lguest_user.c revision df60aeef4f4fe0645d9a195a7689005520422de5
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, pagetable, entry point and kernel address 4 * offset. A read will run the Guest until something happens, such as a signal 5 * or the Guest doing a NOTIFY out to the Launcher. :*/ 6#include <linux/uaccess.h> 7#include <linux/miscdevice.h> 8#include <linux/fs.h> 9#include <linux/sched.h> 10#include <linux/eventfd.h> 11#include <linux/file.h> 12#include "lg.h" 13 14/*L:055 When something happens, the Waker process needs a way to stop the 15 * kernel running the Guest and return to the Launcher. So the Waker writes 16 * LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher 17 * has done whatever needs attention, it writes LHREQ_BREAK and "0" to release 18 * the Waker. */ 19static int break_guest_out(struct lg_cpu *cpu, const unsigned long __user*input) 20{ 21 unsigned long on; 22 23 /* Fetch whether they're turning break on or off. */ 24 if (get_user(on, input) != 0) 25 return -EFAULT; 26 27 if (on) { 28 cpu->break_out = 1; 29 if (!wake_up_process(cpu->tsk)) 30 kick_process(cpu->tsk); 31 /* Wait for them to reset it */ 32 return wait_event_interruptible(cpu->break_wq, !cpu->break_out); 33 } else { 34 cpu->break_out = 0; 35 wake_up(&cpu->break_wq); 36 return 0; 37 } 38} 39 40bool send_notify_to_eventfd(struct lg_cpu *cpu) 41{ 42 unsigned int i; 43 struct lg_eventfd_map *map; 44 45 /* lg->eventfds is RCU-protected */ 46 rcu_read_lock(); 47 map = rcu_dereference(cpu->lg->eventfds); 48 for (i = 0; i < map->num; i++) { 49 if (map->map[i].addr == cpu->pending_notify) { 50 eventfd_signal(map->map[i].event, 1); 51 cpu->pending_notify = 0; 52 break; 53 } 54 } 55 rcu_read_unlock(); 56 return cpu->pending_notify == 0; 57} 58 59static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) 60{ 61 struct lg_eventfd_map *new, *old = lg->eventfds; 62 63 if (!addr) 64 return -EINVAL; 65 66 /* Replace the old array with the new one, carefully: others can 67 * be accessing it at the same time */ 68 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), 69 GFP_KERNEL); 70 if (!new) 71 return -ENOMEM; 72 73 /* First make identical copy. */ 74 memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); 75 new->num = old->num; 76 77 /* Now append new entry. */ 78 new->map[new->num].addr = addr; 79 new->map[new->num].event = eventfd_fget(fd); 80 if (IS_ERR(new->map[new->num].event)) { 81 kfree(new); 82 return PTR_ERR(new->map[new->num].event); 83 } 84 new->num++; 85 86 /* Now put new one in place. */ 87 rcu_assign_pointer(lg->eventfds, new); 88 89 /* We're not in a big hurry. Wait until noone's looking at old 90 * version, then delete it. */ 91 synchronize_rcu(); 92 kfree(old); 93 94 return 0; 95} 96 97static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) 98{ 99 unsigned long addr, fd; 100 int err; 101 102 if (get_user(addr, input) != 0) 103 return -EFAULT; 104 input++; 105 if (get_user(fd, input) != 0) 106 return -EFAULT; 107 108 mutex_lock(&lguest_lock); 109 err = add_eventfd(lg, addr, fd); 110 mutex_unlock(&lguest_lock); 111 112 return 0; 113} 114 115/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt 116 * number to /dev/lguest. */ 117static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) 118{ 119 unsigned long irq; 120 121 if (get_user(irq, input) != 0) 122 return -EFAULT; 123 if (irq >= LGUEST_IRQS) 124 return -EINVAL; 125 126 set_interrupt(cpu, irq); 127 return 0; 128} 129 130/*L:040 Once our Guest is initialized, the Launcher makes it run by reading 131 * from /dev/lguest. */ 132static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) 133{ 134 struct lguest *lg = file->private_data; 135 struct lg_cpu *cpu; 136 unsigned int cpu_id = *o; 137 138 /* You must write LHREQ_INITIALIZE first! */ 139 if (!lg) 140 return -EINVAL; 141 142 /* Watch out for arbitrary vcpu indexes! */ 143 if (cpu_id >= lg->nr_cpus) 144 return -EINVAL; 145 146 cpu = &lg->cpus[cpu_id]; 147 148 /* If you're not the task which owns the Guest, go away. */ 149 if (current != cpu->tsk) 150 return -EPERM; 151 152 /* If the Guest is already dead, we indicate why */ 153 if (lg->dead) { 154 size_t len; 155 156 /* lg->dead either contains an error code, or a string. */ 157 if (IS_ERR(lg->dead)) 158 return PTR_ERR(lg->dead); 159 160 /* We can only return as much as the buffer they read with. */ 161 len = min(size, strlen(lg->dead)+1); 162 if (copy_to_user(user, lg->dead, len) != 0) 163 return -EFAULT; 164 return len; 165 } 166 167 /* If we returned from read() last time because the Guest sent I/O, 168 * clear the flag. */ 169 if (cpu->pending_notify) 170 cpu->pending_notify = 0; 171 172 /* Run the Guest until something interesting happens. */ 173 return run_guest(cpu, (unsigned long __user *)user); 174} 175 176/*L:025 This actually initializes a CPU. For the moment, a Guest is only 177 * uniprocessor, so "id" is always 0. */ 178static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) 179{ 180 /* We have a limited number the number of CPUs in the lguest struct. */ 181 if (id >= ARRAY_SIZE(cpu->lg->cpus)) 182 return -EINVAL; 183 184 /* Set up this CPU's id, and pointer back to the lguest struct. */ 185 cpu->id = id; 186 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); 187 cpu->lg->nr_cpus++; 188 189 /* Each CPU has a timer it can set. */ 190 init_clockdev(cpu); 191 192 /* We need a complete page for the Guest registers: they are accessible 193 * to the Guest and we can only grant it access to whole pages. */ 194 cpu->regs_page = get_zeroed_page(GFP_KERNEL); 195 if (!cpu->regs_page) 196 return -ENOMEM; 197 198 /* We actually put the registers at the bottom of the page. */ 199 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); 200 201 /* Now we initialize the Guest's registers, handing it the start 202 * address. */ 203 lguest_arch_setup_regs(cpu, start_ip); 204 205 /* Initialize the queue for the Waker to wait on */ 206 init_waitqueue_head(&cpu->break_wq); 207 208 /* We keep a pointer to the Launcher task (ie. current task) for when 209 * other Guests want to wake this one (eg. console input). */ 210 cpu->tsk = current; 211 212 /* We need to keep a pointer to the Launcher's memory map, because if 213 * the Launcher dies we need to clean it up. If we don't keep a 214 * reference, it is destroyed before close() is called. */ 215 cpu->mm = get_task_mm(cpu->tsk); 216 217 /* We remember which CPU's pages this Guest used last, for optimization 218 * when the same Guest runs on the same CPU twice. */ 219 cpu->last_pages = NULL; 220 221 /* No error == success. */ 222 return 0; 223} 224 225/*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit) 226 * values (in addition to the LHREQ_INITIALIZE value). These are: 227 * 228 * base: The start of the Guest-physical memory inside the Launcher memory. 229 * 230 * pfnlimit: The highest (Guest-physical) page number the Guest should be 231 * allowed to access. The Guest memory lives inside the Launcher, so it sets 232 * this to ensure the Guest can only reach its own memory. 233 * 234 * start: The first instruction to execute ("eip" in x86-speak). 235 */ 236static int initialize(struct file *file, const unsigned long __user *input) 237{ 238 /* "struct lguest" contains everything we (the Host) know about a 239 * Guest. */ 240 struct lguest *lg; 241 int err; 242 unsigned long args[3]; 243 244 /* We grab the Big Lguest lock, which protects against multiple 245 * simultaneous initializations. */ 246 mutex_lock(&lguest_lock); 247 /* You can't initialize twice! Close the device and start again... */ 248 if (file->private_data) { 249 err = -EBUSY; 250 goto unlock; 251 } 252 253 if (copy_from_user(args, input, sizeof(args)) != 0) { 254 err = -EFAULT; 255 goto unlock; 256 } 257 258 lg = kzalloc(sizeof(*lg), GFP_KERNEL); 259 if (!lg) { 260 err = -ENOMEM; 261 goto unlock; 262 } 263 264 lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); 265 if (!lg->eventfds) { 266 err = -ENOMEM; 267 goto free_lg; 268 } 269 lg->eventfds->num = 0; 270 271 /* Populate the easy fields of our "struct lguest" */ 272 lg->mem_base = (void __user *)args[0]; 273 lg->pfn_limit = args[1]; 274 275 /* This is the first cpu (cpu 0) and it will start booting at args[2] */ 276 err = lg_cpu_start(&lg->cpus[0], 0, args[2]); 277 if (err) 278 goto free_eventfds; 279 280 /* Initialize the Guest's shadow page tables, using the toplevel 281 * address the Launcher gave us. This allocates memory, so can fail. */ 282 err = init_guest_pagetable(lg); 283 if (err) 284 goto free_regs; 285 286 /* We keep our "struct lguest" in the file's private_data. */ 287 file->private_data = lg; 288 289 mutex_unlock(&lguest_lock); 290 291 /* And because this is a write() call, we return the length used. */ 292 return sizeof(args); 293 294free_regs: 295 /* FIXME: This should be in free_vcpu */ 296 free_page(lg->cpus[0].regs_page); 297free_eventfds: 298 kfree(lg->eventfds); 299free_lg: 300 kfree(lg); 301unlock: 302 mutex_unlock(&lguest_lock); 303 return err; 304} 305 306/*L:010 The first operation the Launcher does must be a write. All writes 307 * start with an unsigned long number: for the first write this must be 308 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 309 * writes of other values to send interrupts. 310 * 311 * Note that we overload the "offset" in the /dev/lguest file to indicate what 312 * CPU number we're dealing with. Currently this is always 0, since we only 313 * support uniprocessor Guests, but you can see the beginnings of SMP support 314 * here. */ 315static ssize_t write(struct file *file, const char __user *in, 316 size_t size, loff_t *off) 317{ 318 /* Once the Guest is initialized, we hold the "struct lguest" in the 319 * file private data. */ 320 struct lguest *lg = file->private_data; 321 const unsigned long __user *input = (const unsigned long __user *)in; 322 unsigned long req; 323 struct lg_cpu *uninitialized_var(cpu); 324 unsigned int cpu_id = *off; 325 326 /* The first value tells us what this request is. */ 327 if (get_user(req, input) != 0) 328 return -EFAULT; 329 input++; 330 331 /* If you haven't initialized, you must do that first. */ 332 if (req != LHREQ_INITIALIZE) { 333 if (!lg || (cpu_id >= lg->nr_cpus)) 334 return -EINVAL; 335 cpu = &lg->cpus[cpu_id]; 336 337 /* Once the Guest is dead, you can only read() why it died. */ 338 if (lg->dead) 339 return -ENOENT; 340 } 341 342 switch (req) { 343 case LHREQ_INITIALIZE: 344 return initialize(file, input); 345 case LHREQ_IRQ: 346 return user_send_irq(cpu, input); 347 case LHREQ_BREAK: 348 return break_guest_out(cpu, input); 349 case LHREQ_EVENTFD: 350 return attach_eventfd(lg, input); 351 default: 352 return -EINVAL; 353 } 354} 355 356/*L:060 The final piece of interface code is the close() routine. It reverses 357 * everything done in initialize(). This is usually called because the 358 * Launcher exited. 359 * 360 * Note that the close routine returns 0 or a negative error number: it can't 361 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 362 * letting them do it. :*/ 363static int close(struct inode *inode, struct file *file) 364{ 365 struct lguest *lg = file->private_data; 366 unsigned int i; 367 368 /* If we never successfully initialized, there's nothing to clean up */ 369 if (!lg) 370 return 0; 371 372 /* We need the big lock, to protect from inter-guest I/O and other 373 * Launchers initializing guests. */ 374 mutex_lock(&lguest_lock); 375 376 /* Free up the shadow page tables for the Guest. */ 377 free_guest_pagetable(lg); 378 379 for (i = 0; i < lg->nr_cpus; i++) { 380 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 381 hrtimer_cancel(&lg->cpus[i].hrt); 382 /* We can free up the register page we allocated. */ 383 free_page(lg->cpus[i].regs_page); 384 /* Now all the memory cleanups are done, it's safe to release 385 * the Launcher's memory management structure. */ 386 mmput(lg->cpus[i].mm); 387 } 388 389 /* Release any eventfds they registered. */ 390 for (i = 0; i < lg->eventfds->num; i++) 391 fput(lg->eventfds->map[i].event); 392 kfree(lg->eventfds); 393 394 /* If lg->dead doesn't contain an error code it will be NULL or a 395 * kmalloc()ed string, either of which is ok to hand to kfree(). */ 396 if (!IS_ERR(lg->dead)) 397 kfree(lg->dead); 398 /* Free the memory allocated to the lguest_struct */ 399 kfree(lg); 400 /* Release lock and exit. */ 401 mutex_unlock(&lguest_lock); 402 403 return 0; 404} 405 406/*L:000 407 * Welcome to our journey through the Launcher! 408 * 409 * The Launcher is the Host userspace program which sets up, runs and services 410 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 411 * doing things are inaccurate: the Launcher does all the device handling for 412 * the Guest, but the Guest can't know that. 413 * 414 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 415 * shall see more of that later. 416 * 417 * We begin our understanding with the Host kernel interface which the Launcher 418 * uses: reading and writing a character device called /dev/lguest. All the 419 * work happens in the read(), write() and close() routines: */ 420static struct file_operations lguest_fops = { 421 .owner = THIS_MODULE, 422 .release = close, 423 .write = write, 424 .read = read, 425}; 426 427/* This is a textbook example of a "misc" character device. Populate a "struct 428 * miscdevice" and register it with misc_register(). */ 429static struct miscdevice lguest_dev = { 430 .minor = MISC_DYNAMIC_MINOR, 431 .name = "lguest", 432 .fops = &lguest_fops, 433}; 434 435int __init lguest_device_init(void) 436{ 437 return misc_register(&lguest_dev); 438} 439 440void __exit lguest_device_remove(void) 441{ 442 misc_deregister(&lguest_dev); 443} 444