drivers/lguest/core.c

   1 /*P:400 This contains run_guest() which actually calls into the Host<->Guest
   2  * Switcher and analyzes the return, such as determining if the Guest wants the
   3  * Host to do something.  This file also contains useful helper routines. :*/
   4 #include <linux/module.h>
   5 #include <linux/stringify.h>
   6 #include <linux/stddef.h>
   7 #include <linux/io.h>
   8 #include <linux/mm.h>
   9 #include <linux/vmalloc.h>
  10 #include <linux/cpu.h>
  11 #include <linux/freezer.h>
  12 #include <linux/highmem.h>
  13 #include <asm/paravirt.h>
  14 #include <asm/pgtable.h>
  15 #include <asm/uaccess.h>
  16 #include <asm/poll.h>
  17 #include <asm/asm-offsets.h>
  18 #include "lg.h"
  19
  20
  21 static struct vm_struct *switcher_vma;
  22 static struct page **switcher_page;
  23
  24 /* This One Big lock protects all inter-guest data structures. */
  25 DEFINE_MUTEX(lguest_lock);
  26
  27 /*H:010 We need to set up the Switcher at a high virtual address.  Remember the
  28  * Switcher is a few hundred bytes of assembler code which actually changes the
  29  * CPU to run the Guest, and then changes back to the Host when a trap or
  30  * interrupt happens.
  31  *
  32  * The Switcher code must be at the same virtual address in the Guest as the
  33  * Host since it will be running as the switchover occurs.
  34  *
  35  * Trying to map memory at a particular address is an unusual thing to do, so
  36  * it's not a simple one-liner. */
  37 static __init int map_switcher(void)
  38 {
  39         int i, err;
  40         struct page **pagep;
  41
  42         /*
  43          * Map the Switcher in to high memory.
  44          *
  45          * It turns out that if we choose the address 0xFFC00000 (4MB under the
  46          * top virtual address), it makes setting up the page tables really
  47          * easy.
  48          */
  49
  50         /* We allocate an array of struct page pointers.  map_vm_area() wants
  51          * this, rather than just an array of pages. */
  52         switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
  53                                 GFP_KERNEL);
  54         if (!switcher_page) {
  55                 err = -ENOMEM;
  56                 goto out;
  57         }
  58
  59         /* Now we actually allocate the pages.  The Guest will see these pages,
  60          * so we make sure they're zeroed. */
  61         for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
  62                 unsigned long addr = get_zeroed_page(GFP_KERNEL);
  63                 if (!addr) {
  64                         err = -ENOMEM;
  65                         goto free_some_pages;
  66                 }
  67                 switcher_page[i] = virt_to_page(addr);
  68         }
  69
  70         /* First we check that the Switcher won't overlap the fixmap area at
  71          * the top of memory.  It's currently nowhere near, but it could have
  72          * very strange effects if it ever happened. */
  73         if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
  74                 err = -ENOMEM;
  75                 printk("lguest: mapping switcher would thwack fixmap\n");
  76                 goto free_pages;
  77         }
  78
  79         /* Now we reserve the "virtual memory area" we want: 0xFFC00000
  80          * (SWITCHER_ADDR).  We might not get it in theory, but in practice
  81          * it's worked so far.  The end address needs +1 because __get_vm_area
  82          * allocates an extra guard page, so we need space for that. */
  83         switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
  84                                      VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
  85                                      + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
  86         if (!switcher_vma) {
  87                 err = -ENOMEM;
  88                 printk("lguest: could not map switcher pages high\n");
  89                 goto free_pages;
  90         }
  91
  92         /* This code actually sets up the pages we've allocated to appear at
  93          * SWITCHER_ADDR.  map_vm_area() takes the vma we allocated above, the
  94          * kind of pages we're mapping (kernel pages), and a pointer to our
  95          * array of struct pages.  It increments that pointer, but we don't
  96          * care. */
  97         pagep = switcher_page;
  98         err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep);
  99         if (err) {
 100                 printk("lguest: map_vm_area failed: %i\n", err);
 101                 goto free_vma;
 102         }
 103
 104         /* Now the Switcher is mapped at the right address, we can't fail!
 105          * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
 106         memcpy(switcher_vma->addr, start_switcher_text,
 107                end_switcher_text - start_switcher_text);
 108
 109         printk(KERN_INFO "lguest: mapped switcher at %p\n",
 110                switcher_vma->addr);
 111         /* And we succeeded... */
 112         return 0;
 113
 114 free_vma:
 115         vunmap(switcher_vma->addr);
 116 free_pages:
 117         i = TOTAL_SWITCHER_PAGES;
 118 free_some_pages:
 119         for (--i; i >= 0; i--)
 120                 __free_pages(switcher_page[i], 0);
 121         kfree(switcher_page);
 122 out:
 123         return err;
 124 }
 125 /*:*/
 126
 127 /* Cleaning up the mapping when the module is unloaded is almost...
 128  * too easy. */
 129 static void unmap_switcher(void)
 130 {
 131         unsigned int i;
 132
 133         /* vunmap() undoes *both* map_vm_area() and __get_vm_area(). */
 134         vunmap(switcher_vma->addr);
 135         /* Now we just need to free the pages we copied the switcher into */
 136         for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
 137                 __free_pages(switcher_page[i], 0);
 138         kfree(switcher_page);
 139 }
 140
 141 /*H:032
 142  * Dealing With Guest Memory.
 143  *
 144  * Before we go too much further into the Host, we need to grok the routines
 145  * we use to deal with Guest memory.
 146  *
 147  * When the Guest gives us (what it thinks is) a physical address, we can use
 148  * the normal copy_from_user() & copy_to_user() on the corresponding place in
 149  * the memory region allocated by the Launcher.
 150  *
 151  * But we can't trust the Guest: it might be trying to access the Launcher
 152  * code.  We have to check that the range is below the pfn_limit the Launcher
 153  * gave us.  We have to make sure that addr + len doesn't give us a false
 154  * positive by overflowing, too. */
 155 bool lguest_address_ok(const struct lguest *lg,
 156                        unsigned long addr, unsigned long len)
 157 {
 158         return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
 159 }
 160
 161 /* This routine copies memory from the Guest.  Here we can see how useful the
 162  * kill_lguest() routine we met in the Launcher can be: we return a random
 163  * value (all zeroes) instead of needing to return an error. */
 164 void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
 165 {
 166         if (!lguest_address_ok(cpu->lg, addr, bytes)
 167             || copy_from_user(b, cpu->lg->mem_base + addr, bytes) != 0) {
 168                 /* copy_from_user should do this, but as we rely on it... */
 169                 memset(b, 0, bytes);
 170                 kill_guest(cpu, "bad read address %#lx len %u", addr, bytes);
 171         }
 172 }
 173
 174 /* This is the write (copy into Guest) version. */
 175 void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
 176                unsigned bytes)
 177 {
 178         if (!lguest_address_ok(cpu->lg, addr, bytes)
 179             || copy_to_user(cpu->lg->mem_base + addr, b, bytes) != 0)
 180                 kill_guest(cpu, "bad write address %#lx len %u", addr, bytes);
 181 }
 182 /*:*/
 183
 184 /*H:030 Let's jump straight to the the main loop which runs the Guest.
 185  * Remember, this is called by the Launcher reading /dev/lguest, and we keep
 186  * going around and around until something interesting happens. */
 187 int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
 188 {
 189         /* We stop running once the Guest is dead. */
 190         while (!cpu->lg->dead) {
 191                 unsigned int irq;
 192                 bool more;
 193
 194                 /* First we run any hypercalls the Guest wants done. */
 195                 if (cpu->hcall)
 196                         do_hypercalls(cpu);
 197
 198                 /* It's possible the Guest did a NOTIFY hypercall to the
 199                  * Launcher, in which case we return from the read() now. */
 200                 if (cpu->pending_notify) {
 201                         if (!send_notify_to_eventfd(cpu)) {
 202                                 if (put_user(cpu->pending_notify, user))
 203                                         return -EFAULT;
 204                                 return sizeof(cpu->pending_notify);
 205                         }
 206                 }
 207
 208                 /* Check for signals */
 209                 if (signal_pending(current))
 210                         return -ERESTARTSYS;
 211
 212                 /* Check if there are any interrupts which can be delivered now:
 213                  * if so, this sets up the hander to be executed when we next
 214                  * run the Guest. */
 215                 irq = interrupt_pending(cpu, &more);
 216                 if (irq < LGUEST_IRQS)
 217                         try_deliver_interrupt(cpu, irq, more);
 218
 219                 /* All long-lived kernel loops need to check with this horrible
 220                  * thing called the freezer.  If the Host is trying to suspend,
 221                  * it stops us. */
 222                 try_to_freeze();
 223
 224                 /* Just make absolutely sure the Guest is still alive.  One of
 225                  * those hypercalls could have been fatal, for example. */
 226                 if (cpu->lg->dead)
 227                         break;
 228
 229                 /* If the Guest asked to be stopped, we sleep.  The Guest's
 230                  * clock timer will wake us. */
 231                 if (cpu->halted) {
 232                         set_current_state(TASK_INTERRUPTIBLE);
 233                         /* Just before we sleep, make sure no interrupt snuck in
 234                          * which we should be doing. */
 235                         if (interrupt_pending(cpu, &more) < LGUEST_IRQS)
 236                                 set_current_state(TASK_RUNNING);
 237                         else
 238                                 schedule();
 239                         continue;
 240                 }
 241
 242                 /* OK, now we're ready to jump into the Guest.  First we put up
 243                  * the "Do Not Disturb" sign: */
 244                 local_irq_disable();
 245
 246                 /* Actually run the Guest until something happens. */
 247                 lguest_arch_run_guest(cpu);
 248
 249                 /* Now we're ready to be interrupted or moved to other CPUs */
 250                 local_irq_enable();
 251
 252                 /* Now we deal with whatever happened to the Guest. */
 253                 lguest_arch_handle_trap(cpu);
 254         }
 255
 256         /* Special case: Guest is 'dead' but wants a reboot. */
 257         if (cpu->lg->dead == ERR_PTR(-ERESTART))
 258                 return -ERESTART;
 259
 260         /* The Guest is dead => "No such file or directory" */
 261         return -ENOENT;
 262 }
 263
 264 /*H:000
 265  * Welcome to the Host!
 266  *
 267  * By this point your brain has been tickled by the Guest code and numbed by
 268  * the Launcher code; prepare for it to be stretched by the Host code.  This is
 269  * the heart.  Let's begin at the initialization routine for the Host's lg
 270  * module.
 271  */
 272 static int __init init(void)
 273 {
 274         int err;
 275
 276         /* Lguest can't run under Xen, VMI or itself.  It does Tricky Stuff. */
 277         if (paravirt_enabled()) {
 278                 printk("lguest is afraid of being a guest\n");
 279                 return -EPERM;
 280         }
 281
 282         /* First we put the Switcher up in very high virtual memory. */
 283         err = map_switcher();
 284         if (err)
 285                 goto out;
 286
 287         /* Now we set up the pagetable implementation for the Guests. */
 288         err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
 289         if (err)
 290                 goto unmap;
 291
 292         /* We might need to reserve an interrupt vector. */
 293         err = init_interrupts();
 294         if (err)
 295                 goto free_pgtables;
 296
 297         /* /dev/lguest needs to be registered. */
 298         err = lguest_device_init();
 299         if (err)
 300                 goto free_interrupts;
 301
 302         /* Finally we do some architecture-specific setup. */
 303         lguest_arch_host_init();
 304
 305         /* All good! */
 306         return 0;
 307
 308 free_interrupts:
 309         free_interrupts();
 310 free_pgtables:
 311         free_pagetables();
 312 unmap:
 313         unmap_switcher();
 314 out:
 315         return err;
 316 }
 317
 318 /* Cleaning up is just the same code, backwards.  With a little French. */
 319 static void __exit fini(void)
 320 {
 321         lguest_device_remove();
 322         free_interrupts();
 323         free_pagetables();
 324         unmap_switcher();
 325
 326         lguest_arch_host_fini();
 327 }
 328 /*:*/
 329
 330 /* The Host side of lguest can be a module.  This is a nice way for people to
 331  * play with it.  */
 332 module_init(init);
 333 module_exit(fini);
 334 MODULE_LICENSE("GPL");
 335 MODULE_AUTHOR("Rusty Russell <rusty@rustcorp.com.au>");