The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/imgact_elf.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    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: releng/8.3/sys/kern/imgact_elf.c 217442 2011-01-15 09:25:19Z kib $");
   33 
   34 #include "opt_compat.h"
   35 
   36 #include <sys/param.h>
   37 #include <sys/exec.h>
   38 #include <sys/fcntl.h>
   39 #include <sys/imgact.h>
   40 #include <sys/imgact_elf.h>
   41 #include <sys/kernel.h>
   42 #include <sys/lock.h>
   43 #include <sys/malloc.h>
   44 #include <sys/mount.h>
   45 #include <sys/mutex.h>
   46 #include <sys/mman.h>
   47 #include <sys/namei.h>
   48 #include <sys/pioctl.h>
   49 #include <sys/proc.h>
   50 #include <sys/procfs.h>
   51 #include <sys/resourcevar.h>
   52 #include <sys/sf_buf.h>
   53 #include <sys/systm.h>
   54 #include <sys/signalvar.h>
   55 #include <sys/stat.h>
   56 #include <sys/sx.h>
   57 #include <sys/syscall.h>
   58 #include <sys/sysctl.h>
   59 #include <sys/sysent.h>
   60 #include <sys/vnode.h>
   61 
   62 #include <vm/vm.h>
   63 #include <vm/vm_kern.h>
   64 #include <vm/vm_param.h>
   65 #include <vm/pmap.h>
   66 #include <vm/vm_map.h>
   67 #include <vm/vm_object.h>
   68 #include <vm/vm_extern.h>
   69 
   70 #include <machine/elf.h>
   71 #include <machine/md_var.h>
   72 
   73 #define OLD_EI_BRAND    8
   74 
   75 static int __elfN(check_header)(const Elf_Ehdr *hdr);
   76 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
   77     const char *interp, int32_t *osrel);
   78 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
   79     u_long *entry, size_t pagesize);
   80 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
   81     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
   82     vm_prot_t prot, size_t pagesize);
   83 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
   84 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
   85     int32_t *osrel);
   86 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
   87 static boolean_t __elfN(check_note)(struct image_params *imgp,
   88     Elf_Brandnote *checknote, int32_t *osrel);
   89 static vm_prot_t __elfN(trans_prot)(Elf_Word);
   90 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
   91 
   92 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
   93     "");
   94 
   95 int __elfN(fallback_brand) = -1;
   96 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
   97     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
   98     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
   99 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
  100     &__elfN(fallback_brand));
  101 
  102 static int elf_legacy_coredump = 0;
  103 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 
  104     &elf_legacy_coredump, 0, "");
  105 
  106 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
  107 
  108 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
  109 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
  110 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
  111 
  112 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
  113 
  114 Elf_Brandnote __elfN(freebsd_brandnote) = {
  115         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
  116         .hdr.n_descsz   = sizeof(int32_t),
  117         .hdr.n_type     = 1,
  118         .vendor         = FREEBSD_ABI_VENDOR,
  119         .flags          = BN_TRANSLATE_OSREL,
  120         .trans_osrel    = __elfN(freebsd_trans_osrel)
  121 };
  122 
  123 static boolean_t
  124 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
  125 {
  126         uintptr_t p;
  127 
  128         p = (uintptr_t)(note + 1);
  129         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  130         *osrel = *(const int32_t *)(p);
  131 
  132         return (TRUE);
  133 }
  134 
  135 static const char GNU_ABI_VENDOR[] = "GNU";
  136 static int GNU_KFREEBSD_ABI_DESC = 3;
  137 
  138 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
  139         .hdr.n_namesz   = sizeof(GNU_ABI_VENDOR),
  140         .hdr.n_descsz   = 16,   /* XXX at least 16 */
  141         .hdr.n_type     = 1,
  142         .vendor         = GNU_ABI_VENDOR,
  143         .flags          = BN_TRANSLATE_OSREL,
  144         .trans_osrel    = kfreebsd_trans_osrel
  145 };
  146 
  147 static boolean_t
  148 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
  149 {
  150         const Elf32_Word *desc;
  151         uintptr_t p;
  152 
  153         p = (uintptr_t)(note + 1);
  154         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  155 
  156         desc = (const Elf32_Word *)p;
  157         if (desc[0] != GNU_KFREEBSD_ABI_DESC)
  158                 return (FALSE);
  159 
  160         /*
  161          * Debian GNU/kFreeBSD embed the earliest compatible kernel version
  162          * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
  163          */
  164         *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
  165 
  166         return (TRUE);
  167 }
  168 
  169 int
  170 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
  171 {
  172         int i;
  173 
  174         for (i = 0; i < MAX_BRANDS; i++) {
  175                 if (elf_brand_list[i] == NULL) {
  176                         elf_brand_list[i] = entry;
  177                         break;
  178                 }
  179         }
  180         if (i == MAX_BRANDS) {
  181                 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
  182                         __func__, entry);
  183                 return (-1);
  184         }
  185         return (0);
  186 }
  187 
  188 int
  189 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
  190 {
  191         int i;
  192 
  193         for (i = 0; i < MAX_BRANDS; i++) {
  194                 if (elf_brand_list[i] == entry) {
  195                         elf_brand_list[i] = NULL;
  196                         break;
  197                 }
  198         }
  199         if (i == MAX_BRANDS)
  200                 return (-1);
  201         return (0);
  202 }
  203 
  204 int
  205 __elfN(brand_inuse)(Elf_Brandinfo *entry)
  206 {
  207         struct proc *p;
  208         int rval = FALSE;
  209 
  210         sx_slock(&allproc_lock);
  211         FOREACH_PROC_IN_SYSTEM(p) {
  212                 if (p->p_sysent == entry->sysvec) {
  213                         rval = TRUE;
  214                         break;
  215                 }
  216         }
  217         sx_sunlock(&allproc_lock);
  218 
  219         return (rval);
  220 }
  221 
  222 static Elf_Brandinfo *
  223 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
  224     int32_t *osrel)
  225 {
  226         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  227         Elf_Brandinfo *bi;
  228         boolean_t ret;
  229         int i;
  230 
  231         /*
  232          * We support four types of branding -- (1) the ELF EI_OSABI field
  233          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
  234          * branding w/in the ELF header, (3) path of the `interp_path'
  235          * field, and (4) the ".note.ABI-tag" ELF section.
  236          */
  237 
  238         /* Look for an ".note.ABI-tag" ELF section */
  239         for (i = 0; i < MAX_BRANDS; i++) {
  240                 bi = elf_brand_list[i];
  241                 if (bi == NULL)
  242                         continue;
  243                 if (hdr->e_machine == bi->machine && (bi->flags &
  244                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
  245                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
  246                         if (ret)
  247                                 return (bi);
  248                 }
  249         }
  250 
  251         /* If the executable has a brand, search for it in the brand list. */
  252         for (i = 0; i < MAX_BRANDS; i++) {
  253                 bi = elf_brand_list[i];
  254                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  255                         continue;
  256                 if (hdr->e_machine == bi->machine &&
  257                     (hdr->e_ident[EI_OSABI] == bi->brand ||
  258                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
  259                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
  260                         return (bi);
  261         }
  262 
  263         /* Lacking a known brand, search for a recognized interpreter. */
  264         if (interp != NULL) {
  265                 for (i = 0; i < MAX_BRANDS; i++) {
  266                         bi = elf_brand_list[i];
  267                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  268                                 continue;
  269                         if (hdr->e_machine == bi->machine &&
  270                             strcmp(interp, bi->interp_path) == 0)
  271                                 return (bi);
  272                 }
  273         }
  274 
  275         /* Lacking a recognized interpreter, try the default brand */
  276         for (i = 0; i < MAX_BRANDS; i++) {
  277                 bi = elf_brand_list[i];
  278                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  279                         continue;
  280                 if (hdr->e_machine == bi->machine &&
  281                     __elfN(fallback_brand) == bi->brand)
  282                         return (bi);
  283         }
  284         return (NULL);
  285 }
  286 
  287 static int
  288 __elfN(check_header)(const Elf_Ehdr *hdr)
  289 {
  290         Elf_Brandinfo *bi;
  291         int i;
  292 
  293         if (!IS_ELF(*hdr) ||
  294             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
  295             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
  296             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
  297             hdr->e_phentsize != sizeof(Elf_Phdr) ||
  298             hdr->e_version != ELF_TARG_VER)
  299                 return (ENOEXEC);
  300 
  301         /*
  302          * Make sure we have at least one brand for this machine.
  303          */
  304 
  305         for (i = 0; i < MAX_BRANDS; i++) {
  306                 bi = elf_brand_list[i];
  307                 if (bi != NULL && bi->machine == hdr->e_machine)
  308                         break;
  309         }
  310         if (i == MAX_BRANDS)
  311                 return (ENOEXEC);
  312 
  313         return (0);
  314 }
  315 
  316 static int
  317 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  318     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
  319 {
  320         struct sf_buf *sf;
  321         int error;
  322         vm_offset_t off;
  323 
  324         /*
  325          * Create the page if it doesn't exist yet. Ignore errors.
  326          */
  327         vm_map_lock(map);
  328         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
  329             VM_PROT_ALL, VM_PROT_ALL, 0);
  330         vm_map_unlock(map);
  331 
  332         /*
  333          * Find the page from the underlying object.
  334          */
  335         if (object) {
  336                 sf = vm_imgact_map_page(object, offset);
  337                 if (sf == NULL)
  338                         return (KERN_FAILURE);
  339                 off = offset - trunc_page(offset);
  340                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
  341                     end - start);
  342                 vm_imgact_unmap_page(sf);
  343                 if (error) {
  344                         return (KERN_FAILURE);
  345                 }
  346         }
  347 
  348         return (KERN_SUCCESS);
  349 }
  350 
  351 static int
  352 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  353     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
  354 {
  355         struct sf_buf *sf;
  356         vm_offset_t off;
  357         vm_size_t sz;
  358         int error, rv;
  359 
  360         if (start != trunc_page(start)) {
  361                 rv = __elfN(map_partial)(map, object, offset, start,
  362                     round_page(start), prot);
  363                 if (rv)
  364                         return (rv);
  365                 offset += round_page(start) - start;
  366                 start = round_page(start);
  367         }
  368         if (end != round_page(end)) {
  369                 rv = __elfN(map_partial)(map, object, offset +
  370                     trunc_page(end) - start, trunc_page(end), end, prot);
  371                 if (rv)
  372                         return (rv);
  373                 end = trunc_page(end);
  374         }
  375         if (end > start) {
  376                 if (offset & PAGE_MASK) {
  377                         /*
  378                          * The mapping is not page aligned. This means we have
  379                          * to copy the data. Sigh.
  380                          */
  381                         rv = vm_map_find(map, NULL, 0, &start, end - start,
  382                             FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
  383                         if (rv)
  384                                 return (rv);
  385                         if (object == NULL)
  386                                 return (KERN_SUCCESS);
  387                         for (; start < end; start += sz) {
  388                                 sf = vm_imgact_map_page(object, offset);
  389                                 if (sf == NULL)
  390                                         return (KERN_FAILURE);
  391                                 off = offset - trunc_page(offset);
  392                                 sz = end - start;
  393                                 if (sz > PAGE_SIZE - off)
  394                                         sz = PAGE_SIZE - off;
  395                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  396                                     (caddr_t)start, sz);
  397                                 vm_imgact_unmap_page(sf);
  398                                 if (error) {
  399                                         return (KERN_FAILURE);
  400                                 }
  401                                 offset += sz;
  402                         }
  403                         rv = KERN_SUCCESS;
  404                 } else {
  405                         vm_object_reference(object);
  406                         vm_map_lock(map);
  407                         rv = vm_map_insert(map, object, offset, start, end,
  408                             prot, VM_PROT_ALL, cow);
  409                         vm_map_unlock(map);
  410                         if (rv != KERN_SUCCESS)
  411                                 vm_object_deallocate(object);
  412                 }
  413                 return (rv);
  414         } else {
  415                 return (KERN_SUCCESS);
  416         }
  417 }
  418 
  419 static int
  420 __elfN(load_section)(struct vmspace *vmspace,
  421         vm_object_t object, vm_offset_t offset,
  422         caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
  423         size_t pagesize)
  424 {
  425         struct sf_buf *sf;
  426         size_t map_len;
  427         vm_offset_t map_addr;
  428         int error, rv, cow;
  429         size_t copy_len;
  430         vm_offset_t file_addr;
  431 
  432         /*
  433          * It's necessary to fail if the filsz + offset taken from the
  434          * header is greater than the actual file pager object's size.
  435          * If we were to allow this, then the vm_map_find() below would
  436          * walk right off the end of the file object and into the ether.
  437          *
  438          * While I'm here, might as well check for something else that
  439          * is invalid: filsz cannot be greater than memsz.
  440          */
  441         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
  442             filsz > memsz) {
  443                 uprintf("elf_load_section: truncated ELF file\n");
  444                 return (ENOEXEC);
  445         }
  446 
  447         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
  448         file_addr = trunc_page_ps(offset, pagesize);
  449 
  450         /*
  451          * We have two choices.  We can either clear the data in the last page
  452          * of an oversized mapping, or we can start the anon mapping a page
  453          * early and copy the initialized data into that first page.  We
  454          * choose the second..
  455          */
  456         if (memsz > filsz)
  457                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
  458         else
  459                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
  460 
  461         if (map_len != 0) {
  462                 /* cow flags: don't dump readonly sections in core */
  463                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
  464                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
  465 
  466                 rv = __elfN(map_insert)(&vmspace->vm_map,
  467                                       object,
  468                                       file_addr,        /* file offset */
  469                                       map_addr,         /* virtual start */
  470                                       map_addr + map_len,/* virtual end */
  471                                       prot,
  472                                       cow);
  473                 if (rv != KERN_SUCCESS)
  474                         return (EINVAL);
  475 
  476                 /* we can stop now if we've covered it all */
  477                 if (memsz == filsz) {
  478                         return (0);
  479                 }
  480         }
  481 
  482 
  483         /*
  484          * We have to get the remaining bit of the file into the first part
  485          * of the oversized map segment.  This is normally because the .data
  486          * segment in the file is extended to provide bss.  It's a neat idea
  487          * to try and save a page, but it's a pain in the behind to implement.
  488          */
  489         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
  490         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
  491         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
  492             map_addr;
  493 
  494         /* This had damn well better be true! */
  495         if (map_len != 0) {
  496                 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
  497                     map_addr + map_len, VM_PROT_ALL, 0);
  498                 if (rv != KERN_SUCCESS) {
  499                         return (EINVAL);
  500                 }
  501         }
  502 
  503         if (copy_len != 0) {
  504                 vm_offset_t off;
  505 
  506                 sf = vm_imgact_map_page(object, offset + filsz);
  507                 if (sf == NULL)
  508                         return (EIO);
  509 
  510                 /* send the page fragment to user space */
  511                 off = trunc_page_ps(offset + filsz, pagesize) -
  512                     trunc_page(offset + filsz);
  513                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  514                     (caddr_t)map_addr, copy_len);
  515                 vm_imgact_unmap_page(sf);
  516                 if (error) {
  517                         return (error);
  518                 }
  519         }
  520 
  521         /*
  522          * set it to the specified protection.
  523          * XXX had better undo the damage from pasting over the cracks here!
  524          */
  525         vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
  526             round_page(map_addr + map_len),  prot, FALSE);
  527 
  528         return (0);
  529 }
  530 
  531 /*
  532  * Load the file "file" into memory.  It may be either a shared object
  533  * or an executable.
  534  *
  535  * The "addr" reference parameter is in/out.  On entry, it specifies
  536  * the address where a shared object should be loaded.  If the file is
  537  * an executable, this value is ignored.  On exit, "addr" specifies
  538  * where the file was actually loaded.
  539  *
  540  * The "entry" reference parameter is out only.  On exit, it specifies
  541  * the entry point for the loaded file.
  542  */
  543 static int
  544 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  545         u_long *entry, size_t pagesize)
  546 {
  547         struct {
  548                 struct nameidata nd;
  549                 struct vattr attr;
  550                 struct image_params image_params;
  551         } *tempdata;
  552         const Elf_Ehdr *hdr = NULL;
  553         const Elf_Phdr *phdr = NULL;
  554         struct nameidata *nd;
  555         struct vmspace *vmspace = p->p_vmspace;
  556         struct vattr *attr;
  557         struct image_params *imgp;
  558         vm_prot_t prot;
  559         u_long rbase;
  560         u_long base_addr = 0;
  561         int vfslocked, error, i, numsegs;
  562 
  563         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
  564         nd = &tempdata->nd;
  565         attr = &tempdata->attr;
  566         imgp = &tempdata->image_params;
  567 
  568         /*
  569          * Initialize part of the common data
  570          */
  571         imgp->proc = p;
  572         imgp->attr = attr;
  573         imgp->firstpage = NULL;
  574         imgp->image_header = NULL;
  575         imgp->object = NULL;
  576         imgp->execlabel = NULL;
  577 
  578         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
  579             curthread);
  580         vfslocked = 0;
  581         if ((error = namei(nd)) != 0) {
  582                 nd->ni_vp = NULL;
  583                 goto fail;
  584         }
  585         vfslocked = NDHASGIANT(nd);
  586         NDFREE(nd, NDF_ONLY_PNBUF);
  587         imgp->vp = nd->ni_vp;
  588 
  589         /*
  590          * Check permissions, modes, uid, etc on the file, and "open" it.
  591          */
  592         error = exec_check_permissions(imgp);
  593         if (error)
  594                 goto fail;
  595 
  596         error = exec_map_first_page(imgp);
  597         if (error)
  598                 goto fail;
  599 
  600         /*
  601          * Also make certain that the interpreter stays the same, so set
  602          * its VV_TEXT flag, too.
  603          */
  604         nd->ni_vp->v_vflag |= VV_TEXT;
  605 
  606         imgp->object = nd->ni_vp->v_object;
  607 
  608         hdr = (const Elf_Ehdr *)imgp->image_header;
  609         if ((error = __elfN(check_header)(hdr)) != 0)
  610                 goto fail;
  611         if (hdr->e_type == ET_DYN)
  612                 rbase = *addr;
  613         else if (hdr->e_type == ET_EXEC)
  614                 rbase = 0;
  615         else {
  616                 error = ENOEXEC;
  617                 goto fail;
  618         }
  619 
  620         /* Only support headers that fit within first page for now      */
  621         /*    (multiplication of two Elf_Half fields will not overflow) */
  622         if ((hdr->e_phoff > PAGE_SIZE) ||
  623             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
  624                 error = ENOEXEC;
  625                 goto fail;
  626         }
  627 
  628         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  629         if (!aligned(phdr, Elf_Addr)) {
  630                 error = ENOEXEC;
  631                 goto fail;
  632         }
  633 
  634         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
  635                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
  636                         /* Loadable segment */
  637                         prot = __elfN(trans_prot)(phdr[i].p_flags);
  638                         if ((error = __elfN(load_section)(vmspace,
  639                             imgp->object, phdr[i].p_offset,
  640                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
  641                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  642                             pagesize)) != 0)
  643                                 goto fail;
  644                         /*
  645                          * Establish the base address if this is the
  646                          * first segment.
  647                          */
  648                         if (numsegs == 0)
  649                                 base_addr = trunc_page(phdr[i].p_vaddr +
  650                                     rbase);
  651                         numsegs++;
  652                 }
  653         }
  654         *addr = base_addr;
  655         *entry = (unsigned long)hdr->e_entry + rbase;
  656 
  657 fail:
  658         if (imgp->firstpage)
  659                 exec_unmap_first_page(imgp);
  660 
  661         if (nd->ni_vp)
  662                 vput(nd->ni_vp);
  663 
  664         VFS_UNLOCK_GIANT(vfslocked);
  665         free(tempdata, M_TEMP);
  666 
  667         return (error);
  668 }
  669 
  670 static int
  671 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
  672 {
  673         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  674         const Elf_Phdr *phdr;
  675         Elf_Auxargs *elf_auxargs;
  676         struct vmspace *vmspace;
  677         vm_prot_t prot;
  678         u_long text_size = 0, data_size = 0, total_size = 0;
  679         u_long text_addr = 0, data_addr = 0;
  680         u_long seg_size, seg_addr;
  681         u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
  682         int32_t osrel = 0;
  683         int error = 0, i, n;
  684         const char *interp = NULL, *newinterp = NULL;
  685         Elf_Brandinfo *brand_info;
  686         char *path;
  687         struct sysentvec *sv;
  688 
  689         /*
  690          * Do we have a valid ELF header ?
  691          *
  692          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
  693          * if particular brand doesn't support it.
  694          */
  695         if (__elfN(check_header)(hdr) != 0 ||
  696             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
  697                 return (-1);
  698 
  699         /*
  700          * From here on down, we return an errno, not -1, as we've
  701          * detected an ELF file.
  702          */
  703 
  704         if ((hdr->e_phoff > PAGE_SIZE) ||
  705             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
  706                 /* Only support headers in first page for now */
  707                 return (ENOEXEC);
  708         }
  709         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  710         if (!aligned(phdr, Elf_Addr))
  711                 return (ENOEXEC);
  712         n = 0;
  713         baddr = 0;
  714         for (i = 0; i < hdr->e_phnum; i++) {
  715                 if (phdr[i].p_type == PT_LOAD) {
  716                         if (n == 0)
  717                                 baddr = phdr[i].p_vaddr;
  718                         n++;
  719                         continue;
  720                 }
  721                 if (phdr[i].p_type == PT_INTERP) {
  722                         /* Path to interpreter */
  723                         if (phdr[i].p_filesz > MAXPATHLEN ||
  724                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
  725                                 return (ENOEXEC);
  726                         interp = imgp->image_header + phdr[i].p_offset;
  727                         continue;
  728                 }
  729         }
  730 
  731         brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
  732         if (brand_info == NULL) {
  733                 uprintf("ELF binary type \"%u\" not known.\n",
  734                     hdr->e_ident[EI_OSABI]);
  735                 return (ENOEXEC);
  736         }
  737         if (hdr->e_type == ET_DYN) {
  738                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
  739                         return (ENOEXEC);
  740                 /*
  741                  * Honour the base load address from the dso if it is
  742                  * non-zero for some reason.
  743                  */
  744                 if (baddr == 0)
  745                         et_dyn_addr = ET_DYN_LOAD_ADDR;
  746                 else
  747                         et_dyn_addr = 0;
  748         } else
  749                 et_dyn_addr = 0;
  750         sv = brand_info->sysvec;
  751         if (interp != NULL && brand_info->interp_newpath != NULL)
  752                 newinterp = brand_info->interp_newpath;
  753 
  754         /*
  755          * Avoid a possible deadlock if the current address space is destroyed
  756          * and that address space maps the locked vnode.  In the common case,
  757          * the locked vnode's v_usecount is decremented but remains greater
  758          * than zero.  Consequently, the vnode lock is not needed by vrele().
  759          * However, in cases where the vnode lock is external, such as nullfs,
  760          * v_usecount may become zero.
  761          */
  762         VOP_UNLOCK(imgp->vp, 0);
  763 
  764         error = exec_new_vmspace(imgp, sv);
  765         imgp->proc->p_sysent = sv;
  766 
  767         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  768         if (error)
  769                 return (error);
  770 
  771         vmspace = imgp->proc->p_vmspace;
  772 
  773         for (i = 0; i < hdr->e_phnum; i++) {
  774                 switch (phdr[i].p_type) {
  775                 case PT_LOAD:   /* Loadable segment */
  776                         if (phdr[i].p_memsz == 0)
  777                                 break;
  778                         prot = __elfN(trans_prot)(phdr[i].p_flags);
  779 
  780 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
  781                         /*
  782                          * Some x86 binaries assume read == executable,
  783                          * notably the M3 runtime and therefore cvsup
  784                          */
  785                         if (prot & VM_PROT_READ)
  786                                 prot |= VM_PROT_EXECUTE;
  787 #endif
  788 
  789                         if ((error = __elfN(load_section)(vmspace,
  790                             imgp->object, phdr[i].p_offset,
  791                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
  792                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  793                             sv->sv_pagesize)) != 0)
  794                                 return (error);
  795 
  796                         /*
  797                          * If this segment contains the program headers,
  798                          * remember their virtual address for the AT_PHDR
  799                          * aux entry. Static binaries don't usually include
  800                          * a PT_PHDR entry.
  801                          */
  802                         if (phdr[i].p_offset == 0 &&
  803                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
  804                                 <= phdr[i].p_filesz)
  805                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
  806                                     et_dyn_addr;
  807 
  808                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
  809                         seg_size = round_page(phdr[i].p_memsz +
  810                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
  811 
  812                         /*
  813                          * Is this .text or .data?  We can't use
  814                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
  815                          * alpha terribly and possibly does other bad
  816                          * things so we stick to the old way of figuring
  817                          * it out:  If the segment contains the program
  818                          * entry point, it's a text segment, otherwise it
  819                          * is a data segment.
  820                          *
  821                          * Note that obreak() assumes that data_addr + 
  822                          * data_size == end of data load area, and the ELF
  823                          * file format expects segments to be sorted by
  824                          * address.  If multiple data segments exist, the
  825                          * last one will be used.
  826                          */
  827                         if (hdr->e_entry >= phdr[i].p_vaddr &&
  828                             hdr->e_entry < (phdr[i].p_vaddr +
  829                             phdr[i].p_memsz)) {
  830                                 text_size = seg_size;
  831                                 text_addr = seg_addr;
  832                                 entry = (u_long)hdr->e_entry + et_dyn_addr;
  833                         } else {
  834                                 data_size = seg_size;
  835                                 data_addr = seg_addr;
  836                         }
  837                         total_size += seg_size;
  838                         break;
  839                 case PT_PHDR:   /* Program header table info */
  840                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
  841                         break;
  842                 default:
  843                         break;
  844                 }
  845         }
  846         
  847         if (data_addr == 0 && data_size == 0) {
  848                 data_addr = text_addr;
  849                 data_size = text_size;
  850         }
  851 
  852         /*
  853          * Check limits.  It should be safe to check the
  854          * limits after loading the segments since we do
  855          * not actually fault in all the segments pages.
  856          */
  857         PROC_LOCK(imgp->proc);
  858         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
  859             text_size > maxtsiz ||
  860             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
  861                 PROC_UNLOCK(imgp->proc);
  862                 return (ENOMEM);
  863         }
  864 
  865         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
  866         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
  867         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
  868         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
  869 
  870         /*
  871          * We load the dynamic linker where a userland call
  872          * to mmap(0, ...) would put it.  The rationale behind this
  873          * calculation is that it leaves room for the heap to grow to
  874          * its maximum allowed size.
  875          */
  876         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
  877             lim_max(imgp->proc, RLIMIT_DATA));
  878         PROC_UNLOCK(imgp->proc);
  879 
  880         imgp->entry_addr = entry;
  881 
  882         if (interp != NULL) {
  883                 int have_interp = FALSE;
  884                 VOP_UNLOCK(imgp->vp, 0);
  885                 if (brand_info->emul_path != NULL &&
  886                     brand_info->emul_path[0] != '\0') {
  887                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
  888                         snprintf(path, MAXPATHLEN, "%s%s",
  889                             brand_info->emul_path, interp);
  890                         error = __elfN(load_file)(imgp->proc, path, &addr,
  891                             &imgp->entry_addr, sv->sv_pagesize);
  892                         free(path, M_TEMP);
  893                         if (error == 0)
  894                                 have_interp = TRUE;
  895                 }
  896                 if (!have_interp && newinterp != NULL) {
  897                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
  898                             &imgp->entry_addr, sv->sv_pagesize);
  899                         if (error == 0)
  900                                 have_interp = TRUE;
  901                 }
  902                 if (!have_interp) {
  903                         error = __elfN(load_file)(imgp->proc, interp, &addr,
  904                             &imgp->entry_addr, sv->sv_pagesize);
  905                 }
  906                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  907                 if (error != 0) {
  908                         uprintf("ELF interpreter %s not found\n", interp);
  909                         return (error);
  910                 }
  911         } else
  912                 addr = et_dyn_addr;
  913 
  914         /*
  915          * Construct auxargs table (used by the fixup routine)
  916          */
  917         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
  918         elf_auxargs->execfd = -1;
  919         elf_auxargs->phdr = proghdr;
  920         elf_auxargs->phent = hdr->e_phentsize;
  921         elf_auxargs->phnum = hdr->e_phnum;
  922         elf_auxargs->pagesz = PAGE_SIZE;
  923         elf_auxargs->base = addr;
  924         elf_auxargs->flags = 0;
  925         elf_auxargs->entry = entry;
  926 
  927         imgp->auxargs = elf_auxargs;
  928         imgp->interpreted = 0;
  929         imgp->proc->p_osrel = osrel;
  930 
  931         return (error);
  932 }
  933 
  934 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
  935 
  936 int
  937 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
  938 {
  939         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
  940         Elf_Addr *base;
  941         Elf_Addr *pos;
  942 
  943         base = (Elf_Addr *)*stack_base;
  944         pos = base + (imgp->args->argc + imgp->args->envc + 2);
  945 
  946         if (args->execfd != -1)
  947                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
  948         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
  949         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
  950         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
  951         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
  952         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
  953         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
  954         AUXARGS_ENTRY(pos, AT_BASE, args->base);
  955         if (imgp->execpathp != 0)
  956                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
  957         AUXARGS_ENTRY(pos, AT_NULL, 0);
  958 
  959         free(imgp->auxargs, M_TEMP);
  960         imgp->auxargs = NULL;
  961 
  962         base--;
  963         suword(base, (long)imgp->args->argc);
  964         *stack_base = (register_t *)base;
  965         return (0);
  966 }
  967 
  968 /*
  969  * Code for generating ELF core dumps.
  970  */
  971 
  972 typedef void (*segment_callback)(vm_map_entry_t, void *);
  973 
  974 /* Closure for cb_put_phdr(). */
  975 struct phdr_closure {
  976         Elf_Phdr *phdr;         /* Program header to fill in */
  977         Elf_Off offset;         /* Offset of segment in core file */
  978 };
  979 
  980 /* Closure for cb_size_segment(). */
  981 struct sseg_closure {
  982         int count;              /* Count of writable segments. */
  983         size_t size;            /* Total size of all writable segments. */
  984 };
  985 
  986 static void cb_put_phdr(vm_map_entry_t, void *);
  987 static void cb_size_segment(vm_map_entry_t, void *);
  988 static void each_writable_segment(struct thread *, segment_callback, void *);
  989 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
  990     int, void *, size_t);
  991 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
  992 static void __elfN(putnote)(void *, size_t *, const char *, int,
  993     const void *, size_t);
  994 
  995 int
  996 __elfN(coredump)(td, vp, limit)
  997         struct thread *td;
  998         struct vnode *vp;
  999         off_t limit;
 1000 {
 1001         struct ucred *cred = td->td_ucred;
 1002         int error = 0;
 1003         struct sseg_closure seginfo;
 1004         void *hdr;
 1005         size_t hdrsize;
 1006 
 1007         /* Size the program segments. */
 1008         seginfo.count = 0;
 1009         seginfo.size = 0;
 1010         each_writable_segment(td, cb_size_segment, &seginfo);
 1011 
 1012         /*
 1013          * Calculate the size of the core file header area by making
 1014          * a dry run of generating it.  Nothing is written, but the
 1015          * size is calculated.
 1016          */
 1017         hdrsize = 0;
 1018         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 1019 
 1020         if (hdrsize + seginfo.size >= limit)
 1021                 return (EFAULT);
 1022 
 1023         /*
 1024          * Allocate memory for building the header, fill it up,
 1025          * and write it out.
 1026          */
 1027         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 1028         if (hdr == NULL) {
 1029                 return (EINVAL);
 1030         }
 1031         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 1032 
 1033         /* Write the contents of all of the writable segments. */
 1034         if (error == 0) {
 1035                 Elf_Phdr *php;
 1036                 off_t offset;
 1037                 int i;
 1038 
 1039                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 1040                 offset = hdrsize;
 1041                 for (i = 0; i < seginfo.count; i++) {
 1042                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 1043                             (caddr_t)(uintptr_t)php->p_vaddr,
 1044                             php->p_filesz, offset, UIO_USERSPACE,
 1045                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1046                             curthread);
 1047                         if (error != 0)
 1048                                 break;
 1049                         offset += php->p_filesz;
 1050                         php++;
 1051                 }
 1052         }
 1053         free(hdr, M_TEMP);
 1054 
 1055         return (error);
 1056 }
 1057 
 1058 /*
 1059  * A callback for each_writable_segment() to write out the segment's
 1060  * program header entry.
 1061  */
 1062 static void
 1063 cb_put_phdr(entry, closure)
 1064         vm_map_entry_t entry;
 1065         void *closure;
 1066 {
 1067         struct phdr_closure *phc = (struct phdr_closure *)closure;
 1068         Elf_Phdr *phdr = phc->phdr;
 1069 
 1070         phc->offset = round_page(phc->offset);
 1071 
 1072         phdr->p_type = PT_LOAD;
 1073         phdr->p_offset = phc->offset;
 1074         phdr->p_vaddr = entry->start;
 1075         phdr->p_paddr = 0;
 1076         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 1077         phdr->p_align = PAGE_SIZE;
 1078         phdr->p_flags = __elfN(untrans_prot)(entry->protection);
 1079 
 1080         phc->offset += phdr->p_filesz;
 1081         phc->phdr++;
 1082 }
 1083 
 1084 /*
 1085  * A callback for each_writable_segment() to gather information about
 1086  * the number of segments and their total size.
 1087  */
 1088 static void
 1089 cb_size_segment(entry, closure)
 1090         vm_map_entry_t entry;
 1091         void *closure;
 1092 {
 1093         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 1094 
 1095         ssc->count++;
 1096         ssc->size += entry->end - entry->start;
 1097 }
 1098 
 1099 /*
 1100  * For each writable segment in the process's memory map, call the given
 1101  * function with a pointer to the map entry and some arbitrary
 1102  * caller-supplied data.
 1103  */
 1104 static void
 1105 each_writable_segment(td, func, closure)
 1106         struct thread *td;
 1107         segment_callback func;
 1108         void *closure;
 1109 {
 1110         struct proc *p = td->td_proc;
 1111         vm_map_t map = &p->p_vmspace->vm_map;
 1112         vm_map_entry_t entry;
 1113         vm_object_t backing_object, object;
 1114         boolean_t ignore_entry;
 1115 
 1116         vm_map_lock_read(map);
 1117         for (entry = map->header.next; entry != &map->header;
 1118             entry = entry->next) {
 1119                 /*
 1120                  * Don't dump inaccessible mappings, deal with legacy
 1121                  * coredump mode.
 1122                  *
 1123                  * Note that read-only segments related to the elf binary
 1124                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
 1125                  * need to arbitrarily ignore such segments.
 1126                  */
 1127                 if (elf_legacy_coredump) {
 1128                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
 1129                                 continue;
 1130                 } else {
 1131                         if ((entry->protection & VM_PROT_ALL) == 0)
 1132                                 continue;
 1133                 }
 1134 
 1135                 /*
 1136                  * Dont include memory segment in the coredump if
 1137                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
 1138                  * madvise(2).  Do not dump submaps (i.e. parts of the
 1139                  * kernel map).
 1140                  */
 1141                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
 1142                         continue;
 1143 
 1144                 if ((object = entry->object.vm_object) == NULL)
 1145                         continue;
 1146 
 1147                 /* Ignore memory-mapped devices and such things. */
 1148                 VM_OBJECT_LOCK(object);
 1149                 while ((backing_object = object->backing_object) != NULL) {
 1150                         VM_OBJECT_LOCK(backing_object);
 1151                         VM_OBJECT_UNLOCK(object);
 1152                         object = backing_object;
 1153                 }
 1154                 ignore_entry = object->type != OBJT_DEFAULT &&
 1155                     object->type != OBJT_SWAP && object->type != OBJT_VNODE;
 1156                 VM_OBJECT_UNLOCK(object);
 1157                 if (ignore_entry)
 1158                         continue;
 1159 
 1160                 (*func)(entry, closure);
 1161         }
 1162         vm_map_unlock_read(map);
 1163 }
 1164 
 1165 /*
 1166  * Write the core file header to the file, including padding up to
 1167  * the page boundary.
 1168  */
 1169 static int
 1170 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
 1171         struct thread *td;
 1172         struct vnode *vp;
 1173         struct ucred *cred;
 1174         int numsegs;
 1175         size_t hdrsize;
 1176         void *hdr;
 1177 {
 1178         size_t off;
 1179 
 1180         /* Fill in the header. */
 1181         bzero(hdr, hdrsize);
 1182         off = 0;
 1183         __elfN(puthdr)(td, hdr, &off, numsegs);
 1184 
 1185         /* Write it to the core file. */
 1186         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
 1187             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1188             td));
 1189 }
 1190 
 1191 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
 1192 #include <compat/freebsd32/freebsd32.h>
 1193 
 1194 typedef struct prstatus32 elf_prstatus_t;
 1195 typedef struct prpsinfo32 elf_prpsinfo_t;
 1196 typedef struct fpreg32 elf_prfpregset_t;
 1197 typedef struct fpreg32 elf_fpregset_t;
 1198 typedef struct reg32 elf_gregset_t;
 1199 typedef struct thrmisc32 elf_thrmisc_t;
 1200 #else
 1201 typedef prstatus_t elf_prstatus_t;
 1202 typedef prpsinfo_t elf_prpsinfo_t;
 1203 typedef prfpregset_t elf_prfpregset_t;
 1204 typedef prfpregset_t elf_fpregset_t;
 1205 typedef gregset_t elf_gregset_t;
 1206 typedef thrmisc_t elf_thrmisc_t;
 1207 #endif
 1208 
 1209 static void
 1210 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
 1211 {
 1212         struct {
 1213                 elf_prstatus_t status;
 1214                 elf_prfpregset_t fpregset;
 1215                 elf_prpsinfo_t psinfo;
 1216                 elf_thrmisc_t thrmisc;
 1217         } *tempdata;
 1218         elf_prstatus_t *status;
 1219         elf_prfpregset_t *fpregset;
 1220         elf_prpsinfo_t *psinfo;
 1221         elf_thrmisc_t *thrmisc;
 1222         struct proc *p;
 1223         struct thread *thr;
 1224         size_t ehoff, noteoff, notesz, phoff;
 1225 
 1226         p = td->td_proc;
 1227 
 1228         ehoff = *off;
 1229         *off += sizeof(Elf_Ehdr);
 1230 
 1231         phoff = *off;
 1232         *off += (numsegs + 1) * sizeof(Elf_Phdr);
 1233 
 1234         noteoff = *off;
 1235         /*
 1236          * Don't allocate space for the notes if we're just calculating
 1237          * the size of the header. We also don't collect the data.
 1238          */
 1239         if (dst != NULL) {
 1240                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
 1241                 status = &tempdata->status;
 1242                 fpregset = &tempdata->fpregset;
 1243                 psinfo = &tempdata->psinfo;
 1244                 thrmisc = &tempdata->thrmisc;
 1245         } else {
 1246                 tempdata = NULL;
 1247                 status = NULL;
 1248                 fpregset = NULL;
 1249                 psinfo = NULL;
 1250                 thrmisc = NULL;
 1251         }
 1252 
 1253         if (dst != NULL) {
 1254                 psinfo->pr_version = PRPSINFO_VERSION;
 1255                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
 1256                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
 1257                 /*
 1258                  * XXX - We don't fill in the command line arguments properly
 1259                  * yet.
 1260                  */
 1261                 strlcpy(psinfo->pr_psargs, p->p_comm,
 1262                     sizeof(psinfo->pr_psargs));
 1263         }
 1264         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
 1265             sizeof *psinfo);
 1266 
 1267         /*
 1268          * To have the debugger select the right thread (LWP) as the initial
 1269          * thread, we dump the state of the thread passed to us in td first.
 1270          * This is the thread that causes the core dump and thus likely to
 1271          * be the right thread one wants to have selected in the debugger.
 1272          */
 1273         thr = td;
 1274         while (thr != NULL) {
 1275                 if (dst != NULL) {
 1276                         status->pr_version = PRSTATUS_VERSION;
 1277                         status->pr_statussz = sizeof(elf_prstatus_t);
 1278                         status->pr_gregsetsz = sizeof(elf_gregset_t);
 1279                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
 1280                         status->pr_osreldate = osreldate;
 1281                         status->pr_cursig = p->p_sig;
 1282                         status->pr_pid = thr->td_tid;
 1283 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
 1284                         fill_regs32(thr, &status->pr_reg);
 1285                         fill_fpregs32(thr, fpregset);
 1286 #else
 1287                         fill_regs(thr, &status->pr_reg);
 1288                         fill_fpregs(thr, fpregset);
 1289 #endif
 1290                         memset(&thrmisc->_pad, 0, sizeof (thrmisc->_pad));
 1291                         strcpy(thrmisc->pr_tname, thr->td_name);
 1292                 }
 1293                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
 1294                     sizeof *status);
 1295                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
 1296                     sizeof *fpregset);
 1297                 __elfN(putnote)(dst, off, "FreeBSD", NT_THRMISC, thrmisc,
 1298                     sizeof *thrmisc);
 1299                 /*
 1300                  * Allow for MD specific notes, as well as any MD
 1301                  * specific preparations for writing MI notes.
 1302                  */
 1303                 __elfN(dump_thread)(thr, dst, off);
 1304 
 1305                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
 1306                     TAILQ_NEXT(thr, td_plist);
 1307                 if (thr == td)
 1308                         thr = TAILQ_NEXT(thr, td_plist);
 1309         }
 1310 
 1311         notesz = *off - noteoff;
 1312 
 1313         if (dst != NULL)
 1314                 free(tempdata, M_TEMP);
 1315 
 1316         /* Align up to a page boundary for the program segments. */
 1317         *off = round_page(*off);
 1318 
 1319         if (dst != NULL) {
 1320                 Elf_Ehdr *ehdr;
 1321                 Elf_Phdr *phdr;
 1322                 struct phdr_closure phc;
 1323 
 1324                 /*
 1325                  * Fill in the ELF header.
 1326                  */
 1327                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
 1328                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
 1329                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
 1330                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
 1331                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
 1332                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
 1333                 ehdr->e_ident[EI_DATA] = ELF_DATA;
 1334                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 1335                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
 1336                 ehdr->e_ident[EI_ABIVERSION] = 0;
 1337                 ehdr->e_ident[EI_PAD] = 0;
 1338                 ehdr->e_type = ET_CORE;
 1339 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
 1340                 ehdr->e_machine = ELF_ARCH32;
 1341 #else
 1342                 ehdr->e_machine = ELF_ARCH;
 1343 #endif
 1344                 ehdr->e_version = EV_CURRENT;
 1345                 ehdr->e_entry = 0;
 1346                 ehdr->e_phoff = phoff;
 1347                 ehdr->e_flags = 0;
 1348                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
 1349                 ehdr->e_phentsize = sizeof(Elf_Phdr);
 1350                 ehdr->e_phnum = numsegs + 1;
 1351                 ehdr->e_shentsize = sizeof(Elf_Shdr);
 1352                 ehdr->e_shnum = 0;
 1353                 ehdr->e_shstrndx = SHN_UNDEF;
 1354 
 1355                 /*
 1356                  * Fill in the program header entries.
 1357                  */
 1358                 phdr = (Elf_Phdr *)((char *)dst + phoff);
 1359 
 1360                 /* The note segement. */
 1361                 phdr->p_type = PT_NOTE;
 1362                 phdr->p_offset = noteoff;
 1363                 phdr->p_vaddr = 0;
 1364                 phdr->p_paddr = 0;
 1365                 phdr->p_filesz = notesz;
 1366                 phdr->p_memsz = 0;
 1367                 phdr->p_flags = 0;
 1368                 phdr->p_align = 0;
 1369                 phdr++;
 1370 
 1371                 /* All the writable segments from the program. */
 1372                 phc.phdr = phdr;
 1373                 phc.offset = *off;
 1374                 each_writable_segment(td, cb_put_phdr, &phc);
 1375         }
 1376 }
 1377 
 1378 static void
 1379 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
 1380     const void *desc, size_t descsz)
 1381 {
 1382         Elf_Note note;
 1383 
 1384         note.n_namesz = strlen(name) + 1;
 1385         note.n_descsz = descsz;
 1386         note.n_type = type;
 1387         if (dst != NULL)
 1388                 bcopy(&note, (char *)dst + *off, sizeof note);
 1389         *off += sizeof note;
 1390         if (dst != NULL)
 1391                 bcopy(name, (char *)dst + *off, note.n_namesz);
 1392         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 1393         if (dst != NULL)
 1394                 bcopy(desc, (char *)dst + *off, note.n_descsz);
 1395         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 1396 }
 1397 
 1398 /*
 1399  * Try to find the appropriate ABI-note section for checknote,
 1400  * fetch the osreldate for binary from the ELF OSABI-note. Only the
 1401  * first page of the image is searched, the same as for headers.
 1402  */
 1403 static boolean_t
 1404 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
 1405     int32_t *osrel)
 1406 {
 1407         const Elf_Note *note, *note0, *note_end;
 1408         const Elf_Phdr *phdr, *pnote;
 1409         const Elf_Ehdr *hdr;
 1410         const char *note_name;
 1411         int i;
 1412 
 1413         pnote = NULL;
 1414         hdr = (const Elf_Ehdr *)imgp->image_header;
 1415         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 1416 
 1417         for (i = 0; i < hdr->e_phnum; i++) {
 1418                 if (phdr[i].p_type == PT_NOTE) {
 1419                         pnote = &phdr[i];
 1420                         break;
 1421                 }
 1422         }
 1423 
 1424         if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
 1425             pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
 1426                 return (FALSE);
 1427 
 1428         note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
 1429         note_end = (const Elf_Note *)(imgp->image_header +
 1430             pnote->p_offset + pnote->p_filesz);
 1431         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
 1432                 if (!aligned(note, Elf32_Addr))
 1433                         return (FALSE);
 1434                 if (note->n_namesz != checknote->hdr.n_namesz ||
 1435                     note->n_descsz != checknote->hdr.n_descsz ||
 1436                     note->n_type != checknote->hdr.n_type)
 1437                         goto nextnote;
 1438                 note_name = (const char *)(note + 1);
 1439                 if (strncmp(checknote->vendor, note_name,
 1440                     checknote->hdr.n_namesz) != 0)
 1441                         goto nextnote;
 1442 
 1443                 /*
 1444                  * Fetch the osreldate for binary
 1445                  * from the ELF OSABI-note if necessary.
 1446                  */
 1447                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
 1448                     checknote->trans_osrel != NULL)
 1449                         return (checknote->trans_osrel(note, osrel));
 1450                 return (TRUE);
 1451 
 1452 nextnote:
 1453                 note = (const Elf_Note *)((const char *)(note + 1) +
 1454                     roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
 1455                     roundup2(note->n_descsz, sizeof(Elf32_Addr)));
 1456         }
 1457 
 1458         return (FALSE);
 1459 }
 1460 
 1461 /*
 1462  * Tell kern_execve.c about it, with a little help from the linker.
 1463  */
 1464 static struct execsw __elfN(execsw) = {
 1465         __CONCAT(exec_, __elfN(imgact)),
 1466         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
 1467 };
 1468 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
 1469 
 1470 static vm_prot_t
 1471 __elfN(trans_prot)(Elf_Word flags)
 1472 {
 1473         vm_prot_t prot;
 1474 
 1475         prot = 0;
 1476         if (flags & PF_X)
 1477                 prot |= VM_PROT_EXECUTE;
 1478         if (flags & PF_W)
 1479                 prot |= VM_PROT_WRITE;
 1480         if (flags & PF_R)
 1481                 prot |= VM_PROT_READ;
 1482         return (prot);
 1483 }
 1484 
 1485 static Elf_Word
 1486 __elfN(untrans_prot)(vm_prot_t prot)
 1487 {
 1488         Elf_Word flags;
 1489 
 1490         flags = 0;
 1491         if (prot & VM_PROT_EXECUTE)
 1492                 flags |= PF_X;
 1493         if (prot & VM_PROT_READ)
 1494                 flags |= PF_R;
 1495         if (prot & VM_PROT_WRITE)
 1496                 flags |= PF_W;
 1497         return (flags);
 1498 }

Cache object: c51d8c34470844a87b8af189a9fb1d13


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.