xref: /freebsd/sys/net/vnet.c (revision 089104e0e01f080c9cd45dc5f34c4f636dea4ca7)
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 /*
224  * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
225  */
226 static void vnet_sysinit(void);
227 static void vnet_sysuninit(void);
228 
229 #ifdef DDB
230 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
231 #endif
232 
233 /*
234  * Allocate a virtual network stack.
235  */
236 struct vnet *
237 vnet_alloc(void)
238 {
239 	struct vnet *vnet;
240 
241 	SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
242 	vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
243 	vnet->vnet_magic_n = VNET_MAGIC_N;
244 	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
245 
246 	/*
247 	 * Allocate storage for virtualized global variables and copy in
248 	 * initial values from our 'master' copy.
249 	 */
250 	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
251 	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
252 
253 	/*
254 	 * All use of vnet-specific data will immediately subtract VNET_START
255 	 * from the base memory pointer, so pre-calculate that now to avoid
256 	 * it on each use.
257 	 */
258 	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
259 
260 	/* Initialize / attach vnet module instances. */
261 	CURVNET_SET_QUIET(vnet);
262 	vnet_sysinit();
263 	CURVNET_RESTORE();
264 
265 	VNET_LIST_WLOCK();
266 	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
267 	VNET_LIST_WUNLOCK();
268 
269 	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
270 	return (vnet);
271 }
272 
273 /*
274  * Destroy a virtual network stack.
275  */
276 void
277 vnet_destroy(struct vnet *vnet)
278 {
279 
280 	SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
281 	KASSERT(vnet->vnet_sockcnt == 0,
282 	    ("%s: vnet still has sockets", __func__));
283 
284 	VNET_LIST_WLOCK();
285 	LIST_REMOVE(vnet, vnet_le);
286 	VNET_LIST_WUNLOCK();
287 
288 	/* Signal that VNET is being shutdown. */
289 	vnet->vnet_shutdown = true;
290 
291 	CURVNET_SET_QUIET(vnet);
292 	sx_xlock(&ifnet_detach_sxlock);
293 	vnet_sysuninit();
294 	sx_xunlock(&ifnet_detach_sxlock);
295 	CURVNET_RESTORE();
296 
297 	/*
298 	 * Release storage for the virtual network stack instance.
299 	 */
300 	free(vnet->vnet_data_mem, M_VNET_DATA);
301 	vnet->vnet_data_mem = NULL;
302 	vnet->vnet_data_base = 0;
303 	vnet->vnet_magic_n = 0xdeadbeef;
304 	free(vnet, M_VNET);
305 	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
306 }
307 
308 /*
309  * Boot time initialization and allocation of virtual network stacks.
310  */
311 static void
312 vnet_init_prelink(void *arg __unused)
313 {
314 
315 	rw_init(&vnet_rwlock, "vnet_rwlock");
316 	sx_init(&vnet_sxlock, "vnet_sxlock");
317 	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
318 	LIST_INIT(&vnet_head);
319 }
320 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
321     vnet_init_prelink, NULL);
322 
323 static void
324 vnet0_init(void *arg __unused)
325 {
326 
327 	if (bootverbose)
328 		printf("VIMAGE (virtualized network stack) enabled\n");
329 
330 	/*
331 	 * We MUST clear curvnet in vi_init_done() before going SMP,
332 	 * otherwise CURVNET_SET() macros would scream about unnecessary
333 	 * curvnet recursions.
334 	 */
335 	curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
336 }
337 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
338 
339 static void
340 vnet_init_done(void *unused __unused)
341 {
342 
343 	curvnet = NULL;
344 }
345 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done,
346     NULL);
347 
348 /*
349  * Once on boot, initialize the modspace freelist to entirely cover modspace.
350  */
351 static void
352 vnet_data_startup(void *dummy __unused)
353 {
354 	struct vnet_data_free *df;
355 
356 	df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
357 	df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
358 	df->vnd_len = VNET_MODMIN;
359 	TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
360 	sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
361 }
362 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL);
363 
364 /* Dummy VNET_SYSINIT to make sure we always reach the final end state. */
365 static void
366 vnet_sysinit_done(void *unused __unused)
367 {
368 
369 	return;
370 }
371 VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY,
372     vnet_sysinit_done, NULL);
373 
374 /*
375  * When a module is loaded and requires storage for a virtualized global
376  * variable, allocate space from the modspace free list.  This interface
377  * should be used only by the kernel linker.
378  */
379 void *
380 vnet_data_alloc(int size)
381 {
382 	struct vnet_data_free *df;
383 	void *s;
384 
385 	s = NULL;
386 	size = roundup2(size, sizeof(void *));
387 	sx_xlock(&vnet_data_free_lock);
388 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
389 		if (df->vnd_len < size)
390 			continue;
391 		if (df->vnd_len == size) {
392 			s = (void *)df->vnd_start;
393 			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
394 			free(df, M_VNET_DATA_FREE);
395 			break;
396 		}
397 		s = (void *)df->vnd_start;
398 		df->vnd_len -= size;
399 		df->vnd_start = df->vnd_start + size;
400 		break;
401 	}
402 	sx_xunlock(&vnet_data_free_lock);
403 
404 	return (s);
405 }
406 
407 /*
408  * Free space for a virtualized global variable on module unload.
409  */
410 void
411 vnet_data_free(void *start_arg, int size)
412 {
413 	struct vnet_data_free *df;
414 	struct vnet_data_free *dn;
415 	uintptr_t start;
416 	uintptr_t end;
417 
418 	size = roundup2(size, sizeof(void *));
419 	start = (uintptr_t)start_arg;
420 	end = start + size;
421 	/*
422 	 * Free a region of space and merge it with as many neighbors as
423 	 * possible.  Keeping the list sorted simplifies this operation.
424 	 */
425 	sx_xlock(&vnet_data_free_lock);
426 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
427 		if (df->vnd_start > end)
428 			break;
429 		/*
430 		 * If we expand at the end of an entry we may have to merge
431 		 * it with the one following it as well.
432 		 */
433 		if (df->vnd_start + df->vnd_len == start) {
434 			df->vnd_len += size;
435 			dn = TAILQ_NEXT(df, vnd_link);
436 			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
437 				df->vnd_len += dn->vnd_len;
438 				TAILQ_REMOVE(&vnet_data_free_head, dn,
439 				    vnd_link);
440 				free(dn, M_VNET_DATA_FREE);
441 			}
442 			sx_xunlock(&vnet_data_free_lock);
443 			return;
444 		}
445 		if (df->vnd_start == end) {
446 			df->vnd_start = start;
447 			df->vnd_len += size;
448 			sx_xunlock(&vnet_data_free_lock);
449 			return;
450 		}
451 	}
452 	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
453 	dn->vnd_start = start;
454 	dn->vnd_len = size;
455 	if (df)
456 		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
457 	else
458 		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
459 	sx_xunlock(&vnet_data_free_lock);
460 }
461 
462 /*
463  * When a new virtualized global variable has been allocated, propagate its
464  * initial value to each already-allocated virtual network stack instance.
465  */
466 void
467 vnet_data_copy(void *start, int size)
468 {
469 	struct vnet *vnet;
470 
471 	VNET_LIST_RLOCK();
472 	LIST_FOREACH(vnet, &vnet_head, vnet_le)
473 		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
474 		    (uintptr_t)start), start, size);
475 	VNET_LIST_RUNLOCK();
476 }
477 
478 /*
479  * Support for special SYSINIT handlers registered via VNET_SYSINIT()
480  * and VNET_SYSUNINIT().
481  */
482 void
483 vnet_register_sysinit(void *arg)
484 {
485 	struct vnet_sysinit *vs, *vs2;
486 	struct vnet *vnet;
487 
488 	vs = arg;
489 	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
490 
491 	/* Add the constructor to the global list of vnet constructors. */
492 	VNET_SYSINIT_WLOCK();
493 	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
494 		if (vs2->subsystem > vs->subsystem)
495 			break;
496 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
497 			break;
498 	}
499 	if (vs2 != NULL)
500 		TAILQ_INSERT_BEFORE(vs2, vs, link);
501 	else
502 		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
503 
504 	/*
505 	 * Invoke the constructor on all the existing vnets when it is
506 	 * registered.
507 	 */
508 	VNET_FOREACH(vnet) {
509 		CURVNET_SET_QUIET(vnet);
510 		vs->func(vs->arg);
511 		CURVNET_RESTORE();
512 	}
513 	VNET_SYSINIT_WUNLOCK();
514 }
515 
516 void
517 vnet_deregister_sysinit(void *arg)
518 {
519 	struct vnet_sysinit *vs;
520 
521 	vs = arg;
522 
523 	/* Remove the constructor from the global list of vnet constructors. */
524 	VNET_SYSINIT_WLOCK();
525 	TAILQ_REMOVE(&vnet_constructors, vs, link);
526 	VNET_SYSINIT_WUNLOCK();
527 }
528 
529 void
530 vnet_register_sysuninit(void *arg)
531 {
532 	struct vnet_sysinit *vs, *vs2;
533 
534 	vs = arg;
535 
536 	/* Add the destructor to the global list of vnet destructors. */
537 	VNET_SYSINIT_WLOCK();
538 	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
539 		if (vs2->subsystem > vs->subsystem)
540 			break;
541 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
542 			break;
543 	}
544 	if (vs2 != NULL)
545 		TAILQ_INSERT_BEFORE(vs2, vs, link);
546 	else
547 		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
548 	VNET_SYSINIT_WUNLOCK();
549 }
550 
551 void
552 vnet_deregister_sysuninit(void *arg)
553 {
554 	struct vnet_sysinit *vs;
555 	struct vnet *vnet;
556 
557 	vs = arg;
558 
559 	/*
560 	 * Invoke the destructor on all the existing vnets when it is
561 	 * deregistered.
562 	 */
563 	VNET_SYSINIT_WLOCK();
564 	VNET_FOREACH(vnet) {
565 		CURVNET_SET_QUIET(vnet);
566 		vs->func(vs->arg);
567 		CURVNET_RESTORE();
568 	}
569 
570 	/* Remove the destructor from the global list of vnet destructors. */
571 	TAILQ_REMOVE(&vnet_destructors, vs, link);
572 	VNET_SYSINIT_WUNLOCK();
573 }
574 
575 /*
576  * Invoke all registered vnet constructors on the current vnet.  Used during
577  * vnet construction.  The caller is responsible for ensuring the new vnet is
578  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
579  */
580 static void
581 vnet_sysinit(void)
582 {
583 	struct vnet_sysinit *vs;
584 
585 	VNET_SYSINIT_RLOCK();
586 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
587 		curvnet->vnet_state = vs->subsystem;
588 		vs->func(vs->arg);
589 	}
590 	VNET_SYSINIT_RUNLOCK();
591 }
592 
593 /*
594  * Invoke all registered vnet destructors on the current vnet.  Used during
595  * vnet destruction.  The caller is responsible for ensuring the dying vnet
596  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
597  */
598 static void
599 vnet_sysuninit(void)
600 {
601 	struct vnet_sysinit *vs;
602 
603 	VNET_SYSINIT_RLOCK();
604 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
605 	    link) {
606 		curvnet->vnet_state = vs->subsystem;
607 		vs->func(vs->arg);
608 	}
609 	VNET_SYSINIT_RUNLOCK();
610 }
611 
612 /*
613  * EVENTHANDLER(9) extensions.
614  */
615 /*
616  * Invoke the eventhandler function originally registered with the possibly
617  * registered argument for all virtual network stack instances.
618  *
619  * This iterator can only be used for eventhandlers that do not take any
620  * additional arguments, as we do ignore the variadic arguments from the
621  * EVENTHANDLER_INVOKE() call.
622  */
623 void
624 vnet_global_eventhandler_iterator_func(void *arg, ...)
625 {
626 	VNET_ITERATOR_DECL(vnet_iter);
627 	struct eventhandler_entry_vimage *v_ee;
628 
629 	/*
630 	 * There is a bug here in that we should actually cast things to
631 	 * (struct eventhandler_entry_ ## name *)  but that's not easily
632 	 * possible in here so just re-using the variadic version we
633 	 * defined for the generic vimage case.
634 	 */
635 	v_ee = arg;
636 	VNET_LIST_RLOCK();
637 	VNET_FOREACH(vnet_iter) {
638 		CURVNET_SET(vnet_iter);
639 		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
640 		CURVNET_RESTORE();
641 	}
642 	VNET_LIST_RUNLOCK();
643 }
644 
645 #ifdef VNET_DEBUG
646 struct vnet_recursion {
647 	SLIST_ENTRY(vnet_recursion)	 vnr_le;
648 	const char			*prev_fn;
649 	const char			*where_fn;
650 	int				 where_line;
651 	struct vnet			*old_vnet;
652 	struct vnet			*new_vnet;
653 };
654 
655 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
656     SLIST_HEAD_INITIALIZER(vnet_recursions);
657 
658 static void
659 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
660 {
661 
662 	if (!brief)
663 		printf("CURVNET_SET() recursion in ");
664 	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
665 	    vnr->prev_fn);
666 	if (brief)
667 		printf(", ");
668 	else
669 		printf("\n    ");
670 	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
671 }
672 
673 void
674 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
675 {
676 	struct vnet_recursion *vnr;
677 
678 	/* Skip already logged recursion events. */
679 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
680 		if (vnr->prev_fn == old_fn &&
681 		    vnr->where_fn == curthread->td_vnet_lpush &&
682 		    vnr->where_line == line &&
683 		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
684 			return;
685 
686 	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
687 	if (vnr == NULL)
688 		panic("%s: malloc failed", __func__);
689 	vnr->prev_fn = old_fn;
690 	vnr->where_fn = curthread->td_vnet_lpush;
691 	vnr->where_line = line;
692 	vnr->old_vnet = old_vnet;
693 	vnr->new_vnet = curvnet;
694 
695 	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
696 
697 	vnet_print_recursion(vnr, 0);
698 #ifdef KDB
699 	kdb_backtrace();
700 #endif
701 }
702 #endif /* VNET_DEBUG */
703 
704 /*
705  * DDB(4).
706  */
707 #ifdef DDB
708 static void
709 db_vnet_print(struct vnet *vnet)
710 {
711 
712 	db_printf("vnet            = %p\n", vnet);
713 	db_printf(" vnet_magic_n   = %#08x (%s, orig %#08x)\n",
714 	    vnet->vnet_magic_n,
715 	    (vnet->vnet_magic_n == VNET_MAGIC_N) ?
716 		"ok" : "mismatch", VNET_MAGIC_N);
717 	db_printf(" vnet_ifcnt     = %u\n", vnet->vnet_ifcnt);
718 	db_printf(" vnet_sockcnt   = %u\n", vnet->vnet_sockcnt);
719 	db_printf(" vnet_data_mem  = %p\n", vnet->vnet_data_mem);
720 	db_printf(" vnet_data_base = %#jx\n",
721 	    (uintmax_t)vnet->vnet_data_base);
722 	db_printf(" vnet_state     = %#08x\n", vnet->vnet_state);
723 	db_printf(" vnet_shutdown  = %#03x\n", vnet->vnet_shutdown);
724 	db_printf("\n");
725 }
726 
727 DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets)
728 {
729 	VNET_ITERATOR_DECL(vnet_iter);
730 
731 	VNET_FOREACH(vnet_iter) {
732 		db_vnet_print(vnet_iter);
733 		if (db_pager_quit)
734 			break;
735 	}
736 }
737 
738 DB_SHOW_COMMAND(vnet, db_show_vnet)
739 {
740 
741 	if (!have_addr) {
742 		db_printf("usage: show vnet <struct vnet *>\n");
743 		return;
744 	}
745 
746 	db_vnet_print((struct vnet *)addr);
747 }
748 
749 static void
750 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
751 {
752 	const char *vsname, *funcname;
753 	c_db_sym_t sym;
754 	db_expr_t  offset;
755 
756 #define xprint(...)							\
757 	if (ddb)							\
758 		db_printf(__VA_ARGS__);					\
759 	else								\
760 		printf(__VA_ARGS__)
761 
762 	if (vs == NULL) {
763 		xprint("%s: no vnet_sysinit * given\n", __func__);
764 		return;
765 	}
766 
767 	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
768 	db_symbol_values(sym, &vsname, NULL);
769 	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
770 	db_symbol_values(sym, &funcname, NULL);
771 	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
772 	xprint("  %#08x %#08x\n", vs->subsystem, vs->order);
773 	xprint("  %p(%s)(%p)\n",
774 	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
775 #undef xprint
776 }
777 
778 DB_SHOW_COMMAND_FLAGS(vnet_sysinit, db_show_vnet_sysinit, DB_CMD_MEMSAFE)
779 {
780 	struct vnet_sysinit *vs;
781 
782 	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
783 	db_printf("  Subsystem  Order\n");
784 	db_printf("  Function(Name)(Arg)\n");
785 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
786 		db_show_vnet_print_vs(vs, 1);
787 		if (db_pager_quit)
788 			break;
789 	}
790 }
791 
792 DB_SHOW_COMMAND_FLAGS(vnet_sysuninit, db_show_vnet_sysuninit, DB_CMD_MEMSAFE)
793 {
794 	struct vnet_sysinit *vs;
795 
796 	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
797 	db_printf("  Subsystem  Order\n");
798 	db_printf("  Function(Name)(Arg)\n");
799 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
800 	    link) {
801 		db_show_vnet_print_vs(vs, 1);
802 		if (db_pager_quit)
803 			break;
804 	}
805 }
806 
807 #ifdef VNET_DEBUG
808 DB_SHOW_COMMAND_FLAGS(vnetrcrs, db_show_vnetrcrs, DB_CMD_MEMSAFE)
809 {
810 	struct vnet_recursion *vnr;
811 
812 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
813 		vnet_print_recursion(vnr, 1);
814 }
815 #endif
816 #endif /* DDB */
817