Commit | Line | Data |
---|---|---|
f938d2c8 | 1 | /*P:100 This is the Launcher code, a simple program which lays out the |
a6bd8e13 RR |
2 | * "physical" memory for the new Guest by mapping the kernel image and |
3 | * the virtual devices, then opens /dev/lguest to tell the kernel | |
4 | * about the Guest and control it. :*/ | |
8ca47e00 RR |
5 | #define _LARGEFILE64_SOURCE |
6 | #define _GNU_SOURCE | |
7 | #include <stdio.h> | |
8 | #include <string.h> | |
9 | #include <unistd.h> | |
10 | #include <err.h> | |
11 | #include <stdint.h> | |
12 | #include <stdlib.h> | |
13 | #include <elf.h> | |
14 | #include <sys/mman.h> | |
6649bb7a | 15 | #include <sys/param.h> |
8ca47e00 RR |
16 | #include <sys/types.h> |
17 | #include <sys/stat.h> | |
18 | #include <sys/wait.h> | |
659a0e66 | 19 | #include <sys/eventfd.h> |
8ca47e00 RR |
20 | #include <fcntl.h> |
21 | #include <stdbool.h> | |
22 | #include <errno.h> | |
23 | #include <ctype.h> | |
24 | #include <sys/socket.h> | |
25 | #include <sys/ioctl.h> | |
26 | #include <sys/time.h> | |
27 | #include <time.h> | |
28 | #include <netinet/in.h> | |
29 | #include <net/if.h> | |
30 | #include <linux/sockios.h> | |
31 | #include <linux/if_tun.h> | |
32 | #include <sys/uio.h> | |
33 | #include <termios.h> | |
34 | #include <getopt.h> | |
35 | #include <zlib.h> | |
17cbca2b RR |
36 | #include <assert.h> |
37 | #include <sched.h> | |
a586d4f6 RR |
38 | #include <limits.h> |
39 | #include <stddef.h> | |
a161883a | 40 | #include <signal.h> |
b45d8cb0 | 41 | #include "linux/lguest_launcher.h" |
17cbca2b RR |
42 | #include "linux/virtio_config.h" |
43 | #include "linux/virtio_net.h" | |
44 | #include "linux/virtio_blk.h" | |
45 | #include "linux/virtio_console.h" | |
28fd6d7f | 46 | #include "linux/virtio_rng.h" |
17cbca2b | 47 | #include "linux/virtio_ring.h" |
d5d02d6d | 48 | #include "asm/bootparam.h" |
a6bd8e13 | 49 | /*L:110 We can ignore the 39 include files we need for this program, but I do |
db24e8c2 RR |
50 | * want to draw attention to the use of kernel-style types. |
51 | * | |
52 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
53 | * like these abbreviations, so we define them here. Note that u64 is always | |
54 | * unsigned long long, which works on all Linux systems: this means that we can | |
55 | * use %llu in printf for any u64. */ | |
56 | typedef unsigned long long u64; | |
57 | typedef uint32_t u32; | |
58 | typedef uint16_t u16; | |
59 | typedef uint8_t u8; | |
dde79789 | 60 | /*:*/ |
8ca47e00 RR |
61 | |
62 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ | |
8ca47e00 RR |
63 | #define BRIDGE_PFX "bridge:" |
64 | #ifndef SIOCBRADDIF | |
65 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
66 | #endif | |
3c6b5bfa RR |
67 | /* We can have up to 256 pages for devices. */ |
68 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
69 | /* This will occupy 3 pages: it must be a power of 2. */ |
70 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 71 | |
dde79789 RR |
72 | /*L:120 verbose is both a global flag and a macro. The C preprocessor allows |
73 | * this, and although I wouldn't recommend it, it works quite nicely here. */ | |
8ca47e00 RR |
74 | static bool verbose; |
75 | #define verbose(args...) \ | |
76 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
77 | /*:*/ |
78 | ||
3c6b5bfa RR |
79 | /* The pointer to the start of guest memory. */ |
80 | static void *guest_base; | |
81 | /* The maximum guest physical address allowed, and maximum possible. */ | |
82 | static unsigned long guest_limit, guest_max; | |
56739c80 RR |
83 | /* The /dev/lguest file descriptor. */ |
84 | static int lguest_fd; | |
8ca47e00 | 85 | |
e3283fa0 GOC |
86 | /* a per-cpu variable indicating whose vcpu is currently running */ |
87 | static unsigned int __thread cpu_id; | |
88 | ||
dde79789 | 89 | /* This is our list of devices. */ |
8ca47e00 RR |
90 | struct device_list |
91 | { | |
17cbca2b RR |
92 | /* Counter to assign interrupt numbers. */ |
93 | unsigned int next_irq; | |
94 | ||
95 | /* Counter to print out convenient device numbers. */ | |
96 | unsigned int device_num; | |
97 | ||
dde79789 | 98 | /* The descriptor page for the devices. */ |
17cbca2b RR |
99 | u8 *descpage; |
100 | ||
dde79789 | 101 | /* A single linked list of devices. */ |
8ca47e00 | 102 | struct device *dev; |
a586d4f6 RR |
103 | /* And a pointer to the last device for easy append and also for |
104 | * configuration appending. */ | |
105 | struct device *lastdev; | |
8ca47e00 RR |
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 | 116 | |
713b15b3 | 117 | /* The device's descriptor, as mapped into the Guest. */ |
8ca47e00 | 118 | struct lguest_device_desc *desc; |
17cbca2b | 119 | |
713b15b3 RR |
120 | /* We can't trust desc values once Guest has booted: we use these. */ |
121 | unsigned int feature_len; | |
122 | unsigned int num_vq; | |
123 | ||
17cbca2b RR |
124 | /* The name of this device, for --verbose. */ |
125 | const char *name; | |
8ca47e00 | 126 | |
17cbca2b RR |
127 | /* Any queues attached to this device */ |
128 | struct virtqueue *vq; | |
8ca47e00 | 129 | |
659a0e66 RR |
130 | /* Is it operational */ |
131 | bool running; | |
a007a751 | 132 | |
8ca47e00 RR |
133 | /* Device-specific data. */ |
134 | void *priv; | |
135 | }; | |
136 | ||
17cbca2b RR |
137 | /* The virtqueue structure describes a queue attached to a device. */ |
138 | struct virtqueue | |
139 | { | |
140 | struct virtqueue *next; | |
141 | ||
142 | /* Which device owns me. */ | |
143 | struct device *dev; | |
144 | ||
145 | /* The configuration for this queue. */ | |
146 | struct lguest_vqconfig config; | |
147 | ||
148 | /* The actual ring of buffers. */ | |
149 | struct vring vring; | |
150 | ||
151 | /* Last available index we saw. */ | |
152 | u16 last_avail_idx; | |
153 | ||
95c517c0 RR |
154 | /* How many are used since we sent last irq? */ |
155 | unsigned int pending_used; | |
156 | ||
659a0e66 RR |
157 | /* Eventfd where Guest notifications arrive. */ |
158 | int eventfd; | |
20887611 | 159 | |
659a0e66 RR |
160 | /* Function for the thread which is servicing this virtqueue. */ |
161 | void (*service)(struct virtqueue *vq); | |
162 | pid_t thread; | |
17cbca2b RR |
163 | }; |
164 | ||
ec04b13f BR |
165 | /* Remember the arguments to the program so we can "reboot" */ |
166 | static char **main_args; | |
167 | ||
659a0e66 RR |
168 | /* The original tty settings to restore on exit. */ |
169 | static struct termios orig_term; | |
170 | ||
f7027c63 RR |
171 | /* We have to be careful with barriers: our devices are all run in separate |
172 | * threads and so we need to make sure that changes visible to the Guest happen | |
173 | * in precise order. */ | |
174 | #define wmb() __asm__ __volatile__("" : : : "memory") | |
b60da13f | 175 | #define mb() __asm__ __volatile__("" : : : "memory") |
17cbca2b RR |
176 | |
177 | /* Convert an iovec element to the given type. | |
178 | * | |
179 | * This is a fairly ugly trick: we need to know the size of the type and | |
180 | * alignment requirement to check the pointer is kosher. It's also nice to | |
181 | * have the name of the type in case we report failure. | |
182 | * | |
183 | * Typing those three things all the time is cumbersome and error prone, so we | |
184 | * have a macro which sets them all up and passes to the real function. */ | |
185 | #define convert(iov, type) \ | |
186 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) | |
187 | ||
188 | static void *_convert(struct iovec *iov, size_t size, size_t align, | |
189 | const char *name) | |
190 | { | |
191 | if (iov->iov_len != size) | |
192 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); | |
193 | if ((unsigned long)iov->iov_base % align != 0) | |
194 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); | |
195 | return iov->iov_base; | |
196 | } | |
197 | ||
b5111790 RR |
198 | /* Wrapper for the last available index. Makes it easier to change. */ |
199 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
200 | ||
17cbca2b RR |
201 | /* The virtio configuration space is defined to be little-endian. x86 is |
202 | * little-endian too, but it's nice to be explicit so we have these helpers. */ | |
203 | #define cpu_to_le16(v16) (v16) | |
204 | #define cpu_to_le32(v32) (v32) | |
205 | #define cpu_to_le64(v64) (v64) | |
206 | #define le16_to_cpu(v16) (v16) | |
207 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 208 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 209 | |
28fd6d7f RR |
210 | /* Is this iovec empty? */ |
211 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
212 | { | |
213 | unsigned int i; | |
214 | ||
215 | for (i = 0; i < num_iov; i++) | |
216 | if (iov[i].iov_len) | |
217 | return false; | |
218 | return true; | |
219 | } | |
220 | ||
221 | /* Take len bytes from the front of this iovec. */ | |
222 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) | |
223 | { | |
224 | unsigned int i; | |
225 | ||
226 | for (i = 0; i < num_iov; i++) { | |
227 | unsigned int used; | |
228 | ||
229 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
230 | iov[i].iov_base += used; | |
231 | iov[i].iov_len -= used; | |
232 | len -= used; | |
233 | } | |
234 | assert(len == 0); | |
235 | } | |
236 | ||
6e5aa7ef RR |
237 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
238 | static u8 *get_feature_bits(struct device *dev) | |
239 | { | |
240 | return (u8 *)(dev->desc + 1) | |
713b15b3 | 241 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
6e5aa7ef RR |
242 | } |
243 | ||
3c6b5bfa RR |
244 | /*L:100 The Launcher code itself takes us out into userspace, that scary place |
245 | * where pointers run wild and free! Unfortunately, like most userspace | |
246 | * programs, it's quite boring (which is why everyone likes to hack on the | |
247 | * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it | |
248 | * will get you through this section. Or, maybe not. | |
249 | * | |
250 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
251 | * memory and stores it in "guest_base". In other words, Guest physical == | |
252 | * Launcher virtual with an offset. | |
253 | * | |
254 | * This can be tough to get your head around, but usually it just means that we | |
255 | * use these trivial conversion functions when the Guest gives us it's | |
256 | * "physical" addresses: */ | |
257 | static void *from_guest_phys(unsigned long addr) | |
258 | { | |
259 | return guest_base + addr; | |
260 | } | |
261 | ||
262 | static unsigned long to_guest_phys(const void *addr) | |
263 | { | |
264 | return (addr - guest_base); | |
265 | } | |
266 | ||
dde79789 RR |
267 | /*L:130 |
268 | * Loading the Kernel. | |
269 | * | |
270 | * We start with couple of simple helper routines. open_or_die() avoids | |
271 | * error-checking code cluttering the callers: */ | |
8ca47e00 RR |
272 | static int open_or_die(const char *name, int flags) |
273 | { | |
274 | int fd = open(name, flags); | |
275 | if (fd < 0) | |
276 | err(1, "Failed to open %s", name); | |
277 | return fd; | |
278 | } | |
279 | ||
3c6b5bfa RR |
280 | /* map_zeroed_pages() takes a number of pages. */ |
281 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 282 | { |
3c6b5bfa RR |
283 | int fd = open_or_die("/dev/zero", O_RDONLY); |
284 | void *addr; | |
8ca47e00 | 285 | |
dde79789 | 286 | /* We use a private mapping (ie. if we write to the page, it will be |
3c6b5bfa RR |
287 | * copied). */ |
288 | addr = mmap(NULL, getpagesize() * num, | |
289 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); | |
290 | if (addr == MAP_FAILED) | |
291 | err(1, "Mmaping %u pages of /dev/zero", num); | |
34bdaab4 | 292 | close(fd); |
3c6b5bfa RR |
293 | |
294 | return addr; | |
295 | } | |
296 | ||
297 | /* Get some more pages for a device. */ | |
298 | static void *get_pages(unsigned int num) | |
299 | { | |
300 | void *addr = from_guest_phys(guest_limit); | |
301 | ||
302 | guest_limit += num * getpagesize(); | |
303 | if (guest_limit > guest_max) | |
304 | errx(1, "Not enough memory for devices"); | |
305 | return addr; | |
8ca47e00 RR |
306 | } |
307 | ||
6649bb7a RM |
308 | /* This routine is used to load the kernel or initrd. It tries mmap, but if |
309 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), | |
310 | * it falls back to reading the memory in. */ | |
311 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) | |
312 | { | |
313 | ssize_t r; | |
314 | ||
315 | /* We map writable even though for some segments are marked read-only. | |
316 | * The kernel really wants to be writable: it patches its own | |
317 | * instructions. | |
318 | * | |
319 | * MAP_PRIVATE means that the page won't be copied until a write is | |
320 | * done to it. This allows us to share untouched memory between | |
321 | * Guests. */ | |
322 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, | |
323 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) | |
324 | return; | |
325 | ||
326 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
327 | r = pread(fd, addr, len, offset); | |
328 | if (r != len) | |
329 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
330 | } | |
331 | ||
dde79789 RR |
332 | /* This routine takes an open vmlinux image, which is in ELF, and maps it into |
333 | * the Guest memory. ELF = Embedded Linking Format, which is the format used | |
334 | * by all modern binaries on Linux including the kernel. | |
335 | * | |
336 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
337 | * address. We use the physical address; the Guest will map itself to the |
338 | * virtual address. | |
dde79789 RR |
339 | * |
340 | * We return the starting address. */ | |
47436aa4 | 341 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 342 | { |
8ca47e00 RR |
343 | Elf32_Phdr phdr[ehdr->e_phnum]; |
344 | unsigned int i; | |
8ca47e00 | 345 | |
dde79789 RR |
346 | /* Sanity checks on the main ELF header: an x86 executable with a |
347 | * reasonable number of correctly-sized program headers. */ | |
8ca47e00 RR |
348 | if (ehdr->e_type != ET_EXEC |
349 | || ehdr->e_machine != EM_386 | |
350 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
351 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
352 | errx(1, "Malformed elf header"); | |
353 | ||
dde79789 RR |
354 | /* An ELF executable contains an ELF header and a number of "program" |
355 | * headers which indicate which parts ("segments") of the program to | |
356 | * load where. */ | |
357 | ||
358 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
359 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
360 | err(1, "Seeking to program headers"); | |
361 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
362 | err(1, "Reading program headers"); | |
363 | ||
dde79789 | 364 | /* Try all the headers: there are usually only three. A read-only one, |
a6bd8e13 | 365 | * a read-write one, and a "note" section which we don't load. */ |
8ca47e00 | 366 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 367 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
368 | if (phdr[i].p_type != PT_LOAD) |
369 | continue; | |
370 | ||
371 | verbose("Section %i: size %i addr %p\n", | |
372 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
373 | ||
6649bb7a | 374 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 375 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 376 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
377 | } |
378 | ||
814a0e5c RR |
379 | /* The entry point is given in the ELF header. */ |
380 | return ehdr->e_entry; | |
8ca47e00 RR |
381 | } |
382 | ||
dde79789 | 383 | /*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're |
5bbf89fc RR |
384 | * supposed to jump into it and it will unpack itself. We used to have to |
385 | * perform some hairy magic because the unpacking code scared me. | |
dde79789 | 386 | * |
5bbf89fc RR |
387 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
388 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
389 | * the funky header so we know where in the file to load, and away we go! */ | |
47436aa4 | 390 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 391 | { |
43d33b21 | 392 | struct boot_params boot; |
5bbf89fc RR |
393 | int r; |
394 | /* Modern bzImages get loaded at 1M. */ | |
395 | void *p = from_guest_phys(0x100000); | |
396 | ||
397 | /* Go back to the start of the file and read the header. It should be | |
71cced6e | 398 | * a Linux boot header (see Documentation/x86/i386/boot.txt) */ |
5bbf89fc | 399 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 400 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 401 | |
43d33b21 RR |
402 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
403 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
404 | errx(1, "This doesn't look like a bzImage to me"); |
405 | ||
43d33b21 RR |
406 | /* Skip over the extra sectors of the header. */ |
407 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
408 | |
409 | /* Now read everything into memory. in nice big chunks. */ | |
410 | while ((r = read(fd, p, 65536)) > 0) | |
411 | p += r; | |
412 | ||
43d33b21 RR |
413 | /* Finally, code32_start tells us where to enter the kernel. */ |
414 | return boot.hdr.code32_start; | |
8ca47e00 RR |
415 | } |
416 | ||
dde79789 | 417 | /*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels |
e1e72965 RR |
418 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
419 | * work, we can load those, too. */ | |
47436aa4 | 420 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
421 | { |
422 | Elf32_Ehdr hdr; | |
423 | ||
dde79789 | 424 | /* Read in the first few bytes. */ |
8ca47e00 RR |
425 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
426 | err(1, "Reading kernel"); | |
427 | ||
dde79789 | 428 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 429 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 430 | return map_elf(fd, &hdr); |
8ca47e00 | 431 | |
a6bd8e13 | 432 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 433 | return load_bzimage(fd); |
8ca47e00 RR |
434 | } |
435 | ||
dde79789 RR |
436 | /* This is a trivial little helper to align pages. Andi Kleen hated it because |
437 | * it calls getpagesize() twice: "it's dumb code." | |
438 | * | |
439 | * Kernel guys get really het up about optimization, even when it's not | |
440 | * necessary. I leave this code as a reaction against that. */ | |
8ca47e00 RR |
441 | static inline unsigned long page_align(unsigned long addr) |
442 | { | |
dde79789 | 443 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
444 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
445 | } | |
446 | ||
dde79789 RR |
447 | /*L:180 An "initial ram disk" is a disk image loaded into memory along with |
448 | * the kernel which the kernel can use to boot from without needing any | |
449 | * drivers. Most distributions now use this as standard: the initrd contains | |
450 | * the code to load the appropriate driver modules for the current machine. | |
451 | * | |
452 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
453 | * kernels. He sent me this (and tells me when I break it). */ | |
8ca47e00 RR |
454 | static unsigned long load_initrd(const char *name, unsigned long mem) |
455 | { | |
456 | int ifd; | |
457 | struct stat st; | |
458 | unsigned long len; | |
8ca47e00 RR |
459 | |
460 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 461 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
462 | if (fstat(ifd, &st) < 0) |
463 | err(1, "fstat() on initrd '%s'", name); | |
464 | ||
6649bb7a RM |
465 | /* We map the initrd at the top of memory, but mmap wants it to be |
466 | * page-aligned, so we round the size up for that. */ | |
8ca47e00 | 467 | len = page_align(st.st_size); |
3c6b5bfa | 468 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
dde79789 RR |
469 | /* Once a file is mapped, you can close the file descriptor. It's a |
470 | * little odd, but quite useful. */ | |
8ca47e00 | 471 | close(ifd); |
6649bb7a | 472 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
473 | |
474 | /* We return the initrd size. */ | |
8ca47e00 RR |
475 | return len; |
476 | } | |
e1e72965 | 477 | /*:*/ |
8ca47e00 | 478 | |
dde79789 RR |
479 | /* Simple routine to roll all the commandline arguments together with spaces |
480 | * between them. */ | |
8ca47e00 RR |
481 | static void concat(char *dst, char *args[]) |
482 | { | |
483 | unsigned int i, len = 0; | |
484 | ||
485 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
486 | if (i) { |
487 | strcat(dst+len, " "); | |
488 | len++; | |
489 | } | |
8ca47e00 | 490 | strcpy(dst+len, args[i]); |
1ef36fa6 | 491 | len += strlen(args[i]); |
8ca47e00 RR |
492 | } |
493 | /* In case it's empty. */ | |
494 | dst[len] = '\0'; | |
495 | } | |
496 | ||
e1e72965 RR |
497 | /*L:185 This is where we actually tell the kernel to initialize the Guest. We |
498 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: | |
58a24566 MZ |
499 | * the base of Guest "physical" memory, the top physical page to allow and the |
500 | * entry point for the Guest. */ | |
56739c80 | 501 | static void tell_kernel(unsigned long start) |
8ca47e00 | 502 | { |
511801dc JS |
503 | unsigned long args[] = { LHREQ_INITIALIZE, |
504 | (unsigned long)guest_base, | |
58a24566 | 505 | guest_limit / getpagesize(), start }; |
3c6b5bfa RR |
506 | verbose("Guest: %p - %p (%#lx)\n", |
507 | guest_base, guest_base + guest_limit, guest_limit); | |
56739c80 RR |
508 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
509 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 510 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 511 | } |
dde79789 | 512 | /*:*/ |
8ca47e00 | 513 | |
e1e72965 | 514 | /* |
dde79789 RR |
515 | * Device Handling. |
516 | * | |
e1e72965 | 517 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 518 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 519 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
520 | * if something funny is going on: |
521 | */ | |
8ca47e00 RR |
522 | static void *_check_pointer(unsigned long addr, unsigned int size, |
523 | unsigned int line) | |
524 | { | |
dde79789 RR |
525 | /* We have to separately check addr and addr+size, because size could |
526 | * be huge and addr + size might wrap around. */ | |
3c6b5bfa | 527 | if (addr >= guest_limit || addr + size >= guest_limit) |
17cbca2b | 528 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
dde79789 RR |
529 | /* We return a pointer for the caller's convenience, now we know it's |
530 | * safe to use. */ | |
3c6b5bfa | 531 | return from_guest_phys(addr); |
8ca47e00 | 532 | } |
dde79789 | 533 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
534 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
535 | ||
e1e72965 RR |
536 | /* Each buffer in the virtqueues is actually a chain of descriptors. This |
537 | * function returns the next descriptor in the chain, or vq->vring.num if we're | |
538 | * at the end. */ | |
d1f0132e MM |
539 | static unsigned next_desc(struct vring_desc *desc, |
540 | unsigned int i, unsigned int max) | |
17cbca2b RR |
541 | { |
542 | unsigned int next; | |
543 | ||
544 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
545 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
546 | return max; | |
17cbca2b RR |
547 | |
548 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 549 | next = desc[i].next; |
17cbca2b RR |
550 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
551 | wmb(); | |
552 | ||
d1f0132e | 553 | if (next >= max) |
17cbca2b RR |
554 | errx(1, "Desc next is %u", next); |
555 | ||
556 | return next; | |
557 | } | |
558 | ||
38bc2b8c RR |
559 | /* This actually sends the interrupt for this virtqueue */ |
560 | static void trigger_irq(struct virtqueue *vq) | |
561 | { | |
562 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; | |
563 | ||
95c517c0 RR |
564 | /* Don't inform them if nothing used. */ |
565 | if (!vq->pending_used) | |
566 | return; | |
567 | vq->pending_used = 0; | |
568 | ||
38bc2b8c RR |
569 | /* If they don't want an interrupt, don't send one, unless empty. */ |
570 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) | |
571 | && lg_last_avail(vq) != vq->vring.avail->idx) | |
572 | return; | |
573 | ||
574 | /* Send the Guest an interrupt tell them we used something up. */ | |
575 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
576 | err(1, "Triggering irq %i", vq->config.irq); | |
577 | } | |
578 | ||
17cbca2b RR |
579 | /* This looks in the virtqueue and for the first available buffer, and converts |
580 | * it to an iovec for convenient access. Since descriptors consist of some | |
581 | * number of output then some number of input descriptors, it's actually two | |
582 | * iovecs, but we pack them into one and note how many of each there were. | |
583 | * | |
659a0e66 RR |
584 | * This function returns the descriptor number found. */ |
585 | static unsigned wait_for_vq_desc(struct virtqueue *vq, | |
586 | struct iovec iov[], | |
587 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 588 | { |
d1f0132e MM |
589 | unsigned int i, head, max; |
590 | struct vring_desc *desc; | |
659a0e66 RR |
591 | u16 last_avail = lg_last_avail(vq); |
592 | ||
593 | while (last_avail == vq->vring.avail->idx) { | |
594 | u64 event; | |
595 | ||
38bc2b8c RR |
596 | /* OK, tell Guest about progress up to now. */ |
597 | trigger_irq(vq); | |
598 | ||
b60da13f RR |
599 | /* OK, now we need to know about added descriptors. */ |
600 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
601 | ||
602 | /* They could have slipped one in as we were doing that: make | |
603 | * sure it's written, then check again. */ | |
604 | mb(); | |
605 | if (last_avail != vq->vring.avail->idx) { | |
606 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
607 | break; | |
608 | } | |
609 | ||
659a0e66 RR |
610 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
611 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
612 | errx(1, "Event read failed?"); | |
b60da13f RR |
613 | |
614 | /* We don't need to be notified again. */ | |
615 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 616 | } |
17cbca2b RR |
617 | |
618 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 619 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
17cbca2b | 620 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 621 | last_avail, vq->vring.avail->idx); |
17cbca2b | 622 | |
17cbca2b RR |
623 | /* Grab the next descriptor number they're advertising, and increment |
624 | * the index we've seen. */ | |
b5111790 RR |
625 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
626 | lg_last_avail(vq)++; | |
17cbca2b RR |
627 | |
628 | /* If their number is silly, that's a fatal mistake. */ | |
629 | if (head >= vq->vring.num) | |
630 | errx(1, "Guest says index %u is available", head); | |
631 | ||
632 | /* When we start there are none of either input nor output. */ | |
633 | *out_num = *in_num = 0; | |
634 | ||
d1f0132e MM |
635 | max = vq->vring.num; |
636 | desc = vq->vring.desc; | |
17cbca2b | 637 | i = head; |
d1f0132e MM |
638 | |
639 | /* If this is an indirect entry, then this buffer contains a descriptor | |
640 | * table which we handle as if it's any normal descriptor chain. */ | |
641 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { | |
642 | if (desc[i].len % sizeof(struct vring_desc)) | |
643 | errx(1, "Invalid size for indirect buffer table"); | |
644 | ||
645 | max = desc[i].len / sizeof(struct vring_desc); | |
646 | desc = check_pointer(desc[i].addr, desc[i].len); | |
647 | i = 0; | |
648 | } | |
649 | ||
17cbca2b RR |
650 | do { |
651 | /* Grab the first descriptor, and check it's OK. */ | |
d1f0132e | 652 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 653 | iov[*out_num + *in_num].iov_base |
d1f0132e | 654 | = check_pointer(desc[i].addr, desc[i].len); |
17cbca2b | 655 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 656 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
657 | (*in_num)++; |
658 | else { | |
659 | /* If it's an output descriptor, they're all supposed | |
660 | * to come before any input descriptors. */ | |
661 | if (*in_num) | |
662 | errx(1, "Descriptor has out after in"); | |
663 | (*out_num)++; | |
664 | } | |
665 | ||
666 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 667 | if (*out_num + *in_num > max) |
17cbca2b | 668 | errx(1, "Looped descriptor"); |
d1f0132e | 669 | } while ((i = next_desc(desc, i, max)) != max); |
dde79789 | 670 | |
17cbca2b | 671 | return head; |
8ca47e00 RR |
672 | } |
673 | ||
e1e72965 | 674 | /* After we've used one of their buffers, we tell them about it. We'll then |
17cbca2b RR |
675 | * want to send them an interrupt, using trigger_irq(). */ |
676 | static void add_used(struct virtqueue *vq, unsigned int head, int len) | |
8ca47e00 | 677 | { |
17cbca2b RR |
678 | struct vring_used_elem *used; |
679 | ||
e1e72965 RR |
680 | /* The virtqueue contains a ring of used buffers. Get a pointer to the |
681 | * next entry in that used ring. */ | |
17cbca2b RR |
682 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
683 | used->id = head; | |
684 | used->len = len; | |
685 | /* Make sure buffer is written before we update index. */ | |
686 | wmb(); | |
687 | vq->vring.used->idx++; | |
95c517c0 | 688 | vq->pending_used++; |
8ca47e00 RR |
689 | } |
690 | ||
17cbca2b | 691 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 692 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 693 | { |
17cbca2b | 694 | add_used(vq, head, len); |
56739c80 | 695 | trigger_irq(vq); |
8ca47e00 RR |
696 | } |
697 | ||
e1e72965 RR |
698 | /* |
699 | * The Console | |
700 | * | |
659a0e66 | 701 | * We associate some data with the console for our exit hack. */ |
8ca47e00 RR |
702 | struct console_abort |
703 | { | |
dde79789 | 704 | /* How many times have they hit ^C? */ |
8ca47e00 | 705 | int count; |
dde79789 | 706 | /* When did they start? */ |
8ca47e00 RR |
707 | struct timeval start; |
708 | }; | |
709 | ||
dde79789 | 710 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 711 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 712 | { |
8ca47e00 | 713 | int len; |
17cbca2b | 714 | unsigned int head, in_num, out_num; |
659a0e66 RR |
715 | struct console_abort *abort = vq->dev->priv; |
716 | struct iovec iov[vq->vring.num]; | |
56ae43df | 717 | |
659a0e66 RR |
718 | /* Make sure there's a descriptor waiting. */ |
719 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); | |
56ae43df | 720 | if (out_num) |
17cbca2b | 721 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 722 | |
659a0e66 RR |
723 | /* Read it in. */ |
724 | len = readv(STDIN_FILENO, iov, in_num); | |
8ca47e00 | 725 | if (len <= 0) { |
659a0e66 | 726 | /* Ran out of input? */ |
8ca47e00 | 727 | warnx("Failed to get console input, ignoring console."); |
659a0e66 RR |
728 | /* For simplicity, dying threads kill the whole Launcher. So |
729 | * just nap here. */ | |
730 | for (;;) | |
731 | pause(); | |
8ca47e00 RR |
732 | } |
733 | ||
659a0e66 | 734 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 735 | |
dde79789 RR |
736 | /* Three ^C within one second? Exit. |
737 | * | |
659a0e66 RR |
738 | * This is such a hack, but works surprisingly well. Each ^C has to |
739 | * be in a buffer by itself, so they can't be too fast. But we check | |
740 | * that we get three within about a second, so they can't be too | |
741 | * slow. */ | |
742 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { | |
8ca47e00 | 743 | abort->count = 0; |
659a0e66 RR |
744 | return; |
745 | } | |
8ca47e00 | 746 | |
659a0e66 RR |
747 | abort->count++; |
748 | if (abort->count == 1) | |
749 | gettimeofday(&abort->start, NULL); | |
750 | else if (abort->count == 3) { | |
751 | struct timeval now; | |
752 | gettimeofday(&now, NULL); | |
753 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
754 | if (now.tv_sec <= abort->start.tv_sec+1) | |
755 | kill(0, SIGINT); | |
756 | abort->count = 0; | |
757 | } | |
8ca47e00 RR |
758 | } |
759 | ||
659a0e66 RR |
760 | /* This is the routine which handles console output (ie. stdout). */ |
761 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 762 | { |
17cbca2b | 763 | unsigned int head, out, in; |
17cbca2b RR |
764 | struct iovec iov[vq->vring.num]; |
765 | ||
659a0e66 RR |
766 | head = wait_for_vq_desc(vq, iov, &out, &in); |
767 | if (in) | |
768 | errx(1, "Input buffers in console output queue?"); | |
769 | while (!iov_empty(iov, out)) { | |
770 | int len = writev(STDOUT_FILENO, iov, out); | |
771 | if (len <= 0) | |
772 | err(1, "Write to stdout gave %i", len); | |
773 | iov_consume(iov, out, len); | |
17cbca2b | 774 | } |
38bc2b8c | 775 | add_used(vq, head, 0); |
a161883a RR |
776 | } |
777 | ||
e1e72965 RR |
778 | /* |
779 | * The Network | |
780 | * | |
781 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 782 | * and write them to /dev/net/tun. |
a6bd8e13 | 783 | */ |
659a0e66 RR |
784 | struct net_info { |
785 | int tunfd; | |
786 | }; | |
787 | ||
788 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 789 | { |
659a0e66 RR |
790 | struct net_info *net_info = vq->dev->priv; |
791 | unsigned int head, out, in; | |
17cbca2b | 792 | struct iovec iov[vq->vring.num]; |
a161883a | 793 | |
659a0e66 RR |
794 | head = wait_for_vq_desc(vq, iov, &out, &in); |
795 | if (in) | |
796 | errx(1, "Input buffers in net output queue?"); | |
797 | if (writev(net_info->tunfd, iov, out) < 0) | |
798 | errx(1, "Write to tun failed?"); | |
38bc2b8c | 799 | add_used(vq, head, 0); |
8ca47e00 RR |
800 | } |
801 | ||
4a8962e2 RR |
802 | /* Will reading from this file descriptor block? */ |
803 | static bool will_block(int fd) | |
804 | { | |
805 | fd_set fdset; | |
806 | struct timeval zero = { 0, 0 }; | |
807 | FD_ZERO(&fdset); | |
808 | FD_SET(fd, &fdset); | |
809 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
810 | } | |
811 | ||
659a0e66 | 812 | /* This is where we handle packets coming in from the tun device to our |
17cbca2b | 813 | * Guest. */ |
659a0e66 | 814 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 815 | { |
8ca47e00 | 816 | int len; |
659a0e66 RR |
817 | unsigned int head, out, in; |
818 | struct iovec iov[vq->vring.num]; | |
819 | struct net_info *net_info = vq->dev->priv; | |
820 | ||
821 | head = wait_for_vq_desc(vq, iov, &out, &in); | |
822 | if (out) | |
823 | errx(1, "Output buffers in net input queue?"); | |
4a8962e2 RR |
824 | |
825 | /* Deliver interrupt now, since we're about to sleep. */ | |
826 | if (vq->pending_used && will_block(net_info->tunfd)) | |
827 | trigger_irq(vq); | |
828 | ||
659a0e66 | 829 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 830 | if (len <= 0) |
659a0e66 | 831 | err(1, "Failed to read from tun."); |
4a8962e2 | 832 | add_used(vq, head, len); |
659a0e66 | 833 | } |
dde79789 | 834 | |
659a0e66 RR |
835 | /* This is the helper to create threads. */ |
836 | static int do_thread(void *_vq) | |
837 | { | |
838 | struct virtqueue *vq = _vq; | |
17cbca2b | 839 | |
659a0e66 RR |
840 | for (;;) |
841 | vq->service(vq); | |
842 | return 0; | |
843 | } | |
17cbca2b | 844 | |
659a0e66 RR |
845 | /* When a child dies, we kill our entire process group with SIGTERM. This |
846 | * also has the side effect that the shell restores the console for us! */ | |
847 | static void kill_launcher(int signal) | |
848 | { | |
849 | kill(0, SIGTERM); | |
8ca47e00 RR |
850 | } |
851 | ||
659a0e66 | 852 | static void reset_device(struct device *dev) |
56ae43df | 853 | { |
659a0e66 RR |
854 | struct virtqueue *vq; |
855 | ||
856 | verbose("Resetting device %s\n", dev->name); | |
857 | ||
858 | /* Clear any features they've acked. */ | |
859 | memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); | |
860 | ||
861 | /* We're going to be explicitly killing threads, so ignore them. */ | |
862 | signal(SIGCHLD, SIG_IGN); | |
863 | ||
864 | /* Zero out the virtqueues, get rid of their threads */ | |
865 | for (vq = dev->vq; vq; vq = vq->next) { | |
866 | if (vq->thread != (pid_t)-1) { | |
867 | kill(vq->thread, SIGTERM); | |
868 | waitpid(vq->thread, NULL, 0); | |
869 | vq->thread = (pid_t)-1; | |
870 | } | |
871 | memset(vq->vring.desc, 0, | |
872 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); | |
873 | lg_last_avail(vq) = 0; | |
874 | } | |
875 | dev->running = false; | |
876 | ||
877 | /* Now we care if threads die. */ | |
878 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
879 | } |
880 | ||
659a0e66 | 881 | static void create_thread(struct virtqueue *vq) |
5dae785a | 882 | { |
659a0e66 RR |
883 | /* Create stack for thread and run it. Since stack grows |
884 | * upwards, we point the stack pointer to the end of this | |
885 | * region. */ | |
886 | char *stack = malloc(32768); | |
887 | unsigned long args[] = { LHREQ_EVENTFD, | |
888 | vq->config.pfn*getpagesize(), 0 }; | |
889 | ||
890 | /* Create a zero-initialized eventfd. */ | |
891 | vq->eventfd = eventfd(0, 0); | |
892 | if (vq->eventfd < 0) | |
893 | err(1, "Creating eventfd"); | |
894 | args[2] = vq->eventfd; | |
895 | ||
896 | /* Attach an eventfd to this virtqueue: it will go off | |
897 | * when the Guest does an LHCALL_NOTIFY for this vq. */ | |
898 | if (write(lguest_fd, &args, sizeof(args)) != 0) | |
899 | err(1, "Attaching eventfd"); | |
900 | ||
901 | /* CLONE_VM: because it has to access the Guest memory, and | |
902 | * SIGCHLD so we get a signal if it dies. */ | |
903 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); | |
904 | if (vq->thread == (pid_t)-1) | |
905 | err(1, "Creating clone"); | |
906 | /* We close our local copy, now the child has it. */ | |
907 | close(vq->eventfd); | |
5dae785a RR |
908 | } |
909 | ||
659a0e66 | 910 | static void start_device(struct device *dev) |
6e5aa7ef | 911 | { |
659a0e66 | 912 | unsigned int i; |
6e5aa7ef RR |
913 | struct virtqueue *vq; |
914 | ||
659a0e66 RR |
915 | verbose("Device %s OK: offered", dev->name); |
916 | for (i = 0; i < dev->feature_len; i++) | |
917 | verbose(" %02x", get_feature_bits(dev)[i]); | |
918 | verbose(", accepted"); | |
919 | for (i = 0; i < dev->feature_len; i++) | |
920 | verbose(" %02x", get_feature_bits(dev) | |
921 | [dev->feature_len+i]); | |
922 | ||
923 | for (vq = dev->vq; vq; vq = vq->next) { | |
924 | if (vq->service) | |
925 | create_thread(vq); | |
926 | } | |
927 | dev->running = true; | |
928 | } | |
929 | ||
930 | static void cleanup_devices(void) | |
931 | { | |
932 | struct device *dev; | |
933 | ||
934 | for (dev = devices.dev; dev; dev = dev->next) | |
935 | reset_device(dev); | |
6e5aa7ef | 936 | |
659a0e66 RR |
937 | /* If we saved off the original terminal settings, restore them now. */ |
938 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
939 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
940 | } | |
6e5aa7ef | 941 | |
659a0e66 RR |
942 | /* When the Guest tells us they updated the status field, we handle it. */ |
943 | static void update_device_status(struct device *dev) | |
944 | { | |
945 | /* A zero status is a reset, otherwise it's a set of flags. */ | |
946 | if (dev->desc->status == 0) | |
947 | reset_device(dev); | |
948 | else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { | |
a007a751 | 949 | warnx("Device %s configuration FAILED", dev->name); |
659a0e66 RR |
950 | if (dev->running) |
951 | reset_device(dev); | |
a007a751 | 952 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { |
659a0e66 RR |
953 | if (!dev->running) |
954 | start_device(dev); | |
6e5aa7ef RR |
955 | } |
956 | } | |
957 | ||
17cbca2b | 958 | /* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ |
56739c80 | 959 | static void handle_output(unsigned long addr) |
8ca47e00 RR |
960 | { |
961 | struct device *i; | |
17cbca2b | 962 | |
659a0e66 | 963 | /* Check each device. */ |
17cbca2b | 964 | for (i = devices.dev; i; i = i->next) { |
659a0e66 RR |
965 | struct virtqueue *vq; |
966 | ||
a007a751 | 967 | /* Notifications to device descriptors update device status. */ |
6e5aa7ef | 968 | if (from_guest_phys(addr) == i->desc) { |
a007a751 | 969 | update_device_status(i); |
6e5aa7ef RR |
970 | return; |
971 | } | |
972 | ||
659a0e66 | 973 | /* Devices *can* be used before status is set to DRIVER_OK. */ |
17cbca2b | 974 | for (vq = i->vq; vq; vq = vq->next) { |
659a0e66 | 975 | if (addr != vq->config.pfn*getpagesize()) |
6e5aa7ef | 976 | continue; |
659a0e66 RR |
977 | if (i->running) |
978 | errx(1, "Notification on running %s", i->name); | |
979 | start_device(i); | |
6e5aa7ef | 980 | return; |
8ca47e00 RR |
981 | } |
982 | } | |
dde79789 | 983 | |
17cbca2b RR |
984 | /* Early console write is done using notify on a nul-terminated string |
985 | * in Guest memory. */ | |
986 | if (addr >= guest_limit) | |
987 | errx(1, "Bad NOTIFY %#lx", addr); | |
988 | ||
989 | write(STDOUT_FILENO, from_guest_phys(addr), | |
990 | strnlen(from_guest_phys(addr), guest_limit - addr)); | |
8ca47e00 RR |
991 | } |
992 | ||
dde79789 RR |
993 | /*L:190 |
994 | * Device Setup | |
995 | * | |
996 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
997 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
998 | * routines to allocate and manage them. |
999 | */ | |
8ca47e00 | 1000 | |
a586d4f6 RR |
1001 | /* The layout of the device page is a "struct lguest_device_desc" followed by a |
1002 | * number of virtqueue descriptors, then two sets of feature bits, then an | |
1003 | * array of configuration bytes. This routine returns the configuration | |
1004 | * pointer. */ | |
1005 | static u8 *device_config(const struct device *dev) | |
1006 | { | |
1007 | return (void *)(dev->desc + 1) | |
713b15b3 RR |
1008 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
1009 | + dev->feature_len * 2; | |
17cbca2b RR |
1010 | } |
1011 | ||
a586d4f6 RR |
1012 | /* This routine allocates a new "struct lguest_device_desc" from descriptor |
1013 | * table page just above the Guest's normal memory. It returns a pointer to | |
1014 | * that descriptor. */ | |
1015 | static struct lguest_device_desc *new_dev_desc(u16 type) | |
17cbca2b | 1016 | { |
a586d4f6 RR |
1017 | struct lguest_device_desc d = { .type = type }; |
1018 | void *p; | |
17cbca2b | 1019 | |
a586d4f6 RR |
1020 | /* Figure out where the next device config is, based on the last one. */ |
1021 | if (devices.lastdev) | |
1022 | p = device_config(devices.lastdev) | |
1023 | + devices.lastdev->desc->config_len; | |
1024 | else | |
1025 | p = devices.descpage; | |
17cbca2b | 1026 | |
a586d4f6 RR |
1027 | /* We only have one page for all the descriptors. */ |
1028 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) | |
1029 | errx(1, "Too many devices"); | |
17cbca2b | 1030 | |
a586d4f6 RR |
1031 | /* p might not be aligned, so we memcpy in. */ |
1032 | return memcpy(p, &d, sizeof(d)); | |
17cbca2b RR |
1033 | } |
1034 | ||
a586d4f6 RR |
1035 | /* Each device descriptor is followed by the description of its virtqueues. We |
1036 | * specify how many descriptors the virtqueue is to have. */ | |
17cbca2b | 1037 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
659a0e66 | 1038 | void (*service)(struct virtqueue *)) |
17cbca2b RR |
1039 | { |
1040 | unsigned int pages; | |
1041 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
1042 | void *p; | |
1043 | ||
a6bd8e13 | 1044 | /* First we need some memory for this virtqueue. */ |
2966af73 | 1045 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
42b36cc0 | 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; | |
659a0e66 RR |
1053 | vq->service = service; |
1054 | vq->thread = (pid_t)-1; | |
d1c856e0 | 1055 | |
17cbca2b RR |
1056 | /* Initialize the configuration. */ |
1057 | vq->config.num = num_descs; | |
1058 | vq->config.irq = devices.next_irq++; | |
1059 | vq->config.pfn = to_guest_phys(p) / getpagesize(); | |
1060 | ||
1061 | /* Initialize the vring. */ | |
2966af73 | 1062 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
17cbca2b | 1063 | |
a586d4f6 RR |
1064 | /* Append virtqueue to this device's descriptor. We use |
1065 | * device_config() to get the end of the device's current virtqueues; | |
1066 | * we check that we haven't added any config or feature information | |
1067 | * yet, otherwise we'd be overwriting them. */ | |
1068 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); | |
1069 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); | |
713b15b3 | 1070 | dev->num_vq++; |
a586d4f6 RR |
1071 | dev->desc->num_vq++; |
1072 | ||
1073 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); | |
17cbca2b RR |
1074 | |
1075 | /* Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
1076 | * second. */ | |
1077 | for (i = &dev->vq; *i; i = &(*i)->next); | |
1078 | *i = vq; | |
8ca47e00 RR |
1079 | } |
1080 | ||
6e5aa7ef | 1081 | /* The first half of the feature bitmask is for us to advertise features. The |
a6bd8e13 | 1082 | * second half is for the Guest to accept features. */ |
a586d4f6 RR |
1083 | static void add_feature(struct device *dev, unsigned bit) |
1084 | { | |
6e5aa7ef | 1085 | u8 *features = get_feature_bits(dev); |
a586d4f6 RR |
1086 | |
1087 | /* We can't extend the feature bits once we've added config bytes */ | |
1088 | if (dev->desc->feature_len <= bit / CHAR_BIT) { | |
1089 | assert(dev->desc->config_len == 0); | |
713b15b3 | 1090 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
a586d4f6 RR |
1091 | } |
1092 | ||
a586d4f6 RR |
1093 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
1094 | } | |
1095 | ||
1096 | /* This routine sets the configuration fields for an existing device's | |
1097 | * descriptor. It only works for the last device, but that's OK because that's | |
1098 | * how we use it. */ | |
1099 | static void set_config(struct device *dev, unsigned len, const void *conf) | |
1100 | { | |
1101 | /* Check we haven't overflowed our single page. */ | |
1102 | if (device_config(dev) + len > devices.descpage + getpagesize()) | |
1103 | errx(1, "Too many devices"); | |
1104 | ||
1105 | /* Copy in the config information, and store the length. */ | |
1106 | memcpy(device_config(dev), conf, len); | |
1107 | dev->desc->config_len = len; | |
1108 | } | |
1109 | ||
17cbca2b | 1110 | /* This routine does all the creation and setup of a new device, including |
a6bd8e13 RR |
1111 | * calling new_dev_desc() to allocate the descriptor and device memory. |
1112 | * | |
1113 | * See what I mean about userspace being boring? */ | |
659a0e66 | 1114 | static struct device *new_device(const char *name, u16 type) |
8ca47e00 RR |
1115 | { |
1116 | struct device *dev = malloc(sizeof(*dev)); | |
1117 | ||
dde79789 | 1118 | /* Now we populate the fields one at a time. */ |
17cbca2b | 1119 | dev->desc = new_dev_desc(type); |
17cbca2b | 1120 | dev->name = name; |
d1c856e0 | 1121 | dev->vq = NULL; |
713b15b3 RR |
1122 | dev->feature_len = 0; |
1123 | dev->num_vq = 0; | |
659a0e66 | 1124 | dev->running = false; |
a586d4f6 RR |
1125 | |
1126 | /* Append to device list. Prepending to a single-linked list is | |
1127 | * easier, but the user expects the devices to be arranged on the bus | |
1128 | * in command-line order. The first network device on the command line | |
1129 | * is eth0, the first block device /dev/vda, etc. */ | |
1130 | if (devices.lastdev) | |
1131 | devices.lastdev->next = dev; | |
1132 | else | |
1133 | devices.dev = dev; | |
1134 | devices.lastdev = dev; | |
1135 | ||
8ca47e00 RR |
1136 | return dev; |
1137 | } | |
1138 | ||
dde79789 RR |
1139 | /* Our first setup routine is the console. It's a fairly simple device, but |
1140 | * UNIX tty handling makes it uglier than it could be. */ | |
17cbca2b | 1141 | static void setup_console(void) |
8ca47e00 RR |
1142 | { |
1143 | struct device *dev; | |
1144 | ||
dde79789 | 1145 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
1146 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
1147 | struct termios term = orig_term; | |
dde79789 RR |
1148 | /* Then we turn off echo, line buffering and ^C etc. We want a |
1149 | * raw input stream to the Guest. */ | |
8ca47e00 RR |
1150 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
1151 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
1152 | } |
1153 | ||
659a0e66 RR |
1154 | dev = new_device("console", VIRTIO_ID_CONSOLE); |
1155 | ||
dde79789 | 1156 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
1157 | dev->priv = malloc(sizeof(struct console_abort)); |
1158 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 1159 | |
56ae43df RR |
1160 | /* The console needs two virtqueues: the input then the output. When |
1161 | * they put something the input queue, we make sure we're listening to | |
1162 | * stdin. When they put something in the output queue, we write it to | |
e1e72965 | 1163 | * stdout. */ |
659a0e66 RR |
1164 | add_virtqueue(dev, VIRTQUEUE_NUM, console_input); |
1165 | add_virtqueue(dev, VIRTQUEUE_NUM, console_output); | |
17cbca2b | 1166 | |
659a0e66 | 1167 | verbose("device %u: console\n", ++devices.device_num); |
8ca47e00 | 1168 | } |
17cbca2b | 1169 | /*:*/ |
8ca47e00 | 1170 | |
17cbca2b RR |
1171 | /*M:010 Inter-guest networking is an interesting area. Simplest is to have a |
1172 | * --sharenet=<name> option which opens or creates a named pipe. This can be | |
1173 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 1174 | * |
17cbca2b RR |
1175 | * More sopisticated is to use one of the tools developed for project like UML |
1176 | * to do networking. | |
dde79789 | 1177 | * |
17cbca2b RR |
1178 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
1179 | * completely generic ("here's my vring, attach to your vring") and would work | |
1180 | * for any traffic. Of course, namespace and permissions issues need to be | |
1181 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
1182 | * multiple inter-guest channels behind one interface, although it would | |
1183 | * require some manner of hotplugging new virtio channels. | |
1184 | * | |
1185 | * Finally, we could implement a virtio network switch in the kernel. :*/ | |
8ca47e00 RR |
1186 | |
1187 | static u32 str2ip(const char *ipaddr) | |
1188 | { | |
dec6a2be | 1189 | unsigned int b[4]; |
8ca47e00 | 1190 | |
dec6a2be MM |
1191 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
1192 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
1193 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
1194 | } | |
1195 | ||
1196 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
1197 | { | |
1198 | unsigned int m[6]; | |
1199 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
1200 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
1201 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
1202 | mac[0] = m[0]; | |
1203 | mac[1] = m[1]; | |
1204 | mac[2] = m[2]; | |
1205 | mac[3] = m[3]; | |
1206 | mac[4] = m[4]; | |
1207 | mac[5] = m[5]; | |
8ca47e00 RR |
1208 | } |
1209 | ||
dde79789 RR |
1210 | /* This code is "adapted" from libbridge: it attaches the Host end of the |
1211 | * network device to the bridge device specified by the command line. | |
1212 | * | |
1213 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
1214 | * dislike bridging), and I just try not to break it. */ | |
8ca47e00 RR |
1215 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
1216 | { | |
1217 | int ifidx; | |
1218 | struct ifreq ifr; | |
1219 | ||
1220 | if (!*br_name) | |
1221 | errx(1, "must specify bridge name"); | |
1222 | ||
1223 | ifidx = if_nametoindex(if_name); | |
1224 | if (!ifidx) | |
1225 | errx(1, "interface %s does not exist!", if_name); | |
1226 | ||
1227 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 1228 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
1229 | ifr.ifr_ifindex = ifidx; |
1230 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
1231 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
1232 | } | |
1233 | ||
dde79789 RR |
1234 | /* This sets up the Host end of the network device with an IP address, brings |
1235 | * it up so packets will flow, the copies the MAC address into the hwaddr | |
17cbca2b | 1236 | * pointer. */ |
dec6a2be | 1237 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
1238 | { |
1239 | struct ifreq ifr; | |
1240 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; | |
1241 | ||
1242 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
1243 | strcpy(ifr.ifr_name, tapif); |
1244 | ||
1245 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
8ca47e00 RR |
1246 | sin->sin_family = AF_INET; |
1247 | sin->sin_addr.s_addr = htonl(ipaddr); | |
1248 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) | |
dec6a2be | 1249 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
1250 | ifr.ifr_flags = IFF_UP; |
1251 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
1252 | err(1, "Bringing interface %s up", tapif); |
1253 | } | |
1254 | ||
dec6a2be | 1255 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 1256 | { |
8ca47e00 | 1257 | struct ifreq ifr; |
dec6a2be MM |
1258 | int netfd; |
1259 | ||
1260 | /* Start with this zeroed. Messy but sure. */ | |
1261 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 1262 | |
dde79789 RR |
1263 | /* We open the /dev/net/tun device and tell it we want a tap device. A |
1264 | * tap device is like a tun device, only somehow different. To tell | |
1265 | * the truth, I completely blundered my way through this code, but it | |
1266 | * works now! */ | |
8ca47e00 | 1267 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 1268 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
1269 | strcpy(ifr.ifr_name, "tap%d"); |
1270 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
1271 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 1272 | |
398f187d RR |
1273 | if (ioctl(netfd, TUNSETOFFLOAD, |
1274 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
1275 | err(1, "Could not set features for tun device"); | |
1276 | ||
dde79789 RR |
1277 | /* We don't need checksums calculated for packets coming in this |
1278 | * device: trust us! */ | |
8ca47e00 RR |
1279 | ioctl(netfd, TUNSETNOCSUM, 1); |
1280 | ||
dec6a2be MM |
1281 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
1282 | return netfd; | |
1283 | } | |
1284 | ||
1285 | /*L:195 Our network is a Host<->Guest network. This can either use bridging or | |
1286 | * routing, but the principle is the same: it uses the "tun" device to inject | |
1287 | * packets into the Host as if they came in from a normal network card. We | |
1288 | * just shunt packets between the Guest and the tun device. */ | |
1289 | static void setup_tun_net(char *arg) | |
1290 | { | |
1291 | struct device *dev; | |
659a0e66 RR |
1292 | struct net_info *net_info = malloc(sizeof(*net_info)); |
1293 | int ipfd; | |
dec6a2be MM |
1294 | u32 ip = INADDR_ANY; |
1295 | bool bridging = false; | |
1296 | char tapif[IFNAMSIZ], *p; | |
1297 | struct virtio_net_config conf; | |
1298 | ||
659a0e66 | 1299 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 1300 | |
17cbca2b | 1301 | /* First we create a new network device. */ |
659a0e66 RR |
1302 | dev = new_device("net", VIRTIO_ID_NET); |
1303 | dev->priv = net_info; | |
dde79789 | 1304 | |
56ae43df RR |
1305 | /* Network devices need a receive and a send queue, just like |
1306 | * console. */ | |
659a0e66 RR |
1307 | add_virtqueue(dev, VIRTQUEUE_NUM, net_input); |
1308 | add_virtqueue(dev, VIRTQUEUE_NUM, net_output); | |
8ca47e00 | 1309 | |
dde79789 RR |
1310 | /* We need a socket to perform the magic network ioctls to bring up the |
1311 | * tap interface, connect to the bridge etc. Any socket will do! */ | |
8ca47e00 RR |
1312 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
1313 | if (ipfd < 0) | |
1314 | err(1, "opening IP socket"); | |
1315 | ||
dde79789 | 1316 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 1317 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
1318 | arg += strlen(BRIDGE_PFX); |
1319 | bridging = true; | |
1320 | } | |
1321 | ||
1322 | /* A mac address may follow the bridge name or IP address */ | |
1323 | p = strchr(arg, ':'); | |
1324 | if (p) { | |
1325 | str2mac(p+1, conf.mac); | |
40c42076 | 1326 | add_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 1327 | *p = '\0'; |
dec6a2be MM |
1328 | } |
1329 | ||
1330 | /* arg is now either an IP address or a bridge name */ | |
1331 | if (bridging) | |
1332 | add_to_bridge(ipfd, tapif, arg); | |
1333 | else | |
8ca47e00 RR |
1334 | ip = str2ip(arg); |
1335 | ||
dec6a2be MM |
1336 | /* Set up the tun device. */ |
1337 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 1338 | |
20887611 | 1339 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
398f187d RR |
1340 | /* Expect Guest to handle everything except UFO */ |
1341 | add_feature(dev, VIRTIO_NET_F_CSUM); | |
1342 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
398f187d RR |
1343 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
1344 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
1345 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
1346 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
1347 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
1348 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e MM |
1349 | /* We handle indirect ring entries */ |
1350 | add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); | |
a586d4f6 | 1351 | set_config(dev, sizeof(conf), &conf); |
8ca47e00 | 1352 | |
a586d4f6 | 1353 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
1354 | close(ipfd); |
1355 | ||
dec6a2be MM |
1356 | devices.device_num++; |
1357 | ||
1358 | if (bridging) | |
1359 | verbose("device %u: tun %s attached to bridge: %s\n", | |
1360 | devices.device_num, tapif, arg); | |
1361 | else | |
1362 | verbose("device %u: tun %s: %s\n", | |
1363 | devices.device_num, tapif, arg); | |
8ca47e00 | 1364 | } |
17cbca2b | 1365 | |
e1e72965 RR |
1366 | /* Our block (disk) device should be really simple: the Guest asks for a block |
1367 | * number and we read or write that position in the file. Unfortunately, that | |
1368 | * was amazingly slow: the Guest waits until the read is finished before | |
1369 | * running anything else, even if it could have been doing useful work. | |
17cbca2b | 1370 | * |
e1e72965 RR |
1371 | * We could use async I/O, except it's reputed to suck so hard that characters |
1372 | * actually go missing from your code when you try to use it. | |
17cbca2b RR |
1373 | * |
1374 | * So we farm the I/O out to thread, and communicate with it via a pipe. */ | |
1375 | ||
e1e72965 | 1376 | /* This hangs off device->priv. */ |
17cbca2b RR |
1377 | struct vblk_info |
1378 | { | |
1379 | /* The size of the file. */ | |
1380 | off64_t len; | |
1381 | ||
1382 | /* The file descriptor for the file. */ | |
1383 | int fd; | |
1384 | ||
1385 | /* IO thread listens on this file descriptor [0]. */ | |
1386 | int workpipe[2]; | |
1387 | ||
1388 | /* IO thread writes to this file descriptor to mark it done, then | |
1389 | * Launcher triggers interrupt to Guest. */ | |
1390 | int done_fd; | |
1391 | }; | |
1392 | ||
e1e72965 RR |
1393 | /*L:210 |
1394 | * The Disk | |
1395 | * | |
1396 | * Remember that the block device is handled by a separate I/O thread. We head | |
1397 | * straight into the core of that thread here: | |
1398 | */ | |
659a0e66 | 1399 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 1400 | { |
659a0e66 | 1401 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b RR |
1402 | unsigned int head, out_num, in_num, wlen; |
1403 | int ret; | |
cb38fa23 | 1404 | u8 *in; |
17cbca2b | 1405 | struct virtio_blk_outhdr *out; |
659a0e66 | 1406 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
1407 | off64_t off; |
1408 | ||
659a0e66 RR |
1409 | /* Get the next request. */ |
1410 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); | |
17cbca2b | 1411 | |
e1e72965 RR |
1412 | /* Every block request should contain at least one output buffer |
1413 | * (detailing the location on disk and the type of request) and one | |
1414 | * input buffer (to hold the result). */ | |
17cbca2b RR |
1415 | if (out_num == 0 || in_num == 0) |
1416 | errx(1, "Bad virtblk cmd %u out=%u in=%u", | |
1417 | head, out_num, in_num); | |
1418 | ||
1419 | out = convert(&iov[0], struct virtio_blk_outhdr); | |
cb38fa23 | 1420 | in = convert(&iov[out_num+in_num-1], u8); |
17cbca2b RR |
1421 | off = out->sector * 512; |
1422 | ||
e1e72965 RR |
1423 | /* The block device implements "barriers", where the Guest indicates |
1424 | * that it wants all previous writes to occur before this write. We | |
1425 | * don't have a way of asking our kernel to do a barrier, so we just | |
1426 | * synchronize all the data in the file. Pretty poor, no? */ | |
17cbca2b RR |
1427 | if (out->type & VIRTIO_BLK_T_BARRIER) |
1428 | fdatasync(vblk->fd); | |
1429 | ||
e1e72965 RR |
1430 | /* In general the virtio block driver is allowed to try SCSI commands. |
1431 | * It'd be nice if we supported eject, for example, but we don't. */ | |
17cbca2b RR |
1432 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
1433 | fprintf(stderr, "Scsi commands unsupported\n"); | |
cb38fa23 | 1434 | *in = VIRTIO_BLK_S_UNSUPP; |
1200e646 | 1435 | wlen = sizeof(*in); |
17cbca2b RR |
1436 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
1437 | /* Write */ | |
1438 | ||
1439 | /* Move to the right location in the block file. This can fail | |
1440 | * if they try to write past end. */ | |
1441 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1442 | err(1, "Bad seek to sector %llu", out->sector); | |
1443 | ||
1444 | ret = writev(vblk->fd, iov+1, out_num-1); | |
1445 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); | |
1446 | ||
1447 | /* Grr... Now we know how long the descriptor they sent was, we | |
1448 | * make sure they didn't try to write over the end of the block | |
1449 | * file (possibly extending it). */ | |
1450 | if (ret > 0 && off + ret > vblk->len) { | |
1451 | /* Trim it back to the correct length */ | |
1452 | ftruncate64(vblk->fd, vblk->len); | |
1453 | /* Die, bad Guest, die. */ | |
1454 | errx(1, "Write past end %llu+%u", off, ret); | |
1455 | } | |
1200e646 | 1456 | wlen = sizeof(*in); |
cb38fa23 | 1457 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b RR |
1458 | } else { |
1459 | /* Read */ | |
1460 | ||
1461 | /* Move to the right location in the block file. This can fail | |
1462 | * if they try to read past end. */ | |
1463 | if (lseek64(vblk->fd, off, SEEK_SET) != off) | |
1464 | err(1, "Bad seek to sector %llu", out->sector); | |
1465 | ||
1466 | ret = readv(vblk->fd, iov+1, in_num-1); | |
1467 | verbose("READ from sector %llu: %i\n", out->sector, ret); | |
1468 | if (ret >= 0) { | |
1200e646 | 1469 | wlen = sizeof(*in) + ret; |
cb38fa23 | 1470 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 1471 | } else { |
1200e646 | 1472 | wlen = sizeof(*in); |
cb38fa23 | 1473 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
1474 | } |
1475 | } | |
1476 | ||
d1881d31 RR |
1477 | /* OK, so we noted that it was pretty poor to use an fdatasync as a |
1478 | * barrier. But Christoph Hellwig points out that we need a sync | |
1479 | * *afterwards* as well: "Barriers specify no reordering to the front | |
1480 | * or the back." And Jens Axboe confirmed it, so here we are: */ | |
1481 | if (out->type & VIRTIO_BLK_T_BARRIER) | |
1482 | fdatasync(vblk->fd); | |
1483 | ||
38bc2b8c | 1484 | add_used(vq, head, wlen); |
17cbca2b RR |
1485 | } |
1486 | ||
e1e72965 | 1487 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
1488 | static void setup_block_file(const char *filename) |
1489 | { | |
17cbca2b RR |
1490 | struct device *dev; |
1491 | struct vblk_info *vblk; | |
a586d4f6 | 1492 | struct virtio_blk_config conf; |
17cbca2b | 1493 | |
17cbca2b | 1494 | /* The device responds to return from I/O thread. */ |
659a0e66 | 1495 | dev = new_device("block", VIRTIO_ID_BLOCK); |
17cbca2b | 1496 | |
e1e72965 | 1497 | /* The device has one virtqueue, where the Guest places requests. */ |
659a0e66 | 1498 | add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); |
17cbca2b RR |
1499 | |
1500 | /* Allocate the room for our own bookkeeping */ | |
1501 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
1502 | ||
1503 | /* First we open the file and store the length. */ | |
1504 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
1505 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
1506 | ||
a586d4f6 RR |
1507 | /* We support barriers. */ |
1508 | add_feature(dev, VIRTIO_BLK_F_BARRIER); | |
1509 | ||
17cbca2b | 1510 | /* Tell Guest how many sectors this device has. */ |
a586d4f6 | 1511 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b RR |
1512 | |
1513 | /* Tell Guest not to put in too many descriptors at once: two are used | |
1514 | * for the in and out elements. */ | |
a586d4f6 RR |
1515 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
1516 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); | |
1517 | ||
1518 | set_config(dev, sizeof(conf), &conf); | |
17cbca2b | 1519 | |
17cbca2b | 1520 | verbose("device %u: virtblock %llu sectors\n", |
659a0e66 | 1521 | ++devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 1522 | } |
28fd6d7f | 1523 | |
659a0e66 RR |
1524 | struct rng_info { |
1525 | int rfd; | |
1526 | }; | |
1527 | ||
28fd6d7f RR |
1528 | /* Our random number generator device reads from /dev/random into the Guest's |
1529 | * input buffers. The usual case is that the Guest doesn't want random numbers | |
1530 | * and so has no buffers although /dev/random is still readable, whereas | |
1531 | * console is the reverse. | |
1532 | * | |
1533 | * The same logic applies, however. */ | |
659a0e66 | 1534 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
1535 | { |
1536 | int len; | |
1537 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
1538 | struct rng_info *rng_info = vq->dev->priv; |
1539 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
1540 | |
1541 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 1542 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f RR |
1543 | if (out_num) |
1544 | errx(1, "Output buffers in rng?"); | |
1545 | ||
1546 | /* This is why we convert to iovecs: the readv() call uses them, and so | |
1547 | * it reads straight into the Guest's buffer. We loop to make sure we | |
1548 | * fill it. */ | |
1549 | while (!iov_empty(iov, in_num)) { | |
659a0e66 | 1550 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f RR |
1551 | if (len <= 0) |
1552 | err(1, "Read from /dev/random gave %i", len); | |
1553 | iov_consume(iov, in_num, len); | |
1554 | totlen += len; | |
1555 | } | |
1556 | ||
1557 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 1558 | add_used(vq, head, totlen); |
28fd6d7f RR |
1559 | } |
1560 | ||
1561 | /* And this creates a "hardware" random number device for the Guest. */ | |
1562 | static void setup_rng(void) | |
1563 | { | |
1564 | struct device *dev; | |
659a0e66 | 1565 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 1566 | |
659a0e66 | 1567 | rng_info->rfd = open_or_die("/dev/random", O_RDONLY); |
28fd6d7f RR |
1568 | |
1569 | /* The device responds to return from I/O thread. */ | |
659a0e66 RR |
1570 | dev = new_device("rng", VIRTIO_ID_RNG); |
1571 | dev->priv = rng_info; | |
28fd6d7f RR |
1572 | |
1573 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
659a0e66 | 1574 | add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); |
28fd6d7f RR |
1575 | |
1576 | verbose("device %u: rng\n", devices.device_num++); | |
1577 | } | |
a6bd8e13 | 1578 | /* That's the end of device setup. */ |
ec04b13f | 1579 | |
a6bd8e13 | 1580 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
1581 | static void __attribute__((noreturn)) restart_guest(void) |
1582 | { | |
1583 | unsigned int i; | |
1584 | ||
8c79873d RR |
1585 | /* Since we don't track all open fds, we simply close everything beyond |
1586 | * stderr. */ | |
ec04b13f BR |
1587 | for (i = 3; i < FD_SETSIZE; i++) |
1588 | close(i); | |
8c79873d | 1589 | |
659a0e66 RR |
1590 | /* Reset all the devices (kills all threads). */ |
1591 | cleanup_devices(); | |
1592 | ||
ec04b13f BR |
1593 | execv(main_args[0], main_args); |
1594 | err(1, "Could not exec %s", main_args[0]); | |
1595 | } | |
8ca47e00 | 1596 | |
a6bd8e13 | 1597 | /*L:220 Finally we reach the core of the Launcher which runs the Guest, serves |
dde79789 | 1598 | * its input and output, and finally, lays it to rest. */ |
56739c80 | 1599 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
1600 | { |
1601 | for (;;) { | |
17cbca2b | 1602 | unsigned long notify_addr; |
8ca47e00 RR |
1603 | int readval; |
1604 | ||
1605 | /* We read from the /dev/lguest device to run the Guest. */ | |
e3283fa0 GOC |
1606 | readval = pread(lguest_fd, ¬ify_addr, |
1607 | sizeof(notify_addr), cpu_id); | |
8ca47e00 | 1608 | |
17cbca2b RR |
1609 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
1610 | if (readval == sizeof(notify_addr)) { | |
1611 | verbose("Notify on address %#lx\n", notify_addr); | |
56739c80 | 1612 | handle_output(notify_addr); |
dde79789 | 1613 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
1614 | } else if (errno == ENOENT) { |
1615 | char reason[1024] = { 0 }; | |
e3283fa0 | 1616 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 1617 | errx(1, "%s", reason); |
ec04b13f BR |
1618 | /* ERESTART means that we need to reboot the guest */ |
1619 | } else if (errno == ERESTART) { | |
1620 | restart_guest(); | |
659a0e66 RR |
1621 | /* Anything else means a bug or incompatible change. */ |
1622 | } else | |
8ca47e00 | 1623 | err(1, "Running guest failed"); |
8ca47e00 RR |
1624 | } |
1625 | } | |
a6bd8e13 | 1626 | /*L:240 |
e1e72965 RR |
1627 | * This is the end of the Launcher. The good news: we are over halfway |
1628 | * through! The bad news: the most fiendish part of the code still lies ahead | |
1629 | * of us. | |
dde79789 | 1630 | * |
e1e72965 RR |
1631 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
1632 | * "make Host". | |
1633 | :*/ | |
8ca47e00 RR |
1634 | |
1635 | static struct option opts[] = { | |
1636 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
1637 | { "tunnet", 1, NULL, 't' }, |
1638 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 1639 | { "rng", 0, NULL, 'r' }, |
8ca47e00 RR |
1640 | { "initrd", 1, NULL, 'i' }, |
1641 | { NULL }, | |
1642 | }; | |
1643 | static void usage(void) | |
1644 | { | |
1645 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 1646 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
1647 | "|--block=<filename>|--initrd=<filename>]...\n" |
1648 | "<mem-in-mb> vmlinux [args...]"); | |
1649 | } | |
1650 | ||
3c6b5bfa | 1651 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
1652 | int main(int argc, char *argv[]) |
1653 | { | |
47436aa4 RR |
1654 | /* Memory, top-level pagetable, code startpoint and size of the |
1655 | * (optional) initrd. */ | |
58a24566 | 1656 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
1657 | /* Two temporaries. */ |
1658 | int i, c; | |
3c6b5bfa | 1659 | /* The boot information for the Guest. */ |
43d33b21 | 1660 | struct boot_params *boot; |
dde79789 | 1661 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
1662 | const char *initrd_name = NULL; |
1663 | ||
ec04b13f BR |
1664 | /* Save the args: we "reboot" by execing ourselves again. */ |
1665 | main_args = argv; | |
ec04b13f | 1666 | |
659a0e66 RR |
1667 | /* First we initialize the device list. We keep a pointer to the last |
1668 | * device, and the next interrupt number to use for devices (1: | |
1669 | * remember that 0 is used by the timer). */ | |
a586d4f6 | 1670 | devices.lastdev = NULL; |
17cbca2b | 1671 | devices.next_irq = 1; |
8ca47e00 | 1672 | |
e3283fa0 | 1673 | cpu_id = 0; |
dde79789 RR |
1674 | /* We need to know how much memory so we can set up the device |
1675 | * descriptor and memory pages for the devices as we parse the command | |
1676 | * line. So we quickly look through the arguments to find the amount | |
1677 | * of memory now. */ | |
6570c459 RR |
1678 | for (i = 1; i < argc; i++) { |
1679 | if (argv[i][0] != '-') { | |
3c6b5bfa RR |
1680 | mem = atoi(argv[i]) * 1024 * 1024; |
1681 | /* We start by mapping anonymous pages over all of | |
1682 | * guest-physical memory range. This fills it with 0, | |
1683 | * and ensures that the Guest won't be killed when it | |
1684 | * tries to access it. */ | |
1685 | guest_base = map_zeroed_pages(mem / getpagesize() | |
1686 | + DEVICE_PAGES); | |
1687 | guest_limit = mem; | |
1688 | guest_max = mem + DEVICE_PAGES*getpagesize(); | |
17cbca2b | 1689 | devices.descpage = get_pages(1); |
6570c459 RR |
1690 | break; |
1691 | } | |
1692 | } | |
dde79789 RR |
1693 | |
1694 | /* The options are fairly straight-forward */ | |
8ca47e00 RR |
1695 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
1696 | switch (c) { | |
1697 | case 'v': | |
1698 | verbose = true; | |
1699 | break; | |
8ca47e00 | 1700 | case 't': |
17cbca2b | 1701 | setup_tun_net(optarg); |
8ca47e00 RR |
1702 | break; |
1703 | case 'b': | |
17cbca2b | 1704 | setup_block_file(optarg); |
8ca47e00 | 1705 | break; |
28fd6d7f RR |
1706 | case 'r': |
1707 | setup_rng(); | |
1708 | break; | |
8ca47e00 RR |
1709 | case 'i': |
1710 | initrd_name = optarg; | |
1711 | break; | |
1712 | default: | |
1713 | warnx("Unknown argument %s", argv[optind]); | |
1714 | usage(); | |
1715 | } | |
1716 | } | |
dde79789 RR |
1717 | /* After the other arguments we expect memory and kernel image name, |
1718 | * followed by command line arguments for the kernel. */ | |
8ca47e00 RR |
1719 | if (optind + 2 > argc) |
1720 | usage(); | |
1721 | ||
3c6b5bfa RR |
1722 | verbose("Guest base is at %p\n", guest_base); |
1723 | ||
dde79789 | 1724 | /* We always have a console device */ |
17cbca2b | 1725 | setup_console(); |
8ca47e00 | 1726 | |
8ca47e00 | 1727 | /* Now we load the kernel */ |
47436aa4 | 1728 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 1729 | |
3c6b5bfa RR |
1730 | /* Boot information is stashed at physical address 0 */ |
1731 | boot = from_guest_phys(0); | |
1732 | ||
dde79789 | 1733 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
1734 | if (initrd_name) { |
1735 | initrd_size = load_initrd(initrd_name, mem); | |
dde79789 RR |
1736 | /* These are the location in the Linux boot header where the |
1737 | * start and size of the initrd are expected to be found. */ | |
43d33b21 RR |
1738 | boot->hdr.ramdisk_image = mem - initrd_size; |
1739 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 1740 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 1741 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
1742 | } |
1743 | ||
dde79789 RR |
1744 | /* The Linux boot header contains an "E820" memory map: ours is a |
1745 | * simple, single region. */ | |
43d33b21 RR |
1746 | boot->e820_entries = 1; |
1747 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
dde79789 | 1748 | /* The boot header contains a command line pointer: we put the command |
43d33b21 RR |
1749 | * line after the boot header. */ |
1750 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); | |
e1e72965 | 1751 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 1752 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 1753 | |
814a0e5c | 1754 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 1755 | boot->hdr.version = 0x207; |
814a0e5c RR |
1756 | |
1757 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 1758 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 1759 | |
43d33b21 RR |
1760 | /* Tell the entry path not to try to reload segment registers. */ |
1761 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 1762 | |
dde79789 RR |
1763 | /* We tell the kernel to initialize the Guest: this returns the open |
1764 | * /dev/lguest file descriptor. */ | |
56739c80 | 1765 | tell_kernel(start); |
dde79789 | 1766 | |
659a0e66 RR |
1767 | /* Ensure that we terminate if a child dies. */ |
1768 | signal(SIGCHLD, kill_launcher); | |
1769 | ||
1770 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
1771 | atexit(cleanup_devices); | |
8ca47e00 | 1772 | |
dde79789 | 1773 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 1774 | run_guest(); |
8ca47e00 | 1775 | } |
f56a384e RR |
1776 | /*:*/ |
1777 | ||
1778 | /*M:999 | |
1779 | * Mastery is done: you now know everything I do. | |
1780 | * | |
1781 | * But surely you have seen code, features and bugs in your wanderings which | |
1782 | * you now yearn to attack? That is the real game, and I look forward to you | |
1783 | * patching and forking lguest into the Your-Name-Here-visor. | |
1784 | * | |
1785 | * Farewell, and good coding! | |
1786 | * Rusty Russell. | |
1787 | */ |