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                 if (!wake_up_process(cpu->tsk))
  28                         kick_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
  49         set_interrupt(cpu, irq);
  50         return 0;
  51 }
  52
  53 /*L:040 Once our Guest is initialized, the Launcher makes it run by reading
  54  * from /dev/lguest. */
  55 static 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 sent I/O,
  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
  99 /*L:025 This actually initializes a CPU.  For the moment, a Guest is only
 100  * uniprocessor, so "id" is always 0. */
 101 static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
 102 {
 103         /* We have a limited number the number of CPUs in the lguest struct. */
 104         if (id >= ARRAY_SIZE(cpu->lg->cpus))
 105                 return -EINVAL;
 106
 107         /* Set up this CPU's id, and pointer back to the lguest struct. */
 108         cpu->id = id;
 109         cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
 110         cpu->lg->nr_cpus++;
 111
 112         /* Each CPU has a timer it can set. */
 113         init_clockdev(cpu);
 114
 115         /* We need a complete page for the Guest registers: they are accessible
 116          * to the Guest and we can only grant it access to whole pages. */
 117         cpu->regs_page = get_zeroed_page(GFP_KERNEL);
 118         if (!cpu->regs_page)
 119                 return -ENOMEM;
 120
 121         /* We actually put the registers at the bottom of the page. */
 122         cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
 123
 124         /* Now we initialize the Guest's registers, handing it the start
 125          * address. */
 126         lguest_arch_setup_regs(cpu, start_ip);
 127
 128         /* Initialize the queue for the Waker to wait on */
 129         init_waitqueue_head(&cpu->break_wq);
 130
 131         /* We keep a pointer to the Launcher task (ie. current task) for when
 132          * other Guests want to wake this one (eg. console input). */
 133         cpu->tsk = current;
 134
 135         /* We need to keep a pointer to the Launcher's memory map, because if
 136          * the Launcher dies we need to clean it up.  If we don't keep a
 137          * reference, it is destroyed before close() is called. */
 138         cpu->mm = get_task_mm(cpu->tsk);
 139
 140         /* We remember which CPU's pages this Guest used last, for optimization
 141          * when the same Guest runs on the same CPU twice. */
 142         cpu->last_pages = NULL;
 143
 144         /* No error == success. */
 145         return 0;
 146 }
 147
 148 /*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit)
 149  * values (in addition to the LHREQ_INITIALIZE value).  These are:
 150  *
 151  * base: The start of the Guest-physical memory inside the Launcher memory.
 152  *
 153  * pfnlimit: The highest (Guest-physical) page number the Guest should be
 154  * allowed to access.  The Guest memory lives inside the Launcher, so it sets
 155  * this to ensure the Guest can only reach its own memory.
 156  *
 157  * start: The first instruction to execute ("eip" in x86-speak).
 158  */
 159 static int initialize(struct file *file, const unsigned long __user *input)
 160 {
 161         /* "struct lguest" contains everything we (the Host) know about a
 162          * Guest. */
 163         struct lguest *lg;
 164         int err;
 165         unsigned long args[3];
 166
 167         /* We grab the Big Lguest lock, which protects against multiple
 168          * simultaneous initializations. */
 169         mutex_lock(&lguest_lock);
 170         /* You can't initialize twice!  Close the device and start again... */
 171         if (file->private_data) {
 172                 err = -EBUSY;
 173                 goto unlock;
 174         }
 175
 176         if (copy_from_user(args, input, sizeof(args)) != 0) {
 177                 err = -EFAULT;
 178                 goto unlock;
 179         }
 180
 181         lg = kzalloc(sizeof(*lg), GFP_KERNEL);
 182         if (!lg) {
 183                 err = -ENOMEM;
 184                 goto unlock;
 185         }
 186
 187         /* Populate the easy fields of our "struct lguest" */
 188         lg->mem_base = (void __user *)args[0];
 189         lg->pfn_limit = args[1];
 190
 191         /* This is the first cpu (cpu 0) and it will start booting at args[2] */
 192         err = lg_cpu_start(&lg->cpus[0], 0, args[2]);
 193         if (err)
 194                 goto release_guest;
 195
 196         /* Initialize the Guest's shadow page tables, using the toplevel
 197          * address the Launcher gave us.  This allocates memory, so can fail. */
 198         err = init_guest_pagetable(lg);
 199         if (err)
 200                 goto free_regs;
 201
 202         /* We keep our "struct lguest" in the file's private_data. */
 203         file->private_data = lg;
 204
 205         mutex_unlock(&lguest_lock);
 206
 207         /* And because this is a write() call, we return the length used. */
 208         return sizeof(args);
 209
 210 free_regs:
 211         /* FIXME: This should be in free_vcpu */
 212         free_page(lg->cpus[0].regs_page);
 213 release_guest:
 214         kfree(lg);
 215 unlock:
 216         mutex_unlock(&lguest_lock);
 217         return err;
 218 }
 219
 220 /*L:010 The first operation the Launcher does must be a write.  All writes
 221  * start with an unsigned long number: for the first write this must be
 222  * LHREQ_INITIALIZE to set up the Guest.  After that the Launcher can use
 223  * writes of other values to send interrupts.
 224  *
 225  * Note that we overload the "offset" in the /dev/lguest file to indicate what
 226  * CPU number we're dealing with.  Currently this is always 0, since we only
 227  * support uniprocessor Guests, but you can see the beginnings of SMP support
 228  * here. */
 229 static ssize_t write(struct file *file, const char __user *in,
 230                      size_t size, loff_t *off)
 231 {
 232         /* Once the Guest is initialized, we hold the "struct lguest" in the
 233          * file private data. */
 234         struct lguest *lg = file->private_data;
 235         const unsigned long __user *input = (const unsigned long __user *)in;
 236         unsigned long req;
 237         struct lg_cpu *uninitialized_var(cpu);
 238         unsigned int cpu_id = *off;
 239
 240         /* The first value tells us what this request is. */
 241         if (get_user(req, input) != 0)
 242                 return -EFAULT;
 243         input++;
 244
 245         /* If you haven't initialized, you must do that first. */
 246         if (req != LHREQ_INITIALIZE) {
 247                 if (!lg || (cpu_id >= lg->nr_cpus))
 248                         return -EINVAL;
 249                 cpu = &lg->cpus[cpu_id];
 250
 251                 /* Once the Guest is dead, you can only read() why it died. */
 252                 if (lg->dead)
 253                         return -ENOENT;
 254         }
 255
 256         switch (req) {
 257         case LHREQ_INITIALIZE:
 258                 return initialize(file, input);
 259         case LHREQ_IRQ:
 260                 return user_send_irq(cpu, input);
 261         case LHREQ_BREAK:
 262                 return break_guest_out(cpu, input);
 263         default:
 264                 return -EINVAL;
 265         }
 266 }
 267
 268 /*L:060 The final piece of interface code is the close() routine.  It reverses
 269  * everything done in initialize().  This is usually called because the
 270  * Launcher exited.
 271  *
 272  * Note that the close routine returns 0 or a negative error number: it can't
 273  * really fail, but it can whine.  I blame Sun for this wart, and K&R C for
 274  * letting them do it. :*/
 275 static int close(struct inode *inode, struct file *file)
 276 {
 277         struct lguest *lg = file->private_data;
 278         unsigned int i;
 279
 280         /* If we never successfully initialized, there's nothing to clean up */
 281         if (!lg)
 282                 return 0;
 283
 284         /* We need the big lock, to protect from inter-guest I/O and other
 285          * Launchers initializing guests. */
 286         mutex_lock(&lguest_lock);
 287
 288         /* Free up the shadow page tables for the Guest. */
 289         free_guest_pagetable(lg);
 290
 291         for (i = 0; i < lg->nr_cpus; i++) {
 292                 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
 293                 hrtimer_cancel(&lg->cpus[i].hrt);
 294                 /* We can free up the register page we allocated. */
 295                 free_page(lg->cpus[i].regs_page);
 296                 /* Now all the memory cleanups are done, it's safe to release
 297                  * the Launcher's memory management structure. */
 298                 mmput(lg->cpus[i].mm);
 299         }
 300         /* If lg->dead doesn't contain an error code it will be NULL or a
 301          * kmalloc()ed string, either of which is ok to hand to kfree(). */
 302         if (!IS_ERR(lg->dead))
 303                 kfree(lg->dead);
 304         /* Free the memory allocated to the lguest_struct */
 305         kfree(lg);
 306         /* Release lock and exit. */
 307         mutex_unlock(&lguest_lock);
 308
 309         return 0;
 310 }
 311
 312 /*L:000
 313  * Welcome to our journey through the Launcher!
 314  *
 315  * The Launcher is the Host userspace program which sets up, runs and services
 316  * the Guest.  In fact, many comments in the Drivers which refer to "the Host"
 317  * doing things are inaccurate: the Launcher does all the device handling for
 318  * the Guest, but the Guest can't know that.
 319  *
 320  * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
 321  * shall see more of that later.
 322  *
 323  * We begin our understanding with the Host kernel interface which the Launcher
 324  * uses: reading and writing a character device called /dev/lguest.  All the
 325  * work happens in the read(), write() and close() routines: */
 326 static struct file_operations lguest_fops = {
 327         .owner   = THIS_MODULE,
 328         .release = close,
 329         .write   = write,
 330         .read    = read,
 331 };
 332
 333 /* This is a textbook example of a "misc" character device.  Populate a "struct
 334  * miscdevice" and register it with misc_register(). */
 335 static struct miscdevice lguest_dev = {
 336         .minor  = MISC_DYNAMIC_MINOR,
 337         .name   = "lguest",
 338         .fops   = &lguest_fops,
 339 };
 340
 341 int __init lguest_device_init(void)
 342 {
 343         return misc_register(&lguest_dev);
 344 }
 345
 346 void __exit lguest_device_remove(void)
 347 {
 348         misc_deregister(&lguest_dev);
 349 }