1 /*P:100 This is the Launcher code, a simple program which lays out the
2 * "physical" memory for the new Guest by mapping the kernel image and
3 * the virtual devices, then opens /dev/lguest to tell the kernel
4 * about the Guest and control it. :*/
5 #define _LARGEFILE64_SOURCE
15 #include <sys/param.h>
16 #include <sys/types.h>
19 #include <sys/eventfd.h>
24 #include <sys/socket.h>
25 #include <sys/ioctl.h>
28 #include <netinet/in.h>
30 #include <linux/sockios.h>
31 #include <linux/if_tun.h>
41 #include "linux/lguest_launcher.h"
42 #include "linux/virtio_config.h"
43 #include "linux/virtio_net.h"
44 #include "linux/virtio_blk.h"
45 #include "linux/virtio_console.h"
46 #include "linux/virtio_rng.h"
47 #include "linux/virtio_ring.h"
48 #include "asm/bootparam.h"
49 /*L:110 We can ignore the 39 include files we need for this program, but I do
50 * want to draw attention to the use of kernel-style types.
52 * As Linus said, "C is a Spartan language, and so should your naming be." I
53 * like these abbreviations, so we define them here. Note that u64 is always
54 * unsigned long long, which works on all Linux systems: this means that we can
55 * use %llu in printf for any u64. */
56 typedef unsigned long long u64
;
62 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
63 #define BRIDGE_PFX "bridge:"
65 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
67 /* We can have up to 256 pages for devices. */
68 #define DEVICE_PAGES 256
69 /* This will occupy 3 pages: it must be a power of 2. */
70 #define VIRTQUEUE_NUM 256
72 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
73 * this, and although I wouldn't recommend it, it works quite nicely here. */
75 #define verbose(args...) \
76 do { if (verbose) printf(args); } while(0)
79 /* The pointer to the start of guest memory. */
80 static void *guest_base
;
81 /* The maximum guest physical address allowed, and maximum possible. */
82 static unsigned long guest_limit
, guest_max
;
83 /* The /dev/lguest file descriptor. */
86 /* a per-cpu variable indicating whose vcpu is currently running */
87 static unsigned int __thread cpu_id
;
89 /* This is our list of devices. */
92 /* Counter to assign interrupt numbers. */
93 unsigned int next_irq
;
95 /* Counter to print out convenient device numbers. */
96 unsigned int device_num
;
98 /* The descriptor page for the devices. */
101 /* A single linked list of devices. */
103 /* And a pointer to the last device for easy append and also for
104 * configuration appending. */
105 struct device
*lastdev
;
108 /* The list of Guest devices, based on command line arguments. */
109 static struct device_list devices
;
111 /* The device structure describes a single device. */
114 /* The linked-list pointer. */
117 /* The device's descriptor, as mapped into the Guest. */
118 struct lguest_device_desc
*desc
;
120 /* We can't trust desc values once Guest has booted: we use these. */
121 unsigned int feature_len
;
124 /* The name of this device, for --verbose. */
127 /* Any queues attached to this device */
128 struct virtqueue
*vq
;
130 /* Is it operational */
133 /* Device-specific data. */
137 /* The virtqueue structure describes a queue attached to a device. */
140 struct virtqueue
*next
;
142 /* Which device owns me. */
145 /* The configuration for this queue. */
146 struct lguest_vqconfig config
;
148 /* The actual ring of buffers. */
151 /* Last available index we saw. */
154 /* Eventfd where Guest notifications arrive. */
157 /* Function for the thread which is servicing this virtqueue. */
158 void (*service
)(struct virtqueue
*vq
);
162 /* Remember the arguments to the program so we can "reboot" */
163 static char **main_args
;
165 /* The original tty settings to restore on exit. */
166 static struct termios orig_term
;
168 /* We have to be careful with barriers: our devices are all run in separate
169 * threads and so we need to make sure that changes visible to the Guest happen
170 * in precise order. */
171 #define wmb() __asm__ __volatile__("" : : : "memory")
173 /* Convert an iovec element to the given type.
175 * This is a fairly ugly trick: we need to know the size of the type and
176 * alignment requirement to check the pointer is kosher. It's also nice to
177 * have the name of the type in case we report failure.
179 * Typing those three things all the time is cumbersome and error prone, so we
180 * have a macro which sets them all up and passes to the real function. */
181 #define convert(iov, type) \
182 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
184 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
187 if (iov
->iov_len
!= size
)
188 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
189 if ((unsigned long)iov
->iov_base
% align
!= 0)
190 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
191 return iov
->iov_base
;
194 /* Wrapper for the last available index. Makes it easier to change. */
195 #define lg_last_avail(vq) ((vq)->last_avail_idx)
197 /* The virtio configuration space is defined to be little-endian. x86 is
198 * little-endian too, but it's nice to be explicit so we have these helpers. */
199 #define cpu_to_le16(v16) (v16)
200 #define cpu_to_le32(v32) (v32)
201 #define cpu_to_le64(v64) (v64)
202 #define le16_to_cpu(v16) (v16)
203 #define le32_to_cpu(v32) (v32)
204 #define le64_to_cpu(v64) (v64)
206 /* Is this iovec empty? */
207 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
211 for (i
= 0; i
< num_iov
; i
++)
217 /* Take len bytes from the front of this iovec. */
218 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
222 for (i
= 0; i
< num_iov
; i
++) {
225 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
226 iov
[i
].iov_base
+= used
;
227 iov
[i
].iov_len
-= used
;
233 /* The device virtqueue descriptors are followed by feature bitmasks. */
234 static u8
*get_feature_bits(struct device
*dev
)
236 return (u8
*)(dev
->desc
+ 1)
237 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
240 /*L:100 The Launcher code itself takes us out into userspace, that scary place
241 * where pointers run wild and free! Unfortunately, like most userspace
242 * programs, it's quite boring (which is why everyone likes to hack on the
243 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
244 * will get you through this section. Or, maybe not.
246 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
247 * memory and stores it in "guest_base". In other words, Guest physical ==
248 * Launcher virtual with an offset.
250 * This can be tough to get your head around, but usually it just means that we
251 * use these trivial conversion functions when the Guest gives us it's
252 * "physical" addresses: */
253 static void *from_guest_phys(unsigned long addr
)
255 return guest_base
+ addr
;
258 static unsigned long to_guest_phys(const void *addr
)
260 return (addr
- guest_base
);
264 * Loading the Kernel.
266 * We start with couple of simple helper routines. open_or_die() avoids
267 * error-checking code cluttering the callers: */
268 static int open_or_die(const char *name
, int flags
)
270 int fd
= open(name
, flags
);
272 err(1, "Failed to open %s", name
);
276 /* map_zeroed_pages() takes a number of pages. */
277 static void *map_zeroed_pages(unsigned int num
)
279 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
284 addr
= mmap(NULL
, getpagesize() * num
,
285 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
286 if (addr
== MAP_FAILED
)
287 err(1, "Mmaping %u pages of /dev/zero", num
);
293 /* Get some more pages for a device. */
294 static void *get_pages(unsigned int num
)
296 void *addr
= from_guest_phys(guest_limit
);
298 guest_limit
+= num
* getpagesize();
299 if (guest_limit
> guest_max
)
300 errx(1, "Not enough memory for devices");
304 /* This routine is used to load the kernel or initrd. It tries mmap, but if
305 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
306 * it falls back to reading the memory in. */
307 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
311 /* We map writable even though for some segments are marked read-only.
312 * The kernel really wants to be writable: it patches its own
315 * MAP_PRIVATE means that the page won't be copied until a write is
316 * done to it. This allows us to share untouched memory between
318 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
319 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
322 /* pread does a seek and a read in one shot: saves a few lines. */
323 r
= pread(fd
, addr
, len
, offset
);
325 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
328 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
329 * the Guest memory. ELF = Embedded Linking Format, which is the format used
330 * by all modern binaries on Linux including the kernel.
332 * The ELF headers give *two* addresses: a physical address, and a virtual
333 * address. We use the physical address; the Guest will map itself to the
336 * We return the starting address. */
337 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
339 Elf32_Phdr phdr
[ehdr
->e_phnum
];
342 /* Sanity checks on the main ELF header: an x86 executable with a
343 * reasonable number of correctly-sized program headers. */
344 if (ehdr
->e_type
!= ET_EXEC
345 || ehdr
->e_machine
!= EM_386
346 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
347 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
348 errx(1, "Malformed elf header");
350 /* An ELF executable contains an ELF header and a number of "program"
351 * headers which indicate which parts ("segments") of the program to
354 /* We read in all the program headers at once: */
355 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
356 err(1, "Seeking to program headers");
357 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
358 err(1, "Reading program headers");
360 /* Try all the headers: there are usually only three. A read-only one,
361 * a read-write one, and a "note" section which we don't load. */
362 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
363 /* If this isn't a loadable segment, we ignore it */
364 if (phdr
[i
].p_type
!= PT_LOAD
)
367 verbose("Section %i: size %i addr %p\n",
368 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
370 /* We map this section of the file at its physical address. */
371 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
372 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
375 /* The entry point is given in the ELF header. */
376 return ehdr
->e_entry
;
379 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
380 * supposed to jump into it and it will unpack itself. We used to have to
381 * perform some hairy magic because the unpacking code scared me.
383 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
384 * a small patch to jump over the tricky bits in the Guest, so now we just read
385 * the funky header so we know where in the file to load, and away we go! */
386 static unsigned long load_bzimage(int fd
)
388 struct boot_params boot
;
390 /* Modern bzImages get loaded at 1M. */
391 void *p
= from_guest_phys(0x100000);
393 /* Go back to the start of the file and read the header. It should be
394 * a Linux boot header (see Documentation/x86/i386/boot.txt) */
395 lseek(fd
, 0, SEEK_SET
);
396 read(fd
, &boot
, sizeof(boot
));
398 /* Inside the setup_hdr, we expect the magic "HdrS" */
399 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
400 errx(1, "This doesn't look like a bzImage to me");
402 /* Skip over the extra sectors of the header. */
403 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
405 /* Now read everything into memory. in nice big chunks. */
406 while ((r
= read(fd
, p
, 65536)) > 0)
409 /* Finally, code32_start tells us where to enter the kernel. */
410 return boot
.hdr
.code32_start
;
413 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
414 * come wrapped up in the self-decompressing "bzImage" format. With a little
415 * work, we can load those, too. */
416 static unsigned long load_kernel(int fd
)
420 /* Read in the first few bytes. */
421 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
422 err(1, "Reading kernel");
424 /* If it's an ELF file, it starts with "\177ELF" */
425 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
426 return map_elf(fd
, &hdr
);
428 /* Otherwise we assume it's a bzImage, and try to load it. */
429 return load_bzimage(fd
);
432 /* This is a trivial little helper to align pages. Andi Kleen hated it because
433 * it calls getpagesize() twice: "it's dumb code."
435 * Kernel guys get really het up about optimization, even when it's not
436 * necessary. I leave this code as a reaction against that. */
437 static inline unsigned long page_align(unsigned long addr
)
439 /* Add upwards and truncate downwards. */
440 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
443 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
444 * the kernel which the kernel can use to boot from without needing any
445 * drivers. Most distributions now use this as standard: the initrd contains
446 * the code to load the appropriate driver modules for the current machine.
448 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
449 * kernels. He sent me this (and tells me when I break it). */
450 static unsigned long load_initrd(const char *name
, unsigned long mem
)
456 ifd
= open_or_die(name
, O_RDONLY
);
457 /* fstat() is needed to get the file size. */
458 if (fstat(ifd
, &st
) < 0)
459 err(1, "fstat() on initrd '%s'", name
);
461 /* We map the initrd at the top of memory, but mmap wants it to be
462 * page-aligned, so we round the size up for that. */
463 len
= page_align(st
.st_size
);
464 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
465 /* Once a file is mapped, you can close the file descriptor. It's a
466 * little odd, but quite useful. */
468 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
470 /* We return the initrd size. */
475 /* Simple routine to roll all the commandline arguments together with spaces
477 static void concat(char *dst
, char *args
[])
479 unsigned int i
, len
= 0;
481 for (i
= 0; args
[i
]; i
++) {
483 strcat(dst
+len
, " ");
486 strcpy(dst
+len
, args
[i
]);
487 len
+= strlen(args
[i
]);
489 /* In case it's empty. */
493 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
494 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
495 * the base of Guest "physical" memory, the top physical page to allow and the
496 * entry point for the Guest. */
497 static void tell_kernel(unsigned long start
)
499 unsigned long args
[] = { LHREQ_INITIALIZE
,
500 (unsigned long)guest_base
,
501 guest_limit
/ getpagesize(), start
};
502 verbose("Guest: %p - %p (%#lx)\n",
503 guest_base
, guest_base
+ guest_limit
, guest_limit
);
504 lguest_fd
= open_or_die("/dev/lguest", O_RDWR
);
505 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
506 err(1, "Writing to /dev/lguest");
513 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
514 * We need to make sure it's not trying to reach into the Launcher itself, so
515 * we have a convenient routine which checks it and exits with an error message
516 * if something funny is going on:
518 static void *_check_pointer(unsigned long addr
, unsigned int size
,
521 /* We have to separately check addr and addr+size, because size could
522 * be huge and addr + size might wrap around. */
523 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
524 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
525 /* We return a pointer for the caller's convenience, now we know it's
527 return from_guest_phys(addr
);
529 /* A macro which transparently hands the line number to the real function. */
530 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
532 /* Each buffer in the virtqueues is actually a chain of descriptors. This
533 * function returns the next descriptor in the chain, or vq->vring.num if we're
535 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
539 /* If this descriptor says it doesn't chain, we're done. */
540 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
541 return vq
->vring
.num
;
543 /* Check they're not leading us off end of descriptors. */
544 next
= vq
->vring
.desc
[i
].next
;
545 /* Make sure compiler knows to grab that: we don't want it changing! */
548 if (next
>= vq
->vring
.num
)
549 errx(1, "Desc next is %u", next
);
554 /* This looks in the virtqueue and for the first available buffer, and converts
555 * it to an iovec for convenient access. Since descriptors consist of some
556 * number of output then some number of input descriptors, it's actually two
557 * iovecs, but we pack them into one and note how many of each there were.
559 * This function returns the descriptor number found. */
560 static unsigned wait_for_vq_desc(struct virtqueue
*vq
,
562 unsigned int *out_num
, unsigned int *in_num
)
564 unsigned int i
, head
;
565 u16 last_avail
= lg_last_avail(vq
);
567 while (last_avail
== vq
->vring
.avail
->idx
) {
570 /* Nothing new? Wait for eventfd to tell us they refilled. */
571 if (read(vq
->eventfd
, &event
, sizeof(event
)) != sizeof(event
))
572 errx(1, "Event read failed?");
575 /* Check it isn't doing very strange things with descriptor numbers. */
576 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
577 errx(1, "Guest moved used index from %u to %u",
578 last_avail
, vq
->vring
.avail
->idx
);
580 /* Grab the next descriptor number they're advertising, and increment
581 * the index we've seen. */
582 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
585 /* If their number is silly, that's a fatal mistake. */
586 if (head
>= vq
->vring
.num
)
587 errx(1, "Guest says index %u is available", head
);
589 /* When we start there are none of either input nor output. */
590 *out_num
= *in_num
= 0;
594 /* Grab the first descriptor, and check it's OK. */
595 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
596 iov
[*out_num
+ *in_num
].iov_base
597 = check_pointer(vq
->vring
.desc
[i
].addr
,
598 vq
->vring
.desc
[i
].len
);
599 /* If this is an input descriptor, increment that count. */
600 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
603 /* If it's an output descriptor, they're all supposed
604 * to come before any input descriptors. */
606 errx(1, "Descriptor has out after in");
610 /* If we've got too many, that implies a descriptor loop. */
611 if (*out_num
+ *in_num
> vq
->vring
.num
)
612 errx(1, "Looped descriptor");
613 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
618 /* After we've used one of their buffers, we tell them about it. We'll then
619 * want to send them an interrupt, using trigger_irq(). */
620 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
622 struct vring_used_elem
*used
;
624 /* The virtqueue contains a ring of used buffers. Get a pointer to the
625 * next entry in that used ring. */
626 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
629 /* Make sure buffer is written before we update index. */
631 vq
->vring
.used
->idx
++;
634 /* This actually sends the interrupt for this virtqueue */
635 static void trigger_irq(struct virtqueue
*vq
)
637 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
639 /* If they don't want an interrupt, don't send one, unless empty. */
640 if ((vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
641 && lg_last_avail(vq
) != vq
->vring
.avail
->idx
)
644 /* Send the Guest an interrupt tell them we used something up. */
645 if (write(lguest_fd
, buf
, sizeof(buf
)) != 0)
646 err(1, "Triggering irq %i", vq
->config
.irq
);
649 /* And here's the combo meal deal. Supersize me! */
650 static void add_used_and_trigger(struct virtqueue
*vq
, unsigned head
, int len
)
652 add_used(vq
, head
, len
);
659 * We associate some data with the console for our exit hack. */
662 /* How many times have they hit ^C? */
664 /* When did they start? */
665 struct timeval start
;
668 /* This is the routine which handles console input (ie. stdin). */
669 static void console_input(struct virtqueue
*vq
)
672 unsigned int head
, in_num
, out_num
;
673 struct console_abort
*abort
= vq
->dev
->priv
;
674 struct iovec iov
[vq
->vring
.num
];
676 /* Make sure there's a descriptor waiting. */
677 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
679 errx(1, "Output buffers in console in queue?");
682 len
= readv(STDIN_FILENO
, iov
, in_num
);
684 /* Ran out of input? */
685 warnx("Failed to get console input, ignoring console.");
686 /* For simplicity, dying threads kill the whole Launcher. So
692 add_used_and_trigger(vq
, head
, len
);
694 /* Three ^C within one second? Exit.
696 * This is such a hack, but works surprisingly well. Each ^C has to
697 * be in a buffer by itself, so they can't be too fast. But we check
698 * that we get three within about a second, so they can't be too
700 if (len
!= 1 || ((char *)iov
[0].iov_base
)[0] != 3) {
706 if (abort
->count
== 1)
707 gettimeofday(&abort
->start
, NULL
);
708 else if (abort
->count
== 3) {
710 gettimeofday(&now
, NULL
);
711 /* Kill all Launcher processes with SIGINT, like normal ^C */
712 if (now
.tv_sec
<= abort
->start
.tv_sec
+1)
718 /* This is the routine which handles console output (ie. stdout). */
719 static void console_output(struct virtqueue
*vq
)
721 unsigned int head
, out
, in
;
722 struct iovec iov
[vq
->vring
.num
];
724 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
726 errx(1, "Input buffers in console output queue?");
727 while (!iov_empty(iov
, out
)) {
728 int len
= writev(STDOUT_FILENO
, iov
, out
);
730 err(1, "Write to stdout gave %i", len
);
731 iov_consume(iov
, out
, len
);
733 add_used_and_trigger(vq
, head
, 0);
739 * Handling output for network is also simple: we get all the output buffers
740 * and write them to /dev/net/tun.
746 static void net_output(struct virtqueue
*vq
)
748 struct net_info
*net_info
= vq
->dev
->priv
;
749 unsigned int head
, out
, in
;
750 struct iovec iov
[vq
->vring
.num
];
752 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
754 errx(1, "Input buffers in net output queue?");
755 if (writev(net_info
->tunfd
, iov
, out
) < 0)
756 errx(1, "Write to tun failed?");
757 add_used_and_trigger(vq
, head
, 0);
760 /* This is where we handle packets coming in from the tun device to our
762 static void net_input(struct virtqueue
*vq
)
765 unsigned int head
, out
, in
;
766 struct iovec iov
[vq
->vring
.num
];
767 struct net_info
*net_info
= vq
->dev
->priv
;
769 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
771 errx(1, "Output buffers in net input queue?");
772 len
= readv(net_info
->tunfd
, iov
, in
);
774 err(1, "Failed to read from tun.");
775 add_used_and_trigger(vq
, head
, len
);
778 /* This is the helper to create threads. */
779 static int do_thread(void *_vq
)
781 struct virtqueue
*vq
= _vq
;
788 /* When a child dies, we kill our entire process group with SIGTERM. This
789 * also has the side effect that the shell restores the console for us! */
790 static void kill_launcher(int signal
)
795 static void reset_device(struct device
*dev
)
797 struct virtqueue
*vq
;
799 verbose("Resetting device %s\n", dev
->name
);
801 /* Clear any features they've acked. */
802 memset(get_feature_bits(dev
) + dev
->feature_len
, 0, dev
->feature_len
);
804 /* We're going to be explicitly killing threads, so ignore them. */
805 signal(SIGCHLD
, SIG_IGN
);
807 /* Zero out the virtqueues, get rid of their threads */
808 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
809 if (vq
->thread
!= (pid_t
)-1) {
810 kill(vq
->thread
, SIGTERM
);
811 waitpid(vq
->thread
, NULL
, 0);
812 vq
->thread
= (pid_t
)-1;
814 memset(vq
->vring
.desc
, 0,
815 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
816 lg_last_avail(vq
) = 0;
818 dev
->running
= false;
820 /* Now we care if threads die. */
821 signal(SIGCHLD
, (void *)kill_launcher
);
824 static void create_thread(struct virtqueue
*vq
)
826 /* Create stack for thread and run it. Since stack grows
827 * upwards, we point the stack pointer to the end of this
829 char *stack
= malloc(32768);
830 unsigned long args
[] = { LHREQ_EVENTFD
,
831 vq
->config
.pfn
*getpagesize(), 0 };
833 /* Create a zero-initialized eventfd. */
834 vq
->eventfd
= eventfd(0, 0);
836 err(1, "Creating eventfd");
837 args
[2] = vq
->eventfd
;
839 /* Attach an eventfd to this virtqueue: it will go off
840 * when the Guest does an LHCALL_NOTIFY for this vq. */
841 if (write(lguest_fd
, &args
, sizeof(args
)) != 0)
842 err(1, "Attaching eventfd");
844 /* CLONE_VM: because it has to access the Guest memory, and
845 * SIGCHLD so we get a signal if it dies. */
846 vq
->thread
= clone(do_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, vq
);
847 if (vq
->thread
== (pid_t
)-1)
848 err(1, "Creating clone");
849 /* We close our local copy, now the child has it. */
853 static void start_device(struct device
*dev
)
856 struct virtqueue
*vq
;
858 verbose("Device %s OK: offered", dev
->name
);
859 for (i
= 0; i
< dev
->feature_len
; i
++)
860 verbose(" %02x", get_feature_bits(dev
)[i
]);
861 verbose(", accepted");
862 for (i
= 0; i
< dev
->feature_len
; i
++)
863 verbose(" %02x", get_feature_bits(dev
)
864 [dev
->feature_len
+i
]);
866 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
873 static void cleanup_devices(void)
877 for (dev
= devices
.dev
; dev
; dev
= dev
->next
)
880 /* If we saved off the original terminal settings, restore them now. */
881 if (orig_term
.c_lflag
& (ISIG
|ICANON
|ECHO
))
882 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
885 /* When the Guest tells us they updated the status field, we handle it. */
886 static void update_device_status(struct device
*dev
)
888 /* A zero status is a reset, otherwise it's a set of flags. */
889 if (dev
->desc
->status
== 0)
891 else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
892 warnx("Device %s configuration FAILED", dev
->name
);
895 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
901 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
902 static void handle_output(unsigned long addr
)
906 /* Check each device. */
907 for (i
= devices
.dev
; i
; i
= i
->next
) {
908 struct virtqueue
*vq
;
910 /* Notifications to device descriptors update device status. */
911 if (from_guest_phys(addr
) == i
->desc
) {
912 update_device_status(i
);
916 /* Devices *can* be used before status is set to DRIVER_OK. */
917 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
918 if (addr
!= vq
->config
.pfn
*getpagesize())
921 errx(1, "Notification on running %s", i
->name
);
927 /* Early console write is done using notify on a nul-terminated string
928 * in Guest memory. */
929 if (addr
>= guest_limit
)
930 errx(1, "Bad NOTIFY %#lx", addr
);
932 write(STDOUT_FILENO
, from_guest_phys(addr
),
933 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
939 * All devices need a descriptor so the Guest knows it exists, and a "struct
940 * device" so the Launcher can keep track of it. We have common helper
941 * routines to allocate and manage them.
944 /* The layout of the device page is a "struct lguest_device_desc" followed by a
945 * number of virtqueue descriptors, then two sets of feature bits, then an
946 * array of configuration bytes. This routine returns the configuration
948 static u8
*device_config(const struct device
*dev
)
950 return (void *)(dev
->desc
+ 1)
951 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
952 + dev
->feature_len
* 2;
955 /* This routine allocates a new "struct lguest_device_desc" from descriptor
956 * table page just above the Guest's normal memory. It returns a pointer to
957 * that descriptor. */
958 static struct lguest_device_desc
*new_dev_desc(u16 type
)
960 struct lguest_device_desc d
= { .type
= type
};
963 /* Figure out where the next device config is, based on the last one. */
965 p
= device_config(devices
.lastdev
)
966 + devices
.lastdev
->desc
->config_len
;
968 p
= devices
.descpage
;
970 /* We only have one page for all the descriptors. */
971 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
972 errx(1, "Too many devices");
974 /* p might not be aligned, so we memcpy in. */
975 return memcpy(p
, &d
, sizeof(d
));
978 /* Each device descriptor is followed by the description of its virtqueues. We
979 * specify how many descriptors the virtqueue is to have. */
980 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
981 void (*service
)(struct virtqueue
*))
984 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
987 /* First we need some memory for this virtqueue. */
988 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
990 p
= get_pages(pages
);
992 /* Initialize the virtqueue */
994 vq
->last_avail_idx
= 0;
996 vq
->service
= service
;
997 vq
->thread
= (pid_t
)-1;
999 /* Initialize the configuration. */
1000 vq
->config
.num
= num_descs
;
1001 vq
->config
.irq
= devices
.next_irq
++;
1002 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1004 /* Initialize the vring. */
1005 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1007 /* Append virtqueue to this device's descriptor. We use
1008 * device_config() to get the end of the device's current virtqueues;
1009 * we check that we haven't added any config or feature information
1010 * yet, otherwise we'd be overwriting them. */
1011 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1012 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1014 dev
->desc
->num_vq
++;
1016 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1018 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1020 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1024 /* The first half of the feature bitmask is for us to advertise features. The
1025 * second half is for the Guest to accept features. */
1026 static void add_feature(struct device
*dev
, unsigned bit
)
1028 u8
*features
= get_feature_bits(dev
);
1030 /* We can't extend the feature bits once we've added config bytes */
1031 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1032 assert(dev
->desc
->config_len
== 0);
1033 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1036 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1039 /* This routine sets the configuration fields for an existing device's
1040 * descriptor. It only works for the last device, but that's OK because that's
1042 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1044 /* Check we haven't overflowed our single page. */
1045 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1046 errx(1, "Too many devices");
1048 /* Copy in the config information, and store the length. */
1049 memcpy(device_config(dev
), conf
, len
);
1050 dev
->desc
->config_len
= len
;
1053 /* This routine does all the creation and setup of a new device, including
1054 * calling new_dev_desc() to allocate the descriptor and device memory.
1056 * See what I mean about userspace being boring? */
1057 static struct device
*new_device(const char *name
, u16 type
)
1059 struct device
*dev
= malloc(sizeof(*dev
));
1061 /* Now we populate the fields one at a time. */
1062 dev
->desc
= new_dev_desc(type
);
1065 dev
->feature_len
= 0;
1067 dev
->running
= false;
1069 /* Append to device list. Prepending to a single-linked list is
1070 * easier, but the user expects the devices to be arranged on the bus
1071 * in command-line order. The first network device on the command line
1072 * is eth0, the first block device /dev/vda, etc. */
1073 if (devices
.lastdev
)
1074 devices
.lastdev
->next
= dev
;
1077 devices
.lastdev
= dev
;
1082 /* Our first setup routine is the console. It's a fairly simple device, but
1083 * UNIX tty handling makes it uglier than it could be. */
1084 static void setup_console(void)
1088 /* If we can save the initial standard input settings... */
1089 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1090 struct termios term
= orig_term
;
1091 /* Then we turn off echo, line buffering and ^C etc. We want a
1092 * raw input stream to the Guest. */
1093 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1094 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1097 dev
= new_device("console", VIRTIO_ID_CONSOLE
);
1099 /* We store the console state in dev->priv, and initialize it. */
1100 dev
->priv
= malloc(sizeof(struct console_abort
));
1101 ((struct console_abort
*)dev
->priv
)->count
= 0;
1103 /* The console needs two virtqueues: the input then the output. When
1104 * they put something the input queue, we make sure we're listening to
1105 * stdin. When they put something in the output queue, we write it to
1107 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_input
);
1108 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_output
);
1110 verbose("device %u: console\n", ++devices
.device_num
);
1114 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1115 * --sharenet=<name> option which opens or creates a named pipe. This can be
1116 * used to send packets to another guest in a 1:1 manner.
1118 * More sopisticated is to use one of the tools developed for project like UML
1121 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1122 * completely generic ("here's my vring, attach to your vring") and would work
1123 * for any traffic. Of course, namespace and permissions issues need to be
1124 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1125 * multiple inter-guest channels behind one interface, although it would
1126 * require some manner of hotplugging new virtio channels.
1128 * Finally, we could implement a virtio network switch in the kernel. :*/
1130 static u32
str2ip(const char *ipaddr
)
1134 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1135 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1136 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1139 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1142 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1143 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1144 errx(1, "Failed to parse mac address '%s'", macaddr
);
1153 /* This code is "adapted" from libbridge: it attaches the Host end of the
1154 * network device to the bridge device specified by the command line.
1156 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1157 * dislike bridging), and I just try not to break it. */
1158 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1164 errx(1, "must specify bridge name");
1166 ifidx
= if_nametoindex(if_name
);
1168 errx(1, "interface %s does not exist!", if_name
);
1170 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1171 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1172 ifr
.ifr_ifindex
= ifidx
;
1173 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1174 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1177 /* This sets up the Host end of the network device with an IP address, brings
1178 * it up so packets will flow, the copies the MAC address into the hwaddr
1180 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1183 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1185 memset(&ifr
, 0, sizeof(ifr
));
1186 strcpy(ifr
.ifr_name
, tapif
);
1188 /* Don't read these incantations. Just cut & paste them like I did! */
1189 sin
->sin_family
= AF_INET
;
1190 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1191 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1192 err(1, "Setting %s interface address", tapif
);
1193 ifr
.ifr_flags
= IFF_UP
;
1194 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1195 err(1, "Bringing interface %s up", tapif
);
1198 static int get_tun_device(char tapif
[IFNAMSIZ
])
1203 /* Start with this zeroed. Messy but sure. */
1204 memset(&ifr
, 0, sizeof(ifr
));
1206 /* We open the /dev/net/tun device and tell it we want a tap device. A
1207 * tap device is like a tun device, only somehow different. To tell
1208 * the truth, I completely blundered my way through this code, but it
1210 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1211 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1212 strcpy(ifr
.ifr_name
, "tap%d");
1213 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1214 err(1, "configuring /dev/net/tun");
1216 if (ioctl(netfd
, TUNSETOFFLOAD
,
1217 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1218 err(1, "Could not set features for tun device");
1220 /* We don't need checksums calculated for packets coming in this
1221 * device: trust us! */
1222 ioctl(netfd
, TUNSETNOCSUM
, 1);
1224 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1228 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1229 * routing, but the principle is the same: it uses the "tun" device to inject
1230 * packets into the Host as if they came in from a normal network card. We
1231 * just shunt packets between the Guest and the tun device. */
1232 static void setup_tun_net(char *arg
)
1235 struct net_info
*net_info
= malloc(sizeof(*net_info
));
1237 u32 ip
= INADDR_ANY
;
1238 bool bridging
= false;
1239 char tapif
[IFNAMSIZ
], *p
;
1240 struct virtio_net_config conf
;
1242 net_info
->tunfd
= get_tun_device(tapif
);
1244 /* First we create a new network device. */
1245 dev
= new_device("net", VIRTIO_ID_NET
);
1246 dev
->priv
= net_info
;
1248 /* Network devices need a receive and a send queue, just like
1250 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_input
);
1251 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_output
);
1253 /* We need a socket to perform the magic network ioctls to bring up the
1254 * tap interface, connect to the bridge etc. Any socket will do! */
1255 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1257 err(1, "opening IP socket");
1259 /* If the command line was --tunnet=bridge:<name> do bridging. */
1260 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1261 arg
+= strlen(BRIDGE_PFX
);
1265 /* A mac address may follow the bridge name or IP address */
1266 p
= strchr(arg
, ':');
1268 str2mac(p
+1, conf
.mac
);
1269 add_feature(dev
, VIRTIO_NET_F_MAC
);
1273 /* arg is now either an IP address or a bridge name */
1275 add_to_bridge(ipfd
, tapif
, arg
);
1279 /* Set up the tun device. */
1280 configure_device(ipfd
, tapif
, ip
);
1282 add_feature(dev
, VIRTIO_F_NOTIFY_ON_EMPTY
);
1283 /* Expect Guest to handle everything except UFO */
1284 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1285 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1286 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1287 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1288 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1289 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1290 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1291 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1292 set_config(dev
, sizeof(conf
), &conf
);
1294 /* We don't need the socket any more; setup is done. */
1297 devices
.device_num
++;
1300 verbose("device %u: tun %s attached to bridge: %s\n",
1301 devices
.device_num
, tapif
, arg
);
1303 verbose("device %u: tun %s: %s\n",
1304 devices
.device_num
, tapif
, arg
);
1307 /* Our block (disk) device should be really simple: the Guest asks for a block
1308 * number and we read or write that position in the file. Unfortunately, that
1309 * was amazingly slow: the Guest waits until the read is finished before
1310 * running anything else, even if it could have been doing useful work.
1312 * We could use async I/O, except it's reputed to suck so hard that characters
1313 * actually go missing from your code when you try to use it.
1315 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1317 /* This hangs off device->priv. */
1320 /* The size of the file. */
1323 /* The file descriptor for the file. */
1326 /* IO thread listens on this file descriptor [0]. */
1329 /* IO thread writes to this file descriptor to mark it done, then
1330 * Launcher triggers interrupt to Guest. */
1337 * Remember that the block device is handled by a separate I/O thread. We head
1338 * straight into the core of that thread here:
1340 static void blk_request(struct virtqueue
*vq
)
1342 struct vblk_info
*vblk
= vq
->dev
->priv
;
1343 unsigned int head
, out_num
, in_num
, wlen
;
1346 struct virtio_blk_outhdr
*out
;
1347 struct iovec iov
[vq
->vring
.num
];
1350 /* Get the next request. */
1351 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1353 /* Every block request should contain at least one output buffer
1354 * (detailing the location on disk and the type of request) and one
1355 * input buffer (to hold the result). */
1356 if (out_num
== 0 || in_num
== 0)
1357 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1358 head
, out_num
, in_num
);
1360 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1361 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1362 off
= out
->sector
* 512;
1364 /* The block device implements "barriers", where the Guest indicates
1365 * that it wants all previous writes to occur before this write. We
1366 * don't have a way of asking our kernel to do a barrier, so we just
1367 * synchronize all the data in the file. Pretty poor, no? */
1368 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1369 fdatasync(vblk
->fd
);
1371 /* In general the virtio block driver is allowed to try SCSI commands.
1372 * It'd be nice if we supported eject, for example, but we don't. */
1373 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1374 fprintf(stderr
, "Scsi commands unsupported\n");
1375 *in
= VIRTIO_BLK_S_UNSUPP
;
1377 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1380 /* Move to the right location in the block file. This can fail
1381 * if they try to write past end. */
1382 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1383 err(1, "Bad seek to sector %llu", out
->sector
);
1385 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1386 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1388 /* Grr... Now we know how long the descriptor they sent was, we
1389 * make sure they didn't try to write over the end of the block
1390 * file (possibly extending it). */
1391 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1392 /* Trim it back to the correct length */
1393 ftruncate64(vblk
->fd
, vblk
->len
);
1394 /* Die, bad Guest, die. */
1395 errx(1, "Write past end %llu+%u", off
, ret
);
1398 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1402 /* Move to the right location in the block file. This can fail
1403 * if they try to read past end. */
1404 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1405 err(1, "Bad seek to sector %llu", out
->sector
);
1407 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1408 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1410 wlen
= sizeof(*in
) + ret
;
1411 *in
= VIRTIO_BLK_S_OK
;
1414 *in
= VIRTIO_BLK_S_IOERR
;
1418 /* OK, so we noted that it was pretty poor to use an fdatasync as a
1419 * barrier. But Christoph Hellwig points out that we need a sync
1420 * *afterwards* as well: "Barriers specify no reordering to the front
1421 * or the back." And Jens Axboe confirmed it, so here we are: */
1422 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1423 fdatasync(vblk
->fd
);
1425 add_used_and_trigger(vq
, head
, wlen
);
1428 /*L:198 This actually sets up a virtual block device. */
1429 static void setup_block_file(const char *filename
)
1432 struct vblk_info
*vblk
;
1433 struct virtio_blk_config conf
;
1435 /* The device responds to return from I/O thread. */
1436 dev
= new_device("block", VIRTIO_ID_BLOCK
);
1438 /* The device has one virtqueue, where the Guest places requests. */
1439 add_virtqueue(dev
, VIRTQUEUE_NUM
, blk_request
);
1441 /* Allocate the room for our own bookkeeping */
1442 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1444 /* First we open the file and store the length. */
1445 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1446 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1448 /* We support barriers. */
1449 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1451 /* Tell Guest how many sectors this device has. */
1452 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1454 /* Tell Guest not to put in too many descriptors at once: two are used
1455 * for the in and out elements. */
1456 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1457 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1459 set_config(dev
, sizeof(conf
), &conf
);
1461 verbose("device %u: virtblock %llu sectors\n",
1462 ++devices
.device_num
, le64_to_cpu(conf
.capacity
));
1469 /* Our random number generator device reads from /dev/random into the Guest's
1470 * input buffers. The usual case is that the Guest doesn't want random numbers
1471 * and so has no buffers although /dev/random is still readable, whereas
1472 * console is the reverse.
1474 * The same logic applies, however. */
1475 static void rng_input(struct virtqueue
*vq
)
1478 unsigned int head
, in_num
, out_num
, totlen
= 0;
1479 struct rng_info
*rng_info
= vq
->dev
->priv
;
1480 struct iovec iov
[vq
->vring
.num
];
1482 /* First we need a buffer from the Guests's virtqueue. */
1483 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1485 errx(1, "Output buffers in rng?");
1487 /* This is why we convert to iovecs: the readv() call uses them, and so
1488 * it reads straight into the Guest's buffer. We loop to make sure we
1490 while (!iov_empty(iov
, in_num
)) {
1491 len
= readv(rng_info
->rfd
, iov
, in_num
);
1493 err(1, "Read from /dev/random gave %i", len
);
1494 iov_consume(iov
, in_num
, len
);
1498 /* Tell the Guest about the new input. */
1499 add_used_and_trigger(vq
, head
, totlen
);
1502 /* And this creates a "hardware" random number device for the Guest. */
1503 static void setup_rng(void)
1506 struct rng_info
*rng_info
= malloc(sizeof(*rng_info
));
1508 rng_info
->rfd
= open_or_die("/dev/random", O_RDONLY
);
1510 /* The device responds to return from I/O thread. */
1511 dev
= new_device("rng", VIRTIO_ID_RNG
);
1512 dev
->priv
= rng_info
;
1514 /* The device has one virtqueue, where the Guest places inbufs. */
1515 add_virtqueue(dev
, VIRTQUEUE_NUM
, rng_input
);
1517 verbose("device %u: rng\n", devices
.device_num
++);
1519 /* That's the end of device setup. */
1521 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1522 static void __attribute__((noreturn
)) restart_guest(void)
1526 /* Since we don't track all open fds, we simply close everything beyond
1528 for (i
= 3; i
< FD_SETSIZE
; i
++)
1531 /* Reset all the devices (kills all threads). */
1534 execv(main_args
[0], main_args
);
1535 err(1, "Could not exec %s", main_args
[0]);
1538 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1539 * its input and output, and finally, lays it to rest. */
1540 static void __attribute__((noreturn
)) run_guest(void)
1543 unsigned long notify_addr
;
1546 /* We read from the /dev/lguest device to run the Guest. */
1547 readval
= pread(lguest_fd
, ¬ify_addr
,
1548 sizeof(notify_addr
), cpu_id
);
1550 /* One unsigned long means the Guest did HCALL_NOTIFY */
1551 if (readval
== sizeof(notify_addr
)) {
1552 verbose("Notify on address %#lx\n", notify_addr
);
1553 handle_output(notify_addr
);
1554 /* ENOENT means the Guest died. Reading tells us why. */
1555 } else if (errno
== ENOENT
) {
1556 char reason
[1024] = { 0 };
1557 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1558 errx(1, "%s", reason
);
1559 /* ERESTART means that we need to reboot the guest */
1560 } else if (errno
== ERESTART
) {
1562 /* Anything else means a bug or incompatible change. */
1564 err(1, "Running guest failed");
1568 * This is the end of the Launcher. The good news: we are over halfway
1569 * through! The bad news: the most fiendish part of the code still lies ahead
1572 * Are you ready? Take a deep breath and join me in the core of the Host, in
1576 static struct option opts
[] = {
1577 { "verbose", 0, NULL
, 'v' },
1578 { "tunnet", 1, NULL
, 't' },
1579 { "block", 1, NULL
, 'b' },
1580 { "rng", 0, NULL
, 'r' },
1581 { "initrd", 1, NULL
, 'i' },
1584 static void usage(void)
1586 errx(1, "Usage: lguest [--verbose] "
1587 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1588 "|--block=<filename>|--initrd=<filename>]...\n"
1589 "<mem-in-mb> vmlinux [args...]");
1592 /*L:105 The main routine is where the real work begins: */
1593 int main(int argc
, char *argv
[])
1595 /* Memory, top-level pagetable, code startpoint and size of the
1596 * (optional) initrd. */
1597 unsigned long mem
= 0, start
, initrd_size
= 0;
1598 /* Two temporaries. */
1600 /* The boot information for the Guest. */
1601 struct boot_params
*boot
;
1602 /* If they specify an initrd file to load. */
1603 const char *initrd_name
= NULL
;
1605 /* Save the args: we "reboot" by execing ourselves again. */
1608 /* First we initialize the device list. We keep a pointer to the last
1609 * device, and the next interrupt number to use for devices (1:
1610 * remember that 0 is used by the timer). */
1611 devices
.lastdev
= NULL
;
1612 devices
.next_irq
= 1;
1615 /* We need to know how much memory so we can set up the device
1616 * descriptor and memory pages for the devices as we parse the command
1617 * line. So we quickly look through the arguments to find the amount
1619 for (i
= 1; i
< argc
; i
++) {
1620 if (argv
[i
][0] != '-') {
1621 mem
= atoi(argv
[i
]) * 1024 * 1024;
1622 /* We start by mapping anonymous pages over all of
1623 * guest-physical memory range. This fills it with 0,
1624 * and ensures that the Guest won't be killed when it
1625 * tries to access it. */
1626 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1629 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1630 devices
.descpage
= get_pages(1);
1635 /* The options are fairly straight-forward */
1636 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1642 setup_tun_net(optarg
);
1645 setup_block_file(optarg
);
1651 initrd_name
= optarg
;
1654 warnx("Unknown argument %s", argv
[optind
]);
1658 /* After the other arguments we expect memory and kernel image name,
1659 * followed by command line arguments for the kernel. */
1660 if (optind
+ 2 > argc
)
1663 verbose("Guest base is at %p\n", guest_base
);
1665 /* We always have a console device */
1668 /* Now we load the kernel */
1669 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1671 /* Boot information is stashed at physical address 0 */
1672 boot
= from_guest_phys(0);
1674 /* Map the initrd image if requested (at top of physical memory) */
1676 initrd_size
= load_initrd(initrd_name
, mem
);
1677 /* These are the location in the Linux boot header where the
1678 * start and size of the initrd are expected to be found. */
1679 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1680 boot
->hdr
.ramdisk_size
= initrd_size
;
1681 /* The bootloader type 0xFF means "unknown"; that's OK. */
1682 boot
->hdr
.type_of_loader
= 0xFF;
1685 /* The Linux boot header contains an "E820" memory map: ours is a
1686 * simple, single region. */
1687 boot
->e820_entries
= 1;
1688 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1689 /* The boot header contains a command line pointer: we put the command
1690 * line after the boot header. */
1691 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1692 /* We use a simple helper to copy the arguments separated by spaces. */
1693 concat((char *)(boot
+ 1), argv
+optind
+2);
1695 /* Boot protocol version: 2.07 supports the fields for lguest. */
1696 boot
->hdr
.version
= 0x207;
1698 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1699 boot
->hdr
.hardware_subarch
= 1;
1701 /* Tell the entry path not to try to reload segment registers. */
1702 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1704 /* We tell the kernel to initialize the Guest: this returns the open
1705 * /dev/lguest file descriptor. */
1708 /* Ensure that we terminate if a child dies. */
1709 signal(SIGCHLD
, kill_launcher
);
1711 /* If we exit via err(), this kills all the threads, restores tty. */
1712 atexit(cleanup_devices
);
1714 /* Finally, run the Guest. This doesn't return. */
1720 * Mastery is done: you now know everything I do.
1722 * But surely you have seen code, features and bugs in your wanderings which
1723 * you now yearn to attack? That is the real game, and I look forward to you
1724 * patching and forking lguest into the Your-Name-Here-visor.
1726 * Farewell, and good coding!