| 1 | /*P:200 This contains all the /dev/lguest code, whereby the userspace |
| 2 | * launcher controls and communicates with the Guest. For example, |
| 3 | * the first write will tell us the Guest's memory layout and entry |
| 4 | * point. A read will run the Guest until something happens, such as |
| 5 | * a signal or the Guest accessing a device. |
| 6 | :*/ |
| 7 | #include <linux/uaccess.h> |
| 8 | #include <linux/miscdevice.h> |
| 9 | #include <linux/fs.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/file.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/export.h> |
| 14 | #include "lg.h" |
| 15 | |
| 16 | /*L:052 |
| 17 | The Launcher can get the registers, and also set some of them. |
| 18 | */ |
| 19 | static int getreg_setup(struct lg_cpu *cpu, const unsigned long __user *input) |
| 20 | { |
| 21 | unsigned long which; |
| 22 | |
| 23 | /* We re-use the ptrace structure to specify which register to read. */ |
| 24 | if (get_user(which, input) != 0) |
| 25 | return -EFAULT; |
| 26 | |
| 27 | /* |
| 28 | * We set up the cpu register pointer, and their next read will |
| 29 | * actually get the value (instead of running the guest). |
| 30 | * |
| 31 | * The last argument 'true' says we can access any register. |
| 32 | */ |
| 33 | cpu->reg_read = lguest_arch_regptr(cpu, which, true); |
| 34 | if (!cpu->reg_read) |
| 35 | return -ENOENT; |
| 36 | |
| 37 | /* And because this is a write() call, we return the length used. */ |
| 38 | return sizeof(unsigned long) * 2; |
| 39 | } |
| 40 | |
| 41 | static int setreg(struct lg_cpu *cpu, const unsigned long __user *input) |
| 42 | { |
| 43 | unsigned long which, value, *reg; |
| 44 | |
| 45 | /* We re-use the ptrace structure to specify which register to read. */ |
| 46 | if (get_user(which, input) != 0) |
| 47 | return -EFAULT; |
| 48 | input++; |
| 49 | if (get_user(value, input) != 0) |
| 50 | return -EFAULT; |
| 51 | |
| 52 | /* The last argument 'false' means we can't access all registers. */ |
| 53 | reg = lguest_arch_regptr(cpu, which, false); |
| 54 | if (!reg) |
| 55 | return -ENOENT; |
| 56 | |
| 57 | *reg = value; |
| 58 | |
| 59 | /* And because this is a write() call, we return the length used. */ |
| 60 | return sizeof(unsigned long) * 3; |
| 61 | } |
| 62 | |
| 63 | /*L:050 |
| 64 | * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt |
| 65 | * number to /dev/lguest. |
| 66 | */ |
| 67 | static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) |
| 68 | { |
| 69 | unsigned long irq; |
| 70 | |
| 71 | if (get_user(irq, input) != 0) |
| 72 | return -EFAULT; |
| 73 | if (irq >= LGUEST_IRQS) |
| 74 | return -EINVAL; |
| 75 | |
| 76 | /* |
| 77 | * Next time the Guest runs, the core code will see if it can deliver |
| 78 | * this interrupt. |
| 79 | */ |
| 80 | set_interrupt(cpu, irq); |
| 81 | return 0; |
| 82 | } |
| 83 | |
| 84 | /*L:053 |
| 85 | * Deliver a trap: this is used by the Launcher if it can't emulate |
| 86 | * an instruction. |
| 87 | */ |
| 88 | static int trap(struct lg_cpu *cpu, const unsigned long __user *input) |
| 89 | { |
| 90 | unsigned long trapnum; |
| 91 | |
| 92 | if (get_user(trapnum, input) != 0) |
| 93 | return -EFAULT; |
| 94 | |
| 95 | if (!deliver_trap(cpu, trapnum)) |
| 96 | return -EINVAL; |
| 97 | |
| 98 | return 0; |
| 99 | } |
| 100 | |
| 101 | /*L:040 |
| 102 | * Once our Guest is initialized, the Launcher makes it run by reading |
| 103 | * from /dev/lguest. |
| 104 | */ |
| 105 | static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) |
| 106 | { |
| 107 | struct lguest *lg = file->private_data; |
| 108 | struct lg_cpu *cpu; |
| 109 | unsigned int cpu_id = *o; |
| 110 | |
| 111 | /* You must write LHREQ_INITIALIZE first! */ |
| 112 | if (!lg) |
| 113 | return -EINVAL; |
| 114 | |
| 115 | /* Watch out for arbitrary vcpu indexes! */ |
| 116 | if (cpu_id >= lg->nr_cpus) |
| 117 | return -EINVAL; |
| 118 | |
| 119 | cpu = &lg->cpus[cpu_id]; |
| 120 | |
| 121 | /* If you're not the task which owns the Guest, go away. */ |
| 122 | if (current != cpu->tsk) |
| 123 | return -EPERM; |
| 124 | |
| 125 | /* If the Guest is already dead, we indicate why */ |
| 126 | if (lg->dead) { |
| 127 | size_t len; |
| 128 | |
| 129 | /* lg->dead either contains an error code, or a string. */ |
| 130 | if (IS_ERR(lg->dead)) |
| 131 | return PTR_ERR(lg->dead); |
| 132 | |
| 133 | /* We can only return as much as the buffer they read with. */ |
| 134 | len = min(size, strlen(lg->dead)+1); |
| 135 | if (copy_to_user(user, lg->dead, len) != 0) |
| 136 | return -EFAULT; |
| 137 | return len; |
| 138 | } |
| 139 | |
| 140 | /* |
| 141 | * If we returned from read() last time because the Guest sent I/O, |
| 142 | * clear the flag. |
| 143 | */ |
| 144 | if (cpu->pending.trap) |
| 145 | cpu->pending.trap = 0; |
| 146 | |
| 147 | /* Run the Guest until something interesting happens. */ |
| 148 | return run_guest(cpu, (unsigned long __user *)user); |
| 149 | } |
| 150 | |
| 151 | /*L:025 |
| 152 | * This actually initializes a CPU. For the moment, a Guest is only |
| 153 | * uniprocessor, so "id" is always 0. |
| 154 | */ |
| 155 | static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) |
| 156 | { |
| 157 | /* We have a limited number of CPUs in the lguest struct. */ |
| 158 | if (id >= ARRAY_SIZE(cpu->lg->cpus)) |
| 159 | return -EINVAL; |
| 160 | |
| 161 | /* Set up this CPU's id, and pointer back to the lguest struct. */ |
| 162 | cpu->id = id; |
| 163 | cpu->lg = container_of(cpu, struct lguest, cpus[id]); |
| 164 | cpu->lg->nr_cpus++; |
| 165 | |
| 166 | /* Each CPU has a timer it can set. */ |
| 167 | init_clockdev(cpu); |
| 168 | |
| 169 | /* |
| 170 | * We need a complete page for the Guest registers: they are accessible |
| 171 | * to the Guest and we can only grant it access to whole pages. |
| 172 | */ |
| 173 | cpu->regs_page = get_zeroed_page(GFP_KERNEL); |
| 174 | if (!cpu->regs_page) |
| 175 | return -ENOMEM; |
| 176 | |
| 177 | /* We actually put the registers at the end of the page. */ |
| 178 | cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); |
| 179 | |
| 180 | /* |
| 181 | * Now we initialize the Guest's registers, handing it the start |
| 182 | * address. |
| 183 | */ |
| 184 | lguest_arch_setup_regs(cpu, start_ip); |
| 185 | |
| 186 | /* |
| 187 | * We keep a pointer to the Launcher task (ie. current task) for when |
| 188 | * other Guests want to wake this one (eg. console input). |
| 189 | */ |
| 190 | cpu->tsk = current; |
| 191 | |
| 192 | /* |
| 193 | * We need to keep a pointer to the Launcher's memory map, because if |
| 194 | * the Launcher dies we need to clean it up. If we don't keep a |
| 195 | * reference, it is destroyed before close() is called. |
| 196 | */ |
| 197 | cpu->mm = get_task_mm(cpu->tsk); |
| 198 | |
| 199 | /* |
| 200 | * We remember which CPU's pages this Guest used last, for optimization |
| 201 | * when the same Guest runs on the same CPU twice. |
| 202 | */ |
| 203 | cpu->last_pages = NULL; |
| 204 | |
| 205 | /* No error == success. */ |
| 206 | return 0; |
| 207 | } |
| 208 | |
| 209 | /*L:020 |
| 210 | * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in |
| 211 | * addition to the LHREQ_INITIALIZE value). These are: |
| 212 | * |
| 213 | * base: The start of the Guest-physical memory inside the Launcher memory. |
| 214 | * |
| 215 | * pfnlimit: The highest (Guest-physical) page number the Guest should be |
| 216 | * allowed to access. The Guest memory lives inside the Launcher, so it sets |
| 217 | * this to ensure the Guest can only reach its own memory. |
| 218 | * |
| 219 | * start: The first instruction to execute ("eip" in x86-speak). |
| 220 | */ |
| 221 | static int initialize(struct file *file, const unsigned long __user *input) |
| 222 | { |
| 223 | /* "struct lguest" contains all we (the Host) know about a Guest. */ |
| 224 | struct lguest *lg; |
| 225 | int err; |
| 226 | unsigned long args[4]; |
| 227 | |
| 228 | /* |
| 229 | * We grab the Big Lguest lock, which protects against multiple |
| 230 | * simultaneous initializations. |
| 231 | */ |
| 232 | mutex_lock(&lguest_lock); |
| 233 | /* You can't initialize twice! Close the device and start again... */ |
| 234 | if (file->private_data) { |
| 235 | err = -EBUSY; |
| 236 | goto unlock; |
| 237 | } |
| 238 | |
| 239 | if (copy_from_user(args, input, sizeof(args)) != 0) { |
| 240 | err = -EFAULT; |
| 241 | goto unlock; |
| 242 | } |
| 243 | |
| 244 | lg = kzalloc(sizeof(*lg), GFP_KERNEL); |
| 245 | if (!lg) { |
| 246 | err = -ENOMEM; |
| 247 | goto unlock; |
| 248 | } |
| 249 | |
| 250 | /* Populate the easy fields of our "struct lguest" */ |
| 251 | lg->mem_base = (void __user *)args[0]; |
| 252 | lg->pfn_limit = args[1]; |
| 253 | lg->device_limit = args[3]; |
| 254 | |
| 255 | /* This is the first cpu (cpu 0) and it will start booting at args[2] */ |
| 256 | err = lg_cpu_start(&lg->cpus[0], 0, args[2]); |
| 257 | if (err) |
| 258 | goto free_lg; |
| 259 | |
| 260 | /* |
| 261 | * Initialize the Guest's shadow page tables. This allocates |
| 262 | * memory, so can fail. |
| 263 | */ |
| 264 | err = init_guest_pagetable(lg); |
| 265 | if (err) |
| 266 | goto free_regs; |
| 267 | |
| 268 | /* We keep our "struct lguest" in the file's private_data. */ |
| 269 | file->private_data = lg; |
| 270 | |
| 271 | mutex_unlock(&lguest_lock); |
| 272 | |
| 273 | /* And because this is a write() call, we return the length used. */ |
| 274 | return sizeof(args); |
| 275 | |
| 276 | free_regs: |
| 277 | /* FIXME: This should be in free_vcpu */ |
| 278 | free_page(lg->cpus[0].regs_page); |
| 279 | free_lg: |
| 280 | kfree(lg); |
| 281 | unlock: |
| 282 | mutex_unlock(&lguest_lock); |
| 283 | return err; |
| 284 | } |
| 285 | |
| 286 | /*L:010 |
| 287 | * The first operation the Launcher does must be a write. All writes |
| 288 | * start with an unsigned long number: for the first write this must be |
| 289 | * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use |
| 290 | * writes of other values to send interrupts or set up receipt of notifications. |
| 291 | * |
| 292 | * Note that we overload the "offset" in the /dev/lguest file to indicate what |
| 293 | * CPU number we're dealing with. Currently this is always 0 since we only |
| 294 | * support uniprocessor Guests, but you can see the beginnings of SMP support |
| 295 | * here. |
| 296 | */ |
| 297 | static ssize_t write(struct file *file, const char __user *in, |
| 298 | size_t size, loff_t *off) |
| 299 | { |
| 300 | /* |
| 301 | * Once the Guest is initialized, we hold the "struct lguest" in the |
| 302 | * file private data. |
| 303 | */ |
| 304 | struct lguest *lg = file->private_data; |
| 305 | const unsigned long __user *input = (const unsigned long __user *)in; |
| 306 | unsigned long req; |
| 307 | struct lg_cpu *uninitialized_var(cpu); |
| 308 | unsigned int cpu_id = *off; |
| 309 | |
| 310 | /* The first value tells us what this request is. */ |
| 311 | if (get_user(req, input) != 0) |
| 312 | return -EFAULT; |
| 313 | input++; |
| 314 | |
| 315 | /* If you haven't initialized, you must do that first. */ |
| 316 | if (req != LHREQ_INITIALIZE) { |
| 317 | if (!lg || (cpu_id >= lg->nr_cpus)) |
| 318 | return -EINVAL; |
| 319 | cpu = &lg->cpus[cpu_id]; |
| 320 | |
| 321 | /* Once the Guest is dead, you can only read() why it died. */ |
| 322 | if (lg->dead) |
| 323 | return -ENOENT; |
| 324 | } |
| 325 | |
| 326 | switch (req) { |
| 327 | case LHREQ_INITIALIZE: |
| 328 | return initialize(file, input); |
| 329 | case LHREQ_IRQ: |
| 330 | return user_send_irq(cpu, input); |
| 331 | case LHREQ_GETREG: |
| 332 | return getreg_setup(cpu, input); |
| 333 | case LHREQ_SETREG: |
| 334 | return setreg(cpu, input); |
| 335 | case LHREQ_TRAP: |
| 336 | return trap(cpu, input); |
| 337 | default: |
| 338 | return -EINVAL; |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | static int open(struct inode *inode, struct file *file) |
| 343 | { |
| 344 | file->private_data = NULL; |
| 345 | |
| 346 | return 0; |
| 347 | } |
| 348 | |
| 349 | /*L:060 |
| 350 | * The final piece of interface code is the close() routine. It reverses |
| 351 | * everything done in initialize(). This is usually called because the |
| 352 | * Launcher exited. |
| 353 | * |
| 354 | * Note that the close routine returns 0 or a negative error number: it can't |
| 355 | * really fail, but it can whine. I blame Sun for this wart, and K&R C for |
| 356 | * letting them do it. |
| 357 | :*/ |
| 358 | static int close(struct inode *inode, struct file *file) |
| 359 | { |
| 360 | struct lguest *lg = file->private_data; |
| 361 | unsigned int i; |
| 362 | |
| 363 | /* If we never successfully initialized, there's nothing to clean up */ |
| 364 | if (!lg) |
| 365 | return 0; |
| 366 | |
| 367 | /* |
| 368 | * We need the big lock, to protect from inter-guest I/O and other |
| 369 | * Launchers initializing guests. |
| 370 | */ |
| 371 | mutex_lock(&lguest_lock); |
| 372 | |
| 373 | /* Free up the shadow page tables for the Guest. */ |
| 374 | free_guest_pagetable(lg); |
| 375 | |
| 376 | for (i = 0; i < lg->nr_cpus; i++) { |
| 377 | /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ |
| 378 | hrtimer_cancel(&lg->cpus[i].hrt); |
| 379 | /* We can free up the register page we allocated. */ |
| 380 | free_page(lg->cpus[i].regs_page); |
| 381 | /* |
| 382 | * Now all the memory cleanups are done, it's safe to release |
| 383 | * the Launcher's memory management structure. |
| 384 | */ |
| 385 | mmput(lg->cpus[i].mm); |
| 386 | } |
| 387 | |
| 388 | /* |
| 389 | * If lg->dead doesn't contain an error code it will be NULL or a |
| 390 | * kmalloc()ed string, either of which is ok to hand to kfree(). |
| 391 | */ |
| 392 | if (!IS_ERR(lg->dead)) |
| 393 | kfree(lg->dead); |
| 394 | /* Free the memory allocated to the lguest_struct */ |
| 395 | kfree(lg); |
| 396 | /* Release lock and exit. */ |
| 397 | mutex_unlock(&lguest_lock); |
| 398 | |
| 399 | return 0; |
| 400 | } |
| 401 | |
| 402 | /*L:000 |
| 403 | * Welcome to our journey through the Launcher! |
| 404 | * |
| 405 | * The Launcher is the Host userspace program which sets up, runs and services |
| 406 | * the Guest. In fact, many comments in the Drivers which refer to "the Host" |
| 407 | * doing things are inaccurate: the Launcher does all the device handling for |
| 408 | * the Guest, but the Guest can't know that. |
| 409 | * |
| 410 | * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we |
| 411 | * shall see more of that later. |
| 412 | * |
| 413 | * We begin our understanding with the Host kernel interface which the Launcher |
| 414 | * uses: reading and writing a character device called /dev/lguest. All the |
| 415 | * work happens in the read(), write() and close() routines: |
| 416 | */ |
| 417 | static const struct file_operations lguest_fops = { |
| 418 | .owner = THIS_MODULE, |
| 419 | .open = open, |
| 420 | .release = close, |
| 421 | .write = write, |
| 422 | .read = read, |
| 423 | .llseek = default_llseek, |
| 424 | }; |
| 425 | /*:*/ |
| 426 | |
| 427 | /* |
| 428 | * This is a textbook example of a "misc" character device. Populate a "struct |
| 429 | * miscdevice" and register it with misc_register(). |
| 430 | */ |
| 431 | static struct miscdevice lguest_dev = { |
| 432 | .minor = MISC_DYNAMIC_MINOR, |
| 433 | .name = "lguest", |
| 434 | .fops = &lguest_fops, |
| 435 | }; |
| 436 | |
| 437 | int __init lguest_device_init(void) |
| 438 | { |
| 439 | return misc_register(&lguest_dev); |
| 440 | } |
| 441 | |
| 442 | void __exit lguest_device_remove(void) |
| 443 | { |
| 444 | misc_deregister(&lguest_dev); |
| 445 | } |