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>
23 #include <sys/socket.h>
24 #include <sys/ioctl.h>
27 #include <netinet/in.h>
29 #include <linux/sockios.h>
30 #include <linux/if_tun.h>
40 #include "linux/lguest_launcher.h"
41 #include "linux/virtio_config.h"
42 #include "linux/virtio_net.h"
43 #include "linux/virtio_blk.h"
44 #include "linux/virtio_console.h"
45 #include "linux/virtio_rng.h"
46 #include "linux/virtio_ring.h"
47 #include "asm/bootparam.h"
48 /*L:110 We can ignore the 39 include files we need for this program, but I do
49 * want to draw attention to the use of kernel-style types.
51 * As Linus said, "C is a Spartan language, and so should your naming be." I
52 * like these abbreviations, so we define them here. Note that u64 is always
53 * unsigned long long, which works on all Linux systems: this means that we can
54 * use %llu in printf for any u64. */
55 typedef unsigned long long u64
;
61 #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 /* File descriptors for the Waker. */
85 /* The pointer to the start of guest memory. */
86 static void *guest_base
;
87 /* The maximum guest physical address allowed, and maximum possible. */
88 static unsigned long guest_limit
, guest_max
;
89 /* The pipe for signal hander to write to. */
90 static int timeoutpipe
[2];
91 static unsigned int timeout_usec
= 500;
93 /* a per-cpu variable indicating whose vcpu is currently running */
94 static unsigned int __thread cpu_id
;
96 /* This is our list of devices. */
99 /* Summary information about the devices in our list: ready to pass to
100 * select() to ask which need servicing.*/
104 /* Counter to assign interrupt numbers. */
105 unsigned int next_irq
;
107 /* Counter to print out convenient device numbers. */
108 unsigned int device_num
;
110 /* The descriptor page for the devices. */
113 /* A single linked list of devices. */
115 /* And a pointer to the last device for easy append and also for
116 * configuration appending. */
117 struct device
*lastdev
;
120 /* The list of Guest devices, based on command line arguments. */
121 static struct device_list devices
;
123 /* The device structure describes a single device. */
126 /* The linked-list pointer. */
129 /* The device's descriptor, as mapped into the Guest. */
130 struct lguest_device_desc
*desc
;
132 /* We can't trust desc values once Guest has booted: we use these. */
133 unsigned int feature_len
;
136 /* The name of this device, for --verbose. */
139 /* If handle_input is set, it wants to be called when this file
140 * descriptor is ready. */
142 bool (*handle_input
)(int fd
, struct device
*me
);
144 /* Any queues attached to this device */
145 struct virtqueue
*vq
;
147 /* Handle status being finalized (ie. feature bits stable). */
148 void (*ready
)(struct device
*me
);
150 /* Device-specific data. */
154 /* The virtqueue structure describes a queue attached to a device. */
157 struct virtqueue
*next
;
159 /* Which device owns me. */
162 /* The configuration for this queue. */
163 struct lguest_vqconfig config
;
165 /* The actual ring of buffers. */
168 /* Last available index we saw. */
171 /* The routine to call when the Guest pings us, or timeout. */
172 void (*handle_output
)(int fd
, struct virtqueue
*me
, bool timeout
);
174 /* Outstanding buffers */
175 unsigned int inflight
;
177 /* Is this blocked awaiting a timer? */
181 /* Remember the arguments to the program so we can "reboot" */
182 static char **main_args
;
184 /* Since guest is UP and we don't run at the same time, we don't need barriers.
185 * But I include them in the code in case others copy it. */
188 /* Convert an iovec element to the given type.
190 * This is a fairly ugly trick: we need to know the size of the type and
191 * alignment requirement to check the pointer is kosher. It's also nice to
192 * have the name of the type in case we report failure.
194 * Typing those three things all the time is cumbersome and error prone, so we
195 * have a macro which sets them all up and passes to the real function. */
196 #define convert(iov, type) \
197 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
199 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
202 if (iov
->iov_len
!= size
)
203 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
204 if ((unsigned long)iov
->iov_base
% align
!= 0)
205 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
206 return iov
->iov_base
;
209 /* Wrapper for the last available index. Makes it easier to change. */
210 #define lg_last_avail(vq) ((vq)->last_avail_idx)
212 /* The virtio configuration space is defined to be little-endian. x86 is
213 * little-endian too, but it's nice to be explicit so we have these helpers. */
214 #define cpu_to_le16(v16) (v16)
215 #define cpu_to_le32(v32) (v32)
216 #define cpu_to_le64(v64) (v64)
217 #define le16_to_cpu(v16) (v16)
218 #define le32_to_cpu(v32) (v32)
219 #define le64_to_cpu(v64) (v64)
221 /* Is this iovec empty? */
222 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
226 for (i
= 0; i
< num_iov
; i
++)
232 /* Take len bytes from the front of this iovec. */
233 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
237 for (i
= 0; i
< num_iov
; i
++) {
240 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
241 iov
[i
].iov_base
+= used
;
242 iov
[i
].iov_len
-= used
;
248 /* The device virtqueue descriptors are followed by feature bitmasks. */
249 static u8
*get_feature_bits(struct device
*dev
)
251 return (u8
*)(dev
->desc
+ 1)
252 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
255 /*L:100 The Launcher code itself takes us out into userspace, that scary place
256 * where pointers run wild and free! Unfortunately, like most userspace
257 * programs, it's quite boring (which is why everyone likes to hack on the
258 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
259 * will get you through this section. Or, maybe not.
261 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
262 * memory and stores it in "guest_base". In other words, Guest physical ==
263 * Launcher virtual with an offset.
265 * This can be tough to get your head around, but usually it just means that we
266 * use these trivial conversion functions when the Guest gives us it's
267 * "physical" addresses: */
268 static void *from_guest_phys(unsigned long addr
)
270 return guest_base
+ addr
;
273 static unsigned long to_guest_phys(const void *addr
)
275 return (addr
- guest_base
);
279 * Loading the Kernel.
281 * We start with couple of simple helper routines. open_or_die() avoids
282 * error-checking code cluttering the callers: */
283 static int open_or_die(const char *name
, int flags
)
285 int fd
= open(name
, flags
);
287 err(1, "Failed to open %s", name
);
291 /* map_zeroed_pages() takes a number of pages. */
292 static void *map_zeroed_pages(unsigned int num
)
294 int fd
= open_or_die("/dev/zero", O_RDONLY
);
297 /* We use a private mapping (ie. if we write to the page, it will be
299 addr
= mmap(NULL
, getpagesize() * num
,
300 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
301 if (addr
== MAP_FAILED
)
302 err(1, "Mmaping %u pages of /dev/zero", num
);
308 /* Get some more pages for a device. */
309 static void *get_pages(unsigned int num
)
311 void *addr
= from_guest_phys(guest_limit
);
313 guest_limit
+= num
* getpagesize();
314 if (guest_limit
> guest_max
)
315 errx(1, "Not enough memory for devices");
319 /* This routine is used to load the kernel or initrd. It tries mmap, but if
320 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
321 * it falls back to reading the memory in. */
322 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
326 /* We map writable even though for some segments are marked read-only.
327 * The kernel really wants to be writable: it patches its own
330 * MAP_PRIVATE means that the page won't be copied until a write is
331 * done to it. This allows us to share untouched memory between
333 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
334 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
337 /* pread does a seek and a read in one shot: saves a few lines. */
338 r
= pread(fd
, addr
, len
, offset
);
340 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
343 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
344 * the Guest memory. ELF = Embedded Linking Format, which is the format used
345 * by all modern binaries on Linux including the kernel.
347 * The ELF headers give *two* addresses: a physical address, and a virtual
348 * address. We use the physical address; the Guest will map itself to the
351 * We return the starting address. */
352 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
354 Elf32_Phdr phdr
[ehdr
->e_phnum
];
357 /* Sanity checks on the main ELF header: an x86 executable with a
358 * reasonable number of correctly-sized program headers. */
359 if (ehdr
->e_type
!= ET_EXEC
360 || ehdr
->e_machine
!= EM_386
361 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
362 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
363 errx(1, "Malformed elf header");
365 /* An ELF executable contains an ELF header and a number of "program"
366 * headers which indicate which parts ("segments") of the program to
369 /* We read in all the program headers at once: */
370 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
371 err(1, "Seeking to program headers");
372 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
373 err(1, "Reading program headers");
375 /* Try all the headers: there are usually only three. A read-only one,
376 * a read-write one, and a "note" section which we don't load. */
377 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
378 /* If this isn't a loadable segment, we ignore it */
379 if (phdr
[i
].p_type
!= PT_LOAD
)
382 verbose("Section %i: size %i addr %p\n",
383 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
385 /* We map this section of the file at its physical address. */
386 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
387 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
390 /* The entry point is given in the ELF header. */
391 return ehdr
->e_entry
;
394 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
395 * supposed to jump into it and it will unpack itself. We used to have to
396 * perform some hairy magic because the unpacking code scared me.
398 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
399 * a small patch to jump over the tricky bits in the Guest, so now we just read
400 * the funky header so we know where in the file to load, and away we go! */
401 static unsigned long load_bzimage(int fd
)
403 struct boot_params boot
;
405 /* Modern bzImages get loaded at 1M. */
406 void *p
= from_guest_phys(0x100000);
408 /* Go back to the start of the file and read the header. It should be
409 * a Linux boot header (see Documentation/x86/i386/boot.txt) */
410 lseek(fd
, 0, SEEK_SET
);
411 read(fd
, &boot
, sizeof(boot
));
413 /* Inside the setup_hdr, we expect the magic "HdrS" */
414 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
415 errx(1, "This doesn't look like a bzImage to me");
417 /* Skip over the extra sectors of the header. */
418 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
420 /* Now read everything into memory. in nice big chunks. */
421 while ((r
= read(fd
, p
, 65536)) > 0)
424 /* Finally, code32_start tells us where to enter the kernel. */
425 return boot
.hdr
.code32_start
;
428 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
429 * come wrapped up in the self-decompressing "bzImage" format. With a little
430 * work, we can load those, too. */
431 static unsigned long load_kernel(int fd
)
435 /* Read in the first few bytes. */
436 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
437 err(1, "Reading kernel");
439 /* If it's an ELF file, it starts with "\177ELF" */
440 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
441 return map_elf(fd
, &hdr
);
443 /* Otherwise we assume it's a bzImage, and try to load it. */
444 return load_bzimage(fd
);
447 /* This is a trivial little helper to align pages. Andi Kleen hated it because
448 * it calls getpagesize() twice: "it's dumb code."
450 * Kernel guys get really het up about optimization, even when it's not
451 * necessary. I leave this code as a reaction against that. */
452 static inline unsigned long page_align(unsigned long addr
)
454 /* Add upwards and truncate downwards. */
455 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
458 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
459 * the kernel which the kernel can use to boot from without needing any
460 * drivers. Most distributions now use this as standard: the initrd contains
461 * the code to load the appropriate driver modules for the current machine.
463 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
464 * kernels. He sent me this (and tells me when I break it). */
465 static unsigned long load_initrd(const char *name
, unsigned long mem
)
471 ifd
= open_or_die(name
, O_RDONLY
);
472 /* fstat() is needed to get the file size. */
473 if (fstat(ifd
, &st
) < 0)
474 err(1, "fstat() on initrd '%s'", name
);
476 /* We map the initrd at the top of memory, but mmap wants it to be
477 * page-aligned, so we round the size up for that. */
478 len
= page_align(st
.st_size
);
479 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
480 /* Once a file is mapped, you can close the file descriptor. It's a
481 * little odd, but quite useful. */
483 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
485 /* We return the initrd size. */
490 /* Simple routine to roll all the commandline arguments together with spaces
492 static void concat(char *dst
, char *args
[])
494 unsigned int i
, len
= 0;
496 for (i
= 0; args
[i
]; i
++) {
498 strcat(dst
+len
, " ");
501 strcpy(dst
+len
, args
[i
]);
502 len
+= strlen(args
[i
]);
504 /* In case it's empty. */
508 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
509 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
510 * the base of Guest "physical" memory, the top physical page to allow and the
511 * entry point for the Guest. */
512 static int tell_kernel(unsigned long start
)
514 unsigned long args
[] = { LHREQ_INITIALIZE
,
515 (unsigned long)guest_base
,
516 guest_limit
/ getpagesize(), start
};
519 verbose("Guest: %p - %p (%#lx)\n",
520 guest_base
, guest_base
+ guest_limit
, guest_limit
);
521 fd
= open_or_die("/dev/lguest", O_RDWR
);
522 if (write(fd
, args
, sizeof(args
)) < 0)
523 err(1, "Writing to /dev/lguest");
525 /* We return the /dev/lguest file descriptor to control this Guest */
530 static void add_device_fd(int fd
)
532 FD_SET(fd
, &devices
.infds
);
533 if (fd
> devices
.max_infd
)
534 devices
.max_infd
= fd
;
540 * With console, block and network devices, we can have lots of input which we
541 * need to process. We could try to tell the kernel what file descriptors to
542 * watch, but handing a file descriptor mask through to the kernel is fairly
545 * Instead, we clone off a thread which watches the file descriptors and writes
546 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
547 * stop running the Guest. This causes the Launcher to return from the
548 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
549 * the LHREQ_BREAK and wake us up again.
551 * This, of course, is merely a different *kind* of icky.
553 * Given my well-known antipathy to threads, I'd prefer to use processes. But
554 * it's easier to share Guest memory with threads, and trivial to share the
555 * devices.infds as the Launcher changes it.
557 static int waker(void *unused
)
559 /* Close the write end of the pipe: only the Launcher has it open. */
560 close(waker_fds
.pipe
[1]);
563 fd_set rfds
= devices
.infds
;
564 unsigned long args
[] = { LHREQ_BREAK
, 1 };
565 unsigned int maxfd
= devices
.max_infd
;
567 /* We also listen to the pipe from the Launcher. */
568 FD_SET(waker_fds
.pipe
[0], &rfds
);
569 if (waker_fds
.pipe
[0] > maxfd
)
570 maxfd
= waker_fds
.pipe
[0];
572 /* Wait until input is ready from one of the devices. */
573 select(maxfd
+1, &rfds
, NULL
, NULL
, NULL
);
575 /* Message from Launcher? */
576 if (FD_ISSET(waker_fds
.pipe
[0], &rfds
)) {
578 /* If this fails, then assume Launcher has exited.
579 * Don't do anything on exit: we're just a thread! */
580 if (read(waker_fds
.pipe
[0], &c
, 1) != 1)
585 /* Send LHREQ_BREAK command to snap the Launcher out of it. */
586 pwrite(waker_fds
.lguest_fd
, args
, sizeof(args
), cpu_id
);
591 /* This routine just sets up a pipe to the Waker process. */
592 static void setup_waker(int lguest_fd
)
594 /* This pipe is closed when Launcher dies, telling Waker. */
595 if (pipe(waker_fds
.pipe
) != 0)
596 err(1, "Creating pipe for Waker");
598 /* Waker also needs to know the lguest fd */
599 waker_fds
.lguest_fd
= lguest_fd
;
601 if (clone(waker
, malloc(4096) + 4096, CLONE_VM
| SIGCHLD
, NULL
) == -1)
602 err(1, "Creating Waker");
608 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
609 * We need to make sure it's not trying to reach into the Launcher itself, so
610 * we have a convenient routine which checks it and exits with an error message
611 * if something funny is going on:
613 static void *_check_pointer(unsigned long addr
, unsigned int size
,
616 /* We have to separately check addr and addr+size, because size could
617 * be huge and addr + size might wrap around. */
618 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
619 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
620 /* We return a pointer for the caller's convenience, now we know it's
622 return from_guest_phys(addr
);
624 /* A macro which transparently hands the line number to the real function. */
625 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
627 /* Each buffer in the virtqueues is actually a chain of descriptors. This
628 * function returns the next descriptor in the chain, or vq->vring.num if we're
630 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
634 /* If this descriptor says it doesn't chain, we're done. */
635 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
636 return vq
->vring
.num
;
638 /* Check they're not leading us off end of descriptors. */
639 next
= vq
->vring
.desc
[i
].next
;
640 /* Make sure compiler knows to grab that: we don't want it changing! */
643 if (next
>= vq
->vring
.num
)
644 errx(1, "Desc next is %u", next
);
649 /* This looks in the virtqueue and for the first available buffer, and converts
650 * it to an iovec for convenient access. Since descriptors consist of some
651 * number of output then some number of input descriptors, it's actually two
652 * iovecs, but we pack them into one and note how many of each there were.
654 * This function returns the descriptor number found, or vq->vring.num (which
655 * is never a valid descriptor number) if none was found. */
656 static unsigned get_vq_desc(struct virtqueue
*vq
,
658 unsigned int *out_num
, unsigned int *in_num
)
660 unsigned int i
, head
;
663 /* Check it isn't doing very strange things with descriptor numbers. */
664 last_avail
= lg_last_avail(vq
);
665 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
666 errx(1, "Guest moved used index from %u to %u",
667 last_avail
, vq
->vring
.avail
->idx
);
669 /* If there's nothing new since last we looked, return invalid. */
670 if (vq
->vring
.avail
->idx
== last_avail
)
671 return vq
->vring
.num
;
673 /* Grab the next descriptor number they're advertising, and increment
674 * the index we've seen. */
675 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
678 /* If their number is silly, that's a fatal mistake. */
679 if (head
>= vq
->vring
.num
)
680 errx(1, "Guest says index %u is available", head
);
682 /* When we start there are none of either input nor output. */
683 *out_num
= *in_num
= 0;
687 /* Grab the first descriptor, and check it's OK. */
688 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
689 iov
[*out_num
+ *in_num
].iov_base
690 = check_pointer(vq
->vring
.desc
[i
].addr
,
691 vq
->vring
.desc
[i
].len
);
692 /* If this is an input descriptor, increment that count. */
693 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
696 /* If it's an output descriptor, they're all supposed
697 * to come before any input descriptors. */
699 errx(1, "Descriptor has out after in");
703 /* If we've got too many, that implies a descriptor loop. */
704 if (*out_num
+ *in_num
> vq
->vring
.num
)
705 errx(1, "Looped descriptor");
706 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
712 /* After we've used one of their buffers, we tell them about it. We'll then
713 * want to send them an interrupt, using trigger_irq(). */
714 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
716 struct vring_used_elem
*used
;
718 /* The virtqueue contains a ring of used buffers. Get a pointer to the
719 * next entry in that used ring. */
720 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
723 /* Make sure buffer is written before we update index. */
725 vq
->vring
.used
->idx
++;
729 /* This actually sends the interrupt for this virtqueue */
730 static void trigger_irq(int fd
, struct virtqueue
*vq
)
732 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
734 /* If they don't want an interrupt, don't send one, unless empty. */
735 if ((vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
739 /* Send the Guest an interrupt tell them we used something up. */
740 if (write(fd
, buf
, sizeof(buf
)) != 0)
741 err(1, "Triggering irq %i", vq
->config
.irq
);
744 /* And here's the combo meal deal. Supersize me! */
745 static void add_used_and_trigger(int fd
, struct virtqueue
*vq
,
746 unsigned int head
, int len
)
748 add_used(vq
, head
, len
);
755 * Here is the input terminal setting we save, and the routine to restore them
756 * on exit so the user gets their terminal back. */
757 static struct termios orig_term
;
758 static void restore_term(void)
760 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
763 /* We associate some data with the console for our exit hack. */
766 /* How many times have they hit ^C? */
768 /* When did they start? */
769 struct timeval start
;
772 /* This is the routine which handles console input (ie. stdin). */
773 static bool handle_console_input(int fd
, struct device
*dev
)
776 unsigned int head
, in_num
, out_num
;
777 struct iovec iov
[dev
->vq
->vring
.num
];
778 struct console_abort
*abort
= dev
->priv
;
780 /* First we need a console buffer from the Guests's input virtqueue. */
781 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
783 /* If they're not ready for input, stop listening to this file
784 * descriptor. We'll start again once they add an input buffer. */
785 if (head
== dev
->vq
->vring
.num
)
789 errx(1, "Output buffers in console in queue?");
791 /* This is why we convert to iovecs: the readv() call uses them, and so
792 * it reads straight into the Guest's buffer. */
793 len
= readv(dev
->fd
, iov
, in_num
);
795 /* This implies that the console is closed, is /dev/null, or
796 * something went terribly wrong. */
797 warnx("Failed to get console input, ignoring console.");
798 /* Put the input terminal back. */
800 /* Remove callback from input vq, so it doesn't restart us. */
801 dev
->vq
->handle_output
= NULL
;
802 /* Stop listening to this fd: don't call us again. */
806 /* Tell the Guest about the new input. */
807 add_used_and_trigger(fd
, dev
->vq
, head
, len
);
809 /* Three ^C within one second? Exit.
811 * This is such a hack, but works surprisingly well. Each ^C has to be
812 * in a buffer by itself, so they can't be too fast. But we check that
813 * we get three within about a second, so they can't be too slow. */
814 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
816 gettimeofday(&abort
->start
, NULL
);
817 else if (abort
->count
== 3) {
819 gettimeofday(&now
, NULL
);
820 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
821 unsigned long args
[] = { LHREQ_BREAK
, 0 };
822 /* Close the fd so Waker will know it has to
824 close(waker_fds
.pipe
[1]);
825 /* Just in case Waker is blocked in BREAK, send
827 write(fd
, args
, sizeof(args
));
833 /* Any other key resets the abort counter. */
836 /* Everything went OK! */
840 /* Handling output for console is simple: we just get all the output buffers
841 * and write them to stdout. */
842 static void handle_console_output(int fd
, struct virtqueue
*vq
, bool timeout
)
844 unsigned int head
, out
, in
;
846 struct iovec iov
[vq
->vring
.num
];
848 /* Keep getting output buffers from the Guest until we run out. */
849 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
851 errx(1, "Input buffers in output queue?");
852 len
= writev(STDOUT_FILENO
, iov
, out
);
853 add_used_and_trigger(fd
, vq
, head
, len
);
857 /* This is called when we no longer want to hear about Guest changes to a
858 * virtqueue. This is more efficient in high-traffic cases, but it means we
859 * have to set a timer to check if any more changes have occurred. */
860 static void block_vq(struct virtqueue
*vq
)
862 struct itimerval itm
;
864 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
867 itm
.it_interval
.tv_sec
= 0;
868 itm
.it_interval
.tv_usec
= 0;
869 itm
.it_value
.tv_sec
= 0;
870 itm
.it_value
.tv_usec
= timeout_usec
;
872 setitimer(ITIMER_REAL
, &itm
, NULL
);
878 * Handling output for network is also simple: we get all the output buffers
879 * and write them (ignoring the first element) to this device's file descriptor
882 static void handle_net_output(int fd
, struct virtqueue
*vq
, bool timeout
)
884 unsigned int head
, out
, in
, num
= 0;
886 struct iovec iov
[vq
->vring
.num
];
887 static int last_timeout_num
;
889 /* Keep getting output buffers from the Guest until we run out. */
890 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
892 errx(1, "Input buffers in output queue?");
893 len
= writev(vq
->dev
->fd
, iov
, out
);
895 err(1, "Writing network packet to tun");
896 add_used_and_trigger(fd
, vq
, head
, len
);
900 /* Block further kicks and set up a timer if we saw anything. */
904 /* We never quite know how long should we wait before we check the
905 * queue again for more packets. We start at 500 microseconds, and if
906 * we get fewer packets than last time, we assume we made the timeout
907 * too small and increase it by 10 microseconds. Otherwise, we drop it
908 * by one microsecond every time. It seems to work well enough. */
910 if (num
< last_timeout_num
)
912 else if (timeout_usec
> 1)
914 last_timeout_num
= num
;
918 /* This is where we handle a packet coming in from the tun device to our
920 static bool handle_tun_input(int fd
, struct device
*dev
)
922 unsigned int head
, in_num
, out_num
;
924 struct iovec iov
[dev
->vq
->vring
.num
];
926 /* First we need a network buffer from the Guests's recv virtqueue. */
927 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
928 if (head
== dev
->vq
->vring
.num
) {
929 /* Now, it's expected that if we try to send a packet too
930 * early, the Guest won't be ready yet. Wait until the device
931 * status says it's ready. */
932 /* FIXME: Actually want DRIVER_ACTIVE here. */
934 /* Now tell it we want to know if new things appear. */
935 dev
->vq
->vring
.used
->flags
&= ~VRING_USED_F_NO_NOTIFY
;
938 /* We'll turn this back on if input buffers are registered. */
941 errx(1, "Output buffers in network recv queue?");
943 /* Read the packet from the device directly into the Guest's buffer. */
944 len
= readv(dev
->fd
, iov
, in_num
);
946 err(1, "reading network");
948 /* Tell the Guest about the new packet. */
949 add_used_and_trigger(fd
, dev
->vq
, head
, len
);
951 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
952 ((u8
*)iov
[1].iov_base
)[0], ((u8
*)iov
[1].iov_base
)[1],
953 head
!= dev
->vq
->vring
.num
? "sent" : "discarded");
959 /*L:215 This is the callback attached to the network and console input
960 * virtqueues: it ensures we try again, in case we stopped console or net
961 * delivery because Guest didn't have any buffers. */
962 static void enable_fd(int fd
, struct virtqueue
*vq
, bool timeout
)
964 add_device_fd(vq
->dev
->fd
);
965 /* Snap the Waker out of its select loop. */
966 write(waker_fds
.pipe
[1], "", 1);
969 static void net_enable_fd(int fd
, struct virtqueue
*vq
, bool timeout
)
971 /* We don't need to know again when Guest refills receive buffer. */
972 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
973 enable_fd(fd
, vq
, timeout
);
976 /* When the Guest tells us they updated the status field, we handle it. */
977 static void update_device_status(struct device
*dev
)
979 struct virtqueue
*vq
;
981 /* This is a reset. */
982 if (dev
->desc
->status
== 0) {
983 verbose("Resetting device %s\n", dev
->name
);
985 /* Clear any features they've acked. */
986 memset(get_feature_bits(dev
) + dev
->feature_len
, 0,
989 /* Zero out the virtqueues. */
990 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
991 memset(vq
->vring
.desc
, 0,
992 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
993 lg_last_avail(vq
) = 0;
995 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
996 warnx("Device %s configuration FAILED", dev
->name
);
997 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
1000 verbose("Device %s OK: offered", dev
->name
);
1001 for (i
= 0; i
< dev
->feature_len
; i
++)
1002 verbose(" %02x", get_feature_bits(dev
)[i
]);
1003 verbose(", accepted");
1004 for (i
= 0; i
< dev
->feature_len
; i
++)
1005 verbose(" %02x", get_feature_bits(dev
)
1006 [dev
->feature_len
+i
]);
1013 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1014 static void handle_output(int fd
, unsigned long addr
)
1017 struct virtqueue
*vq
;
1019 /* Check each device and virtqueue. */
1020 for (i
= devices
.dev
; i
; i
= i
->next
) {
1021 /* Notifications to device descriptors update device status. */
1022 if (from_guest_phys(addr
) == i
->desc
) {
1023 update_device_status(i
);
1027 /* Notifications to virtqueues mean output has occurred. */
1028 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1029 if (vq
->config
.pfn
!= addr
/getpagesize())
1032 /* Guest should acknowledge (and set features!) before
1033 * using the device. */
1034 if (i
->desc
->status
== 0) {
1035 warnx("%s gave early output", i
->name
);
1039 if (strcmp(vq
->dev
->name
, "console") != 0)
1040 verbose("Output to %s\n", vq
->dev
->name
);
1041 if (vq
->handle_output
)
1042 vq
->handle_output(fd
, vq
, false);
1047 /* Early console write is done using notify on a nul-terminated string
1048 * in Guest memory. */
1049 if (addr
>= guest_limit
)
1050 errx(1, "Bad NOTIFY %#lx", addr
);
1052 write(STDOUT_FILENO
, from_guest_phys(addr
),
1053 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1056 static void handle_timeout(int fd
)
1060 struct virtqueue
*vq
;
1062 /* Clear the pipe */
1063 read(timeoutpipe
[0], buf
, sizeof(buf
));
1065 /* Check each device and virtqueue: flush blocked ones. */
1066 for (i
= devices
.dev
; i
; i
= i
->next
) {
1067 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1071 vq
->vring
.used
->flags
&= ~VRING_USED_F_NO_NOTIFY
;
1072 vq
->blocked
= false;
1073 if (vq
->handle_output
)
1074 vq
->handle_output(fd
, vq
, true);
1079 /* This is called when the Waker wakes us up: check for incoming file
1081 static void handle_input(int fd
)
1083 /* select() wants a zeroed timeval to mean "don't wait". */
1084 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
1088 fd_set fds
= devices
.infds
;
1091 num
= select(devices
.max_infd
+1, &fds
, NULL
, NULL
, &poll
);
1092 /* Could get interrupted */
1095 /* If nothing is ready, we're done. */
1099 /* Otherwise, call the device(s) which have readable file
1100 * descriptors and a method of handling them. */
1101 for (i
= devices
.dev
; i
; i
= i
->next
) {
1102 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
1103 if (i
->handle_input(fd
, i
))
1106 /* If handle_input() returns false, it means we
1107 * should no longer service it. Networking and
1108 * console do this when there's no input
1109 * buffers to deliver into. Console also uses
1110 * it when it discovers that stdin is closed. */
1111 FD_CLR(i
->fd
, &devices
.infds
);
1115 /* Is this the timeout fd? */
1116 if (FD_ISSET(timeoutpipe
[0], &fds
))
1124 * All devices need a descriptor so the Guest knows it exists, and a "struct
1125 * device" so the Launcher can keep track of it. We have common helper
1126 * routines to allocate and manage them.
1129 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1130 * number of virtqueue descriptors, then two sets of feature bits, then an
1131 * array of configuration bytes. This routine returns the configuration
1133 static u8
*device_config(const struct device
*dev
)
1135 return (void *)(dev
->desc
+ 1)
1136 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
1137 + dev
->feature_len
* 2;
1140 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1141 * table page just above the Guest's normal memory. It returns a pointer to
1142 * that descriptor. */
1143 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1145 struct lguest_device_desc d
= { .type
= type
};
1148 /* Figure out where the next device config is, based on the last one. */
1149 if (devices
.lastdev
)
1150 p
= device_config(devices
.lastdev
)
1151 + devices
.lastdev
->desc
->config_len
;
1153 p
= devices
.descpage
;
1155 /* We only have one page for all the descriptors. */
1156 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1157 errx(1, "Too many devices");
1159 /* p might not be aligned, so we memcpy in. */
1160 return memcpy(p
, &d
, sizeof(d
));
1163 /* Each device descriptor is followed by the description of its virtqueues. We
1164 * specify how many descriptors the virtqueue is to have. */
1165 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1166 void (*handle_output
)(int, struct virtqueue
*, bool))
1169 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1172 /* First we need some memory for this virtqueue. */
1173 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
1175 p
= get_pages(pages
);
1177 /* Initialize the virtqueue */
1179 vq
->last_avail_idx
= 0;
1182 vq
->blocked
= false;
1184 /* Initialize the configuration. */
1185 vq
->config
.num
= num_descs
;
1186 vq
->config
.irq
= devices
.next_irq
++;
1187 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1189 /* Initialize the vring. */
1190 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1192 /* Append virtqueue to this device's descriptor. We use
1193 * device_config() to get the end of the device's current virtqueues;
1194 * we check that we haven't added any config or feature information
1195 * yet, otherwise we'd be overwriting them. */
1196 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1197 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1199 dev
->desc
->num_vq
++;
1201 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1203 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1205 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1208 /* Set the routine to call when the Guest does something to this
1210 vq
->handle_output
= handle_output
;
1212 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1213 * don't have a handler */
1215 vq
->vring
.used
->flags
= VRING_USED_F_NO_NOTIFY
;
1218 /* The first half of the feature bitmask is for us to advertise features. The
1219 * second half is for the Guest to accept features. */
1220 static void add_feature(struct device
*dev
, unsigned bit
)
1222 u8
*features
= get_feature_bits(dev
);
1224 /* We can't extend the feature bits once we've added config bytes */
1225 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1226 assert(dev
->desc
->config_len
== 0);
1227 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1230 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1233 /* This routine sets the configuration fields for an existing device's
1234 * descriptor. It only works for the last device, but that's OK because that's
1236 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1238 /* Check we haven't overflowed our single page. */
1239 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1240 errx(1, "Too many devices");
1242 /* Copy in the config information, and store the length. */
1243 memcpy(device_config(dev
), conf
, len
);
1244 dev
->desc
->config_len
= len
;
1247 /* This routine does all the creation and setup of a new device, including
1248 * calling new_dev_desc() to allocate the descriptor and device memory.
1250 * See what I mean about userspace being boring? */
1251 static struct device
*new_device(const char *name
, u16 type
, int fd
,
1252 bool (*handle_input
)(int, struct device
*))
1254 struct device
*dev
= malloc(sizeof(*dev
));
1256 /* Now we populate the fields one at a time. */
1258 /* If we have an input handler for this file descriptor, then we add it
1259 * to the device_list's fdset and maxfd. */
1261 add_device_fd(dev
->fd
);
1262 dev
->desc
= new_dev_desc(type
);
1263 dev
->handle_input
= handle_input
;
1267 dev
->feature_len
= 0;
1270 /* Append to device list. Prepending to a single-linked list is
1271 * easier, but the user expects the devices to be arranged on the bus
1272 * in command-line order. The first network device on the command line
1273 * is eth0, the first block device /dev/vda, etc. */
1274 if (devices
.lastdev
)
1275 devices
.lastdev
->next
= dev
;
1278 devices
.lastdev
= dev
;
1283 /* Our first setup routine is the console. It's a fairly simple device, but
1284 * UNIX tty handling makes it uglier than it could be. */
1285 static void setup_console(void)
1289 /* If we can save the initial standard input settings... */
1290 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1291 struct termios term
= orig_term
;
1292 /* Then we turn off echo, line buffering and ^C etc. We want a
1293 * raw input stream to the Guest. */
1294 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1295 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1296 /* If we exit gracefully, the original settings will be
1297 * restored so the user can see what they're typing. */
1298 atexit(restore_term
);
1301 dev
= new_device("console", VIRTIO_ID_CONSOLE
,
1302 STDIN_FILENO
, handle_console_input
);
1303 /* We store the console state in dev->priv, and initialize it. */
1304 dev
->priv
= malloc(sizeof(struct console_abort
));
1305 ((struct console_abort
*)dev
->priv
)->count
= 0;
1307 /* The console needs two virtqueues: the input then the output. When
1308 * they put something the input queue, we make sure we're listening to
1309 * stdin. When they put something in the output queue, we write it to
1311 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1312 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_console_output
);
1314 verbose("device %u: console\n", devices
.device_num
++);
1318 static void timeout_alarm(int sig
)
1320 write(timeoutpipe
[1], "", 1);
1323 static void setup_timeout(void)
1325 if (pipe(timeoutpipe
) != 0)
1326 err(1, "Creating timeout pipe");
1328 if (fcntl(timeoutpipe
[1], F_SETFL
,
1329 fcntl(timeoutpipe
[1], F_GETFL
) | O_NONBLOCK
) != 0)
1330 err(1, "Making timeout pipe nonblocking");
1332 add_device_fd(timeoutpipe
[0]);
1333 signal(SIGALRM
, timeout_alarm
);
1336 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1337 * --sharenet=<name> option which opens or creates a named pipe. This can be
1338 * used to send packets to another guest in a 1:1 manner.
1340 * More sopisticated is to use one of the tools developed for project like UML
1343 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1344 * completely generic ("here's my vring, attach to your vring") and would work
1345 * for any traffic. Of course, namespace and permissions issues need to be
1346 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1347 * multiple inter-guest channels behind one interface, although it would
1348 * require some manner of hotplugging new virtio channels.
1350 * Finally, we could implement a virtio network switch in the kernel. :*/
1352 static u32
str2ip(const char *ipaddr
)
1356 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1357 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1358 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1361 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1364 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1365 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1366 errx(1, "Failed to parse mac address '%s'", macaddr
);
1375 /* This code is "adapted" from libbridge: it attaches the Host end of the
1376 * network device to the bridge device specified by the command line.
1378 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1379 * dislike bridging), and I just try not to break it. */
1380 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1386 errx(1, "must specify bridge name");
1388 ifidx
= if_nametoindex(if_name
);
1390 errx(1, "interface %s does not exist!", if_name
);
1392 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1393 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1394 ifr
.ifr_ifindex
= ifidx
;
1395 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1396 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1399 /* This sets up the Host end of the network device with an IP address, brings
1400 * it up so packets will flow, the copies the MAC address into the hwaddr
1402 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1405 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1407 memset(&ifr
, 0, sizeof(ifr
));
1408 strcpy(ifr
.ifr_name
, tapif
);
1410 /* Don't read these incantations. Just cut & paste them like I did! */
1411 sin
->sin_family
= AF_INET
;
1412 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1413 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1414 err(1, "Setting %s interface address", tapif
);
1415 ifr
.ifr_flags
= IFF_UP
;
1416 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1417 err(1, "Bringing interface %s up", tapif
);
1420 static int get_tun_device(char tapif
[IFNAMSIZ
])
1425 /* Start with this zeroed. Messy but sure. */
1426 memset(&ifr
, 0, sizeof(ifr
));
1428 /* We open the /dev/net/tun device and tell it we want a tap device. A
1429 * tap device is like a tun device, only somehow different. To tell
1430 * the truth, I completely blundered my way through this code, but it
1432 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1433 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1434 strcpy(ifr
.ifr_name
, "tap%d");
1435 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1436 err(1, "configuring /dev/net/tun");
1438 if (ioctl(netfd
, TUNSETOFFLOAD
,
1439 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1440 err(1, "Could not set features for tun device");
1442 /* We don't need checksums calculated for packets coming in this
1443 * device: trust us! */
1444 ioctl(netfd
, TUNSETNOCSUM
, 1);
1446 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1450 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1451 * routing, but the principle is the same: it uses the "tun" device to inject
1452 * packets into the Host as if they came in from a normal network card. We
1453 * just shunt packets between the Guest and the tun device. */
1454 static void setup_tun_net(char *arg
)
1458 u32 ip
= INADDR_ANY
;
1459 bool bridging
= false;
1460 char tapif
[IFNAMSIZ
], *p
;
1461 struct virtio_net_config conf
;
1463 netfd
= get_tun_device(tapif
);
1465 /* First we create a new network device. */
1466 dev
= new_device("net", VIRTIO_ID_NET
, netfd
, handle_tun_input
);
1468 /* Network devices need a receive and a send queue, just like
1470 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_enable_fd
);
1471 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_net_output
);
1473 /* We need a socket to perform the magic network ioctls to bring up the
1474 * tap interface, connect to the bridge etc. Any socket will do! */
1475 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1477 err(1, "opening IP socket");
1479 /* If the command line was --tunnet=bridge:<name> do bridging. */
1480 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1481 arg
+= strlen(BRIDGE_PFX
);
1485 /* A mac address may follow the bridge name or IP address */
1486 p
= strchr(arg
, ':');
1488 str2mac(p
+1, conf
.mac
);
1489 add_feature(dev
, VIRTIO_NET_F_MAC
);
1493 /* arg is now either an IP address or a bridge name */
1495 add_to_bridge(ipfd
, tapif
, arg
);
1499 /* Set up the tun device. */
1500 configure_device(ipfd
, tapif
, ip
);
1502 add_feature(dev
, VIRTIO_F_NOTIFY_ON_EMPTY
);
1503 /* Expect Guest to handle everything except UFO */
1504 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1505 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1506 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1507 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1508 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1509 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1510 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1511 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1512 set_config(dev
, sizeof(conf
), &conf
);
1514 /* We don't need the socket any more; setup is done. */
1517 devices
.device_num
++;
1520 verbose("device %u: tun %s attached to bridge: %s\n",
1521 devices
.device_num
, tapif
, arg
);
1523 verbose("device %u: tun %s: %s\n",
1524 devices
.device_num
, tapif
, arg
);
1527 /* Our block (disk) device should be really simple: the Guest asks for a block
1528 * number and we read or write that position in the file. Unfortunately, that
1529 * was amazingly slow: the Guest waits until the read is finished before
1530 * running anything else, even if it could have been doing useful work.
1532 * We could use async I/O, except it's reputed to suck so hard that characters
1533 * actually go missing from your code when you try to use it.
1535 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1537 /* This hangs off device->priv. */
1540 /* The size of the file. */
1543 /* The file descriptor for the file. */
1546 /* IO thread listens on this file descriptor [0]. */
1549 /* IO thread writes to this file descriptor to mark it done, then
1550 * Launcher triggers interrupt to Guest. */
1557 * Remember that the block device is handled by a separate I/O thread. We head
1558 * straight into the core of that thread here:
1560 static bool service_io(struct device
*dev
)
1562 struct vblk_info
*vblk
= dev
->priv
;
1563 unsigned int head
, out_num
, in_num
, wlen
;
1566 struct virtio_blk_outhdr
*out
;
1567 struct iovec iov
[dev
->vq
->vring
.num
];
1570 /* See if there's a request waiting. If not, nothing to do. */
1571 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1572 if (head
== dev
->vq
->vring
.num
)
1575 /* Every block request should contain at least one output buffer
1576 * (detailing the location on disk and the type of request) and one
1577 * input buffer (to hold the result). */
1578 if (out_num
== 0 || in_num
== 0)
1579 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1580 head
, out_num
, in_num
);
1582 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1583 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1584 off
= out
->sector
* 512;
1586 /* The block device implements "barriers", where the Guest indicates
1587 * that it wants all previous writes to occur before this write. We
1588 * don't have a way of asking our kernel to do a barrier, so we just
1589 * synchronize all the data in the file. Pretty poor, no? */
1590 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1591 fdatasync(vblk
->fd
);
1593 /* In general the virtio block driver is allowed to try SCSI commands.
1594 * It'd be nice if we supported eject, for example, but we don't. */
1595 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1596 fprintf(stderr
, "Scsi commands unsupported\n");
1597 *in
= VIRTIO_BLK_S_UNSUPP
;
1599 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1602 /* Move to the right location in the block file. This can fail
1603 * if they try to write past end. */
1604 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1605 err(1, "Bad seek to sector %llu", out
->sector
);
1607 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1608 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1610 /* Grr... Now we know how long the descriptor they sent was, we
1611 * make sure they didn't try to write over the end of the block
1612 * file (possibly extending it). */
1613 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1614 /* Trim it back to the correct length */
1615 ftruncate64(vblk
->fd
, vblk
->len
);
1616 /* Die, bad Guest, die. */
1617 errx(1, "Write past end %llu+%u", off
, ret
);
1620 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1624 /* Move to the right location in the block file. This can fail
1625 * if they try to read past end. */
1626 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1627 err(1, "Bad seek to sector %llu", out
->sector
);
1629 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1630 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1632 wlen
= sizeof(*in
) + ret
;
1633 *in
= VIRTIO_BLK_S_OK
;
1636 *in
= VIRTIO_BLK_S_IOERR
;
1640 /* OK, so we noted that it was pretty poor to use an fdatasync as a
1641 * barrier. But Christoph Hellwig points out that we need a sync
1642 * *afterwards* as well: "Barriers specify no reordering to the front
1643 * or the back." And Jens Axboe confirmed it, so here we are: */
1644 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1645 fdatasync(vblk
->fd
);
1647 /* We can't trigger an IRQ, because we're not the Launcher. It does
1648 * that when we tell it we're done. */
1649 add_used(dev
->vq
, head
, wlen
);
1653 /* This is the thread which actually services the I/O. */
1654 static int io_thread(void *_dev
)
1656 struct device
*dev
= _dev
;
1657 struct vblk_info
*vblk
= dev
->priv
;
1660 /* Close other side of workpipe so we get 0 read when main dies. */
1661 close(vblk
->workpipe
[1]);
1662 /* Close the other side of the done_fd pipe. */
1665 /* When this read fails, it means Launcher died, so we follow. */
1666 while (read(vblk
->workpipe
[0], &c
, 1) == 1) {
1667 /* We acknowledge each request immediately to reduce latency,
1668 * rather than waiting until we've done them all. I haven't
1669 * measured to see if it makes any difference.
1671 * That would be an interesting test, wouldn't it? You could
1672 * also try having more than one I/O thread. */
1673 while (service_io(dev
))
1674 write(vblk
->done_fd
, &c
, 1);
1679 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1680 * when that thread tells us it's completed some I/O. */
1681 static bool handle_io_finish(int fd
, struct device
*dev
)
1685 /* If the I/O thread died, presumably it printed the error, so we
1687 if (read(dev
->fd
, &c
, 1) != 1)
1690 /* It did some work, so trigger the irq. */
1691 trigger_irq(fd
, dev
->vq
);
1695 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1696 static void handle_virtblk_output(int fd
, struct virtqueue
*vq
, bool timeout
)
1698 struct vblk_info
*vblk
= vq
->dev
->priv
;
1701 /* Wake up I/O thread and tell it to go to work! */
1702 if (write(vblk
->workpipe
[1], &c
, 1) != 1)
1703 /* Presumably it indicated why it died. */
1707 /*L:198 This actually sets up a virtual block device. */
1708 static void setup_block_file(const char *filename
)
1712 struct vblk_info
*vblk
;
1714 struct virtio_blk_config conf
;
1716 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1719 /* The device responds to return from I/O thread. */
1720 dev
= new_device("block", VIRTIO_ID_BLOCK
, p
[0], handle_io_finish
);
1722 /* The device has one virtqueue, where the Guest places requests. */
1723 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_virtblk_output
);
1725 /* Allocate the room for our own bookkeeping */
1726 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1728 /* First we open the file and store the length. */
1729 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1730 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1732 /* We support barriers. */
1733 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1735 /* Tell Guest how many sectors this device has. */
1736 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1738 /* Tell Guest not to put in too many descriptors at once: two are used
1739 * for the in and out elements. */
1740 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1741 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1743 set_config(dev
, sizeof(conf
), &conf
);
1745 /* The I/O thread writes to this end of the pipe when done. */
1746 vblk
->done_fd
= p
[1];
1748 /* This is the second pipe, which is how we tell the I/O thread about
1750 pipe(vblk
->workpipe
);
1752 /* Create stack for thread and run it. Since stack grows upwards, we
1753 * point the stack pointer to the end of this region. */
1754 stack
= malloc(32768);
1755 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1756 * becoming a zombie. */
1757 if (clone(io_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, dev
) == -1)
1758 err(1, "Creating clone");
1760 /* We don't need to keep the I/O thread's end of the pipes open. */
1761 close(vblk
->done_fd
);
1762 close(vblk
->workpipe
[0]);
1764 verbose("device %u: virtblock %llu sectors\n",
1765 devices
.device_num
, le64_to_cpu(conf
.capacity
));
1768 /* Our random number generator device reads from /dev/random into the Guest's
1769 * input buffers. The usual case is that the Guest doesn't want random numbers
1770 * and so has no buffers although /dev/random is still readable, whereas
1771 * console is the reverse.
1773 * The same logic applies, however. */
1774 static bool handle_rng_input(int fd
, struct device
*dev
)
1777 unsigned int head
, in_num
, out_num
, totlen
= 0;
1778 struct iovec iov
[dev
->vq
->vring
.num
];
1780 /* First we need a buffer from the Guests's virtqueue. */
1781 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1783 /* If they're not ready for input, stop listening to this file
1784 * descriptor. We'll start again once they add an input buffer. */
1785 if (head
== dev
->vq
->vring
.num
)
1789 errx(1, "Output buffers in rng?");
1791 /* This is why we convert to iovecs: the readv() call uses them, and so
1792 * it reads straight into the Guest's buffer. We loop to make sure we
1794 while (!iov_empty(iov
, in_num
)) {
1795 len
= readv(dev
->fd
, iov
, in_num
);
1797 err(1, "Read from /dev/random gave %i", len
);
1798 iov_consume(iov
, in_num
, len
);
1802 /* Tell the Guest about the new input. */
1803 add_used_and_trigger(fd
, dev
->vq
, head
, totlen
);
1805 /* Everything went OK! */
1809 /* And this creates a "hardware" random number device for the Guest. */
1810 static void setup_rng(void)
1815 fd
= open_or_die("/dev/random", O_RDONLY
);
1817 /* The device responds to return from I/O thread. */
1818 dev
= new_device("rng", VIRTIO_ID_RNG
, fd
, handle_rng_input
);
1820 /* The device has one virtqueue, where the Guest places inbufs. */
1821 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1823 verbose("device %u: rng\n", devices
.device_num
++);
1825 /* That's the end of device setup. */
1827 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1828 static void __attribute__((noreturn
)) restart_guest(void)
1832 /* Since we don't track all open fds, we simply close everything beyond
1834 for (i
= 3; i
< FD_SETSIZE
; i
++)
1837 /* The exec automatically gets rid of the I/O and Waker threads. */
1838 execv(main_args
[0], main_args
);
1839 err(1, "Could not exec %s", main_args
[0]);
1842 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1843 * its input and output, and finally, lays it to rest. */
1844 static void __attribute__((noreturn
)) run_guest(int lguest_fd
)
1847 unsigned long args
[] = { LHREQ_BREAK
, 0 };
1848 unsigned long notify_addr
;
1851 /* We read from the /dev/lguest device to run the Guest. */
1852 readval
= pread(lguest_fd
, ¬ify_addr
,
1853 sizeof(notify_addr
), cpu_id
);
1855 /* One unsigned long means the Guest did HCALL_NOTIFY */
1856 if (readval
== sizeof(notify_addr
)) {
1857 verbose("Notify on address %#lx\n", notify_addr
);
1858 handle_output(lguest_fd
, notify_addr
);
1860 /* ENOENT means the Guest died. Reading tells us why. */
1861 } else if (errno
== ENOENT
) {
1862 char reason
[1024] = { 0 };
1863 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1864 errx(1, "%s", reason
);
1865 /* ERESTART means that we need to reboot the guest */
1866 } else if (errno
== ERESTART
) {
1868 /* EAGAIN means a signal (timeout).
1869 * Anything else means a bug or incompatible change. */
1870 } else if (errno
!= EAGAIN
)
1871 err(1, "Running guest failed");
1873 /* Only service input on thread for CPU 0. */
1877 /* Service input, then unset the BREAK to release the Waker. */
1878 handle_input(lguest_fd
);
1879 if (pwrite(lguest_fd
, args
, sizeof(args
), cpu_id
) < 0)
1880 err(1, "Resetting break");
1884 * This is the end of the Launcher. The good news: we are over halfway
1885 * through! The bad news: the most fiendish part of the code still lies ahead
1888 * Are you ready? Take a deep breath and join me in the core of the Host, in
1892 static struct option opts
[] = {
1893 { "verbose", 0, NULL
, 'v' },
1894 { "tunnet", 1, NULL
, 't' },
1895 { "block", 1, NULL
, 'b' },
1896 { "rng", 0, NULL
, 'r' },
1897 { "initrd", 1, NULL
, 'i' },
1900 static void usage(void)
1902 errx(1, "Usage: lguest [--verbose] "
1903 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1904 "|--block=<filename>|--initrd=<filename>]...\n"
1905 "<mem-in-mb> vmlinux [args...]");
1908 /*L:105 The main routine is where the real work begins: */
1909 int main(int argc
, char *argv
[])
1911 /* Memory, top-level pagetable, code startpoint and size of the
1912 * (optional) initrd. */
1913 unsigned long mem
= 0, start
, initrd_size
= 0;
1914 /* Two temporaries and the /dev/lguest file descriptor. */
1915 int i
, c
, lguest_fd
;
1916 /* The boot information for the Guest. */
1917 struct boot_params
*boot
;
1918 /* If they specify an initrd file to load. */
1919 const char *initrd_name
= NULL
;
1921 /* Save the args: we "reboot" by execing ourselves again. */
1923 /* We don't "wait" for the children, so prevent them from becoming
1925 signal(SIGCHLD
, SIG_IGN
);
1927 /* First we initialize the device list. Since console and network
1928 * device receive input from a file descriptor, we keep an fdset
1929 * (infds) and the maximum fd number (max_infd) with the head of the
1930 * list. We also keep a pointer to the last device. Finally, we keep
1931 * the next interrupt number to use for devices (1: remember that 0 is
1932 * used by the timer). */
1933 FD_ZERO(&devices
.infds
);
1934 devices
.max_infd
= -1;
1935 devices
.lastdev
= NULL
;
1936 devices
.next_irq
= 1;
1939 /* We need to know how much memory so we can set up the device
1940 * descriptor and memory pages for the devices as we parse the command
1941 * line. So we quickly look through the arguments to find the amount
1943 for (i
= 1; i
< argc
; i
++) {
1944 if (argv
[i
][0] != '-') {
1945 mem
= atoi(argv
[i
]) * 1024 * 1024;
1946 /* We start by mapping anonymous pages over all of
1947 * guest-physical memory range. This fills it with 0,
1948 * and ensures that the Guest won't be killed when it
1949 * tries to access it. */
1950 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1953 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1954 devices
.descpage
= get_pages(1);
1959 /* The options are fairly straight-forward */
1960 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1966 setup_tun_net(optarg
);
1969 setup_block_file(optarg
);
1975 initrd_name
= optarg
;
1978 warnx("Unknown argument %s", argv
[optind
]);
1982 /* After the other arguments we expect memory and kernel image name,
1983 * followed by command line arguments for the kernel. */
1984 if (optind
+ 2 > argc
)
1987 verbose("Guest base is at %p\n", guest_base
);
1989 /* We always have a console device */
1992 /* We can timeout waiting for Guest network transmit. */
1995 /* Now we load the kernel */
1996 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1998 /* Boot information is stashed at physical address 0 */
1999 boot
= from_guest_phys(0);
2001 /* Map the initrd image if requested (at top of physical memory) */
2003 initrd_size
= load_initrd(initrd_name
, mem
);
2004 /* These are the location in the Linux boot header where the
2005 * start and size of the initrd are expected to be found. */
2006 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
2007 boot
->hdr
.ramdisk_size
= initrd_size
;
2008 /* The bootloader type 0xFF means "unknown"; that's OK. */
2009 boot
->hdr
.type_of_loader
= 0xFF;
2012 /* The Linux boot header contains an "E820" memory map: ours is a
2013 * simple, single region. */
2014 boot
->e820_entries
= 1;
2015 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
2016 /* The boot header contains a command line pointer: we put the command
2017 * line after the boot header. */
2018 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
2019 /* We use a simple helper to copy the arguments separated by spaces. */
2020 concat((char *)(boot
+ 1), argv
+optind
+2);
2022 /* Boot protocol version: 2.07 supports the fields for lguest. */
2023 boot
->hdr
.version
= 0x207;
2025 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2026 boot
->hdr
.hardware_subarch
= 1;
2028 /* Tell the entry path not to try to reload segment registers. */
2029 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
2031 /* We tell the kernel to initialize the Guest: this returns the open
2032 * /dev/lguest file descriptor. */
2033 lguest_fd
= tell_kernel(start
);
2035 /* We clone off a thread, which wakes the Launcher whenever one of the
2036 * input file descriptors needs attention. We call this the Waker, and
2037 * we'll cover it in a moment. */
2038 setup_waker(lguest_fd
);
2040 /* Finally, run the Guest. This doesn't return. */
2041 run_guest(lguest_fd
);
2046 * Mastery is done: you now know everything I do.
2048 * But surely you have seen code, features and bugs in your wanderings which
2049 * you now yearn to attack? That is the real game, and I look forward to you
2050 * patching and forking lguest into the Your-Name-Here-visor.
2052 * Farewell, and good coding!