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_capsicum.h"
35 #include "opt_compat.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
59 #include <sys/sf_buf.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
84 #include <machine/acle-compat.h>
87 #define ELF_NOTE_ROUNDSIZE 4
88 #define OLD_EI_BRAND 8
90 static int __elfN(check_header)(const Elf_Ehdr *hdr);
91 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
92 const char *interp, int interp_name_len, int32_t *osrel);
93 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 u_long *entry, size_t pagesize);
95 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
96 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
101 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103 Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
110 #define CORE_BUF_SIZE (16 * 1024)
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119 &elf_legacy_coredump, 0, "");
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136 "enable execution from readable segments");
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
143 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
150 .hdr.n_descsz = sizeof(int32_t),
151 .hdr.n_type = NT_FREEBSD_ABI_TAG,
152 .vendor = FREEBSD_ABI_VENDOR,
153 .flags = BN_TRANSLATE_OSREL,
154 .trans_osrel = __elfN(freebsd_trans_osrel)
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 p = (uintptr_t)(note + 1);
163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 *osrel = *(const int32_t *)(p);
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
174 .hdr.n_descsz = 16, /* XXX at least 16 */
176 .vendor = GNU_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = kfreebsd_trans_osrel
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 const Elf32_Word *desc;
187 p = (uintptr_t)(note + 1);
188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 desc = (const Elf32_Word *)p;
191 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 for (i = 0; i < MAX_BRANDS; i++) {
209 if (elf_brand_list[i] == NULL) {
210 elf_brand_list[i] = entry;
214 if (i == MAX_BRANDS) {
215 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 for (i = 0; i < MAX_BRANDS; i++) {
228 if (elf_brand_list[i] == entry) {
229 elf_brand_list[i] = NULL;
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
244 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
246 if (p->p_sysent == entry->sysvec) {
251 sx_sunlock(&allproc_lock);
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258 int interp_name_len, int32_t *osrel)
260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 Elf_Brandinfo *bi, *bi_m;
266 * We support four types of branding -- (1) the ELF EI_OSABI field
267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 * branding w/in the ELF header, (3) path of the `interp_path'
269 * field, and (4) the ".note.ABI-tag" ELF section.
272 /* Look for an ".note.ABI-tag" ELF section */
274 for (i = 0; i < MAX_BRANDS; i++) {
275 bi = elf_brand_list[i];
278 if (hdr->e_machine == bi->machine && (bi->flags &
279 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
280 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
281 /* Give brand a chance to veto check_note's guess */
282 if (ret && bi->header_supported)
283 ret = bi->header_supported(imgp);
285 * If note checker claimed the binary, but the
286 * interpreter path in the image does not
287 * match default one for the brand, try to
288 * search for other brands with the same
289 * interpreter. Either there is better brand
290 * with the right interpreter, or, failing
291 * this, we return first brand which accepted
292 * our note and, optionally, header.
294 if (ret && bi_m == NULL && (strlen(bi->interp_path) +
295 1 != interp_name_len || strncmp(interp,
296 bi->interp_path, interp_name_len) != 0)) {
307 /* If the executable has a brand, search for it in the brand list. */
308 for (i = 0; i < MAX_BRANDS; i++) {
309 bi = elf_brand_list[i];
310 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
312 if (hdr->e_machine == bi->machine &&
313 (hdr->e_ident[EI_OSABI] == bi->brand ||
314 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
315 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
316 /* Looks good, but give brand a chance to veto */
317 if (!bi->header_supported || bi->header_supported(imgp))
322 /* No known brand, see if the header is recognized by any brand */
323 for (i = 0; i < MAX_BRANDS; i++) {
324 bi = elf_brand_list[i];
325 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
326 bi->header_supported == NULL)
328 if (hdr->e_machine == bi->machine) {
329 ret = bi->header_supported(imgp);
335 /* Lacking a known brand, search for a recognized interpreter. */
336 if (interp != NULL) {
337 for (i = 0; i < MAX_BRANDS; i++) {
338 bi = elf_brand_list[i];
339 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
341 if (hdr->e_machine == bi->machine &&
342 /* ELF image p_filesz includes terminating zero */
343 strlen(bi->interp_path) + 1 == interp_name_len &&
344 strncmp(interp, bi->interp_path, interp_name_len)
350 /* Lacking a recognized interpreter, try the default brand */
351 for (i = 0; i < MAX_BRANDS; i++) {
352 bi = elf_brand_list[i];
353 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
355 if (hdr->e_machine == bi->machine &&
356 __elfN(fallback_brand) == bi->brand)
363 __elfN(check_header)(const Elf_Ehdr *hdr)
369 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
370 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
371 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
372 hdr->e_phentsize != sizeof(Elf_Phdr) ||
373 hdr->e_version != ELF_TARG_VER)
377 * Make sure we have at least one brand for this machine.
380 for (i = 0; i < MAX_BRANDS; i++) {
381 bi = elf_brand_list[i];
382 if (bi != NULL && bi->machine == hdr->e_machine)
392 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
393 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
400 * Create the page if it doesn't exist yet. Ignore errors.
403 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
404 VM_PROT_ALL, VM_PROT_ALL, 0);
408 * Find the page from the underlying object.
411 sf = vm_imgact_map_page(object, offset);
413 return (KERN_FAILURE);
414 off = offset - trunc_page(offset);
415 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
417 vm_imgact_unmap_page(sf);
419 return (KERN_FAILURE);
423 return (KERN_SUCCESS);
427 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
428 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
435 if (start != trunc_page(start)) {
436 rv = __elfN(map_partial)(map, object, offset, start,
437 round_page(start), prot);
440 offset += round_page(start) - start;
441 start = round_page(start);
443 if (end != round_page(end)) {
444 rv = __elfN(map_partial)(map, object, offset +
445 trunc_page(end) - start, trunc_page(end), end, prot);
448 end = trunc_page(end);
451 if (offset & PAGE_MASK) {
453 * The mapping is not page aligned. This means we have
454 * to copy the data. Sigh.
456 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
457 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
462 return (KERN_SUCCESS);
463 for (; start < end; start += sz) {
464 sf = vm_imgact_map_page(object, offset);
466 return (KERN_FAILURE);
467 off = offset - trunc_page(offset);
469 if (sz > PAGE_SIZE - off)
470 sz = PAGE_SIZE - off;
471 error = copyout((caddr_t)sf_buf_kva(sf) + off,
473 vm_imgact_unmap_page(sf);
475 return (KERN_FAILURE);
481 vm_object_reference(object);
483 rv = vm_map_insert(map, object, offset, start, end,
484 prot, VM_PROT_ALL, cow);
486 if (rv != KERN_SUCCESS)
487 vm_object_deallocate(object);
491 return (KERN_SUCCESS);
496 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
497 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
504 vm_offset_t map_addr;
507 vm_offset_t file_addr;
510 * It's necessary to fail if the filsz + offset taken from the
511 * header is greater than the actual file pager object's size.
512 * If we were to allow this, then the vm_map_find() below would
513 * walk right off the end of the file object and into the ether.
515 * While I'm here, might as well check for something else that
516 * is invalid: filsz cannot be greater than memsz.
518 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
519 uprintf("elf_load_section: truncated ELF file\n");
523 object = imgp->object;
524 map = &imgp->proc->p_vmspace->vm_map;
525 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
526 file_addr = trunc_page_ps(offset, pagesize);
529 * We have two choices. We can either clear the data in the last page
530 * of an oversized mapping, or we can start the anon mapping a page
531 * early and copy the initialized data into that first page. We
532 * choose the second..
535 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
537 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
540 /* cow flags: don't dump readonly sections in core */
541 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
542 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
544 rv = __elfN(map_insert)(map,
546 file_addr, /* file offset */
547 map_addr, /* virtual start */
548 map_addr + map_len,/* virtual end */
551 if (rv != KERN_SUCCESS)
554 /* we can stop now if we've covered it all */
555 if (memsz == filsz) {
562 * We have to get the remaining bit of the file into the first part
563 * of the oversized map segment. This is normally because the .data
564 * segment in the file is extended to provide bss. It's a neat idea
565 * to try and save a page, but it's a pain in the behind to implement.
567 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
568 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
569 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
572 /* This had damn well better be true! */
574 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
575 map_len, VM_PROT_ALL, 0);
576 if (rv != KERN_SUCCESS) {
584 sf = vm_imgact_map_page(object, offset + filsz);
588 /* send the page fragment to user space */
589 off = trunc_page_ps(offset + filsz, pagesize) -
590 trunc_page(offset + filsz);
591 error = copyout((caddr_t)sf_buf_kva(sf) + off,
592 (caddr_t)map_addr, copy_len);
593 vm_imgact_unmap_page(sf);
600 * set it to the specified protection.
601 * XXX had better undo the damage from pasting over the cracks here!
603 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
604 map_len), prot, FALSE);
610 * Load the file "file" into memory. It may be either a shared object
613 * The "addr" reference parameter is in/out. On entry, it specifies
614 * the address where a shared object should be loaded. If the file is
615 * an executable, this value is ignored. On exit, "addr" specifies
616 * where the file was actually loaded.
618 * The "entry" reference parameter is out only. On exit, it specifies
619 * the entry point for the loaded file.
622 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
623 u_long *entry, size_t pagesize)
628 struct image_params image_params;
630 const Elf_Ehdr *hdr = NULL;
631 const Elf_Phdr *phdr = NULL;
632 struct nameidata *nd;
634 struct image_params *imgp;
637 u_long base_addr = 0;
638 int error, i, numsegs;
640 #ifdef CAPABILITY_MODE
642 * XXXJA: This check can go away once we are sufficiently confident
643 * that the checks in namei() are correct.
645 if (IN_CAPABILITY_MODE(curthread))
649 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
651 attr = &tempdata->attr;
652 imgp = &tempdata->image_params;
655 * Initialize part of the common data
659 imgp->firstpage = NULL;
660 imgp->image_header = NULL;
662 imgp->execlabel = NULL;
664 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
665 if ((error = namei(nd)) != 0) {
669 NDFREE(nd, NDF_ONLY_PNBUF);
670 imgp->vp = nd->ni_vp;
673 * Check permissions, modes, uid, etc on the file, and "open" it.
675 error = exec_check_permissions(imgp);
679 error = exec_map_first_page(imgp);
684 * Also make certain that the interpreter stays the same, so set
685 * its VV_TEXT flag, too.
687 VOP_SET_TEXT(nd->ni_vp);
689 imgp->object = nd->ni_vp->v_object;
691 hdr = (const Elf_Ehdr *)imgp->image_header;
692 if ((error = __elfN(check_header)(hdr)) != 0)
694 if (hdr->e_type == ET_DYN)
696 else if (hdr->e_type == ET_EXEC)
703 /* Only support headers that fit within first page for now */
704 if ((hdr->e_phoff > PAGE_SIZE) ||
705 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
710 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
711 if (!aligned(phdr, Elf_Addr)) {
716 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
717 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
718 /* Loadable segment */
719 prot = __elfN(trans_prot)(phdr[i].p_flags);
720 error = __elfN(load_section)(imgp, phdr[i].p_offset,
721 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
722 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
726 * Establish the base address if this is the
730 base_addr = trunc_page(phdr[i].p_vaddr +
736 *entry = (unsigned long)hdr->e_entry + rbase;
740 exec_unmap_first_page(imgp);
745 free(tempdata, M_TEMP);
751 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
755 const Elf_Phdr *phdr;
756 Elf_Auxargs *elf_auxargs;
757 struct vmspace *vmspace;
758 const char *err_str, *newinterp;
759 char *interp, *interp_buf, *path;
760 Elf_Brandinfo *brand_info;
761 struct sysentvec *sv;
763 u_long text_size, data_size, total_size, text_addr, data_addr;
764 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
766 int error, i, n, interp_name_len, have_interp;
768 hdr = (const Elf_Ehdr *)imgp->image_header;
771 * Do we have a valid ELF header ?
773 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
774 * if particular brand doesn't support it.
776 if (__elfN(check_header)(hdr) != 0 ||
777 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
781 * From here on down, we return an errno, not -1, as we've
782 * detected an ELF file.
785 if ((hdr->e_phoff > PAGE_SIZE) ||
786 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
787 /* Only support headers in first page for now */
788 uprintf("Program headers not in the first page\n");
791 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
792 if (!aligned(phdr, Elf_Addr)) {
793 uprintf("Unaligned program headers\n");
800 text_size = data_size = total_size = text_addr = data_addr = 0;
803 err_str = newinterp = NULL;
804 interp = interp_buf = NULL;
807 for (i = 0; i < hdr->e_phnum; i++) {
808 switch (phdr[i].p_type) {
811 baddr = phdr[i].p_vaddr;
815 /* Path to interpreter */
816 if (phdr[i].p_filesz > MAXPATHLEN) {
817 uprintf("Invalid PT_INTERP\n");
821 if (interp != NULL) {
822 uprintf("Multiple PT_INTERP headers\n");
826 interp_name_len = phdr[i].p_filesz;
827 if (phdr[i].p_offset > PAGE_SIZE ||
828 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
829 VOP_UNLOCK(imgp->vp, 0);
830 interp_buf = malloc(interp_name_len + 1, M_TEMP,
832 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
833 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
834 interp_name_len, phdr[i].p_offset,
835 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
838 uprintf("i/o error PT_INTERP\n");
841 interp_buf[interp_name_len] = '\0';
844 interp = __DECONST(char *, imgp->image_header) +
851 __elfN(trans_prot)(phdr[i].p_flags);
852 imgp->stack_sz = phdr[i].p_memsz;
857 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
859 if (brand_info == NULL) {
860 uprintf("ELF binary type \"%u\" not known.\n",
861 hdr->e_ident[EI_OSABI]);
865 if (hdr->e_type == ET_DYN) {
866 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
867 uprintf("Cannot execute shared object\n");
872 * Honour the base load address from the dso if it is
873 * non-zero for some reason.
876 et_dyn_addr = ET_DYN_LOAD_ADDR;
881 sv = brand_info->sysvec;
882 if (interp != NULL && brand_info->interp_newpath != NULL)
883 newinterp = brand_info->interp_newpath;
886 * Avoid a possible deadlock if the current address space is destroyed
887 * and that address space maps the locked vnode. In the common case,
888 * the locked vnode's v_usecount is decremented but remains greater
889 * than zero. Consequently, the vnode lock is not needed by vrele().
890 * However, in cases where the vnode lock is external, such as nullfs,
891 * v_usecount may become zero.
893 * The VV_TEXT flag prevents modifications to the executable while
894 * the vnode is unlocked.
896 VOP_UNLOCK(imgp->vp, 0);
898 error = exec_new_vmspace(imgp, sv);
899 imgp->proc->p_sysent = sv;
901 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
905 for (i = 0; i < hdr->e_phnum; i++) {
906 switch (phdr[i].p_type) {
907 case PT_LOAD: /* Loadable segment */
908 if (phdr[i].p_memsz == 0)
910 prot = __elfN(trans_prot)(phdr[i].p_flags);
911 error = __elfN(load_section)(imgp, phdr[i].p_offset,
912 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
913 phdr[i].p_memsz, phdr[i].p_filesz, prot,
919 * If this segment contains the program headers,
920 * remember their virtual address for the AT_PHDR
921 * aux entry. Static binaries don't usually include
924 if (phdr[i].p_offset == 0 &&
925 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
927 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
930 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
931 seg_size = round_page(phdr[i].p_memsz +
932 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
935 * Make the largest executable segment the official
936 * text segment and all others data.
938 * Note that obreak() assumes that data_addr +
939 * data_size == end of data load area, and the ELF
940 * file format expects segments to be sorted by
941 * address. If multiple data segments exist, the
942 * last one will be used.
945 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
946 text_size = seg_size;
947 text_addr = seg_addr;
949 data_size = seg_size;
950 data_addr = seg_addr;
952 total_size += seg_size;
954 case PT_PHDR: /* Program header table info */
955 proghdr = phdr[i].p_vaddr + et_dyn_addr;
962 if (data_addr == 0 && data_size == 0) {
963 data_addr = text_addr;
964 data_size = text_size;
967 entry = (u_long)hdr->e_entry + et_dyn_addr;
970 * Check limits. It should be safe to check the
971 * limits after loading the segments since we do
972 * not actually fault in all the segments pages.
974 PROC_LOCK(imgp->proc);
975 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
976 err_str = "Data segment size exceeds process limit";
977 else if (text_size > maxtsiz)
978 err_str = "Text segment size exceeds system limit";
979 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
980 err_str = "Total segment size exceeds process limit";
981 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
982 err_str = "Data segment size exceeds resource limit";
983 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
984 err_str = "Total segment size exceeds resource limit";
985 if (err_str != NULL) {
986 PROC_UNLOCK(imgp->proc);
987 uprintf("%s\n", err_str);
992 vmspace = imgp->proc->p_vmspace;
993 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
994 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
995 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
996 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
999 * We load the dynamic linker where a userland call
1000 * to mmap(0, ...) would put it. The rationale behind this
1001 * calculation is that it leaves room for the heap to grow to
1002 * its maximum allowed size.
1004 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1006 PROC_UNLOCK(imgp->proc);
1008 imgp->entry_addr = entry;
1010 if (interp != NULL) {
1011 have_interp = FALSE;
1012 VOP_UNLOCK(imgp->vp, 0);
1013 if (brand_info->emul_path != NULL &&
1014 brand_info->emul_path[0] != '\0') {
1015 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1016 snprintf(path, MAXPATHLEN, "%s%s",
1017 brand_info->emul_path, interp);
1018 error = __elfN(load_file)(imgp->proc, path, &addr,
1019 &imgp->entry_addr, sv->sv_pagesize);
1024 if (!have_interp && newinterp != NULL &&
1025 (brand_info->interp_path == NULL ||
1026 strcmp(interp, brand_info->interp_path) == 0)) {
1027 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1028 &imgp->entry_addr, sv->sv_pagesize);
1033 error = __elfN(load_file)(imgp->proc, interp, &addr,
1034 &imgp->entry_addr, sv->sv_pagesize);
1036 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1038 uprintf("ELF interpreter %s not found, error %d\n",
1046 * Construct auxargs table (used by the fixup routine)
1048 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1049 elf_auxargs->execfd = -1;
1050 elf_auxargs->phdr = proghdr;
1051 elf_auxargs->phent = hdr->e_phentsize;
1052 elf_auxargs->phnum = hdr->e_phnum;
1053 elf_auxargs->pagesz = PAGE_SIZE;
1054 elf_auxargs->base = addr;
1055 elf_auxargs->flags = 0;
1056 elf_auxargs->entry = entry;
1057 elf_auxargs->hdr_eflags = hdr->e_flags;
1059 imgp->auxargs = elf_auxargs;
1060 imgp->interpreted = 0;
1061 imgp->reloc_base = addr;
1062 imgp->proc->p_osrel = osrel;
1065 free(interp_buf, M_TEMP);
1069 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1072 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1074 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1078 base = (Elf_Addr *)*stack_base;
1079 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1081 if (args->execfd != -1)
1082 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1083 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1084 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1085 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1086 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1087 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1088 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1089 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1091 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1093 if (imgp->execpathp != 0)
1094 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1095 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1096 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1097 if (imgp->canary != 0) {
1098 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1099 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1101 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1102 if (imgp->pagesizes != 0) {
1103 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1104 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1106 if (imgp->sysent->sv_timekeep_base != 0) {
1107 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1108 imgp->sysent->sv_timekeep_base);
1110 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1111 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1112 imgp->sysent->sv_stackprot);
1113 AUXARGS_ENTRY(pos, AT_NULL, 0);
1115 free(imgp->auxargs, M_TEMP);
1116 imgp->auxargs = NULL;
1119 suword(base, (long)imgp->args->argc);
1120 *stack_base = (register_t *)base;
1125 * Code for generating ELF core dumps.
1128 typedef void (*segment_callback)(vm_map_entry_t, void *);
1130 /* Closure for cb_put_phdr(). */
1131 struct phdr_closure {
1132 Elf_Phdr *phdr; /* Program header to fill in */
1133 Elf_Off offset; /* Offset of segment in core file */
1136 /* Closure for cb_size_segment(). */
1137 struct sseg_closure {
1138 int count; /* Count of writable segments. */
1139 size_t size; /* Total size of all writable segments. */
1142 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1145 int type; /* Note type. */
1146 outfunc_t outfunc; /* Output function. */
1147 void *outarg; /* Argument for the output function. */
1148 size_t outsize; /* Output size. */
1149 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1152 TAILQ_HEAD(note_info_list, note_info);
1154 /* Coredump output parameters. */
1155 struct coredump_params {
1157 struct ucred *active_cred;
1158 struct ucred *file_cred;
1161 struct gzio_stream *gzs;
1164 static void cb_put_phdr(vm_map_entry_t, void *);
1165 static void cb_size_segment(vm_map_entry_t, void *);
1166 static int core_write(struct coredump_params *, void *, size_t, off_t,
1168 static void each_writable_segment(struct thread *, segment_callback, void *);
1169 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1170 struct note_info_list *, size_t);
1171 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1173 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1174 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1175 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1176 static int sbuf_drain_core_output(void *, const char *, int);
1177 static int sbuf_drain_count(void *arg, const char *data, int len);
1179 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1180 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1181 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1182 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1183 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1184 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1185 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1186 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1187 static void note_procstat_files(void *, struct sbuf *, size_t *);
1188 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1189 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1190 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1191 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1192 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1195 extern int compress_user_cores_gzlevel;
1198 * Write out a core segment to the compression stream.
1201 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1207 chunk_len = MIN(len, CORE_BUF_SIZE);
1208 copyin(base, buf, chunk_len);
1209 error = gzio_write(p->gzs, buf, chunk_len);
1219 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1222 return (core_write((struct coredump_params *)arg, base, len, offset,
1228 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1232 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1233 seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1234 p->active_cred, p->file_cred, NULL, p->td));
1238 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1244 return (compress_chunk(p, base, tmpbuf, len));
1246 return (core_write(p, base, len, offset, UIO_USERSPACE));
1250 * Drain into a core file.
1253 sbuf_drain_core_output(void *arg, const char *data, int len)
1255 struct coredump_params *p;
1258 p = (struct coredump_params *)arg;
1261 * Some kern_proc out routines that print to this sbuf may
1262 * call us with the process lock held. Draining with the
1263 * non-sleepable lock held is unsafe. The lock is needed for
1264 * those routines when dumping a live process. In our case we
1265 * can safely release the lock before draining and acquire
1268 locked = PROC_LOCKED(p->td->td_proc);
1270 PROC_UNLOCK(p->td->td_proc);
1273 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1276 error = core_write(p, __DECONST(void *, data), len, p->offset,
1279 PROC_LOCK(p->td->td_proc);
1287 * Drain into a counter.
1290 sbuf_drain_count(void *arg, const char *data __unused, int len)
1294 sizep = (size_t *)arg;
1300 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1302 struct ucred *cred = td->td_ucred;
1304 struct sseg_closure seginfo;
1305 struct note_info_list notelst;
1306 struct coredump_params params;
1307 struct note_info *ninfo;
1309 size_t hdrsize, notesz, coresize;
1313 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1317 TAILQ_INIT(¬elst);
1319 /* Size the program segments. */
1322 each_writable_segment(td, cb_size_segment, &seginfo);
1325 * Collect info about the core file header area.
1327 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1328 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1329 coresize = round_page(hdrsize + notesz) + seginfo.size;
1331 /* Set up core dump parameters. */
1333 params.active_cred = cred;
1334 params.file_cred = NOCRED;
1341 PROC_LOCK(td->td_proc);
1342 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1343 PROC_UNLOCK(td->td_proc);
1350 if (coresize >= limit) {
1356 /* Create a compression stream if necessary. */
1358 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1359 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1360 if (params.gzs == NULL) {
1364 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1369 * Allocate memory for building the header, fill it up,
1370 * and write it out following the notes.
1372 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1377 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1380 /* Write the contents of all of the writable segments. */
1386 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1387 offset = round_page(hdrsize + notesz);
1388 for (i = 0; i < seginfo.count; i++) {
1389 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1390 php->p_filesz, offset, ¶ms, tmpbuf);
1393 offset += php->p_filesz;
1397 if (error == 0 && compress)
1398 error = gzio_flush(params.gzs);
1403 "Failed to write core file for process %s (error %d)\n",
1404 curproc->p_comm, error);
1410 free(tmpbuf, M_TEMP);
1411 if (params.gzs != NULL)
1412 gzio_fini(params.gzs);
1415 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1416 TAILQ_REMOVE(¬elst, ninfo, link);
1417 free(ninfo, M_TEMP);
1426 * A callback for each_writable_segment() to write out the segment's
1427 * program header entry.
1430 cb_put_phdr(entry, closure)
1431 vm_map_entry_t entry;
1434 struct phdr_closure *phc = (struct phdr_closure *)closure;
1435 Elf_Phdr *phdr = phc->phdr;
1437 phc->offset = round_page(phc->offset);
1439 phdr->p_type = PT_LOAD;
1440 phdr->p_offset = phc->offset;
1441 phdr->p_vaddr = entry->start;
1443 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1444 phdr->p_align = PAGE_SIZE;
1445 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1447 phc->offset += phdr->p_filesz;
1452 * A callback for each_writable_segment() to gather information about
1453 * the number of segments and their total size.
1456 cb_size_segment(entry, closure)
1457 vm_map_entry_t entry;
1460 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1463 ssc->size += entry->end - entry->start;
1467 * For each writable segment in the process's memory map, call the given
1468 * function with a pointer to the map entry and some arbitrary
1469 * caller-supplied data.
1472 each_writable_segment(td, func, closure)
1474 segment_callback func;
1477 struct proc *p = td->td_proc;
1478 vm_map_t map = &p->p_vmspace->vm_map;
1479 vm_map_entry_t entry;
1480 vm_object_t backing_object, object;
1481 boolean_t ignore_entry;
1483 vm_map_lock_read(map);
1484 for (entry = map->header.next; entry != &map->header;
1485 entry = entry->next) {
1487 * Don't dump inaccessible mappings, deal with legacy
1490 * Note that read-only segments related to the elf binary
1491 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1492 * need to arbitrarily ignore such segments.
1494 if (elf_legacy_coredump) {
1495 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1498 if ((entry->protection & VM_PROT_ALL) == 0)
1503 * Dont include memory segment in the coredump if
1504 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1505 * madvise(2). Do not dump submaps (i.e. parts of the
1508 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1511 if ((object = entry->object.vm_object) == NULL)
1514 /* Ignore memory-mapped devices and such things. */
1515 VM_OBJECT_RLOCK(object);
1516 while ((backing_object = object->backing_object) != NULL) {
1517 VM_OBJECT_RLOCK(backing_object);
1518 VM_OBJECT_RUNLOCK(object);
1519 object = backing_object;
1521 ignore_entry = object->type != OBJT_DEFAULT &&
1522 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1523 object->type != OBJT_PHYS;
1524 VM_OBJECT_RUNLOCK(object);
1528 (*func)(entry, closure);
1530 vm_map_unlock_read(map);
1534 * Write the core file header to the file, including padding up to
1535 * the page boundary.
1538 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1539 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1541 struct note_info *ninfo;
1545 /* Fill in the header. */
1546 bzero(hdr, hdrsize);
1547 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1549 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1550 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1551 sbuf_start_section(sb, NULL);
1552 sbuf_bcat(sb, hdr, hdrsize);
1553 TAILQ_FOREACH(ninfo, notelst, link)
1554 __elfN(putnote)(ninfo, sb);
1555 /* Align up to a page boundary for the program segments. */
1556 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1557 error = sbuf_finish(sb);
1564 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1574 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1577 * To have the debugger select the right thread (LWP) as the initial
1578 * thread, we dump the state of the thread passed to us in td first.
1579 * This is the thread that causes the core dump and thus likely to
1580 * be the right thread one wants to have selected in the debugger.
1583 while (thr != NULL) {
1584 size += register_note(list, NT_PRSTATUS,
1585 __elfN(note_prstatus), thr);
1586 size += register_note(list, NT_FPREGSET,
1587 __elfN(note_fpregset), thr);
1588 size += register_note(list, NT_THRMISC,
1589 __elfN(note_thrmisc), thr);
1590 size += register_note(list, -1,
1591 __elfN(note_threadmd), thr);
1593 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1594 TAILQ_NEXT(thr, td_plist);
1596 thr = TAILQ_NEXT(thr, td_plist);
1599 size += register_note(list, NT_PROCSTAT_PROC,
1600 __elfN(note_procstat_proc), p);
1601 size += register_note(list, NT_PROCSTAT_FILES,
1602 note_procstat_files, p);
1603 size += register_note(list, NT_PROCSTAT_VMMAP,
1604 note_procstat_vmmap, p);
1605 size += register_note(list, NT_PROCSTAT_GROUPS,
1606 note_procstat_groups, p);
1607 size += register_note(list, NT_PROCSTAT_UMASK,
1608 note_procstat_umask, p);
1609 size += register_note(list, NT_PROCSTAT_RLIMIT,
1610 note_procstat_rlimit, p);
1611 size += register_note(list, NT_PROCSTAT_OSREL,
1612 note_procstat_osrel, p);
1613 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1614 __elfN(note_procstat_psstrings), p);
1615 size += register_note(list, NT_PROCSTAT_AUXV,
1616 __elfN(note_procstat_auxv), p);
1622 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1627 struct phdr_closure phc;
1629 ehdr = (Elf_Ehdr *)hdr;
1630 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1632 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1633 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1634 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1635 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1636 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1637 ehdr->e_ident[EI_DATA] = ELF_DATA;
1638 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1639 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1640 ehdr->e_ident[EI_ABIVERSION] = 0;
1641 ehdr->e_ident[EI_PAD] = 0;
1642 ehdr->e_type = ET_CORE;
1643 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1644 ehdr->e_machine = ELF_ARCH32;
1646 ehdr->e_machine = ELF_ARCH;
1648 ehdr->e_version = EV_CURRENT;
1650 ehdr->e_phoff = sizeof(Elf_Ehdr);
1652 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1653 ehdr->e_phentsize = sizeof(Elf_Phdr);
1654 ehdr->e_phnum = numsegs + 1;
1655 ehdr->e_shentsize = sizeof(Elf_Shdr);
1657 ehdr->e_shstrndx = SHN_UNDEF;
1660 * Fill in the program header entries.
1663 /* The note segement. */
1664 phdr->p_type = PT_NOTE;
1665 phdr->p_offset = hdrsize;
1668 phdr->p_filesz = notesz;
1670 phdr->p_flags = PF_R;
1671 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1674 /* All the writable segments from the program. */
1676 phc.offset = round_page(hdrsize + notesz);
1677 each_writable_segment(td, cb_put_phdr, &phc);
1681 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1683 struct note_info *ninfo;
1684 size_t size, notesize;
1687 out(arg, NULL, &size);
1688 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1690 ninfo->outfunc = out;
1691 ninfo->outarg = arg;
1692 ninfo->outsize = size;
1693 TAILQ_INSERT_TAIL(list, ninfo, link);
1698 notesize = sizeof(Elf_Note) + /* note header */
1699 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1701 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1707 append_note_data(const void *src, void *dst, size_t len)
1711 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1713 bcopy(src, dst, len);
1714 bzero((char *)dst + len, padded_len - len);
1716 return (padded_len);
1720 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1728 note = (Elf_Note *)buf;
1729 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1730 note->n_descsz = size;
1731 note->n_type = type;
1732 buf += sizeof(*note);
1733 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1734 sizeof(FREEBSD_ABI_VENDOR));
1735 append_note_data(src, buf, size);
1740 notesize = sizeof(Elf_Note) + /* note header */
1741 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1743 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1749 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1752 ssize_t old_len, sect_len;
1753 size_t new_len, descsz, i;
1755 if (ninfo->type == -1) {
1756 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1760 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1761 note.n_descsz = ninfo->outsize;
1762 note.n_type = ninfo->type;
1764 sbuf_bcat(sb, ¬e, sizeof(note));
1765 sbuf_start_section(sb, &old_len);
1766 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1767 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1768 if (note.n_descsz == 0)
1770 sbuf_start_section(sb, &old_len);
1771 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1772 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1776 new_len = (size_t)sect_len;
1777 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1778 if (new_len < descsz) {
1780 * It is expected that individual note emitters will correctly
1781 * predict their expected output size and fill up to that size
1782 * themselves, padding in a format-specific way if needed.
1783 * However, in case they don't, just do it here with zeros.
1785 for (i = 0; i < descsz - new_len; i++)
1787 } else if (new_len > descsz) {
1789 * We can't always truncate sb -- we may have drained some
1792 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1793 "read it (%zu > %zu). Since it is longer than "
1794 "expected, this coredump's notes are corrupt. THIS "
1795 "IS A BUG in the note_procstat routine for type %u.\n",
1796 __func__, (unsigned)note.n_type, new_len, descsz,
1797 (unsigned)note.n_type));
1802 * Miscellaneous note out functions.
1805 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1806 #include <compat/freebsd32/freebsd32.h>
1808 typedef struct prstatus32 elf_prstatus_t;
1809 typedef struct prpsinfo32 elf_prpsinfo_t;
1810 typedef struct fpreg32 elf_prfpregset_t;
1811 typedef struct fpreg32 elf_fpregset_t;
1812 typedef struct reg32 elf_gregset_t;
1813 typedef struct thrmisc32 elf_thrmisc_t;
1814 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1815 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1816 typedef uint32_t elf_ps_strings_t;
1818 typedef prstatus_t elf_prstatus_t;
1819 typedef prpsinfo_t elf_prpsinfo_t;
1820 typedef prfpregset_t elf_prfpregset_t;
1821 typedef prfpregset_t elf_fpregset_t;
1822 typedef gregset_t elf_gregset_t;
1823 typedef thrmisc_t elf_thrmisc_t;
1824 #define ELF_KERN_PROC_MASK 0
1825 typedef struct kinfo_proc elf_kinfo_proc_t;
1826 typedef vm_offset_t elf_ps_strings_t;
1830 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1833 elf_prpsinfo_t *psinfo;
1835 p = (struct proc *)arg;
1837 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1838 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1839 psinfo->pr_version = PRPSINFO_VERSION;
1840 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1841 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1843 * XXX - We don't fill in the command line arguments properly
1846 strlcpy(psinfo->pr_psargs, p->p_comm,
1847 sizeof(psinfo->pr_psargs));
1849 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1850 free(psinfo, M_TEMP);
1852 *sizep = sizeof(*psinfo);
1856 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1859 elf_prstatus_t *status;
1861 td = (struct thread *)arg;
1863 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1864 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1865 status->pr_version = PRSTATUS_VERSION;
1866 status->pr_statussz = sizeof(elf_prstatus_t);
1867 status->pr_gregsetsz = sizeof(elf_gregset_t);
1868 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1869 status->pr_osreldate = osreldate;
1870 status->pr_cursig = td->td_proc->p_sig;
1871 status->pr_pid = td->td_tid;
1872 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1873 fill_regs32(td, &status->pr_reg);
1875 fill_regs(td, &status->pr_reg);
1877 sbuf_bcat(sb, status, sizeof(*status));
1878 free(status, M_TEMP);
1880 *sizep = sizeof(*status);
1884 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1887 elf_prfpregset_t *fpregset;
1889 td = (struct thread *)arg;
1891 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1892 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1893 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1894 fill_fpregs32(td, fpregset);
1896 fill_fpregs(td, fpregset);
1898 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1899 free(fpregset, M_TEMP);
1901 *sizep = sizeof(*fpregset);
1905 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1908 elf_thrmisc_t thrmisc;
1910 td = (struct thread *)arg;
1912 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1913 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1914 strcpy(thrmisc.pr_tname, td->td_name);
1915 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1917 *sizep = sizeof(thrmisc);
1921 * Allow for MD specific notes, as well as any MD
1922 * specific preparations for writing MI notes.
1925 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1931 td = (struct thread *)arg;
1933 if (size != 0 && sb != NULL)
1934 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1938 __elfN(dump_thread)(td, buf, &size);
1939 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1940 if (size != 0 && sb != NULL)
1941 sbuf_bcat(sb, buf, size);
1946 #ifdef KINFO_PROC_SIZE
1947 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1951 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1957 p = (struct proc *)arg;
1958 size = sizeof(structsize) + p->p_numthreads *
1959 sizeof(elf_kinfo_proc_t);
1962 KASSERT(*sizep == size, ("invalid size"));
1963 structsize = sizeof(elf_kinfo_proc_t);
1964 sbuf_bcat(sb, &structsize, sizeof(structsize));
1965 sx_slock(&proctree_lock);
1967 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1968 sx_sunlock(&proctree_lock);
1973 #ifdef KINFO_FILE_SIZE
1974 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1978 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1981 size_t size, sect_sz, i;
1982 ssize_t start_len, sect_len;
1983 int structsize, filedesc_flags;
1985 if (coredump_pack_fileinfo)
1986 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1990 p = (struct proc *)arg;
1991 structsize = sizeof(struct kinfo_file);
1994 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1995 sbuf_set_drain(sb, sbuf_drain_count, &size);
1996 sbuf_bcat(sb, &structsize, sizeof(structsize));
1998 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2003 sbuf_start_section(sb, &start_len);
2005 sbuf_bcat(sb, &structsize, sizeof(structsize));
2007 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2010 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2015 KASSERT(sect_sz <= *sizep,
2016 ("kern_proc_filedesc_out did not respect maxlen; "
2017 "requested %zu, got %zu", *sizep - sizeof(structsize),
2018 sect_sz - sizeof(structsize)));
2020 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2025 #ifdef KINFO_VMENTRY_SIZE
2026 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2030 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2034 int structsize, vmmap_flags;
2036 if (coredump_pack_vmmapinfo)
2037 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2041 p = (struct proc *)arg;
2042 structsize = sizeof(struct kinfo_vmentry);
2045 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2046 sbuf_set_drain(sb, sbuf_drain_count, &size);
2047 sbuf_bcat(sb, &structsize, sizeof(structsize));
2049 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2054 sbuf_bcat(sb, &structsize, sizeof(structsize));
2056 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2062 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2068 p = (struct proc *)arg;
2069 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2071 KASSERT(*sizep == size, ("invalid size"));
2072 structsize = sizeof(gid_t);
2073 sbuf_bcat(sb, &structsize, sizeof(structsize));
2074 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2081 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2087 p = (struct proc *)arg;
2088 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2090 KASSERT(*sizep == size, ("invalid size"));
2091 structsize = sizeof(p->p_fd->fd_cmask);
2092 sbuf_bcat(sb, &structsize, sizeof(structsize));
2093 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2099 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2102 struct rlimit rlim[RLIM_NLIMITS];
2106 p = (struct proc *)arg;
2107 size = sizeof(structsize) + sizeof(rlim);
2109 KASSERT(*sizep == size, ("invalid size"));
2110 structsize = sizeof(rlim);
2111 sbuf_bcat(sb, &structsize, sizeof(structsize));
2113 for (i = 0; i < RLIM_NLIMITS; i++)
2114 lim_rlimit_proc(p, i, &rlim[i]);
2116 sbuf_bcat(sb, rlim, sizeof(rlim));
2122 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2128 p = (struct proc *)arg;
2129 size = sizeof(structsize) + sizeof(p->p_osrel);
2131 KASSERT(*sizep == size, ("invalid size"));
2132 structsize = sizeof(p->p_osrel);
2133 sbuf_bcat(sb, &structsize, sizeof(structsize));
2134 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2140 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2143 elf_ps_strings_t ps_strings;
2147 p = (struct proc *)arg;
2148 size = sizeof(structsize) + sizeof(ps_strings);
2150 KASSERT(*sizep == size, ("invalid size"));
2151 structsize = sizeof(ps_strings);
2152 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2153 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2155 ps_strings = p->p_sysent->sv_psstrings;
2157 sbuf_bcat(sb, &structsize, sizeof(structsize));
2158 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2164 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2170 p = (struct proc *)arg;
2173 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2174 sbuf_set_drain(sb, sbuf_drain_count, &size);
2175 sbuf_bcat(sb, &structsize, sizeof(structsize));
2177 proc_getauxv(curthread, p, sb);
2183 structsize = sizeof(Elf_Auxinfo);
2184 sbuf_bcat(sb, &structsize, sizeof(structsize));
2186 proc_getauxv(curthread, p, sb);
2192 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2193 int32_t *osrel, const Elf_Phdr *pnote)
2195 const Elf_Note *note, *note0, *note_end;
2196 const char *note_name;
2201 /* We need some limit, might as well use PAGE_SIZE. */
2202 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2204 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2205 if (pnote->p_offset > PAGE_SIZE ||
2206 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2207 VOP_UNLOCK(imgp->vp, 0);
2208 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2209 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2210 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2211 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2212 curthread->td_ucred, NOCRED, NULL, curthread);
2214 uprintf("i/o error PT_NOTE\n");
2218 note = note0 = (const Elf_Note *)buf;
2219 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2221 note = note0 = (const Elf_Note *)(imgp->image_header +
2223 note_end = (const Elf_Note *)(imgp->image_header +
2224 pnote->p_offset + pnote->p_filesz);
2227 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2228 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2229 (const char *)note < sizeof(Elf_Note)) {
2233 if (note->n_namesz != checknote->hdr.n_namesz ||
2234 note->n_descsz != checknote->hdr.n_descsz ||
2235 note->n_type != checknote->hdr.n_type)
2237 note_name = (const char *)(note + 1);
2238 if (note_name + checknote->hdr.n_namesz >=
2239 (const char *)note_end || strncmp(checknote->vendor,
2240 note_name, checknote->hdr.n_namesz) != 0)
2244 * Fetch the osreldate for binary
2245 * from the ELF OSABI-note if necessary.
2247 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2248 checknote->trans_osrel != NULL) {
2249 res = checknote->trans_osrel(note, osrel);
2255 note = (const Elf_Note *)((const char *)(note + 1) +
2256 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2257 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2266 * Try to find the appropriate ABI-note section for checknote,
2267 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2268 * first page of the image is searched, the same as for headers.
2271 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2274 const Elf_Phdr *phdr;
2275 const Elf_Ehdr *hdr;
2278 hdr = (const Elf_Ehdr *)imgp->image_header;
2279 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2281 for (i = 0; i < hdr->e_phnum; i++) {
2282 if (phdr[i].p_type == PT_NOTE &&
2283 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2291 * Tell kern_execve.c about it, with a little help from the linker.
2293 static struct execsw __elfN(execsw) = {
2294 __CONCAT(exec_, __elfN(imgact)),
2295 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2297 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2300 __elfN(trans_prot)(Elf_Word flags)
2306 prot |= VM_PROT_EXECUTE;
2308 prot |= VM_PROT_WRITE;
2310 prot |= VM_PROT_READ;
2311 #if __ELF_WORD_SIZE == 32
2312 #if defined(__amd64__)
2313 if (i386_read_exec && (flags & PF_R))
2314 prot |= VM_PROT_EXECUTE;
2321 __elfN(untrans_prot)(vm_prot_t prot)
2326 if (prot & VM_PROT_EXECUTE)
2328 if (prot & VM_PROT_READ)
2330 if (prot & VM_PROT_WRITE)