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  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  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  -  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: stable/7/sys/kern/imgact_elf.c 202646 2010-01-19 19:59:03Z 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_trace = 0;
  106 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
  107 
  108 static int elf_legacy_coredump = 0;
  109 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW, 
  110     &elf_legacy_coredump, 0, "");
  111 
  112 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
  113 
  114 #define trunc_page_ps(va, ps)   ((va) & ~(ps - 1))
  115 #define round_page_ps(va, ps)   (((va) + (ps - 1)) & ~(ps - 1))
  116 #define aligned(a, t)   (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
  117 
  118 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
  119 
  120 Elf_Brandnote __elfN(freebsd_brandnote) = {
  121         .hdr.n_namesz   = sizeof(FREEBSD_ABI_VENDOR),
  122         .hdr.n_descsz   = sizeof(int32_t),
  123         .hdr.n_type     = 1,
  124         .vendor         = FREEBSD_ABI_VENDOR,
  125         .flags          = BN_TRANSLATE_OSREL,
  126         .trans_osrel    = __elfN(freebsd_trans_osrel)
  127 };
  128 
  129 static boolean_t
  130 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
  131 {
  132         uintptr_t p;
  133 
  134         p = (uintptr_t)(note + 1);
  135         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  136         *osrel = *(const int32_t *)(p);
  137 
  138         return (TRUE);
  139 }
  140 
  141 static const char GNU_ABI_VENDOR[] = "GNU";
  142 static int GNU_KFREEBSD_ABI_DESC = 3;
  143 
  144 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
  145         .hdr.n_namesz   = sizeof(GNU_ABI_VENDOR),
  146         .hdr.n_descsz   = 16,   /* XXX at least 16 */
  147         .hdr.n_type     = 1,
  148         .vendor         = GNU_ABI_VENDOR,
  149         .flags          = BN_TRANSLATE_OSREL,
  150         .trans_osrel    = kfreebsd_trans_osrel
  151 };
  152 
  153 static boolean_t
  154 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
  155 {
  156         const Elf32_Word *desc;
  157         uintptr_t p;
  158 
  159         p = (uintptr_t)(note + 1);
  160         p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
  161 
  162         desc = (const Elf32_Word *)p;
  163         if (desc[0] != GNU_KFREEBSD_ABI_DESC)
  164                 return (FALSE);
  165 
  166         /*
  167          * Debian GNU/kFreeBSD embed the earliest compatible kernel version
  168          * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
  169          */
  170         *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
  171 
  172         return (TRUE);
  173 }
  174 
  175 int
  176 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
  177 {
  178         int i;
  179 
  180         for (i = 0; i < MAX_BRANDS; i++) {
  181                 if (elf_brand_list[i] == NULL) {
  182                         elf_brand_list[i] = entry;
  183                         break;
  184                 }
  185         }
  186         if (i == MAX_BRANDS)
  187                 return (-1);
  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         if (curthread->td_proc != p)
  567                 panic("elf_load_file - thread");        /* XXXKSE DIAGNOSTIC */
  568 
  569         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
  570         nd = &tempdata->nd;
  571         attr = &tempdata->attr;
  572         imgp = &tempdata->image_params;
  573 
  574         /*
  575          * Initialize part of the common data
  576          */
  577         imgp->proc = p;
  578         imgp->attr = attr;
  579         imgp->firstpage = NULL;
  580         imgp->image_header = NULL;
  581         imgp->object = NULL;
  582         imgp->execlabel = NULL;
  583 
  584         /* XXXKSE */
  585         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
  586             curthread);
  587         vfslocked = 0;
  588         if ((error = namei(nd)) != 0) {
  589                 nd->ni_vp = NULL;
  590                 goto fail;
  591         }
  592         vfslocked = NDHASGIANT(nd);
  593         NDFREE(nd, NDF_ONLY_PNBUF);
  594         imgp->vp = nd->ni_vp;
  595 
  596         /*
  597          * Check permissions, modes, uid, etc on the file, and "open" it.
  598          */
  599         error = exec_check_permissions(imgp);
  600         if (error)
  601                 goto fail;
  602 
  603         error = exec_map_first_page(imgp);
  604         if (error)
  605                 goto fail;
  606 
  607         /*
  608          * Also make certain that the interpreter stays the same, so set
  609          * its VV_TEXT flag, too.
  610          */
  611         nd->ni_vp->v_vflag |= VV_TEXT;
  612 
  613         imgp->object = nd->ni_vp->v_object;
  614 
  615         hdr = (const Elf_Ehdr *)imgp->image_header;
  616         if ((error = __elfN(check_header)(hdr)) != 0)
  617                 goto fail;
  618         if (hdr->e_type == ET_DYN)
  619                 rbase = *addr;
  620         else if (hdr->e_type == ET_EXEC)
  621                 rbase = 0;
  622         else {
  623                 error = ENOEXEC;
  624                 goto fail;
  625         }
  626 
  627         /* Only support headers that fit within first page for now      */
  628         /*    (multiplication of two Elf_Half fields will not overflow) */
  629         if ((hdr->e_phoff > PAGE_SIZE) ||
  630             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
  631                 error = ENOEXEC;
  632                 goto fail;
  633         }
  634 
  635         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  636         if (!aligned(phdr, Elf_Addr)) {
  637                 error = ENOEXEC;
  638                 goto fail;
  639         }
  640 
  641         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
  642                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
  643                         /* Loadable segment */
  644                         prot = 0;
  645                         if (phdr[i].p_flags & PF_X)
  646                                 prot |= VM_PROT_EXECUTE;
  647                         if (phdr[i].p_flags & PF_W)
  648                                 prot |= VM_PROT_WRITE;
  649                         if (phdr[i].p_flags & PF_R)
  650                                 prot |= VM_PROT_READ;
  651 
  652                         if ((error = __elfN(load_section)(vmspace,
  653                             imgp->object, phdr[i].p_offset,
  654                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
  655                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  656                             pagesize)) != 0)
  657                                 goto fail;
  658                         /*
  659                          * Establish the base address if this is the
  660                          * first segment.
  661                          */
  662                         if (numsegs == 0)
  663                                 base_addr = trunc_page(phdr[i].p_vaddr +
  664                                     rbase);
  665                         numsegs++;
  666                 }
  667         }
  668         *addr = base_addr;
  669         *entry = (unsigned long)hdr->e_entry + rbase;
  670 
  671 fail:
  672         if (imgp->firstpage)
  673                 exec_unmap_first_page(imgp);
  674 
  675         if (nd->ni_vp)
  676                 vput(nd->ni_vp);
  677 
  678         VFS_UNLOCK_GIANT(vfslocked);
  679         free(tempdata, M_TEMP);
  680 
  681         return (error);
  682 }
  683 
  684 static int
  685 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
  686 {
  687         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  688         const Elf_Phdr *phdr;
  689         Elf_Auxargs *elf_auxargs;
  690         struct vmspace *vmspace;
  691         vm_prot_t prot;
  692         u_long text_size = 0, data_size = 0, total_size = 0;
  693         u_long text_addr = 0, data_addr = 0;
  694         u_long seg_size, seg_addr;
  695         u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
  696         int32_t osrel = 0;
  697         int error = 0, i, n;
  698         const char *interp = NULL, *newinterp = NULL;
  699         Elf_Brandinfo *brand_info;
  700         char *path;
  701         struct thread *td = curthread;
  702         struct sysentvec *sv;
  703 
  704         /*
  705          * Do we have a valid ELF header ?
  706          *
  707          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
  708          * if particular brand doesn't support it.
  709          */
  710         if (__elfN(check_header)(hdr) != 0 ||
  711             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
  712                 return (-1);
  713 
  714         /*
  715          * From here on down, we return an errno, not -1, as we've
  716          * detected an ELF file.
  717          */
  718 
  719         if ((hdr->e_phoff > PAGE_SIZE) ||
  720             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
  721                 /* Only support headers in first page for now */
  722                 return (ENOEXEC);
  723         }
  724         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  725         if (!aligned(phdr, Elf_Addr))
  726                 return (ENOEXEC);
  727         n = 0;
  728         baddr = 0;
  729         for (i = 0; i < hdr->e_phnum; i++) {
  730                 if (phdr[i].p_type == PT_LOAD) {
  731                         if (n == 0)
  732                                 baddr = phdr[i].p_vaddr;
  733                         n++;
  734                         continue;
  735                 }
  736                 if (phdr[i].p_type == PT_INTERP) {
  737                         /* Path to interpreter */
  738                         if (phdr[i].p_filesz > MAXPATHLEN ||
  739                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
  740                                 return (ENOEXEC);
  741                         interp = imgp->image_header + phdr[i].p_offset;
  742                         continue;
  743                 }
  744         }
  745 
  746         brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
  747         if (brand_info == NULL) {
  748                 uprintf("ELF binary type \"%u\" not known.\n",
  749                     hdr->e_ident[EI_OSABI]);
  750                 return (ENOEXEC);
  751         }
  752         if (hdr->e_type == ET_DYN) {
  753                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
  754                         return (ENOEXEC);
  755                 /*
  756                  * Honour the base load address from the dso if it is
  757                  * non-zero for some reason.
  758                  */
  759                 if (baddr == 0)
  760                         et_dyn_addr = ET_DYN_LOAD_ADDR;
  761                 else
  762                         et_dyn_addr = 0;
  763         } else
  764                 et_dyn_addr = 0;
  765         sv = brand_info->sysvec;
  766         if (interp != NULL && brand_info->interp_newpath != NULL)
  767                 newinterp = brand_info->interp_newpath;
  768 
  769         /*
  770          * Avoid a possible deadlock if the current address space is destroyed
  771          * and that address space maps the locked vnode.  In the common case,
  772          * the locked vnode's v_usecount is decremented but remains greater
  773          * than zero.  Consequently, the vnode lock is not needed by vrele().
  774          * However, in cases where the vnode lock is external, such as nullfs,
  775          * v_usecount may become zero.
  776          */
  777         VOP_UNLOCK(imgp->vp, 0, td);
  778 
  779         error = exec_new_vmspace(imgp, sv);
  780         imgp->proc->p_sysent = sv;
  781 
  782         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
  783         if (error)
  784                 return (error);
  785 
  786         vmspace = imgp->proc->p_vmspace;
  787 
  788         for (i = 0; i < hdr->e_phnum; i++) {
  789                 switch (phdr[i].p_type) {
  790                 case PT_LOAD:   /* Loadable segment */
  791                         if (phdr[i].p_memsz == 0)
  792                                 break;
  793                         prot = 0;
  794                         if (phdr[i].p_flags & PF_X)
  795                                 prot |= VM_PROT_EXECUTE;
  796                         if (phdr[i].p_flags & PF_W)
  797                                 prot |= VM_PROT_WRITE;
  798                         if (phdr[i].p_flags & PF_R)
  799                                 prot |= VM_PROT_READ;
  800 
  801 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
  802                         /*
  803                          * Some x86 binaries assume read == executable,
  804                          * notably the M3 runtime and therefore cvsup
  805                          */
  806                         if (prot & VM_PROT_READ)
  807                                 prot |= VM_PROT_EXECUTE;
  808 #endif
  809 
  810                         if ((error = __elfN(load_section)(vmspace,
  811                             imgp->object, phdr[i].p_offset,
  812                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
  813                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  814                             sv->sv_pagesize)) != 0)
  815                                 return (error);
  816 
  817                         /*
  818                          * If this segment contains the program headers,
  819                          * remember their virtual address for the AT_PHDR
  820                          * aux entry. Static binaries don't usually include
  821                          * a PT_PHDR entry.
  822                          */
  823                         if (phdr[i].p_offset == 0 &&
  824                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
  825                                 <= phdr[i].p_filesz)
  826                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
  827                                     et_dyn_addr;
  828 
  829                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
  830                         seg_size = round_page(phdr[i].p_memsz +
  831                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
  832 
  833                         /*
  834                          * Is this .text or .data?  We can't use
  835                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
  836                          * alpha terribly and possibly does other bad
  837                          * things so we stick to the old way of figuring
  838                          * it out:  If the segment contains the program
  839                          * entry point, it's a text segment, otherwise it
  840                          * is a data segment.
  841                          *
  842                          * Note that obreak() assumes that data_addr + 
  843                          * data_size == end of data load area, and the ELF
  844                          * file format expects segments to be sorted by
  845                          * address.  If multiple data segments exist, the
  846                          * last one will be used.
  847                          */
  848                         if (hdr->e_entry >= phdr[i].p_vaddr &&
  849                             hdr->e_entry < (phdr[i].p_vaddr +
  850                             phdr[i].p_memsz)) {
  851                                 text_size = seg_size;
  852                                 text_addr = seg_addr;
  853                                 entry = (u_long)hdr->e_entry + et_dyn_addr;
  854                         } else {
  855                                 data_size = seg_size;
  856                                 data_addr = seg_addr;
  857                         }
  858                         total_size += seg_size;
  859                         break;
  860                 case PT_PHDR:   /* Program header table info */
  861                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
  862                         break;
  863                 default:
  864                         break;
  865                 }
  866         }
  867         
  868         if (data_addr == 0 && data_size == 0) {
  869                 data_addr = text_addr;
  870                 data_size = text_size;
  871         }
  872 
  873         /*
  874          * Check limits.  It should be safe to check the
  875          * limits after loading the segments since we do
  876          * not actually fault in all the segments pages.
  877          */
  878         PROC_LOCK(imgp->proc);
  879         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
  880             text_size > maxtsiz ||
  881             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
  882                 PROC_UNLOCK(imgp->proc);
  883                 return (ENOMEM);
  884         }
  885 
  886         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
  887         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
  888         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
  889         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
  890 
  891         /*
  892          * We load the dynamic linker where a userland call
  893          * to mmap(0, ...) would put it.  The rationale behind this
  894          * calculation is that it leaves room for the heap to grow to
  895          * its maximum allowed size.
  896          */
  897         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
  898             lim_max(imgp->proc, RLIMIT_DATA));
  899         PROC_UNLOCK(imgp->proc);
  900 
  901         imgp->entry_addr = entry;
  902 
  903         if (interp != NULL) {
  904                 int have_interp = FALSE;
  905                 VOP_UNLOCK(imgp->vp, 0, td);
  906                 if (brand_info->emul_path != NULL &&
  907                     brand_info->emul_path[0] != '\0') {
  908                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
  909                         snprintf(path, MAXPATHLEN, "%s%s",
  910                             brand_info->emul_path, interp);
  911                         error = __elfN(load_file)(imgp->proc, path, &addr,
  912                             &imgp->entry_addr, sv->sv_pagesize);
  913                         free(path, M_TEMP);
  914                         if (error == 0)
  915                                 have_interp = TRUE;
  916                 }
  917                 if (!have_interp && newinterp != NULL) {
  918                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
  919                             &imgp->entry_addr, sv->sv_pagesize);
  920                         if (error == 0)
  921                                 have_interp = TRUE;
  922                 }
  923                 if (!have_interp) {
  924                         error = __elfN(load_file)(imgp->proc, interp, &addr,
  925                             &imgp->entry_addr, sv->sv_pagesize);
  926                 }
  927                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
  928                 if (error != 0) {
  929                         uprintf("ELF interpreter %s not found\n", interp);
  930                         return (error);
  931                 }
  932         } else
  933                 addr = et_dyn_addr;
  934 
  935         /*
  936          * Construct auxargs table (used by the fixup routine)
  937          */
  938         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
  939         elf_auxargs->execfd = -1;
  940         elf_auxargs->phdr = proghdr;
  941         elf_auxargs->phent = hdr->e_phentsize;
  942         elf_auxargs->phnum = hdr->e_phnum;
  943         elf_auxargs->pagesz = PAGE_SIZE;
  944         elf_auxargs->base = addr;
  945         elf_auxargs->flags = 0;
  946         elf_auxargs->entry = entry;
  947         elf_auxargs->trace = elf_trace;
  948 
  949         imgp->auxargs = elf_auxargs;
  950         imgp->interpreted = 0;
  951         imgp->proc->p_osrel = osrel;
  952 
  953         return (error);
  954 }
  955 
  956 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
  957 
  958 int
  959 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
  960 {
  961         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
  962         Elf_Addr *base;
  963         Elf_Addr *pos;
  964 
  965         base = (Elf_Addr *)*stack_base;
  966         pos = base + (imgp->args->argc + imgp->args->envc + 2);
  967 
  968         if (args->execfd != -1) {
  969                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
  970         }
  971         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
  972         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
  973         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
  974         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
  975         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
  976         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
  977         AUXARGS_ENTRY(pos, AT_BASE, args->base);
  978         if (imgp->execpathp != 0)
  979                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
  980         AUXARGS_ENTRY(pos, AT_NULL, 0);
  981 
  982         free(imgp->auxargs, M_TEMP);
  983         imgp->auxargs = NULL;
  984 
  985         base--;
  986         suword(base, (long)imgp->args->argc);
  987         *stack_base = (register_t *)base;
  988         return (0);
  989 }
  990 
  991 /*
  992  * Code for generating ELF core dumps.
  993  */
  994 
  995 typedef void (*segment_callback)(vm_map_entry_t, void *);
  996 
  997 /* Closure for cb_put_phdr(). */
  998 struct phdr_closure {
  999         Elf_Phdr *phdr;         /* Program header to fill in */
 1000         Elf_Off offset;         /* Offset of segment in core file */
 1001 };
 1002 
 1003 /* Closure for cb_size_segment(). */
 1004 struct sseg_closure {
 1005         int count;              /* Count of writable segments. */
 1006         size_t size;            /* Total size of all writable segments. */
 1007 };
 1008 
 1009 static void cb_put_phdr(vm_map_entry_t, void *);
 1010 static void cb_size_segment(vm_map_entry_t, void *);
 1011 static void each_writable_segment(struct thread *, segment_callback, void *);
 1012 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
 1013     int, void *, size_t);
 1014 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
 1015 static void __elfN(putnote)(void *, size_t *, const char *, int,
 1016     const void *, size_t);
 1017 
 1018 int
 1019 __elfN(coredump)(td, vp, limit)
 1020         struct thread *td;
 1021         struct vnode *vp;
 1022         off_t limit;
 1023 {
 1024         struct ucred *cred = td->td_ucred;
 1025         int error = 0;
 1026         struct sseg_closure seginfo;
 1027         void *hdr;
 1028         size_t hdrsize;
 1029 
 1030         /* Size the program segments. */
 1031         seginfo.count = 0;
 1032         seginfo.size = 0;
 1033         each_writable_segment(td, cb_size_segment, &seginfo);
 1034 
 1035         /*
 1036          * Calculate the size of the core file header area by making
 1037          * a dry run of generating it.  Nothing is written, but the
 1038          * size is calculated.
 1039          */
 1040         hdrsize = 0;
 1041         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 1042 
 1043         if (hdrsize + seginfo.size >= limit)
 1044                 return (EFAULT);
 1045 
 1046         /*
 1047          * Allocate memory for building the header, fill it up,
 1048          * and write it out.
 1049          */
 1050         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 1051         if (hdr == NULL) {
 1052                 return (EINVAL);
 1053         }
 1054         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 1055 
 1056         /* Write the contents of all of the writable segments. */
 1057         if (error == 0) {
 1058                 Elf_Phdr *php;
 1059                 off_t offset;
 1060                 int i;
 1061 
 1062                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 1063                 offset = hdrsize;
 1064                 for (i = 0; i < seginfo.count; i++) {
 1065                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 1066                             (caddr_t)(uintptr_t)php->p_vaddr,
 1067                             php->p_filesz, offset, UIO_USERSPACE,
 1068                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1069                             curthread); /* XXXKSE */
 1070                         if (error != 0)
 1071                                 break;
 1072                         offset += php->p_filesz;
 1073                         php++;
 1074                 }
 1075         }
 1076         free(hdr, M_TEMP);
 1077 
 1078         return (error);
 1079 }
 1080 
 1081 /*
 1082  * A callback for each_writable_segment() to write out the segment's
 1083  * program header entry.
 1084  */
 1085 static void
 1086 cb_put_phdr(entry, closure)
 1087         vm_map_entry_t entry;
 1088         void *closure;
 1089 {
 1090         struct phdr_closure *phc = (struct phdr_closure *)closure;
 1091         Elf_Phdr *phdr = phc->phdr;
 1092 
 1093         phc->offset = round_page(phc->offset);
 1094 
 1095         phdr->p_type = PT_LOAD;
 1096         phdr->p_offset = phc->offset;
 1097         phdr->p_vaddr = entry->start;
 1098         phdr->p_paddr = 0;
 1099         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 1100         phdr->p_align = PAGE_SIZE;
 1101         phdr->p_flags = 0;
 1102         if (entry->protection & VM_PROT_READ)
 1103                 phdr->p_flags |= PF_R;
 1104         if (entry->protection & VM_PROT_WRITE)
 1105                 phdr->p_flags |= PF_W;
 1106         if (entry->protection & VM_PROT_EXECUTE)
 1107                 phdr->p_flags |= PF_X;
 1108 
 1109         phc->offset += phdr->p_filesz;
 1110         phc->phdr++;
 1111 }
 1112 
 1113 /*
 1114  * A callback for each_writable_segment() to gather information about
 1115  * the number of segments and their total size.
 1116  */
 1117 static void
 1118 cb_size_segment(entry, closure)
 1119         vm_map_entry_t entry;
 1120         void *closure;
 1121 {
 1122         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 1123 
 1124         ssc->count++;
 1125         ssc->size += entry->end - entry->start;
 1126 }
 1127 
 1128 /*
 1129  * For each writable segment in the process's memory map, call the given
 1130  * function with a pointer to the map entry and some arbitrary
 1131  * caller-supplied data.
 1132  */
 1133 static void
 1134 each_writable_segment(td, func, closure)
 1135         struct thread *td;
 1136         segment_callback func;
 1137         void *closure;
 1138 {
 1139         struct proc *p = td->td_proc;
 1140         vm_map_t map = &p->p_vmspace->vm_map;
 1141         vm_map_entry_t entry;
 1142         vm_object_t backing_object, object;
 1143         boolean_t ignore_entry;
 1144 
 1145         vm_map_lock_read(map);
 1146         for (entry = map->header.next; entry != &map->header;
 1147             entry = entry->next) {
 1148                 /*
 1149                  * Don't dump inaccessible mappings, deal with legacy
 1150                  * coredump mode.
 1151                  *
 1152                  * Note that read-only segments related to the elf binary
 1153                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
 1154                  * need to arbitrarily ignore such segments.
 1155                  */
 1156                 if (elf_legacy_coredump) {
 1157                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
 1158                                 continue;
 1159                 } else {
 1160                         if ((entry->protection & VM_PROT_ALL) == 0)
 1161                                 continue;
 1162                 }
 1163 
 1164                 /*
 1165                  * Dont include memory segment in the coredump if
 1166                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
 1167                  * madvise(2).  Do not dump submaps (i.e. parts of the
 1168                  * kernel map).
 1169                  */
 1170                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
 1171                         continue;
 1172 
 1173                 if ((object = entry->object.vm_object) == NULL)
 1174                         continue;
 1175 
 1176                 /* Ignore memory-mapped devices and such things. */
 1177                 VM_OBJECT_LOCK(object);
 1178                 while ((backing_object = object->backing_object) != NULL) {
 1179                         VM_OBJECT_LOCK(backing_object);
 1180                         VM_OBJECT_UNLOCK(object);
 1181                         object = backing_object;
 1182                 }
 1183                 ignore_entry = object->type != OBJT_DEFAULT &&
 1184                     object->type != OBJT_SWAP && object->type != OBJT_VNODE;
 1185                 VM_OBJECT_UNLOCK(object);
 1186                 if (ignore_entry)
 1187                         continue;
 1188 
 1189                 (*func)(entry, closure);
 1190         }
 1191         vm_map_unlock_read(map);
 1192 }
 1193 
 1194 /*
 1195  * Write the core file header to the file, including padding up to
 1196  * the page boundary.
 1197  */
 1198 static int
 1199 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
 1200         struct thread *td;
 1201         struct vnode *vp;
 1202         struct ucred *cred;
 1203         int numsegs;
 1204         size_t hdrsize;
 1205         void *hdr;
 1206 {
 1207         size_t off;
 1208 
 1209         /* Fill in the header. */
 1210         bzero(hdr, hdrsize);
 1211         off = 0;
 1212         __elfN(puthdr)(td, hdr, &off, numsegs);
 1213 
 1214         /* Write it to the core file. */
 1215         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
 1216             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1217             td)); /* XXXKSE */
 1218 }
 1219 
 1220 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1221 typedef struct prstatus32 elf_prstatus_t;
 1222 typedef struct prpsinfo32 elf_prpsinfo_t;
 1223 typedef struct fpreg32 elf_prfpregset_t;
 1224 typedef struct fpreg32 elf_fpregset_t;
 1225 typedef struct reg32 elf_gregset_t;
 1226 #else
 1227 typedef prstatus_t elf_prstatus_t;
 1228 typedef prpsinfo_t elf_prpsinfo_t;
 1229 typedef prfpregset_t elf_prfpregset_t;
 1230 typedef prfpregset_t elf_fpregset_t;
 1231 typedef gregset_t elf_gregset_t;
 1232 #endif
 1233 
 1234 static void
 1235 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
 1236 {
 1237         struct {
 1238                 elf_prstatus_t status;
 1239                 elf_prfpregset_t fpregset;
 1240                 elf_prpsinfo_t psinfo;
 1241         } *tempdata;
 1242         elf_prstatus_t *status;
 1243         elf_prfpregset_t *fpregset;
 1244         elf_prpsinfo_t *psinfo;
 1245         struct proc *p;
 1246         struct thread *thr;
 1247         size_t ehoff, noteoff, notesz, phoff;
 1248 
 1249         p = td->td_proc;
 1250 
 1251         ehoff = *off;
 1252         *off += sizeof(Elf_Ehdr);
 1253 
 1254         phoff = *off;
 1255         *off += (numsegs + 1) * sizeof(Elf_Phdr);
 1256 
 1257         noteoff = *off;
 1258         /*
 1259          * Don't allocate space for the notes if we're just calculating
 1260          * the size of the header. We also don't collect the data.
 1261          */
 1262         if (dst != NULL) {
 1263                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
 1264                 status = &tempdata->status;
 1265                 fpregset = &tempdata->fpregset;
 1266                 psinfo = &tempdata->psinfo;
 1267         } else {
 1268                 tempdata = NULL;
 1269                 status = NULL;
 1270                 fpregset = NULL;
 1271                 psinfo = NULL;
 1272         }
 1273 
 1274         if (dst != NULL) {
 1275                 psinfo->pr_version = PRPSINFO_VERSION;
 1276                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
 1277                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
 1278                 /*
 1279                  * XXX - We don't fill in the command line arguments properly
 1280                  * yet.
 1281                  */
 1282                 strlcpy(psinfo->pr_psargs, p->p_comm,
 1283                     sizeof(psinfo->pr_psargs));
 1284         }
 1285         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
 1286             sizeof *psinfo);
 1287 
 1288         /*
 1289          * To have the debugger select the right thread (LWP) as the initial
 1290          * thread, we dump the state of the thread passed to us in td first.
 1291          * This is the thread that causes the core dump and thus likely to
 1292          * be the right thread one wants to have selected in the debugger.
 1293          */
 1294         thr = td;
 1295         while (thr != NULL) {
 1296                 if (dst != NULL) {
 1297                         status->pr_version = PRSTATUS_VERSION;
 1298                         status->pr_statussz = sizeof(elf_prstatus_t);
 1299                         status->pr_gregsetsz = sizeof(elf_gregset_t);
 1300                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
 1301                         status->pr_osreldate = osreldate;
 1302                         status->pr_cursig = p->p_sig;
 1303                         status->pr_pid = thr->td_tid;
 1304 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1305                         fill_regs32(thr, &status->pr_reg);
 1306                         fill_fpregs32(thr, fpregset);
 1307 #else
 1308                         fill_regs(thr, &status->pr_reg);
 1309                         fill_fpregs(thr, fpregset);
 1310 #endif
 1311                 }
 1312                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
 1313                     sizeof *status);
 1314                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
 1315                     sizeof *fpregset);
 1316                 /*
 1317                  * Allow for MD specific notes, as well as any MD
 1318                  * specific preparations for writing MI notes.
 1319                  */
 1320                 __elfN(dump_thread)(thr, dst, off);
 1321 
 1322                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
 1323                     TAILQ_NEXT(thr, td_plist);
 1324                 if (thr == td)
 1325                         thr = TAILQ_NEXT(thr, td_plist);
 1326         }
 1327 
 1328         notesz = *off - noteoff;
 1329 
 1330         if (dst != NULL)
 1331                 free(tempdata, M_TEMP);
 1332 
 1333         /* Align up to a page boundary for the program segments. */
 1334         *off = round_page(*off);
 1335 
 1336         if (dst != NULL) {
 1337                 Elf_Ehdr *ehdr;
 1338                 Elf_Phdr *phdr;
 1339                 struct phdr_closure phc;
 1340 
 1341                 /*
 1342                  * Fill in the ELF header.
 1343                  */
 1344                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
 1345                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
 1346                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
 1347                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
 1348                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
 1349                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
 1350                 ehdr->e_ident[EI_DATA] = ELF_DATA;
 1351                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 1352                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
 1353                 ehdr->e_ident[EI_ABIVERSION] = 0;
 1354                 ehdr->e_ident[EI_PAD] = 0;
 1355                 ehdr->e_type = ET_CORE;
 1356 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1357                 ehdr->e_machine = EM_386;
 1358 #else
 1359                 ehdr->e_machine = ELF_ARCH;
 1360 #endif
 1361                 ehdr->e_version = EV_CURRENT;
 1362                 ehdr->e_entry = 0;
 1363                 ehdr->e_phoff = phoff;
 1364                 ehdr->e_flags = 0;
 1365                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
 1366                 ehdr->e_phentsize = sizeof(Elf_Phdr);
 1367                 ehdr->e_phnum = numsegs + 1;
 1368                 ehdr->e_shentsize = sizeof(Elf_Shdr);
 1369                 ehdr->e_shnum = 0;
 1370                 ehdr->e_shstrndx = SHN_UNDEF;
 1371 
 1372                 /*
 1373                  * Fill in the program header entries.
 1374                  */
 1375                 phdr = (Elf_Phdr *)((char *)dst + phoff);
 1376 
 1377                 /* The note segement. */
 1378                 phdr->p_type = PT_NOTE;
 1379                 phdr->p_offset = noteoff;
 1380                 phdr->p_vaddr = 0;
 1381                 phdr->p_paddr = 0;
 1382                 phdr->p_filesz = notesz;
 1383                 phdr->p_memsz = 0;
 1384                 phdr->p_flags = 0;
 1385                 phdr->p_align = 0;
 1386                 phdr++;
 1387 
 1388                 /* All the writable segments from the program. */
 1389                 phc.phdr = phdr;
 1390                 phc.offset = *off;
 1391                 each_writable_segment(td, cb_put_phdr, &phc);
 1392         }
 1393 }
 1394 
 1395 static void
 1396 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
 1397     const void *desc, size_t descsz)
 1398 {
 1399         Elf_Note note;
 1400 
 1401         note.n_namesz = strlen(name) + 1;
 1402         note.n_descsz = descsz;
 1403         note.n_type = type;
 1404         if (dst != NULL)
 1405                 bcopy(&note, (char *)dst + *off, sizeof note);
 1406         *off += sizeof note;
 1407         if (dst != NULL)
 1408                 bcopy(name, (char *)dst + *off, note.n_namesz);
 1409         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 1410         if (dst != NULL)
 1411                 bcopy(desc, (char *)dst + *off, note.n_descsz);
 1412         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 1413 }
 1414 
 1415 /*
 1416  * Try to find the appropriate ABI-note section for checknote,
 1417  * fetch the osreldate for binary from the ELF OSABI-note. Only the
 1418  * first page of the image is searched, the same as for headers.
 1419  */
 1420 static boolean_t
 1421 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
 1422     int32_t *osrel)
 1423 {
 1424         const Elf_Note *note, *note0, *note_end;
 1425         const Elf_Phdr *phdr, *pnote;
 1426         const Elf_Ehdr *hdr;
 1427         const char *note_name;
 1428         int i;
 1429 
 1430         pnote = NULL;
 1431         hdr = (const Elf_Ehdr *)imgp->image_header;
 1432         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 1433 
 1434         for (i = 0; i < hdr->e_phnum; i++) {
 1435                 if (phdr[i].p_type == PT_NOTE) {
 1436                         pnote = &phdr[i];
 1437                         break;
 1438                 }
 1439         }
 1440 
 1441         if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
 1442             pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
 1443                 return (FALSE);
 1444 
 1445         note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
 1446         note_end = (const Elf_Note *)(imgp->image_header +
 1447             pnote->p_offset + pnote->p_filesz);
 1448         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
 1449                 if (!aligned(note, Elf32_Addr))
 1450                         return (FALSE);
 1451                 if (note->n_namesz != checknote->hdr.n_namesz ||
 1452                     note->n_descsz != checknote->hdr.n_descsz ||
 1453                     note->n_type != checknote->hdr.n_type)
 1454                         goto nextnote;
 1455                 note_name = (const char *)(note + 1);
 1456                 if (strncmp(checknote->vendor, note_name,
 1457                     checknote->hdr.n_namesz) != 0)
 1458                         goto nextnote;
 1459 
 1460                 /*
 1461                  * Fetch the osreldate for binary
 1462                  * from the ELF OSABI-note if necessary.
 1463                  */
 1464                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
 1465                     checknote->trans_osrel != NULL)
 1466                         return (checknote->trans_osrel(note, osrel));
 1467                 return (TRUE);
 1468 
 1469 nextnote:
 1470                 note = (const Elf_Note *)((const char *)(note + 1) +
 1471                     roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
 1472                     roundup2(note->n_descsz, sizeof(Elf32_Addr)));
 1473         }
 1474 
 1475         return (FALSE);
 1476 }
 1477 
 1478 /*
 1479  * Tell kern_execve.c about it, with a little help from the linker.
 1480  */
 1481 static struct execsw __elfN(execsw) = {
 1482         __CONCAT(exec_, __elfN(imgact)),
 1483         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
 1484 };
 1485 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));

Cache object: 4e6c91dda301e9b6a96556af0ba358f0


[ 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.