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_trace = 0;
103 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
105 static int elf_legacy_coredump = 0;
106 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
107 &elf_legacy_coredump, 0, "");
109 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
111 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
112 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
113 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
115 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
117 Elf_Brandnote __elfN(freebsd_brandnote) = {
118 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
119 .hdr.n_descsz = sizeof(int32_t),
121 .vendor = FREEBSD_ABI_VENDOR,
122 .flags = BN_CAN_FETCH_OSREL
126 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
130 for (i = 0; i < MAX_BRANDS; i++) {
131 if (elf_brand_list[i] == NULL) {
132 elf_brand_list[i] = entry;
142 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
146 for (i = 0; i < MAX_BRANDS; i++) {
147 if (elf_brand_list[i] == entry) {
148 elf_brand_list[i] = NULL;
158 __elfN(brand_inuse)(Elf_Brandinfo *entry)
163 sx_slock(&allproc_lock);
164 FOREACH_PROC_IN_SYSTEM(p) {
165 if (p->p_sysent == entry->sysvec) {
170 sx_sunlock(&allproc_lock);
175 static Elf_Brandinfo *
176 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
179 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
185 * We support four types of branding -- (1) the ELF EI_OSABI field
186 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
187 * branding w/in the ELF header, (3) path of the `interp_path'
188 * field, and (4) the ".note.ABI-tag" ELF section.
191 /* Look for an ".note.ABI-tag" ELF section */
192 for (i = 0; i < MAX_BRANDS; i++) {
193 bi = elf_brand_list[i];
194 if (bi != NULL && hdr->e_machine == bi->machine &&
195 (bi->flags & BI_BRAND_NOTE) != 0) {
196 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
202 /* If the executable has a brand, search for it in the brand list. */
203 for (i = 0; i < MAX_BRANDS; i++) {
204 bi = elf_brand_list[i];
205 if (bi != NULL && hdr->e_machine == bi->machine &&
206 (hdr->e_ident[EI_OSABI] == bi->brand ||
207 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
208 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
212 /* Lacking a known brand, search for a recognized interpreter. */
213 if (interp != NULL) {
214 for (i = 0; i < MAX_BRANDS; i++) {
215 bi = elf_brand_list[i];
216 if (bi != NULL && hdr->e_machine == bi->machine &&
217 strcmp(interp, bi->interp_path) == 0)
222 /* Lacking a recognized interpreter, try the default brand */
223 for (i = 0; i < MAX_BRANDS; i++) {
224 bi = elf_brand_list[i];
225 if (bi != NULL && hdr->e_machine == bi->machine &&
226 __elfN(fallback_brand) == bi->brand)
233 __elfN(check_header)(const Elf_Ehdr *hdr)
239 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
240 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
241 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
242 hdr->e_phentsize != sizeof(Elf_Phdr) ||
243 hdr->e_version != ELF_TARG_VER)
247 * Make sure we have at least one brand for this machine.
250 for (i = 0; i < MAX_BRANDS; i++) {
251 bi = elf_brand_list[i];
252 if (bi != NULL && bi->machine == hdr->e_machine)
262 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
263 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
270 * Create the page if it doesn't exist yet. Ignore errors.
273 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
274 VM_PROT_ALL, VM_PROT_ALL, 0);
278 * Find the page from the underlying object.
281 sf = vm_imgact_map_page(object, offset);
283 return (KERN_FAILURE);
284 off = offset - trunc_page(offset);
285 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
287 vm_imgact_unmap_page(sf);
289 return (KERN_FAILURE);
293 return (KERN_SUCCESS);
297 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
298 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
305 if (start != trunc_page(start)) {
306 rv = __elfN(map_partial)(map, object, offset, start,
307 round_page(start), prot);
310 offset += round_page(start) - start;
311 start = round_page(start);
313 if (end != round_page(end)) {
314 rv = __elfN(map_partial)(map, object, offset +
315 trunc_page(end) - start, trunc_page(end), end, prot);
318 end = trunc_page(end);
321 if (offset & PAGE_MASK) {
323 * The mapping is not page aligned. This means we have
324 * to copy the data. Sigh.
326 rv = vm_map_find(map, NULL, 0, &start, end - start,
327 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
331 return (KERN_SUCCESS);
332 for (; start < end; start += sz) {
333 sf = vm_imgact_map_page(object, offset);
335 return (KERN_FAILURE);
336 off = offset - trunc_page(offset);
338 if (sz > PAGE_SIZE - off)
339 sz = PAGE_SIZE - off;
340 error = copyout((caddr_t)sf_buf_kva(sf) + off,
342 vm_imgact_unmap_page(sf);
344 return (KERN_FAILURE);
350 vm_object_reference(object);
352 rv = vm_map_insert(map, object, offset, start, end,
353 prot, VM_PROT_ALL, cow);
355 if (rv != KERN_SUCCESS)
356 vm_object_deallocate(object);
360 return (KERN_SUCCESS);
365 __elfN(load_section)(struct vmspace *vmspace,
366 vm_object_t object, vm_offset_t offset,
367 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
372 vm_offset_t map_addr;
375 vm_offset_t file_addr;
378 * It's necessary to fail if the filsz + offset taken from the
379 * header is greater than the actual file pager object's size.
380 * If we were to allow this, then the vm_map_find() below would
381 * walk right off the end of the file object and into the ether.
383 * While I'm here, might as well check for something else that
384 * is invalid: filsz cannot be greater than memsz.
386 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
388 uprintf("elf_load_section: truncated ELF file\n");
392 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
393 file_addr = trunc_page_ps(offset, pagesize);
396 * We have two choices. We can either clear the data in the last page
397 * of an oversized mapping, or we can start the anon mapping a page
398 * early and copy the initialized data into that first page. We
399 * choose the second..
402 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
404 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
407 /* cow flags: don't dump readonly sections in core */
408 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
409 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
411 rv = __elfN(map_insert)(&vmspace->vm_map,
413 file_addr, /* file offset */
414 map_addr, /* virtual start */
415 map_addr + map_len,/* virtual end */
418 if (rv != KERN_SUCCESS)
421 /* we can stop now if we've covered it all */
422 if (memsz == filsz) {
429 * We have to get the remaining bit of the file into the first part
430 * of the oversized map segment. This is normally because the .data
431 * segment in the file is extended to provide bss. It's a neat idea
432 * to try and save a page, but it's a pain in the behind to implement.
434 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
435 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
436 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
439 /* This had damn well better be true! */
441 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
442 map_addr + map_len, VM_PROT_ALL, 0);
443 if (rv != KERN_SUCCESS) {
451 sf = vm_imgact_map_page(object, offset + filsz);
455 /* send the page fragment to user space */
456 off = trunc_page_ps(offset + filsz, pagesize) -
457 trunc_page(offset + filsz);
458 error = copyout((caddr_t)sf_buf_kva(sf) + off,
459 (caddr_t)map_addr, copy_len);
460 vm_imgact_unmap_page(sf);
467 * set it to the specified protection.
468 * XXX had better undo the damage from pasting over the cracks here!
470 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
471 round_page(map_addr + map_len), prot, FALSE);
477 * Load the file "file" into memory. It may be either a shared object
480 * The "addr" reference parameter is in/out. On entry, it specifies
481 * the address where a shared object should be loaded. If the file is
482 * an executable, this value is ignored. On exit, "addr" specifies
483 * where the file was actually loaded.
485 * The "entry" reference parameter is out only. On exit, it specifies
486 * the entry point for the loaded file.
489 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
490 u_long *entry, size_t pagesize)
495 struct image_params image_params;
497 const Elf_Ehdr *hdr = NULL;
498 const Elf_Phdr *phdr = NULL;
499 struct nameidata *nd;
500 struct vmspace *vmspace = p->p_vmspace;
502 struct image_params *imgp;
505 u_long base_addr = 0;
506 int vfslocked, error, i, numsegs;
508 if (curthread->td_proc != p)
509 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
511 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
513 attr = &tempdata->attr;
514 imgp = &tempdata->image_params;
517 * Initialize part of the common data
521 imgp->firstpage = NULL;
522 imgp->image_header = NULL;
524 imgp->execlabel = NULL;
527 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
530 if ((error = namei(nd)) != 0) {
534 vfslocked = NDHASGIANT(nd);
535 NDFREE(nd, NDF_ONLY_PNBUF);
536 imgp->vp = nd->ni_vp;
539 * Check permissions, modes, uid, etc on the file, and "open" it.
541 error = exec_check_permissions(imgp);
545 error = exec_map_first_page(imgp);
550 * Also make certain that the interpreter stays the same, so set
551 * its VV_TEXT flag, too.
553 nd->ni_vp->v_vflag |= VV_TEXT;
555 imgp->object = nd->ni_vp->v_object;
557 hdr = (const Elf_Ehdr *)imgp->image_header;
558 if ((error = __elfN(check_header)(hdr)) != 0)
560 if (hdr->e_type == ET_DYN)
562 else if (hdr->e_type == ET_EXEC)
569 /* Only support headers that fit within first page for now */
570 /* (multiplication of two Elf_Half fields will not overflow) */
571 if ((hdr->e_phoff > PAGE_SIZE) ||
572 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
577 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
578 if (!aligned(phdr, Elf_Addr)) {
583 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
584 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
586 if (phdr[i].p_flags & PF_X)
587 prot |= VM_PROT_EXECUTE;
588 if (phdr[i].p_flags & PF_W)
589 prot |= VM_PROT_WRITE;
590 if (phdr[i].p_flags & PF_R)
591 prot |= VM_PROT_READ;
593 if ((error = __elfN(load_section)(vmspace,
594 imgp->object, phdr[i].p_offset,
595 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
596 phdr[i].p_memsz, phdr[i].p_filesz, prot,
600 * Establish the base address if this is the
604 base_addr = trunc_page(phdr[i].p_vaddr +
610 *entry = (unsigned long)hdr->e_entry + rbase;
614 exec_unmap_first_page(imgp);
619 VFS_UNLOCK_GIANT(vfslocked);
620 free(tempdata, M_TEMP);
626 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
628 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
629 const Elf_Phdr *phdr;
630 Elf_Auxargs *elf_auxargs;
631 struct vmspace *vmspace;
633 u_long text_size = 0, data_size = 0, total_size = 0;
634 u_long text_addr = 0, data_addr = 0;
635 u_long seg_size, seg_addr;
636 u_long addr, entry = 0, proghdr = 0;
639 const char *interp = NULL, *newinterp = NULL;
640 Elf_Brandinfo *brand_info;
642 struct thread *td = curthread;
643 struct sysentvec *sv;
646 * Do we have a valid ELF header ?
648 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
649 * if particular brand doesn't support it.
651 if (__elfN(check_header)(hdr) != 0 ||
652 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
656 * From here on down, we return an errno, not -1, as we've
657 * detected an ELF file.
660 if ((hdr->e_phoff > PAGE_SIZE) ||
661 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
662 /* Only support headers in first page for now */
665 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
666 if (!aligned(phdr, Elf_Addr))
668 for (i = 0; i < hdr->e_phnum; i++) {
669 if (phdr[i].p_type == PT_INTERP) {
670 /* Path to interpreter */
671 if (phdr[i].p_filesz > MAXPATHLEN ||
672 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
674 interp = imgp->image_header + phdr[i].p_offset;
679 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
680 if (brand_info == NULL) {
681 uprintf("ELF binary type \"%u\" not known.\n",
682 hdr->e_ident[EI_OSABI]);
685 if (hdr->e_type == ET_DYN &&
686 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
688 sv = brand_info->sysvec;
689 if (interp != NULL && brand_info->interp_newpath != NULL)
690 newinterp = brand_info->interp_newpath;
693 * Avoid a possible deadlock if the current address space is destroyed
694 * and that address space maps the locked vnode. In the common case,
695 * the locked vnode's v_usecount is decremented but remains greater
696 * than zero. Consequently, the vnode lock is not needed by vrele().
697 * However, in cases where the vnode lock is external, such as nullfs,
698 * v_usecount may become zero.
700 VOP_UNLOCK(imgp->vp, 0, td);
702 error = exec_new_vmspace(imgp, sv);
703 imgp->proc->p_sysent = sv;
705 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
709 vmspace = imgp->proc->p_vmspace;
711 for (i = 0; i < hdr->e_phnum; i++) {
712 switch (phdr[i].p_type) {
713 case PT_LOAD: /* Loadable segment */
715 if (phdr[i].p_flags & PF_X)
716 prot |= VM_PROT_EXECUTE;
717 if (phdr[i].p_flags & PF_W)
718 prot |= VM_PROT_WRITE;
719 if (phdr[i].p_flags & PF_R)
720 prot |= VM_PROT_READ;
722 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
724 * Some x86 binaries assume read == executable,
725 * notably the M3 runtime and therefore cvsup
727 if (prot & VM_PROT_READ)
728 prot |= VM_PROT_EXECUTE;
731 if ((error = __elfN(load_section)(vmspace,
732 imgp->object, phdr[i].p_offset,
733 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
734 phdr[i].p_memsz, phdr[i].p_filesz, prot,
735 sv->sv_pagesize)) != 0)
739 * If this segment contains the program headers,
740 * remember their virtual address for the AT_PHDR
741 * aux entry. Static binaries don't usually include
744 if (phdr[i].p_offset == 0 &&
745 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
747 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
749 seg_addr = trunc_page(phdr[i].p_vaddr);
750 seg_size = round_page(phdr[i].p_memsz +
751 phdr[i].p_vaddr - seg_addr);
754 * Is this .text or .data? We can't use
755 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
756 * alpha terribly and possibly does other bad
757 * things so we stick to the old way of figuring
758 * it out: If the segment contains the program
759 * entry point, it's a text segment, otherwise it
762 * Note that obreak() assumes that data_addr +
763 * data_size == end of data load area, and the ELF
764 * file format expects segments to be sorted by
765 * address. If multiple data segments exist, the
766 * last one will be used.
768 if (hdr->e_entry >= phdr[i].p_vaddr &&
769 hdr->e_entry < (phdr[i].p_vaddr +
771 text_size = seg_size;
772 text_addr = seg_addr;
773 entry = (u_long)hdr->e_entry;
775 data_size = seg_size;
776 data_addr = seg_addr;
778 total_size += seg_size;
780 case PT_PHDR: /* Program header table info */
781 proghdr = phdr[i].p_vaddr;
788 if (data_addr == 0 && data_size == 0) {
789 data_addr = text_addr;
790 data_size = text_size;
794 * Check limits. It should be safe to check the
795 * limits after loading the segments since we do
796 * not actually fault in all the segments pages.
798 PROC_LOCK(imgp->proc);
799 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
800 text_size > maxtsiz ||
801 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
802 PROC_UNLOCK(imgp->proc);
806 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
807 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
808 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
809 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
812 * We load the dynamic linker where a userland call
813 * to mmap(0, ...) would put it. The rationale behind this
814 * calculation is that it leaves room for the heap to grow to
815 * its maximum allowed size.
817 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
818 lim_max(imgp->proc, RLIMIT_DATA));
819 PROC_UNLOCK(imgp->proc);
821 imgp->entry_addr = entry;
823 if (interp != NULL) {
824 int have_interp = FALSE;
825 VOP_UNLOCK(imgp->vp, 0, td);
826 if (brand_info->emul_path != NULL &&
827 brand_info->emul_path[0] != '\0') {
828 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
829 snprintf(path, MAXPATHLEN, "%s%s",
830 brand_info->emul_path, interp);
831 error = __elfN(load_file)(imgp->proc, path, &addr,
832 &imgp->entry_addr, sv->sv_pagesize);
837 if (!have_interp && newinterp != NULL) {
838 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
839 &imgp->entry_addr, sv->sv_pagesize);
844 error = __elfN(load_file)(imgp->proc, interp, &addr,
845 &imgp->entry_addr, sv->sv_pagesize);
847 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
849 uprintf("ELF interpreter %s not found\n", interp);
856 * Construct auxargs table (used by the fixup routine)
858 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
859 elf_auxargs->execfd = -1;
860 elf_auxargs->phdr = proghdr;
861 elf_auxargs->phent = hdr->e_phentsize;
862 elf_auxargs->phnum = hdr->e_phnum;
863 elf_auxargs->pagesz = PAGE_SIZE;
864 elf_auxargs->base = addr;
865 elf_auxargs->flags = 0;
866 elf_auxargs->entry = entry;
867 elf_auxargs->trace = elf_trace;
869 imgp->auxargs = elf_auxargs;
870 imgp->interpreted = 0;
871 imgp->proc->p_osrel = osrel;
876 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
879 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
881 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
885 base = (Elf_Addr *)*stack_base;
886 pos = base + (imgp->args->argc + imgp->args->envc + 2);
889 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
891 if (args->execfd != -1) {
892 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
894 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
895 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
896 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
897 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
898 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
899 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
900 AUXARGS_ENTRY(pos, AT_BASE, args->base);
901 AUXARGS_ENTRY(pos, AT_NULL, 0);
903 free(imgp->auxargs, M_TEMP);
904 imgp->auxargs = NULL;
907 suword(base, (long)imgp->args->argc);
908 *stack_base = (register_t *)base;
913 * Code for generating ELF core dumps.
916 typedef void (*segment_callback)(vm_map_entry_t, void *);
918 /* Closure for cb_put_phdr(). */
919 struct phdr_closure {
920 Elf_Phdr *phdr; /* Program header to fill in */
921 Elf_Off offset; /* Offset of segment in core file */
924 /* Closure for cb_size_segment(). */
925 struct sseg_closure {
926 int count; /* Count of writable segments. */
927 size_t size; /* Total size of all writable segments. */
930 static void cb_put_phdr(vm_map_entry_t, void *);
931 static void cb_size_segment(vm_map_entry_t, void *);
932 static void each_writable_segment(struct thread *, segment_callback, void *);
933 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
934 int, void *, size_t);
935 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
936 static void __elfN(putnote)(void *, size_t *, const char *, int,
937 const void *, size_t);
940 __elfN(coredump)(td, vp, limit)
945 struct ucred *cred = td->td_ucred;
947 struct sseg_closure seginfo;
951 /* Size the program segments. */
954 each_writable_segment(td, cb_size_segment, &seginfo);
957 * Calculate the size of the core file header area by making
958 * a dry run of generating it. Nothing is written, but the
959 * size is calculated.
962 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
964 if (hdrsize + seginfo.size >= limit)
968 * Allocate memory for building the header, fill it up,
971 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
975 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
977 /* Write the contents of all of the writable segments. */
983 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
985 for (i = 0; i < seginfo.count; i++) {
986 error = vn_rdwr_inchunks(UIO_WRITE, vp,
987 (caddr_t)(uintptr_t)php->p_vaddr,
988 php->p_filesz, offset, UIO_USERSPACE,
989 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
990 curthread); /* XXXKSE */
993 offset += php->p_filesz;
1003 * A callback for each_writable_segment() to write out the segment's
1004 * program header entry.
1007 cb_put_phdr(entry, closure)
1008 vm_map_entry_t entry;
1011 struct phdr_closure *phc = (struct phdr_closure *)closure;
1012 Elf_Phdr *phdr = phc->phdr;
1014 phc->offset = round_page(phc->offset);
1016 phdr->p_type = PT_LOAD;
1017 phdr->p_offset = phc->offset;
1018 phdr->p_vaddr = entry->start;
1020 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1021 phdr->p_align = PAGE_SIZE;
1023 if (entry->protection & VM_PROT_READ)
1024 phdr->p_flags |= PF_R;
1025 if (entry->protection & VM_PROT_WRITE)
1026 phdr->p_flags |= PF_W;
1027 if (entry->protection & VM_PROT_EXECUTE)
1028 phdr->p_flags |= PF_X;
1030 phc->offset += phdr->p_filesz;
1035 * A callback for each_writable_segment() to gather information about
1036 * the number of segments and their total size.
1039 cb_size_segment(entry, closure)
1040 vm_map_entry_t entry;
1043 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1046 ssc->size += entry->end - entry->start;
1050 * For each writable segment in the process's memory map, call the given
1051 * function with a pointer to the map entry and some arbitrary
1052 * caller-supplied data.
1055 each_writable_segment(td, func, closure)
1057 segment_callback func;
1060 struct proc *p = td->td_proc;
1061 vm_map_t map = &p->p_vmspace->vm_map;
1062 vm_map_entry_t entry;
1063 vm_object_t backing_object, object;
1064 boolean_t ignore_entry;
1066 vm_map_lock_read(map);
1067 for (entry = map->header.next; entry != &map->header;
1068 entry = entry->next) {
1070 * Don't dump inaccessible mappings, deal with legacy
1073 * Note that read-only segments related to the elf binary
1074 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1075 * need to arbitrarily ignore such segments.
1077 if (elf_legacy_coredump) {
1078 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1081 if ((entry->protection & VM_PROT_ALL) == 0)
1086 * Dont include memory segment in the coredump if
1087 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1088 * madvise(2). Do not dump submaps (i.e. parts of the
1091 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1094 if ((object = entry->object.vm_object) == NULL)
1097 /* Ignore memory-mapped devices and such things. */
1098 VM_OBJECT_LOCK(object);
1099 while ((backing_object = object->backing_object) != NULL) {
1100 VM_OBJECT_LOCK(backing_object);
1101 VM_OBJECT_UNLOCK(object);
1102 object = backing_object;
1104 ignore_entry = object->type != OBJT_DEFAULT &&
1105 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1106 VM_OBJECT_UNLOCK(object);
1110 (*func)(entry, closure);
1112 vm_map_unlock_read(map);
1116 * Write the core file header to the file, including padding up to
1117 * the page boundary.
1120 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1130 /* Fill in the header. */
1131 bzero(hdr, hdrsize);
1133 __elfN(puthdr)(td, hdr, &off, numsegs);
1135 /* Write it to the core file. */
1136 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1137 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1141 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1142 typedef struct prstatus32 elf_prstatus_t;
1143 typedef struct prpsinfo32 elf_prpsinfo_t;
1144 typedef struct fpreg32 elf_prfpregset_t;
1145 typedef struct fpreg32 elf_fpregset_t;
1146 typedef struct reg32 elf_gregset_t;
1148 typedef prstatus_t elf_prstatus_t;
1149 typedef prpsinfo_t elf_prpsinfo_t;
1150 typedef prfpregset_t elf_prfpregset_t;
1151 typedef prfpregset_t elf_fpregset_t;
1152 typedef gregset_t elf_gregset_t;
1156 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1159 elf_prstatus_t status;
1160 elf_prfpregset_t fpregset;
1161 elf_prpsinfo_t psinfo;
1163 elf_prstatus_t *status;
1164 elf_prfpregset_t *fpregset;
1165 elf_prpsinfo_t *psinfo;
1168 size_t ehoff, noteoff, notesz, phoff;
1173 *off += sizeof(Elf_Ehdr);
1176 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1180 * Don't allocate space for the notes if we're just calculating
1181 * the size of the header. We also don't collect the data.
1184 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1185 status = &tempdata->status;
1186 fpregset = &tempdata->fpregset;
1187 psinfo = &tempdata->psinfo;
1196 psinfo->pr_version = PRPSINFO_VERSION;
1197 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1198 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1200 * XXX - We don't fill in the command line arguments properly
1203 strlcpy(psinfo->pr_psargs, p->p_comm,
1204 sizeof(psinfo->pr_psargs));
1206 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1210 * To have the debugger select the right thread (LWP) as the initial
1211 * thread, we dump the state of the thread passed to us in td first.
1212 * This is the thread that causes the core dump and thus likely to
1213 * be the right thread one wants to have selected in the debugger.
1216 while (thr != NULL) {
1218 status->pr_version = PRSTATUS_VERSION;
1219 status->pr_statussz = sizeof(elf_prstatus_t);
1220 status->pr_gregsetsz = sizeof(elf_gregset_t);
1221 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1222 status->pr_osreldate = osreldate;
1223 status->pr_cursig = p->p_sig;
1224 status->pr_pid = thr->td_tid;
1225 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1226 fill_regs32(thr, &status->pr_reg);
1227 fill_fpregs32(thr, fpregset);
1229 fill_regs(thr, &status->pr_reg);
1230 fill_fpregs(thr, fpregset);
1233 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1235 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1238 * Allow for MD specific notes, as well as any MD
1239 * specific preparations for writing MI notes.
1241 __elfN(dump_thread)(thr, dst, off);
1243 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1244 TAILQ_NEXT(thr, td_plist);
1246 thr = TAILQ_NEXT(thr, td_plist);
1249 notesz = *off - noteoff;
1252 free(tempdata, M_TEMP);
1254 /* Align up to a page boundary for the program segments. */
1255 *off = round_page(*off);
1260 struct phdr_closure phc;
1263 * Fill in the ELF header.
1265 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1266 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1267 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1268 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1269 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1270 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1271 ehdr->e_ident[EI_DATA] = ELF_DATA;
1272 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1273 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1274 ehdr->e_ident[EI_ABIVERSION] = 0;
1275 ehdr->e_ident[EI_PAD] = 0;
1276 ehdr->e_type = ET_CORE;
1277 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1278 ehdr->e_machine = EM_386;
1280 ehdr->e_machine = ELF_ARCH;
1282 ehdr->e_version = EV_CURRENT;
1284 ehdr->e_phoff = phoff;
1286 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1287 ehdr->e_phentsize = sizeof(Elf_Phdr);
1288 ehdr->e_phnum = numsegs + 1;
1289 ehdr->e_shentsize = sizeof(Elf_Shdr);
1291 ehdr->e_shstrndx = SHN_UNDEF;
1294 * Fill in the program header entries.
1296 phdr = (Elf_Phdr *)((char *)dst + phoff);
1298 /* The note segement. */
1299 phdr->p_type = PT_NOTE;
1300 phdr->p_offset = noteoff;
1303 phdr->p_filesz = notesz;
1309 /* All the writable segments from the program. */
1312 each_writable_segment(td, cb_put_phdr, &phc);
1317 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1318 const void *desc, size_t descsz)
1322 note.n_namesz = strlen(name) + 1;
1323 note.n_descsz = descsz;
1326 bcopy(¬e, (char *)dst + *off, sizeof note);
1327 *off += sizeof note;
1329 bcopy(name, (char *)dst + *off, note.n_namesz);
1330 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1332 bcopy(desc, (char *)dst + *off, note.n_descsz);
1333 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1337 * Try to find the appropriate ABI-note section for checknote,
1338 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1339 * first page of the image is searched, the same as for headers.
1342 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1345 const Elf_Note *note, *note0, *note_end;
1346 const Elf_Phdr *phdr, *pnote;
1347 const Elf_Ehdr *hdr;
1348 const char *note_name;
1352 hdr = (const Elf_Ehdr *)imgp->image_header;
1353 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1355 for (i = 0; i < hdr->e_phnum; i++) {
1356 if (phdr[i].p_type == PT_NOTE) {
1362 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1363 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
1366 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1367 note_end = (const Elf_Note *)(imgp->image_header +
1368 pnote->p_offset + pnote->p_filesz);
1369 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1370 if (!aligned(note, Elf32_Addr))
1372 if (note->n_namesz != checknote->hdr.n_namesz ||
1373 note->n_descsz != checknote->hdr.n_descsz ||
1374 note->n_type != checknote->hdr.n_type)
1376 note_name = (const char *)(note + 1);
1377 if (strncmp(checknote->vendor, note_name,
1378 checknote->hdr.n_namesz) != 0)
1382 * Fetch the osreldate for binary
1383 * from the ELF OSABI-note if necessary.
1385 if ((checknote->flags & BN_CAN_FETCH_OSREL) != 0 &&
1387 *osrel = *(const int32_t *) (note_name +
1388 roundup2(checknote->hdr.n_namesz,
1389 sizeof(Elf32_Addr)));
1393 note = (const Elf_Note *)((const char *)(note + 1) +
1394 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1395 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1402 * Tell kern_execve.c about it, with a little help from the linker.
1404 static struct execsw __elfN(execsw) = {
1405 __CONCAT(exec_, __elfN(imgact)),
1406 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1408 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));