1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
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#include "opt_ddb.h"
40#include "opt_kdb.h"
41
42#include <sys/param.h>
43#include <sys/kdb.h>
44#include <sys/kernel.h>
45#include <sys/jail.h>
46#include <sys/sdt.h>
47#include <sys/systm.h>
48#include <sys/sysctl.h>
49#include <sys/eventhandler.h>
50#include <sys/lock.h>
51#include <sys/malloc.h>
52#include <sys/proc.h>
53#include <sys/socket.h>
54#include <sys/sx.h>
55#include <sys/sysctl.h>
56
57#include <machine/stdarg.h>
58
59#ifdef DDB
60#include <ddb/ddb.h>
61#include <ddb/db_sym.h>
62#endif
63
64#include <net/if.h>
65#include <net/if_var.h>
66#include <net/vnet.h>
67
68/*-
69 * This file implements core functions for virtual network stacks:
70 *
71 * - Virtual network stack management functions.
72 *
73 * - Virtual network stack memory allocator, which virtualizes global
74 *   variables in the network stack
75 *
76 * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
77 *   to register startup/shutdown events to be run for each virtual network
78 *   stack instance.
79 */
80
81FEATURE(vimage, "VIMAGE kernel virtualization");
82
83static MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
84
85/*
86 * The virtual network stack list has two read-write locks, one sleepable and
87 * the other not, so that the list can be stablized and walked in a variety
88 * of network stack contexts.  Both must be acquired exclusively to modify
89 * the list, but a read lock of either lock is sufficient to walk the list.
90 */
91struct rwlock		vnet_rwlock;
92struct sx		vnet_sxlock;
93
94#define	VNET_LIST_WLOCK() do {						\
95	sx_xlock(&vnet_sxlock);						\
96	rw_wlock(&vnet_rwlock);						\
97} while (0)
98
99#define	VNET_LIST_WUNLOCK() do {					\
100	rw_wunlock(&vnet_rwlock);					\
101	sx_xunlock(&vnet_sxlock);					\
102} while (0)
103
104struct vnet_list_head vnet_head;
105struct vnet *vnet0;
106
107/*
108 * The virtual network stack allocator provides storage for virtualized
109 * global variables.  These variables are defined/declared using the
110 * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet'
111 * linker set.  The details of the implementation are somewhat subtle, but
112 * allow the majority of most network subsystems to maintain
113 * virtualization-agnostic.
114 *
115 * The virtual network stack allocator handles variables in the base kernel
116 * vs. modules in similar but different ways.  In both cases, virtualized
117 * global variables are marked as such by being declared to be part of the
118 * vnet linker set.  These "master" copies of global variables serve two
119 * functions:
120 *
121 * (1) They contain static initialization or "default" values for global
122 *     variables which will be propagated to each virtual network stack
123 *     instance when created.  As with normal global variables, they default
124 *     to zero-filled.
125 *
126 * (2) They act as unique global names by which the variable can be referred
127 *     to, regardless of network stack instance.  The single global symbol
128 *     will be used to calculate the location of a per-virtual instance
129 *     variable at run-time.
130 *
131 * Each virtual network stack instance has a complete copy of each
132 * virtualized global variable, stored in a malloc'd block of memory
133 * referred to by vnet->vnet_data_mem.  Critical to the design is that each
134 * per-instance memory block is laid out identically to the master block so
135 * that the offset of each global variable is the same across all blocks.  To
136 * optimize run-time access, a precalculated 'base' address,
137 * vnet->vnet_data_base, is stored in each vnet, and is the amount that can
138 * be added to the address of a 'master' instance of a variable to get to the
139 * per-vnet instance.
140 *
141 * Virtualized global variables are handled in a similar manner, but as each
142 * module has its own 'set_vnet' linker set, and we want to keep all
143 * virtualized globals togther, we reserve space in the kernel's linker set
144 * for potential module variables using a per-vnet character array,
145 * 'modspace'.  The virtual network stack allocator maintains a free list to
146 * track what space in the array is free (all, initially) and as modules are
147 * linked, allocates portions of the space to specific globals.  The kernel
148 * module linker queries the virtual network stack allocator and will
149 * bind references of the global to the location during linking.  It also
150 * calls into the virtual network stack allocator, once the memory is
151 * initialized, in order to propagate the new static initializations to all
152 * existing virtual network stack instances so that the soon-to-be executing
153 * module will find every network stack instance with proper default values.
154 */
155
156/*
157 * Number of bytes of data in the 'set_vnet' linker set, and hence the total
158 * size of all kernel virtualized global variables, and the malloc(9) type
159 * that will be used to allocate it.
160 */
161#define	VNET_BYTES	(VNET_STOP - VNET_START)
162
163static MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
164
165/*
166 * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of
167 * global variables across all loaded modules.  As this actually sizes an
168 * array declared as a virtualized global variable in the kernel itself, and
169 * we want the virtualized global variable space to be page-sized, we may
170 * have more space than that in practice.
171 */
172#define	VNET_MODMIN	(8 * PAGE_SIZE)
173#define	VNET_SIZE	roundup2(VNET_BYTES, PAGE_SIZE)
174
175/*
176 * Space to store virtualized global variables from loadable kernel modules,
177 * and the free list to manage it.
178 */
179VNET_DEFINE_STATIC(char, modspace[VNET_MODMIN] __aligned(__alignof(void *)));
180
181/*
182 * A copy of the initial values of all virtualized global variables.
183 */
184static uintptr_t vnet_init_var;
185
186/*
187 * Global lists of subsystem constructor and destructors for vnets.  They are
188 * registered via VNET_SYSINIT() and VNET_SYSUNINIT().  Both lists are
189 * protected by the vnet_sysinit_sxlock global lock.
190 */
191static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
192	TAILQ_HEAD_INITIALIZER(vnet_constructors);
193static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
194	TAILQ_HEAD_INITIALIZER(vnet_destructors);
195
196struct sx		vnet_sysinit_sxlock;
197
198#define	VNET_SYSINIT_WLOCK()	sx_xlock(&vnet_sysinit_sxlock);
199#define	VNET_SYSINIT_WUNLOCK()	sx_xunlock(&vnet_sysinit_sxlock);
200#define	VNET_SYSINIT_RLOCK()	sx_slock(&vnet_sysinit_sxlock);
201#define	VNET_SYSINIT_RUNLOCK()	sx_sunlock(&vnet_sysinit_sxlock);
202
203struct vnet_data_free {
204	uintptr_t	vnd_start;
205	int		vnd_len;
206	TAILQ_ENTRY(vnet_data_free) vnd_link;
207};
208
209static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
210    "VNET resource accounting");
211static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
212	    TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
213static struct sx vnet_data_free_lock;
214
215SDT_PROVIDER_DEFINE(vnet);
216SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
217SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
218    "struct vnet *");
219SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
220    "int", "struct vnet *");
221SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
222    "int", "struct vnet *");
223SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
224    "int");
225
226/*
227 * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
228 */
229static void vnet_sysinit(void);
230static void vnet_sysuninit(void);
231
232#ifdef DDB
233static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
234#endif
235
236/*
237 * Allocate a virtual network stack.
238 */
239struct vnet *
240vnet_alloc(void)
241{
242	struct vnet *vnet;
243
244	SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
245	vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
246	vnet->vnet_magic_n = VNET_MAGIC_N;
247	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
248
249	/*
250	 * Allocate storage for virtualized global variables and copy in
251	 * initial values from our 'master' copy.
252	 */
253	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
254	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
255
256	/*
257	 * All use of vnet-specific data will immediately subtract VNET_START
258	 * from the base memory pointer, so pre-calculate that now to avoid
259	 * it on each use.
260	 */
261	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
262
263	/* Initialize / attach vnet module instances. */
264	CURVNET_SET_QUIET(vnet);
265	vnet_sysinit();
266	CURVNET_RESTORE();
267
268	VNET_LIST_WLOCK();
269	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
270	VNET_LIST_WUNLOCK();
271
272	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
273	return (vnet);
274}
275
276/*
277 * Destroy a virtual network stack.
278 */
279void
280vnet_destroy(struct vnet *vnet)
281{
282
283	SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
284	KASSERT(vnet->vnet_sockcnt == 0,
285	    ("%s: vnet still has sockets", __func__));
286
287	VNET_LIST_WLOCK();
288	LIST_REMOVE(vnet, vnet_le);
289	VNET_LIST_WUNLOCK();
290
291	/* Signal that VNET is being shutdown. */
292	vnet->vnet_shutdown = true;
293
294	CURVNET_SET_QUIET(vnet);
295	sx_xlock(&ifnet_detach_sxlock);
296	vnet_sysuninit();
297	sx_xunlock(&ifnet_detach_sxlock);
298	CURVNET_RESTORE();
299
300	/*
301	 * Release storage for the virtual network stack instance.
302	 */
303	free(vnet->vnet_data_mem, M_VNET_DATA);
304	vnet->vnet_data_mem = NULL;
305	vnet->vnet_data_base = 0;
306	vnet->vnet_magic_n = 0xdeadbeef;
307	free(vnet, M_VNET);
308	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
309}
310
311/*
312 * Boot time initialization and allocation of virtual network stacks.
313 */
314static void
315vnet_init_prelink(void *arg __unused)
316{
317
318	rw_init(&vnet_rwlock, "vnet_rwlock");
319	sx_init(&vnet_sxlock, "vnet_sxlock");
320	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
321	LIST_INIT(&vnet_head);
322}
323SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
324    vnet_init_prelink, NULL);
325
326static void
327vnet0_init(void *arg __unused)
328{
329
330	if (bootverbose)
331		printf("VIMAGE (virtualized network stack) enabled\n");
332
333	/*
334	 * We MUST clear curvnet in vi_init_done() before going SMP,
335	 * otherwise CURVNET_SET() macros would scream about unnecessary
336	 * curvnet recursions.
337	 */
338	curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
339}
340SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
341
342static void
343vnet_init_done(void *unused __unused)
344{
345
346	curvnet = NULL;
347}
348SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done,
349    NULL);
350
351/*
352 * Once on boot, initialize the modspace freelist to entirely cover modspace.
353 */
354static void
355vnet_data_startup(void *dummy __unused)
356{
357	struct vnet_data_free *df;
358
359	df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
360	df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
361	df->vnd_len = VNET_MODMIN;
362	TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
363	sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
364	vnet_init_var = (uintptr_t)malloc(VNET_BYTES, M_VNET_DATA, M_WAITOK);
365}
366SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL);
367
368/* Dummy VNET_SYSINIT to make sure we always reach the final end state. */
369static void
370vnet_sysinit_done(void *unused __unused)
371{
372
373	return;
374}
375VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY,
376    vnet_sysinit_done, NULL);
377
378/*
379 * When a module is loaded and requires storage for a virtualized global
380 * variable, allocate space from the modspace free list.  This interface
381 * should be used only by the kernel linker.
382 */
383void *
384vnet_data_alloc(int size)
385{
386	struct vnet_data_free *df;
387	void *s;
388
389	s = NULL;
390	size = roundup2(size, sizeof(void *));
391	sx_xlock(&vnet_data_free_lock);
392	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
393		if (df->vnd_len < size)
394			continue;
395		if (df->vnd_len == size) {
396			s = (void *)df->vnd_start;
397			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
398			free(df, M_VNET_DATA_FREE);
399			break;
400		}
401		s = (void *)df->vnd_start;
402		df->vnd_len -= size;
403		df->vnd_start = df->vnd_start + size;
404		break;
405	}
406	sx_xunlock(&vnet_data_free_lock);
407
408	return (s);
409}
410
411/*
412 * Free space for a virtualized global variable on module unload.
413 */
414void
415vnet_data_free(void *start_arg, int size)
416{
417	struct vnet_data_free *df;
418	struct vnet_data_free *dn;
419	uintptr_t start;
420	uintptr_t end;
421
422	size = roundup2(size, sizeof(void *));
423	start = (uintptr_t)start_arg;
424	end = start + size;
425	/*
426	 * Free a region of space and merge it with as many neighbors as
427	 * possible.  Keeping the list sorted simplifies this operation.
428	 */
429	sx_xlock(&vnet_data_free_lock);
430	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
431		if (df->vnd_start > end)
432			break;
433		/*
434		 * If we expand at the end of an entry we may have to merge
435		 * it with the one following it as well.
436		 */
437		if (df->vnd_start + df->vnd_len == start) {
438			df->vnd_len += size;
439			dn = TAILQ_NEXT(df, vnd_link);
440			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
441				df->vnd_len += dn->vnd_len;
442				TAILQ_REMOVE(&vnet_data_free_head, dn,
443				    vnd_link);
444				free(dn, M_VNET_DATA_FREE);
445			}
446			sx_xunlock(&vnet_data_free_lock);
447			return;
448		}
449		if (df->vnd_start == end) {
450			df->vnd_start = start;
451			df->vnd_len += size;
452			sx_xunlock(&vnet_data_free_lock);
453			return;
454		}
455	}
456	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
457	dn->vnd_start = start;
458	dn->vnd_len = size;
459	if (df)
460		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
461	else
462		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
463	sx_xunlock(&vnet_data_free_lock);
464}
465
466/*
467 * When a new virtualized global variable has been allocated, propagate its
468 * initial value to each already-allocated virtual network stack instance.
469 */
470void
471vnet_data_copy(void *start, int size)
472{
473	struct vnet *vnet;
474
475	VNET_LIST_RLOCK();
476	LIST_FOREACH(vnet, &vnet_head, vnet_le)
477		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
478		    (uintptr_t)start), start, size);
479	VNET_LIST_RUNLOCK();
480}
481
482/*
483 * Save a copy of the initial values of virtualized global variables.
484 */
485void
486vnet_save_init(void *start, size_t size)
487{
488	MPASS(vnet_init_var != 0);
489	MPASS(VNET_START <= (uintptr_t)start &&
490	    (uintptr_t)start + size <= VNET_STOP);
491	memcpy((void *)(vnet_init_var + ((uintptr_t)start - VNET_START)),
492	    start, size);
493}
494
495/*
496 * Restore the 'master' copies of virtualized global variables to theirs
497 * initial values.
498 */
499void
500vnet_restore_init(void *start, size_t size)
501{
502	MPASS(vnet_init_var != 0);
503	MPASS(VNET_START <= (uintptr_t)start &&
504	    (uintptr_t)start + size <= VNET_STOP);
505	memcpy(start,
506	    (void *)(vnet_init_var + ((uintptr_t)start - VNET_START)), size);
507}
508
509/*
510 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
511 * and VNET_SYSUNINIT().
512 */
513void
514vnet_register_sysinit(void *arg)
515{
516	struct vnet_sysinit *vs, *vs2;
517	struct vnet *vnet;
518
519	vs = arg;
520	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
521
522	/* Add the constructor to the global list of vnet constructors. */
523	VNET_SYSINIT_WLOCK();
524	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
525		if (vs2->subsystem > vs->subsystem)
526			break;
527		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
528			break;
529	}
530	if (vs2 != NULL)
531		TAILQ_INSERT_BEFORE(vs2, vs, link);
532	else
533		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
534
535	/*
536	 * Invoke the constructor on all the existing vnets when it is
537	 * registered.
538	 */
539	VNET_LIST_RLOCK();
540	VNET_FOREACH(vnet) {
541		CURVNET_SET_QUIET(vnet);
542		vs->func(vs->arg);
543		CURVNET_RESTORE();
544	}
545	VNET_LIST_RUNLOCK();
546	VNET_SYSINIT_WUNLOCK();
547}
548
549void
550vnet_deregister_sysinit(void *arg)
551{
552	struct vnet_sysinit *vs;
553
554	vs = arg;
555
556	/* Remove the constructor from the global list of vnet constructors. */
557	VNET_SYSINIT_WLOCK();
558	TAILQ_REMOVE(&vnet_constructors, vs, link);
559	VNET_SYSINIT_WUNLOCK();
560}
561
562void
563vnet_register_sysuninit(void *arg)
564{
565	struct vnet_sysinit *vs, *vs2;
566
567	vs = arg;
568
569	/* Add the destructor to the global list of vnet destructors. */
570	VNET_SYSINIT_WLOCK();
571	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
572		if (vs2->subsystem > vs->subsystem)
573			break;
574		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
575			break;
576	}
577	if (vs2 != NULL)
578		TAILQ_INSERT_BEFORE(vs2, vs, link);
579	else
580		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
581	VNET_SYSINIT_WUNLOCK();
582}
583
584void
585vnet_deregister_sysuninit(void *arg)
586{
587	struct vnet_sysinit *vs;
588	struct vnet *vnet;
589
590	vs = arg;
591
592	/*
593	 * Invoke the destructor on all the existing vnets when it is
594	 * deregistered.
595	 */
596	VNET_SYSINIT_WLOCK();
597	VNET_LIST_RLOCK();
598	VNET_FOREACH(vnet) {
599		CURVNET_SET_QUIET(vnet);
600		vs->func(vs->arg);
601		CURVNET_RESTORE();
602	}
603
604	/* Remove the destructor from the global list of vnet destructors. */
605	TAILQ_REMOVE(&vnet_destructors, vs, link);
606	VNET_SYSINIT_WUNLOCK();
607	VNET_LIST_RUNLOCK();
608}
609
610/*
611 * Invoke all registered vnet constructors on the current vnet.  Used during
612 * vnet construction.  The caller is responsible for ensuring the new vnet is
613 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
614 */
615static void
616vnet_sysinit(void)
617{
618	struct vnet_sysinit *vs;
619
620	VNET_SYSINIT_RLOCK();
621	TAILQ_FOREACH(vs, &vnet_constructors, link) {
622		curvnet->vnet_state = vs->subsystem;
623		vs->func(vs->arg);
624	}
625	VNET_SYSINIT_RUNLOCK();
626}
627
628/*
629 * Invoke all registered vnet destructors on the current vnet.  Used during
630 * vnet destruction.  The caller is responsible for ensuring the dying vnet
631 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
632 */
633static void
634vnet_sysuninit(void)
635{
636	struct vnet_sysinit *vs;
637
638	VNET_SYSINIT_RLOCK();
639	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
640	    link) {
641		curvnet->vnet_state = vs->subsystem;
642		vs->func(vs->arg);
643	}
644	VNET_SYSINIT_RUNLOCK();
645}
646
647/*
648 * EVENTHANDLER(9) extensions.
649 */
650/*
651 * Invoke the eventhandler function originally registered with the possibly
652 * registered argument for all virtual network stack instances.
653 *
654 * This iterator can only be used for eventhandlers that do not take any
655 * additional arguments, as we do ignore the variadic arguments from the
656 * EVENTHANDLER_INVOKE() call.
657 */
658void
659vnet_global_eventhandler_iterator_func(void *arg, ...)
660{
661	VNET_ITERATOR_DECL(vnet_iter);
662	struct eventhandler_entry_vimage *v_ee;
663
664	/*
665	 * There is a bug here in that we should actually cast things to
666	 * (struct eventhandler_entry_ ## name *)  but that's not easily
667	 * possible in here so just re-using the variadic version we
668	 * defined for the generic vimage case.
669	 */
670	v_ee = arg;
671	VNET_LIST_RLOCK();
672	VNET_FOREACH(vnet_iter) {
673		CURVNET_SET(vnet_iter);
674		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
675		CURVNET_RESTORE();
676	}
677	VNET_LIST_RUNLOCK();
678}
679
680#ifdef VNET_DEBUG
681struct vnet_recursion {
682	SLIST_ENTRY(vnet_recursion)	 vnr_le;
683	const char			*prev_fn;
684	const char			*where_fn;
685	int				 where_line;
686	struct vnet			*old_vnet;
687	struct vnet			*new_vnet;
688};
689
690static SLIST_HEAD(, vnet_recursion) vnet_recursions =
691    SLIST_HEAD_INITIALIZER(vnet_recursions);
692
693static void
694vnet_print_recursion(struct vnet_recursion *vnr, int brief)
695{
696
697	if (!brief)
698		printf("CURVNET_SET() recursion in ");
699	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
700	    vnr->prev_fn);
701	if (brief)
702		printf(", ");
703	else
704		printf("\n    ");
705	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
706}
707
708void
709vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
710{
711	struct vnet_recursion *vnr;
712
713	/* Skip already logged recursion events. */
714	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
715		if (vnr->prev_fn == old_fn &&
716		    vnr->where_fn == curthread->td_vnet_lpush &&
717		    vnr->where_line == line &&
718		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
719			return;
720
721	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
722	if (vnr == NULL)
723		panic("%s: malloc failed", __func__);
724	vnr->prev_fn = old_fn;
725	vnr->where_fn = curthread->td_vnet_lpush;
726	vnr->where_line = line;
727	vnr->old_vnet = old_vnet;
728	vnr->new_vnet = curvnet;
729
730	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
731
732	vnet_print_recursion(vnr, 0);
733#ifdef KDB
734	kdb_backtrace();
735#endif
736}
737#endif /* VNET_DEBUG */
738
739/*
740 * DDB(4).
741 */
742#ifdef DDB
743static void
744db_vnet_print(struct vnet *vnet)
745{
746
747	db_printf("vnet            = %p\n", vnet);
748	db_printf(" vnet_magic_n   = %#08x (%s, orig %#08x)\n",
749	    vnet->vnet_magic_n,
750	    (vnet->vnet_magic_n == VNET_MAGIC_N) ?
751		"ok" : "mismatch", VNET_MAGIC_N);
752	db_printf(" vnet_ifcnt     = %u\n", vnet->vnet_ifcnt);
753	db_printf(" vnet_sockcnt   = %u\n", vnet->vnet_sockcnt);
754	db_printf(" vnet_data_mem  = %p\n", vnet->vnet_data_mem);
755	db_printf(" vnet_data_base = %#jx\n",
756	    (uintmax_t)vnet->vnet_data_base);
757	db_printf(" vnet_state     = %#08x\n", vnet->vnet_state);
758	db_printf(" vnet_shutdown  = %#03x\n", vnet->vnet_shutdown);
759	db_printf("\n");
760}
761
762DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets)
763{
764	VNET_ITERATOR_DECL(vnet_iter);
765
766	VNET_FOREACH(vnet_iter) {
767		db_vnet_print(vnet_iter);
768		if (db_pager_quit)
769			break;
770	}
771}
772
773DB_SHOW_COMMAND(vnet, db_show_vnet)
774{
775
776	if (!have_addr) {
777		db_printf("usage: show vnet <struct vnet *>\n");
778		return;
779	}
780
781	db_vnet_print((struct vnet *)addr);
782}
783
784static void
785db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
786{
787	const char *vsname, *funcname;
788	c_db_sym_t sym;
789	db_expr_t  offset;
790
791#define xprint(...)							\
792	if (ddb)							\
793		db_printf(__VA_ARGS__);					\
794	else								\
795		printf(__VA_ARGS__)
796
797	if (vs == NULL) {
798		xprint("%s: no vnet_sysinit * given\n", __func__);
799		return;
800	}
801
802	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
803	db_symbol_values(sym, &vsname, NULL);
804	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
805	db_symbol_values(sym, &funcname, NULL);
806	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
807	xprint("  %#08x %#08x\n", vs->subsystem, vs->order);
808	xprint("  %p(%s)(%p)\n",
809	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
810#undef xprint
811}
812
813DB_SHOW_COMMAND_FLAGS(vnet_sysinit, db_show_vnet_sysinit, DB_CMD_MEMSAFE)
814{
815	struct vnet_sysinit *vs;
816
817	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
818	db_printf("  Subsystem  Order\n");
819	db_printf("  Function(Name)(Arg)\n");
820	TAILQ_FOREACH(vs, &vnet_constructors, link) {
821		db_show_vnet_print_vs(vs, 1);
822		if (db_pager_quit)
823			break;
824	}
825}
826
827DB_SHOW_COMMAND_FLAGS(vnet_sysuninit, db_show_vnet_sysuninit, DB_CMD_MEMSAFE)
828{
829	struct vnet_sysinit *vs;
830
831	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
832	db_printf("  Subsystem  Order\n");
833	db_printf("  Function(Name)(Arg)\n");
834	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
835	    link) {
836		db_show_vnet_print_vs(vs, 1);
837		if (db_pager_quit)
838			break;
839	}
840}
841
842#ifdef VNET_DEBUG
843DB_SHOW_COMMAND_FLAGS(vnetrcrs, db_show_vnetrcrs, DB_CMD_MEMSAFE)
844{
845	struct vnet_recursion *vnr;
846
847	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
848		vnet_print_recursion(vnr, 1);
849}
850#endif
851#endif /* DDB */
852