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 /* Does Guest want an intrrupt on empty? */
141 /* Device-specific data. */
145 /* The virtqueue structure describes a queue attached to a device. */
147 struct virtqueue *next;
149 /* Which device owns me. */
152 /* The configuration for this queue. */
153 struct lguest_vqconfig config;
155 /* The actual ring of buffers. */
158 /* Last available index we saw. */
161 /* How many are used since we sent last irq? */
162 unsigned int pending_used;
164 /* Eventfd where Guest notifications arrive. */
167 /* Function for the thread which is servicing this virtqueue. */
168 void (*service)(struct virtqueue *vq);
172 /* Remember the arguments to the program so we can "reboot" */
173 static char **main_args;
175 /* The original tty settings to restore on exit. */
176 static struct termios orig_term;
179 * We have to be careful with barriers: our devices are all run in separate
180 * threads and so we need to make sure that changes visible to the Guest happen
183 #define wmb() __asm__ __volatile__("" : : : "memory")
184 #define mb() __asm__ __volatile__("" : : : "memory")
187 * Convert an iovec element to the given type.
189 * This is a fairly ugly trick: we need to know the size of the type and
190 * alignment requirement to check the pointer is kosher. It's also nice to
191 * have the name of the type in case we report failure.
193 * Typing those three things all the time is cumbersome and error prone, so we
194 * have a macro which sets them all up and passes to the real function.
196 #define convert(iov, type) \
197 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
199 static void *_convert(struct iovec *iov, size_t size, size_t align,
202 if (iov->iov_len != size)
203 errx(1, "Bad iovec size %zu for %s", iov->iov_len, name);
204 if ((unsigned long)iov->iov_base % align != 0)
205 errx(1, "Bad alignment %p for %s", iov->iov_base, name);
206 return iov->iov_base;
209 /* Wrapper for the last available index. Makes it easier to change. */
210 #define lg_last_avail(vq) ((vq)->last_avail_idx)
213 * The virtio configuration space is defined to be little-endian. x86 is
214 * little-endian too, but it's nice to be explicit so we have these helpers.
216 #define cpu_to_le16(v16) (v16)
217 #define cpu_to_le32(v32) (v32)
218 #define cpu_to_le64(v64) (v64)
219 #define le16_to_cpu(v16) (v16)
220 #define le32_to_cpu(v32) (v32)
221 #define le64_to_cpu(v64) (v64)
223 /* Is this iovec empty? */
224 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
228 for (i = 0; i < num_iov; i++)
234 /* Take len bytes from the front of this iovec. */
235 static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len)
239 for (i = 0; i < num_iov; i++) {
242 used = iov[i].iov_len < len ? iov[i].iov_len : len;
243 iov[i].iov_base += used;
244 iov[i].iov_len -= used;
250 /* The device virtqueue descriptors are followed by feature bitmasks. */
251 static u8 *get_feature_bits(struct device *dev)
253 return (u8 *)(dev->desc + 1)
254 + dev->num_vq * sizeof(struct lguest_vqconfig);
258 * The Launcher code itself takes us out into userspace, that scary place where
259 * pointers run wild and free! Unfortunately, like most userspace programs,
260 * it's quite boring (which is why everyone likes to hack on the kernel!).
261 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
262 * you through this section. Or, maybe not.
264 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
265 * memory and stores it in "guest_base". In other words, Guest physical ==
266 * Launcher virtual with an offset.
268 * This can be tough to get your head around, but usually it just means that we
269 * use these trivial conversion functions when the Guest gives us its
270 * "physical" addresses:
272 static void *from_guest_phys(unsigned long addr)
274 return guest_base + addr;
277 static unsigned long to_guest_phys(const void *addr)
279 return (addr - guest_base);
283 * Loading the Kernel.
285 * We start with couple of simple helper routines. open_or_die() avoids
286 * error-checking code cluttering the callers:
288 static int open_or_die(const char *name, int flags)
290 int fd = open(name, flags);
292 err(1, "Failed to open %s", name);
296 /* map_zeroed_pages() takes a number of pages. */
297 static void *map_zeroed_pages(unsigned int num)
299 int fd = open_or_die("/dev/zero", O_RDONLY);
303 * We use a private mapping (ie. if we write to the page, it will be
306 addr = mmap(NULL, getpagesize() * num,
307 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
308 if (addr == MAP_FAILED)
309 err(1, "Mmapping %u pages of /dev/zero", num);
312 * One neat mmap feature is that you can close the fd, and it
320 /* Get some more pages for a device. */
321 static void *get_pages(unsigned int num)
323 void *addr = from_guest_phys(guest_limit);
325 guest_limit += num * getpagesize();
326 if (guest_limit > guest_max)
327 errx(1, "Not enough memory for devices");
332 * This routine is used to load the kernel or initrd. It tries mmap, but if
333 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
334 * it falls back to reading the memory in.
336 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
341 * We map writable even though for some segments are marked read-only.
342 * The kernel really wants to be writable: it patches its own
345 * MAP_PRIVATE means that the page won't be copied until a write is
346 * done to it. This allows us to share untouched memory between
349 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
350 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
353 /* pread does a seek and a read in one shot: saves a few lines. */
354 r = pread(fd, addr, len, offset);
356 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
360 * This routine takes an open vmlinux image, which is in ELF, and maps it into
361 * the Guest memory. ELF = Embedded Linking Format, which is the format used
362 * by all modern binaries on Linux including the kernel.
364 * The ELF headers give *two* addresses: a physical address, and a virtual
365 * address. We use the physical address; the Guest will map itself to the
368 * We return the starting address.
370 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
372 Elf32_Phdr phdr[ehdr->e_phnum];
376 * Sanity checks on the main ELF header: an x86 executable with a
377 * reasonable number of correctly-sized program headers.
379 if (ehdr->e_type != ET_EXEC
380 || ehdr->e_machine != EM_386
381 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
382 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
383 errx(1, "Malformed elf header");
386 * An ELF executable contains an ELF header and a number of "program"
387 * headers which indicate which parts ("segments") of the program to
391 /* We read in all the program headers at once: */
392 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
393 err(1, "Seeking to program headers");
394 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
395 err(1, "Reading program headers");
398 * Try all the headers: there are usually only three. A read-only one,
399 * a read-write one, and a "note" section which we don't load.
401 for (i = 0; i < ehdr->e_phnum; i++) {
402 /* If this isn't a loadable segment, we ignore it */
403 if (phdr[i].p_type != PT_LOAD)
406 verbose("Section %i: size %i addr %p\n",
407 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
409 /* We map this section of the file at its physical address. */
410 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
411 phdr[i].p_offset, phdr[i].p_filesz);
414 /* The entry point is given in the ELF header. */
415 return ehdr->e_entry;
419 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
420 * to jump into it and it will unpack itself. We used to have to perform some
421 * hairy magic because the unpacking code scared me.
423 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
424 * a small patch to jump over the tricky bits in the Guest, so now we just read
425 * the funky header so we know where in the file to load, and away we go!
427 static unsigned long load_bzimage(int fd)
429 struct boot_params boot;
431 /* Modern bzImages get loaded at 1M. */
432 void *p = from_guest_phys(0x100000);
435 * Go back to the start of the file and read the header. It should be
436 * a Linux boot header (see Documentation/x86/i386/boot.txt)
438 lseek(fd, 0, SEEK_SET);
439 read(fd, &boot, sizeof(boot));
441 /* Inside the setup_hdr, we expect the magic "HdrS" */
442 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
443 errx(1, "This doesn't look like a bzImage to me");
445 /* Skip over the extra sectors of the header. */
446 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
448 /* Now read everything into memory. in nice big chunks. */
449 while ((r = read(fd, p, 65536)) > 0)
452 /* Finally, code32_start tells us where to enter the kernel. */
453 return boot.hdr.code32_start;
457 * Loading the kernel is easy when it's a "vmlinux", but most kernels
458 * come wrapped up in the self-decompressing "bzImage" format. With a little
459 * work, we can load those, too.
461 static unsigned long load_kernel(int fd)
465 /* Read in the first few bytes. */
466 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
467 err(1, "Reading kernel");
469 /* If it's an ELF file, it starts with "\177ELF" */
470 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
471 return map_elf(fd, &hdr);
473 /* Otherwise we assume it's a bzImage, and try to load it. */
474 return load_bzimage(fd);
478 * This is a trivial little helper to align pages. Andi Kleen hated it because
479 * it calls getpagesize() twice: "it's dumb code."
481 * Kernel guys get really het up about optimization, even when it's not
482 * necessary. I leave this code as a reaction against that.
484 static inline unsigned long page_align(unsigned long addr)
486 /* Add upwards and truncate downwards. */
487 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
491 * An "initial ram disk" is a disk image loaded into memory along with the
492 * kernel which the kernel can use to boot from without needing any drivers.
493 * Most distributions now use this as standard: the initrd contains the code to
494 * load the appropriate driver modules for the current machine.
496 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
497 * kernels. He sent me this (and tells me when I break it).
499 static unsigned long load_initrd(const char *name, unsigned long mem)
505 ifd = open_or_die(name, O_RDONLY);
506 /* fstat() is needed to get the file size. */
507 if (fstat(ifd, &st) < 0)
508 err(1, "fstat() on initrd '%s'", name);
511 * We map the initrd at the top of memory, but mmap wants it to be
512 * page-aligned, so we round the size up for that.
514 len = page_align(st.st_size);
515 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
517 * Once a file is mapped, you can close the file descriptor. It's a
518 * little odd, but quite useful.
521 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
523 /* We return the initrd size. */
529 * Simple routine to roll all the commandline arguments together with spaces
532 static void concat(char *dst, char *args[])
534 unsigned int i, len = 0;
536 for (i = 0; args[i]; i++) {
538 strcat(dst+len, " ");
541 strcpy(dst+len, args[i]);
542 len += strlen(args[i]);
544 /* In case it's empty. */
549 * This is where we actually tell the kernel to initialize the Guest. We
550 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
551 * the base of Guest "physical" memory, the top physical page to allow and the
552 * entry point for the Guest.
554 static void tell_kernel(unsigned long start)
556 unsigned long args[] = { LHREQ_INITIALIZE,
557 (unsigned long)guest_base,
558 guest_limit / getpagesize(), start };
559 verbose("Guest: %p - %p (%#lx)\n",
560 guest_base, guest_base + guest_limit, guest_limit);
561 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
562 if (write(lguest_fd, args, sizeof(args)) < 0)
563 err(1, "Writing to /dev/lguest");
570 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
571 * We need to make sure it's not trying to reach into the Launcher itself, so
572 * we have a convenient routine which checks it and exits with an error message
573 * if something funny is going on:
575 static void *_check_pointer(unsigned long addr, unsigned int size,
579 * We have to separately check addr and addr+size, because size could
580 * be huge and addr + size might wrap around.
582 if (addr >= guest_limit || addr + size >= guest_limit)
583 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
585 * We return a pointer for the caller's convenience, now we know it's
588 return from_guest_phys(addr);
590 /* A macro which transparently hands the line number to the real function. */
591 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
594 * Each buffer in the virtqueues is actually a chain of descriptors. This
595 * function returns the next descriptor in the chain, or vq->vring.num if we're
598 static unsigned next_desc(struct vring_desc *desc,
599 unsigned int i, unsigned int max)
603 /* If this descriptor says it doesn't chain, we're done. */
604 if (!(desc[i].flags & VRING_DESC_F_NEXT))
607 /* Check they're not leading us off end of descriptors. */
609 /* Make sure compiler knows to grab that: we don't want it changing! */
613 errx(1, "Desc next is %u", next);
619 * This actually sends the interrupt for this virtqueue, if we've used a
622 static void trigger_irq(struct virtqueue *vq)
624 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
626 /* Don't inform them if nothing used. */
627 if (!vq->pending_used)
629 vq->pending_used = 0;
631 /* If they don't want an interrupt, don't send one... */
632 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
633 /* ... unless they've asked us to force one on empty. */
634 if (!vq->dev->irq_on_empty
635 || lg_last_avail(vq) != vq->vring.avail->idx)
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);
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.
650 * This function waits if necessary, and returns the descriptor number found.
652 static unsigned wait_for_vq_desc(struct virtqueue *vq,
654 unsigned int *out_num, unsigned int *in_num)
656 unsigned int i, head, max;
657 struct vring_desc *desc;
658 u16 last_avail = lg_last_avail(vq);
660 /* There's nothing available? */
661 while (last_avail == vq->vring.avail->idx) {
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.
670 /* OK, now we need to know about added descriptors. */
671 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
674 * They could have slipped one in as we were doing that: make
675 * sure it's written, then check again.
678 if (last_avail != vq->vring.avail->idx) {
679 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
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?");
687 /* We don't need to be notified again. */
688 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
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);
697 * Grab the next descriptor number they're advertising, and increment
698 * the index we've seen.
700 head = vq->vring.avail->ring[last_avail % vq->vring.num];
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);
707 /* When we start there are none of either input nor output. */
708 *out_num = *in_num = 0;
711 desc = vq->vring.desc;
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.
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");
722 max = desc[i].len / sizeof(struct vring_desc);
723 desc = check_pointer(desc[i].addr, desc[i].len);
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)
737 * If it's an output descriptor, they're all supposed
738 * to come before any input descriptors.
741 errx(1, "Descriptor has out after in");
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);
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.
758 static void add_used(struct virtqueue *vq, unsigned int head, int len)
760 struct vring_used_elem *used;
763 * The virtqueue contains a ring of used buffers. Get a pointer to the
764 * next entry in that used ring.
766 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
769 /* Make sure buffer is written before we update index. */
771 vq->vring.used->idx++;
775 /* And here's the combo meal deal. Supersize me! */
776 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
778 add_used(vq, head, len);
785 * We associate some data with the console for our exit hack.
787 struct console_abort {
788 /* How many times have they hit ^C? */
790 /* When did they start? */
791 struct timeval start;
794 /* This is the routine which handles console input (ie. stdin). */
795 static void console_input(struct virtqueue *vq)
798 unsigned int head, in_num, out_num;
799 struct console_abort *abort = vq->dev->priv;
800 struct iovec iov[vq->vring.num];
802 /* Make sure there's a descriptor available. */
803 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
805 errx(1, "Output buffers in console in queue?");
807 /* Read into it. This is where we usually wait. */
808 len = readv(STDIN_FILENO, iov, in_num);
810 /* Ran out of input? */
811 warnx("Failed to get console input, ignoring console.");
813 * For simplicity, dying threads kill the whole Launcher. So
820 /* Tell the Guest we used a buffer. */
821 add_used_and_trigger(vq, head, len);
824 * Three ^C within one second? Exit.
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
831 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
837 if (abort->count == 1)
838 gettimeofday(&abort->start, NULL);
839 else if (abort->count == 3) {
841 gettimeofday(&now, NULL);
842 /* Kill all Launcher processes with SIGINT, like normal ^C */
843 if (now.tv_sec <= abort->start.tv_sec+1)
849 /* This is the routine which handles console output (ie. stdout). */
850 static void console_output(struct virtqueue *vq)
852 unsigned int head, out, in;
853 struct iovec iov[vq->vring.num];
855 /* We usually wait in here, for the Guest to give us something. */
856 head = wait_for_vq_desc(vq, iov, &out, &in);
858 errx(1, "Input buffers in console output queue?");
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);
864 err(1, "Write to stdout gave %i", len);
865 iov_consume(iov, out, len);
869 * We're finished with that buffer: if we're going to sleep,
870 * wait_for_vq_desc() will prod the Guest with an interrupt.
872 add_used(vq, head, 0);
878 * Handling output for network is also simple: we get all the output buffers
879 * and write them to /dev/net/tun.
885 static void net_output(struct virtqueue *vq)
887 struct net_info *net_info = vq->dev->priv;
888 unsigned int head, out, in;
889 struct iovec iov[vq->vring.num];
891 /* We usually wait in here for the Guest to give us a packet. */
892 head = wait_for_vq_desc(vq, iov, &out, &in);
894 errx(1, "Input buffers in net output queue?");
896 * Send the whole thing through to /dev/net/tun. It expects the exact
897 * same format: what a coincidence!
899 if (writev(net_info->tunfd, iov, out) < 0)
900 errx(1, "Write to tun failed?");
903 * Done with that one; wait_for_vq_desc() will send the interrupt if
904 * all packets are processed.
906 add_used(vq, head, 0);
910 * Handling network input is a bit trickier, because I've tried to optimize it.
912 * First we have a helper routine which tells is if from this file descriptor
913 * (ie. the /dev/net/tun device) will block:
915 static bool will_block(int fd)
918 struct timeval zero = { 0, 0 };
921 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
925 * This handles packets coming in from the tun device to our Guest. Like all
926 * service routines, it gets called again as soon as it returns, so you don't
927 * see a while(1) loop here.
929 static void net_input(struct virtqueue *vq)
932 unsigned int head, out, in;
933 struct iovec iov[vq->vring.num];
934 struct net_info *net_info = vq->dev->priv;
937 * Get a descriptor to write an incoming packet into. This will also
938 * send an interrupt if they're out of descriptors.
940 head = wait_for_vq_desc(vq, iov, &out, &in);
942 errx(1, "Output buffers in net input queue?");
945 * If it looks like we'll block reading from the tun device, send them
948 if (vq->pending_used && will_block(net_info->tunfd))
952 * Read in the packet. This is where we normally wait (when there's no
953 * incoming network traffic).
955 len = readv(net_info->tunfd, iov, in);
957 err(1, "Failed to read from tun.");
960 * Mark that packet buffer as used, but don't interrupt here. We want
961 * to wait until we've done as much work as we can.
963 add_used(vq, head, len);
967 /* This is the helper to create threads: run the service routine in a loop. */
968 static int do_thread(void *_vq)
970 struct virtqueue *vq = _vq;
978 * When a child dies, we kill our entire process group with SIGTERM. This
979 * also has the side effect that the shell restores the console for us!
981 static void kill_launcher(int signal)
986 static void reset_device(struct device *dev)
988 struct virtqueue *vq;
990 verbose("Resetting device %s\n", dev->name);
992 /* Clear any features they've acked. */
993 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
995 /* We're going to be explicitly killing threads, so ignore them. */
996 signal(SIGCHLD, SIG_IGN);
998 /* Zero out the virtqueues, get rid of their threads */
999 for (vq = dev->vq; vq; vq = vq->next) {
1000 if (vq->thread != (pid_t)-1) {
1001 kill(vq->thread, SIGTERM);
1002 waitpid(vq->thread, NULL, 0);
1003 vq->thread = (pid_t)-1;
1005 memset(vq->vring.desc, 0,
1006 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1007 lg_last_avail(vq) = 0;
1009 dev->running = false;
1011 /* Now we care if threads die. */
1012 signal(SIGCHLD, (void *)kill_launcher);
1016 * This actually creates the thread which services the virtqueue for a device.
1018 static void create_thread(struct virtqueue *vq)
1021 * Create stack for thread. Since the stack grows upwards, we point
1022 * the stack pointer to the end of this region.
1024 char *stack = malloc(32768);
1025 unsigned long args[] = { LHREQ_EVENTFD,
1026 vq->config.pfn*getpagesize(), 0 };
1028 /* Create a zero-initialized eventfd. */
1029 vq->eventfd = eventfd(0, 0);
1030 if (vq->eventfd < 0)
1031 err(1, "Creating eventfd");
1032 args[2] = vq->eventfd;
1035 * Attach an eventfd to this virtqueue: it will go off when the Guest
1036 * does an LHCALL_NOTIFY for this vq.
1038 if (write(lguest_fd, &args, sizeof(args)) != 0)
1039 err(1, "Attaching eventfd");
1042 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1043 * we get a signal if it dies.
1045 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1046 if (vq->thread == (pid_t)-1)
1047 err(1, "Creating clone");
1049 /* We close our local copy now the child has it. */
1053 static bool accepted_feature(struct device *dev, unsigned int bit)
1055 const u8 *features = get_feature_bits(dev) + dev->feature_len;
1057 if (dev->feature_len < bit / CHAR_BIT)
1059 return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT));
1062 static void start_device(struct device *dev)
1065 struct virtqueue *vq;
1067 verbose("Device %s OK: offered", dev->name);
1068 for (i = 0; i < dev->feature_len; i++)
1069 verbose(" %02x", get_feature_bits(dev)[i]);
1070 verbose(", accepted");
1071 for (i = 0; i < dev->feature_len; i++)
1072 verbose(" %02x", get_feature_bits(dev)
1073 [dev->feature_len+i]);
1075 dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1077 for (vq = dev->vq; vq; vq = vq->next) {
1081 dev->running = true;
1084 static void cleanup_devices(void)
1088 for (dev = devices.dev; dev; dev = dev->next)
1091 /* If we saved off the original terminal settings, restore them now. */
1092 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1093 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1096 /* When the Guest tells us they updated the status field, we handle it. */
1097 static void update_device_status(struct device *dev)
1099 /* A zero status is a reset, otherwise it's a set of flags. */
1100 if (dev->desc->status == 0)
1102 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1103 warnx("Device %s configuration FAILED", dev->name);
1106 } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) {
1113 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1114 * particular, it's used to notify us of device status changes during boot.
1116 static void handle_output(unsigned long addr)
1120 /* Check each device. */
1121 for (i = devices.dev; i; i = i->next) {
1122 struct virtqueue *vq;
1125 * Notifications to device descriptors mean they updated the
1128 if (from_guest_phys(addr) == i->desc) {
1129 update_device_status(i);
1134 * Devices *can* be used before status is set to DRIVER_OK.
1135 * The original plan was that they would never do this: they
1136 * would always finish setting up their status bits before
1137 * actually touching the virtqueues. In practice, we allowed
1138 * them to, and they do (eg. the disk probes for partition
1139 * tables as part of initialization).
1141 * If we see this, we start the device: once it's running, we
1142 * expect the device to catch all the notifications.
1144 for (vq = i->vq; vq; vq = vq->next) {
1145 if (addr != vq->config.pfn*getpagesize())
1148 errx(1, "Notification on running %s", i->name);
1149 /* This just calls create_thread() for each virtqueue */
1156 * Early console write is done using notify on a nul-terminated string
1157 * in Guest memory. It's also great for hacking debugging messages
1160 if (addr >= guest_limit)
1161 errx(1, "Bad NOTIFY %#lx", addr);
1163 write(STDOUT_FILENO, from_guest_phys(addr),
1164 strnlen(from_guest_phys(addr), guest_limit - addr));
1170 * All devices need a descriptor so the Guest knows it exists, and a "struct
1171 * device" so the Launcher can keep track of it. We have common helper
1172 * routines to allocate and manage them.
1176 * The layout of the device page is a "struct lguest_device_desc" followed by a
1177 * number of virtqueue descriptors, then two sets of feature bits, then an
1178 * array of configuration bytes. This routine returns the configuration
1181 static u8 *device_config(const struct device *dev)
1183 return (void *)(dev->desc + 1)
1184 + dev->num_vq * sizeof(struct lguest_vqconfig)
1185 + dev->feature_len * 2;
1189 * This routine allocates a new "struct lguest_device_desc" from descriptor
1190 * table page just above the Guest's normal memory. It returns a pointer to
1193 static struct lguest_device_desc *new_dev_desc(u16 type)
1195 struct lguest_device_desc d = { .type = type };
1198 /* Figure out where the next device config is, based on the last one. */
1199 if (devices.lastdev)
1200 p = device_config(devices.lastdev)
1201 + devices.lastdev->desc->config_len;
1203 p = devices.descpage;
1205 /* We only have one page for all the descriptors. */
1206 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1207 errx(1, "Too many devices");
1209 /* p might not be aligned, so we memcpy in. */
1210 return memcpy(p, &d, sizeof(d));
1214 * Each device descriptor is followed by the description of its virtqueues. We
1215 * specify how many descriptors the virtqueue is to have.
1217 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1218 void (*service)(struct virtqueue *))
1221 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1224 /* First we need some memory for this virtqueue. */
1225 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1227 p = get_pages(pages);
1229 /* Initialize the virtqueue */
1231 vq->last_avail_idx = 0;
1235 * This is the routine the service thread will run, and its Process ID
1236 * once it's running.
1238 vq->service = service;
1239 vq->thread = (pid_t)-1;
1241 /* Initialize the configuration. */
1242 vq->config.num = num_descs;
1243 vq->config.irq = devices.next_irq++;
1244 vq->config.pfn = to_guest_phys(p) / getpagesize();
1246 /* Initialize the vring. */
1247 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1250 * Append virtqueue to this device's descriptor. We use
1251 * device_config() to get the end of the device's current virtqueues;
1252 * we check that we haven't added any config or feature information
1253 * yet, otherwise we'd be overwriting them.
1255 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1256 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1258 dev->desc->num_vq++;
1260 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1263 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1266 for (i = &dev->vq; *i; i = &(*i)->next);
1271 * The first half of the feature bitmask is for us to advertise features. The
1272 * second half is for the Guest to accept features.
1274 static void add_feature(struct device *dev, unsigned bit)
1276 u8 *features = get_feature_bits(dev);
1278 /* We can't extend the feature bits once we've added config bytes */
1279 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1280 assert(dev->desc->config_len == 0);
1281 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1284 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1288 * This routine sets the configuration fields for an existing device's
1289 * descriptor. It only works for the last device, but that's OK because that's
1292 static void set_config(struct device *dev, unsigned len, const void *conf)
1294 /* Check we haven't overflowed our single page. */
1295 if (device_config(dev) + len > devices.descpage + getpagesize())
1296 errx(1, "Too many devices");
1298 /* Copy in the config information, and store the length. */
1299 memcpy(device_config(dev), conf, len);
1300 dev->desc->config_len = len;
1302 /* Size must fit in config_len field (8 bits)! */
1303 assert(dev->desc->config_len == len);
1307 * This routine does all the creation and setup of a new device, including
1308 * calling new_dev_desc() to allocate the descriptor and device memory. We
1309 * don't actually start the service threads until later.
1311 * See what I mean about userspace being boring?
1313 static struct device *new_device(const char *name, u16 type)
1315 struct device *dev = malloc(sizeof(*dev));
1317 /* Now we populate the fields one at a time. */
1318 dev->desc = new_dev_desc(type);
1321 dev->feature_len = 0;
1323 dev->running = false;
1326 * Append to device list. Prepending to a single-linked list is
1327 * easier, but the user expects the devices to be arranged on the bus
1328 * in command-line order. The first network device on the command line
1329 * is eth0, the first block device /dev/vda, etc.
1331 if (devices.lastdev)
1332 devices.lastdev->next = dev;
1335 devices.lastdev = dev;
1341 * Our first setup routine is the console. It's a fairly simple device, but
1342 * UNIX tty handling makes it uglier than it could be.
1344 static void setup_console(void)
1348 /* If we can save the initial standard input settings... */
1349 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1350 struct termios term = orig_term;
1352 * Then we turn off echo, line buffering and ^C etc: We want a
1353 * raw input stream to the Guest.
1355 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1356 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1359 dev = new_device("console", VIRTIO_ID_CONSOLE);
1361 /* We store the console state in dev->priv, and initialize it. */
1362 dev->priv = malloc(sizeof(struct console_abort));
1363 ((struct console_abort *)dev->priv)->count = 0;
1366 * The console needs two virtqueues: the input then the output. When
1367 * they put something the input queue, we make sure we're listening to
1368 * stdin. When they put something in the output queue, we write it to
1371 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1372 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1374 verbose("device %u: console\n", ++devices.device_num);
1379 * Inter-guest networking is an interesting area. Simplest is to have a
1380 * --sharenet=<name> option which opens or creates a named pipe. This can be
1381 * used to send packets to another guest in a 1:1 manner.
1383 * More sopisticated is to use one of the tools developed for project like UML
1386 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1387 * completely generic ("here's my vring, attach to your vring") and would work
1388 * for any traffic. Of course, namespace and permissions issues need to be
1389 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1390 * multiple inter-guest channels behind one interface, although it would
1391 * require some manner of hotplugging new virtio channels.
1393 * Finally, we could implement a virtio network switch in the kernel.
1396 static u32 str2ip(const char *ipaddr)
1400 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1401 errx(1, "Failed to parse IP address '%s'", ipaddr);
1402 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1405 static void str2mac(const char *macaddr, unsigned char mac[6])
1408 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1409 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1410 errx(1, "Failed to parse mac address '%s'", macaddr);
1420 * This code is "adapted" from libbridge: it attaches the Host end of the
1421 * network device to the bridge device specified by the command line.
1423 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1424 * dislike bridging), and I just try not to break it.
1426 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1432 errx(1, "must specify bridge name");
1434 ifidx = if_nametoindex(if_name);
1436 errx(1, "interface %s does not exist!", if_name);
1438 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1439 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1440 ifr.ifr_ifindex = ifidx;
1441 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1442 err(1, "can't add %s to bridge %s", if_name, br_name);
1446 * This sets up the Host end of the network device with an IP address, brings
1447 * it up so packets will flow, the copies the MAC address into the hwaddr
1450 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1453 struct sockaddr_in sin;
1455 memset(&ifr, 0, sizeof(ifr));
1456 strcpy(ifr.ifr_name, tapif);
1458 /* Don't read these incantations. Just cut & paste them like I did! */
1459 sin.sin_family = AF_INET;
1460 sin.sin_addr.s_addr = htonl(ipaddr);
1461 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1462 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1463 err(1, "Setting %s interface address", tapif);
1464 ifr.ifr_flags = IFF_UP;
1465 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1466 err(1, "Bringing interface %s up", tapif);
1469 static int get_tun_device(char tapif[IFNAMSIZ])
1474 /* Start with this zeroed. Messy but sure. */
1475 memset(&ifr, 0, sizeof(ifr));
1478 * We open the /dev/net/tun device and tell it we want a tap device. A
1479 * tap device is like a tun device, only somehow different. To tell
1480 * the truth, I completely blundered my way through this code, but it
1483 netfd = open_or_die("/dev/net/tun", O_RDWR);
1484 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1485 strcpy(ifr.ifr_name, "tap%d");
1486 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1487 err(1, "configuring /dev/net/tun");
1489 if (ioctl(netfd, TUNSETOFFLOAD,
1490 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1491 err(1, "Could not set features for tun device");
1494 * We don't need checksums calculated for packets coming in this
1497 ioctl(netfd, TUNSETNOCSUM, 1);
1499 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1504 * Our network is a Host<->Guest network. This can either use bridging or
1505 * routing, but the principle is the same: it uses the "tun" device to inject
1506 * packets into the Host as if they came in from a normal network card. We
1507 * just shunt packets between the Guest and the tun device.
1509 static void setup_tun_net(char *arg)
1512 struct net_info *net_info = malloc(sizeof(*net_info));
1514 u32 ip = INADDR_ANY;
1515 bool bridging = false;
1516 char tapif[IFNAMSIZ], *p;
1517 struct virtio_net_config conf;
1519 net_info->tunfd = get_tun_device(tapif);
1521 /* First we create a new network device. */
1522 dev = new_device("net", VIRTIO_ID_NET);
1523 dev->priv = net_info;
1525 /* Network devices need a recv and a send queue, just like console. */
1526 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1527 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1530 * We need a socket to perform the magic network ioctls to bring up the
1531 * tap interface, connect to the bridge etc. Any socket will do!
1533 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1535 err(1, "opening IP socket");
1537 /* If the command line was --tunnet=bridge:<name> do bridging. */
1538 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1539 arg += strlen(BRIDGE_PFX);
1543 /* A mac address may follow the bridge name or IP address */
1544 p = strchr(arg, ':');
1546 str2mac(p+1, conf.mac);
1547 add_feature(dev, VIRTIO_NET_F_MAC);
1551 /* arg is now either an IP address or a bridge name */
1553 add_to_bridge(ipfd, tapif, arg);
1557 /* Set up the tun device. */
1558 configure_device(ipfd, tapif, ip);
1560 add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY);
1561 /* Expect Guest to handle everything except UFO */
1562 add_feature(dev, VIRTIO_NET_F_CSUM);
1563 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1564 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1565 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1566 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1567 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1568 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1569 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1570 /* We handle indirect ring entries */
1571 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1572 set_config(dev, sizeof(conf), &conf);
1574 /* We don't need the socket any more; setup is done. */
1577 devices.device_num++;
1580 verbose("device %u: tun %s attached to bridge: %s\n",
1581 devices.device_num, tapif, arg);
1583 verbose("device %u: tun %s: %s\n",
1584 devices.device_num, tapif, arg);
1588 /* This hangs off device->priv. */
1590 /* The size of the file. */
1593 /* The file descriptor for the file. */
1601 * The disk only has one virtqueue, so it only has one thread. It is really
1602 * simple: the Guest asks for a block number and we read or write that position
1605 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1606 * slow: the Guest waits until the read is finished before running anything
1607 * else, even if it could have been doing useful work.
1609 * We could have used async I/O, except it's reputed to suck so hard that
1610 * characters actually go missing from your code when you try to use it.
1612 static void blk_request(struct virtqueue *vq)
1614 struct vblk_info *vblk = vq->dev->priv;
1615 unsigned int head, out_num, in_num, wlen;
1618 struct virtio_blk_outhdr *out;
1619 struct iovec iov[vq->vring.num];
1623 * Get the next request, where we normally wait. It triggers the
1624 * interrupt to acknowledge previously serviced requests (if any).
1626 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1629 * Every block request should contain at least one output buffer
1630 * (detailing the location on disk and the type of request) and one
1631 * input buffer (to hold the result).
1633 if (out_num == 0 || in_num == 0)
1634 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1635 head, out_num, in_num);
1637 out = convert(&iov[0], struct virtio_blk_outhdr);
1638 in = convert(&iov[out_num+in_num-1], u8);
1640 * For historical reasons, block operations are expressed in 512 byte
1643 off = out->sector * 512;
1646 * In general the virtio block driver is allowed to try SCSI commands.
1647 * It'd be nice if we supported eject, for example, but we don't.
1649 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1650 fprintf(stderr, "Scsi commands unsupported\n");
1651 *in = VIRTIO_BLK_S_UNSUPP;
1653 } else if (out->type & VIRTIO_BLK_T_OUT) {
1657 * Move to the right location in the block file. This can fail
1658 * if they try to write past end.
1660 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1661 err(1, "Bad seek to sector %llu", out->sector);
1663 ret = writev(vblk->fd, iov+1, out_num-1);
1664 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1667 * Grr... Now we know how long the descriptor they sent was, we
1668 * make sure they didn't try to write over the end of the block
1669 * file (possibly extending it).
1671 if (ret > 0 && off + ret > vblk->len) {
1672 /* Trim it back to the correct length */
1673 ftruncate64(vblk->fd, vblk->len);
1674 /* Die, bad Guest, die. */
1675 errx(1, "Write past end %llu+%u", off, ret);
1679 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1680 } else if (out->type & VIRTIO_BLK_T_FLUSH) {
1682 ret = fdatasync(vblk->fd);
1683 verbose("FLUSH fdatasync: %i\n", ret);
1685 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1690 * Move to the right location in the block file. This can fail
1691 * if they try to read past end.
1693 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1694 err(1, "Bad seek to sector %llu", out->sector);
1696 ret = readv(vblk->fd, iov+1, in_num-1);
1697 verbose("READ from sector %llu: %i\n", out->sector, ret);
1699 wlen = sizeof(*in) + ret;
1700 *in = VIRTIO_BLK_S_OK;
1703 *in = VIRTIO_BLK_S_IOERR;
1707 /* Finished that request. */
1708 add_used(vq, head, wlen);
1711 /*L:198 This actually sets up a virtual block device. */
1712 static void setup_block_file(const char *filename)
1715 struct vblk_info *vblk;
1716 struct virtio_blk_config conf;
1718 /* Creat the device. */
1719 dev = new_device("block", VIRTIO_ID_BLOCK);
1721 /* The device has one virtqueue, where the Guest places requests. */
1722 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1724 /* Allocate the room for our own bookkeeping */
1725 vblk = dev->priv = malloc(sizeof(*vblk));
1727 /* First we open the file and store the length. */
1728 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1729 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1731 /* We support FLUSH. */
1732 add_feature(dev, VIRTIO_BLK_F_FLUSH);
1734 /* Tell Guest how many sectors this device has. */
1735 conf.capacity = cpu_to_le64(vblk->len / 512);
1738 * Tell Guest not to put in too many descriptors at once: two are used
1739 * for the in and out elements.
1741 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1742 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1744 /* Don't try to put whole struct: we have 8 bit limit. */
1745 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1747 verbose("device %u: virtblock %llu sectors\n",
1748 ++devices.device_num, le64_to_cpu(conf.capacity));
1752 * Our random number generator device reads from /dev/random into the Guest's
1753 * input buffers. The usual case is that the Guest doesn't want random numbers
1754 * and so has no buffers although /dev/random is still readable, whereas
1755 * console is the reverse.
1757 * The same logic applies, however.
1763 static void rng_input(struct virtqueue *vq)
1766 unsigned int head, in_num, out_num, totlen = 0;
1767 struct rng_info *rng_info = vq->dev->priv;
1768 struct iovec iov[vq->vring.num];
1770 /* First we need a buffer from the Guests's virtqueue. */
1771 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1773 errx(1, "Output buffers in rng?");
1776 * Just like the console write, we loop to cover the whole iovec.
1777 * In this case, short reads actually happen quite a bit.
1779 while (!iov_empty(iov, in_num)) {
1780 len = readv(rng_info->rfd, iov, in_num);
1782 err(1, "Read from /dev/random gave %i", len);
1783 iov_consume(iov, in_num, len);
1787 /* Tell the Guest about the new input. */
1788 add_used(vq, head, totlen);
1792 * This creates a "hardware" random number device for the Guest.
1794 static void setup_rng(void)
1797 struct rng_info *rng_info = malloc(sizeof(*rng_info));
1799 /* Our device's privat info simply contains the /dev/random fd. */
1800 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1802 /* Create the new device. */
1803 dev = new_device("rng", VIRTIO_ID_RNG);
1804 dev->priv = rng_info;
1806 /* The device has one virtqueue, where the Guest places inbufs. */
1807 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1809 verbose("device %u: rng\n", devices.device_num++);
1811 /* That's the end of device setup. */
1813 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1814 static void __attribute__((noreturn)) restart_guest(void)
1819 * Since we don't track all open fds, we simply close everything beyond
1822 for (i = 3; i < FD_SETSIZE; i++)
1825 /* Reset all the devices (kills all threads). */
1828 execv(main_args[0], main_args);
1829 err(1, "Could not exec %s", main_args[0]);
1833 * Finally we reach the core of the Launcher which runs the Guest, serves
1834 * its input and output, and finally, lays it to rest.
1836 static void __attribute__((noreturn)) run_guest(void)
1839 unsigned long notify_addr;
1842 /* We read from the /dev/lguest device to run the Guest. */
1843 readval = pread(lguest_fd, ¬ify_addr,
1844 sizeof(notify_addr), cpu_id);
1846 /* One unsigned long means the Guest did HCALL_NOTIFY */
1847 if (readval == sizeof(notify_addr)) {
1848 verbose("Notify on address %#lx\n", notify_addr);
1849 handle_output(notify_addr);
1850 /* ENOENT means the Guest died. Reading tells us why. */
1851 } else if (errno == ENOENT) {
1852 char reason[1024] = { 0 };
1853 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1854 errx(1, "%s", reason);
1855 /* ERESTART means that we need to reboot the guest */
1856 } else if (errno == ERESTART) {
1858 /* Anything else means a bug or incompatible change. */
1860 err(1, "Running guest failed");
1864 * This is the end of the Launcher. The good news: we are over halfway
1865 * through! The bad news: the most fiendish part of the code still lies ahead
1868 * Are you ready? Take a deep breath and join me in the core of the Host, in
1872 static struct option opts[] = {
1873 { "verbose", 0, NULL, 'v' },
1874 { "tunnet", 1, NULL, 't' },
1875 { "block", 1, NULL, 'b' },
1876 { "rng", 0, NULL, 'r' },
1877 { "initrd", 1, NULL, 'i' },
1878 { "username", 1, NULL, 'u' },
1879 { "chroot", 1, NULL, 'c' },
1882 static void usage(void)
1884 errx(1, "Usage: lguest [--verbose] "
1885 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1886 "|--block=<filename>|--initrd=<filename>]...\n"
1887 "<mem-in-mb> vmlinux [args...]");
1890 /*L:105 The main routine is where the real work begins: */
1891 int main(int argc, char *argv[])
1893 /* Memory, code startpoint and size of the (optional) initrd. */
1894 unsigned long mem = 0, start, initrd_size = 0;
1895 /* Two temporaries. */
1897 /* The boot information for the Guest. */
1898 struct boot_params *boot;
1899 /* If they specify an initrd file to load. */
1900 const char *initrd_name = NULL;
1902 /* Password structure for initgroups/setres[gu]id */
1903 struct passwd *user_details = NULL;
1905 /* Directory to chroot to */
1906 char *chroot_path = NULL;
1908 /* Save the args: we "reboot" by execing ourselves again. */
1912 * First we initialize the device list. We keep a pointer to the last
1913 * device, and the next interrupt number to use for devices (1:
1914 * remember that 0 is used by the timer).
1916 devices.lastdev = NULL;
1917 devices.next_irq = 1;
1919 /* We're CPU 0. In fact, that's the only CPU possible right now. */
1923 * We need to know how much memory so we can set up the device
1924 * descriptor and memory pages for the devices as we parse the command
1925 * line. So we quickly look through the arguments to find the amount
1928 for (i = 1; i < argc; i++) {
1929 if (argv[i][0] != '-') {
1930 mem = atoi(argv[i]) * 1024 * 1024;
1932 * We start by mapping anonymous pages over all of
1933 * guest-physical memory range. This fills it with 0,
1934 * and ensures that the Guest won't be killed when it
1935 * tries to access it.
1937 guest_base = map_zeroed_pages(mem / getpagesize()
1940 guest_max = mem + DEVICE_PAGES*getpagesize();
1941 devices.descpage = get_pages(1);
1946 /* The options are fairly straight-forward */
1947 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
1953 setup_tun_net(optarg);
1956 setup_block_file(optarg);
1962 initrd_name = optarg;
1965 user_details = getpwnam(optarg);
1967 err(1, "getpwnam failed, incorrect username?");
1970 chroot_path = optarg;
1973 warnx("Unknown argument %s", argv[optind]);
1978 * After the other arguments we expect memory and kernel image name,
1979 * followed by command line arguments for the kernel.
1981 if (optind + 2 > argc)
1984 verbose("Guest base is at %p\n", guest_base);
1986 /* We always have a console device */
1989 /* Now we load the kernel */
1990 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
1992 /* Boot information is stashed at physical address 0 */
1993 boot = from_guest_phys(0);
1995 /* Map the initrd image if requested (at top of physical memory) */
1997 initrd_size = load_initrd(initrd_name, mem);
1999 * These are the location in the Linux boot header where the
2000 * start and size of the initrd are expected to be found.
2002 boot->hdr.ramdisk_image = mem - initrd_size;
2003 boot->hdr.ramdisk_size = initrd_size;
2004 /* The bootloader type 0xFF means "unknown"; that's OK. */
2005 boot->hdr.type_of_loader = 0xFF;
2009 * The Linux boot header contains an "E820" memory map: ours is a
2010 * simple, single region.
2012 boot->e820_entries = 1;
2013 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
2015 * The boot header contains a command line pointer: we put the command
2016 * line after the boot header.
2018 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2019 /* We use a simple helper to copy the arguments separated by spaces. */
2020 concat((char *)(boot + 1), argv+optind+2);
2022 /* Boot protocol version: 2.07 supports the fields for lguest. */
2023 boot->hdr.version = 0x207;
2025 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2026 boot->hdr.hardware_subarch = 1;
2028 /* Tell the entry path not to try to reload segment registers. */
2029 boot->hdr.loadflags |= KEEP_SEGMENTS;
2032 * We tell the kernel to initialize the Guest: this returns the open
2033 * /dev/lguest file descriptor.
2037 /* Ensure that we terminate if a device-servicing child dies. */
2038 signal(SIGCHLD, kill_launcher);
2040 /* If we exit via err(), this kills all the threads, restores tty. */
2041 atexit(cleanup_devices);
2043 /* If requested, chroot to a directory */
2045 if (chroot(chroot_path) != 0)
2046 err(1, "chroot(\"%s\") failed", chroot_path);
2048 if (chdir("/") != 0)
2049 err(1, "chdir(\"/\") failed");
2051 verbose("chroot done\n");
2054 /* If requested, drop privileges */
2059 u = user_details->pw_uid;
2060 g = user_details->pw_gid;
2062 if (initgroups(user_details->pw_name, g) != 0)
2063 err(1, "initgroups failed");
2065 if (setresgid(g, g, g) != 0)
2066 err(1, "setresgid failed");
2068 if (setresuid(u, u, u) != 0)
2069 err(1, "setresuid failed");
2071 verbose("Dropping privileges completed\n");
2074 /* Finally, run the Guest. This doesn't return. */
2080 * Mastery is done: you now know everything I do.
2082 * But surely you have seen code, features and bugs in your wanderings which
2083 * you now yearn to attack? That is the real game, and I look forward to you
2084 * patching and forking lguest into the Your-Name-Here-visor.
2086 * Farewell, and good coding!