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 = LIST_HEAD_INITIALIZER(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 * A copy of the initial values of all virtualized global variables.
183 */
184 static 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 */
191 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
192 TAILQ_HEAD_INITIALIZER(vnet_constructors);
193 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
194 TAILQ_HEAD_INITIALIZER(vnet_destructors);
195
196 struct 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
203 struct vnet_data_free {
204 uintptr_t vnd_start;
205 int vnd_len;
206 TAILQ_ENTRY(vnet_data_free) vnd_link;
207 };
208
209 static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free",
210 "VNET resource accounting");
211 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
212 TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
213 static struct sx vnet_data_free_lock;
214
215 SDT_PROVIDER_DEFINE(vnet);
216 SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int");
217 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int",
218 "struct vnet *");
219 SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return,
220 "int", "struct vnet *");
221 SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry,
222 "int", "struct vnet *");
223 SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return,
224 "int");
225
226 /*
227 * Run per-vnet sysinits or sysuninits during vnet creation/destruction.
228 */
229 static void vnet_sysinit(void);
230 static void vnet_sysuninit(void);
231
232 #ifdef DDB
233 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
234 #endif
235
236 /*
237 * Allocate a virtual network stack.
238 */
239 struct vnet *
vnet_alloc(void)240 vnet_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 */
279 void
vnet_destroy(struct vnet * vnet)280 vnet_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 */
314 static void
vnet_init_prelink(void * arg __unused)315 vnet_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 }
322 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
323 vnet_init_prelink, NULL);
324
325 static void
vnet0_init(void * arg __unused)326 vnet0_init(void *arg __unused)
327 {
328
329 if (bootverbose)
330 printf("VIMAGE (virtualized network stack) enabled\n");
331
332 /*
333 * We MUST clear curvnet in vi_init_done() before going SMP,
334 * otherwise CURVNET_SET() macros would scream about unnecessary
335 * curvnet recursions.
336 */
337 curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
338 }
339 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
340
341 static void
vnet_init_done(void * unused __unused)342 vnet_init_done(void *unused __unused)
343 {
344
345 curvnet = NULL;
346 }
347 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done,
348 NULL);
349
350 /*
351 * Once on boot, initialize the modspace freelist to entirely cover modspace.
352 */
353 static void
vnet_data_startup(void * dummy __unused)354 vnet_data_startup(void *dummy __unused)
355 {
356 struct vnet_data_free *df;
357
358 df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
359 df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
360 df->vnd_len = VNET_MODMIN;
361 TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
362 sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
363 vnet_init_var = (uintptr_t)malloc(VNET_BYTES, M_VNET_DATA, M_WAITOK);
364 }
365 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL);
366
367 /* Dummy VNET_SYSINIT to make sure we always reach the final end state. */
368 static void
vnet_sysinit_done(void * unused __unused)369 vnet_sysinit_done(void *unused __unused)
370 {
371
372 return;
373 }
374 VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY,
375 vnet_sysinit_done, NULL);
376
377 /*
378 * When a module is loaded and requires storage for a virtualized global
379 * variable, allocate space from the modspace free list. This interface
380 * should be used only by the kernel linker.
381 */
382 void *
vnet_data_alloc(int size)383 vnet_data_alloc(int size)
384 {
385 struct vnet_data_free *df;
386 void *s;
387
388 s = NULL;
389 size = roundup2(size, sizeof(void *));
390 sx_xlock(&vnet_data_free_lock);
391 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
392 if (df->vnd_len < size)
393 continue;
394 if (df->vnd_len == size) {
395 s = (void *)df->vnd_start;
396 TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
397 free(df, M_VNET_DATA_FREE);
398 break;
399 }
400 s = (void *)df->vnd_start;
401 df->vnd_len -= size;
402 df->vnd_start = df->vnd_start + size;
403 break;
404 }
405 sx_xunlock(&vnet_data_free_lock);
406
407 return (s);
408 }
409
410 /*
411 * Free space for a virtualized global variable on module unload.
412 */
413 void
vnet_data_free(void * start_arg,int size)414 vnet_data_free(void *start_arg, int size)
415 {
416 struct vnet_data_free *df;
417 struct vnet_data_free *dn;
418 uintptr_t start;
419 uintptr_t end;
420
421 size = roundup2(size, sizeof(void *));
422 start = (uintptr_t)start_arg;
423 end = start + size;
424 /*
425 * Free a region of space and merge it with as many neighbors as
426 * possible. Keeping the list sorted simplifies this operation.
427 */
428 sx_xlock(&vnet_data_free_lock);
429 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
430 if (df->vnd_start > end)
431 break;
432 /*
433 * If we expand at the end of an entry we may have to merge
434 * it with the one following it as well.
435 */
436 if (df->vnd_start + df->vnd_len == start) {
437 df->vnd_len += size;
438 dn = TAILQ_NEXT(df, vnd_link);
439 if (df->vnd_start + df->vnd_len == dn->vnd_start) {
440 df->vnd_len += dn->vnd_len;
441 TAILQ_REMOVE(&vnet_data_free_head, dn,
442 vnd_link);
443 free(dn, M_VNET_DATA_FREE);
444 }
445 sx_xunlock(&vnet_data_free_lock);
446 return;
447 }
448 if (df->vnd_start == end) {
449 df->vnd_start = start;
450 df->vnd_len += size;
451 sx_xunlock(&vnet_data_free_lock);
452 return;
453 }
454 }
455 dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
456 dn->vnd_start = start;
457 dn->vnd_len = size;
458 if (df)
459 TAILQ_INSERT_BEFORE(df, dn, vnd_link);
460 else
461 TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
462 sx_xunlock(&vnet_data_free_lock);
463 }
464
465 /*
466 * When a new virtualized global variable has been allocated, propagate its
467 * initial value to each already-allocated virtual network stack instance.
468 */
469 void
vnet_data_copy(void * start,int size)470 vnet_data_copy(void *start, int size)
471 {
472 struct vnet *vnet;
473
474 VNET_LIST_RLOCK();
475 LIST_FOREACH(vnet, &vnet_head, vnet_le)
476 memcpy((void *)((uintptr_t)vnet->vnet_data_base +
477 (uintptr_t)start), start, size);
478 VNET_LIST_RUNLOCK();
479 }
480
481 /*
482 * Save a copy of the initial values of virtualized global variables.
483 */
484 void
vnet_save_init(void * start,size_t size)485 vnet_save_init(void *start, size_t size)
486 {
487 MPASS(vnet_init_var != 0);
488 MPASS(VNET_START <= (uintptr_t)start &&
489 (uintptr_t)start + size <= VNET_STOP);
490 memcpy((void *)(vnet_init_var + ((uintptr_t)start - VNET_START)),
491 start, size);
492 }
493
494 /*
495 * Restore the 'master' copies of virtualized global variables to theirs
496 * initial values.
497 */
498 void
vnet_restore_init(void * start,size_t size)499 vnet_restore_init(void *start, size_t size)
500 {
501 MPASS(vnet_init_var != 0);
502 MPASS(VNET_START <= (uintptr_t)start &&
503 (uintptr_t)start + size <= VNET_STOP);
504 memcpy(start,
505 (void *)(vnet_init_var + ((uintptr_t)start - VNET_START)), size);
506 }
507
508 /*
509 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
510 * and VNET_SYSUNINIT().
511 */
512 void
vnet_register_sysinit(void * arg)513 vnet_register_sysinit(void *arg)
514 {
515 struct vnet_sysinit *vs, *vs2;
516 struct vnet *vnet;
517
518 vs = arg;
519 KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
520
521 /* Add the constructor to the global list of vnet constructors. */
522 VNET_SYSINIT_WLOCK();
523 TAILQ_FOREACH(vs2, &vnet_constructors, link) {
524 if (vs2->subsystem > vs->subsystem)
525 break;
526 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
527 break;
528 }
529 if (vs2 != NULL)
530 TAILQ_INSERT_BEFORE(vs2, vs, link);
531 else
532 TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
533
534 /*
535 * Invoke the constructor on all the existing vnets when it is
536 * registered.
537 */
538 VNET_LIST_RLOCK();
539 VNET_FOREACH(vnet) {
540 CURVNET_SET_QUIET(vnet);
541 vs->func(vs->arg);
542 CURVNET_RESTORE();
543 }
544 VNET_LIST_RUNLOCK();
545 VNET_SYSINIT_WUNLOCK();
546 }
547
548 void
vnet_deregister_sysinit(void * arg)549 vnet_deregister_sysinit(void *arg)
550 {
551 struct vnet_sysinit *vs;
552
553 vs = arg;
554
555 /* Remove the constructor from the global list of vnet constructors. */
556 VNET_SYSINIT_WLOCK();
557 TAILQ_REMOVE(&vnet_constructors, vs, link);
558 VNET_SYSINIT_WUNLOCK();
559 }
560
561 void
vnet_register_sysuninit(void * arg)562 vnet_register_sysuninit(void *arg)
563 {
564 struct vnet_sysinit *vs, *vs2;
565
566 vs = arg;
567
568 /* Add the destructor to the global list of vnet destructors. */
569 VNET_SYSINIT_WLOCK();
570 TAILQ_FOREACH(vs2, &vnet_destructors, link) {
571 if (vs2->subsystem > vs->subsystem)
572 break;
573 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
574 break;
575 }
576 if (vs2 != NULL)
577 TAILQ_INSERT_BEFORE(vs2, vs, link);
578 else
579 TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
580 VNET_SYSINIT_WUNLOCK();
581 }
582
583 void
vnet_deregister_sysuninit(void * arg)584 vnet_deregister_sysuninit(void *arg)
585 {
586 struct vnet_sysinit *vs;
587 struct vnet *vnet;
588
589 vs = arg;
590
591 /*
592 * Invoke the destructor on all the existing vnets when it is
593 * deregistered.
594 */
595 VNET_SYSINIT_WLOCK();
596 VNET_LIST_RLOCK();
597 VNET_FOREACH(vnet) {
598 CURVNET_SET_QUIET(vnet);
599 vs->func(vs->arg);
600 CURVNET_RESTORE();
601 }
602
603 /* Remove the destructor from the global list of vnet destructors. */
604 TAILQ_REMOVE(&vnet_destructors, vs, link);
605 VNET_SYSINIT_WUNLOCK();
606 VNET_LIST_RUNLOCK();
607 }
608
609 /*
610 * Invoke all registered vnet constructors on the current vnet. Used during
611 * vnet construction. The caller is responsible for ensuring the new vnet is
612 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
613 */
614 static void
vnet_sysinit(void)615 vnet_sysinit(void)
616 {
617 struct vnet_sysinit *vs;
618
619 VNET_SYSINIT_RLOCK();
620 TAILQ_FOREACH(vs, &vnet_constructors, link) {
621 curvnet->vnet_state = vs->subsystem;
622 vs->func(vs->arg);
623 }
624 VNET_SYSINIT_RUNLOCK();
625 }
626
627 /*
628 * Invoke all registered vnet destructors on the current vnet. Used during
629 * vnet destruction. The caller is responsible for ensuring the dying vnet
630 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
631 */
632 static void
vnet_sysuninit(void)633 vnet_sysuninit(void)
634 {
635 struct vnet_sysinit *vs;
636
637 VNET_SYSINIT_RLOCK();
638 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
639 link) {
640 curvnet->vnet_state = vs->subsystem;
641 vs->func(vs->arg);
642 }
643 VNET_SYSINIT_RUNLOCK();
644 }
645
646 /*
647 * EVENTHANDLER(9) extensions.
648 */
649 /*
650 * Invoke the eventhandler function originally registered with the possibly
651 * registered argument for all virtual network stack instances.
652 *
653 * This iterator can only be used for eventhandlers that do not take any
654 * additional arguments, as we do ignore the variadic arguments from the
655 * EVENTHANDLER_INVOKE() call.
656 */
657 void
vnet_global_eventhandler_iterator_func(void * arg,...)658 vnet_global_eventhandler_iterator_func(void *arg, ...)
659 {
660 VNET_ITERATOR_DECL(vnet_iter);
661 struct eventhandler_entry_vimage *v_ee;
662
663 /*
664 * There is a bug here in that we should actually cast things to
665 * (struct eventhandler_entry_ ## name *) but that's not easily
666 * possible in here so just re-using the variadic version we
667 * defined for the generic vimage case.
668 */
669 v_ee = arg;
670 VNET_LIST_RLOCK();
671 VNET_FOREACH(vnet_iter) {
672 CURVNET_SET(vnet_iter);
673 ((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
674 CURVNET_RESTORE();
675 }
676 VNET_LIST_RUNLOCK();
677 }
678
679 #ifdef VNET_DEBUG
680 struct vnet_recursion {
681 SLIST_ENTRY(vnet_recursion) vnr_le;
682 const char *prev_fn;
683 const char *where_fn;
684 int where_line;
685 struct vnet *old_vnet;
686 struct vnet *new_vnet;
687 };
688
689 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
690 SLIST_HEAD_INITIALIZER(vnet_recursions);
691
692 static void
vnet_print_recursion(struct vnet_recursion * vnr,int brief)693 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
694 {
695
696 if (!brief)
697 printf("CURVNET_SET() recursion in ");
698 printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
699 vnr->prev_fn);
700 if (brief)
701 printf(", ");
702 else
703 printf("\n ");
704 printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
705 }
706
707 void
vnet_log_recursion(struct vnet * old_vnet,const char * old_fn,int line)708 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
709 {
710 struct vnet_recursion *vnr;
711
712 /* Skip already logged recursion events. */
713 SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
714 if (vnr->prev_fn == old_fn &&
715 vnr->where_fn == curthread->td_vnet_lpush &&
716 vnr->where_line == line &&
717 (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
718 return;
719
720 vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
721 if (vnr == NULL)
722 panic("%s: malloc failed", __func__);
723 vnr->prev_fn = old_fn;
724 vnr->where_fn = curthread->td_vnet_lpush;
725 vnr->where_line = line;
726 vnr->old_vnet = old_vnet;
727 vnr->new_vnet = curvnet;
728
729 SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
730
731 vnet_print_recursion(vnr, 0);
732 #ifdef KDB
733 kdb_backtrace();
734 #endif
735 }
736 #endif /* VNET_DEBUG */
737
738 /*
739 * DDB(4).
740 */
741 #ifdef DDB
742 static void
db_vnet_print(struct vnet * vnet)743 db_vnet_print(struct vnet *vnet)
744 {
745
746 db_printf("vnet = %p\n", vnet);
747 db_printf(" vnet_magic_n = %#08x (%s, orig %#08x)\n",
748 vnet->vnet_magic_n,
749 (vnet->vnet_magic_n == VNET_MAGIC_N) ?
750 "ok" : "mismatch", VNET_MAGIC_N);
751 db_printf(" vnet_ifcnt = %u\n", vnet->vnet_ifcnt);
752 db_printf(" vnet_sockcnt = %u\n", vnet->vnet_sockcnt);
753 db_printf(" vnet_data_mem = %p\n", vnet->vnet_data_mem);
754 db_printf(" vnet_data_base = %#jx\n",
755 (uintmax_t)vnet->vnet_data_base);
756 db_printf(" vnet_state = %#08x\n", vnet->vnet_state);
757 db_printf(" vnet_shutdown = %#03x\n", vnet->vnet_shutdown);
758 db_printf("\n");
759 }
760
DB_SHOW_ALL_COMMAND(vnets,db_show_all_vnets)761 DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets)
762 {
763 VNET_ITERATOR_DECL(vnet_iter);
764
765 VNET_FOREACH(vnet_iter) {
766 db_vnet_print(vnet_iter);
767 if (db_pager_quit)
768 break;
769 }
770 }
771
DB_SHOW_COMMAND(vnet,db_show_vnet)772 DB_SHOW_COMMAND(vnet, db_show_vnet)
773 {
774
775 if (!have_addr) {
776 db_printf("usage: show vnet <struct vnet *>\n");
777 return;
778 }
779
780 db_vnet_print((struct vnet *)addr);
781 }
782
783 static void
db_show_vnet_print_vs(struct vnet_sysinit * vs,int ddb)784 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
785 {
786 const char *vsname, *funcname;
787 c_db_sym_t sym;
788 db_expr_t offset;
789
790 #define xprint(...) do { \
791 if (ddb) \
792 db_printf(__VA_ARGS__); \
793 else \
794 printf(__VA_ARGS__); \
795 } while (0)
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
DB_SHOW_COMMAND_FLAGS(vnet_sysinit,db_show_vnet_sysinit,DB_CMD_MEMSAFE)813 DB_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
DB_SHOW_COMMAND_FLAGS(vnet_sysuninit,db_show_vnet_sysuninit,DB_CMD_MEMSAFE)827 DB_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
DB_SHOW_COMMAND_FLAGS(vnetrcrs,db_show_vnetrcrs,DB_CMD_MEMSAFE)843 DB_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