2 * This is the Launcher code, a simple program which lays out the "physical"
3 * memory for the new Guest by mapping the kernel image and the virtual
4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
7 #define _LARGEFILE64_SOURCE
17 #include <sys/param.h>
18 #include <sys/types.h>
21 #include <sys/eventfd.h>
26 #include <sys/socket.h>
27 #include <sys/ioctl.h>
30 #include <netinet/in.h>
32 #include <linux/sockios.h>
33 #include <linux/if_tun.h>
45 #include <linux/virtio_config.h>
46 #include <linux/virtio_net.h>
47 #include <linux/virtio_blk.h>
48 #include <linux/virtio_console.h>
49 #include <linux/virtio_rng.h>
50 #include <linux/virtio_ring.h>
51 #include <asm/bootparam.h>
52 #include "../../../include/linux/lguest_launcher.h"
54 * We can ignore the 42 include files we need for this program, but I do want
55 * to draw attention to the use of kernel-style types.
57 * As Linus said, "C is a Spartan language, and so should your naming be." I
58 * like these abbreviations, so we define them here. Note that u64 is always
59 * unsigned long long, which works on all Linux systems: this means that we can
60 * use %llu in printf for any u64.
62 typedef unsigned long long u64
;
68 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
69 #define BRIDGE_PFX "bridge:"
71 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
73 /* We can have up to 256 pages for devices. */
74 #define DEVICE_PAGES 256
75 /* This will occupy 3 pages: it must be a power of 2. */
76 #define VIRTQUEUE_NUM 256
79 * verbose is both a global flag and a macro. The C preprocessor allows
80 * this, and although I wouldn't recommend it, it works quite nicely here.
83 #define verbose(args...) \
84 do { if (verbose) printf(args); } while(0)
87 /* The pointer to the start of guest memory. */
88 static void *guest_base
;
89 /* The maximum guest physical address allowed, and maximum possible. */
90 static unsigned long guest_limit
, guest_max
;
91 /* The /dev/lguest file descriptor. */
94 /* a per-cpu variable indicating whose vcpu is currently running */
95 static unsigned int __thread cpu_id
;
97 /* This is our list of devices. */
99 /* Counter to assign interrupt numbers. */
100 unsigned int next_irq
;
102 /* Counter to print out convenient device numbers. */
103 unsigned int device_num
;
105 /* The descriptor page for the devices. */
108 /* A single linked list of devices. */
110 /* And a pointer to the last device for easy append. */
111 struct device
*lastdev
;
114 /* The list of Guest devices, based on command line arguments. */
115 static struct device_list devices
;
117 /* The device structure describes a single device. */
119 /* The linked-list pointer. */
122 /* The device's descriptor, as mapped into the Guest. */
123 struct lguest_device_desc
*desc
;
125 /* We can't trust desc values once Guest has booted: we use these. */
126 unsigned int feature_len
;
129 /* The name of this device, for --verbose. */
132 /* Any queues attached to this device */
133 struct virtqueue
*vq
;
135 /* Is it operational */
138 /* Device-specific data. */
142 /* The virtqueue structure describes a queue attached to a device. */
144 struct virtqueue
*next
;
146 /* Which device owns me. */
149 /* The configuration for this queue. */
150 struct lguest_vqconfig config
;
152 /* The actual ring of buffers. */
155 /* Last available index we saw. */
158 /* How many are used since we sent last irq? */
159 unsigned int pending_used
;
161 /* Eventfd where Guest notifications arrive. */
164 /* Function for the thread which is servicing this virtqueue. */
165 void (*service
)(struct virtqueue
*vq
);
169 /* Remember the arguments to the program so we can "reboot" */
170 static char **main_args
;
172 /* The original tty settings to restore on exit. */
173 static struct termios orig_term
;
176 * We have to be careful with barriers: our devices are all run in separate
177 * threads and so we need to make sure that changes visible to the Guest happen
180 #define wmb() __asm__ __volatile__("" : : : "memory")
181 #define mb() __asm__ __volatile__("" : : : "memory")
184 * Convert an iovec element to the given type.
186 * This is a fairly ugly trick: we need to know the size of the type and
187 * alignment requirement to check the pointer is kosher. It's also nice to
188 * have the name of the type in case we report failure.
190 * Typing those three things all the time is cumbersome and error prone, so we
191 * have a macro which sets them all up and passes to the real function.
193 #define convert(iov, type) \
194 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
196 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
199 if (iov
->iov_len
!= size
)
200 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
201 if ((unsigned long)iov
->iov_base
% align
!= 0)
202 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
203 return iov
->iov_base
;
206 /* Wrapper for the last available index. Makes it easier to change. */
207 #define lg_last_avail(vq) ((vq)->last_avail_idx)
210 * The virtio configuration space is defined to be little-endian. x86 is
211 * little-endian too, but it's nice to be explicit so we have these helpers.
213 #define cpu_to_le16(v16) (v16)
214 #define cpu_to_le32(v32) (v32)
215 #define cpu_to_le64(v64) (v64)
216 #define le16_to_cpu(v16) (v16)
217 #define le32_to_cpu(v32) (v32)
218 #define le64_to_cpu(v64) (v64)
220 /* Is this iovec empty? */
221 static bool iov_empty(const struct iovec iov
[], unsigned int num_iov
)
225 for (i
= 0; i
< num_iov
; i
++)
231 /* Take len bytes from the front of this iovec. */
232 static void iov_consume(struct iovec iov
[], unsigned num_iov
, unsigned len
)
236 for (i
= 0; i
< num_iov
; i
++) {
239 used
= iov
[i
].iov_len
< len
? iov
[i
].iov_len
: len
;
240 iov
[i
].iov_base
+= used
;
241 iov
[i
].iov_len
-= used
;
247 /* The device virtqueue descriptors are followed by feature bitmasks. */
248 static u8
*get_feature_bits(struct device
*dev
)
250 return (u8
*)(dev
->desc
+ 1)
251 + dev
->num_vq
* sizeof(struct lguest_vqconfig
);
255 * The Launcher code itself takes us out into userspace, that scary place where
256 * pointers run wild and free! Unfortunately, like most userspace programs,
257 * it's quite boring (which is why everyone likes to hack on the kernel!).
258 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
259 * you through this section. Or, maybe not.
261 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
262 * memory and stores it in "guest_base". In other words, Guest physical ==
263 * Launcher virtual with an offset.
265 * This can be tough to get your head around, but usually it just means that we
266 * use these trivial conversion functions when the Guest gives us its
267 * "physical" addresses:
269 static void *from_guest_phys(unsigned long addr
)
271 return guest_base
+ addr
;
274 static unsigned long to_guest_phys(const void *addr
)
276 return (addr
- guest_base
);
280 * Loading the Kernel.
282 * We start with couple of simple helper routines. open_or_die() avoids
283 * error-checking code cluttering the callers:
285 static int open_or_die(const char *name
, int flags
)
287 int fd
= open(name
, flags
);
289 err(1, "Failed to open %s", name
);
293 /* map_zeroed_pages() takes a number of pages. */
294 static void *map_zeroed_pages(unsigned int num
)
296 int fd
= open_or_die("/dev/zero", O_RDONLY
);
300 * We use a private mapping (ie. if we write to the page, it will be
301 * copied). We allocate an extra two pages PROT_NONE to act as guard
302 * pages against read/write attempts that exceed allocated space.
304 addr
= mmap(NULL
, getpagesize() * (num
+2),
305 PROT_NONE
, MAP_PRIVATE
, fd
, 0);
307 if (addr
== MAP_FAILED
)
308 err(1, "Mmapping %u pages of /dev/zero", num
);
310 if (mprotect(addr
+ getpagesize(), getpagesize() * num
,
311 PROT_READ
|PROT_WRITE
) == -1)
312 err(1, "mprotect rw %u pages failed", num
);
315 * One neat mmap feature is that you can close the fd, and it
320 /* Return address after PROT_NONE page */
321 return addr
+ getpagesize();
324 /* Get some more pages for a device. */
325 static void *get_pages(unsigned int num
)
327 void *addr
= from_guest_phys(guest_limit
);
329 guest_limit
+= num
* getpagesize();
330 if (guest_limit
> guest_max
)
331 errx(1, "Not enough memory for devices");
336 * This routine is used to load the kernel or initrd. It tries mmap, but if
337 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
338 * it falls back to reading the memory in.
340 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
345 * We map writable even though for some segments are marked read-only.
346 * The kernel really wants to be writable: it patches its own
349 * MAP_PRIVATE means that the page won't be copied until a write is
350 * done to it. This allows us to share untouched memory between
353 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
,
354 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
357 /* pread does a seek and a read in one shot: saves a few lines. */
358 r
= pread(fd
, addr
, len
, offset
);
360 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
364 * This routine takes an open vmlinux image, which is in ELF, and maps it into
365 * the Guest memory. ELF = Embedded Linking Format, which is the format used
366 * by all modern binaries on Linux including the kernel.
368 * The ELF headers give *two* addresses: a physical address, and a virtual
369 * address. We use the physical address; the Guest will map itself to the
372 * We return the starting address.
374 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
376 Elf32_Phdr phdr
[ehdr
->e_phnum
];
380 * Sanity checks on the main ELF header: an x86 executable with a
381 * reasonable number of correctly-sized program headers.
383 if (ehdr
->e_type
!= ET_EXEC
384 || ehdr
->e_machine
!= EM_386
385 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
386 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
387 errx(1, "Malformed elf header");
390 * An ELF executable contains an ELF header and a number of "program"
391 * headers which indicate which parts ("segments") of the program to
395 /* We read in all the program headers at once: */
396 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
397 err(1, "Seeking to program headers");
398 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
399 err(1, "Reading program headers");
402 * Try all the headers: there are usually only three. A read-only one,
403 * a read-write one, and a "note" section which we don't load.
405 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
406 /* If this isn't a loadable segment, we ignore it */
407 if (phdr
[i
].p_type
!= PT_LOAD
)
410 verbose("Section %i: size %i addr %p\n",
411 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
413 /* We map this section of the file at its physical address. */
414 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
415 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
418 /* The entry point is given in the ELF header. */
419 return ehdr
->e_entry
;
423 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
424 * to jump into it and it will unpack itself. We used to have to perform some
425 * hairy magic because the unpacking code scared me.
427 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
428 * a small patch to jump over the tricky bits in the Guest, so now we just read
429 * the funky header so we know where in the file to load, and away we go!
431 static unsigned long load_bzimage(int fd
)
433 struct boot_params boot
;
435 /* Modern bzImages get loaded at 1M. */
436 void *p
= from_guest_phys(0x100000);
439 * Go back to the start of the file and read the header. It should be
440 * a Linux boot header (see Documentation/x86/i386/boot.txt)
442 lseek(fd
, 0, SEEK_SET
);
443 read(fd
, &boot
, sizeof(boot
));
445 /* Inside the setup_hdr, we expect the magic "HdrS" */
446 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
447 errx(1, "This doesn't look like a bzImage to me");
449 /* Skip over the extra sectors of the header. */
450 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
452 /* Now read everything into memory. in nice big chunks. */
453 while ((r
= read(fd
, p
, 65536)) > 0)
456 /* Finally, code32_start tells us where to enter the kernel. */
457 return boot
.hdr
.code32_start
;
461 * Loading the kernel is easy when it's a "vmlinux", but most kernels
462 * come wrapped up in the self-decompressing "bzImage" format. With a little
463 * work, we can load those, too.
465 static unsigned long load_kernel(int fd
)
469 /* Read in the first few bytes. */
470 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
471 err(1, "Reading kernel");
473 /* If it's an ELF file, it starts with "\177ELF" */
474 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
475 return map_elf(fd
, &hdr
);
477 /* Otherwise we assume it's a bzImage, and try to load it. */
478 return load_bzimage(fd
);
482 * This is a trivial little helper to align pages. Andi Kleen hated it because
483 * it calls getpagesize() twice: "it's dumb code."
485 * Kernel guys get really het up about optimization, even when it's not
486 * necessary. I leave this code as a reaction against that.
488 static inline unsigned long page_align(unsigned long addr
)
490 /* Add upwards and truncate downwards. */
491 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
495 * An "initial ram disk" is a disk image loaded into memory along with the
496 * kernel which the kernel can use to boot from without needing any drivers.
497 * Most distributions now use this as standard: the initrd contains the code to
498 * load the appropriate driver modules for the current machine.
500 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
501 * kernels. He sent me this (and tells me when I break it).
503 static unsigned long load_initrd(const char *name
, unsigned long mem
)
509 ifd
= open_or_die(name
, O_RDONLY
);
510 /* fstat() is needed to get the file size. */
511 if (fstat(ifd
, &st
) < 0)
512 err(1, "fstat() on initrd '%s'", name
);
515 * We map the initrd at the top of memory, but mmap wants it to be
516 * page-aligned, so we round the size up for that.
518 len
= page_align(st
.st_size
);
519 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
521 * Once a file is mapped, you can close the file descriptor. It's a
522 * little odd, but quite useful.
525 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
527 /* We return the initrd size. */
533 * Simple routine to roll all the commandline arguments together with spaces
536 static void concat(char *dst
, char *args
[])
538 unsigned int i
, len
= 0;
540 for (i
= 0; args
[i
]; i
++) {
542 strcat(dst
+len
, " ");
545 strcpy(dst
+len
, args
[i
]);
546 len
+= strlen(args
[i
]);
548 /* In case it's empty. */
553 * This is where we actually tell the kernel to initialize the Guest. We
554 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
555 * the base of Guest "physical" memory, the top physical page to allow and the
556 * entry point for the Guest.
558 static void tell_kernel(unsigned long start
)
560 unsigned long args
[] = { LHREQ_INITIALIZE
,
561 (unsigned long)guest_base
,
562 guest_limit
/ getpagesize(), start
};
563 verbose("Guest: %p - %p (%#lx)\n",
564 guest_base
, guest_base
+ guest_limit
, guest_limit
);
565 lguest_fd
= open_or_die("/dev/lguest", O_RDWR
);
566 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
567 err(1, "Writing to /dev/lguest");
574 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
575 * We need to make sure it's not trying to reach into the Launcher itself, so
576 * we have a convenient routine which checks it and exits with an error message
577 * if something funny is going on:
579 static void *_check_pointer(unsigned long addr
, unsigned int size
,
583 * Check if the requested address and size exceeds the allocated memory,
584 * or addr + size wraps around.
586 if ((addr
+ size
) > guest_limit
|| (addr
+ size
) < addr
)
587 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
589 * We return a pointer for the caller's convenience, now we know it's
592 return from_guest_phys(addr
);
594 /* A macro which transparently hands the line number to the real function. */
595 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
598 * Each buffer in the virtqueues is actually a chain of descriptors. This
599 * function returns the next descriptor in the chain, or vq->vring.num if we're
602 static unsigned next_desc(struct vring_desc
*desc
,
603 unsigned int i
, unsigned int max
)
607 /* If this descriptor says it doesn't chain, we're done. */
608 if (!(desc
[i
].flags
& VRING_DESC_F_NEXT
))
611 /* Check they're not leading us off end of descriptors. */
613 /* Make sure compiler knows to grab that: we don't want it changing! */
617 errx(1, "Desc next is %u", next
);
623 * This actually sends the interrupt for this virtqueue, if we've used a
626 static void trigger_irq(struct virtqueue
*vq
)
628 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
630 /* Don't inform them if nothing used. */
631 if (!vq
->pending_used
)
633 vq
->pending_used
= 0;
635 /* If they don't want an interrupt, don't send one... */
636 if (vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
) {
640 /* Send the Guest an interrupt tell them we used something up. */
641 if (write(lguest_fd
, buf
, sizeof(buf
)) != 0)
642 err(1, "Triggering irq %i", vq
->config
.irq
);
646 * This looks in the virtqueue for the first available buffer, and converts
647 * it to an iovec for convenient access. Since descriptors consist of some
648 * number of output then some number of input descriptors, it's actually two
649 * iovecs, but we pack them into one and note how many of each there were.
651 * This function waits if necessary, and returns the descriptor number found.
653 static unsigned wait_for_vq_desc(struct virtqueue
*vq
,
655 unsigned int *out_num
, unsigned int *in_num
)
657 unsigned int i
, head
, max
;
658 struct vring_desc
*desc
;
659 u16 last_avail
= lg_last_avail(vq
);
661 /* There's nothing available? */
662 while (last_avail
== vq
->vring
.avail
->idx
) {
666 * Since we're about to sleep, now is a good time to tell the
667 * Guest about what we've used up to now.
671 /* OK, now we need to know about added descriptors. */
672 vq
->vring
.used
->flags
&= ~VRING_USED_F_NO_NOTIFY
;
675 * They could have slipped one in as we were doing that: make
676 * sure it's written, then check again.
679 if (last_avail
!= vq
->vring
.avail
->idx
) {
680 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
684 /* Nothing new? Wait for eventfd to tell us they refilled. */
685 if (read(vq
->eventfd
, &event
, sizeof(event
)) != sizeof(event
))
686 errx(1, "Event read failed?");
688 /* We don't need to be notified again. */
689 vq
->vring
.used
->flags
|= VRING_USED_F_NO_NOTIFY
;
692 /* Check it isn't doing very strange things with descriptor numbers. */
693 if ((u16
)(vq
->vring
.avail
->idx
- last_avail
) > vq
->vring
.num
)
694 errx(1, "Guest moved used index from %u to %u",
695 last_avail
, vq
->vring
.avail
->idx
);
698 * Grab the next descriptor number they're advertising, and increment
699 * the index we've seen.
701 head
= vq
->vring
.avail
->ring
[last_avail
% vq
->vring
.num
];
704 /* If their number is silly, that's a fatal mistake. */
705 if (head
>= vq
->vring
.num
)
706 errx(1, "Guest says index %u is available", head
);
708 /* When we start there are none of either input nor output. */
709 *out_num
= *in_num
= 0;
712 desc
= vq
->vring
.desc
;
716 * If this is an indirect entry, then this buffer contains a descriptor
717 * table which we handle as if it's any normal descriptor chain.
719 if (desc
[i
].flags
& VRING_DESC_F_INDIRECT
) {
720 if (desc
[i
].len
% sizeof(struct vring_desc
))
721 errx(1, "Invalid size for indirect buffer table");
723 max
= desc
[i
].len
/ sizeof(struct vring_desc
);
724 desc
= check_pointer(desc
[i
].addr
, desc
[i
].len
);
729 /* Grab the first descriptor, and check it's OK. */
730 iov
[*out_num
+ *in_num
].iov_len
= desc
[i
].len
;
731 iov
[*out_num
+ *in_num
].iov_base
732 = check_pointer(desc
[i
].addr
, desc
[i
].len
);
733 /* If this is an input descriptor, increment that count. */
734 if (desc
[i
].flags
& VRING_DESC_F_WRITE
)
738 * If it's an output descriptor, they're all supposed
739 * to come before any input descriptors.
742 errx(1, "Descriptor has out after in");
746 /* If we've got too many, that implies a descriptor loop. */
747 if (*out_num
+ *in_num
> max
)
748 errx(1, "Looped descriptor");
749 } while ((i
= next_desc(desc
, i
, max
)) != max
);
755 * After we've used one of their buffers, we tell the Guest about it. Sometime
756 * later we'll want to send them an interrupt using trigger_irq(); note that
757 * wait_for_vq_desc() does that for us if it has to wait.
759 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
761 struct vring_used_elem
*used
;
764 * The virtqueue contains a ring of used buffers. Get a pointer to the
765 * next entry in that used ring.
767 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
770 /* Make sure buffer is written before we update index. */
772 vq
->vring
.used
->idx
++;
776 /* And here's the combo meal deal. Supersize me! */
777 static void add_used_and_trigger(struct virtqueue
*vq
, unsigned head
, int len
)
779 add_used(vq
, head
, len
);
786 * We associate some data with the console for our exit hack.
788 struct console_abort
{
789 /* How many times have they hit ^C? */
791 /* When did they start? */
792 struct timeval start
;
795 /* This is the routine which handles console input (ie. stdin). */
796 static void console_input(struct virtqueue
*vq
)
799 unsigned int head
, in_num
, out_num
;
800 struct console_abort
*abort
= vq
->dev
->priv
;
801 struct iovec iov
[vq
->vring
.num
];
803 /* Make sure there's a descriptor available. */
804 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
806 errx(1, "Output buffers in console in queue?");
808 /* Read into it. This is where we usually wait. */
809 len
= readv(STDIN_FILENO
, iov
, in_num
);
811 /* Ran out of input? */
812 warnx("Failed to get console input, ignoring console.");
814 * For simplicity, dying threads kill the whole Launcher. So
821 /* Tell the Guest we used a buffer. */
822 add_used_and_trigger(vq
, head
, len
);
825 * Three ^C within one second? Exit.
827 * This is such a hack, but works surprisingly well. Each ^C has to
828 * be in a buffer by itself, so they can't be too fast. But we check
829 * that we get three within about a second, so they can't be too
832 if (len
!= 1 || ((char *)iov
[0].iov_base
)[0] != 3) {
838 if (abort
->count
== 1)
839 gettimeofday(&abort
->start
, NULL
);
840 else if (abort
->count
== 3) {
842 gettimeofday(&now
, NULL
);
843 /* Kill all Launcher processes with SIGINT, like normal ^C */
844 if (now
.tv_sec
<= abort
->start
.tv_sec
+1)
850 /* This is the routine which handles console output (ie. stdout). */
851 static void console_output(struct virtqueue
*vq
)
853 unsigned int head
, out
, in
;
854 struct iovec iov
[vq
->vring
.num
];
856 /* We usually wait in here, for the Guest to give us something. */
857 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
859 errx(1, "Input buffers in console output queue?");
861 /* writev can return a partial write, so we loop here. */
862 while (!iov_empty(iov
, out
)) {
863 int len
= writev(STDOUT_FILENO
, iov
, out
);
865 err(1, "Write to stdout gave %i", len
);
866 iov_consume(iov
, out
, len
);
870 * We're finished with that buffer: if we're going to sleep,
871 * wait_for_vq_desc() will prod the Guest with an interrupt.
873 add_used(vq
, head
, 0);
879 * Handling output for network is also simple: we get all the output buffers
880 * and write them to /dev/net/tun.
886 static void net_output(struct virtqueue
*vq
)
888 struct net_info
*net_info
= vq
->dev
->priv
;
889 unsigned int head
, out
, in
;
890 struct iovec iov
[vq
->vring
.num
];
892 /* We usually wait in here for the Guest to give us a packet. */
893 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
895 errx(1, "Input buffers in net output queue?");
897 * Send the whole thing through to /dev/net/tun. It expects the exact
898 * same format: what a coincidence!
900 if (writev(net_info
->tunfd
, iov
, out
) < 0)
901 errx(1, "Write to tun failed?");
904 * Done with that one; wait_for_vq_desc() will send the interrupt if
905 * all packets are processed.
907 add_used(vq
, head
, 0);
911 * Handling network input is a bit trickier, because I've tried to optimize it.
913 * First we have a helper routine which tells is if from this file descriptor
914 * (ie. the /dev/net/tun device) will block:
916 static bool will_block(int fd
)
919 struct timeval zero
= { 0, 0 };
922 return select(fd
+1, &fdset
, NULL
, NULL
, &zero
) != 1;
926 * This handles packets coming in from the tun device to our Guest. Like all
927 * service routines, it gets called again as soon as it returns, so you don't
928 * see a while(1) loop here.
930 static void net_input(struct virtqueue
*vq
)
933 unsigned int head
, out
, in
;
934 struct iovec iov
[vq
->vring
.num
];
935 struct net_info
*net_info
= vq
->dev
->priv
;
938 * Get a descriptor to write an incoming packet into. This will also
939 * send an interrupt if they're out of descriptors.
941 head
= wait_for_vq_desc(vq
, iov
, &out
, &in
);
943 errx(1, "Output buffers in net input queue?");
946 * If it looks like we'll block reading from the tun device, send them
949 if (vq
->pending_used
&& will_block(net_info
->tunfd
))
953 * Read in the packet. This is where we normally wait (when there's no
954 * incoming network traffic).
956 len
= readv(net_info
->tunfd
, iov
, in
);
958 err(1, "Failed to read from tun.");
961 * Mark that packet buffer as used, but don't interrupt here. We want
962 * to wait until we've done as much work as we can.
964 add_used(vq
, head
, len
);
968 /* This is the helper to create threads: run the service routine in a loop. */
969 static int do_thread(void *_vq
)
971 struct virtqueue
*vq
= _vq
;
979 * When a child dies, we kill our entire process group with SIGTERM. This
980 * also has the side effect that the shell restores the console for us!
982 static void kill_launcher(int signal
)
987 static void reset_device(struct device
*dev
)
989 struct virtqueue
*vq
;
991 verbose("Resetting device %s\n", dev
->name
);
993 /* Clear any features they've acked. */
994 memset(get_feature_bits(dev
) + dev
->feature_len
, 0, dev
->feature_len
);
996 /* We're going to be explicitly killing threads, so ignore them. */
997 signal(SIGCHLD
, SIG_IGN
);
999 /* Zero out the virtqueues, get rid of their threads */
1000 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
1001 if (vq
->thread
!= (pid_t
)-1) {
1002 kill(vq
->thread
, SIGTERM
);
1003 waitpid(vq
->thread
, NULL
, 0);
1004 vq
->thread
= (pid_t
)-1;
1006 memset(vq
->vring
.desc
, 0,
1007 vring_size(vq
->config
.num
, LGUEST_VRING_ALIGN
));
1008 lg_last_avail(vq
) = 0;
1010 dev
->running
= false;
1012 /* Now we care if threads die. */
1013 signal(SIGCHLD
, (void *)kill_launcher
);
1017 * This actually creates the thread which services the virtqueue for a device.
1019 static void create_thread(struct virtqueue
*vq
)
1022 * Create stack for thread. Since the stack grows upwards, we point
1023 * the stack pointer to the end of this region.
1025 char *stack
= malloc(32768);
1026 unsigned long args
[] = { LHREQ_EVENTFD
,
1027 vq
->config
.pfn
*getpagesize(), 0 };
1029 /* Create a zero-initialized eventfd. */
1030 vq
->eventfd
= eventfd(0, 0);
1031 if (vq
->eventfd
< 0)
1032 err(1, "Creating eventfd");
1033 args
[2] = vq
->eventfd
;
1036 * Attach an eventfd to this virtqueue: it will go off when the Guest
1037 * does an LHCALL_NOTIFY for this vq.
1039 if (write(lguest_fd
, &args
, sizeof(args
)) != 0)
1040 err(1, "Attaching eventfd");
1043 * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so
1044 * we get a signal if it dies.
1046 vq
->thread
= clone(do_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, vq
);
1047 if (vq
->thread
== (pid_t
)-1)
1048 err(1, "Creating clone");
1050 /* We close our local copy now the child has it. */
1054 static void start_device(struct device
*dev
)
1057 struct virtqueue
*vq
;
1059 verbose("Device %s OK: offered", dev
->name
);
1060 for (i
= 0; i
< dev
->feature_len
; i
++)
1061 verbose(" %02x", get_feature_bits(dev
)[i
]);
1062 verbose(", accepted");
1063 for (i
= 0; i
< dev
->feature_len
; i
++)
1064 verbose(" %02x", get_feature_bits(dev
)
1065 [dev
->feature_len
+i
]);
1067 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
1071 dev
->running
= true;
1074 static void cleanup_devices(void)
1078 for (dev
= devices
.dev
; dev
; dev
= dev
->next
)
1081 /* If we saved off the original terminal settings, restore them now. */
1082 if (orig_term
.c_lflag
& (ISIG
|ICANON
|ECHO
))
1083 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
1086 /* When the Guest tells us they updated the status field, we handle it. */
1087 static void update_device_status(struct device
*dev
)
1089 /* A zero status is a reset, otherwise it's a set of flags. */
1090 if (dev
->desc
->status
== 0)
1092 else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
1093 warnx("Device %s configuration FAILED", dev
->name
);
1096 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
1103 * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In
1104 * particular, it's used to notify us of device status changes during boot.
1106 static void handle_output(unsigned long addr
)
1110 /* Check each device. */
1111 for (i
= devices
.dev
; i
; i
= i
->next
) {
1112 struct virtqueue
*vq
;
1115 * Notifications to device descriptors mean they updated the
1118 if (from_guest_phys(addr
) == i
->desc
) {
1119 update_device_status(i
);
1124 * Devices *can* be used before status is set to DRIVER_OK.
1125 * The original plan was that they would never do this: they
1126 * would always finish setting up their status bits before
1127 * actually touching the virtqueues. In practice, we allowed
1128 * them to, and they do (eg. the disk probes for partition
1129 * tables as part of initialization).
1131 * If we see this, we start the device: once it's running, we
1132 * expect the device to catch all the notifications.
1134 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
1135 if (addr
!= vq
->config
.pfn
*getpagesize())
1138 errx(1, "Notification on running %s", i
->name
);
1139 /* This just calls create_thread() for each virtqueue */
1146 * Early console write is done using notify on a nul-terminated string
1147 * in Guest memory. It's also great for hacking debugging messages
1150 if (addr
>= guest_limit
)
1151 errx(1, "Bad NOTIFY %#lx", addr
);
1153 write(STDOUT_FILENO
, from_guest_phys(addr
),
1154 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1160 * All devices need a descriptor so the Guest knows it exists, and a "struct
1161 * device" so the Launcher can keep track of it. We have common helper
1162 * routines to allocate and manage them.
1166 * The layout of the device page is a "struct lguest_device_desc" followed by a
1167 * number of virtqueue descriptors, then two sets of feature bits, then an
1168 * array of configuration bytes. This routine returns the configuration
1171 static u8
*device_config(const struct device
*dev
)
1173 return (void *)(dev
->desc
+ 1)
1174 + dev
->num_vq
* sizeof(struct lguest_vqconfig
)
1175 + dev
->feature_len
* 2;
1179 * This routine allocates a new "struct lguest_device_desc" from descriptor
1180 * table page just above the Guest's normal memory. It returns a pointer to
1183 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1185 struct lguest_device_desc d
= { .type
= type
};
1188 /* Figure out where the next device config is, based on the last one. */
1189 if (devices
.lastdev
)
1190 p
= device_config(devices
.lastdev
)
1191 + devices
.lastdev
->desc
->config_len
;
1193 p
= devices
.descpage
;
1195 /* We only have one page for all the descriptors. */
1196 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1197 errx(1, "Too many devices");
1199 /* p might not be aligned, so we memcpy in. */
1200 return memcpy(p
, &d
, sizeof(d
));
1204 * Each device descriptor is followed by the description of its virtqueues. We
1205 * specify how many descriptors the virtqueue is to have.
1207 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1208 void (*service
)(struct virtqueue
*))
1211 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1214 /* First we need some memory for this virtqueue. */
1215 pages
= (vring_size(num_descs
, LGUEST_VRING_ALIGN
) + getpagesize() - 1)
1217 p
= get_pages(pages
);
1219 /* Initialize the virtqueue */
1221 vq
->last_avail_idx
= 0;
1225 * This is the routine the service thread will run, and its Process ID
1226 * once it's running.
1228 vq
->service
= service
;
1229 vq
->thread
= (pid_t
)-1;
1231 /* Initialize the configuration. */
1232 vq
->config
.num
= num_descs
;
1233 vq
->config
.irq
= devices
.next_irq
++;
1234 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1236 /* Initialize the vring. */
1237 vring_init(&vq
->vring
, num_descs
, p
, LGUEST_VRING_ALIGN
);
1240 * Append virtqueue to this device's descriptor. We use
1241 * device_config() to get the end of the device's current virtqueues;
1242 * we check that we haven't added any config or feature information
1243 * yet, otherwise we'd be overwriting them.
1245 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1246 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1248 dev
->desc
->num_vq
++;
1250 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1253 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1256 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1261 * The first half of the feature bitmask is for us to advertise features. The
1262 * second half is for the Guest to accept features.
1264 static void add_feature(struct device
*dev
, unsigned bit
)
1266 u8
*features
= get_feature_bits(dev
);
1268 /* We can't extend the feature bits once we've added config bytes */
1269 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1270 assert(dev
->desc
->config_len
== 0);
1271 dev
->feature_len
= dev
->desc
->feature_len
= (bit
/CHAR_BIT
) + 1;
1274 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1278 * This routine sets the configuration fields for an existing device's
1279 * descriptor. It only works for the last device, but that's OK because that's
1282 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1284 /* Check we haven't overflowed our single page. */
1285 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1286 errx(1, "Too many devices");
1288 /* Copy in the config information, and store the length. */
1289 memcpy(device_config(dev
), conf
, len
);
1290 dev
->desc
->config_len
= len
;
1292 /* Size must fit in config_len field (8 bits)! */
1293 assert(dev
->desc
->config_len
== len
);
1297 * This routine does all the creation and setup of a new device, including
1298 * calling new_dev_desc() to allocate the descriptor and device memory. We
1299 * don't actually start the service threads until later.
1301 * See what I mean about userspace being boring?
1303 static struct device
*new_device(const char *name
, u16 type
)
1305 struct device
*dev
= malloc(sizeof(*dev
));
1307 /* Now we populate the fields one at a time. */
1308 dev
->desc
= new_dev_desc(type
);
1311 dev
->feature_len
= 0;
1313 dev
->running
= false;
1316 * Append to device list. Prepending to a single-linked list is
1317 * easier, but the user expects the devices to be arranged on the bus
1318 * in command-line order. The first network device on the command line
1319 * is eth0, the first block device /dev/vda, etc.
1321 if (devices
.lastdev
)
1322 devices
.lastdev
->next
= dev
;
1325 devices
.lastdev
= dev
;
1331 * Our first setup routine is the console. It's a fairly simple device, but
1332 * UNIX tty handling makes it uglier than it could be.
1334 static void setup_console(void)
1338 /* If we can save the initial standard input settings... */
1339 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1340 struct termios term
= orig_term
;
1342 * Then we turn off echo, line buffering and ^C etc: We want a
1343 * raw input stream to the Guest.
1345 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1346 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1349 dev
= new_device("console", VIRTIO_ID_CONSOLE
);
1351 /* We store the console state in dev->priv, and initialize it. */
1352 dev
->priv
= malloc(sizeof(struct console_abort
));
1353 ((struct console_abort
*)dev
->priv
)->count
= 0;
1356 * The console needs two virtqueues: the input then the output. When
1357 * they put something the input queue, we make sure we're listening to
1358 * stdin. When they put something in the output queue, we write it to
1361 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_input
);
1362 add_virtqueue(dev
, VIRTQUEUE_NUM
, console_output
);
1364 verbose("device %u: console\n", ++devices
.device_num
);
1369 * Inter-guest networking is an interesting area. Simplest is to have a
1370 * --sharenet=<name> option which opens or creates a named pipe. This can be
1371 * used to send packets to another guest in a 1:1 manner.
1373 * More sopisticated is to use one of the tools developed for project like UML
1376 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1377 * completely generic ("here's my vring, attach to your vring") and would work
1378 * for any traffic. Of course, namespace and permissions issues need to be
1379 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1380 * multiple inter-guest channels behind one interface, although it would
1381 * require some manner of hotplugging new virtio channels.
1383 * Finally, we could implement a virtio network switch in the kernel.
1386 static u32
str2ip(const char *ipaddr
)
1390 if (sscanf(ipaddr
, "%u.%u.%u.%u", &b
[0], &b
[1], &b
[2], &b
[3]) != 4)
1391 errx(1, "Failed to parse IP address '%s'", ipaddr
);
1392 return (b
[0] << 24) | (b
[1] << 16) | (b
[2] << 8) | b
[3];
1395 static void str2mac(const char *macaddr
, unsigned char mac
[6])
1398 if (sscanf(macaddr
, "%02x:%02x:%02x:%02x:%02x:%02x",
1399 &m
[0], &m
[1], &m
[2], &m
[3], &m
[4], &m
[5]) != 6)
1400 errx(1, "Failed to parse mac address '%s'", macaddr
);
1410 * This code is "adapted" from libbridge: it attaches the Host end of the
1411 * network device to the bridge device specified by the command line.
1413 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1414 * dislike bridging), and I just try not to break it.
1416 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1422 errx(1, "must specify bridge name");
1424 ifidx
= if_nametoindex(if_name
);
1426 errx(1, "interface %s does not exist!", if_name
);
1428 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1429 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1430 ifr
.ifr_ifindex
= ifidx
;
1431 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1432 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1436 * This sets up the Host end of the network device with an IP address, brings
1437 * it up so packets will flow, the copies the MAC address into the hwaddr
1440 static void configure_device(int fd
, const char *tapif
, u32 ipaddr
)
1443 struct sockaddr_in sin
;
1445 memset(&ifr
, 0, sizeof(ifr
));
1446 strcpy(ifr
.ifr_name
, tapif
);
1448 /* Don't read these incantations. Just cut & paste them like I did! */
1449 sin
.sin_family
= AF_INET
;
1450 sin
.sin_addr
.s_addr
= htonl(ipaddr
);
1451 memcpy(&ifr
.ifr_addr
, &sin
, sizeof(sin
));
1452 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1453 err(1, "Setting %s interface address", tapif
);
1454 ifr
.ifr_flags
= IFF_UP
;
1455 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1456 err(1, "Bringing interface %s up", tapif
);
1459 static int get_tun_device(char tapif
[IFNAMSIZ
])
1464 /* Start with this zeroed. Messy but sure. */
1465 memset(&ifr
, 0, sizeof(ifr
));
1468 * We open the /dev/net/tun device and tell it we want a tap device. A
1469 * tap device is like a tun device, only somehow different. To tell
1470 * the truth, I completely blundered my way through this code, but it
1473 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1474 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
;
1475 strcpy(ifr
.ifr_name
, "tap%d");
1476 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1477 err(1, "configuring /dev/net/tun");
1479 if (ioctl(netfd
, TUNSETOFFLOAD
,
1480 TUN_F_CSUM
|TUN_F_TSO4
|TUN_F_TSO6
|TUN_F_TSO_ECN
) != 0)
1481 err(1, "Could not set features for tun device");
1484 * We don't need checksums calculated for packets coming in this
1487 ioctl(netfd
, TUNSETNOCSUM
, 1);
1489 memcpy(tapif
, ifr
.ifr_name
, IFNAMSIZ
);
1494 * Our network is a Host<->Guest network. This can either use bridging or
1495 * routing, but the principle is the same: it uses the "tun" device to inject
1496 * packets into the Host as if they came in from a normal network card. We
1497 * just shunt packets between the Guest and the tun device.
1499 static void setup_tun_net(char *arg
)
1502 struct net_info
*net_info
= malloc(sizeof(*net_info
));
1504 u32 ip
= INADDR_ANY
;
1505 bool bridging
= false;
1506 char tapif
[IFNAMSIZ
], *p
;
1507 struct virtio_net_config conf
;
1509 net_info
->tunfd
= get_tun_device(tapif
);
1511 /* First we create a new network device. */
1512 dev
= new_device("net", VIRTIO_ID_NET
);
1513 dev
->priv
= net_info
;
1515 /* Network devices need a recv and a send queue, just like console. */
1516 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_input
);
1517 add_virtqueue(dev
, VIRTQUEUE_NUM
, net_output
);
1520 * We need a socket to perform the magic network ioctls to bring up the
1521 * tap interface, connect to the bridge etc. Any socket will do!
1523 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1525 err(1, "opening IP socket");
1527 /* If the command line was --tunnet=bridge:<name> do bridging. */
1528 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1529 arg
+= strlen(BRIDGE_PFX
);
1533 /* A mac address may follow the bridge name or IP address */
1534 p
= strchr(arg
, ':');
1536 str2mac(p
+1, conf
.mac
);
1537 add_feature(dev
, VIRTIO_NET_F_MAC
);
1541 /* arg is now either an IP address or a bridge name */
1543 add_to_bridge(ipfd
, tapif
, arg
);
1547 /* Set up the tun device. */
1548 configure_device(ipfd
, tapif
, ip
);
1550 /* Expect Guest to handle everything except UFO */
1551 add_feature(dev
, VIRTIO_NET_F_CSUM
);
1552 add_feature(dev
, VIRTIO_NET_F_GUEST_CSUM
);
1553 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO4
);
1554 add_feature(dev
, VIRTIO_NET_F_GUEST_TSO6
);
1555 add_feature(dev
, VIRTIO_NET_F_GUEST_ECN
);
1556 add_feature(dev
, VIRTIO_NET_F_HOST_TSO4
);
1557 add_feature(dev
, VIRTIO_NET_F_HOST_TSO6
);
1558 add_feature(dev
, VIRTIO_NET_F_HOST_ECN
);
1559 /* We handle indirect ring entries */
1560 add_feature(dev
, VIRTIO_RING_F_INDIRECT_DESC
);
1561 set_config(dev
, sizeof(conf
), &conf
);
1563 /* We don't need the socket any more; setup is done. */
1566 devices
.device_num
++;
1569 verbose("device %u: tun %s attached to bridge: %s\n",
1570 devices
.device_num
, tapif
, arg
);
1572 verbose("device %u: tun %s: %s\n",
1573 devices
.device_num
, tapif
, arg
);
1577 /* This hangs off device->priv. */
1579 /* The size of the file. */
1582 /* The file descriptor for the file. */
1590 * The disk only has one virtqueue, so it only has one thread. It is really
1591 * simple: the Guest asks for a block number and we read or write that position
1594 * Before we serviced each virtqueue in a separate thread, that was unacceptably
1595 * slow: the Guest waits until the read is finished before running anything
1596 * else, even if it could have been doing useful work.
1598 * We could have used async I/O, except it's reputed to suck so hard that
1599 * characters actually go missing from your code when you try to use it.
1601 static void blk_request(struct virtqueue
*vq
)
1603 struct vblk_info
*vblk
= vq
->dev
->priv
;
1604 unsigned int head
, out_num
, in_num
, wlen
;
1607 struct virtio_blk_outhdr
*out
;
1608 struct iovec iov
[vq
->vring
.num
];
1612 * Get the next request, where we normally wait. It triggers the
1613 * interrupt to acknowledge previously serviced requests (if any).
1615 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1618 * Every block request should contain at least one output buffer
1619 * (detailing the location on disk and the type of request) and one
1620 * input buffer (to hold the result).
1622 if (out_num
== 0 || in_num
== 0)
1623 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1624 head
, out_num
, in_num
);
1626 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1627 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1629 * For historical reasons, block operations are expressed in 512 byte
1632 off
= out
->sector
* 512;
1635 * In general the virtio block driver is allowed to try SCSI commands.
1636 * It'd be nice if we supported eject, for example, but we don't.
1638 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1639 fprintf(stderr
, "Scsi commands unsupported\n");
1640 *in
= VIRTIO_BLK_S_UNSUPP
;
1642 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1646 * Move to the right location in the block file. This can fail
1647 * if they try to write past end.
1649 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1650 err(1, "Bad seek to sector %llu", out
->sector
);
1652 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1653 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1656 * Grr... Now we know how long the descriptor they sent was, we
1657 * make sure they didn't try to write over the end of the block
1658 * file (possibly extending it).
1660 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1661 /* Trim it back to the correct length */
1662 ftruncate64(vblk
->fd
, vblk
->len
);
1663 /* Die, bad Guest, die. */
1664 errx(1, "Write past end %llu+%u", off
, ret
);
1668 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1669 } else if (out
->type
& VIRTIO_BLK_T_FLUSH
) {
1671 ret
= fdatasync(vblk
->fd
);
1672 verbose("FLUSH fdatasync: %i\n", ret
);
1674 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1679 * Move to the right location in the block file. This can fail
1680 * if they try to read past end.
1682 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1683 err(1, "Bad seek to sector %llu", out
->sector
);
1685 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1686 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1688 wlen
= sizeof(*in
) + ret
;
1689 *in
= VIRTIO_BLK_S_OK
;
1692 *in
= VIRTIO_BLK_S_IOERR
;
1696 /* Finished that request. */
1697 add_used(vq
, head
, wlen
);
1700 /*L:198 This actually sets up a virtual block device. */
1701 static void setup_block_file(const char *filename
)
1704 struct vblk_info
*vblk
;
1705 struct virtio_blk_config conf
;
1707 /* Creat the device. */
1708 dev
= new_device("block", VIRTIO_ID_BLOCK
);
1710 /* The device has one virtqueue, where the Guest places requests. */
1711 add_virtqueue(dev
, VIRTQUEUE_NUM
, blk_request
);
1713 /* Allocate the room for our own bookkeeping */
1714 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1716 /* First we open the file and store the length. */
1717 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1718 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1720 /* We support FLUSH. */
1721 add_feature(dev
, VIRTIO_BLK_F_FLUSH
);
1723 /* Tell Guest how many sectors this device has. */
1724 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1727 * Tell Guest not to put in too many descriptors at once: two are used
1728 * for the in and out elements.
1730 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1731 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1733 /* Don't try to put whole struct: we have 8 bit limit. */
1734 set_config(dev
, offsetof(struct virtio_blk_config
, geometry
), &conf
);
1736 verbose("device %u: virtblock %llu sectors\n",
1737 ++devices
.device_num
, le64_to_cpu(conf
.capacity
));
1741 * Our random number generator device reads from /dev/random into the Guest's
1742 * input buffers. The usual case is that the Guest doesn't want random numbers
1743 * and so has no buffers although /dev/random is still readable, whereas
1744 * console is the reverse.
1746 * The same logic applies, however.
1752 static void rng_input(struct virtqueue
*vq
)
1755 unsigned int head
, in_num
, out_num
, totlen
= 0;
1756 struct rng_info
*rng_info
= vq
->dev
->priv
;
1757 struct iovec iov
[vq
->vring
.num
];
1759 /* First we need a buffer from the Guests's virtqueue. */
1760 head
= wait_for_vq_desc(vq
, iov
, &out_num
, &in_num
);
1762 errx(1, "Output buffers in rng?");
1765 * Just like the console write, we loop to cover the whole iovec.
1766 * In this case, short reads actually happen quite a bit.
1768 while (!iov_empty(iov
, in_num
)) {
1769 len
= readv(rng_info
->rfd
, iov
, in_num
);
1771 err(1, "Read from /dev/random gave %i", len
);
1772 iov_consume(iov
, in_num
, len
);
1776 /* Tell the Guest about the new input. */
1777 add_used(vq
, head
, totlen
);
1781 * This creates a "hardware" random number device for the Guest.
1783 static void setup_rng(void)
1786 struct rng_info
*rng_info
= malloc(sizeof(*rng_info
));
1788 /* Our device's privat info simply contains the /dev/random fd. */
1789 rng_info
->rfd
= open_or_die("/dev/random", O_RDONLY
);
1791 /* Create the new device. */
1792 dev
= new_device("rng", VIRTIO_ID_RNG
);
1793 dev
->priv
= rng_info
;
1795 /* The device has one virtqueue, where the Guest places inbufs. */
1796 add_virtqueue(dev
, VIRTQUEUE_NUM
, rng_input
);
1798 verbose("device %u: rng\n", devices
.device_num
++);
1800 /* That's the end of device setup. */
1802 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1803 static void __attribute__((noreturn
)) restart_guest(void)
1808 * Since we don't track all open fds, we simply close everything beyond
1811 for (i
= 3; i
< FD_SETSIZE
; i
++)
1814 /* Reset all the devices (kills all threads). */
1817 execv(main_args
[0], main_args
);
1818 err(1, "Could not exec %s", main_args
[0]);
1822 * Finally we reach the core of the Launcher which runs the Guest, serves
1823 * its input and output, and finally, lays it to rest.
1825 static void __attribute__((noreturn
)) run_guest(void)
1828 unsigned long notify_addr
;
1831 /* We read from the /dev/lguest device to run the Guest. */
1832 readval
= pread(lguest_fd
, ¬ify_addr
,
1833 sizeof(notify_addr
), cpu_id
);
1835 /* One unsigned long means the Guest did HCALL_NOTIFY */
1836 if (readval
== sizeof(notify_addr
)) {
1837 verbose("Notify on address %#lx\n", notify_addr
);
1838 handle_output(notify_addr
);
1839 /* ENOENT means the Guest died. Reading tells us why. */
1840 } else if (errno
== ENOENT
) {
1841 char reason
[1024] = { 0 };
1842 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1843 errx(1, "%s", reason
);
1844 /* ERESTART means that we need to reboot the guest */
1845 } else if (errno
== ERESTART
) {
1847 /* Anything else means a bug or incompatible change. */
1849 err(1, "Running guest failed");
1853 * This is the end of the Launcher. The good news: we are over halfway
1854 * through! The bad news: the most fiendish part of the code still lies ahead
1857 * Are you ready? Take a deep breath and join me in the core of the Host, in
1861 static struct option opts
[] = {
1862 { "verbose", 0, NULL
, 'v' },
1863 { "tunnet", 1, NULL
, 't' },
1864 { "block", 1, NULL
, 'b' },
1865 { "rng", 0, NULL
, 'r' },
1866 { "initrd", 1, NULL
, 'i' },
1867 { "username", 1, NULL
, 'u' },
1868 { "chroot", 1, NULL
, 'c' },
1871 static void usage(void)
1873 errx(1, "Usage: lguest [--verbose] "
1874 "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n"
1875 "|--block=<filename>|--initrd=<filename>]...\n"
1876 "<mem-in-mb> vmlinux [args...]");
1879 /*L:105 The main routine is where the real work begins: */
1880 int main(int argc
, char *argv
[])
1882 /* Memory, code startpoint and size of the (optional) initrd. */
1883 unsigned long mem
= 0, start
, initrd_size
= 0;
1884 /* Two temporaries. */
1886 /* The boot information for the Guest. */
1887 struct boot_params
*boot
;
1888 /* If they specify an initrd file to load. */
1889 const char *initrd_name
= NULL
;
1891 /* Password structure for initgroups/setres[gu]id */
1892 struct passwd
*user_details
= NULL
;
1894 /* Directory to chroot to */
1895 char *chroot_path
= NULL
;
1897 /* Save the args: we "reboot" by execing ourselves again. */
1901 * First we initialize the device list. We keep a pointer to the last
1902 * device, and the next interrupt number to use for devices (1:
1903 * remember that 0 is used by the timer).
1905 devices
.lastdev
= NULL
;
1906 devices
.next_irq
= 1;
1908 /* We're CPU 0. In fact, that's the only CPU possible right now. */
1912 * We need to know how much memory so we can set up the device
1913 * descriptor and memory pages for the devices as we parse the command
1914 * line. So we quickly look through the arguments to find the amount
1917 for (i
= 1; i
< argc
; i
++) {
1918 if (argv
[i
][0] != '-') {
1919 mem
= atoi(argv
[i
]) * 1024 * 1024;
1921 * We start by mapping anonymous pages over all of
1922 * guest-physical memory range. This fills it with 0,
1923 * and ensures that the Guest won't be killed when it
1924 * tries to access it.
1926 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1929 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1930 devices
.descpage
= get_pages(1);
1935 /* The options are fairly straight-forward */
1936 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1942 setup_tun_net(optarg
);
1945 setup_block_file(optarg
);
1951 initrd_name
= optarg
;
1954 user_details
= getpwnam(optarg
);
1956 err(1, "getpwnam failed, incorrect username?");
1959 chroot_path
= optarg
;
1962 warnx("Unknown argument %s", argv
[optind
]);
1967 * After the other arguments we expect memory and kernel image name,
1968 * followed by command line arguments for the kernel.
1970 if (optind
+ 2 > argc
)
1973 verbose("Guest base is at %p\n", guest_base
);
1975 /* We always have a console device */
1978 /* Now we load the kernel */
1979 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1981 /* Boot information is stashed at physical address 0 */
1982 boot
= from_guest_phys(0);
1984 /* Map the initrd image if requested (at top of physical memory) */
1986 initrd_size
= load_initrd(initrd_name
, mem
);
1988 * These are the location in the Linux boot header where the
1989 * start and size of the initrd are expected to be found.
1991 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1992 boot
->hdr
.ramdisk_size
= initrd_size
;
1993 /* The bootloader type 0xFF means "unknown"; that's OK. */
1994 boot
->hdr
.type_of_loader
= 0xFF;
1998 * The Linux boot header contains an "E820" memory map: ours is a
1999 * simple, single region.
2001 boot
->e820_entries
= 1;
2002 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
2004 * The boot header contains a command line pointer: we put the command
2005 * line after the boot header.
2007 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
2008 /* We use a simple helper to copy the arguments separated by spaces. */
2009 concat((char *)(boot
+ 1), argv
+optind
+2);
2011 /* Boot protocol version: 2.07 supports the fields for lguest. */
2012 boot
->hdr
.version
= 0x207;
2014 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
2015 boot
->hdr
.hardware_subarch
= 1;
2017 /* Tell the entry path not to try to reload segment registers. */
2018 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
2021 * We tell the kernel to initialize the Guest: this returns the open
2022 * /dev/lguest file descriptor.
2026 /* Ensure that we terminate if a device-servicing child dies. */
2027 signal(SIGCHLD
, kill_launcher
);
2029 /* If we exit via err(), this kills all the threads, restores tty. */
2030 atexit(cleanup_devices
);
2032 /* If requested, chroot to a directory */
2034 if (chroot(chroot_path
) != 0)
2035 err(1, "chroot(\"%s\") failed", chroot_path
);
2037 if (chdir("/") != 0)
2038 err(1, "chdir(\"/\") failed");
2040 verbose("chroot done\n");
2043 /* If requested, drop privileges */
2048 u
= user_details
->pw_uid
;
2049 g
= user_details
->pw_gid
;
2051 if (initgroups(user_details
->pw_name
, g
) != 0)
2052 err(1, "initgroups failed");
2054 if (setresgid(g
, g
, g
) != 0)
2055 err(1, "setresgid failed");
2057 if (setresuid(u
, u
, u
) != 0)
2058 err(1, "setresuid failed");
2060 verbose("Dropping privileges completed\n");
2063 /* Finally, run the Guest. This doesn't return. */
2069 * Mastery is done: you now know everything I do.
2071 * But surely you have seen code, features and bugs in your wanderings which
2072 * you now yearn to attack? That is the real game, and I look forward to you
2073 * patching and forking lguest into the Your-Name-Here-visor.
2075 * Farewell, and good coding!