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> | |
17cbca2b RR |
37 | #include <assert.h> |
38 | #include <sched.h> | |
a586d4f6 RR |
39 | #include <limits.h> |
40 | #include <stddef.h> | |
a161883a | 41 | #include <signal.h> |
8aeb36e8 PS |
42 | #include <pwd.h> |
43 | #include <grp.h> | |
44 | ||
f846619e RR |
45 | #include <linux/virtio_config.h> |
46 | #include <linux/virtio_net.h> | |
47 | #include <linux/virtio_blk.h> | |
48 | #include <linux/virtio_console.h> | |
49 | #include <linux/virtio_rng.h> | |
50 | #include <linux/virtio_ring.h> | |
51 | #include <asm/bootparam.h> | |
52 | #include "../../include/linux/lguest_launcher.h" | |
2e04ef76 | 53 | /*L:110 |
a91d74a3 | 54 | * We can ignore the 42 include files we need for this program, but I do want |
2e04ef76 | 55 | * to draw attention to the use of kernel-style types. |
db24e8c2 RR |
56 | * |
57 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
58 | * like these abbreviations, so we define them here. Note that u64 is always | |
59 | * unsigned long long, which works on all Linux systems: this means that we can | |
2e04ef76 RR |
60 | * use %llu in printf for any u64. |
61 | */ | |
db24e8c2 RR |
62 | typedef unsigned long long u64; |
63 | typedef uint32_t u32; | |
64 | typedef uint16_t u16; | |
65 | typedef uint8_t u8; | |
dde79789 | 66 | /*:*/ |
8ca47e00 RR |
67 | |
68 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
8ca47e00 RR |
69 | #define BRIDGE_PFX "bridge:" |
70 | #ifndef SIOCBRADDIF | |
71 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
72 | #endif | |
3c6b5bfa RR |
73 | /* We can have up to 256 pages for devices. */ |
74 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
75 | /* This will occupy 3 pages: it must be a power of 2. */ |
76 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 77 | |
2e04ef76 RR |
78 | /*L:120 |
79 | * verbose is both a global flag and a macro. The C preprocessor allows | |
80 | * this, and although I wouldn't recommend it, it works quite nicely here. | |
81 | */ | |
8ca47e00 RR |
82 | static bool verbose; |
83 | #define verbose(args...) \ | |
84 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
85 | /*:*/ |
86 | ||
3c6b5bfa RR |
87 | /* The pointer to the start of guest memory. */ |
88 | static void *guest_base; | |
89 | /* The maximum guest physical address allowed, and maximum possible. */ | |
90 | static unsigned long guest_limit, guest_max; | |
56739c80 RR |
91 | /* The /dev/lguest file descriptor. */ |
92 | static int lguest_fd; | |
8ca47e00 | 93 | |
e3283fa0 GOC |
94 | /* a per-cpu variable indicating whose vcpu is currently running */ |
95 | static unsigned int __thread cpu_id; | |
96 | ||
dde79789 | 97 | /* This is our list of devices. */ |
1842f23c | 98 | struct device_list { |
17cbca2b RR |
99 | /* Counter to assign interrupt numbers. */ |
100 | unsigned int next_irq; | |
101 | ||
102 | /* Counter to print out convenient device numbers. */ | |
103 | unsigned int device_num; | |
104 | ||
dde79789 | 105 | /* The descriptor page for the devices. */ |
17cbca2b RR |
106 | u8 *descpage; |
107 | ||
dde79789 | 108 | /* A single linked list of devices. */ |
8ca47e00 | 109 | struct device *dev; |
2e04ef76 | 110 | /* And a pointer to the last device for easy append. */ |
a586d4f6 | 111 | struct device *lastdev; |
8ca47e00 RR |
112 | }; |
113 | ||
17cbca2b RR |
114 | /* The list of Guest devices, based on command line arguments. */ |
115 | static struct device_list devices; | |
116 | ||
dde79789 | 117 | /* The device structure describes a single device. */ |
1842f23c | 118 | struct device { |
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 | |
ca60a42c RR |
138 | /* Does Guest want an intrrupt on empty? */ |
139 | bool irq_on_empty; | |
140 | ||
8ca47e00 RR |
141 | /* Device-specific data. */ |
142 | void *priv; | |
143 | }; | |
144 | ||
17cbca2b | 145 | /* The virtqueue structure describes a queue attached to a device. */ |
1842f23c | 146 | struct virtqueue { |
17cbca2b RR |
147 | struct virtqueue *next; |
148 | ||
149 | /* Which device owns me. */ | |
150 | struct device *dev; | |
151 | ||
152 | /* The configuration for this queue. */ | |
153 | struct lguest_vqconfig config; | |
154 | ||
155 | /* The actual ring of buffers. */ | |
156 | struct vring vring; | |
157 | ||
158 | /* Last available index we saw. */ | |
159 | u16 last_avail_idx; | |
160 | ||
95c517c0 RR |
161 | /* How many are used since we sent last irq? */ |
162 | unsigned int pending_used; | |
163 | ||
659a0e66 RR |
164 | /* Eventfd where Guest notifications arrive. */ |
165 | int eventfd; | |
20887611 | 166 | |
659a0e66 RR |
167 | /* Function for the thread which is servicing this virtqueue. */ |
168 | void (*service)(struct virtqueue *vq); | |
169 | pid_t thread; | |
17cbca2b RR |
170 | }; |
171 | ||
ec04b13f BR |
172 | /* Remember the arguments to the program so we can "reboot" */ |
173 | static char **main_args; | |
174 | ||
659a0e66 RR |
175 | /* The original tty settings to restore on exit. */ |
176 | static struct termios orig_term; | |
177 | ||
2e04ef76 RR |
178 | /* |
179 | * We have to be careful with barriers: our devices are all run in separate | |
f7027c63 | 180 | * threads and so we need to make sure that changes visible to the Guest happen |
2e04ef76 RR |
181 | * in precise order. |
182 | */ | |
f7027c63 | 183 | #define wmb() __asm__ __volatile__("" : : : "memory") |
b60da13f | 184 | #define mb() __asm__ __volatile__("" : : : "memory") |
17cbca2b | 185 | |
2e04ef76 RR |
186 | /* |
187 | * Convert an iovec element to the given type. | |
17cbca2b RR |
188 | * |
189 | * This is a fairly ugly trick: we need to know the size of the type and | |
190 | * alignment requirement to check the pointer is kosher. It's also nice to | |
191 | * have the name of the type in case we report failure. | |
192 | * | |
193 | * Typing those three things all the time is cumbersome and error prone, so we | |
2e04ef76 RR |
194 | * have a macro which sets them all up and passes to the real function. |
195 | */ | |
17cbca2b RR |
196 | #define convert(iov, type) \ |
197 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
198 | ||
199 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
200 | const char *name) | |
201 | { | |
202 | if (iov->iov_len != size) | |
203 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
204 | if ((unsigned long)iov->iov_base % align != 0) | |
205 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
206 | return iov->iov_base; | |
207 | } | |
208 | ||
b5111790 RR |
209 | /* Wrapper for the last available index. Makes it easier to change. */ |
210 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
211 | ||
2e04ef76 RR |
212 | /* |
213 | * The virtio configuration space is defined to be little-endian. x86 is | |
214 | * little-endian too, but it's nice to be explicit so we have these helpers. | |
215 | */ | |
17cbca2b RR |
216 | #define cpu_to_le16(v16) (v16) |
217 | #define cpu_to_le32(v32) (v32) | |
218 | #define cpu_to_le64(v64) (v64) | |
219 | #define le16_to_cpu(v16) (v16) | |
220 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 221 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 222 | |
28fd6d7f RR |
223 | /* Is this iovec empty? */ |
224 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
225 | { | |
226 | unsigned int i; | |
227 | ||
228 | for (i = 0; i < num_iov; i++) | |
229 | if (iov[i].iov_len) | |
230 | return false; | |
231 | return true; | |
232 | } | |
233 | ||
234 | /* Take len bytes from the front of this iovec. */ | |
235 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) | |
236 | { | |
237 | unsigned int i; | |
238 | ||
239 | for (i = 0; i < num_iov; i++) { | |
240 | unsigned int used; | |
241 | ||
242 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
243 | iov[i].iov_base += used; | |
244 | iov[i].iov_len -= used; | |
245 | len -= used; | |
246 | } | |
247 | assert(len == 0); | |
248 | } | |
249 | ||
6e5aa7ef RR |
250 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
251 | static u8 *get_feature_bits(struct device *dev) | |
252 | { | |
253 | return (u8 *)(dev->desc + 1) | |
713b15b3 | 254 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
6e5aa7ef RR |
255 | } |
256 | ||
2e04ef76 RR |
257 | /*L:100 |
258 | * The Launcher code itself takes us out into userspace, that scary place where | |
259 | * pointers run wild and free! Unfortunately, like most userspace programs, | |
260 | * it's quite boring (which is why everyone likes to hack on the kernel!). | |
261 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get | |
262 | * you through this section. Or, maybe not. | |
3c6b5bfa RR |
263 | * |
264 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
265 | * memory and stores it in "guest_base". In other words, Guest physical == | |
266 | * Launcher virtual with an offset. | |
267 | * | |
268 | * This can be tough to get your head around, but usually it just means that we | |
a33f3224 | 269 | * use these trivial conversion functions when the Guest gives us its |
2e04ef76 RR |
270 | * "physical" addresses: |
271 | */ | |
3c6b5bfa RR |
272 | static void *from_guest_phys(unsigned long addr) |
273 | { | |
274 | return guest_base + addr; | |
275 | } | |
276 | ||
277 | static unsigned long to_guest_phys(const void *addr) | |
278 | { | |
279 | return (addr - guest_base); | |
280 | } | |
281 | ||
dde79789 RR |
282 | /*L:130 |
283 | * Loading the Kernel. | |
284 | * | |
285 | * We start with couple of simple helper routines. open_or_die() avoids | |
2e04ef76 RR |
286 | * error-checking code cluttering the callers: |
287 | */ | |
8ca47e00 RR |
288 | static int open_or_die(const char *name, int flags) |
289 | { | |
290 | int fd = open(name, flags); | |
291 | if (fd < 0) | |
292 | err(1, "Failed to open %s", name); | |
293 | return fd; | |
294 | } | |
295 | ||
3c6b5bfa RR |
296 | /* map_zeroed_pages() takes a number of pages. */ |
297 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 298 | { |
3c6b5bfa RR |
299 | int fd = open_or_die("/dev/zero", O_RDONLY); |
300 | void *addr; | |
8ca47e00 | 301 | |
2e04ef76 RR |
302 | /* |
303 | * We use a private mapping (ie. if we write to the page, it will be | |
304 | * copied). | |
305 | */ | |
3c6b5bfa RR |
306 | addr = mmap(NULL, getpagesize() * num, |
307 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
308 | if (addr == MAP_FAILED) | |
af901ca1 | 309 | err(1, "Mmapping %u pages of /dev/zero", num); |
a91d74a3 RR |
310 | |
311 | /* | |
312 | * One neat mmap feature is that you can close the fd, and it | |
313 | * stays mapped. | |
314 | */ | |
34bdaab4 | 315 | close(fd); |
3c6b5bfa RR |
316 | |
317 | return addr; | |
318 | } | |
319 | ||
320 | /* Get some more pages for a device. */ | |
321 | static void *get_pages(unsigned int num) | |
322 | { | |
323 | void *addr = from_guest_phys(guest_limit); | |
324 | ||
325 | guest_limit += num * getpagesize(); | |
326 | if (guest_limit > guest_max) | |
327 | errx(1, "Not enough memory for devices"); | |
328 | return addr; | |
8ca47e00 RR |
329 | } |
330 | ||
2e04ef76 RR |
331 | /* |
332 | * This routine is used to load the kernel or initrd. It tries mmap, but if | |
6649bb7a | 333 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
2e04ef76 RR |
334 | * it falls back to reading the memory in. |
335 | */ | |
6649bb7a RM |
336 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
337 | { | |
338 | ssize_t r; | |
339 | ||
2e04ef76 RR |
340 | /* |
341 | * We map writable even though for some segments are marked read-only. | |
6649bb7a RM |
342 | * The kernel really wants to be writable: it patches its own |
343 | * instructions. | |
344 | * | |
345 | * MAP_PRIVATE means that the page won't be copied until a write is | |
346 | * done to it. This allows us to share untouched memory between | |
2e04ef76 RR |
347 | * Guests. |
348 | */ | |
6649bb7a RM |
349 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, |
350 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
351 | return; | |
352 | ||
353 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
354 | r = pread(fd, addr, len, offset); | |
355 | if (r != len) | |
356 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
357 | } | |
358 | ||
2e04ef76 RR |
359 | /* |
360 | * This routine takes an open vmlinux image, which is in ELF, and maps it into | |
dde79789 RR |
361 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
362 | * by all modern binaries on Linux including the kernel. | |
363 | * | |
364 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
365 | * address. We use the physical address; the Guest will map itself to the |
366 | * virtual address. | |
dde79789 | 367 | * |
2e04ef76 RR |
368 | * We return the starting address. |
369 | */ | |
47436aa4 | 370 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 371 | { |
8ca47e00 RR |
372 | Elf32_Phdr phdr[ehdr->e_phnum]; |
373 | unsigned int i; | |
8ca47e00 | 374 | |
2e04ef76 RR |
375 | /* |
376 | * Sanity checks on the main ELF header: an x86 executable with a | |
377 | * reasonable number of correctly-sized program headers. | |
378 | */ | |
8ca47e00 RR |
379 | if (ehdr->e_type != ET_EXEC |
380 | || ehdr->e_machine != EM_386 | |
381 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
382 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
383 | errx(1, "Malformed elf header"); | |
384 | ||
2e04ef76 RR |
385 | /* |
386 | * An ELF executable contains an ELF header and a number of "program" | |
dde79789 | 387 | * headers which indicate which parts ("segments") of the program to |
2e04ef76 RR |
388 | * load where. |
389 | */ | |
dde79789 RR |
390 | |
391 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
392 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
393 | err(1, "Seeking to program headers"); | |
394 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
395 | err(1, "Reading program headers"); | |
396 | ||
2e04ef76 RR |
397 | /* |
398 | * Try all the headers: there are usually only three. A read-only one, | |
399 | * a read-write one, and a "note" section which we don't load. | |
400 | */ | |
8ca47e00 | 401 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 402 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
403 | if (phdr[i].p_type != PT_LOAD) |
404 | continue; | |
405 | ||
406 | verbose("Section %i: size %i addr %p\n", | |
407 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
408 | ||
6649bb7a | 409 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 410 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 411 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
412 | } |
413 | ||
814a0e5c RR |
414 | /* The entry point is given in the ELF header. */ |
415 | return ehdr->e_entry; | |
8ca47e00 RR |
416 | } |
417 | ||
2e04ef76 RR |
418 | /*L:150 |
419 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed | |
420 | * to jump into it and it will unpack itself. We used to have to perform some | |
421 | * hairy magic because the unpacking code scared me. | |
dde79789 | 422 | * |
5bbf89fc RR |
423 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
424 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
2e04ef76 RR |
425 | * the funky header so we know where in the file to load, and away we go! |
426 | */ | |
47436aa4 | 427 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 428 | { |
43d33b21 | 429 | struct boot_params boot; |
5bbf89fc RR |
430 | int r; |
431 | /* Modern bzImages get loaded at 1M. */ | |
432 | void *p = from_guest_phys(0x100000); | |
433 | ||
2e04ef76 RR |
434 | /* |
435 | * Go back to the start of the file and read the header. It should be | |
436 | * a Linux boot header (see Documentation/x86/i386/boot.txt) | |
437 | */ | |
5bbf89fc | 438 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 439 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 440 | |
43d33b21 RR |
441 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
442 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
443 | errx(1, "This doesn't look like a bzImage to me"); |
444 | ||
43d33b21 RR |
445 | /* Skip over the extra sectors of the header. */ |
446 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
447 | |
448 | /* Now read everything into memory. in nice big chunks. */ | |
449 | while ((r = read(fd, p, 65536)) > 0) | |
450 | p += r; | |
451 | ||
43d33b21 RR |
452 | /* Finally, code32_start tells us where to enter the kernel. */ |
453 | return boot.hdr.code32_start; | |
8ca47e00 RR |
454 | } |
455 | ||
2e04ef76 RR |
456 | /*L:140 |
457 | * Loading the kernel is easy when it's a "vmlinux", but most kernels | |
e1e72965 | 458 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
2e04ef76 RR |
459 | * work, we can load those, too. |
460 | */ | |
47436aa4 | 461 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
462 | { |
463 | Elf32_Ehdr hdr; | |
464 | ||
dde79789 | 465 | /* Read in the first few bytes. */ |
8ca47e00 RR |
466 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
467 | err(1, "Reading kernel"); | |
468 | ||
dde79789 | 469 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 470 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 471 | return map_elf(fd, &hdr); |
8ca47e00 | 472 | |
a6bd8e13 | 473 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 474 | return load_bzimage(fd); |
8ca47e00 RR |
475 | } |
476 | ||
2e04ef76 RR |
477 | /* |
478 | * This is a trivial little helper to align pages. Andi Kleen hated it because | |
dde79789 RR |
479 | * it calls getpagesize() twice: "it's dumb code." |
480 | * | |
481 | * Kernel guys get really het up about optimization, even when it's not | |
2e04ef76 RR |
482 | * necessary. I leave this code as a reaction against that. |
483 | */ | |
8ca47e00 RR |
484 | static inline unsigned long page_align(unsigned long addr) |
485 | { | |
dde79789 | 486 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
487 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
488 | } | |
489 | ||
2e04ef76 RR |
490 | /*L:180 |
491 | * An "initial ram disk" is a disk image loaded into memory along with the | |
492 | * kernel which the kernel can use to boot from without needing any drivers. | |
493 | * Most distributions now use this as standard: the initrd contains the code to | |
494 | * load the appropriate driver modules for the current machine. | |
dde79789 RR |
495 | * |
496 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
2e04ef76 RR |
497 | * kernels. He sent me this (and tells me when I break it). |
498 | */ | |
8ca47e00 RR |
499 | static unsigned long load_initrd(const char *name, unsigned long mem) |
500 | { | |
501 | int ifd; | |
502 | struct stat st; | |
503 | unsigned long len; | |
8ca47e00 RR |
504 | |
505 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 506 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
507 | if (fstat(ifd, &st) < 0) |
508 | err(1, "fstat() on initrd '%s'", name); | |
509 | ||
2e04ef76 RR |
510 | /* |
511 | * We map the initrd at the top of memory, but mmap wants it to be | |
512 | * page-aligned, so we round the size up for that. | |
513 | */ | |
8ca47e00 | 514 | len = page_align(st.st_size); |
3c6b5bfa | 515 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
2e04ef76 RR |
516 | /* |
517 | * Once a file is mapped, you can close the file descriptor. It's a | |
518 | * little odd, but quite useful. | |
519 | */ | |
8ca47e00 | 520 | close(ifd); |
6649bb7a | 521 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
522 | |
523 | /* We return the initrd size. */ | |
8ca47e00 RR |
524 | return len; |
525 | } | |
e1e72965 | 526 | /*:*/ |
8ca47e00 | 527 | |
2e04ef76 RR |
528 | /* |
529 | * Simple routine to roll all the commandline arguments together with spaces | |
530 | * between them. | |
531 | */ | |
8ca47e00 RR |
532 | static void concat(char *dst, char *args[]) |
533 | { | |
534 | unsigned int i, len = 0; | |
535 | ||
536 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
537 | if (i) { |
538 | strcat(dst+len, " "); | |
539 | len++; | |
540 | } | |
8ca47e00 | 541 | strcpy(dst+len, args[i]); |
1ef36fa6 | 542 | len += strlen(args[i]); |
8ca47e00 RR |
543 | } |
544 | /* In case it's empty. */ | |
545 | dst[len] = '\0'; | |
546 | } | |
547 | ||
2e04ef76 RR |
548 | /*L:185 |
549 | * This is where we actually tell the kernel to initialize the Guest. We | |
e1e72965 | 550 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
58a24566 | 551 | * the base of Guest "physical" memory, the top physical page to allow and the |
2e04ef76 RR |
552 | * entry point for the Guest. |
553 | */ | |
56739c80 | 554 | static void tell_kernel(unsigned long start) |
8ca47e00 | 555 | { |
511801dc JS |
556 | unsigned long args[] = { LHREQ_INITIALIZE, |
557 | (unsigned long)guest_base, | |
58a24566 | 558 | guest_limit / getpagesize(), start }; |
3c6b5bfa RR |
559 | verbose("Guest: %p - %p (%#lx)\n", |
560 | guest_base, guest_base + guest_limit, guest_limit); | |
56739c80 RR |
561 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
562 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 563 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 564 | } |
dde79789 | 565 | /*:*/ |
8ca47e00 | 566 | |
a91d74a3 | 567 | /*L:200 |
dde79789 RR |
568 | * Device Handling. |
569 | * | |
e1e72965 | 570 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 571 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 572 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
573 | * if something funny is going on: |
574 | */ | |
8ca47e00 RR |
575 | static void *_check_pointer(unsigned long addr, unsigned int size, |
576 | unsigned int line) | |
577 | { | |
2e04ef76 RR |
578 | /* |
579 | * We have to separately check addr and addr+size, because size could | |
580 | * be huge and addr + size might wrap around. | |
581 | */ | |
3c6b5bfa | 582 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 583 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
2e04ef76 RR |
584 | /* |
585 | * We return a pointer for the caller's convenience, now we know it's | |
586 | * safe to use. | |
587 | */ | |
3c6b5bfa | 588 | return from_guest_phys(addr); |
8ca47e00 | 589 | } |
dde79789 | 590 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
591 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
592 | ||
2e04ef76 RR |
593 | /* |
594 | * Each buffer in the virtqueues is actually a chain of descriptors. This | |
e1e72965 | 595 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
2e04ef76 RR |
596 | * at the end. |
597 | */ | |
d1f0132e MM |
598 | static unsigned next_desc(struct vring_desc *desc, |
599 | unsigned int i, unsigned int max) | |
17cbca2b RR |
600 | { |
601 | unsigned int next; | |
602 | ||
603 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
604 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
605 | return max; | |
17cbca2b RR |
606 | |
607 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 608 | next = desc[i].next; |
17cbca2b RR |
609 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
610 | wmb(); | |
611 | ||
d1f0132e | 612 | if (next >= max) |
17cbca2b RR |
613 | errx(1, "Desc next is %u", next); |
614 | ||
615 | return next; | |
616 | } | |
617 | ||
a91d74a3 RR |
618 | /* |
619 | * This actually sends the interrupt for this virtqueue, if we've used a | |
620 | * buffer. | |
621 | */ | |
38bc2b8c RR |
622 | static void trigger_irq(struct virtqueue *vq) |
623 | { | |
624 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; | |
625 | ||
95c517c0 RR |
626 | /* Don't inform them if nothing used. */ |
627 | if (!vq->pending_used) | |
628 | return; | |
629 | vq->pending_used = 0; | |
630 | ||
ca60a42c RR |
631 | /* If they don't want an interrupt, don't send one... */ |
632 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) { | |
633 | /* ... unless they've asked us to force one on empty. */ | |
634 | if (!vq->dev->irq_on_empty | |
635 | || lg_last_avail(vq) != vq->vring.avail->idx) | |
636 | return; | |
637 | } | |
38bc2b8c RR |
638 | |
639 | /* Send the Guest an interrupt tell them we used something up. */ | |
640 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
641 | err(1, "Triggering irq %i", vq->config.irq); | |
642 | } | |
643 | ||
2e04ef76 | 644 | /* |
a91d74a3 | 645 | * This looks in the virtqueue for the first available buffer, and converts |
17cbca2b RR |
646 | * it to an iovec for convenient access. Since descriptors consist of some |
647 | * number of output then some number of input descriptors, it's actually two | |
648 | * iovecs, but we pack them into one and note how many of each there were. | |
649 | * | |
a91d74a3 | 650 | * This function waits if necessary, and returns the descriptor number found. |
2e04ef76 | 651 | */ |
659a0e66 RR |
652 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
653 | struct iovec iov[], | |
654 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 655 | { |
d1f0132e MM |
656 | unsigned int i, head, max; |
657 | struct vring_desc *desc; | |
659a0e66 RR |
658 | u16 last_avail = lg_last_avail(vq); |
659 | ||
a91d74a3 | 660 | /* There's nothing available? */ |
659a0e66 RR |
661 | while (last_avail == vq->vring.avail->idx) { |
662 | u64 event; | |
663 | ||
a91d74a3 RR |
664 | /* |
665 | * Since we're about to sleep, now is a good time to tell the | |
666 | * Guest about what we've used up to now. | |
667 | */ | |
38bc2b8c RR |
668 | trigger_irq(vq); |
669 | ||
b60da13f RR |
670 | /* OK, now we need to know about added descriptors. */ |
671 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
672 | ||
2e04ef76 RR |
673 | /* |
674 | * They could have slipped one in as we were doing that: make | |
675 | * sure it's written, then check again. | |
676 | */ | |
b60da13f RR |
677 | mb(); |
678 | if (last_avail != vq->vring.avail->idx) { | |
679 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
680 | break; | |
681 | } | |
682 | ||
659a0e66 RR |
683 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
684 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
685 | errx(1, "Event read failed?"); | |
b60da13f RR |
686 | |
687 | /* We don't need to be notified again. */ | |
688 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 689 | } |
17cbca2b RR |
690 | |
691 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 692 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
17cbca2b | 693 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 694 | last_avail, vq->vring.avail->idx); |
17cbca2b | 695 | |
2e04ef76 RR |
696 | /* |
697 | * Grab the next descriptor number they're advertising, and increment | |
698 | * the index we've seen. | |
699 | */ | |
b5111790 RR |
700 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
701 | lg_last_avail(vq)++; | |
17cbca2b RR |
702 | |
703 | /* If their number is silly, that's a fatal mistake. */ | |
704 | if (head >= vq->vring.num) | |
705 | errx(1, "Guest says index %u is available", head); | |
706 | ||
707 | /* When we start there are none of either input nor output. */ | |
708 | *out_num = *in_num = 0; | |
709 | ||
d1f0132e MM |
710 | max = vq->vring.num; |
711 | desc = vq->vring.desc; | |
17cbca2b | 712 | i = head; |
d1f0132e | 713 | |
2e04ef76 RR |
714 | /* |
715 | * If this is an indirect entry, then this buffer contains a descriptor | |
716 | * table which we handle as if it's any normal descriptor chain. | |
717 | */ | |
d1f0132e MM |
718 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { |
719 | if (desc[i].len % sizeof(struct vring_desc)) | |
720 | errx(1, "Invalid size for indirect buffer table"); | |
721 | ||
722 | max = desc[i].len / sizeof(struct vring_desc); | |
723 | desc = check_pointer(desc[i].addr, desc[i].len); | |
724 | i = 0; | |
725 | } | |
726 | ||
17cbca2b RR |
727 | do { |
728 | /* Grab the first descriptor, and check it's OK. */ | |
d1f0132e | 729 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 730 | iov[*out_num + *in_num].iov_base |
d1f0132e | 731 | = check_pointer(desc[i].addr, desc[i].len); |
17cbca2b | 732 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 733 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
734 | (*in_num)++; |
735 | else { | |
2e04ef76 RR |
736 | /* |
737 | * If it's an output descriptor, they're all supposed | |
738 | * to come before any input descriptors. | |
739 | */ | |
17cbca2b RR |
740 | if (*in_num) |
741 | errx(1, "Descriptor has out after in"); | |
742 | (*out_num)++; | |
743 | } | |
744 | ||
745 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 746 | if (*out_num + *in_num > max) |
17cbca2b | 747 | errx(1, "Looped descriptor"); |
d1f0132e | 748 | } while ((i = next_desc(desc, i, max)) != max); |
dde79789 | 749 | |
17cbca2b | 750 | return head; |
8ca47e00 RR |
751 | } |
752 | ||
2e04ef76 | 753 | /* |
a91d74a3 RR |
754 | * After we've used one of their buffers, we tell the Guest about it. Sometime |
755 | * later we'll want to send them an interrupt using trigger_irq(); note that | |
756 | * wait_for_vq_desc() does that for us if it has to wait. | |
2e04ef76 | 757 | */ |
17cbca2b | 758 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
8ca47e00 | 759 | { |
17cbca2b RR |
760 | struct vring_used_elem *used; |
761 | ||
2e04ef76 RR |
762 | /* |
763 | * The virtqueue contains a ring of used buffers. Get a pointer to the | |
764 | * next entry in that used ring. | |
765 | */ | |
17cbca2b RR |
766 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
767 | used->id = head; | |
768 | used->len = len; | |
769 | /* Make sure buffer is written before we update index. */ | |
770 | wmb(); | |
771 | vq->vring.used->idx++; | |
95c517c0 | 772 | vq->pending_used++; |
8ca47e00 RR |
773 | } |
774 | ||
17cbca2b | 775 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 776 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 777 | { |
17cbca2b | 778 | add_used(vq, head, len); |
56739c80 | 779 | trigger_irq(vq); |
8ca47e00 RR |
780 | } |
781 | ||
e1e72965 RR |
782 | /* |
783 | * The Console | |
784 | * | |
2e04ef76 RR |
785 | * We associate some data with the console for our exit hack. |
786 | */ | |
1842f23c | 787 | struct console_abort { |
dde79789 | 788 | /* How many times have they hit ^C? */ |
8ca47e00 | 789 | int count; |
dde79789 | 790 | /* When did they start? */ |
8ca47e00 RR |
791 | struct timeval start; |
792 | }; | |
793 | ||
dde79789 | 794 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 795 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 796 | { |
8ca47e00 | 797 | int len; |
17cbca2b | 798 | unsigned int head, in_num, out_num; |
659a0e66 RR |
799 | struct console_abort *abort = vq->dev->priv; |
800 | struct iovec iov[vq->vring.num]; | |
56ae43df | 801 | |
a91d74a3 | 802 | /* Make sure there's a descriptor available. */ |
659a0e66 | 803 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
56ae43df | 804 | if (out_num) |
17cbca2b | 805 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 806 | |
a91d74a3 | 807 | /* Read into it. This is where we usually wait. */ |
659a0e66 | 808 | len = readv(STDIN_FILENO, iov, in_num); |
8ca47e00 | 809 | if (len <= 0) { |
659a0e66 | 810 | /* Ran out of input? */ |
8ca47e00 | 811 | warnx("Failed to get console input, ignoring console."); |
2e04ef76 RR |
812 | /* |
813 | * For simplicity, dying threads kill the whole Launcher. So | |
814 | * just nap here. | |
815 | */ | |
659a0e66 RR |
816 | for (;;) |
817 | pause(); | |
8ca47e00 RR |
818 | } |
819 | ||
a91d74a3 | 820 | /* Tell the Guest we used a buffer. */ |
659a0e66 | 821 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 822 | |
2e04ef76 RR |
823 | /* |
824 | * Three ^C within one second? Exit. | |
dde79789 | 825 | * |
659a0e66 RR |
826 | * This is such a hack, but works surprisingly well. Each ^C has to |
827 | * be in a buffer by itself, so they can't be too fast. But we check | |
828 | * that we get three within about a second, so they can't be too | |
2e04ef76 RR |
829 | * slow. |
830 | */ | |
659a0e66 | 831 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
8ca47e00 | 832 | abort->count = 0; |
659a0e66 RR |
833 | return; |
834 | } | |
8ca47e00 | 835 | |
659a0e66 RR |
836 | abort->count++; |
837 | if (abort->count == 1) | |
838 | gettimeofday(&abort->start, NULL); | |
839 | else if (abort->count == 3) { | |
840 | struct timeval now; | |
841 | gettimeofday(&now, NULL); | |
842 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
843 | if (now.tv_sec <= abort->start.tv_sec+1) | |
844 | kill(0, SIGINT); | |
845 | abort->count = 0; | |
846 | } | |
8ca47e00 RR |
847 | } |
848 | ||
659a0e66 RR |
849 | /* This is the routine which handles console output (ie. stdout). */ |
850 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 851 | { |
17cbca2b | 852 | unsigned int head, out, in; |
17cbca2b RR |
853 | struct iovec iov[vq->vring.num]; |
854 | ||
a91d74a3 | 855 | /* We usually wait in here, for the Guest to give us something. */ |
659a0e66 RR |
856 | head = wait_for_vq_desc(vq, iov, &out, &in); |
857 | if (in) | |
858 | errx(1, "Input buffers in console output queue?"); | |
a91d74a3 RR |
859 | |
860 | /* writev can return a partial write, so we loop here. */ | |
659a0e66 RR |
861 | while (!iov_empty(iov, out)) { |
862 | int len = writev(STDOUT_FILENO, iov, out); | |
863 | if (len <= 0) | |
864 | err(1, "Write to stdout gave %i", len); | |
865 | iov_consume(iov, out, len); | |
17cbca2b | 866 | } |
a91d74a3 RR |
867 | |
868 | /* | |
869 | * We're finished with that buffer: if we're going to sleep, | |
870 | * wait_for_vq_desc() will prod the Guest with an interrupt. | |
871 | */ | |
38bc2b8c | 872 | add_used(vq, head, 0); |
a161883a RR |
873 | } |
874 | ||
e1e72965 RR |
875 | /* |
876 | * The Network | |
877 | * | |
878 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 879 | * and write them to /dev/net/tun. |
a6bd8e13 | 880 | */ |
659a0e66 RR |
881 | struct net_info { |
882 | int tunfd; | |
883 | }; | |
884 | ||
885 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 886 | { |
659a0e66 RR |
887 | struct net_info *net_info = vq->dev->priv; |
888 | unsigned int head, out, in; | |
17cbca2b | 889 | struct iovec iov[vq->vring.num]; |
a161883a | 890 | |
a91d74a3 | 891 | /* We usually wait in here for the Guest to give us a packet. */ |
659a0e66 RR |
892 | head = wait_for_vq_desc(vq, iov, &out, &in); |
893 | if (in) | |
894 | errx(1, "Input buffers in net output queue?"); | |
a91d74a3 RR |
895 | /* |
896 | * Send the whole thing through to /dev/net/tun. It expects the exact | |
897 | * same format: what a coincidence! | |
898 | */ | |
659a0e66 RR |
899 | if (writev(net_info->tunfd, iov, out) < 0) |
900 | errx(1, "Write to tun failed?"); | |
a91d74a3 RR |
901 | |
902 | /* | |
903 | * Done with that one; wait_for_vq_desc() will send the interrupt if | |
904 | * all packets are processed. | |
905 | */ | |
38bc2b8c | 906 | add_used(vq, head, 0); |
8ca47e00 RR |
907 | } |
908 | ||
a91d74a3 RR |
909 | /* |
910 | * Handling network input is a bit trickier, because I've tried to optimize it. | |
911 | * | |
912 | * First we have a helper routine which tells is if from this file descriptor | |
913 | * (ie. the /dev/net/tun device) will block: | |
914 | */ | |
4a8962e2 RR |
915 | static bool will_block(int fd) |
916 | { | |
917 | fd_set fdset; | |
918 | struct timeval zero = { 0, 0 }; | |
919 | FD_ZERO(&fdset); | |
920 | FD_SET(fd, &fdset); | |
921 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
922 | } | |
923 | ||
a91d74a3 RR |
924 | /* |
925 | * This handles packets coming in from the tun device to our Guest. Like all | |
926 | * service routines, it gets called again as soon as it returns, so you don't | |
927 | * see a while(1) loop here. | |
928 | */ | |
659a0e66 | 929 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 930 | { |
8ca47e00 | 931 | int len; |
659a0e66 RR |
932 | unsigned int head, out, in; |
933 | struct iovec iov[vq->vring.num]; | |
934 | struct net_info *net_info = vq->dev->priv; | |
935 | ||
a91d74a3 RR |
936 | /* |
937 | * Get a descriptor to write an incoming packet into. This will also | |
938 | * send an interrupt if they're out of descriptors. | |
939 | */ | |
659a0e66 RR |
940 | head = wait_for_vq_desc(vq, iov, &out, &in); |
941 | if (out) | |
942 | errx(1, "Output buffers in net input queue?"); | |
4a8962e2 | 943 | |
a91d74a3 RR |
944 | /* |
945 | * If it looks like we'll block reading from the tun device, send them | |
946 | * an interrupt. | |
947 | */ | |
4a8962e2 RR |
948 | if (vq->pending_used && will_block(net_info->tunfd)) |
949 | trigger_irq(vq); | |
950 | ||
a91d74a3 RR |
951 | /* |
952 | * Read in the packet. This is where we normally wait (when there's no | |
953 | * incoming network traffic). | |
954 | */ | |
659a0e66 | 955 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 956 | if (len <= 0) |
659a0e66 | 957 | err(1, "Failed to read from tun."); |
a91d74a3 RR |
958 | |
959 | /* | |
960 | * Mark that packet buffer as used, but don't interrupt here. We want | |
961 | * to wait until we've done as much work as we can. | |
962 | */ | |
4a8962e2 | 963 | add_used(vq, head, len); |
659a0e66 | 964 | } |
a91d74a3 | 965 | /*:*/ |
dde79789 | 966 | |
a91d74a3 | 967 | /* This is the helper to create threads: run the service routine in a loop. */ |
659a0e66 RR |
968 | static int do_thread(void *_vq) |
969 | { | |
970 | struct virtqueue *vq = _vq; | |
17cbca2b | 971 | |
659a0e66 RR |
972 | for (;;) |
973 | vq->service(vq); | |
974 | return 0; | |
975 | } | |
17cbca2b | 976 | |
2e04ef76 RR |
977 | /* |
978 | * When a child dies, we kill our entire process group with SIGTERM. This | |
979 | * also has the side effect that the shell restores the console for us! | |
980 | */ | |
659a0e66 RR |
981 | static void kill_launcher(int signal) |
982 | { | |
983 | kill(0, SIGTERM); | |
8ca47e00 RR |
984 | } |
985 | ||
659a0e66 | 986 | static void reset_device(struct device *dev) |
56ae43df | 987 | { |
659a0e66 RR |
988 | struct virtqueue *vq; |
989 | ||
990 | verbose("Resetting device %s\n", dev->name); | |
991 | ||
992 | /* Clear any features they've acked. */ | |
993 | memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); | |
994 | ||
995 | /* We're going to be explicitly killing threads, so ignore them. */ | |
996 | signal(SIGCHLD, SIG_IGN); | |
997 | ||
998 | /* Zero out the virtqueues, get rid of their threads */ | |
999 | for (vq = dev->vq; vq; vq = vq->next) { | |
1000 | if (vq->thread != (pid_t)-1) { | |
1001 | kill(vq->thread, SIGTERM); | |
1002 | waitpid(vq->thread, NULL, 0); | |
1003 | vq->thread = (pid_t)-1; | |
1004 | } | |
1005 | memset(vq->vring.desc, 0, | |
1006 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); | |
1007 | lg_last_avail(vq) = 0; | |
1008 | } | |
1009 | dev->running = false; | |
1010 | ||
1011 | /* Now we care if threads die. */ | |
1012 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
1013 | } |
1014 | ||
a91d74a3 RR |
1015 | /*L:216 |
1016 | * This actually creates the thread which services the virtqueue for a device. | |
1017 | */ | |
659a0e66 | 1018 | static void create_thread(struct virtqueue *vq) |
5dae785a | 1019 | { |
2e04ef76 | 1020 | /* |
a91d74a3 RR |
1021 | * Create stack for thread. Since the stack grows upwards, we point |
1022 | * the stack pointer to the end of this region. | |
2e04ef76 | 1023 | */ |
659a0e66 RR |
1024 | char *stack = malloc(32768); |
1025 | unsigned long args[] = { LHREQ_EVENTFD, | |
1026 | vq->config.pfn*getpagesize(), 0 }; | |
1027 | ||
1028 | /* Create a zero-initialized eventfd. */ | |
1029 | vq->eventfd = eventfd(0, 0); | |
1030 | if (vq->eventfd < 0) | |
1031 | err(1, "Creating eventfd"); | |
1032 | args[2] = vq->eventfd; | |
1033 | ||
a91d74a3 RR |
1034 | /* |
1035 | * Attach an eventfd to this virtqueue: it will go off when the Guest | |
1036 | * does an LHCALL_NOTIFY for this vq. | |
1037 | */ | |
659a0e66 RR |
1038 | if (write(lguest_fd, &args, sizeof(args)) != 0) |
1039 | err(1, "Attaching eventfd"); | |
1040 | ||
a91d74a3 RR |
1041 | /* |
1042 | * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so | |
1043 | * we get a signal if it dies. | |
1044 | */ | |
659a0e66 RR |
1045 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); |
1046 | if (vq->thread == (pid_t)-1) | |
1047 | err(1, "Creating clone"); | |
a91d74a3 RR |
1048 | |
1049 | /* We close our local copy now the child has it. */ | |
659a0e66 | 1050 | close(vq->eventfd); |
5dae785a RR |
1051 | } |
1052 | ||
ca60a42c RR |
1053 | static bool accepted_feature(struct device *dev, unsigned int bit) |
1054 | { | |
1055 | const u8 *features = get_feature_bits(dev) + dev->feature_len; | |
1056 | ||
1057 | if (dev->feature_len < bit / CHAR_BIT) | |
1058 | return false; | |
1059 | return features[bit / CHAR_BIT] & (1 << (bit % CHAR_BIT)); | |
1060 | } | |
1061 | ||
659a0e66 | 1062 | static void start_device(struct device *dev) |
6e5aa7ef | 1063 | { |
659a0e66 | 1064 | unsigned int i; |
6e5aa7ef RR |
1065 | struct virtqueue *vq; |
1066 | ||
659a0e66 RR |
1067 | verbose("Device %s OK: offered", dev->name); |
1068 | for (i = 0; i < dev->feature_len; i++) | |
1069 | verbose(" %02x", get_feature_bits(dev)[i]); | |
1070 | verbose(", accepted"); | |
1071 | for (i = 0; i < dev->feature_len; i++) | |
1072 | verbose(" %02x", get_feature_bits(dev) | |
1073 | [dev->feature_len+i]); | |
1074 | ||
ca60a42c RR |
1075 | dev->irq_on_empty = accepted_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
1076 | ||
659a0e66 RR |
1077 | for (vq = dev->vq; vq; vq = vq->next) { |
1078 | if (vq->service) | |
1079 | create_thread(vq); | |
1080 | } | |
1081 | dev->running = true; | |
1082 | } | |
1083 | ||
1084 | static void cleanup_devices(void) | |
1085 | { | |
1086 | struct device *dev; | |
1087 | ||
1088 | for (dev = devices.dev; dev; dev = dev->next) | |
1089 | reset_device(dev); | |
6e5aa7ef | 1090 | |
659a0e66 RR |
1091 | /* If we saved off the original terminal settings, restore them now. */ |
1092 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
1093 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
1094 | } | |
6e5aa7ef | 1095 | |
659a0e66 RR |
1096 | /* When the Guest tells us they updated the status field, we handle it. */ |
1097 | static void update_device_status(struct device *dev) | |
1098 | { | |
1099 | /* A zero status is a reset, otherwise it's a set of flags. */ | |
1100 | if (dev->desc->status == 0) | |
1101 | reset_device(dev); | |
1102 | else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
a007a751 | 1103 | warnx("Device %s configuration FAILED", dev->name); |
659a0e66 RR |
1104 | if (dev->running) |
1105 | reset_device(dev); | |
a007a751 | 1106 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { |
659a0e66 RR |
1107 | if (!dev->running) |
1108 | start_device(dev); | |
6e5aa7ef RR |
1109 | } |
1110 | } | |
1111 | ||
a91d74a3 RR |
1112 | /*L:215 |
1113 | * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In | |
1114 | * particular, it's used to notify us of device status changes during boot. | |
1115 | */ | |
56739c80 | 1116 | static void handle_output(unsigned long addr) |
8ca47e00 RR |
1117 | { |
1118 | struct device *i; | |
17cbca2b | 1119 | |
659a0e66 | 1120 | /* Check each device. */ |
17cbca2b | 1121 | for (i = devices.dev; i; i = i->next) { |
659a0e66 RR |
1122 | struct virtqueue *vq; |
1123 | ||
a91d74a3 RR |
1124 | /* |
1125 | * Notifications to device descriptors mean they updated the | |
1126 | * device status. | |
1127 | */ | |
6e5aa7ef | 1128 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1129 | update_device_status(i); |
6e5aa7ef RR |
1130 | return; |
1131 | } | |
1132 | ||
a91d74a3 RR |
1133 | /* |
1134 | * Devices *can* be used before status is set to DRIVER_OK. | |
1135 | * The original plan was that they would never do this: they | |
1136 | * would always finish setting up their status bits before | |
1137 | * actually touching the virtqueues. In practice, we allowed | |
1138 | * them to, and they do (eg. the disk probes for partition | |
1139 | * tables as part of initialization). | |
1140 | * | |
1141 | * If we see this, we start the device: once it's running, we | |
1142 | * expect the device to catch all the notifications. | |
1143 | */ | |
17cbca2b | 1144 | for (vq = i->vq; vq; vq = vq->next) { |
659a0e66 | 1145 | if (addr != vq->config.pfn*getpagesize()) |
6e5aa7ef | 1146 | continue; |
659a0e66 RR |
1147 | if (i->running) |
1148 | errx(1, "Notification on running %s", i->name); | |
a91d74a3 | 1149 | /* This just calls create_thread() for each virtqueue */ |
659a0e66 | 1150 | start_device(i); |
6e5aa7ef | 1151 | return; |
8ca47e00 RR |
1152 | } |
1153 | } | |
dde79789 | 1154 | |
2e04ef76 RR |
1155 | /* |
1156 | * Early console write is done using notify on a nul-terminated string | |
1157 | * in Guest memory. It's also great for hacking debugging messages | |
1158 | * into a Guest. | |
1159 | */ | |
17cbca2b RR |
1160 | if (addr >= guest_limit) |
1161 | errx(1, "Bad NOTIFY %#lx", addr); | |
1162 | ||
1163 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1164 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1165 | } |
1166 | ||
dde79789 RR |
1167 | /*L:190 |
1168 | * Device Setup | |
1169 | * | |
1170 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1171 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1172 | * routines to allocate and manage them. |
1173 | */ | |
8ca47e00 | 1174 | |
2e04ef76 RR |
1175 | /* |
1176 | * The layout of the device page is a "struct lguest_device_desc" followed by a | |
a586d4f6 RR |
1177 | * number of virtqueue descriptors, then two sets of feature bits, then an |
1178 | * array of configuration bytes. This routine returns the configuration | |
2e04ef76 RR |
1179 | * pointer. |
1180 | */ | |
a586d4f6 RR |
1181 | static u8 *device_config(const struct device *dev) |
1182 | { | |
1183 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1184 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1185 | + dev->feature_len * 2; | |
17cbca2b RR |
1186 | } |
1187 | ||
2e04ef76 RR |
1188 | /* |
1189 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
a586d4f6 | 1190 | * table page just above the Guest's normal memory. It returns a pointer to |
2e04ef76 RR |
1191 | * that descriptor. |
1192 | */ | |
a586d4f6 | 1193 | static struct lguest_device_desc *new_dev_desc(u16 type) |
17cbca2b | 1194 | { |
a586d4f6 RR |
1195 | struct lguest_device_desc d = { .type = type }; |
1196 | void *p; | |
17cbca2b | 1197 | |
a586d4f6 RR |
1198 | /* Figure out where the next device config is, based on the last one. */ |
1199 | if (devices.lastdev) | |
1200 | p = device_config(devices.lastdev) | |
1201 | + devices.lastdev->desc->config_len; | |
1202 | else | |
1203 | p = devices.descpage; | |
17cbca2b | 1204 | |
a586d4f6 RR |
1205 | /* We only have one page for all the descriptors. */ |
1206 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1207 | errx(1, "Too many devices"); | |
17cbca2b | 1208 | |
a586d4f6 RR |
1209 | /* p might not be aligned, so we memcpy in. */ |
1210 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1211 | } |
1212 | ||
2e04ef76 RR |
1213 | /* |
1214 | * Each device descriptor is followed by the description of its virtqueues. We | |
1215 | * specify how many descriptors the virtqueue is to have. | |
1216 | */ | |
17cbca2b | 1217 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
659a0e66 | 1218 | void (*service)(struct virtqueue *)) |
17cbca2b RR |
1219 | { |
1220 | unsigned int pages; | |
1221 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1222 | void *p; | |
1223 | ||
a6bd8e13 | 1224 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1225 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 1226 | / getpagesize(); |
17cbca2b RR |
1227 | p = get_pages(pages); |
1228 | ||
d1c856e0 RR |
1229 | /* Initialize the virtqueue */ |
1230 | vq->next = NULL; | |
1231 | vq->last_avail_idx = 0; | |
1232 | vq->dev = dev; | |
a91d74a3 RR |
1233 | |
1234 | /* | |
1235 | * This is the routine the service thread will run, and its Process ID | |
1236 | * once it's running. | |
1237 | */ | |
659a0e66 RR |
1238 | vq->service = service; |
1239 | vq->thread = (pid_t)-1; | |
d1c856e0 | 1240 | |
17cbca2b RR |
1241 | /* Initialize the configuration. */ |
1242 | vq->config.num = num_descs; | |
1243 | vq->config.irq = devices.next_irq++; | |
1244 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1245 | ||
1246 | /* Initialize the vring. */ | |
2966af73 | 1247 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1248 | |
2e04ef76 RR |
1249 | /* |
1250 | * Append virtqueue to this device's descriptor. We use | |
a586d4f6 RR |
1251 | * device_config() to get the end of the device's current virtqueues; |
1252 | * we check that we haven't added any config or feature information | |
2e04ef76 RR |
1253 | * yet, otherwise we'd be overwriting them. |
1254 | */ | |
a586d4f6 RR |
1255 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); |
1256 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1257 | dev->num_vq++; |
a586d4f6 RR |
1258 | dev->desc->num_vq++; |
1259 | ||
1260 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b | 1261 | |
2e04ef76 RR |
1262 | /* |
1263 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1264 | * second. | |
1265 | */ | |
17cbca2b RR |
1266 | for (i = &dev->vq; *i; i = &(*i)->next); |
1267 | *i = vq; | |
8ca47e00 RR |
1268 | } |
1269 | ||
2e04ef76 RR |
1270 | /* |
1271 | * The first half of the feature bitmask is for us to advertise features. The | |
1272 | * second half is for the Guest to accept features. | |
1273 | */ | |
a586d4f6 RR |
1274 | static void add_feature(struct device *dev, unsigned bit) |
1275 | { | |
6e5aa7ef | 1276 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1277 | |
1278 | /* We can't extend the feature bits once we've added config bytes */ | |
1279 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1280 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1281 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1282 | } |
1283 | ||
a586d4f6 RR |
1284 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1285 | } | |
1286 | ||
2e04ef76 RR |
1287 | /* |
1288 | * This routine sets the configuration fields for an existing device's | |
a586d4f6 | 1289 | * descriptor. It only works for the last device, but that's OK because that's |
2e04ef76 RR |
1290 | * how we use it. |
1291 | */ | |
a586d4f6 RR |
1292 | static void set_config(struct device *dev, unsigned len, const void *conf) |
1293 | { | |
1294 | /* Check we haven't overflowed our single page. */ | |
1295 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1296 | errx(1, "Too many devices"); | |
1297 | ||
1298 | /* Copy in the config information, and store the length. */ | |
1299 | memcpy(device_config(dev), conf, len); | |
1300 | dev->desc->config_len = len; | |
8ef562d1 RR |
1301 | |
1302 | /* Size must fit in config_len field (8 bits)! */ | |
1303 | assert(dev->desc->config_len == len); | |
a586d4f6 RR |
1304 | } |
1305 | ||
2e04ef76 RR |
1306 | /* |
1307 | * This routine does all the creation and setup of a new device, including | |
a91d74a3 RR |
1308 | * calling new_dev_desc() to allocate the descriptor and device memory. We |
1309 | * don't actually start the service threads until later. | |
a6bd8e13 | 1310 | * |
2e04ef76 RR |
1311 | * See what I mean about userspace being boring? |
1312 | */ | |
659a0e66 | 1313 | static struct device *new_device(const char *name, u16 type) |
8ca47e00 RR |
1314 | { |
1315 | struct device *dev = malloc(sizeof(*dev)); | |
1316 | ||
dde79789 | 1317 | /* Now we populate the fields one at a time. */ |
17cbca2b | 1318 | dev->desc = new_dev_desc(type); |
17cbca2b | 1319 | dev->name = name; |
d1c856e0 | 1320 | dev->vq = NULL; |
713b15b3 RR |
1321 | dev->feature_len = 0; |
1322 | dev->num_vq = 0; | |
659a0e66 | 1323 | dev->running = false; |
a586d4f6 | 1324 | |
2e04ef76 RR |
1325 | /* |
1326 | * Append to device list. Prepending to a single-linked list is | |
a586d4f6 RR |
1327 | * easier, but the user expects the devices to be arranged on the bus |
1328 | * in command-line order. The first network device on the command line | |
2e04ef76 RR |
1329 | * is eth0, the first block device /dev/vda, etc. |
1330 | */ | |
a586d4f6 RR |
1331 | if (devices.lastdev) |
1332 | devices.lastdev->next = dev; | |
1333 | else | |
1334 | devices.dev = dev; | |
1335 | devices.lastdev = dev; | |
1336 | ||
8ca47e00 RR |
1337 | return dev; |
1338 | } | |
1339 | ||
2e04ef76 RR |
1340 | /* |
1341 | * Our first setup routine is the console. It's a fairly simple device, but | |
1342 | * UNIX tty handling makes it uglier than it could be. | |
1343 | */ | |
17cbca2b | 1344 | static void setup_console(void) |
8ca47e00 RR |
1345 | { |
1346 | struct device *dev; | |
1347 | ||
dde79789 | 1348 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1349 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1350 | struct termios term = orig_term; | |
2e04ef76 RR |
1351 | /* |
1352 | * Then we turn off echo, line buffering and ^C etc: We want a | |
1353 | * raw input stream to the Guest. | |
1354 | */ | |
8ca47e00 RR |
1355 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1356 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
1357 | } |
1358 | ||
659a0e66 RR |
1359 | dev = new_device("console", VIRTIO_ID_CONSOLE); |
1360 | ||
dde79789 | 1361 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1362 | dev->priv = malloc(sizeof(struct console_abort)); |
1363 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1364 | |
2e04ef76 RR |
1365 | /* |
1366 | * The console needs two virtqueues: the input then the output. When | |
56ae43df RR |
1367 | * they put something the input queue, we make sure we're listening to |
1368 | * stdin. When they put something in the output queue, we write it to | |
2e04ef76 RR |
1369 | * stdout. |
1370 | */ | |
659a0e66 RR |
1371 | add_virtqueue(dev, VIRTQUEUE_NUM, console_input); |
1372 | add_virtqueue(dev, VIRTQUEUE_NUM, console_output); | |
17cbca2b | 1373 | |
659a0e66 | 1374 | verbose("device %u: console\n", ++devices.device_num); |
8ca47e00 | 1375 | } |
17cbca2b | 1376 | /*:*/ |
8ca47e00 | 1377 | |
2e04ef76 RR |
1378 | /*M:010 |
1379 | * Inter-guest networking is an interesting area. Simplest is to have a | |
17cbca2b RR |
1380 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
1381 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1382 | * |
17cbca2b RR |
1383 | * More sopisticated is to use one of the tools developed for project like UML |
1384 | * to do networking. | |
dde79789 | 1385 | * |
17cbca2b RR |
1386 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1387 | * completely generic ("here's my vring, attach to your vring") and would work | |
1388 | * for any traffic. Of course, namespace and permissions issues need to be | |
1389 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1390 | * multiple inter-guest channels behind one interface, although it would | |
1391 | * require some manner of hotplugging new virtio channels. | |
1392 | * | |
2e04ef76 RR |
1393 | * Finally, we could implement a virtio network switch in the kernel. |
1394 | :*/ | |
8ca47e00 RR |
1395 | |
1396 | static u32 str2ip(const char *ipaddr) | |
1397 | { | |
dec6a2be | 1398 | unsigned int b[4]; |
8ca47e00 | 1399 | |
dec6a2be MM |
1400 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1401 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1402 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1403 | } | |
1404 | ||
1405 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1406 | { | |
1407 | unsigned int m[6]; | |
1408 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1409 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1410 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1411 | mac[0] = m[0]; | |
1412 | mac[1] = m[1]; | |
1413 | mac[2] = m[2]; | |
1414 | mac[3] = m[3]; | |
1415 | mac[4] = m[4]; | |
1416 | mac[5] = m[5]; | |
8ca47e00 RR |
1417 | } |
1418 | ||
2e04ef76 RR |
1419 | /* |
1420 | * This code is "adapted" from libbridge: it attaches the Host end of the | |
dde79789 RR |
1421 | * network device to the bridge device specified by the command line. |
1422 | * | |
1423 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
2e04ef76 RR |
1424 | * dislike bridging), and I just try not to break it. |
1425 | */ | |
8ca47e00 RR |
1426 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1427 | { | |
1428 | int ifidx; | |
1429 | struct ifreq ifr; | |
1430 | ||
1431 | if (!*br_name) | |
1432 | errx(1, "must specify bridge name"); | |
1433 | ||
1434 | ifidx = if_nametoindex(if_name); | |
1435 | if (!ifidx) | |
1436 | errx(1, "interface %s does not exist!", if_name); | |
1437 | ||
1438 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1439 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1440 | ifr.ifr_ifindex = ifidx; |
1441 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1442 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1443 | } | |
1444 | ||
2e04ef76 RR |
1445 | /* |
1446 | * This sets up the Host end of the network device with an IP address, brings | |
dde79789 | 1447 | * it up so packets will flow, the copies the MAC address into the hwaddr |
2e04ef76 RR |
1448 | * pointer. |
1449 | */ | |
dec6a2be | 1450 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1451 | { |
1452 | struct ifreq ifr; | |
f846619e | 1453 | struct sockaddr_in sin; |
8ca47e00 RR |
1454 | |
1455 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1456 | strcpy(ifr.ifr_name, tapif); |
1457 | ||
1458 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
f846619e RR |
1459 | sin.sin_family = AF_INET; |
1460 | sin.sin_addr.s_addr = htonl(ipaddr); | |
1461 | memcpy(&ifr.ifr_addr, &sin, sizeof(sin)); | |
8ca47e00 | 1462 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
dec6a2be | 1463 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1464 | ifr.ifr_flags = IFF_UP; |
1465 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1466 | err(1, "Bringing interface %s up", tapif); |
1467 | } | |
1468 | ||
dec6a2be | 1469 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1470 | { |
8ca47e00 | 1471 | struct ifreq ifr; |
dec6a2be MM |
1472 | int netfd; |
1473 | ||
1474 | /* Start with this zeroed. Messy but sure. */ | |
1475 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1476 | |
2e04ef76 RR |
1477 | /* |
1478 | * We open the /dev/net/tun device and tell it we want a tap device. A | |
dde79789 RR |
1479 | * tap device is like a tun device, only somehow different. To tell |
1480 | * the truth, I completely blundered my way through this code, but it | |
2e04ef76 RR |
1481 | * works now! |
1482 | */ | |
8ca47e00 | 1483 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1484 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1485 | strcpy(ifr.ifr_name, "tap%d"); |
1486 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1487 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1488 | |
398f187d RR |
1489 | if (ioctl(netfd, TUNSETOFFLOAD, |
1490 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1491 | err(1, "Could not set features for tun device"); | |
1492 | ||
2e04ef76 RR |
1493 | /* |
1494 | * We don't need checksums calculated for packets coming in this | |
1495 | * device: trust us! | |
1496 | */ | |
8ca47e00 RR |
1497 | ioctl(netfd, TUNSETNOCSUM, 1); |
1498 | ||
dec6a2be MM |
1499 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1500 | return netfd; | |
1501 | } | |
1502 | ||
2e04ef76 RR |
1503 | /*L:195 |
1504 | * Our network is a Host<->Guest network. This can either use bridging or | |
dec6a2be MM |
1505 | * routing, but the principle is the same: it uses the "tun" device to inject |
1506 | * packets into the Host as if they came in from a normal network card. We | |
2e04ef76 RR |
1507 | * just shunt packets between the Guest and the tun device. |
1508 | */ | |
dec6a2be MM |
1509 | static void setup_tun_net(char *arg) |
1510 | { | |
1511 | struct device *dev; | |
659a0e66 RR |
1512 | struct net_info *net_info = malloc(sizeof(*net_info)); |
1513 | int ipfd; | |
dec6a2be MM |
1514 | u32 ip = INADDR_ANY; |
1515 | bool bridging = false; | |
1516 | char tapif[IFNAMSIZ], *p; | |
1517 | struct virtio_net_config conf; | |
1518 | ||
659a0e66 | 1519 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 1520 | |
17cbca2b | 1521 | /* First we create a new network device. */ |
659a0e66 RR |
1522 | dev = new_device("net", VIRTIO_ID_NET); |
1523 | dev->priv = net_info; | |
dde79789 | 1524 | |
2e04ef76 | 1525 | /* Network devices need a recv and a send queue, just like console. */ |
659a0e66 RR |
1526 | add_virtqueue(dev, VIRTQUEUE_NUM, net_input); |
1527 | add_virtqueue(dev, VIRTQUEUE_NUM, net_output); | |
8ca47e00 | 1528 | |
2e04ef76 RR |
1529 | /* |
1530 | * We need a socket to perform the magic network ioctls to bring up the | |
1531 | * tap interface, connect to the bridge etc. Any socket will do! | |
1532 | */ | |
8ca47e00 RR |
1533 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1534 | if (ipfd < 0) | |
1535 | err(1, "opening IP socket"); | |
1536 | ||
dde79789 | 1537 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1538 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1539 | arg += strlen(BRIDGE_PFX); |
1540 | bridging = true; | |
1541 | } | |
1542 | ||
1543 | /* A mac address may follow the bridge name or IP address */ | |
1544 | p = strchr(arg, ':'); | |
1545 | if (p) { | |
1546 | str2mac(p+1, conf.mac); | |
40c42076 | 1547 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1548 | *p = '\0'; |
dec6a2be MM |
1549 | } |
1550 | ||
1551 | /* arg is now either an IP address or a bridge name */ | |
1552 | if (bridging) | |
1553 | add_to_bridge(ipfd, tapif, arg); | |
1554 | else | |
8ca47e00 RR |
1555 | ip = str2ip(arg); |
1556 | ||
dec6a2be MM |
1557 | /* Set up the tun device. */ |
1558 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1559 | |
20887611 | 1560 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1561 | /* Expect Guest to handle everything except UFO */ |
1562 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1563 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1564 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1565 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1566 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1567 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1568 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1569 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e MM |
1570 | /* We handle indirect ring entries */ |
1571 | add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); | |
a586d4f6 | 1572 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1573 | |
a586d4f6 | 1574 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1575 | close(ipfd); |
1576 | ||
dec6a2be MM |
1577 | devices.device_num++; |
1578 | ||
1579 | if (bridging) | |
1580 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1581 | devices.device_num, tapif, arg); | |
1582 | else | |
1583 | verbose("device %u: tun %s: %s\n", | |
1584 | devices.device_num, tapif, arg); | |
8ca47e00 | 1585 | } |
a91d74a3 | 1586 | /*:*/ |
17cbca2b | 1587 | |
e1e72965 | 1588 | /* This hangs off device->priv. */ |
1842f23c | 1589 | struct vblk_info { |
17cbca2b RR |
1590 | /* The size of the file. */ |
1591 | off64_t len; | |
1592 | ||
1593 | /* The file descriptor for the file. */ | |
1594 | int fd; | |
1595 | ||
17cbca2b RR |
1596 | }; |
1597 | ||
e1e72965 RR |
1598 | /*L:210 |
1599 | * The Disk | |
1600 | * | |
a91d74a3 RR |
1601 | * The disk only has one virtqueue, so it only has one thread. It is really |
1602 | * simple: the Guest asks for a block number and we read or write that position | |
1603 | * in the file. | |
1604 | * | |
1605 | * Before we serviced each virtqueue in a separate thread, that was unacceptably | |
1606 | * slow: the Guest waits until the read is finished before running anything | |
1607 | * else, even if it could have been doing useful work. | |
1608 | * | |
1609 | * We could have used async I/O, except it's reputed to suck so hard that | |
1610 | * characters actually go missing from your code when you try to use it. | |
e1e72965 | 1611 | */ |
659a0e66 | 1612 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 1613 | { |
659a0e66 | 1614 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b RR |
1615 | unsigned int head, out_num, in_num, wlen; |
1616 | int ret; | |
cb38fa23 | 1617 | u8 *in; |
17cbca2b | 1618 | struct virtio_blk_outhdr *out; |
659a0e66 | 1619 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
1620 | off64_t off; |
1621 | ||
a91d74a3 RR |
1622 | /* |
1623 | * Get the next request, where we normally wait. It triggers the | |
1624 | * interrupt to acknowledge previously serviced requests (if any). | |
1625 | */ | |
659a0e66 | 1626 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
17cbca2b | 1627 | |
2e04ef76 RR |
1628 | /* |
1629 | * Every block request should contain at least one output buffer | |
e1e72965 | 1630 | * (detailing the location on disk and the type of request) and one |
2e04ef76 RR |
1631 | * input buffer (to hold the result). |
1632 | */ | |
17cbca2b RR |
1633 | if (out_num == 0 || in_num == 0) |
1634 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1635 | head, out_num, in_num); | |
1636 | ||
1637 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1638 | in = convert(&iov[out_num+in_num-1], u8); |
a91d74a3 RR |
1639 | /* |
1640 | * For historical reasons, block operations are expressed in 512 byte | |
1641 | * "sectors". | |
1642 | */ | |
17cbca2b RR |
1643 | off = out->sector * 512; |
1644 | ||
2e04ef76 RR |
1645 | /* |
1646 | * In general the virtio block driver is allowed to try SCSI commands. | |
1647 | * It'd be nice if we supported eject, for example, but we don't. | |
1648 | */ | |
17cbca2b RR |
1649 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1650 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1651 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1652 | wlen = sizeof(*in); |
17cbca2b | 1653 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
2e04ef76 RR |
1654 | /* |
1655 | * Write | |
1656 | * | |
1657 | * Move to the right location in the block file. This can fail | |
1658 | * if they try to write past end. | |
1659 | */ | |
17cbca2b RR |
1660 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
1661 | err(1, "Bad seek to sector %llu", out->sector); | |
1662 | ||
1663 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1664 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1665 | ||
2e04ef76 RR |
1666 | /* |
1667 | * Grr... Now we know how long the descriptor they sent was, we | |
17cbca2b | 1668 | * make sure they didn't try to write over the end of the block |
2e04ef76 RR |
1669 | * file (possibly extending it). |
1670 | */ | |
17cbca2b RR |
1671 | if (ret > 0 && off + ret > vblk->len) { |
1672 | /* Trim it back to the correct length */ | |
1673 | ftruncate64(vblk->fd, vblk->len); | |
1674 | /* Die, bad Guest, die. */ | |
1675 | errx(1, "Write past end %llu+%u", off, ret); | |
1676 | } | |
7bc9fdda TH |
1677 | |
1678 | wlen = sizeof(*in); | |
1679 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
1680 | } else if (out->type & VIRTIO_BLK_T_FLUSH) { | |
1681 | /* Flush */ | |
1682 | ret = fdatasync(vblk->fd); | |
1683 | verbose("FLUSH fdatasync: %i\n", ret); | |
1200e646 | 1684 | wlen = sizeof(*in); |
cb38fa23 | 1685 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b | 1686 | } else { |
2e04ef76 RR |
1687 | /* |
1688 | * Read | |
1689 | * | |
1690 | * Move to the right location in the block file. This can fail | |
1691 | * if they try to read past end. | |
1692 | */ | |
17cbca2b RR |
1693 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
1694 | err(1, "Bad seek to sector %llu", out->sector); | |
1695 | ||
1696 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1697 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1698 | if (ret >= 0) { | |
1200e646 | 1699 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1700 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1701 | } else { |
1200e646 | 1702 | wlen = sizeof(*in); |
cb38fa23 | 1703 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1704 | } |
1705 | } | |
1706 | ||
a91d74a3 | 1707 | /* Finished that request. */ |
38bc2b8c | 1708 | add_used(vq, head, wlen); |
17cbca2b RR |
1709 | } |
1710 | ||
e1e72965 | 1711 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1712 | static void setup_block_file(const char *filename) |
1713 | { | |
17cbca2b RR |
1714 | struct device *dev; |
1715 | struct vblk_info *vblk; | |
a586d4f6 | 1716 | struct virtio_blk_config conf; |
17cbca2b | 1717 | |
2e04ef76 | 1718 | /* Creat the device. */ |
659a0e66 | 1719 | dev = new_device("block", VIRTIO_ID_BLOCK); |
17cbca2b | 1720 | |
e1e72965 | 1721 | /* The device has one virtqueue, where the Guest places requests. */ |
659a0e66 | 1722 | add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); |
17cbca2b RR |
1723 | |
1724 | /* Allocate the room for our own bookkeeping */ | |
1725 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1726 | ||
1727 | /* First we open the file and store the length. */ | |
1728 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1729 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1730 | ||
7bc9fdda TH |
1731 | /* We support FLUSH. */ |
1732 | add_feature(dev, VIRTIO_BLK_F_FLUSH); | |
a586d4f6 | 1733 | |
17cbca2b | 1734 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1735 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b | 1736 | |
2e04ef76 RR |
1737 | /* |
1738 | * Tell Guest not to put in too many descriptors at once: two are used | |
1739 | * for the in and out elements. | |
1740 | */ | |
a586d4f6 RR |
1741 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1742 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1743 | ||
8ef562d1 RR |
1744 | /* Don't try to put whole struct: we have 8 bit limit. */ |
1745 | set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf); | |
17cbca2b | 1746 | |
17cbca2b | 1747 | verbose("device %u: virtblock %llu sectors\n", |
659a0e66 | 1748 | ++devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1749 | } |
28fd6d7f | 1750 | |
2e04ef76 RR |
1751 | /*L:211 |
1752 | * Our random number generator device reads from /dev/random into the Guest's | |
28fd6d7f RR |
1753 | * input buffers. The usual case is that the Guest doesn't want random numbers |
1754 | * and so has no buffers although /dev/random is still readable, whereas | |
1755 | * console is the reverse. | |
1756 | * | |
2e04ef76 RR |
1757 | * The same logic applies, however. |
1758 | */ | |
1759 | struct rng_info { | |
1760 | int rfd; | |
1761 | }; | |
1762 | ||
659a0e66 | 1763 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
1764 | { |
1765 | int len; | |
1766 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
1767 | struct rng_info *rng_info = vq->dev->priv; |
1768 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
1769 | |
1770 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 1771 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f RR |
1772 | if (out_num) |
1773 | errx(1, "Output buffers in rng?"); | |
1774 | ||
2e04ef76 | 1775 | /* |
a91d74a3 RR |
1776 | * Just like the console write, we loop to cover the whole iovec. |
1777 | * In this case, short reads actually happen quite a bit. | |
2e04ef76 | 1778 | */ |
28fd6d7f | 1779 | while (!iov_empty(iov, in_num)) { |
659a0e66 | 1780 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f RR |
1781 | if (len <= 0) |
1782 | err(1, "Read from /dev/random gave %i", len); | |
1783 | iov_consume(iov, in_num, len); | |
1784 | totlen += len; | |
1785 | } | |
1786 | ||
1787 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 1788 | add_used(vq, head, totlen); |
28fd6d7f RR |
1789 | } |
1790 | ||
2e04ef76 RR |
1791 | /*L:199 |
1792 | * This creates a "hardware" random number device for the Guest. | |
1793 | */ | |
28fd6d7f RR |
1794 | static void setup_rng(void) |
1795 | { | |
1796 | struct device *dev; | |
659a0e66 | 1797 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 1798 | |
2e04ef76 | 1799 | /* Our device's privat info simply contains the /dev/random fd. */ |
659a0e66 | 1800 | rng_info->rfd = open_or_die("/dev/random", O_RDONLY); |
28fd6d7f | 1801 | |
2e04ef76 | 1802 | /* Create the new device. */ |
659a0e66 RR |
1803 | dev = new_device("rng", VIRTIO_ID_RNG); |
1804 | dev->priv = rng_info; | |
28fd6d7f RR |
1805 | |
1806 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
659a0e66 | 1807 | add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); |
28fd6d7f RR |
1808 | |
1809 | verbose("device %u: rng\n", devices.device_num++); | |
1810 | } | |
a6bd8e13 | 1811 | /* That's the end of device setup. */ |
ec04b13f | 1812 | |
a6bd8e13 | 1813 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1814 | static void __attribute__((noreturn)) restart_guest(void) |
1815 | { | |
1816 | unsigned int i; | |
1817 | ||
2e04ef76 RR |
1818 | /* |
1819 | * Since we don't track all open fds, we simply close everything beyond | |
1820 | * stderr. | |
1821 | */ | |
ec04b13f BR |
1822 | for (i = 3; i < FD_SETSIZE; i++) |
1823 | close(i); | |
8c79873d | 1824 | |
659a0e66 RR |
1825 | /* Reset all the devices (kills all threads). */ |
1826 | cleanup_devices(); | |
1827 | ||
ec04b13f BR |
1828 | execv(main_args[0], main_args); |
1829 | err(1, "Could not exec %s", main_args[0]); | |
1830 | } | |
8ca47e00 | 1831 | |
2e04ef76 RR |
1832 | /*L:220 |
1833 | * Finally we reach the core of the Launcher which runs the Guest, serves | |
1834 | * its input and output, and finally, lays it to rest. | |
1835 | */ | |
56739c80 | 1836 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
1837 | { |
1838 | for (;;) { | |
17cbca2b | 1839 | unsigned long notify_addr; |
8ca47e00 RR |
1840 | int readval; |
1841 | ||
1842 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1843 | readval = pread(lguest_fd, ¬ify_addr, |
1844 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1845 | |
17cbca2b RR |
1846 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1847 | if (readval == sizeof(notify_addr)) { | |
1848 | verbose("Notify on address %#lx\n", notify_addr); | |
56739c80 | 1849 | handle_output(notify_addr); |
dde79789 | 1850 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1851 | } else if (errno == ENOENT) { |
1852 | char reason[1024] = { 0 }; | |
e3283fa0 | 1853 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1854 | errx(1, "%s", reason); |
ec04b13f BR |
1855 | /* ERESTART means that we need to reboot the guest */ |
1856 | } else if (errno == ERESTART) { | |
1857 | restart_guest(); | |
659a0e66 RR |
1858 | /* Anything else means a bug or incompatible change. */ |
1859 | } else | |
8ca47e00 | 1860 | err(1, "Running guest failed"); |
8ca47e00 RR |
1861 | } |
1862 | } | |
a6bd8e13 | 1863 | /*L:240 |
e1e72965 RR |
1864 | * This is the end of the Launcher. The good news: we are over halfway |
1865 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1866 | * of us. | |
dde79789 | 1867 | * |
e1e72965 RR |
1868 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1869 | * "make Host". | |
2e04ef76 | 1870 | :*/ |
8ca47e00 RR |
1871 | |
1872 | static struct option opts[] = { | |
1873 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1874 | { "tunnet", 1, NULL, 't' }, |
1875 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1876 | { "rng", 0, NULL, 'r' }, |
8ca47e00 | 1877 | { "initrd", 1, NULL, 'i' }, |
8aeb36e8 PS |
1878 | { "username", 1, NULL, 'u' }, |
1879 | { "chroot", 1, NULL, 'c' }, | |
8ca47e00 RR |
1880 | { NULL }, |
1881 | }; | |
1882 | static void usage(void) | |
1883 | { | |
1884 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1885 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1886 | "|--block=<filename>|--initrd=<filename>]...\n" |
1887 | "<mem-in-mb> vmlinux [args...]"); | |
1888 | } | |
1889 | ||
3c6b5bfa | 1890 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1891 | int main(int argc, char *argv[]) |
1892 | { | |
2e04ef76 | 1893 | /* Memory, code startpoint and size of the (optional) initrd. */ |
58a24566 | 1894 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
1895 | /* Two temporaries. */ |
1896 | int i, c; | |
3c6b5bfa | 1897 | /* The boot information for the Guest. */ |
43d33b21 | 1898 | struct boot_params *boot; |
dde79789 | 1899 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1900 | const char *initrd_name = NULL; |
1901 | ||
8aeb36e8 PS |
1902 | /* Password structure for initgroups/setres[gu]id */ |
1903 | struct passwd *user_details = NULL; | |
1904 | ||
1905 | /* Directory to chroot to */ | |
1906 | char *chroot_path = NULL; | |
1907 | ||
ec04b13f BR |
1908 | /* Save the args: we "reboot" by execing ourselves again. */ |
1909 | main_args = argv; | |
ec04b13f | 1910 | |
2e04ef76 RR |
1911 | /* |
1912 | * First we initialize the device list. We keep a pointer to the last | |
659a0e66 | 1913 | * device, and the next interrupt number to use for devices (1: |
2e04ef76 RR |
1914 | * remember that 0 is used by the timer). |
1915 | */ | |
a586d4f6 | 1916 | devices.lastdev = NULL; |
17cbca2b | 1917 | devices.next_irq = 1; |
8ca47e00 | 1918 | |
a91d74a3 | 1919 | /* We're CPU 0. In fact, that's the only CPU possible right now. */ |
e3283fa0 | 1920 | cpu_id = 0; |
a91d74a3 | 1921 | |
2e04ef76 RR |
1922 | /* |
1923 | * We need to know how much memory so we can set up the device | |
dde79789 RR |
1924 | * descriptor and memory pages for the devices as we parse the command |
1925 | * line. So we quickly look through the arguments to find the amount | |
2e04ef76 RR |
1926 | * of memory now. |
1927 | */ | |
6570c459 RR |
1928 | for (i = 1; i < argc; i++) { |
1929 | if (argv[i][0] != '-') { | |
3c6b5bfa | 1930 | mem = atoi(argv[i]) * 1024 * 1024; |
2e04ef76 RR |
1931 | /* |
1932 | * We start by mapping anonymous pages over all of | |
3c6b5bfa RR |
1933 | * guest-physical memory range. This fills it with 0, |
1934 | * and ensures that the Guest won't be killed when it | |
2e04ef76 RR |
1935 | * tries to access it. |
1936 | */ | |
3c6b5bfa RR |
1937 | guest_base = map_zeroed_pages(mem / getpagesize() |
1938 | + DEVICE_PAGES); | |
1939 | guest_limit = mem; | |
1940 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1941 | devices.descpage = get_pages(1); |
6570c459 RR |
1942 | break; |
1943 | } | |
1944 | } | |
dde79789 RR |
1945 | |
1946 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1947 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1948 | switch (c) { | |
1949 | case 'v': | |
1950 | verbose = true; | |
1951 | break; | |
8ca47e00 | 1952 | case 't': |
17cbca2b | 1953 | setup_tun_net(optarg); |
8ca47e00 RR |
1954 | break; |
1955 | case 'b': | |
17cbca2b | 1956 | setup_block_file(optarg); |
8ca47e00 | 1957 | break; |
28fd6d7f RR |
1958 | case 'r': |
1959 | setup_rng(); | |
1960 | break; | |
8ca47e00 RR |
1961 | case 'i': |
1962 | initrd_name = optarg; | |
1963 | break; | |
8aeb36e8 PS |
1964 | case 'u': |
1965 | user_details = getpwnam(optarg); | |
1966 | if (!user_details) | |
1967 | err(1, "getpwnam failed, incorrect username?"); | |
1968 | break; | |
1969 | case 'c': | |
1970 | chroot_path = optarg; | |
1971 | break; | |
8ca47e00 RR |
1972 | default: |
1973 | warnx("Unknown argument %s", argv[optind]); | |
1974 | usage(); | |
1975 | } | |
1976 | } | |
2e04ef76 RR |
1977 | /* |
1978 | * After the other arguments we expect memory and kernel image name, | |
1979 | * followed by command line arguments for the kernel. | |
1980 | */ | |
8ca47e00 RR |
1981 | if (optind + 2 > argc) |
1982 | usage(); | |
1983 | ||
3c6b5bfa RR |
1984 | verbose("Guest base is at %p\n", guest_base); |
1985 | ||
dde79789 | 1986 | /* We always have a console device */ |
17cbca2b | 1987 | setup_console(); |
8ca47e00 | 1988 | |
8ca47e00 | 1989 | /* Now we load the kernel */ |
47436aa4 | 1990 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1991 | |
3c6b5bfa RR |
1992 | /* Boot information is stashed at physical address 0 */ |
1993 | boot = from_guest_phys(0); | |
1994 | ||
dde79789 | 1995 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1996 | if (initrd_name) { |
1997 | initrd_size = load_initrd(initrd_name, mem); | |
2e04ef76 RR |
1998 | /* |
1999 | * These are the location in the Linux boot header where the | |
2000 | * start and size of the initrd are expected to be found. | |
2001 | */ | |
43d33b21 RR |
2002 | boot->hdr.ramdisk_image = mem - initrd_size; |
2003 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 2004 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 2005 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
2006 | } |
2007 | ||
2e04ef76 RR |
2008 | /* |
2009 | * The Linux boot header contains an "E820" memory map: ours is a | |
2010 | * simple, single region. | |
2011 | */ | |
43d33b21 RR |
2012 | boot->e820_entries = 1; |
2013 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
2e04ef76 RR |
2014 | /* |
2015 | * The boot header contains a command line pointer: we put the command | |
2016 | * line after the boot header. | |
2017 | */ | |
43d33b21 | 2018 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
e1e72965 | 2019 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 2020 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 2021 | |
814a0e5c | 2022 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 2023 | boot->hdr.version = 0x207; |
814a0e5c RR |
2024 | |
2025 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 2026 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 2027 | |
43d33b21 RR |
2028 | /* Tell the entry path not to try to reload segment registers. */ |
2029 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 2030 | |
2e04ef76 RR |
2031 | /* |
2032 | * We tell the kernel to initialize the Guest: this returns the open | |
2033 | * /dev/lguest file descriptor. | |
2034 | */ | |
56739c80 | 2035 | tell_kernel(start); |
dde79789 | 2036 | |
a91d74a3 | 2037 | /* Ensure that we terminate if a device-servicing child dies. */ |
659a0e66 RR |
2038 | signal(SIGCHLD, kill_launcher); |
2039 | ||
2040 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
2041 | atexit(cleanup_devices); | |
8ca47e00 | 2042 | |
8aeb36e8 PS |
2043 | /* If requested, chroot to a directory */ |
2044 | if (chroot_path) { | |
2045 | if (chroot(chroot_path) != 0) | |
2046 | err(1, "chroot(\"%s\") failed", chroot_path); | |
2047 | ||
2048 | if (chdir("/") != 0) | |
2049 | err(1, "chdir(\"/\") failed"); | |
2050 | ||
2051 | verbose("chroot done\n"); | |
2052 | } | |
2053 | ||
2054 | /* If requested, drop privileges */ | |
2055 | if (user_details) { | |
2056 | uid_t u; | |
2057 | gid_t g; | |
2058 | ||
2059 | u = user_details->pw_uid; | |
2060 | g = user_details->pw_gid; | |
2061 | ||
2062 | if (initgroups(user_details->pw_name, g) != 0) | |
2063 | err(1, "initgroups failed"); | |
2064 | ||
2065 | if (setresgid(g, g, g) != 0) | |
2066 | err(1, "setresgid failed"); | |
2067 | ||
2068 | if (setresuid(u, u, u) != 0) | |
2069 | err(1, "setresuid failed"); | |
2070 | ||
2071 | verbose("Dropping privileges completed\n"); | |
2072 | } | |
2073 | ||
dde79789 | 2074 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 2075 | run_guest(); |
8ca47e00 | 2076 | } |
f56a384e RR |
2077 | /*:*/ |
2078 | ||
2079 | /*M:999 | |
2080 | * Mastery is done: you now know everything I do. | |
2081 | * | |
2082 | * But surely you have seen code, features and bugs in your wanderings which | |
2083 | * you now yearn to attack? That is the real game, and I look forward to you | |
2084 | * patching and forking lguest into the Your-Name-Here-visor. | |
2085 | * | |
2086 | * Farewell, and good coding! | |
2087 | * Rusty Russell. | |
2088 | */ |