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_ring.h"
45 #include "asm-x86/bootparam.h"
46 /*L:110 We can ignore the 39 include files we need for this program, but I do
47 * want to draw attention to the use of kernel-style types.
49 * As Linus said, "C is a Spartan language, and so should your naming be." I
50 * like these abbreviations, so we define them here. Note that u64 is always
51 * unsigned long long, which works on all Linux systems: this means that we can
52 * use %llu in printf for any u64. */
53 typedef unsigned long long u64
;
59 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
61 #define BRIDGE_PFX "bridge:"
63 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
65 /* We can have up to 256 pages for devices. */
66 #define DEVICE_PAGES 256
67 /* This will occupy 2 pages: it must be a power of 2. */
68 #define VIRTQUEUE_NUM 128
70 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
71 * this, and although I wouldn't recommend it, it works quite nicely here. */
73 #define verbose(args...) \
74 do { if (verbose) printf(args); } while(0)
77 /* The pipe to send commands to the waker process */
79 /* The pointer to the start of guest memory. */
80 static void *guest_base
;
81 /* The maximum guest physical address allowed, and maximum possible. */
82 static unsigned long guest_limit
, guest_max
;
84 /* a per-cpu variable indicating whose vcpu is currently running */
85 static unsigned int __thread cpu_id
;
87 /* This is our list of devices. */
90 /* Summary information about the devices in our list: ready to pass to
91 * select() to ask which need servicing.*/
95 /* Counter to assign interrupt numbers. */
96 unsigned int next_irq
;
98 /* Counter to print out convenient device numbers. */
99 unsigned int device_num
;
101 /* The descriptor page for the devices. */
104 /* A single linked list of devices. */
106 /* And a pointer to the last device for easy append and also for
107 * configuration appending. */
108 struct device
*lastdev
;
111 /* The list of Guest devices, based on command line arguments. */
112 static struct device_list devices
;
114 /* The device structure describes a single device. */
117 /* The linked-list pointer. */
120 /* The this device's descriptor, as mapped into the Guest. */
121 struct lguest_device_desc
*desc
;
123 /* The name of this device, for --verbose. */
126 /* If handle_input is set, it wants to be called when this file
127 * descriptor is ready. */
129 bool (*handle_input
)(int fd
, struct device
*me
);
131 /* Any queues attached to this device */
132 struct virtqueue
*vq
;
134 /* Handle status being finalized (ie. feature bits stable). */
135 void (*ready
)(struct device
*me
);
137 /* Device-specific data. */
141 /* 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 /* The routine to call when the Guest pings us. */
159 void (*handle_output
)(int fd
, struct virtqueue
*me
);
162 /* Remember the arguments to the program so we can "reboot" */
163 static char **main_args
;
165 /* Since guest is UP and we don't run at the same time, we don't need barriers.
166 * But I include them in the code in case others copy it. */
169 /* Convert an iovec element to the given type.
171 * This is a fairly ugly trick: we need to know the size of the type and
172 * alignment requirement to check the pointer is kosher. It's also nice to
173 * have the name of the type in case we report failure.
175 * Typing those three things all the time is cumbersome and error prone, so we
176 * have a macro which sets them all up and passes to the real function. */
177 #define convert(iov, type) \
178 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
180 static void *_convert(struct iovec
*iov
, size_t size
, size_t align
,
183 if (iov
->iov_len
!= size
)
184 errx(1, "Bad iovec size %zu for %s", iov
->iov_len
, name
);
185 if ((unsigned long)iov
->iov_base
% align
!= 0)
186 errx(1, "Bad alignment %p for %s", iov
->iov_base
, name
);
187 return iov
->iov_base
;
190 /* The virtio configuration space is defined to be little-endian. x86 is
191 * little-endian too, but it's nice to be explicit so we have these helpers. */
192 #define cpu_to_le16(v16) (v16)
193 #define cpu_to_le32(v32) (v32)
194 #define cpu_to_le64(v64) (v64)
195 #define le16_to_cpu(v16) (v16)
196 #define le32_to_cpu(v32) (v32)
197 #define le64_to_cpu(v64) (v64)
199 /* The device virtqueue descriptors are followed by feature bitmasks. */
200 static u8
*get_feature_bits(struct device
*dev
)
202 return (u8
*)(dev
->desc
+ 1)
203 + dev
->desc
->num_vq
* sizeof(struct lguest_vqconfig
);
206 /*L:100 The Launcher code itself takes us out into userspace, that scary place
207 * where pointers run wild and free! Unfortunately, like most userspace
208 * programs, it's quite boring (which is why everyone likes to hack on the
209 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
210 * will get you through this section. Or, maybe not.
212 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
213 * memory and stores it in "guest_base". In other words, Guest physical ==
214 * Launcher virtual with an offset.
216 * This can be tough to get your head around, but usually it just means that we
217 * use these trivial conversion functions when the Guest gives us it's
218 * "physical" addresses: */
219 static void *from_guest_phys(unsigned long addr
)
221 return guest_base
+ addr
;
224 static unsigned long to_guest_phys(const void *addr
)
226 return (addr
- guest_base
);
230 * Loading the Kernel.
232 * We start with couple of simple helper routines. open_or_die() avoids
233 * error-checking code cluttering the callers: */
234 static int open_or_die(const char *name
, int flags
)
236 int fd
= open(name
, flags
);
238 err(1, "Failed to open %s", name
);
242 /* map_zeroed_pages() takes a number of pages. */
243 static void *map_zeroed_pages(unsigned int num
)
245 int fd
= open_or_die("/dev/zero", O_RDONLY
);
248 /* We use a private mapping (ie. if we write to the page, it will be
250 addr
= mmap(NULL
, getpagesize() * num
,
251 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
252 if (addr
== MAP_FAILED
)
253 err(1, "Mmaping %u pages of /dev/zero", num
);
258 /* Get some more pages for a device. */
259 static void *get_pages(unsigned int num
)
261 void *addr
= from_guest_phys(guest_limit
);
263 guest_limit
+= num
* getpagesize();
264 if (guest_limit
> guest_max
)
265 errx(1, "Not enough memory for devices");
269 /* This routine is used to load the kernel or initrd. It tries mmap, but if
270 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
271 * it falls back to reading the memory in. */
272 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
276 /* We map writable even though for some segments are marked read-only.
277 * The kernel really wants to be writable: it patches its own
280 * MAP_PRIVATE means that the page won't be copied until a write is
281 * done to it. This allows us to share untouched memory between
283 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
284 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
287 /* pread does a seek and a read in one shot: saves a few lines. */
288 r
= pread(fd
, addr
, len
, offset
);
290 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
293 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
294 * the Guest memory. ELF = Embedded Linking Format, which is the format used
295 * by all modern binaries on Linux including the kernel.
297 * The ELF headers give *two* addresses: a physical address, and a virtual
298 * address. We use the physical address; the Guest will map itself to the
301 * We return the starting address. */
302 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
)
304 Elf32_Phdr phdr
[ehdr
->e_phnum
];
307 /* Sanity checks on the main ELF header: an x86 executable with a
308 * reasonable number of correctly-sized program headers. */
309 if (ehdr
->e_type
!= ET_EXEC
310 || ehdr
->e_machine
!= EM_386
311 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
312 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
313 errx(1, "Malformed elf header");
315 /* An ELF executable contains an ELF header and a number of "program"
316 * headers which indicate which parts ("segments") of the program to
319 /* We read in all the program headers at once: */
320 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
321 err(1, "Seeking to program headers");
322 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
323 err(1, "Reading program headers");
325 /* Try all the headers: there are usually only three. A read-only one,
326 * a read-write one, and a "note" section which we don't load. */
327 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
328 /* If this isn't a loadable segment, we ignore it */
329 if (phdr
[i
].p_type
!= PT_LOAD
)
332 verbose("Section %i: size %i addr %p\n",
333 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
335 /* We map this section of the file at its physical address. */
336 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
337 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
340 /* The entry point is given in the ELF header. */
341 return ehdr
->e_entry
;
344 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
345 * supposed to jump into it and it will unpack itself. We used to have to
346 * perform some hairy magic because the unpacking code scared me.
348 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
349 * a small patch to jump over the tricky bits in the Guest, so now we just read
350 * the funky header so we know where in the file to load, and away we go! */
351 static unsigned long load_bzimage(int fd
)
353 struct boot_params boot
;
355 /* Modern bzImages get loaded at 1M. */
356 void *p
= from_guest_phys(0x100000);
358 /* Go back to the start of the file and read the header. It should be
359 * a Linux boot header (see Documentation/i386/boot.txt) */
360 lseek(fd
, 0, SEEK_SET
);
361 read(fd
, &boot
, sizeof(boot
));
363 /* Inside the setup_hdr, we expect the magic "HdrS" */
364 if (memcmp(&boot
.hdr
.header
, "HdrS", 4) != 0)
365 errx(1, "This doesn't look like a bzImage to me");
367 /* Skip over the extra sectors of the header. */
368 lseek(fd
, (boot
.hdr
.setup_sects
+1) * 512, SEEK_SET
);
370 /* Now read everything into memory. in nice big chunks. */
371 while ((r
= read(fd
, p
, 65536)) > 0)
374 /* Finally, code32_start tells us where to enter the kernel. */
375 return boot
.hdr
.code32_start
;
378 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
379 * come wrapped up in the self-decompressing "bzImage" format. With a little
380 * work, we can load those, too. */
381 static unsigned long load_kernel(int fd
)
385 /* Read in the first few bytes. */
386 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
387 err(1, "Reading kernel");
389 /* If it's an ELF file, it starts with "\177ELF" */
390 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
391 return map_elf(fd
, &hdr
);
393 /* Otherwise we assume it's a bzImage, and try to load it. */
394 return load_bzimage(fd
);
397 /* This is a trivial little helper to align pages. Andi Kleen hated it because
398 * it calls getpagesize() twice: "it's dumb code."
400 * Kernel guys get really het up about optimization, even when it's not
401 * necessary. I leave this code as a reaction against that. */
402 static inline unsigned long page_align(unsigned long addr
)
404 /* Add upwards and truncate downwards. */
405 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
408 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
409 * the kernel which the kernel can use to boot from without needing any
410 * drivers. Most distributions now use this as standard: the initrd contains
411 * the code to load the appropriate driver modules for the current machine.
413 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
414 * kernels. He sent me this (and tells me when I break it). */
415 static unsigned long load_initrd(const char *name
, unsigned long mem
)
421 ifd
= open_or_die(name
, O_RDONLY
);
422 /* fstat() is needed to get the file size. */
423 if (fstat(ifd
, &st
) < 0)
424 err(1, "fstat() on initrd '%s'", name
);
426 /* We map the initrd at the top of memory, but mmap wants it to be
427 * page-aligned, so we round the size up for that. */
428 len
= page_align(st
.st_size
);
429 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
430 /* Once a file is mapped, you can close the file descriptor. It's a
431 * little odd, but quite useful. */
433 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
435 /* We return the initrd size. */
439 /* Once we know how much memory we have we can construct simple linear page
440 * tables which set virtual == physical which will get the Guest far enough
441 * into the boot to create its own.
443 * We lay them out of the way, just below the initrd (which is why we need to
444 * know its size here). */
445 static unsigned long setup_pagetables(unsigned long mem
,
446 unsigned long initrd_size
)
448 unsigned long *pgdir
, *linear
;
449 unsigned int mapped_pages
, i
, linear_pages
;
450 unsigned int ptes_per_page
= getpagesize()/sizeof(void *);
452 mapped_pages
= mem
/getpagesize();
454 /* Each PTE page can map ptes_per_page pages: how many do we need? */
455 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
457 /* We put the toplevel page directory page at the top of memory. */
458 pgdir
= from_guest_phys(mem
) - initrd_size
- getpagesize();
460 /* Now we use the next linear_pages pages as pte pages */
461 linear
= (void *)pgdir
- linear_pages
*getpagesize();
463 /* Linear mapping is easy: put every page's address into the mapping in
464 * order. PAGE_PRESENT contains the flags Present, Writable and
466 for (i
= 0; i
< mapped_pages
; i
++)
467 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
469 /* The top level points to the linear page table pages above. */
470 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
471 pgdir
[i
/ptes_per_page
]
472 = ((to_guest_phys(linear
) + i
*sizeof(void *))
476 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
477 mapped_pages
, linear_pages
, to_guest_phys(linear
));
479 /* We return the top level (guest-physical) address: the kernel needs
480 * to know where it is. */
481 return to_guest_phys(pgdir
);
485 /* Simple routine to roll all the commandline arguments together with spaces
487 static void concat(char *dst
, char *args
[])
489 unsigned int i
, len
= 0;
491 for (i
= 0; args
[i
]; i
++) {
493 strcat(dst
+len
, " ");
496 strcpy(dst
+len
, args
[i
]);
497 len
+= strlen(args
[i
]);
499 /* In case it's empty. */
503 /*L:185 This is where we actually tell the kernel to initialize the Guest. We
504 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
505 * the base of Guest "physical" memory, the top physical page to allow, the
506 * top level pagetable and the entry point for the Guest. */
507 static int tell_kernel(unsigned long pgdir
, unsigned long start
)
509 unsigned long args
[] = { LHREQ_INITIALIZE
,
510 (unsigned long)guest_base
,
511 guest_limit
/ getpagesize(), pgdir
, start
};
514 verbose("Guest: %p - %p (%#lx)\n",
515 guest_base
, guest_base
+ guest_limit
, guest_limit
);
516 fd
= open_or_die("/dev/lguest", O_RDWR
);
517 if (write(fd
, args
, sizeof(args
)) < 0)
518 err(1, "Writing to /dev/lguest");
520 /* We return the /dev/lguest file descriptor to control this Guest */
525 static void add_device_fd(int fd
)
527 FD_SET(fd
, &devices
.infds
);
528 if (fd
> devices
.max_infd
)
529 devices
.max_infd
= fd
;
535 * With console, block and network devices, we can have lots of input which we
536 * need to process. We could try to tell the kernel what file descriptors to
537 * watch, but handing a file descriptor mask through to the kernel is fairly
540 * Instead, we fork off a process which watches the file descriptors and writes
541 * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host
542 * stop running the Guest. This causes the Launcher to return from the
543 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
544 * the LHREQ_BREAK and wake us up again.
546 * This, of course, is merely a different *kind* of icky.
548 static void wake_parent(int pipefd
, int lguest_fd
)
550 /* Add the pipe from the Launcher to the fdset in the device_list, so
551 * we watch it, too. */
552 add_device_fd(pipefd
);
555 fd_set rfds
= devices
.infds
;
556 unsigned long args
[] = { LHREQ_BREAK
, 1 };
558 /* Wait until input is ready from one of the devices. */
559 select(devices
.max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
560 /* Is it a message from the Launcher? */
561 if (FD_ISSET(pipefd
, &rfds
)) {
563 /* If read() returns 0, it means the Launcher has
564 * exited. We silently follow. */
565 if (read(pipefd
, &fd
, sizeof(fd
)) == 0)
567 /* Otherwise it's telling us to change what file
568 * descriptors we're to listen to. Positive means
569 * listen to a new one, negative means stop
572 FD_SET(fd
, &devices
.infds
);
574 FD_CLR(-fd
- 1, &devices
.infds
);
575 } else /* Send LHREQ_BREAK command. */
576 pwrite(lguest_fd
, args
, sizeof(args
), cpu_id
);
580 /* This routine just sets up a pipe to the Waker process. */
581 static int setup_waker(int lguest_fd
)
583 int pipefd
[2], child
;
585 /* We create a pipe to talk to the Waker, and also so it knows when the
586 * Launcher dies (and closes pipe). */
593 /* We are the Waker: close the "writing" end of our copy of the
594 * pipe and start waiting for input. */
596 wake_parent(pipefd
[0], lguest_fd
);
598 /* Close the reading end of our copy of the pipe. */
601 /* Here is the fd used to talk to the waker. */
608 * When the Guest gives us a buffer, it sends an array of addresses and sizes.
609 * We need to make sure it's not trying to reach into the Launcher itself, so
610 * we have a convenient routine which checks it and exits with an error message
611 * if something funny is going on:
613 static void *_check_pointer(unsigned long addr
, unsigned int size
,
616 /* We have to separately check addr and addr+size, because size could
617 * be huge and addr + size might wrap around. */
618 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
619 errx(1, "%s:%i: Invalid address %#lx", __FILE__
, line
, addr
);
620 /* We return a pointer for the caller's convenience, now we know it's
622 return from_guest_phys(addr
);
624 /* A macro which transparently hands the line number to the real function. */
625 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
627 /* Each buffer in the virtqueues is actually a chain of descriptors. This
628 * function returns the next descriptor in the chain, or vq->vring.num if we're
630 static unsigned next_desc(struct virtqueue
*vq
, unsigned int i
)
634 /* If this descriptor says it doesn't chain, we're done. */
635 if (!(vq
->vring
.desc
[i
].flags
& VRING_DESC_F_NEXT
))
636 return vq
->vring
.num
;
638 /* Check they're not leading us off end of descriptors. */
639 next
= vq
->vring
.desc
[i
].next
;
640 /* Make sure compiler knows to grab that: we don't want it changing! */
643 if (next
>= vq
->vring
.num
)
644 errx(1, "Desc next is %u", next
);
649 /* This looks in the virtqueue and for the first available buffer, and converts
650 * it to an iovec for convenient access. Since descriptors consist of some
651 * number of output then some number of input descriptors, it's actually two
652 * iovecs, but we pack them into one and note how many of each there were.
654 * This function returns the descriptor number found, or vq->vring.num (which
655 * is never a valid descriptor number) if none was found. */
656 static unsigned get_vq_desc(struct virtqueue
*vq
,
658 unsigned int *out_num
, unsigned int *in_num
)
660 unsigned int i
, head
;
662 /* Check it isn't doing very strange things with descriptor numbers. */
663 if ((u16
)(vq
->vring
.avail
->idx
- vq
->last_avail_idx
) > vq
->vring
.num
)
664 errx(1, "Guest moved used index from %u to %u",
665 vq
->last_avail_idx
, vq
->vring
.avail
->idx
);
667 /* If there's nothing new since last we looked, return invalid. */
668 if (vq
->vring
.avail
->idx
== vq
->last_avail_idx
)
669 return vq
->vring
.num
;
671 /* Grab the next descriptor number they're advertising, and increment
672 * the index we've seen. */
673 head
= vq
->vring
.avail
->ring
[vq
->last_avail_idx
++ % vq
->vring
.num
];
675 /* If their number is silly, that's a fatal mistake. */
676 if (head
>= vq
->vring
.num
)
677 errx(1, "Guest says index %u is available", head
);
679 /* When we start there are none of either input nor output. */
680 *out_num
= *in_num
= 0;
684 /* Grab the first descriptor, and check it's OK. */
685 iov
[*out_num
+ *in_num
].iov_len
= vq
->vring
.desc
[i
].len
;
686 iov
[*out_num
+ *in_num
].iov_base
687 = check_pointer(vq
->vring
.desc
[i
].addr
,
688 vq
->vring
.desc
[i
].len
);
689 /* If this is an input descriptor, increment that count. */
690 if (vq
->vring
.desc
[i
].flags
& VRING_DESC_F_WRITE
)
693 /* If it's an output descriptor, they're all supposed
694 * to come before any input descriptors. */
696 errx(1, "Descriptor has out after in");
700 /* If we've got too many, that implies a descriptor loop. */
701 if (*out_num
+ *in_num
> vq
->vring
.num
)
702 errx(1, "Looped descriptor");
703 } while ((i
= next_desc(vq
, i
)) != vq
->vring
.num
);
708 /* After we've used one of their buffers, we tell them about it. We'll then
709 * want to send them an interrupt, using trigger_irq(). */
710 static void add_used(struct virtqueue
*vq
, unsigned int head
, int len
)
712 struct vring_used_elem
*used
;
714 /* The virtqueue contains a ring of used buffers. Get a pointer to the
715 * next entry in that used ring. */
716 used
= &vq
->vring
.used
->ring
[vq
->vring
.used
->idx
% vq
->vring
.num
];
719 /* Make sure buffer is written before we update index. */
721 vq
->vring
.used
->idx
++;
724 /* This actually sends the interrupt for this virtqueue */
725 static void trigger_irq(int fd
, struct virtqueue
*vq
)
727 unsigned long buf
[] = { LHREQ_IRQ
, vq
->config
.irq
};
729 /* If they don't want an interrupt, don't send one. */
730 if (vq
->vring
.avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)
733 /* Send the Guest an interrupt tell them we used something up. */
734 if (write(fd
, buf
, sizeof(buf
)) != 0)
735 err(1, "Triggering irq %i", vq
->config
.irq
);
738 /* And here's the combo meal deal. Supersize me! */
739 static void add_used_and_trigger(int fd
, struct virtqueue
*vq
,
740 unsigned int head
, int len
)
742 add_used(vq
, head
, len
);
749 * Here is the input terminal setting we save, and the routine to restore them
750 * on exit so the user gets their terminal back. */
751 static struct termios orig_term
;
752 static void restore_term(void)
754 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
757 /* We associate some data with the console for our exit hack. */
760 /* How many times have they hit ^C? */
762 /* When did they start? */
763 struct timeval start
;
766 /* This is the routine which handles console input (ie. stdin). */
767 static bool handle_console_input(int fd
, struct device
*dev
)
770 unsigned int head
, in_num
, out_num
;
771 struct iovec iov
[dev
->vq
->vring
.num
];
772 struct console_abort
*abort
= dev
->priv
;
774 /* First we need a console buffer from the Guests's input virtqueue. */
775 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
777 /* If they're not ready for input, stop listening to this file
778 * descriptor. We'll start again once they add an input buffer. */
779 if (head
== dev
->vq
->vring
.num
)
783 errx(1, "Output buffers in console in queue?");
785 /* This is why we convert to iovecs: the readv() call uses them, and so
786 * it reads straight into the Guest's buffer. */
787 len
= readv(dev
->fd
, iov
, in_num
);
789 /* This implies that the console is closed, is /dev/null, or
790 * something went terribly wrong. */
791 warnx("Failed to get console input, ignoring console.");
792 /* Put the input terminal back. */
794 /* Remove callback from input vq, so it doesn't restart us. */
795 dev
->vq
->handle_output
= NULL
;
796 /* Stop listening to this fd: don't call us again. */
800 /* Tell the Guest about the new input. */
801 add_used_and_trigger(fd
, dev
->vq
, head
, len
);
803 /* Three ^C within one second? Exit.
805 * This is such a hack, but works surprisingly well. Each ^C has to be
806 * in a buffer by itself, so they can't be too fast. But we check that
807 * we get three within about a second, so they can't be too slow. */
808 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
810 gettimeofday(&abort
->start
, NULL
);
811 else if (abort
->count
== 3) {
813 gettimeofday(&now
, NULL
);
814 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
815 unsigned long args
[] = { LHREQ_BREAK
, 0 };
816 /* Close the fd so Waker will know it has to
819 /* Just in case waker is blocked in BREAK, send
821 write(fd
, args
, sizeof(args
));
827 /* Any other key resets the abort counter. */
830 /* Everything went OK! */
834 /* Handling output for console is simple: we just get all the output buffers
835 * and write them to stdout. */
836 static void handle_console_output(int fd
, struct virtqueue
*vq
)
838 unsigned int head
, out
, in
;
840 struct iovec iov
[vq
->vring
.num
];
842 /* Keep getting output buffers from the Guest until we run out. */
843 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
845 errx(1, "Input buffers in output queue?");
846 len
= writev(STDOUT_FILENO
, iov
, out
);
847 add_used_and_trigger(fd
, vq
, head
, len
);
854 * Handling output for network is also simple: we get all the output buffers
855 * and write them (ignoring the first element) to this device's file descriptor
858 static void handle_net_output(int fd
, struct virtqueue
*vq
)
860 unsigned int head
, out
, in
;
862 struct iovec iov
[vq
->vring
.num
];
864 /* Keep getting output buffers from the Guest until we run out. */
865 while ((head
= get_vq_desc(vq
, iov
, &out
, &in
)) != vq
->vring
.num
) {
867 errx(1, "Input buffers in output queue?");
868 /* Check header, but otherwise ignore it (we told the Guest we
869 * supported no features, so it shouldn't have anything
871 (void)convert(&iov
[0], struct virtio_net_hdr
);
872 len
= writev(vq
->dev
->fd
, iov
+1, out
-1);
873 add_used_and_trigger(fd
, vq
, head
, len
);
877 /* This is where we handle a packet coming in from the tun device to our
879 static bool handle_tun_input(int fd
, struct device
*dev
)
881 unsigned int head
, in_num
, out_num
;
883 struct iovec iov
[dev
->vq
->vring
.num
];
884 struct virtio_net_hdr
*hdr
;
886 /* First we need a network buffer from the Guests's recv virtqueue. */
887 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
888 if (head
== dev
->vq
->vring
.num
) {
889 /* Now, it's expected that if we try to send a packet too
890 * early, the Guest won't be ready yet. Wait until the device
891 * status says it's ready. */
892 /* FIXME: Actually want DRIVER_ACTIVE here. */
893 if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
)
894 warn("network: no dma buffer!");
895 /* We'll turn this back on if input buffers are registered. */
898 errx(1, "Output buffers in network recv queue?");
900 /* First element is the header: we set it to 0 (no features). */
901 hdr
= convert(&iov
[0], struct virtio_net_hdr
);
903 hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
905 /* Read the packet from the device directly into the Guest's buffer. */
906 len
= readv(dev
->fd
, iov
+1, in_num
-1);
908 err(1, "reading network");
910 /* Tell the Guest about the new packet. */
911 add_used_and_trigger(fd
, dev
->vq
, head
, sizeof(*hdr
) + len
);
913 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
914 ((u8
*)iov
[1].iov_base
)[0], ((u8
*)iov
[1].iov_base
)[1],
915 head
!= dev
->vq
->vring
.num
? "sent" : "discarded");
921 /*L:215 This is the callback attached to the network and console input
922 * virtqueues: it ensures we try again, in case we stopped console or net
923 * delivery because Guest didn't have any buffers. */
924 static void enable_fd(int fd
, struct virtqueue
*vq
)
926 add_device_fd(vq
->dev
->fd
);
927 /* Tell waker to listen to it again */
928 write(waker_fd
, &vq
->dev
->fd
, sizeof(vq
->dev
->fd
));
931 /* When the Guest tells us they updated the status field, we handle it. */
932 static void update_device_status(struct device
*dev
)
934 struct virtqueue
*vq
;
936 /* This is a reset. */
937 if (dev
->desc
->status
== 0) {
938 verbose("Resetting device %s\n", dev
->name
);
940 /* Clear any features they've acked. */
941 memset(get_feature_bits(dev
) + dev
->desc
->feature_len
, 0,
942 dev
->desc
->feature_len
);
944 /* Zero out the virtqueues. */
945 for (vq
= dev
->vq
; vq
; vq
= vq
->next
) {
946 memset(vq
->vring
.desc
, 0,
947 vring_size(vq
->config
.num
, getpagesize()));
948 vq
->last_avail_idx
= 0;
950 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_FAILED
) {
951 warnx("Device %s configuration FAILED", dev
->name
);
952 } else if (dev
->desc
->status
& VIRTIO_CONFIG_S_DRIVER_OK
) {
955 verbose("Device %s OK: offered", dev
->name
);
956 for (i
= 0; i
< dev
->desc
->feature_len
; i
++)
957 verbose(" %08x", get_feature_bits(dev
)[i
]);
958 verbose(", accepted");
959 for (i
= 0; i
< dev
->desc
->feature_len
; i
++)
960 verbose(" %08x", get_feature_bits(dev
)
961 [dev
->desc
->feature_len
+i
]);
968 /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
969 static void handle_output(int fd
, unsigned long addr
)
972 struct virtqueue
*vq
;
974 /* Check each device and virtqueue. */
975 for (i
= devices
.dev
; i
; i
= i
->next
) {
976 /* Notifications to device descriptors update device status. */
977 if (from_guest_phys(addr
) == i
->desc
) {
978 update_device_status(i
);
982 /* Notifications to virtqueues mean output has occurred. */
983 for (vq
= i
->vq
; vq
; vq
= vq
->next
) {
984 if (vq
->config
.pfn
!= addr
/getpagesize())
987 /* Guest should acknowledge (and set features!) before
988 * using the device. */
989 if (i
->desc
->status
== 0) {
990 warnx("%s gave early output", i
->name
);
994 if (strcmp(vq
->dev
->name
, "console") != 0)
995 verbose("Output to %s\n", vq
->dev
->name
);
996 if (vq
->handle_output
)
997 vq
->handle_output(fd
, vq
);
1002 /* Early console write is done using notify on a nul-terminated string
1003 * in Guest memory. */
1004 if (addr
>= guest_limit
)
1005 errx(1, "Bad NOTIFY %#lx", addr
);
1007 write(STDOUT_FILENO
, from_guest_phys(addr
),
1008 strnlen(from_guest_phys(addr
), guest_limit
- addr
));
1011 /* This is called when the Waker wakes us up: check for incoming file
1013 static void handle_input(int fd
)
1015 /* select() wants a zeroed timeval to mean "don't wait". */
1016 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
1020 fd_set fds
= devices
.infds
;
1022 /* If nothing is ready, we're done. */
1023 if (select(devices
.max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
1026 /* Otherwise, call the device(s) which have readable file
1027 * descriptors and a method of handling them. */
1028 for (i
= devices
.dev
; i
; i
= i
->next
) {
1029 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
1031 if (i
->handle_input(fd
, i
))
1034 /* If handle_input() returns false, it means we
1035 * should no longer service it. Networking and
1036 * console do this when there's no input
1037 * buffers to deliver into. Console also uses
1038 * it when it discovers that stdin is closed. */
1039 FD_CLR(i
->fd
, &devices
.infds
);
1040 /* Tell waker to ignore it too, by sending a
1041 * negative fd number (-1, since 0 is a valid
1043 dev_fd
= -i
->fd
- 1;
1044 write(waker_fd
, &dev_fd
, sizeof(dev_fd
));
1053 * All devices need a descriptor so the Guest knows it exists, and a "struct
1054 * device" so the Launcher can keep track of it. We have common helper
1055 * routines to allocate and manage them.
1058 /* The layout of the device page is a "struct lguest_device_desc" followed by a
1059 * number of virtqueue descriptors, then two sets of feature bits, then an
1060 * array of configuration bytes. This routine returns the configuration
1062 static u8
*device_config(const struct device
*dev
)
1064 return (void *)(dev
->desc
+ 1)
1065 + dev
->desc
->num_vq
* sizeof(struct lguest_vqconfig
)
1066 + dev
->desc
->feature_len
* 2;
1069 /* This routine allocates a new "struct lguest_device_desc" from descriptor
1070 * table page just above the Guest's normal memory. It returns a pointer to
1071 * that descriptor. */
1072 static struct lguest_device_desc
*new_dev_desc(u16 type
)
1074 struct lguest_device_desc d
= { .type
= type
};
1077 /* Figure out where the next device config is, based on the last one. */
1078 if (devices
.lastdev
)
1079 p
= device_config(devices
.lastdev
)
1080 + devices
.lastdev
->desc
->config_len
;
1082 p
= devices
.descpage
;
1084 /* We only have one page for all the descriptors. */
1085 if (p
+ sizeof(d
) > (void *)devices
.descpage
+ getpagesize())
1086 errx(1, "Too many devices");
1088 /* p might not be aligned, so we memcpy in. */
1089 return memcpy(p
, &d
, sizeof(d
));
1092 /* Each device descriptor is followed by the description of its virtqueues. We
1093 * specify how many descriptors the virtqueue is to have. */
1094 static void add_virtqueue(struct device
*dev
, unsigned int num_descs
,
1095 void (*handle_output
)(int fd
, struct virtqueue
*me
))
1098 struct virtqueue
**i
, *vq
= malloc(sizeof(*vq
));
1101 /* First we need some memory for this virtqueue. */
1102 pages
= (vring_size(num_descs
, getpagesize()) + getpagesize() - 1)
1104 p
= get_pages(pages
);
1106 /* Initialize the virtqueue */
1108 vq
->last_avail_idx
= 0;
1111 /* Initialize the configuration. */
1112 vq
->config
.num
= num_descs
;
1113 vq
->config
.irq
= devices
.next_irq
++;
1114 vq
->config
.pfn
= to_guest_phys(p
) / getpagesize();
1116 /* Initialize the vring. */
1117 vring_init(&vq
->vring
, num_descs
, p
, getpagesize());
1119 /* Append virtqueue to this device's descriptor. We use
1120 * device_config() to get the end of the device's current virtqueues;
1121 * we check that we haven't added any config or feature information
1122 * yet, otherwise we'd be overwriting them. */
1123 assert(dev
->desc
->config_len
== 0 && dev
->desc
->feature_len
== 0);
1124 memcpy(device_config(dev
), &vq
->config
, sizeof(vq
->config
));
1125 dev
->desc
->num_vq
++;
1127 verbose("Virtqueue page %#lx\n", to_guest_phys(p
));
1129 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1131 for (i
= &dev
->vq
; *i
; i
= &(*i
)->next
);
1134 /* Set the routine to call when the Guest does something to this
1136 vq
->handle_output
= handle_output
;
1138 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1139 * don't have a handler */
1141 vq
->vring
.used
->flags
= VRING_USED_F_NO_NOTIFY
;
1144 /* The first half of the feature bitmask is for us to advertise features. The
1145 * second half is for the Guest to accept features. */
1146 static void add_feature(struct device
*dev
, unsigned bit
)
1148 u8
*features
= get_feature_bits(dev
);
1150 /* We can't extend the feature bits once we've added config bytes */
1151 if (dev
->desc
->feature_len
<= bit
/ CHAR_BIT
) {
1152 assert(dev
->desc
->config_len
== 0);
1153 dev
->desc
->feature_len
= (bit
/ CHAR_BIT
) + 1;
1156 features
[bit
/ CHAR_BIT
] |= (1 << (bit
% CHAR_BIT
));
1159 /* This routine sets the configuration fields for an existing device's
1160 * descriptor. It only works for the last device, but that's OK because that's
1162 static void set_config(struct device
*dev
, unsigned len
, const void *conf
)
1164 /* Check we haven't overflowed our single page. */
1165 if (device_config(dev
) + len
> devices
.descpage
+ getpagesize())
1166 errx(1, "Too many devices");
1168 /* Copy in the config information, and store the length. */
1169 memcpy(device_config(dev
), conf
, len
);
1170 dev
->desc
->config_len
= len
;
1173 /* This routine does all the creation and setup of a new device, including
1174 * calling new_dev_desc() to allocate the descriptor and device memory.
1176 * See what I mean about userspace being boring? */
1177 static struct device
*new_device(const char *name
, u16 type
, int fd
,
1178 bool (*handle_input
)(int, struct device
*))
1180 struct device
*dev
= malloc(sizeof(*dev
));
1182 /* Now we populate the fields one at a time. */
1184 /* If we have an input handler for this file descriptor, then we add it
1185 * to the device_list's fdset and maxfd. */
1187 add_device_fd(dev
->fd
);
1188 dev
->desc
= new_dev_desc(type
);
1189 dev
->handle_input
= handle_input
;
1194 /* Append to device list. Prepending to a single-linked list is
1195 * easier, but the user expects the devices to be arranged on the bus
1196 * in command-line order. The first network device on the command line
1197 * is eth0, the first block device /dev/vda, etc. */
1198 if (devices
.lastdev
)
1199 devices
.lastdev
->next
= dev
;
1202 devices
.lastdev
= dev
;
1207 /* Our first setup routine is the console. It's a fairly simple device, but
1208 * UNIX tty handling makes it uglier than it could be. */
1209 static void setup_console(void)
1213 /* If we can save the initial standard input settings... */
1214 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1215 struct termios term
= orig_term
;
1216 /* Then we turn off echo, line buffering and ^C etc. We want a
1217 * raw input stream to the Guest. */
1218 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1219 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1220 /* If we exit gracefully, the original settings will be
1221 * restored so the user can see what they're typing. */
1222 atexit(restore_term
);
1225 dev
= new_device("console", VIRTIO_ID_CONSOLE
,
1226 STDIN_FILENO
, handle_console_input
);
1227 /* We store the console state in dev->priv, and initialize it. */
1228 dev
->priv
= malloc(sizeof(struct console_abort
));
1229 ((struct console_abort
*)dev
->priv
)->count
= 0;
1231 /* The console needs two virtqueues: the input then the output. When
1232 * they put something the input queue, we make sure we're listening to
1233 * stdin. When they put something in the output queue, we write it to
1235 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1236 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_console_output
);
1238 verbose("device %u: console\n", devices
.device_num
++);
1242 /*M:010 Inter-guest networking is an interesting area. Simplest is to have a
1243 * --sharenet=<name> option which opens or creates a named pipe. This can be
1244 * used to send packets to another guest in a 1:1 manner.
1246 * More sopisticated is to use one of the tools developed for project like UML
1249 * Faster is to do virtio bonding in kernel. Doing this 1:1 would be
1250 * completely generic ("here's my vring, attach to your vring") and would work
1251 * for any traffic. Of course, namespace and permissions issues need to be
1252 * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide
1253 * multiple inter-guest channels behind one interface, although it would
1254 * require some manner of hotplugging new virtio channels.
1256 * Finally, we could implement a virtio network switch in the kernel. :*/
1258 static u32
str2ip(const char *ipaddr
)
1260 unsigned int byte
[4];
1262 sscanf(ipaddr
, "%u.%u.%u.%u", &byte
[0], &byte
[1], &byte
[2], &byte
[3]);
1263 return (byte
[0] << 24) | (byte
[1] << 16) | (byte
[2] << 8) | byte
[3];
1266 /* This code is "adapted" from libbridge: it attaches the Host end of the
1267 * network device to the bridge device specified by the command line.
1269 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1270 * dislike bridging), and I just try not to break it. */
1271 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1277 errx(1, "must specify bridge name");
1279 ifidx
= if_nametoindex(if_name
);
1281 errx(1, "interface %s does not exist!", if_name
);
1283 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1284 ifr
.ifr_ifindex
= ifidx
;
1285 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1286 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1289 /* This sets up the Host end of the network device with an IP address, brings
1290 * it up so packets will flow, the copies the MAC address into the hwaddr
1292 static void configure_device(int fd
, const char *devname
, u32 ipaddr
,
1293 unsigned char hwaddr
[6])
1296 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1298 /* Don't read these incantations. Just cut & paste them like I did! */
1299 memset(&ifr
, 0, sizeof(ifr
));
1300 strcpy(ifr
.ifr_name
, devname
);
1301 sin
->sin_family
= AF_INET
;
1302 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1303 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1304 err(1, "Setting %s interface address", devname
);
1305 ifr
.ifr_flags
= IFF_UP
;
1306 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1307 err(1, "Bringing interface %s up", devname
);
1309 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1310 * above). IF means Interface, and HWADDR is hardware address.
1312 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1313 err(1, "getting hw address for %s", devname
);
1314 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1317 /*L:195 Our network is a Host<->Guest network. This can either use bridging or
1318 * routing, but the principle is the same: it uses the "tun" device to inject
1319 * packets into the Host as if they came in from a normal network card. We
1320 * just shunt packets between the Guest and the tun device. */
1321 static void setup_tun_net(const char *arg
)
1327 const char *br_name
= NULL
;
1328 struct virtio_net_config conf
;
1330 /* We open the /dev/net/tun device and tell it we want a tap device. A
1331 * tap device is like a tun device, only somehow different. To tell
1332 * the truth, I completely blundered my way through this code, but it
1334 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1335 memset(&ifr
, 0, sizeof(ifr
));
1336 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1337 strcpy(ifr
.ifr_name
, "tap%d");
1338 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1339 err(1, "configuring /dev/net/tun");
1340 /* We don't need checksums calculated for packets coming in this
1341 * device: trust us! */
1342 ioctl(netfd
, TUNSETNOCSUM
, 1);
1344 /* First we create a new network device. */
1345 dev
= new_device("net", VIRTIO_ID_NET
, netfd
, handle_tun_input
);
1347 /* Network devices need a receive and a send queue, just like
1349 add_virtqueue(dev
, VIRTQUEUE_NUM
, enable_fd
);
1350 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_net_output
);
1352 /* We need a socket to perform the magic network ioctls to bring up the
1353 * tap interface, connect to the bridge etc. Any socket will do! */
1354 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1356 err(1, "opening IP socket");
1358 /* If the command line was --tunnet=bridge:<name> do bridging. */
1359 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1361 br_name
= arg
+ strlen(BRIDGE_PFX
);
1362 add_to_bridge(ipfd
, ifr
.ifr_name
, br_name
);
1363 } else /* It is an IP address to set up the device with */
1366 /* Set up the tun device, and get the mac address for the interface. */
1367 configure_device(ipfd
, ifr
.ifr_name
, ip
, conf
.mac
);
1369 /* Tell Guest what MAC address to use. */
1370 add_feature(dev
, VIRTIO_NET_F_MAC
);
1371 set_config(dev
, sizeof(conf
), &conf
);
1373 /* We don't need the socket any more; setup is done. */
1376 verbose("device %u: tun net %u.%u.%u.%u\n",
1377 devices
.device_num
++,
1378 (u8
)(ip
>>24),(u8
)(ip
>>16),(u8
)(ip
>>8),(u8
)ip
);
1380 verbose("attached to bridge: %s\n", br_name
);
1383 /* Our block (disk) device should be really simple: the Guest asks for a block
1384 * number and we read or write that position in the file. Unfortunately, that
1385 * was amazingly slow: the Guest waits until the read is finished before
1386 * running anything else, even if it could have been doing useful work.
1388 * We could use async I/O, except it's reputed to suck so hard that characters
1389 * actually go missing from your code when you try to use it.
1391 * So we farm the I/O out to thread, and communicate with it via a pipe. */
1393 /* This hangs off device->priv. */
1396 /* The size of the file. */
1399 /* The file descriptor for the file. */
1402 /* IO thread listens on this file descriptor [0]. */
1405 /* IO thread writes to this file descriptor to mark it done, then
1406 * Launcher triggers interrupt to Guest. */
1413 * Remember that the block device is handled by a separate I/O thread. We head
1414 * straight into the core of that thread here:
1416 static bool service_io(struct device
*dev
)
1418 struct vblk_info
*vblk
= dev
->priv
;
1419 unsigned int head
, out_num
, in_num
, wlen
;
1422 struct virtio_blk_outhdr
*out
;
1423 struct iovec iov
[dev
->vq
->vring
.num
];
1426 /* See if there's a request waiting. If not, nothing to do. */
1427 head
= get_vq_desc(dev
->vq
, iov
, &out_num
, &in_num
);
1428 if (head
== dev
->vq
->vring
.num
)
1431 /* Every block request should contain at least one output buffer
1432 * (detailing the location on disk and the type of request) and one
1433 * input buffer (to hold the result). */
1434 if (out_num
== 0 || in_num
== 0)
1435 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1436 head
, out_num
, in_num
);
1438 out
= convert(&iov
[0], struct virtio_blk_outhdr
);
1439 in
= convert(&iov
[out_num
+in_num
-1], u8
);
1440 off
= out
->sector
* 512;
1442 /* The block device implements "barriers", where the Guest indicates
1443 * that it wants all previous writes to occur before this write. We
1444 * don't have a way of asking our kernel to do a barrier, so we just
1445 * synchronize all the data in the file. Pretty poor, no? */
1446 if (out
->type
& VIRTIO_BLK_T_BARRIER
)
1447 fdatasync(vblk
->fd
);
1449 /* In general the virtio block driver is allowed to try SCSI commands.
1450 * It'd be nice if we supported eject, for example, but we don't. */
1451 if (out
->type
& VIRTIO_BLK_T_SCSI_CMD
) {
1452 fprintf(stderr
, "Scsi commands unsupported\n");
1453 *in
= VIRTIO_BLK_S_UNSUPP
;
1455 } else if (out
->type
& VIRTIO_BLK_T_OUT
) {
1458 /* Move to the right location in the block file. This can fail
1459 * if they try to write past end. */
1460 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1461 err(1, "Bad seek to sector %llu", out
->sector
);
1463 ret
= writev(vblk
->fd
, iov
+1, out_num
-1);
1464 verbose("WRITE to sector %llu: %i\n", out
->sector
, ret
);
1466 /* Grr... Now we know how long the descriptor they sent was, we
1467 * make sure they didn't try to write over the end of the block
1468 * file (possibly extending it). */
1469 if (ret
> 0 && off
+ ret
> vblk
->len
) {
1470 /* Trim it back to the correct length */
1471 ftruncate64(vblk
->fd
, vblk
->len
);
1472 /* Die, bad Guest, die. */
1473 errx(1, "Write past end %llu+%u", off
, ret
);
1476 *in
= (ret
>= 0 ? VIRTIO_BLK_S_OK
: VIRTIO_BLK_S_IOERR
);
1480 /* Move to the right location in the block file. This can fail
1481 * if they try to read past end. */
1482 if (lseek64(vblk
->fd
, off
, SEEK_SET
) != off
)
1483 err(1, "Bad seek to sector %llu", out
->sector
);
1485 ret
= readv(vblk
->fd
, iov
+1, in_num
-1);
1486 verbose("READ from sector %llu: %i\n", out
->sector
, ret
);
1488 wlen
= sizeof(*in
) + ret
;
1489 *in
= VIRTIO_BLK_S_OK
;
1492 *in
= VIRTIO_BLK_S_IOERR
;
1496 /* We can't trigger an IRQ, because we're not the Launcher. It does
1497 * that when we tell it we're done. */
1498 add_used(dev
->vq
, head
, wlen
);
1502 /* This is the thread which actually services the I/O. */
1503 static int io_thread(void *_dev
)
1505 struct device
*dev
= _dev
;
1506 struct vblk_info
*vblk
= dev
->priv
;
1509 /* Close other side of workpipe so we get 0 read when main dies. */
1510 close(vblk
->workpipe
[1]);
1511 /* Close the other side of the done_fd pipe. */
1514 /* When this read fails, it means Launcher died, so we follow. */
1515 while (read(vblk
->workpipe
[0], &c
, 1) == 1) {
1516 /* We acknowledge each request immediately to reduce latency,
1517 * rather than waiting until we've done them all. I haven't
1518 * measured to see if it makes any difference.
1520 * That would be an interesting test, wouldn't it? You could
1521 * also try having more than one I/O thread. */
1522 while (service_io(dev
))
1523 write(vblk
->done_fd
, &c
, 1);
1528 /* Now we've seen the I/O thread, we return to the Launcher to see what happens
1529 * when that thread tells us it's completed some I/O. */
1530 static bool handle_io_finish(int fd
, struct device
*dev
)
1534 /* If the I/O thread died, presumably it printed the error, so we
1536 if (read(dev
->fd
, &c
, 1) != 1)
1539 /* It did some work, so trigger the irq. */
1540 trigger_irq(fd
, dev
->vq
);
1544 /* When the Guest submits some I/O, we just need to wake the I/O thread. */
1545 static void handle_virtblk_output(int fd
, struct virtqueue
*vq
)
1547 struct vblk_info
*vblk
= vq
->dev
->priv
;
1550 /* Wake up I/O thread and tell it to go to work! */
1551 if (write(vblk
->workpipe
[1], &c
, 1) != 1)
1552 /* Presumably it indicated why it died. */
1556 /*L:198 This actually sets up a virtual block device. */
1557 static void setup_block_file(const char *filename
)
1561 struct vblk_info
*vblk
;
1563 struct virtio_blk_config conf
;
1565 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1568 /* The device responds to return from I/O thread. */
1569 dev
= new_device("block", VIRTIO_ID_BLOCK
, p
[0], handle_io_finish
);
1571 /* The device has one virtqueue, where the Guest places requests. */
1572 add_virtqueue(dev
, VIRTQUEUE_NUM
, handle_virtblk_output
);
1574 /* Allocate the room for our own bookkeeping */
1575 vblk
= dev
->priv
= malloc(sizeof(*vblk
));
1577 /* First we open the file and store the length. */
1578 vblk
->fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
);
1579 vblk
->len
= lseek64(vblk
->fd
, 0, SEEK_END
);
1581 /* We support barriers. */
1582 add_feature(dev
, VIRTIO_BLK_F_BARRIER
);
1584 /* Tell Guest how many sectors this device has. */
1585 conf
.capacity
= cpu_to_le64(vblk
->len
/ 512);
1587 /* Tell Guest not to put in too many descriptors at once: two are used
1588 * for the in and out elements. */
1589 add_feature(dev
, VIRTIO_BLK_F_SEG_MAX
);
1590 conf
.seg_max
= cpu_to_le32(VIRTQUEUE_NUM
- 2);
1592 set_config(dev
, sizeof(conf
), &conf
);
1594 /* The I/O thread writes to this end of the pipe when done. */
1595 vblk
->done_fd
= p
[1];
1597 /* This is the second pipe, which is how we tell the I/O thread about
1599 pipe(vblk
->workpipe
);
1601 /* Create stack for thread and run it. Since stack grows upwards, we
1602 * point the stack pointer to the end of this region. */
1603 stack
= malloc(32768);
1604 /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from
1605 * becoming a zombie. */
1606 if (clone(io_thread
, stack
+ 32768, CLONE_VM
| SIGCHLD
, dev
) == -1)
1607 err(1, "Creating clone");
1609 /* We don't need to keep the I/O thread's end of the pipes open. */
1610 close(vblk
->done_fd
);
1611 close(vblk
->workpipe
[0]);
1613 verbose("device %u: virtblock %llu sectors\n",
1614 devices
.device_num
, le64_to_cpu(conf
.capacity
));
1616 /* That's the end of device setup. */
1618 /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */
1619 static void __attribute__((noreturn
)) restart_guest(void)
1623 /* Closing pipes causes the Waker thread and io_threads to die, and
1624 * closing /dev/lguest cleans up the Guest. Since we don't track all
1625 * open fds, we simply close everything beyond stderr. */
1626 for (i
= 3; i
< FD_SETSIZE
; i
++)
1628 execv(main_args
[0], main_args
);
1629 err(1, "Could not exec %s", main_args
[0]);
1632 /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves
1633 * its input and output, and finally, lays it to rest. */
1634 static void __attribute__((noreturn
)) run_guest(int lguest_fd
)
1637 unsigned long args
[] = { LHREQ_BREAK
, 0 };
1638 unsigned long notify_addr
;
1641 /* We read from the /dev/lguest device to run the Guest. */
1642 readval
= pread(lguest_fd
, ¬ify_addr
,
1643 sizeof(notify_addr
), cpu_id
);
1645 /* One unsigned long means the Guest did HCALL_NOTIFY */
1646 if (readval
== sizeof(notify_addr
)) {
1647 verbose("Notify on address %#lx\n", notify_addr
);
1648 handle_output(lguest_fd
, notify_addr
);
1650 /* ENOENT means the Guest died. Reading tells us why. */
1651 } else if (errno
== ENOENT
) {
1652 char reason
[1024] = { 0 };
1653 pread(lguest_fd
, reason
, sizeof(reason
)-1, cpu_id
);
1654 errx(1, "%s", reason
);
1655 /* ERESTART means that we need to reboot the guest */
1656 } else if (errno
== ERESTART
) {
1658 /* EAGAIN means the Waker wanted us to look at some input.
1659 * Anything else means a bug or incompatible change. */
1660 } else if (errno
!= EAGAIN
)
1661 err(1, "Running guest failed");
1663 /* Only service input on thread for CPU 0. */
1667 /* Service input, then unset the BREAK to release the Waker. */
1668 handle_input(lguest_fd
);
1669 if (pwrite(lguest_fd
, args
, sizeof(args
), cpu_id
) < 0)
1670 err(1, "Resetting break");
1674 * This is the end of the Launcher. The good news: we are over halfway
1675 * through! The bad news: the most fiendish part of the code still lies ahead
1678 * Are you ready? Take a deep breath and join me in the core of the Host, in
1682 static struct option opts
[] = {
1683 { "verbose", 0, NULL
, 'v' },
1684 { "tunnet", 1, NULL
, 't' },
1685 { "block", 1, NULL
, 'b' },
1686 { "initrd", 1, NULL
, 'i' },
1689 static void usage(void)
1691 errx(1, "Usage: lguest [--verbose] "
1692 "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1693 "|--block=<filename>|--initrd=<filename>]...\n"
1694 "<mem-in-mb> vmlinux [args...]");
1697 /*L:105 The main routine is where the real work begins: */
1698 int main(int argc
, char *argv
[])
1700 /* Memory, top-level pagetable, code startpoint and size of the
1701 * (optional) initrd. */
1702 unsigned long mem
= 0, pgdir
, start
, initrd_size
= 0;
1703 /* Two temporaries and the /dev/lguest file descriptor. */
1704 int i
, c
, lguest_fd
;
1705 /* The boot information for the Guest. */
1706 struct boot_params
*boot
;
1707 /* If they specify an initrd file to load. */
1708 const char *initrd_name
= NULL
;
1710 /* Save the args: we "reboot" by execing ourselves again. */
1712 /* We don't "wait" for the children, so prevent them from becoming
1714 signal(SIGCHLD
, SIG_IGN
);
1716 /* First we initialize the device list. Since console and network
1717 * device receive input from a file descriptor, we keep an fdset
1718 * (infds) and the maximum fd number (max_infd) with the head of the
1719 * list. We also keep a pointer to the last device. Finally, we keep
1720 * the next interrupt number to use for devices (1: remember that 0 is
1721 * used by the timer). */
1722 FD_ZERO(&devices
.infds
);
1723 devices
.max_infd
= -1;
1724 devices
.lastdev
= NULL
;
1725 devices
.next_irq
= 1;
1728 /* We need to know how much memory so we can set up the device
1729 * descriptor and memory pages for the devices as we parse the command
1730 * line. So we quickly look through the arguments to find the amount
1732 for (i
= 1; i
< argc
; i
++) {
1733 if (argv
[i
][0] != '-') {
1734 mem
= atoi(argv
[i
]) * 1024 * 1024;
1735 /* We start by mapping anonymous pages over all of
1736 * guest-physical memory range. This fills it with 0,
1737 * and ensures that the Guest won't be killed when it
1738 * tries to access it. */
1739 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1742 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1743 devices
.descpage
= get_pages(1);
1748 /* The options are fairly straight-forward */
1749 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1755 setup_tun_net(optarg
);
1758 setup_block_file(optarg
);
1761 initrd_name
= optarg
;
1764 warnx("Unknown argument %s", argv
[optind
]);
1768 /* After the other arguments we expect memory and kernel image name,
1769 * followed by command line arguments for the kernel. */
1770 if (optind
+ 2 > argc
)
1773 verbose("Guest base is at %p\n", guest_base
);
1775 /* We always have a console device */
1778 /* Now we load the kernel */
1779 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
));
1781 /* Boot information is stashed at physical address 0 */
1782 boot
= from_guest_phys(0);
1784 /* Map the initrd image if requested (at top of physical memory) */
1786 initrd_size
= load_initrd(initrd_name
, mem
);
1787 /* These are the location in the Linux boot header where the
1788 * start and size of the initrd are expected to be found. */
1789 boot
->hdr
.ramdisk_image
= mem
- initrd_size
;
1790 boot
->hdr
.ramdisk_size
= initrd_size
;
1791 /* The bootloader type 0xFF means "unknown"; that's OK. */
1792 boot
->hdr
.type_of_loader
= 0xFF;
1795 /* Set up the initial linear pagetables, starting below the initrd. */
1796 pgdir
= setup_pagetables(mem
, initrd_size
);
1798 /* The Linux boot header contains an "E820" memory map: ours is a
1799 * simple, single region. */
1800 boot
->e820_entries
= 1;
1801 boot
->e820_map
[0] = ((struct e820entry
) { 0, mem
, E820_RAM
});
1802 /* The boot header contains a command line pointer: we put the command
1803 * line after the boot header. */
1804 boot
->hdr
.cmd_line_ptr
= to_guest_phys(boot
+ 1);
1805 /* We use a simple helper to copy the arguments separated by spaces. */
1806 concat((char *)(boot
+ 1), argv
+optind
+2);
1808 /* Boot protocol version: 2.07 supports the fields for lguest. */
1809 boot
->hdr
.version
= 0x207;
1811 /* The hardware_subarch value of "1" tells the Guest it's an lguest. */
1812 boot
->hdr
.hardware_subarch
= 1;
1814 /* Tell the entry path not to try to reload segment registers. */
1815 boot
->hdr
.loadflags
|= KEEP_SEGMENTS
;
1817 /* We tell the kernel to initialize the Guest: this returns the open
1818 * /dev/lguest file descriptor. */
1819 lguest_fd
= tell_kernel(pgdir
, start
);
1821 /* We fork off a child process, which wakes the Launcher whenever one
1822 * of the input file descriptors needs attention. We call this the
1823 * Waker, and we'll cover it in a moment. */
1824 waker_fd
= setup_waker(lguest_fd
);
1826 /* Finally, run the Guest. This doesn't return. */
1827 run_guest(lguest_fd
);
1832 * Mastery is done: you now know everything I do.
1834 * But surely you have seen code, features and bugs in your wanderings which
1835 * you now yearn to attack? That is the real game, and I look forward to you
1836 * patching and forking lguest into the Your-Name-Here-visor.
1838 * Farewell, and good coding!