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 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static bool __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 static size_t __elfN(prepare_register_notes)(struct thread *td,
106 struct note_info_list *list, struct thread *target_td);
108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
109 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119 &elf_legacy_coredump, 0,
120 "include all and only RW pages in core dumps");
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124 (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 defined(__amd64__)
135 static int __elfN(vdso) = 1;
136 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
137 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
138 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
140 static int __elfN(vdso) = 0;
143 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
144 int i386_read_exec = 0;
145 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
146 "enable execution from readable segments");
149 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
151 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
156 val = __elfN(pie_base);
157 error = sysctl_handle_long(oidp, &val, 0, req);
158 if (error != 0 || req->newptr == NULL)
160 if ((val & PAGE_MASK) != 0)
162 __elfN(pie_base) = val;
165 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
166 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
167 sysctl_pie_base, "LU",
168 "PIE load base without randomization");
170 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
176 * While for 64-bit machines ASLR works properly, there are
177 * still some problems when using 32-bit architectures. For this
178 * reason ASLR is only enabled by default when running native
179 * 64-bit non-PIE executables.
181 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
182 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
183 &__elfN(aslr_enabled), 0,
184 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
185 ": enable address map randomization");
188 * Enable ASLR only for 64-bit PIE binaries by default.
190 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
191 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
192 &__elfN(pie_aslr_enabled), 0,
193 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
194 ": enable address map randomization for PIE binaries");
197 * Sbrk is now deprecated and it can be assumed, that in most
198 * cases it will not be used anyway. This setting is valid only
199 * for the ASLR enabled and allows for utilizing the bss grow region.
201 static int __elfN(aslr_honor_sbrk) = 0;
202 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
203 &__elfN(aslr_honor_sbrk), 0,
204 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
206 static int __elfN(aslr_stack) = 1;
207 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
208 &__elfN(aslr_stack), 0,
209 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
210 ": enable stack address randomization");
212 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64;
213 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN,
214 &__elfN(aslr_shared_page), 0,
215 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
216 ": enable shared page address randomization");
218 static int __elfN(sigfastblock) = 1;
219 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
220 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
221 "enable sigfastblock for new processes");
223 static bool __elfN(allow_wx) = true;
224 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
225 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
226 "Allow pages to be mapped simultaneously writable and executable");
228 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
230 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
232 Elf_Brandnote __elfN(freebsd_brandnote) = {
233 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
234 .hdr.n_descsz = sizeof(int32_t),
235 .hdr.n_type = NT_FREEBSD_ABI_TAG,
236 .vendor = FREEBSD_ABI_VENDOR,
237 .flags = BN_TRANSLATE_OSREL,
238 .trans_osrel = __elfN(freebsd_trans_osrel)
242 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
246 p = (uintptr_t)(note + 1);
247 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
248 *osrel = *(const int32_t *)(p);
253 static const char GNU_ABI_VENDOR[] = "GNU";
254 static int GNU_KFREEBSD_ABI_DESC = 3;
256 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
257 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
258 .hdr.n_descsz = 16, /* XXX at least 16 */
260 .vendor = GNU_ABI_VENDOR,
261 .flags = BN_TRANSLATE_OSREL,
262 .trans_osrel = kfreebsd_trans_osrel
266 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
268 const Elf32_Word *desc;
271 p = (uintptr_t)(note + 1);
272 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
274 desc = (const Elf32_Word *)p;
275 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
279 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
280 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
282 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
288 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
292 for (i = 0; i < MAX_BRANDS; i++) {
293 if (elf_brand_list[i] == NULL) {
294 elf_brand_list[i] = entry;
298 if (i == MAX_BRANDS) {
299 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
307 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
311 for (i = 0; i < MAX_BRANDS; i++) {
312 if (elf_brand_list[i] == entry) {
313 elf_brand_list[i] = NULL;
323 __elfN(brand_inuse)(Elf_Brandinfo *entry)
328 sx_slock(&allproc_lock);
329 FOREACH_PROC_IN_SYSTEM(p) {
330 if (p->p_sysent == entry->sysvec) {
335 sx_sunlock(&allproc_lock);
340 static Elf_Brandinfo *
341 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
342 int32_t *osrel, uint32_t *fctl0)
344 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
345 Elf_Brandinfo *bi, *bi_m;
347 int i, interp_name_len;
349 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
352 * We support four types of branding -- (1) the ELF EI_OSABI field
353 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
354 * branding w/in the ELF header, (3) path of the `interp_path'
355 * field, and (4) the ".note.ABI-tag" ELF section.
358 /* Look for an ".note.ABI-tag" ELF section */
360 for (i = 0; i < MAX_BRANDS; i++) {
361 bi = elf_brand_list[i];
364 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
366 if (hdr->e_machine == bi->machine && (bi->flags &
367 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
371 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
373 /* Give brand a chance to veto check_note's guess */
374 if (ret && bi->header_supported) {
375 ret = bi->header_supported(imgp, osrel,
376 has_fctl0 ? fctl0 : NULL);
379 * If note checker claimed the binary, but the
380 * interpreter path in the image does not
381 * match default one for the brand, try to
382 * search for other brands with the same
383 * interpreter. Either there is better brand
384 * with the right interpreter, or, failing
385 * this, we return first brand which accepted
386 * our note and, optionally, header.
388 if (ret && bi_m == NULL && interp != NULL &&
389 (bi->interp_path == NULL ||
390 (strlen(bi->interp_path) + 1 != interp_name_len ||
391 strncmp(interp, bi->interp_path, interp_name_len)
403 /* If the executable has a brand, search for it in the brand list. */
404 for (i = 0; i < MAX_BRANDS; i++) {
405 bi = elf_brand_list[i];
406 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
407 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
409 if (hdr->e_machine == bi->machine &&
410 (hdr->e_ident[EI_OSABI] == bi->brand ||
411 (bi->compat_3_brand != NULL &&
412 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
413 bi->compat_3_brand) == 0))) {
414 /* Looks good, but give brand a chance to veto */
415 if (bi->header_supported == NULL ||
416 bi->header_supported(imgp, NULL, NULL)) {
418 * Again, prefer strictly matching
421 if (interp_name_len == 0 &&
422 bi->interp_path == NULL)
424 if (bi->interp_path != NULL &&
425 strlen(bi->interp_path) + 1 ==
426 interp_name_len && strncmp(interp,
427 bi->interp_path, interp_name_len) == 0)
437 /* No known brand, see if the header is recognized by any brand */
438 for (i = 0; i < MAX_BRANDS; i++) {
439 bi = elf_brand_list[i];
440 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
441 bi->header_supported == NULL)
443 if (hdr->e_machine == bi->machine) {
444 ret = bi->header_supported(imgp, NULL, NULL);
450 /* Lacking a known brand, search for a recognized interpreter. */
451 if (interp != NULL) {
452 for (i = 0; i < MAX_BRANDS; i++) {
453 bi = elf_brand_list[i];
454 if (bi == NULL || (bi->flags &
455 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
458 if (hdr->e_machine == bi->machine &&
459 bi->interp_path != NULL &&
460 /* ELF image p_filesz includes terminating zero */
461 strlen(bi->interp_path) + 1 == interp_name_len &&
462 strncmp(interp, bi->interp_path, interp_name_len)
463 == 0 && (bi->header_supported == NULL ||
464 bi->header_supported(imgp, NULL, NULL)))
469 /* Lacking a recognized interpreter, try the default brand */
470 for (i = 0; i < MAX_BRANDS; i++) {
471 bi = elf_brand_list[i];
472 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
473 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
475 if (hdr->e_machine == bi->machine &&
476 __elfN(fallback_brand) == bi->brand &&
477 (bi->header_supported == NULL ||
478 bi->header_supported(imgp, NULL, NULL)))
485 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
487 return (hdr->e_phoff <= PAGE_SIZE &&
488 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
492 __elfN(check_header)(const Elf_Ehdr *hdr)
498 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
499 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
500 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
501 hdr->e_phentsize != sizeof(Elf_Phdr) ||
502 hdr->e_version != ELF_TARG_VER)
506 * Make sure we have at least one brand for this machine.
509 for (i = 0; i < MAX_BRANDS; i++) {
510 bi = elf_brand_list[i];
511 if (bi != NULL && bi->machine == hdr->e_machine)
521 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
522 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
529 * Create the page if it doesn't exist yet. Ignore errors.
531 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
532 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
535 * Find the page from the underlying object.
537 if (object != NULL) {
538 sf = vm_imgact_map_page(object, offset);
540 return (KERN_FAILURE);
541 off = offset - trunc_page(offset);
542 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
544 vm_imgact_unmap_page(sf);
546 return (KERN_FAILURE);
549 return (KERN_SUCCESS);
553 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
554 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
560 int error, locked, rv;
562 if (start != trunc_page(start)) {
563 rv = __elfN(map_partial)(map, object, offset, start,
564 round_page(start), prot);
565 if (rv != KERN_SUCCESS)
567 offset += round_page(start) - start;
568 start = round_page(start);
570 if (end != round_page(end)) {
571 rv = __elfN(map_partial)(map, object, offset +
572 trunc_page(end) - start, trunc_page(end), end, prot);
573 if (rv != KERN_SUCCESS)
575 end = trunc_page(end);
578 return (KERN_SUCCESS);
579 if ((offset & PAGE_MASK) != 0) {
581 * The mapping is not page aligned. This means that we have
584 rv = vm_map_fixed(map, NULL, 0, start, end - start,
585 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
586 if (rv != KERN_SUCCESS)
589 return (KERN_SUCCESS);
590 for (; start < end; start += sz) {
591 sf = vm_imgact_map_page(object, offset);
593 return (KERN_FAILURE);
594 off = offset - trunc_page(offset);
596 if (sz > PAGE_SIZE - off)
597 sz = PAGE_SIZE - off;
598 error = copyout((caddr_t)sf_buf_kva(sf) + off,
600 vm_imgact_unmap_page(sf);
602 return (KERN_FAILURE);
606 vm_object_reference(object);
607 rv = vm_map_fixed(map, object, offset, start, end - start,
608 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
609 (object != NULL ? MAP_VN_EXEC : 0));
610 if (rv != KERN_SUCCESS) {
611 locked = VOP_ISLOCKED(imgp->vp);
612 VOP_UNLOCK(imgp->vp);
613 vm_object_deallocate(object);
614 vn_lock(imgp->vp, locked | LK_RETRY);
616 } else if (object != NULL) {
617 MPASS(imgp->vp->v_object == object);
618 VOP_SET_TEXT_CHECKED(imgp->vp);
621 return (KERN_SUCCESS);
625 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
626 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
632 vm_offset_t map_addr;
635 vm_ooffset_t file_addr;
638 * It's necessary to fail if the filsz + offset taken from the
639 * header is greater than the actual file pager object's size.
640 * If we were to allow this, then the vm_map_find() below would
641 * walk right off the end of the file object and into the ether.
643 * While I'm here, might as well check for something else that
644 * is invalid: filsz cannot be greater than memsz.
646 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
648 uprintf("elf_load_section: truncated ELF file\n");
652 object = imgp->object;
653 map = &imgp->proc->p_vmspace->vm_map;
654 map_addr = trunc_page((vm_offset_t)vmaddr);
655 file_addr = trunc_page(offset);
658 * We have two choices. We can either clear the data in the last page
659 * of an oversized mapping, or we can start the anon mapping a page
660 * early and copy the initialized data into that first page. We
665 else if (memsz > filsz)
666 map_len = trunc_page(offset + filsz) - file_addr;
668 map_len = round_page(offset + filsz) - file_addr;
671 /* cow flags: don't dump readonly sections in core */
672 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
673 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
675 rv = __elfN(map_insert)(imgp, map, object, file_addr,
676 map_addr, map_addr + map_len, prot, cow);
677 if (rv != KERN_SUCCESS)
680 /* we can stop now if we've covered it all */
686 * We have to get the remaining bit of the file into the first part
687 * of the oversized map segment. This is normally because the .data
688 * segment in the file is extended to provide bss. It's a neat idea
689 * to try and save a page, but it's a pain in the behind to implement.
691 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
693 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
694 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
696 /* This had damn well better be true! */
698 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
699 map_addr + map_len, prot, 0);
700 if (rv != KERN_SUCCESS)
705 sf = vm_imgact_map_page(object, offset + filsz);
709 /* send the page fragment to user space */
710 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
712 vm_imgact_unmap_page(sf);
718 * Remove write access to the page if it was only granted by map_insert
721 if ((prot & VM_PROT_WRITE) == 0)
722 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
723 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
729 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
730 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
737 ASSERT_VOP_LOCKED(imgp->vp, __func__);
742 for (i = 0; i < hdr->e_phnum; i++) {
743 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
746 /* Loadable segment */
747 prot = __elfN(trans_prot)(phdr[i].p_flags);
748 error = __elfN(load_section)(imgp, phdr[i].p_offset,
749 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
750 phdr[i].p_memsz, phdr[i].p_filesz, prot);
755 * Establish the base address if this is the first segment.
758 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
763 if (base_addrp != NULL)
764 *base_addrp = base_addr;
770 * Load the file "file" into memory. It may be either a shared object
773 * The "addr" reference parameter is in/out. On entry, it specifies
774 * the address where a shared object should be loaded. If the file is
775 * an executable, this value is ignored. On exit, "addr" specifies
776 * where the file was actually loaded.
778 * The "entry" reference parameter is out only. On exit, it specifies
779 * the entry point for the loaded file.
782 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
788 struct image_params image_params;
790 const Elf_Ehdr *hdr = NULL;
791 const Elf_Phdr *phdr = NULL;
792 struct nameidata *nd;
794 struct image_params *imgp;
796 u_long base_addr = 0;
799 #ifdef CAPABILITY_MODE
801 * XXXJA: This check can go away once we are sufficiently confident
802 * that the checks in namei() are correct.
804 if (IN_CAPABILITY_MODE(curthread))
808 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
810 attr = &tempdata->attr;
811 imgp = &tempdata->image_params;
814 * Initialize part of the common data
819 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
821 if ((error = namei(nd)) != 0) {
826 imgp->vp = nd->ni_vp;
829 * Check permissions, modes, uid, etc on the file, and "open" it.
831 error = exec_check_permissions(imgp);
835 error = exec_map_first_page(imgp);
839 imgp->object = nd->ni_vp->v_object;
841 hdr = (const Elf_Ehdr *)imgp->image_header;
842 if ((error = __elfN(check_header)(hdr)) != 0)
844 if (hdr->e_type == ET_DYN)
846 else if (hdr->e_type == ET_EXEC)
853 /* Only support headers that fit within first page for now */
854 if (!__elfN(phdr_in_zero_page)(hdr)) {
859 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
860 if (!aligned(phdr, Elf_Addr)) {
865 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
870 *entry = (unsigned long)hdr->e_entry + rbase;
874 exec_unmap_first_page(imgp);
878 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
881 free(tempdata, M_TEMP);
887 * Select randomized valid address in the map map, between minv and
888 * maxv, with specified alignment. The [minv, maxv) range must belong
889 * to the map. Note that function only allocates the address, it is
890 * up to caller to clamp maxv in a way that the final allocation
891 * length fit into the map.
893 * Result is returned in *resp, error code indicates that arguments
894 * did not pass sanity checks for overflow and range correctness.
897 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
898 u_int align, u_long *resp)
902 MPASS(vm_map_min(map) <= minv);
904 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
905 uprintf("Invalid ELF segments layout\n");
909 arc4rand(&rbase, sizeof(rbase), 0);
910 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
911 res &= ~((u_long)align - 1);
916 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
917 res, minv, maxv, rbase));
919 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
920 res, maxv, minv, rbase));
927 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
928 const Elf_Phdr *phdr, u_long et_dyn_addr)
930 struct vmspace *vmspace;
932 u_long text_size, data_size, total_size, text_addr, data_addr;
933 u_long seg_size, seg_addr;
937 text_size = data_size = total_size = text_addr = data_addr = 0;
939 for (i = 0; i < hdr->e_phnum; i++) {
940 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
943 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
944 seg_size = round_page(phdr[i].p_memsz +
945 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
948 * Make the largest executable segment the official
949 * text segment and all others data.
951 * Note that obreak() assumes that data_addr + data_size == end
952 * of data load area, and the ELF file format expects segments
953 * to be sorted by address. If multiple data segments exist,
954 * the last one will be used.
957 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
958 text_size = seg_size;
959 text_addr = seg_addr;
961 data_size = seg_size;
962 data_addr = seg_addr;
964 total_size += seg_size;
967 if (data_addr == 0 && data_size == 0) {
968 data_addr = text_addr;
969 data_size = text_size;
973 * Check limits. It should be safe to check the
974 * limits after loading the segments since we do
975 * not actually fault in all the segments pages.
977 PROC_LOCK(imgp->proc);
978 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
979 err_str = "Data segment size exceeds process limit";
980 else if (text_size > maxtsiz)
981 err_str = "Text segment size exceeds system limit";
982 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
983 err_str = "Total segment size exceeds process limit";
984 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
985 err_str = "Data segment size exceeds resource limit";
986 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
987 err_str = "Total segment size exceeds resource limit";
988 PROC_UNLOCK(imgp->proc);
989 if (err_str != NULL) {
990 uprintf("%s\n", err_str);
994 vmspace = imgp->proc->p_vmspace;
995 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
996 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
997 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
998 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1004 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
1005 char **interpp, bool *free_interpp)
1009 int error, interp_name_len;
1011 KASSERT(phdr->p_type == PT_INTERP,
1012 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1013 ASSERT_VOP_LOCKED(imgp->vp, __func__);
1017 /* Path to interpreter */
1018 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1019 uprintf("Invalid PT_INTERP\n");
1023 interp_name_len = phdr->p_filesz;
1024 if (phdr->p_offset > PAGE_SIZE ||
1025 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1027 * The vnode lock might be needed by the pagedaemon to
1028 * clean pages owned by the vnode. Do not allow sleep
1029 * waiting for memory with the vnode locked, instead
1030 * try non-sleepable allocation first, and if it
1031 * fails, go to the slow path were we drop the lock
1032 * and do M_WAITOK. A text reference prevents
1033 * modifications to the vnode content.
1035 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1036 if (interp == NULL) {
1037 VOP_UNLOCK(imgp->vp);
1038 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1039 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1042 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1043 interp_name_len, phdr->p_offset,
1044 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1047 free(interp, M_TEMP);
1048 uprintf("i/o error PT_INTERP %d\n", error);
1051 interp[interp_name_len] = '\0';
1054 *free_interpp = true;
1058 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1059 if (interp[interp_name_len - 1] != '\0') {
1060 uprintf("Invalid PT_INTERP\n");
1065 *free_interpp = false;
1070 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1071 const char *interp, u_long *addr, u_long *entry)
1076 if (brand_info->emul_path != NULL &&
1077 brand_info->emul_path[0] != '\0') {
1078 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1079 snprintf(path, MAXPATHLEN, "%s%s",
1080 brand_info->emul_path, interp);
1081 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1087 if (brand_info->interp_newpath != NULL &&
1088 (brand_info->interp_path == NULL ||
1089 strcmp(interp, brand_info->interp_path) == 0)) {
1090 error = __elfN(load_file)(imgp->proc,
1091 brand_info->interp_newpath, addr, entry);
1096 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1100 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1105 * Impossible et_dyn_addr initial value indicating that the real base
1106 * must be calculated later with some randomization applied.
1108 #define ET_DYN_ADDR_RAND 1
1111 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1114 const Elf_Ehdr *hdr;
1115 const Elf_Phdr *phdr;
1116 Elf_Auxargs *elf_auxargs;
1117 struct vmspace *vmspace;
1120 Elf_Brandinfo *brand_info;
1121 struct sysentvec *sv;
1122 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1123 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1129 hdr = (const Elf_Ehdr *)imgp->image_header;
1132 * Do we have a valid ELF header ?
1134 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1135 * if particular brand doesn't support it.
1137 if (__elfN(check_header)(hdr) != 0 ||
1138 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1142 * From here on down, we return an errno, not -1, as we've
1143 * detected an ELF file.
1146 if (!__elfN(phdr_in_zero_page)(hdr)) {
1147 uprintf("Program headers not in the first page\n");
1150 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1151 if (!aligned(phdr, Elf_Addr)) {
1152 uprintf("Unaligned program headers\n");
1160 entry = proghdr = 0;
1162 free_interp = false;
1166 * Somewhat arbitrary, limit accepted max alignment for the
1167 * loadable segment to the max supported superpage size. Too
1168 * large alignment requests are not useful and are indicators
1169 * of corrupted or outright malicious binary.
1171 maxalign = PAGE_SIZE;
1172 maxsalign = PAGE_SIZE * 1024;
1173 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1174 if (pagesizes[i] > maxsalign)
1175 maxsalign = pagesizes[i];
1180 for (i = 0; i < hdr->e_phnum; i++) {
1181 switch (phdr[i].p_type) {
1184 baddr = phdr[i].p_vaddr;
1185 if (!powerof2(phdr[i].p_align) ||
1186 phdr[i].p_align > maxsalign) {
1187 uprintf("Invalid segment alignment\n");
1191 if (phdr[i].p_align > maxalign)
1192 maxalign = phdr[i].p_align;
1193 if (mapsz + phdr[i].p_memsz < mapsz) {
1194 uprintf("Mapsize overflow\n");
1198 mapsz += phdr[i].p_memsz;
1202 * If this segment contains the program headers,
1203 * remember their virtual address for the AT_PHDR
1204 * aux entry. Static binaries don't usually include
1207 if (phdr[i].p_offset == 0 &&
1208 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1210 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1213 /* Path to interpreter */
1214 if (interp != NULL) {
1215 uprintf("Multiple PT_INTERP headers\n");
1219 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1225 if (__elfN(nxstack))
1227 __elfN(trans_prot)(phdr[i].p_flags);
1228 imgp->stack_sz = phdr[i].p_memsz;
1230 case PT_PHDR: /* Program header table info */
1231 proghdr = phdr[i].p_vaddr;
1236 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1237 if (brand_info == NULL) {
1238 uprintf("ELF binary type \"%u\" not known.\n",
1239 hdr->e_ident[EI_OSABI]);
1243 sv = brand_info->sysvec;
1245 if (hdr->e_type == ET_DYN) {
1246 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1247 uprintf("Cannot execute shared object\n");
1252 * Honour the base load address from the dso if it is
1253 * non-zero for some reason.
1256 if ((sv->sv_flags & SV_ASLR) == 0 ||
1257 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1258 et_dyn_addr = __elfN(pie_base);
1259 else if ((__elfN(pie_aslr_enabled) &&
1260 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1261 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1262 et_dyn_addr = ET_DYN_ADDR_RAND;
1264 et_dyn_addr = __elfN(pie_base);
1269 * Avoid a possible deadlock if the current address space is destroyed
1270 * and that address space maps the locked vnode. In the common case,
1271 * the locked vnode's v_usecount is decremented but remains greater
1272 * than zero. Consequently, the vnode lock is not needed by vrele().
1273 * However, in cases where the vnode lock is external, such as nullfs,
1274 * v_usecount may become zero.
1276 * The VV_TEXT flag prevents modifications to the executable while
1277 * the vnode is unlocked.
1279 VOP_UNLOCK(imgp->vp);
1282 * Decide whether to enable randomization of user mappings.
1283 * First, reset user preferences for the setid binaries.
1284 * Then, account for the support of the randomization by the
1285 * ABI, by user preferences, and make special treatment for
1288 if (imgp->credential_setid) {
1289 PROC_LOCK(imgp->proc);
1290 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1291 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1292 PROC_UNLOCK(imgp->proc);
1294 if ((sv->sv_flags & SV_ASLR) == 0 ||
1295 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1296 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1297 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1298 ("et_dyn_addr == RAND and !ASLR"));
1299 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1300 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1301 et_dyn_addr == ET_DYN_ADDR_RAND) {
1302 imgp->map_flags |= MAP_ASLR;
1304 * If user does not care about sbrk, utilize the bss
1305 * grow region for mappings as well. We can select
1306 * the base for the image anywere and still not suffer
1307 * from the fragmentation.
1309 if (!__elfN(aslr_honor_sbrk) ||
1310 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1311 imgp->map_flags |= MAP_ASLR_IGNSTART;
1312 if (__elfN(aslr_stack))
1313 imgp->map_flags |= MAP_ASLR_STACK;
1314 if (__elfN(aslr_shared_page))
1315 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE;
1318 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1319 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1320 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1321 imgp->map_flags |= MAP_WXORX;
1323 error = exec_new_vmspace(imgp, sv);
1325 imgp->proc->p_sysent = sv;
1326 imgp->proc->p_elf_brandinfo = brand_info;
1328 vmspace = imgp->proc->p_vmspace;
1329 map = &vmspace->vm_map;
1330 maxv = sv->sv_usrstack;
1331 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1332 maxv -= lim_max(td, RLIMIT_STACK);
1333 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1334 uprintf("Excessive mapping size\n");
1338 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1339 KASSERT((map->flags & MAP_ASLR) != 0,
1340 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1341 error = __CONCAT(rnd_, __elfN(base))(map,
1342 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1343 /* reserve half of the address space to interpreter */
1344 maxv / 2, maxalign, &et_dyn_addr);
1347 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1351 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1355 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1360 * We load the dynamic linker where a userland call
1361 * to mmap(0, ...) would put it. The rationale behind this
1362 * calculation is that it leaves room for the heap to grow to
1363 * its maximum allowed size.
1365 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1367 if ((map->flags & MAP_ASLR) != 0) {
1368 maxv1 = maxv / 2 + addr / 2;
1369 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1370 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1371 pagesizes[1] : pagesizes[0], &anon_loc);
1374 map->anon_loc = anon_loc;
1376 map->anon_loc = addr;
1379 entry = (u_long)hdr->e_entry + et_dyn_addr;
1380 imgp->entry_addr = entry;
1382 if (interp != NULL) {
1383 VOP_UNLOCK(imgp->vp);
1384 if ((map->flags & MAP_ASLR) != 0) {
1385 /* Assume that interpreter fits into 1/4 of AS */
1386 maxv1 = maxv / 2 + addr / 2;
1387 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1388 maxv1, PAGE_SIZE, &addr);
1391 error = __elfN(load_interp)(imgp, brand_info, interp,
1392 &addr, &imgp->entry_addr);
1394 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1400 error = exec_map_stack(imgp);
1405 * Construct auxargs table (used by the copyout_auxargs routine)
1407 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1408 if (elf_auxargs == NULL) {
1409 VOP_UNLOCK(imgp->vp);
1410 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1411 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1413 elf_auxargs->execfd = -1;
1414 elf_auxargs->phdr = proghdr + et_dyn_addr;
1415 elf_auxargs->phent = hdr->e_phentsize;
1416 elf_auxargs->phnum = hdr->e_phnum;
1417 elf_auxargs->pagesz = PAGE_SIZE;
1418 elf_auxargs->base = addr;
1419 elf_auxargs->flags = 0;
1420 elf_auxargs->entry = entry;
1421 elf_auxargs->hdr_eflags = hdr->e_flags;
1423 imgp->auxargs = elf_auxargs;
1424 imgp->interpreted = 0;
1425 imgp->reloc_base = addr;
1426 imgp->proc->p_osrel = osrel;
1427 imgp->proc->p_fctl0 = fctl0;
1428 imgp->proc->p_elf_flags = hdr->e_flags;
1431 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1433 free(interp, M_TEMP);
1437 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1440 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1442 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1443 Elf_Auxinfo *argarray, *pos;
1444 struct vmspace *vmspace;
1447 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1450 vmspace = imgp->proc->p_vmspace;
1452 if (args->execfd != -1)
1453 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1454 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1455 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1456 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1457 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1458 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1459 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1460 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1461 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1462 if (imgp->execpathp != 0)
1463 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1464 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1465 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1466 if (imgp->canary != 0) {
1467 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1468 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1470 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1471 if (imgp->pagesizes != 0) {
1472 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1473 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1475 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) {
1476 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1477 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset);
1479 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1480 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1481 imgp->sysent->sv_stackprot);
1482 if (imgp->sysent->sv_hwcap != NULL)
1483 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1484 if (imgp->sysent->sv_hwcap2 != NULL)
1485 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1486 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1487 ELF_BSDF_SIGFASTBLK : 0);
1488 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1489 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1490 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1491 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1492 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1493 #ifdef RANDOM_FENESTRASX
1494 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) {
1495 AUXARGS_ENTRY(pos, AT_FXRNG,
1496 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset);
1499 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) {
1500 AUXARGS_ENTRY(pos, AT_KPRELOAD,
1501 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset);
1503 AUXARGS_ENTRY(pos, AT_NULL, 0);
1505 free(imgp->auxargs, M_TEMP);
1506 imgp->auxargs = NULL;
1507 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1509 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1510 free(argarray, M_TEMP);
1515 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1519 base = (Elf_Addr *)*stack_base;
1521 if (elf_suword(base, imgp->args->argc) == -1)
1523 *stack_base = (uintptr_t)base;
1528 * Code for generating ELF core dumps.
1531 typedef void (*segment_callback)(vm_map_entry_t, void *);
1533 /* Closure for cb_put_phdr(). */
1534 struct phdr_closure {
1535 Elf_Phdr *phdr; /* Program header to fill in */
1536 Elf_Off offset; /* Offset of segment in core file */
1540 int type; /* Note type. */
1541 struct regset *regset; /* Register set. */
1542 outfunc_t outfunc; /* Output function. */
1543 void *outarg; /* Argument for the output function. */
1544 size_t outsize; /* Output size. */
1545 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1548 TAILQ_HEAD(note_info_list, note_info);
1550 extern int compress_user_cores;
1551 extern int compress_user_cores_level;
1553 static void cb_put_phdr(vm_map_entry_t, void *);
1554 static void cb_size_segment(vm_map_entry_t, void *);
1555 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1557 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1558 struct note_info_list *, size_t, int);
1559 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1561 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1562 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1563 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1564 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1565 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1566 static void note_procstat_files(void *, struct sbuf *, size_t *);
1567 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1568 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1569 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1570 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1571 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1574 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1577 return (core_write((struct coredump_params *)arg, base, len, offset,
1578 UIO_SYSSPACE, NULL));
1582 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1584 struct ucred *cred = td->td_ucred;
1585 int compm, error = 0;
1586 struct sseg_closure seginfo;
1587 struct note_info_list notelst;
1588 struct coredump_params params;
1589 struct note_info *ninfo;
1591 size_t hdrsize, notesz, coresize;
1595 TAILQ_INIT(¬elst);
1597 /* Size the program segments. */
1598 __elfN(size_segments)(td, &seginfo, flags);
1601 * Collect info about the core file header area.
1603 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1604 if (seginfo.count + 1 >= PN_XNUM)
1605 hdrsize += sizeof(Elf_Shdr);
1606 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1607 coresize = round_page(hdrsize + notesz) + seginfo.size;
1609 /* Set up core dump parameters. */
1611 params.active_cred = cred;
1612 params.file_cred = NOCRED;
1619 PROC_LOCK(td->td_proc);
1620 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1621 PROC_UNLOCK(td->td_proc);
1628 if (coresize >= limit) {
1633 /* Create a compression stream if necessary. */
1634 compm = compress_user_cores;
1635 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1637 compm = COMPRESS_GZIP;
1639 params.comp = compressor_init(core_compressed_write,
1640 compm, CORE_BUF_SIZE,
1641 compress_user_cores_level, ¶ms);
1642 if (params.comp == NULL) {
1646 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1650 * Allocate memory for building the header, fill it up,
1651 * and write it out following the notes.
1653 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1654 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1657 /* Write the contents of all of the writable segments. */
1663 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1664 offset = round_page(hdrsize + notesz);
1665 for (i = 0; i < seginfo.count; i++) {
1666 error = core_output((char *)(uintptr_t)php->p_vaddr,
1667 php->p_filesz, offset, ¶ms, tmpbuf);
1670 offset += php->p_filesz;
1673 if (error == 0 && params.comp != NULL)
1674 error = compressor_flush(params.comp);
1678 "Failed to write core file for process %s (error %d)\n",
1679 curproc->p_comm, error);
1683 free(tmpbuf, M_TEMP);
1684 if (params.comp != NULL)
1685 compressor_fini(params.comp);
1686 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1687 TAILQ_REMOVE(¬elst, ninfo, link);
1688 free(ninfo, M_TEMP);
1697 * A callback for each_dumpable_segment() to write out the segment's
1698 * program header entry.
1701 cb_put_phdr(vm_map_entry_t entry, void *closure)
1703 struct phdr_closure *phc = (struct phdr_closure *)closure;
1704 Elf_Phdr *phdr = phc->phdr;
1706 phc->offset = round_page(phc->offset);
1708 phdr->p_type = PT_LOAD;
1709 phdr->p_offset = phc->offset;
1710 phdr->p_vaddr = entry->start;
1712 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1713 phdr->p_align = PAGE_SIZE;
1714 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1716 phc->offset += phdr->p_filesz;
1721 * A callback for each_dumpable_segment() to gather information about
1722 * the number of segments and their total size.
1725 cb_size_segment(vm_map_entry_t entry, void *closure)
1727 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1730 ssc->size += entry->end - entry->start;
1734 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1740 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1744 * For each writable segment in the process's memory map, call the given
1745 * function with a pointer to the map entry and some arbitrary
1746 * caller-supplied data.
1749 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1752 struct proc *p = td->td_proc;
1753 vm_map_t map = &p->p_vmspace->vm_map;
1754 vm_map_entry_t entry;
1755 vm_object_t backing_object, object;
1758 vm_map_lock_read(map);
1759 VM_MAP_ENTRY_FOREACH(entry, map) {
1761 * Don't dump inaccessible mappings, deal with legacy
1764 * Note that read-only segments related to the elf binary
1765 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1766 * need to arbitrarily ignore such segments.
1768 if ((flags & SVC_ALL) == 0) {
1769 if (elf_legacy_coredump) {
1770 if ((entry->protection & VM_PROT_RW) !=
1774 if ((entry->protection & VM_PROT_ALL) == 0)
1780 * Dont include memory segment in the coredump if
1781 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1782 * madvise(2). Do not dump submaps (i.e. parts of the
1785 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1787 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1788 (flags & SVC_ALL) == 0)
1790 if ((object = entry->object.vm_object) == NULL)
1793 /* Ignore memory-mapped devices and such things. */
1794 VM_OBJECT_RLOCK(object);
1795 while ((backing_object = object->backing_object) != NULL) {
1796 VM_OBJECT_RLOCK(backing_object);
1797 VM_OBJECT_RUNLOCK(object);
1798 object = backing_object;
1800 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1801 VM_OBJECT_RUNLOCK(object);
1805 (*func)(entry, closure);
1807 vm_map_unlock_read(map);
1811 * Write the core file header to the file, including padding up to
1812 * the page boundary.
1815 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1816 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1819 struct note_info *ninfo;
1823 /* Fill in the header. */
1824 bzero(hdr, hdrsize);
1825 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1827 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1828 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1829 sbuf_start_section(sb, NULL);
1830 sbuf_bcat(sb, hdr, hdrsize);
1831 TAILQ_FOREACH(ninfo, notelst, link)
1832 __elfN(putnote)(p->td, ninfo, sb);
1833 /* Align up to a page boundary for the program segments. */
1834 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1835 error = sbuf_finish(sb);
1842 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1852 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1853 __elfN(note_prpsinfo), p);
1856 * To have the debugger select the right thread (LWP) as the initial
1857 * thread, we dump the state of the thread passed to us in td first.
1858 * This is the thread that causes the core dump and thus likely to
1859 * be the right thread one wants to have selected in the debugger.
1862 while (thr != NULL) {
1863 size += __elfN(prepare_register_notes)(td, list, thr);
1864 size += __elfN(register_note)(td, list, -1,
1865 __elfN(note_threadmd), thr);
1867 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1868 TAILQ_NEXT(thr, td_plist);
1870 thr = TAILQ_NEXT(thr, td_plist);
1873 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1874 __elfN(note_procstat_proc), p);
1875 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1876 note_procstat_files, p);
1877 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1878 note_procstat_vmmap, p);
1879 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1880 note_procstat_groups, p);
1881 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1882 note_procstat_umask, p);
1883 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1884 note_procstat_rlimit, p);
1885 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1886 note_procstat_osrel, p);
1887 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1888 __elfN(note_procstat_psstrings), p);
1889 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1890 __elfN(note_procstat_auxv), p);
1896 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1897 size_t notesz, int flags)
1902 struct phdr_closure phc;
1905 ehdr = (Elf_Ehdr *)hdr;
1906 bi = td->td_proc->p_elf_brandinfo;
1908 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1909 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1910 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1911 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1912 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1913 ehdr->e_ident[EI_DATA] = ELF_DATA;
1914 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1915 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1916 ehdr->e_ident[EI_ABIVERSION] = 0;
1917 ehdr->e_ident[EI_PAD] = 0;
1918 ehdr->e_type = ET_CORE;
1919 ehdr->e_machine = bi->machine;
1920 ehdr->e_version = EV_CURRENT;
1922 ehdr->e_phoff = sizeof(Elf_Ehdr);
1923 ehdr->e_flags = td->td_proc->p_elf_flags;
1924 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1925 ehdr->e_phentsize = sizeof(Elf_Phdr);
1926 ehdr->e_shentsize = sizeof(Elf_Shdr);
1927 ehdr->e_shstrndx = SHN_UNDEF;
1928 if (numsegs + 1 < PN_XNUM) {
1929 ehdr->e_phnum = numsegs + 1;
1932 ehdr->e_phnum = PN_XNUM;
1935 ehdr->e_shoff = ehdr->e_phoff +
1936 (numsegs + 1) * ehdr->e_phentsize;
1937 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1938 ("e_shoff: %zu, hdrsize - shdr: %zu",
1939 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1941 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1942 memset(shdr, 0, sizeof(*shdr));
1944 * A special first section is used to hold large segment and
1945 * section counts. This was proposed by Sun Microsystems in
1946 * Solaris and has been adopted by Linux; the standard ELF
1947 * tools are already familiar with the technique.
1949 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1950 * (or 12-7 depending on the version of the document) for more
1953 shdr->sh_type = SHT_NULL;
1954 shdr->sh_size = ehdr->e_shnum;
1955 shdr->sh_link = ehdr->e_shstrndx;
1956 shdr->sh_info = numsegs + 1;
1960 * Fill in the program header entries.
1962 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1964 /* The note segement. */
1965 phdr->p_type = PT_NOTE;
1966 phdr->p_offset = hdrsize;
1969 phdr->p_filesz = notesz;
1971 phdr->p_flags = PF_R;
1972 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1975 /* All the writable segments from the program. */
1977 phc.offset = round_page(hdrsize + notesz);
1978 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1982 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1983 struct regset *regset, struct thread *target_td)
1985 const struct sysentvec *sv;
1986 struct note_info *ninfo;
1987 size_t size, notesize;
1990 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1993 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1994 ninfo->type = regset->note;
1995 ninfo->regset = regset;
1996 ninfo->outarg = target_td;
1997 ninfo->outsize = size;
1998 TAILQ_INSERT_TAIL(list, ninfo, link);
2000 sv = td->td_proc->p_sysent;
2001 notesize = sizeof(Elf_Note) + /* note header */
2002 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2004 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2010 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2011 int type, outfunc_t out, void *arg)
2013 const struct sysentvec *sv;
2014 struct note_info *ninfo;
2015 size_t size, notesize;
2017 sv = td->td_proc->p_sysent;
2019 out(arg, NULL, &size);
2020 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2022 ninfo->outfunc = out;
2023 ninfo->outarg = arg;
2024 ninfo->outsize = size;
2025 TAILQ_INSERT_TAIL(list, ninfo, link);
2030 notesize = sizeof(Elf_Note) + /* note header */
2031 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2033 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2039 append_note_data(const void *src, void *dst, size_t len)
2043 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2045 bcopy(src, dst, len);
2046 bzero((char *)dst + len, padded_len - len);
2048 return (padded_len);
2052 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2060 note = (Elf_Note *)buf;
2061 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2062 note->n_descsz = size;
2063 note->n_type = type;
2064 buf += sizeof(*note);
2065 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2066 sizeof(FREEBSD_ABI_VENDOR));
2067 append_note_data(src, buf, size);
2072 notesize = sizeof(Elf_Note) + /* note header */
2073 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2075 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2081 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2084 const struct sysentvec *sv;
2085 ssize_t old_len, sect_len;
2086 size_t new_len, descsz, i;
2088 if (ninfo->type == -1) {
2089 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2093 sv = td->td_proc->p_sysent;
2095 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2096 note.n_descsz = ninfo->outsize;
2097 note.n_type = ninfo->type;
2099 sbuf_bcat(sb, ¬e, sizeof(note));
2100 sbuf_start_section(sb, &old_len);
2101 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2102 strlen(sv->sv_elf_core_abi_vendor) + 1);
2103 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2104 if (note.n_descsz == 0)
2106 sbuf_start_section(sb, &old_len);
2107 if (ninfo->regset != NULL) {
2108 struct regset *regset = ninfo->regset;
2111 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2112 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2113 sbuf_bcat(sb, buf, ninfo->outsize);
2116 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2117 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2121 new_len = (size_t)sect_len;
2122 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2123 if (new_len < descsz) {
2125 * It is expected that individual note emitters will correctly
2126 * predict their expected output size and fill up to that size
2127 * themselves, padding in a format-specific way if needed.
2128 * However, in case they don't, just do it here with zeros.
2130 for (i = 0; i < descsz - new_len; i++)
2132 } else if (new_len > descsz) {
2134 * We can't always truncate sb -- we may have drained some
2137 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2138 "read it (%zu > %zu). Since it is longer than "
2139 "expected, this coredump's notes are corrupt. THIS "
2140 "IS A BUG in the note_procstat routine for type %u.\n",
2141 __func__, (unsigned)note.n_type, new_len, descsz,
2142 (unsigned)note.n_type));
2147 * Miscellaneous note out functions.
2150 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2151 #include <compat/freebsd32/freebsd32.h>
2152 #include <compat/freebsd32/freebsd32_signal.h>
2154 typedef struct prstatus32 elf_prstatus_t;
2155 typedef struct prpsinfo32 elf_prpsinfo_t;
2156 typedef struct fpreg32 elf_prfpregset_t;
2157 typedef struct fpreg32 elf_fpregset_t;
2158 typedef struct reg32 elf_gregset_t;
2159 typedef struct thrmisc32 elf_thrmisc_t;
2160 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2161 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2162 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2163 typedef uint32_t elf_ps_strings_t;
2165 typedef prstatus_t elf_prstatus_t;
2166 typedef prpsinfo_t elf_prpsinfo_t;
2167 typedef prfpregset_t elf_prfpregset_t;
2168 typedef prfpregset_t elf_fpregset_t;
2169 typedef gregset_t elf_gregset_t;
2170 typedef thrmisc_t elf_thrmisc_t;
2171 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2172 #define ELF_KERN_PROC_MASK 0
2173 typedef struct kinfo_proc elf_kinfo_proc_t;
2174 typedef vm_offset_t elf_ps_strings_t;
2178 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2184 elf_prpsinfo_t *psinfo;
2189 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2190 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2191 psinfo->pr_version = PRPSINFO_VERSION;
2192 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2193 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2195 if (p->p_args != NULL) {
2196 len = sizeof(psinfo->pr_psargs) - 1;
2197 if (len > p->p_args->ar_length)
2198 len = p->p_args->ar_length;
2199 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2205 sbuf_new(&sbarg, psinfo->pr_psargs,
2206 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2207 error = proc_getargv(curthread, p, &sbarg);
2209 if (sbuf_finish(&sbarg) == 0)
2210 len = sbuf_len(&sbarg) - 1;
2212 len = sizeof(psinfo->pr_psargs) - 1;
2213 sbuf_delete(&sbarg);
2215 if (error || len == 0)
2216 strlcpy(psinfo->pr_psargs, p->p_comm,
2217 sizeof(psinfo->pr_psargs));
2219 KASSERT(len < sizeof(psinfo->pr_psargs),
2220 ("len is too long: %zu vs %zu", len,
2221 sizeof(psinfo->pr_psargs)));
2222 cp = psinfo->pr_psargs;
2225 cp = memchr(cp, '\0', end - cp);
2231 psinfo->pr_pid = p->p_pid;
2232 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2233 free(psinfo, M_TEMP);
2235 *sizep = sizeof(*psinfo);
2239 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2242 elf_prstatus_t *status;
2245 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2248 memset(status, 0, *sizep);
2249 status->pr_version = PRSTATUS_VERSION;
2250 status->pr_statussz = sizeof(elf_prstatus_t);
2251 status->pr_gregsetsz = sizeof(elf_gregset_t);
2252 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2253 status->pr_osreldate = osreldate;
2254 status->pr_cursig = td->td_proc->p_sig;
2255 status->pr_pid = td->td_tid;
2256 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2257 fill_regs32(td, &status->pr_reg);
2259 fill_regs(td, &status->pr_reg);
2262 *sizep = sizeof(*status);
2267 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2270 elf_prstatus_t *status;
2272 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2274 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2275 set_regs32(td, &status->pr_reg);
2277 set_regs(td, &status->pr_reg);
2282 static struct regset __elfN(regset_prstatus) = {
2283 .note = NT_PRSTATUS,
2284 .size = sizeof(elf_prstatus_t),
2285 .get = __elfN(get_prstatus),
2286 .set = __elfN(set_prstatus),
2288 ELF_REGSET(__elfN(regset_prstatus));
2291 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2294 elf_prfpregset_t *fpregset;
2297 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2300 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2301 fill_fpregs32(td, fpregset);
2303 fill_fpregs(td, fpregset);
2306 *sizep = sizeof(*fpregset);
2311 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2314 elf_prfpregset_t *fpregset;
2317 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2318 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2319 set_fpregs32(td, fpregset);
2321 set_fpregs(td, fpregset);
2326 static struct regset __elfN(regset_fpregset) = {
2327 .note = NT_FPREGSET,
2328 .size = sizeof(elf_prfpregset_t),
2329 .get = __elfN(get_fpregset),
2330 .set = __elfN(set_fpregset),
2332 ELF_REGSET(__elfN(regset_fpregset));
2335 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2338 elf_thrmisc_t *thrmisc;
2341 KASSERT(*sizep == sizeof(*thrmisc),
2342 ("%s: invalid size", __func__));
2344 bzero(thrmisc, sizeof(*thrmisc));
2345 strcpy(thrmisc->pr_tname, td->td_name);
2347 *sizep = sizeof(*thrmisc);
2351 static struct regset __elfN(regset_thrmisc) = {
2353 .size = sizeof(elf_thrmisc_t),
2354 .get = __elfN(get_thrmisc),
2356 ELF_REGSET(__elfN(regset_thrmisc));
2359 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2366 size = sizeof(structsize) + sizeof(pl);
2368 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2369 structsize = sizeof(pl);
2370 memcpy(buf, &structsize, sizeof(structsize));
2371 bzero(&pl, sizeof(pl));
2372 pl.pl_lwpid = td->td_tid;
2373 pl.pl_event = PL_EVENT_NONE;
2374 pl.pl_sigmask = td->td_sigmask;
2375 pl.pl_siglist = td->td_siglist;
2376 if (td->td_si.si_signo != 0) {
2377 pl.pl_event = PL_EVENT_SIGNAL;
2378 pl.pl_flags |= PL_FLAG_SI;
2379 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2380 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2382 pl.pl_siginfo = td->td_si;
2385 strcpy(pl.pl_tdname, td->td_name);
2386 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2387 memcpy((int *)buf + 1, &pl, sizeof(pl));
2393 static struct regset __elfN(regset_lwpinfo) = {
2394 .note = NT_PTLWPINFO,
2395 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2396 .get = __elfN(get_lwpinfo),
2398 ELF_REGSET(__elfN(regset_lwpinfo));
2401 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2402 struct thread *target_td)
2404 struct sysentvec *sv = td->td_proc->p_sysent;
2405 struct regset **regsetp, **regset_end, *regset;
2410 /* NT_PRSTATUS must be the first register set note. */
2411 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2414 regsetp = sv->sv_regset_begin;
2415 if (regsetp == NULL) {
2416 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2417 size += __elfN(register_regset_note)(td, list,
2418 &__elfN(regset_fpregset), target_td);
2421 regset_end = sv->sv_regset_end;
2422 MPASS(regset_end != NULL);
2423 for (; regsetp < regset_end; regsetp++) {
2425 if (regset->note == NT_PRSTATUS)
2427 size += __elfN(register_regset_note)(td, list, regset,
2434 * Allow for MD specific notes, as well as any MD
2435 * specific preparations for writing MI notes.
2438 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2444 td = (struct thread *)arg;
2446 if (size != 0 && sb != NULL)
2447 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2451 __elfN(dump_thread)(td, buf, &size);
2452 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2453 if (size != 0 && sb != NULL)
2454 sbuf_bcat(sb, buf, size);
2459 #ifdef KINFO_PROC_SIZE
2460 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2464 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2471 size = sizeof(structsize) + p->p_numthreads *
2472 sizeof(elf_kinfo_proc_t);
2475 KASSERT(*sizep == size, ("invalid size"));
2476 structsize = sizeof(elf_kinfo_proc_t);
2477 sbuf_bcat(sb, &structsize, sizeof(structsize));
2478 sx_slock(&proctree_lock);
2480 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2481 sx_sunlock(&proctree_lock);
2486 #ifdef KINFO_FILE_SIZE
2487 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2491 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2494 size_t size, sect_sz, i;
2495 ssize_t start_len, sect_len;
2496 int structsize, filedesc_flags;
2498 if (coredump_pack_fileinfo)
2499 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2504 structsize = sizeof(struct kinfo_file);
2507 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2508 sbuf_set_drain(sb, sbuf_count_drain, &size);
2509 sbuf_bcat(sb, &structsize, sizeof(structsize));
2511 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2516 sbuf_start_section(sb, &start_len);
2518 sbuf_bcat(sb, &structsize, sizeof(structsize));
2520 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2523 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2528 KASSERT(sect_sz <= *sizep,
2529 ("kern_proc_filedesc_out did not respect maxlen; "
2530 "requested %zu, got %zu", *sizep - sizeof(structsize),
2531 sect_sz - sizeof(structsize)));
2533 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2538 #ifdef KINFO_VMENTRY_SIZE
2539 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2543 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2547 int structsize, vmmap_flags;
2549 if (coredump_pack_vmmapinfo)
2550 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2555 structsize = sizeof(struct kinfo_vmentry);
2558 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2559 sbuf_set_drain(sb, sbuf_count_drain, &size);
2560 sbuf_bcat(sb, &structsize, sizeof(structsize));
2562 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2567 sbuf_bcat(sb, &structsize, sizeof(structsize));
2569 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2575 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2582 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2584 KASSERT(*sizep == size, ("invalid size"));
2585 structsize = sizeof(gid_t);
2586 sbuf_bcat(sb, &structsize, sizeof(structsize));
2587 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2594 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2601 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2603 KASSERT(*sizep == size, ("invalid size"));
2604 structsize = sizeof(p->p_pd->pd_cmask);
2605 sbuf_bcat(sb, &structsize, sizeof(structsize));
2606 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2612 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2615 struct rlimit rlim[RLIM_NLIMITS];
2620 size = sizeof(structsize) + sizeof(rlim);
2622 KASSERT(*sizep == size, ("invalid size"));
2623 structsize = sizeof(rlim);
2624 sbuf_bcat(sb, &structsize, sizeof(structsize));
2626 for (i = 0; i < RLIM_NLIMITS; i++)
2627 lim_rlimit_proc(p, i, &rlim[i]);
2629 sbuf_bcat(sb, rlim, sizeof(rlim));
2635 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2642 size = sizeof(structsize) + sizeof(p->p_osrel);
2644 KASSERT(*sizep == size, ("invalid size"));
2645 structsize = sizeof(p->p_osrel);
2646 sbuf_bcat(sb, &structsize, sizeof(structsize));
2647 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2653 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2656 elf_ps_strings_t ps_strings;
2661 size = sizeof(structsize) + sizeof(ps_strings);
2663 KASSERT(*sizep == size, ("invalid size"));
2664 structsize = sizeof(ps_strings);
2665 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2666 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2668 ps_strings = PROC_PS_STRINGS(p);
2670 sbuf_bcat(sb, &structsize, sizeof(structsize));
2671 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2677 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2686 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2688 sbuf_set_drain(sb, sbuf_count_drain, &size);
2689 sbuf_bcat(sb, &structsize, sizeof(structsize));
2691 proc_getauxv(curthread, p, sb);
2697 structsize = sizeof(Elf_Auxinfo);
2698 sbuf_bcat(sb, &structsize, sizeof(structsize));
2700 proc_getauxv(curthread, p, sb);
2706 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2707 const char *note_vendor, const Elf_Phdr *pnote,
2708 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2710 const Elf_Note *note, *note0, *note_end;
2711 const char *note_name;
2716 /* We need some limit, might as well use PAGE_SIZE. */
2717 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2719 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2720 if (pnote->p_offset > PAGE_SIZE ||
2721 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2722 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2724 VOP_UNLOCK(imgp->vp);
2725 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2726 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2728 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2729 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2730 curthread->td_ucred, NOCRED, NULL, curthread);
2732 uprintf("i/o error PT_NOTE\n");
2735 note = note0 = (const Elf_Note *)buf;
2736 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2738 note = note0 = (const Elf_Note *)(imgp->image_header +
2740 note_end = (const Elf_Note *)(imgp->image_header +
2741 pnote->p_offset + pnote->p_filesz);
2744 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2745 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2746 (const char *)note < sizeof(Elf_Note)) {
2749 if (note->n_namesz != checknote->n_namesz ||
2750 note->n_descsz != checknote->n_descsz ||
2751 note->n_type != checknote->n_type)
2753 note_name = (const char *)(note + 1);
2754 if (note_name + checknote->n_namesz >=
2755 (const char *)note_end || strncmp(note_vendor,
2756 note_name, checknote->n_namesz) != 0)
2759 if (cb(note, cb_arg, &res))
2762 note = (const Elf_Note *)((const char *)(note + 1) +
2763 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2764 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2773 struct brandnote_cb_arg {
2774 Elf_Brandnote *brandnote;
2779 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2781 struct brandnote_cb_arg *arg;
2786 * Fetch the osreldate for binary from the ELF OSABI-note if
2789 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2790 arg->brandnote->trans_osrel != NULL ?
2791 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2796 static Elf_Note fctl_note = {
2797 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2798 .n_descsz = sizeof(uint32_t),
2799 .n_type = NT_FREEBSD_FEATURE_CTL,
2802 struct fctl_cb_arg {
2808 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2810 struct fctl_cb_arg *arg;
2811 const Elf32_Word *desc;
2815 p = (uintptr_t)(note + 1);
2816 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2817 desc = (const Elf32_Word *)p;
2818 *arg->has_fctl0 = true;
2819 *arg->fctl0 = desc[0];
2825 * Try to find the appropriate ABI-note section for checknote, fetch
2826 * the osreldate and feature control flags for binary from the ELF
2827 * OSABI-note. Only the first page of the image is searched, the same
2831 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2832 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2834 const Elf_Phdr *phdr;
2835 const Elf_Ehdr *hdr;
2836 struct brandnote_cb_arg b_arg;
2837 struct fctl_cb_arg f_arg;
2840 hdr = (const Elf_Ehdr *)imgp->image_header;
2841 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2842 b_arg.brandnote = brandnote;
2843 b_arg.osrel = osrel;
2844 f_arg.has_fctl0 = has_fctl0;
2845 f_arg.fctl0 = fctl0;
2847 for (i = 0; i < hdr->e_phnum; i++) {
2848 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2849 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2851 for (j = 0; j < hdr->e_phnum; j++) {
2852 if (phdr[j].p_type == PT_NOTE &&
2853 __elfN(parse_notes)(imgp, &fctl_note,
2854 FREEBSD_ABI_VENDOR, &phdr[j],
2855 note_fctl_cb, &f_arg))
2866 * Tell kern_execve.c about it, with a little help from the linker.
2868 static struct execsw __elfN(execsw) = {
2869 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2870 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2872 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2875 __elfN(trans_prot)(Elf_Word flags)
2881 prot |= VM_PROT_EXECUTE;
2883 prot |= VM_PROT_WRITE;
2885 prot |= VM_PROT_READ;
2886 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2887 if (i386_read_exec && (flags & PF_R))
2888 prot |= VM_PROT_EXECUTE;
2894 __elfN(untrans_prot)(vm_prot_t prot)
2899 if (prot & VM_PROT_EXECUTE)
2901 if (prot & VM_PROT_READ)
2903 if (prot & VM_PROT_WRITE)