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>
22 // On Solaris 10/x86, <signal.h> indirectly includes <ia32/sys/reg.h>, which
23 // defines PC since g++ predefines __EXTENSIONS__. Undef here to avoid clash
24 // with PC member of class _Jv_BytecodeVerifier below.
29 #include <java/lang/Class.h>
30 #include <java/lang/VerifyError.h>
31 #include <java/lang/Throwable.h>
32 #include <java/lang/reflect/Modifier.h>
33 #include <java/lang/StringBuffer.h>
37 #endif /* VERIFY_DEBUG */
40 static void debug_print (const char *fmt
, ...)
41 __attribute__ ((format (printf
, 1, 2)));
44 debug_print (const char *fmt
, ...)
49 vfprintf (stderr
, fmt
, ap
);
51 #endif /* VERIFY_DEBUG */
54 class _Jv_BytecodeVerifier
58 static const int FLAG_INSN_START
= 1;
59 static const int FLAG_BRANCH_TARGET
= 2;
64 struct subr_entry_info
;
66 struct ref_intersection
;
70 // The PC corresponding to the start of the current instruction.
73 // The current state of the stack, locals, etc.
76 // We store the state at branch targets, for merging. This holds
80 // We keep a linked list of all the PCs which we must reverify.
81 // The link is done using the PC values. This is the head of the
85 // We keep some flags for each instruction. The values are the
86 // FLAG_* constants defined above.
89 // We need to keep track of which instructions can call a given
90 // subroutine. FIXME: this is inefficient. We keep a linked list
91 // of all calling `jsr's at at each jsr target.
94 // We keep a linked list of entries which map each `ret' instruction
95 // to its unique subroutine entry point. We expect that there won't
96 // be many `ret' instructions, so a linked list is ok.
97 subr_entry_info
*entry_points
;
99 // The bytecode itself.
100 unsigned char *bytecode
;
102 _Jv_InterpException
*exception
;
105 jclass current_class
;
107 _Jv_InterpMethod
*current_method
;
109 // A linked list of utf8 objects we allocate. This is really ugly,
110 // but without this our utf8 objects would be collected.
111 linked_utf8
*utf8_list
;
113 // A linked list of all ref_intersection objects we allocate.
114 ref_intersection
*isect_list
;
122 _Jv_Utf8Const
*make_utf8_const (char *s
, int len
)
124 _Jv_Utf8Const
*val
= _Jv_makeUtf8Const (s
, len
);
125 _Jv_Utf8Const
*r
= (_Jv_Utf8Const
*) _Jv_Malloc (sizeof (_Jv_Utf8Const
)
128 r
->length
= val
->length
;
130 memcpy (r
->data
, val
->data
, val
->length
+ 1);
132 linked_utf8
*lu
= (linked_utf8
*) _Jv_Malloc (sizeof (linked_utf8
));
134 lu
->next
= utf8_list
;
140 __attribute__ ((__noreturn__
)) void verify_fail (char *s
, jint pc
= -1)
142 using namespace java::lang
;
143 StringBuffer
*buf
= new StringBuffer ();
145 buf
->append (JvNewStringLatin1 ("verification failed"));
150 buf
->append (JvNewStringLatin1 (" at PC "));
154 _Jv_InterpMethod
*method
= current_method
;
155 buf
->append (JvNewStringLatin1 (" in "));
156 buf
->append (current_class
->getName());
157 buf
->append ((jchar
) ':');
158 buf
->append (JvNewStringUTF (method
->get_method()->name
->data
));
159 buf
->append ((jchar
) '(');
160 buf
->append (JvNewStringUTF (method
->get_method()->signature
->data
));
161 buf
->append ((jchar
) ')');
163 buf
->append (JvNewStringLatin1 (": "));
164 buf
->append (JvNewStringLatin1 (s
));
165 throw new java::lang::VerifyError (buf
->toString ());
168 // This enum holds a list of tags for all the different types we
169 // need to handle. Reference types are treated specially by the
175 // The values for primitive types are chosen to correspond to values
176 // specified to newarray.
186 // Used when overwriting second word of a double or long in the
187 // local variables. Also used after merging local variable states
188 // to indicate an unusable value.
193 // There is an obscure special case which requires us to note when
194 // a local variable has not been used by a subroutine. See
195 // push_jump_merge for more information.
196 unused_by_subroutine_type
,
198 // Everything after `reference_type' must be a reference type.
201 uninitialized_reference_type
204 // This represents a merged class type. Some verifiers (including
205 // earlier versions of this one) will compute the intersection of
206 // two class types when merging states. However, this loses
207 // critical information about interfaces implemented by the various
208 // classes. So instead we keep track of all the actual classes that
210 struct ref_intersection
212 // Whether or not this type has been resolved.
218 // For a resolved reference type, this is a pointer to the class.
220 // For other reference types, this it the name of the class.
224 // Link to the next reference in the intersection.
225 ref_intersection
*ref_next
;
227 // This is used to keep track of all the allocated
228 // ref_intersection objects, so we can free them.
229 // FIXME: we should allocate these in chunks.
230 ref_intersection
*alloc_next
;
232 ref_intersection (jclass klass
, _Jv_BytecodeVerifier
*verifier
)
237 alloc_next
= verifier
->isect_list
;
238 verifier
->isect_list
= this;
241 ref_intersection (_Jv_Utf8Const
*name
, _Jv_BytecodeVerifier
*verifier
)
246 alloc_next
= verifier
->isect_list
;
247 verifier
->isect_list
= this;
250 ref_intersection (ref_intersection
*dup
, ref_intersection
*tail
,
251 _Jv_BytecodeVerifier
*verifier
)
254 is_resolved
= dup
->is_resolved
;
256 alloc_next
= verifier
->isect_list
;
257 verifier
->isect_list
= this;
260 bool equals (ref_intersection
*other
, _Jv_BytecodeVerifier
*verifier
)
262 if (! is_resolved
&& ! other
->is_resolved
263 && _Jv_equalUtf8Consts (data
.name
, other
->data
.name
))
267 if (! other
->is_resolved
)
268 other
->resolve (verifier
);
269 return data
.klass
== other
->data
.klass
;
272 // Merge THIS type into OTHER, returning the result. This will
273 // return OTHER if all the classes in THIS already appear in
275 ref_intersection
*merge (ref_intersection
*other
,
276 _Jv_BytecodeVerifier
*verifier
)
278 ref_intersection
*tail
= other
;
279 for (ref_intersection
*self
= this; self
!= NULL
; self
= self
->ref_next
)
282 for (ref_intersection
*iter
= other
; iter
!= NULL
;
283 iter
= iter
->ref_next
)
285 if (iter
->equals (self
, verifier
))
293 tail
= new ref_intersection (self
, tail
, verifier
);
298 void resolve (_Jv_BytecodeVerifier
*verifier
)
303 using namespace java::lang
;
304 java::lang::ClassLoader
*loader
305 = verifier
->current_class
->getClassLoaderInternal();
306 // We might see either kind of name. Sigh.
307 if (data
.name
->data
[0] == 'L'
308 && data
.name
->data
[data
.name
->length
- 1] == ';')
309 data
.klass
= _Jv_FindClassFromSignature (data
.name
->data
, loader
);
311 data
.klass
= Class::forName (_Jv_NewStringUtf8Const (data
.name
),
316 // See if an object of type OTHER can be assigned to an object of
317 // type *THIS. This might resolve classes in one chain or the
319 bool compatible (ref_intersection
*other
,
320 _Jv_BytecodeVerifier
*verifier
)
322 ref_intersection
*self
= this;
324 for (; self
!= NULL
; self
= self
->ref_next
)
326 ref_intersection
*other_iter
= other
;
328 for (; other_iter
!= NULL
; other_iter
= other_iter
->ref_next
)
330 // Avoid resolving if possible.
331 if (! self
->is_resolved
332 && ! other_iter
->is_resolved
333 && _Jv_equalUtf8Consts (self
->data
.name
,
334 other_iter
->data
.name
))
337 if (! self
->is_resolved
)
338 self
->resolve(verifier
);
339 if (! other_iter
->is_resolved
)
340 other_iter
->resolve(verifier
);
342 if (! is_assignable_from_slow (self
->data
.klass
,
343 other_iter
->data
.klass
))
353 // assert (ref_next == NULL);
355 return data
.klass
->isArray ();
357 return data
.name
->data
[0] == '[';
360 bool isinterface (_Jv_BytecodeVerifier
*verifier
)
362 // assert (ref_next == NULL);
365 return data
.klass
->isInterface ();
368 bool isabstract (_Jv_BytecodeVerifier
*verifier
)
370 // assert (ref_next == NULL);
373 using namespace java::lang::reflect
;
374 return Modifier::isAbstract (data
.klass
->getModifiers ());
377 jclass
getclass (_Jv_BytecodeVerifier
*verifier
)
384 int count_dimensions ()
389 jclass k
= data
.klass
;
390 while (k
->isArray ())
392 k
= k
->getComponentType ();
398 char *p
= data
.name
->data
;
405 void *operator new (size_t bytes
)
407 return _Jv_Malloc (bytes
);
410 void operator delete (void *mem
)
416 // Return the type_val corresponding to a primitive signature
417 // character. For instance `I' returns `int.class'.
418 type_val
get_type_val_for_signature (jchar sig
)
451 verify_fail ("invalid signature");
456 // Return the type_val corresponding to a primitive class.
457 type_val
get_type_val_for_signature (jclass k
)
459 return get_type_val_for_signature ((jchar
) k
->method_count
);
462 // This is like _Jv_IsAssignableFrom, but it works even if SOURCE or
463 // TARGET haven't been prepared.
464 static bool is_assignable_from_slow (jclass target
, jclass source
)
466 // First, strip arrays.
467 while (target
->isArray ())
469 // If target is array, source must be as well.
470 if (! source
->isArray ())
472 target
= target
->getComponentType ();
473 source
= source
->getComponentType ();
477 if (target
== &java::lang::Object::class$
)
482 if (source
== target
)
485 if (target
->isPrimitive () || source
->isPrimitive ())
488 if (target
->isInterface ())
490 for (int i
= 0; i
< source
->interface_count
; ++i
)
492 // We use a recursive call because we also need to
493 // check superinterfaces.
494 if (is_assignable_from_slow (target
, source
->interfaces
[i
]))
498 source
= source
->getSuperclass ();
500 while (source
!= NULL
);
505 // This is used to keep track of which `jsr's correspond to a given
509 // PC of the instruction just after the jsr.
515 // This is used to keep track of which subroutine entry point
516 // corresponds to which `ret' instruction.
517 struct subr_entry_info
519 // PC of the subroutine entry point.
521 // PC of the `ret' instruction.
524 subr_entry_info
*next
;
527 // The `type' class is used to represent a single type in the
534 // For reference types, the representation of the type.
535 ref_intersection
*klass
;
537 // This is used when constructing a new object. It is the PC of the
538 // `new' instruction which created the object. We use the special
539 // value -2 to mean that this is uninitialized, and the special
540 // value -1 for the case where the current method is itself the
544 static const int UNINIT
= -2;
545 static const int SELF
= -1;
547 // Basic constructor.
550 key
= unsuitable_type
;
555 // Make a new instance given the type tag. We assume a generic
556 // `reference_type' means Object.
560 // For reference_type, if KLASS==NULL then that means we are
561 // looking for a generic object of any kind, including an
562 // uninitialized reference.
567 // Make a new instance given a class.
568 type (jclass k
, _Jv_BytecodeVerifier
*verifier
)
570 key
= reference_type
;
571 klass
= new ref_intersection (k
, verifier
);
575 // Make a new instance given the name of a class.
576 type (_Jv_Utf8Const
*n
, _Jv_BytecodeVerifier
*verifier
)
578 key
= reference_type
;
579 klass
= new ref_intersection (n
, verifier
);
591 // These operators are required because libgcj can't link in
593 void *operator new[] (size_t bytes
)
595 return _Jv_Malloc (bytes
);
598 void operator delete[] (void *mem
)
603 type
& operator= (type_val k
)
611 type
& operator= (const type
& t
)
619 // Promote a numeric type.
622 if (key
== boolean_type
|| key
== char_type
623 || key
== byte_type
|| key
== short_type
)
628 // Mark this type as the uninitialized result of `new'.
629 void set_uninitialized (int npc
, _Jv_BytecodeVerifier
*verifier
)
631 if (key
== reference_type
)
632 key
= uninitialized_reference_type
;
634 verifier
->verify_fail ("internal error in type::uninitialized");
638 // Mark this type as now initialized.
639 void set_initialized (int npc
)
641 if (npc
!= UNINIT
&& pc
== npc
&& key
== uninitialized_reference_type
)
643 key
= reference_type
;
649 // Return true if an object of type K can be assigned to a variable
650 // of type *THIS. Handle various special cases too. Might modify
651 // *THIS or K. Note however that this does not perform numeric
653 bool compatible (type
&k
, _Jv_BytecodeVerifier
*verifier
)
655 // Any type is compatible with the unsuitable type.
656 if (key
== unsuitable_type
)
659 if (key
< reference_type
|| k
.key
< reference_type
)
662 // The `null' type is convertible to any initialized reference
664 if (key
== null_type
)
665 return k
.key
!= uninitialized_reference_type
;
666 if (k
.key
== null_type
)
667 return key
!= uninitialized_reference_type
;
669 // A special case for a generic reference.
673 verifier
->verify_fail ("programmer error in type::compatible");
675 // An initialized type and an uninitialized type are not
677 if (isinitialized () != k
.isinitialized ())
680 // Two uninitialized objects are compatible if either:
681 // * The PCs are identical, or
682 // * One PC is UNINIT.
683 if (! isinitialized ())
685 if (pc
!= k
.pc
&& pc
!= UNINIT
&& k
.pc
!= UNINIT
)
689 return klass
->compatible(k
.klass
, verifier
);
694 return key
== void_type
;
699 return key
== long_type
|| key
== double_type
;
702 // Return number of stack or local variable slots taken by this
706 return iswide () ? 2 : 1;
709 bool isarray () const
711 // We treat null_type as not an array. This is ok based on the
712 // current uses of this method.
713 if (key
== reference_type
)
714 return klass
->isarray ();
720 return key
== null_type
;
723 bool isinterface (_Jv_BytecodeVerifier
*verifier
)
725 if (key
!= reference_type
)
727 return klass
->isinterface (verifier
);
730 bool isabstract (_Jv_BytecodeVerifier
*verifier
)
732 if (key
!= reference_type
)
734 return klass
->isabstract (verifier
);
737 // Return the element type of an array.
738 type
element_type (_Jv_BytecodeVerifier
*verifier
)
740 if (key
!= reference_type
)
741 verifier
->verify_fail ("programmer error in type::element_type()", -1);
743 jclass k
= klass
->getclass (verifier
)->getComponentType ();
744 if (k
->isPrimitive ())
745 return type (verifier
->get_type_val_for_signature (k
));
746 return type (k
, verifier
);
749 // Return the array type corresponding to an initialized
750 // reference. We could expand this to work for other kinds of
751 // types, but currently we don't need to.
752 type
to_array (_Jv_BytecodeVerifier
*verifier
)
754 if (key
!= reference_type
)
755 verifier
->verify_fail ("internal error in type::to_array()");
757 jclass k
= klass
->getclass (verifier
);
758 return type (_Jv_GetArrayClass (k
, k
->getClassLoaderInternal()),
762 bool isreference () const
764 return key
>= reference_type
;
772 bool isinitialized () const
774 return key
== reference_type
|| key
== null_type
;
777 bool isresolved () const
779 return (key
== reference_type
781 || key
== uninitialized_reference_type
);
784 void verify_dimensions (int ndims
, _Jv_BytecodeVerifier
*verifier
)
786 // The way this is written, we don't need to check isarray().
787 if (key
!= reference_type
)
788 verifier
->verify_fail ("internal error in verify_dimensions: not a reference type");
790 if (klass
->count_dimensions () < ndims
)
791 verifier
->verify_fail ("array type has fewer dimensions than required");
794 // Merge OLD_TYPE into this. On error throw exception.
795 bool merge (type
& old_type
, bool local_semantics
,
796 _Jv_BytecodeVerifier
*verifier
)
798 bool changed
= false;
799 bool refo
= old_type
.isreference ();
800 bool refn
= isreference ();
803 if (old_type
.key
== null_type
)
805 else if (key
== null_type
)
810 else if (isinitialized () != old_type
.isinitialized ())
811 verifier
->verify_fail ("merging initialized and uninitialized types");
814 if (! isinitialized ())
818 else if (old_type
.pc
== UNINIT
)
820 else if (pc
!= old_type
.pc
)
821 verifier
->verify_fail ("merging different uninitialized types");
824 ref_intersection
*merged
= old_type
.klass
->merge (klass
,
833 else if (refo
|| refn
|| key
!= old_type
.key
)
837 // If we're merging into an "unused" slot, then we
838 // simply accept whatever we're merging from.
839 if (key
== unused_by_subroutine_type
)
844 else if (old_type
.key
== unused_by_subroutine_type
)
848 // If we already have an `unsuitable' type, then we
849 // don't need to change again.
850 else if (key
!= unsuitable_type
)
852 key
= unsuitable_type
;
857 verifier
->verify_fail ("unmergeable type");
863 void print (void) const
868 case boolean_type
: c
= 'Z'; break;
869 case byte_type
: c
= 'B'; break;
870 case char_type
: c
= 'C'; break;
871 case short_type
: c
= 'S'; break;
872 case int_type
: c
= 'I'; break;
873 case long_type
: c
= 'J'; break;
874 case float_type
: c
= 'F'; break;
875 case double_type
: c
= 'D'; break;
876 case void_type
: c
= 'V'; break;
877 case unsuitable_type
: c
= '-'; break;
878 case return_address_type
: c
= 'r'; break;
879 case continuation_type
: c
= '+'; break;
880 case unused_by_subroutine_type
: c
= '_'; break;
881 case reference_type
: c
= 'L'; break;
882 case null_type
: c
= '@'; break;
883 case uninitialized_reference_type
: c
= 'U'; break;
885 debug_print ("%c", c
);
887 #endif /* VERIFY_DEBUG */
890 // This class holds all the state information we need for a given
894 // The current top of the stack, in terms of slots.
896 // The current depth of the stack. This will be larger than
897 // STACKTOP when wide types are on the stack.
901 // The local variables.
903 // Flags are used in subroutines to keep track of which local
904 // variables have been accessed. They are also used after
906 // If not 0, then we are in a subroutine. The value is the PC of
907 // the subroutine's entry point. We can use 0 as an exceptional
908 // value because PC=0 can never be a subroutine.
910 // This is used to keep a linked list of all the states which
911 // require re-verification. We use the PC to keep track.
913 // We keep track of the type of `this' specially. This is used to
914 // ensure that an instance initializer invokes another initializer
915 // on `this' before returning. We must keep track of this
916 // specially because otherwise we might be confused by code which
917 // assigns to locals[0] (overwriting `this') and then returns
918 // without really initializing.
920 // This is a list of all subroutines that have been seen at this
921 // point. Ordinarily this is NULL; it is only allocated and used
922 // in relatively weird situations involving non-ret exit from a
923 // subroutine. We have to keep track of this in this way to avoid
924 // endless recursion in these cases.
925 subr_info
*seen_subrs
;
927 // INVALID marks a state which is not on the linked list of states
928 // requiring reverification.
929 static const int INVALID
= -1;
930 // NO_NEXT marks the state at the end of the reverification list.
931 static const int NO_NEXT
= -2;
933 // This is used to mark the stack depth at the instruction just
934 // after a `jsr' when we haven't yet processed the corresponding
935 // `ret'. See handle_jsr_insn for more information.
936 static const int NO_STACK
= -1;
938 // This flag indicates that the local was changed in this
940 static const int FLAG_CHANGED
= 1;
941 // This is set only on the flags of the state of an instruction
942 // directly following a "jsr". It indicates that the local
943 // variable was changed by the subroutine corresponding to the
945 static const int FLAG_USED
= 2;
956 state (int max_stack
, int max_locals
)
961 stack
= new type
[max_stack
];
962 for (int i
= 0; i
< max_stack
; ++i
)
963 stack
[i
] = unsuitable_type
;
964 locals
= new type
[max_locals
];
965 flags
= (char *) _Jv_Malloc (sizeof (char) * max_locals
);
967 for (int i
= 0; i
< max_locals
; ++i
)
969 locals
[i
] = unsuitable_type
;
976 state (const state
*orig
, int max_stack
, int max_locals
,
977 bool ret_semantics
= false)
979 stack
= new type
[max_stack
];
980 locals
= new type
[max_locals
];
981 flags
= (char *) _Jv_Malloc (sizeof (char) * max_locals
);
983 copy (orig
, max_stack
, max_locals
, ret_semantics
);
998 void *operator new[] (size_t bytes
)
1000 return _Jv_Malloc (bytes
);
1003 void operator delete[] (void *mem
)
1008 void *operator new (size_t bytes
)
1010 return _Jv_Malloc (bytes
);
1013 void operator delete (void *mem
)
1020 subr_info
*info
= seen_subrs
;
1021 while (info
!= NULL
)
1023 subr_info
*next
= info
->next
;
1030 void copy (const state
*copy
, int max_stack
, int max_locals
,
1031 bool ret_semantics
= false)
1033 stacktop
= copy
->stacktop
;
1034 stackdepth
= copy
->stackdepth
;
1035 subroutine
= copy
->subroutine
;
1036 for (int i
= 0; i
< max_stack
; ++i
)
1037 stack
[i
] = copy
->stack
[i
];
1038 for (int i
= 0; i
< max_locals
; ++i
)
1040 // See push_jump_merge to understand this case.
1043 if ((copy
->flags
[i
] & FLAG_CHANGED
))
1045 // Changed in the subroutine, so we copy it here.
1046 locals
[i
] = copy
->locals
[i
];
1047 flags
[i
] |= FLAG_USED
;
1051 // Not changed in the subroutine. Use a special
1052 // type so the coming merge will overwrite.
1053 locals
[i
] = type (unused_by_subroutine_type
);
1057 locals
[i
] = copy
->locals
[i
];
1059 // Clear the flag unconditionally just so printouts look ok,
1060 // then only set it if we're still in a subroutine and it
1061 // did in fact change.
1062 flags
[i
] &= ~FLAG_CHANGED
;
1063 if (subroutine
&& (copy
->flags
[i
] & FLAG_CHANGED
) != 0)
1064 flags
[i
] |= FLAG_CHANGED
;
1068 if (copy
->seen_subrs
)
1070 for (subr_info
*info
= copy
->seen_subrs
;
1071 info
!= NULL
; info
= info
->next
)
1072 add_subr (info
->pc
);
1075 this_type
= copy
->this_type
;
1076 // Don't modify `next'.
1079 // Modify this state to reflect entry to an exception handler.
1080 void set_exception (type t
, int max_stack
)
1085 for (int i
= stacktop
; i
< max_stack
; ++i
)
1086 stack
[i
] = unsuitable_type
;
1089 // Modify this state to reflect entry into a subroutine.
1090 void enter_subroutine (int npc
, int max_locals
)
1093 // Mark all items as unchanged. Each subroutine needs to keep
1094 // track of its `changed' state independently. In the case of
1095 // nested subroutines, this information will be merged back into
1096 // parent by the `ret'.
1097 for (int i
= 0; i
< max_locals
; ++i
)
1098 flags
[i
] &= ~FLAG_CHANGED
;
1101 // Indicate that we've been in this this subroutine.
1102 void add_subr (int pc
)
1104 subr_info
*n
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
1106 n
->next
= seen_subrs
;
1110 // Merge STATE_OLD into this state. Destructively modifies this
1111 // state. Returns true if the new state was in fact changed.
1112 // Will throw an exception if the states are not mergeable.
1113 bool merge (state
*state_old
, bool ret_semantics
,
1114 int max_locals
, _Jv_BytecodeVerifier
*verifier
,
1115 bool jsr_semantics
= false)
1117 bool changed
= false;
1119 // Special handling for `this'. If one or the other is
1120 // uninitialized, then the merge is uninitialized.
1121 if (this_type
.isinitialized ())
1122 this_type
= state_old
->this_type
;
1124 // Merge subroutine states. Here we just keep track of what
1125 // subroutine we think we're in. We only check for a merge
1126 // (which is invalid) when we see a `ret'.
1127 if (subroutine
== state_old
->subroutine
)
1131 else if (subroutine
== 0)
1133 subroutine
= state_old
->subroutine
;
1138 // If the subroutines differ, and we haven't seen this
1139 // subroutine before, indicate that the state changed. This
1140 // is needed to detect when subroutines have merged.
1142 for (subr_info
*info
= seen_subrs
; info
!= NULL
; info
= info
->next
)
1144 if (info
->pc
== state_old
->subroutine
)
1152 add_subr (state_old
->subroutine
);
1157 // Merge stacks, including special handling for NO_STACK case.
1158 // If the destination is NO_STACK, this means it is the
1159 // instruction following a "jsr" and has not yet been processed
1160 // in any way. In this situation, if we are currently
1161 // processing a "ret", then we must *copy* any locals changed in
1162 // the subroutine into the current state. Merging in this
1163 // situation is incorrect because the locals we've noted didn't
1164 // come real program flow, they are just an artifact of how
1165 // we've chosen to handle the post-jsr state.
1166 bool copy_in_locals
= ret_semantics
&& stacktop
== NO_STACK
;
1168 if (state_old
->stacktop
== NO_STACK
)
1170 // This can happen if we're doing a pass-through jsr merge.
1171 // Here we can just ignore the stack.
1173 else if (stacktop
== NO_STACK
)
1175 stacktop
= state_old
->stacktop
;
1176 stackdepth
= state_old
->stackdepth
;
1177 for (int i
= 0; i
< stacktop
; ++i
)
1178 stack
[i
] = state_old
->stack
[i
];
1181 else if (state_old
->stacktop
!= stacktop
)
1182 verifier
->verify_fail ("stack sizes differ");
1185 for (int i
= 0; i
< state_old
->stacktop
; ++i
)
1187 if (stack
[i
].merge (state_old
->stack
[i
], false, verifier
))
1192 // Merge local variables.
1193 for (int i
= 0; i
< max_locals
; ++i
)
1195 // If we're not processing a `ret', then we merge every
1196 // local variable. If we are processing a `ret', then we
1197 // only merge locals which changed in the subroutine. When
1198 // processing a `ret', STATE_OLD is the state at the point
1199 // of the `ret', and THIS is the state just after the `jsr'.
1200 // See comment above for explanation of COPY_IN_LOCALS.
1203 if ((state_old
->flags
[i
] & FLAG_CHANGED
) != 0)
1205 locals
[i
] = state_old
->locals
[i
];
1207 // There's no point in calling note_variable here,
1208 // since we call it under the same condition before
1212 else if (jsr_semantics
&& (flags
[i
] & FLAG_USED
) != 0)
1214 // We are processing the "pass-through" part of a jsr
1215 // statement. In this particular case, the local was
1216 // changed by the subroutine. So, we have no work to
1217 // do, as the pre-jsr value does not survive the
1220 else if (! ret_semantics
1221 || (state_old
->flags
[i
] & FLAG_CHANGED
) != 0)
1223 // If we have ordinary (not ret) semantics, then we have
1224 // merging flow control, so we merge types. Or, we have
1225 // jsr pass-through semantics and the type survives the
1226 // subroutine (see above), so again we merge. Or,
1227 // finally, we have ret semantics and this value did
1228 // change, in which case we merge the change from the
1229 // subroutine into the post-jsr instruction.
1230 if (locals
[i
].merge (state_old
->locals
[i
], true, verifier
))
1232 // Note that we don't call `note_variable' here.
1233 // This change doesn't represent a real change to a
1234 // local, but rather a merge artifact. If we're in
1235 // a subroutine which is called with two
1236 // incompatible types in a slot that is unused by
1237 // the subroutine, then we don't want to mark that
1238 // variable as having been modified.
1243 // If we're in a subroutine, we must compute the union of
1244 // all the changed local variables.
1245 if ((state_old
->flags
[i
] & FLAG_CHANGED
) != 0)
1248 // If we're returning from a subroutine, we must mark the
1249 // post-jsr instruction with information about what changed,
1250 // so that future "pass-through" jsr merges work correctly.
1251 if (ret_semantics
&& (state_old
->flags
[i
] & FLAG_CHANGED
) != 0)
1252 flags
[i
] |= FLAG_USED
;
1258 // Throw an exception if there is an uninitialized object on the
1259 // stack or in a local variable. EXCEPTION_SEMANTICS controls
1260 // whether we're using backwards-branch or exception-handing
1262 void check_no_uninitialized_objects (int max_locals
,
1263 _Jv_BytecodeVerifier
*verifier
,
1264 bool exception_semantics
= false)
1266 if (! exception_semantics
)
1268 for (int i
= 0; i
< stacktop
; ++i
)
1269 if (stack
[i
].isreference () && ! stack
[i
].isinitialized ())
1270 verifier
->verify_fail ("uninitialized object on stack");
1273 for (int i
= 0; i
< max_locals
; ++i
)
1274 if (locals
[i
].isreference () && ! locals
[i
].isinitialized ())
1275 verifier
->verify_fail ("uninitialized object in local variable");
1277 check_this_initialized (verifier
);
1280 // Ensure that `this' has been initialized.
1281 void check_this_initialized (_Jv_BytecodeVerifier
*verifier
)
1283 if (this_type
.isreference () && ! this_type
.isinitialized ())
1284 verifier
->verify_fail ("`this' is uninitialized");
1287 // Set type of `this'.
1288 void set_this_type (const type
&k
)
1293 // Note that a local variable was modified.
1294 void note_variable (int index
)
1297 flags
[index
] |= FLAG_CHANGED
;
1300 // Mark each `new'd object we know of that was allocated at PC as
1302 void set_initialized (int pc
, int max_locals
)
1304 for (int i
= 0; i
< stacktop
; ++i
)
1305 stack
[i
].set_initialized (pc
);
1306 for (int i
= 0; i
< max_locals
; ++i
)
1307 locals
[i
].set_initialized (pc
);
1308 this_type
.set_initialized (pc
);
1311 // Return true if this state is the unmerged result of a `ret'.
1312 bool is_unmerged_ret_state (int max_locals
) const
1314 if (stacktop
== NO_STACK
)
1316 for (int i
= 0; i
< max_locals
; ++i
)
1318 if (locals
[i
].key
== unused_by_subroutine_type
)
1325 void print (const char *leader
, int pc
,
1326 int max_stack
, int max_locals
) const
1328 debug_print ("%s [%4d]: [stack] ", leader
, pc
);
1330 for (i
= 0; i
< stacktop
; ++i
)
1332 for (; i
< max_stack
; ++i
)
1334 debug_print (" [local] ");
1335 for (i
= 0; i
< max_locals
; ++i
)
1338 if ((flags
[i
] & FLAG_USED
) != 0)
1339 debug_print ((flags
[i
] & FLAG_CHANGED
) ? ">" : "<");
1341 debug_print ((flags
[i
] & FLAG_CHANGED
) ? "+" : " ");
1343 if (subroutine
== 0)
1344 debug_print (" | None");
1346 debug_print (" | %4d", subroutine
);
1347 debug_print (" | %p\n", this);
1350 inline void print (const char *, int, int, int) const
1353 #endif /* VERIFY_DEBUG */
1358 if (current_state
->stacktop
<= 0)
1359 verify_fail ("stack empty");
1360 type r
= current_state
->stack
[--current_state
->stacktop
];
1361 current_state
->stackdepth
-= r
.depth ();
1362 if (current_state
->stackdepth
< 0)
1363 verify_fail ("stack empty", start_PC
);
1369 type r
= pop_raw ();
1371 verify_fail ("narrow pop of wide type");
1375 type
pop_type (type match
)
1378 type t
= pop_raw ();
1379 if (! match
.compatible (t
, this))
1380 verify_fail ("incompatible type on stack");
1384 // Pop a reference which is guaranteed to be initialized. MATCH
1385 // doesn't have to be a reference type; in this case this acts like
1387 type
pop_init_ref (type match
)
1389 type t
= pop_raw ();
1390 if (t
.isreference () && ! t
.isinitialized ())
1391 verify_fail ("initialized reference required");
1392 else if (! match
.compatible (t
, this))
1393 verify_fail ("incompatible type on stack");
1397 // Pop a reference type or a return address.
1398 type
pop_ref_or_return ()
1400 type t
= pop_raw ();
1401 if (! t
.isreference () && t
.key
!= return_address_type
)
1402 verify_fail ("expected reference or return address on stack");
1406 void push_type (type t
)
1408 // If T is a numeric type like short, promote it to int.
1411 int depth
= t
.depth ();
1412 if (current_state
->stackdepth
+ depth
> current_method
->max_stack
)
1413 verify_fail ("stack overflow");
1414 current_state
->stack
[current_state
->stacktop
++] = t
;
1415 current_state
->stackdepth
+= depth
;
1418 void set_variable (int index
, type t
)
1420 // If T is a numeric type like short, promote it to int.
1423 int depth
= t
.depth ();
1424 if (index
> current_method
->max_locals
- depth
)
1425 verify_fail ("invalid local variable");
1426 current_state
->locals
[index
] = t
;
1427 current_state
->note_variable (index
);
1431 current_state
->locals
[index
+ 1] = continuation_type
;
1432 current_state
->note_variable (index
+ 1);
1434 if (index
> 0 && current_state
->locals
[index
- 1].iswide ())
1436 current_state
->locals
[index
- 1] = unsuitable_type
;
1437 // There's no need to call note_variable here.
1441 type
get_variable (int index
, type t
)
1443 int depth
= t
.depth ();
1444 if (index
> current_method
->max_locals
- depth
)
1445 verify_fail ("invalid local variable");
1446 if (! t
.compatible (current_state
->locals
[index
], this))
1447 verify_fail ("incompatible type in local variable");
1450 type
t (continuation_type
);
1451 if (! current_state
->locals
[index
+ 1].compatible (t
, this))
1452 verify_fail ("invalid local variable");
1454 return current_state
->locals
[index
];
1457 // Make sure ARRAY is an array type and that its elements are
1458 // compatible with type ELEMENT. Returns the actual element type.
1459 type
require_array_type (type array
, type element
)
1461 // An odd case. Here we just pretend that everything went ok. If
1462 // the requested element type is some kind of reference, return
1463 // the null type instead.
1464 if (array
.isnull ())
1465 return element
.isreference () ? type (null_type
) : element
;
1467 if (! array
.isarray ())
1468 verify_fail ("array required");
1470 type t
= array
.element_type (this);
1471 if (! element
.compatible (t
, this))
1473 // Special case for byte arrays, which must also be boolean
1476 if (element
.key
== byte_type
)
1478 type
e2 (boolean_type
);
1479 ok
= e2
.compatible (t
, this);
1482 verify_fail ("incompatible array element type");
1485 // Return T and not ELEMENT, because T might be specialized.
1491 if (PC
>= current_method
->code_length
)
1492 verify_fail ("premature end of bytecode");
1493 return (jint
) bytecode
[PC
++] & 0xff;
1498 jint b1
= get_byte ();
1499 jint b2
= get_byte ();
1500 return (jint
) ((b1
<< 8) | b2
) & 0xffff;
1505 jint b1
= get_byte ();
1506 jint b2
= get_byte ();
1507 jshort s
= (b1
<< 8) | b2
;
1513 jint b1
= get_byte ();
1514 jint b2
= get_byte ();
1515 jint b3
= get_byte ();
1516 jint b4
= get_byte ();
1517 return (b1
<< 24) | (b2
<< 16) | (b3
<< 8) | b4
;
1520 int compute_jump (int offset
)
1522 int npc
= start_PC
+ offset
;
1523 if (npc
< 0 || npc
>= current_method
->code_length
)
1524 verify_fail ("branch out of range", start_PC
);
1528 // Merge the indicated state into the state at the branch target and
1529 // schedule a new PC if there is a change. If RET_SEMANTICS is
1530 // true, then we are merging from a `ret' instruction into the
1531 // instruction after a `jsr'. This is a special case with its own
1532 // modified semantics. If JSR_SEMANTICS is true, then we're merging
1533 // some type information from a "jsr" instruction to the immediately
1534 // following instruction. In this situation we have to be careful
1535 // not to merge local variables whose values are modified by the
1536 // subroutine we're about to call.
1537 void push_jump_merge (int npc
, state
*nstate
,
1538 bool ret_semantics
= false,
1539 bool jsr_semantics
= false)
1541 bool changed
= true;
1542 if (states
[npc
] == NULL
)
1544 // There's a weird situation here. If are examining the
1545 // branch that results from a `ret', and there is not yet a
1546 // state available at the branch target (the instruction just
1547 // after the `jsr'), then we have to construct a special kind
1548 // of state at that point for future merging. This special
1549 // state has the type `unused_by_subroutine_type' in each slot
1550 // which was not modified by the subroutine.
1551 states
[npc
] = new state (nstate
, current_method
->max_stack
,
1552 current_method
->max_locals
, ret_semantics
);
1553 debug_print ("== New state in push_jump_merge (ret_semantics = %s)\n",
1554 ret_semantics
? "true" : "false");
1555 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1556 current_method
->max_locals
);
1560 debug_print ("== Merge states in push_jump_merge\n");
1561 nstate
->print ("Frm", start_PC
, current_method
->max_stack
,
1562 current_method
->max_locals
);
1563 states
[npc
]->print (" To", npc
, current_method
->max_stack
,
1564 current_method
->max_locals
);
1565 changed
= states
[npc
]->merge (nstate
, ret_semantics
,
1566 current_method
->max_locals
, this,
1568 states
[npc
]->print ("New", npc
, current_method
->max_stack
,
1569 current_method
->max_locals
);
1572 if (changed
&& states
[npc
]->next
== state::INVALID
)
1574 // The merge changed the state, and the new PC isn't yet on our
1575 // list of PCs to re-verify.
1576 states
[npc
]->next
= next_verify_pc
;
1577 next_verify_pc
= npc
;
1581 void push_jump (int offset
)
1583 int npc
= compute_jump (offset
);
1585 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1586 push_jump_merge (npc
, current_state
);
1589 void push_exception_jump (type t
, int pc
)
1591 current_state
->check_no_uninitialized_objects (current_method
->max_locals
,
1593 state
s (current_state
, current_method
->max_stack
,
1594 current_method
->max_locals
);
1595 if (current_method
->max_stack
< 1)
1596 verify_fail ("stack overflow at exception handler");
1597 s
.set_exception (t
, current_method
->max_stack
);
1598 push_jump_merge (pc
, &s
);
1603 int *prev_loc
= &next_verify_pc
;
1604 int npc
= next_verify_pc
;
1606 while (npc
!= state::NO_NEXT
)
1608 // If the next available PC is an unmerged `ret' state, then
1609 // we aren't yet ready to handle it. That's because we would
1610 // need all kind of special cases to do so. So instead we
1611 // defer this jump until after we've processed it via a
1612 // fall-through. This has to happen because the instruction
1613 // before this one must be a `jsr'.
1614 if (! states
[npc
]->is_unmerged_ret_state (current_method
->max_locals
))
1616 *prev_loc
= states
[npc
]->next
;
1617 states
[npc
]->next
= state::INVALID
;
1621 prev_loc
= &states
[npc
]->next
;
1622 npc
= states
[npc
]->next
;
1625 // Note that we might have gotten here even when there are
1626 // remaining states to process. That can happen if we find a
1627 // `jsr' without a `ret'.
1628 return state::NO_NEXT
;
1631 void invalidate_pc ()
1633 PC
= state::NO_NEXT
;
1636 void note_branch_target (int pc
, bool is_jsr_target
= false)
1638 // Don't check `pc <= PC', because we've advanced PC after
1639 // fetching the target and we haven't yet checked the next
1641 if (pc
< PC
&& ! (flags
[pc
] & FLAG_INSN_START
))
1642 verify_fail ("branch not to instruction start", start_PC
);
1643 flags
[pc
] |= FLAG_BRANCH_TARGET
;
1646 // Record the jsr which called this instruction.
1647 subr_info
*info
= (subr_info
*) _Jv_Malloc (sizeof (subr_info
));
1649 info
->next
= jsr_ptrs
[pc
];
1650 jsr_ptrs
[pc
] = info
;
1654 void skip_padding ()
1656 while ((PC
% 4) > 0)
1657 if (get_byte () != 0)
1658 verify_fail ("found nonzero padding byte");
1661 // Return the subroutine to which the instruction at PC belongs.
1662 int get_subroutine (int pc
)
1664 if (states
[pc
] == NULL
)
1666 return states
[pc
]->subroutine
;
1669 // Do the work for a `ret' instruction. INDEX is the index into the
1671 void handle_ret_insn (int index
)
1673 get_variable (index
, return_address_type
);
1675 int csub
= current_state
->subroutine
;
1677 verify_fail ("no subroutine");
1679 // Check to see if we've merged subroutines.
1680 subr_entry_info
*entry
;
1681 for (entry
= entry_points
; entry
!= NULL
; entry
= entry
->next
)
1683 if (entry
->ret_pc
== start_PC
)
1688 entry
= (subr_entry_info
*) _Jv_Malloc (sizeof (subr_entry_info
));
1690 entry
->ret_pc
= start_PC
;
1691 entry
->next
= entry_points
;
1692 entry_points
= entry
;
1694 else if (entry
->pc
!= csub
)
1695 verify_fail ("subroutines merged");
1697 for (subr_info
*subr
= jsr_ptrs
[csub
]; subr
!= NULL
; subr
= subr
->next
)
1699 // We might be returning to a `jsr' that is at the end of the
1700 // bytecode. This is ok if we never return from the called
1701 // subroutine, but if we see this here it is an error.
1702 if (subr
->pc
>= current_method
->code_length
)
1703 verify_fail ("fell off end");
1705 // Temporarily modify the current state so it looks like we're
1706 // in the enclosing context.
1707 current_state
->subroutine
= get_subroutine (subr
->pc
);
1709 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1710 push_jump_merge (subr
->pc
, current_state
, true);
1713 current_state
->subroutine
= csub
;
1717 // We're in the subroutine SUB, calling a subroutine at DEST. Make
1718 // sure this subroutine isn't already on the stack.
1719 void check_nonrecursive_call (int sub
, int dest
)
1724 verify_fail ("recursive subroutine call");
1725 for (subr_info
*info
= jsr_ptrs
[sub
]; info
!= NULL
; info
= info
->next
)
1726 check_nonrecursive_call (get_subroutine (info
->pc
), dest
);
1729 void handle_jsr_insn (int offset
)
1731 int npc
= compute_jump (offset
);
1734 current_state
->check_no_uninitialized_objects (current_method
->max_locals
, this);
1735 check_nonrecursive_call (current_state
->subroutine
, npc
);
1737 // Modify our state as appropriate for entry into a subroutine.
1738 push_type (return_address_type
);
1739 push_jump_merge (npc
, current_state
);
1741 pop_type (return_address_type
);
1743 // On entry to the subroutine, the subroutine number must be set
1744 // and the locals must be marked as cleared. We do this after
1745 // merging state so that we don't erroneously "notice" a variable
1746 // change merely on entry.
1747 states
[npc
]->enter_subroutine (npc
, current_method
->max_locals
);
1749 // Indicate that we don't know the stack depth of the instruction
1750 // following the `jsr'. The idea here is that we need to merge
1751 // the local variable state across the jsr, but the subroutine
1752 // might change the stack depth, so we can't make any assumptions
1753 // about it. So we have yet another special case. We know that
1754 // at this point PC points to the instruction after the jsr. Note
1755 // that it is ok to have a `jsr' at the end of the bytecode,
1756 // provided that the called subroutine never returns. So, we have
1757 // a special case here and another one when we handle the ret.
1758 if (PC
< current_method
->code_length
)
1760 current_state
->stacktop
= state::NO_STACK
;
1761 push_jump_merge (PC
, current_state
, false, true);
1766 jclass
construct_primitive_array_type (type_val prim
)
1772 k
= JvPrimClass (boolean
);
1775 k
= JvPrimClass (char);
1778 k
= JvPrimClass (float);
1781 k
= JvPrimClass (double);
1784 k
= JvPrimClass (byte
);
1787 k
= JvPrimClass (short);
1790 k
= JvPrimClass (int);
1793 k
= JvPrimClass (long);
1796 // These aren't used here but we call them out to avoid
1799 case unsuitable_type
:
1800 case return_address_type
:
1801 case continuation_type
:
1802 case unused_by_subroutine_type
:
1803 case reference_type
:
1805 case uninitialized_reference_type
:
1807 verify_fail ("unknown type in construct_primitive_array_type");
1809 k
= _Jv_GetArrayClass (k
, NULL
);
1813 // This pass computes the location of branch targets and also
1814 // instruction starts.
1815 void branch_prepass ()
1817 flags
= (char *) _Jv_Malloc (current_method
->code_length
);
1818 jsr_ptrs
= (subr_info
**) _Jv_Malloc (sizeof (subr_info
*)
1819 * current_method
->code_length
);
1821 for (int i
= 0; i
< current_method
->code_length
; ++i
)
1827 bool last_was_jsr
= false;
1830 while (PC
< current_method
->code_length
)
1832 // Set `start_PC' early so that error checking can have the
1835 flags
[PC
] |= FLAG_INSN_START
;
1837 // If the previous instruction was a jsr, then the next
1838 // instruction is a branch target -- the branch being the
1839 // corresponding `ret'.
1841 note_branch_target (PC
);
1842 last_was_jsr
= false;
1844 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
1848 case op_aconst_null
:
1984 case op_monitorenter
:
1985 case op_monitorexit
:
1993 case op_arraylength
:
2025 case op_invokespecial
:
2026 case op_invokestatic
:
2027 case op_invokevirtual
:
2031 case op_multianewarray
:
2037 last_was_jsr
= true;
2056 note_branch_target (compute_jump (get_short ()), last_was_jsr
);
2059 case op_tableswitch
:
2062 note_branch_target (compute_jump (get_int ()));
2063 jint low
= get_int ();
2064 jint hi
= get_int ();
2066 verify_fail ("invalid tableswitch", start_PC
);
2067 for (int i
= low
; i
<= hi
; ++i
)
2068 note_branch_target (compute_jump (get_int ()));
2072 case op_lookupswitch
:
2075 note_branch_target (compute_jump (get_int ()));
2076 int npairs
= get_int ();
2078 verify_fail ("too few pairs in lookupswitch", start_PC
);
2079 while (npairs
-- > 0)
2082 note_branch_target (compute_jump (get_int ()));
2087 case op_invokeinterface
:
2095 opcode
= (java_opcode
) get_byte ();
2097 if (opcode
== op_iinc
)
2103 last_was_jsr
= true;
2106 note_branch_target (compute_jump (get_int ()), last_was_jsr
);
2109 // These are unused here, but we call them out explicitly
2110 // so that -Wswitch-enum doesn't complain.
2116 case op_putstatic_1
:
2117 case op_putstatic_2
:
2118 case op_putstatic_4
:
2119 case op_putstatic_8
:
2120 case op_putstatic_a
:
2122 case op_getfield_2s
:
2123 case op_getfield_2u
:
2127 case op_getstatic_1
:
2128 case op_getstatic_2s
:
2129 case op_getstatic_2u
:
2130 case op_getstatic_4
:
2131 case op_getstatic_8
:
2132 case op_getstatic_a
:
2134 verify_fail ("unrecognized instruction in branch_prepass",
2138 // See if any previous branch tried to branch to the middle of
2139 // this instruction.
2140 for (int pc
= start_PC
+ 1; pc
< PC
; ++pc
)
2142 if ((flags
[pc
] & FLAG_BRANCH_TARGET
))
2143 verify_fail ("branch to middle of instruction", pc
);
2147 // Verify exception handlers.
2148 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
2150 if (! (flags
[exception
[i
].handler_pc
.i
] & FLAG_INSN_START
))
2151 verify_fail ("exception handler not at instruction start",
2152 exception
[i
].handler_pc
.i
);
2153 if (! (flags
[exception
[i
].start_pc
.i
] & FLAG_INSN_START
))
2154 verify_fail ("exception start not at instruction start",
2155 exception
[i
].start_pc
.i
);
2156 if (exception
[i
].end_pc
.i
!= current_method
->code_length
2157 && ! (flags
[exception
[i
].end_pc
.i
] & FLAG_INSN_START
))
2158 verify_fail ("exception end not at instruction start",
2159 exception
[i
].end_pc
.i
);
2161 flags
[exception
[i
].handler_pc
.i
] |= FLAG_BRANCH_TARGET
;
2165 void check_pool_index (int index
)
2167 if (index
< 0 || index
>= current_class
->constants
.size
)
2168 verify_fail ("constant pool index out of range", start_PC
);
2171 type
check_class_constant (int index
)
2173 check_pool_index (index
);
2174 _Jv_Constants
*pool
= ¤t_class
->constants
;
2175 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedClass
)
2176 return type (pool
->data
[index
].clazz
, this);
2177 else if (pool
->tags
[index
] == JV_CONSTANT_Class
)
2178 return type (pool
->data
[index
].utf8
, this);
2179 verify_fail ("expected class constant", start_PC
);
2182 type
check_constant (int index
)
2184 check_pool_index (index
);
2185 _Jv_Constants
*pool
= ¤t_class
->constants
;
2186 if (pool
->tags
[index
] == JV_CONSTANT_ResolvedString
2187 || pool
->tags
[index
] == JV_CONSTANT_String
)
2188 return type (&java::lang::String::class$
, this);
2189 else if (pool
->tags
[index
] == JV_CONSTANT_Integer
)
2190 return type (int_type
);
2191 else if (pool
->tags
[index
] == JV_CONSTANT_Float
)
2192 return type (float_type
);
2193 verify_fail ("String, int, or float constant expected", start_PC
);
2196 type
check_wide_constant (int index
)
2198 check_pool_index (index
);
2199 _Jv_Constants
*pool
= ¤t_class
->constants
;
2200 if (pool
->tags
[index
] == JV_CONSTANT_Long
)
2201 return type (long_type
);
2202 else if (pool
->tags
[index
] == JV_CONSTANT_Double
)
2203 return type (double_type
);
2204 verify_fail ("long or double constant expected", start_PC
);
2207 // Helper for both field and method. These are laid out the same in
2208 // the constant pool.
2209 type
handle_field_or_method (int index
, int expected
,
2210 _Jv_Utf8Const
**name
,
2211 _Jv_Utf8Const
**fmtype
)
2213 check_pool_index (index
);
2214 _Jv_Constants
*pool
= ¤t_class
->constants
;
2215 if (pool
->tags
[index
] != expected
)
2216 verify_fail ("didn't see expected constant", start_PC
);
2217 // Once we know we have a Fieldref or Methodref we assume that it
2218 // is correctly laid out in the constant pool. I think the code
2219 // in defineclass.cc guarantees this.
2220 _Jv_ushort class_index
, name_and_type_index
;
2221 _Jv_loadIndexes (&pool
->data
[index
],
2223 name_and_type_index
);
2224 _Jv_ushort name_index
, desc_index
;
2225 _Jv_loadIndexes (&pool
->data
[name_and_type_index
],
2226 name_index
, desc_index
);
2228 *name
= pool
->data
[name_index
].utf8
;
2229 *fmtype
= pool
->data
[desc_index
].utf8
;
2231 return check_class_constant (class_index
);
2234 // Return field's type, compute class' type if requested.
2235 type
check_field_constant (int index
, type
*class_type
= NULL
)
2237 _Jv_Utf8Const
*name
, *field_type
;
2238 type ct
= handle_field_or_method (index
,
2239 JV_CONSTANT_Fieldref
,
2240 &name
, &field_type
);
2243 if (field_type
->data
[0] == '[' || field_type
->data
[0] == 'L')
2244 return type (field_type
, this);
2245 return get_type_val_for_signature (field_type
->data
[0]);
2248 type
check_method_constant (int index
, bool is_interface
,
2249 _Jv_Utf8Const
**method_name
,
2250 _Jv_Utf8Const
**method_signature
)
2252 return handle_field_or_method (index
,
2254 ? JV_CONSTANT_InterfaceMethodref
2255 : JV_CONSTANT_Methodref
),
2256 method_name
, method_signature
);
2259 type
get_one_type (char *&p
)
2277 _Jv_Utf8Const
*name
= make_utf8_const (start
, p
- start
);
2278 return type (name
, this);
2281 // Casting to jchar here is ok since we are looking at an ASCII
2283 type_val rt
= get_type_val_for_signature (jchar (v
));
2285 if (arraycount
== 0)
2287 // Callers of this function eventually push their arguments on
2288 // the stack. So, promote them here.
2289 return type (rt
).promote ();
2292 jclass k
= construct_primitive_array_type (rt
);
2293 while (--arraycount
> 0)
2294 k
= _Jv_GetArrayClass (k
, NULL
);
2295 return type (k
, this);
2298 void compute_argument_types (_Jv_Utf8Const
*signature
,
2301 char *p
= signature
->data
;
2307 types
[i
++] = get_one_type (p
);
2310 type
compute_return_type (_Jv_Utf8Const
*signature
)
2312 char *p
= signature
->data
;
2316 return get_one_type (p
);
2319 void check_return_type (type onstack
)
2321 type rt
= compute_return_type (current_method
->self
->signature
);
2322 if (! rt
.compatible (onstack
, this))
2323 verify_fail ("incompatible return type");
2326 // Initialize the stack for the new method. Returns true if this
2327 // method is an instance initializer.
2328 bool initialize_stack ()
2331 bool is_init
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2333 bool is_clinit
= _Jv_equalUtf8Consts (current_method
->self
->name
,
2336 using namespace java::lang::reflect
;
2337 if (! Modifier::isStatic (current_method
->self
->accflags
))
2339 type
kurr (current_class
, this);
2342 kurr
.set_uninitialized (type::SELF
, this);
2346 verify_fail ("<clinit> method must be static");
2347 set_variable (0, kurr
);
2348 current_state
->set_this_type (kurr
);
2354 verify_fail ("<init> method must be non-static");
2357 // We have to handle wide arguments specially here.
2358 int arg_count
= _Jv_count_arguments (current_method
->self
->signature
);
2359 type arg_types
[arg_count
];
2360 compute_argument_types (current_method
->self
->signature
, arg_types
);
2361 for (int i
= 0; i
< arg_count
; ++i
)
2363 set_variable (var
, arg_types
[i
]);
2365 if (arg_types
[i
].iswide ())
2372 void verify_instructions_0 ()
2374 current_state
= new state (current_method
->max_stack
,
2375 current_method
->max_locals
);
2380 // True if we are verifying an instance initializer.
2381 bool this_is_init
= initialize_stack ();
2383 states
= (state
**) _Jv_Malloc (sizeof (state
*)
2384 * current_method
->code_length
);
2385 for (int i
= 0; i
< current_method
->code_length
; ++i
)
2388 next_verify_pc
= state::NO_NEXT
;
2392 // If the PC was invalidated, get a new one from the work list.
2393 if (PC
== state::NO_NEXT
)
2396 if (PC
== state::INVALID
)
2397 verify_fail ("can't happen: saw state::INVALID");
2398 if (PC
== state::NO_NEXT
)
2400 debug_print ("== State pop from pending list\n");
2401 // Set up the current state.
2402 current_state
->copy (states
[PC
], current_method
->max_stack
,
2403 current_method
->max_locals
);
2407 // Control can't fall off the end of the bytecode. We
2408 // only need to check this in the fall-through case,
2409 // because branch bounds are checked when they are
2411 if (PC
>= current_method
->code_length
)
2412 verify_fail ("fell off end");
2414 // We only have to do this checking in the situation where
2415 // control flow falls through from the previous
2416 // instruction. Otherwise merging is done at the time we
2418 if (states
[PC
] != NULL
)
2420 // We've already visited this instruction. So merge
2421 // the states together. If this yields no change then
2422 // we don't have to re-verify. However, if the new
2423 // state is an the result of an unmerged `ret', we
2424 // must continue through it.
2425 debug_print ("== Fall through merge\n");
2426 states
[PC
]->print ("Old", PC
, current_method
->max_stack
,
2427 current_method
->max_locals
);
2428 current_state
->print ("Cur", PC
, current_method
->max_stack
,
2429 current_method
->max_locals
);
2430 if (! current_state
->merge (states
[PC
], false,
2431 current_method
->max_locals
, this)
2432 && ! states
[PC
]->is_unmerged_ret_state (current_method
->max_locals
))
2434 debug_print ("== Fall through optimization\n");
2438 // Save a copy of it for later.
2439 states
[PC
]->copy (current_state
, current_method
->max_stack
,
2440 current_method
->max_locals
);
2441 current_state
->print ("New", PC
, current_method
->max_stack
,
2442 current_method
->max_locals
);
2446 // We only have to keep saved state at branch targets. If
2447 // we're at a branch target and the state here hasn't been set
2448 // yet, we set it now.
2449 if (states
[PC
] == NULL
&& (flags
[PC
] & FLAG_BRANCH_TARGET
))
2451 states
[PC
] = new state (current_state
, current_method
->max_stack
,
2452 current_method
->max_locals
);
2455 // Set this before handling exceptions so that debug output is
2459 // Update states for all active exception handlers. Ordinarily
2460 // there are not many exception handlers. So we simply run
2461 // through them all.
2462 for (int i
= 0; i
< current_method
->exc_count
; ++i
)
2464 if (PC
>= exception
[i
].start_pc
.i
&& PC
< exception
[i
].end_pc
.i
)
2466 type
handler (&java::lang::Throwable::class$
, this);
2467 if (exception
[i
].handler_type
.i
!= 0)
2468 handler
= check_class_constant (exception
[i
].handler_type
.i
);
2469 push_exception_jump (handler
, exception
[i
].handler_pc
.i
);
2473 current_state
->print (" ", PC
, current_method
->max_stack
,
2474 current_method
->max_locals
);
2475 java_opcode opcode
= (java_opcode
) bytecode
[PC
++];
2481 case op_aconst_null
:
2482 push_type (null_type
);
2492 push_type (int_type
);
2497 push_type (long_type
);
2503 push_type (float_type
);
2508 push_type (double_type
);
2513 push_type (int_type
);
2518 push_type (int_type
);
2522 push_type (check_constant (get_byte ()));
2525 push_type (check_constant (get_ushort ()));
2528 push_type (check_wide_constant (get_ushort ()));
2532 push_type (get_variable (get_byte (), int_type
));
2535 push_type (get_variable (get_byte (), long_type
));
2538 push_type (get_variable (get_byte (), float_type
));
2541 push_type (get_variable (get_byte (), double_type
));
2544 push_type (get_variable (get_byte (), reference_type
));
2551 push_type (get_variable (opcode
- op_iload_0
, int_type
));
2557 push_type (get_variable (opcode
- op_lload_0
, long_type
));
2563 push_type (get_variable (opcode
- op_fload_0
, float_type
));
2569 push_type (get_variable (opcode
- op_dload_0
, double_type
));
2575 push_type (get_variable (opcode
- op_aload_0
, reference_type
));
2578 pop_type (int_type
);
2579 push_type (require_array_type (pop_init_ref (reference_type
),
2583 pop_type (int_type
);
2584 push_type (require_array_type (pop_init_ref (reference_type
),
2588 pop_type (int_type
);
2589 push_type (require_array_type (pop_init_ref (reference_type
),
2593 pop_type (int_type
);
2594 push_type (require_array_type (pop_init_ref (reference_type
),
2598 pop_type (int_type
);
2599 push_type (require_array_type (pop_init_ref (reference_type
),
2603 pop_type (int_type
);
2604 require_array_type (pop_init_ref (reference_type
), byte_type
);
2605 push_type (int_type
);
2608 pop_type (int_type
);
2609 require_array_type (pop_init_ref (reference_type
), char_type
);
2610 push_type (int_type
);
2613 pop_type (int_type
);
2614 require_array_type (pop_init_ref (reference_type
), short_type
);
2615 push_type (int_type
);
2618 set_variable (get_byte (), pop_type (int_type
));
2621 set_variable (get_byte (), pop_type (long_type
));
2624 set_variable (get_byte (), pop_type (float_type
));
2627 set_variable (get_byte (), pop_type (double_type
));
2630 set_variable (get_byte (), pop_ref_or_return ());
2636 set_variable (opcode
- op_istore_0
, pop_type (int_type
));
2642 set_variable (opcode
- op_lstore_0
, pop_type (long_type
));
2648 set_variable (opcode
- op_fstore_0
, pop_type (float_type
));
2654 set_variable (opcode
- op_dstore_0
, pop_type (double_type
));
2660 set_variable (opcode
- op_astore_0
, pop_ref_or_return ());
2663 pop_type (int_type
);
2664 pop_type (int_type
);
2665 require_array_type (pop_init_ref (reference_type
), int_type
);
2668 pop_type (long_type
);
2669 pop_type (int_type
);
2670 require_array_type (pop_init_ref (reference_type
), long_type
);
2673 pop_type (float_type
);
2674 pop_type (int_type
);
2675 require_array_type (pop_init_ref (reference_type
), float_type
);
2678 pop_type (double_type
);
2679 pop_type (int_type
);
2680 require_array_type (pop_init_ref (reference_type
), double_type
);
2683 pop_type (reference_type
);
2684 pop_type (int_type
);
2685 require_array_type (pop_init_ref (reference_type
), reference_type
);
2688 pop_type (int_type
);
2689 pop_type (int_type
);
2690 require_array_type (pop_init_ref (reference_type
), byte_type
);
2693 pop_type (int_type
);
2694 pop_type (int_type
);
2695 require_array_type (pop_init_ref (reference_type
), char_type
);
2698 pop_type (int_type
);
2699 pop_type (int_type
);
2700 require_array_type (pop_init_ref (reference_type
), short_type
);
2707 type t
= pop_raw ();
2731 type t2
= pop_raw ();
2746 type t
= pop_raw ();
2761 type t1
= pop_raw ();
2778 type t1
= pop_raw ();
2781 type t2
= pop_raw ();
2799 type t3
= pop_raw ();
2837 pop_type (int_type
);
2838 push_type (pop_type (int_type
));
2848 pop_type (long_type
);
2849 push_type (pop_type (long_type
));
2854 pop_type (int_type
);
2855 push_type (pop_type (long_type
));
2862 pop_type (float_type
);
2863 push_type (pop_type (float_type
));
2870 pop_type (double_type
);
2871 push_type (pop_type (double_type
));
2877 push_type (pop_type (int_type
));
2880 push_type (pop_type (long_type
));
2883 push_type (pop_type (float_type
));
2886 push_type (pop_type (double_type
));
2889 get_variable (get_byte (), int_type
);
2893 pop_type (int_type
);
2894 push_type (long_type
);
2897 pop_type (int_type
);
2898 push_type (float_type
);
2901 pop_type (int_type
);
2902 push_type (double_type
);
2905 pop_type (long_type
);
2906 push_type (int_type
);
2909 pop_type (long_type
);
2910 push_type (float_type
);
2913 pop_type (long_type
);
2914 push_type (double_type
);
2917 pop_type (float_type
);
2918 push_type (int_type
);
2921 pop_type (float_type
);
2922 push_type (long_type
);
2925 pop_type (float_type
);
2926 push_type (double_type
);
2929 pop_type (double_type
);
2930 push_type (int_type
);
2933 pop_type (double_type
);
2934 push_type (long_type
);
2937 pop_type (double_type
);
2938 push_type (float_type
);
2941 pop_type (long_type
);
2942 pop_type (long_type
);
2943 push_type (int_type
);
2947 pop_type (float_type
);
2948 pop_type (float_type
);
2949 push_type (int_type
);
2953 pop_type (double_type
);
2954 pop_type (double_type
);
2955 push_type (int_type
);
2963 pop_type (int_type
);
2964 push_jump (get_short ());
2972 pop_type (int_type
);
2973 pop_type (int_type
);
2974 push_jump (get_short ());
2978 pop_type (reference_type
);
2979 pop_type (reference_type
);
2980 push_jump (get_short ());
2983 push_jump (get_short ());
2987 handle_jsr_insn (get_short ());
2990 handle_ret_insn (get_byte ());
2992 case op_tableswitch
:
2994 pop_type (int_type
);
2996 push_jump (get_int ());
2997 jint low
= get_int ();
2998 jint high
= get_int ();
2999 // Already checked LOW -vs- HIGH.
3000 for (int i
= low
; i
<= high
; ++i
)
3001 push_jump (get_int ());
3006 case op_lookupswitch
:
3008 pop_type (int_type
);
3010 push_jump (get_int ());
3011 jint npairs
= get_int ();
3012 // Already checked NPAIRS >= 0.
3014 for (int i
= 0; i
< npairs
; ++i
)
3016 jint key
= get_int ();
3017 if (i
> 0 && key
<= lastkey
)
3018 verify_fail ("lookupswitch pairs unsorted", start_PC
);
3020 push_jump (get_int ());
3026 check_return_type (pop_type (int_type
));
3030 check_return_type (pop_type (long_type
));
3034 check_return_type (pop_type (float_type
));
3038 check_return_type (pop_type (double_type
));
3042 check_return_type (pop_init_ref (reference_type
));
3046 // We only need to check this when the return type is
3047 // void, because all instance initializers return void.
3049 current_state
->check_this_initialized (this);
3050 check_return_type (void_type
);
3054 push_type (check_field_constant (get_ushort ()));
3057 pop_type (check_field_constant (get_ushort ()));
3062 type field
= check_field_constant (get_ushort (), &klass
);
3070 type field
= check_field_constant (get_ushort (), &klass
);
3073 // We have an obscure special case here: we can use
3074 // `putfield' on a field declared in this class, even if
3075 // `this' has not yet been initialized.
3076 if (! current_state
->this_type
.isinitialized ()
3077 && current_state
->this_type
.pc
== type::SELF
)
3078 klass
.set_uninitialized (type::SELF
, this);
3083 case op_invokevirtual
:
3084 case op_invokespecial
:
3085 case op_invokestatic
:
3086 case op_invokeinterface
:
3088 _Jv_Utf8Const
*method_name
, *method_signature
;
3090 = check_method_constant (get_ushort (),
3091 opcode
== op_invokeinterface
,
3094 // NARGS is only used when we're processing
3095 // invokeinterface. It is simplest for us to compute it
3096 // here and then verify it later.
3098 if (opcode
== op_invokeinterface
)
3100 nargs
= get_byte ();
3101 if (get_byte () != 0)
3102 verify_fail ("invokeinterface dummy byte is wrong");
3105 bool is_init
= false;
3106 if (_Jv_equalUtf8Consts (method_name
, gcj::init_name
))
3109 if (opcode
!= op_invokespecial
)
3110 verify_fail ("can't invoke <init>");
3112 else if (method_name
->data
[0] == '<')
3113 verify_fail ("can't invoke method starting with `<'");
3115 // Pop arguments and check types.
3116 int arg_count
= _Jv_count_arguments (method_signature
);
3117 type arg_types
[arg_count
];
3118 compute_argument_types (method_signature
, arg_types
);
3119 for (int i
= arg_count
- 1; i
>= 0; --i
)
3121 // This is only used for verifying the byte for
3123 nargs
-= arg_types
[i
].depth ();
3124 pop_init_ref (arg_types
[i
]);
3127 if (opcode
== op_invokeinterface
3129 verify_fail ("wrong argument count for invokeinterface");
3131 if (opcode
!= op_invokestatic
)
3133 type t
= class_type
;
3136 // In this case the PC doesn't matter.
3137 t
.set_uninitialized (type::UNINIT
, this);
3138 // FIXME: check to make sure that the <init>
3139 // call is to the right class.
3140 // It must either be super or an exact class
3143 type raw
= pop_raw ();
3144 if (! t
.compatible (raw
, this))
3145 verify_fail ("incompatible type on stack");
3148 current_state
->set_initialized (raw
.get_pc (),
3149 current_method
->max_locals
);
3152 type rt
= compute_return_type (method_signature
);
3160 type t
= check_class_constant (get_ushort ());
3161 if (t
.isarray () || t
.isinterface (this) || t
.isabstract (this))
3162 verify_fail ("type is array, interface, or abstract");
3163 t
.set_uninitialized (start_PC
, this);
3170 int atype
= get_byte ();
3171 // We intentionally have chosen constants to make this
3173 if (atype
< boolean_type
|| atype
> long_type
)
3174 verify_fail ("type not primitive", start_PC
);
3175 pop_type (int_type
);
3176 type
t (construct_primitive_array_type (type_val (atype
)), this);
3181 pop_type (int_type
);
3182 push_type (check_class_constant (get_ushort ()).to_array (this));
3184 case op_arraylength
:
3186 type t
= pop_init_ref (reference_type
);
3187 if (! t
.isarray () && ! t
.isnull ())
3188 verify_fail ("array type expected");
3189 push_type (int_type
);
3193 pop_type (type (&java::lang::Throwable::class$
, this));
3197 pop_init_ref (reference_type
);
3198 push_type (check_class_constant (get_ushort ()));
3201 pop_init_ref (reference_type
);
3202 check_class_constant (get_ushort ());
3203 push_type (int_type
);
3205 case op_monitorenter
:
3206 pop_init_ref (reference_type
);
3208 case op_monitorexit
:
3209 pop_init_ref (reference_type
);
3213 switch (get_byte ())
3216 push_type (get_variable (get_ushort (), int_type
));
3219 push_type (get_variable (get_ushort (), long_type
));
3222 push_type (get_variable (get_ushort (), float_type
));
3225 push_type (get_variable (get_ushort (), double_type
));
3228 push_type (get_variable (get_ushort (), reference_type
));
3231 set_variable (get_ushort (), pop_type (int_type
));
3234 set_variable (get_ushort (), pop_type (long_type
));
3237 set_variable (get_ushort (), pop_type (float_type
));
3240 set_variable (get_ushort (), pop_type (double_type
));
3243 set_variable (get_ushort (), pop_init_ref (reference_type
));
3246 handle_ret_insn (get_short ());
3249 get_variable (get_ushort (), int_type
);
3253 verify_fail ("unrecognized wide instruction", start_PC
);
3257 case op_multianewarray
:
3259 type atype
= check_class_constant (get_ushort ());
3260 int dim
= get_byte ();
3262 verify_fail ("too few dimensions to multianewarray", start_PC
);
3263 atype
.verify_dimensions (dim
, this);
3264 for (int i
= 0; i
< dim
; ++i
)
3265 pop_type (int_type
);
3271 pop_type (reference_type
);
3272 push_jump (get_short ());
3275 push_jump (get_int ());
3279 handle_jsr_insn (get_int ());
3282 // These are unused here, but we call them out explicitly
3283 // so that -Wswitch-enum doesn't complain.
3289 case op_putstatic_1
:
3290 case op_putstatic_2
:
3291 case op_putstatic_4
:
3292 case op_putstatic_8
:
3293 case op_putstatic_a
:
3295 case op_getfield_2s
:
3296 case op_getfield_2u
:
3300 case op_getstatic_1
:
3301 case op_getstatic_2s
:
3302 case op_getstatic_2u
:
3303 case op_getstatic_4
:
3304 case op_getstatic_8
:
3305 case op_getstatic_a
:
3307 // Unrecognized opcode.
3308 verify_fail ("unrecognized instruction in verify_instructions_0",
3316 void verify_instructions ()
3319 verify_instructions_0 ();
3322 _Jv_BytecodeVerifier (_Jv_InterpMethod
*m
)
3324 // We just print the text as utf-8. This is just for debugging
3326 debug_print ("--------------------------------\n");
3327 debug_print ("-- Verifying method `%s'\n", m
->self
->name
->data
);
3330 bytecode
= m
->bytecode ();
3331 exception
= m
->exceptions ();
3332 current_class
= m
->defining_class
;
3339 entry_points
= NULL
;
3342 ~_Jv_BytecodeVerifier ()
3351 for (int i
= 0; i
< current_method
->code_length
; ++i
)
3353 if (jsr_ptrs
[i
] != NULL
)
3355 subr_info
*info
= jsr_ptrs
[i
];
3356 while (info
!= NULL
)
3358 subr_info
*next
= info
->next
;
3364 _Jv_Free (jsr_ptrs
);
3367 while (utf8_list
!= NULL
)
3369 linked_utf8
*n
= utf8_list
->next
;
3370 _Jv_Free (utf8_list
->val
);
3371 _Jv_Free (utf8_list
);
3375 while (entry_points
!= NULL
)
3377 subr_entry_info
*next
= entry_points
->next
;
3378 _Jv_Free (entry_points
);
3379 entry_points
= next
;
3382 while (isect_list
!= NULL
)
3384 ref_intersection
*next
= isect_list
->alloc_next
;
3392 _Jv_VerifyMethod (_Jv_InterpMethod
*meth
)
3394 _Jv_BytecodeVerifier
v (meth
);
3395 v
.verify_instructions ();
3397 #endif /* INTERPRETER */