// defineclass.cc - defining a class from .class format.
-/* Copyright (C) 2001 Free Software Foundation
+/* Copyright (C) 2001, 2002 Free Software Foundation
This file is part of libgcj.
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/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
-// * class loader madness
-// * Lots and lots of debugging and testing
-// * type representation is still ugly. look for the big switches
-// * at least one GC problem :-(
+static void debug_print (const char *fmt, ...)
+ __attribute__ ((format (printf, 1, 2)));
-// This is global because __attribute__ doesn't seem to work on static
-// methods.
-static void verify_fail (char *msg, jint pc = -1)
- __attribute__ ((__noreturn__));
+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 subr_entry_info;
+ struct linked_utf8;
// The current PC.
int PC;
// of all calling `jsr's at at each jsr target.
subr_info **jsr_ptrs;
- // The current top of the stack, in terms of slots.
- int stacktop;
- // The current depth of the stack. This will be larger than
- // STACKTOP when wide types are on the stack.
- int stackdepth;
+ // We keep a linked list of entries which map each `ret' instruction
+ // to its unique subroutine entry point. We expect that there won't
+ // be many `ret' instructions, so a linked list is ok.
+ subr_entry_info *entry_points;
// The bytecode itself.
unsigned char *bytecode;
// 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,
// Return the type_val corresponding to a primitive signature
// character. For instance `I' returns `int.class'.
- static type_val get_type_val_for_signature (jchar sig)
+ type_val get_type_val_for_signature (jchar sig)
{
type_val rt;
switch (sig)
}
// Return the type_val corresponding to a primitive class.
- static type_val get_type_val_for_signature (jclass k)
+ type_val get_type_val_for_signature (jclass k)
{
return get_type_val_for_signature ((jchar) k->method_count);
}
if (target->isPrimitive () || source->isPrimitive ())
return false;
- // _Jv_IsAssignableFrom can handle a target which is an
- // interface even if it hasn't been prepared.
- if ((target->state > JV_STATE_LINKED || target->isInterface ())
- && source->state > JV_STATE_LINKED)
- return _Jv_IsAssignableFrom (target, source);
-
if (target->isArray ())
{
if (! source->isArray ())
if (is_assignable_from_slow (target, source->interfaces[i]))
return true;
}
- return false;
+ source = source->getSuperclass ();
+ if (source == NULL)
+ return false;
}
+ // We must do this check before we check to see if SOURCE is
+ // an interface. This way we know that any interface is
+ // assignable to an Object.
else if (target == &java::lang::Object::class$)
return true;
- else if (source->isInterface ()
- || source == &java::lang::Object::class$)
+ else if (source->isInterface ())
+ {
+ for (int i = 0; i < target->interface_count; ++i)
+ {
+ // We use a recursive call because we also need to
+ // check superinterfaces.
+ if (is_assignable_from_slow (target->interfaces[i], source))
+ return true;
+ }
+ target = target->getSuperclass ();
+ if (target == NULL)
+ return false;
+ }
+ else if (source == &java::lang::Object::class$)
return false;
else
source = source->getSuperclass ();
subr_info *next;
};
+ // This is used to keep track of which subroutine entry point
+ // corresponds to which `ret' instruction.
+ struct subr_entry_info
+ {
+ // PC of the subroutine entry point.
+ int pc;
+ // PC of the `ret' instruction.
+ int ret_pc;
+ // Link.
+ subr_entry_info *next;
+ };
+
// The `type' class is used to represent a single type in the
// verifier.
struct type
}
// If *THIS is an unresolved reference type, resolve it.
- void resolve ()
+ void resolve (_Jv_BytecodeVerifier *verifier)
{
if (key != unresolved_reference_type
&& key != uninitialized_unresolved_reference_type)
return;
- // FIXME: class loader
using namespace java::lang;
+ java::lang::ClassLoader *loader
+ = verifier->current_class->getClassLoaderInternal();
// We might see either kind of name. Sigh.
if (data.name->data[0] == 'L'
&& data.name->data[data.name->length - 1] == ';')
- data.klass = _Jv_FindClassFromSignature (data.name->data, NULL);
+ data.klass = _Jv_FindClassFromSignature (data.name->data, loader);
else
data.klass = Class::forName (_Jv_NewStringUtf8Const (data.name),
- false, NULL);
+ false, loader);
key = (key == unresolved_reference_type
? reference_type
: uninitialized_reference_type);
}
// Mark this type as the uninitialized result of `new'.
- void set_uninitialized (int npc)
+ void set_uninitialized (int npc, _Jv_BytecodeVerifier *verifier)
{
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");
+ verifier->verify_fail ("internal error in type::uninitialized");
pc = npc;
}
// of type *THIS. Handle various special cases too. Might modify
// *THIS or K. Note however that this does not perform numeric
// promotion.
- bool compatible (type &k)
+ bool compatible (type &k, _Jv_BytecodeVerifier *verifier)
{
// Any type is compatible with the unsuitable type.
if (key == unsuitable_type)
if (key < reference_type || k.key < reference_type)
return key == k.key;
- // The `null' type is convertible to any reference type.
- // FIXME: is this correct for THIS?
+ // The `null' type is convertible to any initialized reference
+ // type.
if (key == null_type || k.key == null_type)
return true;
return true;
// We must resolve both types and check assignability.
- resolve ();
- k.resolve ();
+ resolve (verifier);
+ k.resolve (verifier);
return is_assignable_from_slow (data.klass, k.data.klass);
}
return false;
}
- bool isinterface ()
+ bool isnull () const
+ {
+ return key == null_type;
+ }
+
+ bool isinterface (_Jv_BytecodeVerifier *verifier)
{
- resolve ();
+ resolve (verifier);
if (key != reference_type)
return false;
return data.klass->isInterface ();
}
- bool isabstract ()
+ bool isabstract (_Jv_BytecodeVerifier *verifier)
{
- resolve ();
+ resolve (verifier);
if (key != reference_type)
return false;
using namespace java::lang::reflect;
}
// Return the element type of an array.
- type element_type ()
+ type element_type (_Jv_BytecodeVerifier *verifier)
{
// FIXME: maybe should do string manipulation here.
- resolve ();
+ resolve (verifier);
if (key != reference_type)
- verify_fail ("programmer error in type::element_type()");
+ verifier->verify_fail ("programmer error in type::element_type()", -1);
jclass k = data.klass->getComponentType ();
if (k->isPrimitive ())
- return type (get_type_val_for_signature (k));
+ return type (verifier->get_type_val_for_signature (k));
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 ()
+ type to_array (_Jv_BytecodeVerifier *verifier)
{
// Resolving isn't ideal, because it might force us to load
// another class, but it's easy. FIXME?
if (key == unresolved_reference_type)
- resolve ();
+ resolve (verifier);
if (key == reference_type)
return type (_Jv_GetArrayClass (data.klass,
- data.klass->getClassLoader ()));
+ data.klass->getClassLoaderInternal()));
else
- verify_fail ("internal error in type::to_array()");
+ verifier->verify_fail ("internal error in type::to_array()");
}
bool isreference () const
|| key == uninitialized_reference_type);
}
- void verify_dimensions (int ndims)
+ void verify_dimensions (int ndims, _Jv_BytecodeVerifier *verifier)
{
// The way this is written, we don't need to check isarray().
if (key == reference_type)
}
if (ndims > 0)
- verify_fail ("array type has fewer dimensions than required");
+ verifier->verify_fail ("array type has fewer dimensions than required");
}
// Merge OLD_TYPE into this. On error throw exception.
- bool merge (type& old_type, bool local_semantics = false)
+ bool merge (type& old_type, bool local_semantics,
+ _Jv_BytecodeVerifier *verifier)
{
bool changed = false;
bool refo = old_type.isreference ();
changed = true;
}
else if (isinitialized () != old_type.isinitialized ())
- verify_fail ("merging initialized and uninitialized types");
+ verifier->verify_fail ("merging initialized and uninitialized types");
else
{
if (! isinitialized ())
else if (old_type.pc == UNINIT)
;
else if (pc != old_type.pc)
- verify_fail ("merging different uninitialized types");
+ verifier->verify_fail ("merging different uninitialized types");
}
if (! isresolved ()
}
else
{
- resolve ();
- old_type.resolve ();
+ resolve (verifier);
+ old_type.resolve (verifier);
jclass k = data.klass;
jclass oldk = old_type.data.klass;
oldk = oldk->getComponentType ();
}
- // This loop will end when we hit Object.
- while (true)
+ // Ordinarily this terminates when we hit Object...
+ while (k != NULL)
{
if (is_assignable_from_slow (k, oldk))
break;
k = k->getSuperclass ();
changed = true;
}
+ // ... but K could have been an interface, in which
+ // case we'll end up here. We just convert this
+ // into Object.
+ if (k == NULL)
+ k = &java::lang::Object::class$;
if (changed)
{
while (arraycount > 0)
{
- // FIXME: Class loader.
- k = _Jv_GetArrayClass (k, NULL);
+ java::lang::ClassLoader *loader
+ = verifier->current_class->getClassLoaderInternal();
+ k = _Jv_GetArrayClass (k, loader);
--arraycount;
}
data.klass = k;
{
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");
+ verifier->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
// location.
struct state
{
- // Current top of stack.
+ // The current top of the stack, in terms of slots.
int stacktop;
- // Current stack depth. This is like the top of stack but it
- // includes wide variable information.
+ // The current depth of the stack. This will be larger than
+ // STACKTOP when wide types are on the stack.
int stackdepth;
// The stack.
type *stack;
// 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.
// NO_NEXT marks the state at the end of the reverification list.
static const int NO_NEXT = -2;
+ // This is used to mark the stack depth at the instruction just
+ // after a `jsr' when we haven't yet processed the corresponding
+ // `ret'. See handle_jsr_insn for more information.
+ static const int NO_STACK = -1;
+
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'.
}
stack[0] = t;
for (int i = stacktop; i < max_stack; ++i)
stack[i] = unsuitable_type;
+ }
- // FIXME: subroutine handling?
+ // Modify this state to reflect entry into a subroutine.
+ void enter_subroutine (int npc, int max_locals)
+ {
+ subroutine = npc;
+ // Mark all items as unchanged. Each subroutine needs to keep
+ // track of its `changed' state independently. In the case of
+ // nested subroutines, this information will be merged back into
+ // parent by the `ret'.
+ for (int i = 0; i < max_locals; ++i)
+ local_changed[i] = false;
}
- // 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)
+ int max_locals, _Jv_BytecodeVerifier *verifier)
{
bool changed = false;
- // Merge subroutine states. *THIS and *STATE_OLD must be in the
- // same subroutine. Also, recursive subroutine calls must be
- // avoided.
+ // 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. Here we just keep track of what
+ // subroutine we think we're in. We only check for a merge
+ // (which is invalid) when we see a `ret'.
if (subroutine == state_old->subroutine)
{
// Nothing.
changed = true;
}
else
- verify_fail ("subroutines merged");
+ {
+ // If the subroutines differ, indicate that the state
+ // changed. This is needed to detect when subroutines have
+ // merged.
+ changed = true;
+ }
- // Merge stacks.
- if (state_old->stacktop != stacktop)
- verify_fail ("stack sizes differ");
- for (int i = 0; i < state_old->stacktop; ++i)
+ // Merge stacks. Special handling for NO_STACK case.
+ if (state_old->stacktop == NO_STACK)
{
- if (stack[i].merge (state_old->stack[i]))
- changed = true;
+ // Nothing to do in this case; we don't care about modifying
+ // the old state.
+ }
+ else if (stacktop == NO_STACK)
+ {
+ stacktop = state_old->stacktop;
+ stackdepth = state_old->stackdepth;
+ for (int i = 0; i < stacktop; ++i)
+ stack[i] = state_old->stack[i];
+ changed = true;
+ }
+ else if (state_old->stacktop != stacktop)
+ verifier->verify_fail ("stack sizes differ");
+ else
+ {
+ for (int i = 0; i < state_old->stacktop; ++i)
+ {
+ if (stack[i].merge (state_old->stack[i], false, verifier))
+ changed = true;
+ }
}
// Merge local variables.
for (int i = 0; i < max_locals; ++i)
{
- if (! ret_semantics || local_changed[i])
+ // If we're not processing a `ret', then we merge every
+ // local variable. If we are processing a `ret', then we
+ // only merge locals which changed in the subroutine. When
+ // processing a `ret', STATE_OLD is the state at the point
+ // of the `ret', and THIS is the state just after the `jsr'.
+ if (! ret_semantics || state_old->local_changed[i])
{
- if (locals[i].merge (state_old->locals[i], true))
+ if (locals[i].merge (state_old->locals[i], true, verifier))
{
+ // Note that we don't call `note_variable' here.
+ // This change doesn't represent a real change to a
+ // local, but rather a merge artifact. If we're in
+ // a subroutine which is called with two
+ // incompatible types in a slot that is unused by
+ // the subroutine, then we don't want to mark that
+ // variable as having been modified.
changed = true;
- note_variable (i);
}
}
// whether we're using backwards-branch or exception-handing
// semantics.
void check_no_uninitialized_objects (int max_locals,
+ _Jv_BytecodeVerifier *verifier,
bool exception_semantics = false)
{
if (! exception_semantics)
{
for (int i = 0; i < stacktop; ++i)
if (stack[i].isreference () && ! stack[i].isinitialized ())
- verify_fail ("uninitialized object on stack");
+ verifier->verify_fail ("uninitialized object on stack");
}
for (int i = 0; i < max_locals; ++i)
if (locals[i].isreference () && ! locals[i].isinitialized ())
- verify_fail ("uninitialized object in local variable");
+ verifier->verify_fail ("uninitialized object in local variable");
+
+ check_this_initialized (verifier);
+ }
+
+ // Ensure that `this' has been initialized.
+ void check_this_initialized (_Jv_BytecodeVerifier *verifier)
+ {
+ if (this_type.isreference () && ! this_type.isinitialized ())
+ verifier->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 accessed or modified.
+ // 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
+ {
+ if (stacktop == NO_STACK)
+ return true;
+ 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 (local_changed[i] ? "+" : " ");
+ }
+ if (subroutine == 0)
+ debug_print (" | None");
+ else
+ debug_print (" | %4d", subroutine);
+ 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", start_PC);
+ verify_fail ("stack empty");
type r = current_state->stack[--current_state->stacktop];
current_state->stackdepth -= r.depth ();
if (current_state->stackdepth < 0)
{
type r = pop_raw ();
if (r.iswide ())
- verify_fail ("narrow pop of wide type", start_PC);
+ verify_fail ("narrow pop of wide type");
return r;
}
{
type r = pop_raw ();
if (! r.iswide ())
- verify_fail ("wide pop of narrow type", start_PC);
+ verify_fail ("wide pop of narrow type");
return r;
}
{
match.promote ();
type t = pop_raw ();
- if (! match.compatible (t))
- verify_fail ("incompatible type on stack", start_PC);
+ if (! match.compatible (t, this))
+ verify_fail ("incompatible type on stack");
+ return t;
+ }
+
+ // Pop a reference which is guaranteed to be initialized. MATCH
+ // doesn't have to be a reference type; in this case this acts like
+ // pop_type.
+ type pop_init_ref (type match)
+ {
+ type t = pop_raw ();
+ if (t.isreference () && ! t.isinitialized ())
+ verify_fail ("initialized reference required");
+ else if (! match.compatible (t, this))
+ verify_fail ("incompatible type on stack");
+ 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");
return t;
}
{
int depth = t.depth ();
if (index > current_method->max_locals - depth)
- verify_fail ("invalid local variable", start_PC);
- if (! t.compatible (current_state->locals[index]))
- verify_fail ("incompatible type in local variable", start_PC);
+ verify_fail ("invalid local variable");
+ if (! t.compatible (current_state->locals[index], this))
+ verify_fail ("incompatible type in local variable");
if (depth == 2)
{
type t (continuation_type);
- if (! current_state->locals[index + 1].compatible (t))
- verify_fail ("invalid local variable", start_PC);
+ if (! current_state->locals[index + 1].compatible (t, this))
+ verify_fail ("invalid local variable");
}
- current_state->note_variable (index);
return current_state->locals[index];
}
// compatible with type ELEMENT. Returns the actual element type.
type require_array_type (type array, type element)
{
+ // An odd case. Here we just pretend that everything went ok. If
+ // the requested element type is some kind of reference, return
+ // the null type instead.
+ if (array.isnull ())
+ return element.isreference () ? type (null_type) : element;
+
if (! array.isarray ())
verify_fail ("array required");
- type t = array.element_type ();
- if (! element.compatible (t))
+ type t = array.element_type (this);
+ if (! element.compatible (t, this))
{
// Special case for byte arrays, which must also be boolean
// arrays.
if (element.key == byte_type)
{
type e2 (boolean_type);
- ok = e2.compatible (t);
+ ok = e2.compatible (t, this);
}
if (! ok)
verify_fail ("incompatible array element type");
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, this);
+ 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, this);
push_jump_merge (npc, current_state);
}
void push_exception_jump (type t, int pc)
{
- current_state->check_no_uninitialized_objects (current_method->max_stack,
- true);
+ current_state->check_no_uninitialized_objects (current_method->max_locals,
+ this, true);
state s (current_state, current_method->max_stack,
current_method->max_locals);
+ if (current_method->max_stack < 1)
+ verify_fail ("stack overflow at exception handler");
s.set_exception (t, current_method->max_stack);
push_jump_merge (pc, &s);
}
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;
+
+ // Note that we might have gotten here even when there are
+ // remaining states to process. That can happen if we find a
+ // `jsr' without a `ret'.
+ 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;
}
}
if (csub == 0)
verify_fail ("no subroutine");
+ // Check to see if we've merged subroutines.
+ subr_entry_info *entry;
+ for (entry = entry_points; entry != NULL; entry = entry->next)
+ {
+ if (entry->ret_pc == start_PC)
+ break;
+ }
+ if (entry == NULL)
+ {
+ entry = (subr_entry_info *) _Jv_Malloc (sizeof (subr_entry_info));
+ entry->pc = csub;
+ entry->ret_pc = start_PC;
+ entry->next = entry_points;
+ entry_points = entry;
+ }
+ else if (entry->pc != csub)
+ verify_fail ("subroutines merged");
+
for (subr_info *subr = jsr_ptrs[csub]; subr != NULL; subr = subr->next)
{
// Temporarily modify the current state so it looks like we're
// 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, this);
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, this);
check_nonrecursive_call (current_state->subroutine, npc);
- // Temporarily modify the current state so that it looks like we are
- // in the subroutine.
+ // Modify our state as appropriate for entry into a subroutine.
push_type (return_address_type);
- int save = current_state->subroutine;
- current_state->subroutine = npc;
-
- // Merge into the subroutine.
push_jump_merge (npc, current_state);
-
- // Undo our modifications.
- current_state->subroutine = save;
+ // Clean up.
pop_type (return_address_type);
+
+ // On entry to the subroutine, the subroutine number must be set
+ // and the locals must be marked as cleared. We do this after
+ // merging state so that we don't erroneously "notice" a variable
+ // change merely on entry.
+ states[npc]->enter_subroutine (npc, current_method->max_locals);
+
+ // Indicate that we don't know the stack depth of the instruction
+ // following the `jsr'. The idea here is that we need to merge
+ // the local variable state across the jsr, but the subroutine
+ // might change the stack depth, so we can't make any assumptions
+ // about it. So we have yet another special case. We know that
+ // at this point PC points to the instruction after the jsr.
+
+ // FIXME: what if we have a jsr at the end of the code, but that
+ // jsr has no corresponding ret? Is this verifiable, or is it
+ // not? If it is then we need a special case here.
+ if (PC >= current_method->code_length)
+ verify_fail ("fell off end");
+
+ current_state->stacktop = state::NO_STACK;
+ push_jump_merge (PC, current_state);
+ invalidate_pc ();
}
jclass construct_primitive_array_type (type_val prim)
case long_type:
k = JvPrimClass (long);
break;
+
+ // These aren't used here but we call them out to avoid
+ // warnings.
+ case void_type:
+ case unsuitable_type:
+ case return_address_type:
+ case continuation_type:
+ case unused_by_subroutine_type:
+ case reference_type:
+ case null_type:
+ case unresolved_reference_type:
+ case uninitialized_reference_type:
+ case uninitialized_unresolved_reference_type:
default:
verify_fail ("unknown type in construct_primitive_array_type");
}
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_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_wide:
{
- opcode = get_byte ();
+ opcode = (java_opcode) get_byte ();
get_short ();
- if (opcode == (unsigned char) op_iinc)
+ if (opcode == op_iinc)
get_short ();
}
break;
note_branch_target (compute_jump (get_int ()), last_was_jsr);
break;
+ // These are unused here, but we call them out explicitly
+ // so that -Wswitch-enum doesn't complain.
+ case op_putfield_1:
+ case op_putfield_2:
+ case op_putfield_4:
+ case op_putfield_8:
+ case op_putfield_a:
+ case op_putstatic_1:
+ case op_putstatic_2:
+ case op_putstatic_4:
+ case op_putstatic_8:
+ case op_putstatic_a:
+ case op_getfield_1:
+ case op_getfield_2s:
+ case op_getfield_2u:
+ case op_getfield_4:
+ case op_getfield_8:
+ case op_getfield_a:
+ case op_getstatic_1:
+ case op_getstatic_2s:
+ case op_getstatic_2u:
+ case op_getstatic_4:
+ case op_getstatic_8:
+ case op_getstatic_a:
default:
verify_fail ("unrecognized instruction in branch_prepass",
start_PC);
// Verify exception handlers.
for (int i = 0; i < current_method->exc_count; ++i)
{
- if (! (flags[exception[i].handler_pc] & FLAG_INSN_START))
+ if (! (flags[exception[i].handler_pc.i] & FLAG_INSN_START))
verify_fail ("exception handler not at instruction start",
- exception[i].handler_pc);
- if (exception[i].start_pc > exception[i].end_pc)
- verify_fail ("exception range inverted");
- if (! (flags[exception[i].start_pc] & FLAG_INSN_START))
+ exception[i].handler_pc.i);
+ if (! (flags[exception[i].start_pc.i] & FLAG_INSN_START))
verify_fail ("exception start not at instruction start",
- exception[i].start_pc);
- else if (! (flags[exception[i].end_pc] & FLAG_INSN_START))
+ exception[i].start_pc.i);
+ if (exception[i].end_pc.i != current_method->code_length
+ && ! (flags[exception[i].end_pc.i] & FLAG_INSN_START))
verify_fail ("exception end not at instruction start",
- exception[i].end_pc);
+ exception[i].end_pc.i);
- flags[exception[i].handler_pc] |= FLAG_BRANCH_TARGET;
+ flags[exception[i].handler_pc.i] |= FLAG_BRANCH_TARGET;
}
}
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);
}
void check_return_type (type onstack)
{
type rt = compute_return_type (current_method->self->signature);
- if (! rt.compatible (onstack))
- verify_fail ("incompatible return type", start_PC);
+ if (! rt.compatible (onstack, this))
+ verify_fail ("incompatible return type");
+ }
+
+ // 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, this);
+ 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 ()
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;
- }
-
- // 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;
- }
- }
+ // 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", start_PC);
+ verify_fail ("can't happen: saw state::INVALID");
if (PC == state::NO_NEXT)
break;
+ debug_print ("== State pop from pending list\n");
// 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, this)
+ && ! 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.
for (int i = 0; i < current_method->exc_count; ++i)
{
- if (PC >= exception[i].start_pc && PC < exception[i].end_pc)
+ if (PC >= exception[i].start_pc.i && PC < exception[i].end_pc.i)
{
- type handler = reference_type;
- if (exception[i].handler_type != 0)
- handler = check_class_constant (exception[i].handler_type);
- push_exception_jump (handler, exception[i].handler_pc);
+ type handler (&java::lang::Throwable::class$);
+ if (exception[i].handler_type.i != 0)
+ handler = check_class_constant (exception[i].handler_type.i);
+ push_exception_jump (handler, exception[i].handler_pc.i);
}
}
- 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:
break;
case op_iaload:
pop_type (int_type);
- push_type (require_array_type (pop_type (reference_type),
+ push_type (require_array_type (pop_init_ref (reference_type),
int_type));
break;
case op_laload:
pop_type (int_type);
- push_type (require_array_type (pop_type (reference_type),
+ push_type (require_array_type (pop_init_ref (reference_type),
long_type));
break;
case op_faload:
pop_type (int_type);
- push_type (require_array_type (pop_type (reference_type),
+ push_type (require_array_type (pop_init_ref (reference_type),
float_type));
break;
case op_daload:
pop_type (int_type);
- push_type (require_array_type (pop_type (reference_type),
+ push_type (require_array_type (pop_init_ref (reference_type),
double_type));
break;
case op_aaload:
pop_type (int_type);
- push_type (require_array_type (pop_type (reference_type),
+ push_type (require_array_type (pop_init_ref (reference_type),
reference_type));
break;
case op_baload:
pop_type (int_type);
- require_array_type (pop_type (reference_type), byte_type);
+ require_array_type (pop_init_ref (reference_type), byte_type);
push_type (int_type);
break;
case op_caload:
pop_type (int_type);
- require_array_type (pop_type (reference_type), char_type);
+ require_array_type (pop_init_ref (reference_type), char_type);
push_type (int_type);
break;
case op_saload:
pop_type (int_type);
- require_array_type (pop_type (reference_type), short_type);
+ require_array_type (pop_init_ref (reference_type), short_type);
push_type (int_type);
break;
case op_istore:
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);
pop_type (int_type);
- require_array_type (pop_type (reference_type), int_type);
+ require_array_type (pop_init_ref (reference_type), int_type);
break;
case op_lastore:
pop_type (long_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), long_type);
+ require_array_type (pop_init_ref (reference_type), long_type);
break;
case op_fastore:
pop_type (float_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), float_type);
+ require_array_type (pop_init_ref (reference_type), float_type);
break;
case op_dastore:
pop_type (double_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), double_type);
+ require_array_type (pop_init_ref (reference_type), double_type);
break;
case op_aastore:
pop_type (reference_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), reference_type);
+ require_array_type (pop_init_ref (reference_type), reference_type);
break;
case op_bastore:
pop_type (int_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), byte_type);
+ require_array_type (pop_init_ref (reference_type), byte_type);
break;
case op_castore:
pop_type (int_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), char_type);
+ require_array_type (pop_init_ref (reference_type), char_type);
break;
case op_sastore:
pop_type (int_type);
pop_type (int_type);
- require_array_type (pop_type (reference_type), short_type);
+ require_array_type (pop_init_ref (reference_type), short_type);
break;
case op_pop:
pop32 ();
push_type (t);
push_type (t2);
}
+ else
+ push_type (t);
push_type (t);
}
break;
break;
case op_dup2_x2:
{
- // FIXME
type t1 = pop_raw ();
if (t1.iswide ())
{
invalidate_pc ();
break;
case op_areturn:
- check_return_type (pop_type (reference_type));
+ check_return_type (pop_init_ref (reference_type));
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 (this);
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, this);
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)
+ // NARGS is only used when we're processing
+ // invokeinterface. It is simplest for us to compute it
+ // here and then verify it later.
+ int nargs = 0;
+ if (opcode == op_invokeinterface)
{
- int nargs = get_byte ();
- if (nargs == 0)
- verify_fail ("too few arguments to invokeinterface",
- start_PC);
+ nargs = get_byte ();
if (get_byte () != 0)
- verify_fail ("invokeinterface dummy byte is wrong",
- start_PC);
- if (nargs - 1 != arg_count)
- verify_fail ("wrong argument count for invokeinterface",
- start_PC);
+ verify_fail ("invokeinterface dummy byte is wrong");
}
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>", start_PC);
+ if (opcode != op_invokespecial)
+ verify_fail ("can't invoke <init>");
}
else if (method_name->data[0] == '<')
- verify_fail ("can't invoke method starting with `<'",
- start_PC);
+ verify_fail ("can't invoke method starting with `<'");
// Pop arguments and check types.
+ int arg_count = _Jv_count_arguments (method_signature);
type arg_types[arg_count];
compute_argument_types (method_signature, arg_types);
for (int i = arg_count - 1; i >= 0; --i)
- pop_type (arg_types[i]);
+ {
+ // This is only used for verifying the byte for
+ // invokeinterface.
+ nargs -= arg_types[i].depth ();
+ pop_init_ref (arg_types[i]);
+ }
+
+ if (opcode == op_invokeinterface
+ && nargs != 1)
+ verify_fail ("wrong argument count for invokeinterface");
- if (opcode != (unsigned char) op_invokestatic)
+ if (opcode != op_invokestatic)
{
type t = class_type;
if (is_init)
{
// In this case the PC doesn't matter.
- t.set_uninitialized (type::UNINIT);
+ t.set_uninitialized (type::UNINIT, this);
+ }
+ type raw = pop_raw ();
+ bool ok = false;
+ if (! is_init && ! raw.isinitialized ())
+ {
+ // This is a failure.
+ }
+ else if (is_init && raw.isnull ())
+ {
+ // Another failure.
+ }
+ else if (t.compatible (raw, this))
+ {
+ ok = true;
}
- t = pop_type (t);
+ else if (opcode == op_invokeinterface)
+ {
+ // This is a hack. We might have merged two
+ // items and gotten `Object'. This can happen
+ // because we don't keep track of where merges
+ // come from. This is safe as long as the
+ // interpreter checks interfaces at runtime.
+ type obj (&java::lang::Object::class$);
+ ok = raw.compatible (obj, this);
+ }
+
+ if (! ok)
+ verify_fail ("incompatible type on stack");
+
if (is_init)
- current_state->set_initialized (t.get_pc (),
+ current_state->set_initialized (raw.get_pc (),
current_method->max_locals);
}
case op_new:
{
type t = check_class_constant (get_ushort ());
- if (t.isarray () || t.isinterface () || t.isabstract ())
- verify_fail ("type is array, interface, or abstract",
- start_PC);
- t.set_uninitialized (start_PC);
+ if (t.isarray () || t.isinterface (this) || t.isabstract (this))
+ verify_fail ("type is array, interface, or abstract");
+ t.set_uninitialized (start_PC, this);
push_type (t);
}
break;
break;
case op_anewarray:
pop_type (int_type);
- push_type (check_class_constant (get_ushort ()).to_array ());
+ push_type (check_class_constant (get_ushort ()).to_array (this));
break;
case op_arraylength:
{
- type t = pop_type (reference_type);
- if (! t.isarray ())
- verify_fail ("array type expected", start_PC);
+ type t = pop_init_ref (reference_type);
+ if (! t.isarray () && ! t.isnull ())
+ verify_fail ("array type expected");
push_type (int_type);
}
break;
invalidate_pc ();
break;
case op_checkcast:
- pop_type (reference_type);
+ pop_init_ref (reference_type);
push_type (check_class_constant (get_ushort ()));
break;
case op_instanceof:
- pop_type (reference_type);
+ pop_init_ref (reference_type);
check_class_constant (get_ushort ());
push_type (int_type);
break;
case op_monitorenter:
- pop_type (reference_type);
+ pop_init_ref (reference_type);
break;
case op_monitorexit:
- pop_type (reference_type);
+ pop_init_ref (reference_type);
break;
case op_wide:
{
set_variable (get_ushort (), pop_type (double_type));
break;
case op_astore:
- set_variable (get_ushort (), pop_type (reference_type));
+ set_variable (get_ushort (), pop_init_ref (reference_type));
break;
case op_ret:
handle_ret_insn (get_short ());
int dim = get_byte ();
if (dim < 1)
verify_fail ("too few dimensions to multianewarray", start_PC);
- atype.verify_dimensions (dim);
+ atype.verify_dimensions (dim, this);
for (int i = 0; i < dim; ++i)
pop_type (int_type);
push_type (atype);
handle_jsr_insn (get_int ());
break;
+ // These are unused here, but we call them out explicitly
+ // so that -Wswitch-enum doesn't complain.
+ case op_putfield_1:
+ case op_putfield_2:
+ case op_putfield_4:
+ case op_putfield_8:
+ case op_putfield_a:
+ case op_putstatic_1:
+ case op_putstatic_2:
+ case op_putstatic_4:
+ case op_putstatic_8:
+ case op_putstatic_a:
+ case op_getfield_1:
+ case op_getfield_2s:
+ case op_getfield_2u:
+ case op_getfield_4:
+ case op_getfield_8:
+ case op_getfield_a:
+ case op_getstatic_1:
+ case op_getstatic_2s:
+ case op_getstatic_2u:
+ case op_getstatic_4:
+ case op_getstatic_8:
+ case op_getstatic_a:
default:
// Unrecognized opcode.
verify_fail ("unrecognized instruction in verify_instructions_0",
}
}
+ __attribute__ ((__noreturn__)) void verify_fail (char *s, jint pc = -1)
+ {
+ using namespace java::lang;
+ StringBuffer *buf = new StringBuffer ();
+
+ buf->append (JvNewStringLatin1 ("verification failed"));
+ if (pc == -1)
+ pc = start_PC;
+ if (pc != -1)
+ {
+ buf->append (JvNewStringLatin1 (" at PC "));
+ buf->append (pc);
+ }
+
+ _Jv_InterpMethod *method = current_method;
+ buf->append (JvNewStringLatin1 (" in "));
+ buf->append (current_class->getName());
+ buf->append ((jchar) ':');
+ buf->append (JvNewStringUTF (method->get_method()->name->data));
+ buf->append ((jchar) '(');
+ buf->append (JvNewStringUTF (method->get_method()->signature->data));
+ buf->append ((jchar) ')');
+
+ buf->append (JvNewStringLatin1 (": "));
+ buf->append (JvNewStringLatin1 (s));
+ throw new java::lang::VerifyError (buf->toString ());
+ }
+
public:
void verify_instructions ()
_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;
+ entry_points = NULL;
}
~_Jv_BytecodeVerifier ()
_Jv_Free (states);
if (flags)
_Jv_Free (flags);
+
if (jsr_ptrs)
- _Jv_Free (jsr_ptrs);
+ {
+ for (int i = 0; i < current_method->code_length; ++i)
+ {
+ if (jsr_ptrs[i] != NULL)
+ {
+ subr_info *info = jsr_ptrs[i];
+ while (info != NULL)
+ {
+ subr_info *next = info->next;
+ _Jv_Free (info);
+ info = next;
+ }
+ }
+ }
+ _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;
+ }
+
+ while (entry_points != NULL)
+ {
+ subr_entry_info *next = entry_points->next;
+ _Jv_Free (entry_points);
+ entry_points = next;
+ }
}
};
_Jv_BytecodeVerifier v (meth);
v.verify_instructions ();
}
-
-// FIXME: add more info, like PC, when required.
-static void
-verify_fail (char *s, jint pc)
-{
- 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
