void *, int);
325 static int viona_ioc_ring_init(viona_link_t *, void *, int);
326 static int viona_ioc_ring_set_state(viona_link_t *, void *, int);
327 static int viona_ioc_ring_get_state(viona_link_t *, void *, int);
328 static int viona_ioc_ring_reset(viona_link_t *, uint_t);
329 static int viona_ioc_ring_kick(viona_link_t *, uint_t);
330 static int viona_ioc_ring_pause(viona_link_t *, uint_t);
331 static int viona_ioc_ring_set_msi(viona_link_t *, void *, int);
332 static int viona_ioc_ring_intr_clear(viona_link_t *, uint_t);
333 static int viona_ioc_intr_poll(viona_link_t *, void *, int, int *);
334 
335 static void viona_params_get_defaults(viona_link_params_t *);
336 
337 static struct cb_ops viona_cb_ops = {
338 	viona_open,
339 	viona_close,
340 	nodev,
341 	nodev,
342 	nodev,
343 	nodev,
344 	nodev,
345 	viona_ioctl,
346 	nodev,
347 	nodev,
348 	nodev,
349 	viona_chpoll,
350 	ddi_prop_op,
351 	0,
352 	D_MP | D_NEW | D_HOTPLUG,
353 	CB_REV,
354 	nodev,
355 	nodev
356 };
357 
358 static struct dev_ops viona_ops = {
359 	DEVO_REV,
360 	0,
361 	viona_info,
362 	nulldev,
363 	nulldev,
364 	viona_attach,
365 	viona_detach,
366 	nodev,
367 	&viona_cb_ops,
368 	NULL,
369 	ddi_power,
370 	ddi_quiesce_not_needed
371 };
372 
373 static struct modldrv modldrv = {
374 	&mod_driverops,
375 	VIONA_NAME,
376 	&viona_ops,
377 };
378 
379 static struct modlinkage modlinkage = {
380 	MODREV_1, &modldrv, NULL
381 };
382 
383 int
_init(void)384 _init(void)
385 {
386 	int ret;
387 
388 	ret = ddi_soft_state_init(&viona_state, sizeof (viona_soft_state_t), 0);
389 	if (ret != 0) {
390 		return (ret);
391 	}
392 
393 	viona_minors = id_space_create("viona_minors",
394 	    VIONA_CTL_MINOR + 1, UINT16_MAX);
395 	viona_rx_init();
396 	mutex_init(&viona_force_copy_lock, NULL, MUTEX_DRIVER, NULL);
397 
398 	ret = mod_install(&modlinkage);
399 	if (ret != 0) {
400 		ddi_soft_state_fini(&viona_state);
401 		id_space_destroy(viona_minors);
402 		viona_rx_fini();
403 		mutex_destroy(&viona_force_copy_lock);
404 	}
405 
406 	return (ret);
407 }
408 
409 int
_fini(void)410 _fini(void)
411 {
412 	int ret;
413 
414 	ret = mod_remove(&modlinkage);
415 	if (ret != 0) {
416 		return (ret);
417 	}
418 
419 	ddi_soft_state_fini(&viona_state);
420 	id_space_destroy(viona_minors);
421 	viona_rx_fini();
422 	mutex_destroy(&viona_force_copy_lock);
423 
424 	return (ret);
425 }
426 
427 int
_info(struct modinfo * modinfop)428 _info(struct modinfo *modinfop)
429 {
430 	return (mod_info(&modlinkage, modinfop));
431 }
432 
433 /* ARGSUSED */
434 static int
viona_info(dev_info_t * dip,ddi_info_cmd_t cmd,void * arg,void ** result)435 viona_info(dev_info_t *dip, ddi_info_cmd_t cmd, void *arg, void **result)
436 {
437 	int error;
438 
439 	switch (cmd) {
440 	case DDI_INFO_DEVT2DEVINFO:
441 		*result = (void *)viona_dip;
442 		error = DDI_SUCCESS;
443 		break;
444 	case DDI_INFO_DEVT2INSTANCE:
445 		*result = (void *)0;
446 		error = DDI_SUCCESS;
447 		break;
448 	default:
449 		error = DDI_FAILURE;
450 		break;
451 	}
452 	return (error);
453 }
454 
455 static int
viona_attach(dev_info_t * dip,ddi_attach_cmd_t cmd)456 viona_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
457 {
458 	if (cmd != DDI_ATTACH) {
459 		return (DDI_FAILURE);
460 	}
461 
462 	if (ddi_create_minor_node(dip, "viona", S_IFCHR, VIONA_CTL_MINOR,
463 	    DDI_PSEUDO, 0) != DDI_SUCCESS) {
464 		return (DDI_FAILURE);
465 	}
466 
467 	viona_neti_attach();
468 
469 	viona_dip = dip;
470 	ddi_report_dev(viona_dip);
471 
472 	return (DDI_SUCCESS);
473 }
474 
475 static int
viona_detach(dev_info_t * dip,ddi_detach_cmd_t cmd)476 viona_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
477 {
478 	dev_info_t *old_dip = viona_dip;
479 
480 	if (cmd != DDI_DETACH) {
481 		return (DDI_FAILURE);
482 	}
483 
484 	VERIFY(old_dip != NULL);
485 
486 	viona_neti_detach();
487 	viona_dip = NULL;
488 	ddi_remove_minor_node(old_dip, NULL);
489 
490 	return (DDI_SUCCESS);
491 }
492 
493 static int
viona_open(dev_t * devp,int flag,int otype,cred_t * credp)494 viona_open(dev_t *devp, int flag, int otype, cred_t *credp)
495 {
496 	int	minor;
497 	viona_soft_state_t *ss;
498 
499 	if (otype != OTYP_CHR) {
500 		return (EINVAL);
501 	}
502 #if 0
503 	/*
504 	 * XXX-mg: drv_priv() is wrong, but I'm not sure what is right.
505 	 * Should the check be at open() or ioctl()?
506 	 */
507 	if (drv_priv(credp) != 0) {
508 		return (EPERM);
509 	}
510 #endif
511 	if (getminor(*devp) != VIONA_CTL_MINOR) {
512 		return (ENXIO);
513 	}
514 
515 	minor = id_alloc_nosleep(viona_minors);
516 	if (minor == -1) {
517 		/* All minors are busy */
518 		return (EBUSY);
519 	}
520 	if (ddi_soft_state_zalloc(viona_state, minor) != DDI_SUCCESS) {
521 		id_free(viona_minors, minor);
522 		return (ENOMEM);
523 	}
524 
525 	ss = ddi_get_soft_state(viona_state, minor);
526 	mutex_init(&ss->ss_lock, NULL, MUTEX_DEFAULT, NULL);
527 	ss->ss_minor = minor;
528 	*devp = makedevice(getmajor(*devp), minor);
529 
530 	return (0);
531 }
532 
533 static int
viona_close(dev_t dev,int flag,int otype,cred_t * credp)534 viona_close(dev_t dev, int flag, int otype, cred_t *credp)
535 {
536 	int			minor;
537 	viona_soft_state_t	*ss;
538 
539 	if (otype != OTYP_CHR) {
540 		return (EINVAL);
541 	}
542 
543 	minor = getminor(dev);
544 
545 	ss = ddi_get_soft_state(viona_state, minor);
546 	if (ss == NULL) {
547 		return (ENXIO);
548 	}
549 
550 	VERIFY0(viona_ioc_delete(ss, B_TRUE));
551 	VERIFY(!list_link_active(&ss->ss_node));
552 	ddi_soft_state_free(viona_state, minor);
553 	id_free(viona_minors, minor);
554 
555 	return (0);
556 }
557 
558 static int
viona_ioctl(dev_t dev,int cmd,intptr_t data,int md,cred_t * cr,int * rv)559 viona_ioctl(dev_t dev, int cmd, intptr_t data, int md, cred_t *cr, int *rv)
560 {
561 	viona_soft_state_t *ss;
562 	void *dptr = (void *)data;
563 	int err = 0, val;
564 	viona_link_t *link;
565 
566 	ss = ddi_get_soft_state(viona_state, getminor(dev));
567 	if (ss == NULL) {
568 		return (ENXIO);
569 	}
570 
571 	switch (cmd) {
572 	case VNA_IOC_CREATE:
573 		return (viona_ioc_create(ss, dptr, md, cr));
574 	case VNA_IOC_DELETE:
575 		return (viona_ioc_delete(ss, B_FALSE));
576 	case VNA_IOC_VERSION:
577 		*rv = VIONA_CURRENT_INTERFACE_VERSION;
578 		return (0);
579 	case VNA_IOC_DEFAULT_PARAMS:
580 		/*
581 		 * With a NULL link parameter, viona_ioc_get_params() will emit
582 		 * the default parameters with the same error-handling behavior
583 		 * as VNA_IOC_GET_PARAMS.
584 		 */
585 		return (viona_ioc_get_params(NULL, dptr, md));
586 	default:
587 		break;
588 	}
589 
590 	mutex_enter(&ss->ss_lock);
591 	if ((link = ss->ss_link) == NULL || link->l_destroyed ||
592 	    vmm_drv_release_reqd(link->l_vm_hold)) {
593 		mutex_exit(&ss->ss_lock);
594 		return (ENXIO);
595 	}
596 
597 	switch (cmd) {
598 	case VNA_IOC_GET_FEATURES:
599 		val = VIONA_S_HOSTCAPS | link->l_features_hw;
600 		if (ddi_copyout(&val, dptr, sizeof (val), md) != 0) {
601 			err = EFAULT;
602 		}
603 		break;
604 	case VNA_IOC_SET_FEATURES:
605 		if (ddi_copyin(dptr, &val, sizeof (val), md) != 0) {
606 			err = EFAULT;
607 			break;
608 		}
609 		val &= (VIONA_S_HOSTCAPS | link->l_features_hw);
610 
611 		if ((val & VIRTIO_NET_F_CSUM) == 0)
612 			val &= ~VIRTIO_NET_F_HOST_TSO4;
613 
614 		if ((val & VIRTIO_NET_F_GUEST_CSUM) == 0)
615 			val &= ~VIRTIO_NET_F_GUEST_TSO4;
616 
617 		link->l_features = val;
618 		break;
619 	case VNA_IOC_RING_INIT:
620 		err = viona_ioc_ring_init(link, dptr, md);
621 		break;
622 	case VNA_IOC_RING_RESET:
623 		err = viona_ioc_ring_reset(link, (uint_t)data);
624 		break;
625 	case VNA_IOC_RING_KICK:
626 		err = viona_ioc_ring_kick(link, (uint_t)data);
627 		break;
628 	case VNA_IOC_RING_SET_MSI:
629 		err = viona_ioc_ring_set_msi(link, dptr, md);
630 		break;
631 	case VNA_IOC_RING_INTR_CLR:
632 		err = viona_ioc_ring_intr_clear(link, (uint_t)data);
633 		break;
634 	case VNA_IOC_RING_SET_STATE:
635 		err = viona_ioc_ring_set_state(link, dptr, md);
636 		break;
637 	case VNA_IOC_RING_GET_STATE:
638 		err = viona_ioc_ring_get_state(link, dptr, md);
639 		break;
640 	case VNA_IOC_RING_PAUSE:
641 		err = viona_ioc_ring_pause(link, (uint_t)data);
642 		break;
643 
644 	case VNA_IOC_INTR_POLL:
645 		err = viona_ioc_intr_poll(link, dptr, md, rv);
646 		break;
647 	case VNA_IOC_SET_NOTIFY_IOP:
648 		if (data < 0 || data > UINT16_MAX) {
649 			err = EINVAL;
650 			break;
651 		}
652 		err = viona_ioc_set_notify_ioport(link, (uint16_t)data);
653 		break;
654 	case VNA_IOC_SET_PROMISC:
655 		err = viona_ioc_set_promisc(link, (viona_promisc_t)data);
656 		break;
657 	case VNA_IOC_GET_PARAMS:
658 		err = viona_ioc_get_params(link, dptr, md);
659 		break;
660 	case VNA_IOC_SET_PARAMS:
661 		err = viona_ioc_set_params(link, dptr, md);
662 		break;
663 	default:
664 		err = ENOTTY;
665 		break;
666 	}
667 
668 	mutex_exit(&ss->ss_lock);
669 	return (err);
670 }
671 
672 static int
viona_chpoll(dev_t dev,short events,int anyyet,short * reventsp,struct pollhead ** phpp)673 viona_chpoll(dev_t dev, short events, int anyyet, short *reventsp,
674     struct pollhead **phpp)
675 {
676 	viona_soft_state_t *ss;
677 	viona_link_t *link;
678 
679 	ss = ddi_get_soft_state(viona_state, getminor(dev));
680 	if (ss == NULL) {
681 		return (ENXIO);
682 	}
683 
684 	mutex_enter(&ss->ss_lock);
685 	if ((link = ss->ss_link) == NULL || link->l_destroyed) {
686 		mutex_exit(&ss->ss_lock);
687 		return (ENXIO);
688 	}
689 
690 	*reventsp = 0;
691 	if ((events & POLLRDBAND) != 0) {
692 		for (uint_t i = 0; i < VIONA_VQ_MAX; i++) {
693 			if (link->l_vrings[i].vr_intr_enabled != 0) {
694 				*reventsp |= POLLRDBAND;
695 				break;
696 			}
697 		}
698 	}
699 	if ((*reventsp == 0 && !anyyet) || (events & POLLET)) {
700 		*phpp = &link->l_pollhead;
701 	}
702 	mutex_exit(&ss->ss_lock);
703 
704 	return (0);
705 }
706 
707 static void
viona_get_mac_capab(viona_link_t * link)708 viona_get_mac_capab(viona_link_t *link)
709 {
710 	mac_handle_t mh = link->l_mh;
711 	uint32_t cap = 0;
712 	mac_capab_lso_t lso_cap;
713 
714 	link->l_features_hw = 0;
715 	if (mac_capab_get(mh, MAC_CAPAB_HCKSUM, &cap)) {
716 		/*
717 		 * Only report HW checksum ability if the underlying MAC
718 		 * resource is capable of populating the L4 header.
719 		 */
720 		if ((cap & VIONA_CAP_HCKSUM_INTEREST) != 0) {
721 			link->l_features_hw |= VIRTIO_NET_F_CSUM;
722 		}
723 		link->l_cap_csum = cap;
724 	}
725 
726 	if ((link->l_features_hw & VIRTIO_NET_F_CSUM) &&
727 	    mac_capab_get(mh, MAC_CAPAB_LSO, &lso_cap)) {
728 		/*
729 		 * Virtio doesn't allow for negotiating a maximum LSO
730 		 * packet size. We have to assume that the guest may
731 		 * send a maximum length IP packet. Make sure the
732 		 * underlying MAC can handle an LSO of this size.
733 		 */
734 		if ((lso_cap.lso_flags & LSO_TX_BASIC_TCP_IPV4) &&
735 		    lso_cap.lso_basic_tcp_ipv4.lso_max >= IP_MAXPACKET)
736 			link->l_features_hw |= VIRTIO_NET_F_HOST_TSO4;
737 	}
738 }
739 
740 static int
viona_kstat_update(kstat_t * ksp,int rw)741 viona_kstat_update(kstat_t *ksp, int rw)
742 {
743 	viona_link_t *link = ksp->ks_private;
744 	viona_kstats_t *vk = ksp->ks_data;
745 
746 	/*
747 	 * Avoid the potential for mangled values due to a racing consolidation
748 	 * of stats for a ring by performing the kstat update with l_stats_lock
749 	 * held while adding up the central (link) and ring values.
750 	 */
751 	mutex_enter(&link->l_stats_lock);
752 
753 	const viona_transfer_stats_t *ring_stats =
754 	    &link->l_vrings[VIONA_VQ_RX].vr_stats;
755 	const viona_transfer_stats_t *link_stats = &link->l_stats.vls_rx;
756 
757 	vk->vk_rx_packets.value.ui64 =
758 	    link_stats->vts_packets + ring_stats->vts_packets;
759 	vk->vk_rx_bytes.value.ui64 =
760 	    link_stats->vts_bytes + ring_stats->vts_bytes;
761 	vk->vk_rx_errors.value.ui64 =
762 	    link_stats->vts_errors + ring_stats->vts_errors;
763 	vk->vk_rx_drops.value.ui64 =
764 	    link_stats->vts_drops + ring_stats->vts_drops;
765 
766 	ring_stats = &link->l_vrings[VIONA_VQ_TX].vr_stats;
767 	link_stats = &link->l_stats.vls_tx;
768 
769 	vk->vk_tx_packets.value.ui64 =
770 	    link_stats->vts_packets + ring_stats->vts_packets;
771 	vk->vk_tx_bytes.value.ui64 =
772 	    link_stats->vts_bytes + ring_stats->vts_bytes;
773 	vk->vk_tx_errors.value.ui64 =
774 	    link_stats->vts_errors + ring_stats->vts_errors;
775 	vk->vk_tx_drops.value.ui64 =
776 	    link_stats->vts_drops + ring_stats->vts_drops;
777 
778 	mutex_exit(&link->l_stats_lock);
779 
780 	return (0);
781 }
782 
783 static int
viona_kstat_init(viona_soft_state_t * ss,const cred_t * cr)784 viona_kstat_init(viona_soft_state_t *ss, const cred_t *cr)
785 {
786 	zoneid_t zid = crgetzoneid(cr);
787 	kstat_t *ksp;
788 
789 	ASSERT(MUTEX_HELD(&ss->ss_lock));
790 	ASSERT3P(ss->ss_kstat, ==, NULL);
791 
792 	ksp = kstat_create_zone(VIONA_MODULE_NAME, ss->ss_minor,
793 	    VIONA_KSTAT_NAME, VIONA_KSTAT_CLASS, KSTAT_TYPE_NAMED,
794 	    sizeof (viona_kstats_t) / sizeof (kstat_named_t), 0, zid);
795 
796 	if (ksp == NULL) {
797 		/*
798 		 * Without detail from kstat_create_zone(), assume that resource
799 		 * exhaustion is to blame for the failure.
800 		 */
801 		return (ENOMEM);
802 	}
803 	ss->ss_kstat = ksp;
804 
805 	/*
806 	 * If this instance is associated with a non-global zone, make its
807 	 * kstats visible from the GZ.
808 	 */
809 	if (zid != GLOBAL_ZONEID) {
810 		kstat_zone_add(ss->ss_kstat, GLOBAL_ZONEID);
811 	}
812 
813 	viona_kstats_t *vk = ksp->ks_data;
814 
815 	kstat_named_init(&vk->vk_rx_packets, "rx_packets", KSTAT_DATA_UINT64);
816 	kstat_named_init(&vk->vk_rx_bytes, "rx_bytes", KSTAT_DATA_UINT64);
817 	kstat_named_init(&vk->vk_rx_errors, "rx_errors", KSTAT_DATA_UINT64);
818 	kstat_named_init(&vk->vk_rx_drops, "rx_drops", KSTAT_DATA_UINT64);
819 	kstat_named_init(&vk->vk_tx_packets, "tx_packets", KSTAT_DATA_UINT64);
820 	kstat_named_init(&vk->vk_tx_bytes, "tx_bytes", KSTAT_DATA_UINT64);
821 	kstat_named_init(&vk->vk_tx_errors, "tx_errors", KSTAT_DATA_UINT64);
822 	kstat_named_init(&vk->vk_tx_drops, "tx_drops", KSTAT_DATA_UINT64);
823 	ksp->ks_private = ss->ss_link;
824 	ksp->ks_update = viona_kstat_update;
825 
826 	kstat_install(ss->ss_kstat);
827 	return (0);
828 }
829 
830 static void
viona_kstat_fini(viona_soft_state_t * ss)831 viona_kstat_fini(viona_soft_state_t *ss)
832 {
833 	ASSERT(MUTEX_HELD(&ss->ss_lock));
834 
835 	if (ss->ss_kstat != NULL) {
836 		kstat_delete(ss->ss_kstat);
837 		ss->ss_kstat = NULL;
838 	}
839 }
840 
841 static int
viona_ioc_create(viona_soft_state_t * ss,void * dptr,int md,cred_t * cr)842 viona_ioc_create(viona_soft_state_t *ss, void *dptr, int md, cred_t *cr)
843 {
844 	vioc_create_t	kvc;
845 	viona_link_t	*link = NULL;
846 	char		cli_name[MAXNAMELEN];
847 	int		err = 0;
848 	file_t		*fp;
849 	vmm_hold_t	*hold = NULL;
850 	viona_neti_t	*nip = NULL;
851 	zoneid_t	zid;
852 	mac_diag_t	mac_diag = MAC_DIAG_NONE;
853 	boolean_t	rings_allocd = B_FALSE;
854 
855 	ASSERT(MUTEX_NOT_HELD(&ss->ss_lock));
856 
857 	if (ddi_copyin(dptr, &kvc, sizeof (kvc), md) != 0) {
858 		return (EFAULT);
859 	}
860 
861 	zid = crgetzoneid(cr);
862 	nip = viona_neti_lookup_by_zid(zid);
863 	if (nip == NULL) {
864 		return (EIO);
865 	}
866 
867 	if (!nip->vni_nethook.vnh_hooked) {
868 		viona_neti_rele(nip);
869 		return (EIO);
870 	}
871 
872 	mutex_enter(&ss->ss_lock);
873 	if (ss->ss_link != NULL) {
874 		mutex_exit(&ss->ss_lock);
875 		viona_neti_rele(nip);
876 		return (EEXIST);
877 	}
878 
879 	if ((fp = getf(kvc.c_vmfd)) == NULL) {
880 		err = EBADF;
881 		goto bail;
882 	}
883 	err = vmm_drv_hold(fp, cr, &hold);
884 	releasef(kvc.c_vmfd);
885 	if (err != 0) {
886 		goto bail;
887 	}
888 
889 	link = kmem_zalloc(sizeof (viona_link_t), KM_SLEEP);
890 	link->l_linkid = kvc.c_linkid;
891 	link->l_vm_hold = hold;
892 
893 	err = mac_open_by_linkid(link->l_linkid, &link->l_mh);
894 	if (err != 0) {
895 		goto bail;
896 	}
897 
898 	viona_get_mac_capab(link);
899 	viona_params_get_defaults(&link->l_params);
900 
901 	(void) snprintf(cli_name, sizeof (cli_name), "%s-%d", VIONA_MODULE_NAME,
902 	    link->l_linkid);
903 	err = mac_client_open(link->l_mh, &link->l_mch, cli_name, 0);
904 	if (err != 0) {
905 		goto bail;
906 	}
907 
908 	err = mac_unicast_add(link->l_mch, NULL, MAC_UNICAST_PRIMARY,
909 	    &link->l_muh, VLAN_ID_NONE, &mac_diag);
910 	if (err != 0) {
911 		goto bail;
912 	}
913 
914 	viona_ring_alloc(link, &link->l_vrings[VIONA_VQ_RX]);
915 	viona_ring_alloc(link, &link->l_vrings[VIONA_VQ_TX]);
916 	rings_allocd = B_TRUE;
917 
918 	/*
919 	 * Default to passing up all multicast traffic in addition to
920 	 * classified unicast. Guests which have support will change this
921 	 * if they need to via the virtio net control queue; guests without
922 	 * support generally still want to see multicast.
923 	 */
924 	link->l_promisc = VIONA_PROMISC_MULTI;
925 	if ((err = viona_rx_set(link, link->l_promisc)) != 0) {
926 		goto bail;
927 	}
928 
929 	link->l_neti = nip;
930 	ss->ss_link = link;
931 
932 	if ((err = viona_kstat_init(ss, cr)) != 0) {
933 		goto bail;
934 	}
935 
936 	mutex_exit(&ss->ss_lock);
937 
938 	mutex_enter(&nip->vni_lock);
939 	list_insert_tail(&nip->vni_dev_list, ss);
940 	mutex_exit(&nip->vni_lock);
941 
942 	return (0);
943 
944 bail:
945 	if (link != NULL) {
946 		viona_rx_clear(link);
947 		if (link->l_mch != NULL) {
948 			if (link->l_muh != NULL) {
949 				VERIFY0(mac_unicast_remove(link->l_mch,
950 				    link->l_muh));
951 				link->l_muh = NULL;
952 			}
953 			mac_client_close(link->l_mch, 0);
954 		}
955 		if (link->l_mh != NULL) {
956 			mac_close(link->l_mh);
957 		}
958 		if (rings_allocd) {
959 			viona_ring_free(&link->l_vrings[VIONA_VQ_RX]);
960 			viona_ring_free(&link->l_vrings[VIONA_VQ_TX]);
961 		}
962 		kmem_free(link, sizeof (viona_link_t));
963 		ss->ss_link = NULL;
964 	}
965 	if (hold != NULL) {
966 		vmm_drv_rele(hold);
967 	}
968 	viona_neti_rele(nip);
969 
970 	mutex_exit(&ss->ss_lock);
971 	return (err);
972 }
973 
974 static int
viona_ioc_delete(viona_soft_state_t * ss,boolean_t on_close)975 viona_ioc_delete(viona_soft_state_t *ss, boolean_t on_close)
976 {
977 	viona_link_t *link;
978 	viona_neti_t *nip = NULL;
979 
980 	mutex_enter(&ss->ss_lock);
981 	if ((link = ss->ss_link) == NULL) {
982 		/* Link destruction already complete */
983 		mutex_exit(&ss->ss_lock);
984 		return (0);
985 	}
986 
987 	if (link->l_destroyed) {
988 		/*
989 		 * Link destruction has been started by another thread, but has
990 		 * not completed.  This condition should be impossible to
991 		 * encounter when performing the on-close destroy of the link,
992 		 * since racing ioctl accessors must necessarily be absent.
993 		 */
994 		VERIFY(!on_close);
995 		mutex_exit(&ss->ss_lock);
996 		return (EAGAIN);
997 	}
998 	/*
999 	 * The link deletion cannot fail after this point, continuing until its
1000 	 * successful completion is reached.
1001 	 */
1002 	link->l_destroyed = B_TRUE;
1003 
1004 	/*
1005 	 * Tear down the IO port hook so it cannot be used to kick any of the
1006 	 * rings which are about to be reset and stopped.
1007 	 */
1008 	VERIFY0(viona_ioc_set_notify_ioport(link, 0));
1009 	mutex_exit(&ss->ss_lock);
1010 
1011 	/*
1012 	 * Return the rings to their reset state, ignoring any possible
1013 	 * interruptions from signals.
1014 	 */
1015 	VERIFY0(viona_ring_reset(&link->l_vrings[VIONA_VQ_RX], B_FALSE));
1016 	VERIFY0(viona_ring_reset(&link->l_vrings[VIONA_VQ_TX], B_FALSE));
1017 
1018 	mutex_enter(&ss->ss_lock);
1019 	viona_kstat_fini(ss);
1020 	if (link->l_mch != NULL) {
1021 		/* Unhook the receive callbacks and close out the client */
1022 		viona_rx_clear(link);
1023 		if (link->l_muh != NULL) {
1024 			VERIFY0(mac_unicast_remove(link->l_mch, link->l_muh));
1025 			link->l_muh = NULL;
1026 		}
1027 		mac_client_close(link->l_mch, 0);
1028 	}
1029 	if (link->l_mh != NULL) {
1030 		mac_close(link->l_mh);
1031 	}
1032 	if (link->l_vm_hold != NULL) {
1033 		vmm_drv_rele(link->l_vm_hold);
1034 		link->l_vm_hold = NULL;
1035 	}
1036 
1037 	nip = link->l_neti;
1038 	link->l_neti = NULL;
1039 
1040 	viona_ring_free(&link->l_vrings[VIONA_VQ_RX]);
1041 	viona_ring_free(&link->l_vrings[VIONA_VQ_TX]);
1042 	pollhead_clean(&link->l_pollhead);
1043 	ss->ss_link = NULL;
1044 	mutex_exit(&ss->ss_lock);
1045 
1046 	mutex_enter(&nip->vni_lock);
1047 	list_remove(&nip->vni_dev_list, ss);
1048 	mutex_exit(&nip->vni_lock);
1049 
1050 	viona_neti_rele(nip);
1051 
1052 	kmem_free(link, sizeof (viona_link_t));
1053 	return (0);
1054 }
1055 
1056 static int
viona_ioc_ring_init(viona_link_t * link,void * udata,int md)1057 viona_ioc_ring_init(viona_link_t *link, void *udata, int md)
1058 {
1059 	vioc_ring_init_t kri;
1060 	int err;
1061 
1062 	if (ddi_copyin(udata, &kri, sizeof (kri), md) != 0) {
1063 		return (EFAULT);
1064 	}
1065 	const struct viona_ring_params params = {
1066 		.vrp_pa = kri.ri_qaddr,
1067 		.vrp_size = kri.ri_qsize,
1068 		.vrp_avail_idx = 0,
1069 		.vrp_used_idx = 0,
1070 	};
1071 
1072 	err = viona_ring_init(link, kri.ri_index, &params);
1073 
1074 	return (err);
1075 }
1076 
1077 static int
viona_ioc_ring_set_state(viona_link_t * link,void * udata,int md)1078 viona_ioc_ring_set_state(viona_link_t *link, void *udata, int md)
1079 {
1080 	vioc_ring_state_t krs;
1081 	int err;
1082 
1083 	if (ddi_copyin(udata, &krs, sizeof (krs), md) != 0) {
1084 		return (EFAULT);
1085 	}
1086 	const struct viona_ring_params params = {
1087 		.vrp_pa = krs.vrs_qaddr,
1088 		.vrp_size = krs.vrs_qsize,
1089 		.vrp_avail_idx = krs.vrs_avail_idx,
1090 		.vrp_used_idx = krs.vrs_used_idx,
1091 	};
1092 
1093 	err = viona_ring_init(link, krs.vrs_index, &params);
1094 
1095 	return (err);
1096 }
1097 
1098 static int
viona_ioc_ring_get_state(viona_link_t * link,void * udata,int md)1099 viona_ioc_ring_get_state(viona_link_t *link, void *udata, int md)
1100 {
1101 	vioc_ring_state_t krs;
1102 
1103 	if (ddi_copyin(udata, &krs, sizeof (krs), md) != 0) {
1104 		return (EFAULT);
1105 	}
1106 
1107 	struct viona_ring_params params;
1108 	int err = viona_ring_get_state(link, krs.vrs_index, &params);
1109 	if (err != 0) {
1110 		return (err);
1111 	}
1112 	krs.vrs_qsize = params.vrp_size;
1113 	krs.vrs_qaddr = params.vrp_pa;
1114 	krs.vrs_avail_idx = params.vrp_avail_idx;
1115 	krs.vrs_used_idx = params.vrp_used_idx;
1116 
1117 	if (ddi_copyout(&krs, udata, sizeof (krs), md) != 0) {
1118 		return (EFAULT);
1119 	}
1120 	return (0);
1121 }
1122 
1123 static int
viona_ioc_ring_reset(viona_link_t * link,uint_t idx)1124 viona_ioc_ring_reset(viona_link_t *link, uint_t idx)
1125 {
1126 	viona_vring_t *ring;
1127 
1128 	if (idx >= VIONA_VQ_MAX) {
1129 		return (EINVAL);
1130 	}
1131 	ring = &link->l_vrings[idx];
1132 
1133 	return (viona_ring_reset(ring, B_TRUE));
1134 }
1135 
1136 static int
viona_ioc_ring_kick(viona_link_t * link,uint_t idx)1137 viona_ioc_ring_kick(viona_link_t *link, uint_t idx)
1138 {
1139 	viona_vring_t *ring;
1140 	int err;
1141 
1142 	if (idx >= VIONA_VQ_MAX) {
1143 		return (EINVAL);
1144 	}
1145 	ring = &link->l_vrings[idx];
1146 
1147 	mutex_enter(&ring->vr_lock);
1148 	switch (ring->vr_state) {
1149 	case VRS_SETUP:
1150 		/*
1151 		 * An early kick to a ring which is starting its worker thread
1152 		 * is fine.  Once that thread is active, it will process the
1153 		 * start-up request immediately.
1154 		 */
1155 		/* FALLTHROUGH */
1156 	case VRS_INIT:
1157 		ring->vr_state_flags |= VRSF_REQ_START;
1158 		/* FALLTHROUGH */
1159 	case VRS_RUN:
1160 		cv_broadcast(&ring->vr_cv);
1161 		err = 0;
1162 		break;
1163 	default:
1164 		err = EBUSY;
1165 		break;
1166 	}
1167 	mutex_exit(&ring->vr_lock);
1168 
1169 	return (err);
1170 }
1171 
1172 static int
viona_ioc_ring_pause(viona_link_t * link,uint_t idx)1173 viona_ioc_ring_pause(viona_link_t *link, uint_t idx)
1174 {
1175 	if (idx >= VIONA_VQ_MAX) {
1176 		return (EINVAL);
1177 	}
1178 
1179 	viona_vring_t *ring = &link->l_vrings[idx];
1180 	return (viona_ring_pause(ring));
1181 }
1182 
1183 static int
viona_ioc_ring_set_msi(viona_link_t * link,void * data,int md)1184 viona_ioc_ring_set_msi(viona_link_t *link, void *data, int md)
1185 {
1186 	vioc_ring_msi_t vrm;
1187 	viona_vring_t *ring;
1188 
1189 	if (ddi_copyin(data, &vrm, sizeof (vrm), md) != 0) {
1190 		return (EFAULT);
1191 	}
1192 	if (vrm.rm_index >= VIONA_VQ_MAX) {
1193 		return (EINVAL);
1194 	}
1195 
1196 	ring = &link->l_vrings[vrm.rm_index];
1197 	mutex_enter(&ring->vr_lock);
1198 	ring->vr_msi_addr = vrm.rm_addr;
1199 	ring->vr_msi_msg = vrm.rm_msg;
1200 	mutex_exit(&ring->vr_lock);
1201 
1202 	return (0);
1203 }
1204 
1205 static int
viona_notify_iop(void * arg,bool in,uint16_t port,uint8_t bytes,uint32_t * val)1206 viona_notify_iop(void *arg, bool in, uint16_t port, uint8_t bytes,
1207     uint32_t *val)
1208 {
1209 	viona_link_t *link = (viona_link_t *)arg;
1210 
1211 	/*
1212 	 * If the request is a read (in/ins), or direct at a port other than
1213 	 * what we expect to be registered on, ignore it.
1214 	 */
1215 	if (in || port != link->l_notify_ioport) {
1216 		return (ESRCH);
1217 	}
1218 
1219 	/* Let userspace handle notifications for rings other than RX/TX. */
1220 	const uint16_t vq = *val;
1221 	if (vq >= VIONA_VQ_MAX) {
1222 		return (ESRCH);
1223 	}
1224 
1225 	viona_vring_t *ring = &link->l_vrings[vq];
1226 	int res = 0;
1227 
1228 	mutex_enter(&ring->vr_lock);
1229 	if (ring->vr_state == VRS_RUN) {
1230 		cv_broadcast(&ring->vr_cv);
1231 	} else {
1232 		res = ESRCH;
1233 	}
1234 	mutex_exit(&ring->vr_lock);
1235 
1236 	return (res);
1237 }
1238 
1239 static int
viona_ioc_set_notify_ioport(viona_link_t * link,uint16_t ioport)1240 viona_ioc_set_notify_ioport(viona_link_t *link, uint16_t ioport)
1241 {
1242 	int err = 0;
1243 
1244 	if (link->l_notify_ioport != 0) {
1245 		vmm_drv_ioport_unhook(link->l_vm_hold, &link->l_notify_cookie);
1246 		link->l_notify_ioport = 0;
1247 	}
1248 
1249 	if (ioport != 0) {
1250 		err = vmm_drv_ioport_hook(link->l_vm_hold, ioport,
1251 		    viona_notify_iop, (void *)link, &link->l_notify_cookie);
1252 		if (err == 0) {
1253 			link->l_notify_ioport = ioport;
1254 		}
1255 	}
1256 	return (err);
1257 }
1258 
1259 static int
viona_ioc_set_promisc(viona_link_t * link,viona_promisc_t mode)1260 viona_ioc_set_promisc(viona_link_t *link, viona_promisc_t mode)
1261 {
1262 	int err;
1263 
1264 	if (mode >= VIONA_PROMISC_MAX) {
1265 		return (EINVAL);
1266 	}
1267 
1268 	if (mode == link->l_promisc) {
1269 		return (0);
1270 	}
1271 
1272 	if ((err = viona_rx_set(link, mode)) != 0) {
1273 		return (err);
1274 	}
1275 
1276 	link->l_promisc = mode;
1277 	return (0);
1278 }
1279 
1280 #define	PARAM_NM_TX_COPY_DATA	"tx_copy_data"
1281 #define	PARAM_NM_TX_HEADER_PAD	"tx_header_pad"
1282 
1283 #define	PARAM_ERR_INVALID_TYPE	"invalid type"
1284 #define	PARAM_ERR_OUT_OF_RANGE	"value out of range"
1285 #define	PARAM_ERR_UNK_KEY	"unknown key"
1286 
1287 static nvlist_t *
viona_params_to_nvlist(const viona_link_params_t * vlp)1288 viona_params_to_nvlist(const viona_link_params_t *vlp)
1289 {
1290 	nvlist_t *nvl = fnvlist_alloc();
1291 
1292 	fnvlist_add_boolean_value(nvl, PARAM_NM_TX_COPY_DATA,
1293 	    vlp->vlp_tx_copy_data);
1294 	fnvlist_add_uint16(nvl, PARAM_NM_TX_HEADER_PAD,
1295 	    vlp->vlp_tx_header_pad);
1296 
1297 	return (nvl);
1298 }
1299 
1300 static nvlist_t *
viona_params_from_nvlist(nvlist_t * nvl,viona_link_params_t * vlp)1301 viona_params_from_nvlist(nvlist_t *nvl, viona_link_params_t *vlp)
1302 {
1303 	nvlist_t *nverr = fnvlist_alloc();
1304 	nvpair_t *nvp = NULL;
1305 
1306 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
1307 		const char *name = nvpair_name(nvp);
1308 		const data_type_t dtype = nvpair_type(nvp);
1309 
1310 		if (strcmp(name, PARAM_NM_TX_COPY_DATA) == 0) {
1311 			if (dtype == DATA_TYPE_BOOLEAN_VALUE) {
1312 				vlp->vlp_tx_copy_data =
1313 				    fnvpair_value_boolean_value(nvp);
1314 			} else {
1315 				fnvlist_add_string(nverr, name,
1316 				    PARAM_ERR_INVALID_TYPE);
1317 			}
1318 			continue;
1319 		}
1320 		if (strcmp(name, PARAM_NM_TX_HEADER_PAD) == 0) {
1321 			if (dtype == DATA_TYPE_UINT16) {
1322 				uint16_t value = fnvpair_value_uint16(nvp);
1323 
1324 				if (value > viona_max_header_pad) {
1325 					fnvlist_add_string(nverr, name,
1326 					    PARAM_ERR_OUT_OF_RANGE);
1327 				} else {
1328 					vlp->vlp_tx_header_pad = value;
1329 				}
1330 			} else {
1331 				fnvlist_add_string(nverr, name,
1332 				    PARAM_ERR_INVALID_TYPE);
1333 			}
1334 			continue;
1335 		}
1336 
1337 		/* Reject parameters we do not recognize */
1338 		fnvlist_add_string(nverr, name, PARAM_ERR_UNK_KEY);
1339 	}
1340 
1341 	if (!nvlist_empty(nverr)) {
1342 		return (nverr);
1343 	}
1344 
1345 	nvlist_free(nverr);
1346 	return (NULL);
1347 }
1348 
1349 static void
viona_params_get_defaults(viona_link_params_t * vlp)1350 viona_params_get_defaults(viona_link_params_t *vlp)
1351 {
1352 	vlp->vlp_tx_copy_data = viona_tx_copy_needed();
1353 	vlp->vlp_tx_header_pad = 0;
1354 }
1355 
1356 static int
viona_ioc_get_params(viona_link_t * link,void * udata,int md)1357 viona_ioc_get_params(viona_link_t *link, void *udata, int md)
1358 {
1359 	vioc_get_params_t vgp;
1360 	int err = 0;
1361 
1362 	if (ddi_copyin(udata, &vgp, sizeof (vgp), md) != 0) {
1363 		return (EFAULT);
1364 	}
1365 
1366 	nvlist_t *nvl = NULL;
1367 	if (link != NULL) {
1368 		nvl = viona_params_to_nvlist(&link->l_params);
1369 	} else {
1370 		viona_link_params_t vlp = { 0 };
1371 
1372 		viona_params_get_defaults(&vlp);
1373 		nvl = viona_params_to_nvlist(&vlp);
1374 	}
1375 
1376 	VERIFY(nvl != NULL);
1377 
1378 	size_t packed_sz;
1379 	void *packed = fnvlist_pack(nvl, &packed_sz);
1380 	nvlist_free(nvl);
1381 
1382 	if (packed_sz > vgp.vgp_param_sz) {
1383 		err = E2BIG;
1384 	}
1385 	/* Communicate size, even if the data will not fit */
1386 	vgp.vgp_param_sz = packed_sz;
1387 
1388 	if (err == 0 &&
1389 	    ddi_copyout(packed, vgp.vgp_param, packed_sz, md) != 0) {
1390 		err = EFAULT;
1391 	}
1392 	kmem_free(packed, packed_sz);
1393 
1394 	if (ddi_copyout(&vgp, udata, sizeof (vgp), md) != 0) {
1395 		if (err != 0) {
1396 			err = EFAULT;
1397 		}
1398 	}
1399 
1400 	return (err);
1401 }
1402 
1403 static int
viona_ioc_set_params(viona_link_t * link,void * udata,int md)1404 viona_ioc_set_params(viona_link_t *link, void *udata, int md)
1405 {
1406 	vioc_set_params_t vsp;
1407 	int err = 0;
1408 	nvlist_t *nverr = NULL;
1409 
1410 	if (ddi_copyin(udata, &vsp, sizeof (vsp), md) != 0) {
1411 		return (EFAULT);
1412 	}
1413 
1414 	if (vsp.vsp_param_sz > VIONA_MAX_PARAM_NVLIST_SZ) {
1415 		err = E2BIG;
1416 		goto done;
1417 	} else if (vsp.vsp_param_sz == 0) {
1418 		/*
1419 		 * There is no reason to make this ioctl call with no actual
1420 		 * parameters to be changed.
1421 		 */
1422 		err = EINVAL;
1423 		goto done;
1424 	}
1425 
1426 	const size_t packed_sz = vsp.vsp_param_sz;
1427 	void *packed = kmem_alloc(packed_sz, KM_SLEEP);
1428 	if (ddi_copyin(vsp.vsp_param, packed, packed_sz, md) != 0) {
1429 		kmem_free(packed, packed_sz);
1430 		err = EFAULT;
1431 		goto done;
1432 	}
1433 
1434 	nvlist_t *parsed = NULL;
1435 	if (nvlist_unpack(packed, packed_sz, &parsed, KM_SLEEP) == 0) {
1436 		/* Use the existing parameters as a starting point */
1437 		viona_link_params_t new_params;
1438 		bcopy(&link->l_params, &new_params,
1439 		    sizeof (new_params));
1440 
1441 		nverr = viona_params_from_nvlist(parsed, &new_params);
1442 		if (nverr == NULL) {
1443 			/*
1444 			 * Only apply the updated parameters if there
1445 			 * were no errors during parsing.
1446 			 */
1447 			bcopy(&new_params, &link->l_params,
1448 			    sizeof (new_params));
1449 		} else {
1450 			err = EINVAL;
1451 		}
1452 
1453 	} else {
1454 		err = EINVAL;
1455 	}
1456 	nvlist_free(parsed);
1457 	kmem_free(packed, packed_sz);
1458 
1459 done:
1460 	if (nverr != NULL) {
1461 		size_t err_packed_sz;
1462 		void *err_packed = fnvlist_pack(nverr, &err_packed_sz);
1463 
1464 		if (err_packed_sz > vsp.vsp_error_sz) {
1465 			if (err != 0) {
1466 				err = E2BIG;
1467 			}
1468 		} else if (ddi_copyout(err_packed, vsp.vsp_error,
1469 		    err_packed_sz, md) != 0 && err == 0) {
1470 			err = EFAULT;
1471 		}
1472 		vsp.vsp_error_sz = err_packed_sz;
1473 
1474 		nvlist_free(nverr);
1475 		kmem_free(err_packed, err_packed_sz);
1476 	} else {
1477 		/*
1478 		 * If there are no detailed per-field errors, it is important to
1479 		 * communicate that absense to userspace.
1480 		 */
1481 		vsp.vsp_error_sz = 0;
1482 	}
1483 
1484 	if (ddi_copyout(&vsp, udata, sizeof (vsp), md) != 0 && err == 0) {
1485 		err = EFAULT;
1486 	}
1487 
1488 	return (err);
1489 }
1490 
1491 static int
viona_ioc_ring_intr_clear(viona_link_t * link,uint_t idx)1492 viona_ioc_ring_intr_clear(viona_link_t *link, uint_t idx)
1493 {
1494 	if (idx >= VIONA_VQ_MAX) {
1495 		return (EINVAL);
1496 	}
1497 
1498 	link->l_vrings[idx].vr_intr_enabled = 0;
1499 	return (0);
1500 }
1501 
1502 static int
viona_ioc_intr_poll(viona_link_t * link,void * udata,int md,int * rv)1503 viona_ioc_intr_poll(viona_link_t *link, void *udata, int md, int *rv)
1504 {
1505 	uint_t cnt = 0;
1506 	vioc_intr_poll_t vip;
1507 
1508 	for (uint_t i = 0; i < VIONA_VQ_MAX; i++) {
1509 		uint_t val = link->l_vrings[i].vr_intr_enabled;
1510 
1511 		vip.vip_status[i] = val;
1512 		if (val != 0) {
1513 			cnt++;
1514 		}
1515 	}
1516 
1517 	if (ddi_copyout(&vip, udata, sizeof (vip), md) != 0) {
1518 		return (EFAULT);
1519 	}
1520 	*rv = (int)cnt;
1521 	return (0);
1522 }
1523