2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
7 #define _LARGEFILE64_SOURCE
17 #include <sys/param.h>
18 #include <sys/types.h>
21 #include <sys/eventfd.h>
26 #include <sys/socket.h>
27 #include <sys/ioctl.h>
30 #include <netinet/in.h>
32 #include <linux/sockios.h>
33 #include <linux/if_tun.h>
45 #include <linux/virtio_config.h>
46 #include <linux/virtio_net.h>
47 #include <linux/virtio_blk.h>
48 #include <linux/virtio_console.h>
49 #include <linux/virtio_rng.h>
50 #include <linux/virtio_ring.h>
51 #include <asm/bootparam.h>
52 #include "../../include/linux/lguest_launcher.h"
54 * We can ignore the 43 include files we need for this program, but I do want
55 * to draw attention to the use of kernel-style types.
57 * As Linus said, "C is a Spartan language, and so should your naming be." I
58 * like these abbreviations, so we define them here. Note that u64 is always
59 * unsigned long long, which works on all Linux systems: this means that we can
60 * use %llu in printf for any u64.
62 typedef unsigned long long u64
;
68 #define BRIDGE_PFX "bridge:"
70 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
72 /* We can have up to 256 pages for devices. */
73 #define DEVICE_PAGES 256
74 /* This will occupy 3 pages: it must be a power of 2. */
75 #define VIRTQUEUE_NUM 256
78 * verbose is both a global flag and a macro. The C preprocessor allows
79 * this, and although I wouldn't recommend it, it works quite nicely here.
82 #define verbose(args...) \
83 do { if (verbose) printf(args); } while(0)
86 /* The pointer to the start of guest memory. */
87 static void *guest_base
;
88 /* The maximum guest physical address allowed, and maximum possible. */
89 static unsigned long guest_limit
, guest_max
;
90 /* The /dev/lguest file descriptor. */
93 /* a per-cpu variable indicating whose vcpu is currently running */
94 static unsigned int __thread cpu_id
;
96 /* This is our list of devices. */
98 /* Counter to assign interrupt numbers. */
99 unsigned int next_irq
;
101 /* Counter to print out convenient device numbers. */
102 unsigned int device_num
;
104 /* The descriptor page for the devices. */
107 /* A single linked list of devices. */
109 /* And a pointer to the last device for easy append. */
110 struct device
*lastdev
;
113 /* The list of Guest devices, based on command line arguments. */
114 static struct device_list devices
;
116 /* The device structure describes a single device. */
118 /* The linked-list pointer. */
121 /* The device's descriptor, as mapped into the Guest. */
122 struct lguest_device_desc
*desc
;
124 /* We can't trust desc values once Guest has booted: we use these. */
125 unsigned int feature_len
;
128 /* The name of this device, for --verbose. */
131 /* Any queues attached to this device */
132 struct virtqueue
*vq
;
134 /* Is it operational */
137 /* Device-specific data. */
141 /* The virtqueue structure describes a queue attached to a device. */
143 struct virtqueue
*next
;
145 /* Which device owns me. */
148 /* The configuration for this queue. */
149 struct lguest_vqconfig config
;
151 /* The actual ring of buffers. */
154 /* Last available index we saw. */
157 /* How many are used since we sent last irq? */
158 unsigned int pending_used
;
160 /* Eventfd where Guest notifications arrive. */
163 /* Function for the thread which is servicing this virtqueue. */
164 void (*service
)(struct virtqueue
*vq
);
168 /* Remember the arguments to the program so we can "reboot" */
169 static char **main_args
;
171 /* The original tty settings to restore on exit. */
172 static struct termios orig_term
;
175 * We have to be careful with barriers: our devices are all run in separate
176 * threads and so we need to make sure that changes visible to the Guest happen
179 #define wmb() __asm__ __volatile__("" : : : "memory")
180 #define mb() __asm__ __volatile__("" : : : "memory")
182 /* Wrapper for the last available index. Makes it easier to change. */
183 #define lg_last_avail(vq) ((vq)->last_avail_idx)
186 * The virtio configuration space is defined to be little-endian. x86 is
187 * little-endian too, but it's nice to be explicit so we have these helpers.
189 #define cpu_to_le16(v16) (v16)
190 #define cpu_to_le32(v32) (v32)
191 #define cpu_to_le64(v64) (v64)
192 #define le16_to_cpu(v16) (v16)
193 #define le32_to_cpu(v32) (v32)
194 #define le64_to_cpu(v64) (v64)
196 /* Is this iovec empty? */
197 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
201 for (i
= 0; i
< num_iov
; i
++)
207 /* Take len bytes from the front of this iovec. */
208 static void iov_consume(struct iovec iov
[], unsigned num_iov
,
209 void *dest
, unsigned len
)
213 for (i
= 0; i
< num_iov
; i
++) {
216 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
218 memcpy(dest
, iov
[i
].iov_base
, used
);
221 iov
[i
].iov_base
+= used
;
222 iov
[i
].iov_len
-= used
;
226 errx(1, "iovec too short!");
229 /* The device virtqueue descriptors are followed by feature bitmasks. */
230 static u8
*get_feature_bits(struct device
*dev
)
232 return (u8
*)(dev
->desc
+ 1)
233 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
237 * The Launcher code itself takes us out into userspace, that scary place where
238 * pointers run wild and free! Unfortunately, like most userspace programs,
239 * it's quite boring (which is why everyone likes to hack on the kernel!).
240 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
241 * you through this section. Or, maybe not.
243 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
244 * memory and stores it in "guest_base". In other words, Guest physical ==
245 * Launcher virtual with an offset.
247 * This can be tough to get your head around, but usually it just means that we
248 * use these trivial conversion functions when the Guest gives us its
249 * "physical" addresses:
251 static void *from_guest_phys(unsigned long addr
)
253 return guest_base
+ addr
;
256 static unsigned long to_guest_phys(const void *addr
)
258 return (addr
- guest_base
);
262 * Loading the Kernel.
264 * We start with couple of simple helper routines. open_or_die() avoids
265 * error-checking code cluttering the callers:
267 static int open_or_die(const char *name
, int flags
)
269 int fd
= open(name
, flags
);
271 err(1, "Failed to open %s", name
);
275 /* map_zeroed_pages() takes a number of pages. */
276 static void *map_zeroed_pages(unsigned int num
)
278 int fd
= open_or_die("/dev/zero", O_RDONLY
);
282 * We use a private mapping (ie. if we write to the page, it will be
283 * copied). We allocate an extra two pages PROT_NONE to act as guard
284 * pages against read/write attempts that exceed allocated space.
286 addr
= mmap(NULL
, getpagesize() * (num
+2),
287 PROT_NONE
, MAP_PRIVATE
, fd
, 0);
289 if (addr
== MAP_FAILED
)
290 err(1, "Mmapping %u pages of /dev/zero", num
);
292 if (mprotect(addr
+ getpagesize(), getpagesize() * num
,
293 PROT_READ
|PROT_WRITE
) == -1)
294 err(1, "mprotect rw %u pages failed", num
);
297 * One neat mmap feature is that you can close the fd, and it
302 /* Return address after PROT_NONE page */
303 return addr
+ getpagesize();
306 /* Get some more pages for a device. */
307 static void *get_pages(unsigned int num
)
309 void *addr
= from_guest_phys(guest_limit
);
311 guest_limit
+= num
* getpagesize();
312 if (guest_limit
> guest_max
)
313 errx(1, "Not enough memory for devices");
318 * This routine is used to load the kernel or initrd. It tries mmap, but if
319 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
320 * it falls back to reading the memory in.
322 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
327 * We map writable even though for some segments are marked read-only.
328 * The kernel really wants to be writable: it patches its own
331 * MAP_PRIVATE means that the page won't be copied until a write is
332 * done to it. This allows us to share untouched memory between
335 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
,
336 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
339 /* pread does a seek and a read in one shot: saves a few lines. */
340 r
= pread(fd
, addr
, len
, offset
);
342 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
346 * This routine takes an open vmlinux image, which is in ELF, and maps it into
347 * the Guest memory. ELF = Embedded Linking Format, which is the format used
348 * by all modern binaries on Linux including the kernel.
350 * The ELF headers give *two* addresses: a physical address, and a virtual
351 * address. We use the physical address; the Guest will map itself to the
354 * We return the starting address.
356 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
358 Elf32_Phdr phdr
[ehdr
->e_phnum
];
362 * Sanity checks on the main ELF header: an x86 executable with a
363 * reasonable number of correctly-sized program headers.
365 if (ehdr
->e_type
!= ET_EXEC
366 || ehdr
->e_machine
!= EM_386
367 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
368 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
369 errx(1, "Malformed elf header");
372 * An ELF executable contains an ELF header and a number of "program"
373 * headers which indicate which parts ("segments") of the program to
377 /* We read in all the program headers at once: */
378 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
379 err(1, "Seeking to program headers");
380 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
381 err(1, "Reading program headers");
384 * Try all the headers: there are usually only three. A read-only one,
385 * a read-write one, and a "note" section which we don't load.
387 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
388 /* If this isn't a loadable segment, we ignore it */
389 if (phdr
[i
].p_type
!= PT_LOAD
)
392 verbose("Section %i: size %i addr %p\n",
393 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
395 /* We map this section of the file at its physical address. */
396 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
397 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
400 /* The entry point is given in the ELF header. */
401 return ehdr
->e_entry
;
405 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
406 * to jump into it and it will unpack itself. We used to have to perform some
407 * hairy magic because the unpacking code scared me.
409 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
410 * a small patch to jump over the tricky bits in the Guest, so now we just read
411 * the funky header so we know where in the file to load, and away we go!
413 static unsigned long load_bzimage(int fd
)
415 struct boot_params boot
;
417 /* Modern bzImages get loaded at 1M. */
418 void *p
= from_guest_phys(0x100000);
421 * Go back to the start of the file and read the header. It should be
422 * a Linux boot header (see Documentation/x86/boot.txt)
424 lseek(fd
, 0, SEEK_SET
);
425 read(fd
, &boot
, sizeof(boot
));
427 /* Inside the setup_hdr, we expect the magic "HdrS" */
428 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
429 errx(1, "This doesn't look like a bzImage to me");
431 /* Skip over the extra sectors of the header. */
432 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
434 /* Now read everything into memory. in nice big chunks. */
435 while ((r
= read(fd
, p
, 65536)) > 0)
438 /* Finally, code32_start tells us where to enter the kernel. */
439 return boot
.hdr
.code32_start
;
443 * Loading the kernel is easy when it's a "vmlinux", but most kernels
444 * come wrapped up in the self-decompressing "bzImage" format. With a little
445 * work, we can load those, too.
447 static unsigned long load_kernel(int fd
)
451 /* Read in the first few bytes. */
452 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
453 err(1, "Reading kernel");
455 /* If it's an ELF file, it starts with "\177ELF" */
456 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
457 return map_elf(fd
, &hdr
);
459 /* Otherwise we assume it's a bzImage, and try to load it. */
460 return load_bzimage(fd
);
464 * This is a trivial little helper to align pages. Andi Kleen hated it because
465 * it calls getpagesize() twice: "it's dumb code."
467 * Kernel guys get really het up about optimization, even when it's not
468 * necessary. I leave this code as a reaction against that.
470 static inline unsigned long page_align(unsigned long addr
)
472 /* Add upwards and truncate downwards. */
473 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
477 * An "initial ram disk" is a disk image loaded into memory along with the
478 * kernel which the kernel can use to boot from without needing any drivers.
479 * Most distributions now use this as standard: the initrd contains the code to
480 * load the appropriate driver modules for the current machine.
482 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
483 * kernels. He sent me this (and tells me when I break it).
485 static unsigned long load_initrd(const char *name
, unsigned long mem
)
491 ifd
= open_or_die(name
, O_RDONLY
);
492 /* fstat() is needed to get the file size. */
493 if (fstat(ifd
, &st
) < 0)
494 err(1, "fstat() on initrd '%s'", name
);
497 * We map the initrd at the top of memory, but mmap wants it to be
498 * page-aligned, so we round the size up for that.
500 len
= page_align(st
.st_size
);
501 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
503 * Once a file is mapped, you can close the file descriptor. It's a
504 * little odd, but quite useful.
507 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
509 /* We return the initrd size. */
515 * Simple routine to roll all the commandline arguments together with spaces
518 static void concat(char *dst
, char *args
[])
520 unsigned int i
, len
= 0;
522 for (i
= 0; args
[i
]; i
++) {
524 strcat(dst
+len
, " ");
527 strcpy(dst
+len
, args
[i
]);
528 len
+= strlen(args
[i
]);
530 /* In case it's empty. */
535 * This is where we actually tell the kernel to initialize the Guest. We
536 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
537 * the base of Guest "physical" memory, the top physical page to allow and the
538 * entry point for the Guest.
540 static void tell_kernel(unsigned long start
)
542 unsigned long args
[] = { LHREQ_INITIALIZE
,
543 (unsigned long)guest_base
,
544 guest_limit
/ getpagesize(), start
};
545 verbose("Guest: %p - %p (%#lx)\n",
546 guest_base
, guest_base
+ guest_limit
, guest_limit
);
547 lguest_fd
= open_or_die("/dev/lguest", O_RDWR
);
548 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
549 err(1, "Writing to /dev/lguest");
556 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
557 * We need to make sure it's not trying to reach into the Launcher itself, so
558 * we have a convenient routine which checks it and exits with an error message
559 * if something funny is going on:
561 static void *_check_pointer(unsigned long addr
, unsigned int size
,
565 * Check if the requested address and size exceeds the allocated memory,
566 * or addr + size wraps around.
568 if ((addr
+ size
) > guest_limit
|| (addr
+ size
) < addr
)
569 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
571 * We return a pointer for the caller's convenience, now we know it's
574 return from_guest_phys(addr
);
576 /* A macro which transparently hands the line number to the real function. */
577 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
580 * Each buffer in the virtqueues is actually a chain of descriptors. This
581 * function returns the next descriptor in the chain, or vq->vring.num if we're
584 static unsigned next_desc(struct vring_desc
*desc
,
585 unsigned int i
, unsigned int max
)
589 /* If this descriptor says it doesn't chain, we're done. */
590 if (!(desc
[i
].flags
& VRING_DESC_F_NEXT
))
593 /* Check they're not leading us off end of descriptors. */
595 /* Make sure compiler knows to grab that: we don't want it changing! */
599 errx(1, "Desc next is %u", next
);
605 * This actually sends the interrupt for this virtqueue, if we've used a
608 static void trigger_irq(struct virtqueue
*vq
)
610 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
612 /* Don't inform them if nothing used. */
613 if (!vq
->pending_used
)
615 vq
->pending_used
= 0;
617 /* If they don't want an interrupt, don't send one... */
618 if (vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
) {
622 /* Send the Guest an interrupt tell them we used something up. */
623 if (write(lguest_fd
, buf
, sizeof(buf
)) != 0)
624 err(1, "Triggering irq %i", vq
->config
.irq
);
628 * This looks in the virtqueue for the first available buffer, and converts
629 * it to an iovec for convenient access. Since descriptors consist of some
630 * number of output then some number of input descriptors, it's actually two
631 * iovecs, but we pack them into one and note how many of each there were.
633 * This function waits if necessary, and returns the descriptor number found.
635 static unsigned wait_for_vq_desc(struct virtqueue
*vq
,
637 unsigned int *out_num
, unsigned int *in_num
)
639 unsigned int i
, head
, max
;
640 struct vring_desc
*desc
;
641 u16 last_avail
= lg_last_avail(vq
);
643 /* There's nothing available? */
644 while (last_avail
== vq
->vring
.avail
->idx
) {
648 * Since we're about to sleep, now is a good time to tell the
649 * Guest about what we've used up to now.
653 /* OK, now we need to know about added descriptors. */
654 vq
->vring
.used
->flags
&= ~VRING_USED_F_NO_NOTIFY
;
657 * They could have slipped one in as we were doing that: make
658 * sure it's written, then check again.
661 if (last_avail
!= vq
->vring
.avail
->idx
) {
662 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
666 /* Nothing new? Wait for eventfd to tell us they refilled. */
667 if (read(vq
->eventfd
, &event
, sizeof(event
)) != sizeof(event
))
668 errx(1, "Event read failed?");
670 /* We don't need to be notified again. */
671 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
674 /* Check it isn't doing very strange things with descriptor numbers. */
675 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
676 errx(1, "Guest moved used index from %u to %u",
677 last_avail
, vq
->vring
.avail
->idx
);
680 * Grab the next descriptor number they're advertising, and increment
681 * the index we've seen.
683 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
686 /* If their number is silly, that's a fatal mistake. */
687 if (head
>= vq
->vring
.num
)
688 errx(1, "Guest says index %u is available", head
);
690 /* When we start there are none of either input nor output. */
691 *out_num
= *in_num
= 0;
694 desc
= vq
->vring
.desc
;
698 * If this is an indirect entry, then this buffer contains a descriptor
699 * table which we handle as if it's any normal descriptor chain.
701 if (desc
[i
].flags
& VRING_DESC_F_INDIRECT
) {
702 if (desc
[i
].len
% sizeof(struct vring_desc
))
703 errx(1, "Invalid size for indirect buffer table");
705 max
= desc
[i
].len
/ sizeof(struct vring_desc
);
706 desc
= check_pointer(desc
[i
].addr
, desc
[i
].len
);
711 /* Grab the first descriptor, and check it's OK. */
712 iov
[*out_num
+ *in_num
].iov_len
= desc
[i
].len
;
713 iov
[*out_num
+ *in_num
].iov_base
714 = check_pointer(desc
[i
].addr
, desc
[i
].len
);
715 /* If this is an input descriptor, increment that count. */
716 if (desc
[i
].flags
& VRING_DESC_F_WRITE
)
720 * If it's an output descriptor, they're all supposed
721 * to come before any input descriptors.
724 errx(1, "Descriptor has out after in");
728 /* If we've got too many, that implies a descriptor loop. */
729 if (*out_num
+ *in_num
> max
)
730 errx(1, "Looped descriptor");
731 } while ((i
= next_desc(desc
, i
, max
)) != max
);
737 * After we've used one of their buffers, we tell the Guest about it. Sometime
738 * later we'll want to send them an interrupt using trigger_irq(); note that
739 * wait_for_vq_desc() does that for us if it has to wait.
741 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
743 struct vring_used_elem
*used
;
746 * The virtqueue contains a ring of used buffers. Get a pointer to the
747 * next entry in that used ring.
749 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
752 /* Make sure buffer is written before we update index. */
754 vq
->vring
.used
->idx
++;
758 /* And here's the combo meal deal. Supersize me! */
759 static void add_used_and_trigger(struct virtqueue
*vq
, unsigned head
, int len
)
761 add_used(vq
, head
, len
);
768 * We associate some data with the console for our exit hack.
770 struct console_abort
{
771 /* How many times have they hit ^C? */
773 /* When did they start? */
774 struct timeval start
;
777 /* This is the routine which handles console input (ie. stdin). */
778 static void console_input(struct virtqueue
*vq
)
781 unsigned int head
, in_num
, out_num
;
782 struct console_abort
*abort
= vq
->dev
->priv
;
783 struct iovec iov
[vq
->vring
.num
];
785 /* Make sure there's a descriptor available. */
786 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
788 errx(1, "Output buffers in console in queue?");
790 /* Read into it. This is where we usually wait. */
791 len
= readv(STDIN_FILENO
, iov
, in_num
);
793 /* Ran out of input? */
794 warnx("Failed to get console input, ignoring console.");
796 * For simplicity, dying threads kill the whole Launcher. So
803 /* Tell the Guest we used a buffer. */
804 add_used_and_trigger(vq
, head
, len
);
807 * Three ^C within one second? Exit.
809 * This is such a hack, but works surprisingly well. Each ^C has to
810 * be in a buffer by itself, so they can't be too fast. But we check
811 * that we get three within about a second, so they can't be too
814 if (len
!= 1 || ((char *)iov
[0].iov_base
)[0] != 3) {
820 if (abort
->count
== 1)
821 gettimeofday(&abort
->start
, NULL
);
822 else if (abort
->count
== 3) {
824 gettimeofday(&now
, NULL
);
825 /* Kill all Launcher processes with SIGINT, like normal ^C */
826 if (now
.tv_sec
<= abort
->start
.tv_sec
+1)
832 /* This is the routine which handles console output (ie. stdout). */
833 static void console_output(struct virtqueue
*vq
)
835 unsigned int head
, out
, in
;
836 struct iovec iov
[vq
->vring
.num
];
838 /* We usually wait in here, for the Guest to give us something. */
839 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
841 errx(1, "Input buffers in console output queue?");
843 /* writev can return a partial write, so we loop here. */
844 while (!iov_empty(iov
, out
)) {
845 int len
= writev(STDOUT_FILENO
, iov
, out
);
847 warn("Write to stdout gave %i (%d)", len
, errno
);
850 iov_consume(iov
, out
, NULL
, len
);
854 * We're finished with that buffer: if we're going to sleep,
855 * wait_for_vq_desc() will prod the Guest with an interrupt.
857 add_used(vq
, head
, 0);
863 * Handling output for network is also simple: we get all the output buffers
864 * and write them to /dev/net/tun.
870 static void net_output(struct virtqueue
*vq
)
872 struct net_info
*net_info
= vq
->dev
->priv
;
873 unsigned int head
, out
, in
;
874 struct iovec iov
[vq
->vring
.num
];
876 /* We usually wait in here for the Guest to give us a packet. */
877 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
879 errx(1, "Input buffers in net output queue?");
881 * Send the whole thing through to /dev/net/tun. It expects the exact
882 * same format: what a coincidence!
884 if (writev(net_info
->tunfd
, iov
, out
) < 0)
885 warnx("Write to tun failed (%d)?", errno
);
888 * Done with that one; wait_for_vq_desc() will send the interrupt if
889 * all packets are processed.
891 add_used(vq
, head
, 0);
895 * Handling network input is a bit trickier, because I've tried to optimize it.
897 * First we have a helper routine which tells is if from this file descriptor
898 * (ie. the /dev/net/tun device) will block:
900 static bool will_block(int fd
)
903 struct timeval zero
= { 0, 0 };
906 return select(fd
+1, &fdset
, NULL
, NULL
, &zero
) != 1;
910 * This handles packets coming in from the tun device to our Guest. Like all
911 * service routines, it gets called again as soon as it returns, so you don't
912 * see a while(1) loop here.
914 static void net_input(struct virtqueue
*vq
)
917 unsigned int head
, out
, in
;
918 struct iovec iov
[vq
->vring
.num
];
919 struct net_info
*net_info
= vq
->dev
->priv
;
922 * Get a descriptor to write an incoming packet into. This will also
923 * send an interrupt if they're out of descriptors.
925 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
927 errx(1, "Output buffers in net input queue?");
930 * If it looks like we'll block reading from the tun device, send them
933 if (vq
->pending_used
&& will_block(net_info
->tunfd
))
937 * Read in the packet. This is where we normally wait (when there's no
938 * incoming network traffic).
940 len
= readv(net_info
->tunfd
, iov
, in
);
942 warn("Failed to read from tun (%d).", errno
);
945 * Mark that packet buffer as used, but don't interrupt here. We want
946 * to wait until we've done as much work as we can.
948 add_used(vq
, head
, len
);
952 /* This is the helper to create threads: run the service routine in a loop. */
953 static int do_thread(void *_vq
)
955 struct virtqueue
*vq
= _vq
;
963 * When a child dies, we kill our entire process group with SIGTERM. This
964 * also has the side effect that the shell restores the console for us!
966 static void kill_launcher(int signal
)
971 static void reset_device(struct device
*dev
)
973 struct virtqueue
*vq
;
975 verbose("Resetting device %s\n", dev
->name
);
977 /* Clear any features they've acked. */
978 memset(get_feature_bits(dev
) + dev
->feature_len
, 0, dev
->feature_len
);
980 /* We're going to be explicitly killing threads, so ignore them. */
981 signal(SIGCHLD
, SIG_IGN
);
983 /* Zero out the virtqueues, get rid of their threads */
984 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
985 if (vq
->thread
!= (pid_t
)-1) {
986 kill(vq
->thread
, SIGTERM
);
987 waitpid(vq
->thread
, NULL
, 0);
988 vq
->thread
= (pid_t
)-1;
990 memset(vq
->vring
.desc
, 0,
991 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
992 lg_last_avail(vq
) = 0;
994 dev
->running
= false;
996 /* Now we care if threads die. */
997 signal(SIGCHLD
, (void *)kill_launcher
);
1001 * This actually creates the thread which services the virtqueue for a device.
1003 static void create_thread(struct virtqueue
*vq
)
1006 * Create stack for thread. Since the stack grows upwards, we point
1007 * the stack pointer to the end of this region.
1009 char *stack
= malloc(32768);
1010 unsigned long args
[] = { LHREQ_EVENTFD
,
1011 vq
->config
.pfn
*getpagesize(), 0 };
1013 /* Create a zero-initialized eventfd. */
1014 vq
->eventfd
= eventfd(0, 0);
1015 if (vq
->eventfd
< 0)
1016 err(1, "Creating eventfd");
1017 args
[2] = vq
->eventfd
;
1020 * Attach an eventfd to this virtqueue: it will go off when the Guest
1021 * does an LHCALL_NOTIFY for this vq.
1023 if (write(lguest_fd
, &args
, sizeof(args
)) != 0)
1024 err(1, "Attaching eventfd");
1027 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1028 * we get a signal if it dies.
1030 vq
->thread
= clone(do_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, vq
);
1031 if (vq
->thread
== (pid_t
)-1)
1032 err(1, "Creating clone");
1034 /* We close our local copy now the child has it. */
1038 static void start_device(struct device
*dev
)
1041 struct virtqueue
*vq
;
1043 verbose("Device %s OK: offered", dev
->name
);
1044 for (i
= 0; i
< dev
->feature_len
; i
++)
1045 verbose(" %02x", get_feature_bits(dev
)[i
]);
1046 verbose(", accepted");
1047 for (i
= 0; i
< dev
->feature_len
; i
++)
1048 verbose(" %02x", get_feature_bits(dev
)
1049 [dev
->feature_len
+i
]);
1051 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
1055 dev
->running
= true;
1058 static void cleanup_devices(void)
1062 for (dev
= devices
.dev
; dev
; dev
= dev
->next
)
1065 /* If we saved off the original terminal settings, restore them now. */
1066 if (orig_term
.c_lflag
& (ISIG
|ICANON
|ECHO
))
1067 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
1070 /* When the Guest tells us they updated the status field, we handle it. */
1071 static void update_device_status(struct device
*dev
)
1073 /* A zero status is a reset, otherwise it's a set of flags. */
1074 if (dev
->desc
->status
== 0)
1076 else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
1077 warnx("Device %s configuration FAILED", dev
->name
);
1082 err(1, "Device %s features finalized twice", dev
->name
);
1088 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1089 * particular, it's used to notify us of device status changes during boot.
1091 static void handle_output(unsigned long addr
)
1095 /* Check each device. */
1096 for (i
= devices
.dev
; i
; i
= i
->next
) {
1097 struct virtqueue
*vq
;
1100 * Notifications to device descriptors mean they updated the
1103 if (from_guest_phys(addr
) == i
->desc
) {
1104 update_device_status(i
);
1108 /* Devices should not be used before features are finalized. */
1109 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1110 if (addr
!= vq
->config
.pfn
*getpagesize())
1112 errx(1, "Notification on %s before setup!", i
->name
);
1117 * Early console write is done using notify on a nul-terminated string
1118 * in Guest memory. It's also great for hacking debugging messages
1121 if (addr
>= guest_limit
)
1122 errx(1, "Bad NOTIFY %#lx", addr
);
1124 write(STDOUT_FILENO
, from_guest_phys(addr
),
1125 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1131 * All devices need a descriptor so the Guest knows it exists, and a "struct
1132 * device" so the Launcher can keep track of it. We have common helper
1133 * routines to allocate and manage them.
1137 * The layout of the device page is a "struct lguest_device_desc" followed by a
1138 * number of virtqueue descriptors, then two sets of feature bits, then an
1139 * array of configuration bytes. This routine returns the configuration
1142 static u8
*device_config(const struct device
*dev
)
1144 return (void *)(dev
->desc
+ 1)
1145 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
1146 + dev
->feature_len
* 2;
1150 * This routine allocates a new "struct lguest_device_desc" from descriptor
1151 * table page just above the Guest's normal memory. It returns a pointer to
1154 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1156 struct lguest_device_desc d
= { .type
= type
};
1159 /* Figure out where the next device config is, based on the last one. */
1160 if (devices
.lastdev
)
1161 p
= device_config(devices
.lastdev
)
1162 + devices
.lastdev
->desc
->config_len
;
1164 p
= devices
.descpage
;
1166 /* We only have one page for all the descriptors. */
1167 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1168 errx(1, "Too many devices");
1170 /* p might not be aligned, so we memcpy in. */
1171 return memcpy(p
, &d
, sizeof(d
));
1175 * Each device descriptor is followed by the description of its virtqueues. We
1176 * specify how many descriptors the virtqueue is to have.
1178 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1179 void (*service
)(struct virtqueue
*))
1182 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1185 /* First we need some memory for this virtqueue. */
1186 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
1188 p
= get_pages(pages
);
1190 /* Initialize the virtqueue */
1192 vq
->last_avail_idx
= 0;
1196 * This is the routine the service thread will run, and its Process ID
1197 * once it's running.
1199 vq
->service
= service
;
1200 vq
->thread
= (pid_t
)-1;
1202 /* Initialize the configuration. */
1203 vq
->config
.num
= num_descs
;
1204 vq
->config
.irq
= devices
.next_irq
++;
1205 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1207 /* Initialize the vring. */
1208 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1211 * Append virtqueue to this device's descriptor. We use
1212 * device_config() to get the end of the device's current virtqueues;
1213 * we check that we haven't added any config or feature information
1214 * yet, otherwise we'd be overwriting them.
1216 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1217 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1219 dev
->desc
->num_vq
++;
1221 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1224 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1227 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1232 * The first half of the feature bitmask is for us to advertise features. The
1233 * second half is for the Guest to accept features.
1235 static void add_feature(struct device
*dev
, unsigned bit
)
1237 u8
*features
= get_feature_bits(dev
);
1239 /* We can't extend the feature bits once we've added config bytes */
1240 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1241 assert(dev
->desc
->config_len
== 0);
1242 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1245 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1249 * This routine sets the configuration fields for an existing device's
1250 * descriptor. It only works for the last device, but that's OK because that's
1253 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1255 /* Check we haven't overflowed our single page. */
1256 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1257 errx(1, "Too many devices");
1259 /* Copy in the config information, and store the length. */
1260 memcpy(device_config(dev
), conf
, len
);
1261 dev
->desc
->config_len
= len
;
1263 /* Size must fit in config_len field (8 bits)! */
1264 assert(dev
->desc
->config_len
== len
);
1268 * This routine does all the creation and setup of a new device, including
1269 * calling new_dev_desc() to allocate the descriptor and device memory. We
1270 * don't actually start the service threads until later.
1272 * See what I mean about userspace being boring?
1274 static struct device
*new_device(const char *name
, u16 type
)
1276 struct device
*dev
= malloc(sizeof(*dev
));
1278 /* Now we populate the fields one at a time. */
1279 dev
->desc
= new_dev_desc(type
);
1282 dev
->feature_len
= 0;
1284 dev
->running
= false;
1288 * Append to device list. Prepending to a single-linked list is
1289 * easier, but the user expects the devices to be arranged on the bus
1290 * in command-line order. The first network device on the command line
1291 * is eth0, the first block device /dev/vda, etc.
1293 if (devices
.lastdev
)
1294 devices
.lastdev
->next
= dev
;
1297 devices
.lastdev
= dev
;
1303 * Our first setup routine is the console. It's a fairly simple device, but
1304 * UNIX tty handling makes it uglier than it could be.
1306 static void setup_console(void)
1310 /* If we can save the initial standard input settings... */
1311 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1312 struct termios term
= orig_term
;
1314 * Then we turn off echo, line buffering and ^C etc: We want a
1315 * raw input stream to the Guest.
1317 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1318 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1321 dev
= new_device("console", VIRTIO_ID_CONSOLE
);
1323 /* We store the console state in dev->priv, and initialize it. */
1324 dev
->priv
= malloc(sizeof(struct console_abort
));
1325 ((struct console_abort
*)dev
->priv
)->count
= 0;
1328 * The console needs two virtqueues: the input then the output. When
1329 * they put something the input queue, we make sure we're listening to
1330 * stdin. When they put something in the output queue, we write it to
1333 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_input
);
1334 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_output
);
1336 verbose("device %u: console\n", ++devices
.device_num
);
1341 * Inter-guest networking is an interesting area. Simplest is to have a
1342 * --sharenet=<name> option which opens or creates a named pipe. This can be
1343 * used to send packets to another guest in a 1:1 manner.
1345 * More sophisticated is to use one of the tools developed for project like UML
1348 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1349 * completely generic ("here's my vring, attach to your vring") and would work
1350 * for any traffic. Of course, namespace and permissions issues need to be
1351 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1352 * multiple inter-guest channels behind one interface, although it would
1353 * require some manner of hotplugging new virtio channels.
1355 * Finally, we could use a virtio network switch in the kernel, ie. vhost.
1358 static u32
str2ip(const char *ipaddr
)
1362 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1363 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1364 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1367 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1370 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1371 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1372 errx(1, "Failed to parse mac address '%s'", macaddr
);
1382 * This code is "adapted" from libbridge: it attaches the Host end of the
1383 * network device to the bridge device specified by the command line.
1385 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1386 * dislike bridging), and I just try not to break it.
1388 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1394 errx(1, "must specify bridge name");
1396 ifidx
= if_nametoindex(if_name
);
1398 errx(1, "interface %s does not exist!", if_name
);
1400 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1401 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1402 ifr
.ifr_ifindex
= ifidx
;
1403 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1404 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1408 * This sets up the Host end of the network device with an IP address, brings
1409 * it up so packets will flow, the copies the MAC address into the hwaddr
1412 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1415 struct sockaddr_in sin
;
1417 memset(&ifr
, 0, sizeof(ifr
));
1418 strcpy(ifr
.ifr_name
, tapif
);
1420 /* Don't read these incantations. Just cut & paste them like I did! */
1421 sin
.sin_family
= AF_INET
;
1422 sin
.sin_addr
.s_addr
= htonl(ipaddr
);
1423 memcpy(&ifr
.ifr_addr
, &sin
, sizeof(sin
));
1424 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1425 err(1, "Setting %s interface address", tapif
);
1426 ifr
.ifr_flags
= IFF_UP
;
1427 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1428 err(1, "Bringing interface %s up", tapif
);
1431 static int get_tun_device(char tapif
[IFNAMSIZ
])
1436 /* Start with this zeroed. Messy but sure. */
1437 memset(&ifr
, 0, sizeof(ifr
));
1440 * We open the /dev/net/tun device and tell it we want a tap device. A
1441 * tap device is like a tun device, only somehow different. To tell
1442 * the truth, I completely blundered my way through this code, but it
1445 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1446 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1447 strcpy(ifr
.ifr_name
, "tap%d");
1448 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1449 err(1, "configuring /dev/net/tun");
1451 if (ioctl(netfd
, TUNSETOFFLOAD
,
1452 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1453 err(1, "Could not set features for tun device");
1456 * We don't need checksums calculated for packets coming in this
1459 ioctl(netfd
, TUNSETNOCSUM
, 1);
1461 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1466 * Our network is a Host<->Guest network. This can either use bridging or
1467 * routing, but the principle is the same: it uses the "tun" device to inject
1468 * packets into the Host as if they came in from a normal network card. We
1469 * just shunt packets between the Guest and the tun device.
1471 static void setup_tun_net(char *arg
)
1474 struct net_info
*net_info
= malloc(sizeof(*net_info
));
1476 u32 ip
= INADDR_ANY
;
1477 bool bridging
= false;
1478 char tapif
[IFNAMSIZ
], *p
;
1479 struct virtio_net_config conf
;
1481 net_info
->tunfd
= get_tun_device(tapif
);
1483 /* First we create a new network device. */
1484 dev
= new_device("net", VIRTIO_ID_NET
);
1485 dev
->priv
= net_info
;
1487 /* Network devices need a recv and a send queue, just like console. */
1488 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_input
);
1489 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_output
);
1492 * We need a socket to perform the magic network ioctls to bring up the
1493 * tap interface, connect to the bridge etc. Any socket will do!
1495 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1497 err(1, "opening IP socket");
1499 /* If the command line was --tunnet=bridge:<name> do bridging. */
1500 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1501 arg
+= strlen(BRIDGE_PFX
);
1505 /* A mac address may follow the bridge name or IP address */
1506 p
= strchr(arg
, ':');
1508 str2mac(p
+1, conf
.mac
);
1509 add_feature(dev
, VIRTIO_NET_F_MAC
);
1513 /* arg is now either an IP address or a bridge name */
1515 add_to_bridge(ipfd
, tapif
, arg
);
1519 /* Set up the tun device. */
1520 configure_device(ipfd
, tapif
, ip
);
1522 /* Expect Guest to handle everything except UFO */
1523 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1524 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1525 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1526 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1527 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1528 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1529 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1530 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1531 /* We handle indirect ring entries */
1532 add_feature(dev
, VIRTIO_RING_F_INDIRECT_DESC
);
1533 set_config(dev
, sizeof(conf
), &conf
);
1535 /* We don't need the socket any more; setup is done. */
1538 devices
.device_num
++;
1541 verbose("device %u: tun %s attached to bridge: %s\n",
1542 devices
.device_num
, tapif
, arg
);
1544 verbose("device %u: tun %s: %s\n",
1545 devices
.device_num
, tapif
, arg
);
1549 /* This hangs off device->priv. */
1551 /* The size of the file. */
1554 /* The file descriptor for the file. */
1562 * The disk only has one virtqueue, so it only has one thread. It is really
1563 * simple: the Guest asks for a block number and we read or write that position
1566 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1567 * slow: the Guest waits until the read is finished before running anything
1568 * else, even if it could have been doing useful work.
1570 * We could have used async I/O, except it's reputed to suck so hard that
1571 * characters actually go missing from your code when you try to use it.
1573 static void blk_request(struct virtqueue
*vq
)
1575 struct vblk_info
*vblk
= vq
->dev
->priv
;
1576 unsigned int head
, out_num
, in_num
, wlen
;
1579 struct virtio_blk_outhdr out
;
1580 struct iovec iov
[vq
->vring
.num
];
1584 * Get the next request, where we normally wait. It triggers the
1585 * interrupt to acknowledge previously serviced requests (if any).
1587 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1589 /* Copy the output header from the front of the iov (adjusts iov) */
1590 iov_consume(iov
, out_num
, &out
, sizeof(out
));
1592 /* Find and trim end of iov input array, for our status byte. */
1594 for (i
= out_num
+ in_num
- 1; i
>= out_num
; i
--) {
1595 if (iov
[i
].iov_len
> 0) {
1596 in
= iov
[i
].iov_base
+ iov
[i
].iov_len
- 1;
1602 errx(1, "Bad virtblk cmd with no room for status");
1605 * For historical reasons, block operations are expressed in 512 byte
1608 off
= out
.sector
* 512;
1611 * In general the virtio block driver is allowed to try SCSI commands.
1612 * It'd be nice if we supported eject, for example, but we don't.
1614 if (out
.type
& VIRTIO_BLK_T_SCSI_CMD
) {
1615 fprintf(stderr
, "Scsi commands unsupported\n");
1616 *in
= VIRTIO_BLK_S_UNSUPP
;
1618 } else if (out
.type
& VIRTIO_BLK_T_OUT
) {
1622 * Move to the right location in the block file. This can fail
1623 * if they try to write past end.
1625 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1626 err(1, "Bad seek to sector %llu", out
.sector
);
1628 ret
= writev(vblk
->fd
, iov
, out_num
);
1629 verbose("WRITE to sector %llu: %i\n", out
.sector
, ret
);
1632 * Grr... Now we know how long the descriptor they sent was, we
1633 * make sure they didn't try to write over the end of the block
1634 * file (possibly extending it).
1636 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1637 /* Trim it back to the correct length */
1638 ftruncate64(vblk
->fd
, vblk
->len
);
1639 /* Die, bad Guest, die. */
1640 errx(1, "Write past end %llu+%u", off
, ret
);
1644 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1645 } else if (out
.type
& VIRTIO_BLK_T_FLUSH
) {
1647 ret
= fdatasync(vblk
->fd
);
1648 verbose("FLUSH fdatasync: %i\n", ret
);
1650 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1655 * Move to the right location in the block file. This can fail
1656 * if they try to read past end.
1658 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1659 err(1, "Bad seek to sector %llu", out
.sector
);
1661 ret
= readv(vblk
->fd
, iov
+ out_num
, in_num
);
1663 wlen
= sizeof(*in
) + ret
;
1664 *in
= VIRTIO_BLK_S_OK
;
1667 *in
= VIRTIO_BLK_S_IOERR
;
1671 /* Finished that request. */
1672 add_used(vq
, head
, wlen
);
1675 /*L:198 This actually sets up a virtual block device. */
1676 static void setup_block_file(const char *filename
)
1679 struct vblk_info
*vblk
;
1680 struct virtio_blk_config conf
;
1682 /* Creat the device. */
1683 dev
= new_device("block", VIRTIO_ID_BLOCK
);
1685 /* The device has one virtqueue, where the Guest places requests. */
1686 add_virtqueue(dev
, VIRTQUEUE_NUM
, blk_request
);
1688 /* Allocate the room for our own bookkeeping */
1689 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1691 /* First we open the file and store the length. */
1692 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1693 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1695 /* We support FLUSH. */
1696 add_feature(dev
, VIRTIO_BLK_F_FLUSH
);
1698 /* Tell Guest how many sectors this device has. */
1699 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1702 * Tell Guest not to put in too many descriptors at once: two are used
1703 * for the in and out elements.
1705 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1706 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1708 /* Don't try to put whole struct: we have 8 bit limit. */
1709 set_config(dev
, offsetof(struct virtio_blk_config
, geometry
), &conf
);
1711 verbose("device %u: virtblock %llu sectors\n",
1712 ++devices
.device_num
, le64_to_cpu(conf
.capacity
));
1716 * Our random number generator device reads from /dev/random into the Guest's
1717 * input buffers. The usual case is that the Guest doesn't want random numbers
1718 * and so has no buffers although /dev/random is still readable, whereas
1719 * console is the reverse.
1721 * The same logic applies, however.
1727 static void rng_input(struct virtqueue
*vq
)
1730 unsigned int head
, in_num
, out_num
, totlen
= 0;
1731 struct rng_info
*rng_info
= vq
->dev
->priv
;
1732 struct iovec iov
[vq
->vring
.num
];
1734 /* First we need a buffer from the Guests's virtqueue. */
1735 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1737 errx(1, "Output buffers in rng?");
1740 * Just like the console write, we loop to cover the whole iovec.
1741 * In this case, short reads actually happen quite a bit.
1743 while (!iov_empty(iov
, in_num
)) {
1744 len
= readv(rng_info
->rfd
, iov
, in_num
);
1746 err(1, "Read from /dev/random gave %i", len
);
1747 iov_consume(iov
, in_num
, NULL
, len
);
1751 /* Tell the Guest about the new input. */
1752 add_used(vq
, head
, totlen
);
1756 * This creates a "hardware" random number device for the Guest.
1758 static void setup_rng(void)
1761 struct rng_info
*rng_info
= malloc(sizeof(*rng_info
));
1763 /* Our device's privat info simply contains the /dev/random fd. */
1764 rng_info
->rfd
= open_or_die("/dev/random", O_RDONLY
);
1766 /* Create the new device. */
1767 dev
= new_device("rng", VIRTIO_ID_RNG
);
1768 dev
->priv
= rng_info
;
1770 /* The device has one virtqueue, where the Guest places inbufs. */
1771 add_virtqueue(dev
, VIRTQUEUE_NUM
, rng_input
);
1773 verbose("device %u: rng\n", devices
.device_num
++);
1775 /* That's the end of device setup. */
1777 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1778 static void __attribute__((noreturn
)) restart_guest(void)
1783 * Since we don't track all open fds, we simply close everything beyond
1786 for (i
= 3; i
< FD_SETSIZE
; i
++)
1789 /* Reset all the devices (kills all threads). */
1792 execv(main_args
[0], main_args
);
1793 err(1, "Could not exec %s", main_args
[0]);
1797 * Finally we reach the core of the Launcher which runs the Guest, serves
1798 * its input and output, and finally, lays it to rest.
1800 static void __attribute__((noreturn
)) run_guest(void)
1803 unsigned long notify_addr
;
1806 /* We read from the /dev/lguest device to run the Guest. */
1807 readval
= pread(lguest_fd
, ¬ify_addr
,
1808 sizeof(notify_addr
), cpu_id
);
1810 /* One unsigned long means the Guest did HCALL_NOTIFY */
1811 if (readval
== sizeof(notify_addr
)) {
1812 verbose("Notify on address %#lx\n", notify_addr
);
1813 handle_output(notify_addr
);
1814 /* ENOENT means the Guest died. Reading tells us why. */
1815 } else if (errno
== ENOENT
) {
1816 char reason
[1024] = { 0 };
1817 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1818 errx(1, "%s", reason
);
1819 /* ERESTART means that we need to reboot the guest */
1820 } else if (errno
== ERESTART
) {
1822 /* Anything else means a bug or incompatible change. */
1824 err(1, "Running guest failed");
1828 * This is the end of the Launcher. The good news: we are over halfway
1829 * through! The bad news: the most fiendish part of the code still lies ahead
1832 * Are you ready? Take a deep breath and join me in the core of the Host, in
1836 static struct option opts
[] = {
1837 { "verbose", 0, NULL
, 'v' },
1838 { "tunnet", 1, NULL
, 't' },
1839 { "block", 1, NULL
, 'b' },
1840 { "rng", 0, NULL
, 'r' },
1841 { "initrd", 1, NULL
, 'i' },
1842 { "username", 1, NULL
, 'u' },
1843 { "chroot", 1, NULL
, 'c' },
1846 static void usage(void)
1848 errx(1, "Usage: lguest [--verbose] "
1849 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1850 "|--block=<filename>|--initrd=<filename>]...\n"
1851 "<mem-in-mb> vmlinux [args...]");
1854 /*L:105 The main routine is where the real work begins: */
1855 int main(int argc
, char *argv
[])
1857 /* Memory, code startpoint and size of the (optional) initrd. */
1858 unsigned long mem
= 0, start
, initrd_size
= 0;
1859 /* Two temporaries. */
1861 /* The boot information for the Guest. */
1862 struct boot_params
*boot
;
1863 /* If they specify an initrd file to load. */
1864 const char *initrd_name
= NULL
;
1866 /* Password structure for initgroups/setres[gu]id */
1867 struct passwd
*user_details
= NULL
;
1869 /* Directory to chroot to */
1870 char *chroot_path
= NULL
;
1872 /* Save the args: we "reboot" by execing ourselves again. */
1876 * First we initialize the device list. We keep a pointer to the last
1877 * device, and the next interrupt number to use for devices (1:
1878 * remember that 0 is used by the timer).
1880 devices
.lastdev
= NULL
;
1881 devices
.next_irq
= 1;
1883 /* We're CPU 0. In fact, that's the only CPU possible right now. */
1887 * We need to know how much memory so we can set up the device
1888 * descriptor and memory pages for the devices as we parse the command
1889 * line. So we quickly look through the arguments to find the amount
1892 for (i
= 1; i
< argc
; i
++) {
1893 if (argv
[i
][0] != '-') {
1894 mem
= atoi(argv
[i
]) * 1024 * 1024;
1896 * We start by mapping anonymous pages over all of
1897 * guest-physical memory range. This fills it with 0,
1898 * and ensures that the Guest won't be killed when it
1899 * tries to access it.
1901 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1904 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1905 devices
.descpage
= get_pages(1);
1910 /* The options are fairly straight-forward */
1911 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1917 setup_tun_net(optarg
);
1920 setup_block_file(optarg
);
1926 initrd_name
= optarg
;
1929 user_details
= getpwnam(optarg
);
1931 err(1, "getpwnam failed, incorrect username?");
1934 chroot_path
= optarg
;
1937 warnx("Unknown argument %s", argv
[optind
]);
1942 * After the other arguments we expect memory and kernel image name,
1943 * followed by command line arguments for the kernel.
1945 if (optind
+ 2 > argc
)
1948 verbose("Guest base is at %p\n", guest_base
);
1950 /* We always have a console device */
1953 /* Now we load the kernel */
1954 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1956 /* Boot information is stashed at physical address 0 */
1957 boot
= from_guest_phys(0);
1959 /* Map the initrd image if requested (at top of physical memory) */
1961 initrd_size
= load_initrd(initrd_name
, mem
);
1963 * These are the location in the Linux boot header where the
1964 * start and size of the initrd are expected to be found.
1966 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1967 boot
->hdr
.ramdisk_size
= initrd_size
;
1968 /* The bootloader type 0xFF means "unknown"; that's OK. */
1969 boot
->hdr
.type_of_loader
= 0xFF;
1973 * The Linux boot header contains an "E820" memory map: ours is a
1974 * simple, single region.
1976 boot
->e820_entries
= 1;
1977 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1979 * The boot header contains a command line pointer: we put the command
1980 * line after the boot header.
1982 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1983 /* We use a simple helper to copy the arguments separated by spaces. */
1984 concat((char *)(boot
+ 1), argv
+optind
+2);
1986 /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */
1987 boot
->hdr
.kernel_alignment
= 0x1000000;
1989 /* Boot protocol version: 2.07 supports the fields for lguest. */
1990 boot
->hdr
.version
= 0x207;
1992 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1993 boot
->hdr
.hardware_subarch
= 1;
1995 /* Tell the entry path not to try to reload segment registers. */
1996 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1998 /* We tell the kernel to initialize the Guest. */
2001 /* Ensure that we terminate if a device-servicing child dies. */
2002 signal(SIGCHLD
, kill_launcher
);
2004 /* If we exit via err(), this kills all the threads, restores tty. */
2005 atexit(cleanup_devices
);
2007 /* If requested, chroot to a directory */
2009 if (chroot(chroot_path
) != 0)
2010 err(1, "chroot(\"%s\") failed", chroot_path
);
2012 if (chdir("/") != 0)
2013 err(1, "chdir(\"/\") failed");
2015 verbose("chroot done\n");
2018 /* If requested, drop privileges */
2023 u
= user_details
->pw_uid
;
2024 g
= user_details
->pw_gid
;
2026 if (initgroups(user_details
->pw_name
, g
) != 0)
2027 err(1, "initgroups failed");
2029 if (setresgid(g
, g
, g
) != 0)
2030 err(1, "setresgid failed");
2032 if (setresuid(u
, u
, u
) != 0)
2033 err(1, "setresuid failed");
2035 verbose("Dropping privileges completed\n");
2038 /* Finally, run the Guest. This doesn't return. */
2044 * Mastery is done: you now know everything I do.
2046 * But surely you have seen code, features and bugs in your wanderings which
2047 * you now yearn to attack? That is the real game, and I look forward to you
2048 * patching and forking lguest into the Your-Name-Here-visor.
2050 * Farewell, and good coding!