lguest_user.c revision a53a35a8b485b9c16b73e5177bddaa4321971199
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 lguest *lg, 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 lg->break_out = 1; 26 /* Pop it out of the Guest (may be running on different CPU) */ 27 wake_up_process(lg->tsk); 28 /* Wait for them to reset it */ 29 return wait_event_interruptible(lg->break_wq, !lg->break_out); 30 } else { 31 lg->break_out = 0; 32 wake_up(&lg->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 != lg->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 (lg->pending_notify) 93 lg->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 return 0; 123} 124 125/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit) 126 * values (in addition to the LHREQ_INITIALIZE value). These are: 127 * 128 * base: The start of the Guest-physical memory inside the Launcher memory. 129 * 130 * pfnlimit: The highest (Guest-physical) page number the Guest should be 131 * allowed to access. The Guest memory lives inside the Launcher, so it sets 132 * this to ensure the Guest can only reach its own memory. 133 * 134 * pgdir: The (Guest-physical) address of the top of the initial Guest 135 * pagetables (which are set up by the Launcher). 136 * 137 * start: The first instruction to execute ("eip" in x86-speak). 138 */ 139static int initialize(struct file *file, const unsigned long __user *input) 140{ 141 /* "struct lguest" contains everything we (the Host) know about a 142 * Guest. */ 143 struct lguest *lg; 144 int err; 145 unsigned long args[4]; 146 147 /* We grab the Big Lguest lock, which protects against multiple 148 * simultaneous initializations. */ 149 mutex_lock(&lguest_lock); 150 /* You can't initialize twice! Close the device and start again... */ 151 if (file->private_data) { 152 err = -EBUSY; 153 goto unlock; 154 } 155 156 if (copy_from_user(args, input, sizeof(args)) != 0) { 157 err = -EFAULT; 158 goto unlock; 159 } 160 161 lg = kzalloc(sizeof(*lg), GFP_KERNEL); 162 if (!lg) { 163 err = -ENOMEM; 164 goto unlock; 165 } 166 167 /* Populate the easy fields of our "struct lguest" */ 168 lg->mem_base = (void __user *)(long)args[0]; 169 lg->pfn_limit = args[1]; 170 171 /* This is the first cpu */ 172 err = lg_cpu_start(&lg->cpus[0], 0, args[3]); 173 if (err) 174 goto release_guest; 175 176 /* Initialize the Guest's shadow page tables, using the toplevel 177 * address the Launcher gave us. This allocates memory, so can 178 * fail. */ 179 err = init_guest_pagetable(lg, args[2]); 180 if (err) 181 goto free_regs; 182 183 /* We keep a pointer to the Launcher task (ie. current task) for when 184 * other Guests want to wake this one (inter-Guest I/O). */ 185 lg->tsk = current; 186 /* We need to keep a pointer to the Launcher's memory map, because if 187 * the Launcher dies we need to clean it up. If we don't keep a 188 * reference, it is destroyed before close() is called. */ 189 lg->mm = get_task_mm(lg->tsk); 190 191 /* Initialize the queue for the waker to wait on */ 192 init_waitqueue_head(&lg->break_wq); 193 194 /* We remember which CPU's pages this Guest used last, for optimization 195 * when the same Guest runs on the same CPU twice. */ 196 lg->last_pages = NULL; 197 198 /* We keep our "struct lguest" in the file's private_data. */ 199 file->private_data = lg; 200 201 mutex_unlock(&lguest_lock); 202 203 /* And because this is a write() call, we return the length used. */ 204 return sizeof(args); 205 206free_regs: 207 /* FIXME: This should be in free_vcpu */ 208 free_page(lg->cpus[0].regs_page); 209release_guest: 210 kfree(lg); 211unlock: 212 mutex_unlock(&lguest_lock); 213 return err; 214} 215 216/*L:010 The first operation the Launcher does must be a write. All writes 217 * start with an unsigned long number: for the first write this must be 218 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 219 * writes of other values to send interrupts. */ 220static ssize_t write(struct file *file, const char __user *in, 221 size_t size, loff_t *off) 222{ 223 /* Once the guest is initialized, we hold the "struct lguest" in the 224 * file private data. */ 225 struct lguest *lg = file->private_data; 226 const unsigned long __user *input = (const unsigned long __user *)in; 227 unsigned long req; 228 struct lg_cpu *uninitialized_var(cpu); 229 unsigned int cpu_id = *off; 230 231 if (get_user(req, input) != 0) 232 return -EFAULT; 233 input++; 234 235 /* If you haven't initialized, you must do that first. */ 236 if (req != LHREQ_INITIALIZE) { 237 if (!lg || (cpu_id >= lg->nr_cpus)) 238 return -EINVAL; 239 cpu = &lg->cpus[cpu_id]; 240 if (!cpu) 241 return -EINVAL; 242 } 243 244 /* Once the Guest is dead, all you can do is read() why it died. */ 245 if (lg && lg->dead) 246 return -ENOENT; 247 248 /* If you're not the task which owns the Guest, you can only break */ 249 if (lg && current != lg->tsk && req != LHREQ_BREAK) 250 return -EPERM; 251 252 switch (req) { 253 case LHREQ_INITIALIZE: 254 return initialize(file, input); 255 case LHREQ_IRQ: 256 return user_send_irq(cpu, input); 257 case LHREQ_BREAK: 258 return break_guest_out(lg, input); 259 default: 260 return -EINVAL; 261 } 262} 263 264/*L:060 The final piece of interface code is the close() routine. It reverses 265 * everything done in initialize(). This is usually called because the 266 * Launcher exited. 267 * 268 * Note that the close routine returns 0 or a negative error number: it can't 269 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 270 * letting them do it. :*/ 271static int close(struct inode *inode, struct file *file) 272{ 273 struct lguest *lg = file->private_data; 274 unsigned int i; 275 276 /* If we never successfully initialized, there's nothing to clean up */ 277 if (!lg) 278 return 0; 279 280 /* We need the big lock, to protect from inter-guest I/O and other 281 * Launchers initializing guests. */ 282 mutex_lock(&lguest_lock); 283 for (i = 0; i < lg->nr_cpus; i++) { 284 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */ 285 hrtimer_cancel(&lg->cpus[i].hrt); 286 /* We can free up the register page we allocated. */ 287 free_page(lg->cpus[i].regs_page); 288 } 289 /* Free up the shadow page tables for the Guest. */ 290 free_guest_pagetable(lg); 291 /* Now all the memory cleanups are done, it's safe to release the 292 * Launcher's memory management structure. */ 293 mmput(lg->mm); 294 /* If lg->dead doesn't contain an error code it will be NULL or a 295 * kmalloc()ed string, either of which is ok to hand to kfree(). */ 296 if (!IS_ERR(lg->dead)) 297 kfree(lg->dead); 298 /* We clear the entire structure, which also marks it as free for the 299 * next user. */ 300 memset(lg, 0, sizeof(*lg)); 301 /* Release lock and exit. */ 302 mutex_unlock(&lguest_lock); 303 304 return 0; 305} 306 307/*L:000 308 * Welcome to our journey through the Launcher! 309 * 310 * The Launcher is the Host userspace program which sets up, runs and services 311 * the Guest. In fact, many comments in the Drivers which refer to "the Host" 312 * doing things are inaccurate: the Launcher does all the device handling for 313 * the Guest, but the Guest can't know that. 314 * 315 * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we 316 * shall see more of that later. 317 * 318 * We begin our understanding with the Host kernel interface which the Launcher 319 * uses: reading and writing a character device called /dev/lguest. All the 320 * work happens in the read(), write() and close() routines: */ 321static struct file_operations lguest_fops = { 322 .owner = THIS_MODULE, 323 .release = close, 324 .write = write, 325 .read = read, 326}; 327 328/* This is a textbook example of a "misc" character device. Populate a "struct 329 * miscdevice" and register it with misc_register(). */ 330static struct miscdevice lguest_dev = { 331 .minor = MISC_DYNAMIC_MINOR, 332 .name = "lguest", 333 .fops = &lguest_fops, 334}; 335 336int __init lguest_device_init(void) 337{ 338 return misc_register(&lguest_dev); 339} 340 341void __exit lguest_device_remove(void) 342{ 343 misc_deregister(&lguest_dev); 344} 345