usr.bin/gcore/elfcore.c

   1 /*-
   2  * Copyright (c) 1998 John D. Polstra
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  *
  26  * $FreeBSD: src/usr.bin/gcore/elfcore.c,v 1.5.2.3 2001/07/11 23:59:11 obrien Exp $
  27  */
  28
  29 #define _KERNEL_STRUCTURES
  30
  31 #include <sys/param.h>
  32 #include <sys/lock.h>
  33 #include <sys/procfs.h>
  34 #include <machine/elf.h>
  35 #include <vm/vm_param.h>
  36 #include <vm/vm.h>
  37 #include <vm/pmap.h>
  38 #include <vm/vm_map.h>
  39 #include <err.h>
  40 #include <errno.h>
  41 #include <fcntl.h>
  42 #include <stdio.h>
  43 #include <stdlib.h>
  44 #include <string.h>
  45 #include <unistd.h>
  46
  47 #include "extern.h"
  48
  49 extern ssize_t gcore_seg_limit;
  50 extern int gcore_verbose;
  51
  52 /*
  53  * Code for generating ELF core dumps.
  54  */
  55
  56 typedef void (*segment_callback)(vm_map_entry_t, void *);
  57
  58 /* Closure for cb_put_phdr(). */
  59 struct phdr_closure {
  60         Elf_Phdr *phdr;         /* Program header to fill in */
  61         Elf_Off offset;         /* Offset of segment in core file */
  62 };
  63
  64 /* Closure for cb_size_segment(). */
  65 struct sseg_closure {
  66         int count;              /* Count of writable segments. */
  67         size_t size;            /* Total size of all writable segments. */
  68 };
  69
  70 static void cb_put_phdr(vm_map_entry_t, void *);
  71 static void cb_size_segment(vm_map_entry_t, void *);
  72 static void each_writable_segment(vm_map_entry_t, segment_callback,
  73     void *closure);
  74 static void elf_corehdr(int fd, pid_t, vm_map_entry_t, int numsegs,
  75     void *hdr, size_t hdrsize);
  76 static void elf_puthdr(vm_map_entry_t, void *, size_t *,
  77     const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int numsegs);
  78 static void elf_putnote(void *dst, size_t *off, const char *name, int type,
  79     const void *desc, size_t descsz);
  80 static void freemap(vm_map_entry_t);
  81 static void readhdrinfo(pid_t, prstatus_t *, prfpregset_t *, prpsinfo_t *);
  82 static vm_map_entry_t readmap(pid_t);
  83
  84 /*
  85  * Write an ELF coredump for the given pid to the given fd.
  86  */
  87 void
  88 elf_coredump(int fd, pid_t pid)
  89 {
  90         vm_map_entry_t map;
  91         struct sseg_closure seginfo;
  92         void *hdr;
  93         size_t hdrsize;
  94         char memname[64];
  95         int memfd;
  96         Elf_Phdr *php;
  97         int i;
  98
  99         /* Get the program's memory map. */
 100         map = readmap(pid);
 101
 102         /* Size the program segments. */
 103         seginfo.count = 0;
 104         seginfo.size = 0;
 105         each_writable_segment(map, cb_size_segment, &seginfo);
 106
 107         /*
 108          * Calculate the size of the core file header area by making
 109          * a dry run of generating it.  Nothing is written, but the
 110          * size is calculated.
 111          */
 112         hdrsize = 0;
 113         elf_puthdr(map, NULL, &hdrsize, NULL, NULL, NULL, seginfo.count);
 114
 115         /*
 116          * Allocate memory for building the header, fill it up,
 117          * and write it out.
 118          */
 119         hdr = malloc(hdrsize);
 120         if ((hdr = malloc(hdrsize)) == NULL)
 121                 errx(1, "out of memory");
 122         elf_corehdr(fd, pid, map, seginfo.count, hdr, hdrsize);
 123
 124         /* Write the contents of all of the writable segments. */
 125         snprintf(memname, sizeof memname, "/proc/%d/mem", pid);
 126         if ((memfd = open(memname, O_RDONLY)) == -1)
 127                 err(1, "cannot open %s", memname);
 128
 129         php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 130         for (i = 0;  i < seginfo.count;  i++) {
 131                 long nleft = php->p_filesz;
 132
 133                 lseek(memfd, (off_t)php->p_vaddr, SEEK_SET);
 134                 while (nleft > 0) {
 135                         char buf[65536];
 136                         ssize_t nwant;
 137                         ssize_t ngot;
 138
 139                         nwant = (ssize_t)nleft;
 140                         if (nwant > sizeof buf)
 141                                 nwant = sizeof buf;
 142                         ngot = read(memfd, buf, nwant);
 143                         if (ngot == -1)
 144                                 err(1, "read from %s", memname);
 145                         if (ngot < nwant)
 146                                 errx(1, "short read from %s:"
 147                                     " wanted %zd, got %zd\n", memname,
 148                                     nwant, ngot);
 149                         ngot = write(fd, buf, nwant);
 150                         if (ngot == -1)
 151                                 err(1, "write of segment %d failed", i);
 152                         if (ngot != nwant)
 153                                 errx(1, "short write");
 154                         nleft -= nwant;
 155                 }
 156                 php++;
 157         }
 158         close(memfd);
 159         free(hdr);
 160         freemap(map);
 161 }
 162
 163 /*
 164  * A callback for each_writable_segment() to write out the segment's
 165  * program header entry.
 166  */
 167 static void
 168 cb_put_phdr(vm_map_entry_t entry, void *closure)
 169 {
 170         struct phdr_closure *phc = (struct phdr_closure *)closure;
 171         Elf_Phdr *phdr = phc->phdr;
 172
 173         phc->offset = round_page(phc->offset);
 174
 175         phdr->p_type = PT_LOAD;
 176         phdr->p_offset = phc->offset;
 177         phdr->p_vaddr = entry->ba.start;
 178         phdr->p_paddr = 0;
 179         phdr->p_filesz = phdr->p_memsz = entry->ba.end - entry->ba.start;
 180         phdr->p_align = PAGE_SIZE;
 181         phdr->p_flags = 0;
 182         if (entry->protection & VM_PROT_READ)
 183                 phdr->p_flags |= PF_R;
 184         if (entry->protection & VM_PROT_WRITE)
 185                 phdr->p_flags |= PF_W;
 186         if (entry->protection & VM_PROT_EXECUTE)
 187                 phdr->p_flags |= PF_X;
 188
 189         phc->offset += phdr->p_filesz;
 190         phc->phdr++;
 191 }
 192
 193 /*
 194  * A callback for each_writable_segment() to gather information about
 195  * the number of segments and their total size.
 196  */
 197 static void
 198 cb_size_segment(vm_map_entry_t entry, void *closure)
 199 {
 200         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 201
 202         ssc->count++;
 203         ssc->size += entry->ba.end - entry->ba.start;
 204 }
 205
 206 /*
 207  * For each segment in the given memory map, call the given function
 208  * with a pointer to the map entry and some arbitrary caller-supplied
 209  * data.
 210  */
 211 static void
 212 each_writable_segment(vm_map_entry_t map, segment_callback func, void *closure)
 213 {
 214         vm_map_entry_t entry;
 215
 216         for (entry = map; entry; entry = entry->rb_entry.rbe_parent)
 217                 (*func)(entry, closure);
 218 }
 219
 220 /*
 221  * Write the core file header to the file, including padding up to
 222  * the page boundary.
 223  */
 224 static void
 225 elf_corehdr(int fd, pid_t pid, vm_map_entry_t map, int numsegs, void *hdr,
 226     size_t hdrsize)
 227 {
 228         size_t off;
 229         prstatus_t status;
 230         prfpregset_t fpregset;
 231         prpsinfo_t psinfo;
 232
 233         /* Gather the information for the header. */
 234         readhdrinfo(pid, &status, &fpregset, &psinfo);
 235
 236         /* Fill in the header. */
 237         memset(hdr, 0, hdrsize);
 238         off = 0;
 239         elf_puthdr(map, hdr, &off, &status, &fpregset, &psinfo, numsegs);
 240
 241         /* Write it to the core file. */
 242         if (write(fd, hdr, hdrsize) == -1)
 243                 err(1, "write");
 244 }
 245
 246 /*
 247  * Generate the ELF coredump header into the buffer at "dst".  "dst" may
 248  * be NULL, in which case the header is sized but not actually generated.
 249  */
 250 static void
 251 elf_puthdr(vm_map_entry_t map, void *dst, size_t *off, const prstatus_t *status,
 252     const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
 253 {
 254         size_t ehoff;
 255         size_t phoff;
 256         size_t noteoff;
 257         size_t notesz;
 258
 259         ehoff = *off;
 260         *off += sizeof(Elf_Ehdr);
 261
 262         phoff = *off;
 263         *off += (numsegs + 1) * sizeof(Elf_Phdr);
 264
 265         noteoff = *off;
 266         elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
 267             sizeof *status);
 268         elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
 269             sizeof *fpregset);
 270         elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
 271             sizeof *psinfo);
 272         notesz = *off - noteoff;
 273
 274         /* Align up to a page boundary for the program segments. */
 275         *off = round_page(*off);
 276
 277         if (dst != NULL) {
 278                 Elf_Ehdr *ehdr;
 279                 Elf_Phdr *phdr;
 280                 struct phdr_closure phc;
 281
 282                 /*
 283                  * Fill in the ELF header.
 284                  */
 285                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
 286                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
 287                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
 288                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
 289                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
 290                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
 291                 ehdr->e_ident[EI_DATA] = ELF_DATA;
 292                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 293                 ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE;
 294                 ehdr->e_ident[EI_ABIVERSION] = 0;
 295                 ehdr->e_ident[EI_PAD] = 0;
 296                 ehdr->e_type = ET_CORE;
 297                 ehdr->e_machine = ELF_ARCH;
 298                 ehdr->e_version = EV_CURRENT;
 299                 ehdr->e_entry = 0;
 300                 ehdr->e_phoff = phoff;
 301                 ehdr->e_flags = 0;
 302                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
 303                 ehdr->e_phentsize = sizeof(Elf_Phdr);
 304                 ehdr->e_phnum = numsegs + 1;
 305                 ehdr->e_shentsize = sizeof(Elf_Shdr);
 306                 ehdr->e_shnum = 0;
 307                 ehdr->e_shstrndx = SHN_UNDEF;
 308
 309                 /*
 310                  * Fill in the program header entries.
 311                  */
 312                 phdr = (Elf_Phdr *)((char *)dst + phoff);
 313
 314                 /* The note segement. */
 315                 phdr->p_type = PT_NOTE;
 316                 phdr->p_offset = noteoff;
 317                 phdr->p_vaddr = 0;
 318                 phdr->p_paddr = 0;
 319                 phdr->p_filesz = notesz;
 320                 phdr->p_memsz = 0;
 321                 phdr->p_flags = 0;
 322                 phdr->p_align = 0;
 323                 phdr++;
 324
 325                 /* All the writable segments from the program. */
 326                 phc.phdr = phdr;
 327                 phc.offset = *off;
 328                 each_writable_segment(map, cb_put_phdr, &phc);
 329         }
 330 }
 331
 332 /*
 333  * Emit one note section to "dst", or just size it if "dst" is NULL.
 334  */
 335 static void
 336 elf_putnote(void *dst, size_t *off, const char *name, int type,
 337     const void *desc, size_t descsz)
 338 {
 339         Elf_Note note;
 340
 341         note.n_namesz = strlen(name) + 1;
 342         note.n_descsz = descsz;
 343         note.n_type = type;
 344         if (dst != NULL)
 345                 bcopy(&note, (char *)dst + *off, sizeof note);
 346         *off += sizeof note;
 347         if (dst != NULL)
 348                 bcopy(name, (char *)dst + *off, note.n_namesz);
 349         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 350         if (dst != NULL)
 351                 bcopy(desc, (char *)dst + *off, note.n_descsz);
 352         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 353 }
 354
 355 /*
 356  * Free the memory map.
 357  */
 358 static void
 359 freemap(vm_map_entry_t map)
 360 {
 361         while (map != NULL) {
 362                 vm_map_entry_t next = map->rb_entry.rbe_parent;
 363                 free(map);
 364                 map = next;
 365         }
 366 }
 367
 368 /*
 369  * Read the process information necessary to fill in the core file's header.
 370  */
 371 static void
 372 readhdrinfo(pid_t pid, prstatus_t *status, prfpregset_t *fpregset,
 373     prpsinfo_t *psinfo)
 374 {
 375         char name[64];
 376         char line[256];
 377         int fd;
 378         int i;
 379         int n;
 380
 381         memset(status, 0, sizeof *status);
 382         status->pr_version = PRSTATUS_VERSION;
 383         status->pr_statussz = sizeof(prstatus_t);
 384         status->pr_gregsetsz = sizeof(gregset_t);
 385         status->pr_fpregsetsz = sizeof(fpregset_t);
 386         status->pr_osreldate = __DragonFly_version;
 387         status->pr_pid = pid;
 388
 389         memset(fpregset, 0, sizeof *fpregset);
 390
 391         memset(psinfo, 0, sizeof *psinfo);
 392         psinfo->pr_version = PRPSINFO_VERSION;
 393         psinfo->pr_psinfosz = sizeof(prpsinfo_t);
 394
 395         /* Read the general registers. */
 396         snprintf(name, sizeof name, "/proc/%d/regs", pid);
 397         if ((fd = open(name, O_RDONLY)) == -1)
 398                 err(1, "cannot open %s", name);
 399         if ((n = read(fd, &status->pr_reg, sizeof status->pr_reg)) == -1)
 400                 err(1, "read error from %s", name);
 401         if (n < sizeof status->pr_reg)
 402                 errx(1, "short read from %s: wanted %zu, got %d",
 403                      name, sizeof(status->pr_reg), n);
 404         close(fd);
 405
 406         /* Read the floating point registers. */
 407         snprintf(name, sizeof name, "/proc/%d/fpregs", pid);
 408         if ((fd = open(name, O_RDONLY)) == -1)
 409                 err(1, "cannot open %s", name);
 410         if ((n = read(fd, fpregset, sizeof *fpregset)) == -1)
 411                 err(1, "read error from %s", name);
 412         if (n < sizeof *fpregset)
 413                 errx(1, "short read from %s: wanted %zu, got %d",
 414                      name, sizeof(*fpregset), n);
 415         close(fd);
 416
 417         /* Read and parse the process status. */
 418         snprintf(name, sizeof name, "/proc/%d/status", pid);
 419         if ((fd = open(name, O_RDONLY)) == -1)
 420                 err(1, "cannot open %s", name);
 421         if ((n = read(fd, line, sizeof line - 1)) == -1)
 422                 err(1, "read error from %s", name);
 423         if (n > MAXCOMLEN)
 424                 n = MAXCOMLEN;
 425         for (i = 0;  i < n && line[i] != ' ';  i++)
 426                 psinfo->pr_fname[i] = line[i];
 427         strncpy(psinfo->pr_psargs, psinfo->pr_fname, PRARGSZ);
 428         close(fd);
 429 }
 430
 431 /*
 432  * Read the process's memory map using procfs, and return a list of
 433  * VM map entries.  Only the non-device read/writable segments are
 434  * returned.  The map entries in the list aren't fully filled in; only
 435  * the items we need are present.
 436  */
 437 static vm_map_entry_t
 438 readmap(pid_t pid)
 439 {
 440         char mapname[64];
 441         int mapfd;
 442         ssize_t mapsize;
 443         size_t bufsize;
 444         char *mapbuf;
 445         int pos;
 446         vm_map_entry_t map;
 447         vm_map_entry_t *linkp;
 448
 449         snprintf(mapname, sizeof mapname, "/proc/%d/map", pid);
 450         if ((mapfd = open(mapname, O_RDONLY)) == -1)
 451                 err(1, "cannot open %s", mapname);
 452
 453         /*
 454          * Procfs requires (for consistency) that the entire memory map
 455          * be read with a single read() call.  Start with a reasonbly sized
 456          * buffer, and double it until it is big enough.
 457          */
 458         bufsize = 65536;
 459         mapbuf = NULL;
 460         for ( ; ; ) {
 461                 if ((mapbuf = realloc(mapbuf, bufsize + 1)) == NULL)
 462                         errx(1, "out of memory");
 463                 mapsize = read(mapfd, mapbuf, bufsize);
 464                 if ((mapsize != -1 || errno != EFBIG) && mapsize != bufsize) {
 465                         break;
 466                 }
 467                 bufsize *= 2;
 468                 /* This lseek shouldn't be necessary, but it is. */
 469                 lseek(mapfd, (off_t)0, SEEK_SET);
 470         }
 471         if (mapsize == -1)
 472                 err(1, "read error from %s", mapname);
 473         if (mapsize == 0)
 474                 errx(1, "empty map file %s", mapname);
 475         mapbuf[mapsize] = 0;
 476         close(mapfd);
 477
 478         pos = 0;
 479         map = NULL;
 480         linkp = &map;
 481         while (pos < mapsize) {
 482                 vm_map_entry_t ent;
 483                 vm_offset_t start;
 484                 vm_offset_t end;
 485                 char prot[4];
 486                 char type[16];
 487                 int n;
 488                 int len;
 489                 int skipme = 0;
 490
 491                 len = 0;
 492                 n = sscanf(mapbuf + pos, "%lx %lx %*d %*d %*x %3[-rwx]"
 493                     " %*d %*d %*x %*s %*s %15s %*s%*[\n]%n",
 494                     &start, &end, prot, type, &len);
 495                 if (n != 4)
 496                         errx(1, "ill-formed line in %s: '%s'",
 497                             mapname, mapbuf);
 498                 pos += len;
 499
 500                 /* Ignore segments of the wrong kind, and unwritable ones */
 501                 if (gcore_seg_limit >= 0 && end - start > gcore_seg_limit) {
 502                         skipme = 1;
 503                 }
 504                 if (strncmp(prot, "rw", 2) != 0 ||
 505                     (strcmp(type, "default") != 0 &&
 506                     strcmp(type, "vnode") != 0 &&
 507                     strcmp(type, "swap") != 0)) {
 508                         skipme = 2;
 509                 }
 510                 if (gcore_verbose || skipme == 1)
 511                         printf("%016lx-%016lx (%ldM) %s,%s",
 512                                start, end,
 513                                 (end - start) / (1024 * 1024),
 514                                 prot, type);
 515                 if (skipme) {
 516                         if (skipme == 1)
 517                                 printf(" (ignored - seglimit)\n");
 518                         else if (gcore_verbose)
 519                                 printf(" (ignored)\n");
 520                         continue;
 521                 }
 522                 if (gcore_verbose)
 523                         printf("\n");
 524
 525                 if ((ent = (vm_map_entry_t)calloc(1, sizeof *ent)) == NULL)
 526                         errx(1, "out of memory");
 527                 ent->ba.start = start;
 528                 ent->ba.end = end;
 529                 ent->protection = VM_PROT_READ | VM_PROT_WRITE;
 530                 if (prot[2] == 'x')
 531                     ent->protection |= VM_PROT_EXECUTE;
 532
 533                 *linkp = ent;
 534                 linkp = &ent->rb_entry.rbe_parent;
 535         }
 536         free(mapbuf);
 537         return map;
 538 }