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
7 #define _LARGEFILE64_SOURCE
17 #include <sys/param.h>
18 #include <sys/types.h>
21 #include <sys/eventfd.h>
26 #include <sys/socket.h>
27 #include <sys/ioctl.h>
30 #include <netinet/in.h>
32 #include <linux/sockios.h>
33 #include <linux/if_tun.h>
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"
54 * We can ignore the 42 include files we need for this program, but I do want
55 * to draw attention to the use of kernel-style types.
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.
62 typedef unsigned long long u64;
68 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
69 #define BRIDGE_PFX "bridge:"
71 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
73 /* We can have up to 256 pages for devices. */
74 #define DEVICE_PAGES 256
75 /* This will occupy 3 pages: it must be a power of 2. */
76 #define VIRTQUEUE_NUM 256
79 * verbose is both a global flag and a macro. The C preprocessor allows
80 * this, and although I wouldn't recommend it, it works quite nicely here.
83 #define verbose(args...) \
84 do { if (verbose) printf(args); } while(0)
87 /* The pointer to the start of guest memory. */
88 static void *guest_base;
89 /* The maximum guest physical address allowed, and maximum possible. */
90 static unsigned long guest_limit, guest_max;
91 /* The /dev/lguest file descriptor. */
94 /* a per-cpu variable indicating whose vcpu is currently running */
95 static unsigned int __thread cpu_id;
97 /* This is our list of devices. */
99 /* Counter to assign interrupt numbers. */
100 unsigned int next_irq;
102 /* Counter to print out convenient device numbers. */
103 unsigned int device_num;
105 /* The descriptor page for the devices. */
108 /* A single linked list of devices. */
110 /* And a pointer to the last device for easy append. */
111 struct device *lastdev;
114 /* The list of Guest devices, based on command line arguments. */
115 static struct device_list devices;
117 /* The device structure describes a single device. */
119 /* The linked-list pointer. */
122 /* The device's descriptor, as mapped into the Guest. */
123 struct lguest_device_desc *desc;
125 /* We can't trust desc values once Guest has booted: we use these. */
126 unsigned int feature_len;
129 /* The name of this device, for --verbose. */
132 /* Any queues attached to this device */
133 struct virtqueue *vq;
135 /* Is it operational */
138 /* Device-specific data. */
142 /* The virtqueue structure describes a queue attached to a device. */
144 struct virtqueue *next;
146 /* Which device owns me. */
149 /* The configuration for this queue. */
150 struct lguest_vqconfig config;
152 /* The actual ring of buffers. */
155 /* Last available index we saw. */
158 /* How many are used since we sent last irq? */
159 unsigned int pending_used;
161 /* Eventfd where Guest notifications arrive. */
164 /* Function for the thread which is servicing this virtqueue. */
165 void (*service)(struct virtqueue *vq);
169 /* Remember the arguments to the program so we can "reboot" */
170 static char **main_args;
172 /* The original tty settings to restore on exit. */
173 static struct termios orig_term;
176 * We have to be careful with barriers: our devices are all run in separate
177 * threads and so we need to make sure that changes visible to the Guest happen
180 #define wmb() __asm__ __volatile__("" : : : "memory")
181 #define mb() __asm__ __volatile__("" : : : "memory")
184 * Convert an iovec element to the given type.
186 * This is a fairly ugly trick: we need to know the size of the type and
187 * alignment requirement to check the pointer is kosher. It's also nice to
188 * have the name of the type in case we report failure.
190 * Typing those three things all the time is cumbersome and error prone, so we
191 * have a macro which sets them all up and passes to the real function.
193 #define convert(iov, type) \
194 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
196 static void *_convert(struct iovec *iov, size_t size, size_t align,
199 if (iov->iov_len != size)
200 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
201 if ((unsigned long)iov->iov_base % align != 0)
202 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
203 return iov->iov_base;
206 /* Wrapper for the last available index. Makes it easier to change. */
207 #define lg_last_avail(vq) ((vq)->last_avail_idx)
210 * The virtio configuration space is defined to be little-endian. x86 is
211 * little-endian too, but it's nice to be explicit so we have these helpers.
213 #define cpu_to_le16(v16) (v16)
214 #define cpu_to_le32(v32) (v32)
215 #define cpu_to_le64(v64) (v64)
216 #define le16_to_cpu(v16) (v16)
217 #define le32_to_cpu(v32) (v32)
218 #define le64_to_cpu(v64) (v64)
220 /* Is this iovec empty? */
221 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
225 for (i = 0; i < num_iov; i++)
231 /* Take len bytes from the front of this iovec. */
232 static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
236 for (i = 0; i < num_iov; i++) {
239 used = iov[i].iov_len < len ? iov[i].iov_len : len;
240 iov[i].iov_base += used;
241 iov[i].iov_len -= used;
247 /* The device virtqueue descriptors are followed by feature bitmasks. */
248 static u8 *get_feature_bits(struct device *dev)
250 return (u8 *)(dev->desc + 1)
251 + dev->num_vq * sizeof(struct lguest_vqconfig);
255 * The Launcher code itself takes us out into userspace, that scary place where
256 * pointers run wild and free! Unfortunately, like most userspace programs,
257 * it's quite boring (which is why everyone likes to hack on the kernel!).
258 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
259 * you through this section. Or, maybe not.
261 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
262 * memory and stores it in "guest_base". In other words, Guest physical ==
263 * Launcher virtual with an offset.
265 * This can be tough to get your head around, but usually it just means that we
266 * use these trivial conversion functions when the Guest gives us its
267 * "physical" addresses:
269 static void *from_guest_phys(unsigned long addr)
271 return guest_base + addr;
274 static unsigned long to_guest_phys(const void *addr)
276 return (addr - guest_base);
280 * Loading the Kernel.
282 * We start with couple of simple helper routines. open_or_die() avoids
283 * error-checking code cluttering the callers:
285 static int open_or_die(const char *name, int flags)
287 int fd = open(name, flags);
289 err(1, "Failed to open %s", name);
293 /* map_zeroed_pages() takes a number of pages. */
294 static void *map_zeroed_pages(unsigned int num)
296 int fd = open_or_die("/dev/zero", O_RDONLY);
300 * We use a private mapping (ie. if we write to the page, it will be
301 * copied). We allocate an extra two pages PROT_NONE to act as guard
302 * pages against read/write attempts that exceed allocated space.
304 addr = mmap(NULL, getpagesize() * (num+2),
305 PROT_NONE, MAP_PRIVATE, fd, 0);
307 if (addr == MAP_FAILED)
308 err(1, "Mmapping %u pages of /dev/zero", num);
310 if (mprotect(addr + getpagesize(), getpagesize() * num,
311 PROT_READ|PROT_WRITE) == -1)
312 err(1, "mprotect rw %u pages failed", num);
315 * One neat mmap feature is that you can close the fd, and it
320 /* Return address after PROT_NONE page */
321 return addr + getpagesize();
324 /* Get some more pages for a device. */
325 static void *get_pages(unsigned int num)
327 void *addr = from_guest_phys(guest_limit);
329 guest_limit += num * getpagesize();
330 if (guest_limit > guest_max)
331 errx(1, "Not enough memory for devices");
336 * This routine is used to load the kernel or initrd. It tries mmap, but if
337 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
338 * it falls back to reading the memory in.
340 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
345 * We map writable even though for some segments are marked read-only.
346 * The kernel really wants to be writable: it patches its own
349 * MAP_PRIVATE means that the page won't be copied until a write is
350 * done to it. This allows us to share untouched memory between
353 if (mmap(addr, len, PROT_READ|PROT_WRITE,
354 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
357 /* pread does a seek and a read in one shot: saves a few lines. */
358 r = pread(fd, addr, len, offset);
360 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
364 * This routine takes an open vmlinux image, which is in ELF, and maps it into
365 * the Guest memory. ELF = Embedded Linking Format, which is the format used
366 * by all modern binaries on Linux including the kernel.
368 * The ELF headers give *two* addresses: a physical address, and a virtual
369 * address. We use the physical address; the Guest will map itself to the
372 * We return the starting address.
374 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
376 Elf32_Phdr phdr[ehdr->e_phnum];
380 * Sanity checks on the main ELF header: an x86 executable with a
381 * reasonable number of correctly-sized program headers.
383 if (ehdr->e_type != ET_EXEC
384 || ehdr->e_machine != EM_386
385 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
386 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
387 errx(1, "Malformed elf header");
390 * An ELF executable contains an ELF header and a number of "program"
391 * headers which indicate which parts ("segments") of the program to
395 /* We read in all the program headers at once: */
396 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
397 err(1, "Seeking to program headers");
398 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
399 err(1, "Reading program headers");
402 * Try all the headers: there are usually only three. A read-only one,
403 * a read-write one, and a "note" section which we don't load.
405 for (i = 0; i < ehdr->e_phnum; i++) {
406 /* If this isn't a loadable segment, we ignore it */
407 if (phdr[i].p_type != PT_LOAD)
410 verbose("Section %i: size %i addr %p\n",
411 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
413 /* We map this section of the file at its physical address. */
414 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
415 phdr[i].p_offset, phdr[i].p_filesz);
418 /* The entry point is given in the ELF header. */
419 return ehdr->e_entry;
423 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
424 * to jump into it and it will unpack itself. We used to have to perform some
425 * hairy magic because the unpacking code scared me.
427 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
428 * a small patch to jump over the tricky bits in the Guest, so now we just read
429 * the funky header so we know where in the file to load, and away we go!
431 static unsigned long load_bzimage(int fd)
433 struct boot_params boot;
435 /* Modern bzImages get loaded at 1M. */
436 void *p = from_guest_phys(0x100000);
439 * Go back to the start of the file and read the header. It should be
440 * a Linux boot header (see Documentation/x86/i386/boot.txt)
442 lseek(fd, 0, SEEK_SET);
443 read(fd, &boot, sizeof(boot));
445 /* Inside the setup_hdr, we expect the magic "HdrS" */
446 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
447 errx(1, "This doesn't look like a bzImage to me");
449 /* Skip over the extra sectors of the header. */
450 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
452 /* Now read everything into memory. in nice big chunks. */
453 while ((r = read(fd, p, 65536)) > 0)
456 /* Finally, code32_start tells us where to enter the kernel. */
457 return boot.hdr.code32_start;
461 * Loading the kernel is easy when it's a "vmlinux", but most kernels
462 * come wrapped up in the self-decompressing "bzImage" format. With a little
463 * work, we can load those, too.
465 static unsigned long load_kernel(int fd)
469 /* Read in the first few bytes. */
470 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
471 err(1, "Reading kernel");
473 /* If it's an ELF file, it starts with "\177ELF" */
474 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
475 return map_elf(fd, &hdr);
477 /* Otherwise we assume it's a bzImage, and try to load it. */
478 return load_bzimage(fd);
482 * This is a trivial little helper to align pages. Andi Kleen hated it because
483 * it calls getpagesize() twice: "it's dumb code."
485 * Kernel guys get really het up about optimization, even when it's not
486 * necessary. I leave this code as a reaction against that.
488 static inline unsigned long page_align(unsigned long addr)
490 /* Add upwards and truncate downwards. */
491 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
495 * An "initial ram disk" is a disk image loaded into memory along with the
496 * kernel which the kernel can use to boot from without needing any drivers.
497 * Most distributions now use this as standard: the initrd contains the code to
498 * load the appropriate driver modules for the current machine.
500 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
501 * kernels. He sent me this (and tells me when I break it).
503 static unsigned long load_initrd(const char *name, unsigned long mem)
509 ifd = open_or_die(name, O_RDONLY);
510 /* fstat() is needed to get the file size. */
511 if (fstat(ifd, &st) < 0)
512 err(1, "fstat() on initrd '%s'", name);
515 * We map the initrd at the top of memory, but mmap wants it to be
516 * page-aligned, so we round the size up for that.
518 len = page_align(st.st_size);
519 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
521 * Once a file is mapped, you can close the file descriptor. It's a
522 * little odd, but quite useful.
525 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
527 /* We return the initrd size. */
533 * Simple routine to roll all the commandline arguments together with spaces
536 static void concat(char *dst, char *args[])
538 unsigned int i, len = 0;
540 for (i = 0; args[i]; i++) {
542 strcat(dst+len, " ");
545 strcpy(dst+len, args[i]);
546 len += strlen(args[i]);
548 /* In case it's empty. */
553 * This is where we actually tell the kernel to initialize the Guest. We
554 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
555 * the base of Guest "physical" memory, the top physical page to allow and the
556 * entry point for the Guest.
558 static void tell_kernel(unsigned long start)
560 unsigned long args[] = { LHREQ_INITIALIZE,
561 (unsigned long)guest_base,
562 guest_limit / getpagesize(), start };
563 verbose("Guest: %p - %p (%#lx)\n",
564 guest_base, guest_base + guest_limit, guest_limit);
565 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
566 if (write(lguest_fd, args, sizeof(args)) < 0)
567 err(1, "Writing to /dev/lguest");
574 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
575 * We need to make sure it's not trying to reach into the Launcher itself, so
576 * we have a convenient routine which checks it and exits with an error message
577 * if something funny is going on:
579 static void *_check_pointer(unsigned long addr, unsigned int size,
583 * Check if the requested address and size exceeds the allocated memory,
584 * or addr + size wraps around.
586 if ((addr + size) > guest_limit || (addr + size) < addr)
587 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
589 * We return a pointer for the caller's convenience, now we know it's
592 return from_guest_phys(addr);
594 /* A macro which transparently hands the line number to the real function. */
595 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
598 * Each buffer in the virtqueues is actually a chain of descriptors. This
599 * function returns the next descriptor in the chain, or vq->vring.num if we're
602 static unsigned next_desc(struct vring_desc *desc,
603 unsigned int i, unsigned int max)
607 /* If this descriptor says it doesn't chain, we're done. */
608 if (!(desc[i].flags & VRING_DESC_F_NEXT))
611 /* Check they're not leading us off end of descriptors. */
613 /* Make sure compiler knows to grab that: we don't want it changing! */
617 errx(1, "Desc next is %u", next);
623 * This actually sends the interrupt for this virtqueue, if we've used a
626 static void trigger_irq(struct virtqueue *vq)
628 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
630 /* Don't inform them if nothing used. */
631 if (!vq->pending_used)
633 vq->pending_used = 0;
635 /* If they don't want an interrupt, don't send one... */
636 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
640 /* Send the Guest an interrupt tell them we used something up. */
641 if (write(lguest_fd, buf, sizeof(buf)) != 0)
642 err(1, "Triggering irq %i", vq->config.irq);
646 * This looks in the virtqueue for the first available buffer, and converts
647 * it to an iovec for convenient access. Since descriptors consist of some
648 * number of output then some number of input descriptors, it's actually two
649 * iovecs, but we pack them into one and note how many of each there were.
651 * This function waits if necessary, and returns the descriptor number found.
653 static unsigned wait_for_vq_desc(struct virtqueue *vq,
655 unsigned int *out_num, unsigned int *in_num)
657 unsigned int i, head, max;
658 struct vring_desc *desc;
659 u16 last_avail = lg_last_avail(vq);
661 /* There's nothing available? */
662 while (last_avail == vq->vring.avail->idx) {
666 * Since we're about to sleep, now is a good time to tell the
667 * Guest about what we've used up to now.
671 /* OK, now we need to know about added descriptors. */
672 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
675 * They could have slipped one in as we were doing that: make
676 * sure it's written, then check again.
679 if (last_avail != vq->vring.avail->idx) {
680 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
684 /* Nothing new? Wait for eventfd to tell us they refilled. */
685 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
686 errx(1, "Event read failed?");
688 /* We don't need to be notified again. */
689 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
692 /* Check it isn't doing very strange things with descriptor numbers. */
693 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
694 errx(1, "Guest moved used index from %u to %u",
695 last_avail, vq->vring.avail->idx);
698 * Grab the next descriptor number they're advertising, and increment
699 * the index we've seen.
701 head = vq->vring.avail->ring[last_avail % vq->vring.num];
704 /* If their number is silly, that's a fatal mistake. */
705 if (head >= vq->vring.num)
706 errx(1, "Guest says index %u is available", head);
708 /* When we start there are none of either input nor output. */
709 *out_num = *in_num = 0;
712 desc = vq->vring.desc;
716 * If this is an indirect entry, then this buffer contains a descriptor
717 * table which we handle as if it's any normal descriptor chain.
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");
723 max = desc[i].len / sizeof(struct vring_desc);
724 desc = check_pointer(desc[i].addr, desc[i].len);
729 /* Grab the first descriptor, and check it's OK. */
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 /* If this is an input descriptor, increment that count. */
734 if (desc[i].flags & VRING_DESC_F_WRITE)
738 * If it's an output descriptor, they're all supposed
739 * to come before any input descriptors.
742 errx(1, "Descriptor has out after in");
746 /* If we've got too many, that implies a descriptor loop. */
747 if (*out_num + *in_num > max)
748 errx(1, "Looped descriptor");
749 } while ((i = next_desc(desc, i, max)) != max);
755 * After we've used one of their buffers, we tell the Guest about it. Sometime
756 * later we'll want to send them an interrupt using trigger_irq(); note that
757 * wait_for_vq_desc() does that for us if it has to wait.
759 static void add_used(struct virtqueue *vq, unsigned int head, int len)
761 struct vring_used_elem *used;
764 * The virtqueue contains a ring of used buffers. Get a pointer to the
765 * next entry in that used ring.
767 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
770 /* Make sure buffer is written before we update index. */
772 vq->vring.used->idx++;
776 /* And here's the combo meal deal. Supersize me! */
777 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
779 add_used(vq, head, len);
786 * We associate some data with the console for our exit hack.
788 struct console_abort {
789 /* How many times have they hit ^C? */
791 /* When did they start? */
792 struct timeval start;
795 /* This is the routine which handles console input (ie. stdin). */
796 static void console_input(struct virtqueue *vq)
799 unsigned int head, in_num, out_num;
800 struct console_abort *abort = vq->dev->priv;
801 struct iovec iov[vq->vring.num];
803 /* Make sure there's a descriptor available. */
804 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
806 errx(1, "Output buffers in console in queue?");
808 /* Read into it. This is where we usually wait. */
809 len = readv(STDIN_FILENO, iov, in_num);
811 /* Ran out of input? */
812 warnx("Failed to get console input, ignoring console.");
814 * For simplicity, dying threads kill the whole Launcher. So
821 /* Tell the Guest we used a buffer. */
822 add_used_and_trigger(vq, head, len);
825 * Three ^C within one second? Exit.
827 * This is such a hack, but works surprisingly well. Each ^C has to
828 * be in a buffer by itself, so they can't be too fast. But we check
829 * that we get three within about a second, so they can't be too
832 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
838 if (abort->count == 1)
839 gettimeofday(&abort->start, NULL);
840 else if (abort->count == 3) {
842 gettimeofday(&now, NULL);
843 /* Kill all Launcher processes with SIGINT, like normal ^C */
844 if (now.tv_sec <= abort->start.tv_sec+1)
850 /* This is the routine which handles console output (ie. stdout). */
851 static void console_output(struct virtqueue *vq)
853 unsigned int head, out, in;
854 struct iovec iov[vq->vring.num];
856 /* We usually wait in here, for the Guest to give us something. */
857 head = wait_for_vq_desc(vq, iov, &out, &in);
859 errx(1, "Input buffers in console output queue?");
861 /* writev can return a partial write, so we loop here. */
862 while (!iov_empty(iov, out)) {
863 int len = writev(STDOUT_FILENO, iov, out);
865 warn("Write to stdout gave %i (%d)", len, errno);
868 iov_consume(iov, out, len);
872 * We're finished with that buffer: if we're going to sleep,
873 * wait_for_vq_desc() will prod the Guest with an interrupt.
875 add_used(vq, head, 0);
881 * Handling output for network is also simple: we get all the output buffers
882 * and write them to /dev/net/tun.
888 static void net_output(struct virtqueue *vq)
890 struct net_info *net_info = vq->dev->priv;
891 unsigned int head, out, in;
892 struct iovec iov[vq->vring.num];
894 /* We usually wait in here for the Guest to give us a packet. */
895 head = wait_for_vq_desc(vq, iov, &out, &in);
897 errx(1, "Input buffers in net output queue?");
899 * Send the whole thing through to /dev/net/tun. It expects the exact
900 * same format: what a coincidence!
902 if (writev(net_info->tunfd, iov, out) < 0)
903 warnx("Write to tun failed (%d)?", errno);
906 * Done with that one; wait_for_vq_desc() will send the interrupt if
907 * all packets are processed.
909 add_used(vq, head, 0);
913 * Handling network input is a bit trickier, because I've tried to optimize it.
915 * First we have a helper routine which tells is if from this file descriptor
916 * (ie. the /dev/net/tun device) will block:
918 static bool will_block(int fd)
921 struct timeval zero = { 0, 0 };
924 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
928 * This handles packets coming in from the tun device to our Guest. Like all
929 * service routines, it gets called again as soon as it returns, so you don't
930 * see a while(1) loop here.
932 static void net_input(struct virtqueue *vq)
935 unsigned int head, out, in;
936 struct iovec iov[vq->vring.num];
937 struct net_info *net_info = vq->dev->priv;
940 * Get a descriptor to write an incoming packet into. This will also
941 * send an interrupt if they're out of descriptors.
943 head = wait_for_vq_desc(vq, iov, &out, &in);
945 errx(1, "Output buffers in net input queue?");
948 * If it looks like we'll block reading from the tun device, send them
951 if (vq->pending_used && will_block(net_info->tunfd))
955 * Read in the packet. This is where we normally wait (when there's no
956 * incoming network traffic).
958 len = readv(net_info->tunfd, iov, in);
960 warn("Failed to read from tun (%d).", errno);
963 * Mark that packet buffer as used, but don't interrupt here. We want
964 * to wait until we've done as much work as we can.
966 add_used(vq, head, len);
970 /* This is the helper to create threads: run the service routine in a loop. */
971 static int do_thread(void *_vq)
973 struct virtqueue *vq = _vq;
981 * When a child dies, we kill our entire process group with SIGTERM. This
982 * also has the side effect that the shell restores the console for us!
984 static void kill_launcher(int signal)
989 static void reset_device(struct device *dev)
991 struct virtqueue *vq;
993 verbose("Resetting device %s\n", dev->name);
995 /* Clear any features they've acked. */
996 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
998 /* We're going to be explicitly killing threads, so ignore them. */
999 signal(SIGCHLD, SIG_IGN);
1001 /* Zero out the virtqueues, get rid of their threads */
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;
1008 memset(vq->vring.desc, 0,
1009 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1010 lg_last_avail(vq) = 0;
1012 dev->running = false;
1014 /* Now we care if threads die. */
1015 signal(SIGCHLD, (void *)kill_launcher);
1019 * This actually creates the thread which services the virtqueue for a device.
1021 static void create_thread(struct virtqueue *vq)
1024 * Create stack for thread. Since the stack grows upwards, we point
1025 * the stack pointer to the end of this region.
1027 char *stack = malloc(32768);
1028 unsigned long args[] = { LHREQ_EVENTFD,
1029 vq->config.pfn*getpagesize(), 0 };
1031 /* Create a zero-initialized eventfd. */
1032 vq->eventfd = eventfd(0, 0);
1033 if (vq->eventfd < 0)
1034 err(1, "Creating eventfd");
1035 args[2] = vq->eventfd;
1038 * Attach an eventfd to this virtqueue: it will go off when the Guest
1039 * does an LHCALL_NOTIFY for this vq.
1041 if (write(lguest_fd, &args, sizeof(args)) != 0)
1042 err(1, "Attaching eventfd");
1045 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1046 * we get a signal if it dies.
1048 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1049 if (vq->thread == (pid_t)-1)
1050 err(1, "Creating clone");
1052 /* We close our local copy now the child has it. */
1056 static void start_device(struct device *dev)
1059 struct virtqueue *vq;
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]);
1069 for (vq = dev->vq; vq; vq = vq->next) {
1073 dev->running = true;
1076 static void cleanup_devices(void)
1080 for (dev = devices.dev; dev; dev = dev->next)
1083 /* If we saved off the original terminal settings, restore them now. */
1084 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1085 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1088 /* When the Guest tells us they updated the status field, we handle it. */
1089 static void update_device_status(struct device *dev)
1091 /* A zero status is a reset, otherwise it's a set of flags. */
1092 if (dev->desc->status == 0)
1094 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1095 warnx("Device %s configuration FAILED", dev->name);
1100 err(1, "Device %s features finalized twice", dev->name);
1106 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1107 * particular, it's used to notify us of device status changes during boot.
1109 static void handle_output(unsigned long addr)
1113 /* Check each device. */
1114 for (i = devices.dev; i; i = i->next) {
1115 struct virtqueue *vq;
1118 * Notifications to device descriptors mean they updated the
1121 if (from_guest_phys(addr) == i->desc) {
1122 update_device_status(i);
1126 /* Devices should not be used before features are finalized. */
1127 for (vq = i->vq; vq; vq = vq->next) {
1128 if (addr != vq->config.pfn*getpagesize())
1130 errx(1, "Notification on %s before setup!", i->name);
1135 * Early console write is done using notify on a nul-terminated string
1136 * in Guest memory. It's also great for hacking debugging messages
1139 if (addr >= guest_limit)
1140 errx(1, "Bad NOTIFY %#lx", addr);
1142 write(STDOUT_FILENO, from_guest_phys(addr),
1143 strnlen(from_guest_phys(addr), guest_limit - addr));
1149 * All devices need a descriptor so the Guest knows it exists, and a "struct
1150 * device" so the Launcher can keep track of it. We have common helper
1151 * routines to allocate and manage them.
1155 * The layout of the device page is a "struct lguest_device_desc" followed by a
1156 * number of virtqueue descriptors, then two sets of feature bits, then an
1157 * array of configuration bytes. This routine returns the configuration
1160 static u8 *device_config(const struct device *dev)
1162 return (void *)(dev->desc + 1)
1163 + dev->num_vq * sizeof(struct lguest_vqconfig)
1164 + dev->feature_len * 2;
1168 * This routine allocates a new "struct lguest_device_desc" from descriptor
1169 * table page just above the Guest's normal memory. It returns a pointer to
1172 static struct lguest_device_desc *new_dev_desc(u16 type)
1174 struct lguest_device_desc d = { .type = type };
1177 /* Figure out where the next device config is, based on the last one. */
1178 if (devices.lastdev)
1179 p = device_config(devices.lastdev)
1180 + devices.lastdev->desc->config_len;
1182 p = devices.descpage;
1184 /* We only have one page for all the descriptors. */
1185 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1186 errx(1, "Too many devices");
1188 /* p might not be aligned, so we memcpy in. */
1189 return memcpy(p, &d, sizeof(d));
1193 * Each device descriptor is followed by the description of its virtqueues. We
1194 * specify how many descriptors the virtqueue is to have.
1196 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1197 void (*service)(struct virtqueue *))
1200 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1203 /* First we need some memory for this virtqueue. */
1204 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1206 p = get_pages(pages);
1208 /* Initialize the virtqueue */
1210 vq->last_avail_idx = 0;
1214 * This is the routine the service thread will run, and its Process ID
1215 * once it's running.
1217 vq->service = service;
1218 vq->thread = (pid_t)-1;
1220 /* Initialize the configuration. */
1221 vq->config.num = num_descs;
1222 vq->config.irq = devices.next_irq++;
1223 vq->config.pfn = to_guest_phys(p) / getpagesize();
1225 /* Initialize the vring. */
1226 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1229 * Append virtqueue to this device's descriptor. We use
1230 * device_config() to get the end of the device's current virtqueues;
1231 * we check that we haven't added any config or feature information
1232 * yet, otherwise we'd be overwriting them.
1234 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1235 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1237 dev->desc->num_vq++;
1239 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1242 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1245 for (i = &dev->vq; *i; i = &(*i)->next);
1250 * The first half of the feature bitmask is for us to advertise features. The
1251 * second half is for the Guest to accept features.
1253 static void add_feature(struct device *dev, unsigned bit)
1255 u8 *features = get_feature_bits(dev);
1257 /* We can't extend the feature bits once we've added config bytes */
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;
1263 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1267 * This routine sets the configuration fields for an existing device's
1268 * descriptor. It only works for the last device, but that's OK because that's
1271 static void set_config(struct device *dev, unsigned len, const void *conf)
1273 /* Check we haven't overflowed our single page. */
1274 if (device_config(dev) + len > devices.descpage + getpagesize())
1275 errx(1, "Too many devices");
1277 /* Copy in the config information, and store the length. */
1278 memcpy(device_config(dev), conf, len);
1279 dev->desc->config_len = len;
1281 /* Size must fit in config_len field (8 bits)! */
1282 assert(dev->desc->config_len == len);
1286 * This routine does all the creation and setup of a new device, including
1287 * calling new_dev_desc() to allocate the descriptor and device memory. We
1288 * don't actually start the service threads until later.
1290 * See what I mean about userspace being boring?
1292 static struct device *new_device(const char *name, u16 type)
1294 struct device *dev = malloc(sizeof(*dev));
1296 /* Now we populate the fields one at a time. */
1297 dev->desc = new_dev_desc(type);
1300 dev->feature_len = 0;
1302 dev->running = false;
1305 * Append to device list. Prepending to a single-linked list is
1306 * easier, but the user expects the devices to be arranged on the bus
1307 * in command-line order. The first network device on the command line
1308 * is eth0, the first block device /dev/vda, etc.
1310 if (devices.lastdev)
1311 devices.lastdev->next = dev;
1314 devices.lastdev = dev;
1320 * Our first setup routine is the console. It's a fairly simple device, but
1321 * UNIX tty handling makes it uglier than it could be.
1323 static void setup_console(void)
1327 /* If we can save the initial standard input settings... */
1328 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1329 struct termios term = orig_term;
1331 * Then we turn off echo, line buffering and ^C etc: We want a
1332 * raw input stream to the Guest.
1334 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1335 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1338 dev = new_device("console", VIRTIO_ID_CONSOLE);
1340 /* We store the console state in dev->priv, and initialize it. */
1341 dev->priv = malloc(sizeof(struct console_abort));
1342 ((struct console_abort *)dev->priv)->count = 0;
1345 * The console needs two virtqueues: the input then the output. When
1346 * they put something the input queue, we make sure we're listening to
1347 * stdin. When they put something in the output queue, we write it to
1350 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1351 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1353 verbose("device %u: console\n", ++devices.device_num);
1358 * Inter-guest networking is an interesting area. Simplest is to have a
1359 * --sharenet=<name> option which opens or creates a named pipe. This can be
1360 * used to send packets to another guest in a 1:1 manner.
1362 * More sopisticated is to use one of the tools developed for project like UML
1365 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1366 * completely generic ("here's my vring, attach to your vring") and would work
1367 * for any traffic. Of course, namespace and permissions issues need to be
1368 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1369 * multiple inter-guest channels behind one interface, although it would
1370 * require some manner of hotplugging new virtio channels.
1372 * Finally, we could implement a virtio network switch in the kernel.
1375 static u32 str2ip(const char *ipaddr)
1379 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1380 errx(1, "Failed to parse IP address '%s'", ipaddr);
1381 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1384 static void str2mac(const char *macaddr, unsigned char mac[6])
1387 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1388 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1389 errx(1, "Failed to parse mac address '%s'", macaddr);
1399 * This code is "adapted" from libbridge: it attaches the Host end of the
1400 * network device to the bridge device specified by the command line.
1402 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1403 * dislike bridging), and I just try not to break it.
1405 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1411 errx(1, "must specify bridge name");
1413 ifidx = if_nametoindex(if_name);
1415 errx(1, "interface %s does not exist!", if_name);
1417 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1418 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1419 ifr.ifr_ifindex = ifidx;
1420 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1421 err(1, "can't add %s to bridge %s", if_name, br_name);
1425 * This sets up the Host end of the network device with an IP address, brings
1426 * it up so packets will flow, the copies the MAC address into the hwaddr
1429 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1432 struct sockaddr_in sin;
1434 memset(&ifr, 0, sizeof(ifr));
1435 strcpy(ifr.ifr_name, tapif);
1437 /* Don't read these incantations. Just cut & paste them like I did! */
1438 sin.sin_family = AF_INET;
1439 sin.sin_addr.s_addr = htonl(ipaddr);
1440 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1441 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1442 err(1, "Setting %s interface address", tapif);
1443 ifr.ifr_flags = IFF_UP;
1444 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1445 err(1, "Bringing interface %s up", tapif);
1448 static int get_tun_device(char tapif[IFNAMSIZ])
1453 /* Start with this zeroed. Messy but sure. */
1454 memset(&ifr, 0, sizeof(ifr));
1457 * We open the /dev/net/tun device and tell it we want a tap device. A
1458 * tap device is like a tun device, only somehow different. To tell
1459 * the truth, I completely blundered my way through this code, but it
1462 netfd = open_or_die("/dev/net/tun", O_RDWR);
1463 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1464 strcpy(ifr.ifr_name, "tap%d");
1465 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1466 err(1, "configuring /dev/net/tun");
1468 if (ioctl(netfd, TUNSETOFFLOAD,
1469 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1470 err(1, "Could not set features for tun device");
1473 * We don't need checksums calculated for packets coming in this
1476 ioctl(netfd, TUNSETNOCSUM, 1);
1478 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1483 * Our network is a Host<->Guest network. This can either use bridging or
1484 * routing, but the principle is the same: it uses the "tun" device to inject
1485 * packets into the Host as if they came in from a normal network card. We
1486 * just shunt packets between the Guest and the tun device.
1488 static void setup_tun_net(char *arg)
1491 struct net_info *net_info = malloc(sizeof(*net_info));
1493 u32 ip = INADDR_ANY;
1494 bool bridging = false;
1495 char tapif[IFNAMSIZ], *p;
1496 struct virtio_net_config conf;
1498 net_info->tunfd = get_tun_device(tapif);
1500 /* First we create a new network device. */
1501 dev = new_device("net", VIRTIO_ID_NET);
1502 dev->priv = net_info;
1504 /* Network devices need a recv and a send queue, just like console. */
1505 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1506 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1509 * We need a socket to perform the magic network ioctls to bring up the
1510 * tap interface, connect to the bridge etc. Any socket will do!
1512 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1514 err(1, "opening IP socket");
1516 /* If the command line was --tunnet=bridge:<name> do bridging. */
1517 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1518 arg += strlen(BRIDGE_PFX);
1522 /* A mac address may follow the bridge name or IP address */
1523 p = strchr(arg, ':');
1525 str2mac(p+1, conf.mac);
1526 add_feature(dev, VIRTIO_NET_F_MAC);
1530 /* arg is now either an IP address or a bridge name */
1532 add_to_bridge(ipfd, tapif, arg);
1536 /* Set up the tun device. */
1537 configure_device(ipfd, tapif, ip);
1539 /* Expect Guest to handle everything except UFO */
1540 add_feature(dev, VIRTIO_NET_F_CSUM);
1541 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1542 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1543 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1544 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1545 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1546 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1547 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1548 /* We handle indirect ring entries */
1549 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1550 set_config(dev, sizeof(conf), &conf);
1552 /* We don't need the socket any more; setup is done. */
1555 devices.device_num++;
1558 verbose("device %u: tun %s attached to bridge: %s\n",
1559 devices.device_num, tapif, arg);
1561 verbose("device %u: tun %s: %s\n",
1562 devices.device_num, tapif, arg);
1566 /* This hangs off device->priv. */
1568 /* The size of the file. */
1571 /* The file descriptor for the file. */
1579 * The disk only has one virtqueue, so it only has one thread. It is really
1580 * simple: the Guest asks for a block number and we read or write that position
1583 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1584 * slow: the Guest waits until the read is finished before running anything
1585 * else, even if it could have been doing useful work.
1587 * We could have used async I/O, except it's reputed to suck so hard that
1588 * characters actually go missing from your code when you try to use it.
1590 static void blk_request(struct virtqueue *vq)
1592 struct vblk_info *vblk = vq->dev->priv;
1593 unsigned int head, out_num, in_num, wlen;
1596 struct virtio_blk_outhdr *out;
1597 struct iovec iov[vq->vring.num];
1601 * Get the next request, where we normally wait. It triggers the
1602 * interrupt to acknowledge previously serviced requests (if any).
1604 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1607 * Every block request should contain at least one output buffer
1608 * (detailing the location on disk and the type of request) and one
1609 * input buffer (to hold the result).
1611 if (out_num == 0 || in_num == 0)
1612 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1613 head, out_num, in_num);
1615 out = convert(&iov[0], struct virtio_blk_outhdr);
1616 in = convert(&iov[out_num+in_num-1], u8);
1618 * For historical reasons, block operations are expressed in 512 byte
1621 off = out->sector * 512;
1624 * In general the virtio block driver is allowed to try SCSI commands.
1625 * It'd be nice if we supported eject, for example, but we don't.
1627 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1628 fprintf(stderr, "Scsi commands unsupported\n");
1629 *in = VIRTIO_BLK_S_UNSUPP;
1631 } else if (out->type & VIRTIO_BLK_T_OUT) {
1635 * Move to the right location in the block file. This can fail
1636 * if they try to write past end.
1638 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1639 err(1, "Bad seek to sector %llu", out->sector);
1641 ret = writev(vblk->fd, iov+1, out_num-1);
1642 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1645 * Grr... Now we know how long the descriptor they sent was, we
1646 * make sure they didn't try to write over the end of the block
1647 * file (possibly extending it).
1649 if (ret > 0 && off + ret > vblk->len) {
1650 /* Trim it back to the correct length */
1651 ftruncate64(vblk->fd, vblk->len);
1652 /* Die, bad Guest, die. */
1653 errx(1, "Write past end %llu+%u", off, ret);
1657 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1658 } else if (out->type & VIRTIO_BLK_T_FLUSH) {
1660 ret = fdatasync(vblk->fd);
1661 verbose("FLUSH fdatasync: %i\n", ret);
1663 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1668 * Move to the right location in the block file. This can fail
1669 * if they try to read past end.
1671 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1672 err(1, "Bad seek to sector %llu", out->sector);
1674 ret = readv(vblk->fd, iov+1, in_num-1);
1675 verbose("READ from sector %llu: %i\n", out->sector, ret);
1677 wlen = sizeof(*in) + ret;
1678 *in = VIRTIO_BLK_S_OK;
1681 *in = VIRTIO_BLK_S_IOERR;
1685 /* Finished that request. */
1686 add_used(vq, head, wlen);
1689 /*L:198 This actually sets up a virtual block device. */
1690 static void setup_block_file(const char *filename)
1693 struct vblk_info *vblk;
1694 struct virtio_blk_config conf;
1696 /* Creat the device. */
1697 dev = new_device("block", VIRTIO_ID_BLOCK);
1699 /* The device has one virtqueue, where the Guest places requests. */
1700 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1702 /* Allocate the room for our own bookkeeping */
1703 vblk = dev->priv = malloc(sizeof(*vblk));
1705 /* First we open the file and store the length. */
1706 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1707 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1709 /* We support FLUSH. */
1710 add_feature(dev, VIRTIO_BLK_F_FLUSH);
1712 /* Tell Guest how many sectors this device has. */
1713 conf.capacity = cpu_to_le64(vblk->len / 512);
1716 * Tell Guest not to put in too many descriptors at once: two are used
1717 * for the in and out elements.
1719 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1720 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1722 /* Don't try to put whole struct: we have 8 bit limit. */
1723 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1725 verbose("device %u: virtblock %llu sectors\n",
1726 ++devices.device_num, le64_to_cpu(conf.capacity));
1730 * Our random number generator device reads from /dev/random into the Guest's
1731 * input buffers. The usual case is that the Guest doesn't want random numbers
1732 * and so has no buffers although /dev/random is still readable, whereas
1733 * console is the reverse.
1735 * The same logic applies, however.
1741 static void rng_input(struct virtqueue *vq)
1744 unsigned int head, in_num, out_num, totlen = 0;
1745 struct rng_info *rng_info = vq->dev->priv;
1746 struct iovec iov[vq->vring.num];
1748 /* First we need a buffer from the Guests's virtqueue. */
1749 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1751 errx(1, "Output buffers in rng?");
1754 * Just like the console write, we loop to cover the whole iovec.
1755 * In this case, short reads actually happen quite a bit.
1757 while (!iov_empty(iov, in_num)) {
1758 len = readv(rng_info->rfd, iov, in_num);
1760 err(1, "Read from /dev/random gave %i", len);
1761 iov_consume(iov, in_num, len);
1765 /* Tell the Guest about the new input. */
1766 add_used(vq, head, totlen);
1770 * This creates a "hardware" random number device for the Guest.
1772 static void setup_rng(void)
1775 struct rng_info *rng_info = malloc(sizeof(*rng_info));
1777 /* Our device's privat info simply contains the /dev/random fd. */
1778 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1780 /* Create the new device. */
1781 dev = new_device("rng", VIRTIO_ID_RNG);
1782 dev->priv = rng_info;
1784 /* The device has one virtqueue, where the Guest places inbufs. */
1785 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1787 verbose("device %u: rng\n", devices.device_num++);
1789 /* That's the end of device setup. */
1791 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1792 static void __attribute__((noreturn)) restart_guest(void)
1797 * Since we don't track all open fds, we simply close everything beyond
1800 for (i = 3; i < FD_SETSIZE; i++)
1803 /* Reset all the devices (kills all threads). */
1806 execv(main_args[0], main_args);
1807 err(1, "Could not exec %s", main_args[0]);
1811 * Finally we reach the core of the Launcher which runs the Guest, serves
1812 * its input and output, and finally, lays it to rest.
1814 static void __attribute__((noreturn)) run_guest(void)
1817 unsigned long notify_addr;
1820 /* We read from the /dev/lguest device to run the Guest. */
1821 readval = pread(lguest_fd, ¬ify_addr,
1822 sizeof(notify_addr), cpu_id);
1824 /* One unsigned long means the Guest did HCALL_NOTIFY */
1825 if (readval == sizeof(notify_addr)) {
1826 verbose("Notify on address %#lx\n", notify_addr);
1827 handle_output(notify_addr);
1828 /* ENOENT means the Guest died. Reading tells us why. */
1829 } else if (errno == ENOENT) {
1830 char reason[1024] = { 0 };
1831 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1832 errx(1, "%s", reason);
1833 /* ERESTART means that we need to reboot the guest */
1834 } else if (errno == ERESTART) {
1836 /* Anything else means a bug or incompatible change. */
1838 err(1, "Running guest failed");
1842 * This is the end of the Launcher. The good news: we are over halfway
1843 * through! The bad news: the most fiendish part of the code still lies ahead
1846 * Are you ready? Take a deep breath and join me in the core of the Host, in
1850 static struct option opts[] = {
1851 { "verbose", 0, NULL, 'v' },
1852 { "tunnet", 1, NULL, 't' },
1853 { "block", 1, NULL, 'b' },
1854 { "rng", 0, NULL, 'r' },
1855 { "initrd", 1, NULL, 'i' },
1856 { "username", 1, NULL, 'u' },
1857 { "chroot", 1, NULL, 'c' },
1860 static void usage(void)
1862 errx(1, "Usage: lguest [--verbose] "
1863 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1864 "|--block=<filename>|--initrd=<filename>]...\n"
1865 "<mem-in-mb> vmlinux [args...]");
1868 /*L:105 The main routine is where the real work begins: */
1869 int main(int argc, char *argv[])
1871 /* Memory, code startpoint and size of the (optional) initrd. */
1872 unsigned long mem = 0, start, initrd_size = 0;
1873 /* Two temporaries. */
1875 /* The boot information for the Guest. */
1876 struct boot_params *boot;
1877 /* If they specify an initrd file to load. */
1878 const char *initrd_name = NULL;
1880 /* Password structure for initgroups/setres[gu]id */
1881 struct passwd *user_details = NULL;
1883 /* Directory to chroot to */
1884 char *chroot_path = NULL;
1886 /* Save the args: we "reboot" by execing ourselves again. */
1890 * First we initialize the device list. We keep a pointer to the last
1891 * device, and the next interrupt number to use for devices (1:
1892 * remember that 0 is used by the timer).
1894 devices.lastdev = NULL;
1895 devices.next_irq = 1;
1897 /* We're CPU 0. In fact, that's the only CPU possible right now. */
1901 * We need to know how much memory so we can set up the device
1902 * descriptor and memory pages for the devices as we parse the command
1903 * line. So we quickly look through the arguments to find the amount
1906 for (i = 1; i < argc; i++) {
1907 if (argv[i][0] != '-') {
1908 mem = atoi(argv[i]) * 1024 * 1024;
1910 * We start by mapping anonymous pages over all of
1911 * guest-physical memory range. This fills it with 0,
1912 * and ensures that the Guest won't be killed when it
1913 * tries to access it.
1915 guest_base = map_zeroed_pages(mem / getpagesize()
1918 guest_max = mem + DEVICE_PAGES*getpagesize();
1919 devices.descpage = get_pages(1);
1924 /* The options are fairly straight-forward */
1925 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1931 setup_tun_net(optarg);
1934 setup_block_file(optarg);
1940 initrd_name = optarg;
1943 user_details = getpwnam(optarg);
1945 err(1, "getpwnam failed, incorrect username?");
1948 chroot_path = optarg;
1951 warnx("Unknown argument %s", argv[optind]);
1956 * After the other arguments we expect memory and kernel image name,
1957 * followed by command line arguments for the kernel.
1959 if (optind + 2 > argc)
1962 verbose("Guest base is at %p\n", guest_base);
1964 /* We always have a console device */
1967 /* Now we load the kernel */
1968 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1970 /* Boot information is stashed at physical address 0 */
1971 boot = from_guest_phys(0);
1973 /* Map the initrd image if requested (at top of physical memory) */
1975 initrd_size = load_initrd(initrd_name, mem);
1977 * These are the location in the Linux boot header where the
1978 * start and size of the initrd are expected to be found.
1980 boot->hdr.ramdisk_image = mem - initrd_size;
1981 boot->hdr.ramdisk_size = initrd_size;
1982 /* The bootloader type 0xFF means "unknown"; that's OK. */
1983 boot->hdr.type_of_loader = 0xFF;
1987 * The Linux boot header contains an "E820" memory map: ours is a
1988 * simple, single region.
1990 boot->e820_entries = 1;
1991 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1993 * The boot header contains a command line pointer: we put the command
1994 * line after the boot header.
1996 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1997 /* We use a simple helper to copy the arguments separated by spaces. */
1998 concat((char *)(boot + 1), argv+optind+2);
2000 /* Boot protocol version: 2.07 supports the fields for lguest. */
2001 boot->hdr.version = 0x207;
2003 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2004 boot->hdr.hardware_subarch = 1;
2006 /* Tell the entry path not to try to reload segment registers. */
2007 boot->hdr.loadflags |= KEEP_SEGMENTS;
2010 * We tell the kernel to initialize the Guest: this returns the open
2011 * /dev/lguest file descriptor.
2015 /* Ensure that we terminate if a device-servicing child dies. */
2016 signal(SIGCHLD, kill_launcher);
2018 /* If we exit via err(), this kills all the threads, restores tty. */
2019 atexit(cleanup_devices);
2021 /* If requested, chroot to a directory */
2023 if (chroot(chroot_path) != 0)
2024 err(1, "chroot(\"%s\") failed", chroot_path);
2026 if (chdir("/") != 0)
2027 err(1, "chdir(\"/\") failed");
2029 verbose("chroot done\n");
2032 /* If requested, drop privileges */
2037 u = user_details->pw_uid;
2038 g = user_details->pw_gid;
2040 if (initgroups(user_details->pw_name, g) != 0)
2041 err(1, "initgroups failed");
2043 if (setresgid(g, g, g) != 0)
2044 err(1, "setresgid failed");
2046 if (setresuid(u, u, u) != 0)
2047 err(1, "setresuid failed");
2049 verbose("Dropping privileges completed\n");
2052 /* Finally, run the Guest. This doesn't return. */
2058 * Mastery is done: you now know everything I do.
2060 * But surely you have seen code, features and bugs in your wanderings which
2061 * you now yearn to attack? That is the real game, and I look forward to you
2062 * patching and forking lguest into the Your-Name-Here-visor.
2064 * Farewell, and good coding!