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