| 1 | /*P:100 |
| 2 | * This is the Launcher code, a simple program which lays out the "physical" |
| 3 | * memory for the new Guest by mapping the kernel image and the virtual |
| 4 | * devices, then opens /dev/lguest to tell the kernel about the Guest and |
| 5 | * control it. |
| 6 | :*/ |
| 7 | #define _LARGEFILE64_SOURCE |
| 8 | #define _GNU_SOURCE |
| 9 | #include <stdio.h> |
| 10 | #include <string.h> |
| 11 | #include <unistd.h> |
| 12 | #include <err.h> |
| 13 | #include <stdint.h> |
| 14 | #include <stdlib.h> |
| 15 | #include <elf.h> |
| 16 | #include <sys/mman.h> |
| 17 | #include <sys/param.h> |
| 18 | #include <sys/types.h> |
| 19 | #include <sys/stat.h> |
| 20 | #include <sys/wait.h> |
| 21 | #include <sys/eventfd.h> |
| 22 | #include <fcntl.h> |
| 23 | #include <stdbool.h> |
| 24 | #include <errno.h> |
| 25 | #include <ctype.h> |
| 26 | #include <sys/socket.h> |
| 27 | #include <sys/ioctl.h> |
| 28 | #include <sys/time.h> |
| 29 | #include <time.h> |
| 30 | #include <netinet/in.h> |
| 31 | #include <net/if.h> |
| 32 | #include <linux/sockios.h> |
| 33 | #include <linux/if_tun.h> |
| 34 | #include <sys/uio.h> |
| 35 | #include <termios.h> |
| 36 | #include <getopt.h> |
| 37 | #include <zlib.h> |
| 38 | #include <assert.h> |
| 39 | #include <sched.h> |
| 40 | #include <limits.h> |
| 41 | #include <stddef.h> |
| 42 | #include <signal.h> |
| 43 | #include "linux/lguest_launcher.h" |
| 44 | #include "linux/virtio_config.h" |
| 45 | #include "linux/virtio_net.h" |
| 46 | #include "linux/virtio_blk.h" |
| 47 | #include "linux/virtio_console.h" |
| 48 | #include "linux/virtio_rng.h" |
| 49 | #include "linux/virtio_ring.h" |
| 50 | #include "asm/bootparam.h" |
| 51 | /*L:110 |
| 52 | * We can ignore the 42 include files we need for this program, but I do want |
| 53 | * to draw attention to the use of kernel-style types. |
| 54 | * |
| 55 | * As Linus said, "C is a Spartan language, and so should your naming be." I |
| 56 | * like these abbreviations, so we define them here. Note that u64 is always |
| 57 | * unsigned long long, which works on all Linux systems: this means that we can |
| 58 | * use %llu in printf for any u64. |
| 59 | */ |
| 60 | typedef unsigned long long u64; |
| 61 | typedef uint32_t u32; |
| 62 | typedef uint16_t u16; |
| 63 | typedef uint8_t u8; |
| 64 | /*:*/ |
| 65 | |
| 66 | #define PAGE_PRESENT 0x7 /* Present, RW, Execute */ |
| 67 | #define BRIDGE_PFX "bridge:" |
| 68 | #ifndef SIOCBRADDIF |
| 69 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ |
| 70 | #endif |
| 71 | /* We can have up to 256 pages for devices. */ |
| 72 | #define DEVICE_PAGES 256 |
| 73 | /* This will occupy 3 pages: it must be a power of 2. */ |
| 74 | #define VIRTQUEUE_NUM 256 |
| 75 | |
| 76 | /*L:120 |
| 77 | * verbose is both a global flag and a macro. The C preprocessor allows |
| 78 | * this, and although I wouldn't recommend it, it works quite nicely here. |
| 79 | */ |
| 80 | static bool verbose; |
| 81 | #define verbose(args...) \ |
| 82 | do { if (verbose) printf(args); } while(0) |
| 83 | /*:*/ |
| 84 | |
| 85 | /* The pointer to the start of guest memory. */ |
| 86 | static void *guest_base; |
| 87 | /* The maximum guest physical address allowed, and maximum possible. */ |
| 88 | static unsigned long guest_limit, guest_max; |
| 89 | /* The /dev/lguest file descriptor. */ |
| 90 | static int lguest_fd; |
| 91 | |
| 92 | /* a per-cpu variable indicating whose vcpu is currently running */ |
| 93 | static unsigned int __thread cpu_id; |
| 94 | |
| 95 | /* This is our list of devices. */ |
| 96 | struct device_list { |
| 97 | /* Counter to assign interrupt numbers. */ |
| 98 | unsigned int next_irq; |
| 99 | |
| 100 | /* Counter to print out convenient device numbers. */ |
| 101 | unsigned int device_num; |
| 102 | |
| 103 | /* The descriptor page for the devices. */ |
| 104 | u8 *descpage; |
| 105 | |
| 106 | /* A single linked list of devices. */ |
| 107 | struct device *dev; |
| 108 | /* And a pointer to the last device for easy append. */ |
| 109 | struct device *lastdev; |
| 110 | }; |
| 111 | |
| 112 | /* The list of Guest devices, based on command line arguments. */ |
| 113 | static struct device_list devices; |
| 114 | |
| 115 | /* The device structure describes a single device. */ |
| 116 | struct device { |
| 117 | /* The linked-list pointer. */ |
| 118 | struct device *next; |
| 119 | |
| 120 | /* The device's descriptor, as mapped into the Guest. */ |
| 121 | struct lguest_device_desc *desc; |
| 122 | |
| 123 | /* We can't trust desc values once Guest has booted: we use these. */ |
| 124 | unsigned int feature_len; |
| 125 | unsigned int num_vq; |
| 126 | |
| 127 | /* The name of this device, for --verbose. */ |
| 128 | const char *name; |
| 129 | |
| 130 | /* Any queues attached to this device */ |
| 131 | struct virtqueue *vq; |
| 132 | |
| 133 | /* Is it operational */ |
| 134 | bool running; |
| 135 | |
| 136 | /* Device-specific data. */ |
| 137 | void *priv; |
| 138 | }; |
| 139 | |
| 140 | /* The virtqueue structure describes a queue attached to a device. */ |
| 141 | struct virtqueue { |
| 142 | struct virtqueue *next; |
| 143 | |
| 144 | /* Which device owns me. */ |
| 145 | struct device *dev; |
| 146 | |
| 147 | /* The configuration for this queue. */ |
| 148 | struct lguest_vqconfig config; |
| 149 | |
| 150 | /* The actual ring of buffers. */ |
| 151 | struct vring vring; |
| 152 | |
| 153 | /* Last available index we saw. */ |
| 154 | u16 last_avail_idx; |
| 155 | |
| 156 | /* How many are used since we sent last irq? */ |
| 157 | unsigned int pending_used; |
| 158 | |
| 159 | /* Eventfd where Guest notifications arrive. */ |
| 160 | int eventfd; |
| 161 | |
| 162 | /* Function for the thread which is servicing this virtqueue. */ |
| 163 | void (*service)(struct virtqueue *vq); |
| 164 | pid_t thread; |
| 165 | }; |
| 166 | |
| 167 | /* Remember the arguments to the program so we can "reboot" */ |
| 168 | static char **main_args; |
| 169 | |
| 170 | /* The original tty settings to restore on exit. */ |
| 171 | static struct termios orig_term; |
| 172 | |
| 173 | /* |
| 174 | * We have to be careful with barriers: our devices are all run in separate |
| 175 | * threads and so we need to make sure that changes visible to the Guest happen |
| 176 | * in precise order. |
| 177 | */ |
| 178 | #define wmb() __asm__ __volatile__("" : : : "memory") |
| 179 | #define mb() __asm__ __volatile__("" : : : "memory") |
| 180 | |
| 181 | /* |
| 182 | * Convert an iovec element to the given type. |
| 183 | * |
| 184 | * This is a fairly ugly trick: we need to know the size of the type and |
| 185 | * alignment requirement to check the pointer is kosher. It's also nice to |
| 186 | * have the name of the type in case we report failure. |
| 187 | * |
| 188 | * Typing those three things all the time is cumbersome and error prone, so we |
| 189 | * have a macro which sets them all up and passes to the real function. |
| 190 | */ |
| 191 | #define convert(iov, type) \ |
| 192 | ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) |
| 193 | |
| 194 | static void *_convert(struct iovec *iov, size_t size, size_t align, |
| 195 | const char *name) |
| 196 | { |
| 197 | if (iov->iov_len != size) |
| 198 | errx(1, "Bad iovec size %zu for %s", iov->iov_len, name); |
| 199 | if ((unsigned long)iov->iov_base % align != 0) |
| 200 | errx(1, "Bad alignment %p for %s", iov->iov_base, name); |
| 201 | return iov->iov_base; |
| 202 | } |
| 203 | |
| 204 | /* Wrapper for the last available index. Makes it easier to change. */ |
| 205 | #define lg_last_avail(vq) ((vq)->last_avail_idx) |
| 206 | |
| 207 | /* |
| 208 | * The virtio configuration space is defined to be little-endian. x86 is |
| 209 | * little-endian too, but it's nice to be explicit so we have these helpers. |
| 210 | */ |
| 211 | #define cpu_to_le16(v16) (v16) |
| 212 | #define cpu_to_le32(v32) (v32) |
| 213 | #define cpu_to_le64(v64) (v64) |
| 214 | #define le16_to_cpu(v16) (v16) |
| 215 | #define le32_to_cpu(v32) (v32) |
| 216 | #define le64_to_cpu(v64) (v64) |
| 217 | |
| 218 | /* Is this iovec empty? */ |
| 219 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) |
| 220 | { |
| 221 | unsigned int i; |
| 222 | |
| 223 | for (i = 0; i < num_iov; i++) |
| 224 | if (iov[i].iov_len) |
| 225 | return false; |
| 226 | return true; |
| 227 | } |
| 228 | |
| 229 | /* Take len bytes from the front of this iovec. */ |
| 230 | static void iov_consume(struct iovec iov[], unsigned num_iov, unsigned len) |
| 231 | { |
| 232 | unsigned int i; |
| 233 | |
| 234 | for (i = 0; i < num_iov; i++) { |
| 235 | unsigned int used; |
| 236 | |
| 237 | used = iov[i].iov_len < len ? iov[i].iov_len : len; |
| 238 | iov[i].iov_base += used; |
| 239 | iov[i].iov_len -= used; |
| 240 | len -= used; |
| 241 | } |
| 242 | assert(len == 0); |
| 243 | } |
| 244 | |
| 245 | /* The device virtqueue descriptors are followed by feature bitmasks. */ |
| 246 | static u8 *get_feature_bits(struct device *dev) |
| 247 | { |
| 248 | return (u8 *)(dev->desc + 1) |
| 249 | + dev->num_vq * sizeof(struct lguest_vqconfig); |
| 250 | } |
| 251 | |
| 252 | /*L:100 |
| 253 | * The Launcher code itself takes us out into userspace, that scary place where |
| 254 | * pointers run wild and free! Unfortunately, like most userspace programs, |
| 255 | * it's quite boring (which is why everyone likes to hack on the kernel!). |
| 256 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get |
| 257 | * you through this section. Or, maybe not. |
| 258 | * |
| 259 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" |
| 260 | * memory and stores it in "guest_base". In other words, Guest physical == |
| 261 | * Launcher virtual with an offset. |
| 262 | * |
| 263 | * This can be tough to get your head around, but usually it just means that we |
| 264 | * use these trivial conversion functions when the Guest gives us it's |
| 265 | * "physical" addresses: |
| 266 | */ |
| 267 | static void *from_guest_phys(unsigned long addr) |
| 268 | { |
| 269 | return guest_base + addr; |
| 270 | } |
| 271 | |
| 272 | static unsigned long to_guest_phys(const void *addr) |
| 273 | { |
| 274 | return (addr - guest_base); |
| 275 | } |
| 276 | |
| 277 | /*L:130 |
| 278 | * Loading the Kernel. |
| 279 | * |
| 280 | * We start with couple of simple helper routines. open_or_die() avoids |
| 281 | * error-checking code cluttering the callers: |
| 282 | */ |
| 283 | static int open_or_die(const char *name, int flags) |
| 284 | { |
| 285 | int fd = open(name, flags); |
| 286 | if (fd < 0) |
| 287 | err(1, "Failed to open %s", name); |
| 288 | return fd; |
| 289 | } |
| 290 | |
| 291 | /* map_zeroed_pages() takes a number of pages. */ |
| 292 | static void *map_zeroed_pages(unsigned int num) |
| 293 | { |
| 294 | int fd = open_or_die("/dev/zero", O_RDONLY); |
| 295 | void *addr; |
| 296 | |
| 297 | /* |
| 298 | * We use a private mapping (ie. if we write to the page, it will be |
| 299 | * copied). |
| 300 | */ |
| 301 | addr = mmap(NULL, getpagesize() * num, |
| 302 | PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); |
| 303 | if (addr == MAP_FAILED) |
| 304 | err(1, "Mmaping %u pages of /dev/zero", num); |
| 305 | |
| 306 | /* |
| 307 | * One neat mmap feature is that you can close the fd, and it |
| 308 | * stays mapped. |
| 309 | */ |
| 310 | close(fd); |
| 311 | |
| 312 | return addr; |
| 313 | } |
| 314 | |
| 315 | /* Get some more pages for a device. */ |
| 316 | static void *get_pages(unsigned int num) |
| 317 | { |
| 318 | void *addr = from_guest_phys(guest_limit); |
| 319 | |
| 320 | guest_limit += num * getpagesize(); |
| 321 | if (guest_limit > guest_max) |
| 322 | errx(1, "Not enough memory for devices"); |
| 323 | return addr; |
| 324 | } |
| 325 | |
| 326 | /* |
| 327 | * This routine is used to load the kernel or initrd. It tries mmap, but if |
| 328 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
| 329 | * it falls back to reading the memory in. |
| 330 | */ |
| 331 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
| 332 | { |
| 333 | ssize_t r; |
| 334 | |
| 335 | /* |
| 336 | * We map writable even though for some segments are marked read-only. |
| 337 | * The kernel really wants to be writable: it patches its own |
| 338 | * instructions. |
| 339 | * |
| 340 | * MAP_PRIVATE means that the page won't be copied until a write is |
| 341 | * done to it. This allows us to share untouched memory between |
| 342 | * Guests. |
| 343 | */ |
| 344 | if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, |
| 345 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) |
| 346 | return; |
| 347 | |
| 348 | /* pread does a seek and a read in one shot: saves a few lines. */ |
| 349 | r = pread(fd, addr, len, offset); |
| 350 | if (r != len) |
| 351 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); |
| 352 | } |
| 353 | |
| 354 | /* |
| 355 | * This routine takes an open vmlinux image, which is in ELF, and maps it into |
| 356 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
| 357 | * by all modern binaries on Linux including the kernel. |
| 358 | * |
| 359 | * The ELF headers give *two* addresses: a physical address, and a virtual |
| 360 | * address. We use the physical address; the Guest will map itself to the |
| 361 | * virtual address. |
| 362 | * |
| 363 | * We return the starting address. |
| 364 | */ |
| 365 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
| 366 | { |
| 367 | Elf32_Phdr phdr[ehdr->e_phnum]; |
| 368 | unsigned int i; |
| 369 | |
| 370 | /* |
| 371 | * Sanity checks on the main ELF header: an x86 executable with a |
| 372 | * reasonable number of correctly-sized program headers. |
| 373 | */ |
| 374 | if (ehdr->e_type != ET_EXEC |
| 375 | || ehdr->e_machine != EM_386 |
| 376 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) |
| 377 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) |
| 378 | errx(1, "Malformed elf header"); |
| 379 | |
| 380 | /* |
| 381 | * An ELF executable contains an ELF header and a number of "program" |
| 382 | * headers which indicate which parts ("segments") of the program to |
| 383 | * load where. |
| 384 | */ |
| 385 | |
| 386 | /* We read in all the program headers at once: */ |
| 387 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
| 388 | err(1, "Seeking to program headers"); |
| 389 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) |
| 390 | err(1, "Reading program headers"); |
| 391 | |
| 392 | /* |
| 393 | * Try all the headers: there are usually only three. A read-only one, |
| 394 | * a read-write one, and a "note" section which we don't load. |
| 395 | */ |
| 396 | for (i = 0; i < ehdr->e_phnum; i++) { |
| 397 | /* If this isn't a loadable segment, we ignore it */ |
| 398 | if (phdr[i].p_type != PT_LOAD) |
| 399 | continue; |
| 400 | |
| 401 | verbose("Section %i: size %i addr %p\n", |
| 402 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); |
| 403 | |
| 404 | /* We map this section of the file at its physical address. */ |
| 405 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
| 406 | phdr[i].p_offset, phdr[i].p_filesz); |
| 407 | } |
| 408 | |
| 409 | /* The entry point is given in the ELF header. */ |
| 410 | return ehdr->e_entry; |
| 411 | } |
| 412 | |
| 413 | /*L:150 |
| 414 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed |
| 415 | * to jump into it and it will unpack itself. We used to have to perform some |
| 416 | * hairy magic because the unpacking code scared me. |
| 417 | * |
| 418 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
| 419 | * a small patch to jump over the tricky bits in the Guest, so now we just read |
| 420 | * the funky header so we know where in the file to load, and away we go! |
| 421 | */ |
| 422 | static unsigned long load_bzimage(int fd) |
| 423 | { |
| 424 | struct boot_params boot; |
| 425 | int r; |
| 426 | /* Modern bzImages get loaded at 1M. */ |
| 427 | void *p = from_guest_phys(0x100000); |
| 428 | |
| 429 | /* |
| 430 | * Go back to the start of the file and read the header. It should be |
| 431 | * a Linux boot header (see Documentation/x86/i386/boot.txt) |
| 432 | */ |
| 433 | lseek(fd, 0, SEEK_SET); |
| 434 | read(fd, &boot, sizeof(boot)); |
| 435 | |
| 436 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
| 437 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) |
| 438 | errx(1, "This doesn't look like a bzImage to me"); |
| 439 | |
| 440 | /* Skip over the extra sectors of the header. */ |
| 441 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); |
| 442 | |
| 443 | /* Now read everything into memory. in nice big chunks. */ |
| 444 | while ((r = read(fd, p, 65536)) > 0) |
| 445 | p += r; |
| 446 | |
| 447 | /* Finally, code32_start tells us where to enter the kernel. */ |
| 448 | return boot.hdr.code32_start; |
| 449 | } |
| 450 | |
| 451 | /*L:140 |
| 452 | * Loading the kernel is easy when it's a "vmlinux", but most kernels |
| 453 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
| 454 | * work, we can load those, too. |
| 455 | */ |
| 456 | static unsigned long load_kernel(int fd) |
| 457 | { |
| 458 | Elf32_Ehdr hdr; |
| 459 | |
| 460 | /* Read in the first few bytes. */ |
| 461 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
| 462 | err(1, "Reading kernel"); |
| 463 | |
| 464 | /* If it's an ELF file, it starts with "\177ELF" */ |
| 465 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
| 466 | return map_elf(fd, &hdr); |
| 467 | |
| 468 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
| 469 | return load_bzimage(fd); |
| 470 | } |
| 471 | |
| 472 | /* |
| 473 | * This is a trivial little helper to align pages. Andi Kleen hated it because |
| 474 | * it calls getpagesize() twice: "it's dumb code." |
| 475 | * |
| 476 | * Kernel guys get really het up about optimization, even when it's not |
| 477 | * necessary. I leave this code as a reaction against that. |
| 478 | */ |
| 479 | static inline unsigned long page_align(unsigned long addr) |
| 480 | { |
| 481 | /* Add upwards and truncate downwards. */ |
| 482 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
| 483 | } |
| 484 | |
| 485 | /*L:180 |
| 486 | * An "initial ram disk" is a disk image loaded into memory along with the |
| 487 | * kernel which the kernel can use to boot from without needing any drivers. |
| 488 | * Most distributions now use this as standard: the initrd contains the code to |
| 489 | * load the appropriate driver modules for the current machine. |
| 490 | * |
| 491 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its |
| 492 | * kernels. He sent me this (and tells me when I break it). |
| 493 | */ |
| 494 | static unsigned long load_initrd(const char *name, unsigned long mem) |
| 495 | { |
| 496 | int ifd; |
| 497 | struct stat st; |
| 498 | unsigned long len; |
| 499 | |
| 500 | ifd = open_or_die(name, O_RDONLY); |
| 501 | /* fstat() is needed to get the file size. */ |
| 502 | if (fstat(ifd, &st) < 0) |
| 503 | err(1, "fstat() on initrd '%s'", name); |
| 504 | |
| 505 | /* |
| 506 | * We map the initrd at the top of memory, but mmap wants it to be |
| 507 | * page-aligned, so we round the size up for that. |
| 508 | */ |
| 509 | len = page_align(st.st_size); |
| 510 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
| 511 | /* |
| 512 | * Once a file is mapped, you can close the file descriptor. It's a |
| 513 | * little odd, but quite useful. |
| 514 | */ |
| 515 | close(ifd); |
| 516 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
| 517 | |
| 518 | /* We return the initrd size. */ |
| 519 | return len; |
| 520 | } |
| 521 | /*:*/ |
| 522 | |
| 523 | /* |
| 524 | * Simple routine to roll all the commandline arguments together with spaces |
| 525 | * between them. |
| 526 | */ |
| 527 | static void concat(char *dst, char *args[]) |
| 528 | { |
| 529 | unsigned int i, len = 0; |
| 530 | |
| 531 | for (i = 0; args[i]; i++) { |
| 532 | if (i) { |
| 533 | strcat(dst+len, " "); |
| 534 | len++; |
| 535 | } |
| 536 | strcpy(dst+len, args[i]); |
| 537 | len += strlen(args[i]); |
| 538 | } |
| 539 | /* In case it's empty. */ |
| 540 | dst[len] = '\0'; |
| 541 | } |
| 542 | |
| 543 | /*L:185 |
| 544 | * This is where we actually tell the kernel to initialize the Guest. We |
| 545 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
| 546 | * the base of Guest "physical" memory, the top physical page to allow and the |
| 547 | * entry point for the Guest. |
| 548 | */ |
| 549 | static void tell_kernel(unsigned long start) |
| 550 | { |
| 551 | unsigned long args[] = { LHREQ_INITIALIZE, |
| 552 | (unsigned long)guest_base, |
| 553 | guest_limit / getpagesize(), start }; |
| 554 | verbose("Guest: %p - %p (%#lx)\n", |
| 555 | guest_base, guest_base + guest_limit, guest_limit); |
| 556 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
| 557 | if (write(lguest_fd, args, sizeof(args)) < 0) |
| 558 | err(1, "Writing to /dev/lguest"); |
| 559 | } |
| 560 | /*:*/ |
| 561 | |
| 562 | /*L:200 |
| 563 | * Device Handling. |
| 564 | * |
| 565 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
| 566 | * We need to make sure it's not trying to reach into the Launcher itself, so |
| 567 | * we have a convenient routine which checks it and exits with an error message |
| 568 | * if something funny is going on: |
| 569 | */ |
| 570 | static void *_check_pointer(unsigned long addr, unsigned int size, |
| 571 | unsigned int line) |
| 572 | { |
| 573 | /* |
| 574 | * We have to separately check addr and addr+size, because size could |
| 575 | * be huge and addr + size might wrap around. |
| 576 | */ |
| 577 | if (addr >= guest_limit || addr + size >= guest_limit) |
| 578 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
| 579 | /* |
| 580 | * We return a pointer for the caller's convenience, now we know it's |
| 581 | * safe to use. |
| 582 | */ |
| 583 | return from_guest_phys(addr); |
| 584 | } |
| 585 | /* A macro which transparently hands the line number to the real function. */ |
| 586 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
| 587 | |
| 588 | /* |
| 589 | * Each buffer in the virtqueues is actually a chain of descriptors. This |
| 590 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
| 591 | * at the end. |
| 592 | */ |
| 593 | static unsigned next_desc(struct vring_desc *desc, |
| 594 | unsigned int i, unsigned int max) |
| 595 | { |
| 596 | unsigned int next; |
| 597 | |
| 598 | /* If this descriptor says it doesn't chain, we're done. */ |
| 599 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
| 600 | return max; |
| 601 | |
| 602 | /* Check they're not leading us off end of descriptors. */ |
| 603 | next = desc[i].next; |
| 604 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
| 605 | wmb(); |
| 606 | |
| 607 | if (next >= max) |
| 608 | errx(1, "Desc next is %u", next); |
| 609 | |
| 610 | return next; |
| 611 | } |
| 612 | |
| 613 | /* |
| 614 | * This actually sends the interrupt for this virtqueue, if we've used a |
| 615 | * buffer. |
| 616 | */ |
| 617 | static void trigger_irq(struct virtqueue *vq) |
| 618 | { |
| 619 | unsigned long buf[] = { LHREQ_IRQ, vq->config.irq }; |
| 620 | |
| 621 | /* Don't inform them if nothing used. */ |
| 622 | if (!vq->pending_used) |
| 623 | return; |
| 624 | vq->pending_used = 0; |
| 625 | |
| 626 | /* If they don't want an interrupt, don't send one, unless empty. */ |
| 627 | if ((vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) |
| 628 | && lg_last_avail(vq) != vq->vring.avail->idx) |
| 629 | return; |
| 630 | |
| 631 | /* Send the Guest an interrupt tell them we used something up. */ |
| 632 | if (write(lguest_fd, buf, sizeof(buf)) != 0) |
| 633 | err(1, "Triggering irq %i", vq->config.irq); |
| 634 | } |
| 635 | |
| 636 | /* |
| 637 | * This looks in the virtqueue for the first available buffer, and converts |
| 638 | * it to an iovec for convenient access. Since descriptors consist of some |
| 639 | * number of output then some number of input descriptors, it's actually two |
| 640 | * iovecs, but we pack them into one and note how many of each there were. |
| 641 | * |
| 642 | * This function waits if necessary, and returns the descriptor number found. |
| 643 | */ |
| 644 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
| 645 | struct iovec iov[], |
| 646 | unsigned int *out_num, unsigned int *in_num) |
| 647 | { |
| 648 | unsigned int i, head, max; |
| 649 | struct vring_desc *desc; |
| 650 | u16 last_avail = lg_last_avail(vq); |
| 651 | |
| 652 | /* There's nothing available? */ |
| 653 | while (last_avail == vq->vring.avail->idx) { |
| 654 | u64 event; |
| 655 | |
| 656 | /* |
| 657 | * Since we're about to sleep, now is a good time to tell the |
| 658 | * Guest about what we've used up to now. |
| 659 | */ |
| 660 | trigger_irq(vq); |
| 661 | |
| 662 | /* OK, now we need to know about added descriptors. */ |
| 663 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; |
| 664 | |
| 665 | /* |
| 666 | * They could have slipped one in as we were doing that: make |
| 667 | * sure it's written, then check again. |
| 668 | */ |
| 669 | mb(); |
| 670 | if (last_avail != vq->vring.avail->idx) { |
| 671 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; |
| 672 | break; |
| 673 | } |
| 674 | |
| 675 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
| 676 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) |
| 677 | errx(1, "Event read failed?"); |
| 678 | |
| 679 | /* We don't need to be notified again. */ |
| 680 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; |
| 681 | } |
| 682 | |
| 683 | /* Check it isn't doing very strange things with descriptor numbers. */ |
| 684 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
| 685 | errx(1, "Guest moved used index from %u to %u", |
| 686 | last_avail, vq->vring.avail->idx); |
| 687 | |
| 688 | /* |
| 689 | * Grab the next descriptor number they're advertising, and increment |
| 690 | * the index we've seen. |
| 691 | */ |
| 692 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
| 693 | lg_last_avail(vq)++; |
| 694 | |
| 695 | /* If their number is silly, that's a fatal mistake. */ |
| 696 | if (head >= vq->vring.num) |
| 697 | errx(1, "Guest says index %u is available", head); |
| 698 | |
| 699 | /* When we start there are none of either input nor output. */ |
| 700 | *out_num = *in_num = 0; |
| 701 | |
| 702 | max = vq->vring.num; |
| 703 | desc = vq->vring.desc; |
| 704 | i = head; |
| 705 | |
| 706 | /* |
| 707 | * If this is an indirect entry, then this buffer contains a descriptor |
| 708 | * table which we handle as if it's any normal descriptor chain. |
| 709 | */ |
| 710 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { |
| 711 | if (desc[i].len % sizeof(struct vring_desc)) |
| 712 | errx(1, "Invalid size for indirect buffer table"); |
| 713 | |
| 714 | max = desc[i].len / sizeof(struct vring_desc); |
| 715 | desc = check_pointer(desc[i].addr, desc[i].len); |
| 716 | i = 0; |
| 717 | } |
| 718 | |
| 719 | do { |
| 720 | /* Grab the first descriptor, and check it's OK. */ |
| 721 | iov[*out_num + *in_num].iov_len = desc[i].len; |
| 722 | iov[*out_num + *in_num].iov_base |
| 723 | = check_pointer(desc[i].addr, desc[i].len); |
| 724 | /* If this is an input descriptor, increment that count. */ |
| 725 | if (desc[i].flags & VRING_DESC_F_WRITE) |
| 726 | (*in_num)++; |
| 727 | else { |
| 728 | /* |
| 729 | * If it's an output descriptor, they're all supposed |
| 730 | * to come before any input descriptors. |
| 731 | */ |
| 732 | if (*in_num) |
| 733 | errx(1, "Descriptor has out after in"); |
| 734 | (*out_num)++; |
| 735 | } |
| 736 | |
| 737 | /* If we've got too many, that implies a descriptor loop. */ |
| 738 | if (*out_num + *in_num > max) |
| 739 | errx(1, "Looped descriptor"); |
| 740 | } while ((i = next_desc(desc, i, max)) != max); |
| 741 | |
| 742 | return head; |
| 743 | } |
| 744 | |
| 745 | /* |
| 746 | * After we've used one of their buffers, we tell the Guest about it. Sometime |
| 747 | * later we'll want to send them an interrupt using trigger_irq(); note that |
| 748 | * wait_for_vq_desc() does that for us if it has to wait. |
| 749 | */ |
| 750 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
| 751 | { |
| 752 | struct vring_used_elem *used; |
| 753 | |
| 754 | /* |
| 755 | * The virtqueue contains a ring of used buffers. Get a pointer to the |
| 756 | * next entry in that used ring. |
| 757 | */ |
| 758 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
| 759 | used->id = head; |
| 760 | used->len = len; |
| 761 | /* Make sure buffer is written before we update index. */ |
| 762 | wmb(); |
| 763 | vq->vring.used->idx++; |
| 764 | vq->pending_used++; |
| 765 | } |
| 766 | |
| 767 | /* And here's the combo meal deal. Supersize me! */ |
| 768 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
| 769 | { |
| 770 | add_used(vq, head, len); |
| 771 | trigger_irq(vq); |
| 772 | } |
| 773 | |
| 774 | /* |
| 775 | * The Console |
| 776 | * |
| 777 | * We associate some data with the console for our exit hack. |
| 778 | */ |
| 779 | struct console_abort { |
| 780 | /* How many times have they hit ^C? */ |
| 781 | int count; |
| 782 | /* When did they start? */ |
| 783 | struct timeval start; |
| 784 | }; |
| 785 | |
| 786 | /* This is the routine which handles console input (ie. stdin). */ |
| 787 | static void console_input(struct virtqueue *vq) |
| 788 | { |
| 789 | int len; |
| 790 | unsigned int head, in_num, out_num; |
| 791 | struct console_abort *abort = vq->dev->priv; |
| 792 | struct iovec iov[vq->vring.num]; |
| 793 | |
| 794 | /* Make sure there's a descriptor available. */ |
| 795 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
| 796 | if (out_num) |
| 797 | errx(1, "Output buffers in console in queue?"); |
| 798 | |
| 799 | /* Read into it. This is where we usually wait. */ |
| 800 | len = readv(STDIN_FILENO, iov, in_num); |
| 801 | if (len <= 0) { |
| 802 | /* Ran out of input? */ |
| 803 | warnx("Failed to get console input, ignoring console."); |
| 804 | /* |
| 805 | * For simplicity, dying threads kill the whole Launcher. So |
| 806 | * just nap here. |
| 807 | */ |
| 808 | for (;;) |
| 809 | pause(); |
| 810 | } |
| 811 | |
| 812 | /* Tell the Guest we used a buffer. */ |
| 813 | add_used_and_trigger(vq, head, len); |
| 814 | |
| 815 | /* |
| 816 | * Three ^C within one second? Exit. |
| 817 | * |
| 818 | * This is such a hack, but works surprisingly well. Each ^C has to |
| 819 | * be in a buffer by itself, so they can't be too fast. But we check |
| 820 | * that we get three within about a second, so they can't be too |
| 821 | * slow. |
| 822 | */ |
| 823 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
| 824 | abort->count = 0; |
| 825 | return; |
| 826 | } |
| 827 | |
| 828 | abort->count++; |
| 829 | if (abort->count == 1) |
| 830 | gettimeofday(&abort->start, NULL); |
| 831 | else if (abort->count == 3) { |
| 832 | struct timeval now; |
| 833 | gettimeofday(&now, NULL); |
| 834 | /* Kill all Launcher processes with SIGINT, like normal ^C */ |
| 835 | if (now.tv_sec <= abort->start.tv_sec+1) |
| 836 | kill(0, SIGINT); |
| 837 | abort->count = 0; |
| 838 | } |
| 839 | } |
| 840 | |
| 841 | /* This is the routine which handles console output (ie. stdout). */ |
| 842 | static void console_output(struct virtqueue *vq) |
| 843 | { |
| 844 | unsigned int head, out, in; |
| 845 | struct iovec iov[vq->vring.num]; |
| 846 | |
| 847 | /* We usually wait in here, for the Guest to give us something. */ |
| 848 | head = wait_for_vq_desc(vq, iov, &out, &in); |
| 849 | if (in) |
| 850 | errx(1, "Input buffers in console output queue?"); |
| 851 | |
| 852 | /* writev can return a partial write, so we loop here. */ |
| 853 | while (!iov_empty(iov, out)) { |
| 854 | int len = writev(STDOUT_FILENO, iov, out); |
| 855 | if (len <= 0) |
| 856 | err(1, "Write to stdout gave %i", len); |
| 857 | iov_consume(iov, out, len); |
| 858 | } |
| 859 | |
| 860 | /* |
| 861 | * We're finished with that buffer: if we're going to sleep, |
| 862 | * wait_for_vq_desc() will prod the Guest with an interrupt. |
| 863 | */ |
| 864 | add_used(vq, head, 0); |
| 865 | } |
| 866 | |
| 867 | /* |
| 868 | * The Network |
| 869 | * |
| 870 | * Handling output for network is also simple: we get all the output buffers |
| 871 | * and write them to /dev/net/tun. |
| 872 | */ |
| 873 | struct net_info { |
| 874 | int tunfd; |
| 875 | }; |
| 876 | |
| 877 | static void net_output(struct virtqueue *vq) |
| 878 | { |
| 879 | struct net_info *net_info = vq->dev->priv; |
| 880 | unsigned int head, out, in; |
| 881 | struct iovec iov[vq->vring.num]; |
| 882 | |
| 883 | /* We usually wait in here for the Guest to give us a packet. */ |
| 884 | head = wait_for_vq_desc(vq, iov, &out, &in); |
| 885 | if (in) |
| 886 | errx(1, "Input buffers in net output queue?"); |
| 887 | /* |
| 888 | * Send the whole thing through to /dev/net/tun. It expects the exact |
| 889 | * same format: what a coincidence! |
| 890 | */ |
| 891 | if (writev(net_info->tunfd, iov, out) < 0) |
| 892 | errx(1, "Write to tun failed?"); |
| 893 | |
| 894 | /* |
| 895 | * Done with that one; wait_for_vq_desc() will send the interrupt if |
| 896 | * all packets are processed. |
| 897 | */ |
| 898 | add_used(vq, head, 0); |
| 899 | } |
| 900 | |
| 901 | /* |
| 902 | * Handling network input is a bit trickier, because I've tried to optimize it. |
| 903 | * |
| 904 | * First we have a helper routine which tells is if from this file descriptor |
| 905 | * (ie. the /dev/net/tun device) will block: |
| 906 | */ |
| 907 | static bool will_block(int fd) |
| 908 | { |
| 909 | fd_set fdset; |
| 910 | struct timeval zero = { 0, 0 }; |
| 911 | FD_ZERO(&fdset); |
| 912 | FD_SET(fd, &fdset); |
| 913 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; |
| 914 | } |
| 915 | |
| 916 | /* |
| 917 | * This handles packets coming in from the tun device to our Guest. Like all |
| 918 | * service routines, it gets called again as soon as it returns, so you don't |
| 919 | * see a while(1) loop here. |
| 920 | */ |
| 921 | static void net_input(struct virtqueue *vq) |
| 922 | { |
| 923 | int len; |
| 924 | unsigned int head, out, in; |
| 925 | struct iovec iov[vq->vring.num]; |
| 926 | struct net_info *net_info = vq->dev->priv; |
| 927 | |
| 928 | /* |
| 929 | * Get a descriptor to write an incoming packet into. This will also |
| 930 | * send an interrupt if they're out of descriptors. |
| 931 | */ |
| 932 | head = wait_for_vq_desc(vq, iov, &out, &in); |
| 933 | if (out) |
| 934 | errx(1, "Output buffers in net input queue?"); |
| 935 | |
| 936 | /* |
| 937 | * If it looks like we'll block reading from the tun device, send them |
| 938 | * an interrupt. |
| 939 | */ |
| 940 | if (vq->pending_used && will_block(net_info->tunfd)) |
| 941 | trigger_irq(vq); |
| 942 | |
| 943 | /* |
| 944 | * Read in the packet. This is where we normally wait (when there's no |
| 945 | * incoming network traffic). |
| 946 | */ |
| 947 | len = readv(net_info->tunfd, iov, in); |
| 948 | if (len <= 0) |
| 949 | err(1, "Failed to read from tun."); |
| 950 | |
| 951 | /* |
| 952 | * Mark that packet buffer as used, but don't interrupt here. We want |
| 953 | * to wait until we've done as much work as we can. |
| 954 | */ |
| 955 | add_used(vq, head, len); |
| 956 | } |
| 957 | /*:*/ |
| 958 | |
| 959 | /* This is the helper to create threads: run the service routine in a loop. */ |
| 960 | static int do_thread(void *_vq) |
| 961 | { |
| 962 | struct virtqueue *vq = _vq; |
| 963 | |
| 964 | for (;;) |
| 965 | vq->service(vq); |
| 966 | return 0; |
| 967 | } |
| 968 | |
| 969 | /* |
| 970 | * When a child dies, we kill our entire process group with SIGTERM. This |
| 971 | * also has the side effect that the shell restores the console for us! |
| 972 | */ |
| 973 | static void kill_launcher(int signal) |
| 974 | { |
| 975 | kill(0, SIGTERM); |
| 976 | } |
| 977 | |
| 978 | static void reset_device(struct device *dev) |
| 979 | { |
| 980 | struct virtqueue *vq; |
| 981 | |
| 982 | verbose("Resetting device %s\n", dev->name); |
| 983 | |
| 984 | /* Clear any features they've acked. */ |
| 985 | memset(get_feature_bits(dev) + dev->feature_len, 0, dev->feature_len); |
| 986 | |
| 987 | /* We're going to be explicitly killing threads, so ignore them. */ |
| 988 | signal(SIGCHLD, SIG_IGN); |
| 989 | |
| 990 | /* Zero out the virtqueues, get rid of their threads */ |
| 991 | for (vq = dev->vq; vq; vq = vq->next) { |
| 992 | if (vq->thread != (pid_t)-1) { |
| 993 | kill(vq->thread, SIGTERM); |
| 994 | waitpid(vq->thread, NULL, 0); |
| 995 | vq->thread = (pid_t)-1; |
| 996 | } |
| 997 | memset(vq->vring.desc, 0, |
| 998 | vring_size(vq->config.num, LGUEST_VRING_ALIGN)); |
| 999 | lg_last_avail(vq) = 0; |
| 1000 | } |
| 1001 | dev->running = false; |
| 1002 | |
| 1003 | /* Now we care if threads die. */ |
| 1004 | signal(SIGCHLD, (void *)kill_launcher); |
| 1005 | } |
| 1006 | |
| 1007 | /*L:216 |
| 1008 | * This actually creates the thread which services the virtqueue for a device. |
| 1009 | */ |
| 1010 | static void create_thread(struct virtqueue *vq) |
| 1011 | { |
| 1012 | /* |
| 1013 | * Create stack for thread. Since the stack grows upwards, we point |
| 1014 | * the stack pointer to the end of this region. |
| 1015 | */ |
| 1016 | char *stack = malloc(32768); |
| 1017 | unsigned long args[] = { LHREQ_EVENTFD, |
| 1018 | vq->config.pfn*getpagesize(), 0 }; |
| 1019 | |
| 1020 | /* Create a zero-initialized eventfd. */ |
| 1021 | vq->eventfd = eventfd(0, 0); |
| 1022 | if (vq->eventfd < 0) |
| 1023 | err(1, "Creating eventfd"); |
| 1024 | args[2] = vq->eventfd; |
| 1025 | |
| 1026 | /* |
| 1027 | * Attach an eventfd to this virtqueue: it will go off when the Guest |
| 1028 | * does an LHCALL_NOTIFY for this vq. |
| 1029 | */ |
| 1030 | if (write(lguest_fd, &args, sizeof(args)) != 0) |
| 1031 | err(1, "Attaching eventfd"); |
| 1032 | |
| 1033 | /* |
| 1034 | * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so |
| 1035 | * we get a signal if it dies. |
| 1036 | */ |
| 1037 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); |
| 1038 | if (vq->thread == (pid_t)-1) |
| 1039 | err(1, "Creating clone"); |
| 1040 | |
| 1041 | /* We close our local copy now the child has it. */ |
| 1042 | close(vq->eventfd); |
| 1043 | } |
| 1044 | |
| 1045 | static void start_device(struct device *dev) |
| 1046 | { |
| 1047 | unsigned int i; |
| 1048 | struct virtqueue *vq; |
| 1049 | |
| 1050 | verbose("Device %s OK: offered", dev->name); |
| 1051 | for (i = 0; i < dev->feature_len; i++) |
| 1052 | verbose(" %02x", get_feature_bits(dev)[i]); |
| 1053 | verbose(", accepted"); |
| 1054 | for (i = 0; i < dev->feature_len; i++) |
| 1055 | verbose(" %02x", get_feature_bits(dev) |
| 1056 | [dev->feature_len+i]); |
| 1057 | |
| 1058 | for (vq = dev->vq; vq; vq = vq->next) { |
| 1059 | if (vq->service) |
| 1060 | create_thread(vq); |
| 1061 | } |
| 1062 | dev->running = true; |
| 1063 | } |
| 1064 | |
| 1065 | static void cleanup_devices(void) |
| 1066 | { |
| 1067 | struct device *dev; |
| 1068 | |
| 1069 | for (dev = devices.dev; dev; dev = dev->next) |
| 1070 | reset_device(dev); |
| 1071 | |
| 1072 | /* If we saved off the original terminal settings, restore them now. */ |
| 1073 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) |
| 1074 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); |
| 1075 | } |
| 1076 | |
| 1077 | /* When the Guest tells us they updated the status field, we handle it. */ |
| 1078 | static void update_device_status(struct device *dev) |
| 1079 | { |
| 1080 | /* A zero status is a reset, otherwise it's a set of flags. */ |
| 1081 | if (dev->desc->status == 0) |
| 1082 | reset_device(dev); |
| 1083 | else if (dev->desc->status & VIRTIO_CONFIG_S_FAILED) { |
| 1084 | warnx("Device %s configuration FAILED", dev->name); |
| 1085 | if (dev->running) |
| 1086 | reset_device(dev); |
| 1087 | } else if (dev->desc->status & VIRTIO_CONFIG_S_DRIVER_OK) { |
| 1088 | if (!dev->running) |
| 1089 | start_device(dev); |
| 1090 | } |
| 1091 | } |
| 1092 | |
| 1093 | /*L:215 |
| 1094 | * This is the generic routine we call when the Guest uses LHCALL_NOTIFY. In |
| 1095 | * particular, it's used to notify us of device status changes during boot. |
| 1096 | */ |
| 1097 | static void handle_output(unsigned long addr) |
| 1098 | { |
| 1099 | struct device *i; |
| 1100 | |
| 1101 | /* Check each device. */ |
| 1102 | for (i = devices.dev; i; i = i->next) { |
| 1103 | struct virtqueue *vq; |
| 1104 | |
| 1105 | /* |
| 1106 | * Notifications to device descriptors mean they updated the |
| 1107 | * device status. |
| 1108 | */ |
| 1109 | if (from_guest_phys(addr) == i->desc) { |
| 1110 | update_device_status(i); |
| 1111 | return; |
| 1112 | } |
| 1113 | |
| 1114 | /* |
| 1115 | * Devices *can* be used before status is set to DRIVER_OK. |
| 1116 | * The original plan was that they would never do this: they |
| 1117 | * would always finish setting up their status bits before |
| 1118 | * actually touching the virtqueues. In practice, we allowed |
| 1119 | * them to, and they do (eg. the disk probes for partition |
| 1120 | * tables as part of initialization). |
| 1121 | * |
| 1122 | * If we see this, we start the device: once it's running, we |
| 1123 | * expect the device to catch all the notifications. |
| 1124 | */ |
| 1125 | for (vq = i->vq; vq; vq = vq->next) { |
| 1126 | if (addr != vq->config.pfn*getpagesize()) |
| 1127 | continue; |
| 1128 | if (i->running) |
| 1129 | errx(1, "Notification on running %s", i->name); |
| 1130 | /* This just calls create_thread() for each virtqueue */ |
| 1131 | start_device(i); |
| 1132 | return; |
| 1133 | } |
| 1134 | } |
| 1135 | |
| 1136 | /* |
| 1137 | * Early console write is done using notify on a nul-terminated string |
| 1138 | * in Guest memory. It's also great for hacking debugging messages |
| 1139 | * into a Guest. |
| 1140 | */ |
| 1141 | if (addr >= guest_limit) |
| 1142 | errx(1, "Bad NOTIFY %#lx", addr); |
| 1143 | |
| 1144 | write(STDOUT_FILENO, from_guest_phys(addr), |
| 1145 | strnlen(from_guest_phys(addr), guest_limit - addr)); |
| 1146 | } |
| 1147 | |
| 1148 | /*L:190 |
| 1149 | * Device Setup |
| 1150 | * |
| 1151 | * All devices need a descriptor so the Guest knows it exists, and a "struct |
| 1152 | * device" so the Launcher can keep track of it. We have common helper |
| 1153 | * routines to allocate and manage them. |
| 1154 | */ |
| 1155 | |
| 1156 | /* |
| 1157 | * The layout of the device page is a "struct lguest_device_desc" followed by a |
| 1158 | * number of virtqueue descriptors, then two sets of feature bits, then an |
| 1159 | * array of configuration bytes. This routine returns the configuration |
| 1160 | * pointer. |
| 1161 | */ |
| 1162 | static u8 *device_config(const struct device *dev) |
| 1163 | { |
| 1164 | return (void *)(dev->desc + 1) |
| 1165 | + dev->num_vq * sizeof(struct lguest_vqconfig) |
| 1166 | + dev->feature_len * 2; |
| 1167 | } |
| 1168 | |
| 1169 | /* |
| 1170 | * This routine allocates a new "struct lguest_device_desc" from descriptor |
| 1171 | * table page just above the Guest's normal memory. It returns a pointer to |
| 1172 | * that descriptor. |
| 1173 | */ |
| 1174 | static struct lguest_device_desc *new_dev_desc(u16 type) |
| 1175 | { |
| 1176 | struct lguest_device_desc d = { .type = type }; |
| 1177 | void *p; |
| 1178 | |
| 1179 | /* Figure out where the next device config is, based on the last one. */ |
| 1180 | if (devices.lastdev) |
| 1181 | p = device_config(devices.lastdev) |
| 1182 | + devices.lastdev->desc->config_len; |
| 1183 | else |
| 1184 | p = devices.descpage; |
| 1185 | |
| 1186 | /* We only have one page for all the descriptors. */ |
| 1187 | if (p + sizeof(d) > (void *)devices.descpage + getpagesize()) |
| 1188 | errx(1, "Too many devices"); |
| 1189 | |
| 1190 | /* p might not be aligned, so we memcpy in. */ |
| 1191 | return memcpy(p, &d, sizeof(d)); |
| 1192 | } |
| 1193 | |
| 1194 | /* |
| 1195 | * Each device descriptor is followed by the description of its virtqueues. We |
| 1196 | * specify how many descriptors the virtqueue is to have. |
| 1197 | */ |
| 1198 | static void add_virtqueue(struct device *dev, unsigned int num_descs, |
| 1199 | void (*service)(struct virtqueue *)) |
| 1200 | { |
| 1201 | unsigned int pages; |
| 1202 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); |
| 1203 | void *p; |
| 1204 | |
| 1205 | /* First we need some memory for this virtqueue. */ |
| 1206 | pages = (vring_size(num_descs, LGUEST_VRING_ALIGN) + getpagesize() - 1) |
| 1207 | / getpagesize(); |
| 1208 | p = get_pages(pages); |
| 1209 | |
| 1210 | /* Initialize the virtqueue */ |
| 1211 | vq->next = NULL; |
| 1212 | vq->last_avail_idx = 0; |
| 1213 | vq->dev = dev; |
| 1214 | |
| 1215 | /* |
| 1216 | * This is the routine the service thread will run, and its Process ID |
| 1217 | * once it's running. |
| 1218 | */ |
| 1219 | vq->service = service; |
| 1220 | vq->thread = (pid_t)-1; |
| 1221 | |
| 1222 | /* Initialize the configuration. */ |
| 1223 | vq->config.num = num_descs; |
| 1224 | vq->config.irq = devices.next_irq++; |
| 1225 | vq->config.pfn = to_guest_phys(p) / getpagesize(); |
| 1226 | |
| 1227 | /* Initialize the vring. */ |
| 1228 | vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); |
| 1229 | |
| 1230 | /* |
| 1231 | * Append virtqueue to this device's descriptor. We use |
| 1232 | * device_config() to get the end of the device's current virtqueues; |
| 1233 | * we check that we haven't added any config or feature information |
| 1234 | * yet, otherwise we'd be overwriting them. |
| 1235 | */ |
| 1236 | assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); |
| 1237 | memcpy(device_config(dev), &vq->config, sizeof(vq->config)); |
| 1238 | dev->num_vq++; |
| 1239 | dev->desc->num_vq++; |
| 1240 | |
| 1241 | verbose("Virtqueue page %#lx\n", to_guest_phys(p)); |
| 1242 | |
| 1243 | /* |
| 1244 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is |
| 1245 | * second. |
| 1246 | */ |
| 1247 | for (i = &dev->vq; *i; i = &(*i)->next); |
| 1248 | *i = vq; |
| 1249 | } |
| 1250 | |
| 1251 | /* |
| 1252 | * The first half of the feature bitmask is for us to advertise features. The |
| 1253 | * second half is for the Guest to accept features. |
| 1254 | */ |
| 1255 | static void add_feature(struct device *dev, unsigned bit) |
| 1256 | { |
| 1257 | u8 *features = get_feature_bits(dev); |
| 1258 | |
| 1259 | /* We can't extend the feature bits once we've added config bytes */ |
| 1260 | if (dev->desc->feature_len <= bit / CHAR_BIT) { |
| 1261 | assert(dev->desc->config_len == 0); |
| 1262 | dev->feature_len = dev->desc->feature_len = (bit/CHAR_BIT) + 1; |
| 1263 | } |
| 1264 | |
| 1265 | features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); |
| 1266 | } |
| 1267 | |
| 1268 | /* |
| 1269 | * This routine sets the configuration fields for an existing device's |
| 1270 | * descriptor. It only works for the last device, but that's OK because that's |
| 1271 | * how we use it. |
| 1272 | */ |
| 1273 | static void set_config(struct device *dev, unsigned len, const void *conf) |
| 1274 | { |
| 1275 | /* Check we haven't overflowed our single page. */ |
| 1276 | if (device_config(dev) + len > devices.descpage + getpagesize()) |
| 1277 | errx(1, "Too many devices"); |
| 1278 | |
| 1279 | /* Copy in the config information, and store the length. */ |
| 1280 | memcpy(device_config(dev), conf, len); |
| 1281 | dev->desc->config_len = len; |
| 1282 | |
| 1283 | /* Size must fit in config_len field (8 bits)! */ |
| 1284 | assert(dev->desc->config_len == len); |
| 1285 | } |
| 1286 | |
| 1287 | /* |
| 1288 | * This routine does all the creation and setup of a new device, including |
| 1289 | * calling new_dev_desc() to allocate the descriptor and device memory. We |
| 1290 | * don't actually start the service threads until later. |
| 1291 | * |
| 1292 | * See what I mean about userspace being boring? |
| 1293 | */ |
| 1294 | static struct device *new_device(const char *name, u16 type) |
| 1295 | { |
| 1296 | struct device *dev = malloc(sizeof(*dev)); |
| 1297 | |
| 1298 | /* Now we populate the fields one at a time. */ |
| 1299 | dev->desc = new_dev_desc(type); |
| 1300 | dev->name = name; |
| 1301 | dev->vq = NULL; |
| 1302 | dev->feature_len = 0; |
| 1303 | dev->num_vq = 0; |
| 1304 | dev->running = false; |
| 1305 | |
| 1306 | /* |
| 1307 | * Append to device list. Prepending to a single-linked list is |
| 1308 | * easier, but the user expects the devices to be arranged on the bus |
| 1309 | * in command-line order. The first network device on the command line |
| 1310 | * is eth0, the first block device /dev/vda, etc. |
| 1311 | */ |
| 1312 | if (devices.lastdev) |
| 1313 | devices.lastdev->next = dev; |
| 1314 | else |
| 1315 | devices.dev = dev; |
| 1316 | devices.lastdev = dev; |
| 1317 | |
| 1318 | return dev; |
| 1319 | } |
| 1320 | |
| 1321 | /* |
| 1322 | * Our first setup routine is the console. It's a fairly simple device, but |
| 1323 | * UNIX tty handling makes it uglier than it could be. |
| 1324 | */ |
| 1325 | static void setup_console(void) |
| 1326 | { |
| 1327 | struct device *dev; |
| 1328 | |
| 1329 | /* If we can save the initial standard input settings... */ |
| 1330 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
| 1331 | struct termios term = orig_term; |
| 1332 | /* |
| 1333 | * Then we turn off echo, line buffering and ^C etc: We want a |
| 1334 | * raw input stream to the Guest. |
| 1335 | */ |
| 1336 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
| 1337 | tcsetattr(STDIN_FILENO, TCSANOW, &term); |
| 1338 | } |
| 1339 | |
| 1340 | dev = new_device("console", VIRTIO_ID_CONSOLE); |
| 1341 | |
| 1342 | /* We store the console state in dev->priv, and initialize it. */ |
| 1343 | dev->priv = malloc(sizeof(struct console_abort)); |
| 1344 | ((struct console_abort *)dev->priv)->count = 0; |
| 1345 | |
| 1346 | /* |
| 1347 | * The console needs two virtqueues: the input then the output. When |
| 1348 | * they put something the input queue, we make sure we're listening to |
| 1349 | * stdin. When they put something in the output queue, we write it to |
| 1350 | * stdout. |
| 1351 | */ |
| 1352 | add_virtqueue(dev, VIRTQUEUE_NUM, console_input); |
| 1353 | add_virtqueue(dev, VIRTQUEUE_NUM, console_output); |
| 1354 | |
| 1355 | verbose("device %u: console\n", ++devices.device_num); |
| 1356 | } |
| 1357 | /*:*/ |
| 1358 | |
| 1359 | /*M:010 |
| 1360 | * Inter-guest networking is an interesting area. Simplest is to have a |
| 1361 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
| 1362 | * used to send packets to another guest in a 1:1 manner. |
| 1363 | * |
| 1364 | * More sopisticated is to use one of the tools developed for project like UML |
| 1365 | * to do networking. |
| 1366 | * |
| 1367 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
| 1368 | * completely generic ("here's my vring, attach to your vring") and would work |
| 1369 | * for any traffic. Of course, namespace and permissions issues need to be |
| 1370 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide |
| 1371 | * multiple inter-guest channels behind one interface, although it would |
| 1372 | * require some manner of hotplugging new virtio channels. |
| 1373 | * |
| 1374 | * Finally, we could implement a virtio network switch in the kernel. |
| 1375 | :*/ |
| 1376 | |
| 1377 | static u32 str2ip(const char *ipaddr) |
| 1378 | { |
| 1379 | unsigned int b[4]; |
| 1380 | |
| 1381 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
| 1382 | errx(1, "Failed to parse IP address '%s'", ipaddr); |
| 1383 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; |
| 1384 | } |
| 1385 | |
| 1386 | static void str2mac(const char *macaddr, unsigned char mac[6]) |
| 1387 | { |
| 1388 | unsigned int m[6]; |
| 1389 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", |
| 1390 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) |
| 1391 | errx(1, "Failed to parse mac address '%s'", macaddr); |
| 1392 | mac[0] = m[0]; |
| 1393 | mac[1] = m[1]; |
| 1394 | mac[2] = m[2]; |
| 1395 | mac[3] = m[3]; |
| 1396 | mac[4] = m[4]; |
| 1397 | mac[5] = m[5]; |
| 1398 | } |
| 1399 | |
| 1400 | /* |
| 1401 | * This code is "adapted" from libbridge: it attaches the Host end of the |
| 1402 | * network device to the bridge device specified by the command line. |
| 1403 | * |
| 1404 | * This is yet another James Morris contribution (I'm an IP-level guy, so I |
| 1405 | * dislike bridging), and I just try not to break it. |
| 1406 | */ |
| 1407 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
| 1408 | { |
| 1409 | int ifidx; |
| 1410 | struct ifreq ifr; |
| 1411 | |
| 1412 | if (!*br_name) |
| 1413 | errx(1, "must specify bridge name"); |
| 1414 | |
| 1415 | ifidx = if_nametoindex(if_name); |
| 1416 | if (!ifidx) |
| 1417 | errx(1, "interface %s does not exist!", if_name); |
| 1418 | |
| 1419 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); |
| 1420 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
| 1421 | ifr.ifr_ifindex = ifidx; |
| 1422 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) |
| 1423 | err(1, "can't add %s to bridge %s", if_name, br_name); |
| 1424 | } |
| 1425 | |
| 1426 | /* |
| 1427 | * This sets up the Host end of the network device with an IP address, brings |
| 1428 | * it up so packets will flow, the copies the MAC address into the hwaddr |
| 1429 | * pointer. |
| 1430 | */ |
| 1431 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
| 1432 | { |
| 1433 | struct ifreq ifr; |
| 1434 | struct sockaddr_in *sin = (struct sockaddr_in *)&ifr.ifr_addr; |
| 1435 | |
| 1436 | memset(&ifr, 0, sizeof(ifr)); |
| 1437 | strcpy(ifr.ifr_name, tapif); |
| 1438 | |
| 1439 | /* Don't read these incantations. Just cut & paste them like I did! */ |
| 1440 | sin->sin_family = AF_INET; |
| 1441 | sin->sin_addr.s_addr = htonl(ipaddr); |
| 1442 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
| 1443 | err(1, "Setting %s interface address", tapif); |
| 1444 | ifr.ifr_flags = IFF_UP; |
| 1445 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) |
| 1446 | err(1, "Bringing interface %s up", tapif); |
| 1447 | } |
| 1448 | |
| 1449 | static int get_tun_device(char tapif[IFNAMSIZ]) |
| 1450 | { |
| 1451 | struct ifreq ifr; |
| 1452 | int netfd; |
| 1453 | |
| 1454 | /* Start with this zeroed. Messy but sure. */ |
| 1455 | memset(&ifr, 0, sizeof(ifr)); |
| 1456 | |
| 1457 | /* |
| 1458 | * We open the /dev/net/tun device and tell it we want a tap device. A |
| 1459 | * tap device is like a tun device, only somehow different. To tell |
| 1460 | * the truth, I completely blundered my way through this code, but it |
| 1461 | * works now! |
| 1462 | */ |
| 1463 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
| 1464 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
| 1465 | strcpy(ifr.ifr_name, "tap%d"); |
| 1466 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) |
| 1467 | err(1, "configuring /dev/net/tun"); |
| 1468 | |
| 1469 | if (ioctl(netfd, TUNSETOFFLOAD, |
| 1470 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) |
| 1471 | err(1, "Could not set features for tun device"); |
| 1472 | |
| 1473 | /* |
| 1474 | * We don't need checksums calculated for packets coming in this |
| 1475 | * device: trust us! |
| 1476 | */ |
| 1477 | ioctl(netfd, TUNSETNOCSUM, 1); |
| 1478 | |
| 1479 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
| 1480 | return netfd; |
| 1481 | } |
| 1482 | |
| 1483 | /*L:195 |
| 1484 | * Our network is a Host<->Guest network. This can either use bridging or |
| 1485 | * routing, but the principle is the same: it uses the "tun" device to inject |
| 1486 | * packets into the Host as if they came in from a normal network card. We |
| 1487 | * just shunt packets between the Guest and the tun device. |
| 1488 | */ |
| 1489 | static void setup_tun_net(char *arg) |
| 1490 | { |
| 1491 | struct device *dev; |
| 1492 | struct net_info *net_info = malloc(sizeof(*net_info)); |
| 1493 | int ipfd; |
| 1494 | u32 ip = INADDR_ANY; |
| 1495 | bool bridging = false; |
| 1496 | char tapif[IFNAMSIZ], *p; |
| 1497 | struct virtio_net_config conf; |
| 1498 | |
| 1499 | net_info->tunfd = get_tun_device(tapif); |
| 1500 | |
| 1501 | /* First we create a new network device. */ |
| 1502 | dev = new_device("net", VIRTIO_ID_NET); |
| 1503 | dev->priv = net_info; |
| 1504 | |
| 1505 | /* Network devices need a recv and a send queue, just like console. */ |
| 1506 | add_virtqueue(dev, VIRTQUEUE_NUM, net_input); |
| 1507 | add_virtqueue(dev, VIRTQUEUE_NUM, net_output); |
| 1508 | |
| 1509 | /* |
| 1510 | * We need a socket to perform the magic network ioctls to bring up the |
| 1511 | * tap interface, connect to the bridge etc. Any socket will do! |
| 1512 | */ |
| 1513 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
| 1514 | if (ipfd < 0) |
| 1515 | err(1, "opening IP socket"); |
| 1516 | |
| 1517 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
| 1518 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
| 1519 | arg += strlen(BRIDGE_PFX); |
| 1520 | bridging = true; |
| 1521 | } |
| 1522 | |
| 1523 | /* A mac address may follow the bridge name or IP address */ |
| 1524 | p = strchr(arg, ':'); |
| 1525 | if (p) { |
| 1526 | str2mac(p+1, conf.mac); |
| 1527 | add_feature(dev, VIRTIO_NET_F_MAC); |
| 1528 | *p = '\0'; |
| 1529 | } |
| 1530 | |
| 1531 | /* arg is now either an IP address or a bridge name */ |
| 1532 | if (bridging) |
| 1533 | add_to_bridge(ipfd, tapif, arg); |
| 1534 | else |
| 1535 | ip = str2ip(arg); |
| 1536 | |
| 1537 | /* Set up the tun device. */ |
| 1538 | configure_device(ipfd, tapif, ip); |
| 1539 | |
| 1540 | add_feature(dev, VIRTIO_F_NOTIFY_ON_EMPTY); |
| 1541 | /* Expect Guest to handle everything except UFO */ |
| 1542 | add_feature(dev, VIRTIO_NET_F_CSUM); |
| 1543 | add_feature(dev, VIRTIO_NET_F_GUEST_CSUM); |
| 1544 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO4); |
| 1545 | add_feature(dev, VIRTIO_NET_F_GUEST_TSO6); |
| 1546 | add_feature(dev, VIRTIO_NET_F_GUEST_ECN); |
| 1547 | add_feature(dev, VIRTIO_NET_F_HOST_TSO4); |
| 1548 | add_feature(dev, VIRTIO_NET_F_HOST_TSO6); |
| 1549 | add_feature(dev, VIRTIO_NET_F_HOST_ECN); |
| 1550 | /* We handle indirect ring entries */ |
| 1551 | add_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); |
| 1552 | set_config(dev, sizeof(conf), &conf); |
| 1553 | |
| 1554 | /* We don't need the socket any more; setup is done. */ |
| 1555 | close(ipfd); |
| 1556 | |
| 1557 | devices.device_num++; |
| 1558 | |
| 1559 | if (bridging) |
| 1560 | verbose("device %u: tun %s attached to bridge: %s\n", |
| 1561 | devices.device_num, tapif, arg); |
| 1562 | else |
| 1563 | verbose("device %u: tun %s: %s\n", |
| 1564 | devices.device_num, tapif, arg); |
| 1565 | } |
| 1566 | /*:*/ |
| 1567 | |
| 1568 | /* This hangs off device->priv. */ |
| 1569 | struct vblk_info { |
| 1570 | /* The size of the file. */ |
| 1571 | off64_t len; |
| 1572 | |
| 1573 | /* The file descriptor for the file. */ |
| 1574 | int fd; |
| 1575 | |
| 1576 | }; |
| 1577 | |
| 1578 | /*L:210 |
| 1579 | * The Disk |
| 1580 | * |
| 1581 | * The disk only has one virtqueue, so it only has one thread. It is really |
| 1582 | * simple: the Guest asks for a block number and we read or write that position |
| 1583 | * in the file. |
| 1584 | * |
| 1585 | * Before we serviced each virtqueue in a separate thread, that was unacceptably |
| 1586 | * slow: the Guest waits until the read is finished before running anything |
| 1587 | * else, even if it could have been doing useful work. |
| 1588 | * |
| 1589 | * We could have used async I/O, except it's reputed to suck so hard that |
| 1590 | * characters actually go missing from your code when you try to use it. |
| 1591 | */ |
| 1592 | static void blk_request(struct virtqueue *vq) |
| 1593 | { |
| 1594 | struct vblk_info *vblk = vq->dev->priv; |
| 1595 | unsigned int head, out_num, in_num, wlen; |
| 1596 | int ret; |
| 1597 | u8 *in; |
| 1598 | struct virtio_blk_outhdr *out; |
| 1599 | struct iovec iov[vq->vring.num]; |
| 1600 | off64_t off; |
| 1601 | |
| 1602 | /* |
| 1603 | * Get the next request, where we normally wait. It triggers the |
| 1604 | * interrupt to acknowledge previously serviced requests (if any). |
| 1605 | */ |
| 1606 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
| 1607 | |
| 1608 | /* |
| 1609 | * Every block request should contain at least one output buffer |
| 1610 | * (detailing the location on disk and the type of request) and one |
| 1611 | * input buffer (to hold the result). |
| 1612 | */ |
| 1613 | if (out_num == 0 || in_num == 0) |
| 1614 | errx(1, "Bad virtblk cmd %u out=%u in=%u", |
| 1615 | head, out_num, in_num); |
| 1616 | |
| 1617 | out = convert(&iov[0], struct virtio_blk_outhdr); |
| 1618 | in = convert(&iov[out_num+in_num-1], u8); |
| 1619 | /* |
| 1620 | * For historical reasons, block operations are expressed in 512 byte |
| 1621 | * "sectors". |
| 1622 | */ |
| 1623 | off = out->sector * 512; |
| 1624 | |
| 1625 | /* |
| 1626 | * The block device implements "barriers", where the Guest indicates |
| 1627 | * that it wants all previous writes to occur before this write. We |
| 1628 | * don't have a way of asking our kernel to do a barrier, so we just |
| 1629 | * synchronize all the data in the file. Pretty poor, no? |
| 1630 | */ |
| 1631 | if (out->type & VIRTIO_BLK_T_BARRIER) |
| 1632 | fdatasync(vblk->fd); |
| 1633 | |
| 1634 | /* |
| 1635 | * In general the virtio block driver is allowed to try SCSI commands. |
| 1636 | * It'd be nice if we supported eject, for example, but we don't. |
| 1637 | */ |
| 1638 | if (out->type & VIRTIO_BLK_T_SCSI_CMD) { |
| 1639 | fprintf(stderr, "Scsi commands unsupported\n"); |
| 1640 | *in = VIRTIO_BLK_S_UNSUPP; |
| 1641 | wlen = sizeof(*in); |
| 1642 | } else if (out->type & VIRTIO_BLK_T_OUT) { |
| 1643 | /* |
| 1644 | * Write |
| 1645 | * |
| 1646 | * Move to the right location in the block file. This can fail |
| 1647 | * if they try to write past end. |
| 1648 | */ |
| 1649 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
| 1650 | err(1, "Bad seek to sector %llu", out->sector); |
| 1651 | |
| 1652 | ret = writev(vblk->fd, iov+1, out_num-1); |
| 1653 | verbose("WRITE to sector %llu: %i\n", out->sector, ret); |
| 1654 | |
| 1655 | /* |
| 1656 | * Grr... Now we know how long the descriptor they sent was, we |
| 1657 | * make sure they didn't try to write over the end of the block |
| 1658 | * file (possibly extending it). |
| 1659 | */ |
| 1660 | if (ret > 0 && off + ret > vblk->len) { |
| 1661 | /* Trim it back to the correct length */ |
| 1662 | ftruncate64(vblk->fd, vblk->len); |
| 1663 | /* Die, bad Guest, die. */ |
| 1664 | errx(1, "Write past end %llu+%u", off, ret); |
| 1665 | } |
| 1666 | wlen = sizeof(*in); |
| 1667 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
| 1668 | } else { |
| 1669 | /* |
| 1670 | * Read |
| 1671 | * |
| 1672 | * Move to the right location in the block file. This can fail |
| 1673 | * if they try to read past end. |
| 1674 | */ |
| 1675 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
| 1676 | err(1, "Bad seek to sector %llu", out->sector); |
| 1677 | |
| 1678 | ret = readv(vblk->fd, iov+1, in_num-1); |
| 1679 | verbose("READ from sector %llu: %i\n", out->sector, ret); |
| 1680 | if (ret >= 0) { |
| 1681 | wlen = sizeof(*in) + ret; |
| 1682 | *in = VIRTIO_BLK_S_OK; |
| 1683 | } else { |
| 1684 | wlen = sizeof(*in); |
| 1685 | *in = VIRTIO_BLK_S_IOERR; |
| 1686 | } |
| 1687 | } |
| 1688 | |
| 1689 | /* |
| 1690 | * OK, so we noted that it was pretty poor to use an fdatasync as a |
| 1691 | * barrier. But Christoph Hellwig points out that we need a sync |
| 1692 | * *afterwards* as well: "Barriers specify no reordering to the front |
| 1693 | * or the back." And Jens Axboe confirmed it, so here we are: |
| 1694 | */ |
| 1695 | if (out->type & VIRTIO_BLK_T_BARRIER) |
| 1696 | fdatasync(vblk->fd); |
| 1697 | |
| 1698 | /* Finished that request. */ |
| 1699 | add_used(vq, head, wlen); |
| 1700 | } |
| 1701 | |
| 1702 | /*L:198 This actually sets up a virtual block device. */ |
| 1703 | static void setup_block_file(const char *filename) |
| 1704 | { |
| 1705 | struct device *dev; |
| 1706 | struct vblk_info *vblk; |
| 1707 | struct virtio_blk_config conf; |
| 1708 | |
| 1709 | /* Creat the device. */ |
| 1710 | dev = new_device("block", VIRTIO_ID_BLOCK); |
| 1711 | |
| 1712 | /* The device has one virtqueue, where the Guest places requests. */ |
| 1713 | add_virtqueue(dev, VIRTQUEUE_NUM, blk_request); |
| 1714 | |
| 1715 | /* Allocate the room for our own bookkeeping */ |
| 1716 | vblk = dev->priv = malloc(sizeof(*vblk)); |
| 1717 | |
| 1718 | /* First we open the file and store the length. */ |
| 1719 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); |
| 1720 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); |
| 1721 | |
| 1722 | /* We support barriers. */ |
| 1723 | add_feature(dev, VIRTIO_BLK_F_BARRIER); |
| 1724 | |
| 1725 | /* Tell Guest how many sectors this device has. */ |
| 1726 | conf.capacity = cpu_to_le64(vblk->len / 512); |
| 1727 | |
| 1728 | /* |
| 1729 | * Tell Guest not to put in too many descriptors at once: two are used |
| 1730 | * for the in and out elements. |
| 1731 | */ |
| 1732 | add_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
| 1733 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); |
| 1734 | |
| 1735 | /* Don't try to put whole struct: we have 8 bit limit. */ |
| 1736 | set_config(dev, offsetof(struct virtio_blk_config, geometry), &conf); |
| 1737 | |
| 1738 | verbose("device %u: virtblock %llu sectors\n", |
| 1739 | ++devices.device_num, le64_to_cpu(conf.capacity)); |
| 1740 | } |
| 1741 | |
| 1742 | /*L:211 |
| 1743 | * Our random number generator device reads from /dev/random into the Guest's |
| 1744 | * input buffers. The usual case is that the Guest doesn't want random numbers |
| 1745 | * and so has no buffers although /dev/random is still readable, whereas |
| 1746 | * console is the reverse. |
| 1747 | * |
| 1748 | * The same logic applies, however. |
| 1749 | */ |
| 1750 | struct rng_info { |
| 1751 | int rfd; |
| 1752 | }; |
| 1753 | |
| 1754 | static void rng_input(struct virtqueue *vq) |
| 1755 | { |
| 1756 | int len; |
| 1757 | unsigned int head, in_num, out_num, totlen = 0; |
| 1758 | struct rng_info *rng_info = vq->dev->priv; |
| 1759 | struct iovec iov[vq->vring.num]; |
| 1760 | |
| 1761 | /* First we need a buffer from the Guests's virtqueue. */ |
| 1762 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
| 1763 | if (out_num) |
| 1764 | errx(1, "Output buffers in rng?"); |
| 1765 | |
| 1766 | /* |
| 1767 | * Just like the console write, we loop to cover the whole iovec. |
| 1768 | * In this case, short reads actually happen quite a bit. |
| 1769 | */ |
| 1770 | while (!iov_empty(iov, in_num)) { |
| 1771 | len = readv(rng_info->rfd, iov, in_num); |
| 1772 | if (len <= 0) |
| 1773 | err(1, "Read from /dev/random gave %i", len); |
| 1774 | iov_consume(iov, in_num, len); |
| 1775 | totlen += len; |
| 1776 | } |
| 1777 | |
| 1778 | /* Tell the Guest about the new input. */ |
| 1779 | add_used(vq, head, totlen); |
| 1780 | } |
| 1781 | |
| 1782 | /*L:199 |
| 1783 | * This creates a "hardware" random number device for the Guest. |
| 1784 | */ |
| 1785 | static void setup_rng(void) |
| 1786 | { |
| 1787 | struct device *dev; |
| 1788 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
| 1789 | |
| 1790 | /* Our device's privat info simply contains the /dev/random fd. */ |
| 1791 | rng_info->rfd = open_or_die("/dev/random", O_RDONLY); |
| 1792 | |
| 1793 | /* Create the new device. */ |
| 1794 | dev = new_device("rng", VIRTIO_ID_RNG); |
| 1795 | dev->priv = rng_info; |
| 1796 | |
| 1797 | /* The device has one virtqueue, where the Guest places inbufs. */ |
| 1798 | add_virtqueue(dev, VIRTQUEUE_NUM, rng_input); |
| 1799 | |
| 1800 | verbose("device %u: rng\n", devices.device_num++); |
| 1801 | } |
| 1802 | /* That's the end of device setup. */ |
| 1803 | |
| 1804 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
| 1805 | static void __attribute__((noreturn)) restart_guest(void) |
| 1806 | { |
| 1807 | unsigned int i; |
| 1808 | |
| 1809 | /* |
| 1810 | * Since we don't track all open fds, we simply close everything beyond |
| 1811 | * stderr. |
| 1812 | */ |
| 1813 | for (i = 3; i < FD_SETSIZE; i++) |
| 1814 | close(i); |
| 1815 | |
| 1816 | /* Reset all the devices (kills all threads). */ |
| 1817 | cleanup_devices(); |
| 1818 | |
| 1819 | execv(main_args[0], main_args); |
| 1820 | err(1, "Could not exec %s", main_args[0]); |
| 1821 | } |
| 1822 | |
| 1823 | /*L:220 |
| 1824 | * Finally we reach the core of the Launcher which runs the Guest, serves |
| 1825 | * its input and output, and finally, lays it to rest. |
| 1826 | */ |
| 1827 | static void __attribute__((noreturn)) run_guest(void) |
| 1828 | { |
| 1829 | for (;;) { |
| 1830 | unsigned long notify_addr; |
| 1831 | int readval; |
| 1832 | |
| 1833 | /* We read from the /dev/lguest device to run the Guest. */ |
| 1834 | readval = pread(lguest_fd, ¬ify_addr, |
| 1835 | sizeof(notify_addr), cpu_id); |
| 1836 | |
| 1837 | /* One unsigned long means the Guest did HCALL_NOTIFY */ |
| 1838 | if (readval == sizeof(notify_addr)) { |
| 1839 | verbose("Notify on address %#lx\n", notify_addr); |
| 1840 | handle_output(notify_addr); |
| 1841 | /* ENOENT means the Guest died. Reading tells us why. */ |
| 1842 | } else if (errno == ENOENT) { |
| 1843 | char reason[1024] = { 0 }; |
| 1844 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
| 1845 | errx(1, "%s", reason); |
| 1846 | /* ERESTART means that we need to reboot the guest */ |
| 1847 | } else if (errno == ERESTART) { |
| 1848 | restart_guest(); |
| 1849 | /* Anything else means a bug or incompatible change. */ |
| 1850 | } else |
| 1851 | err(1, "Running guest failed"); |
| 1852 | } |
| 1853 | } |
| 1854 | /*L:240 |
| 1855 | * This is the end of the Launcher. The good news: we are over halfway |
| 1856 | * through! The bad news: the most fiendish part of the code still lies ahead |
| 1857 | * of us. |
| 1858 | * |
| 1859 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
| 1860 | * "make Host". |
| 1861 | :*/ |
| 1862 | |
| 1863 | static struct option opts[] = { |
| 1864 | { "verbose", 0, NULL, 'v' }, |
| 1865 | { "tunnet", 1, NULL, 't' }, |
| 1866 | { "block", 1, NULL, 'b' }, |
| 1867 | { "rng", 0, NULL, 'r' }, |
| 1868 | { "initrd", 1, NULL, 'i' }, |
| 1869 | { NULL }, |
| 1870 | }; |
| 1871 | static void usage(void) |
| 1872 | { |
| 1873 | errx(1, "Usage: lguest [--verbose] " |
| 1874 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
| 1875 | "|--block=<filename>|--initrd=<filename>]...\n" |
| 1876 | "<mem-in-mb> vmlinux [args...]"); |
| 1877 | } |
| 1878 | |
| 1879 | /*L:105 The main routine is where the real work begins: */ |
| 1880 | int main(int argc, char *argv[]) |
| 1881 | { |
| 1882 | /* Memory, code startpoint and size of the (optional) initrd. */ |
| 1883 | unsigned long mem = 0, start, initrd_size = 0; |
| 1884 | /* Two temporaries. */ |
| 1885 | int i, c; |
| 1886 | /* The boot information for the Guest. */ |
| 1887 | struct boot_params *boot; |
| 1888 | /* If they specify an initrd file to load. */ |
| 1889 | const char *initrd_name = NULL; |
| 1890 | |
| 1891 | /* Save the args: we "reboot" by execing ourselves again. */ |
| 1892 | main_args = argv; |
| 1893 | |
| 1894 | /* |
| 1895 | * First we initialize the device list. We keep a pointer to the last |
| 1896 | * device, and the next interrupt number to use for devices (1: |
| 1897 | * remember that 0 is used by the timer). |
| 1898 | */ |
| 1899 | devices.lastdev = NULL; |
| 1900 | devices.next_irq = 1; |
| 1901 | |
| 1902 | /* We're CPU 0. In fact, that's the only CPU possible right now. */ |
| 1903 | cpu_id = 0; |
| 1904 | |
| 1905 | /* |
| 1906 | * We need to know how much memory so we can set up the device |
| 1907 | * descriptor and memory pages for the devices as we parse the command |
| 1908 | * line. So we quickly look through the arguments to find the amount |
| 1909 | * of memory now. |
| 1910 | */ |
| 1911 | for (i = 1; i < argc; i++) { |
| 1912 | if (argv[i][0] != '-') { |
| 1913 | mem = atoi(argv[i]) * 1024 * 1024; |
| 1914 | /* |
| 1915 | * We start by mapping anonymous pages over all of |
| 1916 | * guest-physical memory range. This fills it with 0, |
| 1917 | * and ensures that the Guest won't be killed when it |
| 1918 | * tries to access it. |
| 1919 | */ |
| 1920 | guest_base = map_zeroed_pages(mem / getpagesize() |
| 1921 | + DEVICE_PAGES); |
| 1922 | guest_limit = mem; |
| 1923 | guest_max = mem + DEVICE_PAGES*getpagesize(); |
| 1924 | devices.descpage = get_pages(1); |
| 1925 | break; |
| 1926 | } |
| 1927 | } |
| 1928 | |
| 1929 | /* The options are fairly straight-forward */ |
| 1930 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
| 1931 | switch (c) { |
| 1932 | case 'v': |
| 1933 | verbose = true; |
| 1934 | break; |
| 1935 | case 't': |
| 1936 | setup_tun_net(optarg); |
| 1937 | break; |
| 1938 | case 'b': |
| 1939 | setup_block_file(optarg); |
| 1940 | break; |
| 1941 | case 'r': |
| 1942 | setup_rng(); |
| 1943 | break; |
| 1944 | case 'i': |
| 1945 | initrd_name = optarg; |
| 1946 | break; |
| 1947 | default: |
| 1948 | warnx("Unknown argument %s", argv[optind]); |
| 1949 | usage(); |
| 1950 | } |
| 1951 | } |
| 1952 | /* |
| 1953 | * After the other arguments we expect memory and kernel image name, |
| 1954 | * followed by command line arguments for the kernel. |
| 1955 | */ |
| 1956 | if (optind + 2 > argc) |
| 1957 | usage(); |
| 1958 | |
| 1959 | verbose("Guest base is at %p\n", guest_base); |
| 1960 | |
| 1961 | /* We always have a console device */ |
| 1962 | setup_console(); |
| 1963 | |
| 1964 | /* Now we load the kernel */ |
| 1965 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
| 1966 | |
| 1967 | /* Boot information is stashed at physical address 0 */ |
| 1968 | boot = from_guest_phys(0); |
| 1969 | |
| 1970 | /* Map the initrd image if requested (at top of physical memory) */ |
| 1971 | if (initrd_name) { |
| 1972 | initrd_size = load_initrd(initrd_name, mem); |
| 1973 | /* |
| 1974 | * These are the location in the Linux boot header where the |
| 1975 | * start and size of the initrd are expected to be found. |
| 1976 | */ |
| 1977 | boot->hdr.ramdisk_image = mem - initrd_size; |
| 1978 | boot->hdr.ramdisk_size = initrd_size; |
| 1979 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
| 1980 | boot->hdr.type_of_loader = 0xFF; |
| 1981 | } |
| 1982 | |
| 1983 | /* |
| 1984 | * The Linux boot header contains an "E820" memory map: ours is a |
| 1985 | * simple, single region. |
| 1986 | */ |
| 1987 | boot->e820_entries = 1; |
| 1988 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); |
| 1989 | /* |
| 1990 | * The boot header contains a command line pointer: we put the command |
| 1991 | * line after the boot header. |
| 1992 | */ |
| 1993 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
| 1994 | /* We use a simple helper to copy the arguments separated by spaces. */ |
| 1995 | concat((char *)(boot + 1), argv+optind+2); |
| 1996 | |
| 1997 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
| 1998 | boot->hdr.version = 0x207; |
| 1999 | |
| 2000 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ |
| 2001 | boot->hdr.hardware_subarch = 1; |
| 2002 | |
| 2003 | /* Tell the entry path not to try to reload segment registers. */ |
| 2004 | boot->hdr.loadflags |= KEEP_SEGMENTS; |
| 2005 | |
| 2006 | /* |
| 2007 | * We tell the kernel to initialize the Guest: this returns the open |
| 2008 | * /dev/lguest file descriptor. |
| 2009 | */ |
| 2010 | tell_kernel(start); |
| 2011 | |
| 2012 | /* Ensure that we terminate if a device-servicing child dies. */ |
| 2013 | signal(SIGCHLD, kill_launcher); |
| 2014 | |
| 2015 | /* If we exit via err(), this kills all the threads, restores tty. */ |
| 2016 | atexit(cleanup_devices); |
| 2017 | |
| 2018 | /* Finally, run the Guest. This doesn't return. */ |
| 2019 | run_guest(); |
| 2020 | } |
| 2021 | /*:*/ |
| 2022 | |
| 2023 | /*M:999 |
| 2024 | * Mastery is done: you now know everything I do. |
| 2025 | * |
| 2026 | * But surely you have seen code, features and bugs in your wanderings which |
| 2027 | * you now yearn to attack? That is the real game, and I look forward to you |
| 2028 | * patching and forking lguest into the Your-Name-Here-visor. |
| 2029 | * |
| 2030 | * Farewell, and good coding! |
| 2031 | * Rusty Russell. |
| 2032 | */ |