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/systm.h>
53 #include <sys/signalvar.h>
56 #include <sys/syscall.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/vnode.h>
62 #include <vm/vm_kern.h>
63 #include <vm/vm_param.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
69 #include <machine/elf.h>
70 #include <machine/md_var.h>
72 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
73 #include <machine/fpu.h>
74 #include <compat/ia32/ia32_reg.h>
77 #define OLD_EI_BRAND 8
79 static int __elfN(check_header)(const Elf_Ehdr *hdr);
80 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
82 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
83 u_long *entry, size_t pagesize);
84 static int __elfN(load_section)(struct proc *p,
85 struct vmspace *vmspace, struct vnode *vp, 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];
110 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
114 for (i = 0; i < MAX_BRANDS; i++) {
115 if (elf_brand_list[i] == NULL) {
116 elf_brand_list[i] = entry;
126 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
130 for (i = 0; i < MAX_BRANDS; i++) {
131 if (elf_brand_list[i] == entry) {
132 elf_brand_list[i] = NULL;
142 __elfN(brand_inuse)(Elf_Brandinfo *entry)
147 sx_slock(&allproc_lock);
148 LIST_FOREACH(p, &allproc, p_list) {
149 if (p->p_sysent == entry->sysvec) {
154 sx_sunlock(&allproc_lock);
159 static Elf_Brandinfo *
160 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
166 * We support three types of branding -- (1) the ELF EI_OSABI field
167 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
168 * branding w/in the ELF header, and (3) path of the `interp_path'
169 * field. We should also look for an ".note.ABI-tag" ELF section now
170 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
173 /* If the executable has a brand, search for it in the brand list. */
174 for (i = 0; i < MAX_BRANDS; i++) {
175 bi = elf_brand_list[i];
176 if (bi != NULL && hdr->e_machine == bi->machine &&
177 (hdr->e_ident[EI_OSABI] == bi->brand ||
178 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
179 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
183 /* Lacking a known brand, search for a recognized interpreter. */
184 if (interp != NULL) {
185 for (i = 0; i < MAX_BRANDS; i++) {
186 bi = elf_brand_list[i];
187 if (bi != NULL && hdr->e_machine == bi->machine &&
188 strcmp(interp, bi->interp_path) == 0)
193 /* Lacking a recognized interpreter, try the default brand */
194 for (i = 0; i < MAX_BRANDS; i++) {
195 bi = elf_brand_list[i];
196 if (bi != NULL && hdr->e_machine == bi->machine &&
197 __elfN(fallback_brand) == bi->brand)
204 __elfN(check_header)(const Elf_Ehdr *hdr)
210 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
211 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
212 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
213 hdr->e_phentsize != sizeof(Elf_Phdr) ||
214 hdr->e_version != ELF_TARG_VER)
218 * Make sure we have at least one brand for this machine.
221 for (i = 0; i < MAX_BRANDS; i++) {
222 bi = elf_brand_list[i];
223 if (bi != NULL && bi->machine == hdr->e_machine)
233 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
234 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
239 vm_offset_t data_buf = 0;
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), max,
250 * Find the page from the underlying object.
253 vm_object_reference(object);
254 rv = vm_map_find(exec_map,
262 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
263 if (rv != KERN_SUCCESS) {
264 vm_object_deallocate(object);
268 off = offset - trunc_page(offset);
269 error = copyout((caddr_t)data_buf + off, (caddr_t)start,
271 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
273 return (KERN_FAILURE);
277 return (KERN_SUCCESS);
281 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
282 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
283 vm_prot_t max, int cow)
285 vm_offset_t data_buf, off;
289 if (start != trunc_page(start)) {
290 rv = __elfN(map_partial)(map, object, offset, start,
291 round_page(start), prot, max);
294 offset += round_page(start) - start;
295 start = round_page(start);
297 if (end != round_page(end)) {
298 rv = __elfN(map_partial)(map, object, offset +
299 trunc_page(end) - start, trunc_page(end), end, prot, max);
302 end = trunc_page(end);
305 if (offset & PAGE_MASK) {
307 * The mapping is not page aligned. This means we have
308 * to copy the data. Sigh.
310 rv = vm_map_find(map, 0, 0, &start, end - start,
311 FALSE, prot, max, 0);
315 while (start < end) {
316 vm_object_reference(object);
317 rv = vm_map_find(exec_map,
326 | MAP_PREFAULT_PARTIAL));
327 if (rv != KERN_SUCCESS) {
328 vm_object_deallocate(object);
331 off = offset - trunc_page(offset);
335 error = copyout((caddr_t)data_buf + off,
337 vm_map_remove(exec_map, data_buf,
338 data_buf + 2 * PAGE_SIZE);
340 return (KERN_FAILURE);
347 rv = vm_map_insert(map, object, offset, start, end,
353 return (KERN_SUCCESS);
358 __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
359 struct vnode *vp, vm_object_t object, vm_offset_t offset,
360 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
364 vm_offset_t map_addr;
367 vm_offset_t file_addr;
368 vm_offset_t data_buf = 0;
373 * It's necessary to fail if the filsz + offset taken from the
374 * header is greater than the actual file pager object's size.
375 * If we were to allow this, then the vm_map_find() below would
376 * walk right off the end of the file object and into the ether.
378 * While I'm here, might as well check for something else that
379 * is invalid: filsz cannot be greater than memsz.
381 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
383 uprintf("elf_load_section: truncated ELF file\n");
387 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
388 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
390 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
391 file_addr = trunc_page_ps(offset, pagesize);
394 * We have two choices. We can either clear the data in the last page
395 * of an oversized mapping, or we can start the anon mapping a page
396 * early and copy the initialized data into that first page. We
397 * choose the second..
400 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
402 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
405 vm_object_reference(object);
407 /* cow flags: don't dump readonly sections in core */
408 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
409 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
411 rv = __elfN(map_insert)(&vmspace->vm_map,
413 file_addr, /* file offset */
414 map_addr, /* virtual start */
415 map_addr + map_len,/* virtual end */
419 if (rv != KERN_SUCCESS) {
420 vm_object_deallocate(object);
424 /* we can stop now if we've covered it all */
425 if (memsz == filsz) {
432 * We have to get the remaining bit of the file into the first part
433 * of the oversized map segment. This is normally because the .data
434 * segment in the file is extended to provide bss. It's a neat idea
435 * to try and save a page, but it's a pain in the behind to implement.
437 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
438 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
439 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
442 /* This had damn well better be true! */
444 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
445 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0);
446 if (rv != KERN_SUCCESS) {
453 vm_object_reference(object);
454 rv = vm_map_find(exec_map,
456 trunc_page(offset + filsz),
462 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
463 if (rv != KERN_SUCCESS) {
464 vm_object_deallocate(object);
468 /* send the page fragment to user space */
469 off = trunc_page_ps(offset + filsz, pagesize) -
470 trunc_page(offset + filsz);
471 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr,
473 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
480 * set it to the specified protection.
481 * XXX had better undo the damage from pasting over the cracks here!
483 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
484 round_page(map_addr + map_len), prot, FALSE);
490 * Load the file "file" into memory. It may be either a shared object
493 * The "addr" reference parameter is in/out. On entry, it specifies
494 * the address where a shared object should be loaded. If the file is
495 * an executable, this value is ignored. On exit, "addr" specifies
496 * where the file was actually loaded.
498 * The "entry" reference parameter is out only. On exit, it specifies
499 * the entry point for the loaded file.
502 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
503 u_long *entry, size_t pagesize)
508 struct image_params image_params;
510 const Elf_Ehdr *hdr = NULL;
511 const Elf_Phdr *phdr = NULL;
512 struct nameidata *nd;
513 struct vmspace *vmspace = p->p_vmspace;
515 struct image_params *imgp;
518 u_long base_addr = 0;
519 int vfslocked, error, i, numsegs;
521 if (curthread->td_proc != p)
522 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
524 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
526 attr = &tempdata->attr;
527 imgp = &tempdata->image_params;
530 * Initialize part of the common data
534 imgp->firstpage = NULL;
535 imgp->image_header = NULL;
537 imgp->execlabel = NULL;
540 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
543 if ((error = namei(nd)) != 0) {
547 vfslocked = NDHASGIANT(nd);
548 NDFREE(nd, NDF_ONLY_PNBUF);
549 imgp->vp = nd->ni_vp;
552 * Check permissions, modes, uid, etc on the file, and "open" it.
554 error = exec_check_permissions(imgp);
556 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
560 error = exec_map_first_page(imgp);
562 * Also make certain that the interpreter stays the same, so set
563 * its VV_TEXT flag, too.
566 nd->ni_vp->v_vflag |= VV_TEXT;
568 imgp->object = nd->ni_vp->v_object;
569 vm_object_reference(imgp->object);
571 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
575 hdr = (const Elf_Ehdr *)imgp->image_header;
576 if ((error = __elfN(check_header)(hdr)) != 0)
578 if (hdr->e_type == ET_DYN)
580 else if (hdr->e_type == ET_EXEC)
587 /* Only support headers that fit within first page for now */
588 /* (multiplication of two Elf_Half fields will not overflow) */
589 if ((hdr->e_phoff > PAGE_SIZE) ||
590 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
595 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
597 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
598 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
600 if (phdr[i].p_flags & PF_X)
601 prot |= VM_PROT_EXECUTE;
602 if (phdr[i].p_flags & PF_W)
603 prot |= VM_PROT_WRITE;
604 if (phdr[i].p_flags & PF_R)
605 prot |= VM_PROT_READ;
607 if ((error = __elfN(load_section)(p, vmspace,
608 nd->ni_vp, imgp->object, phdr[i].p_offset,
609 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
610 phdr[i].p_memsz, phdr[i].p_filesz, prot,
614 * Establish the base address if this is the
618 base_addr = trunc_page(phdr[i].p_vaddr +
624 *entry = (unsigned long)hdr->e_entry + rbase;
628 exec_unmap_first_page(imgp);
630 vm_object_deallocate(imgp->object);
635 VFS_UNLOCK_GIANT(vfslocked);
636 free(tempdata, M_TEMP);
642 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
644 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
645 const Elf_Phdr *phdr;
646 Elf_Auxargs *elf_auxargs = NULL;
647 struct vmspace *vmspace;
649 u_long text_size = 0, data_size = 0, total_size = 0;
650 u_long text_addr = 0, data_addr = 0;
651 u_long seg_size, seg_addr;
652 u_long addr, entry = 0, proghdr = 0;
654 const char *interp = NULL;
655 Elf_Brandinfo *brand_info;
657 struct thread *td = curthread;
658 struct sysentvec *sv;
661 * Do we have a valid ELF header ?
663 if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC)
667 * From here on down, we return an errno, not -1, as we've
668 * detected an ELF file.
671 if ((hdr->e_phoff > PAGE_SIZE) ||
672 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
673 /* Only support headers in first page for now */
676 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
679 * From this point on, we may have resources that need to be freed.
682 VOP_UNLOCK(imgp->vp, 0, td);
684 for (i = 0; i < hdr->e_phnum; i++) {
685 switch (phdr[i].p_type) {
686 case PT_INTERP: /* Path to interpreter */
687 if (phdr[i].p_filesz > MAXPATHLEN ||
688 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
692 interp = imgp->image_header + phdr[i].p_offset;
699 brand_info = __elfN(get_brandinfo)(hdr, interp);
700 if (brand_info == NULL) {
701 uprintf("ELF binary type \"%u\" not known.\n",
702 hdr->e_ident[EI_OSABI]);
706 sv = brand_info->sysvec;
707 if (interp != NULL && brand_info->interp_newpath != NULL)
708 interp = brand_info->interp_newpath;
710 exec_new_vmspace(imgp, sv);
712 vmspace = imgp->proc->p_vmspace;
714 for (i = 0; i < hdr->e_phnum; i++) {
715 switch (phdr[i].p_type) {
716 case PT_LOAD: /* Loadable segment */
718 if (phdr[i].p_flags & PF_X)
719 prot |= VM_PROT_EXECUTE;
720 if (phdr[i].p_flags & PF_W)
721 prot |= VM_PROT_WRITE;
722 if (phdr[i].p_flags & PF_R)
723 prot |= VM_PROT_READ;
725 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
727 * Some x86 binaries assume read == executable,
728 * notably the M3 runtime and therefore cvsup
730 if (prot & VM_PROT_READ)
731 prot |= VM_PROT_EXECUTE;
734 if ((error = __elfN(load_section)(imgp->proc, vmspace,
735 imgp->vp, imgp->object, phdr[i].p_offset,
736 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
737 phdr[i].p_memsz, phdr[i].p_filesz, prot,
738 sv->sv_pagesize)) != 0)
742 * If this segment contains the program headers,
743 * remember their virtual address for the AT_PHDR
744 * aux entry. Static binaries don't usually include
747 if (phdr[i].p_offset == 0 &&
748 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
750 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
752 seg_addr = trunc_page(phdr[i].p_vaddr);
753 seg_size = round_page(phdr[i].p_memsz +
754 phdr[i].p_vaddr - seg_addr);
757 * Is this .text or .data? We can't use
758 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
759 * alpha terribly and possibly does other bad
760 * things so we stick to the old way of figuring
761 * it out: If the segment contains the program
762 * entry point, it's a text segment, otherwise it
765 * Note that obreak() assumes that data_addr +
766 * data_size == end of data load area, and the ELF
767 * file format expects segments to be sorted by
768 * address. If multiple data segments exist, the
769 * last one will be used.
771 if (hdr->e_entry >= phdr[i].p_vaddr &&
772 hdr->e_entry < (phdr[i].p_vaddr +
774 text_size = seg_size;
775 text_addr = seg_addr;
776 entry = (u_long)hdr->e_entry;
778 data_size = seg_size;
779 data_addr = seg_addr;
781 total_size += seg_size;
783 case PT_PHDR: /* Program header table info */
784 proghdr = phdr[i].p_vaddr;
791 if (data_addr == 0 && data_size == 0) {
792 data_addr = text_addr;
793 data_size = text_size;
797 * Check limits. It should be safe to check the
798 * limits after loading the segments since we do
799 * not actually fault in all the segments pages.
801 PROC_LOCK(imgp->proc);
802 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
803 text_size > maxtsiz ||
804 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
805 PROC_UNLOCK(imgp->proc);
810 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
811 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
812 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
813 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
816 * We load the dynamic linker where a userland call
817 * to mmap(0, ...) would put it. The rationale behind this
818 * calculation is that it leaves room for the heap to grow to
819 * its maximum allowed size.
821 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
822 lim_max(imgp->proc, RLIMIT_DATA));
823 PROC_UNLOCK(imgp->proc);
825 imgp->entry_addr = entry;
827 imgp->proc->p_sysent = sv;
828 if (interp != NULL && brand_info->emul_path != NULL &&
829 brand_info->emul_path[0] != '\0') {
830 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
831 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path,
833 error = __elfN(load_file)(imgp->proc, path, &addr,
834 &imgp->entry_addr, sv->sv_pagesize);
839 if (interp != NULL) {
840 error = __elfN(load_file)(imgp->proc, interp, &addr,
841 &imgp->entry_addr, sv->sv_pagesize);
843 uprintf("ELF interpreter %s not found\n", interp);
849 * Construct auxargs table (used by the fixup routine)
851 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
852 elf_auxargs->execfd = -1;
853 elf_auxargs->phdr = proghdr;
854 elf_auxargs->phent = hdr->e_phentsize;
855 elf_auxargs->phnum = hdr->e_phnum;
856 elf_auxargs->pagesz = PAGE_SIZE;
857 elf_auxargs->base = addr;
858 elf_auxargs->flags = 0;
859 elf_auxargs->entry = entry;
860 elf_auxargs->trace = elf_trace;
862 imgp->auxargs = elf_auxargs;
863 imgp->interpreted = 0;
866 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
870 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
873 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
875 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
879 base = (Elf_Addr *)*stack_base;
880 pos = base + (imgp->args->argc + imgp->args->envc + 2);
883 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
885 if (args->execfd != -1) {
886 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
888 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
889 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
890 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
891 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
892 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
893 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
894 AUXARGS_ENTRY(pos, AT_BASE, args->base);
895 AUXARGS_ENTRY(pos, AT_NULL, 0);
897 free(imgp->auxargs, M_TEMP);
898 imgp->auxargs = NULL;
901 suword(base, (long)imgp->args->argc);
902 *stack_base = (register_t *)base;
907 * Code for generating ELF core dumps.
910 typedef void (*segment_callback)(vm_map_entry_t, void *);
912 /* Closure for cb_put_phdr(). */
913 struct phdr_closure {
914 Elf_Phdr *phdr; /* Program header to fill in */
915 Elf_Off offset; /* Offset of segment in core file */
918 /* Closure for cb_size_segment(). */
919 struct sseg_closure {
920 int count; /* Count of writable segments. */
921 size_t size; /* Total size of all writable segments. */
924 static void cb_put_phdr(vm_map_entry_t, void *);
925 static void cb_size_segment(vm_map_entry_t, void *);
926 static void each_writable_segment(struct thread *, segment_callback, void *);
927 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
928 int, void *, size_t);
929 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
930 static void __elfN(putnote)(void *, size_t *, const char *, int,
931 const void *, size_t);
933 extern int osreldate;
936 __elfN(coredump)(td, vp, limit)
941 struct ucred *cred = td->td_ucred;
943 struct sseg_closure seginfo;
947 /* Size the program segments. */
950 each_writable_segment(td, cb_size_segment, &seginfo);
953 * Calculate the size of the core file header area by making
954 * a dry run of generating it. Nothing is written, but the
955 * size is calculated.
958 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
960 if (hdrsize + seginfo.size >= limit)
964 * Allocate memory for building the header, fill it up,
967 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
971 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
973 /* Write the contents of all of the writable segments. */
979 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
981 for (i = 0; i < seginfo.count; i++) {
982 error = vn_rdwr_inchunks(UIO_WRITE, vp,
983 (caddr_t)(uintptr_t)php->p_vaddr,
984 php->p_filesz, offset, UIO_USERSPACE,
985 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
986 curthread); /* XXXKSE */
989 offset += php->p_filesz;
999 * A callback for each_writable_segment() to write out the segment's
1000 * program header entry.
1003 cb_put_phdr(entry, closure)
1004 vm_map_entry_t entry;
1007 struct phdr_closure *phc = (struct phdr_closure *)closure;
1008 Elf_Phdr *phdr = phc->phdr;
1010 phc->offset = round_page(phc->offset);
1012 phdr->p_type = PT_LOAD;
1013 phdr->p_offset = phc->offset;
1014 phdr->p_vaddr = entry->start;
1016 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1017 phdr->p_align = PAGE_SIZE;
1019 if (entry->protection & VM_PROT_READ)
1020 phdr->p_flags |= PF_R;
1021 if (entry->protection & VM_PROT_WRITE)
1022 phdr->p_flags |= PF_W;
1023 if (entry->protection & VM_PROT_EXECUTE)
1024 phdr->p_flags |= PF_X;
1026 phc->offset += phdr->p_filesz;
1031 * A callback for each_writable_segment() to gather information about
1032 * the number of segments and their total size.
1035 cb_size_segment(entry, closure)
1036 vm_map_entry_t entry;
1039 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1042 ssc->size += entry->end - entry->start;
1046 * For each writable segment in the process's memory map, call the given
1047 * function with a pointer to the map entry and some arbitrary
1048 * caller-supplied data.
1051 each_writable_segment(td, func, closure)
1053 segment_callback func;
1056 struct proc *p = td->td_proc;
1057 vm_map_t map = &p->p_vmspace->vm_map;
1058 vm_map_entry_t entry;
1060 for (entry = map->header.next; entry != &map->header;
1061 entry = entry->next) {
1065 * Don't dump inaccessible mappings, deal with legacy
1068 * Note that read-only segments related to the elf binary
1069 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1070 * need to arbitrarily ignore such segments.
1072 if (elf_legacy_coredump) {
1073 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1076 if ((entry->protection & VM_PROT_ALL) == 0)
1081 * Dont include memory segment in the coredump if
1082 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1083 * madvise(2). Do not dump submaps (i.e. parts of the
1086 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1089 if ((obj = entry->object.vm_object) == NULL)
1092 /* Find the deepest backing object. */
1093 while (obj->backing_object != NULL)
1094 obj = obj->backing_object;
1096 /* Ignore memory-mapped devices and such things. */
1097 if (obj->type != OBJT_DEFAULT &&
1098 obj->type != OBJT_SWAP &&
1099 obj->type != OBJT_VNODE)
1102 (*func)(entry, closure);
1107 * Write the core file header to the file, including padding up to
1108 * the page boundary.
1111 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1121 /* Fill in the header. */
1122 bzero(hdr, hdrsize);
1124 __elfN(puthdr)(td, hdr, &off, numsegs);
1126 /* Write it to the core file. */
1127 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1128 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1132 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1133 typedef struct prstatus32 elf_prstatus_t;
1134 typedef struct prpsinfo32 elf_prpsinfo_t;
1135 typedef struct fpreg32 elf_prfpregset_t;
1136 typedef struct fpreg32 elf_fpregset_t;
1137 typedef struct reg32 elf_gregset_t;
1139 typedef prstatus_t elf_prstatus_t;
1140 typedef prpsinfo_t elf_prpsinfo_t;
1141 typedef prfpregset_t elf_prfpregset_t;
1142 typedef prfpregset_t elf_fpregset_t;
1143 typedef gregset_t elf_gregset_t;
1147 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1150 elf_prstatus_t status;
1151 elf_prfpregset_t fpregset;
1152 elf_prpsinfo_t psinfo;
1154 elf_prstatus_t *status;
1155 elf_prfpregset_t *fpregset;
1156 elf_prpsinfo_t *psinfo;
1159 size_t ehoff, noteoff, notesz, phoff;
1164 *off += sizeof(Elf_Ehdr);
1167 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1171 * Don't allocate space for the notes if we're just calculating
1172 * the size of the header. We also don't collect the data.
1175 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1176 status = &tempdata->status;
1177 fpregset = &tempdata->fpregset;
1178 psinfo = &tempdata->psinfo;
1187 psinfo->pr_version = PRPSINFO_VERSION;
1188 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1189 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1191 * XXX - We don't fill in the command line arguments properly
1194 strlcpy(psinfo->pr_psargs, p->p_comm,
1195 sizeof(psinfo->pr_psargs));
1197 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1201 * To have the debugger select the right thread (LWP) as the initial
1202 * thread, we dump the state of the thread passed to us in td first.
1203 * This is the thread that causes the core dump and thus likely to
1204 * be the right thread one wants to have selected in the debugger.
1207 while (thr != NULL) {
1209 status->pr_version = PRSTATUS_VERSION;
1210 status->pr_statussz = sizeof(elf_prstatus_t);
1211 status->pr_gregsetsz = sizeof(elf_gregset_t);
1212 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1213 status->pr_osreldate = osreldate;
1214 status->pr_cursig = p->p_sig;
1215 status->pr_pid = thr->td_tid;
1216 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1217 fill_regs32(thr, &status->pr_reg);
1218 fill_fpregs32(thr, fpregset);
1220 fill_regs(thr, &status->pr_reg);
1221 fill_fpregs(thr, fpregset);
1224 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1226 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1229 * Allow for MD specific notes, as well as any MD
1230 * specific preparations for writing MI notes.
1232 __elfN(dump_thread)(thr, dst, off);
1234 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1235 TAILQ_NEXT(thr, td_plist);
1237 thr = TAILQ_NEXT(thr, td_plist);
1240 notesz = *off - noteoff;
1243 free(tempdata, M_TEMP);
1245 /* Align up to a page boundary for the program segments. */
1246 *off = round_page(*off);
1251 struct phdr_closure phc;
1254 * Fill in the ELF header.
1256 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1257 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1258 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1259 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1260 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1261 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1262 ehdr->e_ident[EI_DATA] = ELF_DATA;
1263 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1264 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1265 ehdr->e_ident[EI_ABIVERSION] = 0;
1266 ehdr->e_ident[EI_PAD] = 0;
1267 ehdr->e_type = ET_CORE;
1268 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1269 ehdr->e_machine = EM_386;
1271 ehdr->e_machine = ELF_ARCH;
1273 ehdr->e_version = EV_CURRENT;
1275 ehdr->e_phoff = phoff;
1277 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1278 ehdr->e_phentsize = sizeof(Elf_Phdr);
1279 ehdr->e_phnum = numsegs + 1;
1280 ehdr->e_shentsize = sizeof(Elf_Shdr);
1282 ehdr->e_shstrndx = SHN_UNDEF;
1285 * Fill in the program header entries.
1287 phdr = (Elf_Phdr *)((char *)dst + phoff);
1289 /* The note segement. */
1290 phdr->p_type = PT_NOTE;
1291 phdr->p_offset = noteoff;
1294 phdr->p_filesz = notesz;
1300 /* All the writable segments from the program. */
1303 each_writable_segment(td, cb_put_phdr, &phc);
1308 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1309 const void *desc, size_t descsz)
1313 note.n_namesz = strlen(name) + 1;
1314 note.n_descsz = descsz;
1317 bcopy(¬e, (char *)dst + *off, sizeof note);
1318 *off += sizeof note;
1320 bcopy(name, (char *)dst + *off, note.n_namesz);
1321 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1323 bcopy(desc, (char *)dst + *off, note.n_descsz);
1324 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1328 * Tell kern_execve.c about it, with a little help from the linker.
1330 static struct execsw __elfN(execsw) = {
1331 __CONCAT(exec_, __elfN(imgact)),
1332 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1334 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));