Commit | Line | Data |
---|---|---|
f938d2c8 RR |
1 | /*P:100 This is the Launcher code, a simple program which lays out the |
2 | * "physical" memory for the new Guest by mapping the kernel image and the | |
3 | * virtual devices, then reads repeatedly from /dev/lguest to run the Guest. | |
3c6b5bfa | 4 | :*/ |
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> | |
37 | /*L:110 We can ignore the 30 include files we need for this program, but I do | |
dde79789 RR |
38 | * want to draw attention to the use of kernel-style types. |
39 | * | |
40 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
41 | * like these abbreviations and the header we need uses them, so we define them | |
42 | * here. | |
43 | */ | |
8ca47e00 RR |
44 | typedef unsigned long long u64; |
45 | typedef uint32_t u32; | |
46 | typedef uint16_t u16; | |
47 | typedef uint8_t u8; | |
b45d8cb0 | 48 | #include "linux/lguest_launcher.h" |
17cbca2b RR |
49 | #include "linux/pci_ids.h" |
50 | #include "linux/virtio_config.h" | |
51 | #include "linux/virtio_net.h" | |
52 | #include "linux/virtio_blk.h" | |
53 | #include "linux/virtio_console.h" | |
54 | #include "linux/virtio_ring.h" | |
b45d8cb0 | 55 | #include "asm-x86/e820.h" |
dde79789 | 56 | /*:*/ |
8ca47e00 RR |
57 | |
58 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
59 | #define NET_PEERNUM 1 | |
60 | #define BRIDGE_PFX "bridge:" | |
61 | #ifndef SIOCBRADDIF | |
62 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
63 | #endif | |
3c6b5bfa RR |
64 | /* We can have up to 256 pages for devices. */ |
65 | #define DEVICE_PAGES 256 | |
17cbca2b RR |
66 | /* This fits nicely in a single 4096-byte page. */ |
67 | #define VIRTQUEUE_NUM 127 | |
8ca47e00 | 68 | |
dde79789 RR |
69 | /*L:120 verbose is both a global flag and a macro. The C preprocessor allows |
70 | * this, and although I wouldn't recommend it, it works quite nicely here. */ | |
8ca47e00 RR |
71 | static bool verbose; |
72 | #define verbose(args...) \ | |
73 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
74 | /*:*/ |
75 | ||
76 | /* The pipe to send commands to the waker process */ | |
8ca47e00 | 77 | static int waker_fd; |
3c6b5bfa RR |
78 | /* The pointer to the start of guest memory. */ |
79 | static void *guest_base; | |
80 | /* The maximum guest physical address allowed, and maximum possible. */ | |
81 | static unsigned long guest_limit, guest_max; | |
8ca47e00 | 82 | |
dde79789 | 83 | /* This is our list of devices. */ |
8ca47e00 RR |
84 | struct device_list |
85 | { | |
dde79789 RR |
86 | /* Summary information about the devices in our list: ready to pass to |
87 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
88 | fd_set infds; |
89 | int max_infd; | |
90 | ||
17cbca2b RR |
91 | /* Counter to assign interrupt numbers. */ |
92 | unsigned int next_irq; | |
93 | ||
94 | /* Counter to print out convenient device numbers. */ | |
95 | unsigned int device_num; | |
96 | ||
dde79789 | 97 | /* The descriptor page for the devices. */ |
17cbca2b RR |
98 | u8 *descpage; |
99 | ||
100 | /* The tail of the last descriptor. */ | |
101 | unsigned int desc_used; | |
dde79789 RR |
102 | |
103 | /* A single linked list of devices. */ | |
8ca47e00 | 104 | struct device *dev; |
dde79789 | 105 | /* ... And an end pointer so we can easily append new devices */ |
8ca47e00 RR |
106 | struct device **lastdev; |
107 | }; | |
108 | ||
17cbca2b RR |
109 | /* The list of Guest devices, based on command line arguments. */ |
110 | static struct device_list devices; | |
111 | ||
dde79789 | 112 | /* The device structure describes a single device. */ |
8ca47e00 RR |
113 | struct device |
114 | { | |
dde79789 | 115 | /* The linked-list pointer. */ |
8ca47e00 | 116 | struct device *next; |
17cbca2b RR |
117 | |
118 | /* The this device's descriptor, as mapped into the Guest. */ | |
8ca47e00 | 119 | struct lguest_device_desc *desc; |
17cbca2b RR |
120 | |
121 | /* The name of this device, for --verbose. */ | |
122 | const char *name; | |
8ca47e00 | 123 | |
dde79789 RR |
124 | /* If handle_input is set, it wants to be called when this file |
125 | * descriptor is ready. */ | |
8ca47e00 RR |
126 | int fd; |
127 | bool (*handle_input)(int fd, struct device *me); | |
128 | ||
17cbca2b RR |
129 | /* Any queues attached to this device */ |
130 | struct virtqueue *vq; | |
8ca47e00 RR |
131 | |
132 | /* Device-specific data. */ | |
133 | void *priv; | |
134 | }; | |
135 | ||
17cbca2b RR |
136 | /* The virtqueue structure describes a queue attached to a device. */ |
137 | struct virtqueue | |
138 | { | |
139 | struct virtqueue *next; | |
140 | ||
141 | /* Which device owns me. */ | |
142 | struct device *dev; | |
143 | ||
144 | /* The configuration for this queue. */ | |
145 | struct lguest_vqconfig config; | |
146 | ||
147 | /* The actual ring of buffers. */ | |
148 | struct vring vring; | |
149 | ||
150 | /* Last available index we saw. */ | |
151 | u16 last_avail_idx; | |
152 | ||
153 | /* The routine to call when the Guest pings us. */ | |
154 | void (*handle_output)(int fd, struct virtqueue *me); | |
155 | }; | |
156 | ||
157 | /* Since guest is UP and we don't run at the same time, we don't need barriers. | |
158 | * But I include them in the code in case others copy it. */ | |
159 | #define wmb() | |
160 | ||
161 | /* Convert an iovec element to the given type. | |
162 | * | |
163 | * This is a fairly ugly trick: we need to know the size of the type and | |
164 | * alignment requirement to check the pointer is kosher. It's also nice to | |
165 | * have the name of the type in case we report failure. | |
166 | * | |
167 | * Typing those three things all the time is cumbersome and error prone, so we | |
168 | * have a macro which sets them all up and passes to the real function. */ | |
169 | #define convert(iov, type) \ | |
170 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
171 | ||
172 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
173 | const char *name) | |
174 | { | |
175 | if (iov->iov_len != size) | |
176 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
177 | if ((unsigned long)iov->iov_base % align != 0) | |
178 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
179 | return iov->iov_base; | |
180 | } | |
181 | ||
182 | /* The virtio configuration space is defined to be little-endian. x86 is | |
183 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
184 | #define cpu_to_le16(v16) (v16) | |
185 | #define cpu_to_le32(v32) (v32) | |
186 | #define cpu_to_le64(v64) (v64) | |
187 | #define le16_to_cpu(v16) (v16) | |
188 | #define le32_to_cpu(v32) (v32) | |
189 | #define le64_to_cpu(v32) (v64) | |
190 | ||
3c6b5bfa RR |
191 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
192 | * where pointers run wild and free! Unfortunately, like most userspace | |
193 | * programs, it's quite boring (which is why everyone likes to hack on the | |
194 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
195 | * will get you through this section. Or, maybe not. | |
196 | * | |
197 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
198 | * memory and stores it in "guest_base". In other words, Guest physical == | |
199 | * Launcher virtual with an offset. | |
200 | * | |
201 | * This can be tough to get your head around, but usually it just means that we | |
202 | * use these trivial conversion functions when the Guest gives us it's | |
203 | * "physical" addresses: */ | |
204 | static void *from_guest_phys(unsigned long addr) | |
205 | { | |
206 | return guest_base + addr; | |
207 | } | |
208 | ||
209 | static unsigned long to_guest_phys(const void *addr) | |
210 | { | |
211 | return (addr - guest_base); | |
212 | } | |
213 | ||
dde79789 RR |
214 | /*L:130 |
215 | * Loading the Kernel. | |
216 | * | |
217 | * We start with couple of simple helper routines. open_or_die() avoids | |
218 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
219 | static int open_or_die(const char *name, int flags) |
220 | { | |
221 | int fd = open(name, flags); | |
222 | if (fd < 0) | |
223 | err(1, "Failed to open %s", name); | |
224 | return fd; | |
225 | } | |
226 | ||
3c6b5bfa RR |
227 | /* map_zeroed_pages() takes a number of pages. */ |
228 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 229 | { |
3c6b5bfa RR |
230 | int fd = open_or_die("/dev/zero", O_RDONLY); |
231 | void *addr; | |
8ca47e00 | 232 | |
dde79789 | 233 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
234 | * copied). */ |
235 | addr = mmap(NULL, getpagesize() * num, | |
236 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
237 | if (addr == MAP_FAILED) | |
238 | err(1, "Mmaping %u pages of /dev/zero", num); | |
239 | ||
240 | return addr; | |
241 | } | |
242 | ||
243 | /* Get some more pages for a device. */ | |
244 | static void *get_pages(unsigned int num) | |
245 | { | |
246 | void *addr = from_guest_phys(guest_limit); | |
247 | ||
248 | guest_limit += num * getpagesize(); | |
249 | if (guest_limit > guest_max) | |
250 | errx(1, "Not enough memory for devices"); | |
251 | return addr; | |
8ca47e00 RR |
252 | } |
253 | ||
dde79789 RR |
254 | /* To find out where to start we look for the magic Guest string, which marks |
255 | * the code we see in lguest_asm.S. This is a hack which we are currently | |
256 | * plotting to replace with the normal Linux entry point. */ | |
47436aa4 | 257 | static unsigned long entry_point(const void *start, const void *end) |
8ca47e00 | 258 | { |
3c6b5bfa | 259 | const void *p; |
8ca47e00 | 260 | |
47436aa4 RR |
261 | /* The scan gives us the physical starting address. We boot with |
262 | * pagetables set up with virtual and physical the same, so that's | |
263 | * OK. */ | |
8ca47e00 RR |
264 | for (p = start; p < end; p++) |
265 | if (memcmp(p, "GenuineLguest", strlen("GenuineLguest")) == 0) | |
47436aa4 | 266 | return to_guest_phys(p + strlen("GenuineLguest")); |
8ca47e00 | 267 | |
babed5c0 | 268 | errx(1, "Is this image a genuine lguest?"); |
8ca47e00 RR |
269 | } |
270 | ||
6649bb7a RM |
271 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
272 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
273 | * it falls back to reading the memory in. */ | |
274 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
275 | { | |
276 | ssize_t r; | |
277 | ||
278 | /* We map writable even though for some segments are marked read-only. | |
279 | * The kernel really wants to be writable: it patches its own | |
280 | * instructions. | |
281 | * | |
282 | * MAP_PRIVATE means that the page won't be copied until a write is | |
283 | * done to it. This allows us to share untouched memory between | |
284 | * Guests. */ | |
285 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
286 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
287 | return; | |
288 | ||
289 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
290 | r = pread(fd, addr, len, offset); | |
291 | if (r != len) | |
292 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
293 | } | |
294 | ||
dde79789 RR |
295 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
296 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
297 | * by all modern binaries on Linux including the kernel. | |
298 | * | |
299 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
300 | * address. We use the physical address; the Guest will map itself to the |
301 | * virtual address. | |
dde79789 RR |
302 | * |
303 | * We return the starting address. */ | |
47436aa4 | 304 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 305 | { |
3c6b5bfa | 306 | void *start = (void *)-1, *end = NULL; |
8ca47e00 RR |
307 | Elf32_Phdr phdr[ehdr->e_phnum]; |
308 | unsigned int i; | |
8ca47e00 | 309 | |
dde79789 RR |
310 | /* Sanity checks on the main ELF header: an x86 executable with a |
311 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
312 | if (ehdr->e_type != ET_EXEC |
313 | || ehdr->e_machine != EM_386 | |
314 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
315 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
316 | errx(1, "Malformed elf header"); | |
317 | ||
dde79789 RR |
318 | /* An ELF executable contains an ELF header and a number of "program" |
319 | * headers which indicate which parts ("segments") of the program to | |
320 | * load where. */ | |
321 | ||
322 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
323 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
324 | err(1, "Seeking to program headers"); | |
325 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
326 | err(1, "Reading program headers"); | |
327 | ||
dde79789 RR |
328 | /* Try all the headers: there are usually only three. A read-only one, |
329 | * a read-write one, and a "note" section which isn't loadable. */ | |
8ca47e00 | 330 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 331 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
332 | if (phdr[i].p_type != PT_LOAD) |
333 | continue; | |
334 | ||
335 | verbose("Section %i: size %i addr %p\n", | |
336 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
337 | ||
dde79789 RR |
338 | /* We track the first and last address we mapped, so we can |
339 | * tell entry_point() where to scan. */ | |
3c6b5bfa RR |
340 | if (from_guest_phys(phdr[i].p_paddr) < start) |
341 | start = from_guest_phys(phdr[i].p_paddr); | |
342 | if (from_guest_phys(phdr[i].p_paddr) + phdr[i].p_filesz > end) | |
343 | end=from_guest_phys(phdr[i].p_paddr)+phdr[i].p_filesz; | |
8ca47e00 | 344 | |
6649bb7a | 345 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 346 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 347 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
348 | } |
349 | ||
47436aa4 | 350 | return entry_point(start, end); |
8ca47e00 RR |
351 | } |
352 | ||
dde79789 RR |
353 | /*L:160 Unfortunately the entire ELF image isn't compressed: the segments |
354 | * which need loading are extracted and compressed raw. This denies us the | |
355 | * information we need to make a fully-general loader. */ | |
47436aa4 | 356 | static unsigned long unpack_bzimage(int fd) |
8ca47e00 RR |
357 | { |
358 | gzFile f; | |
359 | int ret, len = 0; | |
dde79789 RR |
360 | /* A bzImage always gets loaded at physical address 1M. This is |
361 | * actually configurable as CONFIG_PHYSICAL_START, but as the comment | |
362 | * there says, "Don't change this unless you know what you are doing". | |
363 | * Indeed. */ | |
3c6b5bfa | 364 | void *img = from_guest_phys(0x100000); |
8ca47e00 | 365 | |
dde79789 RR |
366 | /* gzdopen takes our file descriptor (carefully placed at the start of |
367 | * the GZIP header we found) and returns a gzFile. */ | |
8ca47e00 | 368 | f = gzdopen(fd, "rb"); |
dde79789 | 369 | /* We read it into memory in 64k chunks until we hit the end. */ |
8ca47e00 RR |
370 | while ((ret = gzread(f, img + len, 65536)) > 0) |
371 | len += ret; | |
372 | if (ret < 0) | |
373 | err(1, "reading image from bzImage"); | |
374 | ||
375 | verbose("Unpacked size %i addr %p\n", len, img); | |
dde79789 | 376 | |
47436aa4 | 377 | return entry_point(img, img + len); |
8ca47e00 RR |
378 | } |
379 | ||
dde79789 RR |
380 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
381 | * supposed to jump into it and it will unpack itself. We can't do that | |
382 | * because the Guest can't run the unpacking code, and adding features to | |
383 | * lguest kills puppies, so we don't want to. | |
384 | * | |
385 | * The bzImage is formed by putting the decompressing code in front of the | |
386 | * compressed kernel code. So we can simple scan through it looking for the | |
387 | * first "gzip" header, and start decompressing from there. */ | |
47436aa4 | 388 | static unsigned long load_bzimage(int fd) |
8ca47e00 RR |
389 | { |
390 | unsigned char c; | |
391 | int state = 0; | |
392 | ||
dde79789 | 393 | /* GZIP header is 0x1F 0x8B <method> <flags>... <compressed-by>. */ |
8ca47e00 RR |
394 | while (read(fd, &c, 1) == 1) { |
395 | switch (state) { | |
396 | case 0: | |
397 | if (c == 0x1F) | |
398 | state++; | |
399 | break; | |
400 | case 1: | |
401 | if (c == 0x8B) | |
402 | state++; | |
403 | else | |
404 | state = 0; | |
405 | break; | |
406 | case 2 ... 8: | |
407 | state++; | |
408 | break; | |
409 | case 9: | |
dde79789 | 410 | /* Seek back to the start of the gzip header. */ |
8ca47e00 | 411 | lseek(fd, -10, SEEK_CUR); |
dde79789 RR |
412 | /* One final check: "compressed under UNIX". */ |
413 | if (c != 0x03) | |
8ca47e00 RR |
414 | state = -1; |
415 | else | |
47436aa4 | 416 | return unpack_bzimage(fd); |
8ca47e00 RR |
417 | } |
418 | } | |
419 | errx(1, "Could not find kernel in bzImage"); | |
420 | } | |
421 | ||
dde79789 RR |
422 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
423 | * come wrapped up in the self-decompressing "bzImage" format. With some funky | |
424 | * coding, we can load those, too. */ | |
47436aa4 | 425 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
426 | { |
427 | Elf32_Ehdr hdr; | |
428 | ||
dde79789 | 429 | /* Read in the first few bytes. */ |
8ca47e00 RR |
430 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
431 | err(1, "Reading kernel"); | |
432 | ||
dde79789 | 433 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 434 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 435 | return map_elf(fd, &hdr); |
8ca47e00 | 436 | |
dde79789 | 437 | /* Otherwise we assume it's a bzImage, and try to unpack it */ |
47436aa4 | 438 | return load_bzimage(fd); |
8ca47e00 RR |
439 | } |
440 | ||
dde79789 RR |
441 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
442 | * it calls getpagesize() twice: "it's dumb code." | |
443 | * | |
444 | * Kernel guys get really het up about optimization, even when it's not | |
445 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
446 | static inline unsigned long page_align(unsigned long addr) |
447 | { | |
dde79789 | 448 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
449 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
450 | } | |
451 | ||
dde79789 RR |
452 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
453 | * the kernel which the kernel can use to boot from without needing any | |
454 | * drivers. Most distributions now use this as standard: the initrd contains | |
455 | * the code to load the appropriate driver modules for the current machine. | |
456 | * | |
457 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
458 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
459 | static unsigned long load_initrd(const char *name, unsigned long mem) |
460 | { | |
461 | int ifd; | |
462 | struct stat st; | |
463 | unsigned long len; | |
8ca47e00 RR |
464 | |
465 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 466 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
467 | if (fstat(ifd, &st) < 0) |
468 | err(1, "fstat() on initrd '%s'", name); | |
469 | ||
6649bb7a RM |
470 | /* We map the initrd at the top of memory, but mmap wants it to be |
471 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 472 | len = page_align(st.st_size); |
3c6b5bfa | 473 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
474 | /* Once a file is mapped, you can close the file descriptor. It's a |
475 | * little odd, but quite useful. */ | |
8ca47e00 | 476 | close(ifd); |
6649bb7a | 477 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
478 | |
479 | /* We return the initrd size. */ | |
8ca47e00 RR |
480 | return len; |
481 | } | |
482 | ||
47436aa4 RR |
483 | /* Once we know how much memory we have, we can construct simple linear page |
484 | * tables which set virtual == physical which will get the Guest far enough | |
3c6b5bfa | 485 | * into the boot to create its own. |
dde79789 RR |
486 | * |
487 | * We lay them out of the way, just below the initrd (which is why we need to | |
488 | * know its size). */ | |
8ca47e00 | 489 | static unsigned long setup_pagetables(unsigned long mem, |
47436aa4 | 490 | unsigned long initrd_size) |
8ca47e00 | 491 | { |
511801dc | 492 | unsigned long *pgdir, *linear; |
8ca47e00 | 493 | unsigned int mapped_pages, i, linear_pages; |
511801dc | 494 | unsigned int ptes_per_page = getpagesize()/sizeof(void *); |
8ca47e00 | 495 | |
47436aa4 | 496 | mapped_pages = mem/getpagesize(); |
8ca47e00 | 497 | |
dde79789 | 498 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
8ca47e00 RR |
499 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
500 | ||
dde79789 | 501 | /* We put the toplevel page directory page at the top of memory. */ |
3c6b5bfa | 502 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
dde79789 RR |
503 | |
504 | /* Now we use the next linear_pages pages as pte pages */ | |
8ca47e00 RR |
505 | linear = (void *)pgdir - linear_pages*getpagesize(); |
506 | ||
dde79789 RR |
507 | /* Linear mapping is easy: put every page's address into the mapping in |
508 | * order. PAGE_PRESENT contains the flags Present, Writable and | |
509 | * Executable. */ | |
8ca47e00 RR |
510 | for (i = 0; i < mapped_pages; i++) |
511 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); | |
512 | ||
47436aa4 | 513 | /* The top level points to the linear page table pages above. */ |
8ca47e00 | 514 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
47436aa4 | 515 | pgdir[i/ptes_per_page] |
511801dc | 516 | = ((to_guest_phys(linear) + i*sizeof(void *)) |
3c6b5bfa | 517 | | PAGE_PRESENT); |
8ca47e00 RR |
518 | } |
519 | ||
3c6b5bfa RR |
520 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
521 | mapped_pages, linear_pages, to_guest_phys(linear)); | |
8ca47e00 | 522 | |
dde79789 RR |
523 | /* We return the top level (guest-physical) address: the kernel needs |
524 | * to know where it is. */ | |
3c6b5bfa | 525 | return to_guest_phys(pgdir); |
8ca47e00 RR |
526 | } |
527 | ||
dde79789 RR |
528 | /* Simple routine to roll all the commandline arguments together with spaces |
529 | * between them. */ | |
8ca47e00 RR |
530 | static void concat(char *dst, char *args[]) |
531 | { | |
532 | unsigned int i, len = 0; | |
533 | ||
534 | for (i = 0; args[i]; i++) { | |
535 | strcpy(dst+len, args[i]); | |
536 | strcat(dst+len, " "); | |
537 | len += strlen(args[i]) + 1; | |
538 | } | |
539 | /* In case it's empty. */ | |
540 | dst[len] = '\0'; | |
541 | } | |
542 | ||
dde79789 RR |
543 | /* This is where we actually tell the kernel to initialize the Guest. We saw |
544 | * the arguments it expects when we looked at initialize() in lguest_user.c: | |
3c6b5bfa | 545 | * the base of guest "physical" memory, the top physical page to allow, the |
47436aa4 RR |
546 | * top level pagetable and the entry point for the Guest. */ |
547 | static int tell_kernel(unsigned long pgdir, unsigned long start) | |
8ca47e00 | 548 | { |
511801dc JS |
549 | unsigned long args[] = { LHREQ_INITIALIZE, |
550 | (unsigned long)guest_base, | |
47436aa4 | 551 | guest_limit / getpagesize(), pgdir, start }; |
8ca47e00 RR |
552 | int fd; |
553 | ||
3c6b5bfa RR |
554 | verbose("Guest: %p - %p (%#lx)\n", |
555 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
556 | fd = open_or_die("/dev/lguest", O_RDWR); |
557 | if (write(fd, args, sizeof(args)) < 0) | |
558 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
559 | |
560 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
561 | return fd; |
562 | } | |
dde79789 | 563 | /*:*/ |
8ca47e00 | 564 | |
17cbca2b | 565 | static void add_device_fd(int fd) |
8ca47e00 | 566 | { |
17cbca2b RR |
567 | FD_SET(fd, &devices.infds); |
568 | if (fd > devices.max_infd) | |
569 | devices.max_infd = fd; | |
8ca47e00 RR |
570 | } |
571 | ||
dde79789 RR |
572 | /*L:200 |
573 | * The Waker. | |
574 | * | |
575 | * With a console and network devices, we can have lots of input which we need | |
576 | * to process. We could try to tell the kernel what file descriptors to watch, | |
577 | * but handing a file descriptor mask through to the kernel is fairly icky. | |
578 | * | |
579 | * Instead, we fork off a process which watches the file descriptors and writes | |
580 | * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host | |
581 | * loop to stop running the Guest. This causes it to return from the | |
582 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset | |
583 | * the LHREQ_BREAK and wake us up again. | |
584 | * | |
585 | * This, of course, is merely a different *kind* of icky. | |
586 | */ | |
17cbca2b | 587 | static void wake_parent(int pipefd, int lguest_fd) |
8ca47e00 | 588 | { |
dde79789 RR |
589 | /* Add the pipe from the Launcher to the fdset in the device_list, so |
590 | * we watch it, too. */ | |
17cbca2b | 591 | add_device_fd(pipefd); |
8ca47e00 RR |
592 | |
593 | for (;;) { | |
17cbca2b | 594 | fd_set rfds = devices.infds; |
511801dc | 595 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8ca47e00 | 596 | |
dde79789 | 597 | /* Wait until input is ready from one of the devices. */ |
17cbca2b | 598 | select(devices.max_infd+1, &rfds, NULL, NULL, NULL); |
dde79789 | 599 | /* Is it a message from the Launcher? */ |
8ca47e00 | 600 | if (FD_ISSET(pipefd, &rfds)) { |
56ae43df | 601 | int fd; |
dde79789 RR |
602 | /* If read() returns 0, it means the Launcher has |
603 | * exited. We silently follow. */ | |
56ae43df | 604 | if (read(pipefd, &fd, sizeof(fd)) == 0) |
8ca47e00 | 605 | exit(0); |
56ae43df RR |
606 | /* Otherwise it's telling us to change what file |
607 | * descriptors we're to listen to. */ | |
608 | if (fd >= 0) | |
609 | FD_SET(fd, &devices.infds); | |
610 | else | |
611 | FD_CLR(-fd - 1, &devices.infds); | |
dde79789 | 612 | } else /* Send LHREQ_BREAK command. */ |
8ca47e00 RR |
613 | write(lguest_fd, args, sizeof(args)); |
614 | } | |
615 | } | |
616 | ||
dde79789 | 617 | /* This routine just sets up a pipe to the Waker process. */ |
17cbca2b | 618 | static int setup_waker(int lguest_fd) |
8ca47e00 RR |
619 | { |
620 | int pipefd[2], child; | |
621 | ||
dde79789 RR |
622 | /* We create a pipe to talk to the waker, and also so it knows when the |
623 | * Launcher dies (and closes pipe). */ | |
8ca47e00 RR |
624 | pipe(pipefd); |
625 | child = fork(); | |
626 | if (child == -1) | |
627 | err(1, "forking"); | |
628 | ||
629 | if (child == 0) { | |
dde79789 | 630 | /* Close the "writing" end of our copy of the pipe */ |
8ca47e00 | 631 | close(pipefd[1]); |
17cbca2b | 632 | wake_parent(pipefd[0], lguest_fd); |
8ca47e00 | 633 | } |
dde79789 | 634 | /* Close the reading end of our copy of the pipe. */ |
8ca47e00 RR |
635 | close(pipefd[0]); |
636 | ||
dde79789 | 637 | /* Here is the fd used to talk to the waker. */ |
8ca47e00 RR |
638 | return pipefd[1]; |
639 | } | |
640 | ||
dde79789 RR |
641 | /*L:210 |
642 | * Device Handling. | |
643 | * | |
644 | * When the Guest sends DMA to us, it sends us an array of addresses and sizes. | |
645 | * We need to make sure it's not trying to reach into the Launcher itself, so | |
646 | * we have a convenient routine which check it and exits with an error message | |
647 | * if something funny is going on: | |
648 | */ | |
8ca47e00 RR |
649 | static void *_check_pointer(unsigned long addr, unsigned int size, |
650 | unsigned int line) | |
651 | { | |
dde79789 RR |
652 | /* We have to separately check addr and addr+size, because size could |
653 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 654 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 655 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
656 | /* We return a pointer for the caller's convenience, now we know it's |
657 | * safe to use. */ | |
3c6b5bfa | 658 | return from_guest_phys(addr); |
8ca47e00 | 659 | } |
dde79789 | 660 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
661 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
662 | ||
17cbca2b RR |
663 | /* This function returns the next descriptor in the chain, or vq->vring.num. */ |
664 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) | |
665 | { | |
666 | unsigned int next; | |
667 | ||
668 | /* If this descriptor says it doesn't chain, we're done. */ | |
669 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
670 | return vq->vring.num; | |
671 | ||
672 | /* Check they're not leading us off end of descriptors. */ | |
673 | next = vq->vring.desc[i].next; | |
674 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
675 | wmb(); | |
676 | ||
677 | if (next >= vq->vring.num) | |
678 | errx(1, "Desc next is %u", next); | |
679 | ||
680 | return next; | |
681 | } | |
682 | ||
683 | /* This looks in the virtqueue and for the first available buffer, and converts | |
684 | * it to an iovec for convenient access. Since descriptors consist of some | |
685 | * number of output then some number of input descriptors, it's actually two | |
686 | * iovecs, but we pack them into one and note how many of each there were. | |
687 | * | |
688 | * This function returns the descriptor number found, or vq->vring.num (which | |
689 | * is never a valid descriptor number) if none was found. */ | |
690 | static unsigned get_vq_desc(struct virtqueue *vq, | |
691 | struct iovec iov[], | |
692 | unsigned int *out_num, unsigned int *in_num) | |
693 | { | |
694 | unsigned int i, head; | |
695 | ||
696 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
697 | if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num) | |
698 | errx(1, "Guest moved used index from %u to %u", | |
699 | vq->last_avail_idx, vq->vring.avail->idx); | |
700 | ||
701 | /* If there's nothing new since last we looked, return invalid. */ | |
702 | if (vq->vring.avail->idx == vq->last_avail_idx) | |
703 | return vq->vring.num; | |
704 | ||
705 | /* Grab the next descriptor number they're advertising, and increment | |
706 | * the index we've seen. */ | |
707 | head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num]; | |
708 | ||
709 | /* If their number is silly, that's a fatal mistake. */ | |
710 | if (head >= vq->vring.num) | |
711 | errx(1, "Guest says index %u is available", head); | |
712 | ||
713 | /* When we start there are none of either input nor output. */ | |
714 | *out_num = *in_num = 0; | |
715 | ||
716 | i = head; | |
717 | do { | |
718 | /* Grab the first descriptor, and check it's OK. */ | |
719 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
720 | iov[*out_num + *in_num].iov_base | |
721 | = check_pointer(vq->vring.desc[i].addr, | |
722 | vq->vring.desc[i].len); | |
723 | /* If this is an input descriptor, increment that count. */ | |
724 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
725 | (*in_num)++; | |
726 | else { | |
727 | /* If it's an output descriptor, they're all supposed | |
728 | * to come before any input descriptors. */ | |
729 | if (*in_num) | |
730 | errx(1, "Descriptor has out after in"); | |
731 | (*out_num)++; | |
732 | } | |
733 | ||
734 | /* If we've got too many, that implies a descriptor loop. */ | |
735 | if (*out_num + *in_num > vq->vring.num) | |
736 | errx(1, "Looped descriptor"); | |
737 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 738 | |
17cbca2b | 739 | return head; |
8ca47e00 RR |
740 | } |
741 | ||
17cbca2b RR |
742 | /* Once we've used one of their buffers, we tell them about it. We'll then |
743 | * want to send them an interrupt, using trigger_irq(). */ | |
744 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 745 | { |
17cbca2b RR |
746 | struct vring_used_elem *used; |
747 | ||
748 | /* Get a pointer to the next entry in the used ring. */ | |
749 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; | |
750 | used->id = head; | |
751 | used->len = len; | |
752 | /* Make sure buffer is written before we update index. */ | |
753 | wmb(); | |
754 | vq->vring.used->idx++; | |
8ca47e00 RR |
755 | } |
756 | ||
17cbca2b RR |
757 | /* This actually sends the interrupt for this virtqueue */ |
758 | static void trigger_irq(int fd, struct virtqueue *vq) | |
8ca47e00 | 759 | { |
17cbca2b RR |
760 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
761 | ||
762 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
763 | return; | |
764 | ||
765 | /* Send the Guest an interrupt tell them we used something up. */ | |
8ca47e00 | 766 | if (write(fd, buf, sizeof(buf)) != 0) |
17cbca2b | 767 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
768 | } |
769 | ||
17cbca2b RR |
770 | /* And here's the combo meal deal. Supersize me! */ |
771 | static void add_used_and_trigger(int fd, struct virtqueue *vq, | |
772 | unsigned int head, int len) | |
8ca47e00 | 773 | { |
17cbca2b RR |
774 | add_used(vq, head, len); |
775 | trigger_irq(fd, vq); | |
8ca47e00 RR |
776 | } |
777 | ||
dde79789 RR |
778 | /* Here is the input terminal setting we save, and the routine to restore them |
779 | * on exit so the user can see what they type next. */ | |
8ca47e00 RR |
780 | static struct termios orig_term; |
781 | static void restore_term(void) | |
782 | { | |
783 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
784 | } | |
785 | ||
dde79789 | 786 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
787 | struct console_abort |
788 | { | |
dde79789 | 789 | /* How many times have they hit ^C? */ |
8ca47e00 | 790 | int count; |
dde79789 | 791 | /* When did they start? */ |
8ca47e00 RR |
792 | struct timeval start; |
793 | }; | |
794 | ||
dde79789 | 795 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
796 | static bool handle_console_input(int fd, struct device *dev) |
797 | { | |
8ca47e00 | 798 | int len; |
17cbca2b RR |
799 | unsigned int head, in_num, out_num; |
800 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
801 | struct console_abort *abort = dev->priv; |
802 | ||
17cbca2b RR |
803 | /* First we need a console buffer from the Guests's input virtqueue. */ |
804 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
805 | |
806 | /* If they're not ready for input, stop listening to this file | |
807 | * descriptor. We'll start again once they add an input buffer. */ | |
808 | if (head == dev->vq->vring.num) | |
809 | return false; | |
810 | ||
811 | if (out_num) | |
17cbca2b | 812 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 813 | |
dde79789 RR |
814 | /* This is why we convert to iovecs: the readv() call uses them, and so |
815 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 816 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 817 | if (len <= 0) { |
dde79789 | 818 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 819 | * something went terribly wrong. */ |
8ca47e00 | 820 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 821 | /* Put the input terminal back. */ |
17cbca2b | 822 | restore_term(); |
56ae43df RR |
823 | /* Remove callback from input vq, so it doesn't restart us. */ |
824 | dev->vq->handle_output = NULL; | |
825 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 826 | return false; |
8ca47e00 RR |
827 | } |
828 | ||
56ae43df RR |
829 | /* Tell the Guest about the new input. */ |
830 | add_used_and_trigger(fd, dev->vq, head, len); | |
8ca47e00 | 831 | |
dde79789 RR |
832 | /* Three ^C within one second? Exit. |
833 | * | |
834 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
835 | * in a buffer by itself, so they can't be too fast. But we check that | |
836 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
837 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
838 | if (!abort->count++) | |
839 | gettimeofday(&abort->start, NULL); | |
840 | else if (abort->count == 3) { | |
841 | struct timeval now; | |
842 | gettimeofday(&now, NULL); | |
843 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 844 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
845 | /* Close the fd so Waker will know it has to |
846 | * exit. */ | |
8ca47e00 | 847 | close(waker_fd); |
dde79789 RR |
848 | /* Just in case waker is blocked in BREAK, send |
849 | * unbreak now. */ | |
8ca47e00 RR |
850 | write(fd, args, sizeof(args)); |
851 | exit(2); | |
852 | } | |
853 | abort->count = 0; | |
854 | } | |
855 | } else | |
dde79789 | 856 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
857 | abort->count = 0; |
858 | ||
dde79789 | 859 | /* Everything went OK! */ |
8ca47e00 RR |
860 | return true; |
861 | } | |
862 | ||
17cbca2b RR |
863 | /* Handling output for console is simple: we just get all the output buffers |
864 | * and write them to stdout. */ | |
865 | static void handle_console_output(int fd, struct virtqueue *vq) | |
8ca47e00 | 866 | { |
17cbca2b RR |
867 | unsigned int head, out, in; |
868 | int len; | |
869 | struct iovec iov[vq->vring.num]; | |
870 | ||
871 | /* Keep getting output buffers from the Guest until we run out. */ | |
872 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
873 | if (in) | |
874 | errx(1, "Input buffers in output queue?"); | |
875 | len = writev(STDOUT_FILENO, iov, out); | |
876 | add_used_and_trigger(fd, vq, head, len); | |
877 | } | |
8ca47e00 RR |
878 | } |
879 | ||
17cbca2b RR |
880 | /* Handling output for network is also simple: we get all the output buffers |
881 | * and write them (ignoring the first element) to this device's file descriptor | |
882 | * (stdout). */ | |
883 | static void handle_net_output(int fd, struct virtqueue *vq) | |
8ca47e00 | 884 | { |
17cbca2b RR |
885 | unsigned int head, out, in; |
886 | int len; | |
887 | struct iovec iov[vq->vring.num]; | |
888 | ||
889 | /* Keep getting output buffers from the Guest until we run out. */ | |
890 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
891 | if (in) | |
892 | errx(1, "Input buffers in output queue?"); | |
893 | /* Check header, but otherwise ignore it (we said we supported | |
894 | * no features). */ | |
895 | (void)convert(&iov[0], struct virtio_net_hdr); | |
896 | len = writev(vq->dev->fd, iov+1, out-1); | |
897 | add_used_and_trigger(fd, vq, head, len); | |
898 | } | |
8ca47e00 RR |
899 | } |
900 | ||
17cbca2b RR |
901 | /* This is where we handle a packet coming in from the tun device to our |
902 | * Guest. */ | |
8ca47e00 RR |
903 | static bool handle_tun_input(int fd, struct device *dev) |
904 | { | |
17cbca2b | 905 | unsigned int head, in_num, out_num; |
8ca47e00 | 906 | int len; |
17cbca2b RR |
907 | struct iovec iov[dev->vq->vring.num]; |
908 | struct virtio_net_hdr *hdr; | |
8ca47e00 | 909 | |
17cbca2b RR |
910 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
911 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
912 | if (head == dev->vq->vring.num) { | |
dde79789 | 913 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
914 | * early, the Guest won't be ready yet. Wait until the device |
915 | * status says it's ready. */ | |
916 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
917 | if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) | |
8ca47e00 | 918 | warn("network: no dma buffer!"); |
56ae43df RR |
919 | /* We'll turn this back on if input buffers are registered. */ |
920 | return false; | |
17cbca2b RR |
921 | } else if (out_num) |
922 | errx(1, "Output buffers in network recv queue?"); | |
923 | ||
924 | /* First element is the header: we set it to 0 (no features). */ | |
925 | hdr = convert(&iov[0], struct virtio_net_hdr); | |
926 | hdr->flags = 0; | |
927 | hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
8ca47e00 | 928 | |
dde79789 | 929 | /* Read the packet from the device directly into the Guest's buffer. */ |
17cbca2b | 930 | len = readv(dev->fd, iov+1, in_num-1); |
8ca47e00 RR |
931 | if (len <= 0) |
932 | err(1, "reading network"); | |
dde79789 | 933 | |
56ae43df RR |
934 | /* Tell the Guest about the new packet. */ |
935 | add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len); | |
17cbca2b | 936 | |
8ca47e00 | 937 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
938 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
939 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
940 | ||
dde79789 | 941 | /* All good. */ |
8ca47e00 RR |
942 | return true; |
943 | } | |
944 | ||
56ae43df RR |
945 | /* This callback ensures we try again, in case we stopped console or net |
946 | * delivery because Guest didn't have any buffers. */ | |
947 | static void enable_fd(int fd, struct virtqueue *vq) | |
948 | { | |
949 | add_device_fd(vq->dev->fd); | |
950 | /* Tell waker to listen to it again */ | |
951 | write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd)); | |
952 | } | |
953 | ||
17cbca2b RR |
954 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
955 | static void handle_output(int fd, unsigned long addr) | |
8ca47e00 RR |
956 | { |
957 | struct device *i; | |
17cbca2b RR |
958 | struct virtqueue *vq; |
959 | ||
960 | /* Check each virtqueue. */ | |
961 | for (i = devices.dev; i; i = i->next) { | |
962 | for (vq = i->vq; vq; vq = vq->next) { | |
963 | if (vq->config.pfn == addr/getpagesize() | |
964 | && vq->handle_output) { | |
965 | verbose("Output to %s\n", vq->dev->name); | |
966 | vq->handle_output(fd, vq); | |
967 | return; | |
968 | } | |
8ca47e00 RR |
969 | } |
970 | } | |
dde79789 | 971 | |
17cbca2b RR |
972 | /* Early console write is done using notify on a nul-terminated string |
973 | * in Guest memory. */ | |
974 | if (addr >= guest_limit) | |
975 | errx(1, "Bad NOTIFY %#lx", addr); | |
976 | ||
977 | write(STDOUT_FILENO, from_guest_phys(addr), | |
978 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
979 | } |
980 | ||
dde79789 RR |
981 | /* This is called when the waker wakes us up: check for incoming file |
982 | * descriptors. */ | |
17cbca2b | 983 | static void handle_input(int fd) |
8ca47e00 | 984 | { |
dde79789 | 985 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
986 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
987 | ||
988 | for (;;) { | |
989 | struct device *i; | |
17cbca2b | 990 | fd_set fds = devices.infds; |
8ca47e00 | 991 | |
dde79789 | 992 | /* If nothing is ready, we're done. */ |
17cbca2b | 993 | if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0) |
8ca47e00 RR |
994 | break; |
995 | ||
dde79789 RR |
996 | /* Otherwise, call the device(s) which have readable |
997 | * file descriptors and a method of handling them. */ | |
17cbca2b | 998 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 999 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56ae43df RR |
1000 | int dev_fd; |
1001 | if (i->handle_input(fd, i)) | |
1002 | continue; | |
1003 | ||
dde79789 | 1004 | /* If handle_input() returns false, it means we |
56ae43df RR |
1005 | * should no longer service it. Networking and |
1006 | * console do this when there's no input | |
1007 | * buffers to deliver into. Console also uses | |
1008 | * it when it discovers that stdin is | |
1009 | * closed. */ | |
1010 | FD_CLR(i->fd, &devices.infds); | |
1011 | /* Tell waker to ignore it too, by sending a | |
1012 | * negative fd number (-1, since 0 is a valid | |
1013 | * FD number). */ | |
1014 | dev_fd = -i->fd - 1; | |
1015 | write(waker_fd, &dev_fd, sizeof(dev_fd)); | |
8ca47e00 RR |
1016 | } |
1017 | } | |
1018 | } | |
1019 | } | |
1020 | ||
dde79789 RR |
1021 | /*L:190 |
1022 | * Device Setup | |
1023 | * | |
1024 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
1025 | * device" so the Launcher can keep track of it. We have common helper | |
1026 | * routines to allocate them. | |
1027 | * | |
1028 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
17cbca2b RR |
1029 | * table just above the Guest's normal memory. It returns a pointer to that |
1030 | * descriptor. */ | |
1031 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
8ca47e00 | 1032 | { |
17cbca2b | 1033 | struct lguest_device_desc *d; |
8ca47e00 | 1034 | |
17cbca2b RR |
1035 | /* We only have one page for all the descriptors. */ |
1036 | if (devices.desc_used + sizeof(*d) > getpagesize()) | |
1037 | errx(1, "Too many devices"); | |
1038 | ||
1039 | /* We don't need to set config_len or status: page is 0 already. */ | |
1040 | d = (void *)devices.descpage + devices.desc_used; | |
1041 | d->type = type; | |
1042 | devices.desc_used += sizeof(*d); | |
1043 | ||
1044 | return d; | |
1045 | } | |
1046 | ||
1047 | /* Each device descriptor is followed by some configuration information. | |
1048 | * The first byte is a "status" byte for the Guest to report what's happening. | |
1049 | * After that are fields: u8 type, u8 len, [... len bytes...]. | |
1050 | * | |
1051 | * This routine adds a new field to an existing device's descriptor. It only | |
1052 | * works for the last device, but that's OK because that's how we use it. */ | |
1053 | static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c) | |
1054 | { | |
1055 | /* This is the last descriptor, right? */ | |
1056 | assert(devices.descpage + devices.desc_used | |
1057 | == (u8 *)(dev->desc + 1) + dev->desc->config_len); | |
1058 | ||
1059 | /* We only have one page of device descriptions. */ | |
1060 | if (devices.desc_used + 2 + len > getpagesize()) | |
1061 | errx(1, "Too many devices"); | |
1062 | ||
1063 | /* Copy in the new config header: type then length. */ | |
1064 | devices.descpage[devices.desc_used++] = type; | |
1065 | devices.descpage[devices.desc_used++] = len; | |
1066 | memcpy(devices.descpage + devices.desc_used, c, len); | |
1067 | devices.desc_used += len; | |
1068 | ||
1069 | /* Update the device descriptor length: two byte head then data. */ | |
1070 | dev->desc->config_len += 2 + len; | |
1071 | } | |
1072 | ||
1073 | /* This routine adds a virtqueue to a device. We specify how many descriptors | |
1074 | * the virtqueue is to have. */ | |
1075 | static void add_virtqueue(struct device *dev, unsigned int num_descs, | |
1076 | void (*handle_output)(int fd, struct virtqueue *me)) | |
1077 | { | |
1078 | unsigned int pages; | |
1079 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1080 | void *p; | |
1081 | ||
1082 | /* First we need some pages for this virtqueue. */ | |
1083 | pages = (vring_size(num_descs) + getpagesize() - 1) / getpagesize(); | |
1084 | p = get_pages(pages); | |
1085 | ||
1086 | /* Initialize the configuration. */ | |
1087 | vq->config.num = num_descs; | |
1088 | vq->config.irq = devices.next_irq++; | |
1089 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1090 | ||
1091 | /* Initialize the vring. */ | |
1092 | vring_init(&vq->vring, num_descs, p); | |
1093 | ||
1094 | /* Add the configuration information to this device's descriptor. */ | |
1095 | add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE, | |
1096 | sizeof(vq->config), &vq->config); | |
1097 | ||
1098 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1099 | * second. */ | |
1100 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1101 | *i = vq; | |
1102 | ||
1103 | /* Link virtqueue back to device. */ | |
1104 | vq->dev = dev; | |
1105 | ||
1106 | /* Set up handler. */ | |
1107 | vq->handle_output = handle_output; | |
1108 | if (!handle_output) | |
1109 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1110 | } |
1111 | ||
17cbca2b RR |
1112 | /* This routine does all the creation and setup of a new device, including |
1113 | * caling new_dev_desc() to allocate the descriptor and device memory. */ | |
1114 | static struct device *new_device(const char *name, u16 type, int fd, | |
1115 | bool (*handle_input)(int, struct device *)) | |
8ca47e00 RR |
1116 | { |
1117 | struct device *dev = malloc(sizeof(*dev)); | |
1118 | ||
dde79789 RR |
1119 | /* Append to device list. Prepending to a single-linked list is |
1120 | * easier, but the user expects the devices to be arranged on the bus | |
1121 | * in command-line order. The first network device on the command line | |
1122 | * is eth0, the first block device /dev/lgba, etc. */ | |
17cbca2b | 1123 | *devices.lastdev = dev; |
8ca47e00 | 1124 | dev->next = NULL; |
17cbca2b | 1125 | devices.lastdev = &dev->next; |
8ca47e00 | 1126 | |
dde79789 | 1127 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1128 | dev->fd = fd; |
dde79789 RR |
1129 | /* If we have an input handler for this file descriptor, then we add it |
1130 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1131 | if (handle_input) |
17cbca2b RR |
1132 | add_device_fd(dev->fd); |
1133 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1134 | dev->handle_input = handle_input; |
17cbca2b | 1135 | dev->name = name; |
8ca47e00 RR |
1136 | return dev; |
1137 | } | |
1138 | ||
dde79789 RR |
1139 | /* Our first setup routine is the console. It's a fairly simple device, but |
1140 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1141 | static void setup_console(void) |
8ca47e00 RR |
1142 | { |
1143 | struct device *dev; | |
1144 | ||
dde79789 | 1145 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1146 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1147 | struct termios term = orig_term; | |
dde79789 RR |
1148 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1149 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1150 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1151 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1152 | /* If we exit gracefully, the original settings will be |
1153 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1154 | atexit(restore_term); |
1155 | } | |
1156 | ||
17cbca2b RR |
1157 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1158 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1159 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1160 | dev->priv = malloc(sizeof(struct console_abort)); |
1161 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1162 | |
56ae43df RR |
1163 | /* The console needs two virtqueues: the input then the output. When |
1164 | * they put something the input queue, we make sure we're listening to | |
1165 | * stdin. When they put something in the output queue, we write it to | |
1166 | * stdout. */ | |
1167 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
17cbca2b RR |
1168 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1169 | ||
1170 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1171 | } |
17cbca2b | 1172 | /*:*/ |
8ca47e00 | 1173 | |
17cbca2b RR |
1174 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1175 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1176 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1177 | * |
17cbca2b RR |
1178 | * More sopisticated is to use one of the tools developed for project like UML |
1179 | * to do networking. | |
dde79789 | 1180 | * |
17cbca2b RR |
1181 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1182 | * completely generic ("here's my vring, attach to your vring") and would work | |
1183 | * for any traffic. Of course, namespace and permissions issues need to be | |
1184 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1185 | * multiple inter-guest channels behind one interface, although it would | |
1186 | * require some manner of hotplugging new virtio channels. | |
1187 | * | |
1188 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1189 | |
1190 | static u32 str2ip(const char *ipaddr) | |
1191 | { | |
1192 | unsigned int byte[4]; | |
1193 | ||
1194 | sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]); | |
1195 | return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3]; | |
1196 | } | |
1197 | ||
dde79789 RR |
1198 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1199 | * network device to the bridge device specified by the command line. | |
1200 | * | |
1201 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1202 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1203 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1204 | { | |
1205 | int ifidx; | |
1206 | struct ifreq ifr; | |
1207 | ||
1208 | if (!*br_name) | |
1209 | errx(1, "must specify bridge name"); | |
1210 | ||
1211 | ifidx = if_nametoindex(if_name); | |
1212 | if (!ifidx) | |
1213 | errx(1, "interface %s does not exist!", if_name); | |
1214 | ||
1215 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
1216 | ifr.ifr_ifindex = ifidx; | |
1217 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1218 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1219 | } | |
1220 | ||
dde79789 RR |
1221 | /* This sets up the Host end of the network device with an IP address, brings |
1222 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1223 | * pointer. */ |
8ca47e00 RR |
1224 | static void configure_device(int fd, const char *devname, u32 ipaddr, |
1225 | unsigned char hwaddr[6]) | |
1226 | { | |
1227 | struct ifreq ifr; | |
1228 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1229 | ||
dde79789 | 1230 | /* Don't read these incantations. Just cut & paste them like I did! */ |
8ca47e00 RR |
1231 | memset(&ifr, 0, sizeof(ifr)); |
1232 | strcpy(ifr.ifr_name, devname); | |
1233 | sin->sin_family = AF_INET; | |
1234 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1235 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
1236 | err(1, "Setting %s interface address", devname); | |
1237 | ifr.ifr_flags = IFF_UP; | |
1238 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
1239 | err(1, "Bringing interface %s up", devname); | |
1240 | ||
dde79789 RR |
1241 | /* SIOC stands for Socket I/O Control. G means Get (vs S for Set |
1242 | * above). IF means Interface, and HWADDR is hardware address. | |
1243 | * Simple! */ | |
8ca47e00 RR |
1244 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0) |
1245 | err(1, "getting hw address for %s", devname); | |
8ca47e00 RR |
1246 | memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); |
1247 | } | |
1248 | ||
17cbca2b RR |
1249 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or |
1250 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1251 | * packets into the Host as if they came in from a normal network card. We | |
1252 | * just shunt packets between the Guest and the tun device. */ | |
1253 | static void setup_tun_net(const char *arg) | |
8ca47e00 RR |
1254 | { |
1255 | struct device *dev; | |
1256 | struct ifreq ifr; | |
1257 | int netfd, ipfd; | |
1258 | u32 ip; | |
1259 | const char *br_name = NULL; | |
17cbca2b | 1260 | u8 hwaddr[6]; |
8ca47e00 | 1261 | |
dde79789 RR |
1262 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1263 | * tap device is like a tun device, only somehow different. To tell | |
1264 | * the truth, I completely blundered my way through this code, but it | |
1265 | * works now! */ | |
8ca47e00 RR |
1266 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
1267 | memset(&ifr, 0, sizeof(ifr)); | |
1268 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
1269 | strcpy(ifr.ifr_name, "tap%d"); | |
1270 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1271 | err(1, "configuring /dev/net/tun"); | |
dde79789 RR |
1272 | /* We don't need checksums calculated for packets coming in this |
1273 | * device: trust us! */ | |
8ca47e00 RR |
1274 | ioctl(netfd, TUNSETNOCSUM, 1); |
1275 | ||
17cbca2b RR |
1276 | /* First we create a new network device. */ |
1277 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1278 | |
56ae43df RR |
1279 | /* Network devices need a receive and a send queue, just like |
1280 | * console. */ | |
1281 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
17cbca2b | 1282 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1283 | |
dde79789 RR |
1284 | /* We need a socket to perform the magic network ioctls to bring up the |
1285 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1286 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1287 | if (ipfd < 0) | |
1288 | err(1, "opening IP socket"); | |
1289 | ||
dde79789 | 1290 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 RR |
1291 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
1292 | ip = INADDR_ANY; | |
1293 | br_name = arg + strlen(BRIDGE_PFX); | |
1294 | add_to_bridge(ipfd, ifr.ifr_name, br_name); | |
dde79789 | 1295 | } else /* It is an IP address to set up the device with */ |
8ca47e00 RR |
1296 | ip = str2ip(arg); |
1297 | ||
17cbca2b RR |
1298 | /* Set up the tun device, and get the mac address for the interface. */ |
1299 | configure_device(ipfd, ifr.ifr_name, ip, hwaddr); | |
8ca47e00 | 1300 | |
17cbca2b RR |
1301 | /* Tell Guest what MAC address to use. */ |
1302 | add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr); | |
8ca47e00 | 1303 | |
17cbca2b | 1304 | /* We don't seed the socket any more; setup is done. */ |
8ca47e00 RR |
1305 | close(ipfd); |
1306 | ||
17cbca2b RR |
1307 | verbose("device %u: tun net %u.%u.%u.%u\n", |
1308 | devices.device_num++, | |
1309 | (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip); | |
8ca47e00 RR |
1310 | if (br_name) |
1311 | verbose("attached to bridge: %s\n", br_name); | |
1312 | } | |
17cbca2b RR |
1313 | |
1314 | ||
1315 | /* | |
1316 | * Block device. | |
1317 | * | |
1318 | * Serving a block device is really easy: the Guest asks for a block number and | |
1319 | * we read or write that position in the file. | |
1320 | * | |
1321 | * Unfortunately, this is amazingly slow: the Guest waits until the read is | |
1322 | * finished before running anything else, even if it could be doing useful | |
1323 | * work. We could use async I/O, except it's reputed to suck so hard that | |
1324 | * characters actually go missing from your code when you try to use it. | |
1325 | * | |
1326 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1327 | ||
1328 | /* This hangs off device->priv, with the data. */ | |
1329 | struct vblk_info | |
1330 | { | |
1331 | /* The size of the file. */ | |
1332 | off64_t len; | |
1333 | ||
1334 | /* The file descriptor for the file. */ | |
1335 | int fd; | |
1336 | ||
1337 | /* IO thread listens on this file descriptor [0]. */ | |
1338 | int workpipe[2]; | |
1339 | ||
1340 | /* IO thread writes to this file descriptor to mark it done, then | |
1341 | * Launcher triggers interrupt to Guest. */ | |
1342 | int done_fd; | |
1343 | }; | |
1344 | ||
1345 | /* This is the core of the I/O thread. It returns true if it did something. */ | |
1346 | static bool service_io(struct device *dev) | |
1347 | { | |
1348 | struct vblk_info *vblk = dev->priv; | |
1349 | unsigned int head, out_num, in_num, wlen; | |
1350 | int ret; | |
1351 | struct virtio_blk_inhdr *in; | |
1352 | struct virtio_blk_outhdr *out; | |
1353 | struct iovec iov[dev->vq->vring.num]; | |
1354 | off64_t off; | |
1355 | ||
1356 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
1357 | if (head == dev->vq->vring.num) | |
1358 | return false; | |
1359 | ||
1360 | if (out_num == 0 || in_num == 0) | |
1361 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1362 | head, out_num, in_num); | |
1363 | ||
1364 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
1365 | in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr); | |
1366 | off = out->sector * 512; | |
1367 | ||
1368 | /* This is how we implement barriers. Pretty poor, no? */ | |
1369 | if (out->type & VIRTIO_BLK_T_BARRIER) | |
1370 | fdatasync(vblk->fd); | |
1371 | ||
1372 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { | |
1373 | fprintf(stderr, "Scsi commands unsupported\n"); | |
1374 | in->status = VIRTIO_BLK_S_UNSUPP; | |
1375 | wlen = sizeof(in); | |
1376 | } else if (out->type & VIRTIO_BLK_T_OUT) { | |
1377 | /* Write */ | |
1378 | ||
1379 | /* Move to the right location in the block file. This can fail | |
1380 | * if they try to write past end. */ | |
1381 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1382 | err(1, "Bad seek to sector %llu", out->sector); | |
1383 | ||
1384 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1385 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1386 | ||
1387 | /* Grr... Now we know how long the descriptor they sent was, we | |
1388 | * make sure they didn't try to write over the end of the block | |
1389 | * file (possibly extending it). */ | |
1390 | if (ret > 0 && off + ret > vblk->len) { | |
1391 | /* Trim it back to the correct length */ | |
1392 | ftruncate64(vblk->fd, vblk->len); | |
1393 | /* Die, bad Guest, die. */ | |
1394 | errx(1, "Write past end %llu+%u", off, ret); | |
1395 | } | |
1396 | wlen = sizeof(in); | |
1397 | in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
1398 | } else { | |
1399 | /* Read */ | |
1400 | ||
1401 | /* Move to the right location in the block file. This can fail | |
1402 | * if they try to read past end. */ | |
1403 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1404 | err(1, "Bad seek to sector %llu", out->sector); | |
1405 | ||
1406 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1407 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1408 | if (ret >= 0) { | |
1409 | wlen = sizeof(in) + ret; | |
1410 | in->status = VIRTIO_BLK_S_OK; | |
1411 | } else { | |
1412 | wlen = sizeof(in); | |
1413 | in->status = VIRTIO_BLK_S_IOERR; | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | /* We can't trigger an IRQ, because we're not the Launcher. It does | |
1418 | * that when we tell it we're done. */ | |
1419 | add_used(dev->vq, head, wlen); | |
1420 | return true; | |
1421 | } | |
1422 | ||
1423 | /* This is the thread which actually services the I/O. */ | |
1424 | static int io_thread(void *_dev) | |
1425 | { | |
1426 | struct device *dev = _dev; | |
1427 | struct vblk_info *vblk = dev->priv; | |
1428 | char c; | |
1429 | ||
1430 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1431 | close(vblk->workpipe[1]); | |
1432 | /* Close the other side of the done_fd pipe. */ | |
1433 | close(dev->fd); | |
1434 | ||
1435 | /* When this read fails, it means Launcher died, so we follow. */ | |
1436 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
1437 | /* We acknowledge each request immediately, to reduce latency, | |
1438 | * rather than waiting until we've done them all. I haven't | |
1439 | * measured to see if it makes any difference. */ | |
1440 | while (service_io(dev)) | |
1441 | write(vblk->done_fd, &c, 1); | |
1442 | } | |
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | /* When the thread says some I/O is done, we interrupt the Guest. */ | |
1447 | static bool handle_io_finish(int fd, struct device *dev) | |
1448 | { | |
1449 | char c; | |
1450 | ||
1451 | /* If child died, presumably it printed message. */ | |
1452 | if (read(dev->fd, &c, 1) != 1) | |
1453 | exit(1); | |
1454 | ||
1455 | /* It did some work, so trigger the irq. */ | |
1456 | trigger_irq(fd, dev->vq); | |
1457 | return true; | |
1458 | } | |
1459 | ||
1460 | /* When the Guest submits some I/O, we wake the I/O thread. */ | |
1461 | static void handle_virtblk_output(int fd, struct virtqueue *vq) | |
1462 | { | |
1463 | struct vblk_info *vblk = vq->dev->priv; | |
1464 | char c = 0; | |
1465 | ||
1466 | /* Wake up I/O thread and tell it to go to work! */ | |
1467 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1468 | /* Presumably it indicated why it died. */ | |
1469 | exit(1); | |
1470 | } | |
1471 | ||
1472 | /* This creates a virtual block device. */ | |
1473 | static void setup_block_file(const char *filename) | |
1474 | { | |
1475 | int p[2]; | |
1476 | struct device *dev; | |
1477 | struct vblk_info *vblk; | |
1478 | void *stack; | |
1479 | u64 cap; | |
1480 | unsigned int val; | |
1481 | ||
1482 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1483 | pipe(p); | |
1484 | ||
1485 | /* The device responds to return from I/O thread. */ | |
1486 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1487 | ||
1488 | /* The device has a virtqueue. */ | |
1489 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); | |
1490 | ||
1491 | /* Allocate the room for our own bookkeeping */ | |
1492 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1493 | ||
1494 | /* First we open the file and store the length. */ | |
1495 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1496 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1497 | ||
1498 | /* Tell Guest how many sectors this device has. */ | |
1499 | cap = cpu_to_le64(vblk->len / 512); | |
1500 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap); | |
1501 | ||
1502 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1503 | * for the in and out elements. */ | |
1504 | val = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1505 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val); | |
1506 | ||
1507 | /* The I/O thread writes to this end of the pipe when done. */ | |
1508 | vblk->done_fd = p[1]; | |
1509 | ||
1510 | /* This is how we tell the I/O thread about more work. */ | |
1511 | pipe(vblk->workpipe); | |
1512 | ||
1513 | /* Create stack for thread and run it */ | |
1514 | stack = malloc(32768); | |
1515 | if (clone(io_thread, stack + 32768, CLONE_VM, dev) == -1) | |
1516 | err(1, "Creating clone"); | |
1517 | ||
1518 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1519 | close(vblk->done_fd); | |
1520 | close(vblk->workpipe[0]); | |
1521 | ||
1522 | verbose("device %u: virtblock %llu sectors\n", | |
1523 | devices.device_num, cap); | |
1524 | } | |
dde79789 | 1525 | /* That's the end of device setup. */ |
8ca47e00 | 1526 | |
dde79789 RR |
1527 | /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves |
1528 | * its input and output, and finally, lays it to rest. */ | |
17cbca2b | 1529 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
8ca47e00 RR |
1530 | { |
1531 | for (;;) { | |
511801dc | 1532 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1533 | unsigned long notify_addr; |
8ca47e00 RR |
1534 | int readval; |
1535 | ||
1536 | /* We read from the /dev/lguest device to run the Guest. */ | |
17cbca2b | 1537 | readval = read(lguest_fd, ¬ify_addr, sizeof(notify_addr)); |
8ca47e00 | 1538 | |
17cbca2b RR |
1539 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1540 | if (readval == sizeof(notify_addr)) { | |
1541 | verbose("Notify on address %#lx\n", notify_addr); | |
1542 | handle_output(lguest_fd, notify_addr); | |
8ca47e00 | 1543 | continue; |
dde79789 | 1544 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1545 | } else if (errno == ENOENT) { |
1546 | char reason[1024] = { 0 }; | |
1547 | read(lguest_fd, reason, sizeof(reason)-1); | |
1548 | errx(1, "%s", reason); | |
dde79789 RR |
1549 | /* EAGAIN means the waker wanted us to look at some input. |
1550 | * Anything else means a bug or incompatible change. */ | |
8ca47e00 RR |
1551 | } else if (errno != EAGAIN) |
1552 | err(1, "Running guest failed"); | |
dde79789 RR |
1553 | |
1554 | /* Service input, then unset the BREAK which releases | |
1555 | * the Waker. */ | |
17cbca2b | 1556 | handle_input(lguest_fd); |
8ca47e00 RR |
1557 | if (write(lguest_fd, args, sizeof(args)) < 0) |
1558 | err(1, "Resetting break"); | |
1559 | } | |
1560 | } | |
dde79789 RR |
1561 | /* |
1562 | * This is the end of the Launcher. | |
1563 | * | |
1564 | * But wait! We've seen I/O from the Launcher, and we've seen I/O from the | |
1565 | * Drivers. If we were to see the Host kernel I/O code, our understanding | |
1566 | * would be complete... :*/ | |
8ca47e00 RR |
1567 | |
1568 | static struct option opts[] = { | |
1569 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1570 | { "tunnet", 1, NULL, 't' }, |
1571 | { "block", 1, NULL, 'b' }, | |
1572 | { "initrd", 1, NULL, 'i' }, | |
1573 | { NULL }, | |
1574 | }; | |
1575 | static void usage(void) | |
1576 | { | |
1577 | errx(1, "Usage: lguest [--verbose] " | |
17cbca2b | 1578 | "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n" |
8ca47e00 RR |
1579 | "|--block=<filename>|--initrd=<filename>]...\n" |
1580 | "<mem-in-mb> vmlinux [args...]"); | |
1581 | } | |
1582 | ||
3c6b5bfa | 1583 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1584 | int main(int argc, char *argv[]) |
1585 | { | |
47436aa4 RR |
1586 | /* Memory, top-level pagetable, code startpoint and size of the |
1587 | * (optional) initrd. */ | |
1588 | unsigned long mem = 0, pgdir, start, initrd_size = 0; | |
dde79789 | 1589 | /* A temporary and the /dev/lguest file descriptor. */ |
6570c459 | 1590 | int i, c, lguest_fd; |
3c6b5bfa RR |
1591 | /* The boot information for the Guest. */ |
1592 | void *boot; | |
dde79789 | 1593 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1594 | const char *initrd_name = NULL; |
1595 | ||
dde79789 RR |
1596 | /* First we initialize the device list. Since console and network |
1597 | * device receive input from a file descriptor, we keep an fdset | |
1598 | * (infds) and the maximum fd number (max_infd) with the head of the | |
1599 | * list. We also keep a pointer to the last device, for easy appending | |
17cbca2b RR |
1600 | * to the list. Finally, we keep the next interrupt number to hand out |
1601 | * (1: remember that 0 is used by the timer). */ | |
1602 | FD_ZERO(&devices.infds); | |
1603 | devices.max_infd = -1; | |
1604 | devices.lastdev = &devices.dev; | |
1605 | devices.next_irq = 1; | |
8ca47e00 | 1606 | |
dde79789 RR |
1607 | /* We need to know how much memory so we can set up the device |
1608 | * descriptor and memory pages for the devices as we parse the command | |
1609 | * line. So we quickly look through the arguments to find the amount | |
1610 | * of memory now. */ | |
6570c459 RR |
1611 | for (i = 1; i < argc; i++) { |
1612 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1613 | mem = atoi(argv[i]) * 1024 * 1024; |
1614 | /* We start by mapping anonymous pages over all of | |
1615 | * guest-physical memory range. This fills it with 0, | |
1616 | * and ensures that the Guest won't be killed when it | |
1617 | * tries to access it. */ | |
1618 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1619 | + DEVICE_PAGES); | |
1620 | guest_limit = mem; | |
1621 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1622 | devices.descpage = get_pages(1); |
6570c459 RR |
1623 | break; |
1624 | } | |
1625 | } | |
dde79789 RR |
1626 | |
1627 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1628 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1629 | switch (c) { | |
1630 | case 'v': | |
1631 | verbose = true; | |
1632 | break; | |
8ca47e00 | 1633 | case 't': |
17cbca2b | 1634 | setup_tun_net(optarg); |
8ca47e00 RR |
1635 | break; |
1636 | case 'b': | |
17cbca2b | 1637 | setup_block_file(optarg); |
8ca47e00 RR |
1638 | break; |
1639 | case 'i': | |
1640 | initrd_name = optarg; | |
1641 | break; | |
1642 | default: | |
1643 | warnx("Unknown argument %s", argv[optind]); | |
1644 | usage(); | |
1645 | } | |
1646 | } | |
dde79789 RR |
1647 | /* After the other arguments we expect memory and kernel image name, |
1648 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1649 | if (optind + 2 > argc) |
1650 | usage(); | |
1651 | ||
3c6b5bfa RR |
1652 | verbose("Guest base is at %p\n", guest_base); |
1653 | ||
dde79789 | 1654 | /* We always have a console device */ |
17cbca2b | 1655 | setup_console(); |
8ca47e00 | 1656 | |
8ca47e00 | 1657 | /* Now we load the kernel */ |
47436aa4 | 1658 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1659 | |
3c6b5bfa RR |
1660 | /* Boot information is stashed at physical address 0 */ |
1661 | boot = from_guest_phys(0); | |
1662 | ||
dde79789 | 1663 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1664 | if (initrd_name) { |
1665 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1666 | /* These are the location in the Linux boot header where the |
1667 | * start and size of the initrd are expected to be found. */ | |
8ca47e00 RR |
1668 | *(unsigned long *)(boot+0x218) = mem - initrd_size; |
1669 | *(unsigned long *)(boot+0x21c) = initrd_size; | |
dde79789 | 1670 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
8ca47e00 RR |
1671 | *(unsigned char *)(boot+0x210) = 0xFF; |
1672 | } | |
1673 | ||
dde79789 | 1674 | /* Set up the initial linear pagetables, starting below the initrd. */ |
47436aa4 | 1675 | pgdir = setup_pagetables(mem, initrd_size); |
8ca47e00 | 1676 | |
dde79789 RR |
1677 | /* The Linux boot header contains an "E820" memory map: ours is a |
1678 | * simple, single region. */ | |
8ca47e00 RR |
1679 | *(char*)(boot+E820NR) = 1; |
1680 | *((struct e820entry *)(boot+E820MAP)) | |
1681 | = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 RR |
1682 | /* The boot header contains a command line pointer: we put the command |
1683 | * line after the boot header (at address 4096) */ | |
3c6b5bfa | 1684 | *(u32 *)(boot + 0x228) = 4096; |
8ca47e00 | 1685 | concat(boot + 4096, argv+optind+2); |
dde79789 RR |
1686 | |
1687 | /* The guest type value of "1" tells the Guest it's under lguest. */ | |
8ca47e00 RR |
1688 | *(int *)(boot + 0x23c) = 1; |
1689 | ||
dde79789 RR |
1690 | /* We tell the kernel to initialize the Guest: this returns the open |
1691 | * /dev/lguest file descriptor. */ | |
47436aa4 | 1692 | lguest_fd = tell_kernel(pgdir, start); |
dde79789 RR |
1693 | |
1694 | /* We fork off a child process, which wakes the Launcher whenever one | |
1695 | * of the input file descriptors needs attention. Otherwise we would | |
1696 | * run the Guest until it tries to output something. */ | |
17cbca2b | 1697 | waker_fd = setup_waker(lguest_fd); |
8ca47e00 | 1698 | |
dde79789 | 1699 | /* Finally, run the Guest. This doesn't return. */ |
17cbca2b | 1700 | run_guest(lguest_fd); |
8ca47e00 | 1701 | } |
f56a384e RR |
1702 | /*:*/ |
1703 | ||
1704 | /*M:999 | |
1705 | * Mastery is done: you now know everything I do. | |
1706 | * | |
1707 | * But surely you have seen code, features and bugs in your wanderings which | |
1708 | * you now yearn to attack? That is the real game, and I look forward to you | |
1709 | * patching and forking lguest into the Your-Name-Here-visor. | |
1710 | * | |
1711 | * Farewell, and good coding! | |
1712 | * Rusty Russell. | |
1713 | */ |