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> | |
b45d8cb0 | 37 | #include "linux/lguest_launcher.h" |
17cbca2b RR |
38 | #include "linux/virtio_config.h" |
39 | #include "linux/virtio_net.h" | |
40 | #include "linux/virtio_blk.h" | |
41 | #include "linux/virtio_console.h" | |
42 | #include "linux/virtio_ring.h" | |
43d33b21 | 43 | #include "asm-x86/bootparam.h" |
db24e8c2 RR |
44 | /*L:110 We can ignore the 38 include files we need for this program, but I do |
45 | * want to draw attention to the use of kernel-style types. | |
46 | * | |
47 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
48 | * like these abbreviations, so we define them here. Note that u64 is always | |
49 | * unsigned long long, which works on all Linux systems: this means that we can | |
50 | * use %llu in printf for any u64. */ | |
51 | typedef unsigned long long u64; | |
52 | typedef uint32_t u32; | |
53 | typedef uint16_t u16; | |
54 | typedef uint8_t u8; | |
dde79789 | 55 | /*:*/ |
8ca47e00 RR |
56 | |
57 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
58 | #define NET_PEERNUM 1 | |
59 | #define BRIDGE_PFX "bridge:" | |
60 | #ifndef SIOCBRADDIF | |
61 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
62 | #endif | |
3c6b5bfa RR |
63 | /* We can have up to 256 pages for devices. */ |
64 | #define DEVICE_PAGES 256 | |
17cbca2b RR |
65 | /* This fits nicely in a single 4096-byte page. */ |
66 | #define VIRTQUEUE_NUM 127 | |
8ca47e00 | 67 | |
dde79789 RR |
68 | /*L:120 verbose is both a global flag and a macro. The C preprocessor allows |
69 | * this, and although I wouldn't recommend it, it works quite nicely here. */ | |
8ca47e00 RR |
70 | static bool verbose; |
71 | #define verbose(args...) \ | |
72 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
73 | /*:*/ |
74 | ||
75 | /* The pipe to send commands to the waker process */ | |
8ca47e00 | 76 | static int waker_fd; |
3c6b5bfa RR |
77 | /* The pointer to the start of guest memory. */ |
78 | static void *guest_base; | |
79 | /* The maximum guest physical address allowed, and maximum possible. */ | |
80 | static unsigned long guest_limit, guest_max; | |
8ca47e00 | 81 | |
dde79789 | 82 | /* This is our list of devices. */ |
8ca47e00 RR |
83 | struct device_list |
84 | { | |
dde79789 RR |
85 | /* Summary information about the devices in our list: ready to pass to |
86 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
87 | fd_set infds; |
88 | int max_infd; | |
89 | ||
17cbca2b RR |
90 | /* Counter to assign interrupt numbers. */ |
91 | unsigned int next_irq; | |
92 | ||
93 | /* Counter to print out convenient device numbers. */ | |
94 | unsigned int device_num; | |
95 | ||
dde79789 | 96 | /* The descriptor page for the devices. */ |
17cbca2b RR |
97 | u8 *descpage; |
98 | ||
99 | /* The tail of the last descriptor. */ | |
100 | unsigned int desc_used; | |
dde79789 RR |
101 | |
102 | /* A single linked list of devices. */ | |
8ca47e00 | 103 | struct device *dev; |
dde79789 | 104 | /* ... And an end pointer so we can easily append new devices */ |
8ca47e00 RR |
105 | struct device **lastdev; |
106 | }; | |
107 | ||
17cbca2b RR |
108 | /* The list of Guest devices, based on command line arguments. */ |
109 | static struct device_list devices; | |
110 | ||
dde79789 | 111 | /* The device structure describes a single device. */ |
8ca47e00 RR |
112 | struct device |
113 | { | |
dde79789 | 114 | /* The linked-list pointer. */ |
8ca47e00 | 115 | struct device *next; |
17cbca2b RR |
116 | |
117 | /* The this device's descriptor, as mapped into the Guest. */ | |
8ca47e00 | 118 | struct lguest_device_desc *desc; |
17cbca2b RR |
119 | |
120 | /* The name of this device, for --verbose. */ | |
121 | const char *name; | |
8ca47e00 | 122 | |
dde79789 RR |
123 | /* If handle_input is set, it wants to be called when this file |
124 | * descriptor is ready. */ | |
8ca47e00 RR |
125 | int fd; |
126 | bool (*handle_input)(int fd, struct device *me); | |
127 | ||
17cbca2b RR |
128 | /* Any queues attached to this device */ |
129 | struct virtqueue *vq; | |
8ca47e00 RR |
130 | |
131 | /* Device-specific data. */ | |
132 | void *priv; | |
133 | }; | |
134 | ||
17cbca2b RR |
135 | /* The virtqueue structure describes a queue attached to a device. */ |
136 | struct virtqueue | |
137 | { | |
138 | struct virtqueue *next; | |
139 | ||
140 | /* Which device owns me. */ | |
141 | struct device *dev; | |
142 | ||
143 | /* The configuration for this queue. */ | |
144 | struct lguest_vqconfig config; | |
145 | ||
146 | /* The actual ring of buffers. */ | |
147 | struct vring vring; | |
148 | ||
149 | /* Last available index we saw. */ | |
150 | u16 last_avail_idx; | |
151 | ||
152 | /* The routine to call when the Guest pings us. */ | |
153 | void (*handle_output)(int fd, struct virtqueue *me); | |
154 | }; | |
155 | ||
156 | /* Since guest is UP and we don't run at the same time, we don't need barriers. | |
157 | * But I include them in the code in case others copy it. */ | |
158 | #define wmb() | |
159 | ||
160 | /* Convert an iovec element to the given type. | |
161 | * | |
162 | * This is a fairly ugly trick: we need to know the size of the type and | |
163 | * alignment requirement to check the pointer is kosher. It's also nice to | |
164 | * have the name of the type in case we report failure. | |
165 | * | |
166 | * Typing those three things all the time is cumbersome and error prone, so we | |
167 | * have a macro which sets them all up and passes to the real function. */ | |
168 | #define convert(iov, type) \ | |
169 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
170 | ||
171 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
172 | const char *name) | |
173 | { | |
174 | if (iov->iov_len != size) | |
175 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
176 | if ((unsigned long)iov->iov_base % align != 0) | |
177 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
178 | return iov->iov_base; | |
179 | } | |
180 | ||
181 | /* The virtio configuration space is defined to be little-endian. x86 is | |
182 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
183 | #define cpu_to_le16(v16) (v16) | |
184 | #define cpu_to_le32(v32) (v32) | |
185 | #define cpu_to_le64(v64) (v64) | |
186 | #define le16_to_cpu(v16) (v16) | |
187 | #define le32_to_cpu(v32) (v32) | |
188 | #define le64_to_cpu(v32) (v64) | |
189 | ||
3c6b5bfa RR |
190 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
191 | * where pointers run wild and free! Unfortunately, like most userspace | |
192 | * programs, it's quite boring (which is why everyone likes to hack on the | |
193 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
194 | * will get you through this section. Or, maybe not. | |
195 | * | |
196 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
197 | * memory and stores it in "guest_base". In other words, Guest physical == | |
198 | * Launcher virtual with an offset. | |
199 | * | |
200 | * This can be tough to get your head around, but usually it just means that we | |
201 | * use these trivial conversion functions when the Guest gives us it's | |
202 | * "physical" addresses: */ | |
203 | static void *from_guest_phys(unsigned long addr) | |
204 | { | |
205 | return guest_base + addr; | |
206 | } | |
207 | ||
208 | static unsigned long to_guest_phys(const void *addr) | |
209 | { | |
210 | return (addr - guest_base); | |
211 | } | |
212 | ||
dde79789 RR |
213 | /*L:130 |
214 | * Loading the Kernel. | |
215 | * | |
216 | * We start with couple of simple helper routines. open_or_die() avoids | |
217 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
218 | static int open_or_die(const char *name, int flags) |
219 | { | |
220 | int fd = open(name, flags); | |
221 | if (fd < 0) | |
222 | err(1, "Failed to open %s", name); | |
223 | return fd; | |
224 | } | |
225 | ||
3c6b5bfa RR |
226 | /* map_zeroed_pages() takes a number of pages. */ |
227 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 228 | { |
3c6b5bfa RR |
229 | int fd = open_or_die("/dev/zero", O_RDONLY); |
230 | void *addr; | |
8ca47e00 | 231 | |
dde79789 | 232 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
233 | * copied). */ |
234 | addr = mmap(NULL, getpagesize() * num, | |
235 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
236 | if (addr == MAP_FAILED) | |
237 | err(1, "Mmaping %u pages of /dev/zero", num); | |
238 | ||
239 | return addr; | |
240 | } | |
241 | ||
242 | /* Get some more pages for a device. */ | |
243 | static void *get_pages(unsigned int num) | |
244 | { | |
245 | void *addr = from_guest_phys(guest_limit); | |
246 | ||
247 | guest_limit += num * getpagesize(); | |
248 | if (guest_limit > guest_max) | |
249 | errx(1, "Not enough memory for devices"); | |
250 | return addr; | |
8ca47e00 RR |
251 | } |
252 | ||
6649bb7a RM |
253 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
254 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
255 | * it falls back to reading the memory in. */ | |
256 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
257 | { | |
258 | ssize_t r; | |
259 | ||
260 | /* We map writable even though for some segments are marked read-only. | |
261 | * The kernel really wants to be writable: it patches its own | |
262 | * instructions. | |
263 | * | |
264 | * MAP_PRIVATE means that the page won't be copied until a write is | |
265 | * done to it. This allows us to share untouched memory between | |
266 | * Guests. */ | |
267 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
268 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
269 | return; | |
270 | ||
271 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
272 | r = pread(fd, addr, len, offset); | |
273 | if (r != len) | |
274 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
275 | } | |
276 | ||
dde79789 RR |
277 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
278 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
279 | * by all modern binaries on Linux including the kernel. | |
280 | * | |
281 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
282 | * address. We use the physical address; the Guest will map itself to the |
283 | * virtual address. | |
dde79789 RR |
284 | * |
285 | * We return the starting address. */ | |
47436aa4 | 286 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 287 | { |
8ca47e00 RR |
288 | Elf32_Phdr phdr[ehdr->e_phnum]; |
289 | unsigned int i; | |
8ca47e00 | 290 | |
dde79789 RR |
291 | /* Sanity checks on the main ELF header: an x86 executable with a |
292 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
293 | if (ehdr->e_type != ET_EXEC |
294 | || ehdr->e_machine != EM_386 | |
295 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
296 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
297 | errx(1, "Malformed elf header"); | |
298 | ||
dde79789 RR |
299 | /* An ELF executable contains an ELF header and a number of "program" |
300 | * headers which indicate which parts ("segments") of the program to | |
301 | * load where. */ | |
302 | ||
303 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
304 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
305 | err(1, "Seeking to program headers"); | |
306 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
307 | err(1, "Reading program headers"); | |
308 | ||
dde79789 RR |
309 | /* Try all the headers: there are usually only three. A read-only one, |
310 | * a read-write one, and a "note" section which isn't loadable. */ | |
8ca47e00 | 311 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 312 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
313 | if (phdr[i].p_type != PT_LOAD) |
314 | continue; | |
315 | ||
316 | verbose("Section %i: size %i addr %p\n", | |
317 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
318 | ||
6649bb7a | 319 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 320 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 321 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
322 | } |
323 | ||
814a0e5c RR |
324 | /* The entry point is given in the ELF header. */ |
325 | return ehdr->e_entry; | |
8ca47e00 RR |
326 | } |
327 | ||
dde79789 | 328 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
329 | * supposed to jump into it and it will unpack itself. We used to have to |
330 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 331 | * |
5bbf89fc RR |
332 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
333 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
334 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 335 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 336 | { |
43d33b21 | 337 | struct boot_params boot; |
5bbf89fc RR |
338 | int r; |
339 | /* Modern bzImages get loaded at 1M. */ | |
340 | void *p = from_guest_phys(0x100000); | |
341 | ||
342 | /* Go back to the start of the file and read the header. It should be | |
343 | * a Linux boot header (see Documentation/i386/boot.txt) */ | |
344 | lseek(fd, 0, SEEK_SET); | |
43d33b21 | 345 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 346 | |
43d33b21 RR |
347 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
348 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
349 | errx(1, "This doesn't look like a bzImage to me"); |
350 | ||
43d33b21 RR |
351 | /* Skip over the extra sectors of the header. */ |
352 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
353 | |
354 | /* Now read everything into memory. in nice big chunks. */ | |
355 | while ((r = read(fd, p, 65536)) > 0) | |
356 | p += r; | |
357 | ||
43d33b21 RR |
358 | /* Finally, code32_start tells us where to enter the kernel. */ |
359 | return boot.hdr.code32_start; | |
8ca47e00 RR |
360 | } |
361 | ||
dde79789 RR |
362 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
363 | * come wrapped up in the self-decompressing "bzImage" format. With some funky | |
364 | * coding, we can load those, too. */ | |
47436aa4 | 365 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
366 | { |
367 | Elf32_Ehdr hdr; | |
368 | ||
dde79789 | 369 | /* Read in the first few bytes. */ |
8ca47e00 RR |
370 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
371 | err(1, "Reading kernel"); | |
372 | ||
dde79789 | 373 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 374 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 375 | return map_elf(fd, &hdr); |
8ca47e00 | 376 | |
dde79789 | 377 | /* Otherwise we assume it's a bzImage, and try to unpack it */ |
47436aa4 | 378 | return load_bzimage(fd); |
8ca47e00 RR |
379 | } |
380 | ||
dde79789 RR |
381 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
382 | * it calls getpagesize() twice: "it's dumb code." | |
383 | * | |
384 | * Kernel guys get really het up about optimization, even when it's not | |
385 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
386 | static inline unsigned long page_align(unsigned long addr) |
387 | { | |
dde79789 | 388 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
389 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
390 | } | |
391 | ||
dde79789 RR |
392 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
393 | * the kernel which the kernel can use to boot from without needing any | |
394 | * drivers. Most distributions now use this as standard: the initrd contains | |
395 | * the code to load the appropriate driver modules for the current machine. | |
396 | * | |
397 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
398 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
399 | static unsigned long load_initrd(const char *name, unsigned long mem) |
400 | { | |
401 | int ifd; | |
402 | struct stat st; | |
403 | unsigned long len; | |
8ca47e00 RR |
404 | |
405 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 406 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
407 | if (fstat(ifd, &st) < 0) |
408 | err(1, "fstat() on initrd '%s'", name); | |
409 | ||
6649bb7a RM |
410 | /* We map the initrd at the top of memory, but mmap wants it to be |
411 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 412 | len = page_align(st.st_size); |
3c6b5bfa | 413 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
414 | /* Once a file is mapped, you can close the file descriptor. It's a |
415 | * little odd, but quite useful. */ | |
8ca47e00 | 416 | close(ifd); |
6649bb7a | 417 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
418 | |
419 | /* We return the initrd size. */ | |
8ca47e00 RR |
420 | return len; |
421 | } | |
422 | ||
47436aa4 RR |
423 | /* Once we know how much memory we have, we can construct simple linear page |
424 | * tables which set virtual == physical which will get the Guest far enough | |
3c6b5bfa | 425 | * into the boot to create its own. |
dde79789 RR |
426 | * |
427 | * We lay them out of the way, just below the initrd (which is why we need to | |
428 | * know its size). */ | |
8ca47e00 | 429 | static unsigned long setup_pagetables(unsigned long mem, |
47436aa4 | 430 | unsigned long initrd_size) |
8ca47e00 | 431 | { |
511801dc | 432 | unsigned long *pgdir, *linear; |
8ca47e00 | 433 | unsigned int mapped_pages, i, linear_pages; |
511801dc | 434 | unsigned int ptes_per_page = getpagesize()/sizeof(void *); |
8ca47e00 | 435 | |
47436aa4 | 436 | mapped_pages = mem/getpagesize(); |
8ca47e00 | 437 | |
dde79789 | 438 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
8ca47e00 RR |
439 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
440 | ||
dde79789 | 441 | /* We put the toplevel page directory page at the top of memory. */ |
3c6b5bfa | 442 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
dde79789 RR |
443 | |
444 | /* Now we use the next linear_pages pages as pte pages */ | |
8ca47e00 RR |
445 | linear = (void *)pgdir - linear_pages*getpagesize(); |
446 | ||
dde79789 RR |
447 | /* Linear mapping is easy: put every page's address into the mapping in |
448 | * order. PAGE_PRESENT contains the flags Present, Writable and | |
449 | * Executable. */ | |
8ca47e00 RR |
450 | for (i = 0; i < mapped_pages; i++) |
451 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); | |
452 | ||
47436aa4 | 453 | /* The top level points to the linear page table pages above. */ |
8ca47e00 | 454 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
47436aa4 | 455 | pgdir[i/ptes_per_page] |
511801dc | 456 | = ((to_guest_phys(linear) + i*sizeof(void *)) |
3c6b5bfa | 457 | | PAGE_PRESENT); |
8ca47e00 RR |
458 | } |
459 | ||
3c6b5bfa RR |
460 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
461 | mapped_pages, linear_pages, to_guest_phys(linear)); | |
8ca47e00 | 462 | |
dde79789 RR |
463 | /* We return the top level (guest-physical) address: the kernel needs |
464 | * to know where it is. */ | |
3c6b5bfa | 465 | return to_guest_phys(pgdir); |
8ca47e00 RR |
466 | } |
467 | ||
dde79789 RR |
468 | /* Simple routine to roll all the commandline arguments together with spaces |
469 | * between them. */ | |
8ca47e00 RR |
470 | static void concat(char *dst, char *args[]) |
471 | { | |
472 | unsigned int i, len = 0; | |
473 | ||
474 | for (i = 0; args[i]; i++) { | |
475 | strcpy(dst+len, args[i]); | |
476 | strcat(dst+len, " "); | |
477 | len += strlen(args[i]) + 1; | |
478 | } | |
479 | /* In case it's empty. */ | |
480 | dst[len] = '\0'; | |
481 | } | |
482 | ||
dde79789 RR |
483 | /* This is where we actually tell the kernel to initialize the Guest. We saw |
484 | * the arguments it expects when we looked at initialize() in lguest_user.c: | |
3c6b5bfa | 485 | * the base of guest "physical" memory, the top physical page to allow, the |
47436aa4 RR |
486 | * top level pagetable and the entry point for the Guest. */ |
487 | static int tell_kernel(unsigned long pgdir, unsigned long start) | |
8ca47e00 | 488 | { |
511801dc JS |
489 | unsigned long args[] = { LHREQ_INITIALIZE, |
490 | (unsigned long)guest_base, | |
47436aa4 | 491 | guest_limit / getpagesize(), pgdir, start }; |
8ca47e00 RR |
492 | int fd; |
493 | ||
3c6b5bfa RR |
494 | verbose("Guest: %p - %p (%#lx)\n", |
495 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
496 | fd = open_or_die("/dev/lguest", O_RDWR); |
497 | if (write(fd, args, sizeof(args)) < 0) | |
498 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
499 | |
500 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
501 | return fd; |
502 | } | |
dde79789 | 503 | /*:*/ |
8ca47e00 | 504 | |
17cbca2b | 505 | static void add_device_fd(int fd) |
8ca47e00 | 506 | { |
17cbca2b RR |
507 | FD_SET(fd, &devices.infds); |
508 | if (fd > devices.max_infd) | |
509 | devices.max_infd = fd; | |
8ca47e00 RR |
510 | } |
511 | ||
dde79789 RR |
512 | /*L:200 |
513 | * The Waker. | |
514 | * | |
515 | * With a console and network devices, we can have lots of input which we need | |
516 | * to process. We could try to tell the kernel what file descriptors to watch, | |
517 | * but handing a file descriptor mask through to the kernel is fairly icky. | |
518 | * | |
519 | * Instead, we fork off a process which watches the file descriptors and writes | |
520 | * the LHREQ_BREAK command to the /dev/lguest filedescriptor to tell the Host | |
521 | * loop to stop running the Guest. This causes it to return from the | |
522 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset | |
523 | * the LHREQ_BREAK and wake us up again. | |
524 | * | |
525 | * This, of course, is merely a different *kind* of icky. | |
526 | */ | |
17cbca2b | 527 | static void wake_parent(int pipefd, int lguest_fd) |
8ca47e00 | 528 | { |
dde79789 RR |
529 | /* Add the pipe from the Launcher to the fdset in the device_list, so |
530 | * we watch it, too. */ | |
17cbca2b | 531 | add_device_fd(pipefd); |
8ca47e00 RR |
532 | |
533 | for (;;) { | |
17cbca2b | 534 | fd_set rfds = devices.infds; |
511801dc | 535 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8ca47e00 | 536 | |
dde79789 | 537 | /* Wait until input is ready from one of the devices. */ |
17cbca2b | 538 | select(devices.max_infd+1, &rfds, NULL, NULL, NULL); |
dde79789 | 539 | /* Is it a message from the Launcher? */ |
8ca47e00 | 540 | if (FD_ISSET(pipefd, &rfds)) { |
56ae43df | 541 | int fd; |
dde79789 RR |
542 | /* If read() returns 0, it means the Launcher has |
543 | * exited. We silently follow. */ | |
56ae43df | 544 | if (read(pipefd, &fd, sizeof(fd)) == 0) |
8ca47e00 | 545 | exit(0); |
56ae43df RR |
546 | /* Otherwise it's telling us to change what file |
547 | * descriptors we're to listen to. */ | |
548 | if (fd >= 0) | |
549 | FD_SET(fd, &devices.infds); | |
550 | else | |
551 | FD_CLR(-fd - 1, &devices.infds); | |
dde79789 | 552 | } else /* Send LHREQ_BREAK command. */ |
8ca47e00 RR |
553 | write(lguest_fd, args, sizeof(args)); |
554 | } | |
555 | } | |
556 | ||
dde79789 | 557 | /* This routine just sets up a pipe to the Waker process. */ |
17cbca2b | 558 | static int setup_waker(int lguest_fd) |
8ca47e00 RR |
559 | { |
560 | int pipefd[2], child; | |
561 | ||
dde79789 RR |
562 | /* We create a pipe to talk to the waker, and also so it knows when the |
563 | * Launcher dies (and closes pipe). */ | |
8ca47e00 RR |
564 | pipe(pipefd); |
565 | child = fork(); | |
566 | if (child == -1) | |
567 | err(1, "forking"); | |
568 | ||
569 | if (child == 0) { | |
dde79789 | 570 | /* Close the "writing" end of our copy of the pipe */ |
8ca47e00 | 571 | close(pipefd[1]); |
17cbca2b | 572 | wake_parent(pipefd[0], lguest_fd); |
8ca47e00 | 573 | } |
dde79789 | 574 | /* Close the reading end of our copy of the pipe. */ |
8ca47e00 RR |
575 | close(pipefd[0]); |
576 | ||
dde79789 | 577 | /* Here is the fd used to talk to the waker. */ |
8ca47e00 RR |
578 | return pipefd[1]; |
579 | } | |
580 | ||
dde79789 RR |
581 | /*L:210 |
582 | * Device Handling. | |
583 | * | |
584 | * When the Guest sends DMA to us, it sends us an array of addresses and sizes. | |
585 | * We need to make sure it's not trying to reach into the Launcher itself, so | |
586 | * we have a convenient routine which check it and exits with an error message | |
587 | * if something funny is going on: | |
588 | */ | |
8ca47e00 RR |
589 | static void *_check_pointer(unsigned long addr, unsigned int size, |
590 | unsigned int line) | |
591 | { | |
dde79789 RR |
592 | /* We have to separately check addr and addr+size, because size could |
593 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 594 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 595 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
596 | /* We return a pointer for the caller's convenience, now we know it's |
597 | * safe to use. */ | |
3c6b5bfa | 598 | return from_guest_phys(addr); |
8ca47e00 | 599 | } |
dde79789 | 600 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
601 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
602 | ||
17cbca2b RR |
603 | /* This function returns the next descriptor in the chain, or vq->vring.num. */ |
604 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) | |
605 | { | |
606 | unsigned int next; | |
607 | ||
608 | /* If this descriptor says it doesn't chain, we're done. */ | |
609 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
610 | return vq->vring.num; | |
611 | ||
612 | /* Check they're not leading us off end of descriptors. */ | |
613 | next = vq->vring.desc[i].next; | |
614 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
615 | wmb(); | |
616 | ||
617 | if (next >= vq->vring.num) | |
618 | errx(1, "Desc next is %u", next); | |
619 | ||
620 | return next; | |
621 | } | |
622 | ||
623 | /* This looks in the virtqueue and for the first available buffer, and converts | |
624 | * it to an iovec for convenient access. Since descriptors consist of some | |
625 | * number of output then some number of input descriptors, it's actually two | |
626 | * iovecs, but we pack them into one and note how many of each there were. | |
627 | * | |
628 | * This function returns the descriptor number found, or vq->vring.num (which | |
629 | * is never a valid descriptor number) if none was found. */ | |
630 | static unsigned get_vq_desc(struct virtqueue *vq, | |
631 | struct iovec iov[], | |
632 | unsigned int *out_num, unsigned int *in_num) | |
633 | { | |
634 | unsigned int i, head; | |
635 | ||
636 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
637 | if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num) | |
638 | errx(1, "Guest moved used index from %u to %u", | |
639 | vq->last_avail_idx, vq->vring.avail->idx); | |
640 | ||
641 | /* If there's nothing new since last we looked, return invalid. */ | |
642 | if (vq->vring.avail->idx == vq->last_avail_idx) | |
643 | return vq->vring.num; | |
644 | ||
645 | /* Grab the next descriptor number they're advertising, and increment | |
646 | * the index we've seen. */ | |
647 | head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num]; | |
648 | ||
649 | /* If their number is silly, that's a fatal mistake. */ | |
650 | if (head >= vq->vring.num) | |
651 | errx(1, "Guest says index %u is available", head); | |
652 | ||
653 | /* When we start there are none of either input nor output. */ | |
654 | *out_num = *in_num = 0; | |
655 | ||
656 | i = head; | |
657 | do { | |
658 | /* Grab the first descriptor, and check it's OK. */ | |
659 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
660 | iov[*out_num + *in_num].iov_base | |
661 | = check_pointer(vq->vring.desc[i].addr, | |
662 | vq->vring.desc[i].len); | |
663 | /* If this is an input descriptor, increment that count. */ | |
664 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
665 | (*in_num)++; | |
666 | else { | |
667 | /* If it's an output descriptor, they're all supposed | |
668 | * to come before any input descriptors. */ | |
669 | if (*in_num) | |
670 | errx(1, "Descriptor has out after in"); | |
671 | (*out_num)++; | |
672 | } | |
673 | ||
674 | /* If we've got too many, that implies a descriptor loop. */ | |
675 | if (*out_num + *in_num > vq->vring.num) | |
676 | errx(1, "Looped descriptor"); | |
677 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 678 | |
17cbca2b | 679 | return head; |
8ca47e00 RR |
680 | } |
681 | ||
17cbca2b RR |
682 | /* Once we've used one of their buffers, we tell them about it. We'll then |
683 | * want to send them an interrupt, using trigger_irq(). */ | |
684 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 685 | { |
17cbca2b RR |
686 | struct vring_used_elem *used; |
687 | ||
688 | /* Get a pointer to the next entry in the used ring. */ | |
689 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; | |
690 | used->id = head; | |
691 | used->len = len; | |
692 | /* Make sure buffer is written before we update index. */ | |
693 | wmb(); | |
694 | vq->vring.used->idx++; | |
8ca47e00 RR |
695 | } |
696 | ||
17cbca2b RR |
697 | /* This actually sends the interrupt for this virtqueue */ |
698 | static void trigger_irq(int fd, struct virtqueue *vq) | |
8ca47e00 | 699 | { |
17cbca2b RR |
700 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
701 | ||
702 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
703 | return; | |
704 | ||
705 | /* Send the Guest an interrupt tell them we used something up. */ | |
8ca47e00 | 706 | if (write(fd, buf, sizeof(buf)) != 0) |
17cbca2b | 707 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
708 | } |
709 | ||
17cbca2b RR |
710 | /* And here's the combo meal deal. Supersize me! */ |
711 | static void add_used_and_trigger(int fd, struct virtqueue *vq, | |
712 | unsigned int head, int len) | |
8ca47e00 | 713 | { |
17cbca2b RR |
714 | add_used(vq, head, len); |
715 | trigger_irq(fd, vq); | |
8ca47e00 RR |
716 | } |
717 | ||
dde79789 RR |
718 | /* Here is the input terminal setting we save, and the routine to restore them |
719 | * on exit so the user can see what they type next. */ | |
8ca47e00 RR |
720 | static struct termios orig_term; |
721 | static void restore_term(void) | |
722 | { | |
723 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
724 | } | |
725 | ||
dde79789 | 726 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
727 | struct console_abort |
728 | { | |
dde79789 | 729 | /* How many times have they hit ^C? */ |
8ca47e00 | 730 | int count; |
dde79789 | 731 | /* When did they start? */ |
8ca47e00 RR |
732 | struct timeval start; |
733 | }; | |
734 | ||
dde79789 | 735 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
736 | static bool handle_console_input(int fd, struct device *dev) |
737 | { | |
8ca47e00 | 738 | int len; |
17cbca2b RR |
739 | unsigned int head, in_num, out_num; |
740 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
741 | struct console_abort *abort = dev->priv; |
742 | ||
17cbca2b RR |
743 | /* First we need a console buffer from the Guests's input virtqueue. */ |
744 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
745 | |
746 | /* If they're not ready for input, stop listening to this file | |
747 | * descriptor. We'll start again once they add an input buffer. */ | |
748 | if (head == dev->vq->vring.num) | |
749 | return false; | |
750 | ||
751 | if (out_num) | |
17cbca2b | 752 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 753 | |
dde79789 RR |
754 | /* This is why we convert to iovecs: the readv() call uses them, and so |
755 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 756 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 757 | if (len <= 0) { |
dde79789 | 758 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 759 | * something went terribly wrong. */ |
8ca47e00 | 760 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 761 | /* Put the input terminal back. */ |
17cbca2b | 762 | restore_term(); |
56ae43df RR |
763 | /* Remove callback from input vq, so it doesn't restart us. */ |
764 | dev->vq->handle_output = NULL; | |
765 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 766 | return false; |
8ca47e00 RR |
767 | } |
768 | ||
56ae43df RR |
769 | /* Tell the Guest about the new input. */ |
770 | add_used_and_trigger(fd, dev->vq, head, len); | |
8ca47e00 | 771 | |
dde79789 RR |
772 | /* Three ^C within one second? Exit. |
773 | * | |
774 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
775 | * in a buffer by itself, so they can't be too fast. But we check that | |
776 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
777 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
778 | if (!abort->count++) | |
779 | gettimeofday(&abort->start, NULL); | |
780 | else if (abort->count == 3) { | |
781 | struct timeval now; | |
782 | gettimeofday(&now, NULL); | |
783 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 784 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
785 | /* Close the fd so Waker will know it has to |
786 | * exit. */ | |
8ca47e00 | 787 | close(waker_fd); |
dde79789 RR |
788 | /* Just in case waker is blocked in BREAK, send |
789 | * unbreak now. */ | |
8ca47e00 RR |
790 | write(fd, args, sizeof(args)); |
791 | exit(2); | |
792 | } | |
793 | abort->count = 0; | |
794 | } | |
795 | } else | |
dde79789 | 796 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
797 | abort->count = 0; |
798 | ||
dde79789 | 799 | /* Everything went OK! */ |
8ca47e00 RR |
800 | return true; |
801 | } | |
802 | ||
17cbca2b RR |
803 | /* Handling output for console is simple: we just get all the output buffers |
804 | * and write them to stdout. */ | |
805 | static void handle_console_output(int fd, struct virtqueue *vq) | |
8ca47e00 | 806 | { |
17cbca2b RR |
807 | unsigned int head, out, in; |
808 | int len; | |
809 | struct iovec iov[vq->vring.num]; | |
810 | ||
811 | /* Keep getting output buffers from the Guest until we run out. */ | |
812 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
813 | if (in) | |
814 | errx(1, "Input buffers in output queue?"); | |
815 | len = writev(STDOUT_FILENO, iov, out); | |
816 | add_used_and_trigger(fd, vq, head, len); | |
817 | } | |
8ca47e00 RR |
818 | } |
819 | ||
17cbca2b RR |
820 | /* Handling output for network is also simple: we get all the output buffers |
821 | * and write them (ignoring the first element) to this device's file descriptor | |
822 | * (stdout). */ | |
823 | static void handle_net_output(int fd, struct virtqueue *vq) | |
8ca47e00 | 824 | { |
17cbca2b RR |
825 | unsigned int head, out, in; |
826 | int len; | |
827 | struct iovec iov[vq->vring.num]; | |
828 | ||
829 | /* Keep getting output buffers from the Guest until we run out. */ | |
830 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
831 | if (in) | |
832 | errx(1, "Input buffers in output queue?"); | |
833 | /* Check header, but otherwise ignore it (we said we supported | |
834 | * no features). */ | |
835 | (void)convert(&iov[0], struct virtio_net_hdr); | |
836 | len = writev(vq->dev->fd, iov+1, out-1); | |
837 | add_used_and_trigger(fd, vq, head, len); | |
838 | } | |
8ca47e00 RR |
839 | } |
840 | ||
17cbca2b RR |
841 | /* This is where we handle a packet coming in from the tun device to our |
842 | * Guest. */ | |
8ca47e00 RR |
843 | static bool handle_tun_input(int fd, struct device *dev) |
844 | { | |
17cbca2b | 845 | unsigned int head, in_num, out_num; |
8ca47e00 | 846 | int len; |
17cbca2b RR |
847 | struct iovec iov[dev->vq->vring.num]; |
848 | struct virtio_net_hdr *hdr; | |
8ca47e00 | 849 | |
17cbca2b RR |
850 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
851 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
852 | if (head == dev->vq->vring.num) { | |
dde79789 | 853 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
854 | * early, the Guest won't be ready yet. Wait until the device |
855 | * status says it's ready. */ | |
856 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
857 | if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) | |
8ca47e00 | 858 | warn("network: no dma buffer!"); |
56ae43df RR |
859 | /* We'll turn this back on if input buffers are registered. */ |
860 | return false; | |
17cbca2b RR |
861 | } else if (out_num) |
862 | errx(1, "Output buffers in network recv queue?"); | |
863 | ||
864 | /* First element is the header: we set it to 0 (no features). */ | |
865 | hdr = convert(&iov[0], struct virtio_net_hdr); | |
866 | hdr->flags = 0; | |
867 | hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
8ca47e00 | 868 | |
dde79789 | 869 | /* Read the packet from the device directly into the Guest's buffer. */ |
17cbca2b | 870 | len = readv(dev->fd, iov+1, in_num-1); |
8ca47e00 RR |
871 | if (len <= 0) |
872 | err(1, "reading network"); | |
dde79789 | 873 | |
56ae43df RR |
874 | /* Tell the Guest about the new packet. */ |
875 | add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len); | |
17cbca2b | 876 | |
8ca47e00 | 877 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
878 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
879 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
880 | ||
dde79789 | 881 | /* All good. */ |
8ca47e00 RR |
882 | return true; |
883 | } | |
884 | ||
56ae43df RR |
885 | /* This callback ensures we try again, in case we stopped console or net |
886 | * delivery because Guest didn't have any buffers. */ | |
887 | static void enable_fd(int fd, struct virtqueue *vq) | |
888 | { | |
889 | add_device_fd(vq->dev->fd); | |
890 | /* Tell waker to listen to it again */ | |
891 | write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd)); | |
892 | } | |
893 | ||
17cbca2b RR |
894 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
895 | static void handle_output(int fd, unsigned long addr) | |
8ca47e00 RR |
896 | { |
897 | struct device *i; | |
17cbca2b RR |
898 | struct virtqueue *vq; |
899 | ||
900 | /* Check each virtqueue. */ | |
901 | for (i = devices.dev; i; i = i->next) { | |
902 | for (vq = i->vq; vq; vq = vq->next) { | |
903 | if (vq->config.pfn == addr/getpagesize() | |
904 | && vq->handle_output) { | |
905 | verbose("Output to %s\n", vq->dev->name); | |
906 | vq->handle_output(fd, vq); | |
907 | return; | |
908 | } | |
8ca47e00 RR |
909 | } |
910 | } | |
dde79789 | 911 | |
17cbca2b RR |
912 | /* Early console write is done using notify on a nul-terminated string |
913 | * in Guest memory. */ | |
914 | if (addr >= guest_limit) | |
915 | errx(1, "Bad NOTIFY %#lx", addr); | |
916 | ||
917 | write(STDOUT_FILENO, from_guest_phys(addr), | |
918 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
919 | } |
920 | ||
dde79789 RR |
921 | /* This is called when the waker wakes us up: check for incoming file |
922 | * descriptors. */ | |
17cbca2b | 923 | static void handle_input(int fd) |
8ca47e00 | 924 | { |
dde79789 | 925 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
926 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
927 | ||
928 | for (;;) { | |
929 | struct device *i; | |
17cbca2b | 930 | fd_set fds = devices.infds; |
8ca47e00 | 931 | |
dde79789 | 932 | /* If nothing is ready, we're done. */ |
17cbca2b | 933 | if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0) |
8ca47e00 RR |
934 | break; |
935 | ||
dde79789 RR |
936 | /* Otherwise, call the device(s) which have readable |
937 | * file descriptors and a method of handling them. */ | |
17cbca2b | 938 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 939 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56ae43df RR |
940 | int dev_fd; |
941 | if (i->handle_input(fd, i)) | |
942 | continue; | |
943 | ||
dde79789 | 944 | /* If handle_input() returns false, it means we |
56ae43df RR |
945 | * should no longer service it. Networking and |
946 | * console do this when there's no input | |
947 | * buffers to deliver into. Console also uses | |
948 | * it when it discovers that stdin is | |
949 | * closed. */ | |
950 | FD_CLR(i->fd, &devices.infds); | |
951 | /* Tell waker to ignore it too, by sending a | |
952 | * negative fd number (-1, since 0 is a valid | |
953 | * FD number). */ | |
954 | dev_fd = -i->fd - 1; | |
955 | write(waker_fd, &dev_fd, sizeof(dev_fd)); | |
8ca47e00 RR |
956 | } |
957 | } | |
958 | } | |
959 | } | |
960 | ||
dde79789 RR |
961 | /*L:190 |
962 | * Device Setup | |
963 | * | |
964 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
965 | * device" so the Launcher can keep track of it. We have common helper | |
966 | * routines to allocate them. | |
967 | * | |
968 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
17cbca2b RR |
969 | * table just above the Guest's normal memory. It returns a pointer to that |
970 | * descriptor. */ | |
971 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
8ca47e00 | 972 | { |
17cbca2b | 973 | struct lguest_device_desc *d; |
8ca47e00 | 974 | |
17cbca2b RR |
975 | /* We only have one page for all the descriptors. */ |
976 | if (devices.desc_used + sizeof(*d) > getpagesize()) | |
977 | errx(1, "Too many devices"); | |
978 | ||
979 | /* We don't need to set config_len or status: page is 0 already. */ | |
980 | d = (void *)devices.descpage + devices.desc_used; | |
981 | d->type = type; | |
982 | devices.desc_used += sizeof(*d); | |
983 | ||
984 | return d; | |
985 | } | |
986 | ||
987 | /* Each device descriptor is followed by some configuration information. | |
988 | * The first byte is a "status" byte for the Guest to report what's happening. | |
989 | * After that are fields: u8 type, u8 len, [... len bytes...]. | |
990 | * | |
991 | * This routine adds a new field to an existing device's descriptor. It only | |
992 | * works for the last device, but that's OK because that's how we use it. */ | |
993 | static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c) | |
994 | { | |
995 | /* This is the last descriptor, right? */ | |
996 | assert(devices.descpage + devices.desc_used | |
997 | == (u8 *)(dev->desc + 1) + dev->desc->config_len); | |
998 | ||
999 | /* We only have one page of device descriptions. */ | |
1000 | if (devices.desc_used + 2 + len > getpagesize()) | |
1001 | errx(1, "Too many devices"); | |
1002 | ||
1003 | /* Copy in the new config header: type then length. */ | |
1004 | devices.descpage[devices.desc_used++] = type; | |
1005 | devices.descpage[devices.desc_used++] = len; | |
1006 | memcpy(devices.descpage + devices.desc_used, c, len); | |
1007 | devices.desc_used += len; | |
1008 | ||
1009 | /* Update the device descriptor length: two byte head then data. */ | |
1010 | dev->desc->config_len += 2 + len; | |
1011 | } | |
1012 | ||
1013 | /* This routine adds a virtqueue to a device. We specify how many descriptors | |
1014 | * the virtqueue is to have. */ | |
1015 | static void add_virtqueue(struct device *dev, unsigned int num_descs, | |
1016 | void (*handle_output)(int fd, struct virtqueue *me)) | |
1017 | { | |
1018 | unsigned int pages; | |
1019 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1020 | void *p; | |
1021 | ||
1022 | /* First we need some pages for this virtqueue. */ | |
1023 | pages = (vring_size(num_descs) + getpagesize() - 1) / getpagesize(); | |
1024 | p = get_pages(pages); | |
1025 | ||
1026 | /* Initialize the configuration. */ | |
1027 | vq->config.num = num_descs; | |
1028 | vq->config.irq = devices.next_irq++; | |
1029 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1030 | ||
1031 | /* Initialize the vring. */ | |
1032 | vring_init(&vq->vring, num_descs, p); | |
1033 | ||
1034 | /* Add the configuration information to this device's descriptor. */ | |
1035 | add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE, | |
1036 | sizeof(vq->config), &vq->config); | |
1037 | ||
1038 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1039 | * second. */ | |
1040 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1041 | *i = vq; | |
1042 | ||
1043 | /* Link virtqueue back to device. */ | |
1044 | vq->dev = dev; | |
1045 | ||
1046 | /* Set up handler. */ | |
1047 | vq->handle_output = handle_output; | |
1048 | if (!handle_output) | |
1049 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1050 | } |
1051 | ||
17cbca2b RR |
1052 | /* This routine does all the creation and setup of a new device, including |
1053 | * caling new_dev_desc() to allocate the descriptor and device memory. */ | |
1054 | static struct device *new_device(const char *name, u16 type, int fd, | |
1055 | bool (*handle_input)(int, struct device *)) | |
8ca47e00 RR |
1056 | { |
1057 | struct device *dev = malloc(sizeof(*dev)); | |
1058 | ||
dde79789 RR |
1059 | /* Append to device list. Prepending to a single-linked list is |
1060 | * easier, but the user expects the devices to be arranged on the bus | |
1061 | * in command-line order. The first network device on the command line | |
1062 | * is eth0, the first block device /dev/lgba, etc. */ | |
17cbca2b | 1063 | *devices.lastdev = dev; |
8ca47e00 | 1064 | dev->next = NULL; |
17cbca2b | 1065 | devices.lastdev = &dev->next; |
8ca47e00 | 1066 | |
dde79789 | 1067 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1068 | dev->fd = fd; |
dde79789 RR |
1069 | /* If we have an input handler for this file descriptor, then we add it |
1070 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1071 | if (handle_input) |
17cbca2b RR |
1072 | add_device_fd(dev->fd); |
1073 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1074 | dev->handle_input = handle_input; |
17cbca2b | 1075 | dev->name = name; |
8ca47e00 RR |
1076 | return dev; |
1077 | } | |
1078 | ||
dde79789 RR |
1079 | /* Our first setup routine is the console. It's a fairly simple device, but |
1080 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1081 | static void setup_console(void) |
8ca47e00 RR |
1082 | { |
1083 | struct device *dev; | |
1084 | ||
dde79789 | 1085 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1086 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1087 | struct termios term = orig_term; | |
dde79789 RR |
1088 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1089 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1090 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1091 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1092 | /* If we exit gracefully, the original settings will be |
1093 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1094 | atexit(restore_term); |
1095 | } | |
1096 | ||
17cbca2b RR |
1097 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1098 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1099 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1100 | dev->priv = malloc(sizeof(struct console_abort)); |
1101 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1102 | |
56ae43df RR |
1103 | /* The console needs two virtqueues: the input then the output. When |
1104 | * they put something the input queue, we make sure we're listening to | |
1105 | * stdin. When they put something in the output queue, we write it to | |
1106 | * stdout. */ | |
1107 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
17cbca2b RR |
1108 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1109 | ||
1110 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1111 | } |
17cbca2b | 1112 | /*:*/ |
8ca47e00 | 1113 | |
17cbca2b RR |
1114 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1115 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1116 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1117 | * |
17cbca2b RR |
1118 | * More sopisticated is to use one of the tools developed for project like UML |
1119 | * to do networking. | |
dde79789 | 1120 | * |
17cbca2b RR |
1121 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1122 | * completely generic ("here's my vring, attach to your vring") and would work | |
1123 | * for any traffic. Of course, namespace and permissions issues need to be | |
1124 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1125 | * multiple inter-guest channels behind one interface, although it would | |
1126 | * require some manner of hotplugging new virtio channels. | |
1127 | * | |
1128 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1129 | |
1130 | static u32 str2ip(const char *ipaddr) | |
1131 | { | |
1132 | unsigned int byte[4]; | |
1133 | ||
1134 | sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]); | |
1135 | return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3]; | |
1136 | } | |
1137 | ||
dde79789 RR |
1138 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1139 | * network device to the bridge device specified by the command line. | |
1140 | * | |
1141 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1142 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1143 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1144 | { | |
1145 | int ifidx; | |
1146 | struct ifreq ifr; | |
1147 | ||
1148 | if (!*br_name) | |
1149 | errx(1, "must specify bridge name"); | |
1150 | ||
1151 | ifidx = if_nametoindex(if_name); | |
1152 | if (!ifidx) | |
1153 | errx(1, "interface %s does not exist!", if_name); | |
1154 | ||
1155 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
1156 | ifr.ifr_ifindex = ifidx; | |
1157 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1158 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1159 | } | |
1160 | ||
dde79789 RR |
1161 | /* This sets up the Host end of the network device with an IP address, brings |
1162 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1163 | * pointer. */ |
8ca47e00 RR |
1164 | static void configure_device(int fd, const char *devname, u32 ipaddr, |
1165 | unsigned char hwaddr[6]) | |
1166 | { | |
1167 | struct ifreq ifr; | |
1168 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1169 | ||
dde79789 | 1170 | /* Don't read these incantations. Just cut & paste them like I did! */ |
8ca47e00 RR |
1171 | memset(&ifr, 0, sizeof(ifr)); |
1172 | strcpy(ifr.ifr_name, devname); | |
1173 | sin->sin_family = AF_INET; | |
1174 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1175 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
1176 | err(1, "Setting %s interface address", devname); | |
1177 | ifr.ifr_flags = IFF_UP; | |
1178 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
1179 | err(1, "Bringing interface %s up", devname); | |
1180 | ||
dde79789 RR |
1181 | /* SIOC stands for Socket I/O Control. G means Get (vs S for Set |
1182 | * above). IF means Interface, and HWADDR is hardware address. | |
1183 | * Simple! */ | |
8ca47e00 RR |
1184 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0) |
1185 | err(1, "getting hw address for %s", devname); | |
8ca47e00 RR |
1186 | memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); |
1187 | } | |
1188 | ||
17cbca2b RR |
1189 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or |
1190 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1191 | * packets into the Host as if they came in from a normal network card. We | |
1192 | * just shunt packets between the Guest and the tun device. */ | |
1193 | static void setup_tun_net(const char *arg) | |
8ca47e00 RR |
1194 | { |
1195 | struct device *dev; | |
1196 | struct ifreq ifr; | |
1197 | int netfd, ipfd; | |
1198 | u32 ip; | |
1199 | const char *br_name = NULL; | |
17cbca2b | 1200 | u8 hwaddr[6]; |
8ca47e00 | 1201 | |
dde79789 RR |
1202 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1203 | * tap device is like a tun device, only somehow different. To tell | |
1204 | * the truth, I completely blundered my way through this code, but it | |
1205 | * works now! */ | |
8ca47e00 RR |
1206 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
1207 | memset(&ifr, 0, sizeof(ifr)); | |
1208 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
1209 | strcpy(ifr.ifr_name, "tap%d"); | |
1210 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1211 | err(1, "configuring /dev/net/tun"); | |
dde79789 RR |
1212 | /* We don't need checksums calculated for packets coming in this |
1213 | * device: trust us! */ | |
8ca47e00 RR |
1214 | ioctl(netfd, TUNSETNOCSUM, 1); |
1215 | ||
17cbca2b RR |
1216 | /* First we create a new network device. */ |
1217 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1218 | |
56ae43df RR |
1219 | /* Network devices need a receive and a send queue, just like |
1220 | * console. */ | |
1221 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
17cbca2b | 1222 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1223 | |
dde79789 RR |
1224 | /* We need a socket to perform the magic network ioctls to bring up the |
1225 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1226 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1227 | if (ipfd < 0) | |
1228 | err(1, "opening IP socket"); | |
1229 | ||
dde79789 | 1230 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 RR |
1231 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
1232 | ip = INADDR_ANY; | |
1233 | br_name = arg + strlen(BRIDGE_PFX); | |
1234 | add_to_bridge(ipfd, ifr.ifr_name, br_name); | |
dde79789 | 1235 | } else /* It is an IP address to set up the device with */ |
8ca47e00 RR |
1236 | ip = str2ip(arg); |
1237 | ||
17cbca2b RR |
1238 | /* Set up the tun device, and get the mac address for the interface. */ |
1239 | configure_device(ipfd, ifr.ifr_name, ip, hwaddr); | |
8ca47e00 | 1240 | |
17cbca2b RR |
1241 | /* Tell Guest what MAC address to use. */ |
1242 | add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr); | |
8ca47e00 | 1243 | |
17cbca2b | 1244 | /* We don't seed the socket any more; setup is done. */ |
8ca47e00 RR |
1245 | close(ipfd); |
1246 | ||
17cbca2b RR |
1247 | verbose("device %u: tun net %u.%u.%u.%u\n", |
1248 | devices.device_num++, | |
1249 | (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip); | |
8ca47e00 RR |
1250 | if (br_name) |
1251 | verbose("attached to bridge: %s\n", br_name); | |
1252 | } | |
17cbca2b RR |
1253 | |
1254 | ||
1255 | /* | |
1256 | * Block device. | |
1257 | * | |
1258 | * Serving a block device is really easy: the Guest asks for a block number and | |
1259 | * we read or write that position in the file. | |
1260 | * | |
1261 | * Unfortunately, this is amazingly slow: the Guest waits until the read is | |
1262 | * finished before running anything else, even if it could be doing useful | |
1263 | * work. We could use async I/O, except it's reputed to suck so hard that | |
1264 | * characters actually go missing from your code when you try to use it. | |
1265 | * | |
1266 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1267 | ||
1268 | /* This hangs off device->priv, with the data. */ | |
1269 | struct vblk_info | |
1270 | { | |
1271 | /* The size of the file. */ | |
1272 | off64_t len; | |
1273 | ||
1274 | /* The file descriptor for the file. */ | |
1275 | int fd; | |
1276 | ||
1277 | /* IO thread listens on this file descriptor [0]. */ | |
1278 | int workpipe[2]; | |
1279 | ||
1280 | /* IO thread writes to this file descriptor to mark it done, then | |
1281 | * Launcher triggers interrupt to Guest. */ | |
1282 | int done_fd; | |
1283 | }; | |
1284 | ||
1285 | /* This is the core of the I/O thread. It returns true if it did something. */ | |
1286 | static bool service_io(struct device *dev) | |
1287 | { | |
1288 | struct vblk_info *vblk = dev->priv; | |
1289 | unsigned int head, out_num, in_num, wlen; | |
1290 | int ret; | |
1291 | struct virtio_blk_inhdr *in; | |
1292 | struct virtio_blk_outhdr *out; | |
1293 | struct iovec iov[dev->vq->vring.num]; | |
1294 | off64_t off; | |
1295 | ||
1296 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
1297 | if (head == dev->vq->vring.num) | |
1298 | return false; | |
1299 | ||
1300 | if (out_num == 0 || in_num == 0) | |
1301 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1302 | head, out_num, in_num); | |
1303 | ||
1304 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
1305 | in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr); | |
1306 | off = out->sector * 512; | |
1307 | ||
1308 | /* This is how we implement barriers. Pretty poor, no? */ | |
1309 | if (out->type & VIRTIO_BLK_T_BARRIER) | |
1310 | fdatasync(vblk->fd); | |
1311 | ||
1312 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { | |
1313 | fprintf(stderr, "Scsi commands unsupported\n"); | |
1314 | in->status = VIRTIO_BLK_S_UNSUPP; | |
1315 | wlen = sizeof(in); | |
1316 | } else if (out->type & VIRTIO_BLK_T_OUT) { | |
1317 | /* Write */ | |
1318 | ||
1319 | /* Move to the right location in the block file. This can fail | |
1320 | * if they try to write past end. */ | |
1321 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1322 | err(1, "Bad seek to sector %llu", out->sector); | |
1323 | ||
1324 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1325 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1326 | ||
1327 | /* Grr... Now we know how long the descriptor they sent was, we | |
1328 | * make sure they didn't try to write over the end of the block | |
1329 | * file (possibly extending it). */ | |
1330 | if (ret > 0 && off + ret > vblk->len) { | |
1331 | /* Trim it back to the correct length */ | |
1332 | ftruncate64(vblk->fd, vblk->len); | |
1333 | /* Die, bad Guest, die. */ | |
1334 | errx(1, "Write past end %llu+%u", off, ret); | |
1335 | } | |
1336 | wlen = sizeof(in); | |
1337 | in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
1338 | } else { | |
1339 | /* Read */ | |
1340 | ||
1341 | /* Move to the right location in the block file. This can fail | |
1342 | * if they try to read past end. */ | |
1343 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1344 | err(1, "Bad seek to sector %llu", out->sector); | |
1345 | ||
1346 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1347 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1348 | if (ret >= 0) { | |
1349 | wlen = sizeof(in) + ret; | |
1350 | in->status = VIRTIO_BLK_S_OK; | |
1351 | } else { | |
1352 | wlen = sizeof(in); | |
1353 | in->status = VIRTIO_BLK_S_IOERR; | |
1354 | } | |
1355 | } | |
1356 | ||
1357 | /* We can't trigger an IRQ, because we're not the Launcher. It does | |
1358 | * that when we tell it we're done. */ | |
1359 | add_used(dev->vq, head, wlen); | |
1360 | return true; | |
1361 | } | |
1362 | ||
1363 | /* This is the thread which actually services the I/O. */ | |
1364 | static int io_thread(void *_dev) | |
1365 | { | |
1366 | struct device *dev = _dev; | |
1367 | struct vblk_info *vblk = dev->priv; | |
1368 | char c; | |
1369 | ||
1370 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1371 | close(vblk->workpipe[1]); | |
1372 | /* Close the other side of the done_fd pipe. */ | |
1373 | close(dev->fd); | |
1374 | ||
1375 | /* When this read fails, it means Launcher died, so we follow. */ | |
1376 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
1377 | /* We acknowledge each request immediately, to reduce latency, | |
1378 | * rather than waiting until we've done them all. I haven't | |
1379 | * measured to see if it makes any difference. */ | |
1380 | while (service_io(dev)) | |
1381 | write(vblk->done_fd, &c, 1); | |
1382 | } | |
1383 | return 0; | |
1384 | } | |
1385 | ||
1386 | /* When the thread says some I/O is done, we interrupt the Guest. */ | |
1387 | static bool handle_io_finish(int fd, struct device *dev) | |
1388 | { | |
1389 | char c; | |
1390 | ||
1391 | /* If child died, presumably it printed message. */ | |
1392 | if (read(dev->fd, &c, 1) != 1) | |
1393 | exit(1); | |
1394 | ||
1395 | /* It did some work, so trigger the irq. */ | |
1396 | trigger_irq(fd, dev->vq); | |
1397 | return true; | |
1398 | } | |
1399 | ||
1400 | /* When the Guest submits some I/O, we wake the I/O thread. */ | |
1401 | static void handle_virtblk_output(int fd, struct virtqueue *vq) | |
1402 | { | |
1403 | struct vblk_info *vblk = vq->dev->priv; | |
1404 | char c = 0; | |
1405 | ||
1406 | /* Wake up I/O thread and tell it to go to work! */ | |
1407 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1408 | /* Presumably it indicated why it died. */ | |
1409 | exit(1); | |
1410 | } | |
1411 | ||
1412 | /* This creates a virtual block device. */ | |
1413 | static void setup_block_file(const char *filename) | |
1414 | { | |
1415 | int p[2]; | |
1416 | struct device *dev; | |
1417 | struct vblk_info *vblk; | |
1418 | void *stack; | |
1419 | u64 cap; | |
1420 | unsigned int val; | |
1421 | ||
1422 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1423 | pipe(p); | |
1424 | ||
1425 | /* The device responds to return from I/O thread. */ | |
1426 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1427 | ||
1428 | /* The device has a virtqueue. */ | |
1429 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); | |
1430 | ||
1431 | /* Allocate the room for our own bookkeeping */ | |
1432 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1433 | ||
1434 | /* First we open the file and store the length. */ | |
1435 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1436 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1437 | ||
1438 | /* Tell Guest how many sectors this device has. */ | |
1439 | cap = cpu_to_le64(vblk->len / 512); | |
1440 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap); | |
1441 | ||
1442 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1443 | * for the in and out elements. */ | |
1444 | val = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1445 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val); | |
1446 | ||
1447 | /* The I/O thread writes to this end of the pipe when done. */ | |
1448 | vblk->done_fd = p[1]; | |
1449 | ||
1450 | /* This is how we tell the I/O thread about more work. */ | |
1451 | pipe(vblk->workpipe); | |
1452 | ||
1453 | /* Create stack for thread and run it */ | |
1454 | stack = malloc(32768); | |
1455 | if (clone(io_thread, stack + 32768, CLONE_VM, dev) == -1) | |
1456 | err(1, "Creating clone"); | |
1457 | ||
1458 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1459 | close(vblk->done_fd); | |
1460 | close(vblk->workpipe[0]); | |
1461 | ||
1462 | verbose("device %u: virtblock %llu sectors\n", | |
1463 | devices.device_num, cap); | |
1464 | } | |
dde79789 | 1465 | /* That's the end of device setup. */ |
8ca47e00 | 1466 | |
dde79789 RR |
1467 | /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves |
1468 | * its input and output, and finally, lays it to rest. */ | |
17cbca2b | 1469 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
8ca47e00 RR |
1470 | { |
1471 | for (;;) { | |
511801dc | 1472 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1473 | unsigned long notify_addr; |
8ca47e00 RR |
1474 | int readval; |
1475 | ||
1476 | /* We read from the /dev/lguest device to run the Guest. */ | |
17cbca2b | 1477 | readval = read(lguest_fd, ¬ify_addr, sizeof(notify_addr)); |
8ca47e00 | 1478 | |
17cbca2b RR |
1479 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1480 | if (readval == sizeof(notify_addr)) { | |
1481 | verbose("Notify on address %#lx\n", notify_addr); | |
1482 | handle_output(lguest_fd, notify_addr); | |
8ca47e00 | 1483 | continue; |
dde79789 | 1484 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1485 | } else if (errno == ENOENT) { |
1486 | char reason[1024] = { 0 }; | |
1487 | read(lguest_fd, reason, sizeof(reason)-1); | |
1488 | errx(1, "%s", reason); | |
dde79789 RR |
1489 | /* EAGAIN means the waker wanted us to look at some input. |
1490 | * Anything else means a bug or incompatible change. */ | |
8ca47e00 RR |
1491 | } else if (errno != EAGAIN) |
1492 | err(1, "Running guest failed"); | |
dde79789 RR |
1493 | |
1494 | /* Service input, then unset the BREAK which releases | |
1495 | * the Waker. */ | |
17cbca2b | 1496 | handle_input(lguest_fd); |
8ca47e00 RR |
1497 | if (write(lguest_fd, args, sizeof(args)) < 0) |
1498 | err(1, "Resetting break"); | |
1499 | } | |
1500 | } | |
dde79789 RR |
1501 | /* |
1502 | * This is the end of the Launcher. | |
1503 | * | |
1504 | * But wait! We've seen I/O from the Launcher, and we've seen I/O from the | |
1505 | * Drivers. If we were to see the Host kernel I/O code, our understanding | |
1506 | * would be complete... :*/ | |
8ca47e00 RR |
1507 | |
1508 | static struct option opts[] = { | |
1509 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1510 | { "tunnet", 1, NULL, 't' }, |
1511 | { "block", 1, NULL, 'b' }, | |
1512 | { "initrd", 1, NULL, 'i' }, | |
1513 | { NULL }, | |
1514 | }; | |
1515 | static void usage(void) | |
1516 | { | |
1517 | errx(1, "Usage: lguest [--verbose] " | |
17cbca2b | 1518 | "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n" |
8ca47e00 RR |
1519 | "|--block=<filename>|--initrd=<filename>]...\n" |
1520 | "<mem-in-mb> vmlinux [args...]"); | |
1521 | } | |
1522 | ||
3c6b5bfa | 1523 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1524 | int main(int argc, char *argv[]) |
1525 | { | |
47436aa4 RR |
1526 | /* Memory, top-level pagetable, code startpoint and size of the |
1527 | * (optional) initrd. */ | |
1528 | unsigned long mem = 0, pgdir, start, initrd_size = 0; | |
dde79789 | 1529 | /* A temporary and the /dev/lguest file descriptor. */ |
6570c459 | 1530 | int i, c, lguest_fd; |
3c6b5bfa | 1531 | /* The boot information for the Guest. */ |
43d33b21 | 1532 | struct boot_params *boot; |
dde79789 | 1533 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1534 | const char *initrd_name = NULL; |
1535 | ||
dde79789 RR |
1536 | /* First we initialize the device list. Since console and network |
1537 | * device receive input from a file descriptor, we keep an fdset | |
1538 | * (infds) and the maximum fd number (max_infd) with the head of the | |
1539 | * list. We also keep a pointer to the last device, for easy appending | |
17cbca2b RR |
1540 | * to the list. Finally, we keep the next interrupt number to hand out |
1541 | * (1: remember that 0 is used by the timer). */ | |
1542 | FD_ZERO(&devices.infds); | |
1543 | devices.max_infd = -1; | |
1544 | devices.lastdev = &devices.dev; | |
1545 | devices.next_irq = 1; | |
8ca47e00 | 1546 | |
dde79789 RR |
1547 | /* We need to know how much memory so we can set up the device |
1548 | * descriptor and memory pages for the devices as we parse the command | |
1549 | * line. So we quickly look through the arguments to find the amount | |
1550 | * of memory now. */ | |
6570c459 RR |
1551 | for (i = 1; i < argc; i++) { |
1552 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1553 | mem = atoi(argv[i]) * 1024 * 1024; |
1554 | /* We start by mapping anonymous pages over all of | |
1555 | * guest-physical memory range. This fills it with 0, | |
1556 | * and ensures that the Guest won't be killed when it | |
1557 | * tries to access it. */ | |
1558 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1559 | + DEVICE_PAGES); | |
1560 | guest_limit = mem; | |
1561 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1562 | devices.descpage = get_pages(1); |
6570c459 RR |
1563 | break; |
1564 | } | |
1565 | } | |
dde79789 RR |
1566 | |
1567 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1568 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1569 | switch (c) { | |
1570 | case 'v': | |
1571 | verbose = true; | |
1572 | break; | |
8ca47e00 | 1573 | case 't': |
17cbca2b | 1574 | setup_tun_net(optarg); |
8ca47e00 RR |
1575 | break; |
1576 | case 'b': | |
17cbca2b | 1577 | setup_block_file(optarg); |
8ca47e00 RR |
1578 | break; |
1579 | case 'i': | |
1580 | initrd_name = optarg; | |
1581 | break; | |
1582 | default: | |
1583 | warnx("Unknown argument %s", argv[optind]); | |
1584 | usage(); | |
1585 | } | |
1586 | } | |
dde79789 RR |
1587 | /* After the other arguments we expect memory and kernel image name, |
1588 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1589 | if (optind + 2 > argc) |
1590 | usage(); | |
1591 | ||
3c6b5bfa RR |
1592 | verbose("Guest base is at %p\n", guest_base); |
1593 | ||
dde79789 | 1594 | /* We always have a console device */ |
17cbca2b | 1595 | setup_console(); |
8ca47e00 | 1596 | |
8ca47e00 | 1597 | /* Now we load the kernel */ |
47436aa4 | 1598 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1599 | |
3c6b5bfa RR |
1600 | /* Boot information is stashed at physical address 0 */ |
1601 | boot = from_guest_phys(0); | |
1602 | ||
dde79789 | 1603 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1604 | if (initrd_name) { |
1605 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1606 | /* These are the location in the Linux boot header where the |
1607 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
1608 | boot->hdr.ramdisk_image = mem - initrd_size; |
1609 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 1610 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 1611 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
1612 | } |
1613 | ||
dde79789 | 1614 | /* Set up the initial linear pagetables, starting below the initrd. */ |
47436aa4 | 1615 | pgdir = setup_pagetables(mem, initrd_size); |
8ca47e00 | 1616 | |
dde79789 RR |
1617 | /* The Linux boot header contains an "E820" memory map: ours is a |
1618 | * simple, single region. */ | |
43d33b21 RR |
1619 | boot->e820_entries = 1; |
1620 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 1621 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
1622 | * line after the boot header. */ |
1623 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
1624 | concat((char *)(boot + 1), argv+optind+2); | |
dde79789 | 1625 | |
814a0e5c | 1626 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 1627 | boot->hdr.version = 0x207; |
814a0e5c RR |
1628 | |
1629 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 1630 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 1631 | |
43d33b21 RR |
1632 | /* Tell the entry path not to try to reload segment registers. */ |
1633 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 1634 | |
dde79789 RR |
1635 | /* We tell the kernel to initialize the Guest: this returns the open |
1636 | * /dev/lguest file descriptor. */ | |
47436aa4 | 1637 | lguest_fd = tell_kernel(pgdir, start); |
dde79789 RR |
1638 | |
1639 | /* We fork off a child process, which wakes the Launcher whenever one | |
1640 | * of the input file descriptors needs attention. Otherwise we would | |
1641 | * run the Guest until it tries to output something. */ | |
17cbca2b | 1642 | waker_fd = setup_waker(lguest_fd); |
8ca47e00 | 1643 | |
dde79789 | 1644 | /* Finally, run the Guest. This doesn't return. */ |
17cbca2b | 1645 | run_guest(lguest_fd); |
8ca47e00 | 1646 | } |
f56a384e RR |
1647 | /*:*/ |
1648 | ||
1649 | /*M:999 | |
1650 | * Mastery is done: you now know everything I do. | |
1651 | * | |
1652 | * But surely you have seen code, features and bugs in your wanderings which | |
1653 | * you now yearn to attack? That is the real game, and I look forward to you | |
1654 | * patching and forking lguest into the Your-Name-Here-visor. | |
1655 | * | |
1656 | * Farewell, and good coding! | |
1657 | * Rusty Russell. | |
1658 | */ |