drivers/lguest/segments.c

   1 /*P:600 The x86 architecture has segments, which involve a table of descriptors
   2  * which can be used to do funky things with virtual address interpretation.
   3  * We originally used to use segments so the Guest couldn't alter the
   4  * Guest<->Host Switcher, and then we had to trim Guest segments, and restore
   5  * for userspace per-thread segments, but trim again for on userspace->kernel
   6  * transitions...  This nightmarish creation was contained within this file,
   7  * where we knew not to tread without heavy armament and a change of underwear.
   8  *
   9  * In these modern times, the segment handling code consists of simple sanity
  10  * checks, and the worst you'll experience reading this code is butterfly-rash
  11  * from frolicking through its parklike serenity. :*/
  12 #include "lg.h"
  13
  14 /*H:600
  15  * We've almost completed the Host; there's just one file to go!
  16  *
  17  * Segments & The Global Descriptor Table
  18  *
  19  * (That title sounds like a bad Nerdcore group.  Not to suggest that there are
  20  * any good Nerdcore groups, but in high school a friend of mine had a band
  21  * called Joe Fish and the Chips, so there are definitely worse band names).
  22  *
  23  * To refresh: the GDT is a table of 8-byte values describing segments.  Once
  24  * set up, these segments can be loaded into one of the 6 "segment registers".
  25  *
  26  * GDT entries are passed around as "struct desc_struct"s, which like IDT
  27  * entries are split into two 32-bit members, "a" and "b".  One day, someone
  28  * will clean that up, and be declared a Hero.  (No pressure, I'm just saying).
  29  *
  30  * Anyway, the GDT entry contains a base (the start address of the segment), a
  31  * limit (the size of the segment - 1), and some flags.  Sounds simple, and it
  32  * would be, except those zany Intel engineers decided that it was too boring
  33  * to put the base at one end, the limit at the other, and the flags in
  34  * between.  They decided to shotgun the bits at random throughout the 8 bytes,
  35  * like so:
  36  *
  37  * 0               16                     40       48  52  56     63
  38  * [ limit part 1 ][     base part 1     ][ flags ][li][fl][base ]
  39  *                                                  mit ags part 2
  40  *                                                part 2
  41  *
  42  * As a result, this file contains a certain amount of magic numeracy.  Let's
  43  * begin.
  44  */
  45
  46 /* There are several entries we don't let the Guest set.  The TSS entry is the
  47  * "Task State Segment" which controls all kinds of delicate things.  The
  48  * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the
  49  * the Guest can't be trusted to deal with double faults. */
  50 static int ignored_gdt(unsigned int num)
  51 {
  52         return (num == GDT_ENTRY_TSS
  53                 || num == GDT_ENTRY_LGUEST_CS
  54                 || num == GDT_ENTRY_LGUEST_DS
  55                 || num == GDT_ENTRY_DOUBLEFAULT_TSS);
  56 }
  57
  58 /*H:610 Once the GDT has been changed, we fix the new entries up a little.  We
  59  * don't care if they're invalid: the worst that can happen is a General
  60  * Protection Fault in the Switcher when it restores a Guest segment register
  61  * which tries to use that entry.  Then we kill the Guest for causing such a
  62  * mess: the message will be "unhandled trap 256". */
  63 static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end)
  64 {
  65         unsigned int i;
  66
  67         for (i = start; i < end; i++) {
  68                 /* We never copy these ones to real GDT, so we don't care what
  69                  * they say */
  70                 if (ignored_gdt(i))
  71                         continue;
  72
  73                 /* Segment descriptors contain a privilege level: the Guest is
  74                  * sometimes careless and leaves this as 0, even though it's
  75                  * running at privilege level 1.  If so, we fix it here. */
  76                 if ((lg->gdt[i].b & 0x00006000) == 0)
  77                         lg->gdt[i].b |= (GUEST_PL << 13);
  78
  79                 /* Each descriptor has an "accessed" bit.  If we don't set it
  80                  * now, the CPU will try to set it when the Guest first loads
  81                  * that entry into a segment register.  But the GDT isn't
  82                  * writable by the Guest, so bad things can happen. */
  83                 lg->gdt[i].b |= 0x00000100;
  84         }
  85 }
  86
  87 /* This routine is called at boot or modprobe time for each CPU to set up the
  88  * "constant" GDT entries for Guests running on that CPU. */
  89 void setup_default_gdt_entries(struct lguest_ro_state *state)
  90 {
  91         struct desc_struct *gdt = state->guest_gdt;
  92         unsigned long tss = (unsigned long)&state->guest_tss;
  93
  94         /* The hypervisor segments are full 0-4G segments, privilege level 0 */
  95         gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
  96         gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
  97
  98         /* The TSS segment refers to the TSS entry for this CPU, so we cannot
  99          * copy it from the Guest.  Forgive the magic flags */
 100         gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16);
 101         gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000)
 102                 | ((tss >> 16) & 0x000000FF);
 103 }
 104
 105 /* This routine is called before the Guest is run for the first time. */
 106 void setup_guest_gdt(struct lguest *lg)
 107 {
 108         /* Start with full 0-4G segments... */
 109         lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
 110         lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
 111         /* ...except the Guest is allowed to use them, so set the privilege
 112          * level appropriately in the flags. */
 113         lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
 114         lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
 115 }
 116
 117 /* Like the IDT, we never simply use the GDT the Guest gives us.  We set up the
 118  * GDTs for each CPU, then we copy across the entries each time we want to run
 119  * a different Guest on that CPU. */
 120
 121 /* A partial GDT load, for the three "thead-local storage" entries.  Otherwise
 122  * it's just like load_guest_gdt().  So much, in fact, it would probably be
 123  * neater to have a single hypercall to cover both. */
 124 void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt)
 125 {
 126         unsigned int i;
 127
 128         for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++)
 129                 gdt[i] = lg->gdt[i];
 130 }
 131
 132 /* This is the full version */
 133 void copy_gdt(const struct lguest *lg, struct desc_struct *gdt)
 134 {
 135         unsigned int i;
 136
 137         /* The default entries from setup_default_gdt_entries() are not
 138          * replaced.  See ignored_gdt() above. */
 139         for (i = 0; i < GDT_ENTRIES; i++)
 140                 if (!ignored_gdt(i))
 141                         gdt[i] = lg->gdt[i];
 142 }
 143
 144 /* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */
 145 void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
 146 {
 147         /* We assume the Guest has the same number of GDT entries as the
 148          * Host, otherwise we'd have to dynamically allocate the Guest GDT. */
 149         if (num > ARRAY_SIZE(lg->gdt))
 150                 kill_guest(lg, "too many gdt entries %i", num);
 151
 152         /* We read the whole thing in, then fix it up. */
 153         lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0]));
 154         fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt));
 155         /* Mark that the GDT changed so the core knows it has to copy it again,
 156          * even if the Guest is run on the same CPU. */
 157         lg->changed |= CHANGED_GDT;
 158 }
 159
 160 void guest_load_tls(struct lguest *lg, unsigned long gtls)
 161 {
 162         struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN];
 163
 164         lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES);
 165         fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1);
 166         lg->changed |= CHANGED_GDT_TLS;
 167 }
 168
 169 /*
 170  * With this, we have finished the Host.
 171  *
 172  * Five of the seven parts of our task are complete.  You have made it through
 173  * the Bit of Despair (I think that's somewhere in the page table code,
 174  * myself).
 175  *
 176  * Next, we examine "make Switcher".  It's short, but intense.
 177  */