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>
46 #ifndef VIRTIO_F_ANY_LAYOUT
47 #define VIRTIO_F_ANY_LAYOUT 27
51 * We can ignore the 43 include files we need for this program, but I do want
52 * to draw attention to the use of kernel-style types.
54 * As Linus said, "C is a Spartan language, and so should your naming be." I
55 * like these abbreviations, so we define them here. Note that u64 is always
56 * unsigned long long, which works on all Linux systems: this means that we can
57 * use %llu in printf for any u64.
59 typedef unsigned long long u64;
65 #include <linux/virtio_config.h>
66 #include <linux/virtio_net.h>
67 #include <linux/virtio_blk.h>
68 #include <linux/virtio_console.h>
69 #include <linux/virtio_rng.h>
70 #include <linux/virtio_ring.h>
71 #include <asm/bootparam.h>
72 #include "../../include/linux/lguest_launcher.h"
74 #define BRIDGE_PFX "bridge:"
76 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
78 /* We can have up to 256 pages for devices. */
79 #define DEVICE_PAGES 256
80 /* This will occupy 3 pages: it must be a power of 2. */
81 #define VIRTQUEUE_NUM 256
84 * verbose is both a global flag and a macro. The C preprocessor allows
85 * this, and although I wouldn't recommend it, it works quite nicely here.
88 #define verbose(args...) \
89 do { if (verbose) printf(args); } while(0)
92 /* The pointer to the start of guest memory. */
93 static void *guest_base;
94 /* The maximum guest physical address allowed, and maximum possible. */
95 static unsigned long guest_limit, guest_max;
96 /* The /dev/lguest file descriptor. */
99 /* a per-cpu variable indicating whose vcpu is currently running */
100 static unsigned int __thread cpu_id;
102 /* This is our list of devices. */
104 /* Counter to assign interrupt numbers. */
105 unsigned int next_irq;
107 /* Counter to print out convenient device numbers. */
108 unsigned int device_num;
110 /* The descriptor page for the devices. */
113 /* A single linked list of devices. */
115 /* And a pointer to the last device for easy append. */
116 struct device *lastdev;
119 /* The list of Guest devices, based on command line arguments. */
120 static struct device_list devices;
122 /* The device structure describes a single device. */
124 /* The linked-list pointer. */
127 /* The device's descriptor, as mapped into the Guest. */
128 struct lguest_device_desc *desc;
130 /* We can't trust desc values once Guest has booted: we use these. */
131 unsigned int feature_len;
134 /* The name of this device, for --verbose. */
137 /* Any queues attached to this device */
138 struct virtqueue *vq;
140 /* Is it operational */
143 /* Device-specific data. */
147 /* The virtqueue structure describes a queue attached to a device. */
149 struct virtqueue *next;
151 /* Which device owns me. */
154 /* The configuration for this queue. */
155 struct lguest_vqconfig config;
157 /* The actual ring of buffers. */
160 /* Last available index we saw. */
163 /* How many are used since we sent last irq? */
164 unsigned int pending_used;
166 /* Eventfd where Guest notifications arrive. */
169 /* Function for the thread which is servicing this virtqueue. */
170 void (*service)(struct virtqueue *vq);
174 /* Remember the arguments to the program so we can "reboot" */
175 static char **main_args;
177 /* The original tty settings to restore on exit. */
178 static struct termios orig_term;
181 * We have to be careful with barriers: our devices are all run in separate
182 * threads and so we need to make sure that changes visible to the Guest happen
185 #define wmb() __asm__ __volatile__("" : : : "memory")
186 #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
187 #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
189 /* Wrapper for the last available index. Makes it easier to change. */
190 #define lg_last_avail(vq) ((vq)->last_avail_idx)
193 * The virtio configuration space is defined to be little-endian. x86 is
194 * little-endian too, but it's nice to be explicit so we have these helpers.
196 #define cpu_to_le16(v16) (v16)
197 #define cpu_to_le32(v32) (v32)
198 #define cpu_to_le64(v64) (v64)
199 #define le16_to_cpu(v16) (v16)
200 #define le32_to_cpu(v32) (v32)
201 #define le64_to_cpu(v64) (v64)
203 /* Is this iovec empty? */
204 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
208 for (i = 0; i < num_iov; i++)
214 /* Take len bytes from the front of this iovec. */
215 static void iov_consume(struct iovec iov[], unsigned num_iov,
216 void *dest, unsigned len)
220 for (i = 0; i < num_iov; i++) {
223 used = iov[i].iov_len < len ? iov[i].iov_len : len;
225 memcpy(dest, iov[i].iov_base, used);
228 iov[i].iov_base += used;
229 iov[i].iov_len -= used;
233 errx(1, "iovec too short!");
236 /* The device virtqueue descriptors are followed by feature bitmasks. */
237 static u8 *get_feature_bits(struct device *dev)
239 return (u8 *)(dev->desc + 1)
240 + dev->num_vq * sizeof(struct lguest_vqconfig);
244 * The Launcher code itself takes us out into userspace, that scary place where
245 * pointers run wild and free! Unfortunately, like most userspace programs,
246 * it's quite boring (which is why everyone likes to hack on the kernel!).
247 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
248 * you through this section. Or, maybe not.
250 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
251 * memory and stores it in "guest_base". In other words, Guest physical ==
252 * Launcher virtual with an offset.
254 * This can be tough to get your head around, but usually it just means that we
255 * use these trivial conversion functions when the Guest gives us its
256 * "physical" addresses:
258 static void *from_guest_phys(unsigned long addr)
260 return guest_base + addr;
263 static unsigned long to_guest_phys(const void *addr)
265 return (addr - guest_base);
269 * Loading the Kernel.
271 * We start with couple of simple helper routines. open_or_die() avoids
272 * error-checking code cluttering the callers:
274 static int open_or_die(const char *name, int flags)
276 int fd = open(name, flags);
278 err(1, "Failed to open %s", name);
282 /* map_zeroed_pages() takes a number of pages. */
283 static void *map_zeroed_pages(unsigned int num)
285 int fd = open_or_die("/dev/zero", O_RDONLY);
289 * We use a private mapping (ie. if we write to the page, it will be
290 * copied). We allocate an extra two pages PROT_NONE to act as guard
291 * pages against read/write attempts that exceed allocated space.
293 addr = mmap(NULL, getpagesize() * (num+2),
294 PROT_NONE, MAP_PRIVATE, fd, 0);
296 if (addr == MAP_FAILED)
297 err(1, "Mmapping %u pages of /dev/zero", num);
299 if (mprotect(addr + getpagesize(), getpagesize() * num,
300 PROT_READ|PROT_WRITE) == -1)
301 err(1, "mprotect rw %u pages failed", num);
304 * One neat mmap feature is that you can close the fd, and it
309 /* Return address after PROT_NONE page */
310 return addr + getpagesize();
313 /* Get some more pages for a device. */
314 static void *get_pages(unsigned int num)
316 void *addr = from_guest_phys(guest_limit);
318 guest_limit += num * getpagesize();
319 if (guest_limit > guest_max)
320 errx(1, "Not enough memory for devices");
325 * This routine is used to load the kernel or initrd. It tries mmap, but if
326 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
327 * it falls back to reading the memory in.
329 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
334 * We map writable even though for some segments are marked read-only.
335 * The kernel really wants to be writable: it patches its own
338 * MAP_PRIVATE means that the page won't be copied until a write is
339 * done to it. This allows us to share untouched memory between
342 if (mmap(addr, len, PROT_READ|PROT_WRITE,
343 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
346 /* pread does a seek and a read in one shot: saves a few lines. */
347 r = pread(fd, addr, len, offset);
349 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
353 * This routine takes an open vmlinux image, which is in ELF, and maps it into
354 * the Guest memory. ELF = Embedded Linking Format, which is the format used
355 * by all modern binaries on Linux including the kernel.
357 * The ELF headers give *two* addresses: a physical address, and a virtual
358 * address. We use the physical address; the Guest will map itself to the
361 * We return the starting address.
363 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
365 Elf32_Phdr phdr[ehdr->e_phnum];
369 * Sanity checks on the main ELF header: an x86 executable with a
370 * reasonable number of correctly-sized program headers.
372 if (ehdr->e_type != ET_EXEC
373 || ehdr->e_machine != EM_386
374 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
375 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
376 errx(1, "Malformed elf header");
379 * An ELF executable contains an ELF header and a number of "program"
380 * headers which indicate which parts ("segments") of the program to
384 /* We read in all the program headers at once: */
385 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
386 err(1, "Seeking to program headers");
387 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
388 err(1, "Reading program headers");
391 * Try all the headers: there are usually only three. A read-only one,
392 * a read-write one, and a "note" section which we don't load.
394 for (i = 0; i < ehdr->e_phnum; i++) {
395 /* If this isn't a loadable segment, we ignore it */
396 if (phdr[i].p_type != PT_LOAD)
399 verbose("Section %i: size %i addr %p\n",
400 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
402 /* We map this section of the file at its physical address. */
403 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
404 phdr[i].p_offset, phdr[i].p_filesz);
407 /* The entry point is given in the ELF header. */
408 return ehdr->e_entry;
412 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
413 * to jump into it and it will unpack itself. We used to have to perform some
414 * hairy magic because the unpacking code scared me.
416 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
417 * a small patch to jump over the tricky bits in the Guest, so now we just read
418 * the funky header so we know where in the file to load, and away we go!
420 static unsigned long load_bzimage(int fd)
422 struct boot_params boot;
424 /* Modern bzImages get loaded at 1M. */
425 void *p = from_guest_phys(0x100000);
428 * Go back to the start of the file and read the header. It should be
429 * a Linux boot header (see Documentation/x86/boot.txt)
431 lseek(fd, 0, SEEK_SET);
432 read(fd, &boot, sizeof(boot));
434 /* Inside the setup_hdr, we expect the magic "HdrS" */
435 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
436 errx(1, "This doesn't look like a bzImage to me");
438 /* Skip over the extra sectors of the header. */
439 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
441 /* Now read everything into memory. in nice big chunks. */
442 while ((r = read(fd, p, 65536)) > 0)
445 /* Finally, code32_start tells us where to enter the kernel. */
446 return boot.hdr.code32_start;
450 * Loading the kernel is easy when it's a "vmlinux", but most kernels
451 * come wrapped up in the self-decompressing "bzImage" format. With a little
452 * work, we can load those, too.
454 static unsigned long load_kernel(int fd)
458 /* Read in the first few bytes. */
459 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
460 err(1, "Reading kernel");
462 /* If it's an ELF file, it starts with "\177ELF" */
463 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
464 return map_elf(fd, &hdr);
466 /* Otherwise we assume it's a bzImage, and try to load it. */
467 return load_bzimage(fd);
471 * This is a trivial little helper to align pages. Andi Kleen hated it because
472 * it calls getpagesize() twice: "it's dumb code."
474 * Kernel guys get really het up about optimization, even when it's not
475 * necessary. I leave this code as a reaction against that.
477 static inline unsigned long page_align(unsigned long addr)
479 /* Add upwards and truncate downwards. */
480 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
484 * An "initial ram disk" is a disk image loaded into memory along with the
485 * kernel which the kernel can use to boot from without needing any drivers.
486 * Most distributions now use this as standard: the initrd contains the code to
487 * load the appropriate driver modules for the current machine.
489 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
490 * kernels. He sent me this (and tells me when I break it).
492 static unsigned long load_initrd(const char *name, unsigned long mem)
498 ifd = open_or_die(name, O_RDONLY);
499 /* fstat() is needed to get the file size. */
500 if (fstat(ifd, &st) < 0)
501 err(1, "fstat() on initrd '%s'", name);
504 * We map the initrd at the top of memory, but mmap wants it to be
505 * page-aligned, so we round the size up for that.
507 len = page_align(st.st_size);
508 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
510 * Once a file is mapped, you can close the file descriptor. It's a
511 * little odd, but quite useful.
514 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
516 /* We return the initrd size. */
522 * Simple routine to roll all the commandline arguments together with spaces
525 static void concat(char *dst, char *args[])
527 unsigned int i, len = 0;
529 for (i = 0; args[i]; i++) {
531 strcat(dst+len, " ");
534 strcpy(dst+len, args[i]);
535 len += strlen(args[i]);
537 /* In case it's empty. */
542 * This is where we actually tell the kernel to initialize the Guest. We
543 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
544 * the base of Guest "physical" memory, the top physical page to allow and the
545 * entry point for the Guest.
547 static void tell_kernel(unsigned long start)
549 unsigned long args[] = { LHREQ_INITIALIZE,
550 (unsigned long)guest_base,
551 guest_limit / getpagesize(), start };
552 verbose("Guest: %p - %p (%#lx)\n",
553 guest_base, guest_base + guest_limit, guest_limit);
554 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
555 if (write(lguest_fd, args, sizeof(args)) < 0)
556 err(1, "Writing to /dev/lguest");
563 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
564 * We need to make sure it's not trying to reach into the Launcher itself, so
565 * we have a convenient routine which checks it and exits with an error message
566 * if something funny is going on:
568 static void *_check_pointer(unsigned long addr, unsigned int size,
572 * Check if the requested address and size exceeds the allocated memory,
573 * or addr + size wraps around.
575 if ((addr + size) > guest_limit || (addr + size) < addr)
576 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
578 * We return a pointer for the caller's convenience, now we know it's
581 return from_guest_phys(addr);
583 /* A macro which transparently hands the line number to the real function. */
584 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
587 * Each buffer in the virtqueues is actually a chain of descriptors. This
588 * function returns the next descriptor in the chain, or vq->vring.num if we're
591 static unsigned next_desc(struct vring_desc *desc,
592 unsigned int i, unsigned int max)
596 /* If this descriptor says it doesn't chain, we're done. */
597 if (!(desc[i].flags & VRING_DESC_F_NEXT))
600 /* Check they're not leading us off end of descriptors. */
602 /* Make sure compiler knows to grab that: we don't want it changing! */
606 errx(1, "Desc next is %u", next);
612 * This actually sends the interrupt for this virtqueue, if we've used a
615 static void trigger_irq(struct virtqueue *vq)
617 unsigned long buf[] = { LHREQ_IRQ, vq->config.irq };
619 /* Don't inform them if nothing used. */
620 if (!vq->pending_used)
622 vq->pending_used = 0;
624 /* If they don't want an interrupt, don't send one... */
625 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
629 /* Send the Guest an interrupt tell them we used something up. */
630 if (write(lguest_fd, buf, sizeof(buf)) != 0)
631 err(1, "Triggering irq %i", vq->config.irq);
635 * This looks in the virtqueue for the first available buffer, and converts
636 * it to an iovec for convenient access. Since descriptors consist of some
637 * number of output then some number of input descriptors, it's actually two
638 * iovecs, but we pack them into one and note how many of each there were.
640 * This function waits if necessary, and returns the descriptor number found.
642 static unsigned wait_for_vq_desc(struct virtqueue *vq,
644 unsigned int *out_num, unsigned int *in_num)
646 unsigned int i, head, max;
647 struct vring_desc *desc;
648 u16 last_avail = lg_last_avail(vq);
650 /* There's nothing available? */
651 while (last_avail == vq->vring.avail->idx) {
655 * Since we're about to sleep, now is a good time to tell the
656 * Guest about what we've used up to now.
660 /* OK, now we need to know about added descriptors. */
661 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
664 * They could have slipped one in as we were doing that: make
665 * sure it's written, then check again.
668 if (last_avail != vq->vring.avail->idx) {
669 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
673 /* Nothing new? Wait for eventfd to tell us they refilled. */
674 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
675 errx(1, "Event read failed?");
677 /* We don't need to be notified again. */
678 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
681 /* Check it isn't doing very strange things with descriptor numbers. */
682 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
683 errx(1, "Guest moved used index from %u to %u",
684 last_avail, vq->vring.avail->idx);
687 * Make sure we read the descriptor number *after* we read the ring
688 * update; don't let the cpu or compiler change the order.
693 * Grab the next descriptor number they're advertising, and increment
694 * the index we've seen.
696 head = vq->vring.avail->ring[last_avail % vq->vring.num];
699 /* If their number is silly, that's a fatal mistake. */
700 if (head >= vq->vring.num)
701 errx(1, "Guest says index %u is available", head);
703 /* When we start there are none of either input nor output. */
704 *out_num = *in_num = 0;
707 desc = vq->vring.desc;
711 * We have to read the descriptor after we read the descriptor number,
712 * but there's a data dependency there so the CPU shouldn't reorder
713 * that: no rmb() required.
717 * If this is an indirect entry, then this buffer contains a descriptor
718 * table which we handle as if it's any normal descriptor chain.
720 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
721 if (desc[i].len % sizeof(struct vring_desc))
722 errx(1, "Invalid size for indirect buffer table");
724 max = desc[i].len / sizeof(struct vring_desc);
725 desc = check_pointer(desc[i].addr, desc[i].len);
730 /* Grab the first descriptor, and check it's OK. */
731 iov[*out_num + *in_num].iov_len = desc[i].len;
732 iov[*out_num + *in_num].iov_base
733 = check_pointer(desc[i].addr, desc[i].len);
734 /* If this is an input descriptor, increment that count. */
735 if (desc[i].flags & VRING_DESC_F_WRITE)
739 * If it's an output descriptor, they're all supposed
740 * to come before any input descriptors.
743 errx(1, "Descriptor has out after in");
747 /* If we've got too many, that implies a descriptor loop. */
748 if (*out_num + *in_num > max)
749 errx(1, "Looped descriptor");
750 } while ((i = next_desc(desc, i, max)) != max);
756 * After we've used one of their buffers, we tell the Guest about it. Sometime
757 * later we'll want to send them an interrupt using trigger_irq(); note that
758 * wait_for_vq_desc() does that for us if it has to wait.
760 static void add_used(struct virtqueue *vq, unsigned int head, int len)
762 struct vring_used_elem *used;
765 * The virtqueue contains a ring of used buffers. Get a pointer to the
766 * next entry in that used ring.
768 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
771 /* Make sure buffer is written before we update index. */
773 vq->vring.used->idx++;
777 /* And here's the combo meal deal. Supersize me! */
778 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
780 add_used(vq, head, len);
787 * We associate some data with the console for our exit hack.
789 struct console_abort {
790 /* How many times have they hit ^C? */
792 /* When did they start? */
793 struct timeval start;
796 /* This is the routine which handles console input (ie. stdin). */
797 static void console_input(struct virtqueue *vq)
800 unsigned int head, in_num, out_num;
801 struct console_abort *abort = vq->dev->priv;
802 struct iovec iov[vq->vring.num];
804 /* Make sure there's a descriptor available. */
805 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
807 errx(1, "Output buffers in console in queue?");
809 /* Read into it. This is where we usually wait. */
810 len = readv(STDIN_FILENO, iov, in_num);
812 /* Ran out of input? */
813 warnx("Failed to get console input, ignoring console.");
815 * For simplicity, dying threads kill the whole Launcher. So
822 /* Tell the Guest we used a buffer. */
823 add_used_and_trigger(vq, head, len);
826 * Three ^C within one second? Exit.
828 * This is such a hack, but works surprisingly well. Each ^C has to
829 * be in a buffer by itself, so they can't be too fast. But we check
830 * that we get three within about a second, so they can't be too
833 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
839 if (abort->count == 1)
840 gettimeofday(&abort->start, NULL);
841 else if (abort->count == 3) {
843 gettimeofday(&now, NULL);
844 /* Kill all Launcher processes with SIGINT, like normal ^C */
845 if (now.tv_sec <= abort->start.tv_sec+1)
851 /* This is the routine which handles console output (ie. stdout). */
852 static void console_output(struct virtqueue *vq)
854 unsigned int head, out, in;
855 struct iovec iov[vq->vring.num];
857 /* We usually wait in here, for the Guest to give us something. */
858 head = wait_for_vq_desc(vq, iov, &out, &in);
860 errx(1, "Input buffers in console output queue?");
862 /* writev can return a partial write, so we loop here. */
863 while (!iov_empty(iov, out)) {
864 int len = writev(STDOUT_FILENO, iov, out);
866 warn("Write to stdout gave %i (%d)", len, errno);
869 iov_consume(iov, out, NULL, len);
873 * We're finished with that buffer: if we're going to sleep,
874 * wait_for_vq_desc() will prod the Guest with an interrupt.
876 add_used(vq, head, 0);
882 * Handling output for network is also simple: we get all the output buffers
883 * and write them to /dev/net/tun.
889 static void net_output(struct virtqueue *vq)
891 struct net_info *net_info = vq->dev->priv;
892 unsigned int head, out, in;
893 struct iovec iov[vq->vring.num];
895 /* We usually wait in here for the Guest to give us a packet. */
896 head = wait_for_vq_desc(vq, iov, &out, &in);
898 errx(1, "Input buffers in net output queue?");
900 * Send the whole thing through to /dev/net/tun. It expects the exact
901 * same format: what a coincidence!
903 if (writev(net_info->tunfd, iov, out) < 0)
904 warnx("Write to tun failed (%d)?", errno);
907 * Done with that one; wait_for_vq_desc() will send the interrupt if
908 * all packets are processed.
910 add_used(vq, head, 0);
914 * Handling network input is a bit trickier, because I've tried to optimize it.
916 * First we have a helper routine which tells is if from this file descriptor
917 * (ie. the /dev/net/tun device) will block:
919 static bool will_block(int fd)
922 struct timeval zero = { 0, 0 };
925 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
929 * This handles packets coming in from the tun device to our Guest. Like all
930 * service routines, it gets called again as soon as it returns, so you don't
931 * see a while(1) loop here.
933 static void net_input(struct virtqueue *vq)
936 unsigned int head, out, in;
937 struct iovec iov[vq->vring.num];
938 struct net_info *net_info = vq->dev->priv;
941 * Get a descriptor to write an incoming packet into. This will also
942 * send an interrupt if they're out of descriptors.
944 head = wait_for_vq_desc(vq, iov, &out, &in);
946 errx(1, "Output buffers in net input queue?");
949 * If it looks like we'll block reading from the tun device, send them
952 if (vq->pending_used && will_block(net_info->tunfd))
956 * Read in the packet. This is where we normally wait (when there's no
957 * incoming network traffic).
959 len = readv(net_info->tunfd, iov, in);
961 warn("Failed to read from tun (%d).", errno);
964 * Mark that packet buffer as used, but don't interrupt here. We want
965 * to wait until we've done as much work as we can.
967 add_used(vq, head, len);
971 /* This is the helper to create threads: run the service routine in a loop. */
972 static int do_thread(void *_vq)
974 struct virtqueue *vq = _vq;
982 * When a child dies, we kill our entire process group with SIGTERM. This
983 * also has the side effect that the shell restores the console for us!
985 static void kill_launcher(int signal)
990 static void reset_device(struct device *dev)
992 struct virtqueue *vq;
994 verbose("Resetting device %s\n", dev->name);
996 /* Clear any features they've acked. */
997 memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len);
999 /* We're going to be explicitly killing threads, so ignore them. */
1000 signal(SIGCHLD, SIG_IGN);
1002 /* Zero out the virtqueues, get rid of their threads */
1003 for (vq = dev->vq; vq; vq = vq->next) {
1004 if (vq->thread != (pid_t)-1) {
1005 kill(vq->thread, SIGTERM);
1006 waitpid(vq->thread, NULL, 0);
1007 vq->thread = (pid_t)-1;
1009 memset(vq->vring.desc, 0,
1010 vring_size(vq->config.num, LGUEST_VRING_ALIGN));
1011 lg_last_avail(vq) = 0;
1013 dev->running = false;
1015 /* Now we care if threads die. */
1016 signal(SIGCHLD, (void *)kill_launcher);
1020 * This actually creates the thread which services the virtqueue for a device.
1022 static void create_thread(struct virtqueue *vq)
1025 * Create stack for thread. Since the stack grows upwards, we point
1026 * the stack pointer to the end of this region.
1028 char *stack = malloc(32768);
1029 unsigned long args[] = { LHREQ_EVENTFD,
1030 vq->config.pfn*getpagesize(), 0 };
1032 /* Create a zero-initialized eventfd. */
1033 vq->eventfd = eventfd(0, 0);
1034 if (vq->eventfd < 0)
1035 err(1, "Creating eventfd");
1036 args[2] = vq->eventfd;
1039 * Attach an eventfd to this virtqueue: it will go off when the Guest
1040 * does an LHCALL_NOTIFY for this vq.
1042 if (write(lguest_fd, &args, sizeof(args)) != 0)
1043 err(1, "Attaching eventfd");
1046 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1047 * we get a signal if it dies.
1049 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1050 if (vq->thread == (pid_t)-1)
1051 err(1, "Creating clone");
1053 /* We close our local copy now the child has it. */
1057 static void start_device(struct device *dev)
1060 struct virtqueue *vq;
1062 verbose("Device %s OK: offered", dev->name);
1063 for (i = 0; i < dev->feature_len; i++)
1064 verbose(" %02x", get_feature_bits(dev)[i]);
1065 verbose(", accepted");
1066 for (i = 0; i < dev->feature_len; i++)
1067 verbose(" %02x", get_feature_bits(dev)
1068 [dev->feature_len+i]);
1070 for (vq = dev->vq; vq; vq = vq->next) {
1074 dev->running = true;
1077 static void cleanup_devices(void)
1081 for (dev = devices.dev; dev; dev = dev->next)
1084 /* If we saved off the original terminal settings, restore them now. */
1085 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1086 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1089 /* When the Guest tells us they updated the status field, we handle it. */
1090 static void update_device_status(struct device *dev)
1092 /* A zero status is a reset, otherwise it's a set of flags. */
1093 if (dev->desc->status == 0)
1095 else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) {
1096 warnx("Device %s configuration FAILED", dev->name);
1101 err(1, "Device %s features finalized twice", dev->name);
1107 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1108 * particular, it's used to notify us of device status changes during boot.
1110 static void handle_output(unsigned long addr)
1114 /* Check each device. */
1115 for (i = devices.dev; i; i = i->next) {
1116 struct virtqueue *vq;
1119 * Notifications to device descriptors mean they updated the
1122 if (from_guest_phys(addr) == i->desc) {
1123 update_device_status(i);
1127 /* Devices should not be used before features are finalized. */
1128 for (vq = i->vq; vq; vq = vq->next) {
1129 if (addr != vq->config.pfn*getpagesize())
1131 errx(1, "Notification on %s before setup!", i->name);
1136 * Early console write is done using notify on a nul-terminated string
1137 * in Guest memory. It's also great for hacking debugging messages
1140 if (addr >= guest_limit)
1141 errx(1, "Bad NOTIFY %#lx", addr);
1143 write(STDOUT_FILENO, from_guest_phys(addr),
1144 strnlen(from_guest_phys(addr), guest_limit - addr));
1148 * This is where we emulate a handful of Guest instructions. It's ugly
1149 * and we used to do it in the kernel but it grew over time.
1153 * We use the ptrace syscall's pt_regs struct to talk about registers
1154 * to lguest: these macros convert the names to the offsets.
1156 #define getreg(name) getreg_off(offsetof(struct user_regs_struct, name))
1157 #define setreg(name, val) \
1158 setreg_off(offsetof(struct user_regs_struct, name), (val))
1160 static u32 getreg_off(size_t offset)
1163 unsigned long args[] = { LHREQ_GETREG, offset };
1165 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1166 err(1, "Getting register %u", offset);
1167 if (pread(lguest_fd, &r, sizeof(r), cpu_id) != sizeof(r))
1168 err(1, "Reading register %u", offset);
1173 static void setreg_off(size_t offset, u32 val)
1175 unsigned long args[] = { LHREQ_SETREG, offset, val };
1177 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1178 err(1, "Setting register %u", offset);
1181 static void emulate_insn(const u8 insn[])
1183 unsigned long args[] = { LHREQ_TRAP, 13 };
1184 unsigned int insnlen = 0, in = 0, small_operand = 0, byte_access;
1185 unsigned int eax, port, mask;
1187 * We always return all-ones on IO port reads, which traditionally
1188 * means "there's nothing there".
1190 u32 val = 0xFFFFFFFF;
1193 * This must be the Guest kernel trying to do something, not userspace!
1194 * The bottom two bits of the CS segment register are the privilege
1197 if ((getreg(xcs) & 3) != 0x1)
1200 /* Decoding x86 instructions is icky. */
1203 * Around 2.6.33, the kernel started using an emulation for the
1204 * cmpxchg8b instruction in early boot on many configurations. This
1205 * code isn't paravirtualized, and it tries to disable interrupts.
1206 * Ignore it, which will Mostly Work.
1208 if (insn[insnlen] == 0xfa) {
1209 /* "cli", or Clear Interrupt Enable instruction. Skip it. */
1215 * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out.
1217 if (insn[insnlen] == 0x66) {
1219 /* The instruction is 1 byte so far, read the next byte. */
1223 /* If the lower bit isn't set, it's a single byte access */
1224 byte_access = !(insn[insnlen] & 1);
1227 * Now we can ignore the lower bit and decode the 4 opcodes
1228 * we need to emulate.
1230 switch (insn[insnlen] & 0xFE) {
1231 case 0xE4: /* in <next byte>,%al */
1232 port = insn[insnlen+1];
1236 case 0xEC: /* in (%dx),%al */
1237 port = getreg(edx) & 0xFFFF;
1241 case 0xE6: /* out %al,<next byte> */
1242 port = insn[insnlen+1];
1245 case 0xEE: /* out %al,(%dx) */
1246 port = getreg(edx) & 0xFFFF;
1250 /* OK, we don't know what this is, can't emulate. */
1254 /* Set a mask of the 1, 2 or 4 bytes, depending on size of IO */
1257 else if (small_operand)
1262 /* This is the PS/2 keyboard status; 1 means ready for output */
1267 * If it was an "IN" instruction, they expect the result to be read
1268 * into %eax, so we change %eax.
1273 /* Clear the bits we're about to read */
1275 /* Copy bits in from val. */
1277 /* Now update the register. */
1281 verbose("IO %s of %x to %u: %#08x\n",
1282 in ? "IN" : "OUT", mask, port, eax);
1284 /* Finally, we've "done" the instruction, so move past it. */
1285 setreg(eip, getreg(eip) + insnlen);
1289 /* Inject trap into Guest. */
1290 if (write(lguest_fd, args, sizeof(args)) < 0)
1291 err(1, "Reinjecting trap 13 for fault at %#x", getreg(eip));
1298 * All devices need a descriptor so the Guest knows it exists, and a "struct
1299 * device" so the Launcher can keep track of it. We have common helper
1300 * routines to allocate and manage them.
1304 * The layout of the device page is a "struct lguest_device_desc" followed by a
1305 * number of virtqueue descriptors, then two sets of feature bits, then an
1306 * array of configuration bytes. This routine returns the configuration
1309 static u8 *device_config(const struct device *dev)
1311 return (void *)(dev->desc + 1)
1312 + dev->num_vq * sizeof(struct lguest_vqconfig)
1313 + dev->feature_len * 2;
1317 * This routine allocates a new "struct lguest_device_desc" from descriptor
1318 * table page just above the Guest's normal memory. It returns a pointer to
1321 static struct lguest_device_desc *new_dev_desc(u16 type)
1323 struct lguest_device_desc d = { .type = type };
1326 /* Figure out where the next device config is, based on the last one. */
1327 if (devices.lastdev)
1328 p = device_config(devices.lastdev)
1329 + devices.lastdev->desc->config_len;
1331 p = devices.descpage;
1333 /* We only have one page for all the descriptors. */
1334 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1335 errx(1, "Too many devices");
1337 /* p might not be aligned, so we memcpy in. */
1338 return memcpy(p, &d, sizeof(d));
1342 * Each device descriptor is followed by the description of its virtqueues. We
1343 * specify how many descriptors the virtqueue is to have.
1345 static void add_virtqueue(struct device *dev, unsigned int num_descs,
1346 void (*service)(struct virtqueue *))
1349 struct virtqueue **i, *vq = malloc(sizeof(*vq));
1352 /* First we need some memory for this virtqueue. */
1353 pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1)
1355 p = get_pages(pages);
1357 /* Initialize the virtqueue */
1359 vq->last_avail_idx = 0;
1363 * This is the routine the service thread will run, and its Process ID
1364 * once it's running.
1366 vq->service = service;
1367 vq->thread = (pid_t)-1;
1369 /* Initialize the configuration. */
1370 vq->config.num = num_descs;
1371 vq->config.irq = devices.next_irq++;
1372 vq->config.pfn = to_guest_phys(p) / getpagesize();
1374 /* Initialize the vring. */
1375 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1378 * Append virtqueue to this device's descriptor. We use
1379 * device_config() to get the end of the device's current virtqueues;
1380 * we check that we haven't added any config or feature information
1381 * yet, otherwise we'd be overwriting them.
1383 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1384 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1386 dev->desc->num_vq++;
1388 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1391 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1394 for (i = &dev->vq; *i; i = &(*i)->next);
1399 * The first half of the feature bitmask is for us to advertise features. The
1400 * second half is for the Guest to accept features.
1402 static void add_feature(struct device *dev, unsigned bit)
1404 u8 *features = get_feature_bits(dev);
1406 /* We can't extend the feature bits once we've added config bytes */
1407 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1408 assert(dev->desc->config_len == 0);
1409 dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1;
1412 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1416 * This routine sets the configuration fields for an existing device's
1417 * descriptor. It only works for the last device, but that's OK because that's
1420 static void set_config(struct device *dev, unsigned len, const void *conf)
1422 /* Check we haven't overflowed our single page. */
1423 if (device_config(dev) + len > devices.descpage + getpagesize())
1424 errx(1, "Too many devices");
1426 /* Copy in the config information, and store the length. */
1427 memcpy(device_config(dev), conf, len);
1428 dev->desc->config_len = len;
1430 /* Size must fit in config_len field (8 bits)! */
1431 assert(dev->desc->config_len == len);
1435 * This routine does all the creation and setup of a new device, including
1436 * calling new_dev_desc() to allocate the descriptor and device memory. We
1437 * don't actually start the service threads until later.
1439 * See what I mean about userspace being boring?
1441 static struct device *new_device(const char *name, u16 type)
1443 struct device *dev = malloc(sizeof(*dev));
1445 /* Now we populate the fields one at a time. */
1446 dev->desc = new_dev_desc(type);
1449 dev->feature_len = 0;
1451 dev->running = false;
1455 * Append to device list. Prepending to a single-linked list is
1456 * easier, but the user expects the devices to be arranged on the bus
1457 * in command-line order. The first network device on the command line
1458 * is eth0, the first block device /dev/vda, etc.
1460 if (devices.lastdev)
1461 devices.lastdev->next = dev;
1464 devices.lastdev = dev;
1470 * Our first setup routine is the console. It's a fairly simple device, but
1471 * UNIX tty handling makes it uglier than it could be.
1473 static void setup_console(void)
1477 /* If we can save the initial standard input settings... */
1478 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1479 struct termios term = orig_term;
1481 * Then we turn off echo, line buffering and ^C etc: We want a
1482 * raw input stream to the Guest.
1484 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1485 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1488 dev = new_device("console", VIRTIO_ID_CONSOLE);
1490 /* We store the console state in dev->priv, and initialize it. */
1491 dev->priv = malloc(sizeof(struct console_abort));
1492 ((struct console_abort *)dev->priv)->count = 0;
1495 * The console needs two virtqueues: the input then the output. When
1496 * they put something the input queue, we make sure we're listening to
1497 * stdin. When they put something in the output queue, we write it to
1500 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1501 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1503 verbose("device %u: console\n", ++devices.device_num);
1508 * Inter-guest networking is an interesting area. Simplest is to have a
1509 * --sharenet=<name> option which opens or creates a named pipe. This can be
1510 * used to send packets to another guest in a 1:1 manner.
1512 * More sophisticated is to use one of the tools developed for project like UML
1515 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1516 * completely generic ("here's my vring, attach to your vring") and would work
1517 * for any traffic. Of course, namespace and permissions issues need to be
1518 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1519 * multiple inter-guest channels behind one interface, although it would
1520 * require some manner of hotplugging new virtio channels.
1522 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
1525 static u32 str2ip(const char *ipaddr)
1529 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
1530 errx(1, "Failed to parse IP address '%s'", ipaddr);
1531 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
1534 static void str2mac(const char *macaddr, unsigned char mac[6])
1537 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
1538 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
1539 errx(1, "Failed to parse mac address '%s'", macaddr);
1549 * This code is "adapted" from libbridge: it attaches the Host end of the
1550 * network device to the bridge device specified by the command line.
1552 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1553 * dislike bridging), and I just try not to break it.
1555 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1561 errx(1, "must specify bridge name");
1563 ifidx = if_nametoindex(if_name);
1565 errx(1, "interface %s does not exist!", if_name);
1567 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
1568 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1569 ifr.ifr_ifindex = ifidx;
1570 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
1571 err(1, "can't add %s to bridge %s", if_name, br_name);
1575 * This sets up the Host end of the network device with an IP address, brings
1576 * it up so packets will flow, the copies the MAC address into the hwaddr
1579 static void configure_device(int fd, const char *tapif, u32 ipaddr)
1582 struct sockaddr_in sin;
1584 memset(&ifr, 0, sizeof(ifr));
1585 strcpy(ifr.ifr_name, tapif);
1587 /* Don't read these incantations. Just cut & paste them like I did! */
1588 sin.sin_family = AF_INET;
1589 sin.sin_addr.s_addr = htonl(ipaddr);
1590 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
1591 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
1592 err(1, "Setting %s interface address", tapif);
1593 ifr.ifr_flags = IFF_UP;
1594 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
1595 err(1, "Bringing interface %s up", tapif);
1598 static int get_tun_device(char tapif[IFNAMSIZ])
1603 /* Start with this zeroed. Messy but sure. */
1604 memset(&ifr, 0, sizeof(ifr));
1607 * We open the /dev/net/tun device and tell it we want a tap device. A
1608 * tap device is like a tun device, only somehow different. To tell
1609 * the truth, I completely blundered my way through this code, but it
1612 netfd = open_or_die("/dev/net/tun", O_RDWR);
1613 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1614 strcpy(ifr.ifr_name, "tap%d");
1615 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
1616 err(1, "configuring /dev/net/tun");
1618 if (ioctl(netfd, TUNSETOFFLOAD,
1619 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1620 err(1, "Could not set features for tun device");
1623 * We don't need checksums calculated for packets coming in this
1626 ioctl(netfd, TUNSETNOCSUM, 1);
1628 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1633 * Our network is a Host<->Guest network. This can either use bridging or
1634 * routing, but the principle is the same: it uses the "tun" device to inject
1635 * packets into the Host as if they came in from a normal network card. We
1636 * just shunt packets between the Guest and the tun device.
1638 static void setup_tun_net(char *arg)
1641 struct net_info *net_info = malloc(sizeof(*net_info));
1643 u32 ip = INADDR_ANY;
1644 bool bridging = false;
1645 char tapif[IFNAMSIZ], *p;
1646 struct virtio_net_config conf;
1648 net_info->tunfd = get_tun_device(tapif);
1650 /* First we create a new network device. */
1651 dev = new_device("net", VIRTIO_ID_NET);
1652 dev->priv = net_info;
1654 /* Network devices need a recv and a send queue, just like console. */
1655 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1656 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1659 * We need a socket to perform the magic network ioctls to bring up the
1660 * tap interface, connect to the bridge etc. Any socket will do!
1662 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1664 err(1, "opening IP socket");
1666 /* If the command line was --tunnet=bridge:<name> do bridging. */
1667 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
1668 arg += strlen(BRIDGE_PFX);
1672 /* A mac address may follow the bridge name or IP address */
1673 p = strchr(arg, ':');
1675 str2mac(p+1, conf.mac);
1676 add_feature(dev, VIRTIO_NET_F_MAC);
1680 /* arg is now either an IP address or a bridge name */
1682 add_to_bridge(ipfd, tapif, arg);
1686 /* Set up the tun device. */
1687 configure_device(ipfd, tapif, ip);
1689 /* Expect Guest to handle everything except UFO */
1690 add_feature(dev, VIRTIO_NET_F_CSUM);
1691 add_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
1692 add_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
1693 add_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
1694 add_feature(dev, VIRTIO_NET_F_GUEST_ECN);
1695 add_feature(dev, VIRTIO_NET_F_HOST_TSO4);
1696 add_feature(dev, VIRTIO_NET_F_HOST_TSO6);
1697 add_feature(dev, VIRTIO_NET_F_HOST_ECN);
1698 /* We handle indirect ring entries */
1699 add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
1700 /* We're compliant with the damn spec. */
1701 add_feature(dev, VIRTIO_F_ANY_LAYOUT);
1702 set_config(dev, sizeof(conf), &conf);
1704 /* We don't need the socket any more; setup is done. */
1707 devices.device_num++;
1710 verbose("device %u: tun %s attached to bridge: %s\n",
1711 devices.device_num, tapif, arg);
1713 verbose("device %u: tun %s: %s\n",
1714 devices.device_num, tapif, arg);
1718 /* This hangs off device->priv. */
1720 /* The size of the file. */
1723 /* The file descriptor for the file. */
1731 * The disk only has one virtqueue, so it only has one thread. It is really
1732 * simple: the Guest asks for a block number and we read or write that position
1735 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1736 * slow: the Guest waits until the read is finished before running anything
1737 * else, even if it could have been doing useful work.
1739 * We could have used async I/O, except it's reputed to suck so hard that
1740 * characters actually go missing from your code when you try to use it.
1742 static void blk_request(struct virtqueue *vq)
1744 struct vblk_info *vblk = vq->dev->priv;
1745 unsigned int head, out_num, in_num, wlen;
1748 struct virtio_blk_outhdr out;
1749 struct iovec iov[vq->vring.num];
1753 * Get the next request, where we normally wait. It triggers the
1754 * interrupt to acknowledge previously serviced requests (if any).
1756 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1758 /* Copy the output header from the front of the iov (adjusts iov) */
1759 iov_consume(iov, out_num, &out, sizeof(out));
1761 /* Find and trim end of iov input array, for our status byte. */
1763 for (i = out_num + in_num - 1; i >= out_num; i--) {
1764 if (iov[i].iov_len > 0) {
1765 in = iov[i].iov_base + iov[i].iov_len - 1;
1771 errx(1, "Bad virtblk cmd with no room for status");
1774 * For historical reasons, block operations are expressed in 512 byte
1777 off = out.sector * 512;
1780 * In general the virtio block driver is allowed to try SCSI commands.
1781 * It'd be nice if we supported eject, for example, but we don't.
1783 if (out.type & VIRTIO_BLK_T_SCSI_CMD) {
1784 fprintf(stderr, "Scsi commands unsupported\n");
1785 *in = VIRTIO_BLK_S_UNSUPP;
1787 } else if (out.type & VIRTIO_BLK_T_OUT) {
1791 * Move to the right location in the block file. This can fail
1792 * if they try to write past end.
1794 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1795 err(1, "Bad seek to sector %llu", out.sector);
1797 ret = writev(vblk->fd, iov, out_num);
1798 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
1801 * Grr... Now we know how long the descriptor they sent was, we
1802 * make sure they didn't try to write over the end of the block
1803 * file (possibly extending it).
1805 if (ret > 0 && off + ret > vblk->len) {
1806 /* Trim it back to the correct length */
1807 ftruncate64(vblk->fd, vblk->len);
1808 /* Die, bad Guest, die. */
1809 errx(1, "Write past end %llu+%u", off, ret);
1813 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1814 } else if (out.type & VIRTIO_BLK_T_FLUSH) {
1816 ret = fdatasync(vblk->fd);
1817 verbose("FLUSH fdatasync: %i\n", ret);
1819 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1824 * Move to the right location in the block file. This can fail
1825 * if they try to read past end.
1827 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1828 err(1, "Bad seek to sector %llu", out.sector);
1830 ret = readv(vblk->fd, iov + out_num, in_num);
1832 wlen = sizeof(*in) + ret;
1833 *in = VIRTIO_BLK_S_OK;
1836 *in = VIRTIO_BLK_S_IOERR;
1840 /* Finished that request. */
1841 add_used(vq, head, wlen);
1844 /*L:198 This actually sets up a virtual block device. */
1845 static void setup_block_file(const char *filename)
1848 struct vblk_info *vblk;
1849 struct virtio_blk_config conf;
1851 /* Creat the device. */
1852 dev = new_device("block", VIRTIO_ID_BLOCK);
1854 /* The device has one virtqueue, where the Guest places requests. */
1855 add_virtqueue(dev, VIRTQUEUE_NUM, blk_request);
1857 /* Allocate the room for our own bookkeeping */
1858 vblk = dev->priv = malloc(sizeof(*vblk));
1860 /* First we open the file and store the length. */
1861 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1862 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1864 /* We support FLUSH. */
1865 add_feature(dev, VIRTIO_BLK_F_FLUSH);
1867 /* Tell Guest how many sectors this device has. */
1868 conf.capacity = cpu_to_le64(vblk->len / 512);
1871 * Tell Guest not to put in too many descriptors at once: two are used
1872 * for the in and out elements.
1874 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1875 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1877 /* Don't try to put whole struct: we have 8 bit limit. */
1878 set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf);
1880 verbose("device %u: virtblock %llu sectors\n",
1881 ++devices.device_num, le64_to_cpu(conf.capacity));
1885 * Our random number generator device reads from /dev/urandom into the Guest's
1886 * input buffers. The usual case is that the Guest doesn't want random numbers
1887 * and so has no buffers although /dev/urandom is still readable, whereas
1888 * console is the reverse.
1890 * The same logic applies, however.
1896 static void rng_input(struct virtqueue *vq)
1899 unsigned int head, in_num, out_num, totlen = 0;
1900 struct rng_info *rng_info = vq->dev->priv;
1901 struct iovec iov[vq->vring.num];
1903 /* First we need a buffer from the Guests's virtqueue. */
1904 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1906 errx(1, "Output buffers in rng?");
1909 * Just like the console write, we loop to cover the whole iovec.
1910 * In this case, short reads actually happen quite a bit.
1912 while (!iov_empty(iov, in_num)) {
1913 len = readv(rng_info->rfd, iov, in_num);
1915 err(1, "Read from /dev/urandom gave %i", len);
1916 iov_consume(iov, in_num, NULL, len);
1920 /* Tell the Guest about the new input. */
1921 add_used(vq, head, totlen);
1925 * This creates a "hardware" random number device for the Guest.
1927 static void setup_rng(void)
1930 struct rng_info *rng_info = malloc(sizeof(*rng_info));
1932 /* Our device's private info simply contains the /dev/urandom fd. */
1933 rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY);
1935 /* Create the new device. */
1936 dev = new_device("rng", VIRTIO_ID_RNG);
1937 dev->priv = rng_info;
1939 /* The device has one virtqueue, where the Guest places inbufs. */
1940 add_virtqueue(dev, VIRTQUEUE_NUM, rng_input);
1942 verbose("device %u: rng\n", devices.device_num++);
1944 /* That's the end of device setup. */
1946 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1947 static void __attribute__((noreturn)) restart_guest(void)
1952 * Since we don't track all open fds, we simply close everything beyond
1955 for (i = 3; i < FD_SETSIZE; i++)
1958 /* Reset all the devices (kills all threads). */
1961 execv(main_args[0], main_args);
1962 err(1, "Could not exec %s", main_args[0]);
1966 * Finally we reach the core of the Launcher which runs the Guest, serves
1967 * its input and output, and finally, lays it to rest.
1969 static void __attribute__((noreturn)) run_guest(void)
1972 struct lguest_pending notify;
1975 /* We read from the /dev/lguest device to run the Guest. */
1976 readval = pread(lguest_fd, ¬ify, sizeof(notify), cpu_id);
1978 /* One unsigned long means the Guest did HCALL_NOTIFY */
1979 if (readval == sizeof(notify)) {
1980 if (notify.trap == 0x1F) {
1981 verbose("Notify on address %#08x\n",
1983 handle_output(notify.addr);
1984 } else if (notify.trap == 13) {
1985 verbose("Emulating instruction at %#x\n",
1987 emulate_insn(notify.insn);
1989 errx(1, "Unknown trap %i addr %#08x\n",
1990 notify.trap, notify.addr);
1991 /* ENOENT means the Guest died. Reading tells us why. */
1992 } else if (errno == ENOENT) {
1993 char reason[1024] = { 0 };
1994 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
1995 errx(1, "%s", reason);
1996 /* ERESTART means that we need to reboot the guest */
1997 } else if (errno == ERESTART) {
1999 /* Anything else means a bug or incompatible change. */
2001 err(1, "Running guest failed");
2005 * This is the end of the Launcher. The good news: we are over halfway
2006 * through! The bad news: the most fiendish part of the code still lies ahead
2009 * Are you ready? Take a deep breath and join me in the core of the Host, in
2013 static struct option opts[] = {
2014 { "verbose", 0, NULL, 'v' },
2015 { "tunnet", 1, NULL, 't' },
2016 { "block", 1, NULL, 'b' },
2017 { "rng", 0, NULL, 'r' },
2018 { "initrd", 1, NULL, 'i' },
2019 { "username", 1, NULL, 'u' },
2020 { "chroot", 1, NULL, 'c' },
2023 static void usage(void)
2025 errx(1, "Usage: lguest [--verbose] "
2026 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
2027 "|--block=<filename>|--initrd=<filename>]...\n"
2028 "<mem-in-mb> vmlinux [args...]");
2031 /*L:105 The main routine is where the real work begins: */
2032 int main(int argc, char *argv[])
2034 /* Memory, code startpoint and size of the (optional) initrd. */
2035 unsigned long mem = 0, start, initrd_size = 0;
2036 /* Two temporaries. */
2038 /* The boot information for the Guest. */
2039 struct boot_params *boot;
2040 /* If they specify an initrd file to load. */
2041 const char *initrd_name = NULL;
2043 /* Password structure for initgroups/setres[gu]id */
2044 struct passwd *user_details = NULL;
2046 /* Directory to chroot to */
2047 char *chroot_path = NULL;
2049 /* Save the args: we "reboot" by execing ourselves again. */
2053 * First we initialize the device list. We keep a pointer to the last
2054 * device, and the next interrupt number to use for devices (1:
2055 * remember that 0 is used by the timer).
2057 devices.lastdev = NULL;
2058 devices.next_irq = 1;
2060 /* We're CPU 0. In fact, that's the only CPU possible right now. */
2064 * We need to know how much memory so we can set up the device
2065 * descriptor and memory pages for the devices as we parse the command
2066 * line. So we quickly look through the arguments to find the amount
2069 for (i = 1; i < argc; i++) {
2070 if (argv[i][0] != '-') {
2071 mem = atoi(argv[i]) * 1024 * 1024;
2073 * We start by mapping anonymous pages over all of
2074 * guest-physical memory range. This fills it with 0,
2075 * and ensures that the Guest won't be killed when it
2076 * tries to access it.
2078 guest_base = map_zeroed_pages(mem / getpagesize()
2081 guest_max = mem + DEVICE_PAGES*getpagesize();
2082 devices.descpage = get_pages(1);
2087 /* The options are fairly straight-forward */
2088 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
2094 setup_tun_net(optarg);
2097 setup_block_file(optarg);
2103 initrd_name = optarg;
2106 user_details = getpwnam(optarg);
2108 err(1, "getpwnam failed, incorrect username?");
2111 chroot_path = optarg;
2114 warnx("Unknown argument %s", argv[optind]);
2119 * After the other arguments we expect memory and kernel image name,
2120 * followed by command line arguments for the kernel.
2122 if (optind + 2 > argc)
2125 verbose("Guest base is at %p\n", guest_base);
2127 /* We always have a console device */
2130 /* Now we load the kernel */
2131 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
2133 /* Boot information is stashed at physical address 0 */
2134 boot = from_guest_phys(0);
2136 /* Map the initrd image if requested (at top of physical memory) */
2138 initrd_size = load_initrd(initrd_name, mem);
2140 * These are the location in the Linux boot header where the
2141 * start and size of the initrd are expected to be found.
2143 boot->hdr.ramdisk_image = mem - initrd_size;
2144 boot->hdr.ramdisk_size = initrd_size;
2145 /* The bootloader type 0xFF means "unknown"; that's OK. */
2146 boot->hdr.type_of_loader = 0xFF;
2150 * The Linux boot header contains an "E820" memory map: ours is a
2151 * simple, single region.
2153 boot->e820_entries = 1;
2154 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
2156 * The boot header contains a command line pointer: we put the command
2157 * line after the boot header.
2159 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
2160 /* We use a simple helper to copy the arguments separated by spaces. */
2161 concat((char *)(boot + 1), argv+optind+2);
2163 /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
2164 boot->hdr.kernel_alignment = 0x1000000;
2166 /* Boot protocol version: 2.07 supports the fields for lguest. */
2167 boot->hdr.version = 0x207;
2169 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2170 boot->hdr.hardware_subarch = 1;
2172 /* Tell the entry path not to try to reload segment registers. */
2173 boot->hdr.loadflags |= KEEP_SEGMENTS;
2175 /* We tell the kernel to initialize the Guest. */
2178 /* Ensure that we terminate if a device-servicing child dies. */
2179 signal(SIGCHLD, kill_launcher);
2181 /* If we exit via err(), this kills all the threads, restores tty. */
2182 atexit(cleanup_devices);
2184 /* If requested, chroot to a directory */
2186 if (chroot(chroot_path) != 0)
2187 err(1, "chroot(\"%s\") failed", chroot_path);
2189 if (chdir("/") != 0)
2190 err(1, "chdir(\"/\") failed");
2192 verbose("chroot done\n");
2195 /* If requested, drop privileges */
2200 u = user_details->pw_uid;
2201 g = user_details->pw_gid;
2203 if (initgroups(user_details->pw_name, g) != 0)
2204 err(1, "initgroups failed");
2206 if (setresgid(g, g, g) != 0)
2207 err(1, "setresgid failed");
2209 if (setresuid(u, u, u) != 0)
2210 err(1, "setresuid failed");
2212 verbose("Dropping privileges completed\n");
2215 /* Finally, run the Guest. This doesn't return. */
2221 * Mastery is done: you now know everything I do.
2223 * But surely you have seen code, features and bugs in your wanderings which
2224 * you now yearn to attack? That is the real game, and I look forward to you
2225 * patching and forking lguest into the Your-Name-Here-visor.
2227 * Farewell, and good coding!