xref: /freebsd/sys/net/vnet.c (revision 148a8da8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2004-2009 University of Zagreb
5  * Copyright (c) 2006-2009 FreeBSD Foundation
6  * All rights reserved.
7  *
8  * This software was developed by the University of Zagreb and the
9  * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
10  * FreeBSD Foundation.
11  *
12  * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
13  * Copyright (c) 2009 Robert N. M. Watson
14  * All rights reserved.
15  *
16  * Redistribution and use in source and binary forms, with or without
17  * modification, are permitted provided that the following conditions
18  * are met:
19  * 1. Redistributions of source code must retain the above copyright
20  *    notice, this list of conditions and the following disclaimer.
21  * 2. Redistributions in binary form must reproduce the above copyright
22  *    notice, this list of conditions and the following disclaimer in the
23  *    documentation and/or other materials provided with the distribution.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_ddb.h"
42 #include "opt_kdb.h"
43 
44 #include <sys/param.h>
45 #include <sys/kdb.h>
46 #include <sys/kernel.h>
47 #include <sys/jail.h>
48 #include <sys/sdt.h>
49 #include <sys/systm.h>
50 #include <sys/sysctl.h>
51 #include <sys/eventhandler.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/socket.h>
56 #include <sys/sx.h>
57 #include <sys/sysctl.h>
58 
59 #include <machine/stdarg.h>
60 
61 #ifdef DDB
62 #include <ddb/ddb.h>
63 #include <ddb/db_sym.h>
64 #endif
65 
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/vnet.h>
69 
70 /*-
71  * This file implements core functions for virtual network stacks:
72  *
73  * - Virtual network stack management functions.
74  *
75  * - Virtual network stack memory allocator, which virtualizes global
76  *   variables in the network stack
77  *
78  * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
79  *   to register startup/shutdown events to be run for each virtual network
80  *   stack instance.
81  */
82 
83 FEATURE(vimage, "VIMAGE kernel virtualization");
84 
85 static MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
86 
87 /*
88  * The virtual network stack list has two read-write locks, one sleepable and
89  * the other not, so that the list can be stablized and walked in a variety
90  * of network stack contexts.  Both must be acquired exclusively to modify
91  * the list, but a read lock of either lock is sufficient to walk the list.
92  */
93 struct rwlock		vnet_rwlock;
94 struct sx		vnet_sxlock;
95 
96 #define	VNET_LIST_WLOCK() do {						\
97 	sx_xlock(&vnet_sxlock);						\
98 	rw_wlock(&vnet_rwlock);						\
99 } while (0)
100 
101 #define	VNET_LIST_WUNLOCK() do {					\
102 	rw_wunlock(&vnet_rwlock);					\
103 	sx_xunlock(&vnet_sxlock);					\
104 } while (0)
105 
106 struct vnet_list_head vnet_head;
107 struct vnet *vnet0;
108 
109 /*
110  * The virtual network stack allocator provides storage for virtualized
111  * global variables.  These variables are defined/declared using the
112  * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet'
113  * linker set.  The details of the implementation are somewhat subtle, but
114  * allow the majority of most network subsystems to maintain
115  * virtualization-agnostic.
116  *
117  * The virtual network stack allocator handles variables in the base kernel
118  * vs. modules in similar but different ways.  In both cases, virtualized
119  * global variables are marked as such by being declared to be part of the
120  * vnet linker set.  These "master" copies of global variables serve two
121  * functions:
122  *
123  * (1) They contain static initialization or "default" values for global
124  *     variables which will be propagated to each virtual network stack
125  *     instance when created.  As with normal global variables, they default
126  *     to zero-filled.
127  *
128  * (2) They act as unique global names by which the variable can be referred
129  *     to, regardless of network stack instance.  The single global symbol
130  *     will be used to calculate the location of a per-virtual instance
131  *     variable at run-time.
132  *
133  * Each virtual network stack instance has a complete copy of each
134  * virtualized global variable, stored in a malloc'd block of memory
135  * referred to by vnet->vnet_data_mem.  Critical to the design is that each
136  * per-instance memory block is laid out identically to the master block so
137  * that the offset of each global variable is the same across all blocks.  To
138  * optimize run-time access, a precalculated 'base' address,
139  * vnet->vnet_data_base, is stored in each vnet, and is the amount that can
140  * be added to the address of a 'master' instance of a variable to get to the
141  * per-vnet instance.
142  *
143  * Virtualized global variables are handled in a similar manner, but as each
144  * module has its own 'set_vnet' linker set, and we want to keep all
145  * virtualized globals togther, we reserve space in the kernel's linker set
146  * for potential module variables using a per-vnet character array,
147  * 'modspace'.  The virtual network stack allocator maintains a free list to
148  * track what space in the array is free (all, initially) and as modules are
149  * linked, allocates portions of the space to specific globals.  The kernel
150  * module linker queries the virtual network stack allocator and will
151  * bind references of the global to the location during linking.  It also
152  * calls into the virtual network stack allocator, once the memory is
153  * initialized, in order to propagate the new static initializations to all
154  * existing virtual network stack instances so that the soon-to-be executing
155  * module will find every network stack instance with proper default values.
156  */
157 
158 /*
159  * Number of bytes of data in the 'set_vnet' linker set, and hence the total
160  * size of all kernel virtualized global variables, and the malloc(9) type
161  * that will be used to allocate it.
162  */
163 #define	VNET_BYTES	(VNET_STOP - VNET_START)
164 
165 static MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
166 
167 /*
168  * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of
169  * global variables across all loaded modules.  As this actually sizes an
170  * array declared as a virtualized global variable in the kernel itself, and
171  * we want the virtualized global variable space to be page-sized, we may
172  * have more space than that in practice.
173  */
174 #define	VNET_MODMIN	(8 * PAGE_SIZE)
175 #define	VNET_SIZE	roundup2(VNET_BYTES, PAGE_SIZE)
176 
177 /*
178  * Space to store virtualized global variables from loadable kernel modules,
179  * and the free list to manage it.
180  */
181 VNET_DEFINE_STATIC(char, modspace[VNET_MODMIN] __aligned(__alignof(void *)));
182 
183 /*
184  * Global lists of subsystem constructor and destructors for vnets.  They are
185  * registered via VNET_SYSINIT() and VNET_SYSUNINIT().  Both lists are
186  * protected by the vnet_sysinit_sxlock global lock.
187  */
188 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
189 	TAILQ_HEAD_INITIALIZER(vnet_constructors);
190 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
191 	TAILQ_HEAD_INITIALIZER(vnet_destructors);
192 
193 struct sx		vnet_sysinit_sxlock;
194 
195 #define	VNET_SYSINIT_WLOCK()	sx_xlock(&vnet_sysinit_sxlock);
196 #define	VNET_SYSINIT_WUNLOCK()	sx_xunlock(&vnet_sysinit_sxlock);
197 #define	VNET_SYSINIT_RLOCK()	sx_slock(&vnet_sysinit_sxlock);
198 #define	VNET_SYSINIT_RUNLOCK()	sx_sunlock(&vnet_sysinit_sxlock);
199 
200 struct vnet_data_free {
201 	uintptr_t	vnd_start;
202 	int		vnd_len;
203 	TAILQ_ENTRY(vnet_data_free) vnd_link;
204 };
205 
206 static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
207     "VNET resource accounting");
208 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
209 	    TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
210 static struct sx vnet_data_free_lock;
211 
212 SDT_PROVIDER_DEFINE(vnet);
213 SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
214 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
215     "struct vnet *");
216 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
217     "int", "struct vnet *");
218 SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
219     "int", "struct vnet *");
220 SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
221     "int");
222 
223 #ifdef DDB
224 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
225 #endif
226 
227 /*
228  * Allocate a virtual network stack.
229  */
230 struct vnet *
231 vnet_alloc(void)
232 {
233 	struct vnet *vnet;
234 
235 	SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
236 	vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
237 	vnet->vnet_magic_n = VNET_MAGIC_N;
238 	vnet->vnet_state = 0;
239 	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
240 
241 	/*
242 	 * Allocate storage for virtualized global variables and copy in
243 	 * initial values form our 'master' copy.
244 	 */
245 	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
246 	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
247 
248 	/*
249 	 * All use of vnet-specific data will immediately subtract VNET_START
250 	 * from the base memory pointer, so pre-calculate that now to avoid
251 	 * it on each use.
252 	 */
253 	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
254 
255 	/* Initialize / attach vnet module instances. */
256 	CURVNET_SET_QUIET(vnet);
257 	vnet_sysinit();
258 	CURVNET_RESTORE();
259 
260 	VNET_LIST_WLOCK();
261 	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
262 	VNET_LIST_WUNLOCK();
263 
264 	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
265 	return (vnet);
266 }
267 
268 /*
269  * Destroy a virtual network stack.
270  */
271 void
272 vnet_destroy(struct vnet *vnet)
273 {
274 
275 	SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
276 	KASSERT(vnet->vnet_sockcnt == 0,
277 	    ("%s: vnet still has sockets", __func__));
278 
279 	VNET_LIST_WLOCK();
280 	LIST_REMOVE(vnet, vnet_le);
281 	VNET_LIST_WUNLOCK();
282 
283 	CURVNET_SET_QUIET(vnet);
284 	vnet_sysuninit();
285 	CURVNET_RESTORE();
286 
287 	/*
288 	 * Release storage for the virtual network stack instance.
289 	 */
290 	free(vnet->vnet_data_mem, M_VNET_DATA);
291 	vnet->vnet_data_mem = NULL;
292 	vnet->vnet_data_base = 0;
293 	vnet->vnet_magic_n = 0xdeadbeef;
294 	free(vnet, M_VNET);
295 	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
296 }
297 
298 /*
299  * Boot time initialization and allocation of virtual network stacks.
300  */
301 static void
302 vnet_init_prelink(void *arg __unused)
303 {
304 
305 	rw_init(&vnet_rwlock, "vnet_rwlock");
306 	sx_init(&vnet_sxlock, "vnet_sxlock");
307 	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
308 	LIST_INIT(&vnet_head);
309 }
310 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
311     vnet_init_prelink, NULL);
312 
313 static void
314 vnet0_init(void *arg __unused)
315 {
316 
317 	if (bootverbose)
318 		printf("VIMAGE (virtualized network stack) enabled\n");
319 
320 	/*
321 	 * We MUST clear curvnet in vi_init_done() before going SMP,
322 	 * otherwise CURVNET_SET() macros would scream about unnecessary
323 	 * curvnet recursions.
324 	 */
325 	curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
326 }
327 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
328 
329 static void
330 vnet_init_done(void *unused __unused)
331 {
332 
333 	curvnet = NULL;
334 }
335 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done,
336     NULL);
337 
338 /*
339  * Once on boot, initialize the modspace freelist to entirely cover modspace.
340  */
341 static void
342 vnet_data_startup(void *dummy __unused)
343 {
344 	struct vnet_data_free *df;
345 
346 	df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
347 	df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
348 	df->vnd_len = VNET_MODMIN;
349 	TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
350 	sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
351 }
352 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL);
353 
354 /* Dummy VNET_SYSINIT to make sure we always reach the final end state. */
355 static void
356 vnet_sysinit_done(void *unused __unused)
357 {
358 
359 	return;
360 }
361 VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY,
362     vnet_sysinit_done, NULL);
363 
364 /*
365  * When a module is loaded and requires storage for a virtualized global
366  * variable, allocate space from the modspace free list.  This interface
367  * should be used only by the kernel linker.
368  */
369 void *
370 vnet_data_alloc(int size)
371 {
372 	struct vnet_data_free *df;
373 	void *s;
374 
375 	s = NULL;
376 	size = roundup2(size, sizeof(void *));
377 	sx_xlock(&vnet_data_free_lock);
378 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
379 		if (df->vnd_len < size)
380 			continue;
381 		if (df->vnd_len == size) {
382 			s = (void *)df->vnd_start;
383 			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
384 			free(df, M_VNET_DATA_FREE);
385 			break;
386 		}
387 		s = (void *)df->vnd_start;
388 		df->vnd_len -= size;
389 		df->vnd_start = df->vnd_start + size;
390 		break;
391 	}
392 	sx_xunlock(&vnet_data_free_lock);
393 
394 	return (s);
395 }
396 
397 /*
398  * Free space for a virtualized global variable on module unload.
399  */
400 void
401 vnet_data_free(void *start_arg, int size)
402 {
403 	struct vnet_data_free *df;
404 	struct vnet_data_free *dn;
405 	uintptr_t start;
406 	uintptr_t end;
407 
408 	size = roundup2(size, sizeof(void *));
409 	start = (uintptr_t)start_arg;
410 	end = start + size;
411 	/*
412 	 * Free a region of space and merge it with as many neighbors as
413 	 * possible.  Keeping the list sorted simplifies this operation.
414 	 */
415 	sx_xlock(&vnet_data_free_lock);
416 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
417 		if (df->vnd_start > end)
418 			break;
419 		/*
420 		 * If we expand at the end of an entry we may have to merge
421 		 * it with the one following it as well.
422 		 */
423 		if (df->vnd_start + df->vnd_len == start) {
424 			df->vnd_len += size;
425 			dn = TAILQ_NEXT(df, vnd_link);
426 			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
427 				df->vnd_len += dn->vnd_len;
428 				TAILQ_REMOVE(&vnet_data_free_head, dn,
429 				    vnd_link);
430 				free(dn, M_VNET_DATA_FREE);
431 			}
432 			sx_xunlock(&vnet_data_free_lock);
433 			return;
434 		}
435 		if (df->vnd_start == end) {
436 			df->vnd_start = start;
437 			df->vnd_len += size;
438 			sx_xunlock(&vnet_data_free_lock);
439 			return;
440 		}
441 	}
442 	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
443 	dn->vnd_start = start;
444 	dn->vnd_len = size;
445 	if (df)
446 		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
447 	else
448 		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
449 	sx_xunlock(&vnet_data_free_lock);
450 }
451 
452 /*
453  * When a new virtualized global variable has been allocated, propagate its
454  * initial value to each already-allocated virtual network stack instance.
455  */
456 void
457 vnet_data_copy(void *start, int size)
458 {
459 	struct vnet *vnet;
460 
461 	VNET_LIST_RLOCK();
462 	LIST_FOREACH(vnet, &vnet_head, vnet_le)
463 		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
464 		    (uintptr_t)start), start, size);
465 	VNET_LIST_RUNLOCK();
466 }
467 
468 /*
469  * Support for special SYSINIT handlers registered via VNET_SYSINIT()
470  * and VNET_SYSUNINIT().
471  */
472 void
473 vnet_register_sysinit(void *arg)
474 {
475 	struct vnet_sysinit *vs, *vs2;
476 	struct vnet *vnet;
477 
478 	vs = arg;
479 	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
480 
481 	/* Add the constructor to the global list of vnet constructors. */
482 	VNET_SYSINIT_WLOCK();
483 	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
484 		if (vs2->subsystem > vs->subsystem)
485 			break;
486 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
487 			break;
488 	}
489 	if (vs2 != NULL)
490 		TAILQ_INSERT_BEFORE(vs2, vs, link);
491 	else
492 		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
493 
494 	/*
495 	 * Invoke the constructor on all the existing vnets when it is
496 	 * registered.
497 	 */
498 	VNET_FOREACH(vnet) {
499 		CURVNET_SET_QUIET(vnet);
500 		vs->func(vs->arg);
501 		CURVNET_RESTORE();
502 	}
503 	VNET_SYSINIT_WUNLOCK();
504 }
505 
506 void
507 vnet_deregister_sysinit(void *arg)
508 {
509 	struct vnet_sysinit *vs;
510 
511 	vs = arg;
512 
513 	/* Remove the constructor from the global list of vnet constructors. */
514 	VNET_SYSINIT_WLOCK();
515 	TAILQ_REMOVE(&vnet_constructors, vs, link);
516 	VNET_SYSINIT_WUNLOCK();
517 }
518 
519 void
520 vnet_register_sysuninit(void *arg)
521 {
522 	struct vnet_sysinit *vs, *vs2;
523 
524 	vs = arg;
525 
526 	/* Add the destructor to the global list of vnet destructors. */
527 	VNET_SYSINIT_WLOCK();
528 	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
529 		if (vs2->subsystem > vs->subsystem)
530 			break;
531 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
532 			break;
533 	}
534 	if (vs2 != NULL)
535 		TAILQ_INSERT_BEFORE(vs2, vs, link);
536 	else
537 		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
538 	VNET_SYSINIT_WUNLOCK();
539 }
540 
541 void
542 vnet_deregister_sysuninit(void *arg)
543 {
544 	struct vnet_sysinit *vs;
545 	struct vnet *vnet;
546 
547 	vs = arg;
548 
549 	/*
550 	 * Invoke the destructor on all the existing vnets when it is
551 	 * deregistered.
552 	 */
553 	VNET_SYSINIT_WLOCK();
554 	VNET_FOREACH(vnet) {
555 		CURVNET_SET_QUIET(vnet);
556 		vs->func(vs->arg);
557 		CURVNET_RESTORE();
558 	}
559 
560 	/* Remove the destructor from the global list of vnet destructors. */
561 	TAILQ_REMOVE(&vnet_destructors, vs, link);
562 	VNET_SYSINIT_WUNLOCK();
563 }
564 
565 /*
566  * Invoke all registered vnet constructors on the current vnet.  Used during
567  * vnet construction.  The caller is responsible for ensuring the new vnet is
568  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
569  */
570 void
571 vnet_sysinit(void)
572 {
573 	struct vnet_sysinit *vs;
574 
575 	VNET_SYSINIT_RLOCK();
576 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
577 		curvnet->vnet_state = vs->subsystem;
578 		vs->func(vs->arg);
579 	}
580 	VNET_SYSINIT_RUNLOCK();
581 }
582 
583 /*
584  * Invoke all registered vnet destructors on the current vnet.  Used during
585  * vnet destruction.  The caller is responsible for ensuring the dying vnet
586  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
587  */
588 void
589 vnet_sysuninit(void)
590 {
591 	struct vnet_sysinit *vs;
592 
593 	VNET_SYSINIT_RLOCK();
594 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
595 	    link) {
596 		curvnet->vnet_state = vs->subsystem;
597 		vs->func(vs->arg);
598 	}
599 	VNET_SYSINIT_RUNLOCK();
600 }
601 
602 /*
603  * EVENTHANDLER(9) extensions.
604  */
605 /*
606  * Invoke the eventhandler function originally registered with the possibly
607  * registered argument for all virtual network stack instances.
608  *
609  * This iterator can only be used for eventhandlers that do not take any
610  * additional arguments, as we do ignore the variadic arguments from the
611  * EVENTHANDLER_INVOKE() call.
612  */
613 void
614 vnet_global_eventhandler_iterator_func(void *arg, ...)
615 {
616 	VNET_ITERATOR_DECL(vnet_iter);
617 	struct eventhandler_entry_vimage *v_ee;
618 
619 	/*
620 	 * There is a bug here in that we should actually cast things to
621 	 * (struct eventhandler_entry_ ## name *)  but that's not easily
622 	 * possible in here so just re-using the variadic version we
623 	 * defined for the generic vimage case.
624 	 */
625 	v_ee = arg;
626 	VNET_LIST_RLOCK();
627 	VNET_FOREACH(vnet_iter) {
628 		CURVNET_SET(vnet_iter);
629 		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
630 		CURVNET_RESTORE();
631 	}
632 	VNET_LIST_RUNLOCK();
633 }
634 
635 #ifdef VNET_DEBUG
636 struct vnet_recursion {
637 	SLIST_ENTRY(vnet_recursion)	 vnr_le;
638 	const char			*prev_fn;
639 	const char			*where_fn;
640 	int				 where_line;
641 	struct vnet			*old_vnet;
642 	struct vnet			*new_vnet;
643 };
644 
645 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
646     SLIST_HEAD_INITIALIZER(vnet_recursions);
647 
648 static void
649 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
650 {
651 
652 	if (!brief)
653 		printf("CURVNET_SET() recursion in ");
654 	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
655 	    vnr->prev_fn);
656 	if (brief)
657 		printf(", ");
658 	else
659 		printf("\n    ");
660 	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
661 }
662 
663 void
664 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
665 {
666 	struct vnet_recursion *vnr;
667 
668 	/* Skip already logged recursion events. */
669 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
670 		if (vnr->prev_fn == old_fn &&
671 		    vnr->where_fn == curthread->td_vnet_lpush &&
672 		    vnr->where_line == line &&
673 		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
674 			return;
675 
676 	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
677 	if (vnr == NULL)
678 		panic("%s: malloc failed", __func__);
679 	vnr->prev_fn = old_fn;
680 	vnr->where_fn = curthread->td_vnet_lpush;
681 	vnr->where_line = line;
682 	vnr->old_vnet = old_vnet;
683 	vnr->new_vnet = curvnet;
684 
685 	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
686 
687 	vnet_print_recursion(vnr, 0);
688 #ifdef KDB
689 	kdb_backtrace();
690 #endif
691 }
692 #endif /* VNET_DEBUG */
693 
694 /*
695  * DDB(4).
696  */
697 #ifdef DDB
698 static void
699 db_vnet_print(struct vnet *vnet)
700 {
701 
702 	db_printf("vnet            = %p\n", vnet);
703 	db_printf(" vnet_magic_n   = %#08x (%s, orig %#08x)\n",
704 	    vnet->vnet_magic_n,
705 	    (vnet->vnet_magic_n == VNET_MAGIC_N) ?
706 		"ok" : "mismatch", VNET_MAGIC_N);
707 	db_printf(" vnet_ifcnt     = %u\n", vnet->vnet_ifcnt);
708 	db_printf(" vnet_sockcnt   = %u\n", vnet->vnet_sockcnt);
709 	db_printf(" vnet_data_mem  = %p\n", vnet->vnet_data_mem);
710 	db_printf(" vnet_data_base = %#jx\n",
711 	    (uintmax_t)vnet->vnet_data_base);
712 	db_printf(" vnet_state     = %#08x\n", vnet->vnet_state);
713 	db_printf("\n");
714 }
715 
716 DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets)
717 {
718 	VNET_ITERATOR_DECL(vnet_iter);
719 
720 	VNET_FOREACH(vnet_iter) {
721 		db_vnet_print(vnet_iter);
722 		if (db_pager_quit)
723 			break;
724 	}
725 }
726 
727 DB_SHOW_COMMAND(vnet, db_show_vnet)
728 {
729 
730 	if (!have_addr) {
731 		db_printf("usage: show vnet <struct vnet *>\n");
732 		return;
733 	}
734 
735 	db_vnet_print((struct vnet *)addr);
736 }
737 
738 static void
739 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
740 {
741 	const char *vsname, *funcname;
742 	c_db_sym_t sym;
743 	db_expr_t  offset;
744 
745 #define xprint(...)							\
746 	if (ddb)							\
747 		db_printf(__VA_ARGS__);					\
748 	else								\
749 		printf(__VA_ARGS__)
750 
751 	if (vs == NULL) {
752 		xprint("%s: no vnet_sysinit * given\n", __func__);
753 		return;
754 	}
755 
756 	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
757 	db_symbol_values(sym, &vsname, NULL);
758 	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
759 	db_symbol_values(sym, &funcname, NULL);
760 	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
761 	xprint("  %#08x %#08x\n", vs->subsystem, vs->order);
762 	xprint("  %p(%s)(%p)\n",
763 	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
764 #undef xprint
765 }
766 
767 DB_SHOW_COMMAND(vnet_sysinit, db_show_vnet_sysinit)
768 {
769 	struct vnet_sysinit *vs;
770 
771 	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
772 	db_printf("  Subsystem  Order\n");
773 	db_printf("  Function(Name)(Arg)\n");
774 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
775 		db_show_vnet_print_vs(vs, 1);
776 		if (db_pager_quit)
777 			break;
778 	}
779 }
780 
781 DB_SHOW_COMMAND(vnet_sysuninit, db_show_vnet_sysuninit)
782 {
783 	struct vnet_sysinit *vs;
784 
785 	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
786 	db_printf("  Subsystem  Order\n");
787 	db_printf("  Function(Name)(Arg)\n");
788 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
789 	    link) {
790 		db_show_vnet_print_vs(vs, 1);
791 		if (db_pager_quit)
792 			break;
793 	}
794 }
795 
796 #ifdef VNET_DEBUG
797 DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs)
798 {
799 	struct vnet_recursion *vnr;
800 
801 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
802 		vnet_print_recursion(vnr, 1);
803 }
804 #endif
805 #endif /* DDB */
806