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
  28 #ifndef LISP_FEATURE_WIN32
  29 #ifdef LISP_FEATURE_LINUX
  30 /* For madvise */
  31 # define _BSD_SOURCE
  32 #endif
  33 #include <sys/mman.h>
  34 #endif
  35
  36 #include "os.h"
  37 #include "runtime.h"
  38 #include "globals.h"
  39 #include "core.h"
  40 #include "arch.h"
  41 #include "interr.h"
  42 #include "thread.h"
  43
  44 #include "validate.h"
  45 #include "gc-internal.h"
  46
  47 /* lutex stuff */
  48 #if defined(LISP_FEATURE_SB_THREAD) && defined(LISP_FEATURE_SB_LUTEX)
  49 #include "genesis/sap.h"
  50 #include "pthread-lutex.h"
  51 #endif
  52
  53 #include <errno.h>
  54
  55 #ifdef LISP_FEATURE_SB_CORE_COMPRESSION
  56 # include <zlib.h>
  57 #endif
  58
  59 unsigned char build_id[] =
  60 #include "../../output/build-id.tmp"
  61 ;
  62
  63 int
  64 open_binary(char *filename, int mode)
  65 {
  66 #ifdef LISP_FEATURE_WIN32
  67     mode |= O_BINARY;
  68 #endif
  69
  70     return open(filename, mode);
  71 }
  72
  73
  74 static struct runtime_options *
  75 read_runtime_options(int fd)
  76 {
  77     size_t optarray[RUNTIME_OPTIONS_WORDS];
  78     struct runtime_options *options = NULL;
  79
  80     if (read(fd, optarray, RUNTIME_OPTIONS_WORDS * sizeof(size_t)) !=
  81         RUNTIME_OPTIONS_WORDS * sizeof(size_t)) {
  82         return NULL;
  83     }
  84
  85     if ((RUNTIME_OPTIONS_MAGIC != optarray[0]) || (0 == optarray[1])) {
  86         return NULL;
  87     }
  88
  89     options = successful_malloc(sizeof(struct runtime_options));
  90
  91     options->dynamic_space_size = optarray[2];
  92     options->thread_control_stack_size = optarray[3];
  93
  94     return options;
  95 }
  96
  97 void
  98 maybe_initialize_runtime_options(int fd)
  99 {
 100     struct runtime_options *new_runtime_options;
 101     off_t end_offset = sizeof(lispobj) +
 102         sizeof(os_vm_offset_t) +
 103         (RUNTIME_OPTIONS_WORDS * sizeof(size_t));
 104
 105     lseek(fd, -end_offset, SEEK_END);
 106
 107     if ((new_runtime_options = read_runtime_options(fd))) {
 108         runtime_options = new_runtime_options;
 109     }
 110 }
 111
 112 /* Search 'filename' for an embedded core.  An SBCL core has, at the
 113  * end of the file, a trailer containing optional saved runtime
 114  * options, the start of the core (an os_vm_offset_t), and a final
 115  * signature word (the lispobj CORE_MAGIC).  If this trailer is found
 116  * at the end of the file, the start of the core can be determined
 117  * from the core size.
 118  *
 119  * If an embedded core is present, this returns the offset into the
 120  * file to load the core from, or -1 if no core is present. */
 121 os_vm_offset_t
 122 search_for_embedded_core(char *filename)
 123 {
 124     lispobj header;
 125     os_vm_offset_t lispobj_size = sizeof(lispobj);
 126     os_vm_offset_t trailer_size = lispobj_size + sizeof(os_vm_offset_t);
 127     os_vm_offset_t core_start, pos;
 128     int fd = -1;
 129
 130     if ((fd = open_binary(filename, O_RDONLY)) < 0)
 131         goto lose;
 132
 133     if (read(fd, &header, (size_t)lispobj_size) < lispobj_size)
 134         goto lose;
 135     if (header == CORE_MAGIC) {
 136         /* This file is a real core, not an embedded core.  Return 0 to
 137          * indicate where the core starts, and do not look for runtime
 138          * options in this case. */
 139         return 0;
 140     }
 141
 142     if (lseek(fd, -lispobj_size, SEEK_END) < 0)
 143         goto lose;
 144     if (read(fd, &header, (size_t)lispobj_size) < lispobj_size)
 145         goto lose;
 146
 147     if (header == CORE_MAGIC) {
 148         if (lseek(fd, -trailer_size, SEEK_END) < 0)
 149             goto lose;
 150         if (read(fd, &core_start, sizeof(os_vm_offset_t)) < 0)
 151             goto lose;
 152
 153         if (lseek(fd, core_start, SEEK_SET) < 0)
 154             goto lose;
 155         pos = lseek(fd, 0, SEEK_CUR);
 156
 157         if (read(fd, &header, (size_t)lispobj_size) < lispobj_size)
 158             goto lose;
 159
 160         if (header != CORE_MAGIC)
 161             goto lose;
 162
 163         maybe_initialize_runtime_options(fd);
 164
 165         close(fd);
 166         return pos;
 167     }
 168
 169 lose:
 170     if (fd != -1)
 171         close(fd);
 172
 173     return -1;
 174 }
 175
 176 /* If more platforms doesn't support overlapping mmap rename this
 177  * def to something like ifdef nommapoverlap */
 178 /* currently hpux only */
 179 #ifdef LISP_FEATURE_HPUX
 180 os_vm_address_t copy_core_bytes(int fd, os_vm_offset_t offset,
 181                                 os_vm_address_t addr, int len)
 182 {
 183   unsigned char buf[4096];
 184   int c,x;
 185   int old_fd = lseek(fd, 0, SEEK_CUR);
 186
 187   if(len & (4096-1)){
 188     fprintf(stderr, "cant copy a slice of core because slice-length is not of page size(4096)\n");
 189     exit(-1);
 190   }
 191   if(old_fd < 0){
 192     fprintf(stderr, "cant perform lseek() on corefile\n");
 193   }
 194   lseek(fd, offset, SEEK_SET);
 195   if(fd < 0){
 196     fprintf(stderr, "cant perform lseek(%u,%lu,SEEK_SET) on corefile\n", fd, offset);
 197   }
 198   for(x = 0; x < len; x += 4096){
 199     c = read(fd, buf, 4096);
 200     if(c != 4096){
 201       fprintf(stderr, "cant read memory area from corefile at position %lu, got %d\n", offset + x, c);
 202       exit(-1);
 203     }
 204     memcpy(addr+x, buf, 4096);
 205   }
 206   os_flush_icache(addr, len);
 207   return addr;
 208 }
 209 #endif
 210
 211 #ifdef LISP_FEATURE_SB_CORE_COMPRESSION
 212 # define ZLIB_BUFFER_SIZE (1u<<16)
 213 os_vm_address_t inflate_core_bytes(int fd, os_vm_offset_t offset,
 214                                    os_vm_address_t addr, int len)
 215 {
 216     z_stream stream;
 217     unsigned char buf[ZLIB_BUFFER_SIZE];
 218     int ret;
 219
 220     if (-1 == lseek(fd, offset, SEEK_SET)) {
 221         lose("Unable to lseek() on corefile\n");
 222     }
 223
 224     stream.zalloc = NULL;
 225     stream.zfree = NULL;
 226     stream.opaque = NULL;
 227     stream.avail_in = 0;
 228     stream.next_in = buf;
 229
 230     ret = inflateInit(&stream);
 231     if (ret != Z_OK)
 232         lose("zlib error %i\n", ret);
 233
 234     stream.next_out  = (void*)addr;
 235     stream.avail_out = len;
 236     do {
 237         ssize_t count = read(fd, buf, sizeof(buf));
 238         if (count < 0)
 239             lose("unable to read core file (errno = %i)\n", errno);
 240         stream.next_in = buf;
 241         stream.avail_in = count;
 242         if (count == 0) break;
 243         ret = inflate(&stream, Z_NO_FLUSH);
 244         switch (ret) {
 245         case Z_STREAM_END:
 246             break;
 247         case Z_OK:
 248             if (stream.avail_out == 0)
 249                 lose("Runaway gzipped core directory... aborting\n");
 250             if (stream.avail_in > 0)
 251                 lose("zlib inflate returned without fully"
 252                      "using up input buffer... aborting\n");
 253             break;
 254         default:
 255             lose("zlib inflate error: %i\n", ret);
 256             break;
 257         }
 258     } while (ret != Z_STREAM_END);
 259
 260     if (stream.avail_out > 0) {
 261         if (stream.avail_out >= os_vm_page_size)
 262             fprintf(stderr, "Warning: gzipped core directory significantly"
 263                     "shorter than expected (%lu bytes)", (unsigned long)stream.avail_out);
 264         /* Is this needed? */
 265         memset(stream.next_out, 0, stream.avail_out);
 266     }
 267
 268     inflateEnd(&stream);
 269     return addr;
 270 }
 271 # undef ZLIB_BUFFER_SIZE
 272 #endif
 273
 274 int merge_core_pages = -1;
 275
 276 static void
 277 process_directory(int fd, lispobj *ptr, int count, os_vm_offset_t file_offset)
 278 {
 279     struct ndir_entry *entry;
 280     int compressed;
 281
 282     FSHOW((stderr, "/process_directory(..), count=%d\n", count));
 283
 284     for (entry = (struct ndir_entry *) ptr; --count>= 0; ++entry) {
 285         compressed = 0;
 286         long id = entry->identifier;
 287         if (id <= (MAX_CORE_SPACE_ID | DEFLATED_CORE_SPACE_ID_FLAG)) {
 288             if (id & DEFLATED_CORE_SPACE_ID_FLAG)
 289                 compressed = 1;
 290             id &= ~(DEFLATED_CORE_SPACE_ID_FLAG);
 291         }
 292         long offset = os_vm_page_size * (1 + entry->data_page);
 293         os_vm_address_t addr =
 294             (os_vm_address_t) (os_vm_page_size * entry->address);
 295         lispobj *free_pointer = (lispobj *) addr + entry->nwords;
 296         unsigned long len = os_vm_page_size * entry->page_count;
 297         if (len != 0) {
 298             os_vm_address_t real_addr;
 299             FSHOW((stderr, "/mapping %ld(0x%lx) bytes at 0x%lx\n",
 300                    (long)len, (long)len, (unsigned long)addr));
 301             if (compressed) {
 302 #ifdef LISP_FEATURE_SB_CORE_COMPRESSION
 303                 real_addr = inflate_core_bytes(fd, offset + file_offset, addr, len);
 304 #else
 305                 lose("This runtime was not built with zlib-compressed core support... aborting\n");
 306 #endif
 307             } else {
 308 #ifdef LISP_FEATURE_HPUX
 309                 real_addr = copy_core_bytes(fd, offset + file_offset, addr, len);
 310 #else
 311                 real_addr = os_map(fd, offset + file_offset, addr, len);
 312 #endif
 313             }
 314             if (real_addr != addr) {
 315                 lose("file mapped in wrong place! "
 316                      "(0x%08x != 0x%08lx)\n",
 317                      real_addr,
 318                      addr);
 319             }
 320         }
 321
 322 #ifdef MADV_MERGEABLE
 323         if ((merge_core_pages == 1)
 324             || ((merge_core_pages == -1) && compressed)) {
 325                 madvise(addr, len, MADV_MERGEABLE);
 326         }
 327 #endif
 328
 329         FSHOW((stderr, "/space id = %ld, free pointer = 0x%lx\n",
 330                id, (unsigned long)free_pointer));
 331
 332         switch (id) {
 333         case DYNAMIC_CORE_SPACE_ID:
 334             if (len > dynamic_space_size) {
 335                 fprintf(stderr,
 336                         "dynamic space too small for core: %ldKiB required, %ldKiB available.\n",
 337                         len >> 10,
 338                         (long)dynamic_space_size >> 10);
 339                 exit(1);
 340             }
 341 #ifdef LISP_FEATURE_GENCGC
 342             if (addr != (os_vm_address_t)DYNAMIC_SPACE_START) {
 343                 fprintf(stderr, "in core: 0x%lx; in runtime: 0x%lx \n",
 344                         (long)addr, (long)DYNAMIC_SPACE_START);
 345                 lose("core/runtime address mismatch: DYNAMIC_SPACE_START\n");
 346             }
 347 #else
 348             if ((addr != (os_vm_address_t)DYNAMIC_0_SPACE_START) &&
 349                 (addr != (os_vm_address_t)DYNAMIC_1_SPACE_START)) {
 350                 fprintf(stderr, "in core: 0x%lx; in runtime: 0x%lx or 0x%lx\n",
 351                         (long)addr,
 352                         (long)DYNAMIC_0_SPACE_START,
 353                         (long)DYNAMIC_1_SPACE_START);
 354                 lose("warning: core/runtime address mismatch: DYNAMIC_SPACE_START\n");
 355             }
 356 #endif
 357 #if defined(ALLOCATION_POINTER)
 358             SetSymbolValue(ALLOCATION_POINTER, (lispobj)free_pointer,0);
 359 #else
 360             dynamic_space_free_pointer = free_pointer;
 361 #endif
 362             /* For stop-and-copy GC, this will be whatever the GC was
 363              * using at the time. With GENCGC, this will always be
 364              * space 0. (We checked above that for GENCGC,
 365              * addr==DYNAMIC_SPACE_START.) */
 366             current_dynamic_space = (lispobj *)addr;
 367             break;
 368         case STATIC_CORE_SPACE_ID:
 369             if (addr != (os_vm_address_t)STATIC_SPACE_START) {
 370                 fprintf(stderr, "in core: 0x%lx - in runtime: 0x%lx\n",
 371                         (long)addr, (long)STATIC_SPACE_START);
 372                 lose("core/runtime address mismatch: STATIC_SPACE_START\n");
 373             }
 374             break;
 375         case READ_ONLY_CORE_SPACE_ID:
 376             if (addr != (os_vm_address_t)READ_ONLY_SPACE_START) {
 377                 fprintf(stderr, "in core: 0x%lx - in runtime: 0x%lx\n",
 378                         (long)addr, (long)READ_ONLY_SPACE_START);
 379                 lose("core/runtime address mismatch: READ_ONLY_SPACE_START\n");
 380             }
 381             break;
 382         default:
 383             lose("unknown space ID %ld addr 0x%lx\n", id, (long)addr);
 384         }
 385     }
 386 }
 387
 388 lispobj
 389 load_core_file(char *file, os_vm_offset_t file_offset)
 390 {
 391     lispobj *header, val, len, *ptr, remaining_len;
 392     int fd = open_binary(file, O_RDONLY);
 393     unsigned int count;
 394
 395     lispobj initial_function = NIL;
 396     FSHOW((stderr, "/entering load_core_file(%s)\n", file));
 397     if (fd < 0) {
 398         fprintf(stderr, "could not open file \"%s\"\n", file);
 399         perror("open");
 400         exit(1);
 401     }
 402
 403     lseek(fd, file_offset, SEEK_SET);
 404     header = calloc(os_vm_page_size / sizeof(u32), sizeof(u32));
 405
 406     count = read(fd, header, os_vm_page_size);
 407     if (count < os_vm_page_size) {
 408         lose("premature end of core file\n");
 409     }
 410     SHOW("successfully read first page of core");
 411
 412     ptr = header;
 413     val = *ptr++;
 414
 415     if (val != CORE_MAGIC) {
 416         lose("invalid magic number in core: 0x%lx should have been 0x%x.\n",
 417              val,
 418              CORE_MAGIC);
 419     }
 420     SHOW("found CORE_MAGIC");
 421
 422     while (val != END_CORE_ENTRY_TYPE_CODE) {
 423         val = *ptr++;
 424         len = *ptr++;
 425         remaining_len = len - 2; /* (-2 to cancel the two ++ operations) */
 426         FSHOW((stderr, "/val=0x%ld, remaining_len=0x%ld\n",
 427                (long)val, (long)remaining_len));
 428
 429         switch (val) {
 430
 431         case END_CORE_ENTRY_TYPE_CODE:
 432             SHOW("END_CORE_ENTRY_TYPE_CODE case");
 433             break;
 434
 435         case VERSION_CORE_ENTRY_TYPE_CODE:
 436             SHOW("VERSION_CORE_ENTRY_TYPE_CODE case");
 437             if (*ptr != SBCL_CORE_VERSION_INTEGER) {
 438                 lose("core file version (%d) != runtime library version (%d)\n",
 439                      *ptr,
 440                      SBCL_CORE_VERSION_INTEGER);
 441             }
 442             break;
 443
 444         case BUILD_ID_CORE_ENTRY_TYPE_CODE:
 445             SHOW("BUILD_ID_CORE_ENTRY_TYPE_CODE case");
 446             {
 447                 unsigned int i;
 448
 449                 FSHOW((stderr, "build_id[]=\"%s\"\n", build_id));
 450                 FSHOW((stderr, "remaining_len = %d\n", remaining_len));
 451                 if (remaining_len != strlen((const char *)build_id))
 452                     goto losing_build_id;
 453                 for (i = 0; i < remaining_len; ++i) {
 454                     FSHOW((stderr, "ptr[%d] = char = %d, expected=%d\n",
 455                            i, ptr[i], build_id[i]));
 456                     if (ptr[i] != build_id[i])
 457                         goto losing_build_id;
 458                 }
 459                 break;
 460             losing_build_id:
 461                 /* .core files are not binary-compatible between
 462                  * builds because we can't easily detect whether the
 463                  * sources were patched between the time the
 464                  * dumping-the-.core runtime was built and the time
 465                  * that the loading-the-.core runtime was built.
 466                  *
 467                  * (We could easily detect whether version.lisp-expr
 468                  * was changed, but people experimenting with patches
 469                  * don't necessarily update version.lisp-expr.) */
 470
 471                 lose("can't load .core for different runtime, sorry\n");
 472             }
 473
 474         case NEW_DIRECTORY_CORE_ENTRY_TYPE_CODE:
 475             SHOW("NEW_DIRECTORY_CORE_ENTRY_TYPE_CODE case");
 476             process_directory(fd,
 477                               ptr,
 478 #ifndef LISP_FEATURE_ALPHA
 479                               remaining_len / (sizeof(struct ndir_entry) /
 480                                                sizeof(long)),
 481 #else
 482                               remaining_len / (sizeof(struct ndir_entry) /
 483                                                sizeof(u32)),
 484 #endif
 485                               file_offset);
 486             break;
 487
 488         case INITIAL_FUN_CORE_ENTRY_TYPE_CODE:
 489             SHOW("INITIAL_FUN_CORE_ENTRY_TYPE_CODE case");
 490             initial_function = (lispobj)*ptr;
 491             break;
 492
 493 #if defined(LISP_FEATURE_SB_THREAD) && defined(LISP_FEATURE_SB_LUTEX)
 494         case LUTEX_TABLE_CORE_ENTRY_TYPE_CODE:
 495             SHOW("LUTEX_TABLE_CORE_ENTRY_TYPE_CODE case");
 496             {
 497                 size_t n_lutexes = *ptr;
 498                 size_t fdoffset = (*(ptr + 1) + 1) * (os_vm_page_size);
 499                 size_t data_length = n_lutexes * sizeof(struct sap *);
 500                 struct lutex **lutexes_to_resurrect = malloc(data_length);
 501                 long bytes_read;
 502
 503                 lseek(fd, fdoffset + file_offset, SEEK_SET);
 504
 505                 FSHOW((stderr, "attempting to read %ld lutexes from core\n", n_lutexes));
 506                 bytes_read = read(fd, lutexes_to_resurrect, data_length);
 507
 508                 /* XXX */
 509                 if (bytes_read != data_length) {
 510                     lose("Could not read the lutex table");
 511                 }
 512                 else {
 513                     int i;
 514
 515                     for (i=0; i<n_lutexes; ++i) {
 516                         struct lutex *lutex = lutexes_to_resurrect[i];
 517
 518                         FSHOW((stderr, "re-init'ing lutex @ %p\n", lutex));
 519                         lutex_init((tagged_lutex_t) lutex);
 520                     }
 521
 522                     free(lutexes_to_resurrect);
 523                 }
 524                 break;
 525             }
 526 #endif
 527
 528 #ifdef LISP_FEATURE_GENCGC
 529         case PAGE_TABLE_CORE_ENTRY_TYPE_CODE:
 530         {
 531             size_t size = *ptr;
 532             size_t fdoffset = (*(ptr+1) + 1) * (os_vm_page_size);
 533             size_t offset = 0;
 534             long bytes_read;
 535             unsigned long data[4096];
 536             unsigned long word;
 537             lseek(fd, fdoffset + file_offset, SEEK_SET);
 538             while ((bytes_read = read(fd, data, (size < 4096 ? size : 4096 )))
 539                     > 0)
 540             {
 541                 int i = 0;
 542                 size -= bytes_read;
 543                 while (bytes_read) {
 544                     bytes_read -= sizeof(long);
 545                     /* Ignore all zeroes. The size of the page table
 546                      * core entry was rounded up to os_vm_page_size
 547                      * during the save, and might now have more
 548                      * elements than the page table.
 549                      *
 550                      * The low bits of each word are allocation flags.
 551                      */
 552                     if (word=data[i]) {
 553                         page_table[offset].region_start_offset = word & ~0x03;
 554                         page_table[offset].allocated = word & 0x03;
 555                     }
 556                     i++;
 557                     offset++;
 558                 }
 559             }
 560
 561             gencgc_partial_pickup = 1;
 562             break;
 563         }
 564 #endif
 565         default:
 566             lose("unknown core file entry: %ld\n", (long)val);
 567         }
 568
 569         ptr += remaining_len;
 570         FSHOW((stderr, "/new ptr=%lx\n", (unsigned long)ptr));
 571     }
 572     SHOW("about to free(header)");
 573     free(header);
 574     SHOW("returning from load_core_file(..)");
 575     return initial_function;
 576 }
 577