Commit | Line | Data |
---|---|---|
f938d2c8 | 1 | /*P:100 This is the Launcher code, a simple program which lays out the |
a6bd8e13 RR |
2 | * "physical" memory for the new Guest by mapping the kernel image and |
3 | * the virtual devices, then opens /dev/lguest to tell the kernel | |
4 | * about the Guest and control it. :*/ | |
8ca47e00 RR |
5 | #define _LARGEFILE64_SOURCE |
6 | #define _GNU_SOURCE | |
7 | #include <stdio.h> | |
8 | #include <string.h> | |
9 | #include <unistd.h> | |
10 | #include <err.h> | |
11 | #include <stdint.h> | |
12 | #include <stdlib.h> | |
13 | #include <elf.h> | |
14 | #include <sys/mman.h> | |
6649bb7a | 15 | #include <sys/param.h> |
8ca47e00 RR |
16 | #include <sys/types.h> |
17 | #include <sys/stat.h> | |
18 | #include <sys/wait.h> | |
19 | #include <fcntl.h> | |
20 | #include <stdbool.h> | |
21 | #include <errno.h> | |
22 | #include <ctype.h> | |
23 | #include <sys/socket.h> | |
24 | #include <sys/ioctl.h> | |
25 | #include <sys/time.h> | |
26 | #include <time.h> | |
27 | #include <netinet/in.h> | |
28 | #include <net/if.h> | |
29 | #include <linux/sockios.h> | |
30 | #include <linux/if_tun.h> | |
31 | #include <sys/uio.h> | |
32 | #include <termios.h> | |
33 | #include <getopt.h> | |
34 | #include <zlib.h> | |
17cbca2b RR |
35 | #include <assert.h> |
36 | #include <sched.h> | |
a586d4f6 RR |
37 | #include <limits.h> |
38 | #include <stddef.h> | |
a161883a | 39 | #include <signal.h> |
b45d8cb0 | 40 | #include "linux/lguest_launcher.h" |
17cbca2b RR |
41 | #include "linux/virtio_config.h" |
42 | #include "linux/virtio_net.h" | |
43 | #include "linux/virtio_blk.h" | |
44 | #include "linux/virtio_console.h" | |
28fd6d7f | 45 | #include "linux/virtio_rng.h" |
17cbca2b | 46 | #include "linux/virtio_ring.h" |
d5d02d6d | 47 | #include "asm/bootparam.h" |
a6bd8e13 | 48 | /*L:110 We can ignore the 39 include files we need for this program, but I do |
db24e8c2 RR |
49 | * want to draw attention to the use of kernel-style types. |
50 | * | |
51 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
52 | * like these abbreviations, so we define them here. Note that u64 is always | |
53 | * unsigned long long, which works on all Linux systems: this means that we can | |
54 | * use %llu in printf for any u64. */ | |
55 | typedef unsigned long long u64; | |
56 | typedef uint32_t u32; | |
57 | typedef uint16_t u16; | |
58 | typedef uint8_t u8; | |
dde79789 | 59 | /*:*/ |
8ca47e00 RR |
60 | |
61 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
62 | #define NET_PEERNUM 1 | |
63 | #define BRIDGE_PFX "bridge:" | |
64 | #ifndef SIOCBRADDIF | |
65 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
66 | #endif | |
3c6b5bfa RR |
67 | /* We can have up to 256 pages for devices. */ |
68 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
69 | /* This will occupy 3 pages: it must be a power of 2. */ |
70 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 71 | |
dde79789 RR |
72 | /*L:120 verbose is both a global flag and a macro. The C preprocessor allows |
73 | * this, and although I wouldn't recommend it, it works quite nicely here. */ | |
8ca47e00 RR |
74 | static bool verbose; |
75 | #define verbose(args...) \ | |
76 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
77 | /*:*/ |
78 | ||
8c79873d RR |
79 | /* File descriptors for the Waker. */ |
80 | struct { | |
81 | int pipe[2]; | |
8c79873d RR |
82 | } waker_fds; |
83 | ||
3c6b5bfa RR |
84 | /* The pointer to the start of guest memory. */ |
85 | static void *guest_base; | |
86 | /* The maximum guest physical address allowed, and maximum possible. */ | |
87 | static unsigned long guest_limit, guest_max; | |
a161883a RR |
88 | /* The pipe for signal hander to write to. */ |
89 | static int timeoutpipe[2]; | |
aa124984 | 90 | static unsigned int timeout_usec = 500; |
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. */ |
8ca47e00 RR |
98 | struct device_list |
99 | { | |
dde79789 RR |
100 | /* Summary information about the devices in our list: ready to pass to |
101 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
102 | fd_set infds; |
103 | int max_infd; | |
104 | ||
17cbca2b RR |
105 | /* Counter to assign interrupt numbers. */ |
106 | unsigned int next_irq; | |
107 | ||
108 | /* Counter to print out convenient device numbers. */ | |
109 | unsigned int device_num; | |
110 | ||
dde79789 | 111 | /* The descriptor page for the devices. */ |
17cbca2b RR |
112 | u8 *descpage; |
113 | ||
dde79789 | 114 | /* A single linked list of devices. */ |
8ca47e00 | 115 | struct device *dev; |
a586d4f6 RR |
116 | /* And a pointer to the last device for easy append and also for |
117 | * configuration appending. */ | |
118 | struct device *lastdev; | |
8ca47e00 RR |
119 | }; |
120 | ||
17cbca2b RR |
121 | /* The list of Guest devices, based on command line arguments. */ |
122 | static struct device_list devices; | |
123 | ||
dde79789 | 124 | /* The device structure describes a single device. */ |
8ca47e00 RR |
125 | struct device |
126 | { | |
dde79789 | 127 | /* The linked-list pointer. */ |
8ca47e00 | 128 | struct device *next; |
17cbca2b | 129 | |
713b15b3 | 130 | /* The device's descriptor, as mapped into the Guest. */ |
8ca47e00 | 131 | struct lguest_device_desc *desc; |
17cbca2b | 132 | |
713b15b3 RR |
133 | /* We can't trust desc values once Guest has booted: we use these. */ |
134 | unsigned int feature_len; | |
135 | unsigned int num_vq; | |
136 | ||
17cbca2b RR |
137 | /* The name of this device, for --verbose. */ |
138 | const char *name; | |
8ca47e00 | 139 | |
dde79789 RR |
140 | /* If handle_input is set, it wants to be called when this file |
141 | * descriptor is ready. */ | |
8ca47e00 | 142 | int fd; |
56739c80 | 143 | bool (*handle_input)(struct device *me); |
8ca47e00 | 144 | |
17cbca2b RR |
145 | /* Any queues attached to this device */ |
146 | struct virtqueue *vq; | |
8ca47e00 | 147 | |
a007a751 RR |
148 | /* Handle status being finalized (ie. feature bits stable). */ |
149 | void (*ready)(struct device *me); | |
150 | ||
8ca47e00 RR |
151 | /* Device-specific data. */ |
152 | void *priv; | |
153 | }; | |
154 | ||
17cbca2b RR |
155 | /* The virtqueue structure describes a queue attached to a device. */ |
156 | struct virtqueue | |
157 | { | |
158 | struct virtqueue *next; | |
159 | ||
160 | /* Which device owns me. */ | |
161 | struct device *dev; | |
162 | ||
163 | /* The configuration for this queue. */ | |
164 | struct lguest_vqconfig config; | |
165 | ||
166 | /* The actual ring of buffers. */ | |
167 | struct vring vring; | |
168 | ||
169 | /* Last available index we saw. */ | |
170 | u16 last_avail_idx; | |
171 | ||
a161883a | 172 | /* The routine to call when the Guest pings us, or timeout. */ |
56739c80 | 173 | void (*handle_output)(struct virtqueue *me, bool timeout); |
20887611 RR |
174 | |
175 | /* Outstanding buffers */ | |
176 | unsigned int inflight; | |
a161883a RR |
177 | |
178 | /* Is this blocked awaiting a timer? */ | |
179 | bool blocked; | |
17cbca2b RR |
180 | }; |
181 | ||
ec04b13f BR |
182 | /* Remember the arguments to the program so we can "reboot" */ |
183 | static char **main_args; | |
184 | ||
f7027c63 RR |
185 | /* We have to be careful with barriers: our devices are all run in separate |
186 | * threads and so we need to make sure that changes visible to the Guest happen | |
187 | * in precise order. */ | |
188 | #define wmb() __asm__ __volatile__("" : : : "memory") | |
17cbca2b RR |
189 | |
190 | /* Convert an iovec element to the given type. | |
191 | * | |
192 | * This is a fairly ugly trick: we need to know the size of the type and | |
193 | * alignment requirement to check the pointer is kosher. It's also nice to | |
194 | * have the name of the type in case we report failure. | |
195 | * | |
196 | * Typing those three things all the time is cumbersome and error prone, so we | |
197 | * have a macro which sets them all up and passes to the real function. */ | |
198 | #define convert(iov, type) \ | |
199 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
200 | ||
201 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
202 | const char *name) | |
203 | { | |
204 | if (iov->iov_len != size) | |
205 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
206 | if ((unsigned long)iov->iov_base % align != 0) | |
207 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
208 | return iov->iov_base; | |
209 | } | |
210 | ||
b5111790 RR |
211 | /* Wrapper for the last available index. Makes it easier to change. */ |
212 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
213 | ||
17cbca2b RR |
214 | /* The virtio configuration space is defined to be little-endian. x86 is |
215 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
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 | ||
3c6b5bfa RR |
257 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
258 | * where pointers run wild and free! Unfortunately, like most userspace | |
259 | * programs, it's quite boring (which is why everyone likes to hack on the | |
260 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
261 | * will get you through this section. Or, maybe not. | |
262 | * | |
263 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
264 | * memory and stores it in "guest_base". In other words, Guest physical == | |
265 | * Launcher virtual with an offset. | |
266 | * | |
267 | * This can be tough to get your head around, but usually it just means that we | |
268 | * use these trivial conversion functions when the Guest gives us it's | |
269 | * "physical" addresses: */ | |
270 | static void *from_guest_phys(unsigned long addr) | |
271 | { | |
272 | return guest_base + addr; | |
273 | } | |
274 | ||
275 | static unsigned long to_guest_phys(const void *addr) | |
276 | { | |
277 | return (addr - guest_base); | |
278 | } | |
279 | ||
dde79789 RR |
280 | /*L:130 |
281 | * Loading the Kernel. | |
282 | * | |
283 | * We start with couple of simple helper routines. open_or_die() avoids | |
284 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
285 | static int open_or_die(const char *name, int flags) |
286 | { | |
287 | int fd = open(name, flags); | |
288 | if (fd < 0) | |
289 | err(1, "Failed to open %s", name); | |
290 | return fd; | |
291 | } | |
292 | ||
3c6b5bfa RR |
293 | /* map_zeroed_pages() takes a number of pages. */ |
294 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 295 | { |
3c6b5bfa RR |
296 | int fd = open_or_die("/dev/zero", O_RDONLY); |
297 | void *addr; | |
8ca47e00 | 298 | |
dde79789 | 299 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
300 | * copied). */ |
301 | addr = mmap(NULL, getpagesize() * num, | |
302 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
303 | if (addr == MAP_FAILED) | |
304 | err(1, "Mmaping %u pages of /dev/zero", num); | |
34bdaab4 | 305 | close(fd); |
3c6b5bfa RR |
306 | |
307 | return addr; | |
308 | } | |
309 | ||
310 | /* Get some more pages for a device. */ | |
311 | static void *get_pages(unsigned int num) | |
312 | { | |
313 | void *addr = from_guest_phys(guest_limit); | |
314 | ||
315 | guest_limit += num * getpagesize(); | |
316 | if (guest_limit > guest_max) | |
317 | errx(1, "Not enough memory for devices"); | |
318 | return addr; | |
8ca47e00 RR |
319 | } |
320 | ||
6649bb7a RM |
321 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
322 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
323 | * it falls back to reading the memory in. */ | |
324 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
325 | { | |
326 | ssize_t r; | |
327 | ||
328 | /* We map writable even though for some segments are marked read-only. | |
329 | * The kernel really wants to be writable: it patches its own | |
330 | * instructions. | |
331 | * | |
332 | * MAP_PRIVATE means that the page won't be copied until a write is | |
333 | * done to it. This allows us to share untouched memory between | |
334 | * Guests. */ | |
335 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
336 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
337 | return; | |
338 | ||
339 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
340 | r = pread(fd, addr, len, offset); | |
341 | if (r != len) | |
342 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
343 | } | |
344 | ||
dde79789 RR |
345 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
346 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
347 | * by all modern binaries on Linux including the kernel. | |
348 | * | |
349 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
350 | * address. We use the physical address; the Guest will map itself to the |
351 | * virtual address. | |
dde79789 RR |
352 | * |
353 | * We return the starting address. */ | |
47436aa4 | 354 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 355 | { |
8ca47e00 RR |
356 | Elf32_Phdr phdr[ehdr->e_phnum]; |
357 | unsigned int i; | |
8ca47e00 | 358 | |
dde79789 RR |
359 | /* Sanity checks on the main ELF header: an x86 executable with a |
360 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
361 | if (ehdr->e_type != ET_EXEC |
362 | || ehdr->e_machine != EM_386 | |
363 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
364 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
365 | errx(1, "Malformed elf header"); | |
366 | ||
dde79789 RR |
367 | /* An ELF executable contains an ELF header and a number of "program" |
368 | * headers which indicate which parts ("segments") of the program to | |
369 | * load where. */ | |
370 | ||
371 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
372 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
373 | err(1, "Seeking to program headers"); | |
374 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
375 | err(1, "Reading program headers"); | |
376 | ||
dde79789 | 377 | /* Try all the headers: there are usually only three. A read-only one, |
a6bd8e13 | 378 | * a read-write one, and a "note" section which we don't load. */ |
8ca47e00 | 379 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 380 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
381 | if (phdr[i].p_type != PT_LOAD) |
382 | continue; | |
383 | ||
384 | verbose("Section %i: size %i addr %p\n", | |
385 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
386 | ||
6649bb7a | 387 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 388 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 389 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
390 | } |
391 | ||
814a0e5c RR |
392 | /* The entry point is given in the ELF header. */ |
393 | return ehdr->e_entry; | |
8ca47e00 RR |
394 | } |
395 | ||
dde79789 | 396 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
397 | * supposed to jump into it and it will unpack itself. We used to have to |
398 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 399 | * |
5bbf89fc RR |
400 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
401 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
402 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 403 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 404 | { |
43d33b21 | 405 | struct boot_params boot; |
5bbf89fc RR |
406 | int r; |
407 | /* Modern bzImages get loaded at 1M. */ | |
408 | void *p = from_guest_phys(0x100000); | |
409 | ||
410 | /* Go back to the start of the file and read the header. It should be | |
71cced6e | 411 | * a Linux boot header (see Documentation/x86/i386/boot.txt) */ |
5bbf89fc | 412 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 413 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 414 | |
43d33b21 RR |
415 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
416 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
417 | errx(1, "This doesn't look like a bzImage to me"); |
418 | ||
43d33b21 RR |
419 | /* Skip over the extra sectors of the header. */ |
420 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
421 | |
422 | /* Now read everything into memory. in nice big chunks. */ | |
423 | while ((r = read(fd, p, 65536)) > 0) | |
424 | p += r; | |
425 | ||
43d33b21 RR |
426 | /* Finally, code32_start tells us where to enter the kernel. */ |
427 | return boot.hdr.code32_start; | |
8ca47e00 RR |
428 | } |
429 | ||
dde79789 | 430 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
e1e72965 RR |
431 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
432 | * work, we can load those, too. */ | |
47436aa4 | 433 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
434 | { |
435 | Elf32_Ehdr hdr; | |
436 | ||
dde79789 | 437 | /* Read in the first few bytes. */ |
8ca47e00 RR |
438 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
439 | err(1, "Reading kernel"); | |
440 | ||
dde79789 | 441 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 442 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 443 | return map_elf(fd, &hdr); |
8ca47e00 | 444 | |
a6bd8e13 | 445 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 446 | return load_bzimage(fd); |
8ca47e00 RR |
447 | } |
448 | ||
dde79789 RR |
449 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
450 | * it calls getpagesize() twice: "it's dumb code." | |
451 | * | |
452 | * Kernel guys get really het up about optimization, even when it's not | |
453 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
454 | static inline unsigned long page_align(unsigned long addr) |
455 | { | |
dde79789 | 456 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
457 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
458 | } | |
459 | ||
dde79789 RR |
460 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
461 | * the kernel which the kernel can use to boot from without needing any | |
462 | * drivers. Most distributions now use this as standard: the initrd contains | |
463 | * the code to load the appropriate driver modules for the current machine. | |
464 | * | |
465 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
466 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
467 | static unsigned long load_initrd(const char *name, unsigned long mem) |
468 | { | |
469 | int ifd; | |
470 | struct stat st; | |
471 | unsigned long len; | |
8ca47e00 RR |
472 | |
473 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 474 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
475 | if (fstat(ifd, &st) < 0) |
476 | err(1, "fstat() on initrd '%s'", name); | |
477 | ||
6649bb7a RM |
478 | /* We map the initrd at the top of memory, but mmap wants it to be |
479 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 480 | len = page_align(st.st_size); |
3c6b5bfa | 481 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
482 | /* Once a file is mapped, you can close the file descriptor. It's a |
483 | * little odd, but quite useful. */ | |
8ca47e00 | 484 | close(ifd); |
6649bb7a | 485 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
486 | |
487 | /* We return the initrd size. */ | |
8ca47e00 RR |
488 | return len; |
489 | } | |
e1e72965 | 490 | /*:*/ |
8ca47e00 | 491 | |
dde79789 RR |
492 | /* Simple routine to roll all the commandline arguments together with spaces |
493 | * between them. */ | |
8ca47e00 RR |
494 | static void concat(char *dst, char *args[]) |
495 | { | |
496 | unsigned int i, len = 0; | |
497 | ||
498 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
499 | if (i) { |
500 | strcat(dst+len, " "); | |
501 | len++; | |
502 | } | |
8ca47e00 | 503 | strcpy(dst+len, args[i]); |
1ef36fa6 | 504 | len += strlen(args[i]); |
8ca47e00 RR |
505 | } |
506 | /* In case it's empty. */ | |
507 | dst[len] = '\0'; | |
508 | } | |
509 | ||
e1e72965 RR |
510 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
511 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: | |
58a24566 MZ |
512 | * the base of Guest "physical" memory, the top physical page to allow and the |
513 | * entry point for the Guest. */ | |
56739c80 | 514 | static void tell_kernel(unsigned long start) |
8ca47e00 | 515 | { |
511801dc JS |
516 | unsigned long args[] = { LHREQ_INITIALIZE, |
517 | (unsigned long)guest_base, | |
58a24566 | 518 | guest_limit / getpagesize(), start }; |
3c6b5bfa RR |
519 | verbose("Guest: %p - %p (%#lx)\n", |
520 | guest_base, guest_base + guest_limit, guest_limit); | |
56739c80 RR |
521 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
522 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 523 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 524 | } |
dde79789 | 525 | /*:*/ |
8ca47e00 | 526 | |
17cbca2b | 527 | static void add_device_fd(int fd) |
8ca47e00 | 528 | { |
17cbca2b RR |
529 | FD_SET(fd, &devices.infds); |
530 | if (fd > devices.max_infd) | |
531 | devices.max_infd = fd; | |
8ca47e00 RR |
532 | } |
533 | ||
dde79789 RR |
534 | /*L:200 |
535 | * The Waker. | |
536 | * | |
e1e72965 RR |
537 | * With console, block and network devices, we can have lots of input which we |
538 | * need to process. We could try to tell the kernel what file descriptors to | |
539 | * watch, but handing a file descriptor mask through to the kernel is fairly | |
540 | * icky. | |
dde79789 | 541 | * |
8c79873d | 542 | * Instead, we clone off a thread which watches the file descriptors and writes |
e1e72965 RR |
543 | * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host |
544 | * stop running the Guest. This causes the Launcher to return from the | |
dde79789 RR |
545 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset |
546 | * the LHREQ_BREAK and wake us up again. | |
547 | * | |
548 | * This, of course, is merely a different *kind* of icky. | |
8c79873d RR |
549 | * |
550 | * Given my well-known antipathy to threads, I'd prefer to use processes. But | |
551 | * it's easier to share Guest memory with threads, and trivial to share the | |
552 | * devices.infds as the Launcher changes it. | |
dde79789 | 553 | */ |
8c79873d | 554 | static int waker(void *unused) |
8ca47e00 | 555 | { |
8c79873d RR |
556 | /* Close the write end of the pipe: only the Launcher has it open. */ |
557 | close(waker_fds.pipe[1]); | |
8ca47e00 RR |
558 | |
559 | for (;;) { | |
17cbca2b | 560 | fd_set rfds = devices.infds; |
511801dc | 561 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8c79873d RR |
562 | unsigned int maxfd = devices.max_infd; |
563 | ||
564 | /* We also listen to the pipe from the Launcher. */ | |
565 | FD_SET(waker_fds.pipe[0], &rfds); | |
566 | if (waker_fds.pipe[0] > maxfd) | |
567 | maxfd = waker_fds.pipe[0]; | |
8ca47e00 | 568 | |
dde79789 | 569 | /* Wait until input is ready from one of the devices. */ |
8c79873d RR |
570 | select(maxfd+1, &rfds, NULL, NULL, NULL); |
571 | ||
572 | /* Message from Launcher? */ | |
573 | if (FD_ISSET(waker_fds.pipe[0], &rfds)) { | |
574 | char c; | |
575 | /* If this fails, then assume Launcher has exited. | |
576 | * Don't do anything on exit: we're just a thread! */ | |
577 | if (read(waker_fds.pipe[0], &c, 1) != 1) | |
578 | _exit(0); | |
579 | continue; | |
580 | } | |
581 | ||
582 | /* Send LHREQ_BREAK command to snap the Launcher out of it. */ | |
56739c80 | 583 | pwrite(lguest_fd, args, sizeof(args), cpu_id); |
8ca47e00 | 584 | } |
8c79873d | 585 | return 0; |
8ca47e00 RR |
586 | } |
587 | ||
dde79789 | 588 | /* This routine just sets up a pipe to the Waker process. */ |
56739c80 | 589 | static void setup_waker(void) |
8c79873d RR |
590 | { |
591 | /* This pipe is closed when Launcher dies, telling Waker. */ | |
592 | if (pipe(waker_fds.pipe) != 0) | |
593 | err(1, "Creating pipe for Waker"); | |
8ca47e00 | 594 | |
8c79873d RR |
595 | if (clone(waker, malloc(4096) + 4096, CLONE_VM | SIGCHLD, NULL) == -1) |
596 | err(1, "Creating Waker"); | |
8ca47e00 RR |
597 | } |
598 | ||
e1e72965 | 599 | /* |
dde79789 RR |
600 | * Device Handling. |
601 | * | |
e1e72965 | 602 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 603 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 604 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
605 | * if something funny is going on: |
606 | */ | |
8ca47e00 RR |
607 | static void *_check_pointer(unsigned long addr, unsigned int size, |
608 | unsigned int line) | |
609 | { | |
dde79789 RR |
610 | /* We have to separately check addr and addr+size, because size could |
611 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 612 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 613 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
614 | /* We return a pointer for the caller's convenience, now we know it's |
615 | * safe to use. */ | |
3c6b5bfa | 616 | return from_guest_phys(addr); |
8ca47e00 | 617 | } |
dde79789 | 618 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
619 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
620 | ||
e1e72965 RR |
621 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
622 | * function returns the next descriptor in the chain, or vq->vring.num if we're | |
623 | * at the end. */ | |
17cbca2b RR |
624 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) |
625 | { | |
626 | unsigned int next; | |
627 | ||
628 | /* If this descriptor says it doesn't chain, we're done. */ | |
629 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
630 | return vq->vring.num; | |
631 | ||
632 | /* Check they're not leading us off end of descriptors. */ | |
633 | next = vq->vring.desc[i].next; | |
634 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
635 | wmb(); | |
636 | ||
637 | if (next >= vq->vring.num) | |
638 | errx(1, "Desc next is %u", next); | |
639 | ||
640 | return next; | |
641 | } | |
642 | ||
643 | /* This looks in the virtqueue and for the first available buffer, and converts | |
644 | * it to an iovec for convenient access. Since descriptors consist of some | |
645 | * number of output then some number of input descriptors, it's actually two | |
646 | * iovecs, but we pack them into one and note how many of each there were. | |
647 | * | |
648 | * This function returns the descriptor number found, or vq->vring.num (which | |
649 | * is never a valid descriptor number) if none was found. */ | |
650 | static unsigned get_vq_desc(struct virtqueue *vq, | |
651 | struct iovec iov[], | |
652 | unsigned int *out_num, unsigned int *in_num) | |
653 | { | |
654 | unsigned int i, head; | |
b5111790 | 655 | u16 last_avail; |
17cbca2b RR |
656 | |
657 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 RR |
658 | last_avail = lg_last_avail(vq); |
659 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) | |
17cbca2b | 660 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 661 | last_avail, vq->vring.avail->idx); |
17cbca2b RR |
662 | |
663 | /* If there's nothing new since last we looked, return invalid. */ | |
b5111790 | 664 | if (vq->vring.avail->idx == last_avail) |
17cbca2b RR |
665 | return vq->vring.num; |
666 | ||
667 | /* Grab the next descriptor number they're advertising, and increment | |
668 | * the index we've seen. */ | |
b5111790 RR |
669 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
670 | lg_last_avail(vq)++; | |
17cbca2b RR |
671 | |
672 | /* If their number is silly, that's a fatal mistake. */ | |
673 | if (head >= vq->vring.num) | |
674 | errx(1, "Guest says index %u is available", head); | |
675 | ||
676 | /* When we start there are none of either input nor output. */ | |
677 | *out_num = *in_num = 0; | |
678 | ||
679 | i = head; | |
680 | do { | |
681 | /* Grab the first descriptor, and check it's OK. */ | |
682 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
683 | iov[*out_num + *in_num].iov_base | |
684 | = check_pointer(vq->vring.desc[i].addr, | |
685 | vq->vring.desc[i].len); | |
686 | /* If this is an input descriptor, increment that count. */ | |
687 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
688 | (*in_num)++; | |
689 | else { | |
690 | /* If it's an output descriptor, they're all supposed | |
691 | * to come before any input descriptors. */ | |
692 | if (*in_num) | |
693 | errx(1, "Descriptor has out after in"); | |
694 | (*out_num)++; | |
695 | } | |
696 | ||
697 | /* If we've got too many, that implies a descriptor loop. */ | |
698 | if (*out_num + *in_num > vq->vring.num) | |
699 | errx(1, "Looped descriptor"); | |
700 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 701 | |
20887611 | 702 | vq->inflight++; |
17cbca2b | 703 | return head; |
8ca47e00 RR |
704 | } |
705 | ||
e1e72965 | 706 | /* After we've used one of their buffers, we tell them about it. We'll then |
17cbca2b RR |
707 | * want to send them an interrupt, using trigger_irq(). */ |
708 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 709 | { |
17cbca2b RR |
710 | struct vring_used_elem *used; |
711 | ||
e1e72965 RR |
712 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
713 | * next entry in that used ring. */ | |
17cbca2b RR |
714 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
715 | used->id = head; | |
716 | used->len = len; | |
717 | /* Make sure buffer is written before we update index. */ | |
718 | wmb(); | |
719 | vq->vring.used->idx++; | |
20887611 | 720 | vq->inflight--; |
8ca47e00 RR |
721 | } |
722 | ||
17cbca2b | 723 | /* This actually sends the interrupt for this virtqueue */ |
56739c80 | 724 | static void trigger_irq(struct virtqueue *vq) |
8ca47e00 | 725 | { |
17cbca2b RR |
726 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
727 | ||
20887611 RR |
728 | /* If they don't want an interrupt, don't send one, unless empty. */ |
729 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
730 | && vq->inflight) | |
17cbca2b RR |
731 | return; |
732 | ||
733 | /* Send the Guest an interrupt tell them we used something up. */ | |
56739c80 | 734 | if (write(lguest_fd, buf, sizeof(buf)) != 0) |
17cbca2b | 735 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
736 | } |
737 | ||
17cbca2b | 738 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 739 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 740 | { |
17cbca2b | 741 | add_used(vq, head, len); |
56739c80 | 742 | trigger_irq(vq); |
8ca47e00 RR |
743 | } |
744 | ||
e1e72965 RR |
745 | /* |
746 | * The Console | |
747 | * | |
748 | * Here is the input terminal setting we save, and the routine to restore them | |
749 | * on exit so the user gets their terminal back. */ | |
8ca47e00 RR |
750 | static struct termios orig_term; |
751 | static void restore_term(void) | |
752 | { | |
753 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
754 | } | |
755 | ||
dde79789 | 756 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
757 | struct console_abort |
758 | { | |
dde79789 | 759 | /* How many times have they hit ^C? */ |
8ca47e00 | 760 | int count; |
dde79789 | 761 | /* When did they start? */ |
8ca47e00 RR |
762 | struct timeval start; |
763 | }; | |
764 | ||
dde79789 | 765 | /* This is the routine which handles console input (ie. stdin). */ |
56739c80 | 766 | static bool handle_console_input(struct device *dev) |
8ca47e00 | 767 | { |
8ca47e00 | 768 | int len; |
17cbca2b RR |
769 | unsigned int head, in_num, out_num; |
770 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
771 | struct console_abort *abort = dev->priv; |
772 | ||
17cbca2b RR |
773 | /* First we need a console buffer from the Guests's input virtqueue. */ |
774 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
775 | |
776 | /* If they're not ready for input, stop listening to this file | |
777 | * descriptor. We'll start again once they add an input buffer. */ | |
778 | if (head == dev->vq->vring.num) | |
779 | return false; | |
780 | ||
781 | if (out_num) | |
17cbca2b | 782 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 783 | |
dde79789 RR |
784 | /* This is why we convert to iovecs: the readv() call uses them, and so |
785 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 786 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 787 | if (len <= 0) { |
dde79789 | 788 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 789 | * something went terribly wrong. */ |
8ca47e00 | 790 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 791 | /* Put the input terminal back. */ |
17cbca2b | 792 | restore_term(); |
56ae43df RR |
793 | /* Remove callback from input vq, so it doesn't restart us. */ |
794 | dev->vq->handle_output = NULL; | |
795 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 796 | return false; |
8ca47e00 RR |
797 | } |
798 | ||
56ae43df | 799 | /* Tell the Guest about the new input. */ |
56739c80 | 800 | add_used_and_trigger(dev->vq, head, len); |
8ca47e00 | 801 | |
dde79789 RR |
802 | /* Three ^C within one second? Exit. |
803 | * | |
804 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
805 | * in a buffer by itself, so they can't be too fast. But we check that | |
806 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
807 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
808 | if (!abort->count++) | |
809 | gettimeofday(&abort->start, NULL); | |
810 | else if (abort->count == 3) { | |
811 | struct timeval now; | |
812 | gettimeofday(&now, NULL); | |
813 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 814 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
815 | /* Close the fd so Waker will know it has to |
816 | * exit. */ | |
8c79873d RR |
817 | close(waker_fds.pipe[1]); |
818 | /* Just in case Waker is blocked in BREAK, send | |
dde79789 | 819 | * unbreak now. */ |
56739c80 | 820 | write(lguest_fd, args, sizeof(args)); |
8ca47e00 RR |
821 | exit(2); |
822 | } | |
823 | abort->count = 0; | |
824 | } | |
825 | } else | |
dde79789 | 826 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
827 | abort->count = 0; |
828 | ||
dde79789 | 829 | /* Everything went OK! */ |
8ca47e00 RR |
830 | return true; |
831 | } | |
832 | ||
17cbca2b RR |
833 | /* Handling output for console is simple: we just get all the output buffers |
834 | * and write them to stdout. */ | |
56739c80 | 835 | static void handle_console_output(struct virtqueue *vq, bool timeout) |
8ca47e00 | 836 | { |
17cbca2b RR |
837 | unsigned int head, out, in; |
838 | int len; | |
839 | struct iovec iov[vq->vring.num]; | |
840 | ||
841 | /* Keep getting output buffers from the Guest until we run out. */ | |
842 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
843 | if (in) | |
844 | errx(1, "Input buffers in output queue?"); | |
845 | len = writev(STDOUT_FILENO, iov, out); | |
56739c80 | 846 | add_used_and_trigger(vq, head, len); |
17cbca2b | 847 | } |
8ca47e00 RR |
848 | } |
849 | ||
1dc3e3bc RR |
850 | /* This is called when we no longer want to hear about Guest changes to a |
851 | * virtqueue. This is more efficient in high-traffic cases, but it means we | |
852 | * have to set a timer to check if any more changes have occurred. */ | |
a161883a RR |
853 | static void block_vq(struct virtqueue *vq) |
854 | { | |
855 | struct itimerval itm; | |
856 | ||
857 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
858 | vq->blocked = true; | |
859 | ||
860 | itm.it_interval.tv_sec = 0; | |
861 | itm.it_interval.tv_usec = 0; | |
862 | itm.it_value.tv_sec = 0; | |
aa124984 | 863 | itm.it_value.tv_usec = timeout_usec; |
a161883a RR |
864 | |
865 | setitimer(ITIMER_REAL, &itm, NULL); | |
866 | } | |
867 | ||
e1e72965 RR |
868 | /* |
869 | * The Network | |
870 | * | |
871 | * Handling output for network is also simple: we get all the output buffers | |
17cbca2b | 872 | * and write them (ignoring the first element) to this device's file descriptor |
a6bd8e13 RR |
873 | * (/dev/net/tun). |
874 | */ | |
56739c80 | 875 | static void handle_net_output(struct virtqueue *vq, bool timeout) |
8ca47e00 | 876 | { |
a161883a | 877 | unsigned int head, out, in, num = 0; |
17cbca2b RR |
878 | int len; |
879 | struct iovec iov[vq->vring.num]; | |
aa124984 | 880 | static int last_timeout_num; |
17cbca2b RR |
881 | |
882 | /* Keep getting output buffers from the Guest until we run out. */ | |
883 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
884 | if (in) | |
885 | errx(1, "Input buffers in output queue?"); | |
398f187d RR |
886 | len = writev(vq->dev->fd, iov, out); |
887 | if (len < 0) | |
888 | err(1, "Writing network packet to tun"); | |
56739c80 | 889 | add_used_and_trigger(vq, head, len); |
a161883a | 890 | num++; |
17cbca2b | 891 | } |
a161883a RR |
892 | |
893 | /* Block further kicks and set up a timer if we saw anything. */ | |
894 | if (!timeout && num) | |
895 | block_vq(vq); | |
aa124984 | 896 | |
1dc3e3bc RR |
897 | /* We never quite know how long should we wait before we check the |
898 | * queue again for more packets. We start at 500 microseconds, and if | |
899 | * we get fewer packets than last time, we assume we made the timeout | |
900 | * too small and increase it by 10 microseconds. Otherwise, we drop it | |
901 | * by one microsecond every time. It seems to work well enough. */ | |
aa124984 RR |
902 | if (timeout) { |
903 | if (num < last_timeout_num) | |
904 | timeout_usec += 10; | |
905 | else if (timeout_usec > 1) | |
906 | timeout_usec--; | |
907 | last_timeout_num = num; | |
908 | } | |
8ca47e00 RR |
909 | } |
910 | ||
17cbca2b RR |
911 | /* This is where we handle a packet coming in from the tun device to our |
912 | * Guest. */ | |
56739c80 | 913 | static bool handle_tun_input(struct device *dev) |
8ca47e00 | 914 | { |
17cbca2b | 915 | unsigned int head, in_num, out_num; |
8ca47e00 | 916 | int len; |
17cbca2b | 917 | struct iovec iov[dev->vq->vring.num]; |
8ca47e00 | 918 | |
17cbca2b RR |
919 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
920 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
921 | if (head == dev->vq->vring.num) { | |
dde79789 | 922 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
923 | * early, the Guest won't be ready yet. Wait until the device |
924 | * status says it's ready. */ | |
925 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
5dae785a RR |
926 | |
927 | /* Now tell it we want to know if new things appear. */ | |
928 | dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
929 | wmb(); | |
930 | ||
56ae43df RR |
931 | /* We'll turn this back on if input buffers are registered. */ |
932 | return false; | |
17cbca2b RR |
933 | } else if (out_num) |
934 | errx(1, "Output buffers in network recv queue?"); | |
935 | ||
dde79789 | 936 | /* Read the packet from the device directly into the Guest's buffer. */ |
398f187d | 937 | len = readv(dev->fd, iov, in_num); |
8ca47e00 RR |
938 | if (len <= 0) |
939 | err(1, "reading network"); | |
dde79789 | 940 | |
56ae43df | 941 | /* Tell the Guest about the new packet. */ |
56739c80 | 942 | add_used_and_trigger(dev->vq, head, len); |
17cbca2b | 943 | |
8ca47e00 | 944 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
945 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
946 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
947 | ||
dde79789 | 948 | /* All good. */ |
8ca47e00 RR |
949 | return true; |
950 | } | |
951 | ||
e1e72965 RR |
952 | /*L:215 This is the callback attached to the network and console input |
953 | * virtqueues: it ensures we try again, in case we stopped console or net | |
56ae43df | 954 | * delivery because Guest didn't have any buffers. */ |
56739c80 | 955 | static void enable_fd(struct virtqueue *vq, bool timeout) |
56ae43df RR |
956 | { |
957 | add_device_fd(vq->dev->fd); | |
8c79873d RR |
958 | /* Snap the Waker out of its select loop. */ |
959 | write(waker_fds.pipe[1], "", 1); | |
56ae43df RR |
960 | } |
961 | ||
56739c80 | 962 | static void net_enable_fd(struct virtqueue *vq, bool timeout) |
5dae785a RR |
963 | { |
964 | /* We don't need to know again when Guest refills receive buffer. */ | |
965 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
56739c80 | 966 | enable_fd(vq, timeout); |
5dae785a RR |
967 | } |
968 | ||
a007a751 RR |
969 | /* When the Guest tells us they updated the status field, we handle it. */ |
970 | static void update_device_status(struct device *dev) | |
6e5aa7ef RR |
971 | { |
972 | struct virtqueue *vq; | |
973 | ||
a007a751 RR |
974 | /* This is a reset. */ |
975 | if (dev->desc->status == 0) { | |
976 | verbose("Resetting device %s\n", dev->name); | |
6e5aa7ef | 977 | |
a007a751 | 978 | /* Clear any features they've acked. */ |
713b15b3 RR |
979 | memset(get_feature_bits(dev) + dev->feature_len, 0, |
980 | dev->feature_len); | |
6e5aa7ef | 981 | |
a007a751 RR |
982 | /* Zero out the virtqueues. */ |
983 | for (vq = dev->vq; vq; vq = vq->next) { | |
984 | memset(vq->vring.desc, 0, | |
2966af73 | 985 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); |
b5111790 | 986 | lg_last_avail(vq) = 0; |
a007a751 RR |
987 | } |
988 | } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
989 | warnx("Device %s configuration FAILED", dev->name); | |
990 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { | |
991 | unsigned int i; | |
992 | ||
993 | verbose("Device %s OK: offered", dev->name); | |
713b15b3 | 994 | for (i = 0; i < dev->feature_len; i++) |
32c68e5c | 995 | verbose(" %02x", get_feature_bits(dev)[i]); |
a007a751 | 996 | verbose(", accepted"); |
713b15b3 | 997 | for (i = 0; i < dev->feature_len; i++) |
32c68e5c | 998 | verbose(" %02x", get_feature_bits(dev) |
713b15b3 | 999 | [dev->feature_len+i]); |
a007a751 RR |
1000 | |
1001 | if (dev->ready) | |
1002 | dev->ready(dev); | |
6e5aa7ef RR |
1003 | } |
1004 | } | |
1005 | ||
17cbca2b | 1006 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
56739c80 | 1007 | static void handle_output(unsigned long addr) |
8ca47e00 RR |
1008 | { |
1009 | struct device *i; | |
17cbca2b RR |
1010 | struct virtqueue *vq; |
1011 | ||
6e5aa7ef | 1012 | /* Check each device and virtqueue. */ |
17cbca2b | 1013 | for (i = devices.dev; i; i = i->next) { |
a007a751 | 1014 | /* Notifications to device descriptors update device status. */ |
6e5aa7ef | 1015 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1016 | update_device_status(i); |
6e5aa7ef RR |
1017 | return; |
1018 | } | |
1019 | ||
1020 | /* Notifications to virtqueues mean output has occurred. */ | |
17cbca2b | 1021 | for (vq = i->vq; vq; vq = vq->next) { |
6e5aa7ef RR |
1022 | if (vq->config.pfn != addr/getpagesize()) |
1023 | continue; | |
1024 | ||
1025 | /* Guest should acknowledge (and set features!) before | |
1026 | * using the device. */ | |
1027 | if (i->desc->status == 0) { | |
1028 | warnx("%s gave early output", i->name); | |
17cbca2b RR |
1029 | return; |
1030 | } | |
6e5aa7ef RR |
1031 | |
1032 | if (strcmp(vq->dev->name, "console") != 0) | |
1033 | verbose("Output to %s\n", vq->dev->name); | |
1034 | if (vq->handle_output) | |
56739c80 | 1035 | vq->handle_output(vq, false); |
6e5aa7ef | 1036 | return; |
8ca47e00 RR |
1037 | } |
1038 | } | |
dde79789 | 1039 | |
17cbca2b RR |
1040 | /* Early console write is done using notify on a nul-terminated string |
1041 | * in Guest memory. */ | |
1042 | if (addr >= guest_limit) | |
1043 | errx(1, "Bad NOTIFY %#lx", addr); | |
1044 | ||
1045 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1046 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1047 | } |
1048 | ||
56739c80 | 1049 | static void handle_timeout(void) |
a161883a RR |
1050 | { |
1051 | char buf[32]; | |
1052 | struct device *i; | |
1053 | struct virtqueue *vq; | |
1054 | ||
1055 | /* Clear the pipe */ | |
1056 | read(timeoutpipe[0], buf, sizeof(buf)); | |
1057 | ||
1058 | /* Check each device and virtqueue: flush blocked ones. */ | |
1059 | for (i = devices.dev; i; i = i->next) { | |
1060 | for (vq = i->vq; vq; vq = vq->next) { | |
1061 | if (!vq->blocked) | |
1062 | continue; | |
1063 | ||
1064 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
1065 | vq->blocked = false; | |
1066 | if (vq->handle_output) | |
56739c80 | 1067 | vq->handle_output(vq, true); |
a161883a RR |
1068 | } |
1069 | } | |
1070 | } | |
1071 | ||
e1e72965 | 1072 | /* This is called when the Waker wakes us up: check for incoming file |
dde79789 | 1073 | * descriptors. */ |
56739c80 | 1074 | static void handle_input(void) |
8ca47e00 | 1075 | { |
dde79789 | 1076 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
1077 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
1078 | ||
1079 | for (;;) { | |
1080 | struct device *i; | |
17cbca2b | 1081 | fd_set fds = devices.infds; |
a161883a | 1082 | int num; |
8ca47e00 | 1083 | |
a161883a RR |
1084 | num = select(devices.max_infd+1, &fds, NULL, NULL, &poll); |
1085 | /* Could get interrupted */ | |
1086 | if (num < 0) | |
1087 | continue; | |
dde79789 | 1088 | /* If nothing is ready, we're done. */ |
a161883a | 1089 | if (num == 0) |
8ca47e00 RR |
1090 | break; |
1091 | ||
a6bd8e13 RR |
1092 | /* Otherwise, call the device(s) which have readable file |
1093 | * descriptors and a method of handling them. */ | |
17cbca2b | 1094 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 1095 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56739c80 | 1096 | if (i->handle_input(i)) |
56ae43df RR |
1097 | continue; |
1098 | ||
dde79789 | 1099 | /* If handle_input() returns false, it means we |
56ae43df RR |
1100 | * should no longer service it. Networking and |
1101 | * console do this when there's no input | |
1102 | * buffers to deliver into. Console also uses | |
a6bd8e13 | 1103 | * it when it discovers that stdin is closed. */ |
56ae43df | 1104 | FD_CLR(i->fd, &devices.infds); |
8ca47e00 RR |
1105 | } |
1106 | } | |
a161883a RR |
1107 | |
1108 | /* Is this the timeout fd? */ | |
1109 | if (FD_ISSET(timeoutpipe[0], &fds)) | |
56739c80 | 1110 | handle_timeout(); |
8ca47e00 RR |
1111 | } |
1112 | } | |
1113 | ||
dde79789 RR |
1114 | /*L:190 |
1115 | * Device Setup | |
1116 | * | |
1117 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1118 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1119 | * routines to allocate and manage them. |
1120 | */ | |
8ca47e00 | 1121 | |
a586d4f6 RR |
1122 | /* The layout of the device page is a "struct lguest_device_desc" followed by a |
1123 | * number of virtqueue descriptors, then two sets of feature bits, then an | |
1124 | * array of configuration bytes. This routine returns the configuration | |
1125 | * pointer. */ | |
1126 | static u8 *device_config(const struct device *dev) | |
1127 | { | |
1128 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1129 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1130 | + dev->feature_len * 2; | |
17cbca2b RR |
1131 | } |
1132 | ||
a586d4f6 RR |
1133 | /* This routine allocates a new "struct lguest_device_desc" from descriptor |
1134 | * table page just above the Guest's normal memory. It returns a pointer to | |
1135 | * that descriptor. */ | |
1136 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
17cbca2b | 1137 | { |
a586d4f6 RR |
1138 | struct lguest_device_desc d = { .type = type }; |
1139 | void *p; | |
17cbca2b | 1140 | |
a586d4f6 RR |
1141 | /* Figure out where the next device config is, based on the last one. */ |
1142 | if (devices.lastdev) | |
1143 | p = device_config(devices.lastdev) | |
1144 | + devices.lastdev->desc->config_len; | |
1145 | else | |
1146 | p = devices.descpage; | |
17cbca2b | 1147 | |
a586d4f6 RR |
1148 | /* We only have one page for all the descriptors. */ |
1149 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1150 | errx(1, "Too many devices"); | |
17cbca2b | 1151 | |
a586d4f6 RR |
1152 | /* p might not be aligned, so we memcpy in. */ |
1153 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1154 | } |
1155 | ||
a586d4f6 RR |
1156 | /* Each device descriptor is followed by the description of its virtqueues. We |
1157 | * specify how many descriptors the virtqueue is to have. */ | |
17cbca2b | 1158 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
56739c80 | 1159 | void (*handle_output)(struct virtqueue *, bool)) |
17cbca2b RR |
1160 | { |
1161 | unsigned int pages; | |
1162 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1163 | void *p; | |
1164 | ||
a6bd8e13 | 1165 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1166 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 1167 | / getpagesize(); |
17cbca2b RR |
1168 | p = get_pages(pages); |
1169 | ||
d1c856e0 RR |
1170 | /* Initialize the virtqueue */ |
1171 | vq->next = NULL; | |
1172 | vq->last_avail_idx = 0; | |
1173 | vq->dev = dev; | |
20887611 | 1174 | vq->inflight = 0; |
a161883a | 1175 | vq->blocked = false; |
d1c856e0 | 1176 | |
17cbca2b RR |
1177 | /* Initialize the configuration. */ |
1178 | vq->config.num = num_descs; | |
1179 | vq->config.irq = devices.next_irq++; | |
1180 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1181 | ||
1182 | /* Initialize the vring. */ | |
2966af73 | 1183 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1184 | |
a586d4f6 RR |
1185 | /* Append virtqueue to this device's descriptor. We use |
1186 | * device_config() to get the end of the device's current virtqueues; | |
1187 | * we check that we haven't added any config or feature information | |
1188 | * yet, otherwise we'd be overwriting them. */ | |
1189 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); | |
1190 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1191 | dev->num_vq++; |
a586d4f6 RR |
1192 | dev->desc->num_vq++; |
1193 | ||
1194 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b RR |
1195 | |
1196 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1197 | * second. */ | |
1198 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1199 | *i = vq; | |
1200 | ||
e1e72965 RR |
1201 | /* Set the routine to call when the Guest does something to this |
1202 | * virtqueue. */ | |
17cbca2b | 1203 | vq->handle_output = handle_output; |
e1e72965 | 1204 | |
426e3e0a RR |
1205 | /* As an optimization, set the advisory "Don't Notify Me" flag if we |
1206 | * don't have a handler */ | |
17cbca2b RR |
1207 | if (!handle_output) |
1208 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1209 | } |
1210 | ||
6e5aa7ef | 1211 | /* The first half of the feature bitmask is for us to advertise features. The |
a6bd8e13 | 1212 | * second half is for the Guest to accept features. */ |
a586d4f6 RR |
1213 | static void add_feature(struct device *dev, unsigned bit) |
1214 | { | |
6e5aa7ef | 1215 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1216 | |
1217 | /* We can't extend the feature bits once we've added config bytes */ | |
1218 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1219 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1220 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1221 | } |
1222 | ||
a586d4f6 RR |
1223 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1224 | } | |
1225 | ||
1226 | /* This routine sets the configuration fields for an existing device's | |
1227 | * descriptor. It only works for the last device, but that's OK because that's | |
1228 | * how we use it. */ | |
1229 | static void set_config(struct device *dev, unsigned len, const void *conf) | |
1230 | { | |
1231 | /* Check we haven't overflowed our single page. */ | |
1232 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1233 | errx(1, "Too many devices"); | |
1234 | ||
1235 | /* Copy in the config information, and store the length. */ | |
1236 | memcpy(device_config(dev), conf, len); | |
1237 | dev->desc->config_len = len; | |
1238 | } | |
1239 | ||
17cbca2b | 1240 | /* This routine does all the creation and setup of a new device, including |
a6bd8e13 RR |
1241 | * calling new_dev_desc() to allocate the descriptor and device memory. |
1242 | * | |
1243 | * See what I mean about userspace being boring? */ | |
17cbca2b | 1244 | static struct device *new_device(const char *name, u16 type, int fd, |
56739c80 | 1245 | bool (*handle_input)(struct device *)) |
8ca47e00 RR |
1246 | { |
1247 | struct device *dev = malloc(sizeof(*dev)); | |
1248 | ||
dde79789 | 1249 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1250 | dev->fd = fd; |
dde79789 RR |
1251 | /* If we have an input handler for this file descriptor, then we add it |
1252 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1253 | if (handle_input) |
17cbca2b RR |
1254 | add_device_fd(dev->fd); |
1255 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1256 | dev->handle_input = handle_input; |
17cbca2b | 1257 | dev->name = name; |
d1c856e0 | 1258 | dev->vq = NULL; |
a007a751 | 1259 | dev->ready = NULL; |
713b15b3 RR |
1260 | dev->feature_len = 0; |
1261 | dev->num_vq = 0; | |
a586d4f6 RR |
1262 | |
1263 | /* Append to device list. Prepending to a single-linked list is | |
1264 | * easier, but the user expects the devices to be arranged on the bus | |
1265 | * in command-line order. The first network device on the command line | |
1266 | * is eth0, the first block device /dev/vda, etc. */ | |
1267 | if (devices.lastdev) | |
1268 | devices.lastdev->next = dev; | |
1269 | else | |
1270 | devices.dev = dev; | |
1271 | devices.lastdev = dev; | |
1272 | ||
8ca47e00 RR |
1273 | return dev; |
1274 | } | |
1275 | ||
dde79789 RR |
1276 | /* Our first setup routine is the console. It's a fairly simple device, but |
1277 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1278 | static void setup_console(void) |
8ca47e00 RR |
1279 | { |
1280 | struct device *dev; | |
1281 | ||
dde79789 | 1282 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1283 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1284 | struct termios term = orig_term; | |
dde79789 RR |
1285 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1286 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1287 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1288 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1289 | /* If we exit gracefully, the original settings will be |
1290 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1291 | atexit(restore_term); |
1292 | } | |
1293 | ||
17cbca2b RR |
1294 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1295 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1296 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1297 | dev->priv = malloc(sizeof(struct console_abort)); |
1298 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1299 | |
56ae43df RR |
1300 | /* The console needs two virtqueues: the input then the output. When |
1301 | * they put something the input queue, we make sure we're listening to | |
1302 | * stdin. When they put something in the output queue, we write it to | |
e1e72965 | 1303 | * stdout. */ |
56ae43df | 1304 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
17cbca2b RR |
1305 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1306 | ||
1307 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1308 | } |
17cbca2b | 1309 | /*:*/ |
8ca47e00 | 1310 | |
a161883a RR |
1311 | static void timeout_alarm(int sig) |
1312 | { | |
1313 | write(timeoutpipe[1], "", 1); | |
1314 | } | |
1315 | ||
1316 | static void setup_timeout(void) | |
1317 | { | |
1318 | if (pipe(timeoutpipe) != 0) | |
1319 | err(1, "Creating timeout pipe"); | |
1320 | ||
1321 | if (fcntl(timeoutpipe[1], F_SETFL, | |
1322 | fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0) | |
1323 | err(1, "Making timeout pipe nonblocking"); | |
1324 | ||
1325 | add_device_fd(timeoutpipe[0]); | |
1326 | signal(SIGALRM, timeout_alarm); | |
1327 | } | |
1328 | ||
17cbca2b RR |
1329 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1330 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1331 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1332 | * |
17cbca2b RR |
1333 | * More sopisticated is to use one of the tools developed for project like UML |
1334 | * to do networking. | |
dde79789 | 1335 | * |
17cbca2b RR |
1336 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1337 | * completely generic ("here's my vring, attach to your vring") and would work | |
1338 | * for any traffic. Of course, namespace and permissions issues need to be | |
1339 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1340 | * multiple inter-guest channels behind one interface, although it would | |
1341 | * require some manner of hotplugging new virtio channels. | |
1342 | * | |
1343 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1344 | |
1345 | static u32 str2ip(const char *ipaddr) | |
1346 | { | |
dec6a2be | 1347 | unsigned int b[4]; |
8ca47e00 | 1348 | |
dec6a2be MM |
1349 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1350 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1351 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1352 | } | |
1353 | ||
1354 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1355 | { | |
1356 | unsigned int m[6]; | |
1357 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1358 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1359 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1360 | mac[0] = m[0]; | |
1361 | mac[1] = m[1]; | |
1362 | mac[2] = m[2]; | |
1363 | mac[3] = m[3]; | |
1364 | mac[4] = m[4]; | |
1365 | mac[5] = m[5]; | |
8ca47e00 RR |
1366 | } |
1367 | ||
dde79789 RR |
1368 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1369 | * network device to the bridge device specified by the command line. | |
1370 | * | |
1371 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1372 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1373 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1374 | { | |
1375 | int ifidx; | |
1376 | struct ifreq ifr; | |
1377 | ||
1378 | if (!*br_name) | |
1379 | errx(1, "must specify bridge name"); | |
1380 | ||
1381 | ifidx = if_nametoindex(if_name); | |
1382 | if (!ifidx) | |
1383 | errx(1, "interface %s does not exist!", if_name); | |
1384 | ||
1385 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1386 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1387 | ifr.ifr_ifindex = ifidx; |
1388 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1389 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1390 | } | |
1391 | ||
dde79789 RR |
1392 | /* This sets up the Host end of the network device with an IP address, brings |
1393 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1394 | * pointer. */ |
dec6a2be | 1395 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1396 | { |
1397 | struct ifreq ifr; | |
1398 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1399 | ||
1400 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1401 | strcpy(ifr.ifr_name, tapif); |
1402 | ||
1403 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
8ca47e00 RR |
1404 | sin->sin_family = AF_INET; |
1405 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1406 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
dec6a2be | 1407 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1408 | ifr.ifr_flags = IFF_UP; |
1409 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1410 | err(1, "Bringing interface %s up", tapif); |
1411 | } | |
1412 | ||
dec6a2be | 1413 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1414 | { |
8ca47e00 | 1415 | struct ifreq ifr; |
dec6a2be MM |
1416 | int netfd; |
1417 | ||
1418 | /* Start with this zeroed. Messy but sure. */ | |
1419 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1420 | |
dde79789 RR |
1421 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1422 | * tap device is like a tun device, only somehow different. To tell | |
1423 | * the truth, I completely blundered my way through this code, but it | |
1424 | * works now! */ | |
8ca47e00 | 1425 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1426 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1427 | strcpy(ifr.ifr_name, "tap%d"); |
1428 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1429 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1430 | |
398f187d RR |
1431 | if (ioctl(netfd, TUNSETOFFLOAD, |
1432 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1433 | err(1, "Could not set features for tun device"); | |
1434 | ||
dde79789 RR |
1435 | /* We don't need checksums calculated for packets coming in this |
1436 | * device: trust us! */ | |
8ca47e00 RR |
1437 | ioctl(netfd, TUNSETNOCSUM, 1); |
1438 | ||
dec6a2be MM |
1439 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1440 | return netfd; | |
1441 | } | |
1442 | ||
1443 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or | |
1444 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1445 | * packets into the Host as if they came in from a normal network card. We | |
1446 | * just shunt packets between the Guest and the tun device. */ | |
1447 | static void setup_tun_net(char *arg) | |
1448 | { | |
1449 | struct device *dev; | |
1450 | int netfd, ipfd; | |
1451 | u32 ip = INADDR_ANY; | |
1452 | bool bridging = false; | |
1453 | char tapif[IFNAMSIZ], *p; | |
1454 | struct virtio_net_config conf; | |
1455 | ||
1456 | netfd = get_tun_device(tapif); | |
1457 | ||
17cbca2b RR |
1458 | /* First we create a new network device. */ |
1459 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1460 | |
56ae43df RR |
1461 | /* Network devices need a receive and a send queue, just like |
1462 | * console. */ | |
5dae785a | 1463 | add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd); |
17cbca2b | 1464 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1465 | |
dde79789 RR |
1466 | /* We need a socket to perform the magic network ioctls to bring up the |
1467 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1468 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1469 | if (ipfd < 0) | |
1470 | err(1, "opening IP socket"); | |
1471 | ||
dde79789 | 1472 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1473 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1474 | arg += strlen(BRIDGE_PFX); |
1475 | bridging = true; | |
1476 | } | |
1477 | ||
1478 | /* A mac address may follow the bridge name or IP address */ | |
1479 | p = strchr(arg, ':'); | |
1480 | if (p) { | |
1481 | str2mac(p+1, conf.mac); | |
40c42076 | 1482 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1483 | *p = '\0'; |
dec6a2be MM |
1484 | } |
1485 | ||
1486 | /* arg is now either an IP address or a bridge name */ | |
1487 | if (bridging) | |
1488 | add_to_bridge(ipfd, tapif, arg); | |
1489 | else | |
8ca47e00 RR |
1490 | ip = str2ip(arg); |
1491 | ||
dec6a2be MM |
1492 | /* Set up the tun device. */ |
1493 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1494 | |
20887611 | 1495 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1496 | /* Expect Guest to handle everything except UFO */ |
1497 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1498 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1499 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1500 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1501 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1502 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1503 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1504 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
a586d4f6 | 1505 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1506 | |
a586d4f6 | 1507 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1508 | close(ipfd); |
1509 | ||
dec6a2be MM |
1510 | devices.device_num++; |
1511 | ||
1512 | if (bridging) | |
1513 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1514 | devices.device_num, tapif, arg); | |
1515 | else | |
1516 | verbose("device %u: tun %s: %s\n", | |
1517 | devices.device_num, tapif, arg); | |
8ca47e00 | 1518 | } |
17cbca2b | 1519 | |
e1e72965 RR |
1520 | /* Our block (disk) device should be really simple: the Guest asks for a block |
1521 | * number and we read or write that position in the file. Unfortunately, that | |
1522 | * was amazingly slow: the Guest waits until the read is finished before | |
1523 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1524 | * |
e1e72965 RR |
1525 | * We could use async I/O, except it's reputed to suck so hard that characters |
1526 | * actually go missing from your code when you try to use it. | |
17cbca2b RR |
1527 | * |
1528 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1529 | ||
e1e72965 | 1530 | /* This hangs off device->priv. */ |
17cbca2b RR |
1531 | struct vblk_info |
1532 | { | |
1533 | /* The size of the file. */ | |
1534 | off64_t len; | |
1535 | ||
1536 | /* The file descriptor for the file. */ | |
1537 | int fd; | |
1538 | ||
1539 | /* IO thread listens on this file descriptor [0]. */ | |
1540 | int workpipe[2]; | |
1541 | ||
1542 | /* IO thread writes to this file descriptor to mark it done, then | |
1543 | * Launcher triggers interrupt to Guest. */ | |
1544 | int done_fd; | |
1545 | }; | |
1546 | ||
e1e72965 RR |
1547 | /*L:210 |
1548 | * The Disk | |
1549 | * | |
1550 | * Remember that the block device is handled by a separate I/O thread. We head | |
1551 | * straight into the core of that thread here: | |
1552 | */ | |
17cbca2b RR |
1553 | static bool service_io(struct device *dev) |
1554 | { | |
1555 | struct vblk_info *vblk = dev->priv; | |
1556 | unsigned int head, out_num, in_num, wlen; | |
1557 | int ret; | |
cb38fa23 | 1558 | u8 *in; |
17cbca2b RR |
1559 | struct virtio_blk_outhdr *out; |
1560 | struct iovec iov[dev->vq->vring.num]; | |
1561 | off64_t off; | |
1562 | ||
e1e72965 | 1563 | /* See if there's a request waiting. If not, nothing to do. */ |
17cbca2b RR |
1564 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
1565 | if (head == dev->vq->vring.num) | |
1566 | return false; | |
1567 | ||
e1e72965 RR |
1568 | /* Every block request should contain at least one output buffer |
1569 | * (detailing the location on disk and the type of request) and one | |
1570 | * input buffer (to hold the result). */ | |
17cbca2b RR |
1571 | if (out_num == 0 || in_num == 0) |
1572 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1573 | head, out_num, in_num); | |
1574 | ||
1575 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1576 | in = convert(&iov[out_num+in_num-1], u8); |
17cbca2b RR |
1577 | off = out->sector * 512; |
1578 | ||
e1e72965 RR |
1579 | /* The block device implements "barriers", where the Guest indicates |
1580 | * that it wants all previous writes to occur before this write. We | |
1581 | * don't have a way of asking our kernel to do a barrier, so we just | |
1582 | * synchronize all the data in the file. Pretty poor, no? */ | |
17cbca2b RR |
1583 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1584 | fdatasync(vblk->fd); | |
1585 | ||
e1e72965 RR |
1586 | /* In general the virtio block driver is allowed to try SCSI commands. |
1587 | * It'd be nice if we supported eject, for example, but we don't. */ | |
17cbca2b RR |
1588 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1589 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1590 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1591 | wlen = sizeof(*in); |
17cbca2b RR |
1592 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
1593 | /* Write */ | |
1594 | ||
1595 | /* Move to the right location in the block file. This can fail | |
1596 | * if they try to write past end. */ | |
1597 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1598 | err(1, "Bad seek to sector %llu", out->sector); | |
1599 | ||
1600 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1601 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1602 | ||
1603 | /* Grr... Now we know how long the descriptor they sent was, we | |
1604 | * make sure they didn't try to write over the end of the block | |
1605 | * file (possibly extending it). */ | |
1606 | if (ret > 0 && off + ret > vblk->len) { | |
1607 | /* Trim it back to the correct length */ | |
1608 | ftruncate64(vblk->fd, vblk->len); | |
1609 | /* Die, bad Guest, die. */ | |
1610 | errx(1, "Write past end %llu+%u", off, ret); | |
1611 | } | |
1200e646 | 1612 | wlen = sizeof(*in); |
cb38fa23 | 1613 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b RR |
1614 | } else { |
1615 | /* Read */ | |
1616 | ||
1617 | /* Move to the right location in the block file. This can fail | |
1618 | * if they try to read past end. */ | |
1619 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1620 | err(1, "Bad seek to sector %llu", out->sector); | |
1621 | ||
1622 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1623 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1624 | if (ret >= 0) { | |
1200e646 | 1625 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1626 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1627 | } else { |
1200e646 | 1628 | wlen = sizeof(*in); |
cb38fa23 | 1629 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1630 | } |
1631 | } | |
1632 | ||
d1881d31 RR |
1633 | /* OK, so we noted that it was pretty poor to use an fdatasync as a |
1634 | * barrier. But Christoph Hellwig points out that we need a sync | |
1635 | * *afterwards* as well: "Barriers specify no reordering to the front | |
1636 | * or the back." And Jens Axboe confirmed it, so here we are: */ | |
1637 | if (out->type & VIRTIO_BLK_T_BARRIER) | |
1638 | fdatasync(vblk->fd); | |
1639 | ||
17cbca2b RR |
1640 | /* We can't trigger an IRQ, because we're not the Launcher. It does |
1641 | * that when we tell it we're done. */ | |
1642 | add_used(dev->vq, head, wlen); | |
1643 | return true; | |
1644 | } | |
1645 | ||
1646 | /* This is the thread which actually services the I/O. */ | |
1647 | static int io_thread(void *_dev) | |
1648 | { | |
1649 | struct device *dev = _dev; | |
1650 | struct vblk_info *vblk = dev->priv; | |
1651 | char c; | |
1652 | ||
1653 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1654 | close(vblk->workpipe[1]); | |
1655 | /* Close the other side of the done_fd pipe. */ | |
1656 | close(dev->fd); | |
1657 | ||
1658 | /* When this read fails, it means Launcher died, so we follow. */ | |
1659 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
e1e72965 | 1660 | /* We acknowledge each request immediately to reduce latency, |
17cbca2b | 1661 | * rather than waiting until we've done them all. I haven't |
a6bd8e13 RR |
1662 | * measured to see if it makes any difference. |
1663 | * | |
1664 | * That would be an interesting test, wouldn't it? You could | |
1665 | * also try having more than one I/O thread. */ | |
17cbca2b RR |
1666 | while (service_io(dev)) |
1667 | write(vblk->done_fd, &c, 1); | |
1668 | } | |
1669 | return 0; | |
1670 | } | |
1671 | ||
e1e72965 | 1672 | /* Now we've seen the I/O thread, we return to the Launcher to see what happens |
a6bd8e13 | 1673 | * when that thread tells us it's completed some I/O. */ |
56739c80 | 1674 | static bool handle_io_finish(struct device *dev) |
17cbca2b RR |
1675 | { |
1676 | char c; | |
1677 | ||
e1e72965 RR |
1678 | /* If the I/O thread died, presumably it printed the error, so we |
1679 | * simply exit. */ | |
17cbca2b RR |
1680 | if (read(dev->fd, &c, 1) != 1) |
1681 | exit(1); | |
1682 | ||
1683 | /* It did some work, so trigger the irq. */ | |
56739c80 | 1684 | trigger_irq(dev->vq); |
17cbca2b RR |
1685 | return true; |
1686 | } | |
1687 | ||
e1e72965 | 1688 | /* When the Guest submits some I/O, we just need to wake the I/O thread. */ |
56739c80 | 1689 | static void handle_virtblk_output(struct virtqueue *vq, bool timeout) |
17cbca2b RR |
1690 | { |
1691 | struct vblk_info *vblk = vq->dev->priv; | |
1692 | char c = 0; | |
1693 | ||
1694 | /* Wake up I/O thread and tell it to go to work! */ | |
1695 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1696 | /* Presumably it indicated why it died. */ | |
1697 | exit(1); | |
1698 | } | |
1699 | ||
e1e72965 | 1700 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1701 | static void setup_block_file(const char *filename) |
1702 | { | |
1703 | int p[2]; | |
1704 | struct device *dev; | |
1705 | struct vblk_info *vblk; | |
1706 | void *stack; | |
a586d4f6 | 1707 | struct virtio_blk_config conf; |
17cbca2b RR |
1708 | |
1709 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1710 | pipe(p); | |
1711 | ||
1712 | /* The device responds to return from I/O thread. */ | |
1713 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1714 | ||
e1e72965 | 1715 | /* The device has one virtqueue, where the Guest places requests. */ |
17cbca2b RR |
1716 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); |
1717 | ||
1718 | /* Allocate the room for our own bookkeeping */ | |
1719 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1720 | ||
1721 | /* First we open the file and store the length. */ | |
1722 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1723 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1724 | ||
a586d4f6 RR |
1725 | /* We support barriers. */ |
1726 | add_feature(dev, VIRTIO_BLK_F_BARRIER); | |
1727 | ||
17cbca2b | 1728 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1729 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b RR |
1730 | |
1731 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1732 | * for the in and out elements. */ | |
a586d4f6 RR |
1733 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1734 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1735 | ||
1736 | set_config(dev, sizeof(conf), &conf); | |
17cbca2b RR |
1737 | |
1738 | /* The I/O thread writes to this end of the pipe when done. */ | |
1739 | vblk->done_fd = p[1]; | |
1740 | ||
e1e72965 RR |
1741 | /* This is the second pipe, which is how we tell the I/O thread about |
1742 | * more work. */ | |
17cbca2b RR |
1743 | pipe(vblk->workpipe); |
1744 | ||
a6bd8e13 RR |
1745 | /* Create stack for thread and run it. Since stack grows upwards, we |
1746 | * point the stack pointer to the end of this region. */ | |
17cbca2b | 1747 | stack = malloc(32768); |
ec04b13f BR |
1748 | /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from |
1749 | * becoming a zombie. */ | |
a6bd8e13 | 1750 | if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1) |
17cbca2b RR |
1751 | err(1, "Creating clone"); |
1752 | ||
1753 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1754 | close(vblk->done_fd); | |
1755 | close(vblk->workpipe[0]); | |
1756 | ||
1757 | verbose("device %u: virtblock %llu sectors\n", | |
a586d4f6 | 1758 | devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1759 | } |
28fd6d7f RR |
1760 | |
1761 | /* Our random number generator device reads from /dev/random into the Guest's | |
1762 | * input buffers. The usual case is that the Guest doesn't want random numbers | |
1763 | * and so has no buffers although /dev/random is still readable, whereas | |
1764 | * console is the reverse. | |
1765 | * | |
1766 | * The same logic applies, however. */ | |
56739c80 | 1767 | static bool handle_rng_input(struct device *dev) |
28fd6d7f RR |
1768 | { |
1769 | int len; | |
1770 | unsigned int head, in_num, out_num, totlen = 0; | |
1771 | struct iovec iov[dev->vq->vring.num]; | |
1772 | ||
1773 | /* First we need a buffer from the Guests's virtqueue. */ | |
1774 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
1775 | ||
1776 | /* If they're not ready for input, stop listening to this file | |
1777 | * descriptor. We'll start again once they add an input buffer. */ | |
1778 | if (head == dev->vq->vring.num) | |
1779 | return false; | |
1780 | ||
1781 | if (out_num) | |
1782 | errx(1, "Output buffers in rng?"); | |
1783 | ||
1784 | /* This is why we convert to iovecs: the readv() call uses them, and so | |
1785 | * it reads straight into the Guest's buffer. We loop to make sure we | |
1786 | * fill it. */ | |
1787 | while (!iov_empty(iov, in_num)) { | |
1788 | len = readv(dev->fd, iov, in_num); | |
1789 | if (len <= 0) | |
1790 | err(1, "Read from /dev/random gave %i", len); | |
1791 | iov_consume(iov, in_num, len); | |
1792 | totlen += len; | |
1793 | } | |
1794 | ||
1795 | /* Tell the Guest about the new input. */ | |
56739c80 | 1796 | add_used_and_trigger(dev->vq, head, totlen); |
28fd6d7f RR |
1797 | |
1798 | /* Everything went OK! */ | |
1799 | return true; | |
1800 | } | |
1801 | ||
1802 | /* And this creates a "hardware" random number device for the Guest. */ | |
1803 | static void setup_rng(void) | |
1804 | { | |
1805 | struct device *dev; | |
1806 | int fd; | |
1807 | ||
1808 | fd = open_or_die("/dev/random", O_RDONLY); | |
1809 | ||
1810 | /* The device responds to return from I/O thread. */ | |
1811 | dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input); | |
1812 | ||
1813 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
1814 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
1815 | ||
1816 | verbose("device %u: rng\n", devices.device_num++); | |
1817 | } | |
a6bd8e13 | 1818 | /* That's the end of device setup. */ |
ec04b13f | 1819 | |
a6bd8e13 | 1820 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1821 | static void __attribute__((noreturn)) restart_guest(void) |
1822 | { | |
1823 | unsigned int i; | |
1824 | ||
8c79873d RR |
1825 | /* Since we don't track all open fds, we simply close everything beyond |
1826 | * stderr. */ | |
ec04b13f BR |
1827 | for (i = 3; i < FD_SETSIZE; i++) |
1828 | close(i); | |
8c79873d RR |
1829 | |
1830 | /* The exec automatically gets rid of the I/O and Waker threads. */ | |
ec04b13f BR |
1831 | execv(main_args[0], main_args); |
1832 | err(1, "Could not exec %s", main_args[0]); | |
1833 | } | |
8ca47e00 | 1834 | |
a6bd8e13 | 1835 | /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves |
dde79789 | 1836 | * its input and output, and finally, lays it to rest. */ |
56739c80 | 1837 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
1838 | { |
1839 | for (;;) { | |
511801dc | 1840 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1841 | unsigned long notify_addr; |
8ca47e00 RR |
1842 | int readval; |
1843 | ||
1844 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1845 | readval = pread(lguest_fd, ¬ify_addr, |
1846 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1847 | |
17cbca2b RR |
1848 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1849 | if (readval == sizeof(notify_addr)) { | |
1850 | verbose("Notify on address %#lx\n", notify_addr); | |
56739c80 | 1851 | handle_output(notify_addr); |
8ca47e00 | 1852 | continue; |
dde79789 | 1853 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1854 | } else if (errno == ENOENT) { |
1855 | char reason[1024] = { 0 }; | |
e3283fa0 | 1856 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1857 | errx(1, "%s", reason); |
ec04b13f BR |
1858 | /* ERESTART means that we need to reboot the guest */ |
1859 | } else if (errno == ERESTART) { | |
1860 | restart_guest(); | |
a161883a | 1861 | /* EAGAIN means a signal (timeout). |
dde79789 | 1862 | * Anything else means a bug or incompatible change. */ |
8ca47e00 RR |
1863 | } else if (errno != EAGAIN) |
1864 | err(1, "Running guest failed"); | |
dde79789 | 1865 | |
e3283fa0 GOC |
1866 | /* Only service input on thread for CPU 0. */ |
1867 | if (cpu_id != 0) | |
1868 | continue; | |
1869 | ||
e1e72965 | 1870 | /* Service input, then unset the BREAK to release the Waker. */ |
56739c80 | 1871 | handle_input(); |
e3283fa0 | 1872 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) |
8ca47e00 RR |
1873 | err(1, "Resetting break"); |
1874 | } | |
1875 | } | |
a6bd8e13 | 1876 | /*L:240 |
e1e72965 RR |
1877 | * This is the end of the Launcher. The good news: we are over halfway |
1878 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1879 | * of us. | |
dde79789 | 1880 | * |
e1e72965 RR |
1881 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1882 | * "make Host". | |
1883 | :*/ | |
8ca47e00 RR |
1884 | |
1885 | static struct option opts[] = { | |
1886 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1887 | { "tunnet", 1, NULL, 't' }, |
1888 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1889 | { "rng", 0, NULL, 'r' }, |
8ca47e00 RR |
1890 | { "initrd", 1, NULL, 'i' }, |
1891 | { NULL }, | |
1892 | }; | |
1893 | static void usage(void) | |
1894 | { | |
1895 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1896 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1897 | "|--block=<filename>|--initrd=<filename>]...\n" |
1898 | "<mem-in-mb> vmlinux [args...]"); | |
1899 | } | |
1900 | ||
3c6b5bfa | 1901 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1902 | int main(int argc, char *argv[]) |
1903 | { | |
47436aa4 RR |
1904 | /* Memory, top-level pagetable, code startpoint and size of the |
1905 | * (optional) initrd. */ | |
58a24566 | 1906 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
1907 | /* Two temporaries. */ |
1908 | int i, c; | |
3c6b5bfa | 1909 | /* The boot information for the Guest. */ |
43d33b21 | 1910 | struct boot_params *boot; |
dde79789 | 1911 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1912 | const char *initrd_name = NULL; |
1913 | ||
ec04b13f BR |
1914 | /* Save the args: we "reboot" by execing ourselves again. */ |
1915 | main_args = argv; | |
1916 | /* We don't "wait" for the children, so prevent them from becoming | |
1917 | * zombies. */ | |
1918 | signal(SIGCHLD, SIG_IGN); | |
1919 | ||
dde79789 RR |
1920 | /* First we initialize the device list. Since console and network |
1921 | * device receive input from a file descriptor, we keep an fdset | |
1922 | * (infds) and the maximum fd number (max_infd) with the head of the | |
a586d4f6 | 1923 | * list. We also keep a pointer to the last device. Finally, we keep |
a6bd8e13 RR |
1924 | * the next interrupt number to use for devices (1: remember that 0 is |
1925 | * used by the timer). */ | |
17cbca2b RR |
1926 | FD_ZERO(&devices.infds); |
1927 | devices.max_infd = -1; | |
a586d4f6 | 1928 | devices.lastdev = NULL; |
17cbca2b | 1929 | devices.next_irq = 1; |
8ca47e00 | 1930 | |
e3283fa0 | 1931 | cpu_id = 0; |
dde79789 RR |
1932 | /* We need to know how much memory so we can set up the device |
1933 | * descriptor and memory pages for the devices as we parse the command | |
1934 | * line. So we quickly look through the arguments to find the amount | |
1935 | * of memory now. */ | |
6570c459 RR |
1936 | for (i = 1; i < argc; i++) { |
1937 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1938 | mem = atoi(argv[i]) * 1024 * 1024; |
1939 | /* We start by mapping anonymous pages over all of | |
1940 | * guest-physical memory range. This fills it with 0, | |
1941 | * and ensures that the Guest won't be killed when it | |
1942 | * tries to access it. */ | |
1943 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1944 | + DEVICE_PAGES); | |
1945 | guest_limit = mem; | |
1946 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1947 | devices.descpage = get_pages(1); |
6570c459 RR |
1948 | break; |
1949 | } | |
1950 | } | |
dde79789 RR |
1951 | |
1952 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1953 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1954 | switch (c) { | |
1955 | case 'v': | |
1956 | verbose = true; | |
1957 | break; | |
8ca47e00 | 1958 | case 't': |
17cbca2b | 1959 | setup_tun_net(optarg); |
8ca47e00 RR |
1960 | break; |
1961 | case 'b': | |
17cbca2b | 1962 | setup_block_file(optarg); |
8ca47e00 | 1963 | break; |
28fd6d7f RR |
1964 | case 'r': |
1965 | setup_rng(); | |
1966 | break; | |
8ca47e00 RR |
1967 | case 'i': |
1968 | initrd_name = optarg; | |
1969 | break; | |
1970 | default: | |
1971 | warnx("Unknown argument %s", argv[optind]); | |
1972 | usage(); | |
1973 | } | |
1974 | } | |
dde79789 RR |
1975 | /* After the other arguments we expect memory and kernel image name, |
1976 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1977 | if (optind + 2 > argc) |
1978 | usage(); | |
1979 | ||
3c6b5bfa RR |
1980 | verbose("Guest base is at %p\n", guest_base); |
1981 | ||
dde79789 | 1982 | /* We always have a console device */ |
17cbca2b | 1983 | setup_console(); |
8ca47e00 | 1984 | |
a161883a RR |
1985 | /* We can timeout waiting for Guest network transmit. */ |
1986 | setup_timeout(); | |
1987 | ||
8ca47e00 | 1988 | /* Now we load the kernel */ |
47436aa4 | 1989 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1990 | |
3c6b5bfa RR |
1991 | /* Boot information is stashed at physical address 0 */ |
1992 | boot = from_guest_phys(0); | |
1993 | ||
dde79789 | 1994 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1995 | if (initrd_name) { |
1996 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1997 | /* These are the location in the Linux boot header where the |
1998 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
1999 | boot->hdr.ramdisk_image = mem - initrd_size; |
2000 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 2001 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 2002 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
2003 | } |
2004 | ||
dde79789 RR |
2005 | /* The Linux boot header contains an "E820" memory map: ours is a |
2006 | * simple, single region. */ | |
43d33b21 RR |
2007 | boot->e820_entries = 1; |
2008 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 2009 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
2010 | * line after the boot header. */ |
2011 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
e1e72965 | 2012 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 2013 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 2014 | |
814a0e5c | 2015 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 2016 | boot->hdr.version = 0x207; |
814a0e5c RR |
2017 | |
2018 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 2019 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 2020 | |
43d33b21 RR |
2021 | /* Tell the entry path not to try to reload segment registers. */ |
2022 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 2023 | |
dde79789 RR |
2024 | /* We tell the kernel to initialize the Guest: this returns the open |
2025 | * /dev/lguest file descriptor. */ | |
56739c80 | 2026 | tell_kernel(start); |
dde79789 | 2027 | |
8c79873d RR |
2028 | /* We clone off a thread, which wakes the Launcher whenever one of the |
2029 | * input file descriptors needs attention. We call this the Waker, and | |
2030 | * we'll cover it in a moment. */ | |
56739c80 | 2031 | setup_waker(); |
8ca47e00 | 2032 | |
dde79789 | 2033 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 2034 | run_guest(); |
8ca47e00 | 2035 | } |
f56a384e RR |
2036 | /*:*/ |
2037 | ||
2038 | /*M:999 | |
2039 | * Mastery is done: you now know everything I do. | |
2040 | * | |
2041 | * But surely you have seen code, features and bugs in your wanderings which | |
2042 | * you now yearn to attack? That is the real game, and I look forward to you | |
2043 | * patching and forking lguest into the Your-Name-Here-visor. | |
2044 | * | |
2045 | * Farewell, and good coding! | |
2046 | * Rusty Russell. | |
2047 | */ |