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/pci_regs.h>
47 #ifndef VIRTIO_F_ANY_LAYOUT
48 #define VIRTIO_F_ANY_LAYOUT 27
52 * We can ignore the 43 include files we need for this program, but I do want
53 * to draw attention to the use of kernel-style types.
55 * As Linus said, "C is a Spartan language, and so should your naming be." I
56 * like these abbreviations, so we define them here. Note that u64 is always
57 * unsigned long long, which works on all Linux systems: this means that we can
58 * use %llu in printf for any u64.
60 typedef unsigned long long u64;
66 #define VIRTIO_CONFIG_NO_LEGACY
67 #define VIRTIO_PCI_NO_LEGACY
68 #define VIRTIO_BLK_NO_LEGACY
70 /* Use in-kernel ones, which defines VIRTIO_F_VERSION_1 */
71 #include "../../include/uapi/linux/virtio_config.h"
72 #include "../../include/uapi/linux/virtio_net.h"
73 #include "../../include/uapi/linux/virtio_blk.h"
74 #include "../../include/uapi/linux/virtio_console.h"
75 #include "../../include/uapi/linux/virtio_rng.h"
76 #include <linux/virtio_ring.h>
77 #include "../../include/uapi/linux/virtio_pci.h"
78 #include <asm/bootparam.h>
79 #include "../../include/linux/lguest_launcher.h"
81 #define BRIDGE_PFX "bridge:"
83 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
85 /* We can have up to 256 pages for devices. */
86 #define DEVICE_PAGES 256
87 /* This will occupy 3 pages: it must be a power of 2. */
88 #define VIRTQUEUE_NUM 256
91 * verbose is both a global flag and a macro. The C preprocessor allows
92 * this, and although I wouldn't recommend it, it works quite nicely here.
95 #define verbose(args...) \
96 do { if (verbose) printf(args); } while(0)
99 /* The pointer to the start of guest memory. */
100 static void *guest_base;
101 /* The maximum guest physical address allowed, and maximum possible. */
102 static unsigned long guest_limit, guest_max, guest_mmio;
103 /* The /dev/lguest file descriptor. */
104 static int lguest_fd;
106 /* a per-cpu variable indicating whose vcpu is currently running */
107 static unsigned int __thread cpu_id;
109 /* 5 bit device number in the PCI_CONFIG_ADDR => 32 only */
110 #define MAX_PCI_DEVICES 32
112 /* This is our list of devices. */
114 /* Counter to assign interrupt numbers. */
115 unsigned int next_irq;
117 /* Counter to print out convenient device numbers. */
118 unsigned int device_num;
121 struct device *pci[MAX_PCI_DEVICES];
124 /* The list of Guest devices, based on command line arguments. */
125 static struct device_list devices;
127 struct virtio_pci_cfg_cap {
128 struct virtio_pci_cap cap;
129 u32 pci_cfg_data; /* Data for BAR access. */
132 struct virtio_pci_mmio {
133 struct virtio_pci_common_cfg cfg;
137 /* Device-specific configuration follows this. */
140 /* This is the layout (little-endian) of the PCI config space. */
142 u16 vendor_id, device_id;
144 u8 revid, prog_if, subclass, class;
145 u8 cacheline_size, lat_timer, header_type, bist;
148 u16 subsystem_vendor_id, subsystem_device_id;
149 u32 expansion_rom_addr;
150 u8 capabilities, reserved1[3];
152 u8 irq_line, irq_pin, min_grant, max_latency;
154 /* Now, this is the linked capability list. */
155 struct virtio_pci_cap common;
156 struct virtio_pci_notify_cap notify;
157 struct virtio_pci_cap isr;
158 struct virtio_pci_cap device;
159 struct virtio_pci_cfg_cap cfg_access;
162 /* The device structure describes a single device. */
164 /* The name of this device, for --verbose. */
167 /* Any queues attached to this device */
168 struct virtqueue *vq;
170 /* Is it operational */
173 /* PCI configuration */
175 struct pci_config config;
176 u32 config_words[sizeof(struct pci_config) / sizeof(u32)];
179 /* Features we offer, and those accepted. */
180 u64 features, features_accepted;
182 /* Device-specific config hangs off the end of this. */
183 struct virtio_pci_mmio *mmio;
185 /* PCI MMIO resources (all in BAR0) */
189 /* Device-specific data. */
193 /* The virtqueue structure describes a queue attached to a device. */
195 struct virtqueue *next;
197 /* Which device owns me. */
200 /* The actual ring of buffers. */
203 /* The information about this virtqueue (we only use queue_size on) */
204 struct virtio_pci_common_cfg pci_config;
206 /* Last available index we saw. */
209 /* How many are used since we sent last irq? */
210 unsigned int pending_used;
212 /* Eventfd where Guest notifications arrive. */
215 /* Function for the thread which is servicing this virtqueue. */
216 void (*service)(struct virtqueue *vq);
220 /* Remember the arguments to the program so we can "reboot" */
221 static char **main_args;
223 /* The original tty settings to restore on exit. */
224 static struct termios orig_term;
227 * We have to be careful with barriers: our devices are all run in separate
228 * threads and so we need to make sure that changes visible to the Guest happen
231 #define wmb() __asm__ __volatile__("" : : : "memory")
232 #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
233 #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory")
235 /* Wrapper for the last available index. Makes it easier to change. */
236 #define lg_last_avail(vq) ((vq)->last_avail_idx)
239 * The virtio configuration space is defined to be little-endian. x86 is
240 * little-endian too, but it's nice to be explicit so we have these helpers.
242 #define cpu_to_le16(v16) (v16)
243 #define cpu_to_le32(v32) (v32)
244 #define cpu_to_le64(v64) (v64)
245 #define le16_to_cpu(v16) (v16)
246 #define le32_to_cpu(v32) (v32)
247 #define le64_to_cpu(v64) (v64)
249 /* Is this iovec empty? */
250 static bool iov_empty(const struct iovec iov[], unsigned int num_iov)
254 for (i = 0; i < num_iov; i++)
260 /* Take len bytes from the front of this iovec. */
261 static void iov_consume(struct iovec iov[], unsigned num_iov,
262 void *dest, unsigned len)
266 for (i = 0; i < num_iov; i++) {
269 used = iov[i].iov_len < len ? iov[i].iov_len : len;
271 memcpy(dest, iov[i].iov_base, used);
274 iov[i].iov_base += used;
275 iov[i].iov_len -= used;
279 errx(1, "iovec too short!");
283 * The Launcher code itself takes us out into userspace, that scary place where
284 * pointers run wild and free! Unfortunately, like most userspace programs,
285 * it's quite boring (which is why everyone likes to hack on the kernel!).
286 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
287 * you through this section. Or, maybe not.
289 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
290 * memory and stores it in "guest_base". In other words, Guest physical ==
291 * Launcher virtual with an offset.
293 * This can be tough to get your head around, but usually it just means that we
294 * use these trivial conversion functions when the Guest gives us its
295 * "physical" addresses:
297 static void *from_guest_phys(unsigned long addr)
299 return guest_base + addr;
302 static unsigned long to_guest_phys(const void *addr)
304 return (addr - guest_base);
308 * Loading the Kernel.
310 * We start with couple of simple helper routines. open_or_die() avoids
311 * error-checking code cluttering the callers:
313 static int open_or_die(const char *name, int flags)
315 int fd = open(name, flags);
317 err(1, "Failed to open %s", name);
321 /* map_zeroed_pages() takes a number of pages. */
322 static void *map_zeroed_pages(unsigned int num)
324 int fd = open_or_die("/dev/zero", O_RDONLY);
328 * We use a private mapping (ie. if we write to the page, it will be
329 * copied). We allocate an extra two pages PROT_NONE to act as guard
330 * pages against read/write attempts that exceed allocated space.
332 addr = mmap(NULL, getpagesize() * (num+2),
333 PROT_NONE, MAP_PRIVATE, fd, 0);
335 if (addr == MAP_FAILED)
336 err(1, "Mmapping %u pages of /dev/zero", num);
338 if (mprotect(addr + getpagesize(), getpagesize() * num,
339 PROT_READ|PROT_WRITE) == -1)
340 err(1, "mprotect rw %u pages failed", num);
343 * One neat mmap feature is that you can close the fd, and it
348 /* Return address after PROT_NONE page */
349 return addr + getpagesize();
352 /* Get some bytes which won't be mapped into the guest. */
353 static unsigned long get_mmio_region(size_t size)
355 unsigned long addr = guest_mmio;
361 /* Size has to be a power of 2 (and multiple of 16) */
362 for (i = 1; i < size; i <<= 1);
370 * This routine is used to load the kernel or initrd. It tries mmap, but if
371 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
372 * it falls back to reading the memory in.
374 static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
379 * We map writable even though for some segments are marked read-only.
380 * The kernel really wants to be writable: it patches its own
383 * MAP_PRIVATE means that the page won't be copied until a write is
384 * done to it. This allows us to share untouched memory between
387 if (mmap(addr, len, PROT_READ|PROT_WRITE,
388 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
391 /* pread does a seek and a read in one shot: saves a few lines. */
392 r = pread(fd, addr, len, offset);
394 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
398 * This routine takes an open vmlinux image, which is in ELF, and maps it into
399 * the Guest memory. ELF = Embedded Linking Format, which is the format used
400 * by all modern binaries on Linux including the kernel.
402 * The ELF headers give *two* addresses: a physical address, and a virtual
403 * address. We use the physical address; the Guest will map itself to the
406 * We return the starting address.
408 static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
410 Elf32_Phdr phdr[ehdr->e_phnum];
414 * Sanity checks on the main ELF header: an x86 executable with a
415 * reasonable number of correctly-sized program headers.
417 if (ehdr->e_type != ET_EXEC
418 || ehdr->e_machine != EM_386
419 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
420 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
421 errx(1, "Malformed elf header");
424 * An ELF executable contains an ELF header and a number of "program"
425 * headers which indicate which parts ("segments") of the program to
429 /* We read in all the program headers at once: */
430 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
431 err(1, "Seeking to program headers");
432 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
433 err(1, "Reading program headers");
436 * Try all the headers: there are usually only three. A read-only one,
437 * a read-write one, and a "note" section which we don't load.
439 for (i = 0; i < ehdr->e_phnum; i++) {
440 /* If this isn't a loadable segment, we ignore it */
441 if (phdr[i].p_type != PT_LOAD)
444 verbose("Section %i: size %i addr %p\n",
445 i, phdr[i].p_memsz, (void *)phdr[i].p_paddr);
447 /* We map this section of the file at its physical address. */
448 map_at(elf_fd, from_guest_phys(phdr[i].p_paddr),
449 phdr[i].p_offset, phdr[i].p_filesz);
452 /* The entry point is given in the ELF header. */
453 return ehdr->e_entry;
457 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
458 * to jump into it and it will unpack itself. We used to have to perform some
459 * hairy magic because the unpacking code scared me.
461 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
462 * a small patch to jump over the tricky bits in the Guest, so now we just read
463 * the funky header so we know where in the file to load, and away we go!
465 static unsigned long load_bzimage(int fd)
467 struct boot_params boot;
469 /* Modern bzImages get loaded at 1M. */
470 void *p = from_guest_phys(0x100000);
473 * Go back to the start of the file and read the header. It should be
474 * a Linux boot header (see Documentation/x86/boot.txt)
476 lseek(fd, 0, SEEK_SET);
477 read(fd, &boot, sizeof(boot));
479 /* Inside the setup_hdr, we expect the magic "HdrS" */
480 if (memcmp(&boot.hdr.header, "HdrS", 4) != 0)
481 errx(1, "This doesn't look like a bzImage to me");
483 /* Skip over the extra sectors of the header. */
484 lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET);
486 /* Now read everything into memory. in nice big chunks. */
487 while ((r = read(fd, p, 65536)) > 0)
490 /* Finally, code32_start tells us where to enter the kernel. */
491 return boot.hdr.code32_start;
495 * Loading the kernel is easy when it's a "vmlinux", but most kernels
496 * come wrapped up in the self-decompressing "bzImage" format. With a little
497 * work, we can load those, too.
499 static unsigned long load_kernel(int fd)
503 /* Read in the first few bytes. */
504 if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
505 err(1, "Reading kernel");
507 /* If it's an ELF file, it starts with "\177ELF" */
508 if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0)
509 return map_elf(fd, &hdr);
511 /* Otherwise we assume it's a bzImage, and try to load it. */
512 return load_bzimage(fd);
516 * This is a trivial little helper to align pages. Andi Kleen hated it because
517 * it calls getpagesize() twice: "it's dumb code."
519 * Kernel guys get really het up about optimization, even when it's not
520 * necessary. I leave this code as a reaction against that.
522 static inline unsigned long page_align(unsigned long addr)
524 /* Add upwards and truncate downwards. */
525 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
529 * An "initial ram disk" is a disk image loaded into memory along with the
530 * kernel which the kernel can use to boot from without needing any drivers.
531 * Most distributions now use this as standard: the initrd contains the code to
532 * load the appropriate driver modules for the current machine.
534 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
535 * kernels. He sent me this (and tells me when I break it).
537 static unsigned long load_initrd(const char *name, unsigned long mem)
543 ifd = open_or_die(name, O_RDONLY);
544 /* fstat() is needed to get the file size. */
545 if (fstat(ifd, &st) < 0)
546 err(1, "fstat() on initrd '%s'", name);
549 * We map the initrd at the top of memory, but mmap wants it to be
550 * page-aligned, so we round the size up for that.
552 len = page_align(st.st_size);
553 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
555 * Once a file is mapped, you can close the file descriptor. It's a
556 * little odd, but quite useful.
559 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
561 /* We return the initrd size. */
567 * Simple routine to roll all the commandline arguments together with spaces
570 static void concat(char *dst, char *args[])
572 unsigned int i, len = 0;
574 for (i = 0; args[i]; i++) {
576 strcat(dst+len, " ");
579 strcpy(dst+len, args[i]);
580 len += strlen(args[i]);
582 /* In case it's empty. */
587 * This is where we actually tell the kernel to initialize the Guest. We
588 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
589 * the base of Guest "physical" memory, the top physical page to allow and the
590 * entry point for the Guest.
592 static void tell_kernel(unsigned long start)
594 unsigned long args[] = { LHREQ_INITIALIZE,
595 (unsigned long)guest_base,
596 guest_limit / getpagesize(), start,
597 (guest_mmio+getpagesize()-1) / getpagesize() };
598 verbose("Guest: %p - %p (%#lx, MMIO %#lx)\n",
599 guest_base, guest_base + guest_limit,
600 guest_limit, guest_mmio);
601 lguest_fd = open_or_die("/dev/lguest", O_RDWR);
602 if (write(lguest_fd, args, sizeof(args)) < 0)
603 err(1, "Writing to /dev/lguest");
610 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
611 * We need to make sure it's not trying to reach into the Launcher itself, so
612 * we have a convenient routine which checks it and exits with an error message
613 * if something funny is going on:
615 static void *_check_pointer(unsigned long addr, unsigned int size,
619 * Check if the requested address and size exceeds the allocated memory,
620 * or addr + size wraps around.
622 if ((addr + size) > guest_limit || (addr + size) < addr)
623 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
625 * We return a pointer for the caller's convenience, now we know it's
628 return from_guest_phys(addr);
630 /* A macro which transparently hands the line number to the real function. */
631 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
634 * Each buffer in the virtqueues is actually a chain of descriptors. This
635 * function returns the next descriptor in the chain, or vq->vring.num if we're
638 static unsigned next_desc(struct vring_desc *desc,
639 unsigned int i, unsigned int max)
643 /* If this descriptor says it doesn't chain, we're done. */
644 if (!(desc[i].flags & VRING_DESC_F_NEXT))
647 /* Check they're not leading us off end of descriptors. */
649 /* Make sure compiler knows to grab that: we don't want it changing! */
653 errx(1, "Desc next is %u", next);
659 * This actually sends the interrupt for this virtqueue, if we've used a
662 static void trigger_irq(struct virtqueue *vq)
664 unsigned long buf[] = { LHREQ_IRQ, vq->dev->config.irq_line };
666 /* Don't inform them if nothing used. */
667 if (!vq->pending_used)
669 vq->pending_used = 0;
671 /* If they don't want an interrupt, don't send one... */
672 if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) {
679 * If MSI-X capability is disabled, the device MUST set the Queue
680 * Interrupt bit in ISR status before sending a virtqueue notification
683 vq->dev->mmio->isr = 0x1;
685 /* Send the Guest an interrupt tell them we used something up. */
686 if (write(lguest_fd, buf, sizeof(buf)) != 0)
687 err(1, "Triggering irq %i", vq->dev->config.irq_line);
691 * This looks in the virtqueue for the first available buffer, and converts
692 * it to an iovec for convenient access. Since descriptors consist of some
693 * number of output then some number of input descriptors, it's actually two
694 * iovecs, but we pack them into one and note how many of each there were.
696 * This function waits if necessary, and returns the descriptor number found.
698 static unsigned wait_for_vq_desc(struct virtqueue *vq,
700 unsigned int *out_num, unsigned int *in_num)
702 unsigned int i, head, max;
703 struct vring_desc *desc;
704 u16 last_avail = lg_last_avail(vq);
706 /* There's nothing available? */
707 while (last_avail == vq->vring.avail->idx) {
711 * Since we're about to sleep, now is a good time to tell the
712 * Guest about what we've used up to now.
716 /* OK, now we need to know about added descriptors. */
717 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
720 * They could have slipped one in as we were doing that: make
721 * sure it's written, then check again.
724 if (last_avail != vq->vring.avail->idx) {
725 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
729 /* Nothing new? Wait for eventfd to tell us they refilled. */
730 if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event))
731 errx(1, "Event read failed?");
733 /* We don't need to be notified again. */
734 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
737 /* Check it isn't doing very strange things with descriptor numbers. */
738 if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num)
739 errx(1, "Guest moved used index from %u to %u",
740 last_avail, vq->vring.avail->idx);
743 * Make sure we read the descriptor number *after* we read the ring
744 * update; don't let the cpu or compiler change the order.
749 * Grab the next descriptor number they're advertising, and increment
750 * the index we've seen.
752 head = vq->vring.avail->ring[last_avail % vq->vring.num];
755 /* If their number is silly, that's a fatal mistake. */
756 if (head >= vq->vring.num)
757 errx(1, "Guest says index %u is available", head);
759 /* When we start there are none of either input nor output. */
760 *out_num = *in_num = 0;
763 desc = vq->vring.desc;
767 * We have to read the descriptor after we read the descriptor number,
768 * but there's a data dependency there so the CPU shouldn't reorder
769 * that: no rmb() required.
774 * If this is an indirect entry, then this buffer contains a
775 * descriptor table which we handle as if it's any normal
778 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
779 if (desc[i].len % sizeof(struct vring_desc))
780 errx(1, "Invalid size for indirect buffer table");
782 max = desc[i].len / sizeof(struct vring_desc);
783 desc = check_pointer(desc[i].addr, desc[i].len);
787 /* Grab the first descriptor, and check it's OK. */
788 iov[*out_num + *in_num].iov_len = desc[i].len;
789 iov[*out_num + *in_num].iov_base
790 = check_pointer(desc[i].addr, desc[i].len);
791 /* If this is an input descriptor, increment that count. */
792 if (desc[i].flags & VRING_DESC_F_WRITE)
796 * If it's an output descriptor, they're all supposed
797 * to come before any input descriptors.
800 errx(1, "Descriptor has out after in");
804 /* If we've got too many, that implies a descriptor loop. */
805 if (*out_num + *in_num > max)
806 errx(1, "Looped descriptor");
807 } while ((i = next_desc(desc, i, max)) != max);
813 * After we've used one of their buffers, we tell the Guest about it. Sometime
814 * later we'll want to send them an interrupt using trigger_irq(); note that
815 * wait_for_vq_desc() does that for us if it has to wait.
817 static void add_used(struct virtqueue *vq, unsigned int head, int len)
819 struct vring_used_elem *used;
822 * The virtqueue contains a ring of used buffers. Get a pointer to the
823 * next entry in that used ring.
825 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
828 /* Make sure buffer is written before we update index. */
830 vq->vring.used->idx++;
834 /* And here's the combo meal deal. Supersize me! */
835 static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
837 add_used(vq, head, len);
844 * We associate some data with the console for our exit hack.
846 struct console_abort {
847 /* How many times have they hit ^C? */
849 /* When did they start? */
850 struct timeval start;
853 /* This is the routine which handles console input (ie. stdin). */
854 static void console_input(struct virtqueue *vq)
857 unsigned int head, in_num, out_num;
858 struct console_abort *abort = vq->dev->priv;
859 struct iovec iov[vq->vring.num];
861 /* Make sure there's a descriptor available. */
862 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
864 errx(1, "Output buffers in console in queue?");
866 /* Read into it. This is where we usually wait. */
867 len = readv(STDIN_FILENO, iov, in_num);
869 /* Ran out of input? */
870 warnx("Failed to get console input, ignoring console.");
872 * For simplicity, dying threads kill the whole Launcher. So
879 /* Tell the Guest we used a buffer. */
880 add_used_and_trigger(vq, head, len);
883 * Three ^C within one second? Exit.
885 * This is such a hack, but works surprisingly well. Each ^C has to
886 * be in a buffer by itself, so they can't be too fast. But we check
887 * that we get three within about a second, so they can't be too
890 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
896 if (abort->count == 1)
897 gettimeofday(&abort->start, NULL);
898 else if (abort->count == 3) {
900 gettimeofday(&now, NULL);
901 /* Kill all Launcher processes with SIGINT, like normal ^C */
902 if (now.tv_sec <= abort->start.tv_sec+1)
908 /* This is the routine which handles console output (ie. stdout). */
909 static void console_output(struct virtqueue *vq)
911 unsigned int head, out, in;
912 struct iovec iov[vq->vring.num];
914 /* We usually wait in here, for the Guest to give us something. */
915 head = wait_for_vq_desc(vq, iov, &out, &in);
917 errx(1, "Input buffers in console output queue?");
919 /* writev can return a partial write, so we loop here. */
920 while (!iov_empty(iov, out)) {
921 int len = writev(STDOUT_FILENO, iov, out);
923 warn("Write to stdout gave %i (%d)", len, errno);
926 iov_consume(iov, out, NULL, len);
930 * We're finished with that buffer: if we're going to sleep,
931 * wait_for_vq_desc() will prod the Guest with an interrupt.
933 add_used(vq, head, 0);
939 * Handling output for network is also simple: we get all the output buffers
940 * and write them to /dev/net/tun.
946 static void net_output(struct virtqueue *vq)
948 struct net_info *net_info = vq->dev->priv;
949 unsigned int head, out, in;
950 struct iovec iov[vq->vring.num];
952 /* We usually wait in here for the Guest to give us a packet. */
953 head = wait_for_vq_desc(vq, iov, &out, &in);
955 errx(1, "Input buffers in net output queue?");
957 * Send the whole thing through to /dev/net/tun. It expects the exact
958 * same format: what a coincidence!
960 if (writev(net_info->tunfd, iov, out) < 0)
961 warnx("Write to tun failed (%d)?", errno);
964 * Done with that one; wait_for_vq_desc() will send the interrupt if
965 * all packets are processed.
967 add_used(vq, head, 0);
971 * Handling network input is a bit trickier, because I've tried to optimize it.
973 * First we have a helper routine which tells is if from this file descriptor
974 * (ie. the /dev/net/tun device) will block:
976 static bool will_block(int fd)
979 struct timeval zero = { 0, 0 };
982 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
986 * This handles packets coming in from the tun device to our Guest. Like all
987 * service routines, it gets called again as soon as it returns, so you don't
988 * see a while(1) loop here.
990 static void net_input(struct virtqueue *vq)
993 unsigned int head, out, in;
994 struct iovec iov[vq->vring.num];
995 struct net_info *net_info = vq->dev->priv;
998 * Get a descriptor to write an incoming packet into. This will also
999 * send an interrupt if they're out of descriptors.
1001 head = wait_for_vq_desc(vq, iov, &out, &in);
1003 errx(1, "Output buffers in net input queue?");
1006 * If it looks like we'll block reading from the tun device, send them
1009 if (vq->pending_used && will_block(net_info->tunfd))
1013 * Read in the packet. This is where we normally wait (when there's no
1014 * incoming network traffic).
1016 len = readv(net_info->tunfd, iov, in);
1018 warn("Failed to read from tun (%d).", errno);
1021 * Mark that packet buffer as used, but don't interrupt here. We want
1022 * to wait until we've done as much work as we can.
1024 add_used(vq, head, len);
1028 /* This is the helper to create threads: run the service routine in a loop. */
1029 static int do_thread(void *_vq)
1031 struct virtqueue *vq = _vq;
1039 * When a child dies, we kill our entire process group with SIGTERM. This
1040 * also has the side effect that the shell restores the console for us!
1042 static void kill_launcher(int signal)
1047 static void reset_vq_pci_config(struct virtqueue *vq)
1049 vq->pci_config.queue_size = VIRTQUEUE_NUM;
1050 vq->pci_config.queue_enable = 0;
1053 static void reset_device(struct device *dev)
1055 struct virtqueue *vq;
1057 verbose("Resetting device %s\n", dev->name);
1059 /* Clear any features they've acked. */
1060 dev->features_accepted = 0;
1062 /* We're going to be explicitly killing threads, so ignore them. */
1063 signal(SIGCHLD, SIG_IGN);
1068 * The device MUST present a 0 in queue_enable on reset.
1070 * This means we set it here, and reset the saved ones in every vq.
1072 dev->mmio->cfg.queue_enable = 0;
1074 /* Get rid of the virtqueue threads */
1075 for (vq = dev->vq; vq; vq = vq->next) {
1076 vq->last_avail_idx = 0;
1077 reset_vq_pci_config(vq);
1078 if (vq->thread != (pid_t)-1) {
1079 kill(vq->thread, SIGTERM);
1080 waitpid(vq->thread, NULL, 0);
1081 vq->thread = (pid_t)-1;
1084 dev->running = false;
1086 /* Now we care if threads die. */
1087 signal(SIGCHLD, (void *)kill_launcher);
1090 static void cleanup_devices(void)
1094 for (i = 1; i < MAX_PCI_DEVICES; i++) {
1095 struct device *d = devices.pci[i];
1101 /* If we saved off the original terminal settings, restore them now. */
1102 if (orig_term.c_lflag & (ISIG|ICANON|ECHO))
1103 tcsetattr(STDIN_FILENO, TCSANOW, &orig_term);
1107 * We do PCI. This is mainly done to let us test the kernel virtio PCI
1111 /* Linux expects a PCI host bridge: ours is a dummy, and first on the bus. */
1112 static struct device pci_host_bridge;
1114 static void init_pci_host_bridge(void)
1116 pci_host_bridge.name = "PCI Host Bridge";
1117 pci_host_bridge.config.class = 0x06; /* bridge */
1118 pci_host_bridge.config.subclass = 0; /* host bridge */
1119 devices.pci[0] = &pci_host_bridge;
1122 /* The IO ports used to read the PCI config space. */
1123 #define PCI_CONFIG_ADDR 0xCF8
1124 #define PCI_CONFIG_DATA 0xCFC
1127 * Not really portable, but does help readability: this is what the Guest
1128 * writes to the PCI_CONFIG_ADDR IO port.
1130 union pci_config_addr {
1134 unsigned funcnum: 3;
1137 unsigned reserved: 7;
1138 unsigned enabled : 1;
1144 * We cache what they wrote to the address port, so we know what they're
1145 * talking about when they access the data port.
1147 static union pci_config_addr pci_config_addr;
1149 static struct device *find_pci_device(unsigned int index)
1151 return devices.pci[index];
1154 /* PCI can do 1, 2 and 4 byte reads; we handle that here. */
1155 static void ioread(u16 off, u32 v, u32 mask, u32 *val)
1158 assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF);
1159 *val = (v >> (off * 8)) & mask;
1162 /* PCI can do 1, 2 and 4 byte writes; we handle that here. */
1163 static void iowrite(u16 off, u32 v, u32 mask, u32 *dst)
1166 assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF);
1167 *dst &= ~(mask << (off * 8));
1168 *dst |= (v & mask) << (off * 8);
1172 * Where PCI_CONFIG_DATA accesses depends on the previous write to
1175 static struct device *dev_and_reg(u32 *reg)
1177 if (!pci_config_addr.bits.enabled)
1180 if (pci_config_addr.bits.funcnum != 0)
1183 if (pci_config_addr.bits.busnum != 0)
1186 if (pci_config_addr.bits.offset * 4 >= sizeof(struct pci_config))
1189 *reg = pci_config_addr.bits.offset;
1190 return find_pci_device(pci_config_addr.bits.devnum);
1194 * We can get invalid combinations of values while they're writing, so we
1195 * only fault if they try to write with some invalid bar/offset/length.
1197 static bool valid_bar_access(struct device *d,
1198 struct virtio_pci_cfg_cap *cfg_access)
1200 /* We only have 1 bar (BAR0) */
1201 if (cfg_access->cap.bar != 0)
1204 /* Check it's within BAR0. */
1205 if (cfg_access->cap.offset >= d->mmio_size
1206 || cfg_access->cap.offset + cfg_access->cap.length > d->mmio_size)
1209 /* Check length is 1, 2 or 4. */
1210 if (cfg_access->cap.length != 1
1211 && cfg_access->cap.length != 2
1212 && cfg_access->cap.length != 4)
1218 * The driver MUST NOT write a cap.offset which is not a multiple of
1219 * cap.length (ie. all accesses MUST be aligned).
1221 if (cfg_access->cap.offset % cfg_access->cap.length != 0)
1224 /* Return pointer into word in BAR0. */
1228 /* Is this accessing the PCI config address port?. */
1229 static bool is_pci_addr_port(u16 port)
1231 return port >= PCI_CONFIG_ADDR && port < PCI_CONFIG_ADDR + 4;
1234 static bool pci_addr_iowrite(u16 port, u32 mask, u32 val)
1236 iowrite(port - PCI_CONFIG_ADDR, val, mask,
1237 &pci_config_addr.val);
1238 verbose("PCI%s: %#x/%x: bus %u dev %u func %u reg %u\n",
1239 pci_config_addr.bits.enabled ? "" : " DISABLED",
1241 pci_config_addr.bits.busnum,
1242 pci_config_addr.bits.devnum,
1243 pci_config_addr.bits.funcnum,
1244 pci_config_addr.bits.offset);
1248 static void pci_addr_ioread(u16 port, u32 mask, u32 *val)
1250 ioread(port - PCI_CONFIG_ADDR, pci_config_addr.val, mask, val);
1253 /* Is this accessing the PCI config data port?. */
1254 static bool is_pci_data_port(u16 port)
1256 return port >= PCI_CONFIG_DATA && port < PCI_CONFIG_DATA + 4;
1259 static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask);
1261 static bool pci_data_iowrite(u16 port, u32 mask, u32 val)
1264 struct device *d = dev_and_reg(®);
1266 /* Complain if they don't belong to a device. */
1270 /* They can do 1 byte writes, etc. */
1271 portoff = port - PCI_CONFIG_DATA;
1274 * PCI uses a weird way to determine the BAR size: the OS
1275 * writes all 1's, and sees which ones stick.
1277 if (&d->config_words[reg] == &d->config.bar[0]) {
1280 iowrite(portoff, val, mask, &d->config.bar[0]);
1281 for (i = 0; (1 << i) < d->mmio_size; i++)
1282 d->config.bar[0] &= ~(1 << i);
1284 } else if ((&d->config_words[reg] > &d->config.bar[0]
1285 && &d->config_words[reg] <= &d->config.bar[6])
1286 || &d->config_words[reg] == &d->config.expansion_rom_addr) {
1287 /* Allow writing to any other BAR, or expansion ROM */
1288 iowrite(portoff, val, mask, &d->config_words[reg]);
1290 /* We let them overide latency timer and cacheline size */
1291 } else if (&d->config_words[reg] == (void *)&d->config.cacheline_size) {
1292 /* Only let them change the first two fields. */
1293 if (mask == 0xFFFFFFFF)
1295 iowrite(portoff, val, mask, &d->config_words[reg]);
1297 } else if (&d->config_words[reg] == (void *)&d->config.command
1298 && mask == 0xFFFF) {
1299 /* Ignore command writes. */
1301 } else if (&d->config_words[reg]
1302 == (void *)&d->config.cfg_access.cap.bar
1303 || &d->config_words[reg]
1304 == &d->config.cfg_access.cap.length
1305 || &d->config_words[reg]
1306 == &d->config.cfg_access.cap.offset) {
1309 * The VIRTIO_PCI_CAP_PCI_CFG capability
1310 * provides a backdoor to access the MMIO
1311 * regions without mapping them. Weird, but
1314 iowrite(portoff, val, mask, &d->config_words[reg]);
1316 } else if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) {
1322 * Upon detecting driver write access to pci_cfg_data, the
1323 * device MUST execute a write access at offset cap.offset at
1324 * BAR selected by cap.bar using the first cap.length bytes
1325 * from pci_cfg_data.
1329 if (!valid_bar_access(d, &d->config.cfg_access))
1332 iowrite(portoff, val, mask, &d->config.cfg_access.pci_cfg_data);
1335 * Now emulate a write. The mask we use is set by
1336 * len, *not* this write!
1338 write_mask = (1ULL<<(8*d->config.cfg_access.cap.length)) - 1;
1339 verbose("Window writing %#x/%#x to bar %u, offset %u len %u\n",
1340 d->config.cfg_access.pci_cfg_data, write_mask,
1341 d->config.cfg_access.cap.bar,
1342 d->config.cfg_access.cap.offset,
1343 d->config.cfg_access.cap.length);
1345 emulate_mmio_write(d, d->config.cfg_access.cap.offset,
1346 d->config.cfg_access.pci_cfg_data,
1354 * The driver MUST NOT write into any field of the capability
1355 * structure, with the exception of those with cap_type
1356 * VIRTIO_PCI_CAP_PCI_CFG...
1361 static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask);
1363 static void pci_data_ioread(u16 port, u32 mask, u32 *val)
1366 struct device *d = dev_and_reg(®);
1371 /* Read through the PCI MMIO access window is special */
1372 if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) {
1378 * Upon detecting driver read access to pci_cfg_data, the
1379 * device MUST execute a read access of length cap.length at
1380 * offset cap.offset at BAR selected by cap.bar and store the
1381 * first cap.length bytes in pci_cfg_data.
1384 if (!valid_bar_access(d, &d->config.cfg_access))
1385 errx(1, "Invalid cfg_access to bar%u, offset %u len %u",
1386 d->config.cfg_access.cap.bar,
1387 d->config.cfg_access.cap.offset,
1388 d->config.cfg_access.cap.length);
1391 * Read into the window. The mask we use is set by
1392 * len, *not* this read!
1394 read_mask = (1ULL<<(8*d->config.cfg_access.cap.length))-1;
1395 d->config.cfg_access.pci_cfg_data
1396 = emulate_mmio_read(d,
1397 d->config.cfg_access.cap.offset,
1399 verbose("Window read %#x/%#x from bar %u, offset %u len %u\n",
1400 d->config.cfg_access.pci_cfg_data, read_mask,
1401 d->config.cfg_access.cap.bar,
1402 d->config.cfg_access.cap.offset,
1403 d->config.cfg_access.cap.length);
1405 ioread(port - PCI_CONFIG_DATA, d->config_words[reg], mask, val);
1409 * This is where we emulate a handful of Guest instructions. It's ugly
1410 * and we used to do it in the kernel but it grew over time.
1414 * We use the ptrace syscall's pt_regs struct to talk about registers
1415 * to lguest: these macros convert the names to the offsets.
1417 #define getreg(name) getreg_off(offsetof(struct user_regs_struct, name))
1418 #define setreg(name, val) \
1419 setreg_off(offsetof(struct user_regs_struct, name), (val))
1421 static u32 getreg_off(size_t offset)
1424 unsigned long args[] = { LHREQ_GETREG, offset };
1426 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1427 err(1, "Getting register %u", offset);
1428 if (pread(lguest_fd, &r, sizeof(r), cpu_id) != sizeof(r))
1429 err(1, "Reading register %u", offset);
1434 static void setreg_off(size_t offset, u32 val)
1436 unsigned long args[] = { LHREQ_SETREG, offset, val };
1438 if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0)
1439 err(1, "Setting register %u", offset);
1442 /* Get register by instruction encoding */
1443 static u32 getreg_num(unsigned regnum, u32 mask)
1445 /* 8 bit ops use regnums 4-7 for high parts of word */
1446 if (mask == 0xFF && (regnum & 0x4))
1447 return getreg_num(regnum & 0x3, 0xFFFF) >> 8;
1450 case 0: return getreg(eax) & mask;
1451 case 1: return getreg(ecx) & mask;
1452 case 2: return getreg(edx) & mask;
1453 case 3: return getreg(ebx) & mask;
1454 case 4: return getreg(esp) & mask;
1455 case 5: return getreg(ebp) & mask;
1456 case 6: return getreg(esi) & mask;
1457 case 7: return getreg(edi) & mask;
1462 /* Set register by instruction encoding */
1463 static void setreg_num(unsigned regnum, u32 val, u32 mask)
1465 /* Don't try to set bits out of range */
1466 assert(~(val & ~mask));
1468 /* 8 bit ops use regnums 4-7 for high parts of word */
1469 if (mask == 0xFF && (regnum & 0x4)) {
1470 /* Construct the 16 bits we want. */
1471 val = (val << 8) | getreg_num(regnum & 0x3, 0xFF);
1472 setreg_num(regnum & 0x3, val, 0xFFFF);
1477 case 0: setreg(eax, val | (getreg(eax) & ~mask)); return;
1478 case 1: setreg(ecx, val | (getreg(ecx) & ~mask)); return;
1479 case 2: setreg(edx, val | (getreg(edx) & ~mask)); return;
1480 case 3: setreg(ebx, val | (getreg(ebx) & ~mask)); return;
1481 case 4: setreg(esp, val | (getreg(esp) & ~mask)); return;
1482 case 5: setreg(ebp, val | (getreg(ebp) & ~mask)); return;
1483 case 6: setreg(esi, val | (getreg(esi) & ~mask)); return;
1484 case 7: setreg(edi, val | (getreg(edi) & ~mask)); return;
1489 /* Get bytes of displacement appended to instruction, from r/m encoding */
1490 static u32 insn_displacement_len(u8 mod_reg_rm)
1492 /* Switch on the mod bits */
1493 switch (mod_reg_rm >> 6) {
1495 /* If mod == 0, and r/m == 101, 16-bit displacement follows */
1496 if ((mod_reg_rm & 0x7) == 0x5)
1498 /* Normally, mod == 0 means no literal displacement */
1501 /* One byte displacement */
1504 /* Four byte displacement */
1513 static void emulate_insn(const u8 insn[])
1515 unsigned long args[] = { LHREQ_TRAP, 13 };
1516 unsigned int insnlen = 0, in = 0, small_operand = 0, byte_access;
1517 unsigned int eax, port, mask;
1519 * Default is to return all-ones on IO port reads, which traditionally
1520 * means "there's nothing there".
1522 u32 val = 0xFFFFFFFF;
1525 * This must be the Guest kernel trying to do something, not userspace!
1526 * The bottom two bits of the CS segment register are the privilege
1529 if ((getreg(xcs) & 3) != 0x1)
1532 /* Decoding x86 instructions is icky. */
1535 * Around 2.6.33, the kernel started using an emulation for the
1536 * cmpxchg8b instruction in early boot on many configurations. This
1537 * code isn't paravirtualized, and it tries to disable interrupts.
1538 * Ignore it, which will Mostly Work.
1540 if (insn[insnlen] == 0xfa) {
1541 /* "cli", or Clear Interrupt Enable instruction. Skip it. */
1547 * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out.
1549 if (insn[insnlen] == 0x66) {
1551 /* The instruction is 1 byte so far, read the next byte. */
1555 /* If the lower bit isn't set, it's a single byte access */
1556 byte_access = !(insn[insnlen] & 1);
1559 * Now we can ignore the lower bit and decode the 4 opcodes
1560 * we need to emulate.
1562 switch (insn[insnlen] & 0xFE) {
1563 case 0xE4: /* in <next byte>,%al */
1564 port = insn[insnlen+1];
1568 case 0xEC: /* in (%dx),%al */
1569 port = getreg(edx) & 0xFFFF;
1573 case 0xE6: /* out %al,<next byte> */
1574 port = insn[insnlen+1];
1577 case 0xEE: /* out %al,(%dx) */
1578 port = getreg(edx) & 0xFFFF;
1582 /* OK, we don't know what this is, can't emulate. */
1586 /* Set a mask of the 1, 2 or 4 bytes, depending on size of IO */
1589 else if (small_operand)
1595 * If it was an "IN" instruction, they expect the result to be read
1596 * into %eax, so we change %eax.
1601 /* This is the PS/2 keyboard status; 1 means ready for output */
1604 else if (is_pci_addr_port(port))
1605 pci_addr_ioread(port, mask, &val);
1606 else if (is_pci_data_port(port))
1607 pci_data_ioread(port, mask, &val);
1609 /* Clear the bits we're about to read */
1611 /* Copy bits in from val. */
1613 /* Now update the register. */
1616 if (is_pci_addr_port(port)) {
1617 if (!pci_addr_iowrite(port, mask, eax))
1619 } else if (is_pci_data_port(port)) {
1620 if (!pci_data_iowrite(port, mask, eax))
1623 /* There are many other ports, eg. CMOS clock, serial
1624 * and parallel ports, so we ignore them all. */
1627 verbose("IO %s of %x to %u: %#08x\n",
1628 in ? "IN" : "OUT", mask, port, eax);
1630 /* Finally, we've "done" the instruction, so move past it. */
1631 setreg(eip, getreg(eip) + insnlen);
1635 warnx("Attempt to %s port %u (%#x mask)",
1636 in ? "read from" : "write to", port, mask);
1639 /* Inject trap into Guest. */
1640 if (write(lguest_fd, args, sizeof(args)) < 0)
1641 err(1, "Reinjecting trap 13 for fault at %#x", getreg(eip));
1644 static struct device *find_mmio_region(unsigned long paddr, u32 *off)
1648 for (i = 1; i < MAX_PCI_DEVICES; i++) {
1649 struct device *d = devices.pci[i];
1653 if (paddr < d->mmio_addr)
1655 if (paddr >= d->mmio_addr + d->mmio_size)
1657 *off = paddr - d->mmio_addr;
1663 /* FIXME: Use vq array. */
1664 static struct virtqueue *vq_by_num(struct device *d, u32 num)
1666 struct virtqueue *vq = d->vq;
1674 static void save_vq_config(const struct virtio_pci_common_cfg *cfg,
1675 struct virtqueue *vq)
1677 vq->pci_config = *cfg;
1680 static void restore_vq_config(struct virtio_pci_common_cfg *cfg,
1681 struct virtqueue *vq)
1683 /* Only restore the per-vq part */
1684 size_t off = offsetof(struct virtio_pci_common_cfg, queue_size);
1686 memcpy((void *)cfg + off, (void *)&vq->pci_config + off,
1687 sizeof(*cfg) - off);
1691 * When they enable the virtqueue, we check that their setup is valid.
1693 static void enable_virtqueue(struct device *d, struct virtqueue *vq)
1696 * Create stack for thread. Since the stack grows upwards, we point
1697 * the stack pointer to the end of this region.
1699 char *stack = malloc(32768);
1701 /* Because lguest is 32 bit, all the descriptor high bits must be 0 */
1702 if (vq->pci_config.queue_desc_hi
1703 || vq->pci_config.queue_avail_hi
1704 || vq->pci_config.queue_used_hi)
1705 errx(1, "%s: invalid 64-bit queue address", d->name);
1707 /* Initialize the virtqueue and check they're all in range. */
1708 vq->vring.num = vq->pci_config.queue_size;
1709 vq->vring.desc = check_pointer(vq->pci_config.queue_desc_lo,
1710 sizeof(*vq->vring.desc) * vq->vring.num);
1711 vq->vring.avail = check_pointer(vq->pci_config.queue_avail_lo,
1712 sizeof(*vq->vring.avail)
1713 + (sizeof(vq->vring.avail->ring[0])
1715 vq->vring.used = check_pointer(vq->pci_config.queue_used_lo,
1716 sizeof(*vq->vring.used)
1717 + (sizeof(vq->vring.used->ring[0])
1721 /* Create a zero-initialized eventfd. */
1722 vq->eventfd = eventfd(0, 0);
1723 if (vq->eventfd < 0)
1724 err(1, "Creating eventfd");
1727 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1728 * we get a signal if it dies.
1730 vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq);
1731 if (vq->thread == (pid_t)-1)
1732 err(1, "Creating clone");
1735 static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask)
1737 struct virtqueue *vq;
1740 case offsetof(struct virtio_pci_mmio, cfg.device_feature_select):
1744 * The device MUST present the feature bits it is offering in
1745 * device_feature, starting at bit device_feature_select ∗ 32
1746 * for any device_feature_select written by the driver
1749 d->mmio->cfg.device_feature = d->features;
1751 d->mmio->cfg.device_feature = (d->features >> 32);
1753 d->mmio->cfg.device_feature = 0;
1754 goto write_through32;
1755 case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select):
1757 errx(1, "%s: Unexpected driver select %u",
1759 goto write_through32;
1760 case offsetof(struct virtio_pci_mmio, cfg.guest_feature):
1761 if (d->mmio->cfg.guest_feature_select == 0) {
1762 d->features_accepted &= ~((u64)0xFFFFFFFF);
1763 d->features_accepted |= val;
1765 assert(d->mmio->cfg.guest_feature_select == 1);
1766 d->features_accepted &= 0xFFFFFFFF;
1767 d->features_accepted |= ((u64)val) << 32;
1769 if (d->features_accepted & ~d->features)
1770 errx(1, "%s: over-accepted features %#llx of %#llx",
1771 d->name, d->features_accepted, d->features);
1772 goto write_through32;
1773 case offsetof(struct virtio_pci_mmio, cfg.device_status):
1774 verbose("%s: device status -> %#x\n", d->name, val);
1778 * The device MUST reset when 0 is written to device_status,
1779 * and present a 0 in device_status once that is done.
1783 goto write_through8;
1784 case offsetof(struct virtio_pci_mmio, cfg.queue_select):
1785 vq = vq_by_num(d, val);
1789 * The device MUST present a 0 in queue_size if the virtqueue
1790 * corresponding to the current queue_select is unavailable.
1793 d->mmio->cfg.queue_size = 0;
1794 goto write_through16;
1796 /* Save registers for old vq, if it was a valid vq */
1797 if (d->mmio->cfg.queue_size)
1798 save_vq_config(&d->mmio->cfg,
1799 vq_by_num(d, d->mmio->cfg.queue_select));
1800 /* Restore the registers for the queue they asked for */
1801 restore_vq_config(&d->mmio->cfg, vq);
1802 goto write_through16;
1803 case offsetof(struct virtio_pci_mmio, cfg.queue_size):
1807 * The driver MUST NOT write a value which is not a power of 2
1811 errx(1, "%s: invalid queue size %u\n", d->name, val);
1812 if (d->mmio->cfg.queue_enable)
1813 errx(1, "%s: changing queue size on live device",
1815 goto write_through16;
1816 case offsetof(struct virtio_pci_mmio, cfg.queue_msix_vector):
1817 errx(1, "%s: attempt to set MSIX vector to %u",
1819 case offsetof(struct virtio_pci_mmio, cfg.queue_enable):
1823 * The driver MUST NOT write a 0 to queue_enable.
1826 errx(1, "%s: setting queue_enable to %u", d->name, val);
1827 d->mmio->cfg.queue_enable = val;
1828 save_vq_config(&d->mmio->cfg,
1829 vq_by_num(d, d->mmio->cfg.queue_select));
1833 * The driver MUST configure the other virtqueue fields before
1834 * enabling the virtqueue with queue_enable.
1836 enable_virtqueue(d, vq_by_num(d, d->mmio->cfg.queue_select));
1837 goto write_through16;
1838 case offsetof(struct virtio_pci_mmio, cfg.queue_notify_off):
1839 errx(1, "%s: attempt to write to queue_notify_off", d->name);
1840 case offsetof(struct virtio_pci_mmio, cfg.queue_desc_lo):
1841 case offsetof(struct virtio_pci_mmio, cfg.queue_desc_hi):
1842 case offsetof(struct virtio_pci_mmio, cfg.queue_avail_lo):
1843 case offsetof(struct virtio_pci_mmio, cfg.queue_avail_hi):
1844 case offsetof(struct virtio_pci_mmio, cfg.queue_used_lo):
1845 case offsetof(struct virtio_pci_mmio, cfg.queue_used_hi):
1849 * The driver MUST configure the other virtqueue fields before
1850 * enabling the virtqueue with queue_enable.
1852 if (d->mmio->cfg.queue_enable)
1853 errx(1, "%s: changing queue on live device",
1855 goto write_through32;
1856 case offsetof(struct virtio_pci_mmio, notify):
1857 vq = vq_by_num(d, val);
1859 errx(1, "Invalid vq notification on %u", val);
1860 /* Notify the process handling this vq by adding 1 to eventfd */
1861 write(vq->eventfd, "\1\0\0\0\0\0\0\0", 8);
1862 goto write_through16;
1863 case offsetof(struct virtio_pci_mmio, isr):
1864 errx(1, "%s: Unexpected write to isr", d->name);
1865 /* Weird corner case: write to emerg_wr of console */
1866 case sizeof(struct virtio_pci_mmio)
1867 + offsetof(struct virtio_console_config, emerg_wr):
1868 if (strcmp(d->name, "console") == 0) {
1870 write(STDOUT_FILENO, &c, 1);
1871 goto write_through32;
1873 /* Fall through... */
1878 * The driver MUST NOT write to device_feature, num_queues,
1879 * config_generation or queue_notify_off.
1881 errx(1, "%s: Unexpected write to offset %u", d->name, off);
1888 * The driver MUST access each field using the “natural” access
1889 * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for
1890 * 16-bit fields and 8-bit accesses for 8-bit fields.
1893 if (mask != 0xFFFFFFFF) {
1894 errx(1, "%s: non-32-bit write to offset %u (%#x)",
1895 d->name, off, getreg(eip));
1898 memcpy((char *)d->mmio + off, &val, 4);
1903 errx(1, "%s: non-16-bit (%#x) write to offset %u (%#x)",
1904 d->name, mask, off, getreg(eip));
1905 memcpy((char *)d->mmio + off, &val, 2);
1910 errx(1, "%s: non-8-bit write to offset %u (%#x)",
1911 d->name, off, getreg(eip));
1912 memcpy((char *)d->mmio + off, &val, 1);
1916 static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask)
1922 case offsetof(struct virtio_pci_mmio, cfg.device_feature_select):
1923 case offsetof(struct virtio_pci_mmio, cfg.device_feature):
1924 case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select):
1925 case offsetof(struct virtio_pci_mmio, cfg.guest_feature):
1926 goto read_through32;
1927 case offsetof(struct virtio_pci_mmio, cfg.msix_config):
1928 errx(1, "%s: read of msix_config", d->name);
1929 case offsetof(struct virtio_pci_mmio, cfg.num_queues):
1930 goto read_through16;
1931 case offsetof(struct virtio_pci_mmio, cfg.device_status):
1932 case offsetof(struct virtio_pci_mmio, cfg.config_generation):
1936 * The device MUST present a changed config_generation after
1937 * the driver has read a device-specific configuration value
1938 * which has changed since any part of the device-specific
1939 * configuration was last read.
1941 * This is simple: none of our devices change config, so this
1945 case offsetof(struct virtio_pci_mmio, notify):
1946 goto read_through16;
1947 case offsetof(struct virtio_pci_mmio, isr):
1949 errx(1, "%s: non-8-bit read from offset %u (%#x)",
1950 d->name, off, getreg(eip));
1955 * The device MUST reset ISR status to 0 on driver read.
1959 case offsetof(struct virtio_pci_mmio, padding):
1960 errx(1, "%s: read from padding (%#x)",
1961 d->name, getreg(eip));
1963 /* Read from device config space, beware unaligned overflow */
1964 if (off > d->mmio_size - 4)
1965 errx(1, "%s: read past end (%#x)",
1966 d->name, getreg(eip));
1967 if (mask == 0xFFFFFFFF)
1968 goto read_through32;
1969 else if (mask == 0xFFFF)
1970 goto read_through16;
1978 * The driver MUST access each field using the “natural” access
1979 * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for
1980 * 16-bit fields and 8-bit accesses for 8-bit fields.
1983 if (mask != 0xFFFFFFFF)
1984 errx(1, "%s: non-32-bit read to offset %u (%#x)",
1985 d->name, off, getreg(eip));
1986 memcpy(&val, (char *)d->mmio + off, 4);
1991 errx(1, "%s: non-16-bit read to offset %u (%#x)",
1992 d->name, off, getreg(eip));
1993 memcpy(&val, (char *)d->mmio + off, 2);
1998 errx(1, "%s: non-8-bit read to offset %u (%#x)",
1999 d->name, off, getreg(eip));
2000 memcpy(&val, (char *)d->mmio + off, 1);
2004 static void emulate_mmio(unsigned long paddr, const u8 *insn)
2006 u32 val, off, mask = 0xFFFFFFFF, insnlen = 0;
2007 struct device *d = find_mmio_region(paddr, &off);
2008 unsigned long args[] = { LHREQ_TRAP, 14 };
2011 warnx("MMIO touching %#08lx (not a device)", paddr);
2015 /* Prefix makes it a 16 bit op */
2016 if (insn[0] == 0x66) {
2022 if (insn[insnlen] == 0x89) {
2023 /* Next byte is r/m byte: bits 3-5 are register. */
2024 val = getreg_num((insn[insnlen+1] >> 3) & 0x7, mask);
2025 emulate_mmio_write(d, off, val, mask);
2026 insnlen += 2 + insn_displacement_len(insn[insnlen+1]);
2027 } else if (insn[insnlen] == 0x8b) { /* ioread */
2028 /* Next byte is r/m byte: bits 3-5 are register. */
2029 val = emulate_mmio_read(d, off, mask);
2030 setreg_num((insn[insnlen+1] >> 3) & 0x7, val, mask);
2031 insnlen += 2 + insn_displacement_len(insn[insnlen+1]);
2032 } else if (insn[0] == 0x88) { /* 8-bit iowrite */
2034 /* Next byte is r/m byte: bits 3-5 are register. */
2035 val = getreg_num((insn[1] >> 3) & 0x7, mask);
2036 emulate_mmio_write(d, off, val, mask);
2037 insnlen = 2 + insn_displacement_len(insn[1]);
2038 } else if (insn[0] == 0x8a) { /* 8-bit ioread */
2040 val = emulate_mmio_read(d, off, mask);
2041 setreg_num((insn[1] >> 3) & 0x7, val, mask);
2042 insnlen = 2 + insn_displacement_len(insn[1]);
2044 warnx("Unknown MMIO instruction touching %#08lx:"
2045 " %02x %02x %02x %02x at %u",
2046 paddr, insn[0], insn[1], insn[2], insn[3], getreg(eip));
2048 /* Inject trap into Guest. */
2049 if (write(lguest_fd, args, sizeof(args)) < 0)
2050 err(1, "Reinjecting trap 14 for fault at %#x",
2055 /* Finally, we've "done" the instruction, so move past it. */
2056 setreg(eip, getreg(eip) + insnlen);
2062 * All devices need a descriptor so the Guest knows it exists, and a "struct
2063 * device" so the Launcher can keep track of it. We have common helper
2064 * routines to allocate and manage them.
2066 static void add_pci_virtqueue(struct device *dev,
2067 void (*service)(struct virtqueue *))
2069 struct virtqueue **i, *vq = malloc(sizeof(*vq));
2071 /* Initialize the virtqueue */
2073 vq->last_avail_idx = 0;
2077 * This is the routine the service thread will run, and its Process ID
2078 * once it's running.
2080 vq->service = service;
2081 vq->thread = (pid_t)-1;
2083 /* Initialize the configuration. */
2084 reset_vq_pci_config(vq);
2085 vq->pci_config.queue_notify_off = 0;
2087 /* Add one to the number of queues */
2088 vq->dev->mmio->cfg.num_queues++;
2091 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
2094 for (i = &dev->vq; *i; i = &(*i)->next);
2098 /* The Guest accesses the feature bits via the PCI common config MMIO region */
2099 static void add_pci_feature(struct device *dev, unsigned bit)
2101 dev->features |= (1ULL << bit);
2104 /* For devices with no config. */
2105 static void no_device_config(struct device *dev)
2107 dev->mmio_addr = get_mmio_region(dev->mmio_size);
2109 dev->config.bar[0] = dev->mmio_addr;
2110 /* Bottom 4 bits must be zero */
2111 assert(~(dev->config.bar[0] & 0xF));
2114 /* This puts the device config into BAR0 */
2115 static void set_device_config(struct device *dev, const void *conf, size_t len)
2118 dev->mmio_size += len;
2119 dev->mmio = realloc(dev->mmio, dev->mmio_size);
2120 memcpy(dev->mmio + 1, conf, len);
2125 * The device MUST present at least one VIRTIO_PCI_CAP_DEVICE_CFG
2126 * capability for any device type which has a device-specific
2129 /* Hook up device cfg */
2130 dev->config.cfg_access.cap.cap_next
2131 = offsetof(struct pci_config, device);
2136 * The offset for the device-specific configuration MUST be 4-byte
2139 assert(dev->config.cfg_access.cap.cap_next % 4 == 0);
2141 /* Fix up device cfg field length. */
2142 dev->config.device.length = len;
2144 /* The rest is the same as the no-config case */
2145 no_device_config(dev);
2148 static void init_cap(struct virtio_pci_cap *cap, size_t caplen, int type,
2149 size_t bar_offset, size_t bar_bytes, u8 next)
2151 cap->cap_vndr = PCI_CAP_ID_VNDR;
2152 cap->cap_next = next;
2153 cap->cap_len = caplen;
2154 cap->cfg_type = type;
2156 memset(cap->padding, 0, sizeof(cap->padding));
2157 cap->offset = bar_offset;
2158 cap->length = bar_bytes;
2162 * This sets up the pci_config structure, as defined in the virtio 1.0
2163 * standard (and PCI standard).
2165 static void init_pci_config(struct pci_config *pci, u16 type,
2166 u8 class, u8 subclass)
2168 size_t bar_offset, bar_len;
2173 * The device MUST either present notify_off_multiplier as an even
2174 * power of 2, or present notify_off_multiplier as 0.
2176 memset(pci, 0, sizeof(*pci));
2178 /* 4.1.2.1: Devices MUST have the PCI Vendor ID 0x1AF4 */
2179 pci->vendor_id = 0x1AF4;
2180 /* 4.1.2.1: ... PCI Device ID calculated by adding 0x1040 ... */
2181 pci->device_id = 0x1040 + type;
2184 * PCI have specific codes for different types of devices.
2185 * Linux doesn't care, but it's a good clue for people looking
2189 pci->subclass = subclass;
2194 * Non-transitional devices SHOULD have a PCI Revision ID of 1 or
2202 * Non-transitional devices SHOULD have a PCI Subsystem Device ID of
2205 pci->subsystem_device_id = 0x40;
2207 /* We use our dummy interrupt controller, and irq_line is the irq */
2208 pci->irq_line = devices.next_irq++;
2211 /* Support for extended capabilities. */
2212 pci->status = (1 << 4);
2218 * The device MUST present at least one common configuration
2221 pci->capabilities = offsetof(struct pci_config, common);
2223 /* 4.1.4.3.1 ... offset MUST be 4-byte aligned. */
2224 assert(pci->capabilities % 4 == 0);
2226 bar_offset = offsetof(struct virtio_pci_mmio, cfg);
2227 bar_len = sizeof(((struct virtio_pci_mmio *)0)->cfg);
2228 init_cap(&pci->common, sizeof(pci->common), VIRTIO_PCI_CAP_COMMON_CFG,
2229 bar_offset, bar_len,
2230 offsetof(struct pci_config, notify));
2235 * The device MUST present at least one notification capability.
2237 bar_offset += bar_len;
2238 bar_len = sizeof(((struct virtio_pci_mmio *)0)->notify);
2243 * The cap.offset MUST be 2-byte aligned.
2245 assert(pci->common.cap_next % 2 == 0);
2247 /* FIXME: Use a non-zero notify_off, for per-queue notification? */
2251 * The value cap.length presented by the device MUST be at least 2 and
2252 * MUST be large enough to support queue notification offsets for all
2253 * supported queues in all possible configurations.
2255 assert(bar_len >= 2);
2257 init_cap(&pci->notify.cap, sizeof(pci->notify),
2258 VIRTIO_PCI_CAP_NOTIFY_CFG,
2259 bar_offset, bar_len,
2260 offsetof(struct pci_config, isr));
2262 bar_offset += bar_len;
2263 bar_len = sizeof(((struct virtio_pci_mmio *)0)->isr);
2267 * The device MUST present at least one VIRTIO_PCI_CAP_ISR_CFG
2270 init_cap(&pci->isr, sizeof(pci->isr),
2271 VIRTIO_PCI_CAP_ISR_CFG,
2272 bar_offset, bar_len,
2273 offsetof(struct pci_config, cfg_access));
2278 * The device MUST present at least one VIRTIO_PCI_CAP_PCI_CFG
2281 /* This doesn't have any presence in the BAR */
2282 init_cap(&pci->cfg_access.cap, sizeof(pci->cfg_access),
2283 VIRTIO_PCI_CAP_PCI_CFG,
2286 bar_offset += bar_len + sizeof(((struct virtio_pci_mmio *)0)->padding);
2287 assert(bar_offset == sizeof(struct virtio_pci_mmio));
2290 * This gets sewn in and length set in set_device_config().
2291 * Some devices don't have a device configuration interface, so
2292 * we never expose this if we don't call set_device_config().
2294 init_cap(&pci->device, sizeof(pci->device), VIRTIO_PCI_CAP_DEVICE_CFG,
2299 * This routine does all the creation and setup of a new device, but we don't
2300 * actually place the MMIO region until we know the size (if any) of the
2301 * device-specific config. And we don't actually start the service threads
2304 * See what I mean about userspace being boring?
2306 static struct device *new_pci_device(const char *name, u16 type,
2307 u8 class, u8 subclass)
2309 struct device *dev = malloc(sizeof(*dev));
2311 /* Now we populate the fields one at a time. */
2314 dev->running = false;
2315 dev->mmio_size = sizeof(struct virtio_pci_mmio);
2316 dev->mmio = calloc(1, dev->mmio_size);
2317 dev->features = (u64)1 << VIRTIO_F_VERSION_1;
2318 dev->features_accepted = 0;
2320 if (devices.device_num + 1 >= MAX_PCI_DEVICES)
2321 errx(1, "Can only handle 31 PCI devices");
2323 init_pci_config(&dev->config, type, class, subclass);
2324 assert(!devices.pci[devices.device_num+1]);
2325 devices.pci[++devices.device_num] = dev;
2331 * Our first setup routine is the console. It's a fairly simple device, but
2332 * UNIX tty handling makes it uglier than it could be.
2334 static void setup_console(void)
2337 struct virtio_console_config conf;
2339 /* If we can save the initial standard input settings... */
2340 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
2341 struct termios term = orig_term;
2343 * Then we turn off echo, line buffering and ^C etc: We want a
2344 * raw input stream to the Guest.
2346 term.c_lflag &= ~(ISIG|ICANON|ECHO);
2347 tcsetattr(STDIN_FILENO, TCSANOW, &term);
2350 dev = new_pci_device("console", VIRTIO_ID_CONSOLE, 0x07, 0x00);
2352 /* We store the console state in dev->priv, and initialize it. */
2353 dev->priv = malloc(sizeof(struct console_abort));
2354 ((struct console_abort *)dev->priv)->count = 0;
2357 * The console needs two virtqueues: the input then the output. When
2358 * they put something the input queue, we make sure we're listening to
2359 * stdin. When they put something in the output queue, we write it to
2362 add_pci_virtqueue(dev, console_input);
2363 add_pci_virtqueue(dev, console_output);
2365 /* We need a configuration area for the emerg_wr early writes. */
2366 add_pci_feature(dev, VIRTIO_CONSOLE_F_EMERG_WRITE);
2367 set_device_config(dev, &conf, sizeof(conf));
2369 verbose("device %u: console\n", devices.device_num);
2374 * Inter-guest networking is an interesting area. Simplest is to have a
2375 * --sharenet=<name> option which opens or creates a named pipe. This can be
2376 * used to send packets to another guest in a 1:1 manner.
2378 * More sophisticated is to use one of the tools developed for project like UML
2381 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
2382 * completely generic ("here's my vring, attach to your vring") and would work
2383 * for any traffic. Of course, namespace and permissions issues need to be
2384 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
2385 * multiple inter-guest channels behind one interface, although it would
2386 * require some manner of hotplugging new virtio channels.
2388 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
2391 static u32 str2ip(const char *ipaddr)
2395 if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4)
2396 errx(1, "Failed to parse IP address '%s'", ipaddr);
2397 return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3];
2400 static void str2mac(const char *macaddr, unsigned char mac[6])
2403 if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x",
2404 &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6)
2405 errx(1, "Failed to parse mac address '%s'", macaddr);
2415 * This code is "adapted" from libbridge: it attaches the Host end of the
2416 * network device to the bridge device specified by the command line.
2418 * This is yet another James Morris contribution (I'm an IP-level guy, so I
2419 * dislike bridging), and I just try not to break it.
2421 static void add_to_bridge(int fd, const char *if_name, const char *br_name)
2427 errx(1, "must specify bridge name");
2429 ifidx = if_nametoindex(if_name);
2431 errx(1, "interface %s does not exist!", if_name);
2433 strncpy(ifr.ifr_name, br_name, IFNAMSIZ);
2434 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2435 ifr.ifr_ifindex = ifidx;
2436 if (ioctl(fd, SIOCBRADDIF, &ifr) < 0)
2437 err(1, "can't add %s to bridge %s", if_name, br_name);
2441 * This sets up the Host end of the network device with an IP address, brings
2442 * it up so packets will flow, the copies the MAC address into the hwaddr
2445 static void configure_device(int fd, const char *tapif, u32 ipaddr)
2448 struct sockaddr_in sin;
2450 memset(&ifr, 0, sizeof(ifr));
2451 strcpy(ifr.ifr_name, tapif);
2453 /* Don't read these incantations. Just cut & paste them like I did! */
2454 sin.sin_family = AF_INET;
2455 sin.sin_addr.s_addr = htonl(ipaddr);
2456 memcpy(&ifr.ifr_addr, &sin, sizeof(sin));
2457 if (ioctl(fd, SIOCSIFADDR, &ifr) != 0)
2458 err(1, "Setting %s interface address", tapif);
2459 ifr.ifr_flags = IFF_UP;
2460 if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0)
2461 err(1, "Bringing interface %s up", tapif);
2464 static int get_tun_device(char tapif[IFNAMSIZ])
2470 /* Start with this zeroed. Messy but sure. */
2471 memset(&ifr, 0, sizeof(ifr));
2474 * We open the /dev/net/tun device and tell it we want a tap device. A
2475 * tap device is like a tun device, only somehow different. To tell
2476 * the truth, I completely blundered my way through this code, but it
2479 netfd = open_or_die("/dev/net/tun", O_RDWR);
2480 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
2481 strcpy(ifr.ifr_name, "tap%d");
2482 if (ioctl(netfd, TUNSETIFF, &ifr) != 0)
2483 err(1, "configuring /dev/net/tun");
2485 if (ioctl(netfd, TUNSETOFFLOAD,
2486 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
2487 err(1, "Could not set features for tun device");
2490 * We don't need checksums calculated for packets coming in this
2493 ioctl(netfd, TUNSETNOCSUM, 1);
2496 * In virtio before 1.0 (aka legacy virtio), we added a 16-bit
2497 * field at the end of the network header iff
2498 * VIRTIO_NET_F_MRG_RXBUF was negotiated. For virtio 1.0,
2499 * that became the norm, but we need to tell the tun device
2500 * about our expanded header (which is called
2501 * virtio_net_hdr_mrg_rxbuf in the legacy system).
2503 vnet_hdr_sz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2504 if (ioctl(netfd, TUNSETVNETHDRSZ, &vnet_hdr_sz) != 0)
2505 err(1, "Setting tun header size to %u", vnet_hdr_sz);
2507 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
2512 * Our network is a Host<->Guest network. This can either use bridging or
2513 * routing, but the principle is the same: it uses the "tun" device to inject
2514 * packets into the Host as if they came in from a normal network card. We
2515 * just shunt packets between the Guest and the tun device.
2517 static void setup_tun_net(char *arg)
2520 struct net_info *net_info = malloc(sizeof(*net_info));
2522 u32 ip = INADDR_ANY;
2523 bool bridging = false;
2524 char tapif[IFNAMSIZ], *p;
2525 struct virtio_net_config conf;
2527 net_info->tunfd = get_tun_device(tapif);
2529 /* First we create a new network device. */
2530 dev = new_pci_device("net", VIRTIO_ID_NET, 0x02, 0x00);
2531 dev->priv = net_info;
2533 /* Network devices need a recv and a send queue, just like console. */
2534 add_pci_virtqueue(dev, net_input);
2535 add_pci_virtqueue(dev, net_output);
2538 * We need a socket to perform the magic network ioctls to bring up the
2539 * tap interface, connect to the bridge etc. Any socket will do!
2541 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
2543 err(1, "opening IP socket");
2545 /* If the command line was --tunnet=bridge:<name> do bridging. */
2546 if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) {
2547 arg += strlen(BRIDGE_PFX);
2551 /* A mac address may follow the bridge name or IP address */
2552 p = strchr(arg, ':');
2554 str2mac(p+1, conf.mac);
2555 add_pci_feature(dev, VIRTIO_NET_F_MAC);
2559 /* arg is now either an IP address or a bridge name */
2561 add_to_bridge(ipfd, tapif, arg);
2565 /* Set up the tun device. */
2566 configure_device(ipfd, tapif, ip);
2568 /* Expect Guest to handle everything except UFO */
2569 add_pci_feature(dev, VIRTIO_NET_F_CSUM);
2570 add_pci_feature(dev, VIRTIO_NET_F_GUEST_CSUM);
2571 add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO4);
2572 add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO6);
2573 add_pci_feature(dev, VIRTIO_NET_F_GUEST_ECN);
2574 add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO4);
2575 add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO6);
2576 add_pci_feature(dev, VIRTIO_NET_F_HOST_ECN);
2577 /* We handle indirect ring entries */
2578 add_pci_feature(dev, VIRTIO_RING_F_INDIRECT_DESC);
2579 set_device_config(dev, &conf, sizeof(conf));
2581 /* We don't need the socket any more; setup is done. */
2585 verbose("device %u: tun %s attached to bridge: %s\n",
2586 devices.device_num, tapif, arg);
2588 verbose("device %u: tun %s: %s\n",
2589 devices.device_num, tapif, arg);
2593 /* This hangs off device->priv. */
2595 /* The size of the file. */
2598 /* The file descriptor for the file. */
2606 * The disk only has one virtqueue, so it only has one thread. It is really
2607 * simple: the Guest asks for a block number and we read or write that position
2610 * Before we serviced each virtqueue in a separate thread, that was unacceptably
2611 * slow: the Guest waits until the read is finished before running anything
2612 * else, even if it could have been doing useful work.
2614 * We could have used async I/O, except it's reputed to suck so hard that
2615 * characters actually go missing from your code when you try to use it.
2617 static void blk_request(struct virtqueue *vq)
2619 struct vblk_info *vblk = vq->dev->priv;
2620 unsigned int head, out_num, in_num, wlen;
2623 struct virtio_blk_outhdr out;
2624 struct iovec iov[vq->vring.num];
2628 * Get the next request, where we normally wait. It triggers the
2629 * interrupt to acknowledge previously serviced requests (if any).
2631 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
2633 /* Copy the output header from the front of the iov (adjusts iov) */
2634 iov_consume(iov, out_num, &out, sizeof(out));
2636 /* Find and trim end of iov input array, for our status byte. */
2638 for (i = out_num + in_num - 1; i >= out_num; i--) {
2639 if (iov[i].iov_len > 0) {
2640 in = iov[i].iov_base + iov[i].iov_len - 1;
2646 errx(1, "Bad virtblk cmd with no room for status");
2649 * For historical reasons, block operations are expressed in 512 byte
2652 off = out.sector * 512;
2654 if (out.type & VIRTIO_BLK_T_OUT) {
2658 * Move to the right location in the block file. This can fail
2659 * if they try to write past end.
2661 if (lseek64(vblk->fd, off, SEEK_SET) != off)
2662 err(1, "Bad seek to sector %llu", out.sector);
2664 ret = writev(vblk->fd, iov, out_num);
2665 verbose("WRITE to sector %llu: %i\n", out.sector, ret);
2668 * Grr... Now we know how long the descriptor they sent was, we
2669 * make sure they didn't try to write over the end of the block
2670 * file (possibly extending it).
2672 if (ret > 0 && off + ret > vblk->len) {
2673 /* Trim it back to the correct length */
2674 ftruncate64(vblk->fd, vblk->len);
2675 /* Die, bad Guest, die. */
2676 errx(1, "Write past end %llu+%u", off, ret);
2680 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
2681 } else if (out.type & VIRTIO_BLK_T_FLUSH) {
2683 ret = fdatasync(vblk->fd);
2684 verbose("FLUSH fdatasync: %i\n", ret);
2686 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
2691 * Move to the right location in the block file. This can fail
2692 * if they try to read past end.
2694 if (lseek64(vblk->fd, off, SEEK_SET) != off)
2695 err(1, "Bad seek to sector %llu", out.sector);
2697 ret = readv(vblk->fd, iov + out_num, in_num);
2699 wlen = sizeof(*in) + ret;
2700 *in = VIRTIO_BLK_S_OK;
2703 *in = VIRTIO_BLK_S_IOERR;
2707 /* Finished that request. */
2708 add_used(vq, head, wlen);
2711 /*L:198 This actually sets up a virtual block device. */
2712 static void setup_block_file(const char *filename)
2715 struct vblk_info *vblk;
2716 struct virtio_blk_config conf;
2718 /* Create the device. */
2719 dev = new_pci_device("block", VIRTIO_ID_BLOCK, 0x01, 0x80);
2721 /* The device has one virtqueue, where the Guest places requests. */
2722 add_pci_virtqueue(dev, blk_request);
2724 /* Allocate the room for our own bookkeeping */
2725 vblk = dev->priv = malloc(sizeof(*vblk));
2727 /* First we open the file and store the length. */
2728 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
2729 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
2731 /* Tell Guest how many sectors this device has. */
2732 conf.capacity = cpu_to_le64(vblk->len / 512);
2735 * Tell Guest not to put in too many descriptors at once: two are used
2736 * for the in and out elements.
2738 add_pci_feature(dev, VIRTIO_BLK_F_SEG_MAX);
2739 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
2741 set_device_config(dev, &conf, sizeof(struct virtio_blk_config));
2743 verbose("device %u: virtblock %llu sectors\n",
2744 devices.device_num, le64_to_cpu(conf.capacity));
2748 * Our random number generator device reads from /dev/urandom into the Guest's
2749 * input buffers. The usual case is that the Guest doesn't want random numbers
2750 * and so has no buffers although /dev/urandom is still readable, whereas
2751 * console is the reverse.
2753 * The same logic applies, however.
2759 static void rng_input(struct virtqueue *vq)
2762 unsigned int head, in_num, out_num, totlen = 0;
2763 struct rng_info *rng_info = vq->dev->priv;
2764 struct iovec iov[vq->vring.num];
2766 /* First we need a buffer from the Guests's virtqueue. */
2767 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
2769 errx(1, "Output buffers in rng?");
2772 * Just like the console write, we loop to cover the whole iovec.
2773 * In this case, short reads actually happen quite a bit.
2775 while (!iov_empty(iov, in_num)) {
2776 len = readv(rng_info->rfd, iov, in_num);
2778 err(1, "Read from /dev/urandom gave %i", len);
2779 iov_consume(iov, in_num, NULL, len);
2783 /* Tell the Guest about the new input. */
2784 add_used(vq, head, totlen);
2788 * This creates a "hardware" random number device for the Guest.
2790 static void setup_rng(void)
2793 struct rng_info *rng_info = malloc(sizeof(*rng_info));
2795 /* Our device's private info simply contains the /dev/urandom fd. */
2796 rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY);
2798 /* Create the new device. */
2799 dev = new_pci_device("rng", VIRTIO_ID_RNG, 0xff, 0);
2800 dev->priv = rng_info;
2802 /* The device has one virtqueue, where the Guest places inbufs. */
2803 add_pci_virtqueue(dev, rng_input);
2805 /* We don't have any configuration space */
2806 no_device_config(dev);
2808 verbose("device %u: rng\n", devices.device_num);
2810 /* That's the end of device setup. */
2812 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
2813 static void __attribute__((noreturn)) restart_guest(void)
2818 * Since we don't track all open fds, we simply close everything beyond
2821 for (i = 3; i < FD_SETSIZE; i++)
2824 /* Reset all the devices (kills all threads). */
2827 execv(main_args[0], main_args);
2828 err(1, "Could not exec %s", main_args[0]);
2832 * Finally we reach the core of the Launcher which runs the Guest, serves
2833 * its input and output, and finally, lays it to rest.
2835 static void __attribute__((noreturn)) run_guest(void)
2838 struct lguest_pending notify;
2841 /* We read from the /dev/lguest device to run the Guest. */
2842 readval = pread(lguest_fd, ¬ify, sizeof(notify), cpu_id);
2843 if (readval == sizeof(notify)) {
2844 if (notify.trap == 13) {
2845 verbose("Emulating instruction at %#x\n",
2847 emulate_insn(notify.insn);
2848 } else if (notify.trap == 14) {
2849 verbose("Emulating MMIO at %#x\n",
2851 emulate_mmio(notify.addr, notify.insn);
2853 errx(1, "Unknown trap %i addr %#08x\n",
2854 notify.trap, notify.addr);
2855 /* ENOENT means the Guest died. Reading tells us why. */
2856 } else if (errno == ENOENT) {
2857 char reason[1024] = { 0 };
2858 pread(lguest_fd, reason, sizeof(reason)-1, cpu_id);
2859 errx(1, "%s", reason);
2860 /* ERESTART means that we need to reboot the guest */
2861 } else if (errno == ERESTART) {
2863 /* Anything else means a bug or incompatible change. */
2865 err(1, "Running guest failed");
2869 * This is the end of the Launcher. The good news: we are over halfway
2870 * through! The bad news: the most fiendish part of the code still lies ahead
2873 * Are you ready? Take a deep breath and join me in the core of the Host, in
2877 static struct option opts[] = {
2878 { "verbose", 0, NULL, 'v' },
2879 { "tunnet", 1, NULL, 't' },
2880 { "block", 1, NULL, 'b' },
2881 { "rng", 0, NULL, 'r' },
2882 { "initrd", 1, NULL, 'i' },
2883 { "username", 1, NULL, 'u' },
2884 { "chroot", 1, NULL, 'c' },
2887 static void usage(void)
2889 errx(1, "Usage: lguest [--verbose] "
2890 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
2891 "|--block=<filename>|--initrd=<filename>]...\n"
2892 "<mem-in-mb> vmlinux [args...]");
2895 /*L:105 The main routine is where the real work begins: */
2896 int main(int argc, char *argv[])
2898 /* Memory, code startpoint and size of the (optional) initrd. */
2899 unsigned long mem = 0, start, initrd_size = 0;
2900 /* Two temporaries. */
2902 /* The boot information for the Guest. */
2903 struct boot_params *boot;
2904 /* If they specify an initrd file to load. */
2905 const char *initrd_name = NULL;
2907 /* Password structure for initgroups/setres[gu]id */
2908 struct passwd *user_details = NULL;
2910 /* Directory to chroot to */
2911 char *chroot_path = NULL;
2913 /* Save the args: we "reboot" by execing ourselves again. */
2917 * First we initialize the device list. We remember next interrupt
2918 * number to use for devices (1: remember that 0 is used by the timer).
2920 devices.next_irq = 1;
2922 /* We're CPU 0. In fact, that's the only CPU possible right now. */
2926 * We need to know how much memory so we can set up the device
2927 * descriptor and memory pages for the devices as we parse the command
2928 * line. So we quickly look through the arguments to find the amount
2931 for (i = 1; i < argc; i++) {
2932 if (argv[i][0] != '-') {
2933 mem = atoi(argv[i]) * 1024 * 1024;
2935 * We start by mapping anonymous pages over all of
2936 * guest-physical memory range. This fills it with 0,
2937 * and ensures that the Guest won't be killed when it
2938 * tries to access it.
2940 guest_base = map_zeroed_pages(mem / getpagesize()
2943 guest_max = guest_mmio = mem + DEVICE_PAGES*getpagesize();
2948 /* We always have a console device, and it's always device 1. */
2951 /* The options are fairly straight-forward */
2952 while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) {
2958 setup_tun_net(optarg);
2961 setup_block_file(optarg);
2967 initrd_name = optarg;
2970 user_details = getpwnam(optarg);
2972 err(1, "getpwnam failed, incorrect username?");
2975 chroot_path = optarg;
2978 warnx("Unknown argument %s", argv[optind]);
2983 * After the other arguments we expect memory and kernel image name,
2984 * followed by command line arguments for the kernel.
2986 if (optind + 2 > argc)
2989 verbose("Guest base is at %p\n", guest_base);
2991 /* Initialize the (fake) PCI host bridge device. */
2992 init_pci_host_bridge();
2994 /* Now we load the kernel */
2995 start = load_kernel(open_or_die(argv[optind+1], O_RDONLY));
2997 /* Boot information is stashed at physical address 0 */
2998 boot = from_guest_phys(0);
3000 /* Map the initrd image if requested (at top of physical memory) */
3002 initrd_size = load_initrd(initrd_name, mem);
3004 * These are the location in the Linux boot header where the
3005 * start and size of the initrd are expected to be found.
3007 boot->hdr.ramdisk_image = mem - initrd_size;
3008 boot->hdr.ramdisk_size = initrd_size;
3009 /* The bootloader type 0xFF means "unknown"; that's OK. */
3010 boot->hdr.type_of_loader = 0xFF;
3014 * The Linux boot header contains an "E820" memory map: ours is a
3015 * simple, single region.
3017 boot->e820_entries = 1;
3018 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
3020 * The boot header contains a command line pointer: we put the command
3021 * line after the boot header.
3023 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
3024 /* We use a simple helper to copy the arguments separated by spaces. */
3025 concat((char *)(boot + 1), argv+optind+2);
3027 /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
3028 boot->hdr.kernel_alignment = 0x1000000;
3030 /* Boot protocol version: 2.07 supports the fields for lguest. */
3031 boot->hdr.version = 0x207;
3033 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
3034 boot->hdr.hardware_subarch = 1;
3036 /* Tell the entry path not to try to reload segment registers. */
3037 boot->hdr.loadflags |= KEEP_SEGMENTS;
3039 /* We tell the kernel to initialize the Guest. */
3042 /* Ensure that we terminate if a device-servicing child dies. */
3043 signal(SIGCHLD, kill_launcher);
3045 /* If we exit via err(), this kills all the threads, restores tty. */
3046 atexit(cleanup_devices);
3048 /* If requested, chroot to a directory */
3050 if (chroot(chroot_path) != 0)
3051 err(1, "chroot(\"%s\") failed", chroot_path);
3053 if (chdir("/") != 0)
3054 err(1, "chdir(\"/\") failed");
3056 verbose("chroot done\n");
3059 /* If requested, drop privileges */
3064 u = user_details->pw_uid;
3065 g = user_details->pw_gid;
3067 if (initgroups(user_details->pw_name, g) != 0)
3068 err(1, "initgroups failed");
3070 if (setresgid(g, g, g) != 0)
3071 err(1, "setresgid failed");
3073 if (setresuid(u, u, u) != 0)
3074 err(1, "setresuid failed");
3076 verbose("Dropping privileges completed\n");
3079 /* Finally, run the Guest. This doesn't return. */
3085 * Mastery is done: you now know everything I do.
3087 * But surely you have seen code, features and bugs in your wanderings which
3088 * you now yearn to attack? That is the real game, and I look forward to you
3089 * patching and forking lguest into the Your-Name-Here-visor.
3091 * Farewell, and good coding!