1
2
3
4
5
6
7#define _LARGEFILE64_SOURCE
8#define _GNU_SOURCE
9#include <stdio.h>
10#include <string.h>
11#include <unistd.h>
12#include <err.h>
13#include <stdint.h>
14#include <stdlib.h>
15#include <elf.h>
16#include <sys/mman.h>
17#include <sys/param.h>
18#include <sys/types.h>
19#include <sys/stat.h>
20#include <sys/wait.h>
21#include <sys/eventfd.h>
22#include <fcntl.h>
23#include <stdbool.h>
24#include <errno.h>
25#include <ctype.h>
26#include <sys/socket.h>
27#include <sys/ioctl.h>
28#include <sys/time.h>
29#include <time.h>
30#include <netinet/in.h>
31#include <net/if.h>
32#include <linux/sockios.h>
33#include <linux/if_tun.h>
34#include <sys/uio.h>
35#include <termios.h>
36#include <getopt.h>
37#include <assert.h>
38#include <sched.h>
39#include <limits.h>
40#include <stddef.h>
41#include <signal.h>
42#include <pwd.h>
43#include <grp.h>
44
45#ifndef VIRTIO_F_ANY_LAYOUT
46#define VIRTIO_F_ANY_LAYOUT 27
47#endif
48
49
50
51
52
53
54
55
56
57
58typedef unsigned long long u64;
59typedef uint32_t u32;
60typedef uint16_t u16;
61typedef uint8_t u8;
62
63
64#include <linux/virtio_config.h>
65#include <linux/virtio_net.h>
66#include <linux/virtio_blk.h>
67#include <linux/virtio_console.h>
68#include <linux/virtio_rng.h>
69#include <linux/virtio_ring.h>
70#include <asm/bootparam.h>
71#include "../../include/linux/lguest_launcher.h"
72
73#define BRIDGE_PFX "bridge:"
74#ifndef SIOCBRADDIF
75#define SIOCBRADDIF 0x89a2
76#endif
77
78#define DEVICE_PAGES 256
79
80#define VIRTQUEUE_NUM 256
81
82
83
84
85
86static bool verbose;
87#define verbose(args...) \
88 do { if (verbose) printf(args); } while(0)
89
90
91
92static void *guest_base;
93
94static unsigned long guest_limit, guest_max;
95
96static int lguest_fd;
97
98
99static unsigned int __thread cpu_id;
100
101
102struct device_list {
103
104 unsigned int next_irq;
105
106
107 unsigned int device_num;
108
109
110 u8 *descpage;
111
112
113 struct device *dev;
114
115 struct device *lastdev;
116};
117
118
119static struct device_list devices;
120
121
122struct device {
123
124 struct device *next;
125
126
127 struct lguest_device_desc *desc;
128
129
130 unsigned int feature_len;
131 unsigned int num_vq;
132
133
134 const char *name;
135
136
137 struct virtqueue *vq;
138
139
140 bool running;
141
142
143 void *priv;
144};
145
146
147struct virtqueue {
148 struct virtqueue *next;
149
150
151 struct device *dev;
152
153
154 struct lguest_vqconfig config;
155
156
157 struct vring vring;
158
159
160 u16 last_avail_idx;
161
162
163 unsigned int pending_used;
164
165
166 int eventfd;
167
168
169 void (*service)(struct virtqueue *vq);
170 pid_t thread;
171};
172
173
174static char **main_args;
175
176
177static struct termios orig_term;
178
179
180
181
182
183
184#define wmb() __asm__ __volatile__("" : : : "memory")
185#define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
186#define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
187
188
189#define lg_last_avail(vq) ((vq)->last_avail_idx)
190
191
192
193
194
195#define cpu_to_le16(v16) (v16)
196#define cpu_to_le32(v32) (v32)
197#define cpu_to_le64(v64) (v64)
198#define le16_to_cpu(v16) (v16)
199#define le32_to_cpu(v32) (v32)
200#define le64_to_cpu(v64) (v64)
201
202
203static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
204{
205 unsigned int i;
206
207 for (i = 0; i < num_iov; i++)
208 if (iov[i].iov_len)
209 return false;
210 return true;
211}
212
213
214static void iov_consume(struct iovec iov[], unsigned num_iov,
215 void *dest, unsigned len)
216{
217 unsigned int i;
218
219 for (i = 0; i < num_iov; i++) {
220 unsigned int used;
221
222 used = iov[i].iov_len < len ? iov[i].iov_len : len;
223 if (dest) {
224 memcpy(dest, iov[i].iov_base, used);
225 dest += used;
226 }
227 iov[i].iov_base += used;
228 iov[i].iov_len -= used;
229 len -= used;
230 }
231 if (len != 0)
232 errx(1, "iovec too short!");
233}
234
235
236static u8 *get_feature_bits(struct device *dev)
237{
238 return (u8 *)(dev->desc + 1)
239 + dev->num_vq * sizeof(struct lguest_vqconfig);
240}
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257static void *from_guest_phys(unsigned long addr)
258{
259 return guest_base + addr;
260}
261
262static unsigned long to_guest_phys(const void *addr)
263{
264 return (addr - guest_base);
265}
266
267
268
269
270
271
272
273static int open_or_die(const char *name, int flags)
274{
275 int fd = open(name, flags);
276 if (fd < 0)
277 err(1, "Failed to open %s", name);
278 return fd;
279}
280
281
282static void *map_zeroed_pages(unsigned int num)
283{
284 int fd = open_or_die("/dev/zero", O_RDONLY);
285 void *addr;
286
287
288
289
290
291
292 addr = mmap(NULL, getpagesize() * (num+2),
293 PROT_NONE, MAP_PRIVATE, fd, 0);
294
295 if (addr == MAP_FAILED)
296 err(1, "Mmapping %u pages of /dev/zero", num);
297
298 if (mprotect(addr + getpagesize(), getpagesize() * num,
299 PROT_READ|PROT_WRITE) == -1)
300 err(1, "mprotect rw %u pages failed", num);
301
302
303
304
305
306 close(fd);
307
308
309 return addr + getpagesize();
310}
311
312
313static void *get_pages(unsigned int num)
314{
315 void *addr = from_guest_phys(guest_limit);
316
317 guest_limit += num * getpagesize();
318 if (guest_limit > guest_max)
319 errx(1, "Not enough memory for devices");
320 return addr;
321}
322
323
324
325
326
327
328static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
329{
330 ssize_t r;
331
332
333
334
335
336
337
338
339
340
341 if (mmap(addr, len, PROT_READ|PROT_WRITE,
342 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
343 return;
344
345
346 r = pread(fd, addr, len, offset);
347 if (r != len)
348 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
349}
350
351
352
353
354
355
356
357
358
359
360
361
362static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
363{
364 Elf32_Phdr phdr[ehdr->e_phnum];
365 unsigned int i;
366
367
368
369
370
371 if (ehdr->e_type != ET_EXEC
372 || ehdr->e_machine != EM_386
373 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
374 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
375 errx(1, "Malformed elf header");
376
377
378
379
380
381
382
383
384 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
385 err(1, "Seeking to program headers");
386 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
387 err(1, "Reading program headers");
388
389
390
391
392
393 for (i = 0; i < ehdr->e_phnum; i++) {
394
395 if (phdr[i].p_type != PT_LOAD)
396 continue;
397
398 verbose("Section %i: size %i addr %p\n",
399 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
400
401
402 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
403 phdr[i].p_offset, phdr[i].p_filesz);
404 }
405
406
407 return ehdr->e_entry;
408}
409
410
411
412
413
414
415
416
417
418
419static unsigned long load_bzimage(int fd)
420{
421 struct boot_params boot;
422 int r;
423
424 void *p = from_guest_phys(0x100000);
425
426
427
428
429
430 lseek(fd, 0, SEEK_SET);
431 read(fd, &boot, sizeof(boot));
432
433
434 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
435 errx(1, "This doesn't look like a bzImage to me");
436
437
438 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
439
440
441 while ((r = read(fd, p, 65536)) > 0)
442 p += r;
443
444
445 return boot.hdr.code32_start;
446}
447
448
449
450
451
452
453static unsigned long load_kernel(int fd)
454{
455 Elf32_Ehdr hdr;
456
457
458 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
459 err(1, "Reading kernel");
460
461
462 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
463 return map_elf(fd, &hdr);
464
465
466 return load_bzimage(fd);
467}
468
469
470
471
472
473
474
475
476static inline unsigned long page_align(unsigned long addr)
477{
478
479 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
480}
481
482
483
484
485
486
487
488
489
490
491static unsigned long load_initrd(const char *name, unsigned long mem)
492{
493 int ifd;
494 struct stat st;
495 unsigned long len;
496
497 ifd = open_or_die(name, O_RDONLY);
498
499 if (fstat(ifd, &st) < 0)
500 err(1, "fstat() on initrd '%s'", name);
501
502
503
504
505
506 len = page_align(st.st_size);
507 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
508
509
510
511
512 close(ifd);
513 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
514
515
516 return len;
517}
518
519
520
521
522
523
524static void concat(char *dst, char *args[])
525{
526 unsigned int i, len = 0;
527
528 for (i = 0; args[i]; i++) {
529 if (i) {
530 strcat(dst+len, " ");
531 len++;
532 }
533 strcpy(dst+len, args[i]);
534 len += strlen(args[i]);
535 }
536
537 dst[len] = '\0';
538}
539
540
541
542
543
544
545
546static void tell_kernel(unsigned long start)
547{
548 unsigned long args[] = { LHREQ_INITIALIZE,
549 (unsigned long)guest_base,
550 guest_limit / getpagesize(), start };
551 verbose("Guest: %p - %p (%#lx)\n",
552 guest_base, guest_base + guest_limit, guest_limit);
553 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
554 if (write(lguest_fd, args, sizeof(args)) < 0)
555 err(1, "Writing to /dev/lguest");
556}
557
558
559
560
561
562
563
564
565
566
567static void *_check_pointer(unsigned long addr, unsigned int size,
568 unsigned int line)
569{
570
571
572
573
574 if ((addr + size) > guest_limit || (addr + size) < addr)
575 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
576
577
578
579
580 return from_guest_phys(addr);
581}
582
583#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
584
585
586
587
588
589
590static unsigned next_desc(struct vring_desc *desc,
591 unsigned int i, unsigned int max)
592{
593 unsigned int next;
594
595
596 if (!(desc[i].flags & VRING_DESC_F_NEXT))
597 return max;
598
599
600 next = desc[i].next;
601
602 wmb();
603
604 if (next >= max)
605 errx(1, "Desc next is %u", next);
606
607 return next;
608}
609
610
611
612
613
614static void trigger_irq(struct virtqueue *vq)
615{
616 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
617
618
619 if (!vq->pending_used)
620 return;
621 vq->pending_used = 0;
622
623
624 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
625 return;
626 }
627
628
629 if (write(lguest_fd, buf, sizeof(buf)) != 0)
630 err(1, "Triggering irq %i", vq->config.irq);
631}
632
633
634
635
636
637
638
639
640
641static unsigned wait_for_vq_desc(struct virtqueue *vq,
642 struct iovec iov[],
643 unsigned int *out_num, unsigned int *in_num)
644{
645 unsigned int i, head, max;
646 struct vring_desc *desc;
647 u16 last_avail = lg_last_avail(vq);
648
649
650 while (last_avail == vq->vring.avail->idx) {
651 u64 event;
652
653
654
655
656
657 trigger_irq(vq);
658
659
660 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
661
662
663
664
665
666 mb();
667 if (last_avail != vq->vring.avail->idx) {
668 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
669 break;
670 }
671
672
673 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
674 errx(1, "Event read failed?");
675
676
677 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
678 }
679
680
681 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
682 errx(1, "Guest moved used index from %u to %u",
683 last_avail, vq->vring.avail->idx);
684
685
686
687
688
689 rmb();
690
691
692
693
694
695 head = vq->vring.avail->ring[last_avail % vq->vring.num];
696 lg_last_avail(vq)++;
697
698
699 if (head >= vq->vring.num)
700 errx(1, "Guest says index %u is available", head);
701
702
703 *out_num = *in_num = 0;
704
705 max = vq->vring.num;
706 desc = vq->vring.desc;
707 i = head;
708
709
710
711
712
713
714
715
716
717
718
719 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
720 if (desc[i].len % sizeof(struct vring_desc))
721 errx(1, "Invalid size for indirect buffer table");
722
723 max = desc[i].len / sizeof(struct vring_desc);
724 desc = check_pointer(desc[i].addr, desc[i].len);
725 i = 0;
726 }
727
728 do {
729
730 iov[*out_num + *in_num].iov_len = desc[i].len;
731 iov[*out_num + *in_num].iov_base
732 = check_pointer(desc[i].addr, desc[i].len);
733
734 if (desc[i].flags & VRING_DESC_F_WRITE)
735 (*in_num)++;
736 else {
737
738
739
740
741 if (*in_num)
742 errx(1, "Descriptor has out after in");
743 (*out_num)++;
744 }
745
746
747 if (*out_num + *in_num > max)
748 errx(1, "Looped descriptor");
749 } while ((i = next_desc(desc, i, max)) != max);
750
751 return head;
752}
753
754
755
756
757
758
759static void add_used(struct virtqueue *vq, unsigned int head, int len)
760{
761 struct vring_used_elem *used;
762
763
764
765
766
767 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
768 used->id = head;
769 used->len = len;
770
771 wmb();
772 vq->vring.used->idx++;
773 vq->pending_used++;
774}
775
776
777static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
778{
779 add_used(vq, head, len);
780 trigger_irq(vq);
781}
782
783
784
785
786
787
788struct console_abort {
789
790 int count;
791
792 struct timeval start;
793};
794
795
796static void console_input(struct virtqueue *vq)
797{
798 int len;
799 unsigned int head, in_num, out_num;
800 struct console_abort *abort = vq->dev->priv;
801 struct iovec iov[vq->vring.num];
802
803
804 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
805 if (out_num)
806 errx(1, "Output buffers in console in queue?");
807
808
809 len = readv(STDIN_FILENO, iov, in_num);
810 if (len <= 0) {
811
812 warnx("Failed to get console input, ignoring console.");
813
814
815
816
817 for (;;)
818 pause();
819 }
820
821
822 add_used_and_trigger(vq, head, len);
823
824
825
826
827
828
829
830
831
832 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
833 abort->count = 0;
834 return;
835 }
836
837 abort->count++;
838 if (abort->count == 1)
839 gettimeofday(&abort->start, NULL);
840 else if (abort->count == 3) {
841 struct timeval now;
842 gettimeofday(&now, NULL);
843
844 if (now.tv_sec <= abort->start.tv_sec+1)
845 kill(0, SIGINT);
846 abort->count = 0;
847 }
848}
849
850
851static void console_output(struct virtqueue *vq)
852{
853 unsigned int head, out, in;
854 struct iovec iov[vq->vring.num];
855
856
857 head = wait_for_vq_desc(vq, iov, &out, &in);
858 if (in)
859 errx(1, "Input buffers in console output queue?");
860
861
862 while (!iov_empty(iov, out)) {
863 int len = writev(STDOUT_FILENO, iov, out);
864 if (len <= 0) {
865 warn("Write to stdout gave %i (%d)", len, errno);
866 break;
867 }
868 iov_consume(iov, out, NULL, len);
869 }
870
871
872
873
874
875 add_used(vq, head, 0);
876}
877
878
879
880
881
882
883
884struct net_info {
885 int tunfd;
886};
887
888static void net_output(struct virtqueue *vq)
889{
890 struct net_info *net_info = vq->dev->priv;
891 unsigned int head, out, in;
892 struct iovec iov[vq->vring.num];
893
894
895 head = wait_for_vq_desc(vq, iov, &out, &in);
896 if (in)
897 errx(1, "Input buffers in net output queue?");
898
899
900
901
902 if (writev(net_info->tunfd, iov, out) < 0)
903 warnx("Write to tun failed (%d)?", errno);
904
905
906
907
908
909 add_used(vq, head, 0);
910}
911
912
913
914
915
916
917
918static bool will_block(int fd)
919{
920 fd_set fdset;
921 struct timeval zero = { 0, 0 };
922 FD_ZERO(&fdset);
923 FD_SET(fd, &fdset);
924 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
925}
926
927
928
929
930
931
932static void net_input(struct virtqueue *vq)
933{
934 int len;
935 unsigned int head, out, in;
936 struct iovec iov[vq->vring.num];
937 struct net_info *net_info = vq->dev->priv;
938
939
940
941
942
943 head = wait_for_vq_desc(vq, iov, &out, &in);
944 if (out)
945 errx(1, "Output buffers in net input queue?");
946
947
948
949
950
951 if (vq->pending_used && will_block(net_info->tunfd))
952 trigger_irq(vq);
953
954
955
956
957
958 len = readv(net_info->tunfd, iov, in);
959 if (len <= 0)
960 warn("Failed to read from tun (%d).", errno);
961
962
963
964
965
966 add_used(vq, head, len);
967}
968
969
970
971static int do_thread(void *_vq)
972{
973 struct virtqueue *vq = _vq;
974
975 for (;;)
976 vq->service(vq);
977 return 0;
978}
979
980
981
982
983
984static void kill_launcher(int signal)
985{
986 kill(0, SIGTERM);
987}
988
989static void reset_device(struct device *dev)
990{
991 struct virtqueue *vq;
992
993 verbose("Resetting device %s\n", dev->name);
994
995
996 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
997
998
999 signal(SIGCHLD, SIG_IGN);
1000
1001
1002 for (vq = dev->vq; vq; vq = vq->next) {
1003 if (vq->thread != (pid_t)-1) {
1004 kill(vq->thread, SIGTERM);
1005 waitpid(vq->thread, NULL, 0);
1006 vq->thread = (pid_t)-1;
1007 }
1008 memset(vq->vring.desc, 0,
1009 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1010 lg_last_avail(vq) = 0;
1011 }
1012 dev->running = false;
1013
1014
1015 signal(SIGCHLD, (void *)kill_launcher);
1016}
1017
1018
1019
1020
1021static void create_thread(struct virtqueue *vq)
1022{
1023
1024
1025
1026
1027 char *stack = malloc(32768);
1028 unsigned long args[] = { LHREQ_EVENTFD,
1029 vq->config.pfn*getpagesize(), 0 };
1030
1031
1032 vq->eventfd = eventfd(0, 0);
1033 if (vq->eventfd < 0)
1034 err(1, "Creating eventfd");
1035 args[2] = vq->eventfd;
1036
1037
1038
1039
1040
1041 if (write(lguest_fd, &args, sizeof(args)) != 0)
1042 err(1, "Attaching eventfd");
1043
1044
1045
1046
1047
1048 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1049 if (vq->thread == (pid_t)-1)
1050 err(1, "Creating clone");
1051
1052
1053 close(vq->eventfd);
1054}
1055
1056static void start_device(struct device *dev)
1057{
1058 unsigned int i;
1059 struct virtqueue *vq;
1060
1061 verbose("Device %s OK: offered", dev->name);
1062 for (i = 0; i < dev->feature_len; i++)
1063 verbose(" %02x", get_feature_bits(dev)[i]);
1064 verbose(", accepted");
1065 for (i = 0; i < dev->feature_len; i++)
1066 verbose(" %02x", get_feature_bits(dev)
1067 [dev->feature_len+i]);
1068
1069 for (vq = dev->vq; vq; vq = vq->next) {
1070 if (vq->service)
1071 create_thread(vq);
1072 }
1073 dev->running = true;
1074}
1075
1076static void cleanup_devices(void)
1077{
1078 struct device *dev;
1079
1080 for (dev = devices.dev; dev; dev = dev->next)
1081 reset_device(dev);
1082
1083
1084 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1085 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1086}
1087
1088
1089static void update_device_status(struct device *dev)
1090{
1091
1092 if (dev->desc->status == 0)
1093 reset_device(dev);
1094 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1095 warnx("Device %s configuration FAILED", dev->name);
1096 if (dev->running)
1097 reset_device(dev);
1098 } else {
1099 if (dev->running)
1100 err(1, "Device %s features finalized twice", dev->name);
1101 start_device(dev);
1102 }
1103}
1104
1105
1106
1107
1108
1109static void handle_output(unsigned long addr)
1110{
1111 struct device *i;
1112
1113
1114 for (i = devices.dev; i; i = i->next) {
1115 struct virtqueue *vq;
1116
1117
1118
1119
1120
1121 if (from_guest_phys(addr) == i->desc) {
1122 update_device_status(i);
1123 return;
1124 }
1125
1126
1127 for (vq = i->vq; vq; vq = vq->next) {
1128 if (addr != vq->config.pfn*getpagesize())
1129 continue;
1130 errx(1, "Notification on %s before setup!", i->name);
1131 }
1132 }
1133
1134
1135
1136
1137
1138
1139 if (addr >= guest_limit)
1140 errx(1, "Bad NOTIFY %#lx", addr);
1141
1142 write(STDOUT_FILENO, from_guest_phys(addr),
1143 strnlen(from_guest_phys(addr), guest_limit - addr));
1144}
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160static u8 *device_config(const struct device *dev)
1161{
1162 return (void *)(dev->desc + 1)
1163 + dev->num_vq * sizeof(struct lguest_vqconfig)
1164 + dev->feature_len * 2;
1165}
1166
1167
1168
1169
1170
1171
1172static struct lguest_device_desc *new_dev_desc(u16 type)
1173{
1174 struct lguest_device_desc d = { .type = type };
1175 void *p;
1176
1177
1178 if (devices.lastdev)
1179 p = device_config(devices.lastdev)
1180 + devices.lastdev->desc->config_len;
1181 else
1182 p = devices.descpage;
1183
1184
1185 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1186 errx(1, "Too many devices");
1187
1188
1189 return memcpy(p, &d, sizeof(d));
1190}
1191
1192
1193
1194
1195
1196static void add_virtqueue(struct device *dev, unsigned int num_descs,
1197 void (*service)(struct virtqueue *))
1198{
1199 unsigned int pages;
1200 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1201 void *p;
1202
1203
1204 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1205 / getpagesize();
1206 p = get_pages(pages);
1207
1208
1209 vq->next = NULL;
1210 vq->last_avail_idx = 0;
1211 vq->dev = dev;
1212
1213
1214
1215
1216
1217 vq->service = service;
1218 vq->thread = (pid_t)-1;
1219
1220
1221 vq->config.num = num_descs;
1222 vq->config.irq = devices.next_irq++;
1223 vq->config.pfn = to_guest_phys(p) / getpagesize();
1224
1225
1226 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1227
1228
1229
1230
1231
1232
1233
1234 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1235 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1236 dev->num_vq++;
1237 dev->desc->num_vq++;
1238
1239 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1240
1241
1242
1243
1244
1245 for (i = &dev->vq; *i; i = &(*i)->next);
1246 *i = vq;
1247}
1248
1249
1250
1251
1252
1253static void add_feature(struct device *dev, unsigned bit)
1254{
1255 u8 *features = get_feature_bits(dev);
1256
1257
1258 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1259 assert(dev->desc->config_len == 0);
1260 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1261 }
1262
1263 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1264}
1265
1266
1267
1268
1269
1270
1271static void set_config(struct device *dev, unsigned len, const void *conf)
1272{
1273
1274 if (device_config(dev) + len > devices.descpage + getpagesize())
1275 errx(1, "Too many devices");
1276
1277
1278 memcpy(device_config(dev), conf, len);
1279 dev->desc->config_len = len;
1280
1281
1282 assert(dev->desc->config_len == len);
1283}
1284
1285
1286
1287
1288
1289
1290
1291
1292static struct device *new_device(const char *name, u16 type)
1293{
1294 struct device *dev = malloc(sizeof(*dev));
1295
1296
1297 dev->desc = new_dev_desc(type);
1298 dev->name = name;
1299 dev->vq = NULL;
1300 dev->feature_len = 0;
1301 dev->num_vq = 0;
1302 dev->running = false;
1303 dev->next = NULL;
1304
1305
1306
1307
1308
1309
1310
1311 if (devices.lastdev)
1312 devices.lastdev->next = dev;
1313 else
1314 devices.dev = dev;
1315 devices.lastdev = dev;
1316
1317 return dev;
1318}
1319
1320
1321
1322
1323
1324static void setup_console(void)
1325{
1326 struct device *dev;
1327
1328
1329 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1330 struct termios term = orig_term;
1331
1332
1333
1334
1335 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1336 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1337 }
1338
1339 dev = new_device("console", VIRTIO_ID_CONSOLE);
1340
1341
1342 dev->priv = malloc(sizeof(struct console_abort));
1343 ((struct console_abort *)dev->priv)->count = 0;
1344
1345
1346
1347
1348
1349
1350
1351 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1352 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1353
1354 verbose("device %u: console\n", ++devices.device_num);
1355}
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376static u32 str2ip(const char *ipaddr)
1377{
1378 unsigned int b[4];
1379
1380 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1381 errx(1, "Failed to parse IP address '%s'", ipaddr);
1382 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1383}
1384
1385static void str2mac(const char *macaddr, unsigned char mac[6])
1386{
1387 unsigned int m[6];
1388 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1389 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1390 errx(1, "Failed to parse mac address '%s'", macaddr);
1391 mac[0] = m[0];
1392 mac[1] = m[1];
1393 mac[2] = m[2];
1394 mac[3] = m[3];
1395 mac[4] = m[4];
1396 mac[5] = m[5];
1397}
1398
1399
1400
1401
1402
1403
1404
1405
1406static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1407{
1408 int ifidx;
1409 struct ifreq ifr;
1410
1411 if (!*br_name)
1412 errx(1, "must specify bridge name");
1413
1414 ifidx = if_nametoindex(if_name);
1415 if (!ifidx)
1416 errx(1, "interface %s does not exist!", if_name);
1417
1418 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1419 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1420 ifr.ifr_ifindex = ifidx;
1421 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1422 err(1, "can't add %s to bridge %s", if_name, br_name);
1423}
1424
1425
1426
1427
1428
1429
1430static void configure_device(int fd, const char *tapif, u32 ipaddr)
1431{
1432 struct ifreq ifr;
1433 struct sockaddr_in sin;
1434
1435 memset(&ifr, 0, sizeof(ifr));
1436 strcpy(ifr.ifr_name, tapif);
1437
1438
1439 sin.sin_family = AF_INET;
1440 sin.sin_addr.s_addr = htonl(ipaddr);
1441 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1442 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1443 err(1, "Setting %s interface address", tapif);
1444 ifr.ifr_flags = IFF_UP;
1445 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1446 err(1, "Bringing interface %s up", tapif);
1447}
1448
1449static int get_tun_device(char tapif[IFNAMSIZ])
1450{
1451 struct ifreq ifr;
1452 int netfd;
1453
1454
1455 memset(&ifr, 0, sizeof(ifr));
1456
1457
1458
1459
1460
1461
1462
1463 netfd = open_or_die("/dev/net/tun", O_RDWR);
1464 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1465 strcpy(ifr.ifr_name, "tap%d");
1466 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1467 err(1, "configuring /dev/net/tun");
1468
1469 if (ioctl(netfd, TUNSETOFFLOAD,
1470 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1471 err(1, "Could not set features for tun device");
1472
1473
1474
1475
1476
1477 ioctl(netfd, TUNSETNOCSUM, 1);
1478
1479 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1480 return netfd;
1481}
1482
1483
1484
1485
1486
1487
1488
1489static void setup_tun_net(char *arg)
1490{
1491 struct device *dev;
1492 struct net_info *net_info = malloc(sizeof(*net_info));
1493 int ipfd;
1494 u32 ip = INADDR_ANY;
1495 bool bridging = false;
1496 char tapif[IFNAMSIZ], *p;
1497 struct virtio_net_config conf;
1498
1499 net_info->tunfd = get_tun_device(tapif);
1500
1501
1502 dev = new_device("net", VIRTIO_ID_NET);
1503 dev->priv = net_info;
1504
1505
1506 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1507 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1508
1509
1510
1511
1512
1513 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1514 if (ipfd < 0)
1515 err(1, "opening IP socket");
1516
1517
1518 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1519 arg += strlen(BRIDGE_PFX);
1520 bridging = true;
1521 }
1522
1523
1524 p = strchr(arg, ':');
1525 if (p) {
1526 str2mac(p+1, conf.mac);
1527 add_feature(dev, VIRTIO_NET_F_MAC);
1528 *p = '\0';
1529 }
1530
1531
1532 if (bridging)
1533 add_to_bridge(ipfd, tapif, arg);
1534 else
1535 ip = str2ip(arg);
1536
1537
1538 configure_device(ipfd, tapif, ip);
1539
1540
1541 add_feature(dev, VIRTIO_NET_F_CSUM);
1542 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1543 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1544 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1545 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1546 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1547 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1548 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1549
1550 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1551
1552 add_feature(dev, VIRTIO_F_ANY_LAYOUT);
1553 set_config(dev, sizeof(conf), &conf);
1554
1555
1556 close(ipfd);
1557
1558 devices.device_num++;
1559
1560 if (bridging)
1561 verbose("device %u: tun %s attached to bridge: %s\n",
1562 devices.device_num, tapif, arg);
1563 else
1564 verbose("device %u: tun %s: %s\n",
1565 devices.device_num, tapif, arg);
1566}
1567
1568
1569
1570struct vblk_info {
1571
1572 off64_t len;
1573
1574
1575 int fd;
1576
1577};
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593static void blk_request(struct virtqueue *vq)
1594{
1595 struct vblk_info *vblk = vq->dev->priv;
1596 unsigned int head, out_num, in_num, wlen;
1597 int ret, i;
1598 u8 *in;
1599 struct virtio_blk_outhdr out;
1600 struct iovec iov[vq->vring.num];
1601 off64_t off;
1602
1603
1604
1605
1606
1607 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1608
1609
1610 iov_consume(iov, out_num, &out, sizeof(out));
1611
1612
1613 in = NULL;
1614 for (i = out_num + in_num - 1; i >= out_num; i--) {
1615 if (iov[i].iov_len > 0) {
1616 in = iov[i].iov_base + iov[i].iov_len - 1;
1617 iov[i].iov_len--;
1618 break;
1619 }
1620 }
1621 if (!in)
1622 errx(1, "Bad virtblk cmd with no room for status");
1623
1624
1625
1626
1627
1628 off = out.sector * 512;
1629
1630
1631
1632
1633
1634 if (out.type & VIRTIO_BLK_T_SCSI_CMD) {
1635 fprintf(stderr, "Scsi commands unsupported\n");
1636 *in = VIRTIO_BLK_S_UNSUPP;
1637 wlen = sizeof(*in);
1638 } else if (out.type & VIRTIO_BLK_T_OUT) {
1639
1640
1641
1642
1643
1644
1645 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1646 err(1, "Bad seek to sector %llu", out.sector);
1647
1648 ret = writev(vblk->fd, iov, out_num);
1649 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
1650
1651
1652
1653
1654
1655
1656 if (ret > 0 && off + ret > vblk->len) {
1657
1658 ftruncate64(vblk->fd, vblk->len);
1659
1660 errx(1, "Write past end %llu+%u", off, ret);
1661 }
1662
1663 wlen = sizeof(*in);
1664 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1665 } else if (out.type & VIRTIO_BLK_T_FLUSH) {
1666
1667 ret = fdatasync(vblk->fd);
1668 verbose("FLUSH fdatasync: %i\n", ret);
1669 wlen = sizeof(*in);
1670 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1671 } else {
1672
1673
1674
1675
1676
1677
1678 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1679 err(1, "Bad seek to sector %llu", out.sector);
1680
1681 ret = readv(vblk->fd, iov + out_num, in_num);
1682 if (ret >= 0) {
1683 wlen = sizeof(*in) + ret;
1684 *in = VIRTIO_BLK_S_OK;
1685 } else {
1686 wlen = sizeof(*in);
1687 *in = VIRTIO_BLK_S_IOERR;
1688 }
1689 }
1690
1691
1692 add_used(vq, head, wlen);
1693}
1694
1695
1696static void setup_block_file(const char *filename)
1697{
1698 struct device *dev;
1699 struct vblk_info *vblk;
1700 struct virtio_blk_config conf;
1701
1702
1703 dev = new_device("block", VIRTIO_ID_BLOCK);
1704
1705
1706 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1707
1708
1709 vblk = dev->priv = malloc(sizeof(*vblk));
1710
1711
1712 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1713 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1714
1715
1716 add_feature(dev, VIRTIO_BLK_F_FLUSH);
1717
1718
1719 conf.capacity = cpu_to_le64(vblk->len / 512);
1720
1721
1722
1723
1724
1725 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1726 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1727
1728
1729 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1730
1731 verbose("device %u: virtblock %llu sectors\n",
1732 ++devices.device_num, le64_to_cpu(conf.capacity));
1733}
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743struct rng_info {
1744 int rfd;
1745};
1746
1747static void rng_input(struct virtqueue *vq)
1748{
1749 int len;
1750 unsigned int head, in_num, out_num, totlen = 0;
1751 struct rng_info *rng_info = vq->dev->priv;
1752 struct iovec iov[vq->vring.num];
1753
1754
1755 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1756 if (out_num)
1757 errx(1, "Output buffers in rng?");
1758
1759
1760
1761
1762
1763 while (!iov_empty(iov, in_num)) {
1764 len = readv(rng_info->rfd, iov, in_num);
1765 if (len <= 0)
1766 err(1, "Read from /dev/random gave %i", len);
1767 iov_consume(iov, in_num, NULL, len);
1768 totlen += len;
1769 }
1770
1771
1772 add_used(vq, head, totlen);
1773}
1774
1775
1776
1777
1778static void setup_rng(void)
1779{
1780 struct device *dev;
1781 struct rng_info *rng_info = malloc(sizeof(*rng_info));
1782
1783
1784 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1785
1786
1787 dev = new_device("rng", VIRTIO_ID_RNG);
1788 dev->priv = rng_info;
1789
1790
1791 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1792
1793 verbose("device %u: rng\n", devices.device_num++);
1794}
1795
1796
1797
1798static void __attribute__((noreturn)) restart_guest(void)
1799{
1800 unsigned int i;
1801
1802
1803
1804
1805
1806 for (i = 3; i < FD_SETSIZE; i++)
1807 close(i);
1808
1809
1810 cleanup_devices();
1811
1812 execv(main_args[0], main_args);
1813 err(1, "Could not exec %s", main_args[0]);
1814}
1815
1816
1817
1818
1819
1820static void __attribute__((noreturn)) run_guest(void)
1821{
1822 for (;;) {
1823 unsigned long notify_addr;
1824 int readval;
1825
1826
1827 readval = pread(lguest_fd, ¬ify_addr,
1828 sizeof(notify_addr), cpu_id);
1829
1830
1831 if (readval == sizeof(notify_addr)) {
1832 verbose("Notify on address %#lx\n", notify_addr);
1833 handle_output(notify_addr);
1834
1835 } else if (errno == ENOENT) {
1836 char reason[1024] = { 0 };
1837 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1838 errx(1, "%s", reason);
1839
1840 } else if (errno == ERESTART) {
1841 restart_guest();
1842
1843 } else
1844 err(1, "Running guest failed");
1845 }
1846}
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856static struct option opts[] = {
1857 { "verbose", 0, NULL, 'v' },
1858 { "tunnet", 1, NULL, 't' },
1859 { "block", 1, NULL, 'b' },
1860 { "rng", 0, NULL, 'r' },
1861 { "initrd", 1, NULL, 'i' },
1862 { "username", 1, NULL, 'u' },
1863 { "chroot", 1, NULL, 'c' },
1864 { NULL },
1865};
1866static void usage(void)
1867{
1868 errx(1, "Usage: lguest [--verbose] "
1869 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1870 "|--block=<filename>|--initrd=<filename>]...\n"
1871 "<mem-in-mb> vmlinux [args...]");
1872}
1873
1874
1875int main(int argc, char *argv[])
1876{
1877
1878 unsigned long mem = 0, start, initrd_size = 0;
1879
1880 int i, c;
1881
1882 struct boot_params *boot;
1883
1884 const char *initrd_name = NULL;
1885
1886
1887 struct passwd *user_details = NULL;
1888
1889
1890 char *chroot_path = NULL;
1891
1892
1893 main_args = argv;
1894
1895
1896
1897
1898
1899
1900 devices.lastdev = NULL;
1901 devices.next_irq = 1;
1902
1903
1904 cpu_id = 0;
1905
1906
1907
1908
1909
1910
1911
1912 for (i = 1; i < argc; i++) {
1913 if (argv[i][0] != '-') {
1914 mem = atoi(argv[i]) * 1024 * 1024;
1915
1916
1917
1918
1919
1920
1921 guest_base = map_zeroed_pages(mem / getpagesize()
1922 + DEVICE_PAGES);
1923 guest_limit = mem;
1924 guest_max = mem + DEVICE_PAGES*getpagesize();
1925 devices.descpage = get_pages(1);
1926 break;
1927 }
1928 }
1929
1930
1931 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1932 switch (c) {
1933 case 'v':
1934 verbose = true;
1935 break;
1936 case 't':
1937 setup_tun_net(optarg);
1938 break;
1939 case 'b':
1940 setup_block_file(optarg);
1941 break;
1942 case 'r':
1943 setup_rng();
1944 break;
1945 case 'i':
1946 initrd_name = optarg;
1947 break;
1948 case 'u':
1949 user_details = getpwnam(optarg);
1950 if (!user_details)
1951 err(1, "getpwnam failed, incorrect username?");
1952 break;
1953 case 'c':
1954 chroot_path = optarg;
1955 break;
1956 default:
1957 warnx("Unknown argument %s", argv[optind]);
1958 usage();
1959 }
1960 }
1961
1962
1963
1964
1965 if (optind + 2 > argc)
1966 usage();
1967
1968 verbose("Guest base is at %p\n", guest_base);
1969
1970
1971 setup_console();
1972
1973
1974 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1975
1976
1977 boot = from_guest_phys(0);
1978
1979
1980 if (initrd_name) {
1981 initrd_size = load_initrd(initrd_name, mem);
1982
1983
1984
1985
1986 boot->hdr.ramdisk_image = mem - initrd_size;
1987 boot->hdr.ramdisk_size = initrd_size;
1988
1989 boot->hdr.type_of_loader = 0xFF;
1990 }
1991
1992
1993
1994
1995
1996 boot->e820_entries = 1;
1997 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1998
1999
2000
2001
2002 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2003
2004 concat((char *)(boot + 1), argv+optind+2);
2005
2006
2007 boot->hdr.kernel_alignment = 0x1000000;
2008
2009
2010 boot->hdr.version = 0x207;
2011
2012
2013 boot->hdr.hardware_subarch = 1;
2014
2015
2016 boot->hdr.loadflags |= KEEP_SEGMENTS;
2017
2018
2019 tell_kernel(start);
2020
2021
2022 signal(SIGCHLD, kill_launcher);
2023
2024
2025 atexit(cleanup_devices);
2026
2027
2028 if (chroot_path) {
2029 if (chroot(chroot_path) != 0)
2030 err(1, "chroot(\"%s\") failed", chroot_path);
2031
2032 if (chdir("/") != 0)
2033 err(1, "chdir(\"/\") failed");
2034
2035 verbose("chroot done\n");
2036 }
2037
2038
2039 if (user_details) {
2040 uid_t u;
2041 gid_t g;
2042
2043 u = user_details->pw_uid;
2044 g = user_details->pw_gid;
2045
2046 if (initgroups(user_details->pw_name, g) != 0)
2047 err(1, "initgroups failed");
2048
2049 if (setresgid(g, g, g) != 0)
2050 err(1, "setresgid failed");
2051
2052 if (setresuid(u, u, u) != 0)
2053 err(1, "setresuid failed");
2054
2055 verbose("Dropping privileges completed\n");
2056 }
2057
2058
2059 run_guest();
2060}
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073