Root/tools/lguest/lguest.c

1/*P:100
2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5 * control it.
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#include <linux/virtio_config.h>
46#include <linux/virtio_net.h>
47#include <linux/virtio_blk.h>
48#include <linux/virtio_console.h>
49#include <linux/virtio_rng.h>
50#include <linux/virtio_ring.h>
51#include <asm/bootparam.h>
52#include "../../include/linux/lguest_launcher.h"
53/*L:110
54 * We can ignore the 43 include files we need for this program, but I do want
55 * to draw attention to the use of kernel-style types.
56 *
57 * As Linus said, "C is a Spartan language, and so should your naming be." I
58 * like these abbreviations, so we define them here. Note that u64 is always
59 * unsigned long long, which works on all Linux systems: this means that we can
60 * use %llu in printf for any u64.
61 */
62typedef unsigned long long u64;
63typedef uint32_t u32;
64typedef uint16_t u16;
65typedef uint8_t u8;
66/*:*/
67
68#define BRIDGE_PFX "bridge:"
69#ifndef SIOCBRADDIF
70#define SIOCBRADDIF 0x89a2 /* add interface to bridge */
71#endif
72/* We can have up to 256 pages for devices. */
73#define DEVICE_PAGES 256
74/* This will occupy 3 pages: it must be a power of 2. */
75#define VIRTQUEUE_NUM 256
76
77/*L:120
78 * verbose is both a global flag and a macro. The C preprocessor allows
79 * this, and although I wouldn't recommend it, it works quite nicely here.
80 */
81static bool verbose;
82#define verbose(args...) \
83    do { if (verbose) printf(args); } while(0)
84/*:*/
85
86/* The pointer to the start of guest memory. */
87static void *guest_base;
88/* The maximum guest physical address allowed, and maximum possible. */
89static unsigned long guest_limit, guest_max;
90/* The /dev/lguest file descriptor. */
91static int lguest_fd;
92
93/* a per-cpu variable indicating whose vcpu is currently running */
94static unsigned int __thread cpu_id;
95
96/* This is our list of devices. */
97struct device_list {
98    /* Counter to assign interrupt numbers. */
99    unsigned int next_irq;
100
101    /* Counter to print out convenient device numbers. */
102    unsigned int device_num;
103
104    /* The descriptor page for the devices. */
105    u8 *descpage;
106
107    /* A single linked list of devices. */
108    struct device *dev;
109    /* And a pointer to the last device for easy append. */
110    struct device *lastdev;
111};
112
113/* The list of Guest devices, based on command line arguments. */
114static struct device_list devices;
115
116/* The device structure describes a single device. */
117struct device {
118    /* The linked-list pointer. */
119    struct device *next;
120
121    /* The device's descriptor, as mapped into the Guest. */
122    struct lguest_device_desc *desc;
123
124    /* We can't trust desc values once Guest has booted: we use these. */
125    unsigned int feature_len;
126    unsigned int num_vq;
127
128    /* The name of this device, for --verbose. */
129    const char *name;
130
131    /* Any queues attached to this device */
132    struct virtqueue *vq;
133
134    /* Is it operational */
135    bool running;
136
137    /* Device-specific data. */
138    void *priv;
139};
140
141/* The virtqueue structure describes a queue attached to a device. */
142struct virtqueue {
143    struct virtqueue *next;
144
145    /* Which device owns me. */
146    struct device *dev;
147
148    /* The configuration for this queue. */
149    struct lguest_vqconfig config;
150
151    /* The actual ring of buffers. */
152    struct vring vring;
153
154    /* Last available index we saw. */
155    u16 last_avail_idx;
156
157    /* How many are used since we sent last irq? */
158    unsigned int pending_used;
159
160    /* Eventfd where Guest notifications arrive. */
161    int eventfd;
162
163    /* Function for the thread which is servicing this virtqueue. */
164    void (*service)(struct virtqueue *vq);
165    pid_t thread;
166};
167
168/* Remember the arguments to the program so we can "reboot" */
169static char **main_args;
170
171/* The original tty settings to restore on exit. */
172static struct termios orig_term;
173
174/*
175 * We have to be careful with barriers: our devices are all run in separate
176 * threads and so we need to make sure that changes visible to the Guest happen
177 * in precise order.
178 */
179#define wmb() __asm__ __volatile__("" : : : "memory")
180#define mb() __asm__ __volatile__("" : : : "memory")
181
182/*
183 * Convert an iovec element to the given type.
184 *
185 * This is a fairly ugly trick: we need to know the size of the type and
186 * alignment requirement to check the pointer is kosher. It's also nice to
187 * have the name of the type in case we report failure.
188 *
189 * Typing those three things all the time is cumbersome and error prone, so we
190 * have a macro which sets them all up and passes to the real function.
191 */
192#define convert(iov, type) \
193    ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
194
195static void *_convert(struct iovec *iov, size_t size, size_t align,
196              const char *name)
197{
198    if (iov->iov_len != size)
199        errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
200    if ((unsigned long)iov->iov_base % align != 0)
201        errx(1, "Bad alignment %p for %s", iov->iov_base, name);
202    return iov->iov_base;
203}
204
205/* Wrapper for the last available index. Makes it easier to change. */
206#define lg_last_avail(vq) ((vq)->last_avail_idx)
207
208/*
209 * The virtio configuration space is defined to be little-endian. x86 is
210 * little-endian too, but it's nice to be explicit so we have these helpers.
211 */
212#define cpu_to_le16(v16) (v16)
213#define cpu_to_le32(v32) (v32)
214#define cpu_to_le64(v64) (v64)
215#define le16_to_cpu(v16) (v16)
216#define le32_to_cpu(v32) (v32)
217#define le64_to_cpu(v64) (v64)
218
219/* Is this iovec empty? */
220static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
221{
222    unsigned int i;
223
224    for (i = 0; i < num_iov; i++)
225        if (iov[i].iov_len)
226            return false;
227    return true;
228}
229
230/* Take len bytes from the front of this iovec. */
231static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
232{
233    unsigned int i;
234
235    for (i = 0; i < num_iov; i++) {
236        unsigned int used;
237
238        used = iov[i].iov_len < len ? iov[i].iov_len : len;
239        iov[i].iov_base += used;
240        iov[i].iov_len -= used;
241        len -= used;
242    }
243    assert(len == 0);
244}
245
246/* The device virtqueue descriptors are followed by feature bitmasks. */
247static u8 *get_feature_bits(struct device *dev)
248{
249    return (u8 *)(dev->desc + 1)
250        + dev->num_vq * sizeof(struct lguest_vqconfig);
251}
252
253/*L:100
254 * The Launcher code itself takes us out into userspace, that scary place where
255 * pointers run wild and free! Unfortunately, like most userspace programs,
256 * it's quite boring (which is why everyone likes to hack on the kernel!).
257 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
258 * you through this section. Or, maybe not.
259 *
260 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
261 * memory and stores it in "guest_base". In other words, Guest physical ==
262 * Launcher virtual with an offset.
263 *
264 * This can be tough to get your head around, but usually it just means that we
265 * use these trivial conversion functions when the Guest gives us its
266 * "physical" addresses:
267 */
268static void *from_guest_phys(unsigned long addr)
269{
270    return guest_base + addr;
271}
272
273static unsigned long to_guest_phys(const void *addr)
274{
275    return (addr - guest_base);
276}
277
278/*L:130
279 * Loading the Kernel.
280 *
281 * We start with couple of simple helper routines. open_or_die() avoids
282 * error-checking code cluttering the callers:
283 */
284static int open_or_die(const char *name, int flags)
285{
286    int fd = open(name, flags);
287    if (fd < 0)
288        err(1, "Failed to open %s", name);
289    return fd;
290}
291
292/* map_zeroed_pages() takes a number of pages. */
293static void *map_zeroed_pages(unsigned int num)
294{
295    int fd = open_or_die("/dev/zero", O_RDONLY);
296    void *addr;
297
298    /*
299     * We use a private mapping (ie. if we write to the page, it will be
300     * copied). We allocate an extra two pages PROT_NONE to act as guard
301     * pages against read/write attempts that exceed allocated space.
302     */
303    addr = mmap(NULL, getpagesize() * (num+2),
304            PROT_NONE, MAP_PRIVATE, fd, 0);
305
306    if (addr == MAP_FAILED)
307        err(1, "Mmapping %u pages of /dev/zero", num);
308
309    if (mprotect(addr + getpagesize(), getpagesize() * num,
310             PROT_READ|PROT_WRITE) == -1)
311        err(1, "mprotect rw %u pages failed", num);
312
313    /*
314     * One neat mmap feature is that you can close the fd, and it
315     * stays mapped.
316     */
317    close(fd);
318
319    /* Return address after PROT_NONE page */
320    return addr + getpagesize();
321}
322
323/* Get some more pages for a device. */
324static void *get_pages(unsigned int num)
325{
326    void *addr = from_guest_phys(guest_limit);
327
328    guest_limit += num * getpagesize();
329    if (guest_limit > guest_max)
330        errx(1, "Not enough memory for devices");
331    return addr;
332}
333
334/*
335 * This routine is used to load the kernel or initrd. It tries mmap, but if
336 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
337 * it falls back to reading the memory in.
338 */
339static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
340{
341    ssize_t r;
342
343    /*
344     * We map writable even though for some segments are marked read-only.
345     * The kernel really wants to be writable: it patches its own
346     * instructions.
347     *
348     * MAP_PRIVATE means that the page won't be copied until a write is
349     * done to it. This allows us to share untouched memory between
350     * Guests.
351     */
352    if (mmap(addr, len, PROT_READ|PROT_WRITE,
353         MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
354        return;
355
356    /* pread does a seek and a read in one shot: saves a few lines. */
357    r = pread(fd, addr, len, offset);
358    if (r != len)
359        err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
360}
361
362/*
363 * This routine takes an open vmlinux image, which is in ELF, and maps it into
364 * the Guest memory. ELF = Embedded Linking Format, which is the format used
365 * by all modern binaries on Linux including the kernel.
366 *
367 * The ELF headers give *two* addresses: a physical address, and a virtual
368 * address. We use the physical address; the Guest will map itself to the
369 * virtual address.
370 *
371 * We return the starting address.
372 */
373static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
374{
375    Elf32_Phdr phdr[ehdr->e_phnum];
376    unsigned int i;
377
378    /*
379     * Sanity checks on the main ELF header: an x86 executable with a
380     * reasonable number of correctly-sized program headers.
381     */
382    if (ehdr->e_type != ET_EXEC
383        || ehdr->e_machine != EM_386
384        || ehdr->e_phentsize != sizeof(Elf32_Phdr)
385        || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
386        errx(1, "Malformed elf header");
387
388    /*
389     * An ELF executable contains an ELF header and a number of "program"
390     * headers which indicate which parts ("segments") of the program to
391     * load where.
392     */
393
394    /* We read in all the program headers at once: */
395    if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
396        err(1, "Seeking to program headers");
397    if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
398        err(1, "Reading program headers");
399
400    /*
401     * Try all the headers: there are usually only three. A read-only one,
402     * a read-write one, and a "note" section which we don't load.
403     */
404    for (i = 0; i < ehdr->e_phnum; i++) {
405        /* If this isn't a loadable segment, we ignore it */
406        if (phdr[i].p_type != PT_LOAD)
407            continue;
408
409        verbose("Section %i: size %i addr %p\n",
410            i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
411
412        /* We map this section of the file at its physical address. */
413        map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
414               phdr[i].p_offset, phdr[i].p_filesz);
415    }
416
417    /* The entry point is given in the ELF header. */
418    return ehdr->e_entry;
419}
420
421/*L:150
422 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
423 * to jump into it and it will unpack itself. We used to have to perform some
424 * hairy magic because the unpacking code scared me.
425 *
426 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
427 * a small patch to jump over the tricky bits in the Guest, so now we just read
428 * the funky header so we know where in the file to load, and away we go!
429 */
430static unsigned long load_bzimage(int fd)
431{
432    struct boot_params boot;
433    int r;
434    /* Modern bzImages get loaded at 1M. */
435    void *p = from_guest_phys(0x100000);
436
437    /*
438     * Go back to the start of the file and read the header. It should be
439     * a Linux boot header (see Documentation/x86/boot.txt)
440     */
441    lseek(fd, 0, SEEK_SET);
442    read(fd, &boot, sizeof(boot));
443
444    /* Inside the setup_hdr, we expect the magic "HdrS" */
445    if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
446        errx(1, "This doesn't look like a bzImage to me");
447
448    /* Skip over the extra sectors of the header. */
449    lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
450
451    /* Now read everything into memory. in nice big chunks. */
452    while ((r = read(fd, p, 65536)) > 0)
453        p += r;
454
455    /* Finally, code32_start tells us where to enter the kernel. */
456    return boot.hdr.code32_start;
457}
458
459/*L:140
460 * Loading the kernel is easy when it's a "vmlinux", but most kernels
461 * come wrapped up in the self-decompressing "bzImage" format. With a little
462 * work, we can load those, too.
463 */
464static unsigned long load_kernel(int fd)
465{
466    Elf32_Ehdr hdr;
467
468    /* Read in the first few bytes. */
469    if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
470        err(1, "Reading kernel");
471
472    /* If it's an ELF file, it starts with "\177ELF" */
473    if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
474        return map_elf(fd, &hdr);
475
476    /* Otherwise we assume it's a bzImage, and try to load it. */
477    return load_bzimage(fd);
478}
479
480/*
481 * This is a trivial little helper to align pages. Andi Kleen hated it because
482 * it calls getpagesize() twice: "it's dumb code."
483 *
484 * Kernel guys get really het up about optimization, even when it's not
485 * necessary. I leave this code as a reaction against that.
486 */
487static inline unsigned long page_align(unsigned long addr)
488{
489    /* Add upwards and truncate downwards. */
490    return ((addr + getpagesize()-1) & ~(getpagesize()-1));
491}
492
493/*L:180
494 * An "initial ram disk" is a disk image loaded into memory along with the
495 * kernel which the kernel can use to boot from without needing any drivers.
496 * Most distributions now use this as standard: the initrd contains the code to
497 * load the appropriate driver modules for the current machine.
498 *
499 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
500 * kernels. He sent me this (and tells me when I break it).
501 */
502static unsigned long load_initrd(const char *name, unsigned long mem)
503{
504    int ifd;
505    struct stat st;
506    unsigned long len;
507
508    ifd = open_or_die(name, O_RDONLY);
509    /* fstat() is needed to get the file size. */
510    if (fstat(ifd, &st) < 0)
511        err(1, "fstat() on initrd '%s'", name);
512
513    /*
514     * We map the initrd at the top of memory, but mmap wants it to be
515     * page-aligned, so we round the size up for that.
516     */
517    len = page_align(st.st_size);
518    map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
519    /*
520     * Once a file is mapped, you can close the file descriptor. It's a
521     * little odd, but quite useful.
522     */
523    close(ifd);
524    verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
525
526    /* We return the initrd size. */
527    return len;
528}
529/*:*/
530
531/*
532 * Simple routine to roll all the commandline arguments together with spaces
533 * between them.
534 */
535static void concat(char *dst, char *args[])
536{
537    unsigned int i, len = 0;
538
539    for (i = 0; args[i]; i++) {
540        if (i) {
541            strcat(dst+len, " ");
542            len++;
543        }
544        strcpy(dst+len, args[i]);
545        len += strlen(args[i]);
546    }
547    /* In case it's empty. */
548    dst[len] = '\0';
549}
550
551/*L:185
552 * This is where we actually tell the kernel to initialize the Guest. We
553 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
554 * the base of Guest "physical" memory, the top physical page to allow and the
555 * entry point for the Guest.
556 */
557static void tell_kernel(unsigned long start)
558{
559    unsigned long args[] = { LHREQ_INITIALIZE,
560                 (unsigned long)guest_base,
561                 guest_limit / getpagesize(), start };
562    verbose("Guest: %p - %p (%#lx)\n",
563        guest_base, guest_base + guest_limit, guest_limit);
564    lguest_fd = open_or_die("/dev/lguest", O_RDWR);
565    if (write(lguest_fd, args, sizeof(args)) < 0)
566        err(1, "Writing to /dev/lguest");
567}
568/*:*/
569
570/*L:200
571 * Device Handling.
572 *
573 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
574 * We need to make sure it's not trying to reach into the Launcher itself, so
575 * we have a convenient routine which checks it and exits with an error message
576 * if something funny is going on:
577 */
578static void *_check_pointer(unsigned long addr, unsigned int size,
579                unsigned int line)
580{
581    /*
582     * Check if the requested address and size exceeds the allocated memory,
583     * or addr + size wraps around.
584     */
585    if ((addr + size) > guest_limit || (addr + size) < addr)
586        errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
587    /*
588     * We return a pointer for the caller's convenience, now we know it's
589     * safe to use.
590     */
591    return from_guest_phys(addr);
592}
593/* A macro which transparently hands the line number to the real function. */
594#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
595
596/*
597 * Each buffer in the virtqueues is actually a chain of descriptors. This
598 * function returns the next descriptor in the chain, or vq->vring.num if we're
599 * at the end.
600 */
601static unsigned next_desc(struct vring_desc *desc,
602              unsigned int i, unsigned int max)
603{
604    unsigned int next;
605
606    /* If this descriptor says it doesn't chain, we're done. */
607    if (!(desc[i].flags & VRING_DESC_F_NEXT))
608        return max;
609
610    /* Check they're not leading us off end of descriptors. */
611    next = desc[i].next;
612    /* Make sure compiler knows to grab that: we don't want it changing! */
613    wmb();
614
615    if (next >= max)
616        errx(1, "Desc next is %u", next);
617
618    return next;
619}
620
621/*
622 * This actually sends the interrupt for this virtqueue, if we've used a
623 * buffer.
624 */
625static void trigger_irq(struct virtqueue *vq)
626{
627    unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
628
629    /* Don't inform them if nothing used. */
630    if (!vq->pending_used)
631        return;
632    vq->pending_used = 0;
633
634    /* If they don't want an interrupt, don't send one... */
635    if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
636        return;
637    }
638
639    /* Send the Guest an interrupt tell them we used something up. */
640    if (write(lguest_fd, buf, sizeof(buf)) != 0)
641        err(1, "Triggering irq %i", vq->config.irq);
642}
643
644/*
645 * This looks in the virtqueue for the first available buffer, and converts
646 * it to an iovec for convenient access. Since descriptors consist of some
647 * number of output then some number of input descriptors, it's actually two
648 * iovecs, but we pack them into one and note how many of each there were.
649 *
650 * This function waits if necessary, and returns the descriptor number found.
651 */
652static unsigned wait_for_vq_desc(struct virtqueue *vq,
653                 struct iovec iov[],
654                 unsigned int *out_num, unsigned int *in_num)
655{
656    unsigned int i, head, max;
657    struct vring_desc *desc;
658    u16 last_avail = lg_last_avail(vq);
659
660    /* There's nothing available? */
661    while (last_avail == vq->vring.avail->idx) {
662        u64 event;
663
664        /*
665         * Since we're about to sleep, now is a good time to tell the
666         * Guest about what we've used up to now.
667         */
668        trigger_irq(vq);
669
670        /* OK, now we need to know about added descriptors. */
671        vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
672
673        /*
674         * They could have slipped one in as we were doing that: make
675         * sure it's written, then check again.
676         */
677        mb();
678        if (last_avail != vq->vring.avail->idx) {
679            vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
680            break;
681        }
682
683        /* Nothing new? Wait for eventfd to tell us they refilled. */
684        if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
685            errx(1, "Event read failed?");
686
687        /* We don't need to be notified again. */
688        vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
689    }
690
691    /* Check it isn't doing very strange things with descriptor numbers. */
692    if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
693        errx(1, "Guest moved used index from %u to %u",
694             last_avail, vq->vring.avail->idx);
695
696    /*
697     * Grab the next descriptor number they're advertising, and increment
698     * the index we've seen.
699     */
700    head = vq->vring.avail->ring[last_avail % vq->vring.num];
701    lg_last_avail(vq)++;
702
703    /* If their number is silly, that's a fatal mistake. */
704    if (head >= vq->vring.num)
705        errx(1, "Guest says index %u is available", head);
706
707    /* When we start there are none of either input nor output. */
708    *out_num = *in_num = 0;
709
710    max = vq->vring.num;
711    desc = vq->vring.desc;
712    i = head;
713
714    /*
715     * If this is an indirect entry, then this buffer contains a descriptor
716     * table which we handle as if it's any normal descriptor chain.
717     */
718    if (desc[i].flags & VRING_DESC_F_INDIRECT) {
719        if (desc[i].len % sizeof(struct vring_desc))
720            errx(1, "Invalid size for indirect buffer table");
721
722        max = desc[i].len / sizeof(struct vring_desc);
723        desc = check_pointer(desc[i].addr, desc[i].len);
724        i = 0;
725    }
726
727    do {
728        /* Grab the first descriptor, and check it's OK. */
729        iov[*out_num + *in_num].iov_len = desc[i].len;
730        iov[*out_num + *in_num].iov_base
731            = check_pointer(desc[i].addr, desc[i].len);
732        /* If this is an input descriptor, increment that count. */
733        if (desc[i].flags & VRING_DESC_F_WRITE)
734            (*in_num)++;
735        else {
736            /*
737             * If it's an output descriptor, they're all supposed
738             * to come before any input descriptors.
739             */
740            if (*in_num)
741                errx(1, "Descriptor has out after in");
742            (*out_num)++;
743        }
744
745        /* If we've got too many, that implies a descriptor loop. */
746        if (*out_num + *in_num > max)
747            errx(1, "Looped descriptor");
748    } while ((i = next_desc(desc, i, max)) != max);
749
750    return head;
751}
752
753/*
754 * After we've used one of their buffers, we tell the Guest about it. Sometime
755 * later we'll want to send them an interrupt using trigger_irq(); note that
756 * wait_for_vq_desc() does that for us if it has to wait.
757 */
758static void add_used(struct virtqueue *vq, unsigned int head, int len)
759{
760    struct vring_used_elem *used;
761
762    /*
763     * The virtqueue contains a ring of used buffers. Get a pointer to the
764     * next entry in that used ring.
765     */
766    used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
767    used->id = head;
768    used->len = len;
769    /* Make sure buffer is written before we update index. */
770    wmb();
771    vq->vring.used->idx++;
772    vq->pending_used++;
773}
774
775/* And here's the combo meal deal. Supersize me! */
776static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
777{
778    add_used(vq, head, len);
779    trigger_irq(vq);
780}
781
782/*
783 * The Console
784 *
785 * We associate some data with the console for our exit hack.
786 */
787struct console_abort {
788    /* How many times have they hit ^C? */
789    int count;
790    /* When did they start? */
791    struct timeval start;
792};
793
794/* This is the routine which handles console input (ie. stdin). */
795static void console_input(struct virtqueue *vq)
796{
797    int len;
798    unsigned int head, in_num, out_num;
799    struct console_abort *abort = vq->dev->priv;
800    struct iovec iov[vq->vring.num];
801
802    /* Make sure there's a descriptor available. */
803    head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
804    if (out_num)
805        errx(1, "Output buffers in console in queue?");
806
807    /* Read into it. This is where we usually wait. */
808    len = readv(STDIN_FILENO, iov, in_num);
809    if (len <= 0) {
810        /* Ran out of input? */
811        warnx("Failed to get console input, ignoring console.");
812        /*
813         * For simplicity, dying threads kill the whole Launcher. So
814         * just nap here.
815         */
816        for (;;)
817            pause();
818    }
819
820    /* Tell the Guest we used a buffer. */
821    add_used_and_trigger(vq, head, len);
822
823    /*
824     * Three ^C within one second? Exit.
825     *
826     * This is such a hack, but works surprisingly well. Each ^C has to
827     * be in a buffer by itself, so they can't be too fast. But we check
828     * that we get three within about a second, so they can't be too
829     * slow.
830     */
831    if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
832        abort->count = 0;
833        return;
834    }
835
836    abort->count++;
837    if (abort->count == 1)
838        gettimeofday(&abort->start, NULL);
839    else if (abort->count == 3) {
840        struct timeval now;
841        gettimeofday(&now, NULL);
842        /* Kill all Launcher processes with SIGINT, like normal ^C */
843        if (now.tv_sec <= abort->start.tv_sec+1)
844            kill(0, SIGINT);
845        abort->count = 0;
846    }
847}
848
849/* This is the routine which handles console output (ie. stdout). */
850static void console_output(struct virtqueue *vq)
851{
852    unsigned int head, out, in;
853    struct iovec iov[vq->vring.num];
854
855    /* We usually wait in here, for the Guest to give us something. */
856    head = wait_for_vq_desc(vq, iov, &out, &in);
857    if (in)
858        errx(1, "Input buffers in console output queue?");
859
860    /* writev can return a partial write, so we loop here. */
861    while (!iov_empty(iov, out)) {
862        int len = writev(STDOUT_FILENO, iov, out);
863        if (len <= 0) {
864            warn("Write to stdout gave %i (%d)", len, errno);
865            break;
866        }
867        iov_consume(iov, out, len);
868    }
869
870    /*
871     * We're finished with that buffer: if we're going to sleep,
872     * wait_for_vq_desc() will prod the Guest with an interrupt.
873     */
874    add_used(vq, head, 0);
875}
876
877/*
878 * The Network
879 *
880 * Handling output for network is also simple: we get all the output buffers
881 * and write them to /dev/net/tun.
882 */
883struct net_info {
884    int tunfd;
885};
886
887static void net_output(struct virtqueue *vq)
888{
889    struct net_info *net_info = vq->dev->priv;
890    unsigned int head, out, in;
891    struct iovec iov[vq->vring.num];
892
893    /* We usually wait in here for the Guest to give us a packet. */
894    head = wait_for_vq_desc(vq, iov, &out, &in);
895    if (in)
896        errx(1, "Input buffers in net output queue?");
897    /*
898     * Send the whole thing through to /dev/net/tun. It expects the exact
899     * same format: what a coincidence!
900     */
901    if (writev(net_info->tunfd, iov, out) < 0)
902        warnx("Write to tun failed (%d)?", errno);
903
904    /*
905     * Done with that one; wait_for_vq_desc() will send the interrupt if
906     * all packets are processed.
907     */
908    add_used(vq, head, 0);
909}
910
911/*
912 * Handling network input is a bit trickier, because I've tried to optimize it.
913 *
914 * First we have a helper routine which tells is if from this file descriptor
915 * (ie. the /dev/net/tun device) will block:
916 */
917static bool will_block(int fd)
918{
919    fd_set fdset;
920    struct timeval zero = { 0, 0 };
921    FD_ZERO(&fdset);
922    FD_SET(fd, &fdset);
923    return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
924}
925
926/*
927 * This handles packets coming in from the tun device to our Guest. Like all
928 * service routines, it gets called again as soon as it returns, so you don't
929 * see a while(1) loop here.
930 */
931static void net_input(struct virtqueue *vq)
932{
933    int len;
934    unsigned int head, out, in;
935    struct iovec iov[vq->vring.num];
936    struct net_info *net_info = vq->dev->priv;
937
938    /*
939     * Get a descriptor to write an incoming packet into. This will also
940     * send an interrupt if they're out of descriptors.
941     */
942    head = wait_for_vq_desc(vq, iov, &out, &in);
943    if (out)
944        errx(1, "Output buffers in net input queue?");
945
946    /*
947     * If it looks like we'll block reading from the tun device, send them
948     * an interrupt.
949     */
950    if (vq->pending_used && will_block(net_info->tunfd))
951        trigger_irq(vq);
952
953    /*
954     * Read in the packet. This is where we normally wait (when there's no
955     * incoming network traffic).
956     */
957    len = readv(net_info->tunfd, iov, in);
958    if (len <= 0)
959        warn("Failed to read from tun (%d).", errno);
960
961    /*
962     * Mark that packet buffer as used, but don't interrupt here. We want
963     * to wait until we've done as much work as we can.
964     */
965    add_used(vq, head, len);
966}
967/*:*/
968
969/* This is the helper to create threads: run the service routine in a loop. */
970static int do_thread(void *_vq)
971{
972    struct virtqueue *vq = _vq;
973
974    for (;;)
975        vq->service(vq);
976    return 0;
977}
978
979/*
980 * When a child dies, we kill our entire process group with SIGTERM. This
981 * also has the side effect that the shell restores the console for us!
982 */
983static void kill_launcher(int signal)
984{
985    kill(0, SIGTERM);
986}
987
988static void reset_device(struct device *dev)
989{
990    struct virtqueue *vq;
991
992    verbose("Resetting device %s\n", dev->name);
993
994    /* Clear any features they've acked. */
995    memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
996
997    /* We're going to be explicitly killing threads, so ignore them. */
998    signal(SIGCHLD, SIG_IGN);
999
1000    /* Zero out the virtqueues, get rid of their threads */
1001    for (vq = dev->vq; vq; vq = vq->next) {
1002        if (vq->thread != (pid_t)-1) {
1003            kill(vq->thread, SIGTERM);
1004            waitpid(vq->thread, NULL, 0);
1005            vq->thread = (pid_t)-1;
1006        }
1007        memset(vq->vring.desc, 0,
1008               vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1009        lg_last_avail(vq) = 0;
1010    }
1011    dev->running = false;
1012
1013    /* Now we care if threads die. */
1014    signal(SIGCHLD, (void *)kill_launcher);
1015}
1016
1017/*L:216
1018 * This actually creates the thread which services the virtqueue for a device.
1019 */
1020static void create_thread(struct virtqueue *vq)
1021{
1022    /*
1023     * Create stack for thread. Since the stack grows upwards, we point
1024     * the stack pointer to the end of this region.
1025     */
1026    char *stack = malloc(32768);
1027    unsigned long args[] = { LHREQ_EVENTFD,
1028                 vq->config.pfn*getpagesize(), 0 };
1029
1030    /* Create a zero-initialized eventfd. */
1031    vq->eventfd = eventfd(0, 0);
1032    if (vq->eventfd < 0)
1033        err(1, "Creating eventfd");
1034    args[2] = vq->eventfd;
1035
1036    /*
1037     * Attach an eventfd to this virtqueue: it will go off when the Guest
1038     * does an LHCALL_NOTIFY for this vq.
1039     */
1040    if (write(lguest_fd, &args, sizeof(args)) != 0)
1041        err(1, "Attaching eventfd");
1042
1043    /*
1044     * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1045     * we get a signal if it dies.
1046     */
1047    vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1048    if (vq->thread == (pid_t)-1)
1049        err(1, "Creating clone");
1050
1051    /* We close our local copy now the child has it. */
1052    close(vq->eventfd);
1053}
1054
1055static void start_device(struct device *dev)
1056{
1057    unsigned int i;
1058    struct virtqueue *vq;
1059
1060    verbose("Device %s OK: offered", dev->name);
1061    for (i = 0; i < dev->feature_len; i++)
1062        verbose(" %02x", get_feature_bits(dev)[i]);
1063    verbose(", accepted");
1064    for (i = 0; i < dev->feature_len; i++)
1065        verbose(" %02x", get_feature_bits(dev)
1066            [dev->feature_len+i]);
1067
1068    for (vq = dev->vq; vq; vq = vq->next) {
1069        if (vq->service)
1070            create_thread(vq);
1071    }
1072    dev->running = true;
1073}
1074
1075static void cleanup_devices(void)
1076{
1077    struct device *dev;
1078
1079    for (dev = devices.dev; dev; dev = dev->next)
1080        reset_device(dev);
1081
1082    /* If we saved off the original terminal settings, restore them now. */
1083    if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1084        tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1085}
1086
1087/* When the Guest tells us they updated the status field, we handle it. */
1088static void update_device_status(struct device *dev)
1089{
1090    /* A zero status is a reset, otherwise it's a set of flags. */
1091    if (dev->desc->status == 0)
1092        reset_device(dev);
1093    else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1094        warnx("Device %s configuration FAILED", dev->name);
1095        if (dev->running)
1096            reset_device(dev);
1097    } else {
1098        if (dev->running)
1099            err(1, "Device %s features finalized twice", dev->name);
1100        start_device(dev);
1101    }
1102}
1103
1104/*L:215
1105 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1106 * particular, it's used to notify us of device status changes during boot.
1107 */
1108static void handle_output(unsigned long addr)
1109{
1110    struct device *i;
1111
1112    /* Check each device. */
1113    for (i = devices.dev; i; i = i->next) {
1114        struct virtqueue *vq;
1115
1116        /*
1117         * Notifications to device descriptors mean they updated the
1118         * device status.
1119         */
1120        if (from_guest_phys(addr) == i->desc) {
1121            update_device_status(i);
1122            return;
1123        }
1124
1125        /* Devices should not be used before features are finalized. */
1126        for (vq = i->vq; vq; vq = vq->next) {
1127            if (addr != vq->config.pfn*getpagesize())
1128                continue;
1129            errx(1, "Notification on %s before setup!", i->name);
1130        }
1131    }
1132
1133    /*
1134     * Early console write is done using notify on a nul-terminated string
1135     * in Guest memory. It's also great for hacking debugging messages
1136     * into a Guest.
1137     */
1138    if (addr >= guest_limit)
1139        errx(1, "Bad NOTIFY %#lx", addr);
1140
1141    write(STDOUT_FILENO, from_guest_phys(addr),
1142          strnlen(from_guest_phys(addr), guest_limit - addr));
1143}
1144
1145/*L:190
1146 * Device Setup
1147 *
1148 * All devices need a descriptor so the Guest knows it exists, and a "struct
1149 * device" so the Launcher can keep track of it. We have common helper
1150 * routines to allocate and manage them.
1151 */
1152
1153/*
1154 * The layout of the device page is a "struct lguest_device_desc" followed by a
1155 * number of virtqueue descriptors, then two sets of feature bits, then an
1156 * array of configuration bytes. This routine returns the configuration
1157 * pointer.
1158 */
1159static u8 *device_config(const struct device *dev)
1160{
1161    return (void *)(dev->desc + 1)
1162        + dev->num_vq * sizeof(struct lguest_vqconfig)
1163        + dev->feature_len * 2;
1164}
1165
1166/*
1167 * This routine allocates a new "struct lguest_device_desc" from descriptor
1168 * table page just above the Guest's normal memory. It returns a pointer to
1169 * that descriptor.
1170 */
1171static struct lguest_device_desc *new_dev_desc(u16 type)
1172{
1173    struct lguest_device_desc d = { .type = type };
1174    void *p;
1175
1176    /* Figure out where the next device config is, based on the last one. */
1177    if (devices.lastdev)
1178        p = device_config(devices.lastdev)
1179            + devices.lastdev->desc->config_len;
1180    else
1181        p = devices.descpage;
1182
1183    /* We only have one page for all the descriptors. */
1184    if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1185        errx(1, "Too many devices");
1186
1187    /* p might not be aligned, so we memcpy in. */
1188    return memcpy(p, &d, sizeof(d));
1189}
1190
1191/*
1192 * Each device descriptor is followed by the description of its virtqueues. We
1193 * specify how many descriptors the virtqueue is to have.
1194 */
1195static void add_virtqueue(struct device *dev, unsigned int num_descs,
1196              void (*service)(struct virtqueue *))
1197{
1198    unsigned int pages;
1199    struct virtqueue **i, *vq = malloc(sizeof(*vq));
1200    void *p;
1201
1202    /* First we need some memory for this virtqueue. */
1203    pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1204        / getpagesize();
1205    p = get_pages(pages);
1206
1207    /* Initialize the virtqueue */
1208    vq->next = NULL;
1209    vq->last_avail_idx = 0;
1210    vq->dev = dev;
1211
1212    /*
1213     * This is the routine the service thread will run, and its Process ID
1214     * once it's running.
1215     */
1216    vq->service = service;
1217    vq->thread = (pid_t)-1;
1218
1219    /* Initialize the configuration. */
1220    vq->config.num = num_descs;
1221    vq->config.irq = devices.next_irq++;
1222    vq->config.pfn = to_guest_phys(p) / getpagesize();
1223
1224    /* Initialize the vring. */
1225    vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1226
1227    /*
1228     * Append virtqueue to this device's descriptor. We use
1229     * device_config() to get the end of the device's current virtqueues;
1230     * we check that we haven't added any config or feature information
1231     * yet, otherwise we'd be overwriting them.
1232     */
1233    assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1234    memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1235    dev->num_vq++;
1236    dev->desc->num_vq++;
1237
1238    verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1239
1240    /*
1241     * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1242     * second.
1243     */
1244    for (i = &dev->vq; *i; i = &(*i)->next);
1245    *i = vq;
1246}
1247
1248/*
1249 * The first half of the feature bitmask is for us to advertise features. The
1250 * second half is for the Guest to accept features.
1251 */
1252static void add_feature(struct device *dev, unsigned bit)
1253{
1254    u8 *features = get_feature_bits(dev);
1255
1256    /* We can't extend the feature bits once we've added config bytes */
1257    if (dev->desc->feature_len <= bit / CHAR_BIT) {
1258        assert(dev->desc->config_len == 0);
1259        dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1260    }
1261
1262    features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1263}
1264
1265/*
1266 * This routine sets the configuration fields for an existing device's
1267 * descriptor. It only works for the last device, but that's OK because that's
1268 * how we use it.
1269 */
1270static void set_config(struct device *dev, unsigned len, const void *conf)
1271{
1272    /* Check we haven't overflowed our single page. */
1273    if (device_config(dev) + len > devices.descpage + getpagesize())
1274        errx(1, "Too many devices");
1275
1276    /* Copy in the config information, and store the length. */
1277    memcpy(device_config(dev), conf, len);
1278    dev->desc->config_len = len;
1279
1280    /* Size must fit in config_len field (8 bits)! */
1281    assert(dev->desc->config_len == len);
1282}
1283
1284/*
1285 * This routine does all the creation and setup of a new device, including
1286 * calling new_dev_desc() to allocate the descriptor and device memory. We
1287 * don't actually start the service threads until later.
1288 *
1289 * See what I mean about userspace being boring?
1290 */
1291static struct device *new_device(const char *name, u16 type)
1292{
1293    struct device *dev = malloc(sizeof(*dev));
1294
1295    /* Now we populate the fields one at a time. */
1296    dev->desc = new_dev_desc(type);
1297    dev->name = name;
1298    dev->vq = NULL;
1299    dev->feature_len = 0;
1300    dev->num_vq = 0;
1301    dev->running = false;
1302
1303    /*
1304     * Append to device list. Prepending to a single-linked list is
1305     * easier, but the user expects the devices to be arranged on the bus
1306     * in command-line order. The first network device on the command line
1307     * is eth0, the first block device /dev/vda, etc.
1308     */
1309    if (devices.lastdev)
1310        devices.lastdev->next = dev;
1311    else
1312        devices.dev = dev;
1313    devices.lastdev = dev;
1314
1315    return dev;
1316}
1317
1318/*
1319 * Our first setup routine is the console. It's a fairly simple device, but
1320 * UNIX tty handling makes it uglier than it could be.
1321 */
1322static void setup_console(void)
1323{
1324    struct device *dev;
1325
1326    /* If we can save the initial standard input settings... */
1327    if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1328        struct termios term = orig_term;
1329        /*
1330         * Then we turn off echo, line buffering and ^C etc: We want a
1331         * raw input stream to the Guest.
1332         */
1333        term.c_lflag &= ~(ISIG|ICANON|ECHO);
1334        tcsetattr(STDIN_FILENO, TCSANOW, &term);
1335    }
1336
1337    dev = new_device("console", VIRTIO_ID_CONSOLE);
1338
1339    /* We store the console state in dev->priv, and initialize it. */
1340    dev->priv = malloc(sizeof(struct console_abort));
1341    ((struct console_abort *)dev->priv)->count = 0;
1342
1343    /*
1344     * The console needs two virtqueues: the input then the output. When
1345     * they put something the input queue, we make sure we're listening to
1346     * stdin. When they put something in the output queue, we write it to
1347     * stdout.
1348     */
1349    add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1350    add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1351
1352    verbose("device %u: console\n", ++devices.device_num);
1353}
1354/*:*/
1355
1356/*M:010
1357 * Inter-guest networking is an interesting area. Simplest is to have a
1358 * --sharenet=<name> option which opens or creates a named pipe. This can be
1359 * used to send packets to another guest in a 1:1 manner.
1360 *
1361 * More sophisticated is to use one of the tools developed for project like UML
1362 * to do networking.
1363 *
1364 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1365 * completely generic ("here's my vring, attach to your vring") and would work
1366 * for any traffic. Of course, namespace and permissions issues need to be
1367 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1368 * multiple inter-guest channels behind one interface, although it would
1369 * require some manner of hotplugging new virtio channels.
1370 *
1371 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
1372:*/
1373
1374static u32 str2ip(const char *ipaddr)
1375{
1376    unsigned int b[4];
1377
1378    if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1379        errx(1, "Failed to parse IP address '%s'", ipaddr);
1380    return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1381}
1382
1383static void str2mac(const char *macaddr, unsigned char mac[6])
1384{
1385    unsigned int m[6];
1386    if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1387           &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1388        errx(1, "Failed to parse mac address '%s'", macaddr);
1389    mac[0] = m[0];
1390    mac[1] = m[1];
1391    mac[2] = m[2];
1392    mac[3] = m[3];
1393    mac[4] = m[4];
1394    mac[5] = m[5];
1395}
1396
1397/*
1398 * This code is "adapted" from libbridge: it attaches the Host end of the
1399 * network device to the bridge device specified by the command line.
1400 *
1401 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1402 * dislike bridging), and I just try not to break it.
1403 */
1404static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1405{
1406    int ifidx;
1407    struct ifreq ifr;
1408
1409    if (!*br_name)
1410        errx(1, "must specify bridge name");
1411
1412    ifidx = if_nametoindex(if_name);
1413    if (!ifidx)
1414        errx(1, "interface %s does not exist!", if_name);
1415
1416    strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1417    ifr.ifr_name[IFNAMSIZ-1] = '\0';
1418    ifr.ifr_ifindex = ifidx;
1419    if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1420        err(1, "can't add %s to bridge %s", if_name, br_name);
1421}
1422
1423/*
1424 * This sets up the Host end of the network device with an IP address, brings
1425 * it up so packets will flow, the copies the MAC address into the hwaddr
1426 * pointer.
1427 */
1428static void configure_device(int fd, const char *tapif, u32 ipaddr)
1429{
1430    struct ifreq ifr;
1431    struct sockaddr_in sin;
1432
1433    memset(&ifr, 0, sizeof(ifr));
1434    strcpy(ifr.ifr_name, tapif);
1435
1436    /* Don't read these incantations. Just cut & paste them like I did! */
1437    sin.sin_family = AF_INET;
1438    sin.sin_addr.s_addr = htonl(ipaddr);
1439    memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1440    if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1441        err(1, "Setting %s interface address", tapif);
1442    ifr.ifr_flags = IFF_UP;
1443    if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1444        err(1, "Bringing interface %s up", tapif);
1445}
1446
1447static int get_tun_device(char tapif[IFNAMSIZ])
1448{
1449    struct ifreq ifr;
1450    int netfd;
1451
1452    /* Start with this zeroed. Messy but sure. */
1453    memset(&ifr, 0, sizeof(ifr));
1454
1455    /*
1456     * We open the /dev/net/tun device and tell it we want a tap device. A
1457     * tap device is like a tun device, only somehow different. To tell
1458     * the truth, I completely blundered my way through this code, but it
1459     * works now!
1460     */
1461    netfd = open_or_die("/dev/net/tun", O_RDWR);
1462    ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1463    strcpy(ifr.ifr_name, "tap%d");
1464    if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1465        err(1, "configuring /dev/net/tun");
1466
1467    if (ioctl(netfd, TUNSETOFFLOAD,
1468          TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1469        err(1, "Could not set features for tun device");
1470
1471    /*
1472     * We don't need checksums calculated for packets coming in this
1473     * device: trust us!
1474     */
1475    ioctl(netfd, TUNSETNOCSUM, 1);
1476
1477    memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1478    return netfd;
1479}
1480
1481/*L:195
1482 * Our network is a Host<->Guest network. This can either use bridging or
1483 * routing, but the principle is the same: it uses the "tun" device to inject
1484 * packets into the Host as if they came in from a normal network card. We
1485 * just shunt packets between the Guest and the tun device.
1486 */
1487static void setup_tun_net(char *arg)
1488{
1489    struct device *dev;
1490    struct net_info *net_info = malloc(sizeof(*net_info));
1491    int ipfd;
1492    u32 ip = INADDR_ANY;
1493    bool bridging = false;
1494    char tapif[IFNAMSIZ], *p;
1495    struct virtio_net_config conf;
1496
1497    net_info->tunfd = get_tun_device(tapif);
1498
1499    /* First we create a new network device. */
1500    dev = new_device("net", VIRTIO_ID_NET);
1501    dev->priv = net_info;
1502
1503    /* Network devices need a recv and a send queue, just like console. */
1504    add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1505    add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1506
1507    /*
1508     * We need a socket to perform the magic network ioctls to bring up the
1509     * tap interface, connect to the bridge etc. Any socket will do!
1510     */
1511    ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1512    if (ipfd < 0)
1513        err(1, "opening IP socket");
1514
1515    /* If the command line was --tunnet=bridge:<name> do bridging. */
1516    if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1517        arg += strlen(BRIDGE_PFX);
1518        bridging = true;
1519    }
1520
1521    /* A mac address may follow the bridge name or IP address */
1522    p = strchr(arg, ':');
1523    if (p) {
1524        str2mac(p+1, conf.mac);
1525        add_feature(dev, VIRTIO_NET_F_MAC);
1526        *p = '\0';
1527    }
1528
1529    /* arg is now either an IP address or a bridge name */
1530    if (bridging)
1531        add_to_bridge(ipfd, tapif, arg);
1532    else
1533        ip = str2ip(arg);
1534
1535    /* Set up the tun device. */
1536    configure_device(ipfd, tapif, ip);
1537
1538    /* Expect Guest to handle everything except UFO */
1539    add_feature(dev, VIRTIO_NET_F_CSUM);
1540    add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1541    add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1542    add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1543    add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1544    add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1545    add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1546    add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1547    /* We handle indirect ring entries */
1548    add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1549    set_config(dev, sizeof(conf), &conf);
1550
1551    /* We don't need the socket any more; setup is done. */
1552    close(ipfd);
1553
1554    devices.device_num++;
1555
1556    if (bridging)
1557        verbose("device %u: tun %s attached to bridge: %s\n",
1558            devices.device_num, tapif, arg);
1559    else
1560        verbose("device %u: tun %s: %s\n",
1561            devices.device_num, tapif, arg);
1562}
1563/*:*/
1564
1565/* This hangs off device->priv. */
1566struct vblk_info {
1567    /* The size of the file. */
1568    off64_t len;
1569
1570    /* The file descriptor for the file. */
1571    int fd;
1572
1573};
1574
1575/*L:210
1576 * The Disk
1577 *
1578 * The disk only has one virtqueue, so it only has one thread. It is really
1579 * simple: the Guest asks for a block number and we read or write that position
1580 * in the file.
1581 *
1582 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1583 * slow: the Guest waits until the read is finished before running anything
1584 * else, even if it could have been doing useful work.
1585 *
1586 * We could have used async I/O, except it's reputed to suck so hard that
1587 * characters actually go missing from your code when you try to use it.
1588 */
1589static void blk_request(struct virtqueue *vq)
1590{
1591    struct vblk_info *vblk = vq->dev->priv;
1592    unsigned int head, out_num, in_num, wlen;
1593    int ret;
1594    u8 *in;
1595    struct virtio_blk_outhdr *out;
1596    struct iovec iov[vq->vring.num];
1597    off64_t off;
1598
1599    /*
1600     * Get the next request, where we normally wait. It triggers the
1601     * interrupt to acknowledge previously serviced requests (if any).
1602     */
1603    head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1604
1605    /*
1606     * Every block request should contain at least one output buffer
1607     * (detailing the location on disk and the type of request) and one
1608     * input buffer (to hold the result).
1609     */
1610    if (out_num == 0 || in_num == 0)
1611        errx(1, "Bad virtblk cmd %u out=%u in=%u",
1612             head, out_num, in_num);
1613
1614    out = convert(&iov[0], struct virtio_blk_outhdr);
1615    in = convert(&iov[out_num+in_num-1], u8);
1616    /*
1617     * For historical reasons, block operations are expressed in 512 byte
1618     * "sectors".
1619     */
1620    off = out->sector * 512;
1621
1622    /*
1623     * In general the virtio block driver is allowed to try SCSI commands.
1624     * It'd be nice if we supported eject, for example, but we don't.
1625     */
1626    if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1627        fprintf(stderr, "Scsi commands unsupported\n");
1628        *in = VIRTIO_BLK_S_UNSUPP;
1629        wlen = sizeof(*in);
1630    } else if (out->type & VIRTIO_BLK_T_OUT) {
1631        /*
1632         * Write
1633         *
1634         * Move to the right location in the block file. This can fail
1635         * if they try to write past end.
1636         */
1637        if (lseek64(vblk->fd, off, SEEK_SET) != off)
1638            err(1, "Bad seek to sector %llu", out->sector);
1639
1640        ret = writev(vblk->fd, iov+1, out_num-1);
1641        verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1642
1643        /*
1644         * Grr... Now we know how long the descriptor they sent was, we
1645         * make sure they didn't try to write over the end of the block
1646         * file (possibly extending it).
1647         */
1648        if (ret > 0 && off + ret > vblk->len) {
1649            /* Trim it back to the correct length */
1650            ftruncate64(vblk->fd, vblk->len);
1651            /* Die, bad Guest, die. */
1652            errx(1, "Write past end %llu+%u", off, ret);
1653        }
1654
1655        wlen = sizeof(*in);
1656        *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1657    } else if (out->type & VIRTIO_BLK_T_FLUSH) {
1658        /* Flush */
1659        ret = fdatasync(vblk->fd);
1660        verbose("FLUSH fdatasync: %i\n", ret);
1661        wlen = sizeof(*in);
1662        *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1663    } else {
1664        /*
1665         * Read
1666         *
1667         * Move to the right location in the block file. This can fail
1668         * if they try to read past end.
1669         */
1670        if (lseek64(vblk->fd, off, SEEK_SET) != off)
1671            err(1, "Bad seek to sector %llu", out->sector);
1672
1673        ret = readv(vblk->fd, iov+1, in_num-1);
1674        verbose("READ from sector %llu: %i\n", out->sector, ret);
1675        if (ret >= 0) {
1676            wlen = sizeof(*in) + ret;
1677            *in = VIRTIO_BLK_S_OK;
1678        } else {
1679            wlen = sizeof(*in);
1680            *in = VIRTIO_BLK_S_IOERR;
1681        }
1682    }
1683
1684    /* Finished that request. */
1685    add_used(vq, head, wlen);
1686}
1687
1688/*L:198 This actually sets up a virtual block device. */
1689static void setup_block_file(const char *filename)
1690{
1691    struct device *dev;
1692    struct vblk_info *vblk;
1693    struct virtio_blk_config conf;
1694
1695    /* Creat the device. */
1696    dev = new_device("block", VIRTIO_ID_BLOCK);
1697
1698    /* The device has one virtqueue, where the Guest places requests. */
1699    add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1700
1701    /* Allocate the room for our own bookkeeping */
1702    vblk = dev->priv = malloc(sizeof(*vblk));
1703
1704    /* First we open the file and store the length. */
1705    vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1706    vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1707
1708    /* We support FLUSH. */
1709    add_feature(dev, VIRTIO_BLK_F_FLUSH);
1710
1711    /* Tell Guest how many sectors this device has. */
1712    conf.capacity = cpu_to_le64(vblk->len / 512);
1713
1714    /*
1715     * Tell Guest not to put in too many descriptors at once: two are used
1716     * for the in and out elements.
1717     */
1718    add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1719    conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1720
1721    /* Don't try to put whole struct: we have 8 bit limit. */
1722    set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1723
1724    verbose("device %u: virtblock %llu sectors\n",
1725        ++devices.device_num, le64_to_cpu(conf.capacity));
1726}
1727
1728/*L:211
1729 * Our random number generator device reads from /dev/random into the Guest's
1730 * input buffers. The usual case is that the Guest doesn't want random numbers
1731 * and so has no buffers although /dev/random is still readable, whereas
1732 * console is the reverse.
1733 *
1734 * The same logic applies, however.
1735 */
1736struct rng_info {
1737    int rfd;
1738};
1739
1740static void rng_input(struct virtqueue *vq)
1741{
1742    int len;
1743    unsigned int head, in_num, out_num, totlen = 0;
1744    struct rng_info *rng_info = vq->dev->priv;
1745    struct iovec iov[vq->vring.num];
1746
1747    /* First we need a buffer from the Guests's virtqueue. */
1748    head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1749    if (out_num)
1750        errx(1, "Output buffers in rng?");
1751
1752    /*
1753     * Just like the console write, we loop to cover the whole iovec.
1754     * In this case, short reads actually happen quite a bit.
1755     */
1756    while (!iov_empty(iov, in_num)) {
1757        len = readv(rng_info->rfd, iov, in_num);
1758        if (len <= 0)
1759            err(1, "Read from /dev/random gave %i", len);
1760        iov_consume(iov, in_num, len);
1761        totlen += len;
1762    }
1763
1764    /* Tell the Guest about the new input. */
1765    add_used(vq, head, totlen);
1766}
1767
1768/*L:199
1769 * This creates a "hardware" random number device for the Guest.
1770 */
1771static void setup_rng(void)
1772{
1773    struct device *dev;
1774    struct rng_info *rng_info = malloc(sizeof(*rng_info));
1775
1776    /* Our device's privat info simply contains the /dev/random fd. */
1777    rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1778
1779    /* Create the new device. */
1780    dev = new_device("rng", VIRTIO_ID_RNG);
1781    dev->priv = rng_info;
1782
1783    /* The device has one virtqueue, where the Guest places inbufs. */
1784    add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1785
1786    verbose("device %u: rng\n", devices.device_num++);
1787}
1788/* That's the end of device setup. */
1789
1790/*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1791static void __attribute__((noreturn)) restart_guest(void)
1792{
1793    unsigned int i;
1794
1795    /*
1796     * Since we don't track all open fds, we simply close everything beyond
1797     * stderr.
1798     */
1799    for (i = 3; i < FD_SETSIZE; i++)
1800        close(i);
1801
1802    /* Reset all the devices (kills all threads). */
1803    cleanup_devices();
1804
1805    execv(main_args[0], main_args);
1806    err(1, "Could not exec %s", main_args[0]);
1807}
1808
1809/*L:220
1810 * Finally we reach the core of the Launcher which runs the Guest, serves
1811 * its input and output, and finally, lays it to rest.
1812 */
1813static void __attribute__((noreturn)) run_guest(void)
1814{
1815    for (;;) {
1816        unsigned long notify_addr;
1817        int readval;
1818
1819        /* We read from the /dev/lguest device to run the Guest. */
1820        readval = pread(lguest_fd, &notify_addr,
1821                sizeof(notify_addr), cpu_id);
1822
1823        /* One unsigned long means the Guest did HCALL_NOTIFY */
1824        if (readval == sizeof(notify_addr)) {
1825            verbose("Notify on address %#lx\n", notify_addr);
1826            handle_output(notify_addr);
1827        /* ENOENT means the Guest died. Reading tells us why. */
1828        } else if (errno == ENOENT) {
1829            char reason[1024] = { 0 };
1830            pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1831            errx(1, "%s", reason);
1832        /* ERESTART means that we need to reboot the guest */
1833        } else if (errno == ERESTART) {
1834            restart_guest();
1835        /* Anything else means a bug or incompatible change. */
1836        } else
1837            err(1, "Running guest failed");
1838    }
1839}
1840/*L:240
1841 * This is the end of the Launcher. The good news: we are over halfway
1842 * through! The bad news: the most fiendish part of the code still lies ahead
1843 * of us.
1844 *
1845 * Are you ready? Take a deep breath and join me in the core of the Host, in
1846 * "make Host".
1847:*/
1848
1849static struct option opts[] = {
1850    { "verbose", 0, NULL, 'v' },
1851    { "tunnet", 1, NULL, 't' },
1852    { "block", 1, NULL, 'b' },
1853    { "rng", 0, NULL, 'r' },
1854    { "initrd", 1, NULL, 'i' },
1855    { "username", 1, NULL, 'u' },
1856    { "chroot", 1, NULL, 'c' },
1857    { NULL },
1858};
1859static void usage(void)
1860{
1861    errx(1, "Usage: lguest [--verbose] "
1862         "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1863         "|--block=<filename>|--initrd=<filename>]...\n"
1864         "<mem-in-mb> vmlinux [args...]");
1865}
1866
1867/*L:105 The main routine is where the real work begins: */
1868int main(int argc, char *argv[])
1869{
1870    /* Memory, code startpoint and size of the (optional) initrd. */
1871    unsigned long mem = 0, start, initrd_size = 0;
1872    /* Two temporaries. */
1873    int i, c;
1874    /* The boot information for the Guest. */
1875    struct boot_params *boot;
1876    /* If they specify an initrd file to load. */
1877    const char *initrd_name = NULL;
1878
1879    /* Password structure for initgroups/setres[gu]id */
1880    struct passwd *user_details = NULL;
1881
1882    /* Directory to chroot to */
1883    char *chroot_path = NULL;
1884
1885    /* Save the args: we "reboot" by execing ourselves again. */
1886    main_args = argv;
1887
1888    /*
1889     * First we initialize the device list. We keep a pointer to the last
1890     * device, and the next interrupt number to use for devices (1:
1891     * remember that 0 is used by the timer).
1892     */
1893    devices.lastdev = NULL;
1894    devices.next_irq = 1;
1895
1896    /* We're CPU 0. In fact, that's the only CPU possible right now. */
1897    cpu_id = 0;
1898
1899    /*
1900     * We need to know how much memory so we can set up the device
1901     * descriptor and memory pages for the devices as we parse the command
1902     * line. So we quickly look through the arguments to find the amount
1903     * of memory now.
1904     */
1905    for (i = 1; i < argc; i++) {
1906        if (argv[i][0] != '-') {
1907            mem = atoi(argv[i]) * 1024 * 1024;
1908            /*
1909             * We start by mapping anonymous pages over all of
1910             * guest-physical memory range. This fills it with 0,
1911             * and ensures that the Guest won't be killed when it
1912             * tries to access it.
1913             */
1914            guest_base = map_zeroed_pages(mem / getpagesize()
1915                              + DEVICE_PAGES);
1916            guest_limit = mem;
1917            guest_max = mem + DEVICE_PAGES*getpagesize();
1918            devices.descpage = get_pages(1);
1919            break;
1920        }
1921    }
1922
1923    /* The options are fairly straight-forward */
1924    while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1925        switch (c) {
1926        case 'v':
1927            verbose = true;
1928            break;
1929        case 't':
1930            setup_tun_net(optarg);
1931            break;
1932        case 'b':
1933            setup_block_file(optarg);
1934            break;
1935        case 'r':
1936            setup_rng();
1937            break;
1938        case 'i':
1939            initrd_name = optarg;
1940            break;
1941        case 'u':
1942            user_details = getpwnam(optarg);
1943            if (!user_details)
1944                err(1, "getpwnam failed, incorrect username?");
1945            break;
1946        case 'c':
1947            chroot_path = optarg;
1948            break;
1949        default:
1950            warnx("Unknown argument %s", argv[optind]);
1951            usage();
1952        }
1953    }
1954    /*
1955     * After the other arguments we expect memory and kernel image name,
1956     * followed by command line arguments for the kernel.
1957     */
1958    if (optind + 2 > argc)
1959        usage();
1960
1961    verbose("Guest base is at %p\n", guest_base);
1962
1963    /* We always have a console device */
1964    setup_console();
1965
1966    /* Now we load the kernel */
1967    start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1968
1969    /* Boot information is stashed at physical address 0 */
1970    boot = from_guest_phys(0);
1971
1972    /* Map the initrd image if requested (at top of physical memory) */
1973    if (initrd_name) {
1974        initrd_size = load_initrd(initrd_name, mem);
1975        /*
1976         * These are the location in the Linux boot header where the
1977         * start and size of the initrd are expected to be found.
1978         */
1979        boot->hdr.ramdisk_image = mem - initrd_size;
1980        boot->hdr.ramdisk_size = initrd_size;
1981        /* The bootloader type 0xFF means "unknown"; that's OK. */
1982        boot->hdr.type_of_loader = 0xFF;
1983    }
1984
1985    /*
1986     * The Linux boot header contains an "E820" memory map: ours is a
1987     * simple, single region.
1988     */
1989    boot->e820_entries = 1;
1990    boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1991    /*
1992     * The boot header contains a command line pointer: we put the command
1993     * line after the boot header.
1994     */
1995    boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1996    /* We use a simple helper to copy the arguments separated by spaces. */
1997    concat((char *)(boot + 1), argv+optind+2);
1998
1999    /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
2000    boot->hdr.kernel_alignment = 0x1000000;
2001
2002    /* Boot protocol version: 2.07 supports the fields for lguest. */
2003    boot->hdr.version = 0x207;
2004
2005    /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2006    boot->hdr.hardware_subarch = 1;
2007
2008    /* Tell the entry path not to try to reload segment registers. */
2009    boot->hdr.loadflags |= KEEP_SEGMENTS;
2010
2011    /* We tell the kernel to initialize the Guest. */
2012    tell_kernel(start);
2013
2014    /* Ensure that we terminate if a device-servicing child dies. */
2015    signal(SIGCHLD, kill_launcher);
2016
2017    /* If we exit via err(), this kills all the threads, restores tty. */
2018    atexit(cleanup_devices);
2019
2020    /* If requested, chroot to a directory */
2021    if (chroot_path) {
2022        if (chroot(chroot_path) != 0)
2023            err(1, "chroot(\"%s\") failed", chroot_path);
2024
2025        if (chdir("/") != 0)
2026            err(1, "chdir(\"/\") failed");
2027
2028        verbose("chroot done\n");
2029    }
2030
2031    /* If requested, drop privileges */
2032    if (user_details) {
2033        uid_t u;
2034        gid_t g;
2035
2036        u = user_details->pw_uid;
2037        g = user_details->pw_gid;
2038
2039        if (initgroups(user_details->pw_name, g) != 0)
2040            err(1, "initgroups failed");
2041
2042        if (setresgid(g, g, g) != 0)
2043            err(1, "setresgid failed");
2044
2045        if (setresuid(u, u, u) != 0)
2046            err(1, "setresuid failed");
2047
2048        verbose("Dropping privileges completed\n");
2049    }
2050
2051    /* Finally, run the Guest. This doesn't return. */
2052    run_guest();
2053}
2054/*:*/
2055
2056/*M:999
2057 * Mastery is done: you now know everything I do.
2058 *
2059 * But surely you have seen code, features and bugs in your wanderings which
2060 * you now yearn to attack? That is the real game, and I look forward to you
2061 * patching and forking lguest into the Your-Name-Here-visor.
2062 *
2063 * Farewell, and good coding!
2064 * Rusty Russell.
2065 */
2066

Archive Download this file



interactive