1 /*
  2  * Copyright (c) 2020, 2023, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 #ifndef SHARE_CDS_ARCHIVEBUILDER_HPP
 26 #define SHARE_CDS_ARCHIVEBUILDER_HPP
 27 
 28 #include "cds/archiveUtils.hpp"
 29 #include "cds/dumpAllocStats.hpp"
 30 #include "memory/metaspaceClosure.hpp"
 31 #include "oops/array.hpp"
 32 #include "oops/klass.hpp"
 33 #include "runtime/os.hpp"
 34 #include "utilities/bitMap.hpp"
 35 #include "utilities/growableArray.hpp"
 36 #include "utilities/resizeableResourceHash.hpp"
 37 #include "utilities/resourceHash.hpp"
 38 
 39 class ArchiveHeapInfo;
 40 class CHeapBitMap;
 41 class FileMapInfo;
 42 class Klass;
 43 class MemRegion;
 44 class Symbol;
 45 
 46 // Metaspace::allocate() requires that all blocks must be aligned with KlassAlignmentInBytes.
 47 // We enforce the same alignment rule in blocks allocated from the shared space.
 48 const int SharedSpaceObjectAlignment = KlassAlignmentInBytes;
 49 
 50 // Overview of CDS archive creation (for both static and dynamic dump):
 51 //
 52 // [1] Load all classes (static dump: from the classlist, dynamic dump: as part of app execution)
 53 // [2] Allocate "output buffer"
 54 // [3] Copy contents of the 2 "core" regions (rw/ro) into the output buffer.
 55 //       - allocate the cpp vtables in rw (static dump only)
 56 //       - memcpy the MetaspaceObjs into rw/ro:
 57 //         dump_rw_region();
 58 //         dump_ro_region();
 59 //       - fix all the pointers in the MetaspaceObjs to point to the copies
 60 //         relocate_metaspaceobj_embedded_pointers()
 61 // [4] Copy symbol table, dictionary, etc, into the ro region
 62 // [5] Relocate all the pointers in rw/ro, so that the archive can be mapped to
 63 //     the "requested" location without runtime relocation. See relocate_to_requested()
 64 //
 65 // "source" vs "buffered" vs "requested"
 66 //
 67 // The ArchiveBuilder deals with three types of addresses.
 68 //
 69 // "source":    These are the addresses of objects created in step [1] above. They are the actual
 70 //              InstanceKlass*, Method*, etc, of the Java classes that are loaded for executing
 71 //              Java bytecodes in the JVM process that's dumping the CDS archive.
 72 //
 73 //              It may be necessary to contiue Java execution after ArchiveBuilder is finished.
 74 //              Therefore, we don't modify any of the "source" objects.
 75 //
 76 // "buffered":  The "source" objects that are deemed archivable are copied into a temporary buffer.
 77 //              Objects in the buffer are modified in steps [2, 3, 4] (e.g., unshareable info is
 78 //              removed, pointers are relocated, etc) to prepare them to be loaded at runtime.
 79 //
 80 // "requested": These are the addreses where the "buffered" objects should be loaded at runtime.
 81 //              When the "buffered" objects are written into the archive file, their addresses
 82 //              are adjusted in step [5] such that the lowest of these objects would be mapped
 83 //              at SharedBaseAddress.
 84 //
 85 // Translation between "source" and "buffered" addresses is done with two hashtables:
 86 //     _src_obj_table          : "source"   -> "buffered"
 87 //     _buffered_to_src_table  : "buffered" -> "source"
 88 //
 89 // Translation between "buffered" and "requested" addresses is done with a simple shift:
 90 //    buffered_address + _buffer_to_requested_delta == requested_address
 91 //
 92 class ArchiveBuilder : public StackObj {
 93 protected:
 94   DumpRegion* _current_dump_space;
 95   address _buffer_bottom;                      // for writing the contents of rw/ro regions
 96   address _last_verified_top;
 97   int _num_dump_regions_used;
 98   size_t _other_region_used_bytes;
 99 
100   // These are the addresses where we will request the static and dynamic archives to be
101   // mapped at run time. If the request fails (due to ASLR), we will map the archives at
102   // os-selected addresses.
103   address _requested_static_archive_bottom;     // This is determined solely by the value of
104                                                 // SharedBaseAddress during -Xshare:dump.
105   address _requested_static_archive_top;
106   address _requested_dynamic_archive_bottom;    // Used only during dynamic dump. It's placed
107                                                 // immediately above _requested_static_archive_top.
108   address _requested_dynamic_archive_top;
109 
110   // (Used only during dynamic dump) where the static archive is actually mapped. This
111   // may be different than _requested_static_archive_{bottom,top} due to ASLR
112   address _mapped_static_archive_bottom;
113   address _mapped_static_archive_top;
114 
115   intx _buffer_to_requested_delta;
116 
117   DumpRegion* current_dump_space() const {  return _current_dump_space;  }
118 
119 public:
120   enum FollowMode {
121     make_a_copy, point_to_it, set_to_null
122   };
123 
124 private:
125   class SourceObjInfo {
126     uintx _ptrmap_start;     // The bit-offset of the start of this object (inclusive)
127     uintx _ptrmap_end;       // The bit-offset of the end   of this object (exclusive)
128     bool _read_only;
129     FollowMode _follow_mode;
130     int _size_in_bytes;
131     MetaspaceObj::Type _msotype;
132     address _source_addr;    // The source object to be copied.
133     address _buffered_addr;  // The copy of this object insider the buffer.
134   public:
135     SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
136       _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _follow_mode(follow_mode),
137       _size_in_bytes(ref->size() * BytesPerWord), _msotype(ref->msotype()),
138       _source_addr(ref->obj()) {
139       if (follow_mode == point_to_it) {
140         _buffered_addr = ref->obj();
141       } else {
142         _buffered_addr = nullptr;
143       }
144     }
145     SourceObjInfo(address src, address buf) {
146       _source_addr = src;
147       _buffered_addr = buf;
148     }
149 
150     // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
151     //   src = address of a Method or InstanceKlass that has been regenerated.
152     //   renegerated_obj_info = info for the regenerated version of src.
153     SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
154       _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
155       _follow_mode(renegerated_obj_info->_follow_mode),
156       _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
157       _source_addr(src),  _buffered_addr(renegerated_obj_info->_buffered_addr) {}
158 
159     bool should_copy() const { return _follow_mode == make_a_copy; }
160     void set_buffered_addr(address addr)  {
161       assert(should_copy(), "must be");
162       assert(_buffered_addr == nullptr, "cannot be copied twice");
163       assert(addr != nullptr, "must be a valid copy");
164       _buffered_addr = addr;
165     }
166     void set_ptrmap_start(uintx v) { _ptrmap_start = v;    }
167     void set_ptrmap_end(uintx v)   { _ptrmap_end = v;      }
168     uintx ptrmap_start()  const    { return _ptrmap_start; } // inclusive
169     uintx ptrmap_end()    const    { return _ptrmap_end;   } // exclusive
170     bool read_only()      const    { return _read_only;    }
171     int size_in_bytes()   const    { return _size_in_bytes; }
172     address source_addr() const    { return _source_addr; }
173     address buffered_addr() const  {
174       if (_follow_mode != set_to_null) {
175         assert(_buffered_addr != nullptr, "must be initialized");
176       }
177       return _buffered_addr;
178     }
179     MetaspaceObj::Type msotype() const { return _msotype; }
180   };
181 
182   class SourceObjList {
183     uintx _total_bytes;
184     GrowableArray<SourceObjInfo*>* _objs;     // Source objects to be archived
185     CHeapBitMap _ptrmap;                      // Marks the addresses of the pointer fields
186                                               // in the source objects
187   public:
188     SourceObjList();
189     ~SourceObjList();
190 
191     GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
192 
193     void append(SourceObjInfo* src_info);
194     void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
195     void relocate(int i, ArchiveBuilder* builder);
196 
197     // convenience accessor
198     SourceObjInfo* at(int i) const { return objs()->at(i); }
199   };
200 
201   class CDSMapLogger;
202 
203   static const int INITIAL_TABLE_SIZE = 15889;
204   static const int MAX_TABLE_SIZE     = 1000000;
205 
206   ReservedSpace _shared_rs;
207   VirtualSpace _shared_vs;
208 
209   DumpRegion _rw_region;
210   DumpRegion _ro_region;
211   DumpRegion _cc_region;
212   CHeapBitMap _ptrmap;    // bitmap used by ArchivePtrMarker
213 
214   SourceObjList _rw_src_objs;                 // objs to put in rw region
215   SourceObjList _ro_src_objs;                 // objs to put in ro region
216   ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
217   ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
218   GrowableArray<Klass*>* _klasses;
219   GrowableArray<Symbol*>* _symbols;
220 
221   // statistics
222   DumpAllocStats _alloc_stats;
223   size_t _total_heap_region_size;
224 
225   void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
226   void print_bitmap_region_stats(size_t size, size_t total_size);
227   void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
228 
229   // For global access.
230   static ArchiveBuilder* _current;
231 
232 public:
233   // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
234   // These are usually for misc tables that are allocated in the RO space.
235   class OtherROAllocMark {
236     char* _oldtop;
237   public:
238     OtherROAllocMark() {
239       _oldtop = _current->_ro_region.top();
240     }
241     ~OtherROAllocMark();
242   };
243 
244 private:
245   FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
246 
247   void iterate_sorted_roots(MetaspaceClosure* it);
248   void sort_klasses();
249   static int compare_symbols_by_address(Symbol** a, Symbol** b);
250   static int compare_klass_by_name(Klass** a, Klass** b);
251   void update_hidden_class_loader_type(InstanceKlass* ik) NOT_CDS_JAVA_HEAP_RETURN;
252 
253   void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
254   void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
255 
256   void relocate_embedded_pointers(SourceObjList* src_objs);
257 
258   bool is_excluded(Klass* k);
259   void clean_up_src_obj_table();
260 
261 protected:
262   virtual void iterate_roots(MetaspaceClosure* it) = 0;
263 
264   // Conservative estimate for number of bytes needed for:
265   size_t _estimated_metaspaceobj_bytes;   // all archived MetaspaceObj's.
266   size_t _estimated_hashtable_bytes;     // symbol table and dictionaries
267 
268   static const int _total_dump_regions = 2;
269 
270   size_t estimate_archive_size();
271 
272   void start_dump_space(DumpRegion* next);
273   void verify_estimate_size(size_t estimate, const char* which);
274 
275 public:
276   address reserve_buffer();
277 
278   address buffer_bottom()                    const { return _buffer_bottom;                       }
279   address buffer_top()                       const { return (address)current_dump_space()->top(); }
280   address requested_static_archive_bottom()  const { return  _requested_static_archive_bottom;    }
281   address mapped_static_archive_bottom()     const { return  _mapped_static_archive_bottom;       }
282   intx buffer_to_requested_delta()           const { return _buffer_to_requested_delta;           }
283 
284   bool is_in_buffer_space(address p) const {
285     if (current_dump_space() == nullptr) {
286       return false;
287     }
288     return (buffer_bottom() <= p && p < buffer_top());
289   }
290 
291   template <typename T> bool is_in_requested_static_archive(T p) const {
292     return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
293   }
294 
295   template <typename T> bool is_in_mapped_static_archive(T p) const {
296     return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
297   }
298 
299   template <typename T> bool is_in_buffer_space(T obj) const {
300     return is_in_buffer_space(address(obj));
301   }
302 
303   template <typename T> T to_requested(T obj) const {
304     assert(is_in_buffer_space(obj), "must be");
305     return (T)(address(obj) + _buffer_to_requested_delta);
306   }
307 
308   static intx get_buffer_to_requested_delta() {
309     return current()->buffer_to_requested_delta();
310   }
311 
312   inline static u4 to_offset_u4(uintx offset) {
313     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
314     return (u4)offset;
315   }
316 
317 public:
318   static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
319 
320   // The address p points to an object inside the output buffer. When the archive is mapped
321   // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
322   uintx buffer_to_offset(address p) const;
323 
324   // Same as buffer_to_offset, except that the address p points to either (a) an object
325   // inside the output buffer, or (b), an object in the currently mapped static archive.
326   uintx any_to_offset(address p) const;
327 
328   template <typename T>
329   u4 buffer_to_offset_u4(T p) const {
330     uintx offset = buffer_to_offset((address)p);
331     return to_offset_u4(offset);
332   }
333 
334   template <typename T>
335   u4 any_to_offset_u4(T p) const {
336     uintx offset = any_to_offset((address)p);
337     return to_offset_u4(offset);
338   }
339 
340 public:
341   ArchiveBuilder();
342   ~ArchiveBuilder();
343 
344   void gather_klasses_and_symbols();
345   void gather_source_objs();
346   bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
347   bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
348   void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
349   static void serialize_dynamic_archivable_items(SerializeClosure* soc);
350 
351   DumpRegion* rw_region() { return &_rw_region; }
352   DumpRegion* ro_region() { return &_ro_region; }
353   DumpRegion* cc_region() { return &_cc_region; }
354 
355   void start_cc_region();
356   void end_cc_region();
357 
358   static char* rw_region_alloc(size_t num_bytes) {
359     return current()->rw_region()->allocate(num_bytes);
360   }
361   static char* ro_region_alloc(size_t num_bytes) {
362     return current()->ro_region()->allocate(num_bytes);
363   }
364   static char* cc_region_alloc(size_t num_bytes) {
365     return current()->cc_region()->allocate(num_bytes);
366   }
367 
368   template <typename T>
369   static Array<T>* new_ro_array(int length) {
370     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
371     Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
372     array->initialize(length);
373     return array;
374   }
375 
376   template <typename T>
377   static Array<T>* new_rw_array(int length) {
378     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
379     Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
380     array->initialize(length);
381     return array;
382   }
383 
384   template <typename T>
385   static size_t ro_array_bytesize(int length) {
386     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
387     return align_up(byte_size, SharedSpaceObjectAlignment);
388   }
389 
390   char* ro_strdup(const char* s);
391 
392   void dump_rw_metadata();
393   void dump_ro_metadata();
394   void relocate_metaspaceobj_embedded_pointers();
395   void record_regenerated_object(address orig_src_obj, address regen_src_obj);
396   void make_klasses_shareable();
397   void relocate_to_requested();
398   void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
399   void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
400                     bool read_only,  bool allow_exec);
401 
402   void write_pointer_in_buffer(address* ptr_location, address src_addr);
403   template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
404     write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
405   }
406 
407   void mark_and_relocate_to_buffered_addr(address* ptr_location);
408   template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
409     mark_and_relocate_to_buffered_addr((address*)ptr_location);
410   }
411 
412   bool has_been_archived(address src_addr) const;
413   address get_buffered_addr(address src_addr) const;
414   template <typename T> T get_buffered_addr(T src_addr) const {
415     return (T)get_buffered_addr((address)src_addr);
416   }
417 
418   address get_source_addr(address buffered_addr) const;
419   template <typename T> T get_source_addr(T buffered_addr) const {
420     return (T)get_source_addr((address)buffered_addr);
421   }
422 
423   // All klasses and symbols that will be copied into the archive
424   GrowableArray<Klass*>*  klasses() const { return _klasses; }
425   GrowableArray<Symbol*>* symbols() const { return _symbols; }
426 
427   static bool is_active() {
428     CDS_ONLY(return (_current != nullptr));
429     NOT_CDS(return false;)
430   }
431 
432   static ArchiveBuilder* current() {
433     assert(_current != nullptr, "ArchiveBuilder must be active");
434     return _current;
435   }
436 
437   static DumpAllocStats* alloc_stats() {
438     return &(current()->_alloc_stats);
439   }
440 
441   static CompactHashtableStats* symbol_stats() {
442     return alloc_stats()->symbol_stats();
443   }
444 
445   static CompactHashtableStats* string_stats() {
446     return alloc_stats()->string_stats();
447   }
448 
449   narrowKlass get_requested_narrow_klass(Klass* k);
450 
451   static Klass* get_buffered_klass(Klass* src_klass) {
452     Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
453     assert(klass != nullptr && klass->is_klass(), "must be");
454     return klass;
455   }
456 
457   static Symbol* get_buffered_symbol(Symbol* src_symbol) {
458     return (Symbol*)current()->get_buffered_addr((address)src_symbol);
459   }
460 
461   void print_stats();
462   void report_out_of_space(const char* name, size_t needed_bytes);
463 };
464 
465 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP