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, int interp_name_len, 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);
89 static vm_prot_t __elfN(trans_prot)(Elf_Word);
90 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
92 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
95 int __elfN(fallback_brand) = -1;
96 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
97 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
98 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
99 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
100 &__elfN(fallback_brand));
102 static int elf_legacy_coredump = 0;
103 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
104 &elf_legacy_coredump, 0, "");
106 #if __ELF_WORD_SIZE == 32
107 #if defined(__amd64__) || defined(__ia64__)
108 int i386_read_exec = 0;
109 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
110 "enable execution from readable segments");
114 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
116 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
117 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
118 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
120 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
122 Elf_Brandnote __elfN(freebsd_brandnote) = {
123 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
124 .hdr.n_descsz = sizeof(int32_t),
126 .vendor = FREEBSD_ABI_VENDOR,
127 .flags = BN_TRANSLATE_OSREL,
128 .trans_osrel = __elfN(freebsd_trans_osrel)
132 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
136 p = (uintptr_t)(note + 1);
137 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
138 *osrel = *(const int32_t *)(p);
143 static const char GNU_ABI_VENDOR[] = "GNU";
144 static int GNU_KFREEBSD_ABI_DESC = 3;
146 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
147 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
148 .hdr.n_descsz = 16, /* XXX at least 16 */
150 .vendor = GNU_ABI_VENDOR,
151 .flags = BN_TRANSLATE_OSREL,
152 .trans_osrel = kfreebsd_trans_osrel
156 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
158 const Elf32_Word *desc;
161 p = (uintptr_t)(note + 1);
162 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
164 desc = (const Elf32_Word *)p;
165 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
169 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
170 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
172 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
178 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
182 for (i = 0; i < MAX_BRANDS; i++) {
183 if (elf_brand_list[i] == NULL) {
184 elf_brand_list[i] = entry;
188 if (i == MAX_BRANDS) {
189 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
197 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
201 for (i = 0; i < MAX_BRANDS; i++) {
202 if (elf_brand_list[i] == entry) {
203 elf_brand_list[i] = NULL;
213 __elfN(brand_inuse)(Elf_Brandinfo *entry)
218 sx_slock(&allproc_lock);
219 FOREACH_PROC_IN_SYSTEM(p) {
220 if (p->p_sysent == entry->sysvec) {
225 sx_sunlock(&allproc_lock);
230 static Elf_Brandinfo *
231 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
232 int interp_name_len, int32_t *osrel)
234 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
240 * We support four types of branding -- (1) the ELF EI_OSABI field
241 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
242 * branding w/in the ELF header, (3) path of the `interp_path'
243 * field, and (4) the ".note.ABI-tag" ELF section.
246 /* Look for an ".note.ABI-tag" ELF section */
247 for (i = 0; i < MAX_BRANDS; i++) {
248 bi = elf_brand_list[i];
251 if (hdr->e_machine == bi->machine && (bi->flags &
252 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
253 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
259 /* If the executable has a brand, search for it in the brand list. */
260 for (i = 0; i < MAX_BRANDS; i++) {
261 bi = elf_brand_list[i];
262 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
264 if (hdr->e_machine == bi->machine &&
265 (hdr->e_ident[EI_OSABI] == bi->brand ||
266 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
267 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
271 /* Lacking a known brand, search for a recognized interpreter. */
272 if (interp != NULL) {
273 for (i = 0; i < MAX_BRANDS; i++) {
274 bi = elf_brand_list[i];
275 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
277 if (hdr->e_machine == bi->machine &&
278 /* ELF image p_filesz includes terminating zero */
279 strlen(bi->interp_path) + 1 == interp_name_len &&
280 strncmp(interp, bi->interp_path, interp_name_len)
286 /* Lacking a recognized interpreter, try the default brand */
287 for (i = 0; i < MAX_BRANDS; i++) {
288 bi = elf_brand_list[i];
289 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
291 if (hdr->e_machine == bi->machine &&
292 __elfN(fallback_brand) == bi->brand)
299 __elfN(check_header)(const Elf_Ehdr *hdr)
305 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
306 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
307 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
308 hdr->e_phentsize != sizeof(Elf_Phdr) ||
309 hdr->e_version != ELF_TARG_VER)
313 * Make sure we have at least one brand for this machine.
316 for (i = 0; i < MAX_BRANDS; i++) {
317 bi = elf_brand_list[i];
318 if (bi != NULL && bi->machine == hdr->e_machine)
328 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
329 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
336 * Create the page if it doesn't exist yet. Ignore errors.
339 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
340 VM_PROT_ALL, VM_PROT_ALL, 0);
344 * Find the page from the underlying object.
347 sf = vm_imgact_map_page(object, offset);
349 return (KERN_FAILURE);
350 off = offset - trunc_page(offset);
351 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
353 vm_imgact_unmap_page(sf);
355 return (KERN_FAILURE);
359 return (KERN_SUCCESS);
363 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
364 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
371 if (start != trunc_page(start)) {
372 rv = __elfN(map_partial)(map, object, offset, start,
373 round_page(start), prot);
376 offset += round_page(start) - start;
377 start = round_page(start);
379 if (end != round_page(end)) {
380 rv = __elfN(map_partial)(map, object, offset +
381 trunc_page(end) - start, trunc_page(end), end, prot);
384 end = trunc_page(end);
387 if (offset & PAGE_MASK) {
389 * The mapping is not page aligned. This means we have
390 * to copy the data. Sigh.
392 rv = vm_map_find(map, NULL, 0, &start, end - start,
393 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
397 return (KERN_SUCCESS);
398 for (; start < end; start += sz) {
399 sf = vm_imgact_map_page(object, offset);
401 return (KERN_FAILURE);
402 off = offset - trunc_page(offset);
404 if (sz > PAGE_SIZE - off)
405 sz = PAGE_SIZE - off;
406 error = copyout((caddr_t)sf_buf_kva(sf) + off,
408 vm_imgact_unmap_page(sf);
410 return (KERN_FAILURE);
416 vm_object_reference(object);
418 rv = vm_map_insert(map, object, offset, start, end,
419 prot, VM_PROT_ALL, cow);
421 if (rv != KERN_SUCCESS)
422 vm_object_deallocate(object);
426 return (KERN_SUCCESS);
431 __elfN(load_section)(struct vmspace *vmspace,
432 vm_object_t object, vm_offset_t offset,
433 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
438 vm_offset_t map_addr;
441 vm_offset_t file_addr;
444 * It's necessary to fail if the filsz + offset taken from the
445 * header is greater than the actual file pager object's size.
446 * If we were to allow this, then the vm_map_find() below would
447 * walk right off the end of the file object and into the ether.
449 * While I'm here, might as well check for something else that
450 * is invalid: filsz cannot be greater than memsz.
452 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
454 uprintf("elf_load_section: truncated ELF file\n");
458 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
459 file_addr = trunc_page_ps(offset, pagesize);
462 * We have two choices. We can either clear the data in the last page
463 * of an oversized mapping, or we can start the anon mapping a page
464 * early and copy the initialized data into that first page. We
465 * choose the second..
468 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
470 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
473 /* cow flags: don't dump readonly sections in core */
474 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
475 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
477 rv = __elfN(map_insert)(&vmspace->vm_map,
479 file_addr, /* file offset */
480 map_addr, /* virtual start */
481 map_addr + map_len,/* virtual end */
484 if (rv != KERN_SUCCESS)
487 /* we can stop now if we've covered it all */
488 if (memsz == filsz) {
495 * We have to get the remaining bit of the file into the first part
496 * of the oversized map segment. This is normally because the .data
497 * segment in the file is extended to provide bss. It's a neat idea
498 * to try and save a page, but it's a pain in the behind to implement.
500 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
501 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
502 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
505 /* This had damn well better be true! */
507 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
508 map_addr + map_len, VM_PROT_ALL, 0);
509 if (rv != KERN_SUCCESS) {
517 sf = vm_imgact_map_page(object, offset + filsz);
521 /* send the page fragment to user space */
522 off = trunc_page_ps(offset + filsz, pagesize) -
523 trunc_page(offset + filsz);
524 error = copyout((caddr_t)sf_buf_kva(sf) + off,
525 (caddr_t)map_addr, copy_len);
526 vm_imgact_unmap_page(sf);
533 * set it to the specified protection.
534 * XXX had better undo the damage from pasting over the cracks here!
536 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
537 round_page(map_addr + map_len), prot, FALSE);
543 * Load the file "file" into memory. It may be either a shared object
546 * The "addr" reference parameter is in/out. On entry, it specifies
547 * the address where a shared object should be loaded. If the file is
548 * an executable, this value is ignored. On exit, "addr" specifies
549 * where the file was actually loaded.
551 * The "entry" reference parameter is out only. On exit, it specifies
552 * the entry point for the loaded file.
555 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
556 u_long *entry, size_t pagesize)
561 struct image_params image_params;
563 const Elf_Ehdr *hdr = NULL;
564 const Elf_Phdr *phdr = NULL;
565 struct nameidata *nd;
566 struct vmspace *vmspace = p->p_vmspace;
568 struct image_params *imgp;
571 u_long base_addr = 0;
572 int vfslocked, error, i, numsegs;
574 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
576 attr = &tempdata->attr;
577 imgp = &tempdata->image_params;
580 * Initialize part of the common data
584 imgp->firstpage = NULL;
585 imgp->image_header = NULL;
587 imgp->execlabel = NULL;
589 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
592 if ((error = namei(nd)) != 0) {
596 vfslocked = NDHASGIANT(nd);
597 NDFREE(nd, NDF_ONLY_PNBUF);
598 imgp->vp = nd->ni_vp;
601 * Check permissions, modes, uid, etc on the file, and "open" it.
603 error = exec_check_permissions(imgp);
607 error = exec_map_first_page(imgp);
612 * Also make certain that the interpreter stays the same, so set
613 * its VV_TEXT flag, too.
615 nd->ni_vp->v_vflag |= VV_TEXT;
617 imgp->object = nd->ni_vp->v_object;
619 hdr = (const Elf_Ehdr *)imgp->image_header;
620 if ((error = __elfN(check_header)(hdr)) != 0)
622 if (hdr->e_type == ET_DYN)
624 else if (hdr->e_type == ET_EXEC)
631 /* Only support headers that fit within first page for now */
632 if ((hdr->e_phoff > PAGE_SIZE) ||
633 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
638 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
639 if (!aligned(phdr, Elf_Addr)) {
644 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
645 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
646 /* Loadable segment */
647 prot = __elfN(trans_prot)(phdr[i].p_flags);
648 if ((error = __elfN(load_section)(vmspace,
649 imgp->object, phdr[i].p_offset,
650 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
651 phdr[i].p_memsz, phdr[i].p_filesz, prot,
655 * Establish the base address if this is the
659 base_addr = trunc_page(phdr[i].p_vaddr +
665 *entry = (unsigned long)hdr->e_entry + rbase;
669 exec_unmap_first_page(imgp);
674 VFS_UNLOCK_GIANT(vfslocked);
675 free(tempdata, M_TEMP);
681 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
683 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
684 const Elf_Phdr *phdr;
685 Elf_Auxargs *elf_auxargs;
686 struct vmspace *vmspace;
688 u_long text_size = 0, data_size = 0, total_size = 0;
689 u_long text_addr = 0, data_addr = 0;
690 u_long seg_size, seg_addr;
691 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
693 int error = 0, i, n, interp_name_len = 0;
694 const char *interp = NULL, *newinterp = NULL;
695 Elf_Brandinfo *brand_info;
697 struct sysentvec *sv;
700 * Do we have a valid ELF header ?
702 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
703 * if particular brand doesn't support it.
705 if (__elfN(check_header)(hdr) != 0 ||
706 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
710 * From here on down, we return an errno, not -1, as we've
711 * detected an ELF file.
714 if ((hdr->e_phoff > PAGE_SIZE) ||
715 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
716 /* Only support headers in first page for now */
719 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
720 if (!aligned(phdr, Elf_Addr))
724 for (i = 0; i < hdr->e_phnum; i++) {
725 if (phdr[i].p_type == PT_LOAD) {
727 baddr = phdr[i].p_vaddr;
731 if (phdr[i].p_type == PT_INTERP) {
732 /* Path to interpreter */
733 if (phdr[i].p_filesz > MAXPATHLEN ||
734 phdr[i].p_offset > PAGE_SIZE ||
735 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
737 interp = imgp->image_header + phdr[i].p_offset;
738 interp_name_len = phdr[i].p_filesz;
743 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
745 if (brand_info == NULL) {
746 uprintf("ELF binary type \"%u\" not known.\n",
747 hdr->e_ident[EI_OSABI]);
750 if (hdr->e_type == ET_DYN) {
751 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
754 * Honour the base load address from the dso if it is
755 * non-zero for some reason.
758 et_dyn_addr = ET_DYN_LOAD_ADDR;
763 sv = brand_info->sysvec;
764 if (interp != NULL && brand_info->interp_newpath != NULL)
765 newinterp = brand_info->interp_newpath;
768 * Avoid a possible deadlock if the current address space is destroyed
769 * and that address space maps the locked vnode. In the common case,
770 * the locked vnode's v_usecount is decremented but remains greater
771 * than zero. Consequently, the vnode lock is not needed by vrele().
772 * However, in cases where the vnode lock is external, such as nullfs,
773 * v_usecount may become zero.
775 VOP_UNLOCK(imgp->vp, 0);
777 error = exec_new_vmspace(imgp, sv);
778 imgp->proc->p_sysent = sv;
780 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
784 vmspace = imgp->proc->p_vmspace;
786 for (i = 0; i < hdr->e_phnum; i++) {
787 switch (phdr[i].p_type) {
788 case PT_LOAD: /* Loadable segment */
789 if (phdr[i].p_memsz == 0)
791 prot = __elfN(trans_prot)(phdr[i].p_flags);
793 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
795 * Some x86 binaries assume read == executable,
796 * notably the M3 runtime and therefore cvsup
798 if (prot & VM_PROT_READ)
799 prot |= VM_PROT_EXECUTE;
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)
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
815 if (phdr[i].p_offset == 0 &&
816 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
818 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
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);
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
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.
840 if (hdr->e_entry >= phdr[i].p_vaddr &&
841 hdr->e_entry < (phdr[i].p_vaddr +
843 text_size = seg_size;
844 text_addr = seg_addr;
845 entry = (u_long)hdr->e_entry + et_dyn_addr;
847 data_size = seg_size;
848 data_addr = seg_addr;
850 total_size += seg_size;
852 case PT_PHDR: /* Program header table info */
853 proghdr = phdr[i].p_vaddr + et_dyn_addr;
860 if (data_addr == 0 && data_size == 0) {
861 data_addr = text_addr;
862 data_size = text_size;
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.
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);
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;
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.
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);
893 imgp->entry_addr = entry;
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);
909 if (!have_interp && newinterp != NULL) {
910 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
911 &imgp->entry_addr, sv->sv_pagesize);
916 error = __elfN(load_file)(imgp->proc, interp, &addr,
917 &imgp->entry_addr, sv->sv_pagesize);
919 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
921 uprintf("ELF interpreter %s not found\n", interp);
928 * Construct auxargs table (used by the fixup routine)
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;
940 imgp->auxargs = elf_auxargs;
941 imgp->interpreted = 0;
942 imgp->proc->p_osrel = osrel;
947 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
950 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
952 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
956 base = (Elf_Addr *)*stack_base;
957 pos = base + (imgp->args->argc + imgp->args->envc + 2);
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);
972 free(imgp->auxargs, M_TEMP);
973 imgp->auxargs = NULL;
976 suword(base, (long)imgp->args->argc);
977 *stack_base = (register_t *)base;
982 * Code for generating ELF core dumps.
985 typedef void (*segment_callback)(vm_map_entry_t, void *);
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 */
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. */
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);
1009 __elfN(coredump)(td, vp, limit)
1014 struct ucred *cred = td->td_ucred;
1016 struct sseg_closure seginfo;
1020 /* Size the program segments. */
1023 each_writable_segment(td, cb_size_segment, &seginfo);
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.
1031 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1033 if (hdrsize + seginfo.size >= limit)
1037 * Allocate memory for building the header, fill it up,
1040 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1044 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
1046 /* Write the contents of all of the writable segments. */
1052 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
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,
1062 offset += php->p_filesz;
1072 * A callback for each_writable_segment() to write out the segment's
1073 * program header entry.
1076 cb_put_phdr(entry, closure)
1077 vm_map_entry_t entry;
1080 struct phdr_closure *phc = (struct phdr_closure *)closure;
1081 Elf_Phdr *phdr = phc->phdr;
1083 phc->offset = round_page(phc->offset);
1085 phdr->p_type = PT_LOAD;
1086 phdr->p_offset = phc->offset;
1087 phdr->p_vaddr = entry->start;
1089 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1090 phdr->p_align = PAGE_SIZE;
1091 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1093 phc->offset += phdr->p_filesz;
1098 * A callback for each_writable_segment() to gather information about
1099 * the number of segments and their total size.
1102 cb_size_segment(entry, closure)
1103 vm_map_entry_t entry;
1106 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1109 ssc->size += entry->end - entry->start;
1113 * For each writable segment in the process's memory map, call the given
1114 * function with a pointer to the map entry and some arbitrary
1115 * caller-supplied data.
1118 each_writable_segment(td, func, closure)
1120 segment_callback func;
1123 struct proc *p = td->td_proc;
1124 vm_map_t map = &p->p_vmspace->vm_map;
1125 vm_map_entry_t entry;
1126 vm_object_t backing_object, object;
1127 boolean_t ignore_entry;
1129 vm_map_lock_read(map);
1130 for (entry = map->header.next; entry != &map->header;
1131 entry = entry->next) {
1133 * Don't dump inaccessible mappings, deal with legacy
1136 * Note that read-only segments related to the elf binary
1137 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1138 * need to arbitrarily ignore such segments.
1140 if (elf_legacy_coredump) {
1141 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1144 if ((entry->protection & VM_PROT_ALL) == 0)
1149 * Dont include memory segment in the coredump if
1150 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1151 * madvise(2). Do not dump submaps (i.e. parts of the
1154 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1157 if ((object = entry->object.vm_object) == NULL)
1160 /* Ignore memory-mapped devices and such things. */
1161 VM_OBJECT_LOCK(object);
1162 while ((backing_object = object->backing_object) != NULL) {
1163 VM_OBJECT_LOCK(backing_object);
1164 VM_OBJECT_UNLOCK(object);
1165 object = backing_object;
1167 ignore_entry = object->type != OBJT_DEFAULT &&
1168 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1169 VM_OBJECT_UNLOCK(object);
1173 (*func)(entry, closure);
1175 vm_map_unlock_read(map);
1179 * Write the core file header to the file, including padding up to
1180 * the page boundary.
1183 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1193 /* Fill in the header. */
1194 bzero(hdr, hdrsize);
1196 __elfN(puthdr)(td, hdr, &off, numsegs);
1198 /* Write it to the core file. */
1199 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1200 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1204 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1205 #include <compat/freebsd32/freebsd32.h>
1207 typedef struct prstatus32 elf_prstatus_t;
1208 typedef struct prpsinfo32 elf_prpsinfo_t;
1209 typedef struct fpreg32 elf_prfpregset_t;
1210 typedef struct fpreg32 elf_fpregset_t;
1211 typedef struct reg32 elf_gregset_t;
1212 typedef struct thrmisc32 elf_thrmisc_t;
1214 typedef prstatus_t elf_prstatus_t;
1215 typedef prpsinfo_t elf_prpsinfo_t;
1216 typedef prfpregset_t elf_prfpregset_t;
1217 typedef prfpregset_t elf_fpregset_t;
1218 typedef gregset_t elf_gregset_t;
1219 typedef thrmisc_t elf_thrmisc_t;
1223 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1226 elf_prstatus_t status;
1227 elf_prfpregset_t fpregset;
1228 elf_prpsinfo_t psinfo;
1229 elf_thrmisc_t thrmisc;
1231 elf_prstatus_t *status;
1232 elf_prfpregset_t *fpregset;
1233 elf_prpsinfo_t *psinfo;
1234 elf_thrmisc_t *thrmisc;
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;
1257 thrmisc = &tempdata->thrmisc;
1267 psinfo->pr_version = PRPSINFO_VERSION;
1268 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1269 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1271 * XXX - We don't fill in the command line arguments properly
1274 strlcpy(psinfo->pr_psargs, p->p_comm,
1275 sizeof(psinfo->pr_psargs));
1277 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1281 * To have the debugger select the right thread (LWP) as the initial
1282 * thread, we dump the state of the thread passed to us in td first.
1283 * This is the thread that causes the core dump and thus likely to
1284 * be the right thread one wants to have selected in the debugger.
1287 while (thr != NULL) {
1289 status->pr_version = PRSTATUS_VERSION;
1290 status->pr_statussz = sizeof(elf_prstatus_t);
1291 status->pr_gregsetsz = sizeof(elf_gregset_t);
1292 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1293 status->pr_osreldate = osreldate;
1294 status->pr_cursig = p->p_sig;
1295 status->pr_pid = thr->td_tid;
1296 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1297 fill_regs32(thr, &status->pr_reg);
1298 fill_fpregs32(thr, fpregset);
1300 fill_regs(thr, &status->pr_reg);
1301 fill_fpregs(thr, fpregset);
1303 memset(&thrmisc->_pad, 0, sizeof (thrmisc->_pad));
1304 strcpy(thrmisc->pr_tname, thr->td_name);
1306 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1308 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1310 __elfN(putnote)(dst, off, "FreeBSD", NT_THRMISC, thrmisc,
1313 * Allow for MD specific notes, as well as any MD
1314 * specific preparations for writing MI notes.
1316 __elfN(dump_thread)(thr, dst, off);
1318 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1319 TAILQ_NEXT(thr, td_plist);
1321 thr = TAILQ_NEXT(thr, td_plist);
1324 notesz = *off - noteoff;
1327 free(tempdata, M_TEMP);
1329 /* Align up to a page boundary for the program segments. */
1330 *off = round_page(*off);
1335 struct phdr_closure phc;
1338 * Fill in the ELF header.
1340 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1341 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1342 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1343 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1344 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1345 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1346 ehdr->e_ident[EI_DATA] = ELF_DATA;
1347 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1348 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1349 ehdr->e_ident[EI_ABIVERSION] = 0;
1350 ehdr->e_ident[EI_PAD] = 0;
1351 ehdr->e_type = ET_CORE;
1352 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1353 ehdr->e_machine = ELF_ARCH32;
1355 ehdr->e_machine = ELF_ARCH;
1357 ehdr->e_version = EV_CURRENT;
1359 ehdr->e_phoff = phoff;
1361 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1362 ehdr->e_phentsize = sizeof(Elf_Phdr);
1363 ehdr->e_phnum = numsegs + 1;
1364 ehdr->e_shentsize = sizeof(Elf_Shdr);
1366 ehdr->e_shstrndx = SHN_UNDEF;
1369 * Fill in the program header entries.
1371 phdr = (Elf_Phdr *)((char *)dst + phoff);
1373 /* The note segement. */
1374 phdr->p_type = PT_NOTE;
1375 phdr->p_offset = noteoff;
1378 phdr->p_filesz = notesz;
1384 /* All the writable segments from the program. */
1387 each_writable_segment(td, cb_put_phdr, &phc);
1392 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1393 const void *desc, size_t descsz)
1397 note.n_namesz = strlen(name) + 1;
1398 note.n_descsz = descsz;
1401 bcopy(¬e, (char *)dst + *off, sizeof note);
1402 *off += sizeof note;
1404 bcopy(name, (char *)dst + *off, note.n_namesz);
1405 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1407 bcopy(desc, (char *)dst + *off, note.n_descsz);
1408 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1412 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1413 int32_t *osrel, const Elf_Phdr *pnote)
1415 const Elf_Note *note, *note0, *note_end;
1416 const char *note_name;
1419 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
1420 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
1423 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1424 note_end = (const Elf_Note *)(imgp->image_header +
1425 pnote->p_offset + pnote->p_filesz);
1426 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1427 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
1428 (const char *)note < sizeof(Elf_Note))
1430 if (note->n_namesz != checknote->hdr.n_namesz ||
1431 note->n_descsz != checknote->hdr.n_descsz ||
1432 note->n_type != checknote->hdr.n_type)
1434 note_name = (const char *)(note + 1);
1435 if (note_name + checknote->hdr.n_namesz >=
1436 (const char *)note_end || strncmp(checknote->vendor,
1437 note_name, checknote->hdr.n_namesz) != 0)
1441 * Fetch the osreldate for binary
1442 * from the ELF OSABI-note if necessary.
1444 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1445 checknote->trans_osrel != NULL)
1446 return (checknote->trans_osrel(note, osrel));
1450 note = (const Elf_Note *)((const char *)(note + 1) +
1451 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1452 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1459 * Try to find the appropriate ABI-note section for checknote,
1460 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1461 * first page of the image is searched, the same as for headers.
1464 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1467 const Elf_Phdr *phdr;
1468 const Elf_Ehdr *hdr;
1471 hdr = (const Elf_Ehdr *)imgp->image_header;
1472 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1474 for (i = 0; i < hdr->e_phnum; i++) {
1475 if (phdr[i].p_type == PT_NOTE &&
1476 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
1484 * Tell kern_execve.c about it, with a little help from the linker.
1486 static struct execsw __elfN(execsw) = {
1487 __CONCAT(exec_, __elfN(imgact)),
1488 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1490 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
1493 __elfN(trans_prot)(Elf_Word flags)
1499 prot |= VM_PROT_EXECUTE;
1501 prot |= VM_PROT_WRITE;
1503 prot |= VM_PROT_READ;
1504 #if __ELF_WORD_SIZE == 32
1505 #if defined(__amd64__) || defined(__ia64__)
1506 if (i386_read_exec && (flags & PF_R))
1507 prot |= VM_PROT_EXECUTE;
1514 __elfN(untrans_prot)(vm_prot_t prot)
1519 if (prot & VM_PROT_EXECUTE)
1521 if (prot & VM_PROT_READ)
1523 if (prot & VM_PROT_WRITE)