xref: /freebsd/sys/net/vnet.c (revision 315ee00f)
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 
81 FEATURE(vimage, "VIMAGE kernel virtualization");
82 
83 static 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  */
91 struct rwlock		vnet_rwlock;
92 struct 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 
104 struct vnet_list_head vnet_head;
105 struct 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 
163 static 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  */
179 VNET_DEFINE_STATIC(char, modspace[VNET_MODMIN] __aligned(__alignof(void *)));
180 
181 /*
182  * Global lists of subsystem constructor and destructors for vnets.  They are
183  * registered via VNET_SYSINIT() and VNET_SYSUNINIT().  Both lists are
184  * protected by the vnet_sysinit_sxlock global lock.
185  */
186 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
187 	TAILQ_HEAD_INITIALIZER(vnet_constructors);
188 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
189 	TAILQ_HEAD_INITIALIZER(vnet_destructors);
190 
191 struct sx		vnet_sysinit_sxlock;
192 
193 #define	VNET_SYSINIT_WLOCK()	sx_xlock(&vnet_sysinit_sxlock);
194 #define	VNET_SYSINIT_WUNLOCK()	sx_xunlock(&vnet_sysinit_sxlock);
195 #define	VNET_SYSINIT_RLOCK()	sx_slock(&vnet_sysinit_sxlock);
196 #define	VNET_SYSINIT_RUNLOCK()	sx_sunlock(&vnet_sysinit_sxlock);
197 
198 struct vnet_data_free {
199 	uintptr_t	vnd_start;
200 	int		vnd_len;
201 	TAILQ_ENTRY(vnet_data_free) vnd_link;
202 };
203 
204 static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
205     "VNET resource accounting");
206 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
207 	    TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
208 static struct sx vnet_data_free_lock;
209 
210 SDT_PROVIDER_DEFINE(vnet);
211 SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
212 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
213     "struct vnet *");
214 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
215     "int", "struct vnet *");
216 SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
217     "int", "struct vnet *");
218 SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
219     "int");
220 
221 /*
222  * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
223  */
224 static void vnet_sysinit(void);
225 static void vnet_sysuninit(void);
226 
227 #ifdef DDB
228 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
229 #endif
230 
231 /*
232  * Allocate a virtual network stack.
233  */
234 struct vnet *
235 vnet_alloc(void)
236 {
237 	struct vnet *vnet;
238 
239 	SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
240 	vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
241 	vnet->vnet_magic_n = VNET_MAGIC_N;
242 	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
243 
244 	/*
245 	 * Allocate storage for virtualized global variables and copy in
246 	 * initial values from our 'master' copy.
247 	 */
248 	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
249 	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
250 
251 	/*
252 	 * All use of vnet-specific data will immediately subtract VNET_START
253 	 * from the base memory pointer, so pre-calculate that now to avoid
254 	 * it on each use.
255 	 */
256 	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
257 
258 	/* Initialize / attach vnet module instances. */
259 	CURVNET_SET_QUIET(vnet);
260 	vnet_sysinit();
261 	CURVNET_RESTORE();
262 
263 	VNET_LIST_WLOCK();
264 	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
265 	VNET_LIST_WUNLOCK();
266 
267 	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
268 	return (vnet);
269 }
270 
271 /*
272  * Destroy a virtual network stack.
273  */
274 void
275 vnet_destroy(struct vnet *vnet)
276 {
277 
278 	SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
279 	KASSERT(vnet->vnet_sockcnt == 0,
280 	    ("%s: vnet still has sockets", __func__));
281 
282 	VNET_LIST_WLOCK();
283 	LIST_REMOVE(vnet, vnet_le);
284 	VNET_LIST_WUNLOCK();
285 
286 	/* Signal that VNET is being shutdown. */
287 	vnet->vnet_shutdown = true;
288 
289 	CURVNET_SET_QUIET(vnet);
290 	sx_xlock(&ifnet_detach_sxlock);
291 	vnet_sysuninit();
292 	sx_xunlock(&ifnet_detach_sxlock);
293 	CURVNET_RESTORE();
294 
295 	/*
296 	 * Release storage for the virtual network stack instance.
297 	 */
298 	free(vnet->vnet_data_mem, M_VNET_DATA);
299 	vnet->vnet_data_mem = NULL;
300 	vnet->vnet_data_base = 0;
301 	vnet->vnet_magic_n = 0xdeadbeef;
302 	free(vnet, M_VNET);
303 	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
304 }
305 
306 /*
307  * Boot time initialization and allocation of virtual network stacks.
308  */
309 static void
310 vnet_init_prelink(void *arg __unused)
311 {
312 
313 	rw_init(&vnet_rwlock, "vnet_rwlock");
314 	sx_init(&vnet_sxlock, "vnet_sxlock");
315 	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
316 	LIST_INIT(&vnet_head);
317 }
318 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
319     vnet_init_prelink, NULL);
320 
321 static void
322 vnet0_init(void *arg __unused)
323 {
324 
325 	if (bootverbose)
326 		printf("VIMAGE (virtualized network stack) enabled\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 }
335 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
336 
337 static void
338 vnet_init_done(void *unused __unused)
339 {
340 
341 	curvnet = NULL;
342 }
343 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done,
344     NULL);
345 
346 /*
347  * Once on boot, initialize the modspace freelist to entirely cover modspace.
348  */
349 static void
350 vnet_data_startup(void *dummy __unused)
351 {
352 	struct vnet_data_free *df;
353 
354 	df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
355 	df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
356 	df->vnd_len = VNET_MODMIN;
357 	TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
358 	sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
359 }
360 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL);
361 
362 /* Dummy VNET_SYSINIT to make sure we always reach the final end state. */
363 static void
364 vnet_sysinit_done(void *unused __unused)
365 {
366 
367 	return;
368 }
369 VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY,
370     vnet_sysinit_done, NULL);
371 
372 /*
373  * When a module is loaded and requires storage for a virtualized global
374  * variable, allocate space from the modspace free list.  This interface
375  * should be used only by the kernel linker.
376  */
377 void *
378 vnet_data_alloc(int size)
379 {
380 	struct vnet_data_free *df;
381 	void *s;
382 
383 	s = NULL;
384 	size = roundup2(size, sizeof(void *));
385 	sx_xlock(&vnet_data_free_lock);
386 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
387 		if (df->vnd_len < size)
388 			continue;
389 		if (df->vnd_len == size) {
390 			s = (void *)df->vnd_start;
391 			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
392 			free(df, M_VNET_DATA_FREE);
393 			break;
394 		}
395 		s = (void *)df->vnd_start;
396 		df->vnd_len -= size;
397 		df->vnd_start = df->vnd_start + size;
398 		break;
399 	}
400 	sx_xunlock(&vnet_data_free_lock);
401 
402 	return (s);
403 }
404 
405 /*
406  * Free space for a virtualized global variable on module unload.
407  */
408 void
409 vnet_data_free(void *start_arg, int size)
410 {
411 	struct vnet_data_free *df;
412 	struct vnet_data_free *dn;
413 	uintptr_t start;
414 	uintptr_t end;
415 
416 	size = roundup2(size, sizeof(void *));
417 	start = (uintptr_t)start_arg;
418 	end = start + size;
419 	/*
420 	 * Free a region of space and merge it with as many neighbors as
421 	 * possible.  Keeping the list sorted simplifies this operation.
422 	 */
423 	sx_xlock(&vnet_data_free_lock);
424 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
425 		if (df->vnd_start > end)
426 			break;
427 		/*
428 		 * If we expand at the end of an entry we may have to merge
429 		 * it with the one following it as well.
430 		 */
431 		if (df->vnd_start + df->vnd_len == start) {
432 			df->vnd_len += size;
433 			dn = TAILQ_NEXT(df, vnd_link);
434 			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
435 				df->vnd_len += dn->vnd_len;
436 				TAILQ_REMOVE(&vnet_data_free_head, dn,
437 				    vnd_link);
438 				free(dn, M_VNET_DATA_FREE);
439 			}
440 			sx_xunlock(&vnet_data_free_lock);
441 			return;
442 		}
443 		if (df->vnd_start == end) {
444 			df->vnd_start = start;
445 			df->vnd_len += size;
446 			sx_xunlock(&vnet_data_free_lock);
447 			return;
448 		}
449 	}
450 	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
451 	dn->vnd_start = start;
452 	dn->vnd_len = size;
453 	if (df)
454 		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
455 	else
456 		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
457 	sx_xunlock(&vnet_data_free_lock);
458 }
459 
460 /*
461  * When a new virtualized global variable has been allocated, propagate its
462  * initial value to each already-allocated virtual network stack instance.
463  */
464 void
465 vnet_data_copy(void *start, int size)
466 {
467 	struct vnet *vnet;
468 
469 	VNET_LIST_RLOCK();
470 	LIST_FOREACH(vnet, &vnet_head, vnet_le)
471 		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
472 		    (uintptr_t)start), start, size);
473 	VNET_LIST_RUNLOCK();
474 }
475 
476 /*
477  * Support for special SYSINIT handlers registered via VNET_SYSINIT()
478  * and VNET_SYSUNINIT().
479  */
480 void
481 vnet_register_sysinit(void *arg)
482 {
483 	struct vnet_sysinit *vs, *vs2;
484 	struct vnet *vnet;
485 
486 	vs = arg;
487 	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
488 
489 	/* Add the constructor to the global list of vnet constructors. */
490 	VNET_SYSINIT_WLOCK();
491 	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
492 		if (vs2->subsystem > vs->subsystem)
493 			break;
494 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
495 			break;
496 	}
497 	if (vs2 != NULL)
498 		TAILQ_INSERT_BEFORE(vs2, vs, link);
499 	else
500 		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
501 
502 	/*
503 	 * Invoke the constructor on all the existing vnets when it is
504 	 * registered.
505 	 */
506 	VNET_FOREACH(vnet) {
507 		CURVNET_SET_QUIET(vnet);
508 		vs->func(vs->arg);
509 		CURVNET_RESTORE();
510 	}
511 	VNET_SYSINIT_WUNLOCK();
512 }
513 
514 void
515 vnet_deregister_sysinit(void *arg)
516 {
517 	struct vnet_sysinit *vs;
518 
519 	vs = arg;
520 
521 	/* Remove the constructor from the global list of vnet constructors. */
522 	VNET_SYSINIT_WLOCK();
523 	TAILQ_REMOVE(&vnet_constructors, vs, link);
524 	VNET_SYSINIT_WUNLOCK();
525 }
526 
527 void
528 vnet_register_sysuninit(void *arg)
529 {
530 	struct vnet_sysinit *vs, *vs2;
531 
532 	vs = arg;
533 
534 	/* Add the destructor to the global list of vnet destructors. */
535 	VNET_SYSINIT_WLOCK();
536 	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
537 		if (vs2->subsystem > vs->subsystem)
538 			break;
539 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
540 			break;
541 	}
542 	if (vs2 != NULL)
543 		TAILQ_INSERT_BEFORE(vs2, vs, link);
544 	else
545 		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
546 	VNET_SYSINIT_WUNLOCK();
547 }
548 
549 void
550 vnet_deregister_sysuninit(void *arg)
551 {
552 	struct vnet_sysinit *vs;
553 	struct vnet *vnet;
554 
555 	vs = arg;
556 
557 	/*
558 	 * Invoke the destructor on all the existing vnets when it is
559 	 * deregistered.
560 	 */
561 	VNET_SYSINIT_WLOCK();
562 	VNET_FOREACH(vnet) {
563 		CURVNET_SET_QUIET(vnet);
564 		vs->func(vs->arg);
565 		CURVNET_RESTORE();
566 	}
567 
568 	/* Remove the destructor from the global list of vnet destructors. */
569 	TAILQ_REMOVE(&vnet_destructors, vs, link);
570 	VNET_SYSINIT_WUNLOCK();
571 }
572 
573 /*
574  * Invoke all registered vnet constructors on the current vnet.  Used during
575  * vnet construction.  The caller is responsible for ensuring the new vnet is
576  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
577  */
578 static void
579 vnet_sysinit(void)
580 {
581 	struct vnet_sysinit *vs;
582 
583 	VNET_SYSINIT_RLOCK();
584 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
585 		curvnet->vnet_state = vs->subsystem;
586 		vs->func(vs->arg);
587 	}
588 	VNET_SYSINIT_RUNLOCK();
589 }
590 
591 /*
592  * Invoke all registered vnet destructors on the current vnet.  Used during
593  * vnet destruction.  The caller is responsible for ensuring the dying vnet
594  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
595  */
596 static void
597 vnet_sysuninit(void)
598 {
599 	struct vnet_sysinit *vs;
600 
601 	VNET_SYSINIT_RLOCK();
602 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
603 	    link) {
604 		curvnet->vnet_state = vs->subsystem;
605 		vs->func(vs->arg);
606 	}
607 	VNET_SYSINIT_RUNLOCK();
608 }
609 
610 /*
611  * EVENTHANDLER(9) extensions.
612  */
613 /*
614  * Invoke the eventhandler function originally registered with the possibly
615  * registered argument for all virtual network stack instances.
616  *
617  * This iterator can only be used for eventhandlers that do not take any
618  * additional arguments, as we do ignore the variadic arguments from the
619  * EVENTHANDLER_INVOKE() call.
620  */
621 void
622 vnet_global_eventhandler_iterator_func(void *arg, ...)
623 {
624 	VNET_ITERATOR_DECL(vnet_iter);
625 	struct eventhandler_entry_vimage *v_ee;
626 
627 	/*
628 	 * There is a bug here in that we should actually cast things to
629 	 * (struct eventhandler_entry_ ## name *)  but that's not easily
630 	 * possible in here so just re-using the variadic version we
631 	 * defined for the generic vimage case.
632 	 */
633 	v_ee = arg;
634 	VNET_LIST_RLOCK();
635 	VNET_FOREACH(vnet_iter) {
636 		CURVNET_SET(vnet_iter);
637 		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
638 		CURVNET_RESTORE();
639 	}
640 	VNET_LIST_RUNLOCK();
641 }
642 
643 #ifdef VNET_DEBUG
644 struct vnet_recursion {
645 	SLIST_ENTRY(vnet_recursion)	 vnr_le;
646 	const char			*prev_fn;
647 	const char			*where_fn;
648 	int				 where_line;
649 	struct vnet			*old_vnet;
650 	struct vnet			*new_vnet;
651 };
652 
653 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
654     SLIST_HEAD_INITIALIZER(vnet_recursions);
655 
656 static void
657 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
658 {
659 
660 	if (!brief)
661 		printf("CURVNET_SET() recursion in ");
662 	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
663 	    vnr->prev_fn);
664 	if (brief)
665 		printf(", ");
666 	else
667 		printf("\n    ");
668 	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
669 }
670 
671 void
672 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
673 {
674 	struct vnet_recursion *vnr;
675 
676 	/* Skip already logged recursion events. */
677 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
678 		if (vnr->prev_fn == old_fn &&
679 		    vnr->where_fn == curthread->td_vnet_lpush &&
680 		    vnr->where_line == line &&
681 		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
682 			return;
683 
684 	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
685 	if (vnr == NULL)
686 		panic("%s: malloc failed", __func__);
687 	vnr->prev_fn = old_fn;
688 	vnr->where_fn = curthread->td_vnet_lpush;
689 	vnr->where_line = line;
690 	vnr->old_vnet = old_vnet;
691 	vnr->new_vnet = curvnet;
692 
693 	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
694 
695 	vnet_print_recursion(vnr, 0);
696 #ifdef KDB
697 	kdb_backtrace();
698 #endif
699 }
700 #endif /* VNET_DEBUG */
701 
702 /*
703  * DDB(4).
704  */
705 #ifdef DDB
706 static void
707 db_vnet_print(struct vnet *vnet)
708 {
709 
710 	db_printf("vnet            = %p\n", vnet);
711 	db_printf(" vnet_magic_n   = %#08x (%s, orig %#08x)\n",
712 	    vnet->vnet_magic_n,
713 	    (vnet->vnet_magic_n == VNET_MAGIC_N) ?
714 		"ok" : "mismatch", VNET_MAGIC_N);
715 	db_printf(" vnet_ifcnt     = %u\n", vnet->vnet_ifcnt);
716 	db_printf(" vnet_sockcnt   = %u\n", vnet->vnet_sockcnt);
717 	db_printf(" vnet_data_mem  = %p\n", vnet->vnet_data_mem);
718 	db_printf(" vnet_data_base = %#jx\n",
719 	    (uintmax_t)vnet->vnet_data_base);
720 	db_printf(" vnet_state     = %#08x\n", vnet->vnet_state);
721 	db_printf(" vnet_shutdown  = %#03x\n", vnet->vnet_shutdown);
722 	db_printf("\n");
723 }
724 
725 DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets)
726 {
727 	VNET_ITERATOR_DECL(vnet_iter);
728 
729 	VNET_FOREACH(vnet_iter) {
730 		db_vnet_print(vnet_iter);
731 		if (db_pager_quit)
732 			break;
733 	}
734 }
735 
736 DB_SHOW_COMMAND(vnet, db_show_vnet)
737 {
738 
739 	if (!have_addr) {
740 		db_printf("usage: show vnet <struct vnet *>\n");
741 		return;
742 	}
743 
744 	db_vnet_print((struct vnet *)addr);
745 }
746 
747 static void
748 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
749 {
750 	const char *vsname, *funcname;
751 	c_db_sym_t sym;
752 	db_expr_t  offset;
753 
754 #define xprint(...)							\
755 	if (ddb)							\
756 		db_printf(__VA_ARGS__);					\
757 	else								\
758 		printf(__VA_ARGS__)
759 
760 	if (vs == NULL) {
761 		xprint("%s: no vnet_sysinit * given\n", __func__);
762 		return;
763 	}
764 
765 	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
766 	db_symbol_values(sym, &vsname, NULL);
767 	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
768 	db_symbol_values(sym, &funcname, NULL);
769 	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
770 	xprint("  %#08x %#08x\n", vs->subsystem, vs->order);
771 	xprint("  %p(%s)(%p)\n",
772 	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
773 #undef xprint
774 }
775 
776 DB_SHOW_COMMAND_FLAGS(vnet_sysinit, db_show_vnet_sysinit, DB_CMD_MEMSAFE)
777 {
778 	struct vnet_sysinit *vs;
779 
780 	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
781 	db_printf("  Subsystem  Order\n");
782 	db_printf("  Function(Name)(Arg)\n");
783 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
784 		db_show_vnet_print_vs(vs, 1);
785 		if (db_pager_quit)
786 			break;
787 	}
788 }
789 
790 DB_SHOW_COMMAND_FLAGS(vnet_sysuninit, db_show_vnet_sysuninit, DB_CMD_MEMSAFE)
791 {
792 	struct vnet_sysinit *vs;
793 
794 	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
795 	db_printf("  Subsystem  Order\n");
796 	db_printf("  Function(Name)(Arg)\n");
797 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
798 	    link) {
799 		db_show_vnet_print_vs(vs, 1);
800 		if (db_pager_quit)
801 			break;
802 	}
803 }
804 
805 #ifdef VNET_DEBUG
806 DB_SHOW_COMMAND_FLAGS(vnetrcrs, db_show_vnetrcrs, DB_CMD_MEMSAFE)
807 {
808 	struct vnet_recursion *vnr;
809 
810 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
811 		vnet_print_recursion(vnr, 1);
812 }
813 #endif
814 #endif /* DDB */
815