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>
43 #include "linux/lguest_launcher.h"
44 #include "linux/virtio_config.h"
45 #include "linux/virtio_net.h"
46 #include "linux/virtio_blk.h"
47 #include "linux/virtio_console.h"
48 #include "linux/virtio_rng.h"
49 #include "linux/virtio_ring.h"
50 #include "asm/bootparam.h"
52 * We can ignore the 39 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 PAGE_PRESENT 0x7 /* Present, RW, Execute */
67 #define BRIDGE_PFX "bridge:"
69 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
71 /* We can have up to 256 pages for devices. */
72 #define DEVICE_PAGES 256
73 /* This will occupy 3 pages: it must be a power of 2. */
74 #define VIRTQUEUE_NUM 256
77 * verbose is both a global flag and a macro. The C preprocessor allows
78 * this, and although I wouldn't recommend it, it works quite nicely here.
81 #define verbose(args...) \
82 do { if (verbose) printf(args); } while(0)
85 /* The pointer to the start of guest memory. */
86 static void *guest_base
;
87 /* The maximum guest physical address allowed, and maximum possible. */
88 static unsigned long guest_limit
, guest_max
;
89 /* The /dev/lguest file descriptor. */
92 /* a per-cpu variable indicating whose vcpu is currently running */
93 static unsigned int __thread cpu_id
;
95 /* 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. */
119 /* The linked-list pointer. */
122 /* The device's descriptor, as mapped into the Guest. */
123 struct lguest_device_desc
*desc
;
125 /* We can't trust desc values once Guest has booted: we use these. */
126 unsigned int feature_len
;
129 /* The name of this device, for --verbose. */
132 /* Any queues attached to this device */
133 struct virtqueue
*vq
;
135 /* Is it operational */
138 /* Device-specific data. */
142 /* The virtqueue structure describes a queue attached to a device. */
145 struct virtqueue
*next
;
147 /* Which device owns me. */
150 /* The configuration for this queue. */
151 struct lguest_vqconfig config
;
153 /* The actual ring of buffers. */
156 /* Last available index we saw. */
159 /* How many are used since we sent last irq? */
160 unsigned int pending_used
;
162 /* Eventfd where Guest notifications arrive. */
165 /* Function for the thread which is servicing this virtqueue. */
166 void (*service
)(struct virtqueue
*vq
);
170 /* Remember the arguments to the program so we can "reboot" */
171 static char **main_args
;
173 /* The original tty settings to restore on exit. */
174 static struct termios orig_term
;
177 * We have to be careful with barriers: our devices are all run in separate
178 * threads and so we need to make sure that changes visible to the Guest happen
181 #define wmb() __asm__ __volatile__("" : : : "memory")
182 #define mb() __asm__ __volatile__("" : : : "memory")
185 * Convert an iovec element to the given type.
187 * This is a fairly ugly trick: we need to know the size of the type and
188 * alignment requirement to check the pointer is kosher. It's also nice to
189 * have the name of the type in case we report failure.
191 * Typing those three things all the time is cumbersome and error prone, so we
192 * have a macro which sets them all up and passes to the real function.
194 #define convert(iov, type) \
195 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
197 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
200 if (iov
->iov_len
!= size
)
201 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
202 if ((unsigned long)iov
->iov_base
% align
!= 0)
203 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
204 return iov
->iov_base
;
207 /* Wrapper for the last available index. Makes it easier to change. */
208 #define lg_last_avail(vq) ((vq)->last_avail_idx)
211 * The virtio configuration space is defined to be little-endian. x86 is
212 * little-endian too, but it's nice to be explicit so we have these helpers.
214 #define cpu_to_le16(v16) (v16)
215 #define cpu_to_le32(v32) (v32)
216 #define cpu_to_le64(v64) (v64)
217 #define le16_to_cpu(v16) (v16)
218 #define le32_to_cpu(v32) (v32)
219 #define le64_to_cpu(v64) (v64)
221 /* Is this iovec empty? */
222 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
226 for (i
= 0; i
< num_iov
; i
++)
232 /* Take len bytes from the front of this iovec. */
233 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
237 for (i
= 0; i
< num_iov
; i
++) {
240 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
241 iov
[i
].iov_base
+= used
;
242 iov
[i
].iov_len
-= used
;
248 /* The device virtqueue descriptors are followed by feature bitmasks. */
249 static u8
*get_feature_bits(struct device
*dev
)
251 return (u8
*)(dev
->desc
+ 1)
252 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
256 * The Launcher code itself takes us out into userspace, that scary place where
257 * pointers run wild and free! Unfortunately, like most userspace programs,
258 * it's quite boring (which is why everyone likes to hack on the kernel!).
259 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
260 * you through this section. Or, maybe not.
262 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
263 * memory and stores it in "guest_base". In other words, Guest physical ==
264 * Launcher virtual with an offset.
266 * This can be tough to get your head around, but usually it just means that we
267 * use these trivial conversion functions when the Guest gives us it's
268 * "physical" addresses:
270 static void *from_guest_phys(unsigned long addr
)
272 return guest_base
+ addr
;
275 static unsigned long to_guest_phys(const void *addr
)
277 return (addr
- guest_base
);
281 * Loading the Kernel.
283 * We start with couple of simple helper routines. open_or_die() avoids
284 * error-checking code cluttering the callers:
286 static int open_or_die(const char *name
, int flags
)
288 int fd
= open(name
, flags
);
290 err(1, "Failed to open %s", name
);
294 /* map_zeroed_pages() takes a number of pages. */
295 static void *map_zeroed_pages(unsigned int num
)
297 int fd
= open_or_die("/dev/zero", O_RDONLY
);
301 * We use a private mapping (ie. if we write to the page, it will be
304 addr
= mmap(NULL
, getpagesize() * num
,
305 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
306 if (addr
== MAP_FAILED
)
307 err(1, "Mmaping %u pages of /dev/zero", num
);
313 /* Get some more pages for a device. */
314 static void *get_pages(unsigned int num
)
316 void *addr
= from_guest_phys(guest_limit
);
318 guest_limit
+= num
* getpagesize();
319 if (guest_limit
> guest_max
)
320 errx(1, "Not enough memory for devices");
325 * This routine is used to load the kernel or initrd. It tries mmap, but if
326 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
327 * it falls back to reading the memory in.
329 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
334 * We map writable even though for some segments are marked read-only.
335 * The kernel really wants to be writable: it patches its own
338 * MAP_PRIVATE means that the page won't be copied until a write is
339 * done to it. This allows us to share untouched memory between
342 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
343 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
346 /* pread does a seek and a read in one shot: saves a few lines. */
347 r
= pread(fd
, addr
, len
, offset
);
349 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
353 * This routine takes an open vmlinux image, which is in ELF, and maps it into
354 * the Guest memory. ELF = Embedded Linking Format, which is the format used
355 * by all modern binaries on Linux including the kernel.
357 * The ELF headers give *two* addresses: a physical address, and a virtual
358 * address. We use the physical address; the Guest will map itself to the
361 * We return the starting address.
363 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
365 Elf32_Phdr phdr
[ehdr
->e_phnum
];
369 * Sanity checks on the main ELF header: an x86 executable with a
370 * reasonable number of correctly-sized program headers.
372 if (ehdr
->e_type
!= ET_EXEC
373 || ehdr
->e_machine
!= EM_386
374 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
375 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
376 errx(1, "Malformed elf header");
379 * An ELF executable contains an ELF header and a number of "program"
380 * headers which indicate which parts ("segments") of the program to
384 /* We read in all the program headers at once: */
385 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
386 err(1, "Seeking to program headers");
387 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
388 err(1, "Reading program headers");
391 * Try all the headers: there are usually only three. A read-only one,
392 * a read-write one, and a "note" section which we don't load.
394 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
395 /* If this isn't a loadable segment, we ignore it */
396 if (phdr
[i
].p_type
!= PT_LOAD
)
399 verbose("Section %i: size %i addr %p\n",
400 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
402 /* We map this section of the file at its physical address. */
403 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
404 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
407 /* The entry point is given in the ELF header. */
408 return ehdr
->e_entry
;
412 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
413 * to jump into it and it will unpack itself. We used to have to perform some
414 * hairy magic because the unpacking code scared me.
416 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
417 * a small patch to jump over the tricky bits in the Guest, so now we just read
418 * the funky header so we know where in the file to load, and away we go!
420 static unsigned long load_bzimage(int fd
)
422 struct boot_params boot
;
424 /* Modern bzImages get loaded at 1M. */
425 void *p
= from_guest_phys(0x100000);
428 * Go back to the start of the file and read the header. It should be
429 * a Linux boot header (see Documentation/x86/i386/boot.txt)
431 lseek(fd
, 0, SEEK_SET
);
432 read(fd
, &boot
, sizeof(boot
));
434 /* Inside the setup_hdr, we expect the magic "HdrS" */
435 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
436 errx(1, "This doesn't look like a bzImage to me");
438 /* Skip over the extra sectors of the header. */
439 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
441 /* Now read everything into memory. in nice big chunks. */
442 while ((r
= read(fd
, p
, 65536)) > 0)
445 /* Finally, code32_start tells us where to enter the kernel. */
446 return boot
.hdr
.code32_start
;
450 * Loading the kernel is easy when it's a "vmlinux", but most kernels
451 * come wrapped up in the self-decompressing "bzImage" format. With a little
452 * work, we can load those, too.
454 static unsigned long load_kernel(int fd
)
458 /* Read in the first few bytes. */
459 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
460 err(1, "Reading kernel");
462 /* If it's an ELF file, it starts with "\177ELF" */
463 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
464 return map_elf(fd
, &hdr
);
466 /* Otherwise we assume it's a bzImage, and try to load it. */
467 return load_bzimage(fd
);
471 * This is a trivial little helper to align pages. Andi Kleen hated it because
472 * it calls getpagesize() twice: "it's dumb code."
474 * Kernel guys get really het up about optimization, even when it's not
475 * necessary. I leave this code as a reaction against that.
477 static inline unsigned long page_align(unsigned long addr
)
479 /* Add upwards and truncate downwards. */
480 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
484 * An "initial ram disk" is a disk image loaded into memory along with the
485 * kernel which the kernel can use to boot from without needing any drivers.
486 * Most distributions now use this as standard: the initrd contains the code to
487 * load the appropriate driver modules for the current machine.
489 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
490 * kernels. He sent me this (and tells me when I break it).
492 static unsigned long load_initrd(const char *name
, unsigned long mem
)
498 ifd
= open_or_die(name
, O_RDONLY
);
499 /* fstat() is needed to get the file size. */
500 if (fstat(ifd
, &st
) < 0)
501 err(1, "fstat() on initrd '%s'", name
);
504 * We map the initrd at the top of memory, but mmap wants it to be
505 * page-aligned, so we round the size up for that.
507 len
= page_align(st
.st_size
);
508 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
510 * Once a file is mapped, you can close the file descriptor. It's a
511 * little odd, but quite useful.
514 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
516 /* We return the initrd size. */
522 * Simple routine to roll all the commandline arguments together with spaces
525 static void concat(char *dst
, char *args
[])
527 unsigned int i
, len
= 0;
529 for (i
= 0; args
[i
]; i
++) {
531 strcat(dst
+len
, " ");
534 strcpy(dst
+len
, args
[i
]);
535 len
+= strlen(args
[i
]);
537 /* In case it's empty. */
542 * This is where we actually tell the kernel to initialize the Guest. We
543 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
544 * the base of Guest "physical" memory, the top physical page to allow and the
545 * entry point for the Guest.
547 static void tell_kernel(unsigned long start
)
549 unsigned long args
[] = { LHREQ_INITIALIZE
,
550 (unsigned long)guest_base
,
551 guest_limit
/ getpagesize(), start
};
552 verbose("Guest: %p - %p (%#lx)\n",
553 guest_base
, guest_base
+ guest_limit
, guest_limit
);
554 lguest_fd
= open_or_die("/dev/lguest", O_RDWR
);
555 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
556 err(1, "Writing to /dev/lguest");
563 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
564 * We need to make sure it's not trying to reach into the Launcher itself, so
565 * we have a convenient routine which checks it and exits with an error message
566 * if something funny is going on:
568 static void *_check_pointer(unsigned long addr
, unsigned int size
,
572 * We have to separately check addr and addr+size, because size could
573 * be huge and addr + size might wrap around.
575 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
576 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
578 * We return a pointer for the caller's convenience, now we know it's
581 return from_guest_phys(addr
);
583 /* A macro which transparently hands the line number to the real function. */
584 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
587 * Each buffer in the virtqueues is actually a chain of descriptors. This
588 * function returns the next descriptor in the chain, or vq->vring.num if we're
591 static unsigned next_desc(struct vring_desc
*desc
,
592 unsigned int i
, unsigned int max
)
596 /* If this descriptor says it doesn't chain, we're done. */
597 if (!(desc
[i
].flags
& VRING_DESC_F_NEXT
))
600 /* Check they're not leading us off end of descriptors. */
602 /* Make sure compiler knows to grab that: we don't want it changing! */
606 errx(1, "Desc next is %u", next
);
611 /* This actually sends the interrupt for this virtqueue */
612 static void trigger_irq(struct virtqueue
*vq
)
614 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
616 /* Don't inform them if nothing used. */
617 if (!vq
->pending_used
)
619 vq
->pending_used
= 0;
621 /* If they don't want an interrupt, don't send one, unless empty. */
622 if ((vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
623 && lg_last_avail(vq
) != vq
->vring
.avail
->idx
)
626 /* Send the Guest an interrupt tell them we used something up. */
627 if (write(lguest_fd
, buf
, sizeof(buf
)) != 0)
628 err(1, "Triggering irq %i", vq
->config
.irq
);
632 * This looks in the virtqueue and for the first available buffer, and converts
633 * it to an iovec for convenient access. Since descriptors consist of some
634 * number of output then some number of input descriptors, it's actually two
635 * iovecs, but we pack them into one and note how many of each there were.
637 * This function returns the descriptor number found.
639 static unsigned wait_for_vq_desc(struct virtqueue
*vq
,
641 unsigned int *out_num
, unsigned int *in_num
)
643 unsigned int i
, head
, max
;
644 struct vring_desc
*desc
;
645 u16 last_avail
= lg_last_avail(vq
);
647 while (last_avail
== vq
->vring
.avail
->idx
) {
650 /* OK, tell Guest about progress 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 them about it. We'll then
738 * want to send them an interrupt, using trigger_irq().
740 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
742 struct vring_used_elem
*used
;
745 * The virtqueue contains a ring of used buffers. Get a pointer to the
746 * next entry in that used ring.
748 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
751 /* Make sure buffer is written before we update index. */
753 vq
->vring
.used
->idx
++;
757 /* And here's the combo meal deal. Supersize me! */
758 static void add_used_and_trigger(struct virtqueue
*vq
, unsigned head
, int len
)
760 add_used(vq
, head
, len
);
767 * We associate some data with the console for our exit hack.
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 waiting. */
786 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
788 errx(1, "Output buffers in console in queue?");
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 add_used_and_trigger(vq
, head
, len
);
806 * Three ^C within one second? Exit.
808 * This is such a hack, but works surprisingly well. Each ^C has to
809 * be in a buffer by itself, so they can't be too fast. But we check
810 * that we get three within about a second, so they can't be too
813 if (len
!= 1 || ((char *)iov
[0].iov_base
)[0] != 3) {
819 if (abort
->count
== 1)
820 gettimeofday(&abort
->start
, NULL
);
821 else if (abort
->count
== 3) {
823 gettimeofday(&now
, NULL
);
824 /* Kill all Launcher processes with SIGINT, like normal ^C */
825 if (now
.tv_sec
<= abort
->start
.tv_sec
+1)
831 /* This is the routine which handles console output (ie. stdout). */
832 static void console_output(struct virtqueue
*vq
)
834 unsigned int head
, out
, in
;
835 struct iovec iov
[vq
->vring
.num
];
837 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
839 errx(1, "Input buffers in console output queue?");
840 while (!iov_empty(iov
, out
)) {
841 int len
= writev(STDOUT_FILENO
, iov
, out
);
843 err(1, "Write to stdout gave %i", len
);
844 iov_consume(iov
, out
, len
);
846 add_used(vq
, head
, 0);
852 * Handling output for network is also simple: we get all the output buffers
853 * and write them to /dev/net/tun.
859 static void net_output(struct virtqueue
*vq
)
861 struct net_info
*net_info
= vq
->dev
->priv
;
862 unsigned int head
, out
, in
;
863 struct iovec iov
[vq
->vring
.num
];
865 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
867 errx(1, "Input buffers in net output queue?");
868 if (writev(net_info
->tunfd
, iov
, out
) < 0)
869 errx(1, "Write to tun failed?");
870 add_used(vq
, head
, 0);
873 /* Will reading from this file descriptor block? */
874 static bool will_block(int fd
)
877 struct timeval zero
= { 0, 0 };
880 return select(fd
+1, &fdset
, NULL
, NULL
, &zero
) != 1;
883 /* This handles packets coming in from the tun device to our Guest. */
884 static void net_input(struct virtqueue
*vq
)
887 unsigned int head
, out
, in
;
888 struct iovec iov
[vq
->vring
.num
];
889 struct net_info
*net_info
= vq
->dev
->priv
;
891 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
893 errx(1, "Output buffers in net input queue?");
895 /* Deliver interrupt now, since we're about to sleep. */
896 if (vq
->pending_used
&& will_block(net_info
->tunfd
))
899 len
= readv(net_info
->tunfd
, iov
, in
);
901 err(1, "Failed to read from tun.");
902 add_used(vq
, head
, len
);
905 /* This is the helper to create threads. */
906 static int do_thread(void *_vq
)
908 struct virtqueue
*vq
= _vq
;
916 * When a child dies, we kill our entire process group with SIGTERM. This
917 * also has the side effect that the shell restores the console for us!
919 static void kill_launcher(int signal
)
924 static void reset_device(struct device
*dev
)
926 struct virtqueue
*vq
;
928 verbose("Resetting device %s\n", dev
->name
);
930 /* Clear any features they've acked. */
931 memset(get_feature_bits(dev
) + dev
->feature_len
, 0, dev
->feature_len
);
933 /* We're going to be explicitly killing threads, so ignore them. */
934 signal(SIGCHLD
, SIG_IGN
);
936 /* Zero out the virtqueues, get rid of their threads */
937 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
938 if (vq
->thread
!= (pid_t
)-1) {
939 kill(vq
->thread
, SIGTERM
);
940 waitpid(vq
->thread
, NULL
, 0);
941 vq
->thread
= (pid_t
)-1;
943 memset(vq
->vring
.desc
, 0,
944 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
945 lg_last_avail(vq
) = 0;
947 dev
->running
= false;
949 /* Now we care if threads die. */
950 signal(SIGCHLD
, (void *)kill_launcher
);
953 static void create_thread(struct virtqueue
*vq
)
956 * Create stack for thread and run it. Since the stack grows upwards,
957 * we point the stack pointer to the end of this region.
959 char *stack
= malloc(32768);
960 unsigned long args
[] = { LHREQ_EVENTFD
,
961 vq
->config
.pfn
*getpagesize(), 0 };
963 /* Create a zero-initialized eventfd. */
964 vq
->eventfd
= eventfd(0, 0);
966 err(1, "Creating eventfd");
967 args
[2] = vq
->eventfd
;
969 /* Attach an eventfd to this virtqueue: it will go off
970 * when the Guest does an LHCALL_NOTIFY for this vq. */
971 if (write(lguest_fd
, &args
, sizeof(args
)) != 0)
972 err(1, "Attaching eventfd");
974 /* CLONE_VM: because it has to access the Guest memory, and
975 * SIGCHLD so we get a signal if it dies. */
976 vq
->thread
= clone(do_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, vq
);
977 if (vq
->thread
== (pid_t
)-1)
978 err(1, "Creating clone");
979 /* We close our local copy, now the child has it. */
983 static void start_device(struct device
*dev
)
986 struct virtqueue
*vq
;
988 verbose("Device %s OK: offered", dev
->name
);
989 for (i
= 0; i
< dev
->feature_len
; i
++)
990 verbose(" %02x", get_feature_bits(dev
)[i
]);
991 verbose(", accepted");
992 for (i
= 0; i
< dev
->feature_len
; i
++)
993 verbose(" %02x", get_feature_bits(dev
)
994 [dev
->feature_len
+i
]);
996 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
1000 dev
->running
= true;
1003 static void cleanup_devices(void)
1007 for (dev
= devices
.dev
; dev
; dev
= dev
->next
)
1010 /* If we saved off the original terminal settings, restore them now. */
1011 if (orig_term
.c_lflag
& (ISIG
|ICANON
|ECHO
))
1012 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
1015 /* When the Guest tells us they updated the status field, we handle it. */
1016 static void update_device_status(struct device
*dev
)
1018 /* A zero status is a reset, otherwise it's a set of flags. */
1019 if (dev
->desc
->status
== 0)
1021 else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
1022 warnx("Device %s configuration FAILED", dev
->name
);
1025 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
1031 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1032 static void handle_output(unsigned long addr
)
1036 /* Check each device. */
1037 for (i
= devices
.dev
; i
; i
= i
->next
) {
1038 struct virtqueue
*vq
;
1040 /* Notifications to device descriptors update device status. */
1041 if (from_guest_phys(addr
) == i
->desc
) {
1042 update_device_status(i
);
1046 /* Devices *can* be used before status is set to DRIVER_OK. */
1047 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1048 if (addr
!= vq
->config
.pfn
*getpagesize())
1051 errx(1, "Notification on running %s", i
->name
);
1058 * Early console write is done using notify on a nul-terminated string
1059 * in Guest memory. It's also great for hacking debugging messages
1062 if (addr
>= guest_limit
)
1063 errx(1, "Bad NOTIFY %#lx", addr
);
1065 write(STDOUT_FILENO
, from_guest_phys(addr
),
1066 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1072 * All devices need a descriptor so the Guest knows it exists, and a "struct
1073 * device" so the Launcher can keep track of it. We have common helper
1074 * routines to allocate and manage them.
1078 * The layout of the device page is a "struct lguest_device_desc" followed by a
1079 * number of virtqueue descriptors, then two sets of feature bits, then an
1080 * array of configuration bytes. This routine returns the configuration
1083 static u8
*device_config(const struct device
*dev
)
1085 return (void *)(dev
->desc
+ 1)
1086 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
1087 + dev
->feature_len
* 2;
1091 * This routine allocates a new "struct lguest_device_desc" from descriptor
1092 * table page just above the Guest's normal memory. It returns a pointer to
1095 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1097 struct lguest_device_desc d
= { .type
= type
};
1100 /* Figure out where the next device config is, based on the last one. */
1101 if (devices
.lastdev
)
1102 p
= device_config(devices
.lastdev
)
1103 + devices
.lastdev
->desc
->config_len
;
1105 p
= devices
.descpage
;
1107 /* We only have one page for all the descriptors. */
1108 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1109 errx(1, "Too many devices");
1111 /* p might not be aligned, so we memcpy in. */
1112 return memcpy(p
, &d
, sizeof(d
));
1116 * Each device descriptor is followed by the description of its virtqueues. We
1117 * specify how many descriptors the virtqueue is to have.
1119 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1120 void (*service
)(struct virtqueue
*))
1123 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1126 /* First we need some memory for this virtqueue. */
1127 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
1129 p
= get_pages(pages
);
1131 /* Initialize the virtqueue */
1133 vq
->last_avail_idx
= 0;
1135 vq
->service
= service
;
1136 vq
->thread
= (pid_t
)-1;
1138 /* Initialize the configuration. */
1139 vq
->config
.num
= num_descs
;
1140 vq
->config
.irq
= devices
.next_irq
++;
1141 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1143 /* Initialize the vring. */
1144 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1147 * Append virtqueue to this device's descriptor. We use
1148 * device_config() to get the end of the device's current virtqueues;
1149 * we check that we haven't added any config or feature information
1150 * yet, otherwise we'd be overwriting them.
1152 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1153 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1155 dev
->desc
->num_vq
++;
1157 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1160 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1163 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1168 * The first half of the feature bitmask is for us to advertise features. The
1169 * second half is for the Guest to accept features.
1171 static void add_feature(struct device
*dev
, unsigned bit
)
1173 u8
*features
= get_feature_bits(dev
);
1175 /* We can't extend the feature bits once we've added config bytes */
1176 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1177 assert(dev
->desc
->config_len
== 0);
1178 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1181 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1185 * This routine sets the configuration fields for an existing device's
1186 * descriptor. It only works for the last device, but that's OK because that's
1189 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1191 /* Check we haven't overflowed our single page. */
1192 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1193 errx(1, "Too many devices");
1195 /* Copy in the config information, and store the length. */
1196 memcpy(device_config(dev
), conf
, len
);
1197 dev
->desc
->config_len
= len
;
1199 /* Size must fit in config_len field (8 bits)! */
1200 assert(dev
->desc
->config_len
== len
);
1204 * This routine does all the creation and setup of a new device, including
1205 * calling new_dev_desc() to allocate the descriptor and device memory.
1207 * See what I mean about userspace being boring?
1209 static struct device
*new_device(const char *name
, u16 type
)
1211 struct device
*dev
= malloc(sizeof(*dev
));
1213 /* Now we populate the fields one at a time. */
1214 dev
->desc
= new_dev_desc(type
);
1217 dev
->feature_len
= 0;
1219 dev
->running
= false;
1222 * Append to device list. Prepending to a single-linked list is
1223 * easier, but the user expects the devices to be arranged on the bus
1224 * in command-line order. The first network device on the command line
1225 * is eth0, the first block device /dev/vda, etc.
1227 if (devices
.lastdev
)
1228 devices
.lastdev
->next
= dev
;
1231 devices
.lastdev
= dev
;
1237 * Our first setup routine is the console. It's a fairly simple device, but
1238 * UNIX tty handling makes it uglier than it could be.
1240 static void setup_console(void)
1244 /* If we can save the initial standard input settings... */
1245 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1246 struct termios term
= orig_term
;
1248 * Then we turn off echo, line buffering and ^C etc: We want a
1249 * raw input stream to the Guest.
1251 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1252 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1255 dev
= new_device("console", VIRTIO_ID_CONSOLE
);
1257 /* We store the console state in dev->priv, and initialize it. */
1258 dev
->priv
= malloc(sizeof(struct console_abort
));
1259 ((struct console_abort
*)dev
->priv
)->count
= 0;
1262 * The console needs two virtqueues: the input then the output. When
1263 * they put something the input queue, we make sure we're listening to
1264 * stdin. When they put something in the output queue, we write it to
1267 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_input
);
1268 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_output
);
1270 verbose("device %u: console\n", ++devices
.device_num
);
1275 * Inter-guest networking is an interesting area. Simplest is to have a
1276 * --sharenet=<name> option which opens or creates a named pipe. This can be
1277 * used to send packets to another guest in a 1:1 manner.
1279 * More sopisticated is to use one of the tools developed for project like UML
1282 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1283 * completely generic ("here's my vring, attach to your vring") and would work
1284 * for any traffic. Of course, namespace and permissions issues need to be
1285 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1286 * multiple inter-guest channels behind one interface, although it would
1287 * require some manner of hotplugging new virtio channels.
1289 * Finally, we could implement a virtio network switch in the kernel.
1292 static u32
str2ip(const char *ipaddr
)
1296 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1297 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1298 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1301 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1304 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1305 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1306 errx(1, "Failed to parse mac address '%s'", macaddr
);
1316 * This code is "adapted" from libbridge: it attaches the Host end of the
1317 * network device to the bridge device specified by the command line.
1319 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1320 * dislike bridging), and I just try not to break it.
1322 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1328 errx(1, "must specify bridge name");
1330 ifidx
= if_nametoindex(if_name
);
1332 errx(1, "interface %s does not exist!", if_name
);
1334 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1335 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1336 ifr
.ifr_ifindex
= ifidx
;
1337 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1338 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1342 * This sets up the Host end of the network device with an IP address, brings
1343 * it up so packets will flow, the copies the MAC address into the hwaddr
1346 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1349 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1351 memset(&ifr
, 0, sizeof(ifr
));
1352 strcpy(ifr
.ifr_name
, tapif
);
1354 /* Don't read these incantations. Just cut & paste them like I did! */
1355 sin
->sin_family
= AF_INET
;
1356 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1357 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1358 err(1, "Setting %s interface address", tapif
);
1359 ifr
.ifr_flags
= IFF_UP
;
1360 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1361 err(1, "Bringing interface %s up", tapif
);
1364 static int get_tun_device(char tapif
[IFNAMSIZ
])
1369 /* Start with this zeroed. Messy but sure. */
1370 memset(&ifr
, 0, sizeof(ifr
));
1373 * We open the /dev/net/tun device and tell it we want a tap device. A
1374 * tap device is like a tun device, only somehow different. To tell
1375 * the truth, I completely blundered my way through this code, but it
1378 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1379 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1380 strcpy(ifr
.ifr_name
, "tap%d");
1381 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1382 err(1, "configuring /dev/net/tun");
1384 if (ioctl(netfd
, TUNSETOFFLOAD
,
1385 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1386 err(1, "Could not set features for tun device");
1389 * We don't need checksums calculated for packets coming in this
1392 ioctl(netfd
, TUNSETNOCSUM
, 1);
1394 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1399 * Our network is a Host<->Guest network. This can either use bridging or
1400 * routing, but the principle is the same: it uses the "tun" device to inject
1401 * packets into the Host as if they came in from a normal network card. We
1402 * just shunt packets between the Guest and the tun device.
1404 static void setup_tun_net(char *arg
)
1407 struct net_info
*net_info
= malloc(sizeof(*net_info
));
1409 u32 ip
= INADDR_ANY
;
1410 bool bridging
= false;
1411 char tapif
[IFNAMSIZ
], *p
;
1412 struct virtio_net_config conf
;
1414 net_info
->tunfd
= get_tun_device(tapif
);
1416 /* First we create a new network device. */
1417 dev
= new_device("net", VIRTIO_ID_NET
);
1418 dev
->priv
= net_info
;
1420 /* Network devices need a recv and a send queue, just like console. */
1421 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_input
);
1422 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_output
);
1425 * We need a socket to perform the magic network ioctls to bring up the
1426 * tap interface, connect to the bridge etc. Any socket will do!
1428 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1430 err(1, "opening IP socket");
1432 /* If the command line was --tunnet=bridge:<name> do bridging. */
1433 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1434 arg
+= strlen(BRIDGE_PFX
);
1438 /* A mac address may follow the bridge name or IP address */
1439 p
= strchr(arg
, ':');
1441 str2mac(p
+1, conf
.mac
);
1442 add_feature(dev
, VIRTIO_NET_F_MAC
);
1446 /* arg is now either an IP address or a bridge name */
1448 add_to_bridge(ipfd
, tapif
, arg
);
1452 /* Set up the tun device. */
1453 configure_device(ipfd
, tapif
, ip
);
1455 add_feature(dev
, VIRTIO_F_NOTIFY_ON_EMPTY
);
1456 /* Expect Guest to handle everything except UFO */
1457 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1458 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1459 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1460 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1461 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1462 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1463 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1464 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1465 /* We handle indirect ring entries */
1466 add_feature(dev
, VIRTIO_RING_F_INDIRECT_DESC
);
1467 set_config(dev
, sizeof(conf
), &conf
);
1469 /* We don't need the socket any more; setup is done. */
1472 devices
.device_num
++;
1475 verbose("device %u: tun %s attached to bridge: %s\n",
1476 devices
.device_num
, tapif
, arg
);
1478 verbose("device %u: tun %s: %s\n",
1479 devices
.device_num
, tapif
, arg
);
1483 * Our block (disk) device should be really simple: the Guest asks for a block
1484 * number and we read or write that position in the file. Unfortunately, that
1485 * was amazingly slow: the Guest waits until the read is finished before
1486 * running anything else, even if it could have been doing useful work.
1488 * We could use async I/O, except it's reputed to suck so hard that characters
1489 * actually go missing from your code when you try to use it.
1491 * So this was one reason why lguest now does all virtqueue servicing in
1492 * separate threads: it's more efficient and more like a real device.
1495 /* This hangs off device->priv. */
1498 /* The size of the file. */
1501 /* The file descriptor for the file. */
1504 /* IO thread listens on this file descriptor [0]. */
1507 /* IO thread writes to this file descriptor to mark it done, then
1508 * Launcher triggers interrupt to Guest. */
1515 * Remember that the block device is handled by a separate I/O thread. We head
1516 * straight into the core of that thread here:
1518 static void blk_request(struct virtqueue
*vq
)
1520 struct vblk_info
*vblk
= vq
->dev
->priv
;
1521 unsigned int head
, out_num
, in_num
, wlen
;
1524 struct virtio_blk_outhdr
*out
;
1525 struct iovec iov
[vq
->vring
.num
];
1528 /* Get the next request. */
1529 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1532 * Every block request should contain at least one output buffer
1533 * (detailing the location on disk and the type of request) and one
1534 * input buffer (to hold the result).
1536 if (out_num
== 0 || in_num
== 0)
1537 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1538 head
, out_num
, in_num
);
1540 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1541 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1542 off
= out
->sector
* 512;
1545 * The block device implements "barriers", where the Guest indicates
1546 * that it wants all previous writes to occur before this write. We
1547 * don't have a way of asking our kernel to do a barrier, so we just
1548 * synchronize all the data in the file. Pretty poor, no?
1550 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1551 fdatasync(vblk
->fd
);
1554 * In general the virtio block driver is allowed to try SCSI commands.
1555 * It'd be nice if we supported eject, for example, but we don't.
1557 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1558 fprintf(stderr
, "Scsi commands unsupported\n");
1559 *in
= VIRTIO_BLK_S_UNSUPP
;
1561 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1565 * Move to the right location in the block file. This can fail
1566 * if they try to write past end.
1568 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1569 err(1, "Bad seek to sector %llu", out
->sector
);
1571 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1572 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1575 * Grr... Now we know how long the descriptor they sent was, we
1576 * make sure they didn't try to write over the end of the block
1577 * file (possibly extending it).
1579 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1580 /* Trim it back to the correct length */
1581 ftruncate64(vblk
->fd
, vblk
->len
);
1582 /* Die, bad Guest, die. */
1583 errx(1, "Write past end %llu+%u", off
, ret
);
1586 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1591 * Move to the right location in the block file. This can fail
1592 * if they try to read past end.
1594 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1595 err(1, "Bad seek to sector %llu", out
->sector
);
1597 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1598 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1600 wlen
= sizeof(*in
) + ret
;
1601 *in
= VIRTIO_BLK_S_OK
;
1604 *in
= VIRTIO_BLK_S_IOERR
;
1609 * OK, so we noted that it was pretty poor to use an fdatasync as a
1610 * barrier. But Christoph Hellwig points out that we need a sync
1611 * *afterwards* as well: "Barriers specify no reordering to the front
1612 * or the back." And Jens Axboe confirmed it, so here we are:
1614 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1615 fdatasync(vblk
->fd
);
1617 add_used(vq
, head
, wlen
);
1620 /*L:198 This actually sets up a virtual block device. */
1621 static void setup_block_file(const char *filename
)
1624 struct vblk_info
*vblk
;
1625 struct virtio_blk_config conf
;
1627 /* Creat the device. */
1628 dev
= new_device("block", VIRTIO_ID_BLOCK
);
1630 /* The device has one virtqueue, where the Guest places requests. */
1631 add_virtqueue(dev
, VIRTQUEUE_NUM
, blk_request
);
1633 /* Allocate the room for our own bookkeeping */
1634 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1636 /* First we open the file and store the length. */
1637 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1638 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1640 /* We support barriers. */
1641 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1643 /* Tell Guest how many sectors this device has. */
1644 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1647 * Tell Guest not to put in too many descriptors at once: two are used
1648 * for the in and out elements.
1650 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1651 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1653 /* Don't try to put whole struct: we have 8 bit limit. */
1654 set_config(dev
, offsetof(struct virtio_blk_config
, geometry
), &conf
);
1656 verbose("device %u: virtblock %llu sectors\n",
1657 ++devices
.device_num
, le64_to_cpu(conf
.capacity
));
1661 * Our random number generator device reads from /dev/random into the Guest's
1662 * input buffers. The usual case is that the Guest doesn't want random numbers
1663 * and so has no buffers although /dev/random is still readable, whereas
1664 * console is the reverse.
1666 * The same logic applies, however.
1672 static void rng_input(struct virtqueue
*vq
)
1675 unsigned int head
, in_num
, out_num
, totlen
= 0;
1676 struct rng_info
*rng_info
= vq
->dev
->priv
;
1677 struct iovec iov
[vq
->vring
.num
];
1679 /* First we need a buffer from the Guests's virtqueue. */
1680 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1682 errx(1, "Output buffers in rng?");
1685 * This is why we convert to iovecs: the readv() call uses them, and so
1686 * it reads straight into the Guest's buffer. We loop to make sure we
1689 while (!iov_empty(iov
, in_num
)) {
1690 len
= readv(rng_info
->rfd
, iov
, in_num
);
1692 err(1, "Read from /dev/random gave %i", len
);
1693 iov_consume(iov
, in_num
, len
);
1697 /* Tell the Guest about the new input. */
1698 add_used(vq
, head
, totlen
);
1702 * This creates a "hardware" random number device for the Guest.
1704 static void setup_rng(void)
1707 struct rng_info
*rng_info
= malloc(sizeof(*rng_info
));
1709 /* Our device's privat info simply contains the /dev/random fd. */
1710 rng_info
->rfd
= open_or_die("/dev/random", O_RDONLY
);
1712 /* Create the new device. */
1713 dev
= new_device("rng", VIRTIO_ID_RNG
);
1714 dev
->priv
= rng_info
;
1716 /* The device has one virtqueue, where the Guest places inbufs. */
1717 add_virtqueue(dev
, VIRTQUEUE_NUM
, rng_input
);
1719 verbose("device %u: rng\n", devices
.device_num
++);
1721 /* That's the end of device setup. */
1723 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1724 static void __attribute__((noreturn
)) restart_guest(void)
1729 * Since we don't track all open fds, we simply close everything beyond
1732 for (i
= 3; i
< FD_SETSIZE
; i
++)
1735 /* Reset all the devices (kills all threads). */
1738 execv(main_args
[0], main_args
);
1739 err(1, "Could not exec %s", main_args
[0]);
1743 * Finally we reach the core of the Launcher which runs the Guest, serves
1744 * its input and output, and finally, lays it to rest.
1746 static void __attribute__((noreturn
)) run_guest(void)
1749 unsigned long notify_addr
;
1752 /* We read from the /dev/lguest device to run the Guest. */
1753 readval
= pread(lguest_fd
, ¬ify_addr
,
1754 sizeof(notify_addr
), cpu_id
);
1756 /* One unsigned long means the Guest did HCALL_NOTIFY */
1757 if (readval
== sizeof(notify_addr
)) {
1758 verbose("Notify on address %#lx\n", notify_addr
);
1759 handle_output(notify_addr
);
1760 /* ENOENT means the Guest died. Reading tells us why. */
1761 } else if (errno
== ENOENT
) {
1762 char reason
[1024] = { 0 };
1763 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1764 errx(1, "%s", reason
);
1765 /* ERESTART means that we need to reboot the guest */
1766 } else if (errno
== ERESTART
) {
1768 /* Anything else means a bug or incompatible change. */
1770 err(1, "Running guest failed");
1774 * This is the end of the Launcher. The good news: we are over halfway
1775 * through! The bad news: the most fiendish part of the code still lies ahead
1778 * Are you ready? Take a deep breath and join me in the core of the Host, in
1782 static struct option opts
[] = {
1783 { "verbose", 0, NULL
, 'v' },
1784 { "tunnet", 1, NULL
, 't' },
1785 { "block", 1, NULL
, 'b' },
1786 { "rng", 0, NULL
, 'r' },
1787 { "initrd", 1, NULL
, 'i' },
1790 static void usage(void)
1792 errx(1, "Usage: lguest [--verbose] "
1793 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1794 "|--block=<filename>|--initrd=<filename>]...\n"
1795 "<mem-in-mb> vmlinux [args...]");
1798 /*L:105 The main routine is where the real work begins: */
1799 int main(int argc
, char *argv
[])
1801 /* Memory, code startpoint and size of the (optional) initrd. */
1802 unsigned long mem
= 0, start
, initrd_size
= 0;
1803 /* Two temporaries. */
1805 /* The boot information for the Guest. */
1806 struct boot_params
*boot
;
1807 /* If they specify an initrd file to load. */
1808 const char *initrd_name
= NULL
;
1810 /* Save the args: we "reboot" by execing ourselves again. */
1814 * First we initialize the device list. We keep a pointer to the last
1815 * device, and the next interrupt number to use for devices (1:
1816 * remember that 0 is used by the timer).
1818 devices
.lastdev
= NULL
;
1819 devices
.next_irq
= 1;
1823 * We need to know how much memory so we can set up the device
1824 * descriptor and memory pages for the devices as we parse the command
1825 * line. So we quickly look through the arguments to find the amount
1828 for (i
= 1; i
< argc
; i
++) {
1829 if (argv
[i
][0] != '-') {
1830 mem
= atoi(argv
[i
]) * 1024 * 1024;
1832 * We start by mapping anonymous pages over all of
1833 * guest-physical memory range. This fills it with 0,
1834 * and ensures that the Guest won't be killed when it
1835 * tries to access it.
1837 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1840 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1841 devices
.descpage
= get_pages(1);
1846 /* The options are fairly straight-forward */
1847 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1853 setup_tun_net(optarg
);
1856 setup_block_file(optarg
);
1862 initrd_name
= optarg
;
1865 warnx("Unknown argument %s", argv
[optind
]);
1870 * After the other arguments we expect memory and kernel image name,
1871 * followed by command line arguments for the kernel.
1873 if (optind
+ 2 > argc
)
1876 verbose("Guest base is at %p\n", guest_base
);
1878 /* We always have a console device */
1881 /* Now we load the kernel */
1882 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1884 /* Boot information is stashed at physical address 0 */
1885 boot
= from_guest_phys(0);
1887 /* Map the initrd image if requested (at top of physical memory) */
1889 initrd_size
= load_initrd(initrd_name
, mem
);
1891 * These are the location in the Linux boot header where the
1892 * start and size of the initrd are expected to be found.
1894 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1895 boot
->hdr
.ramdisk_size
= initrd_size
;
1896 /* The bootloader type 0xFF means "unknown"; that's OK. */
1897 boot
->hdr
.type_of_loader
= 0xFF;
1901 * The Linux boot header contains an "E820" memory map: ours is a
1902 * simple, single region.
1904 boot
->e820_entries
= 1;
1905 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1907 * The boot header contains a command line pointer: we put the command
1908 * line after the boot header.
1910 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1911 /* We use a simple helper to copy the arguments separated by spaces. */
1912 concat((char *)(boot
+ 1), argv
+optind
+2);
1914 /* Boot protocol version: 2.07 supports the fields for lguest. */
1915 boot
->hdr
.version
= 0x207;
1917 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1918 boot
->hdr
.hardware_subarch
= 1;
1920 /* Tell the entry path not to try to reload segment registers. */
1921 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1924 * We tell the kernel to initialize the Guest: this returns the open
1925 * /dev/lguest file descriptor.
1929 /* Ensure that we terminate if a child dies. */
1930 signal(SIGCHLD
, kill_launcher
);
1932 /* If we exit via err(), this kills all the threads, restores tty. */
1933 atexit(cleanup_devices
);
1935 /* Finally, run the Guest. This doesn't return. */
1941 * Mastery is done: you now know everything I do.
1943 * But surely you have seen code, features and bugs in your wanderings which
1944 * you now yearn to attack? That is the real game, and I look forward to you
1945 * patching and forking lguest into the Your-Name-Here-visor.
1947 * Farewell, and good coding!