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; /* no of logical CPU bits - only SMP */
124 int vkernel_B_arg; /* 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);
146 static void handle_term(int);
151 static char **save_av;
154 * Kernel startup for virtual kernels - standard main()
156 int main(int ac, char **av) {
157 char *memImageFile = NULL;
158 char *netifFile[VKNETIF_MAX];
159 char *diskFile[VKDISK_MAX];
160 char *cdFile[VKDISK_MAX];
165 int netifFileNum = 0;
168 int bootOnDisk = -1; /* set below to vcd (0) or vkd (1) */
176 int dflag = 0; /* disable vmm */
177 int real_vkernel_enable;
188 * Currently a bad hack but rtld-elf needs LD_SHAREDLIB_BASE to
189 * be set to force it to mmap() shared libraries into low memory,
190 * so our module loader can link against the related symbols.
192 if (getenv("LD_SHAREDLIB_BASE") == NULL) {
193 setenv("LD_SHAREDLIB_BASE", "0x10000000", 1);
195 fprintf(stderr, "Must run %s with full path\n", av[0]);
199 while ((childpid = fork()) != 0) {
201 bzero(&sa, sizeof(sa));
202 sigemptyset(&sa.sa_mask);
203 sa.sa_handler = SIG_IGN;
204 sigaction(SIGINT, &sa, NULL);
205 sigaction(SIGQUIT, &sa, NULL);
206 sigaction(SIGHUP, &sa, NULL);
209 * Forward SIGTERM to the child so that
210 * the shutdown process initiates correctly.
212 sa.sa_handler = handle_term;
213 sigaction(SIGTERM, &sa, NULL);
216 * Wait for child to terminate, exit if
217 * someone stole our child.
219 while (waitpid(childpid, &status, 0) != childpid) {
223 if (WEXITSTATUS(status) != EX_VKERNEL_REBOOT)
238 kernel_mem_readonly = 1;
242 lwp_cpu_lock = LCL_NONE;
244 real_vkernel_enable = 0;
245 vsize = sizeof(real_vkernel_enable);
246 sysctlbyname("vm.vkernel_enable", &real_vkernel_enable, &vsize, NULL,0);
248 if (real_vkernel_enable == 0) {
249 errx(1, "vm.vkernel_enable is 0, must be set "
250 "to 1 to execute a vkernel!");
254 vsize = sizeof(real_ncpus);
255 sysctlbyname("hw.ncpu", &real_ncpus, &vsize, NULL, 0);
260 while ((c = getopt(ac, av, "c:hsvl:m:n:r:e:i:p:I:Ud")) != -1) {
267 * name=value:name=value:name=value...
270 * Allow values to be quoted but note that shells
271 * may remove the quotes, so using this feature
272 * to embed colons may require a backslash.
279 kern_envp = malloc(kenv_size);
280 if (kern_envp == NULL)
281 errx(1, "Couldn't allocate %zd bytes for kern_envp", kenv_size);
283 kenv_size2 = kenv_size + n + 1;
285 if ((tmp = realloc(kern_envp, kenv_size2)) == NULL)
286 errx(1, "Couldn't reallocate %zd bytes for kern_envp", kenv_size2);
288 kenv_size = kenv_size2;
291 for (i = 0, j = pos; i < n; ++i) {
292 if (optarg[i] == '"')
294 else if (optarg[i] == '\'')
296 else if (isq == 0 && optarg[i] == ':')
299 kern_envp[j++] = optarg[i];
306 boothowto |= RB_SINGLE;
312 memImageFile = optarg;
315 if (netifFileNum < VKNETIF_MAX)
316 netifFile[netifFileNum++] = strdup(optarg);
321 if (diskFileNum + cdFileNum < VKDISK_MAX)
322 diskFile[diskFileNum++] = strdup(optarg);
327 if (diskFileNum + cdFileNum < VKDISK_MAX)
328 cdFile[cdFileNum++] = strdup(optarg);
331 Maxmem_bytes = strtoull(optarg, &suffix, 0);
348 usage_err("Bad maxmem option");
356 if (strncmp("map", optarg, 3) == 0) {
357 lwp_cpu_lock = LCL_PER_CPU;
358 if (optarg[3] == ',') {
359 next_cpu = strtol(optarg+4, &endp, 0);
361 usage_err("Bad target CPU number at '%s'", endp);
365 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
366 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
367 } else if (strncmp("any", optarg, 3) == 0) {
368 lwp_cpu_lock = LCL_NONE;
370 lwp_cpu_lock = LCL_SINGLE_CPU;
371 next_cpu = strtol(optarg, &endp, 0);
373 usage_err("Bad target CPU number at '%s'", endp);
374 if (next_cpu < 0 || next_cpu > real_ncpus - 1)
375 usage_err("Bad target CPU, valid range is 0-%d", real_ncpus - 1);
380 * This value is set up by mp_start(), don't just
383 tok = strtok(optarg, ":");
384 optcpus = strtol(tok, NULL, 0);
385 if (optcpus < 1 || optcpus > MAXCPU)
386 usage_err("Bad ncpus, valid range is 1-%d", MAXCPU);
388 /* :lbits argument */
389 tok = strtok(NULL, ":");
391 vkernel_b_arg = strtol(tok, NULL, 0);
393 /* :cbits argument */
394 tok = strtok(NULL, ":");
396 vkernel_B_arg = strtol(tok, NULL, 0);
405 kernel_mem_readonly = 0;
418 vsize = sizeof(vmm_enabled);
419 sysctlbyname("hw.vmm.enable", &vmm_enabled, &vsize, NULL, 0);
420 vmm_enabled = (vmm_enabled && !dflag);
425 /* use a MAP_ANON directly */
426 init_kern_memory_vmm();
428 init_sys_memory(memImageFile);
436 vmm_guest = VMM_GUEST_VKERNEL;
441 vsize = sizeof(tsc_present);
442 sysctlbyname("hw.tsc_present", &tsc_present, &vsize, NULL, 0);
443 vsize = sizeof(tsc_invariant);
444 sysctlbyname("hw.tsc_invariant", &tsc_invariant, &vsize, NULL, 0);
445 vsize = sizeof(tsc_mpsync);
446 sysctlbyname("hw.tsc_mpsync", &tsc_mpsync, &vsize, NULL, 0);
447 vsize = sizeof(tsc_frequency);
448 sysctlbyname("hw.tsc_frequency", &tsc_frequency, &vsize, NULL, 0);
450 cpu_feature |= CPUID_TSC;
455 vsize = sizeof(supports_sse);
457 sysctlbyname("hw.instruction_sse", &supports_sse, &vsize, NULL, 0);
458 sysctlbyname("hw.mxcsr_mask", &mxcsr_mask, &msize, NULL, 0);
459 init_fpu(supports_sse);
461 cpu_feature |= CPUID_SSE | CPUID_FXSR;
464 * We boot from the first installed disk.
466 if (bootOnDisk == 1) {
467 init_disk(diskFile, diskFileNum, VKD_DISK);
468 init_disk(cdFile, cdFileNum, VKD_CD);
470 init_disk(cdFile, cdFileNum, VKD_CD);
471 init_disk(diskFile, diskFileNum, VKD_DISK);
474 init_netif(netifFile, netifFileNum);
481 /* SIGTERM handler */
490 * Initialize system memory. This is the virtual kernel's 'RAM'.
494 init_sys_memory(char *imageFile)
501 * Figure out the system memory image size. If an image file was
502 * specified and -m was not specified, use the image file's size.
504 if (imageFile && stat(imageFile, &st) == 0 && Maxmem_bytes == 0)
505 Maxmem_bytes = (vm_paddr_t)st.st_size;
506 if ((imageFile == NULL || stat(imageFile, &st) < 0) &&
508 errx(1, "Cannot create new memory file %s unless "
509 "system memory size is specified with -m",
515 * Maxmem must be known at this time
517 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
518 errx(1, "Bad maxmem specification: 64MB minimum, "
519 "multiples of %dMB only",
520 SEG_SIZE / 1024 / 1024);
525 * Generate an image file name if necessary, then open/create the
526 * file exclusively locked. Do not allow multiple virtual kernels
527 * to use the same image file.
529 * Don't iterate through a million files if we do not have write
530 * access to the directory, stop if our open() failed on a
531 * non-existant file. Otherwise opens can fail for any number
533 if (imageFile == NULL) {
534 for (i = 0; i < 1000000; ++i) {
535 asprintf(&imageFile, "/var/vkernel/memimg.%06d", i);
537 O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
538 if (fd < 0 && stat(imageFile, &st) == 0) {
545 fd = open(imageFile, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0644);
547 fprintf(stderr, "Using memory file: %s\n", imageFile);
548 if (fd < 0 || fstat(fd, &st) < 0) {
549 err(1, "Unable to open/create %s", imageFile);
554 * Truncate or extend the file as necessary. Clean out the contents
555 * of the file, we want it to be full of holes so we don't waste
556 * time reading in data from an old file that we no longer care
560 ftruncate(fd, Maxmem_bytes);
563 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
568 * Initialize kernel memory. This reserves kernel virtual memory by using
574 init_kern_memory(void)
581 * Memory map our kernel virtual memory space. Note that the
582 * kernel image itself is not made part of this memory for the
585 * The memory map must be segment-aligned so we can properly
588 * If the system kernel has a different MAXDSIZ, it might not
589 * be possible to map kernel memory in its prefered location.
590 * Try a number of different locations.
593 base = mmap((void*)KERNEL_KVA_START, KERNEL_KVA_SIZE, PROT_READ|PROT_WRITE,
594 MAP_FILE|MAP_SHARED|MAP_VPAGETABLE|MAP_FIXED|MAP_TRYFIXED,
595 MemImageFd, (off_t)KERNEL_KVA_START);
597 if (base == MAP_FAILED) {
598 err(1, "Unable to mmap() kernel virtual memory!");
601 madvise(base, KERNEL_KVA_SIZE, MADV_NOSYNC);
602 KvaStart = (vm_offset_t)base;
603 KvaSize = KERNEL_KVA_SIZE;
604 KvaEnd = KvaStart + KvaSize;
606 /* cannot use kprintf yet */
607 printf("KVM mapped at %p-%p\n", (void *)KvaStart, (void *)KvaEnd);
610 dmap_min_address = mmap(0, DMAP_SIZE, PROT_READ|PROT_WRITE,
611 MAP_NOCORE|MAP_NOSYNC|MAP_SHARED,
613 if (dmap_min_address == MAP_FAILED) {
614 err(1, "Unable to mmap() kernel DMAP region!");
619 * Bootstrap the kernel_pmap
622 pmap_bootstrap((vm_paddr_t *)&firstfree, (int64_t)base);
624 mcontrol(base, KERNEL_KVA_SIZE, MADV_SETMAP,
625 0 | VPTE_RW | VPTE_V);
628 * phys_avail[] represents unallocated physical memory. MI code
629 * will use phys_avail[] to create the vm_page array.
631 phys_avail[0] = (vm_paddr_t)firstfree;
632 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
633 phys_avail[1] = Maxmem_bytes;
637 * (virtual_start, virtual_end) represent unallocated kernel virtual
638 * memory. MI code will create kernel_map using these parameters.
640 virtual_start = KvaStart + (long)firstfree;
641 virtual_start = (virtual_start + PAGE_MASK) & ~(vm_offset_t)PAGE_MASK;
642 virtual_end = KvaStart + KERNEL_KVA_SIZE;
646 * pmap_growkernel() will set the correct value.
651 * Allocate space for process 0's UAREA.
653 proc0paddr = (void *)virtual_start;
654 for (i = 0; i < UPAGES; ++i) {
655 pmap_kenter_quick(virtual_start, phys_avail[0]);
656 virtual_start += PAGE_SIZE;
657 phys_avail[0] += PAGE_SIZE;
663 crashdumpmap = virtual_start;
664 virtual_start += MAXDUMPPGS * PAGE_SIZE;
667 * msgbufp maps the system message buffer
669 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
670 msgbufp = (void *)virtual_start;
671 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
672 pmap_kenter_quick(virtual_start, phys_avail[0]);
673 virtual_start += PAGE_SIZE;
674 phys_avail[0] += PAGE_SIZE;
676 msgbufinit(msgbufp, MSGBUF_SIZE);
679 * used by kern_memio for /dev/mem access
681 ptvmmap = (caddr_t)virtual_start;
682 virtual_start += PAGE_SIZE;
687 init_kern_memory_vmm(void)
691 struct vmm_guest_options options;
694 KvaStart = (vm_offset_t)KERNEL_KVA_START;
695 KvaSize = KERNEL_KVA_SIZE;
696 KvaEnd = KvaStart + KvaSize;
698 Maxmem = Maxmem_bytes >> PAGE_SHIFT;
701 if (Maxmem_bytes < 64 * 1024 * 1024 || (Maxmem_bytes & SEG_MASK)) {
702 errx(1, "Bad maxmem specification: 64MB minimum, "
703 "multiples of %dMB only",
704 SEG_SIZE / 1024 / 1024);
708 /* Call the vmspace_create to allocate the internal
709 * vkernel structures. Won't do anything else (no new
712 if (vmspace_create(NULL, 0, NULL) < 0)
713 panic("vmspace_create() failed");
717 * MAP_ANON the region of the VKERNEL phyisical memory
718 * (known as GPA - Guest Physical Address
720 dmap_address = mmap(NULL, Maxmem_bytes, PROT_READ|PROT_WRITE|PROT_EXEC,
721 MAP_ANON|MAP_SHARED, -1, 0);
722 if (dmap_address == MAP_FAILED) {
723 err(1, "Unable to mmap() RAM region!");
727 /* Alloc a new stack in the lowmem */
728 vkernel_stack = mmap(NULL, KERNEL_STACK_SIZE,
729 PROT_READ|PROT_WRITE|PROT_EXEC,
731 if (vkernel_stack == MAP_FAILED) {
732 err(1, "Unable to allocate stack\n");
736 * Bootstrap the kernel_pmap
738 firstfree = dmap_address;
739 dmap_min_address = NULL; /* VIRT == PHYS in the first 512G */
740 pmap_bootstrap((vm_paddr_t *)&firstfree, (uint64_t)KvaStart);
745 options.guest_cr3 = (register_t) KPML4phys;
746 options.new_stack = (uint64_t) vkernel_stack + KERNEL_STACK_SIZE;
748 if (vmm_guest_ctl(VMM_GUEST_RUN, &options)) {
749 err(1, "Unable to enter VMM mode.");
753 * phys_avail[] represents unallocated physical memory. MI code
754 * will use phys_avail[] to create the vm_page array.
756 phys_avail[0] = (vm_paddr_t)firstfree;
757 phys_avail[0] = (phys_avail[0] + PAGE_MASK) & ~(vm_paddr_t)PAGE_MASK;
758 phys_avail[1] = (vm_paddr_t)dmap_address + Maxmem_bytes;
761 * pmap_growkernel() will set the correct value.
766 * Allocate space for process 0's UAREA.
768 proc0paddr = (void *)virtual_start;
769 for (i = 0; i < UPAGES; ++i) {
770 pmap_kenter_quick(virtual_start, phys_avail[0]);
771 virtual_start += PAGE_SIZE;
772 phys_avail[0] += PAGE_SIZE;
778 crashdumpmap = virtual_start;
779 virtual_start += MAXDUMPPGS * PAGE_SIZE;
782 * msgbufp maps the system message buffer
784 assert((MSGBUF_SIZE & PAGE_MASK) == 0);
785 msgbufp = (void *)virtual_start;
786 for (i = 0; i < (MSGBUF_SIZE >> PAGE_SHIFT); ++i) {
788 pmap_kenter_quick(virtual_start, phys_avail[0]);
789 virtual_start += PAGE_SIZE;
790 phys_avail[0] += PAGE_SIZE;
793 msgbufinit(msgbufp, MSGBUF_SIZE);
796 * used by kern_memio for /dev/mem access
798 ptvmmap = (caddr_t)virtual_start;
799 virtual_start += PAGE_SIZE;
801 printf("vmm: Hardware pagetable enabled for guest\n");
806 * Map the per-cpu globaldata for cpu #0. Allocate the space using
807 * virtual_start and phys_avail[0]
811 init_globaldata(void)
818 * Reserve enough KVA to cover possible cpus. This is a considerable
819 * amount of KVA since the privatespace structure includes two
820 * whole page table mappings.
822 virtual_start = (virtual_start + SEG_MASK) & ~(vm_offset_t)SEG_MASK;
823 CPU_prvspace = (void *)virtual_start;
824 virtual_start += sizeof(struct privatespace) * SMP_MAXCPU;
827 * Allocate enough physical memory to cover the mdglobaldata
828 * portion of the space and the idle stack and map the pages
829 * into KVA. For cpu #0 only.
831 for (i = 0; i < sizeof(struct mdglobaldata); i += PAGE_SIZE) {
833 va = (vm_offset_t)&CPU_prvspace[0].mdglobaldata + i;
834 pmap_kenter_quick(va, pa);
835 phys_avail[0] += PAGE_SIZE;
837 for (i = 0; i < sizeof(CPU_prvspace[0].idlestack); i += PAGE_SIZE) {
839 va = (vm_offset_t)&CPU_prvspace[0].idlestack + i;
840 pmap_kenter_quick(va, pa);
841 phys_avail[0] += PAGE_SIZE;
845 * Setup the %gs for cpu #0. The mycpu macro works after this
846 * point. Note that %fs is used by pthreads.
848 tls_set_gs(&CPU_prvspace[0], sizeof(struct privatespace));
853 * Initialize pool tokens and other necessary locks
860 * Get the initial mplock with a count of 1 for the BSP.
861 * This uses a LOGICAL cpu ID, ie BSP == 0.
863 cpu_get_initial_mplock();
865 /* our token pool needs to work early */
866 lwkt_token_pool_init();
872 * Initialize very low level systems including thread0, proc0, etc.
878 struct mdglobaldata *gd;
880 gd = &CPU_prvspace[0].mdglobaldata;
881 bzero(gd, sizeof(*gd));
883 gd->mi.gd_curthread = &thread0;
884 thread0.td_gd = &gd->mi;
886 ncpus2 = 1; /* rounded down power of 2 */
887 ncpus_fit = 1; /* rounded up power of 2 */
888 /* ncpus2_mask and ncpus_fit_mask are 0 */
890 gd->mi.gd_prvspace = &CPU_prvspace[0];
891 mi_gdinit(&gd->mi, 0);
893 mi_proc0init(&gd->mi, proc0paddr);
894 lwp0.lwp_md.md_regs = &proc0_tf;
899 #if 0 /* #ifdef DDB */
901 if (boothowto & RB_KDB)
902 Debugger("Boot flags requested debugger");
906 initializecpu(); /* Initialize CPU registers */
908 init_param2((phys_avail[1] - phys_avail[0]) / PAGE_SIZE);
912 * Map the message buffer
914 for (off = 0; off < round_page(MSGBUF_SIZE); off += PAGE_SIZE)
915 pmap_kenter((vm_offset_t)msgbufp + off, avail_end + off);
916 msgbufinit(msgbufp, MSGBUF_SIZE);
919 thread0.td_pcb_cr3 ... MMU
920 lwp0.lwp_md.md_regs = &proc0_tf;
925 * Filesystem image paths for the virtual kernel are optional.
926 * If specified they each should point to a disk image,
927 * the first of which will become the root disk.
929 * The virtual kernel caches data from our 'disk' just like a normal kernel,
930 * so we do not really want the real kernel to cache the data too. Use
931 * O_DIRECT to remove the duplication.
935 init_disk(char *diskExp[], int diskFileNum, enum vkdisk_type type)
940 if (diskFileNum == 0)
943 for(i=0; i < diskFileNum; i++){
948 warnx("Invalid argument to '-r'");
952 * Check for a serial number for the virtual disk
953 * passed from the command line.
958 if (DiskNum < VKDISK_MAX) {
960 struct vkdisk_info* info = NULL;
964 if (type == VKD_DISK)
965 fd = open(fname, O_RDWR|O_DIRECT, 0644);
967 fd = open(fname, O_RDONLY|O_DIRECT, 0644);
968 if (fd < 0 || fstat(fd, &st) < 0) {
969 err(1, "Unable to open/create %s", fname);
972 if (S_ISREG(st.st_mode)) {
973 if (flock(fd, LOCK_EX|LOCK_NB) < 0) {
974 errx(1, "Disk image %s is already "
980 info = &DiskInfo[DiskNum];
986 memcpy(info->fname, fname, l);
989 if ((info->serno = malloc(SERNOLEN)) != NULL)
990 strlcpy(info->serno, serno, SERNOLEN);
992 warnx("Couldn't allocate memory for the operation");
996 if (type == VKD_CD) {
997 rootdevnames[0] = "cd9660:vcd0";
998 } else if (type == VKD_DISK) {
999 rootdevnames[0] = "ufs:vkd0s0a";
1000 rootdevnames[1] = "ufs:vkd0s1a";
1006 warnx("vkd%d (%s) > VKDISK_MAX", DiskNum, fname);
1014 netif_set_tapflags(int tap_unit, int f, int s)
1019 bzero(&ifr, sizeof(ifr));
1021 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1022 if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
1023 warn("tap%d: ioctl(SIOCGIFFLAGS) failed", tap_unit);
1030 * If the flags are already set/cleared, then we return
1031 * immediately to avoid extra syscalls
1033 flags = (ifr.ifr_flags & 0xffff) | (ifr.ifr_flagshigh << 16);
1035 /* Turn off flags */
1037 if ((flags & f) == 0)
1048 * Fix up ifreq.ifr_name, since it may be trashed
1049 * in previous ioctl(SIOCGIFFLAGS)
1051 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "tap%d", tap_unit);
1053 ifr.ifr_flags = flags & 0xffff;
1054 ifr.ifr_flagshigh = flags >> 16;
1055 if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
1056 warn("tap%d: ioctl(SIOCSIFFLAGS) failed", tap_unit);
1064 netif_set_tapaddr(int tap_unit, in_addr_t addr, in_addr_t mask, int s)
1066 struct ifaliasreq ifra;
1067 struct sockaddr_in *in;
1069 bzero(&ifra, sizeof(ifra));
1070 snprintf(ifra.ifra_name, sizeof(ifra.ifra_name), "tap%d", tap_unit);
1073 in = (struct sockaddr_in *)&ifra.ifra_addr;
1074 in->sin_family = AF_INET;
1075 in->sin_len = sizeof(*in);
1076 in->sin_addr.s_addr = addr;
1080 in = (struct sockaddr_in *)&ifra.ifra_mask;
1081 in->sin_len = sizeof(*in);
1082 in->sin_addr.s_addr = mask;
1085 if (ioctl(s, SIOCAIFADDR, &ifra) < 0) {
1086 warn("tap%d: ioctl(SIOCAIFADDR) failed", tap_unit);
1094 netif_add_tap2brg(int tap_unit, const char *ifbridge, int s)
1099 bzero(&ifbr, sizeof(ifbr));
1100 snprintf(ifbr.ifbr_ifsname, sizeof(ifbr.ifbr_ifsname),
1103 bzero(&ifd, sizeof(ifd));
1104 strlcpy(ifd.ifd_name, ifbridge, sizeof(ifd.ifd_name));
1105 ifd.ifd_cmd = BRDGADD;
1106 ifd.ifd_len = sizeof(ifbr);
1107 ifd.ifd_data = &ifbr;
1109 if (ioctl(s, SIOCSDRVSPEC, &ifd) < 0) {
1111 * 'errno == EEXIST' means that the tap(4) is already
1112 * a member of the bridge(4)
1114 if (errno != EEXIST) {
1115 warn("ioctl(%s, SIOCSDRVSPEC) failed", ifbridge);
1122 #define TAPDEV_OFLAGS (O_RDWR | O_NONBLOCK)
1125 * Locate the first unused tap(4) device file if auto mode is requested,
1126 * or open the user supplied device file, and bring up the corresponding
1129 * NOTE: Only tap(4) device file is supported currently
1133 netif_open_tap(const char *netif, int *tap_unit, int s)
1135 char tap_dev[MAXPATHLEN];
1142 if (strcmp(netif, "auto") == 0) {
1144 * Find first unused tap(4) device file
1146 tap_fd = open("/dev/tap", TAPDEV_OFLAGS);
1148 warnc(errno, "Unable to find a free tap(4)");
1153 * User supplied tap(4) device file or unix socket.
1155 if (netif[0] == '/') /* Absolute path */
1156 strlcpy(tap_dev, netif, sizeof(tap_dev));
1158 snprintf(tap_dev, sizeof(tap_dev), "/dev/%s", netif);
1160 tap_fd = open(tap_dev, TAPDEV_OFLAGS);
1163 * If we cannot open normally try to connect to it.
1166 tap_fd = unix_connect(tap_dev);
1169 warn("Unable to open %s", tap_dev);
1175 * Check whether the device file is a tap(4)
1177 if (fstat(tap_fd, &st) < 0) {
1179 } else if (S_ISCHR(st.st_mode)) {
1180 dname = fdevname(tap_fd);
1182 dname = strstr(dname, "tap");
1185 * Bring up the corresponding tap(4) interface
1187 *tap_unit = strtol(dname + 3, NULL, 10);
1188 printf("TAP UNIT %d\n", *tap_unit);
1189 if (netif_set_tapflags(*tap_unit, IFF_UP, s) == 0)
1196 } else if (S_ISSOCK(st.st_mode)) {
1198 * Special socket connection (typically to vknet). We
1199 * do not have to do anything.
1207 warnx("%s is not a tap(4) device or socket", tap_dev);
1216 unix_connect(const char *path)
1218 struct sockaddr_un sunx;
1221 int sndbuf = 262144;
1224 snprintf(sunx.sun_path, sizeof(sunx.sun_path), "%s", path);
1225 len = offsetof(struct sockaddr_un, sun_path[strlen(sunx.sun_path)]);
1226 ++len; /* include nul */
1227 sunx.sun_family = AF_UNIX;
1230 net_fd = socket(AF_UNIX, SOCK_SEQPACKET, 0);
1233 if (connect(net_fd, (void *)&sunx, len) < 0) {
1237 setsockopt(net_fd, SOL_SOCKET, SO_SNDBUF, &sndbuf, sizeof(sndbuf));
1238 if (fstat(net_fd, &st) == 0)
1239 printf("Network socket buffer: %d bytes\n", st.st_blksize);
1240 fcntl(net_fd, F_SETFL, O_NONBLOCK);
1246 #undef TAPDEV_OFLAGS
1249 * Following syntax is supported,
1250 * 1) x.x.x.x tap(4)'s address is x.x.x.x
1252 * 2) x.x.x.x/z tap(4)'s address is x.x.x.x
1253 * tap(4)'s netmask len is z
1255 * 3) x.x.x.x:y.y.y.y tap(4)'s address is x.x.x.x
1256 * pseudo netif's address is y.y.y.y
1258 * 4) x.x.x.x:y.y.y.y/z tap(4)'s address is x.x.x.x
1259 * pseudo netif's address is y.y.y.y
1260 * tap(4) and pseudo netif's netmask len are z
1262 * 5) bridgeX tap(4) will be added to bridgeX
1264 * 6) bridgeX:y.y.y.y tap(4) will be added to bridgeX
1265 * pseudo netif's address is y.y.y.y
1267 * 7) bridgeX:y.y.y.y/z tap(4) will be added to bridgeX
1268 * pseudo netif's address is y.y.y.y
1269 * pseudo netif's netmask len is z
1273 netif_init_tap(int tap_unit, in_addr_t *addr, in_addr_t *mask, int s)
1275 in_addr_t tap_addr, netmask, netif_addr;
1276 int next_netif_addr;
1277 char *tok, *masklen_str, *ifbridge;
1282 tok = strtok(NULL, ":/");
1285 * Nothing special, simply use tap(4) as backend
1290 if (inet_pton(AF_INET, tok, &tap_addr) > 0) {
1292 * tap(4)'s address is supplied
1297 * If there is next token, then it may be pseudo
1298 * netif's address or netmask len for tap(4)
1300 next_netif_addr = 0;
1303 * Not tap(4)'s address, assume it as a bridge(4)
1310 * If there is next token, then it must be pseudo
1313 next_netif_addr = 1;
1316 netmask = netif_addr = 0;
1318 tok = strtok(NULL, ":/");
1322 if (inet_pton(AF_INET, tok, &netif_addr) <= 0) {
1323 if (next_netif_addr) {
1324 warnx("Invalid pseudo netif address: %s", tok);
1330 * Current token is not address, then it must be netmask len
1335 * Current token is pseudo netif address, if there is next token
1336 * it must be netmask len
1338 masklen_str = strtok(NULL, "/");
1341 /* Calculate netmask */
1342 if (masklen_str != NULL) {
1345 masklen = strtoul(masklen_str, NULL, 10);
1346 if (masklen < 32 && masklen > 0) {
1347 netmask = htonl(~((1LL << (32 - masklen)) - 1)
1350 warnx("Invalid netmask len: %lu", masklen);
1355 /* Make sure there is no more token left */
1356 if (strtok(NULL, ":/") != NULL) {
1357 warnx("Invalid argument to '-I'");
1364 } else if (ifbridge == NULL) {
1365 /* Set tap(4) address/netmask */
1366 if (netif_set_tapaddr(tap_unit, tap_addr, netmask, s) < 0)
1369 /* Tie tap(4) to bridge(4) */
1370 if (netif_add_tap2brg(tap_unit, ifbridge, s) < 0)
1380 * NetifInfo[] will be filled for pseudo netif initialization.
1381 * NetifNum will be bumped to reflect the number of valid entries
1386 init_netif(char *netifExp[], int netifExpNum)
1391 if (netifExpNum == 0)
1394 s = socket(AF_INET, SOCK_DGRAM, 0); /* for ioctl(SIOC) */
1398 for (i = 0; i < netifExpNum; ++i) {
1399 struct vknetif_info *info;
1400 in_addr_t netif_addr, netif_mask;
1401 int tap_fd, tap_unit;
1404 /* Extract MAC address if there is one */
1408 netif = strtok(netifExp[i], ":");
1409 if (netif == NULL) {
1410 warnx("Invalid argument to '-I'");
1415 * Open tap(4) device file and bring up the
1416 * corresponding interface
1418 tap_fd = netif_open_tap(netif, &tap_unit, s);
1423 * Initialize tap(4) and get address/netmask
1426 * NB: Rest part of netifExp[i] is passed
1427 * to netif_init_tap() implicitly.
1429 if (netif_init_tap(tap_unit, &netif_addr, &netif_mask, s) < 0) {
1431 * NB: Closing tap(4) device file will bring
1432 * down the corresponding interface
1438 info = &NetifInfo[NetifNum];
1439 bzero(info, sizeof(*info));
1440 info->tap_fd = tap_fd;
1441 info->tap_unit = tap_unit;
1442 info->netif_addr = netif_addr;
1443 info->netif_mask = netif_mask;
1445 * If tmp isn't NULL it means a MAC could have been
1446 * specified so attempt to convert it.
1447 * Setting enaddr to NULL will tell vke_attach() we
1448 * need a pseudo-random MAC address.
1451 if ((info->enaddr = malloc(ETHER_ADDR_LEN)) == NULL)
1452 warnx("Couldn't allocate memory for the operation");
1454 if ((kether_aton(tmp, info->enaddr)) == NULL) {
1456 info->enaddr = NULL;
1462 if (NetifNum >= VKNETIF_MAX) /* XXX will this happen? */
1469 * Create the pid file and leave it open and locked while the vkernel is
1470 * running. This allows a script to use /usr/bin/lockf to probe whether
1471 * a vkernel is still running (so as not to accidently kill an unrelated
1472 * process from a stale pid file).
1481 if (pid_file != NULL) {
1482 snprintf(buf, sizeof(buf), "%ld\n", (long)getpid());
1483 fd = open(pid_file, O_RDWR|O_CREAT|O_EXLOCK|O_NONBLOCK, 0666);
1485 if (errno == EWOULDBLOCK) {
1486 perror("Failed to lock pidfile, "
1487 "vkernel already running");
1489 perror("Failed to create pidfile");
1494 write(fd, buf, strlen(buf));
1495 /* leave the file open to maintain the lock */
1503 if (pid_file != NULL) {
1504 if (unlink(pid_file) < 0)
1505 perror("Warning: couldn't remove pidfile");
1511 usage_err(const char *ctl, ...)
1516 vfprintf(stderr, ctl, va);
1518 fprintf(stderr, "\n");
1524 usage_help(_Bool help)
1526 fprintf(stderr, "Usage: %s [-hsUvd] [-c file] [-e name=value:name=value:...]\n"
1527 "\t[-i file] [-I interface[:address1[:address2][/netmask]]] [-l cpulock]\n"
1528 "\t[-m size] [-n numcpus[:lbits[:cbits]]]\n"
1529 "\t[-p file] [-r file]\n", save_av[0]);
1532 fprintf(stderr, "\nArguments:\n"
1533 "\t-c\tSpecify a readonly CD-ROM image file to be used by the kernel.\n"
1534 "\t-e\tSpecify an environment to be used by the kernel.\n"
1535 "\t-h\tThis list of options.\n"
1536 "\t-i\tSpecify a memory image file to be used by the virtual kernel.\n"
1537 "\t-I\tCreate a virtual network device.\n"
1538 "\t-l\tSpecify which, if any, real CPUs to lock virtual CPUs to.\n"
1539 "\t-m\tSpecify the amount of memory to be used by the kernel in bytes.\n"
1540 "\t-n\tSpecify the number of CPUs and the topology you wish to emulate:\n"
1541 "\t\t\tnumcpus - number of cpus\n"
1542 "\t\t\tlbits - specify the number of bits within APICID(=CPUID)\n"
1543 "\t\t\t needed for representing the logical ID.\n"
1544 "\t\t\t Controls the number of threads/core:\n"
1545 "\t\t\t (0 bits - 1 thread, 1 bit - 2 threads).\n"
1546 "\t\t\tcbits - specify the number of bits within APICID(=CPUID)\n"
1547 "\t\t\t needed for representing the core ID.\n"
1548 "\t\t\t Controls the number of cores/package:\n"
1549 "\t\t\t (0 bits - 1 core, 1 bit - 2 cores).\n"
1550 "\t-p\tSpecify a file in which to store the process ID.\n"
1551 "\t-r\tSpecify a R/W disk image file to be used by the kernel.\n"
1552 "\t-s\tBoot into single-user mode.\n"
1553 "\t-U\tEnable writing to kernel memory and module loading.\n"
1554 "\t-v\tTurn on verbose booting.\n");
1562 kprintf("cpu reset, rebooting vkernel\n");
1565 exit(EX_VKERNEL_REBOOT);
1572 kprintf("cpu halt, exiting vkernel\n");
1580 switch(lwp_cpu_lock) {
1583 kprintf("Locking CPU%d to real cpu %d\n",
1585 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1587 if (next_cpu >= real_ncpus)
1590 case LCL_SINGLE_CPU:
1592 kprintf("Locking CPU%d to real cpu %d\n",
1594 usched_set(getpid(), USCHED_SET_CPU, &next_cpu, sizeof(next_cpu));
1597 /* do not map virtual cpus to real cpus */
1603 * Allocate and free memory for module loading. The loaded module
1604 * has to be placed somewhere near the current kernel binary load
1605 * point or the relocations will not work.
1607 * I'm not sure why this isn't working.
1610 vkernel_module_memory_alloc(vm_offset_t *basep, size_t bytes)
1614 xtra = (PAGE_SIZE - (vm_offset_t)sbrk(0)) & PAGE_MASK;
1615 *basep = (vm_offset_t)sbrk(xtra + bytes) + xtra;
1616 bzero((void *)*basep, bytes);
1618 *basep = (vm_offset_t)mmap((void *)0x000000000, bytes,
1619 PROT_READ|PROT_WRITE|PROT_EXEC,
1620 MAP_ANON|MAP_SHARED, -1, 0);
1621 if ((void *)*basep == MAP_FAILED)
1628 vkernel_module_memory_free(vm_offset_t base, size_t bytes)
1632 munmap((void *)base, bytes);