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>
39 #include "linux/lguest_launcher.h"
40 #include "linux/virtio_config.h"
41 #include "linux/virtio_net.h"
42 #include "linux/virtio_blk.h"
43 #include "linux/virtio_console.h"
44 #include "linux/virtio_rng.h"
45 #include "linux/virtio_ring.h"
46 #include "asm-x86/bootparam.h"
47 /*L:110 We can ignore the 39 include files we need for this program, but I do
48 * want to draw attention to the use of kernel-style types.
50 * As Linus said, "C is a Spartan language, and so should your naming be." I
51 * like these abbreviations, so we define them here. Note that u64 is always
52 * unsigned long long, which works on all Linux systems: this means that we can
53 * use %llu in printf for any u64. */
54 typedef unsigned long long u64
;
60 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
62 #define BRIDGE_PFX "bridge:"
64 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
66 /* We can have up to 256 pages for devices. */
67 #define DEVICE_PAGES 256
68 /* This will occupy 2 pages: it must be a power of 2. */
69 #define VIRTQUEUE_NUM 128
71 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
72 * this, and although I wouldn't recommend it, it works quite nicely here. */
74 #define verbose(args...) \
75 do { if (verbose) printf(args); } while(0)
78 /* The pipe to send commands to the waker process */
80 /* The pointer to the start of guest memory. */
81 static void *guest_base
;
82 /* The maximum guest physical address allowed, and maximum possible. */
83 static unsigned long guest_limit
, guest_max
;
85 /* a per-cpu variable indicating whose vcpu is currently running */
86 static unsigned int __thread cpu_id
;
88 /* This is our list of devices. */
91 /* Summary information about the devices in our list: ready to pass to
92 * select() to ask which need servicing.*/
96 /* Counter to assign interrupt numbers. */
97 unsigned int next_irq
;
99 /* Counter to print out convenient device numbers. */
100 unsigned int device_num
;
102 /* The descriptor page for the devices. */
105 /* A single linked list of devices. */
107 /* And a pointer to the last device for easy append and also for
108 * configuration appending. */
109 struct device
*lastdev
;
112 /* The list of Guest devices, based on command line arguments. */
113 static struct device_list devices
;
115 /* The device structure describes a single device. */
118 /* The linked-list pointer. */
121 /* The this device's descriptor, as mapped into the Guest. */
122 struct lguest_device_desc
*desc
;
124 /* The name of this device, for --verbose. */
127 /* If handle_input is set, it wants to be called when this file
128 * descriptor is ready. */
130 bool (*handle_input
)(int fd
, struct device
*me
);
132 /* Any queues attached to this device */
133 struct virtqueue
*vq
;
135 /* Handle status being finalized (ie. feature bits stable). */
136 void (*ready
)(struct device
*me
);
138 /* Device-specific data. */
142 /* The virtqueue structure describes a queue attached to a device. */
145 struct virtqueue
*next
;
147 /* Which device owns me. */
150 /* The configuration for this queue. */
151 struct lguest_vqconfig config
;
153 /* The actual ring of buffers. */
156 /* Last available index we saw. */
159 /* The routine to call when the Guest pings us. */
160 void (*handle_output
)(int fd
, struct virtqueue
*me
);
162 /* Outstanding buffers */
163 unsigned int inflight
;
166 /* Remember the arguments to the program so we can "reboot" */
167 static char **main_args
;
169 /* Since guest is UP and we don't run at the same time, we don't need barriers.
170 * But I include them in the code in case others copy it. */
173 /* Convert an iovec element to the given type.
175 * This is a fairly ugly trick: we need to know the size of the type and
176 * alignment requirement to check the pointer is kosher. It's also nice to
177 * have the name of the type in case we report failure.
179 * Typing those three things all the time is cumbersome and error prone, so we
180 * have a macro which sets them all up and passes to the real function. */
181 #define convert(iov, type) \
182 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
184 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
187 if (iov
->iov_len
!= size
)
188 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
189 if ((unsigned long)iov
->iov_base
% align
!= 0)
190 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
191 return iov
->iov_base
;
194 /* Wrapper for the last available index. Makes it easier to change. */
195 #define lg_last_avail(vq) ((vq)->last_avail_idx)
197 /* The virtio configuration space is defined to be little-endian. x86 is
198 * little-endian too, but it's nice to be explicit so we have these helpers. */
199 #define cpu_to_le16(v16) (v16)
200 #define cpu_to_le32(v32) (v32)
201 #define cpu_to_le64(v64) (v64)
202 #define le16_to_cpu(v16) (v16)
203 #define le32_to_cpu(v32) (v32)
204 #define le64_to_cpu(v64) (v64)
206 /* Is this iovec empty? */
207 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
211 for (i
= 0; i
< num_iov
; i
++)
217 /* Take len bytes from the front of this iovec. */
218 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
222 for (i
= 0; i
< num_iov
; i
++) {
225 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
226 iov
[i
].iov_base
+= used
;
227 iov
[i
].iov_len
-= used
;
233 /* The device virtqueue descriptors are followed by feature bitmasks. */
234 static u8
*get_feature_bits(struct device
*dev
)
236 return (u8
*)(dev
->desc
+ 1)
237 + dev
->desc
->num_vq
* sizeof(struct lguest_vqconfig
);
240 /*L:100 The Launcher code itself takes us out into userspace, that scary place
241 * where pointers run wild and free! Unfortunately, like most userspace
242 * programs, it's quite boring (which is why everyone likes to hack on the
243 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
244 * will get you through this section. Or, maybe not.
246 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
247 * memory and stores it in "guest_base". In other words, Guest physical ==
248 * Launcher virtual with an offset.
250 * This can be tough to get your head around, but usually it just means that we
251 * use these trivial conversion functions when the Guest gives us it's
252 * "physical" addresses: */
253 static void *from_guest_phys(unsigned long addr
)
255 return guest_base
+ addr
;
258 static unsigned long to_guest_phys(const void *addr
)
260 return (addr
- guest_base
);
264 * Loading the Kernel.
266 * We start with couple of simple helper routines. open_or_die() avoids
267 * error-checking code cluttering the callers: */
268 static int open_or_die(const char *name
, int flags
)
270 int fd
= open(name
, flags
);
272 err(1, "Failed to open %s", name
);
276 /* map_zeroed_pages() takes a number of pages. */
277 static void *map_zeroed_pages(unsigned int num
)
279 int fd
= open_or_die("/dev/zero", O_RDONLY
);
282 /* We use a private mapping (ie. if we write to the page, it will be
284 addr
= mmap(NULL
, getpagesize() * num
,
285 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
286 if (addr
== MAP_FAILED
)
287 err(1, "Mmaping %u pages of /dev/zero", num
);
293 /* Get some more pages for a device. */
294 static void *get_pages(unsigned int num
)
296 void *addr
= from_guest_phys(guest_limit
);
298 guest_limit
+= num
* getpagesize();
299 if (guest_limit
> guest_max
)
300 errx(1, "Not enough memory for devices");
304 /* This routine is used to load the kernel or initrd. It tries mmap, but if
305 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
306 * it falls back to reading the memory in. */
307 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
311 /* We map writable even though for some segments are marked read-only.
312 * The kernel really wants to be writable: it patches its own
315 * MAP_PRIVATE means that the page won't be copied until a write is
316 * done to it. This allows us to share untouched memory between
318 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
319 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
322 /* pread does a seek and a read in one shot: saves a few lines. */
323 r
= pread(fd
, addr
, len
, offset
);
325 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
328 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
329 * the Guest memory. ELF = Embedded Linking Format, which is the format used
330 * by all modern binaries on Linux including the kernel.
332 * The ELF headers give *two* addresses: a physical address, and a virtual
333 * address. We use the physical address; the Guest will map itself to the
336 * We return the starting address. */
337 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
339 Elf32_Phdr phdr
[ehdr
->e_phnum
];
342 /* Sanity checks on the main ELF header: an x86 executable with a
343 * reasonable number of correctly-sized program headers. */
344 if (ehdr
->e_type
!= ET_EXEC
345 || ehdr
->e_machine
!= EM_386
346 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
347 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
348 errx(1, "Malformed elf header");
350 /* An ELF executable contains an ELF header and a number of "program"
351 * headers which indicate which parts ("segments") of the program to
354 /* We read in all the program headers at once: */
355 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
356 err(1, "Seeking to program headers");
357 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
358 err(1, "Reading program headers");
360 /* Try all the headers: there are usually only three. A read-only one,
361 * a read-write one, and a "note" section which we don't load. */
362 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
363 /* If this isn't a loadable segment, we ignore it */
364 if (phdr
[i
].p_type
!= PT_LOAD
)
367 verbose("Section %i: size %i addr %p\n",
368 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
370 /* We map this section of the file at its physical address. */
371 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
372 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
375 /* The entry point is given in the ELF header. */
376 return ehdr
->e_entry
;
379 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
380 * supposed to jump into it and it will unpack itself. We used to have to
381 * perform some hairy magic because the unpacking code scared me.
383 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
384 * a small patch to jump over the tricky bits in the Guest, so now we just read
385 * the funky header so we know where in the file to load, and away we go! */
386 static unsigned long load_bzimage(int fd
)
388 struct boot_params boot
;
390 /* Modern bzImages get loaded at 1M. */
391 void *p
= from_guest_phys(0x100000);
393 /* Go back to the start of the file and read the header. It should be
394 * a Linux boot header (see Documentation/i386/boot.txt) */
395 lseek(fd
, 0, SEEK_SET
);
396 read(fd
, &boot
, sizeof(boot
));
398 /* Inside the setup_hdr, we expect the magic "HdrS" */
399 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
400 errx(1, "This doesn't look like a bzImage to me");
402 /* Skip over the extra sectors of the header. */
403 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
405 /* Now read everything into memory. in nice big chunks. */
406 while ((r
= read(fd
, p
, 65536)) > 0)
409 /* Finally, code32_start tells us where to enter the kernel. */
410 return boot
.hdr
.code32_start
;
413 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
414 * come wrapped up in the self-decompressing "bzImage" format. With a little
415 * work, we can load those, too. */
416 static unsigned long load_kernel(int fd
)
420 /* Read in the first few bytes. */
421 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
422 err(1, "Reading kernel");
424 /* If it's an ELF file, it starts with "\177ELF" */
425 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
426 return map_elf(fd
, &hdr
);
428 /* Otherwise we assume it's a bzImage, and try to load it. */
429 return load_bzimage(fd
);
432 /* This is a trivial little helper to align pages. Andi Kleen hated it because
433 * it calls getpagesize() twice: "it's dumb code."
435 * Kernel guys get really het up about optimization, even when it's not
436 * necessary. I leave this code as a reaction against that. */
437 static inline unsigned long page_align(unsigned long addr
)
439 /* Add upwards and truncate downwards. */
440 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
443 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
444 * the kernel which the kernel can use to boot from without needing any
445 * drivers. Most distributions now use this as standard: the initrd contains
446 * the code to load the appropriate driver modules for the current machine.
448 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
449 * kernels. He sent me this (and tells me when I break it). */
450 static unsigned long load_initrd(const char *name
, unsigned long mem
)
456 ifd
= open_or_die(name
, O_RDONLY
);
457 /* fstat() is needed to get the file size. */
458 if (fstat(ifd
, &st
) < 0)
459 err(1, "fstat() on initrd '%s'", name
);
461 /* We map the initrd at the top of memory, but mmap wants it to be
462 * page-aligned, so we round the size up for that. */
463 len
= page_align(st
.st_size
);
464 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
465 /* Once a file is mapped, you can close the file descriptor. It's a
466 * little odd, but quite useful. */
468 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
470 /* We return the initrd size. */
474 /* Once we know how much memory we have we can construct simple linear page
475 * tables which set virtual == physical which will get the Guest far enough
476 * into the boot to create its own.
478 * We lay them out of the way, just below the initrd (which is why we need to
479 * know its size here). */
480 static unsigned long setup_pagetables(unsigned long mem
,
481 unsigned long initrd_size
)
483 unsigned long *pgdir
, *linear
;
484 unsigned int mapped_pages
, i
, linear_pages
;
485 unsigned int ptes_per_page
= getpagesize()/sizeof(void *);
487 mapped_pages
= mem
/getpagesize();
489 /* Each PTE page can map ptes_per_page pages: how many do we need? */
490 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
492 /* We put the toplevel page directory page at the top of memory. */
493 pgdir
= from_guest_phys(mem
) - initrd_size
- getpagesize();
495 /* Now we use the next linear_pages pages as pte pages */
496 linear
= (void *)pgdir
- linear_pages
*getpagesize();
498 /* Linear mapping is easy: put every page's address into the mapping in
499 * order. PAGE_PRESENT contains the flags Present, Writable and
501 for (i
= 0; i
< mapped_pages
; i
++)
502 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
504 /* The top level points to the linear page table pages above. */
505 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
506 pgdir
[i
/ptes_per_page
]
507 = ((to_guest_phys(linear
) + i
*sizeof(void *))
511 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
512 mapped_pages
, linear_pages
, to_guest_phys(linear
));
514 /* We return the top level (guest-physical) address: the kernel needs
515 * to know where it is. */
516 return to_guest_phys(pgdir
);
520 /* Simple routine to roll all the commandline arguments together with spaces
522 static void concat(char *dst
, char *args
[])
524 unsigned int i
, len
= 0;
526 for (i
= 0; args
[i
]; i
++) {
528 strcat(dst
+len
, " ");
531 strcpy(dst
+len
, args
[i
]);
532 len
+= strlen(args
[i
]);
534 /* In case it's empty. */
538 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
539 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
540 * the base of Guest "physical" memory, the top physical page to allow, the
541 * top level pagetable and the entry point for the Guest. */
542 static int tell_kernel(unsigned long pgdir
, unsigned long start
)
544 unsigned long args
[] = { LHREQ_INITIALIZE
,
545 (unsigned long)guest_base
,
546 guest_limit
/ getpagesize(), pgdir
, start
};
549 verbose("Guest: %p - %p (%#lx)\n",
550 guest_base
, guest_base
+ guest_limit
, guest_limit
);
551 fd
= open_or_die("/dev/lguest", O_RDWR
);
552 if (write(fd
, args
, sizeof(args
)) < 0)
553 err(1, "Writing to /dev/lguest");
555 /* We return the /dev/lguest file descriptor to control this Guest */
560 static void add_device_fd(int fd
)
562 FD_SET(fd
, &devices
.infds
);
563 if (fd
> devices
.max_infd
)
564 devices
.max_infd
= fd
;
570 * With console, block and network devices, we can have lots of input which we
571 * need to process. We could try to tell the kernel what file descriptors to
572 * watch, but handing a file descriptor mask through to the kernel is fairly
575 * Instead, we fork off a process which watches the file descriptors and writes
576 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
577 * stop running the Guest. This causes the Launcher to return from the
578 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
579 * the LHREQ_BREAK and wake us up again.
581 * This, of course, is merely a different *kind* of icky.
583 static void wake_parent(int pipefd
, int lguest_fd
)
585 /* Add the pipe from the Launcher to the fdset in the device_list, so
586 * we watch it, too. */
587 add_device_fd(pipefd
);
590 fd_set rfds
= devices
.infds
;
591 unsigned long args
[] = { LHREQ_BREAK
, 1 };
593 /* Wait until input is ready from one of the devices. */
594 select(devices
.max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
595 /* Is it a message from the Launcher? */
596 if (FD_ISSET(pipefd
, &rfds
)) {
598 /* If read() returns 0, it means the Launcher has
599 * exited. We silently follow. */
600 if (read(pipefd
, &fd
, sizeof(fd
)) == 0)
602 /* Otherwise it's telling us to change what file
603 * descriptors we're to listen to. Positive means
604 * listen to a new one, negative means stop
607 FD_SET(fd
, &devices
.infds
);
609 FD_CLR(-fd
- 1, &devices
.infds
);
610 } else /* Send LHREQ_BREAK command. */
611 pwrite(lguest_fd
, args
, sizeof(args
), cpu_id
);
615 /* This routine just sets up a pipe to the Waker process. */
616 static int setup_waker(int lguest_fd
)
618 int pipefd
[2], child
;
620 /* We create a pipe to talk to the Waker, and also so it knows when the
621 * Launcher dies (and closes pipe). */
628 /* We are the Waker: close the "writing" end of our copy of the
629 * pipe and start waiting for input. */
631 wake_parent(pipefd
[0], lguest_fd
);
633 /* Close the reading end of our copy of the pipe. */
636 /* Here is the fd used to talk to the waker. */
643 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
644 * We need to make sure it's not trying to reach into the Launcher itself, so
645 * we have a convenient routine which checks it and exits with an error message
646 * if something funny is going on:
648 static void *_check_pointer(unsigned long addr
, unsigned int size
,
651 /* We have to separately check addr and addr+size, because size could
652 * be huge and addr + size might wrap around. */
653 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
654 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
655 /* We return a pointer for the caller's convenience, now we know it's
657 return from_guest_phys(addr
);
659 /* A macro which transparently hands the line number to the real function. */
660 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
662 /* Each buffer in the virtqueues is actually a chain of descriptors. This
663 * function returns the next descriptor in the chain, or vq->vring.num if we're
665 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
669 /* If this descriptor says it doesn't chain, we're done. */
670 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
671 return vq
->vring
.num
;
673 /* Check they're not leading us off end of descriptors. */
674 next
= vq
->vring
.desc
[i
].next
;
675 /* Make sure compiler knows to grab that: we don't want it changing! */
678 if (next
>= vq
->vring
.num
)
679 errx(1, "Desc next is %u", next
);
684 /* This looks in the virtqueue and for the first available buffer, and converts
685 * it to an iovec for convenient access. Since descriptors consist of some
686 * number of output then some number of input descriptors, it's actually two
687 * iovecs, but we pack them into one and note how many of each there were.
689 * This function returns the descriptor number found, or vq->vring.num (which
690 * is never a valid descriptor number) if none was found. */
691 static unsigned get_vq_desc(struct virtqueue
*vq
,
693 unsigned int *out_num
, unsigned int *in_num
)
695 unsigned int i
, head
;
698 /* Check it isn't doing very strange things with descriptor numbers. */
699 last_avail
= lg_last_avail(vq
);
700 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
701 errx(1, "Guest moved used index from %u to %u",
702 last_avail
, vq
->vring
.avail
->idx
);
704 /* If there's nothing new since last we looked, return invalid. */
705 if (vq
->vring
.avail
->idx
== last_avail
)
706 return vq
->vring
.num
;
708 /* Grab the next descriptor number they're advertising, and increment
709 * the index we've seen. */
710 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
713 /* If their number is silly, that's a fatal mistake. */
714 if (head
>= vq
->vring
.num
)
715 errx(1, "Guest says index %u is available", head
);
717 /* When we start there are none of either input nor output. */
718 *out_num
= *in_num
= 0;
722 /* Grab the first descriptor, and check it's OK. */
723 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
724 iov
[*out_num
+ *in_num
].iov_base
725 = check_pointer(vq
->vring
.desc
[i
].addr
,
726 vq
->vring
.desc
[i
].len
);
727 /* If this is an input descriptor, increment that count. */
728 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
731 /* If it's an output descriptor, they're all supposed
732 * to come before any input descriptors. */
734 errx(1, "Descriptor has out after in");
738 /* If we've got too many, that implies a descriptor loop. */
739 if (*out_num
+ *in_num
> vq
->vring
.num
)
740 errx(1, "Looped descriptor");
741 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
747 /* After we've used one of their buffers, we tell them about it. We'll then
748 * want to send them an interrupt, using trigger_irq(). */
749 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
751 struct vring_used_elem
*used
;
753 /* The virtqueue contains a ring of used buffers. Get a pointer to the
754 * next entry in that used ring. */
755 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
758 /* Make sure buffer is written before we update index. */
760 vq
->vring
.used
->idx
++;
764 /* This actually sends the interrupt for this virtqueue */
765 static void trigger_irq(int fd
, struct virtqueue
*vq
)
767 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
769 /* If they don't want an interrupt, don't send one, unless empty. */
770 if ((vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
774 /* Send the Guest an interrupt tell them we used something up. */
775 if (write(fd
, buf
, sizeof(buf
)) != 0)
776 err(1, "Triggering irq %i", vq
->config
.irq
);
779 /* And here's the combo meal deal. Supersize me! */
780 static void add_used_and_trigger(int fd
, struct virtqueue
*vq
,
781 unsigned int head
, int len
)
783 add_used(vq
, head
, len
);
790 * Here is the input terminal setting we save, and the routine to restore them
791 * on exit so the user gets their terminal back. */
792 static struct termios orig_term
;
793 static void restore_term(void)
795 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
798 /* We associate some data with the console for our exit hack. */
801 /* How many times have they hit ^C? */
803 /* When did they start? */
804 struct timeval start
;
807 /* This is the routine which handles console input (ie. stdin). */
808 static bool handle_console_input(int fd
, struct device
*dev
)
811 unsigned int head
, in_num
, out_num
;
812 struct iovec iov
[dev
->vq
->vring
.num
];
813 struct console_abort
*abort
= dev
->priv
;
815 /* First we need a console buffer from the Guests's input virtqueue. */
816 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
818 /* If they're not ready for input, stop listening to this file
819 * descriptor. We'll start again once they add an input buffer. */
820 if (head
== dev
->vq
->vring
.num
)
824 errx(1, "Output buffers in console in queue?");
826 /* This is why we convert to iovecs: the readv() call uses them, and so
827 * it reads straight into the Guest's buffer. */
828 len
= readv(dev
->fd
, iov
, in_num
);
830 /* This implies that the console is closed, is /dev/null, or
831 * something went terribly wrong. */
832 warnx("Failed to get console input, ignoring console.");
833 /* Put the input terminal back. */
835 /* Remove callback from input vq, so it doesn't restart us. */
836 dev
->vq
->handle_output
= NULL
;
837 /* Stop listening to this fd: don't call us again. */
841 /* Tell the Guest about the new input. */
842 add_used_and_trigger(fd
, dev
->vq
, head
, len
);
844 /* Three ^C within one second? Exit.
846 * This is such a hack, but works surprisingly well. Each ^C has to be
847 * in a buffer by itself, so they can't be too fast. But we check that
848 * we get three within about a second, so they can't be too slow. */
849 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
851 gettimeofday(&abort
->start
, NULL
);
852 else if (abort
->count
== 3) {
854 gettimeofday(&now
, NULL
);
855 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
856 unsigned long args
[] = { LHREQ_BREAK
, 0 };
857 /* Close the fd so Waker will know it has to
860 /* Just in case waker is blocked in BREAK, send
862 write(fd
, args
, sizeof(args
));
868 /* Any other key resets the abort counter. */
871 /* Everything went OK! */
875 /* Handling output for console is simple: we just get all the output buffers
876 * and write them to stdout. */
877 static void handle_console_output(int fd
, struct virtqueue
*vq
)
879 unsigned int head
, out
, in
;
881 struct iovec iov
[vq
->vring
.num
];
883 /* Keep getting output buffers from the Guest until we run out. */
884 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
886 errx(1, "Input buffers in output queue?");
887 len
= writev(STDOUT_FILENO
, iov
, out
);
888 add_used_and_trigger(fd
, vq
, head
, len
);
895 * Handling output for network is also simple: we get all the output buffers
896 * and write them (ignoring the first element) to this device's file descriptor
899 static void handle_net_output(int fd
, struct virtqueue
*vq
)
901 unsigned int head
, out
, in
;
903 struct iovec iov
[vq
->vring
.num
];
905 /* Keep getting output buffers from the Guest until we run out. */
906 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
908 errx(1, "Input buffers in output queue?");
909 /* Check header, but otherwise ignore it (we told the Guest we
910 * supported no features, so it shouldn't have anything
912 (void)convert(&iov
[0], struct virtio_net_hdr
);
913 len
= writev(vq
->dev
->fd
, iov
+1, out
-1);
914 add_used_and_trigger(fd
, vq
, head
, len
);
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
];
925 struct virtio_net_hdr
*hdr
;
927 /* First we need a network buffer from the Guests's recv virtqueue. */
928 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
929 if (head
== dev
->vq
->vring
.num
) {
930 /* Now, it's expected that if we try to send a packet too
931 * early, the Guest won't be ready yet. Wait until the device
932 * status says it's ready. */
933 /* FIXME: Actually want DRIVER_ACTIVE here. */
934 if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)
935 warn("network: no dma buffer!");
937 /* Now tell it we want to know if new things appear. */
938 dev
->vq
->vring
.used
->flags
&= ~VRING_USED_F_NO_NOTIFY
;
941 /* We'll turn this back on if input buffers are registered. */
944 errx(1, "Output buffers in network recv queue?");
946 /* First element is the header: we set it to 0 (no features). */
947 hdr
= convert(&iov
[0], struct virtio_net_hdr
);
949 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
951 /* Read the packet from the device directly into the Guest's buffer. */
952 len
= readv(dev
->fd
, iov
+1, in_num
-1);
954 err(1, "reading network");
956 /* Tell the Guest about the new packet. */
957 add_used_and_trigger(fd
, dev
->vq
, head
, sizeof(*hdr
) + len
);
959 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
960 ((u8
*)iov
[1].iov_base
)[0], ((u8
*)iov
[1].iov_base
)[1],
961 head
!= dev
->vq
->vring
.num
? "sent" : "discarded");
967 /*L:215 This is the callback attached to the network and console input
968 * virtqueues: it ensures we try again, in case we stopped console or net
969 * delivery because Guest didn't have any buffers. */
970 static void enable_fd(int fd
, struct virtqueue
*vq
)
972 add_device_fd(vq
->dev
->fd
);
973 /* Tell waker to listen to it again */
974 write(waker_fd
, &vq
->dev
->fd
, sizeof(vq
->dev
->fd
));
977 static void net_enable_fd(int fd
, struct virtqueue
*vq
)
979 /* We don't need to know again when Guest refills receive buffer. */
980 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
984 /* When the Guest tells us they updated the status field, we handle it. */
985 static void update_device_status(struct device
*dev
)
987 struct virtqueue
*vq
;
989 /* This is a reset. */
990 if (dev
->desc
->status
== 0) {
991 verbose("Resetting device %s\n", dev
->name
);
993 /* Clear any features they've acked. */
994 memset(get_feature_bits(dev
) + dev
->desc
->feature_len
, 0,
995 dev
->desc
->feature_len
);
997 /* Zero out the virtqueues. */
998 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
999 memset(vq
->vring
.desc
, 0,
1000 vring_size(vq
->config
.num
, getpagesize()));
1001 lg_last_avail(vq
) = 0;
1003 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
1004 warnx("Device %s configuration FAILED", dev
->name
);
1005 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
1008 verbose("Device %s OK: offered", dev
->name
);
1009 for (i
= 0; i
< dev
->desc
->feature_len
; i
++)
1010 verbose(" %02x", get_feature_bits(dev
)[i
]);
1011 verbose(", accepted");
1012 for (i
= 0; i
< dev
->desc
->feature_len
; i
++)
1013 verbose(" %02x", get_feature_bits(dev
)
1014 [dev
->desc
->feature_len
+i
]);
1021 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
1022 static void handle_output(int fd
, unsigned long addr
)
1025 struct virtqueue
*vq
;
1027 /* Check each device and virtqueue. */
1028 for (i
= devices
.dev
; i
; i
= i
->next
) {
1029 /* Notifications to device descriptors update device status. */
1030 if (from_guest_phys(addr
) == i
->desc
) {
1031 update_device_status(i
);
1035 /* Notifications to virtqueues mean output has occurred. */
1036 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1037 if (vq
->config
.pfn
!= addr
/getpagesize())
1040 /* Guest should acknowledge (and set features!) before
1041 * using the device. */
1042 if (i
->desc
->status
== 0) {
1043 warnx("%s gave early output", i
->name
);
1047 if (strcmp(vq
->dev
->name
, "console") != 0)
1048 verbose("Output to %s\n", vq
->dev
->name
);
1049 if (vq
->handle_output
)
1050 vq
->handle_output(fd
, vq
);
1055 /* Early console write is done using notify on a nul-terminated string
1056 * in Guest memory. */
1057 if (addr
>= guest_limit
)
1058 errx(1, "Bad NOTIFY %#lx", addr
);
1060 write(STDOUT_FILENO
, from_guest_phys(addr
),
1061 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1064 /* This is called when the Waker wakes us up: check for incoming file
1066 static void handle_input(int fd
)
1068 /* select() wants a zeroed timeval to mean "don't wait". */
1069 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
1073 fd_set fds
= devices
.infds
;
1075 /* If nothing is ready, we're done. */
1076 if (select(devices
.max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
1079 /* Otherwise, call the device(s) which have readable file
1080 * descriptors and a method of handling them. */
1081 for (i
= devices
.dev
; i
; i
= i
->next
) {
1082 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
1084 if (i
->handle_input(fd
, i
))
1087 /* If handle_input() returns false, it means we
1088 * should no longer service it. Networking and
1089 * console do this when there's no input
1090 * buffers to deliver into. Console also uses
1091 * it when it discovers that stdin is closed. */
1092 FD_CLR(i
->fd
, &devices
.infds
);
1093 /* Tell waker to ignore it too, by sending a
1094 * negative fd number (-1, since 0 is a valid
1096 dev_fd
= -i
->fd
- 1;
1097 write(waker_fd
, &dev_fd
, sizeof(dev_fd
));
1106 * All devices need a descriptor so the Guest knows it exists, and a "struct
1107 * device" so the Launcher can keep track of it. We have common helper
1108 * routines to allocate and manage them.
1111 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1112 * number of virtqueue descriptors, then two sets of feature bits, then an
1113 * array of configuration bytes. This routine returns the configuration
1115 static u8
*device_config(const struct device
*dev
)
1117 return (void *)(dev
->desc
+ 1)
1118 + dev
->desc
->num_vq
* sizeof(struct lguest_vqconfig
)
1119 + dev
->desc
->feature_len
* 2;
1122 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1123 * table page just above the Guest's normal memory. It returns a pointer to
1124 * that descriptor. */
1125 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1127 struct lguest_device_desc d
= { .type
= type
};
1130 /* Figure out where the next device config is, based on the last one. */
1131 if (devices
.lastdev
)
1132 p
= device_config(devices
.lastdev
)
1133 + devices
.lastdev
->desc
->config_len
;
1135 p
= devices
.descpage
;
1137 /* We only have one page for all the descriptors. */
1138 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1139 errx(1, "Too many devices");
1141 /* p might not be aligned, so we memcpy in. */
1142 return memcpy(p
, &d
, sizeof(d
));
1145 /* Each device descriptor is followed by the description of its virtqueues. We
1146 * specify how many descriptors the virtqueue is to have. */
1147 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1148 void (*handle_output
)(int fd
, struct virtqueue
*me
))
1151 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1154 /* First we need some memory for this virtqueue. */
1155 pages
= (vring_size(num_descs
, getpagesize()) + getpagesize() - 1)
1157 p
= get_pages(pages
);
1159 /* Initialize the virtqueue */
1161 vq
->last_avail_idx
= 0;
1165 /* Initialize the configuration. */
1166 vq
->config
.num
= num_descs
;
1167 vq
->config
.irq
= devices
.next_irq
++;
1168 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1170 /* Initialize the vring. */
1171 vring_init(&vq
->vring
, num_descs
, p
, getpagesize());
1173 /* Append virtqueue to this device's descriptor. We use
1174 * device_config() to get the end of the device's current virtqueues;
1175 * we check that we haven't added any config or feature information
1176 * yet, otherwise we'd be overwriting them. */
1177 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1178 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1179 dev
->desc
->num_vq
++;
1181 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1183 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1185 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1188 /* Set the routine to call when the Guest does something to this
1190 vq
->handle_output
= handle_output
;
1192 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1193 * don't have a handler */
1195 vq
->vring
.used
->flags
= VRING_USED_F_NO_NOTIFY
;
1198 /* The first half of the feature bitmask is for us to advertise features. The
1199 * second half is for the Guest to accept features. */
1200 static void add_feature(struct device
*dev
, unsigned bit
)
1202 u8
*features
= get_feature_bits(dev
);
1204 /* We can't extend the feature bits once we've added config bytes */
1205 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1206 assert(dev
->desc
->config_len
== 0);
1207 dev
->desc
->feature_len
= (bit
/ CHAR_BIT
) + 1;
1210 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1213 /* This routine sets the configuration fields for an existing device's
1214 * descriptor. It only works for the last device, but that's OK because that's
1216 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1218 /* Check we haven't overflowed our single page. */
1219 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1220 errx(1, "Too many devices");
1222 /* Copy in the config information, and store the length. */
1223 memcpy(device_config(dev
), conf
, len
);
1224 dev
->desc
->config_len
= len
;
1227 /* This routine does all the creation and setup of a new device, including
1228 * calling new_dev_desc() to allocate the descriptor and device memory.
1230 * See what I mean about userspace being boring? */
1231 static struct device
*new_device(const char *name
, u16 type
, int fd
,
1232 bool (*handle_input
)(int, struct device
*))
1234 struct device
*dev
= malloc(sizeof(*dev
));
1236 /* Now we populate the fields one at a time. */
1238 /* If we have an input handler for this file descriptor, then we add it
1239 * to the device_list's fdset and maxfd. */
1241 add_device_fd(dev
->fd
);
1242 dev
->desc
= new_dev_desc(type
);
1243 dev
->handle_input
= handle_input
;
1248 /* Append to device list. Prepending to a single-linked list is
1249 * easier, but the user expects the devices to be arranged on the bus
1250 * in command-line order. The first network device on the command line
1251 * is eth0, the first block device /dev/vda, etc. */
1252 if (devices
.lastdev
)
1253 devices
.lastdev
->next
= dev
;
1256 devices
.lastdev
= dev
;
1261 /* Our first setup routine is the console. It's a fairly simple device, but
1262 * UNIX tty handling makes it uglier than it could be. */
1263 static void setup_console(void)
1267 /* If we can save the initial standard input settings... */
1268 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1269 struct termios term
= orig_term
;
1270 /* Then we turn off echo, line buffering and ^C etc. We want a
1271 * raw input stream to the Guest. */
1272 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1273 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1274 /* If we exit gracefully, the original settings will be
1275 * restored so the user can see what they're typing. */
1276 atexit(restore_term
);
1279 dev
= new_device("console", VIRTIO_ID_CONSOLE
,
1280 STDIN_FILENO
, handle_console_input
);
1281 /* We store the console state in dev->priv, and initialize it. */
1282 dev
->priv
= malloc(sizeof(struct console_abort
));
1283 ((struct console_abort
*)dev
->priv
)->count
= 0;
1285 /* The console needs two virtqueues: the input then the output. When
1286 * they put something the input queue, we make sure we're listening to
1287 * stdin. When they put something in the output queue, we write it to
1289 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1290 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_console_output
);
1292 verbose("device %u: console\n", devices
.device_num
++);
1296 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1297 * --sharenet=<name> option which opens or creates a named pipe. This can be
1298 * used to send packets to another guest in a 1:1 manner.
1300 * More sopisticated is to use one of the tools developed for project like UML
1303 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1304 * completely generic ("here's my vring, attach to your vring") and would work
1305 * for any traffic. Of course, namespace and permissions issues need to be
1306 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1307 * multiple inter-guest channels behind one interface, although it would
1308 * require some manner of hotplugging new virtio channels.
1310 * Finally, we could implement a virtio network switch in the kernel. :*/
1312 static u32
str2ip(const char *ipaddr
)
1316 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1317 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1318 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1321 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1324 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1325 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1326 errx(1, "Failed to parse mac address '%s'", macaddr
);
1335 /* This code is "adapted" from libbridge: it attaches the Host end of the
1336 * network device to the bridge device specified by the command line.
1338 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1339 * dislike bridging), and I just try not to break it. */
1340 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1346 errx(1, "must specify bridge name");
1348 ifidx
= if_nametoindex(if_name
);
1350 errx(1, "interface %s does not exist!", if_name
);
1352 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1353 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1354 ifr
.ifr_ifindex
= ifidx
;
1355 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1356 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1359 /* This sets up the Host end of the network device with an IP address, brings
1360 * it up so packets will flow, the copies the MAC address into the hwaddr
1362 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1365 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1367 memset(&ifr
, 0, sizeof(ifr
));
1368 strcpy(ifr
.ifr_name
, tapif
);
1370 /* Don't read these incantations. Just cut & paste them like I did! */
1371 sin
->sin_family
= AF_INET
;
1372 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1373 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1374 err(1, "Setting %s interface address", tapif
);
1375 ifr
.ifr_flags
= IFF_UP
;
1376 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1377 err(1, "Bringing interface %s up", tapif
);
1380 static void get_mac(int fd
, const char *tapif
, unsigned char hwaddr
[6])
1384 memset(&ifr
, 0, sizeof(ifr
));
1385 strcpy(ifr
.ifr_name
, tapif
);
1387 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1388 * above). IF means Interface, and HWADDR is hardware address.
1390 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1391 err(1, "getting hw address for %s", tapif
);
1392 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1395 static int get_tun_device(char tapif
[IFNAMSIZ
])
1400 /* Start with this zeroed. Messy but sure. */
1401 memset(&ifr
, 0, sizeof(ifr
));
1403 /* We open the /dev/net/tun device and tell it we want a tap device. A
1404 * tap device is like a tun device, only somehow different. To tell
1405 * the truth, I completely blundered my way through this code, but it
1407 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1408 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1409 strcpy(ifr
.ifr_name
, "tap%d");
1410 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1411 err(1, "configuring /dev/net/tun");
1413 /* We don't need checksums calculated for packets coming in this
1414 * device: trust us! */
1415 ioctl(netfd
, TUNSETNOCSUM
, 1);
1417 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1421 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1422 * routing, but the principle is the same: it uses the "tun" device to inject
1423 * packets into the Host as if they came in from a normal network card. We
1424 * just shunt packets between the Guest and the tun device. */
1425 static void setup_tun_net(char *arg
)
1429 u32 ip
= INADDR_ANY
;
1430 bool bridging
= false;
1431 char tapif
[IFNAMSIZ
], *p
;
1432 struct virtio_net_config conf
;
1434 netfd
= get_tun_device(tapif
);
1436 /* First we create a new network device. */
1437 dev
= new_device("net", VIRTIO_ID_NET
, netfd
, handle_tun_input
);
1439 /* Network devices need a receive and a send queue, just like
1441 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_enable_fd
);
1442 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_net_output
);
1444 /* We need a socket to perform the magic network ioctls to bring up the
1445 * tap interface, connect to the bridge etc. Any socket will do! */
1446 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1448 err(1, "opening IP socket");
1450 /* If the command line was --tunnet=bridge:<name> do bridging. */
1451 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1452 arg
+= strlen(BRIDGE_PFX
);
1456 /* A mac address may follow the bridge name or IP address */
1457 p
= strchr(arg
, ':');
1459 str2mac(p
+1, conf
.mac
);
1462 p
= arg
+ strlen(arg
);
1463 /* None supplied; query the randomly assigned mac. */
1464 get_mac(ipfd
, tapif
, conf
.mac
);
1467 /* arg is now either an IP address or a bridge name */
1469 add_to_bridge(ipfd
, tapif
, arg
);
1473 /* Set up the tun device. */
1474 configure_device(ipfd
, tapif
, ip
);
1476 /* Tell Guest what MAC address to use. */
1477 add_feature(dev
, VIRTIO_NET_F_MAC
);
1478 add_feature(dev
, VIRTIO_F_NOTIFY_ON_EMPTY
);
1479 set_config(dev
, sizeof(conf
), &conf
);
1481 /* We don't need the socket any more; setup is done. */
1484 devices
.device_num
++;
1487 verbose("device %u: tun %s attached to bridge: %s\n",
1488 devices
.device_num
, tapif
, arg
);
1490 verbose("device %u: tun %s: %s\n",
1491 devices
.device_num
, tapif
, arg
);
1494 /* Our block (disk) device should be really simple: the Guest asks for a block
1495 * number and we read or write that position in the file. Unfortunately, that
1496 * was amazingly slow: the Guest waits until the read is finished before
1497 * running anything else, even if it could have been doing useful work.
1499 * We could use async I/O, except it's reputed to suck so hard that characters
1500 * actually go missing from your code when you try to use it.
1502 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1504 /* This hangs off device->priv. */
1507 /* The size of the file. */
1510 /* The file descriptor for the file. */
1513 /* IO thread listens on this file descriptor [0]. */
1516 /* IO thread writes to this file descriptor to mark it done, then
1517 * Launcher triggers interrupt to Guest. */
1524 * Remember that the block device is handled by a separate I/O thread. We head
1525 * straight into the core of that thread here:
1527 static bool service_io(struct device
*dev
)
1529 struct vblk_info
*vblk
= dev
->priv
;
1530 unsigned int head
, out_num
, in_num
, wlen
;
1533 struct virtio_blk_outhdr
*out
;
1534 struct iovec iov
[dev
->vq
->vring
.num
];
1537 /* See if there's a request waiting. If not, nothing to do. */
1538 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1539 if (head
== dev
->vq
->vring
.num
)
1542 /* Every block request should contain at least one output buffer
1543 * (detailing the location on disk and the type of request) and one
1544 * input buffer (to hold the result). */
1545 if (out_num
== 0 || in_num
== 0)
1546 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1547 head
, out_num
, in_num
);
1549 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1550 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1551 off
= out
->sector
* 512;
1553 /* The block device implements "barriers", where the Guest indicates
1554 * that it wants all previous writes to occur before this write. We
1555 * don't have a way of asking our kernel to do a barrier, so we just
1556 * synchronize all the data in the file. Pretty poor, no? */
1557 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1558 fdatasync(vblk
->fd
);
1560 /* In general the virtio block driver is allowed to try SCSI commands.
1561 * It'd be nice if we supported eject, for example, but we don't. */
1562 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1563 fprintf(stderr
, "Scsi commands unsupported\n");
1564 *in
= VIRTIO_BLK_S_UNSUPP
;
1566 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1569 /* Move to the right location in the block file. This can fail
1570 * if they try to write past end. */
1571 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1572 err(1, "Bad seek to sector %llu", out
->sector
);
1574 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1575 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1577 /* Grr... Now we know how long the descriptor they sent was, we
1578 * make sure they didn't try to write over the end of the block
1579 * file (possibly extending it). */
1580 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1581 /* Trim it back to the correct length */
1582 ftruncate64(vblk
->fd
, vblk
->len
);
1583 /* Die, bad Guest, die. */
1584 errx(1, "Write past end %llu+%u", off
, ret
);
1587 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1591 /* Move to the right location in the block file. This can fail
1592 * if they try to read past end. */
1593 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1594 err(1, "Bad seek to sector %llu", out
->sector
);
1596 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1597 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1599 wlen
= sizeof(*in
) + ret
;
1600 *in
= VIRTIO_BLK_S_OK
;
1603 *in
= VIRTIO_BLK_S_IOERR
;
1607 /* We can't trigger an IRQ, because we're not the Launcher. It does
1608 * that when we tell it we're done. */
1609 add_used(dev
->vq
, head
, wlen
);
1613 /* This is the thread which actually services the I/O. */
1614 static int io_thread(void *_dev
)
1616 struct device
*dev
= _dev
;
1617 struct vblk_info
*vblk
= dev
->priv
;
1620 /* Close other side of workpipe so we get 0 read when main dies. */
1621 close(vblk
->workpipe
[1]);
1622 /* Close the other side of the done_fd pipe. */
1625 /* When this read fails, it means Launcher died, so we follow. */
1626 while (read(vblk
->workpipe
[0], &c
, 1) == 1) {
1627 /* We acknowledge each request immediately to reduce latency,
1628 * rather than waiting until we've done them all. I haven't
1629 * measured to see if it makes any difference.
1631 * That would be an interesting test, wouldn't it? You could
1632 * also try having more than one I/O thread. */
1633 while (service_io(dev
))
1634 write(vblk
->done_fd
, &c
, 1);
1639 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1640 * when that thread tells us it's completed some I/O. */
1641 static bool handle_io_finish(int fd
, struct device
*dev
)
1645 /* If the I/O thread died, presumably it printed the error, so we
1647 if (read(dev
->fd
, &c
, 1) != 1)
1650 /* It did some work, so trigger the irq. */
1651 trigger_irq(fd
, dev
->vq
);
1655 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1656 static void handle_virtblk_output(int fd
, struct virtqueue
*vq
)
1658 struct vblk_info
*vblk
= vq
->dev
->priv
;
1661 /* Wake up I/O thread and tell it to go to work! */
1662 if (write(vblk
->workpipe
[1], &c
, 1) != 1)
1663 /* Presumably it indicated why it died. */
1667 /*L:198 This actually sets up a virtual block device. */
1668 static void setup_block_file(const char *filename
)
1672 struct vblk_info
*vblk
;
1674 struct virtio_blk_config conf
;
1676 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1679 /* The device responds to return from I/O thread. */
1680 dev
= new_device("block", VIRTIO_ID_BLOCK
, p
[0], handle_io_finish
);
1682 /* The device has one virtqueue, where the Guest places requests. */
1683 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_virtblk_output
);
1685 /* Allocate the room for our own bookkeeping */
1686 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1688 /* First we open the file and store the length. */
1689 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1690 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1692 /* We support barriers. */
1693 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1695 /* Tell Guest how many sectors this device has. */
1696 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1698 /* Tell Guest not to put in too many descriptors at once: two are used
1699 * for the in and out elements. */
1700 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1701 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1703 set_config(dev
, sizeof(conf
), &conf
);
1705 /* The I/O thread writes to this end of the pipe when done. */
1706 vblk
->done_fd
= p
[1];
1708 /* This is the second pipe, which is how we tell the I/O thread about
1710 pipe(vblk
->workpipe
);
1712 /* Create stack for thread and run it. Since stack grows upwards, we
1713 * point the stack pointer to the end of this region. */
1714 stack
= malloc(32768);
1715 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1716 * becoming a zombie. */
1717 if (clone(io_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, dev
) == -1)
1718 err(1, "Creating clone");
1720 /* We don't need to keep the I/O thread's end of the pipes open. */
1721 close(vblk
->done_fd
);
1722 close(vblk
->workpipe
[0]);
1724 verbose("device %u: virtblock %llu sectors\n",
1725 devices
.device_num
, le64_to_cpu(conf
.capacity
));
1728 /* Our random number generator device reads from /dev/random into the Guest's
1729 * input buffers. The usual case is that the Guest doesn't want random numbers
1730 * and so has no buffers although /dev/random is still readable, whereas
1731 * console is the reverse.
1733 * The same logic applies, however. */
1734 static bool handle_rng_input(int fd
, struct device
*dev
)
1737 unsigned int head
, in_num
, out_num
, totlen
= 0;
1738 struct iovec iov
[dev
->vq
->vring
.num
];
1740 /* First we need a buffer from the Guests's virtqueue. */
1741 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1743 /* If they're not ready for input, stop listening to this file
1744 * descriptor. We'll start again once they add an input buffer. */
1745 if (head
== dev
->vq
->vring
.num
)
1749 errx(1, "Output buffers in rng?");
1751 /* This is why we convert to iovecs: the readv() call uses them, and so
1752 * it reads straight into the Guest's buffer. We loop to make sure we
1754 while (!iov_empty(iov
, in_num
)) {
1755 len
= readv(dev
->fd
, iov
, in_num
);
1757 err(1, "Read from /dev/random gave %i", len
);
1758 iov_consume(iov
, in_num
, len
);
1762 /* Tell the Guest about the new input. */
1763 add_used_and_trigger(fd
, dev
->vq
, head
, totlen
);
1765 /* Everything went OK! */
1769 /* And this creates a "hardware" random number device for the Guest. */
1770 static void setup_rng(void)
1775 fd
= open_or_die("/dev/random", O_RDONLY
);
1777 /* The device responds to return from I/O thread. */
1778 dev
= new_device("rng", VIRTIO_ID_RNG
, fd
, handle_rng_input
);
1780 /* The device has one virtqueue, where the Guest places inbufs. */
1781 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1783 verbose("device %u: rng\n", devices
.device_num
++);
1785 /* That's the end of device setup. */
1787 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1788 static void __attribute__((noreturn
)) restart_guest(void)
1792 /* Closing pipes causes the Waker thread and io_threads to die, and
1793 * closing /dev/lguest cleans up the Guest. Since we don't track all
1794 * open fds, we simply close everything beyond stderr. */
1795 for (i
= 3; i
< FD_SETSIZE
; i
++)
1797 execv(main_args
[0], main_args
);
1798 err(1, "Could not exec %s", main_args
[0]);
1801 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1802 * its input and output, and finally, lays it to rest. */
1803 static void __attribute__((noreturn
)) run_guest(int lguest_fd
)
1806 unsigned long args
[] = { LHREQ_BREAK
, 0 };
1807 unsigned long notify_addr
;
1810 /* We read from the /dev/lguest device to run the Guest. */
1811 readval
= pread(lguest_fd
, ¬ify_addr
,
1812 sizeof(notify_addr
), cpu_id
);
1814 /* One unsigned long means the Guest did HCALL_NOTIFY */
1815 if (readval
== sizeof(notify_addr
)) {
1816 verbose("Notify on address %#lx\n", notify_addr
);
1817 handle_output(lguest_fd
, notify_addr
);
1819 /* ENOENT means the Guest died. Reading tells us why. */
1820 } else if (errno
== ENOENT
) {
1821 char reason
[1024] = { 0 };
1822 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1823 errx(1, "%s", reason
);
1824 /* ERESTART means that we need to reboot the guest */
1825 } else if (errno
== ERESTART
) {
1827 /* EAGAIN means the Waker wanted us to look at some input.
1828 * Anything else means a bug or incompatible change. */
1829 } else if (errno
!= EAGAIN
)
1830 err(1, "Running guest failed");
1832 /* Only service input on thread for CPU 0. */
1836 /* Service input, then unset the BREAK to release the Waker. */
1837 handle_input(lguest_fd
);
1838 if (pwrite(lguest_fd
, args
, sizeof(args
), cpu_id
) < 0)
1839 err(1, "Resetting break");
1843 * This is the end of the Launcher. The good news: we are over halfway
1844 * through! The bad news: the most fiendish part of the code still lies ahead
1847 * Are you ready? Take a deep breath and join me in the core of the Host, in
1851 static struct option opts
[] = {
1852 { "verbose", 0, NULL
, 'v' },
1853 { "tunnet", 1, NULL
, 't' },
1854 { "block", 1, NULL
, 'b' },
1855 { "rng", 0, NULL
, 'r' },
1856 { "initrd", 1, NULL
, 'i' },
1859 static void usage(void)
1861 errx(1, "Usage: lguest [--verbose] "
1862 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1863 "|--block=<filename>|--initrd=<filename>]...\n"
1864 "<mem-in-mb> vmlinux [args...]");
1867 /*L:105 The main routine is where the real work begins: */
1868 int main(int argc
, char *argv
[])
1870 /* Memory, top-level pagetable, code startpoint and size of the
1871 * (optional) initrd. */
1872 unsigned long mem
= 0, pgdir
, start
, initrd_size
= 0;
1873 /* Two temporaries and the /dev/lguest file descriptor. */
1874 int i
, c
, lguest_fd
;
1875 /* The boot information for the Guest. */
1876 struct boot_params
*boot
;
1877 /* If they specify an initrd file to load. */
1878 const char *initrd_name
= NULL
;
1880 /* Save the args: we "reboot" by execing ourselves again. */
1882 /* We don't "wait" for the children, so prevent them from becoming
1884 signal(SIGCHLD
, SIG_IGN
);
1886 /* First we initialize the device list. Since console and network
1887 * device receive input from a file descriptor, we keep an fdset
1888 * (infds) and the maximum fd number (max_infd) with the head of the
1889 * list. We also keep a pointer to the last device. Finally, we keep
1890 * the next interrupt number to use for devices (1: remember that 0 is
1891 * used by the timer). */
1892 FD_ZERO(&devices
.infds
);
1893 devices
.max_infd
= -1;
1894 devices
.lastdev
= NULL
;
1895 devices
.next_irq
= 1;
1898 /* We need to know how much memory so we can set up the device
1899 * descriptor and memory pages for the devices as we parse the command
1900 * line. So we quickly look through the arguments to find the amount
1902 for (i
= 1; i
< argc
; i
++) {
1903 if (argv
[i
][0] != '-') {
1904 mem
= atoi(argv
[i
]) * 1024 * 1024;
1905 /* We start by mapping anonymous pages over all of
1906 * guest-physical memory range. This fills it with 0,
1907 * and ensures that the Guest won't be killed when it
1908 * tries to access it. */
1909 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1912 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1913 devices
.descpage
= get_pages(1);
1918 /* The options are fairly straight-forward */
1919 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1925 setup_tun_net(optarg
);
1928 setup_block_file(optarg
);
1934 initrd_name
= optarg
;
1937 warnx("Unknown argument %s", argv
[optind
]);
1941 /* After the other arguments we expect memory and kernel image name,
1942 * followed by command line arguments for the kernel. */
1943 if (optind
+ 2 > argc
)
1946 verbose("Guest base is at %p\n", guest_base
);
1948 /* We always have a console device */
1951 /* Now we load the kernel */
1952 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1954 /* Boot information is stashed at physical address 0 */
1955 boot
= from_guest_phys(0);
1957 /* Map the initrd image if requested (at top of physical memory) */
1959 initrd_size
= load_initrd(initrd_name
, mem
);
1960 /* These are the location in the Linux boot header where the
1961 * start and size of the initrd are expected to be found. */
1962 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1963 boot
->hdr
.ramdisk_size
= initrd_size
;
1964 /* The bootloader type 0xFF means "unknown"; that's OK. */
1965 boot
->hdr
.type_of_loader
= 0xFF;
1968 /* Set up the initial linear pagetables, starting below the initrd. */
1969 pgdir
= setup_pagetables(mem
, initrd_size
);
1971 /* The Linux boot header contains an "E820" memory map: ours is a
1972 * simple, single region. */
1973 boot
->e820_entries
= 1;
1974 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1975 /* The boot header contains a command line pointer: we put the command
1976 * line after the boot header. */
1977 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1978 /* We use a simple helper to copy the arguments separated by spaces. */
1979 concat((char *)(boot
+ 1), argv
+optind
+2);
1981 /* Boot protocol version: 2.07 supports the fields for lguest. */
1982 boot
->hdr
.version
= 0x207;
1984 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1985 boot
->hdr
.hardware_subarch
= 1;
1987 /* Tell the entry path not to try to reload segment registers. */
1988 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1990 /* We tell the kernel to initialize the Guest: this returns the open
1991 * /dev/lguest file descriptor. */
1992 lguest_fd
= tell_kernel(pgdir
, start
);
1994 /* We fork off a child process, which wakes the Launcher whenever one
1995 * of the input file descriptors needs attention. We call this the
1996 * Waker, and we'll cover it in a moment. */
1997 waker_fd
= setup_waker(lguest_fd
);
1999 /* Finally, run the Guest. This doesn't return. */
2000 run_guest(lguest_fd
);
2005 * Mastery is done: you now know everything I do.
2007 * But surely you have seen code, features and bugs in your wanderings which
2008 * you now yearn to attack? That is the real game, and I look forward to you
2009 * patching and forking lguest into the Your-Name-Here-visor.
2011 * Farewell, and good coding!