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(freebsd_trans_osrel)(const Elf_Note *note,
91 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
92 static boolean_t __elfN(check_note)(struct image_params *imgp,
93 Elf_Brandnote *checknote, int32_t *osrel);
95 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
98 int __elfN(fallback_brand) = -1;
99 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
100 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
101 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
102 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
103 &__elfN(fallback_brand));
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_TRANSLATE_OSREL,
123 .trans_osrel = __elfN(freebsd_trans_osrel)
127 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
131 p = (uintptr_t)(note + 1);
132 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
133 *osrel = *(const int32_t *)(p);
138 static const char GNU_ABI_VENDOR[] = "GNU";
139 static int GNU_KFREEBSD_ABI_DESC = 3;
141 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
142 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
143 .hdr.n_descsz = 16, /* XXX at least 16 */
145 .vendor = GNU_ABI_VENDOR,
146 .flags = BN_TRANSLATE_OSREL,
147 .trans_osrel = kfreebsd_trans_osrel
151 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
153 const Elf32_Word *desc;
156 p = (uintptr_t)(note + 1);
157 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
159 desc = (const Elf32_Word *)p;
160 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
164 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
165 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
167 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
173 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
177 for (i = 0; i < MAX_BRANDS; i++) {
178 if (elf_brand_list[i] == NULL) {
179 elf_brand_list[i] = entry;
189 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
193 for (i = 0; i < MAX_BRANDS; i++) {
194 if (elf_brand_list[i] == entry) {
195 elf_brand_list[i] = NULL;
205 __elfN(brand_inuse)(Elf_Brandinfo *entry)
210 sx_slock(&allproc_lock);
211 FOREACH_PROC_IN_SYSTEM(p) {
212 if (p->p_sysent == entry->sysvec) {
217 sx_sunlock(&allproc_lock);
222 static Elf_Brandinfo *
223 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
226 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
232 * We support four types of branding -- (1) the ELF EI_OSABI field
233 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
234 * branding w/in the ELF header, (3) path of the `interp_path'
235 * field, and (4) the ".note.ABI-tag" ELF section.
238 /* Look for an ".note.ABI-tag" ELF section */
239 for (i = 0; i < MAX_BRANDS; i++) {
240 bi = elf_brand_list[i];
243 if (hdr->e_machine == bi->machine && (bi->flags &
244 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
245 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
251 /* If the executable has a brand, search for it in the brand list. */
252 for (i = 0; i < MAX_BRANDS; i++) {
253 bi = elf_brand_list[i];
254 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
256 if (hdr->e_machine == bi->machine &&
257 (hdr->e_ident[EI_OSABI] == bi->brand ||
258 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
259 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
263 /* Lacking a known brand, search for a recognized interpreter. */
264 if (interp != NULL) {
265 for (i = 0; i < MAX_BRANDS; i++) {
266 bi = elf_brand_list[i];
267 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
269 if (hdr->e_machine == bi->machine &&
270 strcmp(interp, bi->interp_path) == 0)
275 /* Lacking a recognized interpreter, try the default brand */
276 for (i = 0; i < MAX_BRANDS; i++) {
277 bi = elf_brand_list[i];
278 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
280 if (hdr->e_machine == bi->machine &&
281 __elfN(fallback_brand) == bi->brand)
288 __elfN(check_header)(const Elf_Ehdr *hdr)
294 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
295 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
296 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
297 hdr->e_phentsize != sizeof(Elf_Phdr) ||
298 hdr->e_version != ELF_TARG_VER)
302 * Make sure we have at least one brand for this machine.
305 for (i = 0; i < MAX_BRANDS; i++) {
306 bi = elf_brand_list[i];
307 if (bi != NULL && bi->machine == hdr->e_machine)
317 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
318 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
325 * Create the page if it doesn't exist yet. Ignore errors.
328 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
329 VM_PROT_ALL, VM_PROT_ALL, 0);
333 * Find the page from the underlying object.
336 sf = vm_imgact_map_page(object, offset);
338 return (KERN_FAILURE);
339 off = offset - trunc_page(offset);
340 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
342 vm_imgact_unmap_page(sf);
344 return (KERN_FAILURE);
348 return (KERN_SUCCESS);
352 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
353 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
360 if (start != trunc_page(start)) {
361 rv = __elfN(map_partial)(map, object, offset, start,
362 round_page(start), prot);
365 offset += round_page(start) - start;
366 start = round_page(start);
368 if (end != round_page(end)) {
369 rv = __elfN(map_partial)(map, object, offset +
370 trunc_page(end) - start, trunc_page(end), end, prot);
373 end = trunc_page(end);
376 if (offset & PAGE_MASK) {
378 * The mapping is not page aligned. This means we have
379 * to copy the data. Sigh.
381 rv = vm_map_find(map, NULL, 0, &start, end - start,
382 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
386 return (KERN_SUCCESS);
387 for (; start < end; start += sz) {
388 sf = vm_imgact_map_page(object, offset);
390 return (KERN_FAILURE);
391 off = offset - trunc_page(offset);
393 if (sz > PAGE_SIZE - off)
394 sz = PAGE_SIZE - off;
395 error = copyout((caddr_t)sf_buf_kva(sf) + off,
397 vm_imgact_unmap_page(sf);
399 return (KERN_FAILURE);
405 vm_object_reference(object);
407 rv = vm_map_insert(map, object, offset, start, end,
408 prot, VM_PROT_ALL, cow);
410 if (rv != KERN_SUCCESS)
411 vm_object_deallocate(object);
415 return (KERN_SUCCESS);
420 __elfN(load_section)(struct vmspace *vmspace,
421 vm_object_t object, vm_offset_t offset,
422 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
427 vm_offset_t map_addr;
430 vm_offset_t file_addr;
433 * It's necessary to fail if the filsz + offset taken from the
434 * header is greater than the actual file pager object's size.
435 * If we were to allow this, then the vm_map_find() below would
436 * walk right off the end of the file object and into the ether.
438 * While I'm here, might as well check for something else that
439 * is invalid: filsz cannot be greater than memsz.
441 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
443 uprintf("elf_load_section: truncated ELF file\n");
447 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
448 file_addr = trunc_page_ps(offset, pagesize);
451 * We have two choices. We can either clear the data in the last page
452 * of an oversized mapping, or we can start the anon mapping a page
453 * early and copy the initialized data into that first page. We
454 * choose the second..
457 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
459 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
462 /* cow flags: don't dump readonly sections in core */
463 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
464 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
466 rv = __elfN(map_insert)(&vmspace->vm_map,
468 file_addr, /* file offset */
469 map_addr, /* virtual start */
470 map_addr + map_len,/* virtual end */
473 if (rv != KERN_SUCCESS)
476 /* we can stop now if we've covered it all */
477 if (memsz == filsz) {
484 * We have to get the remaining bit of the file into the first part
485 * of the oversized map segment. This is normally because the .data
486 * segment in the file is extended to provide bss. It's a neat idea
487 * to try and save a page, but it's a pain in the behind to implement.
489 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
490 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
491 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
494 /* This had damn well better be true! */
496 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
497 map_addr + map_len, VM_PROT_ALL, 0);
498 if (rv != KERN_SUCCESS) {
506 sf = vm_imgact_map_page(object, offset + filsz);
510 /* send the page fragment to user space */
511 off = trunc_page_ps(offset + filsz, pagesize) -
512 trunc_page(offset + filsz);
513 error = copyout((caddr_t)sf_buf_kva(sf) + off,
514 (caddr_t)map_addr, copy_len);
515 vm_imgact_unmap_page(sf);
522 * set it to the specified protection.
523 * XXX had better undo the damage from pasting over the cracks here!
525 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
526 round_page(map_addr + map_len), prot, FALSE);
532 * Load the file "file" into memory. It may be either a shared object
535 * The "addr" reference parameter is in/out. On entry, it specifies
536 * the address where a shared object should be loaded. If the file is
537 * an executable, this value is ignored. On exit, "addr" specifies
538 * where the file was actually loaded.
540 * The "entry" reference parameter is out only. On exit, it specifies
541 * the entry point for the loaded file.
544 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
545 u_long *entry, size_t pagesize)
550 struct image_params image_params;
552 const Elf_Ehdr *hdr = NULL;
553 const Elf_Phdr *phdr = NULL;
554 struct nameidata *nd;
555 struct vmspace *vmspace = p->p_vmspace;
557 struct image_params *imgp;
560 u_long base_addr = 0;
561 int vfslocked, error, i, numsegs;
563 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
565 attr = &tempdata->attr;
566 imgp = &tempdata->image_params;
569 * Initialize part of the common data
573 imgp->firstpage = NULL;
574 imgp->image_header = NULL;
576 imgp->execlabel = NULL;
578 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
581 if ((error = namei(nd)) != 0) {
585 vfslocked = NDHASGIANT(nd);
586 NDFREE(nd, NDF_ONLY_PNBUF);
587 imgp->vp = nd->ni_vp;
590 * Check permissions, modes, uid, etc on the file, and "open" it.
592 error = exec_check_permissions(imgp);
596 error = exec_map_first_page(imgp);
601 * Also make certain that the interpreter stays the same, so set
602 * its VV_TEXT flag, too.
604 nd->ni_vp->v_vflag |= VV_TEXT;
606 imgp->object = nd->ni_vp->v_object;
608 hdr = (const Elf_Ehdr *)imgp->image_header;
609 if ((error = __elfN(check_header)(hdr)) != 0)
611 if (hdr->e_type == ET_DYN)
613 else if (hdr->e_type == ET_EXEC)
620 /* Only support headers that fit within first page for now */
621 /* (multiplication of two Elf_Half fields will not overflow) */
622 if ((hdr->e_phoff > PAGE_SIZE) ||
623 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
628 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
629 if (!aligned(phdr, Elf_Addr)) {
634 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
635 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
637 if (phdr[i].p_flags & PF_X)
638 prot |= VM_PROT_EXECUTE;
639 if (phdr[i].p_flags & PF_W)
640 prot |= VM_PROT_WRITE;
641 if (phdr[i].p_flags & PF_R)
642 prot |= VM_PROT_READ;
644 if ((error = __elfN(load_section)(vmspace,
645 imgp->object, phdr[i].p_offset,
646 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
647 phdr[i].p_memsz, phdr[i].p_filesz, prot,
651 * Establish the base address if this is the
655 base_addr = trunc_page(phdr[i].p_vaddr +
661 *entry = (unsigned long)hdr->e_entry + rbase;
665 exec_unmap_first_page(imgp);
670 VFS_UNLOCK_GIANT(vfslocked);
671 free(tempdata, M_TEMP);
677 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
679 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
680 const Elf_Phdr *phdr;
681 Elf_Auxargs *elf_auxargs;
682 struct vmspace *vmspace;
684 u_long text_size = 0, data_size = 0, total_size = 0;
685 u_long text_addr = 0, data_addr = 0;
686 u_long seg_size, seg_addr;
687 u_long addr, entry = 0, proghdr = 0;
690 const char *interp = NULL, *newinterp = NULL;
691 Elf_Brandinfo *brand_info;
693 struct sysentvec *sv;
696 * Do we have a valid ELF header ?
698 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
699 * if particular brand doesn't support it.
701 if (__elfN(check_header)(hdr) != 0 ||
702 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
706 * From here on down, we return an errno, not -1, as we've
707 * detected an ELF file.
710 if ((hdr->e_phoff > PAGE_SIZE) ||
711 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
712 /* Only support headers in first page for now */
715 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
716 if (!aligned(phdr, Elf_Addr))
718 for (i = 0; i < hdr->e_phnum; i++) {
719 if (phdr[i].p_type == PT_INTERP) {
720 /* Path to interpreter */
721 if (phdr[i].p_filesz > MAXPATHLEN ||
722 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
724 interp = imgp->image_header + phdr[i].p_offset;
729 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
730 if (brand_info == NULL) {
731 uprintf("ELF binary type \"%u\" not known.\n",
732 hdr->e_ident[EI_OSABI]);
735 if (hdr->e_type == ET_DYN &&
736 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
738 sv = brand_info->sysvec;
739 if (interp != NULL && brand_info->interp_newpath != NULL)
740 newinterp = brand_info->interp_newpath;
743 * Avoid a possible deadlock if the current address space is destroyed
744 * and that address space maps the locked vnode. In the common case,
745 * the locked vnode's v_usecount is decremented but remains greater
746 * than zero. Consequently, the vnode lock is not needed by vrele().
747 * However, in cases where the vnode lock is external, such as nullfs,
748 * v_usecount may become zero.
750 VOP_UNLOCK(imgp->vp, 0);
752 error = exec_new_vmspace(imgp, sv);
753 imgp->proc->p_sysent = sv;
755 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
759 vmspace = imgp->proc->p_vmspace;
761 for (i = 0; i < hdr->e_phnum; i++) {
762 switch (phdr[i].p_type) {
763 case PT_LOAD: /* Loadable segment */
765 if (phdr[i].p_flags & PF_X)
766 prot |= VM_PROT_EXECUTE;
767 if (phdr[i].p_flags & PF_W)
768 prot |= VM_PROT_WRITE;
769 if (phdr[i].p_flags & PF_R)
770 prot |= VM_PROT_READ;
772 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
774 * Some x86 binaries assume read == executable,
775 * notably the M3 runtime and therefore cvsup
777 if (prot & VM_PROT_READ)
778 prot |= VM_PROT_EXECUTE;
781 if ((error = __elfN(load_section)(vmspace,
782 imgp->object, phdr[i].p_offset,
783 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
784 phdr[i].p_memsz, phdr[i].p_filesz, prot,
785 sv->sv_pagesize)) != 0)
789 * If this segment contains the program headers,
790 * remember their virtual address for the AT_PHDR
791 * aux entry. Static binaries don't usually include
794 if (phdr[i].p_offset == 0 &&
795 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
797 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
799 seg_addr = trunc_page(phdr[i].p_vaddr);
800 seg_size = round_page(phdr[i].p_memsz +
801 phdr[i].p_vaddr - seg_addr);
804 * Is this .text or .data? We can't use
805 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
806 * alpha terribly and possibly does other bad
807 * things so we stick to the old way of figuring
808 * it out: If the segment contains the program
809 * entry point, it's a text segment, otherwise it
812 * Note that obreak() assumes that data_addr +
813 * data_size == end of data load area, and the ELF
814 * file format expects segments to be sorted by
815 * address. If multiple data segments exist, the
816 * last one will be used.
818 if (hdr->e_entry >= phdr[i].p_vaddr &&
819 hdr->e_entry < (phdr[i].p_vaddr +
821 text_size = seg_size;
822 text_addr = seg_addr;
823 entry = (u_long)hdr->e_entry;
825 data_size = seg_size;
826 data_addr = seg_addr;
828 total_size += seg_size;
830 case PT_PHDR: /* Program header table info */
831 proghdr = phdr[i].p_vaddr;
838 if (data_addr == 0 && data_size == 0) {
839 data_addr = text_addr;
840 data_size = text_size;
844 * Check limits. It should be safe to check the
845 * limits after loading the segments since we do
846 * not actually fault in all the segments pages.
848 PROC_LOCK(imgp->proc);
849 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
850 text_size > maxtsiz ||
851 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
852 PROC_UNLOCK(imgp->proc);
856 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
857 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
858 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
859 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
862 * We load the dynamic linker where a userland call
863 * to mmap(0, ...) would put it. The rationale behind this
864 * calculation is that it leaves room for the heap to grow to
865 * its maximum allowed size.
867 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
868 lim_max(imgp->proc, RLIMIT_DATA));
869 PROC_UNLOCK(imgp->proc);
871 imgp->entry_addr = entry;
873 if (interp != NULL) {
874 int have_interp = FALSE;
875 VOP_UNLOCK(imgp->vp, 0);
876 if (brand_info->emul_path != NULL &&
877 brand_info->emul_path[0] != '\0') {
878 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
879 snprintf(path, MAXPATHLEN, "%s%s",
880 brand_info->emul_path, interp);
881 error = __elfN(load_file)(imgp->proc, path, &addr,
882 &imgp->entry_addr, sv->sv_pagesize);
887 if (!have_interp && newinterp != NULL) {
888 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
889 &imgp->entry_addr, sv->sv_pagesize);
894 error = __elfN(load_file)(imgp->proc, interp, &addr,
895 &imgp->entry_addr, sv->sv_pagesize);
897 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
899 uprintf("ELF interpreter %s not found\n", interp);
906 * Construct auxargs table (used by the fixup routine)
908 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
909 elf_auxargs->execfd = -1;
910 elf_auxargs->phdr = proghdr;
911 elf_auxargs->phent = hdr->e_phentsize;
912 elf_auxargs->phnum = hdr->e_phnum;
913 elf_auxargs->pagesz = PAGE_SIZE;
914 elf_auxargs->base = addr;
915 elf_auxargs->flags = 0;
916 elf_auxargs->entry = entry;
918 imgp->auxargs = elf_auxargs;
919 imgp->interpreted = 0;
920 imgp->proc->p_osrel = osrel;
925 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
928 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
930 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
934 base = (Elf_Addr *)*stack_base;
935 pos = base + (imgp->args->argc + imgp->args->envc + 2);
937 if (args->execfd != -1)
938 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
939 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
940 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
941 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
942 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
943 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
944 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
945 AUXARGS_ENTRY(pos, AT_BASE, args->base);
946 if (imgp->execpathp != 0)
947 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
948 AUXARGS_ENTRY(pos, AT_NULL, 0);
950 free(imgp->auxargs, M_TEMP);
951 imgp->auxargs = NULL;
954 suword(base, (long)imgp->args->argc);
955 *stack_base = (register_t *)base;
960 * Code for generating ELF core dumps.
963 typedef void (*segment_callback)(vm_map_entry_t, void *);
965 /* Closure for cb_put_phdr(). */
966 struct phdr_closure {
967 Elf_Phdr *phdr; /* Program header to fill in */
968 Elf_Off offset; /* Offset of segment in core file */
971 /* Closure for cb_size_segment(). */
972 struct sseg_closure {
973 int count; /* Count of writable segments. */
974 size_t size; /* Total size of all writable segments. */
977 static void cb_put_phdr(vm_map_entry_t, void *);
978 static void cb_size_segment(vm_map_entry_t, void *);
979 static void each_writable_segment(struct thread *, segment_callback, void *);
980 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
981 int, void *, size_t);
982 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
983 static void __elfN(putnote)(void *, size_t *, const char *, int,
984 const void *, size_t);
987 __elfN(coredump)(td, vp, limit)
992 struct ucred *cred = td->td_ucred;
994 struct sseg_closure seginfo;
998 /* Size the program segments. */
1001 each_writable_segment(td, cb_size_segment, &seginfo);
1004 * Calculate the size of the core file header area by making
1005 * a dry run of generating it. Nothing is written, but the
1006 * size is calculated.
1009 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1011 if (hdrsize + seginfo.size >= limit)
1015 * Allocate memory for building the header, fill it up,
1018 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1022 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
1024 /* Write the contents of all of the writable segments. */
1030 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1032 for (i = 0; i < seginfo.count; i++) {
1033 error = vn_rdwr_inchunks(UIO_WRITE, vp,
1034 (caddr_t)(uintptr_t)php->p_vaddr,
1035 php->p_filesz, offset, UIO_USERSPACE,
1036 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1040 offset += php->p_filesz;
1050 * A callback for each_writable_segment() to write out the segment's
1051 * program header entry.
1054 cb_put_phdr(entry, closure)
1055 vm_map_entry_t entry;
1058 struct phdr_closure *phc = (struct phdr_closure *)closure;
1059 Elf_Phdr *phdr = phc->phdr;
1061 phc->offset = round_page(phc->offset);
1063 phdr->p_type = PT_LOAD;
1064 phdr->p_offset = phc->offset;
1065 phdr->p_vaddr = entry->start;
1067 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1068 phdr->p_align = PAGE_SIZE;
1070 if (entry->protection & VM_PROT_READ)
1071 phdr->p_flags |= PF_R;
1072 if (entry->protection & VM_PROT_WRITE)
1073 phdr->p_flags |= PF_W;
1074 if (entry->protection & VM_PROT_EXECUTE)
1075 phdr->p_flags |= PF_X;
1077 phc->offset += phdr->p_filesz;
1082 * A callback for each_writable_segment() to gather information about
1083 * the number of segments and their total size.
1086 cb_size_segment(entry, closure)
1087 vm_map_entry_t entry;
1090 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1093 ssc->size += entry->end - entry->start;
1097 * For each writable segment in the process's memory map, call the given
1098 * function with a pointer to the map entry and some arbitrary
1099 * caller-supplied data.
1102 each_writable_segment(td, func, closure)
1104 segment_callback func;
1107 struct proc *p = td->td_proc;
1108 vm_map_t map = &p->p_vmspace->vm_map;
1109 vm_map_entry_t entry;
1110 vm_object_t backing_object, object;
1111 boolean_t ignore_entry;
1113 vm_map_lock_read(map);
1114 for (entry = map->header.next; entry != &map->header;
1115 entry = entry->next) {
1117 * Don't dump inaccessible mappings, deal with legacy
1120 * Note that read-only segments related to the elf binary
1121 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1122 * need to arbitrarily ignore such segments.
1124 if (elf_legacy_coredump) {
1125 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1128 if ((entry->protection & VM_PROT_ALL) == 0)
1133 * Dont include memory segment in the coredump if
1134 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1135 * madvise(2). Do not dump submaps (i.e. parts of the
1138 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1141 if ((object = entry->object.vm_object) == NULL)
1144 /* Ignore memory-mapped devices and such things. */
1145 VM_OBJECT_LOCK(object);
1146 while ((backing_object = object->backing_object) != NULL) {
1147 VM_OBJECT_LOCK(backing_object);
1148 VM_OBJECT_UNLOCK(object);
1149 object = backing_object;
1151 ignore_entry = object->type != OBJT_DEFAULT &&
1152 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1153 VM_OBJECT_UNLOCK(object);
1157 (*func)(entry, closure);
1159 vm_map_unlock_read(map);
1163 * Write the core file header to the file, including padding up to
1164 * the page boundary.
1167 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1177 /* Fill in the header. */
1178 bzero(hdr, hdrsize);
1180 __elfN(puthdr)(td, hdr, &off, numsegs);
1182 /* Write it to the core file. */
1183 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1184 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1188 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1189 typedef struct prstatus32 elf_prstatus_t;
1190 typedef struct prpsinfo32 elf_prpsinfo_t;
1191 typedef struct fpreg32 elf_prfpregset_t;
1192 typedef struct fpreg32 elf_fpregset_t;
1193 typedef struct reg32 elf_gregset_t;
1195 typedef prstatus_t elf_prstatus_t;
1196 typedef prpsinfo_t elf_prpsinfo_t;
1197 typedef prfpregset_t elf_prfpregset_t;
1198 typedef prfpregset_t elf_fpregset_t;
1199 typedef gregset_t elf_gregset_t;
1203 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1206 elf_prstatus_t status;
1207 elf_prfpregset_t fpregset;
1208 elf_prpsinfo_t psinfo;
1210 elf_prstatus_t *status;
1211 elf_prfpregset_t *fpregset;
1212 elf_prpsinfo_t *psinfo;
1215 size_t ehoff, noteoff, notesz, phoff;
1220 *off += sizeof(Elf_Ehdr);
1223 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1227 * Don't allocate space for the notes if we're just calculating
1228 * the size of the header. We also don't collect the data.
1231 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1232 status = &tempdata->status;
1233 fpregset = &tempdata->fpregset;
1234 psinfo = &tempdata->psinfo;
1243 psinfo->pr_version = PRPSINFO_VERSION;
1244 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1245 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1247 * XXX - We don't fill in the command line arguments properly
1250 strlcpy(psinfo->pr_psargs, p->p_comm,
1251 sizeof(psinfo->pr_psargs));
1253 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1257 * To have the debugger select the right thread (LWP) as the initial
1258 * thread, we dump the state of the thread passed to us in td first.
1259 * This is the thread that causes the core dump and thus likely to
1260 * be the right thread one wants to have selected in the debugger.
1263 while (thr != NULL) {
1265 status->pr_version = PRSTATUS_VERSION;
1266 status->pr_statussz = sizeof(elf_prstatus_t);
1267 status->pr_gregsetsz = sizeof(elf_gregset_t);
1268 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1269 status->pr_osreldate = osreldate;
1270 status->pr_cursig = p->p_sig;
1271 status->pr_pid = thr->td_tid;
1272 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1273 fill_regs32(thr, &status->pr_reg);
1274 fill_fpregs32(thr, fpregset);
1276 fill_regs(thr, &status->pr_reg);
1277 fill_fpregs(thr, fpregset);
1280 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1282 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1285 * Allow for MD specific notes, as well as any MD
1286 * specific preparations for writing MI notes.
1288 __elfN(dump_thread)(thr, dst, off);
1290 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1291 TAILQ_NEXT(thr, td_plist);
1293 thr = TAILQ_NEXT(thr, td_plist);
1296 notesz = *off - noteoff;
1299 free(tempdata, M_TEMP);
1301 /* Align up to a page boundary for the program segments. */
1302 *off = round_page(*off);
1307 struct phdr_closure phc;
1310 * Fill in the ELF header.
1312 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1313 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1314 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1315 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1316 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1317 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1318 ehdr->e_ident[EI_DATA] = ELF_DATA;
1319 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1320 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1321 ehdr->e_ident[EI_ABIVERSION] = 0;
1322 ehdr->e_ident[EI_PAD] = 0;
1323 ehdr->e_type = ET_CORE;
1324 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1325 ehdr->e_machine = EM_386;
1327 ehdr->e_machine = ELF_ARCH;
1329 ehdr->e_version = EV_CURRENT;
1331 ehdr->e_phoff = phoff;
1333 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1334 ehdr->e_phentsize = sizeof(Elf_Phdr);
1335 ehdr->e_phnum = numsegs + 1;
1336 ehdr->e_shentsize = sizeof(Elf_Shdr);
1338 ehdr->e_shstrndx = SHN_UNDEF;
1341 * Fill in the program header entries.
1343 phdr = (Elf_Phdr *)((char *)dst + phoff);
1345 /* The note segement. */
1346 phdr->p_type = PT_NOTE;
1347 phdr->p_offset = noteoff;
1350 phdr->p_filesz = notesz;
1356 /* All the writable segments from the program. */
1359 each_writable_segment(td, cb_put_phdr, &phc);
1364 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1365 const void *desc, size_t descsz)
1369 note.n_namesz = strlen(name) + 1;
1370 note.n_descsz = descsz;
1373 bcopy(¬e, (char *)dst + *off, sizeof note);
1374 *off += sizeof note;
1376 bcopy(name, (char *)dst + *off, note.n_namesz);
1377 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1379 bcopy(desc, (char *)dst + *off, note.n_descsz);
1380 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1384 * Try to find the appropriate ABI-note section for checknote,
1385 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1386 * first page of the image is searched, the same as for headers.
1389 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1392 const Elf_Note *note, *note0, *note_end;
1393 const Elf_Phdr *phdr, *pnote;
1394 const Elf_Ehdr *hdr;
1395 const char *note_name;
1399 hdr = (const Elf_Ehdr *)imgp->image_header;
1400 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1402 for (i = 0; i < hdr->e_phnum; i++) {
1403 if (phdr[i].p_type == PT_NOTE) {
1409 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1410 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
1413 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1414 note_end = (const Elf_Note *)(imgp->image_header +
1415 pnote->p_offset + pnote->p_filesz);
1416 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1417 if (!aligned(note, Elf32_Addr))
1419 if (note->n_namesz != checknote->hdr.n_namesz ||
1420 note->n_descsz != checknote->hdr.n_descsz ||
1421 note->n_type != checknote->hdr.n_type)
1423 note_name = (const char *)(note + 1);
1424 if (strncmp(checknote->vendor, note_name,
1425 checknote->hdr.n_namesz) != 0)
1429 * Fetch the osreldate for binary
1430 * from the ELF OSABI-note if necessary.
1432 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1433 checknote->trans_osrel != NULL)
1434 return (checknote->trans_osrel(note, osrel));
1438 note = (const Elf_Note *)((const char *)(note + 1) +
1439 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1440 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1447 * Tell kern_execve.c about it, with a little help from the linker.
1449 static struct execsw __elfN(execsw) = {
1450 __CONCAT(exec_, __elfN(imgact)),
1451 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1453 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));