2 * Copyright (c) 2006 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include <sys/types.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
41 #include <sys/random.h>
42 #include <sys/vkernel.h>
44 #include <sys/reboot.h>
46 #include <sys/msgbuf.h>
47 #include <sys/vmspace.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/sysctl.h>
52 #include <vm/vm_page.h>
53 #include <sys/mplock2.h>
55 #include <machine/cpu.h>
56 #include <machine/globaldata.h>
57 #include <machine/tls.h>
58 #include <machine/md_var.h>
59 #include <machine/vmparam.h>
60 #include <cpu/specialreg.h>
63 #include <net/if_arp.h>
64 #include <net/ethernet.h>
65 #include <net/bridge/if_bridgevar.h>
66 #include <netinet/in.h>
67 #include <arpa/inet.h>
68 #include <net/if_var.h>
82 vm_paddr_t phys_avail[16];
84 vm_paddr_t Maxmem_bytes;
87 struct vkdisk_info DiskInfo[VKDISK_MAX];
89 struct vknetif_info NetifInfo[VKNETIF_MAX];
95 vm_offset_t virtual_start;
96 vm_offset_t virtual_end;
97 vm_offset_t virtual2_start;
98 vm_offset_t virtual2_end;
99 vm_offset_t kernel_vm_end;
100 vm_offset_t crashdumpmap;
101 vm_offset_t clean_sva;
102 vm_offset_t clean_eva;
103 struct msgbuf *msgbufp;
106 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
107 u_int cpu_feature; /* XXX */
111 int64_t tsc_frequency;
112 int optcpus; /* number of cpus - see mp_start() */
113 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
114 int real_ncpus; /* number of real CPUs */
115 int next_cpu; /* next real CPU to lock a virtual CPU to */
116 int vkernel_b_arg; /* no of logical CPU bits - only SMP */
117 int vkernel_B_arg; /* no of core bits - only SMP */
119 int via_feature_xcrypt = 0; /* XXX */
120 int via_feature_rng = 0; /* XXX */
122 struct privatespace *CPU_prvspace;
124 static struct trapframe proc0_tf;
125 static void *proc0paddr;
127 static void init_sys_memory(char *imageFile);
128 static void init_kern_memory(void);
129 static void init_globaldata(void);
130 static void init_vkernel(void);
131 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
132 static void init_netif(char *netifExp[], int netifFileNum);
133 static void writepid(void);
134 static void cleanpid(void);
135 static int unix_connect(const char *path);
136 static void usage_err(const char *ctl, ...);
137 static void usage_help(_Bool);
138 static void init_locks(void);
141 static char **save_av;
144 * Kernel startup for virtual kernels - standard main()
147 main(int ac, char **av)
149 char *memImageFile = NULL;
150 char *netifFile[VKNETIF_MAX];
151 char *diskFile[VKDISK_MAX];
152 char *cdFile[VKDISK_MAX];
157 int netifFileNum = 0;
160 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
168 int dflag = 0; /* disable vmm */
169 int real_vkernel_enable;
184 kernel_mem_readonly = 1;
188 lwp_cpu_lock = LCL_NONE;
190 real_vkernel_enable = 0;
191 vsize = sizeof(real_vkernel_enable);
192 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
194 if (real_vkernel_enable == 0) {
195 errx(1, "vm.vkernel_enable is 0, must be set "
196 "to 1 to execute a vkernel!");
200 vsize = sizeof(real_ncpus);
201 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
206 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:Ud")) != -1) {
209 printf("vmm: No need to disable. Hardware pagetable "
210 "is not available in vkernel32.\n");
215 * name=value:name=value:name=value...
218 * Allow values to be quoted but note that shells
219 * may remove the quotes, so using this feature
220 * to embed colons may require a backslash.
227 kern_envp = malloc(kenv_size);
228 if (kern_envp == NULL)
229 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
231 kenv_size2 = kenv_size + n + 1;
233 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
234 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
236 kenv_size = kenv_size2;
239 for (i = 0, j = pos; i < n; ++i) {
240 if (optarg[i] == '"')
242 else if (optarg[i] == '\'')
244 else if (isq == 0 && optarg[i] == ':')
247 kern_envp[j++] = optarg[i];
254 boothowto |= RB_SINGLE;
260 memImageFile = optarg;
263 if (netifFileNum < VKNETIF_MAX)
264 netifFile[netifFileNum++] = strdup(optarg);
269 if (diskFileNum + cdFileNum < VKDISK_MAX)
270 diskFile[diskFileNum++] = strdup(optarg);
275 if (diskFileNum + cdFileNum < VKDISK_MAX)
276 cdFile[cdFileNum++] = strdup(optarg);
279 Maxmem_bytes = strtoull(optarg, &suffix, 0);
296 usage_err("Bad maxmem option");
304 if (strncmp("map", optarg, 3) == 0) {
305 lwp_cpu_lock = LCL_PER_CPU;
306 if (optarg[3] == ',') {
307 next_cpu = strtol(optarg+4, &endp, 0);
309 usage_err("Bad target CPU number at '%s'", endp);
313 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
314 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
315 } else if (strncmp("any", optarg, 3) == 0) {
316 lwp_cpu_lock = LCL_NONE;
318 lwp_cpu_lock = LCL_SINGLE_CPU;
319 next_cpu = strtol(optarg, &endp, 0);
321 usage_err("Bad target CPU number at '%s'", endp);
322 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
323 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
328 * This value is set up by mp_start(), don't just
331 tok = strtok(optarg, ":");
332 optcpus = strtol(tok, NULL, 0);
333 if (optcpus < 1 || optcpus > MAXCPU)
334 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
336 /* :core_bits argument */
337 tok = strtok(NULL, ":");
339 vkernel_b_arg = strtol(tok, NULL, 0);
341 /* :logical_CPU_bits argument */
342 tok = strtok(NULL, ":");
344 vkernel_B_arg = strtol(tok, NULL, 0);
353 kernel_mem_readonly = 0;
365 init_sys_memory(memImageFile);
377 vsize = sizeof(tsc_present);
378 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
379 vsize = sizeof(tsc_invariant);
380 sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0);
381 vsize = sizeof(tsc_mpsync);
382 sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0);
383 vsize = sizeof(tsc_frequency);
384 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
386 cpu_feature |= CPUID_TSC;
391 vsize = sizeof(supports_sse);
393 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
394 init_fpu(supports_sse);
396 cpu_feature |= CPUID_SSE | CPUID_FXSR;
399 * We boot from the first installed disk.
401 if (bootOnDisk == 1) {
402 init_disk(diskFile, diskFileNum, VKD_DISK);
403 init_disk(cdFile, cdFileNum, VKD_CD);
405 init_disk(cdFile, cdFileNum, VKD_CD);
406 init_disk(diskFile, diskFileNum, VKD_DISK);
408 init_netif(netifFile, netifFileNum);
416 * Initialize system memory. This is the virtual kernel's 'RAM'.
420 init_sys_memory(char *imageFile)
427 * Figure out the system memory image size. If an image file was
428 * specified and -m was not specified, use the image file's size.
430 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
431 Maxmem_bytes = (vm_paddr_t)st.st_size;
432 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
434 errx(1, "Cannot create new memory file %s unless "
435 "system memory size is specified with -m",
441 * Maxmem must be known at this time
443 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
444 errx(1, "Bad maxmem specification: 32MB minimum, "
445 "multiples of %dMB only",
446 SEG_SIZE / 1024 / 1024);
451 * Generate an image file name if necessary, then open/create the
452 * file exclusively locked. Do not allow multiple virtual kernels
453 * to use the same image file.
455 * Don't iterate through a million files if we do not have write
456 * access to the directory, stop if our open() failed on a
457 * non-existant file. Otherwise opens can fail for any number
458 * of reasons (lock failed, file not owned or writable by us, etc).
460 if (imageFile == NULL) {
461 for (i = 0; i < 1000000; ++i) {
462 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
464 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
465 if (fd < 0 && stat(imageFile, &st) == 0) {
472 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
474 printf("Using memory file: %s\n", imageFile);
475 if (fd < 0 || fstat(fd, &st) < 0) {
476 err(1, "Unable to open/create %s", imageFile);
481 * Truncate or extend the file as necessary. Clean out the contents
482 * of the file, we want it to be full of holes so we don't waste
483 * time reading in data from an old file that we no longer care
487 ftruncate(fd, Maxmem_bytes);
490 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
495 * Initialize pool tokens and other necessary locks
502 * Get the initial mplock with a count of 1 for the BSP.
503 * This uses a LOGICAL cpu ID, ie BSP == 0.
505 cpu_get_initial_mplock();
507 /* our token pool needs to work early */
508 lwkt_token_pool_init();
513 * Initialize kernel memory. This reserves kernel virtual memory by using
519 init_kern_memory(void)
525 char *topofstack = &dummy;
530 * Memory map our kernel virtual memory space. Note that the
531 * kernel image itself is not made part of this memory for the
534 * The memory map must be segment-aligned so we can properly
537 * If the system kernel has a different MAXDSIZ, it might not
538 * be possible to map kernel memory in its prefered location.
539 * Try a number of different locations.
541 try = (void *)0x40000000;
543 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
544 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
545 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
549 if (base != MAP_FAILED)
550 munmap(base, KERNEL_KVA_SIZE);
551 try = (char *)try + 0x10000000;
554 err(1, "Unable to mmap() kernel virtual memory!");
557 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
558 KvaStart = (vm_offset_t)base;
559 KvaSize = KERNEL_KVA_SIZE;
560 KvaEnd = KvaStart + KvaSize;
562 /* cannot use kprintf yet */
563 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
566 * Create a top-level page table self-mapping itself.
568 * Initialize the page directory at physical page index 0 to point
569 * to an array of page table pages starting at physical page index 1
571 lseek(MemImageFd, 0L, 0);
572 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
573 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_RW;
574 write(MemImageFd, &pte, sizeof(pte));
578 * Initialize the PTEs in the page table pages required to map the
579 * page table itself. This includes mapping the page directory page
580 * at the base so we go one more loop then normal.
582 lseek(MemImageFd, PAGE_SIZE, 0);
583 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
584 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_RW;
585 write(MemImageFd, &pte, sizeof(pte));
589 * Initialize remaining PTEs to 0. We may be reusing a memory image
590 * file. This is approximately a megabyte.
592 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
593 zero = malloc(PAGE_SIZE);
594 bzero(zero, PAGE_SIZE);
596 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
597 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
602 * Enable the page table and calculate pointers to our self-map
603 * for easy kernel page table manipulation.
605 * KernelPTA must be offset so we can do direct VA translations
607 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
608 0 | VPTE_RW | VPTE_V);
609 KernelPTD = (vpte_t *)base; /* pg directory */
610 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
611 KernelPTA -= KvaStart >> PAGE_SHIFT;
614 * phys_avail[] represents unallocated physical memory. MI code
615 * will use phys_avail[] to create the vm_page array.
617 phys_avail[0] = PAGE_SIZE +
618 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
619 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
620 phys_avail[1] = Maxmem_bytes;
623 * (virtual_start, virtual_end) represent unallocated kernel virtual
624 * memory. MI code will create kernel_map using these parameters.
626 virtual_start = KvaStart + PAGE_SIZE +
627 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
628 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
629 virtual_end = KvaStart + KERNEL_KVA_SIZE;
632 * kernel_vm_end could be set to virtual_end but we want some
633 * indication of how much of the kernel_map we've used, so
634 * set it low and let pmap_growkernel increase it even though we
635 * don't need to create any new page table pages.
637 kernel_vm_end = virtual_start;
640 * Allocate space for process 0's UAREA.
642 proc0paddr = (void *)virtual_start;
643 for (i = 0; i < UPAGES; ++i) {
644 pmap_kenter_quick(virtual_start, phys_avail[0]);
645 virtual_start += PAGE_SIZE;
646 phys_avail[0] += PAGE_SIZE;
652 crashdumpmap = virtual_start;
653 virtual_start += MAXDUMPPGS * PAGE_SIZE;
656 * msgbufp maps the system message buffer
658 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
659 msgbufp = (void *)virtual_start;
660 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
661 pmap_kenter_quick(virtual_start, phys_avail[0]);
662 virtual_start += PAGE_SIZE;
663 phys_avail[0] += PAGE_SIZE;
665 msgbufinit(msgbufp, MSGBUF_SIZE);
668 * used by kern_memio for /dev/mem access
670 ptvmmap = (caddr_t)virtual_start;
671 virtual_start += PAGE_SIZE;
674 * Bootstrap the kernel_pmap
680 * Map the per-cpu globaldata for cpu #0. Allocate the space using
681 * virtual_start and phys_avail[0]
685 init_globaldata(void)
692 * Reserve enough KVA to cover possible cpus. This is a considerable
693 * amount of KVA since the privatespace structure includes two
694 * whole page table mappings.
696 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
697 CPU_prvspace = (void *)virtual_start;
698 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
701 * Allocate enough physical memory to cover the mdglobaldata
702 * portion of the space and the idle stack and map the pages
703 * into KVA. For cpu #0 only.
705 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
707 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
708 pmap_kenter_quick(va, pa);
709 phys_avail[0] += PAGE_SIZE;
711 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
713 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
714 pmap_kenter_quick(va, pa);
715 phys_avail[0] += PAGE_SIZE;
719 * Setup the %fs for cpu #0. The mycpu macro works after this
720 * point. Note that %gs is used by pthreads.
722 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
726 * Initialize very low level systems including thread0, proc0, etc.
732 struct mdglobaldata *gd;
734 gd = &CPU_prvspace[0].mdglobaldata;
735 bzero(gd, sizeof(*gd));
737 gd->mi.gd_curthread = &thread0;
738 thread0.td_gd = &gd->mi;
740 ncpus2 = 1; /* rounded down power of 2 */
741 ncpus_fit = 1; /* rounded up power of 2 */
742 /* ncpus2_mask and ncpus_fit_mask are 0 */
744 gd->mi.gd_prvspace = &CPU_prvspace[0];
745 mi_gdinit(&gd->mi, 0);
747 mi_proc0init(&gd->mi, proc0paddr);
748 lwp0.lwp_md.md_regs = &proc0_tf;
753 #if 0 /* #ifdef DDB */
755 if (boothowto & RB_KDB)
756 Debugger("Boot flags requested debugger");
760 initializecpu(); /* Initialize CPU registers */
762 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
766 * Map the message buffer
768 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
769 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
770 msgbufinit(msgbufp, MSGBUF_SIZE);
773 thread0.td_pcb_cr3 ... MMU
774 lwp0.lwp_md.md_regs = &proc0_tf;
779 * Filesystem image paths for the virtual kernel are optional.
780 * If specified they each should point to a disk image,
781 * the first of which will become the root disk.
783 * The virtual kernel caches data from our 'disk' just like a normal kernel,
784 * so we do not really want the real kernel to cache the data too. Use
785 * O_DIRECT to remove the duplication.
789 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
794 if (diskFileNum == 0)
797 for(i=0; i < diskFileNum; i++){
802 warnx("Invalid argument to '-r'");
806 * Check for a serial number for the virtual disk
807 * passed from the command line.
812 if (DiskNum < VKDISK_MAX) {
814 struct vkdisk_info* info = NULL;
818 if (type == VKD_DISK)
819 fd = open(fname, O_RDWR|O_DIRECT, 0644);
821 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
822 if (fd < 0 || fstat(fd, &st) < 0) {
823 err(1, "Unable to open/create %s", fname);
826 if (S_ISREG(st.st_mode)) {
827 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
828 errx(1, "Disk image %s is already "
834 info = &DiskInfo[DiskNum];
840 memcpy(info->fname, fname, l);
843 if ((info->serno = malloc(SERNOLEN)) != NULL)
844 strlcpy(info->serno, serno, SERNOLEN);
846 warnx("Couldn't allocate memory for the operation");
850 if (type == VKD_CD) {
851 rootdevnames[0] = "cd9660:vcd0";
852 } else if (type == VKD_DISK) {
853 rootdevnames[0] = "ufs:vkd0s0a";
854 rootdevnames[1] = "ufs:vkd0s1a";
860 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
868 netif_set_tapflags(int tap_unit, int f, int s)
873 bzero(&ifr, sizeof(ifr));
875 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
876 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
877 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
884 * If the flags are already set/cleared, then we return
885 * immediately to avoid extra syscalls
887 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
891 if ((flags & f) == 0)
902 * Fix up ifreq.ifr_name, since it may be trashed
903 * in previous ioctl(SIOCGIFFLAGS)
905 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
907 ifr.ifr_flags = flags & 0xffff;
908 ifr.ifr_flagshigh = flags >> 16;
909 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
910 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
918 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
920 struct ifaliasreq ifra;
921 struct sockaddr_in *in;
923 bzero(&ifra, sizeof(ifra));
924 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
927 in = (struct sockaddr_in *)&ifra.ifra_addr;
928 in->sin_family = AF_INET;
929 in->sin_len = sizeof(*in);
930 in->sin_addr.s_addr = addr;
934 in = (struct sockaddr_in *)&ifra.ifra_mask;
935 in->sin_len = sizeof(*in);
936 in->sin_addr.s_addr = mask;
939 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
940 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
948 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
953 bzero(&ifbr, sizeof(ifbr));
954 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
957 bzero(&ifd, sizeof(ifd));
958 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
959 ifd.ifd_cmd = BRDGADD;
960 ifd.ifd_len = sizeof(ifbr);
961 ifd.ifd_data = &ifbr;
963 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
965 * 'errno == EEXIST' means that the tap(4) is already
966 * a member of the bridge(4)
968 if (errno != EEXIST) {
969 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
976 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
979 * Locate the first unused tap(4) device file if auto mode is requested,
980 * or open the user supplied device file, and bring up the corresponding
983 * NOTE: Only tap(4) device file is supported currently
987 netif_open_tap(const char *netif, int *tap_unit, int s)
989 char tap_dev[MAXPATHLEN];
996 if (strcmp(netif, "auto") == 0) {
998 * Find first unused tap(4) device file
1000 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
1002 warnc(errno, "Unable to find a free tap(4)");
1007 * User supplied tap(4) device file or unix socket.
1009 if (netif[0] == '/') /* Absolute path */
1010 strlcpy(tap_dev, netif, sizeof(tap_dev));
1012 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
1014 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
1017 * If we cannot open normally try to connect to it.
1020 tap_fd = unix_connect(tap_dev);
1023 warn("Unable to open %s", tap_dev);
1029 * Check whether the device file is a tap(4)
1031 if (fstat(tap_fd, &st) < 0) {
1033 } else if (S_ISCHR(st.st_mode)) {
1034 dname = fdevname(tap_fd);
1036 dname = strstr(dname, "tap");
1039 * Bring up the corresponding tap(4) interface
1041 *tap_unit = strtol(dname + 3, NULL, 10);
1042 printf("TAP UNIT %d\n", *tap_unit);
1043 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1050 } else if (S_ISSOCK(st.st_mode)) {
1052 * Special socket connection (typically to vknet). We
1053 * do not have to do anything.
1061 warnx("%s is not a tap(4) device or socket", tap_dev);
1070 unix_connect(const char *path)
1072 struct sockaddr_un sunx;
1075 int sndbuf = 262144;
1078 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1079 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1080 ++len; /* include nul */
1081 sunx.sun_family = AF_UNIX;
1084 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1087 if (connect(net_fd, (void *)&sunx, len) < 0) {
1091 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1092 if (fstat(net_fd, &st) == 0)
1093 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1094 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1100 #undef TAPDEV_OFLAGS
1103 * Following syntax is supported,
1104 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1106 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1107 * tap(4)'s netmask len is z
1109 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1110 * pseudo netif's address is y.y.y.y
1112 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1113 * pseudo netif's address is y.y.y.y
1114 * tap(4) and pseudo netif's netmask len are z
1116 * 5) bridgeX tap(4) will be added to bridgeX
1118 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1119 * pseudo netif's address is y.y.y.y
1121 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1122 * pseudo netif's address is y.y.y.y
1123 * pseudo netif's netmask len is z
1127 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1129 in_addr_t tap_addr, netmask, netif_addr;
1130 int next_netif_addr;
1131 char *tok, *masklen_str, *ifbridge;
1136 tok = strtok(NULL, ":/");
1139 * Nothing special, simply use tap(4) as backend
1144 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1146 * tap(4)'s address is supplied
1151 * If there is next token, then it may be pseudo
1152 * netif's address or netmask len for tap(4)
1154 next_netif_addr = 0;
1157 * Not tap(4)'s address, assume it as a bridge(4)
1164 * If there is next token, then it must be pseudo
1167 next_netif_addr = 1;
1170 netmask = netif_addr = 0;
1172 tok = strtok(NULL, ":/");
1176 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1177 if (next_netif_addr) {
1178 warnx("Invalid pseudo netif address: %s", tok);
1184 * Current token is not address, then it must be netmask len
1189 * Current token is pseudo netif address, if there is next token
1190 * it must be netmask len
1192 masklen_str = strtok(NULL, "/");
1195 /* Calculate netmask */
1196 if (masklen_str != NULL) {
1199 masklen = strtoul(masklen_str, NULL, 10);
1200 if (masklen < 32 && masklen > 0) {
1201 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1204 warnx("Invalid netmask len: %lu", masklen);
1209 /* Make sure there is no more token left */
1210 if (strtok(NULL, ":/") != NULL) {
1211 warnx("Invalid argument to '-I'");
1218 } else if (ifbridge == NULL) {
1219 /* Set tap(4) address/netmask */
1220 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1223 /* Tie tap(4) to bridge(4) */
1224 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1234 * NetifInfo[] will be filled for pseudo netif initialization.
1235 * NetifNum will be bumped to reflect the number of valid entries
1240 init_netif(char *netifExp[], int netifExpNum)
1245 if (netifExpNum == 0)
1248 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1252 for (i = 0; i < netifExpNum; ++i) {
1253 struct vknetif_info *info;
1254 in_addr_t netif_addr, netif_mask;
1255 int tap_fd, tap_unit;
1258 /* Extract MAC address if there is one */
1262 netif = strtok(netifExp[i], ":");
1263 if (netif == NULL) {
1264 warnx("Invalid argument to '-I'");
1269 * Open tap(4) device file and bring up the
1270 * corresponding interface
1272 tap_fd = netif_open_tap(netif, &tap_unit, s);
1277 * Initialize tap(4) and get address/netmask
1280 * NB: Rest part of netifExp[i] is passed
1281 * to netif_init_tap() implicitly.
1283 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1285 * NB: Closing tap(4) device file will bring
1286 * down the corresponding interface
1292 info = &NetifInfo[NetifNum];
1293 bzero(info, sizeof(*info));
1294 info->tap_fd = tap_fd;
1295 info->tap_unit = tap_unit;
1296 info->netif_addr = netif_addr;
1297 info->netif_mask = netif_mask;
1299 * If tmp isn't NULL it means a MAC could have been
1300 * specified so attempt to convert it.
1301 * Setting enaddr to NULL will tell vke_attach() we
1302 * need a pseudo-random MAC address.
1305 if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL)
1306 warnx("Couldn't allocate memory for the operation");
1308 if ((kether_aton(tmp, info->enaddr)) == NULL) {
1310 info->enaddr = NULL;
1316 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1323 * Create the pid file and leave it open and locked while the vkernel is
1324 * running. This allows a script to use /usr/bin/lockf to probe whether
1325 * a vkernel is still running (so as not to accidently kill an unrelated
1326 * process from a stale pid file).
1335 if (pid_file != NULL) {
1336 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1337 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1339 if (errno == EWOULDBLOCK) {
1340 perror("Failed to lock pidfile, "
1341 "vkernel already running");
1343 perror("Failed to create pidfile");
1348 write(fd, buf, strlen(buf));
1349 /* leave the file open to maintain the lock */
1357 if (pid_file != NULL) {
1358 if (unlink(pid_file) < 0)
1359 perror("Warning: couldn't remove pidfile");
1365 usage_err(const char *ctl, ...)
1370 vfprintf(stderr, ctl, va);
1372 fprintf(stderr, "\n");
1378 usage_help(_Bool help)
1380 fprintf(stderr, "Usage: %s [-hsUvd] [-c file] [-e name=value:name=value:...]\n"
1381 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1382 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1383 "\t[-p file] [-r file]\n", save_av[0]);
1386 fprintf(stderr, "\nArguments:\n"
1387 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1388 "\t-e\tSpecify an environment to be used by the kernel.\n"
1389 "\t-h\tThis list of options.\n"
1390 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1391 "\t-I\tCreate a virtual network device.\n"
1392 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1393 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1394 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1395 "\t \t- numcpus - number of cpus\n"
1396 "\t \t- :lbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1397 "\t \tthe logical ID. Controls the number of threads/core (0bits - 1 thread, 1bit - 2 threads).\n"
1398 "\t \t- :cbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1399 "\t \tthe core ID. Controls the number of core/package (0bits - 1 core, 1bit - 2 cores).\n"
1400 "\t-p\tSpecify a file in which to store the process ID.\n"
1401 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n"
1402 "\t-s\tBoot into single-user mode.\n"
1403 "\t-U\tEnable writing to kernel memory and module loading.\n"
1404 "\t-v\tTurn on verbose booting.\n");
1412 kprintf("cpu reset, rebooting vkernel\n");
1415 execv(save_av[0], save_av);
1421 kprintf("cpu halt, exiting vkernel\n");
1429 switch(lwp_cpu_lock) {
1432 kprintf("Locking CPU%d to real cpu %d\n",
1434 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1436 if (next_cpu >= real_ncpus)
1439 case LCL_SINGLE_CPU:
1441 kprintf("Locking CPU%d to real cpu %d\n",
1443 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1446 /* do not map virtual cpus to real cpus */