lguest_user.c revision 25985edcedea6396277003854657b5f3cb31a628
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 <linux/slab.h> 14#include "lg.h" 15 16/*L:056 17 * Before we move on, let's jump ahead and look at what the kernel does when 18 * it needs to look up the eventfds. That will complete our picture of how we 19 * use RCU. 20 * 21 * The notification value is in cpu->pending_notify: we return true if it went 22 * to an eventfd. 23 */ 24bool send_notify_to_eventfd(struct lg_cpu *cpu) 25{ 26 unsigned int i; 27 struct lg_eventfd_map *map; 28 29 /* 30 * This "rcu_read_lock()" helps track when someone is still looking at 31 * the (RCU-using) eventfds array. It's not actually a lock at all; 32 * indeed it's a noop in many configurations. (You didn't expect me to 33 * explain all the RCU secrets here, did you?) 34 */ 35 rcu_read_lock(); 36 /* 37 * rcu_dereference is the counter-side of rcu_assign_pointer(); it 38 * makes sure we don't access the memory pointed to by 39 * cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy, 40 * but Alpha allows this! Paul McKenney points out that a really 41 * aggressive compiler could have the same effect: 42 * http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html 43 * 44 * So play safe, use rcu_dereference to get the rcu-protected pointer: 45 */ 46 map = rcu_dereference(cpu->lg->eventfds); 47 /* 48 * Simple array search: even if they add an eventfd while we do this, 49 * we'll continue to use the old array and just won't see the new one. 50 */ 51 for (i = 0; i < map->num; i++) { 52 if (map->map[i].addr == cpu->pending_notify) { 53 eventfd_signal(map->map[i].event, 1); 54 cpu->pending_notify = 0; 55 break; 56 } 57 } 58 /* We're done with the rcu-protected variable cpu->lg->eventfds. */ 59 rcu_read_unlock(); 60 61 /* If we cleared the notification, it's because we found a match. */ 62 return cpu->pending_notify == 0; 63} 64 65/*L:055 66 * One of the more tricksy tricks in the Linux Kernel is a technique called 67 * Read Copy Update. Since one point of lguest is to teach lguest journeyers 68 * about kernel coding, I use it here. (In case you're curious, other purposes 69 * include learning about virtualization and instilling a deep appreciation for 70 * simplicity and puppies). 71 * 72 * We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we 73 * add new eventfds without ever blocking readers from accessing the array. 74 * The current Launcher only does this during boot, so that never happens. But 75 * Read Copy Update is cool, and adding a lock risks damaging even more puppies 76 * than this code does. 77 * 78 * We allocate a brand new one-larger array, copy the old one and add our new 79 * element. Then we make the lg eventfd pointer point to the new array. 80 * That's the easy part: now we need to free the old one, but we need to make 81 * sure no slow CPU somewhere is still looking at it. That's what 82 * synchronize_rcu does for us: waits until every CPU has indicated that it has 83 * moved on to know it's no longer using the old one. 84 * 85 * If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update. 86 */ 87static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) 88{ 89 struct lg_eventfd_map *new, *old = lg->eventfds; 90 91 /* 92 * We don't allow notifications on value 0 anyway (pending_notify of 93 * 0 means "nothing pending"). 94 */ 95 if (!addr) 96 return -EINVAL; 97 98 /* 99 * Replace the old array with the new one, carefully: others can 100 * be accessing it at the same time. 101 */ 102 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), 103 GFP_KERNEL); 104 if (!new) 105 return -ENOMEM; 106 107 /* First make identical copy. */ 108 memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); 109 new->num = old->num; 110 111 /* Now append new entry. */ 112 new->map[new->num].addr = addr; 113 new->map[new->num].event = eventfd_ctx_fdget(fd); 114 if (IS_ERR(new->map[new->num].event)) { 115 int err = PTR_ERR(new->map[new->num].event); 116 kfree(new); 117 return err; 118 } 119 new->num++; 120 121 /* 122 * Now put new one in place: rcu_assign_pointer() is a fancy way of 123 * doing "lg->eventfds = new", but it uses memory barriers to make 124 * absolutely sure that the contents of "new" written above is nailed 125 * down before we actually do the assignment. 126 * 127 * We have to think about these kinds of things when we're operating on 128 * live data without locks. 129 */ 130 rcu_assign_pointer(lg->eventfds, new); 131 132 /* 133 * We're not in a big hurry. Wait until no one's looking at old 134 * version, then free it. 135 */ 136 synchronize_rcu(); 137 kfree(old); 138 139 return 0; 140} 141 142/*L:052 143 * Receiving notifications from the Guest is usually done by attaching a 144 * particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will 145 * become readable when the Guest does an LHCALL_NOTIFY with that value. 146 * 147 * This is really convenient for processing each virtqueue in a separate 148 * thread. 149 */ 150static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) 151{ 152 unsigned long addr, fd; 153 int err; 154 155 if (get_user(addr, input) != 0) 156 return -EFAULT; 157 input++; 158 if (get_user(fd, input) != 0) 159 return -EFAULT; 160 161 /* 162 * Just make sure two callers don't add eventfds at once. We really 163 * only need to lock against callers adding to the same Guest, so using 164 * the Big Lguest Lock is overkill. But this is setup, not a fast path. 165 */ 166 mutex_lock(&lguest_lock); 167 err = add_eventfd(lg, addr, fd); 168 mutex_unlock(&lguest_lock); 169 170 return err; 171} 172 173/*L:050 174 * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt 175 * number to /dev/lguest. 176 */ 177static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) 178{ 179 unsigned long irq; 180 181 if (get_user(irq, input) != 0) 182 return -EFAULT; 183 if (irq >= LGUEST_IRQS) 184 return -EINVAL; 185 186 /* 187 * Next time the Guest runs, the core code will see if it can deliver 188 * this interrupt. 189 */ 190 set_interrupt(cpu, irq); 191 return 0; 192} 193 194/*L:040 195 * Once our Guest is initialized, the Launcher makes it run by reading 196 * from /dev/lguest. 197 */ 198static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) 199{ 200 struct lguest *lg = file->private_data; 201 struct lg_cpu *cpu; 202 unsigned int cpu_id = *o; 203 204 /* You must write LHREQ_INITIALIZE first! */ 205 if (!lg) 206 return -EINVAL; 207 208 /* Watch out for arbitrary vcpu indexes! */ 209 if (cpu_id >= lg->nr_cpus) 210 return -EINVAL; 211 212 cpu = &lg->cpus[cpu_id]; 213 214 /* If you're not the task which owns the Guest, go away. */ 215 if (current != cpu->tsk) 216 return -EPERM; 217 218 /* If the Guest is already dead, we indicate why */ 219 if (lg->dead) { 220 size_t len; 221 222 /* lg->dead either contains an error code, or a string. */ 223 if (IS_ERR(lg->dead)) 224 return PTR_ERR(lg->dead); 225 226 /* We can only return as much as the buffer they read with. */ 227 len = min(size, strlen(lg->dead)+1); 228 if (copy_to_user(user, lg->dead, len) != 0) 229 return -EFAULT; 230 return len; 231 } 232 233 /* 234 * If we returned from read() last time because the Guest sent I/O, 235 * clear the flag. 236 */ 237 if (cpu->pending_notify) 238 cpu->pending_notify = 0; 239 240 /* Run the Guest until something interesting happens. */ 241 return run_guest(cpu, (unsigned long __user *)user); 242} 243 244/*L:025 245 * This actually initializes a CPU. For the moment, a Guest is only 246 * uniprocessor, so "id" is always 0. 247 */ 248static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) 249{ 250 /* We have a limited number the number of CPUs in the lguest struct. */ 251 if (id >= ARRAY_SIZE(cpu->lg->cpus)) 252 return -EINVAL; 253 254 /* Set up this CPU's id, and pointer back to the lguest struct. */ 255 cpu->id = id; 256 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); 257 cpu->lg->nr_cpus++; 258 259 /* Each CPU has a timer it can set. */ 260 init_clockdev(cpu); 261 262 /* 263 * We need a complete page for the Guest registers: they are accessible 264 * to the Guest and we can only grant it access to whole pages. 265 */ 266 cpu->regs_page = get_zeroed_page(GFP_KERNEL); 267 if (!cpu->regs_page) 268 return -ENOMEM; 269 270 /* We actually put the registers at the bottom of the page. */ 271 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); 272 273 /* 274 * Now we initialize the Guest's registers, handing it the start 275 * address. 276 */ 277 lguest_arch_setup_regs(cpu, start_ip); 278 279 /* 280 * We keep a pointer to the Launcher task (ie. current task) for when 281 * other Guests want to wake this one (eg. console input). 282 */ 283 cpu->tsk = current; 284 285 /* 286 * We need to keep a pointer to the Launcher's memory map, because if 287 * the Launcher dies we need to clean it up. If we don't keep a 288 * reference, it is destroyed before close() is called. 289 */ 290 cpu->mm = get_task_mm(cpu->tsk); 291 292 /* 293 * We remember which CPU's pages this Guest used last, for optimization 294 * when the same Guest runs on the same CPU twice. 295 */ 296 cpu->last_pages = NULL; 297 298 /* No error == success. */ 299 return 0; 300} 301 302/*L:020 303 * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in 304 * addition to the LHREQ_INITIALIZE value). These are: 305 * 306 * base: The start of the Guest-physical memory inside the Launcher memory. 307 * 308 * pfnlimit: The highest (Guest-physical) page number the Guest should be 309 * allowed to access. The Guest memory lives inside the Launcher, so it sets 310 * this to ensure the Guest can only reach its own memory. 311 * 312 * start: The first instruction to execute ("eip" in x86-speak). 313 */ 314static int initialize(struct file *file, const unsigned long __user *input) 315{ 316 /* "struct lguest" contains all we (the Host) know about a Guest. */ 317 struct lguest *lg; 318 int err; 319 unsigned long args[3]; 320 321 /* 322 * We grab the Big Lguest lock, which protects against multiple 323 * simultaneous initializations. 324 */ 325 mutex_lock(&lguest_lock); 326 /* You can't initialize twice! Close the device and start again... */ 327 if (file->private_data) { 328 err = -EBUSY; 329 goto unlock; 330 } 331 332 if (copy_from_user(args, input, sizeof(args)) != 0) { 333 err = -EFAULT; 334 goto unlock; 335 } 336 337 lg = kzalloc(sizeof(*lg), GFP_KERNEL); 338 if (!lg) { 339 err = -ENOMEM; 340 goto unlock; 341 } 342 343 lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); 344 if (!lg->eventfds) { 345 err = -ENOMEM; 346 goto free_lg; 347 } 348 lg->eventfds->num = 0; 349 350 /* Populate the easy fields of our "struct lguest" */ 351 lg->mem_base = (void __user *)args[0]; 352 lg->pfn_limit = args[1]; 353 354 /* This is the first cpu (cpu 0) and it will start booting at args[2] */ 355 err = lg_cpu_start(&lg->cpus[0], 0, args[2]); 356 if (err) 357 goto free_eventfds; 358 359 /* 360 * Initialize the Guest's shadow page tables, using the toplevel 361 * address the Launcher gave us. This allocates memory, so can fail. 362 */ 363 err = init_guest_pagetable(lg); 364 if (err) 365 goto free_regs; 366 367 /* We keep our "struct lguest" in the file's private_data. */ 368 file->private_data = lg; 369 370 mutex_unlock(&lguest_lock); 371 372 /* And because this is a write() call, we return the length used. */ 373 return sizeof(args); 374 375free_regs: 376 /* FIXME: This should be in free_vcpu */ 377 free_page(lg->cpus[0].regs_page); 378free_eventfds: 379 kfree(lg->eventfds); 380free_lg: 381 kfree(lg); 382unlock: 383 mutex_unlock(&lguest_lock); 384 return err; 385} 386 387/*L:010 388 * The first operation the Launcher does must be a write. All writes 389 * start with an unsigned long number: for the first write this must be 390 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 391 * writes of other values to send interrupts or set up receipt of notifications. 392 * 393 * Note that we overload the "offset" in the /dev/lguest file to indicate what 394 * CPU number we're dealing with. Currently this is always 0 since we only 395 * support uniprocessor Guests, but you can see the beginnings of SMP support 396 * here. 397 */ 398static ssize_t write(struct file *file, const char __user *in, 399 size_t size, loff_t *off) 400{ 401 /* 402 * Once the Guest is initialized, we hold the "struct lguest" in the 403 * file private data. 404 */ 405 struct lguest *lg = file->private_data; 406 const unsigned long __user *input = (const unsigned long __user *)in; 407 unsigned long req; 408 struct lg_cpu *uninitialized_var(cpu); 409 unsigned int cpu_id = *off; 410 411 /* The first value tells us what this request is. */ 412 if (get_user(req, input) != 0) 413 return -EFAULT; 414 input++; 415 416 /* If you haven't initialized, you must do that first. */ 417 if (req != LHREQ_INITIALIZE) { 418 if (!lg || (cpu_id >= lg->nr_cpus)) 419 return -EINVAL; 420 cpu = &lg->cpus[cpu_id]; 421 422 /* Once the Guest is dead, you can only read() why it died. */ 423 if (lg->dead) 424 return -ENOENT; 425 } 426 427 switch (req) { 428 case LHREQ_INITIALIZE: 429 return initialize(file, input); 430 case LHREQ_IRQ: 431 return user_send_irq(cpu, input); 432 case LHREQ_EVENTFD: 433 return attach_eventfd(lg, input); 434 default: 435 return -EINVAL; 436 } 437} 438 439/*L:060 440 * The final piece of interface code is the close() routine. It reverses 441 * everything done in initialize(). This is usually called because the 442 * Launcher exited. 443 * 444 * Note that the close routine returns 0 or a negative error number: it can't 445 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 446 * letting them do it. 447:*/ 448static int close(struct inode *inode, struct file *file) 449{ 450 struct lguest *lg = file->private_data; 451 unsigned int i; 452 453 /* If we never successfully initialized, there's nothing to clean up */ 454 if (!lg) 455 return 0; 456 457 /* 458 * We need the big lock, to protect from inter-guest I/O and other 459 * Launchers initializing guests. 460 */ 461 mutex_lock(&lguest_lock); 462 463 /* Free up the shadow page tables for the Guest. */ 464 free_guest_pagetable(lg); 465 466 for (i = 0; i < lg->nr_cpus; i++) { 467 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 468 hrtimer_cancel(&lg->cpus[i].hrt); 469 /* We can free up the register page we allocated. */ 470 free_page(lg->cpus[i].regs_page); 471 /* 472 * Now all the memory cleanups are done, it's safe to release 473 * the Launcher's memory management structure. 474 */ 475 mmput(lg->cpus[i].mm); 476 } 477 478 /* Release any eventfds they registered. */ 479 for (i = 0; i < lg->eventfds->num; i++) 480 eventfd_ctx_put(lg->eventfds->map[i].event); 481 kfree(lg->eventfds); 482 483 /* 484 * If lg->dead doesn't contain an error code it will be NULL or a 485 * kmalloc()ed string, either of which is ok to hand to kfree(). 486 */ 487 if (!IS_ERR(lg->dead)) 488 kfree(lg->dead); 489 /* Free the memory allocated to the lguest_struct */ 490 kfree(lg); 491 /* Release lock and exit. */ 492 mutex_unlock(&lguest_lock); 493 494 return 0; 495} 496 497/*L:000 498 * Welcome to our journey through the Launcher! 499 * 500 * The Launcher is the Host userspace program which sets up, runs and services 501 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 502 * doing things are inaccurate: the Launcher does all the device handling for 503 * the Guest, but the Guest can't know that. 504 * 505 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 506 * shall see more of that later. 507 * 508 * We begin our understanding with the Host kernel interface which the Launcher 509 * uses: reading and writing a character device called /dev/lguest. All the 510 * work happens in the read(), write() and close() routines: 511 */ 512static const struct file_operations lguest_fops = { 513 .owner = THIS_MODULE, 514 .release = close, 515 .write = write, 516 .read = read, 517 .llseek = default_llseek, 518}; 519 520/* 521 * This is a textbook example of a "misc" character device. Populate a "struct 522 * miscdevice" and register it with misc_register(). 523 */ 524static struct miscdevice lguest_dev = { 525 .minor = MISC_DYNAMIC_MINOR, 526 .name = "lguest", 527 .fops = &lguest_fops, 528}; 529 530int __init lguest_device_init(void) 531{ 532 return misc_register(&lguest_dev); 533} 534 535void __exit lguest_device_remove(void) 536{ 537 misc_deregister(&lguest_dev); 538} 539