1/*-
2 * Copyright (c) 2004-2009 University of Zagreb
3 * Copyright (c) 2006-2009 FreeBSD Foundation
4 * All rights reserved.
5 *
6 * This software was developed by the University of Zagreb and the
7 * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
8 * FreeBSD Foundation.
9 *
10 * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
11 * Copyright (c) 2009 Robert N. M. Watson
12 * All rights reserved.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * 1. Redistributions of source code must retain the above copyright
18 *    notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 *    notice, this list of conditions and the following disclaimer in the
21 *    documentation and/or other materials provided with the distribution.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD$");
38
39#include "opt_ddb.h"
40#include "opt_kdb.h"
41#include "opt_kdtrace.h"
42
43#include <sys/param.h>
44#include <sys/kdb.h>
45#include <sys/kernel.h>
46#include <sys/jail.h>
47#include <sys/sdt.h>
48#include <sys/systm.h>
49#include <sys/sysctl.h>
50#include <sys/eventhandler.h>
51#include <sys/lock.h>
52#include <sys/malloc.h>
53#include <sys/proc.h>
54#include <sys/socket.h>
55#include <sys/sx.h>
56#include <sys/sysctl.h>
57
58#include <machine/stdarg.h>
59
60#ifdef DDB
61#include <ddb/ddb.h>
62#include <ddb/db_sym.h>
63#endif
64
65#include <net/if.h>
66#include <net/if_var.h>
67#include <net/vnet.h>
68
69/*-
70 * This file implements core functions for virtual network stacks:
71 *
72 * - Virtual network stack management functions.
73 *
74 * - Virtual network stack memory allocator, which virtualizes global
75 *   variables in the network stack
76 *
77 * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
78 *   to register startup/shutdown events to be run for each virtual network
79 *   stack instance.
80 */
81
82FEATURE(vimage, "VIMAGE kernel virtualization");
83
84static MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
85
86/*
87 * The virtual network stack list has two read-write locks, one sleepable and
88 * the other not, so that the list can be stablized and walked in a variety
89 * of network stack contexts.  Both must be acquired exclusively to modify
90 * the list, but a read lock of either lock is sufficient to walk the list.
91 */
92struct rwlock		vnet_rwlock;
93struct sx		vnet_sxlock;
94
95#define	VNET_LIST_WLOCK() do {						\
96	sx_xlock(&vnet_sxlock);						\
97	rw_wlock(&vnet_rwlock);						\
98} while (0)
99
100#define	VNET_LIST_WUNLOCK() do {					\
101	rw_wunlock(&vnet_rwlock);					\
102	sx_xunlock(&vnet_sxlock);					\
103} while (0)
104
105struct vnet_list_head vnet_head;
106struct vnet *vnet0;
107
108/*
109 * The virtual network stack allocator provides storage for virtualized
110 * global variables.  These variables are defined/declared using the
111 * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet'
112 * linker set.  The details of the implementation are somewhat subtle, but
113 * allow the majority of most network subsystems to maintain
114 * virtualization-agnostic.
115 *
116 * The virtual network stack allocator handles variables in the base kernel
117 * vs. modules in similar but different ways.  In both cases, virtualized
118 * global variables are marked as such by being declared to be part of the
119 * vnet linker set.  These "master" copies of global variables serve two
120 * functions:
121 *
122 * (1) They contain static initialization or "default" values for global
123 *     variables which will be propagated to each virtual network stack
124 *     instance when created.  As with normal global variables, they default
125 *     to zero-filled.
126 *
127 * (2) They act as unique global names by which the variable can be referred
128 *     to, regardless of network stack instance.  The single global symbol
129 *     will be used to calculate the location of a per-virtual instance
130 *     variable at run-time.
131 *
132 * Each virtual network stack instance has a complete copy of each
133 * virtualized global variable, stored in a malloc'd block of memory
134 * referred to by vnet->vnet_data_mem.  Critical to the design is that each
135 * per-instance memory block is laid out identically to the master block so
136 * that the offset of each global variable is the same across all blocks.  To
137 * optimize run-time access, a precalculated 'base' address,
138 * vnet->vnet_data_base, is stored in each vnet, and is the amount that can
139 * be added to the address of a 'master' instance of a variable to get to the
140 * per-vnet instance.
141 *
142 * Virtualized global variables are handled in a similar manner, but as each
143 * module has its own 'set_vnet' linker set, and we want to keep all
144 * virtualized globals togther, we reserve space in the kernel's linker set
145 * for potential module variables using a per-vnet character array,
146 * 'modspace'.  The virtual network stack allocator maintains a free list to
147 * track what space in the array is free (all, initially) and as modules are
148 * linked, allocates portions of the space to specific globals.  The kernel
149 * module linker queries the virtual network stack allocator and will
150 * bind references of the global to the location during linking.  It also
151 * calls into the virtual network stack allocator, once the memory is
152 * initialized, in order to propagate the new static initializations to all
153 * existing virtual network stack instances so that the soon-to-be executing
154 * module will find every network stack instance with proper default values.
155 */
156
157/*
158 * Number of bytes of data in the 'set_vnet' linker set, and hence the total
159 * size of all kernel virtualized global variables, and the malloc(9) type
160 * that will be used to allocate it.
161 */
162#define	VNET_BYTES	(VNET_STOP - VNET_START)
163
164static MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
165
166/*
167 * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of
168 * global variables across all loaded modules.  As this actually sizes an
169 * array declared as a virtualized global variable in the kernel itself, and
170 * we want the virtualized global variable space to be page-sized, we may
171 * have more space than that in practice.
172 */
173#define	VNET_MODMIN	8192
174#define	VNET_SIZE	roundup2(VNET_BYTES, PAGE_SIZE)
175#define	VNET_MODSIZE	(VNET_SIZE - (VNET_BYTES - VNET_MODMIN))
176
177/*
178 * Space to store virtualized global variables from loadable kernel modules,
179 * and the free list to manage it.
180 */
181static VNET_DEFINE(char, modspace[VNET_MODMIN]);
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 */
188static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
189	TAILQ_HEAD_INITIALIZER(vnet_constructors);
190static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
191	TAILQ_HEAD_INITIALIZER(vnet_destructors);
192
193struct 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
200struct vnet_data_free {
201	uintptr_t	vnd_start;
202	int		vnd_len;
203	TAILQ_ENTRY(vnet_data_free) vnd_link;
204};
205
206static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
207    "VNET resource accounting");
208static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
209	    TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
210static struct sx vnet_data_free_lock;
211
212SDT_PROVIDER_DEFINE(vnet);
213SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
214SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
215    "struct vnet *");
216SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
217    "int", "struct vnet *");
218SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
219    "int", "struct vnet *");
220SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
221    "int");
222
223#ifdef DDB
224static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
225#endif
226
227/*
228 * Allocate a virtual network stack.
229 */
230struct vnet *
231vnet_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	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
239
240	/*
241	 * Allocate storage for virtualized global variables and copy in
242	 * initial values form our 'master' copy.
243	 */
244	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
245	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
246
247	/*
248	 * All use of vnet-specific data will immediately subtract VNET_START
249	 * from the base memory pointer, so pre-calculate that now to avoid
250	 * it on each use.
251	 */
252	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
253
254	/* Initialize / attach vnet module instances. */
255	CURVNET_SET_QUIET(vnet);
256	vnet_sysinit();
257	CURVNET_RESTORE();
258
259	VNET_LIST_WLOCK();
260	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
261	VNET_LIST_WUNLOCK();
262
263	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
264	return (vnet);
265}
266
267/*
268 * Destroy a virtual network stack.
269 */
270void
271vnet_destroy(struct vnet *vnet)
272{
273	struct ifnet *ifp, *nifp;
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
285	/* Return all inherited interfaces to their parent vnets. */
286	TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
287		if (ifp->if_home_vnet != ifp->if_vnet)
288			if_vmove(ifp, ifp->if_home_vnet);
289	}
290
291	vnet_sysuninit();
292	CURVNET_RESTORE();
293
294	/*
295	 * Release storage for the virtual network stack instance.
296	 */
297	free(vnet->vnet_data_mem, M_VNET_DATA);
298	vnet->vnet_data_mem = NULL;
299	vnet->vnet_data_base = 0;
300	vnet->vnet_magic_n = 0xdeadbeef;
301	free(vnet, M_VNET);
302	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
303}
304
305/*
306 * Boot time initialization and allocation of virtual network stacks.
307 */
308static void
309vnet_init_prelink(void *arg)
310{
311
312	rw_init(&vnet_rwlock, "vnet_rwlock");
313	sx_init(&vnet_sxlock, "vnet_sxlock");
314	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
315	LIST_INIT(&vnet_head);
316}
317SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
318    vnet_init_prelink, NULL);
319
320static void
321vnet0_init(void *arg)
322{
323
324	/* Warn people before take off - in case we crash early. */
325	printf("WARNING: VIMAGE (virtualized network stack) is a highly "
326	    "experimental feature.\n");
327
328	/*
329	 * We MUST clear curvnet in vi_init_done() before going SMP,
330	 * otherwise CURVNET_SET() macros would scream about unnecessary
331	 * curvnet recursions.
332	 */
333	curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
334}
335SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
336
337static void
338vnet_init_done(void *unused)
339{
340
341	curvnet = NULL;
342}
343
344SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, vnet_init_done,
345    NULL);
346
347/*
348 * Once on boot, initialize the modspace freelist to entirely cover modspace.
349 */
350static void
351vnet_data_startup(void *dummy __unused)
352{
353	struct vnet_data_free *df;
354
355	df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
356	df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
357	df->vnd_len = VNET_MODMIN;
358	TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
359	sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
360}
361SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, 0);
362
363/*
364 * When a module is loaded and requires storage for a virtualized global
365 * variable, allocate space from the modspace free list.  This interface
366 * should be used only by the kernel linker.
367 */
368void *
369vnet_data_alloc(int size)
370{
371	struct vnet_data_free *df;
372	void *s;
373
374	s = NULL;
375	size = roundup2(size, sizeof(void *));
376	sx_xlock(&vnet_data_free_lock);
377	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
378		if (df->vnd_len < size)
379			continue;
380		if (df->vnd_len == size) {
381			s = (void *)df->vnd_start;
382			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
383			free(df, M_VNET_DATA_FREE);
384			break;
385		}
386		s = (void *)df->vnd_start;
387		df->vnd_len -= size;
388		df->vnd_start = df->vnd_start + size;
389		break;
390	}
391	sx_xunlock(&vnet_data_free_lock);
392
393	return (s);
394}
395
396/*
397 * Free space for a virtualized global variable on module unload.
398 */
399void
400vnet_data_free(void *start_arg, int size)
401{
402	struct vnet_data_free *df;
403	struct vnet_data_free *dn;
404	uintptr_t start;
405	uintptr_t end;
406
407	size = roundup2(size, sizeof(void *));
408	start = (uintptr_t)start_arg;
409	end = start + size;
410	/*
411	 * Free a region of space and merge it with as many neighbors as
412	 * possible.  Keeping the list sorted simplifies this operation.
413	 */
414	sx_xlock(&vnet_data_free_lock);
415	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
416		if (df->vnd_start > end)
417			break;
418		/*
419		 * If we expand at the end of an entry we may have to merge
420		 * it with the one following it as well.
421		 */
422		if (df->vnd_start + df->vnd_len == start) {
423			df->vnd_len += size;
424			dn = TAILQ_NEXT(df, vnd_link);
425			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
426				df->vnd_len += dn->vnd_len;
427				TAILQ_REMOVE(&vnet_data_free_head, dn,
428				    vnd_link);
429				free(dn, M_VNET_DATA_FREE);
430			}
431			sx_xunlock(&vnet_data_free_lock);
432			return;
433		}
434		if (df->vnd_start == end) {
435			df->vnd_start = start;
436			df->vnd_len += size;
437			sx_xunlock(&vnet_data_free_lock);
438			return;
439		}
440	}
441	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
442	dn->vnd_start = start;
443	dn->vnd_len = size;
444	if (df)
445		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
446	else
447		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
448	sx_xunlock(&vnet_data_free_lock);
449}
450
451/*
452 * When a new virtualized global variable has been allocated, propagate its
453 * initial value to each already-allocated virtual network stack instance.
454 */
455void
456vnet_data_copy(void *start, int size)
457{
458	struct vnet *vnet;
459
460	VNET_LIST_RLOCK();
461	LIST_FOREACH(vnet, &vnet_head, vnet_le)
462		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
463		    (uintptr_t)start), start, size);
464	VNET_LIST_RUNLOCK();
465}
466
467/*
468 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
469 * and VNET_SYSUNINIT().
470 */
471void
472vnet_register_sysinit(void *arg)
473{
474	struct vnet_sysinit *vs, *vs2;
475	struct vnet *vnet;
476
477	vs = arg;
478	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
479
480	/* Add the constructor to the global list of vnet constructors. */
481	VNET_SYSINIT_WLOCK();
482	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
483		if (vs2->subsystem > vs->subsystem)
484			break;
485		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
486			break;
487	}
488	if (vs2 != NULL)
489		TAILQ_INSERT_BEFORE(vs2, vs, link);
490	else
491		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
492
493	/*
494	 * Invoke the constructor on all the existing vnets when it is
495	 * registered.
496	 */
497	VNET_FOREACH(vnet) {
498		CURVNET_SET_QUIET(vnet);
499		vs->func(vs->arg);
500		CURVNET_RESTORE();
501	}
502	VNET_SYSINIT_WUNLOCK();
503}
504
505void
506vnet_deregister_sysinit(void *arg)
507{
508	struct vnet_sysinit *vs;
509
510	vs = arg;
511
512	/* Remove the constructor from the global list of vnet constructors. */
513	VNET_SYSINIT_WLOCK();
514	TAILQ_REMOVE(&vnet_constructors, vs, link);
515	VNET_SYSINIT_WUNLOCK();
516}
517
518void
519vnet_register_sysuninit(void *arg)
520{
521	struct vnet_sysinit *vs, *vs2;
522
523	vs = arg;
524
525	/* Add the destructor to the global list of vnet destructors. */
526	VNET_SYSINIT_WLOCK();
527	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
528		if (vs2->subsystem > vs->subsystem)
529			break;
530		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
531			break;
532	}
533	if (vs2 != NULL)
534		TAILQ_INSERT_BEFORE(vs2, vs, link);
535	else
536		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
537	VNET_SYSINIT_WUNLOCK();
538}
539
540void
541vnet_deregister_sysuninit(void *arg)
542{
543	struct vnet_sysinit *vs;
544	struct vnet *vnet;
545
546	vs = arg;
547
548	/*
549	 * Invoke the destructor on all the existing vnets when it is
550	 * deregistered.
551	 */
552	VNET_SYSINIT_WLOCK();
553	VNET_FOREACH(vnet) {
554		CURVNET_SET_QUIET(vnet);
555		vs->func(vs->arg);
556		CURVNET_RESTORE();
557	}
558
559	/* Remove the destructor from the global list of vnet destructors. */
560	TAILQ_REMOVE(&vnet_destructors, vs, link);
561	VNET_SYSINIT_WUNLOCK();
562}
563
564/*
565 * Invoke all registered vnet constructors on the current vnet.  Used during
566 * vnet construction.  The caller is responsible for ensuring the new vnet is
567 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
568 */
569void
570vnet_sysinit(void)
571{
572	struct vnet_sysinit *vs;
573
574	VNET_SYSINIT_RLOCK();
575	TAILQ_FOREACH(vs, &vnet_constructors, link) {
576		vs->func(vs->arg);
577	}
578	VNET_SYSINIT_RUNLOCK();
579}
580
581/*
582 * Invoke all registered vnet destructors on the current vnet.  Used during
583 * vnet destruction.  The caller is responsible for ensuring the dying vnet
584 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
585 */
586void
587vnet_sysuninit(void)
588{
589	struct vnet_sysinit *vs;
590
591	VNET_SYSINIT_RLOCK();
592	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
593	    link) {
594		vs->func(vs->arg);
595	}
596	VNET_SYSINIT_RUNLOCK();
597}
598
599/*
600 * EVENTHANDLER(9) extensions.
601 */
602/*
603 * Invoke the eventhandler function originally registered with the possibly
604 * registered argument for all virtual network stack instances.
605 *
606 * This iterator can only be used for eventhandlers that do not take any
607 * additional arguments, as we do ignore the variadic arguments from the
608 * EVENTHANDLER_INVOKE() call.
609 */
610void
611vnet_global_eventhandler_iterator_func(void *arg, ...)
612{
613	VNET_ITERATOR_DECL(vnet_iter);
614	struct eventhandler_entry_vimage *v_ee;
615
616	/*
617	 * There is a bug here in that we should actually cast things to
618	 * (struct eventhandler_entry_ ## name *)  but that's not easily
619	 * possible in here so just re-using the variadic version we
620	 * defined for the generic vimage case.
621	 */
622	v_ee = arg;
623	VNET_LIST_RLOCK();
624	VNET_FOREACH(vnet_iter) {
625		CURVNET_SET(vnet_iter);
626		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
627		CURVNET_RESTORE();
628	}
629	VNET_LIST_RUNLOCK();
630}
631
632#ifdef VNET_DEBUG
633struct vnet_recursion {
634	SLIST_ENTRY(vnet_recursion)	 vnr_le;
635	const char			*prev_fn;
636	const char			*where_fn;
637	int				 where_line;
638	struct vnet			*old_vnet;
639	struct vnet			*new_vnet;
640};
641
642static SLIST_HEAD(, vnet_recursion) vnet_recursions =
643    SLIST_HEAD_INITIALIZER(vnet_recursions);
644
645static void
646vnet_print_recursion(struct vnet_recursion *vnr, int brief)
647{
648
649	if (!brief)
650		printf("CURVNET_SET() recursion in ");
651	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
652	    vnr->prev_fn);
653	if (brief)
654		printf(", ");
655	else
656		printf("\n    ");
657	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
658}
659
660void
661vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
662{
663	struct vnet_recursion *vnr;
664
665	/* Skip already logged recursion events. */
666	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
667		if (vnr->prev_fn == old_fn &&
668		    vnr->where_fn == curthread->td_vnet_lpush &&
669		    vnr->where_line == line &&
670		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
671			return;
672
673	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
674	if (vnr == NULL)
675		panic("%s: malloc failed", __func__);
676	vnr->prev_fn = old_fn;
677	vnr->where_fn = curthread->td_vnet_lpush;
678	vnr->where_line = line;
679	vnr->old_vnet = old_vnet;
680	vnr->new_vnet = curvnet;
681
682	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
683
684	vnet_print_recursion(vnr, 0);
685#ifdef KDB
686	kdb_backtrace();
687#endif
688}
689#endif /* VNET_DEBUG */
690
691/*
692 * DDB(4).
693 */
694#ifdef DDB
695DB_SHOW_COMMAND(vnets, db_show_vnets)
696{
697	VNET_ITERATOR_DECL(vnet_iter);
698
699	VNET_FOREACH(vnet_iter) {
700		db_printf("vnet            = %p\n", vnet_iter);
701		db_printf(" vnet_magic_n   = 0x%x (%s, orig 0x%x)\n",
702		    vnet_iter->vnet_magic_n,
703		    (vnet_iter->vnet_magic_n == VNET_MAGIC_N) ?
704			"ok" : "mismatch", VNET_MAGIC_N);
705		db_printf(" vnet_ifcnt     = %u\n", vnet_iter->vnet_ifcnt);
706		db_printf(" vnet_sockcnt   = %u\n", vnet_iter->vnet_sockcnt);
707		db_printf(" vnet_data_mem  = %p\n", vnet_iter->vnet_data_mem);
708		db_printf(" vnet_data_base = 0x%jx\n",
709		    (uintmax_t)vnet_iter->vnet_data_base);
710		db_printf("\n");
711		if (db_pager_quit)
712			break;
713	}
714}
715
716static void
717db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
718{
719	const char *vsname, *funcname;
720	c_db_sym_t sym;
721	db_expr_t  offset;
722
723#define xprint(...)							\
724	if (ddb)							\
725		db_printf(__VA_ARGS__);					\
726	else								\
727		printf(__VA_ARGS__)
728
729	if (vs == NULL) {
730		xprint("%s: no vnet_sysinit * given\n", __func__);
731		return;
732	}
733
734	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
735	db_symbol_values(sym, &vsname, NULL);
736	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
737	db_symbol_values(sym, &funcname, NULL);
738	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
739	xprint("  0x%08x 0x%08x\n", vs->subsystem, vs->order);
740	xprint("  %p(%s)(%p)\n",
741	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
742#undef xprint
743}
744
745DB_SHOW_COMMAND(vnet_sysinit, db_show_vnet_sysinit)
746{
747	struct vnet_sysinit *vs;
748
749	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
750	db_printf("  Subsystem  Order\n");
751	db_printf("  Function(Name)(Arg)\n");
752	TAILQ_FOREACH(vs, &vnet_constructors, link) {
753		db_show_vnet_print_vs(vs, 1);
754		if (db_pager_quit)
755			break;
756	}
757}
758
759DB_SHOW_COMMAND(vnet_sysuninit, db_show_vnet_sysuninit)
760{
761	struct vnet_sysinit *vs;
762
763	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
764	db_printf("  Subsystem  Order\n");
765	db_printf("  Function(Name)(Arg)\n");
766	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
767	    link) {
768		db_show_vnet_print_vs(vs, 1);
769		if (db_pager_quit)
770			break;
771	}
772}
773
774#ifdef VNET_DEBUG
775DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs)
776{
777	struct vnet_recursion *vnr;
778
779	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
780		vnet_print_recursion(vnr, 1);
781}
782#endif
783#endif /* DDB */
784