1 /*
  2  * Copyright (c) 2020, 2022, 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 struct ArchiveHeapBitmapInfo;
 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 SpecialRefInfo {
126     // We have a "special pointer" of the given _type at _field_offset of _src_obj.
127     // See MetaspaceClosure::push_special().
128     MetaspaceClosure::SpecialRef _type;
129     address _src_obj;
130     size_t _field_offset;
131     DEBUG_ONLY(size_t _src_obj_size_in_bytes;)
132 
133   public:
134     SpecialRefInfo() {}
135     SpecialRefInfo(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset, size_t src_obj_size_in_bytes)
136       : _type(type), _src_obj(src_obj), _field_offset(field_offset) {
137       DEBUG_ONLY(_src_obj_size_in_bytes = src_obj_size_in_bytes);
138     }
139 
140     MetaspaceClosure::SpecialRef type() const { return _type;         }
141     address src_obj()                   const { return _src_obj;      }
142     size_t field_offset()               const { return _field_offset; }
143 
144     DEBUG_ONLY(size_t src_obj_size_in_bytes() const { return _src_obj_size_in_bytes; })
145   };
146 
147   class SourceObjInfo {
148     MetaspaceClosure::Ref* _ref; // The object that's copied into the buffer
149     uintx _ptrmap_start;     // The bit-offset of the start of this object (inclusive)
150     uintx _ptrmap_end;       // The bit-offset of the end   of this object (exclusive)
151     bool _read_only;
152     FollowMode _follow_mode;
153     int _size_in_bytes;
154     MetaspaceObj::Type _msotype;
155     address _source_addr;    // The value of the source object (_ref->obj()) when this
156                              // SourceObjInfo was created. Note that _ref->obj() may change
157                              // later if _ref is relocated.
158     address _buffered_addr;  // The copy of _ref->obj() insider the buffer.
159   public:
160     SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
161       _ref(ref), _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _follow_mode(follow_mode),
162       _size_in_bytes(ref->size() * BytesPerWord), _msotype(ref->msotype()),
163       _source_addr(ref->obj()) {
164       if (follow_mode == point_to_it) {
165         _buffered_addr = ref->obj();
166       } else {
167         _buffered_addr = NULL;
168       }
169     }
170 
171     bool should_copy() const { return _follow_mode == make_a_copy; }
172     MetaspaceClosure::Ref* ref() const { return  _ref; }
173     void set_buffered_addr(address addr)  {
174       assert(should_copy(), "must be");
175       assert(_buffered_addr == NULL, "cannot be copied twice");
176       assert(addr != NULL, "must be a valid copy");
177       _buffered_addr = addr;
178     }
179     void set_ptrmap_start(uintx v) { _ptrmap_start = v;    }
180     void set_ptrmap_end(uintx v)   { _ptrmap_end = v;      }
181     uintx ptrmap_start()  const    { return _ptrmap_start; } // inclusive
182     uintx ptrmap_end()    const    { return _ptrmap_end;   } // exclusive
183     bool read_only()      const    { return _read_only;    }
184     int size_in_bytes()   const    { return _size_in_bytes; }
185     address source_addr() const    { return _source_addr; }
186     address buffered_addr() const  { return _buffered_addr; }
187     MetaspaceObj::Type msotype() const { return _msotype; }
188 
189     // convenience accessor
190     address obj() const { return ref()->obj(); }
191   };
192 
193   class SourceObjList {
194     uintx _total_bytes;
195     GrowableArray<SourceObjInfo*>* _objs;     // Source objects to be archived
196     CHeapBitMap _ptrmap;                      // Marks the addresses of the pointer fields
197                                               // in the source objects
198   public:
199     SourceObjList();
200     ~SourceObjList();
201 
202     GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
203 
204     void append(MetaspaceClosure::Ref* enclosing_ref, SourceObjInfo* src_info);
205     void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
206     void relocate(int i, ArchiveBuilder* builder);
207 
208     // convenience accessor
209     SourceObjInfo* at(int i) const { return objs()->at(i); }
210   };
211 
212   class SrcObjTableCleaner {
213   public:
214     bool do_entry(address key, const SourceObjInfo& value) {
215       delete value.ref();
216       return true;
217     }
218   };
219 
220   class CDSMapLogger;
221 
222   static const int INITIAL_TABLE_SIZE = 15889;
223   static const int MAX_TABLE_SIZE     = 1000000;
224 
225   ReservedSpace _shared_rs;
226   VirtualSpace _shared_vs;
227 
228   DumpRegion _rw_region;
229   DumpRegion _ro_region;
230   CHeapBitMap _ptrmap;    // bitmap used by ArchivePtrMarker
231 
232   SourceObjList _rw_src_objs;                 // objs to put in rw region
233   SourceObjList _ro_src_objs;                 // objs to put in ro region
234   ResizeableResourceHashtable<address, SourceObjInfo, ResourceObj::C_HEAP, mtClassShared> _src_obj_table;
235   ResizeableResourceHashtable<address, address, ResourceObj::C_HEAP, mtClassShared> _buffered_to_src_table;
236   GrowableArray<Klass*>* _klasses;
237   GrowableArray<Symbol*>* _symbols;
238   GrowableArray<SpecialRefInfo>* _special_refs;
239 
240   // statistics
241   DumpAllocStats _alloc_stats;
242   size_t _total_closed_heap_region_size;
243   size_t _total_open_heap_region_size;
244 
245   void print_region_stats(FileMapInfo *map_info,
246                           GrowableArray<MemRegion>* closed_heap_regions,
247                           GrowableArray<MemRegion>* open_heap_regions);
248   void print_bitmap_region_stats(size_t size, size_t total_size);
249   void print_heap_region_stats(GrowableArray<MemRegion>* regions,
250                                const char *name, size_t total_size);
251 
252   // For global access.
253   static ArchiveBuilder* _current;
254 
255 public:
256   // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
257   // These are usually for misc tables that are allocated in the RO space.
258   class OtherROAllocMark {
259     char* _oldtop;
260   public:
261     OtherROAllocMark() {
262       _oldtop = _current->_ro_region.top();
263     }
264     ~OtherROAllocMark();
265   };
266 
267 private:
268   bool is_dumping_full_module_graph();
269   FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
270 
271   void iterate_sorted_roots(MetaspaceClosure* it, bool is_relocating_pointers);
272   void sort_symbols_and_fix_hash();
273   void sort_klasses();
274   static int compare_symbols_by_address(Symbol** a, Symbol** b);
275   static int compare_klass_by_name(Klass** a, Klass** b);
276 
277   void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
278   void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
279 
280   void update_special_refs();
281   void relocate_embedded_pointers(SourceObjList* src_objs);
282 
283   bool is_excluded(Klass* k);
284   void clean_up_src_obj_table();
285 
286 protected:
287   virtual void iterate_roots(MetaspaceClosure* it, bool is_relocating_pointers) = 0;
288 
289   // Conservative estimate for number of bytes needed for:
290   size_t _estimated_metaspaceobj_bytes;   // all archived MetaspaceObj's.
291   size_t _estimated_hashtable_bytes;     // symbol table and dictionaries
292 
293   static const int _total_dump_regions = 2;
294 
295   size_t estimate_archive_size();
296 
297   void start_dump_space(DumpRegion* next);
298   void verify_estimate_size(size_t estimate, const char* which);
299 
300 public:
301   address reserve_buffer();
302 
303   address buffer_bottom()                    const { return _buffer_bottom;                       }
304   address buffer_top()                       const { return (address)current_dump_space()->top(); }
305   address requested_static_archive_bottom()  const { return  _requested_static_archive_bottom;    }
306   address mapped_static_archive_bottom()     const { return  _mapped_static_archive_bottom;       }
307   intx buffer_to_requested_delta()           const { return _buffer_to_requested_delta;           }
308 
309   bool is_in_buffer_space(address p) const {
310     return (buffer_bottom() <= p && p < buffer_top());
311   }
312 
313   template <typename T> bool is_in_requested_static_archive(T p) const {
314     return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
315   }
316 
317   template <typename T> bool is_in_mapped_static_archive(T p) const {
318     return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
319   }
320 
321   template <typename T> bool is_in_buffer_space(T obj) const {
322     return is_in_buffer_space(address(obj));
323   }
324 
325   template <typename T> T to_requested(T obj) const {
326     assert(is_in_buffer_space(obj), "must be");
327     return (T)(address(obj) + _buffer_to_requested_delta);
328   }
329 
330   static intx get_buffer_to_requested_delta() {
331     return current()->buffer_to_requested_delta();
332   }
333 
334 public:
335   static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
336 
337   // The address p points to an object inside the output buffer. When the archive is mapped
338   // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
339   uintx buffer_to_offset(address p) const;
340 
341   // Same as buffer_to_offset, except that the address p points to either (a) an object
342   // inside the output buffer, or (b), an object in the currently mapped static archive.
343   uintx any_to_offset(address p) const;
344 
345   template <typename T>
346   u4 buffer_to_offset_u4(T p) const {
347     uintx offset = buffer_to_offset((address)p);
348     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
349     return (u4)offset;
350   }
351 
352   template <typename T>
353   u4 any_to_offset_u4(T p) const {
354     uintx offset = any_to_offset((address)p);
355     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset " INTPTR_FORMAT, offset);
356     return (u4)offset;
357   }
358 
359   static void assert_is_vm_thread() PRODUCT_RETURN;
360 
361 public:
362   ArchiveBuilder();
363   ~ArchiveBuilder();
364 
365   void gather_klasses_and_symbols();
366   void gather_source_objs();
367   bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
368   bool gather_one_source_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref, bool read_only);
369   void add_special_ref(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset, size_t src_obj_size_in_bytes) {
370     _special_refs->append(SpecialRefInfo(type, src_obj, field_offset, src_obj_size_in_bytes));
371   }
372   void remember_embedded_pointer_in_copied_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref);
373 
374   DumpRegion* rw_region() { return &_rw_region; }
375   DumpRegion* ro_region() { return &_ro_region; }
376 
377   static char* rw_region_alloc(size_t num_bytes) {
378     return current()->rw_region()->allocate(num_bytes);
379   }
380   static char* ro_region_alloc(size_t num_bytes) {
381     return current()->ro_region()->allocate(num_bytes);
382   }
383 
384   template <typename T>
385   static Array<T>* new_ro_array(int length) {
386     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
387     Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
388     array->initialize(length);
389     return array;
390   }
391 
392   template <typename T>
393   static Array<T>* new_rw_array(int length) {
394     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
395     Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
396     array->initialize(length);
397     return array;
398   }
399 
400   template <typename T>
401   static size_t ro_array_bytesize(int length) {
402     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
403     return align_up(byte_size, SharedSpaceObjectAlignment);
404   }
405 
406   void dump_rw_metadata();
407   void dump_ro_metadata();
408   void relocate_metaspaceobj_embedded_pointers();
409   void relocate_roots();
410   void relocate_vm_classes();
411   void make_klasses_shareable();
412   void relocate_to_requested();
413   void write_archive(FileMapInfo* mapinfo,
414                      GrowableArray<MemRegion>* closed_heap_regions,
415                      GrowableArray<MemRegion>* open_heap_regions,
416                      GrowableArray<ArchiveHeapBitmapInfo>* closed_heap_oopmaps,
417                      GrowableArray<ArchiveHeapBitmapInfo>* open_heap_oopmaps);
418   void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
419                     bool read_only,  bool allow_exec);
420 
421   address get_buffered_addr(address src_addr) const;
422   address get_source_addr(address buffered_addr) const;
423   template <typename T> T get_source_addr(T buffered_addr) const {
424     return (T)get_source_addr((address)buffered_addr);
425   }
426 
427   // All klasses and symbols that will be copied into the archive
428   GrowableArray<Klass*>*  klasses() const { return _klasses; }
429   GrowableArray<Symbol*>* symbols() const { return _symbols; }
430 
431   static bool is_active() {
432     return (_current != NULL);
433   }
434 
435   static ArchiveBuilder* current() {
436     assert_is_vm_thread();
437     assert(_current != NULL, "ArchiveBuilder must be active");
438     return _current;
439   }
440 
441   static DumpAllocStats* alloc_stats() {
442     return &(current()->_alloc_stats);
443   }
444 
445   static CompactHashtableStats* symbol_stats() {
446     return alloc_stats()->symbol_stats();
447   }
448 
449   static CompactHashtableStats* string_stats() {
450     return alloc_stats()->string_stats();
451   }
452 
453   void relocate_klass_ptr_of_oop(oop o);
454 
455   static Klass* get_buffered_klass(Klass* src_klass) {
456     Klass* klass = (Klass*)current()->get_buffered_addr((address)src_klass);
457     assert(klass != NULL && klass->is_klass(), "must be");
458     return klass;
459   }
460 
461   static Symbol* get_buffered_symbol(Symbol* src_symbol) {
462     return (Symbol*)current()->get_buffered_addr((address)src_symbol);
463   }
464 
465   void print_stats();
466   void report_out_of_space(const char* name, size_t needed_bytes);
467 };
468 
469 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP