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" |
43d33b21 | 47 | #include "asm-x86/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 | |
42b36cc0 RR |
69 | /* This will occupy 2 pages: it must be a power of 2. */ |
70 | #define VIRTQUEUE_NUM 128 | |
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 | ||
79 | /* The pipe to send commands to the waker process */ | |
8ca47e00 | 80 | static int waker_fd; |
3c6b5bfa RR |
81 | /* The pointer to the start of guest memory. */ |
82 | static void *guest_base; | |
83 | /* The maximum guest physical address allowed, and maximum possible. */ | |
84 | static unsigned long guest_limit, guest_max; | |
a161883a RR |
85 | /* The pipe for signal hander to write to. */ |
86 | static int timeoutpipe[2]; | |
aa124984 | 87 | static unsigned int timeout_usec = 500; |
8ca47e00 | 88 | |
e3283fa0 GOC |
89 | /* a per-cpu variable indicating whose vcpu is currently running */ |
90 | static unsigned int __thread cpu_id; | |
91 | ||
dde79789 | 92 | /* This is our list of devices. */ |
8ca47e00 RR |
93 | struct device_list |
94 | { | |
dde79789 RR |
95 | /* Summary information about the devices in our list: ready to pass to |
96 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
97 | fd_set infds; |
98 | int max_infd; | |
99 | ||
17cbca2b RR |
100 | /* Counter to assign interrupt numbers. */ |
101 | unsigned int next_irq; | |
102 | ||
103 | /* Counter to print out convenient device numbers. */ | |
104 | unsigned int device_num; | |
105 | ||
dde79789 | 106 | /* The descriptor page for the devices. */ |
17cbca2b RR |
107 | u8 *descpage; |
108 | ||
dde79789 | 109 | /* A single linked list of devices. */ |
8ca47e00 | 110 | struct device *dev; |
a586d4f6 RR |
111 | /* And a pointer to the last device for easy append and also for |
112 | * configuration appending. */ | |
113 | struct device *lastdev; | |
8ca47e00 RR |
114 | }; |
115 | ||
17cbca2b RR |
116 | /* The list of Guest devices, based on command line arguments. */ |
117 | static struct device_list devices; | |
118 | ||
dde79789 | 119 | /* The device structure describes a single device. */ |
8ca47e00 RR |
120 | struct device |
121 | { | |
dde79789 | 122 | /* The linked-list pointer. */ |
8ca47e00 | 123 | struct device *next; |
17cbca2b RR |
124 | |
125 | /* The this device's descriptor, as mapped into the Guest. */ | |
8ca47e00 | 126 | struct lguest_device_desc *desc; |
17cbca2b RR |
127 | |
128 | /* The name of this device, for --verbose. */ | |
129 | const char *name; | |
8ca47e00 | 130 | |
dde79789 RR |
131 | /* If handle_input is set, it wants to be called when this file |
132 | * descriptor is ready. */ | |
8ca47e00 RR |
133 | int fd; |
134 | bool (*handle_input)(int fd, struct device *me); | |
135 | ||
17cbca2b RR |
136 | /* Any queues attached to this device */ |
137 | struct virtqueue *vq; | |
8ca47e00 | 138 | |
a007a751 RR |
139 | /* Handle status being finalized (ie. feature bits stable). */ |
140 | void (*ready)(struct device *me); | |
141 | ||
8ca47e00 RR |
142 | /* Device-specific data. */ |
143 | void *priv; | |
144 | }; | |
145 | ||
17cbca2b RR |
146 | /* The virtqueue structure describes a queue attached to a device. */ |
147 | struct virtqueue | |
148 | { | |
149 | struct virtqueue *next; | |
150 | ||
151 | /* Which device owns me. */ | |
152 | struct device *dev; | |
153 | ||
154 | /* The configuration for this queue. */ | |
155 | struct lguest_vqconfig config; | |
156 | ||
157 | /* The actual ring of buffers. */ | |
158 | struct vring vring; | |
159 | ||
160 | /* Last available index we saw. */ | |
161 | u16 last_avail_idx; | |
162 | ||
a161883a RR |
163 | /* The routine to call when the Guest pings us, or timeout. */ |
164 | void (*handle_output)(int fd, struct virtqueue *me, bool timeout); | |
20887611 RR |
165 | |
166 | /* Outstanding buffers */ | |
167 | unsigned int inflight; | |
a161883a RR |
168 | |
169 | /* Is this blocked awaiting a timer? */ | |
170 | bool blocked; | |
17cbca2b RR |
171 | }; |
172 | ||
ec04b13f BR |
173 | /* Remember the arguments to the program so we can "reboot" */ |
174 | static char **main_args; | |
175 | ||
17cbca2b RR |
176 | /* Since guest is UP and we don't run at the same time, we don't need barriers. |
177 | * But I include them in the code in case others copy it. */ | |
178 | #define wmb() | |
179 | ||
180 | /* Convert an iovec element to the given type. | |
181 | * | |
182 | * This is a fairly ugly trick: we need to know the size of the type and | |
183 | * alignment requirement to check the pointer is kosher. It's also nice to | |
184 | * have the name of the type in case we report failure. | |
185 | * | |
186 | * Typing those three things all the time is cumbersome and error prone, so we | |
187 | * have a macro which sets them all up and passes to the real function. */ | |
188 | #define convert(iov, type) \ | |
189 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
190 | ||
191 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
192 | const char *name) | |
193 | { | |
194 | if (iov->iov_len != size) | |
195 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
196 | if ((unsigned long)iov->iov_base % align != 0) | |
197 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
198 | return iov->iov_base; | |
199 | } | |
200 | ||
b5111790 RR |
201 | /* Wrapper for the last available index. Makes it easier to change. */ |
202 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
203 | ||
17cbca2b RR |
204 | /* The virtio configuration space is defined to be little-endian. x86 is |
205 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
206 | #define cpu_to_le16(v16) (v16) | |
207 | #define cpu_to_le32(v32) (v32) | |
208 | #define cpu_to_le64(v64) (v64) | |
209 | #define le16_to_cpu(v16) (v16) | |
210 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 211 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 212 | |
28fd6d7f RR |
213 | /* Is this iovec empty? */ |
214 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
215 | { | |
216 | unsigned int i; | |
217 | ||
218 | for (i = 0; i < num_iov; i++) | |
219 | if (iov[i].iov_len) | |
220 | return false; | |
221 | return true; | |
222 | } | |
223 | ||
224 | /* Take len bytes from the front of this iovec. */ | |
225 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) | |
226 | { | |
227 | unsigned int i; | |
228 | ||
229 | for (i = 0; i < num_iov; i++) { | |
230 | unsigned int used; | |
231 | ||
232 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
233 | iov[i].iov_base += used; | |
234 | iov[i].iov_len -= used; | |
235 | len -= used; | |
236 | } | |
237 | assert(len == 0); | |
238 | } | |
239 | ||
6e5aa7ef RR |
240 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
241 | static u8 *get_feature_bits(struct device *dev) | |
242 | { | |
243 | return (u8 *)(dev->desc + 1) | |
244 | + dev->desc->num_vq * sizeof(struct lguest_vqconfig); | |
245 | } | |
246 | ||
3c6b5bfa RR |
247 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
248 | * where pointers run wild and free! Unfortunately, like most userspace | |
249 | * programs, it's quite boring (which is why everyone likes to hack on the | |
250 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
251 | * will get you through this section. Or, maybe not. | |
252 | * | |
253 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
254 | * memory and stores it in "guest_base". In other words, Guest physical == | |
255 | * Launcher virtual with an offset. | |
256 | * | |
257 | * This can be tough to get your head around, but usually it just means that we | |
258 | * use these trivial conversion functions when the Guest gives us it's | |
259 | * "physical" addresses: */ | |
260 | static void *from_guest_phys(unsigned long addr) | |
261 | { | |
262 | return guest_base + addr; | |
263 | } | |
264 | ||
265 | static unsigned long to_guest_phys(const void *addr) | |
266 | { | |
267 | return (addr - guest_base); | |
268 | } | |
269 | ||
dde79789 RR |
270 | /*L:130 |
271 | * Loading the Kernel. | |
272 | * | |
273 | * We start with couple of simple helper routines. open_or_die() avoids | |
274 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
275 | static int open_or_die(const char *name, int flags) |
276 | { | |
277 | int fd = open(name, flags); | |
278 | if (fd < 0) | |
279 | err(1, "Failed to open %s", name); | |
280 | return fd; | |
281 | } | |
282 | ||
3c6b5bfa RR |
283 | /* map_zeroed_pages() takes a number of pages. */ |
284 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 285 | { |
3c6b5bfa RR |
286 | int fd = open_or_die("/dev/zero", O_RDONLY); |
287 | void *addr; | |
8ca47e00 | 288 | |
dde79789 | 289 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
290 | * copied). */ |
291 | addr = mmap(NULL, getpagesize() * num, | |
292 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
293 | if (addr == MAP_FAILED) | |
294 | err(1, "Mmaping %u pages of /dev/zero", num); | |
34bdaab4 | 295 | close(fd); |
3c6b5bfa RR |
296 | |
297 | return addr; | |
298 | } | |
299 | ||
300 | /* Get some more pages for a device. */ | |
301 | static void *get_pages(unsigned int num) | |
302 | { | |
303 | void *addr = from_guest_phys(guest_limit); | |
304 | ||
305 | guest_limit += num * getpagesize(); | |
306 | if (guest_limit > guest_max) | |
307 | errx(1, "Not enough memory for devices"); | |
308 | return addr; | |
8ca47e00 RR |
309 | } |
310 | ||
6649bb7a RM |
311 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
312 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
313 | * it falls back to reading the memory in. */ | |
314 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
315 | { | |
316 | ssize_t r; | |
317 | ||
318 | /* We map writable even though for some segments are marked read-only. | |
319 | * The kernel really wants to be writable: it patches its own | |
320 | * instructions. | |
321 | * | |
322 | * MAP_PRIVATE means that the page won't be copied until a write is | |
323 | * done to it. This allows us to share untouched memory between | |
324 | * Guests. */ | |
325 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
326 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
327 | return; | |
328 | ||
329 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
330 | r = pread(fd, addr, len, offset); | |
331 | if (r != len) | |
332 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
333 | } | |
334 | ||
dde79789 RR |
335 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
336 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
337 | * by all modern binaries on Linux including the kernel. | |
338 | * | |
339 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
340 | * address. We use the physical address; the Guest will map itself to the |
341 | * virtual address. | |
dde79789 RR |
342 | * |
343 | * We return the starting address. */ | |
47436aa4 | 344 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 345 | { |
8ca47e00 RR |
346 | Elf32_Phdr phdr[ehdr->e_phnum]; |
347 | unsigned int i; | |
8ca47e00 | 348 | |
dde79789 RR |
349 | /* Sanity checks on the main ELF header: an x86 executable with a |
350 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
351 | if (ehdr->e_type != ET_EXEC |
352 | || ehdr->e_machine != EM_386 | |
353 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
354 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
355 | errx(1, "Malformed elf header"); | |
356 | ||
dde79789 RR |
357 | /* An ELF executable contains an ELF header and a number of "program" |
358 | * headers which indicate which parts ("segments") of the program to | |
359 | * load where. */ | |
360 | ||
361 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
362 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
363 | err(1, "Seeking to program headers"); | |
364 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
365 | err(1, "Reading program headers"); | |
366 | ||
dde79789 | 367 | /* Try all the headers: there are usually only three. A read-only one, |
a6bd8e13 | 368 | * a read-write one, and a "note" section which we don't load. */ |
8ca47e00 | 369 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 370 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
371 | if (phdr[i].p_type != PT_LOAD) |
372 | continue; | |
373 | ||
374 | verbose("Section %i: size %i addr %p\n", | |
375 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
376 | ||
6649bb7a | 377 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 378 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 379 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
380 | } |
381 | ||
814a0e5c RR |
382 | /* The entry point is given in the ELF header. */ |
383 | return ehdr->e_entry; | |
8ca47e00 RR |
384 | } |
385 | ||
dde79789 | 386 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
387 | * supposed to jump into it and it will unpack itself. We used to have to |
388 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 389 | * |
5bbf89fc RR |
390 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
391 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
392 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 393 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 394 | { |
43d33b21 | 395 | struct boot_params boot; |
5bbf89fc RR |
396 | int r; |
397 | /* Modern bzImages get loaded at 1M. */ | |
398 | void *p = from_guest_phys(0x100000); | |
399 | ||
400 | /* Go back to the start of the file and read the header. It should be | |
401 | * a Linux boot header (see Documentation/i386/boot.txt) */ | |
402 | lseek(fd, 0, SEEK_SET); | |
43d33b21 | 403 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 404 | |
43d33b21 RR |
405 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
406 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
407 | errx(1, "This doesn't look like a bzImage to me"); |
408 | ||
43d33b21 RR |
409 | /* Skip over the extra sectors of the header. */ |
410 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
411 | |
412 | /* Now read everything into memory. in nice big chunks. */ | |
413 | while ((r = read(fd, p, 65536)) > 0) | |
414 | p += r; | |
415 | ||
43d33b21 RR |
416 | /* Finally, code32_start tells us where to enter the kernel. */ |
417 | return boot.hdr.code32_start; | |
8ca47e00 RR |
418 | } |
419 | ||
dde79789 | 420 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
e1e72965 RR |
421 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
422 | * work, we can load those, too. */ | |
47436aa4 | 423 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
424 | { |
425 | Elf32_Ehdr hdr; | |
426 | ||
dde79789 | 427 | /* Read in the first few bytes. */ |
8ca47e00 RR |
428 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
429 | err(1, "Reading kernel"); | |
430 | ||
dde79789 | 431 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 432 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 433 | return map_elf(fd, &hdr); |
8ca47e00 | 434 | |
a6bd8e13 | 435 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 436 | return load_bzimage(fd); |
8ca47e00 RR |
437 | } |
438 | ||
dde79789 RR |
439 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
440 | * it calls getpagesize() twice: "it's dumb code." | |
441 | * | |
442 | * Kernel guys get really het up about optimization, even when it's not | |
443 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
444 | static inline unsigned long page_align(unsigned long addr) |
445 | { | |
dde79789 | 446 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
447 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
448 | } | |
449 | ||
dde79789 RR |
450 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
451 | * the kernel which the kernel can use to boot from without needing any | |
452 | * drivers. Most distributions now use this as standard: the initrd contains | |
453 | * the code to load the appropriate driver modules for the current machine. | |
454 | * | |
455 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
456 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
457 | static unsigned long load_initrd(const char *name, unsigned long mem) |
458 | { | |
459 | int ifd; | |
460 | struct stat st; | |
461 | unsigned long len; | |
8ca47e00 RR |
462 | |
463 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 464 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
465 | if (fstat(ifd, &st) < 0) |
466 | err(1, "fstat() on initrd '%s'", name); | |
467 | ||
6649bb7a RM |
468 | /* We map the initrd at the top of memory, but mmap wants it to be |
469 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 470 | len = page_align(st.st_size); |
3c6b5bfa | 471 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
472 | /* Once a file is mapped, you can close the file descriptor. It's a |
473 | * little odd, but quite useful. */ | |
8ca47e00 | 474 | close(ifd); |
6649bb7a | 475 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
476 | |
477 | /* We return the initrd size. */ | |
8ca47e00 RR |
478 | return len; |
479 | } | |
480 | ||
a6bd8e13 | 481 | /* Once we know how much memory we have we can construct simple linear page |
47436aa4 | 482 | * tables which set virtual == physical which will get the Guest far enough |
3c6b5bfa | 483 | * into the boot to create its own. |
dde79789 RR |
484 | * |
485 | * We lay them out of the way, just below the initrd (which is why we need to | |
a6bd8e13 | 486 | * know its size here). */ |
8ca47e00 | 487 | static unsigned long setup_pagetables(unsigned long mem, |
47436aa4 | 488 | unsigned long initrd_size) |
8ca47e00 | 489 | { |
511801dc | 490 | unsigned long *pgdir, *linear; |
8ca47e00 | 491 | unsigned int mapped_pages, i, linear_pages; |
511801dc | 492 | unsigned int ptes_per_page = getpagesize()/sizeof(void *); |
8ca47e00 | 493 | |
47436aa4 | 494 | mapped_pages = mem/getpagesize(); |
8ca47e00 | 495 | |
dde79789 | 496 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
8ca47e00 RR |
497 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
498 | ||
dde79789 | 499 | /* We put the toplevel page directory page at the top of memory. */ |
3c6b5bfa | 500 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
dde79789 RR |
501 | |
502 | /* Now we use the next linear_pages pages as pte pages */ | |
8ca47e00 RR |
503 | linear = (void *)pgdir - linear_pages*getpagesize(); |
504 | ||
dde79789 RR |
505 | /* Linear mapping is easy: put every page's address into the mapping in |
506 | * order. PAGE_PRESENT contains the flags Present, Writable and | |
507 | * Executable. */ | |
8ca47e00 RR |
508 | for (i = 0; i < mapped_pages; i++) |
509 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); | |
510 | ||
47436aa4 | 511 | /* The top level points to the linear page table pages above. */ |
8ca47e00 | 512 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
47436aa4 | 513 | pgdir[i/ptes_per_page] |
511801dc | 514 | = ((to_guest_phys(linear) + i*sizeof(void *)) |
3c6b5bfa | 515 | | PAGE_PRESENT); |
8ca47e00 RR |
516 | } |
517 | ||
3c6b5bfa RR |
518 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
519 | mapped_pages, linear_pages, to_guest_phys(linear)); | |
8ca47e00 | 520 | |
dde79789 RR |
521 | /* We return the top level (guest-physical) address: the kernel needs |
522 | * to know where it is. */ | |
3c6b5bfa | 523 | return to_guest_phys(pgdir); |
8ca47e00 | 524 | } |
e1e72965 | 525 | /*:*/ |
8ca47e00 | 526 | |
dde79789 RR |
527 | /* Simple routine to roll all the commandline arguments together with spaces |
528 | * between them. */ | |
8ca47e00 RR |
529 | static void concat(char *dst, char *args[]) |
530 | { | |
531 | unsigned int i, len = 0; | |
532 | ||
533 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
534 | if (i) { |
535 | strcat(dst+len, " "); | |
536 | len++; | |
537 | } | |
8ca47e00 | 538 | strcpy(dst+len, args[i]); |
1ef36fa6 | 539 | len += strlen(args[i]); |
8ca47e00 RR |
540 | } |
541 | /* In case it's empty. */ | |
542 | dst[len] = '\0'; | |
543 | } | |
544 | ||
e1e72965 RR |
545 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
546 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: | |
547 | * the base of Guest "physical" memory, the top physical page to allow, the | |
47436aa4 RR |
548 | * top level pagetable and the entry point for the Guest. */ |
549 | static int tell_kernel(unsigned long pgdir, unsigned long start) | |
8ca47e00 | 550 | { |
511801dc JS |
551 | unsigned long args[] = { LHREQ_INITIALIZE, |
552 | (unsigned long)guest_base, | |
47436aa4 | 553 | guest_limit / getpagesize(), pgdir, start }; |
8ca47e00 RR |
554 | int fd; |
555 | ||
3c6b5bfa RR |
556 | verbose("Guest: %p - %p (%#lx)\n", |
557 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
558 | fd = open_or_die("/dev/lguest", O_RDWR); |
559 | if (write(fd, args, sizeof(args)) < 0) | |
560 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
561 | |
562 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
563 | return fd; |
564 | } | |
dde79789 | 565 | /*:*/ |
8ca47e00 | 566 | |
17cbca2b | 567 | static void add_device_fd(int fd) |
8ca47e00 | 568 | { |
17cbca2b RR |
569 | FD_SET(fd, &devices.infds); |
570 | if (fd > devices.max_infd) | |
571 | devices.max_infd = fd; | |
8ca47e00 RR |
572 | } |
573 | ||
dde79789 RR |
574 | /*L:200 |
575 | * The Waker. | |
576 | * | |
e1e72965 RR |
577 | * With console, block and network devices, we can have lots of input which we |
578 | * need to process. We could try to tell the kernel what file descriptors to | |
579 | * watch, but handing a file descriptor mask through to the kernel is fairly | |
580 | * icky. | |
dde79789 RR |
581 | * |
582 | * Instead, we fork off a process which watches the file descriptors and writes | |
e1e72965 RR |
583 | * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host |
584 | * stop running the Guest. This causes the Launcher to return from the | |
dde79789 RR |
585 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset |
586 | * the LHREQ_BREAK and wake us up again. | |
587 | * | |
588 | * This, of course, is merely a different *kind* of icky. | |
589 | */ | |
17cbca2b | 590 | static void wake_parent(int pipefd, int lguest_fd) |
8ca47e00 | 591 | { |
dde79789 RR |
592 | /* Add the pipe from the Launcher to the fdset in the device_list, so |
593 | * we watch it, too. */ | |
17cbca2b | 594 | add_device_fd(pipefd); |
8ca47e00 RR |
595 | |
596 | for (;;) { | |
17cbca2b | 597 | fd_set rfds = devices.infds; |
511801dc | 598 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8ca47e00 | 599 | |
dde79789 | 600 | /* Wait until input is ready from one of the devices. */ |
17cbca2b | 601 | select(devices.max_infd+1, &rfds, NULL, NULL, NULL); |
dde79789 | 602 | /* Is it a message from the Launcher? */ |
8ca47e00 | 603 | if (FD_ISSET(pipefd, &rfds)) { |
56ae43df | 604 | int fd; |
dde79789 RR |
605 | /* If read() returns 0, it means the Launcher has |
606 | * exited. We silently follow. */ | |
56ae43df | 607 | if (read(pipefd, &fd, sizeof(fd)) == 0) |
8ca47e00 | 608 | exit(0); |
56ae43df | 609 | /* Otherwise it's telling us to change what file |
e1e72965 RR |
610 | * descriptors we're to listen to. Positive means |
611 | * listen to a new one, negative means stop | |
612 | * listening. */ | |
56ae43df RR |
613 | if (fd >= 0) |
614 | FD_SET(fd, &devices.infds); | |
615 | else | |
616 | FD_CLR(-fd - 1, &devices.infds); | |
dde79789 | 617 | } else /* Send LHREQ_BREAK command. */ |
e3283fa0 | 618 | pwrite(lguest_fd, args, sizeof(args), cpu_id); |
8ca47e00 RR |
619 | } |
620 | } | |
621 | ||
dde79789 | 622 | /* This routine just sets up a pipe to the Waker process. */ |
17cbca2b | 623 | static int setup_waker(int lguest_fd) |
8ca47e00 RR |
624 | { |
625 | int pipefd[2], child; | |
626 | ||
e1e72965 | 627 | /* We create a pipe to talk to the Waker, and also so it knows when the |
dde79789 | 628 | * Launcher dies (and closes pipe). */ |
8ca47e00 RR |
629 | pipe(pipefd); |
630 | child = fork(); | |
631 | if (child == -1) | |
632 | err(1, "forking"); | |
633 | ||
634 | if (child == 0) { | |
e1e72965 RR |
635 | /* We are the Waker: close the "writing" end of our copy of the |
636 | * pipe and start waiting for input. */ | |
8ca47e00 | 637 | close(pipefd[1]); |
17cbca2b | 638 | wake_parent(pipefd[0], lguest_fd); |
8ca47e00 | 639 | } |
dde79789 | 640 | /* Close the reading end of our copy of the pipe. */ |
8ca47e00 RR |
641 | close(pipefd[0]); |
642 | ||
dde79789 | 643 | /* Here is the fd used to talk to the waker. */ |
8ca47e00 RR |
644 | return pipefd[1]; |
645 | } | |
646 | ||
e1e72965 | 647 | /* |
dde79789 RR |
648 | * Device Handling. |
649 | * | |
e1e72965 | 650 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 651 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 652 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
653 | * if something funny is going on: |
654 | */ | |
8ca47e00 RR |
655 | static void *_check_pointer(unsigned long addr, unsigned int size, |
656 | unsigned int line) | |
657 | { | |
dde79789 RR |
658 | /* We have to separately check addr and addr+size, because size could |
659 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 660 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 661 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
662 | /* We return a pointer for the caller's convenience, now we know it's |
663 | * safe to use. */ | |
3c6b5bfa | 664 | return from_guest_phys(addr); |
8ca47e00 | 665 | } |
dde79789 | 666 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
667 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
668 | ||
e1e72965 RR |
669 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
670 | * function returns the next descriptor in the chain, or vq->vring.num if we're | |
671 | * at the end. */ | |
17cbca2b RR |
672 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) |
673 | { | |
674 | unsigned int next; | |
675 | ||
676 | /* If this descriptor says it doesn't chain, we're done. */ | |
677 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
678 | return vq->vring.num; | |
679 | ||
680 | /* Check they're not leading us off end of descriptors. */ | |
681 | next = vq->vring.desc[i].next; | |
682 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
683 | wmb(); | |
684 | ||
685 | if (next >= vq->vring.num) | |
686 | errx(1, "Desc next is %u", next); | |
687 | ||
688 | return next; | |
689 | } | |
690 | ||
691 | /* This looks in the virtqueue and for the first available buffer, and converts | |
692 | * it to an iovec for convenient access. Since descriptors consist of some | |
693 | * number of output then some number of input descriptors, it's actually two | |
694 | * iovecs, but we pack them into one and note how many of each there were. | |
695 | * | |
696 | * This function returns the descriptor number found, or vq->vring.num (which | |
697 | * is never a valid descriptor number) if none was found. */ | |
698 | static unsigned get_vq_desc(struct virtqueue *vq, | |
699 | struct iovec iov[], | |
700 | unsigned int *out_num, unsigned int *in_num) | |
701 | { | |
702 | unsigned int i, head; | |
b5111790 | 703 | u16 last_avail; |
17cbca2b RR |
704 | |
705 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 RR |
706 | last_avail = lg_last_avail(vq); |
707 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) | |
17cbca2b | 708 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 709 | last_avail, vq->vring.avail->idx); |
17cbca2b RR |
710 | |
711 | /* If there's nothing new since last we looked, return invalid. */ | |
b5111790 | 712 | if (vq->vring.avail->idx == last_avail) |
17cbca2b RR |
713 | return vq->vring.num; |
714 | ||
715 | /* Grab the next descriptor number they're advertising, and increment | |
716 | * the index we've seen. */ | |
b5111790 RR |
717 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
718 | lg_last_avail(vq)++; | |
17cbca2b RR |
719 | |
720 | /* If their number is silly, that's a fatal mistake. */ | |
721 | if (head >= vq->vring.num) | |
722 | errx(1, "Guest says index %u is available", head); | |
723 | ||
724 | /* When we start there are none of either input nor output. */ | |
725 | *out_num = *in_num = 0; | |
726 | ||
727 | i = head; | |
728 | do { | |
729 | /* Grab the first descriptor, and check it's OK. */ | |
730 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
731 | iov[*out_num + *in_num].iov_base | |
732 | = check_pointer(vq->vring.desc[i].addr, | |
733 | vq->vring.desc[i].len); | |
734 | /* If this is an input descriptor, increment that count. */ | |
735 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
736 | (*in_num)++; | |
737 | else { | |
738 | /* If it's an output descriptor, they're all supposed | |
739 | * to come before any input descriptors. */ | |
740 | if (*in_num) | |
741 | errx(1, "Descriptor has out after in"); | |
742 | (*out_num)++; | |
743 | } | |
744 | ||
745 | /* If we've got too many, that implies a descriptor loop. */ | |
746 | if (*out_num + *in_num > vq->vring.num) | |
747 | errx(1, "Looped descriptor"); | |
748 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 749 | |
20887611 | 750 | vq->inflight++; |
17cbca2b | 751 | return head; |
8ca47e00 RR |
752 | } |
753 | ||
e1e72965 | 754 | /* After we've used one of their buffers, we tell them about it. We'll then |
17cbca2b RR |
755 | * want to send them an interrupt, using trigger_irq(). */ |
756 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 757 | { |
17cbca2b RR |
758 | struct vring_used_elem *used; |
759 | ||
e1e72965 RR |
760 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
761 | * next entry in that used ring. */ | |
17cbca2b RR |
762 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
763 | used->id = head; | |
764 | used->len = len; | |
765 | /* Make sure buffer is written before we update index. */ | |
766 | wmb(); | |
767 | vq->vring.used->idx++; | |
20887611 | 768 | vq->inflight--; |
8ca47e00 RR |
769 | } |
770 | ||
17cbca2b RR |
771 | /* This actually sends the interrupt for this virtqueue */ |
772 | static void trigger_irq(int fd, struct virtqueue *vq) | |
8ca47e00 | 773 | { |
17cbca2b RR |
774 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
775 | ||
20887611 RR |
776 | /* If they don't want an interrupt, don't send one, unless empty. */ |
777 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
778 | && vq->inflight) | |
17cbca2b RR |
779 | return; |
780 | ||
781 | /* Send the Guest an interrupt tell them we used something up. */ | |
8ca47e00 | 782 | if (write(fd, buf, sizeof(buf)) != 0) |
17cbca2b | 783 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
784 | } |
785 | ||
17cbca2b RR |
786 | /* And here's the combo meal deal. Supersize me! */ |
787 | static void add_used_and_trigger(int fd, struct virtqueue *vq, | |
788 | unsigned int head, int len) | |
8ca47e00 | 789 | { |
17cbca2b RR |
790 | add_used(vq, head, len); |
791 | trigger_irq(fd, vq); | |
8ca47e00 RR |
792 | } |
793 | ||
e1e72965 RR |
794 | /* |
795 | * The Console | |
796 | * | |
797 | * Here is the input terminal setting we save, and the routine to restore them | |
798 | * on exit so the user gets their terminal back. */ | |
8ca47e00 RR |
799 | static struct termios orig_term; |
800 | static void restore_term(void) | |
801 | { | |
802 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
803 | } | |
804 | ||
dde79789 | 805 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
806 | struct console_abort |
807 | { | |
dde79789 | 808 | /* How many times have they hit ^C? */ |
8ca47e00 | 809 | int count; |
dde79789 | 810 | /* When did they start? */ |
8ca47e00 RR |
811 | struct timeval start; |
812 | }; | |
813 | ||
dde79789 | 814 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
815 | static bool handle_console_input(int fd, struct device *dev) |
816 | { | |
8ca47e00 | 817 | int len; |
17cbca2b RR |
818 | unsigned int head, in_num, out_num; |
819 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
820 | struct console_abort *abort = dev->priv; |
821 | ||
17cbca2b RR |
822 | /* First we need a console buffer from the Guests's input virtqueue. */ |
823 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
824 | |
825 | /* If they're not ready for input, stop listening to this file | |
826 | * descriptor. We'll start again once they add an input buffer. */ | |
827 | if (head == dev->vq->vring.num) | |
828 | return false; | |
829 | ||
830 | if (out_num) | |
17cbca2b | 831 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 832 | |
dde79789 RR |
833 | /* This is why we convert to iovecs: the readv() call uses them, and so |
834 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 835 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 836 | if (len <= 0) { |
dde79789 | 837 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 838 | * something went terribly wrong. */ |
8ca47e00 | 839 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 840 | /* Put the input terminal back. */ |
17cbca2b | 841 | restore_term(); |
56ae43df RR |
842 | /* Remove callback from input vq, so it doesn't restart us. */ |
843 | dev->vq->handle_output = NULL; | |
844 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 845 | return false; |
8ca47e00 RR |
846 | } |
847 | ||
56ae43df RR |
848 | /* Tell the Guest about the new input. */ |
849 | add_used_and_trigger(fd, dev->vq, head, len); | |
8ca47e00 | 850 | |
dde79789 RR |
851 | /* Three ^C within one second? Exit. |
852 | * | |
853 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
854 | * in a buffer by itself, so they can't be too fast. But we check that | |
855 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
856 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
857 | if (!abort->count++) | |
858 | gettimeofday(&abort->start, NULL); | |
859 | else if (abort->count == 3) { | |
860 | struct timeval now; | |
861 | gettimeofday(&now, NULL); | |
862 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 863 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
864 | /* Close the fd so Waker will know it has to |
865 | * exit. */ | |
8ca47e00 | 866 | close(waker_fd); |
dde79789 RR |
867 | /* Just in case waker is blocked in BREAK, send |
868 | * unbreak now. */ | |
8ca47e00 RR |
869 | write(fd, args, sizeof(args)); |
870 | exit(2); | |
871 | } | |
872 | abort->count = 0; | |
873 | } | |
874 | } else | |
dde79789 | 875 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
876 | abort->count = 0; |
877 | ||
dde79789 | 878 | /* Everything went OK! */ |
8ca47e00 RR |
879 | return true; |
880 | } | |
881 | ||
17cbca2b RR |
882 | /* Handling output for console is simple: we just get all the output buffers |
883 | * and write them to stdout. */ | |
a161883a | 884 | static void handle_console_output(int fd, struct virtqueue *vq, bool timeout) |
8ca47e00 | 885 | { |
17cbca2b RR |
886 | unsigned int head, out, in; |
887 | int len; | |
888 | struct iovec iov[vq->vring.num]; | |
889 | ||
890 | /* Keep getting output buffers from the Guest until we run out. */ | |
891 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
892 | if (in) | |
893 | errx(1, "Input buffers in output queue?"); | |
894 | len = writev(STDOUT_FILENO, iov, out); | |
895 | add_used_and_trigger(fd, vq, head, len); | |
896 | } | |
8ca47e00 RR |
897 | } |
898 | ||
a161883a RR |
899 | static void block_vq(struct virtqueue *vq) |
900 | { | |
901 | struct itimerval itm; | |
902 | ||
903 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
904 | vq->blocked = true; | |
905 | ||
906 | itm.it_interval.tv_sec = 0; | |
907 | itm.it_interval.tv_usec = 0; | |
908 | itm.it_value.tv_sec = 0; | |
aa124984 | 909 | itm.it_value.tv_usec = timeout_usec; |
a161883a RR |
910 | |
911 | setitimer(ITIMER_REAL, &itm, NULL); | |
912 | } | |
913 | ||
e1e72965 RR |
914 | /* |
915 | * The Network | |
916 | * | |
917 | * Handling output for network is also simple: we get all the output buffers | |
17cbca2b | 918 | * and write them (ignoring the first element) to this device's file descriptor |
a6bd8e13 RR |
919 | * (/dev/net/tun). |
920 | */ | |
a161883a | 921 | static void handle_net_output(int fd, struct virtqueue *vq, bool timeout) |
8ca47e00 | 922 | { |
a161883a | 923 | unsigned int head, out, in, num = 0; |
17cbca2b RR |
924 | int len; |
925 | struct iovec iov[vq->vring.num]; | |
aa124984 | 926 | static int last_timeout_num; |
17cbca2b RR |
927 | |
928 | /* Keep getting output buffers from the Guest until we run out. */ | |
929 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
930 | if (in) | |
931 | errx(1, "Input buffers in output queue?"); | |
398f187d RR |
932 | len = writev(vq->dev->fd, iov, out); |
933 | if (len < 0) | |
934 | err(1, "Writing network packet to tun"); | |
17cbca2b | 935 | add_used_and_trigger(fd, vq, head, len); |
a161883a | 936 | num++; |
17cbca2b | 937 | } |
a161883a RR |
938 | |
939 | /* Block further kicks and set up a timer if we saw anything. */ | |
940 | if (!timeout && num) | |
941 | block_vq(vq); | |
aa124984 RR |
942 | |
943 | if (timeout) { | |
944 | if (num < last_timeout_num) | |
945 | timeout_usec += 10; | |
946 | else if (timeout_usec > 1) | |
947 | timeout_usec--; | |
948 | last_timeout_num = num; | |
949 | } | |
8ca47e00 RR |
950 | } |
951 | ||
17cbca2b RR |
952 | /* This is where we handle a packet coming in from the tun device to our |
953 | * Guest. */ | |
8ca47e00 RR |
954 | static bool handle_tun_input(int fd, struct device *dev) |
955 | { | |
17cbca2b | 956 | unsigned int head, in_num, out_num; |
8ca47e00 | 957 | int len; |
17cbca2b | 958 | struct iovec iov[dev->vq->vring.num]; |
8ca47e00 | 959 | |
17cbca2b RR |
960 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
961 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
962 | if (head == dev->vq->vring.num) { | |
dde79789 | 963 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
964 | * early, the Guest won't be ready yet. Wait until the device |
965 | * status says it's ready. */ | |
966 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
5dae785a RR |
967 | |
968 | /* Now tell it we want to know if new things appear. */ | |
969 | dev->vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
970 | wmb(); | |
971 | ||
56ae43df RR |
972 | /* We'll turn this back on if input buffers are registered. */ |
973 | return false; | |
17cbca2b RR |
974 | } else if (out_num) |
975 | errx(1, "Output buffers in network recv queue?"); | |
976 | ||
dde79789 | 977 | /* Read the packet from the device directly into the Guest's buffer. */ |
398f187d | 978 | len = readv(dev->fd, iov, in_num); |
8ca47e00 RR |
979 | if (len <= 0) |
980 | err(1, "reading network"); | |
dde79789 | 981 | |
56ae43df | 982 | /* Tell the Guest about the new packet. */ |
398f187d | 983 | add_used_and_trigger(fd, dev->vq, head, len); |
17cbca2b | 984 | |
8ca47e00 | 985 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
986 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
987 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
988 | ||
dde79789 | 989 | /* All good. */ |
8ca47e00 RR |
990 | return true; |
991 | } | |
992 | ||
e1e72965 RR |
993 | /*L:215 This is the callback attached to the network and console input |
994 | * virtqueues: it ensures we try again, in case we stopped console or net | |
56ae43df | 995 | * delivery because Guest didn't have any buffers. */ |
a161883a | 996 | static void enable_fd(int fd, struct virtqueue *vq, bool timeout) |
56ae43df RR |
997 | { |
998 | add_device_fd(vq->dev->fd); | |
999 | /* Tell waker to listen to it again */ | |
1000 | write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd)); | |
1001 | } | |
1002 | ||
a161883a | 1003 | static void net_enable_fd(int fd, struct virtqueue *vq, bool timeout) |
5dae785a RR |
1004 | { |
1005 | /* We don't need to know again when Guest refills receive buffer. */ | |
1006 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
a161883a | 1007 | enable_fd(fd, vq, timeout); |
5dae785a RR |
1008 | } |
1009 | ||
a007a751 RR |
1010 | /* When the Guest tells us they updated the status field, we handle it. */ |
1011 | static void update_device_status(struct device *dev) | |
6e5aa7ef RR |
1012 | { |
1013 | struct virtqueue *vq; | |
1014 | ||
a007a751 RR |
1015 | /* This is a reset. */ |
1016 | if (dev->desc->status == 0) { | |
1017 | verbose("Resetting device %s\n", dev->name); | |
6e5aa7ef | 1018 | |
a007a751 RR |
1019 | /* Clear any features they've acked. */ |
1020 | memset(get_feature_bits(dev) + dev->desc->feature_len, 0, | |
1021 | dev->desc->feature_len); | |
6e5aa7ef | 1022 | |
a007a751 RR |
1023 | /* Zero out the virtqueues. */ |
1024 | for (vq = dev->vq; vq; vq = vq->next) { | |
1025 | memset(vq->vring.desc, 0, | |
1026 | vring_size(vq->config.num, getpagesize())); | |
b5111790 | 1027 | lg_last_avail(vq) = 0; |
a007a751 RR |
1028 | } |
1029 | } else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
1030 | warnx("Device %s configuration FAILED", dev->name); | |
1031 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { | |
1032 | unsigned int i; | |
1033 | ||
1034 | verbose("Device %s OK: offered", dev->name); | |
1035 | for (i = 0; i < dev->desc->feature_len; i++) | |
32c68e5c | 1036 | verbose(" %02x", get_feature_bits(dev)[i]); |
a007a751 RR |
1037 | verbose(", accepted"); |
1038 | for (i = 0; i < dev->desc->feature_len; i++) | |
32c68e5c | 1039 | verbose(" %02x", get_feature_bits(dev) |
a007a751 RR |
1040 | [dev->desc->feature_len+i]); |
1041 | ||
1042 | if (dev->ready) | |
1043 | dev->ready(dev); | |
6e5aa7ef RR |
1044 | } |
1045 | } | |
1046 | ||
17cbca2b RR |
1047 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
1048 | static void handle_output(int fd, unsigned long addr) | |
8ca47e00 RR |
1049 | { |
1050 | struct device *i; | |
17cbca2b RR |
1051 | struct virtqueue *vq; |
1052 | ||
6e5aa7ef | 1053 | /* Check each device and virtqueue. */ |
17cbca2b | 1054 | for (i = devices.dev; i; i = i->next) { |
a007a751 | 1055 | /* Notifications to device descriptors update device status. */ |
6e5aa7ef | 1056 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 1057 | update_device_status(i); |
6e5aa7ef RR |
1058 | return; |
1059 | } | |
1060 | ||
1061 | /* Notifications to virtqueues mean output has occurred. */ | |
17cbca2b | 1062 | for (vq = i->vq; vq; vq = vq->next) { |
6e5aa7ef RR |
1063 | if (vq->config.pfn != addr/getpagesize()) |
1064 | continue; | |
1065 | ||
1066 | /* Guest should acknowledge (and set features!) before | |
1067 | * using the device. */ | |
1068 | if (i->desc->status == 0) { | |
1069 | warnx("%s gave early output", i->name); | |
17cbca2b RR |
1070 | return; |
1071 | } | |
6e5aa7ef RR |
1072 | |
1073 | if (strcmp(vq->dev->name, "console") != 0) | |
1074 | verbose("Output to %s\n", vq->dev->name); | |
1075 | if (vq->handle_output) | |
a161883a | 1076 | vq->handle_output(fd, vq, false); |
6e5aa7ef | 1077 | return; |
8ca47e00 RR |
1078 | } |
1079 | } | |
dde79789 | 1080 | |
17cbca2b RR |
1081 | /* Early console write is done using notify on a nul-terminated string |
1082 | * in Guest memory. */ | |
1083 | if (addr >= guest_limit) | |
1084 | errx(1, "Bad NOTIFY %#lx", addr); | |
1085 | ||
1086 | write(STDOUT_FILENO, from_guest_phys(addr), | |
1087 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
1088 | } |
1089 | ||
a161883a RR |
1090 | static void handle_timeout(int fd) |
1091 | { | |
1092 | char buf[32]; | |
1093 | struct device *i; | |
1094 | struct virtqueue *vq; | |
1095 | ||
1096 | /* Clear the pipe */ | |
1097 | read(timeoutpipe[0], buf, sizeof(buf)); | |
1098 | ||
1099 | /* Check each device and virtqueue: flush blocked ones. */ | |
1100 | for (i = devices.dev; i; i = i->next) { | |
1101 | for (vq = i->vq; vq; vq = vq->next) { | |
1102 | if (!vq->blocked) | |
1103 | continue; | |
1104 | ||
1105 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
1106 | vq->blocked = false; | |
1107 | if (vq->handle_output) | |
1108 | vq->handle_output(fd, vq, true); | |
1109 | } | |
1110 | } | |
1111 | } | |
1112 | ||
e1e72965 | 1113 | /* This is called when the Waker wakes us up: check for incoming file |
dde79789 | 1114 | * descriptors. */ |
17cbca2b | 1115 | static void handle_input(int fd) |
8ca47e00 | 1116 | { |
dde79789 | 1117 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
1118 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
1119 | ||
1120 | for (;;) { | |
1121 | struct device *i; | |
17cbca2b | 1122 | fd_set fds = devices.infds; |
a161883a | 1123 | int num; |
8ca47e00 | 1124 | |
a161883a RR |
1125 | num = select(devices.max_infd+1, &fds, NULL, NULL, &poll); |
1126 | /* Could get interrupted */ | |
1127 | if (num < 0) | |
1128 | continue; | |
dde79789 | 1129 | /* If nothing is ready, we're done. */ |
a161883a | 1130 | if (num == 0) |
8ca47e00 RR |
1131 | break; |
1132 | ||
a6bd8e13 RR |
1133 | /* Otherwise, call the device(s) which have readable file |
1134 | * descriptors and a method of handling them. */ | |
17cbca2b | 1135 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 1136 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56ae43df RR |
1137 | int dev_fd; |
1138 | if (i->handle_input(fd, i)) | |
1139 | continue; | |
1140 | ||
dde79789 | 1141 | /* If handle_input() returns false, it means we |
56ae43df RR |
1142 | * should no longer service it. Networking and |
1143 | * console do this when there's no input | |
1144 | * buffers to deliver into. Console also uses | |
a6bd8e13 | 1145 | * it when it discovers that stdin is closed. */ |
56ae43df RR |
1146 | FD_CLR(i->fd, &devices.infds); |
1147 | /* Tell waker to ignore it too, by sending a | |
1148 | * negative fd number (-1, since 0 is a valid | |
1149 | * FD number). */ | |
1150 | dev_fd = -i->fd - 1; | |
1151 | write(waker_fd, &dev_fd, sizeof(dev_fd)); | |
8ca47e00 RR |
1152 | } |
1153 | } | |
a161883a RR |
1154 | |
1155 | /* Is this the timeout fd? */ | |
1156 | if (FD_ISSET(timeoutpipe[0], &fds)) | |
1157 | handle_timeout(fd); | |
8ca47e00 RR |
1158 | } |
1159 | } | |
1160 | ||
dde79789 RR |
1161 | /*L:190 |
1162 | * Device Setup | |
1163 | * | |
1164 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1165 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
1166 | * routines to allocate and manage them. |
1167 | */ | |
8ca47e00 | 1168 | |
a586d4f6 RR |
1169 | /* The layout of the device page is a "struct lguest_device_desc" followed by a |
1170 | * number of virtqueue descriptors, then two sets of feature bits, then an | |
1171 | * array of configuration bytes. This routine returns the configuration | |
1172 | * pointer. */ | |
1173 | static u8 *device_config(const struct device *dev) | |
1174 | { | |
1175 | return (void *)(dev->desc + 1) | |
1176 | + dev->desc->num_vq * sizeof(struct lguest_vqconfig) | |
1177 | + dev->desc->feature_len * 2; | |
17cbca2b RR |
1178 | } |
1179 | ||
a586d4f6 RR |
1180 | /* This routine allocates a new "struct lguest_device_desc" from descriptor |
1181 | * table page just above the Guest's normal memory. It returns a pointer to | |
1182 | * that descriptor. */ | |
1183 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
17cbca2b | 1184 | { |
a586d4f6 RR |
1185 | struct lguest_device_desc d = { .type = type }; |
1186 | void *p; | |
17cbca2b | 1187 | |
a586d4f6 RR |
1188 | /* Figure out where the next device config is, based on the last one. */ |
1189 | if (devices.lastdev) | |
1190 | p = device_config(devices.lastdev) | |
1191 | + devices.lastdev->desc->config_len; | |
1192 | else | |
1193 | p = devices.descpage; | |
17cbca2b | 1194 | |
a586d4f6 RR |
1195 | /* We only have one page for all the descriptors. */ |
1196 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1197 | errx(1, "Too many devices"); | |
17cbca2b | 1198 | |
a586d4f6 RR |
1199 | /* p might not be aligned, so we memcpy in. */ |
1200 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1201 | } |
1202 | ||
a586d4f6 RR |
1203 | /* Each device descriptor is followed by the description of its virtqueues. We |
1204 | * specify how many descriptors the virtqueue is to have. */ | |
17cbca2b | 1205 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
a161883a | 1206 | void (*handle_output)(int, struct virtqueue *, bool)) |
17cbca2b RR |
1207 | { |
1208 | unsigned int pages; | |
1209 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1210 | void *p; | |
1211 | ||
a6bd8e13 | 1212 | /* First we need some memory for this virtqueue. */ |
42b36cc0 RR |
1213 | pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1) |
1214 | / getpagesize(); | |
17cbca2b RR |
1215 | p = get_pages(pages); |
1216 | ||
d1c856e0 RR |
1217 | /* Initialize the virtqueue */ |
1218 | vq->next = NULL; | |
1219 | vq->last_avail_idx = 0; | |
1220 | vq->dev = dev; | |
20887611 | 1221 | vq->inflight = 0; |
a161883a | 1222 | vq->blocked = false; |
d1c856e0 | 1223 | |
17cbca2b RR |
1224 | /* Initialize the configuration. */ |
1225 | vq->config.num = num_descs; | |
1226 | vq->config.irq = devices.next_irq++; | |
1227 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1228 | ||
1229 | /* Initialize the vring. */ | |
42b36cc0 | 1230 | vring_init(&vq->vring, num_descs, p, getpagesize()); |
17cbca2b | 1231 | |
a586d4f6 RR |
1232 | /* Append virtqueue to this device's descriptor. We use |
1233 | * device_config() to get the end of the device's current virtqueues; | |
1234 | * we check that we haven't added any config or feature information | |
1235 | * yet, otherwise we'd be overwriting them. */ | |
1236 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); | |
1237 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
1238 | dev->desc->num_vq++; | |
1239 | ||
1240 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b RR |
1241 | |
1242 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1243 | * second. */ | |
1244 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1245 | *i = vq; | |
1246 | ||
e1e72965 RR |
1247 | /* Set the routine to call when the Guest does something to this |
1248 | * virtqueue. */ | |
17cbca2b | 1249 | vq->handle_output = handle_output; |
e1e72965 | 1250 | |
426e3e0a RR |
1251 | /* As an optimization, set the advisory "Don't Notify Me" flag if we |
1252 | * don't have a handler */ | |
17cbca2b RR |
1253 | if (!handle_output) |
1254 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1255 | } |
1256 | ||
6e5aa7ef | 1257 | /* The first half of the feature bitmask is for us to advertise features. The |
a6bd8e13 | 1258 | * second half is for the Guest to accept features. */ |
a586d4f6 RR |
1259 | static void add_feature(struct device *dev, unsigned bit) |
1260 | { | |
6e5aa7ef | 1261 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1262 | |
1263 | /* We can't extend the feature bits once we've added config bytes */ | |
1264 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1265 | assert(dev->desc->config_len == 0); | |
1266 | dev->desc->feature_len = (bit / CHAR_BIT) + 1; | |
1267 | } | |
1268 | ||
a586d4f6 RR |
1269 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1270 | } | |
1271 | ||
1272 | /* This routine sets the configuration fields for an existing device's | |
1273 | * descriptor. It only works for the last device, but that's OK because that's | |
1274 | * how we use it. */ | |
1275 | static void set_config(struct device *dev, unsigned len, const void *conf) | |
1276 | { | |
1277 | /* Check we haven't overflowed our single page. */ | |
1278 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1279 | errx(1, "Too many devices"); | |
1280 | ||
1281 | /* Copy in the config information, and store the length. */ | |
1282 | memcpy(device_config(dev), conf, len); | |
1283 | dev->desc->config_len = len; | |
1284 | } | |
1285 | ||
17cbca2b | 1286 | /* This routine does all the creation and setup of a new device, including |
a6bd8e13 RR |
1287 | * calling new_dev_desc() to allocate the descriptor and device memory. |
1288 | * | |
1289 | * See what I mean about userspace being boring? */ | |
17cbca2b RR |
1290 | static struct device *new_device(const char *name, u16 type, int fd, |
1291 | bool (*handle_input)(int, struct device *)) | |
8ca47e00 RR |
1292 | { |
1293 | struct device *dev = malloc(sizeof(*dev)); | |
1294 | ||
dde79789 | 1295 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1296 | dev->fd = fd; |
dde79789 RR |
1297 | /* If we have an input handler for this file descriptor, then we add it |
1298 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1299 | if (handle_input) |
17cbca2b RR |
1300 | add_device_fd(dev->fd); |
1301 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1302 | dev->handle_input = handle_input; |
17cbca2b | 1303 | dev->name = name; |
d1c856e0 | 1304 | dev->vq = NULL; |
a007a751 | 1305 | dev->ready = NULL; |
a586d4f6 RR |
1306 | |
1307 | /* Append to device list. Prepending to a single-linked list is | |
1308 | * easier, but the user expects the devices to be arranged on the bus | |
1309 | * in command-line order. The first network device on the command line | |
1310 | * is eth0, the first block device /dev/vda, etc. */ | |
1311 | if (devices.lastdev) | |
1312 | devices.lastdev->next = dev; | |
1313 | else | |
1314 | devices.dev = dev; | |
1315 | devices.lastdev = dev; | |
1316 | ||
8ca47e00 RR |
1317 | return dev; |
1318 | } | |
1319 | ||
dde79789 RR |
1320 | /* Our first setup routine is the console. It's a fairly simple device, but |
1321 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1322 | static void setup_console(void) |
8ca47e00 RR |
1323 | { |
1324 | struct device *dev; | |
1325 | ||
dde79789 | 1326 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1327 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1328 | struct termios term = orig_term; | |
dde79789 RR |
1329 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1330 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1331 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1332 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1333 | /* If we exit gracefully, the original settings will be |
1334 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1335 | atexit(restore_term); |
1336 | } | |
1337 | ||
17cbca2b RR |
1338 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1339 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1340 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1341 | dev->priv = malloc(sizeof(struct console_abort)); |
1342 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1343 | |
56ae43df RR |
1344 | /* The console needs two virtqueues: the input then the output. When |
1345 | * they put something the input queue, we make sure we're listening to | |
1346 | * stdin. When they put something in the output queue, we write it to | |
e1e72965 | 1347 | * stdout. */ |
56ae43df | 1348 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
17cbca2b RR |
1349 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1350 | ||
1351 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1352 | } |
17cbca2b | 1353 | /*:*/ |
8ca47e00 | 1354 | |
a161883a RR |
1355 | static void timeout_alarm(int sig) |
1356 | { | |
1357 | write(timeoutpipe[1], "", 1); | |
1358 | } | |
1359 | ||
1360 | static void setup_timeout(void) | |
1361 | { | |
1362 | if (pipe(timeoutpipe) != 0) | |
1363 | err(1, "Creating timeout pipe"); | |
1364 | ||
1365 | if (fcntl(timeoutpipe[1], F_SETFL, | |
1366 | fcntl(timeoutpipe[1], F_GETFL) | O_NONBLOCK) != 0) | |
1367 | err(1, "Making timeout pipe nonblocking"); | |
1368 | ||
1369 | add_device_fd(timeoutpipe[0]); | |
1370 | signal(SIGALRM, timeout_alarm); | |
1371 | } | |
1372 | ||
17cbca2b RR |
1373 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1374 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1375 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1376 | * |
17cbca2b RR |
1377 | * More sopisticated is to use one of the tools developed for project like UML |
1378 | * to do networking. | |
dde79789 | 1379 | * |
17cbca2b RR |
1380 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1381 | * completely generic ("here's my vring, attach to your vring") and would work | |
1382 | * for any traffic. Of course, namespace and permissions issues need to be | |
1383 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1384 | * multiple inter-guest channels behind one interface, although it would | |
1385 | * require some manner of hotplugging new virtio channels. | |
1386 | * | |
1387 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1388 | |
1389 | static u32 str2ip(const char *ipaddr) | |
1390 | { | |
dec6a2be | 1391 | unsigned int b[4]; |
8ca47e00 | 1392 | |
dec6a2be MM |
1393 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1394 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1395 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1396 | } | |
1397 | ||
1398 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1399 | { | |
1400 | unsigned int m[6]; | |
1401 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1402 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1403 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1404 | mac[0] = m[0]; | |
1405 | mac[1] = m[1]; | |
1406 | mac[2] = m[2]; | |
1407 | mac[3] = m[3]; | |
1408 | mac[4] = m[4]; | |
1409 | mac[5] = m[5]; | |
8ca47e00 RR |
1410 | } |
1411 | ||
dde79789 RR |
1412 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1413 | * network device to the bridge device specified by the command line. | |
1414 | * | |
1415 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1416 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1417 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1418 | { | |
1419 | int ifidx; | |
1420 | struct ifreq ifr; | |
1421 | ||
1422 | if (!*br_name) | |
1423 | errx(1, "must specify bridge name"); | |
1424 | ||
1425 | ifidx = if_nametoindex(if_name); | |
1426 | if (!ifidx) | |
1427 | errx(1, "interface %s does not exist!", if_name); | |
1428 | ||
1429 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1430 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1431 | ifr.ifr_ifindex = ifidx; |
1432 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1433 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1434 | } | |
1435 | ||
dde79789 RR |
1436 | /* This sets up the Host end of the network device with an IP address, brings |
1437 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1438 | * pointer. */ |
dec6a2be | 1439 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1440 | { |
1441 | struct ifreq ifr; | |
1442 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1443 | ||
1444 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1445 | strcpy(ifr.ifr_name, tapif); |
1446 | ||
1447 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
8ca47e00 RR |
1448 | sin->sin_family = AF_INET; |
1449 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1450 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
dec6a2be | 1451 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1452 | ifr.ifr_flags = IFF_UP; |
1453 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1454 | err(1, "Bringing interface %s up", tapif); |
1455 | } | |
1456 | ||
1457 | static void get_mac(int fd, const char *tapif, unsigned char hwaddr[6]) | |
1458 | { | |
1459 | struct ifreq ifr; | |
1460 | ||
1461 | memset(&ifr, 0, sizeof(ifr)); | |
1462 | strcpy(ifr.ifr_name, tapif); | |
8ca47e00 | 1463 | |
dde79789 RR |
1464 | /* SIOC stands for Socket I/O Control. G means Get (vs S for Set |
1465 | * above). IF means Interface, and HWADDR is hardware address. | |
1466 | * Simple! */ | |
8ca47e00 | 1467 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0) |
dec6a2be | 1468 | err(1, "getting hw address for %s", tapif); |
8ca47e00 RR |
1469 | memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); |
1470 | } | |
1471 | ||
dec6a2be | 1472 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1473 | { |
8ca47e00 | 1474 | struct ifreq ifr; |
dec6a2be MM |
1475 | int netfd; |
1476 | ||
1477 | /* Start with this zeroed. Messy but sure. */ | |
1478 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1479 | |
dde79789 RR |
1480 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1481 | * tap device is like a tun device, only somehow different. To tell | |
1482 | * the truth, I completely blundered my way through this code, but it | |
1483 | * works now! */ | |
8ca47e00 | 1484 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1485 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1486 | strcpy(ifr.ifr_name, "tap%d"); |
1487 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1488 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1489 | |
398f187d RR |
1490 | if (ioctl(netfd, TUNSETOFFLOAD, |
1491 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1492 | err(1, "Could not set features for tun device"); | |
1493 | ||
dde79789 RR |
1494 | /* We don't need checksums calculated for packets coming in this |
1495 | * device: trust us! */ | |
8ca47e00 RR |
1496 | ioctl(netfd, TUNSETNOCSUM, 1); |
1497 | ||
dec6a2be MM |
1498 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1499 | return netfd; | |
1500 | } | |
1501 | ||
1502 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or | |
1503 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1504 | * packets into the Host as if they came in from a normal network card. We | |
1505 | * just shunt packets between the Guest and the tun device. */ | |
1506 | static void setup_tun_net(char *arg) | |
1507 | { | |
1508 | struct device *dev; | |
1509 | int netfd, ipfd; | |
1510 | u32 ip = INADDR_ANY; | |
1511 | bool bridging = false; | |
1512 | char tapif[IFNAMSIZ], *p; | |
1513 | struct virtio_net_config conf; | |
1514 | ||
1515 | netfd = get_tun_device(tapif); | |
1516 | ||
17cbca2b RR |
1517 | /* First we create a new network device. */ |
1518 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1519 | |
56ae43df RR |
1520 | /* Network devices need a receive and a send queue, just like |
1521 | * console. */ | |
5dae785a | 1522 | add_virtqueue(dev, VIRTQUEUE_NUM, net_enable_fd); |
17cbca2b | 1523 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1524 | |
dde79789 RR |
1525 | /* We need a socket to perform the magic network ioctls to bring up the |
1526 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1527 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1528 | if (ipfd < 0) | |
1529 | err(1, "opening IP socket"); | |
1530 | ||
dde79789 | 1531 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1532 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1533 | arg += strlen(BRIDGE_PFX); |
1534 | bridging = true; | |
1535 | } | |
1536 | ||
1537 | /* A mac address may follow the bridge name or IP address */ | |
1538 | p = strchr(arg, ':'); | |
1539 | if (p) { | |
1540 | str2mac(p+1, conf.mac); | |
1541 | *p = '\0'; | |
1542 | } else { | |
1543 | p = arg + strlen(arg); | |
1544 | /* None supplied; query the randomly assigned mac. */ | |
1545 | get_mac(ipfd, tapif, conf.mac); | |
1546 | } | |
1547 | ||
1548 | /* arg is now either an IP address or a bridge name */ | |
1549 | if (bridging) | |
1550 | add_to_bridge(ipfd, tapif, arg); | |
1551 | else | |
8ca47e00 RR |
1552 | ip = str2ip(arg); |
1553 | ||
dec6a2be MM |
1554 | /* Set up the tun device. */ |
1555 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1556 | |
17cbca2b | 1557 | /* Tell Guest what MAC address to use. */ |
a586d4f6 | 1558 | add_feature(dev, VIRTIO_NET_F_MAC); |
20887611 | 1559 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1560 | /* Expect Guest to handle everything except UFO */ |
1561 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1562 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
1563 | add_feature(dev, VIRTIO_NET_F_MAC); | |
1564 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); | |
1565 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1566 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1567 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1568 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1569 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
a586d4f6 | 1570 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1571 | |
a586d4f6 | 1572 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1573 | close(ipfd); |
1574 | ||
dec6a2be MM |
1575 | devices.device_num++; |
1576 | ||
1577 | if (bridging) | |
1578 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1579 | devices.device_num, tapif, arg); | |
1580 | else | |
1581 | verbose("device %u: tun %s: %s\n", | |
1582 | devices.device_num, tapif, arg); | |
8ca47e00 | 1583 | } |
17cbca2b | 1584 | |
e1e72965 RR |
1585 | /* Our block (disk) device should be really simple: the Guest asks for a block |
1586 | * number and we read or write that position in the file. Unfortunately, that | |
1587 | * was amazingly slow: the Guest waits until the read is finished before | |
1588 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1589 | * |
e1e72965 RR |
1590 | * We could use async I/O, except it's reputed to suck so hard that characters |
1591 | * actually go missing from your code when you try to use it. | |
17cbca2b RR |
1592 | * |
1593 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1594 | ||
e1e72965 | 1595 | /* This hangs off device->priv. */ |
17cbca2b RR |
1596 | struct vblk_info |
1597 | { | |
1598 | /* The size of the file. */ | |
1599 | off64_t len; | |
1600 | ||
1601 | /* The file descriptor for the file. */ | |
1602 | int fd; | |
1603 | ||
1604 | /* IO thread listens on this file descriptor [0]. */ | |
1605 | int workpipe[2]; | |
1606 | ||
1607 | /* IO thread writes to this file descriptor to mark it done, then | |
1608 | * Launcher triggers interrupt to Guest. */ | |
1609 | int done_fd; | |
1610 | }; | |
1611 | ||
e1e72965 RR |
1612 | /*L:210 |
1613 | * The Disk | |
1614 | * | |
1615 | * Remember that the block device is handled by a separate I/O thread. We head | |
1616 | * straight into the core of that thread here: | |
1617 | */ | |
17cbca2b RR |
1618 | static bool service_io(struct device *dev) |
1619 | { | |
1620 | struct vblk_info *vblk = dev->priv; | |
1621 | unsigned int head, out_num, in_num, wlen; | |
1622 | int ret; | |
cb38fa23 | 1623 | u8 *in; |
17cbca2b RR |
1624 | struct virtio_blk_outhdr *out; |
1625 | struct iovec iov[dev->vq->vring.num]; | |
1626 | off64_t off; | |
1627 | ||
e1e72965 | 1628 | /* See if there's a request waiting. If not, nothing to do. */ |
17cbca2b RR |
1629 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
1630 | if (head == dev->vq->vring.num) | |
1631 | return false; | |
1632 | ||
e1e72965 RR |
1633 | /* Every block request should contain at least one output buffer |
1634 | * (detailing the location on disk and the type of request) and one | |
1635 | * input buffer (to hold the result). */ | |
17cbca2b RR |
1636 | if (out_num == 0 || in_num == 0) |
1637 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1638 | head, out_num, in_num); | |
1639 | ||
1640 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1641 | in = convert(&iov[out_num+in_num-1], u8); |
17cbca2b RR |
1642 | off = out->sector * 512; |
1643 | ||
e1e72965 RR |
1644 | /* The block device implements "barriers", where the Guest indicates |
1645 | * that it wants all previous writes to occur before this write. We | |
1646 | * don't have a way of asking our kernel to do a barrier, so we just | |
1647 | * synchronize all the data in the file. Pretty poor, no? */ | |
17cbca2b RR |
1648 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1649 | fdatasync(vblk->fd); | |
1650 | ||
e1e72965 RR |
1651 | /* In general the virtio block driver is allowed to try SCSI commands. |
1652 | * It'd be nice if we supported eject, for example, but we don't. */ | |
17cbca2b RR |
1653 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1654 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1655 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1656 | wlen = sizeof(*in); |
17cbca2b RR |
1657 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
1658 | /* Write */ | |
1659 | ||
1660 | /* Move to the right location in the block file. This can fail | |
1661 | * if they try to write past end. */ | |
1662 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1663 | err(1, "Bad seek to sector %llu", out->sector); | |
1664 | ||
1665 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1666 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1667 | ||
1668 | /* Grr... Now we know how long the descriptor they sent was, we | |
1669 | * make sure they didn't try to write over the end of the block | |
1670 | * file (possibly extending it). */ | |
1671 | if (ret > 0 && off + ret > vblk->len) { | |
1672 | /* Trim it back to the correct length */ | |
1673 | ftruncate64(vblk->fd, vblk->len); | |
1674 | /* Die, bad Guest, die. */ | |
1675 | errx(1, "Write past end %llu+%u", off, ret); | |
1676 | } | |
1200e646 | 1677 | wlen = sizeof(*in); |
cb38fa23 | 1678 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b RR |
1679 | } else { |
1680 | /* Read */ | |
1681 | ||
1682 | /* Move to the right location in the block file. This can fail | |
1683 | * if they try to read past end. */ | |
1684 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1685 | err(1, "Bad seek to sector %llu", out->sector); | |
1686 | ||
1687 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1688 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1689 | if (ret >= 0) { | |
1200e646 | 1690 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1691 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1692 | } else { |
1200e646 | 1693 | wlen = sizeof(*in); |
cb38fa23 | 1694 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1695 | } |
1696 | } | |
1697 | ||
1698 | /* We can't trigger an IRQ, because we're not the Launcher. It does | |
1699 | * that when we tell it we're done. */ | |
1700 | add_used(dev->vq, head, wlen); | |
1701 | return true; | |
1702 | } | |
1703 | ||
1704 | /* This is the thread which actually services the I/O. */ | |
1705 | static int io_thread(void *_dev) | |
1706 | { | |
1707 | struct device *dev = _dev; | |
1708 | struct vblk_info *vblk = dev->priv; | |
1709 | char c; | |
1710 | ||
1711 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1712 | close(vblk->workpipe[1]); | |
1713 | /* Close the other side of the done_fd pipe. */ | |
1714 | close(dev->fd); | |
1715 | ||
1716 | /* When this read fails, it means Launcher died, so we follow. */ | |
1717 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
e1e72965 | 1718 | /* We acknowledge each request immediately to reduce latency, |
17cbca2b | 1719 | * rather than waiting until we've done them all. I haven't |
a6bd8e13 RR |
1720 | * measured to see if it makes any difference. |
1721 | * | |
1722 | * That would be an interesting test, wouldn't it? You could | |
1723 | * also try having more than one I/O thread. */ | |
17cbca2b RR |
1724 | while (service_io(dev)) |
1725 | write(vblk->done_fd, &c, 1); | |
1726 | } | |
1727 | return 0; | |
1728 | } | |
1729 | ||
e1e72965 | 1730 | /* Now we've seen the I/O thread, we return to the Launcher to see what happens |
a6bd8e13 | 1731 | * when that thread tells us it's completed some I/O. */ |
17cbca2b RR |
1732 | static bool handle_io_finish(int fd, struct device *dev) |
1733 | { | |
1734 | char c; | |
1735 | ||
e1e72965 RR |
1736 | /* If the I/O thread died, presumably it printed the error, so we |
1737 | * simply exit. */ | |
17cbca2b RR |
1738 | if (read(dev->fd, &c, 1) != 1) |
1739 | exit(1); | |
1740 | ||
1741 | /* It did some work, so trigger the irq. */ | |
1742 | trigger_irq(fd, dev->vq); | |
1743 | return true; | |
1744 | } | |
1745 | ||
e1e72965 | 1746 | /* When the Guest submits some I/O, we just need to wake the I/O thread. */ |
a161883a | 1747 | static void handle_virtblk_output(int fd, struct virtqueue *vq, bool timeout) |
17cbca2b RR |
1748 | { |
1749 | struct vblk_info *vblk = vq->dev->priv; | |
1750 | char c = 0; | |
1751 | ||
1752 | /* Wake up I/O thread and tell it to go to work! */ | |
1753 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1754 | /* Presumably it indicated why it died. */ | |
1755 | exit(1); | |
1756 | } | |
1757 | ||
e1e72965 | 1758 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1759 | static void setup_block_file(const char *filename) |
1760 | { | |
1761 | int p[2]; | |
1762 | struct device *dev; | |
1763 | struct vblk_info *vblk; | |
1764 | void *stack; | |
a586d4f6 | 1765 | struct virtio_blk_config conf; |
17cbca2b RR |
1766 | |
1767 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1768 | pipe(p); | |
1769 | ||
1770 | /* The device responds to return from I/O thread. */ | |
1771 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1772 | ||
e1e72965 | 1773 | /* The device has one virtqueue, where the Guest places requests. */ |
17cbca2b RR |
1774 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); |
1775 | ||
1776 | /* Allocate the room for our own bookkeeping */ | |
1777 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1778 | ||
1779 | /* First we open the file and store the length. */ | |
1780 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1781 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1782 | ||
a586d4f6 RR |
1783 | /* We support barriers. */ |
1784 | add_feature(dev, VIRTIO_BLK_F_BARRIER); | |
1785 | ||
17cbca2b | 1786 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1787 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b RR |
1788 | |
1789 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1790 | * for the in and out elements. */ | |
a586d4f6 RR |
1791 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1792 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1793 | ||
1794 | set_config(dev, sizeof(conf), &conf); | |
17cbca2b RR |
1795 | |
1796 | /* The I/O thread writes to this end of the pipe when done. */ | |
1797 | vblk->done_fd = p[1]; | |
1798 | ||
e1e72965 RR |
1799 | /* This is the second pipe, which is how we tell the I/O thread about |
1800 | * more work. */ | |
17cbca2b RR |
1801 | pipe(vblk->workpipe); |
1802 | ||
a6bd8e13 RR |
1803 | /* Create stack for thread and run it. Since stack grows upwards, we |
1804 | * point the stack pointer to the end of this region. */ | |
17cbca2b | 1805 | stack = malloc(32768); |
ec04b13f BR |
1806 | /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from |
1807 | * becoming a zombie. */ | |
a6bd8e13 | 1808 | if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1) |
17cbca2b RR |
1809 | err(1, "Creating clone"); |
1810 | ||
1811 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1812 | close(vblk->done_fd); | |
1813 | close(vblk->workpipe[0]); | |
1814 | ||
1815 | verbose("device %u: virtblock %llu sectors\n", | |
a586d4f6 | 1816 | devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1817 | } |
28fd6d7f RR |
1818 | |
1819 | /* Our random number generator device reads from /dev/random into the Guest's | |
1820 | * input buffers. The usual case is that the Guest doesn't want random numbers | |
1821 | * and so has no buffers although /dev/random is still readable, whereas | |
1822 | * console is the reverse. | |
1823 | * | |
1824 | * The same logic applies, however. */ | |
1825 | static bool handle_rng_input(int fd, struct device *dev) | |
1826 | { | |
1827 | int len; | |
1828 | unsigned int head, in_num, out_num, totlen = 0; | |
1829 | struct iovec iov[dev->vq->vring.num]; | |
1830 | ||
1831 | /* First we need a buffer from the Guests's virtqueue. */ | |
1832 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
1833 | ||
1834 | /* If they're not ready for input, stop listening to this file | |
1835 | * descriptor. We'll start again once they add an input buffer. */ | |
1836 | if (head == dev->vq->vring.num) | |
1837 | return false; | |
1838 | ||
1839 | if (out_num) | |
1840 | errx(1, "Output buffers in rng?"); | |
1841 | ||
1842 | /* This is why we convert to iovecs: the readv() call uses them, and so | |
1843 | * it reads straight into the Guest's buffer. We loop to make sure we | |
1844 | * fill it. */ | |
1845 | while (!iov_empty(iov, in_num)) { | |
1846 | len = readv(dev->fd, iov, in_num); | |
1847 | if (len <= 0) | |
1848 | err(1, "Read from /dev/random gave %i", len); | |
1849 | iov_consume(iov, in_num, len); | |
1850 | totlen += len; | |
1851 | } | |
1852 | ||
1853 | /* Tell the Guest about the new input. */ | |
1854 | add_used_and_trigger(fd, dev->vq, head, totlen); | |
1855 | ||
1856 | /* Everything went OK! */ | |
1857 | return true; | |
1858 | } | |
1859 | ||
1860 | /* And this creates a "hardware" random number device for the Guest. */ | |
1861 | static void setup_rng(void) | |
1862 | { | |
1863 | struct device *dev; | |
1864 | int fd; | |
1865 | ||
1866 | fd = open_or_die("/dev/random", O_RDONLY); | |
1867 | ||
1868 | /* The device responds to return from I/O thread. */ | |
1869 | dev = new_device("rng", VIRTIO_ID_RNG, fd, handle_rng_input); | |
1870 | ||
1871 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
1872 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
1873 | ||
1874 | verbose("device %u: rng\n", devices.device_num++); | |
1875 | } | |
a6bd8e13 | 1876 | /* That's the end of device setup. */ |
ec04b13f | 1877 | |
a6bd8e13 | 1878 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1879 | static void __attribute__((noreturn)) restart_guest(void) |
1880 | { | |
1881 | unsigned int i; | |
1882 | ||
a6bd8e13 | 1883 | /* Closing pipes causes the Waker thread and io_threads to die, and |
ec04b13f BR |
1884 | * closing /dev/lguest cleans up the Guest. Since we don't track all |
1885 | * open fds, we simply close everything beyond stderr. */ | |
1886 | for (i = 3; i < FD_SETSIZE; i++) | |
1887 | close(i); | |
1888 | execv(main_args[0], main_args); | |
1889 | err(1, "Could not exec %s", main_args[0]); | |
1890 | } | |
8ca47e00 | 1891 | |
a6bd8e13 | 1892 | /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves |
dde79789 | 1893 | * its input and output, and finally, lays it to rest. */ |
17cbca2b | 1894 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
8ca47e00 RR |
1895 | { |
1896 | for (;;) { | |
511801dc | 1897 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1898 | unsigned long notify_addr; |
8ca47e00 RR |
1899 | int readval; |
1900 | ||
1901 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1902 | readval = pread(lguest_fd, ¬ify_addr, |
1903 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1904 | |
17cbca2b RR |
1905 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1906 | if (readval == sizeof(notify_addr)) { | |
1907 | verbose("Notify on address %#lx\n", notify_addr); | |
1908 | handle_output(lguest_fd, notify_addr); | |
8ca47e00 | 1909 | continue; |
dde79789 | 1910 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1911 | } else if (errno == ENOENT) { |
1912 | char reason[1024] = { 0 }; | |
e3283fa0 | 1913 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1914 | errx(1, "%s", reason); |
ec04b13f BR |
1915 | /* ERESTART means that we need to reboot the guest */ |
1916 | } else if (errno == ERESTART) { | |
1917 | restart_guest(); | |
a161883a | 1918 | /* EAGAIN means a signal (timeout). |
dde79789 | 1919 | * Anything else means a bug or incompatible change. */ |
8ca47e00 RR |
1920 | } else if (errno != EAGAIN) |
1921 | err(1, "Running guest failed"); | |
dde79789 | 1922 | |
e3283fa0 GOC |
1923 | /* Only service input on thread for CPU 0. */ |
1924 | if (cpu_id != 0) | |
1925 | continue; | |
1926 | ||
e1e72965 | 1927 | /* Service input, then unset the BREAK to release the Waker. */ |
17cbca2b | 1928 | handle_input(lguest_fd); |
e3283fa0 | 1929 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) |
8ca47e00 RR |
1930 | err(1, "Resetting break"); |
1931 | } | |
1932 | } | |
a6bd8e13 | 1933 | /*L:240 |
e1e72965 RR |
1934 | * This is the end of the Launcher. The good news: we are over halfway |
1935 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1936 | * of us. | |
dde79789 | 1937 | * |
e1e72965 RR |
1938 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1939 | * "make Host". | |
1940 | :*/ | |
8ca47e00 RR |
1941 | |
1942 | static struct option opts[] = { | |
1943 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1944 | { "tunnet", 1, NULL, 't' }, |
1945 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1946 | { "rng", 0, NULL, 'r' }, |
8ca47e00 RR |
1947 | { "initrd", 1, NULL, 'i' }, |
1948 | { NULL }, | |
1949 | }; | |
1950 | static void usage(void) | |
1951 | { | |
1952 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1953 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1954 | "|--block=<filename>|--initrd=<filename>]...\n" |
1955 | "<mem-in-mb> vmlinux [args...]"); | |
1956 | } | |
1957 | ||
3c6b5bfa | 1958 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1959 | int main(int argc, char *argv[]) |
1960 | { | |
47436aa4 RR |
1961 | /* Memory, top-level pagetable, code startpoint and size of the |
1962 | * (optional) initrd. */ | |
1963 | unsigned long mem = 0, pgdir, start, initrd_size = 0; | |
e1e72965 | 1964 | /* Two temporaries and the /dev/lguest file descriptor. */ |
6570c459 | 1965 | int i, c, lguest_fd; |
3c6b5bfa | 1966 | /* The boot information for the Guest. */ |
43d33b21 | 1967 | struct boot_params *boot; |
dde79789 | 1968 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1969 | const char *initrd_name = NULL; |
1970 | ||
ec04b13f BR |
1971 | /* Save the args: we "reboot" by execing ourselves again. */ |
1972 | main_args = argv; | |
1973 | /* We don't "wait" for the children, so prevent them from becoming | |
1974 | * zombies. */ | |
1975 | signal(SIGCHLD, SIG_IGN); | |
1976 | ||
dde79789 RR |
1977 | /* First we initialize the device list. Since console and network |
1978 | * device receive input from a file descriptor, we keep an fdset | |
1979 | * (infds) and the maximum fd number (max_infd) with the head of the | |
a586d4f6 | 1980 | * list. We also keep a pointer to the last device. Finally, we keep |
a6bd8e13 RR |
1981 | * the next interrupt number to use for devices (1: remember that 0 is |
1982 | * used by the timer). */ | |
17cbca2b RR |
1983 | FD_ZERO(&devices.infds); |
1984 | devices.max_infd = -1; | |
a586d4f6 | 1985 | devices.lastdev = NULL; |
17cbca2b | 1986 | devices.next_irq = 1; |
8ca47e00 | 1987 | |
e3283fa0 | 1988 | cpu_id = 0; |
dde79789 RR |
1989 | /* We need to know how much memory so we can set up the device |
1990 | * descriptor and memory pages for the devices as we parse the command | |
1991 | * line. So we quickly look through the arguments to find the amount | |
1992 | * of memory now. */ | |
6570c459 RR |
1993 | for (i = 1; i < argc; i++) { |
1994 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1995 | mem = atoi(argv[i]) * 1024 * 1024; |
1996 | /* We start by mapping anonymous pages over all of | |
1997 | * guest-physical memory range. This fills it with 0, | |
1998 | * and ensures that the Guest won't be killed when it | |
1999 | * tries to access it. */ | |
2000 | guest_base = map_zeroed_pages(mem / getpagesize() | |
2001 | + DEVICE_PAGES); | |
2002 | guest_limit = mem; | |
2003 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 2004 | devices.descpage = get_pages(1); |
6570c459 RR |
2005 | break; |
2006 | } | |
2007 | } | |
dde79789 RR |
2008 | |
2009 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
2010 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
2011 | switch (c) { | |
2012 | case 'v': | |
2013 | verbose = true; | |
2014 | break; | |
8ca47e00 | 2015 | case 't': |
17cbca2b | 2016 | setup_tun_net(optarg); |
8ca47e00 RR |
2017 | break; |
2018 | case 'b': | |
17cbca2b | 2019 | setup_block_file(optarg); |
8ca47e00 | 2020 | break; |
28fd6d7f RR |
2021 | case 'r': |
2022 | setup_rng(); | |
2023 | break; | |
8ca47e00 RR |
2024 | case 'i': |
2025 | initrd_name = optarg; | |
2026 | break; | |
2027 | default: | |
2028 | warnx("Unknown argument %s", argv[optind]); | |
2029 | usage(); | |
2030 | } | |
2031 | } | |
dde79789 RR |
2032 | /* After the other arguments we expect memory and kernel image name, |
2033 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
2034 | if (optind + 2 > argc) |
2035 | usage(); | |
2036 | ||
3c6b5bfa RR |
2037 | verbose("Guest base is at %p\n", guest_base); |
2038 | ||
dde79789 | 2039 | /* We always have a console device */ |
17cbca2b | 2040 | setup_console(); |
8ca47e00 | 2041 | |
a161883a RR |
2042 | /* We can timeout waiting for Guest network transmit. */ |
2043 | setup_timeout(); | |
2044 | ||
8ca47e00 | 2045 | /* Now we load the kernel */ |
47436aa4 | 2046 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 2047 | |
3c6b5bfa RR |
2048 | /* Boot information is stashed at physical address 0 */ |
2049 | boot = from_guest_phys(0); | |
2050 | ||
dde79789 | 2051 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
2052 | if (initrd_name) { |
2053 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
2054 | /* These are the location in the Linux boot header where the |
2055 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
2056 | boot->hdr.ramdisk_image = mem - initrd_size; |
2057 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 2058 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 2059 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
2060 | } |
2061 | ||
dde79789 | 2062 | /* Set up the initial linear pagetables, starting below the initrd. */ |
47436aa4 | 2063 | pgdir = setup_pagetables(mem, initrd_size); |
8ca47e00 | 2064 | |
dde79789 RR |
2065 | /* The Linux boot header contains an "E820" memory map: ours is a |
2066 | * simple, single region. */ | |
43d33b21 RR |
2067 | boot->e820_entries = 1; |
2068 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 2069 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
2070 | * line after the boot header. */ |
2071 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
e1e72965 | 2072 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 2073 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 2074 | |
814a0e5c | 2075 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 2076 | boot->hdr.version = 0x207; |
814a0e5c RR |
2077 | |
2078 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 2079 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 2080 | |
43d33b21 RR |
2081 | /* Tell the entry path not to try to reload segment registers. */ |
2082 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 2083 | |
dde79789 RR |
2084 | /* We tell the kernel to initialize the Guest: this returns the open |
2085 | * /dev/lguest file descriptor. */ | |
47436aa4 | 2086 | lguest_fd = tell_kernel(pgdir, start); |
dde79789 RR |
2087 | |
2088 | /* We fork off a child process, which wakes the Launcher whenever one | |
a6bd8e13 RR |
2089 | * of the input file descriptors needs attention. We call this the |
2090 | * Waker, and we'll cover it in a moment. */ | |
17cbca2b | 2091 | waker_fd = setup_waker(lguest_fd); |
8ca47e00 | 2092 | |
dde79789 | 2093 | /* Finally, run the Guest. This doesn't return. */ |
17cbca2b | 2094 | run_guest(lguest_fd); |
8ca47e00 | 2095 | } |
f56a384e RR |
2096 | /*:*/ |
2097 | ||
2098 | /*M:999 | |
2099 | * Mastery is done: you now know everything I do. | |
2100 | * | |
2101 | * But surely you have seen code, features and bugs in your wanderings which | |
2102 | * you now yearn to attack? That is the real game, and I look forward to you | |
2103 | * patching and forking lguest into the Your-Name-Here-visor. | |
2104 | * | |
2105 | * Farewell, and good coding! | |
2106 | * Rusty Russell. | |
2107 | */ |