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(check_note)(struct image_params *imgp,
90 Elf_Brandnote *checknote, int32_t *osrel);
92 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
95 int __elfN(fallback_brand) = -1;
96 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
97 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
98 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
99 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
100 &__elfN(fallback_brand));
102 static int elf_legacy_coredump = 0;
103 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
104 &elf_legacy_coredump, 0, "");
106 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
108 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
109 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
110 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
112 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
114 Elf_Brandnote __elfN(freebsd_brandnote) = {
115 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
116 .hdr.n_descsz = sizeof(int32_t),
118 .vendor = FREEBSD_ABI_VENDOR,
119 .flags = BN_CAN_FETCH_OSREL
123 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
127 for (i = 0; i < MAX_BRANDS; i++) {
128 if (elf_brand_list[i] == NULL) {
129 elf_brand_list[i] = entry;
139 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
143 for (i = 0; i < MAX_BRANDS; i++) {
144 if (elf_brand_list[i] == entry) {
145 elf_brand_list[i] = NULL;
155 __elfN(brand_inuse)(Elf_Brandinfo *entry)
160 sx_slock(&allproc_lock);
161 FOREACH_PROC_IN_SYSTEM(p) {
162 if (p->p_sysent == entry->sysvec) {
167 sx_sunlock(&allproc_lock);
172 static Elf_Brandinfo *
173 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
176 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
182 * We support four types of branding -- (1) the ELF EI_OSABI field
183 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
184 * branding w/in the ELF header, (3) path of the `interp_path'
185 * field, and (4) the ".note.ABI-tag" ELF section.
188 /* Look for an ".note.ABI-tag" ELF section */
189 for (i = 0; i < MAX_BRANDS; i++) {
190 bi = elf_brand_list[i];
191 if (bi != NULL && hdr->e_machine == bi->machine &&
192 bi->brand_note != NULL) {
193 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
199 /* If the executable has a brand, search for it in the brand list. */
200 for (i = 0; i < MAX_BRANDS; i++) {
201 bi = elf_brand_list[i];
202 if (bi != NULL && hdr->e_machine == bi->machine &&
203 (hdr->e_ident[EI_OSABI] == bi->brand ||
204 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
205 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
209 /* Lacking a known brand, search for a recognized interpreter. */
210 if (interp != NULL) {
211 for (i = 0; i < MAX_BRANDS; i++) {
212 bi = elf_brand_list[i];
213 if (bi != NULL && hdr->e_machine == bi->machine &&
214 strcmp(interp, bi->interp_path) == 0)
219 /* Lacking a recognized interpreter, try the default brand */
220 for (i = 0; i < MAX_BRANDS; i++) {
221 bi = elf_brand_list[i];
222 if (bi != NULL && hdr->e_machine == bi->machine &&
223 __elfN(fallback_brand) == bi->brand)
230 __elfN(check_header)(const Elf_Ehdr *hdr)
236 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
237 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
238 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
239 hdr->e_phentsize != sizeof(Elf_Phdr) ||
240 hdr->e_version != ELF_TARG_VER)
244 * Make sure we have at least one brand for this machine.
247 for (i = 0; i < MAX_BRANDS; i++) {
248 bi = elf_brand_list[i];
249 if (bi != NULL && bi->machine == hdr->e_machine)
259 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
260 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
267 * Create the page if it doesn't exist yet. Ignore errors.
270 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
271 VM_PROT_ALL, VM_PROT_ALL, 0);
275 * Find the page from the underlying object.
278 sf = vm_imgact_map_page(object, offset);
280 return (KERN_FAILURE);
281 off = offset - trunc_page(offset);
282 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
284 vm_imgact_unmap_page(sf);
286 return (KERN_FAILURE);
290 return (KERN_SUCCESS);
294 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
295 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
302 if (start != trunc_page(start)) {
303 rv = __elfN(map_partial)(map, object, offset, start,
304 round_page(start), prot);
307 offset += round_page(start) - start;
308 start = round_page(start);
310 if (end != round_page(end)) {
311 rv = __elfN(map_partial)(map, object, offset +
312 trunc_page(end) - start, trunc_page(end), end, prot);
315 end = trunc_page(end);
318 if (offset & PAGE_MASK) {
320 * The mapping is not page aligned. This means we have
321 * to copy the data. Sigh.
323 rv = vm_map_find(map, NULL, 0, &start, end - start,
324 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
328 return (KERN_SUCCESS);
329 for (; start < end; start += sz) {
330 sf = vm_imgact_map_page(object, offset);
332 return (KERN_FAILURE);
333 off = offset - trunc_page(offset);
335 if (sz > PAGE_SIZE - off)
336 sz = PAGE_SIZE - off;
337 error = copyout((caddr_t)sf_buf_kva(sf) + off,
339 vm_imgact_unmap_page(sf);
341 return (KERN_FAILURE);
347 vm_object_reference(object);
349 rv = vm_map_insert(map, object, offset, start, end,
350 prot, VM_PROT_ALL, cow);
352 if (rv != KERN_SUCCESS)
353 vm_object_deallocate(object);
357 return (KERN_SUCCESS);
362 __elfN(load_section)(struct vmspace *vmspace,
363 vm_object_t object, vm_offset_t offset,
364 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
369 vm_offset_t map_addr;
372 vm_offset_t file_addr;
375 * It's necessary to fail if the filsz + offset taken from the
376 * header is greater than the actual file pager object's size.
377 * If we were to allow this, then the vm_map_find() below would
378 * walk right off the end of the file object and into the ether.
380 * While I'm here, might as well check for something else that
381 * is invalid: filsz cannot be greater than memsz.
383 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
385 uprintf("elf_load_section: truncated ELF file\n");
389 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
390 file_addr = trunc_page_ps(offset, pagesize);
393 * We have two choices. We can either clear the data in the last page
394 * of an oversized mapping, or we can start the anon mapping a page
395 * early and copy the initialized data into that first page. We
396 * choose the second..
399 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
401 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
404 /* cow flags: don't dump readonly sections in core */
405 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
406 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
408 rv = __elfN(map_insert)(&vmspace->vm_map,
410 file_addr, /* file offset */
411 map_addr, /* virtual start */
412 map_addr + map_len,/* virtual end */
415 if (rv != KERN_SUCCESS)
418 /* we can stop now if we've covered it all */
419 if (memsz == filsz) {
426 * We have to get the remaining bit of the file into the first part
427 * of the oversized map segment. This is normally because the .data
428 * segment in the file is extended to provide bss. It's a neat idea
429 * to try and save a page, but it's a pain in the behind to implement.
431 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
432 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
433 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
436 /* This had damn well better be true! */
438 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
439 map_addr + map_len, VM_PROT_ALL, 0);
440 if (rv != KERN_SUCCESS) {
448 sf = vm_imgact_map_page(object, offset + filsz);
452 /* send the page fragment to user space */
453 off = trunc_page_ps(offset + filsz, pagesize) -
454 trunc_page(offset + filsz);
455 error = copyout((caddr_t)sf_buf_kva(sf) + off,
456 (caddr_t)map_addr, copy_len);
457 vm_imgact_unmap_page(sf);
464 * set it to the specified protection.
465 * XXX had better undo the damage from pasting over the cracks here!
467 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
468 round_page(map_addr + map_len), prot, FALSE);
474 * Load the file "file" into memory. It may be either a shared object
477 * The "addr" reference parameter is in/out. On entry, it specifies
478 * the address where a shared object should be loaded. If the file is
479 * an executable, this value is ignored. On exit, "addr" specifies
480 * where the file was actually loaded.
482 * The "entry" reference parameter is out only. On exit, it specifies
483 * the entry point for the loaded file.
486 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
487 u_long *entry, size_t pagesize)
492 struct image_params image_params;
494 const Elf_Ehdr *hdr = NULL;
495 const Elf_Phdr *phdr = NULL;
496 struct nameidata *nd;
497 struct vmspace *vmspace = p->p_vmspace;
499 struct image_params *imgp;
502 u_long base_addr = 0;
503 int vfslocked, error, i, numsegs;
505 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
507 attr = &tempdata->attr;
508 imgp = &tempdata->image_params;
511 * Initialize part of the common data
515 imgp->firstpage = NULL;
516 imgp->image_header = NULL;
518 imgp->execlabel = NULL;
520 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
523 if ((error = namei(nd)) != 0) {
527 vfslocked = NDHASGIANT(nd);
528 NDFREE(nd, NDF_ONLY_PNBUF);
529 imgp->vp = nd->ni_vp;
532 * Check permissions, modes, uid, etc on the file, and "open" it.
534 error = exec_check_permissions(imgp);
538 error = exec_map_first_page(imgp);
543 * Also make certain that the interpreter stays the same, so set
544 * its VV_TEXT flag, too.
546 nd->ni_vp->v_vflag |= VV_TEXT;
548 imgp->object = nd->ni_vp->v_object;
550 hdr = (const Elf_Ehdr *)imgp->image_header;
551 if ((error = __elfN(check_header)(hdr)) != 0)
553 if (hdr->e_type == ET_DYN)
555 else if (hdr->e_type == ET_EXEC)
562 /* Only support headers that fit within first page for now */
563 /* (multiplication of two Elf_Half fields will not overflow) */
564 if ((hdr->e_phoff > PAGE_SIZE) ||
565 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
570 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
571 if (!aligned(phdr, Elf_Addr)) {
576 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
577 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
579 if (phdr[i].p_flags & PF_X)
580 prot |= VM_PROT_EXECUTE;
581 if (phdr[i].p_flags & PF_W)
582 prot |= VM_PROT_WRITE;
583 if (phdr[i].p_flags & PF_R)
584 prot |= VM_PROT_READ;
586 if ((error = __elfN(load_section)(vmspace,
587 imgp->object, phdr[i].p_offset,
588 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
589 phdr[i].p_memsz, phdr[i].p_filesz, prot,
593 * Establish the base address if this is the
597 base_addr = trunc_page(phdr[i].p_vaddr +
603 *entry = (unsigned long)hdr->e_entry + rbase;
607 exec_unmap_first_page(imgp);
612 VFS_UNLOCK_GIANT(vfslocked);
613 free(tempdata, M_TEMP);
619 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
621 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
622 const Elf_Phdr *phdr;
623 Elf_Auxargs *elf_auxargs;
624 struct vmspace *vmspace;
626 u_long text_size = 0, data_size = 0, total_size = 0;
627 u_long text_addr = 0, data_addr = 0;
628 u_long seg_size, seg_addr;
629 u_long addr, entry = 0, proghdr = 0;
632 const char *interp = NULL, *newinterp = NULL;
633 Elf_Brandinfo *brand_info;
635 struct sysentvec *sv;
638 * Do we have a valid ELF header ?
640 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
641 * if particular brand doesn't support it.
643 if (__elfN(check_header)(hdr) != 0 ||
644 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
648 * From here on down, we return an errno, not -1, as we've
649 * detected an ELF file.
652 if ((hdr->e_phoff > PAGE_SIZE) ||
653 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
654 /* Only support headers in first page for now */
657 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
658 if (!aligned(phdr, Elf_Addr))
660 for (i = 0; i < hdr->e_phnum; i++) {
661 if (phdr[i].p_type == PT_INTERP) {
662 /* Path to interpreter */
663 if (phdr[i].p_filesz > MAXPATHLEN ||
664 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
666 interp = imgp->image_header + phdr[i].p_offset;
671 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
672 if (brand_info == NULL) {
673 uprintf("ELF binary type \"%u\" not known.\n",
674 hdr->e_ident[EI_OSABI]);
677 if (hdr->e_type == ET_DYN &&
678 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
680 sv = brand_info->sysvec;
681 if (interp != NULL && brand_info->interp_newpath != NULL)
682 newinterp = brand_info->interp_newpath;
685 * Avoid a possible deadlock if the current address space is destroyed
686 * and that address space maps the locked vnode. In the common case,
687 * the locked vnode's v_usecount is decremented but remains greater
688 * than zero. Consequently, the vnode lock is not needed by vrele().
689 * However, in cases where the vnode lock is external, such as nullfs,
690 * v_usecount may become zero.
692 VOP_UNLOCK(imgp->vp, 0);
694 error = exec_new_vmspace(imgp, sv);
695 imgp->proc->p_sysent = sv;
697 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
701 vmspace = imgp->proc->p_vmspace;
703 for (i = 0; i < hdr->e_phnum; i++) {
704 switch (phdr[i].p_type) {
705 case PT_LOAD: /* Loadable segment */
707 if (phdr[i].p_flags & PF_X)
708 prot |= VM_PROT_EXECUTE;
709 if (phdr[i].p_flags & PF_W)
710 prot |= VM_PROT_WRITE;
711 if (phdr[i].p_flags & PF_R)
712 prot |= VM_PROT_READ;
714 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
716 * Some x86 binaries assume read == executable,
717 * notably the M3 runtime and therefore cvsup
719 if (prot & VM_PROT_READ)
720 prot |= VM_PROT_EXECUTE;
723 if ((error = __elfN(load_section)(vmspace,
724 imgp->object, phdr[i].p_offset,
725 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
726 phdr[i].p_memsz, phdr[i].p_filesz, prot,
727 sv->sv_pagesize)) != 0)
731 * If this segment contains the program headers,
732 * remember their virtual address for the AT_PHDR
733 * aux entry. Static binaries don't usually include
736 if (phdr[i].p_offset == 0 &&
737 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
739 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
741 seg_addr = trunc_page(phdr[i].p_vaddr);
742 seg_size = round_page(phdr[i].p_memsz +
743 phdr[i].p_vaddr - seg_addr);
746 * Is this .text or .data? We can't use
747 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
748 * alpha terribly and possibly does other bad
749 * things so we stick to the old way of figuring
750 * it out: If the segment contains the program
751 * entry point, it's a text segment, otherwise it
754 * Note that obreak() assumes that data_addr +
755 * data_size == end of data load area, and the ELF
756 * file format expects segments to be sorted by
757 * address. If multiple data segments exist, the
758 * last one will be used.
760 if (hdr->e_entry >= phdr[i].p_vaddr &&
761 hdr->e_entry < (phdr[i].p_vaddr +
763 text_size = seg_size;
764 text_addr = seg_addr;
765 entry = (u_long)hdr->e_entry;
767 data_size = seg_size;
768 data_addr = seg_addr;
770 total_size += seg_size;
772 case PT_PHDR: /* Program header table info */
773 proghdr = phdr[i].p_vaddr;
780 if (data_addr == 0 && data_size == 0) {
781 data_addr = text_addr;
782 data_size = text_size;
786 * Check limits. It should be safe to check the
787 * limits after loading the segments since we do
788 * not actually fault in all the segments pages.
790 PROC_LOCK(imgp->proc);
791 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
792 text_size > maxtsiz ||
793 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
794 PROC_UNLOCK(imgp->proc);
798 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
799 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
800 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
801 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
804 * We load the dynamic linker where a userland call
805 * to mmap(0, ...) would put it. The rationale behind this
806 * calculation is that it leaves room for the heap to grow to
807 * its maximum allowed size.
809 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
810 lim_max(imgp->proc, RLIMIT_DATA));
811 PROC_UNLOCK(imgp->proc);
813 imgp->entry_addr = entry;
815 if (interp != NULL) {
816 int have_interp = FALSE;
817 VOP_UNLOCK(imgp->vp, 0);
818 if (brand_info->emul_path != NULL &&
819 brand_info->emul_path[0] != '\0') {
820 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
821 snprintf(path, MAXPATHLEN, "%s%s",
822 brand_info->emul_path, interp);
823 error = __elfN(load_file)(imgp->proc, path, &addr,
824 &imgp->entry_addr, sv->sv_pagesize);
829 if (!have_interp && newinterp != NULL) {
830 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
831 &imgp->entry_addr, sv->sv_pagesize);
836 error = __elfN(load_file)(imgp->proc, interp, &addr,
837 &imgp->entry_addr, sv->sv_pagesize);
839 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
841 uprintf("ELF interpreter %s not found\n", interp);
848 * Construct auxargs table (used by the fixup routine)
850 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
851 elf_auxargs->execfd = -1;
852 elf_auxargs->phdr = proghdr;
853 elf_auxargs->phent = hdr->e_phentsize;
854 elf_auxargs->phnum = hdr->e_phnum;
855 elf_auxargs->pagesz = PAGE_SIZE;
856 elf_auxargs->base = addr;
857 elf_auxargs->flags = 0;
858 elf_auxargs->entry = entry;
860 imgp->auxargs = elf_auxargs;
861 imgp->interpreted = 0;
862 imgp->proc->p_osrel = osrel;
867 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
870 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
872 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
876 base = (Elf_Addr *)*stack_base;
877 pos = base + (imgp->args->argc + imgp->args->envc + 2);
879 if (args->execfd != -1)
880 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
881 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
882 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
883 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
884 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
885 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
886 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
887 AUXARGS_ENTRY(pos, AT_BASE, args->base);
888 AUXARGS_ENTRY(pos, AT_NULL, 0);
890 free(imgp->auxargs, M_TEMP);
891 imgp->auxargs = NULL;
894 suword(base, (long)imgp->args->argc);
895 *stack_base = (register_t *)base;
900 * Code for generating ELF core dumps.
903 typedef void (*segment_callback)(vm_map_entry_t, void *);
905 /* Closure for cb_put_phdr(). */
906 struct phdr_closure {
907 Elf_Phdr *phdr; /* Program header to fill in */
908 Elf_Off offset; /* Offset of segment in core file */
911 /* Closure for cb_size_segment(). */
912 struct sseg_closure {
913 int count; /* Count of writable segments. */
914 size_t size; /* Total size of all writable segments. */
917 static void cb_put_phdr(vm_map_entry_t, void *);
918 static void cb_size_segment(vm_map_entry_t, void *);
919 static void each_writable_segment(struct thread *, segment_callback, void *);
920 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
921 int, void *, size_t);
922 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
923 static void __elfN(putnote)(void *, size_t *, const char *, int,
924 const void *, size_t);
927 __elfN(coredump)(td, vp, limit)
932 struct ucred *cred = td->td_ucred;
934 struct sseg_closure seginfo;
938 /* Size the program segments. */
941 each_writable_segment(td, cb_size_segment, &seginfo);
944 * Calculate the size of the core file header area by making
945 * a dry run of generating it. Nothing is written, but the
946 * size is calculated.
949 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
951 if (hdrsize + seginfo.size >= limit)
955 * Allocate memory for building the header, fill it up,
958 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
962 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
964 /* Write the contents of all of the writable segments. */
970 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
972 for (i = 0; i < seginfo.count; i++) {
973 error = vn_rdwr_inchunks(UIO_WRITE, vp,
974 (caddr_t)(uintptr_t)php->p_vaddr,
975 php->p_filesz, offset, UIO_USERSPACE,
976 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
980 offset += php->p_filesz;
990 * A callback for each_writable_segment() to write out the segment's
991 * program header entry.
994 cb_put_phdr(entry, closure)
995 vm_map_entry_t entry;
998 struct phdr_closure *phc = (struct phdr_closure *)closure;
999 Elf_Phdr *phdr = phc->phdr;
1001 phc->offset = round_page(phc->offset);
1003 phdr->p_type = PT_LOAD;
1004 phdr->p_offset = phc->offset;
1005 phdr->p_vaddr = entry->start;
1007 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1008 phdr->p_align = PAGE_SIZE;
1010 if (entry->protection & VM_PROT_READ)
1011 phdr->p_flags |= PF_R;
1012 if (entry->protection & VM_PROT_WRITE)
1013 phdr->p_flags |= PF_W;
1014 if (entry->protection & VM_PROT_EXECUTE)
1015 phdr->p_flags |= PF_X;
1017 phc->offset += phdr->p_filesz;
1022 * A callback for each_writable_segment() to gather information about
1023 * the number of segments and their total size.
1026 cb_size_segment(entry, closure)
1027 vm_map_entry_t entry;
1030 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1033 ssc->size += entry->end - entry->start;
1037 * For each writable segment in the process's memory map, call the given
1038 * function with a pointer to the map entry and some arbitrary
1039 * caller-supplied data.
1042 each_writable_segment(td, func, closure)
1044 segment_callback func;
1047 struct proc *p = td->td_proc;
1048 vm_map_t map = &p->p_vmspace->vm_map;
1049 vm_map_entry_t entry;
1050 vm_object_t backing_object, object;
1051 boolean_t ignore_entry;
1053 vm_map_lock_read(map);
1054 for (entry = map->header.next; entry != &map->header;
1055 entry = entry->next) {
1057 * Don't dump inaccessible mappings, deal with legacy
1060 * Note that read-only segments related to the elf binary
1061 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1062 * need to arbitrarily ignore such segments.
1064 if (elf_legacy_coredump) {
1065 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1068 if ((entry->protection & VM_PROT_ALL) == 0)
1073 * Dont include memory segment in the coredump if
1074 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1075 * madvise(2). Do not dump submaps (i.e. parts of the
1078 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1081 if ((object = entry->object.vm_object) == NULL)
1084 /* Ignore memory-mapped devices and such things. */
1085 VM_OBJECT_LOCK(object);
1086 while ((backing_object = object->backing_object) != NULL) {
1087 VM_OBJECT_LOCK(backing_object);
1088 VM_OBJECT_UNLOCK(object);
1089 object = backing_object;
1091 ignore_entry = object->type != OBJT_DEFAULT &&
1092 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1093 VM_OBJECT_UNLOCK(object);
1097 (*func)(entry, closure);
1099 vm_map_unlock_read(map);
1103 * Write the core file header to the file, including padding up to
1104 * the page boundary.
1107 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1117 /* Fill in the header. */
1118 bzero(hdr, hdrsize);
1120 __elfN(puthdr)(td, hdr, &off, numsegs);
1122 /* Write it to the core file. */
1123 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1124 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1128 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1129 typedef struct prstatus32 elf_prstatus_t;
1130 typedef struct prpsinfo32 elf_prpsinfo_t;
1131 typedef struct fpreg32 elf_prfpregset_t;
1132 typedef struct fpreg32 elf_fpregset_t;
1133 typedef struct reg32 elf_gregset_t;
1135 typedef prstatus_t elf_prstatus_t;
1136 typedef prpsinfo_t elf_prpsinfo_t;
1137 typedef prfpregset_t elf_prfpregset_t;
1138 typedef prfpregset_t elf_fpregset_t;
1139 typedef gregset_t elf_gregset_t;
1143 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1146 elf_prstatus_t status;
1147 elf_prfpregset_t fpregset;
1148 elf_prpsinfo_t psinfo;
1150 elf_prstatus_t *status;
1151 elf_prfpregset_t *fpregset;
1152 elf_prpsinfo_t *psinfo;
1155 size_t ehoff, noteoff, notesz, phoff;
1160 *off += sizeof(Elf_Ehdr);
1163 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1167 * Don't allocate space for the notes if we're just calculating
1168 * the size of the header. We also don't collect the data.
1171 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1172 status = &tempdata->status;
1173 fpregset = &tempdata->fpregset;
1174 psinfo = &tempdata->psinfo;
1183 psinfo->pr_version = PRPSINFO_VERSION;
1184 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1185 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1187 * XXX - We don't fill in the command line arguments properly
1190 strlcpy(psinfo->pr_psargs, p->p_comm,
1191 sizeof(psinfo->pr_psargs));
1193 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1197 * To have the debugger select the right thread (LWP) as the initial
1198 * thread, we dump the state of the thread passed to us in td first.
1199 * This is the thread that causes the core dump and thus likely to
1200 * be the right thread one wants to have selected in the debugger.
1203 while (thr != NULL) {
1205 status->pr_version = PRSTATUS_VERSION;
1206 status->pr_statussz = sizeof(elf_prstatus_t);
1207 status->pr_gregsetsz = sizeof(elf_gregset_t);
1208 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1209 status->pr_osreldate = osreldate;
1210 status->pr_cursig = p->p_sig;
1211 status->pr_pid = thr->td_tid;
1212 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1213 fill_regs32(thr, &status->pr_reg);
1214 fill_fpregs32(thr, fpregset);
1216 fill_regs(thr, &status->pr_reg);
1217 fill_fpregs(thr, fpregset);
1220 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1222 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1225 * Allow for MD specific notes, as well as any MD
1226 * specific preparations for writing MI notes.
1228 __elfN(dump_thread)(thr, dst, off);
1230 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1231 TAILQ_NEXT(thr, td_plist);
1233 thr = TAILQ_NEXT(thr, td_plist);
1236 notesz = *off - noteoff;
1239 free(tempdata, M_TEMP);
1241 /* Align up to a page boundary for the program segments. */
1242 *off = round_page(*off);
1247 struct phdr_closure phc;
1250 * Fill in the ELF header.
1252 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1253 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1254 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1255 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1256 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1257 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1258 ehdr->e_ident[EI_DATA] = ELF_DATA;
1259 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1260 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1261 ehdr->e_ident[EI_ABIVERSION] = 0;
1262 ehdr->e_ident[EI_PAD] = 0;
1263 ehdr->e_type = ET_CORE;
1264 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1265 ehdr->e_machine = EM_386;
1267 ehdr->e_machine = ELF_ARCH;
1269 ehdr->e_version = EV_CURRENT;
1271 ehdr->e_phoff = phoff;
1273 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1274 ehdr->e_phentsize = sizeof(Elf_Phdr);
1275 ehdr->e_phnum = numsegs + 1;
1276 ehdr->e_shentsize = sizeof(Elf_Shdr);
1278 ehdr->e_shstrndx = SHN_UNDEF;
1281 * Fill in the program header entries.
1283 phdr = (Elf_Phdr *)((char *)dst + phoff);
1285 /* The note segement. */
1286 phdr->p_type = PT_NOTE;
1287 phdr->p_offset = noteoff;
1290 phdr->p_filesz = notesz;
1296 /* All the writable segments from the program. */
1299 each_writable_segment(td, cb_put_phdr, &phc);
1304 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1305 const void *desc, size_t descsz)
1309 note.n_namesz = strlen(name) + 1;
1310 note.n_descsz = descsz;
1313 bcopy(¬e, (char *)dst + *off, sizeof note);
1314 *off += sizeof note;
1316 bcopy(name, (char *)dst + *off, note.n_namesz);
1317 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1319 bcopy(desc, (char *)dst + *off, note.n_descsz);
1320 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1324 * Try to find the appropriate ABI-note section for checknote,
1325 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1326 * first page of the image is searched, the same as for headers.
1329 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1332 const Elf_Note *note, *note_end;
1333 const Elf_Phdr *phdr, *pnote;
1334 const Elf_Ehdr *hdr;
1335 const char *note_name;
1339 hdr = (const Elf_Ehdr *)imgp->image_header;
1340 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1342 for (i = 0; i < hdr->e_phnum; i++) {
1343 if (phdr[i].p_type == PT_NOTE) {
1349 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1350 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
1353 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1354 if (!aligned(note, Elf32_Addr))
1356 note_end = (const Elf_Note *)(imgp->image_header +
1357 pnote->p_offset + pnote->p_filesz);
1358 while (note < note_end) {
1359 if (note->n_namesz != checknote->hdr.n_namesz ||
1360 note->n_descsz != checknote->hdr.n_descsz ||
1361 note->n_type != checknote->hdr.n_type)
1363 note_name = (const char *)(note + 1);
1364 if (strncmp(checknote->vendor, note_name,
1365 checknote->hdr.n_namesz) != 0)
1369 * Fetch the osreldate for binary
1370 * from the ELF OSABI-note if necessary.
1372 if ((checknote->flags & BN_CAN_FETCH_OSREL) != 0 &&
1374 *osrel = *(const int32_t *) (note_name +
1375 roundup2(checknote->hdr.n_namesz,
1376 sizeof(Elf32_Addr)));
1380 note = (const Elf_Note *)((const char *)(note + 1) +
1381 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1382 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1389 * Tell kern_execve.c about it, with a little help from the linker.
1391 static struct execsw __elfN(execsw) = {
1392 __CONCAT(exec_, __elfN(imgact)),
1393 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1395 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));