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 the
3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest.
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>
37 /*L:110 We can ignore the 30 include files we need for this program, but I do
38 * want to draw attention to the use of kernel-style types.
40 * As Linus said, "C is a Spartan language, and so should your naming be." I
41 * like these abbreviations and the header we need uses them, so we define them
44 typedef unsigned long long u64
;
48 #include "linux/lguest_launcher.h"
49 #include "linux/pci_ids.h"
50 #include "linux/virtio_config.h"
51 #include "linux/virtio_net.h"
52 #include "linux/virtio_blk.h"
53 #include "linux/virtio_console.h"
54 #include "linux/virtio_ring.h"
55 #include "asm-x86/e820.h"
58 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
60 #define BRIDGE_PFX "bridge:"
62 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
64 /* We can have up to 256 pages for devices. */
65 #define DEVICE_PAGES 256
66 /* This fits nicely in a single 4096-byte page. */
67 #define VIRTQUEUE_NUM 127
69 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
70 * this, and although I wouldn't recommend it, it works quite nicely here. */
72 #define verbose(args...) \
73 do { if (verbose) printf(args); } while(0)
76 /* The pipe to send commands to the waker process */
78 /* The pointer to the start of guest memory. */
79 static void *guest_base
;
80 /* The maximum guest physical address allowed, and maximum possible. */
81 static unsigned long guest_limit
, guest_max
;
83 /* This is our list of devices. */
86 /* Summary information about the devices in our list: ready to pass to
87 * select() to ask which need servicing.*/
91 /* Counter to assign interrupt numbers. */
92 unsigned int next_irq
;
94 /* Counter to print out convenient device numbers. */
95 unsigned int device_num
;
97 /* The descriptor page for the devices. */
100 /* The tail of the last descriptor. */
101 unsigned int desc_used
;
103 /* A single linked list of devices. */
105 /* ... And an end pointer so we can easily append new devices */
106 struct device
**lastdev
;
109 /* The list of Guest devices, based on command line arguments. */
110 static struct device_list devices
;
112 /* The device structure describes a single device. */
115 /* The linked-list pointer. */
118 /* The this device's descriptor, as mapped into the Guest. */
119 struct lguest_device_desc
*desc
;
121 /* The name of this device, for --verbose. */
124 /* If handle_input is set, it wants to be called when this file
125 * descriptor is ready. */
127 bool (*handle_input
)(int fd
, struct device
*me
);
129 /* Any queues attached to this device */
130 struct virtqueue
*vq
;
132 /* Device-specific data. */
136 /* The virtqueue structure describes a queue attached to a device. */
139 struct virtqueue
*next
;
141 /* Which device owns me. */
144 /* The configuration for this queue. */
145 struct lguest_vqconfig config
;
147 /* The actual ring of buffers. */
150 /* Last available index we saw. */
153 /* The routine to call when the Guest pings us. */
154 void (*handle_output
)(int fd
, struct virtqueue
*me
);
157 /* Since guest is UP and we don't run at the same time, we don't need barriers.
158 * But I include them in the code in case others copy it. */
161 /* Convert an iovec element to the given type.
163 * This is a fairly ugly trick: we need to know the size of the type and
164 * alignment requirement to check the pointer is kosher. It's also nice to
165 * have the name of the type in case we report failure.
167 * Typing those three things all the time is cumbersome and error prone, so we
168 * have a macro which sets them all up and passes to the real function. */
169 #define convert(iov, type) \
170 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
172 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
175 if (iov
->iov_len
!= size
)
176 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
177 if ((unsigned long)iov
->iov_base
% align
!= 0)
178 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
179 return iov
->iov_base
;
182 /* The virtio configuration space is defined to be little-endian. x86 is
183 * little-endian too, but it's nice to be explicit so we have these helpers. */
184 #define cpu_to_le16(v16) (v16)
185 #define cpu_to_le32(v32) (v32)
186 #define cpu_to_le64(v64) (v64)
187 #define le16_to_cpu(v16) (v16)
188 #define le32_to_cpu(v32) (v32)
189 #define le64_to_cpu(v32) (v64)
191 /*L:100 The Launcher code itself takes us out into userspace, that scary place
192 * where pointers run wild and free! Unfortunately, like most userspace
193 * programs, it's quite boring (which is why everyone likes to hack on the
194 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
195 * will get you through this section. Or, maybe not.
197 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
198 * memory and stores it in "guest_base". In other words, Guest physical ==
199 * Launcher virtual with an offset.
201 * This can be tough to get your head around, but usually it just means that we
202 * use these trivial conversion functions when the Guest gives us it's
203 * "physical" addresses: */
204 static void *from_guest_phys(unsigned long addr
)
206 return guest_base
+ addr
;
209 static unsigned long to_guest_phys(const void *addr
)
211 return (addr
- guest_base
);
215 * Loading the Kernel.
217 * We start with couple of simple helper routines. open_or_die() avoids
218 * error-checking code cluttering the callers: */
219 static int open_or_die(const char *name
, int flags
)
221 int fd
= open(name
, flags
);
223 err(1, "Failed to open %s", name
);
227 /* map_zeroed_pages() takes a number of pages. */
228 static void *map_zeroed_pages(unsigned int num
)
230 int fd
= open_or_die("/dev/zero", O_RDONLY
);
233 /* We use a private mapping (ie. if we write to the page, it will be
235 addr
= mmap(NULL
, getpagesize() * num
,
236 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
237 if (addr
== MAP_FAILED
)
238 err(1, "Mmaping %u pages of /dev/zero", num
);
243 /* Get some more pages for a device. */
244 static void *get_pages(unsigned int num
)
246 void *addr
= from_guest_phys(guest_limit
);
248 guest_limit
+= num
* getpagesize();
249 if (guest_limit
> guest_max
)
250 errx(1, "Not enough memory for devices");
254 /* This routine is used to load the kernel or initrd. It tries mmap, but if
255 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
256 * it falls back to reading the memory in. */
257 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
261 /* We map writable even though for some segments are marked read-only.
262 * The kernel really wants to be writable: it patches its own
265 * MAP_PRIVATE means that the page won't be copied until a write is
266 * done to it. This allows us to share untouched memory between
268 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
269 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
272 /* pread does a seek and a read in one shot: saves a few lines. */
273 r
= pread(fd
, addr
, len
, offset
);
275 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
278 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
279 * the Guest memory. ELF = Embedded Linking Format, which is the format used
280 * by all modern binaries on Linux including the kernel.
282 * The ELF headers give *two* addresses: a physical address, and a virtual
283 * address. We use the physical address; the Guest will map itself to the
286 * We return the starting address. */
287 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
289 Elf32_Phdr phdr
[ehdr
->e_phnum
];
292 /* Sanity checks on the main ELF header: an x86 executable with a
293 * reasonable number of correctly-sized program headers. */
294 if (ehdr
->e_type
!= ET_EXEC
295 || ehdr
->e_machine
!= EM_386
296 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
297 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
298 errx(1, "Malformed elf header");
300 /* An ELF executable contains an ELF header and a number of "program"
301 * headers which indicate which parts ("segments") of the program to
304 /* We read in all the program headers at once: */
305 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
306 err(1, "Seeking to program headers");
307 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
308 err(1, "Reading program headers");
310 /* Try all the headers: there are usually only three. A read-only one,
311 * a read-write one, and a "note" section which isn't loadable. */
312 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
313 /* If this isn't a loadable segment, we ignore it */
314 if (phdr
[i
].p_type
!= PT_LOAD
)
317 verbose("Section %i: size %i addr %p\n",
318 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
320 /* We map this section of the file at its physical address. */
321 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
322 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
325 /* The entry point is given in the ELF header. */
326 return ehdr
->e_entry
;
329 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
330 * supposed to jump into it and it will unpack itself. We used to have to
331 * perform some hairy magic because the unpacking code scared me.
333 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
334 * a small patch to jump over the tricky bits in the Guest, so now we just read
335 * the funky header so we know where in the file to load, and away we go! */
336 static unsigned long load_bzimage(int fd
)
340 /* Modern bzImages get loaded at 1M. */
341 void *p
= from_guest_phys(0x100000);
343 /* Go back to the start of the file and read the header. It should be
344 * a Linux boot header (see Documentation/i386/boot.txt) */
345 lseek(fd
, 0, SEEK_SET
);
346 read(fd
, hdr
, sizeof(hdr
));
348 /* At offset 0x202, we expect the magic "HdrS" */
349 if (memcmp(hdr
+ 0x202, "HdrS", 4) != 0)
350 errx(1, "This doesn't look like a bzImage to me");
352 /* The byte at 0x1F1 tells us how many extra sectors of
353 * header: skip over them all. */
354 lseek(fd
, (unsigned long)(hdr
[0x1F1]+1) * 512, SEEK_SET
);
356 /* Now read everything into memory. in nice big chunks. */
357 while ((r
= read(fd
, p
, 65536)) > 0)
360 /* Finally, 0x214 tells us where to start the kernel. */
361 return *(unsigned long *)&hdr
[0x214];
364 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
365 * come wrapped up in the self-decompressing "bzImage" format. With some funky
366 * coding, we can load those, too. */
367 static unsigned long load_kernel(int fd
)
371 /* Read in the first few bytes. */
372 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
373 err(1, "Reading kernel");
375 /* If it's an ELF file, it starts with "\177ELF" */
376 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
377 return map_elf(fd
, &hdr
);
379 /* Otherwise we assume it's a bzImage, and try to unpack it */
380 return load_bzimage(fd
);
383 /* This is a trivial little helper to align pages. Andi Kleen hated it because
384 * it calls getpagesize() twice: "it's dumb code."
386 * Kernel guys get really het up about optimization, even when it's not
387 * necessary. I leave this code as a reaction against that. */
388 static inline unsigned long page_align(unsigned long addr
)
390 /* Add upwards and truncate downwards. */
391 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
394 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
395 * the kernel which the kernel can use to boot from without needing any
396 * drivers. Most distributions now use this as standard: the initrd contains
397 * the code to load the appropriate driver modules for the current machine.
399 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
400 * kernels. He sent me this (and tells me when I break it). */
401 static unsigned long load_initrd(const char *name
, unsigned long mem
)
407 ifd
= open_or_die(name
, O_RDONLY
);
408 /* fstat() is needed to get the file size. */
409 if (fstat(ifd
, &st
) < 0)
410 err(1, "fstat() on initrd '%s'", name
);
412 /* We map the initrd at the top of memory, but mmap wants it to be
413 * page-aligned, so we round the size up for that. */
414 len
= page_align(st
.st_size
);
415 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
416 /* Once a file is mapped, you can close the file descriptor. It's a
417 * little odd, but quite useful. */
419 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
421 /* We return the initrd size. */
425 /* Once we know how much memory we have, we can construct simple linear page
426 * tables which set virtual == physical which will get the Guest far enough
427 * into the boot to create its own.
429 * We lay them out of the way, just below the initrd (which is why we need to
431 static unsigned long setup_pagetables(unsigned long mem
,
432 unsigned long initrd_size
)
434 unsigned long *pgdir
, *linear
;
435 unsigned int mapped_pages
, i
, linear_pages
;
436 unsigned int ptes_per_page
= getpagesize()/sizeof(void *);
438 mapped_pages
= mem
/getpagesize();
440 /* Each PTE page can map ptes_per_page pages: how many do we need? */
441 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
443 /* We put the toplevel page directory page at the top of memory. */
444 pgdir
= from_guest_phys(mem
) - initrd_size
- getpagesize();
446 /* Now we use the next linear_pages pages as pte pages */
447 linear
= (void *)pgdir
- linear_pages
*getpagesize();
449 /* Linear mapping is easy: put every page's address into the mapping in
450 * order. PAGE_PRESENT contains the flags Present, Writable and
452 for (i
= 0; i
< mapped_pages
; i
++)
453 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
455 /* The top level points to the linear page table pages above. */
456 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
457 pgdir
[i
/ptes_per_page
]
458 = ((to_guest_phys(linear
) + i
*sizeof(void *))
462 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
463 mapped_pages
, linear_pages
, to_guest_phys(linear
));
465 /* We return the top level (guest-physical) address: the kernel needs
466 * to know where it is. */
467 return to_guest_phys(pgdir
);
470 /* Simple routine to roll all the commandline arguments together with spaces
472 static void concat(char *dst
, char *args
[])
474 unsigned int i
, len
= 0;
476 for (i
= 0; args
[i
]; i
++) {
477 strcpy(dst
+len
, args
[i
]);
478 strcat(dst
+len
, " ");
479 len
+= strlen(args
[i
]) + 1;
481 /* In case it's empty. */
485 /* This is where we actually tell the kernel to initialize the Guest. We saw
486 * the arguments it expects when we looked at initialize() in lguest_user.c:
487 * the base of guest "physical" memory, the top physical page to allow, the
488 * top level pagetable and the entry point for the Guest. */
489 static int tell_kernel(unsigned long pgdir
, unsigned long start
)
491 unsigned long args
[] = { LHREQ_INITIALIZE
,
492 (unsigned long)guest_base
,
493 guest_limit
/ getpagesize(), pgdir
, start
};
496 verbose("Guest: %p - %p (%#lx)\n",
497 guest_base
, guest_base
+ guest_limit
, guest_limit
);
498 fd
= open_or_die("/dev/lguest", O_RDWR
);
499 if (write(fd
, args
, sizeof(args
)) < 0)
500 err(1, "Writing to /dev/lguest");
502 /* We return the /dev/lguest file descriptor to control this Guest */
507 static void add_device_fd(int fd
)
509 FD_SET(fd
, &devices
.infds
);
510 if (fd
> devices
.max_infd
)
511 devices
.max_infd
= fd
;
517 * With a console and network devices, we can have lots of input which we need
518 * to process. We could try to tell the kernel what file descriptors to watch,
519 * but handing a file descriptor mask through to the kernel is fairly icky.
521 * Instead, we fork off a process which watches the file descriptors and writes
522 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
523 * loop to stop running the Guest. This causes it to return from the
524 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
525 * the LHREQ_BREAK and wake us up again.
527 * This, of course, is merely a different *kind* of icky.
529 static void wake_parent(int pipefd
, int lguest_fd
)
531 /* Add the pipe from the Launcher to the fdset in the device_list, so
532 * we watch it, too. */
533 add_device_fd(pipefd
);
536 fd_set rfds
= devices
.infds
;
537 unsigned long args
[] = { LHREQ_BREAK
, 1 };
539 /* Wait until input is ready from one of the devices. */
540 select(devices
.max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
541 /* Is it a message from the Launcher? */
542 if (FD_ISSET(pipefd
, &rfds
)) {
544 /* If read() returns 0, it means the Launcher has
545 * exited. We silently follow. */
546 if (read(pipefd
, &fd
, sizeof(fd
)) == 0)
548 /* Otherwise it's telling us to change what file
549 * descriptors we're to listen to. */
551 FD_SET(fd
, &devices
.infds
);
553 FD_CLR(-fd
- 1, &devices
.infds
);
554 } else /* Send LHREQ_BREAK command. */
555 write(lguest_fd
, args
, sizeof(args
));
559 /* This routine just sets up a pipe to the Waker process. */
560 static int setup_waker(int lguest_fd
)
562 int pipefd
[2], child
;
564 /* We create a pipe to talk to the waker, and also so it knows when the
565 * Launcher dies (and closes pipe). */
572 /* Close the "writing" end of our copy of the pipe */
574 wake_parent(pipefd
[0], lguest_fd
);
576 /* Close the reading end of our copy of the pipe. */
579 /* Here is the fd used to talk to the waker. */
586 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
587 * We need to make sure it's not trying to reach into the Launcher itself, so
588 * we have a convenient routine which check it and exits with an error message
589 * if something funny is going on:
591 static void *_check_pointer(unsigned long addr
, unsigned int size
,
594 /* We have to separately check addr and addr+size, because size could
595 * be huge and addr + size might wrap around. */
596 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
597 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
598 /* We return a pointer for the caller's convenience, now we know it's
600 return from_guest_phys(addr
);
602 /* A macro which transparently hands the line number to the real function. */
603 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
605 /* This function returns the next descriptor in the chain, or vq->vring.num. */
606 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
610 /* If this descriptor says it doesn't chain, we're done. */
611 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
612 return vq
->vring
.num
;
614 /* Check they're not leading us off end of descriptors. */
615 next
= vq
->vring
.desc
[i
].next
;
616 /* Make sure compiler knows to grab that: we don't want it changing! */
619 if (next
>= vq
->vring
.num
)
620 errx(1, "Desc next is %u", next
);
625 /* This looks in the virtqueue and for the first available buffer, and converts
626 * it to an iovec for convenient access. Since descriptors consist of some
627 * number of output then some number of input descriptors, it's actually two
628 * iovecs, but we pack them into one and note how many of each there were.
630 * This function returns the descriptor number found, or vq->vring.num (which
631 * is never a valid descriptor number) if none was found. */
632 static unsigned get_vq_desc(struct virtqueue
*vq
,
634 unsigned int *out_num
, unsigned int *in_num
)
636 unsigned int i
, head
;
638 /* Check it isn't doing very strange things with descriptor numbers. */
639 if ((u16
)(vq
->vring
.avail
->idx
- vq
->last_avail_idx
) > vq
->vring
.num
)
640 errx(1, "Guest moved used index from %u to %u",
641 vq
->last_avail_idx
, vq
->vring
.avail
->idx
);
643 /* If there's nothing new since last we looked, return invalid. */
644 if (vq
->vring
.avail
->idx
== vq
->last_avail_idx
)
645 return vq
->vring
.num
;
647 /* Grab the next descriptor number they're advertising, and increment
648 * the index we've seen. */
649 head
= vq
->vring
.avail
->ring
[vq
->last_avail_idx
++ % vq
->vring
.num
];
651 /* If their number is silly, that's a fatal mistake. */
652 if (head
>= vq
->vring
.num
)
653 errx(1, "Guest says index %u is available", head
);
655 /* When we start there are none of either input nor output. */
656 *out_num
= *in_num
= 0;
660 /* Grab the first descriptor, and check it's OK. */
661 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
662 iov
[*out_num
+ *in_num
].iov_base
663 = check_pointer(vq
->vring
.desc
[i
].addr
,
664 vq
->vring
.desc
[i
].len
);
665 /* If this is an input descriptor, increment that count. */
666 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
669 /* If it's an output descriptor, they're all supposed
670 * to come before any input descriptors. */
672 errx(1, "Descriptor has out after in");
676 /* If we've got too many, that implies a descriptor loop. */
677 if (*out_num
+ *in_num
> vq
->vring
.num
)
678 errx(1, "Looped descriptor");
679 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
684 /* Once we've used one of their buffers, we tell them about it. We'll then
685 * want to send them an interrupt, using trigger_irq(). */
686 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
688 struct vring_used_elem
*used
;
690 /* Get a pointer to the next entry in the used ring. */
691 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
694 /* Make sure buffer is written before we update index. */
696 vq
->vring
.used
->idx
++;
699 /* This actually sends the interrupt for this virtqueue */
700 static void trigger_irq(int fd
, struct virtqueue
*vq
)
702 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
704 if (vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
707 /* Send the Guest an interrupt tell them we used something up. */
708 if (write(fd
, buf
, sizeof(buf
)) != 0)
709 err(1, "Triggering irq %i", vq
->config
.irq
);
712 /* And here's the combo meal deal. Supersize me! */
713 static void add_used_and_trigger(int fd
, struct virtqueue
*vq
,
714 unsigned int head
, int len
)
716 add_used(vq
, head
, len
);
720 /* Here is the input terminal setting we save, and the routine to restore them
721 * on exit so the user can see what they type next. */
722 static struct termios orig_term
;
723 static void restore_term(void)
725 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
728 /* We associate some data with the console for our exit hack. */
731 /* How many times have they hit ^C? */
733 /* When did they start? */
734 struct timeval start
;
737 /* This is the routine which handles console input (ie. stdin). */
738 static bool handle_console_input(int fd
, struct device
*dev
)
741 unsigned int head
, in_num
, out_num
;
742 struct iovec iov
[dev
->vq
->vring
.num
];
743 struct console_abort
*abort
= dev
->priv
;
745 /* First we need a console buffer from the Guests's input virtqueue. */
746 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
748 /* If they're not ready for input, stop listening to this file
749 * descriptor. We'll start again once they add an input buffer. */
750 if (head
== dev
->vq
->vring
.num
)
754 errx(1, "Output buffers in console in queue?");
756 /* This is why we convert to iovecs: the readv() call uses them, and so
757 * it reads straight into the Guest's buffer. */
758 len
= readv(dev
->fd
, iov
, in_num
);
760 /* This implies that the console is closed, is /dev/null, or
761 * something went terribly wrong. */
762 warnx("Failed to get console input, ignoring console.");
763 /* Put the input terminal back. */
765 /* Remove callback from input vq, so it doesn't restart us. */
766 dev
->vq
->handle_output
= NULL
;
767 /* Stop listening to this fd: don't call us again. */
771 /* Tell the Guest about the new input. */
772 add_used_and_trigger(fd
, dev
->vq
, head
, len
);
774 /* Three ^C within one second? Exit.
776 * This is such a hack, but works surprisingly well. Each ^C has to be
777 * in a buffer by itself, so they can't be too fast. But we check that
778 * we get three within about a second, so they can't be too slow. */
779 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
781 gettimeofday(&abort
->start
, NULL
);
782 else if (abort
->count
== 3) {
784 gettimeofday(&now
, NULL
);
785 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
786 unsigned long args
[] = { LHREQ_BREAK
, 0 };
787 /* Close the fd so Waker will know it has to
790 /* Just in case waker is blocked in BREAK, send
792 write(fd
, args
, sizeof(args
));
798 /* Any other key resets the abort counter. */
801 /* Everything went OK! */
805 /* Handling output for console is simple: we just get all the output buffers
806 * and write them to stdout. */
807 static void handle_console_output(int fd
, struct virtqueue
*vq
)
809 unsigned int head
, out
, in
;
811 struct iovec iov
[vq
->vring
.num
];
813 /* Keep getting output buffers from the Guest until we run out. */
814 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
816 errx(1, "Input buffers in output queue?");
817 len
= writev(STDOUT_FILENO
, iov
, out
);
818 add_used_and_trigger(fd
, vq
, head
, len
);
822 /* Handling output for network is also simple: we get all the output buffers
823 * and write them (ignoring the first element) to this device's file descriptor
825 static void handle_net_output(int fd
, struct virtqueue
*vq
)
827 unsigned int head
, out
, in
;
829 struct iovec iov
[vq
->vring
.num
];
831 /* Keep getting output buffers from the Guest until we run out. */
832 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
834 errx(1, "Input buffers in output queue?");
835 /* Check header, but otherwise ignore it (we said we supported
837 (void)convert(&iov
[0], struct virtio_net_hdr
);
838 len
= writev(vq
->dev
->fd
, iov
+1, out
-1);
839 add_used_and_trigger(fd
, vq
, head
, len
);
843 /* This is where we handle a packet coming in from the tun device to our
845 static bool handle_tun_input(int fd
, struct device
*dev
)
847 unsigned int head
, in_num
, out_num
;
849 struct iovec iov
[dev
->vq
->vring
.num
];
850 struct virtio_net_hdr
*hdr
;
852 /* First we need a network buffer from the Guests's recv virtqueue. */
853 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
854 if (head
== dev
->vq
->vring
.num
) {
855 /* Now, it's expected that if we try to send a packet too
856 * early, the Guest won't be ready yet. Wait until the device
857 * status says it's ready. */
858 /* FIXME: Actually want DRIVER_ACTIVE here. */
859 if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)
860 warn("network: no dma buffer!");
861 /* We'll turn this back on if input buffers are registered. */
864 errx(1, "Output buffers in network recv queue?");
866 /* First element is the header: we set it to 0 (no features). */
867 hdr
= convert(&iov
[0], struct virtio_net_hdr
);
869 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
871 /* Read the packet from the device directly into the Guest's buffer. */
872 len
= readv(dev
->fd
, iov
+1, in_num
-1);
874 err(1, "reading network");
876 /* Tell the Guest about the new packet. */
877 add_used_and_trigger(fd
, dev
->vq
, head
, sizeof(*hdr
) + len
);
879 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
880 ((u8
*)iov
[1].iov_base
)[0], ((u8
*)iov
[1].iov_base
)[1],
881 head
!= dev
->vq
->vring
.num
? "sent" : "discarded");
887 /* This callback ensures we try again, in case we stopped console or net
888 * delivery because Guest didn't have any buffers. */
889 static void enable_fd(int fd
, struct virtqueue
*vq
)
891 add_device_fd(vq
->dev
->fd
);
892 /* Tell waker to listen to it again */
893 write(waker_fd
, &vq
->dev
->fd
, sizeof(vq
->dev
->fd
));
896 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
897 static void handle_output(int fd
, unsigned long addr
)
900 struct virtqueue
*vq
;
902 /* Check each virtqueue. */
903 for (i
= devices
.dev
; i
; i
= i
->next
) {
904 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
905 if (vq
->config
.pfn
== addr
/getpagesize()
906 && vq
->handle_output
) {
907 verbose("Output to %s\n", vq
->dev
->name
);
908 vq
->handle_output(fd
, vq
);
914 /* Early console write is done using notify on a nul-terminated string
915 * in Guest memory. */
916 if (addr
>= guest_limit
)
917 errx(1, "Bad NOTIFY %#lx", addr
);
919 write(STDOUT_FILENO
, from_guest_phys(addr
),
920 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
923 /* This is called when the waker wakes us up: check for incoming file
925 static void handle_input(int fd
)
927 /* select() wants a zeroed timeval to mean "don't wait". */
928 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
932 fd_set fds
= devices
.infds
;
934 /* If nothing is ready, we're done. */
935 if (select(devices
.max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
938 /* Otherwise, call the device(s) which have readable
939 * file descriptors and a method of handling them. */
940 for (i
= devices
.dev
; i
; i
= i
->next
) {
941 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
943 if (i
->handle_input(fd
, i
))
946 /* If handle_input() returns false, it means we
947 * should no longer service it. Networking and
948 * console do this when there's no input
949 * buffers to deliver into. Console also uses
950 * it when it discovers that stdin is
952 FD_CLR(i
->fd
, &devices
.infds
);
953 /* Tell waker to ignore it too, by sending a
954 * negative fd number (-1, since 0 is a valid
957 write(waker_fd
, &dev_fd
, sizeof(dev_fd
));
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 them.
970 * This routine allocates a new "struct lguest_device_desc" from descriptor
971 * table just above the Guest's normal memory. It returns a pointer to that
973 static struct lguest_device_desc
*new_dev_desc(u16 type
)
975 struct lguest_device_desc
*d
;
977 /* We only have one page for all the descriptors. */
978 if (devices
.desc_used
+ sizeof(*d
) > getpagesize())
979 errx(1, "Too many devices");
981 /* We don't need to set config_len or status: page is 0 already. */
982 d
= (void *)devices
.descpage
+ devices
.desc_used
;
984 devices
.desc_used
+= sizeof(*d
);
989 /* Each device descriptor is followed by some configuration information.
990 * The first byte is a "status" byte for the Guest to report what's happening.
991 * After that are fields: u8 type, u8 len, [... len bytes...].
993 * This routine adds a new field to an existing device's descriptor. It only
994 * works for the last device, but that's OK because that's how we use it. */
995 static void add_desc_field(struct device
*dev
, u8 type
, u8 len
, const void *c
)
997 /* This is the last descriptor, right? */
998 assert(devices
.descpage
+ devices
.desc_used
999 == (u8
*)(dev
->desc
+ 1) + dev
->desc
->config_len
);
1001 /* We only have one page of device descriptions. */
1002 if (devices
.desc_used
+ 2 + len
> getpagesize())
1003 errx(1, "Too many devices");
1005 /* Copy in the new config header: type then length. */
1006 devices
.descpage
[devices
.desc_used
++] = type
;
1007 devices
.descpage
[devices
.desc_used
++] = len
;
1008 memcpy(devices
.descpage
+ devices
.desc_used
, c
, len
);
1009 devices
.desc_used
+= len
;
1011 /* Update the device descriptor length: two byte head then data. */
1012 dev
->desc
->config_len
+= 2 + len
;
1015 /* This routine adds a virtqueue to a device. We specify how many descriptors
1016 * the virtqueue is to have. */
1017 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1018 void (*handle_output
)(int fd
, struct virtqueue
*me
))
1021 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1024 /* First we need some pages for this virtqueue. */
1025 pages
= (vring_size(num_descs
) + getpagesize() - 1) / getpagesize();
1026 p
= get_pages(pages
);
1028 /* Initialize the configuration. */
1029 vq
->config
.num
= num_descs
;
1030 vq
->config
.irq
= devices
.next_irq
++;
1031 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1033 /* Initialize the vring. */
1034 vring_init(&vq
->vring
, num_descs
, p
);
1036 /* Add the configuration information to this device's descriptor. */
1037 add_desc_field(dev
, VIRTIO_CONFIG_F_VIRTQUEUE
,
1038 sizeof(vq
->config
), &vq
->config
);
1040 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1042 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1045 /* Link virtqueue back to device. */
1048 /* Set up handler. */
1049 vq
->handle_output
= handle_output
;
1051 vq
->vring
.used
->flags
= VRING_USED_F_NO_NOTIFY
;
1054 /* This routine does all the creation and setup of a new device, including
1055 * caling new_dev_desc() to allocate the descriptor and device memory. */
1056 static struct device
*new_device(const char *name
, u16 type
, int fd
,
1057 bool (*handle_input
)(int, struct device
*))
1059 struct device
*dev
= malloc(sizeof(*dev
));
1061 /* Append to device list. Prepending to a single-linked list is
1062 * easier, but the user expects the devices to be arranged on the bus
1063 * in command-line order. The first network device on the command line
1064 * is eth0, the first block device /dev/lgba, etc. */
1065 *devices
.lastdev
= dev
;
1067 devices
.lastdev
= &dev
->next
;
1069 /* Now we populate the fields one at a time. */
1071 /* If we have an input handler for this file descriptor, then we add it
1072 * to the device_list's fdset and maxfd. */
1074 add_device_fd(dev
->fd
);
1075 dev
->desc
= new_dev_desc(type
);
1076 dev
->handle_input
= handle_input
;
1081 /* Our first setup routine is the console. It's a fairly simple device, but
1082 * UNIX tty handling makes it uglier than it could be. */
1083 static void setup_console(void)
1087 /* If we can save the initial standard input settings... */
1088 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1089 struct termios term
= orig_term
;
1090 /* Then we turn off echo, line buffering and ^C etc. We want a
1091 * raw input stream to the Guest. */
1092 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1093 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1094 /* If we exit gracefully, the original settings will be
1095 * restored so the user can see what they're typing. */
1096 atexit(restore_term
);
1099 dev
= new_device("console", VIRTIO_ID_CONSOLE
,
1100 STDIN_FILENO
, handle_console_input
);
1101 /* We store the console state in dev->priv, and initialize it. */
1102 dev
->priv
= malloc(sizeof(struct console_abort
));
1103 ((struct console_abort
*)dev
->priv
)->count
= 0;
1105 /* The console needs two virtqueues: the input then the output. When
1106 * they put something the input queue, we make sure we're listening to
1107 * stdin. When they put something in the output queue, we write it to
1109 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1110 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_console_output
);
1112 verbose("device %u: console\n", devices
.device_num
++);
1116 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1117 * --sharenet=<name> option which opens or creates a named pipe. This can be
1118 * used to send packets to another guest in a 1:1 manner.
1120 * More sopisticated is to use one of the tools developed for project like UML
1123 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1124 * completely generic ("here's my vring, attach to your vring") and would work
1125 * for any traffic. Of course, namespace and permissions issues need to be
1126 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1127 * multiple inter-guest channels behind one interface, although it would
1128 * require some manner of hotplugging new virtio channels.
1130 * Finally, we could implement a virtio network switch in the kernel. :*/
1132 static u32
str2ip(const char *ipaddr
)
1134 unsigned int byte
[4];
1136 sscanf(ipaddr
, "%u.%u.%u.%u", &byte
[0], &byte
[1], &byte
[2], &byte
[3]);
1137 return (byte
[0] << 24) | (byte
[1] << 16) | (byte
[2] << 8) | byte
[3];
1140 /* This code is "adapted" from libbridge: it attaches the Host end of the
1141 * network device to the bridge device specified by the command line.
1143 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1144 * dislike bridging), and I just try not to break it. */
1145 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1151 errx(1, "must specify bridge name");
1153 ifidx
= if_nametoindex(if_name
);
1155 errx(1, "interface %s does not exist!", if_name
);
1157 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1158 ifr
.ifr_ifindex
= ifidx
;
1159 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1160 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1163 /* This sets up the Host end of the network device with an IP address, brings
1164 * it up so packets will flow, the copies the MAC address into the hwaddr
1166 static void configure_device(int fd
, const char *devname
, u32 ipaddr
,
1167 unsigned char hwaddr
[6])
1170 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1172 /* Don't read these incantations. Just cut & paste them like I did! */
1173 memset(&ifr
, 0, sizeof(ifr
));
1174 strcpy(ifr
.ifr_name
, devname
);
1175 sin
->sin_family
= AF_INET
;
1176 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1177 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1178 err(1, "Setting %s interface address", devname
);
1179 ifr
.ifr_flags
= IFF_UP
;
1180 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1181 err(1, "Bringing interface %s up", devname
);
1183 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1184 * above). IF means Interface, and HWADDR is hardware address.
1186 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1187 err(1, "getting hw address for %s", devname
);
1188 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1191 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1192 * routing, but the principle is the same: it uses the "tun" device to inject
1193 * packets into the Host as if they came in from a normal network card. We
1194 * just shunt packets between the Guest and the tun device. */
1195 static void setup_tun_net(const char *arg
)
1201 const char *br_name
= NULL
;
1204 /* We open the /dev/net/tun device and tell it we want a tap device. A
1205 * tap device is like a tun device, only somehow different. To tell
1206 * the truth, I completely blundered my way through this code, but it
1208 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1209 memset(&ifr
, 0, sizeof(ifr
));
1210 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1211 strcpy(ifr
.ifr_name
, "tap%d");
1212 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1213 err(1, "configuring /dev/net/tun");
1214 /* We don't need checksums calculated for packets coming in this
1215 * device: trust us! */
1216 ioctl(netfd
, TUNSETNOCSUM
, 1);
1218 /* First we create a new network device. */
1219 dev
= new_device("net", VIRTIO_ID_NET
, netfd
, handle_tun_input
);
1221 /* Network devices need a receive and a send queue, just like
1223 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1224 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_net_output
);
1226 /* We need a socket to perform the magic network ioctls to bring up the
1227 * tap interface, connect to the bridge etc. Any socket will do! */
1228 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1230 err(1, "opening IP socket");
1232 /* If the command line was --tunnet=bridge:<name> do bridging. */
1233 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1235 br_name
= arg
+ strlen(BRIDGE_PFX
);
1236 add_to_bridge(ipfd
, ifr
.ifr_name
, br_name
);
1237 } else /* It is an IP address to set up the device with */
1240 /* Set up the tun device, and get the mac address for the interface. */
1241 configure_device(ipfd
, ifr
.ifr_name
, ip
, hwaddr
);
1243 /* Tell Guest what MAC address to use. */
1244 add_desc_field(dev
, VIRTIO_CONFIG_NET_MAC_F
, sizeof(hwaddr
), hwaddr
);
1246 /* We don't seed the socket any more; setup is done. */
1249 verbose("device %u: tun net %u.%u.%u.%u\n",
1250 devices
.device_num
++,
1251 (u8
)(ip
>>24),(u8
)(ip
>>16),(u8
)(ip
>>8),(u8
)ip
);
1253 verbose("attached to bridge: %s\n", br_name
);
1260 * Serving a block device is really easy: the Guest asks for a block number and
1261 * we read or write that position in the file.
1263 * Unfortunately, this is amazingly slow: the Guest waits until the read is
1264 * finished before running anything else, even if it could be doing useful
1265 * work. We could use async I/O, except it's reputed to suck so hard that
1266 * characters actually go missing from your code when you try to use it.
1268 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1270 /* This hangs off device->priv, with the data. */
1273 /* The size of the file. */
1276 /* The file descriptor for the file. */
1279 /* IO thread listens on this file descriptor [0]. */
1282 /* IO thread writes to this file descriptor to mark it done, then
1283 * Launcher triggers interrupt to Guest. */
1287 /* This is the core of the I/O thread. It returns true if it did something. */
1288 static bool service_io(struct device
*dev
)
1290 struct vblk_info
*vblk
= dev
->priv
;
1291 unsigned int head
, out_num
, in_num
, wlen
;
1293 struct virtio_blk_inhdr
*in
;
1294 struct virtio_blk_outhdr
*out
;
1295 struct iovec iov
[dev
->vq
->vring
.num
];
1298 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1299 if (head
== dev
->vq
->vring
.num
)
1302 if (out_num
== 0 || in_num
== 0)
1303 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1304 head
, out_num
, in_num
);
1306 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1307 in
= convert(&iov
[out_num
+in_num
-1], struct virtio_blk_inhdr
);
1308 off
= out
->sector
* 512;
1310 /* This is how we implement barriers. Pretty poor, no? */
1311 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1312 fdatasync(vblk
->fd
);
1314 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1315 fprintf(stderr
, "Scsi commands unsupported\n");
1316 in
->status
= VIRTIO_BLK_S_UNSUPP
;
1318 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1321 /* Move to the right location in the block file. This can fail
1322 * if they try to write past end. */
1323 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1324 err(1, "Bad seek to sector %llu", out
->sector
);
1326 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1327 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1329 /* Grr... Now we know how long the descriptor they sent was, we
1330 * make sure they didn't try to write over the end of the block
1331 * file (possibly extending it). */
1332 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1333 /* Trim it back to the correct length */
1334 ftruncate64(vblk
->fd
, vblk
->len
);
1335 /* Die, bad Guest, die. */
1336 errx(1, "Write past end %llu+%u", off
, ret
);
1339 in
->status
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1343 /* Move to the right location in the block file. This can fail
1344 * if they try to read past end. */
1345 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1346 err(1, "Bad seek to sector %llu", out
->sector
);
1348 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1349 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1351 wlen
= sizeof(in
) + ret
;
1352 in
->status
= VIRTIO_BLK_S_OK
;
1355 in
->status
= VIRTIO_BLK_S_IOERR
;
1359 /* We can't trigger an IRQ, because we're not the Launcher. It does
1360 * that when we tell it we're done. */
1361 add_used(dev
->vq
, head
, wlen
);
1365 /* This is the thread which actually services the I/O. */
1366 static int io_thread(void *_dev
)
1368 struct device
*dev
= _dev
;
1369 struct vblk_info
*vblk
= dev
->priv
;
1372 /* Close other side of workpipe so we get 0 read when main dies. */
1373 close(vblk
->workpipe
[1]);
1374 /* Close the other side of the done_fd pipe. */
1377 /* When this read fails, it means Launcher died, so we follow. */
1378 while (read(vblk
->workpipe
[0], &c
, 1) == 1) {
1379 /* We acknowledge each request immediately, to reduce latency,
1380 * rather than waiting until we've done them all. I haven't
1381 * measured to see if it makes any difference. */
1382 while (service_io(dev
))
1383 write(vblk
->done_fd
, &c
, 1);
1388 /* When the thread says some I/O is done, we interrupt the Guest. */
1389 static bool handle_io_finish(int fd
, struct device
*dev
)
1393 /* If child died, presumably it printed message. */
1394 if (read(dev
->fd
, &c
, 1) != 1)
1397 /* It did some work, so trigger the irq. */
1398 trigger_irq(fd
, dev
->vq
);
1402 /* When the Guest submits some I/O, we wake the I/O thread. */
1403 static void handle_virtblk_output(int fd
, struct virtqueue
*vq
)
1405 struct vblk_info
*vblk
= vq
->dev
->priv
;
1408 /* Wake up I/O thread and tell it to go to work! */
1409 if (write(vblk
->workpipe
[1], &c
, 1) != 1)
1410 /* Presumably it indicated why it died. */
1414 /* This creates a virtual block device. */
1415 static void setup_block_file(const char *filename
)
1419 struct vblk_info
*vblk
;
1424 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1427 /* The device responds to return from I/O thread. */
1428 dev
= new_device("block", VIRTIO_ID_BLOCK
, p
[0], handle_io_finish
);
1430 /* The device has a virtqueue. */
1431 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_virtblk_output
);
1433 /* Allocate the room for our own bookkeeping */
1434 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1436 /* First we open the file and store the length. */
1437 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1438 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1440 /* Tell Guest how many sectors this device has. */
1441 cap
= cpu_to_le64(vblk
->len
/ 512);
1442 add_desc_field(dev
, VIRTIO_CONFIG_BLK_F_CAPACITY
, sizeof(cap
), &cap
);
1444 /* Tell Guest not to put in too many descriptors at once: two are used
1445 * for the in and out elements. */
1446 val
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1447 add_desc_field(dev
, VIRTIO_CONFIG_BLK_F_SEG_MAX
, sizeof(val
), &val
);
1449 /* The I/O thread writes to this end of the pipe when done. */
1450 vblk
->done_fd
= p
[1];
1452 /* This is how we tell the I/O thread about more work. */
1453 pipe(vblk
->workpipe
);
1455 /* Create stack for thread and run it */
1456 stack
= malloc(32768);
1457 if (clone(io_thread
, stack
+ 32768, CLONE_VM
, dev
) == -1)
1458 err(1, "Creating clone");
1460 /* We don't need to keep the I/O thread's end of the pipes open. */
1461 close(vblk
->done_fd
);
1462 close(vblk
->workpipe
[0]);
1464 verbose("device %u: virtblock %llu sectors\n",
1465 devices
.device_num
, cap
);
1467 /* That's the end of device setup. */
1469 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1470 * its input and output, and finally, lays it to rest. */
1471 static void __attribute__((noreturn
)) run_guest(int lguest_fd
)
1474 unsigned long args
[] = { LHREQ_BREAK
, 0 };
1475 unsigned long notify_addr
;
1478 /* We read from the /dev/lguest device to run the Guest. */
1479 readval
= read(lguest_fd
, ¬ify_addr
, sizeof(notify_addr
));
1481 /* One unsigned long means the Guest did HCALL_NOTIFY */
1482 if (readval
== sizeof(notify_addr
)) {
1483 verbose("Notify on address %#lx\n", notify_addr
);
1484 handle_output(lguest_fd
, notify_addr
);
1486 /* ENOENT means the Guest died. Reading tells us why. */
1487 } else if (errno
== ENOENT
) {
1488 char reason
[1024] = { 0 };
1489 read(lguest_fd
, reason
, sizeof(reason
)-1);
1490 errx(1, "%s", reason
);
1491 /* EAGAIN means the waker wanted us to look at some input.
1492 * Anything else means a bug or incompatible change. */
1493 } else if (errno
!= EAGAIN
)
1494 err(1, "Running guest failed");
1496 /* Service input, then unset the BREAK which releases
1498 handle_input(lguest_fd
);
1499 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
1500 err(1, "Resetting break");
1504 * This is the end of the Launcher.
1506 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1507 * Drivers. If we were to see the Host kernel I/O code, our understanding
1508 * would be complete... :*/
1510 static struct option opts
[] = {
1511 { "verbose", 0, NULL
, 'v' },
1512 { "tunnet", 1, NULL
, 't' },
1513 { "block", 1, NULL
, 'b' },
1514 { "initrd", 1, NULL
, 'i' },
1517 static void usage(void)
1519 errx(1, "Usage: lguest [--verbose] "
1520 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1521 "|--block=<filename>|--initrd=<filename>]...\n"
1522 "<mem-in-mb> vmlinux [args...]");
1525 /*L:105 The main routine is where the real work begins: */
1526 int main(int argc
, char *argv
[])
1528 /* Memory, top-level pagetable, code startpoint and size of the
1529 * (optional) initrd. */
1530 unsigned long mem
= 0, pgdir
, start
, initrd_size
= 0;
1531 /* A temporary and the /dev/lguest file descriptor. */
1532 int i
, c
, lguest_fd
;
1533 /* The boot information for the Guest. */
1535 /* If they specify an initrd file to load. */
1536 const char *initrd_name
= NULL
;
1538 /* First we initialize the device list. Since console and network
1539 * device receive input from a file descriptor, we keep an fdset
1540 * (infds) and the maximum fd number (max_infd) with the head of the
1541 * list. We also keep a pointer to the last device, for easy appending
1542 * to the list. Finally, we keep the next interrupt number to hand out
1543 * (1: remember that 0 is used by the timer). */
1544 FD_ZERO(&devices
.infds
);
1545 devices
.max_infd
= -1;
1546 devices
.lastdev
= &devices
.dev
;
1547 devices
.next_irq
= 1;
1549 /* We need to know how much memory so we can set up the device
1550 * descriptor and memory pages for the devices as we parse the command
1551 * line. So we quickly look through the arguments to find the amount
1553 for (i
= 1; i
< argc
; i
++) {
1554 if (argv
[i
][0] != '-') {
1555 mem
= atoi(argv
[i
]) * 1024 * 1024;
1556 /* We start by mapping anonymous pages over all of
1557 * guest-physical memory range. This fills it with 0,
1558 * and ensures that the Guest won't be killed when it
1559 * tries to access it. */
1560 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1563 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1564 devices
.descpage
= get_pages(1);
1569 /* The options are fairly straight-forward */
1570 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1576 setup_tun_net(optarg
);
1579 setup_block_file(optarg
);
1582 initrd_name
= optarg
;
1585 warnx("Unknown argument %s", argv
[optind
]);
1589 /* After the other arguments we expect memory and kernel image name,
1590 * followed by command line arguments for the kernel. */
1591 if (optind
+ 2 > argc
)
1594 verbose("Guest base is at %p\n", guest_base
);
1596 /* We always have a console device */
1599 /* Now we load the kernel */
1600 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1602 /* Boot information is stashed at physical address 0 */
1603 boot
= from_guest_phys(0);
1605 /* Map the initrd image if requested (at top of physical memory) */
1607 initrd_size
= load_initrd(initrd_name
, mem
);
1608 /* These are the location in the Linux boot header where the
1609 * start and size of the initrd are expected to be found. */
1610 *(unsigned long *)(boot
+0x218) = mem
- initrd_size
;
1611 *(unsigned long *)(boot
+0x21c) = initrd_size
;
1612 /* The bootloader type 0xFF means "unknown"; that's OK. */
1613 *(unsigned char *)(boot
+0x210) = 0xFF;
1616 /* Set up the initial linear pagetables, starting below the initrd. */
1617 pgdir
= setup_pagetables(mem
, initrd_size
);
1619 /* The Linux boot header contains an "E820" memory map: ours is a
1620 * simple, single region. */
1621 *(char*)(boot
+E820NR
) = 1;
1622 *((struct e820entry
*)(boot
+E820MAP
))
1623 = ((struct e820entry
) { 0, mem
, E820_RAM
});
1624 /* The boot header contains a command line pointer: we put the command
1625 * line after the boot header (at address 4096) */
1626 *(u32
*)(boot
+ 0x228) = 4096;
1627 concat(boot
+ 4096, argv
+optind
+2);
1629 /* Boot protocol version: 2.07 supports the fields for lguest. */
1630 *(u16
*)(boot
+ 0x206) = 0x207;
1632 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1633 *(u32
*)(boot
+ 0x23c) = 1;
1635 /* Set bit 6 of the loadflags (aka. KEEP_SEGMENTS) so the entry path
1636 * does not try to reload segment registers. */
1637 *(u8
*)(boot
+ 0x211) |= (1 << 6);
1639 /* We tell the kernel to initialize the Guest: this returns the open
1640 * /dev/lguest file descriptor. */
1641 lguest_fd
= tell_kernel(pgdir
, start
);
1643 /* We fork off a child process, which wakes the Launcher whenever one
1644 * of the input file descriptors needs attention. Otherwise we would
1645 * run the Guest until it tries to output something. */
1646 waker_fd
= setup_waker(lguest_fd
);
1648 /* Finally, run the Guest. This doesn't return. */
1649 run_guest(lguest_fd
);
1654 * Mastery is done: you now know everything I do.
1656 * But surely you have seen code, features and bugs in your wanderings which
1657 * you now yearn to attack? That is the real game, and I look forward to you
1658 * patching and forking lguest into the Your-Name-Here-visor.
1660 * Farewell, and good coding!