2 * Copyright (c) 2006 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include <sys/types.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
41 #include <sys/random.h>
42 #include <sys/vkernel.h>
44 #include <sys/reboot.h>
46 #include <sys/msgbuf.h>
47 #include <sys/vmspace.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/sysctl.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_map.h>
54 #include <sys/mplock2.h>
58 #include <machine/cpu.h>
59 #include <machine/globaldata.h>
60 #include <machine/tls.h>
61 #include <machine/md_var.h>
62 #include <machine/vmparam.h>
63 #include <cpu/specialreg.h>
66 #include <net/if_arp.h>
67 #include <net/ethernet.h>
68 #include <net/bridge/if_bridgevar.h>
69 #include <netinet/in.h>
70 #include <arpa/inet.h>
71 #include <net/if_var.h>
85 #define EX_VKERNEL_REBOOT 32
87 vm_paddr_t phys_avail[16];
89 vm_paddr_t Maxmem_bytes;
92 struct vkdisk_info DiskInfo[VKDISK_MAX];
94 struct vknetif_info NetifInfo[VKNETIF_MAX];
100 vm_offset_t virtual_start;
101 vm_offset_t virtual_end;
102 vm_offset_t virtual2_start;
103 vm_offset_t virtual2_end;
104 vm_offset_t kernel_vm_end;
105 vm_offset_t crashdumpmap;
106 vm_offset_t clean_sva;
107 vm_offset_t clean_eva;
108 struct msgbuf *msgbufp;
111 vpte_t *KernelPTA; /* Warning: Offset for direct VA translation */
112 void *dmap_min_address;
114 u_int cpu_feature; /* XXX */
118 int64_t tsc_frequency;
119 int optcpus; /* number of cpus - see mp_start() */
120 int lwp_cpu_lock; /* if/how to lock virtual CPUs to real CPUs */
121 int real_ncpus; /* number of real CPUs */
122 int next_cpu; /* next real CPU to lock a virtual CPU to */
123 int vkernel_b_arg; /* -b argument - no of logical CPU bits - only SMP */
124 int vkernel_B_arg; /* -B argument - no of core bits - only SMP */
125 int vmm_enabled; /* VMM HW assisted enable */
126 struct privatespace *CPU_prvspace;
128 extern uint64_t KPML4phys; /* phys addr of kernel level 4 */
130 static struct trapframe proc0_tf;
131 static void *proc0paddr;
133 static void init_sys_memory(char *imageFile);
134 static void init_kern_memory(void);
135 static void init_kern_memory_vmm(void);
136 static void init_globaldata(void);
137 static void init_vkernel(void);
138 static void init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type);
139 static void init_netif(char *netifExp[], int netifFileNum);
140 static void writepid(void);
141 static void cleanpid(void);
142 static int unix_connect(const char *path);
143 static void usage_err(const char *ctl, ...);
144 static void usage_help(_Bool);
145 static void init_locks(void);
148 static char **save_av;
151 * Kernel startup for virtual kernels - standard main()
153 int main(int ac, char **av) {
154 char *memImageFile = NULL;
155 char *netifFile[VKNETIF_MAX];
156 char *diskFile[VKDISK_MAX];
157 char *cdFile[VKDISK_MAX];
162 int netifFileNum = 0;
165 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
173 int real_vkernel_enable;
183 * Currently a bad hack but rtld-elf needs LD_SHAREDLIB_BASE to
184 * be set to force it to mmap() shared libraries into low memory,
185 * so our module loader can link against the related symbols.
187 if (getenv("LD_SHAREDLIB_BASE") == NULL) {
188 setenv("LD_SHAREDLIB_BASE", "0x10000000", 1);
190 fprintf(stderr, "Must run %s with full path\n", av[0]);
194 while ((pid = fork()) != 0) {
196 bzero(&sa, sizeof(sa));
197 sigemptyset(&sa.sa_mask);
198 sa.sa_handler = SIG_IGN;
199 sigaction(SIGINT, &sa, NULL);
200 sigaction(SIGQUIT, &sa, NULL);
201 sigaction(SIGHUP, &sa, NULL);
204 * Wait for child to terminate, exit if
205 * someone stole our child.
207 while (waitpid(pid, &status, 0) != pid) {
211 if (WEXITSTATUS(status) != EX_VKERNEL_REBOOT)
226 kernel_mem_readonly = 1;
230 lwp_cpu_lock = LCL_NONE;
232 real_vkernel_enable = 0;
233 vsize = sizeof(real_vkernel_enable);
234 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
236 if (real_vkernel_enable == 0) {
237 errx(1, "vm.vkernel_enable is 0, must be set "
238 "to 1 to execute a vkernel!");
242 vsize = sizeof(real_ncpus);
243 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
248 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:Ub:B:")) != -1) {
252 * name=value:name=value:name=value...
255 * Allow values to be quoted but note that shells
256 * may remove the quotes, so using this feature
257 * to embed colons may require a backslash.
264 kern_envp = malloc(kenv_size);
265 if (kern_envp == NULL)
266 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
268 kenv_size2 = kenv_size + n + 1;
270 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
271 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
273 kenv_size = kenv_size2;
276 for (i = 0, j = pos; i < n; ++i) {
277 if (optarg[i] == '"')
279 else if (optarg[i] == '\'')
281 else if (isq == 0 && optarg[i] == ':')
284 kern_envp[j++] = optarg[i];
291 boothowto |= RB_SINGLE;
297 memImageFile = optarg;
300 if (netifFileNum < VKNETIF_MAX)
301 netifFile[netifFileNum++] = strdup(optarg);
306 if (diskFileNum + cdFileNum < VKDISK_MAX)
307 diskFile[diskFileNum++] = strdup(optarg);
312 if (diskFileNum + cdFileNum < VKDISK_MAX)
313 cdFile[cdFileNum++] = strdup(optarg);
316 Maxmem_bytes = strtoull(optarg, &suffix, 0);
333 usage_err("Bad maxmem option");
341 if (strncmp("map", optarg, 3) == 0) {
342 lwp_cpu_lock = LCL_PER_CPU;
343 if (optarg[3] == ',') {
344 next_cpu = strtol(optarg+4, &endp, 0);
346 usage_err("Bad target CPU number at '%s'", endp);
350 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
351 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
352 } else if (strncmp("any", optarg, 3) == 0) {
353 lwp_cpu_lock = LCL_NONE;
355 lwp_cpu_lock = LCL_SINGLE_CPU;
356 next_cpu = strtol(optarg, &endp, 0);
358 usage_err("Bad target CPU number at '%s'", endp);
359 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
360 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
365 * This value is set up by mp_start(), don't just
368 tok = strtok(optarg, ":");
369 optcpus = strtol(tok, NULL, 0);
370 if (optcpus < 1 || optcpus > MAXCPU)
371 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
373 /* :lbits argument */
374 tok = strtok(NULL, ":");
376 vkernel_b_arg = strtol(tok, NULL, 0);
378 /* :cbits argument */
379 tok = strtok(NULL, ":");
381 vkernel_B_arg = strtol(tok, NULL, 0);
390 kernel_mem_readonly = 0;
403 vsize = sizeof(vmm_enabled);
404 sysctlbyname("hw.vmm.enable", &vmm_enabled, &vsize, NULL, 0);
409 /* use a MAP_ANON directly */
410 init_kern_memory_vmm();
412 init_sys_memory(memImageFile);
425 vsize = sizeof(tsc_present);
426 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
427 vsize = sizeof(tsc_invariant);
428 sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0);
429 vsize = sizeof(tsc_mpsync);
430 sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0);
431 vsize = sizeof(tsc_frequency);
432 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
434 cpu_feature |= CPUID_TSC;
439 vsize = sizeof(supports_sse);
441 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
442 init_fpu(supports_sse);
444 cpu_feature |= CPUID_SSE | CPUID_FXSR;
447 * We boot from the first installed disk.
449 if (bootOnDisk == 1) {
450 init_disk(diskFile, diskFileNum, VKD_DISK);
451 init_disk(cdFile, cdFileNum, VKD_CD);
453 init_disk(cdFile, cdFileNum, VKD_CD);
454 init_disk(diskFile, diskFileNum, VKD_DISK);
457 init_netif(netifFile, netifFileNum);
465 * Initialize system memory. This is the virtual kernel's 'RAM'.
469 init_sys_memory(char *imageFile)
476 * Figure out the system memory image size. If an image file was
477 * specified and -m was not specified, use the image file's size.
479 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
480 Maxmem_bytes = (vm_paddr_t)st.st_size;
481 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
483 errx(1, "Cannot create new memory file %s unless "
484 "system memory size is specified with -m",
490 * Maxmem must be known at this time
492 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
493 errx(1, "Bad maxmem specification: 64MB minimum, "
494 "multiples of %dMB only",
495 SEG_SIZE / 1024 / 1024);
500 * Generate an image file name if necessary, then open/create the
501 * file exclusively locked. Do not allow multiple virtual kernels
502 * to use the same image file.
504 * Don't iterate through a million files if we do not have write
505 * access to the directory, stop if our open() failed on a
506 * non-existant file. Otherwise opens can fail for any number
508 if (imageFile == NULL) {
509 for (i = 0; i < 1000000; ++i) {
510 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
512 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
513 if (fd < 0 && stat(imageFile, &st) == 0) {
520 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
522 fprintf(stderr, "Using memory file: %s\n", imageFile);
523 if (fd < 0 || fstat(fd, &st) < 0) {
524 err(1, "Unable to open/create %s", imageFile);
529 * Truncate or extend the file as necessary. Clean out the contents
530 * of the file, we want it to be full of holes so we don't waste
531 * time reading in data from an old file that we no longer care
535 ftruncate(fd, Maxmem_bytes);
538 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
543 * Initialize kernel memory. This reserves kernel virtual memory by using
549 init_kern_memory(void)
556 * Memory map our kernel virtual memory space. Note that the
557 * kernel image itself is not made part of this memory for the
560 * The memory map must be segment-aligned so we can properly
563 * If the system kernel has a different MAXDSIZ, it might not
564 * be possible to map kernel memory in its prefered location.
565 * Try a number of different locations.
568 base = mmap((void*)KERNEL_KVA_START, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
569 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED|MAP_TRYFIXED,
570 MemImageFd, (off_t)KERNEL_KVA_START);
572 if (base == MAP_FAILED) {
573 err(1, "Unable to mmap() kernel virtual memory!");
576 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
577 KvaStart = (vm_offset_t)base;
578 KvaSize = KERNEL_KVA_SIZE;
579 KvaEnd = KvaStart + KvaSize;
581 /* cannot use kprintf yet */
582 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
585 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE,
586 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED,
588 if (dmap_min_address == MAP_FAILED) {
589 err(1, "Unable to mmap() kernel DMAP region!");
594 * Bootstrap the kernel_pmap
597 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base);
599 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
600 0 | VPTE_RW | VPTE_V);
603 * phys_avail[] represents unallocated physical memory. MI code
604 * will use phys_avail[] to create the vm_page array.
606 phys_avail[0] = (vm_paddr_t)firstfree;
607 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
608 phys_avail[1] = Maxmem_bytes;
612 * (virtual_start, virtual_end) represent unallocated kernel virtual
613 * memory. MI code will create kernel_map using these parameters.
615 virtual_start = KvaStart + (long)firstfree;
616 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
617 virtual_end = KvaStart + KERNEL_KVA_SIZE;
621 * pmap_growkernel() will set the correct value.
626 * Allocate space for process 0's UAREA.
628 proc0paddr = (void *)virtual_start;
629 for (i = 0; i < UPAGES; ++i) {
630 pmap_kenter_quick(virtual_start, phys_avail[0]);
631 virtual_start += PAGE_SIZE;
632 phys_avail[0] += PAGE_SIZE;
638 crashdumpmap = virtual_start;
639 virtual_start += MAXDUMPPGS * PAGE_SIZE;
642 * msgbufp maps the system message buffer
644 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
645 msgbufp = (void *)virtual_start;
646 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
647 pmap_kenter_quick(virtual_start, phys_avail[0]);
648 virtual_start += PAGE_SIZE;
649 phys_avail[0] += PAGE_SIZE;
651 msgbufinit(msgbufp, MSGBUF_SIZE);
654 * used by kern_memio for /dev/mem access
656 ptvmmap = (caddr_t)virtual_start;
657 virtual_start += PAGE_SIZE;
662 init_kern_memory_vmm(void)
666 struct guest_options options;
669 KvaStart = (vm_offset_t)KERNEL_KVA_START;
670 KvaSize = KERNEL_KVA_SIZE;
671 KvaEnd = KvaStart + KvaSize;
673 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
676 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
677 errx(1, "Bad maxmem specification: 64MB minimum, "
678 "multiples of %dMB only",
679 SEG_SIZE / 1024 / 1024);
683 /* Call the vmspace_create to allocate the internal
684 * vkernel structures. Won't do anything else (no new
687 if (vmspace_create(NULL, 0, NULL) < 0)
688 panic("vmspace_create() failed");
692 * MAP_ANON the region of the VKERNEL phyisical memory
693 * (known as GPA - Guest Physical Address
695 dmap_address = mmap(NULL, Maxmem_bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
696 MAP_ANON|MAP_SHARED, -1, 0);
697 if (dmap_address == MAP_FAILED) {
698 err(1, "Unable to mmap() RAM region!");
702 /* Alloc a new stack in the lowmem */
703 vkernel_stack = mmap(NULL, KERNEL_STACK_SIZE,
704 PROT_READ|PROT_WRITE|PROT_EXEC,
706 if (vkernel_stack == MAP_FAILED) {
707 err(1, "Unable to allocate stack\n");
711 * Bootstrap the kernel_pmap
713 firstfree = dmap_address;
714 dmap_min_address = NULL; /* VIRT == PHYS in the first 512G */
715 pmap_bootstrap((vm_paddr_t *)&firstfree, (uint64_t)KvaStart);
720 options.guest_cr3 = (register_t) KPML4phys;
721 options.new_stack = (uint64_t) vkernel_stack + KERNEL_STACK_SIZE;
723 if (vmm_guest_ctl(VMM_GUEST_RUN, &options)) {
724 err(1, "Unable to enter VMM mode.");
728 * phys_avail[] represents unallocated physical memory. MI code
729 * will use phys_avail[] to create the vm_page array.
731 phys_avail[0] = (vm_paddr_t)firstfree;
732 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
733 phys_avail[1] = (vm_paddr_t)dmap_address + Maxmem_bytes;
736 * pmap_growkernel() will set the correct value.
741 * Allocate space for process 0's UAREA.
743 proc0paddr = (void *)virtual_start;
744 for (i = 0; i < UPAGES; ++i) {
745 pmap_kenter_quick(virtual_start, phys_avail[0]);
746 virtual_start += PAGE_SIZE;
747 phys_avail[0] += PAGE_SIZE;
753 crashdumpmap = virtual_start;
754 virtual_start += MAXDUMPPGS * PAGE_SIZE;
757 * msgbufp maps the system message buffer
759 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
760 msgbufp = (void *)virtual_start;
761 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
763 pmap_kenter_quick(virtual_start, phys_avail[0]);
764 virtual_start += PAGE_SIZE;
765 phys_avail[0] += PAGE_SIZE;
768 msgbufinit(msgbufp, MSGBUF_SIZE);
771 * used by kern_memio for /dev/mem access
773 ptvmmap = (caddr_t)virtual_start;
774 virtual_start += PAGE_SIZE;
776 printf("vmm: Hardware pagetable enabled for guest\n");
781 * Map the per-cpu globaldata for cpu #0. Allocate the space using
782 * virtual_start and phys_avail[0]
786 init_globaldata(void)
793 * Reserve enough KVA to cover possible cpus. This is a considerable
794 * amount of KVA since the privatespace structure includes two
795 * whole page table mappings.
797 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
798 CPU_prvspace = (void *)virtual_start;
799 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
802 * Allocate enough physical memory to cover the mdglobaldata
803 * portion of the space and the idle stack and map the pages
804 * into KVA. For cpu #0 only.
806 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
808 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
809 pmap_kenter_quick(va, pa);
810 phys_avail[0] += PAGE_SIZE;
812 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
814 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
815 pmap_kenter_quick(va, pa);
816 phys_avail[0] += PAGE_SIZE;
820 * Setup the %gs for cpu #0. The mycpu macro works after this
821 * point. Note that %fs is used by pthreads.
823 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace));
828 * Initialize pool tokens and other necessary locks
835 * Get the initial mplock with a count of 1 for the BSP.
836 * This uses a LOGICAL cpu ID, ie BSP == 0.
838 cpu_get_initial_mplock();
840 /* our token pool needs to work early */
841 lwkt_token_pool_init();
847 * Initialize very low level systems including thread0, proc0, etc.
853 struct mdglobaldata *gd;
855 gd = &CPU_prvspace[0].mdglobaldata;
856 bzero(gd, sizeof(*gd));
858 gd->mi.gd_curthread = &thread0;
859 thread0.td_gd = &gd->mi;
861 ncpus2 = 1; /* rounded down power of 2 */
862 ncpus_fit = 1; /* rounded up power of 2 */
863 /* ncpus2_mask and ncpus_fit_mask are 0 */
865 gd->mi.gd_prvspace = &CPU_prvspace[0];
866 mi_gdinit(&gd->mi, 0);
868 mi_proc0init(&gd->mi, proc0paddr);
869 lwp0.lwp_md.md_regs = &proc0_tf;
874 #if 0 /* #ifdef DDB */
876 if (boothowto & RB_KDB)
877 Debugger("Boot flags requested debugger");
881 initializecpu(); /* Initialize CPU registers */
883 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
887 * Map the message buffer
889 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
890 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
891 msgbufinit(msgbufp, MSGBUF_SIZE);
894 thread0.td_pcb_cr3 ... MMU
895 lwp0.lwp_md.md_regs = &proc0_tf;
900 * Filesystem image paths for the virtual kernel are optional.
901 * If specified they each should point to a disk image,
902 * the first of which will become the root disk.
904 * The virtual kernel caches data from our 'disk' just like a normal kernel,
905 * so we do not really want the real kernel to cache the data too. Use
906 * O_DIRECT to remove the duplication.
910 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
915 if (diskFileNum == 0)
918 for(i=0; i < diskFileNum; i++){
923 warnx("Invalid argument to '-r'");
927 * Check for a serial number for the virtual disk
928 * passed from the command line.
933 if (DiskNum < VKDISK_MAX) {
935 struct vkdisk_info* info = NULL;
939 if (type == VKD_DISK)
940 fd = open(fname, O_RDWR|O_DIRECT, 0644);
942 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
943 if (fd < 0 || fstat(fd, &st) < 0) {
944 err(1, "Unable to open/create %s", fname);
947 if (S_ISREG(st.st_mode)) {
948 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
949 errx(1, "Disk image %s is already "
955 info = &DiskInfo[DiskNum];
961 memcpy(info->fname, fname, l);
964 if ((info->serno = malloc(SERNOLEN)) != NULL)
965 strlcpy(info->serno, serno, SERNOLEN);
967 warnx("Couldn't allocate memory for the operation");
971 if (type == VKD_CD) {
972 rootdevnames[0] = "cd9660:vcd0a";
973 } else if (type == VKD_DISK) {
974 rootdevnames[0] = "ufs:vkd0s0a";
975 rootdevnames[1] = "ufs:vkd0s1a";
981 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
989 netif_set_tapflags(int tap_unit, int f, int s)
994 bzero(&ifr, sizeof(ifr));
996 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
997 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
998 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
1005 * If the flags are already set/cleared, then we return
1006 * immediately to avoid extra syscalls
1008 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
1010 /* Turn off flags */
1012 if ((flags & f) == 0)
1023 * Fix up ifreq.ifr_name, since it may be trashed
1024 * in previous ioctl(SIOCGIFFLAGS)
1026 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1028 ifr.ifr_flags = flags & 0xffff;
1029 ifr.ifr_flagshigh = flags >> 16;
1030 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
1031 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
1039 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
1041 struct ifaliasreq ifra;
1042 struct sockaddr_in *in;
1044 bzero(&ifra, sizeof(ifra));
1045 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
1048 in = (struct sockaddr_in *)&ifra.ifra_addr;
1049 in->sin_family = AF_INET;
1050 in->sin_len = sizeof(*in);
1051 in->sin_addr.s_addr = addr;
1055 in = (struct sockaddr_in *)&ifra.ifra_mask;
1056 in->sin_len = sizeof(*in);
1057 in->sin_addr.s_addr = mask;
1060 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
1061 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
1069 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
1074 bzero(&ifbr, sizeof(ifbr));
1075 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
1078 bzero(&ifd, sizeof(ifd));
1079 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
1080 ifd.ifd_cmd = BRDGADD;
1081 ifd.ifd_len = sizeof(ifbr);
1082 ifd.ifd_data = &ifbr;
1084 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
1086 * 'errno == EEXIST' means that the tap(4) is already
1087 * a member of the bridge(4)
1089 if (errno != EEXIST) {
1090 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
1097 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
1100 * Locate the first unused tap(4) device file if auto mode is requested,
1101 * or open the user supplied device file, and bring up the corresponding
1104 * NOTE: Only tap(4) device file is supported currently
1108 netif_open_tap(const char *netif, int *tap_unit, int s)
1110 char tap_dev[MAXPATHLEN];
1117 if (strcmp(netif, "auto") == 0) {
1119 * Find first unused tap(4) device file
1121 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
1123 warnc(errno, "Unable to find a free tap(4)");
1128 * User supplied tap(4) device file or unix socket.
1130 if (netif[0] == '/') /* Absolute path */
1131 strlcpy(tap_dev, netif, sizeof(tap_dev));
1133 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
1135 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
1138 * If we cannot open normally try to connect to it.
1141 tap_fd = unix_connect(tap_dev);
1144 warn("Unable to open %s", tap_dev);
1150 * Check whether the device file is a tap(4)
1152 if (fstat(tap_fd, &st) < 0) {
1154 } else if (S_ISCHR(st.st_mode)) {
1155 dname = fdevname(tap_fd);
1157 dname = strstr(dname, "tap");
1160 * Bring up the corresponding tap(4) interface
1162 *tap_unit = strtol(dname + 3, NULL, 10);
1163 printf("TAP UNIT %d\n", *tap_unit);
1164 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1171 } else if (S_ISSOCK(st.st_mode)) {
1173 * Special socket connection (typically to vknet). We
1174 * do not have to do anything.
1182 warnx("%s is not a tap(4) device or socket", tap_dev);
1191 unix_connect(const char *path)
1193 struct sockaddr_un sunx;
1196 int sndbuf = 262144;
1199 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1200 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1201 ++len; /* include nul */
1202 sunx.sun_family = AF_UNIX;
1205 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1208 if (connect(net_fd, (void *)&sunx, len) < 0) {
1212 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1213 if (fstat(net_fd, &st) == 0)
1214 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1215 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1221 #undef TAPDEV_OFLAGS
1224 * Following syntax is supported,
1225 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1227 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1228 * tap(4)'s netmask len is z
1230 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1231 * pseudo netif's address is y.y.y.y
1233 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1234 * pseudo netif's address is y.y.y.y
1235 * tap(4) and pseudo netif's netmask len are z
1237 * 5) bridgeX tap(4) will be added to bridgeX
1239 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1240 * pseudo netif's address is y.y.y.y
1242 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1243 * pseudo netif's address is y.y.y.y
1244 * pseudo netif's netmask len is z
1248 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1250 in_addr_t tap_addr, netmask, netif_addr;
1251 int next_netif_addr;
1252 char *tok, *masklen_str, *ifbridge;
1257 tok = strtok(NULL, ":/");
1260 * Nothing special, simply use tap(4) as backend
1265 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1267 * tap(4)'s address is supplied
1272 * If there is next token, then it may be pseudo
1273 * netif's address or netmask len for tap(4)
1275 next_netif_addr = 0;
1278 * Not tap(4)'s address, assume it as a bridge(4)
1285 * If there is next token, then it must be pseudo
1288 next_netif_addr = 1;
1291 netmask = netif_addr = 0;
1293 tok = strtok(NULL, ":/");
1297 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1298 if (next_netif_addr) {
1299 warnx("Invalid pseudo netif address: %s", tok);
1305 * Current token is not address, then it must be netmask len
1310 * Current token is pseudo netif address, if there is next token
1311 * it must be netmask len
1313 masklen_str = strtok(NULL, "/");
1316 /* Calculate netmask */
1317 if (masklen_str != NULL) {
1320 masklen = strtoul(masklen_str, NULL, 10);
1321 if (masklen < 32 && masklen > 0) {
1322 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1325 warnx("Invalid netmask len: %lu", masklen);
1330 /* Make sure there is no more token left */
1331 if (strtok(NULL, ":/") != NULL) {
1332 warnx("Invalid argument to '-I'");
1339 } else if (ifbridge == NULL) {
1340 /* Set tap(4) address/netmask */
1341 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1344 /* Tie tap(4) to bridge(4) */
1345 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1355 * NetifInfo[] will be filled for pseudo netif initialization.
1356 * NetifNum will be bumped to reflect the number of valid entries
1361 init_netif(char *netifExp[], int netifExpNum)
1366 if (netifExpNum == 0)
1369 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1373 for (i = 0; i < netifExpNum; ++i) {
1374 struct vknetif_info *info;
1375 in_addr_t netif_addr, netif_mask;
1376 int tap_fd, tap_unit;
1379 /* Extract MAC address if there is one */
1383 netif = strtok(netifExp[i], ":");
1384 if (netif == NULL) {
1385 warnx("Invalid argument to '-I'");
1390 * Open tap(4) device file and bring up the
1391 * corresponding interface
1393 tap_fd = netif_open_tap(netif, &tap_unit, s);
1398 * Initialize tap(4) and get address/netmask
1401 * NB: Rest part of netifExp[i] is passed
1402 * to netif_init_tap() implicitly.
1404 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1406 * NB: Closing tap(4) device file will bring
1407 * down the corresponding interface
1413 info = &NetifInfo[NetifNum];
1414 bzero(info, sizeof(*info));
1415 info->tap_fd = tap_fd;
1416 info->tap_unit = tap_unit;
1417 info->netif_addr = netif_addr;
1418 info->netif_mask = netif_mask;
1420 * If tmp isn't NULL it means a MAC could have been
1421 * specified so attempt to convert it.
1422 * Setting enaddr to NULL will tell vke_attach() we
1423 * need a pseudo-random MAC address.
1426 if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL)
1427 warnx("Couldn't allocate memory for the operation");
1429 if ((kether_aton(tmp, info->enaddr)) == NULL) {
1431 info->enaddr = NULL;
1437 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1444 * Create the pid file and leave it open and locked while the vkernel is
1445 * running. This allows a script to use /usr/bin/lockf to probe whether
1446 * a vkernel is still running (so as not to accidently kill an unrelated
1447 * process from a stale pid file).
1456 if (pid_file != NULL) {
1457 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1458 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1460 if (errno == EWOULDBLOCK) {
1461 perror("Failed to lock pidfile, "
1462 "vkernel already running");
1464 perror("Failed to create pidfile");
1469 write(fd, buf, strlen(buf));
1470 /* leave the file open to maintain the lock */
1478 if (pid_file != NULL) {
1479 if (unlink(pid_file) < 0)
1480 perror("Warning: couldn't remove pidfile");
1486 usage_err(const char *ctl, ...)
1491 vfprintf(stderr, ctl, va);
1493 fprintf(stderr, "\n");
1499 usage_help(_Bool help)
1501 fprintf(stderr, "Usage: %s [-hsUv] [-c file] [-e name=value:name=value:...]\n"
1502 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1503 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1504 "\t[-p file] [-r file]\n", save_av[0]);
1507 fprintf(stderr, "\nArguments:\n"
1508 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1509 "\t-e\tSpecify an environment to be used by the kernel.\n"
1510 "\t-h\tThis list of options.\n"
1511 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1512 "\t-I\tCreate a virtual network device.\n"
1513 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1514 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1515 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1516 "\t \t- numcpus - number of cpus\n"
1517 "\t \t- :lbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1518 "\t \t the logical ID. Controls the number of threads/core (0bits - 1 thread, 1bit - 2 threads).\n"
1519 "\t \t- :cbits - specify the number of bits within APICID(=CPUID) needed for representing\n"
1520 "\t \t the core ID. Controls the number of core/package (0bits - 1 core, 1bit - 2 cores).\n"
1521 "\t-p\tSpecify a file in which to store the process ID.\n"
1522 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n"
1523 "\t-s\tBoot into single-user mode.\n"
1524 "\t-U\tEnable writing to kernel memory and module loading.\n"
1525 "\t-v\tTurn on verbose booting.\n");
1533 kprintf("cpu reset, rebooting vkernel\n");
1536 exit(EX_VKERNEL_REBOOT);
1543 kprintf("cpu halt, exiting vkernel\n");
1551 switch(lwp_cpu_lock) {
1554 kprintf("Locking CPU%d to real cpu %d\n",
1556 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1558 if (next_cpu >= real_ncpus)
1561 case LCL_SINGLE_CPU:
1563 kprintf("Locking CPU%d to real cpu %d\n",
1565 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1568 /* do not map virtual cpus to real cpus */
1574 * Allocate and free memory for module loading. The loaded module
1575 * has to be placed somewhere near the current kernel binary load
1576 * point or the relocations will not work.
1578 * I'm not sure why this isn't working.
1581 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes)
1585 xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK;
1586 *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra;
1587 bzero((void *)*basep, bytes);
1589 *basep = (vm_offset_t)mmap((void *)0x000000000, bytes,
1590 PROT_READ|PROT_WRITE|PROT_EXEC,
1591 MAP_ANON|MAP_SHARED, -1, 0);
1592 if ((void *)*basep == MAP_FAILED)
1599 vkernel_module_memory_free(vm_offset_t base, size_t bytes)
1603 munmap((void *)base, bytes);