2 * Copyright (c) 2004-2009 University of Zagreb
3 * Copyright (c) 2006-2009 FreeBSD Foundation
6 * This software was developed by the University of Zagreb and the
7 * FreeBSD Foundation under sponsorship by the Stichting NLnet and the
10 * Copyright (c) 2009 Jeffrey Roberson <jeff@freebsd.org>
11 * Copyright (c) 2009 Robert N. M. Watson
12 * All rights reserved.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
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.
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
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
42 #include <sys/param.h>
44 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/eventhandler.h>
51 #include <sys/malloc.h>
53 #include <sys/socket.h>
55 #include <sys/sysctl.h>
57 #include <machine/stdarg.h>
61 #include <ddb/db_sym.h>
65 #include <net/if_var.h>
69 * This file implements core functions for virtual network stacks:
71 * - Virtual network stack management functions.
73 * - Virtual network stack memory allocator, which virtualizes global
74 * variables in the network stack
76 * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
77 * to register startup/shutdown events to be run for each virtual network
81 FEATURE(vimage, "VIMAGE kernel virtualization");
83 static MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
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.
91 struct rwlock vnet_rwlock;
92 struct sx vnet_sxlock;
94 #define VNET_LIST_WLOCK() do { \
95 sx_xlock(&vnet_sxlock); \
96 rw_wlock(&vnet_rwlock); \
99 #define VNET_LIST_WUNLOCK() do { \
100 rw_wunlock(&vnet_rwlock); \
101 sx_xunlock(&vnet_sxlock); \
104 struct vnet_list_head vnet_head;
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.
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
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
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.
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
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.
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.
161 #define VNET_BYTES (VNET_STOP - VNET_START)
163 static MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
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.
172 #define VNET_MODMIN 8192
173 #define VNET_SIZE roundup2(VNET_BYTES, PAGE_SIZE)
176 * Space to store virtualized global variables from loadable kernel modules,
177 * and the free list to manage it.
179 static VNET_DEFINE(char, modspace[VNET_MODMIN]);
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.
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);
191 struct sx vnet_sysinit_sxlock;
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);
198 struct vnet_data_free {
201 TAILQ_ENTRY(vnet_data_free) vnd_link;
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;
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",
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,
222 static void db_show_vnet_print_vs(struct vnet_sysinit *, int);
226 * Allocate a virtual network stack.
233 SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__);
234 vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
235 vnet->vnet_magic_n = VNET_MAGIC_N;
236 SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet);
239 * Allocate storage for virtualized global variables and copy in
240 * initial values form our 'master' copy.
242 vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
243 memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
246 * All use of vnet-specific data will immediately subtract VNET_START
247 * from the base memory pointer, so pre-calculate that now to avoid
250 vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
252 /* Initialize / attach vnet module instances. */
253 CURVNET_SET_QUIET(vnet);
258 LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
261 SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet);
266 * Destroy a virtual network stack.
269 vnet_destroy(struct vnet *vnet)
271 struct ifnet *ifp, *nifp;
273 SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet);
274 KASSERT(vnet->vnet_sockcnt == 0,
275 ("%s: vnet still has sockets", __func__));
278 LIST_REMOVE(vnet, vnet_le);
281 CURVNET_SET_QUIET(vnet);
283 /* Return all inherited interfaces to their parent vnets. */
284 TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
285 if (ifp->if_home_vnet != ifp->if_vnet)
286 if_vmove(ifp, ifp->if_home_vnet);
293 * Release storage for the virtual network stack instance.
295 free(vnet->vnet_data_mem, M_VNET_DATA);
296 vnet->vnet_data_mem = NULL;
297 vnet->vnet_data_base = 0;
298 vnet->vnet_magic_n = 0xdeadbeef;
300 SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__);
304 * Boot time initialization and allocation of virtual network stacks.
307 vnet_init_prelink(void *arg)
310 rw_init(&vnet_rwlock, "vnet_rwlock");
311 sx_init(&vnet_sxlock, "vnet_sxlock");
312 sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
313 LIST_INIT(&vnet_head);
315 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
316 vnet_init_prelink, NULL);
319 vnet0_init(void *arg)
322 /* Warn people before take off - in case we crash early. */
323 printf("WARNING: VIMAGE (virtualized network stack) is a highly "
324 "experimental feature.\n");
327 * We MUST clear curvnet in vi_init_done() before going SMP,
328 * otherwise CURVNET_SET() macros would scream about unnecessary
329 * curvnet recursions.
331 curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
333 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
336 vnet_init_done(void *unused)
342 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, vnet_init_done,
346 * Once on boot, initialize the modspace freelist to entirely cover modspace.
349 vnet_data_startup(void *dummy __unused)
351 struct vnet_data_free *df;
353 df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
354 df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
355 df->vnd_len = VNET_MODMIN;
356 TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
357 sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
359 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, 0);
362 * When a module is loaded and requires storage for a virtualized global
363 * variable, allocate space from the modspace free list. This interface
364 * should be used only by the kernel linker.
367 vnet_data_alloc(int size)
369 struct vnet_data_free *df;
373 size = roundup2(size, sizeof(void *));
374 sx_xlock(&vnet_data_free_lock);
375 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
376 if (df->vnd_len < size)
378 if (df->vnd_len == size) {
379 s = (void *)df->vnd_start;
380 TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
381 free(df, M_VNET_DATA_FREE);
384 s = (void *)df->vnd_start;
386 df->vnd_start = df->vnd_start + size;
389 sx_xunlock(&vnet_data_free_lock);
395 * Free space for a virtualized global variable on module unload.
398 vnet_data_free(void *start_arg, int size)
400 struct vnet_data_free *df;
401 struct vnet_data_free *dn;
405 size = roundup2(size, sizeof(void *));
406 start = (uintptr_t)start_arg;
409 * Free a region of space and merge it with as many neighbors as
410 * possible. Keeping the list sorted simplifies this operation.
412 sx_xlock(&vnet_data_free_lock);
413 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
414 if (df->vnd_start > end)
417 * If we expand at the end of an entry we may have to merge
418 * it with the one following it as well.
420 if (df->vnd_start + df->vnd_len == start) {
422 dn = TAILQ_NEXT(df, vnd_link);
423 if (df->vnd_start + df->vnd_len == dn->vnd_start) {
424 df->vnd_len += dn->vnd_len;
425 TAILQ_REMOVE(&vnet_data_free_head, dn,
427 free(dn, M_VNET_DATA_FREE);
429 sx_xunlock(&vnet_data_free_lock);
432 if (df->vnd_start == end) {
433 df->vnd_start = start;
435 sx_xunlock(&vnet_data_free_lock);
439 dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
440 dn->vnd_start = start;
443 TAILQ_INSERT_BEFORE(df, dn, vnd_link);
445 TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
446 sx_xunlock(&vnet_data_free_lock);
450 * When a new virtualized global variable has been allocated, propagate its
451 * initial value to each already-allocated virtual network stack instance.
454 vnet_data_copy(void *start, int size)
459 LIST_FOREACH(vnet, &vnet_head, vnet_le)
460 memcpy((void *)((uintptr_t)vnet->vnet_data_base +
461 (uintptr_t)start), start, size);
466 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
467 * and VNET_SYSUNINIT().
470 vnet_register_sysinit(void *arg)
472 struct vnet_sysinit *vs, *vs2;
476 KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
478 /* Add the constructor to the global list of vnet constructors. */
479 VNET_SYSINIT_WLOCK();
480 TAILQ_FOREACH(vs2, &vnet_constructors, link) {
481 if (vs2->subsystem > vs->subsystem)
483 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
487 TAILQ_INSERT_BEFORE(vs2, vs, link);
489 TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
492 * Invoke the constructor on all the existing vnets when it is
496 CURVNET_SET_QUIET(vnet);
500 VNET_SYSINIT_WUNLOCK();
504 vnet_deregister_sysinit(void *arg)
506 struct vnet_sysinit *vs;
510 /* Remove the constructor from the global list of vnet constructors. */
511 VNET_SYSINIT_WLOCK();
512 TAILQ_REMOVE(&vnet_constructors, vs, link);
513 VNET_SYSINIT_WUNLOCK();
517 vnet_register_sysuninit(void *arg)
519 struct vnet_sysinit *vs, *vs2;
523 /* Add the destructor to the global list of vnet destructors. */
524 VNET_SYSINIT_WLOCK();
525 TAILQ_FOREACH(vs2, &vnet_destructors, link) {
526 if (vs2->subsystem > vs->subsystem)
528 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
532 TAILQ_INSERT_BEFORE(vs2, vs, link);
534 TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
535 VNET_SYSINIT_WUNLOCK();
539 vnet_deregister_sysuninit(void *arg)
541 struct vnet_sysinit *vs;
547 * Invoke the destructor on all the existing vnets when it is
550 VNET_SYSINIT_WLOCK();
552 CURVNET_SET_QUIET(vnet);
557 /* Remove the destructor from the global list of vnet destructors. */
558 TAILQ_REMOVE(&vnet_destructors, vs, link);
559 VNET_SYSINIT_WUNLOCK();
563 * Invoke all registered vnet constructors on the current vnet. Used during
564 * vnet construction. The caller is responsible for ensuring the new vnet is
565 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
570 struct vnet_sysinit *vs;
572 VNET_SYSINIT_RLOCK();
573 TAILQ_FOREACH(vs, &vnet_constructors, link) {
576 VNET_SYSINIT_RUNLOCK();
580 * Invoke all registered vnet destructors on the current vnet. Used during
581 * vnet destruction. The caller is responsible for ensuring the dying vnet
582 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
587 struct vnet_sysinit *vs;
589 VNET_SYSINIT_RLOCK();
590 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
594 VNET_SYSINIT_RUNLOCK();
598 * EVENTHANDLER(9) extensions.
601 * Invoke the eventhandler function originally registered with the possibly
602 * registered argument for all virtual network stack instances.
604 * This iterator can only be used for eventhandlers that do not take any
605 * additional arguments, as we do ignore the variadic arguments from the
606 * EVENTHANDLER_INVOKE() call.
609 vnet_global_eventhandler_iterator_func(void *arg, ...)
611 VNET_ITERATOR_DECL(vnet_iter);
612 struct eventhandler_entry_vimage *v_ee;
615 * There is a bug here in that we should actually cast things to
616 * (struct eventhandler_entry_ ## name *) but that's not easily
617 * possible in here so just re-using the variadic version we
618 * defined for the generic vimage case.
622 VNET_FOREACH(vnet_iter) {
623 CURVNET_SET(vnet_iter);
624 ((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg);
631 struct vnet_recursion {
632 SLIST_ENTRY(vnet_recursion) vnr_le;
634 const char *where_fn;
636 struct vnet *old_vnet;
637 struct vnet *new_vnet;
640 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
641 SLIST_HEAD_INITIALIZER(vnet_recursions);
644 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
648 printf("CURVNET_SET() recursion in ");
649 printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
655 printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
659 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
661 struct vnet_recursion *vnr;
663 /* Skip already logged recursion events. */
664 SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
665 if (vnr->prev_fn == old_fn &&
666 vnr->where_fn == curthread->td_vnet_lpush &&
667 vnr->where_line == line &&
668 (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
671 vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
673 panic("%s: malloc failed", __func__);
674 vnr->prev_fn = old_fn;
675 vnr->where_fn = curthread->td_vnet_lpush;
676 vnr->where_line = line;
677 vnr->old_vnet = old_vnet;
678 vnr->new_vnet = curvnet;
680 SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
682 vnet_print_recursion(vnr, 0);
687 #endif /* VNET_DEBUG */
693 DB_SHOW_COMMAND(vnets, db_show_vnets)
695 VNET_ITERATOR_DECL(vnet_iter);
697 VNET_FOREACH(vnet_iter) {
698 db_printf("vnet = %p\n", vnet_iter);
699 db_printf(" vnet_magic_n = 0x%x (%s, orig 0x%x)\n",
700 vnet_iter->vnet_magic_n,
701 (vnet_iter->vnet_magic_n == VNET_MAGIC_N) ?
702 "ok" : "mismatch", VNET_MAGIC_N);
703 db_printf(" vnet_ifcnt = %u\n", vnet_iter->vnet_ifcnt);
704 db_printf(" vnet_sockcnt = %u\n", vnet_iter->vnet_sockcnt);
705 db_printf(" vnet_data_mem = %p\n", vnet_iter->vnet_data_mem);
706 db_printf(" vnet_data_base = 0x%jx\n",
707 (uintmax_t)vnet_iter->vnet_data_base);
715 db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb)
717 const char *vsname, *funcname;
721 #define xprint(...) \
723 db_printf(__VA_ARGS__); \
728 xprint("%s: no vnet_sysinit * given\n", __func__);
732 sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset);
733 db_symbol_values(sym, &vsname, NULL);
734 sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset);
735 db_symbol_values(sym, &funcname, NULL);
736 xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs);
737 xprint(" 0x%08x 0x%08x\n", vs->subsystem, vs->order);
738 xprint(" %p(%s)(%p)\n",
739 vs->func, (funcname != NULL) ? funcname : "", vs->arg);
743 DB_SHOW_COMMAND(vnet_sysinit, db_show_vnet_sysinit)
745 struct vnet_sysinit *vs;
747 db_printf("VNET_SYSINIT vs Name(Ptr)\n");
748 db_printf(" Subsystem Order\n");
749 db_printf(" Function(Name)(Arg)\n");
750 TAILQ_FOREACH(vs, &vnet_constructors, link) {
751 db_show_vnet_print_vs(vs, 1);
757 DB_SHOW_COMMAND(vnet_sysuninit, db_show_vnet_sysuninit)
759 struct vnet_sysinit *vs;
761 db_printf("VNET_SYSUNINIT vs Name(Ptr)\n");
762 db_printf(" Subsystem Order\n");
763 db_printf(" Function(Name)(Arg)\n");
764 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
766 db_show_vnet_print_vs(vs, 1);
773 DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs)
775 struct vnet_recursion *vnr;
777 SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
778 vnet_print_recursion(vnr, 1);