2 * Copyright (c) 1995-1996 Søren Schmidt
3 * Copyright (c) 1996 Peter Wemm
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer
11 * in this position and unchanged.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 * derived from this software withough specific prior written permission
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 * $FreeBSD: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
30 * $DragonFly: src/sys/kern/imgact_elf.c,v 1.10 2003/10/19 19:24:18 dillon Exp $
33 #include <sys/param.h>
35 #include <sys/fcntl.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
42 #include <sys/systm.h>
44 #include <sys/namei.h>
45 #include <sys/pioctl.h>
46 #include <sys/procfs.h>
47 #include <sys/resourcevar.h>
48 #include <sys/signalvar.h>
50 #include <sys/syscall.h>
51 #include <sys/sysctl.h>
52 #include <sys/sysent.h>
53 #include <sys/vnode.h>
56 #include <vm/vm_kern.h>
57 #include <vm/vm_param.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_extern.h>
64 #include <machine/elf.h>
65 #include <machine/md_var.h>
66 #include <sys/mount.h>
68 #define OLD_EI_BRAND 8
73 static int elf_check_header (const Elf_Ehdr *hdr);
74 static int elf_freebsd_fixup (register_t **stack_base,
75 struct image_params *imgp);
76 static int elf_load_file (struct proc *p, const char *file, u_long *addr,
78 static int elf_load_section (struct proc *p,
79 struct vmspace *vmspace, struct vnode *vp,
80 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
82 static int exec_elf_imgact (struct image_params *imgp);
84 static int elf_trace = 0;
85 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
86 static int elf_legacy_coredump = 0;
87 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
88 &elf_legacy_coredump, 0, "");
90 static struct sysentvec elf_freebsd_sysvec = {
110 static Elf_Brandinfo freebsd_brand_info = {
114 "/usr/libexec/ld-elf.so.1",
117 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
120 NULL, NULL, NULL, NULL
124 elf_insert_brand_entry(Elf_Brandinfo *entry)
128 for (i=1; i<MAX_BRANDS; i++) {
129 if (elf_brand_list[i] == NULL) {
130 elf_brand_list[i] = entry;
140 elf_remove_brand_entry(Elf_Brandinfo *entry)
144 for (i=1; i<MAX_BRANDS; i++) {
145 if (elf_brand_list[i] == entry) {
146 elf_brand_list[i] = NULL;
156 elf_brand_inuse(Elf_Brandinfo *entry)
161 FOREACH_PROC_IN_SYSTEM(p) {
162 if (p->p_sysent == entry->sysvec) {
172 elf_check_header(const Elf_Ehdr *hdr)
175 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
176 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
177 hdr->e_ident[EI_VERSION] != EV_CURRENT)
180 if (!ELF_MACHINE_OK(hdr->e_machine))
183 if (hdr->e_version != ELF_TARG_VER)
190 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp,
191 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
195 vm_offset_t map_addr;
200 vm_offset_t file_addr;
201 vm_offset_t data_buf = 0;
203 VOP_GETVOBJECT(vp, &object);
207 * It's necessary to fail if the filsz + offset taken from the
208 * header is greater than the actual file pager object's size.
209 * If we were to allow this, then the vm_map_find() below would
210 * walk right off the end of the file object and into the ether.
212 * While I'm here, might as well check for something else that
213 * is invalid: filsz cannot be greater than memsz.
215 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
217 uprintf("elf_load_section: truncated ELF file\n");
221 map_addr = trunc_page((vm_offset_t)vmaddr);
222 file_addr = trunc_page(offset);
225 * We have two choices. We can either clear the data in the last page
226 * of an oversized mapping, or we can start the anon mapping a page
227 * early and copy the initialized data into that first page. We
228 * choose the second..
231 map_len = trunc_page(offset+filsz) - file_addr;
233 map_len = round_page(offset+filsz) - file_addr;
236 vm_object_reference(object);
238 /* cow flags: don't dump readonly sections in core */
239 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
240 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
242 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
243 vm_map_lock(&vmspace->vm_map);
244 rv = vm_map_insert(&vmspace->vm_map, &count,
246 file_addr, /* file offset */
247 map_addr, /* virtual start */
248 map_addr + map_len,/* virtual end */
252 vm_map_unlock(&vmspace->vm_map);
253 vm_map_entry_release(count);
254 if (rv != KERN_SUCCESS) {
255 vm_object_deallocate(object);
259 /* we can stop now if we've covered it all */
260 if (memsz == filsz) {
267 * We have to get the remaining bit of the file into the first part
268 * of the oversized map segment. This is normally because the .data
269 * segment in the file is extended to provide bss. It's a neat idea
270 * to try and save a page, but it's a pain in the behind to implement.
272 copy_len = (offset + filsz) - trunc_page(offset + filsz);
273 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
274 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
276 /* This had damn well better be true! */
278 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
279 vm_map_lock(&vmspace->vm_map);
280 rv = vm_map_insert(&vmspace->vm_map, &count,
282 map_addr, map_addr + map_len,
283 VM_PROT_ALL, VM_PROT_ALL, 0);
284 vm_map_unlock(&vmspace->vm_map);
285 vm_map_entry_release(count);
286 if (rv != KERN_SUCCESS) {
292 vm_object_reference(object);
293 rv = vm_map_find(exec_map,
295 trunc_page(offset + filsz),
301 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
302 if (rv != KERN_SUCCESS) {
303 vm_object_deallocate(object);
307 /* send the page fragment to user space */
308 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
309 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
316 * set it to the specified protection
318 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot,
325 * Load the file "file" into memory. It may be either a shared object
328 * The "addr" reference parameter is in/out. On entry, it specifies
329 * the address where a shared object should be loaded. If the file is
330 * an executable, this value is ignored. On exit, "addr" specifies
331 * where the file was actually loaded.
333 * The "entry" reference parameter is out only. On exit, it specifies
334 * the entry point for the loaded file.
337 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
342 struct image_params image_params;
344 const Elf_Ehdr *hdr = NULL;
345 const Elf_Phdr *phdr = NULL;
346 struct nameidata *nd;
347 struct vmspace *vmspace = p->p_vmspace;
349 struct image_params *imgp;
352 u_long base_addr = 0;
353 int error, i, numsegs;
354 struct thread *td = p->p_thread;
356 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
358 attr = &tempdata->attr;
359 imgp = &tempdata->image_params;
362 * Initialize part of the common data
367 imgp->firstpage = NULL;
368 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
370 if (imgp->image_header == NULL) {
376 NDINIT(nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW,
377 UIO_SYSSPACE, file, td);
379 if ((error = namei(nd)) != 0) {
383 NDFREE(nd, NDF_ONLY_PNBUF);
384 imgp->vp = nd->ni_vp;
387 * Check permissions, modes, uid, etc on the file, and "open" it.
389 error = exec_check_permissions(imgp);
391 VOP_UNLOCK(nd->ni_vp, 0, td);
395 error = exec_map_first_page(imgp);
397 * Also make certain that the interpreter stays the same, so set
398 * its VTEXT flag, too.
401 nd->ni_vp->v_flag |= VTEXT;
402 VOP_UNLOCK(nd->ni_vp, 0, td);
406 hdr = (const Elf_Ehdr *)imgp->image_header;
407 if ((error = elf_check_header(hdr)) != 0)
409 if (hdr->e_type == ET_DYN)
411 else if (hdr->e_type == ET_EXEC)
418 /* Only support headers that fit within first page for now */
419 if ((hdr->e_phoff > PAGE_SIZE) ||
420 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
425 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
427 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
428 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
430 if (phdr[i].p_flags & PF_X)
431 prot |= VM_PROT_EXECUTE;
432 if (phdr[i].p_flags & PF_W)
433 prot |= VM_PROT_WRITE;
434 if (phdr[i].p_flags & PF_R)
435 prot |= VM_PROT_READ;
437 error = elf_load_section(
438 p, vmspace, nd->ni_vp,
440 (caddr_t)phdr[i].p_vaddr +
443 phdr[i].p_filesz, prot);
447 * Establish the base address if this is the
451 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
456 *entry=(unsigned long)hdr->e_entry + rbase;
460 exec_unmap_first_page(imgp);
461 if (imgp->image_header)
462 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
467 free(tempdata, M_TEMP);
473 * non static, as it can be overridden by start_init()
475 int fallback_elf_brand = -1;
476 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
477 &fallback_elf_brand, -1,
478 "ELF brand of last resort");
481 exec_elf_imgact(struct image_params *imgp)
483 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
484 const Elf_Phdr *phdr;
485 Elf_Auxargs *elf_auxargs = NULL;
486 struct vmspace *vmspace;
488 u_long text_size = 0, data_size = 0, total_size = 0;
489 u_long text_addr = 0, data_addr = 0;
490 u_long seg_size, seg_addr;
491 u_long addr, entry = 0, proghdr = 0;
493 const char *interp = NULL;
494 Elf_Brandinfo *brand_info;
498 * Do we have a valid ELF header ?
500 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
504 * From here on down, we return an errno, not -1, as we've
505 * detected an ELF file.
508 if ((hdr->e_phoff > PAGE_SIZE) ||
509 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
510 /* Only support headers in first page for now */
513 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
516 * From this point on, we may have resources that need to be freed.
519 if ((error = exec_extract_strings(imgp)) != 0)
522 exec_new_vmspace(imgp);
525 * Yeah, I'm paranoid. There is every reason in the world to get
526 * VTEXT now since from here on out, there are places we can have
527 * a context switch. Better safe than sorry; I really don't want
528 * the file to change while it's being loaded.
530 lwkt_gettoken(&imgp->vp->v_interlock);
531 imgp->vp->v_flag |= VTEXT;
532 lwkt_reltoken(&imgp->vp->v_interlock);
534 vmspace = imgp->proc->p_vmspace;
536 for (i = 0; i < hdr->e_phnum; i++) {
537 switch(phdr[i].p_type) {
539 case PT_LOAD: /* Loadable segment */
541 if (phdr[i].p_flags & PF_X)
542 prot |= VM_PROT_EXECUTE;
543 if (phdr[i].p_flags & PF_W)
544 prot |= VM_PROT_WRITE;
545 if (phdr[i].p_flags & PF_R)
546 prot |= VM_PROT_READ;
548 if ((error = elf_load_section(imgp->proc,
551 (caddr_t)phdr[i].p_vaddr,
553 phdr[i].p_filesz, prot)) != 0)
556 seg_addr = trunc_page(phdr[i].p_vaddr);
557 seg_size = round_page(phdr[i].p_memsz +
558 phdr[i].p_vaddr - seg_addr);
561 * Is this .text or .data? We can't use
562 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
563 * alpha terribly and possibly does other bad
564 * things so we stick to the old way of figuring
565 * it out: If the segment contains the program
566 * entry point, it's a text segment, otherwise it
569 * Note that obreak() assumes that data_addr +
570 * data_size == end of data load area, and the ELF
571 * file format expects segments to be sorted by
572 * address. If multiple data segments exist, the
573 * last one will be used.
575 if (hdr->e_entry >= phdr[i].p_vaddr &&
576 hdr->e_entry < (phdr[i].p_vaddr +
578 text_size = seg_size;
579 text_addr = seg_addr;
580 entry = (u_long)hdr->e_entry;
582 data_size = seg_size;
583 data_addr = seg_addr;
585 total_size += seg_size;
588 * Check limits. It should be safe to check the
589 * limits after loading the segment since we do
590 * not actually fault in all the segment's pages.
593 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
594 text_size > maxtsiz ||
596 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
601 case PT_INTERP: /* Path to interpreter */
602 if (phdr[i].p_filesz > MAXPATHLEN ||
603 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
607 interp = imgp->image_header + phdr[i].p_offset;
609 case PT_PHDR: /* Program header table info */
610 proghdr = phdr[i].p_vaddr;
617 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
618 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
619 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
620 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
622 addr = ELF_RTLD_ADDR(vmspace);
624 imgp->entry_addr = entry;
628 /* We support three types of branding -- (1) the ELF EI_OSABI field
629 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
630 * branding w/in the ELF header, and (3) path of the `interp_path'
631 * field. We should also look for an ".note.ABI-tag" ELF section now
632 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
635 /* If the executable has a brand, search for it in the brand list. */
636 if (brand_info == NULL) {
637 for (i = 0; i < MAX_BRANDS; i++) {
638 Elf_Brandinfo *bi = elf_brand_list[i];
641 (hdr->e_ident[EI_OSABI] == bi->brand
643 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
644 bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
651 /* Lacking a known brand, search for a recognized interpreter. */
652 if (brand_info == NULL && interp != NULL) {
653 for (i = 0; i < MAX_BRANDS; i++) {
654 Elf_Brandinfo *bi = elf_brand_list[i];
657 strcmp(interp, bi->interp_path) == 0) {
664 /* Lacking a recognized interpreter, try the default brand */
665 if (brand_info == NULL) {
666 for (i = 0; i < MAX_BRANDS; i++) {
667 Elf_Brandinfo *bi = elf_brand_list[i];
669 if (bi != NULL && fallback_elf_brand == bi->brand) {
676 if (brand_info == NULL) {
677 uprintf("ELF binary type \"%u\" not known.\n",
678 hdr->e_ident[EI_OSABI]);
683 imgp->proc->p_sysent = brand_info->sysvec;
684 if (interp != NULL) {
685 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
686 snprintf(path, MAXPATHLEN, "%s%s",
687 brand_info->emul_path, interp);
688 if ((error = elf_load_file(imgp->proc, path, &addr,
689 &imgp->entry_addr)) != 0) {
690 if ((error = elf_load_file(imgp->proc, interp, &addr,
691 &imgp->entry_addr)) != 0) {
692 uprintf("ELF interpreter %s not found\n", path);
701 * Construct auxargs table (used by the fixup routine)
703 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
704 elf_auxargs->execfd = -1;
705 elf_auxargs->phdr = proghdr;
706 elf_auxargs->phent = hdr->e_phentsize;
707 elf_auxargs->phnum = hdr->e_phnum;
708 elf_auxargs->pagesz = PAGE_SIZE;
709 elf_auxargs->base = addr;
710 elf_auxargs->flags = 0;
711 elf_auxargs->entry = entry;
712 elf_auxargs->trace = elf_trace;
714 imgp->auxargs = elf_auxargs;
715 imgp->interpreted = 0;
722 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
724 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
727 pos = *stack_base + (imgp->argc + imgp->envc + 2);
730 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
732 if (args->execfd != -1) {
733 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
735 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
736 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
737 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
738 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
739 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
740 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
741 AUXARGS_ENTRY(pos, AT_BASE, args->base);
742 AUXARGS_ENTRY(pos, AT_NULL, 0);
744 free(imgp->auxargs, M_TEMP);
745 imgp->auxargs = NULL;
748 suword(*stack_base, (long) imgp->argc);
753 * Code for generating ELF core dumps.
756 typedef void (*segment_callback) (vm_map_entry_t, void *);
758 /* Closure for cb_put_phdr(). */
759 struct phdr_closure {
760 Elf_Phdr *phdr; /* Program header to fill in */
761 Elf_Off offset; /* Offset of segment in core file */
764 /* Closure for cb_size_segment(). */
765 struct sseg_closure {
766 int count; /* Count of writable segments. */
767 size_t size; /* Total size of all writable segments. */
778 static void cb_put_phdr (vm_map_entry_t, void *);
779 static void cb_size_segment (vm_map_entry_t, void *);
780 static void cb_fpcount_segment(vm_map_entry_t, void *);
781 static void cb_put_fp(vm_map_entry_t, void *);
784 static void each_segment (struct proc *, segment_callback,
786 static int elf_corehdr (struct proc *, struct file *, struct ucred *,
787 int, void *, size_t);
788 static void elf_puthdr (struct proc *, void *, size_t *,
789 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int);
790 static void elf_putnote (void *, size_t *, const char *, int,
791 const void *, size_t);
793 static void elf_putsigs(struct proc *, void *, int *);
795 static void elf_puttextvp(struct proc *, void *, int *);
796 static void elf_putfiles(struct proc *, void *, int *);
799 extern int osreldate;
802 elf_coredump(struct proc *p, struct vnode *vp, off_t limit)
807 if ((error = falloc(NULL, &fp, NULL)) != 0)
809 fsetcred(fp, p->p_ucred);
811 fp->f_data = (caddr_t)vp;
812 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
814 fp->f_type = DTYPE_VNODE;
815 VOP_UNLOCK(vp, 0, p->p_thread);
817 error = generic_elf_coredump(p, fp, limit);
821 fp->f_ops = &badfileops;
823 fdrop(fp, p->p_thread);
828 generic_elf_coredump(struct proc *p, struct file *fp, off_t limit)
831 struct ucred *cred = p->p_ucred;
833 struct sseg_closure seginfo;
838 printf("can't dump core - null fp\n");
839 /* Size the program segments. */
842 each_segment(p, cb_size_segment, &seginfo, 1);
845 * Calculate the size of the core file header area by making
846 * a dry run of generating it. Nothing is written, but the
847 * size is calculated.
851 elf_puthdr(p, (void *)NULL, &hdrsize,
852 (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
853 (const prpsinfo_t *)NULL, seginfo.count);
855 if (hdrsize + seginfo.size >= limit)
859 * Allocate memory for building the header, fill it up,
862 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
866 error = elf_corehdr(p, fp, cred, seginfo.count, hdr, hdrsize);
868 /* Write the contents of all of the writable segments. */
874 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
876 for (i = 0; i < seginfo.count; i++) {
878 error = fp_pwrite(fp, (caddr_t)php->p_vaddr, php->p_filesz,
882 offset += php->p_filesz;
892 * A callback for each_segment() to write out the segment's
893 * program header entry.
896 cb_put_phdr(vm_map_entry_t entry, void *closure)
898 struct phdr_closure *phc = (struct phdr_closure *)closure;
899 Elf_Phdr *phdr = phc->phdr;
901 phc->offset = round_page(phc->offset);
903 phdr->p_type = PT_LOAD;
904 phdr->p_offset = phc->offset;
905 phdr->p_vaddr = entry->start;
907 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
908 phdr->p_align = PAGE_SIZE;
910 if (entry->protection & VM_PROT_READ)
911 phdr->p_flags |= PF_R;
912 if (entry->protection & VM_PROT_WRITE)
913 phdr->p_flags |= PF_W;
914 if (entry->protection & VM_PROT_EXECUTE)
915 phdr->p_flags |= PF_X;
917 phc->offset += phdr->p_filesz;
922 * A callback for each_writable_segment() to gather information about
923 * the number of segments and their total size.
926 cb_size_segment(vm_map_entry_t entry, void *closure)
928 struct sseg_closure *ssc = (struct sseg_closure *)closure;
931 ssc->size += entry->end - entry->start;
935 * A callback for each_segment() to gather information about
936 * the number of text segments.
939 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
941 int *count = (int *)closure;
942 if (entry->object.vm_object->type == OBJT_VNODE)
947 cb_put_fp(vm_map_entry_t entry, void *closure)
949 struct fp_closure *fpc = (struct fp_closure *)closure;
950 struct vn_hdr *vnh = fpc->vnh;
951 Elf_Phdr *phdr = &vnh->vnh_phdr;
955 if (entry->object.vm_object->type == OBJT_VNODE) {
956 vp = (struct vnode *)entry->object.vm_object->handle;
958 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
959 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
964 phdr->p_type = PT_LOAD;
965 phdr->p_offset = 0; /* not written to core */
966 phdr->p_vaddr = entry->start;
968 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
969 phdr->p_align = PAGE_SIZE;
971 if (entry->protection & VM_PROT_READ)
972 phdr->p_flags |= PF_R;
973 if (entry->protection & VM_PROT_WRITE)
974 phdr->p_flags |= PF_W;
975 if (entry->protection & VM_PROT_EXECUTE)
976 phdr->p_flags |= PF_X;
986 * For each writable segment in the process's memory map, call the given
987 * function with a pointer to the map entry and some arbitrary
988 * caller-supplied data.
991 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
993 vm_map_t map = &p->p_vmspace->vm_map;
994 vm_map_entry_t entry;
996 for (entry = map->header.next; entry != &map->header;
997 entry = entry->next) {
1001 * Don't dump inaccessible mappings, deal with legacy
1004 * Note that read-only segments related to the elf binary
1005 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1006 * need to arbitrarily ignore such segments.
1008 if (elf_legacy_coredump) {
1009 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1012 if (writable && (entry->protection & VM_PROT_ALL) == 0)
1017 * Dont include memory segment in the coredump if
1018 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1019 * madvise(2). Do not dump submaps (i.e. parts of the
1022 if (writable && entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1025 if ((obj = entry->object.vm_object) == NULL)
1028 /* Find the deepest backing object. */
1029 while (obj->backing_object != NULL)
1030 obj = obj->backing_object;
1032 /* Ignore memory-mapped devices and such things. */
1033 if (obj->type != OBJT_DEFAULT &&
1034 obj->type != OBJT_SWAP &&
1035 obj->type != OBJT_VNODE)
1038 (*func)(entry, closure);
1043 * Write the core file header to the file, including padding up to
1044 * the page boundary.
1047 elf_corehdr(struct proc *p, struct file *fp, struct ucred *cred, int numsegs,
1048 void *hdr, size_t hdrsize)
1052 prfpregset_t fpregset;
1057 prfpregset_t *fpregset;
1060 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
1061 status = &tempdata->status;
1062 fpregset = &tempdata->fpregset;
1063 psinfo = &tempdata->psinfo;
1065 /* Gather the information for the header. */
1066 status->pr_version = PRSTATUS_VERSION;
1067 status->pr_statussz = sizeof(prstatus_t);
1068 status->pr_gregsetsz = sizeof(gregset_t);
1069 status->pr_fpregsetsz = sizeof(fpregset_t);
1070 status->pr_osreldate = osreldate;
1071 status->pr_cursig = p->p_sig;
1072 status->pr_pid = p->p_pid;
1073 fill_regs(p, &status->pr_reg);
1075 fill_fpregs(p, fpregset);
1077 psinfo->pr_version = PRPSINFO_VERSION;
1078 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1079 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
1081 /* XXX - We don't fill in the command line arguments properly yet. */
1082 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1084 /* Fill in the header. */
1085 bzero(hdr, hdrsize);
1087 elf_puthdr(p, hdr, &off, status, fpregset, psinfo, numsegs);
1089 free(tempdata, M_TEMP);
1091 /* Write it to the core file. */
1092 return fp_pwrite(fp, hdr, hdrsize, (off_t)0, &nbytes);
1096 elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
1097 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1105 *off += sizeof(Elf_Ehdr);
1108 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1111 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
1113 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1115 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1117 notesz = *off - noteoff;
1119 /* put extra cruft for dumping process state here
1120 * - we really want it be before all the program
1122 * - we just need to update the offset accordingly
1123 * and GDB will be none the wiser.
1125 elf_puttextvp(p, dst, off);
1126 elf_putsigs(p, dst, off);
1127 elf_putfiles(p, dst, off);
1129 /* Align up to a page boundary for the program segments. */
1130 *off = round_page(*off);
1135 struct phdr_closure phc;
1138 * Fill in the ELF header.
1140 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1141 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1142 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1143 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1144 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1145 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1146 ehdr->e_ident[EI_DATA] = ELF_DATA;
1147 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1148 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1149 ehdr->e_ident[EI_ABIVERSION] = 0;
1150 ehdr->e_ident[EI_PAD] = 0;
1151 ehdr->e_type = ET_CORE;
1152 ehdr->e_machine = ELF_ARCH;
1153 ehdr->e_version = EV_CURRENT;
1155 ehdr->e_phoff = phoff;
1157 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1158 ehdr->e_phentsize = sizeof(Elf_Phdr);
1159 ehdr->e_phnum = numsegs + 1;
1160 ehdr->e_shentsize = sizeof(Elf_Shdr);
1162 ehdr->e_shstrndx = SHN_UNDEF;
1165 * Fill in the program header entries.
1167 phdr = (Elf_Phdr *)((char *)dst + phoff);
1169 /* The note segement. */
1170 phdr->p_type = PT_NOTE;
1171 phdr->p_offset = noteoff;
1174 phdr->p_filesz = notesz;
1180 /* All the writable segments from the program. */
1183 each_segment(p, cb_put_phdr, &phc, 1);
1188 elf_putnote(void *dst, size_t *off, const char *name, int type,
1189 const void *desc, size_t descsz)
1193 note.n_namesz = strlen(name) + 1;
1194 note.n_descsz = descsz;
1197 bcopy(¬e, (char *)dst + *off, sizeof note);
1198 *off += sizeof note;
1200 bcopy(name, (char *)dst + *off, note.n_namesz);
1201 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1203 bcopy(desc, (char *)dst + *off, note.n_descsz);
1204 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1209 elf_putsigs(struct proc *p, void *dst, int *off)
1211 struct ckpt_siginfo *csi;
1213 *off += sizeof(struct ckpt_siginfo);
1216 csi = (struct ckpt_siginfo *)((char *)dst + *off);
1218 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1219 bcopy(p->p_procsig, &csi->csi_procsig, sizeof(struct procsig));
1220 bcopy(p->p_procsig->ps_sigacts, &csi->csi_sigacts, sizeof(struct sigacts));
1221 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1222 csi->csi_sigparent = p->p_sigparent;
1223 *off += sizeof(struct ckpt_siginfo);
1227 elf_putfiles(struct proc *p, void *dst, int *off)
1230 struct ckpt_filehdr *cfh;
1231 struct ckpt_fileinfo *cfi;
1235 * the duplicated loop is gross, but it was the only way
1236 * to eliminate uninitialized variable warnings
1239 cfh = (struct ckpt_filehdr *)((char *)dst + *off);
1240 *off += sizeof(struct ckpt_filehdr);
1241 cfh->cfh_nfiles = 0;
1243 * ignore STDIN/STDERR/STDOUT
1245 for (i = 3; i < p->p_fd->fd_nfiles; i++) {
1246 if ((fp = p->p_fd->fd_ofiles[i]) != NULL && fp->f_type == DTYPE_VNODE) {
1248 printf("saving fd: %d\n", i);
1249 cfi = (struct ckpt_fileinfo *)((char *)dst + *off);
1251 cfi->cfi_flags = fp->f_flag;
1252 cfi->cfi_offset = fp->f_offset;
1253 vp = (struct vnode *)fp->f_data;
1254 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1255 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1257 *off += sizeof(struct ckpt_fileinfo);
1261 *off += sizeof(struct ckpt_filehdr);
1262 for (i = 0; i < p->p_fd->fd_nfiles; i++)
1263 if ((fp = p->p_fd->fd_ofiles[i]) != NULL && fp->f_type == DTYPE_VNODE)
1264 *off += sizeof(struct ckpt_fileinfo);
1270 elf_puttextvp(struct proc *p, void *dst, int *off)
1273 struct fp_closure fpc;
1276 each_segment(p, cb_fpcount_segment, &count, 0);
1277 *off += sizeof(struct vn_hdr)*count + sizeof(int);
1280 vn_count = (int *)((char *)dst + *off);
1281 *off += sizeof(int);
1282 fpc.vnh = (struct vn_hdr *)((char *)dst + *off);
1284 each_segment(p, cb_put_fp, &fpc, 0);
1285 *vn_count = fpc.count;
1286 *off += fpc.count*sizeof(struct vn_hdr);
1291 * Tell kern_execve.c about it, with a little help from the linker.
1293 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1294 EXEC_SET(elf, elf_execsw);