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 proc *p,
86 struct vmspace *vmspace, struct vnode *vp, vm_object_t object,
87 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
88 vm_prot_t prot, size_t pagesize);
89 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
91 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
94 int __elfN(fallback_brand) = -1;
95 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
96 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
97 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
98 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
99 &__elfN(fallback_brand));
101 int __elfN(can_exec_dyn) = 0;
102 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
103 can_exec_dyn, CTLFLAG_RW, &__elfN(can_exec_dyn), 0,
104 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " can exec shared libraries");
106 static int elf_trace = 0;
107 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
109 static int elf_legacy_coredump = 0;
110 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
111 &elf_legacy_coredump, 0, "");
113 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
116 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
120 for (i = 0; i < MAX_BRANDS; i++) {
121 if (elf_brand_list[i] == NULL) {
122 elf_brand_list[i] = entry;
132 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
136 for (i = 0; i < MAX_BRANDS; i++) {
137 if (elf_brand_list[i] == entry) {
138 elf_brand_list[i] = NULL;
148 __elfN(brand_inuse)(Elf_Brandinfo *entry)
153 sx_slock(&allproc_lock);
154 LIST_FOREACH(p, &allproc, p_list) {
155 if (p->p_sysent == entry->sysvec) {
160 sx_sunlock(&allproc_lock);
165 static Elf_Brandinfo *
166 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
172 * We support three types of branding -- (1) the ELF EI_OSABI field
173 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
174 * branding w/in the ELF header, and (3) path of the `interp_path'
175 * field. We should also look for an ".note.ABI-tag" ELF section now
176 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
179 /* If the executable has a brand, search for it in the brand list. */
180 for (i = 0; i < MAX_BRANDS; i++) {
181 bi = elf_brand_list[i];
182 if (bi != NULL && hdr->e_machine == bi->machine &&
183 (hdr->e_ident[EI_OSABI] == bi->brand ||
184 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
185 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
189 /* Lacking a known brand, search for a recognized interpreter. */
190 if (interp != NULL) {
191 for (i = 0; i < MAX_BRANDS; i++) {
192 bi = elf_brand_list[i];
193 if (bi != NULL && hdr->e_machine == bi->machine &&
194 strcmp(interp, bi->interp_path) == 0)
199 /* Lacking a recognized interpreter, try the default brand */
200 for (i = 0; i < MAX_BRANDS; i++) {
201 bi = elf_brand_list[i];
202 if (bi != NULL && hdr->e_machine == bi->machine &&
203 __elfN(fallback_brand) == bi->brand)
210 __elfN(check_header)(const Elf_Ehdr *hdr)
216 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
217 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
218 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
219 hdr->e_phentsize != sizeof(Elf_Phdr) ||
220 hdr->e_version != ELF_TARG_VER)
224 * Make sure we have at least one brand for this machine.
227 for (i = 0; i < MAX_BRANDS; i++) {
228 bi = elf_brand_list[i];
229 if (bi != NULL && bi->machine == hdr->e_machine)
239 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
240 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
248 * Create the page if it doesn't exist yet. Ignore errors.
251 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max,
256 * Find the page from the underlying object.
259 sf = vm_imgact_map_page(object, offset);
261 return (KERN_FAILURE);
262 off = offset - trunc_page(offset);
263 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
265 vm_imgact_unmap_page(sf);
267 return (KERN_FAILURE);
271 return (KERN_SUCCESS);
275 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
276 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
277 vm_prot_t max, int cow)
284 if (start != trunc_page(start)) {
285 rv = __elfN(map_partial)(map, object, offset, start,
286 round_page(start), prot, max);
289 offset += round_page(start) - start;
290 start = round_page(start);
292 if (end != round_page(end)) {
293 rv = __elfN(map_partial)(map, object, offset +
294 trunc_page(end) - start, trunc_page(end), end, prot, max);
297 end = trunc_page(end);
300 if (offset & PAGE_MASK) {
302 * The mapping is not page aligned. This means we have
303 * to copy the data. Sigh.
305 rv = vm_map_find(map, 0, 0, &start, end - start,
306 FALSE, prot, max, 0);
310 return (KERN_SUCCESS);
311 for (; start < end; start += sz) {
312 sf = vm_imgact_map_page(object, offset);
314 return (KERN_FAILURE);
315 off = offset - trunc_page(offset);
317 if (sz > PAGE_SIZE - off)
318 sz = PAGE_SIZE - off;
319 error = copyout((caddr_t)sf_buf_kva(sf) + off,
321 vm_imgact_unmap_page(sf);
323 return (KERN_FAILURE);
330 rv = vm_map_insert(map, object, offset, start, end,
336 return (KERN_SUCCESS);
341 __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
342 struct vnode *vp, vm_object_t object, vm_offset_t offset,
343 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
348 vm_offset_t map_addr;
351 vm_offset_t file_addr;
356 * It's necessary to fail if the filsz + offset taken from the
357 * header is greater than the actual file pager object's size.
358 * If we were to allow this, then the vm_map_find() below would
359 * walk right off the end of the file object and into the ether.
361 * While I'm here, might as well check for something else that
362 * is invalid: filsz cannot be greater than memsz.
364 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
366 uprintf("elf_load_section: truncated ELF file\n");
370 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
371 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
373 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
374 file_addr = trunc_page_ps(offset, pagesize);
377 * We have two choices. We can either clear the data in the last page
378 * of an oversized mapping, or we can start the anon mapping a page
379 * early and copy the initialized data into that first page. We
380 * choose the second..
383 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
385 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
388 vm_object_reference(object);
390 /* cow flags: don't dump readonly sections in core */
391 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
392 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
394 rv = __elfN(map_insert)(&vmspace->vm_map,
396 file_addr, /* file offset */
397 map_addr, /* virtual start */
398 map_addr + map_len,/* virtual end */
402 if (rv != KERN_SUCCESS) {
403 vm_object_deallocate(object);
407 /* we can stop now if we've covered it all */
408 if (memsz == filsz) {
415 * We have to get the remaining bit of the file into the first part
416 * of the oversized map segment. This is normally because the .data
417 * segment in the file is extended to provide bss. It's a neat idea
418 * to try and save a page, but it's a pain in the behind to implement.
420 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
421 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
422 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
425 /* This had damn well better be true! */
427 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
428 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0);
429 if (rv != KERN_SUCCESS) {
437 sf = vm_imgact_map_page(object, offset + filsz);
441 /* send the page fragment to user space */
442 off = trunc_page_ps(offset + filsz, pagesize) -
443 trunc_page(offset + filsz);
444 error = copyout((caddr_t)sf_buf_kva(sf) + off,
445 (caddr_t)map_addr, copy_len);
446 vm_imgact_unmap_page(sf);
453 * set it to the specified protection.
454 * XXX had better undo the damage from pasting over the cracks here!
456 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
457 round_page(map_addr + map_len), prot, FALSE);
463 * Load the file "file" into memory. It may be either a shared object
466 * The "addr" reference parameter is in/out. On entry, it specifies
467 * the address where a shared object should be loaded. If the file is
468 * an executable, this value is ignored. On exit, "addr" specifies
469 * where the file was actually loaded.
471 * The "entry" reference parameter is out only. On exit, it specifies
472 * the entry point for the loaded file.
475 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
476 u_long *entry, size_t pagesize)
481 struct image_params image_params;
483 const Elf_Ehdr *hdr = NULL;
484 const Elf_Phdr *phdr = NULL;
485 struct nameidata *nd;
486 struct vmspace *vmspace = p->p_vmspace;
488 struct image_params *imgp;
491 u_long base_addr = 0;
492 int vfslocked, error, i, numsegs;
494 if (curthread->td_proc != p)
495 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
497 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
499 attr = &tempdata->attr;
500 imgp = &tempdata->image_params;
503 * Initialize part of the common data
507 imgp->firstpage = NULL;
508 imgp->image_header = NULL;
510 imgp->execlabel = NULL;
513 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
516 if ((error = namei(nd)) != 0) {
520 vfslocked = NDHASGIANT(nd);
521 NDFREE(nd, NDF_ONLY_PNBUF);
522 imgp->vp = nd->ni_vp;
525 * Check permissions, modes, uid, etc on the file, and "open" it.
527 error = exec_check_permissions(imgp);
529 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
533 error = exec_map_first_page(imgp);
535 * Also make certain that the interpreter stays the same, so set
536 * its VV_TEXT flag, too.
539 nd->ni_vp->v_vflag |= VV_TEXT;
541 imgp->object = nd->ni_vp->v_object;
542 vm_object_reference(imgp->object);
544 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
548 hdr = (const Elf_Ehdr *)imgp->image_header;
549 if ((error = __elfN(check_header)(hdr)) != 0)
551 if (hdr->e_type == ET_DYN)
553 else if (hdr->e_type == ET_EXEC)
560 /* Only support headers that fit within first page for now */
561 /* (multiplication of two Elf_Half fields will not overflow) */
562 if ((hdr->e_phoff > PAGE_SIZE) ||
563 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
568 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
570 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
571 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
573 if (phdr[i].p_flags & PF_X)
574 prot |= VM_PROT_EXECUTE;
575 if (phdr[i].p_flags & PF_W)
576 prot |= VM_PROT_WRITE;
577 if (phdr[i].p_flags & PF_R)
578 prot |= VM_PROT_READ;
580 if ((error = __elfN(load_section)(p, vmspace,
581 nd->ni_vp, imgp->object, phdr[i].p_offset,
582 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
583 phdr[i].p_memsz, phdr[i].p_filesz, prot,
587 * Establish the base address if this is the
591 base_addr = trunc_page(phdr[i].p_vaddr +
597 *entry = (unsigned long)hdr->e_entry + rbase;
601 exec_unmap_first_page(imgp);
603 vm_object_deallocate(imgp->object);
608 VFS_UNLOCK_GIANT(vfslocked);
609 free(tempdata, M_TEMP);
615 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
617 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
618 const Elf_Phdr *phdr;
619 Elf_Auxargs *elf_auxargs = NULL;
620 struct vmspace *vmspace;
622 u_long text_size = 0, data_size = 0, total_size = 0;
623 u_long text_addr = 0, data_addr = 0;
624 u_long seg_size, seg_addr;
625 u_long addr, entry = 0, proghdr = 0;
627 const char *interp = NULL;
628 Elf_Brandinfo *brand_info;
630 struct thread *td = curthread;
631 struct sysentvec *sv;
634 * Do we have a valid ELF header ?
636 if (__elfN(check_header)(hdr) != 0 || (hdr->e_type != ET_EXEC
637 && (!__elfN(can_exec_dyn) || hdr->e_type != ET_DYN)))
641 * From here on down, we return an errno, not -1, as we've
642 * detected an ELF file.
645 if ((hdr->e_phoff > PAGE_SIZE) ||
646 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
647 /* Only support headers in first page for now */
650 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
653 * From this point on, we may have resources that need to be freed.
656 VOP_UNLOCK(imgp->vp, 0, td);
658 for (i = 0; i < hdr->e_phnum; i++) {
659 switch (phdr[i].p_type) {
660 case PT_INTERP: /* Path to interpreter */
661 if (phdr[i].p_filesz > MAXPATHLEN ||
662 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
666 interp = imgp->image_header + phdr[i].p_offset;
673 brand_info = __elfN(get_brandinfo)(hdr, interp);
674 if (brand_info == NULL) {
675 uprintf("ELF binary type \"%u\" not known.\n",
676 hdr->e_ident[EI_OSABI]);
680 sv = brand_info->sysvec;
681 if (interp != NULL && brand_info->interp_newpath != NULL)
682 interp = brand_info->interp_newpath;
684 exec_new_vmspace(imgp, sv);
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)(imgp->proc, vmspace,
709 imgp->vp, 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;
765 if (data_addr == 0 && data_size == 0) {
766 data_addr = text_addr;
767 data_size = text_size;
771 * Check limits. It should be safe to check the
772 * limits after loading the segments since we do
773 * not actually fault in all the segments pages.
775 PROC_LOCK(imgp->proc);
776 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
777 text_size > maxtsiz ||
778 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
779 PROC_UNLOCK(imgp->proc);
784 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
785 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
786 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
787 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
790 * We load the dynamic linker where a userland call
791 * to mmap(0, ...) would put it. The rationale behind this
792 * calculation is that it leaves room for the heap to grow to
793 * its maximum allowed size.
795 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
796 lim_max(imgp->proc, RLIMIT_DATA));
797 PROC_UNLOCK(imgp->proc);
799 imgp->entry_addr = entry;
801 imgp->proc->p_sysent = sv;
802 if (interp != NULL && brand_info->emul_path != NULL &&
803 brand_info->emul_path[0] != '\0') {
804 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
805 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path,
807 error = __elfN(load_file)(imgp->proc, path, &addr,
808 &imgp->entry_addr, sv->sv_pagesize);
813 if (interp != NULL) {
814 error = __elfN(load_file)(imgp->proc, interp, &addr,
815 &imgp->entry_addr, sv->sv_pagesize);
817 uprintf("ELF interpreter %s not found\n", interp);
823 * Construct auxargs table (used by the fixup routine)
825 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
826 elf_auxargs->execfd = -1;
827 elf_auxargs->phdr = proghdr;
828 elf_auxargs->phent = hdr->e_phentsize;
829 elf_auxargs->phnum = hdr->e_phnum;
830 elf_auxargs->pagesz = PAGE_SIZE;
831 elf_auxargs->base = addr;
832 elf_auxargs->flags = 0;
833 elf_auxargs->entry = entry;
834 elf_auxargs->trace = elf_trace;
836 imgp->auxargs = elf_auxargs;
837 imgp->interpreted = 0;
840 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
844 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
847 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
849 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
853 base = (Elf_Addr *)*stack_base;
854 pos = base + (imgp->args->argc + imgp->args->envc + 2);
857 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
859 if (args->execfd != -1) {
860 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
862 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
863 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
864 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
865 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
866 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
867 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
868 AUXARGS_ENTRY(pos, AT_BASE, args->base);
869 AUXARGS_ENTRY(pos, AT_NULL, 0);
871 free(imgp->auxargs, M_TEMP);
872 imgp->auxargs = NULL;
875 suword(base, (long)imgp->args->argc);
876 *stack_base = (register_t *)base;
881 * Code for generating ELF core dumps.
884 typedef void (*segment_callback)(vm_map_entry_t, void *);
886 /* Closure for cb_put_phdr(). */
887 struct phdr_closure {
888 Elf_Phdr *phdr; /* Program header to fill in */
889 Elf_Off offset; /* Offset of segment in core file */
892 /* Closure for cb_size_segment(). */
893 struct sseg_closure {
894 int count; /* Count of writable segments. */
895 size_t size; /* Total size of all writable segments. */
898 static void cb_put_phdr(vm_map_entry_t, void *);
899 static void cb_size_segment(vm_map_entry_t, void *);
900 static void each_writable_segment(struct thread *, segment_callback, void *);
901 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
902 int, void *, size_t);
903 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
904 static void __elfN(putnote)(void *, size_t *, const char *, int,
905 const void *, size_t);
907 extern int osreldate;
910 __elfN(coredump)(td, vp, limit)
915 struct ucred *cred = td->td_ucred;
917 struct sseg_closure seginfo;
921 /* Size the program segments. */
924 each_writable_segment(td, cb_size_segment, &seginfo);
927 * Calculate the size of the core file header area by making
928 * a dry run of generating it. Nothing is written, but the
929 * size is calculated.
932 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
934 if (hdrsize + seginfo.size >= limit)
938 * Allocate memory for building the header, fill it up,
941 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
945 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
947 /* Write the contents of all of the writable segments. */
953 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
955 for (i = 0; i < seginfo.count; i++) {
956 error = vn_rdwr_inchunks(UIO_WRITE, vp,
957 (caddr_t)(uintptr_t)php->p_vaddr,
958 php->p_filesz, offset, UIO_USERSPACE,
959 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
960 curthread); /* XXXKSE */
963 offset += php->p_filesz;
973 * A callback for each_writable_segment() to write out the segment's
974 * program header entry.
977 cb_put_phdr(entry, closure)
978 vm_map_entry_t entry;
981 struct phdr_closure *phc = (struct phdr_closure *)closure;
982 Elf_Phdr *phdr = phc->phdr;
984 phc->offset = round_page(phc->offset);
986 phdr->p_type = PT_LOAD;
987 phdr->p_offset = phc->offset;
988 phdr->p_vaddr = entry->start;
990 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
991 phdr->p_align = PAGE_SIZE;
993 if (entry->protection & VM_PROT_READ)
994 phdr->p_flags |= PF_R;
995 if (entry->protection & VM_PROT_WRITE)
996 phdr->p_flags |= PF_W;
997 if (entry->protection & VM_PROT_EXECUTE)
998 phdr->p_flags |= PF_X;
1000 phc->offset += phdr->p_filesz;
1005 * A callback for each_writable_segment() to gather information about
1006 * the number of segments and their total size.
1009 cb_size_segment(entry, closure)
1010 vm_map_entry_t entry;
1013 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1016 ssc->size += entry->end - entry->start;
1020 * For each writable segment in the process's memory map, call the given
1021 * function with a pointer to the map entry and some arbitrary
1022 * caller-supplied data.
1025 each_writable_segment(td, func, closure)
1027 segment_callback func;
1030 struct proc *p = td->td_proc;
1031 vm_map_t map = &p->p_vmspace->vm_map;
1032 vm_map_entry_t entry;
1034 for (entry = map->header.next; entry != &map->header;
1035 entry = entry->next) {
1039 * Don't dump inaccessible mappings, deal with legacy
1042 * Note that read-only segments related to the elf binary
1043 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1044 * need to arbitrarily ignore such segments.
1046 if (elf_legacy_coredump) {
1047 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1050 if ((entry->protection & VM_PROT_ALL) == 0)
1055 * Dont include memory segment in the coredump if
1056 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1057 * madvise(2). Do not dump submaps (i.e. parts of the
1060 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1063 if ((obj = entry->object.vm_object) == NULL)
1066 /* Find the deepest backing object. */
1067 while (obj->backing_object != NULL)
1068 obj = obj->backing_object;
1070 /* Ignore memory-mapped devices and such things. */
1071 if (obj->type != OBJT_DEFAULT &&
1072 obj->type != OBJT_SWAP &&
1073 obj->type != OBJT_VNODE)
1076 (*func)(entry, closure);
1081 * Write the core file header to the file, including padding up to
1082 * the page boundary.
1085 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1095 /* Fill in the header. */
1096 bzero(hdr, hdrsize);
1098 __elfN(puthdr)(td, hdr, &off, numsegs);
1100 /* Write it to the core file. */
1101 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1102 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1106 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1107 typedef struct prstatus32 elf_prstatus_t;
1108 typedef struct prpsinfo32 elf_prpsinfo_t;
1109 typedef struct fpreg32 elf_prfpregset_t;
1110 typedef struct fpreg32 elf_fpregset_t;
1111 typedef struct reg32 elf_gregset_t;
1113 typedef prstatus_t elf_prstatus_t;
1114 typedef prpsinfo_t elf_prpsinfo_t;
1115 typedef prfpregset_t elf_prfpregset_t;
1116 typedef prfpregset_t elf_fpregset_t;
1117 typedef gregset_t elf_gregset_t;
1121 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1124 elf_prstatus_t status;
1125 elf_prfpregset_t fpregset;
1126 elf_prpsinfo_t psinfo;
1128 elf_prstatus_t *status;
1129 elf_prfpregset_t *fpregset;
1130 elf_prpsinfo_t *psinfo;
1133 size_t ehoff, noteoff, notesz, phoff;
1138 *off += sizeof(Elf_Ehdr);
1141 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1145 * Don't allocate space for the notes if we're just calculating
1146 * the size of the header. We also don't collect the data.
1149 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1150 status = &tempdata->status;
1151 fpregset = &tempdata->fpregset;
1152 psinfo = &tempdata->psinfo;
1161 psinfo->pr_version = PRPSINFO_VERSION;
1162 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1163 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1165 * XXX - We don't fill in the command line arguments properly
1168 strlcpy(psinfo->pr_psargs, p->p_comm,
1169 sizeof(psinfo->pr_psargs));
1171 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1175 * To have the debugger select the right thread (LWP) as the initial
1176 * thread, we dump the state of the thread passed to us in td first.
1177 * This is the thread that causes the core dump and thus likely to
1178 * be the right thread one wants to have selected in the debugger.
1181 while (thr != NULL) {
1183 status->pr_version = PRSTATUS_VERSION;
1184 status->pr_statussz = sizeof(elf_prstatus_t);
1185 status->pr_gregsetsz = sizeof(elf_gregset_t);
1186 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1187 status->pr_osreldate = osreldate;
1188 status->pr_cursig = p->p_sig;
1189 status->pr_pid = thr->td_tid;
1190 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1191 fill_regs32(thr, &status->pr_reg);
1192 fill_fpregs32(thr, fpregset);
1194 fill_regs(thr, &status->pr_reg);
1195 fill_fpregs(thr, fpregset);
1198 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1200 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1203 * Allow for MD specific notes, as well as any MD
1204 * specific preparations for writing MI notes.
1206 __elfN(dump_thread)(thr, dst, off);
1208 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1209 TAILQ_NEXT(thr, td_plist);
1211 thr = TAILQ_NEXT(thr, td_plist);
1214 notesz = *off - noteoff;
1217 free(tempdata, M_TEMP);
1219 /* Align up to a page boundary for the program segments. */
1220 *off = round_page(*off);
1225 struct phdr_closure phc;
1228 * Fill in the ELF header.
1230 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1231 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1232 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1233 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1234 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1235 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1236 ehdr->e_ident[EI_DATA] = ELF_DATA;
1237 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1238 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1239 ehdr->e_ident[EI_ABIVERSION] = 0;
1240 ehdr->e_ident[EI_PAD] = 0;
1241 ehdr->e_type = ET_CORE;
1242 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1243 ehdr->e_machine = EM_386;
1245 ehdr->e_machine = ELF_ARCH;
1247 ehdr->e_version = EV_CURRENT;
1249 ehdr->e_phoff = phoff;
1251 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1252 ehdr->e_phentsize = sizeof(Elf_Phdr);
1253 ehdr->e_phnum = numsegs + 1;
1254 ehdr->e_shentsize = sizeof(Elf_Shdr);
1256 ehdr->e_shstrndx = SHN_UNDEF;
1259 * Fill in the program header entries.
1261 phdr = (Elf_Phdr *)((char *)dst + phoff);
1263 /* The note segement. */
1264 phdr->p_type = PT_NOTE;
1265 phdr->p_offset = noteoff;
1268 phdr->p_filesz = notesz;
1274 /* All the writable segments from the program. */
1277 each_writable_segment(td, cb_put_phdr, &phc);
1282 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1283 const void *desc, size_t descsz)
1287 note.n_namesz = strlen(name) + 1;
1288 note.n_descsz = descsz;
1291 bcopy(¬e, (char *)dst + *off, sizeof note);
1292 *off += sizeof note;
1294 bcopy(name, (char *)dst + *off, note.n_namesz);
1295 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1297 bcopy(desc, (char *)dst + *off, note.n_descsz);
1298 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1302 * Tell kern_execve.c about it, with a little help from the linker.
1304 static struct execsw __elfN(execsw) = {
1305 __CONCAT(exec_, __elfN(imgact)),
1306 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1308 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));