| 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 doing a NOTIFY out to the Launcher. There is |
| 6 | * also a way for the Launcher to attach eventfds to particular NOTIFY |
| 7 | * values instead of returning from the read() call. |
| 8 | :*/ |
| 9 | #include <linux/uaccess.h> |
| 10 | #include <linux/miscdevice.h> |
| 11 | #include <linux/fs.h> |
| 12 | #include <linux/sched.h> |
| 13 | #include <linux/eventfd.h> |
| 14 | #include <linux/file.h> |
| 15 | #include <linux/slab.h> |
| 16 | #include <linux/export.h> |
| 17 | #include "lg.h" |
| 18 | |
| 19 | /*L:056 |
| 20 | * Before we move on, let's jump ahead and look at what the kernel does when |
| 21 | * it needs to look up the eventfds. That will complete our picture of how we |
| 22 | * use RCU. |
| 23 | * |
| 24 | * The notification value is in cpu->pending_notify: we return true if it went |
| 25 | * to an eventfd. |
| 26 | */ |
| 27 | bool send_notify_to_eventfd(struct lg_cpu *cpu) |
| 28 | { |
| 29 | unsigned int i; |
| 30 | struct lg_eventfd_map *map; |
| 31 | |
| 32 | /* |
| 33 | * This "rcu_read_lock()" helps track when someone is still looking at |
| 34 | * the (RCU-using) eventfds array. It's not actually a lock at all; |
| 35 | * indeed it's a noop in many configurations. (You didn't expect me to |
| 36 | * explain all the RCU secrets here, did you?) |
| 37 | */ |
| 38 | rcu_read_lock(); |
| 39 | /* |
| 40 | * rcu_dereference is the counter-side of rcu_assign_pointer(); it |
| 41 | * makes sure we don't access the memory pointed to by |
| 42 | * cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy, |
| 43 | * but Alpha allows this! Paul McKenney points out that a really |
| 44 | * aggressive compiler could have the same effect: |
| 45 | * http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html |
| 46 | * |
| 47 | * So play safe, use rcu_dereference to get the rcu-protected pointer: |
| 48 | */ |
| 49 | map = rcu_dereference(cpu->lg->eventfds); |
| 50 | /* |
| 51 | * Simple array search: even if they add an eventfd while we do this, |
| 52 | * we'll continue to use the old array and just won't see the new one. |
| 53 | */ |
| 54 | for (i = 0; i < map->num; i++) { |
| 55 | if (map->map[i].addr == cpu->pending_notify) { |
| 56 | eventfd_signal(map->map[i].event, 1); |
| 57 | cpu->pending_notify = 0; |
| 58 | break; |
| 59 | } |
| 60 | } |
| 61 | /* We're done with the rcu-protected variable cpu->lg->eventfds. */ |
| 62 | rcu_read_unlock(); |
| 63 | |
| 64 | /* If we cleared the notification, it's because we found a match. */ |
| 65 | return cpu->pending_notify == 0; |
| 66 | } |
| 67 | |
| 68 | /*L:055 |
| 69 | * One of the more tricksy tricks in the Linux Kernel is a technique called |
| 70 | * Read Copy Update. Since one point of lguest is to teach lguest journeyers |
| 71 | * about kernel coding, I use it here. (In case you're curious, other purposes |
| 72 | * include learning about virtualization and instilling a deep appreciation for |
| 73 | * simplicity and puppies). |
| 74 | * |
| 75 | * We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we |
| 76 | * add new eventfds without ever blocking readers from accessing the array. |
| 77 | * The current Launcher only does this during boot, so that never happens. But |
| 78 | * Read Copy Update is cool, and adding a lock risks damaging even more puppies |
| 79 | * than this code does. |
| 80 | * |
| 81 | * We allocate a brand new one-larger array, copy the old one and add our new |
| 82 | * element. Then we make the lg eventfd pointer point to the new array. |
| 83 | * That's the easy part: now we need to free the old one, but we need to make |
| 84 | * sure no slow CPU somewhere is still looking at it. That's what |
| 85 | * synchronize_rcu does for us: waits until every CPU has indicated that it has |
| 86 | * moved on to know it's no longer using the old one. |
| 87 | * |
| 88 | * If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update. |
| 89 | */ |
| 90 | static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) |
| 91 | { |
| 92 | struct lg_eventfd_map *new, *old = lg->eventfds; |
| 93 | |
| 94 | /* |
| 95 | * We don't allow notifications on value 0 anyway (pending_notify of |
| 96 | * 0 means "nothing pending"). |
| 97 | */ |
| 98 | if (!addr) |
| 99 | return -EINVAL; |
| 100 | |
| 101 | /* |
| 102 | * Replace the old array with the new one, carefully: others can |
| 103 | * be accessing it at the same time. |
| 104 | */ |
| 105 | new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), |
| 106 | GFP_KERNEL); |
| 107 | if (!new) |
| 108 | return -ENOMEM; |
| 109 | |
| 110 | /* First make identical copy. */ |
| 111 | memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); |
| 112 | new->num = old->num; |
| 113 | |
| 114 | /* Now append new entry. */ |
| 115 | new->map[new->num].addr = addr; |
| 116 | new->map[new->num].event = eventfd_ctx_fdget(fd); |
| 117 | if (IS_ERR(new->map[new->num].event)) { |
| 118 | int err = PTR_ERR(new->map[new->num].event); |
| 119 | kfree(new); |
| 120 | return err; |
| 121 | } |
| 122 | new->num++; |
| 123 | |
| 124 | /* |
| 125 | * Now put new one in place: rcu_assign_pointer() is a fancy way of |
| 126 | * doing "lg->eventfds = new", but it uses memory barriers to make |
| 127 | * absolutely sure that the contents of "new" written above is nailed |
| 128 | * down before we actually do the assignment. |
| 129 | * |
| 130 | * We have to think about these kinds of things when we're operating on |
| 131 | * live data without locks. |
| 132 | */ |
| 133 | rcu_assign_pointer(lg->eventfds, new); |
| 134 | |
| 135 | /* |
| 136 | * We're not in a big hurry. Wait until no one's looking at old |
| 137 | * version, then free it. |
| 138 | */ |
| 139 | synchronize_rcu(); |
| 140 | kfree(old); |
| 141 | |
| 142 | return 0; |
| 143 | } |
| 144 | |
| 145 | /*L:052 |
| 146 | * Receiving notifications from the Guest is usually done by attaching a |
| 147 | * particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will |
| 148 | * become readable when the Guest does an LHCALL_NOTIFY with that value. |
| 149 | * |
| 150 | * This is really convenient for processing each virtqueue in a separate |
| 151 | * thread. |
| 152 | */ |
| 153 | static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) |
| 154 | { |
| 155 | unsigned long addr, fd; |
| 156 | int err; |
| 157 | |
| 158 | if (get_user(addr, input) != 0) |
| 159 | return -EFAULT; |
| 160 | input++; |
| 161 | if (get_user(fd, input) != 0) |
| 162 | return -EFAULT; |
| 163 | |
| 164 | /* |
| 165 | * Just make sure two callers don't add eventfds at once. We really |
| 166 | * only need to lock against callers adding to the same Guest, so using |
| 167 | * the Big Lguest Lock is overkill. But this is setup, not a fast path. |
| 168 | */ |
| 169 | mutex_lock(&lguest_lock); |
| 170 | err = add_eventfd(lg, addr, fd); |
| 171 | mutex_unlock(&lguest_lock); |
| 172 | |
| 173 | return err; |
| 174 | } |
| 175 | |
| 176 | /*L:050 |
| 177 | * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt |
| 178 | * number to /dev/lguest. |
| 179 | */ |
| 180 | static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) |
| 181 | { |
| 182 | unsigned long irq; |
| 183 | |
| 184 | if (get_user(irq, input) != 0) |
| 185 | return -EFAULT; |
| 186 | if (irq >= LGUEST_IRQS) |
| 187 | return -EINVAL; |
| 188 | |
| 189 | /* |
| 190 | * Next time the Guest runs, the core code will see if it can deliver |
| 191 | * this interrupt. |
| 192 | */ |
| 193 | set_interrupt(cpu, irq); |
| 194 | return 0; |
| 195 | } |
| 196 | |
| 197 | /*L:040 |
| 198 | * Once our Guest is initialized, the Launcher makes it run by reading |
| 199 | * from /dev/lguest. |
| 200 | */ |
| 201 | static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) |
| 202 | { |
| 203 | struct lguest *lg = file->private_data; |
| 204 | struct lg_cpu *cpu; |
| 205 | unsigned int cpu_id = *o; |
| 206 | |
| 207 | /* You must write LHREQ_INITIALIZE first! */ |
| 208 | if (!lg) |
| 209 | return -EINVAL; |
| 210 | |
| 211 | /* Watch out for arbitrary vcpu indexes! */ |
| 212 | if (cpu_id >= lg->nr_cpus) |
| 213 | return -EINVAL; |
| 214 | |
| 215 | cpu = &lg->cpus[cpu_id]; |
| 216 | |
| 217 | /* If you're not the task which owns the Guest, go away. */ |
| 218 | if (current != cpu->tsk) |
| 219 | return -EPERM; |
| 220 | |
| 221 | /* If the Guest is already dead, we indicate why */ |
| 222 | if (lg->dead) { |
| 223 | size_t len; |
| 224 | |
| 225 | /* lg->dead either contains an error code, or a string. */ |
| 226 | if (IS_ERR(lg->dead)) |
| 227 | return PTR_ERR(lg->dead); |
| 228 | |
| 229 | /* We can only return as much as the buffer they read with. */ |
| 230 | len = min(size, strlen(lg->dead)+1); |
| 231 | if (copy_to_user(user, lg->dead, len) != 0) |
| 232 | return -EFAULT; |
| 233 | return len; |
| 234 | } |
| 235 | |
| 236 | /* |
| 237 | * If we returned from read() last time because the Guest sent I/O, |
| 238 | * clear the flag. |
| 239 | */ |
| 240 | if (cpu->pending_notify) |
| 241 | cpu->pending_notify = 0; |
| 242 | |
| 243 | /* Run the Guest until something interesting happens. */ |
| 244 | return run_guest(cpu, (unsigned long __user *)user); |
| 245 | } |
| 246 | |
| 247 | /*L:025 |
| 248 | * This actually initializes a CPU. For the moment, a Guest is only |
| 249 | * uniprocessor, so "id" is always 0. |
| 250 | */ |
| 251 | static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) |
| 252 | { |
| 253 | /* We have a limited number the number of CPUs in the lguest struct. */ |
| 254 | if (id >= ARRAY_SIZE(cpu->lg->cpus)) |
| 255 | return -EINVAL; |
| 256 | |
| 257 | /* Set up this CPU's id, and pointer back to the lguest struct. */ |
| 258 | cpu->id = id; |
| 259 | cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); |
| 260 | cpu->lg->nr_cpus++; |
| 261 | |
| 262 | /* Each CPU has a timer it can set. */ |
| 263 | init_clockdev(cpu); |
| 264 | |
| 265 | /* |
| 266 | * We need a complete page for the Guest registers: they are accessible |
| 267 | * to the Guest and we can only grant it access to whole pages. |
| 268 | */ |
| 269 | cpu->regs_page = get_zeroed_page(GFP_KERNEL); |
| 270 | if (!cpu->regs_page) |
| 271 | return -ENOMEM; |
| 272 | |
| 273 | /* We actually put the registers at the bottom of the page. */ |
| 274 | cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); |
| 275 | |
| 276 | /* |
| 277 | * Now we initialize the Guest's registers, handing it the start |
| 278 | * address. |
| 279 | */ |
| 280 | lguest_arch_setup_regs(cpu, start_ip); |
| 281 | |
| 282 | /* |
| 283 | * We keep a pointer to the Launcher task (ie. current task) for when |
| 284 | * other Guests want to wake this one (eg. console input). |
| 285 | */ |
| 286 | cpu->tsk = current; |
| 287 | |
| 288 | /* |
| 289 | * We need to keep a pointer to the Launcher's memory map, because if |
| 290 | * the Launcher dies we need to clean it up. If we don't keep a |
| 291 | * reference, it is destroyed before close() is called. |
| 292 | */ |
| 293 | cpu->mm = get_task_mm(cpu->tsk); |
| 294 | |
| 295 | /* |
| 296 | * We remember which CPU's pages this Guest used last, for optimization |
| 297 | * when the same Guest runs on the same CPU twice. |
| 298 | */ |
| 299 | cpu->last_pages = NULL; |
| 300 | |
| 301 | /* No error == success. */ |
| 302 | return 0; |
| 303 | } |
| 304 | |
| 305 | /*L:020 |
| 306 | * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in |
| 307 | * addition to the LHREQ_INITIALIZE value). These are: |
| 308 | * |
| 309 | * base: The start of the Guest-physical memory inside the Launcher memory. |
| 310 | * |
| 311 | * pfnlimit: The highest (Guest-physical) page number the Guest should be |
| 312 | * allowed to access. The Guest memory lives inside the Launcher, so it sets |
| 313 | * this to ensure the Guest can only reach its own memory. |
| 314 | * |
| 315 | * start: The first instruction to execute ("eip" in x86-speak). |
| 316 | */ |
| 317 | static int initialize(struct file *file, const unsigned long __user *input) |
| 318 | { |
| 319 | /* "struct lguest" contains all we (the Host) know about a Guest. */ |
| 320 | struct lguest *lg; |
| 321 | int err; |
| 322 | unsigned long args[3]; |
| 323 | |
| 324 | /* |
| 325 | * We grab the Big Lguest lock, which protects against multiple |
| 326 | * simultaneous initializations. |
| 327 | */ |
| 328 | mutex_lock(&lguest_lock); |
| 329 | /* You can't initialize twice! Close the device and start again... */ |
| 330 | if (file->private_data) { |
| 331 | err = -EBUSY; |
| 332 | goto unlock; |
| 333 | } |
| 334 | |
| 335 | if (copy_from_user(args, input, sizeof(args)) != 0) { |
| 336 | err = -EFAULT; |
| 337 | goto unlock; |
| 338 | } |
| 339 | |
| 340 | lg = kzalloc(sizeof(*lg), GFP_KERNEL); |
| 341 | if (!lg) { |
| 342 | err = -ENOMEM; |
| 343 | goto unlock; |
| 344 | } |
| 345 | |
| 346 | lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); |
| 347 | if (!lg->eventfds) { |
| 348 | err = -ENOMEM; |
| 349 | goto free_lg; |
| 350 | } |
| 351 | lg->eventfds->num = 0; |
| 352 | |
| 353 | /* Populate the easy fields of our "struct lguest" */ |
| 354 | lg->mem_base = (void __user *)args[0]; |
| 355 | lg->pfn_limit = args[1]; |
| 356 | |
| 357 | /* This is the first cpu (cpu 0) and it will start booting at args[2] */ |
| 358 | err = lg_cpu_start(&lg->cpus[0], 0, args[2]); |
| 359 | if (err) |
| 360 | goto free_eventfds; |
| 361 | |
| 362 | /* |
| 363 | * Initialize the Guest's shadow page tables. This allocates |
| 364 | * memory, so can fail. |
| 365 | */ |
| 366 | err = init_guest_pagetable(lg); |
| 367 | if (err) |
| 368 | goto free_regs; |
| 369 | |
| 370 | /* We keep our "struct lguest" in the file's private_data. */ |
| 371 | file->private_data = lg; |
| 372 | |
| 373 | mutex_unlock(&lguest_lock); |
| 374 | |
| 375 | /* And because this is a write() call, we return the length used. */ |
| 376 | return sizeof(args); |
| 377 | |
| 378 | free_regs: |
| 379 | /* FIXME: This should be in free_vcpu */ |
| 380 | free_page(lg->cpus[0].regs_page); |
| 381 | free_eventfds: |
| 382 | kfree(lg->eventfds); |
| 383 | free_lg: |
| 384 | kfree(lg); |
| 385 | unlock: |
| 386 | mutex_unlock(&lguest_lock); |
| 387 | return err; |
| 388 | } |
| 389 | |
| 390 | /*L:010 |
| 391 | * The first operation the Launcher does must be a write. All writes |
| 392 | * start with an unsigned long number: for the first write this must be |
| 393 | * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use |
| 394 | * writes of other values to send interrupts or set up receipt of notifications. |
| 395 | * |
| 396 | * Note that we overload the "offset" in the /dev/lguest file to indicate what |
| 397 | * CPU number we're dealing with. Currently this is always 0 since we only |
| 398 | * support uniprocessor Guests, but you can see the beginnings of SMP support |
| 399 | * here. |
| 400 | */ |
| 401 | static ssize_t write(struct file *file, const char __user *in, |
| 402 | size_t size, loff_t *off) |
| 403 | { |
| 404 | /* |
| 405 | * Once the Guest is initialized, we hold the "struct lguest" in the |
| 406 | * file private data. |
| 407 | */ |
| 408 | struct lguest *lg = file->private_data; |
| 409 | const unsigned long __user *input = (const unsigned long __user *)in; |
| 410 | unsigned long req; |
| 411 | struct lg_cpu *uninitialized_var(cpu); |
| 412 | unsigned int cpu_id = *off; |
| 413 | |
| 414 | /* The first value tells us what this request is. */ |
| 415 | if (get_user(req, input) != 0) |
| 416 | return -EFAULT; |
| 417 | input++; |
| 418 | |
| 419 | /* If you haven't initialized, you must do that first. */ |
| 420 | if (req != LHREQ_INITIALIZE) { |
| 421 | if (!lg || (cpu_id >= lg->nr_cpus)) |
| 422 | return -EINVAL; |
| 423 | cpu = &lg->cpus[cpu_id]; |
| 424 | |
| 425 | /* Once the Guest is dead, you can only read() why it died. */ |
| 426 | if (lg->dead) |
| 427 | return -ENOENT; |
| 428 | } |
| 429 | |
| 430 | switch (req) { |
| 431 | case LHREQ_INITIALIZE: |
| 432 | return initialize(file, input); |
| 433 | case LHREQ_IRQ: |
| 434 | return user_send_irq(cpu, input); |
| 435 | case LHREQ_EVENTFD: |
| 436 | return attach_eventfd(lg, input); |
| 437 | default: |
| 438 | return -EINVAL; |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | /*L:060 |
| 443 | * The final piece of interface code is the close() routine. It reverses |
| 444 | * everything done in initialize(). This is usually called because the |
| 445 | * Launcher exited. |
| 446 | * |
| 447 | * Note that the close routine returns 0 or a negative error number: it can't |
| 448 | * really fail, but it can whine. I blame Sun for this wart, and K&R C for |
| 449 | * letting them do it. |
| 450 | :*/ |
| 451 | static int close(struct inode *inode, struct file *file) |
| 452 | { |
| 453 | struct lguest *lg = file->private_data; |
| 454 | unsigned int i; |
| 455 | |
| 456 | /* If we never successfully initialized, there's nothing to clean up */ |
| 457 | if (!lg) |
| 458 | return 0; |
| 459 | |
| 460 | /* |
| 461 | * We need the big lock, to protect from inter-guest I/O and other |
| 462 | * Launchers initializing guests. |
| 463 | */ |
| 464 | mutex_lock(&lguest_lock); |
| 465 | |
| 466 | /* Free up the shadow page tables for the Guest. */ |
| 467 | free_guest_pagetable(lg); |
| 468 | |
| 469 | for (i = 0; i < lg->nr_cpus; i++) { |
| 470 | /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ |
| 471 | hrtimer_cancel(&lg->cpus[i].hrt); |
| 472 | /* We can free up the register page we allocated. */ |
| 473 | free_page(lg->cpus[i].regs_page); |
| 474 | /* |
| 475 | * Now all the memory cleanups are done, it's safe to release |
| 476 | * the Launcher's memory management structure. |
| 477 | */ |
| 478 | mmput(lg->cpus[i].mm); |
| 479 | } |
| 480 | |
| 481 | /* Release any eventfds they registered. */ |
| 482 | for (i = 0; i < lg->eventfds->num; i++) |
| 483 | eventfd_ctx_put(lg->eventfds->map[i].event); |
| 484 | kfree(lg->eventfds); |
| 485 | |
| 486 | /* |
| 487 | * If lg->dead doesn't contain an error code it will be NULL or a |
| 488 | * kmalloc()ed string, either of which is ok to hand to kfree(). |
| 489 | */ |
| 490 | if (!IS_ERR(lg->dead)) |
| 491 | kfree(lg->dead); |
| 492 | /* Free the memory allocated to the lguest_struct */ |
| 493 | kfree(lg); |
| 494 | /* Release lock and exit. */ |
| 495 | mutex_unlock(&lguest_lock); |
| 496 | |
| 497 | return 0; |
| 498 | } |
| 499 | |
| 500 | /*L:000 |
| 501 | * Welcome to our journey through the Launcher! |
| 502 | * |
| 503 | * The Launcher is the Host userspace program which sets up, runs and services |
| 504 | * the Guest. In fact, many comments in the Drivers which refer to "the Host" |
| 505 | * doing things are inaccurate: the Launcher does all the device handling for |
| 506 | * the Guest, but the Guest can't know that. |
| 507 | * |
| 508 | * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we |
| 509 | * shall see more of that later. |
| 510 | * |
| 511 | * We begin our understanding with the Host kernel interface which the Launcher |
| 512 | * uses: reading and writing a character device called /dev/lguest. All the |
| 513 | * work happens in the read(), write() and close() routines: |
| 514 | */ |
| 515 | static const struct file_operations lguest_fops = { |
| 516 | .owner = THIS_MODULE, |
| 517 | .release = close, |
| 518 | .write = write, |
| 519 | .read = read, |
| 520 | .llseek = default_llseek, |
| 521 | }; |
| 522 | /*:*/ |
| 523 | |
| 524 | /* |
| 525 | * This is a textbook example of a "misc" character device. Populate a "struct |
| 526 | * miscdevice" and register it with misc_register(). |
| 527 | */ |
| 528 | static struct miscdevice lguest_dev = { |
| 529 | .minor = MISC_DYNAMIC_MINOR, |
| 530 | .name = "lguest", |
| 531 | .fops = &lguest_fops, |
| 532 | }; |
| 533 | |
| 534 | int __init lguest_device_init(void) |
| 535 | { |
| 536 | return misc_register(&lguest_dev); |
| 537 | } |
| 538 | |
| 539 | void __exit lguest_device_remove(void) |
| 540 | { |
| 541 | misc_deregister(&lguest_dev); |
| 542 | } |