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>
81 vm_paddr_t phys_avail[16];
83 vm_paddr_t Maxmem_bytes;
86 struct vkdisk_info DiskInfo[VKDISK_MAX];
88 struct vknetif_info NetifInfo[VKNETIF_MAX];
94 vm_offset_t virtual_start;
95 vm_offset_t virtual_end;
96 vm_offset_t virtual2_start;
97 vm_offset_t virtual2_end;
98 vm_offset_t kernel_vm_end;
99 vm_offset_t crashdumpmap;
100 vm_offset_t clean_sva;
101 vm_offset_t clean_eva;
102 struct msgbuf *msgbufp;
105 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
106 u_int cpu_feature; /* XXX */
108 int64_t tsc_frequency;
109 int optcpus; /* number of cpus - see mp_start() */
110 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
111 int real_ncpus; /* number of real CPUs */
112 int next_cpu; /* next real CPU to lock a virtual CPU to */
113 int vkernel_b_arg; /* -b argument - no of logical CPU bits - only SMP */
114 int vkernel_B_arg; /* -B argument - no of core bits - only SMP */
116 int via_feature_xcrypt = 0; /* XXX */
117 int via_feature_rng = 0; /* XXX */
119 struct privatespace *CPU_prvspace;
121 static struct trapframe proc0_tf;
122 static void *proc0paddr;
124 static void init_sys_memory(char *imageFile);
125 static void init_kern_memory(void);
126 static void init_globaldata(void);
127 static void init_vkernel(void);
128 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
129 static void init_netif(char *netifExp[], int netifFileNum);
130 static void writepid(void);
131 static void cleanpid(void);
132 static int unix_connect(const char *path);
133 static void usage_err(const char *ctl, ...);
134 static void usage_help(_Bool);
135 static void init_locks(void);
138 static char **save_av;
141 * Kernel startup for virtual kernels - standard main()
144 main(int ac, char **av)
146 char *memImageFile = NULL;
147 char *netifFile[VKNETIF_MAX];
148 char *diskFile[VKDISK_MAX];
149 char *cdFile[VKDISK_MAX];
154 int netifFileNum = 0;
157 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
165 int real_vkernel_enable;
180 kernel_mem_readonly = 1;
184 lwp_cpu_lock = LCL_NONE;
186 real_vkernel_enable = 0;
187 vsize = sizeof(real_vkernel_enable);
188 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
190 if (real_vkernel_enable == 0) {
191 errx(1, "vm.vkernel_enable is 0, must be set "
192 "to 1 to execute a vkernel!");
196 vsize = sizeof(real_ncpus);
197 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
202 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:Ub:B:")) != -1) {
206 * name=value:name=value:name=value...
209 * Allow values to be quoted but note that shells
210 * may remove the quotes, so using this feature
211 * to embed colons may require a backslash.
218 kern_envp = malloc(kenv_size);
219 if (kern_envp == NULL)
220 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
222 kenv_size2 = kenv_size + n + 1;
224 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
225 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
227 kenv_size = kenv_size2;
230 for (i = 0, j = pos; i < n; ++i) {
231 if (optarg[i] == '"')
233 else if (optarg[i] == '\'')
235 else if (isq == 0 && optarg[i] == ':')
238 kern_envp[j++] = optarg[i];
245 boothowto |= RB_SINGLE;
251 memImageFile = optarg;
254 if (netifFileNum < VKNETIF_MAX)
255 netifFile[netifFileNum++] = strdup(optarg);
260 if (diskFileNum + cdFileNum < VKDISK_MAX)
261 diskFile[diskFileNum++] = strdup(optarg);
266 if (diskFileNum + cdFileNum < VKDISK_MAX)
267 cdFile[cdFileNum++] = strdup(optarg);
270 Maxmem_bytes = strtoull(optarg, &suffix, 0);
287 usage_err("Bad maxmem option");
295 if (strncmp("map", optarg, 3) == 0) {
296 lwp_cpu_lock = LCL_PER_CPU;
297 if (optarg[3] == ',') {
298 next_cpu = strtol(optarg+4, &endp, 0);
300 usage_err("Bad target CPU number at '%s'", endp);
304 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
305 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
306 } else if (strncmp("any", optarg, 3) == 0) {
307 lwp_cpu_lock = LCL_NONE;
309 lwp_cpu_lock = LCL_SINGLE_CPU;
310 next_cpu = strtol(optarg, &endp, 0);
312 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);
319 * This value is set up by mp_start(), don't just
322 tok = strtok(optarg, ":");
323 optcpus = strtol(tok, NULL, 0);
324 if (optcpus < 1 || optcpus > MAXCPU)
325 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
327 /* :core_bits argument */
328 tok = strtok(NULL, ":");
330 vkernel_b_arg = strtol(tok, NULL, 0);
332 /* :logical_CPU_bits argument */
333 tok = strtok(NULL, ":");
335 vkernel_B_arg = strtol(tok, NULL, 0);
344 kernel_mem_readonly = 0;
356 init_sys_memory(memImageFile);
368 vsize = sizeof(tsc_present);
369 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
370 vsize = sizeof(tsc_frequency);
371 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
373 cpu_feature |= CPUID_TSC;
378 vsize = sizeof(supports_sse);
380 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
381 init_fpu(supports_sse);
383 cpu_feature |= CPUID_SSE | CPUID_FXSR;
386 * We boot from the first installed disk.
388 if (bootOnDisk == 1) {
389 init_disk(diskFile, diskFileNum, VKD_DISK);
390 init_disk(cdFile, cdFileNum, VKD_CD);
392 init_disk(cdFile, cdFileNum, VKD_CD);
393 init_disk(diskFile, diskFileNum, VKD_DISK);
395 init_netif(netifFile, netifFileNum);
403 * Initialize system memory. This is the virtual kernel's 'RAM'.
407 init_sys_memory(char *imageFile)
414 * Figure out the system memory image size. If an image file was
415 * specified and -m was not specified, use the image file's size.
417 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
418 Maxmem_bytes = (vm_paddr_t)st.st_size;
419 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
421 errx(1, "Cannot create new memory file %s unless "
422 "system memory size is specified with -m",
428 * Maxmem must be known at this time
430 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
431 errx(1, "Bad maxmem specification: 32MB minimum, "
432 "multiples of %dMB only",
433 SEG_SIZE / 1024 / 1024);
438 * Generate an image file name if necessary, then open/create the
439 * file exclusively locked. Do not allow multiple virtual kernels
440 * to use the same image file.
442 * Don't iterate through a million files if we do not have write
443 * access to the directory, stop if our open() failed on a
444 * non-existant file. Otherwise opens can fail for any number
445 * of reasons (lock failed, file not owned or writable by us, etc).
447 if (imageFile == NULL) {
448 for (i = 0; i < 1000000; ++i) {
449 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
451 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
452 if (fd < 0 && stat(imageFile, &st) == 0) {
459 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
461 printf("Using memory file: %s\n", imageFile);
462 if (fd < 0 || fstat(fd, &st) < 0) {
463 err(1, "Unable to open/create %s", imageFile);
468 * Truncate or extend the file as necessary. Clean out the contents
469 * of the file, we want it to be full of holes so we don't waste
470 * time reading in data from an old file that we no longer care
474 ftruncate(fd, Maxmem_bytes);
477 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
482 * Initialize pool tokens and other necessary locks
489 * Get the initial mplock with a count of 1 for the BSP.
490 * This uses a LOGICAL cpu ID, ie BSP == 0.
492 cpu_get_initial_mplock();
494 /* our token pool needs to work early */
495 lwkt_token_pool_init();
500 * Initialize kernel memory. This reserves kernel virtual memory by using
506 init_kern_memory(void)
512 char *topofstack = &dummy;
517 * Memory map our kernel virtual memory space. Note that the
518 * kernel image itself is not made part of this memory for the
521 * The memory map must be segment-aligned so we can properly
524 * If the system kernel has a different MAXDSIZ, it might not
525 * be possible to map kernel memory in its prefered location.
526 * Try a number of different locations.
528 try = (void *)0x40000000;
530 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
531 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
532 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
536 if (base != MAP_FAILED)
537 munmap(base, KERNEL_KVA_SIZE);
538 try = (char *)try + 0x10000000;
541 err(1, "Unable to mmap() kernel virtual memory!");
544 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
545 KvaStart = (vm_offset_t)base;
546 KvaSize = KERNEL_KVA_SIZE;
547 KvaEnd = KvaStart + KvaSize;
549 /* cannot use kprintf yet */
550 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
553 * Create a top-level page table self-mapping itself.
555 * Initialize the page directory at physical page index 0 to point
556 * to an array of page table pages starting at physical page index 1
558 lseek(MemImageFd, 0L, 0);
559 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
560 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
561 write(MemImageFd, &pte, sizeof(pte));
565 * Initialize the PTEs in the page table pages required to map the
566 * page table itself. This includes mapping the page directory page
567 * at the base so we go one more loop then normal.
569 lseek(MemImageFd, PAGE_SIZE, 0);
570 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
571 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
572 write(MemImageFd, &pte, sizeof(pte));
576 * Initialize remaining PTEs to 0. We may be reusing a memory image
577 * file. This is approximately a megabyte.
579 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
580 zero = malloc(PAGE_SIZE);
581 bzero(zero, PAGE_SIZE);
583 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
584 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
589 * Enable the page table and calculate pointers to our self-map
590 * for easy kernel page table manipulation.
592 * KernelPTA must be offset so we can do direct VA translations
594 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
595 0 | VPTE_R | VPTE_W | VPTE_V);
596 KernelPTD = (vpte_t *)base; /* pg directory */
597 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
598 KernelPTA -= KvaStart >> PAGE_SHIFT;
601 * phys_avail[] represents unallocated physical memory. MI code
602 * will use phys_avail[] to create the vm_page array.
604 phys_avail[0] = PAGE_SIZE +
605 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
606 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
607 phys_avail[1] = Maxmem_bytes;
610 * (virtual_start, virtual_end) represent unallocated kernel virtual
611 * memory. MI code will create kernel_map using these parameters.
613 virtual_start = KvaStart + PAGE_SIZE +
614 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
615 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
616 virtual_end = KvaStart + KERNEL_KVA_SIZE;
619 * kernel_vm_end could be set to virtual_end but we want some
620 * indication of how much of the kernel_map we've used, so
621 * set it low and let pmap_growkernel increase it even though we
622 * don't need to create any new page table pages.
624 kernel_vm_end = virtual_start;
627 * Allocate space for process 0's UAREA.
629 proc0paddr = (void *)virtual_start;
630 for (i = 0; i < UPAGES; ++i) {
631 pmap_kenter_quick(virtual_start, phys_avail[0]);
632 virtual_start += PAGE_SIZE;
633 phys_avail[0] += PAGE_SIZE;
639 crashdumpmap = virtual_start;
640 virtual_start += MAXDUMPPGS * PAGE_SIZE;
643 * msgbufp maps the system message buffer
645 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
646 msgbufp = (void *)virtual_start;
647 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
648 pmap_kenter_quick(virtual_start, phys_avail[0]);
649 virtual_start += PAGE_SIZE;
650 phys_avail[0] += PAGE_SIZE;
652 msgbufinit(msgbufp, MSGBUF_SIZE);
655 * used by kern_memio for /dev/mem access
657 ptvmmap = (caddr_t)virtual_start;
658 virtual_start += PAGE_SIZE;
661 * Bootstrap the kernel_pmap
667 * Map the per-cpu globaldata for cpu #0. Allocate the space using
668 * virtual_start and phys_avail[0]
672 init_globaldata(void)
679 * Reserve enough KVA to cover possible cpus. This is a considerable
680 * amount of KVA since the privatespace structure includes two
681 * whole page table mappings.
683 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
684 CPU_prvspace = (void *)virtual_start;
685 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
688 * Allocate enough physical memory to cover the mdglobaldata
689 * portion of the space and the idle stack and map the pages
690 * into KVA. For cpu #0 only.
692 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
694 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
695 pmap_kenter_quick(va, pa);
696 phys_avail[0] += PAGE_SIZE;
698 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
700 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
701 pmap_kenter_quick(va, pa);
702 phys_avail[0] += PAGE_SIZE;
706 * Setup the %fs for cpu #0. The mycpu macro works after this
707 * point. Note that %gs is used by pthreads.
709 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
713 * Initialize very low level systems including thread0, proc0, etc.
719 struct mdglobaldata *gd;
721 gd = &CPU_prvspace[0].mdglobaldata;
722 bzero(gd, sizeof(*gd));
724 gd->mi.gd_curthread = &thread0;
725 thread0.td_gd = &gd->mi;
727 ncpus2 = 1; /* rounded down power of 2 */
728 ncpus_fit = 1; /* rounded up power of 2 */
729 /* ncpus2_mask and ncpus_fit_mask are 0 */
731 gd->mi.gd_prvspace = &CPU_prvspace[0];
732 mi_gdinit(&gd->mi, 0);
734 mi_proc0init(&gd->mi, proc0paddr);
735 lwp0.lwp_md.md_regs = &proc0_tf;
740 #if 0 /* #ifdef DDB */
742 if (boothowto & RB_KDB)
743 Debugger("Boot flags requested debugger");
747 initializecpu(); /* Initialize CPU registers */
749 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
753 * Map the message buffer
755 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
756 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
757 msgbufinit(msgbufp, MSGBUF_SIZE);
760 thread0.td_pcb_cr3 ... MMU
761 lwp0.lwp_md.md_regs = &proc0_tf;
766 * Filesystem image paths for the virtual kernel are optional.
767 * If specified they each should point to a disk image,
768 * the first of which will become the root disk.
770 * The virtual kernel caches data from our 'disk' just like a normal kernel,
771 * so we do not really want the real kernel to cache the data too. Use
772 * O_DIRECT to remove the duplication.
776 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
780 if (diskFileNum == 0)
783 for(i=0; i < diskFileNum; i++){
788 warnx("Invalid argument to '-r'");
792 if (DiskNum < VKDISK_MAX) {
794 struct vkdisk_info* info = NULL;
798 if (type == VKD_DISK)
799 fd = open(fname, O_RDWR|O_DIRECT, 0644);
801 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
802 if (fd < 0 || fstat(fd, &st) < 0) {
803 err(1, "Unable to open/create %s", fname);
806 if (S_ISREG(st.st_mode)) {
807 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
808 errx(1, "Disk image %s is already "
814 info = &DiskInfo[DiskNum];
820 memcpy(info->fname, fname, l);
823 if (type == VKD_CD) {
824 rootdevnames[0] = "cd9660:vcd0a";
825 } else if (type == VKD_DISK) {
826 rootdevnames[0] = "ufs:vkd0s0a";
827 rootdevnames[1] = "ufs:vkd0s1a";
833 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
841 netif_set_tapflags(int tap_unit, int f, int s)
846 bzero(&ifr, sizeof(ifr));
848 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
849 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
850 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
857 * If the flags are already set/cleared, then we return
858 * immediately to avoid extra syscalls
860 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
864 if ((flags & f) == 0)
875 * Fix up ifreq.ifr_name, since it may be trashed
876 * in previous ioctl(SIOCGIFFLAGS)
878 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
880 ifr.ifr_flags = flags & 0xffff;
881 ifr.ifr_flagshigh = flags >> 16;
882 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
883 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
891 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
893 struct ifaliasreq ifra;
894 struct sockaddr_in *in;
896 bzero(&ifra, sizeof(ifra));
897 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
900 in = (struct sockaddr_in *)&ifra.ifra_addr;
901 in->sin_family = AF_INET;
902 in->sin_len = sizeof(*in);
903 in->sin_addr.s_addr = addr;
907 in = (struct sockaddr_in *)&ifra.ifra_mask;
908 in->sin_len = sizeof(*in);
909 in->sin_addr.s_addr = mask;
912 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
913 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
921 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
926 bzero(&ifbr, sizeof(ifbr));
927 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
930 bzero(&ifd, sizeof(ifd));
931 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
932 ifd.ifd_cmd = BRDGADD;
933 ifd.ifd_len = sizeof(ifbr);
934 ifd.ifd_data = &ifbr;
936 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
938 * 'errno == EEXIST' means that the tap(4) is already
939 * a member of the bridge(4)
941 if (errno != EEXIST) {
942 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
949 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
952 * Locate the first unused tap(4) device file if auto mode is requested,
953 * or open the user supplied device file, and bring up the corresponding
956 * NOTE: Only tap(4) device file is supported currently
960 netif_open_tap(const char *netif, int *tap_unit, int s)
962 char tap_dev[MAXPATHLEN];
969 if (strcmp(netif, "auto") == 0) {
971 * Find first unused tap(4) device file
973 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
975 warnc(errno, "Unable to find a free tap(4)");
980 * User supplied tap(4) device file or unix socket.
982 if (netif[0] == '/') /* Absolute path */
983 strlcpy(tap_dev, netif, sizeof(tap_dev));
985 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
987 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
990 * If we cannot open normally try to connect to it.
993 tap_fd = unix_connect(tap_dev);
996 warn("Unable to open %s", tap_dev);
1002 * Check whether the device file is a tap(4)
1004 if (fstat(tap_fd, &st) < 0) {
1006 } else if (S_ISCHR(st.st_mode)) {
1007 dname = fdevname(tap_fd);
1009 dname = strstr(dname, "tap");
1012 * Bring up the corresponding tap(4) interface
1014 *tap_unit = strtol(dname + 3, NULL, 10);
1015 printf("TAP UNIT %d\n", *tap_unit);
1016 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1023 } else if (S_ISSOCK(st.st_mode)) {
1025 * Special socket connection (typically to vknet). We
1026 * do not have to do anything.
1034 warnx("%s is not a tap(4) device or socket", tap_dev);
1043 unix_connect(const char *path)
1045 struct sockaddr_un sunx;
1048 int sndbuf = 262144;
1051 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1052 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1053 ++len; /* include nul */
1054 sunx.sun_family = AF_UNIX;
1057 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1060 if (connect(net_fd, (void *)&sunx, len) < 0) {
1064 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1065 if (fstat(net_fd, &st) == 0)
1066 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1067 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1073 #undef TAPDEV_OFLAGS
1076 * Following syntax is supported,
1077 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1079 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1080 * tap(4)'s netmask len is z
1082 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1083 * pseudo netif's address is y.y.y.y
1085 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1086 * pseudo netif's address is y.y.y.y
1087 * tap(4) and pseudo netif's netmask len are z
1089 * 5) bridgeX tap(4) will be added to bridgeX
1091 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1092 * pseudo netif's address is y.y.y.y
1094 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1095 * pseudo netif's address is y.y.y.y
1096 * pseudo netif's netmask len is z
1100 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1102 in_addr_t tap_addr, netmask, netif_addr;
1103 int next_netif_addr;
1104 char *tok, *masklen_str, *ifbridge;
1109 tok = strtok(NULL, ":/");
1112 * Nothing special, simply use tap(4) as backend
1117 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1119 * tap(4)'s address is supplied
1124 * If there is next token, then it may be pseudo
1125 * netif's address or netmask len for tap(4)
1127 next_netif_addr = 0;
1130 * Not tap(4)'s address, assume it as a bridge(4)
1137 * If there is next token, then it must be pseudo
1140 next_netif_addr = 1;
1143 netmask = netif_addr = 0;
1145 tok = strtok(NULL, ":/");
1149 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1150 if (next_netif_addr) {
1151 warnx("Invalid pseudo netif address: %s", tok);
1157 * Current token is not address, then it must be netmask len
1162 * Current token is pseudo netif address, if there is next token
1163 * it must be netmask len
1165 masklen_str = strtok(NULL, "/");
1168 /* Calculate netmask */
1169 if (masklen_str != NULL) {
1172 masklen = strtoul(masklen_str, NULL, 10);
1173 if (masklen < 32 && masklen > 0) {
1174 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1177 warnx("Invalid netmask len: %lu", masklen);
1182 /* Make sure there is no more token left */
1183 if (strtok(NULL, ":/") != NULL) {
1184 warnx("Invalid argument to '-I'");
1191 } else if (ifbridge == NULL) {
1192 /* Set tap(4) address/netmask */
1193 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1196 /* Tie tap(4) to bridge(4) */
1197 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1207 * NetifInfo[] will be filled for pseudo netif initialization.
1208 * NetifNum will be bumped to reflect the number of valid entries
1213 init_netif(char *netifExp[], int netifExpNum)
1217 if (netifExpNum == 0)
1220 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1224 for (i = 0; i < netifExpNum; ++i) {
1225 struct vknetif_info *info;
1226 in_addr_t netif_addr, netif_mask;
1227 int tap_fd, tap_unit;
1230 netif = strtok(netifExp[i], ":");
1231 if (netif == NULL) {
1232 warnx("Invalid argument to '-I'");
1237 * Open tap(4) device file and bring up the
1238 * corresponding interface
1240 tap_fd = netif_open_tap(netif, &tap_unit, s);
1245 * Initialize tap(4) and get address/netmask
1248 * NB: Rest part of netifExp[i] is passed
1249 * to netif_init_tap() implicitly.
1251 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1253 * NB: Closing tap(4) device file will bring
1254 * down the corresponding interface
1260 info = &NetifInfo[NetifNum];
1261 info->tap_fd = tap_fd;
1262 info->tap_unit = tap_unit;
1263 info->netif_addr = netif_addr;
1264 info->netif_mask = netif_mask;
1267 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1274 * Create the pid file and leave it open and locked while the vkernel is
1275 * running. This allows a script to use /usr/bin/lockf to probe whether
1276 * a vkernel is still running (so as not to accidently kill an unrelated
1277 * process from a stale pid file).
1286 if (pid_file != NULL) {
1287 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1288 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1290 if (errno == EWOULDBLOCK) {
1291 perror("Failed to lock pidfile, "
1292 "vkernel already running");
1294 perror("Failed to create pidfile");
1299 write(fd, buf, strlen(buf));
1300 /* leave the file open to maintain the lock */
1308 if (pid_file != NULL) {
1309 if (unlink(pid_file) < 0)
1310 perror("Warning: couldn't remove pidfile");
1316 usage_err(const char *ctl, ...)
1321 vfprintf(stderr, ctl, va);
1323 fprintf(stderr, "\n");
1329 usage_help(_Bool help)
1331 fprintf(stderr, "Usage: %s [-hsUv] [-c file] [-e name=value:name=value:...]\n"
1332 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1333 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1334 "\t[-p file] [-r file]\n", save_av[0]);
1337 fprintf(stderr, "\nArguments:\n"
1338 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1339 "\t-e\tSpecify an environment to be used by the kernel.\n"
1340 "\t-h\tThis list of options.\n"
1341 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1342 "\t-I\tCreate a virtual network device.\n"
1343 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1344 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1345 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1346 "\t \t- numcpus - number of cpus\n"
1347 "\t \t- :lbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1348 "\t \tthe logical ID. Controls the number of threads/core (0bits - 1 thread, 1bit - 2 threads).\n"
1349 "\t \t- :cbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1350 "\t \tthe core ID. Controls the number of core/package (0bits - 1 core, 1bit - 2 cores).\n"
1351 "\t-p\tSpecify a file in which to store the process ID.\n"
1352 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n"
1353 "\t-s\tBoot into single-user mode.\n"
1354 "\t-U\tEnable writing to kernel memory and module loading.\n"
1355 "\t-v\tTurn on verbose booting.\n");
1363 kprintf("cpu reset, rebooting vkernel\n");
1366 execv(save_av[0], save_av);
1372 kprintf("cpu halt, exiting vkernel\n");
1380 switch(lwp_cpu_lock) {
1383 kprintf("Locking CPU%d to real cpu %d\n",
1385 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1387 if (next_cpu >= real_ncpus)
1390 case LCL_SINGLE_CPU:
1392 kprintf("Locking CPU%d to real cpu %d\n",
1394 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1397 /* do not map virtual cpus to real cpus */