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 /* How many are used since we sent last irq? */
155 unsigned int pending_used
;
157 /* Eventfd where Guest notifications arrive. */
160 /* Function for the thread which is servicing this virtqueue. */
161 void (*service
)(struct virtqueue
*vq
);
165 /* Remember the arguments to the program so we can "reboot" */
166 static char **main_args
;
168 /* The original tty settings to restore on exit. */
169 static struct termios orig_term
;
171 /* We have to be careful with barriers: our devices are all run in separate
172 * threads and so we need to make sure that changes visible to the Guest happen
173 * in precise order. */
174 #define wmb() __asm__ __volatile__("" : : : "memory")
176 /* Convert an iovec element to the given type.
178 * This is a fairly ugly trick: we need to know the size of the type and
179 * alignment requirement to check the pointer is kosher. It's also nice to
180 * have the name of the type in case we report failure.
182 * Typing those three things all the time is cumbersome and error prone, so we
183 * have a macro which sets them all up and passes to the real function. */
184 #define convert(iov, type) \
185 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
187 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
190 if (iov
->iov_len
!= size
)
191 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
192 if ((unsigned long)iov
->iov_base
% align
!= 0)
193 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
194 return iov
->iov_base
;
197 /* Wrapper for the last available index. Makes it easier to change. */
198 #define lg_last_avail(vq) ((vq)->last_avail_idx)
200 /* The virtio configuration space is defined to be little-endian. x86 is
201 * little-endian too, but it's nice to be explicit so we have these helpers. */
202 #define cpu_to_le16(v16) (v16)
203 #define cpu_to_le32(v32) (v32)
204 #define cpu_to_le64(v64) (v64)
205 #define le16_to_cpu(v16) (v16)
206 #define le32_to_cpu(v32) (v32)
207 #define le64_to_cpu(v64) (v64)
209 /* Is this iovec empty? */
210 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
214 for (i
= 0; i
< num_iov
; i
++)
220 /* Take len bytes from the front of this iovec. */
221 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
225 for (i
= 0; i
< num_iov
; i
++) {
228 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
229 iov
[i
].iov_base
+= used
;
230 iov
[i
].iov_len
-= used
;
236 /* The device virtqueue descriptors are followed by feature bitmasks. */
237 static u8
*get_feature_bits(struct device
*dev
)
239 return (u8
*)(dev
->desc
+ 1)
240 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
243 /*L:100 The Launcher code itself takes us out into userspace, that scary place
244 * where pointers run wild and free! Unfortunately, like most userspace
245 * programs, it's quite boring (which is why everyone likes to hack on the
246 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
247 * will get you through this section. Or, maybe not.
249 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
250 * memory and stores it in "guest_base". In other words, Guest physical ==
251 * Launcher virtual with an offset.
253 * This can be tough to get your head around, but usually it just means that we
254 * use these trivial conversion functions when the Guest gives us it's
255 * "physical" addresses: */
256 static void *from_guest_phys(unsigned long addr
)
258 return guest_base
+ addr
;
261 static unsigned long to_guest_phys(const void *addr
)
263 return (addr
- guest_base
);
267 * Loading the Kernel.
269 * We start with couple of simple helper routines. open_or_die() avoids
270 * error-checking code cluttering the callers: */
271 static int open_or_die(const char *name
, int flags
)
273 int fd
= open(name
, flags
);
275 err(1, "Failed to open %s", name
);
279 /* map_zeroed_pages() takes a number of pages. */
280 static void *map_zeroed_pages(unsigned int num
)
282 int fd
= open_or_die("/dev/zero", O_RDONLY
);
285 /* We use a private mapping (ie. if we write to the page, it will be
287 addr
= mmap(NULL
, getpagesize() * num
,
288 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
289 if (addr
== MAP_FAILED
)
290 err(1, "Mmaping %u pages of /dev/zero", num
);
296 /* Get some more pages for a device. */
297 static void *get_pages(unsigned int num
)
299 void *addr
= from_guest_phys(guest_limit
);
301 guest_limit
+= num
* getpagesize();
302 if (guest_limit
> guest_max
)
303 errx(1, "Not enough memory for devices");
307 /* This routine is used to load the kernel or initrd. It tries mmap, but if
308 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
309 * it falls back to reading the memory in. */
310 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
314 /* We map writable even though for some segments are marked read-only.
315 * The kernel really wants to be writable: it patches its own
318 * MAP_PRIVATE means that the page won't be copied until a write is
319 * done to it. This allows us to share untouched memory between
321 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
322 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
325 /* pread does a seek and a read in one shot: saves a few lines. */
326 r
= pread(fd
, addr
, len
, offset
);
328 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
331 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
332 * the Guest memory. ELF = Embedded Linking Format, which is the format used
333 * by all modern binaries on Linux including the kernel.
335 * The ELF headers give *two* addresses: a physical address, and a virtual
336 * address. We use the physical address; the Guest will map itself to the
339 * We return the starting address. */
340 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
342 Elf32_Phdr phdr
[ehdr
->e_phnum
];
345 /* Sanity checks on the main ELF header: an x86 executable with a
346 * reasonable number of correctly-sized program headers. */
347 if (ehdr
->e_type
!= ET_EXEC
348 || ehdr
->e_machine
!= EM_386
349 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
350 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
351 errx(1, "Malformed elf header");
353 /* An ELF executable contains an ELF header and a number of "program"
354 * headers which indicate which parts ("segments") of the program to
357 /* We read in all the program headers at once: */
358 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
359 err(1, "Seeking to program headers");
360 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
361 err(1, "Reading program headers");
363 /* Try all the headers: there are usually only three. A read-only one,
364 * a read-write one, and a "note" section which we don't load. */
365 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
366 /* If this isn't a loadable segment, we ignore it */
367 if (phdr
[i
].p_type
!= PT_LOAD
)
370 verbose("Section %i: size %i addr %p\n",
371 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
373 /* We map this section of the file at its physical address. */
374 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
375 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
378 /* The entry point is given in the ELF header. */
379 return ehdr
->e_entry
;
382 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
383 * supposed to jump into it and it will unpack itself. We used to have to
384 * perform some hairy magic because the unpacking code scared me.
386 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
387 * a small patch to jump over the tricky bits in the Guest, so now we just read
388 * the funky header so we know where in the file to load, and away we go! */
389 static unsigned long load_bzimage(int fd
)
391 struct boot_params boot
;
393 /* Modern bzImages get loaded at 1M. */
394 void *p
= from_guest_phys(0x100000);
396 /* Go back to the start of the file and read the header. It should be
397 * a Linux boot header (see Documentation/x86/i386/boot.txt) */
398 lseek(fd
, 0, SEEK_SET
);
399 read(fd
, &boot
, sizeof(boot
));
401 /* Inside the setup_hdr, we expect the magic "HdrS" */
402 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
403 errx(1, "This doesn't look like a bzImage to me");
405 /* Skip over the extra sectors of the header. */
406 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
408 /* Now read everything into memory. in nice big chunks. */
409 while ((r
= read(fd
, p
, 65536)) > 0)
412 /* Finally, code32_start tells us where to enter the kernel. */
413 return boot
.hdr
.code32_start
;
416 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
417 * come wrapped up in the self-decompressing "bzImage" format. With a little
418 * work, we can load those, too. */
419 static unsigned long load_kernel(int fd
)
423 /* Read in the first few bytes. */
424 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
425 err(1, "Reading kernel");
427 /* If it's an ELF file, it starts with "\177ELF" */
428 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
429 return map_elf(fd
, &hdr
);
431 /* Otherwise we assume it's a bzImage, and try to load it. */
432 return load_bzimage(fd
);
435 /* This is a trivial little helper to align pages. Andi Kleen hated it because
436 * it calls getpagesize() twice: "it's dumb code."
438 * Kernel guys get really het up about optimization, even when it's not
439 * necessary. I leave this code as a reaction against that. */
440 static inline unsigned long page_align(unsigned long addr
)
442 /* Add upwards and truncate downwards. */
443 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
446 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
447 * the kernel which the kernel can use to boot from without needing any
448 * drivers. Most distributions now use this as standard: the initrd contains
449 * the code to load the appropriate driver modules for the current machine.
451 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
452 * kernels. He sent me this (and tells me when I break it). */
453 static unsigned long load_initrd(const char *name
, unsigned long mem
)
459 ifd
= open_or_die(name
, O_RDONLY
);
460 /* fstat() is needed to get the file size. */
461 if (fstat(ifd
, &st
) < 0)
462 err(1, "fstat() on initrd '%s'", name
);
464 /* We map the initrd at the top of memory, but mmap wants it to be
465 * page-aligned, so we round the size up for that. */
466 len
= page_align(st
.st_size
);
467 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
468 /* Once a file is mapped, you can close the file descriptor. It's a
469 * little odd, but quite useful. */
471 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
473 /* We return the initrd size. */
478 /* Simple routine to roll all the commandline arguments together with spaces
480 static void concat(char *dst
, char *args
[])
482 unsigned int i
, len
= 0;
484 for (i
= 0; args
[i
]; i
++) {
486 strcat(dst
+len
, " ");
489 strcpy(dst
+len
, args
[i
]);
490 len
+= strlen(args
[i
]);
492 /* In case it's empty. */
496 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
497 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
498 * the base of Guest "physical" memory, the top physical page to allow and the
499 * entry point for the Guest. */
500 static void tell_kernel(unsigned long start
)
502 unsigned long args
[] = { LHREQ_INITIALIZE
,
503 (unsigned long)guest_base
,
504 guest_limit
/ getpagesize(), start
};
505 verbose("Guest: %p - %p (%#lx)\n",
506 guest_base
, guest_base
+ guest_limit
, guest_limit
);
507 lguest_fd
= open_or_die("/dev/lguest", O_RDWR
);
508 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
509 err(1, "Writing to /dev/lguest");
516 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
517 * We need to make sure it's not trying to reach into the Launcher itself, so
518 * we have a convenient routine which checks it and exits with an error message
519 * if something funny is going on:
521 static void *_check_pointer(unsigned long addr
, unsigned int size
,
524 /* We have to separately check addr and addr+size, because size could
525 * be huge and addr + size might wrap around. */
526 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
527 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
528 /* We return a pointer for the caller's convenience, now we know it's
530 return from_guest_phys(addr
);
532 /* A macro which transparently hands the line number to the real function. */
533 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
535 /* Each buffer in the virtqueues is actually a chain of descriptors. This
536 * function returns the next descriptor in the chain, or vq->vring.num if we're
538 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
542 /* If this descriptor says it doesn't chain, we're done. */
543 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
544 return vq
->vring
.num
;
546 /* Check they're not leading us off end of descriptors. */
547 next
= vq
->vring
.desc
[i
].next
;
548 /* Make sure compiler knows to grab that: we don't want it changing! */
551 if (next
>= vq
->vring
.num
)
552 errx(1, "Desc next is %u", next
);
557 /* This actually sends the interrupt for this virtqueue */
558 static void trigger_irq(struct virtqueue
*vq
)
560 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
562 /* Don't inform them if nothing used. */
563 if (!vq
->pending_used
)
565 vq
->pending_used
= 0;
567 /* If they don't want an interrupt, don't send one, unless empty. */
568 if ((vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
569 && lg_last_avail(vq
) != vq
->vring
.avail
->idx
)
572 /* Send the Guest an interrupt tell them we used something up. */
573 if (write(lguest_fd
, buf
, sizeof(buf
)) != 0)
574 err(1, "Triggering irq %i", vq
->config
.irq
);
577 /* This looks in the virtqueue and for the first available buffer, and converts
578 * it to an iovec for convenient access. Since descriptors consist of some
579 * number of output then some number of input descriptors, it's actually two
580 * iovecs, but we pack them into one and note how many of each there were.
582 * This function returns the descriptor number found. */
583 static unsigned wait_for_vq_desc(struct virtqueue
*vq
,
585 unsigned int *out_num
, unsigned int *in_num
)
587 unsigned int i
, head
;
588 u16 last_avail
= lg_last_avail(vq
);
590 while (last_avail
== vq
->vring
.avail
->idx
) {
593 /* OK, tell Guest about progress up to now. */
596 /* Nothing new? Wait for eventfd to tell us they refilled. */
597 if (read(vq
->eventfd
, &event
, sizeof(event
)) != sizeof(event
))
598 errx(1, "Event read failed?");
601 /* Check it isn't doing very strange things with descriptor numbers. */
602 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
603 errx(1, "Guest moved used index from %u to %u",
604 last_avail
, vq
->vring
.avail
->idx
);
606 /* Grab the next descriptor number they're advertising, and increment
607 * the index we've seen. */
608 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
611 /* If their number is silly, that's a fatal mistake. */
612 if (head
>= vq
->vring
.num
)
613 errx(1, "Guest says index %u is available", head
);
615 /* When we start there are none of either input nor output. */
616 *out_num
= *in_num
= 0;
620 /* Grab the first descriptor, and check it's OK. */
621 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
622 iov
[*out_num
+ *in_num
].iov_base
623 = check_pointer(vq
->vring
.desc
[i
].addr
,
624 vq
->vring
.desc
[i
].len
);
625 /* If this is an input descriptor, increment that count. */
626 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
629 /* If it's an output descriptor, they're all supposed
630 * to come before any input descriptors. */
632 errx(1, "Descriptor has out after in");
636 /* If we've got too many, that implies a descriptor loop. */
637 if (*out_num
+ *in_num
> vq
->vring
.num
)
638 errx(1, "Looped descriptor");
639 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
644 /* After we've used one of their buffers, we tell them about it. We'll then
645 * want to send them an interrupt, using trigger_irq(). */
646 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
648 struct vring_used_elem
*used
;
650 /* The virtqueue contains a ring of used buffers. Get a pointer to the
651 * next entry in that used ring. */
652 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
655 /* Make sure buffer is written before we update index. */
657 vq
->vring
.used
->idx
++;
661 /* And here's the combo meal deal. Supersize me! */
662 static void add_used_and_trigger(struct virtqueue
*vq
, unsigned head
, int len
)
664 add_used(vq
, head
, len
);
671 * We associate some data with the console for our exit hack. */
674 /* How many times have they hit ^C? */
676 /* When did they start? */
677 struct timeval start
;
680 /* This is the routine which handles console input (ie. stdin). */
681 static void console_input(struct virtqueue
*vq
)
684 unsigned int head
, in_num
, out_num
;
685 struct console_abort
*abort
= vq
->dev
->priv
;
686 struct iovec iov
[vq
->vring
.num
];
688 /* Make sure there's a descriptor waiting. */
689 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
691 errx(1, "Output buffers in console in queue?");
694 len
= readv(STDIN_FILENO
, iov
, in_num
);
696 /* Ran out of input? */
697 warnx("Failed to get console input, ignoring console.");
698 /* For simplicity, dying threads kill the whole Launcher. So
704 add_used_and_trigger(vq
, head
, len
);
706 /* Three ^C within one second? Exit.
708 * This is such a hack, but works surprisingly well. Each ^C has to
709 * be in a buffer by itself, so they can't be too fast. But we check
710 * that we get three within about a second, so they can't be too
712 if (len
!= 1 || ((char *)iov
[0].iov_base
)[0] != 3) {
718 if (abort
->count
== 1)
719 gettimeofday(&abort
->start
, NULL
);
720 else if (abort
->count
== 3) {
722 gettimeofday(&now
, NULL
);
723 /* Kill all Launcher processes with SIGINT, like normal ^C */
724 if (now
.tv_sec
<= abort
->start
.tv_sec
+1)
730 /* This is the routine which handles console output (ie. stdout). */
731 static void console_output(struct virtqueue
*vq
)
733 unsigned int head
, out
, in
;
734 struct iovec iov
[vq
->vring
.num
];
736 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
738 errx(1, "Input buffers in console output queue?");
739 while (!iov_empty(iov
, out
)) {
740 int len
= writev(STDOUT_FILENO
, iov
, out
);
742 err(1, "Write to stdout gave %i", len
);
743 iov_consume(iov
, out
, len
);
745 add_used(vq
, head
, 0);
751 * Handling output for network is also simple: we get all the output buffers
752 * and write them to /dev/net/tun.
758 static void net_output(struct virtqueue
*vq
)
760 struct net_info
*net_info
= vq
->dev
->priv
;
761 unsigned int head
, out
, in
;
762 struct iovec iov
[vq
->vring
.num
];
764 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
766 errx(1, "Input buffers in net output queue?");
767 if (writev(net_info
->tunfd
, iov
, out
) < 0)
768 errx(1, "Write to tun failed?");
769 add_used(vq
, head
, 0);
772 /* Will reading from this file descriptor block? */
773 static bool will_block(int fd
)
776 struct timeval zero
= { 0, 0 };
779 return select(fd
+1, &fdset
, NULL
, NULL
, &zero
) != 1;
782 /* This is where we handle packets coming in from the tun device to our
784 static void net_input(struct virtqueue
*vq
)
787 unsigned int head
, out
, in
;
788 struct iovec iov
[vq
->vring
.num
];
789 struct net_info
*net_info
= vq
->dev
->priv
;
791 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
793 errx(1, "Output buffers in net input queue?");
795 /* Deliver interrupt now, since we're about to sleep. */
796 if (vq
->pending_used
&& will_block(net_info
->tunfd
))
799 len
= readv(net_info
->tunfd
, iov
, in
);
801 err(1, "Failed to read from tun.");
802 add_used(vq
, head
, len
);
805 /* This is the helper to create threads. */
806 static int do_thread(void *_vq
)
808 struct virtqueue
*vq
= _vq
;
815 /* When a child dies, we kill our entire process group with SIGTERM. This
816 * also has the side effect that the shell restores the console for us! */
817 static void kill_launcher(int signal
)
822 static void reset_device(struct device
*dev
)
824 struct virtqueue
*vq
;
826 verbose("Resetting device %s\n", dev
->name
);
828 /* Clear any features they've acked. */
829 memset(get_feature_bits(dev
) + dev
->feature_len
, 0, dev
->feature_len
);
831 /* We're going to be explicitly killing threads, so ignore them. */
832 signal(SIGCHLD
, SIG_IGN
);
834 /* Zero out the virtqueues, get rid of their threads */
835 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
836 if (vq
->thread
!= (pid_t
)-1) {
837 kill(vq
->thread
, SIGTERM
);
838 waitpid(vq
->thread
, NULL
, 0);
839 vq
->thread
= (pid_t
)-1;
841 memset(vq
->vring
.desc
, 0,
842 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
843 lg_last_avail(vq
) = 0;
845 dev
->running
= false;
847 /* Now we care if threads die. */
848 signal(SIGCHLD
, (void *)kill_launcher
);
851 static void create_thread(struct virtqueue
*vq
)
853 /* Create stack for thread and run it. Since stack grows
854 * upwards, we point the stack pointer to the end of this
856 char *stack
= malloc(32768);
857 unsigned long args
[] = { LHREQ_EVENTFD
,
858 vq
->config
.pfn
*getpagesize(), 0 };
860 /* Create a zero-initialized eventfd. */
861 vq
->eventfd
= eventfd(0, 0);
863 err(1, "Creating eventfd");
864 args
[2] = vq
->eventfd
;
866 /* Attach an eventfd to this virtqueue: it will go off
867 * when the Guest does an LHCALL_NOTIFY for this vq. */
868 if (write(lguest_fd
, &args
, sizeof(args
)) != 0)
869 err(1, "Attaching eventfd");
871 /* CLONE_VM: because it has to access the Guest memory, and
872 * SIGCHLD so we get a signal if it dies. */
873 vq
->thread
= clone(do_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, vq
);
874 if (vq
->thread
== (pid_t
)-1)
875 err(1, "Creating clone");
876 /* We close our local copy, now the child has it. */
880 static void start_device(struct device
*dev
)
883 struct virtqueue
*vq
;
885 verbose("Device %s OK: offered", dev
->name
);
886 for (i
= 0; i
< dev
->feature_len
; i
++)
887 verbose(" %02x", get_feature_bits(dev
)[i
]);
888 verbose(", accepted");
889 for (i
= 0; i
< dev
->feature_len
; i
++)
890 verbose(" %02x", get_feature_bits(dev
)
891 [dev
->feature_len
+i
]);
893 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
900 static void cleanup_devices(void)
904 for (dev
= devices
.dev
; dev
; dev
= dev
->next
)
907 /* If we saved off the original terminal settings, restore them now. */
908 if (orig_term
.c_lflag
& (ISIG
|ICANON
|ECHO
))
909 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
912 /* When the Guest tells us they updated the status field, we handle it. */
913 static void update_device_status(struct device
*dev
)
915 /* A zero status is a reset, otherwise it's a set of flags. */
916 if (dev
->desc
->status
== 0)
918 else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
919 warnx("Device %s configuration FAILED", dev
->name
);
922 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
928 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
929 static void handle_output(unsigned long addr
)
933 /* Check each device. */
934 for (i
= devices
.dev
; i
; i
= i
->next
) {
935 struct virtqueue
*vq
;
937 /* Notifications to device descriptors update device status. */
938 if (from_guest_phys(addr
) == i
->desc
) {
939 update_device_status(i
);
943 /* Devices *can* be used before status is set to DRIVER_OK. */
944 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
945 if (addr
!= vq
->config
.pfn
*getpagesize())
948 errx(1, "Notification on running %s", i
->name
);
954 /* Early console write is done using notify on a nul-terminated string
955 * in Guest memory. */
956 if (addr
>= guest_limit
)
957 errx(1, "Bad NOTIFY %#lx", addr
);
959 write(STDOUT_FILENO
, from_guest_phys(addr
),
960 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
966 * All devices need a descriptor so the Guest knows it exists, and a "struct
967 * device" so the Launcher can keep track of it. We have common helper
968 * routines to allocate and manage them.
971 /* The layout of the device page is a "struct lguest_device_desc" followed by a
972 * number of virtqueue descriptors, then two sets of feature bits, then an
973 * array of configuration bytes. This routine returns the configuration
975 static u8
*device_config(const struct device
*dev
)
977 return (void *)(dev
->desc
+ 1)
978 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
979 + dev
->feature_len
* 2;
982 /* This routine allocates a new "struct lguest_device_desc" from descriptor
983 * table page just above the Guest's normal memory. It returns a pointer to
984 * that descriptor. */
985 static struct lguest_device_desc
*new_dev_desc(u16 type
)
987 struct lguest_device_desc d
= { .type
= type
};
990 /* Figure out where the next device config is, based on the last one. */
992 p
= device_config(devices
.lastdev
)
993 + devices
.lastdev
->desc
->config_len
;
995 p
= devices
.descpage
;
997 /* We only have one page for all the descriptors. */
998 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
999 errx(1, "Too many devices");
1001 /* p might not be aligned, so we memcpy in. */
1002 return memcpy(p
, &d
, sizeof(d
));
1005 /* Each device descriptor is followed by the description of its virtqueues. We
1006 * specify how many descriptors the virtqueue is to have. */
1007 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1008 void (*service
)(struct virtqueue
*))
1011 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1014 /* First we need some memory for this virtqueue. */
1015 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
1017 p
= get_pages(pages
);
1019 /* Initialize the virtqueue */
1021 vq
->last_avail_idx
= 0;
1023 vq
->service
= service
;
1024 vq
->thread
= (pid_t
)-1;
1026 /* Initialize the configuration. */
1027 vq
->config
.num
= num_descs
;
1028 vq
->config
.irq
= devices
.next_irq
++;
1029 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1031 /* Initialize the vring. */
1032 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1034 /* Append virtqueue to this device's descriptor. We use
1035 * device_config() to get the end of the device's current virtqueues;
1036 * we check that we haven't added any config or feature information
1037 * yet, otherwise we'd be overwriting them. */
1038 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1039 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1041 dev
->desc
->num_vq
++;
1043 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1045 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1047 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1051 /* The first half of the feature bitmask is for us to advertise features. The
1052 * second half is for the Guest to accept features. */
1053 static void add_feature(struct device
*dev
, unsigned bit
)
1055 u8
*features
= get_feature_bits(dev
);
1057 /* We can't extend the feature bits once we've added config bytes */
1058 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1059 assert(dev
->desc
->config_len
== 0);
1060 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1063 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1066 /* This routine sets the configuration fields for an existing device's
1067 * descriptor. It only works for the last device, but that's OK because that's
1069 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1071 /* Check we haven't overflowed our single page. */
1072 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1073 errx(1, "Too many devices");
1075 /* Copy in the config information, and store the length. */
1076 memcpy(device_config(dev
), conf
, len
);
1077 dev
->desc
->config_len
= len
;
1080 /* This routine does all the creation and setup of a new device, including
1081 * calling new_dev_desc() to allocate the descriptor and device memory.
1083 * See what I mean about userspace being boring? */
1084 static struct device
*new_device(const char *name
, u16 type
)
1086 struct device
*dev
= malloc(sizeof(*dev
));
1088 /* Now we populate the fields one at a time. */
1089 dev
->desc
= new_dev_desc(type
);
1092 dev
->feature_len
= 0;
1094 dev
->running
= false;
1096 /* Append to device list. Prepending to a single-linked list is
1097 * easier, but the user expects the devices to be arranged on the bus
1098 * in command-line order. The first network device on the command line
1099 * is eth0, the first block device /dev/vda, etc. */
1100 if (devices
.lastdev
)
1101 devices
.lastdev
->next
= dev
;
1104 devices
.lastdev
= dev
;
1109 /* Our first setup routine is the console. It's a fairly simple device, but
1110 * UNIX tty handling makes it uglier than it could be. */
1111 static void setup_console(void)
1115 /* If we can save the initial standard input settings... */
1116 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1117 struct termios term
= orig_term
;
1118 /* Then we turn off echo, line buffering and ^C etc. We want a
1119 * raw input stream to the Guest. */
1120 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1121 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1124 dev
= new_device("console", VIRTIO_ID_CONSOLE
);
1126 /* We store the console state in dev->priv, and initialize it. */
1127 dev
->priv
= malloc(sizeof(struct console_abort
));
1128 ((struct console_abort
*)dev
->priv
)->count
= 0;
1130 /* The console needs two virtqueues: the input then the output. When
1131 * they put something the input queue, we make sure we're listening to
1132 * stdin. When they put something in the output queue, we write it to
1134 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_input
);
1135 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_output
);
1137 verbose("device %u: console\n", ++devices
.device_num
);
1141 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1142 * --sharenet=<name> option which opens or creates a named pipe. This can be
1143 * used to send packets to another guest in a 1:1 manner.
1145 * More sopisticated is to use one of the tools developed for project like UML
1148 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1149 * completely generic ("here's my vring, attach to your vring") and would work
1150 * for any traffic. Of course, namespace and permissions issues need to be
1151 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1152 * multiple inter-guest channels behind one interface, although it would
1153 * require some manner of hotplugging new virtio channels.
1155 * Finally, we could implement a virtio network switch in the kernel. :*/
1157 static u32
str2ip(const char *ipaddr
)
1161 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1162 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1163 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1166 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1169 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1170 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1171 errx(1, "Failed to parse mac address '%s'", macaddr
);
1180 /* This code is "adapted" from libbridge: it attaches the Host end of the
1181 * network device to the bridge device specified by the command line.
1183 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1184 * dislike bridging), and I just try not to break it. */
1185 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1191 errx(1, "must specify bridge name");
1193 ifidx
= if_nametoindex(if_name
);
1195 errx(1, "interface %s does not exist!", if_name
);
1197 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1198 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1199 ifr
.ifr_ifindex
= ifidx
;
1200 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1201 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1204 /* This sets up the Host end of the network device with an IP address, brings
1205 * it up so packets will flow, the copies the MAC address into the hwaddr
1207 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1210 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1212 memset(&ifr
, 0, sizeof(ifr
));
1213 strcpy(ifr
.ifr_name
, tapif
);
1215 /* Don't read these incantations. Just cut & paste them like I did! */
1216 sin
->sin_family
= AF_INET
;
1217 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1218 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1219 err(1, "Setting %s interface address", tapif
);
1220 ifr
.ifr_flags
= IFF_UP
;
1221 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1222 err(1, "Bringing interface %s up", tapif
);
1225 static int get_tun_device(char tapif
[IFNAMSIZ
])
1230 /* Start with this zeroed. Messy but sure. */
1231 memset(&ifr
, 0, sizeof(ifr
));
1233 /* We open the /dev/net/tun device and tell it we want a tap device. A
1234 * tap device is like a tun device, only somehow different. To tell
1235 * the truth, I completely blundered my way through this code, but it
1237 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1238 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1239 strcpy(ifr
.ifr_name
, "tap%d");
1240 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1241 err(1, "configuring /dev/net/tun");
1243 if (ioctl(netfd
, TUNSETOFFLOAD
,
1244 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1245 err(1, "Could not set features for tun device");
1247 /* We don't need checksums calculated for packets coming in this
1248 * device: trust us! */
1249 ioctl(netfd
, TUNSETNOCSUM
, 1);
1251 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1255 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1256 * routing, but the principle is the same: it uses the "tun" device to inject
1257 * packets into the Host as if they came in from a normal network card. We
1258 * just shunt packets between the Guest and the tun device. */
1259 static void setup_tun_net(char *arg
)
1262 struct net_info
*net_info
= malloc(sizeof(*net_info
));
1264 u32 ip
= INADDR_ANY
;
1265 bool bridging
= false;
1266 char tapif
[IFNAMSIZ
], *p
;
1267 struct virtio_net_config conf
;
1269 net_info
->tunfd
= get_tun_device(tapif
);
1271 /* First we create a new network device. */
1272 dev
= new_device("net", VIRTIO_ID_NET
);
1273 dev
->priv
= net_info
;
1275 /* Network devices need a receive and a send queue, just like
1277 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_input
);
1278 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_output
);
1280 /* We need a socket to perform the magic network ioctls to bring up the
1281 * tap interface, connect to the bridge etc. Any socket will do! */
1282 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1284 err(1, "opening IP socket");
1286 /* If the command line was --tunnet=bridge:<name> do bridging. */
1287 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1288 arg
+= strlen(BRIDGE_PFX
);
1292 /* A mac address may follow the bridge name or IP address */
1293 p
= strchr(arg
, ':');
1295 str2mac(p
+1, conf
.mac
);
1296 add_feature(dev
, VIRTIO_NET_F_MAC
);
1300 /* arg is now either an IP address or a bridge name */
1302 add_to_bridge(ipfd
, tapif
, arg
);
1306 /* Set up the tun device. */
1307 configure_device(ipfd
, tapif
, ip
);
1309 add_feature(dev
, VIRTIO_F_NOTIFY_ON_EMPTY
);
1310 /* Expect Guest to handle everything except UFO */
1311 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1312 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1313 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1314 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1315 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1316 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1317 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1318 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1319 set_config(dev
, sizeof(conf
), &conf
);
1321 /* We don't need the socket any more; setup is done. */
1324 devices
.device_num
++;
1327 verbose("device %u: tun %s attached to bridge: %s\n",
1328 devices
.device_num
, tapif
, arg
);
1330 verbose("device %u: tun %s: %s\n",
1331 devices
.device_num
, tapif
, arg
);
1334 /* Our block (disk) device should be really simple: the Guest asks for a block
1335 * number and we read or write that position in the file. Unfortunately, that
1336 * was amazingly slow: the Guest waits until the read is finished before
1337 * running anything else, even if it could have been doing useful work.
1339 * We could use async I/O, except it's reputed to suck so hard that characters
1340 * actually go missing from your code when you try to use it.
1342 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1344 /* This hangs off device->priv. */
1347 /* The size of the file. */
1350 /* The file descriptor for the file. */
1353 /* IO thread listens on this file descriptor [0]. */
1356 /* IO thread writes to this file descriptor to mark it done, then
1357 * Launcher triggers interrupt to Guest. */
1364 * Remember that the block device is handled by a separate I/O thread. We head
1365 * straight into the core of that thread here:
1367 static void blk_request(struct virtqueue
*vq
)
1369 struct vblk_info
*vblk
= vq
->dev
->priv
;
1370 unsigned int head
, out_num
, in_num
, wlen
;
1373 struct virtio_blk_outhdr
*out
;
1374 struct iovec iov
[vq
->vring
.num
];
1377 /* Get the next request. */
1378 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1380 /* Every block request should contain at least one output buffer
1381 * (detailing the location on disk and the type of request) and one
1382 * input buffer (to hold the result). */
1383 if (out_num
== 0 || in_num
== 0)
1384 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1385 head
, out_num
, in_num
);
1387 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1388 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1389 off
= out
->sector
* 512;
1391 /* The block device implements "barriers", where the Guest indicates
1392 * that it wants all previous writes to occur before this write. We
1393 * don't have a way of asking our kernel to do a barrier, so we just
1394 * synchronize all the data in the file. Pretty poor, no? */
1395 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1396 fdatasync(vblk
->fd
);
1398 /* In general the virtio block driver is allowed to try SCSI commands.
1399 * It'd be nice if we supported eject, for example, but we don't. */
1400 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1401 fprintf(stderr
, "Scsi commands unsupported\n");
1402 *in
= VIRTIO_BLK_S_UNSUPP
;
1404 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1407 /* Move to the right location in the block file. This can fail
1408 * if they try to write past end. */
1409 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1410 err(1, "Bad seek to sector %llu", out
->sector
);
1412 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1413 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1415 /* Grr... Now we know how long the descriptor they sent was, we
1416 * make sure they didn't try to write over the end of the block
1417 * file (possibly extending it). */
1418 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1419 /* Trim it back to the correct length */
1420 ftruncate64(vblk
->fd
, vblk
->len
);
1421 /* Die, bad Guest, die. */
1422 errx(1, "Write past end %llu+%u", off
, ret
);
1425 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1429 /* Move to the right location in the block file. This can fail
1430 * if they try to read past end. */
1431 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1432 err(1, "Bad seek to sector %llu", out
->sector
);
1434 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1435 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1437 wlen
= sizeof(*in
) + ret
;
1438 *in
= VIRTIO_BLK_S_OK
;
1441 *in
= VIRTIO_BLK_S_IOERR
;
1445 /* OK, so we noted that it was pretty poor to use an fdatasync as a
1446 * barrier. But Christoph Hellwig points out that we need a sync
1447 * *afterwards* as well: "Barriers specify no reordering to the front
1448 * or the back." And Jens Axboe confirmed it, so here we are: */
1449 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1450 fdatasync(vblk
->fd
);
1452 add_used(vq
, head
, wlen
);
1455 /*L:198 This actually sets up a virtual block device. */
1456 static void setup_block_file(const char *filename
)
1459 struct vblk_info
*vblk
;
1460 struct virtio_blk_config conf
;
1462 /* The device responds to return from I/O thread. */
1463 dev
= new_device("block", VIRTIO_ID_BLOCK
);
1465 /* The device has one virtqueue, where the Guest places requests. */
1466 add_virtqueue(dev
, VIRTQUEUE_NUM
, blk_request
);
1468 /* Allocate the room for our own bookkeeping */
1469 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1471 /* First we open the file and store the length. */
1472 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1473 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1475 /* We support barriers. */
1476 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1478 /* Tell Guest how many sectors this device has. */
1479 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1481 /* Tell Guest not to put in too many descriptors at once: two are used
1482 * for the in and out elements. */
1483 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1484 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1486 set_config(dev
, sizeof(conf
), &conf
);
1488 verbose("device %u: virtblock %llu sectors\n",
1489 ++devices
.device_num
, le64_to_cpu(conf
.capacity
));
1496 /* Our random number generator device reads from /dev/random into the Guest's
1497 * input buffers. The usual case is that the Guest doesn't want random numbers
1498 * and so has no buffers although /dev/random is still readable, whereas
1499 * console is the reverse.
1501 * The same logic applies, however. */
1502 static void rng_input(struct virtqueue
*vq
)
1505 unsigned int head
, in_num
, out_num
, totlen
= 0;
1506 struct rng_info
*rng_info
= vq
->dev
->priv
;
1507 struct iovec iov
[vq
->vring
.num
];
1509 /* First we need a buffer from the Guests's virtqueue. */
1510 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1512 errx(1, "Output buffers in rng?");
1514 /* This is why we convert to iovecs: the readv() call uses them, and so
1515 * it reads straight into the Guest's buffer. We loop to make sure we
1517 while (!iov_empty(iov
, in_num
)) {
1518 len
= readv(rng_info
->rfd
, iov
, in_num
);
1520 err(1, "Read from /dev/random gave %i", len
);
1521 iov_consume(iov
, in_num
, len
);
1525 /* Tell the Guest about the new input. */
1526 add_used(vq
, head
, totlen
);
1529 /* And this creates a "hardware" random number device for the Guest. */
1530 static void setup_rng(void)
1533 struct rng_info
*rng_info
= malloc(sizeof(*rng_info
));
1535 rng_info
->rfd
= open_or_die("/dev/random", O_RDONLY
);
1537 /* The device responds to return from I/O thread. */
1538 dev
= new_device("rng", VIRTIO_ID_RNG
);
1539 dev
->priv
= rng_info
;
1541 /* The device has one virtqueue, where the Guest places inbufs. */
1542 add_virtqueue(dev
, VIRTQUEUE_NUM
, rng_input
);
1544 verbose("device %u: rng\n", devices
.device_num
++);
1546 /* That's the end of device setup. */
1548 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1549 static void __attribute__((noreturn
)) restart_guest(void)
1553 /* Since we don't track all open fds, we simply close everything beyond
1555 for (i
= 3; i
< FD_SETSIZE
; i
++)
1558 /* Reset all the devices (kills all threads). */
1561 execv(main_args
[0], main_args
);
1562 err(1, "Could not exec %s", main_args
[0]);
1565 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1566 * its input and output, and finally, lays it to rest. */
1567 static void __attribute__((noreturn
)) run_guest(void)
1570 unsigned long notify_addr
;
1573 /* We read from the /dev/lguest device to run the Guest. */
1574 readval
= pread(lguest_fd
, ¬ify_addr
,
1575 sizeof(notify_addr
), cpu_id
);
1577 /* One unsigned long means the Guest did HCALL_NOTIFY */
1578 if (readval
== sizeof(notify_addr
)) {
1579 verbose("Notify on address %#lx\n", notify_addr
);
1580 handle_output(notify_addr
);
1581 /* ENOENT means the Guest died. Reading tells us why. */
1582 } else if (errno
== ENOENT
) {
1583 char reason
[1024] = { 0 };
1584 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1585 errx(1, "%s", reason
);
1586 /* ERESTART means that we need to reboot the guest */
1587 } else if (errno
== ERESTART
) {
1589 /* Anything else means a bug or incompatible change. */
1591 err(1, "Running guest failed");
1595 * This is the end of the Launcher. The good news: we are over halfway
1596 * through! The bad news: the most fiendish part of the code still lies ahead
1599 * Are you ready? Take a deep breath and join me in the core of the Host, in
1603 static struct option opts
[] = {
1604 { "verbose", 0, NULL
, 'v' },
1605 { "tunnet", 1, NULL
, 't' },
1606 { "block", 1, NULL
, 'b' },
1607 { "rng", 0, NULL
, 'r' },
1608 { "initrd", 1, NULL
, 'i' },
1611 static void usage(void)
1613 errx(1, "Usage: lguest [--verbose] "
1614 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1615 "|--block=<filename>|--initrd=<filename>]...\n"
1616 "<mem-in-mb> vmlinux [args...]");
1619 /*L:105 The main routine is where the real work begins: */
1620 int main(int argc
, char *argv
[])
1622 /* Memory, top-level pagetable, code startpoint and size of the
1623 * (optional) initrd. */
1624 unsigned long mem
= 0, start
, initrd_size
= 0;
1625 /* Two temporaries. */
1627 /* The boot information for the Guest. */
1628 struct boot_params
*boot
;
1629 /* If they specify an initrd file to load. */
1630 const char *initrd_name
= NULL
;
1632 /* Save the args: we "reboot" by execing ourselves again. */
1635 /* First we initialize the device list. We keep a pointer to the last
1636 * device, and the next interrupt number to use for devices (1:
1637 * remember that 0 is used by the timer). */
1638 devices
.lastdev
= NULL
;
1639 devices
.next_irq
= 1;
1642 /* We need to know how much memory so we can set up the device
1643 * descriptor and memory pages for the devices as we parse the command
1644 * line. So we quickly look through the arguments to find the amount
1646 for (i
= 1; i
< argc
; i
++) {
1647 if (argv
[i
][0] != '-') {
1648 mem
= atoi(argv
[i
]) * 1024 * 1024;
1649 /* We start by mapping anonymous pages over all of
1650 * guest-physical memory range. This fills it with 0,
1651 * and ensures that the Guest won't be killed when it
1652 * tries to access it. */
1653 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1656 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1657 devices
.descpage
= get_pages(1);
1662 /* The options are fairly straight-forward */
1663 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1669 setup_tun_net(optarg
);
1672 setup_block_file(optarg
);
1678 initrd_name
= optarg
;
1681 warnx("Unknown argument %s", argv
[optind
]);
1685 /* After the other arguments we expect memory and kernel image name,
1686 * followed by command line arguments for the kernel. */
1687 if (optind
+ 2 > argc
)
1690 verbose("Guest base is at %p\n", guest_base
);
1692 /* We always have a console device */
1695 /* Now we load the kernel */
1696 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1698 /* Boot information is stashed at physical address 0 */
1699 boot
= from_guest_phys(0);
1701 /* Map the initrd image if requested (at top of physical memory) */
1703 initrd_size
= load_initrd(initrd_name
, mem
);
1704 /* These are the location in the Linux boot header where the
1705 * start and size of the initrd are expected to be found. */
1706 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1707 boot
->hdr
.ramdisk_size
= initrd_size
;
1708 /* The bootloader type 0xFF means "unknown"; that's OK. */
1709 boot
->hdr
.type_of_loader
= 0xFF;
1712 /* The Linux boot header contains an "E820" memory map: ours is a
1713 * simple, single region. */
1714 boot
->e820_entries
= 1;
1715 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1716 /* The boot header contains a command line pointer: we put the command
1717 * line after the boot header. */
1718 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1719 /* We use a simple helper to copy the arguments separated by spaces. */
1720 concat((char *)(boot
+ 1), argv
+optind
+2);
1722 /* Boot protocol version: 2.07 supports the fields for lguest. */
1723 boot
->hdr
.version
= 0x207;
1725 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1726 boot
->hdr
.hardware_subarch
= 1;
1728 /* Tell the entry path not to try to reload segment registers. */
1729 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1731 /* We tell the kernel to initialize the Guest: this returns the open
1732 * /dev/lguest file descriptor. */
1735 /* Ensure that we terminate if a child dies. */
1736 signal(SIGCHLD
, kill_launcher
);
1738 /* If we exit via err(), this kills all the threads, restores tty. */
1739 atexit(cleanup_devices
);
1741 /* Finally, run the Guest. This doesn't return. */
1747 * Mastery is done: you now know everything I do.
1749 * But surely you have seen code, features and bugs in your wanderings which
1750 * you now yearn to attack? That is the real game, and I look forward to you
1751 * patching and forking lguest into the Your-Name-Here-visor.
1753 * Farewell, and good coding!