lguest_user.c revision 828c09509b9695271bcbdc53e9fc9a6a737148d2
1/*P:200 This contains all the /dev/lguest code, whereby the userspace launcher 2 * controls and communicates with the Guest. For example, the first write will 3 * tell us the Guest's memory layout and entry point. A read will run the 4 * Guest until something happens, such as a signal or the Guest doing a NOTIFY 5 * out to the Launcher. 6:*/ 7#include <linux/uaccess.h> 8#include <linux/miscdevice.h> 9#include <linux/fs.h> 10#include <linux/sched.h> 11#include <linux/eventfd.h> 12#include <linux/file.h> 13#include "lg.h" 14 15/*L:056 16 * Before we move on, let's jump ahead and look at what the kernel does when 17 * it needs to look up the eventfds. That will complete our picture of how we 18 * use RCU. 19 * 20 * The notification value is in cpu->pending_notify: we return true if it went 21 * to an eventfd. 22 */ 23bool send_notify_to_eventfd(struct lg_cpu *cpu) 24{ 25 unsigned int i; 26 struct lg_eventfd_map *map; 27 28 /* 29 * This "rcu_read_lock()" helps track when someone is still looking at 30 * the (RCU-using) eventfds array. It's not actually a lock at all; 31 * indeed it's a noop in many configurations. (You didn't expect me to 32 * explain all the RCU secrets here, did you?) 33 */ 34 rcu_read_lock(); 35 /* 36 * rcu_dereference is the counter-side of rcu_assign_pointer(); it 37 * makes sure we don't access the memory pointed to by 38 * cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy, 39 * but Alpha allows this! Paul McKenney points out that a really 40 * aggressive compiler could have the same effect: 41 * http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html 42 * 43 * So play safe, use rcu_dereference to get the rcu-protected pointer: 44 */ 45 map = rcu_dereference(cpu->lg->eventfds); 46 /* 47 * Simple array search: even if they add an eventfd while we do this, 48 * we'll continue to use the old array and just won't see the new one. 49 */ 50 for (i = 0; i < map->num; i++) { 51 if (map->map[i].addr == cpu->pending_notify) { 52 eventfd_signal(map->map[i].event, 1); 53 cpu->pending_notify = 0; 54 break; 55 } 56 } 57 /* We're done with the rcu-protected variable cpu->lg->eventfds. */ 58 rcu_read_unlock(); 59 60 /* If we cleared the notification, it's because we found a match. */ 61 return cpu->pending_notify == 0; 62} 63 64/*L:055 65 * One of the more tricksy tricks in the Linux Kernel is a technique called 66 * Read Copy Update. Since one point of lguest is to teach lguest journeyers 67 * about kernel coding, I use it here. (In case you're curious, other purposes 68 * include learning about virtualization and instilling a deep appreciation for 69 * simplicity and puppies). 70 * 71 * We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we 72 * add new eventfds without ever blocking readers from accessing the array. 73 * The current Launcher only does this during boot, so that never happens. But 74 * Read Copy Update is cool, and adding a lock risks damaging even more puppies 75 * than this code does. 76 * 77 * We allocate a brand new one-larger array, copy the old one and add our new 78 * element. Then we make the lg eventfd pointer point to the new array. 79 * That's the easy part: now we need to free the old one, but we need to make 80 * sure no slow CPU somewhere is still looking at it. That's what 81 * synchronize_rcu does for us: waits until every CPU has indicated that it has 82 * moved on to know it's no longer using the old one. 83 * 84 * If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update. 85 */ 86static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) 87{ 88 struct lg_eventfd_map *new, *old = lg->eventfds; 89 90 /* 91 * We don't allow notifications on value 0 anyway (pending_notify of 92 * 0 means "nothing pending"). 93 */ 94 if (!addr) 95 return -EINVAL; 96 97 /* 98 * Replace the old array with the new one, carefully: others can 99 * be accessing it at the same time. 100 */ 101 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), 102 GFP_KERNEL); 103 if (!new) 104 return -ENOMEM; 105 106 /* First make identical copy. */ 107 memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); 108 new->num = old->num; 109 110 /* Now append new entry. */ 111 new->map[new->num].addr = addr; 112 new->map[new->num].event = eventfd_ctx_fdget(fd); 113 if (IS_ERR(new->map[new->num].event)) { 114 int err = PTR_ERR(new->map[new->num].event); 115 kfree(new); 116 return err; 117 } 118 new->num++; 119 120 /* 121 * Now put new one in place: rcu_assign_pointer() is a fancy way of 122 * doing "lg->eventfds = new", but it uses memory barriers to make 123 * absolutely sure that the contents of "new" written above is nailed 124 * down before we actually do the assignment. 125 * 126 * We have to think about these kinds of things when we're operating on 127 * live data without locks. 128 */ 129 rcu_assign_pointer(lg->eventfds, new); 130 131 /* 132 * We're not in a big hurry. Wait until noone's looking at old 133 * version, then free it. 134 */ 135 synchronize_rcu(); 136 kfree(old); 137 138 return 0; 139} 140 141/*L:052 142 * Receiving notifications from the Guest is usually done by attaching a 143 * particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will 144 * become readable when the Guest does an LHCALL_NOTIFY with that value. 145 * 146 * This is really convenient for processing each virtqueue in a separate 147 * thread. 148 */ 149static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) 150{ 151 unsigned long addr, fd; 152 int err; 153 154 if (get_user(addr, input) != 0) 155 return -EFAULT; 156 input++; 157 if (get_user(fd, input) != 0) 158 return -EFAULT; 159 160 /* 161 * Just make sure two callers don't add eventfds at once. We really 162 * only need to lock against callers adding to the same Guest, so using 163 * the Big Lguest Lock is overkill. But this is setup, not a fast path. 164 */ 165 mutex_lock(&lguest_lock); 166 err = add_eventfd(lg, addr, fd); 167 mutex_unlock(&lguest_lock); 168 169 return err; 170} 171 172/*L:050 173 * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt 174 * number to /dev/lguest. 175 */ 176static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) 177{ 178 unsigned long irq; 179 180 if (get_user(irq, input) != 0) 181 return -EFAULT; 182 if (irq >= LGUEST_IRQS) 183 return -EINVAL; 184 185 /* 186 * Next time the Guest runs, the core code will see if it can deliver 187 * this interrupt. 188 */ 189 set_interrupt(cpu, irq); 190 return 0; 191} 192 193/*L:040 194 * Once our Guest is initialized, the Launcher makes it run by reading 195 * from /dev/lguest. 196 */ 197static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) 198{ 199 struct lguest *lg = file->private_data; 200 struct lg_cpu *cpu; 201 unsigned int cpu_id = *o; 202 203 /* You must write LHREQ_INITIALIZE first! */ 204 if (!lg) 205 return -EINVAL; 206 207 /* Watch out for arbitrary vcpu indexes! */ 208 if (cpu_id >= lg->nr_cpus) 209 return -EINVAL; 210 211 cpu = &lg->cpus[cpu_id]; 212 213 /* If you're not the task which owns the Guest, go away. */ 214 if (current != cpu->tsk) 215 return -EPERM; 216 217 /* If the Guest is already dead, we indicate why */ 218 if (lg->dead) { 219 size_t len; 220 221 /* lg->dead either contains an error code, or a string. */ 222 if (IS_ERR(lg->dead)) 223 return PTR_ERR(lg->dead); 224 225 /* We can only return as much as the buffer they read with. */ 226 len = min(size, strlen(lg->dead)+1); 227 if (copy_to_user(user, lg->dead, len) != 0) 228 return -EFAULT; 229 return len; 230 } 231 232 /* 233 * If we returned from read() last time because the Guest sent I/O, 234 * clear the flag. 235 */ 236 if (cpu->pending_notify) 237 cpu->pending_notify = 0; 238 239 /* Run the Guest until something interesting happens. */ 240 return run_guest(cpu, (unsigned long __user *)user); 241} 242 243/*L:025 244 * This actually initializes a CPU. For the moment, a Guest is only 245 * uniprocessor, so "id" is always 0. 246 */ 247static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) 248{ 249 /* We have a limited number the number of CPUs in the lguest struct. */ 250 if (id >= ARRAY_SIZE(cpu->lg->cpus)) 251 return -EINVAL; 252 253 /* Set up this CPU's id, and pointer back to the lguest struct. */ 254 cpu->id = id; 255 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); 256 cpu->lg->nr_cpus++; 257 258 /* Each CPU has a timer it can set. */ 259 init_clockdev(cpu); 260 261 /* 262 * We need a complete page for the Guest registers: they are accessible 263 * to the Guest and we can only grant it access to whole pages. 264 */ 265 cpu->regs_page = get_zeroed_page(GFP_KERNEL); 266 if (!cpu->regs_page) 267 return -ENOMEM; 268 269 /* We actually put the registers at the bottom of the page. */ 270 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); 271 272 /* 273 * Now we initialize the Guest's registers, handing it the start 274 * address. 275 */ 276 lguest_arch_setup_regs(cpu, start_ip); 277 278 /* 279 * We keep a pointer to the Launcher task (ie. current task) for when 280 * other Guests want to wake this one (eg. console input). 281 */ 282 cpu->tsk = current; 283 284 /* 285 * We need to keep a pointer to the Launcher's memory map, because if 286 * the Launcher dies we need to clean it up. If we don't keep a 287 * reference, it is destroyed before close() is called. 288 */ 289 cpu->mm = get_task_mm(cpu->tsk); 290 291 /* 292 * We remember which CPU's pages this Guest used last, for optimization 293 * when the same Guest runs on the same CPU twice. 294 */ 295 cpu->last_pages = NULL; 296 297 /* No error == success. */ 298 return 0; 299} 300 301/*L:020 302 * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in 303 * addition to the LHREQ_INITIALIZE value). These are: 304 * 305 * base: The start of the Guest-physical memory inside the Launcher memory. 306 * 307 * pfnlimit: The highest (Guest-physical) page number the Guest should be 308 * allowed to access. The Guest memory lives inside the Launcher, so it sets 309 * this to ensure the Guest can only reach its own memory. 310 * 311 * start: The first instruction to execute ("eip" in x86-speak). 312 */ 313static int initialize(struct file *file, const unsigned long __user *input) 314{ 315 /* "struct lguest" contains all we (the Host) know about a Guest. */ 316 struct lguest *lg; 317 int err; 318 unsigned long args[3]; 319 320 /* 321 * We grab the Big Lguest lock, which protects against multiple 322 * simultaneous initializations. 323 */ 324 mutex_lock(&lguest_lock); 325 /* You can't initialize twice! Close the device and start again... */ 326 if (file->private_data) { 327 err = -EBUSY; 328 goto unlock; 329 } 330 331 if (copy_from_user(args, input, sizeof(args)) != 0) { 332 err = -EFAULT; 333 goto unlock; 334 } 335 336 lg = kzalloc(sizeof(*lg), GFP_KERNEL); 337 if (!lg) { 338 err = -ENOMEM; 339 goto unlock; 340 } 341 342 lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); 343 if (!lg->eventfds) { 344 err = -ENOMEM; 345 goto free_lg; 346 } 347 lg->eventfds->num = 0; 348 349 /* Populate the easy fields of our "struct lguest" */ 350 lg->mem_base = (void __user *)args[0]; 351 lg->pfn_limit = args[1]; 352 353 /* This is the first cpu (cpu 0) and it will start booting at args[2] */ 354 err = lg_cpu_start(&lg->cpus[0], 0, args[2]); 355 if (err) 356 goto free_eventfds; 357 358 /* 359 * Initialize the Guest's shadow page tables, using the toplevel 360 * address the Launcher gave us. This allocates memory, so can fail. 361 */ 362 err = init_guest_pagetable(lg); 363 if (err) 364 goto free_regs; 365 366 /* We keep our "struct lguest" in the file's private_data. */ 367 file->private_data = lg; 368 369 mutex_unlock(&lguest_lock); 370 371 /* And because this is a write() call, we return the length used. */ 372 return sizeof(args); 373 374free_regs: 375 /* FIXME: This should be in free_vcpu */ 376 free_page(lg->cpus[0].regs_page); 377free_eventfds: 378 kfree(lg->eventfds); 379free_lg: 380 kfree(lg); 381unlock: 382 mutex_unlock(&lguest_lock); 383 return err; 384} 385 386/*L:010 387 * The first operation the Launcher does must be a write. All writes 388 * start with an unsigned long number: for the first write this must be 389 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 390 * writes of other values to send interrupts or set up receipt of notifications. 391 * 392 * Note that we overload the "offset" in the /dev/lguest file to indicate what 393 * CPU number we're dealing with. Currently this is always 0 since we only 394 * support uniprocessor Guests, but you can see the beginnings of SMP support 395 * here. 396 */ 397static ssize_t write(struct file *file, const char __user *in, 398 size_t size, loff_t *off) 399{ 400 /* 401 * Once the Guest is initialized, we hold the "struct lguest" in the 402 * file private data. 403 */ 404 struct lguest *lg = file->private_data; 405 const unsigned long __user *input = (const unsigned long __user *)in; 406 unsigned long req; 407 struct lg_cpu *uninitialized_var(cpu); 408 unsigned int cpu_id = *off; 409 410 /* The first value tells us what this request is. */ 411 if (get_user(req, input) != 0) 412 return -EFAULT; 413 input++; 414 415 /* If you haven't initialized, you must do that first. */ 416 if (req != LHREQ_INITIALIZE) { 417 if (!lg || (cpu_id >= lg->nr_cpus)) 418 return -EINVAL; 419 cpu = &lg->cpus[cpu_id]; 420 421 /* Once the Guest is dead, you can only read() why it died. */ 422 if (lg->dead) 423 return -ENOENT; 424 } 425 426 switch (req) { 427 case LHREQ_INITIALIZE: 428 return initialize(file, input); 429 case LHREQ_IRQ: 430 return user_send_irq(cpu, input); 431 case LHREQ_EVENTFD: 432 return attach_eventfd(lg, input); 433 default: 434 return -EINVAL; 435 } 436} 437 438/*L:060 439 * The final piece of interface code is the close() routine. It reverses 440 * everything done in initialize(). This is usually called because the 441 * Launcher exited. 442 * 443 * Note that the close routine returns 0 or a negative error number: it can't 444 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 445 * letting them do it. 446:*/ 447static int close(struct inode *inode, struct file *file) 448{ 449 struct lguest *lg = file->private_data; 450 unsigned int i; 451 452 /* If we never successfully initialized, there's nothing to clean up */ 453 if (!lg) 454 return 0; 455 456 /* 457 * We need the big lock, to protect from inter-guest I/O and other 458 * Launchers initializing guests. 459 */ 460 mutex_lock(&lguest_lock); 461 462 /* Free up the shadow page tables for the Guest. */ 463 free_guest_pagetable(lg); 464 465 for (i = 0; i < lg->nr_cpus; i++) { 466 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 467 hrtimer_cancel(&lg->cpus[i].hrt); 468 /* We can free up the register page we allocated. */ 469 free_page(lg->cpus[i].regs_page); 470 /* 471 * Now all the memory cleanups are done, it's safe to release 472 * the Launcher's memory management structure. 473 */ 474 mmput(lg->cpus[i].mm); 475 } 476 477 /* Release any eventfds they registered. */ 478 for (i = 0; i < lg->eventfds->num; i++) 479 eventfd_ctx_put(lg->eventfds->map[i].event); 480 kfree(lg->eventfds); 481 482 /* 483 * If lg->dead doesn't contain an error code it will be NULL or a 484 * kmalloc()ed string, either of which is ok to hand to kfree(). 485 */ 486 if (!IS_ERR(lg->dead)) 487 kfree(lg->dead); 488 /* Free the memory allocated to the lguest_struct */ 489 kfree(lg); 490 /* Release lock and exit. */ 491 mutex_unlock(&lguest_lock); 492 493 return 0; 494} 495 496/*L:000 497 * Welcome to our journey through the Launcher! 498 * 499 * The Launcher is the Host userspace program which sets up, runs and services 500 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 501 * doing things are inaccurate: the Launcher does all the device handling for 502 * the Guest, but the Guest can't know that. 503 * 504 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 505 * shall see more of that later. 506 * 507 * We begin our understanding with the Host kernel interface which the Launcher 508 * uses: reading and writing a character device called /dev/lguest. All the 509 * work happens in the read(), write() and close() routines: 510 */ 511static const struct file_operations lguest_fops = { 512 .owner = THIS_MODULE, 513 .release = close, 514 .write = write, 515 .read = read, 516}; 517 518/* 519 * This is a textbook example of a "misc" character device. Populate a "struct 520 * miscdevice" and register it with misc_register(). 521 */ 522static struct miscdevice lguest_dev = { 523 .minor = MISC_DYNAMIC_MINOR, 524 .name = "lguest", 525 .fops = &lguest_fops, 526}; 527 528int __init lguest_device_init(void) 529{ 530 return misc_register(&lguest_dev); 531} 532 533void __exit lguest_device_remove(void) 534{ 535 misc_deregister(&lguest_dev); 536} 537