1 // verify.cc - verify bytecode
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
;
61 struct ref_intersection
;
65 // The PC corresponding to the start of the current instruction.
68 // The current state of the stack, locals, etc.
71 // We store the state at branch targets, for merging. This holds
75 // We keep a linked list of all the PCs which we must reverify.
76 // The link is done using the PC values. This is the head of the
80 // We keep some flags for each instruction. The values are the
81 // FLAG_* constants defined above.
84 // We need to keep track of which instructions can call a given
85 // subroutine. FIXME: this is inefficient. We keep a linked list
86 // of all calling `jsr's at at each jsr target.
89 // We keep a linked list of entries which map each `ret' instruction
90 // to its unique subroutine entry point. We expect that there won't
91 // be many `ret' instructions, so a linked list is ok.
92 subr_entry_info
*entry_points
;
94 // The bytecode itself.
95 unsigned char *bytecode
;
97 _Jv_InterpException
*exception
;
100 jclass current_class
;
102 _Jv_InterpMethod
*current_method
;
104 // A linked list of utf8 objects we allocate. This is really ugly,
105 // but without this our utf8 objects would be collected.
106 linked_utf8
*utf8_list
;
108 // A linked list of all ref_intersection objects we allocate.
109 ref_intersection
*isect_list
;
117 _Jv_Utf8Const
*make_utf8_const (char *s
, int len
)
119 _Jv_Utf8Const
*val
= _Jv_makeUtf8Const (s
, len
);
120 _Jv_Utf8Const
*r
= (_Jv_Utf8Const
*) _Jv_Malloc (sizeof (_Jv_Utf8Const
)
123 r
->length
= val
->length
;
125 memcpy (r
->data
, val
->data
, val
->length
+ 1);
127 linked_utf8
*lu
= (linked_utf8
*) _Jv_Malloc (sizeof (linked_utf8
));
129 lu
->next
= utf8_list
;
135 __attribute__ ((__noreturn__
)) void verify_fail (char *s
, jint pc
= -1)
137 using namespace java::lang
;
138 StringBuffer
*buf
= new StringBuffer ();
140 buf
->append (JvNewStringLatin1 ("verification failed"));
145 buf
->append (JvNewStringLatin1 (" at PC "));
149 _Jv_InterpMethod
*method
= current_method
;
150 buf
->append (JvNewStringLatin1 (" in "));
151 buf
->append (current_class
->getName());
152 buf
->append ((jchar
) ':');
153 buf
->append (JvNewStringUTF (method
->get_method()->name
->data
));
154 buf
->append ((jchar
) '(');
155 buf
->append (JvNewStringUTF (method
->get_method()->signature
->data
));
156 buf
->append ((jchar
) ')');
158 buf
->append (JvNewStringLatin1 (": "));
159 buf
->append (JvNewStringLatin1 (s
));
160 throw new java::lang::VerifyError (buf
->toString ());
163 // This enum holds a list of tags for all the different types we
164 // need to handle. Reference types are treated specially by the
170 // The values for primitive types are chosen to correspond to values
171 // specified to newarray.
181 // Used when overwriting second word of a double or long in the
182 // local variables. Also used after merging local variable states
183 // to indicate an unusable value.
188 // There is an obscure special case which requires us to note when
189 // a local variable has not been used by a subroutine. See
190 // push_jump_merge for more information.
191 unused_by_subroutine_type
,
193 // Everything after `reference_type' must be a reference type.
196 uninitialized_reference_type
199 // This represents a merged class type. Some verifiers (including
200 // earlier versions of this one) will compute the intersection of
201 // two class types when merging states. However, this loses
202 // critical information about interfaces implemented by the various
203 // classes. So instead we keep track of all the actual classes that
205 struct ref_intersection
207 // Whether or not this type has been resolved.
213 // For a resolved reference type, this is a pointer to the class.
215 // For other reference types, this it the name of the class.
219 // Link to the next reference in the intersection.
220 ref_intersection
*ref_next
;
222 // This is used to keep track of all the allocated
223 // ref_intersection objects, so we can free them.
224 // FIXME: we should allocate these in chunks.
225 ref_intersection
*alloc_next
;
227 ref_intersection (jclass klass
, _Jv_BytecodeVerifier
*verifier
)
232 alloc_next
= verifier
->isect_list
;
233 verifier
->isect_list
= this;
236 ref_intersection (_Jv_Utf8Const
*name
, _Jv_BytecodeVerifier
*verifier
)
241 alloc_next
= verifier
->isect_list
;
242 verifier
->isect_list
= this;
245 ref_intersection (ref_intersection
*dup
, ref_intersection
*tail
,
246 _Jv_BytecodeVerifier
*verifier
)
249 is_resolved
= dup
->is_resolved
;
251 alloc_next
= verifier
->isect_list
;
252 verifier
->isect_list
= this;
255 bool equals (ref_intersection
*other
, _Jv_BytecodeVerifier
*verifier
)
257 if (! is_resolved
&& ! other
->is_resolved
258 && _Jv_equalUtf8Consts (data
.name
, other
->data
.name
))
262 if (! other
->is_resolved
)
263 other
->resolve (verifier
);
264 return data
.klass
== other
->data
.klass
;
267 // Merge THIS type into OTHER, returning the result. This will
268 // return OTHER if all the classes in THIS already appear in
270 ref_intersection
*merge (ref_intersection
*other
,
271 _Jv_BytecodeVerifier
*verifier
)
273 ref_intersection
*tail
= other
;
274 for (ref_intersection
*self
= this; self
!= NULL
; self
= self
->ref_next
)
277 for (ref_intersection
*iter
= other
; iter
!= NULL
;
278 iter
= iter
->ref_next
)
280 if (iter
->equals (self
, verifier
))
288 tail
= new ref_intersection (self
, tail
, verifier
);
293 void resolve (_Jv_BytecodeVerifier
*verifier
)
298 using namespace java::lang
;
299 java::lang::ClassLoader
*loader
300 = verifier
->current_class
->getClassLoaderInternal();
301 // We might see either kind of name. Sigh.
302 if (data
.name
->data
[0] == 'L'
303 && data
.name
->data
[data
.name
->length
- 1] == ';')
304 data
.klass
= _Jv_FindClassFromSignature (data
.name
->data
, loader
);
306 data
.klass
= Class::forName (_Jv_NewStringUtf8Const (data
.name
),
311 // See if an object of type OTHER can be assigned to an object of
312 // type *THIS. This might resolve classes in one chain or the
314 bool compatible (ref_intersection
*other
,
315 _Jv_BytecodeVerifier
*verifier
)
317 ref_intersection
*self
= this;
319 for (; self
!= NULL
; self
= self
->ref_next
)
321 ref_intersection
*other_iter
= other
;
323 for (; other_iter
!= NULL
; other_iter
= other_iter
->ref_next
)
325 // Avoid resolving if possible.
326 if (! self
->is_resolved
327 && ! other_iter
->is_resolved
328 && _Jv_equalUtf8Consts (self
->data
.name
,
329 other_iter
->data
.name
))
332 if (! self
->is_resolved
)
333 self
->resolve(verifier
);
334 if (! other_iter
->is_resolved
)
335 other_iter
->resolve(verifier
);
337 if (! is_assignable_from_slow (self
->data
.klass
,
338 other_iter
->data
.klass
))
348 // assert (ref_next == NULL);
350 return data
.klass
->isArray ();
352 return data
.name
->data
[0] == '[';
355 bool isinterface (_Jv_BytecodeVerifier
*verifier
)
357 // assert (ref_next == NULL);
360 return data
.klass
->isInterface ();
363 bool isabstract (_Jv_BytecodeVerifier
*verifier
)
365 // assert (ref_next == NULL);
368 using namespace java::lang::reflect
;
369 return Modifier::isAbstract (data
.klass
->getModifiers ());
372 jclass
getclass (_Jv_BytecodeVerifier
*verifier
)
379 int count_dimensions ()
384 jclass k
= data
.klass
;
385 while (k
->isArray ())
387 k
= k
->getComponentType ();
393 char *p
= data
.name
->data
;
400 void *operator new (size_t bytes
)
402 return _Jv_Malloc (bytes
);
405 void operator delete (void *mem
)
411 // Return the type_val corresponding to a primitive signature
412 // character. For instance `I' returns `int.class'.
413 type_val
get_type_val_for_signature (jchar sig
)
446 verify_fail ("invalid signature");
451 // Return the type_val corresponding to a primitive class.
452 type_val
get_type_val_for_signature (jclass k
)
454 return get_type_val_for_signature ((jchar
) k
->method_count
);
457 // This is like _Jv_IsAssignableFrom, but it works even if SOURCE or
458 // TARGET haven't been prepared.
459 static bool is_assignable_from_slow (jclass target
, jclass source
)
461 // First, strip arrays.
462 while (target
->isArray ())
464 // If target is array, source must be as well.
465 if (! source
->isArray ())
467 target
= target
->getComponentType ();
468 source
= source
->getComponentType ();
472 if (target
== &java::lang::Object::class$
)
477 if (source
== target
)
480 if (target
->isPrimitive () || source
->isPrimitive ())
483 if (target
->isInterface ())
485 for (int i
= 0; i
< source
->interface_count
; ++i
)
487 // We use a recursive call because we also need to
488 // check superinterfaces.
489 if (is_assignable_from_slow (target
, source
->interfaces
[i
]))
493 source
= source
->getSuperclass ();
495 while (source
!= NULL
);
500 // This is used to keep track of which `jsr's correspond to a given
504 // PC of the instruction just after the jsr.
510 // This is used to keep track of which subroutine entry point
511 // corresponds to which `ret' instruction.
512 struct subr_entry_info
514 // PC of the subroutine entry point.
516 // PC of the `ret' instruction.
519 subr_entry_info
*next
;
522 // The `type' class is used to represent a single type in the
529 // For reference types, the representation of the type.
530 ref_intersection
*klass
;
532 // This is used when constructing a new object. It is the PC of the
533 // `new' instruction which created the object. We use the special
534 // value -2 to mean that this is uninitialized, and the special
535 // value -1 for the case where the current method is itself the
539 static const int UNINIT
= -2;
540 static const int SELF
= -1;
542 // Basic constructor.
545 key
= unsuitable_type
;
550 // Make a new instance given the type tag. We assume a generic
551 // `reference_type' means Object.
555 // For reference_type, if KLASS==NULL then that means we are
556 // looking for a generic object of any kind, including an
557 // uninitialized reference.
562 // Make a new instance given a class.
563 type (jclass k
, _Jv_BytecodeVerifier
*verifier
)
565 key
= reference_type
;
566 klass
= new ref_intersection (k
, verifier
);
570 // Make a new instance given the name of a class.
571 type (_Jv_Utf8Const
*n
, _Jv_BytecodeVerifier
*verifier
)
573 key
= reference_type
;
574 klass
= new ref_intersection (n
, verifier
);
586 // These operators are required because libgcj can't link in
588 void *operator new[] (size_t bytes
)
590 return _Jv_Malloc (bytes
);
593 void operator delete[] (void *mem
)
598 type
& operator= (type_val k
)
606 type
& operator= (const type
& t
)
614 // Promote a numeric type.
617 if (key
== boolean_type
|| key
== char_type
618 || key
== byte_type
|| key
== short_type
)
623 // Mark this type as the uninitialized result of `new'.
624 void set_uninitialized (int npc
, _Jv_BytecodeVerifier
*verifier
)
626 if (key
== reference_type
)
627 key
= uninitialized_reference_type
;
629 verifier
->verify_fail ("internal error in type::uninitialized");
633 // Mark this type as now initialized.
634 void set_initialized (int npc
)
636 if (npc
!= UNINIT
&& pc
== npc
&& key
== uninitialized_reference_type
)
638 key
= reference_type
;
644 // Return true if an object of type K can be assigned to a variable
645 // of type *THIS. Handle various special cases too. Might modify
646 // *THIS or K. Note however that this does not perform numeric
648 bool compatible (type
&k
, _Jv_BytecodeVerifier
*verifier
)
650 // Any type is compatible with the unsuitable type.
651 if (key
== unsuitable_type
)
654 if (key
< reference_type
|| k
.key
< reference_type
)
657 // The `null' type is convertible to any initialized reference
659 if (key
== null_type
)
660 return k
.key
!= uninitialized_reference_type
;
661 if (k
.key
== null_type
)
662 return key
!= uninitialized_reference_type
;
664 // A special case for a generic reference.
668 verifier
->verify_fail ("programmer error in type::compatible");
670 // An initialized type and an uninitialized type are not
672 if (isinitialized () != k
.isinitialized ())
675 // Two uninitialized objects are compatible if either:
676 // * The PCs are identical, or
677 // * One PC is UNINIT.
678 if (! isinitialized ())
680 if (pc
!= k
.pc
&& pc
!= UNINIT
&& k
.pc
!= UNINIT
)
684 return klass
->compatible(k
.klass
, verifier
);
689 return key
== void_type
;
694 return key
== long_type
|| key
== double_type
;
697 // Return number of stack or local variable slots taken by this
701 return iswide () ? 2 : 1;
704 bool isarray () const
706 // We treat null_type as not an array. This is ok based on the
707 // current uses of this method.
708 if (key
== reference_type
)
709 return klass
->isarray ();
715 return key
== null_type
;
718 bool isinterface (_Jv_BytecodeVerifier
*verifier
)
720 if (key
!= reference_type
)
722 return klass
->isinterface (verifier
);
725 bool isabstract (_Jv_BytecodeVerifier
*verifier
)
727 if (key
!= reference_type
)
729 return klass
->isabstract (verifier
);
732 // Return the element type of an array.
733 type
element_type (_Jv_BytecodeVerifier
*verifier
)
735 if (key
!= reference_type
)
736 verifier
->verify_fail ("programmer error in type::element_type()", -1);
738 jclass k
= klass
->getclass (verifier
)->getComponentType ();
739 if (k
->isPrimitive ())
740 return type (verifier
->get_type_val_for_signature (k
));
741 return type (k
, verifier
);
744 // Return the array type corresponding to an initialized
745 // reference. We could expand this to work for other kinds of
746 // types, but currently we don't need to.
747 type
to_array (_Jv_BytecodeVerifier
*verifier
)
749 if (key
!= reference_type
)
750 verifier
->verify_fail ("internal error in type::to_array()");
752 jclass k
= klass
->getclass (verifier
);
753 return type (_Jv_GetArrayClass (k
, k
->getClassLoaderInternal()),
757 bool isreference () const
759 return key
>= reference_type
;
767 bool isinitialized () const
769 return key
== reference_type
|| key
== null_type
;
772 bool isresolved () const
774 return (key
== reference_type
776 || key
== uninitialized_reference_type
);
779 void verify_dimensions (int ndims
, _Jv_BytecodeVerifier
*verifier
)
781 // The way this is written, we don't need to check isarray().
782 if (key
!= reference_type
)
783 verifier
->verify_fail ("internal error in verify_dimensions: not a reference type");
785 if (klass
->count_dimensions () < ndims
)
786 verifier
->verify_fail ("array type has fewer dimensions than required");
789 // Merge OLD_TYPE into this. On error throw exception.
790 bool merge (type
& old_type
, bool local_semantics
,
791 _Jv_BytecodeVerifier
*verifier
)
793 bool changed
= false;
794 bool refo
= old_type
.isreference ();
795 bool refn
= isreference ();
798 if (old_type
.key
== null_type
)
800 else if (key
== null_type
)
805 else if (isinitialized () != old_type
.isinitialized ())
806 verifier
->verify_fail ("merging initialized and uninitialized types");
809 if (! isinitialized ())
813 else if (old_type
.pc
== UNINIT
)
815 else if (pc
!= old_type
.pc
)
816 verifier
->verify_fail ("merging different uninitialized types");
819 ref_intersection
*merged
= old_type
.klass
->merge (klass
,
828 else if (refo
|| refn
|| key
!= old_type
.key
)
832 // If we're merging into an "unused" slot, then we
833 // simply accept whatever we're merging from.
834 if (key
== unused_by_subroutine_type
)
839 else if (old_type
.key
== unused_by_subroutine_type
)
843 // If we already have an `unsuitable' type, then we
844 // don't need to change again.
845 else if (key
!= unsuitable_type
)
847 key
= unsuitable_type
;
852 verifier
->verify_fail ("unmergeable type");
858 void print (void) const
863 case boolean_type
: c
= 'Z'; break;
864 case byte_type
: c
= 'B'; break;
865 case char_type
: c
= 'C'; break;
866 case short_type
: c
= 'S'; break;
867 case int_type
: c
= 'I'; break;
868 case long_type
: c
= 'J'; break;
869 case float_type
: c
= 'F'; break;
870 case double_type
: c
= 'D'; break;
871 case void_type
: c
= 'V'; break;
872 case unsuitable_type
: c
= '-'; break;
873 case return_address_type
: c
= 'r'; break;
874 case continuation_type
: c
= '+'; break;
875 case unused_by_subroutine_type
: c
= '_'; break;
876 case reference_type
: c
= 'L'; break;
877 case null_type
: c
= '@'; break;
878 case uninitialized_reference_type
: c
= 'U'; break;
880 debug_print ("%c", c
);
882 #endif /* VERIFY_DEBUG */
885 // This class holds all the state information we need for a given
889 // The current top of the stack, in terms of slots.
891 // The current depth of the stack. This will be larger than
892 // STACKTOP when wide types are on the stack.
896 // The local variables.
898 // This is used in subroutines to keep track of which local
899 // variables have been accessed.
901 // If not 0, then we are in a subroutine. The value is the PC of
902 // the subroutine's entry point. We can use 0 as an exceptional
903 // value because PC=0 can never be a subroutine.
905 // This is used to keep a linked list of all the states which
906 // require re-verification. We use the PC to keep track.
908 // We keep track of the type of `this' specially. This is used to
909 // ensure that an instance initializer invokes another initializer
910 // on `this' before returning. We must keep track of this
911 // specially because otherwise we might be confused by code which
912 // assigns to locals[0] (overwriting `this') and then returns
913 // without really initializing.
915 // This is a list of all subroutines that have been seen at this
916 // point. Ordinarily this is NULL; it is only allocated and used
917 // in relatively weird situations involving non-ret exit from a
918 // subroutine. We have to keep track of this in this way to avoid
919 // endless recursion in these cases.
920 subr_info
*seen_subrs
;
922 // INVALID marks a state which is not on the linked list of states
923 // requiring reverification.
924 static const int INVALID
= -1;
925 // NO_NEXT marks the state at the end of the reverification list.
926 static const int NO_NEXT
= -2;
928 // This is used to mark the stack depth at the instruction just
929 // after a `jsr' when we haven't yet processed the corresponding
930 // `ret'. See handle_jsr_insn for more information.
931 static const int NO_STACK
= -1;
938 local_changed
= NULL
;
942 state (int max_stack
, int max_locals
)
947 stack
= new type
[max_stack
];
948 for (int i
= 0; i
< max_stack
; ++i
)
949 stack
[i
] = unsuitable_type
;
950 locals
= new type
[max_locals
];
951 local_changed
= (bool *) _Jv_Malloc (sizeof (bool) * max_locals
);
953 for (int i
= 0; i
< max_locals
; ++i
)
955 locals
[i
] = unsuitable_type
;
956 local_changed
[i
] = false;
962 state (const state
*orig
, int max_stack
, int max_locals
,
963 bool ret_semantics
= false)
965 stack
= new type
[max_stack
];
966 locals
= new type
[max_locals
];
967 local_changed
= (bool *) _Jv_Malloc (sizeof (bool) * max_locals
);
969 copy (orig
, max_stack
, max_locals
, ret_semantics
);
980 _Jv_Free (local_changed
);
984 void *operator new[] (size_t bytes
)
986 return _Jv_Malloc (bytes
);
989 void operator delete[] (void *mem
)
994 void *operator new (size_t bytes
)
996 return _Jv_Malloc (bytes
);
999 void operator delete (void *mem
)
1006 subr_info
*info
= seen_subrs
;
1007 while (info
!= NULL
)
1009 subr_info
*next
= info
->next
;
1015 void copy (const state
*copy
, int max_stack
, int max_locals
,
1016 bool ret_semantics
= false)
1018 stacktop
= copy
->stacktop
;
1019 stackdepth
= copy
->stackdepth
;
1020 subroutine
= copy
->subroutine
;
1021 for (int i
= 0; i
< max_stack
; ++i
)
1022 stack
[i
] = copy
->stack
[i
];
1023 for (int i
= 0; i
< max_locals
; ++i
)
1025 // See push_jump_merge to understand this case.
1027 locals
[i
] = type (copy
->local_changed
[i
]
1029 : unused_by_subroutine_type
);
1031 locals
[i
] = copy
->locals
[i
];
1032 local_changed
[i
] = subroutine
? copy
->local_changed
[i
] : false;
1036 if (copy
->seen_subrs
)
1038 for (subr_info
*info
= seen_subrs
; info
!= NULL
; info
= info
->next
)
1039 add_subr (info
->pc
);
1044 this_type
= copy
->this_type
;
1045 // Don't modify `next'.
1048 // Modify this state to reflect entry to an exception handler.
1049 void set_exception (type t
, int max_stack
)
1054 for (int i
= stacktop
; i
< max_stack
; ++i
)
1055 stack
[i
] = unsuitable_type
;
1058 // Modify this state to reflect entry into a subroutine.
1059 void enter_subroutine (int npc
, int max_locals
)
1062 // Mark all items as unchanged. Each subroutine needs to keep
1063 // track of its `changed' state independently. In the case of
1064 // nested subroutines, this information will be merged back into
1065 // parent by the `ret'.
1066 for (int i
= 0; i
< max_locals
; ++i
)
1067 local_changed
[i
] = false;
1070 // Indicate that we've been in this this subroutine.
1071 void add_subr (int pc
)
1073 subr_info
*n
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
1075 n
->next
= seen_subrs
;
1079 // Merge STATE_OLD into this state. Destructively modifies this
1080 // state. Returns true if the new state was in fact changed.
1081 // Will throw an exception if the states are not mergeable.
1082 bool merge (state
*state_old
, bool ret_semantics
,
1083 int max_locals
, _Jv_BytecodeVerifier
*verifier
)
1085 bool changed
= false;
1087 // Special handling for `this'. If one or the other is
1088 // uninitialized, then the merge is uninitialized.
1089 if (this_type
.isinitialized ())
1090 this_type
= state_old
->this_type
;
1092 // Merge subroutine states. Here we just keep track of what
1093 // subroutine we think we're in. We only check for a merge
1094 // (which is invalid) when we see a `ret'.
1095 if (subroutine
== state_old
->subroutine
)
1099 else if (subroutine
== 0)
1101 subroutine
= state_old
->subroutine
;
1106 // If the subroutines differ, and we haven't seen this
1107 // subroutine before, indicate that the state changed. This
1108 // is needed to detect when subroutines have merged.
1110 for (subr_info
*info
= seen_subrs
; info
!= NULL
; info
= info
->next
)
1112 if (info
->pc
== state_old
->subroutine
)
1120 add_subr (state_old
->subroutine
);
1125 // Merge stacks. Special handling for NO_STACK case.
1126 if (state_old
->stacktop
== NO_STACK
)
1128 // Nothing to do in this case; we don't care about modifying
1131 else if (stacktop
== NO_STACK
)
1133 stacktop
= state_old
->stacktop
;
1134 stackdepth
= state_old
->stackdepth
;
1135 for (int i
= 0; i
< stacktop
; ++i
)
1136 stack
[i
] = state_old
->stack
[i
];
1139 else if (state_old
->stacktop
!= stacktop
)
1140 verifier
->verify_fail ("stack sizes differ");
1143 for (int i
= 0; i
< state_old
->stacktop
; ++i
)
1145 if (stack
[i
].merge (state_old
->stack
[i
], false, verifier
))
1150 // Merge local variables.
1151 for (int i
= 0; i
< max_locals
; ++i
)
1153 // If we're not processing a `ret', then we merge every
1154 // local variable. If we are processing a `ret', then we
1155 // only merge locals which changed in the subroutine. When
1156 // processing a `ret', STATE_OLD is the state at the point
1157 // of the `ret', and THIS is the state just after the `jsr'.
1158 if (! ret_semantics
|| state_old
->local_changed
[i
])
1160 if (locals
[i
].merge (state_old
->locals
[i
], true, verifier
))
1162 // Note that we don't call `note_variable' here.
1163 // This change doesn't represent a real change to a
1164 // local, but rather a merge artifact. If we're in
1165 // a subroutine which is called with two
1166 // incompatible types in a slot that is unused by
1167 // the subroutine, then we don't want to mark that
1168 // variable as having been modified.
1173 // If we're in a subroutine, we must compute the union of
1174 // all the changed local variables.
1175 if (state_old
->local_changed
[i
])
1182 // Throw an exception if there is an uninitialized object on the
1183 // stack or in a local variable. EXCEPTION_SEMANTICS controls
1184 // whether we're using backwards-branch or exception-handing
1186 void check_no_uninitialized_objects (int max_locals
,
1187 _Jv_BytecodeVerifier
*verifier
,
1188 bool exception_semantics
= false)
1190 if (! exception_semantics
)
1192 for (int i
= 0; i
< stacktop
; ++i
)
1193 if (stack
[i
].isreference () && ! stack
[i
].isinitialized ())
1194 verifier
->verify_fail ("uninitialized object on stack");
1197 for (int i
= 0; i
< max_locals
; ++i
)
1198 if (locals
[i
].isreference () && ! locals
[i
].isinitialized ())
1199 verifier
->verify_fail ("uninitialized object in local variable");
1201 check_this_initialized (verifier
);
1204 // Ensure that `this' has been initialized.
1205 void check_this_initialized (_Jv_BytecodeVerifier
*verifier
)
1207 if (this_type
.isreference () && ! this_type
.isinitialized ())
1208 verifier
->verify_fail ("`this' is uninitialized");
1211 // Set type of `this'.
1212 void set_this_type (const type
&k
)
1217 // Note that a local variable was modified.
1218 void note_variable (int index
)
1221 local_changed
[index
] = true;
1224 // Mark each `new'd object we know of that was allocated at PC as
1226 void set_initialized (int pc
, int max_locals
)
1228 for (int i
= 0; i
< stacktop
; ++i
)
1229 stack
[i
].set_initialized (pc
);
1230 for (int i
= 0; i
< max_locals
; ++i
)
1231 locals
[i
].set_initialized (pc
);
1232 this_type
.set_initialized (pc
);
1235 // Return true if this state is the unmerged result of a `ret'.
1236 bool is_unmerged_ret_state (int max_locals
) const
1238 if (stacktop
== NO_STACK
)
1240 for (int i
= 0; i
< max_locals
; ++i
)
1242 if (locals
[i
].key
== unused_by_subroutine_type
)
1249 void print (const char *leader
, int pc
,
1250 int max_stack
, int max_locals
) const
1252 debug_print ("%s [%4d]: [stack] ", leader
, pc
);
1254 for (i
= 0; i
< stacktop
; ++i
)
1256 for (; i
< max_stack
; ++i
)
1258 debug_print (" [local] ");
1259 for (i
= 0; i
< max_locals
; ++i
)
1262 debug_print (local_changed
[i
] ? "+" : " ");
1264 if (subroutine
== 0)
1265 debug_print (" | None");
1267 debug_print (" | %4d", subroutine
);
1268 debug_print (" | %p\n", this);
1271 inline void print (const char *, int, int, int) const
1274 #endif /* VERIFY_DEBUG */
1279 if (current_state
->stacktop
<= 0)
1280 verify_fail ("stack empty");
1281 type r
= current_state
->stack
[--current_state
->stacktop
];
1282 current_state
->stackdepth
-= r
.depth ();
1283 if (current_state
->stackdepth
< 0)
1284 verify_fail ("stack empty", start_PC
);
1290 type r
= pop_raw ();
1292 verify_fail ("narrow pop of wide type");
1296 type
pop_type (type match
)
1299 type t
= pop_raw ();
1300 if (! match
.compatible (t
, this))
1301 verify_fail ("incompatible type on stack");
1305 // Pop a reference which is guaranteed to be initialized. MATCH
1306 // doesn't have to be a reference type; in this case this acts like
1308 type
pop_init_ref (type match
)
1310 type t
= pop_raw ();
1311 if (t
.isreference () && ! t
.isinitialized ())
1312 verify_fail ("initialized reference required");
1313 else if (! match
.compatible (t
, this))
1314 verify_fail ("incompatible type on stack");
1318 // Pop a reference type or a return address.
1319 type
pop_ref_or_return ()
1321 type t
= pop_raw ();
1322 if (! t
.isreference () && t
.key
!= return_address_type
)
1323 verify_fail ("expected reference or return address on stack");
1327 void push_type (type t
)
1329 // If T is a numeric type like short, promote it to int.
1332 int depth
= t
.depth ();
1333 if (current_state
->stackdepth
+ depth
> current_method
->max_stack
)
1334 verify_fail ("stack overflow");
1335 current_state
->stack
[current_state
->stacktop
++] = t
;
1336 current_state
->stackdepth
+= depth
;
1339 void set_variable (int index
, type t
)
1341 // If T is a numeric type like short, promote it to int.
1344 int depth
= t
.depth ();
1345 if (index
> current_method
->max_locals
- depth
)
1346 verify_fail ("invalid local variable");
1347 current_state
->locals
[index
] = t
;
1348 current_state
->note_variable (index
);
1352 current_state
->locals
[index
+ 1] = continuation_type
;
1353 current_state
->note_variable (index
+ 1);
1355 if (index
> 0 && current_state
->locals
[index
- 1].iswide ())
1357 current_state
->locals
[index
- 1] = unsuitable_type
;
1358 // There's no need to call note_variable here.
1362 type
get_variable (int index
, type t
)
1364 int depth
= t
.depth ();
1365 if (index
> current_method
->max_locals
- depth
)
1366 verify_fail ("invalid local variable");
1367 if (! t
.compatible (current_state
->locals
[index
], this))
1368 verify_fail ("incompatible type in local variable");
1371 type
t (continuation_type
);
1372 if (! current_state
->locals
[index
+ 1].compatible (t
, this))
1373 verify_fail ("invalid local variable");
1375 return current_state
->locals
[index
];
1378 // Make sure ARRAY is an array type and that its elements are
1379 // compatible with type ELEMENT. Returns the actual element type.
1380 type
require_array_type (type array
, type element
)
1382 // An odd case. Here we just pretend that everything went ok. If
1383 // the requested element type is some kind of reference, return
1384 // the null type instead.
1385 if (array
.isnull ())
1386 return element
.isreference () ? type (null_type
) : element
;
1388 if (! array
.isarray ())
1389 verify_fail ("array required");
1391 type t
= array
.element_type (this);
1392 if (! element
.compatible (t
, this))
1394 // Special case for byte arrays, which must also be boolean
1397 if (element
.key
== byte_type
)
1399 type
e2 (boolean_type
);
1400 ok
= e2
.compatible (t
, this);
1403 verify_fail ("incompatible array element type");
1406 // Return T and not ELEMENT, because T might be specialized.
1412 if (PC
>= current_method
->code_length
)
1413 verify_fail ("premature end of bytecode");
1414 return (jint
) bytecode
[PC
++] & 0xff;
1419 jint b1
= get_byte ();
1420 jint b2
= get_byte ();
1421 return (jint
) ((b1
<< 8) | b2
) & 0xffff;
1426 jint b1
= get_byte ();
1427 jint b2
= get_byte ();
1428 jshort s
= (b1
<< 8) | b2
;
1434 jint b1
= get_byte ();
1435 jint b2
= get_byte ();
1436 jint b3
= get_byte ();
1437 jint b4
= get_byte ();
1438 return (b1
<< 24) | (b2
<< 16) | (b3
<< 8) | b4
;
1441 int compute_jump (int offset
)
1443 int npc
= start_PC
+ offset
;
1444 if (npc
< 0 || npc
>= current_method
->code_length
)
1445 verify_fail ("branch out of range", start_PC
);
1449 // Merge the indicated state into the state at the branch target and
1450 // schedule a new PC if there is a change. If RET_SEMANTICS is
1451 // true, then we are merging from a `ret' instruction into the
1452 // instruction after a `jsr'. This is a special case with its own
1453 // modified semantics.
1454 void push_jump_merge (int npc
, state
*nstate
, bool ret_semantics
= false)
1456 bool changed
= true;
1457 if (states
[npc
] == NULL
)
1459 // There's a weird situation here. If are examining the
1460 // branch that results from a `ret', and there is not yet a
1461 // state available at the branch target (the instruction just
1462 // after the `jsr'), then we have to construct a special kind
1463 // of state at that point for future merging. This special
1464 // state has the type `unused_by_subroutine_type' in each slot
1465 // which was not modified by the subroutine.
1466 states
[npc
] = new state (nstate
, current_method
->max_stack
,
1467 current_method
->max_locals
, ret_semantics
);
1468 debug_print ("== New state in push_jump_merge (ret_semantics = %s)\n",
1469 ret_semantics
? "true" : "false");
1470 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1471 current_method
->max_locals
);
1475 debug_print ("== Merge states in push_jump_merge\n");
1476 nstate
->print ("Frm", start_PC
, current_method
->max_stack
,
1477 current_method
->max_locals
);
1478 states
[npc
]->print (" To", npc
, current_method
->max_stack
,
1479 current_method
->max_locals
);
1480 changed
= states
[npc
]->merge (nstate
, ret_semantics
,
1481 current_method
->max_locals
, this);
1482 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1483 current_method
->max_locals
);
1486 if (changed
&& states
[npc
]->next
== state::INVALID
)
1488 // The merge changed the state, and the new PC isn't yet on our
1489 // list of PCs to re-verify.
1490 states
[npc
]->next
= next_verify_pc
;
1491 next_verify_pc
= npc
;
1495 void push_jump (int offset
)
1497 int npc
= compute_jump (offset
);
1499 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1500 push_jump_merge (npc
, current_state
);
1503 void push_exception_jump (type t
, int pc
)
1505 current_state
->check_no_uninitialized_objects (current_method
->max_locals
,
1507 state
s (current_state
, current_method
->max_stack
,
1508 current_method
->max_locals
);
1509 if (current_method
->max_stack
< 1)
1510 verify_fail ("stack overflow at exception handler");
1511 s
.set_exception (t
, current_method
->max_stack
);
1512 push_jump_merge (pc
, &s
);
1517 int *prev_loc
= &next_verify_pc
;
1518 int npc
= next_verify_pc
;
1520 while (npc
!= state::NO_NEXT
)
1522 // If the next available PC is an unmerged `ret' state, then
1523 // we aren't yet ready to handle it. That's because we would
1524 // need all kind of special cases to do so. So instead we
1525 // defer this jump until after we've processed it via a
1526 // fall-through. This has to happen because the instruction
1527 // before this one must be a `jsr'.
1528 if (! states
[npc
]->is_unmerged_ret_state (current_method
->max_locals
))
1530 *prev_loc
= states
[npc
]->next
;
1531 states
[npc
]->next
= state::INVALID
;
1535 prev_loc
= &states
[npc
]->next
;
1536 npc
= states
[npc
]->next
;
1539 // Note that we might have gotten here even when there are
1540 // remaining states to process. That can happen if we find a
1541 // `jsr' without a `ret'.
1542 return state::NO_NEXT
;
1545 void invalidate_pc ()
1547 PC
= state::NO_NEXT
;
1550 void note_branch_target (int pc
, bool is_jsr_target
= false)
1552 // Don't check `pc <= PC', because we've advanced PC after
1553 // fetching the target and we haven't yet checked the next
1555 if (pc
< PC
&& ! (flags
[pc
] & FLAG_INSN_START
))
1556 verify_fail ("branch not to instruction start", start_PC
);
1557 flags
[pc
] |= FLAG_BRANCH_TARGET
;
1560 // Record the jsr which called this instruction.
1561 subr_info
*info
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
1563 info
->next
= jsr_ptrs
[pc
];
1564 jsr_ptrs
[pc
] = info
;
1568 void skip_padding ()
1570 while ((PC
% 4) > 0)
1571 if (get_byte () != 0)
1572 verify_fail ("found nonzero padding byte");
1575 // Return the subroutine to which the instruction at PC belongs.
1576 int get_subroutine (int pc
)
1578 if (states
[pc
] == NULL
)
1580 return states
[pc
]->subroutine
;
1583 // Do the work for a `ret' instruction. INDEX is the index into the
1585 void handle_ret_insn (int index
)
1587 get_variable (index
, return_address_type
);
1589 int csub
= current_state
->subroutine
;
1591 verify_fail ("no subroutine");
1593 // Check to see if we've merged subroutines.
1594 subr_entry_info
*entry
;
1595 for (entry
= entry_points
; entry
!= NULL
; entry
= entry
->next
)
1597 if (entry
->ret_pc
== start_PC
)
1602 entry
= (subr_entry_info
*) _Jv_Malloc (sizeof (subr_entry_info
));
1604 entry
->ret_pc
= start_PC
;
1605 entry
->next
= entry_points
;
1606 entry_points
= entry
;
1608 else if (entry
->pc
!= csub
)
1609 verify_fail ("subroutines merged");
1611 for (subr_info
*subr
= jsr_ptrs
[csub
]; subr
!= NULL
; subr
= subr
->next
)
1613 // We might be returning to a `jsr' that is at the end of the
1614 // bytecode. This is ok if we never return from the called
1615 // subroutine, but if we see this here it is an error.
1616 if (subr
->pc
>= current_method
->code_length
)
1617 verify_fail ("fell off end");
1619 // Temporarily modify the current state so it looks like we're
1620 // in the enclosing context.
1621 current_state
->subroutine
= get_subroutine (subr
->pc
);
1623 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1624 push_jump_merge (subr
->pc
, current_state
, true);
1627 current_state
->subroutine
= csub
;
1631 // We're in the subroutine SUB, calling a subroutine at DEST. Make
1632 // sure this subroutine isn't already on the stack.
1633 void check_nonrecursive_call (int sub
, int dest
)
1638 verify_fail ("recursive subroutine call");
1639 for (subr_info
*info
= jsr_ptrs
[sub
]; info
!= NULL
; info
= info
->next
)
1640 check_nonrecursive_call (get_subroutine (info
->pc
), dest
);
1643 void handle_jsr_insn (int offset
)
1645 int npc
= compute_jump (offset
);
1648 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1649 check_nonrecursive_call (current_state
->subroutine
, npc
);
1651 // Modify our state as appropriate for entry into a subroutine.
1652 push_type (return_address_type
);
1653 push_jump_merge (npc
, current_state
);
1655 pop_type (return_address_type
);
1657 // On entry to the subroutine, the subroutine number must be set
1658 // and the locals must be marked as cleared. We do this after
1659 // merging state so that we don't erroneously "notice" a variable
1660 // change merely on entry.
1661 states
[npc
]->enter_subroutine (npc
, current_method
->max_locals
);
1663 // Indicate that we don't know the stack depth of the instruction
1664 // following the `jsr'. The idea here is that we need to merge
1665 // the local variable state across the jsr, but the subroutine
1666 // might change the stack depth, so we can't make any assumptions
1667 // about it. So we have yet another special case. We know that
1668 // at this point PC points to the instruction after the jsr. Note
1669 // that it is ok to have a `jsr' at the end of the bytecode,
1670 // provided that the called subroutine never returns. So, we have
1671 // a special case here and another one when we handle the ret.
1672 if (PC
< current_method
->code_length
)
1674 current_state
->stacktop
= state::NO_STACK
;
1675 push_jump_merge (PC
, current_state
);
1680 jclass
construct_primitive_array_type (type_val prim
)
1686 k
= JvPrimClass (boolean
);
1689 k
= JvPrimClass (char);
1692 k
= JvPrimClass (float);
1695 k
= JvPrimClass (double);
1698 k
= JvPrimClass (byte
);
1701 k
= JvPrimClass (short);
1704 k
= JvPrimClass (int);
1707 k
= JvPrimClass (long);
1710 // These aren't used here but we call them out to avoid
1713 case unsuitable_type
:
1714 case return_address_type
:
1715 case continuation_type
:
1716 case unused_by_subroutine_type
:
1717 case reference_type
:
1719 case uninitialized_reference_type
:
1721 verify_fail ("unknown type in construct_primitive_array_type");
1723 k
= _Jv_GetArrayClass (k
, NULL
);
1727 // This pass computes the location of branch targets and also
1728 // instruction starts.
1729 void branch_prepass ()
1731 flags
= (char *) _Jv_Malloc (current_method
->code_length
);
1732 jsr_ptrs
= (subr_info
**) _Jv_Malloc (sizeof (subr_info
*)
1733 * current_method
->code_length
);
1735 for (int i
= 0; i
< current_method
->code_length
; ++i
)
1741 bool last_was_jsr
= false;
1744 while (PC
< current_method
->code_length
)
1746 // Set `start_PC' early so that error checking can have the
1749 flags
[PC
] |= FLAG_INSN_START
;
1751 // If the previous instruction was a jsr, then the next
1752 // instruction is a branch target -- the branch being the
1753 // corresponding `ret'.
1755 note_branch_target (PC
);
1756 last_was_jsr
= false;
1758 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
1762 case op_aconst_null
:
1898 case op_monitorenter
:
1899 case op_monitorexit
:
1907 case op_arraylength
:
1939 case op_invokespecial
:
1940 case op_invokestatic
:
1941 case op_invokevirtual
:
1945 case op_multianewarray
:
1951 last_was_jsr
= true;
1970 note_branch_target (compute_jump (get_short ()), last_was_jsr
);
1973 case op_tableswitch
:
1976 note_branch_target (compute_jump (get_int ()));
1977 jint low
= get_int ();
1978 jint hi
= get_int ();
1980 verify_fail ("invalid tableswitch", start_PC
);
1981 for (int i
= low
; i
<= hi
; ++i
)
1982 note_branch_target (compute_jump (get_int ()));
1986 case op_lookupswitch
:
1989 note_branch_target (compute_jump (get_int ()));
1990 int npairs
= get_int ();
1992 verify_fail ("too few pairs in lookupswitch", start_PC
);
1993 while (npairs
-- > 0)
1996 note_branch_target (compute_jump (get_int ()));
2001 case op_invokeinterface
:
2009 opcode
= (java_opcode
) get_byte ();
2011 if (opcode
== op_iinc
)
2017 last_was_jsr
= true;
2020 note_branch_target (compute_jump (get_int ()), last_was_jsr
);
2023 // These are unused here, but we call them out explicitly
2024 // so that -Wswitch-enum doesn't complain.
2030 case op_putstatic_1
:
2031 case op_putstatic_2
:
2032 case op_putstatic_4
:
2033 case op_putstatic_8
:
2034 case op_putstatic_a
:
2036 case op_getfield_2s
:
2037 case op_getfield_2u
:
2041 case op_getstatic_1
:
2042 case op_getstatic_2s
:
2043 case op_getstatic_2u
:
2044 case op_getstatic_4
:
2045 case op_getstatic_8
:
2046 case op_getstatic_a
:
2048 verify_fail ("unrecognized instruction in branch_prepass",
2052 // See if any previous branch tried to branch to the middle of
2053 // this instruction.
2054 for (int pc
= start_PC
+ 1; pc
< PC
; ++pc
)
2056 if ((flags
[pc
] & FLAG_BRANCH_TARGET
))
2057 verify_fail ("branch to middle of instruction", pc
);
2061 // Verify exception handlers.
2062 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
2064 if (! (flags
[exception
[i
].handler_pc
.i
] & FLAG_INSN_START
))
2065 verify_fail ("exception handler not at instruction start",
2066 exception
[i
].handler_pc
.i
);
2067 if (! (flags
[exception
[i
].start_pc
.i
] & FLAG_INSN_START
))
2068 verify_fail ("exception start not at instruction start",
2069 exception
[i
].start_pc
.i
);
2070 if (exception
[i
].end_pc
.i
!= current_method
->code_length
2071 && ! (flags
[exception
[i
].end_pc
.i
] & FLAG_INSN_START
))
2072 verify_fail ("exception end not at instruction start",
2073 exception
[i
].end_pc
.i
);
2075 flags
[exception
[i
].handler_pc
.i
] |= FLAG_BRANCH_TARGET
;
2079 void check_pool_index (int index
)
2081 if (index
< 0 || index
>= current_class
->constants
.size
)
2082 verify_fail ("constant pool index out of range", start_PC
);
2085 type
check_class_constant (int index
)
2087 check_pool_index (index
);
2088 _Jv_Constants
*pool
= ¤t_class
->constants
;
2089 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedClass
)
2090 return type (pool
->data
[index
].clazz
, this);
2091 else if (pool
->tags
[index
] == JV_CONSTANT_Class
)
2092 return type (pool
->data
[index
].utf8
, this);
2093 verify_fail ("expected class constant", start_PC
);
2096 type
check_constant (int index
)
2098 check_pool_index (index
);
2099 _Jv_Constants
*pool
= ¤t_class
->constants
;
2100 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedString
2101 || pool
->tags
[index
] == JV_CONSTANT_String
)
2102 return type (&java::lang::String::class$
, this);
2103 else if (pool
->tags
[index
] == JV_CONSTANT_Integer
)
2104 return type (int_type
);
2105 else if (pool
->tags
[index
] == JV_CONSTANT_Float
)
2106 return type (float_type
);
2107 verify_fail ("String, int, or float constant expected", start_PC
);
2110 type
check_wide_constant (int index
)
2112 check_pool_index (index
);
2113 _Jv_Constants
*pool
= ¤t_class
->constants
;
2114 if (pool
->tags
[index
] == JV_CONSTANT_Long
)
2115 return type (long_type
);
2116 else if (pool
->tags
[index
] == JV_CONSTANT_Double
)
2117 return type (double_type
);
2118 verify_fail ("long or double constant expected", start_PC
);
2121 // Helper for both field and method. These are laid out the same in
2122 // the constant pool.
2123 type
handle_field_or_method (int index
, int expected
,
2124 _Jv_Utf8Const
**name
,
2125 _Jv_Utf8Const
**fmtype
)
2127 check_pool_index (index
);
2128 _Jv_Constants
*pool
= ¤t_class
->constants
;
2129 if (pool
->tags
[index
] != expected
)
2130 verify_fail ("didn't see expected constant", start_PC
);
2131 // Once we know we have a Fieldref or Methodref we assume that it
2132 // is correctly laid out in the constant pool. I think the code
2133 // in defineclass.cc guarantees this.
2134 _Jv_ushort class_index
, name_and_type_index
;
2135 _Jv_loadIndexes (&pool
->data
[index
],
2137 name_and_type_index
);
2138 _Jv_ushort name_index
, desc_index
;
2139 _Jv_loadIndexes (&pool
->data
[name_and_type_index
],
2140 name_index
, desc_index
);
2142 *name
= pool
->data
[name_index
].utf8
;
2143 *fmtype
= pool
->data
[desc_index
].utf8
;
2145 return check_class_constant (class_index
);
2148 // Return field's type, compute class' type if requested.
2149 type
check_field_constant (int index
, type
*class_type
= NULL
)
2151 _Jv_Utf8Const
*name
, *field_type
;
2152 type ct
= handle_field_or_method (index
,
2153 JV_CONSTANT_Fieldref
,
2154 &name
, &field_type
);
2157 if (field_type
->data
[0] == '[' || field_type
->data
[0] == 'L')
2158 return type (field_type
, this);
2159 return get_type_val_for_signature (field_type
->data
[0]);
2162 type
check_method_constant (int index
, bool is_interface
,
2163 _Jv_Utf8Const
**method_name
,
2164 _Jv_Utf8Const
**method_signature
)
2166 return handle_field_or_method (index
,
2168 ? JV_CONSTANT_InterfaceMethodref
2169 : JV_CONSTANT_Methodref
),
2170 method_name
, method_signature
);
2173 type
get_one_type (char *&p
)
2191 _Jv_Utf8Const
*name
= make_utf8_const (start
, p
- start
);
2192 return type (name
, this);
2195 // Casting to jchar here is ok since we are looking at an ASCII
2197 type_val rt
= get_type_val_for_signature (jchar (v
));
2199 if (arraycount
== 0)
2201 // Callers of this function eventually push their arguments on
2202 // the stack. So, promote them here.
2203 return type (rt
).promote ();
2206 jclass k
= construct_primitive_array_type (rt
);
2207 while (--arraycount
> 0)
2208 k
= _Jv_GetArrayClass (k
, NULL
);
2209 return type (k
, this);
2212 void compute_argument_types (_Jv_Utf8Const
*signature
,
2215 char *p
= signature
->data
;
2221 types
[i
++] = get_one_type (p
);
2224 type
compute_return_type (_Jv_Utf8Const
*signature
)
2226 char *p
= signature
->data
;
2230 return get_one_type (p
);
2233 void check_return_type (type onstack
)
2235 type rt
= compute_return_type (current_method
->self
->signature
);
2236 if (! rt
.compatible (onstack
, this))
2237 verify_fail ("incompatible return type");
2240 // Initialize the stack for the new method. Returns true if this
2241 // method is an instance initializer.
2242 bool initialize_stack ()
2245 bool is_init
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2247 bool is_clinit
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2250 using namespace java::lang::reflect
;
2251 if (! Modifier::isStatic (current_method
->self
->accflags
))
2253 type
kurr (current_class
, this);
2256 kurr
.set_uninitialized (type::SELF
, this);
2260 verify_fail ("<clinit> method must be static");
2261 set_variable (0, kurr
);
2262 current_state
->set_this_type (kurr
);
2268 verify_fail ("<init> method must be non-static");
2271 // We have to handle wide arguments specially here.
2272 int arg_count
= _Jv_count_arguments (current_method
->self
->signature
);
2273 type arg_types
[arg_count
];
2274 compute_argument_types (current_method
->self
->signature
, arg_types
);
2275 for (int i
= 0; i
< arg_count
; ++i
)
2277 set_variable (var
, arg_types
[i
]);
2279 if (arg_types
[i
].iswide ())
2286 void verify_instructions_0 ()
2288 current_state
= new state (current_method
->max_stack
,
2289 current_method
->max_locals
);
2294 // True if we are verifying an instance initializer.
2295 bool this_is_init
= initialize_stack ();
2297 states
= (state
**) _Jv_Malloc (sizeof (state
*)
2298 * current_method
->code_length
);
2299 for (int i
= 0; i
< current_method
->code_length
; ++i
)
2302 next_verify_pc
= state::NO_NEXT
;
2306 // If the PC was invalidated, get a new one from the work list.
2307 if (PC
== state::NO_NEXT
)
2310 if (PC
== state::INVALID
)
2311 verify_fail ("can't happen: saw state::INVALID");
2312 if (PC
== state::NO_NEXT
)
2314 debug_print ("== State pop from pending list\n");
2315 // Set up the current state.
2316 current_state
->copy (states
[PC
], current_method
->max_stack
,
2317 current_method
->max_locals
);
2321 // Control can't fall off the end of the bytecode. We
2322 // only need to check this in the fall-through case,
2323 // because branch bounds are checked when they are
2325 if (PC
>= current_method
->code_length
)
2326 verify_fail ("fell off end");
2328 // We only have to do this checking in the situation where
2329 // control flow falls through from the previous
2330 // instruction. Otherwise merging is done at the time we
2332 if (states
[PC
] != NULL
)
2334 // We've already visited this instruction. So merge
2335 // the states together. If this yields no change then
2336 // we don't have to re-verify. However, if the new
2337 // state is an the result of an unmerged `ret', we
2338 // must continue through it.
2339 debug_print ("== Fall through merge\n");
2340 states
[PC
]->print ("Old", PC
, current_method
->max_stack
,
2341 current_method
->max_locals
);
2342 current_state
->print ("Cur", PC
, current_method
->max_stack
,
2343 current_method
->max_locals
);
2344 if (! current_state
->merge (states
[PC
], false,
2345 current_method
->max_locals
, this)
2346 && ! states
[PC
]->is_unmerged_ret_state (current_method
->max_locals
))
2348 debug_print ("== Fall through optimization\n");
2352 // Save a copy of it for later.
2353 states
[PC
]->copy (current_state
, current_method
->max_stack
,
2354 current_method
->max_locals
);
2355 current_state
->print ("New", PC
, current_method
->max_stack
,
2356 current_method
->max_locals
);
2360 // We only have to keep saved state at branch targets. If
2361 // we're at a branch target and the state here hasn't been set
2362 // yet, we set it now.
2363 if (states
[PC
] == NULL
&& (flags
[PC
] & FLAG_BRANCH_TARGET
))
2365 states
[PC
] = new state (current_state
, current_method
->max_stack
,
2366 current_method
->max_locals
);
2369 // Set this before handling exceptions so that debug output is
2373 // Update states for all active exception handlers. Ordinarily
2374 // there are not many exception handlers. So we simply run
2375 // through them all.
2376 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
2378 if (PC
>= exception
[i
].start_pc
.i
&& PC
< exception
[i
].end_pc
.i
)
2380 type
handler (&java::lang::Throwable::class$
, this);
2381 if (exception
[i
].handler_type
.i
!= 0)
2382 handler
= check_class_constant (exception
[i
].handler_type
.i
);
2383 push_exception_jump (handler
, exception
[i
].handler_pc
.i
);
2387 current_state
->print (" ", PC
, current_method
->max_stack
,
2388 current_method
->max_locals
);
2389 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
2395 case op_aconst_null
:
2396 push_type (null_type
);
2406 push_type (int_type
);
2411 push_type (long_type
);
2417 push_type (float_type
);
2422 push_type (double_type
);
2427 push_type (int_type
);
2432 push_type (int_type
);
2436 push_type (check_constant (get_byte ()));
2439 push_type (check_constant (get_ushort ()));
2442 push_type (check_wide_constant (get_ushort ()));
2446 push_type (get_variable (get_byte (), int_type
));
2449 push_type (get_variable (get_byte (), long_type
));
2452 push_type (get_variable (get_byte (), float_type
));
2455 push_type (get_variable (get_byte (), double_type
));
2458 push_type (get_variable (get_byte (), reference_type
));
2465 push_type (get_variable (opcode
- op_iload_0
, int_type
));
2471 push_type (get_variable (opcode
- op_lload_0
, long_type
));
2477 push_type (get_variable (opcode
- op_fload_0
, float_type
));
2483 push_type (get_variable (opcode
- op_dload_0
, double_type
));
2489 push_type (get_variable (opcode
- op_aload_0
, reference_type
));
2492 pop_type (int_type
);
2493 push_type (require_array_type (pop_init_ref (reference_type
),
2497 pop_type (int_type
);
2498 push_type (require_array_type (pop_init_ref (reference_type
),
2502 pop_type (int_type
);
2503 push_type (require_array_type (pop_init_ref (reference_type
),
2507 pop_type (int_type
);
2508 push_type (require_array_type (pop_init_ref (reference_type
),
2512 pop_type (int_type
);
2513 push_type (require_array_type (pop_init_ref (reference_type
),
2517 pop_type (int_type
);
2518 require_array_type (pop_init_ref (reference_type
), byte_type
);
2519 push_type (int_type
);
2522 pop_type (int_type
);
2523 require_array_type (pop_init_ref (reference_type
), char_type
);
2524 push_type (int_type
);
2527 pop_type (int_type
);
2528 require_array_type (pop_init_ref (reference_type
), short_type
);
2529 push_type (int_type
);
2532 set_variable (get_byte (), pop_type (int_type
));
2535 set_variable (get_byte (), pop_type (long_type
));
2538 set_variable (get_byte (), pop_type (float_type
));
2541 set_variable (get_byte (), pop_type (double_type
));
2544 set_variable (get_byte (), pop_ref_or_return ());
2550 set_variable (opcode
- op_istore_0
, pop_type (int_type
));
2556 set_variable (opcode
- op_lstore_0
, pop_type (long_type
));
2562 set_variable (opcode
- op_fstore_0
, pop_type (float_type
));
2568 set_variable (opcode
- op_dstore_0
, pop_type (double_type
));
2574 set_variable (opcode
- op_astore_0
, pop_ref_or_return ());
2577 pop_type (int_type
);
2578 pop_type (int_type
);
2579 require_array_type (pop_init_ref (reference_type
), int_type
);
2582 pop_type (long_type
);
2583 pop_type (int_type
);
2584 require_array_type (pop_init_ref (reference_type
), long_type
);
2587 pop_type (float_type
);
2588 pop_type (int_type
);
2589 require_array_type (pop_init_ref (reference_type
), float_type
);
2592 pop_type (double_type
);
2593 pop_type (int_type
);
2594 require_array_type (pop_init_ref (reference_type
), double_type
);
2597 pop_type (reference_type
);
2598 pop_type (int_type
);
2599 require_array_type (pop_init_ref (reference_type
), reference_type
);
2602 pop_type (int_type
);
2603 pop_type (int_type
);
2604 require_array_type (pop_init_ref (reference_type
), byte_type
);
2607 pop_type (int_type
);
2608 pop_type (int_type
);
2609 require_array_type (pop_init_ref (reference_type
), char_type
);
2612 pop_type (int_type
);
2613 pop_type (int_type
);
2614 require_array_type (pop_init_ref (reference_type
), short_type
);
2621 type t
= pop_raw ();
2645 type t2
= pop_raw ();
2660 type t
= pop_raw ();
2675 type t1
= pop_raw ();
2692 type t1
= pop_raw ();
2695 type t2
= pop_raw ();
2713 type t3
= pop_raw ();
2751 pop_type (int_type
);
2752 push_type (pop_type (int_type
));
2762 pop_type (long_type
);
2763 push_type (pop_type (long_type
));
2768 pop_type (int_type
);
2769 push_type (pop_type (long_type
));
2776 pop_type (float_type
);
2777 push_type (pop_type (float_type
));
2784 pop_type (double_type
);
2785 push_type (pop_type (double_type
));
2791 push_type (pop_type (int_type
));
2794 push_type (pop_type (long_type
));
2797 push_type (pop_type (float_type
));
2800 push_type (pop_type (double_type
));
2803 get_variable (get_byte (), int_type
);
2807 pop_type (int_type
);
2808 push_type (long_type
);
2811 pop_type (int_type
);
2812 push_type (float_type
);
2815 pop_type (int_type
);
2816 push_type (double_type
);
2819 pop_type (long_type
);
2820 push_type (int_type
);
2823 pop_type (long_type
);
2824 push_type (float_type
);
2827 pop_type (long_type
);
2828 push_type (double_type
);
2831 pop_type (float_type
);
2832 push_type (int_type
);
2835 pop_type (float_type
);
2836 push_type (long_type
);
2839 pop_type (float_type
);
2840 push_type (double_type
);
2843 pop_type (double_type
);
2844 push_type (int_type
);
2847 pop_type (double_type
);
2848 push_type (long_type
);
2851 pop_type (double_type
);
2852 push_type (float_type
);
2855 pop_type (long_type
);
2856 pop_type (long_type
);
2857 push_type (int_type
);
2861 pop_type (float_type
);
2862 pop_type (float_type
);
2863 push_type (int_type
);
2867 pop_type (double_type
);
2868 pop_type (double_type
);
2869 push_type (int_type
);
2877 pop_type (int_type
);
2878 push_jump (get_short ());
2886 pop_type (int_type
);
2887 pop_type (int_type
);
2888 push_jump (get_short ());
2892 pop_type (reference_type
);
2893 pop_type (reference_type
);
2894 push_jump (get_short ());
2897 push_jump (get_short ());
2901 handle_jsr_insn (get_short ());
2904 handle_ret_insn (get_byte ());
2906 case op_tableswitch
:
2908 pop_type (int_type
);
2910 push_jump (get_int ());
2911 jint low
= get_int ();
2912 jint high
= get_int ();
2913 // Already checked LOW -vs- HIGH.
2914 for (int i
= low
; i
<= high
; ++i
)
2915 push_jump (get_int ());
2920 case op_lookupswitch
:
2922 pop_type (int_type
);
2924 push_jump (get_int ());
2925 jint npairs
= get_int ();
2926 // Already checked NPAIRS >= 0.
2928 for (int i
= 0; i
< npairs
; ++i
)
2930 jint key
= get_int ();
2931 if (i
> 0 && key
<= lastkey
)
2932 verify_fail ("lookupswitch pairs unsorted", start_PC
);
2934 push_jump (get_int ());
2940 check_return_type (pop_type (int_type
));
2944 check_return_type (pop_type (long_type
));
2948 check_return_type (pop_type (float_type
));
2952 check_return_type (pop_type (double_type
));
2956 check_return_type (pop_init_ref (reference_type
));
2960 // We only need to check this when the return type is
2961 // void, because all instance initializers return void.
2963 current_state
->check_this_initialized (this);
2964 check_return_type (void_type
);
2968 push_type (check_field_constant (get_ushort ()));
2971 pop_type (check_field_constant (get_ushort ()));
2976 type field
= check_field_constant (get_ushort (), &klass
);
2984 type field
= check_field_constant (get_ushort (), &klass
);
2987 // We have an obscure special case here: we can use
2988 // `putfield' on a field declared in this class, even if
2989 // `this' has not yet been initialized.
2990 if (! current_state
->this_type
.isinitialized ()
2991 && current_state
->this_type
.pc
== type::SELF
)
2992 klass
.set_uninitialized (type::SELF
, this);
2997 case op_invokevirtual
:
2998 case op_invokespecial
:
2999 case op_invokestatic
:
3000 case op_invokeinterface
:
3002 _Jv_Utf8Const
*method_name
, *method_signature
;
3004 = check_method_constant (get_ushort (),
3005 opcode
== op_invokeinterface
,
3008 // NARGS is only used when we're processing
3009 // invokeinterface. It is simplest for us to compute it
3010 // here and then verify it later.
3012 if (opcode
== op_invokeinterface
)
3014 nargs
= get_byte ();
3015 if (get_byte () != 0)
3016 verify_fail ("invokeinterface dummy byte is wrong");
3019 bool is_init
= false;
3020 if (_Jv_equalUtf8Consts (method_name
, gcj::init_name
))
3023 if (opcode
!= op_invokespecial
)
3024 verify_fail ("can't invoke <init>");
3026 else if (method_name
->data
[0] == '<')
3027 verify_fail ("can't invoke method starting with `<'");
3029 // Pop arguments and check types.
3030 int arg_count
= _Jv_count_arguments (method_signature
);
3031 type arg_types
[arg_count
];
3032 compute_argument_types (method_signature
, arg_types
);
3033 for (int i
= arg_count
- 1; i
>= 0; --i
)
3035 // This is only used for verifying the byte for
3037 nargs
-= arg_types
[i
].depth ();
3038 pop_init_ref (arg_types
[i
]);
3041 if (opcode
== op_invokeinterface
3043 verify_fail ("wrong argument count for invokeinterface");
3045 if (opcode
!= op_invokestatic
)
3047 type t
= class_type
;
3050 // In this case the PC doesn't matter.
3051 t
.set_uninitialized (type::UNINIT
, this);
3052 // FIXME: check to make sure that the <init>
3053 // call is to the right class.
3054 // It must either be super or an exact class
3057 type raw
= pop_raw ();
3058 if (! t
.compatible (raw
, this))
3059 verify_fail ("incompatible type on stack");
3062 current_state
->set_initialized (raw
.get_pc (),
3063 current_method
->max_locals
);
3066 type rt
= compute_return_type (method_signature
);
3074 type t
= check_class_constant (get_ushort ());
3075 if (t
.isarray () || t
.isinterface (this) || t
.isabstract (this))
3076 verify_fail ("type is array, interface, or abstract");
3077 t
.set_uninitialized (start_PC
, this);
3084 int atype
= get_byte ();
3085 // We intentionally have chosen constants to make this
3087 if (atype
< boolean_type
|| atype
> long_type
)
3088 verify_fail ("type not primitive", start_PC
);
3089 pop_type (int_type
);
3090 type
t (construct_primitive_array_type (type_val (atype
)), this);
3095 pop_type (int_type
);
3096 push_type (check_class_constant (get_ushort ()).to_array (this));
3098 case op_arraylength
:
3100 type t
= pop_init_ref (reference_type
);
3101 if (! t
.isarray () && ! t
.isnull ())
3102 verify_fail ("array type expected");
3103 push_type (int_type
);
3107 pop_type (type (&java::lang::Throwable::class$
, this));
3111 pop_init_ref (reference_type
);
3112 push_type (check_class_constant (get_ushort ()));
3115 pop_init_ref (reference_type
);
3116 check_class_constant (get_ushort ());
3117 push_type (int_type
);
3119 case op_monitorenter
:
3120 pop_init_ref (reference_type
);
3122 case op_monitorexit
:
3123 pop_init_ref (reference_type
);
3127 switch (get_byte ())
3130 push_type (get_variable (get_ushort (), int_type
));
3133 push_type (get_variable (get_ushort (), long_type
));
3136 push_type (get_variable (get_ushort (), float_type
));
3139 push_type (get_variable (get_ushort (), double_type
));
3142 push_type (get_variable (get_ushort (), reference_type
));
3145 set_variable (get_ushort (), pop_type (int_type
));
3148 set_variable (get_ushort (), pop_type (long_type
));
3151 set_variable (get_ushort (), pop_type (float_type
));
3154 set_variable (get_ushort (), pop_type (double_type
));
3157 set_variable (get_ushort (), pop_init_ref (reference_type
));
3160 handle_ret_insn (get_short ());
3163 get_variable (get_ushort (), int_type
);
3167 verify_fail ("unrecognized wide instruction", start_PC
);
3171 case op_multianewarray
:
3173 type atype
= check_class_constant (get_ushort ());
3174 int dim
= get_byte ();
3176 verify_fail ("too few dimensions to multianewarray", start_PC
);
3177 atype
.verify_dimensions (dim
, this);
3178 for (int i
= 0; i
< dim
; ++i
)
3179 pop_type (int_type
);
3185 pop_type (reference_type
);
3186 push_jump (get_short ());
3189 push_jump (get_int ());
3193 handle_jsr_insn (get_int ());
3196 // These are unused here, but we call them out explicitly
3197 // so that -Wswitch-enum doesn't complain.
3203 case op_putstatic_1
:
3204 case op_putstatic_2
:
3205 case op_putstatic_4
:
3206 case op_putstatic_8
:
3207 case op_putstatic_a
:
3209 case op_getfield_2s
:
3210 case op_getfield_2u
:
3214 case op_getstatic_1
:
3215 case op_getstatic_2s
:
3216 case op_getstatic_2u
:
3217 case op_getstatic_4
:
3218 case op_getstatic_8
:
3219 case op_getstatic_a
:
3221 // Unrecognized opcode.
3222 verify_fail ("unrecognized instruction in verify_instructions_0",
3230 void verify_instructions ()
3233 verify_instructions_0 ();
3236 _Jv_BytecodeVerifier (_Jv_InterpMethod
*m
)
3238 // We just print the text as utf-8. This is just for debugging
3240 debug_print ("--------------------------------\n");
3241 debug_print ("-- Verifying method `%s'\n", m
->self
->name
->data
);
3244 bytecode
= m
->bytecode ();
3245 exception
= m
->exceptions ();
3246 current_class
= m
->defining_class
;
3253 entry_points
= NULL
;
3256 ~_Jv_BytecodeVerifier ()
3265 for (int i
= 0; i
< current_method
->code_length
; ++i
)
3267 if (jsr_ptrs
[i
] != NULL
)
3269 subr_info
*info
= jsr_ptrs
[i
];
3270 while (info
!= NULL
)
3272 subr_info
*next
= info
->next
;
3278 _Jv_Free (jsr_ptrs
);
3281 while (utf8_list
!= NULL
)
3283 linked_utf8
*n
= utf8_list
->next
;
3284 _Jv_Free (utf8_list
->val
);
3285 _Jv_Free (utf8_list
);
3289 while (entry_points
!= NULL
)
3291 subr_entry_info
*next
= entry_points
->next
;
3292 _Jv_Free (entry_points
);
3293 entry_points
= next
;
3296 while (isect_list
!= NULL
)
3298 ref_intersection
*next
= isect_list
->alloc_next
;
3306 _Jv_VerifyMethod (_Jv_InterpMethod
*meth
)
3308 _Jv_BytecodeVerifier
v (meth
);
3309 v
.verify_instructions ();
3311 #endif /* INTERPRETER */