drivers/lguest/lguest_user.c

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