Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:100 |
2 | * This is the Launcher code, a simple program which lays out the "physical" | |
3 | * memory for the new Guest by mapping the kernel image and the virtual | |
4 | * devices, then opens /dev/lguest to tell the kernel about the Guest and | |
5 | * control it. | |
6 | :*/ | |
8ca47e00 RR |
7 | #define _LARGEFILE64_SOURCE |
8 | #define _GNU_SOURCE | |
9 | #include <stdio.h> | |
10 | #include <string.h> | |
11 | #include <unistd.h> | |
12 | #include <err.h> | |
13 | #include <stdint.h> | |
14 | #include <stdlib.h> | |
15 | #include <elf.h> | |
16 | #include <sys/mman.h> | |
6649bb7a | 17 | #include <sys/param.h> |
8ca47e00 RR |
18 | #include <sys/types.h> |
19 | #include <sys/stat.h> | |
20 | #include <sys/wait.h> | |
659a0e66 | 21 | #include <sys/eventfd.h> |
8ca47e00 RR |
22 | #include <fcntl.h> |
23 | #include <stdbool.h> | |
24 | #include <errno.h> | |
25 | #include <ctype.h> | |
26 | #include <sys/socket.h> | |
27 | #include <sys/ioctl.h> | |
28 | #include <sys/time.h> | |
29 | #include <time.h> | |
30 | #include <netinet/in.h> | |
31 | #include <net/if.h> | |
32 | #include <linux/sockios.h> | |
33 | #include <linux/if_tun.h> | |
34 | #include <sys/uio.h> | |
35 | #include <termios.h> | |
36 | #include <getopt.h> | |
37 | #include <zlib.h> | |
17cbca2b RR |
38 | #include <assert.h> |
39 | #include <sched.h> | |
a586d4f6 RR |
40 | #include <limits.h> |
41 | #include <stddef.h> | |
a161883a | 42 | #include <signal.h> |
b45d8cb0 | 43 | #include "linux/lguest_launcher.h" |
17cbca2b RR |
44 | #include "linux/virtio_config.h" |
45 | #include "linux/virtio_net.h" | |
46 | #include "linux/virtio_blk.h" | |
47 | #include "linux/virtio_console.h" | |
28fd6d7f | 48 | #include "linux/virtio_rng.h" |
17cbca2b | 49 | #include "linux/virtio_ring.h" |
d5d02d6d | 50 | #include "asm/bootparam.h" |
2e04ef76 RR |
51 | /*L:110 |
52 | * We can ignore the 39 include files we need for this program, but I do want | |
53 | * to draw attention to the use of kernel-style types. | |
db24e8c2 RR |
54 | * |
55 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
56 | * like these abbreviations, so we define them here. Note that u64 is always | |
57 | * unsigned long long, which works on all Linux systems: this means that we can | |
2e04ef76 RR |
58 | * use %llu in printf for any u64. |
59 | */ | |
db24e8c2 RR |
60 | typedef unsigned long long u64; |
61 | typedef uint32_t u32; | |
62 | typedef uint16_t u16; | |
63 | typedef uint8_t u8; | |
dde79789 | 64 | /*:*/ |
8ca47e00 RR |
65 | |
66 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
8ca47e00 RR |
67 | #define BRIDGE_PFX "bridge:" |
68 | #ifndef SIOCBRADDIF | |
69 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
70 | #endif | |
3c6b5bfa RR |
71 | /* We can have up to 256 pages for devices. */ |
72 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
73 | /* This will occupy 3 pages: it must be a power of 2. */ |
74 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 75 | |
2e04ef76 RR |
76 | /*L:120 |
77 | * verbose is both a global flag and a macro. The C preprocessor allows | |
78 | * this, and although I wouldn't recommend it, it works quite nicely here. | |
79 | */ | |
8ca47e00 RR |
80 | static bool verbose; |
81 | #define verbose(args...) \ | |
82 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
83 | /*:*/ |
84 | ||
3c6b5bfa RR |
85 | /* The pointer to the start of guest memory. */ |
86 | static void *guest_base; | |
87 | /* The maximum guest physical address allowed, and maximum possible. */ | |
88 | static unsigned long guest_limit, guest_max; | |
56739c80 RR |
89 | /* The /dev/lguest file descriptor. */ |
90 | static int lguest_fd; | |
8ca47e00 | 91 | |
e3283fa0 GOC |
92 | /* a per-cpu variable indicating whose vcpu is currently running */ |
93 | static unsigned int __thread cpu_id; | |
94 | ||
dde79789 | 95 | /* This is our list of devices. */ |
8ca47e00 RR |
96 | struct device_list |
97 | { | |
17cbca2b RR |
98 | /* Counter to assign interrupt numbers. */ |
99 | unsigned int next_irq; | |
100 | ||
101 | /* Counter to print out convenient device numbers. */ | |
102 | unsigned int device_num; | |
103 | ||
dde79789 | 104 | /* The descriptor page for the devices. */ |
17cbca2b RR |
105 | u8 *descpage; |
106 | ||
dde79789 | 107 | /* A single linked list of devices. */ |
8ca47e00 | 108 | struct device *dev; |
2e04ef76 | 109 | /* And a pointer to the last device for easy append. */ |
a586d4f6 | 110 | struct device *lastdev; |
8ca47e00 RR |
111 | }; |
112 | ||
17cbca2b RR |
113 | /* The list of Guest devices, based on command line arguments. */ |
114 | static struct device_list devices; | |
115 | ||
dde79789 | 116 | /* The device structure describes a single device. */ |
8ca47e00 RR |
117 | struct device |
118 | { | |
dde79789 | 119 | /* The linked-list pointer. */ |
8ca47e00 | 120 | struct device *next; |
17cbca2b | 121 | |
713b15b3 | 122 | /* The device's descriptor, as mapped into the Guest. */ |
8ca47e00 | 123 | struct lguest_device_desc *desc; |
17cbca2b | 124 | |
713b15b3 RR |
125 | /* We can't trust desc values once Guest has booted: we use these. */ |
126 | unsigned int feature_len; | |
127 | unsigned int num_vq; | |
128 | ||
17cbca2b RR |
129 | /* The name of this device, for --verbose. */ |
130 | const char *name; | |
8ca47e00 | 131 | |
17cbca2b RR |
132 | /* Any queues attached to this device */ |
133 | struct virtqueue *vq; | |
8ca47e00 | 134 | |
659a0e66 RR |
135 | /* Is it operational */ |
136 | bool running; | |
a007a751 | 137 | |
8ca47e00 RR |
138 | /* Device-specific data. */ |
139 | void *priv; | |
140 | }; | |
141 | ||
17cbca2b RR |
142 | /* The virtqueue structure describes a queue attached to a device. */ |
143 | struct virtqueue | |
144 | { | |
145 | struct virtqueue *next; | |
146 | ||
147 | /* Which device owns me. */ | |
148 | struct device *dev; | |
149 | ||
150 | /* The configuration for this queue. */ | |
151 | struct lguest_vqconfig config; | |
152 | ||
153 | /* The actual ring of buffers. */ | |
154 | struct vring vring; | |
155 | ||
156 | /* Last available index we saw. */ | |
157 | u16 last_avail_idx; | |
158 | ||
95c517c0 RR |
159 | /* How many are used since we sent last irq? */ |
160 | unsigned int pending_used; | |
161 | ||
659a0e66 RR |
162 | /* Eventfd where Guest notifications arrive. */ |
163 | int eventfd; | |
20887611 | 164 | |
659a0e66 RR |
165 | /* Function for the thread which is servicing this virtqueue. */ |
166 | void (*service)(struct virtqueue *vq); | |
167 | pid_t thread; | |
17cbca2b RR |
168 | }; |
169 | ||
ec04b13f BR |
170 | /* Remember the arguments to the program so we can "reboot" */ |
171 | static char **main_args; | |
172 | ||
659a0e66 RR |
173 | /* The original tty settings to restore on exit. */ |
174 | static struct termios orig_term; | |
175 | ||
2e04ef76 RR |
176 | /* |
177 | * We have to be careful with barriers: our devices are all run in separate | |
f7027c63 | 178 | * threads and so we need to make sure that changes visible to the Guest happen |
2e04ef76 RR |
179 | * in precise order. |
180 | */ | |
f7027c63 | 181 | #define wmb() __asm__ __volatile__("" : : : "memory") |
b60da13f | 182 | #define mb() __asm__ __volatile__("" : : : "memory") |
17cbca2b | 183 | |
2e04ef76 RR |
184 | /* |
185 | * Convert an iovec element to the given type. | |
17cbca2b RR |
186 | * |
187 | * This is a fairly ugly trick: we need to know the size of the type and | |
188 | * alignment requirement to check the pointer is kosher. It's also nice to | |
189 | * have the name of the type in case we report failure. | |
190 | * | |
191 | * Typing those three things all the time is cumbersome and error prone, so we | |
2e04ef76 RR |
192 | * have a macro which sets them all up and passes to the real function. |
193 | */ | |
17cbca2b RR |
194 | #define convert(iov, type) \ |
195 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
196 | ||
197 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
198 | const char *name) | |
199 | { | |
200 | if (iov->iov_len != size) | |
201 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
202 | if ((unsigned long)iov->iov_base % align != 0) | |
203 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
204 | return iov->iov_base; | |
205 | } | |
206 | ||
b5111790 RR |
207 | /* Wrapper for the last available index. Makes it easier to change. */ |
208 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
209 | ||
2e04ef76 RR |
210 | /* |
211 | * The virtio configuration space is defined to be little-endian. x86 is | |
212 | * little-endian too, but it's nice to be explicit so we have these helpers. | |
213 | */ | |
17cbca2b RR |
214 | #define cpu_to_le16(v16) (v16) |
215 | #define cpu_to_le32(v32) (v32) | |
216 | #define cpu_to_le64(v64) (v64) | |
217 | #define le16_to_cpu(v16) (v16) | |
218 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 219 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 220 | |
28fd6d7f RR |
221 | /* Is this iovec empty? */ |
222 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
223 | { | |
224 | unsigned int i; | |
225 | ||
226 | for (i = 0; i < num_iov; i++) | |
227 | if (iov[i].iov_len) | |
228 | return false; | |
229 | return true; | |
230 | } | |
231 | ||
232 | /* Take len bytes from the front of this iovec. */ | |
233 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) | |
234 | { | |
235 | unsigned int i; | |
236 | ||
237 | for (i = 0; i < num_iov; i++) { | |
238 | unsigned int used; | |
239 | ||
240 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
241 | iov[i].iov_base += used; | |
242 | iov[i].iov_len -= used; | |
243 | len -= used; | |
244 | } | |
245 | assert(len == 0); | |
246 | } | |
247 | ||
6e5aa7ef RR |
248 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
249 | static u8 *get_feature_bits(struct device *dev) | |
250 | { | |
251 | return (u8 *)(dev->desc + 1) | |
713b15b3 | 252 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
6e5aa7ef RR |
253 | } |
254 | ||
2e04ef76 RR |
255 | /*L:100 |
256 | * The Launcher code itself takes us out into userspace, that scary place where | |
257 | * pointers run wild and free! Unfortunately, like most userspace programs, | |
258 | * it's quite boring (which is why everyone likes to hack on the kernel!). | |
259 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get | |
260 | * you through this section. Or, maybe not. | |
3c6b5bfa RR |
261 | * |
262 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
263 | * memory and stores it in "guest_base". In other words, Guest physical == | |
264 | * Launcher virtual with an offset. | |
265 | * | |
266 | * This can be tough to get your head around, but usually it just means that we | |
267 | * use these trivial conversion functions when the Guest gives us it's | |
2e04ef76 RR |
268 | * "physical" addresses: |
269 | */ | |
3c6b5bfa RR |
270 | static void *from_guest_phys(unsigned long addr) |
271 | { | |
272 | return guest_base + addr; | |
273 | } | |
274 | ||
275 | static unsigned long to_guest_phys(const void *addr) | |
276 | { | |
277 | return (addr - guest_base); | |
278 | } | |
279 | ||
dde79789 RR |
280 | /*L:130 |
281 | * Loading the Kernel. | |
282 | * | |
283 | * We start with couple of simple helper routines. open_or_die() avoids | |
2e04ef76 RR |
284 | * error-checking code cluttering the callers: |
285 | */ | |
8ca47e00 RR |
286 | static int open_or_die(const char *name, int flags) |
287 | { | |
288 | int fd = open(name, flags); | |
289 | if (fd < 0) | |
290 | err(1, "Failed to open %s", name); | |
291 | return fd; | |
292 | } | |
293 | ||
3c6b5bfa RR |
294 | /* map_zeroed_pages() takes a number of pages. */ |
295 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 296 | { |
3c6b5bfa RR |
297 | int fd = open_or_die("/dev/zero", O_RDONLY); |
298 | void *addr; | |
8ca47e00 | 299 | |
2e04ef76 RR |
300 | /* |
301 | * We use a private mapping (ie. if we write to the page, it will be | |
302 | * copied). | |
303 | */ | |
3c6b5bfa RR |
304 | addr = mmap(NULL, getpagesize() * num, |
305 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
306 | if (addr == MAP_FAILED) | |
307 | err(1, "Mmaping %u pages of /dev/zero", num); | |
34bdaab4 | 308 | close(fd); |
3c6b5bfa RR |
309 | |
310 | return addr; | |
311 | } | |
312 | ||
313 | /* Get some more pages for a device. */ | |
314 | static void *get_pages(unsigned int num) | |
315 | { | |
316 | void *addr = from_guest_phys(guest_limit); | |
317 | ||
318 | guest_limit += num * getpagesize(); | |
319 | if (guest_limit > guest_max) | |
320 | errx(1, "Not enough memory for devices"); | |
321 | return addr; | |
8ca47e00 RR |
322 | } |
323 | ||
2e04ef76 RR |
324 | /* |
325 | * This routine is used to load the kernel or initrd. It tries mmap, but if | |
6649bb7a | 326 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
2e04ef76 RR |
327 | * it falls back to reading the memory in. |
328 | */ | |
6649bb7a RM |
329 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
330 | { | |
331 | ssize_t r; | |
332 | ||
2e04ef76 RR |
333 | /* |
334 | * We map writable even though for some segments are marked read-only. | |
6649bb7a RM |
335 | * The kernel really wants to be writable: it patches its own |
336 | * instructions. | |
337 | * | |
338 | * MAP_PRIVATE means that the page won't be copied until a write is | |
339 | * done to it. This allows us to share untouched memory between | |
2e04ef76 RR |
340 | * Guests. |
341 | */ | |
6649bb7a RM |
342 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, |
343 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
344 | return; | |
345 | ||
346 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
347 | r = pread(fd, addr, len, offset); | |
348 | if (r != len) | |
349 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
350 | } | |
351 | ||
2e04ef76 RR |
352 | /* |
353 | * This routine takes an open vmlinux image, which is in ELF, and maps it into | |
dde79789 RR |
354 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
355 | * by all modern binaries on Linux including the kernel. | |
356 | * | |
357 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
358 | * address. We use the physical address; the Guest will map itself to the |
359 | * virtual address. | |
dde79789 | 360 | * |
2e04ef76 RR |
361 | * We return the starting address. |
362 | */ | |
47436aa4 | 363 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 364 | { |
8ca47e00 RR |
365 | Elf32_Phdr phdr[ehdr->e_phnum]; |
366 | unsigned int i; | |
8ca47e00 | 367 | |
2e04ef76 RR |
368 | /* |
369 | * Sanity checks on the main ELF header: an x86 executable with a | |
370 | * reasonable number of correctly-sized program headers. | |
371 | */ | |
8ca47e00 RR |
372 | if (ehdr->e_type != ET_EXEC |
373 | || ehdr->e_machine != EM_386 | |
374 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
375 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
376 | errx(1, "Malformed elf header"); | |
377 | ||
2e04ef76 RR |
378 | /* |
379 | * An ELF executable contains an ELF header and a number of "program" | |
dde79789 | 380 | * headers which indicate which parts ("segments") of the program to |
2e04ef76 RR |
381 | * load where. |
382 | */ | |
dde79789 RR |
383 | |
384 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
385 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
386 | err(1, "Seeking to program headers"); | |
387 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
388 | err(1, "Reading program headers"); | |
389 | ||
2e04ef76 RR |
390 | /* |
391 | * Try all the headers: there are usually only three. A read-only one, | |
392 | * a read-write one, and a "note" section which we don't load. | |
393 | */ | |
8ca47e00 | 394 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 395 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
396 | if (phdr[i].p_type != PT_LOAD) |
397 | continue; | |
398 | ||
399 | verbose("Section %i: size %i addr %p\n", | |
400 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
401 | ||
6649bb7a | 402 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 403 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 404 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
405 | } |
406 | ||
814a0e5c RR |
407 | /* The entry point is given in the ELF header. */ |
408 | return ehdr->e_entry; | |
8ca47e00 RR |
409 | } |
410 | ||
2e04ef76 RR |
411 | /*L:150 |
412 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed | |
413 | * to jump into it and it will unpack itself. We used to have to perform some | |
414 | * hairy magic because the unpacking code scared me. | |
dde79789 | 415 | * |
5bbf89fc RR |
416 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
417 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
2e04ef76 RR |
418 | * the funky header so we know where in the file to load, and away we go! |
419 | */ | |
47436aa4 | 420 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 421 | { |
43d33b21 | 422 | struct boot_params boot; |
5bbf89fc RR |
423 | int r; |
424 | /* Modern bzImages get loaded at 1M. */ | |
425 | void *p = from_guest_phys(0x100000); | |
426 | ||
2e04ef76 RR |
427 | /* |
428 | * Go back to the start of the file and read the header. It should be | |
429 | * a Linux boot header (see Documentation/x86/i386/boot.txt) | |
430 | */ | |
5bbf89fc | 431 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 432 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 433 | |
43d33b21 RR |
434 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
435 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
436 | errx(1, "This doesn't look like a bzImage to me"); |
437 | ||
43d33b21 RR |
438 | /* Skip over the extra sectors of the header. */ |
439 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
440 | |
441 | /* Now read everything into memory. in nice big chunks. */ | |
442 | while ((r = read(fd, p, 65536)) > 0) | |
443 | p += r; | |
444 | ||
43d33b21 RR |
445 | /* Finally, code32_start tells us where to enter the kernel. */ |
446 | return boot.hdr.code32_start; | |
8ca47e00 RR |
447 | } |
448 | ||
2e04ef76 RR |
449 | /*L:140 |
450 | * Loading the kernel is easy when it's a "vmlinux", but most kernels | |
e1e72965 | 451 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
2e04ef76 RR |
452 | * work, we can load those, too. |
453 | */ | |
47436aa4 | 454 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
455 | { |
456 | Elf32_Ehdr hdr; | |
457 | ||
dde79789 | 458 | /* Read in the first few bytes. */ |
8ca47e00 RR |
459 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
460 | err(1, "Reading kernel"); | |
461 | ||
dde79789 | 462 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 463 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 464 | return map_elf(fd, &hdr); |
8ca47e00 | 465 | |
a6bd8e13 | 466 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 467 | return load_bzimage(fd); |
8ca47e00 RR |
468 | } |
469 | ||
2e04ef76 RR |
470 | /* |
471 | * This is a trivial little helper to align pages. Andi Kleen hated it because | |
dde79789 RR |
472 | * it calls getpagesize() twice: "it's dumb code." |
473 | * | |
474 | * Kernel guys get really het up about optimization, even when it's not | |
2e04ef76 RR |
475 | * necessary. I leave this code as a reaction against that. |
476 | */ | |
8ca47e00 RR |
477 | static inline unsigned long page_align(unsigned long addr) |
478 | { | |
dde79789 | 479 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
480 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
481 | } | |
482 | ||
2e04ef76 RR |
483 | /*L:180 |
484 | * An "initial ram disk" is a disk image loaded into memory along with the | |
485 | * kernel which the kernel can use to boot from without needing any drivers. | |
486 | * Most distributions now use this as standard: the initrd contains the code to | |
487 | * load the appropriate driver modules for the current machine. | |
dde79789 RR |
488 | * |
489 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
2e04ef76 RR |
490 | * kernels. He sent me this (and tells me when I break it). |
491 | */ | |
8ca47e00 RR |
492 | static unsigned long load_initrd(const char *name, unsigned long mem) |
493 | { | |
494 | int ifd; | |
495 | struct stat st; | |
496 | unsigned long len; | |
8ca47e00 RR |
497 | |
498 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 499 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
500 | if (fstat(ifd, &st) < 0) |
501 | err(1, "fstat() on initrd '%s'", name); | |
502 | ||
2e04ef76 RR |
503 | /* |
504 | * We map the initrd at the top of memory, but mmap wants it to be | |
505 | * page-aligned, so we round the size up for that. | |
506 | */ | |
8ca47e00 | 507 | len = page_align(st.st_size); |
3c6b5bfa | 508 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
2e04ef76 RR |
509 | /* |
510 | * Once a file is mapped, you can close the file descriptor. It's a | |
511 | * little odd, but quite useful. | |
512 | */ | |
8ca47e00 | 513 | close(ifd); |
6649bb7a | 514 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
515 | |
516 | /* We return the initrd size. */ | |
8ca47e00 RR |
517 | return len; |
518 | } | |
e1e72965 | 519 | /*:*/ |
8ca47e00 | 520 | |
2e04ef76 RR |
521 | /* |
522 | * Simple routine to roll all the commandline arguments together with spaces | |
523 | * between them. | |
524 | */ | |
8ca47e00 RR |
525 | static void concat(char *dst, char *args[]) |
526 | { | |
527 | unsigned int i, len = 0; | |
528 | ||
529 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
530 | if (i) { |
531 | strcat(dst+len, " "); | |
532 | len++; | |
533 | } | |
8ca47e00 | 534 | strcpy(dst+len, args[i]); |
1ef36fa6 | 535 | len += strlen(args[i]); |
8ca47e00 RR |
536 | } |
537 | /* In case it's empty. */ | |
538 | dst[len] = '\0'; | |
539 | } | |
540 | ||
2e04ef76 RR |
541 | /*L:185 |
542 | * This is where we actually tell the kernel to initialize the Guest. We | |
e1e72965 | 543 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
58a24566 | 544 | * the base of Guest "physical" memory, the top physical page to allow and the |
2e04ef76 RR |
545 | * entry point for the Guest. |
546 | */ | |
56739c80 | 547 | static void tell_kernel(unsigned long start) |
8ca47e00 | 548 | { |
511801dc JS |
549 | unsigned long args[] = { LHREQ_INITIALIZE, |
550 | (unsigned long)guest_base, | |
58a24566 | 551 | guest_limit / getpagesize(), start }; |
3c6b5bfa RR |
552 | verbose("Guest: %p - %p (%#lx)\n", |
553 | guest_base, guest_base + guest_limit, guest_limit); | |
56739c80 RR |
554 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
555 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 556 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 557 | } |
dde79789 | 558 | /*:*/ |
8ca47e00 | 559 | |
e1e72965 | 560 | /* |
dde79789 RR |
561 | * Device Handling. |
562 | * | |
e1e72965 | 563 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 564 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 565 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
566 | * if something funny is going on: |
567 | */ | |
8ca47e00 RR |
568 | static void *_check_pointer(unsigned long addr, unsigned int size, |
569 | unsigned int line) | |
570 | { | |
2e04ef76 RR |
571 | /* |
572 | * We have to separately check addr and addr+size, because size could | |
573 | * be huge and addr + size might wrap around. | |
574 | */ | |
3c6b5bfa | 575 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 576 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
2e04ef76 RR |
577 | /* |
578 | * We return a pointer for the caller's convenience, now we know it's | |
579 | * safe to use. | |
580 | */ | |
3c6b5bfa | 581 | return from_guest_phys(addr); |
8ca47e00 | 582 | } |
dde79789 | 583 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
584 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
585 | ||
2e04ef76 RR |
586 | /* |
587 | * Each buffer in the virtqueues is actually a chain of descriptors. This | |
e1e72965 | 588 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
2e04ef76 RR |
589 | * at the end. |
590 | */ | |
d1f0132e MM |
591 | static unsigned next_desc(struct vring_desc *desc, |
592 | unsigned int i, unsigned int max) | |
17cbca2b RR |
593 | { |
594 | unsigned int next; | |
595 | ||
596 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
597 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
598 | return max; | |
17cbca2b RR |
599 | |
600 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 601 | next = desc[i].next; |
17cbca2b RR |
602 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
603 | wmb(); | |
604 | ||
d1f0132e | 605 | if (next >= max) |
17cbca2b RR |
606 | errx(1, "Desc next is %u", next); |
607 | ||
608 | return next; | |
609 | } | |
610 | ||
38bc2b8c RR |
611 | /* This actually sends the interrupt for this virtqueue */ |
612 | static void trigger_irq(struct virtqueue *vq) | |
613 | { | |
614 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; | |
615 | ||
95c517c0 RR |
616 | /* Don't inform them if nothing used. */ |
617 | if (!vq->pending_used) | |
618 | return; | |
619 | vq->pending_used = 0; | |
620 | ||
38bc2b8c RR |
621 | /* If they don't want an interrupt, don't send one, unless empty. */ |
622 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
623 | && lg_last_avail(vq) != vq->vring.avail->idx) | |
624 | return; | |
625 | ||
626 | /* Send the Guest an interrupt tell them we used something up. */ | |
627 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
628 | err(1, "Triggering irq %i", vq->config.irq); | |
629 | } | |
630 | ||
2e04ef76 RR |
631 | /* |
632 | * This looks in the virtqueue and for the first available buffer, and converts | |
17cbca2b RR |
633 | * it to an iovec for convenient access. Since descriptors consist of some |
634 | * number of output then some number of input descriptors, it's actually two | |
635 | * iovecs, but we pack them into one and note how many of each there were. | |
636 | * | |
2e04ef76 RR |
637 | * This function returns the descriptor number found. |
638 | */ | |
659a0e66 RR |
639 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
640 | struct iovec iov[], | |
641 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 642 | { |
d1f0132e MM |
643 | unsigned int i, head, max; |
644 | struct vring_desc *desc; | |
659a0e66 RR |
645 | u16 last_avail = lg_last_avail(vq); |
646 | ||
647 | while (last_avail == vq->vring.avail->idx) { | |
648 | u64 event; | |
649 | ||
38bc2b8c RR |
650 | /* OK, tell Guest about progress up to now. */ |
651 | trigger_irq(vq); | |
652 | ||
b60da13f RR |
653 | /* OK, now we need to know about added descriptors. */ |
654 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
655 | ||
2e04ef76 RR |
656 | /* |
657 | * They could have slipped one in as we were doing that: make | |
658 | * sure it's written, then check again. | |
659 | */ | |
b60da13f RR |
660 | mb(); |
661 | if (last_avail != vq->vring.avail->idx) { | |
662 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
663 | break; | |
664 | } | |
665 | ||
659a0e66 RR |
666 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
667 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
668 | errx(1, "Event read failed?"); | |
b60da13f RR |
669 | |
670 | /* We don't need to be notified again. */ | |
671 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 672 | } |
17cbca2b RR |
673 | |
674 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 675 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
17cbca2b | 676 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 677 | last_avail, vq->vring.avail->idx); |
17cbca2b | 678 | |
2e04ef76 RR |
679 | /* |
680 | * Grab the next descriptor number they're advertising, and increment | |
681 | * the index we've seen. | |
682 | */ | |
b5111790 RR |
683 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
684 | lg_last_avail(vq)++; | |
17cbca2b RR |
685 | |
686 | /* If their number is silly, that's a fatal mistake. */ | |
687 | if (head >= vq->vring.num) | |
688 | errx(1, "Guest says index %u is available", head); | |
689 | ||
690 | /* When we start there are none of either input nor output. */ | |
691 | *out_num = *in_num = 0; | |
692 | ||
d1f0132e MM |
693 | max = vq->vring.num; |
694 | desc = vq->vring.desc; | |
17cbca2b | 695 | i = head; |
d1f0132e | 696 | |
2e04ef76 RR |
697 | /* |
698 | * If this is an indirect entry, then this buffer contains a descriptor | |
699 | * table which we handle as if it's any normal descriptor chain. | |
700 | */ | |
d1f0132e MM |
701 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { |
702 | if (desc[i].len % sizeof(struct vring_desc)) | |
703 | errx(1, "Invalid size for indirect buffer table"); | |
704 | ||
705 | max = desc[i].len / sizeof(struct vring_desc); | |
706 | desc = check_pointer(desc[i].addr, desc[i].len); | |
707 | i = 0; | |
708 | } | |
709 | ||
17cbca2b RR |
710 | do { |
711 | /* Grab the first descriptor, and check it's OK. */ | |
d1f0132e | 712 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 713 | iov[*out_num + *in_num].iov_base |
d1f0132e | 714 | = check_pointer(desc[i].addr, desc[i].len); |
17cbca2b | 715 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 716 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
717 | (*in_num)++; |
718 | else { | |
2e04ef76 RR |
719 | /* |
720 | * If it's an output descriptor, they're all supposed | |
721 | * to come before any input descriptors. | |
722 | */ | |
17cbca2b RR |
723 | if (*in_num) |
724 | errx(1, "Descriptor has out after in"); | |
725 | (*out_num)++; | |
726 | } | |
727 | ||
728 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 729 | if (*out_num + *in_num > max) |
17cbca2b | 730 | errx(1, "Looped descriptor"); |
d1f0132e | 731 | } while ((i = next_desc(desc, i, max)) != max); |
dde79789 | 732 | |
17cbca2b | 733 | return head; |
8ca47e00 RR |
734 | } |
735 | ||
2e04ef76 RR |
736 | /* |
737 | * After we've used one of their buffers, we tell them about it. We'll then | |
738 | * want to send them an interrupt, using trigger_irq(). | |
739 | */ | |
17cbca2b | 740 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
8ca47e00 | 741 | { |
17cbca2b RR |
742 | struct vring_used_elem *used; |
743 | ||
2e04ef76 RR |
744 | /* |
745 | * The virtqueue contains a ring of used buffers. Get a pointer to the | |
746 | * next entry in that used ring. | |
747 | */ | |
17cbca2b RR |
748 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
749 | used->id = head; | |
750 | used->len = len; | |
751 | /* Make sure buffer is written before we update index. */ | |
752 | wmb(); | |
753 | vq->vring.used->idx++; | |
95c517c0 | 754 | vq->pending_used++; |
8ca47e00 RR |
755 | } |
756 | ||
17cbca2b | 757 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 758 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 759 | { |
17cbca2b | 760 | add_used(vq, head, len); |
56739c80 | 761 | trigger_irq(vq); |
8ca47e00 RR |
762 | } |
763 | ||
e1e72965 RR |
764 | /* |
765 | * The Console | |
766 | * | |
2e04ef76 RR |
767 | * We associate some data with the console for our exit hack. |
768 | */ | |
8ca47e00 RR |
769 | struct console_abort |
770 | { | |
dde79789 | 771 | /* How many times have they hit ^C? */ |
8ca47e00 | 772 | int count; |
dde79789 | 773 | /* When did they start? */ |
8ca47e00 RR |
774 | struct timeval start; |
775 | }; | |
776 | ||
dde79789 | 777 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 778 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 779 | { |
8ca47e00 | 780 | int len; |
17cbca2b | 781 | unsigned int head, in_num, out_num; |
659a0e66 RR |
782 | struct console_abort *abort = vq->dev->priv; |
783 | struct iovec iov[vq->vring.num]; | |
56ae43df | 784 | |
659a0e66 RR |
785 | /* Make sure there's a descriptor waiting. */ |
786 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); | |
56ae43df | 787 | if (out_num) |
17cbca2b | 788 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 789 | |
659a0e66 RR |
790 | /* Read it in. */ |
791 | len = readv(STDIN_FILENO, iov, in_num); | |
8ca47e00 | 792 | if (len <= 0) { |
659a0e66 | 793 | /* Ran out of input? */ |
8ca47e00 | 794 | warnx("Failed to get console input, ignoring console."); |
2e04ef76 RR |
795 | /* |
796 | * For simplicity, dying threads kill the whole Launcher. So | |
797 | * just nap here. | |
798 | */ | |
659a0e66 RR |
799 | for (;;) |
800 | pause(); | |
8ca47e00 RR |
801 | } |
802 | ||
659a0e66 | 803 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 804 | |
2e04ef76 RR |
805 | /* |
806 | * Three ^C within one second? Exit. | |
dde79789 | 807 | * |
659a0e66 RR |
808 | * This is such a hack, but works surprisingly well. Each ^C has to |
809 | * be in a buffer by itself, so they can't be too fast. But we check | |
810 | * that we get three within about a second, so they can't be too | |
2e04ef76 RR |
811 | * slow. |
812 | */ | |
659a0e66 | 813 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
8ca47e00 | 814 | abort->count = 0; |
659a0e66 RR |
815 | return; |
816 | } | |
8ca47e00 | 817 | |
659a0e66 RR |
818 | abort->count++; |
819 | if (abort->count == 1) | |
820 | gettimeofday(&abort->start, NULL); | |
821 | else if (abort->count == 3) { | |
822 | struct timeval now; | |
823 | gettimeofday(&now, NULL); | |
824 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
825 | if (now.tv_sec <= abort->start.tv_sec+1) | |
826 | kill(0, SIGINT); | |
827 | abort->count = 0; | |
828 | } | |
8ca47e00 RR |
829 | } |
830 | ||
659a0e66 RR |
831 | /* This is the routine which handles console output (ie. stdout). */ |
832 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 833 | { |
17cbca2b | 834 | unsigned int head, out, in; |
17cbca2b RR |
835 | struct iovec iov[vq->vring.num]; |
836 | ||
659a0e66 RR |
837 | head = wait_for_vq_desc(vq, iov, &out, &in); |
838 | if (in) | |
839 | errx(1, "Input buffers in console output queue?"); | |
840 | while (!iov_empty(iov, out)) { | |
841 | int len = writev(STDOUT_FILENO, iov, out); | |
842 | if (len <= 0) | |
843 | err(1, "Write to stdout gave %i", len); | |
844 | iov_consume(iov, out, len); | |
17cbca2b | 845 | } |
38bc2b8c | 846 | add_used(vq, head, 0); |
a161883a RR |
847 | } |
848 | ||
e1e72965 RR |
849 | /* |
850 | * The Network | |
851 | * | |
852 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 853 | * and write them to /dev/net/tun. |
a6bd8e13 | 854 | */ |
659a0e66 RR |
855 | struct net_info { |
856 | int tunfd; | |
857 | }; | |
858 | ||
859 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 860 | { |
659a0e66 RR |
861 | struct net_info *net_info = vq->dev->priv; |
862 | unsigned int head, out, in; | |
17cbca2b | 863 | struct iovec iov[vq->vring.num]; |
a161883a | 864 | |
659a0e66 RR |
865 | head = wait_for_vq_desc(vq, iov, &out, &in); |
866 | if (in) | |
867 | errx(1, "Input buffers in net output queue?"); | |
868 | if (writev(net_info->tunfd, iov, out) < 0) | |
869 | errx(1, "Write to tun failed?"); | |
38bc2b8c | 870 | add_used(vq, head, 0); |
8ca47e00 RR |
871 | } |
872 | ||
4a8962e2 RR |
873 | /* Will reading from this file descriptor block? */ |
874 | static bool will_block(int fd) | |
875 | { | |
876 | fd_set fdset; | |
877 | struct timeval zero = { 0, 0 }; | |
878 | FD_ZERO(&fdset); | |
879 | FD_SET(fd, &fdset); | |
880 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
881 | } | |
882 | ||
2e04ef76 | 883 | /* This handles packets coming in from the tun device to our Guest. */ |
659a0e66 | 884 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 885 | { |
8ca47e00 | 886 | int len; |
659a0e66 RR |
887 | unsigned int head, out, in; |
888 | struct iovec iov[vq->vring.num]; | |
889 | struct net_info *net_info = vq->dev->priv; | |
890 | ||
891 | head = wait_for_vq_desc(vq, iov, &out, &in); | |
892 | if (out) | |
893 | errx(1, "Output buffers in net input queue?"); | |
4a8962e2 RR |
894 | |
895 | /* Deliver interrupt now, since we're about to sleep. */ | |
896 | if (vq->pending_used && will_block(net_info->tunfd)) | |
897 | trigger_irq(vq); | |
898 | ||
659a0e66 | 899 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 900 | if (len <= 0) |
659a0e66 | 901 | err(1, "Failed to read from tun."); |
4a8962e2 | 902 | add_used(vq, head, len); |
659a0e66 | 903 | } |
dde79789 | 904 | |
659a0e66 RR |
905 | /* This is the helper to create threads. */ |
906 | static int do_thread(void *_vq) | |
907 | { | |
908 | struct virtqueue *vq = _vq; | |
17cbca2b | 909 | |
659a0e66 RR |
910 | for (;;) |
911 | vq->service(vq); | |
912 | return 0; | |
913 | } | |
17cbca2b | 914 | |
2e04ef76 RR |
915 | /* |
916 | * When a child dies, we kill our entire process group with SIGTERM. This | |
917 | * also has the side effect that the shell restores the console for us! | |
918 | */ | |
659a0e66 RR |
919 | static void kill_launcher(int signal) |
920 | { | |
921 | kill(0, SIGTERM); | |
8ca47e00 RR |
922 | } |
923 | ||
659a0e66 | 924 | static void reset_device(struct device *dev) |
56ae43df | 925 | { |
659a0e66 RR |
926 | struct virtqueue *vq; |
927 | ||
928 | verbose("Resetting device %s\n", dev->name); | |
929 | ||
930 | /* Clear any features they've acked. */ | |
931 | memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); | |
932 | ||
933 | /* We're going to be explicitly killing threads, so ignore them. */ | |
934 | signal(SIGCHLD, SIG_IGN); | |
935 | ||
936 | /* Zero out the virtqueues, get rid of their threads */ | |
937 | for (vq = dev->vq; vq; vq = vq->next) { | |
938 | if (vq->thread != (pid_t)-1) { | |
939 | kill(vq->thread, SIGTERM); | |
940 | waitpid(vq->thread, NULL, 0); | |
941 | vq->thread = (pid_t)-1; | |
942 | } | |
943 | memset(vq->vring.desc, 0, | |
944 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); | |
945 | lg_last_avail(vq) = 0; | |
946 | } | |
947 | dev->running = false; | |
948 | ||
949 | /* Now we care if threads die. */ | |
950 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
951 | } |
952 | ||
659a0e66 | 953 | static void create_thread(struct virtqueue *vq) |
5dae785a | 954 | { |
2e04ef76 RR |
955 | /* |
956 | * Create stack for thread and run it. Since the stack grows upwards, | |
957 | * we point the stack pointer to the end of this region. | |
958 | */ | |
659a0e66 RR |
959 | char *stack = malloc(32768); |
960 | unsigned long args[] = { LHREQ_EVENTFD, | |
961 | vq->config.pfn*getpagesize(), 0 }; | |
962 | ||
963 | /* Create a zero-initialized eventfd. */ | |
964 | vq->eventfd = eventfd(0, 0); | |
965 | if (vq->eventfd < 0) | |
966 | err(1, "Creating eventfd"); | |
967 | args[2] = vq->eventfd; | |
968 | ||
969 | /* Attach an eventfd to this virtqueue: it will go off | |
970 | * when the Guest does an LHCALL_NOTIFY for this vq. */ | |
971 | if (write(lguest_fd, &args, sizeof(args)) != 0) | |
972 | err(1, "Attaching eventfd"); | |
973 | ||
974 | /* CLONE_VM: because it has to access the Guest memory, and | |
975 | * SIGCHLD so we get a signal if it dies. */ | |
976 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); | |
977 | if (vq->thread == (pid_t)-1) | |
978 | err(1, "Creating clone"); | |
979 | /* We close our local copy, now the child has it. */ | |
980 | close(vq->eventfd); | |
5dae785a RR |
981 | } |
982 | ||
659a0e66 | 983 | static void start_device(struct device *dev) |
6e5aa7ef | 984 | { |
659a0e66 | 985 | unsigned int i; |
6e5aa7ef RR |
986 | struct virtqueue *vq; |
987 | ||
659a0e66 RR |
988 | verbose("Device %s OK: offered", dev->name); |
989 | for (i = 0; i < dev->feature_len; i++) | |
990 | verbose(" %02x", get_feature_bits(dev)[i]); | |
991 | verbose(", accepted"); | |
992 | for (i = 0; i < dev->feature_len; i++) | |
993 | verbose(" %02x", get_feature_bits(dev) | |
994 | [dev->feature_len+i]); | |
995 | ||
996 | for (vq = dev->vq; vq; vq = vq->next) { | |
997 | if (vq->service) | |
998 | create_thread(vq); | |
999 | } | |
1000 | dev->running = true; | |
1001 | } | |
1002 | ||
1003 | static void cleanup_devices(void) | |
1004 | { | |
1005 | struct device *dev; | |
1006 | ||
1007 | for (dev = devices.dev; dev; dev = dev->next) | |
1008 | reset_device(dev); | |
6e5aa7ef | 1009 | |
659a0e66 RR |
1010 | /* If we saved off the original terminal settings, restore them now. */ |
1011 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
1012 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
1013 | } | |
6e5aa7ef | 1014 | |
659a0e66 RR |
1015 | /* When the Guest tells us they updated the status field, we handle it. */ |
1016 | static void update_device_status(struct device *dev) | |
1017 | { | |
1018 | /* A zero status is a reset, otherwise it's a set of flags. */ | |
1019 | if (dev->desc->status == 0) | |
1020 | reset_device(dev); | |
1021 | else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
a007a751 | 1022 | warnx("Device %s configuration FAILED", dev->name); |
659a0e66 RR |
1023 | if (dev->running) |
1024 | reset_device(dev); | |
a007a751 | 1025 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { |
659a0e66 RR |
1026 | if (!dev->running) |
1027 | start_device(dev); | |
6e5aa7ef RR |
1028 | } |
1029 | } | |
1030 | ||
17cbca2b | 1031 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
56739c80 | 1032 | static void handle_output(unsigned long addr) |
8ca47e00 RR |
1033 | { |
1034 | struct device *i; | |
17cbca2b | 1035 | |
659a0e66 | 1036 | /* Check each device. */ |
17cbca2b | 1037 | for (i = devices.dev; i; i = i->next) { |
659a0e66 RR |
1038 | struct virtqueue *vq; |
1039 | ||
a007a751 | 1040 | /* Notifications to device descriptors update device status. */ |
6e5aa7ef | 1041 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1042 | update_device_status(i); |
6e5aa7ef RR |
1043 | return; |
1044 | } | |
1045 | ||
659a0e66 | 1046 | /* Devices *can* be used before status is set to DRIVER_OK. */ |
17cbca2b | 1047 | for (vq = i->vq; vq; vq = vq->next) { |
659a0e66 | 1048 | if (addr != vq->config.pfn*getpagesize()) |
6e5aa7ef | 1049 | continue; |
659a0e66 RR |
1050 | if (i->running) |
1051 | errx(1, "Notification on running %s", i->name); | |
1052 | start_device(i); | |
6e5aa7ef | 1053 | return; |
8ca47e00 RR |
1054 | } |
1055 | } | |
dde79789 | 1056 | |
2e04ef76 RR |
1057 | /* |
1058 | * Early console write is done using notify on a nul-terminated string | |
1059 | * in Guest memory. It's also great for hacking debugging messages | |
1060 | * into a Guest. | |
1061 | */ | |
17cbca2b RR |
1062 | if (addr >= guest_limit) |
1063 | errx(1, "Bad NOTIFY %#lx", addr); | |
1064 | ||
1065 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1066 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1067 | } |
1068 | ||
dde79789 RR |
1069 | /*L:190 |
1070 | * Device Setup | |
1071 | * | |
1072 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1073 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1074 | * routines to allocate and manage them. |
1075 | */ | |
8ca47e00 | 1076 | |
2e04ef76 RR |
1077 | /* |
1078 | * The layout of the device page is a "struct lguest_device_desc" followed by a | |
a586d4f6 RR |
1079 | * number of virtqueue descriptors, then two sets of feature bits, then an |
1080 | * array of configuration bytes. This routine returns the configuration | |
2e04ef76 RR |
1081 | * pointer. |
1082 | */ | |
a586d4f6 RR |
1083 | static u8 *device_config(const struct device *dev) |
1084 | { | |
1085 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1086 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1087 | + dev->feature_len * 2; | |
17cbca2b RR |
1088 | } |
1089 | ||
2e04ef76 RR |
1090 | /* |
1091 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
a586d4f6 | 1092 | * table page just above the Guest's normal memory. It returns a pointer to |
2e04ef76 RR |
1093 | * that descriptor. |
1094 | */ | |
a586d4f6 | 1095 | static struct lguest_device_desc *new_dev_desc(u16 type) |
17cbca2b | 1096 | { |
a586d4f6 RR |
1097 | struct lguest_device_desc d = { .type = type }; |
1098 | void *p; | |
17cbca2b | 1099 | |
a586d4f6 RR |
1100 | /* Figure out where the next device config is, based on the last one. */ |
1101 | if (devices.lastdev) | |
1102 | p = device_config(devices.lastdev) | |
1103 | + devices.lastdev->desc->config_len; | |
1104 | else | |
1105 | p = devices.descpage; | |
17cbca2b | 1106 | |
a586d4f6 RR |
1107 | /* We only have one page for all the descriptors. */ |
1108 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1109 | errx(1, "Too many devices"); | |
17cbca2b | 1110 | |
a586d4f6 RR |
1111 | /* p might not be aligned, so we memcpy in. */ |
1112 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1113 | } |
1114 | ||
2e04ef76 RR |
1115 | /* |
1116 | * Each device descriptor is followed by the description of its virtqueues. We | |
1117 | * specify how many descriptors the virtqueue is to have. | |
1118 | */ | |
17cbca2b | 1119 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
659a0e66 | 1120 | void (*service)(struct virtqueue *)) |
17cbca2b RR |
1121 | { |
1122 | unsigned int pages; | |
1123 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1124 | void *p; | |
1125 | ||
a6bd8e13 | 1126 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1127 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 1128 | / getpagesize(); |
17cbca2b RR |
1129 | p = get_pages(pages); |
1130 | ||
d1c856e0 RR |
1131 | /* Initialize the virtqueue */ |
1132 | vq->next = NULL; | |
1133 | vq->last_avail_idx = 0; | |
1134 | vq->dev = dev; | |
659a0e66 RR |
1135 | vq->service = service; |
1136 | vq->thread = (pid_t)-1; | |
d1c856e0 | 1137 | |
17cbca2b RR |
1138 | /* Initialize the configuration. */ |
1139 | vq->config.num = num_descs; | |
1140 | vq->config.irq = devices.next_irq++; | |
1141 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1142 | ||
1143 | /* Initialize the vring. */ | |
2966af73 | 1144 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1145 | |
2e04ef76 RR |
1146 | /* |
1147 | * Append virtqueue to this device's descriptor. We use | |
a586d4f6 RR |
1148 | * device_config() to get the end of the device's current virtqueues; |
1149 | * we check that we haven't added any config or feature information | |
2e04ef76 RR |
1150 | * yet, otherwise we'd be overwriting them. |
1151 | */ | |
a586d4f6 RR |
1152 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); |
1153 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1154 | dev->num_vq++; |
a586d4f6 RR |
1155 | dev->desc->num_vq++; |
1156 | ||
1157 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b | 1158 | |
2e04ef76 RR |
1159 | /* |
1160 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1161 | * second. | |
1162 | */ | |
17cbca2b RR |
1163 | for (i = &dev->vq; *i; i = &(*i)->next); |
1164 | *i = vq; | |
8ca47e00 RR |
1165 | } |
1166 | ||
2e04ef76 RR |
1167 | /* |
1168 | * The first half of the feature bitmask is for us to advertise features. The | |
1169 | * second half is for the Guest to accept features. | |
1170 | */ | |
a586d4f6 RR |
1171 | static void add_feature(struct device *dev, unsigned bit) |
1172 | { | |
6e5aa7ef | 1173 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1174 | |
1175 | /* We can't extend the feature bits once we've added config bytes */ | |
1176 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1177 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1178 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1179 | } |
1180 | ||
a586d4f6 RR |
1181 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1182 | } | |
1183 | ||
2e04ef76 RR |
1184 | /* |
1185 | * This routine sets the configuration fields for an existing device's | |
a586d4f6 | 1186 | * descriptor. It only works for the last device, but that's OK because that's |
2e04ef76 RR |
1187 | * how we use it. |
1188 | */ | |
a586d4f6 RR |
1189 | static void set_config(struct device *dev, unsigned len, const void *conf) |
1190 | { | |
1191 | /* Check we haven't overflowed our single page. */ | |
1192 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1193 | errx(1, "Too many devices"); | |
1194 | ||
1195 | /* Copy in the config information, and store the length. */ | |
1196 | memcpy(device_config(dev), conf, len); | |
1197 | dev->desc->config_len = len; | |
8ef562d1 RR |
1198 | |
1199 | /* Size must fit in config_len field (8 bits)! */ | |
1200 | assert(dev->desc->config_len == len); | |
a586d4f6 RR |
1201 | } |
1202 | ||
2e04ef76 RR |
1203 | /* |
1204 | * This routine does all the creation and setup of a new device, including | |
a6bd8e13 RR |
1205 | * calling new_dev_desc() to allocate the descriptor and device memory. |
1206 | * | |
2e04ef76 RR |
1207 | * See what I mean about userspace being boring? |
1208 | */ | |
659a0e66 | 1209 | static struct device *new_device(const char *name, u16 type) |
8ca47e00 RR |
1210 | { |
1211 | struct device *dev = malloc(sizeof(*dev)); | |
1212 | ||
dde79789 | 1213 | /* Now we populate the fields one at a time. */ |
17cbca2b | 1214 | dev->desc = new_dev_desc(type); |
17cbca2b | 1215 | dev->name = name; |
d1c856e0 | 1216 | dev->vq = NULL; |
713b15b3 RR |
1217 | dev->feature_len = 0; |
1218 | dev->num_vq = 0; | |
659a0e66 | 1219 | dev->running = false; |
a586d4f6 | 1220 | |
2e04ef76 RR |
1221 | /* |
1222 | * Append to device list. Prepending to a single-linked list is | |
a586d4f6 RR |
1223 | * easier, but the user expects the devices to be arranged on the bus |
1224 | * in command-line order. The first network device on the command line | |
2e04ef76 RR |
1225 | * is eth0, the first block device /dev/vda, etc. |
1226 | */ | |
a586d4f6 RR |
1227 | if (devices.lastdev) |
1228 | devices.lastdev->next = dev; | |
1229 | else | |
1230 | devices.dev = dev; | |
1231 | devices.lastdev = dev; | |
1232 | ||
8ca47e00 RR |
1233 | return dev; |
1234 | } | |
1235 | ||
2e04ef76 RR |
1236 | /* |
1237 | * Our first setup routine is the console. It's a fairly simple device, but | |
1238 | * UNIX tty handling makes it uglier than it could be. | |
1239 | */ | |
17cbca2b | 1240 | static void setup_console(void) |
8ca47e00 RR |
1241 | { |
1242 | struct device *dev; | |
1243 | ||
dde79789 | 1244 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1245 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1246 | struct termios term = orig_term; | |
2e04ef76 RR |
1247 | /* |
1248 | * Then we turn off echo, line buffering and ^C etc: We want a | |
1249 | * raw input stream to the Guest. | |
1250 | */ | |
8ca47e00 RR |
1251 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1252 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
1253 | } |
1254 | ||
659a0e66 RR |
1255 | dev = new_device("console", VIRTIO_ID_CONSOLE); |
1256 | ||
dde79789 | 1257 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1258 | dev->priv = malloc(sizeof(struct console_abort)); |
1259 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1260 | |
2e04ef76 RR |
1261 | /* |
1262 | * The console needs two virtqueues: the input then the output. When | |
56ae43df RR |
1263 | * they put something the input queue, we make sure we're listening to |
1264 | * stdin. When they put something in the output queue, we write it to | |
2e04ef76 RR |
1265 | * stdout. |
1266 | */ | |
659a0e66 RR |
1267 | add_virtqueue(dev, VIRTQUEUE_NUM, console_input); |
1268 | add_virtqueue(dev, VIRTQUEUE_NUM, console_output); | |
17cbca2b | 1269 | |
659a0e66 | 1270 | verbose("device %u: console\n", ++devices.device_num); |
8ca47e00 | 1271 | } |
17cbca2b | 1272 | /*:*/ |
8ca47e00 | 1273 | |
2e04ef76 RR |
1274 | /*M:010 |
1275 | * Inter-guest networking is an interesting area. Simplest is to have a | |
17cbca2b RR |
1276 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
1277 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1278 | * |
17cbca2b RR |
1279 | * More sopisticated is to use one of the tools developed for project like UML |
1280 | * to do networking. | |
dde79789 | 1281 | * |
17cbca2b RR |
1282 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1283 | * completely generic ("here's my vring, attach to your vring") and would work | |
1284 | * for any traffic. Of course, namespace and permissions issues need to be | |
1285 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1286 | * multiple inter-guest channels behind one interface, although it would | |
1287 | * require some manner of hotplugging new virtio channels. | |
1288 | * | |
2e04ef76 RR |
1289 | * Finally, we could implement a virtio network switch in the kernel. |
1290 | :*/ | |
8ca47e00 RR |
1291 | |
1292 | static u32 str2ip(const char *ipaddr) | |
1293 | { | |
dec6a2be | 1294 | unsigned int b[4]; |
8ca47e00 | 1295 | |
dec6a2be MM |
1296 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1297 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1298 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1299 | } | |
1300 | ||
1301 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1302 | { | |
1303 | unsigned int m[6]; | |
1304 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1305 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1306 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1307 | mac[0] = m[0]; | |
1308 | mac[1] = m[1]; | |
1309 | mac[2] = m[2]; | |
1310 | mac[3] = m[3]; | |
1311 | mac[4] = m[4]; | |
1312 | mac[5] = m[5]; | |
8ca47e00 RR |
1313 | } |
1314 | ||
2e04ef76 RR |
1315 | /* |
1316 | * This code is "adapted" from libbridge: it attaches the Host end of the | |
dde79789 RR |
1317 | * network device to the bridge device specified by the command line. |
1318 | * | |
1319 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
2e04ef76 RR |
1320 | * dislike bridging), and I just try not to break it. |
1321 | */ | |
8ca47e00 RR |
1322 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1323 | { | |
1324 | int ifidx; | |
1325 | struct ifreq ifr; | |
1326 | ||
1327 | if (!*br_name) | |
1328 | errx(1, "must specify bridge name"); | |
1329 | ||
1330 | ifidx = if_nametoindex(if_name); | |
1331 | if (!ifidx) | |
1332 | errx(1, "interface %s does not exist!", if_name); | |
1333 | ||
1334 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1335 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1336 | ifr.ifr_ifindex = ifidx; |
1337 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1338 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1339 | } | |
1340 | ||
2e04ef76 RR |
1341 | /* |
1342 | * This sets up the Host end of the network device with an IP address, brings | |
dde79789 | 1343 | * it up so packets will flow, the copies the MAC address into the hwaddr |
2e04ef76 RR |
1344 | * pointer. |
1345 | */ | |
dec6a2be | 1346 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1347 | { |
1348 | struct ifreq ifr; | |
1349 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1350 | ||
1351 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1352 | strcpy(ifr.ifr_name, tapif); |
1353 | ||
1354 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
8ca47e00 RR |
1355 | sin->sin_family = AF_INET; |
1356 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1357 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
dec6a2be | 1358 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1359 | ifr.ifr_flags = IFF_UP; |
1360 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1361 | err(1, "Bringing interface %s up", tapif); |
1362 | } | |
1363 | ||
dec6a2be | 1364 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1365 | { |
8ca47e00 | 1366 | struct ifreq ifr; |
dec6a2be MM |
1367 | int netfd; |
1368 | ||
1369 | /* Start with this zeroed. Messy but sure. */ | |
1370 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1371 | |
2e04ef76 RR |
1372 | /* |
1373 | * We open the /dev/net/tun device and tell it we want a tap device. A | |
dde79789 RR |
1374 | * tap device is like a tun device, only somehow different. To tell |
1375 | * the truth, I completely blundered my way through this code, but it | |
2e04ef76 RR |
1376 | * works now! |
1377 | */ | |
8ca47e00 | 1378 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1379 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1380 | strcpy(ifr.ifr_name, "tap%d"); |
1381 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1382 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1383 | |
398f187d RR |
1384 | if (ioctl(netfd, TUNSETOFFLOAD, |
1385 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1386 | err(1, "Could not set features for tun device"); | |
1387 | ||
2e04ef76 RR |
1388 | /* |
1389 | * We don't need checksums calculated for packets coming in this | |
1390 | * device: trust us! | |
1391 | */ | |
8ca47e00 RR |
1392 | ioctl(netfd, TUNSETNOCSUM, 1); |
1393 | ||
dec6a2be MM |
1394 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1395 | return netfd; | |
1396 | } | |
1397 | ||
2e04ef76 RR |
1398 | /*L:195 |
1399 | * Our network is a Host<->Guest network. This can either use bridging or | |
dec6a2be MM |
1400 | * routing, but the principle is the same: it uses the "tun" device to inject |
1401 | * packets into the Host as if they came in from a normal network card. We | |
2e04ef76 RR |
1402 | * just shunt packets between the Guest and the tun device. |
1403 | */ | |
dec6a2be MM |
1404 | static void setup_tun_net(char *arg) |
1405 | { | |
1406 | struct device *dev; | |
659a0e66 RR |
1407 | struct net_info *net_info = malloc(sizeof(*net_info)); |
1408 | int ipfd; | |
dec6a2be MM |
1409 | u32 ip = INADDR_ANY; |
1410 | bool bridging = false; | |
1411 | char tapif[IFNAMSIZ], *p; | |
1412 | struct virtio_net_config conf; | |
1413 | ||
659a0e66 | 1414 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 1415 | |
17cbca2b | 1416 | /* First we create a new network device. */ |
659a0e66 RR |
1417 | dev = new_device("net", VIRTIO_ID_NET); |
1418 | dev->priv = net_info; | |
dde79789 | 1419 | |
2e04ef76 | 1420 | /* Network devices need a recv and a send queue, just like console. */ |
659a0e66 RR |
1421 | add_virtqueue(dev, VIRTQUEUE_NUM, net_input); |
1422 | add_virtqueue(dev, VIRTQUEUE_NUM, net_output); | |
8ca47e00 | 1423 | |
2e04ef76 RR |
1424 | /* |
1425 | * We need a socket to perform the magic network ioctls to bring up the | |
1426 | * tap interface, connect to the bridge etc. Any socket will do! | |
1427 | */ | |
8ca47e00 RR |
1428 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1429 | if (ipfd < 0) | |
1430 | err(1, "opening IP socket"); | |
1431 | ||
dde79789 | 1432 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1433 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1434 | arg += strlen(BRIDGE_PFX); |
1435 | bridging = true; | |
1436 | } | |
1437 | ||
1438 | /* A mac address may follow the bridge name or IP address */ | |
1439 | p = strchr(arg, ':'); | |
1440 | if (p) { | |
1441 | str2mac(p+1, conf.mac); | |
40c42076 | 1442 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1443 | *p = '\0'; |
dec6a2be MM |
1444 | } |
1445 | ||
1446 | /* arg is now either an IP address or a bridge name */ | |
1447 | if (bridging) | |
1448 | add_to_bridge(ipfd, tapif, arg); | |
1449 | else | |
8ca47e00 RR |
1450 | ip = str2ip(arg); |
1451 | ||
dec6a2be MM |
1452 | /* Set up the tun device. */ |
1453 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1454 | |
20887611 | 1455 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1456 | /* Expect Guest to handle everything except UFO */ |
1457 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1458 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1459 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1460 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1461 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1462 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1463 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1464 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e MM |
1465 | /* We handle indirect ring entries */ |
1466 | add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); | |
a586d4f6 | 1467 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1468 | |
a586d4f6 | 1469 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1470 | close(ipfd); |
1471 | ||
dec6a2be MM |
1472 | devices.device_num++; |
1473 | ||
1474 | if (bridging) | |
1475 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1476 | devices.device_num, tapif, arg); | |
1477 | else | |
1478 | verbose("device %u: tun %s: %s\n", | |
1479 | devices.device_num, tapif, arg); | |
8ca47e00 | 1480 | } |
17cbca2b | 1481 | |
2e04ef76 RR |
1482 | /* |
1483 | * Our block (disk) device should be really simple: the Guest asks for a block | |
e1e72965 RR |
1484 | * number and we read or write that position in the file. Unfortunately, that |
1485 | * was amazingly slow: the Guest waits until the read is finished before | |
1486 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1487 | * |
e1e72965 RR |
1488 | * We could use async I/O, except it's reputed to suck so hard that characters |
1489 | * actually go missing from your code when you try to use it. | |
17cbca2b | 1490 | * |
2e04ef76 RR |
1491 | * So this was one reason why lguest now does all virtqueue servicing in |
1492 | * separate threads: it's more efficient and more like a real device. | |
1493 | */ | |
17cbca2b | 1494 | |
e1e72965 | 1495 | /* This hangs off device->priv. */ |
17cbca2b RR |
1496 | struct vblk_info |
1497 | { | |
1498 | /* The size of the file. */ | |
1499 | off64_t len; | |
1500 | ||
1501 | /* The file descriptor for the file. */ | |
1502 | int fd; | |
1503 | ||
1504 | /* IO thread listens on this file descriptor [0]. */ | |
1505 | int workpipe[2]; | |
1506 | ||
1507 | /* IO thread writes to this file descriptor to mark it done, then | |
1508 | * Launcher triggers interrupt to Guest. */ | |
1509 | int done_fd; | |
1510 | }; | |
1511 | ||
e1e72965 RR |
1512 | /*L:210 |
1513 | * The Disk | |
1514 | * | |
1515 | * Remember that the block device is handled by a separate I/O thread. We head | |
1516 | * straight into the core of that thread here: | |
1517 | */ | |
659a0e66 | 1518 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 1519 | { |
659a0e66 | 1520 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b RR |
1521 | unsigned int head, out_num, in_num, wlen; |
1522 | int ret; | |
cb38fa23 | 1523 | u8 *in; |
17cbca2b | 1524 | struct virtio_blk_outhdr *out; |
659a0e66 | 1525 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
1526 | off64_t off; |
1527 | ||
659a0e66 RR |
1528 | /* Get the next request. */ |
1529 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); | |
17cbca2b | 1530 | |
2e04ef76 RR |
1531 | /* |
1532 | * Every block request should contain at least one output buffer | |
e1e72965 | 1533 | * (detailing the location on disk and the type of request) and one |
2e04ef76 RR |
1534 | * input buffer (to hold the result). |
1535 | */ | |
17cbca2b RR |
1536 | if (out_num == 0 || in_num == 0) |
1537 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1538 | head, out_num, in_num); | |
1539 | ||
1540 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1541 | in = convert(&iov[out_num+in_num-1], u8); |
17cbca2b RR |
1542 | off = out->sector * 512; |
1543 | ||
2e04ef76 RR |
1544 | /* |
1545 | * The block device implements "barriers", where the Guest indicates | |
e1e72965 RR |
1546 | * that it wants all previous writes to occur before this write. We |
1547 | * don't have a way of asking our kernel to do a barrier, so we just | |
2e04ef76 RR |
1548 | * synchronize all the data in the file. Pretty poor, no? |
1549 | */ | |
17cbca2b RR |
1550 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1551 | fdatasync(vblk->fd); | |
1552 | ||
2e04ef76 RR |
1553 | /* |
1554 | * In general the virtio block driver is allowed to try SCSI commands. | |
1555 | * It'd be nice if we supported eject, for example, but we don't. | |
1556 | */ | |
17cbca2b RR |
1557 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1558 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1559 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1560 | wlen = sizeof(*in); |
17cbca2b | 1561 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
2e04ef76 RR |
1562 | /* |
1563 | * Write | |
1564 | * | |
1565 | * Move to the right location in the block file. This can fail | |
1566 | * if they try to write past end. | |
1567 | */ | |
17cbca2b RR |
1568 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
1569 | err(1, "Bad seek to sector %llu", out->sector); | |
1570 | ||
1571 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1572 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1573 | ||
2e04ef76 RR |
1574 | /* |
1575 | * Grr... Now we know how long the descriptor they sent was, we | |
17cbca2b | 1576 | * make sure they didn't try to write over the end of the block |
2e04ef76 RR |
1577 | * file (possibly extending it). |
1578 | */ | |
17cbca2b RR |
1579 | if (ret > 0 && off + ret > vblk->len) { |
1580 | /* Trim it back to the correct length */ | |
1581 | ftruncate64(vblk->fd, vblk->len); | |
1582 | /* Die, bad Guest, die. */ | |
1583 | errx(1, "Write past end %llu+%u", off, ret); | |
1584 | } | |
1200e646 | 1585 | wlen = sizeof(*in); |
cb38fa23 | 1586 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b | 1587 | } else { |
2e04ef76 RR |
1588 | /* |
1589 | * Read | |
1590 | * | |
1591 | * Move to the right location in the block file. This can fail | |
1592 | * if they try to read past end. | |
1593 | */ | |
17cbca2b RR |
1594 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
1595 | err(1, "Bad seek to sector %llu", out->sector); | |
1596 | ||
1597 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1598 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1599 | if (ret >= 0) { | |
1200e646 | 1600 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1601 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1602 | } else { |
1200e646 | 1603 | wlen = sizeof(*in); |
cb38fa23 | 1604 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1605 | } |
1606 | } | |
1607 | ||
2e04ef76 RR |
1608 | /* |
1609 | * OK, so we noted that it was pretty poor to use an fdatasync as a | |
d1881d31 RR |
1610 | * barrier. But Christoph Hellwig points out that we need a sync |
1611 | * *afterwards* as well: "Barriers specify no reordering to the front | |
2e04ef76 RR |
1612 | * or the back." And Jens Axboe confirmed it, so here we are: |
1613 | */ | |
d1881d31 RR |
1614 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1615 | fdatasync(vblk->fd); | |
1616 | ||
38bc2b8c | 1617 | add_used(vq, head, wlen); |
17cbca2b RR |
1618 | } |
1619 | ||
e1e72965 | 1620 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1621 | static void setup_block_file(const char *filename) |
1622 | { | |
17cbca2b RR |
1623 | struct device *dev; |
1624 | struct vblk_info *vblk; | |
a586d4f6 | 1625 | struct virtio_blk_config conf; |
17cbca2b | 1626 | |
2e04ef76 | 1627 | /* Creat the device. */ |
659a0e66 | 1628 | dev = new_device("block", VIRTIO_ID_BLOCK); |
17cbca2b | 1629 | |
e1e72965 | 1630 | /* The device has one virtqueue, where the Guest places requests. */ |
659a0e66 | 1631 | add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); |
17cbca2b RR |
1632 | |
1633 | /* Allocate the room for our own bookkeeping */ | |
1634 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1635 | ||
1636 | /* First we open the file and store the length. */ | |
1637 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1638 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1639 | ||
a586d4f6 RR |
1640 | /* We support barriers. */ |
1641 | add_feature(dev, VIRTIO_BLK_F_BARRIER); | |
1642 | ||
17cbca2b | 1643 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1644 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b | 1645 | |
2e04ef76 RR |
1646 | /* |
1647 | * Tell Guest not to put in too many descriptors at once: two are used | |
1648 | * for the in and out elements. | |
1649 | */ | |
a586d4f6 RR |
1650 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1651 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1652 | ||
8ef562d1 RR |
1653 | /* Don't try to put whole struct: we have 8 bit limit. */ |
1654 | set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf); | |
17cbca2b | 1655 | |
17cbca2b | 1656 | verbose("device %u: virtblock %llu sectors\n", |
659a0e66 | 1657 | ++devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1658 | } |
28fd6d7f | 1659 | |
2e04ef76 RR |
1660 | /*L:211 |
1661 | * Our random number generator device reads from /dev/random into the Guest's | |
28fd6d7f RR |
1662 | * input buffers. The usual case is that the Guest doesn't want random numbers |
1663 | * and so has no buffers although /dev/random is still readable, whereas | |
1664 | * console is the reverse. | |
1665 | * | |
2e04ef76 RR |
1666 | * The same logic applies, however. |
1667 | */ | |
1668 | struct rng_info { | |
1669 | int rfd; | |
1670 | }; | |
1671 | ||
659a0e66 | 1672 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
1673 | { |
1674 | int len; | |
1675 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
1676 | struct rng_info *rng_info = vq->dev->priv; |
1677 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
1678 | |
1679 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 1680 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f RR |
1681 | if (out_num) |
1682 | errx(1, "Output buffers in rng?"); | |
1683 | ||
2e04ef76 RR |
1684 | /* |
1685 | * This is why we convert to iovecs: the readv() call uses them, and so | |
28fd6d7f | 1686 | * it reads straight into the Guest's buffer. We loop to make sure we |
2e04ef76 RR |
1687 | * fill it. |
1688 | */ | |
28fd6d7f | 1689 | while (!iov_empty(iov, in_num)) { |
659a0e66 | 1690 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f RR |
1691 | if (len <= 0) |
1692 | err(1, "Read from /dev/random gave %i", len); | |
1693 | iov_consume(iov, in_num, len); | |
1694 | totlen += len; | |
1695 | } | |
1696 | ||
1697 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 1698 | add_used(vq, head, totlen); |
28fd6d7f RR |
1699 | } |
1700 | ||
2e04ef76 RR |
1701 | /*L:199 |
1702 | * This creates a "hardware" random number device for the Guest. | |
1703 | */ | |
28fd6d7f RR |
1704 | static void setup_rng(void) |
1705 | { | |
1706 | struct device *dev; | |
659a0e66 | 1707 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 1708 | |
2e04ef76 | 1709 | /* Our device's privat info simply contains the /dev/random fd. */ |
659a0e66 | 1710 | rng_info->rfd = open_or_die("/dev/random", O_RDONLY); |
28fd6d7f | 1711 | |
2e04ef76 | 1712 | /* Create the new device. */ |
659a0e66 RR |
1713 | dev = new_device("rng", VIRTIO_ID_RNG); |
1714 | dev->priv = rng_info; | |
28fd6d7f RR |
1715 | |
1716 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
659a0e66 | 1717 | add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); |
28fd6d7f RR |
1718 | |
1719 | verbose("device %u: rng\n", devices.device_num++); | |
1720 | } | |
a6bd8e13 | 1721 | /* That's the end of device setup. */ |
ec04b13f | 1722 | |
a6bd8e13 | 1723 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1724 | static void __attribute__((noreturn)) restart_guest(void) |
1725 | { | |
1726 | unsigned int i; | |
1727 | ||
2e04ef76 RR |
1728 | /* |
1729 | * Since we don't track all open fds, we simply close everything beyond | |
1730 | * stderr. | |
1731 | */ | |
ec04b13f BR |
1732 | for (i = 3; i < FD_SETSIZE; i++) |
1733 | close(i); | |
8c79873d | 1734 | |
659a0e66 RR |
1735 | /* Reset all the devices (kills all threads). */ |
1736 | cleanup_devices(); | |
1737 | ||
ec04b13f BR |
1738 | execv(main_args[0], main_args); |
1739 | err(1, "Could not exec %s", main_args[0]); | |
1740 | } | |
8ca47e00 | 1741 | |
2e04ef76 RR |
1742 | /*L:220 |
1743 | * Finally we reach the core of the Launcher which runs the Guest, serves | |
1744 | * its input and output, and finally, lays it to rest. | |
1745 | */ | |
56739c80 | 1746 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
1747 | { |
1748 | for (;;) { | |
17cbca2b | 1749 | unsigned long notify_addr; |
8ca47e00 RR |
1750 | int readval; |
1751 | ||
1752 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1753 | readval = pread(lguest_fd, ¬ify_addr, |
1754 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1755 | |
17cbca2b RR |
1756 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1757 | if (readval == sizeof(notify_addr)) { | |
1758 | verbose("Notify on address %#lx\n", notify_addr); | |
56739c80 | 1759 | handle_output(notify_addr); |
dde79789 | 1760 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1761 | } else if (errno == ENOENT) { |
1762 | char reason[1024] = { 0 }; | |
e3283fa0 | 1763 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1764 | errx(1, "%s", reason); |
ec04b13f BR |
1765 | /* ERESTART means that we need to reboot the guest */ |
1766 | } else if (errno == ERESTART) { | |
1767 | restart_guest(); | |
659a0e66 RR |
1768 | /* Anything else means a bug or incompatible change. */ |
1769 | } else | |
8ca47e00 | 1770 | err(1, "Running guest failed"); |
8ca47e00 RR |
1771 | } |
1772 | } | |
a6bd8e13 | 1773 | /*L:240 |
e1e72965 RR |
1774 | * This is the end of the Launcher. The good news: we are over halfway |
1775 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1776 | * of us. | |
dde79789 | 1777 | * |
e1e72965 RR |
1778 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1779 | * "make Host". | |
2e04ef76 | 1780 | :*/ |
8ca47e00 RR |
1781 | |
1782 | static struct option opts[] = { | |
1783 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1784 | { "tunnet", 1, NULL, 't' }, |
1785 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1786 | { "rng", 0, NULL, 'r' }, |
8ca47e00 RR |
1787 | { "initrd", 1, NULL, 'i' }, |
1788 | { NULL }, | |
1789 | }; | |
1790 | static void usage(void) | |
1791 | { | |
1792 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1793 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1794 | "|--block=<filename>|--initrd=<filename>]...\n" |
1795 | "<mem-in-mb> vmlinux [args...]"); | |
1796 | } | |
1797 | ||
3c6b5bfa | 1798 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1799 | int main(int argc, char *argv[]) |
1800 | { | |
2e04ef76 | 1801 | /* Memory, code startpoint and size of the (optional) initrd. */ |
58a24566 | 1802 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
1803 | /* Two temporaries. */ |
1804 | int i, c; | |
3c6b5bfa | 1805 | /* The boot information for the Guest. */ |
43d33b21 | 1806 | struct boot_params *boot; |
dde79789 | 1807 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1808 | const char *initrd_name = NULL; |
1809 | ||
ec04b13f BR |
1810 | /* Save the args: we "reboot" by execing ourselves again. */ |
1811 | main_args = argv; | |
ec04b13f | 1812 | |
2e04ef76 RR |
1813 | /* |
1814 | * First we initialize the device list. We keep a pointer to the last | |
659a0e66 | 1815 | * device, and the next interrupt number to use for devices (1: |
2e04ef76 RR |
1816 | * remember that 0 is used by the timer). |
1817 | */ | |
a586d4f6 | 1818 | devices.lastdev = NULL; |
17cbca2b | 1819 | devices.next_irq = 1; |
8ca47e00 | 1820 | |
e3283fa0 | 1821 | cpu_id = 0; |
2e04ef76 RR |
1822 | /* |
1823 | * We need to know how much memory so we can set up the device | |
dde79789 RR |
1824 | * descriptor and memory pages for the devices as we parse the command |
1825 | * line. So we quickly look through the arguments to find the amount | |
2e04ef76 RR |
1826 | * of memory now. |
1827 | */ | |
6570c459 RR |
1828 | for (i = 1; i < argc; i++) { |
1829 | if (argv[i][0] != '-') { | |
3c6b5bfa | 1830 | mem = atoi(argv[i]) * 1024 * 1024; |
2e04ef76 RR |
1831 | /* |
1832 | * We start by mapping anonymous pages over all of | |
3c6b5bfa RR |
1833 | * guest-physical memory range. This fills it with 0, |
1834 | * and ensures that the Guest won't be killed when it | |
2e04ef76 RR |
1835 | * tries to access it. |
1836 | */ | |
3c6b5bfa RR |
1837 | guest_base = map_zeroed_pages(mem / getpagesize() |
1838 | + DEVICE_PAGES); | |
1839 | guest_limit = mem; | |
1840 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1841 | devices.descpage = get_pages(1); |
6570c459 RR |
1842 | break; |
1843 | } | |
1844 | } | |
dde79789 RR |
1845 | |
1846 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1847 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1848 | switch (c) { | |
1849 | case 'v': | |
1850 | verbose = true; | |
1851 | break; | |
8ca47e00 | 1852 | case 't': |
17cbca2b | 1853 | setup_tun_net(optarg); |
8ca47e00 RR |
1854 | break; |
1855 | case 'b': | |
17cbca2b | 1856 | setup_block_file(optarg); |
8ca47e00 | 1857 | break; |
28fd6d7f RR |
1858 | case 'r': |
1859 | setup_rng(); | |
1860 | break; | |
8ca47e00 RR |
1861 | case 'i': |
1862 | initrd_name = optarg; | |
1863 | break; | |
1864 | default: | |
1865 | warnx("Unknown argument %s", argv[optind]); | |
1866 | usage(); | |
1867 | } | |
1868 | } | |
2e04ef76 RR |
1869 | /* |
1870 | * After the other arguments we expect memory and kernel image name, | |
1871 | * followed by command line arguments for the kernel. | |
1872 | */ | |
8ca47e00 RR |
1873 | if (optind + 2 > argc) |
1874 | usage(); | |
1875 | ||
3c6b5bfa RR |
1876 | verbose("Guest base is at %p\n", guest_base); |
1877 | ||
dde79789 | 1878 | /* We always have a console device */ |
17cbca2b | 1879 | setup_console(); |
8ca47e00 | 1880 | |
8ca47e00 | 1881 | /* Now we load the kernel */ |
47436aa4 | 1882 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1883 | |
3c6b5bfa RR |
1884 | /* Boot information is stashed at physical address 0 */ |
1885 | boot = from_guest_phys(0); | |
1886 | ||
dde79789 | 1887 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1888 | if (initrd_name) { |
1889 | initrd_size = load_initrd(initrd_name, mem); | |
2e04ef76 RR |
1890 | /* |
1891 | * These are the location in the Linux boot header where the | |
1892 | * start and size of the initrd are expected to be found. | |
1893 | */ | |
43d33b21 RR |
1894 | boot->hdr.ramdisk_image = mem - initrd_size; |
1895 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 1896 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 1897 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
1898 | } |
1899 | ||
2e04ef76 RR |
1900 | /* |
1901 | * The Linux boot header contains an "E820" memory map: ours is a | |
1902 | * simple, single region. | |
1903 | */ | |
43d33b21 RR |
1904 | boot->e820_entries = 1; |
1905 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
2e04ef76 RR |
1906 | /* |
1907 | * The boot header contains a command line pointer: we put the command | |
1908 | * line after the boot header. | |
1909 | */ | |
43d33b21 | 1910 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
e1e72965 | 1911 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 1912 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 1913 | |
814a0e5c | 1914 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 1915 | boot->hdr.version = 0x207; |
814a0e5c RR |
1916 | |
1917 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 1918 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 1919 | |
43d33b21 RR |
1920 | /* Tell the entry path not to try to reload segment registers. */ |
1921 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 1922 | |
2e04ef76 RR |
1923 | /* |
1924 | * We tell the kernel to initialize the Guest: this returns the open | |
1925 | * /dev/lguest file descriptor. | |
1926 | */ | |
56739c80 | 1927 | tell_kernel(start); |
dde79789 | 1928 | |
659a0e66 RR |
1929 | /* Ensure that we terminate if a child dies. */ |
1930 | signal(SIGCHLD, kill_launcher); | |
1931 | ||
1932 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
1933 | atexit(cleanup_devices); | |
8ca47e00 | 1934 | |
dde79789 | 1935 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 1936 | run_guest(); |
8ca47e00 | 1937 | } |
f56a384e RR |
1938 | /*:*/ |
1939 | ||
1940 | /*M:999 | |
1941 | * Mastery is done: you now know everything I do. | |
1942 | * | |
1943 | * But surely you have seen code, features and bugs in your wanderings which | |
1944 | * you now yearn to attack? That is the real game, and I look forward to you | |
1945 | * patching and forking lguest into the Your-Name-Here-visor. | |
1946 | * | |
1947 | * Farewell, and good coding! | |
1948 | * Rusty Russell. | |
1949 | */ |