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 #define _LARGEFILE64_SOURCE
15 #include <sys/param.h>
16 #include <sys/types.h>
23 #include <sys/socket.h>
24 #include <sys/ioctl.h>
27 #include <netinet/in.h>
29 #include <linux/sockios.h>
30 #include <linux/if_tun.h>
35 /*L:110 We can ignore the 28 include files we need for this program, but I do
36 * want to draw attention to the use of kernel-style types.
38 * As Linus said, "C is a Spartan language, and so should your naming be." I
39 * like these abbreviations and the header we need uses them, so we define them
42 typedef unsigned long long u64
;
46 #include "linux/lguest_launcher.h"
47 #include "asm-x86/e820.h"
50 #define PAGE_PRESENT 0x7 /* Present, RW, Execute */
52 #define BRIDGE_PFX "bridge:"
54 #define SIOCBRADDIF 0x89a2 /* add interface to bridge */
56 /* We can have up to 256 pages for devices. */
57 #define DEVICE_PAGES 256
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 pointer to the start of guest memory. */
69 static void *guest_base
;
70 /* The maximum guest physical address allowed, and maximum possible. */
71 static unsigned long guest_limit
, guest_max
;
73 /* This is our list of devices. */
76 /* Summary information about the devices in our list: ready to pass to
77 * select() to ask which need servicing.*/
81 /* The descriptor page for the devices. */
82 struct lguest_device_desc
*descs
;
84 /* A single linked list of devices. */
86 /* ... And an end pointer so we can easily append new devices */
87 struct device
**lastdev
;
90 /* The device structure describes a single device. */
93 /* The linked-list pointer. */
95 /* The descriptor for this device, as mapped into the Guest. */
96 struct lguest_device_desc
*desc
;
97 /* The memory page(s) of this device, if any. Also mapped in Guest. */
100 /* If handle_input is set, it wants to be called when this file
101 * descriptor is ready. */
103 bool (*handle_input
)(int fd
, struct device
*me
);
105 /* If handle_output is set, it wants to be called when the Guest sends
106 * DMA to this key. */
107 unsigned long watch_key
;
108 u32 (*handle_output
)(int fd
, const struct iovec
*iov
,
109 unsigned int num
, struct device
*me
);
111 /* Device-specific data. */
115 /*L:100 The Launcher code itself takes us out into userspace, that scary place
116 * where pointers run wild and free! Unfortunately, like most userspace
117 * programs, it's quite boring (which is why everyone likes to hack on the
118 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it
119 * will get you through this section. Or, maybe not.
121 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
122 * memory and stores it in "guest_base". In other words, Guest physical ==
123 * Launcher virtual with an offset.
125 * This can be tough to get your head around, but usually it just means that we
126 * use these trivial conversion functions when the Guest gives us it's
127 * "physical" addresses: */
128 static void *from_guest_phys(unsigned long addr
)
130 return guest_base
+ addr
;
133 static unsigned long to_guest_phys(const void *addr
)
135 return (addr
- guest_base
);
139 * Loading the Kernel.
141 * We start with couple of simple helper routines. open_or_die() avoids
142 * error-checking code cluttering the callers: */
143 static int open_or_die(const char *name
, int flags
)
145 int fd
= open(name
, flags
);
147 err(1, "Failed to open %s", name
);
151 /* map_zeroed_pages() takes a number of pages. */
152 static void *map_zeroed_pages(unsigned int num
)
154 int fd
= open_or_die("/dev/zero", O_RDONLY
);
157 /* We use a private mapping (ie. if we write to the page, it will be
159 addr
= mmap(NULL
, getpagesize() * num
,
160 PROT_READ
|PROT_WRITE
|PROT_EXEC
, MAP_PRIVATE
, fd
, 0);
161 if (addr
== MAP_FAILED
)
162 err(1, "Mmaping %u pages of /dev/zero", num
);
167 /* Get some more pages for a device. */
168 static void *get_pages(unsigned int num
)
170 void *addr
= from_guest_phys(guest_limit
);
172 guest_limit
+= num
* getpagesize();
173 if (guest_limit
> guest_max
)
174 errx(1, "Not enough memory for devices");
178 /* To find out where to start we look for the magic Guest string, which marks
179 * the code we see in lguest_asm.S. This is a hack which we are currently
180 * plotting to replace with the normal Linux entry point. */
181 static unsigned long entry_point(const void *start
, const void *end
,
182 unsigned long page_offset
)
186 /* The scan gives us the physical starting address. We want the
187 * virtual address in this case, and fortunately, we already figured
188 * out the physical-virtual difference and passed it here in
190 for (p
= start
; p
< end
; p
++)
191 if (memcmp(p
, "GenuineLguest", strlen("GenuineLguest")) == 0)
192 return to_guest_phys(p
+ strlen("GenuineLguest"))
195 errx(1, "Is this image a genuine lguest?");
198 /* This routine is used to load the kernel or initrd. It tries mmap, but if
199 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
200 * it falls back to reading the memory in. */
201 static void map_at(int fd
, void *addr
, unsigned long offset
, unsigned long len
)
205 /* We map writable even though for some segments are marked read-only.
206 * The kernel really wants to be writable: it patches its own
209 * MAP_PRIVATE means that the page won't be copied until a write is
210 * done to it. This allows us to share untouched memory between
212 if (mmap(addr
, len
, PROT_READ
|PROT_WRITE
|PROT_EXEC
,
213 MAP_FIXED
|MAP_PRIVATE
, fd
, offset
) != MAP_FAILED
)
216 /* pread does a seek and a read in one shot: saves a few lines. */
217 r
= pread(fd
, addr
, len
, offset
);
219 err(1, "Reading offset %lu len %lu gave %zi", offset
, len
, r
);
222 /* This routine takes an open vmlinux image, which is in ELF, and maps it into
223 * the Guest memory. ELF = Embedded Linking Format, which is the format used
224 * by all modern binaries on Linux including the kernel.
226 * The ELF headers give *two* addresses: a physical address, and a virtual
227 * address. The Guest kernel expects to be placed in memory at the physical
228 * address, and the page tables set up so it will correspond to that virtual
229 * address. We return the difference between the virtual and physical
230 * addresses in the "page_offset" pointer.
232 * We return the starting address. */
233 static unsigned long map_elf(int elf_fd
, const Elf32_Ehdr
*ehdr
,
234 unsigned long *page_offset
)
236 void *start
= (void *)-1, *end
= NULL
;
237 Elf32_Phdr phdr
[ehdr
->e_phnum
];
240 /* Sanity checks on the main ELF header: an x86 executable with a
241 * reasonable number of correctly-sized program headers. */
242 if (ehdr
->e_type
!= ET_EXEC
243 || ehdr
->e_machine
!= EM_386
244 || ehdr
->e_phentsize
!= sizeof(Elf32_Phdr
)
245 || ehdr
->e_phnum
< 1 || ehdr
->e_phnum
> 65536U/sizeof(Elf32_Phdr
))
246 errx(1, "Malformed elf header");
248 /* An ELF executable contains an ELF header and a number of "program"
249 * headers which indicate which parts ("segments") of the program to
252 /* We read in all the program headers at once: */
253 if (lseek(elf_fd
, ehdr
->e_phoff
, SEEK_SET
) < 0)
254 err(1, "Seeking to program headers");
255 if (read(elf_fd
, phdr
, sizeof(phdr
)) != sizeof(phdr
))
256 err(1, "Reading program headers");
258 /* We don't know page_offset yet. */
261 /* Try all the headers: there are usually only three. A read-only one,
262 * a read-write one, and a "note" section which isn't loadable. */
263 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
264 /* If this isn't a loadable segment, we ignore it */
265 if (phdr
[i
].p_type
!= PT_LOAD
)
268 verbose("Section %i: size %i addr %p\n",
269 i
, phdr
[i
].p_memsz
, (void *)phdr
[i
].p_paddr
);
271 /* We expect a simple linear address space: every segment must
272 * have the same difference between virtual (p_vaddr) and
273 * physical (p_paddr) address. */
275 *page_offset
= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
;
276 else if (*page_offset
!= phdr
[i
].p_vaddr
- phdr
[i
].p_paddr
)
277 errx(1, "Page offset of section %i different", i
);
279 /* We track the first and last address we mapped, so we can
280 * tell entry_point() where to scan. */
281 if (from_guest_phys(phdr
[i
].p_paddr
) < start
)
282 start
= from_guest_phys(phdr
[i
].p_paddr
);
283 if (from_guest_phys(phdr
[i
].p_paddr
) + phdr
[i
].p_filesz
> end
)
284 end
=from_guest_phys(phdr
[i
].p_paddr
)+phdr
[i
].p_filesz
;
286 /* We map this section of the file at its physical address. */
287 map_at(elf_fd
, from_guest_phys(phdr
[i
].p_paddr
),
288 phdr
[i
].p_offset
, phdr
[i
].p_filesz
);
291 return entry_point(start
, end
, *page_offset
);
294 /*L:170 Prepare to be SHOCKED and AMAZED. And possibly a trifle nauseated.
296 * We know that CONFIG_PAGE_OFFSET sets what virtual address the kernel expects
297 * to be. We don't know what that option was, but we can figure it out
298 * approximately by looking at the addresses in the code. I chose the common
299 * case of reading a memory location into the %eax register:
301 * movl <some-address>, %eax
303 * This gets encoded as five bytes: "0xA1 <4-byte-address>". For example,
304 * "0xA1 0x18 0x60 0x47 0xC0" reads the address 0xC0476018 into %eax.
306 * In this example can guess that the kernel was compiled with
307 * CONFIG_PAGE_OFFSET set to 0xC0000000 (it's always a round number). If the
308 * kernel were larger than 16MB, we might see 0xC1 addresses show up, but our
309 * kernel isn't that bloated yet.
311 * Unfortunately, x86 has variable-length instructions, so finding this
312 * particular instruction properly involves writing a disassembler. Instead,
313 * we rely on statistics. We look for "0xA1" and tally the different bytes
314 * which occur 4 bytes later (the "0xC0" in our example above). When one of
315 * those bytes appears three times, we can be reasonably confident that it
316 * forms the start of CONFIG_PAGE_OFFSET.
318 * This is amazingly reliable. */
319 static unsigned long intuit_page_offset(unsigned char *img
, unsigned long len
)
321 unsigned int i
, possibilities
[256] = { 0 };
323 for (i
= 0; i
+ 4 < len
; i
++) {
324 /* mov 0xXXXXXXXX,%eax */
325 if (img
[i
] == 0xA1 && ++possibilities
[img
[i
+4]] > 3)
326 return (unsigned long)img
[i
+4] << 24;
328 errx(1, "could not determine page offset");
331 /*L:160 Unfortunately the entire ELF image isn't compressed: the segments
332 * which need loading are extracted and compressed raw. This denies us the
333 * information we need to make a fully-general loader. */
334 static unsigned long unpack_bzimage(int fd
, unsigned long *page_offset
)
338 /* A bzImage always gets loaded at physical address 1M. This is
339 * actually configurable as CONFIG_PHYSICAL_START, but as the comment
340 * there says, "Don't change this unless you know what you are doing".
342 void *img
= from_guest_phys(0x100000);
344 /* gzdopen takes our file descriptor (carefully placed at the start of
345 * the GZIP header we found) and returns a gzFile. */
346 f
= gzdopen(fd
, "rb");
347 /* We read it into memory in 64k chunks until we hit the end. */
348 while ((ret
= gzread(f
, img
+ len
, 65536)) > 0)
351 err(1, "reading image from bzImage");
353 verbose("Unpacked size %i addr %p\n", len
, img
);
355 /* Without the ELF header, we can't tell virtual-physical gap. This is
356 * CONFIG_PAGE_OFFSET, and people do actually change it. Fortunately,
357 * I have a clever way of figuring it out from the code itself. */
358 *page_offset
= intuit_page_offset(img
, len
);
360 return entry_point(img
, img
+ len
, *page_offset
);
363 /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're
364 * supposed to jump into it and it will unpack itself. We can't do that
365 * because the Guest can't run the unpacking code, and adding features to
366 * lguest kills puppies, so we don't want to.
368 * The bzImage is formed by putting the decompressing code in front of the
369 * compressed kernel code. So we can simple scan through it looking for the
370 * first "gzip" header, and start decompressing from there. */
371 static unsigned long load_bzimage(int fd
, unsigned long *page_offset
)
376 /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */
377 while (read(fd
, &c
, 1) == 1) {
393 /* Seek back to the start of the gzip header. */
394 lseek(fd
, -10, SEEK_CUR
);
395 /* One final check: "compressed under UNIX". */
399 return unpack_bzimage(fd
, page_offset
);
402 errx(1, "Could not find kernel in bzImage");
405 /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels
406 * come wrapped up in the self-decompressing "bzImage" format. With some funky
407 * coding, we can load those, too. */
408 static unsigned long load_kernel(int fd
, unsigned long *page_offset
)
412 /* Read in the first few bytes. */
413 if (read(fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
414 err(1, "Reading kernel");
416 /* If it's an ELF file, it starts with "\177ELF" */
417 if (memcmp(hdr
.e_ident
, ELFMAG
, SELFMAG
) == 0)
418 return map_elf(fd
, &hdr
, page_offset
);
420 /* Otherwise we assume it's a bzImage, and try to unpack it */
421 return load_bzimage(fd
, page_offset
);
424 /* This is a trivial little helper to align pages. Andi Kleen hated it because
425 * it calls getpagesize() twice: "it's dumb code."
427 * Kernel guys get really het up about optimization, even when it's not
428 * necessary. I leave this code as a reaction against that. */
429 static inline unsigned long page_align(unsigned long addr
)
431 /* Add upwards and truncate downwards. */
432 return ((addr
+ getpagesize()-1) & ~(getpagesize()-1));
435 /*L:180 An "initial ram disk" is a disk image loaded into memory along with
436 * the kernel which the kernel can use to boot from without needing any
437 * drivers. Most distributions now use this as standard: the initrd contains
438 * the code to load the appropriate driver modules for the current machine.
440 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
441 * kernels. He sent me this (and tells me when I break it). */
442 static unsigned long load_initrd(const char *name
, unsigned long mem
)
448 ifd
= open_or_die(name
, O_RDONLY
);
449 /* fstat() is needed to get the file size. */
450 if (fstat(ifd
, &st
) < 0)
451 err(1, "fstat() on initrd '%s'", name
);
453 /* We map the initrd at the top of memory, but mmap wants it to be
454 * page-aligned, so we round the size up for that. */
455 len
= page_align(st
.st_size
);
456 map_at(ifd
, from_guest_phys(mem
- len
), 0, st
.st_size
);
457 /* Once a file is mapped, you can close the file descriptor. It's a
458 * little odd, but quite useful. */
460 verbose("mapped initrd %s size=%lu @ %p\n", name
, len
, (void*)mem
-len
);
462 /* We return the initrd size. */
466 /* Once we know the address the Guest kernel expects, we can construct simple
467 * linear page tables for all of memory which will get the Guest far enough
468 * into the boot to create its own.
470 * We lay them out of the way, just below the initrd (which is why we need to
472 static unsigned long setup_pagetables(unsigned long mem
,
473 unsigned long initrd_size
,
474 unsigned long page_offset
)
476 unsigned long *pgdir
, *linear
;
477 unsigned int mapped_pages
, i
, linear_pages
;
478 unsigned int ptes_per_page
= getpagesize()/sizeof(void *);
480 /* Ideally we map all physical memory starting at page_offset.
481 * However, if page_offset is 0xC0000000 we can only map 1G of physical
482 * (0xC0000000 + 1G overflows). */
483 if (mem
<= -page_offset
)
484 mapped_pages
= mem
/getpagesize();
486 mapped_pages
= -page_offset
/getpagesize();
488 /* Each PTE page can map ptes_per_page pages: how many do we need? */
489 linear_pages
= (mapped_pages
+ ptes_per_page
-1)/ptes_per_page
;
491 /* We put the toplevel page directory page at the top of memory. */
492 pgdir
= from_guest_phys(mem
) - initrd_size
- getpagesize();
494 /* Now we use the next linear_pages pages as pte pages */
495 linear
= (void *)pgdir
- linear_pages
*getpagesize();
497 /* Linear mapping is easy: put every page's address into the mapping in
498 * order. PAGE_PRESENT contains the flags Present, Writable and
500 for (i
= 0; i
< mapped_pages
; i
++)
501 linear
[i
] = ((i
* getpagesize()) | PAGE_PRESENT
);
503 /* The top level points to the linear page table pages above. The
504 * entry representing page_offset points to the first one, and they
505 * continue from there. */
506 for (i
= 0; i
< mapped_pages
; i
+= ptes_per_page
) {
507 pgdir
[(i
+ page_offset
/getpagesize())/ptes_per_page
]
508 = ((to_guest_phys(linear
) + i
*sizeof(void *))
512 verbose("Linear mapping of %u pages in %u pte pages at %#lx\n",
513 mapped_pages
, linear_pages
, to_guest_phys(linear
));
515 /* We return the top level (guest-physical) address: the kernel needs
516 * to know where it is. */
517 return to_guest_phys(pgdir
);
520 /* Simple routine to roll all the commandline arguments together with spaces
522 static void concat(char *dst
, char *args
[])
524 unsigned int i
, len
= 0;
526 for (i
= 0; args
[i
]; i
++) {
527 strcpy(dst
+len
, args
[i
]);
528 strcat(dst
+len
, " ");
529 len
+= strlen(args
[i
]) + 1;
531 /* In case it's empty. */
535 /* This is where we actually tell the kernel to initialize the Guest. We saw
536 * the arguments it expects when we looked at initialize() in lguest_user.c:
537 * the base of guest "physical" memory, the top physical page to allow, the
538 * top level pagetable, the entry point and the page_offset constant for the
540 static int tell_kernel(unsigned long pgdir
, unsigned long start
,
541 unsigned long page_offset
)
543 unsigned long args
[] = { LHREQ_INITIALIZE
,
544 (unsigned long)guest_base
,
545 guest_limit
/ getpagesize(),
546 pgdir
, start
, page_offset
};
549 verbose("Guest: %p - %p (%#lx)\n",
550 guest_base
, guest_base
+ guest_limit
, guest_limit
);
551 fd
= open_or_die("/dev/lguest", O_RDWR
);
552 if (write(fd
, args
, sizeof(args
)) < 0)
553 err(1, "Writing to /dev/lguest");
555 /* We return the /dev/lguest file descriptor to control this Guest */
560 static void set_fd(int fd
, struct device_list
*devices
)
562 FD_SET(fd
, &devices
->infds
);
563 if (fd
> devices
->max_infd
)
564 devices
->max_infd
= fd
;
570 * With a console and network devices, we can have lots of input which we need
571 * to process. We could try to tell the kernel what file descriptors to watch,
572 * but handing a file descriptor mask through to the kernel is fairly icky.
574 * Instead, we fork off a process which watches the file descriptors and writes
575 * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host
576 * loop to stop running the Guest. This causes it to return from the
577 * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset
578 * the LHREQ_BREAK and wake us up again.
580 * This, of course, is merely a different *kind* of icky.
582 static void wake_parent(int pipefd
, int lguest_fd
, struct device_list
*devices
)
584 /* Add the pipe from the Launcher to the fdset in the device_list, so
585 * we watch it, too. */
586 set_fd(pipefd
, devices
);
589 fd_set rfds
= devices
->infds
;
590 unsigned long args
[] = { LHREQ_BREAK
, 1 };
592 /* Wait until input is ready from one of the devices. */
593 select(devices
->max_infd
+1, &rfds
, NULL
, NULL
, NULL
);
594 /* Is it a message from the Launcher? */
595 if (FD_ISSET(pipefd
, &rfds
)) {
597 /* If read() returns 0, it means the Launcher has
598 * exited. We silently follow. */
599 if (read(pipefd
, &ignorefd
, sizeof(ignorefd
)) == 0)
601 /* Otherwise it's telling us there's a problem with one
602 * of the devices, and we should ignore that file
603 * descriptor from now on. */
604 FD_CLR(ignorefd
, &devices
->infds
);
605 } else /* Send LHREQ_BREAK command. */
606 write(lguest_fd
, args
, sizeof(args
));
610 /* This routine just sets up a pipe to the Waker process. */
611 static int setup_waker(int lguest_fd
, struct device_list
*device_list
)
613 int pipefd
[2], child
;
615 /* We create a pipe to talk to the waker, and also so it knows when the
616 * Launcher dies (and closes pipe). */
623 /* Close the "writing" end of our copy of the pipe */
625 wake_parent(pipefd
[0], lguest_fd
, device_list
);
627 /* Close the reading end of our copy of the pipe. */
630 /* Here is the fd used to talk to the waker. */
637 * When the Guest sends DMA to us, it sends us an array of addresses and sizes.
638 * We need to make sure it's not trying to reach into the Launcher itself, so
639 * we have a convenient routine which check it and exits with an error message
640 * if something funny is going on:
642 static void *_check_pointer(unsigned long addr
, unsigned int size
,
645 /* We have to separately check addr and addr+size, because size could
646 * be huge and addr + size might wrap around. */
647 if (addr
>= guest_limit
|| addr
+ size
>= guest_limit
)
648 errx(1, "%s:%i: Invalid address %li", __FILE__
, line
, addr
);
649 /* We return a pointer for the caller's convenience, now we know it's
651 return from_guest_phys(addr
);
653 /* A macro which transparently hands the line number to the real function. */
654 #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
656 /* The Guest has given us the address of a "struct lguest_dma". We check it's
657 * OK and convert it to an iovec (which is a simple array of ptr/size
659 static u32
*dma2iov(unsigned long dma
, struct iovec iov
[], unsigned *num
)
662 struct lguest_dma
*udma
;
664 /* First we make sure that the array memory itself is valid. */
665 udma
= check_pointer(dma
, sizeof(*udma
));
666 /* Now we check each element */
667 for (i
= 0; i
< LGUEST_MAX_DMA_SECTIONS
; i
++) {
668 /* A zero length ends the array. */
672 iov
[i
].iov_base
= check_pointer(udma
->addr
[i
], udma
->len
[i
]);
673 iov
[i
].iov_len
= udma
->len
[i
];
677 /* We return the pointer to where the caller should write the amount of
678 * the buffer used. */
679 return &udma
->used_len
;
682 /* This routine gets a DMA buffer from the Guest for a given key, and converts
683 * it to an iovec array. It returns the interrupt the Guest wants when we're
684 * finished, and a pointer to the "used_len" field to fill in. */
685 static u32
*get_dma_buffer(int fd
, void *key
,
686 struct iovec iov
[], unsigned int *num
, u32
*irq
)
688 unsigned long buf
[] = { LHREQ_GETDMA
, to_guest_phys(key
) };
692 /* Ask the kernel for a DMA buffer corresponding to this key. */
693 udma
= write(fd
, buf
, sizeof(buf
));
694 /* They haven't registered any, or they're all used? */
695 if (udma
== (unsigned long)-1)
698 /* Convert it into our iovec array */
699 res
= dma2iov(udma
, iov
, num
);
700 /* The kernel stashes irq in ->used_len to get it out to us. */
702 /* Return a pointer to ((struct lguest_dma *)udma)->used_len. */
706 /* This is a convenient routine to send the Guest an interrupt. */
707 static void trigger_irq(int fd
, u32 irq
)
709 unsigned long buf
[] = { LHREQ_IRQ
, irq
};
710 if (write(fd
, buf
, sizeof(buf
)) != 0)
711 err(1, "Triggering irq %i", irq
);
714 /* This simply sets up an iovec array where we can put data to be discarded.
715 * This happens when the Guest doesn't want or can't handle the input: we have
716 * to get rid of it somewhere, and if we bury it in the ceiling space it will
717 * start to smell after a week. */
718 static void discard_iovec(struct iovec
*iov
, unsigned int *num
)
720 static char discard_buf
[1024];
722 iov
->iov_base
= discard_buf
;
723 iov
->iov_len
= sizeof(discard_buf
);
726 /* Here is the input terminal setting we save, and the routine to restore them
727 * on exit so the user can see what they type next. */
728 static struct termios orig_term
;
729 static void restore_term(void)
731 tcsetattr(STDIN_FILENO
, TCSANOW
, &orig_term
);
734 /* We associate some data with the console for our exit hack. */
737 /* How many times have they hit ^C? */
739 /* When did they start? */
740 struct timeval start
;
743 /* This is the routine which handles console input (ie. stdin). */
744 static bool handle_console_input(int fd
, struct device
*dev
)
749 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
750 struct console_abort
*abort
= dev
->priv
;
752 /* First we get the console buffer from the Guest. The key is dev->mem
753 * which was set to 0 in setup_console(). */
754 lenp
= get_dma_buffer(fd
, dev
->mem
, iov
, &num
, &irq
);
756 /* If it's not ready for input, warn and set up to discard. */
757 warn("console: no dma buffer!");
758 discard_iovec(iov
, &num
);
761 /* This is why we convert to iovecs: the readv() call uses them, and so
762 * it reads straight into the Guest's buffer. */
763 len
= readv(dev
->fd
, iov
, num
);
765 /* This implies that the console is closed, is /dev/null, or
766 * something went terribly wrong. We still go through the rest
767 * of the logic, though, especially the exit handling below. */
768 warnx("Failed to get console input, ignoring console.");
772 /* If we read the data into the Guest, fill in the length and send the
776 trigger_irq(fd
, irq
);
779 /* Three ^C within one second? Exit.
781 * This is such a hack, but works surprisingly well. Each ^C has to be
782 * in a buffer by itself, so they can't be too fast. But we check that
783 * we get three within about a second, so they can't be too slow. */
784 if (len
== 1 && ((char *)iov
[0].iov_base
)[0] == 3) {
786 gettimeofday(&abort
->start
, NULL
);
787 else if (abort
->count
== 3) {
789 gettimeofday(&now
, NULL
);
790 if (now
.tv_sec
<= abort
->start
.tv_sec
+1) {
791 unsigned long args
[] = { LHREQ_BREAK
, 0 };
792 /* Close the fd so Waker will know it has to
795 /* Just in case waker is blocked in BREAK, send
797 write(fd
, args
, sizeof(args
));
803 /* Any other key resets the abort counter. */
806 /* Now, if we didn't read anything, put the input terminal back and
807 * return failure (meaning, don't call us again). */
812 /* Everything went OK! */
816 /* Handling console output is much simpler than input. */
817 static u32
handle_console_output(int fd
, const struct iovec
*iov
,
818 unsigned num
, struct device
*dev
)
820 /* Whatever the Guest sends, write it to standard output. Return the
821 * number of bytes written. */
822 return writev(STDOUT_FILENO
, iov
, num
);
825 /* Guest->Host network output is also pretty easy. */
826 static u32
handle_tun_output(int fd
, const struct iovec
*iov
,
827 unsigned num
, struct device
*dev
)
829 /* We put a flag in the "priv" pointer of the network device, and set
830 * it as soon as we see output. We'll see why in handle_tun_input() */
831 *(bool *)dev
->priv
= true;
832 /* Whatever packet the Guest sent us, write it out to the tun
834 return writev(dev
->fd
, iov
, num
);
837 /* This matches the peer_key() in lguest_net.c. The key for any given slot
838 * is the address of the network device's page plus 4 * the slot number. */
839 static unsigned long peer_offset(unsigned int peernum
)
844 /* This is where we handle a packet coming in from the tun device */
845 static bool handle_tun_input(int fd
, struct device
*dev
)
850 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
852 /* First we get a buffer the Guest has bound to its key. */
853 lenp
= get_dma_buffer(fd
, dev
->mem
+peer_offset(NET_PEERNUM
), iov
, &num
,
856 /* Now, it's expected that if we try to send a packet too
857 * early, the Guest won't be ready yet. This is why we set a
858 * flag when the Guest sends its first packet. If it's sent a
859 * packet we assume it should be ready to receive them.
861 * Actually, this is what the status bits in the descriptor are
862 * for: we should *use* them. FIXME! */
863 if (*(bool *)dev
->priv
)
864 warn("network: no dma buffer!");
865 discard_iovec(iov
, &num
);
868 /* Read the packet from the device directly into the Guest's buffer. */
869 len
= readv(dev
->fd
, iov
, num
);
871 err(1, "reading network");
873 /* Write the used_len, and trigger the interrupt for the Guest */
876 trigger_irq(fd
, irq
);
878 verbose("tun input packet len %i [%02x %02x] (%s)\n", len
,
879 ((u8
*)iov
[0].iov_base
)[0], ((u8
*)iov
[0].iov_base
)[1],
880 lenp
? "sent" : "discarded");
885 /* The last device handling routine is block output: the Guest has sent a DMA
886 * to the block device. It will have placed the command it wants in the
887 * "struct lguest_block_page". */
888 static u32
handle_block_output(int fd
, const struct iovec
*iov
,
889 unsigned num
, struct device
*dev
)
891 struct lguest_block_page
*p
= dev
->mem
;
893 unsigned int len
, reply_num
;
894 struct iovec reply
[LGUEST_MAX_DMA_SECTIONS
];
895 off64_t device_len
, off
= (off64_t
)p
->sector
* 512;
897 /* First we extract the device length from the dev->priv pointer. */
898 device_len
= *(off64_t
*)dev
->priv
;
900 /* We first check that the read or write is within the length of the
902 if (off
>= device_len
)
903 errx(1, "Bad offset %llu vs %llu", off
, device_len
);
904 /* Move to the right location in the block file. This shouldn't fail,
905 * but best to check. */
906 if (lseek64(dev
->fd
, off
, SEEK_SET
) != off
)
907 err(1, "Bad seek to sector %i", p
->sector
);
909 verbose("Block: %s at offset %llu\n", p
->type
? "WRITE" : "READ", off
);
911 /* They were supposed to bind a reply buffer at key equal to the start
912 * of the block device memory. We need this to tell them when the
913 * request is finished. */
914 lenp
= get_dma_buffer(fd
, dev
->mem
, reply
, &reply_num
, &irq
);
916 err(1, "Block request didn't give us a dma buffer");
919 /* A write request. The DMA they sent contained the data, so
921 len
= writev(dev
->fd
, iov
, num
);
922 /* Grr... Now we know how long the "struct lguest_dma" they
923 * sent was, we make sure they didn't try to write over the end
924 * of the block file (possibly extending it). */
925 if (off
+ len
> device_len
) {
926 /* Trim it back to the correct length */
927 ftruncate64(dev
->fd
, device_len
);
928 /* Die, bad Guest, die. */
929 errx(1, "Write past end %llu+%u", off
, len
);
931 /* The reply length is 0: we just send back an empty DMA to
932 * interrupt them and tell them the write is finished. */
935 /* A read request. They sent an empty DMA to start the
936 * request, and we put the read contents into the reply
938 len
= readv(dev
->fd
, reply
, reply_num
);
942 /* The result is 1 (done), 2 if there was an error (short read or
944 p
->result
= 1 + (p
->bytes
!= len
);
945 /* Now tell them we've used their reply buffer. */
946 trigger_irq(fd
, irq
);
948 /* We're supposed to return the number of bytes of the output buffer we
949 * used. But the block device uses the "result" field instead, so we
954 /* This is the generic routine we call when the Guest sends some DMA out. */
955 static void handle_output(int fd
, unsigned long dma
, unsigned long key
,
956 struct device_list
*devices
)
960 struct iovec iov
[LGUEST_MAX_DMA_SECTIONS
];
963 /* Convert the "struct lguest_dma" they're sending to a "struct
965 lenp
= dma2iov(dma
, iov
, &num
);
967 /* Check each device: if they expect output to this key, tell them to
969 for (i
= devices
->dev
; i
; i
= i
->next
) {
970 if (i
->handle_output
&& key
== i
->watch_key
) {
971 /* We write the result straight into the used_len field
973 *lenp
= i
->handle_output(fd
, iov
, num
, i
);
978 /* This can happen: the kernel sends any SEND_DMA which doesn't match
979 * another Guest to us. It could be that another Guest just left a
980 * network, for example. But it's unusual. */
981 warnx("Pending dma %p, key %p", (void *)dma
, (void *)key
);
984 /* This is called when the waker wakes us up: check for incoming file
986 static void handle_input(int fd
, struct device_list
*devices
)
988 /* select() wants a zeroed timeval to mean "don't wait". */
989 struct timeval poll
= { .tv_sec
= 0, .tv_usec
= 0 };
993 fd_set fds
= devices
->infds
;
995 /* If nothing is ready, we're done. */
996 if (select(devices
->max_infd
+1, &fds
, NULL
, NULL
, &poll
) == 0)
999 /* Otherwise, call the device(s) which have readable
1000 * file descriptors and a method of handling them. */
1001 for (i
= devices
->dev
; i
; i
= i
->next
) {
1002 if (i
->handle_input
&& FD_ISSET(i
->fd
, &fds
)) {
1003 /* If handle_input() returns false, it means we
1004 * should no longer service it.
1005 * handle_console_input() does this. */
1006 if (!i
->handle_input(fd
, i
)) {
1007 /* Clear it from the set of input file
1008 * descriptors kept at the head of the
1010 FD_CLR(i
->fd
, &devices
->infds
);
1011 /* Tell waker to ignore it too... */
1012 write(waker_fd
, &i
->fd
, sizeof(i
->fd
));
1022 * All devices need a descriptor so the Guest knows it exists, and a "struct
1023 * device" so the Launcher can keep track of it. We have common helper
1024 * routines to allocate them.
1026 * This routine allocates a new "struct lguest_device_desc" from descriptor
1027 * table in the devices array just above the Guest's normal memory. */
1028 static struct lguest_device_desc
*
1029 new_dev_desc(struct lguest_device_desc
*descs
,
1030 u16 type
, u16 features
, u16 num_pages
)
1034 for (i
= 0; i
< LGUEST_MAX_DEVICES
; i
++) {
1035 if (!descs
[i
].type
) {
1036 descs
[i
].type
= type
;
1037 descs
[i
].features
= features
;
1038 descs
[i
].num_pages
= num_pages
;
1039 /* If they said the device needs memory, we allocate
1043 pa
= to_guest_phys(get_pages(num_pages
));
1044 descs
[i
].pfn
= pa
/ getpagesize();
1049 errx(1, "too many devices");
1052 /* This monster routine does all the creation and setup of a new device,
1053 * including caling new_dev_desc() to allocate the descriptor and device
1055 static struct device
*new_device(struct device_list
*devices
,
1056 u16 type
, u16 num_pages
, u16 features
,
1058 bool (*handle_input
)(int, struct device
*),
1059 unsigned long watch_off
,
1060 u32 (*handle_output
)(int,
1061 const struct iovec
*,
1065 struct device
*dev
= malloc(sizeof(*dev
));
1067 /* Append to device list. Prepending to a single-linked list is
1068 * easier, but the user expects the devices to be arranged on the bus
1069 * in command-line order. The first network device on the command line
1070 * is eth0, the first block device /dev/lgba, etc. */
1071 *devices
->lastdev
= dev
;
1073 devices
->lastdev
= &dev
->next
;
1075 /* Now we populate the fields one at a time. */
1077 /* If we have an input handler for this file descriptor, then we add it
1078 * to the device_list's fdset and maxfd. */
1080 set_fd(dev
->fd
, devices
);
1081 dev
->desc
= new_dev_desc(devices
->descs
, type
, features
, num_pages
);
1082 dev
->mem
= from_guest_phys(dev
->desc
->pfn
* getpagesize());
1083 dev
->handle_input
= handle_input
;
1084 dev
->watch_key
= to_guest_phys(dev
->mem
) + watch_off
;
1085 dev
->handle_output
= handle_output
;
1089 /* Our first setup routine is the console. It's a fairly simple device, but
1090 * UNIX tty handling makes it uglier than it could be. */
1091 static void setup_console(struct device_list
*devices
)
1095 /* If we can save the initial standard input settings... */
1096 if (tcgetattr(STDIN_FILENO
, &orig_term
) == 0) {
1097 struct termios term
= orig_term
;
1098 /* Then we turn off echo, line buffering and ^C etc. We want a
1099 * raw input stream to the Guest. */
1100 term
.c_lflag
&= ~(ISIG
|ICANON
|ECHO
);
1101 tcsetattr(STDIN_FILENO
, TCSANOW
, &term
);
1102 /* If we exit gracefully, the original settings will be
1103 * restored so the user can see what they're typing. */
1104 atexit(restore_term
);
1107 /* We don't currently require any memory for the console, so we ask for
1109 dev
= new_device(devices
, LGUEST_DEVICE_T_CONSOLE
, 0, 0,
1110 STDIN_FILENO
, handle_console_input
,
1111 LGUEST_CONSOLE_DMA_KEY
, handle_console_output
);
1112 /* We store the console state in dev->priv, and initialize it. */
1113 dev
->priv
= malloc(sizeof(struct console_abort
));
1114 ((struct console_abort
*)dev
->priv
)->count
= 0;
1115 verbose("device %p: console\n",
1116 (void *)(dev
->desc
->pfn
* getpagesize()));
1119 /* Setting up a block file is also fairly straightforward. */
1120 static void setup_block_file(const char *filename
, struct device_list
*devices
)
1124 off64_t
*device_len
;
1125 struct lguest_block_page
*p
;
1127 /* We open with O_LARGEFILE because otherwise we get stuck at 2G. We
1128 * open with O_DIRECT because otherwise our benchmarks go much too
1130 fd
= open_or_die(filename
, O_RDWR
|O_LARGEFILE
|O_DIRECT
);
1132 /* We want one page, and have no input handler (the block file never
1133 * has anything interesting to say to us). Our timing will be quite
1134 * random, so it should be a reasonable randomness source. */
1135 dev
= new_device(devices
, LGUEST_DEVICE_T_BLOCK
, 1,
1136 LGUEST_DEVICE_F_RANDOMNESS
,
1137 fd
, NULL
, 0, handle_block_output
);
1139 /* We store the device size in the private area */
1140 device_len
= dev
->priv
= malloc(sizeof(*device_len
));
1141 /* This is the safe way of establishing the size of our device: it
1142 * might be a normal file or an actual block device like /dev/hdb. */
1143 *device_len
= lseek64(fd
, 0, SEEK_END
);
1145 /* The device memory is a "struct lguest_block_page". It's zeroed
1146 * already, we just need to put in the device size. Block devices
1147 * think in sectors (ie. 512 byte chunks), so we translate here. */
1149 p
->num_sectors
= *device_len
/512;
1150 verbose("device %p: block %i sectors\n",
1151 (void *)(dev
->desc
->pfn
* getpagesize()), p
->num_sectors
);
1157 * Setting up network devices is quite a pain, because we have three types.
1158 * First, we have the inter-Guest network. This is a file which is mapped into
1159 * the address space of the Guests who are on the network. Because it is a
1160 * shared mapping, the same page underlies all the devices, and they can send
1161 * DMA to each other.
1163 * Remember from our network driver, the Guest is told what slot in the page it
1164 * is to use. We use exclusive fnctl locks to reserve a slot. If another
1165 * Guest is using a slot, the lock will fail and we try another. Because fnctl
1166 * locks are cleaned up automatically when we die, this cleverly means that our
1167 * reservation on the slot will vanish if we crash. */
1168 static unsigned int find_slot(int netfd
, const char *filename
)
1172 fl
.l_type
= F_WRLCK
;
1173 fl
.l_whence
= SEEK_SET
;
1175 /* Try a 1 byte lock in each possible position number */
1176 for (fl
.l_start
= 0;
1177 fl
.l_start
< getpagesize()/sizeof(struct lguest_net
);
1179 /* If we succeed, return the slot number. */
1180 if (fcntl(netfd
, F_SETLK
, &fl
) == 0)
1183 errx(1, "No free slots in network file %s", filename
);
1186 /* This function sets up the network file */
1187 static void setup_net_file(const char *filename
,
1188 struct device_list
*devices
)
1193 /* We don't use open_or_die() here: for friendliness we create the file
1194 * if it doesn't already exist. */
1195 netfd
= open(filename
, O_RDWR
, 0);
1197 if (errno
== ENOENT
) {
1198 netfd
= open(filename
, O_RDWR
|O_CREAT
, 0600);
1200 /* If we succeeded, initialize the file with a
1202 char page
[getpagesize()];
1203 memset(page
, 0, sizeof(page
));
1204 write(netfd
, page
, sizeof(page
));
1208 err(1, "cannot open net file '%s'", filename
);
1211 /* We need 1 page, and the features indicate the slot to use and that
1212 * no checksum is needed. We never touch this device again; it's
1213 * between the Guests on the network, so we don't register input or
1214 * output handlers. */
1215 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1216 find_slot(netfd
, filename
)|LGUEST_NET_F_NOCSUM
,
1219 /* Map the shared file. */
1220 if (mmap(dev
->mem
, getpagesize(), PROT_READ
|PROT_WRITE
,
1221 MAP_FIXED
|MAP_SHARED
, netfd
, 0) != dev
->mem
)
1222 err(1, "could not mmap '%s'", filename
);
1223 verbose("device %p: shared net %s, peer %i\n",
1224 (void *)(dev
->desc
->pfn
* getpagesize()), filename
,
1225 dev
->desc
->features
& ~LGUEST_NET_F_NOCSUM
);
1229 static u32
str2ip(const char *ipaddr
)
1231 unsigned int byte
[4];
1233 sscanf(ipaddr
, "%u.%u.%u.%u", &byte
[0], &byte
[1], &byte
[2], &byte
[3]);
1234 return (byte
[0] << 24) | (byte
[1] << 16) | (byte
[2] << 8) | byte
[3];
1237 /* This code is "adapted" from libbridge: it attaches the Host end of the
1238 * network device to the bridge device specified by the command line.
1240 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1241 * dislike bridging), and I just try not to break it. */
1242 static void add_to_bridge(int fd
, const char *if_name
, const char *br_name
)
1248 errx(1, "must specify bridge name");
1250 ifidx
= if_nametoindex(if_name
);
1252 errx(1, "interface %s does not exist!", if_name
);
1254 strncpy(ifr
.ifr_name
, br_name
, IFNAMSIZ
);
1255 ifr
.ifr_ifindex
= ifidx
;
1256 if (ioctl(fd
, SIOCBRADDIF
, &ifr
) < 0)
1257 err(1, "can't add %s to bridge %s", if_name
, br_name
);
1260 /* This sets up the Host end of the network device with an IP address, brings
1261 * it up so packets will flow, the copies the MAC address into the hwaddr
1262 * pointer (in practice, the Host's slot in the network device's memory). */
1263 static void configure_device(int fd
, const char *devname
, u32 ipaddr
,
1264 unsigned char hwaddr
[6])
1267 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&ifr
.ifr_addr
;
1269 /* Don't read these incantations. Just cut & paste them like I did! */
1270 memset(&ifr
, 0, sizeof(ifr
));
1271 strcpy(ifr
.ifr_name
, devname
);
1272 sin
->sin_family
= AF_INET
;
1273 sin
->sin_addr
.s_addr
= htonl(ipaddr
);
1274 if (ioctl(fd
, SIOCSIFADDR
, &ifr
) != 0)
1275 err(1, "Setting %s interface address", devname
);
1276 ifr
.ifr_flags
= IFF_UP
;
1277 if (ioctl(fd
, SIOCSIFFLAGS
, &ifr
) != 0)
1278 err(1, "Bringing interface %s up", devname
);
1280 /* SIOC stands for Socket I/O Control. G means Get (vs S for Set
1281 * above). IF means Interface, and HWADDR is hardware address.
1283 if (ioctl(fd
, SIOCGIFHWADDR
, &ifr
) != 0)
1284 err(1, "getting hw address for %s", devname
);
1285 memcpy(hwaddr
, ifr
.ifr_hwaddr
.sa_data
, 6);
1288 /*L:195 The other kind of network is a Host<->Guest network. This can either
1289 * use briding or routing, but the principle is the same: it uses the "tun"
1290 * device to inject packets into the Host as if they came in from a normal
1291 * network card. We just shunt packets between the Guest and the tun
1293 static void setup_tun_net(const char *arg
, struct device_list
*devices
)
1299 const char *br_name
= NULL
;
1301 /* We open the /dev/net/tun device and tell it we want a tap device. A
1302 * tap device is like a tun device, only somehow different. To tell
1303 * the truth, I completely blundered my way through this code, but it
1305 netfd
= open_or_die("/dev/net/tun", O_RDWR
);
1306 memset(&ifr
, 0, sizeof(ifr
));
1307 ifr
.ifr_flags
= IFF_TAP
| IFF_NO_PI
;
1308 strcpy(ifr
.ifr_name
, "tap%d");
1309 if (ioctl(netfd
, TUNSETIFF
, &ifr
) != 0)
1310 err(1, "configuring /dev/net/tun");
1311 /* We don't need checksums calculated for packets coming in this
1312 * device: trust us! */
1313 ioctl(netfd
, TUNSETNOCSUM
, 1);
1315 /* We create the net device with 1 page, using the features field of
1316 * the descriptor to tell the Guest it is in slot 1 (NET_PEERNUM), and
1317 * that the device has fairly random timing. We do *not* specify
1318 * LGUEST_NET_F_NOCSUM: these packets can reach the real world.
1320 * We will put our MAC address is slot 0 for the Guest to see, so
1321 * it will send packets to us using the key "peer_offset(0)": */
1322 dev
= new_device(devices
, LGUEST_DEVICE_T_NET
, 1,
1323 NET_PEERNUM
|LGUEST_DEVICE_F_RANDOMNESS
, netfd
,
1324 handle_tun_input
, peer_offset(0), handle_tun_output
);
1326 /* We keep a flag which says whether we've seen packets come out from
1327 * this network device. */
1328 dev
->priv
= malloc(sizeof(bool));
1329 *(bool *)dev
->priv
= false;
1331 /* We need a socket to perform the magic network ioctls to bring up the
1332 * tap interface, connect to the bridge etc. Any socket will do! */
1333 ipfd
= socket(PF_INET
, SOCK_DGRAM
, IPPROTO_IP
);
1335 err(1, "opening IP socket");
1337 /* If the command line was --tunnet=bridge:<name> do bridging. */
1338 if (!strncmp(BRIDGE_PFX
, arg
, strlen(BRIDGE_PFX
))) {
1340 br_name
= arg
+ strlen(BRIDGE_PFX
);
1341 add_to_bridge(ipfd
, ifr
.ifr_name
, br_name
);
1342 } else /* It is an IP address to set up the device with */
1345 /* We are peer 0, ie. first slot, so we hand dev->mem to this routine
1346 * to write the MAC address at the start of the device memory. */
1347 configure_device(ipfd
, ifr
.ifr_name
, ip
, dev
->mem
);
1349 /* Set "promisc" bit: we want every single packet if we're going to
1350 * bridge to other machines (and otherwise it doesn't matter). */
1351 *((u8
*)dev
->mem
) |= 0x1;
1355 verbose("device %p: tun net %u.%u.%u.%u\n",
1356 (void *)(dev
->desc
->pfn
* getpagesize()),
1357 (u8
)(ip
>>24), (u8
)(ip
>>16), (u8
)(ip
>>8), (u8
)ip
);
1359 verbose("attached to bridge: %s\n", br_name
);
1361 /* That's the end of device setup. */
1363 /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves
1364 * its input and output, and finally, lays it to rest. */
1365 static void __attribute__((noreturn
))
1366 run_guest(int lguest_fd
, struct device_list
*device_list
)
1369 unsigned long args
[] = { LHREQ_BREAK
, 0 };
1370 unsigned long arr
[2];
1373 /* We read from the /dev/lguest device to run the Guest. */
1374 readval
= read(lguest_fd
, arr
, sizeof(arr
));
1376 /* The read can only really return sizeof(arr) (the Guest did a
1377 * SEND_DMA to us), or an error. */
1379 /* For a successful read, arr[0] is the address of the "struct
1380 * lguest_dma", and arr[1] is the key the Guest sent to. */
1381 if (readval
== sizeof(arr
)) {
1382 handle_output(lguest_fd
, arr
[0], arr
[1], device_list
);
1384 /* ENOENT means the Guest died. Reading tells us why. */
1385 } else if (errno
== ENOENT
) {
1386 char reason
[1024] = { 0 };
1387 read(lguest_fd
, reason
, sizeof(reason
)-1);
1388 errx(1, "%s", reason
);
1389 /* EAGAIN means the waker wanted us to look at some input.
1390 * Anything else means a bug or incompatible change. */
1391 } else if (errno
!= EAGAIN
)
1392 err(1, "Running guest failed");
1394 /* Service input, then unset the BREAK which releases
1396 handle_input(lguest_fd
, device_list
);
1397 if (write(lguest_fd
, args
, sizeof(args
)) < 0)
1398 err(1, "Resetting break");
1402 * This is the end of the Launcher.
1404 * But wait! We've seen I/O from the Launcher, and we've seen I/O from the
1405 * Drivers. If we were to see the Host kernel I/O code, our understanding
1406 * would be complete... :*/
1408 static struct option opts
[] = {
1409 { "verbose", 0, NULL
, 'v' },
1410 { "sharenet", 1, NULL
, 's' },
1411 { "tunnet", 1, NULL
, 't' },
1412 { "block", 1, NULL
, 'b' },
1413 { "initrd", 1, NULL
, 'i' },
1416 static void usage(void)
1418 errx(1, "Usage: lguest [--verbose] "
1419 "[--sharenet=<filename>|--tunnet=(<ipaddr>|bridge:<bridgename>)\n"
1420 "|--block=<filename>|--initrd=<filename>]...\n"
1421 "<mem-in-mb> vmlinux [args...]");
1424 /*L:105 The main routine is where the real work begins: */
1425 int main(int argc
, char *argv
[])
1427 /* Memory, top-level pagetable, code startpoint, PAGE_OFFSET and size
1428 * of the (optional) initrd. */
1429 unsigned long mem
= 0, pgdir
, start
, page_offset
, initrd_size
= 0;
1430 /* A temporary and the /dev/lguest file descriptor. */
1431 int i
, c
, lguest_fd
;
1432 /* The list of Guest devices, based on command line arguments. */
1433 struct device_list device_list
;
1434 /* The boot information for the Guest. */
1436 /* If they specify an initrd file to load. */
1437 const char *initrd_name
= NULL
;
1439 /* First we initialize the device list. Since console and network
1440 * device receive input from a file descriptor, we keep an fdset
1441 * (infds) and the maximum fd number (max_infd) with the head of the
1442 * list. We also keep a pointer to the last device, for easy appending
1444 device_list
.max_infd
= -1;
1445 device_list
.dev
= NULL
;
1446 device_list
.lastdev
= &device_list
.dev
;
1447 FD_ZERO(&device_list
.infds
);
1449 /* We need to know how much memory so we can set up the device
1450 * descriptor and memory pages for the devices as we parse the command
1451 * line. So we quickly look through the arguments to find the amount
1453 for (i
= 1; i
< argc
; i
++) {
1454 if (argv
[i
][0] != '-') {
1455 mem
= atoi(argv
[i
]) * 1024 * 1024;
1456 /* We start by mapping anonymous pages over all of
1457 * guest-physical memory range. This fills it with 0,
1458 * and ensures that the Guest won't be killed when it
1459 * tries to access it. */
1460 guest_base
= map_zeroed_pages(mem
/ getpagesize()
1463 guest_max
= mem
+ DEVICE_PAGES
*getpagesize();
1464 device_list
.descs
= get_pages(1);
1469 /* The options are fairly straight-forward */
1470 while ((c
= getopt_long(argc
, argv
, "v", opts
, NULL
)) != EOF
) {
1476 setup_net_file(optarg
, &device_list
);
1479 setup_tun_net(optarg
, &device_list
);
1482 setup_block_file(optarg
, &device_list
);
1485 initrd_name
= optarg
;
1488 warnx("Unknown argument %s", argv
[optind
]);
1492 /* After the other arguments we expect memory and kernel image name,
1493 * followed by command line arguments for the kernel. */
1494 if (optind
+ 2 > argc
)
1497 verbose("Guest base is at %p\n", guest_base
);
1499 /* We always have a console device */
1500 setup_console(&device_list
);
1502 /* Now we load the kernel */
1503 start
= load_kernel(open_or_die(argv
[optind
+1], O_RDONLY
),
1506 /* Boot information is stashed at physical address 0 */
1507 boot
= from_guest_phys(0);
1509 /* Map the initrd image if requested (at top of physical memory) */
1511 initrd_size
= load_initrd(initrd_name
, mem
);
1512 /* These are the location in the Linux boot header where the
1513 * start and size of the initrd are expected to be found. */
1514 *(unsigned long *)(boot
+0x218) = mem
- initrd_size
;
1515 *(unsigned long *)(boot
+0x21c) = initrd_size
;
1516 /* The bootloader type 0xFF means "unknown"; that's OK. */
1517 *(unsigned char *)(boot
+0x210) = 0xFF;
1520 /* Set up the initial linear pagetables, starting below the initrd. */
1521 pgdir
= setup_pagetables(mem
, initrd_size
, page_offset
);
1523 /* The Linux boot header contains an "E820" memory map: ours is a
1524 * simple, single region. */
1525 *(char*)(boot
+E820NR
) = 1;
1526 *((struct e820entry
*)(boot
+E820MAP
))
1527 = ((struct e820entry
) { 0, mem
, E820_RAM
});
1528 /* The boot header contains a command line pointer: we put the command
1529 * line after the boot header (at address 4096) */
1530 *(u32
*)(boot
+ 0x228) = 4096;
1531 concat(boot
+ 4096, argv
+optind
+2);
1533 /* The guest type value of "1" tells the Guest it's under lguest. */
1534 *(int *)(boot
+ 0x23c) = 1;
1536 /* We tell the kernel to initialize the Guest: this returns the open
1537 * /dev/lguest file descriptor. */
1538 lguest_fd
= tell_kernel(pgdir
, start
, page_offset
);
1540 /* We fork off a child process, which wakes the Launcher whenever one
1541 * of the input file descriptors needs attention. Otherwise we would
1542 * run the Guest until it tries to output something. */
1543 waker_fd
= setup_waker(lguest_fd
, &device_list
);
1545 /* Finally, run the Guest. This doesn't return. */
1546 run_guest(lguest_fd
, &device_list
);
1551 * Mastery is done: you now know everything I do.
1553 * But surely you have seen code, features and bugs in your wanderings which
1554 * you now yearn to attack? That is the real game, and I look forward to you
1555 * patching and forking lguest into the Your-Name-Here-visor.
1557 * Farewell, and good coding!