Libgcj License. Please consult the file "LIBGCJ_LICENSE" for
details. */
-// Writte by Tom Tromey <tromey@redhat.com>
+// Written by Tom Tromey <tromey@redhat.com>
+
+// Define VERIFY_DEBUG to enable debugging output.
#include <config.h>
#include <java/lang/VerifyError.h>
#include <java/lang/Throwable.h>
#include <java/lang/reflect/Modifier.h>
+#include <java/lang/StringBuffer.h>
+#ifdef VERIFY_DEBUG
+#include <stdio.h>
+#endif /* VERIFY_DEBUG */
// TO DO
// * read more about when classes must be loaded
-// * there are bugs with boolean arrays?
// * class loader madness
// * Lots and lots of debugging and testing
// * type representation is still ugly. look for the big switches
// This is global because __attribute__ doesn't seem to work on static
// methods.
-static void verify_fail (char *s) __attribute__ ((__noreturn__));
+static void verify_fail (char *msg, jint pc = -1)
+ __attribute__ ((__noreturn__));
+
+static void debug_print (const char *fmt, ...)
+ __attribute__ ((format (printf, 1, 2)));
+
+static inline void
+debug_print (const char *fmt, ...)
+{
+#ifdef VERIFY_DEBUG
+ va_list ap;
+ va_start (ap, fmt);
+ vfprintf (stderr, fmt, ap);
+ va_end (ap);
+#endif /* VERIFY_DEBUG */
+}
class _Jv_BytecodeVerifier
{
static const int FLAG_INSN_START = 1;
static const int FLAG_BRANCH_TARGET = 2;
- static const int FLAG_JSR_TARGET = 4;
struct state;
struct type;
struct subr_info;
+ struct linked_utf8;
// The current PC.
int PC;
// This method.
_Jv_InterpMethod *current_method;
+ // A linked list of utf8 objects we allocate. This is really ugly,
+ // but without this our utf8 objects would be collected.
+ linked_utf8 *utf8_list;
+
+ struct linked_utf8
+ {
+ _Jv_Utf8Const *val;
+ linked_utf8 *next;
+ };
+
+ _Jv_Utf8Const *make_utf8_const (char *s, int len)
+ {
+ _Jv_Utf8Const *val = _Jv_makeUtf8Const (s, len);
+ _Jv_Utf8Const *r = (_Jv_Utf8Const *) _Jv_Malloc (sizeof (_Jv_Utf8Const)
+ + val->length
+ + 1);
+ r->length = val->length;
+ r->hash = val->hash;
+ memcpy (r->data, val->data, val->length + 1);
+
+ linked_utf8 *lu = (linked_utf8 *) _Jv_Malloc (sizeof (linked_utf8));
+ lu->val = r;
+ lu->next = utf8_list;
+ utf8_list = lu;
+
+ return r;
+ }
+
// This enum holds a list of tags for all the different types we
// need to handle. Reference types are treated specially by the
// type class.
return_address_type,
continuation_type,
+ // There is an obscure special case which requires us to note when
+ // a local variable has not been used by a subroutine. See
+ // push_jump_merge for more information.
+ unused_by_subroutine_type,
+
// Everything after `reference_type' must be a reference type.
reference_type,
null_type,
case 'Z':
rt = boolean_type;
break;
+ case 'B':
+ rt = byte_type;
+ break;
case 'C':
rt = char_type;
break;
return get_type_val_for_signature ((jchar) k->method_count);
}
+ // This is like _Jv_IsAssignableFrom, but it works even if SOURCE or
+ // TARGET haven't been prepared.
+ static bool is_assignable_from_slow (jclass target, jclass source)
+ {
+ // This will terminate when SOURCE==Object.
+ while (true)
+ {
+ if (source == target)
+ return true;
+
+ if (target->isPrimitive () || source->isPrimitive ())
+ return false;
+
+ // Check array case first because we can have an array whose
+ // component type is not prepared; _Jv_IsAssignableFrom
+ // doesn't handle this correctly.
+ if (target->isArray ())
+ {
+ if (! source->isArray ())
+ return false;
+ target = target->getComponentType ();
+ source = source->getComponentType ();
+ }
+ // _Jv_IsAssignableFrom can handle a target which is an
+ // interface even if it hasn't been prepared.
+ else if ((target->state > JV_STATE_LINKED || target->isInterface ())
+ && source->state > JV_STATE_LINKED)
+ return _Jv_IsAssignableFrom (target, source);
+ else if (target->isInterface ())
+ {
+ for (int i = 0; i < source->interface_count; ++i)
+ {
+ // We use a recursive call because we also need to
+ // check superinterfaces.
+ if (is_assignable_from_slow (target, source->interfaces[i]))
+ return true;
+ }
+ return false;
+ }
+ else if (target == &java::lang::Object::class$)
+ return true;
+ else if (source->isInterface ()
+ || source == &java::lang::Object::class$)
+ return false;
+ else
+ source = source->getSuperclass ();
+ }
+ }
+
// This is used to keep track of which `jsr's correspond to a given
// jsr target.
struct subr_info
}
// Promote a numeric type.
- void promote ()
+ type &promote ()
{
if (key == boolean_type || key == char_type
|| key == byte_type || key == short_type)
key = int_type;
+ return *this;
}
// If *THIS is an unresolved reference type, resolve it.
}
// Mark this type as the uninitialized result of `new'.
- void set_uninitialized (int pc)
+ void set_uninitialized (int npc)
{
- if (key != reference_type && key != unresolved_reference_type)
+ if (key == reference_type)
+ key = uninitialized_reference_type;
+ else if (key == unresolved_reference_type)
+ key = uninitialized_unresolved_reference_type;
+ else
verify_fail ("internal error in type::uninitialized");
- key = (key == reference_type
- ? uninitialized_reference_type
- : uninitialized_unresolved_reference_type);
- pc = pc;
+ pc = npc;
}
// Mark this type as now initialized.
void set_initialized (int npc)
{
- if (pc == npc)
+ if (npc != UNINIT && pc == npc
+ && (key == uninitialized_reference_type
+ || key == uninitialized_unresolved_reference_type))
{
key = (key == uninitialized_reference_type
? reference_type
// We must resolve both types and check assignability.
resolve ();
k.resolve ();
- return data.klass->isAssignableFrom (k.data.klass);
+ return is_assignable_from_slow (data.klass, k.data.klass);
}
bool isvoid () const
return type (k);
}
+ // Return the array type corresponding to an initialized
+ // reference. We could expand this to work for other kinds of
+ // types, but currently we don't need to.
+ type to_array ()
+ {
+ // Resolving isn't ideal, because it might force us to load
+ // another class, but it's easy. FIXME?
+ if (key == unresolved_reference_type)
+ resolve ();
+
+ if (key == reference_type)
+ return type (_Jv_GetArrayClass (data.klass,
+ data.klass->getClassLoader ()));
+ else
+ verify_fail ("internal error in type::to_array()");
+ }
+
bool isreference () const
{
return key >= reference_type;
// This loop will end when we hit Object.
while (true)
{
- if (k->isAssignableFrom (oldk))
+ if (is_assignable_from_slow (k, oldk))
break;
k = k->getSuperclass ();
changed = true;
{
if (local_semantics)
{
- key = unsuitable_type;
- changed = true;
+ // If we're merging into an "unused" slot, then we
+ // simply accept whatever we're merging from.
+ if (key == unused_by_subroutine_type)
+ {
+ *this = old_type;
+ changed = true;
+ }
+ else if (old_type.key == unused_by_subroutine_type)
+ {
+ // Do nothing.
+ }
+ // If we already have an `unsuitable' type, then we
+ // don't need to change again.
+ else if (key != unsuitable_type)
+ {
+ key = unsuitable_type;
+ changed = true;
+ }
}
else
verify_fail ("unmergeable type");
}
return changed;
}
+
+#ifdef VERIFY_DEBUG
+ void print (void) const
+ {
+ char c = '?';
+ switch (key)
+ {
+ case boolean_type: c = 'Z'; break;
+ case byte_type: c = 'B'; break;
+ case char_type: c = 'C'; break;
+ case short_type: c = 'S'; break;
+ case int_type: c = 'I'; break;
+ case long_type: c = 'J'; break;
+ case float_type: c = 'F'; break;
+ case double_type: c = 'D'; break;
+ case void_type: c = 'V'; break;
+ case unsuitable_type: c = '-'; break;
+ case return_address_type: c = 'r'; break;
+ case continuation_type: c = '+'; break;
+ case unused_by_subroutine_type: c = '_'; break;
+ case reference_type: c = 'L'; break;
+ case null_type: c = '@'; break;
+ case unresolved_reference_type: c = 'l'; break;
+ case uninitialized_reference_type: c = 'U'; break;
+ case uninitialized_unresolved_reference_type: c = 'u'; break;
+ }
+ debug_print ("%c", c);
+ }
+#endif /* VERIFY_DEBUG */
};
// This class holds all the state information we need for a given
// This is used to keep a linked list of all the states which
// require re-verification. We use the PC to keep track.
int next;
+ // We keep track of the type of `this' specially. This is used to
+ // ensure that an instance initializer invokes another initializer
+ // on `this' before returning. We must keep track of this
+ // specially because otherwise we might be confused by code which
+ // assigns to locals[0] (overwriting `this') and then returns
+ // without really initializing.
+ type this_type;
// INVALID marks a state which is not on the linked list of states
// requiring reverification.
static const int NO_NEXT = -2;
state ()
+ : this_type ()
{
stack = NULL;
locals = NULL;
}
state (int max_stack, int max_locals)
+ : this_type ()
{
stacktop = 0;
stackdepth = 0;
subroutine = 0;
}
- state (const state *copy, int max_stack, int max_locals)
+ state (const state *orig, int max_stack, int max_locals,
+ bool ret_semantics = false)
{
stack = new type[max_stack];
locals = new type[max_locals];
local_changed = (bool *) _Jv_Malloc (sizeof (bool) * max_locals);
- *this = *copy;
+ copy (orig, max_stack, max_locals, ret_semantics);
next = INVALID;
}
_Jv_Free (mem);
}
- void copy (const state *copy, int max_stack, int max_locals)
+ void copy (const state *copy, int max_stack, int max_locals,
+ bool ret_semantics = false)
{
stacktop = copy->stacktop;
stackdepth = copy->stackdepth;
stack[i] = copy->stack[i];
for (int i = 0; i < max_locals; ++i)
{
- locals[i] = copy->locals[i];
+ // See push_jump_merge to understand this case.
+ if (ret_semantics)
+ locals[i] = type (copy->local_changed[i]
+ ? unsuitable_type
+ : unused_by_subroutine_type);
+ else
+ locals[i] = copy->locals[i];
local_changed[i] = copy->local_changed[i];
}
+ this_type = copy->this_type;
// Don't modify `next'.
}
// FIXME: subroutine handling?
}
- // Merge STATE into this state. Destructively modifies this state.
- // Returns true if the new state was in fact changed. Will throw an
- // exception if the states are not mergeable.
+ // Merge STATE_OLD into this state. Destructively modifies this
+ // state. Returns true if the new state was in fact changed.
+ // Will throw an exception if the states are not mergeable.
bool merge (state *state_old, bool ret_semantics,
int max_locals)
{
bool changed = false;
+ // Special handling for `this'. If one or the other is
+ // uninitialized, then the merge is uninitialized.
+ if (this_type.isinitialized ())
+ this_type = state_old->this_type;
+
// Merge subroutine states. *THIS and *STATE_OLD must be in the
// same subroutine. Also, recursive subroutine calls must be
// avoided.
for (int i = 0; i < max_locals; ++i)
if (locals[i].isreference () && ! locals[i].isinitialized ())
verify_fail ("uninitialized object in local variable");
+
+ check_this_initialized ();
}
- // Note that a local variable was accessed or modified.
+ // Ensure that `this' has been initialized.
+ void check_this_initialized ()
+ {
+ if (this_type.isreference () && ! this_type.isinitialized ())
+ verify_fail ("`this' is uninitialized");
+ }
+
+ // Set type of `this'.
+ void set_this_type (const type &k)
+ {
+ this_type = k;
+ }
+
+ // Note that a local variable was modified.
void note_variable (int index)
{
if (subroutine > 0)
stack[i].set_initialized (pc);
for (int i = 0; i < max_locals; ++i)
locals[i].set_initialized (pc);
+ this_type.set_initialized (pc);
+ }
+
+ // Return true if this state is the unmerged result of a `ret'.
+ bool is_unmerged_ret_state (int max_locals) const
+ {
+ for (int i = 0; i < max_locals; ++i)
+ {
+ if (locals[i].key == unused_by_subroutine_type)
+ return true;
+ }
+ return false;
+ }
+
+#ifdef VERIFY_DEBUG
+ void print (const char *leader, int pc,
+ int max_stack, int max_locals) const
+ {
+ debug_print ("%s [%4d]: [stack] ", leader, pc);
+ int i;
+ for (i = 0; i < stacktop; ++i)
+ stack[i].print ();
+ for (; i < max_stack; ++i)
+ debug_print (".");
+ debug_print (" [local] ");
+ for (i = 0; i < max_locals; ++i)
+ locals[i].print ();
+ debug_print (" | %p\n", this);
}
+#else
+ inline void print (const char *, int, int, int) const
+ {
+ }
+#endif /* VERIFY_DEBUG */
};
type pop_raw ()
{
if (current_state->stacktop <= 0)
- verify_fail ("stack empty");
+ verify_fail ("stack empty", start_PC);
type r = current_state->stack[--current_state->stacktop];
current_state->stackdepth -= r.depth ();
if (current_state->stackdepth < 0)
- verify_fail ("stack empty");
+ verify_fail ("stack empty", start_PC);
return r;
}
{
type r = pop_raw ();
if (r.iswide ())
- verify_fail ("narrow pop of wide type");
+ verify_fail ("narrow pop of wide type", start_PC);
return r;
}
{
type r = pop_raw ();
if (! r.iswide ())
- verify_fail ("wide pop of narrow type");
+ verify_fail ("wide pop of narrow type", start_PC);
return r;
}
type pop_type (type match)
{
+ match.promote ();
type t = pop_raw ();
if (! match.compatible (t))
- verify_fail ("incompatible type on stack");
+ verify_fail ("incompatible type on stack", start_PC);
+ return t;
+ }
+
+ // Pop a reference type or a return address.
+ type pop_ref_or_return ()
+ {
+ type t = pop_raw ();
+ if (! t.isreference () && t.key != return_address_type)
+ verify_fail ("expected reference or return address on stack", start_PC);
return t;
}
{
int depth = t.depth ();
if (index > current_method->max_locals - depth)
- verify_fail ("invalid local variable");
+ verify_fail ("invalid local variable", start_PC);
if (! t.compatible (current_state->locals[index]))
- verify_fail ("incompatible type in local variable");
+ verify_fail ("incompatible type in local variable", start_PC);
if (depth == 2)
{
type t (continuation_type);
if (! current_state->locals[index + 1].compatible (t))
- verify_fail ("invalid local variable");
+ verify_fail ("invalid local variable", start_PC);
}
- current_state->note_variable (index);
return current_state->locals[index];
}
type t = array.element_type ();
if (! element.compatible (t))
- verify_fail ("incompatible array element type");
+ {
+ // Special case for byte arrays, which must also be boolean
+ // arrays.
+ bool ok = true;
+ if (element.key == byte_type)
+ {
+ type e2 (boolean_type);
+ ok = e2.compatible (t);
+ }
+ if (! ok)
+ verify_fail ("incompatible array element type");
+ }
// Return T and not ELEMENT, because T might be specialized.
return t;
jint get_ushort ()
{
- jbyte b1 = get_byte ();
- jbyte b2 = get_byte ();
+ jint b1 = get_byte ();
+ jint b2 = get_byte ();
return (jint) ((b1 << 8) | b2) & 0xffff;
}
jint get_short ()
{
- jbyte b1 = get_byte ();
- jbyte b2 = get_byte ();
+ jint b1 = get_byte ();
+ jint b2 = get_byte ();
jshort s = (b1 << 8) | b2;
return (jint) s;
}
jint get_int ()
{
- jbyte b1 = get_byte ();
- jbyte b2 = get_byte ();
- jbyte b3 = get_byte ();
- jbyte b4 = get_byte ();
+ jint b1 = get_byte ();
+ jint b2 = get_byte ();
+ jint b3 = get_byte ();
+ jint b4 = get_byte ();
return (b1 << 24) | (b2 << 16) | (b3 << 8) | b4;
}
{
int npc = start_PC + offset;
if (npc < 0 || npc >= current_method->code_length)
- verify_fail ("branch out of range");
+ verify_fail ("branch out of range", start_PC);
return npc;
}
- // Merge the indicated state into a new state and schedule a new PC if
- // there is a change. If RET_SEMANTICS is true, then we are merging
- // from a `ret' instruction into the instruction after a `jsr'. This
- // is a special case with its own modified semantics.
+ // Merge the indicated state into the state at the branch target and
+ // schedule a new PC if there is a change. If RET_SEMANTICS is
+ // true, then we are merging from a `ret' instruction into the
+ // instruction after a `jsr'. This is a special case with its own
+ // modified semantics.
void push_jump_merge (int npc, state *nstate, bool ret_semantics = false)
{
bool changed = true;
if (states[npc] == NULL)
{
- // FIXME: what if we reach this code from a `ret'?
-
+ // There's a weird situation here. If are examining the
+ // branch that results from a `ret', and there is not yet a
+ // state available at the branch target (the instruction just
+ // after the `jsr'), then we have to construct a special kind
+ // of state at that point for future merging. This special
+ // state has the type `unused_by_subroutine_type' in each slot
+ // which was not modified by the subroutine.
states[npc] = new state (nstate, current_method->max_stack,
- current_method->max_locals);
+ current_method->max_locals, ret_semantics);
+ debug_print ("== New state in push_jump_merge\n");
+ states[npc]->print ("New", npc, current_method->max_stack,
+ current_method->max_locals);
}
else
- changed = nstate->merge (states[npc], ret_semantics,
- current_method->max_stack);
+ {
+ debug_print ("== Merge states in push_jump_merge\n");
+ nstate->print ("Frm", start_PC, current_method->max_stack,
+ current_method->max_locals);
+ states[npc]->print (" To", npc, current_method->max_stack,
+ current_method->max_locals);
+ changed = states[npc]->merge (nstate, ret_semantics,
+ current_method->max_locals);
+ states[npc]->print ("New", npc, current_method->max_stack,
+ current_method->max_locals);
+ }
if (changed && states[npc]->next == state::INVALID)
{
{
int npc = compute_jump (offset);
if (npc < PC)
- current_state->check_no_uninitialized_objects (current_method->max_stack);
+ current_state->check_no_uninitialized_objects (current_method->max_locals);
push_jump_merge (npc, current_state);
}
void push_exception_jump (type t, int pc)
{
- current_state->check_no_uninitialized_objects (current_method->max_stack,
+ current_state->check_no_uninitialized_objects (current_method->max_locals,
true);
state s (current_state, current_method->max_stack,
current_method->max_locals);
int pop_jump ()
{
+ int *prev_loc = &next_verify_pc;
int npc = next_verify_pc;
- if (npc != state::NO_NEXT)
+ bool skipped = false;
+
+ while (npc != state::NO_NEXT)
{
- next_verify_pc = states[npc]->next;
- states[npc]->next = state::INVALID;
+ // If the next available PC is an unmerged `ret' state, then
+ // we aren't yet ready to handle it. That's because we would
+ // need all kind of special cases to do so. So instead we
+ // defer this jump until after we've processed it via a
+ // fall-through. This has to happen because the instruction
+ // before this one must be a `jsr'.
+ if (! states[npc]->is_unmerged_ret_state (current_method->max_locals))
+ {
+ *prev_loc = states[npc]->next;
+ states[npc]->next = state::INVALID;
+ return npc;
+ }
+
+ skipped = true;
+ prev_loc = &states[npc]->next;
+ npc = states[npc]->next;
}
- return npc;
+
+ // If we've skipped states and there is nothing else, that's a
+ // bug.
+ if (skipped)
+ verify_fail ("pop_jump: can't happen");
+ return state::NO_NEXT;
}
void invalidate_pc ()
void note_branch_target (int pc, bool is_jsr_target = false)
{
- if (pc <= PC && ! (flags[pc] & FLAG_INSN_START))
- verify_fail ("branch not to instruction start");
+ // Don't check `pc <= PC', because we've advanced PC after
+ // fetching the target and we haven't yet checked the next
+ // instruction.
+ if (pc < PC && ! (flags[pc] & FLAG_INSN_START))
+ verify_fail ("branch not to instruction start", start_PC);
flags[pc] |= FLAG_BRANCH_TARGET;
if (is_jsr_target)
{
info->pc = PC;
info->next = jsr_ptrs[pc];
jsr_ptrs[pc] = info;
- flags[pc] |= FLAG_JSR_TARGET;
}
}
// in the enclosing context.
current_state->subroutine = get_subroutine (subr->pc);
if (subr->pc < PC)
- current_state->check_no_uninitialized_objects (current_method->max_stack);
+ current_state->check_no_uninitialized_objects (current_method->max_locals);
push_jump_merge (subr->pc, current_state, true);
}
int npc = compute_jump (offset);
if (npc < PC)
- current_state->check_no_uninitialized_objects (current_method->max_stack);
+ current_state->check_no_uninitialized_objects (current_method->max_locals);
check_nonrecursive_call (current_state->subroutine, npc);
// Temporarily modify the current state so that it looks like we are
PC = 0;
while (PC < current_method->code_length)
{
+ // Set `start_PC' early so that error checking can have the
+ // correct value.
+ start_PC = PC;
flags[PC] |= FLAG_INSN_START;
// If the previous instruction was a jsr, then the next
note_branch_target (PC);
last_was_jsr = false;
- start_PC = PC;
- unsigned char opcode = bytecode[PC++];
+ java_opcode opcode = (java_opcode) bytecode[PC++];
switch (opcode)
{
case op_nop:
case op_lneg:
case op_fneg:
case op_dneg:
- case op_iinc:
case op_i2l:
case op_i2f:
case op_i2d:
case op_areturn:
case op_return:
case op_athrow:
+ case op_arraylength:
break;
case op_bipush:
- case op_sipush:
case op_ldc:
case op_iload:
case op_lload:
case op_fstore:
case op_dstore:
case op_astore:
- case op_arraylength:
case op_ret:
+ case op_newarray:
get_byte ();
break;
+ case op_iinc:
+ case op_sipush:
case op_ldc_w:
case op_ldc2_w:
case op_getstatic:
case op_putfield:
case op_putstatic:
case op_new:
- case op_newarray:
case op_anewarray:
case op_instanceof:
case op_checkcast:
jint low = get_int ();
jint hi = get_int ();
if (low > hi)
- verify_fail ("invalid tableswitch");
+ verify_fail ("invalid tableswitch", start_PC);
for (int i = low; i <= hi; ++i)
note_branch_target (compute_jump (get_int ()));
}
note_branch_target (compute_jump (get_int ()));
int npairs = get_int ();
if (npairs < 0)
- verify_fail ("too few pairs in lookupswitch");
+ verify_fail ("too few pairs in lookupswitch", start_PC);
while (npairs-- > 0)
{
get_int ();
case op_wide:
{
- opcode = get_byte ();
+ opcode = (java_opcode) get_byte ();
get_short ();
- if (opcode == (unsigned char) op_iinc)
+ if (opcode == op_iinc)
get_short ();
}
break;
break;
default:
- verify_fail ("unrecognized instruction in branch_prepass");
+ verify_fail ("unrecognized instruction in branch_prepass",
+ start_PC);
}
// See if any previous branch tried to branch to the middle of
for (int pc = start_PC + 1; pc < PC; ++pc)
{
if ((flags[pc] & FLAG_BRANCH_TARGET))
- verify_fail ("branch not to instruction start");
+ verify_fail ("branch to middle of instruction", pc);
}
}
for (int i = 0; i < current_method->exc_count; ++i)
{
if (! (flags[exception[i].handler_pc] & FLAG_INSN_START))
- verify_fail ("exception handler not at instruction start");
- if (exception[i].start_pc > exception[i].end_pc)
- verify_fail ("exception range inverted");
- if (! (flags[exception[i].start_pc] & FLAG_INSN_START)
- || ! (flags[exception[i].start_pc] & FLAG_INSN_START))
- verify_fail ("exception endpoint not at instruction start");
+ verify_fail ("exception handler not at instruction start",
+ exception[i].handler_pc);
+ if (! (flags[exception[i].start_pc] & FLAG_INSN_START))
+ verify_fail ("exception start not at instruction start",
+ exception[i].start_pc);
+ if (exception[i].end_pc != current_method->code_length
+ && ! (flags[exception[i].end_pc] & FLAG_INSN_START))
+ verify_fail ("exception end not at instruction start",
+ exception[i].end_pc);
flags[exception[i].handler_pc] |= FLAG_BRANCH_TARGET;
}
void check_pool_index (int index)
{
if (index < 0 || index >= current_class->constants.size)
- verify_fail ("constant pool index out of range");
+ verify_fail ("constant pool index out of range", start_PC);
}
type check_class_constant (int index)
return type (pool->data[index].clazz);
else if (pool->tags[index] == JV_CONSTANT_Class)
return type (pool->data[index].utf8);
- verify_fail ("expected class constant");
+ verify_fail ("expected class constant", start_PC);
}
type check_constant (int index)
return type (int_type);
else if (pool->tags[index] == JV_CONSTANT_Float)
return type (float_type);
- verify_fail ("String, int, or float constant expected");
+ verify_fail ("String, int, or float constant expected", start_PC);
+ }
+
+ type check_wide_constant (int index)
+ {
+ check_pool_index (index);
+ _Jv_Constants *pool = ¤t_class->constants;
+ if (pool->tags[index] == JV_CONSTANT_Long)
+ return type (long_type);
+ else if (pool->tags[index] == JV_CONSTANT_Double)
+ return type (double_type);
+ verify_fail ("long or double constant expected", start_PC);
}
// Helper for both field and method. These are laid out the same in
check_pool_index (index);
_Jv_Constants *pool = ¤t_class->constants;
if (pool->tags[index] != expected)
- verify_fail ("didn't see expected constant");
+ verify_fail ("didn't see expected constant", start_PC);
// Once we know we have a Fieldref or Methodref we assume that it
// is correctly laid out in the constant pool. I think the code
// in defineclass.cc guarantees this.
&name, &field_type);
if (class_type)
*class_type = ct;
- return type (field_type);
+ if (field_type->data[0] == '[' || field_type->data[0] == 'L')
+ return type (field_type);
+ return get_type_val_for_signature (field_type->data[0]);
}
type check_method_constant (int index, bool is_interface,
while (*p != ';')
++p;
++p;
- // FIXME! This will get collected!
- _Jv_Utf8Const *name = _Jv_makeUtf8Const (start, p - start);
+ _Jv_Utf8Const *name = make_utf8_const (start, p - start);
return type (name);
}
type_val rt = get_type_val_for_signature (jchar (v));
if (arraycount == 0)
- return type (rt);
+ {
+ // Callers of this function eventually push their arguments on
+ // the stack. So, promote them here.
+ return type (rt).promote ();
+ }
jclass k = construct_primitive_array_type (rt);
while (--arraycount > 0)
return get_one_type (p);
}
- void check_return_type (type expected)
+ void check_return_type (type onstack)
{
type rt = compute_return_type (current_method->self->signature);
- if (! expected.compatible (rt))
- verify_fail ("incompatible return type");
+ if (! rt.compatible (onstack))
+ verify_fail ("incompatible return type", start_PC);
+ }
+
+ // Initialize the stack for the new method. Returns true if this
+ // method is an instance initializer.
+ bool initialize_stack ()
+ {
+ int var = 0;
+ bool is_init = false;
+
+ using namespace java::lang::reflect;
+ if (! Modifier::isStatic (current_method->self->accflags))
+ {
+ type kurr (current_class);
+ if (_Jv_equalUtf8Consts (current_method->self->name, gcj::init_name))
+ {
+ kurr.set_uninitialized (type::SELF);
+ is_init = true;
+ }
+ set_variable (0, kurr);
+ current_state->set_this_type (kurr);
+ ++var;
+ }
+
+ // We have to handle wide arguments specially here.
+ int arg_count = _Jv_count_arguments (current_method->self->signature);
+ type arg_types[arg_count];
+ compute_argument_types (current_method->self->signature, arg_types);
+ for (int i = 0; i < arg_count; ++i)
+ {
+ set_variable (var, arg_types[i]);
+ ++var;
+ if (arg_types[i].iswide ())
+ ++var;
+ }
+
+ return is_init;
}
void verify_instructions_0 ()
current_method->max_locals);
PC = 0;
+ start_PC = 0;
- {
- int var = 0;
-
- using namespace java::lang::reflect;
- if (! Modifier::isStatic (current_method->self->accflags))
- {
- type kurr (current_class);
- if (_Jv_equalUtf8Consts (current_method->self->name, gcj::init_name))
- kurr.set_uninitialized (type::SELF);
- set_variable (0, kurr);
- ++var;
- }
-
- if (var + _Jv_count_arguments (current_method->self->signature)
- > current_method->max_locals)
- verify_fail ("too many arguments");
- compute_argument_types (current_method->self->signature,
- ¤t_state->locals[var]);
- }
+ // True if we are verifying an instance initializer.
+ bool this_is_init = initialize_stack ();
states = (state **) _Jv_Malloc (sizeof (state *)
* current_method->code_length);
{
PC = pop_jump ();
if (PC == state::INVALID)
- verify_fail ("saw state::INVALID");
+ verify_fail ("can't happen: saw state::INVALID");
if (PC == state::NO_NEXT)
break;
// Set up the current state.
- *current_state = *states[PC];
+ current_state->copy (states[PC], current_method->max_stack,
+ current_method->max_locals);
}
-
- // Control can't fall off the end of the bytecode.
- if (PC >= current_method->code_length)
- verify_fail ("fell off end");
-
- if (states[PC] != NULL)
+ else
{
- // We've already visited this instruction. So merge the
- // states together. If this yields no change then we don't
- // have to re-verify.
- if (! current_state->merge (states[PC], false,
- current_method->max_stack))
+ // Control can't fall off the end of the bytecode. We
+ // only need to check this in the fall-through case,
+ // because branch bounds are checked when they are
+ // pushed.
+ if (PC >= current_method->code_length)
+ verify_fail ("fell off end");
+
+ // We only have to do this checking in the situation where
+ // control flow falls through from the previous
+ // instruction. Otherwise merging is done at the time we
+ // push the branch.
+ if (states[PC] != NULL)
{
- invalidate_pc ();
- continue;
+ // We've already visited this instruction. So merge
+ // the states together. If this yields no change then
+ // we don't have to re-verify. However, if the new
+ // state is an the result of an unmerged `ret', we
+ // must continue through it.
+ debug_print ("== Fall through merge\n");
+ states[PC]->print ("Old", PC, current_method->max_stack,
+ current_method->max_locals);
+ current_state->print ("Cur", PC, current_method->max_stack,
+ current_method->max_locals);
+ if (! current_state->merge (states[PC], false,
+ current_method->max_locals)
+ && ! states[PC]->is_unmerged_ret_state (current_method->max_locals))
+ {
+ debug_print ("== Fall through optimization\n");
+ invalidate_pc ();
+ continue;
+ }
+ // Save a copy of it for later.
+ states[PC]->copy (current_state, current_method->max_stack,
+ current_method->max_locals);
+ current_state->print ("New", PC, current_method->max_stack,
+ current_method->max_locals);
}
- // Save a copy of it for later.
- states[PC]->copy (current_state, current_method->max_stack,
- current_method->max_locals);
}
- else if ((flags[PC] & FLAG_BRANCH_TARGET))
+
+ // We only have to keep saved state at branch targets. If
+ // we're at a branch target and the state here hasn't been set
+ // yet, we set it now.
+ if (states[PC] == NULL && (flags[PC] & FLAG_BRANCH_TARGET))
{
- // We only have to keep saved state at branch targets.
states[PC] = new state (current_state, current_method->max_stack,
current_method->max_locals);
}
+ // Set this before handling exceptions so that debug output is
+ // sane.
+ start_PC = PC;
+
// Update states for all active exception handlers. Ordinarily
// there are not many exception handlers. So we simply run
// through them all.
{
if (PC >= exception[i].start_pc && PC < exception[i].end_pc)
{
- type handler = reference_type;
+ type handler (&java::lang::Throwable::class$);
if (exception[i].handler_type != 0)
handler = check_class_constant (exception[i].handler_type);
push_exception_jump (handler, exception[i].handler_pc);
}
}
- start_PC = PC;
- unsigned char opcode = bytecode[PC++];
+ current_state->print (" ", PC, current_method->max_stack,
+ current_method->max_locals);
+ java_opcode opcode = (java_opcode) bytecode[PC++];
switch (opcode)
{
case op_nop:
push_type (check_constant (get_ushort ()));
break;
case op_ldc2_w:
- push_type (check_constant (get_ushort ()));
+ push_type (check_wide_constant (get_ushort ()));
break;
case op_iload:
set_variable (get_byte (), pop_type (double_type));
break;
case op_astore:
- set_variable (get_byte (), pop_type (reference_type));
+ set_variable (get_byte (), pop_ref_or_return ());
break;
case op_istore_0:
case op_istore_1:
case op_astore_1:
case op_astore_2:
case op_astore_3:
- set_variable (opcode - op_astore_0, pop_type (reference_type));
+ set_variable (opcode - op_astore_0, pop_ref_or_return ());
break;
case op_iastore:
pop_type (int_type);
case op_lmul:
case op_ldiv:
case op_lrem:
- case op_lshl:
- case op_lshr:
- case op_lushr:
case op_land:
case op_lor:
case op_lxor:
pop_type (long_type);
push_type (pop_type (long_type));
break;
+ case op_lshl:
+ case op_lshr:
+ case op_lushr:
+ pop_type (int_type);
+ push_type (pop_type (long_type));
+ break;
case op_fadd:
case op_fsub:
case op_fmul:
{
jint key = get_int ();
if (i > 0 && key <= lastkey)
- verify_fail ("lookupswitch pairs unsorted");
+ verify_fail ("lookupswitch pairs unsorted", start_PC);
lastkey = key;
push_jump (get_int ());
}
invalidate_pc ();
break;
case op_return:
+ // We only need to check this when the return type is
+ // void, because all instance initializers return void.
+ if (this_is_init)
+ current_state->check_this_initialized ();
check_return_type (void_type);
invalidate_pc ();
break;
type klass;
type field = check_field_constant (get_ushort (), &klass);
pop_type (field);
+
+ // We have an obscure special case here: we can use
+ // `putfield' on a field declared in this class, even if
+ // `this' has not yet been initialized.
+ if (! current_state->this_type.isinitialized ()
+ && current_state->this_type.pc == type::SELF)
+ klass.set_uninitialized (type::SELF);
pop_type (klass);
}
break;
_Jv_Utf8Const *method_name, *method_signature;
type class_type
= check_method_constant (get_ushort (),
- opcode == (unsigned char) op_invokeinterface,
+ opcode == op_invokeinterface,
&method_name,
&method_signature);
int arg_count = _Jv_count_arguments (method_signature);
- if (opcode == (unsigned char) op_invokeinterface)
+ if (opcode == op_invokeinterface)
{
int nargs = get_byte ();
if (nargs == 0)
- verify_fail ("too few arguments to invokeinterface");
+ verify_fail ("too few arguments to invokeinterface",
+ start_PC);
if (get_byte () != 0)
- verify_fail ("invokeinterface dummy byte is wrong");
+ verify_fail ("invokeinterface dummy byte is wrong",
+ start_PC);
if (nargs - 1 != arg_count)
- verify_fail ("wrong argument count for invokeinterface");
+ verify_fail ("wrong argument count for invokeinterface",
+ start_PC);
}
bool is_init = false;
if (_Jv_equalUtf8Consts (method_name, gcj::init_name))
{
is_init = true;
- if (opcode != (unsigned char) op_invokespecial)
- verify_fail ("can't invoke <init>");
+ if (opcode != op_invokespecial)
+ verify_fail ("can't invoke <init>", start_PC);
}
else if (method_name->data[0] == '<')
- verify_fail ("can't invoke method starting with `<'");
+ verify_fail ("can't invoke method starting with `<'",
+ start_PC);
// Pop arguments and check types.
type arg_types[arg_count];
for (int i = arg_count - 1; i >= 0; --i)
pop_type (arg_types[i]);
- if (opcode != (unsigned char) op_invokestatic)
+ if (opcode != op_invokestatic)
{
type t = class_type;
if (is_init)
{
type t = check_class_constant (get_ushort ());
if (t.isarray () || t.isinterface () || t.isabstract ())
- verify_fail ("type is array, interface, or abstract");
+ verify_fail ("type is array, interface, or abstract",
+ start_PC);
t.set_uninitialized (start_PC);
push_type (t);
}
// We intentionally have chosen constants to make this
// valid.
if (atype < boolean_type || atype > long_type)
- verify_fail ("type not primitive");
+ verify_fail ("type not primitive", start_PC);
pop_type (int_type);
push_type (construct_primitive_array_type (type_val (atype)));
}
break;
case op_anewarray:
pop_type (int_type);
- push_type (check_class_constant (get_ushort ()));
+ push_type (check_class_constant (get_ushort ()).to_array ());
break;
case op_arraylength:
{
type t = pop_type (reference_type);
if (! t.isarray ())
- verify_fail ("array type expected");
+ verify_fail ("array type expected", start_PC);
push_type (int_type);
}
break;
get_short ();
break;
default:
- verify_fail ("unrecognized wide instruction");
+ verify_fail ("unrecognized wide instruction", start_PC);
}
}
break;
type atype = check_class_constant (get_ushort ());
int dim = get_byte ();
if (dim < 1)
- verify_fail ("too few dimensions to multianewarray");
+ verify_fail ("too few dimensions to multianewarray", start_PC);
atype.verify_dimensions (dim);
for (int i = 0; i < dim; ++i)
pop_type (int_type);
default:
// Unrecognized opcode.
- verify_fail ("unrecognized instruction in verify_instructions_0");
+ verify_fail ("unrecognized instruction in verify_instructions_0",
+ start_PC);
}
}
}
_Jv_BytecodeVerifier (_Jv_InterpMethod *m)
{
+ // We just print the text as utf-8. This is just for debugging
+ // anyway.
+ debug_print ("--------------------------------\n");
+ debug_print ("-- Verifying method `%s'\n", m->self->name->data);
+
current_method = m;
bytecode = m->bytecode ();
exception = m->exceptions ();
states = NULL;
flags = NULL;
jsr_ptrs = NULL;
+ utf8_list = NULL;
}
~_Jv_BytecodeVerifier ()
_Jv_Free (flags);
if (jsr_ptrs)
_Jv_Free (jsr_ptrs);
+ while (utf8_list != NULL)
+ {
+ linked_utf8 *n = utf8_list->next;
+ _Jv_Free (utf8_list->val);
+ _Jv_Free (utf8_list);
+ utf8_list = n;
+ }
}
};
// FIXME: add more info, like PC, when required.
static void
-verify_fail (char *s)
+verify_fail (char *s, jint pc)
{
- char buf[1024];
- strcpy (buf, "verification failed: ");
- strcat (buf, s);
- throw new java::lang::VerifyError (JvNewStringLatin1 (buf));
+ using namespace java::lang;
+ StringBuffer *buf = new StringBuffer ();
+
+ buf->append (JvNewStringLatin1 ("verification failed"));
+ if (pc != -1)
+ {
+ buf->append (JvNewStringLatin1 (" at PC "));
+ buf->append (pc);
+ }
+ buf->append (JvNewStringLatin1 (": "));
+ buf->append (JvNewStringLatin1 (s));
+ throw new java::lang::VerifyError (buf->toString ());
}
#endif /* INTERPRETER */
OSDN Git Service
