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


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

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

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2000 David O'Brien
    3  * Copyright (c) 1995-1996 Søren Schmidt
    4  * Copyright (c) 1996 Peter Wemm
    5  * All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice, this list of conditions and the following disclaimer
   12  *    in this position and unchanged.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 3. The name of the author may not be used to endorse or promote products
   17  *    derived from this software without specific prior written permission
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
   20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
   22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
   23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
   24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
   25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
   26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
   27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
   28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
   29  */
   30 
   31 #include <sys/cdefs.h>
   32 __FBSDID("$FreeBSD: releng/8.0/sys/kern/imgact_elf.c 198284 2009-10-20 13:34:41Z 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                 return (-1);
  185         return (0);
  186 }
  187 
  188 int
  189 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
  190 {
  191         int i;
  192 
  193         for (i = 0; i < MAX_BRANDS; i++) {
  194                 if (elf_brand_list[i] == entry) {
  195                         elf_brand_list[i] = NULL;
  196                         break;
  197                 }
  198         }
  199         if (i == MAX_BRANDS)
  200                 return (-1);
  201         return (0);
  202 }
  203 
  204 int
  205 __elfN(brand_inuse)(Elf_Brandinfo *entry)
  206 {
  207         struct proc *p;
  208         int rval = FALSE;
  209 
  210         sx_slock(&allproc_lock);
  211         FOREACH_PROC_IN_SYSTEM(p) {
  212                 if (p->p_sysent == entry->sysvec) {
  213                         rval = TRUE;
  214                         break;
  215                 }
  216         }
  217         sx_sunlock(&allproc_lock);
  218 
  219         return (rval);
  220 }
  221 
  222 static Elf_Brandinfo *
  223 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
  224     int32_t *osrel)
  225 {
  226         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  227         Elf_Brandinfo *bi;
  228         boolean_t ret;
  229         int i;
  230 
  231         /*
  232          * We support four types of branding -- (1) the ELF EI_OSABI field
  233          * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
  234          * branding w/in the ELF header, (3) path of the `interp_path'
  235          * field, and (4) the ".note.ABI-tag" ELF section.
  236          */
  237 
  238         /* Look for an ".note.ABI-tag" ELF section */
  239         for (i = 0; i < MAX_BRANDS; i++) {
  240                 bi = elf_brand_list[i];
  241                 if (bi == NULL)
  242                         continue;
  243                 if (hdr->e_machine == bi->machine && (bi->flags &
  244                     (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
  245                         ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
  246                         if (ret)
  247                                 return (bi);
  248                 }
  249         }
  250 
  251         /* If the executable has a brand, search for it in the brand list. */
  252         for (i = 0; i < MAX_BRANDS; i++) {
  253                 bi = elf_brand_list[i];
  254                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  255                         continue;
  256                 if (hdr->e_machine == bi->machine &&
  257                     (hdr->e_ident[EI_OSABI] == bi->brand ||
  258                     strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
  259                     bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
  260                         return (bi);
  261         }
  262 
  263         /* Lacking a known brand, search for a recognized interpreter. */
  264         if (interp != NULL) {
  265                 for (i = 0; i < MAX_BRANDS; i++) {
  266                         bi = elf_brand_list[i];
  267                         if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  268                                 continue;
  269                         if (hdr->e_machine == bi->machine &&
  270                             strcmp(interp, bi->interp_path) == 0)
  271                                 return (bi);
  272                 }
  273         }
  274 
  275         /* Lacking a recognized interpreter, try the default brand */
  276         for (i = 0; i < MAX_BRANDS; i++) {
  277                 bi = elf_brand_list[i];
  278                 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
  279                         continue;
  280                 if (hdr->e_machine == bi->machine &&
  281                     __elfN(fallback_brand) == bi->brand)
  282                         return (bi);
  283         }
  284         return (NULL);
  285 }
  286 
  287 static int
  288 __elfN(check_header)(const Elf_Ehdr *hdr)
  289 {
  290         Elf_Brandinfo *bi;
  291         int i;
  292 
  293         if (!IS_ELF(*hdr) ||
  294             hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
  295             hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
  296             hdr->e_ident[EI_VERSION] != EV_CURRENT ||
  297             hdr->e_phentsize != sizeof(Elf_Phdr) ||
  298             hdr->e_version != ELF_TARG_VER)
  299                 return (ENOEXEC);
  300 
  301         /*
  302          * Make sure we have at least one brand for this machine.
  303          */
  304 
  305         for (i = 0; i < MAX_BRANDS; i++) {
  306                 bi = elf_brand_list[i];
  307                 if (bi != NULL && bi->machine == hdr->e_machine)
  308                         break;
  309         }
  310         if (i == MAX_BRANDS)
  311                 return (ENOEXEC);
  312 
  313         return (0);
  314 }
  315 
  316 static int
  317 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  318     vm_offset_t start, vm_offset_t end, vm_prot_t prot)
  319 {
  320         struct sf_buf *sf;
  321         int error;
  322         vm_offset_t off;
  323 
  324         /*
  325          * Create the page if it doesn't exist yet. Ignore errors.
  326          */
  327         vm_map_lock(map);
  328         vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
  329             VM_PROT_ALL, VM_PROT_ALL, 0);
  330         vm_map_unlock(map);
  331 
  332         /*
  333          * Find the page from the underlying object.
  334          */
  335         if (object) {
  336                 sf = vm_imgact_map_page(object, offset);
  337                 if (sf == NULL)
  338                         return (KERN_FAILURE);
  339                 off = offset - trunc_page(offset);
  340                 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
  341                     end - start);
  342                 vm_imgact_unmap_page(sf);
  343                 if (error) {
  344                         return (KERN_FAILURE);
  345                 }
  346         }
  347 
  348         return (KERN_SUCCESS);
  349 }
  350 
  351 static int
  352 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
  353     vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
  354 {
  355         struct sf_buf *sf;
  356         vm_offset_t off;
  357         vm_size_t sz;
  358         int error, rv;
  359 
  360         if (start != trunc_page(start)) {
  361                 rv = __elfN(map_partial)(map, object, offset, start,
  362                     round_page(start), prot);
  363                 if (rv)
  364                         return (rv);
  365                 offset += round_page(start) - start;
  366                 start = round_page(start);
  367         }
  368         if (end != round_page(end)) {
  369                 rv = __elfN(map_partial)(map, object, offset +
  370                     trunc_page(end) - start, trunc_page(end), end, prot);
  371                 if (rv)
  372                         return (rv);
  373                 end = trunc_page(end);
  374         }
  375         if (end > start) {
  376                 if (offset & PAGE_MASK) {
  377                         /*
  378                          * The mapping is not page aligned. This means we have
  379                          * to copy the data. Sigh.
  380                          */
  381                         rv = vm_map_find(map, NULL, 0, &start, end - start,
  382                             FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
  383                         if (rv)
  384                                 return (rv);
  385                         if (object == NULL)
  386                                 return (KERN_SUCCESS);
  387                         for (; start < end; start += sz) {
  388                                 sf = vm_imgact_map_page(object, offset);
  389                                 if (sf == NULL)
  390                                         return (KERN_FAILURE);
  391                                 off = offset - trunc_page(offset);
  392                                 sz = end - start;
  393                                 if (sz > PAGE_SIZE - off)
  394                                         sz = PAGE_SIZE - off;
  395                                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  396                                     (caddr_t)start, sz);
  397                                 vm_imgact_unmap_page(sf);
  398                                 if (error) {
  399                                         return (KERN_FAILURE);
  400                                 }
  401                                 offset += sz;
  402                         }
  403                         rv = KERN_SUCCESS;
  404                 } else {
  405                         vm_object_reference(object);
  406                         vm_map_lock(map);
  407                         rv = vm_map_insert(map, object, offset, start, end,
  408                             prot, VM_PROT_ALL, cow);
  409                         vm_map_unlock(map);
  410                         if (rv != KERN_SUCCESS)
  411                                 vm_object_deallocate(object);
  412                 }
  413                 return (rv);
  414         } else {
  415                 return (KERN_SUCCESS);
  416         }
  417 }
  418 
  419 static int
  420 __elfN(load_section)(struct vmspace *vmspace,
  421         vm_object_t object, vm_offset_t offset,
  422         caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
  423         size_t pagesize)
  424 {
  425         struct sf_buf *sf;
  426         size_t map_len;
  427         vm_offset_t map_addr;
  428         int error, rv, cow;
  429         size_t copy_len;
  430         vm_offset_t file_addr;
  431 
  432         /*
  433          * It's necessary to fail if the filsz + offset taken from the
  434          * header is greater than the actual file pager object's size.
  435          * If we were to allow this, then the vm_map_find() below would
  436          * walk right off the end of the file object and into the ether.
  437          *
  438          * While I'm here, might as well check for something else that
  439          * is invalid: filsz cannot be greater than memsz.
  440          */
  441         if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
  442             filsz > memsz) {
  443                 uprintf("elf_load_section: truncated ELF file\n");
  444                 return (ENOEXEC);
  445         }
  446 
  447         map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
  448         file_addr = trunc_page_ps(offset, pagesize);
  449 
  450         /*
  451          * We have two choices.  We can either clear the data in the last page
  452          * of an oversized mapping, or we can start the anon mapping a page
  453          * early and copy the initialized data into that first page.  We
  454          * choose the second..
  455          */
  456         if (memsz > filsz)
  457                 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
  458         else
  459                 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
  460 
  461         if (map_len != 0) {
  462                 /* cow flags: don't dump readonly sections in core */
  463                 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
  464                     (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
  465 
  466                 rv = __elfN(map_insert)(&vmspace->vm_map,
  467                                       object,
  468                                       file_addr,        /* file offset */
  469                                       map_addr,         /* virtual start */
  470                                       map_addr + map_len,/* virtual end */
  471                                       prot,
  472                                       cow);
  473                 if (rv != KERN_SUCCESS)
  474                         return (EINVAL);
  475 
  476                 /* we can stop now if we've covered it all */
  477                 if (memsz == filsz) {
  478                         return (0);
  479                 }
  480         }
  481 
  482 
  483         /*
  484          * We have to get the remaining bit of the file into the first part
  485          * of the oversized map segment.  This is normally because the .data
  486          * segment in the file is extended to provide bss.  It's a neat idea
  487          * to try and save a page, but it's a pain in the behind to implement.
  488          */
  489         copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
  490         map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
  491         map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
  492             map_addr;
  493 
  494         /* This had damn well better be true! */
  495         if (map_len != 0) {
  496                 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
  497                     map_addr + map_len, VM_PROT_ALL, 0);
  498                 if (rv != KERN_SUCCESS) {
  499                         return (EINVAL);
  500                 }
  501         }
  502 
  503         if (copy_len != 0) {
  504                 vm_offset_t off;
  505 
  506                 sf = vm_imgact_map_page(object, offset + filsz);
  507                 if (sf == NULL)
  508                         return (EIO);
  509 
  510                 /* send the page fragment to user space */
  511                 off = trunc_page_ps(offset + filsz, pagesize) -
  512                     trunc_page(offset + filsz);
  513                 error = copyout((caddr_t)sf_buf_kva(sf) + off,
  514                     (caddr_t)map_addr, copy_len);
  515                 vm_imgact_unmap_page(sf);
  516                 if (error) {
  517                         return (error);
  518                 }
  519         }
  520 
  521         /*
  522          * set it to the specified protection.
  523          * XXX had better undo the damage from pasting over the cracks here!
  524          */
  525         vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
  526             round_page(map_addr + map_len),  prot, FALSE);
  527 
  528         return (0);
  529 }
  530 
  531 /*
  532  * Load the file "file" into memory.  It may be either a shared object
  533  * or an executable.
  534  *
  535  * The "addr" reference parameter is in/out.  On entry, it specifies
  536  * the address where a shared object should be loaded.  If the file is
  537  * an executable, this value is ignored.  On exit, "addr" specifies
  538  * where the file was actually loaded.
  539  *
  540  * The "entry" reference parameter is out only.  On exit, it specifies
  541  * the entry point for the loaded file.
  542  */
  543 static int
  544 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
  545         u_long *entry, size_t pagesize)
  546 {
  547         struct {
  548                 struct nameidata nd;
  549                 struct vattr attr;
  550                 struct image_params image_params;
  551         } *tempdata;
  552         const Elf_Ehdr *hdr = NULL;
  553         const Elf_Phdr *phdr = NULL;
  554         struct nameidata *nd;
  555         struct vmspace *vmspace = p->p_vmspace;
  556         struct vattr *attr;
  557         struct image_params *imgp;
  558         vm_prot_t prot;
  559         u_long rbase;
  560         u_long base_addr = 0;
  561         int vfslocked, error, i, numsegs;
  562 
  563         tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
  564         nd = &tempdata->nd;
  565         attr = &tempdata->attr;
  566         imgp = &tempdata->image_params;
  567 
  568         /*
  569          * Initialize part of the common data
  570          */
  571         imgp->proc = p;
  572         imgp->attr = attr;
  573         imgp->firstpage = NULL;
  574         imgp->image_header = NULL;
  575         imgp->object = NULL;
  576         imgp->execlabel = NULL;
  577 
  578         NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
  579             curthread);
  580         vfslocked = 0;
  581         if ((error = namei(nd)) != 0) {
  582                 nd->ni_vp = NULL;
  583                 goto fail;
  584         }
  585         vfslocked = NDHASGIANT(nd);
  586         NDFREE(nd, NDF_ONLY_PNBUF);
  587         imgp->vp = nd->ni_vp;
  588 
  589         /*
  590          * Check permissions, modes, uid, etc on the file, and "open" it.
  591          */
  592         error = exec_check_permissions(imgp);
  593         if (error)
  594                 goto fail;
  595 
  596         error = exec_map_first_page(imgp);
  597         if (error)
  598                 goto fail;
  599 
  600         /*
  601          * Also make certain that the interpreter stays the same, so set
  602          * its VV_TEXT flag, too.
  603          */
  604         nd->ni_vp->v_vflag |= VV_TEXT;
  605 
  606         imgp->object = nd->ni_vp->v_object;
  607 
  608         hdr = (const Elf_Ehdr *)imgp->image_header;
  609         if ((error = __elfN(check_header)(hdr)) != 0)
  610                 goto fail;
  611         if (hdr->e_type == ET_DYN)
  612                 rbase = *addr;
  613         else if (hdr->e_type == ET_EXEC)
  614                 rbase = 0;
  615         else {
  616                 error = ENOEXEC;
  617                 goto fail;
  618         }
  619 
  620         /* Only support headers that fit within first page for now      */
  621         /*    (multiplication of two Elf_Half fields will not overflow) */
  622         if ((hdr->e_phoff > PAGE_SIZE) ||
  623             (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
  624                 error = ENOEXEC;
  625                 goto fail;
  626         }
  627 
  628         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  629         if (!aligned(phdr, Elf_Addr)) {
  630                 error = ENOEXEC;
  631                 goto fail;
  632         }
  633 
  634         for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
  635                 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
  636                         /* Loadable segment */
  637                         prot = 0;
  638                         if (phdr[i].p_flags & PF_X)
  639                                 prot |= VM_PROT_EXECUTE;
  640                         if (phdr[i].p_flags & PF_W)
  641                                 prot |= VM_PROT_WRITE;
  642                         if (phdr[i].p_flags & PF_R)
  643                                 prot |= VM_PROT_READ;
  644 
  645                         if ((error = __elfN(load_section)(vmspace,
  646                             imgp->object, phdr[i].p_offset,
  647                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
  648                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  649                             pagesize)) != 0)
  650                                 goto fail;
  651                         /*
  652                          * Establish the base address if this is the
  653                          * first segment.
  654                          */
  655                         if (numsegs == 0)
  656                                 base_addr = trunc_page(phdr[i].p_vaddr +
  657                                     rbase);
  658                         numsegs++;
  659                 }
  660         }
  661         *addr = base_addr;
  662         *entry = (unsigned long)hdr->e_entry + rbase;
  663 
  664 fail:
  665         if (imgp->firstpage)
  666                 exec_unmap_first_page(imgp);
  667 
  668         if (nd->ni_vp)
  669                 vput(nd->ni_vp);
  670 
  671         VFS_UNLOCK_GIANT(vfslocked);
  672         free(tempdata, M_TEMP);
  673 
  674         return (error);
  675 }
  676 
  677 static int
  678 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
  679 {
  680         const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
  681         const Elf_Phdr *phdr;
  682         Elf_Auxargs *elf_auxargs;
  683         struct vmspace *vmspace;
  684         vm_prot_t prot;
  685         u_long text_size = 0, data_size = 0, total_size = 0;
  686         u_long text_addr = 0, data_addr = 0;
  687         u_long seg_size, seg_addr;
  688         u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
  689         int32_t osrel = 0;
  690         int error = 0, i, n;
  691         const char *interp = NULL, *newinterp = NULL;
  692         Elf_Brandinfo *brand_info;
  693         char *path;
  694         struct sysentvec *sv;
  695 
  696         /*
  697          * Do we have a valid ELF header ?
  698          *
  699          * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
  700          * if particular brand doesn't support it.
  701          */
  702         if (__elfN(check_header)(hdr) != 0 ||
  703             (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
  704                 return (-1);
  705 
  706         /*
  707          * From here on down, we return an errno, not -1, as we've
  708          * detected an ELF file.
  709          */
  710 
  711         if ((hdr->e_phoff > PAGE_SIZE) ||
  712             (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
  713                 /* Only support headers in first page for now */
  714                 return (ENOEXEC);
  715         }
  716         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
  717         if (!aligned(phdr, Elf_Addr))
  718                 return (ENOEXEC);
  719         n = 0;
  720         baddr = 0;
  721         for (i = 0; i < hdr->e_phnum; i++) {
  722                 if (phdr[i].p_type == PT_LOAD) {
  723                         if (n == 0)
  724                                 baddr = phdr[i].p_vaddr;
  725                         n++;
  726                         continue;
  727                 }
  728                 if (phdr[i].p_type == PT_INTERP) {
  729                         /* Path to interpreter */
  730                         if (phdr[i].p_filesz > MAXPATHLEN ||
  731                             phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
  732                                 return (ENOEXEC);
  733                         interp = imgp->image_header + phdr[i].p_offset;
  734                         continue;
  735                 }
  736         }
  737 
  738         brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
  739         if (brand_info == NULL) {
  740                 uprintf("ELF binary type \"%u\" not known.\n",
  741                     hdr->e_ident[EI_OSABI]);
  742                 return (ENOEXEC);
  743         }
  744         if (hdr->e_type == ET_DYN) {
  745                 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
  746                         return (ENOEXEC);
  747                 /*
  748                  * Honour the base load address from the dso if it is
  749                  * non-zero for some reason.
  750                  */
  751                 if (baddr == 0)
  752                         et_dyn_addr = ET_DYN_LOAD_ADDR;
  753                 else
  754                         et_dyn_addr = 0;
  755         } else
  756                 et_dyn_addr = 0;
  757         sv = brand_info->sysvec;
  758         if (interp != NULL && brand_info->interp_newpath != NULL)
  759                 newinterp = brand_info->interp_newpath;
  760 
  761         /*
  762          * Avoid a possible deadlock if the current address space is destroyed
  763          * and that address space maps the locked vnode.  In the common case,
  764          * the locked vnode's v_usecount is decremented but remains greater
  765          * than zero.  Consequently, the vnode lock is not needed by vrele().
  766          * However, in cases where the vnode lock is external, such as nullfs,
  767          * v_usecount may become zero.
  768          */
  769         VOP_UNLOCK(imgp->vp, 0);
  770 
  771         error = exec_new_vmspace(imgp, sv);
  772         imgp->proc->p_sysent = sv;
  773 
  774         vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  775         if (error)
  776                 return (error);
  777 
  778         vmspace = imgp->proc->p_vmspace;
  779 
  780         for (i = 0; i < hdr->e_phnum; i++) {
  781                 switch (phdr[i].p_type) {
  782                 case PT_LOAD:   /* Loadable segment */
  783                         if (phdr[i].p_memsz == 0)
  784                                 break;
  785                         prot = 0;
  786                         if (phdr[i].p_flags & PF_X)
  787                                 prot |= VM_PROT_EXECUTE;
  788                         if (phdr[i].p_flags & PF_W)
  789                                 prot |= VM_PROT_WRITE;
  790                         if (phdr[i].p_flags & PF_R)
  791                                 prot |= VM_PROT_READ;
  792 
  793 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
  794                         /*
  795                          * Some x86 binaries assume read == executable,
  796                          * notably the M3 runtime and therefore cvsup
  797                          */
  798                         if (prot & VM_PROT_READ)
  799                                 prot |= VM_PROT_EXECUTE;
  800 #endif
  801 
  802                         if ((error = __elfN(load_section)(vmspace,
  803                             imgp->object, phdr[i].p_offset,
  804                             (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
  805                             phdr[i].p_memsz, phdr[i].p_filesz, prot,
  806                             sv->sv_pagesize)) != 0)
  807                                 return (error);
  808 
  809                         /*
  810                          * If this segment contains the program headers,
  811                          * remember their virtual address for the AT_PHDR
  812                          * aux entry. Static binaries don't usually include
  813                          * a PT_PHDR entry.
  814                          */
  815                         if (phdr[i].p_offset == 0 &&
  816                             hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
  817                                 <= phdr[i].p_filesz)
  818                                 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
  819                                     et_dyn_addr;
  820 
  821                         seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
  822                         seg_size = round_page(phdr[i].p_memsz +
  823                             phdr[i].p_vaddr + et_dyn_addr - seg_addr);
  824 
  825                         /*
  826                          * Is this .text or .data?  We can't use
  827                          * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
  828                          * alpha terribly and possibly does other bad
  829                          * things so we stick to the old way of figuring
  830                          * it out:  If the segment contains the program
  831                          * entry point, it's a text segment, otherwise it
  832                          * is a data segment.
  833                          *
  834                          * Note that obreak() assumes that data_addr + 
  835                          * data_size == end of data load area, and the ELF
  836                          * file format expects segments to be sorted by
  837                          * address.  If multiple data segments exist, the
  838                          * last one will be used.
  839                          */
  840                         if (hdr->e_entry >= phdr[i].p_vaddr &&
  841                             hdr->e_entry < (phdr[i].p_vaddr +
  842                             phdr[i].p_memsz)) {
  843                                 text_size = seg_size;
  844                                 text_addr = seg_addr;
  845                                 entry = (u_long)hdr->e_entry + et_dyn_addr;
  846                         } else {
  847                                 data_size = seg_size;
  848                                 data_addr = seg_addr;
  849                         }
  850                         total_size += seg_size;
  851                         break;
  852                 case PT_PHDR:   /* Program header table info */
  853                         proghdr = phdr[i].p_vaddr + et_dyn_addr;
  854                         break;
  855                 default:
  856                         break;
  857                 }
  858         }
  859         
  860         if (data_addr == 0 && data_size == 0) {
  861                 data_addr = text_addr;
  862                 data_size = text_size;
  863         }
  864 
  865         /*
  866          * Check limits.  It should be safe to check the
  867          * limits after loading the segments since we do
  868          * not actually fault in all the segments pages.
  869          */
  870         PROC_LOCK(imgp->proc);
  871         if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
  872             text_size > maxtsiz ||
  873             total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
  874                 PROC_UNLOCK(imgp->proc);
  875                 return (ENOMEM);
  876         }
  877 
  878         vmspace->vm_tsize = text_size >> PAGE_SHIFT;
  879         vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
  880         vmspace->vm_dsize = data_size >> PAGE_SHIFT;
  881         vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
  882 
  883         /*
  884          * We load the dynamic linker where a userland call
  885          * to mmap(0, ...) would put it.  The rationale behind this
  886          * calculation is that it leaves room for the heap to grow to
  887          * its maximum allowed size.
  888          */
  889         addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
  890             lim_max(imgp->proc, RLIMIT_DATA));
  891         PROC_UNLOCK(imgp->proc);
  892 
  893         imgp->entry_addr = entry;
  894 
  895         if (interp != NULL) {
  896                 int have_interp = FALSE;
  897                 VOP_UNLOCK(imgp->vp, 0);
  898                 if (brand_info->emul_path != NULL &&
  899                     brand_info->emul_path[0] != '\0') {
  900                         path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
  901                         snprintf(path, MAXPATHLEN, "%s%s",
  902                             brand_info->emul_path, interp);
  903                         error = __elfN(load_file)(imgp->proc, path, &addr,
  904                             &imgp->entry_addr, sv->sv_pagesize);
  905                         free(path, M_TEMP);
  906                         if (error == 0)
  907                                 have_interp = TRUE;
  908                 }
  909                 if (!have_interp && newinterp != NULL) {
  910                         error = __elfN(load_file)(imgp->proc, newinterp, &addr,
  911                             &imgp->entry_addr, sv->sv_pagesize);
  912                         if (error == 0)
  913                                 have_interp = TRUE;
  914                 }
  915                 if (!have_interp) {
  916                         error = __elfN(load_file)(imgp->proc, interp, &addr,
  917                             &imgp->entry_addr, sv->sv_pagesize);
  918                 }
  919                 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
  920                 if (error != 0) {
  921                         uprintf("ELF interpreter %s not found\n", interp);
  922                         return (error);
  923                 }
  924         } else
  925                 addr = et_dyn_addr;
  926 
  927         /*
  928          * Construct auxargs table (used by the fixup routine)
  929          */
  930         elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
  931         elf_auxargs->execfd = -1;
  932         elf_auxargs->phdr = proghdr;
  933         elf_auxargs->phent = hdr->e_phentsize;
  934         elf_auxargs->phnum = hdr->e_phnum;
  935         elf_auxargs->pagesz = PAGE_SIZE;
  936         elf_auxargs->base = addr;
  937         elf_auxargs->flags = 0;
  938         elf_auxargs->entry = entry;
  939 
  940         imgp->auxargs = elf_auxargs;
  941         imgp->interpreted = 0;
  942         imgp->proc->p_osrel = osrel;
  943 
  944         return (error);
  945 }
  946 
  947 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
  948 
  949 int
  950 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
  951 {
  952         Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
  953         Elf_Addr *base;
  954         Elf_Addr *pos;
  955 
  956         base = (Elf_Addr *)*stack_base;
  957         pos = base + (imgp->args->argc + imgp->args->envc + 2);
  958 
  959         if (args->execfd != -1)
  960                 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
  961         AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
  962         AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
  963         AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
  964         AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
  965         AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
  966         AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
  967         AUXARGS_ENTRY(pos, AT_BASE, args->base);
  968         if (imgp->execpathp != 0)
  969                 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
  970         AUXARGS_ENTRY(pos, AT_NULL, 0);
  971 
  972         free(imgp->auxargs, M_TEMP);
  973         imgp->auxargs = NULL;
  974 
  975         base--;
  976         suword(base, (long)imgp->args->argc);
  977         *stack_base = (register_t *)base;
  978         return (0);
  979 }
  980 
  981 /*
  982  * Code for generating ELF core dumps.
  983  */
  984 
  985 typedef void (*segment_callback)(vm_map_entry_t, void *);
  986 
  987 /* Closure for cb_put_phdr(). */
  988 struct phdr_closure {
  989         Elf_Phdr *phdr;         /* Program header to fill in */
  990         Elf_Off offset;         /* Offset of segment in core file */
  991 };
  992 
  993 /* Closure for cb_size_segment(). */
  994 struct sseg_closure {
  995         int count;              /* Count of writable segments. */
  996         size_t size;            /* Total size of all writable segments. */
  997 };
  998 
  999 static void cb_put_phdr(vm_map_entry_t, void *);
 1000 static void cb_size_segment(vm_map_entry_t, void *);
 1001 static void each_writable_segment(struct thread *, segment_callback, void *);
 1002 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
 1003     int, void *, size_t);
 1004 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
 1005 static void __elfN(putnote)(void *, size_t *, const char *, int,
 1006     const void *, size_t);
 1007 
 1008 int
 1009 __elfN(coredump)(td, vp, limit)
 1010         struct thread *td;
 1011         struct vnode *vp;
 1012         off_t limit;
 1013 {
 1014         struct ucred *cred = td->td_ucred;
 1015         int error = 0;
 1016         struct sseg_closure seginfo;
 1017         void *hdr;
 1018         size_t hdrsize;
 1019 
 1020         /* Size the program segments. */
 1021         seginfo.count = 0;
 1022         seginfo.size = 0;
 1023         each_writable_segment(td, cb_size_segment, &seginfo);
 1024 
 1025         /*
 1026          * Calculate the size of the core file header area by making
 1027          * a dry run of generating it.  Nothing is written, but the
 1028          * size is calculated.
 1029          */
 1030         hdrsize = 0;
 1031         __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
 1032 
 1033         if (hdrsize + seginfo.size >= limit)
 1034                 return (EFAULT);
 1035 
 1036         /*
 1037          * Allocate memory for building the header, fill it up,
 1038          * and write it out.
 1039          */
 1040         hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
 1041         if (hdr == NULL) {
 1042                 return (EINVAL);
 1043         }
 1044         error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
 1045 
 1046         /* Write the contents of all of the writable segments. */
 1047         if (error == 0) {
 1048                 Elf_Phdr *php;
 1049                 off_t offset;
 1050                 int i;
 1051 
 1052                 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
 1053                 offset = hdrsize;
 1054                 for (i = 0; i < seginfo.count; i++) {
 1055                         error = vn_rdwr_inchunks(UIO_WRITE, vp,
 1056                             (caddr_t)(uintptr_t)php->p_vaddr,
 1057                             php->p_filesz, offset, UIO_USERSPACE,
 1058                             IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1059                             curthread);
 1060                         if (error != 0)
 1061                                 break;
 1062                         offset += php->p_filesz;
 1063                         php++;
 1064                 }
 1065         }
 1066         free(hdr, M_TEMP);
 1067 
 1068         return (error);
 1069 }
 1070 
 1071 /*
 1072  * A callback for each_writable_segment() to write out the segment's
 1073  * program header entry.
 1074  */
 1075 static void
 1076 cb_put_phdr(entry, closure)
 1077         vm_map_entry_t entry;
 1078         void *closure;
 1079 {
 1080         struct phdr_closure *phc = (struct phdr_closure *)closure;
 1081         Elf_Phdr *phdr = phc->phdr;
 1082 
 1083         phc->offset = round_page(phc->offset);
 1084 
 1085         phdr->p_type = PT_LOAD;
 1086         phdr->p_offset = phc->offset;
 1087         phdr->p_vaddr = entry->start;
 1088         phdr->p_paddr = 0;
 1089         phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
 1090         phdr->p_align = PAGE_SIZE;
 1091         phdr->p_flags = 0;
 1092         if (entry->protection & VM_PROT_READ)
 1093                 phdr->p_flags |= PF_R;
 1094         if (entry->protection & VM_PROT_WRITE)
 1095                 phdr->p_flags |= PF_W;
 1096         if (entry->protection & VM_PROT_EXECUTE)
 1097                 phdr->p_flags |= PF_X;
 1098 
 1099         phc->offset += phdr->p_filesz;
 1100         phc->phdr++;
 1101 }
 1102 
 1103 /*
 1104  * A callback for each_writable_segment() to gather information about
 1105  * the number of segments and their total size.
 1106  */
 1107 static void
 1108 cb_size_segment(entry, closure)
 1109         vm_map_entry_t entry;
 1110         void *closure;
 1111 {
 1112         struct sseg_closure *ssc = (struct sseg_closure *)closure;
 1113 
 1114         ssc->count++;
 1115         ssc->size += entry->end - entry->start;
 1116 }
 1117 
 1118 /*
 1119  * For each writable segment in the process's memory map, call the given
 1120  * function with a pointer to the map entry and some arbitrary
 1121  * caller-supplied data.
 1122  */
 1123 static void
 1124 each_writable_segment(td, func, closure)
 1125         struct thread *td;
 1126         segment_callback func;
 1127         void *closure;
 1128 {
 1129         struct proc *p = td->td_proc;
 1130         vm_map_t map = &p->p_vmspace->vm_map;
 1131         vm_map_entry_t entry;
 1132         vm_object_t backing_object, object;
 1133         boolean_t ignore_entry;
 1134 
 1135         vm_map_lock_read(map);
 1136         for (entry = map->header.next; entry != &map->header;
 1137             entry = entry->next) {
 1138                 /*
 1139                  * Don't dump inaccessible mappings, deal with legacy
 1140                  * coredump mode.
 1141                  *
 1142                  * Note that read-only segments related to the elf binary
 1143                  * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
 1144                  * need to arbitrarily ignore such segments.
 1145                  */
 1146                 if (elf_legacy_coredump) {
 1147                         if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
 1148                                 continue;
 1149                 } else {
 1150                         if ((entry->protection & VM_PROT_ALL) == 0)
 1151                                 continue;
 1152                 }
 1153 
 1154                 /*
 1155                  * Dont include memory segment in the coredump if
 1156                  * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
 1157                  * madvise(2).  Do not dump submaps (i.e. parts of the
 1158                  * kernel map).
 1159                  */
 1160                 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
 1161                         continue;
 1162 
 1163                 if ((object = entry->object.vm_object) == NULL)
 1164                         continue;
 1165 
 1166                 /* Ignore memory-mapped devices and such things. */
 1167                 VM_OBJECT_LOCK(object);
 1168                 while ((backing_object = object->backing_object) != NULL) {
 1169                         VM_OBJECT_LOCK(backing_object);
 1170                         VM_OBJECT_UNLOCK(object);
 1171                         object = backing_object;
 1172                 }
 1173                 ignore_entry = object->type != OBJT_DEFAULT &&
 1174                     object->type != OBJT_SWAP && object->type != OBJT_VNODE;
 1175                 VM_OBJECT_UNLOCK(object);
 1176                 if (ignore_entry)
 1177                         continue;
 1178 
 1179                 (*func)(entry, closure);
 1180         }
 1181         vm_map_unlock_read(map);
 1182 }
 1183 
 1184 /*
 1185  * Write the core file header to the file, including padding up to
 1186  * the page boundary.
 1187  */
 1188 static int
 1189 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
 1190         struct thread *td;
 1191         struct vnode *vp;
 1192         struct ucred *cred;
 1193         int numsegs;
 1194         size_t hdrsize;
 1195         void *hdr;
 1196 {
 1197         size_t off;
 1198 
 1199         /* Fill in the header. */
 1200         bzero(hdr, hdrsize);
 1201         off = 0;
 1202         __elfN(puthdr)(td, hdr, &off, numsegs);
 1203 
 1204         /* Write it to the core file. */
 1205         return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
 1206             UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
 1207             td));
 1208 }
 1209 
 1210 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1211 typedef struct prstatus32 elf_prstatus_t;
 1212 typedef struct prpsinfo32 elf_prpsinfo_t;
 1213 typedef struct fpreg32 elf_prfpregset_t;
 1214 typedef struct fpreg32 elf_fpregset_t;
 1215 typedef struct reg32 elf_gregset_t;
 1216 #else
 1217 typedef prstatus_t elf_prstatus_t;
 1218 typedef prpsinfo_t elf_prpsinfo_t;
 1219 typedef prfpregset_t elf_prfpregset_t;
 1220 typedef prfpregset_t elf_fpregset_t;
 1221 typedef gregset_t elf_gregset_t;
 1222 #endif
 1223 
 1224 static void
 1225 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
 1226 {
 1227         struct {
 1228                 elf_prstatus_t status;
 1229                 elf_prfpregset_t fpregset;
 1230                 elf_prpsinfo_t psinfo;
 1231         } *tempdata;
 1232         elf_prstatus_t *status;
 1233         elf_prfpregset_t *fpregset;
 1234         elf_prpsinfo_t *psinfo;
 1235         struct proc *p;
 1236         struct thread *thr;
 1237         size_t ehoff, noteoff, notesz, phoff;
 1238 
 1239         p = td->td_proc;
 1240 
 1241         ehoff = *off;
 1242         *off += sizeof(Elf_Ehdr);
 1243 
 1244         phoff = *off;
 1245         *off += (numsegs + 1) * sizeof(Elf_Phdr);
 1246 
 1247         noteoff = *off;
 1248         /*
 1249          * Don't allocate space for the notes if we're just calculating
 1250          * the size of the header. We also don't collect the data.
 1251          */
 1252         if (dst != NULL) {
 1253                 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
 1254                 status = &tempdata->status;
 1255                 fpregset = &tempdata->fpregset;
 1256                 psinfo = &tempdata->psinfo;
 1257         } else {
 1258                 tempdata = NULL;
 1259                 status = NULL;
 1260                 fpregset = NULL;
 1261                 psinfo = NULL;
 1262         }
 1263 
 1264         if (dst != NULL) {
 1265                 psinfo->pr_version = PRPSINFO_VERSION;
 1266                 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
 1267                 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
 1268                 /*
 1269                  * XXX - We don't fill in the command line arguments properly
 1270                  * yet.
 1271                  */
 1272                 strlcpy(psinfo->pr_psargs, p->p_comm,
 1273                     sizeof(psinfo->pr_psargs));
 1274         }
 1275         __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
 1276             sizeof *psinfo);
 1277 
 1278         /*
 1279          * To have the debugger select the right thread (LWP) as the initial
 1280          * thread, we dump the state of the thread passed to us in td first.
 1281          * This is the thread that causes the core dump and thus likely to
 1282          * be the right thread one wants to have selected in the debugger.
 1283          */
 1284         thr = td;
 1285         while (thr != NULL) {
 1286                 if (dst != NULL) {
 1287                         status->pr_version = PRSTATUS_VERSION;
 1288                         status->pr_statussz = sizeof(elf_prstatus_t);
 1289                         status->pr_gregsetsz = sizeof(elf_gregset_t);
 1290                         status->pr_fpregsetsz = sizeof(elf_fpregset_t);
 1291                         status->pr_osreldate = osreldate;
 1292                         status->pr_cursig = p->p_sig;
 1293                         status->pr_pid = thr->td_tid;
 1294 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1295                         fill_regs32(thr, &status->pr_reg);
 1296                         fill_fpregs32(thr, fpregset);
 1297 #else
 1298                         fill_regs(thr, &status->pr_reg);
 1299                         fill_fpregs(thr, fpregset);
 1300 #endif
 1301                 }
 1302                 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
 1303                     sizeof *status);
 1304                 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
 1305                     sizeof *fpregset);
 1306                 /*
 1307                  * Allow for MD specific notes, as well as any MD
 1308                  * specific preparations for writing MI notes.
 1309                  */
 1310                 __elfN(dump_thread)(thr, dst, off);
 1311 
 1312                 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
 1313                     TAILQ_NEXT(thr, td_plist);
 1314                 if (thr == td)
 1315                         thr = TAILQ_NEXT(thr, td_plist);
 1316         }
 1317 
 1318         notesz = *off - noteoff;
 1319 
 1320         if (dst != NULL)
 1321                 free(tempdata, M_TEMP);
 1322 
 1323         /* Align up to a page boundary for the program segments. */
 1324         *off = round_page(*off);
 1325 
 1326         if (dst != NULL) {
 1327                 Elf_Ehdr *ehdr;
 1328                 Elf_Phdr *phdr;
 1329                 struct phdr_closure phc;
 1330 
 1331                 /*
 1332                  * Fill in the ELF header.
 1333                  */
 1334                 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
 1335                 ehdr->e_ident[EI_MAG0] = ELFMAG0;
 1336                 ehdr->e_ident[EI_MAG1] = ELFMAG1;
 1337                 ehdr->e_ident[EI_MAG2] = ELFMAG2;
 1338                 ehdr->e_ident[EI_MAG3] = ELFMAG3;
 1339                 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
 1340                 ehdr->e_ident[EI_DATA] = ELF_DATA;
 1341                 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 1342                 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
 1343                 ehdr->e_ident[EI_ABIVERSION] = 0;
 1344                 ehdr->e_ident[EI_PAD] = 0;
 1345                 ehdr->e_type = ET_CORE;
 1346 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
 1347                 ehdr->e_machine = EM_386;
 1348 #else
 1349                 ehdr->e_machine = ELF_ARCH;
 1350 #endif
 1351                 ehdr->e_version = EV_CURRENT;
 1352                 ehdr->e_entry = 0;
 1353                 ehdr->e_phoff = phoff;
 1354                 ehdr->e_flags = 0;
 1355                 ehdr->e_ehsize = sizeof(Elf_Ehdr);
 1356                 ehdr->e_phentsize = sizeof(Elf_Phdr);
 1357                 ehdr->e_phnum = numsegs + 1;
 1358                 ehdr->e_shentsize = sizeof(Elf_Shdr);
 1359                 ehdr->e_shnum = 0;
 1360                 ehdr->e_shstrndx = SHN_UNDEF;
 1361 
 1362                 /*
 1363                  * Fill in the program header entries.
 1364                  */
 1365                 phdr = (Elf_Phdr *)((char *)dst + phoff);
 1366 
 1367                 /* The note segement. */
 1368                 phdr->p_type = PT_NOTE;
 1369                 phdr->p_offset = noteoff;
 1370                 phdr->p_vaddr = 0;
 1371                 phdr->p_paddr = 0;
 1372                 phdr->p_filesz = notesz;
 1373                 phdr->p_memsz = 0;
 1374                 phdr->p_flags = 0;
 1375                 phdr->p_align = 0;
 1376                 phdr++;
 1377 
 1378                 /* All the writable segments from the program. */
 1379                 phc.phdr = phdr;
 1380                 phc.offset = *off;
 1381                 each_writable_segment(td, cb_put_phdr, &phc);
 1382         }
 1383 }
 1384 
 1385 static void
 1386 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
 1387     const void *desc, size_t descsz)
 1388 {
 1389         Elf_Note note;
 1390 
 1391         note.n_namesz = strlen(name) + 1;
 1392         note.n_descsz = descsz;
 1393         note.n_type = type;
 1394         if (dst != NULL)
 1395                 bcopy(&note, (char *)dst + *off, sizeof note);
 1396         *off += sizeof note;
 1397         if (dst != NULL)
 1398                 bcopy(name, (char *)dst + *off, note.n_namesz);
 1399         *off += roundup2(note.n_namesz, sizeof(Elf_Size));
 1400         if (dst != NULL)
 1401                 bcopy(desc, (char *)dst + *off, note.n_descsz);
 1402         *off += roundup2(note.n_descsz, sizeof(Elf_Size));
 1403 }
 1404 
 1405 /*
 1406  * Try to find the appropriate ABI-note section for checknote,
 1407  * fetch the osreldate for binary from the ELF OSABI-note. Only the
 1408  * first page of the image is searched, the same as for headers.
 1409  */
 1410 static boolean_t
 1411 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
 1412     int32_t *osrel)
 1413 {
 1414         const Elf_Note *note, *note0, *note_end;
 1415         const Elf_Phdr *phdr, *pnote;
 1416         const Elf_Ehdr *hdr;
 1417         const char *note_name;
 1418         int i;
 1419 
 1420         pnote = NULL;
 1421         hdr = (const Elf_Ehdr *)imgp->image_header;
 1422         phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
 1423 
 1424         for (i = 0; i < hdr->e_phnum; i++) {
 1425                 if (phdr[i].p_type == PT_NOTE) {
 1426                         pnote = &phdr[i];
 1427                         break;
 1428                 }
 1429         }
 1430 
 1431         if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
 1432             pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
 1433                 return (FALSE);
 1434 
 1435         note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
 1436         note_end = (const Elf_Note *)(imgp->image_header +
 1437             pnote->p_offset + pnote->p_filesz);
 1438         for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
 1439                 if (!aligned(note, Elf32_Addr))
 1440                         return (FALSE);
 1441                 if (note->n_namesz != checknote->hdr.n_namesz ||
 1442                     note->n_descsz != checknote->hdr.n_descsz ||
 1443                     note->n_type != checknote->hdr.n_type)
 1444                         goto nextnote;
 1445                 note_name = (const char *)(note + 1);
 1446                 if (strncmp(checknote->vendor, note_name,
 1447                     checknote->hdr.n_namesz) != 0)
 1448                         goto nextnote;
 1449 
 1450                 /*
 1451                  * Fetch the osreldate for binary
 1452                  * from the ELF OSABI-note if necessary.
 1453                  */
 1454                 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
 1455                     checknote->trans_osrel != NULL)
 1456                         return (checknote->trans_osrel(note, osrel));
 1457                 return (TRUE);
 1458 
 1459 nextnote:
 1460                 note = (const Elf_Note *)((const char *)(note + 1) +
 1461                     roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
 1462                     roundup2(note->n_descsz, sizeof(Elf32_Addr)));
 1463         }
 1464 
 1465         return (FALSE);
 1466 }
 1467 
 1468 /*
 1469  * Tell kern_execve.c about it, with a little help from the linker.
 1470  */
 1471 static struct execsw __elfN(execsw) = {
 1472         __CONCAT(exec_, __elfN(imgact)),
 1473         __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
 1474 };
 1475 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));

Cache object: 8884c6abbd93c994cac2859182ccf565


[ 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.