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/param.h>
19 #include <sys/types.h>
26 #include <sys/socket.h>
27 #include <sys/ioctl.h>
30 #include <netinet/in.h>
32 #include <linux/sockios.h>
33 #include <linux/if_tun.h>
38 /*L:110 We can ignore the 28 include files we need for this program, but I do
39 * want to draw attention to the use of kernel-style types.
41 * As Linus said, "C is a Spartan language, and so should your naming be." I
42 * like these abbreviations and the header we need uses them, so we define them
45 typedef unsigned long long u64
;
49 #include "linux/lguest_launcher.h"
50 #include "asm-x86/e820.h"
53 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
55 #define BRIDGE_PFX "bridge:"
57 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
60 /*L:120 verbose is both a global flag and a macro. The C preprocessor allows
61 * this, and although I wouldn't recommend it, it works quite nicely here. */
63 #define verbose(args...) \
64 do { if (verbose) printf(args); } while(0)
67 /* The pipe to send commands to the waker process */
69 /* The top of guest physical memory. */
72 /* This is our list of devices. */
75 /* Summary information about the devices in our list: ready to pass to
76 * select() to ask which need servicing.*/
80 /* The descriptor page for the devices. */
81 struct lguest_device_desc
*descs
;
83 /* A single linked list of devices. */
85 /* ... And an end pointer so we can easily append new devices */
86 struct device
**lastdev
;
89 /* The device structure describes a single device. */
92 /* The linked-list pointer. */
94 /* The descriptor for this device, as mapped into the Guest. */
95 struct lguest_device_desc
*desc
;
96 /* The memory page(s) of this device, if any. Also mapped in Guest. */
99 /* If handle_input is set, it wants to be called when this file
100 * descriptor is ready. */
102 bool (*handle_input
)(int fd
, struct device
*me
);
104 /* If handle_output is set, it wants to be called when the Guest sends
105 * DMA to this key. */
106 unsigned long watch_key
;
107 u32 (*handle_output
)(int fd
, const struct iovec
*iov
,
108 unsigned int num
, struct device
*me
);
110 /* Device-specific data. */
115 * Loading the Kernel.
117 * We start with couple of simple helper routines. open_or_die() avoids
118 * error-checking code cluttering the callers: */
119 static int open_or_die(const char *name
, int flags
)
121 int fd
= open(name
, flags
);
123 err(1, "Failed to open %s", name
);
127 /* map_zeroed_pages() takes a (page-aligned) address and a number of pages. */
128 static void *map_zeroed_pages(unsigned long addr
, unsigned int num
)
130 /* We cache the /dev/zero file-descriptor so we only open it once. */
134 fd
= open_or_die("/dev/zero", O_RDONLY
);
136 /* We use a private mapping (ie. if we write to the page, it will be
137 * copied), and obviously we insist that it be mapped where we ask. */
138 if (mmap((void *)addr
, getpagesize() * num
,
139 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_FIXED
|MAP_PRIVATE
, fd
, 0)
141 err(1, "Mmaping %u pages of /dev/zero @%p", num
, (void *)addr
);
143 /* Returning the address is just a courtesy: can simplify callers. */
147 /* To find out where to start we look for the magic Guest string, which marks
148 * the code we see in lguest_asm.S. This is a hack which we are currently
149 * plotting to replace with the normal Linux entry point. */
150 static unsigned long entry_point(void *start
, void *end
,
151 unsigned long page_offset
)
155 /* The scan gives us the physical starting address. We want the
156 * virtual address in this case, and fortunately, we already figured
157 * out the physical-virtual difference and passed it here in
159 for (p
= start
; p
< end
; p
++)
160 if (memcmp(p
, "GenuineLguest", strlen("GenuineLguest")) == 0)
161 return (long)p
+ strlen("GenuineLguest") + page_offset
;
163 errx(1, "Is this image a genuine lguest?");
166 /* This routine is used to load the kernel or initrd. It tries mmap, but if
167 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
168 * it falls back to reading the memory in. */
169 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
173 /* We map writable even though for some segments are marked read-only.
174 * The kernel really wants to be writable: it patches its own
177 * MAP_PRIVATE means that the page won't be copied until a write is
178 * done to it. This allows us to share untouched memory between
180 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
181 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
184 /* pread does a seek and a read in one shot: saves a few lines. */
185 r
= pread(fd
, addr
, len
, offset
);
187 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
190 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
191 * the Guest memory. ELF = Embedded Linking Format, which is the format used
192 * by all modern binaries on Linux including the kernel.
194 * The ELF headers give *two* addresses: a physical address, and a virtual
195 * address. The Guest kernel expects to be placed in memory at the physical
196 * address, and the page tables set up so it will correspond to that virtual
197 * address. We return the difference between the virtual and physical
198 * addresses in the "page_offset" pointer.
200 * We return the starting address. */
201 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
,
202 unsigned long *page_offset
)
204 Elf32_Phdr phdr
[ehdr
->e_phnum
];
206 unsigned long start
= -1UL, end
= 0;
208 /* Sanity checks on the main ELF header: an x86 executable with a
209 * reasonable number of correctly-sized program headers. */
210 if (ehdr
->e_type
!= ET_EXEC
211 || ehdr
->e_machine
!= EM_386
212 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
213 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
214 errx(1, "Malformed elf header");
216 /* An ELF executable contains an ELF header and a number of "program"
217 * headers which indicate which parts ("segments") of the program to
220 /* We read in all the program headers at once: */
221 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
222 err(1, "Seeking to program headers");
223 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
224 err(1, "Reading program headers");
226 /* We don't know page_offset yet. */
229 /* Try all the headers: there are usually only three. A read-only one,
230 * a read-write one, and a "note" section which isn't loadable. */
231 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
232 /* If this isn't a loadable segment, we ignore it */
233 if (phdr
[i
].p_type
!= PT_LOAD
)
236 verbose("Section %i: size %i addr %p\n",
237 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
239 /* We expect a simple linear address space: every segment must
240 * have the same difference between virtual (p_vaddr) and
241 * physical (p_paddr) address. */
243 *page_offset
= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
;
244 else if (*page_offset
!= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
)
245 errx(1, "Page offset of section %i different", i
);
247 /* We track the first and last address we mapped, so we can
248 * tell entry_point() where to scan. */
249 if (phdr
[i
].p_paddr
< start
)
250 start
= phdr
[i
].p_paddr
;
251 if (phdr
[i
].p_paddr
+ phdr
[i
].p_filesz
> end
)
252 end
= phdr
[i
].p_paddr
+ phdr
[i
].p_filesz
;
254 /* We map this section of the file at its physical address. */
255 map_at(elf_fd
, (void *)phdr
[i
].p_paddr
,
256 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
259 return entry_point((void *)start
, (void *)end
, *page_offset
);
262 /*L:170 Prepare to be SHOCKED and AMAZED. And possibly a trifle nauseated.
264 * We know that CONFIG_PAGE_OFFSET sets what virtual address the kernel expects
265 * to be. We don't know what that option was, but we can figure it out
266 * approximately by looking at the addresses in the code. I chose the common
267 * case of reading a memory location into the %eax register:
269 * movl <some-address>, %eax
271 * This gets encoded as five bytes: "0xA1 <4-byte-address>". For example,
272 * "0xA1 0x18 0x60 0x47 0xC0" reads the address 0xC0476018 into %eax.
274 * In this example can guess that the kernel was compiled with
275 * CONFIG_PAGE_OFFSET set to 0xC0000000 (it's always a round number). If the
276 * kernel were larger than 16MB, we might see 0xC1 addresses show up, but our
277 * kernel isn't that bloated yet.
279 * Unfortunately, x86 has variable-length instructions, so finding this
280 * particular instruction properly involves writing a disassembler. Instead,
281 * we rely on statistics. We look for "0xA1" and tally the different bytes
282 * which occur 4 bytes later (the "0xC0" in our example above). When one of
283 * those bytes appears three times, we can be reasonably confident that it
284 * forms the start of CONFIG_PAGE_OFFSET.
286 * This is amazingly reliable. */
287 static unsigned long intuit_page_offset(unsigned char *img
, unsigned long len
)
289 unsigned int i
, possibilities
[256] = { 0 };
291 for (i
= 0; i
+ 4 < len
; i
++) {
292 /* mov 0xXXXXXXXX,%eax */
293 if (img
[i
] == 0xA1 && ++possibilities
[img
[i
+4]] > 3)
294 return (unsigned long)img
[i
+4] << 24;
296 errx(1, "could not determine page offset");
299 /*L:160 Unfortunately the entire ELF image isn't compressed: the segments
300 * which need loading are extracted and compressed raw. This denies us the
301 * information we need to make a fully-general loader. */
302 static unsigned long unpack_bzimage(int fd
, unsigned long *page_offset
)
306 /* A bzImage always gets loaded at physical address 1M. This is
307 * actually configurable as CONFIG_PHYSICAL_START, but as the comment
308 * there says, "Don't change this unless you know what you are doing".
310 void *img
= (void *)0x100000;
312 /* gzdopen takes our file descriptor (carefully placed at the start of
313 * the GZIP header we found) and returns a gzFile. */
314 f
= gzdopen(fd
, "rb");
315 /* We read it into memory in 64k chunks until we hit the end. */
316 while ((ret
= gzread(f
, img
+ len
, 65536)) > 0)
319 err(1, "reading image from bzImage");
321 verbose("Unpacked size %i addr %p\n", len
, img
);
323 /* Without the ELF header, we can't tell virtual-physical gap. This is
324 * CONFIG_PAGE_OFFSET, and people do actually change it. Fortunately,
325 * I have a clever way of figuring it out from the code itself. */
326 *page_offset
= intuit_page_offset(img
, len
);
328 return entry_point(img
, img
+ len
, *page_offset
);
331 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
332 * supposed to jump into it and it will unpack itself. We can't do that
333 * because the Guest can't run the unpacking code, and adding features to
334 * lguest kills puppies, so we don't want to.
336 * The bzImage is formed by putting the decompressing code in front of the
337 * compressed kernel code. So we can simple scan through it looking for the
338 * first "gzip" header, and start decompressing from there. */
339 static unsigned long load_bzimage(int fd
, unsigned long *page_offset
)
344 /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */
345 while (read(fd
, &c
, 1) == 1) {
361 /* Seek back to the start of the gzip header. */
362 lseek(fd
, -10, SEEK_CUR
);
363 /* One final check: "compressed under UNIX". */
367 return unpack_bzimage(fd
, page_offset
);
370 errx(1, "Could not find kernel in bzImage");
373 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
374 * come wrapped up in the self-decompressing "bzImage" format. With some funky
375 * coding, we can load those, too. */
376 static unsigned long load_kernel(int fd
, unsigned long *page_offset
)
380 /* Read in the first few bytes. */
381 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
382 err(1, "Reading kernel");
384 /* If it's an ELF file, it starts with "\177ELF" */
385 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
386 return map_elf(fd
, &hdr
, page_offset
);
388 /* Otherwise we assume it's a bzImage, and try to unpack it */
389 return load_bzimage(fd
, page_offset
);
392 /* This is a trivial little helper to align pages. Andi Kleen hated it because
393 * it calls getpagesize() twice: "it's dumb code."
395 * Kernel guys get really het up about optimization, even when it's not
396 * necessary. I leave this code as a reaction against that. */
397 static inline unsigned long page_align(unsigned long addr
)
399 /* Add upwards and truncate downwards. */
400 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
403 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
404 * the kernel which the kernel can use to boot from without needing any
405 * drivers. Most distributions now use this as standard: the initrd contains
406 * the code to load the appropriate driver modules for the current machine.
408 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
409 * kernels. He sent me this (and tells me when I break it). */
410 static unsigned long load_initrd(const char *name
, unsigned long mem
)
416 ifd
= open_or_die(name
, O_RDONLY
);
417 /* fstat() is needed to get the file size. */
418 if (fstat(ifd
, &st
) < 0)
419 err(1, "fstat() on initrd '%s'", name
);
421 /* We map the initrd at the top of memory, but mmap wants it to be
422 * page-aligned, so we round the size up for that. */
423 len
= page_align(st
.st_size
);
424 map_at(ifd
, (void *)mem
- len
, 0, st
.st_size
);
425 /* Once a file is mapped, you can close the file descriptor. It's a
426 * little odd, but quite useful. */
428 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
430 /* We return the initrd size. */
434 /* Once we know how much memory we have, and the address the Guest kernel
435 * expects, we can construct simple linear page tables which will get the Guest
436 * far enough into the boot to create its own.
438 * We lay them out of the way, just below the initrd (which is why we need to
440 static unsigned long setup_pagetables(unsigned long mem
,
441 unsigned long initrd_size
,
442 unsigned long page_offset
)
445 unsigned int mapped_pages
, i
, linear_pages
;
446 unsigned int ptes_per_page
= getpagesize()/sizeof(u32
);
448 /* Ideally we map all physical memory starting at page_offset.
449 * However, if page_offset is 0xC0000000 we can only map 1G of physical
450 * (0xC0000000 + 1G overflows). */
451 if (mem
<= -page_offset
)
452 mapped_pages
= mem
/getpagesize();
454 mapped_pages
= -page_offset
/getpagesize();
456 /* Each PTE page can map ptes_per_page pages: how many do we need? */
457 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
459 /* We put the toplevel page directory page at the top of memory. */
460 pgdir
= (void *)mem
- initrd_size
- getpagesize();
462 /* Now we use the next linear_pages pages as pte pages */
463 linear
= (void *)pgdir
- linear_pages
*getpagesize();
465 /* Linear mapping is easy: put every page's address into the mapping in
466 * order. PAGE_PRESENT contains the flags Present, Writable and
468 for (i
= 0; i
< mapped_pages
; i
++)
469 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
471 /* The top level points to the linear page table pages above. The
472 * entry representing page_offset points to the first one, and they
473 * continue from there. */
474 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
475 pgdir
[(i
+ page_offset
/getpagesize())/ptes_per_page
]
476 = (((u32
)linear
+ i
*sizeof(u32
)) | PAGE_PRESENT
);
479 verbose("Linear mapping of %u pages in %u pte pages at %p\n",
480 mapped_pages
, linear_pages
, linear
);
482 /* We return the top level (guest-physical) address: the kernel needs
483 * to know where it is. */
484 return (unsigned long)pgdir
;
487 /* Simple routine to roll all the commandline arguments together with spaces
489 static void concat(char *dst
, char *args
[])
491 unsigned int i
, len
= 0;
493 for (i
= 0; args
[i
]; i
++) {
494 strcpy(dst
+len
, args
[i
]);
495 strcat(dst
+len
, " ");
496 len
+= strlen(args
[i
]) + 1;
498 /* In case it's empty. */
502 /* This is where we actually tell the kernel to initialize the Guest. We saw
503 * the arguments it expects when we looked at initialize() in lguest_user.c:
504 * the top physical page to allow, the top level pagetable, the entry point and
505 * the page_offset constant for the Guest. */
506 static int tell_kernel(u32 pgdir
, u32 start
, u32 page_offset
)
508 u32 args
[] = { LHREQ_INITIALIZE
,
509 top
/getpagesize(), pgdir
, start
, page_offset
};
512 fd
= open_or_die("/dev/lguest", O_RDWR
);
513 if (write(fd
, args
, sizeof(args
)) < 0)
514 err(1, "Writing to /dev/lguest");
516 /* We return the /dev/lguest file descriptor to control this Guest */
521 static void set_fd(int fd
, struct device_list
*devices
)
523 FD_SET(fd
, &devices
->infds
);
524 if (fd
> devices
->max_infd
)
525 devices
->max_infd
= fd
;
531 * With a console and network devices, we can have lots of input which we need
532 * to process. We could try to tell the kernel what file descriptors to watch,
533 * but handing a file descriptor mask through to the kernel is fairly icky.
535 * Instead, we fork off a process which watches the file descriptors and writes
536 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
537 * loop to stop running the Guest. This causes it to return from the
538 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
539 * the LHREQ_BREAK and wake us up again.
541 * This, of course, is merely a different *kind* of icky.
543 static void wake_parent(int pipefd
, int lguest_fd
, struct device_list
*devices
)
545 /* Add the pipe from the Launcher to the fdset in the device_list, so
546 * we watch it, too. */
547 set_fd(pipefd
, devices
);
550 fd_set rfds
= devices
->infds
;
551 u32 args
[] = { LHREQ_BREAK
, 1 };
553 /* Wait until input is ready from one of the devices. */
554 select(devices
->max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
555 /* Is it a message from the Launcher? */
556 if (FD_ISSET(pipefd
, &rfds
)) {
558 /* If read() returns 0, it means the Launcher has
559 * exited. We silently follow. */
560 if (read(pipefd
, &ignorefd
, sizeof(ignorefd
)) == 0)
562 /* Otherwise it's telling us there's a problem with one
563 * of the devices, and we should ignore that file
564 * descriptor from now on. */
565 FD_CLR(ignorefd
, &devices
->infds
);
566 } else /* Send LHREQ_BREAK command. */
567 write(lguest_fd
, args
, sizeof(args
));
571 /* This routine just sets up a pipe to the Waker process. */
572 static int setup_waker(int lguest_fd
, struct device_list
*device_list
)
574 int pipefd
[2], child
;
576 /* We create a pipe to talk to the waker, and also so it knows when the
577 * Launcher dies (and closes pipe). */
584 /* Close the "writing" end of our copy of the pipe */
586 wake_parent(pipefd
[0], lguest_fd
, device_list
);
588 /* Close the reading end of our copy of the pipe. */
591 /* Here is the fd used to talk to the waker. */
598 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
599 * We need to make sure it's not trying to reach into the Launcher itself, so
600 * we have a convenient routine which check it and exits with an error message
601 * if something funny is going on:
603 static void *_check_pointer(unsigned long addr
, unsigned int size
,
606 /* We have to separately check addr and addr+size, because size could
607 * be huge and addr + size might wrap around. */
608 if (addr
>= top
|| addr
+ size
>= top
)
609 errx(1, "%s:%i: Invalid address %li", __FILE__
, line
, addr
);
610 /* We return a pointer for the caller's convenience, now we know it's
614 /* A macro which transparently hands the line number to the real function. */
615 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
617 /* The Guest has given us the address of a "struct lguest_dma". We check it's
618 * OK and convert it to an iovec (which is a simple array of ptr/size
620 static u32
*dma2iov(unsigned long dma
, struct iovec iov
[], unsigned *num
)
623 struct lguest_dma
*udma
;
625 /* First we make sure that the array memory itself is valid. */
626 udma
= check_pointer(dma
, sizeof(*udma
));
627 /* Now we check each element */
628 for (i
= 0; i
< LGUEST_MAX_DMA_SECTIONS
; i
++) {
629 /* A zero length ends the array. */
633 iov
[i
].iov_base
= check_pointer(udma
->addr
[i
], udma
->len
[i
]);
634 iov
[i
].iov_len
= udma
->len
[i
];
638 /* We return the pointer to where the caller should write the amount of
639 * the buffer used. */
640 return &udma
->used_len
;
643 /* This routine gets a DMA buffer from the Guest for a given key, and converts
644 * it to an iovec array. It returns the interrupt the Guest wants when we're
645 * finished, and a pointer to the "used_len" field to fill in. */
646 static u32
*get_dma_buffer(int fd
, void *key
,
647 struct iovec iov
[], unsigned int *num
, u32
*irq
)
649 u32 buf
[] = { LHREQ_GETDMA
, (u32
)key
};
653 /* Ask the kernel for a DMA buffer corresponding to this key. */
654 udma
= write(fd
, buf
, sizeof(buf
));
655 /* They haven't registered any, or they're all used? */
656 if (udma
== (unsigned long)-1)
659 /* Convert it into our iovec array */
660 res
= dma2iov(udma
, iov
, num
);
661 /* The kernel stashes irq in ->used_len to get it out to us. */
663 /* Return a pointer to ((struct lguest_dma *)udma)->used_len. */
667 /* This is a convenient routine to send the Guest an interrupt. */
668 static void trigger_irq(int fd
, u32 irq
)
670 u32 buf
[] = { LHREQ_IRQ
, irq
};
671 if (write(fd
, buf
, sizeof(buf
)) != 0)
672 err(1, "Triggering irq %i", irq
);
675 /* This simply sets up an iovec array where we can put data to be discarded.
676 * This happens when the Guest doesn't want or can't handle the input: we have
677 * to get rid of it somewhere, and if we bury it in the ceiling space it will
678 * start to smell after a week. */
679 static void discard_iovec(struct iovec
*iov
, unsigned int *num
)
681 static char discard_buf
[1024];
683 iov
->iov_base
= discard_buf
;
684 iov
->iov_len
= sizeof(discard_buf
);
687 /* Here is the input terminal setting we save, and the routine to restore them
688 * on exit so the user can see what they type next. */
689 static struct termios orig_term
;
690 static void restore_term(void)
692 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
695 /* We associate some data with the console for our exit hack. */
698 /* How many times have they hit ^C? */
700 /* When did they start? */
701 struct timeval start
;
704 /* This is the routine which handles console input (ie. stdin). */
705 static bool handle_console_input(int fd
, struct device
*dev
)
710 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
711 struct console_abort
*abort
= dev
->priv
;
713 /* First we get the console buffer from the Guest. The key is dev->mem
714 * which was set to 0 in setup_console(). */
715 lenp
= get_dma_buffer(fd
, dev
->mem
, iov
, &num
, &irq
);
717 /* If it's not ready for input, warn and set up to discard. */
718 warn("console: no dma buffer!");
719 discard_iovec(iov
, &num
);
722 /* This is why we convert to iovecs: the readv() call uses them, and so
723 * it reads straight into the Guest's buffer. */
724 len
= readv(dev
->fd
, iov
, num
);
726 /* This implies that the console is closed, is /dev/null, or
727 * something went terribly wrong. We still go through the rest
728 * of the logic, though, especially the exit handling below. */
729 warnx("Failed to get console input, ignoring console.");
733 /* If we read the data into the Guest, fill in the length and send the
737 trigger_irq(fd
, irq
);
740 /* Three ^C within one second? Exit.
742 * This is such a hack, but works surprisingly well. Each ^C has to be
743 * in a buffer by itself, so they can't be too fast. But we check that
744 * we get three within about a second, so they can't be too slow. */
745 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
747 gettimeofday(&abort
->start
, NULL
);
748 else if (abort
->count
== 3) {
750 gettimeofday(&now
, NULL
);
751 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
752 u32 args
[] = { LHREQ_BREAK
, 0 };
753 /* Close the fd so Waker will know it has to
756 /* Just in case waker is blocked in BREAK, send
758 write(fd
, args
, sizeof(args
));
764 /* Any other key resets the abort counter. */
767 /* Now, if we didn't read anything, put the input terminal back and
768 * return failure (meaning, don't call us again). */
773 /* Everything went OK! */
777 /* Handling console output is much simpler than input. */
778 static u32
handle_console_output(int fd
, const struct iovec
*iov
,
779 unsigned num
, struct device
*dev
)
781 /* Whatever the Guest sends, write it to standard output. Return the
782 * number of bytes written. */
783 return writev(STDOUT_FILENO
, iov
, num
);
786 /* Guest->Host network output is also pretty easy. */
787 static u32
handle_tun_output(int fd
, const struct iovec
*iov
,
788 unsigned num
, struct device
*dev
)
790 /* We put a flag in the "priv" pointer of the network device, and set
791 * it as soon as we see output. We'll see why in handle_tun_input() */
792 *(bool *)dev
->priv
= true;
793 /* Whatever packet the Guest sent us, write it out to the tun
795 return writev(dev
->fd
, iov
, num
);
798 /* This matches the peer_key() in lguest_net.c. The key for any given slot
799 * is the address of the network device's page plus 4 * the slot number. */
800 static unsigned long peer_offset(unsigned int peernum
)
805 /* This is where we handle a packet coming in from the tun device */
806 static bool handle_tun_input(int fd
, struct device
*dev
)
811 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
813 /* First we get a buffer the Guest has bound to its key. */
814 lenp
= get_dma_buffer(fd
, dev
->mem
+peer_offset(NET_PEERNUM
), iov
, &num
,
817 /* Now, it's expected that if we try to send a packet too
818 * early, the Guest won't be ready yet. This is why we set a
819 * flag when the Guest sends its first packet. If it's sent a
820 * packet we assume it should be ready to receive them.
822 * Actually, this is what the status bits in the descriptor are
823 * for: we should *use* them. FIXME! */
824 if (*(bool *)dev
->priv
)
825 warn("network: no dma buffer!");
826 discard_iovec(iov
, &num
);
829 /* Read the packet from the device directly into the Guest's buffer. */
830 len
= readv(dev
->fd
, iov
, num
);
832 err(1, "reading network");
834 /* Write the used_len, and trigger the interrupt for the Guest */
837 trigger_irq(fd
, irq
);
839 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
840 ((u8
*)iov
[0].iov_base
)[0], ((u8
*)iov
[0].iov_base
)[1],
841 lenp
? "sent" : "discarded");
846 /* The last device handling routine is block output: the Guest has sent a DMA
847 * to the block device. It will have placed the command it wants in the
848 * "struct lguest_block_page". */
849 static u32
handle_block_output(int fd
, const struct iovec
*iov
,
850 unsigned num
, struct device
*dev
)
852 struct lguest_block_page
*p
= dev
->mem
;
854 unsigned int len
, reply_num
;
855 struct iovec reply
[LGUEST_MAX_DMA_SECTIONS
];
856 off64_t device_len
, off
= (off64_t
)p
->sector
* 512;
858 /* First we extract the device length from the dev->priv pointer. */
859 device_len
= *(off64_t
*)dev
->priv
;
861 /* We first check that the read or write is within the length of the
863 if (off
>= device_len
)
864 errx(1, "Bad offset %llu vs %llu", off
, device_len
);
865 /* Move to the right location in the block file. This shouldn't fail,
866 * but best to check. */
867 if (lseek64(dev
->fd
, off
, SEEK_SET
) != off
)
868 err(1, "Bad seek to sector %i", p
->sector
);
870 verbose("Block: %s at offset %llu\n", p
->type
? "WRITE" : "READ", off
);
872 /* They were supposed to bind a reply buffer at key equal to the start
873 * of the block device memory. We need this to tell them when the
874 * request is finished. */
875 lenp
= get_dma_buffer(fd
, dev
->mem
, reply
, &reply_num
, &irq
);
877 err(1, "Block request didn't give us a dma buffer");
880 /* A write request. The DMA they sent contained the data, so
882 len
= writev(dev
->fd
, iov
, num
);
883 /* Grr... Now we know how long the "struct lguest_dma" they
884 * sent was, we make sure they didn't try to write over the end
885 * of the block file (possibly extending it). */
886 if (off
+ len
> device_len
) {
887 /* Trim it back to the correct length */
888 ftruncate64(dev
->fd
, device_len
);
889 /* Die, bad Guest, die. */
890 errx(1, "Write past end %llu+%u", off
, len
);
892 /* The reply length is 0: we just send back an empty DMA to
893 * interrupt them and tell them the write is finished. */
896 /* A read request. They sent an empty DMA to start the
897 * request, and we put the read contents into the reply
899 len
= readv(dev
->fd
, reply
, reply_num
);
903 /* The result is 1 (done), 2 if there was an error (short read or
905 p
->result
= 1 + (p
->bytes
!= len
);
906 /* Now tell them we've used their reply buffer. */
907 trigger_irq(fd
, irq
);
909 /* We're supposed to return the number of bytes of the output buffer we
910 * used. But the block device uses the "result" field instead, so we
915 /* This is the generic routine we call when the Guest sends some DMA out. */
916 static void handle_output(int fd
, unsigned long dma
, unsigned long key
,
917 struct device_list
*devices
)
921 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
924 /* Convert the "struct lguest_dma" they're sending to a "struct
926 lenp
= dma2iov(dma
, iov
, &num
);
928 /* Check each device: if they expect output to this key, tell them to
930 for (i
= devices
->dev
; i
; i
= i
->next
) {
931 if (i
->handle_output
&& key
== i
->watch_key
) {
932 /* We write the result straight into the used_len field
934 *lenp
= i
->handle_output(fd
, iov
, num
, i
);
939 /* This can happen: the kernel sends any SEND_DMA which doesn't match
940 * another Guest to us. It could be that another Guest just left a
941 * network, for example. But it's unusual. */
942 warnx("Pending dma %p, key %p", (void *)dma
, (void *)key
);
945 /* This is called when the waker wakes us up: check for incoming file
947 static void handle_input(int fd
, struct device_list
*devices
)
949 /* select() wants a zeroed timeval to mean "don't wait". */
950 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
954 fd_set fds
= devices
->infds
;
956 /* If nothing is ready, we're done. */
957 if (select(devices
->max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
960 /* Otherwise, call the device(s) which have readable
961 * file descriptors and a method of handling them. */
962 for (i
= devices
->dev
; i
; i
= i
->next
) {
963 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
964 /* If handle_input() returns false, it means we
965 * should no longer service it.
966 * handle_console_input() does this. */
967 if (!i
->handle_input(fd
, i
)) {
968 /* Clear it from the set of input file
969 * descriptors kept at the head of the
971 FD_CLR(i
->fd
, &devices
->infds
);
972 /* Tell waker to ignore it too... */
973 write(waker_fd
, &i
->fd
, sizeof(i
->fd
));
983 * All devices need a descriptor so the Guest knows it exists, and a "struct
984 * device" so the Launcher can keep track of it. We have common helper
985 * routines to allocate them.
987 * This routine allocates a new "struct lguest_device_desc" from descriptor
988 * table in the devices array just above the Guest's normal memory. */
989 static struct lguest_device_desc
*
990 new_dev_desc(struct lguest_device_desc
*descs
,
991 u16 type
, u16 features
, u16 num_pages
)
995 for (i
= 0; i
< LGUEST_MAX_DEVICES
; i
++) {
996 if (!descs
[i
].type
) {
997 descs
[i
].type
= type
;
998 descs
[i
].features
= features
;
999 descs
[i
].num_pages
= num_pages
;
1000 /* If they said the device needs memory, we allocate
1001 * that now, bumping up the top of Guest memory. */
1003 map_zeroed_pages(top
, num_pages
);
1004 descs
[i
].pfn
= top
/getpagesize();
1005 top
+= num_pages
*getpagesize();
1010 errx(1, "too many devices");
1013 /* This monster routine does all the creation and setup of a new device,
1014 * including caling new_dev_desc() to allocate the descriptor and device
1016 static struct device
*new_device(struct device_list
*devices
,
1017 u16 type
, u16 num_pages
, u16 features
,
1019 bool (*handle_input
)(int, struct device
*),
1020 unsigned long watch_off
,
1021 u32 (*handle_output
)(int,
1022 const struct iovec
*,
1026 struct device
*dev
= malloc(sizeof(*dev
));
1028 /* Append to device list. Prepending to a single-linked list is
1029 * easier, but the user expects the devices to be arranged on the bus
1030 * in command-line order. The first network device on the command line
1031 * is eth0, the first block device /dev/lgba, etc. */
1032 *devices
->lastdev
= dev
;
1034 devices
->lastdev
= &dev
->next
;
1036 /* Now we populate the fields one at a time. */
1038 /* If we have an input handler for this file descriptor, then we add it
1039 * to the device_list's fdset and maxfd. */
1041 set_fd(dev
->fd
, devices
);
1042 dev
->desc
= new_dev_desc(devices
->descs
, type
, features
, num_pages
);
1043 dev
->mem
= (void *)(dev
->desc
->pfn
* getpagesize());
1044 dev
->handle_input
= handle_input
;
1045 dev
->watch_key
= (unsigned long)dev
->mem
+ watch_off
;
1046 dev
->handle_output
= handle_output
;
1050 /* Our first setup routine is the console. It's a fairly simple device, but
1051 * UNIX tty handling makes it uglier than it could be. */
1052 static void setup_console(struct device_list
*devices
)
1056 /* If we can save the initial standard input settings... */
1057 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1058 struct termios term
= orig_term
;
1059 /* Then we turn off echo, line buffering and ^C etc. We want a
1060 * raw input stream to the Guest. */
1061 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1062 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1063 /* If we exit gracefully, the original settings will be
1064 * restored so the user can see what they're typing. */
1065 atexit(restore_term
);
1068 /* We don't currently require any memory for the console, so we ask for
1070 dev
= new_device(devices
, LGUEST_DEVICE_T_CONSOLE
, 0, 0,
1071 STDIN_FILENO
, handle_console_input
,
1072 LGUEST_CONSOLE_DMA_KEY
, handle_console_output
);
1073 /* We store the console state in dev->priv, and initialize it. */
1074 dev
->priv
= malloc(sizeof(struct console_abort
));
1075 ((struct console_abort
*)dev
->priv
)->count
= 0;
1076 verbose("device %p: console\n",
1077 (void *)(dev
->desc
->pfn
* getpagesize()));
1080 /* Setting up a block file is also fairly straightforward. */
1081 static void setup_block_file(const char *filename
, struct device_list
*devices
)
1085 off64_t
*device_len
;
1086 struct lguest_block_page
*p
;
1088 /* We open with O_LARGEFILE because otherwise we get stuck at 2G. We
1089 * open with O_DIRECT because otherwise our benchmarks go much too
1091 fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
|O_DIRECT
);
1093 /* We want one page, and have no input handler (the block file never
1094 * has anything interesting to say to us). Our timing will be quite
1095 * random, so it should be a reasonable randomness source. */
1096 dev
= new_device(devices
, LGUEST_DEVICE_T_BLOCK
, 1,
1097 LGUEST_DEVICE_F_RANDOMNESS
,
1098 fd
, NULL
, 0, handle_block_output
);
1100 /* We store the device size in the private area */
1101 device_len
= dev
->priv
= malloc(sizeof(*device_len
));
1102 /* This is the safe way of establishing the size of our device: it
1103 * might be a normal file or an actual block device like /dev/hdb. */
1104 *device_len
= lseek64(fd
, 0, SEEK_END
);
1106 /* The device memory is a "struct lguest_block_page". It's zeroed
1107 * already, we just need to put in the device size. Block devices
1108 * think in sectors (ie. 512 byte chunks), so we translate here. */
1110 p
->num_sectors
= *device_len
/512;
1111 verbose("device %p: block %i sectors\n",
1112 (void *)(dev
->desc
->pfn
* getpagesize()), p
->num_sectors
);
1118 * Setting up network devices is quite a pain, because we have three types.
1119 * First, we have the inter-Guest network. This is a file which is mapped into
1120 * the address space of the Guests who are on the network. Because it is a
1121 * shared mapping, the same page underlies all the devices, and they can send
1122 * DMA to each other.
1124 * Remember from our network driver, the Guest is told what slot in the page it
1125 * is to use. We use exclusive fnctl locks to reserve a slot. If another
1126 * Guest is using a slot, the lock will fail and we try another. Because fnctl
1127 * locks are cleaned up automatically when we die, this cleverly means that our
1128 * reservation on the slot will vanish if we crash. */
1129 static unsigned int find_slot(int netfd
, const char *filename
)
1133 fl
.l_type
= F_WRLCK
;
1134 fl
.l_whence
= SEEK_SET
;
1136 /* Try a 1 byte lock in each possible position number */
1137 for (fl
.l_start
= 0;
1138 fl
.l_start
< getpagesize()/sizeof(struct lguest_net
);
1140 /* If we succeed, return the slot number. */
1141 if (fcntl(netfd
, F_SETLK
, &fl
) == 0)
1144 errx(1, "No free slots in network file %s", filename
);
1147 /* This function sets up the network file */
1148 static void setup_net_file(const char *filename
,
1149 struct device_list
*devices
)
1154 /* We don't use open_or_die() here: for friendliness we create the file
1155 * if it doesn't already exist. */
1156 netfd
= open(filename
, O_RDWR
, 0);
1158 if (errno
== ENOENT
) {
1159 netfd
= open(filename
, O_RDWR
|O_CREAT
, 0600);
1161 /* If we succeeded, initialize the file with a
1163 char page
[getpagesize()];
1164 memset(page
, 0, sizeof(page
));
1165 write(netfd
, page
, sizeof(page
));
1169 err(1, "cannot open net file '%s'", filename
);
1172 /* We need 1 page, and the features indicate the slot to use and that
1173 * no checksum is needed. We never touch this device again; it's
1174 * between the Guests on the network, so we don't register input or
1175 * output handlers. */
1176 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1177 find_slot(netfd
, filename
)|LGUEST_NET_F_NOCSUM
,
1180 /* Map the shared file. */
1181 if (mmap(dev
->mem
, getpagesize(), PROT_READ
|PROT_WRITE
,
1182 MAP_FIXED
|MAP_SHARED
, netfd
, 0) != dev
->mem
)
1183 err(1, "could not mmap '%s'", filename
);
1184 verbose("device %p: shared net %s, peer %i\n",
1185 (void *)(dev
->desc
->pfn
* getpagesize()), filename
,
1186 dev
->desc
->features
& ~LGUEST_NET_F_NOCSUM
);
1190 static u32
str2ip(const char *ipaddr
)
1192 unsigned int byte
[4];
1194 sscanf(ipaddr
, "%u.%u.%u.%u", &byte
[0], &byte
[1], &byte
[2], &byte
[3]);
1195 return (byte
[0] << 24) | (byte
[1] << 16) | (byte
[2] << 8) | byte
[3];
1198 /* This code is "adapted" from libbridge: it attaches the Host end of the
1199 * network device to the bridge device specified by the command line.
1201 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1202 * dislike bridging), and I just try not to break it. */
1203 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1209 errx(1, "must specify bridge name");
1211 ifidx
= if_nametoindex(if_name
);
1213 errx(1, "interface %s does not exist!", if_name
);
1215 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1216 ifr
.ifr_ifindex
= ifidx
;
1217 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1218 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1221 /* This sets up the Host end of the network device with an IP address, brings
1222 * it up so packets will flow, the copies the MAC address into the hwaddr
1223 * pointer (in practice, the Host's slot in the network device's memory). */
1224 static void configure_device(int fd
, const char *devname
, u32 ipaddr
,
1225 unsigned char hwaddr
[6])
1228 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1230 /* Don't read these incantations. Just cut & paste them like I did! */
1231 memset(&ifr
, 0, sizeof(ifr
));
1232 strcpy(ifr
.ifr_name
, devname
);
1233 sin
->sin_family
= AF_INET
;
1234 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1235 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1236 err(1, "Setting %s interface address", devname
);
1237 ifr
.ifr_flags
= IFF_UP
;
1238 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1239 err(1, "Bringing interface %s up", devname
);
1241 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1242 * above). IF means Interface, and HWADDR is hardware address.
1244 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1245 err(1, "getting hw address for %s", devname
);
1246 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1249 /*L:195 The other kind of network is a Host<->Guest network. This can either
1250 * use briding or routing, but the principle is the same: it uses the "tun"
1251 * device to inject packets into the Host as if they came in from a normal
1252 * network card. We just shunt packets between the Guest and the tun
1254 static void setup_tun_net(const char *arg
, struct device_list
*devices
)
1260 const char *br_name
= NULL
;
1262 /* We open the /dev/net/tun device and tell it we want a tap device. A
1263 * tap device is like a tun device, only somehow different. To tell
1264 * the truth, I completely blundered my way through this code, but it
1266 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1267 memset(&ifr
, 0, sizeof(ifr
));
1268 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1269 strcpy(ifr
.ifr_name
, "tap%d");
1270 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1271 err(1, "configuring /dev/net/tun");
1272 /* We don't need checksums calculated for packets coming in this
1273 * device: trust us! */
1274 ioctl(netfd
, TUNSETNOCSUM
, 1);
1276 /* We create the net device with 1 page, using the features field of
1277 * the descriptor to tell the Guest it is in slot 1 (NET_PEERNUM), and
1278 * that the device has fairly random timing. We do *not* specify
1279 * LGUEST_NET_F_NOCSUM: these packets can reach the real world.
1281 * We will put our MAC address is slot 0 for the Guest to see, so
1282 * it will send packets to us using the key "peer_offset(0)": */
1283 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1284 NET_PEERNUM
|LGUEST_DEVICE_F_RANDOMNESS
, netfd
,
1285 handle_tun_input
, peer_offset(0), handle_tun_output
);
1287 /* We keep a flag which says whether we've seen packets come out from
1288 * this network device. */
1289 dev
->priv
= malloc(sizeof(bool));
1290 *(bool *)dev
->priv
= false;
1292 /* We need a socket to perform the magic network ioctls to bring up the
1293 * tap interface, connect to the bridge etc. Any socket will do! */
1294 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1296 err(1, "opening IP socket");
1298 /* If the command line was --tunnet=bridge:<name> do bridging. */
1299 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1301 br_name
= arg
+ strlen(BRIDGE_PFX
);
1302 add_to_bridge(ipfd
, ifr
.ifr_name
, br_name
);
1303 } else /* It is an IP address to set up the device with */
1306 /* We are peer 0, ie. first slot, so we hand dev->mem to this routine
1307 * to write the MAC address at the start of the device memory. */
1308 configure_device(ipfd
, ifr
.ifr_name
, ip
, dev
->mem
);
1310 /* Set "promisc" bit: we want every single packet if we're going to
1311 * bridge to other machines (and otherwise it doesn't matter). */
1312 *((u8
*)dev
->mem
) |= 0x1;
1316 verbose("device %p: tun net %u.%u.%u.%u\n",
1317 (void *)(dev
->desc
->pfn
* getpagesize()),
1318 (u8
)(ip
>>24), (u8
)(ip
>>16), (u8
)(ip
>>8), (u8
)ip
);
1320 verbose("attached to bridge: %s\n", br_name
);
1322 /* That's the end of device setup. */
1324 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1325 * its input and output, and finally, lays it to rest. */
1326 static void __attribute__((noreturn
))
1327 run_guest(int lguest_fd
, struct device_list
*device_list
)
1330 u32 args
[] = { LHREQ_BREAK
, 0 };
1331 unsigned long arr
[2];
1334 /* We read from the /dev/lguest device to run the Guest. */
1335 readval
= read(lguest_fd
, arr
, sizeof(arr
));
1337 /* The read can only really return sizeof(arr) (the Guest did a
1338 * SEND_DMA to us), or an error. */
1340 /* For a successful read, arr[0] is the address of the "struct
1341 * lguest_dma", and arr[1] is the key the Guest sent to. */
1342 if (readval
== sizeof(arr
)) {
1343 handle_output(lguest_fd
, arr
[0], arr
[1], device_list
);
1345 /* ENOENT means the Guest died. Reading tells us why. */
1346 } else if (errno
== ENOENT
) {
1347 char reason
[1024] = { 0 };
1348 read(lguest_fd
, reason
, sizeof(reason
)-1);
1349 errx(1, "%s", reason
);
1350 /* EAGAIN means the waker wanted us to look at some input.
1351 * Anything else means a bug or incompatible change. */
1352 } else if (errno
!= EAGAIN
)
1353 err(1, "Running guest failed");
1355 /* Service input, then unset the BREAK which releases
1357 handle_input(lguest_fd
, device_list
);
1358 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
1359 err(1, "Resetting break");
1363 * This is the end of the Launcher.
1365 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1366 * Drivers. If we were to see the Host kernel I/O code, our understanding
1367 * would be complete... :*/
1369 static struct option opts
[] = {
1370 { "verbose", 0, NULL
, 'v' },
1371 { "sharenet", 1, NULL
, 's' },
1372 { "tunnet", 1, NULL
, 't' },
1373 { "block", 1, NULL
, 'b' },
1374 { "initrd", 1, NULL
, 'i' },
1377 static void usage(void)
1379 errx(1, "Usage: lguest [--verbose] "
1380 "[--sharenet=<filename>|--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1381 "|--block=<filename>|--initrd=<filename>]...\n"
1382 "<mem-in-mb> vmlinux [args...]");
1385 /*L:100 The Launcher code itself takes us out into userspace, that scary place
1386 * where pointers run wild and free! Unfortunately, like most userspace
1387 * programs, it's quite boring (which is why everyone like to hack on the
1388 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
1389 * will get you through this section. Or, maybe not.
1391 * The Launcher binary sits up high, usually starting at address 0xB8000000.
1392 * Everything below this is the "physical" memory for the Guest. For example,
1393 * if the Guest were to write a "1" at physical address 0, we would see a "1"
1394 * in the Launcher at "(int *)0". Guest physical == Launcher virtual.
1396 * This can be tough to get your head around, but usually it just means that we
1397 * don't need to do any conversion when the Guest gives us it's "physical"
1400 int main(int argc
, char *argv
[])
1402 /* Memory, top-level pagetable, code startpoint, PAGE_OFFSET and size
1403 * of the (optional) initrd. */
1404 unsigned long mem
= 0, pgdir
, start
, page_offset
, initrd_size
= 0;
1405 /* A temporary and the /dev/lguest file descriptor. */
1406 int i
, c
, lguest_fd
;
1407 /* The list of Guest devices, based on command line arguments. */
1408 struct device_list device_list
;
1409 /* The boot information for the Guest: at guest-physical address 0. */
1410 void *boot
= (void *)0;
1411 /* If they specify an initrd file to load. */
1412 const char *initrd_name
= NULL
;
1414 /* First we initialize the device list. Since console and network
1415 * device receive input from a file descriptor, we keep an fdset
1416 * (infds) and the maximum fd number (max_infd) with the head of the
1417 * list. We also keep a pointer to the last device, for easy appending
1419 device_list
.max_infd
= -1;
1420 device_list
.dev
= NULL
;
1421 device_list
.lastdev
= &device_list
.dev
;
1422 FD_ZERO(&device_list
.infds
);
1424 /* We need to know how much memory so we can set up the device
1425 * descriptor and memory pages for the devices as we parse the command
1426 * line. So we quickly look through the arguments to find the amount
1428 for (i
= 1; i
< argc
; i
++) {
1429 if (argv
[i
][0] != '-') {
1430 mem
= top
= atoi(argv
[i
]) * 1024 * 1024;
1431 device_list
.descs
= map_zeroed_pages(top
, 1);
1432 top
+= getpagesize();
1437 /* The options are fairly straight-forward */
1438 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1444 setup_net_file(optarg
, &device_list
);
1447 setup_tun_net(optarg
, &device_list
);
1450 setup_block_file(optarg
, &device_list
);
1453 initrd_name
= optarg
;
1456 warnx("Unknown argument %s", argv
[optind
]);
1460 /* After the other arguments we expect memory and kernel image name,
1461 * followed by command line arguments for the kernel. */
1462 if (optind
+ 2 > argc
)
1465 /* We always have a console device */
1466 setup_console(&device_list
);
1468 /* We start by mapping anonymous pages over all of guest-physical
1469 * memory range. This fills it with 0, and ensures that the Guest
1470 * won't be killed when it tries to access it. */
1471 map_zeroed_pages(0, mem
/ getpagesize());
1473 /* Now we load the kernel */
1474 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
),
1477 /* Map the initrd image if requested (at top of physical memory) */
1479 initrd_size
= load_initrd(initrd_name
, mem
);
1480 /* These are the location in the Linux boot header where the
1481 * start and size of the initrd are expected to be found. */
1482 *(unsigned long *)(boot
+0x218) = mem
- initrd_size
;
1483 *(unsigned long *)(boot
+0x21c) = initrd_size
;
1484 /* The bootloader type 0xFF means "unknown"; that's OK. */
1485 *(unsigned char *)(boot
+0x210) = 0xFF;
1488 /* Set up the initial linear pagetables, starting below the initrd. */
1489 pgdir
= setup_pagetables(mem
, initrd_size
, page_offset
);
1491 /* The Linux boot header contains an "E820" memory map: ours is a
1492 * simple, single region. */
1493 *(char*)(boot
+E820NR
) = 1;
1494 *((struct e820entry
*)(boot
+E820MAP
))
1495 = ((struct e820entry
) { 0, mem
, E820_RAM
});
1496 /* The boot header contains a command line pointer: we put the command
1497 * line after the boot header (at address 4096) */
1498 *(void **)(boot
+ 0x228) = boot
+ 4096;
1499 concat(boot
+ 4096, argv
+optind
+2);
1501 /* The guest type value of "1" tells the Guest it's under lguest. */
1502 *(int *)(boot
+ 0x23c) = 1;
1504 /* We tell the kernel to initialize the Guest: this returns the open
1505 * /dev/lguest file descriptor. */
1506 lguest_fd
= tell_kernel(pgdir
, start
, page_offset
);
1508 /* We fork off a child process, which wakes the Launcher whenever one
1509 * of the input file descriptors needs attention. Otherwise we would
1510 * run the Guest until it tries to output something. */
1511 waker_fd
= setup_waker(lguest_fd
, &device_list
);
1513 /* Finally, run the Guest. This doesn't return. */
1514 run_guest(lguest_fd
, &device_list
);
1519 * Mastery is done: you now know everything I do.
1521 * But surely you have seen code, features and bugs in your wanderings which
1522 * you now yearn to attack? That is the real game, and I look forward to you
1523 * patching and forking lguest into the Your-Name-Here-visor.
1525 * Farewell, and good coding!