src/runtime/coreparse.c

   1 /*
   2  * A saved SBCL system is a .core file; the code here helps us accept
   3  * such a file as input.
   4  */
   5
   6 /*
   7  * This software is part of the SBCL system. See the README file for
   8  * more information.
   9  *
  10  * This software is derived from the CMU CL system, which was
  11  * written at Carnegie Mellon University and released into the
  12  * public domain. The software is in the public domain and is
  13  * provided with absolutely no warranty. See the COPYING and CREDITS
  14  * files for more information.
  15  */
  16
  17 #include <stdio.h>
  18 #include <stdlib.h>
  19 #include <string.h>
  20 #include <sys/file.h>
  21 #include <sys/types.h>
  22 #include <sys/stat.h>
  23 #include <fcntl.h>
  24 #include <unistd.h>
  25
  26 #include "sbcl.h"
  27 #include "os.h"
  28 #include "runtime.h"
  29 #include "globals.h"
  30 #include "core.h"
  31 #include "arch.h"
  32 #include "interr.h"
  33 #include "thread.h"
  34
  35 #include "validate.h"
  36 #include "gc-internal.h"
  37
  38 unsigned char build_id[] =
  39 #include "../../output/build-id.tmp"
  40 ;
  41
  42 int
  43 open_binary(char *filename, int mode)
  44 {
  45 #ifdef LISP_FEATURE_WIN32
  46     mode |= O_BINARY;
  47 #endif
  48
  49     return open(filename, mode);
  50 }
  51
  52 /* Search 'filename' for an embedded core.  An SBCL core has, at the
  53  * end of the file, a trailer containing the size of the core (an
  54  * os_vm_offset_t) and a final signature word (the lispobj
  55  * CORE_MAGIC).  If this trailer is found at the end of the file, the
  56  * start of the core can be determined from the core size.
  57  *
  58  * If an embedded core is present, this returns the offset into the
  59  * file to load the core from, or -1 if no core is present. */
  60 os_vm_offset_t
  61 search_for_embedded_core(char *filename)
  62 {
  63     lispobj header;
  64     os_vm_offset_t lispobj_size = sizeof(lispobj);
  65     os_vm_offset_t trailer_size = lispobj_size + sizeof(os_vm_offset_t);
  66     os_vm_offset_t core_size, pos;
  67     int fd = -1;
  68
  69     if ((fd = open_binary(filename, O_RDONLY)) < 0)
  70         goto lose;
  71     if (lseek(fd, -lispobj_size, SEEK_END) < 0)
  72         goto lose;
  73     if (read(fd, &header, (size_t)lispobj_size) < lispobj_size)
  74         goto lose;
  75
  76     if (header == CORE_MAGIC) {
  77         if (lseek(fd, -trailer_size, SEEK_END) < 0)
  78             goto lose;
  79         if (read(fd, &core_size, sizeof(os_vm_offset_t)) < 0)
  80             goto lose;
  81
  82         if (lseek(fd, -(core_size + trailer_size), SEEK_END) < 0)
  83             goto lose;
  84         pos = lseek(fd, 0, SEEK_CUR);
  85
  86         if (read(fd, &header, (size_t)lispobj_size) < lispobj_size)
  87             goto lose;
  88
  89         if (header != CORE_MAGIC)
  90             goto lose;
  91
  92         close(fd);
  93         return pos;
  94     }
  95
  96 lose:
  97     if (fd != -1)
  98         close(fd);
  99
 100     return -1;
 101 }
 102
 103 static void
 104 process_directory(int fd, u32 *ptr, int count, os_vm_offset_t file_offset)
 105 {
 106     struct ndir_entry *entry;
 107
 108     FSHOW((stderr, "/process_directory(..), count=%d\n", count));
 109
 110     for (entry = (struct ndir_entry *) ptr; --count>= 0; ++entry) {
 111
 112         long id = entry->identifier;
 113         long offset = os_vm_page_size * (1 + entry->data_page);
 114         os_vm_address_t addr =
 115             (os_vm_address_t) (os_vm_page_size * entry->address);
 116         lispobj *free_pointer = (lispobj *) addr + entry->nwords;
 117         long len = os_vm_page_size * entry->page_count;
 118
 119         if (len != 0) {
 120             os_vm_address_t real_addr;
 121             FSHOW((stderr, "/mapping %ld(0x%lx) bytes at 0x%lx\n",
 122                    (long)len, (long)len, (unsigned long)addr));
 123             real_addr = os_map(fd, offset + file_offset, addr, len);
 124             if (real_addr != addr) {
 125                 lose("file mapped in wrong place! "
 126                      "(0x%08x != 0x%08lx)\n",
 127                      real_addr,
 128                      addr);
 129             }
 130         }
 131
 132         FSHOW((stderr, "/space id = %ld, free pointer = 0x%lx\n",
 133                id, (unsigned long)free_pointer));
 134
 135         switch (id) {
 136         case DYNAMIC_CORE_SPACE_ID:
 137 #ifdef LISP_FEATURE_GENCGC
 138             if (addr != (os_vm_address_t)DYNAMIC_SPACE_START) {
 139                 fprintf(stderr, "in core: 0x%lx; in runtime: 0x%lx \n",
 140                         (long)addr, (long)DYNAMIC_SPACE_START);
 141                 lose("core/runtime address mismatch: DYNAMIC_SPACE_START\n");
 142             }
 143 #else
 144             if ((addr != (os_vm_address_t)DYNAMIC_0_SPACE_START) &&
 145                 (addr != (os_vm_address_t)DYNAMIC_1_SPACE_START)) {
 146                 fprintf(stderr, "in core: 0x%lx; in runtime: 0x%lx or 0x%lx\n",
 147                         (long)addr,
 148                         (long)DYNAMIC_0_SPACE_START,
 149                         (long)DYNAMIC_1_SPACE_START);
 150                 lose("warning: core/runtime address mismatch: DYNAMIC_SPACE_START\n");
 151             }
 152 #endif
 153 #if defined(ALLOCATION_POINTER)
 154             SetSymbolValue(ALLOCATION_POINTER, (lispobj)free_pointer,0);
 155 #else
 156             dynamic_space_free_pointer = free_pointer;
 157 #endif
 158             /* For stop-and-copy GC, this will be whatever the GC was
 159              * using at the time. With GENCGC, this will always be
 160              * space 0. (We checked above that for GENCGC,
 161              * addr==DYNAMIC_SPACE_START.) */
 162             current_dynamic_space = (lispobj *)addr;
 163             break;
 164         case STATIC_CORE_SPACE_ID:
 165             if (addr != (os_vm_address_t)STATIC_SPACE_START) {
 166                 fprintf(stderr, "in core: 0x%lx - in runtime: 0x%lx\n",
 167                         (long)addr, (long)STATIC_SPACE_START);
 168                 lose("core/runtime address mismatch: STATIC_SPACE_START\n");
 169             }
 170             break;
 171         case READ_ONLY_CORE_SPACE_ID:
 172             if (addr != (os_vm_address_t)READ_ONLY_SPACE_START) {
 173                 fprintf(stderr, "in core: 0x%lx - in runtime: 0x%lx\n",
 174                         (long)addr, (long)READ_ONLY_SPACE_START);
 175                 lose("core/runtime address mismatch: READ_ONLY_SPACE_START\n");
 176             }
 177             break;
 178         default:
 179             lose("unknown space ID %ld addr 0x%lx\n", id, (long)addr);
 180         }
 181     }
 182 }
 183
 184 lispobj
 185 load_core_file(char *file, os_vm_offset_t file_offset)
 186 {
 187     lispobj *header, val, len, *ptr, remaining_len;
 188     int fd = open_binary(file, O_RDONLY), count;
 189
 190     lispobj initial_function = NIL;
 191     FSHOW((stderr, "/entering load_core_file(%s)\n", file));
 192     if (fd < 0) {
 193         fprintf(stderr, "could not open file \"%s\"\n", file);
 194         perror("open");
 195         exit(1);
 196     }
 197
 198     lseek(fd, file_offset, SEEK_SET);
 199     header = calloc(os_vm_page_size / sizeof(u32), sizeof(u32));
 200
 201     count = read(fd, header, os_vm_page_size);
 202     if (count < os_vm_page_size) {
 203         lose("premature end of core file\n");
 204     }
 205     SHOW("successfully read first page of core");
 206
 207     ptr = header;
 208     val = *ptr++;
 209
 210     if (val != CORE_MAGIC) {
 211         lose("invalid magic number in core: 0x%lx should have been 0x%x.\n",
 212              val,
 213              CORE_MAGIC);
 214     }
 215     SHOW("found CORE_MAGIC");
 216
 217     while (val != END_CORE_ENTRY_TYPE_CODE) {
 218         val = *ptr++;
 219         len = *ptr++;
 220         remaining_len = len - 2; /* (-2 to cancel the two ++ operations) */
 221         FSHOW((stderr, "/val=0x%ld, remaining_len=0x%ld\n",
 222                (long)val, (long)remaining_len));
 223
 224         switch (val) {
 225
 226         case END_CORE_ENTRY_TYPE_CODE:
 227             SHOW("END_CORE_ENTRY_TYPE_CODE case");
 228             break;
 229
 230         case VERSION_CORE_ENTRY_TYPE_CODE:
 231             SHOW("VERSION_CORE_ENTRY_TYPE_CODE case");
 232             if (*ptr != SBCL_CORE_VERSION_INTEGER) {
 233                 lose("core file version (%d) != runtime library version (%d)\n",
 234                      *ptr,
 235                      SBCL_CORE_VERSION_INTEGER);
 236             }
 237             break;
 238
 239         case BUILD_ID_CORE_ENTRY_TYPE_CODE:
 240             SHOW("BUILD_ID_CORE_ENTRY_TYPE_CODE case");
 241             {
 242                 int i;
 243
 244                 FSHOW((stderr, "build_id[]=\"%s\"\n", build_id));
 245                 FSHOW((stderr, "remaining_len = %d\n", remaining_len));
 246                 if (remaining_len != strlen((const char *)build_id))
 247                     goto losing_build_id;
 248                 for (i = 0; i < remaining_len; ++i) {
 249                     FSHOW((stderr, "ptr[%d] = char = %d, expected=%d\n",
 250                            i, ptr[i], build_id[i]));
 251                     if (ptr[i] != build_id[i])
 252                         goto losing_build_id;
 253                 }
 254                 break;
 255             losing_build_id:
 256                 /* .core files are not binary-compatible between
 257                  * builds because we can't easily detect whether the
 258                  * sources were patched between the time the
 259                  * dumping-the-.core runtime was built and the time
 260                  * that the loading-the-.core runtime was built.
 261                  *
 262                  * (We could easily detect whether version.lisp-expr
 263                  * was changed, but people experimenting with patches
 264                  * don't necessarily update version.lisp-expr.) */
 265
 266                 lose("can't load .core for different runtime, sorry\n");
 267             }
 268
 269         case NEW_DIRECTORY_CORE_ENTRY_TYPE_CODE:
 270             SHOW("NEW_DIRECTORY_CORE_ENTRY_TYPE_CODE case");
 271             process_directory(fd,
 272                               ptr,
 273 #ifndef LISP_FEATURE_ALPHA
 274                               remaining_len / (sizeof(struct ndir_entry) /
 275                                                sizeof(long)),
 276 #else
 277                               remaining_len / (sizeof(struct ndir_entry) /
 278                                                sizeof(u32)),
 279 #endif
 280                               file_offset);
 281             break;
 282
 283         case INITIAL_FUN_CORE_ENTRY_TYPE_CODE:
 284             SHOW("INITIAL_FUN_CORE_ENTRY_TYPE_CODE case");
 285             initial_function = (lispobj)*ptr;
 286             break;
 287
 288 #ifdef LISP_FEATURE_GENCGC
 289         case PAGE_TABLE_CORE_ENTRY_TYPE_CODE:
 290         {
 291             size_t size = *ptr;
 292             size_t fdoffset = (*(ptr+1) + 1) * (os_vm_page_size);
 293             size_t offset = 0;
 294             long bytes_read;
 295             long data[4096];
 296             lseek(fd, fdoffset + file_offset, SEEK_SET);
 297             while ((bytes_read = read(fd, data, (size < 4096 ? size : 4096 )))
 298                     > 0)
 299             {
 300                 int i = 0;
 301                 size -= bytes_read;
 302                 while (bytes_read) {
 303                     bytes_read -= sizeof(long);
 304                     page_table[offset++].first_object_offset = data[i++];
 305                 }
 306             }
 307
 308             gencgc_partial_pickup = 1;
 309             break;
 310         }
 311 #endif
 312         default:
 313             lose("unknown core file entry: %ld\n", (long)val);
 314         }
 315
 316         ptr += remaining_len;
 317         FSHOW((stderr, "/new ptr=%lx\n", (unsigned long)ptr));
 318     }
 319     SHOW("about to free(header)");
 320     free(header);
 321     SHOW("returning from load_core_file(..)");
 322     return initial_function;
 323 }
 324