2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 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 #include "opt_capsicum.h"
37 #include <sys/param.h>
38 #include <sys/capsicum.h>
39 #include <sys/compressor.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
50 #include <sys/namei.h>
52 #include <sys/procfs.h>
53 #include <sys/ptrace.h>
54 #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, int32_t *osrel, uint32_t *fctl0);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
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);
94 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
95 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
98 static boolean_t __elfN(check_note)(struct image_params *imgp,
99 Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103 static size_t __elfN(prepare_register_notes)(struct thread *td,
104 struct note_info_list *list, struct thread *target_td);
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if defined(__amd64__)
133 static int __elfN(vdso) = 1;
134 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
135 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
136 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
138 static int __elfN(vdso) = 0;
141 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
142 int i386_read_exec = 0;
143 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
144 "enable execution from readable segments");
147 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
149 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
154 val = __elfN(pie_base);
155 error = sysctl_handle_long(oidp, &val, 0, req);
156 if (error != 0 || req->newptr == NULL)
158 if ((val & PAGE_MASK) != 0)
160 __elfN(pie_base) = val;
163 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
164 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
165 sysctl_pie_base, "LU",
166 "PIE load base without randomization");
168 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
174 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures
175 * have limited address space (which can cause issues for applications with
176 * high memory use) so we leave it off there.
178 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
179 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
180 &__elfN(aslr_enabled), 0,
181 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
182 ": enable address map randomization");
185 * Enable ASLR by default for 64-bit PIE binaries.
187 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
188 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
189 &__elfN(pie_aslr_enabled), 0,
190 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
191 ": enable address map randomization for PIE binaries");
194 * Sbrk is deprecated and it can be assumed that in most cases it will not be
195 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR
196 * to use the bss grow region.
198 static int __elfN(aslr_honor_sbrk) = 0;
199 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
200 &__elfN(aslr_honor_sbrk), 0,
201 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
203 static int __elfN(aslr_stack) = 1;
204 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
205 &__elfN(aslr_stack), 0,
206 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
207 ": enable stack address randomization");
209 static int __elfN(sigfastblock) = 1;
210 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
211 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
212 "enable sigfastblock for new processes");
214 static bool __elfN(allow_wx) = true;
215 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
216 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
217 "Allow pages to be mapped simultaneously writable and executable");
219 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
221 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
223 Elf_Brandnote __elfN(freebsd_brandnote) = {
224 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
225 .hdr.n_descsz = sizeof(int32_t),
226 .hdr.n_type = NT_FREEBSD_ABI_TAG,
227 .vendor = FREEBSD_ABI_VENDOR,
228 .flags = BN_TRANSLATE_OSREL,
229 .trans_osrel = __elfN(freebsd_trans_osrel)
233 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
237 p = (uintptr_t)(note + 1);
238 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
239 *osrel = *(const int32_t *)(p);
244 static const char GNU_ABI_VENDOR[] = "GNU";
245 static int GNU_KFREEBSD_ABI_DESC = 3;
247 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
248 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
249 .hdr.n_descsz = 16, /* XXX at least 16 */
251 .vendor = GNU_ABI_VENDOR,
252 .flags = BN_TRANSLATE_OSREL,
253 .trans_osrel = kfreebsd_trans_osrel
257 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
259 const Elf32_Word *desc;
262 p = (uintptr_t)(note + 1);
263 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
265 desc = (const Elf32_Word *)p;
266 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
270 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
271 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
273 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
279 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
283 for (i = 0; i < MAX_BRANDS; i++) {
284 if (elf_brand_list[i] == NULL) {
285 elf_brand_list[i] = entry;
289 if (i == MAX_BRANDS) {
290 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
298 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
302 for (i = 0; i < MAX_BRANDS; i++) {
303 if (elf_brand_list[i] == entry) {
304 elf_brand_list[i] = NULL;
314 __elfN(brand_inuse)(Elf_Brandinfo *entry)
319 sx_slock(&allproc_lock);
320 FOREACH_PROC_IN_SYSTEM(p) {
321 if (p->p_sysent == entry->sysvec) {
326 sx_sunlock(&allproc_lock);
331 static Elf_Brandinfo *
332 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
333 int32_t *osrel, uint32_t *fctl0)
335 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
336 Elf_Brandinfo *bi, *bi_m;
337 boolean_t ret, has_fctl0;
338 int i, interp_name_len;
340 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
343 * We support four types of branding -- (1) the ELF EI_OSABI field
344 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
345 * branding w/in the ELF header, (3) path of the `interp_path'
346 * field, and (4) the ".note.ABI-tag" ELF section.
349 /* Look for an ".note.ABI-tag" ELF section */
351 for (i = 0; i < MAX_BRANDS; i++) {
352 bi = elf_brand_list[i];
355 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
357 if (hdr->e_machine == bi->machine && (bi->flags &
358 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
362 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
364 /* Give brand a chance to veto check_note's guess */
365 if (ret && bi->header_supported) {
366 ret = bi->header_supported(imgp, osrel,
367 has_fctl0 ? fctl0 : NULL);
370 * If note checker claimed the binary, but the
371 * interpreter path in the image does not
372 * match default one for the brand, try to
373 * search for other brands with the same
374 * interpreter. Either there is better brand
375 * with the right interpreter, or, failing
376 * this, we return first brand which accepted
377 * our note and, optionally, header.
379 if (ret && bi_m == NULL && interp != NULL &&
380 (bi->interp_path == NULL ||
381 (strlen(bi->interp_path) + 1 != interp_name_len ||
382 strncmp(interp, bi->interp_path, interp_name_len)
394 /* If the executable has a brand, search for it in the brand list. */
395 for (i = 0; i < MAX_BRANDS; i++) {
396 bi = elf_brand_list[i];
397 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
398 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
400 if (hdr->e_machine == bi->machine &&
401 (hdr->e_ident[EI_OSABI] == bi->brand ||
402 (bi->compat_3_brand != NULL &&
403 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
404 bi->compat_3_brand) == 0))) {
405 /* Looks good, but give brand a chance to veto */
406 if (bi->header_supported == NULL ||
407 bi->header_supported(imgp, NULL, NULL)) {
409 * Again, prefer strictly matching
412 if (interp_name_len == 0 &&
413 bi->interp_path == NULL)
415 if (bi->interp_path != NULL &&
416 strlen(bi->interp_path) + 1 ==
417 interp_name_len && strncmp(interp,
418 bi->interp_path, interp_name_len) == 0)
428 /* No known brand, see if the header is recognized by any brand */
429 for (i = 0; i < MAX_BRANDS; i++) {
430 bi = elf_brand_list[i];
431 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
432 bi->header_supported == NULL)
434 if (hdr->e_machine == bi->machine) {
435 ret = bi->header_supported(imgp, NULL, NULL);
441 /* Lacking a known brand, search for a recognized interpreter. */
442 if (interp != NULL) {
443 for (i = 0; i < MAX_BRANDS; i++) {
444 bi = elf_brand_list[i];
445 if (bi == NULL || (bi->flags &
446 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
449 if (hdr->e_machine == bi->machine &&
450 bi->interp_path != NULL &&
451 /* ELF image p_filesz includes terminating zero */
452 strlen(bi->interp_path) + 1 == interp_name_len &&
453 strncmp(interp, bi->interp_path, interp_name_len)
454 == 0 && (bi->header_supported == NULL ||
455 bi->header_supported(imgp, NULL, NULL)))
460 /* Lacking a recognized interpreter, try the default brand */
461 for (i = 0; i < MAX_BRANDS; i++) {
462 bi = elf_brand_list[i];
463 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
464 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
466 if (hdr->e_machine == bi->machine &&
467 __elfN(fallback_brand) == bi->brand &&
468 (bi->header_supported == NULL ||
469 bi->header_supported(imgp, NULL, NULL)))
476 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
478 return (hdr->e_phoff <= PAGE_SIZE &&
479 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
483 __elfN(check_header)(const Elf_Ehdr *hdr)
489 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
490 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
491 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
492 hdr->e_phentsize != sizeof(Elf_Phdr) ||
493 hdr->e_version != ELF_TARG_VER)
497 * Make sure we have at least one brand for this machine.
500 for (i = 0; i < MAX_BRANDS; i++) {
501 bi = elf_brand_list[i];
502 if (bi != NULL && bi->machine == hdr->e_machine)
512 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
513 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
520 * Create the page if it doesn't exist yet. Ignore errors.
522 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
523 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
526 * Find the page from the underlying object.
528 if (object != NULL) {
529 sf = vm_imgact_map_page(object, offset);
531 return (KERN_FAILURE);
532 off = offset - trunc_page(offset);
533 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
535 vm_imgact_unmap_page(sf);
537 return (KERN_FAILURE);
540 return (KERN_SUCCESS);
544 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
545 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
551 int error, locked, rv;
553 if (start != trunc_page(start)) {
554 rv = __elfN(map_partial)(map, object, offset, start,
555 round_page(start), prot);
556 if (rv != KERN_SUCCESS)
558 offset += round_page(start) - start;
559 start = round_page(start);
561 if (end != round_page(end)) {
562 rv = __elfN(map_partial)(map, object, offset +
563 trunc_page(end) - start, trunc_page(end), end, prot);
564 if (rv != KERN_SUCCESS)
566 end = trunc_page(end);
569 return (KERN_SUCCESS);
570 if ((offset & PAGE_MASK) != 0) {
572 * The mapping is not page aligned. This means that we have
575 rv = vm_map_fixed(map, NULL, 0, start, end - start,
576 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
577 if (rv != KERN_SUCCESS)
580 return (KERN_SUCCESS);
581 for (; start < end; start += sz) {
582 sf = vm_imgact_map_page(object, offset);
584 return (KERN_FAILURE);
585 off = offset - trunc_page(offset);
587 if (sz > PAGE_SIZE - off)
588 sz = PAGE_SIZE - off;
589 error = copyout((caddr_t)sf_buf_kva(sf) + off,
591 vm_imgact_unmap_page(sf);
593 return (KERN_FAILURE);
597 vm_object_reference(object);
598 rv = vm_map_fixed(map, object, offset, start, end - start,
599 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
600 (object != NULL ? MAP_VN_EXEC : 0));
601 if (rv != KERN_SUCCESS) {
602 locked = VOP_ISLOCKED(imgp->vp);
603 VOP_UNLOCK(imgp->vp);
604 vm_object_deallocate(object);
605 vn_lock(imgp->vp, locked | LK_RETRY);
607 } else if (object != NULL) {
608 MPASS(imgp->vp->v_object == object);
609 VOP_SET_TEXT_CHECKED(imgp->vp);
612 return (KERN_SUCCESS);
616 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
617 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
623 vm_offset_t map_addr;
626 vm_ooffset_t file_addr;
629 * It's necessary to fail if the filsz + offset taken from the
630 * header is greater than the actual file pager object's size.
631 * If we were to allow this, then the vm_map_find() below would
632 * walk right off the end of the file object and into the ether.
634 * While I'm here, might as well check for something else that
635 * is invalid: filsz cannot be greater than memsz.
637 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
639 uprintf("elf_load_section: truncated ELF file\n");
643 object = imgp->object;
644 map = &imgp->proc->p_vmspace->vm_map;
645 map_addr = trunc_page((vm_offset_t)vmaddr);
646 file_addr = trunc_page(offset);
649 * We have two choices. We can either clear the data in the last page
650 * of an oversized mapping, or we can start the anon mapping a page
651 * early and copy the initialized data into that first page. We
656 else if (memsz > filsz)
657 map_len = trunc_page(offset + filsz) - file_addr;
659 map_len = round_page(offset + filsz) - file_addr;
662 /* cow flags: don't dump readonly sections in core */
663 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
664 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
666 rv = __elfN(map_insert)(imgp, map, object, file_addr,
667 map_addr, map_addr + map_len, prot, cow);
668 if (rv != KERN_SUCCESS)
671 /* we can stop now if we've covered it all */
677 * We have to get the remaining bit of the file into the first part
678 * of the oversized map segment. This is normally because the .data
679 * segment in the file is extended to provide bss. It's a neat idea
680 * to try and save a page, but it's a pain in the behind to implement.
682 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
684 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
685 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
687 /* This had damn well better be true! */
689 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
690 map_addr + map_len, prot, 0);
691 if (rv != KERN_SUCCESS)
696 sf = vm_imgact_map_page(object, offset + filsz);
700 /* send the page fragment to user space */
701 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
703 vm_imgact_unmap_page(sf);
709 * Remove write access to the page if it was only granted by map_insert
712 if ((prot & VM_PROT_WRITE) == 0)
713 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
714 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
720 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
721 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
728 ASSERT_VOP_LOCKED(imgp->vp, __func__);
733 for (i = 0; i < hdr->e_phnum; i++) {
734 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
737 /* Loadable segment */
738 prot = __elfN(trans_prot)(phdr[i].p_flags);
739 error = __elfN(load_section)(imgp, phdr[i].p_offset,
740 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
741 phdr[i].p_memsz, phdr[i].p_filesz, prot);
746 * Establish the base address if this is the first segment.
749 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
754 if (base_addrp != NULL)
755 *base_addrp = base_addr;
761 * Load the file "file" into memory. It may be either a shared object
764 * The "addr" reference parameter is in/out. On entry, it specifies
765 * the address where a shared object should be loaded. If the file is
766 * an executable, this value is ignored. On exit, "addr" specifies
767 * where the file was actually loaded.
769 * The "entry" reference parameter is out only. On exit, it specifies
770 * the entry point for the loaded file.
773 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
779 struct image_params image_params;
781 const Elf_Ehdr *hdr = NULL;
782 const Elf_Phdr *phdr = NULL;
783 struct nameidata *nd;
785 struct image_params *imgp;
787 u_long base_addr = 0;
790 #ifdef CAPABILITY_MODE
792 * XXXJA: This check can go away once we are sufficiently confident
793 * that the checks in namei() are correct.
795 if (IN_CAPABILITY_MODE(curthread))
799 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
801 attr = &tempdata->attr;
802 imgp = &tempdata->image_params;
805 * Initialize part of the common data
810 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
811 UIO_SYSSPACE, file, curthread);
812 if ((error = namei(nd)) != 0) {
816 NDFREE(nd, NDF_ONLY_PNBUF);
817 imgp->vp = nd->ni_vp;
820 * Check permissions, modes, uid, etc on the file, and "open" it.
822 error = exec_check_permissions(imgp);
826 error = exec_map_first_page(imgp);
830 imgp->object = nd->ni_vp->v_object;
832 hdr = (const Elf_Ehdr *)imgp->image_header;
833 if ((error = __elfN(check_header)(hdr)) != 0)
835 if (hdr->e_type == ET_DYN)
837 else if (hdr->e_type == ET_EXEC)
844 /* Only support headers that fit within first page for now */
845 if (!__elfN(phdr_in_zero_page)(hdr)) {
850 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
851 if (!aligned(phdr, Elf_Addr)) {
856 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
861 *entry = (unsigned long)hdr->e_entry + rbase;
865 exec_unmap_first_page(imgp);
869 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
872 free(tempdata, M_TEMP);
878 * Select randomized valid address in the map map, between minv and
879 * maxv, with specified alignment. The [minv, maxv) range must belong
880 * to the map. Note that function only allocates the address, it is
881 * up to caller to clamp maxv in a way that the final allocation
882 * length fit into the map.
884 * Result is returned in *resp, error code indicates that arguments
885 * did not pass sanity checks for overflow and range correctness.
888 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
889 u_int align, u_long *resp)
893 MPASS(vm_map_min(map) <= minv);
895 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
896 uprintf("Invalid ELF segments layout\n");
900 arc4rand(&rbase, sizeof(rbase), 0);
901 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
902 res &= ~((u_long)align - 1);
907 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
908 res, minv, maxv, rbase));
910 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
911 res, maxv, minv, rbase));
918 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
919 const Elf_Phdr *phdr, u_long et_dyn_addr)
921 struct vmspace *vmspace;
923 u_long text_size, data_size, total_size, text_addr, data_addr;
924 u_long seg_size, seg_addr;
928 text_size = data_size = total_size = text_addr = data_addr = 0;
930 for (i = 0; i < hdr->e_phnum; i++) {
931 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
934 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
935 seg_size = round_page(phdr[i].p_memsz +
936 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
939 * Make the largest executable segment the official
940 * text segment and all others data.
942 * Note that obreak() assumes that data_addr + data_size == end
943 * of data load area, and the ELF file format expects segments
944 * to be sorted by address. If multiple data segments exist,
945 * the last one will be used.
948 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
949 text_size = seg_size;
950 text_addr = seg_addr;
952 data_size = seg_size;
953 data_addr = seg_addr;
955 total_size += seg_size;
958 if (data_addr == 0 && data_size == 0) {
959 data_addr = text_addr;
960 data_size = text_size;
964 * Check limits. It should be safe to check the
965 * limits after loading the segments since we do
966 * not actually fault in all the segments pages.
968 PROC_LOCK(imgp->proc);
969 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
970 err_str = "Data segment size exceeds process limit";
971 else if (text_size > maxtsiz)
972 err_str = "Text segment size exceeds system limit";
973 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
974 err_str = "Total segment size exceeds process limit";
975 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
976 err_str = "Data segment size exceeds resource limit";
977 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
978 err_str = "Total segment size exceeds resource limit";
979 PROC_UNLOCK(imgp->proc);
980 if (err_str != NULL) {
981 uprintf("%s\n", err_str);
985 vmspace = imgp->proc->p_vmspace;
986 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
987 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
988 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
989 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
995 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
996 char **interpp, bool *free_interpp)
1000 int error, interp_name_len;
1002 KASSERT(phdr->p_type == PT_INTERP,
1003 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1004 ASSERT_VOP_LOCKED(imgp->vp, __func__);
1008 /* Path to interpreter */
1009 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1010 uprintf("Invalid PT_INTERP\n");
1014 interp_name_len = phdr->p_filesz;
1015 if (phdr->p_offset > PAGE_SIZE ||
1016 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1018 * The vnode lock might be needed by the pagedaemon to
1019 * clean pages owned by the vnode. Do not allow sleep
1020 * waiting for memory with the vnode locked, instead
1021 * try non-sleepable allocation first, and if it
1022 * fails, go to the slow path were we drop the lock
1023 * and do M_WAITOK. A text reference prevents
1024 * modifications to the vnode content.
1026 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1027 if (interp == NULL) {
1028 VOP_UNLOCK(imgp->vp);
1029 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1030 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1033 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1034 interp_name_len, phdr->p_offset,
1035 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1038 free(interp, M_TEMP);
1039 uprintf("i/o error PT_INTERP %d\n", error);
1042 interp[interp_name_len] = '\0';
1045 *free_interpp = true;
1049 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1050 if (interp[interp_name_len - 1] != '\0') {
1051 uprintf("Invalid PT_INTERP\n");
1056 *free_interpp = false;
1061 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1062 const char *interp, u_long *addr, u_long *entry)
1067 if (brand_info->emul_path != NULL &&
1068 brand_info->emul_path[0] != '\0') {
1069 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1070 snprintf(path, MAXPATHLEN, "%s%s",
1071 brand_info->emul_path, interp);
1072 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1078 if (brand_info->interp_newpath != NULL &&
1079 (brand_info->interp_path == NULL ||
1080 strcmp(interp, brand_info->interp_path) == 0)) {
1081 error = __elfN(load_file)(imgp->proc,
1082 brand_info->interp_newpath, addr, entry);
1087 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1091 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1096 * Impossible et_dyn_addr initial value indicating that the real base
1097 * must be calculated later with some randomization applied.
1099 #define ET_DYN_ADDR_RAND 1
1102 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1105 const Elf_Ehdr *hdr;
1106 const Elf_Phdr *phdr;
1107 Elf_Auxargs *elf_auxargs;
1108 struct vmspace *vmspace;
1111 Elf_Brandinfo *brand_info;
1112 struct sysentvec *sv;
1113 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1114 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1120 hdr = (const Elf_Ehdr *)imgp->image_header;
1123 * Do we have a valid ELF header ?
1125 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1126 * if particular brand doesn't support it.
1128 if (__elfN(check_header)(hdr) != 0 ||
1129 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1133 * From here on down, we return an errno, not -1, as we've
1134 * detected an ELF file.
1137 if (!__elfN(phdr_in_zero_page)(hdr)) {
1138 uprintf("Program headers not in the first page\n");
1141 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1142 if (!aligned(phdr, Elf_Addr)) {
1143 uprintf("Unaligned program headers\n");
1151 entry = proghdr = 0;
1153 free_interp = false;
1157 * Somewhat arbitrary, limit accepted max alignment for the
1158 * loadable segment to the max supported superpage size. Too
1159 * large alignment requests are not useful and are indicators
1160 * of corrupted or outright malicious binary.
1162 maxalign = PAGE_SIZE;
1163 maxsalign = PAGE_SIZE * 1024;
1164 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1165 if (pagesizes[i] > maxsalign)
1166 maxsalign = pagesizes[i];
1171 for (i = 0; i < hdr->e_phnum; i++) {
1172 switch (phdr[i].p_type) {
1175 baddr = phdr[i].p_vaddr;
1176 if (!powerof2(phdr[i].p_align) ||
1177 phdr[i].p_align > maxsalign) {
1178 uprintf("Invalid segment alignment\n");
1182 if (phdr[i].p_align > maxalign)
1183 maxalign = phdr[i].p_align;
1184 if (mapsz + phdr[i].p_memsz < mapsz) {
1185 uprintf("Mapsize overflow\n");
1189 mapsz += phdr[i].p_memsz;
1193 * If this segment contains the program headers,
1194 * remember their virtual address for the AT_PHDR
1195 * aux entry. Static binaries don't usually include
1198 if (phdr[i].p_offset == 0 &&
1199 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1201 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1204 /* Path to interpreter */
1205 if (interp != NULL) {
1206 uprintf("Multiple PT_INTERP headers\n");
1210 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1216 if (__elfN(nxstack)) {
1218 __elfN(trans_prot)(phdr[i].p_flags);
1219 if ((imgp->stack_prot & VM_PROT_RW) !=
1221 uprintf("Invalid PT_GNU_STACK\n");
1226 imgp->stack_sz = phdr[i].p_memsz;
1228 case PT_PHDR: /* Program header table info */
1229 proghdr = phdr[i].p_vaddr;
1234 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1235 if (brand_info == NULL) {
1236 uprintf("ELF binary type \"%u\" not known.\n",
1237 hdr->e_ident[EI_OSABI]);
1241 sv = brand_info->sysvec;
1243 if (hdr->e_type == ET_DYN) {
1244 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1245 uprintf("Cannot execute shared object\n");
1250 * Honour the base load address from the dso if it is
1251 * non-zero for some reason.
1254 if ((sv->sv_flags & SV_ASLR) == 0 ||
1255 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1256 et_dyn_addr = __elfN(pie_base);
1257 else if ((__elfN(pie_aslr_enabled) &&
1258 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1259 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1260 et_dyn_addr = ET_DYN_ADDR_RAND;
1262 et_dyn_addr = __elfN(pie_base);
1267 * Avoid a possible deadlock if the current address space is destroyed
1268 * and that address space maps the locked vnode. In the common case,
1269 * the locked vnode's v_usecount is decremented but remains greater
1270 * than zero. Consequently, the vnode lock is not needed by vrele().
1271 * However, in cases where the vnode lock is external, such as nullfs,
1272 * v_usecount may become zero.
1274 * The VV_TEXT flag prevents modifications to the executable while
1275 * the vnode is unlocked.
1277 VOP_UNLOCK(imgp->vp);
1280 * Decide whether to enable randomization of user mappings.
1281 * First, reset user preferences for the setid binaries.
1282 * Then, account for the support of the randomization by the
1283 * ABI, by user preferences, and make special treatment for
1286 if (imgp->credential_setid) {
1287 PROC_LOCK(imgp->proc);
1288 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1289 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1290 PROC_UNLOCK(imgp->proc);
1292 if ((sv->sv_flags & SV_ASLR) == 0 ||
1293 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1294 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1295 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1296 ("et_dyn_addr == RAND and !ASLR"));
1297 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1298 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1299 et_dyn_addr == ET_DYN_ADDR_RAND) {
1300 imgp->map_flags |= MAP_ASLR;
1302 * If user does not care about sbrk, utilize the bss
1303 * grow region for mappings as well. We can select
1304 * the base for the image anywere and still not suffer
1305 * from the fragmentation.
1307 if (!__elfN(aslr_honor_sbrk) ||
1308 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1309 imgp->map_flags |= MAP_ASLR_IGNSTART;
1310 if (__elfN(aslr_stack))
1311 imgp->map_flags |= MAP_ASLR_STACK;
1314 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1315 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1316 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1317 imgp->map_flags |= MAP_WXORX;
1319 error = exec_new_vmspace(imgp, sv);
1321 imgp->proc->p_sysent = sv;
1322 imgp->proc->p_elf_brandinfo = brand_info;
1324 vmspace = imgp->proc->p_vmspace;
1325 map = &vmspace->vm_map;
1326 maxv = sv->sv_usrstack;
1327 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1328 maxv -= lim_max(td, RLIMIT_STACK);
1329 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1330 uprintf("Excessive mapping size\n");
1334 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1335 KASSERT((map->flags & MAP_ASLR) != 0,
1336 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1337 error = __CONCAT(rnd_, __elfN(base))(map,
1338 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1339 /* reserve half of the address space to interpreter */
1340 maxv / 2, maxalign, &et_dyn_addr);
1343 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1347 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1351 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1356 * We load the dynamic linker where a userland call
1357 * to mmap(0, ...) would put it. The rationale behind this
1358 * calculation is that it leaves room for the heap to grow to
1359 * its maximum allowed size.
1361 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1363 if ((map->flags & MAP_ASLR) != 0) {
1364 maxv1 = maxv / 2 + addr / 2;
1365 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1366 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1367 pagesizes[1] : pagesizes[0], &anon_loc);
1370 map->anon_loc = anon_loc;
1372 map->anon_loc = addr;
1375 entry = (u_long)hdr->e_entry + et_dyn_addr;
1376 imgp->entry_addr = entry;
1378 if (interp != NULL) {
1379 VOP_UNLOCK(imgp->vp);
1380 if ((map->flags & MAP_ASLR) != 0) {
1381 /* Assume that interpreter fits into 1/4 of AS */
1382 maxv1 = maxv / 2 + addr / 2;
1383 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1384 maxv1, PAGE_SIZE, &addr);
1387 error = __elfN(load_interp)(imgp, brand_info, interp,
1388 &addr, &imgp->entry_addr);
1390 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1396 error = exec_map_stack(imgp);
1401 * Construct auxargs table (used by the copyout_auxargs routine)
1403 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1404 if (elf_auxargs == NULL) {
1405 VOP_UNLOCK(imgp->vp);
1406 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1407 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1409 elf_auxargs->execfd = -1;
1410 elf_auxargs->phdr = proghdr + et_dyn_addr;
1411 elf_auxargs->phent = hdr->e_phentsize;
1412 elf_auxargs->phnum = hdr->e_phnum;
1413 elf_auxargs->pagesz = PAGE_SIZE;
1414 elf_auxargs->base = addr;
1415 elf_auxargs->flags = 0;
1416 elf_auxargs->entry = entry;
1417 elf_auxargs->hdr_eflags = hdr->e_flags;
1419 imgp->auxargs = elf_auxargs;
1420 imgp->interpreted = 0;
1421 imgp->reloc_base = addr;
1422 imgp->proc->p_osrel = osrel;
1423 imgp->proc->p_fctl0 = fctl0;
1424 imgp->proc->p_elf_flags = hdr->e_flags;
1427 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1429 free(interp, M_TEMP);
1433 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1436 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1438 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1439 Elf_Auxinfo *argarray, *pos;
1440 struct vmspace *vmspace;
1442 int error, bsdflags, oc;
1444 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1447 vmspace = imgp->proc->p_vmspace;
1449 if (args->execfd != -1)
1450 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1451 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1452 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1453 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1454 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1455 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1456 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1457 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1458 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1459 if (imgp->execpathp != 0)
1460 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1461 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1462 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1463 if (imgp->canary != 0) {
1464 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1465 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1467 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1468 if (imgp->pagesizes != 0) {
1469 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1470 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1472 if (imgp->sysent->sv_timekeep_base != 0) {
1473 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1474 imgp->sysent->sv_timekeep_base);
1476 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1477 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1478 imgp->sysent->sv_stackprot);
1479 if (imgp->sysent->sv_hwcap != NULL)
1480 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1481 if (imgp->sysent->sv_hwcap2 != NULL)
1482 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1484 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0;
1485 oc = atomic_load_int(&vm_overcommit);
1486 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) !=
1487 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0;
1488 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags);
1489 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1490 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1491 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1492 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1493 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1494 if (imgp->sysent->sv_fxrng_gen_base != 0)
1495 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1496 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1497 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1498 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop));
1499 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur;
1500 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz);
1501 AUXARGS_ENTRY(pos, AT_NULL, 0);
1503 free(imgp->auxargs, M_TEMP);
1504 imgp->auxargs = NULL;
1505 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1507 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1508 free(argarray, M_TEMP);
1513 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1517 base = (Elf_Addr *)*stack_base;
1519 if (elf_suword(base, imgp->args->argc) == -1)
1521 *stack_base = (uintptr_t)base;
1526 * Code for generating ELF core dumps.
1529 typedef void (*segment_callback)(vm_map_entry_t, void *);
1531 /* Closure for cb_put_phdr(). */
1532 struct phdr_closure {
1533 Elf_Phdr *phdr; /* Program header to fill in */
1534 Elf_Off offset; /* Offset of segment in core file */
1538 int type; /* Note type. */
1539 struct regset *regset; /* Register set. */
1540 outfunc_t outfunc; /* Output function. */
1541 void *outarg; /* Argument for the output function. */
1542 size_t outsize; /* Output size. */
1543 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1546 TAILQ_HEAD(note_info_list, note_info);
1548 extern int compress_user_cores;
1549 extern int compress_user_cores_level;
1551 static void cb_put_phdr(vm_map_entry_t, void *);
1552 static void cb_size_segment(vm_map_entry_t, void *);
1553 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1555 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1556 struct note_info_list *, size_t, int);
1557 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1559 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1560 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1561 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1562 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1563 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1564 static void note_procstat_files(void *, struct sbuf *, size_t *);
1565 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1566 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1567 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1568 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1569 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1572 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1575 return (core_write((struct coredump_params *)arg, base, len, offset,
1576 UIO_SYSSPACE, NULL));
1580 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1582 struct ucred *cred = td->td_ucred;
1583 int compm, error = 0;
1584 struct sseg_closure seginfo;
1585 struct note_info_list notelst;
1586 struct coredump_params params;
1587 struct note_info *ninfo;
1589 size_t hdrsize, notesz, coresize;
1593 TAILQ_INIT(¬elst);
1595 /* Size the program segments. */
1596 __elfN(size_segments)(td, &seginfo, flags);
1599 * Collect info about the core file header area.
1601 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1602 if (seginfo.count + 1 >= PN_XNUM)
1603 hdrsize += sizeof(Elf_Shdr);
1604 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1605 coresize = round_page(hdrsize + notesz) + seginfo.size;
1607 /* Set up core dump parameters. */
1609 params.active_cred = cred;
1610 params.file_cred = NOCRED;
1617 PROC_LOCK(td->td_proc);
1618 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1619 PROC_UNLOCK(td->td_proc);
1626 if (coresize >= limit) {
1631 /* Create a compression stream if necessary. */
1632 compm = compress_user_cores;
1633 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1635 compm = COMPRESS_GZIP;
1637 params.comp = compressor_init(core_compressed_write,
1638 compm, CORE_BUF_SIZE,
1639 compress_user_cores_level, ¶ms);
1640 if (params.comp == NULL) {
1644 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1648 * Allocate memory for building the header, fill it up,
1649 * and write it out following the notes.
1651 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1652 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1655 /* Write the contents of all of the writable segments. */
1661 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1662 offset = round_page(hdrsize + notesz);
1663 for (i = 0; i < seginfo.count; i++) {
1664 error = core_output((char *)(uintptr_t)php->p_vaddr,
1665 php->p_filesz, offset, ¶ms, tmpbuf);
1668 offset += php->p_filesz;
1671 if (error == 0 && params.comp != NULL)
1672 error = compressor_flush(params.comp);
1676 "Failed to write core file for process %s (error %d)\n",
1677 curproc->p_comm, error);
1681 free(tmpbuf, M_TEMP);
1682 if (params.comp != NULL)
1683 compressor_fini(params.comp);
1684 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1685 TAILQ_REMOVE(¬elst, ninfo, link);
1686 free(ninfo, M_TEMP);
1695 * A callback for each_dumpable_segment() to write out the segment's
1696 * program header entry.
1699 cb_put_phdr(vm_map_entry_t entry, void *closure)
1701 struct phdr_closure *phc = (struct phdr_closure *)closure;
1702 Elf_Phdr *phdr = phc->phdr;
1704 phc->offset = round_page(phc->offset);
1706 phdr->p_type = PT_LOAD;
1707 phdr->p_offset = phc->offset;
1708 phdr->p_vaddr = entry->start;
1710 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1711 phdr->p_align = PAGE_SIZE;
1712 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1714 phc->offset += phdr->p_filesz;
1719 * A callback for each_dumpable_segment() to gather information about
1720 * the number of segments and their total size.
1723 cb_size_segment(vm_map_entry_t entry, void *closure)
1725 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1728 ssc->size += entry->end - entry->start;
1732 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1738 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1742 * For each writable segment in the process's memory map, call the given
1743 * function with a pointer to the map entry and some arbitrary
1744 * caller-supplied data.
1747 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1750 struct proc *p = td->td_proc;
1751 vm_map_t map = &p->p_vmspace->vm_map;
1752 vm_map_entry_t entry;
1753 vm_object_t backing_object, object;
1756 vm_map_lock_read(map);
1757 VM_MAP_ENTRY_FOREACH(entry, map) {
1759 * Don't dump inaccessible mappings, deal with legacy
1762 * Note that read-only segments related to the elf binary
1763 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1764 * need to arbitrarily ignore such segments.
1766 if ((flags & SVC_ALL) == 0) {
1767 if (elf_legacy_coredump) {
1768 if ((entry->protection & VM_PROT_RW) !=
1772 if ((entry->protection & VM_PROT_ALL) == 0)
1778 * Dont include memory segment in the coredump if
1779 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1780 * madvise(2). Do not dump submaps (i.e. parts of the
1783 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1785 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1786 (flags & SVC_ALL) == 0)
1788 if ((object = entry->object.vm_object) == NULL)
1791 /* Ignore memory-mapped devices and such things. */
1792 VM_OBJECT_RLOCK(object);
1793 while ((backing_object = object->backing_object) != NULL) {
1794 VM_OBJECT_RLOCK(backing_object);
1795 VM_OBJECT_RUNLOCK(object);
1796 object = backing_object;
1798 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1799 VM_OBJECT_RUNLOCK(object);
1803 (*func)(entry, closure);
1805 vm_map_unlock_read(map);
1809 * Write the core file header to the file, including padding up to
1810 * the page boundary.
1813 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1814 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1817 struct note_info *ninfo;
1821 /* Fill in the header. */
1822 bzero(hdr, hdrsize);
1823 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1825 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1826 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1827 sbuf_start_section(sb, NULL);
1828 sbuf_bcat(sb, hdr, hdrsize);
1829 TAILQ_FOREACH(ninfo, notelst, link)
1830 __elfN(putnote)(p->td, ninfo, sb);
1831 /* Align up to a page boundary for the program segments. */
1832 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1833 error = sbuf_finish(sb);
1840 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1850 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1851 __elfN(note_prpsinfo), p);
1854 * To have the debugger select the right thread (LWP) as the initial
1855 * thread, we dump the state of the thread passed to us in td first.
1856 * This is the thread that causes the core dump and thus likely to
1857 * be the right thread one wants to have selected in the debugger.
1860 while (thr != NULL) {
1861 size += __elfN(prepare_register_notes)(td, list, thr);
1862 size += __elfN(register_note)(td, list, -1,
1863 __elfN(note_threadmd), thr);
1865 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1866 TAILQ_NEXT(thr, td_plist);
1868 thr = TAILQ_NEXT(thr, td_plist);
1871 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1872 __elfN(note_procstat_proc), p);
1873 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1874 note_procstat_files, p);
1875 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1876 note_procstat_vmmap, p);
1877 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1878 note_procstat_groups, p);
1879 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1880 note_procstat_umask, p);
1881 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1882 note_procstat_rlimit, p);
1883 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1884 note_procstat_osrel, p);
1885 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1886 __elfN(note_procstat_psstrings), p);
1887 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1888 __elfN(note_procstat_auxv), p);
1894 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1895 size_t notesz, int flags)
1900 struct phdr_closure phc;
1903 ehdr = (Elf_Ehdr *)hdr;
1904 bi = td->td_proc->p_elf_brandinfo;
1906 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1907 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1908 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1909 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1910 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1911 ehdr->e_ident[EI_DATA] = ELF_DATA;
1912 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1913 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1914 ehdr->e_ident[EI_ABIVERSION] = 0;
1915 ehdr->e_ident[EI_PAD] = 0;
1916 ehdr->e_type = ET_CORE;
1917 ehdr->e_machine = bi->machine;
1918 ehdr->e_version = EV_CURRENT;
1920 ehdr->e_phoff = sizeof(Elf_Ehdr);
1921 ehdr->e_flags = td->td_proc->p_elf_flags;
1922 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1923 ehdr->e_phentsize = sizeof(Elf_Phdr);
1924 ehdr->e_shentsize = sizeof(Elf_Shdr);
1925 ehdr->e_shstrndx = SHN_UNDEF;
1926 if (numsegs + 1 < PN_XNUM) {
1927 ehdr->e_phnum = numsegs + 1;
1930 ehdr->e_phnum = PN_XNUM;
1933 ehdr->e_shoff = ehdr->e_phoff +
1934 (numsegs + 1) * ehdr->e_phentsize;
1935 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1936 ("e_shoff: %zu, hdrsize - shdr: %zu",
1937 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1939 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1940 memset(shdr, 0, sizeof(*shdr));
1942 * A special first section is used to hold large segment and
1943 * section counts. This was proposed by Sun Microsystems in
1944 * Solaris and has been adopted by Linux; the standard ELF
1945 * tools are already familiar with the technique.
1947 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1948 * (or 12-7 depending on the version of the document) for more
1951 shdr->sh_type = SHT_NULL;
1952 shdr->sh_size = ehdr->e_shnum;
1953 shdr->sh_link = ehdr->e_shstrndx;
1954 shdr->sh_info = numsegs + 1;
1958 * Fill in the program header entries.
1960 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1962 /* The note segement. */
1963 phdr->p_type = PT_NOTE;
1964 phdr->p_offset = hdrsize;
1967 phdr->p_filesz = notesz;
1969 phdr->p_flags = PF_R;
1970 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1973 /* All the writable segments from the program. */
1975 phc.offset = round_page(hdrsize + notesz);
1976 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1980 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1981 struct regset *regset, struct thread *target_td)
1983 const struct sysentvec *sv;
1984 struct note_info *ninfo;
1985 size_t size, notesize;
1988 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1991 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1992 ninfo->type = regset->note;
1993 ninfo->regset = regset;
1994 ninfo->outarg = target_td;
1995 ninfo->outsize = size;
1996 TAILQ_INSERT_TAIL(list, ninfo, link);
1998 sv = td->td_proc->p_sysent;
1999 notesize = sizeof(Elf_Note) + /* note header */
2000 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2002 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2008 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2009 int type, outfunc_t out, void *arg)
2011 const struct sysentvec *sv;
2012 struct note_info *ninfo;
2013 size_t size, notesize;
2015 sv = td->td_proc->p_sysent;
2017 out(arg, NULL, &size);
2018 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2020 ninfo->outfunc = out;
2021 ninfo->outarg = arg;
2022 ninfo->outsize = size;
2023 TAILQ_INSERT_TAIL(list, ninfo, link);
2028 notesize = sizeof(Elf_Note) + /* note header */
2029 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2031 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2037 append_note_data(const void *src, void *dst, size_t len)
2041 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2043 bcopy(src, dst, len);
2044 bzero((char *)dst + len, padded_len - len);
2046 return (padded_len);
2050 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2058 note = (Elf_Note *)buf;
2059 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2060 note->n_descsz = size;
2061 note->n_type = type;
2062 buf += sizeof(*note);
2063 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2064 sizeof(FREEBSD_ABI_VENDOR));
2065 append_note_data(src, buf, size);
2070 notesize = sizeof(Elf_Note) + /* note header */
2071 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2073 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2079 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2082 const struct sysentvec *sv;
2083 ssize_t old_len, sect_len;
2084 size_t new_len, descsz, i;
2086 if (ninfo->type == -1) {
2087 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2091 sv = td->td_proc->p_sysent;
2093 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2094 note.n_descsz = ninfo->outsize;
2095 note.n_type = ninfo->type;
2097 sbuf_bcat(sb, ¬e, sizeof(note));
2098 sbuf_start_section(sb, &old_len);
2099 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2100 strlen(sv->sv_elf_core_abi_vendor) + 1);
2101 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2102 if (note.n_descsz == 0)
2104 sbuf_start_section(sb, &old_len);
2105 if (ninfo->regset != NULL) {
2106 struct regset *regset = ninfo->regset;
2109 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2110 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2111 sbuf_bcat(sb, buf, ninfo->outsize);
2114 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2115 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2119 new_len = (size_t)sect_len;
2120 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2121 if (new_len < descsz) {
2123 * It is expected that individual note emitters will correctly
2124 * predict their expected output size and fill up to that size
2125 * themselves, padding in a format-specific way if needed.
2126 * However, in case they don't, just do it here with zeros.
2128 for (i = 0; i < descsz - new_len; i++)
2130 } else if (new_len > descsz) {
2132 * We can't always truncate sb -- we may have drained some
2135 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2136 "read it (%zu > %zu). Since it is longer than "
2137 "expected, this coredump's notes are corrupt. THIS "
2138 "IS A BUG in the note_procstat routine for type %u.\n",
2139 __func__, (unsigned)note.n_type, new_len, descsz,
2140 (unsigned)note.n_type));
2145 * Miscellaneous note out functions.
2148 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2149 #include <compat/freebsd32/freebsd32.h>
2150 #include <compat/freebsd32/freebsd32_signal.h>
2152 typedef struct prstatus32 elf_prstatus_t;
2153 typedef struct prpsinfo32 elf_prpsinfo_t;
2154 typedef struct fpreg32 elf_prfpregset_t;
2155 typedef struct fpreg32 elf_fpregset_t;
2156 typedef struct reg32 elf_gregset_t;
2157 typedef struct thrmisc32 elf_thrmisc_t;
2158 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2159 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2160 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2161 typedef uint32_t elf_ps_strings_t;
2163 typedef prstatus_t elf_prstatus_t;
2164 typedef prpsinfo_t elf_prpsinfo_t;
2165 typedef prfpregset_t elf_prfpregset_t;
2166 typedef prfpregset_t elf_fpregset_t;
2167 typedef gregset_t elf_gregset_t;
2168 typedef thrmisc_t elf_thrmisc_t;
2169 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2170 #define ELF_KERN_PROC_MASK 0
2171 typedef struct kinfo_proc elf_kinfo_proc_t;
2172 typedef vm_offset_t elf_ps_strings_t;
2176 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2182 elf_prpsinfo_t *psinfo;
2187 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2188 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2189 psinfo->pr_version = PRPSINFO_VERSION;
2190 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2191 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2193 if (p->p_args != NULL) {
2194 len = sizeof(psinfo->pr_psargs) - 1;
2195 if (len > p->p_args->ar_length)
2196 len = p->p_args->ar_length;
2197 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2203 sbuf_new(&sbarg, psinfo->pr_psargs,
2204 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2205 error = proc_getargv(curthread, p, &sbarg);
2207 if (sbuf_finish(&sbarg) == 0) {
2208 len = sbuf_len(&sbarg);
2212 len = sizeof(psinfo->pr_psargs) - 1;
2214 sbuf_delete(&sbarg);
2216 if (error != 0 || len == 0 || (ssize_t)len == -1)
2217 strlcpy(psinfo->pr_psargs, p->p_comm,
2218 sizeof(psinfo->pr_psargs));
2220 KASSERT(len < sizeof(psinfo->pr_psargs),
2221 ("len is too long: %zu vs %zu", len,
2222 sizeof(psinfo->pr_psargs)));
2223 cp = psinfo->pr_psargs;
2226 cp = memchr(cp, '\0', end - cp);
2232 psinfo->pr_pid = p->p_pid;
2233 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2234 free(psinfo, M_TEMP);
2236 *sizep = sizeof(*psinfo);
2240 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2243 elf_prstatus_t *status;
2246 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2249 memset(status, 0, *sizep);
2250 status->pr_version = PRSTATUS_VERSION;
2251 status->pr_statussz = sizeof(elf_prstatus_t);
2252 status->pr_gregsetsz = sizeof(elf_gregset_t);
2253 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2254 status->pr_osreldate = osreldate;
2255 status->pr_cursig = td->td_proc->p_sig;
2256 status->pr_pid = td->td_tid;
2257 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2258 fill_regs32(td, &status->pr_reg);
2260 fill_regs(td, &status->pr_reg);
2263 *sizep = sizeof(*status);
2268 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2271 elf_prstatus_t *status;
2273 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2275 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2276 set_regs32(td, &status->pr_reg);
2278 set_regs(td, &status->pr_reg);
2283 static struct regset __elfN(regset_prstatus) = {
2284 .note = NT_PRSTATUS,
2285 .size = sizeof(elf_prstatus_t),
2286 .get = __elfN(get_prstatus),
2287 .set = __elfN(set_prstatus),
2289 ELF_REGSET(__elfN(regset_prstatus));
2292 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2295 elf_prfpregset_t *fpregset;
2298 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2301 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2302 fill_fpregs32(td, fpregset);
2304 fill_fpregs(td, fpregset);
2307 *sizep = sizeof(*fpregset);
2312 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2315 elf_prfpregset_t *fpregset;
2318 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2319 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2320 set_fpregs32(td, fpregset);
2322 set_fpregs(td, fpregset);
2327 static struct regset __elfN(regset_fpregset) = {
2328 .note = NT_FPREGSET,
2329 .size = sizeof(elf_prfpregset_t),
2330 .get = __elfN(get_fpregset),
2331 .set = __elfN(set_fpregset),
2333 ELF_REGSET(__elfN(regset_fpregset));
2336 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2339 elf_thrmisc_t *thrmisc;
2342 KASSERT(*sizep == sizeof(*thrmisc),
2343 ("%s: invalid size", __func__));
2345 bzero(thrmisc, sizeof(*thrmisc));
2346 strcpy(thrmisc->pr_tname, td->td_name);
2348 *sizep = sizeof(*thrmisc);
2352 static struct regset __elfN(regset_thrmisc) = {
2354 .size = sizeof(elf_thrmisc_t),
2355 .get = __elfN(get_thrmisc),
2357 ELF_REGSET(__elfN(regset_thrmisc));
2360 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2367 size = sizeof(structsize) + sizeof(pl);
2369 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2370 structsize = sizeof(pl);
2371 memcpy(buf, &structsize, sizeof(structsize));
2372 bzero(&pl, sizeof(pl));
2373 pl.pl_lwpid = td->td_tid;
2374 pl.pl_event = PL_EVENT_NONE;
2375 pl.pl_sigmask = td->td_sigmask;
2376 pl.pl_siglist = td->td_siglist;
2377 if (td->td_si.si_signo != 0) {
2378 pl.pl_event = PL_EVENT_SIGNAL;
2379 pl.pl_flags |= PL_FLAG_SI;
2380 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2381 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2383 pl.pl_siginfo = td->td_si;
2386 strcpy(pl.pl_tdname, td->td_name);
2387 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2388 memcpy((int *)buf + 1, &pl, sizeof(pl));
2394 static struct regset __elfN(regset_lwpinfo) = {
2395 .note = NT_PTLWPINFO,
2396 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2397 .get = __elfN(get_lwpinfo),
2399 ELF_REGSET(__elfN(regset_lwpinfo));
2402 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2403 struct thread *target_td)
2405 struct sysentvec *sv = td->td_proc->p_sysent;
2406 struct regset **regsetp, **regset_end, *regset;
2411 /* NT_PRSTATUS must be the first register set note. */
2412 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2415 regsetp = sv->sv_regset_begin;
2416 if (regsetp == NULL) {
2417 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2418 size += __elfN(register_regset_note)(td, list,
2419 &__elfN(regset_fpregset), target_td);
2422 regset_end = sv->sv_regset_end;
2423 MPASS(regset_end != NULL);
2424 for (; regsetp < regset_end; regsetp++) {
2426 if (regset->note == NT_PRSTATUS)
2428 size += __elfN(register_regset_note)(td, list, regset,
2435 * Allow for MD specific notes, as well as any MD
2436 * specific preparations for writing MI notes.
2439 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2445 td = (struct thread *)arg;
2447 if (size != 0 && sb != NULL)
2448 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2452 __elfN(dump_thread)(td, buf, &size);
2453 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2454 if (size != 0 && sb != NULL)
2455 sbuf_bcat(sb, buf, size);
2460 #ifdef KINFO_PROC_SIZE
2461 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2465 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2472 size = sizeof(structsize) + p->p_numthreads *
2473 sizeof(elf_kinfo_proc_t);
2476 KASSERT(*sizep == size, ("invalid size"));
2477 structsize = sizeof(elf_kinfo_proc_t);
2478 sbuf_bcat(sb, &structsize, sizeof(structsize));
2479 sx_slock(&proctree_lock);
2481 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2482 sx_sunlock(&proctree_lock);
2487 #ifdef KINFO_FILE_SIZE
2488 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2492 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2495 size_t size, sect_sz, i;
2496 ssize_t start_len, sect_len;
2497 int structsize, filedesc_flags;
2499 if (coredump_pack_fileinfo)
2500 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2505 structsize = sizeof(struct kinfo_file);
2508 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2509 sbuf_set_drain(sb, sbuf_count_drain, &size);
2510 sbuf_bcat(sb, &structsize, sizeof(structsize));
2512 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2517 sbuf_start_section(sb, &start_len);
2519 sbuf_bcat(sb, &structsize, sizeof(structsize));
2521 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2524 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2529 KASSERT(sect_sz <= *sizep,
2530 ("kern_proc_filedesc_out did not respect maxlen; "
2531 "requested %zu, got %zu", *sizep - sizeof(structsize),
2532 sect_sz - sizeof(structsize)));
2534 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2539 #ifdef KINFO_VMENTRY_SIZE
2540 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2544 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2548 int structsize, vmmap_flags;
2550 if (coredump_pack_vmmapinfo)
2551 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2556 structsize = sizeof(struct kinfo_vmentry);
2559 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2560 sbuf_set_drain(sb, sbuf_count_drain, &size);
2561 sbuf_bcat(sb, &structsize, sizeof(structsize));
2563 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2568 sbuf_bcat(sb, &structsize, sizeof(structsize));
2570 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2576 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2583 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2585 KASSERT(*sizep == size, ("invalid size"));
2586 structsize = sizeof(gid_t);
2587 sbuf_bcat(sb, &structsize, sizeof(structsize));
2588 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2595 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2602 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2604 KASSERT(*sizep == size, ("invalid size"));
2605 structsize = sizeof(p->p_pd->pd_cmask);
2606 sbuf_bcat(sb, &structsize, sizeof(structsize));
2607 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2613 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2616 struct rlimit rlim[RLIM_NLIMITS];
2621 size = sizeof(structsize) + sizeof(rlim);
2623 KASSERT(*sizep == size, ("invalid size"));
2624 structsize = sizeof(rlim);
2625 sbuf_bcat(sb, &structsize, sizeof(structsize));
2627 for (i = 0; i < RLIM_NLIMITS; i++)
2628 lim_rlimit_proc(p, i, &rlim[i]);
2630 sbuf_bcat(sb, rlim, sizeof(rlim));
2636 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2643 size = sizeof(structsize) + sizeof(p->p_osrel);
2645 KASSERT(*sizep == size, ("invalid size"));
2646 structsize = sizeof(p->p_osrel);
2647 sbuf_bcat(sb, &structsize, sizeof(structsize));
2648 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2654 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2657 elf_ps_strings_t ps_strings;
2662 size = sizeof(structsize) + sizeof(ps_strings);
2664 KASSERT(*sizep == size, ("invalid size"));
2665 structsize = sizeof(ps_strings);
2666 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2667 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2669 ps_strings = PROC_PS_STRINGS(p);
2671 sbuf_bcat(sb, &structsize, sizeof(structsize));
2672 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2678 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2687 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2689 sbuf_set_drain(sb, sbuf_count_drain, &size);
2690 sbuf_bcat(sb, &structsize, sizeof(structsize));
2692 proc_getauxv(curthread, p, sb);
2698 structsize = sizeof(Elf_Auxinfo);
2699 sbuf_bcat(sb, &structsize, sizeof(structsize));
2701 proc_getauxv(curthread, p, sb);
2707 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2708 const char *note_vendor, const Elf_Phdr *pnote,
2709 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2711 const Elf_Note *note, *note0, *note_end;
2712 const char *note_name;
2717 /* We need some limit, might as well use PAGE_SIZE. */
2718 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2720 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2721 if (pnote->p_offset > PAGE_SIZE ||
2722 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2723 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2725 VOP_UNLOCK(imgp->vp);
2726 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2727 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2729 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2730 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2731 curthread->td_ucred, NOCRED, NULL, curthread);
2733 uprintf("i/o error PT_NOTE\n");
2736 note = note0 = (const Elf_Note *)buf;
2737 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2739 note = note0 = (const Elf_Note *)(imgp->image_header +
2741 note_end = (const Elf_Note *)(imgp->image_header +
2742 pnote->p_offset + pnote->p_filesz);
2745 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2746 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2747 (const char *)note < sizeof(Elf_Note)) {
2750 if (note->n_namesz != checknote->n_namesz ||
2751 note->n_descsz != checknote->n_descsz ||
2752 note->n_type != checknote->n_type)
2754 note_name = (const char *)(note + 1);
2755 if (note_name + checknote->n_namesz >=
2756 (const char *)note_end || strncmp(note_vendor,
2757 note_name, checknote->n_namesz) != 0)
2760 if (cb(note, cb_arg, &res))
2763 note = (const Elf_Note *)((const char *)(note + 1) +
2764 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2765 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2774 struct brandnote_cb_arg {
2775 Elf_Brandnote *brandnote;
2780 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2782 struct brandnote_cb_arg *arg;
2787 * Fetch the osreldate for binary from the ELF OSABI-note if
2790 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2791 arg->brandnote->trans_osrel != NULL ?
2792 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2797 static Elf_Note fctl_note = {
2798 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2799 .n_descsz = sizeof(uint32_t),
2800 .n_type = NT_FREEBSD_FEATURE_CTL,
2803 struct fctl_cb_arg {
2804 boolean_t *has_fctl0;
2809 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2811 struct fctl_cb_arg *arg;
2812 const Elf32_Word *desc;
2816 p = (uintptr_t)(note + 1);
2817 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2818 desc = (const Elf32_Word *)p;
2819 *arg->has_fctl0 = TRUE;
2820 *arg->fctl0 = desc[0];
2826 * Try to find the appropriate ABI-note section for checknote, fetch
2827 * the osreldate and feature control flags for binary from the ELF
2828 * OSABI-note. Only the first page of the image is searched, the same
2832 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2833 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2835 const Elf_Phdr *phdr;
2836 const Elf_Ehdr *hdr;
2837 struct brandnote_cb_arg b_arg;
2838 struct fctl_cb_arg f_arg;
2841 hdr = (const Elf_Ehdr *)imgp->image_header;
2842 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2843 b_arg.brandnote = brandnote;
2844 b_arg.osrel = osrel;
2845 f_arg.has_fctl0 = has_fctl0;
2846 f_arg.fctl0 = fctl0;
2848 for (i = 0; i < hdr->e_phnum; i++) {
2849 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2850 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2852 for (j = 0; j < hdr->e_phnum; j++) {
2853 if (phdr[j].p_type == PT_NOTE &&
2854 __elfN(parse_notes)(imgp, &fctl_note,
2855 FREEBSD_ABI_VENDOR, &phdr[j],
2856 note_fctl_cb, &f_arg))
2867 * Tell kern_execve.c about it, with a little help from the linker.
2869 static struct execsw __elfN(execsw) = {
2870 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2871 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2873 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2876 __elfN(trans_prot)(Elf_Word flags)
2882 prot |= VM_PROT_EXECUTE;
2884 prot |= VM_PROT_WRITE;
2886 prot |= VM_PROT_READ;
2887 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2888 if (i386_read_exec && (flags & PF_R))
2889 prot |= VM_PROT_EXECUTE;
2895 __elfN(untrans_prot)(vm_prot_t prot)
2900 if (prot & VM_PROT_EXECUTE)
2902 if (prot & VM_PROT_READ)
2904 if (prot & VM_PROT_WRITE)