lguest_user.c revision 5e232f4f428c4266ba5cdae9f23ba19a0913dcf9
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 "lg.h" 10 11/*L:055 When something happens, the Waker process needs a way to stop the 12 * kernel running the Guest and return to the Launcher. So the Waker writes 13 * LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher 14 * has done whatever needs attention, it writes LHREQ_BREAK and "0" to release 15 * the Waker. */ 16static int break_guest_out(struct lg_cpu *cpu, const unsigned long __user*input) 17{ 18 unsigned long on; 19 20 /* Fetch whether they're turning break on or off. */ 21 if (get_user(on, input) != 0) 22 return -EFAULT; 23 24 if (on) { 25 cpu->break_out = 1; 26 /* Pop it out of the Guest (may be running on different CPU) */ 27 wake_up_process(cpu->tsk); 28 /* Wait for them to reset it */ 29 return wait_event_interruptible(cpu->break_wq, !cpu->break_out); 30 } else { 31 cpu->break_out = 0; 32 wake_up(&cpu->break_wq); 33 return 0; 34 } 35} 36 37/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt 38 * number to /dev/lguest. */ 39static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) 40{ 41 unsigned long irq; 42 43 if (get_user(irq, input) != 0) 44 return -EFAULT; 45 if (irq >= LGUEST_IRQS) 46 return -EINVAL; 47 /* Next time the Guest runs, the core code will see if it can deliver 48 * this interrupt. */ 49 set_bit(irq, cpu->irqs_pending); 50 return 0; 51} 52 53/*L:040 Once our Guest is initialized, the Launcher makes it run by reading 54 * from /dev/lguest. */ 55static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) 56{ 57 struct lguest *lg = file->private_data; 58 struct lg_cpu *cpu; 59 unsigned int cpu_id = *o; 60 61 /* You must write LHREQ_INITIALIZE first! */ 62 if (!lg) 63 return -EINVAL; 64 65 /* Watch out for arbitrary vcpu indexes! */ 66 if (cpu_id >= lg->nr_cpus) 67 return -EINVAL; 68 69 cpu = &lg->cpus[cpu_id]; 70 71 /* If you're not the task which owns the Guest, go away. */ 72 if (current != cpu->tsk) 73 return -EPERM; 74 75 /* If the guest is already dead, we indicate why */ 76 if (lg->dead) { 77 size_t len; 78 79 /* lg->dead either contains an error code, or a string. */ 80 if (IS_ERR(lg->dead)) 81 return PTR_ERR(lg->dead); 82 83 /* We can only return as much as the buffer they read with. */ 84 len = min(size, strlen(lg->dead)+1); 85 if (copy_to_user(user, lg->dead, len) != 0) 86 return -EFAULT; 87 return len; 88 } 89 90 /* If we returned from read() last time because the Guest notified, 91 * clear the flag. */ 92 if (cpu->pending_notify) 93 cpu->pending_notify = 0; 94 95 /* Run the Guest until something interesting happens. */ 96 return run_guest(cpu, (unsigned long __user *)user); 97} 98 99static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) 100{ 101 if (id >= NR_CPUS) 102 return -EINVAL; 103 104 cpu->id = id; 105 cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); 106 cpu->lg->nr_cpus++; 107 init_clockdev(cpu); 108 109 /* We need a complete page for the Guest registers: they are accessible 110 * to the Guest and we can only grant it access to whole pages. */ 111 cpu->regs_page = get_zeroed_page(GFP_KERNEL); 112 if (!cpu->regs_page) 113 return -ENOMEM; 114 115 /* We actually put the registers at the bottom of the page. */ 116 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); 117 118 /* Now we initialize the Guest's registers, handing it the start 119 * address. */ 120 lguest_arch_setup_regs(cpu, start_ip); 121 122 /* Initialize the queue for the waker to wait on */ 123 init_waitqueue_head(&cpu->break_wq); 124 125 /* We keep a pointer to the Launcher task (ie. current task) for when 126 * other Guests want to wake this one (inter-Guest I/O). */ 127 cpu->tsk = current; 128 129 /* We need to keep a pointer to the Launcher's memory map, because if 130 * the Launcher dies we need to clean it up. If we don't keep a 131 * reference, it is destroyed before close() is called. */ 132 cpu->mm = get_task_mm(cpu->tsk); 133 134 return 0; 135} 136 137/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit) 138 * values (in addition to the LHREQ_INITIALIZE value). These are: 139 * 140 * base: The start of the Guest-physical memory inside the Launcher memory. 141 * 142 * pfnlimit: The highest (Guest-physical) page number the Guest should be 143 * allowed to access. The Guest memory lives inside the Launcher, so it sets 144 * this to ensure the Guest can only reach its own memory. 145 * 146 * pgdir: The (Guest-physical) address of the top of the initial Guest 147 * pagetables (which are set up by the Launcher). 148 * 149 * start: The first instruction to execute ("eip" in x86-speak). 150 */ 151static int initialize(struct file *file, const unsigned long __user *input) 152{ 153 /* "struct lguest" contains everything we (the Host) know about a 154 * Guest. */ 155 struct lguest *lg; 156 int err; 157 unsigned long args[4]; 158 159 /* We grab the Big Lguest lock, which protects against multiple 160 * simultaneous initializations. */ 161 mutex_lock(&lguest_lock); 162 /* You can't initialize twice! Close the device and start again... */ 163 if (file->private_data) { 164 err = -EBUSY; 165 goto unlock; 166 } 167 168 if (copy_from_user(args, input, sizeof(args)) != 0) { 169 err = -EFAULT; 170 goto unlock; 171 } 172 173 lg = kzalloc(sizeof(*lg), GFP_KERNEL); 174 if (!lg) { 175 err = -ENOMEM; 176 goto unlock; 177 } 178 179 /* Populate the easy fields of our "struct lguest" */ 180 lg->mem_base = (void __user *)(long)args[0]; 181 lg->pfn_limit = args[1]; 182 183 /* This is the first cpu */ 184 err = lg_cpu_start(&lg->cpus[0], 0, args[3]); 185 if (err) 186 goto release_guest; 187 188 /* Initialize the Guest's shadow page tables, using the toplevel 189 * address the Launcher gave us. This allocates memory, so can 190 * fail. */ 191 err = init_guest_pagetable(lg, args[2]); 192 if (err) 193 goto free_regs; 194 195 /* We remember which CPU's pages this Guest used last, for optimization 196 * when the same Guest runs on the same CPU twice. */ 197 lg->last_pages = NULL; 198 199 /* We keep our "struct lguest" in the file's private_data. */ 200 file->private_data = lg; 201 202 mutex_unlock(&lguest_lock); 203 204 /* And because this is a write() call, we return the length used. */ 205 return sizeof(args); 206 207free_regs: 208 /* FIXME: This should be in free_vcpu */ 209 free_page(lg->cpus[0].regs_page); 210release_guest: 211 kfree(lg); 212unlock: 213 mutex_unlock(&lguest_lock); 214 return err; 215} 216 217/*L:010 The first operation the Launcher does must be a write. All writes 218 * start with an unsigned long number: for the first write this must be 219 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 220 * writes of other values to send interrupts. */ 221static ssize_t write(struct file *file, const char __user *in, 222 size_t size, loff_t *off) 223{ 224 /* Once the guest is initialized, we hold the "struct lguest" in the 225 * file private data. */ 226 struct lguest *lg = file->private_data; 227 const unsigned long __user *input = (const unsigned long __user *)in; 228 unsigned long req; 229 struct lg_cpu *uninitialized_var(cpu); 230 unsigned int cpu_id = *off; 231 232 if (get_user(req, input) != 0) 233 return -EFAULT; 234 input++; 235 236 /* If you haven't initialized, you must do that first. */ 237 if (req != LHREQ_INITIALIZE) { 238 if (!lg || (cpu_id >= lg->nr_cpus)) 239 return -EINVAL; 240 cpu = &lg->cpus[cpu_id]; 241 if (!cpu) 242 return -EINVAL; 243 } 244 245 /* Once the Guest is dead, all you can do is read() why it died. */ 246 if (lg && lg->dead) 247 return -ENOENT; 248 249 /* If you're not the task which owns the Guest, you can only break */ 250 if (lg && current != cpu->tsk && req != LHREQ_BREAK) 251 return -EPERM; 252 253 switch (req) { 254 case LHREQ_INITIALIZE: 255 return initialize(file, input); 256 case LHREQ_IRQ: 257 return user_send_irq(cpu, input); 258 case LHREQ_BREAK: 259 return break_guest_out(cpu, input); 260 default: 261 return -EINVAL; 262 } 263} 264 265/*L:060 The final piece of interface code is the close() routine. It reverses 266 * everything done in initialize(). This is usually called because the 267 * Launcher exited. 268 * 269 * Note that the close routine returns 0 or a negative error number: it can't 270 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 271 * letting them do it. :*/ 272static int close(struct inode *inode, struct file *file) 273{ 274 struct lguest *lg = file->private_data; 275 unsigned int i; 276 277 /* If we never successfully initialized, there's nothing to clean up */ 278 if (!lg) 279 return 0; 280 281 /* We need the big lock, to protect from inter-guest I/O and other 282 * Launchers initializing guests. */ 283 mutex_lock(&lguest_lock); 284 285 /* Free up the shadow page tables for the Guest. */ 286 free_guest_pagetable(lg); 287 288 for (i = 0; i < lg->nr_cpus; i++) { 289 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 290 hrtimer_cancel(&lg->cpus[i].hrt); 291 /* We can free up the register page we allocated. */ 292 free_page(lg->cpus[i].regs_page); 293 /* Now all the memory cleanups are done, it's safe to release 294 * the Launcher's memory management structure. */ 295 mmput(lg->cpus[i].mm); 296 } 297 /* If lg->dead doesn't contain an error code it will be NULL or a 298 * kmalloc()ed string, either of which is ok to hand to kfree(). */ 299 if (!IS_ERR(lg->dead)) 300 kfree(lg->dead); 301 /* We clear the entire structure, which also marks it as free for the 302 * next user. */ 303 memset(lg, 0, sizeof(*lg)); 304 /* Release lock and exit. */ 305 mutex_unlock(&lguest_lock); 306 307 return 0; 308} 309 310/*L:000 311 * Welcome to our journey through the Launcher! 312 * 313 * The Launcher is the Host userspace program which sets up, runs and services 314 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 315 * doing things are inaccurate: the Launcher does all the device handling for 316 * the Guest, but the Guest can't know that. 317 * 318 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 319 * shall see more of that later. 320 * 321 * We begin our understanding with the Host kernel interface which the Launcher 322 * uses: reading and writing a character device called /dev/lguest. All the 323 * work happens in the read(), write() and close() routines: */ 324static struct file_operations lguest_fops = { 325 .owner = THIS_MODULE, 326 .release = close, 327 .write = write, 328 .read = read, 329}; 330 331/* This is a textbook example of a "misc" character device. Populate a "struct 332 * miscdevice" and register it with misc_register(). */ 333static struct miscdevice lguest_dev = { 334 .minor = MISC_DYNAMIC_MINOR, 335 .name = "lguest", 336 .fops = &lguest_fops, 337}; 338 339int __init lguest_device_init(void) 340{ 341 return misc_register(&lguest_dev); 342} 343 344void __exit lguest_device_remove(void) 345{ 346 misc_deregister(&lguest_dev); 347} 348