sys/kern/imgact_elf.c

   1 /*-
   2  * Copyright (c) 2000 David O'Brien
   3  * Copyright (c) 1995-1996 Søren Schmidt
   4  * Copyright (c) 1996 Peter Wemm
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer
  12  *    in this position and unchanged.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. The name of the author may not be used to endorse or promote products
  17  *    derived from this software without specific prior written permission
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29  */
  30
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include "opt_capsicum.h"
  35 #include "opt_compat.h"
  36 #include "opt_core.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/capsicum.h>
  40 #include <sys/exec.h>
  41 #include <sys/fcntl.h>
  42 #include <sys/imgact.h>
  43 #include <sys/imgact_elf.h>
  44 #include <sys/jail.h>
  45 #include <sys/kernel.h>
  46 #include <sys/lock.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mount.h>
  49 #include <sys/mman.h>
  50 #include <sys/namei.h>
  51 #include <sys/pioctl.h>
  52 #include <sys/proc.h>
  53 #include <sys/procfs.h>
  54 #include <sys/racct.h>
  55 #include <sys/resourcevar.h>
  56 #include <sys/rwlock.h>
  57 #include <sys/sbuf.h>
  58 #include <sys/sf_buf.h>
  59 #include <sys/smp.h>
  60 #include <sys/systm.h>
  61 #include <sys/signalvar.h>
  62 #include <sys/stat.h>
  63 #include <sys/sx.h>
  64 #include <sys/syscall.h>
  65 #include <sys/sysctl.h>
  66 #include <sys/sysent.h>
  67 #include <sys/vnode.h>
  68 #include <sys/syslog.h>
  69 #include <sys/eventhandler.h>
  70 #include <sys/user.h>
  71
  72 #include <net/zlib.h>
  73
  74 #include <vm/vm.h>
  75 #include <vm/vm_kern.h>
  76 #include <vm/vm_param.h>
  77 #include <vm/pmap.h>
  78 #include <vm/vm_map.h>
  79 #include <vm/vm_object.h>
  80 #include <vm/vm_extern.h>
  81
  82 #include <machine/elf.h>
  83 #include <machine/md_var.h>
  84
  85 #define ELF_NOTE_ROUNDSIZE      4
  86 #define OLD_EI_BRAND    8
  87
  88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
  89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
  90     const char *interp, int interp_name_len, int32_t *osrel);
  91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  92     u_long *entry, size_t pagesize);
  93 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
  94     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
  95     size_t pagesize);
  96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
  97 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
  98     int32_t *osrel);
  99 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
 100 static boolean_t __elfN(check_note)(struct image_params *imgp,
 101     Elf_Brandnote *checknote, int32_t *osrel);
 102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
 103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
 104
 105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
 106     "");
 107
 108 #ifdef COMPRESS_USER_CORES
 109 static int compress_core(gzFile, char *, char *, unsigned int,
 110     struct thread * td);
 111 #endif
 112 #define CORE_BUF_SIZE   (16 * 1024)
 113
 114 int __elfN(fallback_brand) = -1;
 115 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
 116     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
 117     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
 118 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
 119     &__elfN(fallback_brand));
 120
 121 static int elf_legacy_coredump = 0;
 122 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
 123     &elf_legacy_coredump, 0, "");
 124
 125 int __elfN(nxstack) =
 126 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
 127         1;
 128 #else
 129         0;
 130 #endif
 131 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
 132     nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
 133     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
 134
 135 #if __ELF_WORD_SIZE == 32
 136 #if defined(__amd64__) || defined(__ia64__)
 137 int i386_read_exec = 0;
 138 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
 139     "enable execution from readable segments");
 140 #endif
 141 #endif
 142
 143 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
 144
 145 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
 146 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
 147 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
 148
 149 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
 150
 151 Elf_Brandnote __elfN(freebsd_brandnote) = {
 152         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
 153         .hdr.n_descsz   = sizeof(int32_t),
 154         .hdr.n_type     = 1,
 155         .vendor         = FREEBSD_ABI_VENDOR,
 156         .flags          = BN_TRANSLATE_OSREL,
 157         .trans_osrel    = __elfN(freebsd_trans_osrel)
 158 };
 159
 160 static boolean_t
 161 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
 162 {
 163         uintptr_t p;
 164
 165         p = (uintptr_t)(note + 1);
 166         p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
 167         *osrel = *(const int32_t *)(p);
 168
 169         return (TRUE);
 170 }
 171
 172 static const char GNU_ABI_VENDOR[] = "GNU";
 173 static int GNU_KFREEBSD_ABI_DESC = 3;
 174
 175 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
 176         .hdr.n_namesz   = sizeof(GNU_ABI_VENDOR),
 177         .hdr.n_descsz   = 16,   /* XXX at least 16 */
 178         .hdr.n_type     = 1,
 179         .vendor         = GNU_ABI_VENDOR,
 180         .flags          = BN_TRANSLATE_OSREL,
 181         .trans_osrel    = kfreebsd_trans_osrel
 182 };
 183
 184 static boolean_t
 185 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
 186 {
 187         const Elf32_Word *desc;
 188         uintptr_t p;
 189
 190         p = (uintptr_t)(note + 1);
 191         p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
 192
 193         desc = (const Elf32_Word *)p;
 194         if (desc[0] != GNU_KFREEBSD_ABI_DESC)
 195                 return (FALSE);
 196
 197         /*
 198          * Debian GNU/kFreeBSD embed the earliest compatible kernel version
 199          * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
 200          */
 201         *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
 202
 203         return (TRUE);
 204 }
 205
 206 int
 207 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
 208 {
 209         int i;
 210
 211         for (i = 0; i < MAX_BRANDS; i++) {
 212                 if (elf_brand_list[i] == NULL) {
 213                         elf_brand_list[i] = entry;
 214                         break;
 215                 }
 216         }
 217         if (i == MAX_BRANDS) {
 218                 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
 219                         __func__, entry);
 220                 return (-1);
 221         }
 222         return (0);
 223 }
 224
 225 int
 226 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
 227 {
 228         int i;
 229
 230         for (i = 0; i < MAX_BRANDS; i++) {
 231                 if (elf_brand_list[i] == entry) {
 232                         elf_brand_list[i] = NULL;
 233                         break;
 234                 }
 235         }
 236         if (i == MAX_BRANDS)
 237                 return (-1);
 238         return (0);
 239 }
 240
 241 int
 242 __elfN(brand_inuse)(Elf_Brandinfo *entry)
 243 {
 244         struct proc *p;
 245         int rval = FALSE;
 246
 247         sx_slock(&allproc_lock);
 248         FOREACH_PROC_IN_SYSTEM(p) {
 249                 if (p->p_sysent == entry->sysvec) {
 250                         rval = TRUE;
 251                         break;
 252                 }
 253         }
 254         sx_sunlock(&allproc_lock);
 255
 256         return (rval);
 257 }
 258
 259 static Elf_Brandinfo *
 260 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
 261     int interp_name_len, int32_t *osrel)
 262 {
 263         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
 264         Elf_Brandinfo *bi, *bi_m;
 265         boolean_t ret;
 266         int i;
 267
 268         /*
 269          * We support four types of branding -- (1) the ELF EI_OSABI field
 270          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
 271          * branding w/in the ELF header, (3) path of the `interp_path'
 272          * field, and (4) the ".note.ABI-tag" ELF section.
 273          */
 274
 275         /* Look for an ".note.ABI-tag" ELF section */
 276         bi_m = NULL;
 277         for (i = 0; i < MAX_BRANDS; i++) {
 278                 bi = elf_brand_list[i];
 279                 if (bi == NULL)
 280                         continue;
 281                 if (hdr->e_machine == bi->machine && (bi->flags &
 282                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
 283                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
 284                         /*
 285                          * If note checker claimed the binary, but the
 286                          * interpreter path in the image does not
 287                          * match default one for the brand, try to
 288                          * search for other brands with the same
 289                          * interpreter.  Either there is better brand
 290                          * with the right interpreter, or, failing
 291                          * this, we return first brand which accepted
 292                          * our note and, optionally, header.
 293                          */
 294                         if (ret && bi_m == NULL && (strlen(bi->interp_path) +
 295                             1 != interp_name_len || strncmp(interp,
 296                             bi->interp_path, interp_name_len) != 0)) {
 297                                 bi_m = bi;
 298                                 ret = 0;
 299                         }
 300                         if (ret)
 301                                 return (bi);
 302                 }
 303         }
 304         if (bi_m != NULL)
 305                 return (bi_m);
 306
 307         /* If the executable has a brand, search for it in the brand list. */
 308         for (i = 0; i < MAX_BRANDS; i++) {
 309                 bi = elf_brand_list[i];
 310                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 311                         continue;
 312                 if (hdr->e_machine == bi->machine &&
 313                     (hdr->e_ident[EI_OSABI] == bi->brand ||
 314                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
 315                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
 316                         return (bi);
 317         }
 318
 319         /* Lacking a known brand, search for a recognized interpreter. */
 320         if (interp != NULL) {
 321                 for (i = 0; i < MAX_BRANDS; i++) {
 322                         bi = elf_brand_list[i];
 323                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 324                                 continue;
 325                         if (hdr->e_machine == bi->machine &&
 326                             /* ELF image p_filesz includes terminating zero */
 327                             strlen(bi->interp_path) + 1 == interp_name_len &&
 328                             strncmp(interp, bi->interp_path, interp_name_len)
 329                             == 0)
 330                                 return (bi);
 331                 }
 332         }
 333
 334         /* Lacking a recognized interpreter, try the default brand */
 335         for (i = 0; i < MAX_BRANDS; i++) {
 336                 bi = elf_brand_list[i];
 337                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
 338                         continue;
 339                 if (hdr->e_machine == bi->machine &&
 340                     __elfN(fallback_brand) == bi->brand)
 341                         return (bi);
 342         }
 343         return (NULL);
 344 }
 345
 346 static int
 347 __elfN(check_header)(const Elf_Ehdr *hdr)
 348 {
 349         Elf_Brandinfo *bi;
 350         int i;
 351
 352         if (!IS_ELF(*hdr) ||
 353             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
 354             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
 355             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
 356             hdr->e_phentsize != sizeof(Elf_Phdr) ||
 357             hdr->e_version != ELF_TARG_VER)
 358                 return (ENOEXEC);
 359
 360         /*
 361          * Make sure we have at least one brand for this machine.
 362          */
 363
 364         for (i = 0; i < MAX_BRANDS; i++) {
 365                 bi = elf_brand_list[i];
 366                 if (bi != NULL && bi->machine == hdr->e_machine)
 367                         break;
 368         }
 369         if (i == MAX_BRANDS)
 370                 return (ENOEXEC);
 371
 372         return (0);
 373 }
 374
 375 static int
 376 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 377     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
 378 {
 379         struct sf_buf *sf;
 380         int error;
 381         vm_offset_t off;
 382
 383         /*
 384          * Create the page if it doesn't exist yet. Ignore errors.
 385          */
 386         vm_map_lock(map);
 387         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
 388             VM_PROT_ALL, VM_PROT_ALL, 0);
 389         vm_map_unlock(map);
 390
 391         /*
 392          * Find the page from the underlying object.
 393          */
 394         if (object) {
 395                 sf = vm_imgact_map_page(object, offset);
 396                 if (sf == NULL)
 397                         return (KERN_FAILURE);
 398                 off = offset - trunc_page(offset);
 399                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
 400                     end - start);
 401                 vm_imgact_unmap_page(sf);
 402                 if (error) {
 403                         return (KERN_FAILURE);
 404                 }
 405         }
 406
 407         return (KERN_SUCCESS);
 408 }
 409
 410 static int
 411 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
 412     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
 413 {
 414         struct sf_buf *sf;
 415         vm_offset_t off;
 416         vm_size_t sz;
 417         int error, rv;
 418
 419         if (start != trunc_page(start)) {
 420                 rv = __elfN(map_partial)(map, object, offset, start,
 421                     round_page(start), prot);
 422                 if (rv)
 423                         return (rv);
 424                 offset += round_page(start) - start;
 425                 start = round_page(start);
 426         }
 427         if (end != round_page(end)) {
 428                 rv = __elfN(map_partial)(map, object, offset +
 429                     trunc_page(end) - start, trunc_page(end), end, prot);
 430                 if (rv)
 431                         return (rv);
 432                 end = trunc_page(end);
 433         }
 434         if (end > start) {
 435                 if (offset & PAGE_MASK) {
 436                         /*
 437                          * The mapping is not page aligned. This means we have
 438                          * to copy the data. Sigh.
 439                          */
 440                         rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
 441                             VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
 442                             0);
 443                         if (rv != KERN_SUCCESS)
 444                                 return (rv);
 445                         if (object == NULL)
 446                                 return (KERN_SUCCESS);
 447                         for (; start < end; start += sz) {
 448                                 sf = vm_imgact_map_page(object, offset);
 449                                 if (sf == NULL)
 450                                         return (KERN_FAILURE);
 451                                 off = offset - trunc_page(offset);
 452                                 sz = end - start;
 453                                 if (sz > PAGE_SIZE - off)
 454                                         sz = PAGE_SIZE - off;
 455                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
 456                                     (caddr_t)start, sz);
 457                                 vm_imgact_unmap_page(sf);
 458                                 if (error != 0)
 459                                         return (KERN_FAILURE);
 460                                 offset += sz;
 461                         }
 462                         rv = KERN_SUCCESS;
 463                 } else {
 464                         vm_object_reference(object);
 465                         vm_map_lock(map);
 466                         rv = vm_map_insert(map, object, offset, start, end,
 467                             prot, VM_PROT_ALL, cow);
 468                         vm_map_unlock(map);
 469                         if (rv != KERN_SUCCESS)
 470                                 vm_object_deallocate(object);
 471                 }
 472                 return (rv);
 473         } else {
 474                 return (KERN_SUCCESS);
 475         }
 476 }
 477
 478 static int
 479 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
 480     caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
 481     size_t pagesize)
 482 {
 483         struct sf_buf *sf;
 484         size_t map_len;
 485         vm_map_t map;
 486         vm_object_t object;
 487         vm_offset_t map_addr;
 488         int error, rv, cow;
 489         size_t copy_len;
 490         vm_offset_t file_addr;
 491
 492         /*
 493          * It's necessary to fail if the filsz + offset taken from the
 494          * header is greater than the actual file pager object's size.
 495          * If we were to allow this, then the vm_map_find() below would
 496          * walk right off the end of the file object and into the ether.
 497          *
 498          * While I'm here, might as well check for something else that
 499          * is invalid: filsz cannot be greater than memsz.
 500          */
 501         if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
 502                 uprintf("elf_load_section: truncated ELF file\n");
 503                 return (ENOEXEC);
 504         }
 505
 506         object = imgp->object;
 507         map = &imgp->proc->p_vmspace->vm_map;
 508         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
 509         file_addr = trunc_page_ps(offset, pagesize);
 510
 511         /*
 512          * We have two choices.  We can either clear the data in the last page
 513          * of an oversized mapping, or we can start the anon mapping a page
 514          * early and copy the initialized data into that first page.  We
 515          * choose the second..
 516          */
 517         if (memsz > filsz)
 518                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
 519         else
 520                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
 521
 522         if (map_len != 0) {
 523                 /* cow flags: don't dump readonly sections in core */
 524                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
 525                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
 526
 527                 rv = __elfN(map_insert)(map,
 528                                       object,
 529                                       file_addr,        /* file offset */
 530                                       map_addr,         /* virtual start */
 531                                       map_addr + map_len,/* virtual end */
 532                                       prot,
 533                                       cow);
 534                 if (rv != KERN_SUCCESS)
 535                         return (EINVAL);
 536
 537                 /* we can stop now if we've covered it all */
 538                 if (memsz == filsz) {
 539                         return (0);
 540                 }
 541         }
 542
 543
 544         /*
 545          * We have to get the remaining bit of the file into the first part
 546          * of the oversized map segment.  This is normally because the .data
 547          * segment in the file is extended to provide bss.  It's a neat idea
 548          * to try and save a page, but it's a pain in the behind to implement.
 549          */
 550         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
 551         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
 552         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
 553             map_addr;
 554
 555         /* This had damn well better be true! */
 556         if (map_len != 0) {
 557                 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
 558                     map_len, VM_PROT_ALL, 0);
 559                 if (rv != KERN_SUCCESS) {
 560                         return (EINVAL);
 561                 }
 562         }
 563
 564         if (copy_len != 0) {
 565                 vm_offset_t off;
 566
 567                 sf = vm_imgact_map_page(object, offset + filsz);
 568                 if (sf == NULL)
 569                         return (EIO);
 570
 571                 /* send the page fragment to user space */
 572                 off = trunc_page_ps(offset + filsz, pagesize) -
 573                     trunc_page(offset + filsz);
 574                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
 575                     (caddr_t)map_addr, copy_len);
 576                 vm_imgact_unmap_page(sf);
 577                 if (error) {
 578                         return (error);
 579                 }
 580         }
 581
 582         /*
 583          * set it to the specified protection.
 584          * XXX had better undo the damage from pasting over the cracks here!
 585          */
 586         vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
 587             map_len), prot, FALSE);
 588
 589         return (0);
 590 }
 591
 592 /*
 593  * Load the file "file" into memory.  It may be either a shared object
 594  * or an executable.
 595  *
 596  * The "addr" reference parameter is in/out.  On entry, it specifies
 597  * the address where a shared object should be loaded.  If the file is
 598  * an executable, this value is ignored.  On exit, "addr" specifies
 599  * where the file was actually loaded.
 600  *
 601  * The "entry" reference parameter is out only.  On exit, it specifies
 602  * the entry point for the loaded file.
 603  */
 604 static int
 605 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
 606         u_long *entry, size_t pagesize)
 607 {
 608         struct {
 609                 struct nameidata nd;
 610                 struct vattr attr;
 611                 struct image_params image_params;
 612         } *tempdata;
 613         const Elf_Ehdr *hdr = NULL;
 614         const Elf_Phdr *phdr = NULL;
 615         struct nameidata *nd;
 616         struct vattr *attr;
 617         struct image_params *imgp;
 618         vm_prot_t prot;
 619         u_long rbase;
 620         u_long base_addr = 0;
 621         int error, i, numsegs;
 622
 623 #ifdef CAPABILITY_MODE
 624         /*
 625          * XXXJA: This check can go away once we are sufficiently confident
 626          * that the checks in namei() are correct.
 627          */
 628         if (IN_CAPABILITY_MODE(curthread))
 629                 return (ECAPMODE);
 630 #endif
 631
 632         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
 633         nd = &tempdata->nd;
 634         attr = &tempdata->attr;
 635         imgp = &tempdata->image_params;
 636
 637         /*
 638          * Initialize part of the common data
 639          */
 640         imgp->proc = p;
 641         imgp->attr = attr;
 642         imgp->firstpage = NULL;
 643         imgp->image_header = NULL;
 644         imgp->object = NULL;
 645         imgp->execlabel = NULL;
 646
 647         NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
 648         if ((error = namei(nd)) != 0) {
 649                 nd->ni_vp = NULL;
 650                 goto fail;
 651         }
 652         NDFREE(nd, NDF_ONLY_PNBUF);
 653         imgp->vp = nd->ni_vp;
 654
 655         /*
 656          * Check permissions, modes, uid, etc on the file, and "open" it.
 657          */
 658         error = exec_check_permissions(imgp);
 659         if (error)
 660                 goto fail;
 661
 662         error = exec_map_first_page(imgp);
 663         if (error)
 664                 goto fail;
 665
 666         /*
 667          * Also make certain that the interpreter stays the same, so set
 668          * its VV_TEXT flag, too.
 669          */
 670         VOP_SET_TEXT(nd->ni_vp);
 671
 672         imgp->object = nd->ni_vp->v_object;
 673
 674         hdr = (const Elf_Ehdr *)imgp->image_header;
 675         if ((error = __elfN(check_header)(hdr)) != 0)
 676                 goto fail;
 677         if (hdr->e_type == ET_DYN)
 678                 rbase = *addr;
 679         else if (hdr->e_type == ET_EXEC)
 680                 rbase = 0;
 681         else {
 682                 error = ENOEXEC;
 683                 goto fail;
 684         }
 685
 686         /* Only support headers that fit within first page for now      */
 687         if ((hdr->e_phoff > PAGE_SIZE) ||
 688             (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
 689                 error = ENOEXEC;
 690                 goto fail;
 691         }
 692
 693         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 694         if (!aligned(phdr, Elf_Addr)) {
 695                 error = ENOEXEC;
 696                 goto fail;
 697         }
 698
 699         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
 700                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
 701                         /* Loadable segment */
 702                         prot = __elfN(trans_prot)(phdr[i].p_flags);
 703                         error = __elfN(load_section)(imgp, phdr[i].p_offset,
 704                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
 705                             phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
 706                         if (error != 0)
 707                                 goto fail;
 708                         /*
 709                          * Establish the base address if this is the
 710                          * first segment.
 711                          */
 712                         if (numsegs == 0)
 713                                 base_addr = trunc_page(phdr[i].p_vaddr +
 714                                     rbase);
 715                         numsegs++;
 716                 }
 717         }
 718         *addr = base_addr;
 719         *entry = (unsigned long)hdr->e_entry + rbase;
 720
 721 fail:
 722         if (imgp->firstpage)
 723                 exec_unmap_first_page(imgp);
 724
 725         if (nd->ni_vp)
 726                 vput(nd->ni_vp);
 727
 728         free(tempdata, M_TEMP);
 729
 730         return (error);
 731 }
 732
 733 static int
 734 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
 735 {
 736         struct thread *td;
 737         const Elf_Ehdr *hdr;
 738         const Elf_Phdr *phdr;
 739         Elf_Auxargs *elf_auxargs;
 740         struct vmspace *vmspace;
 741         const char *err_str, *newinterp;
 742         char *interp, *interp_buf, *path;
 743         Elf_Brandinfo *brand_info;
 744         struct sysentvec *sv;
 745         vm_prot_t prot;
 746         u_long text_size, data_size, total_size, text_addr, data_addr;
 747         u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
 748         int32_t osrel;
 749         int error, i, n, interp_name_len, have_interp;
 750
 751         hdr = (const Elf_Ehdr *)imgp->image_header;
 752
 753         /*
 754          * Do we have a valid ELF header ?
 755          *
 756          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
 757          * if particular brand doesn't support it.
 758          */
 759         if (__elfN(check_header)(hdr) != 0 ||
 760             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
 761                 return (-1);
 762
 763         /*
 764          * From here on down, we return an errno, not -1, as we've
 765          * detected an ELF file.
 766          */
 767
 768         if ((hdr->e_phoff > PAGE_SIZE) ||
 769             (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
 770                 /* Only support headers in first page for now */
 771                 uprintf("Program headers not in the first page\n");
 772                 return (ENOEXEC);
 773         }
 774         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 775         if (!aligned(phdr, Elf_Addr)) {
 776                 uprintf("Unaligned program headers\n");
 777                 return (ENOEXEC);
 778         }
 779
 780         n = error = 0;
 781         baddr = 0;
 782         osrel = 0;
 783         text_size = data_size = total_size = text_addr = data_addr = 0;
 784         entry = proghdr = 0;
 785         interp_name_len = 0;
 786         err_str = newinterp = NULL;
 787         interp = interp_buf = NULL;
 788         td = curthread;
 789
 790         for (i = 0; i < hdr->e_phnum; i++) {
 791                 switch (phdr[i].p_type) {
 792                 case PT_LOAD:
 793                         if (n == 0)
 794                                 baddr = phdr[i].p_vaddr;
 795                         n++;
 796                         break;
 797                 case PT_INTERP:
 798                         /* Path to interpreter */
 799                         if (phdr[i].p_filesz > MAXPATHLEN) {
 800                                 uprintf("Invalid PT_INTERP\n");
 801                                 error = ENOEXEC;
 802                                 goto ret;
 803                         }
 804                         if (interp != NULL) {
 805                                 uprintf("Multiple PT_INTERP headers\n");
 806                                 error = ENOEXEC;
 807                                 goto ret;
 808                         }
 809                         interp_name_len = phdr[i].p_filesz;
 810                         if (phdr[i].p_offset > PAGE_SIZE ||
 811                             interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
 812                                 VOP_UNLOCK(imgp->vp, 0);
 813                                 interp_buf = malloc(interp_name_len + 1, M_TEMP,
 814                                     M_WAITOK);
 815                                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 816                                 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
 817                                     interp_name_len, phdr[i].p_offset,
 818                                     UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
 819                                     NOCRED, NULL, td);
 820                                 if (error != 0) {
 821                                         uprintf("i/o error PT_INTERP\n");
 822                                         goto ret;
 823                                 }
 824                                 interp_buf[interp_name_len] = '\0';
 825                                 interp = interp_buf;
 826                         } else {
 827                                 interp = __DECONST(char *, imgp->image_header) +
 828                                     phdr[i].p_offset;
 829                         }
 830                         break;
 831                 case PT_GNU_STACK:
 832                         if (__elfN(nxstack))
 833                                 imgp->stack_prot =
 834                                     __elfN(trans_prot)(phdr[i].p_flags);
 835                         imgp->stack_sz = phdr[i].p_memsz;
 836                         break;
 837                 }
 838         }
 839
 840         brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
 841             &osrel);
 842         if (brand_info == NULL) {
 843                 uprintf("ELF binary type \"%u\" not known.\n",
 844                     hdr->e_ident[EI_OSABI]);
 845                 error = ENOEXEC;
 846                 goto ret;
 847         }
 848         if (hdr->e_type == ET_DYN) {
 849                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
 850                         uprintf("Cannot execute shared object\n");
 851                         error = ENOEXEC;
 852                         goto ret;
 853                 }
 854                 /*
 855                  * Honour the base load address from the dso if it is
 856                  * non-zero for some reason.
 857                  */
 858                 if (baddr == 0)
 859                         et_dyn_addr = ET_DYN_LOAD_ADDR;
 860                 else
 861                         et_dyn_addr = 0;
 862         } else
 863                 et_dyn_addr = 0;
 864         sv = brand_info->sysvec;
 865         if (interp != NULL && brand_info->interp_newpath != NULL)
 866                 newinterp = brand_info->interp_newpath;
 867
 868         /*
 869          * Avoid a possible deadlock if the current address space is destroyed
 870          * and that address space maps the locked vnode.  In the common case,
 871          * the locked vnode's v_usecount is decremented but remains greater
 872          * than zero.  Consequently, the vnode lock is not needed by vrele().
 873          * However, in cases where the vnode lock is external, such as nullfs,
 874          * v_usecount may become zero.
 875          *
 876          * The VV_TEXT flag prevents modifications to the executable while
 877          * the vnode is unlocked.
 878          */
 879         VOP_UNLOCK(imgp->vp, 0);
 880
 881         error = exec_new_vmspace(imgp, sv);
 882         imgp->proc->p_sysent = sv;
 883
 884         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
 885         if (error != 0)
 886                 goto ret;
 887
 888         for (i = 0; i < hdr->e_phnum; i++) {
 889                 switch (phdr[i].p_type) {
 890                 case PT_LOAD:   /* Loadable segment */
 891                         if (phdr[i].p_memsz == 0)
 892                                 break;
 893                         prot = __elfN(trans_prot)(phdr[i].p_flags);
 894                         error = __elfN(load_section)(imgp, phdr[i].p_offset,
 895                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
 896                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
 897                             sv->sv_pagesize);
 898                         if (error != 0)
 899                                 goto ret;
 900
 901                         /*
 902                          * If this segment contains the program headers,
 903                          * remember their virtual address for the AT_PHDR
 904                          * aux entry. Static binaries don't usually include
 905                          * a PT_PHDR entry.
 906                          */
 907                         if (phdr[i].p_offset == 0 &&
 908                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
 909                                 <= phdr[i].p_filesz)
 910                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
 911                                     et_dyn_addr;
 912
 913                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
 914                         seg_size = round_page(phdr[i].p_memsz +
 915                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
 916
 917                         /*
 918                          * Make the largest executable segment the official
 919                          * text segment and all others data.
 920                          *
 921                          * Note that obreak() assumes that data_addr +
 922                          * data_size == end of data load area, and the ELF
 923                          * file format expects segments to be sorted by
 924                          * address.  If multiple data segments exist, the
 925                          * last one will be used.
 926                          */
 927
 928                         if (phdr[i].p_flags & PF_X && text_size < seg_size) {
 929                                 text_size = seg_size;
 930                                 text_addr = seg_addr;
 931                         } else {
 932                                 data_size = seg_size;
 933                                 data_addr = seg_addr;
 934                         }
 935                         total_size += seg_size;
 936                         break;
 937                 case PT_PHDR:   /* Program header table info */
 938                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
 939                         break;
 940                 default:
 941                         break;
 942                 }
 943         }
 944
 945         if (data_addr == 0 && data_size == 0) {
 946                 data_addr = text_addr;
 947                 data_size = text_size;
 948         }
 949
 950         entry = (u_long)hdr->e_entry + et_dyn_addr;
 951
 952         /*
 953          * Check limits.  It should be safe to check the
 954          * limits after loading the segments since we do
 955          * not actually fault in all the segments pages.
 956          */
 957         PROC_LOCK(imgp->proc);
 958         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA))
 959                 err_str = "Data segment size exceeds process limit";
 960         else if (text_size > maxtsiz)
 961                 err_str = "Text segment size exceeds system limit";
 962         else if (total_size > lim_cur(imgp->proc, RLIMIT_VMEM))
 963                 err_str = "Total segment size exceeds process limit";
 964         else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
 965                 err_str = "Data segment size exceeds resource limit";
 966         else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
 967                 err_str = "Total segment size exceeds resource limit";
 968         if (err_str != NULL) {
 969                 PROC_UNLOCK(imgp->proc);
 970                 uprintf("%s\n", err_str);
 971                 error = ENOMEM;
 972                 goto ret;
 973         }
 974
 975         vmspace = imgp->proc->p_vmspace;
 976         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
 977         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
 978         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
 979         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
 980
 981         /*
 982          * We load the dynamic linker where a userland call
 983          * to mmap(0, ...) would put it.  The rationale behind this
 984          * calculation is that it leaves room for the heap to grow to
 985          * its maximum allowed size.
 986          */
 987         addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
 988             RLIMIT_DATA));
 989         PROC_UNLOCK(imgp->proc);
 990
 991         imgp->entry_addr = entry;
 992
 993         if (interp != NULL) {
 994                 have_interp = FALSE;
 995                 VOP_UNLOCK(imgp->vp, 0);
 996                 if (brand_info->emul_path != NULL &&
 997                     brand_info->emul_path[0] != '\0') {
 998                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 999                         snprintf(path, MAXPATHLEN, "%s%s",
1000                             brand_info->emul_path, interp);
1001                         error = __elfN(load_file)(imgp->proc, path, &addr,
1002                             &imgp->entry_addr, sv->sv_pagesize);
1003                         free(path, M_TEMP);
1004                         if (error == 0)
1005                                 have_interp = TRUE;
1006                 }
1007                 if (!have_interp && newinterp != NULL &&
1008                     (brand_info->interp_path == NULL ||
1009                     strcmp(interp, brand_info->interp_path) == 0)) {
1010                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1011                             &imgp->entry_addr, sv->sv_pagesize);
1012                         if (error == 0)
1013                                 have_interp = TRUE;
1014                 }
1015                 if (!have_interp) {
1016                         error = __elfN(load_file)(imgp->proc, interp, &addr,
1017                             &imgp->entry_addr, sv->sv_pagesize);
1018                 }
1019                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1020                 if (error != 0) {
1021                         uprintf("ELF interpreter %s not found, error %d\n",
1022                             interp, error);
1023                         goto ret;
1024                 }
1025         } else
1026                 addr = et_dyn_addr;
1027
1028         /*
1029          * Construct auxargs table (used by the fixup routine)
1030          */
1031         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1032         elf_auxargs->execfd = -1;
1033         elf_auxargs->phdr = proghdr;
1034         elf_auxargs->phent = hdr->e_phentsize;
1035         elf_auxargs->phnum = hdr->e_phnum;
1036         elf_auxargs->pagesz = PAGE_SIZE;
1037         elf_auxargs->base = addr;
1038         elf_auxargs->flags = 0;
1039         elf_auxargs->entry = entry;
1040
1041         imgp->auxargs = elf_auxargs;
1042         imgp->interpreted = 0;
1043         imgp->reloc_base = addr;
1044         imgp->proc->p_osrel = osrel;
1045
1046  ret:
1047         free(interp_buf, M_TEMP);
1048         return (error);
1049 }
1050
1051 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1052
1053 int
1054 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1055 {
1056         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1057         Elf_Addr *base;
1058         Elf_Addr *pos;
1059
1060         base = (Elf_Addr *)*stack_base;
1061         pos = base + (imgp->args->argc + imgp->args->envc + 2);
1062
1063         if (args->execfd != -1)
1064                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1065         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1066         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1067         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1068         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1069         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1070         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1071         AUXARGS_ENTRY(pos, AT_BASE, args->base);
1072         if (imgp->execpathp != 0)
1073                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1074         AUXARGS_ENTRY(pos, AT_OSRELDATE,
1075             imgp->proc->p_ucred->cr_prison->pr_osreldate);
1076         if (imgp->canary != 0) {
1077                 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1078                 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1079         }
1080         AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1081         if (imgp->pagesizes != 0) {
1082                 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1083                 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1084         }
1085         if (imgp->sysent->sv_timekeep_base != 0) {
1086                 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1087                     imgp->sysent->sv_timekeep_base);
1088         }
1089         AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1090             != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1091             imgp->sysent->sv_stackprot);
1092         AUXARGS_ENTRY(pos, AT_NULL, 0);
1093
1094         free(imgp->auxargs, M_TEMP);
1095         imgp->auxargs = NULL;
1096
1097         base--;
1098         suword(base, (long)imgp->args->argc);
1099         *stack_base = (register_t *)base;
1100         return (0);
1101 }
1102
1103 /*
1104  * Code for generating ELF core dumps.
1105  */
1106
1107 typedef void (*segment_callback)(vm_map_entry_t, void *);
1108
1109 /* Closure for cb_put_phdr(). */
1110 struct phdr_closure {
1111         Elf_Phdr *phdr;         /* Program header to fill in */
1112         Elf_Off offset;         /* Offset of segment in core file */
1113 };
1114
1115 /* Closure for cb_size_segment(). */
1116 struct sseg_closure {
1117         int count;              /* Count of writable segments. */
1118         size_t size;            /* Total size of all writable segments. */
1119 };
1120
1121 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1122
1123 struct note_info {
1124         int             type;           /* Note type. */
1125         outfunc_t       outfunc;        /* Output function. */
1126         void            *outarg;        /* Argument for the output function. */
1127         size_t          outsize;        /* Output size. */
1128         TAILQ_ENTRY(note_info) link;    /* Link to the next note info. */
1129 };
1130
1131 TAILQ_HEAD(note_info_list, note_info);
1132
1133 static void cb_put_phdr(vm_map_entry_t, void *);
1134 static void cb_size_segment(vm_map_entry_t, void *);
1135 static void each_writable_segment(struct thread *, segment_callback, void *);
1136 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1137     int, void *, size_t, struct note_info_list *, size_t, gzFile);
1138 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1139     size_t *);
1140 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1141 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1142 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1143 static int sbuf_drain_core_output(void *, const char *, int);
1144 static int sbuf_drain_count(void *arg, const char *data, int len);
1145
1146 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1147 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1148 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1154 static void note_procstat_files(void *, struct sbuf *, size_t *);
1155 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1156 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1157 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1158 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1159 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1160
1161 #ifdef COMPRESS_USER_CORES
1162 extern int compress_user_cores;
1163 extern int compress_user_cores_gzlevel;
1164 #endif
1165
1166 static int
1167 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1168     struct ucred *active_cred, struct ucred *file_cred,
1169     struct thread *td, char *core_buf, gzFile gzfile) {
1170
1171         int error;
1172         if (gzfile) {
1173 #ifdef COMPRESS_USER_CORES
1174                 error = compress_core(gzfile, base, core_buf, len, td);
1175 #else
1176                 panic("shouldn't be here");
1177 #endif
1178         } else {
1179                 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1180                     UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1181                     NULL, td);
1182         }
1183         return (error);
1184 }
1185
1186 /* Coredump output parameters for sbuf drain routine. */
1187 struct sbuf_drain_core_params {
1188         off_t           offset;
1189         struct ucred    *active_cred;
1190         struct ucred    *file_cred;
1191         struct thread   *td;
1192         struct vnode    *vp;
1193 #ifdef COMPRESS_USER_CORES
1194         gzFile          gzfile;
1195 #endif
1196 };
1197
1198 /*
1199  * Drain into a core file.
1200  */
1201 static int
1202 sbuf_drain_core_output(void *arg, const char *data, int len)
1203 {
1204         struct sbuf_drain_core_params *p;
1205         int error, locked;
1206
1207         p = (struct sbuf_drain_core_params *)arg;
1208
1209         /*
1210          * Some kern_proc out routines that print to this sbuf may
1211          * call us with the process lock held. Draining with the
1212          * non-sleepable lock held is unsafe. The lock is needed for
1213          * those routines when dumping a live process. In our case we
1214          * can safely release the lock before draining and acquire
1215          * again after.
1216          */
1217         locked = PROC_LOCKED(p->td->td_proc);
1218         if (locked)
1219                 PROC_UNLOCK(p->td->td_proc);
1220 #ifdef COMPRESS_USER_CORES
1221         if (p->gzfile != Z_NULL)
1222                 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1223                     len, p->td);
1224         else
1225 #endif
1226                 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1227                     __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1228                     IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1229                     p->td);
1230         if (locked)
1231                 PROC_LOCK(p->td->td_proc);
1232         if (error != 0)
1233                 return (-error);
1234         p->offset += len;
1235         return (len);
1236 }
1237
1238 /*
1239  * Drain into a counter.
1240  */
1241 static int
1242 sbuf_drain_count(void *arg, const char *data __unused, int len)
1243 {
1244         size_t *sizep;
1245
1246         sizep = (size_t *)arg;
1247         *sizep += len;
1248         return (len);
1249 }
1250
1251 int
1252 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1253 {
1254         struct ucred *cred = td->td_ucred;
1255         int error = 0;
1256         struct sseg_closure seginfo;
1257         struct note_info_list notelst;
1258         struct note_info *ninfo;
1259         void *hdr;
1260         size_t hdrsize, notesz, coresize;
1261
1262         gzFile gzfile = Z_NULL;
1263         char *core_buf = NULL;
1264 #ifdef COMPRESS_USER_CORES
1265         char gzopen_flags[8];
1266         char *p;
1267         int doing_compress = flags & IMGACT_CORE_COMPRESS;
1268 #endif
1269
1270         hdr = NULL;
1271         TAILQ_INIT(&notelst);
1272
1273 #ifdef COMPRESS_USER_CORES
1274         if (doing_compress) {
1275                 p = gzopen_flags;
1276                 *p++ = 'w';
1277                 if (compress_user_cores_gzlevel >= 0 &&
1278                     compress_user_cores_gzlevel <= 9)
1279                         *p++ = '0' + compress_user_cores_gzlevel;
1280                 *p = 0;
1281                 gzfile = gz_open("", gzopen_flags, vp);
1282                 if (gzfile == Z_NULL) {
1283                         error = EFAULT;
1284                         goto done;
1285                 }
1286                 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1287                 if (!core_buf) {
1288                         error = ENOMEM;
1289                         goto done;
1290                 }
1291         }
1292 #endif
1293
1294         /* Size the program segments. */
1295         seginfo.count = 0;
1296         seginfo.size = 0;
1297         each_writable_segment(td, cb_size_segment, &seginfo);
1298
1299         /*
1300          * Collect info about the core file header area.
1301          */
1302         hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1303         __elfN(prepare_notes)(td, &notelst, &notesz);
1304         coresize = round_page(hdrsize + notesz) + seginfo.size;
1305
1306 #ifdef RACCT
1307         if (racct_enable) {
1308                 PROC_LOCK(td->td_proc);
1309                 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1310                 PROC_UNLOCK(td->td_proc);
1311                 if (error != 0) {
1312                         error = EFAULT;
1313                         goto done;
1314                 }
1315         }
1316 #endif
1317         if (coresize >= limit) {
1318                 error = EFAULT;
1319                 goto done;
1320         }
1321
1322         /*
1323          * Allocate memory for building the header, fill it up,
1324          * and write it out following the notes.
1325          */
1326         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1327         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1328             &notelst, notesz, gzfile);
1329
1330         /* Write the contents of all of the writable segments. */
1331         if (error == 0) {
1332                 Elf_Phdr *php;
1333                 off_t offset;
1334                 int i;
1335
1336                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1337                 offset = round_page(hdrsize + notesz);
1338                 for (i = 0; i < seginfo.count; i++) {
1339                         error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1340                             php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1341                         if (error != 0)
1342                                 break;
1343                         offset += php->p_filesz;
1344                         php++;
1345                 }
1346         }
1347         if (error) {
1348                 log(LOG_WARNING,
1349                     "Failed to write core file for process %s (error %d)\n",
1350                     curproc->p_comm, error);
1351         }
1352
1353 done:
1354 #ifdef COMPRESS_USER_CORES
1355         if (core_buf)
1356                 free(core_buf, M_TEMP);
1357         if (gzfile)
1358                 gzclose(gzfile);
1359 #endif
1360         while ((ninfo = TAILQ_FIRST(&notelst)) != NULL) {
1361                 TAILQ_REMOVE(&notelst, ninfo, link);
1362                 free(ninfo, M_TEMP);
1363         }
1364         if (hdr != NULL)
1365                 free(hdr, M_TEMP);
1366
1367         return (error);
1368 }
1369
1370 /*
1371  * A callback for each_writable_segment() to write out the segment's
1372  * program header entry.
1373  */
1374 static void
1375 cb_put_phdr(entry, closure)
1376         vm_map_entry_t entry;
1377         void *closure;
1378 {
1379         struct phdr_closure *phc = (struct phdr_closure *)closure;
1380         Elf_Phdr *phdr = phc->phdr;
1381
1382         phc->offset = round_page(phc->offset);
1383
1384         phdr->p_type = PT_LOAD;
1385         phdr->p_offset = phc->offset;
1386         phdr->p_vaddr = entry->start;
1387         phdr->p_paddr = 0;
1388         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1389         phdr->p_align = PAGE_SIZE;
1390         phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1391
1392         phc->offset += phdr->p_filesz;
1393         phc->phdr++;
1394 }
1395
1396 /*
1397  * A callback for each_writable_segment() to gather information about
1398  * the number of segments and their total size.
1399  */
1400 static void
1401 cb_size_segment(entry, closure)
1402         vm_map_entry_t entry;
1403         void *closure;
1404 {
1405         struct sseg_closure *ssc = (struct sseg_closure *)closure;
1406
1407         ssc->count++;
1408         ssc->size += entry->end - entry->start;
1409 }
1410
1411 /*
1412  * For each writable segment in the process's memory map, call the given
1413  * function with a pointer to the map entry and some arbitrary
1414  * caller-supplied data.
1415  */
1416 static void
1417 each_writable_segment(td, func, closure)
1418         struct thread *td;
1419         segment_callback func;
1420         void *closure;
1421 {
1422         struct proc *p = td->td_proc;
1423         vm_map_t map = &p->p_vmspace->vm_map;
1424         vm_map_entry_t entry;
1425         vm_object_t backing_object, object;
1426         boolean_t ignore_entry;
1427
1428         vm_map_lock_read(map);
1429         for (entry = map->header.next; entry != &map->header;
1430             entry = entry->next) {
1431                 /*
1432                  * Don't dump inaccessible mappings, deal with legacy
1433                  * coredump mode.
1434                  *
1435                  * Note that read-only segments related to the elf binary
1436                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1437                  * need to arbitrarily ignore such segments.
1438                  */
1439                 if (elf_legacy_coredump) {
1440                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1441                                 continue;
1442                 } else {
1443                         if ((entry->protection & VM_PROT_ALL) == 0)
1444                                 continue;
1445                 }
1446
1447                 /*
1448                  * Dont include memory segment in the coredump if
1449                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1450                  * madvise(2).  Do not dump submaps (i.e. parts of the
1451                  * kernel map).
1452                  */
1453                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1454                         continue;
1455
1456                 if ((object = entry->object.vm_object) == NULL)
1457                         continue;
1458
1459                 /* Ignore memory-mapped devices and such things. */
1460                 VM_OBJECT_RLOCK(object);
1461                 while ((backing_object = object->backing_object) != NULL) {
1462                         VM_OBJECT_RLOCK(backing_object);
1463                         VM_OBJECT_RUNLOCK(object);
1464                         object = backing_object;
1465                 }
1466                 ignore_entry = object->type != OBJT_DEFAULT &&
1467                     object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1468                     object->type != OBJT_PHYS;
1469                 VM_OBJECT_RUNLOCK(object);
1470                 if (ignore_entry)
1471                         continue;
1472
1473                 (*func)(entry, closure);
1474         }
1475         vm_map_unlock_read(map);
1476 }
1477
1478 /*
1479  * Write the core file header to the file, including padding up to
1480  * the page boundary.
1481  */
1482 static int
1483 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1484     int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1485     size_t notesz, gzFile gzfile)
1486 {
1487         struct sbuf_drain_core_params params;
1488         struct note_info *ninfo;
1489         struct sbuf *sb;
1490         int error;
1491
1492         /* Fill in the header. */
1493         bzero(hdr, hdrsize);
1494         __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1495
1496         params.offset = 0;
1497         params.active_cred = cred;
1498         params.file_cred = NOCRED;
1499         params.td = td;
1500         params.vp = vp;
1501 #ifdef COMPRESS_USER_CORES
1502         params.gzfile = gzfile;
1503 #endif
1504         sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1505         sbuf_set_drain(sb, sbuf_drain_core_output, &params);
1506         sbuf_start_section(sb, NULL);
1507         sbuf_bcat(sb, hdr, hdrsize);
1508         TAILQ_FOREACH(ninfo, notelst, link)
1509             __elfN(putnote)(ninfo, sb);
1510         /* Align up to a page boundary for the program segments. */
1511         sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1512         error = sbuf_finish(sb);
1513         sbuf_delete(sb);
1514
1515         return (error);
1516 }
1517
1518 static void
1519 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1520     size_t *sizep)
1521 {
1522         struct proc *p;
1523         struct thread *thr;
1524         size_t size;
1525
1526         p = td->td_proc;
1527         size = 0;
1528
1529         size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1530
1531         /*
1532          * To have the debugger select the right thread (LWP) as the initial
1533          * thread, we dump the state of the thread passed to us in td first.
1534          * This is the thread that causes the core dump and thus likely to
1535          * be the right thread one wants to have selected in the debugger.
1536          */
1537         thr = td;
1538         while (thr != NULL) {
1539                 size += register_note(list, NT_PRSTATUS,
1540                     __elfN(note_prstatus), thr);
1541                 size += register_note(list, NT_FPREGSET,
1542                     __elfN(note_fpregset), thr);
1543                 size += register_note(list, NT_THRMISC,
1544                     __elfN(note_thrmisc), thr);
1545                 size += register_note(list, -1,
1546                     __elfN(note_threadmd), thr);
1547
1548                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1549                     TAILQ_NEXT(thr, td_plist);
1550                 if (thr == td)
1551                         thr = TAILQ_NEXT(thr, td_plist);
1552         }
1553
1554         size += register_note(list, NT_PROCSTAT_PROC,
1555             __elfN(note_procstat_proc), p);
1556         size += register_note(list, NT_PROCSTAT_FILES,
1557             note_procstat_files, p);
1558         size += register_note(list, NT_PROCSTAT_VMMAP,
1559             note_procstat_vmmap, p);
1560         size += register_note(list, NT_PROCSTAT_GROUPS,
1561             note_procstat_groups, p);
1562         size += register_note(list, NT_PROCSTAT_UMASK,
1563             note_procstat_umask, p);
1564         size += register_note(list, NT_PROCSTAT_RLIMIT,
1565             note_procstat_rlimit, p);
1566         size += register_note(list, NT_PROCSTAT_OSREL,
1567             note_procstat_osrel, p);
1568         size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1569             __elfN(note_procstat_psstrings), p);
1570         size += register_note(list, NT_PROCSTAT_AUXV,
1571             __elfN(note_procstat_auxv), p);
1572
1573         *sizep = size;
1574 }
1575
1576 static void
1577 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1578     size_t notesz)
1579 {
1580         Elf_Ehdr *ehdr;
1581         Elf_Phdr *phdr;
1582         struct phdr_closure phc;
1583
1584         ehdr = (Elf_Ehdr *)hdr;
1585         phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1586
1587         ehdr->e_ident[EI_MAG0] = ELFMAG0;
1588         ehdr->e_ident[EI_MAG1] = ELFMAG1;
1589         ehdr->e_ident[EI_MAG2] = ELFMAG2;
1590         ehdr->e_ident[EI_MAG3] = ELFMAG3;
1591         ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1592         ehdr->e_ident[EI_DATA] = ELF_DATA;
1593         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1594         ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1595         ehdr->e_ident[EI_ABIVERSION] = 0;
1596         ehdr->e_ident[EI_PAD] = 0;
1597         ehdr->e_type = ET_CORE;
1598 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1599         ehdr->e_machine = ELF_ARCH32;
1600 #else
1601         ehdr->e_machine = ELF_ARCH;
1602 #endif
1603         ehdr->e_version = EV_CURRENT;
1604         ehdr->e_entry = 0;
1605         ehdr->e_phoff = sizeof(Elf_Ehdr);
1606         ehdr->e_flags = 0;
1607         ehdr->e_ehsize = sizeof(Elf_Ehdr);
1608         ehdr->e_phentsize = sizeof(Elf_Phdr);
1609         ehdr->e_phnum = numsegs + 1;
1610         ehdr->e_shentsize = sizeof(Elf_Shdr);
1611         ehdr->e_shnum = 0;
1612         ehdr->e_shstrndx = SHN_UNDEF;
1613
1614         /*
1615          * Fill in the program header entries.
1616          */
1617
1618         /* The note segement. */
1619         phdr->p_type = PT_NOTE;
1620         phdr->p_offset = hdrsize;
1621         phdr->p_vaddr = 0;
1622         phdr->p_paddr = 0;
1623         phdr->p_filesz = notesz;
1624         phdr->p_memsz = 0;
1625         phdr->p_flags = PF_R;
1626         phdr->p_align = ELF_NOTE_ROUNDSIZE;
1627         phdr++;
1628
1629         /* All the writable segments from the program. */
1630         phc.phdr = phdr;
1631         phc.offset = round_page(hdrsize + notesz);
1632         each_writable_segment(td, cb_put_phdr, &phc);
1633 }
1634
1635 static size_t
1636 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1637 {
1638         struct note_info *ninfo;
1639         size_t size, notesize;
1640
1641         size = 0;
1642         out(arg, NULL, &size);
1643         ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1644         ninfo->type = type;
1645         ninfo->outfunc = out;
1646         ninfo->outarg = arg;
1647         ninfo->outsize = size;
1648         TAILQ_INSERT_TAIL(list, ninfo, link);
1649
1650         if (type == -1)
1651                 return (size);
1652
1653         notesize = sizeof(Elf_Note) +           /* note header */
1654             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1655                                                 /* note name */
1656             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1657
1658         return (notesize);
1659 }
1660
1661 static size_t
1662 append_note_data(const void *src, void *dst, size_t len)
1663 {
1664         size_t padded_len;
1665
1666         padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1667         if (dst != NULL) {
1668                 bcopy(src, dst, len);
1669                 bzero((char *)dst + len, padded_len - len);
1670         }
1671         return (padded_len);
1672 }
1673
1674 size_t
1675 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1676 {
1677         Elf_Note *note;
1678         char *buf;
1679         size_t notesize;
1680
1681         buf = dst;
1682         if (buf != NULL) {
1683                 note = (Elf_Note *)buf;
1684                 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1685                 note->n_descsz = size;
1686                 note->n_type = type;
1687                 buf += sizeof(*note);
1688                 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1689                     sizeof(FREEBSD_ABI_VENDOR));
1690                 append_note_data(src, buf, size);
1691                 if (descp != NULL)
1692                         *descp = buf;
1693         }
1694
1695         notesize = sizeof(Elf_Note) +           /* note header */
1696             roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1697                                                 /* note name */
1698             roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1699
1700         return (notesize);
1701 }
1702
1703 static void
1704 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1705 {
1706         Elf_Note note;
1707         ssize_t old_len, sect_len;
1708         size_t new_len, descsz, i;
1709
1710         if (ninfo->type == -1) {
1711                 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1712                 return;
1713         }
1714
1715         note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1716         note.n_descsz = ninfo->outsize;
1717         note.n_type = ninfo->type;
1718
1719         sbuf_bcat(sb, &note, sizeof(note));
1720         sbuf_start_section(sb, &old_len);
1721         sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1722         sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1723         if (note.n_descsz == 0)
1724                 return;
1725         sbuf_start_section(sb, &old_len);
1726         ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1727         sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1728         if (sect_len < 0)
1729                 return;
1730
1731         new_len = (size_t)sect_len;
1732         descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1733         if (new_len < descsz) {
1734                 /*
1735                  * It is expected that individual note emitters will correctly
1736                  * predict their expected output size and fill up to that size
1737                  * themselves, padding in a format-specific way if needed.
1738                  * However, in case they don't, just do it here with zeros.
1739                  */
1740                 for (i = 0; i < descsz - new_len; i++)
1741                         sbuf_putc(sb, 0);
1742         } else if (new_len > descsz) {
1743                 /*
1744                  * We can't always truncate sb -- we may have drained some
1745                  * of it already.
1746                  */
1747                 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1748                     "read it (%zu > %zu).  Since it is longer than "
1749                     "expected, this coredump's notes are corrupt.  THIS "
1750                     "IS A BUG in the note_procstat routine for type %u.\n",
1751                     __func__, (unsigned)note.n_type, new_len, descsz,
1752                     (unsigned)note.n_type));
1753         }
1754 }
1755
1756 /*
1757  * Miscellaneous note out functions.
1758  */
1759
1760 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1761 #include <compat/freebsd32/freebsd32.h>
1762
1763 typedef struct prstatus32 elf_prstatus_t;
1764 typedef struct prpsinfo32 elf_prpsinfo_t;
1765 typedef struct fpreg32 elf_prfpregset_t;
1766 typedef struct fpreg32 elf_fpregset_t;
1767 typedef struct reg32 elf_gregset_t;
1768 typedef struct thrmisc32 elf_thrmisc_t;
1769 #define ELF_KERN_PROC_MASK      KERN_PROC_MASK32
1770 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1771 typedef uint32_t elf_ps_strings_t;
1772 #else
1773 typedef prstatus_t elf_prstatus_t;
1774 typedef prpsinfo_t elf_prpsinfo_t;
1775 typedef prfpregset_t elf_prfpregset_t;
1776 typedef prfpregset_t elf_fpregset_t;
1777 typedef gregset_t elf_gregset_t;
1778 typedef thrmisc_t elf_thrmisc_t;
1779 #define ELF_KERN_PROC_MASK      0
1780 typedef struct kinfo_proc elf_kinfo_proc_t;
1781 typedef vm_offset_t elf_ps_strings_t;
1782 #endif
1783
1784 static void
1785 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1786 {
1787         struct sbuf sbarg;
1788         size_t len;
1789         char *cp, *end;
1790         struct proc *p;
1791         elf_prpsinfo_t *psinfo;
1792         int error;
1793
1794         p = (struct proc *)arg;
1795         if (sb != NULL) {
1796                 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1797                 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1798                 psinfo->pr_version = PRPSINFO_VERSION;
1799                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1800                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1801                 PROC_LOCK(p);
1802                 if (p->p_args != NULL) {
1803                         len = sizeof(psinfo->pr_psargs) - 1;
1804                         if (len > p->p_args->ar_length)
1805                                 len = p->p_args->ar_length;
1806                         memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1807                         PROC_UNLOCK(p);
1808                         error = 0;
1809                 } else {
1810                         _PHOLD(p);
1811                         PROC_UNLOCK(p);
1812                         sbuf_new(&sbarg, psinfo->pr_psargs,
1813                             sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1814                         error = proc_getargv(curthread, p, &sbarg);
1815                         PRELE(p);
1816                         if (sbuf_finish(&sbarg) == 0)
1817                                 len = sbuf_len(&sbarg) - 1;
1818                         else
1819                                 len = sizeof(psinfo->pr_psargs) - 1;
1820                         sbuf_delete(&sbarg);
1821                 }
1822                 if (error || len == 0)
1823                         strlcpy(psinfo->pr_psargs, p->p_comm,
1824                             sizeof(psinfo->pr_psargs));
1825                 else {
1826                         KASSERT(len < sizeof(psinfo->pr_psargs),
1827                             ("len is too long: %zu vs %zu", len,
1828                             sizeof(psinfo->pr_psargs)));
1829                         cp = psinfo->pr_psargs;
1830                         end = cp + len - 1;
1831                         for (;;) {
1832                                 cp = memchr(cp, '\0', end - cp);
1833                                 if (cp == NULL)
1834                                         break;
1835                                 *cp = ' ';
1836                         }
1837                 }
1838                 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1839                 free(psinfo, M_TEMP);
1840         }
1841         *sizep = sizeof(*psinfo);
1842 }
1843
1844 static void
1845 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1846 {
1847         struct thread *td;
1848         elf_prstatus_t *status;
1849
1850         td = (struct thread *)arg;
1851         if (sb != NULL) {
1852                 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1853                 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1854                 status->pr_version = PRSTATUS_VERSION;
1855                 status->pr_statussz = sizeof(elf_prstatus_t);
1856                 status->pr_gregsetsz = sizeof(elf_gregset_t);
1857                 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1858                 status->pr_osreldate = osreldate;
1859                 status->pr_cursig = td->td_proc->p_sig;
1860                 status->pr_pid = td->td_tid;
1861 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1862                 fill_regs32(td, &status->pr_reg);
1863 #else
1864                 fill_regs(td, &status->pr_reg);
1865 #endif
1866                 sbuf_bcat(sb, status, sizeof(*status));
1867                 free(status, M_TEMP);
1868         }
1869         *sizep = sizeof(*status);
1870 }
1871
1872 static void
1873 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1874 {
1875         struct thread *td;
1876         elf_prfpregset_t *fpregset;
1877
1878         td = (struct thread *)arg;
1879         if (sb != NULL) {
1880                 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1881                 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1882 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1883                 fill_fpregs32(td, fpregset);
1884 #else
1885                 fill_fpregs(td, fpregset);
1886 #endif
1887                 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1888                 free(fpregset, M_TEMP);
1889         }
1890         *sizep = sizeof(*fpregset);
1891 }
1892
1893 static void
1894 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1895 {
1896         struct thread *td;
1897         elf_thrmisc_t thrmisc;
1898
1899         td = (struct thread *)arg;
1900         if (sb != NULL) {
1901                 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1902                 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1903                 strcpy(thrmisc.pr_tname, td->td_name);
1904                 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1905         }
1906         *sizep = sizeof(thrmisc);
1907 }
1908
1909 /*
1910  * Allow for MD specific notes, as well as any MD
1911  * specific preparations for writing MI notes.
1912  */
1913 static void
1914 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1915 {
1916         struct thread *td;
1917         void *buf;
1918         size_t size;
1919
1920         td = (struct thread *)arg;
1921         size = *sizep;
1922         if (size != 0 && sb != NULL)
1923                 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1924         else
1925                 buf = NULL;
1926         size = 0;
1927         __elfN(dump_thread)(td, buf, &size);
1928         KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1929         if (size != 0 && sb != NULL)
1930                 sbuf_bcat(sb, buf, size);
1931         free(buf, M_TEMP);
1932         *sizep = size;
1933 }
1934
1935 #ifdef KINFO_PROC_SIZE
1936 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1937 #endif
1938
1939 static void
1940 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1941 {
1942         struct proc *p;
1943         size_t size;
1944         int structsize;
1945
1946         p = (struct proc *)arg;
1947         size = sizeof(structsize) + p->p_numthreads *
1948             sizeof(elf_kinfo_proc_t);
1949
1950         if (sb != NULL) {
1951                 KASSERT(*sizep == size, ("invalid size"));
1952                 structsize = sizeof(elf_kinfo_proc_t);
1953                 sbuf_bcat(sb, &structsize, sizeof(structsize));
1954                 PROC_LOCK(p);
1955                 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1956         }
1957         *sizep = size;
1958 }
1959
1960 #ifdef KINFO_FILE_SIZE
1961 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1962 #endif
1963
1964 static void
1965 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1966 {
1967         struct proc *p;
1968         size_t size, sect_sz, i;
1969         ssize_t start_len, sect_len;
1970         int structsize, filedesc_flags;
1971
1972         if (coredump_pack_fileinfo)
1973                 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1974         else
1975                 filedesc_flags = 0;
1976
1977         p = (struct proc *)arg;
1978         structsize = sizeof(struct kinfo_file);
1979         if (sb == NULL) {
1980                 size = 0;
1981                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1982                 sbuf_set_drain(sb, sbuf_drain_count, &size);
1983                 sbuf_bcat(sb, &structsize, sizeof(structsize));
1984                 PROC_LOCK(p);
1985                 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1986                 sbuf_finish(sb);
1987                 sbuf_delete(sb);
1988                 *sizep = size;
1989         } else {
1990                 sbuf_start_section(sb, &start_len);
1991
1992                 sbuf_bcat(sb, &structsize, sizeof(structsize));
1993                 PROC_LOCK(p);
1994                 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1995                     filedesc_flags);
1996
1997                 sect_len = sbuf_end_section(sb, start_len, 0, 0);
1998                 if (sect_len < 0)
1999                         return;
2000                 sect_sz = sect_len;
2001
2002                 KASSERT(sect_sz <= *sizep,
2003                     ("kern_proc_filedesc_out did not respect maxlen; "
2004                      "requested %zu, got %zu", *sizep - sizeof(structsize),
2005                      sect_sz - sizeof(structsize)));
2006
2007                 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2008                         sbuf_putc(sb, 0);
2009         }
2010 }
2011
2012 #ifdef KINFO_VMENTRY_SIZE
2013 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2014 #endif
2015
2016 static void
2017 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2018 {
2019         struct proc *p;
2020         size_t size;
2021         int structsize, vmmap_flags;
2022
2023         if (coredump_pack_vmmapinfo)
2024                 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2025         else
2026                 vmmap_flags = 0;
2027
2028         p = (struct proc *)arg;
2029         structsize = sizeof(struct kinfo_vmentry);
2030         if (sb == NULL) {
2031                 size = 0;
2032                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2033                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2034                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2035                 PROC_LOCK(p);
2036                 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2037                 sbuf_finish(sb);
2038                 sbuf_delete(sb);
2039                 *sizep = size;
2040         } else {
2041                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2042                 PROC_LOCK(p);
2043                 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2044                     vmmap_flags);
2045         }
2046 }
2047
2048 static void
2049 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2050 {
2051         struct proc *p;
2052         size_t size;
2053         int structsize;
2054
2055         p = (struct proc *)arg;
2056         size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2057         if (sb != NULL) {
2058                 KASSERT(*sizep == size, ("invalid size"));
2059                 structsize = sizeof(gid_t);
2060                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2061                 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2062                     sizeof(gid_t));
2063         }
2064         *sizep = size;
2065 }
2066
2067 static void
2068 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2069 {
2070         struct proc *p;
2071         size_t size;
2072         int structsize;
2073
2074         p = (struct proc *)arg;
2075         size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2076         if (sb != NULL) {
2077                 KASSERT(*sizep == size, ("invalid size"));
2078                 structsize = sizeof(p->p_fd->fd_cmask);
2079                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2080                 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2081         }
2082         *sizep = size;
2083 }
2084
2085 static void
2086 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2087 {
2088         struct proc *p;
2089         struct rlimit rlim[RLIM_NLIMITS];
2090         size_t size;
2091         int structsize, i;
2092
2093         p = (struct proc *)arg;
2094         size = sizeof(structsize) + sizeof(rlim);
2095         if (sb != NULL) {
2096                 KASSERT(*sizep == size, ("invalid size"));
2097                 structsize = sizeof(rlim);
2098                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2099                 PROC_LOCK(p);
2100                 for (i = 0; i < RLIM_NLIMITS; i++)
2101                         lim_rlimit(p, i, &rlim[i]);
2102                 PROC_UNLOCK(p);
2103                 sbuf_bcat(sb, rlim, sizeof(rlim));
2104         }
2105         *sizep = size;
2106 }
2107
2108 static void
2109 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2110 {
2111         struct proc *p;
2112         size_t size;
2113         int structsize;
2114
2115         p = (struct proc *)arg;
2116         size = sizeof(structsize) + sizeof(p->p_osrel);
2117         if (sb != NULL) {
2118                 KASSERT(*sizep == size, ("invalid size"));
2119                 structsize = sizeof(p->p_osrel);
2120                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2121                 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2122         }
2123         *sizep = size;
2124 }
2125
2126 static void
2127 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2128 {
2129         struct proc *p;
2130         elf_ps_strings_t ps_strings;
2131         size_t size;
2132         int structsize;
2133
2134         p = (struct proc *)arg;
2135         size = sizeof(structsize) + sizeof(ps_strings);
2136         if (sb != NULL) {
2137                 KASSERT(*sizep == size, ("invalid size"));
2138                 structsize = sizeof(ps_strings);
2139 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2140                 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2141 #else
2142                 ps_strings = p->p_sysent->sv_psstrings;
2143 #endif
2144                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2145                 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2146         }
2147         *sizep = size;
2148 }
2149
2150 static void
2151 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2152 {
2153         struct proc *p;
2154         size_t size;
2155         int structsize;
2156
2157         p = (struct proc *)arg;
2158         if (sb == NULL) {
2159                 size = 0;
2160                 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2161                 sbuf_set_drain(sb, sbuf_drain_count, &size);
2162                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2163                 PHOLD(p);
2164                 proc_getauxv(curthread, p, sb);
2165                 PRELE(p);
2166                 sbuf_finish(sb);
2167                 sbuf_delete(sb);
2168                 *sizep = size;
2169         } else {
2170                 structsize = sizeof(Elf_Auxinfo);
2171                 sbuf_bcat(sb, &structsize, sizeof(structsize));
2172                 PHOLD(p);
2173                 proc_getauxv(curthread, p, sb);
2174                 PRELE(p);
2175         }
2176 }
2177
2178 static boolean_t
2179 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2180     int32_t *osrel, const Elf_Phdr *pnote)
2181 {
2182         const Elf_Note *note, *note0, *note_end;
2183         const char *note_name;
2184         char *buf;
2185         int i, error;
2186         boolean_t res;
2187
2188         /* We need some limit, might as well use PAGE_SIZE. */
2189         if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2190                 return (FALSE);
2191         ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2192         if (pnote->p_offset > PAGE_SIZE ||
2193             pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2194                 VOP_UNLOCK(imgp->vp, 0);
2195                 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2196                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2197                 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2198                     pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2199                     curthread->td_ucred, NOCRED, NULL, curthread);
2200                 if (error != 0) {
2201                         uprintf("i/o error PT_NOTE\n");
2202                         res = FALSE;
2203                         goto ret;
2204                 }
2205                 note = note0 = (const Elf_Note *)buf;
2206                 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2207         } else {
2208                 note = note0 = (const Elf_Note *)(imgp->image_header +
2209                     pnote->p_offset);
2210                 note_end = (const Elf_Note *)(imgp->image_header +
2211                     pnote->p_offset + pnote->p_filesz);
2212                 buf = NULL;
2213         }
2214         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2215                 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2216                     (const char *)note < sizeof(Elf_Note)) {
2217                         res = FALSE;
2218                         goto ret;
2219                 }
2220                 if (note->n_namesz != checknote->hdr.n_namesz ||
2221                     note->n_descsz != checknote->hdr.n_descsz ||
2222                     note->n_type != checknote->hdr.n_type)
2223                         goto nextnote;
2224                 note_name = (const char *)(note + 1);
2225                 if (note_name + checknote->hdr.n_namesz >=
2226                     (const char *)note_end || strncmp(checknote->vendor,
2227                     note_name, checknote->hdr.n_namesz) != 0)
2228                         goto nextnote;
2229
2230                 /*
2231                  * Fetch the osreldate for binary
2232                  * from the ELF OSABI-note if necessary.
2233                  */
2234                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2235                     checknote->trans_osrel != NULL) {
2236                         res = checknote->trans_osrel(note, osrel);
2237                         goto ret;
2238                 }
2239                 res = TRUE;
2240                 goto ret;
2241 nextnote:
2242                 note = (const Elf_Note *)((const char *)(note + 1) +
2243                     roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2244                     roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2245         }
2246         res = FALSE;
2247 ret:
2248         free(buf, M_TEMP);
2249         return (res);
2250 }
2251
2252 /*
2253  * Try to find the appropriate ABI-note section for checknote,
2254  * fetch the osreldate for binary from the ELF OSABI-note. Only the
2255  * first page of the image is searched, the same as for headers.
2256  */
2257 static boolean_t
2258 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2259     int32_t *osrel)
2260 {
2261         const Elf_Phdr *phdr;
2262         const Elf_Ehdr *hdr;
2263         int i;
2264
2265         hdr = (const Elf_Ehdr *)imgp->image_header;
2266         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2267
2268         for (i = 0; i < hdr->e_phnum; i++) {
2269                 if (phdr[i].p_type == PT_NOTE &&
2270                     __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2271                         return (TRUE);
2272         }
2273         return (FALSE);
2274
2275 }
2276
2277 /*
2278  * Tell kern_execve.c about it, with a little help from the linker.
2279  */
2280 static struct execsw __elfN(execsw) = {
2281         __CONCAT(exec_, __elfN(imgact)),
2282         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2283 };
2284 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2285
2286 #ifdef COMPRESS_USER_CORES
2287 /*
2288  * Compress and write out a core segment for a user process.
2289  *
2290  * 'inbuf' is the starting address of a VM segment in the process' address
2291  * space that is to be compressed and written out to the core file.  'dest_buf'
2292  * is a buffer in the kernel's address space.  The segment is copied from
2293  * 'inbuf' to 'dest_buf' first before being processed by the compression
2294  * routine gzwrite().  This copying is necessary because the content of the VM
2295  * segment may change between the compression pass and the crc-computation pass
2296  * in gzwrite().  This is because realtime threads may preempt the UNIX kernel.
2297  *
2298  * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2299  */
2300 static int
2301 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2302     struct thread *td)
2303 {
2304         int len_compressed;
2305         int error = 0;
2306         unsigned int chunk_len;
2307
2308         while (len) {
2309                 if (inbuf != NULL) {
2310                         chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2311                         copyin(inbuf, dest_buf, chunk_len);
2312                         inbuf += chunk_len;
2313                 } else {
2314                         chunk_len = len;
2315                 }
2316                 len_compressed = gzwrite(file, dest_buf, chunk_len);
2317
2318                 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2319
2320                 if ((unsigned int)len_compressed != chunk_len) {
2321                         log(LOG_WARNING,
2322                             "compress_core: length mismatch (0x%x returned, "
2323                             "0x%x expected)\n", len_compressed, chunk_len);
2324                         EVENTHANDLER_INVOKE(app_coredump_error, td,
2325                             "compress_core: length mismatch %x -> %x",
2326                             chunk_len, len_compressed);
2327                         error = EFAULT;
2328                         break;
2329                 }
2330                 len -= chunk_len;
2331                 maybe_yield();
2332         }
2333
2334         return (error);
2335 }
2336 #endif /* COMPRESS_USER_CORES */
2337
2338 static vm_prot_t
2339 __elfN(trans_prot)(Elf_Word flags)
2340 {
2341         vm_prot_t prot;
2342
2343         prot = 0;
2344         if (flags & PF_X)
2345                 prot |= VM_PROT_EXECUTE;
2346         if (flags & PF_W)
2347                 prot |= VM_PROT_WRITE;
2348         if (flags & PF_R)
2349                 prot |= VM_PROT_READ;
2350 #if __ELF_WORD_SIZE == 32
2351 #if defined(__amd64__) || defined(__ia64__)
2352         if (i386_read_exec && (flags & PF_R))
2353                 prot |= VM_PROT_EXECUTE;
2354 #endif
2355 #endif
2356         return (prot);
2357 }
2358
2359 static Elf_Word
2360 __elfN(untrans_prot)(vm_prot_t prot)
2361 {
2362         Elf_Word flags;
2363
2364         flags = 0;
2365         if (prot & VM_PROT_EXECUTE)
2366                 flags |= PF_X;
2367         if (prot & VM_PROT_READ)
2368                 flags |= PF_R;
2369         if (prot & VM_PROT_WRITE)
2370                 flags |= PF_W;
2371         return (flags);
2372 }