Commit | Line | Data |
---|---|---|
f938d2c8 RR |
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. | |
3c6b5bfa | 4 | :*/ |
8ca47e00 RR |
5 | #define _LARGEFILE64_SOURCE |
6 | #define _GNU_SOURCE | |
7 | #include <stdio.h> | |
8 | #include <string.h> | |
9 | #include <unistd.h> | |
10 | #include <err.h> | |
11 | #include <stdint.h> | |
12 | #include <stdlib.h> | |
13 | #include <elf.h> | |
14 | #include <sys/mman.h> | |
6649bb7a | 15 | #include <sys/param.h> |
8ca47e00 RR |
16 | #include <sys/types.h> |
17 | #include <sys/stat.h> | |
18 | #include <sys/wait.h> | |
19 | #include <fcntl.h> | |
20 | #include <stdbool.h> | |
21 | #include <errno.h> | |
22 | #include <ctype.h> | |
23 | #include <sys/socket.h> | |
24 | #include <sys/ioctl.h> | |
25 | #include <sys/time.h> | |
26 | #include <time.h> | |
27 | #include <netinet/in.h> | |
28 | #include <net/if.h> | |
29 | #include <linux/sockios.h> | |
30 | #include <linux/if_tun.h> | |
31 | #include <sys/uio.h> | |
32 | #include <termios.h> | |
33 | #include <getopt.h> | |
34 | #include <zlib.h> | |
dde79789 RR |
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. | |
37 | * | |
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 | |
40 | * here. | |
41 | */ | |
8ca47e00 RR |
42 | typedef unsigned long long u64; |
43 | typedef uint32_t u32; | |
44 | typedef uint16_t u16; | |
45 | typedef uint8_t u8; | |
b45d8cb0 RR |
46 | #include "linux/lguest_launcher.h" |
47 | #include "asm-x86/e820.h" | |
dde79789 | 48 | /*:*/ |
8ca47e00 RR |
49 | |
50 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
51 | #define NET_PEERNUM 1 | |
52 | #define BRIDGE_PFX "bridge:" | |
53 | #ifndef SIOCBRADDIF | |
54 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
55 | #endif | |
3c6b5bfa RR |
56 | /* We can have up to 256 pages for devices. */ |
57 | #define DEVICE_PAGES 256 | |
8ca47e00 | 58 | |
dde79789 RR |
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. */ | |
8ca47e00 RR |
61 | static bool verbose; |
62 | #define verbose(args...) \ | |
63 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
64 | /*:*/ |
65 | ||
66 | /* The pipe to send commands to the waker process */ | |
8ca47e00 | 67 | static int waker_fd; |
3c6b5bfa RR |
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; | |
8ca47e00 | 72 | |
dde79789 | 73 | /* This is our list of devices. */ |
8ca47e00 RR |
74 | struct device_list |
75 | { | |
dde79789 RR |
76 | /* Summary information about the devices in our list: ready to pass to |
77 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
78 | fd_set infds; |
79 | int max_infd; | |
80 | ||
dde79789 | 81 | /* The descriptor page for the devices. */ |
6570c459 | 82 | struct lguest_device_desc *descs; |
dde79789 RR |
83 | |
84 | /* A single linked list of devices. */ | |
8ca47e00 | 85 | struct device *dev; |
dde79789 | 86 | /* ... And an end pointer so we can easily append new devices */ |
8ca47e00 RR |
87 | struct device **lastdev; |
88 | }; | |
89 | ||
dde79789 | 90 | /* The device structure describes a single device. */ |
8ca47e00 RR |
91 | struct device |
92 | { | |
dde79789 | 93 | /* The linked-list pointer. */ |
8ca47e00 | 94 | struct device *next; |
dde79789 | 95 | /* The descriptor for this device, as mapped into the Guest. */ |
8ca47e00 | 96 | struct lguest_device_desc *desc; |
dde79789 | 97 | /* The memory page(s) of this device, if any. Also mapped in Guest. */ |
8ca47e00 RR |
98 | void *mem; |
99 | ||
dde79789 RR |
100 | /* If handle_input is set, it wants to be called when this file |
101 | * descriptor is ready. */ | |
8ca47e00 RR |
102 | int fd; |
103 | bool (*handle_input)(int fd, struct device *me); | |
104 | ||
dde79789 RR |
105 | /* If handle_output is set, it wants to be called when the Guest sends |
106 | * DMA to this key. */ | |
8ca47e00 RR |
107 | unsigned long watch_key; |
108 | u32 (*handle_output)(int fd, const struct iovec *iov, | |
109 | unsigned int num, struct device *me); | |
110 | ||
111 | /* Device-specific data. */ | |
112 | void *priv; | |
113 | }; | |
114 | ||
3c6b5bfa RR |
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. | |
120 | * | |
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. | |
124 | * | |
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) | |
129 | { | |
130 | return guest_base + addr; | |
131 | } | |
132 | ||
133 | static unsigned long to_guest_phys(const void *addr) | |
134 | { | |
135 | return (addr - guest_base); | |
136 | } | |
137 | ||
dde79789 RR |
138 | /*L:130 |
139 | * Loading the Kernel. | |
140 | * | |
141 | * We start with couple of simple helper routines. open_or_die() avoids | |
142 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
143 | static int open_or_die(const char *name, int flags) |
144 | { | |
145 | int fd = open(name, flags); | |
146 | if (fd < 0) | |
147 | err(1, "Failed to open %s", name); | |
148 | return fd; | |
149 | } | |
150 | ||
3c6b5bfa RR |
151 | /* map_zeroed_pages() takes a number of pages. */ |
152 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 153 | { |
3c6b5bfa RR |
154 | int fd = open_or_die("/dev/zero", O_RDONLY); |
155 | void *addr; | |
8ca47e00 | 156 | |
dde79789 | 157 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
158 | * copied). */ |
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); | |
163 | ||
164 | return addr; | |
165 | } | |
166 | ||
167 | /* Get some more pages for a device. */ | |
168 | static void *get_pages(unsigned int num) | |
169 | { | |
170 | void *addr = from_guest_phys(guest_limit); | |
171 | ||
172 | guest_limit += num * getpagesize(); | |
173 | if (guest_limit > guest_max) | |
174 | errx(1, "Not enough memory for devices"); | |
175 | return addr; | |
8ca47e00 RR |
176 | } |
177 | ||
dde79789 RR |
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. */ | |
3c6b5bfa | 181 | static unsigned long entry_point(const void *start, const void *end, |
8ca47e00 RR |
182 | unsigned long page_offset) |
183 | { | |
3c6b5bfa | 184 | const void *p; |
8ca47e00 | 185 | |
dde79789 RR |
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 | |
189 | * "page_offset". */ | |
8ca47e00 RR |
190 | for (p = start; p < end; p++) |
191 | if (memcmp(p, "GenuineLguest", strlen("GenuineLguest")) == 0) | |
3c6b5bfa RR |
192 | return to_guest_phys(p + strlen("GenuineLguest")) |
193 | + page_offset; | |
8ca47e00 | 194 | |
babed5c0 | 195 | errx(1, "Is this image a genuine lguest?"); |
8ca47e00 RR |
196 | } |
197 | ||
6649bb7a RM |
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) | |
202 | { | |
203 | ssize_t r; | |
204 | ||
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 | |
207 | * instructions. | |
208 | * | |
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 | |
211 | * Guests. */ | |
212 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
213 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
214 | return; | |
215 | ||
216 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
217 | r = pread(fd, addr, len, offset); | |
218 | if (r != len) | |
219 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
220 | } | |
221 | ||
dde79789 RR |
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. | |
225 | * | |
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. | |
231 | * | |
232 | * We return the starting address. */ | |
8ca47e00 RR |
233 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr, |
234 | unsigned long *page_offset) | |
235 | { | |
3c6b5bfa | 236 | void *start = (void *)-1, *end = NULL; |
8ca47e00 RR |
237 | Elf32_Phdr phdr[ehdr->e_phnum]; |
238 | unsigned int i; | |
8ca47e00 | 239 | |
dde79789 RR |
240 | /* Sanity checks on the main ELF header: an x86 executable with a |
241 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
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"); | |
247 | ||
dde79789 RR |
248 | /* An ELF executable contains an ELF header and a number of "program" |
249 | * headers which indicate which parts ("segments") of the program to | |
250 | * load where. */ | |
251 | ||
252 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
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"); | |
257 | ||
dde79789 | 258 | /* We don't know page_offset yet. */ |
8ca47e00 | 259 | *page_offset = 0; |
dde79789 RR |
260 | |
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. */ | |
8ca47e00 | 263 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 264 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
265 | if (phdr[i].p_type != PT_LOAD) |
266 | continue; | |
267 | ||
268 | verbose("Section %i: size %i addr %p\n", | |
269 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
270 | ||
dde79789 RR |
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. */ | |
8ca47e00 RR |
274 | if (!*page_offset) |
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); | |
278 | ||
dde79789 RR |
279 | /* We track the first and last address we mapped, so we can |
280 | * tell entry_point() where to scan. */ | |
3c6b5bfa RR |
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; | |
8ca47e00 | 285 | |
6649bb7a | 286 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 287 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 288 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
289 | } |
290 | ||
3c6b5bfa | 291 | return entry_point(start, end, *page_offset); |
8ca47e00 RR |
292 | } |
293 | ||
dde79789 RR |
294 | /*L:170 Prepare to be SHOCKED and AMAZED. And possibly a trifle nauseated. |
295 | * | |
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: | |
300 | * | |
301 | * movl <some-address>, %eax | |
302 | * | |
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. | |
305 | * | |
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. | |
310 | * | |
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. | |
317 | * | |
318 | * This is amazingly reliable. */ | |
8ca47e00 RR |
319 | static unsigned long intuit_page_offset(unsigned char *img, unsigned long len) |
320 | { | |
321 | unsigned int i, possibilities[256] = { 0 }; | |
322 | ||
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; | |
327 | } | |
328 | errx(1, "could not determine page offset"); | |
329 | } | |
330 | ||
dde79789 RR |
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. */ | |
8ca47e00 RR |
334 | static unsigned long unpack_bzimage(int fd, unsigned long *page_offset) |
335 | { | |
336 | gzFile f; | |
337 | int ret, len = 0; | |
dde79789 RR |
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". | |
341 | * Indeed. */ | |
3c6b5bfa | 342 | void *img = from_guest_phys(0x100000); |
8ca47e00 | 343 | |
dde79789 RR |
344 | /* gzdopen takes our file descriptor (carefully placed at the start of |
345 | * the GZIP header we found) and returns a gzFile. */ | |
8ca47e00 | 346 | f = gzdopen(fd, "rb"); |
dde79789 | 347 | /* We read it into memory in 64k chunks until we hit the end. */ |
8ca47e00 RR |
348 | while ((ret = gzread(f, img + len, 65536)) > 0) |
349 | len += ret; | |
350 | if (ret < 0) | |
351 | err(1, "reading image from bzImage"); | |
352 | ||
353 | verbose("Unpacked size %i addr %p\n", len, img); | |
dde79789 RR |
354 | |
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. */ | |
8ca47e00 RR |
358 | *page_offset = intuit_page_offset(img, len); |
359 | ||
360 | return entry_point(img, img + len, *page_offset); | |
361 | } | |
362 | ||
dde79789 RR |
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. | |
367 | * | |
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. */ | |
8ca47e00 RR |
371 | static unsigned long load_bzimage(int fd, unsigned long *page_offset) |
372 | { | |
373 | unsigned char c; | |
374 | int state = 0; | |
375 | ||
dde79789 | 376 | /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */ |
8ca47e00 RR |
377 | while (read(fd, &c, 1) == 1) { |
378 | switch (state) { | |
379 | case 0: | |
380 | if (c == 0x1F) | |
381 | state++; | |
382 | break; | |
383 | case 1: | |
384 | if (c == 0x8B) | |
385 | state++; | |
386 | else | |
387 | state = 0; | |
388 | break; | |
389 | case 2 ... 8: | |
390 | state++; | |
391 | break; | |
392 | case 9: | |
dde79789 | 393 | /* Seek back to the start of the gzip header. */ |
8ca47e00 | 394 | lseek(fd, -10, SEEK_CUR); |
dde79789 RR |
395 | /* One final check: "compressed under UNIX". */ |
396 | if (c != 0x03) | |
8ca47e00 RR |
397 | state = -1; |
398 | else | |
399 | return unpack_bzimage(fd, page_offset); | |
400 | } | |
401 | } | |
402 | errx(1, "Could not find kernel in bzImage"); | |
403 | } | |
404 | ||
dde79789 RR |
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. */ | |
8ca47e00 RR |
408 | static unsigned long load_kernel(int fd, unsigned long *page_offset) |
409 | { | |
410 | Elf32_Ehdr hdr; | |
411 | ||
dde79789 | 412 | /* Read in the first few bytes. */ |
8ca47e00 RR |
413 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
414 | err(1, "Reading kernel"); | |
415 | ||
dde79789 | 416 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 RR |
417 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
418 | return map_elf(fd, &hdr, page_offset); | |
419 | ||
dde79789 | 420 | /* Otherwise we assume it's a bzImage, and try to unpack it */ |
8ca47e00 RR |
421 | return load_bzimage(fd, page_offset); |
422 | } | |
423 | ||
dde79789 RR |
424 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
425 | * it calls getpagesize() twice: "it's dumb code." | |
426 | * | |
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. */ | |
8ca47e00 RR |
429 | static inline unsigned long page_align(unsigned long addr) |
430 | { | |
dde79789 | 431 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
432 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
433 | } | |
434 | ||
dde79789 RR |
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. | |
439 | * | |
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). */ | |
8ca47e00 RR |
442 | static unsigned long load_initrd(const char *name, unsigned long mem) |
443 | { | |
444 | int ifd; | |
445 | struct stat st; | |
446 | unsigned long len; | |
8ca47e00 RR |
447 | |
448 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 449 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
450 | if (fstat(ifd, &st) < 0) |
451 | err(1, "fstat() on initrd '%s'", name); | |
452 | ||
6649bb7a RM |
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. */ | |
8ca47e00 | 455 | len = page_align(st.st_size); |
3c6b5bfa | 456 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
457 | /* Once a file is mapped, you can close the file descriptor. It's a |
458 | * little odd, but quite useful. */ | |
8ca47e00 | 459 | close(ifd); |
6649bb7a | 460 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
461 | |
462 | /* We return the initrd size. */ | |
8ca47e00 RR |
463 | return len; |
464 | } | |
465 | ||
3c6b5bfa RR |
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. | |
dde79789 RR |
469 | * |
470 | * We lay them out of the way, just below the initrd (which is why we need to | |
471 | * know its size). */ | |
8ca47e00 RR |
472 | static unsigned long setup_pagetables(unsigned long mem, |
473 | unsigned long initrd_size, | |
474 | unsigned long page_offset) | |
475 | { | |
476 | u32 *pgdir, *linear; | |
477 | unsigned int mapped_pages, i, linear_pages; | |
478 | unsigned int ptes_per_page = getpagesize()/sizeof(u32); | |
479 | ||
dde79789 RR |
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). */ | |
8ca47e00 RR |
483 | if (mem <= -page_offset) |
484 | mapped_pages = mem/getpagesize(); | |
485 | else | |
486 | mapped_pages = -page_offset/getpagesize(); | |
487 | ||
dde79789 | 488 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
8ca47e00 RR |
489 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
490 | ||
dde79789 | 491 | /* We put the toplevel page directory page at the top of memory. */ |
3c6b5bfa | 492 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
dde79789 RR |
493 | |
494 | /* Now we use the next linear_pages pages as pte pages */ | |
8ca47e00 RR |
495 | linear = (void *)pgdir - linear_pages*getpagesize(); |
496 | ||
dde79789 RR |
497 | /* Linear mapping is easy: put every page's address into the mapping in |
498 | * order. PAGE_PRESENT contains the flags Present, Writable and | |
499 | * Executable. */ | |
8ca47e00 RR |
500 | for (i = 0; i < mapped_pages; i++) |
501 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); | |
502 | ||
dde79789 RR |
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. */ | |
8ca47e00 RR |
506 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
507 | pgdir[(i + page_offset/getpagesize())/ptes_per_page] | |
3c6b5bfa RR |
508 | = ((to_guest_phys(linear) + i*sizeof(u32)) |
509 | | PAGE_PRESENT); | |
8ca47e00 RR |
510 | } |
511 | ||
3c6b5bfa RR |
512 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
513 | mapped_pages, linear_pages, to_guest_phys(linear)); | |
8ca47e00 | 514 | |
dde79789 RR |
515 | /* We return the top level (guest-physical) address: the kernel needs |
516 | * to know where it is. */ | |
3c6b5bfa | 517 | return to_guest_phys(pgdir); |
8ca47e00 RR |
518 | } |
519 | ||
dde79789 RR |
520 | /* Simple routine to roll all the commandline arguments together with spaces |
521 | * between them. */ | |
8ca47e00 RR |
522 | static void concat(char *dst, char *args[]) |
523 | { | |
524 | unsigned int i, len = 0; | |
525 | ||
526 | for (i = 0; args[i]; i++) { | |
527 | strcpy(dst+len, args[i]); | |
528 | strcat(dst+len, " "); | |
529 | len += strlen(args[i]) + 1; | |
530 | } | |
531 | /* In case it's empty. */ | |
532 | dst[len] = '\0'; | |
533 | } | |
534 | ||
dde79789 RR |
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: | |
3c6b5bfa RR |
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 | |
539 | * Guest. */ | |
8ca47e00 RR |
540 | static int tell_kernel(u32 pgdir, u32 start, u32 page_offset) |
541 | { | |
542 | u32 args[] = { LHREQ_INITIALIZE, | |
3c6b5bfa RR |
543 | (unsigned long)guest_base, |
544 | guest_limit / getpagesize(), | |
545 | pgdir, start, page_offset }; | |
8ca47e00 RR |
546 | int fd; |
547 | ||
3c6b5bfa RR |
548 | verbose("Guest: %p - %p (%#lx)\n", |
549 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
550 | fd = open_or_die("/dev/lguest", O_RDWR); |
551 | if (write(fd, args, sizeof(args)) < 0) | |
552 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
553 | |
554 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
555 | return fd; |
556 | } | |
dde79789 | 557 | /*:*/ |
8ca47e00 RR |
558 | |
559 | static void set_fd(int fd, struct device_list *devices) | |
560 | { | |
561 | FD_SET(fd, &devices->infds); | |
562 | if (fd > devices->max_infd) | |
563 | devices->max_infd = fd; | |
564 | } | |
565 | ||
dde79789 RR |
566 | /*L:200 |
567 | * The Waker. | |
568 | * | |
569 | * With a console and network devices, we can have lots of input which we need | |
570 | * to process. We could try to tell the kernel what file descriptors to watch, | |
571 | * but handing a file descriptor mask through to the kernel is fairly icky. | |
572 | * | |
573 | * Instead, we fork off a process which watches the file descriptors and writes | |
574 | * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host | |
575 | * loop to stop running the Guest. This causes it to return from the | |
576 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset | |
577 | * the LHREQ_BREAK and wake us up again. | |
578 | * | |
579 | * This, of course, is merely a different *kind* of icky. | |
580 | */ | |
8ca47e00 RR |
581 | static void wake_parent(int pipefd, int lguest_fd, struct device_list *devices) |
582 | { | |
dde79789 RR |
583 | /* Add the pipe from the Launcher to the fdset in the device_list, so |
584 | * we watch it, too. */ | |
8ca47e00 RR |
585 | set_fd(pipefd, devices); |
586 | ||
587 | for (;;) { | |
588 | fd_set rfds = devices->infds; | |
589 | u32 args[] = { LHREQ_BREAK, 1 }; | |
590 | ||
dde79789 | 591 | /* Wait until input is ready from one of the devices. */ |
8ca47e00 | 592 | select(devices->max_infd+1, &rfds, NULL, NULL, NULL); |
dde79789 | 593 | /* Is it a message from the Launcher? */ |
8ca47e00 RR |
594 | if (FD_ISSET(pipefd, &rfds)) { |
595 | int ignorefd; | |
dde79789 RR |
596 | /* If read() returns 0, it means the Launcher has |
597 | * exited. We silently follow. */ | |
8ca47e00 RR |
598 | if (read(pipefd, &ignorefd, sizeof(ignorefd)) == 0) |
599 | exit(0); | |
dde79789 RR |
600 | /* Otherwise it's telling us there's a problem with one |
601 | * of the devices, and we should ignore that file | |
602 | * descriptor from now on. */ | |
8ca47e00 | 603 | FD_CLR(ignorefd, &devices->infds); |
dde79789 | 604 | } else /* Send LHREQ_BREAK command. */ |
8ca47e00 RR |
605 | write(lguest_fd, args, sizeof(args)); |
606 | } | |
607 | } | |
608 | ||
dde79789 | 609 | /* This routine just sets up a pipe to the Waker process. */ |
8ca47e00 RR |
610 | static int setup_waker(int lguest_fd, struct device_list *device_list) |
611 | { | |
612 | int pipefd[2], child; | |
613 | ||
dde79789 RR |
614 | /* We create a pipe to talk to the waker, and also so it knows when the |
615 | * Launcher dies (and closes pipe). */ | |
8ca47e00 RR |
616 | pipe(pipefd); |
617 | child = fork(); | |
618 | if (child == -1) | |
619 | err(1, "forking"); | |
620 | ||
621 | if (child == 0) { | |
dde79789 | 622 | /* Close the "writing" end of our copy of the pipe */ |
8ca47e00 RR |
623 | close(pipefd[1]); |
624 | wake_parent(pipefd[0], lguest_fd, device_list); | |
625 | } | |
dde79789 | 626 | /* Close the reading end of our copy of the pipe. */ |
8ca47e00 RR |
627 | close(pipefd[0]); |
628 | ||
dde79789 | 629 | /* Here is the fd used to talk to the waker. */ |
8ca47e00 RR |
630 | return pipefd[1]; |
631 | } | |
632 | ||
dde79789 RR |
633 | /*L:210 |
634 | * Device Handling. | |
635 | * | |
636 | * When the Guest sends DMA to us, it sends us an array of addresses and sizes. | |
637 | * We need to make sure it's not trying to reach into the Launcher itself, so | |
638 | * we have a convenient routine which check it and exits with an error message | |
639 | * if something funny is going on: | |
640 | */ | |
8ca47e00 RR |
641 | static void *_check_pointer(unsigned long addr, unsigned int size, |
642 | unsigned int line) | |
643 | { | |
dde79789 RR |
644 | /* We have to separately check addr and addr+size, because size could |
645 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 646 | if (addr >= guest_limit || addr + size >= guest_limit) |
8ca47e00 | 647 | errx(1, "%s:%i: Invalid address %li", __FILE__, line, addr); |
dde79789 RR |
648 | /* We return a pointer for the caller's convenience, now we know it's |
649 | * safe to use. */ | |
3c6b5bfa | 650 | return from_guest_phys(addr); |
8ca47e00 | 651 | } |
dde79789 | 652 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
653 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
654 | ||
dde79789 RR |
655 | /* The Guest has given us the address of a "struct lguest_dma". We check it's |
656 | * OK and convert it to an iovec (which is a simple array of ptr/size | |
657 | * pairs). */ | |
8ca47e00 RR |
658 | static u32 *dma2iov(unsigned long dma, struct iovec iov[], unsigned *num) |
659 | { | |
660 | unsigned int i; | |
661 | struct lguest_dma *udma; | |
662 | ||
dde79789 | 663 | /* First we make sure that the array memory itself is valid. */ |
8ca47e00 | 664 | udma = check_pointer(dma, sizeof(*udma)); |
dde79789 | 665 | /* Now we check each element */ |
8ca47e00 | 666 | for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) { |
dde79789 | 667 | /* A zero length ends the array. */ |
8ca47e00 RR |
668 | if (!udma->len[i]) |
669 | break; | |
670 | ||
671 | iov[i].iov_base = check_pointer(udma->addr[i], udma->len[i]); | |
672 | iov[i].iov_len = udma->len[i]; | |
673 | } | |
674 | *num = i; | |
dde79789 RR |
675 | |
676 | /* We return the pointer to where the caller should write the amount of | |
677 | * the buffer used. */ | |
8ca47e00 RR |
678 | return &udma->used_len; |
679 | } | |
680 | ||
dde79789 RR |
681 | /* This routine gets a DMA buffer from the Guest for a given key, and converts |
682 | * it to an iovec array. It returns the interrupt the Guest wants when we're | |
683 | * finished, and a pointer to the "used_len" field to fill in. */ | |
8ca47e00 RR |
684 | static u32 *get_dma_buffer(int fd, void *key, |
685 | struct iovec iov[], unsigned int *num, u32 *irq) | |
686 | { | |
3c6b5bfa | 687 | u32 buf[] = { LHREQ_GETDMA, to_guest_phys(key) }; |
8ca47e00 RR |
688 | unsigned long udma; |
689 | u32 *res; | |
690 | ||
dde79789 | 691 | /* Ask the kernel for a DMA buffer corresponding to this key. */ |
8ca47e00 | 692 | udma = write(fd, buf, sizeof(buf)); |
dde79789 | 693 | /* They haven't registered any, or they're all used? */ |
8ca47e00 RR |
694 | if (udma == (unsigned long)-1) |
695 | return NULL; | |
696 | ||
dde79789 | 697 | /* Convert it into our iovec array */ |
8ca47e00 | 698 | res = dma2iov(udma, iov, num); |
dde79789 | 699 | /* The kernel stashes irq in ->used_len to get it out to us. */ |
8ca47e00 | 700 | *irq = *res; |
dde79789 | 701 | /* Return a pointer to ((struct lguest_dma *)udma)->used_len. */ |
8ca47e00 RR |
702 | return res; |
703 | } | |
704 | ||
dde79789 | 705 | /* This is a convenient routine to send the Guest an interrupt. */ |
8ca47e00 RR |
706 | static void trigger_irq(int fd, u32 irq) |
707 | { | |
708 | u32 buf[] = { LHREQ_IRQ, irq }; | |
709 | if (write(fd, buf, sizeof(buf)) != 0) | |
710 | err(1, "Triggering irq %i", irq); | |
711 | } | |
712 | ||
dde79789 RR |
713 | /* This simply sets up an iovec array where we can put data to be discarded. |
714 | * This happens when the Guest doesn't want or can't handle the input: we have | |
715 | * to get rid of it somewhere, and if we bury it in the ceiling space it will | |
716 | * start to smell after a week. */ | |
8ca47e00 RR |
717 | static void discard_iovec(struct iovec *iov, unsigned int *num) |
718 | { | |
719 | static char discard_buf[1024]; | |
720 | *num = 1; | |
721 | iov->iov_base = discard_buf; | |
722 | iov->iov_len = sizeof(discard_buf); | |
723 | } | |
724 | ||
dde79789 RR |
725 | /* Here is the input terminal setting we save, and the routine to restore them |
726 | * on exit so the user can see what they type next. */ | |
8ca47e00 RR |
727 | static struct termios orig_term; |
728 | static void restore_term(void) | |
729 | { | |
730 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
731 | } | |
732 | ||
dde79789 | 733 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
734 | struct console_abort |
735 | { | |
dde79789 | 736 | /* How many times have they hit ^C? */ |
8ca47e00 | 737 | int count; |
dde79789 | 738 | /* When did they start? */ |
8ca47e00 RR |
739 | struct timeval start; |
740 | }; | |
741 | ||
dde79789 | 742 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
743 | static bool handle_console_input(int fd, struct device *dev) |
744 | { | |
745 | u32 irq = 0, *lenp; | |
746 | int len; | |
747 | unsigned int num; | |
748 | struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; | |
749 | struct console_abort *abort = dev->priv; | |
750 | ||
dde79789 RR |
751 | /* First we get the console buffer from the Guest. The key is dev->mem |
752 | * which was set to 0 in setup_console(). */ | |
8ca47e00 RR |
753 | lenp = get_dma_buffer(fd, dev->mem, iov, &num, &irq); |
754 | if (!lenp) { | |
dde79789 | 755 | /* If it's not ready for input, warn and set up to discard. */ |
8ca47e00 RR |
756 | warn("console: no dma buffer!"); |
757 | discard_iovec(iov, &num); | |
758 | } | |
759 | ||
dde79789 RR |
760 | /* This is why we convert to iovecs: the readv() call uses them, and so |
761 | * it reads straight into the Guest's buffer. */ | |
8ca47e00 RR |
762 | len = readv(dev->fd, iov, num); |
763 | if (len <= 0) { | |
dde79789 RR |
764 | /* This implies that the console is closed, is /dev/null, or |
765 | * something went terribly wrong. We still go through the rest | |
766 | * of the logic, though, especially the exit handling below. */ | |
8ca47e00 RR |
767 | warnx("Failed to get console input, ignoring console."); |
768 | len = 0; | |
769 | } | |
770 | ||
dde79789 RR |
771 | /* If we read the data into the Guest, fill in the length and send the |
772 | * interrupt. */ | |
8ca47e00 RR |
773 | if (lenp) { |
774 | *lenp = len; | |
775 | trigger_irq(fd, irq); | |
776 | } | |
777 | ||
dde79789 RR |
778 | /* Three ^C within one second? Exit. |
779 | * | |
780 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
781 | * in a buffer by itself, so they can't be too fast. But we check that | |
782 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
783 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
784 | if (!abort->count++) | |
785 | gettimeofday(&abort->start, NULL); | |
786 | else if (abort->count == 3) { | |
787 | struct timeval now; | |
788 | gettimeofday(&now, NULL); | |
789 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
8ca47e00 | 790 | u32 args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
791 | /* Close the fd so Waker will know it has to |
792 | * exit. */ | |
8ca47e00 | 793 | close(waker_fd); |
dde79789 RR |
794 | /* Just in case waker is blocked in BREAK, send |
795 | * unbreak now. */ | |
8ca47e00 RR |
796 | write(fd, args, sizeof(args)); |
797 | exit(2); | |
798 | } | |
799 | abort->count = 0; | |
800 | } | |
801 | } else | |
dde79789 | 802 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
803 | abort->count = 0; |
804 | ||
dde79789 RR |
805 | /* Now, if we didn't read anything, put the input terminal back and |
806 | * return failure (meaning, don't call us again). */ | |
8ca47e00 RR |
807 | if (!len) { |
808 | restore_term(); | |
809 | return false; | |
810 | } | |
dde79789 | 811 | /* Everything went OK! */ |
8ca47e00 RR |
812 | return true; |
813 | } | |
814 | ||
dde79789 | 815 | /* Handling console output is much simpler than input. */ |
8ca47e00 RR |
816 | static u32 handle_console_output(int fd, const struct iovec *iov, |
817 | unsigned num, struct device*dev) | |
818 | { | |
dde79789 RR |
819 | /* Whatever the Guest sends, write it to standard output. Return the |
820 | * number of bytes written. */ | |
8ca47e00 RR |
821 | return writev(STDOUT_FILENO, iov, num); |
822 | } | |
823 | ||
dde79789 | 824 | /* Guest->Host network output is also pretty easy. */ |
8ca47e00 RR |
825 | static u32 handle_tun_output(int fd, const struct iovec *iov, |
826 | unsigned num, struct device *dev) | |
827 | { | |
dde79789 RR |
828 | /* We put a flag in the "priv" pointer of the network device, and set |
829 | * it as soon as we see output. We'll see why in handle_tun_input() */ | |
8ca47e00 | 830 | *(bool *)dev->priv = true; |
dde79789 RR |
831 | /* Whatever packet the Guest sent us, write it out to the tun |
832 | * device. */ | |
8ca47e00 RR |
833 | return writev(dev->fd, iov, num); |
834 | } | |
835 | ||
dde79789 RR |
836 | /* This matches the peer_key() in lguest_net.c. The key for any given slot |
837 | * is the address of the network device's page plus 4 * the slot number. */ | |
8ca47e00 RR |
838 | static unsigned long peer_offset(unsigned int peernum) |
839 | { | |
840 | return 4 * peernum; | |
841 | } | |
842 | ||
dde79789 | 843 | /* This is where we handle a packet coming in from the tun device */ |
8ca47e00 RR |
844 | static bool handle_tun_input(int fd, struct device *dev) |
845 | { | |
846 | u32 irq = 0, *lenp; | |
847 | int len; | |
848 | unsigned num; | |
849 | struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; | |
850 | ||
dde79789 | 851 | /* First we get a buffer the Guest has bound to its key. */ |
8ca47e00 RR |
852 | lenp = get_dma_buffer(fd, dev->mem+peer_offset(NET_PEERNUM), iov, &num, |
853 | &irq); | |
854 | if (!lenp) { | |
dde79789 RR |
855 | /* Now, it's expected that if we try to send a packet too |
856 | * early, the Guest won't be ready yet. This is why we set a | |
857 | * flag when the Guest sends its first packet. If it's sent a | |
858 | * packet we assume it should be ready to receive them. | |
859 | * | |
860 | * Actually, this is what the status bits in the descriptor are | |
861 | * for: we should *use* them. FIXME! */ | |
8ca47e00 RR |
862 | if (*(bool *)dev->priv) |
863 | warn("network: no dma buffer!"); | |
864 | discard_iovec(iov, &num); | |
865 | } | |
866 | ||
dde79789 | 867 | /* Read the packet from the device directly into the Guest's buffer. */ |
8ca47e00 RR |
868 | len = readv(dev->fd, iov, num); |
869 | if (len <= 0) | |
870 | err(1, "reading network"); | |
dde79789 RR |
871 | |
872 | /* Write the used_len, and trigger the interrupt for the Guest */ | |
8ca47e00 RR |
873 | if (lenp) { |
874 | *lenp = len; | |
875 | trigger_irq(fd, irq); | |
876 | } | |
877 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, | |
878 | ((u8 *)iov[0].iov_base)[0], ((u8 *)iov[0].iov_base)[1], | |
879 | lenp ? "sent" : "discarded"); | |
dde79789 | 880 | /* All good. */ |
8ca47e00 RR |
881 | return true; |
882 | } | |
883 | ||
dde79789 RR |
884 | /* The last device handling routine is block output: the Guest has sent a DMA |
885 | * to the block device. It will have placed the command it wants in the | |
886 | * "struct lguest_block_page". */ | |
8ca47e00 RR |
887 | static u32 handle_block_output(int fd, const struct iovec *iov, |
888 | unsigned num, struct device *dev) | |
889 | { | |
890 | struct lguest_block_page *p = dev->mem; | |
891 | u32 irq, *lenp; | |
892 | unsigned int len, reply_num; | |
893 | struct iovec reply[LGUEST_MAX_DMA_SECTIONS]; | |
894 | off64_t device_len, off = (off64_t)p->sector * 512; | |
895 | ||
dde79789 | 896 | /* First we extract the device length from the dev->priv pointer. */ |
8ca47e00 RR |
897 | device_len = *(off64_t *)dev->priv; |
898 | ||
dde79789 RR |
899 | /* We first check that the read or write is within the length of the |
900 | * block file. */ | |
8ca47e00 | 901 | if (off >= device_len) |
babed5c0 | 902 | errx(1, "Bad offset %llu vs %llu", off, device_len); |
dde79789 RR |
903 | /* Move to the right location in the block file. This shouldn't fail, |
904 | * but best to check. */ | |
8ca47e00 RR |
905 | if (lseek64(dev->fd, off, SEEK_SET) != off) |
906 | err(1, "Bad seek to sector %i", p->sector); | |
907 | ||
908 | verbose("Block: %s at offset %llu\n", p->type ? "WRITE" : "READ", off); | |
909 | ||
dde79789 RR |
910 | /* They were supposed to bind a reply buffer at key equal to the start |
911 | * of the block device memory. We need this to tell them when the | |
912 | * request is finished. */ | |
8ca47e00 RR |
913 | lenp = get_dma_buffer(fd, dev->mem, reply, &reply_num, &irq); |
914 | if (!lenp) | |
915 | err(1, "Block request didn't give us a dma buffer"); | |
916 | ||
917 | if (p->type) { | |
dde79789 RR |
918 | /* A write request. The DMA they sent contained the data, so |
919 | * write it out. */ | |
8ca47e00 | 920 | len = writev(dev->fd, iov, num); |
dde79789 RR |
921 | /* Grr... Now we know how long the "struct lguest_dma" they |
922 | * sent was, we make sure they didn't try to write over the end | |
923 | * of the block file (possibly extending it). */ | |
8ca47e00 | 924 | if (off + len > device_len) { |
dde79789 | 925 | /* Trim it back to the correct length */ |
f6a592e8 | 926 | ftruncate64(dev->fd, device_len); |
dde79789 | 927 | /* Die, bad Guest, die. */ |
8ca47e00 RR |
928 | errx(1, "Write past end %llu+%u", off, len); |
929 | } | |
dde79789 RR |
930 | /* The reply length is 0: we just send back an empty DMA to |
931 | * interrupt them and tell them the write is finished. */ | |
8ca47e00 RR |
932 | *lenp = 0; |
933 | } else { | |
dde79789 RR |
934 | /* A read request. They sent an empty DMA to start the |
935 | * request, and we put the read contents into the reply | |
936 | * buffer. */ | |
8ca47e00 RR |
937 | len = readv(dev->fd, reply, reply_num); |
938 | *lenp = len; | |
939 | } | |
940 | ||
dde79789 RR |
941 | /* The result is 1 (done), 2 if there was an error (short read or |
942 | * write). */ | |
8ca47e00 | 943 | p->result = 1 + (p->bytes != len); |
dde79789 | 944 | /* Now tell them we've used their reply buffer. */ |
8ca47e00 | 945 | trigger_irq(fd, irq); |
dde79789 RR |
946 | |
947 | /* We're supposed to return the number of bytes of the output buffer we | |
948 | * used. But the block device uses the "result" field instead, so we | |
949 | * don't bother. */ | |
8ca47e00 RR |
950 | return 0; |
951 | } | |
952 | ||
dde79789 | 953 | /* This is the generic routine we call when the Guest sends some DMA out. */ |
8ca47e00 RR |
954 | static void handle_output(int fd, unsigned long dma, unsigned long key, |
955 | struct device_list *devices) | |
956 | { | |
957 | struct device *i; | |
958 | u32 *lenp; | |
959 | struct iovec iov[LGUEST_MAX_DMA_SECTIONS]; | |
960 | unsigned num = 0; | |
961 | ||
dde79789 RR |
962 | /* Convert the "struct lguest_dma" they're sending to a "struct |
963 | * iovec". */ | |
8ca47e00 | 964 | lenp = dma2iov(dma, iov, &num); |
dde79789 RR |
965 | |
966 | /* Check each device: if they expect output to this key, tell them to | |
967 | * handle it. */ | |
8ca47e00 RR |
968 | for (i = devices->dev; i; i = i->next) { |
969 | if (i->handle_output && key == i->watch_key) { | |
dde79789 RR |
970 | /* We write the result straight into the used_len field |
971 | * for them. */ | |
8ca47e00 RR |
972 | *lenp = i->handle_output(fd, iov, num, i); |
973 | return; | |
974 | } | |
975 | } | |
dde79789 RR |
976 | |
977 | /* This can happen: the kernel sends any SEND_DMA which doesn't match | |
978 | * another Guest to us. It could be that another Guest just left a | |
979 | * network, for example. But it's unusual. */ | |
8ca47e00 RR |
980 | warnx("Pending dma %p, key %p", (void *)dma, (void *)key); |
981 | } | |
982 | ||
dde79789 RR |
983 | /* This is called when the waker wakes us up: check for incoming file |
984 | * descriptors. */ | |
8ca47e00 RR |
985 | static void handle_input(int fd, struct device_list *devices) |
986 | { | |
dde79789 | 987 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
988 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
989 | ||
990 | for (;;) { | |
991 | struct device *i; | |
992 | fd_set fds = devices->infds; | |
993 | ||
dde79789 | 994 | /* If nothing is ready, we're done. */ |
8ca47e00 RR |
995 | if (select(devices->max_infd+1, &fds, NULL, NULL, &poll) == 0) |
996 | break; | |
997 | ||
dde79789 RR |
998 | /* Otherwise, call the device(s) which have readable |
999 | * file descriptors and a method of handling them. */ | |
8ca47e00 RR |
1000 | for (i = devices->dev; i; i = i->next) { |
1001 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { | |
dde79789 RR |
1002 | /* If handle_input() returns false, it means we |
1003 | * should no longer service it. | |
1004 | * handle_console_input() does this. */ | |
8ca47e00 | 1005 | if (!i->handle_input(fd, i)) { |
dde79789 RR |
1006 | /* Clear it from the set of input file |
1007 | * descriptors kept at the head of the | |
1008 | * device list. */ | |
8ca47e00 RR |
1009 | FD_CLR(i->fd, &devices->infds); |
1010 | /* Tell waker to ignore it too... */ | |
1011 | write(waker_fd, &i->fd, sizeof(i->fd)); | |
1012 | } | |
1013 | } | |
1014 | } | |
1015 | } | |
1016 | } | |
1017 | ||
dde79789 RR |
1018 | /*L:190 |
1019 | * Device Setup | |
1020 | * | |
1021 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1022 | * device" so the Launcher can keep track of it. We have common helper | |
1023 | * routines to allocate them. | |
1024 | * | |
1025 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
1026 | * table in the devices array just above the Guest's normal memory. */ | |
6570c459 RR |
1027 | static struct lguest_device_desc * |
1028 | new_dev_desc(struct lguest_device_desc *descs, | |
1029 | u16 type, u16 features, u16 num_pages) | |
8ca47e00 | 1030 | { |
6570c459 | 1031 | unsigned int i; |
8ca47e00 | 1032 | |
6570c459 RR |
1033 | for (i = 0; i < LGUEST_MAX_DEVICES; i++) { |
1034 | if (!descs[i].type) { | |
1035 | descs[i].type = type; | |
1036 | descs[i].features = features; | |
1037 | descs[i].num_pages = num_pages; | |
dde79789 | 1038 | /* If they said the device needs memory, we allocate |
3c6b5bfa | 1039 | * that now. */ |
6570c459 | 1040 | if (num_pages) { |
3c6b5bfa RR |
1041 | unsigned long pa; |
1042 | pa = to_guest_phys(get_pages(num_pages)); | |
1043 | descs[i].pfn = pa / getpagesize(); | |
6570c459 RR |
1044 | } |
1045 | return &descs[i]; | |
1046 | } | |
1047 | } | |
1048 | errx(1, "too many devices"); | |
8ca47e00 RR |
1049 | } |
1050 | ||
dde79789 RR |
1051 | /* This monster routine does all the creation and setup of a new device, |
1052 | * including caling new_dev_desc() to allocate the descriptor and device | |
1053 | * memory. */ | |
8ca47e00 RR |
1054 | static struct device *new_device(struct device_list *devices, |
1055 | u16 type, u16 num_pages, u16 features, | |
1056 | int fd, | |
1057 | bool (*handle_input)(int, struct device *), | |
1058 | unsigned long watch_off, | |
1059 | u32 (*handle_output)(int, | |
1060 | const struct iovec *, | |
1061 | unsigned, | |
1062 | struct device *)) | |
1063 | { | |
1064 | struct device *dev = malloc(sizeof(*dev)); | |
1065 | ||
dde79789 RR |
1066 | /* Append to device list. Prepending to a single-linked list is |
1067 | * easier, but the user expects the devices to be arranged on the bus | |
1068 | * in command-line order. The first network device on the command line | |
1069 | * is eth0, the first block device /dev/lgba, etc. */ | |
8ca47e00 RR |
1070 | *devices->lastdev = dev; |
1071 | dev->next = NULL; | |
1072 | devices->lastdev = &dev->next; | |
1073 | ||
dde79789 | 1074 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1075 | dev->fd = fd; |
dde79789 RR |
1076 | /* If we have an input handler for this file descriptor, then we add it |
1077 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 RR |
1078 | if (handle_input) |
1079 | set_fd(dev->fd, devices); | |
6570c459 | 1080 | dev->desc = new_dev_desc(devices->descs, type, features, num_pages); |
3c6b5bfa | 1081 | dev->mem = from_guest_phys(dev->desc->pfn * getpagesize()); |
8ca47e00 | 1082 | dev->handle_input = handle_input; |
3c6b5bfa | 1083 | dev->watch_key = to_guest_phys(dev->mem) + watch_off; |
8ca47e00 RR |
1084 | dev->handle_output = handle_output; |
1085 | return dev; | |
1086 | } | |
1087 | ||
dde79789 RR |
1088 | /* Our first setup routine is the console. It's a fairly simple device, but |
1089 | * UNIX tty handling makes it uglier than it could be. */ | |
8ca47e00 RR |
1090 | static void setup_console(struct device_list *devices) |
1091 | { | |
1092 | struct device *dev; | |
1093 | ||
dde79789 | 1094 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1095 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1096 | struct termios term = orig_term; | |
dde79789 RR |
1097 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1098 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1099 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1100 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1101 | /* If we exit gracefully, the original settings will be |
1102 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1103 | atexit(restore_term); |
1104 | } | |
1105 | ||
dde79789 RR |
1106 | /* We don't currently require any memory for the console, so we ask for |
1107 | * 0 pages. */ | |
8ca47e00 RR |
1108 | dev = new_device(devices, LGUEST_DEVICE_T_CONSOLE, 0, 0, |
1109 | STDIN_FILENO, handle_console_input, | |
1110 | LGUEST_CONSOLE_DMA_KEY, handle_console_output); | |
dde79789 | 1111 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1112 | dev->priv = malloc(sizeof(struct console_abort)); |
1113 | ((struct console_abort *)dev->priv)->count = 0; | |
1114 | verbose("device %p: console\n", | |
1115 | (void *)(dev->desc->pfn * getpagesize())); | |
1116 | } | |
1117 | ||
dde79789 | 1118 | /* Setting up a block file is also fairly straightforward. */ |
8ca47e00 RR |
1119 | static void setup_block_file(const char *filename, struct device_list *devices) |
1120 | { | |
1121 | int fd; | |
1122 | struct device *dev; | |
1123 | off64_t *device_len; | |
1124 | struct lguest_block_page *p; | |
1125 | ||
dde79789 RR |
1126 | /* We open with O_LARGEFILE because otherwise we get stuck at 2G. We |
1127 | * open with O_DIRECT because otherwise our benchmarks go much too | |
1128 | * fast. */ | |
8ca47e00 | 1129 | fd = open_or_die(filename, O_RDWR|O_LARGEFILE|O_DIRECT); |
dde79789 RR |
1130 | |
1131 | /* We want one page, and have no input handler (the block file never | |
1132 | * has anything interesting to say to us). Our timing will be quite | |
1133 | * random, so it should be a reasonable randomness source. */ | |
8ca47e00 RR |
1134 | dev = new_device(devices, LGUEST_DEVICE_T_BLOCK, 1, |
1135 | LGUEST_DEVICE_F_RANDOMNESS, | |
1136 | fd, NULL, 0, handle_block_output); | |
dde79789 RR |
1137 | |
1138 | /* We store the device size in the private area */ | |
8ca47e00 | 1139 | device_len = dev->priv = malloc(sizeof(*device_len)); |
dde79789 RR |
1140 | /* This is the safe way of establishing the size of our device: it |
1141 | * might be a normal file or an actual block device like /dev/hdb. */ | |
8ca47e00 | 1142 | *device_len = lseek64(fd, 0, SEEK_END); |
8ca47e00 | 1143 | |
dde79789 RR |
1144 | /* The device memory is a "struct lguest_block_page". It's zeroed |
1145 | * already, we just need to put in the device size. Block devices | |
1146 | * think in sectors (ie. 512 byte chunks), so we translate here. */ | |
1147 | p = dev->mem; | |
8ca47e00 RR |
1148 | p->num_sectors = *device_len/512; |
1149 | verbose("device %p: block %i sectors\n", | |
1150 | (void *)(dev->desc->pfn * getpagesize()), p->num_sectors); | |
1151 | } | |
1152 | ||
dde79789 RR |
1153 | /* |
1154 | * Network Devices. | |
1155 | * | |
1156 | * Setting up network devices is quite a pain, because we have three types. | |
1157 | * First, we have the inter-Guest network. This is a file which is mapped into | |
1158 | * the address space of the Guests who are on the network. Because it is a | |
1159 | * shared mapping, the same page underlies all the devices, and they can send | |
1160 | * DMA to each other. | |
1161 | * | |
1162 | * Remember from our network driver, the Guest is told what slot in the page it | |
1163 | * is to use. We use exclusive fnctl locks to reserve a slot. If another | |
1164 | * Guest is using a slot, the lock will fail and we try another. Because fnctl | |
1165 | * locks are cleaned up automatically when we die, this cleverly means that our | |
1166 | * reservation on the slot will vanish if we crash. */ | |
8ca47e00 RR |
1167 | static unsigned int find_slot(int netfd, const char *filename) |
1168 | { | |
1169 | struct flock fl; | |
1170 | ||
1171 | fl.l_type = F_WRLCK; | |
1172 | fl.l_whence = SEEK_SET; | |
1173 | fl.l_len = 1; | |
dde79789 | 1174 | /* Try a 1 byte lock in each possible position number */ |
8ca47e00 RR |
1175 | for (fl.l_start = 0; |
1176 | fl.l_start < getpagesize()/sizeof(struct lguest_net); | |
1177 | fl.l_start++) { | |
dde79789 | 1178 | /* If we succeed, return the slot number. */ |
8ca47e00 RR |
1179 | if (fcntl(netfd, F_SETLK, &fl) == 0) |
1180 | return fl.l_start; | |
1181 | } | |
1182 | errx(1, "No free slots in network file %s", filename); | |
1183 | } | |
1184 | ||
dde79789 | 1185 | /* This function sets up the network file */ |
8ca47e00 RR |
1186 | static void setup_net_file(const char *filename, |
1187 | struct device_list *devices) | |
1188 | { | |
1189 | int netfd; | |
1190 | struct device *dev; | |
1191 | ||
dde79789 RR |
1192 | /* We don't use open_or_die() here: for friendliness we create the file |
1193 | * if it doesn't already exist. */ | |
8ca47e00 RR |
1194 | netfd = open(filename, O_RDWR, 0); |
1195 | if (netfd < 0) { | |
1196 | if (errno == ENOENT) { | |
1197 | netfd = open(filename, O_RDWR|O_CREAT, 0600); | |
1198 | if (netfd >= 0) { | |
dde79789 RR |
1199 | /* If we succeeded, initialize the file with a |
1200 | * blank page. */ | |
8ca47e00 RR |
1201 | char page[getpagesize()]; |
1202 | memset(page, 0, sizeof(page)); | |
1203 | write(netfd, page, sizeof(page)); | |
1204 | } | |
1205 | } | |
1206 | if (netfd < 0) | |
1207 | err(1, "cannot open net file '%s'", filename); | |
1208 | } | |
1209 | ||
dde79789 RR |
1210 | /* We need 1 page, and the features indicate the slot to use and that |
1211 | * no checksum is needed. We never touch this device again; it's | |
1212 | * between the Guests on the network, so we don't register input or | |
1213 | * output handlers. */ | |
8ca47e00 RR |
1214 | dev = new_device(devices, LGUEST_DEVICE_T_NET, 1, |
1215 | find_slot(netfd, filename)|LGUEST_NET_F_NOCSUM, | |
1216 | -1, NULL, 0, NULL); | |
1217 | ||
dde79789 | 1218 | /* Map the shared file. */ |
8ca47e00 RR |
1219 | if (mmap(dev->mem, getpagesize(), PROT_READ|PROT_WRITE, |
1220 | MAP_FIXED|MAP_SHARED, netfd, 0) != dev->mem) | |
1221 | err(1, "could not mmap '%s'", filename); | |
1222 | verbose("device %p: shared net %s, peer %i\n", | |
1223 | (void *)(dev->desc->pfn * getpagesize()), filename, | |
1224 | dev->desc->features & ~LGUEST_NET_F_NOCSUM); | |
1225 | } | |
dde79789 | 1226 | /*:*/ |
8ca47e00 RR |
1227 | |
1228 | static u32 str2ip(const char *ipaddr) | |
1229 | { | |
1230 | unsigned int byte[4]; | |
1231 | ||
1232 | sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]); | |
1233 | return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3]; | |
1234 | } | |
1235 | ||
dde79789 RR |
1236 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1237 | * network device to the bridge device specified by the command line. | |
1238 | * | |
1239 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1240 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1241 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1242 | { | |
1243 | int ifidx; | |
1244 | struct ifreq ifr; | |
1245 | ||
1246 | if (!*br_name) | |
1247 | errx(1, "must specify bridge name"); | |
1248 | ||
1249 | ifidx = if_nametoindex(if_name); | |
1250 | if (!ifidx) | |
1251 | errx(1, "interface %s does not exist!", if_name); | |
1252 | ||
1253 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
1254 | ifr.ifr_ifindex = ifidx; | |
1255 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1256 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1257 | } | |
1258 | ||
dde79789 RR |
1259 | /* This sets up the Host end of the network device with an IP address, brings |
1260 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
1261 | * pointer (in practice, the Host's slot in the network device's memory). */ | |
8ca47e00 RR |
1262 | static void configure_device(int fd, const char *devname, u32 ipaddr, |
1263 | unsigned char hwaddr[6]) | |
1264 | { | |
1265 | struct ifreq ifr; | |
1266 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1267 | ||
dde79789 | 1268 | /* Don't read these incantations. Just cut & paste them like I did! */ |
8ca47e00 RR |
1269 | memset(&ifr, 0, sizeof(ifr)); |
1270 | strcpy(ifr.ifr_name, devname); | |
1271 | sin->sin_family = AF_INET; | |
1272 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1273 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
1274 | err(1, "Setting %s interface address", devname); | |
1275 | ifr.ifr_flags = IFF_UP; | |
1276 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
1277 | err(1, "Bringing interface %s up", devname); | |
1278 | ||
dde79789 RR |
1279 | /* SIOC stands for Socket I/O Control. G means Get (vs S for Set |
1280 | * above). IF means Interface, and HWADDR is hardware address. | |
1281 | * Simple! */ | |
8ca47e00 RR |
1282 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0) |
1283 | err(1, "getting hw address for %s", devname); | |
8ca47e00 RR |
1284 | memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); |
1285 | } | |
1286 | ||
dde79789 RR |
1287 | /*L:195 The other kind of network is a Host<->Guest network. This can either |
1288 | * use briding or routing, but the principle is the same: it uses the "tun" | |
1289 | * device to inject packets into the Host as if they came in from a normal | |
1290 | * network card. We just shunt packets between the Guest and the tun | |
1291 | * device. */ | |
8ca47e00 RR |
1292 | static void setup_tun_net(const char *arg, struct device_list *devices) |
1293 | { | |
1294 | struct device *dev; | |
1295 | struct ifreq ifr; | |
1296 | int netfd, ipfd; | |
1297 | u32 ip; | |
1298 | const char *br_name = NULL; | |
1299 | ||
dde79789 RR |
1300 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1301 | * tap device is like a tun device, only somehow different. To tell | |
1302 | * the truth, I completely blundered my way through this code, but it | |
1303 | * works now! */ | |
8ca47e00 RR |
1304 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
1305 | memset(&ifr, 0, sizeof(ifr)); | |
1306 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
1307 | strcpy(ifr.ifr_name, "tap%d"); | |
1308 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1309 | err(1, "configuring /dev/net/tun"); | |
dde79789 RR |
1310 | /* We don't need checksums calculated for packets coming in this |
1311 | * device: trust us! */ | |
8ca47e00 RR |
1312 | ioctl(netfd, TUNSETNOCSUM, 1); |
1313 | ||
dde79789 RR |
1314 | /* We create the net device with 1 page, using the features field of |
1315 | * the descriptor to tell the Guest it is in slot 1 (NET_PEERNUM), and | |
1316 | * that the device has fairly random timing. We do *not* specify | |
1317 | * LGUEST_NET_F_NOCSUM: these packets can reach the real world. | |
1318 | * | |
1319 | * We will put our MAC address is slot 0 for the Guest to see, so | |
1320 | * it will send packets to us using the key "peer_offset(0)": */ | |
8ca47e00 RR |
1321 | dev = new_device(devices, LGUEST_DEVICE_T_NET, 1, |
1322 | NET_PEERNUM|LGUEST_DEVICE_F_RANDOMNESS, netfd, | |
1323 | handle_tun_input, peer_offset(0), handle_tun_output); | |
dde79789 RR |
1324 | |
1325 | /* We keep a flag which says whether we've seen packets come out from | |
1326 | * this network device. */ | |
8ca47e00 RR |
1327 | dev->priv = malloc(sizeof(bool)); |
1328 | *(bool *)dev->priv = false; | |
1329 | ||
dde79789 RR |
1330 | /* We need a socket to perform the magic network ioctls to bring up the |
1331 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1332 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1333 | if (ipfd < 0) | |
1334 | err(1, "opening IP socket"); | |
1335 | ||
dde79789 | 1336 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 RR |
1337 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
1338 | ip = INADDR_ANY; | |
1339 | br_name = arg + strlen(BRIDGE_PFX); | |
1340 | add_to_bridge(ipfd, ifr.ifr_name, br_name); | |
dde79789 | 1341 | } else /* It is an IP address to set up the device with */ |
8ca47e00 RR |
1342 | ip = str2ip(arg); |
1343 | ||
dde79789 RR |
1344 | /* We are peer 0, ie. first slot, so we hand dev->mem to this routine |
1345 | * to write the MAC address at the start of the device memory. */ | |
8ca47e00 RR |
1346 | configure_device(ipfd, ifr.ifr_name, ip, dev->mem); |
1347 | ||
dde79789 RR |
1348 | /* Set "promisc" bit: we want every single packet if we're going to |
1349 | * bridge to other machines (and otherwise it doesn't matter). */ | |
8ca47e00 RR |
1350 | *((u8 *)dev->mem) |= 0x1; |
1351 | ||
1352 | close(ipfd); | |
1353 | ||
1354 | verbose("device %p: tun net %u.%u.%u.%u\n", | |
1355 | (void *)(dev->desc->pfn * getpagesize()), | |
1356 | (u8)(ip>>24), (u8)(ip>>16), (u8)(ip>>8), (u8)ip); | |
1357 | if (br_name) | |
1358 | verbose("attached to bridge: %s\n", br_name); | |
1359 | } | |
dde79789 | 1360 | /* That's the end of device setup. */ |
8ca47e00 | 1361 | |
dde79789 RR |
1362 | /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves |
1363 | * its input and output, and finally, lays it to rest. */ | |
8ca47e00 RR |
1364 | static void __attribute__((noreturn)) |
1365 | run_guest(int lguest_fd, struct device_list *device_list) | |
1366 | { | |
1367 | for (;;) { | |
1368 | u32 args[] = { LHREQ_BREAK, 0 }; | |
1369 | unsigned long arr[2]; | |
1370 | int readval; | |
1371 | ||
1372 | /* We read from the /dev/lguest device to run the Guest. */ | |
1373 | readval = read(lguest_fd, arr, sizeof(arr)); | |
1374 | ||
dde79789 RR |
1375 | /* The read can only really return sizeof(arr) (the Guest did a |
1376 | * SEND_DMA to us), or an error. */ | |
1377 | ||
1378 | /* For a successful read, arr[0] is the address of the "struct | |
1379 | * lguest_dma", and arr[1] is the key the Guest sent to. */ | |
8ca47e00 RR |
1380 | if (readval == sizeof(arr)) { |
1381 | handle_output(lguest_fd, arr[0], arr[1], device_list); | |
1382 | continue; | |
dde79789 | 1383 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1384 | } else if (errno == ENOENT) { |
1385 | char reason[1024] = { 0 }; | |
1386 | read(lguest_fd, reason, sizeof(reason)-1); | |
1387 | errx(1, "%s", reason); | |
dde79789 RR |
1388 | /* EAGAIN means the waker wanted us to look at some input. |
1389 | * Anything else means a bug or incompatible change. */ | |
8ca47e00 RR |
1390 | } else if (errno != EAGAIN) |
1391 | err(1, "Running guest failed"); | |
dde79789 RR |
1392 | |
1393 | /* Service input, then unset the BREAK which releases | |
1394 | * the Waker. */ | |
8ca47e00 RR |
1395 | handle_input(lguest_fd, device_list); |
1396 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
1397 | err(1, "Resetting break"); | |
1398 | } | |
1399 | } | |
dde79789 RR |
1400 | /* |
1401 | * This is the end of the Launcher. | |
1402 | * | |
1403 | * But wait! We've seen I/O from the Launcher, and we've seen I/O from the | |
1404 | * Drivers. If we were to see the Host kernel I/O code, our understanding | |
1405 | * would be complete... :*/ | |
8ca47e00 RR |
1406 | |
1407 | static struct option opts[] = { | |
1408 | { "verbose", 0, NULL, 'v' }, | |
1409 | { "sharenet", 1, NULL, 's' }, | |
1410 | { "tunnet", 1, NULL, 't' }, | |
1411 | { "block", 1, NULL, 'b' }, | |
1412 | { "initrd", 1, NULL, 'i' }, | |
1413 | { NULL }, | |
1414 | }; | |
1415 | static void usage(void) | |
1416 | { | |
1417 | errx(1, "Usage: lguest [--verbose] " | |
1418 | "[--sharenet=<filename>|--tunnet=(<ipaddr>|bridge:<bridgename>)\n" | |
1419 | "|--block=<filename>|--initrd=<filename>]...\n" | |
1420 | "<mem-in-mb> vmlinux [args...]"); | |
1421 | } | |
1422 | ||
3c6b5bfa | 1423 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1424 | int main(int argc, char *argv[]) |
1425 | { | |
dde79789 RR |
1426 | /* Memory, top-level pagetable, code startpoint, PAGE_OFFSET and size |
1427 | * of the (optional) initrd. */ | |
6570c459 | 1428 | unsigned long mem = 0, pgdir, start, page_offset, initrd_size = 0; |
dde79789 | 1429 | /* A temporary and the /dev/lguest file descriptor. */ |
6570c459 | 1430 | int i, c, lguest_fd; |
dde79789 | 1431 | /* The list of Guest devices, based on command line arguments. */ |
8ca47e00 | 1432 | struct device_list device_list; |
3c6b5bfa RR |
1433 | /* The boot information for the Guest. */ |
1434 | void *boot; | |
dde79789 | 1435 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1436 | const char *initrd_name = NULL; |
1437 | ||
dde79789 RR |
1438 | /* First we initialize the device list. Since console and network |
1439 | * device receive input from a file descriptor, we keep an fdset | |
1440 | * (infds) and the maximum fd number (max_infd) with the head of the | |
1441 | * list. We also keep a pointer to the last device, for easy appending | |
1442 | * to the list. */ | |
8ca47e00 RR |
1443 | device_list.max_infd = -1; |
1444 | device_list.dev = NULL; | |
1445 | device_list.lastdev = &device_list.dev; | |
1446 | FD_ZERO(&device_list.infds); | |
1447 | ||
dde79789 RR |
1448 | /* We need to know how much memory so we can set up the device |
1449 | * descriptor and memory pages for the devices as we parse the command | |
1450 | * line. So we quickly look through the arguments to find the amount | |
1451 | * of memory now. */ | |
6570c459 RR |
1452 | for (i = 1; i < argc; i++) { |
1453 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1454 | mem = atoi(argv[i]) * 1024 * 1024; |
1455 | /* We start by mapping anonymous pages over all of | |
1456 | * guest-physical memory range. This fills it with 0, | |
1457 | * and ensures that the Guest won't be killed when it | |
1458 | * tries to access it. */ | |
1459 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1460 | + DEVICE_PAGES); | |
1461 | guest_limit = mem; | |
1462 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
1463 | device_list.descs = get_pages(1); | |
6570c459 RR |
1464 | break; |
1465 | } | |
1466 | } | |
dde79789 RR |
1467 | |
1468 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1469 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1470 | switch (c) { | |
1471 | case 'v': | |
1472 | verbose = true; | |
1473 | break; | |
1474 | case 's': | |
1475 | setup_net_file(optarg, &device_list); | |
1476 | break; | |
1477 | case 't': | |
1478 | setup_tun_net(optarg, &device_list); | |
1479 | break; | |
1480 | case 'b': | |
1481 | setup_block_file(optarg, &device_list); | |
1482 | break; | |
1483 | case 'i': | |
1484 | initrd_name = optarg; | |
1485 | break; | |
1486 | default: | |
1487 | warnx("Unknown argument %s", argv[optind]); | |
1488 | usage(); | |
1489 | } | |
1490 | } | |
dde79789 RR |
1491 | /* After the other arguments we expect memory and kernel image name, |
1492 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1493 | if (optind + 2 > argc) |
1494 | usage(); | |
1495 | ||
3c6b5bfa RR |
1496 | verbose("Guest base is at %p\n", guest_base); |
1497 | ||
dde79789 | 1498 | /* We always have a console device */ |
8ca47e00 RR |
1499 | setup_console(&device_list); |
1500 | ||
8ca47e00 RR |
1501 | /* Now we load the kernel */ |
1502 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY), | |
1503 | &page_offset); | |
1504 | ||
3c6b5bfa RR |
1505 | /* Boot information is stashed at physical address 0 */ |
1506 | boot = from_guest_phys(0); | |
1507 | ||
dde79789 | 1508 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1509 | if (initrd_name) { |
1510 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1511 | /* These are the location in the Linux boot header where the |
1512 | * start and size of the initrd are expected to be found. */ | |
8ca47e00 RR |
1513 | *(unsigned long *)(boot+0x218) = mem - initrd_size; |
1514 | *(unsigned long *)(boot+0x21c) = initrd_size; | |
dde79789 | 1515 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
8ca47e00 RR |
1516 | *(unsigned char *)(boot+0x210) = 0xFF; |
1517 | } | |
1518 | ||
dde79789 | 1519 | /* Set up the initial linear pagetables, starting below the initrd. */ |
8ca47e00 RR |
1520 | pgdir = setup_pagetables(mem, initrd_size, page_offset); |
1521 | ||
dde79789 RR |
1522 | /* The Linux boot header contains an "E820" memory map: ours is a |
1523 | * simple, single region. */ | |
8ca47e00 RR |
1524 | *(char*)(boot+E820NR) = 1; |
1525 | *((struct e820entry *)(boot+E820MAP)) | |
1526 | = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 RR |
1527 | /* The boot header contains a command line pointer: we put the command |
1528 | * line after the boot header (at address 4096) */ | |
3c6b5bfa | 1529 | *(u32 *)(boot + 0x228) = 4096; |
8ca47e00 | 1530 | concat(boot + 4096, argv+optind+2); |
dde79789 RR |
1531 | |
1532 | /* The guest type value of "1" tells the Guest it's under lguest. */ | |
8ca47e00 RR |
1533 | *(int *)(boot + 0x23c) = 1; |
1534 | ||
dde79789 RR |
1535 | /* We tell the kernel to initialize the Guest: this returns the open |
1536 | * /dev/lguest file descriptor. */ | |
8ca47e00 | 1537 | lguest_fd = tell_kernel(pgdir, start, page_offset); |
dde79789 RR |
1538 | |
1539 | /* We fork off a child process, which wakes the Launcher whenever one | |
1540 | * of the input file descriptors needs attention. Otherwise we would | |
1541 | * run the Guest until it tries to output something. */ | |
8ca47e00 RR |
1542 | waker_fd = setup_waker(lguest_fd, &device_list); |
1543 | ||
dde79789 | 1544 | /* Finally, run the Guest. This doesn't return. */ |
8ca47e00 RR |
1545 | run_guest(lguest_fd, &device_list); |
1546 | } | |
f56a384e RR |
1547 | /*:*/ |
1548 | ||
1549 | /*M:999 | |
1550 | * Mastery is done: you now know everything I do. | |
1551 | * | |
1552 | * But surely you have seen code, features and bugs in your wanderings which | |
1553 | * you now yearn to attack? That is the real game, and I look forward to you | |
1554 | * patching and forking lguest into the Your-Name-Here-visor. | |
1555 | * | |
1556 | * Farewell, and good coding! | |
1557 | * Rusty Russell. | |
1558 | */ |