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_legacy_coredump = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
102 &elf_legacy_coredump, 0, "");
104 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
106 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
107 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
108 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
111 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
115 for (i = 0; i < MAX_BRANDS; i++) {
116 if (elf_brand_list[i] == NULL) {
117 elf_brand_list[i] = entry;
127 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
131 for (i = 0; i < MAX_BRANDS; i++) {
132 if (elf_brand_list[i] == entry) {
133 elf_brand_list[i] = NULL;
143 __elfN(brand_inuse)(Elf_Brandinfo *entry)
148 sx_slock(&allproc_lock);
149 FOREACH_PROC_IN_SYSTEM(p) {
150 if (p->p_sysent == entry->sysvec) {
155 sx_sunlock(&allproc_lock);
160 static Elf_Brandinfo *
161 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
167 * We support three types of branding -- (1) the ELF EI_OSABI field
168 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
169 * branding w/in the ELF header, and (3) path of the `interp_path'
170 * field. We should also look for an ".note.ABI-tag" ELF section now
171 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
174 /* If the executable has a brand, search for it in the brand list. */
175 for (i = 0; i < MAX_BRANDS; i++) {
176 bi = elf_brand_list[i];
177 if (bi != NULL && hdr->e_machine == bi->machine &&
178 (hdr->e_ident[EI_OSABI] == bi->brand ||
179 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
180 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
184 /* Lacking a known brand, search for a recognized interpreter. */
185 if (interp != NULL) {
186 for (i = 0; i < MAX_BRANDS; i++) {
187 bi = elf_brand_list[i];
188 if (bi != NULL && hdr->e_machine == bi->machine &&
189 strcmp(interp, bi->interp_path) == 0)
194 /* Lacking a recognized interpreter, try the default brand */
195 for (i = 0; i < MAX_BRANDS; i++) {
196 bi = elf_brand_list[i];
197 if (bi != NULL && hdr->e_machine == bi->machine &&
198 __elfN(fallback_brand) == bi->brand)
205 __elfN(check_header)(const Elf_Ehdr *hdr)
211 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
212 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
213 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
214 hdr->e_phentsize != sizeof(Elf_Phdr) ||
215 hdr->e_version != ELF_TARG_VER)
219 * Make sure we have at least one brand for this machine.
222 for (i = 0; i < MAX_BRANDS; i++) {
223 bi = elf_brand_list[i];
224 if (bi != NULL && bi->machine == hdr->e_machine)
234 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
235 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
242 * Create the page if it doesn't exist yet. Ignore errors.
245 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
246 VM_PROT_ALL, VM_PROT_ALL, 0);
250 * Find the page from the underlying object.
253 sf = vm_imgact_map_page(object, offset);
255 return (KERN_FAILURE);
256 off = offset - trunc_page(offset);
257 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
259 vm_imgact_unmap_page(sf);
261 return (KERN_FAILURE);
265 return (KERN_SUCCESS);
269 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
270 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
277 if (start != trunc_page(start)) {
278 rv = __elfN(map_partial)(map, object, offset, start,
279 round_page(start), prot);
282 offset += round_page(start) - start;
283 start = round_page(start);
285 if (end != round_page(end)) {
286 rv = __elfN(map_partial)(map, object, offset +
287 trunc_page(end) - start, trunc_page(end), end, prot);
290 end = trunc_page(end);
293 if (offset & PAGE_MASK) {
295 * The mapping is not page aligned. This means we have
296 * to copy the data. Sigh.
298 rv = vm_map_find(map, NULL, 0, &start, end - start,
299 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
303 return (KERN_SUCCESS);
304 for (; start < end; start += sz) {
305 sf = vm_imgact_map_page(object, offset);
307 return (KERN_FAILURE);
308 off = offset - trunc_page(offset);
310 if (sz > PAGE_SIZE - off)
311 sz = PAGE_SIZE - off;
312 error = copyout((caddr_t)sf_buf_kva(sf) + off,
314 vm_imgact_unmap_page(sf);
316 return (KERN_FAILURE);
322 vm_object_reference(object);
324 rv = vm_map_insert(map, object, offset, start, end,
325 prot, VM_PROT_ALL, cow);
327 if (rv != KERN_SUCCESS)
328 vm_object_deallocate(object);
332 return (KERN_SUCCESS);
337 __elfN(load_section)(struct vmspace *vmspace,
338 vm_object_t object, vm_offset_t offset,
339 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
344 vm_offset_t map_addr;
347 vm_offset_t file_addr;
350 * It's necessary to fail if the filsz + offset taken from the
351 * header is greater than the actual file pager object's size.
352 * If we were to allow this, then the vm_map_find() below would
353 * walk right off the end of the file object and into the ether.
355 * While I'm here, might as well check for something else that
356 * is invalid: filsz cannot be greater than memsz.
358 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
360 uprintf("elf_load_section: truncated ELF file\n");
364 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
365 file_addr = trunc_page_ps(offset, pagesize);
368 * We have two choices. We can either clear the data in the last page
369 * of an oversized mapping, or we can start the anon mapping a page
370 * early and copy the initialized data into that first page. We
371 * choose the second..
374 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
376 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
379 /* cow flags: don't dump readonly sections in core */
380 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
381 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
383 rv = __elfN(map_insert)(&vmspace->vm_map,
385 file_addr, /* file offset */
386 map_addr, /* virtual start */
387 map_addr + map_len,/* virtual end */
390 if (rv != KERN_SUCCESS)
393 /* we can stop now if we've covered it all */
394 if (memsz == filsz) {
401 * We have to get the remaining bit of the file into the first part
402 * of the oversized map segment. This is normally because the .data
403 * segment in the file is extended to provide bss. It's a neat idea
404 * to try and save a page, but it's a pain in the behind to implement.
406 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
407 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
408 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
411 /* This had damn well better be true! */
413 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
414 map_addr + map_len, VM_PROT_ALL, 0);
415 if (rv != KERN_SUCCESS) {
423 sf = vm_imgact_map_page(object, offset + filsz);
427 /* send the page fragment to user space */
428 off = trunc_page_ps(offset + filsz, pagesize) -
429 trunc_page(offset + filsz);
430 error = copyout((caddr_t)sf_buf_kva(sf) + off,
431 (caddr_t)map_addr, copy_len);
432 vm_imgact_unmap_page(sf);
439 * set it to the specified protection.
440 * XXX had better undo the damage from pasting over the cracks here!
442 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
443 round_page(map_addr + map_len), prot, FALSE);
449 * Load the file "file" into memory. It may be either a shared object
452 * The "addr" reference parameter is in/out. On entry, it specifies
453 * the address where a shared object should be loaded. If the file is
454 * an executable, this value is ignored. On exit, "addr" specifies
455 * where the file was actually loaded.
457 * The "entry" reference parameter is out only. On exit, it specifies
458 * the entry point for the loaded file.
461 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
462 u_long *entry, size_t pagesize)
467 struct image_params image_params;
469 const Elf_Ehdr *hdr = NULL;
470 const Elf_Phdr *phdr = NULL;
471 struct nameidata *nd;
472 struct vmspace *vmspace = p->p_vmspace;
474 struct image_params *imgp;
477 u_long base_addr = 0;
478 int vfslocked, error, i, numsegs;
480 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
482 attr = &tempdata->attr;
483 imgp = &tempdata->image_params;
486 * Initialize part of the common data
490 imgp->firstpage = NULL;
491 imgp->image_header = NULL;
493 imgp->execlabel = NULL;
495 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
498 if ((error = namei(nd)) != 0) {
502 vfslocked = NDHASGIANT(nd);
503 NDFREE(nd, NDF_ONLY_PNBUF);
504 imgp->vp = nd->ni_vp;
507 * Check permissions, modes, uid, etc on the file, and "open" it.
509 error = exec_check_permissions(imgp);
513 error = exec_map_first_page(imgp);
518 * Also make certain that the interpreter stays the same, so set
519 * its VV_TEXT flag, too.
521 nd->ni_vp->v_vflag |= VV_TEXT;
523 imgp->object = nd->ni_vp->v_object;
525 hdr = (const Elf_Ehdr *)imgp->image_header;
526 if ((error = __elfN(check_header)(hdr)) != 0)
528 if (hdr->e_type == ET_DYN)
530 else if (hdr->e_type == ET_EXEC)
537 /* Only support headers that fit within first page for now */
538 /* (multiplication of two Elf_Half fields will not overflow) */
539 if ((hdr->e_phoff > PAGE_SIZE) ||
540 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
545 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
546 if (!aligned(phdr, Elf_Addr)) {
551 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
552 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
554 if (phdr[i].p_flags & PF_X)
555 prot |= VM_PROT_EXECUTE;
556 if (phdr[i].p_flags & PF_W)
557 prot |= VM_PROT_WRITE;
558 if (phdr[i].p_flags & PF_R)
559 prot |= VM_PROT_READ;
561 if ((error = __elfN(load_section)(vmspace,
562 imgp->object, phdr[i].p_offset,
563 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
564 phdr[i].p_memsz, phdr[i].p_filesz, prot,
568 * Establish the base address if this is the
572 base_addr = trunc_page(phdr[i].p_vaddr +
578 *entry = (unsigned long)hdr->e_entry + rbase;
582 exec_unmap_first_page(imgp);
587 VFS_UNLOCK_GIANT(vfslocked);
588 free(tempdata, M_TEMP);
593 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
596 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
598 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
599 const Elf_Phdr *phdr, *pnote = NULL;
600 Elf_Auxargs *elf_auxargs;
601 struct vmspace *vmspace;
603 u_long text_size = 0, data_size = 0, total_size = 0;
604 u_long text_addr = 0, data_addr = 0;
605 u_long seg_size, seg_addr;
606 u_long addr, entry = 0, proghdr = 0;
608 const char *interp = NULL, *newinterp = NULL;
609 Elf_Brandinfo *brand_info;
610 const Elf_Note *note, *note_end;
612 const char *note_name;
613 struct sysentvec *sv;
616 * Do we have a valid ELF header ?
618 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
619 * if particular brand doesn't support it.
621 if (__elfN(check_header)(hdr) != 0 ||
622 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
626 * From here on down, we return an errno, not -1, as we've
627 * detected an ELF file.
630 if ((hdr->e_phoff > PAGE_SIZE) ||
631 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
632 /* Only support headers in first page for now */
635 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
636 if (!aligned(phdr, Elf_Addr))
638 for (i = 0; i < hdr->e_phnum; i++) {
639 if (phdr[i].p_type == PT_INTERP) {
640 /* Path to interpreter */
641 if (phdr[i].p_filesz > MAXPATHLEN ||
642 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
644 interp = imgp->image_header + phdr[i].p_offset;
649 brand_info = __elfN(get_brandinfo)(hdr, interp);
650 if (brand_info == NULL) {
651 uprintf("ELF binary type \"%u\" not known.\n",
652 hdr->e_ident[EI_OSABI]);
655 if (hdr->e_type == ET_DYN &&
656 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
658 sv = brand_info->sysvec;
659 if (interp != NULL && brand_info->interp_newpath != NULL)
660 newinterp = brand_info->interp_newpath;
663 * Avoid a possible deadlock if the current address space is destroyed
664 * and that address space maps the locked vnode. In the common case,
665 * the locked vnode's v_usecount is decremented but remains greater
666 * than zero. Consequently, the vnode lock is not needed by vrele().
667 * However, in cases where the vnode lock is external, such as nullfs,
668 * v_usecount may become zero.
670 VOP_UNLOCK(imgp->vp, 0);
672 error = exec_new_vmspace(imgp, sv);
673 imgp->proc->p_sysent = sv;
675 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
679 vmspace = imgp->proc->p_vmspace;
681 for (i = 0; i < hdr->e_phnum; i++) {
682 switch (phdr[i].p_type) {
683 case PT_LOAD: /* Loadable segment */
685 if (phdr[i].p_flags & PF_X)
686 prot |= VM_PROT_EXECUTE;
687 if (phdr[i].p_flags & PF_W)
688 prot |= VM_PROT_WRITE;
689 if (phdr[i].p_flags & PF_R)
690 prot |= VM_PROT_READ;
692 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
694 * Some x86 binaries assume read == executable,
695 * notably the M3 runtime and therefore cvsup
697 if (prot & VM_PROT_READ)
698 prot |= VM_PROT_EXECUTE;
701 if ((error = __elfN(load_section)(vmspace,
702 imgp->object, phdr[i].p_offset,
703 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
704 phdr[i].p_memsz, phdr[i].p_filesz, prot,
705 sv->sv_pagesize)) != 0)
709 * If this segment contains the program headers,
710 * remember their virtual address for the AT_PHDR
711 * aux entry. Static binaries don't usually include
714 if (phdr[i].p_offset == 0 &&
715 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
717 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
719 seg_addr = trunc_page(phdr[i].p_vaddr);
720 seg_size = round_page(phdr[i].p_memsz +
721 phdr[i].p_vaddr - seg_addr);
724 * Is this .text or .data? We can't use
725 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
726 * alpha terribly and possibly does other bad
727 * things so we stick to the old way of figuring
728 * it out: If the segment contains the program
729 * entry point, it's a text segment, otherwise it
732 * Note that obreak() assumes that data_addr +
733 * data_size == end of data load area, and the ELF
734 * file format expects segments to be sorted by
735 * address. If multiple data segments exist, the
736 * last one will be used.
738 if (hdr->e_entry >= phdr[i].p_vaddr &&
739 hdr->e_entry < (phdr[i].p_vaddr +
741 text_size = seg_size;
742 text_addr = seg_addr;
743 entry = (u_long)hdr->e_entry;
745 data_size = seg_size;
746 data_addr = seg_addr;
748 total_size += seg_size;
750 case PT_PHDR: /* Program header table info */
751 proghdr = phdr[i].p_vaddr;
761 if (data_addr == 0 && data_size == 0) {
762 data_addr = text_addr;
763 data_size = text_size;
767 * Check limits. It should be safe to check the
768 * limits after loading the segments since we do
769 * not actually fault in all the segments pages.
771 PROC_LOCK(imgp->proc);
772 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
773 text_size > maxtsiz ||
774 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
775 PROC_UNLOCK(imgp->proc);
779 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
780 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
781 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
782 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
785 * We load the dynamic linker where a userland call
786 * to mmap(0, ...) would put it. The rationale behind this
787 * calculation is that it leaves room for the heap to grow to
788 * its maximum allowed size.
790 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
791 lim_max(imgp->proc, RLIMIT_DATA));
792 PROC_UNLOCK(imgp->proc);
794 imgp->entry_addr = entry;
796 if (interp != NULL) {
797 int have_interp = FALSE;
798 VOP_UNLOCK(imgp->vp, 0);
799 if (brand_info->emul_path != NULL &&
800 brand_info->emul_path[0] != '\0') {
801 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
802 snprintf(path, MAXPATHLEN, "%s%s",
803 brand_info->emul_path, interp);
804 error = __elfN(load_file)(imgp->proc, path, &addr,
805 &imgp->entry_addr, sv->sv_pagesize);
810 if (!have_interp && newinterp != NULL) {
811 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
812 &imgp->entry_addr, sv->sv_pagesize);
817 error = __elfN(load_file)(imgp->proc, interp, &addr,
818 &imgp->entry_addr, sv->sv_pagesize);
820 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
822 uprintf("ELF interpreter %s not found\n", interp);
829 * Construct auxargs table (used by the fixup routine)
831 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
832 elf_auxargs->execfd = -1;
833 elf_auxargs->phdr = proghdr;
834 elf_auxargs->phent = hdr->e_phentsize;
835 elf_auxargs->phnum = hdr->e_phnum;
836 elf_auxargs->pagesz = PAGE_SIZE;
837 elf_auxargs->base = addr;
838 elf_auxargs->flags = 0;
839 elf_auxargs->entry = entry;
841 imgp->auxargs = elf_auxargs;
842 imgp->interpreted = 0;
845 * Try to fetch the osreldate for FreeBSD binary from the ELF
846 * OSABI-note. Only the first page of the image is searched,
847 * the same as for headers.
849 if (pnote != NULL && pnote->p_offset < PAGE_SIZE &&
850 pnote->p_offset + pnote->p_filesz < PAGE_SIZE ) {
851 note = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
852 if (!aligned(note, Elf32_Addr)) {
853 free(imgp->auxargs, M_TEMP);
854 imgp->auxargs = NULL;
857 note_end = (const Elf_Note *)(imgp->image_header + pnote->p_offset +
859 while (note < note_end) {
860 if (note->n_namesz == sizeof(FREEBSD_ABI_VENDOR) &&
861 note->n_descsz == sizeof(int32_t) &&
862 note->n_type == 1 /* ABI_NOTETYPE */) {
863 note_name = (const char *)(note + 1);
864 if (strncmp(FREEBSD_ABI_VENDOR, note_name,
865 sizeof(FREEBSD_ABI_VENDOR)) == 0) {
866 imgp->proc->p_osrel = *(const int32_t *)
868 round_page_ps(sizeof(FREEBSD_ABI_VENDOR),
869 sizeof(Elf32_Addr)));
873 note = (const Elf_Note *)((const char *)(note + 1) +
874 round_page_ps(note->n_namesz, sizeof(Elf32_Addr)) +
875 round_page_ps(note->n_descsz, sizeof(Elf32_Addr)));
882 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
885 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
887 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
891 base = (Elf_Addr *)*stack_base;
892 pos = base + (imgp->args->argc + imgp->args->envc + 2);
894 if (args->execfd != -1)
895 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
896 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
897 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
898 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
899 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
900 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
901 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
902 AUXARGS_ENTRY(pos, AT_BASE, args->base);
903 AUXARGS_ENTRY(pos, AT_NULL, 0);
905 free(imgp->auxargs, M_TEMP);
906 imgp->auxargs = NULL;
909 suword(base, (long)imgp->args->argc);
910 *stack_base = (register_t *)base;
915 * Code for generating ELF core dumps.
918 typedef void (*segment_callback)(vm_map_entry_t, void *);
920 /* Closure for cb_put_phdr(). */
921 struct phdr_closure {
922 Elf_Phdr *phdr; /* Program header to fill in */
923 Elf_Off offset; /* Offset of segment in core file */
926 /* Closure for cb_size_segment(). */
927 struct sseg_closure {
928 int count; /* Count of writable segments. */
929 size_t size; /* Total size of all writable segments. */
932 static void cb_put_phdr(vm_map_entry_t, void *);
933 static void cb_size_segment(vm_map_entry_t, void *);
934 static void each_writable_segment(struct thread *, segment_callback, void *);
935 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
936 int, void *, size_t);
937 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
938 static void __elfN(putnote)(void *, size_t *, const char *, int,
939 const void *, size_t);
942 __elfN(coredump)(td, vp, limit)
947 struct ucred *cred = td->td_ucred;
949 struct sseg_closure seginfo;
953 /* Size the program segments. */
956 each_writable_segment(td, cb_size_segment, &seginfo);
959 * Calculate the size of the core file header area by making
960 * a dry run of generating it. Nothing is written, but the
961 * size is calculated.
964 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
966 if (hdrsize + seginfo.size >= limit)
970 * Allocate memory for building the header, fill it up,
973 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
977 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
979 /* Write the contents of all of the writable segments. */
985 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
987 for (i = 0; i < seginfo.count; i++) {
988 error = vn_rdwr_inchunks(UIO_WRITE, vp,
989 (caddr_t)(uintptr_t)php->p_vaddr,
990 php->p_filesz, offset, UIO_USERSPACE,
991 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
995 offset += php->p_filesz;
1005 * A callback for each_writable_segment() to write out the segment's
1006 * program header entry.
1009 cb_put_phdr(entry, closure)
1010 vm_map_entry_t entry;
1013 struct phdr_closure *phc = (struct phdr_closure *)closure;
1014 Elf_Phdr *phdr = phc->phdr;
1016 phc->offset = round_page(phc->offset);
1018 phdr->p_type = PT_LOAD;
1019 phdr->p_offset = phc->offset;
1020 phdr->p_vaddr = entry->start;
1022 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1023 phdr->p_align = PAGE_SIZE;
1025 if (entry->protection & VM_PROT_READ)
1026 phdr->p_flags |= PF_R;
1027 if (entry->protection & VM_PROT_WRITE)
1028 phdr->p_flags |= PF_W;
1029 if (entry->protection & VM_PROT_EXECUTE)
1030 phdr->p_flags |= PF_X;
1032 phc->offset += phdr->p_filesz;
1037 * A callback for each_writable_segment() to gather information about
1038 * the number of segments and their total size.
1041 cb_size_segment(entry, closure)
1042 vm_map_entry_t entry;
1045 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1048 ssc->size += entry->end - entry->start;
1052 * For each writable segment in the process's memory map, call the given
1053 * function with a pointer to the map entry and some arbitrary
1054 * caller-supplied data.
1057 each_writable_segment(td, func, closure)
1059 segment_callback func;
1062 struct proc *p = td->td_proc;
1063 vm_map_t map = &p->p_vmspace->vm_map;
1064 vm_map_entry_t entry;
1065 vm_object_t backing_object, object;
1066 boolean_t ignore_entry;
1068 vm_map_lock_read(map);
1069 for (entry = map->header.next; entry != &map->header;
1070 entry = entry->next) {
1072 * Don't dump inaccessible mappings, deal with legacy
1075 * Note that read-only segments related to the elf binary
1076 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1077 * need to arbitrarily ignore such segments.
1079 if (elf_legacy_coredump) {
1080 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1083 if ((entry->protection & VM_PROT_ALL) == 0)
1088 * Dont include memory segment in the coredump if
1089 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1090 * madvise(2). Do not dump submaps (i.e. parts of the
1093 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1096 if ((object = entry->object.vm_object) == NULL)
1099 /* Ignore memory-mapped devices and such things. */
1100 VM_OBJECT_LOCK(object);
1101 while ((backing_object = object->backing_object) != NULL) {
1102 VM_OBJECT_LOCK(backing_object);
1103 VM_OBJECT_UNLOCK(object);
1104 object = backing_object;
1106 ignore_entry = object->type != OBJT_DEFAULT &&
1107 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1108 VM_OBJECT_UNLOCK(object);
1112 (*func)(entry, closure);
1114 vm_map_unlock_read(map);
1118 * Write the core file header to the file, including padding up to
1119 * the page boundary.
1122 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1132 /* Fill in the header. */
1133 bzero(hdr, hdrsize);
1135 __elfN(puthdr)(td, hdr, &off, numsegs);
1137 /* Write it to the core file. */
1138 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1139 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1143 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1144 typedef struct prstatus32 elf_prstatus_t;
1145 typedef struct prpsinfo32 elf_prpsinfo_t;
1146 typedef struct fpreg32 elf_prfpregset_t;
1147 typedef struct fpreg32 elf_fpregset_t;
1148 typedef struct reg32 elf_gregset_t;
1150 typedef prstatus_t elf_prstatus_t;
1151 typedef prpsinfo_t elf_prpsinfo_t;
1152 typedef prfpregset_t elf_prfpregset_t;
1153 typedef prfpregset_t elf_fpregset_t;
1154 typedef gregset_t elf_gregset_t;
1158 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1161 elf_prstatus_t status;
1162 elf_prfpregset_t fpregset;
1163 elf_prpsinfo_t psinfo;
1165 elf_prstatus_t *status;
1166 elf_prfpregset_t *fpregset;
1167 elf_prpsinfo_t *psinfo;
1170 size_t ehoff, noteoff, notesz, phoff;
1175 *off += sizeof(Elf_Ehdr);
1178 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1182 * Don't allocate space for the notes if we're just calculating
1183 * the size of the header. We also don't collect the data.
1186 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1187 status = &tempdata->status;
1188 fpregset = &tempdata->fpregset;
1189 psinfo = &tempdata->psinfo;
1198 psinfo->pr_version = PRPSINFO_VERSION;
1199 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1200 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1202 * XXX - We don't fill in the command line arguments properly
1205 strlcpy(psinfo->pr_psargs, p->p_comm,
1206 sizeof(psinfo->pr_psargs));
1208 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1212 * To have the debugger select the right thread (LWP) as the initial
1213 * thread, we dump the state of the thread passed to us in td first.
1214 * This is the thread that causes the core dump and thus likely to
1215 * be the right thread one wants to have selected in the debugger.
1218 while (thr != NULL) {
1220 status->pr_version = PRSTATUS_VERSION;
1221 status->pr_statussz = sizeof(elf_prstatus_t);
1222 status->pr_gregsetsz = sizeof(elf_gregset_t);
1223 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1224 status->pr_osreldate = osreldate;
1225 status->pr_cursig = p->p_sig;
1226 status->pr_pid = thr->td_tid;
1227 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1228 fill_regs32(thr, &status->pr_reg);
1229 fill_fpregs32(thr, fpregset);
1231 fill_regs(thr, &status->pr_reg);
1232 fill_fpregs(thr, fpregset);
1235 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1237 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1240 * Allow for MD specific notes, as well as any MD
1241 * specific preparations for writing MI notes.
1243 __elfN(dump_thread)(thr, dst, off);
1245 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1246 TAILQ_NEXT(thr, td_plist);
1248 thr = TAILQ_NEXT(thr, td_plist);
1251 notesz = *off - noteoff;
1254 free(tempdata, M_TEMP);
1256 /* Align up to a page boundary for the program segments. */
1257 *off = round_page(*off);
1262 struct phdr_closure phc;
1265 * Fill in the ELF header.
1267 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1268 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1269 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1270 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1271 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1272 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1273 ehdr->e_ident[EI_DATA] = ELF_DATA;
1274 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1275 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1276 ehdr->e_ident[EI_ABIVERSION] = 0;
1277 ehdr->e_ident[EI_PAD] = 0;
1278 ehdr->e_type = ET_CORE;
1279 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1280 ehdr->e_machine = EM_386;
1282 ehdr->e_machine = ELF_ARCH;
1284 ehdr->e_version = EV_CURRENT;
1286 ehdr->e_phoff = phoff;
1288 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1289 ehdr->e_phentsize = sizeof(Elf_Phdr);
1290 ehdr->e_phnum = numsegs + 1;
1291 ehdr->e_shentsize = sizeof(Elf_Shdr);
1293 ehdr->e_shstrndx = SHN_UNDEF;
1296 * Fill in the program header entries.
1298 phdr = (Elf_Phdr *)((char *)dst + phoff);
1300 /* The note segement. */
1301 phdr->p_type = PT_NOTE;
1302 phdr->p_offset = noteoff;
1305 phdr->p_filesz = notesz;
1311 /* All the writable segments from the program. */
1314 each_writable_segment(td, cb_put_phdr, &phc);
1319 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1320 const void *desc, size_t descsz)
1324 note.n_namesz = strlen(name) + 1;
1325 note.n_descsz = descsz;
1328 bcopy(¬e, (char *)dst + *off, sizeof note);
1329 *off += sizeof note;
1331 bcopy(name, (char *)dst + *off, note.n_namesz);
1332 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1334 bcopy(desc, (char *)dst + *off, note.n_descsz);
1335 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1339 * Tell kern_execve.c about it, with a little help from the linker.
1341 static struct execsw __elfN(execsw) = {
1342 __CONCAT(exec_, __elfN(imgact)),
1343 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1345 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));