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 | ||
927cfb97 RR |
45 | #ifndef VIRTIO_F_ANY_LAYOUT |
46 | #define VIRTIO_F_ANY_LAYOUT 27 | |
47 | #endif | |
48 | ||
2e04ef76 | 49 | /*L:110 |
9f54288d | 50 | * We can ignore the 43 include files we need for this program, but I do want |
2e04ef76 | 51 | * to draw attention to the use of kernel-style types. |
db24e8c2 RR |
52 | * |
53 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
54 | * like these abbreviations, so we define them here. Note that u64 is always | |
55 | * unsigned long long, which works on all Linux systems: this means that we can | |
2e04ef76 RR |
56 | * use %llu in printf for any u64. |
57 | */ | |
db24e8c2 RR |
58 | typedef unsigned long long u64; |
59 | typedef uint32_t u32; | |
60 | typedef uint16_t u16; | |
61 | typedef uint8_t u8; | |
dde79789 | 62 | /*:*/ |
8ca47e00 | 63 | |
e6dc0418 RR |
64 | #include <linux/virtio_config.h> |
65 | #include <linux/virtio_net.h> | |
66 | #include <linux/virtio_blk.h> | |
67 | #include <linux/virtio_console.h> | |
68 | #include <linux/virtio_rng.h> | |
69 | #include <linux/virtio_ring.h> | |
70 | #include <asm/bootparam.h> | |
71 | #include "../../include/linux/lguest_launcher.h" | |
72 | ||
8ca47e00 RR |
73 | #define BRIDGE_PFX "bridge:" |
74 | #ifndef SIOCBRADDIF | |
75 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
76 | #endif | |
3c6b5bfa RR |
77 | /* We can have up to 256 pages for devices. */ |
78 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
79 | /* This will occupy 3 pages: it must be a power of 2. */ |
80 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 81 | |
2e04ef76 RR |
82 | /*L:120 |
83 | * verbose is both a global flag and a macro. The C preprocessor allows | |
84 | * this, and although I wouldn't recommend it, it works quite nicely here. | |
85 | */ | |
8ca47e00 RR |
86 | static bool verbose; |
87 | #define verbose(args...) \ | |
88 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
89 | /*:*/ |
90 | ||
3c6b5bfa RR |
91 | /* The pointer to the start of guest memory. */ |
92 | static void *guest_base; | |
93 | /* The maximum guest physical address allowed, and maximum possible. */ | |
94 | static unsigned long guest_limit, guest_max; | |
56739c80 RR |
95 | /* The /dev/lguest file descriptor. */ |
96 | static int lguest_fd; | |
8ca47e00 | 97 | |
e3283fa0 GOC |
98 | /* a per-cpu variable indicating whose vcpu is currently running */ |
99 | static unsigned int __thread cpu_id; | |
100 | ||
dde79789 | 101 | /* This is our list of devices. */ |
1842f23c | 102 | struct device_list { |
17cbca2b RR |
103 | /* Counter to assign interrupt numbers. */ |
104 | unsigned int next_irq; | |
105 | ||
106 | /* Counter to print out convenient device numbers. */ | |
107 | unsigned int device_num; | |
108 | ||
dde79789 | 109 | /* The descriptor page for the devices. */ |
17cbca2b RR |
110 | u8 *descpage; |
111 | ||
dde79789 | 112 | /* A single linked list of devices. */ |
8ca47e00 | 113 | struct device *dev; |
2e04ef76 | 114 | /* And a pointer to the last device for easy append. */ |
a586d4f6 | 115 | struct device *lastdev; |
8ca47e00 RR |
116 | }; |
117 | ||
17cbca2b RR |
118 | /* The list of Guest devices, based on command line arguments. */ |
119 | static struct device_list devices; | |
120 | ||
dde79789 | 121 | /* The device structure describes a single device. */ |
1842f23c | 122 | struct device { |
dde79789 | 123 | /* The linked-list pointer. */ |
8ca47e00 | 124 | struct device *next; |
17cbca2b | 125 | |
713b15b3 | 126 | /* The device's descriptor, as mapped into the Guest. */ |
8ca47e00 | 127 | struct lguest_device_desc *desc; |
17cbca2b | 128 | |
713b15b3 RR |
129 | /* We can't trust desc values once Guest has booted: we use these. */ |
130 | unsigned int feature_len; | |
131 | unsigned int num_vq; | |
132 | ||
17cbca2b RR |
133 | /* The name of this device, for --verbose. */ |
134 | const char *name; | |
8ca47e00 | 135 | |
17cbca2b RR |
136 | /* Any queues attached to this device */ |
137 | struct virtqueue *vq; | |
8ca47e00 | 138 | |
659a0e66 RR |
139 | /* Is it operational */ |
140 | bool running; | |
a007a751 | 141 | |
8ca47e00 RR |
142 | /* Device-specific data. */ |
143 | void *priv; | |
144 | }; | |
145 | ||
17cbca2b | 146 | /* The virtqueue structure describes a queue attached to a device. */ |
1842f23c | 147 | struct virtqueue { |
17cbca2b RR |
148 | struct virtqueue *next; |
149 | ||
150 | /* Which device owns me. */ | |
151 | struct device *dev; | |
152 | ||
153 | /* The configuration for this queue. */ | |
154 | struct lguest_vqconfig config; | |
155 | ||
156 | /* The actual ring of buffers. */ | |
157 | struct vring vring; | |
158 | ||
159 | /* Last available index we saw. */ | |
160 | u16 last_avail_idx; | |
161 | ||
95c517c0 RR |
162 | /* How many are used since we sent last irq? */ |
163 | unsigned int pending_used; | |
164 | ||
659a0e66 RR |
165 | /* Eventfd where Guest notifications arrive. */ |
166 | int eventfd; | |
20887611 | 167 | |
659a0e66 RR |
168 | /* Function for the thread which is servicing this virtqueue. */ |
169 | void (*service)(struct virtqueue *vq); | |
170 | pid_t thread; | |
17cbca2b RR |
171 | }; |
172 | ||
ec04b13f BR |
173 | /* Remember the arguments to the program so we can "reboot" */ |
174 | static char **main_args; | |
175 | ||
659a0e66 RR |
176 | /* The original tty settings to restore on exit. */ |
177 | static struct termios orig_term; | |
178 | ||
2e04ef76 RR |
179 | /* |
180 | * We have to be careful with barriers: our devices are all run in separate | |
f7027c63 | 181 | * threads and so we need to make sure that changes visible to the Guest happen |
2e04ef76 RR |
182 | * in precise order. |
183 | */ | |
f7027c63 | 184 | #define wmb() __asm__ __volatile__("" : : : "memory") |
0d69a65e RR |
185 | #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") |
186 | #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") | |
17cbca2b | 187 | |
b5111790 RR |
188 | /* Wrapper for the last available index. Makes it easier to change. */ |
189 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
190 | ||
2e04ef76 RR |
191 | /* |
192 | * The virtio configuration space is defined to be little-endian. x86 is | |
193 | * little-endian too, but it's nice to be explicit so we have these helpers. | |
194 | */ | |
17cbca2b RR |
195 | #define cpu_to_le16(v16) (v16) |
196 | #define cpu_to_le32(v32) (v32) | |
197 | #define cpu_to_le64(v64) (v64) | |
198 | #define le16_to_cpu(v16) (v16) | |
199 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 200 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 201 | |
28fd6d7f RR |
202 | /* Is this iovec empty? */ |
203 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
204 | { | |
205 | unsigned int i; | |
206 | ||
207 | for (i = 0; i < num_iov; i++) | |
208 | if (iov[i].iov_len) | |
209 | return false; | |
210 | return true; | |
211 | } | |
212 | ||
213 | /* Take len bytes from the front of this iovec. */ | |
c0316a94 RR |
214 | static void iov_consume(struct iovec iov[], unsigned num_iov, |
215 | void *dest, unsigned len) | |
28fd6d7f RR |
216 | { |
217 | unsigned int i; | |
218 | ||
219 | for (i = 0; i < num_iov; i++) { | |
220 | unsigned int used; | |
221 | ||
222 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
c0316a94 RR |
223 | if (dest) { |
224 | memcpy(dest, iov[i].iov_base, used); | |
225 | dest += used; | |
226 | } | |
28fd6d7f RR |
227 | iov[i].iov_base += used; |
228 | iov[i].iov_len -= used; | |
229 | len -= used; | |
230 | } | |
c0316a94 RR |
231 | if (len != 0) |
232 | errx(1, "iovec too short!"); | |
28fd6d7f RR |
233 | } |
234 | ||
6e5aa7ef RR |
235 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
236 | static u8 *get_feature_bits(struct device *dev) | |
237 | { | |
238 | return (u8 *)(dev->desc + 1) | |
713b15b3 | 239 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
6e5aa7ef RR |
240 | } |
241 | ||
2e04ef76 RR |
242 | /*L:100 |
243 | * The Launcher code itself takes us out into userspace, that scary place where | |
244 | * pointers run wild and free! Unfortunately, like most userspace programs, | |
245 | * it's quite boring (which is why everyone likes to hack on the kernel!). | |
246 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get | |
247 | * you through this section. Or, maybe not. | |
3c6b5bfa RR |
248 | * |
249 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
250 | * memory and stores it in "guest_base". In other words, Guest physical == | |
251 | * Launcher virtual with an offset. | |
252 | * | |
253 | * This can be tough to get your head around, but usually it just means that we | |
a33f3224 | 254 | * use these trivial conversion functions when the Guest gives us its |
2e04ef76 RR |
255 | * "physical" addresses: |
256 | */ | |
3c6b5bfa RR |
257 | static void *from_guest_phys(unsigned long addr) |
258 | { | |
259 | return guest_base + addr; | |
260 | } | |
261 | ||
262 | static unsigned long to_guest_phys(const void *addr) | |
263 | { | |
264 | return (addr - guest_base); | |
265 | } | |
266 | ||
dde79789 RR |
267 | /*L:130 |
268 | * Loading the Kernel. | |
269 | * | |
270 | * We start with couple of simple helper routines. open_or_die() avoids | |
2e04ef76 RR |
271 | * error-checking code cluttering the callers: |
272 | */ | |
8ca47e00 RR |
273 | static int open_or_die(const char *name, int flags) |
274 | { | |
275 | int fd = open(name, flags); | |
276 | if (fd < 0) | |
277 | err(1, "Failed to open %s", name); | |
278 | return fd; | |
279 | } | |
280 | ||
3c6b5bfa RR |
281 | /* map_zeroed_pages() takes a number of pages. */ |
282 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 283 | { |
3c6b5bfa RR |
284 | int fd = open_or_die("/dev/zero", O_RDONLY); |
285 | void *addr; | |
8ca47e00 | 286 | |
2e04ef76 RR |
287 | /* |
288 | * We use a private mapping (ie. if we write to the page, it will be | |
5230ff0c PS |
289 | * copied). We allocate an extra two pages PROT_NONE to act as guard |
290 | * pages against read/write attempts that exceed allocated space. | |
2e04ef76 | 291 | */ |
5230ff0c PS |
292 | addr = mmap(NULL, getpagesize() * (num+2), |
293 | PROT_NONE, MAP_PRIVATE, fd, 0); | |
294 | ||
3c6b5bfa | 295 | if (addr == MAP_FAILED) |
af901ca1 | 296 | err(1, "Mmapping %u pages of /dev/zero", num); |
a91d74a3 | 297 | |
5230ff0c PS |
298 | if (mprotect(addr + getpagesize(), getpagesize() * num, |
299 | PROT_READ|PROT_WRITE) == -1) | |
300 | err(1, "mprotect rw %u pages failed", num); | |
301 | ||
a91d74a3 RR |
302 | /* |
303 | * One neat mmap feature is that you can close the fd, and it | |
304 | * stays mapped. | |
305 | */ | |
34bdaab4 | 306 | close(fd); |
3c6b5bfa | 307 | |
5230ff0c PS |
308 | /* Return address after PROT_NONE page */ |
309 | return addr + getpagesize(); | |
3c6b5bfa RR |
310 | } |
311 | ||
312 | /* Get some more pages for a device. */ | |
313 | static void *get_pages(unsigned int num) | |
314 | { | |
315 | void *addr = from_guest_phys(guest_limit); | |
316 | ||
317 | guest_limit += num * getpagesize(); | |
318 | if (guest_limit > guest_max) | |
319 | errx(1, "Not enough memory for devices"); | |
320 | return addr; | |
8ca47e00 RR |
321 | } |
322 | ||
2e04ef76 RR |
323 | /* |
324 | * This routine is used to load the kernel or initrd. It tries mmap, but if | |
6649bb7a | 325 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
2e04ef76 RR |
326 | * it falls back to reading the memory in. |
327 | */ | |
6649bb7a RM |
328 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
329 | { | |
330 | ssize_t r; | |
331 | ||
2e04ef76 RR |
332 | /* |
333 | * We map writable even though for some segments are marked read-only. | |
6649bb7a RM |
334 | * The kernel really wants to be writable: it patches its own |
335 | * instructions. | |
336 | * | |
337 | * MAP_PRIVATE means that the page won't be copied until a write is | |
338 | * done to it. This allows us to share untouched memory between | |
2e04ef76 RR |
339 | * Guests. |
340 | */ | |
5230ff0c | 341 | if (mmap(addr, len, PROT_READ|PROT_WRITE, |
6649bb7a RM |
342 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) |
343 | return; | |
344 | ||
345 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
346 | r = pread(fd, addr, len, offset); | |
347 | if (r != len) | |
348 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
349 | } | |
350 | ||
2e04ef76 RR |
351 | /* |
352 | * This routine takes an open vmlinux image, which is in ELF, and maps it into | |
dde79789 RR |
353 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
354 | * by all modern binaries on Linux including the kernel. | |
355 | * | |
356 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
357 | * address. We use the physical address; the Guest will map itself to the |
358 | * virtual address. | |
dde79789 | 359 | * |
2e04ef76 RR |
360 | * We return the starting address. |
361 | */ | |
47436aa4 | 362 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 363 | { |
8ca47e00 RR |
364 | Elf32_Phdr phdr[ehdr->e_phnum]; |
365 | unsigned int i; | |
8ca47e00 | 366 | |
2e04ef76 RR |
367 | /* |
368 | * Sanity checks on the main ELF header: an x86 executable with a | |
369 | * reasonable number of correctly-sized program headers. | |
370 | */ | |
8ca47e00 RR |
371 | if (ehdr->e_type != ET_EXEC |
372 | || ehdr->e_machine != EM_386 | |
373 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
374 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
375 | errx(1, "Malformed elf header"); | |
376 | ||
2e04ef76 RR |
377 | /* |
378 | * An ELF executable contains an ELF header and a number of "program" | |
dde79789 | 379 | * headers which indicate which parts ("segments") of the program to |
2e04ef76 RR |
380 | * load where. |
381 | */ | |
dde79789 RR |
382 | |
383 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
384 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
385 | err(1, "Seeking to program headers"); | |
386 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
387 | err(1, "Reading program headers"); | |
388 | ||
2e04ef76 RR |
389 | /* |
390 | * Try all the headers: there are usually only three. A read-only one, | |
391 | * a read-write one, and a "note" section which we don't load. | |
392 | */ | |
8ca47e00 | 393 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 394 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
395 | if (phdr[i].p_type != PT_LOAD) |
396 | continue; | |
397 | ||
398 | verbose("Section %i: size %i addr %p\n", | |
399 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
400 | ||
6649bb7a | 401 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 402 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 403 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
404 | } |
405 | ||
814a0e5c RR |
406 | /* The entry point is given in the ELF header. */ |
407 | return ehdr->e_entry; | |
8ca47e00 RR |
408 | } |
409 | ||
2e04ef76 RR |
410 | /*L:150 |
411 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed | |
412 | * to jump into it and it will unpack itself. We used to have to perform some | |
413 | * hairy magic because the unpacking code scared me. | |
dde79789 | 414 | * |
5bbf89fc RR |
415 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
416 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
2e04ef76 RR |
417 | * the funky header so we know where in the file to load, and away we go! |
418 | */ | |
47436aa4 | 419 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 420 | { |
43d33b21 | 421 | struct boot_params boot; |
5bbf89fc RR |
422 | int r; |
423 | /* Modern bzImages get loaded at 1M. */ | |
424 | void *p = from_guest_phys(0x100000); | |
425 | ||
2e04ef76 RR |
426 | /* |
427 | * Go back to the start of the file and read the header. It should be | |
395cf969 | 428 | * a Linux boot header (see Documentation/x86/boot.txt) |
2e04ef76 | 429 | */ |
5bbf89fc | 430 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 431 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 432 | |
43d33b21 RR |
433 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
434 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
435 | errx(1, "This doesn't look like a bzImage to me"); |
436 | ||
43d33b21 RR |
437 | /* Skip over the extra sectors of the header. */ |
438 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
439 | |
440 | /* Now read everything into memory. in nice big chunks. */ | |
441 | while ((r = read(fd, p, 65536)) > 0) | |
442 | p += r; | |
443 | ||
43d33b21 RR |
444 | /* Finally, code32_start tells us where to enter the kernel. */ |
445 | return boot.hdr.code32_start; | |
8ca47e00 RR |
446 | } |
447 | ||
2e04ef76 RR |
448 | /*L:140 |
449 | * Loading the kernel is easy when it's a "vmlinux", but most kernels | |
e1e72965 | 450 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
2e04ef76 RR |
451 | * work, we can load those, too. |
452 | */ | |
47436aa4 | 453 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
454 | { |
455 | Elf32_Ehdr hdr; | |
456 | ||
dde79789 | 457 | /* Read in the first few bytes. */ |
8ca47e00 RR |
458 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
459 | err(1, "Reading kernel"); | |
460 | ||
dde79789 | 461 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 462 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 463 | return map_elf(fd, &hdr); |
8ca47e00 | 464 | |
a6bd8e13 | 465 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 466 | return load_bzimage(fd); |
8ca47e00 RR |
467 | } |
468 | ||
2e04ef76 RR |
469 | /* |
470 | * This is a trivial little helper to align pages. Andi Kleen hated it because | |
dde79789 RR |
471 | * it calls getpagesize() twice: "it's dumb code." |
472 | * | |
473 | * Kernel guys get really het up about optimization, even when it's not | |
2e04ef76 RR |
474 | * necessary. I leave this code as a reaction against that. |
475 | */ | |
8ca47e00 RR |
476 | static inline unsigned long page_align(unsigned long addr) |
477 | { | |
dde79789 | 478 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
479 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
480 | } | |
481 | ||
2e04ef76 RR |
482 | /*L:180 |
483 | * An "initial ram disk" is a disk image loaded into memory along with the | |
484 | * kernel which the kernel can use to boot from without needing any drivers. | |
485 | * Most distributions now use this as standard: the initrd contains the code to | |
486 | * load the appropriate driver modules for the current machine. | |
dde79789 RR |
487 | * |
488 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
2e04ef76 RR |
489 | * kernels. He sent me this (and tells me when I break it). |
490 | */ | |
8ca47e00 RR |
491 | static unsigned long load_initrd(const char *name, unsigned long mem) |
492 | { | |
493 | int ifd; | |
494 | struct stat st; | |
495 | unsigned long len; | |
8ca47e00 RR |
496 | |
497 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 498 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
499 | if (fstat(ifd, &st) < 0) |
500 | err(1, "fstat() on initrd '%s'", name); | |
501 | ||
2e04ef76 RR |
502 | /* |
503 | * We map the initrd at the top of memory, but mmap wants it to be | |
504 | * page-aligned, so we round the size up for that. | |
505 | */ | |
8ca47e00 | 506 | len = page_align(st.st_size); |
3c6b5bfa | 507 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
2e04ef76 RR |
508 | /* |
509 | * Once a file is mapped, you can close the file descriptor. It's a | |
510 | * little odd, but quite useful. | |
511 | */ | |
8ca47e00 | 512 | close(ifd); |
6649bb7a | 513 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
514 | |
515 | /* We return the initrd size. */ | |
8ca47e00 RR |
516 | return len; |
517 | } | |
e1e72965 | 518 | /*:*/ |
8ca47e00 | 519 | |
2e04ef76 RR |
520 | /* |
521 | * Simple routine to roll all the commandline arguments together with spaces | |
522 | * between them. | |
523 | */ | |
8ca47e00 RR |
524 | static void concat(char *dst, char *args[]) |
525 | { | |
526 | unsigned int i, len = 0; | |
527 | ||
528 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
529 | if (i) { |
530 | strcat(dst+len, " "); | |
531 | len++; | |
532 | } | |
8ca47e00 | 533 | strcpy(dst+len, args[i]); |
1ef36fa6 | 534 | len += strlen(args[i]); |
8ca47e00 RR |
535 | } |
536 | /* In case it's empty. */ | |
537 | dst[len] = '\0'; | |
538 | } | |
539 | ||
2e04ef76 RR |
540 | /*L:185 |
541 | * This is where we actually tell the kernel to initialize the Guest. We | |
e1e72965 | 542 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
58a24566 | 543 | * the base of Guest "physical" memory, the top physical page to allow and the |
2e04ef76 RR |
544 | * entry point for the Guest. |
545 | */ | |
56739c80 | 546 | static void tell_kernel(unsigned long start) |
8ca47e00 | 547 | { |
511801dc JS |
548 | unsigned long args[] = { LHREQ_INITIALIZE, |
549 | (unsigned long)guest_base, | |
58a24566 | 550 | guest_limit / getpagesize(), start }; |
3c6b5bfa RR |
551 | verbose("Guest: %p - %p (%#lx)\n", |
552 | guest_base, guest_base + guest_limit, guest_limit); | |
56739c80 RR |
553 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
554 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 555 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 556 | } |
dde79789 | 557 | /*:*/ |
8ca47e00 | 558 | |
a91d74a3 | 559 | /*L:200 |
dde79789 RR |
560 | * Device Handling. |
561 | * | |
e1e72965 | 562 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 563 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 564 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
565 | * if something funny is going on: |
566 | */ | |
8ca47e00 RR |
567 | static void *_check_pointer(unsigned long addr, unsigned int size, |
568 | unsigned int line) | |
569 | { | |
2e04ef76 | 570 | /* |
5230ff0c PS |
571 | * Check if the requested address and size exceeds the allocated memory, |
572 | * or addr + size wraps around. | |
2e04ef76 | 573 | */ |
5230ff0c | 574 | if ((addr + size) > guest_limit || (addr + size) < addr) |
17cbca2b | 575 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
2e04ef76 RR |
576 | /* |
577 | * We return a pointer for the caller's convenience, now we know it's | |
578 | * safe to use. | |
579 | */ | |
3c6b5bfa | 580 | return from_guest_phys(addr); |
8ca47e00 | 581 | } |
dde79789 | 582 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
583 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
584 | ||
2e04ef76 RR |
585 | /* |
586 | * Each buffer in the virtqueues is actually a chain of descriptors. This | |
e1e72965 | 587 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
2e04ef76 RR |
588 | * at the end. |
589 | */ | |
d1f0132e MM |
590 | static unsigned next_desc(struct vring_desc *desc, |
591 | unsigned int i, unsigned int max) | |
17cbca2b RR |
592 | { |
593 | unsigned int next; | |
594 | ||
595 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
596 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
597 | return max; | |
17cbca2b RR |
598 | |
599 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 600 | next = desc[i].next; |
17cbca2b RR |
601 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
602 | wmb(); | |
603 | ||
d1f0132e | 604 | if (next >= max) |
17cbca2b RR |
605 | errx(1, "Desc next is %u", next); |
606 | ||
607 | return next; | |
608 | } | |
609 | ||
a91d74a3 RR |
610 | /* |
611 | * This actually sends the interrupt for this virtqueue, if we've used a | |
612 | * buffer. | |
613 | */ | |
38bc2b8c RR |
614 | static void trigger_irq(struct virtqueue *vq) |
615 | { | |
616 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; | |
617 | ||
95c517c0 RR |
618 | /* Don't inform them if nothing used. */ |
619 | if (!vq->pending_used) | |
620 | return; | |
621 | vq->pending_used = 0; | |
622 | ||
ca60a42c RR |
623 | /* If they don't want an interrupt, don't send one... */ |
624 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) { | |
990c91f0 | 625 | return; |
ca60a42c | 626 | } |
38bc2b8c RR |
627 | |
628 | /* Send the Guest an interrupt tell them we used something up. */ | |
629 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
630 | err(1, "Triggering irq %i", vq->config.irq); | |
631 | } | |
632 | ||
2e04ef76 | 633 | /* |
a91d74a3 | 634 | * This looks in the virtqueue for the first available buffer, and converts |
17cbca2b RR |
635 | * it to an iovec for convenient access. Since descriptors consist of some |
636 | * number of output then some number of input descriptors, it's actually two | |
637 | * iovecs, but we pack them into one and note how many of each there were. | |
638 | * | |
a91d74a3 | 639 | * This function waits if necessary, and returns the descriptor number found. |
2e04ef76 | 640 | */ |
659a0e66 RR |
641 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
642 | struct iovec iov[], | |
643 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 644 | { |
d1f0132e MM |
645 | unsigned int i, head, max; |
646 | struct vring_desc *desc; | |
659a0e66 RR |
647 | u16 last_avail = lg_last_avail(vq); |
648 | ||
a91d74a3 | 649 | /* There's nothing available? */ |
659a0e66 RR |
650 | while (last_avail == vq->vring.avail->idx) { |
651 | u64 event; | |
652 | ||
a91d74a3 RR |
653 | /* |
654 | * Since we're about to sleep, now is a good time to tell the | |
655 | * Guest about what we've used up to now. | |
656 | */ | |
38bc2b8c RR |
657 | trigger_irq(vq); |
658 | ||
b60da13f RR |
659 | /* OK, now we need to know about added descriptors. */ |
660 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
661 | ||
2e04ef76 RR |
662 | /* |
663 | * They could have slipped one in as we were doing that: make | |
664 | * sure it's written, then check again. | |
665 | */ | |
b60da13f RR |
666 | mb(); |
667 | if (last_avail != vq->vring.avail->idx) { | |
668 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
669 | break; | |
670 | } | |
671 | ||
659a0e66 RR |
672 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
673 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
674 | errx(1, "Event read failed?"); | |
b60da13f RR |
675 | |
676 | /* We don't need to be notified again. */ | |
677 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 678 | } |
17cbca2b RR |
679 | |
680 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 681 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
17cbca2b | 682 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 683 | last_avail, vq->vring.avail->idx); |
17cbca2b | 684 | |
8fd9a636 RR |
685 | /* |
686 | * Make sure we read the descriptor number *after* we read the ring | |
687 | * update; don't let the cpu or compiler change the order. | |
688 | */ | |
689 | rmb(); | |
690 | ||
2e04ef76 RR |
691 | /* |
692 | * Grab the next descriptor number they're advertising, and increment | |
693 | * the index we've seen. | |
694 | */ | |
b5111790 RR |
695 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
696 | lg_last_avail(vq)++; | |
17cbca2b RR |
697 | |
698 | /* If their number is silly, that's a fatal mistake. */ | |
699 | if (head >= vq->vring.num) | |
700 | errx(1, "Guest says index %u is available", head); | |
701 | ||
702 | /* When we start there are none of either input nor output. */ | |
703 | *out_num = *in_num = 0; | |
704 | ||
d1f0132e MM |
705 | max = vq->vring.num; |
706 | desc = vq->vring.desc; | |
17cbca2b | 707 | i = head; |
d1f0132e | 708 | |
8fd9a636 RR |
709 | /* |
710 | * We have to read the descriptor after we read the descriptor number, | |
711 | * but there's a data dependency there so the CPU shouldn't reorder | |
712 | * that: no rmb() required. | |
713 | */ | |
714 | ||
2e04ef76 RR |
715 | /* |
716 | * If this is an indirect entry, then this buffer contains a descriptor | |
717 | * table which we handle as if it's any normal descriptor chain. | |
718 | */ | |
d1f0132e MM |
719 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { |
720 | if (desc[i].len % sizeof(struct vring_desc)) | |
721 | errx(1, "Invalid size for indirect buffer table"); | |
722 | ||
723 | max = desc[i].len / sizeof(struct vring_desc); | |
724 | desc = check_pointer(desc[i].addr, desc[i].len); | |
725 | i = 0; | |
726 | } | |
727 | ||
17cbca2b RR |
728 | do { |
729 | /* Grab the first descriptor, and check it's OK. */ | |
d1f0132e | 730 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 731 | iov[*out_num + *in_num].iov_base |
d1f0132e | 732 | = check_pointer(desc[i].addr, desc[i].len); |
17cbca2b | 733 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 734 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
735 | (*in_num)++; |
736 | else { | |
2e04ef76 RR |
737 | /* |
738 | * If it's an output descriptor, they're all supposed | |
739 | * to come before any input descriptors. | |
740 | */ | |
17cbca2b RR |
741 | if (*in_num) |
742 | errx(1, "Descriptor has out after in"); | |
743 | (*out_num)++; | |
744 | } | |
745 | ||
746 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 747 | if (*out_num + *in_num > max) |
17cbca2b | 748 | errx(1, "Looped descriptor"); |
d1f0132e | 749 | } while ((i = next_desc(desc, i, max)) != max); |
dde79789 | 750 | |
17cbca2b | 751 | return head; |
8ca47e00 RR |
752 | } |
753 | ||
2e04ef76 | 754 | /* |
a91d74a3 RR |
755 | * After we've used one of their buffers, we tell the Guest about it. Sometime |
756 | * later we'll want to send them an interrupt using trigger_irq(); note that | |
757 | * wait_for_vq_desc() does that for us if it has to wait. | |
2e04ef76 | 758 | */ |
17cbca2b | 759 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
8ca47e00 | 760 | { |
17cbca2b RR |
761 | struct vring_used_elem *used; |
762 | ||
2e04ef76 RR |
763 | /* |
764 | * The virtqueue contains a ring of used buffers. Get a pointer to the | |
765 | * next entry in that used ring. | |
766 | */ | |
17cbca2b RR |
767 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
768 | used->id = head; | |
769 | used->len = len; | |
770 | /* Make sure buffer is written before we update index. */ | |
771 | wmb(); | |
772 | vq->vring.used->idx++; | |
95c517c0 | 773 | vq->pending_used++; |
8ca47e00 RR |
774 | } |
775 | ||
17cbca2b | 776 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 777 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 778 | { |
17cbca2b | 779 | add_used(vq, head, len); |
56739c80 | 780 | trigger_irq(vq); |
8ca47e00 RR |
781 | } |
782 | ||
e1e72965 RR |
783 | /* |
784 | * The Console | |
785 | * | |
2e04ef76 RR |
786 | * We associate some data with the console for our exit hack. |
787 | */ | |
1842f23c | 788 | struct console_abort { |
dde79789 | 789 | /* How many times have they hit ^C? */ |
8ca47e00 | 790 | int count; |
dde79789 | 791 | /* When did they start? */ |
8ca47e00 RR |
792 | struct timeval start; |
793 | }; | |
794 | ||
dde79789 | 795 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 796 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 797 | { |
8ca47e00 | 798 | int len; |
17cbca2b | 799 | unsigned int head, in_num, out_num; |
659a0e66 RR |
800 | struct console_abort *abort = vq->dev->priv; |
801 | struct iovec iov[vq->vring.num]; | |
56ae43df | 802 | |
a91d74a3 | 803 | /* Make sure there's a descriptor available. */ |
659a0e66 | 804 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
56ae43df | 805 | if (out_num) |
17cbca2b | 806 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 807 | |
a91d74a3 | 808 | /* Read into it. This is where we usually wait. */ |
659a0e66 | 809 | len = readv(STDIN_FILENO, iov, in_num); |
8ca47e00 | 810 | if (len <= 0) { |
659a0e66 | 811 | /* Ran out of input? */ |
8ca47e00 | 812 | warnx("Failed to get console input, ignoring console."); |
2e04ef76 RR |
813 | /* |
814 | * For simplicity, dying threads kill the whole Launcher. So | |
815 | * just nap here. | |
816 | */ | |
659a0e66 RR |
817 | for (;;) |
818 | pause(); | |
8ca47e00 RR |
819 | } |
820 | ||
a91d74a3 | 821 | /* Tell the Guest we used a buffer. */ |
659a0e66 | 822 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 823 | |
2e04ef76 RR |
824 | /* |
825 | * Three ^C within one second? Exit. | |
dde79789 | 826 | * |
659a0e66 RR |
827 | * This is such a hack, but works surprisingly well. Each ^C has to |
828 | * be in a buffer by itself, so they can't be too fast. But we check | |
829 | * that we get three within about a second, so they can't be too | |
2e04ef76 RR |
830 | * slow. |
831 | */ | |
659a0e66 | 832 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
8ca47e00 | 833 | abort->count = 0; |
659a0e66 RR |
834 | return; |
835 | } | |
8ca47e00 | 836 | |
659a0e66 RR |
837 | abort->count++; |
838 | if (abort->count == 1) | |
839 | gettimeofday(&abort->start, NULL); | |
840 | else if (abort->count == 3) { | |
841 | struct timeval now; | |
842 | gettimeofday(&now, NULL); | |
843 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
844 | if (now.tv_sec <= abort->start.tv_sec+1) | |
845 | kill(0, SIGINT); | |
846 | abort->count = 0; | |
847 | } | |
8ca47e00 RR |
848 | } |
849 | ||
659a0e66 RR |
850 | /* This is the routine which handles console output (ie. stdout). */ |
851 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 852 | { |
17cbca2b | 853 | unsigned int head, out, in; |
17cbca2b RR |
854 | struct iovec iov[vq->vring.num]; |
855 | ||
a91d74a3 | 856 | /* We usually wait in here, for the Guest to give us something. */ |
659a0e66 RR |
857 | head = wait_for_vq_desc(vq, iov, &out, &in); |
858 | if (in) | |
859 | errx(1, "Input buffers in console output queue?"); | |
a91d74a3 RR |
860 | |
861 | /* writev can return a partial write, so we loop here. */ | |
659a0e66 RR |
862 | while (!iov_empty(iov, out)) { |
863 | int len = writev(STDOUT_FILENO, iov, out); | |
e0377e25 SA |
864 | if (len <= 0) { |
865 | warn("Write to stdout gave %i (%d)", len, errno); | |
866 | break; | |
867 | } | |
c0316a94 | 868 | iov_consume(iov, out, NULL, len); |
17cbca2b | 869 | } |
a91d74a3 RR |
870 | |
871 | /* | |
872 | * We're finished with that buffer: if we're going to sleep, | |
873 | * wait_for_vq_desc() will prod the Guest with an interrupt. | |
874 | */ | |
38bc2b8c | 875 | add_used(vq, head, 0); |
a161883a RR |
876 | } |
877 | ||
e1e72965 RR |
878 | /* |
879 | * The Network | |
880 | * | |
881 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 882 | * and write them to /dev/net/tun. |
a6bd8e13 | 883 | */ |
659a0e66 RR |
884 | struct net_info { |
885 | int tunfd; | |
886 | }; | |
887 | ||
888 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 889 | { |
659a0e66 RR |
890 | struct net_info *net_info = vq->dev->priv; |
891 | unsigned int head, out, in; | |
17cbca2b | 892 | struct iovec iov[vq->vring.num]; |
a161883a | 893 | |
a91d74a3 | 894 | /* We usually wait in here for the Guest to give us a packet. */ |
659a0e66 RR |
895 | head = wait_for_vq_desc(vq, iov, &out, &in); |
896 | if (in) | |
897 | errx(1, "Input buffers in net output queue?"); | |
a91d74a3 RR |
898 | /* |
899 | * Send the whole thing through to /dev/net/tun. It expects the exact | |
900 | * same format: what a coincidence! | |
901 | */ | |
659a0e66 | 902 | if (writev(net_info->tunfd, iov, out) < 0) |
e0377e25 | 903 | warnx("Write to tun failed (%d)?", errno); |
a91d74a3 RR |
904 | |
905 | /* | |
906 | * Done with that one; wait_for_vq_desc() will send the interrupt if | |
907 | * all packets are processed. | |
908 | */ | |
38bc2b8c | 909 | add_used(vq, head, 0); |
8ca47e00 RR |
910 | } |
911 | ||
a91d74a3 RR |
912 | /* |
913 | * Handling network input is a bit trickier, because I've tried to optimize it. | |
914 | * | |
915 | * First we have a helper routine which tells is if from this file descriptor | |
916 | * (ie. the /dev/net/tun device) will block: | |
917 | */ | |
4a8962e2 RR |
918 | static bool will_block(int fd) |
919 | { | |
920 | fd_set fdset; | |
921 | struct timeval zero = { 0, 0 }; | |
922 | FD_ZERO(&fdset); | |
923 | FD_SET(fd, &fdset); | |
924 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
925 | } | |
926 | ||
a91d74a3 RR |
927 | /* |
928 | * This handles packets coming in from the tun device to our Guest. Like all | |
929 | * service routines, it gets called again as soon as it returns, so you don't | |
930 | * see a while(1) loop here. | |
931 | */ | |
659a0e66 | 932 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 933 | { |
8ca47e00 | 934 | int len; |
659a0e66 RR |
935 | unsigned int head, out, in; |
936 | struct iovec iov[vq->vring.num]; | |
937 | struct net_info *net_info = vq->dev->priv; | |
938 | ||
a91d74a3 RR |
939 | /* |
940 | * Get a descriptor to write an incoming packet into. This will also | |
941 | * send an interrupt if they're out of descriptors. | |
942 | */ | |
659a0e66 RR |
943 | head = wait_for_vq_desc(vq, iov, &out, &in); |
944 | if (out) | |
945 | errx(1, "Output buffers in net input queue?"); | |
4a8962e2 | 946 | |
a91d74a3 RR |
947 | /* |
948 | * If it looks like we'll block reading from the tun device, send them | |
949 | * an interrupt. | |
950 | */ | |
4a8962e2 RR |
951 | if (vq->pending_used && will_block(net_info->tunfd)) |
952 | trigger_irq(vq); | |
953 | ||
a91d74a3 RR |
954 | /* |
955 | * Read in the packet. This is where we normally wait (when there's no | |
956 | * incoming network traffic). | |
957 | */ | |
659a0e66 | 958 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 959 | if (len <= 0) |
e0377e25 | 960 | warn("Failed to read from tun (%d).", errno); |
a91d74a3 RR |
961 | |
962 | /* | |
963 | * Mark that packet buffer as used, but don't interrupt here. We want | |
964 | * to wait until we've done as much work as we can. | |
965 | */ | |
4a8962e2 | 966 | add_used(vq, head, len); |
659a0e66 | 967 | } |
a91d74a3 | 968 | /*:*/ |
dde79789 | 969 | |
a91d74a3 | 970 | /* This is the helper to create threads: run the service routine in a loop. */ |
659a0e66 RR |
971 | static int do_thread(void *_vq) |
972 | { | |
973 | struct virtqueue *vq = _vq; | |
17cbca2b | 974 | |
659a0e66 RR |
975 | for (;;) |
976 | vq->service(vq); | |
977 | return 0; | |
978 | } | |
17cbca2b | 979 | |
2e04ef76 RR |
980 | /* |
981 | * When a child dies, we kill our entire process group with SIGTERM. This | |
982 | * also has the side effect that the shell restores the console for us! | |
983 | */ | |
659a0e66 RR |
984 | static void kill_launcher(int signal) |
985 | { | |
986 | kill(0, SIGTERM); | |
8ca47e00 RR |
987 | } |
988 | ||
659a0e66 | 989 | static void reset_device(struct device *dev) |
56ae43df | 990 | { |
659a0e66 RR |
991 | struct virtqueue *vq; |
992 | ||
993 | verbose("Resetting device %s\n", dev->name); | |
994 | ||
995 | /* Clear any features they've acked. */ | |
996 | memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); | |
997 | ||
998 | /* We're going to be explicitly killing threads, so ignore them. */ | |
999 | signal(SIGCHLD, SIG_IGN); | |
1000 | ||
1001 | /* Zero out the virtqueues, get rid of their threads */ | |
1002 | for (vq = dev->vq; vq; vq = vq->next) { | |
1003 | if (vq->thread != (pid_t)-1) { | |
1004 | kill(vq->thread, SIGTERM); | |
1005 | waitpid(vq->thread, NULL, 0); | |
1006 | vq->thread = (pid_t)-1; | |
1007 | } | |
1008 | memset(vq->vring.desc, 0, | |
1009 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); | |
1010 | lg_last_avail(vq) = 0; | |
1011 | } | |
1012 | dev->running = false; | |
1013 | ||
1014 | /* Now we care if threads die. */ | |
1015 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
1016 | } |
1017 | ||
a91d74a3 RR |
1018 | /*L:216 |
1019 | * This actually creates the thread which services the virtqueue for a device. | |
1020 | */ | |
659a0e66 | 1021 | static void create_thread(struct virtqueue *vq) |
5dae785a | 1022 | { |
2e04ef76 | 1023 | /* |
a91d74a3 RR |
1024 | * Create stack for thread. Since the stack grows upwards, we point |
1025 | * the stack pointer to the end of this region. | |
2e04ef76 | 1026 | */ |
659a0e66 RR |
1027 | char *stack = malloc(32768); |
1028 | unsigned long args[] = { LHREQ_EVENTFD, | |
1029 | vq->config.pfn*getpagesize(), 0 }; | |
1030 | ||
1031 | /* Create a zero-initialized eventfd. */ | |
1032 | vq->eventfd = eventfd(0, 0); | |
1033 | if (vq->eventfd < 0) | |
1034 | err(1, "Creating eventfd"); | |
1035 | args[2] = vq->eventfd; | |
1036 | ||
a91d74a3 RR |
1037 | /* |
1038 | * Attach an eventfd to this virtqueue: it will go off when the Guest | |
1039 | * does an LHCALL_NOTIFY for this vq. | |
1040 | */ | |
659a0e66 RR |
1041 | if (write(lguest_fd, &args, sizeof(args)) != 0) |
1042 | err(1, "Attaching eventfd"); | |
1043 | ||
a91d74a3 RR |
1044 | /* |
1045 | * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so | |
1046 | * we get a signal if it dies. | |
1047 | */ | |
659a0e66 RR |
1048 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); |
1049 | if (vq->thread == (pid_t)-1) | |
1050 | err(1, "Creating clone"); | |
a91d74a3 RR |
1051 | |
1052 | /* We close our local copy now the child has it. */ | |
659a0e66 | 1053 | close(vq->eventfd); |
5dae785a RR |
1054 | } |
1055 | ||
659a0e66 | 1056 | static void start_device(struct device *dev) |
6e5aa7ef | 1057 | { |
659a0e66 | 1058 | unsigned int i; |
6e5aa7ef RR |
1059 | struct virtqueue *vq; |
1060 | ||
659a0e66 RR |
1061 | verbose("Device %s OK: offered", dev->name); |
1062 | for (i = 0; i < dev->feature_len; i++) | |
1063 | verbose(" %02x", get_feature_bits(dev)[i]); | |
1064 | verbose(", accepted"); | |
1065 | for (i = 0; i < dev->feature_len; i++) | |
1066 | verbose(" %02x", get_feature_bits(dev) | |
1067 | [dev->feature_len+i]); | |
1068 | ||
1069 | for (vq = dev->vq; vq; vq = vq->next) { | |
1070 | if (vq->service) | |
1071 | create_thread(vq); | |
1072 | } | |
1073 | dev->running = true; | |
1074 | } | |
1075 | ||
1076 | static void cleanup_devices(void) | |
1077 | { | |
1078 | struct device *dev; | |
1079 | ||
1080 | for (dev = devices.dev; dev; dev = dev->next) | |
1081 | reset_device(dev); | |
6e5aa7ef | 1082 | |
659a0e66 RR |
1083 | /* If we saved off the original terminal settings, restore them now. */ |
1084 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
1085 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
1086 | } | |
6e5aa7ef | 1087 | |
659a0e66 RR |
1088 | /* When the Guest tells us they updated the status field, we handle it. */ |
1089 | static void update_device_status(struct device *dev) | |
1090 | { | |
1091 | /* A zero status is a reset, otherwise it's a set of flags. */ | |
1092 | if (dev->desc->status == 0) | |
1093 | reset_device(dev); | |
1094 | else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
a007a751 | 1095 | warnx("Device %s configuration FAILED", dev->name); |
659a0e66 RR |
1096 | if (dev->running) |
1097 | reset_device(dev); | |
3c3ed482 RR |
1098 | } else { |
1099 | if (dev->running) | |
1100 | err(1, "Device %s features finalized twice", dev->name); | |
1101 | start_device(dev); | |
6e5aa7ef RR |
1102 | } |
1103 | } | |
1104 | ||
a91d74a3 RR |
1105 | /*L:215 |
1106 | * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In | |
1107 | * particular, it's used to notify us of device status changes during boot. | |
1108 | */ | |
56739c80 | 1109 | static void handle_output(unsigned long addr) |
8ca47e00 RR |
1110 | { |
1111 | struct device *i; | |
17cbca2b | 1112 | |
659a0e66 | 1113 | /* Check each device. */ |
17cbca2b | 1114 | for (i = devices.dev; i; i = i->next) { |
659a0e66 RR |
1115 | struct virtqueue *vq; |
1116 | ||
a91d74a3 RR |
1117 | /* |
1118 | * Notifications to device descriptors mean they updated the | |
1119 | * device status. | |
1120 | */ | |
6e5aa7ef | 1121 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1122 | update_device_status(i); |
6e5aa7ef RR |
1123 | return; |
1124 | } | |
1125 | ||
3c3ed482 | 1126 | /* Devices should not be used before features are finalized. */ |
17cbca2b | 1127 | for (vq = i->vq; vq; vq = vq->next) { |
659a0e66 | 1128 | if (addr != vq->config.pfn*getpagesize()) |
6e5aa7ef | 1129 | continue; |
3c3ed482 | 1130 | errx(1, "Notification on %s before setup!", i->name); |
8ca47e00 RR |
1131 | } |
1132 | } | |
dde79789 | 1133 | |
2e04ef76 RR |
1134 | /* |
1135 | * Early console write is done using notify on a nul-terminated string | |
1136 | * in Guest memory. It's also great for hacking debugging messages | |
1137 | * into a Guest. | |
1138 | */ | |
17cbca2b RR |
1139 | if (addr >= guest_limit) |
1140 | errx(1, "Bad NOTIFY %#lx", addr); | |
1141 | ||
1142 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1143 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1144 | } |
1145 | ||
dde79789 RR |
1146 | /*L:190 |
1147 | * Device Setup | |
1148 | * | |
1149 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1150 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1151 | * routines to allocate and manage them. |
1152 | */ | |
8ca47e00 | 1153 | |
2e04ef76 RR |
1154 | /* |
1155 | * The layout of the device page is a "struct lguest_device_desc" followed by a | |
a586d4f6 RR |
1156 | * number of virtqueue descriptors, then two sets of feature bits, then an |
1157 | * array of configuration bytes. This routine returns the configuration | |
2e04ef76 RR |
1158 | * pointer. |
1159 | */ | |
a586d4f6 RR |
1160 | static u8 *device_config(const struct device *dev) |
1161 | { | |
1162 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1163 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1164 | + dev->feature_len * 2; | |
17cbca2b RR |
1165 | } |
1166 | ||
2e04ef76 RR |
1167 | /* |
1168 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
a586d4f6 | 1169 | * table page just above the Guest's normal memory. It returns a pointer to |
2e04ef76 RR |
1170 | * that descriptor. |
1171 | */ | |
a586d4f6 | 1172 | static struct lguest_device_desc *new_dev_desc(u16 type) |
17cbca2b | 1173 | { |
a586d4f6 RR |
1174 | struct lguest_device_desc d = { .type = type }; |
1175 | void *p; | |
17cbca2b | 1176 | |
a586d4f6 RR |
1177 | /* Figure out where the next device config is, based on the last one. */ |
1178 | if (devices.lastdev) | |
1179 | p = device_config(devices.lastdev) | |
1180 | + devices.lastdev->desc->config_len; | |
1181 | else | |
1182 | p = devices.descpage; | |
17cbca2b | 1183 | |
a586d4f6 RR |
1184 | /* We only have one page for all the descriptors. */ |
1185 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1186 | errx(1, "Too many devices"); | |
17cbca2b | 1187 | |
a586d4f6 RR |
1188 | /* p might not be aligned, so we memcpy in. */ |
1189 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1190 | } |
1191 | ||
2e04ef76 RR |
1192 | /* |
1193 | * Each device descriptor is followed by the description of its virtqueues. We | |
1194 | * specify how many descriptors the virtqueue is to have. | |
1195 | */ | |
17cbca2b | 1196 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
659a0e66 | 1197 | void (*service)(struct virtqueue *)) |
17cbca2b RR |
1198 | { |
1199 | unsigned int pages; | |
1200 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1201 | void *p; | |
1202 | ||
a6bd8e13 | 1203 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1204 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 1205 | / getpagesize(); |
17cbca2b RR |
1206 | p = get_pages(pages); |
1207 | ||
d1c856e0 RR |
1208 | /* Initialize the virtqueue */ |
1209 | vq->next = NULL; | |
1210 | vq->last_avail_idx = 0; | |
1211 | vq->dev = dev; | |
a91d74a3 RR |
1212 | |
1213 | /* | |
1214 | * This is the routine the service thread will run, and its Process ID | |
1215 | * once it's running. | |
1216 | */ | |
659a0e66 RR |
1217 | vq->service = service; |
1218 | vq->thread = (pid_t)-1; | |
d1c856e0 | 1219 | |
17cbca2b RR |
1220 | /* Initialize the configuration. */ |
1221 | vq->config.num = num_descs; | |
1222 | vq->config.irq = devices.next_irq++; | |
1223 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1224 | ||
1225 | /* Initialize the vring. */ | |
2966af73 | 1226 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1227 | |
2e04ef76 RR |
1228 | /* |
1229 | * Append virtqueue to this device's descriptor. We use | |
a586d4f6 RR |
1230 | * device_config() to get the end of the device's current virtqueues; |
1231 | * we check that we haven't added any config or feature information | |
2e04ef76 RR |
1232 | * yet, otherwise we'd be overwriting them. |
1233 | */ | |
a586d4f6 RR |
1234 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); |
1235 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1236 | dev->num_vq++; |
a586d4f6 RR |
1237 | dev->desc->num_vq++; |
1238 | ||
1239 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b | 1240 | |
2e04ef76 RR |
1241 | /* |
1242 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1243 | * second. | |
1244 | */ | |
17cbca2b RR |
1245 | for (i = &dev->vq; *i; i = &(*i)->next); |
1246 | *i = vq; | |
8ca47e00 RR |
1247 | } |
1248 | ||
2e04ef76 RR |
1249 | /* |
1250 | * The first half of the feature bitmask is for us to advertise features. The | |
1251 | * second half is for the Guest to accept features. | |
1252 | */ | |
a586d4f6 RR |
1253 | static void add_feature(struct device *dev, unsigned bit) |
1254 | { | |
6e5aa7ef | 1255 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1256 | |
1257 | /* We can't extend the feature bits once we've added config bytes */ | |
1258 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1259 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1260 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1261 | } |
1262 | ||
a586d4f6 RR |
1263 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1264 | } | |
1265 | ||
2e04ef76 RR |
1266 | /* |
1267 | * This routine sets the configuration fields for an existing device's | |
a586d4f6 | 1268 | * descriptor. It only works for the last device, but that's OK because that's |
2e04ef76 RR |
1269 | * how we use it. |
1270 | */ | |
a586d4f6 RR |
1271 | static void set_config(struct device *dev, unsigned len, const void *conf) |
1272 | { | |
1273 | /* Check we haven't overflowed our single page. */ | |
1274 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1275 | errx(1, "Too many devices"); | |
1276 | ||
1277 | /* Copy in the config information, and store the length. */ | |
1278 | memcpy(device_config(dev), conf, len); | |
1279 | dev->desc->config_len = len; | |
8ef562d1 RR |
1280 | |
1281 | /* Size must fit in config_len field (8 bits)! */ | |
1282 | assert(dev->desc->config_len == len); | |
a586d4f6 RR |
1283 | } |
1284 | ||
2e04ef76 RR |
1285 | /* |
1286 | * This routine does all the creation and setup of a new device, including | |
a91d74a3 RR |
1287 | * calling new_dev_desc() to allocate the descriptor and device memory. We |
1288 | * don't actually start the service threads until later. | |
a6bd8e13 | 1289 | * |
2e04ef76 RR |
1290 | * See what I mean about userspace being boring? |
1291 | */ | |
659a0e66 | 1292 | static struct device *new_device(const char *name, u16 type) |
8ca47e00 RR |
1293 | { |
1294 | struct device *dev = malloc(sizeof(*dev)); | |
1295 | ||
dde79789 | 1296 | /* Now we populate the fields one at a time. */ |
17cbca2b | 1297 | dev->desc = new_dev_desc(type); |
17cbca2b | 1298 | dev->name = name; |
d1c856e0 | 1299 | dev->vq = NULL; |
713b15b3 RR |
1300 | dev->feature_len = 0; |
1301 | dev->num_vq = 0; | |
659a0e66 | 1302 | dev->running = false; |
ca16f580 | 1303 | dev->next = NULL; |
a586d4f6 | 1304 | |
2e04ef76 RR |
1305 | /* |
1306 | * Append to device list. Prepending to a single-linked list is | |
a586d4f6 RR |
1307 | * easier, but the user expects the devices to be arranged on the bus |
1308 | * in command-line order. The first network device on the command line | |
2e04ef76 RR |
1309 | * is eth0, the first block device /dev/vda, etc. |
1310 | */ | |
a586d4f6 RR |
1311 | if (devices.lastdev) |
1312 | devices.lastdev->next = dev; | |
1313 | else | |
1314 | devices.dev = dev; | |
1315 | devices.lastdev = dev; | |
1316 | ||
8ca47e00 RR |
1317 | return dev; |
1318 | } | |
1319 | ||
2e04ef76 RR |
1320 | /* |
1321 | * Our first setup routine is the console. It's a fairly simple device, but | |
1322 | * UNIX tty handling makes it uglier than it could be. | |
1323 | */ | |
17cbca2b | 1324 | static void setup_console(void) |
8ca47e00 RR |
1325 | { |
1326 | struct device *dev; | |
1327 | ||
dde79789 | 1328 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1329 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1330 | struct termios term = orig_term; | |
2e04ef76 RR |
1331 | /* |
1332 | * Then we turn off echo, line buffering and ^C etc: We want a | |
1333 | * raw input stream to the Guest. | |
1334 | */ | |
8ca47e00 RR |
1335 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1336 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
1337 | } |
1338 | ||
659a0e66 RR |
1339 | dev = new_device("console", VIRTIO_ID_CONSOLE); |
1340 | ||
dde79789 | 1341 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1342 | dev->priv = malloc(sizeof(struct console_abort)); |
1343 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1344 | |
2e04ef76 RR |
1345 | /* |
1346 | * The console needs two virtqueues: the input then the output. When | |
56ae43df RR |
1347 | * they put something the input queue, we make sure we're listening to |
1348 | * stdin. When they put something in the output queue, we write it to | |
2e04ef76 RR |
1349 | * stdout. |
1350 | */ | |
659a0e66 RR |
1351 | add_virtqueue(dev, VIRTQUEUE_NUM, console_input); |
1352 | add_virtqueue(dev, VIRTQUEUE_NUM, console_output); | |
17cbca2b | 1353 | |
659a0e66 | 1354 | verbose("device %u: console\n", ++devices.device_num); |
8ca47e00 | 1355 | } |
17cbca2b | 1356 | /*:*/ |
8ca47e00 | 1357 | |
2e04ef76 RR |
1358 | /*M:010 |
1359 | * Inter-guest networking is an interesting area. Simplest is to have a | |
17cbca2b RR |
1360 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
1361 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1362 | * |
9f54288d | 1363 | * More sophisticated is to use one of the tools developed for project like UML |
17cbca2b | 1364 | * to do networking. |
dde79789 | 1365 | * |
17cbca2b RR |
1366 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1367 | * completely generic ("here's my vring, attach to your vring") and would work | |
1368 | * for any traffic. Of course, namespace and permissions issues need to be | |
1369 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1370 | * multiple inter-guest channels behind one interface, although it would | |
1371 | * require some manner of hotplugging new virtio channels. | |
1372 | * | |
9f54288d | 1373 | * Finally, we could use a virtio network switch in the kernel, ie. vhost. |
2e04ef76 | 1374 | :*/ |
8ca47e00 RR |
1375 | |
1376 | static u32 str2ip(const char *ipaddr) | |
1377 | { | |
dec6a2be | 1378 | unsigned int b[4]; |
8ca47e00 | 1379 | |
dec6a2be MM |
1380 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1381 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1382 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1383 | } | |
1384 | ||
1385 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1386 | { | |
1387 | unsigned int m[6]; | |
1388 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1389 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1390 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1391 | mac[0] = m[0]; | |
1392 | mac[1] = m[1]; | |
1393 | mac[2] = m[2]; | |
1394 | mac[3] = m[3]; | |
1395 | mac[4] = m[4]; | |
1396 | mac[5] = m[5]; | |
8ca47e00 RR |
1397 | } |
1398 | ||
2e04ef76 RR |
1399 | /* |
1400 | * This code is "adapted" from libbridge: it attaches the Host end of the | |
dde79789 RR |
1401 | * network device to the bridge device specified by the command line. |
1402 | * | |
1403 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
2e04ef76 RR |
1404 | * dislike bridging), and I just try not to break it. |
1405 | */ | |
8ca47e00 RR |
1406 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1407 | { | |
1408 | int ifidx; | |
1409 | struct ifreq ifr; | |
1410 | ||
1411 | if (!*br_name) | |
1412 | errx(1, "must specify bridge name"); | |
1413 | ||
1414 | ifidx = if_nametoindex(if_name); | |
1415 | if (!ifidx) | |
1416 | errx(1, "interface %s does not exist!", if_name); | |
1417 | ||
1418 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1419 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1420 | ifr.ifr_ifindex = ifidx; |
1421 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1422 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1423 | } | |
1424 | ||
2e04ef76 RR |
1425 | /* |
1426 | * This sets up the Host end of the network device with an IP address, brings | |
dde79789 | 1427 | * it up so packets will flow, the copies the MAC address into the hwaddr |
2e04ef76 RR |
1428 | * pointer. |
1429 | */ | |
dec6a2be | 1430 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1431 | { |
1432 | struct ifreq ifr; | |
f846619e | 1433 | struct sockaddr_in sin; |
8ca47e00 RR |
1434 | |
1435 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1436 | strcpy(ifr.ifr_name, tapif); |
1437 | ||
1438 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
f846619e RR |
1439 | sin.sin_family = AF_INET; |
1440 | sin.sin_addr.s_addr = htonl(ipaddr); | |
1441 | memcpy(&ifr.ifr_addr, &sin, sizeof(sin)); | |
8ca47e00 | 1442 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
dec6a2be | 1443 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1444 | ifr.ifr_flags = IFF_UP; |
1445 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1446 | err(1, "Bringing interface %s up", tapif); |
1447 | } | |
1448 | ||
dec6a2be | 1449 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1450 | { |
8ca47e00 | 1451 | struct ifreq ifr; |
dec6a2be MM |
1452 | int netfd; |
1453 | ||
1454 | /* Start with this zeroed. Messy but sure. */ | |
1455 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1456 | |
2e04ef76 RR |
1457 | /* |
1458 | * We open the /dev/net/tun device and tell it we want a tap device. A | |
dde79789 RR |
1459 | * tap device is like a tun device, only somehow different. To tell |
1460 | * the truth, I completely blundered my way through this code, but it | |
2e04ef76 RR |
1461 | * works now! |
1462 | */ | |
8ca47e00 | 1463 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1464 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1465 | strcpy(ifr.ifr_name, "tap%d"); |
1466 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1467 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1468 | |
398f187d RR |
1469 | if (ioctl(netfd, TUNSETOFFLOAD, |
1470 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1471 | err(1, "Could not set features for tun device"); | |
1472 | ||
2e04ef76 RR |
1473 | /* |
1474 | * We don't need checksums calculated for packets coming in this | |
1475 | * device: trust us! | |
1476 | */ | |
8ca47e00 RR |
1477 | ioctl(netfd, TUNSETNOCSUM, 1); |
1478 | ||
dec6a2be MM |
1479 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1480 | return netfd; | |
1481 | } | |
1482 | ||
2e04ef76 RR |
1483 | /*L:195 |
1484 | * Our network is a Host<->Guest network. This can either use bridging or | |
dec6a2be MM |
1485 | * routing, but the principle is the same: it uses the "tun" device to inject |
1486 | * packets into the Host as if they came in from a normal network card. We | |
2e04ef76 RR |
1487 | * just shunt packets between the Guest and the tun device. |
1488 | */ | |
dec6a2be MM |
1489 | static void setup_tun_net(char *arg) |
1490 | { | |
1491 | struct device *dev; | |
659a0e66 RR |
1492 | struct net_info *net_info = malloc(sizeof(*net_info)); |
1493 | int ipfd; | |
dec6a2be MM |
1494 | u32 ip = INADDR_ANY; |
1495 | bool bridging = false; | |
1496 | char tapif[IFNAMSIZ], *p; | |
1497 | struct virtio_net_config conf; | |
1498 | ||
659a0e66 | 1499 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 1500 | |
17cbca2b | 1501 | /* First we create a new network device. */ |
659a0e66 RR |
1502 | dev = new_device("net", VIRTIO_ID_NET); |
1503 | dev->priv = net_info; | |
dde79789 | 1504 | |
2e04ef76 | 1505 | /* Network devices need a recv and a send queue, just like console. */ |
659a0e66 RR |
1506 | add_virtqueue(dev, VIRTQUEUE_NUM, net_input); |
1507 | add_virtqueue(dev, VIRTQUEUE_NUM, net_output); | |
8ca47e00 | 1508 | |
2e04ef76 RR |
1509 | /* |
1510 | * We need a socket to perform the magic network ioctls to bring up the | |
1511 | * tap interface, connect to the bridge etc. Any socket will do! | |
1512 | */ | |
8ca47e00 RR |
1513 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1514 | if (ipfd < 0) | |
1515 | err(1, "opening IP socket"); | |
1516 | ||
dde79789 | 1517 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1518 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1519 | arg += strlen(BRIDGE_PFX); |
1520 | bridging = true; | |
1521 | } | |
1522 | ||
1523 | /* A mac address may follow the bridge name or IP address */ | |
1524 | p = strchr(arg, ':'); | |
1525 | if (p) { | |
1526 | str2mac(p+1, conf.mac); | |
40c42076 | 1527 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1528 | *p = '\0'; |
dec6a2be MM |
1529 | } |
1530 | ||
1531 | /* arg is now either an IP address or a bridge name */ | |
1532 | if (bridging) | |
1533 | add_to_bridge(ipfd, tapif, arg); | |
1534 | else | |
8ca47e00 RR |
1535 | ip = str2ip(arg); |
1536 | ||
dec6a2be MM |
1537 | /* Set up the tun device. */ |
1538 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1539 | |
398f187d RR |
1540 | /* Expect Guest to handle everything except UFO */ |
1541 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1542 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1543 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1544 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1545 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1546 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1547 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1548 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e MM |
1549 | /* We handle indirect ring entries */ |
1550 | add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); | |
927cfb97 RR |
1551 | /* We're compliant with the damn spec. */ |
1552 | add_feature(dev, VIRTIO_F_ANY_LAYOUT); | |
a586d4f6 | 1553 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1554 | |
a586d4f6 | 1555 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1556 | close(ipfd); |
1557 | ||
dec6a2be MM |
1558 | devices.device_num++; |
1559 | ||
1560 | if (bridging) | |
1561 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1562 | devices.device_num, tapif, arg); | |
1563 | else | |
1564 | verbose("device %u: tun %s: %s\n", | |
1565 | devices.device_num, tapif, arg); | |
8ca47e00 | 1566 | } |
a91d74a3 | 1567 | /*:*/ |
17cbca2b | 1568 | |
e1e72965 | 1569 | /* This hangs off device->priv. */ |
1842f23c | 1570 | struct vblk_info { |
17cbca2b RR |
1571 | /* The size of the file. */ |
1572 | off64_t len; | |
1573 | ||
1574 | /* The file descriptor for the file. */ | |
1575 | int fd; | |
1576 | ||
17cbca2b RR |
1577 | }; |
1578 | ||
e1e72965 RR |
1579 | /*L:210 |
1580 | * The Disk | |
1581 | * | |
a91d74a3 RR |
1582 | * The disk only has one virtqueue, so it only has one thread. It is really |
1583 | * simple: the Guest asks for a block number and we read or write that position | |
1584 | * in the file. | |
1585 | * | |
1586 | * Before we serviced each virtqueue in a separate thread, that was unacceptably | |
1587 | * slow: the Guest waits until the read is finished before running anything | |
1588 | * else, even if it could have been doing useful work. | |
1589 | * | |
1590 | * We could have used async I/O, except it's reputed to suck so hard that | |
1591 | * characters actually go missing from your code when you try to use it. | |
e1e72965 | 1592 | */ |
659a0e66 | 1593 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 1594 | { |
659a0e66 | 1595 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b | 1596 | unsigned int head, out_num, in_num, wlen; |
c0316a94 | 1597 | int ret, i; |
cb38fa23 | 1598 | u8 *in; |
c0316a94 | 1599 | struct virtio_blk_outhdr out; |
659a0e66 | 1600 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
1601 | off64_t off; |
1602 | ||
a91d74a3 RR |
1603 | /* |
1604 | * Get the next request, where we normally wait. It triggers the | |
1605 | * interrupt to acknowledge previously serviced requests (if any). | |
1606 | */ | |
659a0e66 | 1607 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
17cbca2b | 1608 | |
c0316a94 RR |
1609 | /* Copy the output header from the front of the iov (adjusts iov) */ |
1610 | iov_consume(iov, out_num, &out, sizeof(out)); | |
1611 | ||
1612 | /* Find and trim end of iov input array, for our status byte. */ | |
1613 | in = NULL; | |
1614 | for (i = out_num + in_num - 1; i >= out_num; i--) { | |
1615 | if (iov[i].iov_len > 0) { | |
1616 | in = iov[i].iov_base + iov[i].iov_len - 1; | |
1617 | iov[i].iov_len--; | |
1618 | break; | |
1619 | } | |
1620 | } | |
1621 | if (!in) | |
1622 | errx(1, "Bad virtblk cmd with no room for status"); | |
17cbca2b | 1623 | |
a91d74a3 RR |
1624 | /* |
1625 | * For historical reasons, block operations are expressed in 512 byte | |
1626 | * "sectors". | |
1627 | */ | |
c0316a94 | 1628 | off = out.sector * 512; |
17cbca2b | 1629 | |
2e04ef76 RR |
1630 | /* |
1631 | * In general the virtio block driver is allowed to try SCSI commands. | |
1632 | * It'd be nice if we supported eject, for example, but we don't. | |
1633 | */ | |
c0316a94 | 1634 | if (out.type & VIRTIO_BLK_T_SCSI_CMD) { |
17cbca2b | 1635 | fprintf(stderr, "Scsi commands unsupported\n"); |
cb38fa23 | 1636 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1637 | wlen = sizeof(*in); |
c0316a94 | 1638 | } else if (out.type & VIRTIO_BLK_T_OUT) { |
2e04ef76 RR |
1639 | /* |
1640 | * Write | |
1641 | * | |
1642 | * Move to the right location in the block file. This can fail | |
1643 | * if they try to write past end. | |
1644 | */ | |
17cbca2b | 1645 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 1646 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 1647 | |
c0316a94 RR |
1648 | ret = writev(vblk->fd, iov, out_num); |
1649 | verbose("WRITE to sector %llu: %i\n", out.sector, ret); | |
17cbca2b | 1650 | |
2e04ef76 RR |
1651 | /* |
1652 | * Grr... Now we know how long the descriptor they sent was, we | |
17cbca2b | 1653 | * make sure they didn't try to write over the end of the block |
2e04ef76 RR |
1654 | * file (possibly extending it). |
1655 | */ | |
17cbca2b RR |
1656 | if (ret > 0 && off + ret > vblk->len) { |
1657 | /* Trim it back to the correct length */ | |
1658 | ftruncate64(vblk->fd, vblk->len); | |
1659 | /* Die, bad Guest, die. */ | |
1660 | errx(1, "Write past end %llu+%u", off, ret); | |
1661 | } | |
7bc9fdda TH |
1662 | |
1663 | wlen = sizeof(*in); | |
1664 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
c0316a94 | 1665 | } else if (out.type & VIRTIO_BLK_T_FLUSH) { |
7bc9fdda TH |
1666 | /* Flush */ |
1667 | ret = fdatasync(vblk->fd); | |
1668 | verbose("FLUSH fdatasync: %i\n", ret); | |
1200e646 | 1669 | wlen = sizeof(*in); |
cb38fa23 | 1670 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b | 1671 | } else { |
2e04ef76 RR |
1672 | /* |
1673 | * Read | |
1674 | * | |
1675 | * Move to the right location in the block file. This can fail | |
1676 | * if they try to read past end. | |
1677 | */ | |
17cbca2b | 1678 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 1679 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 1680 | |
c0316a94 | 1681 | ret = readv(vblk->fd, iov + out_num, in_num); |
17cbca2b | 1682 | if (ret >= 0) { |
1200e646 | 1683 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1684 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1685 | } else { |
1200e646 | 1686 | wlen = sizeof(*in); |
cb38fa23 | 1687 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1688 | } |
1689 | } | |
1690 | ||
a91d74a3 | 1691 | /* Finished that request. */ |
38bc2b8c | 1692 | add_used(vq, head, wlen); |
17cbca2b RR |
1693 | } |
1694 | ||
e1e72965 | 1695 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1696 | static void setup_block_file(const char *filename) |
1697 | { | |
17cbca2b RR |
1698 | struct device *dev; |
1699 | struct vblk_info *vblk; | |
a586d4f6 | 1700 | struct virtio_blk_config conf; |
17cbca2b | 1701 | |
2e04ef76 | 1702 | /* Creat the device. */ |
659a0e66 | 1703 | dev = new_device("block", VIRTIO_ID_BLOCK); |
17cbca2b | 1704 | |
e1e72965 | 1705 | /* The device has one virtqueue, where the Guest places requests. */ |
659a0e66 | 1706 | add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); |
17cbca2b RR |
1707 | |
1708 | /* Allocate the room for our own bookkeeping */ | |
1709 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1710 | ||
1711 | /* First we open the file and store the length. */ | |
1712 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1713 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1714 | ||
7bc9fdda TH |
1715 | /* We support FLUSH. */ |
1716 | add_feature(dev, VIRTIO_BLK_F_FLUSH); | |
a586d4f6 | 1717 | |
17cbca2b | 1718 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1719 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b | 1720 | |
2e04ef76 RR |
1721 | /* |
1722 | * Tell Guest not to put in too many descriptors at once: two are used | |
1723 | * for the in and out elements. | |
1724 | */ | |
a586d4f6 RR |
1725 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1726 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1727 | ||
8ef562d1 RR |
1728 | /* Don't try to put whole struct: we have 8 bit limit. */ |
1729 | set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf); | |
17cbca2b | 1730 | |
17cbca2b | 1731 | verbose("device %u: virtblock %llu sectors\n", |
659a0e66 | 1732 | ++devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1733 | } |
28fd6d7f | 1734 | |
2e04ef76 | 1735 | /*L:211 |
a454bb36 | 1736 | * Our random number generator device reads from /dev/urandom into the Guest's |
28fd6d7f | 1737 | * input buffers. The usual case is that the Guest doesn't want random numbers |
a454bb36 | 1738 | * and so has no buffers although /dev/urandom is still readable, whereas |
28fd6d7f RR |
1739 | * console is the reverse. |
1740 | * | |
2e04ef76 RR |
1741 | * The same logic applies, however. |
1742 | */ | |
1743 | struct rng_info { | |
1744 | int rfd; | |
1745 | }; | |
1746 | ||
659a0e66 | 1747 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
1748 | { |
1749 | int len; | |
1750 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
1751 | struct rng_info *rng_info = vq->dev->priv; |
1752 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
1753 | |
1754 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 1755 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f RR |
1756 | if (out_num) |
1757 | errx(1, "Output buffers in rng?"); | |
1758 | ||
2e04ef76 | 1759 | /* |
a91d74a3 RR |
1760 | * Just like the console write, we loop to cover the whole iovec. |
1761 | * In this case, short reads actually happen quite a bit. | |
2e04ef76 | 1762 | */ |
28fd6d7f | 1763 | while (!iov_empty(iov, in_num)) { |
659a0e66 | 1764 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f | 1765 | if (len <= 0) |
a454bb36 | 1766 | err(1, "Read from /dev/urandom gave %i", len); |
c0316a94 | 1767 | iov_consume(iov, in_num, NULL, len); |
28fd6d7f RR |
1768 | totlen += len; |
1769 | } | |
1770 | ||
1771 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 1772 | add_used(vq, head, totlen); |
28fd6d7f RR |
1773 | } |
1774 | ||
2e04ef76 RR |
1775 | /*L:199 |
1776 | * This creates a "hardware" random number device for the Guest. | |
1777 | */ | |
28fd6d7f RR |
1778 | static void setup_rng(void) |
1779 | { | |
1780 | struct device *dev; | |
659a0e66 | 1781 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 1782 | |
a454bb36 RR |
1783 | /* Our device's private info simply contains the /dev/urandom fd. */ |
1784 | rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY); | |
28fd6d7f | 1785 | |
2e04ef76 | 1786 | /* Create the new device. */ |
659a0e66 RR |
1787 | dev = new_device("rng", VIRTIO_ID_RNG); |
1788 | dev->priv = rng_info; | |
28fd6d7f RR |
1789 | |
1790 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
659a0e66 | 1791 | add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); |
28fd6d7f RR |
1792 | |
1793 | verbose("device %u: rng\n", devices.device_num++); | |
1794 | } | |
a6bd8e13 | 1795 | /* That's the end of device setup. */ |
ec04b13f | 1796 | |
a6bd8e13 | 1797 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1798 | static void __attribute__((noreturn)) restart_guest(void) |
1799 | { | |
1800 | unsigned int i; | |
1801 | ||
2e04ef76 RR |
1802 | /* |
1803 | * Since we don't track all open fds, we simply close everything beyond | |
1804 | * stderr. | |
1805 | */ | |
ec04b13f BR |
1806 | for (i = 3; i < FD_SETSIZE; i++) |
1807 | close(i); | |
8c79873d | 1808 | |
659a0e66 RR |
1809 | /* Reset all the devices (kills all threads). */ |
1810 | cleanup_devices(); | |
1811 | ||
ec04b13f BR |
1812 | execv(main_args[0], main_args); |
1813 | err(1, "Could not exec %s", main_args[0]); | |
1814 | } | |
8ca47e00 | 1815 | |
2e04ef76 RR |
1816 | /*L:220 |
1817 | * Finally we reach the core of the Launcher which runs the Guest, serves | |
1818 | * its input and output, and finally, lays it to rest. | |
1819 | */ | |
56739c80 | 1820 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
1821 | { |
1822 | for (;;) { | |
69a09dc1 | 1823 | struct lguest_pending notify; |
8ca47e00 RR |
1824 | int readval; |
1825 | ||
1826 | /* We read from the /dev/lguest device to run the Guest. */ | |
69a09dc1 | 1827 | readval = pread(lguest_fd, ¬ify, sizeof(notify), cpu_id); |
8ca47e00 | 1828 | |
17cbca2b | 1829 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
69a09dc1 RR |
1830 | if (readval == sizeof(notify)) { |
1831 | if (notify.trap == 0x1F) { | |
1832 | verbose("Notify on address %#08x\n", | |
1833 | notify.addr); | |
1834 | handle_output(notify.addr); | |
1835 | } else | |
1836 | errx(1, "Unknown trap %i addr %#08x\n", | |
1837 | notify.trap, notify.addr); | |
dde79789 | 1838 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1839 | } else if (errno == ENOENT) { |
1840 | char reason[1024] = { 0 }; | |
e3283fa0 | 1841 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1842 | errx(1, "%s", reason); |
ec04b13f BR |
1843 | /* ERESTART means that we need to reboot the guest */ |
1844 | } else if (errno == ERESTART) { | |
1845 | restart_guest(); | |
659a0e66 RR |
1846 | /* Anything else means a bug or incompatible change. */ |
1847 | } else | |
8ca47e00 | 1848 | err(1, "Running guest failed"); |
8ca47e00 RR |
1849 | } |
1850 | } | |
a6bd8e13 | 1851 | /*L:240 |
e1e72965 RR |
1852 | * This is the end of the Launcher. The good news: we are over halfway |
1853 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1854 | * of us. | |
dde79789 | 1855 | * |
e1e72965 RR |
1856 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1857 | * "make Host". | |
2e04ef76 | 1858 | :*/ |
8ca47e00 RR |
1859 | |
1860 | static struct option opts[] = { | |
1861 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1862 | { "tunnet", 1, NULL, 't' }, |
1863 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1864 | { "rng", 0, NULL, 'r' }, |
8ca47e00 | 1865 | { "initrd", 1, NULL, 'i' }, |
8aeb36e8 PS |
1866 | { "username", 1, NULL, 'u' }, |
1867 | { "chroot", 1, NULL, 'c' }, | |
8ca47e00 RR |
1868 | { NULL }, |
1869 | }; | |
1870 | static void usage(void) | |
1871 | { | |
1872 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1873 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1874 | "|--block=<filename>|--initrd=<filename>]...\n" |
1875 | "<mem-in-mb> vmlinux [args...]"); | |
1876 | } | |
1877 | ||
3c6b5bfa | 1878 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1879 | int main(int argc, char *argv[]) |
1880 | { | |
2e04ef76 | 1881 | /* Memory, code startpoint and size of the (optional) initrd. */ |
58a24566 | 1882 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
1883 | /* Two temporaries. */ |
1884 | int i, c; | |
3c6b5bfa | 1885 | /* The boot information for the Guest. */ |
43d33b21 | 1886 | struct boot_params *boot; |
dde79789 | 1887 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1888 | const char *initrd_name = NULL; |
1889 | ||
8aeb36e8 PS |
1890 | /* Password structure for initgroups/setres[gu]id */ |
1891 | struct passwd *user_details = NULL; | |
1892 | ||
1893 | /* Directory to chroot to */ | |
1894 | char *chroot_path = NULL; | |
1895 | ||
ec04b13f BR |
1896 | /* Save the args: we "reboot" by execing ourselves again. */ |
1897 | main_args = argv; | |
ec04b13f | 1898 | |
2e04ef76 RR |
1899 | /* |
1900 | * First we initialize the device list. We keep a pointer to the last | |
659a0e66 | 1901 | * device, and the next interrupt number to use for devices (1: |
2e04ef76 RR |
1902 | * remember that 0 is used by the timer). |
1903 | */ | |
a586d4f6 | 1904 | devices.lastdev = NULL; |
17cbca2b | 1905 | devices.next_irq = 1; |
8ca47e00 | 1906 | |
a91d74a3 | 1907 | /* We're CPU 0. In fact, that's the only CPU possible right now. */ |
e3283fa0 | 1908 | cpu_id = 0; |
a91d74a3 | 1909 | |
2e04ef76 RR |
1910 | /* |
1911 | * We need to know how much memory so we can set up the device | |
dde79789 RR |
1912 | * descriptor and memory pages for the devices as we parse the command |
1913 | * line. So we quickly look through the arguments to find the amount | |
2e04ef76 RR |
1914 | * of memory now. |
1915 | */ | |
6570c459 RR |
1916 | for (i = 1; i < argc; i++) { |
1917 | if (argv[i][0] != '-') { | |
3c6b5bfa | 1918 | mem = atoi(argv[i]) * 1024 * 1024; |
2e04ef76 RR |
1919 | /* |
1920 | * We start by mapping anonymous pages over all of | |
3c6b5bfa RR |
1921 | * guest-physical memory range. This fills it with 0, |
1922 | * and ensures that the Guest won't be killed when it | |
2e04ef76 RR |
1923 | * tries to access it. |
1924 | */ | |
3c6b5bfa RR |
1925 | guest_base = map_zeroed_pages(mem / getpagesize() |
1926 | + DEVICE_PAGES); | |
1927 | guest_limit = mem; | |
1928 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1929 | devices.descpage = get_pages(1); |
6570c459 RR |
1930 | break; |
1931 | } | |
1932 | } | |
dde79789 RR |
1933 | |
1934 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1935 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1936 | switch (c) { | |
1937 | case 'v': | |
1938 | verbose = true; | |
1939 | break; | |
8ca47e00 | 1940 | case 't': |
17cbca2b | 1941 | setup_tun_net(optarg); |
8ca47e00 RR |
1942 | break; |
1943 | case 'b': | |
17cbca2b | 1944 | setup_block_file(optarg); |
8ca47e00 | 1945 | break; |
28fd6d7f RR |
1946 | case 'r': |
1947 | setup_rng(); | |
1948 | break; | |
8ca47e00 RR |
1949 | case 'i': |
1950 | initrd_name = optarg; | |
1951 | break; | |
8aeb36e8 PS |
1952 | case 'u': |
1953 | user_details = getpwnam(optarg); | |
1954 | if (!user_details) | |
1955 | err(1, "getpwnam failed, incorrect username?"); | |
1956 | break; | |
1957 | case 'c': | |
1958 | chroot_path = optarg; | |
1959 | break; | |
8ca47e00 RR |
1960 | default: |
1961 | warnx("Unknown argument %s", argv[optind]); | |
1962 | usage(); | |
1963 | } | |
1964 | } | |
2e04ef76 RR |
1965 | /* |
1966 | * After the other arguments we expect memory and kernel image name, | |
1967 | * followed by command line arguments for the kernel. | |
1968 | */ | |
8ca47e00 RR |
1969 | if (optind + 2 > argc) |
1970 | usage(); | |
1971 | ||
3c6b5bfa RR |
1972 | verbose("Guest base is at %p\n", guest_base); |
1973 | ||
dde79789 | 1974 | /* We always have a console device */ |
17cbca2b | 1975 | setup_console(); |
8ca47e00 | 1976 | |
8ca47e00 | 1977 | /* Now we load the kernel */ |
47436aa4 | 1978 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1979 | |
3c6b5bfa RR |
1980 | /* Boot information is stashed at physical address 0 */ |
1981 | boot = from_guest_phys(0); | |
1982 | ||
dde79789 | 1983 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1984 | if (initrd_name) { |
1985 | initrd_size = load_initrd(initrd_name, mem); | |
2e04ef76 RR |
1986 | /* |
1987 | * These are the location in the Linux boot header where the | |
1988 | * start and size of the initrd are expected to be found. | |
1989 | */ | |
43d33b21 RR |
1990 | boot->hdr.ramdisk_image = mem - initrd_size; |
1991 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 1992 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 1993 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
1994 | } |
1995 | ||
2e04ef76 RR |
1996 | /* |
1997 | * The Linux boot header contains an "E820" memory map: ours is a | |
1998 | * simple, single region. | |
1999 | */ | |
43d33b21 RR |
2000 | boot->e820_entries = 1; |
2001 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
2e04ef76 RR |
2002 | /* |
2003 | * The boot header contains a command line pointer: we put the command | |
2004 | * line after the boot header. | |
2005 | */ | |
43d33b21 | 2006 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
e1e72965 | 2007 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 2008 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 2009 | |
e22a5398 RR |
2010 | /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */ |
2011 | boot->hdr.kernel_alignment = 0x1000000; | |
2012 | ||
814a0e5c | 2013 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 2014 | boot->hdr.version = 0x207; |
814a0e5c RR |
2015 | |
2016 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 2017 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 2018 | |
43d33b21 RR |
2019 | /* Tell the entry path not to try to reload segment registers. */ |
2020 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 2021 | |
9f54288d | 2022 | /* We tell the kernel to initialize the Guest. */ |
56739c80 | 2023 | tell_kernel(start); |
dde79789 | 2024 | |
a91d74a3 | 2025 | /* Ensure that we terminate if a device-servicing child dies. */ |
659a0e66 RR |
2026 | signal(SIGCHLD, kill_launcher); |
2027 | ||
2028 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
2029 | atexit(cleanup_devices); | |
8ca47e00 | 2030 | |
8aeb36e8 PS |
2031 | /* If requested, chroot to a directory */ |
2032 | if (chroot_path) { | |
2033 | if (chroot(chroot_path) != 0) | |
2034 | err(1, "chroot(\"%s\") failed", chroot_path); | |
2035 | ||
2036 | if (chdir("/") != 0) | |
2037 | err(1, "chdir(\"/\") failed"); | |
2038 | ||
2039 | verbose("chroot done\n"); | |
2040 | } | |
2041 | ||
2042 | /* If requested, drop privileges */ | |
2043 | if (user_details) { | |
2044 | uid_t u; | |
2045 | gid_t g; | |
2046 | ||
2047 | u = user_details->pw_uid; | |
2048 | g = user_details->pw_gid; | |
2049 | ||
2050 | if (initgroups(user_details->pw_name, g) != 0) | |
2051 | err(1, "initgroups failed"); | |
2052 | ||
2053 | if (setresgid(g, g, g) != 0) | |
2054 | err(1, "setresgid failed"); | |
2055 | ||
2056 | if (setresuid(u, u, u) != 0) | |
2057 | err(1, "setresuid failed"); | |
2058 | ||
2059 | verbose("Dropping privileges completed\n"); | |
2060 | } | |
2061 | ||
dde79789 | 2062 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 2063 | run_guest(); |
8ca47e00 | 2064 | } |
f56a384e RR |
2065 | /*:*/ |
2066 | ||
2067 | /*M:999 | |
2068 | * Mastery is done: you now know everything I do. | |
2069 | * | |
2070 | * But surely you have seen code, features and bugs in your wanderings which | |
2071 | * you now yearn to attack? That is the real game, and I look forward to you | |
2072 | * patching and forking lguest into the Your-Name-Here-visor. | |
2073 | * | |
2074 | * Farewell, and good coding! | |
2075 | * Rusty Russell. | |
2076 | */ |