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 struct privatespace *CPU_prvspace;
112 static struct trapframe proc0_tf;
113 static void *proc0paddr;
115 static void init_sys_memory(char *imageFile);
116 static void init_kern_memory(void);
117 static void init_globaldata(void);
118 static void init_vkernel(void);
119 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
120 static void init_netif(char *netifExp[], int netifFileNum);
121 static void writepid( void );
122 static void cleanpid( void );
123 static int unix_connect(const char *path);
124 static void usage(const char *ctl, ...);
127 static char **save_av;
130 * Kernel startup for virtual kernels - standard main()
133 main(int ac, char **av)
135 char *memImageFile = NULL;
136 char *netifFile[VKNETIF_MAX];
137 char *diskFile[VKDISK_MAX];
138 char *cdFile[VKDISK_MAX];
141 int netifFileNum = 0;
144 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
150 int real_vkernel_enable;
160 kernel_mem_readonly = 1;
164 lwp_cpu_lock = LCL_NONE;
166 real_vkernel_enable = 0;
167 vsize = sizeof(real_vkernel_enable);
168 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
170 if (real_vkernel_enable == 0) {
171 errx(1, "vm.vkernel_enable is 0, must be set "
172 "to 1 to execute a vkernel!");
176 vsize = sizeof(real_ncpus);
177 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
179 while ((c = getopt(ac, av, "c:svl:m:n:r:e:i:p:I:U")) != -1) {
183 * name=value:name=value:name=value...
186 * Allow values to be quoted but note that shells
187 * may remove the quotes, so using this feature
188 * to embed colons may require a backslash.
192 kern_envp = malloc(n + 2);
193 for (i = j = 0; i < n; ++i) {
194 if (optarg[i] == '"')
196 else if (optarg[i] == '\'')
198 else if (isq == 0 && optarg[i] == ':')
201 kern_envp[j++] = optarg[i];
207 boothowto |= RB_SINGLE;
213 memImageFile = optarg;
216 if (netifFileNum < VKNETIF_MAX)
217 netifFile[netifFileNum++] = strdup(optarg);
222 if (diskFileNum + cdFileNum < VKDISK_MAX)
223 diskFile[diskFileNum++] = strdup(optarg);
228 if (diskFileNum + cdFileNum < VKDISK_MAX)
229 cdFile[cdFileNum++] = strdup(optarg);
232 Maxmem_bytes = strtoull(optarg, &suffix, 0);
249 usage("Bad maxmem option");
257 if (strncmp("map", optarg, 3) == 0) {
258 lwp_cpu_lock = LCL_PER_CPU;
259 if (optarg[3] == ',') {
260 next_cpu = strtol(optarg+4, &endp, 0);
262 usage("Bad target CPU number at '%s'", endp);
266 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
267 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
268 } else if (strncmp("any", optarg, 3) == 0) {
269 lwp_cpu_lock = LCL_NONE;
271 lwp_cpu_lock = LCL_SINGLE_CPU;
272 next_cpu = strtol(optarg, &endp, 0);
274 usage("Bad target CPU number at '%s'", endp);
275 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
276 usage("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
281 * This value is set up by mp_start(), don't just
285 optcpus = strtol(optarg, NULL, 0);
286 if (optcpus < 1 || optcpus > MAXCPU)
287 usage("Bad ncpus, valid range is 1-%d", MAXCPU);
289 if (strtol(optarg, NULL, 0) != 1) {
290 usage("You built a UP vkernel, only 1 cpu!");
299 kernel_mem_readonly = 0;
306 init_sys_memory(memImageFile);
316 vsize = sizeof(tsc_present);
317 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
318 vsize = sizeof(tsc_frequency);
319 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
321 cpu_feature |= CPUID_TSC;
326 vsize = sizeof(supports_sse);
328 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
329 init_fpu(supports_sse);
331 cpu_feature |= CPUID_SSE | CPUID_FXSR;
334 * We boot from the first installed disk.
336 if (bootOnDisk == 1) {
337 init_disk(diskFile, diskFileNum, VKD_DISK);
338 init_disk(cdFile, cdFileNum, VKD_CD);
340 init_disk(cdFile, cdFileNum, VKD_CD);
341 init_disk(diskFile, diskFileNum, VKD_DISK);
343 init_netif(netifFile, netifFileNum);
351 * Initialize system memory. This is the virtual kernel's 'RAM'.
355 init_sys_memory(char *imageFile)
362 * Figure out the system memory image size. If an image file was
363 * specified and -m was not specified, use the image file's size.
366 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
367 Maxmem_bytes = (vm_paddr_t)st.st_size;
368 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
370 err(1, "Cannot create new memory file %s unless "
371 "system memory size is specified with -m",
377 * Maxmem must be known at this time
379 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
380 err(1, "Bad maxmem specification: 32MB minimum, "
381 "multiples of %dMB only",
382 SEG_SIZE / 1024 / 1024);
387 * Generate an image file name if necessary, then open/create the
388 * file exclusively locked. Do not allow multiple virtual kernels
389 * to use the same image file.
391 if (imageFile == NULL) {
392 for (i = 0; i < 1000000; ++i) {
393 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
395 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
396 if (fd < 0 && errno == EWOULDBLOCK) {
403 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
405 printf("Using memory file: %s\n", imageFile);
406 if (fd < 0 || fstat(fd, &st) < 0) {
407 err(1, "Unable to open/create %s", imageFile);
412 * Truncate or extend the file as necessary.
414 if (st.st_size > Maxmem_bytes) {
415 ftruncate(fd, Maxmem_bytes);
416 } else if (st.st_size < Maxmem_bytes) {
418 off_t off = st.st_size & ~SEG_MASK;
420 kprintf("%s: Reserving blocks for memory image\n", imageFile);
421 zmem = malloc(SEG_SIZE);
422 bzero(zmem, SEG_SIZE);
423 lseek(fd, off, SEEK_SET);
424 while (off < Maxmem_bytes) {
425 if (write(fd, zmem, SEG_SIZE) != SEG_SIZE) {
426 err(1, "Unable to reserve blocks for memory image");
432 err(1, "Unable to reserve blocks for memory image");
436 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
440 * Initialize kernel memory. This reserves kernel virtual memory by using
446 init_kern_memory(void)
452 char *topofstack = &dummy;
457 * Memory map our kernel virtual memory space. Note that the
458 * kernel image itself is not made part of this memory for the
461 * The memory map must be segment-aligned so we can properly
464 * If the system kernel has a different MAXDSIZ, it might not
465 * be possible to map kernel memory in its prefered location.
466 * Try a number of different locations.
468 try = (void *)0x40000000;
470 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
471 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
472 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
476 if (base != MAP_FAILED)
477 munmap(base, KERNEL_KVA_SIZE);
478 try = (char *)try + 0x10000000;
481 err(1, "Unable to mmap() kernel virtual memory!");
484 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
485 KvaStart = (vm_offset_t)base;
486 KvaSize = KERNEL_KVA_SIZE;
487 KvaEnd = KvaStart + KvaSize;
488 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
491 * Create a top-level page table self-mapping itself.
493 * Initialize the page directory at physical page index 0 to point
494 * to an array of page table pages starting at physical page index 1
496 lseek(MemImageFd, 0L, 0);
497 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
498 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
499 write(MemImageFd, &pte, sizeof(pte));
503 * Initialize the PTEs in the page table pages required to map the
504 * page table itself. This includes mapping the page directory page
505 * at the base so we go one more loop then normal.
507 lseek(MemImageFd, PAGE_SIZE, 0);
508 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
509 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
510 write(MemImageFd, &pte, sizeof(pte));
514 * Initialize remaining PTEs to 0. We may be reusing a memory image
515 * file. This is approximately a megabyte.
517 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
518 zero = malloc(PAGE_SIZE);
519 bzero(zero, PAGE_SIZE);
521 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
522 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
527 * Enable the page table and calculate pointers to our self-map
528 * for easy kernel page table manipulation.
530 * KernelPTA must be offset so we can do direct VA translations
532 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
533 0 | VPTE_R | VPTE_W | VPTE_V);
534 KernelPTD = (vpte_t *)base; /* pg directory */
535 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
536 KernelPTA -= KvaStart >> PAGE_SHIFT;
539 * phys_avail[] represents unallocated physical memory. MI code
540 * will use phys_avail[] to create the vm_page array.
542 phys_avail[0] = PAGE_SIZE +
543 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
544 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
545 phys_avail[1] = Maxmem_bytes;
548 * (virtual_start, virtual_end) represent unallocated kernel virtual
549 * memory. MI code will create kernel_map using these parameters.
551 virtual_start = KvaStart + PAGE_SIZE +
552 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
553 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
554 virtual_end = KvaStart + KERNEL_KVA_SIZE;
557 * kernel_vm_end could be set to virtual_end but we want some
558 * indication of how much of the kernel_map we've used, so
559 * set it low and let pmap_growkernel increase it even though we
560 * don't need to create any new page table pages.
562 kernel_vm_end = virtual_start;
565 * Allocate space for process 0's UAREA.
567 proc0paddr = (void *)virtual_start;
568 for (i = 0; i < UPAGES; ++i) {
569 pmap_kenter_quick(virtual_start, phys_avail[0]);
570 virtual_start += PAGE_SIZE;
571 phys_avail[0] += PAGE_SIZE;
577 crashdumpmap = virtual_start;
578 virtual_start += MAXDUMPPGS * PAGE_SIZE;
581 * msgbufp maps the system message buffer
583 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
584 msgbufp = (void *)virtual_start;
585 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
586 pmap_kenter_quick(virtual_start, phys_avail[0]);
587 virtual_start += PAGE_SIZE;
588 phys_avail[0] += PAGE_SIZE;
590 msgbufinit(msgbufp, MSGBUF_SIZE);
593 * used by kern_memio for /dev/mem access
595 ptvmmap = (caddr_t)virtual_start;
596 virtual_start += PAGE_SIZE;
599 * Bootstrap the kernel_pmap
605 * Map the per-cpu globaldata for cpu #0. Allocate the space using
606 * virtual_start and phys_avail[0]
610 init_globaldata(void)
617 * Reserve enough KVA to cover possible cpus. This is a considerable
618 * amount of KVA since the privatespace structure includes two
619 * whole page table mappings.
621 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
622 CPU_prvspace = (void *)virtual_start;
623 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
626 * Allocate enough physical memory to cover the mdglobaldata
627 * portion of the space and the idle stack and map the pages
628 * into KVA. For cpu #0 only.
630 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
632 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
633 pmap_kenter_quick(va, pa);
634 phys_avail[0] += PAGE_SIZE;
636 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
638 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
639 pmap_kenter_quick(va, pa);
640 phys_avail[0] += PAGE_SIZE;
644 * Setup the %fs for cpu #0. The mycpu macro works after this
645 * point. Note that %gs is used by pthreads.
647 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
651 * Initialize very low level systems including thread0, proc0, etc.
657 struct mdglobaldata *gd;
659 gd = &CPU_prvspace[0].mdglobaldata;
660 bzero(gd, sizeof(*gd));
662 gd->mi.gd_curthread = &thread0;
663 thread0.td_gd = &gd->mi;
665 ncpus2 = 1; /* rounded down power of 2 */
666 ncpus_fit = 1; /* rounded up power of 2 */
667 /* ncpus2_mask and ncpus_fit_mask are 0 */
669 gd->mi.gd_prvspace = &CPU_prvspace[0];
670 mi_gdinit(&gd->mi, 0);
672 mi_proc0init(&gd->mi, proc0paddr);
673 lwp0.lwp_md.md_regs = &proc0_tf;
678 #if 0 /* #ifdef DDB */
680 if (boothowto & RB_KDB)
681 Debugger("Boot flags requested debugger");
685 initializecpu(); /* Initialize CPU registers */
687 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
691 * Map the message buffer
693 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
694 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
695 msgbufinit(msgbufp, MSGBUF_SIZE);
698 thread0.td_pcb_cr3 ... MMU
699 lwp0.lwp_md.md_regs = &proc0_tf;
704 * Filesystem image paths for the virtual kernel are optional.
705 * If specified they each should point to a disk image,
706 * the first of which will become the root disk.
708 * The virtual kernel caches data from our 'disk' just like a normal kernel,
709 * so we do not really want the real kernel to cache the data too. Use
710 * O_DIRECT to remove the duplication.
714 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
718 if (diskFileNum == 0)
721 for(i=0; i < diskFileNum; i++){
726 warnx("Invalid argument to '-r'");
730 if (DiskNum < VKDISK_MAX) {
732 struct vkdisk_info* info = NULL;
736 if (type == VKD_DISK)
737 fd = open(fname, O_RDWR|O_DIRECT|O_EXLOCK|O_NONBLOCK, 0644);
739 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
740 if (fd < 0 || fstat(fd, &st) < 0) {
742 fprintf(stderr, "You may already have a vkernel using this disk image!\n");
743 err(1, "Unable to open/create %s", fname);
746 /* get rid of O_NONBLOCK, keep O_DIRECT */
747 if (type == VKD_DISK)
748 fcntl(fd, F_SETFL, O_DIRECT);
750 info = &DiskInfo[DiskNum];
756 memcpy(info->fname, fname, l);
759 if (type == VKD_CD) {
760 rootdevnames[0] = "cd9660:vcd0a";
761 } else if (type == VKD_DISK) {
762 rootdevnames[0] = "ufs:vkd0s0a";
763 rootdevnames[1] = "ufs:vkd0s1a";
769 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
777 netif_set_tapflags(int tap_unit, int f, int s)
782 bzero(&ifr, sizeof(ifr));
784 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
785 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
786 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
793 * If the flags are already set/cleared, then we return
794 * immediately to avoid extra syscalls
796 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
800 if ((flags & f) == 0)
811 * Fix up ifreq.ifr_name, since it may be trashed
812 * in previous ioctl(SIOCGIFFLAGS)
814 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
816 ifr.ifr_flags = flags & 0xffff;
817 ifr.ifr_flagshigh = flags >> 16;
818 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
819 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
827 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
829 struct ifaliasreq ifra;
830 struct sockaddr_in *in;
832 bzero(&ifra, sizeof(ifra));
833 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
836 in = (struct sockaddr_in *)&ifra.ifra_addr;
837 in->sin_family = AF_INET;
838 in->sin_len = sizeof(*in);
839 in->sin_addr.s_addr = addr;
843 in = (struct sockaddr_in *)&ifra.ifra_mask;
844 in->sin_len = sizeof(*in);
845 in->sin_addr.s_addr = mask;
848 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
849 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
857 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
862 bzero(&ifbr, sizeof(ifbr));
863 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
866 bzero(&ifd, sizeof(ifd));
867 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
868 ifd.ifd_cmd = BRDGADD;
869 ifd.ifd_len = sizeof(ifbr);
870 ifd.ifd_data = &ifbr;
872 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
874 * 'errno == EEXIST' means that the tap(4) is already
875 * a member of the bridge(4)
877 if (errno != EEXIST) {
878 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
885 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
887 /* XXX major()/minor() can't be used in vkernel */
888 #define TAPDEV_MAJOR(x) ((int)(((u_int)(x) >> 8) & 0xff))
889 #define TAPDEV_MINOR(x) ((int)((x) & 0xffff00ff))
891 #ifndef TAP_CDEV_MAJOR
892 #define TAP_CDEV_MAJOR 149
896 * Locate the first unused tap(4) device file if auto mode is requested,
897 * or open the user supplied device file, and bring up the corresponding
900 * NOTE: Only tap(4) device file is supported currently
904 netif_open_tap(const char *netif, int *tap_unit, int s)
906 char tap_dev[MAXPATHLEN];
912 if (strcmp(netif, "auto") == 0) {
917 * Find first unused tap(4) device file
920 snprintf(tap_dev, sizeof(tap_dev), "/dev/tap%d", i);
921 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
922 if (tap_fd >= 0 || errno == ENOENT)
927 warnc(lasterr, "Unable to find a free tap(4)");
932 * User supplied tap(4) device file or unix socket.
934 if (netif[0] == '/') /* Absolute path */
935 strlcpy(tap_dev, netif, sizeof(tap_dev));
937 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
939 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
942 * If we cannot open normally try to connect to it.
945 tap_fd = unix_connect(tap_dev);
948 warn("Unable to open %s", tap_dev);
954 * Check whether the device file is a tap(4)
957 if (fstat(tap_fd, &st) == 0 && S_ISCHR(st.st_mode) &&
958 TAPDEV_MAJOR(st.st_rdev) == TAP_CDEV_MAJOR) {
959 *tap_unit = TAPDEV_MINOR(st.st_rdev);
962 * Bring up the corresponding tap(4) interface
964 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
966 } else if (S_ISSOCK(st.st_mode)) {
968 * Special socket connection (typically to vknet). We
969 * do not have to do anything.
973 warnx("%s is not a tap(4) device", tap_dev);
985 unix_connect(const char *path)
987 struct sockaddr_un sunx;
993 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
994 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
995 ++len; /* include nul */
996 sunx.sun_family = AF_UNIX;
999 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1002 if (connect(net_fd, (void *)&sunx, len) < 0) {
1006 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1007 if (fstat(net_fd, &st) == 0)
1008 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1009 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1015 #undef TAPDEV_OFLAGS
1018 * Following syntax is supported,
1019 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1021 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1022 * tap(4)'s netmask len is z
1024 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1025 * pseudo netif's address is y.y.y.y
1027 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1028 * pseudo netif's address is y.y.y.y
1029 * tap(4) and pseudo netif's netmask len are z
1031 * 5) bridgeX tap(4) will be added to bridgeX
1033 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1034 * pseudo netif's address is y.y.y.y
1036 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1037 * pseudo netif's address is y.y.y.y
1038 * pseudo netif's netmask len is z
1042 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1044 in_addr_t tap_addr, netmask, netif_addr;
1045 int next_netif_addr;
1046 char *tok, *masklen_str, *ifbridge;
1051 tok = strtok(NULL, ":/");
1054 * Nothing special, simply use tap(4) as backend
1059 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1061 * tap(4)'s address is supplied
1066 * If there is next token, then it may be pseudo
1067 * netif's address or netmask len for tap(4)
1069 next_netif_addr = 0;
1072 * Not tap(4)'s address, assume it as a bridge(4)
1079 * If there is next token, then it must be pseudo
1082 next_netif_addr = 1;
1085 netmask = netif_addr = 0;
1087 tok = strtok(NULL, ":/");
1091 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1092 if (next_netif_addr) {
1093 warnx("Invalid pseudo netif address: %s", tok);
1099 * Current token is not address, then it must be netmask len
1104 * Current token is pseudo netif address, if there is next token
1105 * it must be netmask len
1107 masklen_str = strtok(NULL, "/");
1110 /* Calculate netmask */
1111 if (masklen_str != NULL) {
1114 masklen = strtoul(masklen_str, NULL, 10);
1115 if (masklen < 32 && masklen > 0) {
1116 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1119 warnx("Invalid netmask len: %lu", masklen);
1124 /* Make sure there is no more token left */
1125 if (strtok(NULL, ":/") != NULL) {
1126 warnx("Invalid argument to '-I'");
1133 } else if (ifbridge == NULL) {
1134 /* Set tap(4) address/netmask */
1135 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1138 /* Tie tap(4) to bridge(4) */
1139 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1149 * NetifInfo[] will be filled for pseudo netif initialization.
1150 * NetifNum will be bumped to reflect the number of valid entries
1155 init_netif(char *netifExp[], int netifExpNum)
1159 if (netifExpNum == 0)
1162 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1166 for (i = 0; i < netifExpNum; ++i) {
1167 struct vknetif_info *info;
1168 in_addr_t netif_addr, netif_mask;
1169 int tap_fd, tap_unit;
1172 netif = strtok(netifExp[i], ":");
1173 if (netif == NULL) {
1174 warnx("Invalid argument to '-I'");
1179 * Open tap(4) device file and bring up the
1180 * corresponding interface
1182 tap_fd = netif_open_tap(netif, &tap_unit, s);
1187 * Initialize tap(4) and get address/netmask
1190 * NB: Rest part of netifExp[i] is passed
1191 * to netif_init_tap() implicitly.
1193 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1195 * NB: Closing tap(4) device file will bring
1196 * down the corresponding interface
1202 info = &NetifInfo[NetifNum];
1203 info->tap_fd = tap_fd;
1204 info->tap_unit = tap_unit;
1205 info->netif_addr = netif_addr;
1206 info->netif_mask = netif_mask;
1209 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1222 if (pid_file != NULL) {
1224 fp = fopen(pid_file, "w");
1227 fprintf(fp, "%ld\n", (long)self);
1231 perror("Warning: couldn't open pidfile");
1240 if (pid_file != NULL) {
1241 if ( unlink(pid_file) != 0 )
1242 perror("Warning: couldn't remove pidfile");
1248 usage(const char *ctl, ...)
1253 vfprintf(stderr, ctl, va);
1255 fprintf(stderr, "\n");
1262 kprintf("cpu reset, rebooting vkernel\n");
1265 execv(save_av[0], save_av);
1271 kprintf("cpu halt, exiting vkernel\n");
1279 switch(lwp_cpu_lock) {
1282 kprintf("Locking CPU%d to real cpu %d\n",
1284 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1286 if (next_cpu >= real_ncpus)
1289 case LCL_SINGLE_CPU:
1291 kprintf("Locking CPU%d to real cpu %d\n",
1293 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1296 /* do not map virtual cpus to real cpus */