1 /*P:100 This is the Launcher code, a simple program which lays out the
2 * "physical" memory for the new Guest by mapping the kernel image and the
3 * virtual devices, then reads repeatedly from /dev/lguest to run the Guest.
5 * The only trick: the Makefile links it at a high address so it will be clear
6 * of the guest memory region. It means that each Guest cannot have more than
7 * about 2.5G of memory on a normally configured Host. :*/
8 #define _LARGEFILE64_SOURCE
18 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/ioctl.h>
29 #include <netinet/in.h>
31 #include <linux/sockios.h>
32 #include <linux/if_tun.h>
37 /*L:110 We can ignore the 28 include files we need for this program, but I do
38 * want to draw attention to the use of kernel-style types.
40 * As Linus said, "C is a Spartan language, and so should your naming be." I
41 * like these abbreviations and the header we need uses them, so we define them
44 typedef unsigned long long u64
;
48 #include "../../include/linux/lguest_launcher.h"
49 #include "../../include/asm-x86/e820.h"
52 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
54 #define BRIDGE_PFX "bridge:"
56 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
59 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
60 * this, and although I wouldn't recommend it, it works quite nicely here. */
62 #define verbose(args...) \
63 do { if (verbose) printf(args); } while(0)
66 /* The pipe to send commands to the waker process */
68 /* The top of guest physical memory. */
71 /* This is our list of devices. */
74 /* Summary information about the devices in our list: ready to pass to
75 * select() to ask which need servicing.*/
79 /* The descriptor page for the devices. */
80 struct lguest_device_desc
*descs
;
82 /* A single linked list of devices. */
84 /* ... And an end pointer so we can easily append new devices */
85 struct device
**lastdev
;
88 /* The device structure describes a single device. */
91 /* The linked-list pointer. */
93 /* The descriptor for this device, as mapped into the Guest. */
94 struct lguest_device_desc
*desc
;
95 /* The memory page(s) of this device, if any. Also mapped in Guest. */
98 /* If handle_input is set, it wants to be called when this file
99 * descriptor is ready. */
101 bool (*handle_input
)(int fd
, struct device
*me
);
103 /* If handle_output is set, it wants to be called when the Guest sends
104 * DMA to this key. */
105 unsigned long watch_key
;
106 u32 (*handle_output
)(int fd
, const struct iovec
*iov
,
107 unsigned int num
, struct device
*me
);
109 /* Device-specific data. */
114 * Loading the Kernel.
116 * We start with couple of simple helper routines. open_or_die() avoids
117 * error-checking code cluttering the callers: */
118 static int open_or_die(const char *name
, int flags
)
120 int fd
= open(name
, flags
);
122 err(1, "Failed to open %s", name
);
126 /* map_zeroed_pages() takes a (page-aligned) address and a number of pages. */
127 static void *map_zeroed_pages(unsigned long addr
, unsigned int num
)
129 /* We cache the /dev/zero file-descriptor so we only open it once. */
133 fd
= open_or_die("/dev/zero", O_RDONLY
);
135 /* We use a private mapping (ie. if we write to the page, it will be
136 * copied), and obviously we insist that it be mapped where we ask. */
137 if (mmap((void *)addr
, getpagesize() * num
,
138 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_FIXED
|MAP_PRIVATE
, fd
, 0)
140 err(1, "Mmaping %u pages of /dev/zero @%p", num
, (void *)addr
);
142 /* Returning the address is just a courtesy: can simplify callers. */
146 /* To find out where to start we look for the magic Guest string, which marks
147 * the code we see in lguest_asm.S. This is a hack which we are currently
148 * plotting to replace with the normal Linux entry point. */
149 static unsigned long entry_point(void *start
, void *end
,
150 unsigned long page_offset
)
154 /* The scan gives us the physical starting address. We want the
155 * virtual address in this case, and fortunately, we already figured
156 * out the physical-virtual difference and passed it here in
158 for (p
= start
; p
< end
; p
++)
159 if (memcmp(p
, "GenuineLguest", strlen("GenuineLguest")) == 0)
160 return (long)p
+ strlen("GenuineLguest") + page_offset
;
162 err(1, "Is this image a genuine lguest?");
165 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
166 * the Guest memory. ELF = Embedded Linking Format, which is the format used
167 * by all modern binaries on Linux including the kernel.
169 * The ELF headers give *two* addresses: a physical address, and a virtual
170 * address. The Guest kernel expects to be placed in memory at the physical
171 * address, and the page tables set up so it will correspond to that virtual
172 * address. We return the difference between the virtual and physical
173 * addresses in the "page_offset" pointer.
175 * We return the starting address. */
176 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
,
177 unsigned long *page_offset
)
180 Elf32_Phdr phdr
[ehdr
->e_phnum
];
182 unsigned long start
= -1UL, end
= 0;
184 /* Sanity checks on the main ELF header: an x86 executable with a
185 * reasonable number of correctly-sized program headers. */
186 if (ehdr
->e_type
!= ET_EXEC
187 || ehdr
->e_machine
!= EM_386
188 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
189 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
190 errx(1, "Malformed elf header");
192 /* An ELF executable contains an ELF header and a number of "program"
193 * headers which indicate which parts ("segments") of the program to
196 /* We read in all the program headers at once: */
197 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
198 err(1, "Seeking to program headers");
199 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
200 err(1, "Reading program headers");
202 /* We don't know page_offset yet. */
205 /* Try all the headers: there are usually only three. A read-only one,
206 * a read-write one, and a "note" section which isn't loadable. */
207 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
208 /* If this isn't a loadable segment, we ignore it */
209 if (phdr
[i
].p_type
!= PT_LOAD
)
212 verbose("Section %i: size %i addr %p\n",
213 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
215 /* We expect a simple linear address space: every segment must
216 * have the same difference between virtual (p_vaddr) and
217 * physical (p_paddr) address. */
219 *page_offset
= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
;
220 else if (*page_offset
!= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
)
221 errx(1, "Page offset of section %i different", i
);
223 /* We track the first and last address we mapped, so we can
224 * tell entry_point() where to scan. */
225 if (phdr
[i
].p_paddr
< start
)
226 start
= phdr
[i
].p_paddr
;
227 if (phdr
[i
].p_paddr
+ phdr
[i
].p_filesz
> end
)
228 end
= phdr
[i
].p_paddr
+ phdr
[i
].p_filesz
;
230 /* We map this section of the file at its physical address. We
231 * map it read & write even if the header says this segment is
232 * read-only. The kernel really wants to be writable: it
233 * patches its own instructions which would normally be
236 * MAP_PRIVATE means that the page won't be copied until a
237 * write is done to it. This allows us to share much of the
238 * kernel memory between Guests. */
239 addr
= mmap((void *)phdr
[i
].p_paddr
,
241 PROT_READ
|PROT_WRITE
|PROT_EXEC
,
242 MAP_FIXED
|MAP_PRIVATE
,
243 elf_fd
, phdr
[i
].p_offset
);
244 if (addr
!= (void *)phdr
[i
].p_paddr
)
245 err(1, "Mmaping vmlinux seg %i gave %p not %p",
246 i
, addr
, (void *)phdr
[i
].p_paddr
);
249 return entry_point((void *)start
, (void *)end
, *page_offset
);
252 /*L:170 Prepare to be SHOCKED and AMAZED. And possibly a trifle nauseated.
254 * We know that CONFIG_PAGE_OFFSET sets what virtual address the kernel expects
255 * to be. We don't know what that option was, but we can figure it out
256 * approximately by looking at the addresses in the code. I chose the common
257 * case of reading a memory location into the %eax register:
259 * movl <some-address>, %eax
261 * This gets encoded as five bytes: "0xA1 <4-byte-address>". For example,
262 * "0xA1 0x18 0x60 0x47 0xC0" reads the address 0xC0476018 into %eax.
264 * In this example can guess that the kernel was compiled with
265 * CONFIG_PAGE_OFFSET set to 0xC0000000 (it's always a round number). If the
266 * kernel were larger than 16MB, we might see 0xC1 addresses show up, but our
267 * kernel isn't that bloated yet.
269 * Unfortunately, x86 has variable-length instructions, so finding this
270 * particular instruction properly involves writing a disassembler. Instead,
271 * we rely on statistics. We look for "0xA1" and tally the different bytes
272 * which occur 4 bytes later (the "0xC0" in our example above). When one of
273 * those bytes appears three times, we can be reasonably confident that it
274 * forms the start of CONFIG_PAGE_OFFSET.
276 * This is amazingly reliable. */
277 static unsigned long intuit_page_offset(unsigned char *img
, unsigned long len
)
279 unsigned int i
, possibilities
[256] = { 0 };
281 for (i
= 0; i
+ 4 < len
; i
++) {
282 /* mov 0xXXXXXXXX,%eax */
283 if (img
[i
] == 0xA1 && ++possibilities
[img
[i
+4]] > 3)
284 return (unsigned long)img
[i
+4] << 24;
286 errx(1, "could not determine page offset");
289 /*L:160 Unfortunately the entire ELF image isn't compressed: the segments
290 * which need loading are extracted and compressed raw. This denies us the
291 * information we need to make a fully-general loader. */
292 static unsigned long unpack_bzimage(int fd
, unsigned long *page_offset
)
296 /* A bzImage always gets loaded at physical address 1M. This is
297 * actually configurable as CONFIG_PHYSICAL_START, but as the comment
298 * there says, "Don't change this unless you know what you are doing".
300 void *img
= (void *)0x100000;
302 /* gzdopen takes our file descriptor (carefully placed at the start of
303 * the GZIP header we found) and returns a gzFile. */
304 f
= gzdopen(fd
, "rb");
305 /* We read it into memory in 64k chunks until we hit the end. */
306 while ((ret
= gzread(f
, img
+ len
, 65536)) > 0)
309 err(1, "reading image from bzImage");
311 verbose("Unpacked size %i addr %p\n", len
, img
);
313 /* Without the ELF header, we can't tell virtual-physical gap. This is
314 * CONFIG_PAGE_OFFSET, and people do actually change it. Fortunately,
315 * I have a clever way of figuring it out from the code itself. */
316 *page_offset
= intuit_page_offset(img
, len
);
318 return entry_point(img
, img
+ len
, *page_offset
);
321 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
322 * supposed to jump into it and it will unpack itself. We can't do that
323 * because the Guest can't run the unpacking code, and adding features to
324 * lguest kills puppies, so we don't want to.
326 * The bzImage is formed by putting the decompressing code in front of the
327 * compressed kernel code. So we can simple scan through it looking for the
328 * first "gzip" header, and start decompressing from there. */
329 static unsigned long load_bzimage(int fd
, unsigned long *page_offset
)
334 /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */
335 while (read(fd
, &c
, 1) == 1) {
351 /* Seek back to the start of the gzip header. */
352 lseek(fd
, -10, SEEK_CUR
);
353 /* One final check: "compressed under UNIX". */
357 return unpack_bzimage(fd
, page_offset
);
360 errx(1, "Could not find kernel in bzImage");
363 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
364 * come wrapped up in the self-decompressing "bzImage" format. With some funky
365 * coding, we can load those, too. */
366 static unsigned long load_kernel(int fd
, unsigned long *page_offset
)
370 /* Read in the first few bytes. */
371 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
372 err(1, "Reading kernel");
374 /* If it's an ELF file, it starts with "\177ELF" */
375 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
376 return map_elf(fd
, &hdr
, page_offset
);
378 /* Otherwise we assume it's a bzImage, and try to unpack it */
379 return load_bzimage(fd
, page_offset
);
382 /* This is a trivial little helper to align pages. Andi Kleen hated it because
383 * it calls getpagesize() twice: "it's dumb code."
385 * Kernel guys get really het up about optimization, even when it's not
386 * necessary. I leave this code as a reaction against that. */
387 static inline unsigned long page_align(unsigned long addr
)
389 /* Add upwards and truncate downwards. */
390 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
393 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
394 * the kernel which the kernel can use to boot from without needing any
395 * drivers. Most distributions now use this as standard: the initrd contains
396 * the code to load the appropriate driver modules for the current machine.
398 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
399 * kernels. He sent me this (and tells me when I break it). */
400 static unsigned long load_initrd(const char *name
, unsigned long mem
)
407 ifd
= open_or_die(name
, O_RDONLY
);
408 /* fstat() is needed to get the file size. */
409 if (fstat(ifd
, &st
) < 0)
410 err(1, "fstat() on initrd '%s'", name
);
412 /* The length needs to be rounded up to a page size: mmap needs the
413 * address to be page aligned. */
414 len
= page_align(st
.st_size
);
415 /* We map the initrd at the top of memory. */
416 iaddr
= mmap((void *)mem
- len
, st
.st_size
,
417 PROT_READ
|PROT_EXEC
|PROT_WRITE
,
418 MAP_FIXED
|MAP_PRIVATE
, ifd
, 0);
419 if (iaddr
!= (void *)mem
- len
)
420 err(1, "Mmaping initrd '%s' returned %p not %p",
421 name
, iaddr
, (void *)mem
- len
);
422 /* Once a file is mapped, you can close the file descriptor. It's a
423 * little odd, but quite useful. */
425 verbose("mapped initrd %s size=%lu @ %p\n", name
, st
.st_size
, iaddr
);
427 /* We return the initrd size. */
431 /* Once we know how much memory we have, and the address the Guest kernel
432 * expects, we can construct simple linear page tables which will get the Guest
433 * far enough into the boot to create its own.
435 * We lay them out of the way, just below the initrd (which is why we need to
437 static unsigned long setup_pagetables(unsigned long mem
,
438 unsigned long initrd_size
,
439 unsigned long page_offset
)
442 unsigned int mapped_pages
, i
, linear_pages
;
443 unsigned int ptes_per_page
= getpagesize()/sizeof(u32
);
445 /* Ideally we map all physical memory starting at page_offset.
446 * However, if page_offset is 0xC0000000 we can only map 1G of physical
447 * (0xC0000000 + 1G overflows). */
448 if (mem
<= -page_offset
)
449 mapped_pages
= mem
/getpagesize();
451 mapped_pages
= -page_offset
/getpagesize();
453 /* Each PTE page can map ptes_per_page pages: how many do we need? */
454 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
456 /* We put the toplevel page directory page at the top of memory. */
457 pgdir
= (void *)mem
- initrd_size
- getpagesize();
459 /* Now we use the next linear_pages pages as pte pages */
460 linear
= (void *)pgdir
- linear_pages
*getpagesize();
462 /* Linear mapping is easy: put every page's address into the mapping in
463 * order. PAGE_PRESENT contains the flags Present, Writable and
465 for (i
= 0; i
< mapped_pages
; i
++)
466 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
468 /* The top level points to the linear page table pages above. The
469 * entry representing page_offset points to the first one, and they
470 * continue from there. */
471 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
472 pgdir
[(i
+ page_offset
/getpagesize())/ptes_per_page
]
473 = (((u32
)linear
+ i
*sizeof(u32
)) | PAGE_PRESENT
);
476 verbose("Linear mapping of %u pages in %u pte pages at %p\n",
477 mapped_pages
, linear_pages
, linear
);
479 /* We return the top level (guest-physical) address: the kernel needs
480 * to know where it is. */
481 return (unsigned long)pgdir
;
484 /* Simple routine to roll all the commandline arguments together with spaces
486 static void concat(char *dst
, char *args
[])
488 unsigned int i
, len
= 0;
490 for (i
= 0; args
[i
]; i
++) {
491 strcpy(dst
+len
, args
[i
]);
492 strcat(dst
+len
, " ");
493 len
+= strlen(args
[i
]) + 1;
495 /* In case it's empty. */
499 /* This is where we actually tell the kernel to initialize the Guest. We saw
500 * the arguments it expects when we looked at initialize() in lguest_user.c:
501 * the top physical page to allow, the top level pagetable, the entry point and
502 * the page_offset constant for the Guest. */
503 static int tell_kernel(u32 pgdir
, u32 start
, u32 page_offset
)
505 u32 args
[] = { LHREQ_INITIALIZE
,
506 top
/getpagesize(), pgdir
, start
, page_offset
};
509 fd
= open_or_die("/dev/lguest", O_RDWR
);
510 if (write(fd
, args
, sizeof(args
)) < 0)
511 err(1, "Writing to /dev/lguest");
513 /* We return the /dev/lguest file descriptor to control this Guest */
518 static void set_fd(int fd
, struct device_list
*devices
)
520 FD_SET(fd
, &devices
->infds
);
521 if (fd
> devices
->max_infd
)
522 devices
->max_infd
= fd
;
528 * With a console and network devices, we can have lots of input which we need
529 * to process. We could try to tell the kernel what file descriptors to watch,
530 * but handing a file descriptor mask through to the kernel is fairly icky.
532 * Instead, we fork off a process which watches the file descriptors and writes
533 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
534 * loop to stop running the Guest. This causes it to return from the
535 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
536 * the LHREQ_BREAK and wake us up again.
538 * This, of course, is merely a different *kind* of icky.
540 static void wake_parent(int pipefd
, int lguest_fd
, struct device_list
*devices
)
542 /* Add the pipe from the Launcher to the fdset in the device_list, so
543 * we watch it, too. */
544 set_fd(pipefd
, devices
);
547 fd_set rfds
= devices
->infds
;
548 u32 args
[] = { LHREQ_BREAK
, 1 };
550 /* Wait until input is ready from one of the devices. */
551 select(devices
->max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
552 /* Is it a message from the Launcher? */
553 if (FD_ISSET(pipefd
, &rfds
)) {
555 /* If read() returns 0, it means the Launcher has
556 * exited. We silently follow. */
557 if (read(pipefd
, &ignorefd
, sizeof(ignorefd
)) == 0)
559 /* Otherwise it's telling us there's a problem with one
560 * of the devices, and we should ignore that file
561 * descriptor from now on. */
562 FD_CLR(ignorefd
, &devices
->infds
);
563 } else /* Send LHREQ_BREAK command. */
564 write(lguest_fd
, args
, sizeof(args
));
568 /* This routine just sets up a pipe to the Waker process. */
569 static int setup_waker(int lguest_fd
, struct device_list
*device_list
)
571 int pipefd
[2], child
;
573 /* We create a pipe to talk to the waker, and also so it knows when the
574 * Launcher dies (and closes pipe). */
581 /* Close the "writing" end of our copy of the pipe */
583 wake_parent(pipefd
[0], lguest_fd
, device_list
);
585 /* Close the reading end of our copy of the pipe. */
588 /* Here is the fd used to talk to the waker. */
595 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
596 * We need to make sure it's not trying to reach into the Launcher itself, so
597 * we have a convenient routine which check it and exits with an error message
598 * if something funny is going on:
600 static void *_check_pointer(unsigned long addr
, unsigned int size
,
603 /* We have to separately check addr and addr+size, because size could
604 * be huge and addr + size might wrap around. */
605 if (addr
>= top
|| addr
+ size
>= top
)
606 errx(1, "%s:%i: Invalid address %li", __FILE__
, line
, addr
);
607 /* We return a pointer for the caller's convenience, now we know it's
611 /* A macro which transparently hands the line number to the real function. */
612 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
614 /* The Guest has given us the address of a "struct lguest_dma". We check it's
615 * OK and convert it to an iovec (which is a simple array of ptr/size
617 static u32
*dma2iov(unsigned long dma
, struct iovec iov
[], unsigned *num
)
620 struct lguest_dma
*udma
;
622 /* First we make sure that the array memory itself is valid. */
623 udma
= check_pointer(dma
, sizeof(*udma
));
624 /* Now we check each element */
625 for (i
= 0; i
< LGUEST_MAX_DMA_SECTIONS
; i
++) {
626 /* A zero length ends the array. */
630 iov
[i
].iov_base
= check_pointer(udma
->addr
[i
], udma
->len
[i
]);
631 iov
[i
].iov_len
= udma
->len
[i
];
635 /* We return the pointer to where the caller should write the amount of
636 * the buffer used. */
637 return &udma
->used_len
;
640 /* This routine gets a DMA buffer from the Guest for a given key, and converts
641 * it to an iovec array. It returns the interrupt the Guest wants when we're
642 * finished, and a pointer to the "used_len" field to fill in. */
643 static u32
*get_dma_buffer(int fd
, void *key
,
644 struct iovec iov
[], unsigned int *num
, u32
*irq
)
646 u32 buf
[] = { LHREQ_GETDMA
, (u32
)key
};
650 /* Ask the kernel for a DMA buffer corresponding to this key. */
651 udma
= write(fd
, buf
, sizeof(buf
));
652 /* They haven't registered any, or they're all used? */
653 if (udma
== (unsigned long)-1)
656 /* Convert it into our iovec array */
657 res
= dma2iov(udma
, iov
, num
);
658 /* The kernel stashes irq in ->used_len to get it out to us. */
660 /* Return a pointer to ((struct lguest_dma *)udma)->used_len. */
664 /* This is a convenient routine to send the Guest an interrupt. */
665 static void trigger_irq(int fd
, u32 irq
)
667 u32 buf
[] = { LHREQ_IRQ
, irq
};
668 if (write(fd
, buf
, sizeof(buf
)) != 0)
669 err(1, "Triggering irq %i", irq
);
672 /* This simply sets up an iovec array where we can put data to be discarded.
673 * This happens when the Guest doesn't want or can't handle the input: we have
674 * to get rid of it somewhere, and if we bury it in the ceiling space it will
675 * start to smell after a week. */
676 static void discard_iovec(struct iovec
*iov
, unsigned int *num
)
678 static char discard_buf
[1024];
680 iov
->iov_base
= discard_buf
;
681 iov
->iov_len
= sizeof(discard_buf
);
684 /* Here is the input terminal setting we save, and the routine to restore them
685 * on exit so the user can see what they type next. */
686 static struct termios orig_term
;
687 static void restore_term(void)
689 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
692 /* We associate some data with the console for our exit hack. */
695 /* How many times have they hit ^C? */
697 /* When did they start? */
698 struct timeval start
;
701 /* This is the routine which handles console input (ie. stdin). */
702 static bool handle_console_input(int fd
, struct device
*dev
)
707 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
708 struct console_abort
*abort
= dev
->priv
;
710 /* First we get the console buffer from the Guest. The key is dev->mem
711 * which was set to 0 in setup_console(). */
712 lenp
= get_dma_buffer(fd
, dev
->mem
, iov
, &num
, &irq
);
714 /* If it's not ready for input, warn and set up to discard. */
715 warn("console: no dma buffer!");
716 discard_iovec(iov
, &num
);
719 /* This is why we convert to iovecs: the readv() call uses them, and so
720 * it reads straight into the Guest's buffer. */
721 len
= readv(dev
->fd
, iov
, num
);
723 /* This implies that the console is closed, is /dev/null, or
724 * something went terribly wrong. We still go through the rest
725 * of the logic, though, especially the exit handling below. */
726 warnx("Failed to get console input, ignoring console.");
730 /* If we read the data into the Guest, fill in the length and send the
734 trigger_irq(fd
, irq
);
737 /* Three ^C within one second? Exit.
739 * This is such a hack, but works surprisingly well. Each ^C has to be
740 * in a buffer by itself, so they can't be too fast. But we check that
741 * we get three within about a second, so they can't be too slow. */
742 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
744 gettimeofday(&abort
->start
, NULL
);
745 else if (abort
->count
== 3) {
747 gettimeofday(&now
, NULL
);
748 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
749 u32 args
[] = { LHREQ_BREAK
, 0 };
750 /* Close the fd so Waker will know it has to
753 /* Just in case waker is blocked in BREAK, send
755 write(fd
, args
, sizeof(args
));
761 /* Any other key resets the abort counter. */
764 /* Now, if we didn't read anything, put the input terminal back and
765 * return failure (meaning, don't call us again). */
770 /* Everything went OK! */
774 /* Handling console output is much simpler than input. */
775 static u32
handle_console_output(int fd
, const struct iovec
*iov
,
776 unsigned num
, struct device
*dev
)
778 /* Whatever the Guest sends, write it to standard output. Return the
779 * number of bytes written. */
780 return writev(STDOUT_FILENO
, iov
, num
);
783 /* Guest->Host network output is also pretty easy. */
784 static u32
handle_tun_output(int fd
, const struct iovec
*iov
,
785 unsigned num
, struct device
*dev
)
787 /* We put a flag in the "priv" pointer of the network device, and set
788 * it as soon as we see output. We'll see why in handle_tun_input() */
789 *(bool *)dev
->priv
= true;
790 /* Whatever packet the Guest sent us, write it out to the tun
792 return writev(dev
->fd
, iov
, num
);
795 /* This matches the peer_key() in lguest_net.c. The key for any given slot
796 * is the address of the network device's page plus 4 * the slot number. */
797 static unsigned long peer_offset(unsigned int peernum
)
802 /* This is where we handle a packet coming in from the tun device */
803 static bool handle_tun_input(int fd
, struct device
*dev
)
808 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
810 /* First we get a buffer the Guest has bound to its key. */
811 lenp
= get_dma_buffer(fd
, dev
->mem
+peer_offset(NET_PEERNUM
), iov
, &num
,
814 /* Now, it's expected that if we try to send a packet too
815 * early, the Guest won't be ready yet. This is why we set a
816 * flag when the Guest sends its first packet. If it's sent a
817 * packet we assume it should be ready to receive them.
819 * Actually, this is what the status bits in the descriptor are
820 * for: we should *use* them. FIXME! */
821 if (*(bool *)dev
->priv
)
822 warn("network: no dma buffer!");
823 discard_iovec(iov
, &num
);
826 /* Read the packet from the device directly into the Guest's buffer. */
827 len
= readv(dev
->fd
, iov
, num
);
829 err(1, "reading network");
831 /* Write the used_len, and trigger the interrupt for the Guest */
834 trigger_irq(fd
, irq
);
836 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
837 ((u8
*)iov
[0].iov_base
)[0], ((u8
*)iov
[0].iov_base
)[1],
838 lenp
? "sent" : "discarded");
843 /* The last device handling routine is block output: the Guest has sent a DMA
844 * to the block device. It will have placed the command it wants in the
845 * "struct lguest_block_page". */
846 static u32
handle_block_output(int fd
, const struct iovec
*iov
,
847 unsigned num
, struct device
*dev
)
849 struct lguest_block_page
*p
= dev
->mem
;
851 unsigned int len
, reply_num
;
852 struct iovec reply
[LGUEST_MAX_DMA_SECTIONS
];
853 off64_t device_len
, off
= (off64_t
)p
->sector
* 512;
855 /* First we extract the device length from the dev->priv pointer. */
856 device_len
= *(off64_t
*)dev
->priv
;
858 /* We first check that the read or write is within the length of the
860 if (off
>= device_len
)
861 err(1, "Bad offset %llu vs %llu", off
, device_len
);
862 /* Move to the right location in the block file. This shouldn't fail,
863 * but best to check. */
864 if (lseek64(dev
->fd
, off
, SEEK_SET
) != off
)
865 err(1, "Bad seek to sector %i", p
->sector
);
867 verbose("Block: %s at offset %llu\n", p
->type
? "WRITE" : "READ", off
);
869 /* They were supposed to bind a reply buffer at key equal to the start
870 * of the block device memory. We need this to tell them when the
871 * request is finished. */
872 lenp
= get_dma_buffer(fd
, dev
->mem
, reply
, &reply_num
, &irq
);
874 err(1, "Block request didn't give us a dma buffer");
877 /* A write request. The DMA they sent contained the data, so
879 len
= writev(dev
->fd
, iov
, num
);
880 /* Grr... Now we know how long the "struct lguest_dma" they
881 * sent was, we make sure they didn't try to write over the end
882 * of the block file (possibly extending it). */
883 if (off
+ len
> device_len
) {
884 /* Trim it back to the correct length */
885 ftruncate64(dev
->fd
, device_len
);
886 /* Die, bad Guest, die. */
887 errx(1, "Write past end %llu+%u", off
, len
);
889 /* The reply length is 0: we just send back an empty DMA to
890 * interrupt them and tell them the write is finished. */
893 /* A read request. They sent an empty DMA to start the
894 * request, and we put the read contents into the reply
896 len
= readv(dev
->fd
, reply
, reply_num
);
900 /* The result is 1 (done), 2 if there was an error (short read or
902 p
->result
= 1 + (p
->bytes
!= len
);
903 /* Now tell them we've used their reply buffer. */
904 trigger_irq(fd
, irq
);
906 /* We're supposed to return the number of bytes of the output buffer we
907 * used. But the block device uses the "result" field instead, so we
912 /* This is the generic routine we call when the Guest sends some DMA out. */
913 static void handle_output(int fd
, unsigned long dma
, unsigned long key
,
914 struct device_list
*devices
)
918 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
921 /* Convert the "struct lguest_dma" they're sending to a "struct
923 lenp
= dma2iov(dma
, iov
, &num
);
925 /* Check each device: if they expect output to this key, tell them to
927 for (i
= devices
->dev
; i
; i
= i
->next
) {
928 if (i
->handle_output
&& key
== i
->watch_key
) {
929 /* We write the result straight into the used_len field
931 *lenp
= i
->handle_output(fd
, iov
, num
, i
);
936 /* This can happen: the kernel sends any SEND_DMA which doesn't match
937 * another Guest to us. It could be that another Guest just left a
938 * network, for example. But it's unusual. */
939 warnx("Pending dma %p, key %p", (void *)dma
, (void *)key
);
942 /* This is called when the waker wakes us up: check for incoming file
944 static void handle_input(int fd
, struct device_list
*devices
)
946 /* select() wants a zeroed timeval to mean "don't wait". */
947 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
951 fd_set fds
= devices
->infds
;
953 /* If nothing is ready, we're done. */
954 if (select(devices
->max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
957 /* Otherwise, call the device(s) which have readable
958 * file descriptors and a method of handling them. */
959 for (i
= devices
->dev
; i
; i
= i
->next
) {
960 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
961 /* If handle_input() returns false, it means we
962 * should no longer service it.
963 * handle_console_input() does this. */
964 if (!i
->handle_input(fd
, i
)) {
965 /* Clear it from the set of input file
966 * descriptors kept at the head of the
968 FD_CLR(i
->fd
, &devices
->infds
);
969 /* Tell waker to ignore it too... */
970 write(waker_fd
, &i
->fd
, sizeof(i
->fd
));
980 * All devices need a descriptor so the Guest knows it exists, and a "struct
981 * device" so the Launcher can keep track of it. We have common helper
982 * routines to allocate them.
984 * This routine allocates a new "struct lguest_device_desc" from descriptor
985 * table in the devices array just above the Guest's normal memory. */
986 static struct lguest_device_desc
*
987 new_dev_desc(struct lguest_device_desc
*descs
,
988 u16 type
, u16 features
, u16 num_pages
)
992 for (i
= 0; i
< LGUEST_MAX_DEVICES
; i
++) {
993 if (!descs
[i
].type
) {
994 descs
[i
].type
= type
;
995 descs
[i
].features
= features
;
996 descs
[i
].num_pages
= num_pages
;
997 /* If they said the device needs memory, we allocate
998 * that now, bumping up the top of Guest memory. */
1000 map_zeroed_pages(top
, num_pages
);
1001 descs
[i
].pfn
= top
/getpagesize();
1002 top
+= num_pages
*getpagesize();
1007 errx(1, "too many devices");
1010 /* This monster routine does all the creation and setup of a new device,
1011 * including caling new_dev_desc() to allocate the descriptor and device
1013 static struct device
*new_device(struct device_list
*devices
,
1014 u16 type
, u16 num_pages
, u16 features
,
1016 bool (*handle_input
)(int, struct device
*),
1017 unsigned long watch_off
,
1018 u32 (*handle_output
)(int,
1019 const struct iovec
*,
1023 struct device
*dev
= malloc(sizeof(*dev
));
1025 /* Append to device list. Prepending to a single-linked list is
1026 * easier, but the user expects the devices to be arranged on the bus
1027 * in command-line order. The first network device on the command line
1028 * is eth0, the first block device /dev/lgba, etc. */
1029 *devices
->lastdev
= dev
;
1031 devices
->lastdev
= &dev
->next
;
1033 /* Now we populate the fields one at a time. */
1035 /* If we have an input handler for this file descriptor, then we add it
1036 * to the device_list's fdset and maxfd. */
1038 set_fd(dev
->fd
, devices
);
1039 dev
->desc
= new_dev_desc(devices
->descs
, type
, features
, num_pages
);
1040 dev
->mem
= (void *)(dev
->desc
->pfn
* getpagesize());
1041 dev
->handle_input
= handle_input
;
1042 dev
->watch_key
= (unsigned long)dev
->mem
+ watch_off
;
1043 dev
->handle_output
= handle_output
;
1047 /* Our first setup routine is the console. It's a fairly simple device, but
1048 * UNIX tty handling makes it uglier than it could be. */
1049 static void setup_console(struct device_list
*devices
)
1053 /* If we can save the initial standard input settings... */
1054 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1055 struct termios term
= orig_term
;
1056 /* Then we turn off echo, line buffering and ^C etc. We want a
1057 * raw input stream to the Guest. */
1058 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1059 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1060 /* If we exit gracefully, the original settings will be
1061 * restored so the user can see what they're typing. */
1062 atexit(restore_term
);
1065 /* We don't currently require any memory for the console, so we ask for
1067 dev
= new_device(devices
, LGUEST_DEVICE_T_CONSOLE
, 0, 0,
1068 STDIN_FILENO
, handle_console_input
,
1069 LGUEST_CONSOLE_DMA_KEY
, handle_console_output
);
1070 /* We store the console state in dev->priv, and initialize it. */
1071 dev
->priv
= malloc(sizeof(struct console_abort
));
1072 ((struct console_abort
*)dev
->priv
)->count
= 0;
1073 verbose("device %p: console\n",
1074 (void *)(dev
->desc
->pfn
* getpagesize()));
1077 /* Setting up a block file is also fairly straightforward. */
1078 static void setup_block_file(const char *filename
, struct device_list
*devices
)
1082 off64_t
*device_len
;
1083 struct lguest_block_page
*p
;
1085 /* We open with O_LARGEFILE because otherwise we get stuck at 2G. We
1086 * open with O_DIRECT because otherwise our benchmarks go much too
1088 fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
|O_DIRECT
);
1090 /* We want one page, and have no input handler (the block file never
1091 * has anything interesting to say to us). Our timing will be quite
1092 * random, so it should be a reasonable randomness source. */
1093 dev
= new_device(devices
, LGUEST_DEVICE_T_BLOCK
, 1,
1094 LGUEST_DEVICE_F_RANDOMNESS
,
1095 fd
, NULL
, 0, handle_block_output
);
1097 /* We store the device size in the private area */
1098 device_len
= dev
->priv
= malloc(sizeof(*device_len
));
1099 /* This is the safe way of establishing the size of our device: it
1100 * might be a normal file or an actual block device like /dev/hdb. */
1101 *device_len
= lseek64(fd
, 0, SEEK_END
);
1103 /* The device memory is a "struct lguest_block_page". It's zeroed
1104 * already, we just need to put in the device size. Block devices
1105 * think in sectors (ie. 512 byte chunks), so we translate here. */
1107 p
->num_sectors
= *device_len
/512;
1108 verbose("device %p: block %i sectors\n",
1109 (void *)(dev
->desc
->pfn
* getpagesize()), p
->num_sectors
);
1115 * Setting up network devices is quite a pain, because we have three types.
1116 * First, we have the inter-Guest network. This is a file which is mapped into
1117 * the address space of the Guests who are on the network. Because it is a
1118 * shared mapping, the same page underlies all the devices, and they can send
1119 * DMA to each other.
1121 * Remember from our network driver, the Guest is told what slot in the page it
1122 * is to use. We use exclusive fnctl locks to reserve a slot. If another
1123 * Guest is using a slot, the lock will fail and we try another. Because fnctl
1124 * locks are cleaned up automatically when we die, this cleverly means that our
1125 * reservation on the slot will vanish if we crash. */
1126 static unsigned int find_slot(int netfd
, const char *filename
)
1130 fl
.l_type
= F_WRLCK
;
1131 fl
.l_whence
= SEEK_SET
;
1133 /* Try a 1 byte lock in each possible position number */
1134 for (fl
.l_start
= 0;
1135 fl
.l_start
< getpagesize()/sizeof(struct lguest_net
);
1137 /* If we succeed, return the slot number. */
1138 if (fcntl(netfd
, F_SETLK
, &fl
) == 0)
1141 errx(1, "No free slots in network file %s", filename
);
1144 /* This function sets up the network file */
1145 static void setup_net_file(const char *filename
,
1146 struct device_list
*devices
)
1151 /* We don't use open_or_die() here: for friendliness we create the file
1152 * if it doesn't already exist. */
1153 netfd
= open(filename
, O_RDWR
, 0);
1155 if (errno
== ENOENT
) {
1156 netfd
= open(filename
, O_RDWR
|O_CREAT
, 0600);
1158 /* If we succeeded, initialize the file with a
1160 char page
[getpagesize()];
1161 memset(page
, 0, sizeof(page
));
1162 write(netfd
, page
, sizeof(page
));
1166 err(1, "cannot open net file '%s'", filename
);
1169 /* We need 1 page, and the features indicate the slot to use and that
1170 * no checksum is needed. We never touch this device again; it's
1171 * between the Guests on the network, so we don't register input or
1172 * output handlers. */
1173 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1174 find_slot(netfd
, filename
)|LGUEST_NET_F_NOCSUM
,
1177 /* Map the shared file. */
1178 if (mmap(dev
->mem
, getpagesize(), PROT_READ
|PROT_WRITE
,
1179 MAP_FIXED
|MAP_SHARED
, netfd
, 0) != dev
->mem
)
1180 err(1, "could not mmap '%s'", filename
);
1181 verbose("device %p: shared net %s, peer %i\n",
1182 (void *)(dev
->desc
->pfn
* getpagesize()), filename
,
1183 dev
->desc
->features
& ~LGUEST_NET_F_NOCSUM
);
1187 static u32
str2ip(const char *ipaddr
)
1189 unsigned int byte
[4];
1191 sscanf(ipaddr
, "%u.%u.%u.%u", &byte
[0], &byte
[1], &byte
[2], &byte
[3]);
1192 return (byte
[0] << 24) | (byte
[1] << 16) | (byte
[2] << 8) | byte
[3];
1195 /* This code is "adapted" from libbridge: it attaches the Host end of the
1196 * network device to the bridge device specified by the command line.
1198 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1199 * dislike bridging), and I just try not to break it. */
1200 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1206 errx(1, "must specify bridge name");
1208 ifidx
= if_nametoindex(if_name
);
1210 errx(1, "interface %s does not exist!", if_name
);
1212 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1213 ifr
.ifr_ifindex
= ifidx
;
1214 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1215 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1218 /* This sets up the Host end of the network device with an IP address, brings
1219 * it up so packets will flow, the copies the MAC address into the hwaddr
1220 * pointer (in practice, the Host's slot in the network device's memory). */
1221 static void configure_device(int fd
, const char *devname
, u32 ipaddr
,
1222 unsigned char hwaddr
[6])
1225 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1227 /* Don't read these incantations. Just cut & paste them like I did! */
1228 memset(&ifr
, 0, sizeof(ifr
));
1229 strcpy(ifr
.ifr_name
, devname
);
1230 sin
->sin_family
= AF_INET
;
1231 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1232 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1233 err(1, "Setting %s interface address", devname
);
1234 ifr
.ifr_flags
= IFF_UP
;
1235 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1236 err(1, "Bringing interface %s up", devname
);
1238 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1239 * above). IF means Interface, and HWADDR is hardware address.
1241 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1242 err(1, "getting hw address for %s", devname
);
1243 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1246 /*L:195 The other kind of network is a Host<->Guest network. This can either
1247 * use briding or routing, but the principle is the same: it uses the "tun"
1248 * device to inject packets into the Host as if they came in from a normal
1249 * network card. We just shunt packets between the Guest and the tun
1251 static void setup_tun_net(const char *arg
, struct device_list
*devices
)
1257 const char *br_name
= NULL
;
1259 /* We open the /dev/net/tun device and tell it we want a tap device. A
1260 * tap device is like a tun device, only somehow different. To tell
1261 * the truth, I completely blundered my way through this code, but it
1263 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1264 memset(&ifr
, 0, sizeof(ifr
));
1265 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1266 strcpy(ifr
.ifr_name
, "tap%d");
1267 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1268 err(1, "configuring /dev/net/tun");
1269 /* We don't need checksums calculated for packets coming in this
1270 * device: trust us! */
1271 ioctl(netfd
, TUNSETNOCSUM
, 1);
1273 /* We create the net device with 1 page, using the features field of
1274 * the descriptor to tell the Guest it is in slot 1 (NET_PEERNUM), and
1275 * that the device has fairly random timing. We do *not* specify
1276 * LGUEST_NET_F_NOCSUM: these packets can reach the real world.
1278 * We will put our MAC address is slot 0 for the Guest to see, so
1279 * it will send packets to us using the key "peer_offset(0)": */
1280 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1281 NET_PEERNUM
|LGUEST_DEVICE_F_RANDOMNESS
, netfd
,
1282 handle_tun_input
, peer_offset(0), handle_tun_output
);
1284 /* We keep a flag which says whether we've seen packets come out from
1285 * this network device. */
1286 dev
->priv
= malloc(sizeof(bool));
1287 *(bool *)dev
->priv
= false;
1289 /* We need a socket to perform the magic network ioctls to bring up the
1290 * tap interface, connect to the bridge etc. Any socket will do! */
1291 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1293 err(1, "opening IP socket");
1295 /* If the command line was --tunnet=bridge:<name> do bridging. */
1296 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1298 br_name
= arg
+ strlen(BRIDGE_PFX
);
1299 add_to_bridge(ipfd
, ifr
.ifr_name
, br_name
);
1300 } else /* It is an IP address to set up the device with */
1303 /* We are peer 0, ie. first slot, so we hand dev->mem to this routine
1304 * to write the MAC address at the start of the device memory. */
1305 configure_device(ipfd
, ifr
.ifr_name
, ip
, dev
->mem
);
1307 /* Set "promisc" bit: we want every single packet if we're going to
1308 * bridge to other machines (and otherwise it doesn't matter). */
1309 *((u8
*)dev
->mem
) |= 0x1;
1313 verbose("device %p: tun net %u.%u.%u.%u\n",
1314 (void *)(dev
->desc
->pfn
* getpagesize()),
1315 (u8
)(ip
>>24), (u8
)(ip
>>16), (u8
)(ip
>>8), (u8
)ip
);
1317 verbose("attached to bridge: %s\n", br_name
);
1319 /* That's the end of device setup. */
1321 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1322 * its input and output, and finally, lays it to rest. */
1323 static void __attribute__((noreturn
))
1324 run_guest(int lguest_fd
, struct device_list
*device_list
)
1327 u32 args
[] = { LHREQ_BREAK
, 0 };
1328 unsigned long arr
[2];
1331 /* We read from the /dev/lguest device to run the Guest. */
1332 readval
= read(lguest_fd
, arr
, sizeof(arr
));
1334 /* The read can only really return sizeof(arr) (the Guest did a
1335 * SEND_DMA to us), or an error. */
1337 /* For a successful read, arr[0] is the address of the "struct
1338 * lguest_dma", and arr[1] is the key the Guest sent to. */
1339 if (readval
== sizeof(arr
)) {
1340 handle_output(lguest_fd
, arr
[0], arr
[1], device_list
);
1342 /* ENOENT means the Guest died. Reading tells us why. */
1343 } else if (errno
== ENOENT
) {
1344 char reason
[1024] = { 0 };
1345 read(lguest_fd
, reason
, sizeof(reason
)-1);
1346 errx(1, "%s", reason
);
1347 /* EAGAIN means the waker wanted us to look at some input.
1348 * Anything else means a bug or incompatible change. */
1349 } else if (errno
!= EAGAIN
)
1350 err(1, "Running guest failed");
1352 /* Service input, then unset the BREAK which releases
1354 handle_input(lguest_fd
, device_list
);
1355 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
1356 err(1, "Resetting break");
1360 * This is the end of the Launcher.
1362 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1363 * Drivers. If we were to see the Host kernel I/O code, our understanding
1364 * would be complete... :*/
1366 static struct option opts
[] = {
1367 { "verbose", 0, NULL
, 'v' },
1368 { "sharenet", 1, NULL
, 's' },
1369 { "tunnet", 1, NULL
, 't' },
1370 { "block", 1, NULL
, 'b' },
1371 { "initrd", 1, NULL
, 'i' },
1374 static void usage(void)
1376 errx(1, "Usage: lguest [--verbose] "
1377 "[--sharenet=<filename>|--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1378 "|--block=<filename>|--initrd=<filename>]...\n"
1379 "<mem-in-mb> vmlinux [args...]");
1382 /*L:100 The Launcher code itself takes us out into userspace, that scary place
1383 * where pointers run wild and free! Unfortunately, like most userspace
1384 * programs, it's quite boring (which is why everyone like to hack on the
1385 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
1386 * will get you through this section. Or, maybe not.
1388 * The Launcher binary sits up high, usually starting at address 0xB8000000.
1389 * Everything below this is the "physical" memory for the Guest. For example,
1390 * if the Guest were to write a "1" at physical address 0, we would see a "1"
1391 * in the Launcher at "(int *)0". Guest physical == Launcher virtual.
1393 * This can be tough to get your head around, but usually it just means that we
1394 * don't need to do any conversion when the Guest gives us it's "physical"
1397 int main(int argc
, char *argv
[])
1399 /* Memory, top-level pagetable, code startpoint, PAGE_OFFSET and size
1400 * of the (optional) initrd. */
1401 unsigned long mem
= 0, pgdir
, start
, page_offset
, initrd_size
= 0;
1402 /* A temporary and the /dev/lguest file descriptor. */
1403 int i
, c
, lguest_fd
;
1404 /* The list of Guest devices, based on command line arguments. */
1405 struct device_list device_list
;
1406 /* The boot information for the Guest: at guest-physical address 0. */
1407 void *boot
= (void *)0;
1408 /* If they specify an initrd file to load. */
1409 const char *initrd_name
= NULL
;
1411 /* First we initialize the device list. Since console and network
1412 * device receive input from a file descriptor, we keep an fdset
1413 * (infds) and the maximum fd number (max_infd) with the head of the
1414 * list. We also keep a pointer to the last device, for easy appending
1416 device_list
.max_infd
= -1;
1417 device_list
.dev
= NULL
;
1418 device_list
.lastdev
= &device_list
.dev
;
1419 FD_ZERO(&device_list
.infds
);
1421 /* We need to know how much memory so we can set up the device
1422 * descriptor and memory pages for the devices as we parse the command
1423 * line. So we quickly look through the arguments to find the amount
1425 for (i
= 1; i
< argc
; i
++) {
1426 if (argv
[i
][0] != '-') {
1427 mem
= top
= atoi(argv
[i
]) * 1024 * 1024;
1428 device_list
.descs
= map_zeroed_pages(top
, 1);
1429 top
+= getpagesize();
1434 /* The options are fairly straight-forward */
1435 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1441 setup_net_file(optarg
, &device_list
);
1444 setup_tun_net(optarg
, &device_list
);
1447 setup_block_file(optarg
, &device_list
);
1450 initrd_name
= optarg
;
1453 warnx("Unknown argument %s", argv
[optind
]);
1457 /* After the other arguments we expect memory and kernel image name,
1458 * followed by command line arguments for the kernel. */
1459 if (optind
+ 2 > argc
)
1462 /* We always have a console device */
1463 setup_console(&device_list
);
1465 /* We start by mapping anonymous pages over all of guest-physical
1466 * memory range. This fills it with 0, and ensures that the Guest
1467 * won't be killed when it tries to access it. */
1468 map_zeroed_pages(0, mem
/ getpagesize());
1470 /* Now we load the kernel */
1471 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
),
1474 /* Map the initrd image if requested (at top of physical memory) */
1476 initrd_size
= load_initrd(initrd_name
, mem
);
1477 /* These are the location in the Linux boot header where the
1478 * start and size of the initrd are expected to be found. */
1479 *(unsigned long *)(boot
+0x218) = mem
- initrd_size
;
1480 *(unsigned long *)(boot
+0x21c) = initrd_size
;
1481 /* The bootloader type 0xFF means "unknown"; that's OK. */
1482 *(unsigned char *)(boot
+0x210) = 0xFF;
1485 /* Set up the initial linear pagetables, starting below the initrd. */
1486 pgdir
= setup_pagetables(mem
, initrd_size
, page_offset
);
1488 /* The Linux boot header contains an "E820" memory map: ours is a
1489 * simple, single region. */
1490 *(char*)(boot
+E820NR
) = 1;
1491 *((struct e820entry
*)(boot
+E820MAP
))
1492 = ((struct e820entry
) { 0, mem
, E820_RAM
});
1493 /* The boot header contains a command line pointer: we put the command
1494 * line after the boot header (at address 4096) */
1495 *(void **)(boot
+ 0x228) = boot
+ 4096;
1496 concat(boot
+ 4096, argv
+optind
+2);
1498 /* The guest type value of "1" tells the Guest it's under lguest. */
1499 *(int *)(boot
+ 0x23c) = 1;
1501 /* We tell the kernel to initialize the Guest: this returns the open
1502 * /dev/lguest file descriptor. */
1503 lguest_fd
= tell_kernel(pgdir
, start
, page_offset
);
1505 /* We fork off a child process, which wakes the Launcher whenever one
1506 * of the input file descriptors needs attention. Otherwise we would
1507 * run the Guest until it tries to output something. */
1508 waker_fd
= setup_waker(lguest_fd
, &device_list
);
1510 /* Finally, run the Guest. This doesn't return. */
1511 run_guest(lguest_fd
, &device_list
);
1516 * Mastery is done: you now know everything I do.
1518 * But surely you have seen code, features and bugs in your wanderings which
1519 * you now yearn to attack? That is the real game, and I look forward to you
1520 * patching and forking lguest into the Your-Name-Here-visor.
1522 * Farewell, and good coding!