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 <linux/eventfd.h>
  11 #include <linux/file.h>
  12 #include "lg.h"
  13
  14 bool send_notify_to_eventfd(struct lg_cpu *cpu)
  15 {
  16         unsigned int i;
  17         struct lg_eventfd_map *map;
  18
  19         /* lg->eventfds is RCU-protected */
  20         rcu_read_lock();
  21         map = rcu_dereference(cpu->lg->eventfds);
  22         for (i = 0; i < map->num; i++) {
  23                 if (map->map[i].addr == cpu->pending_notify) {
  24                         eventfd_signal(map->map[i].event, 1);
  25                         cpu->pending_notify = 0;
  26                         break;
  27                 }
  28         }
  29         rcu_read_unlock();
  30         return cpu->pending_notify == 0;
  31 }
  32
  33 static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
  34 {
  35         struct lg_eventfd_map *new, *old = lg->eventfds;
  36
  37         if (!addr)
  38                 return -EINVAL;
  39
  40         /* Replace the old array with the new one, carefully: others can
  41          * be accessing it at the same time */
  42         new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1),
  43                       GFP_KERNEL);
  44         if (!new)
  45                 return -ENOMEM;
  46
  47         /* First make identical copy. */
  48         memcpy(new->map, old->map, sizeof(old->map[0]) * old->num);
  49         new->num = old->num;
  50
  51         /* Now append new entry. */
  52         new->map[new->num].addr = addr;
  53         new->map[new->num].event = eventfd_ctx_fdget(fd);
  54         if (IS_ERR(new->map[new->num].event)) {
  55                 int err =  PTR_ERR(new->map[new->num].event);
  56                 kfree(new);
  57                 return err;
  58         }
  59         new->num++;
  60
  61         /* Now put new one in place. */
  62         rcu_assign_pointer(lg->eventfds, new);
  63
  64         /* We're not in a big hurry.  Wait until noone's looking at old
  65          * version, then delete it. */
  66         synchronize_rcu();
  67         kfree(old);
  68
  69         return 0;
  70 }
  71
  72 static int attach_eventfd(struct lguest *lg, const unsigned long __user *input)
  73 {
  74         unsigned long addr, fd;
  75         int err;
  76
  77         if (get_user(addr, input) != 0)
  78                 return -EFAULT;
  79         input++;
  80         if (get_user(fd, input) != 0)
  81                 return -EFAULT;
  82
  83         mutex_lock(&lguest_lock);
  84         err = add_eventfd(lg, addr, fd);
  85         mutex_unlock(&lguest_lock);
  86
  87         return err;
  88 }
  89
  90 /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
  91  * number to /dev/lguest. */
  92 static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
  93 {
  94         unsigned long irq;
  95
  96         if (get_user(irq, input) != 0)
  97                 return -EFAULT;
  98         if (irq >= LGUEST_IRQS)
  99                 return -EINVAL;
 100
 101         set_interrupt(cpu, irq);
 102         return 0;
 103 }
 104
 105 /*L:040 Once our Guest is initialized, the Launcher makes it run by reading
 106  * from /dev/lguest. */
 107 static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
 108 {
 109         struct lguest *lg = file->private_data;
 110         struct lg_cpu *cpu;
 111         unsigned int cpu_id = *o;
 112
 113         /* You must write LHREQ_INITIALIZE first! */
 114         if (!lg)
 115                 return -EINVAL;
 116
 117         /* Watch out for arbitrary vcpu indexes! */
 118         if (cpu_id >= lg->nr_cpus)
 119                 return -EINVAL;
 120
 121         cpu = &lg->cpus[cpu_id];
 122
 123         /* If you're not the task which owns the Guest, go away. */
 124         if (current != cpu->tsk)
 125                 return -EPERM;
 126
 127         /* If the Guest is already dead, we indicate why */
 128         if (lg->dead) {
 129                 size_t len;
 130
 131                 /* lg->dead either contains an error code, or a string. */
 132                 if (IS_ERR(lg->dead))
 133                         return PTR_ERR(lg->dead);
 134
 135                 /* We can only return as much as the buffer they read with. */
 136                 len = min(size, strlen(lg->dead)+1);
 137                 if (copy_to_user(user, lg->dead, len) != 0)
 138                         return -EFAULT;
 139                 return len;
 140         }
 141
 142         /* If we returned from read() last time because the Guest sent I/O,
 143          * clear the flag. */
 144         if (cpu->pending_notify)
 145                 cpu->pending_notify = 0;
 146
 147         /* Run the Guest until something interesting happens. */
 148         return run_guest(cpu, (unsigned long __user *)user);
 149 }
 150
 151 /*L:025 This actually initializes a CPU.  For the moment, a Guest is only
 152  * uniprocessor, so "id" is always 0. */
 153 static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
 154 {
 155         /* We have a limited number the number of CPUs in the lguest struct. */
 156         if (id >= ARRAY_SIZE(cpu->lg->cpus))
 157                 return -EINVAL;
 158
 159         /* Set up this CPU's id, and pointer back to the lguest struct. */
 160         cpu->id = id;
 161         cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
 162         cpu->lg->nr_cpus++;
 163
 164         /* Each CPU has a timer it can set. */
 165         init_clockdev(cpu);
 166
 167         /* We need a complete page for the Guest registers: they are accessible
 168          * to the Guest and we can only grant it access to whole pages. */
 169         cpu->regs_page = get_zeroed_page(GFP_KERNEL);
 170         if (!cpu->regs_page)
 171                 return -ENOMEM;
 172
 173         /* We actually put the registers at the bottom of the page. */
 174         cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
 175
 176         /* Now we initialize the Guest's registers, handing it the start
 177          * address. */
 178         lguest_arch_setup_regs(cpu, start_ip);
 179
 180         /* We keep a pointer to the Launcher task (ie. current task) for when
 181          * other Guests want to wake this one (eg. console input). */
 182         cpu->tsk = current;
 183
 184         /* We need to keep a pointer to the Launcher's memory map, because if
 185          * the Launcher dies we need to clean it up.  If we don't keep a
 186          * reference, it is destroyed before close() is called. */
 187         cpu->mm = get_task_mm(cpu->tsk);
 188
 189         /* We remember which CPU's pages this Guest used last, for optimization
 190          * when the same Guest runs on the same CPU twice. */
 191         cpu->last_pages = NULL;
 192
 193         /* No error == success. */
 194         return 0;
 195 }
 196
 197 /*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit)
 198  * values (in addition to the LHREQ_INITIALIZE value).  These are:
 199  *
 200  * base: The start of the Guest-physical memory inside the Launcher memory.
 201  *
 202  * pfnlimit: The highest (Guest-physical) page number the Guest should be
 203  * allowed to access.  The Guest memory lives inside the Launcher, so it sets
 204  * this to ensure the Guest can only reach its own memory.
 205  *
 206  * start: The first instruction to execute ("eip" in x86-speak).
 207  */
 208 static int initialize(struct file *file, const unsigned long __user *input)
 209 {
 210         /* "struct lguest" contains everything we (the Host) know about a
 211          * Guest. */
 212         struct lguest *lg;
 213         int err;
 214         unsigned long args[3];
 215
 216         /* We grab the Big Lguest lock, which protects against multiple
 217          * simultaneous initializations. */
 218         mutex_lock(&lguest_lock);
 219         /* You can't initialize twice!  Close the device and start again... */
 220         if (file->private_data) {
 221                 err = -EBUSY;
 222                 goto unlock;
 223         }
 224
 225         if (copy_from_user(args, input, sizeof(args)) != 0) {
 226                 err = -EFAULT;
 227                 goto unlock;
 228         }
 229
 230         lg = kzalloc(sizeof(*lg), GFP_KERNEL);
 231         if (!lg) {
 232                 err = -ENOMEM;
 233                 goto unlock;
 234         }
 235
 236         lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL);
 237         if (!lg->eventfds) {
 238                 err = -ENOMEM;
 239                 goto free_lg;
 240         }
 241         lg->eventfds->num = 0;
 242
 243         /* Populate the easy fields of our "struct lguest" */
 244         lg->mem_base = (void __user *)args[0];
 245         lg->pfn_limit = args[1];
 246
 247         /* This is the first cpu (cpu 0) and it will start booting at args[2] */
 248         err = lg_cpu_start(&lg->cpus[0], 0, args[2]);
 249         if (err)
 250                 goto free_eventfds;
 251
 252         /* Initialize the Guest's shadow page tables, using the toplevel
 253          * address the Launcher gave us.  This allocates memory, so can fail. */
 254         err = init_guest_pagetable(lg);
 255         if (err)
 256                 goto free_regs;
 257
 258         /* We keep our "struct lguest" in the file's private_data. */
 259         file->private_data = lg;
 260
 261         mutex_unlock(&lguest_lock);
 262
 263         /* And because this is a write() call, we return the length used. */
 264         return sizeof(args);
 265
 266 free_regs:
 267         /* FIXME: This should be in free_vcpu */
 268         free_page(lg->cpus[0].regs_page);
 269 free_eventfds:
 270         kfree(lg->eventfds);
 271 free_lg:
 272         kfree(lg);
 273 unlock:
 274         mutex_unlock(&lguest_lock);
 275         return err;
 276 }
 277
 278 /*L:010 The first operation the Launcher does must be a write.  All writes
 279  * start with an unsigned long number: for the first write this must be
 280  * LHREQ_INITIALIZE to set up the Guest.  After that the Launcher can use
 281  * writes of other values to send interrupts.
 282  *
 283  * Note that we overload the "offset" in the /dev/lguest file to indicate what
 284  * CPU number we're dealing with.  Currently this is always 0, since we only
 285  * support uniprocessor Guests, but you can see the beginnings of SMP support
 286  * here. */
 287 static ssize_t write(struct file *file, const char __user *in,
 288                      size_t size, loff_t *off)
 289 {
 290         /* Once the Guest is initialized, we hold the "struct lguest" in the
 291          * file private data. */
 292         struct lguest *lg = file->private_data;
 293         const unsigned long __user *input = (const unsigned long __user *)in;
 294         unsigned long req;
 295         struct lg_cpu *uninitialized_var(cpu);
 296         unsigned int cpu_id = *off;
 297
 298         /* The first value tells us what this request is. */
 299         if (get_user(req, input) != 0)
 300                 return -EFAULT;
 301         input++;
 302
 303         /* If you haven't initialized, you must do that first. */
 304         if (req != LHREQ_INITIALIZE) {
 305                 if (!lg || (cpu_id >= lg->nr_cpus))
 306                         return -EINVAL;
 307                 cpu = &lg->cpus[cpu_id];
 308
 309                 /* Once the Guest is dead, you can only read() why it died. */
 310                 if (lg->dead)
 311                         return -ENOENT;
 312         }
 313
 314         switch (req) {
 315         case LHREQ_INITIALIZE:
 316                 return initialize(file, input);
 317         case LHREQ_IRQ:
 318                 return user_send_irq(cpu, input);
 319         case LHREQ_EVENTFD:
 320                 return attach_eventfd(lg, input);
 321         default:
 322                 return -EINVAL;
 323         }
 324 }
 325
 326 /*L:060 The final piece of interface code is the close() routine.  It reverses
 327  * everything done in initialize().  This is usually called because the
 328  * Launcher exited.
 329  *
 330  * Note that the close routine returns 0 or a negative error number: it can't
 331  * really fail, but it can whine.  I blame Sun for this wart, and K&R C for
 332  * letting them do it. :*/
 333 static int close(struct inode *inode, struct file *file)
 334 {
 335         struct lguest *lg = file->private_data;
 336         unsigned int i;
 337
 338         /* If we never successfully initialized, there's nothing to clean up */
 339         if (!lg)
 340                 return 0;
 341
 342         /* We need the big lock, to protect from inter-guest I/O and other
 343          * Launchers initializing guests. */
 344         mutex_lock(&lguest_lock);
 345
 346         /* Free up the shadow page tables for the Guest. */
 347         free_guest_pagetable(lg);
 348
 349         for (i = 0; i < lg->nr_cpus; i++) {
 350                 /* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
 351                 hrtimer_cancel(&lg->cpus[i].hrt);
 352                 /* We can free up the register page we allocated. */
 353                 free_page(lg->cpus[i].regs_page);
 354                 /* Now all the memory cleanups are done, it's safe to release
 355                  * the Launcher's memory management structure. */
 356                 mmput(lg->cpus[i].mm);
 357         }
 358
 359         /* Release any eventfds they registered. */
 360         for (i = 0; i < lg->eventfds->num; i++)
 361                 eventfd_ctx_put(lg->eventfds->map[i].event);
 362         kfree(lg->eventfds);
 363
 364         /* If lg->dead doesn't contain an error code it will be NULL or a
 365          * kmalloc()ed string, either of which is ok to hand to kfree(). */
 366         if (!IS_ERR(lg->dead))
 367                 kfree(lg->dead);
 368         /* Free the memory allocated to the lguest_struct */
 369         kfree(lg);
 370         /* Release lock and exit. */
 371         mutex_unlock(&lguest_lock);
 372
 373         return 0;
 374 }
 375
 376 /*L:000
 377  * Welcome to our journey through the Launcher!
 378  *
 379  * The Launcher is the Host userspace program which sets up, runs and services
 380  * the Guest.  In fact, many comments in the Drivers which refer to "the Host"
 381  * doing things are inaccurate: the Launcher does all the device handling for
 382  * the Guest, but the Guest can't know that.
 383  *
 384  * Just to confuse you: to the Host kernel, the Launcher *is* the Guest and we
 385  * shall see more of that later.
 386  *
 387  * We begin our understanding with the Host kernel interface which the Launcher
 388  * uses: reading and writing a character device called /dev/lguest.  All the
 389  * work happens in the read(), write() and close() routines: */
 390 static struct file_operations lguest_fops = {
 391         .owner   = THIS_MODULE,
 392         .release = close,
 393         .write   = write,
 394         .read    = read,
 395 };
 396
 397 /* This is a textbook example of a "misc" character device.  Populate a "struct
 398  * miscdevice" and register it with misc_register(). */
 399 static struct miscdevice lguest_dev = {
 400         .minor  = MISC_DYNAMIC_MINOR,
 401         .name   = "lguest",
 402         .fops   = &lguest_fops,
 403 };
 404
 405 int __init lguest_device_init(void)
 406 {
 407         return misc_register(&lguest_dev);
 408 }
 409
 410 void __exit lguest_device_remove(void)
 411 {
 412         misc_deregister(&lguest_dev);
 413 }