1 /*- 2 * Copyright (c) 2007 Sandvine Incorporated 3 * Copyright (c) 1998 John D. Polstra 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 #include <sys/cdefs.h> 29 __FBSDID("$FreeBSD$"); 30 31 #include <sys/param.h> 32 #include <sys/procfs.h> 33 #include <sys/ptrace.h> 34 #include <sys/queue.h> 35 #include <sys/linker_set.h> 36 #include <sys/sysctl.h> 37 #include <sys/user.h> 38 #include <sys/wait.h> 39 #include <machine/elf.h> 40 #include <vm/vm_param.h> 41 #include <vm/vm.h> 42 #include <vm/pmap.h> 43 #include <vm/vm_map.h> 44 #include <err.h> 45 #include <errno.h> 46 #include <fcntl.h> 47 #include <stdint.h> 48 #include <stdio.h> 49 #include <stdlib.h> 50 #include <string.h> 51 #include <unistd.h> 52 #include <libutil.h> 53 54 #include "extern.h" 55 56 /* 57 * Code for generating ELF core dumps. 58 */ 59 60 typedef void (*segment_callback)(vm_map_entry_t, void *); 61 62 /* Closure for cb_put_phdr(). */ 63 struct phdr_closure { 64 Elf_Phdr *phdr; /* Program header to fill in */ 65 Elf_Off offset; /* Offset of segment in core file */ 66 }; 67 68 /* Closure for cb_size_segment(). */ 69 struct sseg_closure { 70 int count; /* Count of writable segments. */ 71 size_t size; /* Total size of all writable segments. */ 72 }; 73 74 static void cb_put_phdr(vm_map_entry_t, void *); 75 static void cb_size_segment(vm_map_entry_t, void *); 76 static void each_writable_segment(vm_map_entry_t, segment_callback, 77 void *closure); 78 static void elf_detach(void); /* atexit() handler. */ 79 static void elf_puthdr(pid_t, vm_map_entry_t, void *, size_t *, int numsegs); 80 static void elf_putnote(void *dst, size_t *off, const char *name, int type, 81 const void *desc, size_t descsz); 82 static void freemap(vm_map_entry_t); 83 static vm_map_entry_t readmap(pid_t); 84 85 static pid_t g_pid; /* Pid being dumped, global for elf_detach */ 86 87 static int 88 elf_ident(int efd, pid_t pid __unused, char *binfile __unused) 89 { 90 Elf_Ehdr hdr; 91 int cnt; 92 93 cnt = read(efd, &hdr, sizeof(hdr)); 94 if (cnt != sizeof(hdr)) 95 return (0); 96 if (IS_ELF(hdr)) 97 return (1); 98 return (0); 99 } 100 101 static void 102 elf_detach(void) 103 { 104 105 if (g_pid != 0) 106 ptrace(PT_DETACH, g_pid, (caddr_t)1, 0); 107 } 108 109 /* 110 * Write an ELF coredump for the given pid to the given fd. 111 */ 112 static void 113 elf_coredump(int efd __unused, int fd, pid_t pid) 114 { 115 vm_map_entry_t map; 116 struct sseg_closure seginfo; 117 void *hdr; 118 size_t hdrsize; 119 Elf_Phdr *php; 120 int i; 121 122 /* Attach to process to dump. */ 123 g_pid = pid; 124 if (atexit(elf_detach) != 0) 125 err(1, "atexit"); 126 errno = 0; 127 ptrace(PT_ATTACH, pid, NULL, 0); 128 if (errno) 129 err(1, "PT_ATTACH"); 130 if (waitpid(pid, NULL, 0) == -1) 131 err(1, "waitpid"); 132 133 /* Get the program's memory map. */ 134 map = readmap(pid); 135 136 /* Size the program segments. */ 137 seginfo.count = 0; 138 seginfo.size = 0; 139 each_writable_segment(map, cb_size_segment, &seginfo); 140 141 /* 142 * Calculate the size of the core file header area by making 143 * a dry run of generating it. Nothing is written, but the 144 * size is calculated. 145 */ 146 hdrsize = 0; 147 elf_puthdr(pid, map, NULL, &hdrsize, seginfo.count); 148 149 /* 150 * Allocate memory for building the header, fill it up, 151 * and write it out. 152 */ 153 if ((hdr = calloc(1, hdrsize)) == NULL) 154 errx(1, "out of memory"); 155 156 /* Fill in the header. */ 157 hdrsize = 0; 158 elf_puthdr(pid, map, hdr, &hdrsize, seginfo.count); 159 160 /* Write it to the core file. */ 161 if (write(fd, hdr, hdrsize) == -1) 162 err(1, "write"); 163 164 /* Write the contents of all of the writable segments. */ 165 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1; 166 for (i = 0; i < seginfo.count; i++) { 167 struct ptrace_io_desc iorequest; 168 uintmax_t nleft = php->p_filesz; 169 170 iorequest.piod_op = PIOD_READ_D; 171 iorequest.piod_offs = (caddr_t)php->p_vaddr; 172 while (nleft > 0) { 173 char buf[8*1024]; 174 size_t nwant; 175 ssize_t ngot; 176 177 if (nleft > sizeof(buf)) 178 nwant = sizeof buf; 179 else 180 nwant = nleft; 181 iorequest.piod_addr = buf; 182 iorequest.piod_len = nwant; 183 ptrace(PT_IO, pid, (caddr_t)&iorequest, 0); 184 ngot = iorequest.piod_len; 185 if ((size_t)ngot < nwant) 186 errx(1, "short read wanted %d, got %d", 187 nwant, ngot); 188 ngot = write(fd, buf, nwant); 189 if (ngot == -1) 190 err(1, "write of segment %d failed", i); 191 if ((size_t)ngot != nwant) 192 errx(1, "short write"); 193 nleft -= nwant; 194 iorequest.piod_offs += ngot; 195 } 196 php++; 197 } 198 free(hdr); 199 freemap(map); 200 } 201 202 /* 203 * A callback for each_writable_segment() to write out the segment's 204 * program header entry. 205 */ 206 static void 207 cb_put_phdr(vm_map_entry_t entry, void *closure) 208 { 209 struct phdr_closure *phc = (struct phdr_closure *)closure; 210 Elf_Phdr *phdr = phc->phdr; 211 212 phc->offset = round_page(phc->offset); 213 214 phdr->p_type = PT_LOAD; 215 phdr->p_offset = phc->offset; 216 phdr->p_vaddr = entry->start; 217 phdr->p_paddr = 0; 218 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start; 219 phdr->p_align = PAGE_SIZE; 220 phdr->p_flags = 0; 221 if (entry->protection & VM_PROT_READ) 222 phdr->p_flags |= PF_R; 223 if (entry->protection & VM_PROT_WRITE) 224 phdr->p_flags |= PF_W; 225 if (entry->protection & VM_PROT_EXECUTE) 226 phdr->p_flags |= PF_X; 227 228 phc->offset += phdr->p_filesz; 229 phc->phdr++; 230 } 231 232 /* 233 * A callback for each_writable_segment() to gather information about 234 * the number of segments and their total size. 235 */ 236 static void 237 cb_size_segment(vm_map_entry_t entry, void *closure) 238 { 239 struct sseg_closure *ssc = (struct sseg_closure *)closure; 240 241 ssc->count++; 242 ssc->size += entry->end - entry->start; 243 } 244 245 /* 246 * For each segment in the given memory map, call the given function 247 * with a pointer to the map entry and some arbitrary caller-supplied 248 * data. 249 */ 250 static void 251 each_writable_segment(vm_map_entry_t map, segment_callback func, void *closure) 252 { 253 vm_map_entry_t entry; 254 255 for (entry = map; entry != NULL; entry = entry->next) 256 (*func)(entry, closure); 257 } 258 259 static void 260 elf_getstatus(pid_t pid, prpsinfo_t *psinfo) 261 { 262 struct kinfo_proc kobj; 263 int name[4]; 264 size_t len; 265 266 name[0] = CTL_KERN; 267 name[1] = KERN_PROC; 268 name[2] = KERN_PROC_PID; 269 name[3] = pid; 270 271 len = sizeof(kobj); 272 if (sysctl(name, 4, &kobj, &len, NULL, 0) == -1) 273 err(1, "error accessing kern.proc.pid.%u sysctl", pid); 274 if (kobj.ki_pid != pid) 275 err(1, "error accessing kern.proc.pid.%u sysctl datas", pid); 276 strncpy(psinfo->pr_fname, kobj.ki_comm, MAXCOMLEN); 277 strncpy(psinfo->pr_psargs, psinfo->pr_fname, PRARGSZ); 278 } 279 280 /* 281 * Generate the ELF coredump header into the buffer at "dst". "dst" may 282 * be NULL, in which case the header is sized but not actually generated. 283 */ 284 static void 285 elf_puthdr(pid_t pid, vm_map_entry_t map, void *dst, size_t *off, int numsegs) 286 { 287 struct ptrace_lwpinfo lwpinfo; 288 struct { 289 prstatus_t status; 290 prfpregset_t fpregset; 291 prpsinfo_t psinfo; 292 thrmisc_t thrmisc; 293 } *tempdata; 294 size_t ehoff; 295 size_t phoff; 296 size_t noteoff; 297 size_t notesz; 298 size_t threads; 299 lwpid_t *tids; 300 int i; 301 302 prstatus_t *status; 303 prfpregset_t *fpregset; 304 prpsinfo_t *psinfo; 305 thrmisc_t *thrmisc; 306 307 ehoff = *off; 308 *off += sizeof(Elf_Ehdr); 309 310 phoff = *off; 311 *off += (numsegs + 1) * sizeof(Elf_Phdr); 312 313 noteoff = *off; 314 315 if (dst != NULL) { 316 if ((tempdata = calloc(1, sizeof(*tempdata))) == NULL) 317 errx(1, "out of memory"); 318 status = &tempdata->status; 319 fpregset = &tempdata->fpregset; 320 psinfo = &tempdata->psinfo; 321 thrmisc = &tempdata->thrmisc; 322 } else { 323 tempdata = NULL; 324 status = NULL; 325 fpregset = NULL; 326 psinfo = NULL; 327 thrmisc = NULL; 328 } 329 330 errno = 0; 331 threads = ptrace(PT_GETNUMLWPS, pid, NULL, 0); 332 if (errno) 333 err(1, "PT_GETNUMLWPS"); 334 335 if (dst != NULL) { 336 psinfo->pr_version = PRPSINFO_VERSION; 337 psinfo->pr_psinfosz = sizeof(prpsinfo_t); 338 elf_getstatus(pid, psinfo); 339 340 } 341 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo, 342 sizeof *psinfo); 343 344 if (dst != NULL) { 345 tids = malloc(threads * sizeof(*tids)); 346 if (tids == NULL) 347 errx(1, "out of memory"); 348 errno = 0; 349 ptrace(PT_GETLWPLIST, pid, (void *)tids, threads); 350 if (errno) 351 err(1, "PT_GETLWPLIST"); 352 } 353 for (i = 0; i < threads; ++i) { 354 if (dst != NULL) { 355 status->pr_version = PRSTATUS_VERSION; 356 status->pr_statussz = sizeof(prstatus_t); 357 status->pr_gregsetsz = sizeof(gregset_t); 358 status->pr_fpregsetsz = sizeof(fpregset_t); 359 status->pr_osreldate = __FreeBSD_version; 360 status->pr_pid = tids[i]; 361 362 ptrace(PT_GETREGS, tids[i], (void *)&status->pr_reg, 0); 363 ptrace(PT_GETFPREGS, tids[i], (void *)fpregset, 0); 364 ptrace(PT_LWPINFO, tids[i], (void *)&lwpinfo, 365 sizeof(lwpinfo)); 366 memset(&thrmisc->_pad, 0, sizeof(thrmisc->_pad)); 367 strcpy(thrmisc->pr_tname, lwpinfo.pl_tdname); 368 } 369 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status, 370 sizeof *status); 371 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset, 372 sizeof *fpregset); 373 elf_putnote(dst, off, "FreeBSD", NT_THRMISC, thrmisc, 374 sizeof *thrmisc); 375 } 376 377 notesz = *off - noteoff; 378 379 if (dst != NULL) { 380 free(tids); 381 free(tempdata); 382 } 383 384 /* Align up to a page boundary for the program segments. */ 385 *off = round_page(*off); 386 387 if (dst != NULL) { 388 Elf_Ehdr *ehdr; 389 Elf_Phdr *phdr; 390 struct phdr_closure phc; 391 392 /* 393 * Fill in the ELF header. 394 */ 395 ehdr = (Elf_Ehdr *)((char *)dst + ehoff); 396 ehdr->e_ident[EI_MAG0] = ELFMAG0; 397 ehdr->e_ident[EI_MAG1] = ELFMAG1; 398 ehdr->e_ident[EI_MAG2] = ELFMAG2; 399 ehdr->e_ident[EI_MAG3] = ELFMAG3; 400 ehdr->e_ident[EI_CLASS] = ELF_CLASS; 401 ehdr->e_ident[EI_DATA] = ELF_DATA; 402 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 403 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD; 404 ehdr->e_ident[EI_ABIVERSION] = 0; 405 ehdr->e_ident[EI_PAD] = 0; 406 ehdr->e_type = ET_CORE; 407 ehdr->e_machine = ELF_ARCH; 408 ehdr->e_version = EV_CURRENT; 409 ehdr->e_entry = 0; 410 ehdr->e_phoff = phoff; 411 ehdr->e_flags = 0; 412 ehdr->e_ehsize = sizeof(Elf_Ehdr); 413 ehdr->e_phentsize = sizeof(Elf_Phdr); 414 ehdr->e_phnum = numsegs + 1; 415 ehdr->e_shentsize = sizeof(Elf_Shdr); 416 ehdr->e_shnum = 0; 417 ehdr->e_shstrndx = SHN_UNDEF; 418 419 /* 420 * Fill in the program header entries. 421 */ 422 phdr = (Elf_Phdr *)((char *)dst + phoff); 423 424 /* The note segment. */ 425 phdr->p_type = PT_NOTE; 426 phdr->p_offset = noteoff; 427 phdr->p_vaddr = 0; 428 phdr->p_paddr = 0; 429 phdr->p_filesz = notesz; 430 phdr->p_memsz = 0; 431 phdr->p_flags = 0; 432 phdr->p_align = 0; 433 phdr++; 434 435 /* All the writable segments from the program. */ 436 phc.phdr = phdr; 437 phc.offset = *off; 438 each_writable_segment(map, cb_put_phdr, &phc); 439 } 440 } 441 442 /* 443 * Emit one note section to "dst", or just size it if "dst" is NULL. 444 */ 445 static void 446 elf_putnote(void *dst, size_t *off, const char *name, int type, 447 const void *desc, size_t descsz) 448 { 449 Elf_Note note; 450 451 note.n_namesz = strlen(name) + 1; 452 note.n_descsz = descsz; 453 note.n_type = type; 454 if (dst != NULL) 455 bcopy(¬e, (char *)dst + *off, sizeof note); 456 *off += sizeof note; 457 if (dst != NULL) 458 bcopy(name, (char *)dst + *off, note.n_namesz); 459 *off += roundup2(note.n_namesz, sizeof(Elf_Size)); 460 if (dst != NULL) 461 bcopy(desc, (char *)dst + *off, note.n_descsz); 462 *off += roundup2(note.n_descsz, sizeof(Elf_Size)); 463 } 464 465 /* 466 * Free the memory map. 467 */ 468 static void 469 freemap(vm_map_entry_t map) 470 { 471 472 while (map != NULL) { 473 vm_map_entry_t next = map->next; 474 free(map); 475 map = next; 476 } 477 } 478 479 /* 480 * Read the process's memory map using kinfo_getvmmap(), and return a list of 481 * VM map entries. Only the non-device read/writable segments are 482 * returned. The map entries in the list aren't fully filled in; only 483 * the items we need are present. 484 */ 485 static vm_map_entry_t 486 readmap(pid_t pid) 487 { 488 vm_map_entry_t ent, *linkp, map; 489 struct kinfo_vmentry *vmentl, *kve; 490 int i, nitems; 491 492 vmentl = kinfo_getvmmap(pid, &nitems); 493 if (vmentl == NULL) 494 err(1, "cannot retrieve mappings for %u process", pid); 495 496 map = NULL; 497 linkp = ↦ 498 for (i = 0; i < nitems; i++) { 499 kve = &vmentl[i]; 500 501 /* 502 * Ignore 'malformed' segments or ones representing memory 503 * mapping with MAP_NOCORE on. 504 * If the 'full' support is disabled, just dump the most 505 * meaningful data segments. 506 */ 507 if ((kve->kve_protection & KVME_PROT_READ) == 0 || 508 (kve->kve_flags & KVME_FLAG_NOCOREDUMP) != 0 || 509 kve->kve_type == KVME_TYPE_DEAD || 510 kve->kve_type == KVME_TYPE_UNKNOWN || 511 ((pflags & PFLAGS_FULL) == 0 && 512 kve->kve_type != KVME_TYPE_DEFAULT && 513 kve->kve_type != KVME_TYPE_VNODE && 514 kve->kve_type != KVME_TYPE_SWAP)) 515 continue; 516 517 ent = calloc(1, sizeof(*ent)); 518 if (ent == NULL) 519 errx(1, "out of memory"); 520 ent->start = (vm_offset_t)kve->kve_start; 521 ent->end = (vm_offset_t)kve->kve_end; 522 ent->protection = VM_PROT_READ | VM_PROT_WRITE; 523 if ((kve->kve_protection & KVME_PROT_EXEC) != 0) 524 ent->protection |= VM_PROT_EXECUTE; 525 526 *linkp = ent; 527 linkp = &ent->next; 528 } 529 free(vmentl); 530 return (map); 531 } 532 533 struct dumpers elfdump = { elf_ident, elf_coredump }; 534 TEXT_SET(dumpset, elfdump); 535