drivers/lguest/hypercalls.c

   1 /*P:500
   2  * Just as userspace programs request kernel operations through a system
   3  * call, the Guest requests Host operations through a "hypercall".  You might
   4  * notice this nomenclature doesn't really follow any logic, but the name has
   5  * been around for long enough that we're stuck with it.  As you'd expect, this
   6  * code is basically a one big switch statement.
   7 :*/
   8
   9 /*  Copyright (C) 2006 Rusty Russell IBM Corporation
  10
  11     This program is free software; you can redistribute it and/or modify
  12     it under the terms of the GNU General Public License as published by
  13     the Free Software Foundation; either version 2 of the License, or
  14     (at your option) any later version.
  15
  16     This program is distributed in the hope that it will be useful,
  17     but WITHOUT ANY WARRANTY; without even the implied warranty of
  18     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19     GNU General Public License for more details.
  20
  21     You should have received a copy of the GNU General Public License
  22     along with this program; if not, write to the Free Software
  23     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  24 */
  25 #include <linux/uaccess.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/mm.h>
  28 #include <linux/ktime.h>
  29 #include <asm/page.h>
  30 #include <asm/pgtable.h>
  31 #include "lg.h"
  32
  33 /*H:120
  34  * This is the core hypercall routine: where the Guest gets what it wants.
  35  * Or gets killed.  Or, in the case of LHCALL_SHUTDOWN, both.
  36  */
  37 static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
  38 {
  39         switch (args->arg0) {
  40         case LHCALL_FLUSH_ASYNC:
  41                 /*
  42                  * This call does nothing, except by breaking out of the Guest
  43                  * it makes us process all the asynchronous hypercalls.
  44                  */
  45                 break;
  46         case LHCALL_SEND_INTERRUPTS:
  47                 /*
  48                  * This call does nothing too, but by breaking out of the Guest
  49                  * it makes us process any pending interrupts.
  50                  */
  51                 break;
  52         case LHCALL_LGUEST_INIT:
  53                 /*
  54                  * You can't get here unless you're already initialized.  Don't
  55                  * do that.
  56                  */
  57                 kill_guest(cpu, "already have lguest_data");
  58                 break;
  59         case LHCALL_SHUTDOWN: {
  60                 char msg[128];
  61                 /*
  62                  * Shutdown is such a trivial hypercall that we do it in five
  63                  * lines right here.
  64                  *
  65                  * If the lgread fails, it will call kill_guest() itself; the
  66                  * kill_guest() with the message will be ignored.
  67                  */
  68                 __lgread(cpu, msg, args->arg1, sizeof(msg));
  69                 msg[sizeof(msg)-1] = '\0';
  70                 kill_guest(cpu, "CRASH: %s", msg);
  71                 if (args->arg2 == LGUEST_SHUTDOWN_RESTART)
  72                         cpu->lg->dead = ERR_PTR(-ERESTART);
  73                 break;
  74         }
  75         case LHCALL_FLUSH_TLB:
  76                 /* FLUSH_TLB comes in two flavors, depending on the argument: */
  77                 if (args->arg1)
  78                         guest_pagetable_clear_all(cpu);
  79                 else
  80                         guest_pagetable_flush_user(cpu);
  81                 break;
  82
  83         /*
  84          * All these calls simply pass the arguments through to the right
  85          * routines.
  86          */
  87         case LHCALL_NEW_PGTABLE:
  88                 guest_new_pagetable(cpu, args->arg1);
  89                 break;
  90         case LHCALL_SET_STACK:
  91                 guest_set_stack(cpu, args->arg1, args->arg2, args->arg3);
  92                 break;
  93         case LHCALL_SET_PTE:
  94 #ifdef CONFIG_X86_PAE
  95                 guest_set_pte(cpu, args->arg1, args->arg2,
  96                                 __pte(args->arg3 | (u64)args->arg4 << 32));
  97 #else
  98                 guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3));
  99 #endif
 100                 break;
 101         case LHCALL_SET_PGD:
 102                 guest_set_pgd(cpu->lg, args->arg1, args->arg2);
 103                 break;
 104 #ifdef CONFIG_X86_PAE
 105         case LHCALL_SET_PMD:
 106                 guest_set_pmd(cpu->lg, args->arg1, args->arg2);
 107                 break;
 108 #endif
 109         case LHCALL_SET_CLOCKEVENT:
 110                 guest_set_clockevent(cpu, args->arg1);
 111                 break;
 112         case LHCALL_HALT:
 113                 /* Similarly, this sets the halted flag for run_guest(). */
 114                 cpu->halted = 1;
 115                 break;
 116         default:
 117                 /* It should be an architecture-specific hypercall. */
 118                 if (lguest_arch_do_hcall(cpu, args))
 119                         kill_guest(cpu, "Bad hypercall %li\n", args->arg0);
 120         }
 121 }
 122
 123 /*H:124
 124  * Asynchronous hypercalls are easy: we just look in the array in the
 125  * Guest's "struct lguest_data" to see if any new ones are marked "ready".
 126  *
 127  * We are careful to do these in order: obviously we respect the order the
 128  * Guest put them in the ring, but we also promise the Guest that they will
 129  * happen before any normal hypercall (which is why we check this before
 130  * checking for a normal hcall).
 131  */
 132 static void do_async_hcalls(struct lg_cpu *cpu)
 133 {
 134         unsigned int i;
 135         u8 st[LHCALL_RING_SIZE];
 136
 137         /* For simplicity, we copy the entire call status array in at once. */
 138         if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st)))
 139                 return;
 140
 141         /* We process "struct lguest_data"s hcalls[] ring once. */
 142         for (i = 0; i < ARRAY_SIZE(st); i++) {
 143                 struct hcall_args args;
 144                 /*
 145                  * We remember where we were up to from last time.  This makes
 146                  * sure that the hypercalls are done in the order the Guest
 147                  * places them in the ring.
 148                  */
 149                 unsigned int n = cpu->next_hcall;
 150
 151                 /* 0xFF means there's no call here (yet). */
 152                 if (st[n] == 0xFF)
 153                         break;
 154
 155                 /*
 156                  * OK, we have hypercall.  Increment the "next_hcall" cursor,
 157                  * and wrap back to 0 if we reach the end.
 158                  */
 159                 if (++cpu->next_hcall == LHCALL_RING_SIZE)
 160                         cpu->next_hcall = 0;
 161
 162                 /*
 163                  * Copy the hypercall arguments into a local copy of the
 164                  * hcall_args struct.
 165                  */
 166                 if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n],
 167                                    sizeof(struct hcall_args))) {
 168                         kill_guest(cpu, "Fetching async hypercalls");
 169                         break;
 170                 }
 171
 172                 /* Do the hypercall, same as a normal one. */
 173                 do_hcall(cpu, &args);
 174
 175                 /* Mark the hypercall done. */
 176                 if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) {
 177                         kill_guest(cpu, "Writing result for async hypercall");
 178                         break;
 179                 }
 180
 181                 /*
 182                  * Stop doing hypercalls if they want to notify the Launcher:
 183                  * it needs to service this first.
 184                  */
 185                 if (cpu->pending.trap)
 186                         break;
 187         }
 188 }
 189
 190 /*
 191  * Last of all, we look at what happens first of all.  The very first time the
 192  * Guest makes a hypercall, we end up here to set things up:
 193  */
 194 static void initialize(struct lg_cpu *cpu)
 195 {
 196         /*
 197          * You can't do anything until you're initialized.  The Guest knows the
 198          * rules, so we're unforgiving here.
 199          */
 200         if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) {
 201                 kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0);
 202                 return;
 203         }
 204
 205         if (lguest_arch_init_hypercalls(cpu))
 206                 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
 207
 208         /*
 209          * The Guest tells us where we're not to deliver interrupts by putting
 210          * the instruction address into "struct lguest_data".
 211          */
 212         if (get_user(cpu->lg->noirq_iret, &cpu->lg->lguest_data->noirq_iret))
 213                 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
 214
 215         /*
 216          * We write the current time into the Guest's data page once so it can
 217          * set its clock.
 218          */
 219         write_timestamp(cpu);
 220
 221         /* page_tables.c will also do some setup. */
 222         page_table_guest_data_init(cpu);
 223
 224         /*
 225          * This is the one case where the above accesses might have been the
 226          * first write to a Guest page.  This may have caused a copy-on-write
 227          * fault, but the old page might be (read-only) in the Guest
 228          * pagetable.
 229          */
 230         guest_pagetable_clear_all(cpu);
 231 }
 232 /*:*/
 233
 234 /*M:013
 235  * If a Guest reads from a page (so creates a mapping) that it has never
 236  * written to, and then the Launcher writes to it (ie. the output of a virtual
 237  * device), the Guest will still see the old page.  In practice, this never
 238  * happens: why would the Guest read a page which it has never written to?  But
 239  * a similar scenario might one day bite us, so it's worth mentioning.
 240  *
 241  * Note that if we used a shared anonymous mapping in the Launcher instead of
 242  * mapping /dev/zero private, we wouldn't worry about cop-on-write.  And we
 243  * need that to switch the Launcher to processes (away from threads) anyway.
 244 :*/
 245
 246 /*H:100
 247  * Hypercalls
 248  *
 249  * Remember from the Guest, hypercalls come in two flavors: normal and
 250  * asynchronous.  This file handles both of types.
 251  */
 252 void do_hypercalls(struct lg_cpu *cpu)
 253 {
 254         /* Not initialized yet?  This hypercall must do it. */
 255         if (unlikely(!cpu->lg->lguest_data)) {
 256                 /* Set up the "struct lguest_data" */
 257                 initialize(cpu);
 258                 /* Hcall is done. */
 259                 cpu->hcall = NULL;
 260                 return;
 261         }
 262
 263         /*
 264          * The Guest has initialized.
 265          *
 266          * Look in the hypercall ring for the async hypercalls:
 267          */
 268         do_async_hcalls(cpu);
 269
 270         /*
 271          * If we stopped reading the hypercall ring because the Guest did a
 272          * NOTIFY to the Launcher, we want to return now.  Otherwise we do
 273          * the hypercall.
 274          */
 275         if (!cpu->pending.trap) {
 276                 do_hcall(cpu, cpu->hcall);
 277                 /*
 278                  * Tricky point: we reset the hcall pointer to mark the
 279                  * hypercall as "done".  We use the hcall pointer rather than
 280                  * the trap number to indicate a hypercall is pending.
 281                  * Normally it doesn't matter: the Guest will run again and
 282                  * update the trap number before we come back here.
 283                  *
 284                  * However, if we are signalled or the Guest sends I/O to the
 285                  * Launcher, the run_guest() loop will exit without running the
 286                  * Guest.  When it comes back it would try to re-run the
 287                  * hypercall.  Finding that bug sucked.
 288                  */
 289                 cpu->hcall = NULL;
 290         }
 291 }
 292
 293 /*
 294  * This routine supplies the Guest with time: it's used for wallclock time at
 295  * initial boot and as a rough time source if the TSC isn't available.
 296  */
 297 void write_timestamp(struct lg_cpu *cpu)
 298 {
 299         struct timespec now;
 300         ktime_get_real_ts(&now);
 301         if (copy_to_user(&cpu->lg->lguest_data->time,
 302                          &now, sizeof(struct timespec)))
 303                 kill_guest(cpu, "Writing timestamp");
 304 }