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 | |
42b36cc0 RR |
65 | /* This will occupy 2 pages: it must be a power of 2. */ |
66 | #define VIRTQUEUE_NUM 128 | |
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 | |
e3283fa0 GOC |
82 | /* a per-cpu variable indicating whose vcpu is currently running */ |
83 | static unsigned int __thread cpu_id; | |
84 | ||
dde79789 | 85 | /* This is our list of devices. */ |
8ca47e00 RR |
86 | struct device_list |
87 | { | |
dde79789 RR |
88 | /* Summary information about the devices in our list: ready to pass to |
89 | * select() to ask which need servicing.*/ | |
8ca47e00 RR |
90 | fd_set infds; |
91 | int max_infd; | |
92 | ||
17cbca2b RR |
93 | /* Counter to assign interrupt numbers. */ |
94 | unsigned int next_irq; | |
95 | ||
96 | /* Counter to print out convenient device numbers. */ | |
97 | unsigned int device_num; | |
98 | ||
dde79789 | 99 | /* The descriptor page for the devices. */ |
17cbca2b RR |
100 | u8 *descpage; |
101 | ||
102 | /* The tail of the last descriptor. */ | |
103 | unsigned int desc_used; | |
dde79789 RR |
104 | |
105 | /* A single linked list of devices. */ | |
8ca47e00 | 106 | struct device *dev; |
dde79789 | 107 | /* ... And an end pointer so we can easily append new devices */ |
8ca47e00 RR |
108 | struct device **lastdev; |
109 | }; | |
110 | ||
17cbca2b RR |
111 | /* The list of Guest devices, based on command line arguments. */ |
112 | static struct device_list devices; | |
113 | ||
dde79789 | 114 | /* The device structure describes a single device. */ |
8ca47e00 RR |
115 | struct device |
116 | { | |
dde79789 | 117 | /* The linked-list pointer. */ |
8ca47e00 | 118 | struct device *next; |
17cbca2b RR |
119 | |
120 | /* The this device's descriptor, as mapped into the Guest. */ | |
8ca47e00 | 121 | struct lguest_device_desc *desc; |
17cbca2b RR |
122 | |
123 | /* The name of this device, for --verbose. */ | |
124 | const char *name; | |
8ca47e00 | 125 | |
dde79789 RR |
126 | /* If handle_input is set, it wants to be called when this file |
127 | * descriptor is ready. */ | |
8ca47e00 RR |
128 | int fd; |
129 | bool (*handle_input)(int fd, struct device *me); | |
130 | ||
17cbca2b RR |
131 | /* Any queues attached to this device */ |
132 | struct virtqueue *vq; | |
8ca47e00 RR |
133 | |
134 | /* Device-specific data. */ | |
135 | void *priv; | |
136 | }; | |
137 | ||
17cbca2b RR |
138 | /* The virtqueue structure describes a queue attached to a device. */ |
139 | struct virtqueue | |
140 | { | |
141 | struct virtqueue *next; | |
142 | ||
143 | /* Which device owns me. */ | |
144 | struct device *dev; | |
145 | ||
146 | /* The configuration for this queue. */ | |
147 | struct lguest_vqconfig config; | |
148 | ||
149 | /* The actual ring of buffers. */ | |
150 | struct vring vring; | |
151 | ||
152 | /* Last available index we saw. */ | |
153 | u16 last_avail_idx; | |
154 | ||
155 | /* The routine to call when the Guest pings us. */ | |
156 | void (*handle_output)(int fd, struct virtqueue *me); | |
157 | }; | |
158 | ||
ec04b13f BR |
159 | /* Remember the arguments to the program so we can "reboot" */ |
160 | static char **main_args; | |
161 | ||
17cbca2b RR |
162 | /* Since guest is UP and we don't run at the same time, we don't need barriers. |
163 | * But I include them in the code in case others copy it. */ | |
164 | #define wmb() | |
165 | ||
166 | /* Convert an iovec element to the given type. | |
167 | * | |
168 | * This is a fairly ugly trick: we need to know the size of the type and | |
169 | * alignment requirement to check the pointer is kosher. It's also nice to | |
170 | * have the name of the type in case we report failure. | |
171 | * | |
172 | * Typing those three things all the time is cumbersome and error prone, so we | |
173 | * have a macro which sets them all up and passes to the real function. */ | |
174 | #define convert(iov, type) \ | |
175 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
176 | ||
177 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
178 | const char *name) | |
179 | { | |
180 | if (iov->iov_len != size) | |
181 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
182 | if ((unsigned long)iov->iov_base % align != 0) | |
183 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
184 | return iov->iov_base; | |
185 | } | |
186 | ||
187 | /* The virtio configuration space is defined to be little-endian. x86 is | |
188 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
189 | #define cpu_to_le16(v16) (v16) | |
190 | #define cpu_to_le32(v32) (v32) | |
191 | #define cpu_to_le64(v64) (v64) | |
192 | #define le16_to_cpu(v16) (v16) | |
193 | #define le32_to_cpu(v32) (v32) | |
194 | #define le64_to_cpu(v32) (v64) | |
195 | ||
3c6b5bfa RR |
196 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
197 | * where pointers run wild and free! Unfortunately, like most userspace | |
198 | * programs, it's quite boring (which is why everyone likes to hack on the | |
199 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
200 | * will get you through this section. Or, maybe not. | |
201 | * | |
202 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
203 | * memory and stores it in "guest_base". In other words, Guest physical == | |
204 | * Launcher virtual with an offset. | |
205 | * | |
206 | * This can be tough to get your head around, but usually it just means that we | |
207 | * use these trivial conversion functions when the Guest gives us it's | |
208 | * "physical" addresses: */ | |
209 | static void *from_guest_phys(unsigned long addr) | |
210 | { | |
211 | return guest_base + addr; | |
212 | } | |
213 | ||
214 | static unsigned long to_guest_phys(const void *addr) | |
215 | { | |
216 | return (addr - guest_base); | |
217 | } | |
218 | ||
dde79789 RR |
219 | /*L:130 |
220 | * Loading the Kernel. | |
221 | * | |
222 | * We start with couple of simple helper routines. open_or_die() avoids | |
223 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
224 | static int open_or_die(const char *name, int flags) |
225 | { | |
226 | int fd = open(name, flags); | |
227 | if (fd < 0) | |
228 | err(1, "Failed to open %s", name); | |
229 | return fd; | |
230 | } | |
231 | ||
3c6b5bfa RR |
232 | /* map_zeroed_pages() takes a number of pages. */ |
233 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 234 | { |
3c6b5bfa RR |
235 | int fd = open_or_die("/dev/zero", O_RDONLY); |
236 | void *addr; | |
8ca47e00 | 237 | |
dde79789 | 238 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
239 | * copied). */ |
240 | addr = mmap(NULL, getpagesize() * num, | |
241 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
242 | if (addr == MAP_FAILED) | |
243 | err(1, "Mmaping %u pages of /dev/zero", num); | |
244 | ||
245 | return addr; | |
246 | } | |
247 | ||
248 | /* Get some more pages for a device. */ | |
249 | static void *get_pages(unsigned int num) | |
250 | { | |
251 | void *addr = from_guest_phys(guest_limit); | |
252 | ||
253 | guest_limit += num * getpagesize(); | |
254 | if (guest_limit > guest_max) | |
255 | errx(1, "Not enough memory for devices"); | |
256 | return addr; | |
8ca47e00 RR |
257 | } |
258 | ||
6649bb7a RM |
259 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
260 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
261 | * it falls back to reading the memory in. */ | |
262 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
263 | { | |
264 | ssize_t r; | |
265 | ||
266 | /* We map writable even though for some segments are marked read-only. | |
267 | * The kernel really wants to be writable: it patches its own | |
268 | * instructions. | |
269 | * | |
270 | * MAP_PRIVATE means that the page won't be copied until a write is | |
271 | * done to it. This allows us to share untouched memory between | |
272 | * Guests. */ | |
273 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
274 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
275 | return; | |
276 | ||
277 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
278 | r = pread(fd, addr, len, offset); | |
279 | if (r != len) | |
280 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
281 | } | |
282 | ||
dde79789 RR |
283 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
284 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
285 | * by all modern binaries on Linux including the kernel. | |
286 | * | |
287 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
288 | * address. We use the physical address; the Guest will map itself to the |
289 | * virtual address. | |
dde79789 RR |
290 | * |
291 | * We return the starting address. */ | |
47436aa4 | 292 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 293 | { |
8ca47e00 RR |
294 | Elf32_Phdr phdr[ehdr->e_phnum]; |
295 | unsigned int i; | |
8ca47e00 | 296 | |
dde79789 RR |
297 | /* Sanity checks on the main ELF header: an x86 executable with a |
298 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
299 | if (ehdr->e_type != ET_EXEC |
300 | || ehdr->e_machine != EM_386 | |
301 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
302 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
303 | errx(1, "Malformed elf header"); | |
304 | ||
dde79789 RR |
305 | /* An ELF executable contains an ELF header and a number of "program" |
306 | * headers which indicate which parts ("segments") of the program to | |
307 | * load where. */ | |
308 | ||
309 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
310 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
311 | err(1, "Seeking to program headers"); | |
312 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
313 | err(1, "Reading program headers"); | |
314 | ||
dde79789 RR |
315 | /* Try all the headers: there are usually only three. A read-only one, |
316 | * a read-write one, and a "note" section which isn't loadable. */ | |
8ca47e00 | 317 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 318 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
319 | if (phdr[i].p_type != PT_LOAD) |
320 | continue; | |
321 | ||
322 | verbose("Section %i: size %i addr %p\n", | |
323 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
324 | ||
6649bb7a | 325 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 326 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 327 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
328 | } |
329 | ||
814a0e5c RR |
330 | /* The entry point is given in the ELF header. */ |
331 | return ehdr->e_entry; | |
8ca47e00 RR |
332 | } |
333 | ||
dde79789 | 334 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
335 | * supposed to jump into it and it will unpack itself. We used to have to |
336 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 337 | * |
5bbf89fc RR |
338 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
339 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
340 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 341 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 342 | { |
43d33b21 | 343 | struct boot_params boot; |
5bbf89fc RR |
344 | int r; |
345 | /* Modern bzImages get loaded at 1M. */ | |
346 | void *p = from_guest_phys(0x100000); | |
347 | ||
348 | /* Go back to the start of the file and read the header. It should be | |
349 | * a Linux boot header (see Documentation/i386/boot.txt) */ | |
350 | lseek(fd, 0, SEEK_SET); | |
43d33b21 | 351 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 352 | |
43d33b21 RR |
353 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
354 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
355 | errx(1, "This doesn't look like a bzImage to me"); |
356 | ||
43d33b21 RR |
357 | /* Skip over the extra sectors of the header. */ |
358 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
359 | |
360 | /* Now read everything into memory. in nice big chunks. */ | |
361 | while ((r = read(fd, p, 65536)) > 0) | |
362 | p += r; | |
363 | ||
43d33b21 RR |
364 | /* Finally, code32_start tells us where to enter the kernel. */ |
365 | return boot.hdr.code32_start; | |
8ca47e00 RR |
366 | } |
367 | ||
dde79789 | 368 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
e1e72965 RR |
369 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
370 | * work, we can load those, too. */ | |
47436aa4 | 371 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
372 | { |
373 | Elf32_Ehdr hdr; | |
374 | ||
dde79789 | 375 | /* Read in the first few bytes. */ |
8ca47e00 RR |
376 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
377 | err(1, "Reading kernel"); | |
378 | ||
dde79789 | 379 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 380 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 381 | return map_elf(fd, &hdr); |
8ca47e00 | 382 | |
dde79789 | 383 | /* Otherwise we assume it's a bzImage, and try to unpack it */ |
47436aa4 | 384 | return load_bzimage(fd); |
8ca47e00 RR |
385 | } |
386 | ||
dde79789 RR |
387 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
388 | * it calls getpagesize() twice: "it's dumb code." | |
389 | * | |
390 | * Kernel guys get really het up about optimization, even when it's not | |
391 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
392 | static inline unsigned long page_align(unsigned long addr) |
393 | { | |
dde79789 | 394 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
395 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
396 | } | |
397 | ||
dde79789 RR |
398 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
399 | * the kernel which the kernel can use to boot from without needing any | |
400 | * drivers. Most distributions now use this as standard: the initrd contains | |
401 | * the code to load the appropriate driver modules for the current machine. | |
402 | * | |
403 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
404 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
405 | static unsigned long load_initrd(const char *name, unsigned long mem) |
406 | { | |
407 | int ifd; | |
408 | struct stat st; | |
409 | unsigned long len; | |
8ca47e00 RR |
410 | |
411 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 412 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
413 | if (fstat(ifd, &st) < 0) |
414 | err(1, "fstat() on initrd '%s'", name); | |
415 | ||
6649bb7a RM |
416 | /* We map the initrd at the top of memory, but mmap wants it to be |
417 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 418 | len = page_align(st.st_size); |
3c6b5bfa | 419 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
420 | /* Once a file is mapped, you can close the file descriptor. It's a |
421 | * little odd, but quite useful. */ | |
8ca47e00 | 422 | close(ifd); |
6649bb7a | 423 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
424 | |
425 | /* We return the initrd size. */ | |
8ca47e00 RR |
426 | return len; |
427 | } | |
428 | ||
47436aa4 RR |
429 | /* Once we know how much memory we have, we can construct simple linear page |
430 | * tables which set virtual == physical which will get the Guest far enough | |
3c6b5bfa | 431 | * into the boot to create its own. |
dde79789 RR |
432 | * |
433 | * We lay them out of the way, just below the initrd (which is why we need to | |
434 | * know its size). */ | |
8ca47e00 | 435 | static unsigned long setup_pagetables(unsigned long mem, |
47436aa4 | 436 | unsigned long initrd_size) |
8ca47e00 | 437 | { |
511801dc | 438 | unsigned long *pgdir, *linear; |
8ca47e00 | 439 | unsigned int mapped_pages, i, linear_pages; |
511801dc | 440 | unsigned int ptes_per_page = getpagesize()/sizeof(void *); |
8ca47e00 | 441 | |
47436aa4 | 442 | mapped_pages = mem/getpagesize(); |
8ca47e00 | 443 | |
dde79789 | 444 | /* Each PTE page can map ptes_per_page pages: how many do we need? */ |
8ca47e00 RR |
445 | linear_pages = (mapped_pages + ptes_per_page-1)/ptes_per_page; |
446 | ||
dde79789 | 447 | /* We put the toplevel page directory page at the top of memory. */ |
3c6b5bfa | 448 | pgdir = from_guest_phys(mem) - initrd_size - getpagesize(); |
dde79789 RR |
449 | |
450 | /* Now we use the next linear_pages pages as pte pages */ | |
8ca47e00 RR |
451 | linear = (void *)pgdir - linear_pages*getpagesize(); |
452 | ||
dde79789 RR |
453 | /* Linear mapping is easy: put every page's address into the mapping in |
454 | * order. PAGE_PRESENT contains the flags Present, Writable and | |
455 | * Executable. */ | |
8ca47e00 RR |
456 | for (i = 0; i < mapped_pages; i++) |
457 | linear[i] = ((i * getpagesize()) | PAGE_PRESENT); | |
458 | ||
47436aa4 | 459 | /* The top level points to the linear page table pages above. */ |
8ca47e00 | 460 | for (i = 0; i < mapped_pages; i += ptes_per_page) { |
47436aa4 | 461 | pgdir[i/ptes_per_page] |
511801dc | 462 | = ((to_guest_phys(linear) + i*sizeof(void *)) |
3c6b5bfa | 463 | | PAGE_PRESENT); |
8ca47e00 RR |
464 | } |
465 | ||
3c6b5bfa RR |
466 | verbose("Linear mapping of %u pages in %u pte pages at %#lx\n", |
467 | mapped_pages, linear_pages, to_guest_phys(linear)); | |
8ca47e00 | 468 | |
dde79789 RR |
469 | /* We return the top level (guest-physical) address: the kernel needs |
470 | * to know where it is. */ | |
3c6b5bfa | 471 | return to_guest_phys(pgdir); |
8ca47e00 | 472 | } |
e1e72965 | 473 | /*:*/ |
8ca47e00 | 474 | |
dde79789 RR |
475 | /* Simple routine to roll all the commandline arguments together with spaces |
476 | * between them. */ | |
8ca47e00 RR |
477 | static void concat(char *dst, char *args[]) |
478 | { | |
479 | unsigned int i, len = 0; | |
480 | ||
481 | for (i = 0; args[i]; i++) { | |
482 | strcpy(dst+len, args[i]); | |
483 | strcat(dst+len, " "); | |
484 | len += strlen(args[i]) + 1; | |
485 | } | |
486 | /* In case it's empty. */ | |
487 | dst[len] = '\0'; | |
488 | } | |
489 | ||
e1e72965 RR |
490 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
491 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: | |
492 | * the base of Guest "physical" memory, the top physical page to allow, the | |
47436aa4 RR |
493 | * top level pagetable and the entry point for the Guest. */ |
494 | static int tell_kernel(unsigned long pgdir, unsigned long start) | |
8ca47e00 | 495 | { |
511801dc JS |
496 | unsigned long args[] = { LHREQ_INITIALIZE, |
497 | (unsigned long)guest_base, | |
47436aa4 | 498 | guest_limit / getpagesize(), pgdir, start }; |
8ca47e00 RR |
499 | int fd; |
500 | ||
3c6b5bfa RR |
501 | verbose("Guest: %p - %p (%#lx)\n", |
502 | guest_base, guest_base + guest_limit, guest_limit); | |
8ca47e00 RR |
503 | fd = open_or_die("/dev/lguest", O_RDWR); |
504 | if (write(fd, args, sizeof(args)) < 0) | |
505 | err(1, "Writing to /dev/lguest"); | |
dde79789 RR |
506 | |
507 | /* We return the /dev/lguest file descriptor to control this Guest */ | |
8ca47e00 RR |
508 | return fd; |
509 | } | |
dde79789 | 510 | /*:*/ |
8ca47e00 | 511 | |
17cbca2b | 512 | static void add_device_fd(int fd) |
8ca47e00 | 513 | { |
17cbca2b RR |
514 | FD_SET(fd, &devices.infds); |
515 | if (fd > devices.max_infd) | |
516 | devices.max_infd = fd; | |
8ca47e00 RR |
517 | } |
518 | ||
dde79789 RR |
519 | /*L:200 |
520 | * The Waker. | |
521 | * | |
e1e72965 RR |
522 | * With console, block and network devices, we can have lots of input which we |
523 | * need to process. We could try to tell the kernel what file descriptors to | |
524 | * watch, but handing a file descriptor mask through to the kernel is fairly | |
525 | * icky. | |
dde79789 RR |
526 | * |
527 | * Instead, we fork off a process which watches the file descriptors and writes | |
e1e72965 RR |
528 | * the LHREQ_BREAK command to the /dev/lguest file descriptor to tell the Host |
529 | * stop running the Guest. This causes the Launcher to return from the | |
dde79789 RR |
530 | * /dev/lguest read with -EAGAIN, where it will write to /dev/lguest to reset |
531 | * the LHREQ_BREAK and wake us up again. | |
532 | * | |
533 | * This, of course, is merely a different *kind* of icky. | |
534 | */ | |
17cbca2b | 535 | static void wake_parent(int pipefd, int lguest_fd) |
8ca47e00 | 536 | { |
dde79789 RR |
537 | /* Add the pipe from the Launcher to the fdset in the device_list, so |
538 | * we watch it, too. */ | |
17cbca2b | 539 | add_device_fd(pipefd); |
8ca47e00 RR |
540 | |
541 | for (;;) { | |
17cbca2b | 542 | fd_set rfds = devices.infds; |
511801dc | 543 | unsigned long args[] = { LHREQ_BREAK, 1 }; |
8ca47e00 | 544 | |
dde79789 | 545 | /* Wait until input is ready from one of the devices. */ |
17cbca2b | 546 | select(devices.max_infd+1, &rfds, NULL, NULL, NULL); |
dde79789 | 547 | /* Is it a message from the Launcher? */ |
8ca47e00 | 548 | if (FD_ISSET(pipefd, &rfds)) { |
56ae43df | 549 | int fd; |
dde79789 RR |
550 | /* If read() returns 0, it means the Launcher has |
551 | * exited. We silently follow. */ | |
56ae43df | 552 | if (read(pipefd, &fd, sizeof(fd)) == 0) |
8ca47e00 | 553 | exit(0); |
56ae43df | 554 | /* Otherwise it's telling us to change what file |
e1e72965 RR |
555 | * descriptors we're to listen to. Positive means |
556 | * listen to a new one, negative means stop | |
557 | * listening. */ | |
56ae43df RR |
558 | if (fd >= 0) |
559 | FD_SET(fd, &devices.infds); | |
560 | else | |
561 | FD_CLR(-fd - 1, &devices.infds); | |
dde79789 | 562 | } else /* Send LHREQ_BREAK command. */ |
e3283fa0 | 563 | pwrite(lguest_fd, args, sizeof(args), cpu_id); |
8ca47e00 RR |
564 | } |
565 | } | |
566 | ||
dde79789 | 567 | /* This routine just sets up a pipe to the Waker process. */ |
17cbca2b | 568 | static int setup_waker(int lguest_fd) |
8ca47e00 RR |
569 | { |
570 | int pipefd[2], child; | |
571 | ||
e1e72965 | 572 | /* We create a pipe to talk to the Waker, and also so it knows when the |
dde79789 | 573 | * Launcher dies (and closes pipe). */ |
8ca47e00 RR |
574 | pipe(pipefd); |
575 | child = fork(); | |
576 | if (child == -1) | |
577 | err(1, "forking"); | |
578 | ||
579 | if (child == 0) { | |
e1e72965 RR |
580 | /* We are the Waker: close the "writing" end of our copy of the |
581 | * pipe and start waiting for input. */ | |
8ca47e00 | 582 | close(pipefd[1]); |
17cbca2b | 583 | wake_parent(pipefd[0], lguest_fd); |
8ca47e00 | 584 | } |
dde79789 | 585 | /* Close the reading end of our copy of the pipe. */ |
8ca47e00 RR |
586 | close(pipefd[0]); |
587 | ||
dde79789 | 588 | /* Here is the fd used to talk to the waker. */ |
8ca47e00 RR |
589 | return pipefd[1]; |
590 | } | |
591 | ||
e1e72965 | 592 | /* |
dde79789 RR |
593 | * Device Handling. |
594 | * | |
e1e72965 | 595 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 596 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 597 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
598 | * if something funny is going on: |
599 | */ | |
8ca47e00 RR |
600 | static void *_check_pointer(unsigned long addr, unsigned int size, |
601 | unsigned int line) | |
602 | { | |
dde79789 RR |
603 | /* We have to separately check addr and addr+size, because size could |
604 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 605 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 606 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
607 | /* We return a pointer for the caller's convenience, now we know it's |
608 | * safe to use. */ | |
3c6b5bfa | 609 | return from_guest_phys(addr); |
8ca47e00 | 610 | } |
dde79789 | 611 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
612 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
613 | ||
e1e72965 RR |
614 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
615 | * function returns the next descriptor in the chain, or vq->vring.num if we're | |
616 | * at the end. */ | |
17cbca2b RR |
617 | static unsigned next_desc(struct virtqueue *vq, unsigned int i) |
618 | { | |
619 | unsigned int next; | |
620 | ||
621 | /* If this descriptor says it doesn't chain, we're done. */ | |
622 | if (!(vq->vring.desc[i].flags & VRING_DESC_F_NEXT)) | |
623 | return vq->vring.num; | |
624 | ||
625 | /* Check they're not leading us off end of descriptors. */ | |
626 | next = vq->vring.desc[i].next; | |
627 | /* Make sure compiler knows to grab that: we don't want it changing! */ | |
628 | wmb(); | |
629 | ||
630 | if (next >= vq->vring.num) | |
631 | errx(1, "Desc next is %u", next); | |
632 | ||
633 | return next; | |
634 | } | |
635 | ||
636 | /* This looks in the virtqueue and for the first available buffer, and converts | |
637 | * it to an iovec for convenient access. Since descriptors consist of some | |
638 | * number of output then some number of input descriptors, it's actually two | |
639 | * iovecs, but we pack them into one and note how many of each there were. | |
640 | * | |
641 | * This function returns the descriptor number found, or vq->vring.num (which | |
642 | * is never a valid descriptor number) if none was found. */ | |
643 | static unsigned get_vq_desc(struct virtqueue *vq, | |
644 | struct iovec iov[], | |
645 | unsigned int *out_num, unsigned int *in_num) | |
646 | { | |
647 | unsigned int i, head; | |
648 | ||
649 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
650 | if ((u16)(vq->vring.avail->idx - vq->last_avail_idx) > vq->vring.num) | |
651 | errx(1, "Guest moved used index from %u to %u", | |
652 | vq->last_avail_idx, vq->vring.avail->idx); | |
653 | ||
654 | /* If there's nothing new since last we looked, return invalid. */ | |
655 | if (vq->vring.avail->idx == vq->last_avail_idx) | |
656 | return vq->vring.num; | |
657 | ||
658 | /* Grab the next descriptor number they're advertising, and increment | |
659 | * the index we've seen. */ | |
660 | head = vq->vring.avail->ring[vq->last_avail_idx++ % vq->vring.num]; | |
661 | ||
662 | /* If their number is silly, that's a fatal mistake. */ | |
663 | if (head >= vq->vring.num) | |
664 | errx(1, "Guest says index %u is available", head); | |
665 | ||
666 | /* When we start there are none of either input nor output. */ | |
667 | *out_num = *in_num = 0; | |
668 | ||
669 | i = head; | |
670 | do { | |
671 | /* Grab the first descriptor, and check it's OK. */ | |
672 | iov[*out_num + *in_num].iov_len = vq->vring.desc[i].len; | |
673 | iov[*out_num + *in_num].iov_base | |
674 | = check_pointer(vq->vring.desc[i].addr, | |
675 | vq->vring.desc[i].len); | |
676 | /* If this is an input descriptor, increment that count. */ | |
677 | if (vq->vring.desc[i].flags & VRING_DESC_F_WRITE) | |
678 | (*in_num)++; | |
679 | else { | |
680 | /* If it's an output descriptor, they're all supposed | |
681 | * to come before any input descriptors. */ | |
682 | if (*in_num) | |
683 | errx(1, "Descriptor has out after in"); | |
684 | (*out_num)++; | |
685 | } | |
686 | ||
687 | /* If we've got too many, that implies a descriptor loop. */ | |
688 | if (*out_num + *in_num > vq->vring.num) | |
689 | errx(1, "Looped descriptor"); | |
690 | } while ((i = next_desc(vq, i)) != vq->vring.num); | |
dde79789 | 691 | |
17cbca2b | 692 | return head; |
8ca47e00 RR |
693 | } |
694 | ||
e1e72965 | 695 | /* After we've used one of their buffers, we tell them about it. We'll then |
17cbca2b RR |
696 | * want to send them an interrupt, using trigger_irq(). */ |
697 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 698 | { |
17cbca2b RR |
699 | struct vring_used_elem *used; |
700 | ||
e1e72965 RR |
701 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
702 | * next entry in that used ring. */ | |
17cbca2b RR |
703 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
704 | used->id = head; | |
705 | used->len = len; | |
706 | /* Make sure buffer is written before we update index. */ | |
707 | wmb(); | |
708 | vq->vring.used->idx++; | |
8ca47e00 RR |
709 | } |
710 | ||
17cbca2b RR |
711 | /* This actually sends the interrupt for this virtqueue */ |
712 | static void trigger_irq(int fd, struct virtqueue *vq) | |
8ca47e00 | 713 | { |
17cbca2b RR |
714 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
715 | ||
e1e72965 | 716 | /* If they don't want an interrupt, don't send one. */ |
17cbca2b RR |
717 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) |
718 | return; | |
719 | ||
720 | /* Send the Guest an interrupt tell them we used something up. */ | |
8ca47e00 | 721 | if (write(fd, buf, sizeof(buf)) != 0) |
17cbca2b | 722 | err(1, "Triggering irq %i", vq->config.irq); |
8ca47e00 RR |
723 | } |
724 | ||
17cbca2b RR |
725 | /* And here's the combo meal deal. Supersize me! */ |
726 | static void add_used_and_trigger(int fd, struct virtqueue *vq, | |
727 | unsigned int head, int len) | |
8ca47e00 | 728 | { |
17cbca2b RR |
729 | add_used(vq, head, len); |
730 | trigger_irq(fd, vq); | |
8ca47e00 RR |
731 | } |
732 | ||
e1e72965 RR |
733 | /* |
734 | * The Console | |
735 | * | |
736 | * Here is the input terminal setting we save, and the routine to restore them | |
737 | * on exit so the user gets their terminal back. */ | |
8ca47e00 RR |
738 | static struct termios orig_term; |
739 | static void restore_term(void) | |
740 | { | |
741 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
742 | } | |
743 | ||
dde79789 | 744 | /* We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
745 | struct console_abort |
746 | { | |
dde79789 | 747 | /* How many times have they hit ^C? */ |
8ca47e00 | 748 | int count; |
dde79789 | 749 | /* When did they start? */ |
8ca47e00 RR |
750 | struct timeval start; |
751 | }; | |
752 | ||
dde79789 | 753 | /* This is the routine which handles console input (ie. stdin). */ |
8ca47e00 RR |
754 | static bool handle_console_input(int fd, struct device *dev) |
755 | { | |
8ca47e00 | 756 | int len; |
17cbca2b RR |
757 | unsigned int head, in_num, out_num; |
758 | struct iovec iov[dev->vq->vring.num]; | |
8ca47e00 RR |
759 | struct console_abort *abort = dev->priv; |
760 | ||
17cbca2b RR |
761 | /* First we need a console buffer from the Guests's input virtqueue. */ |
762 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
56ae43df RR |
763 | |
764 | /* If they're not ready for input, stop listening to this file | |
765 | * descriptor. We'll start again once they add an input buffer. */ | |
766 | if (head == dev->vq->vring.num) | |
767 | return false; | |
768 | ||
769 | if (out_num) | |
17cbca2b | 770 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 771 | |
dde79789 RR |
772 | /* This is why we convert to iovecs: the readv() call uses them, and so |
773 | * it reads straight into the Guest's buffer. */ | |
17cbca2b | 774 | len = readv(dev->fd, iov, in_num); |
8ca47e00 | 775 | if (len <= 0) { |
dde79789 | 776 | /* This implies that the console is closed, is /dev/null, or |
17cbca2b | 777 | * something went terribly wrong. */ |
8ca47e00 | 778 | warnx("Failed to get console input, ignoring console."); |
56ae43df | 779 | /* Put the input terminal back. */ |
17cbca2b | 780 | restore_term(); |
56ae43df RR |
781 | /* Remove callback from input vq, so it doesn't restart us. */ |
782 | dev->vq->handle_output = NULL; | |
783 | /* Stop listening to this fd: don't call us again. */ | |
17cbca2b | 784 | return false; |
8ca47e00 RR |
785 | } |
786 | ||
56ae43df RR |
787 | /* Tell the Guest about the new input. */ |
788 | add_used_and_trigger(fd, dev->vq, head, len); | |
8ca47e00 | 789 | |
dde79789 RR |
790 | /* Three ^C within one second? Exit. |
791 | * | |
792 | * This is such a hack, but works surprisingly well. Each ^C has to be | |
793 | * in a buffer by itself, so they can't be too fast. But we check that | |
794 | * we get three within about a second, so they can't be too slow. */ | |
8ca47e00 RR |
795 | if (len == 1 && ((char *)iov[0].iov_base)[0] == 3) { |
796 | if (!abort->count++) | |
797 | gettimeofday(&abort->start, NULL); | |
798 | else if (abort->count == 3) { | |
799 | struct timeval now; | |
800 | gettimeofday(&now, NULL); | |
801 | if (now.tv_sec <= abort->start.tv_sec+1) { | |
511801dc | 802 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
dde79789 RR |
803 | /* Close the fd so Waker will know it has to |
804 | * exit. */ | |
8ca47e00 | 805 | close(waker_fd); |
dde79789 RR |
806 | /* Just in case waker is blocked in BREAK, send |
807 | * unbreak now. */ | |
8ca47e00 RR |
808 | write(fd, args, sizeof(args)); |
809 | exit(2); | |
810 | } | |
811 | abort->count = 0; | |
812 | } | |
813 | } else | |
dde79789 | 814 | /* Any other key resets the abort counter. */ |
8ca47e00 RR |
815 | abort->count = 0; |
816 | ||
dde79789 | 817 | /* Everything went OK! */ |
8ca47e00 RR |
818 | return true; |
819 | } | |
820 | ||
17cbca2b RR |
821 | /* Handling output for console is simple: we just get all the output buffers |
822 | * and write them to stdout. */ | |
823 | static void handle_console_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 | len = writev(STDOUT_FILENO, iov, out); | |
834 | add_used_and_trigger(fd, vq, head, len); | |
835 | } | |
8ca47e00 RR |
836 | } |
837 | ||
e1e72965 RR |
838 | /* |
839 | * The Network | |
840 | * | |
841 | * Handling output for network is also simple: we get all the output buffers | |
17cbca2b RR |
842 | * and write them (ignoring the first element) to this device's file descriptor |
843 | * (stdout). */ | |
844 | static void handle_net_output(int fd, struct virtqueue *vq) | |
8ca47e00 | 845 | { |
17cbca2b RR |
846 | unsigned int head, out, in; |
847 | int len; | |
848 | struct iovec iov[vq->vring.num]; | |
849 | ||
850 | /* Keep getting output buffers from the Guest until we run out. */ | |
851 | while ((head = get_vq_desc(vq, iov, &out, &in)) != vq->vring.num) { | |
852 | if (in) | |
853 | errx(1, "Input buffers in output queue?"); | |
e1e72965 RR |
854 | /* Check header, but otherwise ignore it (we told the Guest we |
855 | * supported no features, so it shouldn't have anything | |
856 | * interesting). */ | |
17cbca2b RR |
857 | (void)convert(&iov[0], struct virtio_net_hdr); |
858 | len = writev(vq->dev->fd, iov+1, out-1); | |
859 | add_used_and_trigger(fd, vq, head, len); | |
860 | } | |
8ca47e00 RR |
861 | } |
862 | ||
17cbca2b RR |
863 | /* This is where we handle a packet coming in from the tun device to our |
864 | * Guest. */ | |
8ca47e00 RR |
865 | static bool handle_tun_input(int fd, struct device *dev) |
866 | { | |
17cbca2b | 867 | unsigned int head, in_num, out_num; |
8ca47e00 | 868 | int len; |
17cbca2b RR |
869 | struct iovec iov[dev->vq->vring.num]; |
870 | struct virtio_net_hdr *hdr; | |
8ca47e00 | 871 | |
17cbca2b RR |
872 | /* First we need a network buffer from the Guests's recv virtqueue. */ |
873 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); | |
874 | if (head == dev->vq->vring.num) { | |
dde79789 | 875 | /* Now, it's expected that if we try to send a packet too |
17cbca2b RR |
876 | * early, the Guest won't be ready yet. Wait until the device |
877 | * status says it's ready. */ | |
878 | /* FIXME: Actually want DRIVER_ACTIVE here. */ | |
879 | if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) | |
8ca47e00 | 880 | warn("network: no dma buffer!"); |
56ae43df RR |
881 | /* We'll turn this back on if input buffers are registered. */ |
882 | return false; | |
17cbca2b RR |
883 | } else if (out_num) |
884 | errx(1, "Output buffers in network recv queue?"); | |
885 | ||
886 | /* First element is the header: we set it to 0 (no features). */ | |
887 | hdr = convert(&iov[0], struct virtio_net_hdr); | |
888 | hdr->flags = 0; | |
889 | hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
8ca47e00 | 890 | |
dde79789 | 891 | /* Read the packet from the device directly into the Guest's buffer. */ |
17cbca2b | 892 | len = readv(dev->fd, iov+1, in_num-1); |
8ca47e00 RR |
893 | if (len <= 0) |
894 | err(1, "reading network"); | |
dde79789 | 895 | |
56ae43df RR |
896 | /* Tell the Guest about the new packet. */ |
897 | add_used_and_trigger(fd, dev->vq, head, sizeof(*hdr) + len); | |
17cbca2b | 898 | |
8ca47e00 | 899 | verbose("tun input packet len %i [%02x %02x] (%s)\n", len, |
17cbca2b RR |
900 | ((u8 *)iov[1].iov_base)[0], ((u8 *)iov[1].iov_base)[1], |
901 | head != dev->vq->vring.num ? "sent" : "discarded"); | |
902 | ||
dde79789 | 903 | /* All good. */ |
8ca47e00 RR |
904 | return true; |
905 | } | |
906 | ||
e1e72965 RR |
907 | /*L:215 This is the callback attached to the network and console input |
908 | * virtqueues: it ensures we try again, in case we stopped console or net | |
56ae43df RR |
909 | * delivery because Guest didn't have any buffers. */ |
910 | static void enable_fd(int fd, struct virtqueue *vq) | |
911 | { | |
912 | add_device_fd(vq->dev->fd); | |
913 | /* Tell waker to listen to it again */ | |
914 | write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd)); | |
915 | } | |
916 | ||
17cbca2b RR |
917 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
918 | static void handle_output(int fd, unsigned long addr) | |
8ca47e00 RR |
919 | { |
920 | struct device *i; | |
17cbca2b RR |
921 | struct virtqueue *vq; |
922 | ||
923 | /* Check each virtqueue. */ | |
924 | for (i = devices.dev; i; i = i->next) { | |
925 | for (vq = i->vq; vq; vq = vq->next) { | |
926 | if (vq->config.pfn == addr/getpagesize() | |
927 | && vq->handle_output) { | |
928 | verbose("Output to %s\n", vq->dev->name); | |
929 | vq->handle_output(fd, vq); | |
930 | return; | |
931 | } | |
8ca47e00 RR |
932 | } |
933 | } | |
dde79789 | 934 | |
17cbca2b RR |
935 | /* Early console write is done using notify on a nul-terminated string |
936 | * in Guest memory. */ | |
937 | if (addr >= guest_limit) | |
938 | errx(1, "Bad NOTIFY %#lx", addr); | |
939 | ||
940 | write(STDOUT_FILENO, from_guest_phys(addr), | |
941 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
942 | } |
943 | ||
e1e72965 | 944 | /* This is called when the Waker wakes us up: check for incoming file |
dde79789 | 945 | * descriptors. */ |
17cbca2b | 946 | static void handle_input(int fd) |
8ca47e00 | 947 | { |
dde79789 | 948 | /* select() wants a zeroed timeval to mean "don't wait". */ |
8ca47e00 RR |
949 | struct timeval poll = { .tv_sec = 0, .tv_usec = 0 }; |
950 | ||
951 | for (;;) { | |
952 | struct device *i; | |
17cbca2b | 953 | fd_set fds = devices.infds; |
8ca47e00 | 954 | |
dde79789 | 955 | /* If nothing is ready, we're done. */ |
17cbca2b | 956 | if (select(devices.max_infd+1, &fds, NULL, NULL, &poll) == 0) |
8ca47e00 RR |
957 | break; |
958 | ||
dde79789 RR |
959 | /* Otherwise, call the device(s) which have readable |
960 | * file descriptors and a method of handling them. */ | |
17cbca2b | 961 | for (i = devices.dev; i; i = i->next) { |
8ca47e00 | 962 | if (i->handle_input && FD_ISSET(i->fd, &fds)) { |
56ae43df RR |
963 | int dev_fd; |
964 | if (i->handle_input(fd, i)) | |
965 | continue; | |
966 | ||
dde79789 | 967 | /* If handle_input() returns false, it means we |
56ae43df RR |
968 | * should no longer service it. Networking and |
969 | * console do this when there's no input | |
970 | * buffers to deliver into. Console also uses | |
971 | * it when it discovers that stdin is | |
972 | * closed. */ | |
973 | FD_CLR(i->fd, &devices.infds); | |
974 | /* Tell waker to ignore it too, by sending a | |
975 | * negative fd number (-1, since 0 is a valid | |
976 | * FD number). */ | |
977 | dev_fd = -i->fd - 1; | |
978 | write(waker_fd, &dev_fd, sizeof(dev_fd)); | |
8ca47e00 RR |
979 | } |
980 | } | |
981 | } | |
982 | } | |
983 | ||
dde79789 RR |
984 | /*L:190 |
985 | * Device Setup | |
986 | * | |
987 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
988 | * device" so the Launcher can keep track of it. We have common helper | |
989 | * routines to allocate them. | |
990 | * | |
991 | * This routine allocates a new "struct lguest_device_desc" from descriptor | |
17cbca2b RR |
992 | * table just above the Guest's normal memory. It returns a pointer to that |
993 | * descriptor. */ | |
994 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
8ca47e00 | 995 | { |
17cbca2b | 996 | struct lguest_device_desc *d; |
8ca47e00 | 997 | |
17cbca2b RR |
998 | /* We only have one page for all the descriptors. */ |
999 | if (devices.desc_used + sizeof(*d) > getpagesize()) | |
1000 | errx(1, "Too many devices"); | |
1001 | ||
1002 | /* We don't need to set config_len or status: page is 0 already. */ | |
1003 | d = (void *)devices.descpage + devices.desc_used; | |
1004 | d->type = type; | |
1005 | devices.desc_used += sizeof(*d); | |
1006 | ||
1007 | return d; | |
1008 | } | |
1009 | ||
1010 | /* Each device descriptor is followed by some configuration information. | |
e1e72965 | 1011 | * Each configuration field looks like: u8 type, u8 len, [... len bytes...]. |
17cbca2b RR |
1012 | * |
1013 | * This routine adds a new field to an existing device's descriptor. It only | |
1014 | * works for the last device, but that's OK because that's how we use it. */ | |
1015 | static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c) | |
1016 | { | |
1017 | /* This is the last descriptor, right? */ | |
1018 | assert(devices.descpage + devices.desc_used | |
1019 | == (u8 *)(dev->desc + 1) + dev->desc->config_len); | |
1020 | ||
1021 | /* We only have one page of device descriptions. */ | |
1022 | if (devices.desc_used + 2 + len > getpagesize()) | |
1023 | errx(1, "Too many devices"); | |
1024 | ||
1025 | /* Copy in the new config header: type then length. */ | |
1026 | devices.descpage[devices.desc_used++] = type; | |
1027 | devices.descpage[devices.desc_used++] = len; | |
1028 | memcpy(devices.descpage + devices.desc_used, c, len); | |
1029 | devices.desc_used += len; | |
1030 | ||
1031 | /* Update the device descriptor length: two byte head then data. */ | |
1032 | dev->desc->config_len += 2 + len; | |
1033 | } | |
1034 | ||
1035 | /* This routine adds a virtqueue to a device. We specify how many descriptors | |
1036 | * the virtqueue is to have. */ | |
1037 | static void add_virtqueue(struct device *dev, unsigned int num_descs, | |
1038 | void (*handle_output)(int fd, struct virtqueue *me)) | |
1039 | { | |
1040 | unsigned int pages; | |
1041 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1042 | void *p; | |
1043 | ||
1044 | /* First we need some pages for this virtqueue. */ | |
42b36cc0 RR |
1045 | pages = (vring_size(num_descs, getpagesize()) + getpagesize() - 1) |
1046 | / getpagesize(); | |
17cbca2b RR |
1047 | p = get_pages(pages); |
1048 | ||
d1c856e0 RR |
1049 | /* Initialize the virtqueue */ |
1050 | vq->next = NULL; | |
1051 | vq->last_avail_idx = 0; | |
1052 | vq->dev = dev; | |
1053 | ||
17cbca2b RR |
1054 | /* Initialize the configuration. */ |
1055 | vq->config.num = num_descs; | |
1056 | vq->config.irq = devices.next_irq++; | |
1057 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1058 | ||
1059 | /* Initialize the vring. */ | |
42b36cc0 | 1060 | vring_init(&vq->vring, num_descs, p, getpagesize()); |
17cbca2b RR |
1061 | |
1062 | /* Add the configuration information to this device's descriptor. */ | |
1063 | add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE, | |
1064 | sizeof(vq->config), &vq->config); | |
1065 | ||
1066 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1067 | * second. */ | |
1068 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1069 | *i = vq; | |
1070 | ||
e1e72965 RR |
1071 | /* Set the routine to call when the Guest does something to this |
1072 | * virtqueue. */ | |
17cbca2b | 1073 | vq->handle_output = handle_output; |
e1e72965 RR |
1074 | |
1075 | /* Set the "Don't Notify Me" flag if we don't have a handler */ | |
17cbca2b RR |
1076 | if (!handle_output) |
1077 | vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; | |
8ca47e00 RR |
1078 | } |
1079 | ||
17cbca2b | 1080 | /* This routine does all the creation and setup of a new device, including |
e1e72965 | 1081 | * calling new_dev_desc() to allocate the descriptor and device memory. */ |
17cbca2b RR |
1082 | static struct device *new_device(const char *name, u16 type, int fd, |
1083 | bool (*handle_input)(int, struct device *)) | |
8ca47e00 RR |
1084 | { |
1085 | struct device *dev = malloc(sizeof(*dev)); | |
1086 | ||
dde79789 RR |
1087 | /* Append to device list. Prepending to a single-linked list is |
1088 | * easier, but the user expects the devices to be arranged on the bus | |
1089 | * in command-line order. The first network device on the command line | |
e1e72965 | 1090 | * is eth0, the first block device /dev/vda, etc. */ |
17cbca2b | 1091 | *devices.lastdev = dev; |
8ca47e00 | 1092 | dev->next = NULL; |
17cbca2b | 1093 | devices.lastdev = &dev->next; |
8ca47e00 | 1094 | |
dde79789 | 1095 | /* Now we populate the fields one at a time. */ |
8ca47e00 | 1096 | dev->fd = fd; |
dde79789 RR |
1097 | /* If we have an input handler for this file descriptor, then we add it |
1098 | * to the device_list's fdset and maxfd. */ | |
8ca47e00 | 1099 | if (handle_input) |
17cbca2b RR |
1100 | add_device_fd(dev->fd); |
1101 | dev->desc = new_dev_desc(type); | |
8ca47e00 | 1102 | dev->handle_input = handle_input; |
17cbca2b | 1103 | dev->name = name; |
d1c856e0 | 1104 | dev->vq = NULL; |
8ca47e00 RR |
1105 | return dev; |
1106 | } | |
1107 | ||
dde79789 RR |
1108 | /* Our first setup routine is the console. It's a fairly simple device, but |
1109 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1110 | static void setup_console(void) |
8ca47e00 RR |
1111 | { |
1112 | struct device *dev; | |
1113 | ||
dde79789 | 1114 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1115 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1116 | struct termios term = orig_term; | |
dde79789 RR |
1117 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1118 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1119 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1120 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
dde79789 RR |
1121 | /* If we exit gracefully, the original settings will be |
1122 | * restored so the user can see what they're typing. */ | |
8ca47e00 RR |
1123 | atexit(restore_term); |
1124 | } | |
1125 | ||
17cbca2b RR |
1126 | dev = new_device("console", VIRTIO_ID_CONSOLE, |
1127 | STDIN_FILENO, handle_console_input); | |
dde79789 | 1128 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1129 | dev->priv = malloc(sizeof(struct console_abort)); |
1130 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1131 | |
56ae43df RR |
1132 | /* The console needs two virtqueues: the input then the output. When |
1133 | * they put something the input queue, we make sure we're listening to | |
1134 | * stdin. When they put something in the output queue, we write it to | |
e1e72965 | 1135 | * stdout. */ |
56ae43df | 1136 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); |
17cbca2b RR |
1137 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_console_output); |
1138 | ||
1139 | verbose("device %u: console\n", devices.device_num++); | |
8ca47e00 | 1140 | } |
17cbca2b | 1141 | /*:*/ |
8ca47e00 | 1142 | |
17cbca2b RR |
1143 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1144 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1145 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1146 | * |
17cbca2b RR |
1147 | * More sopisticated is to use one of the tools developed for project like UML |
1148 | * to do networking. | |
dde79789 | 1149 | * |
17cbca2b RR |
1150 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1151 | * completely generic ("here's my vring, attach to your vring") and would work | |
1152 | * for any traffic. Of course, namespace and permissions issues need to be | |
1153 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1154 | * multiple inter-guest channels behind one interface, although it would | |
1155 | * require some manner of hotplugging new virtio channels. | |
1156 | * | |
1157 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1158 | |
1159 | static u32 str2ip(const char *ipaddr) | |
1160 | { | |
1161 | unsigned int byte[4]; | |
1162 | ||
1163 | sscanf(ipaddr, "%u.%u.%u.%u", &byte[0], &byte[1], &byte[2], &byte[3]); | |
1164 | return (byte[0] << 24) | (byte[1] << 16) | (byte[2] << 8) | byte[3]; | |
1165 | } | |
1166 | ||
dde79789 RR |
1167 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1168 | * network device to the bridge device specified by the command line. | |
1169 | * | |
1170 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1171 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1172 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1173 | { | |
1174 | int ifidx; | |
1175 | struct ifreq ifr; | |
1176 | ||
1177 | if (!*br_name) | |
1178 | errx(1, "must specify bridge name"); | |
1179 | ||
1180 | ifidx = if_nametoindex(if_name); | |
1181 | if (!ifidx) | |
1182 | errx(1, "interface %s does not exist!", if_name); | |
1183 | ||
1184 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
1185 | ifr.ifr_ifindex = ifidx; | |
1186 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1187 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1188 | } | |
1189 | ||
dde79789 RR |
1190 | /* This sets up the Host end of the network device with an IP address, brings |
1191 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1192 | * pointer. */ |
8ca47e00 RR |
1193 | static void configure_device(int fd, const char *devname, u32 ipaddr, |
1194 | unsigned char hwaddr[6]) | |
1195 | { | |
1196 | struct ifreq ifr; | |
1197 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1198 | ||
dde79789 | 1199 | /* Don't read these incantations. Just cut & paste them like I did! */ |
8ca47e00 RR |
1200 | memset(&ifr, 0, sizeof(ifr)); |
1201 | strcpy(ifr.ifr_name, devname); | |
1202 | sin->sin_family = AF_INET; | |
1203 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1204 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
1205 | err(1, "Setting %s interface address", devname); | |
1206 | ifr.ifr_flags = IFF_UP; | |
1207 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
1208 | err(1, "Bringing interface %s up", devname); | |
1209 | ||
dde79789 RR |
1210 | /* SIOC stands for Socket I/O Control. G means Get (vs S for Set |
1211 | * above). IF means Interface, and HWADDR is hardware address. | |
1212 | * Simple! */ | |
8ca47e00 RR |
1213 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0) |
1214 | err(1, "getting hw address for %s", devname); | |
8ca47e00 RR |
1215 | memcpy(hwaddr, ifr.ifr_hwaddr.sa_data, 6); |
1216 | } | |
1217 | ||
17cbca2b RR |
1218 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or |
1219 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1220 | * packets into the Host as if they came in from a normal network card. We | |
1221 | * just shunt packets between the Guest and the tun device. */ | |
1222 | static void setup_tun_net(const char *arg) | |
8ca47e00 RR |
1223 | { |
1224 | struct device *dev; | |
1225 | struct ifreq ifr; | |
1226 | int netfd, ipfd; | |
1227 | u32 ip; | |
1228 | const char *br_name = NULL; | |
17cbca2b | 1229 | u8 hwaddr[6]; |
8ca47e00 | 1230 | |
dde79789 RR |
1231 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1232 | * tap device is like a tun device, only somehow different. To tell | |
1233 | * the truth, I completely blundered my way through this code, but it | |
1234 | * works now! */ | |
8ca47e00 RR |
1235 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
1236 | memset(&ifr, 0, sizeof(ifr)); | |
1237 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
1238 | strcpy(ifr.ifr_name, "tap%d"); | |
1239 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1240 | err(1, "configuring /dev/net/tun"); | |
dde79789 RR |
1241 | /* We don't need checksums calculated for packets coming in this |
1242 | * device: trust us! */ | |
8ca47e00 RR |
1243 | ioctl(netfd, TUNSETNOCSUM, 1); |
1244 | ||
17cbca2b RR |
1245 | /* First we create a new network device. */ |
1246 | dev = new_device("net", VIRTIO_ID_NET, netfd, handle_tun_input); | |
dde79789 | 1247 | |
56ae43df RR |
1248 | /* Network devices need a receive and a send queue, just like |
1249 | * console. */ | |
1250 | add_virtqueue(dev, VIRTQUEUE_NUM, enable_fd); | |
17cbca2b | 1251 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_net_output); |
8ca47e00 | 1252 | |
dde79789 RR |
1253 | /* We need a socket to perform the magic network ioctls to bring up the |
1254 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1255 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1256 | if (ipfd < 0) | |
1257 | err(1, "opening IP socket"); | |
1258 | ||
dde79789 | 1259 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 RR |
1260 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
1261 | ip = INADDR_ANY; | |
1262 | br_name = arg + strlen(BRIDGE_PFX); | |
1263 | add_to_bridge(ipfd, ifr.ifr_name, br_name); | |
dde79789 | 1264 | } else /* It is an IP address to set up the device with */ |
8ca47e00 RR |
1265 | ip = str2ip(arg); |
1266 | ||
17cbca2b RR |
1267 | /* Set up the tun device, and get the mac address for the interface. */ |
1268 | configure_device(ipfd, ifr.ifr_name, ip, hwaddr); | |
8ca47e00 | 1269 | |
17cbca2b RR |
1270 | /* Tell Guest what MAC address to use. */ |
1271 | add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr); | |
8ca47e00 | 1272 | |
17cbca2b | 1273 | /* We don't seed the socket any more; setup is done. */ |
8ca47e00 RR |
1274 | close(ipfd); |
1275 | ||
17cbca2b RR |
1276 | verbose("device %u: tun net %u.%u.%u.%u\n", |
1277 | devices.device_num++, | |
1278 | (u8)(ip>>24),(u8)(ip>>16),(u8)(ip>>8),(u8)ip); | |
8ca47e00 RR |
1279 | if (br_name) |
1280 | verbose("attached to bridge: %s\n", br_name); | |
1281 | } | |
17cbca2b | 1282 | |
e1e72965 RR |
1283 | /* Our block (disk) device should be really simple: the Guest asks for a block |
1284 | * number and we read or write that position in the file. Unfortunately, that | |
1285 | * was amazingly slow: the Guest waits until the read is finished before | |
1286 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1287 | * |
e1e72965 RR |
1288 | * We could use async I/O, except it's reputed to suck so hard that characters |
1289 | * actually go missing from your code when you try to use it. | |
17cbca2b RR |
1290 | * |
1291 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1292 | ||
e1e72965 | 1293 | /* This hangs off device->priv. */ |
17cbca2b RR |
1294 | struct vblk_info |
1295 | { | |
1296 | /* The size of the file. */ | |
1297 | off64_t len; | |
1298 | ||
1299 | /* The file descriptor for the file. */ | |
1300 | int fd; | |
1301 | ||
1302 | /* IO thread listens on this file descriptor [0]. */ | |
1303 | int workpipe[2]; | |
1304 | ||
1305 | /* IO thread writes to this file descriptor to mark it done, then | |
1306 | * Launcher triggers interrupt to Guest. */ | |
1307 | int done_fd; | |
1308 | }; | |
e1e72965 | 1309 | /*:*/ |
17cbca2b | 1310 | |
e1e72965 RR |
1311 | /*L:210 |
1312 | * The Disk | |
1313 | * | |
1314 | * Remember that the block device is handled by a separate I/O thread. We head | |
1315 | * straight into the core of that thread here: | |
1316 | */ | |
17cbca2b RR |
1317 | static bool service_io(struct device *dev) |
1318 | { | |
1319 | struct vblk_info *vblk = dev->priv; | |
1320 | unsigned int head, out_num, in_num, wlen; | |
1321 | int ret; | |
1322 | struct virtio_blk_inhdr *in; | |
1323 | struct virtio_blk_outhdr *out; | |
1324 | struct iovec iov[dev->vq->vring.num]; | |
1325 | off64_t off; | |
1326 | ||
e1e72965 | 1327 | /* See if there's a request waiting. If not, nothing to do. */ |
17cbca2b RR |
1328 | head = get_vq_desc(dev->vq, iov, &out_num, &in_num); |
1329 | if (head == dev->vq->vring.num) | |
1330 | return false; | |
1331 | ||
e1e72965 RR |
1332 | /* Every block request should contain at least one output buffer |
1333 | * (detailing the location on disk and the type of request) and one | |
1334 | * input buffer (to hold the result). */ | |
17cbca2b RR |
1335 | if (out_num == 0 || in_num == 0) |
1336 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1337 | head, out_num, in_num); | |
1338 | ||
1339 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
1340 | in = convert(&iov[out_num+in_num-1], struct virtio_blk_inhdr); | |
1341 | off = out->sector * 512; | |
1342 | ||
e1e72965 RR |
1343 | /* The block device implements "barriers", where the Guest indicates |
1344 | * that it wants all previous writes to occur before this write. We | |
1345 | * don't have a way of asking our kernel to do a barrier, so we just | |
1346 | * synchronize all the data in the file. Pretty poor, no? */ | |
17cbca2b RR |
1347 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1348 | fdatasync(vblk->fd); | |
1349 | ||
e1e72965 RR |
1350 | /* In general the virtio block driver is allowed to try SCSI commands. |
1351 | * It'd be nice if we supported eject, for example, but we don't. */ | |
17cbca2b RR |
1352 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1353 | fprintf(stderr, "Scsi commands unsupported\n"); | |
1354 | in->status = VIRTIO_BLK_S_UNSUPP; | |
1200e646 | 1355 | wlen = sizeof(*in); |
17cbca2b RR |
1356 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
1357 | /* Write */ | |
1358 | ||
1359 | /* Move to the right location in the block file. This can fail | |
1360 | * if they try to write past end. */ | |
1361 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1362 | err(1, "Bad seek to sector %llu", out->sector); | |
1363 | ||
1364 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1365 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1366 | ||
1367 | /* Grr... Now we know how long the descriptor they sent was, we | |
1368 | * make sure they didn't try to write over the end of the block | |
1369 | * file (possibly extending it). */ | |
1370 | if (ret > 0 && off + ret > vblk->len) { | |
1371 | /* Trim it back to the correct length */ | |
1372 | ftruncate64(vblk->fd, vblk->len); | |
1373 | /* Die, bad Guest, die. */ | |
1374 | errx(1, "Write past end %llu+%u", off, ret); | |
1375 | } | |
1200e646 | 1376 | wlen = sizeof(*in); |
17cbca2b RR |
1377 | in->status = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
1378 | } else { | |
1379 | /* Read */ | |
1380 | ||
1381 | /* Move to the right location in the block file. This can fail | |
1382 | * if they try to read past end. */ | |
1383 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1384 | err(1, "Bad seek to sector %llu", out->sector); | |
1385 | ||
1386 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1387 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1388 | if (ret >= 0) { | |
1200e646 | 1389 | wlen = sizeof(*in) + ret; |
17cbca2b RR |
1390 | in->status = VIRTIO_BLK_S_OK; |
1391 | } else { | |
1200e646 | 1392 | wlen = sizeof(*in); |
17cbca2b RR |
1393 | in->status = VIRTIO_BLK_S_IOERR; |
1394 | } | |
1395 | } | |
1396 | ||
1397 | /* We can't trigger an IRQ, because we're not the Launcher. It does | |
1398 | * that when we tell it we're done. */ | |
1399 | add_used(dev->vq, head, wlen); | |
1400 | return true; | |
1401 | } | |
1402 | ||
1403 | /* This is the thread which actually services the I/O. */ | |
1404 | static int io_thread(void *_dev) | |
1405 | { | |
1406 | struct device *dev = _dev; | |
1407 | struct vblk_info *vblk = dev->priv; | |
1408 | char c; | |
1409 | ||
1410 | /* Close other side of workpipe so we get 0 read when main dies. */ | |
1411 | close(vblk->workpipe[1]); | |
1412 | /* Close the other side of the done_fd pipe. */ | |
1413 | close(dev->fd); | |
1414 | ||
1415 | /* When this read fails, it means Launcher died, so we follow. */ | |
1416 | while (read(vblk->workpipe[0], &c, 1) == 1) { | |
e1e72965 | 1417 | /* We acknowledge each request immediately to reduce latency, |
17cbca2b RR |
1418 | * rather than waiting until we've done them all. I haven't |
1419 | * measured to see if it makes any difference. */ | |
1420 | while (service_io(dev)) | |
1421 | write(vblk->done_fd, &c, 1); | |
1422 | } | |
1423 | return 0; | |
1424 | } | |
1425 | ||
e1e72965 RR |
1426 | /* Now we've seen the I/O thread, we return to the Launcher to see what happens |
1427 | * when the thread tells us it's completed some I/O. */ | |
17cbca2b RR |
1428 | static bool handle_io_finish(int fd, struct device *dev) |
1429 | { | |
1430 | char c; | |
1431 | ||
e1e72965 RR |
1432 | /* If the I/O thread died, presumably it printed the error, so we |
1433 | * simply exit. */ | |
17cbca2b RR |
1434 | if (read(dev->fd, &c, 1) != 1) |
1435 | exit(1); | |
1436 | ||
1437 | /* It did some work, so trigger the irq. */ | |
1438 | trigger_irq(fd, dev->vq); | |
1439 | return true; | |
1440 | } | |
1441 | ||
e1e72965 | 1442 | /* When the Guest submits some I/O, we just need to wake the I/O thread. */ |
17cbca2b RR |
1443 | static void handle_virtblk_output(int fd, struct virtqueue *vq) |
1444 | { | |
1445 | struct vblk_info *vblk = vq->dev->priv; | |
1446 | char c = 0; | |
1447 | ||
1448 | /* Wake up I/O thread and tell it to go to work! */ | |
1449 | if (write(vblk->workpipe[1], &c, 1) != 1) | |
1450 | /* Presumably it indicated why it died. */ | |
1451 | exit(1); | |
1452 | } | |
1453 | ||
e1e72965 | 1454 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1455 | static void setup_block_file(const char *filename) |
1456 | { | |
1457 | int p[2]; | |
1458 | struct device *dev; | |
1459 | struct vblk_info *vblk; | |
1460 | void *stack; | |
1461 | u64 cap; | |
1462 | unsigned int val; | |
1463 | ||
1464 | /* This is the pipe the I/O thread will use to tell us I/O is done. */ | |
1465 | pipe(p); | |
1466 | ||
1467 | /* The device responds to return from I/O thread. */ | |
1468 | dev = new_device("block", VIRTIO_ID_BLOCK, p[0], handle_io_finish); | |
1469 | ||
e1e72965 | 1470 | /* The device has one virtqueue, where the Guest places requests. */ |
17cbca2b RR |
1471 | add_virtqueue(dev, VIRTQUEUE_NUM, handle_virtblk_output); |
1472 | ||
1473 | /* Allocate the room for our own bookkeeping */ | |
1474 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1475 | ||
1476 | /* First we open the file and store the length. */ | |
1477 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1478 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1479 | ||
1480 | /* Tell Guest how many sectors this device has. */ | |
1481 | cap = cpu_to_le64(vblk->len / 512); | |
1482 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap); | |
1483 | ||
1484 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1485 | * for the in and out elements. */ | |
1486 | val = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1487 | add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val); | |
1488 | ||
1489 | /* The I/O thread writes to this end of the pipe when done. */ | |
1490 | vblk->done_fd = p[1]; | |
1491 | ||
e1e72965 RR |
1492 | /* This is the second pipe, which is how we tell the I/O thread about |
1493 | * more work. */ | |
17cbca2b RR |
1494 | pipe(vblk->workpipe); |
1495 | ||
1496 | /* Create stack for thread and run it */ | |
1497 | stack = malloc(32768); | |
ec04b13f BR |
1498 | /* SIGCHLD - We dont "wait" for our cloned thread, so prevent it from |
1499 | * becoming a zombie. */ | |
1500 | if (clone(io_thread, stack + 32768, CLONE_VM | SIGCHLD, dev) == -1) | |
17cbca2b RR |
1501 | err(1, "Creating clone"); |
1502 | ||
1503 | /* We don't need to keep the I/O thread's end of the pipes open. */ | |
1504 | close(vblk->done_fd); | |
1505 | close(vblk->workpipe[0]); | |
1506 | ||
1507 | verbose("device %u: virtblock %llu sectors\n", | |
1508 | devices.device_num, cap); | |
1509 | } | |
ec04b13f BR |
1510 | /* That's the end of device setup. :*/ |
1511 | ||
1512 | /* Reboot */ | |
1513 | static void __attribute__((noreturn)) restart_guest(void) | |
1514 | { | |
1515 | unsigned int i; | |
1516 | ||
1517 | /* Closing pipes causes the waker thread and io_threads to die, and | |
1518 | * closing /dev/lguest cleans up the Guest. Since we don't track all | |
1519 | * open fds, we simply close everything beyond stderr. */ | |
1520 | for (i = 3; i < FD_SETSIZE; i++) | |
1521 | close(i); | |
1522 | execv(main_args[0], main_args); | |
1523 | err(1, "Could not exec %s", main_args[0]); | |
1524 | } | |
8ca47e00 | 1525 | |
dde79789 RR |
1526 | /*L:220 Finally we reach the core of the Launcher, which runs the Guest, serves |
1527 | * its input and output, and finally, lays it to rest. */ | |
17cbca2b | 1528 | static void __attribute__((noreturn)) run_guest(int lguest_fd) |
8ca47e00 RR |
1529 | { |
1530 | for (;;) { | |
511801dc | 1531 | unsigned long args[] = { LHREQ_BREAK, 0 }; |
17cbca2b | 1532 | unsigned long notify_addr; |
8ca47e00 RR |
1533 | int readval; |
1534 | ||
1535 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1536 | readval = pread(lguest_fd, ¬ify_addr, |
1537 | sizeof(notify_addr), cpu_id); | |
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 }; | |
e3283fa0 | 1547 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1548 | errx(1, "%s", reason); |
ec04b13f BR |
1549 | /* ERESTART means that we need to reboot the guest */ |
1550 | } else if (errno == ERESTART) { | |
1551 | restart_guest(); | |
e1e72965 | 1552 | /* EAGAIN means the Waker wanted us to look at some input. |
dde79789 | 1553 | * Anything else means a bug or incompatible change. */ |
8ca47e00 RR |
1554 | } else if (errno != EAGAIN) |
1555 | err(1, "Running guest failed"); | |
dde79789 | 1556 | |
e3283fa0 GOC |
1557 | /* Only service input on thread for CPU 0. */ |
1558 | if (cpu_id != 0) | |
1559 | continue; | |
1560 | ||
e1e72965 | 1561 | /* Service input, then unset the BREAK to release the Waker. */ |
17cbca2b | 1562 | handle_input(lguest_fd); |
e3283fa0 | 1563 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) |
8ca47e00 RR |
1564 | err(1, "Resetting break"); |
1565 | } | |
1566 | } | |
dde79789 | 1567 | /* |
e1e72965 RR |
1568 | * This is the end of the Launcher. The good news: we are over halfway |
1569 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1570 | * of us. | |
dde79789 | 1571 | * |
e1e72965 RR |
1572 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1573 | * "make Host". | |
1574 | :*/ | |
8ca47e00 RR |
1575 | |
1576 | static struct option opts[] = { | |
1577 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1578 | { "tunnet", 1, NULL, 't' }, |
1579 | { "block", 1, NULL, 'b' }, | |
1580 | { "initrd", 1, NULL, 'i' }, | |
1581 | { NULL }, | |
1582 | }; | |
1583 | static void usage(void) | |
1584 | { | |
1585 | errx(1, "Usage: lguest [--verbose] " | |
17cbca2b | 1586 | "[--tunnet=(<ipaddr>|bridge:<bridgename>)\n" |
8ca47e00 RR |
1587 | "|--block=<filename>|--initrd=<filename>]...\n" |
1588 | "<mem-in-mb> vmlinux [args...]"); | |
1589 | } | |
1590 | ||
3c6b5bfa | 1591 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1592 | int main(int argc, char *argv[]) |
1593 | { | |
47436aa4 RR |
1594 | /* Memory, top-level pagetable, code startpoint and size of the |
1595 | * (optional) initrd. */ | |
1596 | unsigned long mem = 0, pgdir, start, initrd_size = 0; | |
e1e72965 | 1597 | /* Two temporaries and the /dev/lguest file descriptor. */ |
6570c459 | 1598 | int i, c, lguest_fd; |
3c6b5bfa | 1599 | /* The boot information for the Guest. */ |
43d33b21 | 1600 | struct boot_params *boot; |
dde79789 | 1601 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1602 | const char *initrd_name = NULL; |
1603 | ||
ec04b13f BR |
1604 | /* Save the args: we "reboot" by execing ourselves again. */ |
1605 | main_args = argv; | |
1606 | /* We don't "wait" for the children, so prevent them from becoming | |
1607 | * zombies. */ | |
1608 | signal(SIGCHLD, SIG_IGN); | |
1609 | ||
dde79789 RR |
1610 | /* First we initialize the device list. Since console and network |
1611 | * device receive input from a file descriptor, we keep an fdset | |
1612 | * (infds) and the maximum fd number (max_infd) with the head of the | |
1613 | * list. We also keep a pointer to the last device, for easy appending | |
17cbca2b RR |
1614 | * to the list. Finally, we keep the next interrupt number to hand out |
1615 | * (1: remember that 0 is used by the timer). */ | |
1616 | FD_ZERO(&devices.infds); | |
1617 | devices.max_infd = -1; | |
1618 | devices.lastdev = &devices.dev; | |
1619 | devices.next_irq = 1; | |
8ca47e00 | 1620 | |
e3283fa0 | 1621 | cpu_id = 0; |
dde79789 RR |
1622 | /* We need to know how much memory so we can set up the device |
1623 | * descriptor and memory pages for the devices as we parse the command | |
1624 | * line. So we quickly look through the arguments to find the amount | |
1625 | * of memory now. */ | |
6570c459 RR |
1626 | for (i = 1; i < argc; i++) { |
1627 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1628 | mem = atoi(argv[i]) * 1024 * 1024; |
1629 | /* We start by mapping anonymous pages over all of | |
1630 | * guest-physical memory range. This fills it with 0, | |
1631 | * and ensures that the Guest won't be killed when it | |
1632 | * tries to access it. */ | |
1633 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1634 | + DEVICE_PAGES); | |
1635 | guest_limit = mem; | |
1636 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1637 | devices.descpage = get_pages(1); |
6570c459 RR |
1638 | break; |
1639 | } | |
1640 | } | |
dde79789 RR |
1641 | |
1642 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1643 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1644 | switch (c) { | |
1645 | case 'v': | |
1646 | verbose = true; | |
1647 | break; | |
8ca47e00 | 1648 | case 't': |
17cbca2b | 1649 | setup_tun_net(optarg); |
8ca47e00 RR |
1650 | break; |
1651 | case 'b': | |
17cbca2b | 1652 | setup_block_file(optarg); |
8ca47e00 RR |
1653 | break; |
1654 | case 'i': | |
1655 | initrd_name = optarg; | |
1656 | break; | |
1657 | default: | |
1658 | warnx("Unknown argument %s", argv[optind]); | |
1659 | usage(); | |
1660 | } | |
1661 | } | |
dde79789 RR |
1662 | /* After the other arguments we expect memory and kernel image name, |
1663 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1664 | if (optind + 2 > argc) |
1665 | usage(); | |
1666 | ||
3c6b5bfa RR |
1667 | verbose("Guest base is at %p\n", guest_base); |
1668 | ||
dde79789 | 1669 | /* We always have a console device */ |
17cbca2b | 1670 | setup_console(); |
8ca47e00 | 1671 | |
8ca47e00 | 1672 | /* Now we load the kernel */ |
47436aa4 | 1673 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1674 | |
3c6b5bfa RR |
1675 | /* Boot information is stashed at physical address 0 */ |
1676 | boot = from_guest_phys(0); | |
1677 | ||
dde79789 | 1678 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1679 | if (initrd_name) { |
1680 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1681 | /* These are the location in the Linux boot header where the |
1682 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
1683 | boot->hdr.ramdisk_image = mem - initrd_size; |
1684 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 1685 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 1686 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
1687 | } |
1688 | ||
dde79789 | 1689 | /* Set up the initial linear pagetables, starting below the initrd. */ |
47436aa4 | 1690 | pgdir = setup_pagetables(mem, initrd_size); |
8ca47e00 | 1691 | |
dde79789 RR |
1692 | /* The Linux boot header contains an "E820" memory map: ours is a |
1693 | * simple, single region. */ | |
43d33b21 RR |
1694 | boot->e820_entries = 1; |
1695 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 1696 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
1697 | * line after the boot header. */ |
1698 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
e1e72965 | 1699 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 1700 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 1701 | |
814a0e5c | 1702 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 1703 | boot->hdr.version = 0x207; |
814a0e5c RR |
1704 | |
1705 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 1706 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 1707 | |
43d33b21 RR |
1708 | /* Tell the entry path not to try to reload segment registers. */ |
1709 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 1710 | |
dde79789 RR |
1711 | /* We tell the kernel to initialize the Guest: this returns the open |
1712 | * /dev/lguest file descriptor. */ | |
47436aa4 | 1713 | lguest_fd = tell_kernel(pgdir, start); |
dde79789 RR |
1714 | |
1715 | /* We fork off a child process, which wakes the Launcher whenever one | |
1716 | * of the input file descriptors needs attention. Otherwise we would | |
1717 | * run the Guest until it tries to output something. */ | |
17cbca2b | 1718 | waker_fd = setup_waker(lguest_fd); |
8ca47e00 | 1719 | |
dde79789 | 1720 | /* Finally, run the Guest. This doesn't return. */ |
17cbca2b | 1721 | run_guest(lguest_fd); |
8ca47e00 | 1722 | } |
f56a384e RR |
1723 | /*:*/ |
1724 | ||
1725 | /*M:999 | |
1726 | * Mastery is done: you now know everything I do. | |
1727 | * | |
1728 | * But surely you have seen code, features and bugs in your wanderings which | |
1729 | * you now yearn to attack? That is the real game, and I look forward to you | |
1730 | * patching and forking lguest into the Your-Name-Here-visor. | |
1731 | * | |
1732 | * Farewell, and good coding! | |
1733 | * Rusty Russell. | |
1734 | */ |