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 #define ELF_NOTE_ROUNDSIZE 4
85 #define OLD_EI_BRAND 8
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89 const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
93 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107 #define CORE_BUF_SIZE (16 * 1024)
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
126 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
127 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
128 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
130 #if __ELF_WORD_SIZE == 32
131 #if defined(__amd64__)
132 int i386_read_exec = 0;
133 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
134 "enable execution from readable segments");
138 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
140 #define trunc_page_ps(va, ps) rounddown2(va, ps)
141 #define round_page_ps(va, ps) roundup2(va, ps)
142 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
144 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
146 Elf_Brandnote __elfN(freebsd_brandnote) = {
147 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
148 .hdr.n_descsz = sizeof(int32_t),
149 .hdr.n_type = NT_FREEBSD_ABI_TAG,
150 .vendor = FREEBSD_ABI_VENDOR,
151 .flags = BN_TRANSLATE_OSREL,
152 .trans_osrel = __elfN(freebsd_trans_osrel)
156 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
160 p = (uintptr_t)(note + 1);
161 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
162 *osrel = *(const int32_t *)(p);
167 static const char GNU_ABI_VENDOR[] = "GNU";
168 static int GNU_KFREEBSD_ABI_DESC = 3;
170 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
171 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
172 .hdr.n_descsz = 16, /* XXX at least 16 */
174 .vendor = GNU_ABI_VENDOR,
175 .flags = BN_TRANSLATE_OSREL,
176 .trans_osrel = kfreebsd_trans_osrel
180 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
182 const Elf32_Word *desc;
185 p = (uintptr_t)(note + 1);
186 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
188 desc = (const Elf32_Word *)p;
189 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
193 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
194 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
196 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
202 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
206 for (i = 0; i < MAX_BRANDS; i++) {
207 if (elf_brand_list[i] == NULL) {
208 elf_brand_list[i] = entry;
212 if (i == MAX_BRANDS) {
213 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
221 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
225 for (i = 0; i < MAX_BRANDS; i++) {
226 if (elf_brand_list[i] == entry) {
227 elf_brand_list[i] = NULL;
237 __elfN(brand_inuse)(Elf_Brandinfo *entry)
242 sx_slock(&allproc_lock);
243 FOREACH_PROC_IN_SYSTEM(p) {
244 if (p->p_sysent == entry->sysvec) {
249 sx_sunlock(&allproc_lock);
254 static Elf_Brandinfo *
255 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
256 int interp_name_len, int32_t *osrel)
258 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
259 Elf_Brandinfo *bi, *bi_m;
264 * We support four types of branding -- (1) the ELF EI_OSABI field
265 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
266 * branding w/in the ELF header, (3) path of the `interp_path'
267 * field, and (4) the ".note.ABI-tag" ELF section.
270 /* Look for an ".note.ABI-tag" ELF section */
272 for (i = 0; i < MAX_BRANDS; i++) {
273 bi = elf_brand_list[i];
276 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
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) != 0 ||
311 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
313 if (hdr->e_machine == bi->machine &&
314 (hdr->e_ident[EI_OSABI] == bi->brand ||
315 (bi->compat_3_brand != NULL &&
316 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
317 bi->compat_3_brand) == 0))) {
318 /* Looks good, but give brand a chance to veto */
319 if (!bi->header_supported ||
320 bi->header_supported(imgp)) {
322 * Again, prefer strictly matching
325 if (interp_name_len == 0 &&
326 bi->interp_path == NULL)
328 if (bi->interp_path != NULL &&
329 strlen(bi->interp_path) + 1 ==
330 interp_name_len && strncmp(interp,
331 bi->interp_path, interp_name_len) == 0)
341 /* No known brand, see if the header is recognized by any brand */
342 for (i = 0; i < MAX_BRANDS; i++) {
343 bi = elf_brand_list[i];
344 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
345 bi->header_supported == NULL)
347 if (hdr->e_machine == bi->machine) {
348 ret = bi->header_supported(imgp);
354 /* Lacking a known brand, search for a recognized interpreter. */
355 if (interp != NULL) {
356 for (i = 0; i < MAX_BRANDS; i++) {
357 bi = elf_brand_list[i];
358 if (bi == NULL || (bi->flags &
359 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
362 if (hdr->e_machine == bi->machine &&
363 /* ELF image p_filesz includes terminating zero */
364 strlen(bi->interp_path) + 1 == interp_name_len &&
365 strncmp(interp, bi->interp_path, interp_name_len)
371 /* Lacking a recognized interpreter, try the default brand */
372 for (i = 0; i < MAX_BRANDS; i++) {
373 bi = elf_brand_list[i];
374 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
375 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
377 if (hdr->e_machine == bi->machine &&
378 __elfN(fallback_brand) == bi->brand)
385 __elfN(check_header)(const Elf_Ehdr *hdr)
391 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
392 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
393 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
394 hdr->e_phentsize != sizeof(Elf_Phdr) ||
395 hdr->e_version != ELF_TARG_VER)
399 * Make sure we have at least one brand for this machine.
402 for (i = 0; i < MAX_BRANDS; i++) {
403 bi = elf_brand_list[i];
404 if (bi != NULL && bi->machine == hdr->e_machine)
414 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
415 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
422 * Create the page if it doesn't exist yet. Ignore errors.
424 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
425 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
428 * Find the page from the underlying object.
430 if (object != NULL) {
431 sf = vm_imgact_map_page(object, offset);
433 return (KERN_FAILURE);
434 off = offset - trunc_page(offset);
435 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
437 vm_imgact_unmap_page(sf);
439 return (KERN_FAILURE);
442 return (KERN_SUCCESS);
446 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
447 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
453 int error, locked, rv;
455 if (start != trunc_page(start)) {
456 rv = __elfN(map_partial)(map, object, offset, start,
457 round_page(start), prot);
458 if (rv != KERN_SUCCESS)
460 offset += round_page(start) - start;
461 start = round_page(start);
463 if (end != round_page(end)) {
464 rv = __elfN(map_partial)(map, object, offset +
465 trunc_page(end) - start, trunc_page(end), end, prot);
466 if (rv != KERN_SUCCESS)
468 end = trunc_page(end);
471 return (KERN_SUCCESS);
472 if ((offset & PAGE_MASK) != 0) {
474 * The mapping is not page aligned. This means that we have
477 rv = vm_map_fixed(map, NULL, 0, start, end - start,
478 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
479 if (rv != KERN_SUCCESS)
482 return (KERN_SUCCESS);
483 for (; start < end; start += sz) {
484 sf = vm_imgact_map_page(object, offset);
486 return (KERN_FAILURE);
487 off = offset - trunc_page(offset);
489 if (sz > PAGE_SIZE - off)
490 sz = PAGE_SIZE - off;
491 error = copyout((caddr_t)sf_buf_kva(sf) + off,
493 vm_imgact_unmap_page(sf);
495 return (KERN_FAILURE);
499 vm_object_reference(object);
500 rv = vm_map_fixed(map, object, offset, start, end - start,
501 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
502 if (rv != KERN_SUCCESS) {
503 locked = VOP_ISLOCKED(imgp->vp);
504 VOP_UNLOCK(imgp->vp, 0);
505 vm_object_deallocate(object);
506 vn_lock(imgp->vp, locked | LK_RETRY);
510 return (KERN_SUCCESS);
514 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
515 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
522 vm_offset_t off, map_addr;
525 vm_ooffset_t file_addr;
528 * It's necessary to fail if the filsz + offset taken from the
529 * header is greater than the actual file pager object's size.
530 * If we were to allow this, then the vm_map_find() below would
531 * walk right off the end of the file object and into the ether.
533 * While I'm here, might as well check for something else that
534 * is invalid: filsz cannot be greater than memsz.
536 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
538 uprintf("elf_load_section: truncated ELF file\n");
542 object = imgp->object;
543 map = &imgp->proc->p_vmspace->vm_map;
544 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
545 file_addr = trunc_page_ps(offset, pagesize);
548 * We have two choices. We can either clear the data in the last page
549 * of an oversized mapping, or we can start the anon mapping a page
550 * early and copy the initialized data into that first page. We
555 else if (memsz > filsz)
556 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
558 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
561 /* cow flags: don't dump readonly sections in core */
562 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
563 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
565 rv = __elfN(map_insert)(imgp, map,
567 file_addr, /* file offset */
568 map_addr, /* virtual start */
569 map_addr + map_len,/* virtual end */
572 if (rv != KERN_SUCCESS)
575 /* we can stop now if we've covered it all */
582 * We have to get the remaining bit of the file into the first part
583 * of the oversized map segment. This is normally because the .data
584 * segment in the file is extended to provide bss. It's a neat idea
585 * to try and save a page, but it's a pain in the behind to implement.
587 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
589 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
590 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
593 /* This had damn well better be true! */
595 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
596 map_addr + map_len, prot, 0);
597 if (rv != KERN_SUCCESS)
602 sf = vm_imgact_map_page(object, offset + filsz);
606 /* send the page fragment to user space */
607 off = trunc_page_ps(offset + filsz, pagesize) -
608 trunc_page(offset + filsz);
609 error = copyout((caddr_t)sf_buf_kva(sf) + off,
610 (caddr_t)map_addr, copy_len);
611 vm_imgact_unmap_page(sf);
617 * Remove write access to the page if it was only granted by map_insert
620 if ((prot & VM_PROT_WRITE) == 0)
621 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
622 map_len), prot, FALSE);
628 * Load the file "file" into memory. It may be either a shared object
631 * The "addr" reference parameter is in/out. On entry, it specifies
632 * the address where a shared object should be loaded. If the file is
633 * an executable, this value is ignored. On exit, "addr" specifies
634 * where the file was actually loaded.
636 * The "entry" reference parameter is out only. On exit, it specifies
637 * the entry point for the loaded file.
640 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
641 u_long *entry, size_t pagesize)
646 struct image_params image_params;
648 const Elf_Ehdr *hdr = NULL;
649 const Elf_Phdr *phdr = NULL;
650 struct nameidata *nd;
652 struct image_params *imgp;
655 u_long base_addr = 0;
656 int error, i, numsegs;
658 #ifdef CAPABILITY_MODE
660 * XXXJA: This check can go away once we are sufficiently confident
661 * that the checks in namei() are correct.
663 if (IN_CAPABILITY_MODE(curthread))
667 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
669 attr = &tempdata->attr;
670 imgp = &tempdata->image_params;
673 * Initialize part of the common data
677 imgp->firstpage = NULL;
678 imgp->image_header = NULL;
680 imgp->execlabel = NULL;
682 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
683 if ((error = namei(nd)) != 0) {
687 NDFREE(nd, NDF_ONLY_PNBUF);
688 imgp->vp = nd->ni_vp;
691 * Check permissions, modes, uid, etc on the file, and "open" it.
693 error = exec_check_permissions(imgp);
697 error = exec_map_first_page(imgp);
702 * Also make certain that the interpreter stays the same, so set
703 * its VV_TEXT flag, too.
705 VOP_SET_TEXT(nd->ni_vp);
707 imgp->object = nd->ni_vp->v_object;
709 hdr = (const Elf_Ehdr *)imgp->image_header;
710 if ((error = __elfN(check_header)(hdr)) != 0)
712 if (hdr->e_type == ET_DYN)
714 else if (hdr->e_type == ET_EXEC)
721 /* Only support headers that fit within first page for now */
722 if ((hdr->e_phoff > PAGE_SIZE) ||
723 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
728 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
729 if (!aligned(phdr, Elf_Addr)) {
734 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
735 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
736 /* Loadable segment */
737 prot = __elfN(trans_prot)(phdr[i].p_flags);
738 error = __elfN(load_section)(imgp, phdr[i].p_offset,
739 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
740 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
744 * Establish the base address if this is the
748 base_addr = trunc_page(phdr[i].p_vaddr +
754 *entry = (unsigned long)hdr->e_entry + rbase;
758 exec_unmap_first_page(imgp);
763 free(tempdata, M_TEMP);
769 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
773 const Elf_Phdr *phdr;
774 Elf_Auxargs *elf_auxargs;
775 struct vmspace *vmspace;
776 const char *err_str, *newinterp;
777 char *interp, *interp_buf, *path;
778 Elf_Brandinfo *brand_info;
779 struct sysentvec *sv;
781 u_long text_size, data_size, total_size, text_addr, data_addr;
782 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
784 int error, i, n, interp_name_len, have_interp;
786 hdr = (const Elf_Ehdr *)imgp->image_header;
789 * Do we have a valid ELF header ?
791 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
792 * if particular brand doesn't support it.
794 if (__elfN(check_header)(hdr) != 0 ||
795 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
799 * From here on down, we return an errno, not -1, as we've
800 * detected an ELF file.
803 if ((hdr->e_phoff > PAGE_SIZE) ||
804 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
805 /* Only support headers in first page for now */
806 uprintf("Program headers not in the first page\n");
809 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
810 if (!aligned(phdr, Elf_Addr)) {
811 uprintf("Unaligned program headers\n");
818 text_size = data_size = total_size = text_addr = data_addr = 0;
821 err_str = newinterp = NULL;
822 interp = interp_buf = NULL;
825 for (i = 0; i < hdr->e_phnum; i++) {
826 switch (phdr[i].p_type) {
829 baddr = phdr[i].p_vaddr;
833 /* Path to interpreter */
834 if (phdr[i].p_filesz > MAXPATHLEN) {
835 uprintf("Invalid PT_INTERP\n");
839 if (interp != NULL) {
840 uprintf("Multiple PT_INTERP headers\n");
844 interp_name_len = phdr[i].p_filesz;
845 if (phdr[i].p_offset > PAGE_SIZE ||
846 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
847 VOP_UNLOCK(imgp->vp, 0);
848 interp_buf = malloc(interp_name_len + 1, M_TEMP,
850 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
851 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
852 interp_name_len, phdr[i].p_offset,
853 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
856 uprintf("i/o error PT_INTERP\n");
859 interp_buf[interp_name_len] = '\0';
862 interp = __DECONST(char *, imgp->image_header) +
869 __elfN(trans_prot)(phdr[i].p_flags);
870 imgp->stack_sz = phdr[i].p_memsz;
875 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
877 if (brand_info == NULL) {
878 uprintf("ELF binary type \"%u\" not known.\n",
879 hdr->e_ident[EI_OSABI]);
884 if (hdr->e_type == ET_DYN) {
885 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
886 uprintf("Cannot execute shared object\n");
891 * Honour the base load address from the dso if it is
892 * non-zero for some reason.
895 et_dyn_addr = ET_DYN_LOAD_ADDR;
897 sv = brand_info->sysvec;
898 if (interp != NULL && brand_info->interp_newpath != NULL)
899 newinterp = brand_info->interp_newpath;
902 * Avoid a possible deadlock if the current address space is destroyed
903 * and that address space maps the locked vnode. In the common case,
904 * the locked vnode's v_usecount is decremented but remains greater
905 * than zero. Consequently, the vnode lock is not needed by vrele().
906 * However, in cases where the vnode lock is external, such as nullfs,
907 * v_usecount may become zero.
909 * The VV_TEXT flag prevents modifications to the executable while
910 * the vnode is unlocked.
912 VOP_UNLOCK(imgp->vp, 0);
914 error = exec_new_vmspace(imgp, sv);
915 imgp->proc->p_sysent = sv;
917 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
921 for (i = 0; i < hdr->e_phnum; i++) {
922 switch (phdr[i].p_type) {
923 case PT_LOAD: /* Loadable segment */
924 if (phdr[i].p_memsz == 0)
926 prot = __elfN(trans_prot)(phdr[i].p_flags);
927 error = __elfN(load_section)(imgp, phdr[i].p_offset,
928 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
929 phdr[i].p_memsz, phdr[i].p_filesz, prot,
935 * If this segment contains the program headers,
936 * remember their virtual address for the AT_PHDR
937 * aux entry. Static binaries don't usually include
940 if (phdr[i].p_offset == 0 &&
941 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
943 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
946 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
947 seg_size = round_page(phdr[i].p_memsz +
948 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
951 * Make the largest executable segment the official
952 * text segment and all others data.
954 * Note that obreak() assumes that data_addr +
955 * data_size == end of data load area, and the ELF
956 * file format expects segments to be sorted by
957 * address. If multiple data segments exist, the
958 * last one will be used.
961 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
962 text_size = seg_size;
963 text_addr = seg_addr;
965 data_size = seg_size;
966 data_addr = seg_addr;
968 total_size += seg_size;
970 case PT_PHDR: /* Program header table info */
971 proghdr = phdr[i].p_vaddr + et_dyn_addr;
978 if (data_addr == 0 && data_size == 0) {
979 data_addr = text_addr;
980 data_size = text_size;
983 entry = (u_long)hdr->e_entry + et_dyn_addr;
986 * Check limits. It should be safe to check the
987 * limits after loading the segments since we do
988 * not actually fault in all the segments pages.
990 PROC_LOCK(imgp->proc);
991 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
992 err_str = "Data segment size exceeds process limit";
993 else if (text_size > maxtsiz)
994 err_str = "Text segment size exceeds system limit";
995 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
996 err_str = "Total segment size exceeds process limit";
997 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
998 err_str = "Data segment size exceeds resource limit";
999 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1000 err_str = "Total segment size exceeds resource limit";
1001 if (err_str != NULL) {
1002 PROC_UNLOCK(imgp->proc);
1003 uprintf("%s\n", err_str);
1008 vmspace = imgp->proc->p_vmspace;
1009 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1010 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1011 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1012 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1015 * We load the dynamic linker where a userland call
1016 * to mmap(0, ...) would put it. The rationale behind this
1017 * calculation is that it leaves room for the heap to grow to
1018 * its maximum allowed size.
1020 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1022 PROC_UNLOCK(imgp->proc);
1024 imgp->entry_addr = entry;
1026 if (interp != NULL) {
1027 have_interp = FALSE;
1028 VOP_UNLOCK(imgp->vp, 0);
1029 if (brand_info->emul_path != NULL &&
1030 brand_info->emul_path[0] != '\0') {
1031 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1032 snprintf(path, MAXPATHLEN, "%s%s",
1033 brand_info->emul_path, interp);
1034 error = __elfN(load_file)(imgp->proc, path, &addr,
1035 &imgp->entry_addr, sv->sv_pagesize);
1040 if (!have_interp && newinterp != NULL &&
1041 (brand_info->interp_path == NULL ||
1042 strcmp(interp, brand_info->interp_path) == 0)) {
1043 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1044 &imgp->entry_addr, sv->sv_pagesize);
1049 error = __elfN(load_file)(imgp->proc, interp, &addr,
1050 &imgp->entry_addr, sv->sv_pagesize);
1052 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1054 uprintf("ELF interpreter %s not found, error %d\n",
1062 * Construct auxargs table (used by the fixup routine)
1064 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1065 elf_auxargs->execfd = -1;
1066 elf_auxargs->phdr = proghdr;
1067 elf_auxargs->phent = hdr->e_phentsize;
1068 elf_auxargs->phnum = hdr->e_phnum;
1069 elf_auxargs->pagesz = PAGE_SIZE;
1070 elf_auxargs->base = addr;
1071 elf_auxargs->flags = 0;
1072 elf_auxargs->entry = entry;
1073 elf_auxargs->hdr_eflags = hdr->e_flags;
1075 imgp->auxargs = elf_auxargs;
1076 imgp->interpreted = 0;
1077 imgp->reloc_base = addr;
1078 imgp->proc->p_osrel = osrel;
1079 imgp->proc->p_elf_machine = hdr->e_machine;
1080 imgp->proc->p_elf_flags = hdr->e_flags;
1083 free(interp_buf, M_TEMP);
1087 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1090 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1092 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1096 base = (Elf_Addr *)*stack_base;
1097 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1099 if (args->execfd != -1)
1100 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1101 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1102 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1103 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1104 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1105 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1106 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1107 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1109 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1111 if (imgp->execpathp != 0)
1112 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1113 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1114 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1115 if (imgp->canary != 0) {
1116 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1117 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1119 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1120 if (imgp->pagesizes != 0) {
1121 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1122 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1124 if (imgp->sysent->sv_timekeep_base != 0) {
1125 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1126 imgp->sysent->sv_timekeep_base);
1128 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1129 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1130 imgp->sysent->sv_stackprot);
1131 AUXARGS_ENTRY(pos, AT_NULL, 0);
1133 free(imgp->auxargs, M_TEMP);
1134 imgp->auxargs = NULL;
1137 suword(base, (long)imgp->args->argc);
1138 *stack_base = (register_t *)base;
1143 * Code for generating ELF core dumps.
1146 typedef void (*segment_callback)(vm_map_entry_t, void *);
1148 /* Closure for cb_put_phdr(). */
1149 struct phdr_closure {
1150 Elf_Phdr *phdr; /* Program header to fill in */
1151 Elf_Off offset; /* Offset of segment in core file */
1154 /* Closure for cb_size_segment(). */
1155 struct sseg_closure {
1156 int count; /* Count of writable segments. */
1157 size_t size; /* Total size of all writable segments. */
1160 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1163 int type; /* Note type. */
1164 outfunc_t outfunc; /* Output function. */
1165 void *outarg; /* Argument for the output function. */
1166 size_t outsize; /* Output size. */
1167 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1170 TAILQ_HEAD(note_info_list, note_info);
1172 /* Coredump output parameters. */
1173 struct coredump_params {
1175 struct ucred *active_cred;
1176 struct ucred *file_cred;
1179 struct gzio_stream *gzs;
1182 static void cb_put_phdr(vm_map_entry_t, void *);
1183 static void cb_size_segment(vm_map_entry_t, void *);
1184 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1186 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1187 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1188 struct note_info_list *, size_t);
1189 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1191 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1192 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1193 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1194 static int sbuf_drain_core_output(void *, const char *, int);
1195 static int sbuf_drain_count(void *arg, const char *data, int len);
1197 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1198 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1199 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1200 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1201 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1202 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1203 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1204 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1205 static void note_procstat_files(void *, struct sbuf *, size_t *);
1206 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1207 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1208 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1209 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1210 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1213 extern int compress_user_cores_gzlevel;
1216 * Write out a core segment to the compression stream.
1219 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1225 chunk_len = MIN(len, CORE_BUF_SIZE);
1228 * We can get EFAULT error here.
1229 * In that case zero out the current chunk of the segment.
1231 error = copyin(base, buf, chunk_len);
1233 bzero(buf, chunk_len);
1234 error = gzio_write(p->gzs, buf, chunk_len);
1244 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1247 return (core_write((struct coredump_params *)arg, base, len, offset,
1253 core_write(struct coredump_params *p, const void *base, size_t len,
1254 off_t offset, enum uio_seg seg)
1257 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1258 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1259 p->active_cred, p->file_cred, NULL, p->td));
1263 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1270 return (compress_chunk(p, base, tmpbuf, len));
1273 * EFAULT is a non-fatal error that we can get, for example,
1274 * if the segment is backed by a file but extends beyond its
1277 error = core_write(p, base, len, offset, UIO_USERSPACE);
1278 if (error == EFAULT) {
1279 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1280 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1281 "for process %s\n", base, len, offset, curproc->p_comm);
1284 * Write a "real" zero byte at the end of the target region
1285 * in the case this is the last segment.
1286 * The intermediate space will be implicitly zero-filled.
1288 error = core_write(p, zero_region, 1, offset + len - 1,
1295 * Drain into a core file.
1298 sbuf_drain_core_output(void *arg, const char *data, int len)
1300 struct coredump_params *p;
1303 p = (struct coredump_params *)arg;
1306 * Some kern_proc out routines that print to this sbuf may
1307 * call us with the process lock held. Draining with the
1308 * non-sleepable lock held is unsafe. The lock is needed for
1309 * those routines when dumping a live process. In our case we
1310 * can safely release the lock before draining and acquire
1313 locked = PROC_LOCKED(p->td->td_proc);
1315 PROC_UNLOCK(p->td->td_proc);
1318 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1321 error = core_write(p, __DECONST(void *, data), len, p->offset,
1324 PROC_LOCK(p->td->td_proc);
1332 * Drain into a counter.
1335 sbuf_drain_count(void *arg, const char *data __unused, int len)
1339 sizep = (size_t *)arg;
1345 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1347 struct ucred *cred = td->td_ucred;
1349 struct sseg_closure seginfo;
1350 struct note_info_list notelst;
1351 struct coredump_params params;
1352 struct note_info *ninfo;
1354 size_t hdrsize, notesz, coresize;
1358 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1362 TAILQ_INIT(¬elst);
1364 /* Size the program segments. */
1367 each_dumpable_segment(td, cb_size_segment, &seginfo);
1370 * Collect info about the core file header area.
1372 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1373 if (seginfo.count + 1 >= PN_XNUM)
1374 hdrsize += sizeof(Elf_Shdr);
1375 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1376 coresize = round_page(hdrsize + notesz) + seginfo.size;
1378 /* Set up core dump parameters. */
1380 params.active_cred = cred;
1381 params.file_cred = NOCRED;
1388 PROC_LOCK(td->td_proc);
1389 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1390 PROC_UNLOCK(td->td_proc);
1397 if (coresize >= limit) {
1403 /* Create a compression stream if necessary. */
1405 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1406 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1407 if (params.gzs == NULL) {
1411 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1416 * Allocate memory for building the header, fill it up,
1417 * and write it out following the notes.
1419 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1420 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1423 /* Write the contents of all of the writable segments. */
1429 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1430 offset = round_page(hdrsize + notesz);
1431 for (i = 0; i < seginfo.count; i++) {
1432 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1433 php->p_filesz, offset, ¶ms, tmpbuf);
1436 offset += php->p_filesz;
1440 if (error == 0 && compress)
1441 error = gzio_flush(params.gzs);
1446 "Failed to write core file for process %s (error %d)\n",
1447 curproc->p_comm, error);
1453 free(tmpbuf, M_TEMP);
1454 if (params.gzs != NULL)
1455 gzio_fini(params.gzs);
1458 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1459 TAILQ_REMOVE(¬elst, ninfo, link);
1460 free(ninfo, M_TEMP);
1469 * A callback for each_dumpable_segment() to write out the segment's
1470 * program header entry.
1473 cb_put_phdr(entry, closure)
1474 vm_map_entry_t entry;
1477 struct phdr_closure *phc = (struct phdr_closure *)closure;
1478 Elf_Phdr *phdr = phc->phdr;
1480 phc->offset = round_page(phc->offset);
1482 phdr->p_type = PT_LOAD;
1483 phdr->p_offset = phc->offset;
1484 phdr->p_vaddr = entry->start;
1486 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1487 phdr->p_align = PAGE_SIZE;
1488 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1490 phc->offset += phdr->p_filesz;
1495 * A callback for each_dumpable_segment() to gather information about
1496 * the number of segments and their total size.
1499 cb_size_segment(vm_map_entry_t entry, void *closure)
1501 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1504 ssc->size += entry->end - entry->start;
1508 * For each writable segment in the process's memory map, call the given
1509 * function with a pointer to the map entry and some arbitrary
1510 * caller-supplied data.
1513 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1515 struct proc *p = td->td_proc;
1516 vm_map_t map = &p->p_vmspace->vm_map;
1517 vm_map_entry_t entry;
1518 vm_object_t backing_object, object;
1519 boolean_t ignore_entry;
1521 vm_map_lock_read(map);
1522 for (entry = map->header.next; entry != &map->header;
1523 entry = entry->next) {
1525 * Don't dump inaccessible mappings, deal with legacy
1528 * Note that read-only segments related to the elf binary
1529 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1530 * need to arbitrarily ignore such segments.
1532 if (elf_legacy_coredump) {
1533 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1536 if ((entry->protection & VM_PROT_ALL) == 0)
1541 * Dont include memory segment in the coredump if
1542 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1543 * madvise(2). Do not dump submaps (i.e. parts of the
1546 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1549 if ((object = entry->object.vm_object) == NULL)
1552 /* Ignore memory-mapped devices and such things. */
1553 VM_OBJECT_RLOCK(object);
1554 while ((backing_object = object->backing_object) != NULL) {
1555 VM_OBJECT_RLOCK(backing_object);
1556 VM_OBJECT_RUNLOCK(object);
1557 object = backing_object;
1559 ignore_entry = object->type != OBJT_DEFAULT &&
1560 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1561 object->type != OBJT_PHYS;
1562 VM_OBJECT_RUNLOCK(object);
1566 (*func)(entry, closure);
1568 vm_map_unlock_read(map);
1572 * Write the core file header to the file, including padding up to
1573 * the page boundary.
1576 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1577 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1579 struct note_info *ninfo;
1583 /* Fill in the header. */
1584 bzero(hdr, hdrsize);
1585 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1587 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1588 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1589 sbuf_start_section(sb, NULL);
1590 sbuf_bcat(sb, hdr, hdrsize);
1591 TAILQ_FOREACH(ninfo, notelst, link)
1592 __elfN(putnote)(ninfo, sb);
1593 /* Align up to a page boundary for the program segments. */
1594 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1595 error = sbuf_finish(sb);
1602 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1612 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1615 * To have the debugger select the right thread (LWP) as the initial
1616 * thread, we dump the state of the thread passed to us in td first.
1617 * This is the thread that causes the core dump and thus likely to
1618 * be the right thread one wants to have selected in the debugger.
1621 while (thr != NULL) {
1622 size += register_note(list, NT_PRSTATUS,
1623 __elfN(note_prstatus), thr);
1624 size += register_note(list, NT_FPREGSET,
1625 __elfN(note_fpregset), thr);
1626 size += register_note(list, NT_THRMISC,
1627 __elfN(note_thrmisc), thr);
1628 size += register_note(list, -1,
1629 __elfN(note_threadmd), thr);
1631 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1632 TAILQ_NEXT(thr, td_plist);
1634 thr = TAILQ_NEXT(thr, td_plist);
1637 size += register_note(list, NT_PROCSTAT_PROC,
1638 __elfN(note_procstat_proc), p);
1639 size += register_note(list, NT_PROCSTAT_FILES,
1640 note_procstat_files, p);
1641 size += register_note(list, NT_PROCSTAT_VMMAP,
1642 note_procstat_vmmap, p);
1643 size += register_note(list, NT_PROCSTAT_GROUPS,
1644 note_procstat_groups, p);
1645 size += register_note(list, NT_PROCSTAT_UMASK,
1646 note_procstat_umask, p);
1647 size += register_note(list, NT_PROCSTAT_RLIMIT,
1648 note_procstat_rlimit, p);
1649 size += register_note(list, NT_PROCSTAT_OSREL,
1650 note_procstat_osrel, p);
1651 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1652 __elfN(note_procstat_psstrings), p);
1653 size += register_note(list, NT_PROCSTAT_AUXV,
1654 __elfN(note_procstat_auxv), p);
1660 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1666 struct phdr_closure phc;
1668 ehdr = (Elf_Ehdr *)hdr;
1670 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1671 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1672 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1673 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1674 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1675 ehdr->e_ident[EI_DATA] = ELF_DATA;
1676 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1677 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1678 ehdr->e_ident[EI_ABIVERSION] = 0;
1679 ehdr->e_ident[EI_PAD] = 0;
1680 ehdr->e_type = ET_CORE;
1681 ehdr->e_machine = td->td_proc->p_elf_machine;
1682 ehdr->e_version = EV_CURRENT;
1684 ehdr->e_phoff = sizeof(Elf_Ehdr);
1685 ehdr->e_flags = td->td_proc->p_elf_flags;
1686 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1687 ehdr->e_phentsize = sizeof(Elf_Phdr);
1688 ehdr->e_shentsize = sizeof(Elf_Shdr);
1689 ehdr->e_shstrndx = SHN_UNDEF;
1690 if (numsegs + 1 < PN_XNUM) {
1691 ehdr->e_phnum = numsegs + 1;
1694 ehdr->e_phnum = PN_XNUM;
1697 ehdr->e_shoff = ehdr->e_phoff +
1698 (numsegs + 1) * ehdr->e_phentsize;
1699 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1700 ("e_shoff: %zu, hdrsize - shdr: %zu",
1701 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1703 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1704 memset(shdr, 0, sizeof(*shdr));
1706 * A special first section is used to hold large segment and
1707 * section counts. This was proposed by Sun Microsystems in
1708 * Solaris and has been adopted by Linux; the standard ELF
1709 * tools are already familiar with the technique.
1711 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1712 * (or 12-7 depending on the version of the document) for more
1715 shdr->sh_type = SHT_NULL;
1716 shdr->sh_size = ehdr->e_shnum;
1717 shdr->sh_link = ehdr->e_shstrndx;
1718 shdr->sh_info = numsegs + 1;
1722 * Fill in the program header entries.
1724 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1726 /* The note segement. */
1727 phdr->p_type = PT_NOTE;
1728 phdr->p_offset = hdrsize;
1731 phdr->p_filesz = notesz;
1733 phdr->p_flags = PF_R;
1734 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1737 /* All the writable segments from the program. */
1739 phc.offset = round_page(hdrsize + notesz);
1740 each_dumpable_segment(td, cb_put_phdr, &phc);
1744 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1746 struct note_info *ninfo;
1747 size_t size, notesize;
1750 out(arg, NULL, &size);
1751 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1753 ninfo->outfunc = out;
1754 ninfo->outarg = arg;
1755 ninfo->outsize = size;
1756 TAILQ_INSERT_TAIL(list, ninfo, link);
1761 notesize = sizeof(Elf_Note) + /* note header */
1762 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1764 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1770 append_note_data(const void *src, void *dst, size_t len)
1774 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1776 bcopy(src, dst, len);
1777 bzero((char *)dst + len, padded_len - len);
1779 return (padded_len);
1783 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1791 note = (Elf_Note *)buf;
1792 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1793 note->n_descsz = size;
1794 note->n_type = type;
1795 buf += sizeof(*note);
1796 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1797 sizeof(FREEBSD_ABI_VENDOR));
1798 append_note_data(src, buf, size);
1803 notesize = sizeof(Elf_Note) + /* note header */
1804 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1806 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1812 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1815 ssize_t old_len, sect_len;
1816 size_t new_len, descsz, i;
1818 if (ninfo->type == -1) {
1819 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1823 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1824 note.n_descsz = ninfo->outsize;
1825 note.n_type = ninfo->type;
1827 sbuf_bcat(sb, ¬e, sizeof(note));
1828 sbuf_start_section(sb, &old_len);
1829 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1830 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1831 if (note.n_descsz == 0)
1833 sbuf_start_section(sb, &old_len);
1834 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1835 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1839 new_len = (size_t)sect_len;
1840 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1841 if (new_len < descsz) {
1843 * It is expected that individual note emitters will correctly
1844 * predict their expected output size and fill up to that size
1845 * themselves, padding in a format-specific way if needed.
1846 * However, in case they don't, just do it here with zeros.
1848 for (i = 0; i < descsz - new_len; i++)
1850 } else if (new_len > descsz) {
1852 * We can't always truncate sb -- we may have drained some
1855 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1856 "read it (%zu > %zu). Since it is longer than "
1857 "expected, this coredump's notes are corrupt. THIS "
1858 "IS A BUG in the note_procstat routine for type %u.\n",
1859 __func__, (unsigned)note.n_type, new_len, descsz,
1860 (unsigned)note.n_type));
1865 * Miscellaneous note out functions.
1868 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1869 #include <compat/freebsd32/freebsd32.h>
1871 typedef struct prstatus32 elf_prstatus_t;
1872 typedef struct prpsinfo32 elf_prpsinfo_t;
1873 typedef struct fpreg32 elf_prfpregset_t;
1874 typedef struct fpreg32 elf_fpregset_t;
1875 typedef struct reg32 elf_gregset_t;
1876 typedef struct thrmisc32 elf_thrmisc_t;
1877 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1878 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1879 typedef uint32_t elf_ps_strings_t;
1881 typedef prstatus_t elf_prstatus_t;
1882 typedef prpsinfo_t elf_prpsinfo_t;
1883 typedef prfpregset_t elf_prfpregset_t;
1884 typedef prfpregset_t elf_fpregset_t;
1885 typedef gregset_t elf_gregset_t;
1886 typedef thrmisc_t elf_thrmisc_t;
1887 #define ELF_KERN_PROC_MASK 0
1888 typedef struct kinfo_proc elf_kinfo_proc_t;
1889 typedef vm_offset_t elf_ps_strings_t;
1893 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1899 elf_prpsinfo_t *psinfo;
1902 p = (struct proc *)arg;
1904 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1905 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1906 psinfo->pr_version = PRPSINFO_VERSION;
1907 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1908 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1910 if (p->p_args != NULL) {
1911 len = sizeof(psinfo->pr_psargs) - 1;
1912 if (len > p->p_args->ar_length)
1913 len = p->p_args->ar_length;
1914 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1920 sbuf_new(&sbarg, psinfo->pr_psargs,
1921 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1922 error = proc_getargv(curthread, p, &sbarg);
1924 if (sbuf_finish(&sbarg) == 0)
1925 len = sbuf_len(&sbarg) - 1;
1927 len = sizeof(psinfo->pr_psargs) - 1;
1928 sbuf_delete(&sbarg);
1930 if (error || len == 0)
1931 strlcpy(psinfo->pr_psargs, p->p_comm,
1932 sizeof(psinfo->pr_psargs));
1934 KASSERT(len < sizeof(psinfo->pr_psargs),
1935 ("len is too long: %zu vs %zu", len,
1936 sizeof(psinfo->pr_psargs)));
1937 cp = psinfo->pr_psargs;
1940 cp = memchr(cp, '\0', end - cp);
1946 psinfo->pr_pid = p->p_pid;
1947 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1948 free(psinfo, M_TEMP);
1950 *sizep = sizeof(*psinfo);
1954 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1957 elf_prstatus_t *status;
1959 td = (struct thread *)arg;
1961 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1962 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1963 status->pr_version = PRSTATUS_VERSION;
1964 status->pr_statussz = sizeof(elf_prstatus_t);
1965 status->pr_gregsetsz = sizeof(elf_gregset_t);
1966 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1967 status->pr_osreldate = osreldate;
1968 status->pr_cursig = td->td_proc->p_sig;
1969 status->pr_pid = td->td_tid;
1970 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1971 fill_regs32(td, &status->pr_reg);
1973 fill_regs(td, &status->pr_reg);
1975 sbuf_bcat(sb, status, sizeof(*status));
1976 free(status, M_TEMP);
1978 *sizep = sizeof(*status);
1982 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1985 elf_prfpregset_t *fpregset;
1987 td = (struct thread *)arg;
1989 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1990 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1991 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1992 fill_fpregs32(td, fpregset);
1994 fill_fpregs(td, fpregset);
1996 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1997 free(fpregset, M_TEMP);
1999 *sizep = sizeof(*fpregset);
2003 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2006 elf_thrmisc_t thrmisc;
2008 td = (struct thread *)arg;
2010 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2011 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2012 strcpy(thrmisc.pr_tname, td->td_name);
2013 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2015 *sizep = sizeof(thrmisc);
2019 * Allow for MD specific notes, as well as any MD
2020 * specific preparations for writing MI notes.
2023 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2029 td = (struct thread *)arg;
2031 if (size != 0 && sb != NULL)
2032 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2036 __elfN(dump_thread)(td, buf, &size);
2037 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2038 if (size != 0 && sb != NULL)
2039 sbuf_bcat(sb, buf, size);
2044 #ifdef KINFO_PROC_SIZE
2045 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2049 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2055 p = (struct proc *)arg;
2056 size = sizeof(structsize) + p->p_numthreads *
2057 sizeof(elf_kinfo_proc_t);
2060 KASSERT(*sizep == size, ("invalid size"));
2061 structsize = sizeof(elf_kinfo_proc_t);
2062 sbuf_bcat(sb, &structsize, sizeof(structsize));
2063 sx_slock(&proctree_lock);
2065 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2066 sx_sunlock(&proctree_lock);
2071 #ifdef KINFO_FILE_SIZE
2072 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2076 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2079 size_t size, sect_sz, i;
2080 ssize_t start_len, sect_len;
2081 int structsize, filedesc_flags;
2083 if (coredump_pack_fileinfo)
2084 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2088 p = (struct proc *)arg;
2089 structsize = sizeof(struct kinfo_file);
2092 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2093 sbuf_set_drain(sb, sbuf_drain_count, &size);
2094 sbuf_bcat(sb, &structsize, sizeof(structsize));
2096 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2101 sbuf_start_section(sb, &start_len);
2103 sbuf_bcat(sb, &structsize, sizeof(structsize));
2105 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2108 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2113 KASSERT(sect_sz <= *sizep,
2114 ("kern_proc_filedesc_out did not respect maxlen; "
2115 "requested %zu, got %zu", *sizep - sizeof(structsize),
2116 sect_sz - sizeof(structsize)));
2118 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2123 #ifdef KINFO_VMENTRY_SIZE
2124 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2128 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2132 int structsize, vmmap_flags;
2134 if (coredump_pack_vmmapinfo)
2135 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2139 p = (struct proc *)arg;
2140 structsize = sizeof(struct kinfo_vmentry);
2143 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2144 sbuf_set_drain(sb, sbuf_drain_count, &size);
2145 sbuf_bcat(sb, &structsize, sizeof(structsize));
2147 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2152 sbuf_bcat(sb, &structsize, sizeof(structsize));
2154 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2160 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2166 p = (struct proc *)arg;
2167 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2169 KASSERT(*sizep == size, ("invalid size"));
2170 structsize = sizeof(gid_t);
2171 sbuf_bcat(sb, &structsize, sizeof(structsize));
2172 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2179 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2185 p = (struct proc *)arg;
2186 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2188 KASSERT(*sizep == size, ("invalid size"));
2189 structsize = sizeof(p->p_fd->fd_cmask);
2190 sbuf_bcat(sb, &structsize, sizeof(structsize));
2191 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2197 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2200 struct rlimit rlim[RLIM_NLIMITS];
2204 p = (struct proc *)arg;
2205 size = sizeof(structsize) + sizeof(rlim);
2207 KASSERT(*sizep == size, ("invalid size"));
2208 structsize = sizeof(rlim);
2209 sbuf_bcat(sb, &structsize, sizeof(structsize));
2211 for (i = 0; i < RLIM_NLIMITS; i++)
2212 lim_rlimit_proc(p, i, &rlim[i]);
2214 sbuf_bcat(sb, rlim, sizeof(rlim));
2220 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2226 p = (struct proc *)arg;
2227 size = sizeof(structsize) + sizeof(p->p_osrel);
2229 KASSERT(*sizep == size, ("invalid size"));
2230 structsize = sizeof(p->p_osrel);
2231 sbuf_bcat(sb, &structsize, sizeof(structsize));
2232 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2238 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2241 elf_ps_strings_t ps_strings;
2245 p = (struct proc *)arg;
2246 size = sizeof(structsize) + sizeof(ps_strings);
2248 KASSERT(*sizep == size, ("invalid size"));
2249 structsize = sizeof(ps_strings);
2250 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2251 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2253 ps_strings = p->p_sysent->sv_psstrings;
2255 sbuf_bcat(sb, &structsize, sizeof(structsize));
2256 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2262 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2268 p = (struct proc *)arg;
2271 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2272 sbuf_set_drain(sb, sbuf_drain_count, &size);
2273 sbuf_bcat(sb, &structsize, sizeof(structsize));
2275 proc_getauxv(curthread, p, sb);
2281 structsize = sizeof(Elf_Auxinfo);
2282 sbuf_bcat(sb, &structsize, sizeof(structsize));
2284 proc_getauxv(curthread, p, sb);
2290 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2291 int32_t *osrel, const Elf_Phdr *pnote)
2293 const Elf_Note *note, *note0, *note_end;
2294 const char *note_name;
2299 /* We need some limit, might as well use PAGE_SIZE. */
2300 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2302 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2303 if (pnote->p_offset > PAGE_SIZE ||
2304 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2305 VOP_UNLOCK(imgp->vp, 0);
2306 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2307 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2308 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2309 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2310 curthread->td_ucred, NOCRED, NULL, curthread);
2312 uprintf("i/o error PT_NOTE\n");
2316 note = note0 = (const Elf_Note *)buf;
2317 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2319 note = note0 = (const Elf_Note *)(imgp->image_header +
2321 note_end = (const Elf_Note *)(imgp->image_header +
2322 pnote->p_offset + pnote->p_filesz);
2325 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2326 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2327 (const char *)note < sizeof(Elf_Note)) {
2331 if (note->n_namesz != checknote->hdr.n_namesz ||
2332 note->n_descsz != checknote->hdr.n_descsz ||
2333 note->n_type != checknote->hdr.n_type)
2335 note_name = (const char *)(note + 1);
2336 if (note_name + checknote->hdr.n_namesz >=
2337 (const char *)note_end || strncmp(checknote->vendor,
2338 note_name, checknote->hdr.n_namesz) != 0)
2342 * Fetch the osreldate for binary
2343 * from the ELF OSABI-note if necessary.
2345 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2346 checknote->trans_osrel != NULL) {
2347 res = checknote->trans_osrel(note, osrel);
2353 note = (const Elf_Note *)((const char *)(note + 1) +
2354 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2355 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2364 * Try to find the appropriate ABI-note section for checknote,
2365 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2366 * first page of the image is searched, the same as for headers.
2369 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2372 const Elf_Phdr *phdr;
2373 const Elf_Ehdr *hdr;
2376 hdr = (const Elf_Ehdr *)imgp->image_header;
2377 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2379 for (i = 0; i < hdr->e_phnum; i++) {
2380 if (phdr[i].p_type == PT_NOTE &&
2381 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2389 * Tell kern_execve.c about it, with a little help from the linker.
2391 static struct execsw __elfN(execsw) = {
2392 __CONCAT(exec_, __elfN(imgact)),
2393 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2395 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2398 __elfN(trans_prot)(Elf_Word flags)
2404 prot |= VM_PROT_EXECUTE;
2406 prot |= VM_PROT_WRITE;
2408 prot |= VM_PROT_READ;
2409 #if __ELF_WORD_SIZE == 32
2410 #if defined(__amd64__)
2411 if (i386_read_exec && (flags & PF_R))
2412 prot |= VM_PROT_EXECUTE;
2419 __elfN(untrans_prot)(vm_prot_t prot)
2424 if (prot & VM_PROT_EXECUTE)
2426 if (prot & VM_PROT_READ)
2428 if (prot & VM_PROT_WRITE)