1 // defineclass.cc - defining a class from .class format.
3 /* Copyright (C) 2001, 2002, 2003 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
11 // Written by Tom Tromey <tromey@redhat.com>
13 // Define VERIFY_DEBUG to enable debugging output.
19 #include <java-insns.h>
20 #include <java-interp.h>
24 #include <java/lang/Class.h>
25 #include <java/lang/VerifyError.h>
26 #include <java/lang/Throwable.h>
27 #include <java/lang/reflect/Modifier.h>
28 #include <java/lang/StringBuffer.h>
32 #endif /* VERIFY_DEBUG */
35 static void debug_print (const char *fmt
, ...)
36 __attribute__ ((format (printf
, 1, 2)));
39 debug_print (const char *fmt
, ...)
44 vfprintf (stderr
, fmt
, ap
);
46 #endif /* VERIFY_DEBUG */
49 class _Jv_BytecodeVerifier
53 static const int FLAG_INSN_START
= 1;
54 static const int FLAG_BRANCH_TARGET
= 2;
59 struct subr_entry_info
;
64 // The PC corresponding to the start of the current instruction.
67 // The current state of the stack, locals, etc.
70 // We store the state at branch targets, for merging. This holds
74 // We keep a linked list of all the PCs which we must reverify.
75 // The link is done using the PC values. This is the head of the
79 // We keep some flags for each instruction. The values are the
80 // FLAG_* constants defined above.
83 // We need to keep track of which instructions can call a given
84 // subroutine. FIXME: this is inefficient. We keep a linked list
85 // of all calling `jsr's at at each jsr target.
88 // We keep a linked list of entries which map each `ret' instruction
89 // to its unique subroutine entry point. We expect that there won't
90 // be many `ret' instructions, so a linked list is ok.
91 subr_entry_info
*entry_points
;
93 // The bytecode itself.
94 unsigned char *bytecode
;
96 _Jv_InterpException
*exception
;
101 _Jv_InterpMethod
*current_method
;
103 // A linked list of utf8 objects we allocate. This is really ugly,
104 // but without this our utf8 objects would be collected.
105 linked_utf8
*utf8_list
;
113 _Jv_Utf8Const
*make_utf8_const (char *s
, int len
)
115 _Jv_Utf8Const
*val
= _Jv_makeUtf8Const (s
, len
);
116 _Jv_Utf8Const
*r
= (_Jv_Utf8Const
*) _Jv_Malloc (sizeof (_Jv_Utf8Const
)
119 r
->length
= val
->length
;
121 memcpy (r
->data
, val
->data
, val
->length
+ 1);
123 linked_utf8
*lu
= (linked_utf8
*) _Jv_Malloc (sizeof (linked_utf8
));
125 lu
->next
= utf8_list
;
131 __attribute__ ((__noreturn__
)) void verify_fail (char *s
, jint pc
= -1)
133 using namespace java::lang
;
134 StringBuffer
*buf
= new StringBuffer ();
136 buf
->append (JvNewStringLatin1 ("verification failed"));
141 buf
->append (JvNewStringLatin1 (" at PC "));
145 _Jv_InterpMethod
*method
= current_method
;
146 buf
->append (JvNewStringLatin1 (" in "));
147 buf
->append (current_class
->getName());
148 buf
->append ((jchar
) ':');
149 buf
->append (JvNewStringUTF (method
->get_method()->name
->data
));
150 buf
->append ((jchar
) '(');
151 buf
->append (JvNewStringUTF (method
->get_method()->signature
->data
));
152 buf
->append ((jchar
) ')');
154 buf
->append (JvNewStringLatin1 (": "));
155 buf
->append (JvNewStringLatin1 (s
));
156 throw new java::lang::VerifyError (buf
->toString ());
159 // This enum holds a list of tags for all the different types we
160 // need to handle. Reference types are treated specially by the
166 // The values for primitive types are chosen to correspond to values
167 // specified to newarray.
177 // Used when overwriting second word of a double or long in the
178 // local variables. Also used after merging local variable states
179 // to indicate an unusable value.
184 // There is an obscure special case which requires us to note when
185 // a local variable has not been used by a subroutine. See
186 // push_jump_merge for more information.
187 unused_by_subroutine_type
,
189 // Everything after `reference_type' must be a reference type.
192 unresolved_reference_type
,
193 uninitialized_reference_type
,
194 uninitialized_unresolved_reference_type
197 // Return the type_val corresponding to a primitive signature
198 // character. For instance `I' returns `int.class'.
199 type_val
get_type_val_for_signature (jchar sig
)
232 verify_fail ("invalid signature");
237 // Return the type_val corresponding to a primitive class.
238 type_val
get_type_val_for_signature (jclass k
)
240 return get_type_val_for_signature ((jchar
) k
->method_count
);
243 // This is like _Jv_IsAssignableFrom, but it works even if SOURCE or
244 // TARGET haven't been prepared.
245 static bool is_assignable_from_slow (jclass target
, jclass source
)
247 // This will terminate when SOURCE==Object.
250 if (source
== target
)
253 if (target
->isPrimitive () || source
->isPrimitive ())
256 if (target
->isArray ())
258 if (! source
->isArray ())
260 target
= target
->getComponentType ();
261 source
= source
->getComponentType ();
263 else if (target
->isInterface ())
265 for (int i
= 0; i
< source
->interface_count
; ++i
)
267 // We use a recursive call because we also need to
268 // check superinterfaces.
269 if (is_assignable_from_slow (target
, source
->interfaces
[i
]))
272 source
= source
->getSuperclass ();
276 // We must do this check before we check to see if SOURCE is
277 // an interface. This way we know that any interface is
278 // assignable to an Object.
279 else if (target
== &java::lang::Object::class$
)
281 else if (source
->isInterface ())
283 for (int i
= 0; i
< target
->interface_count
; ++i
)
285 // We use a recursive call because we also need to
286 // check superinterfaces.
287 if (is_assignable_from_slow (target
->interfaces
[i
], source
))
290 target
= target
->getSuperclass ();
294 else if (source
== &java::lang::Object::class$
)
297 source
= source
->getSuperclass ();
301 // This is used to keep track of which `jsr's correspond to a given
305 // PC of the instruction just after the jsr.
311 // This is used to keep track of which subroutine entry point
312 // corresponds to which `ret' instruction.
313 struct subr_entry_info
315 // PC of the subroutine entry point.
317 // PC of the `ret' instruction.
320 subr_entry_info
*next
;
323 // The `type' class is used to represent a single type in the
329 // Some associated data.
332 // For a resolved reference type, this is a pointer to the class.
334 // For other reference types, this it the name of the class.
337 // This is used when constructing a new object. It is the PC of the
338 // `new' instruction which created the object. We use the special
339 // value -2 to mean that this is uninitialized, and the special
340 // value -1 for the case where the current method is itself the
344 static const int UNINIT
= -2;
345 static const int SELF
= -1;
347 // Basic constructor.
350 key
= unsuitable_type
;
355 // Make a new instance given the type tag. We assume a generic
356 // `reference_type' means Object.
361 if (key
== reference_type
)
362 data
.klass
= &java::lang::Object::class$
;
366 // Make a new instance given a class.
369 key
= reference_type
;
374 // Make a new instance given the name of a class.
375 type (_Jv_Utf8Const
*n
)
377 key
= unresolved_reference_type
;
390 // These operators are required because libgcj can't link in
392 void *operator new[] (size_t bytes
)
394 return _Jv_Malloc (bytes
);
397 void operator delete[] (void *mem
)
402 type
& operator= (type_val k
)
410 type
& operator= (const type
& t
)
418 // Promote a numeric type.
421 if (key
== boolean_type
|| key
== char_type
422 || key
== byte_type
|| key
== short_type
)
427 // If *THIS is an unresolved reference type, resolve it.
428 void resolve (_Jv_BytecodeVerifier
*verifier
)
430 if (key
!= unresolved_reference_type
431 && key
!= uninitialized_unresolved_reference_type
)
434 using namespace java::lang
;
435 java::lang::ClassLoader
*loader
436 = verifier
->current_class
->getClassLoaderInternal();
437 // We might see either kind of name. Sigh.
438 if (data
.name
->data
[0] == 'L'
439 && data
.name
->data
[data
.name
->length
- 1] == ';')
440 data
.klass
= _Jv_FindClassFromSignature (data
.name
->data
, loader
);
442 data
.klass
= Class::forName (_Jv_NewStringUtf8Const (data
.name
),
444 key
= (key
== unresolved_reference_type
446 : uninitialized_reference_type
);
449 // Mark this type as the uninitialized result of `new'.
450 void set_uninitialized (int npc
, _Jv_BytecodeVerifier
*verifier
)
452 if (key
== reference_type
)
453 key
= uninitialized_reference_type
;
454 else if (key
== unresolved_reference_type
)
455 key
= uninitialized_unresolved_reference_type
;
457 verifier
->verify_fail ("internal error in type::uninitialized");
461 // Mark this type as now initialized.
462 void set_initialized (int npc
)
464 if (npc
!= UNINIT
&& pc
== npc
465 && (key
== uninitialized_reference_type
466 || key
== uninitialized_unresolved_reference_type
))
468 key
= (key
== uninitialized_reference_type
470 : unresolved_reference_type
);
476 // Return true if an object of type K can be assigned to a variable
477 // of type *THIS. Handle various special cases too. Might modify
478 // *THIS or K. Note however that this does not perform numeric
480 bool compatible (type
&k
, _Jv_BytecodeVerifier
*verifier
)
482 // Any type is compatible with the unsuitable type.
483 if (key
== unsuitable_type
)
486 if (key
< reference_type
|| k
.key
< reference_type
)
489 // The `null' type is convertible to any initialized reference
491 if (key
== null_type
|| k
.key
== null_type
)
494 // Any reference type is convertible to Object. This is a special
495 // case so we don't need to unnecessarily resolve a class.
496 if (key
== reference_type
497 && data
.klass
== &java::lang::Object::class$
)
500 // An initialized type and an uninitialized type are not
502 if (isinitialized () != k
.isinitialized ())
505 // Two uninitialized objects are compatible if either:
506 // * The PCs are identical, or
507 // * One PC is UNINIT.
508 if (! isinitialized ())
510 if (pc
!= k
.pc
&& pc
!= UNINIT
&& k
.pc
!= UNINIT
)
514 // Two unresolved types are equal if their names are the same.
517 && _Jv_equalUtf8Consts (data
.name
, k
.data
.name
))
520 // We must resolve both types and check assignability.
522 k
.resolve (verifier
);
523 return is_assignable_from_slow (data
.klass
, k
.data
.klass
);
528 return key
== void_type
;
533 return key
== long_type
|| key
== double_type
;
536 // Return number of stack or local variable slots taken by this
540 return iswide () ? 2 : 1;
543 bool isarray () const
545 // We treat null_type as not an array. This is ok based on the
546 // current uses of this method.
547 if (key
== reference_type
)
548 return data
.klass
->isArray ();
549 else if (key
== unresolved_reference_type
)
550 return data
.name
->data
[0] == '[';
556 return key
== null_type
;
559 bool isinterface (_Jv_BytecodeVerifier
*verifier
)
562 if (key
!= reference_type
)
564 return data
.klass
->isInterface ();
567 bool isabstract (_Jv_BytecodeVerifier
*verifier
)
570 if (key
!= reference_type
)
572 using namespace java::lang::reflect
;
573 return Modifier::isAbstract (data
.klass
->getModifiers ());
576 // Return the element type of an array.
577 type
element_type (_Jv_BytecodeVerifier
*verifier
)
579 // FIXME: maybe should do string manipulation here.
581 if (key
!= reference_type
)
582 verifier
->verify_fail ("programmer error in type::element_type()", -1);
584 jclass k
= data
.klass
->getComponentType ();
585 if (k
->isPrimitive ())
586 return type (verifier
->get_type_val_for_signature (k
));
590 // Return the array type corresponding to an initialized
591 // reference. We could expand this to work for other kinds of
592 // types, but currently we don't need to.
593 type
to_array (_Jv_BytecodeVerifier
*verifier
)
595 // Resolving isn't ideal, because it might force us to load
596 // another class, but it's easy. FIXME?
597 if (key
== unresolved_reference_type
)
600 if (key
== reference_type
)
601 return type (_Jv_GetArrayClass (data
.klass
,
602 data
.klass
->getClassLoaderInternal()));
604 verifier
->verify_fail ("internal error in type::to_array()");
607 bool isreference () const
609 return key
>= reference_type
;
617 bool isinitialized () const
619 return (key
== reference_type
621 || key
== unresolved_reference_type
);
624 bool isresolved () const
626 return (key
== reference_type
628 || key
== uninitialized_reference_type
);
631 void verify_dimensions (int ndims
, _Jv_BytecodeVerifier
*verifier
)
633 // The way this is written, we don't need to check isarray().
634 if (key
== reference_type
)
636 jclass k
= data
.klass
;
637 while (k
->isArray () && ndims
> 0)
639 k
= k
->getComponentType ();
645 // We know KEY == unresolved_reference_type.
646 char *p
= data
.name
->data
;
647 while (*p
++ == '[' && ndims
-- > 0)
652 verifier
->verify_fail ("array type has fewer dimensions than required");
655 // Merge OLD_TYPE into this. On error throw exception.
656 bool merge (type
& old_type
, bool local_semantics
,
657 _Jv_BytecodeVerifier
*verifier
)
659 bool changed
= false;
660 bool refo
= old_type
.isreference ();
661 bool refn
= isreference ();
664 if (old_type
.key
== null_type
)
666 else if (key
== null_type
)
671 else if (isinitialized () != old_type
.isinitialized ())
672 verifier
->verify_fail ("merging initialized and uninitialized types");
675 if (! isinitialized ())
679 else if (old_type
.pc
== UNINIT
)
681 else if (pc
!= old_type
.pc
)
682 verifier
->verify_fail ("merging different uninitialized types");
686 && ! old_type
.isresolved ()
687 && _Jv_equalUtf8Consts (data
.name
, old_type
.data
.name
))
689 // Types are identical.
694 old_type
.resolve (verifier
);
696 jclass k
= data
.klass
;
697 jclass oldk
= old_type
.data
.klass
;
700 while (k
->isArray () && oldk
->isArray ())
703 k
= k
->getComponentType ();
704 oldk
= oldk
->getComponentType ();
707 // Ordinarily this terminates when we hit Object...
710 if (is_assignable_from_slow (k
, oldk
))
712 k
= k
->getSuperclass ();
715 // ... but K could have been an interface, in which
716 // case we'll end up here. We just convert this
719 k
= &java::lang::Object::class$
;
723 while (arraycount
> 0)
725 java::lang::ClassLoader
*loader
726 = verifier
->current_class
->getClassLoaderInternal();
727 k
= _Jv_GetArrayClass (k
, loader
);
735 else if (refo
|| refn
|| key
!= old_type
.key
)
739 // If we're merging into an "unused" slot, then we
740 // simply accept whatever we're merging from.
741 if (key
== unused_by_subroutine_type
)
746 else if (old_type
.key
== unused_by_subroutine_type
)
750 // If we already have an `unsuitable' type, then we
751 // don't need to change again.
752 else if (key
!= unsuitable_type
)
754 key
= unsuitable_type
;
759 verifier
->verify_fail ("unmergeable type");
765 void print (void) const
770 case boolean_type
: c
= 'Z'; break;
771 case byte_type
: c
= 'B'; break;
772 case char_type
: c
= 'C'; break;
773 case short_type
: c
= 'S'; break;
774 case int_type
: c
= 'I'; break;
775 case long_type
: c
= 'J'; break;
776 case float_type
: c
= 'F'; break;
777 case double_type
: c
= 'D'; break;
778 case void_type
: c
= 'V'; break;
779 case unsuitable_type
: c
= '-'; break;
780 case return_address_type
: c
= 'r'; break;
781 case continuation_type
: c
= '+'; break;
782 case unused_by_subroutine_type
: c
= '_'; break;
783 case reference_type
: c
= 'L'; break;
784 case null_type
: c
= '@'; break;
785 case unresolved_reference_type
: c
= 'l'; break;
786 case uninitialized_reference_type
: c
= 'U'; break;
787 case uninitialized_unresolved_reference_type
: c
= 'u'; break;
789 debug_print ("%c", c
);
791 #endif /* VERIFY_DEBUG */
794 // This class holds all the state information we need for a given
798 // The current top of the stack, in terms of slots.
800 // The current depth of the stack. This will be larger than
801 // STACKTOP when wide types are on the stack.
805 // The local variables.
807 // This is used in subroutines to keep track of which local
808 // variables have been accessed.
810 // If not 0, then we are in a subroutine. The value is the PC of
811 // the subroutine's entry point. We can use 0 as an exceptional
812 // value because PC=0 can never be a subroutine.
814 // This is used to keep a linked list of all the states which
815 // require re-verification. We use the PC to keep track.
817 // We keep track of the type of `this' specially. This is used to
818 // ensure that an instance initializer invokes another initializer
819 // on `this' before returning. We must keep track of this
820 // specially because otherwise we might be confused by code which
821 // assigns to locals[0] (overwriting `this') and then returns
822 // without really initializing.
824 // This is a list of all subroutines that have been seen at this
825 // point. Ordinarily this is NULL; it is only allocated and used
826 // in relatively weird situations involving non-ret exit from a
827 // subroutine. We have to keep track of this in this way to avoid
828 // endless recursion in these cases.
829 subr_info
*seen_subrs
;
831 // INVALID marks a state which is not on the linked list of states
832 // requiring reverification.
833 static const int INVALID
= -1;
834 // NO_NEXT marks the state at the end of the reverification list.
835 static const int NO_NEXT
= -2;
837 // This is used to mark the stack depth at the instruction just
838 // after a `jsr' when we haven't yet processed the corresponding
839 // `ret'. See handle_jsr_insn for more information.
840 static const int NO_STACK
= -1;
847 local_changed
= NULL
;
851 state (int max_stack
, int max_locals
)
856 stack
= new type
[max_stack
];
857 for (int i
= 0; i
< max_stack
; ++i
)
858 stack
[i
] = unsuitable_type
;
859 locals
= new type
[max_locals
];
860 local_changed
= (bool *) _Jv_Malloc (sizeof (bool) * max_locals
);
862 for (int i
= 0; i
< max_locals
; ++i
)
864 locals
[i
] = unsuitable_type
;
865 local_changed
[i
] = false;
871 state (const state
*orig
, int max_stack
, int max_locals
,
872 bool ret_semantics
= false)
874 stack
= new type
[max_stack
];
875 locals
= new type
[max_locals
];
876 local_changed
= (bool *) _Jv_Malloc (sizeof (bool) * max_locals
);
878 copy (orig
, max_stack
, max_locals
, ret_semantics
);
889 _Jv_Free (local_changed
);
893 void *operator new[] (size_t bytes
)
895 return _Jv_Malloc (bytes
);
898 void operator delete[] (void *mem
)
903 void *operator new (size_t bytes
)
905 return _Jv_Malloc (bytes
);
908 void operator delete (void *mem
)
915 subr_info
*info
= seen_subrs
;
918 subr_info
*next
= info
->next
;
924 void copy (const state
*copy
, int max_stack
, int max_locals
,
925 bool ret_semantics
= false)
927 stacktop
= copy
->stacktop
;
928 stackdepth
= copy
->stackdepth
;
929 subroutine
= copy
->subroutine
;
930 for (int i
= 0; i
< max_stack
; ++i
)
931 stack
[i
] = copy
->stack
[i
];
932 for (int i
= 0; i
< max_locals
; ++i
)
934 // See push_jump_merge to understand this case.
936 locals
[i
] = type (copy
->local_changed
[i
]
938 : unused_by_subroutine_type
);
940 locals
[i
] = copy
->locals
[i
];
941 local_changed
[i
] = copy
->local_changed
[i
];
945 if (copy
->seen_subrs
)
947 for (subr_info
*info
= seen_subrs
; info
!= NULL
; info
= info
->next
)
953 this_type
= copy
->this_type
;
954 // Don't modify `next'.
957 // Modify this state to reflect entry to an exception handler.
958 void set_exception (type t
, int max_stack
)
963 for (int i
= stacktop
; i
< max_stack
; ++i
)
964 stack
[i
] = unsuitable_type
;
967 // Modify this state to reflect entry into a subroutine.
968 void enter_subroutine (int npc
, int max_locals
)
971 // Mark all items as unchanged. Each subroutine needs to keep
972 // track of its `changed' state independently. In the case of
973 // nested subroutines, this information will be merged back into
974 // parent by the `ret'.
975 for (int i
= 0; i
< max_locals
; ++i
)
976 local_changed
[i
] = false;
979 // Indicate that we've been in this this subroutine.
980 void add_subr (int pc
)
982 subr_info
*n
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
984 n
->next
= seen_subrs
;
988 // Merge STATE_OLD into this state. Destructively modifies this
989 // state. Returns true if the new state was in fact changed.
990 // Will throw an exception if the states are not mergeable.
991 bool merge (state
*state_old
, bool ret_semantics
,
992 int max_locals
, _Jv_BytecodeVerifier
*verifier
)
994 bool changed
= false;
996 // Special handling for `this'. If one or the other is
997 // uninitialized, then the merge is uninitialized.
998 if (this_type
.isinitialized ())
999 this_type
= state_old
->this_type
;
1001 // Merge subroutine states. Here we just keep track of what
1002 // subroutine we think we're in. We only check for a merge
1003 // (which is invalid) when we see a `ret'.
1004 if (subroutine
== state_old
->subroutine
)
1008 else if (subroutine
== 0)
1010 subroutine
= state_old
->subroutine
;
1015 // If the subroutines differ, and we haven't seen this
1016 // subroutine before, indicate that the state changed. This
1017 // is needed to detect when subroutines have merged.
1019 for (subr_info
*info
= seen_subrs
; info
!= NULL
; info
= info
->next
)
1021 if (info
->pc
== state_old
->subroutine
)
1029 add_subr (state_old
->subroutine
);
1034 // Merge stacks. Special handling for NO_STACK case.
1035 if (state_old
->stacktop
== NO_STACK
)
1037 // Nothing to do in this case; we don't care about modifying
1040 else if (stacktop
== NO_STACK
)
1042 stacktop
= state_old
->stacktop
;
1043 stackdepth
= state_old
->stackdepth
;
1044 for (int i
= 0; i
< stacktop
; ++i
)
1045 stack
[i
] = state_old
->stack
[i
];
1048 else if (state_old
->stacktop
!= stacktop
)
1049 verifier
->verify_fail ("stack sizes differ");
1052 for (int i
= 0; i
< state_old
->stacktop
; ++i
)
1054 if (stack
[i
].merge (state_old
->stack
[i
], false, verifier
))
1059 // Merge local variables.
1060 for (int i
= 0; i
< max_locals
; ++i
)
1062 // If we're not processing a `ret', then we merge every
1063 // local variable. If we are processing a `ret', then we
1064 // only merge locals which changed in the subroutine. When
1065 // processing a `ret', STATE_OLD is the state at the point
1066 // of the `ret', and THIS is the state just after the `jsr'.
1067 if (! ret_semantics
|| state_old
->local_changed
[i
])
1069 if (locals
[i
].merge (state_old
->locals
[i
], true, verifier
))
1071 // Note that we don't call `note_variable' here.
1072 // This change doesn't represent a real change to a
1073 // local, but rather a merge artifact. If we're in
1074 // a subroutine which is called with two
1075 // incompatible types in a slot that is unused by
1076 // the subroutine, then we don't want to mark that
1077 // variable as having been modified.
1082 // If we're in a subroutine, we must compute the union of
1083 // all the changed local variables.
1084 if (state_old
->local_changed
[i
])
1091 // Throw an exception if there is an uninitialized object on the
1092 // stack or in a local variable. EXCEPTION_SEMANTICS controls
1093 // whether we're using backwards-branch or exception-handing
1095 void check_no_uninitialized_objects (int max_locals
,
1096 _Jv_BytecodeVerifier
*verifier
,
1097 bool exception_semantics
= false)
1099 if (! exception_semantics
)
1101 for (int i
= 0; i
< stacktop
; ++i
)
1102 if (stack
[i
].isreference () && ! stack
[i
].isinitialized ())
1103 verifier
->verify_fail ("uninitialized object on stack");
1106 for (int i
= 0; i
< max_locals
; ++i
)
1107 if (locals
[i
].isreference () && ! locals
[i
].isinitialized ())
1108 verifier
->verify_fail ("uninitialized object in local variable");
1110 check_this_initialized (verifier
);
1113 // Ensure that `this' has been initialized.
1114 void check_this_initialized (_Jv_BytecodeVerifier
*verifier
)
1116 if (this_type
.isreference () && ! this_type
.isinitialized ())
1117 verifier
->verify_fail ("`this' is uninitialized");
1120 // Set type of `this'.
1121 void set_this_type (const type
&k
)
1126 // Note that a local variable was modified.
1127 void note_variable (int index
)
1130 local_changed
[index
] = true;
1133 // Mark each `new'd object we know of that was allocated at PC as
1135 void set_initialized (int pc
, int max_locals
)
1137 for (int i
= 0; i
< stacktop
; ++i
)
1138 stack
[i
].set_initialized (pc
);
1139 for (int i
= 0; i
< max_locals
; ++i
)
1140 locals
[i
].set_initialized (pc
);
1141 this_type
.set_initialized (pc
);
1144 // Return true if this state is the unmerged result of a `ret'.
1145 bool is_unmerged_ret_state (int max_locals
) const
1147 if (stacktop
== NO_STACK
)
1149 for (int i
= 0; i
< max_locals
; ++i
)
1151 if (locals
[i
].key
== unused_by_subroutine_type
)
1158 void print (const char *leader
, int pc
,
1159 int max_stack
, int max_locals
) const
1161 debug_print ("%s [%4d]: [stack] ", leader
, pc
);
1163 for (i
= 0; i
< stacktop
; ++i
)
1165 for (; i
< max_stack
; ++i
)
1167 debug_print (" [local] ");
1168 for (i
= 0; i
< max_locals
; ++i
)
1171 debug_print (local_changed
[i
] ? "+" : " ");
1173 if (subroutine
== 0)
1174 debug_print (" | None");
1176 debug_print (" | %4d", subroutine
);
1177 debug_print (" | %p\n", this);
1180 inline void print (const char *, int, int, int) const
1183 #endif /* VERIFY_DEBUG */
1188 if (current_state
->stacktop
<= 0)
1189 verify_fail ("stack empty");
1190 type r
= current_state
->stack
[--current_state
->stacktop
];
1191 current_state
->stackdepth
-= r
.depth ();
1192 if (current_state
->stackdepth
< 0)
1193 verify_fail ("stack empty", start_PC
);
1199 type r
= pop_raw ();
1201 verify_fail ("narrow pop of wide type");
1205 type
pop_type (type match
)
1208 type t
= pop_raw ();
1209 if (! match
.compatible (t
, this))
1210 verify_fail ("incompatible type on stack");
1214 // Pop a reference which is guaranteed to be initialized. MATCH
1215 // doesn't have to be a reference type; in this case this acts like
1217 type
pop_init_ref (type match
)
1219 type t
= pop_raw ();
1220 if (t
.isreference () && ! t
.isinitialized ())
1221 verify_fail ("initialized reference required");
1222 else if (! match
.compatible (t
, this))
1223 verify_fail ("incompatible type on stack");
1227 // Pop a reference type or a return address.
1228 type
pop_ref_or_return ()
1230 type t
= pop_raw ();
1231 if (! t
.isreference () && t
.key
!= return_address_type
)
1232 verify_fail ("expected reference or return address on stack");
1236 void push_type (type t
)
1238 // If T is a numeric type like short, promote it to int.
1241 int depth
= t
.depth ();
1242 if (current_state
->stackdepth
+ depth
> current_method
->max_stack
)
1243 verify_fail ("stack overflow");
1244 current_state
->stack
[current_state
->stacktop
++] = t
;
1245 current_state
->stackdepth
+= depth
;
1248 void set_variable (int index
, type t
)
1250 // If T is a numeric type like short, promote it to int.
1253 int depth
= t
.depth ();
1254 if (index
> current_method
->max_locals
- depth
)
1255 verify_fail ("invalid local variable");
1256 current_state
->locals
[index
] = t
;
1257 current_state
->note_variable (index
);
1261 current_state
->locals
[index
+ 1] = continuation_type
;
1262 current_state
->note_variable (index
+ 1);
1264 if (index
> 0 && current_state
->locals
[index
- 1].iswide ())
1266 current_state
->locals
[index
- 1] = unsuitable_type
;
1267 // There's no need to call note_variable here.
1271 type
get_variable (int index
, type t
)
1273 int depth
= t
.depth ();
1274 if (index
> current_method
->max_locals
- depth
)
1275 verify_fail ("invalid local variable");
1276 if (! t
.compatible (current_state
->locals
[index
], this))
1277 verify_fail ("incompatible type in local variable");
1280 type
t (continuation_type
);
1281 if (! current_state
->locals
[index
+ 1].compatible (t
, this))
1282 verify_fail ("invalid local variable");
1284 return current_state
->locals
[index
];
1287 // Make sure ARRAY is an array type and that its elements are
1288 // compatible with type ELEMENT. Returns the actual element type.
1289 type
require_array_type (type array
, type element
)
1291 // An odd case. Here we just pretend that everything went ok. If
1292 // the requested element type is some kind of reference, return
1293 // the null type instead.
1294 if (array
.isnull ())
1295 return element
.isreference () ? type (null_type
) : element
;
1297 if (! array
.isarray ())
1298 verify_fail ("array required");
1300 type t
= array
.element_type (this);
1301 if (! element
.compatible (t
, this))
1303 // Special case for byte arrays, which must also be boolean
1306 if (element
.key
== byte_type
)
1308 type
e2 (boolean_type
);
1309 ok
= e2
.compatible (t
, this);
1312 verify_fail ("incompatible array element type");
1315 // Return T and not ELEMENT, because T might be specialized.
1321 if (PC
>= current_method
->code_length
)
1322 verify_fail ("premature end of bytecode");
1323 return (jint
) bytecode
[PC
++] & 0xff;
1328 jint b1
= get_byte ();
1329 jint b2
= get_byte ();
1330 return (jint
) ((b1
<< 8) | b2
) & 0xffff;
1335 jint b1
= get_byte ();
1336 jint b2
= get_byte ();
1337 jshort s
= (b1
<< 8) | b2
;
1343 jint b1
= get_byte ();
1344 jint b2
= get_byte ();
1345 jint b3
= get_byte ();
1346 jint b4
= get_byte ();
1347 return (b1
<< 24) | (b2
<< 16) | (b3
<< 8) | b4
;
1350 int compute_jump (int offset
)
1352 int npc
= start_PC
+ offset
;
1353 if (npc
< 0 || npc
>= current_method
->code_length
)
1354 verify_fail ("branch out of range", start_PC
);
1358 // Merge the indicated state into the state at the branch target and
1359 // schedule a new PC if there is a change. If RET_SEMANTICS is
1360 // true, then we are merging from a `ret' instruction into the
1361 // instruction after a `jsr'. This is a special case with its own
1362 // modified semantics.
1363 void push_jump_merge (int npc
, state
*nstate
, bool ret_semantics
= false)
1365 bool changed
= true;
1366 if (states
[npc
] == NULL
)
1368 // There's a weird situation here. If are examining the
1369 // branch that results from a `ret', and there is not yet a
1370 // state available at the branch target (the instruction just
1371 // after the `jsr'), then we have to construct a special kind
1372 // of state at that point for future merging. This special
1373 // state has the type `unused_by_subroutine_type' in each slot
1374 // which was not modified by the subroutine.
1375 states
[npc
] = new state (nstate
, current_method
->max_stack
,
1376 current_method
->max_locals
, ret_semantics
);
1377 debug_print ("== New state in push_jump_merge\n");
1378 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1379 current_method
->max_locals
);
1383 debug_print ("== Merge states in push_jump_merge\n");
1384 nstate
->print ("Frm", start_PC
, current_method
->max_stack
,
1385 current_method
->max_locals
);
1386 states
[npc
]->print (" To", npc
, current_method
->max_stack
,
1387 current_method
->max_locals
);
1388 changed
= states
[npc
]->merge (nstate
, ret_semantics
,
1389 current_method
->max_locals
, this);
1390 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1391 current_method
->max_locals
);
1394 if (changed
&& states
[npc
]->next
== state::INVALID
)
1396 // The merge changed the state, and the new PC isn't yet on our
1397 // list of PCs to re-verify.
1398 states
[npc
]->next
= next_verify_pc
;
1399 next_verify_pc
= npc
;
1403 void push_jump (int offset
)
1405 int npc
= compute_jump (offset
);
1407 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1408 push_jump_merge (npc
, current_state
);
1411 void push_exception_jump (type t
, int pc
)
1413 current_state
->check_no_uninitialized_objects (current_method
->max_locals
,
1415 state
s (current_state
, current_method
->max_stack
,
1416 current_method
->max_locals
);
1417 if (current_method
->max_stack
< 1)
1418 verify_fail ("stack overflow at exception handler");
1419 s
.set_exception (t
, current_method
->max_stack
);
1420 push_jump_merge (pc
, &s
);
1425 int *prev_loc
= &next_verify_pc
;
1426 int npc
= next_verify_pc
;
1428 while (npc
!= state::NO_NEXT
)
1430 // If the next available PC is an unmerged `ret' state, then
1431 // we aren't yet ready to handle it. That's because we would
1432 // need all kind of special cases to do so. So instead we
1433 // defer this jump until after we've processed it via a
1434 // fall-through. This has to happen because the instruction
1435 // before this one must be a `jsr'.
1436 if (! states
[npc
]->is_unmerged_ret_state (current_method
->max_locals
))
1438 *prev_loc
= states
[npc
]->next
;
1439 states
[npc
]->next
= state::INVALID
;
1443 prev_loc
= &states
[npc
]->next
;
1444 npc
= states
[npc
]->next
;
1447 // Note that we might have gotten here even when there are
1448 // remaining states to process. That can happen if we find a
1449 // `jsr' without a `ret'.
1450 return state::NO_NEXT
;
1453 void invalidate_pc ()
1455 PC
= state::NO_NEXT
;
1458 void note_branch_target (int pc
, bool is_jsr_target
= false)
1460 // Don't check `pc <= PC', because we've advanced PC after
1461 // fetching the target and we haven't yet checked the next
1463 if (pc
< PC
&& ! (flags
[pc
] & FLAG_INSN_START
))
1464 verify_fail ("branch not to instruction start", start_PC
);
1465 flags
[pc
] |= FLAG_BRANCH_TARGET
;
1468 // Record the jsr which called this instruction.
1469 subr_info
*info
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
1471 info
->next
= jsr_ptrs
[pc
];
1472 jsr_ptrs
[pc
] = info
;
1476 void skip_padding ()
1478 while ((PC
% 4) > 0)
1479 if (get_byte () != 0)
1480 verify_fail ("found nonzero padding byte");
1483 // Return the subroutine to which the instruction at PC belongs.
1484 int get_subroutine (int pc
)
1486 if (states
[pc
] == NULL
)
1488 return states
[pc
]->subroutine
;
1491 // Do the work for a `ret' instruction. INDEX is the index into the
1493 void handle_ret_insn (int index
)
1495 get_variable (index
, return_address_type
);
1497 int csub
= current_state
->subroutine
;
1499 verify_fail ("no subroutine");
1501 // Check to see if we've merged subroutines.
1502 subr_entry_info
*entry
;
1503 for (entry
= entry_points
; entry
!= NULL
; entry
= entry
->next
)
1505 if (entry
->ret_pc
== start_PC
)
1510 entry
= (subr_entry_info
*) _Jv_Malloc (sizeof (subr_entry_info
));
1512 entry
->ret_pc
= start_PC
;
1513 entry
->next
= entry_points
;
1514 entry_points
= entry
;
1516 else if (entry
->pc
!= csub
)
1517 verify_fail ("subroutines merged");
1519 for (subr_info
*subr
= jsr_ptrs
[csub
]; subr
!= NULL
; subr
= subr
->next
)
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 (subr
->pc
>= current_method
->code_length
)
1525 verify_fail ("fell off end");
1527 // Temporarily modify the current state so it looks like we're
1528 // in the enclosing context.
1529 current_state
->subroutine
= get_subroutine (subr
->pc
);
1531 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1532 push_jump_merge (subr
->pc
, current_state
, true);
1535 current_state
->subroutine
= csub
;
1539 // We're in the subroutine SUB, calling a subroutine at DEST. Make
1540 // sure this subroutine isn't already on the stack.
1541 void check_nonrecursive_call (int sub
, int dest
)
1546 verify_fail ("recursive subroutine call");
1547 for (subr_info
*info
= jsr_ptrs
[sub
]; info
!= NULL
; info
= info
->next
)
1548 check_nonrecursive_call (get_subroutine (info
->pc
), dest
);
1551 void handle_jsr_insn (int offset
)
1553 int npc
= compute_jump (offset
);
1556 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1557 check_nonrecursive_call (current_state
->subroutine
, npc
);
1559 // Modify our state as appropriate for entry into a subroutine.
1560 push_type (return_address_type
);
1561 push_jump_merge (npc
, current_state
);
1563 pop_type (return_address_type
);
1565 // On entry to the subroutine, the subroutine number must be set
1566 // and the locals must be marked as cleared. We do this after
1567 // merging state so that we don't erroneously "notice" a variable
1568 // change merely on entry.
1569 states
[npc
]->enter_subroutine (npc
, current_method
->max_locals
);
1571 // Indicate that we don't know the stack depth of the instruction
1572 // following the `jsr'. The idea here is that we need to merge
1573 // the local variable state across the jsr, but the subroutine
1574 // might change the stack depth, so we can't make any assumptions
1575 // about it. So we have yet another special case. We know that
1576 // at this point PC points to the instruction after the jsr. Note
1577 // that it is ok to have a `jsr' at the end of the bytecode,
1578 // provided that the called subroutine never returns. So, we have
1579 // a special case here and another one when we handle the ret.
1580 if (PC
< current_method
->code_length
)
1582 current_state
->stacktop
= state::NO_STACK
;
1583 push_jump_merge (PC
, current_state
);
1588 jclass
construct_primitive_array_type (type_val prim
)
1594 k
= JvPrimClass (boolean
);
1597 k
= JvPrimClass (char);
1600 k
= JvPrimClass (float);
1603 k
= JvPrimClass (double);
1606 k
= JvPrimClass (byte
);
1609 k
= JvPrimClass (short);
1612 k
= JvPrimClass (int);
1615 k
= JvPrimClass (long);
1618 // These aren't used here but we call them out to avoid
1621 case unsuitable_type
:
1622 case return_address_type
:
1623 case continuation_type
:
1624 case unused_by_subroutine_type
:
1625 case reference_type
:
1627 case unresolved_reference_type
:
1628 case uninitialized_reference_type
:
1629 case uninitialized_unresolved_reference_type
:
1631 verify_fail ("unknown type in construct_primitive_array_type");
1633 k
= _Jv_GetArrayClass (k
, NULL
);
1637 // This pass computes the location of branch targets and also
1638 // instruction starts.
1639 void branch_prepass ()
1641 flags
= (char *) _Jv_Malloc (current_method
->code_length
);
1642 jsr_ptrs
= (subr_info
**) _Jv_Malloc (sizeof (subr_info
*)
1643 * current_method
->code_length
);
1645 for (int i
= 0; i
< current_method
->code_length
; ++i
)
1651 bool last_was_jsr
= false;
1654 while (PC
< current_method
->code_length
)
1656 // Set `start_PC' early so that error checking can have the
1659 flags
[PC
] |= FLAG_INSN_START
;
1661 // If the previous instruction was a jsr, then the next
1662 // instruction is a branch target -- the branch being the
1663 // corresponding `ret'.
1665 note_branch_target (PC
);
1666 last_was_jsr
= false;
1668 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
1672 case op_aconst_null
:
1808 case op_monitorenter
:
1809 case op_monitorexit
:
1817 case op_arraylength
:
1849 case op_invokespecial
:
1850 case op_invokestatic
:
1851 case op_invokevirtual
:
1855 case op_multianewarray
:
1861 last_was_jsr
= true;
1880 note_branch_target (compute_jump (get_short ()), last_was_jsr
);
1883 case op_tableswitch
:
1886 note_branch_target (compute_jump (get_int ()));
1887 jint low
= get_int ();
1888 jint hi
= get_int ();
1890 verify_fail ("invalid tableswitch", start_PC
);
1891 for (int i
= low
; i
<= hi
; ++i
)
1892 note_branch_target (compute_jump (get_int ()));
1896 case op_lookupswitch
:
1899 note_branch_target (compute_jump (get_int ()));
1900 int npairs
= get_int ();
1902 verify_fail ("too few pairs in lookupswitch", start_PC
);
1903 while (npairs
-- > 0)
1906 note_branch_target (compute_jump (get_int ()));
1911 case op_invokeinterface
:
1919 opcode
= (java_opcode
) get_byte ();
1921 if (opcode
== op_iinc
)
1927 last_was_jsr
= true;
1930 note_branch_target (compute_jump (get_int ()), last_was_jsr
);
1933 // These are unused here, but we call them out explicitly
1934 // so that -Wswitch-enum doesn't complain.
1940 case op_putstatic_1
:
1941 case op_putstatic_2
:
1942 case op_putstatic_4
:
1943 case op_putstatic_8
:
1944 case op_putstatic_a
:
1946 case op_getfield_2s
:
1947 case op_getfield_2u
:
1951 case op_getstatic_1
:
1952 case op_getstatic_2s
:
1953 case op_getstatic_2u
:
1954 case op_getstatic_4
:
1955 case op_getstatic_8
:
1956 case op_getstatic_a
:
1958 verify_fail ("unrecognized instruction in branch_prepass",
1962 // See if any previous branch tried to branch to the middle of
1963 // this instruction.
1964 for (int pc
= start_PC
+ 1; pc
< PC
; ++pc
)
1966 if ((flags
[pc
] & FLAG_BRANCH_TARGET
))
1967 verify_fail ("branch to middle of instruction", pc
);
1971 // Verify exception handlers.
1972 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
1974 if (! (flags
[exception
[i
].handler_pc
.i
] & FLAG_INSN_START
))
1975 verify_fail ("exception handler not at instruction start",
1976 exception
[i
].handler_pc
.i
);
1977 if (! (flags
[exception
[i
].start_pc
.i
] & FLAG_INSN_START
))
1978 verify_fail ("exception start not at instruction start",
1979 exception
[i
].start_pc
.i
);
1980 if (exception
[i
].end_pc
.i
!= current_method
->code_length
1981 && ! (flags
[exception
[i
].end_pc
.i
] & FLAG_INSN_START
))
1982 verify_fail ("exception end not at instruction start",
1983 exception
[i
].end_pc
.i
);
1985 flags
[exception
[i
].handler_pc
.i
] |= FLAG_BRANCH_TARGET
;
1989 void check_pool_index (int index
)
1991 if (index
< 0 || index
>= current_class
->constants
.size
)
1992 verify_fail ("constant pool index out of range", start_PC
);
1995 type
check_class_constant (int index
)
1997 check_pool_index (index
);
1998 _Jv_Constants
*pool
= ¤t_class
->constants
;
1999 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedClass
)
2000 return type (pool
->data
[index
].clazz
);
2001 else if (pool
->tags
[index
] == JV_CONSTANT_Class
)
2002 return type (pool
->data
[index
].utf8
);
2003 verify_fail ("expected class constant", start_PC
);
2006 type
check_constant (int index
)
2008 check_pool_index (index
);
2009 _Jv_Constants
*pool
= ¤t_class
->constants
;
2010 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedString
2011 || pool
->tags
[index
] == JV_CONSTANT_String
)
2012 return type (&java::lang::String::class$
);
2013 else if (pool
->tags
[index
] == JV_CONSTANT_Integer
)
2014 return type (int_type
);
2015 else if (pool
->tags
[index
] == JV_CONSTANT_Float
)
2016 return type (float_type
);
2017 verify_fail ("String, int, or float constant expected", start_PC
);
2020 type
check_wide_constant (int index
)
2022 check_pool_index (index
);
2023 _Jv_Constants
*pool
= ¤t_class
->constants
;
2024 if (pool
->tags
[index
] == JV_CONSTANT_Long
)
2025 return type (long_type
);
2026 else if (pool
->tags
[index
] == JV_CONSTANT_Double
)
2027 return type (double_type
);
2028 verify_fail ("long or double constant expected", start_PC
);
2031 // Helper for both field and method. These are laid out the same in
2032 // the constant pool.
2033 type
handle_field_or_method (int index
, int expected
,
2034 _Jv_Utf8Const
**name
,
2035 _Jv_Utf8Const
**fmtype
)
2037 check_pool_index (index
);
2038 _Jv_Constants
*pool
= ¤t_class
->constants
;
2039 if (pool
->tags
[index
] != expected
)
2040 verify_fail ("didn't see expected constant", start_PC
);
2041 // Once we know we have a Fieldref or Methodref we assume that it
2042 // is correctly laid out in the constant pool. I think the code
2043 // in defineclass.cc guarantees this.
2044 _Jv_ushort class_index
, name_and_type_index
;
2045 _Jv_loadIndexes (&pool
->data
[index
],
2047 name_and_type_index
);
2048 _Jv_ushort name_index
, desc_index
;
2049 _Jv_loadIndexes (&pool
->data
[name_and_type_index
],
2050 name_index
, desc_index
);
2052 *name
= pool
->data
[name_index
].utf8
;
2053 *fmtype
= pool
->data
[desc_index
].utf8
;
2055 return check_class_constant (class_index
);
2058 // Return field's type, compute class' type if requested.
2059 type
check_field_constant (int index
, type
*class_type
= NULL
)
2061 _Jv_Utf8Const
*name
, *field_type
;
2062 type ct
= handle_field_or_method (index
,
2063 JV_CONSTANT_Fieldref
,
2064 &name
, &field_type
);
2067 if (field_type
->data
[0] == '[' || field_type
->data
[0] == 'L')
2068 return type (field_type
);
2069 return get_type_val_for_signature (field_type
->data
[0]);
2072 type
check_method_constant (int index
, bool is_interface
,
2073 _Jv_Utf8Const
**method_name
,
2074 _Jv_Utf8Const
**method_signature
)
2076 return handle_field_or_method (index
,
2078 ? JV_CONSTANT_InterfaceMethodref
2079 : JV_CONSTANT_Methodref
),
2080 method_name
, method_signature
);
2083 type
get_one_type (char *&p
)
2101 _Jv_Utf8Const
*name
= make_utf8_const (start
, p
- start
);
2105 // Casting to jchar here is ok since we are looking at an ASCII
2107 type_val rt
= get_type_val_for_signature (jchar (v
));
2109 if (arraycount
== 0)
2111 // Callers of this function eventually push their arguments on
2112 // the stack. So, promote them here.
2113 return type (rt
).promote ();
2116 jclass k
= construct_primitive_array_type (rt
);
2117 while (--arraycount
> 0)
2118 k
= _Jv_GetArrayClass (k
, NULL
);
2122 void compute_argument_types (_Jv_Utf8Const
*signature
,
2125 char *p
= signature
->data
;
2131 types
[i
++] = get_one_type (p
);
2134 type
compute_return_type (_Jv_Utf8Const
*signature
)
2136 char *p
= signature
->data
;
2140 return get_one_type (p
);
2143 void check_return_type (type onstack
)
2145 type rt
= compute_return_type (current_method
->self
->signature
);
2146 if (! rt
.compatible (onstack
, this))
2147 verify_fail ("incompatible return type");
2150 // Initialize the stack for the new method. Returns true if this
2151 // method is an instance initializer.
2152 bool initialize_stack ()
2155 bool is_init
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2157 bool is_clinit
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2160 using namespace java::lang::reflect
;
2161 if (! Modifier::isStatic (current_method
->self
->accflags
))
2163 type
kurr (current_class
);
2166 kurr
.set_uninitialized (type::SELF
, this);
2170 verify_fail ("<clinit> method must be static");
2171 set_variable (0, kurr
);
2172 current_state
->set_this_type (kurr
);
2178 verify_fail ("<init> method must be non-static");
2181 // We have to handle wide arguments specially here.
2182 int arg_count
= _Jv_count_arguments (current_method
->self
->signature
);
2183 type arg_types
[arg_count
];
2184 compute_argument_types (current_method
->self
->signature
, arg_types
);
2185 for (int i
= 0; i
< arg_count
; ++i
)
2187 set_variable (var
, arg_types
[i
]);
2189 if (arg_types
[i
].iswide ())
2196 void verify_instructions_0 ()
2198 current_state
= new state (current_method
->max_stack
,
2199 current_method
->max_locals
);
2204 // True if we are verifying an instance initializer.
2205 bool this_is_init
= initialize_stack ();
2207 states
= (state
**) _Jv_Malloc (sizeof (state
*)
2208 * current_method
->code_length
);
2209 for (int i
= 0; i
< current_method
->code_length
; ++i
)
2212 next_verify_pc
= state::NO_NEXT
;
2216 // If the PC was invalidated, get a new one from the work list.
2217 if (PC
== state::NO_NEXT
)
2220 if (PC
== state::INVALID
)
2221 verify_fail ("can't happen: saw state::INVALID");
2222 if (PC
== state::NO_NEXT
)
2224 debug_print ("== State pop from pending list\n");
2225 // Set up the current state.
2226 current_state
->copy (states
[PC
], current_method
->max_stack
,
2227 current_method
->max_locals
);
2231 // Control can't fall off the end of the bytecode. We
2232 // only need to check this in the fall-through case,
2233 // because branch bounds are checked when they are
2235 if (PC
>= current_method
->code_length
)
2236 verify_fail ("fell off end");
2238 // We only have to do this checking in the situation where
2239 // control flow falls through from the previous
2240 // instruction. Otherwise merging is done at the time we
2242 if (states
[PC
] != NULL
)
2244 // We've already visited this instruction. So merge
2245 // the states together. If this yields no change then
2246 // we don't have to re-verify. However, if the new
2247 // state is an the result of an unmerged `ret', we
2248 // must continue through it.
2249 debug_print ("== Fall through merge\n");
2250 states
[PC
]->print ("Old", PC
, current_method
->max_stack
,
2251 current_method
->max_locals
);
2252 current_state
->print ("Cur", PC
, current_method
->max_stack
,
2253 current_method
->max_locals
);
2254 if (! current_state
->merge (states
[PC
], false,
2255 current_method
->max_locals
, this)
2256 && ! states
[PC
]->is_unmerged_ret_state (current_method
->max_locals
))
2258 debug_print ("== Fall through optimization\n");
2262 // Save a copy of it for later.
2263 states
[PC
]->copy (current_state
, current_method
->max_stack
,
2264 current_method
->max_locals
);
2265 current_state
->print ("New", PC
, current_method
->max_stack
,
2266 current_method
->max_locals
);
2270 // We only have to keep saved state at branch targets. If
2271 // we're at a branch target and the state here hasn't been set
2272 // yet, we set it now.
2273 if (states
[PC
] == NULL
&& (flags
[PC
] & FLAG_BRANCH_TARGET
))
2275 states
[PC
] = new state (current_state
, current_method
->max_stack
,
2276 current_method
->max_locals
);
2279 // Set this before handling exceptions so that debug output is
2283 // Update states for all active exception handlers. Ordinarily
2284 // there are not many exception handlers. So we simply run
2285 // through them all.
2286 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
2288 if (PC
>= exception
[i
].start_pc
.i
&& PC
< exception
[i
].end_pc
.i
)
2290 type
handler (&java::lang::Throwable::class$
);
2291 if (exception
[i
].handler_type
.i
!= 0)
2292 handler
= check_class_constant (exception
[i
].handler_type
.i
);
2293 push_exception_jump (handler
, exception
[i
].handler_pc
.i
);
2297 current_state
->print (" ", PC
, current_method
->max_stack
,
2298 current_method
->max_locals
);
2299 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
2305 case op_aconst_null
:
2306 push_type (null_type
);
2316 push_type (int_type
);
2321 push_type (long_type
);
2327 push_type (float_type
);
2332 push_type (double_type
);
2337 push_type (int_type
);
2342 push_type (int_type
);
2346 push_type (check_constant (get_byte ()));
2349 push_type (check_constant (get_ushort ()));
2352 push_type (check_wide_constant (get_ushort ()));
2356 push_type (get_variable (get_byte (), int_type
));
2359 push_type (get_variable (get_byte (), long_type
));
2362 push_type (get_variable (get_byte (), float_type
));
2365 push_type (get_variable (get_byte (), double_type
));
2368 push_type (get_variable (get_byte (), reference_type
));
2375 push_type (get_variable (opcode
- op_iload_0
, int_type
));
2381 push_type (get_variable (opcode
- op_lload_0
, long_type
));
2387 push_type (get_variable (opcode
- op_fload_0
, float_type
));
2393 push_type (get_variable (opcode
- op_dload_0
, double_type
));
2399 push_type (get_variable (opcode
- op_aload_0
, reference_type
));
2402 pop_type (int_type
);
2403 push_type (require_array_type (pop_init_ref (reference_type
),
2407 pop_type (int_type
);
2408 push_type (require_array_type (pop_init_ref (reference_type
),
2412 pop_type (int_type
);
2413 push_type (require_array_type (pop_init_ref (reference_type
),
2417 pop_type (int_type
);
2418 push_type (require_array_type (pop_init_ref (reference_type
),
2422 pop_type (int_type
);
2423 push_type (require_array_type (pop_init_ref (reference_type
),
2427 pop_type (int_type
);
2428 require_array_type (pop_init_ref (reference_type
), byte_type
);
2429 push_type (int_type
);
2432 pop_type (int_type
);
2433 require_array_type (pop_init_ref (reference_type
), char_type
);
2434 push_type (int_type
);
2437 pop_type (int_type
);
2438 require_array_type (pop_init_ref (reference_type
), short_type
);
2439 push_type (int_type
);
2442 set_variable (get_byte (), pop_type (int_type
));
2445 set_variable (get_byte (), pop_type (long_type
));
2448 set_variable (get_byte (), pop_type (float_type
));
2451 set_variable (get_byte (), pop_type (double_type
));
2454 set_variable (get_byte (), pop_ref_or_return ());
2460 set_variable (opcode
- op_istore_0
, pop_type (int_type
));
2466 set_variable (opcode
- op_lstore_0
, pop_type (long_type
));
2472 set_variable (opcode
- op_fstore_0
, pop_type (float_type
));
2478 set_variable (opcode
- op_dstore_0
, pop_type (double_type
));
2484 set_variable (opcode
- op_astore_0
, pop_ref_or_return ());
2487 pop_type (int_type
);
2488 pop_type (int_type
);
2489 require_array_type (pop_init_ref (reference_type
), int_type
);
2492 pop_type (long_type
);
2493 pop_type (int_type
);
2494 require_array_type (pop_init_ref (reference_type
), long_type
);
2497 pop_type (float_type
);
2498 pop_type (int_type
);
2499 require_array_type (pop_init_ref (reference_type
), float_type
);
2502 pop_type (double_type
);
2503 pop_type (int_type
);
2504 require_array_type (pop_init_ref (reference_type
), double_type
);
2507 pop_type (reference_type
);
2508 pop_type (int_type
);
2509 require_array_type (pop_init_ref (reference_type
), reference_type
);
2512 pop_type (int_type
);
2513 pop_type (int_type
);
2514 require_array_type (pop_init_ref (reference_type
), byte_type
);
2517 pop_type (int_type
);
2518 pop_type (int_type
);
2519 require_array_type (pop_init_ref (reference_type
), char_type
);
2522 pop_type (int_type
);
2523 pop_type (int_type
);
2524 require_array_type (pop_init_ref (reference_type
), short_type
);
2531 type t
= pop_raw ();
2555 type t2
= pop_raw ();
2570 type t
= pop_raw ();
2585 type t1
= pop_raw ();
2602 type t1
= pop_raw ();
2605 type t2
= pop_raw ();
2623 type t3
= pop_raw ();
2661 pop_type (int_type
);
2662 push_type (pop_type (int_type
));
2672 pop_type (long_type
);
2673 push_type (pop_type (long_type
));
2678 pop_type (int_type
);
2679 push_type (pop_type (long_type
));
2686 pop_type (float_type
);
2687 push_type (pop_type (float_type
));
2694 pop_type (double_type
);
2695 push_type (pop_type (double_type
));
2701 push_type (pop_type (int_type
));
2704 push_type (pop_type (long_type
));
2707 push_type (pop_type (float_type
));
2710 push_type (pop_type (double_type
));
2713 get_variable (get_byte (), int_type
);
2717 pop_type (int_type
);
2718 push_type (long_type
);
2721 pop_type (int_type
);
2722 push_type (float_type
);
2725 pop_type (int_type
);
2726 push_type (double_type
);
2729 pop_type (long_type
);
2730 push_type (int_type
);
2733 pop_type (long_type
);
2734 push_type (float_type
);
2737 pop_type (long_type
);
2738 push_type (double_type
);
2741 pop_type (float_type
);
2742 push_type (int_type
);
2745 pop_type (float_type
);
2746 push_type (long_type
);
2749 pop_type (float_type
);
2750 push_type (double_type
);
2753 pop_type (double_type
);
2754 push_type (int_type
);
2757 pop_type (double_type
);
2758 push_type (long_type
);
2761 pop_type (double_type
);
2762 push_type (float_type
);
2765 pop_type (long_type
);
2766 pop_type (long_type
);
2767 push_type (int_type
);
2771 pop_type (float_type
);
2772 pop_type (float_type
);
2773 push_type (int_type
);
2777 pop_type (double_type
);
2778 pop_type (double_type
);
2779 push_type (int_type
);
2787 pop_type (int_type
);
2788 push_jump (get_short ());
2796 pop_type (int_type
);
2797 pop_type (int_type
);
2798 push_jump (get_short ());
2802 pop_type (reference_type
);
2803 pop_type (reference_type
);
2804 push_jump (get_short ());
2807 push_jump (get_short ());
2811 handle_jsr_insn (get_short ());
2814 handle_ret_insn (get_byte ());
2816 case op_tableswitch
:
2818 pop_type (int_type
);
2820 push_jump (get_int ());
2821 jint low
= get_int ();
2822 jint high
= get_int ();
2823 // Already checked LOW -vs- HIGH.
2824 for (int i
= low
; i
<= high
; ++i
)
2825 push_jump (get_int ());
2830 case op_lookupswitch
:
2832 pop_type (int_type
);
2834 push_jump (get_int ());
2835 jint npairs
= get_int ();
2836 // Already checked NPAIRS >= 0.
2838 for (int i
= 0; i
< npairs
; ++i
)
2840 jint key
= get_int ();
2841 if (i
> 0 && key
<= lastkey
)
2842 verify_fail ("lookupswitch pairs unsorted", start_PC
);
2844 push_jump (get_int ());
2850 check_return_type (pop_type (int_type
));
2854 check_return_type (pop_type (long_type
));
2858 check_return_type (pop_type (float_type
));
2862 check_return_type (pop_type (double_type
));
2866 check_return_type (pop_init_ref (reference_type
));
2870 // We only need to check this when the return type is
2871 // void, because all instance initializers return void.
2873 current_state
->check_this_initialized (this);
2874 check_return_type (void_type
);
2878 push_type (check_field_constant (get_ushort ()));
2881 pop_type (check_field_constant (get_ushort ()));
2886 type field
= check_field_constant (get_ushort (), &klass
);
2894 type field
= check_field_constant (get_ushort (), &klass
);
2897 // We have an obscure special case here: we can use
2898 // `putfield' on a field declared in this class, even if
2899 // `this' has not yet been initialized.
2900 if (! current_state
->this_type
.isinitialized ()
2901 && current_state
->this_type
.pc
== type::SELF
)
2902 klass
.set_uninitialized (type::SELF
, this);
2907 case op_invokevirtual
:
2908 case op_invokespecial
:
2909 case op_invokestatic
:
2910 case op_invokeinterface
:
2912 _Jv_Utf8Const
*method_name
, *method_signature
;
2914 = check_method_constant (get_ushort (),
2915 opcode
== op_invokeinterface
,
2918 // NARGS is only used when we're processing
2919 // invokeinterface. It is simplest for us to compute it
2920 // here and then verify it later.
2922 if (opcode
== op_invokeinterface
)
2924 nargs
= get_byte ();
2925 if (get_byte () != 0)
2926 verify_fail ("invokeinterface dummy byte is wrong");
2929 bool is_init
= false;
2930 if (_Jv_equalUtf8Consts (method_name
, gcj::init_name
))
2933 if (opcode
!= op_invokespecial
)
2934 verify_fail ("can't invoke <init>");
2936 else if (method_name
->data
[0] == '<')
2937 verify_fail ("can't invoke method starting with `<'");
2939 // Pop arguments and check types.
2940 int arg_count
= _Jv_count_arguments (method_signature
);
2941 type arg_types
[arg_count
];
2942 compute_argument_types (method_signature
, arg_types
);
2943 for (int i
= arg_count
- 1; i
>= 0; --i
)
2945 // This is only used for verifying the byte for
2947 nargs
-= arg_types
[i
].depth ();
2948 pop_init_ref (arg_types
[i
]);
2951 if (opcode
== op_invokeinterface
2953 verify_fail ("wrong argument count for invokeinterface");
2955 if (opcode
!= op_invokestatic
)
2957 type t
= class_type
;
2960 // In this case the PC doesn't matter.
2961 t
.set_uninitialized (type::UNINIT
, this);
2963 type raw
= pop_raw ();
2965 if (! is_init
&& ! raw
.isinitialized ())
2967 // This is a failure.
2969 else if (is_init
&& raw
.isnull ())
2973 else if (t
.compatible (raw
, this))
2977 else if (opcode
== op_invokeinterface
)
2979 // This is a hack. We might have merged two
2980 // items and gotten `Object'. This can happen
2981 // because we don't keep track of where merges
2982 // come from. This is safe as long as the
2983 // interpreter checks interfaces at runtime.
2984 type
obj (&java::lang::Object::class$
);
2985 ok
= raw
.compatible (obj
, this);
2989 verify_fail ("incompatible type on stack");
2992 current_state
->set_initialized (raw
.get_pc (),
2993 current_method
->max_locals
);
2996 type rt
= compute_return_type (method_signature
);
3004 type t
= check_class_constant (get_ushort ());
3005 if (t
.isarray () || t
.isinterface (this) || t
.isabstract (this))
3006 verify_fail ("type is array, interface, or abstract");
3007 t
.set_uninitialized (start_PC
, this);
3014 int atype
= get_byte ();
3015 // We intentionally have chosen constants to make this
3017 if (atype
< boolean_type
|| atype
> long_type
)
3018 verify_fail ("type not primitive", start_PC
);
3019 pop_type (int_type
);
3020 push_type (construct_primitive_array_type (type_val (atype
)));
3024 pop_type (int_type
);
3025 push_type (check_class_constant (get_ushort ()).to_array (this));
3027 case op_arraylength
:
3029 type t
= pop_init_ref (reference_type
);
3030 if (! t
.isarray () && ! t
.isnull ())
3031 verify_fail ("array type expected");
3032 push_type (int_type
);
3036 pop_type (type (&java::lang::Throwable::class$
));
3040 pop_init_ref (reference_type
);
3041 push_type (check_class_constant (get_ushort ()));
3044 pop_init_ref (reference_type
);
3045 check_class_constant (get_ushort ());
3046 push_type (int_type
);
3048 case op_monitorenter
:
3049 pop_init_ref (reference_type
);
3051 case op_monitorexit
:
3052 pop_init_ref (reference_type
);
3056 switch (get_byte ())
3059 push_type (get_variable (get_ushort (), int_type
));
3062 push_type (get_variable (get_ushort (), long_type
));
3065 push_type (get_variable (get_ushort (), float_type
));
3068 push_type (get_variable (get_ushort (), double_type
));
3071 push_type (get_variable (get_ushort (), reference_type
));
3074 set_variable (get_ushort (), pop_type (int_type
));
3077 set_variable (get_ushort (), pop_type (long_type
));
3080 set_variable (get_ushort (), pop_type (float_type
));
3083 set_variable (get_ushort (), pop_type (double_type
));
3086 set_variable (get_ushort (), pop_init_ref (reference_type
));
3089 handle_ret_insn (get_short ());
3092 get_variable (get_ushort (), int_type
);
3096 verify_fail ("unrecognized wide instruction", start_PC
);
3100 case op_multianewarray
:
3102 type atype
= check_class_constant (get_ushort ());
3103 int dim
= get_byte ();
3105 verify_fail ("too few dimensions to multianewarray", start_PC
);
3106 atype
.verify_dimensions (dim
, this);
3107 for (int i
= 0; i
< dim
; ++i
)
3108 pop_type (int_type
);
3114 pop_type (reference_type
);
3115 push_jump (get_short ());
3118 push_jump (get_int ());
3122 handle_jsr_insn (get_int ());
3125 // These are unused here, but we call them out explicitly
3126 // so that -Wswitch-enum doesn't complain.
3132 case op_putstatic_1
:
3133 case op_putstatic_2
:
3134 case op_putstatic_4
:
3135 case op_putstatic_8
:
3136 case op_putstatic_a
:
3138 case op_getfield_2s
:
3139 case op_getfield_2u
:
3143 case op_getstatic_1
:
3144 case op_getstatic_2s
:
3145 case op_getstatic_2u
:
3146 case op_getstatic_4
:
3147 case op_getstatic_8
:
3148 case op_getstatic_a
:
3150 // Unrecognized opcode.
3151 verify_fail ("unrecognized instruction in verify_instructions_0",
3159 void verify_instructions ()
3162 verify_instructions_0 ();
3165 _Jv_BytecodeVerifier (_Jv_InterpMethod
*m
)
3167 // We just print the text as utf-8. This is just for debugging
3169 debug_print ("--------------------------------\n");
3170 debug_print ("-- Verifying method `%s'\n", m
->self
->name
->data
);
3173 bytecode
= m
->bytecode ();
3174 exception
= m
->exceptions ();
3175 current_class
= m
->defining_class
;
3181 entry_points
= NULL
;
3184 ~_Jv_BytecodeVerifier ()
3193 for (int i
= 0; i
< current_method
->code_length
; ++i
)
3195 if (jsr_ptrs
[i
] != NULL
)
3197 subr_info
*info
= jsr_ptrs
[i
];
3198 while (info
!= NULL
)
3200 subr_info
*next
= info
->next
;
3206 _Jv_Free (jsr_ptrs
);
3209 while (utf8_list
!= NULL
)
3211 linked_utf8
*n
= utf8_list
->next
;
3212 _Jv_Free (utf8_list
->val
);
3213 _Jv_Free (utf8_list
);
3217 while (entry_points
!= NULL
)
3219 subr_entry_info
*next
= entry_points
->next
;
3220 _Jv_Free (entry_points
);
3221 entry_points
= next
;
3227 _Jv_VerifyMethod (_Jv_InterpMethod
*meth
)
3229 _Jv_BytecodeVerifier
v (meth
);
3230 v
.verify_instructions ();
3232 #endif /* INTERPRETER */