gcc/java:
[official-gcc.git] / libjava / verify.cc
blob8b9cfcc759fae5dc8dc828087529535f9e068e30
1 // verify.cc - verify bytecode
3 /* Copyright (C) 2001, 2002, 2003, 2004, 2005 Free Software Foundation
5 This file is part of libgcj.
7 This software is copyrighted work licensed under the terms of the
8 Libgcj License. Please consult the file "LIBGCJ_LICENSE" for
9 details. */
11 // Written by Tom Tromey <tromey@redhat.com>
13 // Define VERIFY_DEBUG to enable debugging output.
15 #include <config.h>
17 #include <jvm.h>
18 #include <gcj/cni.h>
19 #include <java-insns.h>
20 #include <java-interp.h>
22 // On Solaris 10/x86, <signal.h> indirectly includes <ia32/sys/reg.h>, which
23 // defines PC since g++ predefines __EXTENSIONS__. Undef here to avoid clash
24 // with PC member of class _Jv_BytecodeVerifier below.
25 #undef PC
27 #ifdef INTERPRETER
29 #include <java/lang/Class.h>
30 #include <java/lang/VerifyError.h>
31 #include <java/lang/Throwable.h>
32 #include <java/lang/reflect/Modifier.h>
33 #include <java/lang/StringBuffer.h>
34 #include <java/lang/NoClassDefFoundError.h>
36 #ifdef VERIFY_DEBUG
37 #include <stdio.h>
38 #endif /* VERIFY_DEBUG */
41 // This is used to mark states which are not scheduled for
42 // verification.
43 #define INVALID_STATE ((state *) -1)
45 static void debug_print (const char *fmt, ...)
46 __attribute__ ((format (printf, 1, 2)));
48 static inline void
49 debug_print (MAYBE_UNUSED const char *fmt, ...)
51 #ifdef VERIFY_DEBUG
52 va_list ap;
53 va_start (ap, fmt);
54 vfprintf (stderr, fmt, ap);
55 va_end (ap);
56 #endif /* VERIFY_DEBUG */
59 // This started as a fairly ordinary verifier, and for the most part
60 // it remains so. It works in the obvious way, by modeling the effect
61 // of each opcode as it is encountered. For most opcodes, this is a
62 // straightforward operation.
64 // This verifier does not do type merging. It used to, but this
65 // results in difficulty verifying some relatively simple code
66 // involving interfaces, and it pushed some verification work into the
67 // interpreter.
69 // Instead of merging reference types, when we reach a point where two
70 // flows of control merge, we simply keep the union of reference types
71 // from each branch. Then, when we need to verify a fact about a
72 // reference on the stack (e.g., that it is compatible with the
73 // argument type of a method), we check to ensure that all possible
74 // types satisfy the requirement.
76 // Another area this verifier differs from the norm is in its handling
77 // of subroutines. The JVM specification has some confusing things to
78 // say about subroutines. For instance, it makes claims about not
79 // allowing subroutines to merge and it rejects recursive subroutines.
80 // For the most part these are red herrings; we used to try to follow
81 // these things but they lead to problems. For example, the notion of
82 // "being in a subroutine" is not well-defined: is an exception
83 // handler in a subroutine? If you never execute the `ret' but
84 // instead `goto 1' do you remain in the subroutine?
86 // For clarity on what is really required for type safety, read
87 // "Simple Verification Technique for Complex Java Bytecode
88 // Subroutines" by Alessandro Coglio. Among other things this paper
89 // shows that recursive subroutines are not harmful to type safety.
90 // We implement something similar to what he proposes. Note that this
91 // means that this verifier will accept code that is rejected by some
92 // other verifiers.
94 // For those not wanting to read the paper, the basic observation is
95 // that we can maintain split states in subroutines. We maintain one
96 // state for each calling `jsr'. In other words, we re-verify a
97 // subroutine once for each caller, using the exact types held by the
98 // callers (as opposed to the old approach of merging types and
99 // keeping a bitmap registering what did or did not change). This
100 // approach lets us continue to verify correctly even when a
101 // subroutine is exited via `goto' or `athrow' and not `ret'.
103 // In some other areas the JVM specification is (mildly) incorrect,
104 // so we diverge. For instance, you cannot
105 // violate type safety by allocating an object with `new' and then
106 // failing to initialize it, no matter how one branches or where one
107 // stores the uninitialized reference. See "Improving the official
108 // specification of Java bytecode verification" by Alessandro Coglio.
110 // Note that there's no real point in enforcing that padding bytes or
111 // the mystery byte of invokeinterface must be 0, but we do that
112 // regardless.
114 // The verifier is currently neither completely lazy nor eager when it
115 // comes to loading classes. It tries to represent types by name when
116 // possible, and then loads them when it needs to verify a fact about
117 // the type. Checking types by name is valid because we only use
118 // names which come from the current class' constant pool. Since all
119 // such names are looked up using the same class loader, there is no
120 // danger that we might be fooled into comparing different types with
121 // the same name.
123 // In the future we plan to allow for a completely lazy mode of
124 // operation, where the verifier will construct a list of type
125 // assertions to be checked later.
127 // Some test cases for the verifier live in the "verify" module of the
128 // Mauve test suite. However, some of these are presently
129 // (2004-01-20) believed to be incorrect. (More precisely the notion
130 // of "correct" is not well-defined, and this verifier differs from
131 // others while remaining type-safe.) Some other tests live in the
132 // libgcj test suite.
133 class _Jv_BytecodeVerifier
135 private:
137 static const int FLAG_INSN_START = 1;
138 static const int FLAG_BRANCH_TARGET = 2;
140 struct state;
141 struct type;
142 struct linked_utf8;
143 struct ref_intersection;
145 template<typename T>
146 struct linked
148 T *val;
149 linked<T> *next;
152 // The current PC.
153 int PC;
154 // The PC corresponding to the start of the current instruction.
155 int start_PC;
157 // The current state of the stack, locals, etc.
158 state *current_state;
160 // At each branch target we keep a linked list of all the states we
161 // can process at that point. We'll only have multiple states at a
162 // given PC if they both have different return-address types in the
163 // same stack or local slot. This array is indexed by PC and holds
164 // the list of all such states.
165 linked<state> **states;
167 // We keep a linked list of all the states which we must reverify.
168 // This is the head of the list.
169 state *next_verify_state;
171 // We keep some flags for each instruction. The values are the
172 // FLAG_* constants defined above. This is an array indexed by PC.
173 char *flags;
175 // The bytecode itself.
176 unsigned char *bytecode;
177 // The exceptions.
178 _Jv_InterpException *exception;
180 // Defining class.
181 jclass current_class;
182 // This method.
183 _Jv_InterpMethod *current_method;
185 // A linked list of utf8 objects we allocate.
186 linked<_Jv_Utf8Const> *utf8_list;
188 // A linked list of all ref_intersection objects we allocate.
189 ref_intersection *isect_list;
191 // Create a new Utf-8 constant and return it. We do this to avoid
192 // having our Utf-8 constants prematurely collected.
193 _Jv_Utf8Const *make_utf8_const (char *s, int len)
195 linked<_Jv_Utf8Const> *lu = (linked<_Jv_Utf8Const> *)
196 _Jv_Malloc (sizeof (linked<_Jv_Utf8Const>)
197 + _Jv_Utf8Const::space_needed(s, len));
198 _Jv_Utf8Const *r = (_Jv_Utf8Const *) (lu + 1);
199 r->init(s, len);
200 lu->val = r;
201 lu->next = utf8_list;
202 utf8_list = lu;
204 return r;
207 __attribute__ ((__noreturn__)) void verify_fail (char *s, jint pc = -1)
209 using namespace java::lang;
210 StringBuffer *buf = new StringBuffer ();
212 buf->append (JvNewStringLatin1 ("verification failed"));
213 if (pc == -1)
214 pc = start_PC;
215 if (pc != -1)
217 buf->append (JvNewStringLatin1 (" at PC "));
218 buf->append (pc);
221 _Jv_InterpMethod *method = current_method;
222 buf->append (JvNewStringLatin1 (" in "));
223 buf->append (current_class->getName());
224 buf->append ((jchar) ':');
225 buf->append (method->get_method()->name->toString());
226 buf->append ((jchar) '(');
227 buf->append (method->get_method()->signature->toString());
228 buf->append ((jchar) ')');
230 buf->append (JvNewStringLatin1 (": "));
231 buf->append (JvNewStringLatin1 (s));
232 throw new java::lang::VerifyError (buf->toString ());
235 // This enum holds a list of tags for all the different types we
236 // need to handle. Reference types are treated specially by the
237 // type class.
238 enum type_val
240 void_type,
242 // The values for primitive types are chosen to correspond to values
243 // specified to newarray.
244 boolean_type = 4,
245 char_type = 5,
246 float_type = 6,
247 double_type = 7,
248 byte_type = 8,
249 short_type = 9,
250 int_type = 10,
251 long_type = 11,
253 // Used when overwriting second word of a double or long in the
254 // local variables. Also used after merging local variable states
255 // to indicate an unusable value.
256 unsuitable_type,
257 return_address_type,
258 // This is the second word of a two-word value, i.e., a double or
259 // a long.
260 continuation_type,
262 // Everything after `reference_type' must be a reference type.
263 reference_type,
264 null_type,
265 uninitialized_reference_type
268 // This represents a merged class type. Some verifiers (including
269 // earlier versions of this one) will compute the intersection of
270 // two class types when merging states. However, this loses
271 // critical information about interfaces implemented by the various
272 // classes. So instead we keep track of all the actual classes that
273 // have been merged.
274 struct ref_intersection
276 // Whether or not this type has been resolved.
277 bool is_resolved;
279 // Actual type data.
280 union
282 // For a resolved reference type, this is a pointer to the class.
283 jclass klass;
284 // For other reference types, this it the name of the class.
285 _Jv_Utf8Const *name;
286 } data;
288 // Link to the next reference in the intersection.
289 ref_intersection *ref_next;
291 // This is used to keep track of all the allocated
292 // ref_intersection objects, so we can free them.
293 // FIXME: we should allocate these in chunks.
294 ref_intersection *alloc_next;
296 ref_intersection (jclass klass, _Jv_BytecodeVerifier *verifier)
297 : ref_next (NULL)
299 is_resolved = true;
300 data.klass = klass;
301 alloc_next = verifier->isect_list;
302 verifier->isect_list = this;
305 ref_intersection (_Jv_Utf8Const *name, _Jv_BytecodeVerifier *verifier)
306 : ref_next (NULL)
308 is_resolved = false;
309 data.name = name;
310 alloc_next = verifier->isect_list;
311 verifier->isect_list = this;
314 ref_intersection (ref_intersection *dup, ref_intersection *tail,
315 _Jv_BytecodeVerifier *verifier)
316 : ref_next (tail)
318 is_resolved = dup->is_resolved;
319 data = dup->data;
320 alloc_next = verifier->isect_list;
321 verifier->isect_list = this;
324 bool equals (ref_intersection *other, _Jv_BytecodeVerifier *verifier)
326 if (! is_resolved && ! other->is_resolved
327 && _Jv_equalUtf8Consts (data.name, other->data.name))
328 return true;
329 if (! is_resolved)
330 resolve (verifier);
331 if (! other->is_resolved)
332 other->resolve (verifier);
333 return data.klass == other->data.klass;
336 // Merge THIS type into OTHER, returning the result. This will
337 // return OTHER if all the classes in THIS already appear in
338 // OTHER.
339 ref_intersection *merge (ref_intersection *other,
340 _Jv_BytecodeVerifier *verifier)
342 ref_intersection *tail = other;
343 for (ref_intersection *self = this; self != NULL; self = self->ref_next)
345 bool add = true;
346 for (ref_intersection *iter = other; iter != NULL;
347 iter = iter->ref_next)
349 if (iter->equals (self, verifier))
351 add = false;
352 break;
356 if (add)
357 tail = new ref_intersection (self, tail, verifier);
359 return tail;
362 void resolve (_Jv_BytecodeVerifier *verifier)
364 if (is_resolved)
365 return;
367 using namespace java::lang;
368 java::lang::ClassLoader *loader
369 = verifier->current_class->getClassLoaderInternal();
370 // We might see either kind of name. Sigh.
371 if (data.name->first() == 'L' && data.name->limit()[-1] == ';')
373 data.klass = _Jv_FindClassFromSignature (data.name->chars(), loader);
374 if (data.klass == NULL)
375 throw new java::lang::NoClassDefFoundError(data.name->toString());
377 else
378 data.klass = Class::forName (_Jv_NewStringUtf8Const (data.name),
379 false, loader);
380 is_resolved = true;
383 // See if an object of type OTHER can be assigned to an object of
384 // type *THIS. This might resolve classes in one chain or the
385 // other.
386 bool compatible (ref_intersection *other,
387 _Jv_BytecodeVerifier *verifier)
389 ref_intersection *self = this;
391 for (; self != NULL; self = self->ref_next)
393 ref_intersection *other_iter = other;
395 for (; other_iter != NULL; other_iter = other_iter->ref_next)
397 // Avoid resolving if possible.
398 if (! self->is_resolved
399 && ! other_iter->is_resolved
400 && _Jv_equalUtf8Consts (self->data.name,
401 other_iter->data.name))
402 continue;
404 if (! self->is_resolved)
405 self->resolve(verifier);
406 if (! other_iter->is_resolved)
407 other_iter->resolve(verifier);
409 if (! is_assignable_from_slow (self->data.klass,
410 other_iter->data.klass))
411 return false;
415 return true;
418 bool isarray ()
420 // assert (ref_next == NULL);
421 if (is_resolved)
422 return data.klass->isArray ();
423 else
424 return data.name->first() == '[';
427 bool isinterface (_Jv_BytecodeVerifier *verifier)
429 // assert (ref_next == NULL);
430 if (! is_resolved)
431 resolve (verifier);
432 return data.klass->isInterface ();
435 bool isabstract (_Jv_BytecodeVerifier *verifier)
437 // assert (ref_next == NULL);
438 if (! is_resolved)
439 resolve (verifier);
440 using namespace java::lang::reflect;
441 return Modifier::isAbstract (data.klass->getModifiers ());
444 jclass getclass (_Jv_BytecodeVerifier *verifier)
446 if (! is_resolved)
447 resolve (verifier);
448 return data.klass;
451 int count_dimensions ()
453 int ndims = 0;
454 if (is_resolved)
456 jclass k = data.klass;
457 while (k->isArray ())
459 k = k->getComponentType ();
460 ++ndims;
463 else
465 char *p = data.name->chars();
466 while (*p++ == '[')
467 ++ndims;
469 return ndims;
472 void *operator new (size_t bytes)
474 return _Jv_Malloc (bytes);
477 void operator delete (void *mem)
479 _Jv_Free (mem);
483 // Return the type_val corresponding to a primitive signature
484 // character. For instance `I' returns `int.class'.
485 type_val get_type_val_for_signature (jchar sig)
487 type_val rt;
488 switch (sig)
490 case 'Z':
491 rt = boolean_type;
492 break;
493 case 'B':
494 rt = byte_type;
495 break;
496 case 'C':
497 rt = char_type;
498 break;
499 case 'S':
500 rt = short_type;
501 break;
502 case 'I':
503 rt = int_type;
504 break;
505 case 'J':
506 rt = long_type;
507 break;
508 case 'F':
509 rt = float_type;
510 break;
511 case 'D':
512 rt = double_type;
513 break;
514 case 'V':
515 rt = void_type;
516 break;
517 default:
518 verify_fail ("invalid signature");
520 return rt;
523 // Return the type_val corresponding to a primitive class.
524 type_val get_type_val_for_signature (jclass k)
526 return get_type_val_for_signature ((jchar) k->method_count);
529 // This is like _Jv_IsAssignableFrom, but it works even if SOURCE or
530 // TARGET haven't been prepared.
531 static bool is_assignable_from_slow (jclass target, jclass source)
533 // First, strip arrays.
534 while (target->isArray ())
536 // If target is array, source must be as well.
537 if (! source->isArray ())
538 return false;
539 target = target->getComponentType ();
540 source = source->getComponentType ();
543 // Quick success.
544 if (target == &java::lang::Object::class$)
545 return true;
549 if (source == target)
550 return true;
552 if (target->isPrimitive () || source->isPrimitive ())
553 return false;
555 if (target->isInterface ())
557 for (int i = 0; i < source->interface_count; ++i)
559 // We use a recursive call because we also need to
560 // check superinterfaces.
561 if (is_assignable_from_slow (target, source->getInterface (i)))
562 return true;
565 source = source->getSuperclass ();
567 while (source != NULL);
569 return false;
572 // The `type' class is used to represent a single type in the
573 // verifier.
574 struct type
576 // The type key.
577 type_val key;
579 // For reference types, the representation of the type.
580 ref_intersection *klass;
582 // This is used in two situations.
584 // First, when constructing a new object, it is the PC of the
585 // `new' instruction which created the object. We use the special
586 // value UNINIT to mean that this is uninitialized. The special
587 // value SELF is used for the case where the current method is
588 // itself the <init> method. the special value EITHER is used
589 // when we may optionally allow either an uninitialized or
590 // initialized reference to match.
592 // Second, when the key is return_address_type, this holds the PC
593 // of the instruction following the `jsr'.
594 int pc;
596 static const int UNINIT = -2;
597 static const int SELF = -1;
598 static const int EITHER = -3;
600 // Basic constructor.
601 type ()
603 key = unsuitable_type;
604 klass = NULL;
605 pc = UNINIT;
608 // Make a new instance given the type tag. We assume a generic
609 // `reference_type' means Object.
610 type (type_val k)
612 key = k;
613 // For reference_type, if KLASS==NULL then that means we are
614 // looking for a generic object of any kind, including an
615 // uninitialized reference.
616 klass = NULL;
617 pc = UNINIT;
620 // Make a new instance given a class.
621 type (jclass k, _Jv_BytecodeVerifier *verifier)
623 key = reference_type;
624 klass = new ref_intersection (k, verifier);
625 pc = UNINIT;
628 // Make a new instance given the name of a class.
629 type (_Jv_Utf8Const *n, _Jv_BytecodeVerifier *verifier)
631 key = reference_type;
632 klass = new ref_intersection (n, verifier);
633 pc = UNINIT;
636 // Copy constructor.
637 type (const type &t)
639 key = t.key;
640 klass = t.klass;
641 pc = t.pc;
644 // These operators are required because libgcj can't link in
645 // -lstdc++.
646 void *operator new[] (size_t bytes)
648 return _Jv_Malloc (bytes);
651 void operator delete[] (void *mem)
653 _Jv_Free (mem);
656 type& operator= (type_val k)
658 key = k;
659 klass = NULL;
660 pc = UNINIT;
661 return *this;
664 type& operator= (const type& t)
666 key = t.key;
667 klass = t.klass;
668 pc = t.pc;
669 return *this;
672 // Promote a numeric type.
673 type &promote ()
675 if (key == boolean_type || key == char_type
676 || key == byte_type || key == short_type)
677 key = int_type;
678 return *this;
681 // Mark this type as the uninitialized result of `new'.
682 void set_uninitialized (int npc, _Jv_BytecodeVerifier *verifier)
684 if (key == reference_type)
685 key = uninitialized_reference_type;
686 else
687 verifier->verify_fail ("internal error in type::uninitialized");
688 pc = npc;
691 // Mark this type as now initialized.
692 void set_initialized (int npc)
694 if (npc != UNINIT && pc == npc && key == uninitialized_reference_type)
696 key = reference_type;
697 pc = UNINIT;
701 // Mark this type as a particular return address.
702 void set_return_address (int npc)
704 pc = npc;
707 // Return true if this type and type OTHER are considered
708 // mergeable for the purposes of state merging. This is related
709 // to subroutine handling. For this purpose two types are
710 // considered unmergeable if they are both return-addresses but
711 // have different PCs.
712 bool state_mergeable_p (const type &other) const
714 return (key != return_address_type
715 || other.key != return_address_type
716 || pc == other.pc);
719 // Return true if an object of type K can be assigned to a variable
720 // of type *THIS. Handle various special cases too. Might modify
721 // *THIS or K. Note however that this does not perform numeric
722 // promotion.
723 bool compatible (type &k, _Jv_BytecodeVerifier *verifier)
725 // Any type is compatible with the unsuitable type.
726 if (key == unsuitable_type)
727 return true;
729 if (key < reference_type || k.key < reference_type)
730 return key == k.key;
732 // The `null' type is convertible to any initialized reference
733 // type.
734 if (key == null_type)
735 return k.key != uninitialized_reference_type;
736 if (k.key == null_type)
737 return key != uninitialized_reference_type;
739 // A special case for a generic reference.
740 if (klass == NULL)
741 return true;
742 if (k.klass == NULL)
743 verifier->verify_fail ("programmer error in type::compatible");
745 // Handle the special 'EITHER' case, which is only used in a
746 // special case of 'putfield'. Note that we only need to handle
747 // this on the LHS of a check.
748 if (! isinitialized () && pc == EITHER)
750 // If the RHS is uninitialized, it must be an uninitialized
751 // 'this'.
752 if (! k.isinitialized () && k.pc != SELF)
753 return false;
755 else if (isinitialized () != k.isinitialized ())
757 // An initialized type and an uninitialized type are not
758 // otherwise compatible.
759 return false;
761 else
763 // Two uninitialized objects are compatible if either:
764 // * The PCs are identical, or
765 // * One PC is UNINIT.
766 if (! isinitialized ())
768 if (pc != k.pc && pc != UNINIT && k.pc != UNINIT)
769 return false;
773 return klass->compatible(k.klass, verifier);
776 bool equals (const type &other, _Jv_BytecodeVerifier *vfy)
778 // Only works for reference types.
779 if ((key != reference_type
780 && key != uninitialized_reference_type)
781 || (other.key != reference_type
782 && other.key != uninitialized_reference_type))
783 return false;
784 // Only for single-valued types.
785 if (klass->ref_next || other.klass->ref_next)
786 return false;
787 return klass->equals (other.klass, vfy);
790 bool isvoid () const
792 return key == void_type;
795 bool iswide () const
797 return key == long_type || key == double_type;
800 // Return number of stack or local variable slots taken by this
801 // type.
802 int depth () const
804 return iswide () ? 2 : 1;
807 bool isarray () const
809 // We treat null_type as not an array. This is ok based on the
810 // current uses of this method.
811 if (key == reference_type)
812 return klass->isarray ();
813 return false;
816 bool isnull () const
818 return key == null_type;
821 bool isinterface (_Jv_BytecodeVerifier *verifier)
823 if (key != reference_type)
824 return false;
825 return klass->isinterface (verifier);
828 bool isabstract (_Jv_BytecodeVerifier *verifier)
830 if (key != reference_type)
831 return false;
832 return klass->isabstract (verifier);
835 // Return the element type of an array.
836 type element_type (_Jv_BytecodeVerifier *verifier)
838 if (key != reference_type)
839 verifier->verify_fail ("programmer error in type::element_type()", -1);
841 jclass k = klass->getclass (verifier)->getComponentType ();
842 if (k->isPrimitive ())
843 return type (verifier->get_type_val_for_signature (k));
844 return type (k, verifier);
847 // Return the array type corresponding to an initialized
848 // reference. We could expand this to work for other kinds of
849 // types, but currently we don't need to.
850 type to_array (_Jv_BytecodeVerifier *verifier)
852 if (key != reference_type)
853 verifier->verify_fail ("internal error in type::to_array()");
855 jclass k = klass->getclass (verifier);
856 return type (_Jv_GetArrayClass (k, k->getClassLoaderInternal()),
857 verifier);
860 bool isreference () const
862 return key >= reference_type;
865 int get_pc () const
867 return pc;
870 bool isinitialized () const
872 return key == reference_type || key == null_type;
875 bool isresolved () const
877 return (key == reference_type
878 || key == null_type
879 || key == uninitialized_reference_type);
882 void verify_dimensions (int ndims, _Jv_BytecodeVerifier *verifier)
884 // The way this is written, we don't need to check isarray().
885 if (key != reference_type)
886 verifier->verify_fail ("internal error in verify_dimensions:"
887 " not a reference type");
889 if (klass->count_dimensions () < ndims)
890 verifier->verify_fail ("array type has fewer dimensions"
891 " than required");
894 // Merge OLD_TYPE into this. On error throw exception. Return
895 // true if the merge caused a type change.
896 bool merge (type& old_type, bool local_semantics,
897 _Jv_BytecodeVerifier *verifier)
899 bool changed = false;
900 bool refo = old_type.isreference ();
901 bool refn = isreference ();
902 if (refo && refn)
904 if (old_type.key == null_type)
906 else if (key == null_type)
908 *this = old_type;
909 changed = true;
911 else if (isinitialized () != old_type.isinitialized ())
912 verifier->verify_fail ("merging initialized and uninitialized types");
913 else
915 if (! isinitialized ())
917 if (pc == UNINIT)
918 pc = old_type.pc;
919 else if (old_type.pc == UNINIT)
921 else if (pc != old_type.pc)
922 verifier->verify_fail ("merging different uninitialized types");
925 ref_intersection *merged = old_type.klass->merge (klass,
926 verifier);
927 if (merged != klass)
929 klass = merged;
930 changed = true;
934 else if (refo || refn || key != old_type.key)
936 if (local_semantics)
938 // If we already have an `unsuitable' type, then we
939 // don't need to change again.
940 if (key != unsuitable_type)
942 key = unsuitable_type;
943 changed = true;
946 else
947 verifier->verify_fail ("unmergeable type");
949 return changed;
952 #ifdef VERIFY_DEBUG
953 void print (void) const
955 char c = '?';
956 switch (key)
958 case boolean_type: c = 'Z'; break;
959 case byte_type: c = 'B'; break;
960 case char_type: c = 'C'; break;
961 case short_type: c = 'S'; break;
962 case int_type: c = 'I'; break;
963 case long_type: c = 'J'; break;
964 case float_type: c = 'F'; break;
965 case double_type: c = 'D'; break;
966 case void_type: c = 'V'; break;
967 case unsuitable_type: c = '-'; break;
968 case return_address_type: c = 'r'; break;
969 case continuation_type: c = '+'; break;
970 case reference_type: c = 'L'; break;
971 case null_type: c = '@'; break;
972 case uninitialized_reference_type: c = 'U'; break;
974 debug_print ("%c", c);
976 #endif /* VERIFY_DEBUG */
979 // This class holds all the state information we need for a given
980 // location.
981 struct state
983 // The current top of the stack, in terms of slots.
984 int stacktop;
985 // The current depth of the stack. This will be larger than
986 // STACKTOP when wide types are on the stack.
987 int stackdepth;
988 // The stack.
989 type *stack;
990 // The local variables.
991 type *locals;
992 // We keep track of the type of `this' specially. This is used to
993 // ensure that an instance initializer invokes another initializer
994 // on `this' before returning. We must keep track of this
995 // specially because otherwise we might be confused by code which
996 // assigns to locals[0] (overwriting `this') and then returns
997 // without really initializing.
998 type this_type;
1000 // The PC for this state. This is only valid on states which are
1001 // permanently attached to a given PC. For an object like
1002 // `current_state', which is used transiently, this has no
1003 // meaning.
1004 int pc;
1005 // We keep a linked list of all states requiring reverification.
1006 // If this is the special value INVALID_STATE then this state is
1007 // not on the list. NULL marks the end of the linked list.
1008 state *next;
1010 // NO_NEXT is the PC value meaning that a new state must be
1011 // acquired from the verification list.
1012 static const int NO_NEXT = -1;
1014 state ()
1015 : this_type ()
1017 stack = NULL;
1018 locals = NULL;
1019 next = INVALID_STATE;
1022 state (int max_stack, int max_locals)
1023 : this_type ()
1025 stacktop = 0;
1026 stackdepth = 0;
1027 stack = new type[max_stack];
1028 for (int i = 0; i < max_stack; ++i)
1029 stack[i] = unsuitable_type;
1030 locals = new type[max_locals];
1031 for (int i = 0; i < max_locals; ++i)
1032 locals[i] = unsuitable_type;
1033 pc = NO_NEXT;
1034 next = INVALID_STATE;
1037 state (const state *orig, int max_stack, int max_locals)
1039 stack = new type[max_stack];
1040 locals = new type[max_locals];
1041 copy (orig, max_stack, max_locals);
1042 pc = NO_NEXT;
1043 next = INVALID_STATE;
1046 ~state ()
1048 if (stack)
1049 delete[] stack;
1050 if (locals)
1051 delete[] locals;
1054 void *operator new[] (size_t bytes)
1056 return _Jv_Malloc (bytes);
1059 void operator delete[] (void *mem)
1061 _Jv_Free (mem);
1064 void *operator new (size_t bytes)
1066 return _Jv_Malloc (bytes);
1069 void operator delete (void *mem)
1071 _Jv_Free (mem);
1074 void copy (const state *copy, int max_stack, int max_locals)
1076 stacktop = copy->stacktop;
1077 stackdepth = copy->stackdepth;
1078 for (int i = 0; i < max_stack; ++i)
1079 stack[i] = copy->stack[i];
1080 for (int i = 0; i < max_locals; ++i)
1081 locals[i] = copy->locals[i];
1083 this_type = copy->this_type;
1084 // Don't modify `next' or `pc'.
1087 // Modify this state to reflect entry to an exception handler.
1088 void set_exception (type t, int max_stack)
1090 stackdepth = 1;
1091 stacktop = 1;
1092 stack[0] = t;
1093 for (int i = stacktop; i < max_stack; ++i)
1094 stack[i] = unsuitable_type;
1097 inline int get_pc () const
1099 return pc;
1102 void set_pc (int npc)
1104 pc = npc;
1107 // Merge STATE_OLD into this state. Destructively modifies this
1108 // state. Returns true if the new state was in fact changed.
1109 // Will throw an exception if the states are not mergeable.
1110 bool merge (state *state_old, int max_locals,
1111 _Jv_BytecodeVerifier *verifier)
1113 bool changed = false;
1115 // Special handling for `this'. If one or the other is
1116 // uninitialized, then the merge is uninitialized.
1117 if (this_type.isinitialized ())
1118 this_type = state_old->this_type;
1120 // Merge stacks.
1121 if (state_old->stacktop != stacktop) // FIXME stackdepth instead?
1122 verifier->verify_fail ("stack sizes differ");
1123 for (int i = 0; i < state_old->stacktop; ++i)
1125 if (stack[i].merge (state_old->stack[i], false, verifier))
1126 changed = true;
1129 // Merge local variables.
1130 for (int i = 0; i < max_locals; ++i)
1132 if (locals[i].merge (state_old->locals[i], true, verifier))
1133 changed = true;
1136 return changed;
1139 // Ensure that `this' has been initialized.
1140 void check_this_initialized (_Jv_BytecodeVerifier *verifier)
1142 if (this_type.isreference () && ! this_type.isinitialized ())
1143 verifier->verify_fail ("`this' is uninitialized");
1146 // Set type of `this'.
1147 void set_this_type (const type &k)
1149 this_type = k;
1152 // Mark each `new'd object we know of that was allocated at PC as
1153 // initialized.
1154 void set_initialized (int pc, int max_locals)
1156 for (int i = 0; i < stacktop; ++i)
1157 stack[i].set_initialized (pc);
1158 for (int i = 0; i < max_locals; ++i)
1159 locals[i].set_initialized (pc);
1160 this_type.set_initialized (pc);
1163 // This tests to see whether two states can be considered "merge
1164 // compatible". If both states have a return-address in the same
1165 // slot, and the return addresses are different, then they are not
1166 // compatible and we must not try to merge them.
1167 bool state_mergeable_p (state *other, int max_locals,
1168 _Jv_BytecodeVerifier *verifier)
1170 // This is tricky: if the stack sizes differ, then not only are
1171 // these not mergeable, but in fact we should give an error, as
1172 // we've found two execution paths that reach a branch target
1173 // with different stack depths. FIXME stackdepth instead?
1174 if (stacktop != other->stacktop)
1175 verifier->verify_fail ("stack sizes differ");
1177 for (int i = 0; i < stacktop; ++i)
1178 if (! stack[i].state_mergeable_p (other->stack[i]))
1179 return false;
1180 for (int i = 0; i < max_locals; ++i)
1181 if (! locals[i].state_mergeable_p (other->locals[i]))
1182 return false;
1183 return true;
1186 void reverify (_Jv_BytecodeVerifier *verifier)
1188 if (next == INVALID_STATE)
1190 next = verifier->next_verify_state;
1191 verifier->next_verify_state = this;
1195 #ifdef VERIFY_DEBUG
1196 void print (const char *leader, int pc,
1197 int max_stack, int max_locals) const
1199 debug_print ("%s [%4d]: [stack] ", leader, pc);
1200 int i;
1201 for (i = 0; i < stacktop; ++i)
1202 stack[i].print ();
1203 for (; i < max_stack; ++i)
1204 debug_print (".");
1205 debug_print (" [local] ");
1206 for (i = 0; i < max_locals; ++i)
1207 locals[i].print ();
1208 debug_print (" | %p\n", this);
1210 #else
1211 inline void print (const char *, int, int, int) const
1214 #endif /* VERIFY_DEBUG */
1217 type pop_raw ()
1219 if (current_state->stacktop <= 0)
1220 verify_fail ("stack empty");
1221 type r = current_state->stack[--current_state->stacktop];
1222 current_state->stackdepth -= r.depth ();
1223 if (current_state->stackdepth < 0)
1224 verify_fail ("stack empty", start_PC);
1225 return r;
1228 type pop32 ()
1230 type r = pop_raw ();
1231 if (r.iswide ())
1232 verify_fail ("narrow pop of wide type");
1233 return r;
1236 type pop_type (type match)
1238 match.promote ();
1239 type t = pop_raw ();
1240 if (! match.compatible (t, this))
1241 verify_fail ("incompatible type on stack");
1242 return t;
1245 // Pop a reference which is guaranteed to be initialized. MATCH
1246 // doesn't have to be a reference type; in this case this acts like
1247 // pop_type.
1248 type pop_init_ref (type match)
1250 type t = pop_raw ();
1251 if (t.isreference () && ! t.isinitialized ())
1252 verify_fail ("initialized reference required");
1253 else if (! match.compatible (t, this))
1254 verify_fail ("incompatible type on stack");
1255 return t;
1258 // Pop a reference type or a return address.
1259 type pop_ref_or_return ()
1261 type t = pop_raw ();
1262 if (! t.isreference () && t.key != return_address_type)
1263 verify_fail ("expected reference or return address on stack");
1264 return t;
1267 void push_type (type t)
1269 // If T is a numeric type like short, promote it to int.
1270 t.promote ();
1272 int depth = t.depth ();
1273 if (current_state->stackdepth + depth > current_method->max_stack)
1274 verify_fail ("stack overflow");
1275 current_state->stack[current_state->stacktop++] = t;
1276 current_state->stackdepth += depth;
1279 void set_variable (int index, type t)
1281 // If T is a numeric type like short, promote it to int.
1282 t.promote ();
1284 int depth = t.depth ();
1285 if (index > current_method->max_locals - depth)
1286 verify_fail ("invalid local variable");
1287 current_state->locals[index] = t;
1289 if (depth == 2)
1290 current_state->locals[index + 1] = continuation_type;
1291 if (index > 0 && current_state->locals[index - 1].iswide ())
1292 current_state->locals[index - 1] = unsuitable_type;
1295 type get_variable (int index, type t)
1297 int depth = t.depth ();
1298 if (index > current_method->max_locals - depth)
1299 verify_fail ("invalid local variable");
1300 if (! t.compatible (current_state->locals[index], this))
1301 verify_fail ("incompatible type in local variable");
1302 if (depth == 2)
1304 type t (continuation_type);
1305 if (! current_state->locals[index + 1].compatible (t, this))
1306 verify_fail ("invalid local variable");
1308 return current_state->locals[index];
1311 // Make sure ARRAY is an array type and that its elements are
1312 // compatible with type ELEMENT. Returns the actual element type.
1313 type require_array_type (type array, type element)
1315 // An odd case. Here we just pretend that everything went ok. If
1316 // the requested element type is some kind of reference, return
1317 // the null type instead.
1318 if (array.isnull ())
1319 return element.isreference () ? type (null_type) : element;
1321 if (! array.isarray ())
1322 verify_fail ("array required");
1324 type t = array.element_type (this);
1325 if (! element.compatible (t, this))
1327 // Special case for byte arrays, which must also be boolean
1328 // arrays.
1329 bool ok = true;
1330 if (element.key == byte_type)
1332 type e2 (boolean_type);
1333 ok = e2.compatible (t, this);
1335 if (! ok)
1336 verify_fail ("incompatible array element type");
1339 // Return T and not ELEMENT, because T might be specialized.
1340 return t;
1343 jint get_byte ()
1345 if (PC >= current_method->code_length)
1346 verify_fail ("premature end of bytecode");
1347 return (jint) bytecode[PC++] & 0xff;
1350 jint get_ushort ()
1352 jint b1 = get_byte ();
1353 jint b2 = get_byte ();
1354 return (jint) ((b1 << 8) | b2) & 0xffff;
1357 jint get_short ()
1359 jint b1 = get_byte ();
1360 jint b2 = get_byte ();
1361 jshort s = (b1 << 8) | b2;
1362 return (jint) s;
1365 jint get_int ()
1367 jint b1 = get_byte ();
1368 jint b2 = get_byte ();
1369 jint b3 = get_byte ();
1370 jint b4 = get_byte ();
1371 return (b1 << 24) | (b2 << 16) | (b3 << 8) | b4;
1374 int compute_jump (int offset)
1376 int npc = start_PC + offset;
1377 if (npc < 0 || npc >= current_method->code_length)
1378 verify_fail ("branch out of range", start_PC);
1379 return npc;
1382 // Add a new state to the state list at NPC.
1383 state *add_new_state (int npc, state *old_state)
1385 state *new_state = new state (old_state, current_method->max_stack,
1386 current_method->max_locals);
1387 debug_print ("== New state in add_new_state\n");
1388 new_state->print ("New", npc, current_method->max_stack,
1389 current_method->max_locals);
1390 linked<state> *nlink
1391 = (linked<state> *) _Jv_Malloc (sizeof (linked<state>));
1392 nlink->val = new_state;
1393 nlink->next = states[npc];
1394 states[npc] = nlink;
1395 new_state->set_pc (npc);
1396 return new_state;
1399 // Merge the indicated state into the state at the branch target and
1400 // schedule a new PC if there is a change. NPC is the PC of the
1401 // branch target, and FROM_STATE is the state at the source of the
1402 // branch. This method returns true if the destination state
1403 // changed and requires reverification, false otherwise.
1404 void merge_into (int npc, state *from_state)
1406 // Iterate over all target states and merge our state into each,
1407 // if applicable. FIXME one improvement we could make here is
1408 // "state destruction". Merging a new state into an existing one
1409 // might cause a return_address_type to be merged to
1410 // unsuitable_type. In this case the resulting state may now be
1411 // mergeable with other states currently held in parallel at this
1412 // location. So in this situation we could pairwise compare and
1413 // reduce the number of parallel states.
1414 bool applicable = false;
1415 for (linked<state> *iter = states[npc]; iter != NULL; iter = iter->next)
1417 state *new_state = iter->val;
1418 if (new_state->state_mergeable_p (from_state,
1419 current_method->max_locals, this))
1421 applicable = true;
1423 debug_print ("== Merge states in merge_into\n");
1424 from_state->print ("Frm", start_PC, current_method->max_stack,
1425 current_method->max_locals);
1426 new_state->print (" To", npc, current_method->max_stack,
1427 current_method->max_locals);
1428 bool changed = new_state->merge (from_state,
1429 current_method->max_locals,
1430 this);
1431 new_state->print ("New", npc, current_method->max_stack,
1432 current_method->max_locals);
1434 if (changed)
1435 new_state->reverify (this);
1439 if (! applicable)
1441 // Either we don't yet have a state at NPC, or we have a
1442 // return-address type that is in conflict with all existing
1443 // state. So, we need to create a new entry.
1444 state *new_state = add_new_state (npc, from_state);
1445 // A new state added in this way must always be reverified.
1446 new_state->reverify (this);
1450 void push_jump (int offset)
1452 int npc = compute_jump (offset);
1453 // According to the JVM Spec, we need to check for uninitialized
1454 // objects here. However, this does not actually affect type
1455 // safety, and the Eclipse java compiler generates code that
1456 // violates this constraint.
1457 merge_into (npc, current_state);
1460 void push_exception_jump (type t, int pc)
1462 // According to the JVM Spec, we need to check for uninitialized
1463 // objects here. However, this does not actually affect type
1464 // safety, and the Eclipse java compiler generates code that
1465 // violates this constraint.
1466 state s (current_state, current_method->max_stack,
1467 current_method->max_locals);
1468 if (current_method->max_stack < 1)
1469 verify_fail ("stack overflow at exception handler");
1470 s.set_exception (t, current_method->max_stack);
1471 merge_into (pc, &s);
1474 state *pop_jump ()
1476 state *new_state = next_verify_state;
1477 if (new_state == INVALID_STATE)
1478 verify_fail ("programmer error in pop_jump");
1479 if (new_state != NULL)
1481 next_verify_state = new_state->next;
1482 new_state->next = INVALID_STATE;
1484 return new_state;
1487 void invalidate_pc ()
1489 PC = state::NO_NEXT;
1492 void note_branch_target (int pc)
1494 // Don't check `pc <= PC', because we've advanced PC after
1495 // fetching the target and we haven't yet checked the next
1496 // instruction.
1497 if (pc < PC && ! (flags[pc] & FLAG_INSN_START))
1498 verify_fail ("branch not to instruction start", start_PC);
1499 flags[pc] |= FLAG_BRANCH_TARGET;
1502 void skip_padding ()
1504 while ((PC % 4) > 0)
1505 if (get_byte () != 0)
1506 verify_fail ("found nonzero padding byte");
1509 // Do the work for a `ret' instruction. INDEX is the index into the
1510 // local variables.
1511 void handle_ret_insn (int index)
1513 type ret_addr = get_variable (index, return_address_type);
1514 // It would be nice if we could do this. However, the JVM Spec
1515 // doesn't say that this is what happens. It is implied that
1516 // reusing a return address is invalid, but there's no actual
1517 // prohibition against it.
1518 // set_variable (index, unsuitable_type);
1520 int npc = ret_addr.get_pc ();
1521 // We might be returning to a `jsr' that is at the end of the
1522 // bytecode. This is ok if we never return from the called
1523 // subroutine, but if we see this here it is an error.
1524 if (npc >= current_method->code_length)
1525 verify_fail ("fell off end");
1527 // According to the JVM Spec, we need to check for uninitialized
1528 // objects here. However, this does not actually affect type
1529 // safety, and the Eclipse java compiler generates code that
1530 // violates this constraint.
1531 merge_into (npc, current_state);
1532 invalidate_pc ();
1535 void handle_jsr_insn (int offset)
1537 int npc = compute_jump (offset);
1539 // According to the JVM Spec, we need to check for uninitialized
1540 // objects here. However, this does not actually affect type
1541 // safety, and the Eclipse java compiler generates code that
1542 // violates this constraint.
1544 // Modify our state as appropriate for entry into a subroutine.
1545 type ret_addr (return_address_type);
1546 ret_addr.set_return_address (PC);
1547 push_type (ret_addr);
1548 merge_into (npc, current_state);
1549 invalidate_pc ();
1552 jclass construct_primitive_array_type (type_val prim)
1554 jclass k = NULL;
1555 switch (prim)
1557 case boolean_type:
1558 k = JvPrimClass (boolean);
1559 break;
1560 case char_type:
1561 k = JvPrimClass (char);
1562 break;
1563 case float_type:
1564 k = JvPrimClass (float);
1565 break;
1566 case double_type:
1567 k = JvPrimClass (double);
1568 break;
1569 case byte_type:
1570 k = JvPrimClass (byte);
1571 break;
1572 case short_type:
1573 k = JvPrimClass (short);
1574 break;
1575 case int_type:
1576 k = JvPrimClass (int);
1577 break;
1578 case long_type:
1579 k = JvPrimClass (long);
1580 break;
1582 // These aren't used here but we call them out to avoid
1583 // warnings.
1584 case void_type:
1585 case unsuitable_type:
1586 case return_address_type:
1587 case continuation_type:
1588 case reference_type:
1589 case null_type:
1590 case uninitialized_reference_type:
1591 default:
1592 verify_fail ("unknown type in construct_primitive_array_type");
1594 k = _Jv_GetArrayClass (k, NULL);
1595 return k;
1598 // This pass computes the location of branch targets and also
1599 // instruction starts.
1600 void branch_prepass ()
1602 flags = (char *) _Jv_Malloc (current_method->code_length);
1604 for (int i = 0; i < current_method->code_length; ++i)
1605 flags[i] = 0;
1607 PC = 0;
1608 while (PC < current_method->code_length)
1610 // Set `start_PC' early so that error checking can have the
1611 // correct value.
1612 start_PC = PC;
1613 flags[PC] |= FLAG_INSN_START;
1615 java_opcode opcode = (java_opcode) bytecode[PC++];
1616 switch (opcode)
1618 case op_nop:
1619 case op_aconst_null:
1620 case op_iconst_m1:
1621 case op_iconst_0:
1622 case op_iconst_1:
1623 case op_iconst_2:
1624 case op_iconst_3:
1625 case op_iconst_4:
1626 case op_iconst_5:
1627 case op_lconst_0:
1628 case op_lconst_1:
1629 case op_fconst_0:
1630 case op_fconst_1:
1631 case op_fconst_2:
1632 case op_dconst_0:
1633 case op_dconst_1:
1634 case op_iload_0:
1635 case op_iload_1:
1636 case op_iload_2:
1637 case op_iload_3:
1638 case op_lload_0:
1639 case op_lload_1:
1640 case op_lload_2:
1641 case op_lload_3:
1642 case op_fload_0:
1643 case op_fload_1:
1644 case op_fload_2:
1645 case op_fload_3:
1646 case op_dload_0:
1647 case op_dload_1:
1648 case op_dload_2:
1649 case op_dload_3:
1650 case op_aload_0:
1651 case op_aload_1:
1652 case op_aload_2:
1653 case op_aload_3:
1654 case op_iaload:
1655 case op_laload:
1656 case op_faload:
1657 case op_daload:
1658 case op_aaload:
1659 case op_baload:
1660 case op_caload:
1661 case op_saload:
1662 case op_istore_0:
1663 case op_istore_1:
1664 case op_istore_2:
1665 case op_istore_3:
1666 case op_lstore_0:
1667 case op_lstore_1:
1668 case op_lstore_2:
1669 case op_lstore_3:
1670 case op_fstore_0:
1671 case op_fstore_1:
1672 case op_fstore_2:
1673 case op_fstore_3:
1674 case op_dstore_0:
1675 case op_dstore_1:
1676 case op_dstore_2:
1677 case op_dstore_3:
1678 case op_astore_0:
1679 case op_astore_1:
1680 case op_astore_2:
1681 case op_astore_3:
1682 case op_iastore:
1683 case op_lastore:
1684 case op_fastore:
1685 case op_dastore:
1686 case op_aastore:
1687 case op_bastore:
1688 case op_castore:
1689 case op_sastore:
1690 case op_pop:
1691 case op_pop2:
1692 case op_dup:
1693 case op_dup_x1:
1694 case op_dup_x2:
1695 case op_dup2:
1696 case op_dup2_x1:
1697 case op_dup2_x2:
1698 case op_swap:
1699 case op_iadd:
1700 case op_isub:
1701 case op_imul:
1702 case op_idiv:
1703 case op_irem:
1704 case op_ishl:
1705 case op_ishr:
1706 case op_iushr:
1707 case op_iand:
1708 case op_ior:
1709 case op_ixor:
1710 case op_ladd:
1711 case op_lsub:
1712 case op_lmul:
1713 case op_ldiv:
1714 case op_lrem:
1715 case op_lshl:
1716 case op_lshr:
1717 case op_lushr:
1718 case op_land:
1719 case op_lor:
1720 case op_lxor:
1721 case op_fadd:
1722 case op_fsub:
1723 case op_fmul:
1724 case op_fdiv:
1725 case op_frem:
1726 case op_dadd:
1727 case op_dsub:
1728 case op_dmul:
1729 case op_ddiv:
1730 case op_drem:
1731 case op_ineg:
1732 case op_i2b:
1733 case op_i2c:
1734 case op_i2s:
1735 case op_lneg:
1736 case op_fneg:
1737 case op_dneg:
1738 case op_i2l:
1739 case op_i2f:
1740 case op_i2d:
1741 case op_l2i:
1742 case op_l2f:
1743 case op_l2d:
1744 case op_f2i:
1745 case op_f2l:
1746 case op_f2d:
1747 case op_d2i:
1748 case op_d2l:
1749 case op_d2f:
1750 case op_lcmp:
1751 case op_fcmpl:
1752 case op_fcmpg:
1753 case op_dcmpl:
1754 case op_dcmpg:
1755 case op_monitorenter:
1756 case op_monitorexit:
1757 case op_ireturn:
1758 case op_lreturn:
1759 case op_freturn:
1760 case op_dreturn:
1761 case op_areturn:
1762 case op_return:
1763 case op_athrow:
1764 case op_arraylength:
1765 break;
1767 case op_bipush:
1768 case op_ldc:
1769 case op_iload:
1770 case op_lload:
1771 case op_fload:
1772 case op_dload:
1773 case op_aload:
1774 case op_istore:
1775 case op_lstore:
1776 case op_fstore:
1777 case op_dstore:
1778 case op_astore:
1779 case op_ret:
1780 case op_newarray:
1781 get_byte ();
1782 break;
1784 case op_iinc:
1785 case op_sipush:
1786 case op_ldc_w:
1787 case op_ldc2_w:
1788 case op_getstatic:
1789 case op_getfield:
1790 case op_putfield:
1791 case op_putstatic:
1792 case op_new:
1793 case op_anewarray:
1794 case op_instanceof:
1795 case op_checkcast:
1796 case op_invokespecial:
1797 case op_invokestatic:
1798 case op_invokevirtual:
1799 get_short ();
1800 break;
1802 case op_multianewarray:
1803 get_short ();
1804 get_byte ();
1805 break;
1807 case op_jsr:
1808 case op_ifeq:
1809 case op_ifne:
1810 case op_iflt:
1811 case op_ifge:
1812 case op_ifgt:
1813 case op_ifle:
1814 case op_if_icmpeq:
1815 case op_if_icmpne:
1816 case op_if_icmplt:
1817 case op_if_icmpge:
1818 case op_if_icmpgt:
1819 case op_if_icmple:
1820 case op_if_acmpeq:
1821 case op_if_acmpne:
1822 case op_ifnull:
1823 case op_ifnonnull:
1824 case op_goto:
1825 note_branch_target (compute_jump (get_short ()));
1826 break;
1828 case op_tableswitch:
1830 skip_padding ();
1831 note_branch_target (compute_jump (get_int ()));
1832 jint low = get_int ();
1833 jint hi = get_int ();
1834 if (low > hi)
1835 verify_fail ("invalid tableswitch", start_PC);
1836 for (int i = low; i <= hi; ++i)
1837 note_branch_target (compute_jump (get_int ()));
1839 break;
1841 case op_lookupswitch:
1843 skip_padding ();
1844 note_branch_target (compute_jump (get_int ()));
1845 int npairs = get_int ();
1846 if (npairs < 0)
1847 verify_fail ("too few pairs in lookupswitch", start_PC);
1848 while (npairs-- > 0)
1850 get_int ();
1851 note_branch_target (compute_jump (get_int ()));
1854 break;
1856 case op_invokeinterface:
1857 get_short ();
1858 get_byte ();
1859 get_byte ();
1860 break;
1862 case op_wide:
1864 opcode = (java_opcode) get_byte ();
1865 get_short ();
1866 if (opcode == op_iinc)
1867 get_short ();
1869 break;
1871 case op_jsr_w:
1872 case op_goto_w:
1873 note_branch_target (compute_jump (get_int ()));
1874 break;
1876 // These are unused here, but we call them out explicitly
1877 // so that -Wswitch-enum doesn't complain.
1878 case op_putfield_1:
1879 case op_putfield_2:
1880 case op_putfield_4:
1881 case op_putfield_8:
1882 case op_putfield_a:
1883 case op_putstatic_1:
1884 case op_putstatic_2:
1885 case op_putstatic_4:
1886 case op_putstatic_8:
1887 case op_putstatic_a:
1888 case op_getfield_1:
1889 case op_getfield_2s:
1890 case op_getfield_2u:
1891 case op_getfield_4:
1892 case op_getfield_8:
1893 case op_getfield_a:
1894 case op_getstatic_1:
1895 case op_getstatic_2s:
1896 case op_getstatic_2u:
1897 case op_getstatic_4:
1898 case op_getstatic_8:
1899 case op_getstatic_a:
1900 default:
1901 verify_fail ("unrecognized instruction in branch_prepass",
1902 start_PC);
1905 // See if any previous branch tried to branch to the middle of
1906 // this instruction.
1907 for (int pc = start_PC + 1; pc < PC; ++pc)
1909 if ((flags[pc] & FLAG_BRANCH_TARGET))
1910 verify_fail ("branch to middle of instruction", pc);
1914 // Verify exception handlers.
1915 for (int i = 0; i < current_method->exc_count; ++i)
1917 if (! (flags[exception[i].handler_pc.i] & FLAG_INSN_START))
1918 verify_fail ("exception handler not at instruction start",
1919 exception[i].handler_pc.i);
1920 if (! (flags[exception[i].start_pc.i] & FLAG_INSN_START))
1921 verify_fail ("exception start not at instruction start",
1922 exception[i].start_pc.i);
1923 if (exception[i].end_pc.i != current_method->code_length
1924 && ! (flags[exception[i].end_pc.i] & FLAG_INSN_START))
1925 verify_fail ("exception end not at instruction start",
1926 exception[i].end_pc.i);
1928 flags[exception[i].handler_pc.i] |= FLAG_BRANCH_TARGET;
1932 void check_pool_index (int index)
1934 if (index < 0 || index >= current_class->constants.size)
1935 verify_fail ("constant pool index out of range", start_PC);
1938 type check_class_constant (int index)
1940 check_pool_index (index);
1941 _Jv_Constants *pool = &current_class->constants;
1942 if (pool->tags[index] == JV_CONSTANT_ResolvedClass)
1943 return type (pool->data[index].clazz, this);
1944 else if (pool->tags[index] == JV_CONSTANT_Class)
1945 return type (pool->data[index].utf8, this);
1946 verify_fail ("expected class constant", start_PC);
1949 type check_constant (int index)
1951 check_pool_index (index);
1952 _Jv_Constants *pool = &current_class->constants;
1953 int tag = pool->tags[index];
1954 if (tag == JV_CONSTANT_ResolvedString || tag == JV_CONSTANT_String)
1955 return type (&java::lang::String::class$, this);
1956 else if (tag == JV_CONSTANT_Integer)
1957 return type (int_type);
1958 else if (tag == JV_CONSTANT_Float)
1959 return type (float_type);
1960 else if (current_method->is_15
1961 && (tag == JV_CONSTANT_ResolvedClass || tag == JV_CONSTANT_Class))
1962 return type (&java::lang::Class::class$, this);
1963 verify_fail ("String, int, or float constant expected", start_PC);
1966 type check_wide_constant (int index)
1968 check_pool_index (index);
1969 _Jv_Constants *pool = &current_class->constants;
1970 if (pool->tags[index] == JV_CONSTANT_Long)
1971 return type (long_type);
1972 else if (pool->tags[index] == JV_CONSTANT_Double)
1973 return type (double_type);
1974 verify_fail ("long or double constant expected", start_PC);
1977 // Helper for both field and method. These are laid out the same in
1978 // the constant pool.
1979 type handle_field_or_method (int index, int expected,
1980 _Jv_Utf8Const **name,
1981 _Jv_Utf8Const **fmtype)
1983 check_pool_index (index);
1984 _Jv_Constants *pool = &current_class->constants;
1985 if (pool->tags[index] != expected)
1986 verify_fail ("didn't see expected constant", start_PC);
1987 // Once we know we have a Fieldref or Methodref we assume that it
1988 // is correctly laid out in the constant pool. I think the code
1989 // in defineclass.cc guarantees this.
1990 _Jv_ushort class_index, name_and_type_index;
1991 _Jv_loadIndexes (&pool->data[index],
1992 class_index,
1993 name_and_type_index);
1994 _Jv_ushort name_index, desc_index;
1995 _Jv_loadIndexes (&pool->data[name_and_type_index],
1996 name_index, desc_index);
1998 *name = pool->data[name_index].utf8;
1999 *fmtype = pool->data[desc_index].utf8;
2001 return check_class_constant (class_index);
2004 // Return field's type, compute class' type if requested.
2005 // If PUTFIELD is true, use the special 'putfield' semantics.
2006 type check_field_constant (int index, type *class_type = NULL,
2007 bool putfield = false)
2009 _Jv_Utf8Const *name, *field_type;
2010 type ct = handle_field_or_method (index,
2011 JV_CONSTANT_Fieldref,
2012 &name, &field_type);
2013 if (class_type)
2014 *class_type = ct;
2015 type result;
2016 if (field_type->first() == '[' || field_type->first() == 'L')
2017 result = type (field_type, this);
2018 else
2019 result = get_type_val_for_signature (field_type->first());
2021 // We have an obscure special case here: we can use `putfield' on
2022 // a field declared in this class, even if `this' has not yet been
2023 // initialized.
2024 if (putfield
2025 && ! current_state->this_type.isinitialized ()
2026 && current_state->this_type.pc == type::SELF
2027 && current_state->this_type.equals (ct, this)
2028 // We don't look at the signature, figuring that if it is
2029 // wrong we will fail during linking. FIXME?
2030 && _Jv_Linker::has_field_p (current_class, name))
2031 // Note that we don't actually know whether we're going to match
2032 // against 'this' or some other object of the same type. So,
2033 // here we set things up so that it doesn't matter. This relies
2034 // on knowing what our caller is up to.
2035 class_type->set_uninitialized (type::EITHER, this);
2037 return result;
2040 type check_method_constant (int index, bool is_interface,
2041 _Jv_Utf8Const **method_name,
2042 _Jv_Utf8Const **method_signature)
2044 return handle_field_or_method (index,
2045 (is_interface
2046 ? JV_CONSTANT_InterfaceMethodref
2047 : JV_CONSTANT_Methodref),
2048 method_name, method_signature);
2051 type get_one_type (char *&p)
2053 char *start = p;
2055 int arraycount = 0;
2056 while (*p == '[')
2058 ++arraycount;
2059 ++p;
2062 char v = *p++;
2064 if (v == 'L')
2066 while (*p != ';')
2067 ++p;
2068 ++p;
2069 _Jv_Utf8Const *name = make_utf8_const (start, p - start);
2070 return type (name, this);
2073 // Casting to jchar here is ok since we are looking at an ASCII
2074 // character.
2075 type_val rt = get_type_val_for_signature (jchar (v));
2077 if (arraycount == 0)
2079 // Callers of this function eventually push their arguments on
2080 // the stack. So, promote them here.
2081 return type (rt).promote ();
2084 jclass k = construct_primitive_array_type (rt);
2085 while (--arraycount > 0)
2086 k = _Jv_GetArrayClass (k, NULL);
2087 return type (k, this);
2090 void compute_argument_types (_Jv_Utf8Const *signature,
2091 type *types)
2093 char *p = signature->chars();
2095 // Skip `('.
2096 ++p;
2098 int i = 0;
2099 while (*p != ')')
2100 types[i++] = get_one_type (p);
2103 type compute_return_type (_Jv_Utf8Const *signature)
2105 char *p = signature->chars();
2106 while (*p != ')')
2107 ++p;
2108 ++p;
2109 return get_one_type (p);
2112 void check_return_type (type onstack)
2114 type rt = compute_return_type (current_method->self->signature);
2115 if (! rt.compatible (onstack, this))
2116 verify_fail ("incompatible return type");
2119 // Initialize the stack for the new method. Returns true if this
2120 // method is an instance initializer.
2121 bool initialize_stack ()
2123 int var = 0;
2124 bool is_init = _Jv_equalUtf8Consts (current_method->self->name,
2125 gcj::init_name);
2126 bool is_clinit = _Jv_equalUtf8Consts (current_method->self->name,
2127 gcj::clinit_name);
2129 using namespace java::lang::reflect;
2130 if (! Modifier::isStatic (current_method->self->accflags))
2132 type kurr (current_class, this);
2133 if (is_init)
2135 kurr.set_uninitialized (type::SELF, this);
2136 is_init = true;
2138 else if (is_clinit)
2139 verify_fail ("<clinit> method must be static");
2140 set_variable (0, kurr);
2141 current_state->set_this_type (kurr);
2142 ++var;
2144 else
2146 if (is_init)
2147 verify_fail ("<init> method must be non-static");
2150 // We have to handle wide arguments specially here.
2151 int arg_count = _Jv_count_arguments (current_method->self->signature);
2152 type arg_types[arg_count];
2153 compute_argument_types (current_method->self->signature, arg_types);
2154 for (int i = 0; i < arg_count; ++i)
2156 set_variable (var, arg_types[i]);
2157 ++var;
2158 if (arg_types[i].iswide ())
2159 ++var;
2162 return is_init;
2165 void verify_instructions_0 ()
2167 current_state = new state (current_method->max_stack,
2168 current_method->max_locals);
2170 PC = 0;
2171 start_PC = 0;
2173 // True if we are verifying an instance initializer.
2174 bool this_is_init = initialize_stack ();
2176 states = (linked<state> **) _Jv_Malloc (sizeof (linked<state> *)
2177 * current_method->code_length);
2178 for (int i = 0; i < current_method->code_length; ++i)
2179 states[i] = NULL;
2181 next_verify_state = NULL;
2183 while (true)
2185 // If the PC was invalidated, get a new one from the work list.
2186 if (PC == state::NO_NEXT)
2188 state *new_state = pop_jump ();
2189 // If it is null, we're done.
2190 if (new_state == NULL)
2191 break;
2193 PC = new_state->get_pc ();
2194 debug_print ("== State pop from pending list\n");
2195 // Set up the current state.
2196 current_state->copy (new_state, current_method->max_stack,
2197 current_method->max_locals);
2199 else
2201 // We only have to do this checking in the situation where
2202 // control flow falls through from the previous
2203 // instruction. Otherwise merging is done at the time we
2204 // push the branch. Note that we'll catch the
2205 // off-the-end problem just below.
2206 if (PC < current_method->code_length && states[PC] != NULL)
2208 // We've already visited this instruction. So merge
2209 // the states together. It is simplest, but not most
2210 // efficient, to just always invalidate the PC here.
2211 merge_into (PC, current_state);
2212 invalidate_pc ();
2213 continue;
2217 // Control can't fall off the end of the bytecode. We need to
2218 // check this in both cases, not just the fall-through case,
2219 // because we don't check to see whether a `jsr' appears at
2220 // the end of the bytecode until we process a `ret'.
2221 if (PC >= current_method->code_length)
2222 verify_fail ("fell off end");
2224 // We only have to keep saved state at branch targets. If
2225 // we're at a branch target and the state here hasn't been set
2226 // yet, we set it now. You might notice that `ret' targets
2227 // won't necessarily have FLAG_BRANCH_TARGET set. This
2228 // doesn't matter, since those states will be filled in by
2229 // merge_into.
2230 if (states[PC] == NULL && (flags[PC] & FLAG_BRANCH_TARGET))
2231 add_new_state (PC, current_state);
2233 // Set this before handling exceptions so that debug output is
2234 // sane.
2235 start_PC = PC;
2237 // Update states for all active exception handlers. Ordinarily
2238 // there are not many exception handlers. So we simply run
2239 // through them all.
2240 for (int i = 0; i < current_method->exc_count; ++i)
2242 if (PC >= exception[i].start_pc.i && PC < exception[i].end_pc.i)
2244 type handler (&java::lang::Throwable::class$, this);
2245 if (exception[i].handler_type.i != 0)
2246 handler = check_class_constant (exception[i].handler_type.i);
2247 push_exception_jump (handler, exception[i].handler_pc.i);
2251 current_state->print (" ", PC, current_method->max_stack,
2252 current_method->max_locals);
2253 java_opcode opcode = (java_opcode) bytecode[PC++];
2254 switch (opcode)
2256 case op_nop:
2257 break;
2259 case op_aconst_null:
2260 push_type (null_type);
2261 break;
2263 case op_iconst_m1:
2264 case op_iconst_0:
2265 case op_iconst_1:
2266 case op_iconst_2:
2267 case op_iconst_3:
2268 case op_iconst_4:
2269 case op_iconst_5:
2270 push_type (int_type);
2271 break;
2273 case op_lconst_0:
2274 case op_lconst_1:
2275 push_type (long_type);
2276 break;
2278 case op_fconst_0:
2279 case op_fconst_1:
2280 case op_fconst_2:
2281 push_type (float_type);
2282 break;
2284 case op_dconst_0:
2285 case op_dconst_1:
2286 push_type (double_type);
2287 break;
2289 case op_bipush:
2290 get_byte ();
2291 push_type (int_type);
2292 break;
2294 case op_sipush:
2295 get_short ();
2296 push_type (int_type);
2297 break;
2299 case op_ldc:
2300 push_type (check_constant (get_byte ()));
2301 break;
2302 case op_ldc_w:
2303 push_type (check_constant (get_ushort ()));
2304 break;
2305 case op_ldc2_w:
2306 push_type (check_wide_constant (get_ushort ()));
2307 break;
2309 case op_iload:
2310 push_type (get_variable (get_byte (), int_type));
2311 break;
2312 case op_lload:
2313 push_type (get_variable (get_byte (), long_type));
2314 break;
2315 case op_fload:
2316 push_type (get_variable (get_byte (), float_type));
2317 break;
2318 case op_dload:
2319 push_type (get_variable (get_byte (), double_type));
2320 break;
2321 case op_aload:
2322 push_type (get_variable (get_byte (), reference_type));
2323 break;
2325 case op_iload_0:
2326 case op_iload_1:
2327 case op_iload_2:
2328 case op_iload_3:
2329 push_type (get_variable (opcode - op_iload_0, int_type));
2330 break;
2331 case op_lload_0:
2332 case op_lload_1:
2333 case op_lload_2:
2334 case op_lload_3:
2335 push_type (get_variable (opcode - op_lload_0, long_type));
2336 break;
2337 case op_fload_0:
2338 case op_fload_1:
2339 case op_fload_2:
2340 case op_fload_3:
2341 push_type (get_variable (opcode - op_fload_0, float_type));
2342 break;
2343 case op_dload_0:
2344 case op_dload_1:
2345 case op_dload_2:
2346 case op_dload_3:
2347 push_type (get_variable (opcode - op_dload_0, double_type));
2348 break;
2349 case op_aload_0:
2350 case op_aload_1:
2351 case op_aload_2:
2352 case op_aload_3:
2353 push_type (get_variable (opcode - op_aload_0, reference_type));
2354 break;
2355 case op_iaload:
2356 pop_type (int_type);
2357 push_type (require_array_type (pop_init_ref (reference_type),
2358 int_type));
2359 break;
2360 case op_laload:
2361 pop_type (int_type);
2362 push_type (require_array_type (pop_init_ref (reference_type),
2363 long_type));
2364 break;
2365 case op_faload:
2366 pop_type (int_type);
2367 push_type (require_array_type (pop_init_ref (reference_type),
2368 float_type));
2369 break;
2370 case op_daload:
2371 pop_type (int_type);
2372 push_type (require_array_type (pop_init_ref (reference_type),
2373 double_type));
2374 break;
2375 case op_aaload:
2376 pop_type (int_type);
2377 push_type (require_array_type (pop_init_ref (reference_type),
2378 reference_type));
2379 break;
2380 case op_baload:
2381 pop_type (int_type);
2382 require_array_type (pop_init_ref (reference_type), byte_type);
2383 push_type (int_type);
2384 break;
2385 case op_caload:
2386 pop_type (int_type);
2387 require_array_type (pop_init_ref (reference_type), char_type);
2388 push_type (int_type);
2389 break;
2390 case op_saload:
2391 pop_type (int_type);
2392 require_array_type (pop_init_ref (reference_type), short_type);
2393 push_type (int_type);
2394 break;
2395 case op_istore:
2396 set_variable (get_byte (), pop_type (int_type));
2397 break;
2398 case op_lstore:
2399 set_variable (get_byte (), pop_type (long_type));
2400 break;
2401 case op_fstore:
2402 set_variable (get_byte (), pop_type (float_type));
2403 break;
2404 case op_dstore:
2405 set_variable (get_byte (), pop_type (double_type));
2406 break;
2407 case op_astore:
2408 set_variable (get_byte (), pop_ref_or_return ());
2409 break;
2410 case op_istore_0:
2411 case op_istore_1:
2412 case op_istore_2:
2413 case op_istore_3:
2414 set_variable (opcode - op_istore_0, pop_type (int_type));
2415 break;
2416 case op_lstore_0:
2417 case op_lstore_1:
2418 case op_lstore_2:
2419 case op_lstore_3:
2420 set_variable (opcode - op_lstore_0, pop_type (long_type));
2421 break;
2422 case op_fstore_0:
2423 case op_fstore_1:
2424 case op_fstore_2:
2425 case op_fstore_3:
2426 set_variable (opcode - op_fstore_0, pop_type (float_type));
2427 break;
2428 case op_dstore_0:
2429 case op_dstore_1:
2430 case op_dstore_2:
2431 case op_dstore_3:
2432 set_variable (opcode - op_dstore_0, pop_type (double_type));
2433 break;
2434 case op_astore_0:
2435 case op_astore_1:
2436 case op_astore_2:
2437 case op_astore_3:
2438 set_variable (opcode - op_astore_0, pop_ref_or_return ());
2439 break;
2440 case op_iastore:
2441 pop_type (int_type);
2442 pop_type (int_type);
2443 require_array_type (pop_init_ref (reference_type), int_type);
2444 break;
2445 case op_lastore:
2446 pop_type (long_type);
2447 pop_type (int_type);
2448 require_array_type (pop_init_ref (reference_type), long_type);
2449 break;
2450 case op_fastore:
2451 pop_type (float_type);
2452 pop_type (int_type);
2453 require_array_type (pop_init_ref (reference_type), float_type);
2454 break;
2455 case op_dastore:
2456 pop_type (double_type);
2457 pop_type (int_type);
2458 require_array_type (pop_init_ref (reference_type), double_type);
2459 break;
2460 case op_aastore:
2461 pop_type (reference_type);
2462 pop_type (int_type);
2463 require_array_type (pop_init_ref (reference_type), reference_type);
2464 break;
2465 case op_bastore:
2466 pop_type (int_type);
2467 pop_type (int_type);
2468 require_array_type (pop_init_ref (reference_type), byte_type);
2469 break;
2470 case op_castore:
2471 pop_type (int_type);
2472 pop_type (int_type);
2473 require_array_type (pop_init_ref (reference_type), char_type);
2474 break;
2475 case op_sastore:
2476 pop_type (int_type);
2477 pop_type (int_type);
2478 require_array_type (pop_init_ref (reference_type), short_type);
2479 break;
2480 case op_pop:
2481 pop32 ();
2482 break;
2483 case op_pop2:
2485 type t = pop_raw ();
2486 if (! t.iswide ())
2487 pop32 ();
2489 break;
2490 case op_dup:
2492 type t = pop32 ();
2493 push_type (t);
2494 push_type (t);
2496 break;
2497 case op_dup_x1:
2499 type t1 = pop32 ();
2500 type t2 = pop32 ();
2501 push_type (t1);
2502 push_type (t2);
2503 push_type (t1);
2505 break;
2506 case op_dup_x2:
2508 type t1 = pop32 ();
2509 type t2 = pop_raw ();
2510 if (! t2.iswide ())
2512 type t3 = pop32 ();
2513 push_type (t1);
2514 push_type (t3);
2516 else
2517 push_type (t1);
2518 push_type (t2);
2519 push_type (t1);
2521 break;
2522 case op_dup2:
2524 type t = pop_raw ();
2525 if (! t.iswide ())
2527 type t2 = pop32 ();
2528 push_type (t2);
2529 push_type (t);
2530 push_type (t2);
2532 else
2533 push_type (t);
2534 push_type (t);
2536 break;
2537 case op_dup2_x1:
2539 type t1 = pop_raw ();
2540 type t2 = pop32 ();
2541 if (! t1.iswide ())
2543 type t3 = pop32 ();
2544 push_type (t2);
2545 push_type (t1);
2546 push_type (t3);
2548 else
2549 push_type (t1);
2550 push_type (t2);
2551 push_type (t1);
2553 break;
2554 case op_dup2_x2:
2556 type t1 = pop_raw ();
2557 if (t1.iswide ())
2559 type t2 = pop_raw ();
2560 if (t2.iswide ())
2562 push_type (t1);
2563 push_type (t2);
2565 else
2567 type t3 = pop32 ();
2568 push_type (t1);
2569 push_type (t3);
2570 push_type (t2);
2572 push_type (t1);
2574 else
2576 type t2 = pop32 ();
2577 type t3 = pop_raw ();
2578 if (t3.iswide ())
2580 push_type (t2);
2581 push_type (t1);
2583 else
2585 type t4 = pop32 ();
2586 push_type (t2);
2587 push_type (t1);
2588 push_type (t4);
2590 push_type (t3);
2591 push_type (t2);
2592 push_type (t1);
2595 break;
2596 case op_swap:
2598 type t1 = pop32 ();
2599 type t2 = pop32 ();
2600 push_type (t1);
2601 push_type (t2);
2603 break;
2604 case op_iadd:
2605 case op_isub:
2606 case op_imul:
2607 case op_idiv:
2608 case op_irem:
2609 case op_ishl:
2610 case op_ishr:
2611 case op_iushr:
2612 case op_iand:
2613 case op_ior:
2614 case op_ixor:
2615 pop_type (int_type);
2616 push_type (pop_type (int_type));
2617 break;
2618 case op_ladd:
2619 case op_lsub:
2620 case op_lmul:
2621 case op_ldiv:
2622 case op_lrem:
2623 case op_land:
2624 case op_lor:
2625 case op_lxor:
2626 pop_type (long_type);
2627 push_type (pop_type (long_type));
2628 break;
2629 case op_lshl:
2630 case op_lshr:
2631 case op_lushr:
2632 pop_type (int_type);
2633 push_type (pop_type (long_type));
2634 break;
2635 case op_fadd:
2636 case op_fsub:
2637 case op_fmul:
2638 case op_fdiv:
2639 case op_frem:
2640 pop_type (float_type);
2641 push_type (pop_type (float_type));
2642 break;
2643 case op_dadd:
2644 case op_dsub:
2645 case op_dmul:
2646 case op_ddiv:
2647 case op_drem:
2648 pop_type (double_type);
2649 push_type (pop_type (double_type));
2650 break;
2651 case op_ineg:
2652 case op_i2b:
2653 case op_i2c:
2654 case op_i2s:
2655 push_type (pop_type (int_type));
2656 break;
2657 case op_lneg:
2658 push_type (pop_type (long_type));
2659 break;
2660 case op_fneg:
2661 push_type (pop_type (float_type));
2662 break;
2663 case op_dneg:
2664 push_type (pop_type (double_type));
2665 break;
2666 case op_iinc:
2667 get_variable (get_byte (), int_type);
2668 get_byte ();
2669 break;
2670 case op_i2l:
2671 pop_type (int_type);
2672 push_type (long_type);
2673 break;
2674 case op_i2f:
2675 pop_type (int_type);
2676 push_type (float_type);
2677 break;
2678 case op_i2d:
2679 pop_type (int_type);
2680 push_type (double_type);
2681 break;
2682 case op_l2i:
2683 pop_type (long_type);
2684 push_type (int_type);
2685 break;
2686 case op_l2f:
2687 pop_type (long_type);
2688 push_type (float_type);
2689 break;
2690 case op_l2d:
2691 pop_type (long_type);
2692 push_type (double_type);
2693 break;
2694 case op_f2i:
2695 pop_type (float_type);
2696 push_type (int_type);
2697 break;
2698 case op_f2l:
2699 pop_type (float_type);
2700 push_type (long_type);
2701 break;
2702 case op_f2d:
2703 pop_type (float_type);
2704 push_type (double_type);
2705 break;
2706 case op_d2i:
2707 pop_type (double_type);
2708 push_type (int_type);
2709 break;
2710 case op_d2l:
2711 pop_type (double_type);
2712 push_type (long_type);
2713 break;
2714 case op_d2f:
2715 pop_type (double_type);
2716 push_type (float_type);
2717 break;
2718 case op_lcmp:
2719 pop_type (long_type);
2720 pop_type (long_type);
2721 push_type (int_type);
2722 break;
2723 case op_fcmpl:
2724 case op_fcmpg:
2725 pop_type (float_type);
2726 pop_type (float_type);
2727 push_type (int_type);
2728 break;
2729 case op_dcmpl:
2730 case op_dcmpg:
2731 pop_type (double_type);
2732 pop_type (double_type);
2733 push_type (int_type);
2734 break;
2735 case op_ifeq:
2736 case op_ifne:
2737 case op_iflt:
2738 case op_ifge:
2739 case op_ifgt:
2740 case op_ifle:
2741 pop_type (int_type);
2742 push_jump (get_short ());
2743 break;
2744 case op_if_icmpeq:
2745 case op_if_icmpne:
2746 case op_if_icmplt:
2747 case op_if_icmpge:
2748 case op_if_icmpgt:
2749 case op_if_icmple:
2750 pop_type (int_type);
2751 pop_type (int_type);
2752 push_jump (get_short ());
2753 break;
2754 case op_if_acmpeq:
2755 case op_if_acmpne:
2756 pop_type (reference_type);
2757 pop_type (reference_type);
2758 push_jump (get_short ());
2759 break;
2760 case op_goto:
2761 push_jump (get_short ());
2762 invalidate_pc ();
2763 break;
2764 case op_jsr:
2765 handle_jsr_insn (get_short ());
2766 break;
2767 case op_ret:
2768 handle_ret_insn (get_byte ());
2769 break;
2770 case op_tableswitch:
2772 pop_type (int_type);
2773 skip_padding ();
2774 push_jump (get_int ());
2775 jint low = get_int ();
2776 jint high = get_int ();
2777 // Already checked LOW -vs- HIGH.
2778 for (int i = low; i <= high; ++i)
2779 push_jump (get_int ());
2780 invalidate_pc ();
2782 break;
2784 case op_lookupswitch:
2786 pop_type (int_type);
2787 skip_padding ();
2788 push_jump (get_int ());
2789 jint npairs = get_int ();
2790 // Already checked NPAIRS >= 0.
2791 jint lastkey = 0;
2792 for (int i = 0; i < npairs; ++i)
2794 jint key = get_int ();
2795 if (i > 0 && key <= lastkey)
2796 verify_fail ("lookupswitch pairs unsorted", start_PC);
2797 lastkey = key;
2798 push_jump (get_int ());
2800 invalidate_pc ();
2802 break;
2803 case op_ireturn:
2804 check_return_type (pop_type (int_type));
2805 invalidate_pc ();
2806 break;
2807 case op_lreturn:
2808 check_return_type (pop_type (long_type));
2809 invalidate_pc ();
2810 break;
2811 case op_freturn:
2812 check_return_type (pop_type (float_type));
2813 invalidate_pc ();
2814 break;
2815 case op_dreturn:
2816 check_return_type (pop_type (double_type));
2817 invalidate_pc ();
2818 break;
2819 case op_areturn:
2820 check_return_type (pop_init_ref (reference_type));
2821 invalidate_pc ();
2822 break;
2823 case op_return:
2824 // We only need to check this when the return type is
2825 // void, because all instance initializers return void.
2826 if (this_is_init)
2827 current_state->check_this_initialized (this);
2828 check_return_type (void_type);
2829 invalidate_pc ();
2830 break;
2831 case op_getstatic:
2832 push_type (check_field_constant (get_ushort ()));
2833 break;
2834 case op_putstatic:
2835 pop_type (check_field_constant (get_ushort ()));
2836 break;
2837 case op_getfield:
2839 type klass;
2840 type field = check_field_constant (get_ushort (), &klass);
2841 pop_type (klass);
2842 push_type (field);
2844 break;
2845 case op_putfield:
2847 type klass;
2848 type field = check_field_constant (get_ushort (), &klass, true);
2849 pop_type (field);
2850 pop_type (klass);
2852 break;
2854 case op_invokevirtual:
2855 case op_invokespecial:
2856 case op_invokestatic:
2857 case op_invokeinterface:
2859 _Jv_Utf8Const *method_name, *method_signature;
2860 type class_type
2861 = check_method_constant (get_ushort (),
2862 opcode == op_invokeinterface,
2863 &method_name,
2864 &method_signature);
2865 // NARGS is only used when we're processing
2866 // invokeinterface. It is simplest for us to compute it
2867 // here and then verify it later.
2868 int nargs = 0;
2869 if (opcode == op_invokeinterface)
2871 nargs = get_byte ();
2872 if (get_byte () != 0)
2873 verify_fail ("invokeinterface dummy byte is wrong");
2876 bool is_init = false;
2877 if (_Jv_equalUtf8Consts (method_name, gcj::init_name))
2879 is_init = true;
2880 if (opcode != op_invokespecial)
2881 verify_fail ("can't invoke <init>");
2883 else if (method_name->first() == '<')
2884 verify_fail ("can't invoke method starting with `<'");
2886 // Pop arguments and check types.
2887 int arg_count = _Jv_count_arguments (method_signature);
2888 type arg_types[arg_count];
2889 compute_argument_types (method_signature, arg_types);
2890 for (int i = arg_count - 1; i >= 0; --i)
2892 // This is only used for verifying the byte for
2893 // invokeinterface.
2894 nargs -= arg_types[i].depth ();
2895 pop_init_ref (arg_types[i]);
2898 if (opcode == op_invokeinterface
2899 && nargs != 1)
2900 verify_fail ("wrong argument count for invokeinterface");
2902 if (opcode != op_invokestatic)
2904 type t = class_type;
2905 if (is_init)
2907 // In this case the PC doesn't matter.
2908 t.set_uninitialized (type::UNINIT, this);
2909 // FIXME: check to make sure that the <init>
2910 // call is to the right class.
2911 // It must either be super or an exact class
2912 // match.
2914 type raw = pop_raw ();
2915 if (! t.compatible (raw, this))
2916 verify_fail ("incompatible type on stack");
2918 if (is_init)
2919 current_state->set_initialized (raw.get_pc (),
2920 current_method->max_locals);
2923 type rt = compute_return_type (method_signature);
2924 if (! rt.isvoid ())
2925 push_type (rt);
2927 break;
2929 case op_new:
2931 type t = check_class_constant (get_ushort ());
2932 if (t.isarray ())
2933 verify_fail ("type is array");
2934 t.set_uninitialized (start_PC, this);
2935 push_type (t);
2937 break;
2939 case op_newarray:
2941 int atype = get_byte ();
2942 // We intentionally have chosen constants to make this
2943 // valid.
2944 if (atype < boolean_type || atype > long_type)
2945 verify_fail ("type not primitive", start_PC);
2946 pop_type (int_type);
2947 type t (construct_primitive_array_type (type_val (atype)), this);
2948 push_type (t);
2950 break;
2951 case op_anewarray:
2952 pop_type (int_type);
2953 push_type (check_class_constant (get_ushort ()).to_array (this));
2954 break;
2955 case op_arraylength:
2957 type t = pop_init_ref (reference_type);
2958 if (! t.isarray () && ! t.isnull ())
2959 verify_fail ("array type expected");
2960 push_type (int_type);
2962 break;
2963 case op_athrow:
2964 pop_type (type (&java::lang::Throwable::class$, this));
2965 invalidate_pc ();
2966 break;
2967 case op_checkcast:
2968 pop_init_ref (reference_type);
2969 push_type (check_class_constant (get_ushort ()));
2970 break;
2971 case op_instanceof:
2972 pop_init_ref (reference_type);
2973 check_class_constant (get_ushort ());
2974 push_type (int_type);
2975 break;
2976 case op_monitorenter:
2977 pop_init_ref (reference_type);
2978 break;
2979 case op_monitorexit:
2980 pop_init_ref (reference_type);
2981 break;
2982 case op_wide:
2984 switch (get_byte ())
2986 case op_iload:
2987 push_type (get_variable (get_ushort (), int_type));
2988 break;
2989 case op_lload:
2990 push_type (get_variable (get_ushort (), long_type));
2991 break;
2992 case op_fload:
2993 push_type (get_variable (get_ushort (), float_type));
2994 break;
2995 case op_dload:
2996 push_type (get_variable (get_ushort (), double_type));
2997 break;
2998 case op_aload:
2999 push_type (get_variable (get_ushort (), reference_type));
3000 break;
3001 case op_istore:
3002 set_variable (get_ushort (), pop_type (int_type));
3003 break;
3004 case op_lstore:
3005 set_variable (get_ushort (), pop_type (long_type));
3006 break;
3007 case op_fstore:
3008 set_variable (get_ushort (), pop_type (float_type));
3009 break;
3010 case op_dstore:
3011 set_variable (get_ushort (), pop_type (double_type));
3012 break;
3013 case op_astore:
3014 set_variable (get_ushort (), pop_init_ref (reference_type));
3015 break;
3016 case op_ret:
3017 handle_ret_insn (get_short ());
3018 break;
3019 case op_iinc:
3020 get_variable (get_ushort (), int_type);
3021 get_short ();
3022 break;
3023 default:
3024 verify_fail ("unrecognized wide instruction", start_PC);
3027 break;
3028 case op_multianewarray:
3030 type atype = check_class_constant (get_ushort ());
3031 int dim = get_byte ();
3032 if (dim < 1)
3033 verify_fail ("too few dimensions to multianewarray", start_PC);
3034 atype.verify_dimensions (dim, this);
3035 for (int i = 0; i < dim; ++i)
3036 pop_type (int_type);
3037 push_type (atype);
3039 break;
3040 case op_ifnull:
3041 case op_ifnonnull:
3042 pop_type (reference_type);
3043 push_jump (get_short ());
3044 break;
3045 case op_goto_w:
3046 push_jump (get_int ());
3047 invalidate_pc ();
3048 break;
3049 case op_jsr_w:
3050 handle_jsr_insn (get_int ());
3051 break;
3053 // These are unused here, but we call them out explicitly
3054 // so that -Wswitch-enum doesn't complain.
3055 case op_putfield_1:
3056 case op_putfield_2:
3057 case op_putfield_4:
3058 case op_putfield_8:
3059 case op_putfield_a:
3060 case op_putstatic_1:
3061 case op_putstatic_2:
3062 case op_putstatic_4:
3063 case op_putstatic_8:
3064 case op_putstatic_a:
3065 case op_getfield_1:
3066 case op_getfield_2s:
3067 case op_getfield_2u:
3068 case op_getfield_4:
3069 case op_getfield_8:
3070 case op_getfield_a:
3071 case op_getstatic_1:
3072 case op_getstatic_2s:
3073 case op_getstatic_2u:
3074 case op_getstatic_4:
3075 case op_getstatic_8:
3076 case op_getstatic_a:
3077 default:
3078 // Unrecognized opcode.
3079 verify_fail ("unrecognized instruction in verify_instructions_0",
3080 start_PC);
3085 public:
3087 void verify_instructions ()
3089 branch_prepass ();
3090 verify_instructions_0 ();
3093 _Jv_BytecodeVerifier (_Jv_InterpMethod *m)
3095 // We just print the text as utf-8. This is just for debugging
3096 // anyway.
3097 debug_print ("--------------------------------\n");
3098 debug_print ("-- Verifying method `%s'\n", m->self->name->chars());
3100 current_method = m;
3101 bytecode = m->bytecode ();
3102 exception = m->exceptions ();
3103 current_class = m->defining_class;
3105 states = NULL;
3106 flags = NULL;
3107 utf8_list = NULL;
3108 isect_list = NULL;
3111 ~_Jv_BytecodeVerifier ()
3113 if (flags)
3114 _Jv_Free (flags);
3116 while (utf8_list != NULL)
3118 linked<_Jv_Utf8Const> *n = utf8_list->next;
3119 _Jv_Free (utf8_list);
3120 utf8_list = n;
3123 while (isect_list != NULL)
3125 ref_intersection *next = isect_list->alloc_next;
3126 delete isect_list;
3127 isect_list = next;
3130 if (states)
3132 for (int i = 0; i < current_method->code_length; ++i)
3134 linked<state> *iter = states[i];
3135 while (iter != NULL)
3137 linked<state> *next = iter->next;
3138 delete iter->val;
3139 _Jv_Free (iter);
3140 iter = next;
3143 _Jv_Free (states);
3148 void
3149 _Jv_VerifyMethod (_Jv_InterpMethod *meth)
3151 _Jv_BytecodeVerifier v (meth);
3152 v.verify_instructions ();
3155 #endif /* INTERPRETER */