[ 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  -  FREEBSD8  -  FREEBSD7  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  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  -  OPENSOLARIS  -  minix-3-1-1  -  FREEBSD-LIBC  -  FREEBSD7-LIBC  -  FREEBSD6-LIBC  -  GLIBC27 
SearchContext: -  none  -  excerpts  -  bigexcerpts 

    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: head/sys/kern/imgact_elf.c 198202 2009-10-18 12:57:48Z 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 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
   74 #include <machine/fpu.h>
   75 #include <compat/ia32/ia32_reg.h>
   76 #endif
   77 
   78 #define OLD_EI_BRAND    8
   79 
   80 static int __elfN(check_header)(const Elf_Ehdr *hdr);
   81 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
   82     const char *interp, int32_t *osrel);
   83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
   84     u_long *entry, size_t pagesize);
   85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
   86     vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
   87     vm_prot_t prot, size_t pagesize);
   88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
   89 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
   90     int32_t *osrel);
   91 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
   92 static boolean_t __elfN(check_note)(struct image_params *imgp,
   93     Elf_Brandnote *checknote, int32_t *osrel);
   94 
   95 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
   96     "");
   97 
   98 int __elfN(fallback_brand) = -1;
   99 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
  100     fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
  101     __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
  102 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
  103     &__elfN(fallback_brand));
  104 
  105 static int elf_legacy_coredump = 0;
  106 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 
  107     &elf_legacy_coredump, 0, "");
  108 
  109 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
  110 
  111 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
  112 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
  113 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
  114 
  115 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
  116 
  117 Elf_Brandnote __elfN(freebsd_brandnote) = {
  118         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
  119         .hdr.n_descsz   = sizeof(int32_t),
  120         .hdr.n_type     = 1,
  121         .vendor         = FREEBSD_ABI_VENDOR,
  122         .flags          = BN_TRANSLATE_OSREL,
  123         .trans_osrel    = __elfN(freebsd_trans_osrel)
  124 };
  125 
  126 static boolean_t
  127 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
  128 {
  129         uintptr_t p;
  130 
  131         p = (uintptr_t)(note + 1);
  132         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  133         *osrel = *(const int32_t *)(p);
  134 
  135         return (TRUE);
  136 }
  137 
  138 static const char GNU_ABI_VENDOR[] = "GNU";
  139 static int GNU_KFREEBSD_ABI_DESC = 3;
  140 
  141 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
  142         .hdr.n_namesz   = sizeof(GNU_ABI_VENDOR),
  143         .hdr.n_descsz   = 16,   /* XXX at least 16 */
  144         .hdr.n_type     = 1,
  145         .vendor         = GNU_ABI_VENDOR,
  146         .flags          = BN_TRANSLATE_OSREL,
  147         .trans_osrel    = kfreebsd_trans_osrel
  148 };
  149 
  150 static boolean_t
  151 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
  152 {
  153         const Elf32_Word *desc;
  154         uintptr_t p;
  155 
  156         p = (uintptr_t)(note + 1);
  157         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  158 
  159         desc = (const Elf32_Word *)p;
  160         if (desc[0] != GNU_KFREEBSD_ABI_DESC)
  161                 return (FALSE);
  162 
  163         /*
  164          * Debian GNU/kFreeBSD embed the earliest compatible kernel version
  165          * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
  166          */
  167         *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
  168 
  169         return (TRUE);
  170 }
  171 
  172 int
  173 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
  174 {
  175         int i;
  176 
  177         for (i = 0; i < MAX_BRANDS; i++) {
  178                 if (elf_brand_list[i] == NULL) {
  179                         elf_brand_list[i] = entry;
  180                         break;
  181                 }
  182         }
  183         if (i == MAX_BRANDS) {
  184                 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
  185                         __func__, entry);
  186                 return (-1);
  187         }
  188         return (0);
  189 }
  190 
  191 int
  192 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
  193 {
  194         int i;
  195 
  196         for (i = 0; i < MAX_BRANDS; i++) {
  197                 if (elf_brand_list[i] == entry) {
  198                         elf_brand_list[i] = NULL;
  199                         break;
  200                 }
  201         }
  202         if (i == MAX_BRANDS)
  203                 return (-1);
  204         return (0);
  205 }
  206 
  207 int
  208 __elfN(brand_inuse)(Elf_Brandinfo *entry)
  209 {
  210         struct proc *p;
  211         int rval = FALSE;
  212 
  213         sx_slock(&allproc_lock);
  214         FOREACH_PROC_IN_SYSTEM(p) {
  215                 if (p->p_sysent == entry->sysvec) {
  216                         rval = TRUE;
  217                         break;
  218                 }
  219         }
  220         sx_sunlock(&allproc_lock);
  221 
  222         return (rval);
  223 }
  224 
  225 static Elf_Brandinfo *
  226 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
  227     int32_t *osrel)
  228 {
  229         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  230         Elf_Brandinfo *bi;
  231         boolean_t ret;
  232         int i;
  233 
  234         /*
  235          * We support four types of branding -- (1) the ELF EI_OSABI field
  236          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
  237          * branding w/in the ELF header, (3) path of the `interp_path'
  238          * field, and (4) the ".note.ABI-tag" ELF section.
  239          */
  240 
  241         /* Look for an ".note.ABI-tag" ELF section */
  242         for (i = 0; i < MAX_BRANDS; i++) {
  243                 bi = elf_brand_list[i];
  244                 if (bi == NULL)
  245                         continue;
  246                 if (hdr->e_machine == bi->machine && (bi->flags &
  247                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
  248                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
  249                         if (ret)
  250                                 return (bi);
  251                 }
  252         }
  253 
  254         /* If the executable has a brand, search for it in the brand list. */
  255         for (i = 0; i < MAX_BRANDS; i++) {
  256                 bi = elf_brand_list[i];
  257                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  258                         continue;
  259                 if (hdr->e_machine == bi->machine &&
  260                     (hdr->e_ident[EI_OSABI] == bi->brand ||
  261                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
  262                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
  263                         return (bi);
  264         }
  265 
  266         /* Lacking a known brand, search for a recognized interpreter. */
  267         if (interp != NULL) {
  268                 for (i = 0; i < MAX_BRANDS; i++) {
  269                         bi = elf_brand_list[i];
  270                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  271                                 continue;
  272                         if (hdr->e_machine == bi->machine &&
  273                             strcmp(interp, bi->interp_path) == 0)
  274                                 return (bi);
  275                 }
  276         }
  277 
  278         /* Lacking a recognized interpreter, try the default brand */
  279         for (i = 0; i < MAX_BRANDS; i++) {
  280                 bi = elf_brand_list[i];
  281                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  282                         continue;
  283                 if (hdr->e_machine == bi->machine &&
  284                     __elfN(fallback_brand) == bi->brand)
  285                         return (bi);
  286         }
  287         return (NULL);
  288 }
  289 
  290 static int
  291 __elfN(check_header)(const Elf_Ehdr *hdr)
  292 {
  293         Elf_Brandinfo *bi;
  294         int i;
  295 
  296         if (!IS_ELF(*hdr) ||
  297             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
  298             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
  299             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
  300             hdr->e_phentsize != sizeof(Elf_Phdr) ||
  301             hdr->e_version != ELF_TARG_VER)
  302                 return (ENOEXEC);
  303 
  304         /*
  305          * Make sure we have at least one brand for this machine.
  306          */
  307 
  308         for (i = 0; i < MAX_BRANDS; i++) {
  309                 bi = elf_brand_list[i];
  310                 if (bi != NULL && bi->machine == hdr->e_machine)
  311                         break;
  312         }
  313         if (i == MAX_BRANDS)
  314                 return (ENOEXEC);
  315 
  316         return (0);
  317 }
  318 
  319 static int
  320 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  321     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
  322 {
  323         struct sf_buf *sf;
  324         int error;
  325         vm_offset_t off;
  326 
  327         /*
  328          * Create the page if it doesn't exist yet. Ignore errors.
  329          */
  330         vm_map_lock(map);
  331         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
  332             VM_PROT_ALL, VM_PROT_ALL, 0);
  333         vm_map_unlock(map);
  334 
  335         /*
  336          * Find the page from the underlying object.
  337          */
  338         if (object) {
  339                 sf = vm_imgact_map_page(object, offset);
  340                 if (sf == NULL)
  341                         return (KERN_FAILURE);
  342                 off = offset - trunc_page(offset);
  343                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
  344                     end - start);
  345                 vm_imgact_unmap_page(sf);
  346                 if (error) {
  347                         return (KERN_FAILURE);
  348                 }
  349         }
  350 
  351         return (KERN_SUCCESS);
  352 }
  353 
  354 static int
  355 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  356     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
  357 {
  358         struct sf_buf *sf;
  359         vm_offset_t off;
  360         vm_size_t sz;
  361         int error, rv;
  362 
  363         if (start != trunc_page(start)) {
  364                 rv = __elfN(map_partial)(map, object, offset, start,
  365                     round_page(start), prot);
  366                 if (rv)
  367                         return (rv);
  368                 offset += round_page(start) - start;
  369                 start = round_page(start);
  370         }
  371         if (end != round_page(end)) {
  372                 rv = __elfN(map_partial)(map, object, offset +
  373                     trunc_page(end) - start, trunc_page(end), end, prot);
  374                 if (rv)
  375                         return (rv);
  376                 end = trunc_page(end);
  377         }
  378         if (end > start) {
  379                 if (offset & PAGE_MASK) {
  380                         /*
  381                          * The mapping is not page aligned. This means we have
  382                          * to copy the data. Sigh.
  383                          */
  384                         rv = vm_map_find(map, NULL, 0, &start, end - start,
  385                             FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
  386                         if (rv)
  387                                 return (rv);
  388                         if (object == NULL)
  389                                 return (KERN_SUCCESS);
  390                         for (; start < end; start += sz) {
  391                                 sf = vm_imgact_map_page(object, offset);
  392                                 if (sf == NULL)
  393                                         return (KERN_FAILURE);
  394                                 off = offset - trunc_page(offset);
  395                                 sz = end - start;
  396                                 if (sz > PAGE_SIZE - off)
  397                                         sz = PAGE_SIZE - off;
  398                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  399                                     (caddr_t)start, sz);
  400                                 vm_imgact_unmap_page(sf);
  401                                 if (error) {
  402                                         return (KERN_FAILURE);
  403                                 }
  404                                 offset += sz;
  405                         }
  406                         rv = KERN_SUCCESS;
  407                 } else {
  408                         vm_object_reference(object);
  409                         vm_map_lock(map);
  410                         rv = vm_map_insert(map, object, offset, start, end,
  411                             prot, VM_PROT_ALL, cow);
  412                         vm_map_unlock(map);
  413                         if (rv != KERN_SUCCESS)
  414                                 vm_object_deallocate(object);
  415                 }
  416                 return (rv);
  417         } else {
  418                 return (KERN_SUCCESS);
  419         }
  420 }
  421 
  422 static int
  423 __elfN(load_section)(struct vmspace *vmspace,
  424         vm_object_t object, vm_offset_t offset,
  425         caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
  426         size_t pagesize)
  427 {
  428         struct sf_buf *sf;
  429         size_t map_len;
  430         vm_offset_t map_addr;
  431         int error, rv, cow;
  432         size_t copy_len;
  433         vm_offset_t file_addr;
  434 
  435         /*
  436          * It's necessary to fail if the filsz + offset taken from the
  437          * header is greater than the actual file pager object's size.
  438          * If we were to allow this, then the vm_map_find() below would
  439          * walk right off the end of the file object and into the ether.
  440          *
  441          * While I'm here, might as well check for something else that
  442          * is invalid: filsz cannot be greater than memsz.
  443          */
  444         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
  445             filsz > memsz) {
  446                 uprintf("elf_load_section: truncated ELF file\n");
  447                 return (ENOEXEC);
  448         }
  449 
  450         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
  451         file_addr = trunc_page_ps(offset, pagesize);
  452 
  453         /*
  454          * We have two choices.  We can either clear the data in the last page
  455          * of an oversized mapping, or we can start the anon mapping a page
  456          * early and copy the initialized data into that first page.  We
  457          * choose the second..
  458          */
  459         if (memsz > filsz)
  460                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
  461         else
  462                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
  463 
  464         if (map_len != 0) {
  465                 /* cow flags: don't dump readonly sections in core */
  466                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
  467                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
  468 
  469                 rv = __elfN(map_insert)(&vmspace->vm_map,
  470                                       object,
  471                                       file_addr,        /* file offset */
  472                                       map_addr,         /* virtual start */
  473                                       map_addr + map_len,/* virtual end */
  474                                       prot,
  475                                       cow);
  476                 if (rv != KERN_SUCCESS)
  477                         return (EINVAL);
  478 
  479                 /* we can stop now if we've covered it all */
  480                 if (memsz == filsz) {
  481                         return (0);
  482                 }
  483         }
  484 
  485 
  486         /*
  487          * We have to get the remaining bit of the file into the first part
  488          * of the oversized map segment.  This is normally because the .data
  489          * segment in the file is extended to provide bss.  It's a neat idea
  490          * to try and save a page, but it's a pain in the behind to implement.
  491          */
  492         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
  493         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
  494         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
  495             map_addr;
  496 
  497         /* This had damn well better be true! */
  498         if (map_len != 0) {
  499                 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
  500                     map_addr + map_len, VM_PROT_ALL, 0);
  501                 if (rv != KERN_SUCCESS) {
  502                         return (EINVAL);
  503                 }
  504         }
  505 
  506         if (copy_len != 0) {
  507                 vm_offset_t off;
  508 
  509                 sf = vm_imgact_map_page(object, offset + filsz);
  510                 if (sf == NULL)
  511                         return (EIO);
  512 
  513                 /* send the page fragment to user space */
  514                 off = trunc_page_ps(offset + filsz, pagesize) -
  515                     trunc_page(offset + filsz);
  516                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  517                     (caddr_t)map_addr, copy_len);
  518                 vm_imgact_unmap_page(sf);
  519                 if (error) {
  520                         return (error);
  521                 }
  522         }
  523 
  524         /*
  525          * set it to the specified protection.
  526          * XXX had better undo the damage from pasting over the cracks here!
  527          */
  528         vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
  529             round_page(map_addr + map_len),  prot, FALSE);
  530 
  531         return (0);
  532 }
  533 
  534 /*
  535  * Load the file "file" into memory.  It may be either a shared object
  536  * or an executable.
  537  *
  538  * The "addr" reference parameter is in/out.  On entry, it specifies
  539  * the address where a shared object should be loaded.  If the file is
  540  * an executable, this value is ignored.  On exit, "addr" specifies
  541  * where the file was actually loaded.
  542  *
  543  * The "entry" reference parameter is out only.  On exit, it specifies
  544  * the entry point for the loaded file.
  545  */
  546 static int
  547 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  548         u_long *entry, size_t pagesize)
  549 {
  550         struct {
  551                 struct nameidata nd;
  552                 struct vattr attr;
  553                 struct image_params image_params;
  554         } *tempdata;
  555         const Elf_Ehdr *hdr = NULL;
  556         const Elf_Phdr *phdr = NULL;
  557         struct nameidata *nd;
  558         struct vmspace *vmspace = p->p_vmspace;
  559         struct vattr *attr;
  560         struct image_params *imgp;
  561         vm_prot_t prot;
  562         u_long rbase;
  563         u_long base_addr = 0;
  564         int vfslocked, error, i, numsegs;
  565 
  566         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
  567         nd = &tempdata->nd;
  568         attr = &tempdata->attr;
  569         imgp = &tempdata->image_params;
  570 
  571         /*
  572          * Initialize part of the common data
  573          */
  574         imgp->proc = p;
  575         imgp->attr = attr;
  576         imgp->firstpage = NULL;
  577         imgp->image_header = NULL;
  578         imgp->object = NULL;
  579         imgp->execlabel = NULL;
  580 
  581         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
  582             curthread);
  583         vfslocked = 0;
  584         if ((error = namei(nd)) != 0) {
  585                 nd->ni_vp = NULL;
  586                 goto fail;
  587         }
  588         vfslocked = NDHASGIANT(nd);
  589         NDFREE(nd, NDF_ONLY_PNBUF);
  590         imgp->vp = nd->ni_vp;
  591 
  592         /*
  593          * Check permissions, modes, uid, etc on the file, and "open" it.
  594          */
  595         error = exec_check_permissions(imgp);
  596         if (error)
  597                 goto fail;
  598 
  599         error = exec_map_first_page(imgp);
  600         if (error)
  601                 goto fail;
  602 
  603         /*
  604          * Also make certain that the interpreter stays the same, so set
  605          * its VV_TEXT flag, too.
  606          */
  607         nd->ni_vp->v_vflag |= VV_TEXT;
  608 
  609         imgp->object = nd->ni_vp->v_object;
  610 
  611         hdr = (const Elf_Ehdr *)imgp->image_header;
  612         if ((error = __elfN(check_header)(hdr)) != 0)
  613                 goto fail;
  614         if (hdr->e_type == ET_DYN)
  615                 rbase = *addr;
  616         else if (hdr->e_type == ET_EXEC)
  617                 rbase = 0;
  618         else {
  619                 error = ENOEXEC;
  620                 goto fail;
  621         }
  622 
  623         /* Only support headers that fit within first page for now      */
  624         /*    (multiplication of two Elf_Half fields will not overflow) */
  625         if ((hdr->e_phoff > PAGE_SIZE) ||
  626             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
  627                 error = ENOEXEC;
  628                 goto fail;
  629         }
  630 
  631         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  632         if (!aligned(phdr, Elf_Addr)) {
  633                 error = ENOEXEC;
  634                 goto fail;
  635         }
  636 
  637         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
  638                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
  639                         /* Loadable segment */
  640                         prot = 0;
  641                         if (phdr[i].p_flags & PF_X)
  642                                 prot |= VM_PROT_EXECUTE;
  643                         if (phdr[i].p_flags & PF_W)
  644                                 prot |= VM_PROT_WRITE;
  645                         if (phdr[i].p_flags & PF_R)
  646                                 prot |= VM_PROT_READ;
  647 
  648                         if ((error = __elfN(load_section)(vmspace,
  649                             imgp->object, phdr[i].p_offset,
  650                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
  651                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  652                             pagesize)) != 0)
  653                                 goto fail;
  654                         /*
  655                          * Establish the base address if this is the
  656                          * first segment.
  657                          */
  658                         if (numsegs == 0)
  659                                 base_addr = trunc_page(phdr[i].p_vaddr +
  660                                     rbase);
  661                         numsegs++;
  662                 }
  663         }
  664         *addr = base_addr;
  665         *entry = (unsigned long)hdr->e_entry + rbase;
  666 
  667 fail:
  668         if (imgp->firstpage)
  669                 exec_unmap_first_page(imgp);
  670 
  671         if (nd->ni_vp)
  672                 vput(nd->ni_vp);
  673 
  674         VFS_UNLOCK_GIANT(vfslocked);
  675         free(tempdata, M_TEMP);
  676 
  677         return (error);
  678 }
  679 
  680 static int
  681 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
  682 {
  683         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  684         const Elf_Phdr *phdr;
  685         Elf_Auxargs *elf_auxargs;
  686         struct vmspace *vmspace;
  687         vm_prot_t prot;
  688         u_long text_size = 0, data_size = 0, total_size = 0;
  689         u_long text_addr = 0, data_addr = 0;
  690         u_long seg_size, seg_addr;
  691         u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
  692         int32_t osrel = 0;
  693         int error = 0, i, n;
  694         const char *interp = NULL, *newinterp = NULL;
  695         Elf_Brandinfo *brand_info;
  696         char *path;
  697         struct sysentvec *sv;
  698 
  699         /*
  700          * Do we have a valid ELF header ?
  701          *
  702          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
  703          * if particular brand doesn't support it.
  704          */
  705         if (__elfN(check_header)(hdr) != 0 ||
  706             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
  707                 return (-1);
  708 
  709         /*
  710          * From here on down, we return an errno, not -1, as we've
  711          * detected an ELF file.
  712          */
  713 
  714         if ((hdr->e_phoff > PAGE_SIZE) ||
  715             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
  716                 /* Only support headers in first page for now */
  717                 return (ENOEXEC);
  718         }
  719         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  720         if (!aligned(phdr, Elf_Addr))
  721                 return (ENOEXEC);
  722         n = 0;
  723         baddr = 0;
  724         for (i = 0; i < hdr->e_phnum; i++) {
  725                 if (phdr[i].p_type == PT_LOAD) {
  726                         if (n == 0)
  727                                 baddr = phdr[i].p_vaddr;
  728                         n++;
  729                         continue;
  730                 }
  731                 if (phdr[i].p_type == PT_INTERP) {
  732                         /* Path to interpreter */
  733                         if (phdr[i].p_filesz > MAXPATHLEN ||
  734                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
  735                                 return (ENOEXEC);
  736                         interp = imgp->image_header + phdr[i].p_offset;
  737                         continue;
  738                 }
  739         }
  740 
  741         brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
  742         if (brand_info == NULL) {
  743                 uprintf("ELF binary type \"%u\" not known.\n",
  744                     hdr->e_ident[EI_OSABI]);
  745                 return (ENOEXEC);
  746         }
  747         if (hdr->e_type == ET_DYN) {
  748                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
  749                         return (ENOEXEC);
  750                 /*
  751                  * Honour the base load address from the dso if it is
  752                  * non-zero for some reason.
  753                  */
  754                 if (baddr == 0)
  755                         et_dyn_addr = ET_DYN_LOAD_ADDR;
  756                 else
  757                         et_dyn_addr = 0;
  758         } else
  759                 et_dyn_addr = 0;
  760         sv = brand_info->sysvec;
  761         if (interp != NULL && brand_info->interp_newpath != NULL)
  762                 newinterp = brand_info->interp_newpath;
  763 
  764         /*
  765          * Avoid a possible deadlock if the current address space is destroyed
  766          * and that address space maps the locked vnode.  In the common case,
  767          * the locked vnode's v_usecount is decremented but remains greater
  768          * than zero.  Consequently, the vnode lock is not needed by vrele().
  769          * However, in cases where the vnode lock is external, such as nullfs,
  770          * v_usecount may become zero.
  771          */
  772         VOP_UNLOCK(imgp->vp, 0);
  773 
  774         error = exec_new_vmspace(imgp, sv);
  775         imgp->proc->p_sysent = sv;
  776 
  777         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  778         if (error)
  779                 return (error);
  780 
  781         vmspace = imgp->proc->p_vmspace;
  782 
  783         for (i = 0; i < hdr->e_phnum; i++) {
  784                 switch (phdr[i].p_type) {
  785                 case PT_LOAD:   /* Loadable segment */
  786                         if (phdr[i].p_memsz == 0)
  787                                 break;
  788                         prot = 0;
  789                         if (phdr[i].p_flags & PF_X)
  790                                 prot |= VM_PROT_EXECUTE;
  791                         if (phdr[i].p_flags & PF_W)
  792                                 prot |= VM_PROT_WRITE;
  793                         if (phdr[i].p_flags & PF_R)
  794                                 prot |= VM_PROT_READ;
  795 
  796 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
  797                         /*
  798                          * Some x86 binaries assume read == executable,
  799                          * notably the M3 runtime and therefore cvsup
  800                          */
  801                         if (prot & VM_PROT_READ)
  802                                 prot |= VM_PROT_EXECUTE;
  803 #endif
  804 
  805                         if ((error = __elfN(load_section)(vmspace,
  806                             imgp->object, phdr[i].p_offset,
  807                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
  808                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  809                             sv->sv_pagesize)) != 0)
  810                                 return (error);
  811 
  812                         /*
  813                          * If this segment contains the program headers,
  814                          * remember their virtual address for the AT_PHDR
  815                          * aux entry. Static binaries don't usually include
  816                          * a PT_PHDR entry.
  817                          */
  818                         if (phdr[i].p_offset == 0 &&
  819                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
  820                                 <= phdr[i].p_filesz)
  821                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
  822                                     et_dyn_addr;
  823 
  824                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
  825                         seg_size = round_page(phdr[i].p_memsz +
  826                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
  827 
  828                         /*
  829                          * Is this .text or .data?  We can't use
  830                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
  831                          * alpha terribly and possibly does other bad
  832                          * things so we stick to the old way of figuring
  833                          * it out:  If the segment contains the program
  834                          * entry point, it's a text segment, otherwise it
  835                          * is a data segment.
  836                          *
  837                          * Note that obreak() assumes that data_addr + 
  838                          * data_size == end of data load area, and the ELF
  839                          * file format expects segments to be sorted by
  840                          * address.  If multiple data segments exist, the
  841                          * last one will be used.
  842                          */
  843                         if (hdr->e_entry >= phdr[i].p_vaddr &&
  844                             hdr->e_entry < (phdr[i].p_vaddr +
  845                             phdr[i].p_memsz)) {
  846                                 text_size = seg_size;
  847                                 text_addr = seg_addr;
  848                                 entry = (u_long)hdr->e_entry + et_dyn_addr;
  849                         } else {
  850                                 data_size = seg_size;
  851                                 data_addr = seg_addr;
  852                         }
  853                         total_size += seg_size;
  854                         break;
  855                 case PT_PHDR:   /* Program header table info */
  856                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
  857                         break;
  858                 default:
  859                         break;
  860                 }
  861         }
  862         
  863         if (data_addr == 0 && data_size == 0) {
  864                 data_addr = text_addr;
  865                 data_size = text_size;
  866         }
  867 
  868         /*
  869          * Check limits.  It should be safe to check the
  870          * limits after loading the segments since we do
  871          * not actually fault in all the segments pages.
  872          */
  873         PROC_LOCK(imgp->proc);
  874         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
  875             text_size > maxtsiz ||
  876             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
  877                 PROC_UNLOCK(imgp->proc);
  878                 return (ENOMEM);
  879         }
  880 
  881         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
  882         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
  883         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
  884         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
  885 
  886         /*
  887          * We load the dynamic linker where a userland call
  888          * to mmap(0, ...) would put it.  The rationale behind this
  889          * calculation is that it leaves room for the heap to grow to
  890          * its maximum allowed size.
  891          */
  892         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
  893             lim_max(imgp->proc, RLIMIT_DATA));
  894         PROC_UNLOCK(imgp->proc);
  895 
  896         imgp->entry_addr = entry;
  897 
  898         if (interp != NULL) {
  899                 int have_interp = FALSE;
  900                 VOP_UNLOCK(imgp->vp, 0);
  901                 if (brand_info->emul_path != NULL &&
  902                     brand_info->emul_path[0] != '\0') {
  903                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
  904                         snprintf(path, MAXPATHLEN, "%s%s",
  905                             brand_info->emul_path, interp);
  906                         error = __elfN(load_file)(imgp->proc, path, &addr,
  907                             &imgp->entry_addr, sv->sv_pagesize);
  908                         free(path, M_TEMP);
  909                         if (error == 0)
  910                                 have_interp = TRUE;
  911                 }
  912                 if (!have_interp && newinterp != NULL) {
  913                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
  914                             &imgp->entry_addr, sv->sv_pagesize);
  915                         if (error == 0)
  916                                 have_interp = TRUE;
  917                 }
  918                 if (!have_interp) {
  919                         error = __elfN(load_file)(imgp->proc, interp, &addr,
  920                             &imgp->entry_addr, sv->sv_pagesize);
  921                 }
  922                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  923                 if (error != 0) {
  924                         uprintf("ELF interpreter %s not found\n", interp);
  925                         return (error);
  926                 }
  927         } else
  928                 addr = et_dyn_addr;
  929 
  930         /*
  931          * Construct auxargs table (used by the fixup routine)
  932          */
  933         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
  934         elf_auxargs->execfd = -1;
  935         elf_auxargs->phdr = proghdr;
  936         elf_auxargs->phent = hdr->e_phentsize;
  937         elf_auxargs->phnum = hdr->e_phnum;
  938         elf_auxargs->pagesz = PAGE_SIZE;
  939         elf_auxargs->base = addr;
  940         elf_auxargs->flags = 0;
  941         elf_auxargs->entry = entry;
  942 
  943         imgp->auxargs = elf_auxargs;
  944         imgp->interpreted = 0;
  945         imgp->proc->p_osrel = osrel;
  946 
  947         return (error);
  948 }
  949 
  950 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
  951 
  952 int
  953 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
  954 {
  955         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
  956         Elf_Addr *base;
  957         Elf_Addr *pos;
  958 
  959         base = (Elf_Addr *)*stack_base;
  960         pos = base + (imgp->args->argc + imgp->args->envc + 2);
  961 
  962         if (args->execfd != -1)
  963                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
  964         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
  965         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
  966         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
  967         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
  968         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
  969         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
  970         AUXARGS_ENTRY(pos, AT_BASE, args->base);
  971         if (imgp->execpathp != 0)
  972                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
  973         AUXARGS_ENTRY(pos, AT_NULL, 0);
  974 
  975         free(imgp->auxargs, M_TEMP);
  976         imgp->auxargs = NULL;
  977 
  978         base--;
  979         suword(base, (long)imgp->args->argc);
  980         *stack_base = (register_t *)base;
  981         return (0);
  982 }
  983 
  984 /*
  985  * Code for generating ELF core dumps.
  986  */
  987 
  988 typedef void (*segment_callback)(vm_map_entry_t, void *);
  989 
  990 /* Closure for cb_put_phdr(). */
  991 struct phdr_closure {
  992         Elf_Phdr *phdr;         /* Program header to fill in */
  993         Elf_Off offset;         /* Offset of segment in core file */
  994 };
  995 
  996 /* Closure for cb_size_segment(). */
  997 struct sseg_closure {
  998         int count;              /* Count of writable segments. */
  999         size_t size;            /* Total size of all writable segments. */
 1000 };
 1001 
 1002 static void cb_put_phdr(vm_map_entry_t, void *);
 1003 static void cb_size_segment(vm_map_entry_t, void *);
 1004 static void each_writable_segment(struct thread *, segment_callback, void *);
 1005 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
 1006     int, void *, size_t);
 1007 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
 1008 static void __elfN(putnote)(void *, size_t *, const char *, int,
 1009     const void *, size_t);
 1010 
 1011 int
 1012 __elfN(coredump)(td, vp, limit)
 1013         struct thread *td;
 1014         struct vnode *vp;
 1015         off_t limit;
 1016 {
 1017         struct ucred *cred = td->td_ucred;
 1018         int error = 0;
 1019         struct sseg_closure seginfo;
 1020         void *hdr;
 1021         size_t hdrsize;
 1022 
 1023         /* Size the program segments. */
 1024         seginfo.count = 0;
 1025         seginfo.size = 0;
 1026         each_writable_segment(td, cb_size_segment, &seginfo);
 1027 
 1028         /*
 1029          * Calculate the size of the core file header area by making
 1030          * a dry run of generating it.  Nothing is written, but the
 1031          * size is calculated.
 1032          */
 1033         hdrsize = 0;
 1034         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 1035 
 1036         if (hdrsize + seginfo.size >= limit)
 1037                 return (EFAULT);
 1038 
 1039         /*
 1040          * Allocate memory for building the header, fill it up,
 1041          * and write it out.
 1042          */
 1043         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 1044         if (hdr == NULL) {
 1045                 return (EINVAL);
 1046         }
 1047         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 1048 
 1049         /* Write the contents of all of the writable segments. */
 1050         if (error == 0) {
 1051                 Elf_Phdr *php;
 1052                 off_t offset;
 1053                 int i;
 1054 
 1055                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 1056                 offset = hdrsize;
 1057                 for (i = 0; i < seginfo.count; i++) {
 1058                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 1059                             (caddr_t)(uintptr_t)php->p_vaddr,
 1060                             php->p_filesz, offset, UIO_USERSPACE,
 1061                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1062                             curthread);
 1063                         if (error != 0)
 1064                                 break;
 1065                         offset += php->p_filesz;
 1066                         php++;
 1067                 }
 1068         }
 1069         free(hdr, M_TEMP);
 1070 
 1071         return (error);
 1072 }
 1073 
 1074 /*
 1075  * A callback for each_writable_segment() to write out the segment's
 1076  * program header entry.
 1077  */
 1078 static void
 1079 cb_put_phdr(entry, closure)
 1080         vm_map_entry_t entry;
 1081         void *closure;
 1082 {
 1083         struct phdr_closure *phc = (struct phdr_closure *)closure;
 1084         Elf_Phdr *phdr = phc->phdr;
 1085 
 1086         phc->offset = round_page(phc->offset);
 1087 
 1088         phdr->p_type = PT_LOAD;
 1089         phdr->p_offset = phc->offset;
 1090         phdr->p_vaddr = entry->start;
 1091         phdr->p_paddr = 0;
 1092         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 1093         phdr->p_align = PAGE_SIZE;
 1094         phdr->p_flags = 0;
 1095         if (entry->protection & VM_PROT_READ)
 1096                 phdr->p_flags |= PF_R;
 1097         if (entry->protection & VM_PROT_WRITE)
 1098                 phdr->p_flags |= PF_W;
 1099         if (entry->protection & VM_PROT_EXECUTE)
 1100                 phdr->p_flags |= PF_X;
 1101 
 1102         phc->offset += phdr->p_filesz;
 1103         phc->phdr++;
 1104 }
 1105 
 1106 /*
 1107  * A callback for each_writable_segment() to gather information about
 1108  * the number of segments and their total size.
 1109  */
 1110 static void
 1111 cb_size_segment(entry, closure)
 1112         vm_map_entry_t entry;
 1113         void *closure;
 1114 {
 1115         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 1116 
 1117         ssc->count++;
 1118         ssc->size += entry->end - entry->start;
 1119 }
 1120 
 1121 /*
 1122  * For each writable segment in the process's memory map, call the given
 1123  * function with a pointer to the map entry and some arbitrary
 1124  * caller-supplied data.
 1125  */
 1126 static void
 1127 each_writable_segment(td, func, closure)
 1128         struct thread *td;
 1129         segment_callback func;
 1130         void *closure;
 1131 {
 1132         struct proc *p = td->td_proc;
 1133         vm_map_t map = &p->p_vmspace->vm_map;
 1134         vm_map_entry_t entry;
 1135         vm_object_t backing_object, object;
 1136         boolean_t ignore_entry;
 1137 
 1138         vm_map_lock_read(map);
 1139         for (entry = map->header.next; entry != &map->header;
 1140             entry = entry->next) {
 1141                 /*
 1142                  * Don't dump inaccessible mappings, deal with legacy
 1143                  * coredump mode.
 1144                  *
 1145                  * Note that read-only segments related to the elf binary
 1146                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
 1147                  * need to arbitrarily ignore such segments.
 1148                  */
 1149                 if (elf_legacy_coredump) {
 1150                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
 1151                                 continue;
 1152                 } else {
 1153                         if ((entry->protection & VM_PROT_ALL) == 0)
 1154                                 continue;
 1155                 }
 1156 
 1157                 /*
 1158                  * Dont include memory segment in the coredump if
 1159                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
 1160                  * madvise(2).  Do not dump submaps (i.e. parts of the
 1161                  * kernel map).
 1162                  */
 1163                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
 1164                         continue;
 1165 
 1166                 if ((object = entry->object.vm_object) == NULL)
 1167                         continue;
 1168 
 1169                 /* Ignore memory-mapped devices and such things. */
 1170                 VM_OBJECT_LOCK(object);
 1171                 while ((backing_object = object->backing_object) != NULL) {
 1172                         VM_OBJECT_LOCK(backing_object);
 1173                         VM_OBJECT_UNLOCK(object);
 1174                         object = backing_object;
 1175                 }
 1176                 ignore_entry = object->type != OBJT_DEFAULT &&
 1177                     object->type != OBJT_SWAP && object->type != OBJT_VNODE;
 1178                 VM_OBJECT_UNLOCK(object);
 1179                 if (ignore_entry)
 1180                         continue;
 1181 
 1182                 (*func)(entry, closure);
 1183         }
 1184         vm_map_unlock_read(map);
 1185 }
 1186 
 1187 /*
 1188  * Write the core file header to the file, including padding up to
 1189  * the page boundary.
 1190  */
 1191 static int
 1192 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
 1193         struct thread *td;
 1194         struct vnode *vp;
 1195         struct ucred *cred;
 1196         int numsegs;
 1197         size_t hdrsize;
 1198         void *hdr;
 1199 {
 1200         size_t off;
 1201 
 1202         /* Fill in the header. */
 1203         bzero(hdr, hdrsize);
 1204         off = 0;
 1205         __elfN(puthdr)(td, hdr, &off, numsegs);
 1206 
 1207         /* Write it to the core file. */
 1208         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
 1209             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1210             td));
 1211 }
 1212 
 1213 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1214 typedef struct prstatus32 elf_prstatus_t;
 1215 typedef struct prpsinfo32 elf_prpsinfo_t;
 1216 typedef struct fpreg32 elf_prfpregset_t;
 1217 typedef struct fpreg32 elf_fpregset_t;
 1218 typedef struct reg32 elf_gregset_t;
 1219 #else
 1220 typedef prstatus_t elf_prstatus_t;
 1221 typedef prpsinfo_t elf_prpsinfo_t;
 1222 typedef prfpregset_t elf_prfpregset_t;
 1223 typedef prfpregset_t elf_fpregset_t;
 1224 typedef gregset_t elf_gregset_t;
 1225 #endif
 1226 
 1227 static void
 1228 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
 1229 {
 1230         struct {
 1231                 elf_prstatus_t status;
 1232                 elf_prfpregset_t fpregset;
 1233                 elf_prpsinfo_t psinfo;
 1234         } *tempdata;
 1235         elf_prstatus_t *status;
 1236         elf_prfpregset_t *fpregset;
 1237         elf_prpsinfo_t *psinfo;
 1238         struct proc *p;
 1239         struct thread *thr;
 1240         size_t ehoff, noteoff, notesz, phoff;
 1241 
 1242         p = td->td_proc;
 1243 
 1244         ehoff = *off;
 1245         *off += sizeof(Elf_Ehdr);
 1246 
 1247         phoff = *off;
 1248         *off += (numsegs + 1) * sizeof(Elf_Phdr);
 1249 
 1250         noteoff = *off;
 1251         /*
 1252          * Don't allocate space for the notes if we're just calculating
 1253          * the size of the header. We also don't collect the data.
 1254          */
 1255         if (dst != NULL) {
 1256                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
 1257                 status = &tempdata->status;
 1258                 fpregset = &tempdata->fpregset;
 1259                 psinfo = &tempdata->psinfo;
 1260         } else {
 1261                 tempdata = NULL;
 1262                 status = NULL;
 1263                 fpregset = NULL;
 1264                 psinfo = NULL;
 1265         }
 1266 
 1267         if (dst != NULL) {
 1268                 psinfo->pr_version = PRPSINFO_VERSION;
 1269                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
 1270                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
 1271                 /*
 1272                  * XXX - We don't fill in the command line arguments properly
 1273                  * yet.
 1274                  */
 1275                 strlcpy(psinfo->pr_psargs, p->p_comm,
 1276                     sizeof(psinfo->pr_psargs));
 1277         }
 1278         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
 1279             sizeof *psinfo);
 1280 
 1281         /*
 1282          * To have the debugger select the right thread (LWP) as the initial
 1283          * thread, we dump the state of the thread passed to us in td first.
 1284          * This is the thread that causes the core dump and thus likely to
 1285          * be the right thread one wants to have selected in the debugger.
 1286          */
 1287         thr = td;
 1288         while (thr != NULL) {
 1289                 if (dst != NULL) {
 1290                         status->pr_version = PRSTATUS_VERSION;
 1291                         status->pr_statussz = sizeof(elf_prstatus_t);
 1292                         status->pr_gregsetsz = sizeof(elf_gregset_t);
 1293                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
 1294                         status->pr_osreldate = osreldate;
 1295                         status->pr_cursig = p->p_sig;
 1296                         status->pr_pid = thr->td_tid;
 1297 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1298                         fill_regs32(thr, &status->pr_reg);
 1299                         fill_fpregs32(thr, fpregset);
 1300 #else
 1301                         fill_regs(thr, &status->pr_reg);
 1302                         fill_fpregs(thr, fpregset);
 1303 #endif
 1304                 }
 1305                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
 1306                     sizeof *status);
 1307                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
 1308                     sizeof *fpregset);
 1309                 /*
 1310                  * Allow for MD specific notes, as well as any MD
 1311                  * specific preparations for writing MI notes.
 1312                  */
 1313                 __elfN(dump_thread)(thr, dst, off);
 1314 
 1315                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
 1316                     TAILQ_NEXT(thr, td_plist);
 1317                 if (thr == td)
 1318                         thr = TAILQ_NEXT(thr, td_plist);
 1319         }
 1320 
 1321         notesz = *off - noteoff;
 1322 
 1323         if (dst != NULL)
 1324                 free(tempdata, M_TEMP);
 1325 
 1326         /* Align up to a page boundary for the program segments. */
 1327         *off = round_page(*off);
 1328 
 1329         if (dst != NULL) {
 1330                 Elf_Ehdr *ehdr;
 1331                 Elf_Phdr *phdr;
 1332                 struct phdr_closure phc;
 1333 
 1334                 /*
 1335                  * Fill in the ELF header.
 1336                  */
 1337                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
 1338                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
 1339                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
 1340                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
 1341                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
 1342                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
 1343                 ehdr->e_ident[EI_DATA] = ELF_DATA;
 1344                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 1345                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
 1346                 ehdr->e_ident[EI_ABIVERSION] = 0;
 1347                 ehdr->e_ident[EI_PAD] = 0;
 1348                 ehdr->e_type = ET_CORE;
 1349 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1350                 ehdr->e_machine = EM_386;
 1351 #else
 1352                 ehdr->e_machine = ELF_ARCH;
 1353 #endif
 1354                 ehdr->e_version = EV_CURRENT;
 1355                 ehdr->e_entry = 0;
 1356                 ehdr->e_phoff = phoff;
 1357                 ehdr->e_flags = 0;
 1358                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
 1359                 ehdr->e_phentsize = sizeof(Elf_Phdr);
 1360                 ehdr->e_phnum = numsegs + 1;
 1361                 ehdr->e_shentsize = sizeof(Elf_Shdr);
 1362                 ehdr->e_shnum = 0;
 1363                 ehdr->e_shstrndx = SHN_UNDEF;
 1364 
 1365                 /*
 1366                  * Fill in the program header entries.
 1367                  */
 1368                 phdr = (Elf_Phdr *)((char *)dst + phoff);
 1369 
 1370                 /* The note segement. */
 1371                 phdr->p_type = PT_NOTE;
 1372                 phdr->p_offset = noteoff;
 1373                 phdr->p_vaddr = 0;
 1374                 phdr->p_paddr = 0;
 1375                 phdr->p_filesz = notesz;
 1376                 phdr->p_memsz = 0;
 1377                 phdr->p_flags = 0;
 1378                 phdr->p_align = 0;
 1379                 phdr++;
 1380 
 1381                 /* All the writable segments from the program. */
 1382                 phc.phdr = phdr;
 1383                 phc.offset = *off;
 1384                 each_writable_segment(td, cb_put_phdr, &phc);
 1385         }
 1386 }
 1387 
 1388 static void
 1389 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
 1390     const void *desc, size_t descsz)
 1391 {
 1392         Elf_Note note;
 1393 
 1394         note.n_namesz = strlen(name) + 1;
 1395         note.n_descsz = descsz;
 1396         note.n_type = type;
 1397         if (dst != NULL)
 1398                 bcopy(&note, (char *)dst + *off, sizeof note);
 1399         *off += sizeof note;
 1400         if (dst != NULL)
 1401                 bcopy(name, (char *)dst + *off, note.n_namesz);
 1402         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 1403         if (dst != NULL)
 1404                 bcopy(desc, (char *)dst + *off, note.n_descsz);
 1405         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 1406 }
 1407 
 1408 /*
 1409  * Try to find the appropriate ABI-note section for checknote,
 1410  * fetch the osreldate for binary from the ELF OSABI-note. Only the
 1411  * first page of the image is searched, the same as for headers.
 1412  */
 1413 static boolean_t
 1414 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
 1415     int32_t *osrel)
 1416 {
 1417         const Elf_Note *note, *note0, *note_end;
 1418         const Elf_Phdr *phdr, *pnote;
 1419         const Elf_Ehdr *hdr;
 1420         const char *note_name;
 1421         int i;
 1422 
 1423         pnote = NULL;
 1424         hdr = (const Elf_Ehdr *)imgp->image_header;
 1425         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 1426 
 1427         for (i = 0; i < hdr->e_phnum; i++) {
 1428                 if (phdr[i].p_type == PT_NOTE) {
 1429                         pnote = &phdr[i];
 1430                         break;
 1431                 }
 1432         }
 1433 
 1434         if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
 1435             pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
 1436                 return (FALSE);
 1437 
 1438         note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
 1439         note_end = (const Elf_Note *)(imgp->image_header +
 1440             pnote->p_offset + pnote->p_filesz);
 1441         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
 1442                 if (!aligned(note, Elf32_Addr))
 1443                         return (FALSE);
 1444                 if (note->n_namesz != checknote->hdr.n_namesz ||
 1445                     note->n_descsz != checknote->hdr.n_descsz ||
 1446                     note->n_type != checknote->hdr.n_type)
 1447                         goto nextnote;
 1448                 note_name = (const char *)(note + 1);
 1449                 if (strncmp(checknote->vendor, note_name,
 1450                     checknote->hdr.n_namesz) != 0)
 1451                         goto nextnote;
 1452 
 1453                 /*
 1454                  * Fetch the osreldate for binary
 1455                  * from the ELF OSABI-note if necessary.
 1456                  */
 1457                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
 1458                     checknote->trans_osrel != NULL)
 1459                         return (checknote->trans_osrel(note, osrel));
 1460                 return (TRUE);
 1461 
 1462 nextnote:
 1463                 note = (const Elf_Note *)((const char *)(note + 1) +
 1464                     roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
 1465                     roundup2(note->n_descsz, sizeof(Elf32_Addr)));
 1466         }
 1467 
 1468         return (FALSE);
 1469 }
 1470 
 1471 /*
 1472  * Tell kern_execve.c about it, with a little help from the linker.
 1473  */
 1474 static struct execsw __elfN(execsw) = {
 1475         __CONCAT(exec_, __elfN(imgact)),
 1476         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
 1477 };
 1478 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));

Cache object: aee43032b71731dee6b50e04c298aab8


[ 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.