2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
38 #include "opt_compat.h"
41 #include <sys/param.h>
42 #include <sys/capsicum.h>
44 #include <sys/fcntl.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
49 #include <sys/kernel.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
54 #include <sys/namei.h>
55 #include <sys/pioctl.h>
57 #include <sys/procfs.h>
58 #include <sys/ptrace.h>
59 #include <sys/racct.h>
60 #include <sys/resourcevar.h>
61 #include <sys/rwlock.h>
63 #include <sys/sf_buf.h>
65 #include <sys/systm.h>
66 #include <sys/signalvar.h>
69 #include <sys/syscall.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/vnode.h>
73 #include <sys/syslog.h>
74 #include <sys/eventhandler.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_param.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_extern.h>
85 #include <machine/elf.h>
86 #include <machine/md_var.h>
88 #define ELF_NOTE_ROUNDSIZE 4
89 #define OLD_EI_BRAND 8
91 static int __elfN(check_header)(const Elf_Ehdr *hdr);
92 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
93 const char *interp, int interp_name_len, int32_t *osrel);
94 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
95 u_long *entry, size_t pagesize);
96 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
97 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
99 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
100 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
102 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
103 static boolean_t __elfN(check_note)(struct image_params *imgp,
104 Elf_Brandnote *checknote, int32_t *osrel);
105 static vm_prot_t __elfN(trans_prot)(Elf_Word);
106 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
111 #define CORE_BUF_SIZE (16 * 1024)
113 int __elfN(fallback_brand) = -1;
114 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
115 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
116 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
118 static int elf_legacy_coredump = 0;
119 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
120 &elf_legacy_coredump, 0,
121 "include all and only RW pages in core dumps");
123 int __elfN(nxstack) =
124 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
125 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
134 #if __ELF_WORD_SIZE == 32
135 #if defined(__amd64__)
136 int i386_read_exec = 0;
137 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
138 "enable execution from readable segments");
142 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
144 #define trunc_page_ps(va, ps) rounddown2(va, ps)
145 #define round_page_ps(va, ps) roundup2(va, ps)
146 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
148 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
150 Elf_Brandnote __elfN(freebsd_brandnote) = {
151 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
152 .hdr.n_descsz = sizeof(int32_t),
153 .hdr.n_type = NT_FREEBSD_ABI_TAG,
154 .vendor = FREEBSD_ABI_VENDOR,
155 .flags = BN_TRANSLATE_OSREL,
156 .trans_osrel = __elfN(freebsd_trans_osrel)
160 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
164 p = (uintptr_t)(note + 1);
165 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
166 *osrel = *(const int32_t *)(p);
171 static const char GNU_ABI_VENDOR[] = "GNU";
172 static int GNU_KFREEBSD_ABI_DESC = 3;
174 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
175 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
176 .hdr.n_descsz = 16, /* XXX at least 16 */
178 .vendor = GNU_ABI_VENDOR,
179 .flags = BN_TRANSLATE_OSREL,
180 .trans_osrel = kfreebsd_trans_osrel
184 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
186 const Elf32_Word *desc;
189 p = (uintptr_t)(note + 1);
190 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
192 desc = (const Elf32_Word *)p;
193 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
197 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
198 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
200 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
206 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
210 for (i = 0; i < MAX_BRANDS; i++) {
211 if (elf_brand_list[i] == NULL) {
212 elf_brand_list[i] = entry;
216 if (i == MAX_BRANDS) {
217 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
225 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
229 for (i = 0; i < MAX_BRANDS; i++) {
230 if (elf_brand_list[i] == entry) {
231 elf_brand_list[i] = NULL;
241 __elfN(brand_inuse)(Elf_Brandinfo *entry)
246 sx_slock(&allproc_lock);
247 FOREACH_PROC_IN_SYSTEM(p) {
248 if (p->p_sysent == entry->sysvec) {
253 sx_sunlock(&allproc_lock);
258 static Elf_Brandinfo *
259 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
260 int interp_name_len, int32_t *osrel)
262 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
263 Elf_Brandinfo *bi, *bi_m;
268 * We support four types of branding -- (1) the ELF EI_OSABI field
269 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
270 * branding w/in the ELF header, (3) path of the `interp_path'
271 * field, and (4) the ".note.ABI-tag" ELF section.
274 /* Look for an ".note.ABI-tag" ELF section */
276 for (i = 0; i < MAX_BRANDS; i++) {
277 bi = elf_brand_list[i];
280 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
282 if (hdr->e_machine == bi->machine && (bi->flags &
283 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
284 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
285 /* Give brand a chance to veto check_note's guess */
286 if (ret && bi->header_supported)
287 ret = bi->header_supported(imgp);
289 * If note checker claimed the binary, but the
290 * interpreter path in the image does not
291 * match default one for the brand, try to
292 * search for other brands with the same
293 * interpreter. Either there is better brand
294 * with the right interpreter, or, failing
295 * this, we return first brand which accepted
296 * our note and, optionally, header.
298 if (ret && bi_m == NULL && interp != NULL &&
299 (bi->interp_path == NULL ||
300 (strlen(bi->interp_path) + 1 != interp_name_len ||
301 strncmp(interp, bi->interp_path, interp_name_len)
313 /* If the executable has a brand, search for it in the brand list. */
314 for (i = 0; i < MAX_BRANDS; i++) {
315 bi = elf_brand_list[i];
316 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
317 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
319 if (hdr->e_machine == bi->machine &&
320 (hdr->e_ident[EI_OSABI] == bi->brand ||
321 (bi->compat_3_brand != NULL &&
322 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
323 bi->compat_3_brand) == 0))) {
324 /* Looks good, but give brand a chance to veto */
325 if (!bi->header_supported ||
326 bi->header_supported(imgp)) {
328 * Again, prefer strictly matching
331 if (interp_name_len == 0 &&
332 bi->interp_path == NULL)
334 if (bi->interp_path != NULL &&
335 strlen(bi->interp_path) + 1 ==
336 interp_name_len && strncmp(interp,
337 bi->interp_path, interp_name_len) == 0)
347 /* No known brand, see if the header is recognized by any brand */
348 for (i = 0; i < MAX_BRANDS; i++) {
349 bi = elf_brand_list[i];
350 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
351 bi->header_supported == NULL)
353 if (hdr->e_machine == bi->machine) {
354 ret = bi->header_supported(imgp);
360 /* Lacking a known brand, search for a recognized interpreter. */
361 if (interp != NULL) {
362 for (i = 0; i < MAX_BRANDS; i++) {
363 bi = elf_brand_list[i];
364 if (bi == NULL || (bi->flags &
365 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
368 if (hdr->e_machine == bi->machine &&
369 bi->interp_path != NULL &&
370 /* ELF image p_filesz includes terminating zero */
371 strlen(bi->interp_path) + 1 == interp_name_len &&
372 strncmp(interp, bi->interp_path, interp_name_len)
378 /* Lacking a recognized interpreter, try the default brand */
379 for (i = 0; i < MAX_BRANDS; i++) {
380 bi = elf_brand_list[i];
381 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
382 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
384 if (hdr->e_machine == bi->machine &&
385 __elfN(fallback_brand) == bi->brand)
392 __elfN(check_header)(const Elf_Ehdr *hdr)
398 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
399 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
400 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
401 hdr->e_phentsize != sizeof(Elf_Phdr) ||
402 hdr->e_version != ELF_TARG_VER)
406 * Make sure we have at least one brand for this machine.
409 for (i = 0; i < MAX_BRANDS; i++) {
410 bi = elf_brand_list[i];
411 if (bi != NULL && bi->machine == hdr->e_machine)
421 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
422 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
429 * Create the page if it doesn't exist yet. Ignore errors.
431 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
432 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
435 * Find the page from the underlying object.
437 if (object != NULL) {
438 sf = vm_imgact_map_page(object, offset);
440 return (KERN_FAILURE);
441 off = offset - trunc_page(offset);
442 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
444 vm_imgact_unmap_page(sf);
446 return (KERN_FAILURE);
449 return (KERN_SUCCESS);
453 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
454 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
460 int error, locked, rv;
462 if (start != trunc_page(start)) {
463 rv = __elfN(map_partial)(map, object, offset, start,
464 round_page(start), prot);
465 if (rv != KERN_SUCCESS)
467 offset += round_page(start) - start;
468 start = round_page(start);
470 if (end != round_page(end)) {
471 rv = __elfN(map_partial)(map, object, offset +
472 trunc_page(end) - start, trunc_page(end), end, prot);
473 if (rv != KERN_SUCCESS)
475 end = trunc_page(end);
478 return (KERN_SUCCESS);
479 if ((offset & PAGE_MASK) != 0) {
481 * The mapping is not page aligned. This means that we have
484 rv = vm_map_fixed(map, NULL, 0, start, end - start,
485 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
486 if (rv != KERN_SUCCESS)
489 return (KERN_SUCCESS);
490 for (; start < end; start += sz) {
491 sf = vm_imgact_map_page(object, offset);
493 return (KERN_FAILURE);
494 off = offset - trunc_page(offset);
496 if (sz > PAGE_SIZE - off)
497 sz = PAGE_SIZE - off;
498 error = copyout((caddr_t)sf_buf_kva(sf) + off,
500 vm_imgact_unmap_page(sf);
502 return (KERN_FAILURE);
506 vm_object_reference(object);
507 rv = vm_map_fixed(map, object, offset, start, end - start,
508 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
509 if (rv != KERN_SUCCESS) {
510 locked = VOP_ISLOCKED(imgp->vp);
511 VOP_UNLOCK(imgp->vp, 0);
512 vm_object_deallocate(object);
513 vn_lock(imgp->vp, locked | LK_RETRY);
517 return (KERN_SUCCESS);
521 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
522 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
529 vm_offset_t off, map_addr;
532 vm_ooffset_t file_addr;
535 * It's necessary to fail if the filsz + offset taken from the
536 * header is greater than the actual file pager object's size.
537 * If we were to allow this, then the vm_map_find() below would
538 * walk right off the end of the file object and into the ether.
540 * While I'm here, might as well check for something else that
541 * is invalid: filsz cannot be greater than memsz.
543 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
545 uprintf("elf_load_section: truncated ELF file\n");
549 object = imgp->object;
550 map = &imgp->proc->p_vmspace->vm_map;
551 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
552 file_addr = trunc_page_ps(offset, pagesize);
555 * We have two choices. We can either clear the data in the last page
556 * of an oversized mapping, or we can start the anon mapping a page
557 * early and copy the initialized data into that first page. We
562 else if (memsz > filsz)
563 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
565 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
568 /* cow flags: don't dump readonly sections in core */
569 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
570 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
572 rv = __elfN(map_insert)(imgp, map,
574 file_addr, /* file offset */
575 map_addr, /* virtual start */
576 map_addr + map_len,/* virtual end */
579 if (rv != KERN_SUCCESS)
582 /* we can stop now if we've covered it all */
589 * We have to get the remaining bit of the file into the first part
590 * of the oversized map segment. This is normally because the .data
591 * segment in the file is extended to provide bss. It's a neat idea
592 * to try and save a page, but it's a pain in the behind to implement.
594 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
596 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
597 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
600 /* This had damn well better be true! */
602 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
603 map_addr + map_len, prot, 0);
604 if (rv != KERN_SUCCESS)
609 sf = vm_imgact_map_page(object, offset + filsz);
613 /* send the page fragment to user space */
614 off = trunc_page_ps(offset + filsz, pagesize) -
615 trunc_page(offset + filsz);
616 error = copyout((caddr_t)sf_buf_kva(sf) + off,
617 (caddr_t)map_addr, copy_len);
618 vm_imgact_unmap_page(sf);
624 * Remove write access to the page if it was only granted by map_insert
627 if ((prot & VM_PROT_WRITE) == 0)
628 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
629 map_len), prot, FALSE);
635 * Load the file "file" into memory. It may be either a shared object
638 * The "addr" reference parameter is in/out. On entry, it specifies
639 * the address where a shared object should be loaded. If the file is
640 * an executable, this value is ignored. On exit, "addr" specifies
641 * where the file was actually loaded.
643 * The "entry" reference parameter is out only. On exit, it specifies
644 * the entry point for the loaded file.
647 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
648 u_long *entry, size_t pagesize)
653 struct image_params image_params;
655 const Elf_Ehdr *hdr = NULL;
656 const Elf_Phdr *phdr = NULL;
657 struct nameidata *nd;
659 struct image_params *imgp;
662 u_long base_addr = 0;
663 int error, i, numsegs;
665 #ifdef CAPABILITY_MODE
667 * XXXJA: This check can go away once we are sufficiently confident
668 * that the checks in namei() are correct.
670 if (IN_CAPABILITY_MODE(curthread))
674 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
676 attr = &tempdata->attr;
677 imgp = &tempdata->image_params;
680 * Initialize part of the common data
684 imgp->firstpage = NULL;
685 imgp->image_header = NULL;
687 imgp->execlabel = NULL;
689 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
690 if ((error = namei(nd)) != 0) {
694 NDFREE(nd, NDF_ONLY_PNBUF);
695 imgp->vp = nd->ni_vp;
698 * Check permissions, modes, uid, etc on the file, and "open" it.
700 error = exec_check_permissions(imgp);
704 error = exec_map_first_page(imgp);
709 * Also make certain that the interpreter stays the same, so set
710 * its VV_TEXT flag, too.
712 VOP_SET_TEXT(nd->ni_vp);
714 imgp->object = nd->ni_vp->v_object;
716 hdr = (const Elf_Ehdr *)imgp->image_header;
717 if ((error = __elfN(check_header)(hdr)) != 0)
719 if (hdr->e_type == ET_DYN)
721 else if (hdr->e_type == ET_EXEC)
728 /* Only support headers that fit within first page for now */
729 if ((hdr->e_phoff > PAGE_SIZE) ||
730 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
735 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
736 if (!aligned(phdr, Elf_Addr)) {
741 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
742 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
743 /* Loadable segment */
744 prot = __elfN(trans_prot)(phdr[i].p_flags);
745 error = __elfN(load_section)(imgp, phdr[i].p_offset,
746 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
747 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
751 * Establish the base address if this is the
755 base_addr = trunc_page(phdr[i].p_vaddr +
761 *entry = (unsigned long)hdr->e_entry + rbase;
765 exec_unmap_first_page(imgp);
770 free(tempdata, M_TEMP);
776 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
780 const Elf_Phdr *phdr;
781 Elf_Auxargs *elf_auxargs;
782 struct vmspace *vmspace;
783 const char *err_str, *newinterp;
784 char *interp, *interp_buf, *path;
785 Elf_Brandinfo *brand_info;
786 struct sysentvec *sv;
788 u_long text_size, data_size, total_size, text_addr, data_addr;
789 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
791 int error, i, n, interp_name_len, have_interp;
793 hdr = (const Elf_Ehdr *)imgp->image_header;
796 * Do we have a valid ELF header ?
798 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
799 * if particular brand doesn't support it.
801 if (__elfN(check_header)(hdr) != 0 ||
802 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
806 * From here on down, we return an errno, not -1, as we've
807 * detected an ELF file.
810 if ((hdr->e_phoff > PAGE_SIZE) ||
811 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
812 /* Only support headers in first page for now */
813 uprintf("Program headers not in the first page\n");
816 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
817 if (!aligned(phdr, Elf_Addr)) {
818 uprintf("Unaligned program headers\n");
825 text_size = data_size = total_size = text_addr = data_addr = 0;
828 err_str = newinterp = NULL;
829 interp = interp_buf = NULL;
832 for (i = 0; i < hdr->e_phnum; i++) {
833 switch (phdr[i].p_type) {
836 baddr = phdr[i].p_vaddr;
840 /* Path to interpreter */
841 if (phdr[i].p_filesz > MAXPATHLEN) {
842 uprintf("Invalid PT_INTERP\n");
846 if (interp != NULL) {
847 uprintf("Multiple PT_INTERP headers\n");
851 interp_name_len = phdr[i].p_filesz;
852 if (phdr[i].p_offset > PAGE_SIZE ||
853 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
854 VOP_UNLOCK(imgp->vp, 0);
855 interp_buf = malloc(interp_name_len + 1, M_TEMP,
857 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
858 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
859 interp_name_len, phdr[i].p_offset,
860 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
863 uprintf("i/o error PT_INTERP\n");
866 interp_buf[interp_name_len] = '\0';
869 interp = __DECONST(char *, imgp->image_header) +
876 __elfN(trans_prot)(phdr[i].p_flags);
877 imgp->stack_sz = phdr[i].p_memsz;
882 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
884 if (brand_info == NULL) {
885 uprintf("ELF binary type \"%u\" not known.\n",
886 hdr->e_ident[EI_OSABI]);
891 if (hdr->e_type == ET_DYN) {
892 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
893 uprintf("Cannot execute shared object\n");
898 * Honour the base load address from the dso if it is
899 * non-zero for some reason.
902 et_dyn_addr = ET_DYN_LOAD_ADDR;
904 sv = brand_info->sysvec;
905 if (interp != NULL && brand_info->interp_newpath != NULL)
906 newinterp = brand_info->interp_newpath;
909 * Avoid a possible deadlock if the current address space is destroyed
910 * and that address space maps the locked vnode. In the common case,
911 * the locked vnode's v_usecount is decremented but remains greater
912 * than zero. Consequently, the vnode lock is not needed by vrele().
913 * However, in cases where the vnode lock is external, such as nullfs,
914 * v_usecount may become zero.
916 * The VV_TEXT flag prevents modifications to the executable while
917 * the vnode is unlocked.
919 VOP_UNLOCK(imgp->vp, 0);
921 error = exec_new_vmspace(imgp, sv);
922 imgp->proc->p_sysent = sv;
924 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
928 for (i = 0; i < hdr->e_phnum; i++) {
929 switch (phdr[i].p_type) {
930 case PT_LOAD: /* Loadable segment */
931 if (phdr[i].p_memsz == 0)
933 prot = __elfN(trans_prot)(phdr[i].p_flags);
934 error = __elfN(load_section)(imgp, phdr[i].p_offset,
935 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
936 phdr[i].p_memsz, phdr[i].p_filesz, prot,
942 * If this segment contains the program headers,
943 * remember their virtual address for the AT_PHDR
944 * aux entry. Static binaries don't usually include
947 if (phdr[i].p_offset == 0 &&
948 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
950 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
953 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
954 seg_size = round_page(phdr[i].p_memsz +
955 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
958 * Make the largest executable segment the official
959 * text segment and all others data.
961 * Note that obreak() assumes that data_addr +
962 * data_size == end of data load area, and the ELF
963 * file format expects segments to be sorted by
964 * address. If multiple data segments exist, the
965 * last one will be used.
968 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
969 text_size = seg_size;
970 text_addr = seg_addr;
972 data_size = seg_size;
973 data_addr = seg_addr;
975 total_size += seg_size;
977 case PT_PHDR: /* Program header table info */
978 proghdr = phdr[i].p_vaddr + et_dyn_addr;
985 if (data_addr == 0 && data_size == 0) {
986 data_addr = text_addr;
987 data_size = text_size;
990 entry = (u_long)hdr->e_entry + et_dyn_addr;
993 * Check limits. It should be safe to check the
994 * limits after loading the segments since we do
995 * not actually fault in all the segments pages.
997 PROC_LOCK(imgp->proc);
998 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
999 err_str = "Data segment size exceeds process limit";
1000 else if (text_size > maxtsiz)
1001 err_str = "Text segment size exceeds system limit";
1002 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1003 err_str = "Total segment size exceeds process limit";
1004 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1005 err_str = "Data segment size exceeds resource limit";
1006 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1007 err_str = "Total segment size exceeds resource limit";
1008 if (err_str != NULL) {
1009 PROC_UNLOCK(imgp->proc);
1010 uprintf("%s\n", err_str);
1015 vmspace = imgp->proc->p_vmspace;
1016 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1017 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1018 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1019 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1022 * We load the dynamic linker where a userland call
1023 * to mmap(0, ...) would put it. The rationale behind this
1024 * calculation is that it leaves room for the heap to grow to
1025 * its maximum allowed size.
1027 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1029 PROC_UNLOCK(imgp->proc);
1031 imgp->entry_addr = entry;
1033 if (interp != NULL) {
1034 have_interp = FALSE;
1035 VOP_UNLOCK(imgp->vp, 0);
1036 if (brand_info->emul_path != NULL &&
1037 brand_info->emul_path[0] != '\0') {
1038 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1039 snprintf(path, MAXPATHLEN, "%s%s",
1040 brand_info->emul_path, interp);
1041 error = __elfN(load_file)(imgp->proc, path, &addr,
1042 &imgp->entry_addr, sv->sv_pagesize);
1047 if (!have_interp && newinterp != NULL &&
1048 (brand_info->interp_path == NULL ||
1049 strcmp(interp, brand_info->interp_path) == 0)) {
1050 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1051 &imgp->entry_addr, sv->sv_pagesize);
1056 error = __elfN(load_file)(imgp->proc, interp, &addr,
1057 &imgp->entry_addr, sv->sv_pagesize);
1059 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1061 uprintf("ELF interpreter %s not found, error %d\n",
1069 * Construct auxargs table (used by the fixup routine)
1071 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1072 elf_auxargs->execfd = -1;
1073 elf_auxargs->phdr = proghdr;
1074 elf_auxargs->phent = hdr->e_phentsize;
1075 elf_auxargs->phnum = hdr->e_phnum;
1076 elf_auxargs->pagesz = PAGE_SIZE;
1077 elf_auxargs->base = addr;
1078 elf_auxargs->flags = 0;
1079 elf_auxargs->entry = entry;
1080 elf_auxargs->hdr_eflags = hdr->e_flags;
1082 imgp->auxargs = elf_auxargs;
1083 imgp->interpreted = 0;
1084 imgp->reloc_base = addr;
1085 imgp->proc->p_osrel = osrel;
1086 imgp->proc->p_elf_machine = hdr->e_machine;
1087 imgp->proc->p_elf_flags = hdr->e_flags;
1090 free(interp_buf, M_TEMP);
1094 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1097 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1099 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1103 base = (Elf_Addr *)*stack_base;
1104 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1106 if (args->execfd != -1)
1107 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1108 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1109 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1110 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1111 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1112 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1113 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1114 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1115 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1116 if (imgp->execpathp != 0)
1117 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1118 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1119 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1120 if (imgp->canary != 0) {
1121 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1122 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1124 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1125 if (imgp->pagesizes != 0) {
1126 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1127 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1129 if (imgp->sysent->sv_timekeep_base != 0) {
1130 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1131 imgp->sysent->sv_timekeep_base);
1133 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1134 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1135 imgp->sysent->sv_stackprot);
1136 if (imgp->sysent->sv_hwcap != NULL)
1137 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1138 if (imgp->sysent->sv_hwcap2 != NULL)
1139 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1140 AUXARGS_ENTRY(pos, AT_NULL, 0);
1142 free(imgp->auxargs, M_TEMP);
1143 imgp->auxargs = NULL;
1146 suword(base, (long)imgp->args->argc);
1147 *stack_base = (register_t *)base;
1152 * Code for generating ELF core dumps.
1155 typedef void (*segment_callback)(vm_map_entry_t, void *);
1157 /* Closure for cb_put_phdr(). */
1158 struct phdr_closure {
1159 Elf_Phdr *phdr; /* Program header to fill in */
1160 Elf_Off offset; /* Offset of segment in core file */
1163 /* Closure for cb_size_segment(). */
1164 struct sseg_closure {
1165 int count; /* Count of writable segments. */
1166 size_t size; /* Total size of all writable segments. */
1169 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1172 int type; /* Note type. */
1173 outfunc_t outfunc; /* Output function. */
1174 void *outarg; /* Argument for the output function. */
1175 size_t outsize; /* Output size. */
1176 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1179 TAILQ_HEAD(note_info_list, note_info);
1181 /* Coredump output parameters. */
1182 struct coredump_params {
1184 struct ucred *active_cred;
1185 struct ucred *file_cred;
1188 struct gzio_stream *gzs;
1191 static void cb_put_phdr(vm_map_entry_t, void *);
1192 static void cb_size_segment(vm_map_entry_t, void *);
1193 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1195 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1196 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1197 struct note_info_list *, size_t);
1198 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1200 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1201 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1202 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1203 static int sbuf_drain_core_output(void *, const char *, int);
1204 static int sbuf_drain_count(void *arg, const char *data, int len);
1206 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1207 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1208 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1209 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1210 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1211 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1215 static void note_procstat_files(void *, struct sbuf *, size_t *);
1216 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1217 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1218 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1219 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1220 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1223 extern int compress_user_cores_gzlevel;
1226 * Write out a core segment to the compression stream.
1229 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1235 chunk_len = MIN(len, CORE_BUF_SIZE);
1238 * We can get EFAULT error here.
1239 * In that case zero out the current chunk of the segment.
1241 error = copyin(base, buf, chunk_len);
1243 bzero(buf, chunk_len);
1244 error = gzio_write(p->gzs, buf, chunk_len);
1254 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1257 return (core_write((struct coredump_params *)arg, base, len, offset,
1263 core_write(struct coredump_params *p, const void *base, size_t len,
1264 off_t offset, enum uio_seg seg)
1267 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1268 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1269 p->active_cred, p->file_cred, NULL, p->td));
1273 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1280 return (compress_chunk(p, base, tmpbuf, len));
1283 * EFAULT is a non-fatal error that we can get, for example,
1284 * if the segment is backed by a file but extends beyond its
1287 error = core_write(p, base, len, offset, UIO_USERSPACE);
1288 if (error == EFAULT) {
1289 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1290 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1291 "for process %s\n", base, len, offset, curproc->p_comm);
1294 * Write a "real" zero byte at the end of the target region
1295 * in the case this is the last segment.
1296 * The intermediate space will be implicitly zero-filled.
1298 error = core_write(p, zero_region, 1, offset + len - 1,
1305 * Drain into a core file.
1308 sbuf_drain_core_output(void *arg, const char *data, int len)
1310 struct coredump_params *p;
1313 p = (struct coredump_params *)arg;
1316 * Some kern_proc out routines that print to this sbuf may
1317 * call us with the process lock held. Draining with the
1318 * non-sleepable lock held is unsafe. The lock is needed for
1319 * those routines when dumping a live process. In our case we
1320 * can safely release the lock before draining and acquire
1323 locked = PROC_LOCKED(p->td->td_proc);
1325 PROC_UNLOCK(p->td->td_proc);
1328 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1331 error = core_write(p, __DECONST(void *, data), len, p->offset,
1334 PROC_LOCK(p->td->td_proc);
1342 * Drain into a counter.
1345 sbuf_drain_count(void *arg, const char *data __unused, int len)
1349 sizep = (size_t *)arg;
1355 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1357 struct ucred *cred = td->td_ucred;
1359 struct sseg_closure seginfo;
1360 struct note_info_list notelst;
1361 struct coredump_params params;
1362 struct note_info *ninfo;
1364 size_t hdrsize, notesz, coresize;
1368 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1372 TAILQ_INIT(¬elst);
1374 /* Size the program segments. */
1377 each_dumpable_segment(td, cb_size_segment, &seginfo);
1380 * Collect info about the core file header area.
1382 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1383 if (seginfo.count + 1 >= PN_XNUM)
1384 hdrsize += sizeof(Elf_Shdr);
1385 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1386 coresize = round_page(hdrsize + notesz) + seginfo.size;
1388 /* Set up core dump parameters. */
1390 params.active_cred = cred;
1391 params.file_cred = NOCRED;
1398 PROC_LOCK(td->td_proc);
1399 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1400 PROC_UNLOCK(td->td_proc);
1407 if (coresize >= limit) {
1413 /* Create a compression stream if necessary. */
1415 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1416 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1417 if (params.gzs == NULL) {
1421 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1426 * Allocate memory for building the header, fill it up,
1427 * and write it out following the notes.
1429 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1430 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1433 /* Write the contents of all of the writable segments. */
1439 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1440 offset = round_page(hdrsize + notesz);
1441 for (i = 0; i < seginfo.count; i++) {
1442 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1443 php->p_filesz, offset, ¶ms, tmpbuf);
1446 offset += php->p_filesz;
1450 if (error == 0 && compress)
1451 error = gzio_flush(params.gzs);
1456 "Failed to write core file for process %s (error %d)\n",
1457 curproc->p_comm, error);
1463 free(tmpbuf, M_TEMP);
1464 if (params.gzs != NULL)
1465 gzio_fini(params.gzs);
1468 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1469 TAILQ_REMOVE(¬elst, ninfo, link);
1470 free(ninfo, M_TEMP);
1479 * A callback for each_dumpable_segment() to write out the segment's
1480 * program header entry.
1483 cb_put_phdr(entry, closure)
1484 vm_map_entry_t entry;
1487 struct phdr_closure *phc = (struct phdr_closure *)closure;
1488 Elf_Phdr *phdr = phc->phdr;
1490 phc->offset = round_page(phc->offset);
1492 phdr->p_type = PT_LOAD;
1493 phdr->p_offset = phc->offset;
1494 phdr->p_vaddr = entry->start;
1496 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1497 phdr->p_align = PAGE_SIZE;
1498 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1500 phc->offset += phdr->p_filesz;
1505 * A callback for each_dumpable_segment() to gather information about
1506 * the number of segments and their total size.
1509 cb_size_segment(vm_map_entry_t entry, void *closure)
1511 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1514 ssc->size += entry->end - entry->start;
1518 * For each writable segment in the process's memory map, call the given
1519 * function with a pointer to the map entry and some arbitrary
1520 * caller-supplied data.
1523 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1525 struct proc *p = td->td_proc;
1526 vm_map_t map = &p->p_vmspace->vm_map;
1527 vm_map_entry_t entry;
1528 vm_object_t backing_object, object;
1529 boolean_t ignore_entry;
1531 vm_map_lock_read(map);
1532 for (entry = map->header.next; entry != &map->header;
1533 entry = entry->next) {
1535 * Don't dump inaccessible mappings, deal with legacy
1538 * Note that read-only segments related to the elf binary
1539 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1540 * need to arbitrarily ignore such segments.
1542 if (elf_legacy_coredump) {
1543 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1546 if ((entry->protection & VM_PROT_ALL) == 0)
1551 * Dont include memory segment in the coredump if
1552 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1553 * madvise(2). Do not dump submaps (i.e. parts of the
1556 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1559 if ((object = entry->object.vm_object) == NULL)
1562 /* Ignore memory-mapped devices and such things. */
1563 VM_OBJECT_RLOCK(object);
1564 while ((backing_object = object->backing_object) != NULL) {
1565 VM_OBJECT_RLOCK(backing_object);
1566 VM_OBJECT_RUNLOCK(object);
1567 object = backing_object;
1569 ignore_entry = object->type != OBJT_DEFAULT &&
1570 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1571 object->type != OBJT_PHYS;
1572 VM_OBJECT_RUNLOCK(object);
1576 (*func)(entry, closure);
1578 vm_map_unlock_read(map);
1582 * Write the core file header to the file, including padding up to
1583 * the page boundary.
1586 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1587 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1589 struct note_info *ninfo;
1593 /* Fill in the header. */
1594 bzero(hdr, hdrsize);
1595 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1597 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1598 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1599 sbuf_start_section(sb, NULL);
1600 sbuf_bcat(sb, hdr, hdrsize);
1601 TAILQ_FOREACH(ninfo, notelst, link)
1602 __elfN(putnote)(ninfo, sb);
1603 /* Align up to a page boundary for the program segments. */
1604 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1605 error = sbuf_finish(sb);
1612 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1622 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1625 * To have the debugger select the right thread (LWP) as the initial
1626 * thread, we dump the state of the thread passed to us in td first.
1627 * This is the thread that causes the core dump and thus likely to
1628 * be the right thread one wants to have selected in the debugger.
1631 while (thr != NULL) {
1632 size += register_note(list, NT_PRSTATUS,
1633 __elfN(note_prstatus), thr);
1634 size += register_note(list, NT_FPREGSET,
1635 __elfN(note_fpregset), thr);
1636 size += register_note(list, NT_THRMISC,
1637 __elfN(note_thrmisc), thr);
1638 size += register_note(list, NT_PTLWPINFO,
1639 __elfN(note_ptlwpinfo), thr);
1640 size += register_note(list, -1,
1641 __elfN(note_threadmd), thr);
1643 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1644 TAILQ_NEXT(thr, td_plist);
1646 thr = TAILQ_NEXT(thr, td_plist);
1649 size += register_note(list, NT_PROCSTAT_PROC,
1650 __elfN(note_procstat_proc), p);
1651 size += register_note(list, NT_PROCSTAT_FILES,
1652 note_procstat_files, p);
1653 size += register_note(list, NT_PROCSTAT_VMMAP,
1654 note_procstat_vmmap, p);
1655 size += register_note(list, NT_PROCSTAT_GROUPS,
1656 note_procstat_groups, p);
1657 size += register_note(list, NT_PROCSTAT_UMASK,
1658 note_procstat_umask, p);
1659 size += register_note(list, NT_PROCSTAT_RLIMIT,
1660 note_procstat_rlimit, p);
1661 size += register_note(list, NT_PROCSTAT_OSREL,
1662 note_procstat_osrel, p);
1663 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1664 __elfN(note_procstat_psstrings), p);
1665 size += register_note(list, NT_PROCSTAT_AUXV,
1666 __elfN(note_procstat_auxv), p);
1672 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1678 struct phdr_closure phc;
1680 ehdr = (Elf_Ehdr *)hdr;
1682 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1683 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1684 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1685 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1686 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1687 ehdr->e_ident[EI_DATA] = ELF_DATA;
1688 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1689 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1690 ehdr->e_ident[EI_ABIVERSION] = 0;
1691 ehdr->e_ident[EI_PAD] = 0;
1692 ehdr->e_type = ET_CORE;
1693 ehdr->e_machine = td->td_proc->p_elf_machine;
1694 ehdr->e_version = EV_CURRENT;
1696 ehdr->e_phoff = sizeof(Elf_Ehdr);
1697 ehdr->e_flags = td->td_proc->p_elf_flags;
1698 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1699 ehdr->e_phentsize = sizeof(Elf_Phdr);
1700 ehdr->e_shentsize = sizeof(Elf_Shdr);
1701 ehdr->e_shstrndx = SHN_UNDEF;
1702 if (numsegs + 1 < PN_XNUM) {
1703 ehdr->e_phnum = numsegs + 1;
1706 ehdr->e_phnum = PN_XNUM;
1709 ehdr->e_shoff = ehdr->e_phoff +
1710 (numsegs + 1) * ehdr->e_phentsize;
1711 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1712 ("e_shoff: %zu, hdrsize - shdr: %zu",
1713 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1715 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1716 memset(shdr, 0, sizeof(*shdr));
1718 * A special first section is used to hold large segment and
1719 * section counts. This was proposed by Sun Microsystems in
1720 * Solaris and has been adopted by Linux; the standard ELF
1721 * tools are already familiar with the technique.
1723 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1724 * (or 12-7 depending on the version of the document) for more
1727 shdr->sh_type = SHT_NULL;
1728 shdr->sh_size = ehdr->e_shnum;
1729 shdr->sh_link = ehdr->e_shstrndx;
1730 shdr->sh_info = numsegs + 1;
1734 * Fill in the program header entries.
1736 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1738 /* The note segement. */
1739 phdr->p_type = PT_NOTE;
1740 phdr->p_offset = hdrsize;
1743 phdr->p_filesz = notesz;
1745 phdr->p_flags = PF_R;
1746 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1749 /* All the writable segments from the program. */
1751 phc.offset = round_page(hdrsize + notesz);
1752 each_dumpable_segment(td, cb_put_phdr, &phc);
1756 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1758 struct note_info *ninfo;
1759 size_t size, notesize;
1762 out(arg, NULL, &size);
1763 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1765 ninfo->outfunc = out;
1766 ninfo->outarg = arg;
1767 ninfo->outsize = size;
1768 TAILQ_INSERT_TAIL(list, ninfo, link);
1773 notesize = sizeof(Elf_Note) + /* note header */
1774 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1776 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1782 append_note_data(const void *src, void *dst, size_t len)
1786 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1788 bcopy(src, dst, len);
1789 bzero((char *)dst + len, padded_len - len);
1791 return (padded_len);
1795 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1803 note = (Elf_Note *)buf;
1804 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1805 note->n_descsz = size;
1806 note->n_type = type;
1807 buf += sizeof(*note);
1808 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1809 sizeof(FREEBSD_ABI_VENDOR));
1810 append_note_data(src, buf, size);
1815 notesize = sizeof(Elf_Note) + /* note header */
1816 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1818 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1824 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1827 ssize_t old_len, sect_len;
1828 size_t new_len, descsz, i;
1830 if (ninfo->type == -1) {
1831 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1835 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1836 note.n_descsz = ninfo->outsize;
1837 note.n_type = ninfo->type;
1839 sbuf_bcat(sb, ¬e, sizeof(note));
1840 sbuf_start_section(sb, &old_len);
1841 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1842 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1843 if (note.n_descsz == 0)
1845 sbuf_start_section(sb, &old_len);
1846 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1847 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1851 new_len = (size_t)sect_len;
1852 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1853 if (new_len < descsz) {
1855 * It is expected that individual note emitters will correctly
1856 * predict their expected output size and fill up to that size
1857 * themselves, padding in a format-specific way if needed.
1858 * However, in case they don't, just do it here with zeros.
1860 for (i = 0; i < descsz - new_len; i++)
1862 } else if (new_len > descsz) {
1864 * We can't always truncate sb -- we may have drained some
1867 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1868 "read it (%zu > %zu). Since it is longer than "
1869 "expected, this coredump's notes are corrupt. THIS "
1870 "IS A BUG in the note_procstat routine for type %u.\n",
1871 __func__, (unsigned)note.n_type, new_len, descsz,
1872 (unsigned)note.n_type));
1877 * Miscellaneous note out functions.
1880 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1881 #include <compat/freebsd32/freebsd32.h>
1882 #include <compat/freebsd32/freebsd32_signal.h>
1884 typedef struct prstatus32 elf_prstatus_t;
1885 typedef struct prpsinfo32 elf_prpsinfo_t;
1886 typedef struct fpreg32 elf_prfpregset_t;
1887 typedef struct fpreg32 elf_fpregset_t;
1888 typedef struct reg32 elf_gregset_t;
1889 typedef struct thrmisc32 elf_thrmisc_t;
1890 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1891 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1892 typedef uint32_t elf_ps_strings_t;
1894 typedef prstatus_t elf_prstatus_t;
1895 typedef prpsinfo_t elf_prpsinfo_t;
1896 typedef prfpregset_t elf_prfpregset_t;
1897 typedef prfpregset_t elf_fpregset_t;
1898 typedef gregset_t elf_gregset_t;
1899 typedef thrmisc_t elf_thrmisc_t;
1900 #define ELF_KERN_PROC_MASK 0
1901 typedef struct kinfo_proc elf_kinfo_proc_t;
1902 typedef vm_offset_t elf_ps_strings_t;
1906 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1912 elf_prpsinfo_t *psinfo;
1915 p = (struct proc *)arg;
1917 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1918 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1919 psinfo->pr_version = PRPSINFO_VERSION;
1920 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1921 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1923 if (p->p_args != NULL) {
1924 len = sizeof(psinfo->pr_psargs) - 1;
1925 if (len > p->p_args->ar_length)
1926 len = p->p_args->ar_length;
1927 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1933 sbuf_new(&sbarg, psinfo->pr_psargs,
1934 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1935 error = proc_getargv(curthread, p, &sbarg);
1937 if (sbuf_finish(&sbarg) == 0)
1938 len = sbuf_len(&sbarg) - 1;
1940 len = sizeof(psinfo->pr_psargs) - 1;
1941 sbuf_delete(&sbarg);
1943 if (error || len == 0)
1944 strlcpy(psinfo->pr_psargs, p->p_comm,
1945 sizeof(psinfo->pr_psargs));
1947 KASSERT(len < sizeof(psinfo->pr_psargs),
1948 ("len is too long: %zu vs %zu", len,
1949 sizeof(psinfo->pr_psargs)));
1950 cp = psinfo->pr_psargs;
1953 cp = memchr(cp, '\0', end - cp);
1959 psinfo->pr_pid = p->p_pid;
1960 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1961 free(psinfo, M_TEMP);
1963 *sizep = sizeof(*psinfo);
1967 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1970 elf_prstatus_t *status;
1972 td = (struct thread *)arg;
1974 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1975 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1976 status->pr_version = PRSTATUS_VERSION;
1977 status->pr_statussz = sizeof(elf_prstatus_t);
1978 status->pr_gregsetsz = sizeof(elf_gregset_t);
1979 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1980 status->pr_osreldate = osreldate;
1981 status->pr_cursig = td->td_proc->p_sig;
1982 status->pr_pid = td->td_tid;
1983 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1984 fill_regs32(td, &status->pr_reg);
1986 fill_regs(td, &status->pr_reg);
1988 sbuf_bcat(sb, status, sizeof(*status));
1989 free(status, M_TEMP);
1991 *sizep = sizeof(*status);
1995 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1998 elf_prfpregset_t *fpregset;
2000 td = (struct thread *)arg;
2002 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2003 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2004 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2005 fill_fpregs32(td, fpregset);
2007 fill_fpregs(td, fpregset);
2009 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2010 free(fpregset, M_TEMP);
2012 *sizep = sizeof(*fpregset);
2016 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2019 elf_thrmisc_t thrmisc;
2021 td = (struct thread *)arg;
2023 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2024 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2025 strcpy(thrmisc.pr_tname, td->td_name);
2026 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2028 *sizep = sizeof(thrmisc);
2032 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2037 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2038 struct ptrace_lwpinfo32 pl;
2040 struct ptrace_lwpinfo pl;
2043 td = (struct thread *)arg;
2044 size = sizeof(structsize) + sizeof(pl);
2046 KASSERT(*sizep == size, ("invalid size"));
2047 structsize = sizeof(pl);
2048 sbuf_bcat(sb, &structsize, sizeof(structsize));
2049 bzero(&pl, sizeof(pl));
2050 pl.pl_lwpid = td->td_tid;
2051 pl.pl_event = PL_EVENT_NONE;
2052 pl.pl_sigmask = td->td_sigmask;
2053 pl.pl_siglist = td->td_siglist;
2054 if (td->td_si.si_signo != 0) {
2055 pl.pl_event = PL_EVENT_SIGNAL;
2056 pl.pl_flags |= PL_FLAG_SI;
2057 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2058 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2060 pl.pl_siginfo = td->td_si;
2063 strcpy(pl.pl_tdname, td->td_name);
2064 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2065 sbuf_bcat(sb, &pl, sizeof(pl));
2071 * Allow for MD specific notes, as well as any MD
2072 * specific preparations for writing MI notes.
2075 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2081 td = (struct thread *)arg;
2083 if (size != 0 && sb != NULL)
2084 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2088 __elfN(dump_thread)(td, buf, &size);
2089 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2090 if (size != 0 && sb != NULL)
2091 sbuf_bcat(sb, buf, size);
2096 #ifdef KINFO_PROC_SIZE
2097 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2101 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2107 p = (struct proc *)arg;
2108 size = sizeof(structsize) + p->p_numthreads *
2109 sizeof(elf_kinfo_proc_t);
2112 KASSERT(*sizep == size, ("invalid size"));
2113 structsize = sizeof(elf_kinfo_proc_t);
2114 sbuf_bcat(sb, &structsize, sizeof(structsize));
2115 sx_slock(&proctree_lock);
2117 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2118 sx_sunlock(&proctree_lock);
2123 #ifdef KINFO_FILE_SIZE
2124 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2128 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2131 size_t size, sect_sz, i;
2132 ssize_t start_len, sect_len;
2133 int structsize, filedesc_flags;
2135 if (coredump_pack_fileinfo)
2136 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2140 p = (struct proc *)arg;
2141 structsize = sizeof(struct kinfo_file);
2144 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2145 sbuf_set_drain(sb, sbuf_drain_count, &size);
2146 sbuf_bcat(sb, &structsize, sizeof(structsize));
2148 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2153 sbuf_start_section(sb, &start_len);
2155 sbuf_bcat(sb, &structsize, sizeof(structsize));
2157 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2160 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2165 KASSERT(sect_sz <= *sizep,
2166 ("kern_proc_filedesc_out did not respect maxlen; "
2167 "requested %zu, got %zu", *sizep - sizeof(structsize),
2168 sect_sz - sizeof(structsize)));
2170 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2175 #ifdef KINFO_VMENTRY_SIZE
2176 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2180 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2184 int structsize, vmmap_flags;
2186 if (coredump_pack_vmmapinfo)
2187 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2191 p = (struct proc *)arg;
2192 structsize = sizeof(struct kinfo_vmentry);
2195 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2196 sbuf_set_drain(sb, sbuf_drain_count, &size);
2197 sbuf_bcat(sb, &structsize, sizeof(structsize));
2199 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2204 sbuf_bcat(sb, &structsize, sizeof(structsize));
2206 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2212 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2218 p = (struct proc *)arg;
2219 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2221 KASSERT(*sizep == size, ("invalid size"));
2222 structsize = sizeof(gid_t);
2223 sbuf_bcat(sb, &structsize, sizeof(structsize));
2224 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2231 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2237 p = (struct proc *)arg;
2238 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2240 KASSERT(*sizep == size, ("invalid size"));
2241 structsize = sizeof(p->p_fd->fd_cmask);
2242 sbuf_bcat(sb, &structsize, sizeof(structsize));
2243 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2249 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2252 struct rlimit rlim[RLIM_NLIMITS];
2256 p = (struct proc *)arg;
2257 size = sizeof(structsize) + sizeof(rlim);
2259 KASSERT(*sizep == size, ("invalid size"));
2260 structsize = sizeof(rlim);
2261 sbuf_bcat(sb, &structsize, sizeof(structsize));
2263 for (i = 0; i < RLIM_NLIMITS; i++)
2264 lim_rlimit_proc(p, i, &rlim[i]);
2266 sbuf_bcat(sb, rlim, sizeof(rlim));
2272 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2278 p = (struct proc *)arg;
2279 size = sizeof(structsize) + sizeof(p->p_osrel);
2281 KASSERT(*sizep == size, ("invalid size"));
2282 structsize = sizeof(p->p_osrel);
2283 sbuf_bcat(sb, &structsize, sizeof(structsize));
2284 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2290 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2293 elf_ps_strings_t ps_strings;
2297 p = (struct proc *)arg;
2298 size = sizeof(structsize) + sizeof(ps_strings);
2300 KASSERT(*sizep == size, ("invalid size"));
2301 structsize = sizeof(ps_strings);
2302 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2303 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2305 ps_strings = p->p_sysent->sv_psstrings;
2307 sbuf_bcat(sb, &structsize, sizeof(structsize));
2308 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2314 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2320 p = (struct proc *)arg;
2323 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2324 sbuf_set_drain(sb, sbuf_drain_count, &size);
2325 sbuf_bcat(sb, &structsize, sizeof(structsize));
2327 proc_getauxv(curthread, p, sb);
2333 structsize = sizeof(Elf_Auxinfo);
2334 sbuf_bcat(sb, &structsize, sizeof(structsize));
2336 proc_getauxv(curthread, p, sb);
2342 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2343 int32_t *osrel, const Elf_Phdr *pnote)
2345 const Elf_Note *note, *note0, *note_end;
2346 const char *note_name;
2351 /* We need some limit, might as well use PAGE_SIZE. */
2352 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2354 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2355 if (pnote->p_offset > PAGE_SIZE ||
2356 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2357 VOP_UNLOCK(imgp->vp, 0);
2358 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2359 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2360 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2361 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2362 curthread->td_ucred, NOCRED, NULL, curthread);
2364 uprintf("i/o error PT_NOTE\n");
2368 note = note0 = (const Elf_Note *)buf;
2369 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2371 note = note0 = (const Elf_Note *)(imgp->image_header +
2373 note_end = (const Elf_Note *)(imgp->image_header +
2374 pnote->p_offset + pnote->p_filesz);
2377 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2378 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2379 (const char *)note < sizeof(Elf_Note)) {
2383 if (note->n_namesz != checknote->hdr.n_namesz ||
2384 note->n_descsz != checknote->hdr.n_descsz ||
2385 note->n_type != checknote->hdr.n_type)
2387 note_name = (const char *)(note + 1);
2388 if (note_name + checknote->hdr.n_namesz >=
2389 (const char *)note_end || strncmp(checknote->vendor,
2390 note_name, checknote->hdr.n_namesz) != 0)
2394 * Fetch the osreldate for binary
2395 * from the ELF OSABI-note if necessary.
2397 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2398 checknote->trans_osrel != NULL) {
2399 res = checknote->trans_osrel(note, osrel);
2405 note = (const Elf_Note *)((const char *)(note + 1) +
2406 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2407 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2416 * Try to find the appropriate ABI-note section for checknote,
2417 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2418 * first page of the image is searched, the same as for headers.
2421 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2424 const Elf_Phdr *phdr;
2425 const Elf_Ehdr *hdr;
2428 hdr = (const Elf_Ehdr *)imgp->image_header;
2429 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2431 for (i = 0; i < hdr->e_phnum; i++) {
2432 if (phdr[i].p_type == PT_NOTE &&
2433 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2441 * Tell kern_execve.c about it, with a little help from the linker.
2443 static struct execsw __elfN(execsw) = {
2444 __CONCAT(exec_, __elfN(imgact)),
2445 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2447 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2450 __elfN(trans_prot)(Elf_Word flags)
2456 prot |= VM_PROT_EXECUTE;
2458 prot |= VM_PROT_WRITE;
2460 prot |= VM_PROT_READ;
2461 #if __ELF_WORD_SIZE == 32
2462 #if defined(__amd64__)
2463 if (i386_read_exec && (flags & PF_R))
2464 prot |= VM_PROT_EXECUTE;
2471 __elfN(untrans_prot)(vm_prot_t prot)
2476 if (prot & VM_PROT_EXECUTE)
2478 if (prot & VM_PROT_READ)
2480 if (prot & VM_PROT_WRITE)