1 /*
  2  * Copyright (c) 2020, 2024, 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_region;
 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_region() const {  return _current_dump_region;  }
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     bool _has_embedded_pointer;
130     FollowMode _follow_mode;
131     int _size_in_bytes;
132     int _id; // Each object has a unique serial ID, starting from zero. The ID is assigned
133              // when the object is added into _source_objs.
134     MetaspaceObj::Type _msotype;
135     address _source_addr;    // The source object to be copied.
136     address _buffered_addr;  // The copy of this object insider the buffer.
137   public:
138     SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
139       _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _has_embedded_pointer(false), _follow_mode(follow_mode),
140       _size_in_bytes(ref->size() * BytesPerWord), _id(0), _msotype(ref->msotype()),
141       _source_addr(ref->obj()) {
142       if (follow_mode == point_to_it) {
143         _buffered_addr = ref->obj();
144       } else {
145         _buffered_addr = nullptr;
146       }
147     }
148 
149     // This constructor is only used for regenerated objects (created by LambdaFormInvokers, etc).
150     //   src = address of a Method or InstanceKlass that has been regenerated.
151     //   renegerated_obj_info = info for the regenerated version of src.
152     SourceObjInfo(address src, SourceObjInfo* renegerated_obj_info) :
153       _ptrmap_start(0), _ptrmap_end(0), _read_only(false),
154       _follow_mode(renegerated_obj_info->_follow_mode),
155       _size_in_bytes(0), _msotype(renegerated_obj_info->_msotype),
156       _source_addr(src),  _buffered_addr(renegerated_obj_info->_buffered_addr) {}
157 
158     bool should_copy() const { return _follow_mode == make_a_copy; }
159     void set_buffered_addr(address addr)  {
160       assert(should_copy(), "must be");
161       assert(_buffered_addr == nullptr, "cannot be copied twice");
162       assert(addr != nullptr, "must be a valid copy");
163       _buffered_addr = addr;
164     }
165     void set_ptrmap_start(uintx v) { _ptrmap_start = v;    }
166     void set_ptrmap_end(uintx v)   { _ptrmap_end = v;      }
167     uintx ptrmap_start()  const    { return _ptrmap_start; } // inclusive
168     uintx ptrmap_end()    const    { return _ptrmap_end;   } // exclusive
169     bool read_only()      const    { return _read_only;    }
170     bool has_embedded_pointer() const { return _has_embedded_pointer; }
171     void set_has_embedded_pointer()   { _has_embedded_pointer = true; }
172     int size_in_bytes()   const    { return _size_in_bytes; }
173     int id()              const    { return _id; }
174     void set_id(int i)             { _id = i; }
175     address source_addr() const    { return _source_addr; }
176     address buffered_addr() const  {
177       if (_follow_mode != set_to_null) {
178         assert(_buffered_addr != nullptr, "must be initialized");
179       }
180       return _buffered_addr;
181     }
182     MetaspaceObj::Type msotype() const { return _msotype; }
183   };
184 
185   class SourceObjList {
186     uintx _total_bytes;
187     GrowableArray<SourceObjInfo*>* _objs;     // Source objects to be archived
188     CHeapBitMap _ptrmap;                      // Marks the addresses of the pointer fields
189                                               // in the source objects
190   public:
191     SourceObjList();
192     ~SourceObjList();
193 
194     GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
195 
196     void append(SourceObjInfo* src_info);
197     void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
198     void relocate(int i, ArchiveBuilder* builder);
199 
200     // convenience accessor
201     SourceObjInfo* at(int i) const { return objs()->at(i); }
202   };
203 
204   class CDSMapLogger;
205 
206   static const int INITIAL_TABLE_SIZE = 15889;
207   static const int MAX_TABLE_SIZE     = 1000000;
208 
209   ReservedSpace _shared_rs;
210   VirtualSpace _shared_vs;
211 
212   DumpRegion _rw_region;
213   DumpRegion _ro_region;
214 
215   // Combined bitmap to track pointers in both RW and RO regions. This is updated
216   // as objects are copied into RW and RO.
217   CHeapBitMap _ptrmap;
218 
219   // _ptrmap is split into these two bitmaps which are written into the archive.
220   CHeapBitMap _rw_ptrmap;   // marks pointers in the RW region
221   CHeapBitMap _ro_ptrmap;   // marks pointers in the RO region
222 
223   SourceObjList _rw_src_objs;                 // objs to put in rw region
224   SourceObjList _ro_src_objs;                 // objs to put in ro region
225   ResizeableResourceHashtable<address, SourceObjInfo, AnyObj::C_HEAP, mtClassShared> _src_obj_table;
226   ResizeableResourceHashtable<address, address, AnyObj::C_HEAP, mtClassShared> _buffered_to_src_table;
227   GrowableArray<Klass*>* _klasses;
228   GrowableArray<Symbol*>* _symbols;
229   unsigned int _entropy_seed;
230 
231   // statistics
232   DumpAllocStats _alloc_stats;
233   size_t _total_heap_region_size;
234 
235   void print_region_stats(FileMapInfo *map_info, ArchiveHeapInfo* heap_info);
236   void print_bitmap_region_stats(size_t size, size_t total_size);
237   void print_heap_region_stats(ArchiveHeapInfo* heap_info, size_t total_size);
238 
239   // For global access.
240   static ArchiveBuilder* _current;
241 
242 public:
243   // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
244   // These are usually for misc tables that are allocated in the RO space.
245   class OtherROAllocMark {
246     char* _oldtop;
247   public:
248     OtherROAllocMark() {
249       _oldtop = _current->_ro_region.top();
250     }
251     ~OtherROAllocMark();
252   };
253 
254 private:
255   FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
256 
257   void iterate_sorted_roots(MetaspaceClosure* it);
258   void sort_klasses();
259   static int compare_symbols_by_address(Symbol** a, Symbol** b);
260   static int compare_klass_by_name(Klass** a, Klass** b);
261 
262   void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
263   void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
264 
265   void relocate_embedded_pointers(SourceObjList* src_objs);
266 
267   bool is_excluded(Klass* k);
268   void clean_up_src_obj_table();
269 
270 protected:
271   virtual void iterate_roots(MetaspaceClosure* it) = 0;
272 
273   // Conservative estimate for number of bytes needed for:
274   size_t _estimated_metaspaceobj_bytes;   // all archived MetaspaceObj's.
275   size_t _estimated_hashtable_bytes;     // symbol table and dictionaries
276 
277   static const int _total_dump_regions = 2;
278 
279   size_t estimate_archive_size();
280 
281   void start_dump_region(DumpRegion* next);
282   void verify_estimate_size(size_t estimate, const char* which);
283 
284 public:
285   address reserve_buffer();
286 
287   address buffer_bottom()                    const { return _buffer_bottom;                        }
288   address buffer_top()                       const { return (address)current_dump_region()->top(); }
289   address requested_static_archive_bottom()  const { return  _requested_static_archive_bottom;     }
290   address mapped_static_archive_bottom()     const { return  _mapped_static_archive_bottom;        }
291   intx buffer_to_requested_delta()           const { return _buffer_to_requested_delta;            }
292 
293   bool is_in_buffer_space(address p) const {
294     return (buffer_bottom() <= p && p < buffer_top());
295   }
296 
297   template <typename T> bool is_in_requested_static_archive(T p) const {
298     return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
299   }
300 
301   template <typename T> bool is_in_mapped_static_archive(T p) const {
302     return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
303   }
304 
305   template <typename T> bool is_in_buffer_space(T obj) const {
306     return is_in_buffer_space(address(obj));
307   }
308 
309   template <typename T> T to_requested(T obj) const {
310     assert(is_in_buffer_space(obj), "must be");
311     return (T)(address(obj) + _buffer_to_requested_delta);
312   }
313 
314   static intx get_buffer_to_requested_delta() {
315     return current()->buffer_to_requested_delta();
316   }
317 
318   inline static u4 to_offset_u4(uintx offset) {
319     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
320     return (u4)offset;
321   }
322 
323 public:
324   static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
325 
326   // The address p points to an object inside the output buffer. When the archive is mapped
327   // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
328   uintx buffer_to_offset(address p) const;
329 
330   // Same as buffer_to_offset, except that the address p points to either (a) an object
331   // inside the output buffer, or (b), an object in the currently mapped static archive.
332   uintx any_to_offset(address p) const;
333 
334   template <typename T>
335   u4 buffer_to_offset_u4(T p) const {
336     uintx offset = buffer_to_offset((address)p);
337     return to_offset_u4(offset);
338   }
339 
340   template <typename T>
341   u4 any_to_offset_u4(T p) const {
342     uintx offset = any_to_offset((address)p);
343     return to_offset_u4(offset);
344   }
345 
346   static void assert_is_vm_thread() PRODUCT_RETURN;
347 
348 public:
349   ArchiveBuilder();
350   ~ArchiveBuilder();
351 
352   int entropy();
353   void gather_klasses_and_symbols();
354   void gather_source_objs();
355   bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
356   bool gather_one_source_obj(MetaspaceClosure::Ref* ref, bool read_only);
357   void remember_embedded_pointer_in_enclosing_obj(MetaspaceClosure::Ref* ref);
358   static void serialize_dynamic_archivable_items(SerializeClosure* soc);
359 
360   DumpRegion* rw_region() { return &_rw_region; }
361   DumpRegion* ro_region() { return &_ro_region; }
362 
363   static char* rw_region_alloc(size_t num_bytes) {
364     return current()->rw_region()->allocate(num_bytes);
365   }
366   static char* ro_region_alloc(size_t num_bytes) {
367     return current()->ro_region()->allocate(num_bytes);
368   }
369 
370   template <typename T>
371   static Array<T>* new_ro_array(int length) {
372     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
373     Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
374     array->initialize(length);
375     return array;
376   }
377 
378   template <typename T>
379   static Array<T>* new_rw_array(int length) {
380     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
381     Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
382     array->initialize(length);
383     return array;
384   }
385 
386   template <typename T>
387   static size_t ro_array_bytesize(int length) {
388     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
389     return align_up(byte_size, SharedSpaceObjectAlignment);
390   }
391 
392   char* ro_strdup(const char* s);
393 
394   static int compare_src_objs(SourceObjInfo** a, SourceObjInfo** b);
395   void sort_metadata_objs();
396   void dump_rw_metadata();
397   void dump_ro_metadata();
398   void relocate_metaspaceobj_embedded_pointers();
399   void record_regenerated_object(address orig_src_obj, address regen_src_obj);
400   void make_klasses_shareable();
401   void relocate_to_requested();
402   void write_archive(FileMapInfo* mapinfo, ArchiveHeapInfo* heap_info);
403   void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
404                     bool read_only,  bool allow_exec);
405 
406   void write_pointer_in_buffer(address* ptr_location, address src_addr);
407   template <typename T> void write_pointer_in_buffer(T* ptr_location, T src_addr) {
408     write_pointer_in_buffer((address*)ptr_location, (address)src_addr);
409   }
410 
411   void mark_and_relocate_to_buffered_addr(address* ptr_location);
412   template <typename T> void mark_and_relocate_to_buffered_addr(T ptr_location) {
413     mark_and_relocate_to_buffered_addr((address*)ptr_location);
414   }
415 
416   address get_buffered_addr(address src_addr) const;
417   template <typename T> T get_buffered_addr(T src_addr) const {
418     return (T)get_buffered_addr((address)src_addr);
419   }
420 
421   address get_source_addr(address buffered_addr) const;
422   template <typename T> T get_source_addr(T buffered_addr) const {
423     return (T)get_source_addr((address)buffered_addr);
424   }
425 
426   // All klasses and symbols that will be copied into the archive
427   GrowableArray<Klass*>*  klasses() const { return _klasses; }
428   GrowableArray<Symbol*>* symbols() const { return _symbols; }
429 
430   static bool is_active() {
431     return (_current != nullptr);
432   }
433 
434   static ArchiveBuilder* current() {
435     assert_is_vm_thread();
436     assert(_current != nullptr, "ArchiveBuilder must be active");
437     return _current;
438   }
439 
440   static DumpAllocStats* alloc_stats() {
441     return &(current()->_alloc_stats);
442   }
443 
444   static CompactHashtableStats* symbol_stats() {
445     return alloc_stats()->symbol_stats();
446   }
447 
448   static CompactHashtableStats* string_stats() {
449     return alloc_stats()->string_stats();
450   }
451 
452   narrowKlass get_requested_narrow_klass(Klass* k);
453 
454   static Klass* get_buffered_klass(Klass* src_klass) {
455     Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
456     assert(klass != nullptr && klass->is_klass(), "must be");
457     return klass;
458   }
459 
460   static Symbol* get_buffered_symbol(Symbol* src_symbol) {
461     return (Symbol*)current()->get_buffered_addr((address)src_symbol);
462   }
463 
464   void print_stats();
465   void report_out_of_space(const char* name, size_t needed_bytes);
466 };
467 
468 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP