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>
55 #include <sys/mplock2.h>
57 #include <machine/cpu.h>
58 #include <machine/globaldata.h>
59 #include <machine/tls.h>
60 #include <machine/md_var.h>
61 #include <machine/vmparam.h>
62 #include <cpu/specialreg.h>
65 #include <net/if_arp.h>
66 #include <net/ethernet.h>
67 #include <net/bridge/if_bridgevar.h>
68 #include <netinet/in.h>
69 #include <arpa/inet.h>
83 vm_paddr_t phys_avail[16];
85 vm_paddr_t Maxmem_bytes;
88 struct vkdisk_info DiskInfo[VKDISK_MAX];
90 struct vknetif_info NetifInfo[VKNETIF_MAX];
96 vm_offset_t virtual_start;
97 vm_offset_t virtual_end;
98 vm_offset_t virtual2_start;
99 vm_offset_t virtual2_end;
100 vm_offset_t kernel_vm_end;
101 vm_offset_t crashdumpmap;
102 vm_offset_t clean_sva;
103 vm_offset_t clean_eva;
104 struct msgbuf *msgbufp;
107 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
108 u_int cpu_feature; /* XXX */
110 int64_t tsc_frequency;
111 int optcpus; /* number of cpus - see mp_start() */
112 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
113 int real_ncpus; /* number of real CPUs */
114 int next_cpu; /* next real CPU to lock a virtual CPU to */
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);
137 static char **save_av;
140 * Kernel startup for virtual kernels - standard main()
143 main(int ac, char **av)
145 char *memImageFile = NULL;
146 char *netifFile[VKNETIF_MAX];
147 char *diskFile[VKDISK_MAX];
148 char *cdFile[VKDISK_MAX];
151 int netifFileNum = 0;
154 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
160 int real_vkernel_enable;
170 kernel_mem_readonly = 1;
174 lwp_cpu_lock = LCL_NONE;
176 real_vkernel_enable = 0;
177 vsize = sizeof(real_vkernel_enable);
178 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
180 if (real_vkernel_enable == 0) {
181 errx(1, "vm.vkernel_enable is 0, must be set "
182 "to 1 to execute a vkernel!");
186 vsize = sizeof(real_ncpus);
187 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
192 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:U")) != -1) {
196 * name=value:name=value:name=value...
199 * Allow values to be quoted but note that shells
200 * may remove the quotes, so using this feature
201 * to embed colons may require a backslash.
205 kern_envp = malloc(n + 2);
206 for (i = j = 0; i < n; ++i) {
207 if (optarg[i] == '"')
209 else if (optarg[i] == '\'')
211 else if (isq == 0 && optarg[i] == ':')
214 kern_envp[j++] = optarg[i];
220 boothowto |= RB_SINGLE;
226 memImageFile = optarg;
229 if (netifFileNum < VKNETIF_MAX)
230 netifFile[netifFileNum++] = strdup(optarg);
235 if (diskFileNum + cdFileNum < VKDISK_MAX)
236 diskFile[diskFileNum++] = strdup(optarg);
241 if (diskFileNum + cdFileNum < VKDISK_MAX)
242 cdFile[cdFileNum++] = strdup(optarg);
245 Maxmem_bytes = strtoull(optarg, &suffix, 0);
262 usage_err("Bad maxmem option");
270 if (strncmp("map", optarg, 3) == 0) {
271 lwp_cpu_lock = LCL_PER_CPU;
272 if (optarg[3] == ',') {
273 next_cpu = strtol(optarg+4, &endp, 0);
275 usage_err("Bad target CPU number at '%s'", endp);
279 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
280 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
281 } else if (strncmp("any", optarg, 3) == 0) {
282 lwp_cpu_lock = LCL_NONE;
284 lwp_cpu_lock = LCL_SINGLE_CPU;
285 next_cpu = strtol(optarg, &endp, 0);
287 usage_err("Bad target CPU number at '%s'", endp);
288 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
289 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
294 * This value is set up by mp_start(), don't just
298 optcpus = strtol(optarg, NULL, 0);
299 if (optcpus < 1 || optcpus > MAXCPU)
300 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
302 if (strtol(optarg, NULL, 0) != 1) {
303 usage_err("You built a UP vkernel, only 1 cpu!");
312 kernel_mem_readonly = 0;
324 init_sys_memory(memImageFile);
334 vsize = sizeof(tsc_present);
335 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
336 vsize = sizeof(tsc_frequency);
337 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
339 cpu_feature |= CPUID_TSC;
344 vsize = sizeof(supports_sse);
346 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
347 init_fpu(supports_sse);
349 cpu_feature |= CPUID_SSE | CPUID_FXSR;
352 * We boot from the first installed disk.
354 if (bootOnDisk == 1) {
355 init_disk(diskFile, diskFileNum, VKD_DISK);
356 init_disk(cdFile, cdFileNum, VKD_CD);
358 init_disk(cdFile, cdFileNum, VKD_CD);
359 init_disk(diskFile, diskFileNum, VKD_DISK);
361 init_netif(netifFile, netifFileNum);
369 * Initialize system memory. This is the virtual kernel's 'RAM'.
373 init_sys_memory(char *imageFile)
380 * Figure out the system memory image size. If an image file was
381 * specified and -m was not specified, use the image file's size.
383 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
384 Maxmem_bytes = (vm_paddr_t)st.st_size;
385 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
387 errx(1, "Cannot create new memory file %s unless "
388 "system memory size is specified with -m",
394 * Maxmem must be known at this time
396 if (Maxmem_bytes < 32 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
397 errx(1, "Bad maxmem specification: 32MB minimum, "
398 "multiples of %dMB only",
399 SEG_SIZE / 1024 / 1024);
404 * Generate an image file name if necessary, then open/create the
405 * file exclusively locked. Do not allow multiple virtual kernels
406 * to use the same image file.
408 * Don't iterate through a million files if we do not have write
409 * access to the directory, stop if our open() failed on a
410 * non-existant file. Otherwise opens can fail for any number
411 * of reasons (lock failed, file not owned or writable by us, etc).
413 if (imageFile == NULL) {
414 for (i = 0; i < 1000000; ++i) {
415 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
417 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
418 if (fd < 0 && stat(imageFile, &st) == 0) {
425 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
427 printf("Using memory file: %s\n", imageFile);
428 if (fd < 0 || fstat(fd, &st) < 0) {
429 err(1, "Unable to open/create %s", imageFile);
434 * Truncate or extend the file as necessary. Clean out the contents
435 * of the file, we want it to be full of holes so we don't waste
436 * time reading in data from an old file that we no longer care
440 ftruncate(fd, Maxmem_bytes);
443 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
448 * Initialize kernel memory. This reserves kernel virtual memory by using
454 init_kern_memory(void)
460 char *topofstack = &dummy;
465 * Memory map our kernel virtual memory space. Note that the
466 * kernel image itself is not made part of this memory for the
469 * The memory map must be segment-aligned so we can properly
472 * If the system kernel has a different MAXDSIZ, it might not
473 * be possible to map kernel memory in its prefered location.
474 * Try a number of different locations.
476 try = (void *)0x40000000;
478 while ((char *)try + KERNEL_KVA_SIZE < topofstack) {
479 base = mmap(try, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
480 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE,
484 if (base != MAP_FAILED)
485 munmap(base, KERNEL_KVA_SIZE);
486 try = (char *)try + 0x10000000;
489 err(1, "Unable to mmap() kernel virtual memory!");
492 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
493 KvaStart = (vm_offset_t)base;
494 KvaSize = KERNEL_KVA_SIZE;
495 KvaEnd = KvaStart + KvaSize;
497 /* cannot use kprintf yet */
498 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
501 * Create a top-level page table self-mapping itself.
503 * Initialize the page directory at physical page index 0 to point
504 * to an array of page table pages starting at physical page index 1
506 lseek(MemImageFd, 0L, 0);
507 for (i = 0; i < KERNEL_KVA_SIZE / SEG_SIZE; ++i) {
508 pte = ((i + 1) * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
509 write(MemImageFd, &pte, sizeof(pte));
513 * Initialize the PTEs in the page table pages required to map the
514 * page table itself. This includes mapping the page directory page
515 * at the base so we go one more loop then normal.
517 lseek(MemImageFd, PAGE_SIZE, 0);
518 for (i = 0; i <= KERNEL_KVA_SIZE / SEG_SIZE * sizeof(vpte_t); ++i) {
519 pte = (i * PAGE_SIZE) | VPTE_V | VPTE_R | VPTE_W;
520 write(MemImageFd, &pte, sizeof(pte));
524 * Initialize remaining PTEs to 0. We may be reusing a memory image
525 * file. This is approximately a megabyte.
527 i = (KERNEL_KVA_SIZE / PAGE_SIZE - i) * sizeof(pte);
528 zero = malloc(PAGE_SIZE);
529 bzero(zero, PAGE_SIZE);
531 write(MemImageFd, zero, (i > PAGE_SIZE) ? PAGE_SIZE : i);
532 i = i - ((i > PAGE_SIZE) ? PAGE_SIZE : i);
537 * Enable the page table and calculate pointers to our self-map
538 * for easy kernel page table manipulation.
540 * KernelPTA must be offset so we can do direct VA translations
542 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
543 0 | VPTE_R | VPTE_W | VPTE_V);
544 KernelPTD = (vpte_t *)base; /* pg directory */
545 KernelPTA = (vpte_t *)((char *)base + PAGE_SIZE); /* pg table pages */
546 KernelPTA -= KvaStart >> PAGE_SHIFT;
549 * phys_avail[] represents unallocated physical memory. MI code
550 * will use phys_avail[] to create the vm_page array.
552 phys_avail[0] = PAGE_SIZE +
553 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
554 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
555 phys_avail[1] = Maxmem_bytes;
558 * (virtual_start, virtual_end) represent unallocated kernel virtual
559 * memory. MI code will create kernel_map using these parameters.
561 virtual_start = KvaStart + PAGE_SIZE +
562 KERNEL_KVA_SIZE / PAGE_SIZE * sizeof(vpte_t);
563 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
564 virtual_end = KvaStart + KERNEL_KVA_SIZE;
567 * kernel_vm_end could be set to virtual_end but we want some
568 * indication of how much of the kernel_map we've used, so
569 * set it low and let pmap_growkernel increase it even though we
570 * don't need to create any new page table pages.
572 kernel_vm_end = virtual_start;
575 * Allocate space for process 0's UAREA.
577 proc0paddr = (void *)virtual_start;
578 for (i = 0; i < UPAGES; ++i) {
579 pmap_kenter_quick(virtual_start, phys_avail[0]);
580 virtual_start += PAGE_SIZE;
581 phys_avail[0] += PAGE_SIZE;
587 crashdumpmap = virtual_start;
588 virtual_start += MAXDUMPPGS * PAGE_SIZE;
591 * msgbufp maps the system message buffer
593 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
594 msgbufp = (void *)virtual_start;
595 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
596 pmap_kenter_quick(virtual_start, phys_avail[0]);
597 virtual_start += PAGE_SIZE;
598 phys_avail[0] += PAGE_SIZE;
600 msgbufinit(msgbufp, MSGBUF_SIZE);
603 * used by kern_memio for /dev/mem access
605 ptvmmap = (caddr_t)virtual_start;
606 virtual_start += PAGE_SIZE;
609 * Bootstrap the kernel_pmap
615 * Map the per-cpu globaldata for cpu #0. Allocate the space using
616 * virtual_start and phys_avail[0]
620 init_globaldata(void)
627 * Reserve enough KVA to cover possible cpus. This is a considerable
628 * amount of KVA since the privatespace structure includes two
629 * whole page table mappings.
631 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
632 CPU_prvspace = (void *)virtual_start;
633 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
636 * Allocate enough physical memory to cover the mdglobaldata
637 * portion of the space and the idle stack and map the pages
638 * into KVA. For cpu #0 only.
640 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
642 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
643 pmap_kenter_quick(va, pa);
644 phys_avail[0] += PAGE_SIZE;
646 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
648 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
649 pmap_kenter_quick(va, pa);
650 phys_avail[0] += PAGE_SIZE;
654 * Setup the %fs for cpu #0. The mycpu macro works after this
655 * point. Note that %gs is used by pthreads.
657 tls_set_fs(&CPU_prvspace[0], sizeof(struct privatespace));
661 * Initialize very low level systems including thread0, proc0, etc.
667 struct mdglobaldata *gd;
669 gd = &CPU_prvspace[0].mdglobaldata;
670 bzero(gd, sizeof(*gd));
672 gd->mi.gd_curthread = &thread0;
673 thread0.td_gd = &gd->mi;
675 ncpus2 = 1; /* rounded down power of 2 */
676 ncpus_fit = 1; /* rounded up power of 2 */
677 /* ncpus2_mask and ncpus_fit_mask are 0 */
679 gd->mi.gd_prvspace = &CPU_prvspace[0];
680 mi_gdinit(&gd->mi, 0);
682 mi_proc0init(&gd->mi, proc0paddr);
683 lwp0.lwp_md.md_regs = &proc0_tf;
688 * Get the initial mplock with a count of 1 for the BSP.
689 * This uses a LOGICAL cpu ID, ie BSP == 0.
691 cpu_get_initial_mplock();
695 #if 0 /* #ifdef DDB */
697 if (boothowto & RB_KDB)
698 Debugger("Boot flags requested debugger");
702 initializecpu(); /* Initialize CPU registers */
704 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
708 * Map the message buffer
710 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
711 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
712 msgbufinit(msgbufp, MSGBUF_SIZE);
715 thread0.td_pcb_cr3 ... MMU
716 lwp0.lwp_md.md_regs = &proc0_tf;
721 * Filesystem image paths for the virtual kernel are optional.
722 * If specified they each should point to a disk image,
723 * the first of which will become the root disk.
725 * The virtual kernel caches data from our 'disk' just like a normal kernel,
726 * so we do not really want the real kernel to cache the data too. Use
727 * O_DIRECT to remove the duplication.
731 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
735 if (diskFileNum == 0)
738 for(i=0; i < diskFileNum; i++){
743 warnx("Invalid argument to '-r'");
747 if (DiskNum < VKDISK_MAX) {
749 struct vkdisk_info* info = NULL;
753 if (type == VKD_DISK)
754 fd = open(fname, O_RDWR|O_DIRECT, 0644);
756 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
757 if (fd < 0 || fstat(fd, &st) < 0) {
758 err(1, "Unable to open/create %s", fname);
761 if (S_ISREG(st.st_mode)) {
762 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
763 errx(1, "Disk image %s is already "
769 info = &DiskInfo[DiskNum];
775 memcpy(info->fname, fname, l);
778 if (type == VKD_CD) {
779 rootdevnames[0] = "cd9660:vcd0a";
780 } else if (type == VKD_DISK) {
781 rootdevnames[0] = "ufs:vkd0s0a";
782 rootdevnames[1] = "ufs:vkd0s1a";
788 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
796 netif_set_tapflags(int tap_unit, int f, int s)
801 bzero(&ifr, sizeof(ifr));
803 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
804 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
805 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
812 * If the flags are already set/cleared, then we return
813 * immediately to avoid extra syscalls
815 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
819 if ((flags & f) == 0)
830 * Fix up ifreq.ifr_name, since it may be trashed
831 * in previous ioctl(SIOCGIFFLAGS)
833 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
835 ifr.ifr_flags = flags & 0xffff;
836 ifr.ifr_flagshigh = flags >> 16;
837 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
838 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
846 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
848 struct ifaliasreq ifra;
849 struct sockaddr_in *in;
851 bzero(&ifra, sizeof(ifra));
852 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
855 in = (struct sockaddr_in *)&ifra.ifra_addr;
856 in->sin_family = AF_INET;
857 in->sin_len = sizeof(*in);
858 in->sin_addr.s_addr = addr;
862 in = (struct sockaddr_in *)&ifra.ifra_mask;
863 in->sin_len = sizeof(*in);
864 in->sin_addr.s_addr = mask;
867 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
868 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
876 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
881 bzero(&ifbr, sizeof(ifbr));
882 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
885 bzero(&ifd, sizeof(ifd));
886 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
887 ifd.ifd_cmd = BRDGADD;
888 ifd.ifd_len = sizeof(ifbr);
889 ifd.ifd_data = &ifbr;
891 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
893 * 'errno == EEXIST' means that the tap(4) is already
894 * a member of the bridge(4)
896 if (errno != EEXIST) {
897 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
904 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
907 * Locate the first unused tap(4) device file if auto mode is requested,
908 * or open the user supplied device file, and bring up the corresponding
911 * NOTE: Only tap(4) device file is supported currently
915 netif_open_tap(const char *netif, int *tap_unit, int s)
917 char tap_dev[MAXPATHLEN];
924 if (strcmp(netif, "auto") == 0) {
926 * Find first unused tap(4) device file
928 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
930 warnc(errno, "Unable to find a free tap(4)");
935 * User supplied tap(4) device file or unix socket.
937 if (netif[0] == '/') /* Absolute path */
938 strlcpy(tap_dev, netif, sizeof(tap_dev));
940 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
942 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
945 * If we cannot open normally try to connect to it.
948 tap_fd = unix_connect(tap_dev);
951 warn("Unable to open %s", tap_dev);
957 * Check whether the device file is a tap(4)
959 if (fstat(tap_fd, &st) < 0) {
961 } else if (S_ISCHR(st.st_mode)) {
962 dname = fdevname(tap_fd);
964 dname = strstr(dname, "tap");
967 * Bring up the corresponding tap(4) interface
969 *tap_unit = strtol(dname + 3, NULL, 10);
970 printf("TAP UNIT %d\n", *tap_unit);
971 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
978 } else if (S_ISSOCK(st.st_mode)) {
980 * Special socket connection (typically to vknet). We
981 * do not have to do anything.
989 warnx("%s is not a tap(4) device or socket", tap_dev);
998 unix_connect(const char *path)
1000 struct sockaddr_un sunx;
1003 int sndbuf = 262144;
1006 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1007 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1008 ++len; /* include nul */
1009 sunx.sun_family = AF_UNIX;
1012 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1015 if (connect(net_fd, (void *)&sunx, len) < 0) {
1019 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1020 if (fstat(net_fd, &st) == 0)
1021 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1022 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1028 #undef TAPDEV_OFLAGS
1031 * Following syntax is supported,
1032 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1034 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1035 * tap(4)'s netmask len is z
1037 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1038 * pseudo netif's address is y.y.y.y
1040 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1041 * pseudo netif's address is y.y.y.y
1042 * tap(4) and pseudo netif's netmask len are z
1044 * 5) bridgeX tap(4) will be added to bridgeX
1046 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1047 * pseudo netif's address is y.y.y.y
1049 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1050 * pseudo netif's address is y.y.y.y
1051 * pseudo netif's netmask len is z
1055 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1057 in_addr_t tap_addr, netmask, netif_addr;
1058 int next_netif_addr;
1059 char *tok, *masklen_str, *ifbridge;
1064 tok = strtok(NULL, ":/");
1067 * Nothing special, simply use tap(4) as backend
1072 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1074 * tap(4)'s address is supplied
1079 * If there is next token, then it may be pseudo
1080 * netif's address or netmask len for tap(4)
1082 next_netif_addr = 0;
1085 * Not tap(4)'s address, assume it as a bridge(4)
1092 * If there is next token, then it must be pseudo
1095 next_netif_addr = 1;
1098 netmask = netif_addr = 0;
1100 tok = strtok(NULL, ":/");
1104 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1105 if (next_netif_addr) {
1106 warnx("Invalid pseudo netif address: %s", tok);
1112 * Current token is not address, then it must be netmask len
1117 * Current token is pseudo netif address, if there is next token
1118 * it must be netmask len
1120 masklen_str = strtok(NULL, "/");
1123 /* Calculate netmask */
1124 if (masklen_str != NULL) {
1127 masklen = strtoul(masklen_str, NULL, 10);
1128 if (masklen < 32 && masklen > 0) {
1129 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1132 warnx("Invalid netmask len: %lu", masklen);
1137 /* Make sure there is no more token left */
1138 if (strtok(NULL, ":/") != NULL) {
1139 warnx("Invalid argument to '-I'");
1146 } else if (ifbridge == NULL) {
1147 /* Set tap(4) address/netmask */
1148 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1151 /* Tie tap(4) to bridge(4) */
1152 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1162 * NetifInfo[] will be filled for pseudo netif initialization.
1163 * NetifNum will be bumped to reflect the number of valid entries
1168 init_netif(char *netifExp[], int netifExpNum)
1172 if (netifExpNum == 0)
1175 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1179 for (i = 0; i < netifExpNum; ++i) {
1180 struct vknetif_info *info;
1181 in_addr_t netif_addr, netif_mask;
1182 int tap_fd, tap_unit;
1185 netif = strtok(netifExp[i], ":");
1186 if (netif == NULL) {
1187 warnx("Invalid argument to '-I'");
1192 * Open tap(4) device file and bring up the
1193 * corresponding interface
1195 tap_fd = netif_open_tap(netif, &tap_unit, s);
1200 * Initialize tap(4) and get address/netmask
1203 * NB: Rest part of netifExp[i] is passed
1204 * to netif_init_tap() implicitly.
1206 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1208 * NB: Closing tap(4) device file will bring
1209 * down the corresponding interface
1215 info = &NetifInfo[NetifNum];
1216 info->tap_fd = tap_fd;
1217 info->tap_unit = tap_unit;
1218 info->netif_addr = netif_addr;
1219 info->netif_mask = netif_mask;
1222 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1229 * Create the pid file and leave it open and locked while the vkernel is
1230 * running. This allows a script to use /usr/bin/lockf to probe whether
1231 * a vkernel is still running (so as not to accidently kill an unrelated
1232 * process from a stale pid file).
1241 if (pid_file != NULL) {
1242 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1243 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1245 if (errno == EWOULDBLOCK) {
1246 perror("Failed to lock pidfile, "
1247 "vkernel already running");
1249 perror("Failed to create pidfile");
1254 write(fd, buf, strlen(buf));
1255 /* leave the file open to maintain the lock */
1263 if (pid_file != NULL) {
1264 if (unlink(pid_file) < 0)
1265 perror("Warning: couldn't remove pidfile");
1271 usage_err(const char *ctl, ...)
1276 vfprintf(stderr, ctl, va);
1278 fprintf(stderr, "\n");
1284 usage_help(_Bool help)
1286 fprintf(stderr, "Usage: %s [-hsUv] [-c file] [-e name=value:name=value:...]\n"
1287 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1288 "\t[-m size] [-n numcpus] [-p file] [-r file]\n", save_av[0]);
1291 fprintf(stderr, "\nArguments:\n"
1292 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1293 "\t-e\tSpecify an environment to be used by the kernel.\n"
1294 "\t-h\tThis list of options.\n"
1295 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1296 "\t-I\tCreate a virtual network device.\n"
1297 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1298 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1299 "\t-n\tSpecify the number of CPUs you wish to emulate.\n"
1300 "\t-p\tSpecify a file in which to store the process ID.\n"
1301 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n"
1302 "\t-s\tBoot into single-user mode.\n"
1303 "\t-U\tEnable writing to kernel memory and module loading.\n"
1304 "\t-v\tTurn on verbose booting.\n");
1312 kprintf("cpu reset, rebooting vkernel\n");
1315 execv(save_av[0], save_av);
1321 kprintf("cpu halt, exiting vkernel\n");
1329 switch(lwp_cpu_lock) {
1332 kprintf("Locking CPU%d to real cpu %d\n",
1334 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1336 if (next_cpu >= real_ncpus)
1339 case LCL_SINGLE_CPU:
1341 kprintf("Locking CPU%d to real cpu %d\n",
1343 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1346 /* do not map virtual cpus to real cpus */