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
34 * $DragonFly: src/sys/platform/vkernel/platform/init.c,v 1.56 2008/05/27 07:48:00 dillon Exp $
37 #include <sys/types.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
43 #include <sys/random.h>
44 #include <sys/vkernel.h>
46 #include <sys/reboot.h>
48 #include <sys/msgbuf.h>
49 #include <sys/vmspace.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/sysctl.h>
54 #include <vm/vm_page.h>
56 #include <machine/cpu.h>
57 #include <machine/globaldata.h>
58 #include <machine/tls.h>
59 #include <machine/md_var.h>
60 #include <machine/vmparam.h>
61 #include <cpu/specialreg.h>
64 #include <net/if_arp.h>
65 #include <net/ethernet.h>
66 #include <net/bridge/if_bridgevar.h>
67 #include <netinet/in.h>
68 #include <arpa/inet.h>
80 vm_paddr_t phys_avail[16];
82 vm_paddr_t Maxmem_bytes;
85 struct vkdisk_info DiskInfo[VKDISK_MAX];
87 struct vknetif_info NetifInfo[VKNETIF_MAX];
93 vm_offset_t virtual_start;
94 vm_offset_t virtual_end;
95 vm_offset_t virtual2_start;
96 vm_offset_t virtual2_end;
97 vm_offset_t kernel_vm_end;
98 vm_offset_t crashdumpmap;
99 vm_offset_t clean_sva;
100 vm_offset_t clean_eva;
101 struct msgbuf *msgbufp;
104 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
105 u_int cpu_feature; /* XXX */
107 int64_t tsc_frequency;
108 int optcpus; /* number of cpus - see mp_start() */
109 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
110 int real_ncpus; /* number of real CPUs */
111 int next_cpu; /* next real CPU to lock a virtual CPU to */
113 int via_feature_xcrypt = 0; /* XXX */
114 int via_feature_rng = 0; /* XXX */
116 struct privatespace *CPU_prvspace;
118 static struct trapframe proc0_tf;
119 static void *proc0paddr;
121 static void init_sys_memory(char *imageFile);
122 static void init_kern_memory(void);
123 static void init_globaldata(void);
124 static void init_vkernel(void);
125 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
126 static void init_netif(char *netifExp[], int netifFileNum);
127 static void writepid( void );
128 static void cleanpid( void );
129 static int unix_connect(const char *path);
130 static void usage(const char *ctl, ...);
133 static char **save_av;
136 * Kernel startup for virtual kernels - standard main()
139 main(int ac, char **av)
141 char *memImageFile = NULL;
142 char *netifFile[VKNETIF_MAX];
143 char *diskFile[VKDISK_MAX];
144 char *cdFile[VKDISK_MAX];
147 int netifFileNum = 0;
150 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
156 int real_vkernel_enable;
166 kernel_mem_readonly = 1;
170 lwp_cpu_lock = LCL_NONE;
172 real_vkernel_enable = 0;
173 vsize = sizeof(real_vkernel_enable);
174 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
176 if (real_vkernel_enable == 0) {
177 errx(1, "vm.vkernel_enable is 0, must be set "
178 "to 1 to execute a vkernel!");
182 vsize = sizeof(real_ncpus);
183 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
185 while ((c = getopt(ac, av, "c:svl:m:n:r:e:i:p:I:U")) != -1) {
189 * name=value:name=value:name=value...
192 * Allow values to be quoted but note that shells
193 * may remove the quotes, so using this feature
194 * to embed colons may require a backslash.
198 kern_envp = malloc(n + 2);
199 for (i = j = 0; i < n; ++i) {
200 if (optarg[i] == '"')
202 else if (optarg[i] == '\'')
204 else if (isq == 0 && optarg[i] == ':')
207 kern_envp[j++] = optarg[i];
213 boothowto |= RB_SINGLE;
219 memImageFile = optarg;
222 if (netifFileNum < VKNETIF_MAX)
223 netifFile[netifFileNum++] = strdup(optarg);
228 if (diskFileNum + cdFileNum < VKDISK_MAX)
229 diskFile[diskFileNum++] = strdup(optarg);
234 if (diskFileNum + cdFileNum < VKDISK_MAX)
235 cdFile[cdFileNum++] = strdup(optarg);
238 Maxmem_bytes = strtoull(optarg, &suffix, 0);
255 usage("Bad maxmem option");
263 if (strncmp("map", optarg, 3) == 0) {
264 lwp_cpu_lock = LCL_PER_CPU;
265 if (optarg[3] == ',') {
266 next_cpu = strtol(optarg+4, &endp, 0);
268 usage("Bad target CPU number at '%s'", endp);
272 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
273 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
274 } else if (strncmp("any", optarg, 3) == 0) {
275 lwp_cpu_lock = LCL_NONE;
277 lwp_cpu_lock = LCL_SINGLE_CPU;
278 next_cpu = strtol(optarg, &endp, 0);
280 usage("Bad target CPU number at '%s'", endp);
281 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
282 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
287 * This value is set up by mp_start(), don't just
291 optcpus = strtol(optarg, NULL, 0);
292 if (optcpus < 1 || optcpus > MAXCPU)
293 usage("Bad ncpus, valid range is 1-%d", MAXCPU);
295 if (strtol(optarg, NULL, 0) != 1) {
296 usage("You built a UP vkernel, only 1 cpu!");
305 kernel_mem_readonly = 0;
312 init_sys_memory(memImageFile);
322 vsize = sizeof(tsc_present);
323 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
324 vsize = sizeof(tsc_frequency);
325 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
327 cpu_feature |= CPUID_TSC;
332 vsize = sizeof(supports_sse);
334 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
335 init_fpu(supports_sse);
337 cpu_feature |= CPUID_SSE | CPUID_FXSR;
340 * We boot from the first installed disk.
342 if (bootOnDisk == 1) {
343 init_disk(diskFile, diskFileNum, VKD_DISK);
344 init_disk(cdFile, cdFileNum, VKD_CD);
346 init_disk(cdFile, cdFileNum, VKD_CD);
347 init_disk(diskFile, diskFileNum, VKD_DISK);
349 init_netif(netifFile, netifFileNum);
357 * Initialize system memory. This is the virtual kernel's 'RAM'.
361 init_sys_memory(char *imageFile)
368 * Figure out the system memory image size. If an image file was
369 * specified and -m was not specified, use the image file's size.
372 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
373 Maxmem_bytes = (vm_paddr_t)st.st_size;
374 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
376 err(1, "Cannot create new memory file %s unless "
377 "system memory size is specified with -m",
383 * Maxmem must be known at this time
385 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
386 err(1, "Bad maxmem specification: 32MB minimum, "
387 "multiples of %dMB only",
388 SEG_SIZE / 1024 / 1024);
393 * Generate an image file name if necessary, then open/create the
394 * file exclusively locked. Do not allow multiple virtual kernels
395 * to use the same image file.
397 if (imageFile == NULL) {
398 for (i = 0; i < 1000000; ++i) {
399 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
401 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
402 if (fd < 0 && errno == EWOULDBLOCK) {
409 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
411 printf("Using memory file: %s\n", imageFile);
412 if (fd < 0 || fstat(fd, &st) < 0) {
413 err(1, "Unable to open/create %s", imageFile);
418 * Truncate or extend the file as necessary. Clean out the contents
419 * of the file, we want it to be full of holes so we don't waste
420 * time reading in data from an old file that we no longer care
424 ftruncate(fd, Maxmem_bytes);
427 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
432 * Initialize kernel memory. This reserves kernel virtual memory by using
438 init_kern_memory(void)
444 char *topofstack = &dummy;
449 * Memory map our kernel virtual memory space. Note that the
450 * kernel image itself is not made part of this memory for the
453 * The memory map must be segment-aligned so we can properly
456 * If the system kernel has a different MAXDSIZ, it might not
457 * be possible to map kernel memory in its prefered location.
458 * Try a number of different locations.
460 try = (void *)0x40000000;
462 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
463 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
464 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
468 if (base != MAP_FAILED)
469 munmap(base, KERNEL_KVA_SIZE);
470 try = (char *)try + 0x10000000;
473 err(1, "Unable to mmap() kernel virtual memory!");
476 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
477 KvaStart = (vm_offset_t)base;
478 KvaSize = KERNEL_KVA_SIZE;
479 KvaEnd = KvaStart + KvaSize;
481 /* cannot use kprintf yet */
482 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
485 * Create a top-level page table self-mapping itself.
487 * Initialize the page directory at physical page index 0 to point
488 * to an array of page table pages starting at physical page index 1
490 lseek(MemImageFd, 0L, 0);
491 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
492 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
493 write(MemImageFd, &pte, sizeof(pte));
497 * Initialize the PTEs in the page table pages required to map the
498 * page table itself. This includes mapping the page directory page
499 * at the base so we go one more loop then normal.
501 lseek(MemImageFd, PAGE_SIZE, 0);
502 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
503 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
504 write(MemImageFd, &pte, sizeof(pte));
508 * Initialize remaining PTEs to 0. We may be reusing a memory image
509 * file. This is approximately a megabyte.
511 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
512 zero = malloc(PAGE_SIZE);
513 bzero(zero, PAGE_SIZE);
515 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
516 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
521 * Enable the page table and calculate pointers to our self-map
522 * for easy kernel page table manipulation.
524 * KernelPTA must be offset so we can do direct VA translations
526 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
527 0 | VPTE_R | VPTE_W | VPTE_V);
528 KernelPTD = (vpte_t *)base; /* pg directory */
529 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
530 KernelPTA -= KvaStart >> PAGE_SHIFT;
533 * phys_avail[] represents unallocated physical memory. MI code
534 * will use phys_avail[] to create the vm_page array.
536 phys_avail[0] = PAGE_SIZE +
537 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
538 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
539 phys_avail[1] = Maxmem_bytes;
542 * (virtual_start, virtual_end) represent unallocated kernel virtual
543 * memory. MI code will create kernel_map using these parameters.
545 virtual_start = KvaStart + PAGE_SIZE +
546 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
547 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
548 virtual_end = KvaStart + KERNEL_KVA_SIZE;
551 * kernel_vm_end could be set to virtual_end but we want some
552 * indication of how much of the kernel_map we've used, so
553 * set it low and let pmap_growkernel increase it even though we
554 * don't need to create any new page table pages.
556 kernel_vm_end = virtual_start;
559 * Allocate space for process 0's UAREA.
561 proc0paddr = (void *)virtual_start;
562 for (i = 0; i < UPAGES; ++i) {
563 pmap_kenter_quick(virtual_start, phys_avail[0]);
564 virtual_start += PAGE_SIZE;
565 phys_avail[0] += PAGE_SIZE;
571 crashdumpmap = virtual_start;
572 virtual_start += MAXDUMPPGS * PAGE_SIZE;
575 * msgbufp maps the system message buffer
577 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
578 msgbufp = (void *)virtual_start;
579 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
580 pmap_kenter_quick(virtual_start, phys_avail[0]);
581 virtual_start += PAGE_SIZE;
582 phys_avail[0] += PAGE_SIZE;
584 msgbufinit(msgbufp, MSGBUF_SIZE);
587 * used by kern_memio for /dev/mem access
589 ptvmmap = (caddr_t)virtual_start;
590 virtual_start += PAGE_SIZE;
593 * Bootstrap the kernel_pmap
599 * Map the per-cpu globaldata for cpu #0. Allocate the space using
600 * virtual_start and phys_avail[0]
604 init_globaldata(void)
611 * Reserve enough KVA to cover possible cpus. This is a considerable
612 * amount of KVA since the privatespace structure includes two
613 * whole page table mappings.
615 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
616 CPU_prvspace = (void *)virtual_start;
617 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
620 * Allocate enough physical memory to cover the mdglobaldata
621 * portion of the space and the idle stack and map the pages
622 * into KVA. For cpu #0 only.
624 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
626 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
627 pmap_kenter_quick(va, pa);
628 phys_avail[0] += PAGE_SIZE;
630 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
632 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
633 pmap_kenter_quick(va, pa);
634 phys_avail[0] += PAGE_SIZE;
638 * Setup the %fs for cpu #0. The mycpu macro works after this
639 * point. Note that %gs is used by pthreads.
641 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
645 * Initialize very low level systems including thread0, proc0, etc.
651 struct mdglobaldata *gd;
653 gd = &CPU_prvspace[0].mdglobaldata;
654 bzero(gd, sizeof(*gd));
656 gd->mi.gd_curthread = &thread0;
657 thread0.td_gd = &gd->mi;
659 ncpus2 = 1; /* rounded down power of 2 */
660 ncpus_fit = 1; /* rounded up power of 2 */
661 /* ncpus2_mask and ncpus_fit_mask are 0 */
663 gd->mi.gd_prvspace = &CPU_prvspace[0];
664 mi_gdinit(&gd->mi, 0);
666 mi_proc0init(&gd->mi, proc0paddr);
667 lwp0.lwp_md.md_regs = &proc0_tf;
672 #if 0 /* #ifdef DDB */
674 if (boothowto & RB_KDB)
675 Debugger("Boot flags requested debugger");
679 initializecpu(); /* Initialize CPU registers */
681 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
685 * Map the message buffer
687 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
688 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
689 msgbufinit(msgbufp, MSGBUF_SIZE);
692 thread0.td_pcb_cr3 ... MMU
693 lwp0.lwp_md.md_regs = &proc0_tf;
698 * Filesystem image paths for the virtual kernel are optional.
699 * If specified they each should point to a disk image,
700 * the first of which will become the root disk.
702 * The virtual kernel caches data from our 'disk' just like a normal kernel,
703 * so we do not really want the real kernel to cache the data too. Use
704 * O_DIRECT to remove the duplication.
708 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
712 if (diskFileNum == 0)
715 for(i=0; i < diskFileNum; i++){
720 warnx("Invalid argument to '-r'");
724 if (DiskNum < VKDISK_MAX) {
726 struct vkdisk_info* info = NULL;
730 if (type == VKD_DISK)
731 fd = open(fname, O_RDWR|O_DIRECT, 0644);
733 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
734 if (fd < 0 || fstat(fd, &st) < 0) {
735 err(1, "Unable to open/create %s", fname);
738 if (S_ISREG(st.st_mode)) {
739 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
740 errx(1, "Disk image %s is already "
746 info = &DiskInfo[DiskNum];
752 memcpy(info->fname, fname, l);
755 if (type == VKD_CD) {
756 rootdevnames[0] = "cd9660:vcd0a";
757 } else if (type == VKD_DISK) {
758 rootdevnames[0] = "ufs:vkd0s0a";
759 rootdevnames[1] = "ufs:vkd0s1a";
765 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
773 netif_set_tapflags(int tap_unit, int f, int s)
778 bzero(&ifr, sizeof(ifr));
780 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
781 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
782 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
789 * If the flags are already set/cleared, then we return
790 * immediately to avoid extra syscalls
792 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
796 if ((flags & f) == 0)
807 * Fix up ifreq.ifr_name, since it may be trashed
808 * in previous ioctl(SIOCGIFFLAGS)
810 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
812 ifr.ifr_flags = flags & 0xffff;
813 ifr.ifr_flagshigh = flags >> 16;
814 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
815 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
823 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
825 struct ifaliasreq ifra;
826 struct sockaddr_in *in;
828 bzero(&ifra, sizeof(ifra));
829 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
832 in = (struct sockaddr_in *)&ifra.ifra_addr;
833 in->sin_family = AF_INET;
834 in->sin_len = sizeof(*in);
835 in->sin_addr.s_addr = addr;
839 in = (struct sockaddr_in *)&ifra.ifra_mask;
840 in->sin_len = sizeof(*in);
841 in->sin_addr.s_addr = mask;
844 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
845 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
853 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
858 bzero(&ifbr, sizeof(ifbr));
859 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
862 bzero(&ifd, sizeof(ifd));
863 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
864 ifd.ifd_cmd = BRDGADD;
865 ifd.ifd_len = sizeof(ifbr);
866 ifd.ifd_data = &ifbr;
868 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
870 * 'errno == EEXIST' means that the tap(4) is already
871 * a member of the bridge(4)
873 if (errno != EEXIST) {
874 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
881 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
884 * Locate the first unused tap(4) device file if auto mode is requested,
885 * or open the user supplied device file, and bring up the corresponding
888 * NOTE: Only tap(4) device file is supported currently
892 netif_open_tap(const char *netif, int *tap_unit, int s)
894 char tap_dev[MAXPATHLEN];
901 if (strcmp(netif, "auto") == 0) {
903 * Find first unused tap(4) device file
905 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
907 warnc(errno, "Unable to find a free tap(4)");
912 * User supplied tap(4) device file or unix socket.
914 if (netif[0] == '/') /* Absolute path */
915 strlcpy(tap_dev, netif, sizeof(tap_dev));
917 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
919 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
922 * If we cannot open normally try to connect to it.
925 tap_fd = unix_connect(tap_dev);
928 warn("Unable to open %s", tap_dev);
934 * Check whether the device file is a tap(4)
936 if (fstat(tap_fd, &st) < 0) {
938 } else if (S_ISCHR(st.st_mode)) {
939 dname = fdevname(tap_fd);
941 dname = strstr(dname, "tap");
944 * Bring up the corresponding tap(4) interface
946 *tap_unit = strtol(dname + 3, NULL, 10);
947 printf("TAP UNIT %d\n", *tap_unit);
948 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
955 } else if (S_ISSOCK(st.st_mode)) {
957 * Special socket connection (typically to vknet). We
958 * do not have to do anything.
966 warnx("%s is not a tap(4) device or socket", tap_dev);
975 unix_connect(const char *path)
977 struct sockaddr_un sunx;
983 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
984 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
985 ++len; /* include nul */
986 sunx.sun_family = AF_UNIX;
989 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
992 if (connect(net_fd, (void *)&sunx, len) < 0) {
996 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
997 if (fstat(net_fd, &st) == 0)
998 printf("Network socket buffer: %d bytes\n", st.st_blksize);
999 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1005 #undef TAPDEV_OFLAGS
1008 * Following syntax is supported,
1009 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1011 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1012 * tap(4)'s netmask len is z
1014 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1015 * pseudo netif's address is y.y.y.y
1017 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1018 * pseudo netif's address is y.y.y.y
1019 * tap(4) and pseudo netif's netmask len are z
1021 * 5) bridgeX tap(4) will be added to bridgeX
1023 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1024 * pseudo netif's address is y.y.y.y
1026 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1027 * pseudo netif's address is y.y.y.y
1028 * pseudo netif's netmask len is z
1032 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1034 in_addr_t tap_addr, netmask, netif_addr;
1035 int next_netif_addr;
1036 char *tok, *masklen_str, *ifbridge;
1041 tok = strtok(NULL, ":/");
1044 * Nothing special, simply use tap(4) as backend
1049 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1051 * tap(4)'s address is supplied
1056 * If there is next token, then it may be pseudo
1057 * netif's address or netmask len for tap(4)
1059 next_netif_addr = 0;
1062 * Not tap(4)'s address, assume it as a bridge(4)
1069 * If there is next token, then it must be pseudo
1072 next_netif_addr = 1;
1075 netmask = netif_addr = 0;
1077 tok = strtok(NULL, ":/");
1081 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1082 if (next_netif_addr) {
1083 warnx("Invalid pseudo netif address: %s", tok);
1089 * Current token is not address, then it must be netmask len
1094 * Current token is pseudo netif address, if there is next token
1095 * it must be netmask len
1097 masklen_str = strtok(NULL, "/");
1100 /* Calculate netmask */
1101 if (masklen_str != NULL) {
1104 masklen = strtoul(masklen_str, NULL, 10);
1105 if (masklen < 32 && masklen > 0) {
1106 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1109 warnx("Invalid netmask len: %lu", masklen);
1114 /* Make sure there is no more token left */
1115 if (strtok(NULL, ":/") != NULL) {
1116 warnx("Invalid argument to '-I'");
1123 } else if (ifbridge == NULL) {
1124 /* Set tap(4) address/netmask */
1125 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1128 /* Tie tap(4) to bridge(4) */
1129 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1139 * NetifInfo[] will be filled for pseudo netif initialization.
1140 * NetifNum will be bumped to reflect the number of valid entries
1145 init_netif(char *netifExp[], int netifExpNum)
1149 if (netifExpNum == 0)
1152 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1156 for (i = 0; i < netifExpNum; ++i) {
1157 struct vknetif_info *info;
1158 in_addr_t netif_addr, netif_mask;
1159 int tap_fd, tap_unit;
1162 netif = strtok(netifExp[i], ":");
1163 if (netif == NULL) {
1164 warnx("Invalid argument to '-I'");
1169 * Open tap(4) device file and bring up the
1170 * corresponding interface
1172 tap_fd = netif_open_tap(netif, &tap_unit, s);
1177 * Initialize tap(4) and get address/netmask
1180 * NB: Rest part of netifExp[i] is passed
1181 * to netif_init_tap() implicitly.
1183 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1185 * NB: Closing tap(4) device file will bring
1186 * down the corresponding interface
1192 info = &NetifInfo[NetifNum];
1193 info->tap_fd = tap_fd;
1194 info->tap_unit = tap_unit;
1195 info->netif_addr = netif_addr;
1196 info->netif_mask = netif_mask;
1199 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1212 if (pid_file != NULL) {
1214 fp = fopen(pid_file, "w");
1217 fprintf(fp, "%ld\n", (long)self);
1221 perror("Warning: couldn't open pidfile");
1230 if (pid_file != NULL) {
1231 if ( unlink(pid_file) != 0 )
1232 perror("Warning: couldn't remove pidfile");
1238 usage(const char *ctl, ...)
1243 vfprintf(stderr, ctl, va);
1245 fprintf(stderr, "\n");
1252 kprintf("cpu reset, rebooting vkernel\n");
1255 execv(save_av[0], save_av);
1261 kprintf("cpu halt, exiting vkernel\n");
1269 switch(lwp_cpu_lock) {
1272 kprintf("Locking CPU%d to real cpu %d\n",
1274 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1276 if (next_cpu >= real_ncpus)
1279 case LCL_SINGLE_CPU:
1281 kprintf("Locking CPU%d to real cpu %d\n",
1283 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1286 /* do not map virtual cpus to real cpus */