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]; | |
82 | int lguest_fd; | |
83 | } waker_fds; | |
84 | ||
3c6b5bfa RR |
85 | /* The pointer to the start of guest memory. */ |
86 | static void *guest_base; | |
87 | /* The maximum guest physical address allowed, and maximum possible. */ | |
88 | static unsigned long guest_limit, guest_max; | |
a161883a RR |
89 | /* The pipe for signal hander to write to. */ |
90 | static int timeoutpipe[2]; | |
aa124984 | 91 | static unsigned int timeout_usec = 500; |
8ca47e00 | 92 | |
e3283fa0 GOC |
93 | /* a per-cpu variable indicating whose vcpu is currently running */ |
94 | static unsigned int __thread cpu_id; | |
95 | ||
dde79789 | 96 | /* This is our list of devices. */ |
8ca47e00 RR |
97 | struct device_list |
98 | { | |
dde79789 RR |
99 | /* Summary information about the devices in our list: ready to pass to |
100 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
101 | fd_set infds; |
102 | int max_infd; | |
103 | ||
17cbca2b RR |
104 | /* Counter to assign interrupt numbers. */ |
105 | unsigned int next_irq; | |
106 | ||
107 | /* Counter to print out convenient device numbers. */ | |
108 | unsigned int device_num; | |
109 | ||
dde79789 | 110 | /* The descriptor page for the devices. */ |
17cbca2b RR |
111 | u8 *descpage; |
112 | ||
dde79789 | 113 | /* A single linked list of devices. */ |
8ca47e00 | 114 | struct device *dev; |
a586d4f6 RR |
115 | /* And a pointer to the last device for easy append and also for |
116 | * configuration appending. */ | |
117 | struct device *lastdev; | |
8ca47e00 RR |
118 | }; |
119 | ||
17cbca2b RR |
120 | /* The list of Guest devices, based on command line arguments. */ |
121 | static struct device_list devices; | |
122 | ||
dde79789 | 123 | /* The device structure describes a single device. */ |
8ca47e00 RR |
124 | struct device |
125 | { | |
dde79789 | 126 | /* The linked-list pointer. */ |
8ca47e00 | 127 | struct device *next; |
17cbca2b | 128 | |
713b15b3 | 129 | /* The device's descriptor, as mapped into the Guest. */ |
8ca47e00 | 130 | struct lguest_device_desc *desc; |
17cbca2b | 131 | |
713b15b3 RR |
132 | /* We can't trust desc values once Guest has booted: we use these. */ |
133 | unsigned int feature_len; | |
134 | unsigned int num_vq; | |
135 | ||
17cbca2b RR |
136 | /* The name of this device, for --verbose. */ |
137 | const char *name; | |
8ca47e00 | 138 | |
dde79789 RR |
139 | /* If handle_input is set, it wants to be called when this file |
140 | * descriptor is ready. */ | |
8ca47e00 RR |
141 | int fd; |
142 | bool (*handle_input)(int fd, struct device *me); | |
143 | ||
17cbca2b RR |
144 | /* Any queues attached to this device */ |
145 | struct virtqueue *vq; | |
8ca47e00 | 146 | |
a007a751 RR |
147 | /* Handle status being finalized (ie. feature bits stable). */ |
148 | void (*ready)(struct device *me); | |
149 | ||
8ca47e00 RR |
150 | /* Device-specific data. */ |
151 | void *priv; | |
152 | }; | |
153 | ||
17cbca2b RR |
154 | /* The virtqueue structure describes a queue attached to a device. */ |
155 | struct virtqueue | |
156 | { | |
157 | struct virtqueue *next; | |
158 | ||
159 | /* Which device owns me. */ | |
160 | struct device *dev; | |
161 | ||
162 | /* The configuration for this queue. */ | |
163 | struct lguest_vqconfig config; | |
164 | ||
165 | /* The actual ring of buffers. */ | |
166 | struct vring vring; | |
167 | ||
168 | /* Last available index we saw. */ | |
169 | u16 last_avail_idx; | |
170 | ||
a161883a RR |
171 | /* The routine to call when the Guest pings us, or timeout. */ |
172 | void (*handle_output)(int fd, struct virtqueue *me, bool timeout); | |
20887611 RR |
173 | |
174 | /* Outstanding buffers */ | |
175 | unsigned int inflight; | |
a161883a RR |
176 | |
177 | /* Is this blocked awaiting a timer? */ | |
178 | bool blocked; | |
17cbca2b RR |
179 | }; |
180 | ||
ec04b13f BR |
181 | /* Remember the arguments to the program so we can "reboot" */ |
182 | static char **main_args; | |
183 | ||
17cbca2b RR |
184 | /* Since guest is UP and we don't run at the same time, we don't need barriers. |
185 | * But I include them in the code in case others copy it. */ | |
186 | #define wmb() | |
187 | ||
188 | /* Convert an iovec element to the given type. | |
189 | * | |
190 | * This is a fairly ugly trick: we need to know the size of the type and | |
191 | * alignment requirement to check the pointer is kosher. It's also nice to | |
192 | * have the name of the type in case we report failure. | |
193 | * | |
194 | * Typing those three things all the time is cumbersome and error prone, so we | |
195 | * have a macro which sets them all up and passes to the real function. */ | |
196 | #define convert(iov, type) \ | |
197 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
198 | ||
199 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
200 | const char *name) | |
201 | { | |
202 | if (iov->iov_len != size) | |
203 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
204 | if ((unsigned long)iov->iov_base % align != 0) | |
205 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
206 | return iov->iov_base; | |
207 | } | |
208 | ||
b5111790 RR |
209 | /* Wrapper for the last available index. Makes it easier to change. */ |
210 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
211 | ||
17cbca2b RR |
212 | /* The virtio configuration space is defined to be little-endian. x86 is |
213 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
214 | #define cpu_to_le16(v16) (v16) | |
215 | #define cpu_to_le32(v32) (v32) | |
216 | #define cpu_to_le64(v64) (v64) | |
217 | #define le16_to_cpu(v16) (v16) | |
218 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 219 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 220 | |
28fd6d7f RR |
221 | /* Is this iovec empty? */ |
222 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
223 | { | |
224 | unsigned int i; | |
225 | ||
226 | for (i = 0; i < num_iov; i++) | |
227 | if (iov[i].iov_len) | |
228 | return false; | |
229 | return true; | |
230 | } | |
231 | ||
232 | /* Take len bytes from the front of this iovec. */ | |
233 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) | |
234 | { | |
235 | unsigned int i; | |
236 | ||
237 | for (i = 0; i < num_iov; i++) { | |
238 | unsigned int used; | |
239 | ||
240 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
241 | iov[i].iov_base += used; | |
242 | iov[i].iov_len -= used; | |
243 | len -= used; | |
244 | } | |
245 | assert(len == 0); | |
246 | } | |
247 | ||
6e5aa7ef RR |
248 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
249 | static u8 *get_feature_bits(struct device *dev) | |
250 | { | |
251 | return (u8 *)(dev->desc + 1) | |
713b15b3 | 252 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
6e5aa7ef RR |
253 | } |
254 | ||
3c6b5bfa RR |
255 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
256 | * where pointers run wild and free! Unfortunately, like most userspace | |
257 | * programs, it's quite boring (which is why everyone likes to hack on the | |
258 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
259 | * will get you through this section. Or, maybe not. | |
260 | * | |
261 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
262 | * memory and stores it in "guest_base". In other words, Guest physical == | |
263 | * Launcher virtual with an offset. | |
264 | * | |
265 | * This can be tough to get your head around, but usually it just means that we | |
266 | * use these trivial conversion functions when the Guest gives us it's | |
267 | * "physical" addresses: */ | |
268 | static void *from_guest_phys(unsigned long addr) | |
269 | { | |
270 | return guest_base + addr; | |
271 | } | |
272 | ||
273 | static unsigned long to_guest_phys(const void *addr) | |
274 | { | |
275 | return (addr - guest_base); | |
276 | } | |
277 | ||
dde79789 RR |
278 | /*L:130 |
279 | * Loading the Kernel. | |
280 | * | |
281 | * We start with couple of simple helper routines. open_or_die() avoids | |
282 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
283 | static int open_or_die(const char *name, int flags) |
284 | { | |
285 | int fd = open(name, flags); | |
286 | if (fd < 0) | |
287 | err(1, "Failed to open %s", name); | |
288 | return fd; | |
289 | } | |
290 | ||
3c6b5bfa RR |
291 | /* map_zeroed_pages() takes a number of pages. */ |
292 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 293 | { |
3c6b5bfa RR |
294 | int fd = open_or_die("/dev/zero", O_RDONLY); |
295 | void *addr; | |
8ca47e00 | 296 | |
dde79789 | 297 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
298 | * copied). */ |
299 | addr = mmap(NULL, getpagesize() * num, | |
300 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
301 | if (addr == MAP_FAILED) | |
302 | err(1, "Mmaping %u pages of /dev/zero", num); | |
34bdaab4 | 303 | close(fd); |
3c6b5bfa RR |
304 | |
305 | return addr; | |
306 | } | |
307 | ||
308 | /* Get some more pages for a device. */ | |
309 | static void *get_pages(unsigned int num) | |
310 | { | |
311 | void *addr = from_guest_phys(guest_limit); | |
312 | ||
313 | guest_limit += num * getpagesize(); | |
314 | if (guest_limit > guest_max) | |
315 | errx(1, "Not enough memory for devices"); | |
316 | return addr; | |
8ca47e00 RR |
317 | } |
318 | ||
6649bb7a RM |
319 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
320 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
321 | * it falls back to reading the memory in. */ | |
322 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
323 | { | |
324 | ssize_t r; | |
325 | ||
326 | /* We map writable even though for some segments are marked read-only. | |
327 | * The kernel really wants to be writable: it patches its own | |
328 | * instructions. | |
329 | * | |
330 | * MAP_PRIVATE means that the page won't be copied until a write is | |
331 | * done to it. This allows us to share untouched memory between | |
332 | * Guests. */ | |
333 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
334 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
335 | return; | |
336 | ||
337 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
338 | r = pread(fd, addr, len, offset); | |
339 | if (r != len) | |
340 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
341 | } | |
342 | ||
dde79789 RR |
343 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
344 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
345 | * by all modern binaries on Linux including the kernel. | |
346 | * | |
347 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
348 | * address. We use the physical address; the Guest will map itself to the |
349 | * virtual address. | |
dde79789 RR |
350 | * |
351 | * We return the starting address. */ | |
47436aa4 | 352 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 353 | { |
8ca47e00 RR |
354 | Elf32_Phdr phdr[ehdr->e_phnum]; |
355 | unsigned int i; | |
8ca47e00 | 356 | |
dde79789 RR |
357 | /* Sanity checks on the main ELF header: an x86 executable with a |
358 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
359 | if (ehdr->e_type != ET_EXEC |
360 | || ehdr->e_machine != EM_386 | |
361 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
362 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
363 | errx(1, "Malformed elf header"); | |
364 | ||
dde79789 RR |
365 | /* An ELF executable contains an ELF header and a number of "program" |
366 | * headers which indicate which parts ("segments") of the program to | |
367 | * load where. */ | |
368 | ||
369 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
370 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
371 | err(1, "Seeking to program headers"); | |
372 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
373 | err(1, "Reading program headers"); | |
374 | ||
dde79789 | 375 | /* Try all the headers: there are usually only three. A read-only one, |
a6bd8e13 | 376 | * a read-write one, and a "note" section which we don't load. */ |
8ca47e00 | 377 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 378 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
379 | if (phdr[i].p_type != PT_LOAD) |
380 | continue; | |
381 | ||
382 | verbose("Section %i: size %i addr %p\n", | |
383 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
384 | ||
6649bb7a | 385 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 386 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 387 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
388 | } |
389 | ||
814a0e5c RR |
390 | /* The entry point is given in the ELF header. */ |
391 | return ehdr->e_entry; | |
8ca47e00 RR |
392 | } |
393 | ||
dde79789 | 394 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
395 | * supposed to jump into it and it will unpack itself. We used to have to |
396 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 397 | * |
5bbf89fc RR |
398 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
399 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
400 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 401 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 402 | { |
43d33b21 | 403 | struct boot_params boot; |
5bbf89fc RR |
404 | int r; |
405 | /* Modern bzImages get loaded at 1M. */ | |
406 | void *p = from_guest_phys(0x100000); | |
407 | ||
408 | /* Go back to the start of the file and read the header. It should be | |
71cced6e | 409 | * a Linux boot header (see Documentation/x86/i386/boot.txt) */ |
5bbf89fc | 410 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 411 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 412 | |
43d33b21 RR |
413 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
414 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
415 | errx(1, "This doesn't look like a bzImage to me"); |
416 | ||
43d33b21 RR |
417 | /* Skip over the extra sectors of the header. */ |
418 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
419 | |
420 | /* Now read everything into memory. in nice big chunks. */ | |
421 | while ((r = read(fd, p, 65536)) > 0) | |
422 | p += r; | |
423 | ||
43d33b21 RR |
424 | /* Finally, code32_start tells us where to enter the kernel. */ |
425 | return boot.hdr.code32_start; | |
8ca47e00 RR |
426 | } |
427 | ||
dde79789 | 428 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
e1e72965 RR |
429 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
430 | * work, we can load those, too. */ | |
47436aa4 | 431 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
432 | { |
433 | Elf32_Ehdr hdr; | |
434 | ||
dde79789 | 435 | /* Read in the first few bytes. */ |
8ca47e00 RR |
436 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
437 | err(1, "Reading kernel"); | |
438 | ||
dde79789 | 439 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 440 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 441 | return map_elf(fd, &hdr); |
8ca47e00 | 442 | |
a6bd8e13 | 443 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 444 | return load_bzimage(fd); |
8ca47e00 RR |
445 | } |
446 | ||
dde79789 RR |
447 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
448 | * it calls getpagesize() twice: "it's dumb code." | |
449 | * | |
450 | * Kernel guys get really het up about optimization, even when it's not | |
451 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
452 | static inline unsigned long page_align(unsigned long addr) |
453 | { | |
dde79789 | 454 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
455 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
456 | } | |
457 | ||
dde79789 RR |
458 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
459 | * the kernel which the kernel can use to boot from without needing any | |
460 | * drivers. Most distributions now use this as standard: the initrd contains | |
461 | * the code to load the appropriate driver modules for the current machine. | |
462 | * | |
463 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
464 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
465 | static unsigned long load_initrd(const char *name, unsigned long mem) |
466 | { | |
467 | int ifd; | |
468 | struct stat st; | |
469 | unsigned long len; | |
8ca47e00 RR |
470 | |
471 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 472 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
473 | if (fstat(ifd, &st) < 0) |
474 | err(1, "fstat() on initrd '%s'", name); | |
475 | ||
6649bb7a RM |
476 | /* We map the initrd at the top of memory, but mmap wants it to be |
477 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 478 | len = page_align(st.st_size); |
3c6b5bfa | 479 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
480 | /* Once a file is mapped, you can close the file descriptor. It's a |
481 | * little odd, but quite useful. */ | |
8ca47e00 | 482 | close(ifd); |
6649bb7a | 483 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
484 | |
485 | /* We return the initrd size. */ | |
8ca47e00 RR |
486 | return len; |
487 | } | |
e1e72965 | 488 | /*:*/ |
8ca47e00 | 489 | |
dde79789 RR |
490 | /* Simple routine to roll all the commandline arguments together with spaces |
491 | * between them. */ | |
8ca47e00 RR |
492 | static void concat(char *dst, char *args[]) |
493 | { | |
494 | unsigned int i, len = 0; | |
495 | ||
496 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
497 | if (i) { |
498 | strcat(dst+len, " "); | |
499 | len++; | |
500 | } | |
8ca47e00 | 501 | strcpy(dst+len, args[i]); |
1ef36fa6 | 502 | len += strlen(args[i]); |
8ca47e00 RR |
503 | } |
504 | /* In case it's empty. */ | |
505 | dst[len] = '\0'; | |
506 | } | |
507 | ||
e1e72965 RR |
508 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
509 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: | |
58a24566 MZ |
510 | * the base of Guest "physical" memory, the top physical page to allow and the |
511 | * entry point for the Guest. */ | |
512 | static int tell_kernel(unsigned long start) | |
8ca47e00 | 513 | { |
511801dc JS |
514 | unsigned long args[] = { LHREQ_INITIALIZE, |
515 | (unsigned long)guest_base, | |
58a24566 | 516 | guest_limit / getpagesize(), start }; |
8ca47e00 RR |
517 | int fd; |
518 | ||
3c6b5bfa RR |
519 | verbose("Guest: %p - %p (%#lx)\n", |
520 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
521 | fd = open_or_die("/dev/lguest", O_RDWR); |
522 | if (write(fd, args, sizeof(args)) < 0) | |
523 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
524 | |
525 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
526 | return fd; |
527 | } | |
dde79789 | 528 | /*:*/ |
8ca47e00 | 529 | |
17cbca2b | 530 | static void add_device_fd(int fd) |
8ca47e00 | 531 | { |
17cbca2b RR |
532 | FD_SET(fd, &devices.infds); |
533 | if (fd > devices.max_infd) | |
534 | devices.max_infd = fd; | |
8ca47e00 RR |
535 | } |
536 | ||
dde79789 RR |
537 | /*L:200 |
538 | * The Waker. | |
539 | * | |
e1e72965 RR |
540 | * With console, block and network devices, we can have lots of input which we |
541 | * need to process. We could try to tell the kernel what file descriptors to | |
542 | * watch, but handing a file descriptor mask through to the kernel is fairly | |
543 | * icky. | |
dde79789 | 544 | * |
8c79873d | 545 | * Instead, we clone off a thread which watches the file descriptors and writes |
e1e72965 RR |
546 | * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host |
547 | * stop running the Guest. This causes the Launcher to return from the | |
dde79789 RR |
548 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset |
549 | * the LHREQ_BREAK and wake us up again. | |
550 | * | |
551 | * This, of course, is merely a different *kind* of icky. | |
8c79873d RR |
552 | * |
553 | * Given my well-known antipathy to threads, I'd prefer to use processes. But | |
554 | * it's easier to share Guest memory with threads, and trivial to share the | |
555 | * devices.infds as the Launcher changes it. | |
dde79789 | 556 | */ |
8c79873d | 557 | static int waker(void *unused) |
8ca47e00 | 558 | { |
8c79873d RR |
559 | /* Close the write end of the pipe: only the Launcher has it open. */ |
560 | close(waker_fds.pipe[1]); | |
8ca47e00 RR |
561 | |
562 | for (;;) { | |
17cbca2b | 563 | fd_set rfds = devices.infds; |
511801dc | 564 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8c79873d RR |
565 | unsigned int maxfd = devices.max_infd; |
566 | ||
567 | /* We also listen to the pipe from the Launcher. */ | |
568 | FD_SET(waker_fds.pipe[0], &rfds); | |
569 | if (waker_fds.pipe[0] > maxfd) | |
570 | maxfd = waker_fds.pipe[0]; | |
8ca47e00 | 571 | |
dde79789 | 572 | /* Wait until input is ready from one of the devices. */ |
8c79873d RR |
573 | select(maxfd+1, &rfds, NULL, NULL, NULL); |
574 | ||
575 | /* Message from Launcher? */ | |
576 | if (FD_ISSET(waker_fds.pipe[0], &rfds)) { | |
577 | char c; | |
578 | /* If this fails, then assume Launcher has exited. | |
579 | * Don't do anything on exit: we're just a thread! */ | |
580 | if (read(waker_fds.pipe[0], &c, 1) != 1) | |
581 | _exit(0); | |
582 | continue; | |
583 | } | |
584 | ||
585 | /* Send LHREQ_BREAK command to snap the Launcher out of it. */ | |
586 | pwrite(waker_fds.lguest_fd, args, sizeof(args), cpu_id); | |
8ca47e00 | 587 | } |
8c79873d | 588 | return 0; |
8ca47e00 RR |
589 | } |
590 | ||
dde79789 | 591 | /* This routine just sets up a pipe to the Waker process. */ |
8c79873d RR |
592 | static void setup_waker(int lguest_fd) |
593 | { | |
594 | /* This pipe is closed when Launcher dies, telling Waker. */ | |
595 | if (pipe(waker_fds.pipe) != 0) | |
596 | err(1, "Creating pipe for Waker"); | |
8ca47e00 | 597 | |
8c79873d RR |
598 | /* Waker also needs to know the lguest fd */ |
599 | waker_fds.lguest_fd = lguest_fd; | |
600 | ||
601 | if (clone(waker, malloc(4096) + 4096, CLONE_VM | SIGCHLD, NULL) == -1) | |
602 | err(1, "Creating Waker"); | |
8ca47e00 RR |
603 | } |
604 | ||
e1e72965 | 605 | /* |
dde79789 RR |
606 | * Device Handling. |
607 | * | |
e1e72965 | 608 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 609 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 610 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
611 | * if something funny is going on: |
612 | */ | |
8ca47e00 RR |
613 | static void *_check_pointer(unsigned long addr, unsigned int size, |
614 | unsigned int line) | |
615 | { | |
dde79789 RR |
616 | /* We have to separately check addr and addr+size, because size could |
617 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 618 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 619 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
620 | /* We return a pointer for the caller's convenience, now we know it's |
621 | * safe to use. */ | |
3c6b5bfa | 622 | return from_guest_phys(addr); |
8ca47e00 | 623 | } |
dde79789 | 624 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
625 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
626 | ||
e1e72965 RR |
627 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
628 | * function returns the next descriptor in the chain, or vq->vring.num if we're | |
629 | * at the end. */ | |
17cbca2b RR |
630 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) |
631 | { | |
632 | unsigned int next; | |
633 | ||
634 | /* If this descriptor says it doesn't chain, we're done. */ | |
635 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
636 | return vq->vring.num; | |
637 | ||
638 | /* Check they're not leading us off end of descriptors. */ | |
639 | next = vq->vring.desc[i].next; | |
640 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
641 | wmb(); | |
642 | ||
643 | if (next >= vq->vring.num) | |
644 | errx(1, "Desc next is %u", next); | |
645 | ||
646 | return next; | |
647 | } | |
648 | ||
649 | /* This looks in the virtqueue and for the first available buffer, and converts | |
650 | * it to an iovec for convenient access. Since descriptors consist of some | |
651 | * number of output then some number of input descriptors, it's actually two | |
652 | * iovecs, but we pack them into one and note how many of each there were. | |
653 | * | |
654 | * This function returns the descriptor number found, or vq->vring.num (which | |
655 | * is never a valid descriptor number) if none was found. */ | |
656 | static unsigned get_vq_desc(struct virtqueue *vq, | |
657 | struct iovec iov[], | |
658 | unsigned int *out_num, unsigned int *in_num) | |
659 | { | |
660 | unsigned int i, head; | |
b5111790 | 661 | u16 last_avail; |
17cbca2b RR |
662 | |
663 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 RR |
664 | last_avail = lg_last_avail(vq); |
665 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) | |
17cbca2b | 666 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 667 | last_avail, vq->vring.avail->idx); |
17cbca2b RR |
668 | |
669 | /* If there's nothing new since last we looked, return invalid. */ | |
b5111790 | 670 | if (vq->vring.avail->idx == last_avail) |
17cbca2b RR |
671 | return vq->vring.num; |
672 | ||
673 | /* Grab the next descriptor number they're advertising, and increment | |
674 | * the index we've seen. */ | |
b5111790 RR |
675 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
676 | lg_last_avail(vq)++; | |
17cbca2b RR |
677 | |
678 | /* If their number is silly, that's a fatal mistake. */ | |
679 | if (head >= vq->vring.num) | |
680 | errx(1, "Guest says index %u is available", head); | |
681 | ||
682 | /* When we start there are none of either input nor output. */ | |
683 | *out_num = *in_num = 0; | |
684 | ||
685 | i = head; | |
686 | do { | |
687 | /* Grab the first descriptor, and check it's OK. */ | |
688 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
689 | iov[*out_num + *in_num].iov_base | |
690 | = check_pointer(vq->vring.desc[i].addr, | |
691 | vq->vring.desc[i].len); | |
692 | /* If this is an input descriptor, increment that count. */ | |
693 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
694 | (*in_num)++; | |
695 | else { | |
696 | /* If it's an output descriptor, they're all supposed | |
697 | * to come before any input descriptors. */ | |
698 | if (*in_num) | |
699 | errx(1, "Descriptor has out after in"); | |
700 | (*out_num)++; | |
701 | } | |
702 | ||
703 | /* If we've got too many, that implies a descriptor loop. */ | |
704 | if (*out_num + *in_num > vq->vring.num) | |
705 | errx(1, "Looped descriptor"); | |
706 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 707 | |
20887611 | 708 | vq->inflight++; |
17cbca2b | 709 | return head; |
8ca47e00 RR |
710 | } |
711 | ||
e1e72965 | 712 | /* After we've used one of their buffers, we tell them about it. We'll then |
17cbca2b RR |
713 | * want to send them an interrupt, using trigger_irq(). */ |
714 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 715 | { |
17cbca2b RR |
716 | struct vring_used_elem *used; |
717 | ||
e1e72965 RR |
718 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
719 | * next entry in that used ring. */ | |
17cbca2b RR |
720 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
721 | used->id = head; | |
722 | used->len = len; | |
723 | /* Make sure buffer is written before we update index. */ | |
724 | wmb(); | |
725 | vq->vring.used->idx++; | |
20887611 | 726 | vq->inflight--; |
8ca47e00 RR |
727 | } |
728 | ||
17cbca2b RR |
729 | /* This actually sends the interrupt for this virtqueue */ |
730 | static void trigger_irq(int fd, struct virtqueue *vq) | |
8ca47e00 | 731 | { |
17cbca2b RR |
732 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
733 | ||
20887611 RR |
734 | /* If they don't want an interrupt, don't send one, unless empty. */ |
735 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
736 | && vq->inflight) | |
17cbca2b RR |
737 | return; |
738 | ||
739 | /* Send the Guest an interrupt tell them we used something up. */ | |
8ca47e00 | 740 | if (write(fd, buf, sizeof(buf)) != 0) |
17cbca2b | 741 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
742 | } |
743 | ||
17cbca2b RR |
744 | /* And here's the combo meal deal. Supersize me! */ |
745 | static void add_used_and_trigger(int fd, struct virtqueue *vq, | |
746 | unsigned int head, int len) | |
8ca47e00 | 747 | { |
17cbca2b RR |
748 | add_used(vq, head, len); |
749 | trigger_irq(fd, vq); | |
8ca47e00 RR |
750 | } |
751 | ||
e1e72965 RR |
752 | /* |
753 | * The Console | |
754 | * | |
755 | * Here is the input terminal setting we save, and the routine to restore them | |
756 | * on exit so the user gets their terminal back. */ | |
8ca47e00 RR |
757 | static struct termios orig_term; |
758 | static void restore_term(void) | |
759 | { | |
760 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
761 | } | |
762 | ||
dde79789 | 763 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
764 | struct console_abort |
765 | { | |
dde79789 | 766 | /* How many times have they hit ^C? */ |
8ca47e00 | 767 | int count; |
dde79789 | 768 | /* When did they start? */ |
8ca47e00 RR |
769 | struct timeval start; |
770 | }; | |
771 | ||
dde79789 | 772 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
773 | static bool handle_console_input(int fd, struct device *dev) |
774 | { | |
8ca47e00 | 775 | int len; |
17cbca2b RR |
776 | unsigned int head, in_num, out_num; |
777 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
778 | struct console_abort *abort = dev->priv; |
779 | ||
17cbca2b RR |
780 | /* First we need a console buffer from the Guests's input virtqueue. */ |
781 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
782 | |
783 | /* If they're not ready for input, stop listening to this file | |
784 | * descriptor. We'll start again once they add an input buffer. */ | |
785 | if (head == dev->vq->vring.num) | |
786 | return false; | |
787 | ||
788 | if (out_num) | |
17cbca2b | 789 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 790 | |
dde79789 RR |
791 | /* This is why we convert to iovecs: the readv() call uses them, and so |
792 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 793 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 794 | if (len <= 0) { |
dde79789 | 795 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 796 | * something went terribly wrong. */ |
8ca47e00 | 797 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 798 | /* Put the input terminal back. */ |
17cbca2b | 799 | restore_term(); |
56ae43df RR |
800 | /* Remove callback from input vq, so it doesn't restart us. */ |
801 | dev->vq->handle_output = NULL; | |
802 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 803 | return false; |
8ca47e00 RR |
804 | } |
805 | ||
56ae43df RR |
806 | /* Tell the Guest about the new input. */ |
807 | add_used_and_trigger(fd, dev->vq, head, len); | |
8ca47e00 | 808 | |
dde79789 RR |
809 | /* Three ^C within one second? Exit. |
810 | * | |
811 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
812 | * in a buffer by itself, so they can't be too fast. But we check that | |
813 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
814 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
815 | if (!abort->count++) | |
816 | gettimeofday(&abort->start, NULL); | |
817 | else if (abort->count == 3) { | |
818 | struct timeval now; | |
819 | gettimeofday(&now, NULL); | |
820 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 821 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
822 | /* Close the fd so Waker will know it has to |
823 | * exit. */ | |
8c79873d RR |
824 | close(waker_fds.pipe[1]); |
825 | /* Just in case Waker is blocked in BREAK, send | |
dde79789 | 826 | * unbreak now. */ |
8ca47e00 RR |
827 | write(fd, args, sizeof(args)); |
828 | exit(2); | |
829 | } | |
830 | abort->count = 0; | |
831 | } | |
832 | } else | |
dde79789 | 833 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
834 | abort->count = 0; |
835 | ||
dde79789 | 836 | /* Everything went OK! */ |
8ca47e00 RR |
837 | return true; |
838 | } | |
839 | ||
17cbca2b RR |
840 | /* Handling output for console is simple: we just get all the output buffers |
841 | * and write them to stdout. */ | |
a161883a | 842 | static void handle_console_output(int fd, struct virtqueue *vq, bool timeout) |
8ca47e00 | 843 | { |
17cbca2b RR |
844 | unsigned int head, out, in; |
845 | int len; | |
846 | struct iovec iov[vq->vring.num]; | |
847 | ||
848 | /* Keep getting output buffers from the Guest until we run out. */ | |
849 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
850 | if (in) | |
851 | errx(1, "Input buffers in output queue?"); | |
852 | len = writev(STDOUT_FILENO, iov, out); | |
853 | add_used_and_trigger(fd, vq, head, len); | |
854 | } | |
8ca47e00 RR |
855 | } |
856 | ||
1dc3e3bc RR |
857 | /* This is called when we no longer want to hear about Guest changes to a |
858 | * virtqueue. This is more efficient in high-traffic cases, but it means we | |
859 | * have to set a timer to check if any more changes have occurred. */ | |
a161883a RR |
860 | static void block_vq(struct virtqueue *vq) |
861 | { | |
862 | struct itimerval itm; | |
863 | ||
864 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
865 | vq->blocked = true; | |
866 | ||
867 | itm.it_interval.tv_sec = 0; | |
868 | itm.it_interval.tv_usec = 0; | |
869 | itm.it_value.tv_sec = 0; | |
aa124984 | 870 | itm.it_value.tv_usec = timeout_usec; |
a161883a RR |
871 | |
872 | setitimer(ITIMER_REAL, &itm, NULL); | |
873 | } | |
874 | ||
e1e72965 RR |
875 | /* |
876 | * The Network | |
877 | * | |
878 | * Handling output for network is also simple: we get all the output buffers | |
17cbca2b | 879 | * and write them (ignoring the first element) to this device's file descriptor |
a6bd8e13 RR |
880 | * (/dev/net/tun). |
881 | */ | |
a161883a | 882 | static void handle_net_output(int fd, struct virtqueue *vq, bool timeout) |
8ca47e00 | 883 | { |
a161883a | 884 | unsigned int head, out, in, num = 0; |
17cbca2b RR |
885 | int len; |
886 | struct iovec iov[vq->vring.num]; | |
aa124984 | 887 | static int last_timeout_num; |
17cbca2b RR |
888 | |
889 | /* Keep getting output buffers from the Guest until we run out. */ | |
890 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
891 | if (in) | |
892 | errx(1, "Input buffers in output queue?"); | |
398f187d RR |
893 | len = writev(vq->dev->fd, iov, out); |
894 | if (len < 0) | |
895 | err(1, "Writing network packet to tun"); | |
17cbca2b | 896 | add_used_and_trigger(fd, vq, head, len); |
a161883a | 897 | num++; |
17cbca2b | 898 | } |
a161883a RR |
899 | |
900 | /* Block further kicks and set up a timer if we saw anything. */ | |
901 | if (!timeout && num) | |
902 | block_vq(vq); | |
aa124984 | 903 | |
1dc3e3bc RR |
904 | /* We never quite know how long should we wait before we check the |
905 | * queue again for more packets. We start at 500 microseconds, and if | |
906 | * we get fewer packets than last time, we assume we made the timeout | |
907 | * too small and increase it by 10 microseconds. Otherwise, we drop it | |
908 | * by one microsecond every time. It seems to work well enough. */ | |
aa124984 RR |
909 | if (timeout) { |
910 | if (num < last_timeout_num) | |
911 | timeout_usec += 10; | |
912 | else if (timeout_usec > 1) | |
913 | timeout_usec--; | |
914 | last_timeout_num = num; | |
915 | } | |
8ca47e00 RR |
916 | } |
917 | ||
17cbca2b RR |
918 | /* This is where we handle a packet coming in from the tun device to our |
919 | * Guest. */ | |
8ca47e00 RR |
920 | static bool handle_tun_input(int fd, struct device *dev) |
921 | { | |
17cbca2b | 922 | unsigned int head, in_num, out_num; |
8ca47e00 | 923 | int len; |
17cbca2b | 924 | struct iovec iov[dev->vq->vring.num]; |
8ca47e00 | 925 | |
17cbca2b RR |
926 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
927 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
928 | if (head == dev->vq->vring.num) { | |
dde79789 | 929 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
930 | * early, the Guest won't be ready yet. Wait until the device |
931 | * status says it's ready. */ | |
932 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
5dae785a RR |
933 | |
934 | /* Now tell it we want to know if new things appear. */ | |
935 | dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
936 | wmb(); | |
937 | ||
56ae43df RR |
938 | /* We'll turn this back on if input buffers are registered. */ |
939 | return false; | |
17cbca2b RR |
940 | } else if (out_num) |
941 | errx(1, "Output buffers in network recv queue?"); | |
942 | ||
dde79789 | 943 | /* Read the packet from the device directly into the Guest's buffer. */ |
398f187d | 944 | len = readv(dev->fd, iov, in_num); |
8ca47e00 RR |
945 | if (len <= 0) |
946 | err(1, "reading network"); | |
dde79789 | 947 | |
56ae43df | 948 | /* Tell the Guest about the new packet. */ |
398f187d | 949 | add_used_and_trigger(fd, dev->vq, head, len); |
17cbca2b | 950 | |
8ca47e00 | 951 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
952 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
953 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
954 | ||
dde79789 | 955 | /* All good. */ |
8ca47e00 RR |
956 | return true; |
957 | } | |
958 | ||
e1e72965 RR |
959 | /*L:215 This is the callback attached to the network and console input |
960 | * virtqueues: it ensures we try again, in case we stopped console or net | |
56ae43df | 961 | * delivery because Guest didn't have any buffers. */ |
a161883a | 962 | static void enable_fd(int fd, struct virtqueue *vq, bool timeout) |
56ae43df RR |
963 | { |
964 | add_device_fd(vq->dev->fd); | |
8c79873d RR |
965 | /* Snap the Waker out of its select loop. */ |
966 | write(waker_fds.pipe[1], "", 1); | |
56ae43df RR |
967 | } |
968 | ||
a161883a | 969 | static void net_enable_fd(int fd, struct virtqueue *vq, bool timeout) |
5dae785a RR |
970 | { |
971 | /* We don't need to know again when Guest refills receive buffer. */ | |
972 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
a161883a | 973 | enable_fd(fd, vq, timeout); |
5dae785a RR |
974 | } |
975 | ||
a007a751 RR |
976 | /* When the Guest tells us they updated the status field, we handle it. */ |
977 | static void update_device_status(struct device *dev) | |
6e5aa7ef RR |
978 | { |
979 | struct virtqueue *vq; | |
980 | ||
a007a751 RR |
981 | /* This is a reset. */ |
982 | if (dev->desc->status == 0) { | |
983 | verbose("Resetting device %s\n", dev->name); | |
6e5aa7ef | 984 | |
a007a751 | 985 | /* Clear any features they've acked. */ |
713b15b3 RR |
986 | memset(get_feature_bits(dev) + dev->feature_len, 0, |
987 | dev->feature_len); | |
6e5aa7ef | 988 | |
a007a751 RR |
989 | /* Zero out the virtqueues. */ |
990 | for (vq = dev->vq; vq; vq = vq->next) { | |
991 | memset(vq->vring.desc, 0, | |
2966af73 | 992 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); |
b5111790 | 993 | lg_last_avail(vq) = 0; |
a007a751 RR |
994 | } |
995 | } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
996 | warnx("Device %s configuration FAILED", dev->name); | |
997 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { | |
998 | unsigned int i; | |
999 | ||
1000 | verbose("Device %s OK: offered", dev->name); | |
713b15b3 | 1001 | for (i = 0; i < dev->feature_len; i++) |
32c68e5c | 1002 | verbose(" %02x", get_feature_bits(dev)[i]); |
a007a751 | 1003 | verbose(", accepted"); |
713b15b3 | 1004 | for (i = 0; i < dev->feature_len; i++) |
32c68e5c | 1005 | verbose(" %02x", get_feature_bits(dev) |
713b15b3 | 1006 | [dev->feature_len+i]); |
a007a751 RR |
1007 | |
1008 | if (dev->ready) | |
1009 | dev->ready(dev); | |
6e5aa7ef RR |
1010 | } |
1011 | } | |
1012 | ||
17cbca2b RR |
1013 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
1014 | static void handle_output(int fd, unsigned long addr) | |
8ca47e00 RR |
1015 | { |
1016 | struct device *i; | |
17cbca2b RR |
1017 | struct virtqueue *vq; |
1018 | ||
6e5aa7ef | 1019 | /* Check each device and virtqueue. */ |
17cbca2b | 1020 | for (i = devices.dev; i; i = i->next) { |
a007a751 | 1021 | /* Notifications to device descriptors update device status. */ |
6e5aa7ef | 1022 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1023 | update_device_status(i); |
6e5aa7ef RR |
1024 | return; |
1025 | } | |
1026 | ||
1027 | /* Notifications to virtqueues mean output has occurred. */ | |
17cbca2b | 1028 | for (vq = i->vq; vq; vq = vq->next) { |
6e5aa7ef RR |
1029 | if (vq->config.pfn != addr/getpagesize()) |
1030 | continue; | |
1031 | ||
1032 | /* Guest should acknowledge (and set features!) before | |
1033 | * using the device. */ | |
1034 | if (i->desc->status == 0) { | |
1035 | warnx("%s gave early output", i->name); | |
17cbca2b RR |
1036 | return; |
1037 | } | |
6e5aa7ef RR |
1038 | |
1039 | if (strcmp(vq->dev->name, "console") != 0) | |
1040 | verbose("Output to %s\n", vq->dev->name); | |
1041 | if (vq->handle_output) | |
a161883a | 1042 | vq->handle_output(fd, vq, false); |
6e5aa7ef | 1043 | return; |
8ca47e00 RR |
1044 | } |
1045 | } | |
dde79789 | 1046 | |
17cbca2b RR |
1047 | /* Early console write is done using notify on a nul-terminated string |
1048 | * in Guest memory. */ | |
1049 | if (addr >= guest_limit) | |
1050 | errx(1, "Bad NOTIFY %#lx", addr); | |
1051 | ||
1052 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1053 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1054 | } |
1055 | ||
a161883a RR |
1056 | static void handle_timeout(int fd) |
1057 | { | |
1058 | char buf[32]; | |
1059 | struct device *i; | |
1060 | struct virtqueue *vq; | |
1061 | ||
1062 | /* Clear the pipe */ | |
1063 | read(timeoutpipe[0], buf, sizeof(buf)); | |
1064 | ||
1065 | /* Check each device and virtqueue: flush blocked ones. */ | |
1066 | for (i = devices.dev; i; i = i->next) { | |
1067 | for (vq = i->vq; vq; vq = vq->next) { | |
1068 | if (!vq->blocked) | |
1069 | continue; | |
1070 | ||
1071 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
1072 | vq->blocked = false; | |
1073 | if (vq->handle_output) | |
1074 | vq->handle_output(fd, vq, true); | |
1075 | } | |
1076 | } | |
1077 | } | |
1078 | ||
e1e72965 | 1079 | /* This is called when the Waker wakes us up: check for incoming file |
dde79789 | 1080 | * descriptors. */ |
17cbca2b | 1081 | static void handle_input(int fd) |
8ca47e00 | 1082 | { |
dde79789 | 1083 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
1084 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
1085 | ||
1086 | for (;;) { | |
1087 | struct device *i; | |
17cbca2b | 1088 | fd_set fds = devices.infds; |
a161883a | 1089 | int num; |
8ca47e00 | 1090 | |
a161883a RR |
1091 | num = select(devices.max_infd+1, &fds, NULL, NULL, &poll); |
1092 | /* Could get interrupted */ | |
1093 | if (num < 0) | |
1094 | continue; | |
dde79789 | 1095 | /* If nothing is ready, we're done. */ |
a161883a | 1096 | if (num == 0) |
8ca47e00 RR |
1097 | break; |
1098 | ||
a6bd8e13 RR |
1099 | /* Otherwise, call the device(s) which have readable file |
1100 | * descriptors and a method of handling them. */ | |
17cbca2b | 1101 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 1102 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56ae43df RR |
1103 | if (i->handle_input(fd, i)) |
1104 | continue; | |
1105 | ||
dde79789 | 1106 | /* If handle_input() returns false, it means we |
56ae43df RR |
1107 | * should no longer service it. Networking and |
1108 | * console do this when there's no input | |
1109 | * buffers to deliver into. Console also uses | |
a6bd8e13 | 1110 | * it when it discovers that stdin is closed. */ |
56ae43df | 1111 | FD_CLR(i->fd, &devices.infds); |
8ca47e00 RR |
1112 | } |
1113 | } | |
a161883a RR |
1114 | |
1115 | /* Is this the timeout fd? */ | |
1116 | if (FD_ISSET(timeoutpipe[0], &fds)) | |
1117 | handle_timeout(fd); | |
8ca47e00 RR |
1118 | } |
1119 | } | |
1120 | ||
dde79789 RR |
1121 | /*L:190 |
1122 | * Device Setup | |
1123 | * | |
1124 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1125 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1126 | * routines to allocate and manage them. |
1127 | */ | |
8ca47e00 | 1128 | |
a586d4f6 RR |
1129 | /* The layout of the device page is a "struct lguest_device_desc" followed by a |
1130 | * number of virtqueue descriptors, then two sets of feature bits, then an | |
1131 | * array of configuration bytes. This routine returns the configuration | |
1132 | * pointer. */ | |
1133 | static u8 *device_config(const struct device *dev) | |
1134 | { | |
1135 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1136 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1137 | + dev->feature_len * 2; | |
17cbca2b RR |
1138 | } |
1139 | ||
a586d4f6 RR |
1140 | /* This routine allocates a new "struct lguest_device_desc" from descriptor |
1141 | * table page just above the Guest's normal memory. It returns a pointer to | |
1142 | * that descriptor. */ | |
1143 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
17cbca2b | 1144 | { |
a586d4f6 RR |
1145 | struct lguest_device_desc d = { .type = type }; |
1146 | void *p; | |
17cbca2b | 1147 | |
a586d4f6 RR |
1148 | /* Figure out where the next device config is, based on the last one. */ |
1149 | if (devices.lastdev) | |
1150 | p = device_config(devices.lastdev) | |
1151 | + devices.lastdev->desc->config_len; | |
1152 | else | |
1153 | p = devices.descpage; | |
17cbca2b | 1154 | |
a586d4f6 RR |
1155 | /* We only have one page for all the descriptors. */ |
1156 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1157 | errx(1, "Too many devices"); | |
17cbca2b | 1158 | |
a586d4f6 RR |
1159 | /* p might not be aligned, so we memcpy in. */ |
1160 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1161 | } |
1162 | ||
a586d4f6 RR |
1163 | /* Each device descriptor is followed by the description of its virtqueues. We |
1164 | * specify how many descriptors the virtqueue is to have. */ | |
17cbca2b | 1165 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
a161883a | 1166 | void (*handle_output)(int, struct virtqueue *, bool)) |
17cbca2b RR |
1167 | { |
1168 | unsigned int pages; | |
1169 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1170 | void *p; | |
1171 | ||
a6bd8e13 | 1172 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1173 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 1174 | / getpagesize(); |
17cbca2b RR |
1175 | p = get_pages(pages); |
1176 | ||
d1c856e0 RR |
1177 | /* Initialize the virtqueue */ |
1178 | vq->next = NULL; | |
1179 | vq->last_avail_idx = 0; | |
1180 | vq->dev = dev; | |
20887611 | 1181 | vq->inflight = 0; |
a161883a | 1182 | vq->blocked = false; |
d1c856e0 | 1183 | |
17cbca2b RR |
1184 | /* Initialize the configuration. */ |
1185 | vq->config.num = num_descs; | |
1186 | vq->config.irq = devices.next_irq++; | |
1187 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1188 | ||
1189 | /* Initialize the vring. */ | |
2966af73 | 1190 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1191 | |
a586d4f6 RR |
1192 | /* Append virtqueue to this device's descriptor. We use |
1193 | * device_config() to get the end of the device's current virtqueues; | |
1194 | * we check that we haven't added any config or feature information | |
1195 | * yet, otherwise we'd be overwriting them. */ | |
1196 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); | |
1197 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1198 | dev->num_vq++; |
a586d4f6 RR |
1199 | dev->desc->num_vq++; |
1200 | ||
1201 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b RR |
1202 | |
1203 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1204 | * second. */ | |
1205 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1206 | *i = vq; | |
1207 | ||
e1e72965 RR |
1208 | /* Set the routine to call when the Guest does something to this |
1209 | * virtqueue. */ | |
17cbca2b | 1210 | vq->handle_output = handle_output; |
e1e72965 | 1211 | |
426e3e0a RR |
1212 | /* As an optimization, set the advisory "Don't Notify Me" flag if we |
1213 | * don't have a handler */ | |
17cbca2b RR |
1214 | if (!handle_output) |
1215 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1216 | } |
1217 | ||
6e5aa7ef | 1218 | /* The first half of the feature bitmask is for us to advertise features. The |
a6bd8e13 | 1219 | * second half is for the Guest to accept features. */ |
a586d4f6 RR |
1220 | static void add_feature(struct device *dev, unsigned bit) |
1221 | { | |
6e5aa7ef | 1222 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1223 | |
1224 | /* We can't extend the feature bits once we've added config bytes */ | |
1225 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1226 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1227 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1228 | } |
1229 | ||
a586d4f6 RR |
1230 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1231 | } | |
1232 | ||
1233 | /* This routine sets the configuration fields for an existing device's | |
1234 | * descriptor. It only works for the last device, but that's OK because that's | |
1235 | * how we use it. */ | |
1236 | static void set_config(struct device *dev, unsigned len, const void *conf) | |
1237 | { | |
1238 | /* Check we haven't overflowed our single page. */ | |
1239 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1240 | errx(1, "Too many devices"); | |
1241 | ||
1242 | /* Copy in the config information, and store the length. */ | |
1243 | memcpy(device_config(dev), conf, len); | |
1244 | dev->desc->config_len = len; | |
1245 | } | |
1246 | ||
17cbca2b | 1247 | /* This routine does all the creation and setup of a new device, including |
a6bd8e13 RR |
1248 | * calling new_dev_desc() to allocate the descriptor and device memory. |
1249 | * | |
1250 | * See what I mean about userspace being boring? */ | |
17cbca2b RR |
1251 | static struct device *new_device(const char *name, u16 type, int fd, |
1252 | bool (*handle_input)(int, struct device *)) | |
8ca47e00 RR |
1253 | { |
1254 | struct device *dev = malloc(sizeof(*dev)); | |
1255 | ||
dde79789 | 1256 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1257 | dev->fd = fd; |
dde79789 RR |
1258 | /* If we have an input handler for this file descriptor, then we add it |
1259 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1260 | if (handle_input) |
17cbca2b RR |
1261 | add_device_fd(dev->fd); |
1262 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1263 | dev->handle_input = handle_input; |
17cbca2b | 1264 | dev->name = name; |
d1c856e0 | 1265 | dev->vq = NULL; |
a007a751 | 1266 | dev->ready = NULL; |
713b15b3 RR |
1267 | dev->feature_len = 0; |
1268 | dev->num_vq = 0; | |
a586d4f6 RR |
1269 | |
1270 | /* Append to device list. Prepending to a single-linked list is | |
1271 | * easier, but the user expects the devices to be arranged on the bus | |
1272 | * in command-line order. The first network device on the command line | |
1273 | * is eth0, the first block device /dev/vda, etc. */ | |
1274 | if (devices.lastdev) | |
1275 | devices.lastdev->next = dev; | |
1276 | else | |
1277 | devices.dev = dev; | |
1278 | devices.lastdev = dev; | |
1279 | ||
8ca47e00 RR |
1280 | return dev; |
1281 | } | |
1282 | ||
dde79789 RR |
1283 | /* Our first setup routine is the console. It's a fairly simple device, but |
1284 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1285 | static void setup_console(void) |
8ca47e00 RR |
1286 | { |
1287 | struct device *dev; | |
1288 | ||
dde79789 | 1289 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1290 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1291 | struct termios term = orig_term; | |
dde79789 RR |
1292 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1293 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1294 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1295 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1296 | /* If we exit gracefully, the original settings will be |
1297 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1298 | atexit(restore_term); |
1299 | } | |
1300 | ||
17cbca2b RR |
1301 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1302 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1303 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1304 | dev->priv = malloc(sizeof(struct console_abort)); |
1305 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1306 | |
56ae43df RR |
1307 | /* The console needs two virtqueues: the input then the output. When |
1308 | * they put something the input queue, we make sure we're listening to | |
1309 | * stdin. When they put something in the output queue, we write it to | |
e1e72965 | 1310 | * stdout. */ |
56ae43df | 1311 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
17cbca2b RR |
1312 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1313 | ||
1314 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1315 | } |
17cbca2b | 1316 | /*:*/ |
8ca47e00 | 1317 | |
a161883a RR |
1318 | static void timeout_alarm(int sig) |
1319 | { | |
1320 | write(timeoutpipe[1], "", 1); | |
1321 | } | |
1322 | ||
1323 | static void setup_timeout(void) | |
1324 | { | |
1325 | if (pipe(timeoutpipe) != 0) | |
1326 | err(1, "Creating timeout pipe"); | |
1327 | ||
1328 | if (fcntl(timeoutpipe[1], F_SETFL, | |
1329 | fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0) | |
1330 | err(1, "Making timeout pipe nonblocking"); | |
1331 | ||
1332 | add_device_fd(timeoutpipe[0]); | |
1333 | signal(SIGALRM, timeout_alarm); | |
1334 | } | |
1335 | ||
17cbca2b RR |
1336 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1337 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1338 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1339 | * |
17cbca2b RR |
1340 | * More sopisticated is to use one of the tools developed for project like UML |
1341 | * to do networking. | |
dde79789 | 1342 | * |
17cbca2b RR |
1343 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1344 | * completely generic ("here's my vring, attach to your vring") and would work | |
1345 | * for any traffic. Of course, namespace and permissions issues need to be | |
1346 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1347 | * multiple inter-guest channels behind one interface, although it would | |
1348 | * require some manner of hotplugging new virtio channels. | |
1349 | * | |
1350 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1351 | |
1352 | static u32 str2ip(const char *ipaddr) | |
1353 | { | |
dec6a2be | 1354 | unsigned int b[4]; |
8ca47e00 | 1355 | |
dec6a2be MM |
1356 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1357 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1358 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1359 | } | |
1360 | ||
1361 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1362 | { | |
1363 | unsigned int m[6]; | |
1364 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1365 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1366 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1367 | mac[0] = m[0]; | |
1368 | mac[1] = m[1]; | |
1369 | mac[2] = m[2]; | |
1370 | mac[3] = m[3]; | |
1371 | mac[4] = m[4]; | |
1372 | mac[5] = m[5]; | |
8ca47e00 RR |
1373 | } |
1374 | ||
dde79789 RR |
1375 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1376 | * network device to the bridge device specified by the command line. | |
1377 | * | |
1378 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1379 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1380 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1381 | { | |
1382 | int ifidx; | |
1383 | struct ifreq ifr; | |
1384 | ||
1385 | if (!*br_name) | |
1386 | errx(1, "must specify bridge name"); | |
1387 | ||
1388 | ifidx = if_nametoindex(if_name); | |
1389 | if (!ifidx) | |
1390 | errx(1, "interface %s does not exist!", if_name); | |
1391 | ||
1392 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1393 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1394 | ifr.ifr_ifindex = ifidx; |
1395 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1396 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1397 | } | |
1398 | ||
dde79789 RR |
1399 | /* This sets up the Host end of the network device with an IP address, brings |
1400 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1401 | * pointer. */ |
dec6a2be | 1402 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1403 | { |
1404 | struct ifreq ifr; | |
1405 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1406 | ||
1407 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1408 | strcpy(ifr.ifr_name, tapif); |
1409 | ||
1410 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
8ca47e00 RR |
1411 | sin->sin_family = AF_INET; |
1412 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1413 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
dec6a2be | 1414 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1415 | ifr.ifr_flags = IFF_UP; |
1416 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1417 | err(1, "Bringing interface %s up", tapif); |
1418 | } | |
1419 | ||
dec6a2be | 1420 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1421 | { |
8ca47e00 | 1422 | struct ifreq ifr; |
dec6a2be MM |
1423 | int netfd; |
1424 | ||
1425 | /* Start with this zeroed. Messy but sure. */ | |
1426 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1427 | |
dde79789 RR |
1428 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1429 | * tap device is like a tun device, only somehow different. To tell | |
1430 | * the truth, I completely blundered my way through this code, but it | |
1431 | * works now! */ | |
8ca47e00 | 1432 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1433 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1434 | strcpy(ifr.ifr_name, "tap%d"); |
1435 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1436 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1437 | |
398f187d RR |
1438 | if (ioctl(netfd, TUNSETOFFLOAD, |
1439 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1440 | err(1, "Could not set features for tun device"); | |
1441 | ||
dde79789 RR |
1442 | /* We don't need checksums calculated for packets coming in this |
1443 | * device: trust us! */ | |
8ca47e00 RR |
1444 | ioctl(netfd, TUNSETNOCSUM, 1); |
1445 | ||
dec6a2be MM |
1446 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1447 | return netfd; | |
1448 | } | |
1449 | ||
1450 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or | |
1451 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1452 | * packets into the Host as if they came in from a normal network card. We | |
1453 | * just shunt packets between the Guest and the tun device. */ | |
1454 | static void setup_tun_net(char *arg) | |
1455 | { | |
1456 | struct device *dev; | |
1457 | int netfd, ipfd; | |
1458 | u32 ip = INADDR_ANY; | |
1459 | bool bridging = false; | |
1460 | char tapif[IFNAMSIZ], *p; | |
1461 | struct virtio_net_config conf; | |
1462 | ||
1463 | netfd = get_tun_device(tapif); | |
1464 | ||
17cbca2b RR |
1465 | /* First we create a new network device. */ |
1466 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1467 | |
56ae43df RR |
1468 | /* Network devices need a receive and a send queue, just like |
1469 | * console. */ | |
5dae785a | 1470 | add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd); |
17cbca2b | 1471 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1472 | |
dde79789 RR |
1473 | /* We need a socket to perform the magic network ioctls to bring up the |
1474 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1475 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1476 | if (ipfd < 0) | |
1477 | err(1, "opening IP socket"); | |
1478 | ||
dde79789 | 1479 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1480 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1481 | arg += strlen(BRIDGE_PFX); |
1482 | bridging = true; | |
1483 | } | |
1484 | ||
1485 | /* A mac address may follow the bridge name or IP address */ | |
1486 | p = strchr(arg, ':'); | |
1487 | if (p) { | |
1488 | str2mac(p+1, conf.mac); | |
40c42076 | 1489 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1490 | *p = '\0'; |
dec6a2be MM |
1491 | } |
1492 | ||
1493 | /* arg is now either an IP address or a bridge name */ | |
1494 | if (bridging) | |
1495 | add_to_bridge(ipfd, tapif, arg); | |
1496 | else | |
8ca47e00 RR |
1497 | ip = str2ip(arg); |
1498 | ||
dec6a2be MM |
1499 | /* Set up the tun device. */ |
1500 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1501 | |
20887611 | 1502 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1503 | /* Expect Guest to handle everything except UFO */ |
1504 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1505 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1506 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1507 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1508 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1509 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1510 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1511 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
a586d4f6 | 1512 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1513 | |
a586d4f6 | 1514 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1515 | close(ipfd); |
1516 | ||
dec6a2be MM |
1517 | devices.device_num++; |
1518 | ||
1519 | if (bridging) | |
1520 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1521 | devices.device_num, tapif, arg); | |
1522 | else | |
1523 | verbose("device %u: tun %s: %s\n", | |
1524 | devices.device_num, tapif, arg); | |
8ca47e00 | 1525 | } |
17cbca2b | 1526 | |
e1e72965 RR |
1527 | /* Our block (disk) device should be really simple: the Guest asks for a block |
1528 | * number and we read or write that position in the file. Unfortunately, that | |
1529 | * was amazingly slow: the Guest waits until the read is finished before | |
1530 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1531 | * |
e1e72965 RR |
1532 | * We could use async I/O, except it's reputed to suck so hard that characters |
1533 | * actually go missing from your code when you try to use it. | |
17cbca2b RR |
1534 | * |
1535 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1536 | ||
e1e72965 | 1537 | /* This hangs off device->priv. */ |
17cbca2b RR |
1538 | struct vblk_info |
1539 | { | |
1540 | /* The size of the file. */ | |
1541 | off64_t len; | |
1542 | ||
1543 | /* The file descriptor for the file. */ | |
1544 | int fd; | |
1545 | ||
1546 | /* IO thread listens on this file descriptor [0]. */ | |
1547 | int workpipe[2]; | |
1548 | ||
1549 | /* IO thread writes to this file descriptor to mark it done, then | |
1550 | * Launcher triggers interrupt to Guest. */ | |
1551 | int done_fd; | |
1552 | }; | |
1553 | ||
e1e72965 RR |
1554 | /*L:210 |
1555 | * The Disk | |
1556 | * | |
1557 | * Remember that the block device is handled by a separate I/O thread. We head | |
1558 | * straight into the core of that thread here: | |
1559 | */ | |
17cbca2b RR |
1560 | static bool service_io(struct device *dev) |
1561 | { | |
1562 | struct vblk_info *vblk = dev->priv; | |
1563 | unsigned int head, out_num, in_num, wlen; | |
1564 | int ret; | |
cb38fa23 | 1565 | u8 *in; |
17cbca2b RR |
1566 | struct virtio_blk_outhdr *out; |
1567 | struct iovec iov[dev->vq->vring.num]; | |
1568 | off64_t off; | |
1569 | ||
e1e72965 | 1570 | /* See if there's a request waiting. If not, nothing to do. */ |
17cbca2b RR |
1571 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
1572 | if (head == dev->vq->vring.num) | |
1573 | return false; | |
1574 | ||
e1e72965 RR |
1575 | /* Every block request should contain at least one output buffer |
1576 | * (detailing the location on disk and the type of request) and one | |
1577 | * input buffer (to hold the result). */ | |
17cbca2b RR |
1578 | if (out_num == 0 || in_num == 0) |
1579 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1580 | head, out_num, in_num); | |
1581 | ||
1582 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1583 | in = convert(&iov[out_num+in_num-1], u8); |
17cbca2b RR |
1584 | off = out->sector * 512; |
1585 | ||
e1e72965 RR |
1586 | /* The block device implements "barriers", where the Guest indicates |
1587 | * that it wants all previous writes to occur before this write. We | |
1588 | * don't have a way of asking our kernel to do a barrier, so we just | |
1589 | * synchronize all the data in the file. Pretty poor, no? */ | |
17cbca2b RR |
1590 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1591 | fdatasync(vblk->fd); | |
1592 | ||
e1e72965 RR |
1593 | /* In general the virtio block driver is allowed to try SCSI commands. |
1594 | * It'd be nice if we supported eject, for example, but we don't. */ | |
17cbca2b RR |
1595 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1596 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1597 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1598 | wlen = sizeof(*in); |
17cbca2b RR |
1599 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
1600 | /* Write */ | |
1601 | ||
1602 | /* Move to the right location in the block file. This can fail | |
1603 | * if they try to write past end. */ | |
1604 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1605 | err(1, "Bad seek to sector %llu", out->sector); | |
1606 | ||
1607 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1608 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1609 | ||
1610 | /* Grr... Now we know how long the descriptor they sent was, we | |
1611 | * make sure they didn't try to write over the end of the block | |
1612 | * file (possibly extending it). */ | |
1613 | if (ret > 0 && off + ret > vblk->len) { | |
1614 | /* Trim it back to the correct length */ | |
1615 | ftruncate64(vblk->fd, vblk->len); | |
1616 | /* Die, bad Guest, die. */ | |
1617 | errx(1, "Write past end %llu+%u", off, ret); | |
1618 | } | |
1200e646 | 1619 | wlen = sizeof(*in); |
cb38fa23 | 1620 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b RR |
1621 | } else { |
1622 | /* Read */ | |
1623 | ||
1624 | /* Move to the right location in the block file. This can fail | |
1625 | * if they try to read past end. */ | |
1626 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1627 | err(1, "Bad seek to sector %llu", out->sector); | |
1628 | ||
1629 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1630 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1631 | if (ret >= 0) { | |
1200e646 | 1632 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1633 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1634 | } else { |
1200e646 | 1635 | wlen = sizeof(*in); |
cb38fa23 | 1636 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1637 | } |
1638 | } | |
1639 | ||
d1881d31 RR |
1640 | /* OK, so we noted that it was pretty poor to use an fdatasync as a |
1641 | * barrier. But Christoph Hellwig points out that we need a sync | |
1642 | * *afterwards* as well: "Barriers specify no reordering to the front | |
1643 | * or the back." And Jens Axboe confirmed it, so here we are: */ | |
1644 | if (out->type & VIRTIO_BLK_T_BARRIER) | |
1645 | fdatasync(vblk->fd); | |
1646 | ||
17cbca2b RR |
1647 | /* We can't trigger an IRQ, because we're not the Launcher. It does |
1648 | * that when we tell it we're done. */ | |
1649 | add_used(dev->vq, head, wlen); | |
1650 | return true; | |
1651 | } | |
1652 | ||
1653 | /* This is the thread which actually services the I/O. */ | |
1654 | static int io_thread(void *_dev) | |
1655 | { | |
1656 | struct device *dev = _dev; | |
1657 | struct vblk_info *vblk = dev->priv; | |
1658 | char c; | |
1659 | ||
1660 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1661 | close(vblk->workpipe[1]); | |
1662 | /* Close the other side of the done_fd pipe. */ | |
1663 | close(dev->fd); | |
1664 | ||
1665 | /* When this read fails, it means Launcher died, so we follow. */ | |
1666 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
e1e72965 | 1667 | /* We acknowledge each request immediately to reduce latency, |
17cbca2b | 1668 | * rather than waiting until we've done them all. I haven't |
a6bd8e13 RR |
1669 | * measured to see if it makes any difference. |
1670 | * | |
1671 | * That would be an interesting test, wouldn't it? You could | |
1672 | * also try having more than one I/O thread. */ | |
17cbca2b RR |
1673 | while (service_io(dev)) |
1674 | write(vblk->done_fd, &c, 1); | |
1675 | } | |
1676 | return 0; | |
1677 | } | |
1678 | ||
e1e72965 | 1679 | /* Now we've seen the I/O thread, we return to the Launcher to see what happens |
a6bd8e13 | 1680 | * when that thread tells us it's completed some I/O. */ |
17cbca2b RR |
1681 | static bool handle_io_finish(int fd, struct device *dev) |
1682 | { | |
1683 | char c; | |
1684 | ||
e1e72965 RR |
1685 | /* If the I/O thread died, presumably it printed the error, so we |
1686 | * simply exit. */ | |
17cbca2b RR |
1687 | if (read(dev->fd, &c, 1) != 1) |
1688 | exit(1); | |
1689 | ||
1690 | /* It did some work, so trigger the irq. */ | |
1691 | trigger_irq(fd, dev->vq); | |
1692 | return true; | |
1693 | } | |
1694 | ||
e1e72965 | 1695 | /* When the Guest submits some I/O, we just need to wake the I/O thread. */ |
a161883a | 1696 | static void handle_virtblk_output(int fd, struct virtqueue *vq, bool timeout) |
17cbca2b RR |
1697 | { |
1698 | struct vblk_info *vblk = vq->dev->priv; | |
1699 | char c = 0; | |
1700 | ||
1701 | /* Wake up I/O thread and tell it to go to work! */ | |
1702 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1703 | /* Presumably it indicated why it died. */ | |
1704 | exit(1); | |
1705 | } | |
1706 | ||
e1e72965 | 1707 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1708 | static void setup_block_file(const char *filename) |
1709 | { | |
1710 | int p[2]; | |
1711 | struct device *dev; | |
1712 | struct vblk_info *vblk; | |
1713 | void *stack; | |
a586d4f6 | 1714 | struct virtio_blk_config conf; |
17cbca2b RR |
1715 | |
1716 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1717 | pipe(p); | |
1718 | ||
1719 | /* The device responds to return from I/O thread. */ | |
1720 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1721 | ||
e1e72965 | 1722 | /* The device has one virtqueue, where the Guest places requests. */ |
17cbca2b RR |
1723 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); |
1724 | ||
1725 | /* Allocate the room for our own bookkeeping */ | |
1726 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1727 | ||
1728 | /* First we open the file and store the length. */ | |
1729 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1730 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1731 | ||
a586d4f6 RR |
1732 | /* We support barriers. */ |
1733 | add_feature(dev, VIRTIO_BLK_F_BARRIER); | |
1734 | ||
17cbca2b | 1735 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1736 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b RR |
1737 | |
1738 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1739 | * for the in and out elements. */ | |
a586d4f6 RR |
1740 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1741 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1742 | ||
1743 | set_config(dev, sizeof(conf), &conf); | |
17cbca2b RR |
1744 | |
1745 | /* The I/O thread writes to this end of the pipe when done. */ | |
1746 | vblk->done_fd = p[1]; | |
1747 | ||
e1e72965 RR |
1748 | /* This is the second pipe, which is how we tell the I/O thread about |
1749 | * more work. */ | |
17cbca2b RR |
1750 | pipe(vblk->workpipe); |
1751 | ||
a6bd8e13 RR |
1752 | /* Create stack for thread and run it. Since stack grows upwards, we |
1753 | * point the stack pointer to the end of this region. */ | |
17cbca2b | 1754 | stack = malloc(32768); |
ec04b13f BR |
1755 | /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from |
1756 | * becoming a zombie. */ | |
a6bd8e13 | 1757 | if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1) |
17cbca2b RR |
1758 | err(1, "Creating clone"); |
1759 | ||
1760 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1761 | close(vblk->done_fd); | |
1762 | close(vblk->workpipe[0]); | |
1763 | ||
1764 | verbose("device %u: virtblock %llu sectors\n", | |
a586d4f6 | 1765 | devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1766 | } |
28fd6d7f RR |
1767 | |
1768 | /* Our random number generator device reads from /dev/random into the Guest's | |
1769 | * input buffers. The usual case is that the Guest doesn't want random numbers | |
1770 | * and so has no buffers although /dev/random is still readable, whereas | |
1771 | * console is the reverse. | |
1772 | * | |
1773 | * The same logic applies, however. */ | |
1774 | static bool handle_rng_input(int fd, struct device *dev) | |
1775 | { | |
1776 | int len; | |
1777 | unsigned int head, in_num, out_num, totlen = 0; | |
1778 | struct iovec iov[dev->vq->vring.num]; | |
1779 | ||
1780 | /* First we need a buffer from the Guests's virtqueue. */ | |
1781 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
1782 | ||
1783 | /* If they're not ready for input, stop listening to this file | |
1784 | * descriptor. We'll start again once they add an input buffer. */ | |
1785 | if (head == dev->vq->vring.num) | |
1786 | return false; | |
1787 | ||
1788 | if (out_num) | |
1789 | errx(1, "Output buffers in rng?"); | |
1790 | ||
1791 | /* This is why we convert to iovecs: the readv() call uses them, and so | |
1792 | * it reads straight into the Guest's buffer. We loop to make sure we | |
1793 | * fill it. */ | |
1794 | while (!iov_empty(iov, in_num)) { | |
1795 | len = readv(dev->fd, iov, in_num); | |
1796 | if (len <= 0) | |
1797 | err(1, "Read from /dev/random gave %i", len); | |
1798 | iov_consume(iov, in_num, len); | |
1799 | totlen += len; | |
1800 | } | |
1801 | ||
1802 | /* Tell the Guest about the new input. */ | |
1803 | add_used_and_trigger(fd, dev->vq, head, totlen); | |
1804 | ||
1805 | /* Everything went OK! */ | |
1806 | return true; | |
1807 | } | |
1808 | ||
1809 | /* And this creates a "hardware" random number device for the Guest. */ | |
1810 | static void setup_rng(void) | |
1811 | { | |
1812 | struct device *dev; | |
1813 | int fd; | |
1814 | ||
1815 | fd = open_or_die("/dev/random", O_RDONLY); | |
1816 | ||
1817 | /* The device responds to return from I/O thread. */ | |
1818 | dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input); | |
1819 | ||
1820 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
1821 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
1822 | ||
1823 | verbose("device %u: rng\n", devices.device_num++); | |
1824 | } | |
a6bd8e13 | 1825 | /* That's the end of device setup. */ |
ec04b13f | 1826 | |
a6bd8e13 | 1827 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1828 | static void __attribute__((noreturn)) restart_guest(void) |
1829 | { | |
1830 | unsigned int i; | |
1831 | ||
8c79873d RR |
1832 | /* Since we don't track all open fds, we simply close everything beyond |
1833 | * stderr. */ | |
ec04b13f BR |
1834 | for (i = 3; i < FD_SETSIZE; i++) |
1835 | close(i); | |
8c79873d RR |
1836 | |
1837 | /* The exec automatically gets rid of the I/O and Waker threads. */ | |
ec04b13f BR |
1838 | execv(main_args[0], main_args); |
1839 | err(1, "Could not exec %s", main_args[0]); | |
1840 | } | |
8ca47e00 | 1841 | |
a6bd8e13 | 1842 | /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves |
dde79789 | 1843 | * its input and output, and finally, lays it to rest. */ |
17cbca2b | 1844 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
8ca47e00 RR |
1845 | { |
1846 | for (;;) { | |
511801dc | 1847 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1848 | unsigned long notify_addr; |
8ca47e00 RR |
1849 | int readval; |
1850 | ||
1851 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1852 | readval = pread(lguest_fd, ¬ify_addr, |
1853 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1854 | |
17cbca2b RR |
1855 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1856 | if (readval == sizeof(notify_addr)) { | |
1857 | verbose("Notify on address %#lx\n", notify_addr); | |
1858 | handle_output(lguest_fd, notify_addr); | |
8ca47e00 | 1859 | continue; |
dde79789 | 1860 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1861 | } else if (errno == ENOENT) { |
1862 | char reason[1024] = { 0 }; | |
e3283fa0 | 1863 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1864 | errx(1, "%s", reason); |
ec04b13f BR |
1865 | /* ERESTART means that we need to reboot the guest */ |
1866 | } else if (errno == ERESTART) { | |
1867 | restart_guest(); | |
a161883a | 1868 | /* EAGAIN means a signal (timeout). |
dde79789 | 1869 | * Anything else means a bug or incompatible change. */ |
8ca47e00 RR |
1870 | } else if (errno != EAGAIN) |
1871 | err(1, "Running guest failed"); | |
dde79789 | 1872 | |
e3283fa0 GOC |
1873 | /* Only service input on thread for CPU 0. */ |
1874 | if (cpu_id != 0) | |
1875 | continue; | |
1876 | ||
e1e72965 | 1877 | /* Service input, then unset the BREAK to release the Waker. */ |
17cbca2b | 1878 | handle_input(lguest_fd); |
e3283fa0 | 1879 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) |
8ca47e00 RR |
1880 | err(1, "Resetting break"); |
1881 | } | |
1882 | } | |
a6bd8e13 | 1883 | /*L:240 |
e1e72965 RR |
1884 | * This is the end of the Launcher. The good news: we are over halfway |
1885 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1886 | * of us. | |
dde79789 | 1887 | * |
e1e72965 RR |
1888 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1889 | * "make Host". | |
1890 | :*/ | |
8ca47e00 RR |
1891 | |
1892 | static struct option opts[] = { | |
1893 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1894 | { "tunnet", 1, NULL, 't' }, |
1895 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1896 | { "rng", 0, NULL, 'r' }, |
8ca47e00 RR |
1897 | { "initrd", 1, NULL, 'i' }, |
1898 | { NULL }, | |
1899 | }; | |
1900 | static void usage(void) | |
1901 | { | |
1902 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1903 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1904 | "|--block=<filename>|--initrd=<filename>]...\n" |
1905 | "<mem-in-mb> vmlinux [args...]"); | |
1906 | } | |
1907 | ||
3c6b5bfa | 1908 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1909 | int main(int argc, char *argv[]) |
1910 | { | |
47436aa4 RR |
1911 | /* Memory, top-level pagetable, code startpoint and size of the |
1912 | * (optional) initrd. */ | |
58a24566 | 1913 | unsigned long mem = 0, start, initrd_size = 0; |
e1e72965 | 1914 | /* Two temporaries and the /dev/lguest file descriptor. */ |
6570c459 | 1915 | int i, c, lguest_fd; |
3c6b5bfa | 1916 | /* The boot information for the Guest. */ |
43d33b21 | 1917 | struct boot_params *boot; |
dde79789 | 1918 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1919 | const char *initrd_name = NULL; |
1920 | ||
ec04b13f BR |
1921 | /* Save the args: we "reboot" by execing ourselves again. */ |
1922 | main_args = argv; | |
1923 | /* We don't "wait" for the children, so prevent them from becoming | |
1924 | * zombies. */ | |
1925 | signal(SIGCHLD, SIG_IGN); | |
1926 | ||
dde79789 RR |
1927 | /* First we initialize the device list. Since console and network |
1928 | * device receive input from a file descriptor, we keep an fdset | |
1929 | * (infds) and the maximum fd number (max_infd) with the head of the | |
a586d4f6 | 1930 | * list. We also keep a pointer to the last device. Finally, we keep |
a6bd8e13 RR |
1931 | * the next interrupt number to use for devices (1: remember that 0 is |
1932 | * used by the timer). */ | |
17cbca2b RR |
1933 | FD_ZERO(&devices.infds); |
1934 | devices.max_infd = -1; | |
a586d4f6 | 1935 | devices.lastdev = NULL; |
17cbca2b | 1936 | devices.next_irq = 1; |
8ca47e00 | 1937 | |
e3283fa0 | 1938 | cpu_id = 0; |
dde79789 RR |
1939 | /* We need to know how much memory so we can set up the device |
1940 | * descriptor and memory pages for the devices as we parse the command | |
1941 | * line. So we quickly look through the arguments to find the amount | |
1942 | * of memory now. */ | |
6570c459 RR |
1943 | for (i = 1; i < argc; i++) { |
1944 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1945 | mem = atoi(argv[i]) * 1024 * 1024; |
1946 | /* We start by mapping anonymous pages over all of | |
1947 | * guest-physical memory range. This fills it with 0, | |
1948 | * and ensures that the Guest won't be killed when it | |
1949 | * tries to access it. */ | |
1950 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1951 | + DEVICE_PAGES); | |
1952 | guest_limit = mem; | |
1953 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1954 | devices.descpage = get_pages(1); |
6570c459 RR |
1955 | break; |
1956 | } | |
1957 | } | |
dde79789 RR |
1958 | |
1959 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1960 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1961 | switch (c) { | |
1962 | case 'v': | |
1963 | verbose = true; | |
1964 | break; | |
8ca47e00 | 1965 | case 't': |
17cbca2b | 1966 | setup_tun_net(optarg); |
8ca47e00 RR |
1967 | break; |
1968 | case 'b': | |
17cbca2b | 1969 | setup_block_file(optarg); |
8ca47e00 | 1970 | break; |
28fd6d7f RR |
1971 | case 'r': |
1972 | setup_rng(); | |
1973 | break; | |
8ca47e00 RR |
1974 | case 'i': |
1975 | initrd_name = optarg; | |
1976 | break; | |
1977 | default: | |
1978 | warnx("Unknown argument %s", argv[optind]); | |
1979 | usage(); | |
1980 | } | |
1981 | } | |
dde79789 RR |
1982 | /* After the other arguments we expect memory and kernel image name, |
1983 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1984 | if (optind + 2 > argc) |
1985 | usage(); | |
1986 | ||
3c6b5bfa RR |
1987 | verbose("Guest base is at %p\n", guest_base); |
1988 | ||
dde79789 | 1989 | /* We always have a console device */ |
17cbca2b | 1990 | setup_console(); |
8ca47e00 | 1991 | |
a161883a RR |
1992 | /* We can timeout waiting for Guest network transmit. */ |
1993 | setup_timeout(); | |
1994 | ||
8ca47e00 | 1995 | /* Now we load the kernel */ |
47436aa4 | 1996 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1997 | |
3c6b5bfa RR |
1998 | /* Boot information is stashed at physical address 0 */ |
1999 | boot = from_guest_phys(0); | |
2000 | ||
dde79789 | 2001 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
2002 | if (initrd_name) { |
2003 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
2004 | /* These are the location in the Linux boot header where the |
2005 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
2006 | boot->hdr.ramdisk_image = mem - initrd_size; |
2007 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 2008 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 2009 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
2010 | } |
2011 | ||
dde79789 RR |
2012 | /* The Linux boot header contains an "E820" memory map: ours is a |
2013 | * simple, single region. */ | |
43d33b21 RR |
2014 | boot->e820_entries = 1; |
2015 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 2016 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
2017 | * line after the boot header. */ |
2018 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
e1e72965 | 2019 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 2020 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 2021 | |
814a0e5c | 2022 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 2023 | boot->hdr.version = 0x207; |
814a0e5c RR |
2024 | |
2025 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 2026 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 2027 | |
43d33b21 RR |
2028 | /* Tell the entry path not to try to reload segment registers. */ |
2029 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 2030 | |
dde79789 RR |
2031 | /* We tell the kernel to initialize the Guest: this returns the open |
2032 | * /dev/lguest file descriptor. */ | |
58a24566 | 2033 | lguest_fd = tell_kernel(start); |
dde79789 | 2034 | |
8c79873d RR |
2035 | /* We clone off a thread, which wakes the Launcher whenever one of the |
2036 | * input file descriptors needs attention. We call this the Waker, and | |
2037 | * we'll cover it in a moment. */ | |
2038 | setup_waker(lguest_fd); | |
8ca47e00 | 2039 | |
dde79789 | 2040 | /* Finally, run the Guest. This doesn't return. */ |
17cbca2b | 2041 | run_guest(lguest_fd); |
8ca47e00 | 2042 | } |
f56a384e RR |
2043 | /*:*/ |
2044 | ||
2045 | /*M:999 | |
2046 | * Mastery is done: you now know everything I do. | |
2047 | * | |
2048 | * But surely you have seen code, features and bugs in your wanderings which | |
2049 | * you now yearn to attack? That is the real game, and I look forward to you | |
2050 | * patching and forking lguest into the Your-Name-Here-visor. | |
2051 | * | |
2052 | * Farewell, and good coding! | |
2053 | * Rusty Russell. | |
2054 | */ |