xref: /freebsd/sys/net/vnet.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
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 
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	8192
173 #define	VNET_SIZE	roundup2(VNET_BYTES, PAGE_SIZE)
174 #define	VNET_MODSIZE	(VNET_SIZE - (VNET_BYTES - VNET_MODMIN))
175 
176 /*
177  * Space to store virtualized global variables from loadable kernel modules,
178  * and the free list to manage it.
179  */
180 static VNET_DEFINE(char, modspace[VNET_MODMIN]);
181 
182 /*
183  * Global lists of subsystem constructor and destructors for vnets.  They are
184  * registered via VNET_SYSINIT() and VNET_SYSUNINIT().  Both lists are
185  * protected by the vnet_sysinit_sxlock global lock.
186  */
187 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
188 	TAILQ_HEAD_INITIALIZER(vnet_constructors);
189 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
190 	TAILQ_HEAD_INITIALIZER(vnet_destructors);
191 
192 struct sx		vnet_sysinit_sxlock;
193 
194 #define	VNET_SYSINIT_WLOCK()	sx_xlock(&vnet_sysinit_sxlock);
195 #define	VNET_SYSINIT_WUNLOCK()	sx_xunlock(&vnet_sysinit_sxlock);
196 #define	VNET_SYSINIT_RLOCK()	sx_slock(&vnet_sysinit_sxlock);
197 #define	VNET_SYSINIT_RUNLOCK()	sx_sunlock(&vnet_sysinit_sxlock);
198 
199 struct vnet_data_free {
200 	uintptr_t	vnd_start;
201 	int		vnd_len;
202 	TAILQ_ENTRY(vnet_data_free) vnd_link;
203 };
204 
205 static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
206     "VNET resource accounting");
207 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
208 	    TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
209 static struct sx vnet_data_free_lock;
210 
211 SDT_PROVIDER_DEFINE(vnet);
212 SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
213 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
214     "struct vnet *");
215 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
216     "int", "struct vnet *");
217 SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
218     "int", "struct vnet *");
219 SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
220     "int");
221 
222 #ifdef DDB
223 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
224 #endif
225 
226 /*
227  * Allocate a virtual network stack.
228  */
229 struct vnet *
230 vnet_alloc(void)
231 {
232 	struct vnet *vnet;
233 
234 	SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
235 	vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
236 	vnet->vnet_magic_n = VNET_MAGIC_N;
237 	SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
238 
239 	/*
240 	 * Allocate storage for virtualized global variables and copy in
241 	 * initial values form our 'master' copy.
242 	 */
243 	vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
244 	memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
245 
246 	/*
247 	 * All use of vnet-specific data will immediately subtract VNET_START
248 	 * from the base memory pointer, so pre-calculate that now to avoid
249 	 * it on each use.
250 	 */
251 	vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
252 
253 	/* Initialize / attach vnet module instances. */
254 	CURVNET_SET_QUIET(vnet);
255 	vnet_sysinit();
256 	CURVNET_RESTORE();
257 
258 	VNET_LIST_WLOCK();
259 	LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
260 	VNET_LIST_WUNLOCK();
261 
262 	SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
263 	return (vnet);
264 }
265 
266 /*
267  * Destroy a virtual network stack.
268  */
269 void
270 vnet_destroy(struct vnet *vnet)
271 {
272 	struct ifnet *ifp, *nifp;
273 
274 	SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
275 	KASSERT(vnet->vnet_sockcnt == 0,
276 	    ("%s: vnet still has sockets", __func__));
277 
278 	VNET_LIST_WLOCK();
279 	LIST_REMOVE(vnet, vnet_le);
280 	VNET_LIST_WUNLOCK();
281 
282 	CURVNET_SET_QUIET(vnet);
283 
284 	/* Return all inherited interfaces to their parent vnets. */
285 	TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
286 		if (ifp->if_home_vnet != ifp->if_vnet)
287 			if_vmove(ifp, ifp->if_home_vnet);
288 	}
289 
290 	vnet_sysuninit();
291 	CURVNET_RESTORE();
292 
293 	/*
294 	 * Release storage for the virtual network stack instance.
295 	 */
296 	free(vnet->vnet_data_mem, M_VNET_DATA);
297 	vnet->vnet_data_mem = NULL;
298 	vnet->vnet_data_base = 0;
299 	vnet->vnet_magic_n = 0xdeadbeef;
300 	free(vnet, M_VNET);
301 	SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
302 }
303 
304 /*
305  * Boot time initialization and allocation of virtual network stacks.
306  */
307 static void
308 vnet_init_prelink(void *arg)
309 {
310 
311 	rw_init(&vnet_rwlock, "vnet_rwlock");
312 	sx_init(&vnet_sxlock, "vnet_sxlock");
313 	sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
314 	LIST_INIT(&vnet_head);
315 }
316 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
317     vnet_init_prelink, NULL);
318 
319 static void
320 vnet0_init(void *arg)
321 {
322 
323 	/* Warn people before take off - in case we crash early. */
324 	printf("WARNING: VIMAGE (virtualized network stack) is a highly "
325 	    "experimental feature.\n");
326 
327 	/*
328 	 * We MUST clear curvnet in vi_init_done() before going SMP,
329 	 * otherwise CURVNET_SET() macros would scream about unnecessary
330 	 * curvnet recursions.
331 	 */
332 	curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
333 }
334 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
335 
336 static void
337 vnet_init_done(void *unused)
338 {
339 
340 	curvnet = NULL;
341 }
342 
343 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, 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, 0);
361 
362 /*
363  * When a module is loaded and requires storage for a virtualized global
364  * variable, allocate space from the modspace free list.  This interface
365  * should be used only by the kernel linker.
366  */
367 void *
368 vnet_data_alloc(int size)
369 {
370 	struct vnet_data_free *df;
371 	void *s;
372 
373 	s = NULL;
374 	size = roundup2(size, sizeof(void *));
375 	sx_xlock(&vnet_data_free_lock);
376 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
377 		if (df->vnd_len < size)
378 			continue;
379 		if (df->vnd_len == size) {
380 			s = (void *)df->vnd_start;
381 			TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
382 			free(df, M_VNET_DATA_FREE);
383 			break;
384 		}
385 		s = (void *)df->vnd_start;
386 		df->vnd_len -= size;
387 		df->vnd_start = df->vnd_start + size;
388 		break;
389 	}
390 	sx_xunlock(&vnet_data_free_lock);
391 
392 	return (s);
393 }
394 
395 /*
396  * Free space for a virtualized global variable on module unload.
397  */
398 void
399 vnet_data_free(void *start_arg, int size)
400 {
401 	struct vnet_data_free *df;
402 	struct vnet_data_free *dn;
403 	uintptr_t start;
404 	uintptr_t end;
405 
406 	size = roundup2(size, sizeof(void *));
407 	start = (uintptr_t)start_arg;
408 	end = start + size;
409 	/*
410 	 * Free a region of space and merge it with as many neighbors as
411 	 * possible.  Keeping the list sorted simplifies this operation.
412 	 */
413 	sx_xlock(&vnet_data_free_lock);
414 	TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
415 		if (df->vnd_start > end)
416 			break;
417 		/*
418 		 * If we expand at the end of an entry we may have to merge
419 		 * it with the one following it as well.
420 		 */
421 		if (df->vnd_start + df->vnd_len == start) {
422 			df->vnd_len += size;
423 			dn = TAILQ_NEXT(df, vnd_link);
424 			if (df->vnd_start + df->vnd_len == dn->vnd_start) {
425 				df->vnd_len += dn->vnd_len;
426 				TAILQ_REMOVE(&vnet_data_free_head, dn,
427 				    vnd_link);
428 				free(dn, M_VNET_DATA_FREE);
429 			}
430 			sx_xunlock(&vnet_data_free_lock);
431 			return;
432 		}
433 		if (df->vnd_start == end) {
434 			df->vnd_start = start;
435 			df->vnd_len += size;
436 			sx_xunlock(&vnet_data_free_lock);
437 			return;
438 		}
439 	}
440 	dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
441 	dn->vnd_start = start;
442 	dn->vnd_len = size;
443 	if (df)
444 		TAILQ_INSERT_BEFORE(df, dn, vnd_link);
445 	else
446 		TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
447 	sx_xunlock(&vnet_data_free_lock);
448 }
449 
450 /*
451  * When a new virtualized global variable has been allocated, propagate its
452  * initial value to each already-allocated virtual network stack instance.
453  */
454 void
455 vnet_data_copy(void *start, int size)
456 {
457 	struct vnet *vnet;
458 
459 	VNET_LIST_RLOCK();
460 	LIST_FOREACH(vnet, &vnet_head, vnet_le)
461 		memcpy((void *)((uintptr_t)vnet->vnet_data_base +
462 		    (uintptr_t)start), start, size);
463 	VNET_LIST_RUNLOCK();
464 }
465 
466 /*
467  * Support for special SYSINIT handlers registered via VNET_SYSINIT()
468  * and VNET_SYSUNINIT().
469  */
470 void
471 vnet_register_sysinit(void *arg)
472 {
473 	struct vnet_sysinit *vs, *vs2;
474 	struct vnet *vnet;
475 
476 	vs = arg;
477 	KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
478 
479 	/* Add the constructor to the global list of vnet constructors. */
480 	VNET_SYSINIT_WLOCK();
481 	TAILQ_FOREACH(vs2, &vnet_constructors, link) {
482 		if (vs2->subsystem > vs->subsystem)
483 			break;
484 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
485 			break;
486 	}
487 	if (vs2 != NULL)
488 		TAILQ_INSERT_BEFORE(vs2, vs, link);
489 	else
490 		TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
491 
492 	/*
493 	 * Invoke the constructor on all the existing vnets when it is
494 	 * registered.
495 	 */
496 	VNET_FOREACH(vnet) {
497 		CURVNET_SET_QUIET(vnet);
498 		vs->func(vs->arg);
499 		CURVNET_RESTORE();
500 	}
501 	VNET_SYSINIT_WUNLOCK();
502 }
503 
504 void
505 vnet_deregister_sysinit(void *arg)
506 {
507 	struct vnet_sysinit *vs;
508 
509 	vs = arg;
510 
511 	/* Remove the constructor from the global list of vnet constructors. */
512 	VNET_SYSINIT_WLOCK();
513 	TAILQ_REMOVE(&vnet_constructors, vs, link);
514 	VNET_SYSINIT_WUNLOCK();
515 }
516 
517 void
518 vnet_register_sysuninit(void *arg)
519 {
520 	struct vnet_sysinit *vs, *vs2;
521 
522 	vs = arg;
523 
524 	/* Add the destructor to the global list of vnet destructors. */
525 	VNET_SYSINIT_WLOCK();
526 	TAILQ_FOREACH(vs2, &vnet_destructors, link) {
527 		if (vs2->subsystem > vs->subsystem)
528 			break;
529 		if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
530 			break;
531 	}
532 	if (vs2 != NULL)
533 		TAILQ_INSERT_BEFORE(vs2, vs, link);
534 	else
535 		TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
536 	VNET_SYSINIT_WUNLOCK();
537 }
538 
539 void
540 vnet_deregister_sysuninit(void *arg)
541 {
542 	struct vnet_sysinit *vs;
543 	struct vnet *vnet;
544 
545 	vs = arg;
546 
547 	/*
548 	 * Invoke the destructor on all the existing vnets when it is
549 	 * deregistered.
550 	 */
551 	VNET_SYSINIT_WLOCK();
552 	VNET_FOREACH(vnet) {
553 		CURVNET_SET_QUIET(vnet);
554 		vs->func(vs->arg);
555 		CURVNET_RESTORE();
556 	}
557 
558 	/* Remove the destructor from the global list of vnet destructors. */
559 	TAILQ_REMOVE(&vnet_destructors, vs, link);
560 	VNET_SYSINIT_WUNLOCK();
561 }
562 
563 /*
564  * Invoke all registered vnet constructors on the current vnet.  Used during
565  * vnet construction.  The caller is responsible for ensuring the new vnet is
566  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
567  */
568 void
569 vnet_sysinit(void)
570 {
571 	struct vnet_sysinit *vs;
572 
573 	VNET_SYSINIT_RLOCK();
574 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
575 		vs->func(vs->arg);
576 	}
577 	VNET_SYSINIT_RUNLOCK();
578 }
579 
580 /*
581  * Invoke all registered vnet destructors on the current vnet.  Used during
582  * vnet destruction.  The caller is responsible for ensuring the dying vnet
583  * the current vnet and that the vnet_sysinit_sxlock lock is locked.
584  */
585 void
586 vnet_sysuninit(void)
587 {
588 	struct vnet_sysinit *vs;
589 
590 	VNET_SYSINIT_RLOCK();
591 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
592 	    link) {
593 		vs->func(vs->arg);
594 	}
595 	VNET_SYSINIT_RUNLOCK();
596 }
597 
598 /*
599  * EVENTHANDLER(9) extensions.
600  */
601 /*
602  * Invoke the eventhandler function originally registered with the possibly
603  * registered argument for all virtual network stack instances.
604  *
605  * This iterator can only be used for eventhandlers that do not take any
606  * additional arguments, as we do ignore the variadic arguments from the
607  * EVENTHANDLER_INVOKE() call.
608  */
609 void
610 vnet_global_eventhandler_iterator_func(void *arg, ...)
611 {
612 	VNET_ITERATOR_DECL(vnet_iter);
613 	struct eventhandler_entry_vimage *v_ee;
614 
615 	/*
616 	 * There is a bug here in that we should actually cast things to
617 	 * (struct eventhandler_entry_ ## name *)  but that's not easily
618 	 * possible in here so just re-using the variadic version we
619 	 * defined for the generic vimage case.
620 	 */
621 	v_ee = arg;
622 	VNET_LIST_RLOCK();
623 	VNET_FOREACH(vnet_iter) {
624 		CURVNET_SET(vnet_iter);
625 		((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
626 		CURVNET_RESTORE();
627 	}
628 	VNET_LIST_RUNLOCK();
629 }
630 
631 #ifdef VNET_DEBUG
632 struct vnet_recursion {
633 	SLIST_ENTRY(vnet_recursion)	 vnr_le;
634 	const char			*prev_fn;
635 	const char			*where_fn;
636 	int				 where_line;
637 	struct vnet			*old_vnet;
638 	struct vnet			*new_vnet;
639 };
640 
641 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
642     SLIST_HEAD_INITIALIZER(vnet_recursions);
643 
644 static void
645 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
646 {
647 
648 	if (!brief)
649 		printf("CURVNET_SET() recursion in ");
650 	printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
651 	    vnr->prev_fn);
652 	if (brief)
653 		printf(", ");
654 	else
655 		printf("\n    ");
656 	printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
657 }
658 
659 void
660 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
661 {
662 	struct vnet_recursion *vnr;
663 
664 	/* Skip already logged recursion events. */
665 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
666 		if (vnr->prev_fn == old_fn &&
667 		    vnr->where_fn == curthread->td_vnet_lpush &&
668 		    vnr->where_line == line &&
669 		    (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
670 			return;
671 
672 	vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
673 	if (vnr == NULL)
674 		panic("%s: malloc failed", __func__);
675 	vnr->prev_fn = old_fn;
676 	vnr->where_fn = curthread->td_vnet_lpush;
677 	vnr->where_line = line;
678 	vnr->old_vnet = old_vnet;
679 	vnr->new_vnet = curvnet;
680 
681 	SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
682 
683 	vnet_print_recursion(vnr, 0);
684 #ifdef KDB
685 	kdb_backtrace();
686 #endif
687 }
688 #endif /* VNET_DEBUG */
689 
690 /*
691  * DDB(4).
692  */
693 #ifdef DDB
694 DB_SHOW_COMMAND(vnets, db_show_vnets)
695 {
696 	VNET_ITERATOR_DECL(vnet_iter);
697 
698 	VNET_FOREACH(vnet_iter) {
699 		db_printf("vnet            = %p\n", vnet_iter);
700 		db_printf(" vnet_magic_n   = 0x%x (%s, orig 0x%x)\n",
701 		    vnet_iter->vnet_magic_n,
702 		    (vnet_iter->vnet_magic_n == VNET_MAGIC_N) ?
703 			"ok" : "mismatch", VNET_MAGIC_N);
704 		db_printf(" vnet_ifcnt     = %u\n", vnet_iter->vnet_ifcnt);
705 		db_printf(" vnet_sockcnt   = %u\n", vnet_iter->vnet_sockcnt);
706 		db_printf(" vnet_data_mem  = %p\n", vnet_iter->vnet_data_mem);
707 		db_printf(" vnet_data_base = 0x%jx\n",
708 		    (uintmax_t)vnet_iter->vnet_data_base);
709 		db_printf("\n");
710 		if (db_pager_quit)
711 			break;
712 	}
713 }
714 
715 static void
716 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
717 {
718 	const char *vsname, *funcname;
719 	c_db_sym_t sym;
720 	db_expr_t  offset;
721 
722 #define xprint(...)							\
723 	if (ddb)							\
724 		db_printf(__VA_ARGS__);					\
725 	else								\
726 		printf(__VA_ARGS__)
727 
728 	if (vs == NULL) {
729 		xprint("%s: no vnet_sysinit * given\n", __func__);
730 		return;
731 	}
732 
733 	sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
734 	db_symbol_values(sym, &vsname, NULL);
735 	sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
736 	db_symbol_values(sym, &funcname, NULL);
737 	xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
738 	xprint("  0x%08x 0x%08x\n", vs->subsystem, vs->order);
739 	xprint("  %p(%s)(%p)\n",
740 	    vs->func, (funcname != NULL) ? funcname : "", vs->arg);
741 #undef xprint
742 }
743 
744 DB_SHOW_COMMAND(vnet_sysinit, db_show_vnet_sysinit)
745 {
746 	struct vnet_sysinit *vs;
747 
748 	db_printf("VNET_SYSINIT vs Name(Ptr)\n");
749 	db_printf("  Subsystem  Order\n");
750 	db_printf("  Function(Name)(Arg)\n");
751 	TAILQ_FOREACH(vs, &vnet_constructors, link) {
752 		db_show_vnet_print_vs(vs, 1);
753 		if (db_pager_quit)
754 			break;
755 	}
756 }
757 
758 DB_SHOW_COMMAND(vnet_sysuninit, db_show_vnet_sysuninit)
759 {
760 	struct vnet_sysinit *vs;
761 
762 	db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
763 	db_printf("  Subsystem  Order\n");
764 	db_printf("  Function(Name)(Arg)\n");
765 	TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
766 	    link) {
767 		db_show_vnet_print_vs(vs, 1);
768 		if (db_pager_quit)
769 			break;
770 	}
771 }
772 
773 #ifdef VNET_DEBUG
774 DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs)
775 {
776 	struct vnet_recursion *vnr;
777 
778 	SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
779 		vnet_print_recursion(vnr, 1);
780 }
781 #endif
782 #endif /* DDB */
783