1 /*
  2  * Copyright (c) 2020, 2021, 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 ArchiveHeapOopmapInfo;
 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 class ArchiveBuilder : public StackObj {
 65 protected:
 66   DumpRegion* _current_dump_space;
 67   address _buffer_bottom;                      // for writing the contents of rw/ro regions
 68   address _last_verified_top;
 69   int _num_dump_regions_used;
 70   size_t _other_region_used_bytes;
 71 
 72   // These are the addresses where we will request the static and dynamic archives to be
 73   // mapped at run time. If the request fails (due to ASLR), we will map the archives at
 74   // os-selected addresses.
 75   address _requested_static_archive_bottom;     // This is determined solely by the value of
 76                                                 // SharedBaseAddress during -Xshare:dump.
 77   address _requested_static_archive_top;
 78   address _requested_dynamic_archive_bottom;    // Used only during dynamic dump. It's placed
 79                                                 // immediately above _requested_static_archive_top.
 80   address _requested_dynamic_archive_top;
 81 
 82   // (Used only during dynamic dump) where the static archive is actually mapped. This
 83   // may be different than _requested_static_archive_{bottom,top} due to ASLR
 84   address _mapped_static_archive_bottom;
 85   address _mapped_static_archive_top;
 86 
 87   intx _buffer_to_requested_delta;
 88 
 89   DumpRegion* current_dump_space() const {  return _current_dump_space;  }
 90 
 91 public:
 92   enum FollowMode {
 93     make_a_copy, point_to_it, set_to_null
 94   };
 95 
 96 private:
 97   class SpecialRefInfo {
 98     // We have a "special pointer" of the given _type at _field_offset of _src_obj.
 99     // See MetaspaceClosure::push_special().
100     MetaspaceClosure::SpecialRef _type;
101     address _src_obj;
102     size_t _field_offset;
103 
104   public:
105     SpecialRefInfo() {}
106     SpecialRefInfo(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset)
107       : _type(type), _src_obj(src_obj), _field_offset(field_offset) {}
108 
109     MetaspaceClosure::SpecialRef type() const { return _type;         }
110     address src_obj()                   const { return _src_obj;      }
111     size_t field_offset()               const { return _field_offset; }
112   };
113 
114   class SourceObjInfo {
115     MetaspaceClosure::Ref* _ref;
116     uintx _ptrmap_start;     // The bit-offset of the start of this object (inclusive)
117     uintx _ptrmap_end;       // The bit-offset of the end   of this object (exclusive)
118     bool _read_only;
119     FollowMode _follow_mode;
120     int _size_in_bytes;
121     MetaspaceObj::Type _msotype;
122     address _dumped_addr;    // Address this->obj(), as used by the dumped archive.
123     address _orig_obj;       // The value of the original object (_ref->obj()) when this
124                              // SourceObjInfo was created. Note that _ref->obj() may change
125                              // later if _ref is relocated.
126 
127   public:
128     SourceObjInfo(MetaspaceClosure::Ref* ref, bool read_only, FollowMode follow_mode) :
129       _ref(ref), _ptrmap_start(0), _ptrmap_end(0), _read_only(read_only), _follow_mode(follow_mode),
130       _size_in_bytes(ref->size() * BytesPerWord), _msotype(ref->msotype()),
131       _orig_obj(ref->obj()) {
132       if (follow_mode == point_to_it) {
133         _dumped_addr = ref->obj();
134       } else {
135         _dumped_addr = NULL;
136       }
137     }
138 
139     bool should_copy() const { return _follow_mode == make_a_copy; }
140     MetaspaceClosure::Ref* ref() const { return  _ref; }
141     void set_dumped_addr(address dumped_addr)  {
142       assert(should_copy(), "must be");
143       assert(_dumped_addr == NULL, "cannot be copied twice");
144       assert(dumped_addr != NULL, "must be a valid copy");
145       _dumped_addr = dumped_addr;
146     }
147     void set_ptrmap_start(uintx v) { _ptrmap_start = v;    }
148     void set_ptrmap_end(uintx v)   { _ptrmap_end = v;      }
149     uintx ptrmap_start()  const    { return _ptrmap_start; } // inclusive
150     uintx ptrmap_end()    const    { return _ptrmap_end;   } // exclusive
151     bool read_only()      const    { return _read_only;    }
152     int size_in_bytes()   const    { return _size_in_bytes; }
153     address orig_obj()    const    { return _orig_obj; }
154     address dumped_addr() const    { return _dumped_addr; }
155     MetaspaceObj::Type msotype() const { return _msotype; }
156 
157     // convenience accessor
158     address obj() const { return ref()->obj(); }
159   };
160 
161   class SourceObjList {
162     uintx _total_bytes;
163     GrowableArray<SourceObjInfo*>* _objs;     // Source objects to be archived
164     CHeapBitMap _ptrmap;                      // Marks the addresses of the pointer fields
165                                               // in the source objects
166   public:
167     SourceObjList();
168     ~SourceObjList();
169 
170     GrowableArray<SourceObjInfo*>* objs() const { return _objs; }
171 
172     void append(MetaspaceClosure::Ref* enclosing_ref, SourceObjInfo* src_info);
173     void remember_embedded_pointer(SourceObjInfo* pointing_obj, MetaspaceClosure::Ref* ref);
174     void relocate(int i, ArchiveBuilder* builder);
175 
176     // convenience accessor
177     SourceObjInfo* at(int i) const { return objs()->at(i); }
178   };
179 
180   class SrcObjTableCleaner {
181   public:
182     bool do_entry(address key, const SourceObjInfo& value) {
183       delete value.ref();
184       return true;
185     }
186   };
187 
188   class CDSMapLogger;
189 
190   static const int INITIAL_TABLE_SIZE = 15889;
191   static const int MAX_TABLE_SIZE     = 1000000;
192 
193   ReservedSpace _shared_rs;
194   VirtualSpace _shared_vs;
195 
196   DumpRegion _rw_region;
197   DumpRegion _ro_region;
198   CHeapBitMap _ptrmap;    // bitmap used by ArchivePtrMarker
199 
200   SourceObjList _rw_src_objs;                 // objs to put in rw region
201   SourceObjList _ro_src_objs;                 // objs to put in ro region
202   ResizeableResourceHashtable<address, SourceObjInfo, ResourceObj::C_HEAP, mtClassShared> _src_obj_table;
203   GrowableArray<Klass*>* _klasses;
204   GrowableArray<Symbol*>* _symbols;
205   GrowableArray<SpecialRefInfo>* _special_refs;
206 
207   // statistics
208   DumpAllocStats _alloc_stats;
209   size_t _total_closed_heap_region_size;
210   size_t _total_open_heap_region_size;
211 
212   void print_region_stats(FileMapInfo *map_info,
213                           GrowableArray<MemRegion>* closed_heap_regions,
214                           GrowableArray<MemRegion>* open_heap_regions);
215   void print_bitmap_region_stats(size_t size, size_t total_size);
216   void print_heap_region_stats(GrowableArray<MemRegion>* regions,
217                                const char *name, size_t total_size);
218 
219   // For global access.
220   static ArchiveBuilder* _current;
221 
222 public:
223   // Use this when you allocate space outside of ArchiveBuilder::dump_{rw,ro}_region.
224   // These are usually for misc tables that are allocated in the RO space.
225   class OtherROAllocMark {
226     char* _oldtop;
227   public:
228     OtherROAllocMark() {
229       _oldtop = _current->_ro_region.top();
230     }
231     ~OtherROAllocMark();
232   };
233 
234 private:
235   bool is_dumping_full_module_graph();
236   FollowMode get_follow_mode(MetaspaceClosure::Ref *ref);
237 
238   void iterate_sorted_roots(MetaspaceClosure* it, bool is_relocating_pointers);
239   void sort_symbols_and_fix_hash();
240   void sort_klasses();
241   static int compare_symbols_by_address(Symbol** a, Symbol** b);
242   static int compare_klass_by_name(Klass** a, Klass** b);
243 
244   void make_shallow_copies(DumpRegion *dump_region, const SourceObjList* src_objs);
245   void make_shallow_copy(DumpRegion *dump_region, SourceObjInfo* src_info);
246 
247   void update_special_refs();
248   void relocate_embedded_pointers(SourceObjList* src_objs);
249 
250   bool is_excluded(Klass* k);
251   void clean_up_src_obj_table();
252 
253 protected:
254   virtual void iterate_roots(MetaspaceClosure* it, bool is_relocating_pointers) = 0;
255 
256   // Conservative estimate for number of bytes needed for:
257   size_t _estimated_metaspaceobj_bytes;   // all archived MetaspaceObj's.
258   size_t _estimated_hashtable_bytes;     // symbol table and dictionaries
259 
260   static const int _total_dump_regions = 2;
261 
262   size_t estimate_archive_size();
263 
264   void start_dump_space(DumpRegion* next);
265   void verify_estimate_size(size_t estimate, const char* which);
266 
267 public:
268   address reserve_buffer();
269 
270   address buffer_bottom()                    const { return _buffer_bottom;                       }
271   address buffer_top()                       const { return (address)current_dump_space()->top(); }
272   address requested_static_archive_bottom()  const { return  _requested_static_archive_bottom;    }
273   address mapped_static_archive_bottom()     const { return  _mapped_static_archive_bottom;       }
274   intx buffer_to_requested_delta()           const { return _buffer_to_requested_delta;           }
275 
276   bool is_in_buffer_space(address p) const {
277     return (buffer_bottom() <= p && p < buffer_top());
278   }
279 
280   template <typename T> bool is_in_requested_static_archive(T p) const {
281     return _requested_static_archive_bottom <= (address)p && (address)p < _requested_static_archive_top;
282   }
283 
284   template <typename T> bool is_in_mapped_static_archive(T p) const {
285     return _mapped_static_archive_bottom <= (address)p && (address)p < _mapped_static_archive_top;
286   }
287 
288   template <typename T> bool is_in_buffer_space(T obj) const {
289     return is_in_buffer_space(address(obj));
290   }
291 
292   template <typename T> T to_requested(T obj) const {
293     assert(is_in_buffer_space(obj), "must be");
294     return (T)(address(obj) + _buffer_to_requested_delta);
295   }
296 
297   static intx get_buffer_to_requested_delta() {
298     return current()->buffer_to_requested_delta();
299   }
300 
301 public:
302   static const uintx MAX_SHARED_DELTA = 0x7FFFFFFF;
303 
304   // The address p points to an object inside the output buffer. When the archive is mapped
305   // at the requested address, what's the offset of this object from _requested_static_archive_bottom?
306   uintx buffer_to_offset(address p) const;
307 
308   // Same as buffer_to_offset, except that the address p points to either (a) an object
309   // inside the output buffer, or (b), an object in the currently mapped static archive.
310   uintx any_to_offset(address p) const;
311 
312   template <typename T>
313   u4 buffer_to_offset_u4(T p) const {
314     uintx offset = buffer_to_offset((address)p);
315     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset");
316     return (u4)offset;
317   }
318 
319   template <typename T>
320   u4 any_to_offset_u4(T p) const {
321     uintx offset = any_to_offset((address)p);
322     guarantee(offset <= MAX_SHARED_DELTA, "must be 32-bit offset");
323     return (u4)offset;
324   }
325 
326   static void assert_is_vm_thread() PRODUCT_RETURN;
327 
328 public:
329   ArchiveBuilder();
330   ~ArchiveBuilder();
331 
332   void gather_klasses_and_symbols();
333   void gather_source_objs();
334   bool gather_klass_and_symbol(MetaspaceClosure::Ref* ref, bool read_only);
335   bool gather_one_source_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref, bool read_only);
336   void add_special_ref(MetaspaceClosure::SpecialRef type, address src_obj, size_t field_offset);
337   void remember_embedded_pointer_in_copied_obj(MetaspaceClosure::Ref* enclosing_ref, MetaspaceClosure::Ref* ref);
338 
339   DumpRegion* rw_region() { return &_rw_region; }
340   DumpRegion* ro_region() { return &_ro_region; }
341 
342   static char* rw_region_alloc(size_t num_bytes) {
343     return current()->rw_region()->allocate(num_bytes);
344   }
345   static char* ro_region_alloc(size_t num_bytes) {
346     return current()->ro_region()->allocate(num_bytes);
347   }
348 
349   template <typename T>
350   static Array<T>* new_ro_array(int length) {
351     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
352     Array<T>* array = (Array<T>*)ro_region_alloc(byte_size);
353     array->initialize(length);
354     return array;
355   }
356 
357   template <typename T>
358   static Array<T>* new_rw_array(int length) {
359     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
360     Array<T>* array = (Array<T>*)rw_region_alloc(byte_size);
361     array->initialize(length);
362     return array;
363   }
364 
365   template <typename T>
366   static size_t ro_array_bytesize(int length) {
367     size_t byte_size = Array<T>::byte_sizeof(length, sizeof(T));
368     return align_up(byte_size, SharedSpaceObjectAlignment);
369   }
370 
371   void dump_rw_metadata();
372   void dump_ro_metadata();
373   void relocate_metaspaceobj_embedded_pointers();
374   void relocate_roots();
375   void relocate_vm_classes();
376   void make_klasses_shareable();
377   void relocate_to_requested();
378   void write_archive(FileMapInfo* mapinfo,
379                      GrowableArray<MemRegion>* closed_heap_regions,
380                      GrowableArray<MemRegion>* open_heap_regions,
381                      GrowableArray<ArchiveHeapOopmapInfo>* closed_heap_oopmaps,
382                      GrowableArray<ArchiveHeapOopmapInfo>* open_heap_oopmaps);
383   void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
384                     bool read_only,  bool allow_exec);
385 
386   address get_dumped_addr(address src_obj) const;
387 
388   // All klasses and symbols that will be copied into the archive
389   GrowableArray<Klass*>*  klasses() const { return _klasses; }
390   GrowableArray<Symbol*>* symbols() const { return _symbols; }
391 
392   static bool is_active() {
393     return (_current != NULL);
394   }
395 
396   static ArchiveBuilder* current() {
397     assert_is_vm_thread();
398     assert(_current != NULL, "ArchiveBuilder must be active");
399     return _current;
400   }
401 
402   static DumpAllocStats* alloc_stats() {
403     return &(current()->_alloc_stats);
404   }
405 
406   static CompactHashtableStats* symbol_stats() {
407     return alloc_stats()->symbol_stats();
408   }
409 
410   static CompactHashtableStats* string_stats() {
411     return alloc_stats()->string_stats();
412   }
413 
414   void relocate_klass_ptr(oop o);
415 
416   static Klass* get_relocated_klass(Klass* orig_klass) {
417     Klass* klass = (Klass*)current()->get_dumped_addr((address)orig_klass);
418     assert(klass != NULL && klass->is_klass(), "must be");
419     return klass;
420   }
421 
422   static Symbol* get_relocated_symbol(Symbol* orig_symbol) {
423     return (Symbol*)current()->get_dumped_addr((address)orig_symbol);
424   }
425 
426   void print_stats();
427   void report_out_of_space(const char* name, size_t needed_bytes);
428 };
429 
430 #endif // SHARE_CDS_ARCHIVEBUILDER_HPP