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)(const Elf_Ehdr *hdr,
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);
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_trace = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
103 static int elf_legacy_coredump = 0;
104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
105 &elf_legacy_coredump, 0, "");
107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
109 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
110 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
111 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
114 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
118 for (i = 0; i < MAX_BRANDS; i++) {
119 if (elf_brand_list[i] == NULL) {
120 elf_brand_list[i] = entry;
130 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
134 for (i = 0; i < MAX_BRANDS; i++) {
135 if (elf_brand_list[i] == entry) {
136 elf_brand_list[i] = NULL;
146 __elfN(brand_inuse)(Elf_Brandinfo *entry)
151 sx_slock(&allproc_lock);
152 FOREACH_PROC_IN_SYSTEM(p) {
153 if (p->p_sysent == entry->sysvec) {
158 sx_sunlock(&allproc_lock);
163 static Elf_Brandinfo *
164 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
170 * We support three types of branding -- (1) the ELF EI_OSABI field
171 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
172 * branding w/in the ELF header, and (3) path of the `interp_path'
173 * field. We should also look for an ".note.ABI-tag" ELF section now
174 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
177 /* If the executable has a brand, search for it in the brand list. */
178 for (i = 0; i < MAX_BRANDS; i++) {
179 bi = elf_brand_list[i];
180 if (bi != NULL && hdr->e_machine == bi->machine &&
181 (hdr->e_ident[EI_OSABI] == bi->brand ||
182 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
183 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
187 /* Lacking a known brand, search for a recognized interpreter. */
188 if (interp != NULL) {
189 for (i = 0; i < MAX_BRANDS; i++) {
190 bi = elf_brand_list[i];
191 if (bi != NULL && hdr->e_machine == bi->machine &&
192 strcmp(interp, bi->interp_path) == 0)
197 /* Lacking a recognized interpreter, try the default brand */
198 for (i = 0; i < MAX_BRANDS; i++) {
199 bi = elf_brand_list[i];
200 if (bi != NULL && hdr->e_machine == bi->machine &&
201 __elfN(fallback_brand) == bi->brand)
208 __elfN(check_header)(const Elf_Ehdr *hdr)
214 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
215 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
216 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
217 hdr->e_phentsize != sizeof(Elf_Phdr) ||
218 hdr->e_version != ELF_TARG_VER)
222 * Make sure we have at least one brand for this machine.
225 for (i = 0; i < MAX_BRANDS; i++) {
226 bi = elf_brand_list[i];
227 if (bi != NULL && bi->machine == hdr->e_machine)
237 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
238 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
245 * Create the page if it doesn't exist yet. Ignore errors.
248 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
249 VM_PROT_ALL, VM_PROT_ALL, 0);
253 * Find the page from the underlying object.
256 sf = vm_imgact_map_page(object, offset);
258 return (KERN_FAILURE);
259 off = offset - trunc_page(offset);
260 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
262 vm_imgact_unmap_page(sf);
264 return (KERN_FAILURE);
268 return (KERN_SUCCESS);
272 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
273 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
280 if (start != trunc_page(start)) {
281 rv = __elfN(map_partial)(map, object, offset, start,
282 round_page(start), prot);
285 offset += round_page(start) - start;
286 start = round_page(start);
288 if (end != round_page(end)) {
289 rv = __elfN(map_partial)(map, object, offset +
290 trunc_page(end) - start, trunc_page(end), end, prot);
293 end = trunc_page(end);
296 if (offset & PAGE_MASK) {
298 * The mapping is not page aligned. This means we have
299 * to copy the data. Sigh.
301 rv = vm_map_find(map, NULL, 0, &start, end - start,
302 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
306 return (KERN_SUCCESS);
307 for (; start < end; start += sz) {
308 sf = vm_imgact_map_page(object, offset);
310 return (KERN_FAILURE);
311 off = offset - trunc_page(offset);
313 if (sz > PAGE_SIZE - off)
314 sz = PAGE_SIZE - off;
315 error = copyout((caddr_t)sf_buf_kva(sf) + off,
317 vm_imgact_unmap_page(sf);
319 return (KERN_FAILURE);
325 vm_object_reference(object);
327 rv = vm_map_insert(map, object, offset, start, end,
328 prot, VM_PROT_ALL, cow);
330 if (rv != KERN_SUCCESS)
331 vm_object_deallocate(object);
335 return (KERN_SUCCESS);
340 __elfN(load_section)(struct vmspace *vmspace,
341 vm_object_t object, vm_offset_t offset,
342 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
347 vm_offset_t map_addr;
350 vm_offset_t file_addr;
353 * It's necessary to fail if the filsz + offset taken from the
354 * header is greater than the actual file pager object's size.
355 * If we were to allow this, then the vm_map_find() below would
356 * walk right off the end of the file object and into the ether.
358 * While I'm here, might as well check for something else that
359 * is invalid: filsz cannot be greater than memsz.
361 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
363 uprintf("elf_load_section: truncated ELF file\n");
367 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
368 file_addr = trunc_page_ps(offset, pagesize);
371 * We have two choices. We can either clear the data in the last page
372 * of an oversized mapping, or we can start the anon mapping a page
373 * early and copy the initialized data into that first page. We
374 * choose the second..
377 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
379 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
382 /* cow flags: don't dump readonly sections in core */
383 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
384 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
386 rv = __elfN(map_insert)(&vmspace->vm_map,
388 file_addr, /* file offset */
389 map_addr, /* virtual start */
390 map_addr + map_len,/* virtual end */
393 if (rv != KERN_SUCCESS)
396 /* we can stop now if we've covered it all */
397 if (memsz == filsz) {
404 * We have to get the remaining bit of the file into the first part
405 * of the oversized map segment. This is normally because the .data
406 * segment in the file is extended to provide bss. It's a neat idea
407 * to try and save a page, but it's a pain in the behind to implement.
409 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
410 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
411 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
414 /* This had damn well better be true! */
416 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
417 map_addr + map_len, VM_PROT_ALL, 0);
418 if (rv != KERN_SUCCESS) {
426 sf = vm_imgact_map_page(object, offset + filsz);
430 /* send the page fragment to user space */
431 off = trunc_page_ps(offset + filsz, pagesize) -
432 trunc_page(offset + filsz);
433 error = copyout((caddr_t)sf_buf_kva(sf) + off,
434 (caddr_t)map_addr, copy_len);
435 vm_imgact_unmap_page(sf);
442 * set it to the specified protection.
443 * XXX had better undo the damage from pasting over the cracks here!
445 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
446 round_page(map_addr + map_len), prot, FALSE);
452 * Load the file "file" into memory. It may be either a shared object
455 * The "addr" reference parameter is in/out. On entry, it specifies
456 * the address where a shared object should be loaded. If the file is
457 * an executable, this value is ignored. On exit, "addr" specifies
458 * where the file was actually loaded.
460 * The "entry" reference parameter is out only. On exit, it specifies
461 * the entry point for the loaded file.
464 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
465 u_long *entry, size_t pagesize)
470 struct image_params image_params;
472 const Elf_Ehdr *hdr = NULL;
473 const Elf_Phdr *phdr = NULL;
474 struct nameidata *nd;
475 struct vmspace *vmspace = p->p_vmspace;
477 struct image_params *imgp;
480 u_long base_addr = 0;
481 int vfslocked, error, i, numsegs;
483 if (curthread->td_proc != p)
484 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
486 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
488 attr = &tempdata->attr;
489 imgp = &tempdata->image_params;
492 * Initialize part of the common data
496 imgp->firstpage = NULL;
497 imgp->image_header = NULL;
499 imgp->execlabel = NULL;
502 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
505 if ((error = namei(nd)) != 0) {
509 vfslocked = NDHASGIANT(nd);
510 NDFREE(nd, NDF_ONLY_PNBUF);
511 imgp->vp = nd->ni_vp;
514 * Check permissions, modes, uid, etc on the file, and "open" it.
516 error = exec_check_permissions(imgp);
520 error = exec_map_first_page(imgp);
525 * Also make certain that the interpreter stays the same, so set
526 * its VV_TEXT flag, too.
528 nd->ni_vp->v_vflag |= VV_TEXT;
530 imgp->object = nd->ni_vp->v_object;
532 hdr = (const Elf_Ehdr *)imgp->image_header;
533 if ((error = __elfN(check_header)(hdr)) != 0)
535 if (hdr->e_type == ET_DYN)
537 else if (hdr->e_type == ET_EXEC)
544 /* Only support headers that fit within first page for now */
545 /* (multiplication of two Elf_Half fields will not overflow) */
546 if ((hdr->e_phoff > PAGE_SIZE) ||
547 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
552 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
553 if (!aligned(phdr, Elf_Addr)) {
558 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
559 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
561 if (phdr[i].p_flags & PF_X)
562 prot |= VM_PROT_EXECUTE;
563 if (phdr[i].p_flags & PF_W)
564 prot |= VM_PROT_WRITE;
565 if (phdr[i].p_flags & PF_R)
566 prot |= VM_PROT_READ;
568 if ((error = __elfN(load_section)(vmspace,
569 imgp->object, phdr[i].p_offset,
570 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
571 phdr[i].p_memsz, phdr[i].p_filesz, prot,
575 * Establish the base address if this is the
579 base_addr = trunc_page(phdr[i].p_vaddr +
585 *entry = (unsigned long)hdr->e_entry + rbase;
589 exec_unmap_first_page(imgp);
594 VFS_UNLOCK_GIANT(vfslocked);
595 free(tempdata, M_TEMP);
600 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
603 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
605 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
606 const Elf_Phdr *phdr, *pnote = NULL;
607 Elf_Auxargs *elf_auxargs;
608 struct vmspace *vmspace;
610 u_long text_size = 0, data_size = 0, total_size = 0;
611 u_long text_addr = 0, data_addr = 0;
612 u_long seg_size, seg_addr;
613 u_long addr, entry = 0, proghdr = 0;
615 const char *interp = NULL, *newinterp = NULL;
616 Elf_Brandinfo *brand_info;
617 const Elf_Note *note, *note_end;
619 const char *note_name;
620 struct sysentvec *sv;
623 * Do we have a valid ELF header ?
625 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
626 * if particular brand doesn't support it.
628 if (__elfN(check_header)(hdr) != 0 ||
629 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
633 * From here on down, we return an errno, not -1, as we've
634 * detected an ELF file.
637 if ((hdr->e_phoff > PAGE_SIZE) ||
638 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
639 /* Only support headers in first page for now */
642 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
643 if (!aligned(phdr, Elf_Addr))
645 for (i = 0; i < hdr->e_phnum; i++) {
646 if (phdr[i].p_type == PT_INTERP) {
647 /* Path to interpreter */
648 if (phdr[i].p_filesz > MAXPATHLEN ||
649 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
651 interp = imgp->image_header + phdr[i].p_offset;
656 brand_info = __elfN(get_brandinfo)(hdr, interp);
657 if (brand_info == NULL) {
658 uprintf("ELF binary type \"%u\" not known.\n",
659 hdr->e_ident[EI_OSABI]);
662 if (hdr->e_type == ET_DYN &&
663 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
665 sv = brand_info->sysvec;
666 if (interp != NULL && brand_info->interp_newpath != NULL)
667 newinterp = brand_info->interp_newpath;
670 * Avoid a possible deadlock if the current address space is destroyed
671 * and that address space maps the locked vnode. In the common case,
672 * the locked vnode's v_usecount is decremented but remains greater
673 * than zero. Consequently, the vnode lock is not needed by vrele().
674 * However, in cases where the vnode lock is external, such as nullfs,
675 * v_usecount may become zero.
677 VOP_UNLOCK(imgp->vp, 0);
679 error = exec_new_vmspace(imgp, sv);
680 imgp->proc->p_sysent = sv;
682 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
686 vmspace = imgp->proc->p_vmspace;
688 for (i = 0; i < hdr->e_phnum; i++) {
689 switch (phdr[i].p_type) {
690 case PT_LOAD: /* Loadable segment */
692 if (phdr[i].p_flags & PF_X)
693 prot |= VM_PROT_EXECUTE;
694 if (phdr[i].p_flags & PF_W)
695 prot |= VM_PROT_WRITE;
696 if (phdr[i].p_flags & PF_R)
697 prot |= VM_PROT_READ;
699 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
701 * Some x86 binaries assume read == executable,
702 * notably the M3 runtime and therefore cvsup
704 if (prot & VM_PROT_READ)
705 prot |= VM_PROT_EXECUTE;
708 if ((error = __elfN(load_section)(vmspace,
709 imgp->object, phdr[i].p_offset,
710 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
711 phdr[i].p_memsz, phdr[i].p_filesz, prot,
712 sv->sv_pagesize)) != 0)
716 * If this segment contains the program headers,
717 * remember their virtual address for the AT_PHDR
718 * aux entry. Static binaries don't usually include
721 if (phdr[i].p_offset == 0 &&
722 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
724 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
726 seg_addr = trunc_page(phdr[i].p_vaddr);
727 seg_size = round_page(phdr[i].p_memsz +
728 phdr[i].p_vaddr - seg_addr);
731 * Is this .text or .data? We can't use
732 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
733 * alpha terribly and possibly does other bad
734 * things so we stick to the old way of figuring
735 * it out: If the segment contains the program
736 * entry point, it's a text segment, otherwise it
739 * Note that obreak() assumes that data_addr +
740 * data_size == end of data load area, and the ELF
741 * file format expects segments to be sorted by
742 * address. If multiple data segments exist, the
743 * last one will be used.
745 if (hdr->e_entry >= phdr[i].p_vaddr &&
746 hdr->e_entry < (phdr[i].p_vaddr +
748 text_size = seg_size;
749 text_addr = seg_addr;
750 entry = (u_long)hdr->e_entry;
752 data_size = seg_size;
753 data_addr = seg_addr;
755 total_size += seg_size;
757 case PT_PHDR: /* Program header table info */
758 proghdr = phdr[i].p_vaddr;
768 if (data_addr == 0 && data_size == 0) {
769 data_addr = text_addr;
770 data_size = text_size;
774 * Check limits. It should be safe to check the
775 * limits after loading the segments since we do
776 * not actually fault in all the segments pages.
778 PROC_LOCK(imgp->proc);
779 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
780 text_size > maxtsiz ||
781 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
782 PROC_UNLOCK(imgp->proc);
786 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
787 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
788 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
789 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
792 * We load the dynamic linker where a userland call
793 * to mmap(0, ...) would put it. The rationale behind this
794 * calculation is that it leaves room for the heap to grow to
795 * its maximum allowed size.
797 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
798 lim_max(imgp->proc, RLIMIT_DATA));
799 PROC_UNLOCK(imgp->proc);
801 imgp->entry_addr = entry;
803 if (interp != NULL) {
804 int have_interp = FALSE;
805 VOP_UNLOCK(imgp->vp, 0);
806 if (brand_info->emul_path != NULL &&
807 brand_info->emul_path[0] != '\0') {
808 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
809 snprintf(path, MAXPATHLEN, "%s%s",
810 brand_info->emul_path, interp);
811 error = __elfN(load_file)(imgp->proc, path, &addr,
812 &imgp->entry_addr, sv->sv_pagesize);
817 if (!have_interp && newinterp != NULL) {
818 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
819 &imgp->entry_addr, sv->sv_pagesize);
823 error = __elfN(load_file)(imgp->proc, interp, &addr,
824 &imgp->entry_addr, sv->sv_pagesize);
826 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
828 uprintf("ELF interpreter %s not found\n", interp);
834 * Construct auxargs table (used by the fixup routine)
836 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
837 elf_auxargs->execfd = -1;
838 elf_auxargs->phdr = proghdr;
839 elf_auxargs->phent = hdr->e_phentsize;
840 elf_auxargs->phnum = hdr->e_phnum;
841 elf_auxargs->pagesz = PAGE_SIZE;
842 elf_auxargs->base = addr;
843 elf_auxargs->flags = 0;
844 elf_auxargs->entry = entry;
845 elf_auxargs->trace = elf_trace;
847 imgp->auxargs = elf_auxargs;
848 imgp->interpreted = 0;
851 * Try to fetch the osreldate for FreeBSD binary from the ELF
852 * OSABI-note. Only the first page of the image is searched,
853 * the same as for headers.
855 if (pnote != NULL && pnote->p_offset < PAGE_SIZE &&
856 pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) {
857 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
858 if (!aligned(note, Elf32_Addr)) {
859 free(imgp->auxargs, M_TEMP);
860 imgp->auxargs = NULL;
863 note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset +
865 while (note < note_end) {
866 if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) &&
867 note->n_descsz == sizeof(int32_t) &&
868 note->n_type == 1 /* ABI_NOTETYPE */) {
869 note_name = (const char *)(note + 1);
870 if (strncmp(FREEBSD_ABI_VENDOR, note_name,
871 sizeof(FREEBSD_ABI_VENDOR)) == 0) {
872 imgp->proc->p_osrel = *(const int32_t *)
874 round_page_ps(sizeof(FREEBSD_ABI_VENDOR),
875 sizeof(Elf32_Addr)));
879 note = (const Elf_Note *)((const char *)(note + 1) +
880 round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) +
881 round_page_ps(note->n_descsz, sizeof(Elf32_Addr)));
888 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
891 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
893 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
897 base = (Elf_Addr *)*stack_base;
898 pos = base + (imgp->args->argc + imgp->args->envc + 2);
901 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
903 if (args->execfd != -1) {
904 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
906 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
907 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
908 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
909 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
910 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
911 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
912 AUXARGS_ENTRY(pos, AT_BASE, args->base);
913 AUXARGS_ENTRY(pos, AT_NULL, 0);
915 free(imgp->auxargs, M_TEMP);
916 imgp->auxargs = NULL;
919 suword(base, (long)imgp->args->argc);
920 *stack_base = (register_t *)base;
925 * Code for generating ELF core dumps.
928 typedef void (*segment_callback)(vm_map_entry_t, void *);
930 /* Closure for cb_put_phdr(). */
931 struct phdr_closure {
932 Elf_Phdr *phdr; /* Program header to fill in */
933 Elf_Off offset; /* Offset of segment in core file */
936 /* Closure for cb_size_segment(). */
937 struct sseg_closure {
938 int count; /* Count of writable segments. */
939 size_t size; /* Total size of all writable segments. */
942 static void cb_put_phdr(vm_map_entry_t, void *);
943 static void cb_size_segment(vm_map_entry_t, void *);
944 static void each_writable_segment(struct thread *, segment_callback, void *);
945 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
946 int, void *, size_t);
947 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
948 static void __elfN(putnote)(void *, size_t *, const char *, int,
949 const void *, size_t);
952 __elfN(coredump)(td, vp, limit)
957 struct ucred *cred = td->td_ucred;
959 struct sseg_closure seginfo;
963 /* Size the program segments. */
966 each_writable_segment(td, cb_size_segment, &seginfo);
969 * Calculate the size of the core file header area by making
970 * a dry run of generating it. Nothing is written, but the
971 * size is calculated.
974 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
976 if (hdrsize + seginfo.size >= limit)
980 * Allocate memory for building the header, fill it up,
983 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
987 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
989 /* Write the contents of all of the writable segments. */
995 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
997 for (i = 0; i < seginfo.count; i++) {
998 error = vn_rdwr_inchunks(UIO_WRITE, vp,
999 (caddr_t)(uintptr_t)php->p_vaddr,
1000 php->p_filesz, offset, UIO_USERSPACE,
1001 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1002 curthread); /* XXXKSE */
1005 offset += php->p_filesz;
1015 * A callback for each_writable_segment() to write out the segment's
1016 * program header entry.
1019 cb_put_phdr(entry, closure)
1020 vm_map_entry_t entry;
1023 struct phdr_closure *phc = (struct phdr_closure *)closure;
1024 Elf_Phdr *phdr = phc->phdr;
1026 phc->offset = round_page(phc->offset);
1028 phdr->p_type = PT_LOAD;
1029 phdr->p_offset = phc->offset;
1030 phdr->p_vaddr = entry->start;
1032 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1033 phdr->p_align = PAGE_SIZE;
1035 if (entry->protection & VM_PROT_READ)
1036 phdr->p_flags |= PF_R;
1037 if (entry->protection & VM_PROT_WRITE)
1038 phdr->p_flags |= PF_W;
1039 if (entry->protection & VM_PROT_EXECUTE)
1040 phdr->p_flags |= PF_X;
1042 phc->offset += phdr->p_filesz;
1047 * A callback for each_writable_segment() to gather information about
1048 * the number of segments and their total size.
1051 cb_size_segment(entry, closure)
1052 vm_map_entry_t entry;
1055 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1058 ssc->size += entry->end - entry->start;
1062 * For each writable segment in the process's memory map, call the given
1063 * function with a pointer to the map entry and some arbitrary
1064 * caller-supplied data.
1067 each_writable_segment(td, func, closure)
1069 segment_callback func;
1072 struct proc *p = td->td_proc;
1073 vm_map_t map = &p->p_vmspace->vm_map;
1074 vm_map_entry_t entry;
1075 vm_object_t backing_object, object;
1076 boolean_t ignore_entry;
1078 vm_map_lock_read(map);
1079 for (entry = map->header.next; entry != &map->header;
1080 entry = entry->next) {
1082 * Don't dump inaccessible mappings, deal with legacy
1085 * Note that read-only segments related to the elf binary
1086 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1087 * need to arbitrarily ignore such segments.
1089 if (elf_legacy_coredump) {
1090 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1093 if ((entry->protection & VM_PROT_ALL) == 0)
1098 * Dont include memory segment in the coredump if
1099 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1100 * madvise(2). Do not dump submaps (i.e. parts of the
1103 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1106 if ((object = entry->object.vm_object) == NULL)
1109 /* Ignore memory-mapped devices and such things. */
1110 VM_OBJECT_LOCK(object);
1111 while ((backing_object = object->backing_object) != NULL) {
1112 VM_OBJECT_LOCK(backing_object);
1113 VM_OBJECT_UNLOCK(object);
1114 object = backing_object;
1116 ignore_entry = object->type != OBJT_DEFAULT &&
1117 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1118 VM_OBJECT_UNLOCK(object);
1122 (*func)(entry, closure);
1124 vm_map_unlock_read(map);
1128 * Write the core file header to the file, including padding up to
1129 * the page boundary.
1132 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1142 /* Fill in the header. */
1143 bzero(hdr, hdrsize);
1145 __elfN(puthdr)(td, hdr, &off, numsegs);
1147 /* Write it to the core file. */
1148 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1149 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1153 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1154 typedef struct prstatus32 elf_prstatus_t;
1155 typedef struct prpsinfo32 elf_prpsinfo_t;
1156 typedef struct fpreg32 elf_prfpregset_t;
1157 typedef struct fpreg32 elf_fpregset_t;
1158 typedef struct reg32 elf_gregset_t;
1160 typedef prstatus_t elf_prstatus_t;
1161 typedef prpsinfo_t elf_prpsinfo_t;
1162 typedef prfpregset_t elf_prfpregset_t;
1163 typedef prfpregset_t elf_fpregset_t;
1164 typedef gregset_t elf_gregset_t;
1168 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1171 elf_prstatus_t status;
1172 elf_prfpregset_t fpregset;
1173 elf_prpsinfo_t psinfo;
1175 elf_prstatus_t *status;
1176 elf_prfpregset_t *fpregset;
1177 elf_prpsinfo_t *psinfo;
1180 size_t ehoff, noteoff, notesz, phoff;
1185 *off += sizeof(Elf_Ehdr);
1188 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1192 * Don't allocate space for the notes if we're just calculating
1193 * the size of the header. We also don't collect the data.
1196 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1197 status = &tempdata->status;
1198 fpregset = &tempdata->fpregset;
1199 psinfo = &tempdata->psinfo;
1208 psinfo->pr_version = PRPSINFO_VERSION;
1209 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1210 strlcpy(psinfo->pr_fname, td->td_name, sizeof(psinfo->pr_fname));
1212 * XXX - We don't fill in the command line arguments properly
1215 strlcpy(psinfo->pr_psargs, td->td_name,
1216 sizeof(psinfo->pr_psargs));
1218 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1222 * To have the debugger select the right thread (LWP) as the initial
1223 * thread, we dump the state of the thread passed to us in td first.
1224 * This is the thread that causes the core dump and thus likely to
1225 * be the right thread one wants to have selected in the debugger.
1228 while (thr != NULL) {
1230 status->pr_version = PRSTATUS_VERSION;
1231 status->pr_statussz = sizeof(elf_prstatus_t);
1232 status->pr_gregsetsz = sizeof(elf_gregset_t);
1233 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1234 status->pr_osreldate = osreldate;
1235 status->pr_cursig = p->p_sig;
1236 status->pr_pid = thr->td_tid;
1237 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1238 fill_regs32(thr, &status->pr_reg);
1239 fill_fpregs32(thr, fpregset);
1241 fill_regs(thr, &status->pr_reg);
1242 fill_fpregs(thr, fpregset);
1245 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1247 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1250 * Allow for MD specific notes, as well as any MD
1251 * specific preparations for writing MI notes.
1253 __elfN(dump_thread)(thr, dst, off);
1255 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1256 TAILQ_NEXT(thr, td_plist);
1258 thr = TAILQ_NEXT(thr, td_plist);
1261 notesz = *off - noteoff;
1264 free(tempdata, M_TEMP);
1266 /* Align up to a page boundary for the program segments. */
1267 *off = round_page(*off);
1272 struct phdr_closure phc;
1275 * Fill in the ELF header.
1277 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1278 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1279 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1280 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1281 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1282 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1283 ehdr->e_ident[EI_DATA] = ELF_DATA;
1284 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1285 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1286 ehdr->e_ident[EI_ABIVERSION] = 0;
1287 ehdr->e_ident[EI_PAD] = 0;
1288 ehdr->e_type = ET_CORE;
1289 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1290 ehdr->e_machine = EM_386;
1292 ehdr->e_machine = ELF_ARCH;
1294 ehdr->e_version = EV_CURRENT;
1296 ehdr->e_phoff = phoff;
1298 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1299 ehdr->e_phentsize = sizeof(Elf_Phdr);
1300 ehdr->e_phnum = numsegs + 1;
1301 ehdr->e_shentsize = sizeof(Elf_Shdr);
1303 ehdr->e_shstrndx = SHN_UNDEF;
1306 * Fill in the program header entries.
1308 phdr = (Elf_Phdr *)((char *)dst + phoff);
1310 /* The note segement. */
1311 phdr->p_type = PT_NOTE;
1312 phdr->p_offset = noteoff;
1315 phdr->p_filesz = notesz;
1321 /* All the writable segments from the program. */
1324 each_writable_segment(td, cb_put_phdr, &phc);
1329 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1330 const void *desc, size_t descsz)
1334 note.n_namesz = strlen(name) + 1;
1335 note.n_descsz = descsz;
1338 bcopy(¬e, (char *)dst + *off, sizeof note);
1339 *off += sizeof note;
1341 bcopy(name, (char *)dst + *off, note.n_namesz);
1342 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1344 bcopy(desc, (char *)dst + *off, note.n_descsz);
1345 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1349 * Tell kern_execve.c about it, with a little help from the linker.
1351 static struct execsw __elfN(execsw) = {
1352 __CONCAT(exec_, __elfN(imgact)),
1353 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1355 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));