2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_compat.h"
36 #include <sys/param.h>
38 #include <sys/fcntl.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/pioctl.h>
50 #include <sys/procfs.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sf_buf.h>
53 #include <sys/systm.h>
54 #include <sys/signalvar.h>
57 #include <sys/syscall.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/vnode.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_param.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
70 #include <machine/elf.h>
71 #include <machine/md_var.h>
73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
74 #include <machine/fpu.h>
75 #include <compat/ia32/ia32_reg.h>
78 #define OLD_EI_BRAND 8
80 static int __elfN(check_header)(const Elf_Ehdr *hdr);
81 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
84 u_long *entry, size_t pagesize);
85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
86 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
87 vm_prot_t prot, size_t pagesize);
88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
93 int __elfN(fallback_brand) = -1;
94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
95 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
96 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
98 &__elfN(fallback_brand));
100 static int elf_trace = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
103 static int elf_legacy_coredump = 0;
104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
105 &elf_legacy_coredump, 0, "");
107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
109 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
110 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
111 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
114 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
118 for (i = 0; i < MAX_BRANDS; i++) {
119 if (elf_brand_list[i] == NULL) {
120 elf_brand_list[i] = entry;
130 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
134 for (i = 0; i < MAX_BRANDS; i++) {
135 if (elf_brand_list[i] == entry) {
136 elf_brand_list[i] = NULL;
146 __elfN(brand_inuse)(Elf_Brandinfo *entry)
151 sx_slock(&allproc_lock);
152 FOREACH_PROC_IN_SYSTEM(p) {
153 if (p->p_sysent == entry->sysvec) {
158 sx_sunlock(&allproc_lock);
163 static Elf_Brandinfo *
164 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
170 * We support three types of branding -- (1) the ELF EI_OSABI field
171 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
172 * branding w/in the ELF header, and (3) path of the `interp_path'
173 * field. We should also look for an ".note.ABI-tag" ELF section now
174 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
177 /* If the executable has a brand, search for it in the brand list. */
178 for (i = 0; i < MAX_BRANDS; i++) {
179 bi = elf_brand_list[i];
180 if (bi != NULL && hdr->e_machine == bi->machine &&
181 (hdr->e_ident[EI_OSABI] == bi->brand ||
182 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
183 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
187 /* Lacking a known brand, search for a recognized interpreter. */
188 if (interp != NULL) {
189 for (i = 0; i < MAX_BRANDS; i++) {
190 bi = elf_brand_list[i];
191 if (bi != NULL && hdr->e_machine == bi->machine &&
192 strcmp(interp, bi->interp_path) == 0)
197 /* Lacking a recognized interpreter, try the default brand */
198 for (i = 0; i < MAX_BRANDS; i++) {
199 bi = elf_brand_list[i];
200 if (bi != NULL && hdr->e_machine == bi->machine &&
201 __elfN(fallback_brand) == bi->brand)
208 __elfN(check_header)(const Elf_Ehdr *hdr)
214 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
215 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
216 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
217 hdr->e_phentsize != sizeof(Elf_Phdr) ||
218 hdr->e_version != ELF_TARG_VER)
222 * Make sure we have at least one brand for this machine.
225 for (i = 0; i < MAX_BRANDS; i++) {
226 bi = elf_brand_list[i];
227 if (bi != NULL && bi->machine == hdr->e_machine)
237 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
238 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
245 * Create the page if it doesn't exist yet. Ignore errors.
248 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
249 VM_PROT_ALL, VM_PROT_ALL, 0);
253 * Find the page from the underlying object.
256 sf = vm_imgact_map_page(object, offset);
258 return (KERN_FAILURE);
259 off = offset - trunc_page(offset);
260 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
262 vm_imgact_unmap_page(sf);
264 return (KERN_FAILURE);
268 return (KERN_SUCCESS);
272 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
273 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
280 if (start != trunc_page(start)) {
281 rv = __elfN(map_partial)(map, object, offset, start,
282 round_page(start), prot);
285 offset += round_page(start) - start;
286 start = round_page(start);
288 if (end != round_page(end)) {
289 rv = __elfN(map_partial)(map, object, offset +
290 trunc_page(end) - start, trunc_page(end), end, prot);
293 end = trunc_page(end);
296 if (offset & PAGE_MASK) {
298 * The mapping is not page aligned. This means we have
299 * to copy the data. Sigh.
301 rv = vm_map_find(map, NULL, 0, &start, end - start,
302 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
306 return (KERN_SUCCESS);
307 for (; start < end; start += sz) {
308 sf = vm_imgact_map_page(object, offset);
310 return (KERN_FAILURE);
311 off = offset - trunc_page(offset);
313 if (sz > PAGE_SIZE - off)
314 sz = PAGE_SIZE - off;
315 error = copyout((caddr_t)sf_buf_kva(sf) + off,
317 vm_imgact_unmap_page(sf);
319 return (KERN_FAILURE);
325 vm_object_reference(object);
327 rv = vm_map_insert(map, object, offset, start, end,
328 prot, VM_PROT_ALL, cow);
330 if (rv != KERN_SUCCESS)
331 vm_object_deallocate(object);
335 return (KERN_SUCCESS);
340 __elfN(load_section)(struct vmspace *vmspace,
341 vm_object_t object, vm_offset_t offset,
342 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
347 vm_offset_t map_addr;
350 vm_offset_t file_addr;
353 * It's necessary to fail if the filsz + offset taken from the
354 * header is greater than the actual file pager object's size.
355 * If we were to allow this, then the vm_map_find() below would
356 * walk right off the end of the file object and into the ether.
358 * While I'm here, might as well check for something else that
359 * is invalid: filsz cannot be greater than memsz.
361 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
363 uprintf("elf_load_section: truncated ELF file\n");
367 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
368 file_addr = trunc_page_ps(offset, pagesize);
371 * We have two choices. We can either clear the data in the last page
372 * of an oversized mapping, or we can start the anon mapping a page
373 * early and copy the initialized data into that first page. We
374 * choose the second..
377 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
379 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
382 /* cow flags: don't dump readonly sections in core */
383 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
384 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
386 rv = __elfN(map_insert)(&vmspace->vm_map,
388 file_addr, /* file offset */
389 map_addr, /* virtual start */
390 map_addr + map_len,/* virtual end */
393 if (rv != KERN_SUCCESS)
396 /* we can stop now if we've covered it all */
397 if (memsz == filsz) {
404 * We have to get the remaining bit of the file into the first part
405 * of the oversized map segment. This is normally because the .data
406 * segment in the file is extended to provide bss. It's a neat idea
407 * to try and save a page, but it's a pain in the behind to implement.
409 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
410 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
411 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
414 /* This had damn well better be true! */
416 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
417 map_addr + map_len, VM_PROT_ALL, 0);
418 if (rv != KERN_SUCCESS) {
426 sf = vm_imgact_map_page(object, offset + filsz);
430 /* send the page fragment to user space */
431 off = trunc_page_ps(offset + filsz, pagesize) -
432 trunc_page(offset + filsz);
433 error = copyout((caddr_t)sf_buf_kva(sf) + off,
434 (caddr_t)map_addr, copy_len);
435 vm_imgact_unmap_page(sf);
442 * set it to the specified protection.
443 * XXX had better undo the damage from pasting over the cracks here!
445 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
446 round_page(map_addr + map_len), prot, FALSE);
452 * Load the file "file" into memory. It may be either a shared object
455 * The "addr" reference parameter is in/out. On entry, it specifies
456 * the address where a shared object should be loaded. If the file is
457 * an executable, this value is ignored. On exit, "addr" specifies
458 * where the file was actually loaded.
460 * The "entry" reference parameter is out only. On exit, it specifies
461 * the entry point for the loaded file.
464 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
465 u_long *entry, size_t pagesize)
470 struct image_params image_params;
472 const Elf_Ehdr *hdr = NULL;
473 const Elf_Phdr *phdr = NULL;
474 struct nameidata *nd;
475 struct vmspace *vmspace = p->p_vmspace;
477 struct image_params *imgp;
480 u_long base_addr = 0;
481 int vfslocked, error, i, numsegs;
483 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
485 attr = &tempdata->attr;
486 imgp = &tempdata->image_params;
489 * Initialize part of the common data
493 imgp->firstpage = NULL;
494 imgp->image_header = NULL;
496 imgp->execlabel = NULL;
498 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
501 if ((error = namei(nd)) != 0) {
505 vfslocked = NDHASGIANT(nd);
506 NDFREE(nd, NDF_ONLY_PNBUF);
507 imgp->vp = nd->ni_vp;
510 * Check permissions, modes, uid, etc on the file, and "open" it.
512 error = exec_check_permissions(imgp);
516 error = exec_map_first_page(imgp);
521 * Also make certain that the interpreter stays the same, so set
522 * its VV_TEXT flag, too.
524 nd->ni_vp->v_vflag |= VV_TEXT;
526 imgp->object = nd->ni_vp->v_object;
528 hdr = (const Elf_Ehdr *)imgp->image_header;
529 if ((error = __elfN(check_header)(hdr)) != 0)
531 if (hdr->e_type == ET_DYN)
533 else if (hdr->e_type == ET_EXEC)
540 /* Only support headers that fit within first page for now */
541 /* (multiplication of two Elf_Half fields will not overflow) */
542 if ((hdr->e_phoff > PAGE_SIZE) ||
543 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
548 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
549 if (!aligned(phdr, Elf_Addr)) {
554 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
555 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
557 if (phdr[i].p_flags & PF_X)
558 prot |= VM_PROT_EXECUTE;
559 if (phdr[i].p_flags & PF_W)
560 prot |= VM_PROT_WRITE;
561 if (phdr[i].p_flags & PF_R)
562 prot |= VM_PROT_READ;
564 if ((error = __elfN(load_section)(vmspace,
565 imgp->object, phdr[i].p_offset,
566 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
567 phdr[i].p_memsz, phdr[i].p_filesz, prot,
571 * Establish the base address if this is the
575 base_addr = trunc_page(phdr[i].p_vaddr +
581 *entry = (unsigned long)hdr->e_entry + rbase;
585 exec_unmap_first_page(imgp);
590 VFS_UNLOCK_GIANT(vfslocked);
591 free(tempdata, M_TEMP);
596 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
599 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
601 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
602 const Elf_Phdr *phdr, *pnote = NULL;
603 Elf_Auxargs *elf_auxargs;
604 struct vmspace *vmspace;
606 u_long text_size = 0, data_size = 0, total_size = 0;
607 u_long text_addr = 0, data_addr = 0;
608 u_long seg_size, seg_addr;
609 u_long addr, entry = 0, proghdr = 0;
611 const char *interp = NULL, *newinterp = NULL;
612 Elf_Brandinfo *brand_info;
613 const Elf_Note *note, *note_end;
615 const char *note_name;
616 struct sysentvec *sv;
619 * Do we have a valid ELF header ?
621 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
622 * if particular brand doesn't support it.
624 if (__elfN(check_header)(hdr) != 0 ||
625 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
629 * From here on down, we return an errno, not -1, as we've
630 * detected an ELF file.
633 if ((hdr->e_phoff > PAGE_SIZE) ||
634 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
635 /* Only support headers in first page for now */
638 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
639 if (!aligned(phdr, Elf_Addr))
641 for (i = 0; i < hdr->e_phnum; i++) {
642 if (phdr[i].p_type == PT_INTERP) {
643 /* Path to interpreter */
644 if (phdr[i].p_filesz > MAXPATHLEN ||
645 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
647 interp = imgp->image_header + phdr[i].p_offset;
652 brand_info = __elfN(get_brandinfo)(hdr, interp);
653 if (brand_info == NULL) {
654 uprintf("ELF binary type \"%u\" not known.\n",
655 hdr->e_ident[EI_OSABI]);
658 if (hdr->e_type == ET_DYN &&
659 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
661 sv = brand_info->sysvec;
662 if (interp != NULL && brand_info->interp_newpath != NULL)
663 newinterp = brand_info->interp_newpath;
666 * Avoid a possible deadlock if the current address space is destroyed
667 * and that address space maps the locked vnode. In the common case,
668 * the locked vnode's v_usecount is decremented but remains greater
669 * than zero. Consequently, the vnode lock is not needed by vrele().
670 * However, in cases where the vnode lock is external, such as nullfs,
671 * v_usecount may become zero.
673 VOP_UNLOCK(imgp->vp, 0);
675 error = exec_new_vmspace(imgp, sv);
676 imgp->proc->p_sysent = sv;
678 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
682 vmspace = imgp->proc->p_vmspace;
684 for (i = 0; i < hdr->e_phnum; i++) {
685 switch (phdr[i].p_type) {
686 case PT_LOAD: /* Loadable segment */
688 if (phdr[i].p_flags & PF_X)
689 prot |= VM_PROT_EXECUTE;
690 if (phdr[i].p_flags & PF_W)
691 prot |= VM_PROT_WRITE;
692 if (phdr[i].p_flags & PF_R)
693 prot |= VM_PROT_READ;
695 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
697 * Some x86 binaries assume read == executable,
698 * notably the M3 runtime and therefore cvsup
700 if (prot & VM_PROT_READ)
701 prot |= VM_PROT_EXECUTE;
704 if ((error = __elfN(load_section)(vmspace,
705 imgp->object, phdr[i].p_offset,
706 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
707 phdr[i].p_memsz, phdr[i].p_filesz, prot,
708 sv->sv_pagesize)) != 0)
712 * If this segment contains the program headers,
713 * remember their virtual address for the AT_PHDR
714 * aux entry. Static binaries don't usually include
717 if (phdr[i].p_offset == 0 &&
718 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
720 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
722 seg_addr = trunc_page(phdr[i].p_vaddr);
723 seg_size = round_page(phdr[i].p_memsz +
724 phdr[i].p_vaddr - seg_addr);
727 * Is this .text or .data? We can't use
728 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
729 * alpha terribly and possibly does other bad
730 * things so we stick to the old way of figuring
731 * it out: If the segment contains the program
732 * entry point, it's a text segment, otherwise it
735 * Note that obreak() assumes that data_addr +
736 * data_size == end of data load area, and the ELF
737 * file format expects segments to be sorted by
738 * address. If multiple data segments exist, the
739 * last one will be used.
741 if (hdr->e_entry >= phdr[i].p_vaddr &&
742 hdr->e_entry < (phdr[i].p_vaddr +
744 text_size = seg_size;
745 text_addr = seg_addr;
746 entry = (u_long)hdr->e_entry;
748 data_size = seg_size;
749 data_addr = seg_addr;
751 total_size += seg_size;
753 case PT_PHDR: /* Program header table info */
754 proghdr = phdr[i].p_vaddr;
764 if (data_addr == 0 && data_size == 0) {
765 data_addr = text_addr;
766 data_size = text_size;
770 * Check limits. It should be safe to check the
771 * limits after loading the segments since we do
772 * not actually fault in all the segments pages.
774 PROC_LOCK(imgp->proc);
775 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
776 text_size > maxtsiz ||
777 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
778 PROC_UNLOCK(imgp->proc);
782 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
783 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
784 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
785 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
788 * We load the dynamic linker where a userland call
789 * to mmap(0, ...) would put it. The rationale behind this
790 * calculation is that it leaves room for the heap to grow to
791 * its maximum allowed size.
793 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
794 lim_max(imgp->proc, RLIMIT_DATA));
795 PROC_UNLOCK(imgp->proc);
797 imgp->entry_addr = entry;
799 if (interp != NULL) {
800 int have_interp = FALSE;
801 VOP_UNLOCK(imgp->vp, 0);
802 if (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",
806 brand_info->emul_path, interp);
807 error = __elfN(load_file)(imgp->proc, path, &addr,
808 &imgp->entry_addr, sv->sv_pagesize);
813 if (!have_interp && newinterp != NULL) {
814 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
815 &imgp->entry_addr, sv->sv_pagesize);
820 error = __elfN(load_file)(imgp->proc, interp, &addr,
821 &imgp->entry_addr, sv->sv_pagesize);
823 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
825 uprintf("ELF interpreter %s not found\n", interp);
831 * Construct auxargs table (used by the fixup routine)
833 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
834 elf_auxargs->execfd = -1;
835 elf_auxargs->phdr = proghdr;
836 elf_auxargs->phent = hdr->e_phentsize;
837 elf_auxargs->phnum = hdr->e_phnum;
838 elf_auxargs->pagesz = PAGE_SIZE;
839 elf_auxargs->base = addr;
840 elf_auxargs->flags = 0;
841 elf_auxargs->entry = entry;
842 elf_auxargs->trace = elf_trace;
844 imgp->auxargs = elf_auxargs;
845 imgp->interpreted = 0;
848 * Try to fetch the osreldate for FreeBSD binary from the ELF
849 * OSABI-note. Only the first page of the image is searched,
850 * the same as for headers.
852 if (pnote != NULL && pnote->p_offset < PAGE_SIZE &&
853 pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) {
854 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
855 if (!aligned(note, Elf32_Addr)) {
856 free(imgp->auxargs, M_TEMP);
857 imgp->auxargs = NULL;
860 note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset +
862 while (note < note_end) {
863 if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) &&
864 note->n_descsz == sizeof(int32_t) &&
865 note->n_type == 1 /* ABI_NOTETYPE */) {
866 note_name = (const char *)(note + 1);
867 if (strncmp(FREEBSD_ABI_VENDOR, note_name,
868 sizeof(FREEBSD_ABI_VENDOR)) == 0) {
869 imgp->proc->p_osrel = *(const int32_t *)
871 round_page_ps(sizeof(FREEBSD_ABI_VENDOR),
872 sizeof(Elf32_Addr)));
876 note = (const Elf_Note *)((const char *)(note + 1) +
877 round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) +
878 round_page_ps(note->n_descsz, sizeof(Elf32_Addr)));
885 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
888 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
890 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
894 base = (Elf_Addr *)*stack_base;
895 pos = base + (imgp->args->argc + imgp->args->envc + 2);
898 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
900 if (args->execfd != -1) {
901 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
903 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
904 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
905 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
906 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
907 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
908 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
909 AUXARGS_ENTRY(pos, AT_BASE, args->base);
910 AUXARGS_ENTRY(pos, AT_NULL, 0);
912 free(imgp->auxargs, M_TEMP);
913 imgp->auxargs = NULL;
916 suword(base, (long)imgp->args->argc);
917 *stack_base = (register_t *)base;
922 * Code for generating ELF core dumps.
925 typedef void (*segment_callback)(vm_map_entry_t, void *);
927 /* Closure for cb_put_phdr(). */
928 struct phdr_closure {
929 Elf_Phdr *phdr; /* Program header to fill in */
930 Elf_Off offset; /* Offset of segment in core file */
933 /* Closure for cb_size_segment(). */
934 struct sseg_closure {
935 int count; /* Count of writable segments. */
936 size_t size; /* Total size of all writable segments. */
939 static void cb_put_phdr(vm_map_entry_t, void *);
940 static void cb_size_segment(vm_map_entry_t, void *);
941 static void each_writable_segment(struct thread *, segment_callback, void *);
942 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
943 int, void *, size_t);
944 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
945 static void __elfN(putnote)(void *, size_t *, const char *, int,
946 const void *, size_t);
949 __elfN(coredump)(td, vp, limit)
954 struct ucred *cred = td->td_ucred;
956 struct sseg_closure seginfo;
960 /* Size the program segments. */
963 each_writable_segment(td, cb_size_segment, &seginfo);
966 * Calculate the size of the core file header area by making
967 * a dry run of generating it. Nothing is written, but the
968 * size is calculated.
971 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
973 if (hdrsize + seginfo.size >= limit)
977 * Allocate memory for building the header, fill it up,
980 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
984 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
986 /* Write the contents of all of the writable segments. */
992 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
994 for (i = 0; i < seginfo.count; i++) {
995 error = vn_rdwr_inchunks(UIO_WRITE, vp,
996 (caddr_t)(uintptr_t)php->p_vaddr,
997 php->p_filesz, offset, UIO_USERSPACE,
998 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1002 offset += php->p_filesz;
1012 * A callback for each_writable_segment() to write out the segment's
1013 * program header entry.
1016 cb_put_phdr(entry, closure)
1017 vm_map_entry_t entry;
1020 struct phdr_closure *phc = (struct phdr_closure *)closure;
1021 Elf_Phdr *phdr = phc->phdr;
1023 phc->offset = round_page(phc->offset);
1025 phdr->p_type = PT_LOAD;
1026 phdr->p_offset = phc->offset;
1027 phdr->p_vaddr = entry->start;
1029 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1030 phdr->p_align = PAGE_SIZE;
1032 if (entry->protection & VM_PROT_READ)
1033 phdr->p_flags |= PF_R;
1034 if (entry->protection & VM_PROT_WRITE)
1035 phdr->p_flags |= PF_W;
1036 if (entry->protection & VM_PROT_EXECUTE)
1037 phdr->p_flags |= PF_X;
1039 phc->offset += phdr->p_filesz;
1044 * A callback for each_writable_segment() to gather information about
1045 * the number of segments and their total size.
1048 cb_size_segment(entry, closure)
1049 vm_map_entry_t entry;
1052 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1055 ssc->size += entry->end - entry->start;
1059 * For each writable segment in the process's memory map, call the given
1060 * function with a pointer to the map entry and some arbitrary
1061 * caller-supplied data.
1064 each_writable_segment(td, func, closure)
1066 segment_callback func;
1069 struct proc *p = td->td_proc;
1070 vm_map_t map = &p->p_vmspace->vm_map;
1071 vm_map_entry_t entry;
1072 vm_object_t backing_object, object;
1073 boolean_t ignore_entry;
1075 vm_map_lock_read(map);
1076 for (entry = map->header.next; entry != &map->header;
1077 entry = entry->next) {
1079 * Don't dump inaccessible mappings, deal with legacy
1082 * Note that read-only segments related to the elf binary
1083 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1084 * need to arbitrarily ignore such segments.
1086 if (elf_legacy_coredump) {
1087 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1090 if ((entry->protection & VM_PROT_ALL) == 0)
1095 * Dont include memory segment in the coredump if
1096 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1097 * madvise(2). Do not dump submaps (i.e. parts of the
1100 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1103 if ((object = entry->object.vm_object) == NULL)
1106 /* Ignore memory-mapped devices and such things. */
1107 VM_OBJECT_LOCK(object);
1108 while ((backing_object = object->backing_object) != NULL) {
1109 VM_OBJECT_LOCK(backing_object);
1110 VM_OBJECT_UNLOCK(object);
1111 object = backing_object;
1113 ignore_entry = object->type != OBJT_DEFAULT &&
1114 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1115 VM_OBJECT_UNLOCK(object);
1119 (*func)(entry, closure);
1121 vm_map_unlock_read(map);
1125 * Write the core file header to the file, including padding up to
1126 * the page boundary.
1129 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1139 /* Fill in the header. */
1140 bzero(hdr, hdrsize);
1142 __elfN(puthdr)(td, hdr, &off, numsegs);
1144 /* Write it to the core file. */
1145 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1146 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1150 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1151 typedef struct prstatus32 elf_prstatus_t;
1152 typedef struct prpsinfo32 elf_prpsinfo_t;
1153 typedef struct fpreg32 elf_prfpregset_t;
1154 typedef struct fpreg32 elf_fpregset_t;
1155 typedef struct reg32 elf_gregset_t;
1157 typedef prstatus_t elf_prstatus_t;
1158 typedef prpsinfo_t elf_prpsinfo_t;
1159 typedef prfpregset_t elf_prfpregset_t;
1160 typedef prfpregset_t elf_fpregset_t;
1161 typedef gregset_t elf_gregset_t;
1165 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1168 elf_prstatus_t status;
1169 elf_prfpregset_t fpregset;
1170 elf_prpsinfo_t psinfo;
1172 elf_prstatus_t *status;
1173 elf_prfpregset_t *fpregset;
1174 elf_prpsinfo_t *psinfo;
1177 size_t ehoff, noteoff, notesz, phoff;
1182 *off += sizeof(Elf_Ehdr);
1185 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1189 * Don't allocate space for the notes if we're just calculating
1190 * the size of the header. We also don't collect the data.
1193 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1194 status = &tempdata->status;
1195 fpregset = &tempdata->fpregset;
1196 psinfo = &tempdata->psinfo;
1205 psinfo->pr_version = PRPSINFO_VERSION;
1206 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1207 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1209 * XXX - We don't fill in the command line arguments properly
1212 strlcpy(psinfo->pr_psargs, p->p_comm,
1213 sizeof(psinfo->pr_psargs));
1215 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1219 * To have the debugger select the right thread (LWP) as the initial
1220 * thread, we dump the state of the thread passed to us in td first.
1221 * This is the thread that causes the core dump and thus likely to
1222 * be the right thread one wants to have selected in the debugger.
1225 while (thr != NULL) {
1227 status->pr_version = PRSTATUS_VERSION;
1228 status->pr_statussz = sizeof(elf_prstatus_t);
1229 status->pr_gregsetsz = sizeof(elf_gregset_t);
1230 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1231 status->pr_osreldate = osreldate;
1232 status->pr_cursig = p->p_sig;
1233 status->pr_pid = thr->td_tid;
1234 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1235 fill_regs32(thr, &status->pr_reg);
1236 fill_fpregs32(thr, fpregset);
1238 fill_regs(thr, &status->pr_reg);
1239 fill_fpregs(thr, fpregset);
1242 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1244 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1247 * Allow for MD specific notes, as well as any MD
1248 * specific preparations for writing MI notes.
1250 __elfN(dump_thread)(thr, dst, off);
1252 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1253 TAILQ_NEXT(thr, td_plist);
1255 thr = TAILQ_NEXT(thr, td_plist);
1258 notesz = *off - noteoff;
1261 free(tempdata, M_TEMP);
1263 /* Align up to a page boundary for the program segments. */
1264 *off = round_page(*off);
1269 struct phdr_closure phc;
1272 * Fill in the ELF header.
1274 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1275 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1276 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1277 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1278 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1279 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1280 ehdr->e_ident[EI_DATA] = ELF_DATA;
1281 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1282 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1283 ehdr->e_ident[EI_ABIVERSION] = 0;
1284 ehdr->e_ident[EI_PAD] = 0;
1285 ehdr->e_type = ET_CORE;
1286 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1287 ehdr->e_machine = EM_386;
1289 ehdr->e_machine = ELF_ARCH;
1291 ehdr->e_version = EV_CURRENT;
1293 ehdr->e_phoff = phoff;
1295 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1296 ehdr->e_phentsize = sizeof(Elf_Phdr);
1297 ehdr->e_phnum = numsegs + 1;
1298 ehdr->e_shentsize = sizeof(Elf_Shdr);
1300 ehdr->e_shstrndx = SHN_UNDEF;
1303 * Fill in the program header entries.
1305 phdr = (Elf_Phdr *)((char *)dst + phoff);
1307 /* The note segement. */
1308 phdr->p_type = PT_NOTE;
1309 phdr->p_offset = noteoff;
1312 phdr->p_filesz = notesz;
1318 /* All the writable segments from the program. */
1321 each_writable_segment(td, cb_put_phdr, &phc);
1326 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1327 const void *desc, size_t descsz)
1331 note.n_namesz = strlen(name) + 1;
1332 note.n_descsz = descsz;
1335 bcopy(¬e, (char *)dst + *off, sizeof note);
1336 *off += sizeof note;
1338 bcopy(name, (char *)dst + *off, note.n_namesz);
1339 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1341 bcopy(desc, (char *)dst + *off, note.n_descsz);
1342 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1346 * Tell kern_execve.c about it, with a little help from the linker.
1348 static struct execsw __elfN(execsw) = {
1349 __CONCAT(exec_, __elfN(imgact)),
1350 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1352 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));