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 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
74 #include <machine/fpu.h>
75 #include <compat/ia32/ia32_reg.h>
78 #define OLD_EI_BRAND 8
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,
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);
95 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
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));
105 static int elf_legacy_coredump = 0;
106 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
107 &elf_legacy_coredump, 0, "");
109 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
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))
115 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
117 Elf_Brandnote __elfN(freebsd_brandnote) = {
118 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
119 .hdr.n_descsz = sizeof(int32_t),
121 .vendor = FREEBSD_ABI_VENDOR,
122 .flags = BN_TRANSLATE_OSREL,
123 .trans_osrel = __elfN(freebsd_trans_osrel)
127 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
131 p = (uintptr_t)(note + 1);
132 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
133 *osrel = *(const int32_t *)(p);
138 static const char GNU_ABI_VENDOR[] = "GNU";
139 static int GNU_KFREEBSD_ABI_DESC = 3;
141 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
142 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
143 .hdr.n_descsz = 16, /* XXX at least 16 */
145 .vendor = GNU_ABI_VENDOR,
146 .flags = BN_TRANSLATE_OSREL,
147 .trans_osrel = kfreebsd_trans_osrel
151 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
153 const Elf32_Word *desc;
156 p = (uintptr_t)(note + 1);
157 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
159 desc = (const Elf32_Word *)p;
160 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
164 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
165 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
167 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
173 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
177 for (i = 0; i < MAX_BRANDS; i++) {
178 if (elf_brand_list[i] == NULL) {
179 elf_brand_list[i] = entry;
183 if (i == MAX_BRANDS) {
184 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
192 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
196 for (i = 0; i < MAX_BRANDS; i++) {
197 if (elf_brand_list[i] == entry) {
198 elf_brand_list[i] = NULL;
208 __elfN(brand_inuse)(Elf_Brandinfo *entry)
213 sx_slock(&allproc_lock);
214 FOREACH_PROC_IN_SYSTEM(p) {
215 if (p->p_sysent == entry->sysvec) {
220 sx_sunlock(&allproc_lock);
225 static Elf_Brandinfo *
226 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
229 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
235 * We support four types of branding -- (1) the ELF EI_OSABI field
236 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
237 * branding w/in the ELF header, (3) path of the `interp_path'
238 * field, and (4) the ".note.ABI-tag" ELF section.
241 /* Look for an ".note.ABI-tag" ELF section */
242 for (i = 0; i < MAX_BRANDS; i++) {
243 bi = elf_brand_list[i];
246 if (hdr->e_machine == bi->machine && (bi->flags &
247 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
248 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
254 /* If the executable has a brand, search for it in the brand list. */
255 for (i = 0; i < MAX_BRANDS; i++) {
256 bi = elf_brand_list[i];
257 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
259 if (hdr->e_machine == bi->machine &&
260 (hdr->e_ident[EI_OSABI] == bi->brand ||
261 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
262 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
266 /* Lacking a known brand, search for a recognized interpreter. */
267 if (interp != NULL) {
268 for (i = 0; i < MAX_BRANDS; i++) {
269 bi = elf_brand_list[i];
270 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
272 if (hdr->e_machine == bi->machine &&
273 strcmp(interp, bi->interp_path) == 0)
278 /* Lacking a recognized interpreter, try the default brand */
279 for (i = 0; i < MAX_BRANDS; i++) {
280 bi = elf_brand_list[i];
281 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
283 if (hdr->e_machine == bi->machine &&
284 __elfN(fallback_brand) == bi->brand)
291 __elfN(check_header)(const Elf_Ehdr *hdr)
297 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
298 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
299 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
300 hdr->e_phentsize != sizeof(Elf_Phdr) ||
301 hdr->e_version != ELF_TARG_VER)
305 * Make sure we have at least one brand for this machine.
308 for (i = 0; i < MAX_BRANDS; i++) {
309 bi = elf_brand_list[i];
310 if (bi != NULL && bi->machine == hdr->e_machine)
320 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
321 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
328 * Create the page if it doesn't exist yet. Ignore errors.
331 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
332 VM_PROT_ALL, VM_PROT_ALL, 0);
336 * Find the page from the underlying object.
339 sf = vm_imgact_map_page(object, offset);
341 return (KERN_FAILURE);
342 off = offset - trunc_page(offset);
343 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
345 vm_imgact_unmap_page(sf);
347 return (KERN_FAILURE);
351 return (KERN_SUCCESS);
355 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
356 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
363 if (start != trunc_page(start)) {
364 rv = __elfN(map_partial)(map, object, offset, start,
365 round_page(start), prot);
368 offset += round_page(start) - start;
369 start = round_page(start);
371 if (end != round_page(end)) {
372 rv = __elfN(map_partial)(map, object, offset +
373 trunc_page(end) - start, trunc_page(end), end, prot);
376 end = trunc_page(end);
379 if (offset & PAGE_MASK) {
381 * The mapping is not page aligned. This means we have
382 * to copy the data. Sigh.
384 rv = vm_map_find(map, NULL, 0, &start, end - start,
385 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
389 return (KERN_SUCCESS);
390 for (; start < end; start += sz) {
391 sf = vm_imgact_map_page(object, offset);
393 return (KERN_FAILURE);
394 off = offset - trunc_page(offset);
396 if (sz > PAGE_SIZE - off)
397 sz = PAGE_SIZE - off;
398 error = copyout((caddr_t)sf_buf_kva(sf) + off,
400 vm_imgact_unmap_page(sf);
402 return (KERN_FAILURE);
408 vm_object_reference(object);
410 rv = vm_map_insert(map, object, offset, start, end,
411 prot, VM_PROT_ALL, cow);
413 if (rv != KERN_SUCCESS)
414 vm_object_deallocate(object);
418 return (KERN_SUCCESS);
423 __elfN(load_section)(struct vmspace *vmspace,
424 vm_object_t object, vm_offset_t offset,
425 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
430 vm_offset_t map_addr;
433 vm_offset_t file_addr;
436 * It's necessary to fail if the filsz + offset taken from the
437 * header is greater than the actual file pager object's size.
438 * If we were to allow this, then the vm_map_find() below would
439 * walk right off the end of the file object and into the ether.
441 * While I'm here, might as well check for something else that
442 * is invalid: filsz cannot be greater than memsz.
444 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
446 uprintf("elf_load_section: truncated ELF file\n");
450 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
451 file_addr = trunc_page_ps(offset, pagesize);
454 * We have two choices. We can either clear the data in the last page
455 * of an oversized mapping, or we can start the anon mapping a page
456 * early and copy the initialized data into that first page. We
457 * choose the second..
460 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
462 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
465 /* cow flags: don't dump readonly sections in core */
466 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
467 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
469 rv = __elfN(map_insert)(&vmspace->vm_map,
471 file_addr, /* file offset */
472 map_addr, /* virtual start */
473 map_addr + map_len,/* virtual end */
476 if (rv != KERN_SUCCESS)
479 /* we can stop now if we've covered it all */
480 if (memsz == filsz) {
487 * We have to get the remaining bit of the file into the first part
488 * of the oversized map segment. This is normally because the .data
489 * segment in the file is extended to provide bss. It's a neat idea
490 * to try and save a page, but it's a pain in the behind to implement.
492 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
493 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
494 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
497 /* This had damn well better be true! */
499 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
500 map_addr + map_len, VM_PROT_ALL, 0);
501 if (rv != KERN_SUCCESS) {
509 sf = vm_imgact_map_page(object, offset + filsz);
513 /* send the page fragment to user space */
514 off = trunc_page_ps(offset + filsz, pagesize) -
515 trunc_page(offset + filsz);
516 error = copyout((caddr_t)sf_buf_kva(sf) + off,
517 (caddr_t)map_addr, copy_len);
518 vm_imgact_unmap_page(sf);
525 * set it to the specified protection.
526 * XXX had better undo the damage from pasting over the cracks here!
528 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
529 round_page(map_addr + map_len), prot, FALSE);
535 * Load the file "file" into memory. It may be either a shared object
538 * The "addr" reference parameter is in/out. On entry, it specifies
539 * the address where a shared object should be loaded. If the file is
540 * an executable, this value is ignored. On exit, "addr" specifies
541 * where the file was actually loaded.
543 * The "entry" reference parameter is out only. On exit, it specifies
544 * the entry point for the loaded file.
547 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
548 u_long *entry, size_t pagesize)
553 struct image_params image_params;
555 const Elf_Ehdr *hdr = NULL;
556 const Elf_Phdr *phdr = NULL;
557 struct nameidata *nd;
558 struct vmspace *vmspace = p->p_vmspace;
560 struct image_params *imgp;
563 u_long base_addr = 0;
564 int vfslocked, error, i, numsegs;
566 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
568 attr = &tempdata->attr;
569 imgp = &tempdata->image_params;
572 * Initialize part of the common data
576 imgp->firstpage = NULL;
577 imgp->image_header = NULL;
579 imgp->execlabel = NULL;
581 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
584 if ((error = namei(nd)) != 0) {
588 vfslocked = NDHASGIANT(nd);
589 NDFREE(nd, NDF_ONLY_PNBUF);
590 imgp->vp = nd->ni_vp;
593 * Check permissions, modes, uid, etc on the file, and "open" it.
595 error = exec_check_permissions(imgp);
599 error = exec_map_first_page(imgp);
604 * Also make certain that the interpreter stays the same, so set
605 * its VV_TEXT flag, too.
607 nd->ni_vp->v_vflag |= VV_TEXT;
609 imgp->object = nd->ni_vp->v_object;
611 hdr = (const Elf_Ehdr *)imgp->image_header;
612 if ((error = __elfN(check_header)(hdr)) != 0)
614 if (hdr->e_type == ET_DYN)
616 else if (hdr->e_type == ET_EXEC)
623 /* Only support headers that fit within first page for now */
624 /* (multiplication of two Elf_Half fields will not overflow) */
625 if ((hdr->e_phoff > PAGE_SIZE) ||
626 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
631 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
632 if (!aligned(phdr, Elf_Addr)) {
637 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
638 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
639 /* Loadable segment */
641 if (phdr[i].p_flags & PF_X)
642 prot |= VM_PROT_EXECUTE;
643 if (phdr[i].p_flags & PF_W)
644 prot |= VM_PROT_WRITE;
645 if (phdr[i].p_flags & PF_R)
646 prot |= VM_PROT_READ;
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, et_dyn_addr, entry = 0, proghdr = 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 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
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))
722 for (i = 0; i < hdr->e_phnum; i++) {
723 if (phdr[i].p_type == PT_INTERP) {
724 /* Path to interpreter */
725 if (phdr[i].p_filesz > MAXPATHLEN ||
726 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
728 interp = imgp->image_header + phdr[i].p_offset;
733 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
734 if (brand_info == NULL) {
735 uprintf("ELF binary type \"%u\" not known.\n",
736 hdr->e_ident[EI_OSABI]);
739 if (hdr->e_type == ET_DYN) {
740 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
742 et_dyn_addr = ET_DYN_LOAD_ADDR;
745 sv = brand_info->sysvec;
746 if (interp != NULL && brand_info->interp_newpath != NULL)
747 newinterp = brand_info->interp_newpath;
750 * Avoid a possible deadlock if the current address space is destroyed
751 * and that address space maps the locked vnode. In the common case,
752 * the locked vnode's v_usecount is decremented but remains greater
753 * than zero. Consequently, the vnode lock is not needed by vrele().
754 * However, in cases where the vnode lock is external, such as nullfs,
755 * v_usecount may become zero.
757 VOP_UNLOCK(imgp->vp, 0);
759 error = exec_new_vmspace(imgp, sv);
760 imgp->proc->p_sysent = sv;
762 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
766 vmspace = imgp->proc->p_vmspace;
768 for (i = 0; i < hdr->e_phnum; i++) {
769 switch (phdr[i].p_type) {
770 case PT_LOAD: /* Loadable segment */
771 if (phdr[i].p_memsz == 0)
774 if (phdr[i].p_flags & PF_X)
775 prot |= VM_PROT_EXECUTE;
776 if (phdr[i].p_flags & PF_W)
777 prot |= VM_PROT_WRITE;
778 if (phdr[i].p_flags & PF_R)
779 prot |= VM_PROT_READ;
781 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
783 * Some x86 binaries assume read == executable,
784 * notably the M3 runtime and therefore cvsup
786 if (prot & VM_PROT_READ)
787 prot |= VM_PROT_EXECUTE;
790 if ((error = __elfN(load_section)(vmspace,
791 imgp->object, phdr[i].p_offset,
792 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
793 phdr[i].p_memsz, phdr[i].p_filesz, prot,
794 sv->sv_pagesize)) != 0)
798 * If this segment contains the program headers,
799 * remember their virtual address for the AT_PHDR
800 * aux entry. Static binaries don't usually include
803 if (phdr[i].p_offset == 0 &&
804 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
806 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
809 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
810 seg_size = round_page(phdr[i].p_memsz +
811 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
814 * Is this .text or .data? We can't use
815 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
816 * alpha terribly and possibly does other bad
817 * things so we stick to the old way of figuring
818 * it out: If the segment contains the program
819 * entry point, it's a text segment, otherwise it
822 * Note that obreak() assumes that data_addr +
823 * data_size == end of data load area, and the ELF
824 * file format expects segments to be sorted by
825 * address. If multiple data segments exist, the
826 * last one will be used.
828 if (hdr->e_entry >= phdr[i].p_vaddr &&
829 hdr->e_entry < (phdr[i].p_vaddr +
831 text_size = seg_size;
832 text_addr = seg_addr;
833 entry = (u_long)hdr->e_entry + et_dyn_addr;
835 data_size = seg_size;
836 data_addr = seg_addr;
838 total_size += seg_size;
840 case PT_PHDR: /* Program header table info */
841 proghdr = phdr[i].p_vaddr + et_dyn_addr;
848 if (data_addr == 0 && data_size == 0) {
849 data_addr = text_addr;
850 data_size = text_size;
854 * Check limits. It should be safe to check the
855 * limits after loading the segments since we do
856 * not actually fault in all the segments pages.
858 PROC_LOCK(imgp->proc);
859 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
860 text_size > maxtsiz ||
861 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
862 PROC_UNLOCK(imgp->proc);
866 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
867 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
868 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
869 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
872 * We load the dynamic linker where a userland call
873 * to mmap(0, ...) would put it. The rationale behind this
874 * calculation is that it leaves room for the heap to grow to
875 * its maximum allowed size.
877 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
878 lim_max(imgp->proc, RLIMIT_DATA));
879 PROC_UNLOCK(imgp->proc);
881 imgp->entry_addr = entry;
883 if (interp != NULL) {
884 int have_interp = FALSE;
885 VOP_UNLOCK(imgp->vp, 0);
886 if (brand_info->emul_path != NULL &&
887 brand_info->emul_path[0] != '\0') {
888 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
889 snprintf(path, MAXPATHLEN, "%s%s",
890 brand_info->emul_path, interp);
891 error = __elfN(load_file)(imgp->proc, path, &addr,
892 &imgp->entry_addr, sv->sv_pagesize);
897 if (!have_interp && newinterp != NULL) {
898 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
899 &imgp->entry_addr, sv->sv_pagesize);
904 error = __elfN(load_file)(imgp->proc, interp, &addr,
905 &imgp->entry_addr, sv->sv_pagesize);
907 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
909 uprintf("ELF interpreter %s not found\n", interp);
916 * Construct auxargs table (used by the fixup routine)
918 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
919 elf_auxargs->execfd = -1;
920 elf_auxargs->phdr = proghdr;
921 elf_auxargs->phent = hdr->e_phentsize;
922 elf_auxargs->phnum = hdr->e_phnum;
923 elf_auxargs->pagesz = PAGE_SIZE;
924 elf_auxargs->base = addr;
925 elf_auxargs->flags = 0;
926 elf_auxargs->entry = entry;
928 imgp->auxargs = elf_auxargs;
929 imgp->interpreted = 0;
930 imgp->proc->p_osrel = osrel;
935 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
938 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
940 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
944 base = (Elf_Addr *)*stack_base;
945 pos = base + (imgp->args->argc + imgp->args->envc + 2);
947 if (args->execfd != -1)
948 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
949 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
950 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
951 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
952 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
953 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
954 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
955 AUXARGS_ENTRY(pos, AT_BASE, args->base);
956 if (imgp->execpathp != 0)
957 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
958 AUXARGS_ENTRY(pos, AT_NULL, 0);
960 free(imgp->auxargs, M_TEMP);
961 imgp->auxargs = NULL;
964 suword(base, (long)imgp->args->argc);
965 *stack_base = (register_t *)base;
970 * Code for generating ELF core dumps.
973 typedef void (*segment_callback)(vm_map_entry_t, void *);
975 /* Closure for cb_put_phdr(). */
976 struct phdr_closure {
977 Elf_Phdr *phdr; /* Program header to fill in */
978 Elf_Off offset; /* Offset of segment in core file */
981 /* Closure for cb_size_segment(). */
982 struct sseg_closure {
983 int count; /* Count of writable segments. */
984 size_t size; /* Total size of all writable segments. */
987 static void cb_put_phdr(vm_map_entry_t, void *);
988 static void cb_size_segment(vm_map_entry_t, void *);
989 static void each_writable_segment(struct thread *, segment_callback, void *);
990 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
991 int, void *, size_t);
992 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
993 static void __elfN(putnote)(void *, size_t *, const char *, int,
994 const void *, size_t);
997 __elfN(coredump)(td, vp, limit)
1002 struct ucred *cred = td->td_ucred;
1004 struct sseg_closure seginfo;
1008 /* Size the program segments. */
1011 each_writable_segment(td, cb_size_segment, &seginfo);
1014 * Calculate the size of the core file header area by making
1015 * a dry run of generating it. Nothing is written, but the
1016 * size is calculated.
1019 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1021 if (hdrsize + seginfo.size >= limit)
1025 * Allocate memory for building the header, fill it up,
1028 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1032 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
1034 /* Write the contents of all of the writable segments. */
1040 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1042 for (i = 0; i < seginfo.count; i++) {
1043 error = vn_rdwr_inchunks(UIO_WRITE, vp,
1044 (caddr_t)(uintptr_t)php->p_vaddr,
1045 php->p_filesz, offset, UIO_USERSPACE,
1046 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1050 offset += php->p_filesz;
1060 * A callback for each_writable_segment() to write out the segment's
1061 * program header entry.
1064 cb_put_phdr(entry, closure)
1065 vm_map_entry_t entry;
1068 struct phdr_closure *phc = (struct phdr_closure *)closure;
1069 Elf_Phdr *phdr = phc->phdr;
1071 phc->offset = round_page(phc->offset);
1073 phdr->p_type = PT_LOAD;
1074 phdr->p_offset = phc->offset;
1075 phdr->p_vaddr = entry->start;
1077 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1078 phdr->p_align = PAGE_SIZE;
1080 if (entry->protection & VM_PROT_READ)
1081 phdr->p_flags |= PF_R;
1082 if (entry->protection & VM_PROT_WRITE)
1083 phdr->p_flags |= PF_W;
1084 if (entry->protection & VM_PROT_EXECUTE)
1085 phdr->p_flags |= PF_X;
1087 phc->offset += phdr->p_filesz;
1092 * A callback for each_writable_segment() to gather information about
1093 * the number of segments and their total size.
1096 cb_size_segment(entry, closure)
1097 vm_map_entry_t entry;
1100 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1103 ssc->size += entry->end - entry->start;
1107 * For each writable segment in the process's memory map, call the given
1108 * function with a pointer to the map entry and some arbitrary
1109 * caller-supplied data.
1112 each_writable_segment(td, func, closure)
1114 segment_callback func;
1117 struct proc *p = td->td_proc;
1118 vm_map_t map = &p->p_vmspace->vm_map;
1119 vm_map_entry_t entry;
1120 vm_object_t backing_object, object;
1121 boolean_t ignore_entry;
1123 vm_map_lock_read(map);
1124 for (entry = map->header.next; entry != &map->header;
1125 entry = entry->next) {
1127 * Don't dump inaccessible mappings, deal with legacy
1130 * Note that read-only segments related to the elf binary
1131 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1132 * need to arbitrarily ignore such segments.
1134 if (elf_legacy_coredump) {
1135 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1138 if ((entry->protection & VM_PROT_ALL) == 0)
1143 * Dont include memory segment in the coredump if
1144 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1145 * madvise(2). Do not dump submaps (i.e. parts of the
1148 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1151 if ((object = entry->object.vm_object) == NULL)
1154 /* Ignore memory-mapped devices and such things. */
1155 VM_OBJECT_LOCK(object);
1156 while ((backing_object = object->backing_object) != NULL) {
1157 VM_OBJECT_LOCK(backing_object);
1158 VM_OBJECT_UNLOCK(object);
1159 object = backing_object;
1161 ignore_entry = object->type != OBJT_DEFAULT &&
1162 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1163 VM_OBJECT_UNLOCK(object);
1167 (*func)(entry, closure);
1169 vm_map_unlock_read(map);
1173 * Write the core file header to the file, including padding up to
1174 * the page boundary.
1177 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1187 /* Fill in the header. */
1188 bzero(hdr, hdrsize);
1190 __elfN(puthdr)(td, hdr, &off, numsegs);
1192 /* Write it to the core file. */
1193 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1194 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1198 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1199 typedef struct prstatus32 elf_prstatus_t;
1200 typedef struct prpsinfo32 elf_prpsinfo_t;
1201 typedef struct fpreg32 elf_prfpregset_t;
1202 typedef struct fpreg32 elf_fpregset_t;
1203 typedef struct reg32 elf_gregset_t;
1205 typedef prstatus_t elf_prstatus_t;
1206 typedef prpsinfo_t elf_prpsinfo_t;
1207 typedef prfpregset_t elf_prfpregset_t;
1208 typedef prfpregset_t elf_fpregset_t;
1209 typedef gregset_t elf_gregset_t;
1213 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1216 elf_prstatus_t status;
1217 elf_prfpregset_t fpregset;
1218 elf_prpsinfo_t psinfo;
1220 elf_prstatus_t *status;
1221 elf_prfpregset_t *fpregset;
1222 elf_prpsinfo_t *psinfo;
1225 size_t ehoff, noteoff, notesz, phoff;
1230 *off += sizeof(Elf_Ehdr);
1233 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1237 * Don't allocate space for the notes if we're just calculating
1238 * the size of the header. We also don't collect the data.
1241 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1242 status = &tempdata->status;
1243 fpregset = &tempdata->fpregset;
1244 psinfo = &tempdata->psinfo;
1253 psinfo->pr_version = PRPSINFO_VERSION;
1254 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1255 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1257 * XXX - We don't fill in the command line arguments properly
1260 strlcpy(psinfo->pr_psargs, p->p_comm,
1261 sizeof(psinfo->pr_psargs));
1263 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1267 * To have the debugger select the right thread (LWP) as the initial
1268 * thread, we dump the state of the thread passed to us in td first.
1269 * This is the thread that causes the core dump and thus likely to
1270 * be the right thread one wants to have selected in the debugger.
1273 while (thr != NULL) {
1275 status->pr_version = PRSTATUS_VERSION;
1276 status->pr_statussz = sizeof(elf_prstatus_t);
1277 status->pr_gregsetsz = sizeof(elf_gregset_t);
1278 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1279 status->pr_osreldate = osreldate;
1280 status->pr_cursig = p->p_sig;
1281 status->pr_pid = thr->td_tid;
1282 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1283 fill_regs32(thr, &status->pr_reg);
1284 fill_fpregs32(thr, fpregset);
1286 fill_regs(thr, &status->pr_reg);
1287 fill_fpregs(thr, fpregset);
1290 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1292 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1295 * Allow for MD specific notes, as well as any MD
1296 * specific preparations for writing MI notes.
1298 __elfN(dump_thread)(thr, dst, off);
1300 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1301 TAILQ_NEXT(thr, td_plist);
1303 thr = TAILQ_NEXT(thr, td_plist);
1306 notesz = *off - noteoff;
1309 free(tempdata, M_TEMP);
1311 /* Align up to a page boundary for the program segments. */
1312 *off = round_page(*off);
1317 struct phdr_closure phc;
1320 * Fill in the ELF header.
1322 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1323 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1324 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1325 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1326 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1327 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1328 ehdr->e_ident[EI_DATA] = ELF_DATA;
1329 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1330 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1331 ehdr->e_ident[EI_ABIVERSION] = 0;
1332 ehdr->e_ident[EI_PAD] = 0;
1333 ehdr->e_type = ET_CORE;
1334 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1335 ehdr->e_machine = EM_386;
1337 ehdr->e_machine = ELF_ARCH;
1339 ehdr->e_version = EV_CURRENT;
1341 ehdr->e_phoff = phoff;
1343 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1344 ehdr->e_phentsize = sizeof(Elf_Phdr);
1345 ehdr->e_phnum = numsegs + 1;
1346 ehdr->e_shentsize = sizeof(Elf_Shdr);
1348 ehdr->e_shstrndx = SHN_UNDEF;
1351 * Fill in the program header entries.
1353 phdr = (Elf_Phdr *)((char *)dst + phoff);
1355 /* The note segement. */
1356 phdr->p_type = PT_NOTE;
1357 phdr->p_offset = noteoff;
1360 phdr->p_filesz = notesz;
1366 /* All the writable segments from the program. */
1369 each_writable_segment(td, cb_put_phdr, &phc);
1374 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1375 const void *desc, size_t descsz)
1379 note.n_namesz = strlen(name) + 1;
1380 note.n_descsz = descsz;
1383 bcopy(¬e, (char *)dst + *off, sizeof note);
1384 *off += sizeof note;
1386 bcopy(name, (char *)dst + *off, note.n_namesz);
1387 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1389 bcopy(desc, (char *)dst + *off, note.n_descsz);
1390 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1394 * Try to find the appropriate ABI-note section for checknote,
1395 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1396 * first page of the image is searched, the same as for headers.
1399 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1402 const Elf_Note *note, *note0, *note_end;
1403 const Elf_Phdr *phdr, *pnote;
1404 const Elf_Ehdr *hdr;
1405 const char *note_name;
1409 hdr = (const Elf_Ehdr *)imgp->image_header;
1410 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1412 for (i = 0; i < hdr->e_phnum; i++) {
1413 if (phdr[i].p_type == PT_NOTE) {
1419 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1420 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
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))
1429 if (note->n_namesz != checknote->hdr.n_namesz ||
1430 note->n_descsz != checknote->hdr.n_descsz ||
1431 note->n_type != checknote->hdr.n_type)
1433 note_name = (const char *)(note + 1);
1434 if (strncmp(checknote->vendor, note_name,
1435 checknote->hdr.n_namesz) != 0)
1439 * Fetch the osreldate for binary
1440 * from the ELF OSABI-note if necessary.
1442 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1443 checknote->trans_osrel != NULL)
1444 return (checknote->trans_osrel(note, osrel));
1448 note = (const Elf_Note *)((const char *)(note + 1) +
1449 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1450 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1457 * Tell kern_execve.c about it, with a little help from the linker.
1459 static struct execsw __elfN(execsw) = {
1460 __CONCAT(exec_, __elfN(imgact)),
1461 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1463 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));