2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_compat.h"
36 #include <sys/param.h>
38 #include <sys/fcntl.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/pioctl.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>
57 #include <sys/syscall.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/vnode.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_param.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
70 #include <machine/elf.h>
71 #include <machine/md_var.h>
73 #define OLD_EI_BRAND 8
75 static int __elfN(check_header)(const Elf_Ehdr *hdr);
76 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
77 const char *interp, int32_t *osrel);
78 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
79 u_long *entry, size_t pagesize);
80 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
81 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
82 vm_prot_t prot, size_t pagesize);
83 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
84 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
86 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
87 static boolean_t __elfN(check_note)(struct image_params *imgp,
88 Elf_Brandnote *checknote, int32_t *osrel);
90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
93 int __elfN(fallback_brand) = -1;
94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
95 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
96 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
98 &__elfN(fallback_brand));
100 static int elf_legacy_coredump = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
102 &elf_legacy_coredump, 0, "");
104 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
106 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
107 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
108 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
110 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
112 Elf_Brandnote __elfN(freebsd_brandnote) = {
113 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
114 .hdr.n_descsz = sizeof(int32_t),
116 .vendor = FREEBSD_ABI_VENDOR,
117 .flags = BN_TRANSLATE_OSREL,
118 .trans_osrel = __elfN(freebsd_trans_osrel)
122 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
126 p = (uintptr_t)(note + 1);
127 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
128 *osrel = *(const int32_t *)(p);
133 static const char GNU_ABI_VENDOR[] = "GNU";
134 static int GNU_KFREEBSD_ABI_DESC = 3;
136 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
137 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
138 .hdr.n_descsz = 16, /* XXX at least 16 */
140 .vendor = GNU_ABI_VENDOR,
141 .flags = BN_TRANSLATE_OSREL,
142 .trans_osrel = kfreebsd_trans_osrel
146 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
148 const Elf32_Word *desc;
151 p = (uintptr_t)(note + 1);
152 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
154 desc = (const Elf32_Word *)p;
155 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
159 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
160 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
162 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
168 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
172 for (i = 0; i < MAX_BRANDS; i++) {
173 if (elf_brand_list[i] == NULL) {
174 elf_brand_list[i] = entry;
178 if (i == MAX_BRANDS) {
179 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
187 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
191 for (i = 0; i < MAX_BRANDS; i++) {
192 if (elf_brand_list[i] == entry) {
193 elf_brand_list[i] = NULL;
203 __elfN(brand_inuse)(Elf_Brandinfo *entry)
208 sx_slock(&allproc_lock);
209 FOREACH_PROC_IN_SYSTEM(p) {
210 if (p->p_sysent == entry->sysvec) {
215 sx_sunlock(&allproc_lock);
220 static Elf_Brandinfo *
221 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
224 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
230 * We support four types of branding -- (1) the ELF EI_OSABI field
231 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
232 * branding w/in the ELF header, (3) path of the `interp_path'
233 * field, and (4) the ".note.ABI-tag" ELF section.
236 /* Look for an ".note.ABI-tag" ELF section */
237 for (i = 0; i < MAX_BRANDS; i++) {
238 bi = elf_brand_list[i];
241 if (hdr->e_machine == bi->machine && (bi->flags &
242 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
243 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
249 /* If the executable has a brand, search for it in the brand list. */
250 for (i = 0; i < MAX_BRANDS; i++) {
251 bi = elf_brand_list[i];
252 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
254 if (hdr->e_machine == bi->machine &&
255 (hdr->e_ident[EI_OSABI] == bi->brand ||
256 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
257 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
261 /* Lacking a known brand, search for a recognized interpreter. */
262 if (interp != NULL) {
263 for (i = 0; i < MAX_BRANDS; i++) {
264 bi = elf_brand_list[i];
265 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
267 if (hdr->e_machine == bi->machine &&
268 strcmp(interp, bi->interp_path) == 0)
273 /* Lacking a recognized interpreter, try the default brand */
274 for (i = 0; i < MAX_BRANDS; i++) {
275 bi = elf_brand_list[i];
276 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
278 if (hdr->e_machine == bi->machine &&
279 __elfN(fallback_brand) == bi->brand)
286 __elfN(check_header)(const Elf_Ehdr *hdr)
292 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
293 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
294 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
295 hdr->e_phentsize != sizeof(Elf_Phdr) ||
296 hdr->e_version != ELF_TARG_VER)
300 * Make sure we have at least one brand for this machine.
303 for (i = 0; i < MAX_BRANDS; i++) {
304 bi = elf_brand_list[i];
305 if (bi != NULL && bi->machine == hdr->e_machine)
315 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
316 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
323 * Create the page if it doesn't exist yet. Ignore errors.
326 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
327 VM_PROT_ALL, VM_PROT_ALL, 0);
331 * Find the page from the underlying object.
334 sf = vm_imgact_map_page(object, offset);
336 return (KERN_FAILURE);
337 off = offset - trunc_page(offset);
338 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
340 vm_imgact_unmap_page(sf);
342 return (KERN_FAILURE);
346 return (KERN_SUCCESS);
350 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
351 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
358 if (start != trunc_page(start)) {
359 rv = __elfN(map_partial)(map, object, offset, start,
360 round_page(start), prot);
363 offset += round_page(start) - start;
364 start = round_page(start);
366 if (end != round_page(end)) {
367 rv = __elfN(map_partial)(map, object, offset +
368 trunc_page(end) - start, trunc_page(end), end, prot);
371 end = trunc_page(end);
374 if (offset & PAGE_MASK) {
376 * The mapping is not page aligned. This means we have
377 * to copy the data. Sigh.
379 rv = vm_map_find(map, NULL, 0, &start, end - start,
380 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
384 return (KERN_SUCCESS);
385 for (; start < end; start += sz) {
386 sf = vm_imgact_map_page(object, offset);
388 return (KERN_FAILURE);
389 off = offset - trunc_page(offset);
391 if (sz > PAGE_SIZE - off)
392 sz = PAGE_SIZE - off;
393 error = copyout((caddr_t)sf_buf_kva(sf) + off,
395 vm_imgact_unmap_page(sf);
397 return (KERN_FAILURE);
403 vm_object_reference(object);
405 rv = vm_map_insert(map, object, offset, start, end,
406 prot, VM_PROT_ALL, cow);
408 if (rv != KERN_SUCCESS)
409 vm_object_deallocate(object);
413 return (KERN_SUCCESS);
418 __elfN(load_section)(struct vmspace *vmspace,
419 vm_object_t object, vm_offset_t offset,
420 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
425 vm_offset_t map_addr;
428 vm_offset_t file_addr;
431 * It's necessary to fail if the filsz + offset taken from the
432 * header is greater than the actual file pager object's size.
433 * If we were to allow this, then the vm_map_find() below would
434 * walk right off the end of the file object and into the ether.
436 * While I'm here, might as well check for something else that
437 * is invalid: filsz cannot be greater than memsz.
439 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
441 uprintf("elf_load_section: truncated ELF file\n");
445 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
446 file_addr = trunc_page_ps(offset, pagesize);
449 * We have two choices. We can either clear the data in the last page
450 * of an oversized mapping, or we can start the anon mapping a page
451 * early and copy the initialized data into that first page. We
452 * choose the second..
455 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
457 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
460 /* cow flags: don't dump readonly sections in core */
461 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
462 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
464 rv = __elfN(map_insert)(&vmspace->vm_map,
466 file_addr, /* file offset */
467 map_addr, /* virtual start */
468 map_addr + map_len,/* virtual end */
471 if (rv != KERN_SUCCESS)
474 /* we can stop now if we've covered it all */
475 if (memsz == filsz) {
482 * We have to get the remaining bit of the file into the first part
483 * of the oversized map segment. This is normally because the .data
484 * segment in the file is extended to provide bss. It's a neat idea
485 * to try and save a page, but it's a pain in the behind to implement.
487 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
488 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
489 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
492 /* This had damn well better be true! */
494 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
495 map_addr + map_len, VM_PROT_ALL, 0);
496 if (rv != KERN_SUCCESS) {
504 sf = vm_imgact_map_page(object, offset + filsz);
508 /* send the page fragment to user space */
509 off = trunc_page_ps(offset + filsz, pagesize) -
510 trunc_page(offset + filsz);
511 error = copyout((caddr_t)sf_buf_kva(sf) + off,
512 (caddr_t)map_addr, copy_len);
513 vm_imgact_unmap_page(sf);
520 * set it to the specified protection.
521 * XXX had better undo the damage from pasting over the cracks here!
523 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
524 round_page(map_addr + map_len), prot, FALSE);
530 * Load the file "file" into memory. It may be either a shared object
533 * The "addr" reference parameter is in/out. On entry, it specifies
534 * the address where a shared object should be loaded. If the file is
535 * an executable, this value is ignored. On exit, "addr" specifies
536 * where the file was actually loaded.
538 * The "entry" reference parameter is out only. On exit, it specifies
539 * the entry point for the loaded file.
542 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
543 u_long *entry, size_t pagesize)
548 struct image_params image_params;
550 const Elf_Ehdr *hdr = NULL;
551 const Elf_Phdr *phdr = NULL;
552 struct nameidata *nd;
553 struct vmspace *vmspace = p->p_vmspace;
555 struct image_params *imgp;
558 u_long base_addr = 0;
559 int vfslocked, error, i, numsegs;
561 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
563 attr = &tempdata->attr;
564 imgp = &tempdata->image_params;
567 * Initialize part of the common data
571 imgp->firstpage = NULL;
572 imgp->image_header = NULL;
574 imgp->execlabel = NULL;
576 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
579 if ((error = namei(nd)) != 0) {
583 vfslocked = NDHASGIANT(nd);
584 NDFREE(nd, NDF_ONLY_PNBUF);
585 imgp->vp = nd->ni_vp;
588 * Check permissions, modes, uid, etc on the file, and "open" it.
590 error = exec_check_permissions(imgp);
594 error = exec_map_first_page(imgp);
599 * Also make certain that the interpreter stays the same, so set
600 * its VV_TEXT flag, too.
602 nd->ni_vp->v_vflag |= VV_TEXT;
604 imgp->object = nd->ni_vp->v_object;
606 hdr = (const Elf_Ehdr *)imgp->image_header;
607 if ((error = __elfN(check_header)(hdr)) != 0)
609 if (hdr->e_type == ET_DYN)
611 else if (hdr->e_type == ET_EXEC)
618 /* Only support headers that fit within first page for now */
619 /* (multiplication of two Elf_Half fields will not overflow) */
620 if ((hdr->e_phoff > PAGE_SIZE) ||
621 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
626 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
627 if (!aligned(phdr, Elf_Addr)) {
632 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
633 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
634 /* Loadable segment */
636 if (phdr[i].p_flags & PF_X)
637 prot |= VM_PROT_EXECUTE;
638 if (phdr[i].p_flags & PF_W)
639 prot |= VM_PROT_WRITE;
640 if (phdr[i].p_flags & PF_R)
641 prot |= VM_PROT_READ;
643 if ((error = __elfN(load_section)(vmspace,
644 imgp->object, phdr[i].p_offset,
645 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
646 phdr[i].p_memsz, phdr[i].p_filesz, prot,
650 * Establish the base address if this is the
654 base_addr = trunc_page(phdr[i].p_vaddr +
660 *entry = (unsigned long)hdr->e_entry + rbase;
664 exec_unmap_first_page(imgp);
669 VFS_UNLOCK_GIANT(vfslocked);
670 free(tempdata, M_TEMP);
676 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
678 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
679 const Elf_Phdr *phdr;
680 Elf_Auxargs *elf_auxargs;
681 struct vmspace *vmspace;
683 u_long text_size = 0, data_size = 0, total_size = 0;
684 u_long text_addr = 0, data_addr = 0;
685 u_long seg_size, seg_addr;
686 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
689 const char *interp = NULL, *newinterp = NULL;
690 Elf_Brandinfo *brand_info;
692 struct sysentvec *sv;
695 * Do we have a valid ELF header ?
697 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
698 * if particular brand doesn't support it.
700 if (__elfN(check_header)(hdr) != 0 ||
701 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
705 * From here on down, we return an errno, not -1, as we've
706 * detected an ELF file.
709 if ((hdr->e_phoff > PAGE_SIZE) ||
710 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
711 /* Only support headers in first page for now */
714 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
715 if (!aligned(phdr, Elf_Addr))
719 for (i = 0; i < hdr->e_phnum; i++) {
720 if (phdr[i].p_type == PT_LOAD) {
722 baddr = phdr[i].p_vaddr;
726 if (phdr[i].p_type == PT_INTERP) {
727 /* Path to interpreter */
728 if (phdr[i].p_filesz > MAXPATHLEN ||
729 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
731 interp = imgp->image_header + phdr[i].p_offset;
736 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
737 if (brand_info == NULL) {
738 uprintf("ELF binary type \"%u\" not known.\n",
739 hdr->e_ident[EI_OSABI]);
742 if (hdr->e_type == ET_DYN) {
743 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
746 * Honour the base load address from the dso if it is
747 * non-zero for some reason.
750 et_dyn_addr = ET_DYN_LOAD_ADDR;
755 sv = brand_info->sysvec;
756 if (interp != NULL && brand_info->interp_newpath != NULL)
757 newinterp = brand_info->interp_newpath;
760 * Avoid a possible deadlock if the current address space is destroyed
761 * and that address space maps the locked vnode. In the common case,
762 * the locked vnode's v_usecount is decremented but remains greater
763 * than zero. Consequently, the vnode lock is not needed by vrele().
764 * However, in cases where the vnode lock is external, such as nullfs,
765 * v_usecount may become zero.
767 VOP_UNLOCK(imgp->vp, 0);
769 error = exec_new_vmspace(imgp, sv);
770 imgp->proc->p_sysent = sv;
772 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
776 vmspace = imgp->proc->p_vmspace;
778 for (i = 0; i < hdr->e_phnum; i++) {
779 switch (phdr[i].p_type) {
780 case PT_LOAD: /* Loadable segment */
781 if (phdr[i].p_memsz == 0)
784 if (phdr[i].p_flags & PF_X)
785 prot |= VM_PROT_EXECUTE;
786 if (phdr[i].p_flags & PF_W)
787 prot |= VM_PROT_WRITE;
788 if (phdr[i].p_flags & PF_R)
789 prot |= VM_PROT_READ;
791 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
793 * Some x86 binaries assume read == executable,
794 * notably the M3 runtime and therefore cvsup
796 if (prot & VM_PROT_READ)
797 prot |= VM_PROT_EXECUTE;
800 if ((error = __elfN(load_section)(vmspace,
801 imgp->object, phdr[i].p_offset,
802 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
803 phdr[i].p_memsz, phdr[i].p_filesz, prot,
804 sv->sv_pagesize)) != 0)
808 * If this segment contains the program headers,
809 * remember their virtual address for the AT_PHDR
810 * aux entry. Static binaries don't usually include
813 if (phdr[i].p_offset == 0 &&
814 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
816 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
819 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
820 seg_size = round_page(phdr[i].p_memsz +
821 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
824 * Is this .text or .data? We can't use
825 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
826 * alpha terribly and possibly does other bad
827 * things so we stick to the old way of figuring
828 * it out: If the segment contains the program
829 * entry point, it's a text segment, otherwise it
832 * Note that obreak() assumes that data_addr +
833 * data_size == end of data load area, and the ELF
834 * file format expects segments to be sorted by
835 * address. If multiple data segments exist, the
836 * last one will be used.
838 if (hdr->e_entry >= phdr[i].p_vaddr &&
839 hdr->e_entry < (phdr[i].p_vaddr +
841 text_size = seg_size;
842 text_addr = seg_addr;
843 entry = (u_long)hdr->e_entry + et_dyn_addr;
845 data_size = seg_size;
846 data_addr = seg_addr;
848 total_size += seg_size;
850 case PT_PHDR: /* Program header table info */
851 proghdr = phdr[i].p_vaddr + et_dyn_addr;
858 if (data_addr == 0 && data_size == 0) {
859 data_addr = text_addr;
860 data_size = text_size;
864 * Check limits. It should be safe to check the
865 * limits after loading the segments since we do
866 * not actually fault in all the segments pages.
868 PROC_LOCK(imgp->proc);
869 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
870 text_size > maxtsiz ||
871 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
872 PROC_UNLOCK(imgp->proc);
876 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
877 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
878 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
879 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
882 * We load the dynamic linker where a userland call
883 * to mmap(0, ...) would put it. The rationale behind this
884 * calculation is that it leaves room for the heap to grow to
885 * its maximum allowed size.
887 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
888 lim_max(imgp->proc, RLIMIT_DATA));
889 PROC_UNLOCK(imgp->proc);
891 imgp->entry_addr = entry;
893 if (interp != NULL) {
894 int have_interp = FALSE;
895 VOP_UNLOCK(imgp->vp, 0);
896 if (brand_info->emul_path != NULL &&
897 brand_info->emul_path[0] != '\0') {
898 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
899 snprintf(path, MAXPATHLEN, "%s%s",
900 brand_info->emul_path, interp);
901 error = __elfN(load_file)(imgp->proc, path, &addr,
902 &imgp->entry_addr, sv->sv_pagesize);
907 if (!have_interp && newinterp != NULL) {
908 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
909 &imgp->entry_addr, sv->sv_pagesize);
914 error = __elfN(load_file)(imgp->proc, interp, &addr,
915 &imgp->entry_addr, sv->sv_pagesize);
917 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
919 uprintf("ELF interpreter %s not found\n", interp);
926 * Construct auxargs table (used by the fixup routine)
928 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
929 elf_auxargs->execfd = -1;
930 elf_auxargs->phdr = proghdr;
931 elf_auxargs->phent = hdr->e_phentsize;
932 elf_auxargs->phnum = hdr->e_phnum;
933 elf_auxargs->pagesz = PAGE_SIZE;
934 elf_auxargs->base = addr;
935 elf_auxargs->flags = 0;
936 elf_auxargs->entry = entry;
938 imgp->auxargs = elf_auxargs;
939 imgp->interpreted = 0;
940 imgp->proc->p_osrel = osrel;
945 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
948 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
950 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
954 base = (Elf_Addr *)*stack_base;
955 pos = base + (imgp->args->argc + imgp->args->envc + 2);
957 if (args->execfd != -1)
958 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
959 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
960 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
961 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
962 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
963 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
964 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
965 AUXARGS_ENTRY(pos, AT_BASE, args->base);
966 if (imgp->execpathp != 0)
967 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
968 AUXARGS_ENTRY(pos, AT_NULL, 0);
970 free(imgp->auxargs, M_TEMP);
971 imgp->auxargs = NULL;
974 suword(base, (long)imgp->args->argc);
975 *stack_base = (register_t *)base;
980 * Code for generating ELF core dumps.
983 typedef void (*segment_callback)(vm_map_entry_t, void *);
985 /* Closure for cb_put_phdr(). */
986 struct phdr_closure {
987 Elf_Phdr *phdr; /* Program header to fill in */
988 Elf_Off offset; /* Offset of segment in core file */
991 /* Closure for cb_size_segment(). */
992 struct sseg_closure {
993 int count; /* Count of writable segments. */
994 size_t size; /* Total size of all writable segments. */
997 static void cb_put_phdr(vm_map_entry_t, void *);
998 static void cb_size_segment(vm_map_entry_t, void *);
999 static void each_writable_segment(struct thread *, segment_callback, void *);
1000 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1001 int, void *, size_t);
1002 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
1003 static void __elfN(putnote)(void *, size_t *, const char *, int,
1004 const void *, size_t);
1007 __elfN(coredump)(td, vp, limit)
1012 struct ucred *cred = td->td_ucred;
1014 struct sseg_closure seginfo;
1018 /* Size the program segments. */
1021 each_writable_segment(td, cb_size_segment, &seginfo);
1024 * Calculate the size of the core file header area by making
1025 * a dry run of generating it. Nothing is written, but the
1026 * size is calculated.
1029 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1031 if (hdrsize + seginfo.size >= limit)
1035 * Allocate memory for building the header, fill it up,
1038 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1042 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
1044 /* Write the contents of all of the writable segments. */
1050 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1052 for (i = 0; i < seginfo.count; i++) {
1053 error = vn_rdwr_inchunks(UIO_WRITE, vp,
1054 (caddr_t)(uintptr_t)php->p_vaddr,
1055 php->p_filesz, offset, UIO_USERSPACE,
1056 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1060 offset += php->p_filesz;
1070 * A callback for each_writable_segment() to write out the segment's
1071 * program header entry.
1074 cb_put_phdr(entry, closure)
1075 vm_map_entry_t entry;
1078 struct phdr_closure *phc = (struct phdr_closure *)closure;
1079 Elf_Phdr *phdr = phc->phdr;
1081 phc->offset = round_page(phc->offset);
1083 phdr->p_type = PT_LOAD;
1084 phdr->p_offset = phc->offset;
1085 phdr->p_vaddr = entry->start;
1087 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1088 phdr->p_align = PAGE_SIZE;
1090 if (entry->protection & VM_PROT_READ)
1091 phdr->p_flags |= PF_R;
1092 if (entry->protection & VM_PROT_WRITE)
1093 phdr->p_flags |= PF_W;
1094 if (entry->protection & VM_PROT_EXECUTE)
1095 phdr->p_flags |= PF_X;
1097 phc->offset += phdr->p_filesz;
1102 * A callback for each_writable_segment() to gather information about
1103 * the number of segments and their total size.
1106 cb_size_segment(entry, closure)
1107 vm_map_entry_t entry;
1110 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1113 ssc->size += entry->end - entry->start;
1117 * For each writable segment in the process's memory map, call the given
1118 * function with a pointer to the map entry and some arbitrary
1119 * caller-supplied data.
1122 each_writable_segment(td, func, closure)
1124 segment_callback func;
1127 struct proc *p = td->td_proc;
1128 vm_map_t map = &p->p_vmspace->vm_map;
1129 vm_map_entry_t entry;
1130 vm_object_t backing_object, object;
1131 boolean_t ignore_entry;
1133 vm_map_lock_read(map);
1134 for (entry = map->header.next; entry != &map->header;
1135 entry = entry->next) {
1137 * Don't dump inaccessible mappings, deal with legacy
1140 * Note that read-only segments related to the elf binary
1141 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1142 * need to arbitrarily ignore such segments.
1144 if (elf_legacy_coredump) {
1145 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1148 if ((entry->protection & VM_PROT_ALL) == 0)
1153 * Dont include memory segment in the coredump if
1154 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1155 * madvise(2). Do not dump submaps (i.e. parts of the
1158 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1161 if ((object = entry->object.vm_object) == NULL)
1164 /* Ignore memory-mapped devices and such things. */
1165 VM_OBJECT_LOCK(object);
1166 while ((backing_object = object->backing_object) != NULL) {
1167 VM_OBJECT_LOCK(backing_object);
1168 VM_OBJECT_UNLOCK(object);
1169 object = backing_object;
1171 ignore_entry = object->type != OBJT_DEFAULT &&
1172 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1173 VM_OBJECT_UNLOCK(object);
1177 (*func)(entry, closure);
1179 vm_map_unlock_read(map);
1183 * Write the core file header to the file, including padding up to
1184 * the page boundary.
1187 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1197 /* Fill in the header. */
1198 bzero(hdr, hdrsize);
1200 __elfN(puthdr)(td, hdr, &off, numsegs);
1202 /* Write it to the core file. */
1203 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1204 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1208 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1209 #include <compat/freebsd32/freebsd32.h>
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;
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;
1225 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1228 elf_prstatus_t status;
1229 elf_prfpregset_t fpregset;
1230 elf_prpsinfo_t psinfo;
1232 elf_prstatus_t *status;
1233 elf_prfpregset_t *fpregset;
1234 elf_prpsinfo_t *psinfo;
1237 size_t ehoff, noteoff, notesz, phoff;
1242 *off += sizeof(Elf_Ehdr);
1245 *off += (numsegs + 1) * sizeof(Elf_Phdr);
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.
1253 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1254 status = &tempdata->status;
1255 fpregset = &tempdata->fpregset;
1256 psinfo = &tempdata->psinfo;
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));
1269 * XXX - We don't fill in the command line arguments properly
1272 strlcpy(psinfo->pr_psargs, p->p_comm,
1273 sizeof(psinfo->pr_psargs));
1275 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
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.
1285 while (thr != 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_FREEBSD32) && __ELF_WORD_SIZE == 32
1295 fill_regs32(thr, &status->pr_reg);
1296 fill_fpregs32(thr, fpregset);
1298 fill_regs(thr, &status->pr_reg);
1299 fill_fpregs(thr, fpregset);
1302 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1304 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1307 * Allow for MD specific notes, as well as any MD
1308 * specific preparations for writing MI notes.
1310 __elfN(dump_thread)(thr, dst, off);
1312 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1313 TAILQ_NEXT(thr, td_plist);
1315 thr = TAILQ_NEXT(thr, td_plist);
1318 notesz = *off - noteoff;
1321 free(tempdata, M_TEMP);
1323 /* Align up to a page boundary for the program segments. */
1324 *off = round_page(*off);
1329 struct phdr_closure phc;
1332 * Fill in the ELF header.
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_FREEBSD32) && __ELF_WORD_SIZE == 32
1347 ehdr->e_machine = ELF_ARCH32;
1349 ehdr->e_machine = ELF_ARCH;
1351 ehdr->e_version = EV_CURRENT;
1353 ehdr->e_phoff = phoff;
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);
1360 ehdr->e_shstrndx = SHN_UNDEF;
1363 * Fill in the program header entries.
1365 phdr = (Elf_Phdr *)((char *)dst + phoff);
1367 /* The note segement. */
1368 phdr->p_type = PT_NOTE;
1369 phdr->p_offset = noteoff;
1372 phdr->p_filesz = notesz;
1378 /* All the writable segments from the program. */
1381 each_writable_segment(td, cb_put_phdr, &phc);
1386 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1387 const void *desc, size_t descsz)
1391 note.n_namesz = strlen(name) + 1;
1392 note.n_descsz = descsz;
1395 bcopy(¬e, (char *)dst + *off, sizeof note);
1396 *off += sizeof note;
1398 bcopy(name, (char *)dst + *off, note.n_namesz);
1399 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1401 bcopy(desc, (char *)dst + *off, note.n_descsz);
1402 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
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.
1411 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1414 const Elf_Note *note, *note0, *note_end;
1415 const Elf_Phdr *phdr, *pnote;
1416 const Elf_Ehdr *hdr;
1417 const char *note_name;
1421 hdr = (const Elf_Ehdr *)imgp->image_header;
1422 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1424 for (i = 0; i < hdr->e_phnum; i++) {
1425 if (phdr[i].p_type == PT_NOTE) {
1431 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1432 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
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))
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)
1445 note_name = (const char *)(note + 1);
1446 if (strncmp(checknote->vendor, note_name,
1447 checknote->hdr.n_namesz) != 0)
1451 * Fetch the osreldate for binary
1452 * from the ELF OSABI-note if necessary.
1454 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1455 checknote->trans_osrel != NULL)
1456 return (checknote->trans_osrel(note, osrel));
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)));
1469 * Tell kern_execve.c about it, with a little help from the linker.
1471 static struct execsw __elfN(execsw) = {
1472 __CONCAT(exec_, __elfN(imgact)),
1473 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1475 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));