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;
84 struct vkdisk_info DiskInfo[VKDISK_MAX];
86 struct vknetif_info NetifInfo[VKNETIF_MAX];
92 vm_offset_t virtual_start;
93 vm_offset_t virtual_end;
94 vm_offset_t kernel_vm_end;
95 vm_offset_t crashdumpmap;
96 vm_offset_t clean_sva;
97 vm_offset_t clean_eva;
98 struct msgbuf *msgbufp;
101 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
102 u_int cpu_feature; /* XXX */
104 int64_t tsc_frequency;
105 int optcpus; /* number of cpus - see mp_start() */
106 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
107 int real_ncpus; /* number of real CPUs */
108 int next_cpu; /* next real CPU to lock a virtual CPU to */
110 int via_feature_xcrypt = 0; /* XXX */
111 int via_feature_rng = 0; /* XXX */
113 struct privatespace *CPU_prvspace;
115 static struct trapframe proc0_tf;
116 static void *proc0paddr;
118 static void init_sys_memory(char *imageFile);
119 static void init_kern_memory(void);
120 static void init_globaldata(void);
121 static void init_vkernel(void);
122 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
123 static void init_netif(char *netifExp[], int netifFileNum);
124 static void writepid( void );
125 static void cleanpid( void );
126 static int unix_connect(const char *path);
127 static void usage(const char *ctl, ...);
130 static char **save_av;
133 * Kernel startup for virtual kernels - standard main()
136 main(int ac, char **av)
138 char *memImageFile = NULL;
139 char *netifFile[VKNETIF_MAX];
140 char *diskFile[VKDISK_MAX];
141 char *cdFile[VKDISK_MAX];
144 int netifFileNum = 0;
147 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
153 int real_vkernel_enable;
163 kernel_mem_readonly = 1;
167 lwp_cpu_lock = LCL_NONE;
169 real_vkernel_enable = 0;
170 vsize = sizeof(real_vkernel_enable);
171 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
173 if (real_vkernel_enable == 0) {
174 errx(1, "vm.vkernel_enable is 0, must be set "
175 "to 1 to execute a vkernel!");
179 vsize = sizeof(real_ncpus);
180 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
182 while ((c = getopt(ac, av, "c:svl:m:n:r:e:i:p:I:U")) != -1) {
186 * name=value:name=value:name=value...
189 * Allow values to be quoted but note that shells
190 * may remove the quotes, so using this feature
191 * to embed colons may require a backslash.
195 kern_envp = malloc(n + 2);
196 for (i = j = 0; i < n; ++i) {
197 if (optarg[i] == '"')
199 else if (optarg[i] == '\'')
201 else if (isq == 0 && optarg[i] == ':')
204 kern_envp[j++] = optarg[i];
210 boothowto |= RB_SINGLE;
216 memImageFile = optarg;
219 if (netifFileNum < VKNETIF_MAX)
220 netifFile[netifFileNum++] = strdup(optarg);
225 if (diskFileNum + cdFileNum < VKDISK_MAX)
226 diskFile[diskFileNum++] = strdup(optarg);
231 if (diskFileNum + cdFileNum < VKDISK_MAX)
232 cdFile[cdFileNum++] = strdup(optarg);
235 Maxmem_bytes = strtoull(optarg, &suffix, 0);
252 usage("Bad maxmem option");
260 if (strncmp("map", optarg, 3) == 0) {
261 lwp_cpu_lock = LCL_PER_CPU;
262 if (optarg[3] == ',') {
263 next_cpu = strtol(optarg+4, &endp, 0);
265 usage("Bad target CPU number at '%s'", endp);
269 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
270 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
271 } else if (strncmp("any", optarg, 3) == 0) {
272 lwp_cpu_lock = LCL_NONE;
274 lwp_cpu_lock = LCL_SINGLE_CPU;
275 next_cpu = strtol(optarg, &endp, 0);
277 usage("Bad target CPU number at '%s'", endp);
278 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
279 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
284 * This value is set up by mp_start(), don't just
288 optcpus = strtol(optarg, NULL, 0);
289 if (optcpus < 1 || optcpus > MAXCPU)
290 usage("Bad ncpus, valid range is 1-%d", MAXCPU);
292 if (strtol(optarg, NULL, 0) != 1) {
293 usage("You built a UP vkernel, only 1 cpu!");
302 kernel_mem_readonly = 0;
309 init_sys_memory(memImageFile);
319 vsize = sizeof(tsc_present);
320 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
321 vsize = sizeof(tsc_frequency);
322 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
324 cpu_feature |= CPUID_TSC;
329 vsize = sizeof(supports_sse);
331 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
332 init_fpu(supports_sse);
334 cpu_feature |= CPUID_SSE | CPUID_FXSR;
337 * We boot from the first installed disk.
339 if (bootOnDisk == 1) {
340 init_disk(diskFile, diskFileNum, VKD_DISK);
341 init_disk(cdFile, cdFileNum, VKD_CD);
343 init_disk(cdFile, cdFileNum, VKD_CD);
344 init_disk(diskFile, diskFileNum, VKD_DISK);
346 init_netif(netifFile, netifFileNum);
354 * Initialize system memory. This is the virtual kernel's 'RAM'.
358 init_sys_memory(char *imageFile)
365 * Figure out the system memory image size. If an image file was
366 * specified and -m was not specified, use the image file's size.
369 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
370 Maxmem_bytes = (vm_paddr_t)st.st_size;
371 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
373 err(1, "Cannot create new memory file %s unless "
374 "system memory size is specified with -m",
380 * Maxmem must be known at this time
382 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
383 err(1, "Bad maxmem specification: 32MB minimum, "
384 "multiples of %dMB only",
385 SEG_SIZE / 1024 / 1024);
390 * Generate an image file name if necessary, then open/create the
391 * file exclusively locked. Do not allow multiple virtual kernels
392 * to use the same image file.
394 if (imageFile == NULL) {
395 for (i = 0; i < 1000000; ++i) {
396 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
398 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
399 if (fd < 0 && errno == EWOULDBLOCK) {
406 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
408 printf("Using memory file: %s\n", imageFile);
409 if (fd < 0 || fstat(fd, &st) < 0) {
410 err(1, "Unable to open/create %s", imageFile);
415 * Truncate or extend the file as necessary.
417 if (st.st_size > Maxmem_bytes) {
418 ftruncate(fd, Maxmem_bytes);
419 } else if (st.st_size < Maxmem_bytes) {
421 off_t off = st.st_size & ~SEG_MASK;
423 kprintf("%s: Reserving blocks for memory image\n", imageFile);
424 zmem = malloc(SEG_SIZE);
425 bzero(zmem, SEG_SIZE);
426 lseek(fd, off, SEEK_SET);
427 while (off < Maxmem_bytes) {
428 if (write(fd, zmem, SEG_SIZE) != SEG_SIZE) {
429 err(1, "Unable to reserve blocks for memory image");
435 err(1, "Unable to reserve blocks for memory image");
439 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
443 * Initialize kernel memory. This reserves kernel virtual memory by using
449 init_kern_memory(void)
455 char *topofstack = &dummy;
460 * Memory map our kernel virtual memory space. Note that the
461 * kernel image itself is not made part of this memory for the
464 * The memory map must be segment-aligned so we can properly
467 * If the system kernel has a different MAXDSIZ, it might not
468 * be possible to map kernel memory in its prefered location.
469 * Try a number of different locations.
471 try = (void *)0x40000000;
473 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
474 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
475 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
479 if (base != MAP_FAILED)
480 munmap(base, KERNEL_KVA_SIZE);
481 try = (char *)try + 0x10000000;
484 err(1, "Unable to mmap() kernel virtual memory!");
487 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
488 KvaStart = (vm_offset_t)base;
489 KvaSize = KERNEL_KVA_SIZE;
490 KvaEnd = KvaStart + KvaSize;
491 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
494 * Create a top-level page table self-mapping itself.
496 * Initialize the page directory at physical page index 0 to point
497 * to an array of page table pages starting at physical page index 1
499 lseek(MemImageFd, 0L, 0);
500 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
501 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
502 write(MemImageFd, &pte, sizeof(pte));
506 * Initialize the PTEs in the page table pages required to map the
507 * page table itself. This includes mapping the page directory page
508 * at the base so we go one more loop then normal.
510 lseek(MemImageFd, PAGE_SIZE, 0);
511 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
512 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
513 write(MemImageFd, &pte, sizeof(pte));
517 * Initialize remaining PTEs to 0. We may be reusing a memory image
518 * file. This is approximately a megabyte.
520 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
521 zero = malloc(PAGE_SIZE);
522 bzero(zero, PAGE_SIZE);
524 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
525 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
530 * Enable the page table and calculate pointers to our self-map
531 * for easy kernel page table manipulation.
533 * KernelPTA must be offset so we can do direct VA translations
535 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
536 0 | VPTE_R | VPTE_W | VPTE_V);
537 KernelPTD = (vpte_t *)base; /* pg directory */
538 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
539 KernelPTA -= KvaStart >> PAGE_SHIFT;
542 * phys_avail[] represents unallocated physical memory. MI code
543 * will use phys_avail[] to create the vm_page array.
545 phys_avail[0] = PAGE_SIZE +
546 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
547 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
548 phys_avail[1] = Maxmem_bytes;
551 * (virtual_start, virtual_end) represent unallocated kernel virtual
552 * memory. MI code will create kernel_map using these parameters.
554 virtual_start = KvaStart + PAGE_SIZE +
555 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
556 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
557 virtual_end = KvaStart + KERNEL_KVA_SIZE;
560 * kernel_vm_end could be set to virtual_end but we want some
561 * indication of how much of the kernel_map we've used, so
562 * set it low and let pmap_growkernel increase it even though we
563 * don't need to create any new page table pages.
565 kernel_vm_end = virtual_start;
568 * Allocate space for process 0's UAREA.
570 proc0paddr = (void *)virtual_start;
571 for (i = 0; i < UPAGES; ++i) {
572 pmap_kenter_quick(virtual_start, phys_avail[0]);
573 virtual_start += PAGE_SIZE;
574 phys_avail[0] += PAGE_SIZE;
580 crashdumpmap = virtual_start;
581 virtual_start += MAXDUMPPGS * PAGE_SIZE;
584 * msgbufp maps the system message buffer
586 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
587 msgbufp = (void *)virtual_start;
588 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
589 pmap_kenter_quick(virtual_start, phys_avail[0]);
590 virtual_start += PAGE_SIZE;
591 phys_avail[0] += PAGE_SIZE;
593 msgbufinit(msgbufp, MSGBUF_SIZE);
596 * used by kern_memio for /dev/mem access
598 ptvmmap = (caddr_t)virtual_start;
599 virtual_start += PAGE_SIZE;
602 * Bootstrap the kernel_pmap
608 * Map the per-cpu globaldata for cpu #0. Allocate the space using
609 * virtual_start and phys_avail[0]
613 init_globaldata(void)
620 * Reserve enough KVA to cover possible cpus. This is a considerable
621 * amount of KVA since the privatespace structure includes two
622 * whole page table mappings.
624 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
625 CPU_prvspace = (void *)virtual_start;
626 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
629 * Allocate enough physical memory to cover the mdglobaldata
630 * portion of the space and the idle stack and map the pages
631 * into KVA. For cpu #0 only.
633 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
635 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
636 pmap_kenter_quick(va, pa);
637 phys_avail[0] += PAGE_SIZE;
639 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
641 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
642 pmap_kenter_quick(va, pa);
643 phys_avail[0] += PAGE_SIZE;
647 * Setup the %fs for cpu #0. The mycpu macro works after this
648 * point. Note that %gs is used by pthreads.
650 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
654 * Initialize very low level systems including thread0, proc0, etc.
660 struct mdglobaldata *gd;
662 gd = &CPU_prvspace[0].mdglobaldata;
663 bzero(gd, sizeof(*gd));
665 gd->mi.gd_curthread = &thread0;
666 thread0.td_gd = &gd->mi;
668 ncpus2 = 1; /* rounded down power of 2 */
669 ncpus_fit = 1; /* rounded up power of 2 */
670 /* ncpus2_mask and ncpus_fit_mask are 0 */
672 gd->mi.gd_prvspace = &CPU_prvspace[0];
673 mi_gdinit(&gd->mi, 0);
675 mi_proc0init(&gd->mi, proc0paddr);
676 lwp0.lwp_md.md_regs = &proc0_tf;
681 #if 0 /* #ifdef DDB */
683 if (boothowto & RB_KDB)
684 Debugger("Boot flags requested debugger");
688 initializecpu(); /* Initialize CPU registers */
690 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
694 * Map the message buffer
696 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
697 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
698 msgbufinit(msgbufp, MSGBUF_SIZE);
701 thread0.td_pcb_cr3 ... MMU
702 lwp0.lwp_md.md_regs = &proc0_tf;
707 * Filesystem image paths for the virtual kernel are optional.
708 * If specified they each should point to a disk image,
709 * the first of which will become the root disk.
711 * The virtual kernel caches data from our 'disk' just like a normal kernel,
712 * so we do not really want the real kernel to cache the data too. Use
713 * O_DIRECT to remove the duplication.
717 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
721 if (diskFileNum == 0)
724 for(i=0; i < diskFileNum; i++){
729 warnx("Invalid argument to '-r'");
733 if (DiskNum < VKDISK_MAX) {
735 struct vkdisk_info* info = NULL;
739 if (type == VKD_DISK)
740 fd = open(fname, O_RDWR|O_DIRECT|O_EXLOCK|O_NONBLOCK, 0644);
742 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
743 if (fd < 0 || fstat(fd, &st) < 0) {
745 fprintf(stderr, "You may already have a vkernel using this disk image!\n");
746 err(1, "Unable to open/create %s", fname);
749 /* get rid of O_NONBLOCK, keep O_DIRECT */
750 if (type == VKD_DISK)
751 fcntl(fd, F_SETFL, O_DIRECT);
753 info = &DiskInfo[DiskNum];
759 memcpy(info->fname, fname, l);
762 if (type == VKD_CD) {
763 rootdevnames[0] = "cd9660:vcd0a";
764 } else if (type == VKD_DISK) {
765 rootdevnames[0] = "ufs:vkd0s0a";
766 rootdevnames[1] = "ufs:vkd0s1a";
772 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
780 netif_set_tapflags(int tap_unit, int f, int s)
785 bzero(&ifr, sizeof(ifr));
787 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
788 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
789 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
796 * If the flags are already set/cleared, then we return
797 * immediately to avoid extra syscalls
799 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
803 if ((flags & f) == 0)
814 * Fix up ifreq.ifr_name, since it may be trashed
815 * in previous ioctl(SIOCGIFFLAGS)
817 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
819 ifr.ifr_flags = flags & 0xffff;
820 ifr.ifr_flagshigh = flags >> 16;
821 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
822 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
830 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
832 struct ifaliasreq ifra;
833 struct sockaddr_in *in;
835 bzero(&ifra, sizeof(ifra));
836 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
839 in = (struct sockaddr_in *)&ifra.ifra_addr;
840 in->sin_family = AF_INET;
841 in->sin_len = sizeof(*in);
842 in->sin_addr.s_addr = addr;
846 in = (struct sockaddr_in *)&ifra.ifra_mask;
847 in->sin_len = sizeof(*in);
848 in->sin_addr.s_addr = mask;
851 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
852 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
860 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
865 bzero(&ifbr, sizeof(ifbr));
866 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
869 bzero(&ifd, sizeof(ifd));
870 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
871 ifd.ifd_cmd = BRDGADD;
872 ifd.ifd_len = sizeof(ifbr);
873 ifd.ifd_data = &ifbr;
875 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
877 * 'errno == EEXIST' means that the tap(4) is already
878 * a member of the bridge(4)
880 if (errno != EEXIST) {
881 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
888 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
891 * Locate the first unused tap(4) device file if auto mode is requested,
892 * or open the user supplied device file, and bring up the corresponding
895 * NOTE: Only tap(4) device file is supported currently
899 netif_open_tap(const char *netif, int *tap_unit, int s)
901 char tap_dev[MAXPATHLEN];
908 if (strcmp(netif, "auto") == 0) {
910 * Find first unused tap(4) device file
912 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
914 warnc(errno, "Unable to find a free tap(4)");
919 * User supplied tap(4) device file or unix socket.
921 if (netif[0] == '/') /* Absolute path */
922 strlcpy(tap_dev, netif, sizeof(tap_dev));
924 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
926 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
929 * If we cannot open normally try to connect to it.
932 tap_fd = unix_connect(tap_dev);
935 warn("Unable to open %s", tap_dev);
941 * Check whether the device file is a tap(4)
943 if (fstat(tap_fd, &st) < 0) {
945 } else if (S_ISCHR(st.st_mode)) {
946 dname = fdevname(tap_fd);
948 dname = strstr(dname, "tap");
951 * Bring up the corresponding tap(4) interface
953 *tap_unit = strtol(dname + 3, NULL, 10);
954 printf("TAP UNIT %d\n", *tap_unit);
955 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
961 } else if (S_ISSOCK(st.st_mode)) {
963 * Special socket connection (typically to vknet). We
964 * do not have to do anything.
972 warnx("%s is not a tap(4) device or socket", tap_dev);
981 unix_connect(const char *path)
983 struct sockaddr_un sunx;
989 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
990 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
991 ++len; /* include nul */
992 sunx.sun_family = AF_UNIX;
995 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
998 if (connect(net_fd, (void *)&sunx, len) < 0) {
1002 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1003 if (fstat(net_fd, &st) == 0)
1004 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1005 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1011 #undef TAPDEV_OFLAGS
1014 * Following syntax is supported,
1015 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1017 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1018 * tap(4)'s netmask len is z
1020 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1021 * pseudo netif's address is y.y.y.y
1023 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1024 * pseudo netif's address is y.y.y.y
1025 * tap(4) and pseudo netif's netmask len are z
1027 * 5) bridgeX tap(4) will be added to bridgeX
1029 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1030 * pseudo netif's address is y.y.y.y
1032 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1033 * pseudo netif's address is y.y.y.y
1034 * pseudo netif's netmask len is z
1038 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1040 in_addr_t tap_addr, netmask, netif_addr;
1041 int next_netif_addr;
1042 char *tok, *masklen_str, *ifbridge;
1047 tok = strtok(NULL, ":/");
1050 * Nothing special, simply use tap(4) as backend
1055 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1057 * tap(4)'s address is supplied
1062 * If there is next token, then it may be pseudo
1063 * netif's address or netmask len for tap(4)
1065 next_netif_addr = 0;
1068 * Not tap(4)'s address, assume it as a bridge(4)
1075 * If there is next token, then it must be pseudo
1078 next_netif_addr = 1;
1081 netmask = netif_addr = 0;
1083 tok = strtok(NULL, ":/");
1087 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1088 if (next_netif_addr) {
1089 warnx("Invalid pseudo netif address: %s", tok);
1095 * Current token is not address, then it must be netmask len
1100 * Current token is pseudo netif address, if there is next token
1101 * it must be netmask len
1103 masklen_str = strtok(NULL, "/");
1106 /* Calculate netmask */
1107 if (masklen_str != NULL) {
1110 masklen = strtoul(masklen_str, NULL, 10);
1111 if (masklen < 32 && masklen > 0) {
1112 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1115 warnx("Invalid netmask len: %lu", masklen);
1120 /* Make sure there is no more token left */
1121 if (strtok(NULL, ":/") != NULL) {
1122 warnx("Invalid argument to '-I'");
1129 } else if (ifbridge == NULL) {
1130 /* Set tap(4) address/netmask */
1131 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1134 /* Tie tap(4) to bridge(4) */
1135 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1145 * NetifInfo[] will be filled for pseudo netif initialization.
1146 * NetifNum will be bumped to reflect the number of valid entries
1151 init_netif(char *netifExp[], int netifExpNum)
1155 if (netifExpNum == 0)
1158 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1162 for (i = 0; i < netifExpNum; ++i) {
1163 struct vknetif_info *info;
1164 in_addr_t netif_addr, netif_mask;
1165 int tap_fd, tap_unit;
1168 netif = strtok(netifExp[i], ":");
1169 if (netif == NULL) {
1170 warnx("Invalid argument to '-I'");
1175 * Open tap(4) device file and bring up the
1176 * corresponding interface
1178 tap_fd = netif_open_tap(netif, &tap_unit, s);
1183 * Initialize tap(4) and get address/netmask
1186 * NB: Rest part of netifExp[i] is passed
1187 * to netif_init_tap() implicitly.
1189 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1191 * NB: Closing tap(4) device file will bring
1192 * down the corresponding interface
1198 info = &NetifInfo[NetifNum];
1199 info->tap_fd = tap_fd;
1200 info->tap_unit = tap_unit;
1201 info->netif_addr = netif_addr;
1202 info->netif_mask = netif_mask;
1205 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1218 if (pid_file != NULL) {
1220 fp = fopen(pid_file, "w");
1223 fprintf(fp, "%ld\n", (long)self);
1227 perror("Warning: couldn't open pidfile");
1236 if (pid_file != NULL) {
1237 if ( unlink(pid_file) != 0 )
1238 perror("Warning: couldn't remove pidfile");
1244 usage(const char *ctl, ...)
1249 vfprintf(stderr, ctl, va);
1251 fprintf(stderr, "\n");
1258 kprintf("cpu reset, rebooting vkernel\n");
1261 execv(save_av[0], save_av);
1267 kprintf("cpu halt, exiting vkernel\n");
1275 switch(lwp_cpu_lock) {
1278 kprintf("Locking CPU%d to real cpu %d\n",
1280 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1282 if (next_cpu >= real_ncpus)
1285 case LCL_SINGLE_CPU:
1287 kprintf("Locking CPU%d to real cpu %d\n",
1289 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1292 /* do not map virtual cpus to real cpus */