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>
46 #include <sys/systm.h>
47 #include <sys/sysctl.h>
48 #include <sys/linker_set.h>
50 #include <sys/malloc.h>
52 #include <sys/socket.h>
54 #include <sys/sysctl.h>
61 #include <net/if_var.h>
65 * This file implements core functions for virtual network stacks:
67 * - Virtual network stack management functions.
69 * - Virtual network stack memory allocator, which virtualizes global
70 * variables in the network stack
72 * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems
73 * to register startup/shutdown events to be run for each virtual network
77 MALLOC_DEFINE(M_VNET, "vnet", "network stack control block");
80 * The virtual network stack list has two read-write locks, one sleepable and
81 * the other not, so that the list can be stablized and walked in a variety
82 * of network stack contexts. Both must be acquired exclusively to modify
83 * the list, but a read lock of either lock is sufficient to walk the list.
85 struct rwlock vnet_rwlock;
86 struct sx vnet_sxlock;
88 #define VNET_LIST_WLOCK() do { \
89 sx_xlock(&vnet_sxlock); \
90 rw_wlock(&vnet_rwlock); \
93 #define VNET_LIST_WUNLOCK() do { \
94 rw_wunlock(&vnet_rwlock); \
95 sx_xunlock(&vnet_sxlock); \
98 struct vnet_list_head vnet_head;
102 * The virtual network stack allocator provides storage for virtualized
103 * global variables. These variables are defined/declared using the
104 * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet'
105 * linker set. The details of the implementation are somewhat subtle, but
106 * allow the majority of most network subsystems to maintain
107 * virtualization-agnostic.
109 * The virtual network stack allocator handles variables in the base kernel
110 * vs. modules in similar but different ways. In both cases, virtualized
111 * global variables are marked as such by being declared to be part of the
112 * vnet linker set. These "master" copies of global variables serve two
115 * (1) They contain static initialization or "default" values for global
116 * variables which will be propagated to each virtual network stack
117 * instance when created. As with normal global variables, they default
120 * (2) They act as unique global names by which the variable can be referred
121 * to, regardless of network stack instance. The single global symbol
122 * will be used to calculate the location of a per-virtual instance
123 * variable at run-time.
125 * Each virtual network stack instance has a complete copy of each
126 * virtualized global variable, stored in a malloc'd block of memory
127 * referred to by vnet->vnet_data_mem. Critical to the design is that each
128 * per-instance memory block is laid out identically to the master block so
129 * that the offset of each global variable is the same across all blocks. To
130 * optimize run-time access, a precalculated 'base' address,
131 * vnet->vnet_data_base, is stored in each vnet, and is the amount that can
132 * be added to the address of a 'master' instance of a variable to get to the
135 * Virtualized global variables are handled in a similar manner, but as each
136 * module has its own 'set_vnet' linker set, and we want to keep all
137 * virtualized globals togther, we reserve space in the kernel's linker set
138 * for potential module variables using a per-vnet character array,
139 * 'modspace'. The virtual network stack allocator maintains a free list to
140 * track what space in the array is free (all, initially) and as modules are
141 * linked, allocates portions of the space to specific globals. The kernel
142 * module linker queries the virtual network stack allocator and will
143 * bind references of the global to the location during linking. It also
144 * calls into the virtual network stack allocator, once the memory is
145 * initialized, in order to propagate the new static initializations to all
146 * existing virtual network stack instances so that the soon-to-be executing
147 * module will find every network stack instance with proper default values.
151 * Location of the kernel's 'set_vnet' linker set.
153 extern uintptr_t *__start_set_vnet;
154 extern uintptr_t *__stop_set_vnet;
156 #define VNET_START (uintptr_t)&__start_set_vnet
157 #define VNET_STOP (uintptr_t)&__stop_set_vnet
160 * Number of bytes of data in the 'set_vnet' linker set, and hence the total
161 * size of all kernel virtualized global variables, and the malloc(9) type
162 * that will be used to allocate it.
164 #define VNET_BYTES (VNET_STOP - VNET_START)
166 MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data");
169 * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of
170 * global variables across all loaded modules. As this actually sizes an
171 * array declared as a virtualized global variable in the kernel itself, and
172 * we want the virtualized global variable space to be page-sized, we may
173 * have more space than that in practice.
175 #define VNET_MODMIN 8192
176 #define VNET_SIZE roundup2(VNET_BYTES, PAGE_SIZE)
177 #define VNET_MODSIZE (VNET_SIZE - (VNET_BYTES - VNET_MODMIN))
180 * Space to store virtualized global variables from loadable kernel modules,
181 * and the free list to manage it.
183 static VNET_DEFINE(char, modspace[VNET_MODMIN]);
186 * Global lists of subsystem constructor and destructors for vnets. They are
187 * registered via VNET_SYSINIT() and VNET_SYSUNINIT(). Both lists are
188 * protected by the vnet_sysinit_sxlock global lock.
190 static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors =
191 TAILQ_HEAD_INITIALIZER(vnet_constructors);
192 static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors =
193 TAILQ_HEAD_INITIALIZER(vnet_destructors);
195 struct sx vnet_sysinit_sxlock;
197 #define VNET_SYSINIT_WLOCK() sx_xlock(&vnet_sysinit_sxlock);
198 #define VNET_SYSINIT_WUNLOCK() sx_xunlock(&vnet_sysinit_sxlock);
199 #define VNET_SYSINIT_RLOCK() sx_slock(&vnet_sysinit_sxlock);
200 #define VNET_SYSINIT_RUNLOCK() sx_sunlock(&vnet_sysinit_sxlock);
202 struct vnet_data_free {
205 TAILQ_ENTRY(vnet_data_free) vnd_link;
208 MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free", "VNET resource accounting");
209 static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head =
210 TAILQ_HEAD_INITIALIZER(vnet_data_free_head);
211 static struct sx vnet_data_free_lock;
214 * Allocate a virtual network stack.
221 vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO);
222 vnet->vnet_magic_n = VNET_MAGIC_N;
225 * Allocate storage for virtualized global variables and copy in
226 * initial values form our 'master' copy.
228 vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK);
229 memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES);
232 * All use of vnet-specific data will immediately subtract VNET_START
233 * from the base memory pointer, so pre-calculate that now to avoid
236 vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START;
238 /* Initialize / attach vnet module instances. */
239 CURVNET_SET_QUIET(vnet);
244 LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le);
251 * Destroy a virtual network stack.
254 vnet_destroy(struct vnet *vnet)
256 struct ifnet *ifp, *nifp;
258 KASSERT(vnet->vnet_sockcnt == 0,
259 ("%s: vnet still has sockets", __func__));
262 LIST_REMOVE(vnet, vnet_le);
265 CURVNET_SET_QUIET(vnet);
267 /* Return all inherited interfaces to their parent vnets. */
268 TAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) {
269 if (ifp->if_home_vnet != ifp->if_vnet)
270 if_vmove(ifp, ifp->if_home_vnet);
277 * Release storage for the virtual network stack instance.
279 free(vnet->vnet_data_mem, M_VNET_DATA);
280 vnet->vnet_data_mem = NULL;
281 vnet->vnet_data_base = 0;
282 vnet->vnet_magic_n = 0xdeadbeef;
287 * Boot time initialization and allocation of virtual network stacks.
290 vnet_init_prelink(void *arg)
293 rw_init(&vnet_rwlock, "vnet_rwlock");
294 sx_init(&vnet_sxlock, "vnet_sxlock");
295 sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock");
296 LIST_INIT(&vnet_head);
298 SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST,
299 vnet_init_prelink, NULL);
302 vnet0_init(void *arg)
305 /* Warn people before take off - in case we crash early. */
306 printf("WARNING: VIMAGE (virtualized network stack) is a highly "
307 "experimental feature.\n");
310 * We MUST clear curvnet in vi_init_done() before going SMP,
311 * otherwise CURVNET_SET() macros would scream about unnecessary
312 * curvnet recursions.
314 curvnet = prison0.pr_vnet = vnet0 = vnet_alloc();
316 SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL);
319 vnet_init_done(void *unused)
325 SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_FIRST, vnet_init_done,
329 * Once on boot, initialize the modspace freelist to entirely cover modspace.
332 vnet_data_startup(void *dummy __unused)
334 struct vnet_data_free *df;
336 df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
337 df->vnd_start = (uintptr_t)&VNET_NAME(modspace);
338 df->vnd_len = VNET_MODSIZE;
339 TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link);
340 sx_init(&vnet_data_free_lock, "vnet_data alloc lock");
342 SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, 0);
345 * When a module is loaded and requires storage for a virtualized global
346 * variable, allocate space from the modspace free list. This interface
347 * should be used only by the kernel linker.
350 vnet_data_alloc(int size)
352 struct vnet_data_free *df;
356 size = roundup2(size, sizeof(void *));
357 sx_xlock(&vnet_data_free_lock);
358 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
359 if (df->vnd_len < size)
361 if (df->vnd_len == size) {
362 s = (void *)df->vnd_start;
363 TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link);
364 free(df, M_VNET_DATA_FREE);
367 s = (void *)df->vnd_start;
369 df->vnd_start = df->vnd_start + size;
372 sx_xunlock(&vnet_data_free_lock);
378 * Free space for a virtualized global variable on module unload.
381 vnet_data_free(void *start_arg, int size)
383 struct vnet_data_free *df;
384 struct vnet_data_free *dn;
388 size = roundup2(size, sizeof(void *));
389 start = (uintptr_t)start_arg;
392 * Free a region of space and merge it with as many neighbors as
393 * possible. Keeping the list sorted simplifies this operation.
395 sx_xlock(&vnet_data_free_lock);
396 TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) {
397 if (df->vnd_start > end)
400 * If we expand at the end of an entry we may have to merge
401 * it with the one following it as well.
403 if (df->vnd_start + df->vnd_len == start) {
405 dn = TAILQ_NEXT(df, vnd_link);
406 if (df->vnd_start + df->vnd_len == dn->vnd_start) {
407 df->vnd_len += dn->vnd_len;
408 TAILQ_REMOVE(&vnet_data_free_head, dn,
410 free(dn, M_VNET_DATA_FREE);
412 sx_xunlock(&vnet_data_free_lock);
415 if (df->vnd_start == end) {
416 df->vnd_start = start;
418 sx_xunlock(&vnet_data_free_lock);
422 dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO);
423 dn->vnd_start = start;
426 TAILQ_INSERT_BEFORE(df, dn, vnd_link);
428 TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link);
429 sx_xunlock(&vnet_data_free_lock);
433 * When a new virtualized global variable has been allocated, propagate its
434 * initial value to each already-allocated virtual network stack instance.
437 vnet_data_copy(void *start, int size)
442 LIST_FOREACH(vnet, &vnet_head, vnet_le)
443 memcpy((void *)((uintptr_t)vnet->vnet_data_base +
444 (uintptr_t)start), start, size);
449 * Variants on sysctl_handle_foo that know how to handle virtualized global
450 * variables: if 'arg1' is a pointer, then we transform it to the local vnet
454 vnet_sysctl_handle_int(SYSCTL_HANDLER_ARGS)
458 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
459 return (sysctl_handle_int(oidp, arg1, arg2, req));
463 vnet_sysctl_handle_opaque(SYSCTL_HANDLER_ARGS)
467 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
468 return (sysctl_handle_opaque(oidp, arg1, arg2, req));
472 vnet_sysctl_handle_string(SYSCTL_HANDLER_ARGS)
476 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
477 return (sysctl_handle_string(oidp, arg1, arg2, req));
481 vnet_sysctl_handle_uint(SYSCTL_HANDLER_ARGS)
485 arg1 = (void *)(curvnet->vnet_data_base + (uintptr_t)arg1);
486 return (sysctl_handle_int(oidp, arg1, arg2, req));
490 * Support for special SYSINIT handlers registered via VNET_SYSINIT()
491 * and VNET_SYSUNINIT().
494 vnet_register_sysinit(void *arg)
496 struct vnet_sysinit *vs, *vs2;
500 KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early"));
502 /* Add the constructor to the global list of vnet constructors. */
503 VNET_SYSINIT_WLOCK();
504 TAILQ_FOREACH(vs2, &vnet_constructors, link) {
505 if (vs2->subsystem > vs->subsystem)
507 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
511 TAILQ_INSERT_BEFORE(vs2, vs, link);
513 TAILQ_INSERT_TAIL(&vnet_constructors, vs, link);
516 * Invoke the constructor on all the existing vnets when it is
520 CURVNET_SET_QUIET(vnet);
524 VNET_SYSINIT_WUNLOCK();
528 vnet_deregister_sysinit(void *arg)
530 struct vnet_sysinit *vs;
534 /* Remove the constructor from the global list of vnet constructors. */
535 VNET_SYSINIT_WLOCK();
536 TAILQ_REMOVE(&vnet_constructors, vs, link);
537 VNET_SYSINIT_WUNLOCK();
541 vnet_register_sysuninit(void *arg)
543 struct vnet_sysinit *vs, *vs2;
547 /* Add the destructor to the global list of vnet destructors. */
548 VNET_SYSINIT_WLOCK();
549 TAILQ_FOREACH(vs2, &vnet_destructors, link) {
550 if (vs2->subsystem > vs->subsystem)
552 if (vs2->subsystem == vs->subsystem && vs2->order > vs->order)
556 TAILQ_INSERT_BEFORE(vs2, vs, link);
558 TAILQ_INSERT_TAIL(&vnet_destructors, vs, link);
559 VNET_SYSINIT_WUNLOCK();
563 vnet_deregister_sysuninit(void *arg)
565 struct vnet_sysinit *vs;
571 * Invoke the destructor on all the existing vnets when it is
574 VNET_SYSINIT_WLOCK();
576 CURVNET_SET_QUIET(vnet);
581 /* Remove the destructor from the global list of vnet destructors. */
582 TAILQ_REMOVE(&vnet_destructors, vs, link);
583 VNET_SYSINIT_WUNLOCK();
587 * Invoke all registered vnet constructors on the current vnet. Used during
588 * vnet construction. The caller is responsible for ensuring the new vnet is
589 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
594 struct vnet_sysinit *vs;
596 VNET_SYSINIT_RLOCK();
597 TAILQ_FOREACH(vs, &vnet_constructors, link) {
600 VNET_SYSINIT_RUNLOCK();
604 * Invoke all registered vnet destructors on the current vnet. Used during
605 * vnet destruction. The caller is responsible for ensuring the dying vnet
606 * the current vnet and that the vnet_sysinit_sxlock lock is locked.
611 struct vnet_sysinit *vs;
613 VNET_SYSINIT_RLOCK();
614 TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head,
618 VNET_SYSINIT_RUNLOCK();
622 struct vnet_recursion {
623 SLIST_ENTRY(vnet_recursion) vnr_le;
625 const char *where_fn;
627 struct vnet *old_vnet;
628 struct vnet *new_vnet;
631 static SLIST_HEAD(, vnet_recursion) vnet_recursions =
632 SLIST_HEAD_INITIALIZER(vnet_recursions);
635 vnet_print_recursion(struct vnet_recursion *vnr, int brief)
639 printf("CURVNET_SET() recursion in ");
640 printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line,
646 printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet);
650 vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line)
652 struct vnet_recursion *vnr;
654 /* Skip already logged recursion events. */
655 SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
656 if (vnr->prev_fn == old_fn &&
657 vnr->where_fn == curthread->td_vnet_lpush &&
658 vnr->where_line == line &&
659 (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet))
662 vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO);
664 panic("%s: malloc failed", __func__);
665 vnr->prev_fn = old_fn;
666 vnr->where_fn = curthread->td_vnet_lpush;
667 vnr->where_line = line;
668 vnr->old_vnet = old_vnet;
669 vnr->new_vnet = curvnet;
671 SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le);
673 vnet_print_recursion(vnr, 0);
678 #endif /* VNET_DEBUG */
681 DB_SHOW_COMMAND(vnets, db_show_vnets)
683 VNET_ITERATOR_DECL(vnet_iter);
685 VNET_FOREACH(vnet_iter) {
686 db_printf("vnet = %p\n", vnet_iter);
687 db_printf(" vnet_magic_n = 0x%x (%s, orig 0x%x)\n",
688 vnet_iter->vnet_magic_n,
689 (vnet_iter->vnet_magic_n == VNET_MAGIC_N) ?
690 "ok" : "mismatch", VNET_MAGIC_N);
691 db_printf(" vnet_ifcnt = %u\n", vnet_iter->vnet_ifcnt);
692 db_printf(" vnet_sockcnt = %u\n", vnet_iter->vnet_sockcnt);
693 db_printf(" vnet_data_mem = %p\n", vnet_iter->vnet_data_mem);
694 db_printf(" vnet_data_base = 0x%jx\n",
695 (uintmax_t)vnet_iter->vnet_data_base);
703 DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs)
705 struct vnet_recursion *vnr;
707 SLIST_FOREACH(vnr, &vnet_recursions, vnr_le)
708 vnet_print_recursion(vnr, 1);