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