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 "opt_capsicum.h"
36 #include <sys/param.h>
37 #include <sys/capsicum.h>
38 #include <sys/compressor.h>
40 #include <sys/fcntl.h>
41 #include <sys/imgact.h>
42 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
49 #include <sys/namei.h>
51 #include <sys/procfs.h>
52 #include <sys/ptrace.h>
53 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
58 #include <sys/sf_buf.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/syslog.h>
69 #include <sys/eventhandler.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_extern.h>
80 #include <machine/elf.h>
81 #include <machine/md_var.h>
83 #define ELF_NOTE_ROUNDSIZE 4
84 #define OLD_EI_BRAND 8
86 static int __elfN(check_header)(const Elf_Ehdr *hdr);
87 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
88 const char *interp, int32_t *osrel, uint32_t *fctl0);
89 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
92 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
93 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
94 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
96 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
97 static bool __elfN(check_note)(struct image_params *imgp,
98 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
100 static vm_prot_t __elfN(trans_prot)(Elf_Word);
101 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
102 static size_t __elfN(prepare_register_notes)(struct thread *td,
103 struct note_info_list *list, struct thread *target_td);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
106 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 #if defined(__amd64__)
132 static int __elfN(vdso) = 1;
133 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
134 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
135 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
137 static int __elfN(vdso) = 0;
140 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
141 int i386_read_exec = 0;
142 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
143 "enable execution from readable segments");
146 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
148 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
153 val = __elfN(pie_base);
154 error = sysctl_handle_long(oidp, &val, 0, req);
155 if (error != 0 || req->newptr == NULL)
157 if ((val & PAGE_MASK) != 0)
159 __elfN(pie_base) = val;
162 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
163 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
164 sysctl_pie_base, "LU",
165 "PIE load base without randomization");
167 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
173 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures
174 * have limited address space (which can cause issues for applications with
175 * high memory use) so we leave it off there.
177 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
178 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
179 &__elfN(aslr_enabled), 0,
180 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
181 ": enable address map randomization");
184 * Enable ASLR by default for 64-bit PIE binaries.
186 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
187 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
188 &__elfN(pie_aslr_enabled), 0,
189 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
190 ": enable address map randomization for PIE binaries");
193 * Sbrk is deprecated and it can be assumed that in most cases it will not be
194 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR
195 * to use the bss grow region.
197 static int __elfN(aslr_honor_sbrk) = 0;
198 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
199 &__elfN(aslr_honor_sbrk), 0,
200 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
202 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64;
203 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
204 &__elfN(aslr_stack), 0,
205 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
206 ": enable stack address randomization");
208 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64;
209 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN,
210 &__elfN(aslr_shared_page), 0,
211 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
212 ": enable shared page address randomization");
214 static int __elfN(sigfastblock) = 1;
215 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
216 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
217 "enable sigfastblock for new processes");
219 static bool __elfN(allow_wx) = true;
220 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
221 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
222 "Allow pages to be mapped simultaneously writable and executable");
224 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
226 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
228 Elf_Brandnote __elfN(freebsd_brandnote) = {
229 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
230 .hdr.n_descsz = sizeof(int32_t),
231 .hdr.n_type = NT_FREEBSD_ABI_TAG,
232 .vendor = FREEBSD_ABI_VENDOR,
233 .flags = BN_TRANSLATE_OSREL,
234 .trans_osrel = __elfN(freebsd_trans_osrel)
238 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
242 p = (uintptr_t)(note + 1);
243 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
244 *osrel = *(const int32_t *)(p);
249 static int GNU_KFREEBSD_ABI_DESC = 3;
251 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
252 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
253 .hdr.n_descsz = 16, /* XXX at least 16 */
255 .vendor = GNU_ABI_VENDOR,
256 .flags = BN_TRANSLATE_OSREL,
257 .trans_osrel = kfreebsd_trans_osrel
261 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
263 const Elf32_Word *desc;
266 p = (uintptr_t)(note + 1);
267 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
269 desc = (const Elf32_Word *)p;
270 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
274 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
275 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
277 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
283 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
287 for (i = 0; i < MAX_BRANDS; i++) {
288 if (elf_brand_list[i] == NULL) {
289 elf_brand_list[i] = entry;
293 if (i == MAX_BRANDS) {
294 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
302 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
306 for (i = 0; i < MAX_BRANDS; i++) {
307 if (elf_brand_list[i] == entry) {
308 elf_brand_list[i] = NULL;
318 __elfN(brand_inuse)(Elf_Brandinfo *entry)
323 sx_slock(&allproc_lock);
324 FOREACH_PROC_IN_SYSTEM(p) {
325 if (p->p_sysent == entry->sysvec) {
330 sx_sunlock(&allproc_lock);
335 static Elf_Brandinfo *
336 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
337 int32_t *osrel, uint32_t *fctl0)
339 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
340 Elf_Brandinfo *bi, *bi_m;
342 int i, interp_name_len;
344 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
347 * We support four types of branding -- (1) the ELF EI_OSABI field
348 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
349 * branding w/in the ELF header, (3) path of the `interp_path'
350 * field, and (4) the ".note.ABI-tag" ELF section.
353 /* Look for an ".note.ABI-tag" ELF section */
355 for (i = 0; i < MAX_BRANDS; i++) {
356 bi = elf_brand_list[i];
359 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
361 if (hdr->e_machine == bi->machine && (bi->flags &
362 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
366 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
368 /* Give brand a chance to veto check_note's guess */
369 if (ret && bi->header_supported) {
370 ret = bi->header_supported(imgp, osrel,
371 has_fctl0 ? fctl0 : NULL);
374 * If note checker claimed the binary, but the
375 * interpreter path in the image does not
376 * match default one for the brand, try to
377 * search for other brands with the same
378 * interpreter. Either there is better brand
379 * with the right interpreter, or, failing
380 * this, we return first brand which accepted
381 * our note and, optionally, header.
383 if (ret && bi_m == NULL && interp != NULL &&
384 (bi->interp_path == NULL ||
385 (strlen(bi->interp_path) + 1 != interp_name_len ||
386 strncmp(interp, bi->interp_path, interp_name_len)
398 /* If the executable has a brand, search for it in the brand list. */
399 for (i = 0; i < MAX_BRANDS; i++) {
400 bi = elf_brand_list[i];
401 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
402 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
404 if (hdr->e_machine == bi->machine &&
405 (hdr->e_ident[EI_OSABI] == bi->brand ||
406 (bi->compat_3_brand != NULL &&
407 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
408 bi->compat_3_brand) == 0))) {
409 /* Looks good, but give brand a chance to veto */
410 if (bi->header_supported == NULL ||
411 bi->header_supported(imgp, NULL, NULL)) {
413 * Again, prefer strictly matching
416 if (interp_name_len == 0 &&
417 bi->interp_path == NULL)
419 if (bi->interp_path != NULL &&
420 strlen(bi->interp_path) + 1 ==
421 interp_name_len && strncmp(interp,
422 bi->interp_path, interp_name_len) == 0)
432 /* No known brand, see if the header is recognized by any brand */
433 for (i = 0; i < MAX_BRANDS; i++) {
434 bi = elf_brand_list[i];
435 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
436 bi->header_supported == NULL)
438 if (hdr->e_machine == bi->machine) {
439 ret = bi->header_supported(imgp, NULL, NULL);
445 /* Lacking a known brand, search for a recognized interpreter. */
446 if (interp != NULL) {
447 for (i = 0; i < MAX_BRANDS; i++) {
448 bi = elf_brand_list[i];
449 if (bi == NULL || (bi->flags &
450 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
453 if (hdr->e_machine == bi->machine &&
454 bi->interp_path != NULL &&
455 /* ELF image p_filesz includes terminating zero */
456 strlen(bi->interp_path) + 1 == interp_name_len &&
457 strncmp(interp, bi->interp_path, interp_name_len)
458 == 0 && (bi->header_supported == NULL ||
459 bi->header_supported(imgp, NULL, NULL)))
464 /* Lacking a recognized interpreter, try the default brand */
465 for (i = 0; i < MAX_BRANDS; i++) {
466 bi = elf_brand_list[i];
467 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
468 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
470 if (hdr->e_machine == bi->machine &&
471 __elfN(fallback_brand) == bi->brand &&
472 (bi->header_supported == NULL ||
473 bi->header_supported(imgp, NULL, NULL)))
480 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
482 return (hdr->e_phoff <= PAGE_SIZE &&
483 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
487 __elfN(check_header)(const Elf_Ehdr *hdr)
493 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
494 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
495 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
496 hdr->e_phentsize != sizeof(Elf_Phdr) ||
497 hdr->e_version != ELF_TARG_VER)
501 * Make sure we have at least one brand for this machine.
504 for (i = 0; i < MAX_BRANDS; i++) {
505 bi = elf_brand_list[i];
506 if (bi != NULL && bi->machine == hdr->e_machine)
516 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
517 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
524 * Create the page if it doesn't exist yet. Ignore errors.
526 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
527 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
530 * Find the page from the underlying object.
532 if (object != NULL) {
533 sf = vm_imgact_map_page(object, offset);
535 return (KERN_FAILURE);
536 off = offset - trunc_page(offset);
537 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
539 vm_imgact_unmap_page(sf);
541 return (KERN_FAILURE);
544 return (KERN_SUCCESS);
548 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
549 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
555 int error, locked, rv;
557 if (start != trunc_page(start)) {
558 rv = __elfN(map_partial)(map, object, offset, start,
559 round_page(start), prot);
560 if (rv != KERN_SUCCESS)
562 offset += round_page(start) - start;
563 start = round_page(start);
565 if (end != round_page(end)) {
566 rv = __elfN(map_partial)(map, object, offset +
567 trunc_page(end) - start, trunc_page(end), end, prot);
568 if (rv != KERN_SUCCESS)
570 end = trunc_page(end);
573 return (KERN_SUCCESS);
574 if ((offset & PAGE_MASK) != 0) {
576 * The mapping is not page aligned. This means that we have
579 rv = vm_map_fixed(map, NULL, 0, start, end - start,
580 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
581 if (rv != KERN_SUCCESS)
584 return (KERN_SUCCESS);
585 for (; start < end; start += sz) {
586 sf = vm_imgact_map_page(object, offset);
588 return (KERN_FAILURE);
589 off = offset - trunc_page(offset);
591 if (sz > PAGE_SIZE - off)
592 sz = PAGE_SIZE - off;
593 error = copyout((caddr_t)sf_buf_kva(sf) + off,
595 vm_imgact_unmap_page(sf);
597 return (KERN_FAILURE);
601 vm_object_reference(object);
602 rv = vm_map_fixed(map, object, offset, start, end - start,
603 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
604 (object != NULL ? MAP_VN_EXEC : 0));
605 if (rv != KERN_SUCCESS) {
606 locked = VOP_ISLOCKED(imgp->vp);
607 VOP_UNLOCK(imgp->vp);
608 vm_object_deallocate(object);
609 vn_lock(imgp->vp, locked | LK_RETRY);
611 } else if (object != NULL) {
612 MPASS(imgp->vp->v_object == object);
613 VOP_SET_TEXT_CHECKED(imgp->vp);
616 return (KERN_SUCCESS);
620 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
621 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
627 vm_offset_t map_addr;
630 vm_ooffset_t file_addr;
633 * It's necessary to fail if the filsz + offset taken from the
634 * header is greater than the actual file pager object's size.
635 * If we were to allow this, then the vm_map_find() below would
636 * walk right off the end of the file object and into the ether.
638 * While I'm here, might as well check for something else that
639 * is invalid: filsz cannot be greater than memsz.
641 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
643 uprintf("elf_load_section: truncated ELF file\n");
647 object = imgp->object;
648 map = &imgp->proc->p_vmspace->vm_map;
649 map_addr = trunc_page((vm_offset_t)vmaddr);
650 file_addr = trunc_page(offset);
653 * We have two choices. We can either clear the data in the last page
654 * of an oversized mapping, or we can start the anon mapping a page
655 * early and copy the initialized data into that first page. We
660 else if (memsz > filsz)
661 map_len = trunc_page(offset + filsz) - file_addr;
663 map_len = round_page(offset + filsz) - file_addr;
666 /* cow flags: don't dump readonly sections in core */
667 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
668 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
670 rv = __elfN(map_insert)(imgp, map, object, file_addr,
671 map_addr, map_addr + map_len, prot, cow);
672 if (rv != KERN_SUCCESS)
675 /* we can stop now if we've covered it all */
681 * We have to get the remaining bit of the file into the first part
682 * of the oversized map segment. This is normally because the .data
683 * segment in the file is extended to provide bss. It's a neat idea
684 * to try and save a page, but it's a pain in the behind to implement.
686 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
688 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
689 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
691 /* This had damn well better be true! */
693 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
694 map_addr + map_len, prot, 0);
695 if (rv != KERN_SUCCESS)
700 sf = vm_imgact_map_page(object, offset + filsz);
704 /* send the page fragment to user space */
705 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
707 vm_imgact_unmap_page(sf);
713 * Remove write access to the page if it was only granted by map_insert
716 if ((prot & VM_PROT_WRITE) == 0)
717 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
718 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
724 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
725 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
732 ASSERT_VOP_LOCKED(imgp->vp, __func__);
737 for (i = 0; i < hdr->e_phnum; i++) {
738 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
741 /* Loadable segment */
742 prot = __elfN(trans_prot)(phdr[i].p_flags);
743 error = __elfN(load_section)(imgp, phdr[i].p_offset,
744 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
745 phdr[i].p_memsz, phdr[i].p_filesz, prot);
750 * Establish the base address if this is the first segment.
753 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
758 if (base_addrp != NULL)
759 *base_addrp = base_addr;
765 * Load the file "file" into memory. It may be either a shared object
768 * The "addr" reference parameter is in/out. On entry, it specifies
769 * the address where a shared object should be loaded. If the file is
770 * an executable, this value is ignored. On exit, "addr" specifies
771 * where the file was actually loaded.
773 * The "entry" reference parameter is out only. On exit, it specifies
774 * the entry point for the loaded file.
777 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
783 struct image_params image_params;
785 const Elf_Ehdr *hdr = NULL;
786 const Elf_Phdr *phdr = NULL;
787 struct nameidata *nd;
789 struct image_params *imgp;
791 u_long base_addr = 0;
794 #ifdef CAPABILITY_MODE
796 * XXXJA: This check can go away once we are sufficiently confident
797 * that the checks in namei() are correct.
799 if (IN_CAPABILITY_MODE(curthread))
803 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
805 attr = &tempdata->attr;
806 imgp = &tempdata->image_params;
809 * Initialize part of the common data
814 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
816 if ((error = namei(nd)) != 0) {
821 imgp->vp = nd->ni_vp;
824 * Check permissions, modes, uid, etc on the file, and "open" it.
826 error = exec_check_permissions(imgp);
830 error = exec_map_first_page(imgp);
834 imgp->object = nd->ni_vp->v_object;
836 hdr = (const Elf_Ehdr *)imgp->image_header;
837 if ((error = __elfN(check_header)(hdr)) != 0)
839 if (hdr->e_type == ET_DYN)
841 else if (hdr->e_type == ET_EXEC)
848 /* Only support headers that fit within first page for now */
849 if (!__elfN(phdr_in_zero_page)(hdr)) {
854 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
855 if (!aligned(phdr, Elf_Addr)) {
860 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
864 if (p->p_sysent->sv_protect != NULL)
865 p->p_sysent->sv_protect(imgp, SVP_INTERP);
868 *entry = (unsigned long)hdr->e_entry + rbase;
872 exec_unmap_first_page(imgp);
876 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
879 free(tempdata, M_TEMP);
885 * Select randomized valid address in the map map, between minv and
886 * maxv, with specified alignment. The [minv, maxv) range must belong
887 * to the map. Note that function only allocates the address, it is
888 * up to caller to clamp maxv in a way that the final allocation
889 * length fit into the map.
891 * Result is returned in *resp, error code indicates that arguments
892 * did not pass sanity checks for overflow and range correctness.
895 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
896 u_int align, u_long *resp)
900 MPASS(vm_map_min(map) <= minv);
902 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
903 uprintf("Invalid ELF segments layout\n");
907 arc4rand(&rbase, sizeof(rbase), 0);
908 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
909 res &= ~((u_long)align - 1);
914 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
915 res, minv, maxv, rbase));
917 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
918 res, maxv, minv, rbase));
925 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
926 const Elf_Phdr *phdr)
928 struct vmspace *vmspace;
930 u_long text_size, data_size, total_size, text_addr, data_addr;
931 u_long seg_size, seg_addr;
935 text_size = data_size = total_size = text_addr = data_addr = 0;
937 for (i = 0; i < hdr->e_phnum; i++) {
938 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
941 seg_addr = trunc_page(phdr[i].p_vaddr + imgp->et_dyn_addr);
942 seg_size = round_page(phdr[i].p_memsz +
943 phdr[i].p_vaddr + imgp->et_dyn_addr - seg_addr);
946 * Make the largest executable segment the official
947 * text segment and all others data.
949 * Note that obreak() assumes that data_addr + data_size == end
950 * of data load area, and the ELF file format expects segments
951 * to be sorted by address. If multiple data segments exist,
952 * the last one will be used.
955 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
956 text_size = seg_size;
957 text_addr = seg_addr;
959 data_size = seg_size;
960 data_addr = seg_addr;
962 total_size += seg_size;
965 if (data_addr == 0 && data_size == 0) {
966 data_addr = text_addr;
967 data_size = text_size;
971 * Check limits. It should be safe to check the
972 * limits after loading the segments since we do
973 * not actually fault in all the segments pages.
975 PROC_LOCK(imgp->proc);
976 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
977 err_str = "Data segment size exceeds process limit";
978 else if (text_size > maxtsiz)
979 err_str = "Text segment size exceeds system limit";
980 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
981 err_str = "Total segment size exceeds process limit";
982 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
983 err_str = "Data segment size exceeds resource limit";
984 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
985 err_str = "Total segment size exceeds resource limit";
986 PROC_UNLOCK(imgp->proc);
987 if (err_str != NULL) {
988 uprintf("%s\n", err_str);
992 vmspace = imgp->proc->p_vmspace;
993 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
994 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
995 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
996 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1002 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
1003 char **interpp, bool *free_interpp)
1007 int error, interp_name_len;
1009 KASSERT(phdr->p_type == PT_INTERP,
1010 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1011 ASSERT_VOP_LOCKED(imgp->vp, __func__);
1015 /* Path to interpreter */
1016 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1017 uprintf("Invalid PT_INTERP\n");
1021 interp_name_len = phdr->p_filesz;
1022 if (phdr->p_offset > PAGE_SIZE ||
1023 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1025 * The vnode lock might be needed by the pagedaemon to
1026 * clean pages owned by the vnode. Do not allow sleep
1027 * waiting for memory with the vnode locked, instead
1028 * try non-sleepable allocation first, and if it
1029 * fails, go to the slow path were we drop the lock
1030 * and do M_WAITOK. A text reference prevents
1031 * modifications to the vnode content.
1033 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1034 if (interp == NULL) {
1035 VOP_UNLOCK(imgp->vp);
1036 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1037 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1040 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1041 interp_name_len, phdr->p_offset,
1042 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1045 free(interp, M_TEMP);
1046 uprintf("i/o error PT_INTERP %d\n", error);
1049 interp[interp_name_len] = '\0';
1052 *free_interpp = true;
1056 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1057 if (interp[interp_name_len - 1] != '\0') {
1058 uprintf("Invalid PT_INTERP\n");
1063 *free_interpp = false;
1068 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1069 const char *interp, u_long *addr, u_long *entry)
1073 if (brand_info->interp_newpath != NULL &&
1074 (brand_info->interp_path == NULL ||
1075 strcmp(interp, brand_info->interp_path) == 0)) {
1076 error = __elfN(load_file)(imgp->proc,
1077 brand_info->interp_newpath, addr, entry);
1082 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1086 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1091 * Impossible et_dyn_addr initial value indicating that the real base
1092 * must be calculated later with some randomization applied.
1094 #define ET_DYN_ADDR_RAND 1
1097 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1100 const Elf_Ehdr *hdr;
1101 const Elf_Phdr *phdr;
1102 Elf_Auxargs *elf_auxargs;
1103 struct vmspace *vmspace;
1106 Elf_Brandinfo *brand_info;
1107 struct sysentvec *sv;
1108 u_long addr, baddr, entry, proghdr;
1109 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1115 hdr = (const Elf_Ehdr *)imgp->image_header;
1118 * Do we have a valid ELF header ?
1120 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1121 * if particular brand doesn't support it.
1123 if (__elfN(check_header)(hdr) != 0 ||
1124 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1128 * From here on down, we return an errno, not -1, as we've
1129 * detected an ELF file.
1132 if (!__elfN(phdr_in_zero_page)(hdr)) {
1133 uprintf("Program headers not in the first page\n");
1136 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1137 if (!aligned(phdr, Elf_Addr)) {
1138 uprintf("Unaligned program headers\n");
1146 entry = proghdr = 0;
1148 free_interp = false;
1152 * Somewhat arbitrary, limit accepted max alignment for the
1153 * loadable segment to the max supported superpage size. Too
1154 * large alignment requests are not useful and are indicators
1155 * of corrupted or outright malicious binary.
1157 maxalign = PAGE_SIZE;
1158 maxsalign = PAGE_SIZE * 1024;
1159 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1160 if (pagesizes[i] > maxsalign)
1161 maxsalign = pagesizes[i];
1166 for (i = 0; i < hdr->e_phnum; i++) {
1167 switch (phdr[i].p_type) {
1170 baddr = phdr[i].p_vaddr;
1171 if (!powerof2(phdr[i].p_align) ||
1172 phdr[i].p_align > maxsalign) {
1173 uprintf("Invalid segment alignment\n");
1177 if (phdr[i].p_align > maxalign)
1178 maxalign = phdr[i].p_align;
1179 if (mapsz + phdr[i].p_memsz < mapsz) {
1180 uprintf("Mapsize overflow\n");
1184 mapsz += phdr[i].p_memsz;
1188 * If this segment contains the program headers,
1189 * remember their virtual address for the AT_PHDR
1190 * aux entry. Static binaries don't usually include
1193 if (phdr[i].p_offset == 0 &&
1194 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1196 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1199 /* Path to interpreter */
1200 if (interp != NULL) {
1201 uprintf("Multiple PT_INTERP headers\n");
1205 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1211 if (__elfN(nxstack)) {
1213 __elfN(trans_prot)(phdr[i].p_flags);
1214 if ((imgp->stack_prot & VM_PROT_RW) !=
1216 uprintf("Invalid PT_GNU_STACK\n");
1221 imgp->stack_sz = phdr[i].p_memsz;
1223 case PT_PHDR: /* Program header table info */
1224 proghdr = phdr[i].p_vaddr;
1229 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1230 if (brand_info == NULL) {
1231 uprintf("ELF binary type \"%u\" not known.\n",
1232 hdr->e_ident[EI_OSABI]);
1236 sv = brand_info->sysvec;
1237 if (hdr->e_type == ET_DYN) {
1238 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1239 uprintf("Cannot execute shared object\n");
1244 * Honour the base load address from the dso if it is
1245 * non-zero for some reason.
1248 if ((sv->sv_flags & SV_ASLR) == 0 ||
1249 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1250 imgp->et_dyn_addr = __elfN(pie_base);
1251 else if ((__elfN(pie_aslr_enabled) &&
1252 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1253 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1254 imgp->et_dyn_addr = ET_DYN_ADDR_RAND;
1256 imgp->et_dyn_addr = __elfN(pie_base);
1261 * Avoid a possible deadlock if the current address space is destroyed
1262 * and that address space maps the locked vnode. In the common case,
1263 * the locked vnode's v_usecount is decremented but remains greater
1264 * than zero. Consequently, the vnode lock is not needed by vrele().
1265 * However, in cases where the vnode lock is external, such as nullfs,
1266 * v_usecount may become zero.
1268 * The VV_TEXT flag prevents modifications to the executable while
1269 * the vnode is unlocked.
1271 VOP_UNLOCK(imgp->vp);
1274 * Decide whether to enable randomization of user mappings.
1275 * First, reset user preferences for the setid binaries.
1276 * Then, account for the support of the randomization by the
1277 * ABI, by user preferences, and make special treatment for
1280 if (imgp->credential_setid) {
1281 PROC_LOCK(imgp->proc);
1282 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1283 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1284 PROC_UNLOCK(imgp->proc);
1286 if ((sv->sv_flags & SV_ASLR) == 0 ||
1287 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1288 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1289 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND,
1290 ("imgp->et_dyn_addr == RAND and !ASLR"));
1291 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1292 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1293 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) {
1294 imgp->map_flags |= MAP_ASLR;
1296 * If user does not care about sbrk, utilize the bss
1297 * grow region for mappings as well. We can select
1298 * the base for the image anywere and still not suffer
1299 * from the fragmentation.
1301 if (!__elfN(aslr_honor_sbrk) ||
1302 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1303 imgp->map_flags |= MAP_ASLR_IGNSTART;
1304 if (__elfN(aslr_stack))
1305 imgp->map_flags |= MAP_ASLR_STACK;
1306 if (__elfN(aslr_shared_page))
1307 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE;
1310 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1311 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1312 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1313 imgp->map_flags |= MAP_WXORX;
1315 error = exec_new_vmspace(imgp, sv);
1317 imgp->proc->p_sysent = sv;
1318 imgp->proc->p_elf_brandinfo = brand_info;
1320 vmspace = imgp->proc->p_vmspace;
1321 map = &vmspace->vm_map;
1322 maxv = sv->sv_usrstack;
1323 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1324 maxv -= lim_max(td, RLIMIT_STACK);
1325 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1326 uprintf("Excessive mapping size\n");
1330 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) {
1331 KASSERT((map->flags & MAP_ASLR) != 0,
1332 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1333 error = __CONCAT(rnd_, __elfN(base))(map,
1334 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1335 /* reserve half of the address space to interpreter */
1336 maxv / 2, maxalign, &imgp->et_dyn_addr);
1339 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1343 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL);
1347 error = __elfN(enforce_limits)(imgp, hdr, phdr);
1352 * We load the dynamic linker where a userland call
1353 * to mmap(0, ...) would put it. The rationale behind this
1354 * calculation is that it leaves room for the heap to grow to
1355 * its maximum allowed size.
1357 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1359 if ((map->flags & MAP_ASLR) != 0) {
1360 maxv1 = maxv / 2 + addr / 2;
1361 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1362 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1363 pagesizes[1] : pagesizes[0], &anon_loc);
1366 map->anon_loc = anon_loc;
1368 map->anon_loc = addr;
1371 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr;
1372 imgp->entry_addr = entry;
1374 if (sv->sv_protect != NULL)
1375 sv->sv_protect(imgp, SVP_IMAGE);
1377 if (interp != NULL) {
1378 VOP_UNLOCK(imgp->vp);
1379 if ((map->flags & MAP_ASLR) != 0) {
1380 /* Assume that interpreter fits into 1/4 of AS */
1381 maxv1 = maxv / 2 + addr / 2;
1382 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1383 maxv1, PAGE_SIZE, &addr);
1386 error = __elfN(load_interp)(imgp, brand_info, interp,
1387 &addr, &imgp->entry_addr);
1389 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1393 addr = imgp->et_dyn_addr;
1395 error = exec_map_stack(imgp);
1400 * Construct auxargs table (used by the copyout_auxargs routine)
1402 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1403 if (elf_auxargs == NULL) {
1404 VOP_UNLOCK(imgp->vp);
1405 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1406 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1408 elf_auxargs->execfd = -1;
1409 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr;
1410 elf_auxargs->phent = hdr->e_phentsize;
1411 elf_auxargs->phnum = hdr->e_phnum;
1412 elf_auxargs->pagesz = PAGE_SIZE;
1413 elf_auxargs->base = addr;
1414 elf_auxargs->flags = 0;
1415 elf_auxargs->entry = entry;
1416 elf_auxargs->hdr_eflags = hdr->e_flags;
1418 imgp->auxargs = elf_auxargs;
1419 imgp->interpreted = 0;
1420 imgp->reloc_base = addr;
1421 imgp->proc->p_osrel = osrel;
1422 imgp->proc->p_fctl0 = fctl0;
1423 imgp->proc->p_elf_flags = hdr->e_flags;
1426 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1428 free(interp, M_TEMP);
1432 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1435 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1437 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1438 Elf_Auxinfo *argarray, *pos;
1439 struct vmspace *vmspace;
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_flags & SV_TIMEKEEP) != 0) {
1473 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1474 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset);
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 #ifdef RANDOM_FENESTRASX
1495 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) {
1496 AUXARGS_ENTRY(pos, AT_FXRNG,
1497 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset);
1500 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) {
1501 AUXARGS_ENTRY(pos, AT_KPRELOAD,
1502 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset);
1504 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop));
1505 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur;
1506 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz);
1507 AUXARGS_ENTRY(pos, AT_NULL, 0);
1509 free(imgp->auxargs, M_TEMP);
1510 imgp->auxargs = NULL;
1511 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1513 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1514 free(argarray, M_TEMP);
1519 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1523 base = (Elf_Addr *)*stack_base;
1525 if (elf_suword(base, imgp->args->argc) == -1)
1527 *stack_base = (uintptr_t)base;
1532 * Code for generating ELF core dumps.
1535 typedef void (*segment_callback)(vm_map_entry_t, void *);
1537 /* Closure for cb_put_phdr(). */
1538 struct phdr_closure {
1539 Elf_Phdr *phdr; /* Program header to fill in */
1540 Elf_Off offset; /* Offset of segment in core file */
1544 int type; /* Note type. */
1545 struct regset *regset; /* Register set. */
1546 outfunc_t outfunc; /* Output function. */
1547 void *outarg; /* Argument for the output function. */
1548 size_t outsize; /* Output size. */
1549 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1552 TAILQ_HEAD(note_info_list, note_info);
1554 extern int compress_user_cores;
1555 extern int compress_user_cores_level;
1557 static void cb_put_phdr(vm_map_entry_t, void *);
1558 static void cb_size_segment(vm_map_entry_t, void *);
1559 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1561 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1562 struct note_info_list *, size_t, int);
1563 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1565 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1566 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1567 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1568 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1569 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1570 static void note_procstat_files(void *, struct sbuf *, size_t *);
1571 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1572 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1573 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1574 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1575 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1578 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1581 return (core_write((struct coredump_params *)arg, base, len, offset,
1582 UIO_SYSSPACE, NULL));
1586 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1588 struct ucred *cred = td->td_ucred;
1589 int compm, error = 0;
1590 struct sseg_closure seginfo;
1591 struct note_info_list notelst;
1592 struct coredump_params params;
1593 struct note_info *ninfo;
1595 size_t hdrsize, notesz, coresize;
1599 TAILQ_INIT(¬elst);
1601 /* Size the program segments. */
1602 __elfN(size_segments)(td, &seginfo, flags);
1605 * Collect info about the core file header area.
1607 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1608 if (seginfo.count + 1 >= PN_XNUM)
1609 hdrsize += sizeof(Elf_Shdr);
1610 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1611 coresize = round_page(hdrsize + notesz) + seginfo.size;
1613 /* Set up core dump parameters. */
1615 params.active_cred = cred;
1616 params.file_cred = NOCRED;
1623 PROC_LOCK(td->td_proc);
1624 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1625 PROC_UNLOCK(td->td_proc);
1632 if (coresize >= limit) {
1637 /* Create a compression stream if necessary. */
1638 compm = compress_user_cores;
1639 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1641 compm = COMPRESS_GZIP;
1643 params.comp = compressor_init(core_compressed_write,
1644 compm, CORE_BUF_SIZE,
1645 compress_user_cores_level, ¶ms);
1646 if (params.comp == NULL) {
1650 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1654 * Allocate memory for building the header, fill it up,
1655 * and write it out following the notes.
1657 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1658 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1661 /* Write the contents of all of the writable segments. */
1667 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1668 offset = round_page(hdrsize + notesz);
1669 for (i = 0; i < seginfo.count; i++) {
1670 error = core_output((char *)(uintptr_t)php->p_vaddr,
1671 php->p_filesz, offset, ¶ms, tmpbuf);
1674 offset += php->p_filesz;
1677 if (error == 0 && params.comp != NULL)
1678 error = compressor_flush(params.comp);
1682 "Failed to write core file for process %s (error %d)\n",
1683 curproc->p_comm, error);
1687 free(tmpbuf, M_TEMP);
1688 if (params.comp != NULL)
1689 compressor_fini(params.comp);
1690 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1691 TAILQ_REMOVE(¬elst, ninfo, link);
1692 free(ninfo, M_TEMP);
1701 * A callback for each_dumpable_segment() to write out the segment's
1702 * program header entry.
1705 cb_put_phdr(vm_map_entry_t entry, void *closure)
1707 struct phdr_closure *phc = (struct phdr_closure *)closure;
1708 Elf_Phdr *phdr = phc->phdr;
1710 phc->offset = round_page(phc->offset);
1712 phdr->p_type = PT_LOAD;
1713 phdr->p_offset = phc->offset;
1714 phdr->p_vaddr = entry->start;
1716 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1717 phdr->p_align = PAGE_SIZE;
1718 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1720 phc->offset += phdr->p_filesz;
1725 * A callback for each_dumpable_segment() to gather information about
1726 * the number of segments and their total size.
1729 cb_size_segment(vm_map_entry_t entry, void *closure)
1731 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1734 ssc->size += entry->end - entry->start;
1738 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1744 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1748 * For each writable segment in the process's memory map, call the given
1749 * function with a pointer to the map entry and some arbitrary
1750 * caller-supplied data.
1753 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1756 struct proc *p = td->td_proc;
1757 vm_map_t map = &p->p_vmspace->vm_map;
1758 vm_map_entry_t entry;
1759 vm_object_t backing_object, object;
1762 vm_map_lock_read(map);
1763 VM_MAP_ENTRY_FOREACH(entry, map) {
1765 * Don't dump inaccessible mappings, deal with legacy
1768 * Note that read-only segments related to the elf binary
1769 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1770 * need to arbitrarily ignore such segments.
1772 if ((flags & SVC_ALL) == 0) {
1773 if (elf_legacy_coredump) {
1774 if ((entry->protection & VM_PROT_RW) !=
1778 if ((entry->protection & VM_PROT_ALL) == 0)
1784 * Dont include memory segment in the coredump if
1785 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1786 * madvise(2). Do not dump submaps (i.e. parts of the
1789 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1791 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1792 (flags & SVC_ALL) == 0)
1794 if ((object = entry->object.vm_object) == NULL)
1797 /* Ignore memory-mapped devices and such things. */
1798 VM_OBJECT_RLOCK(object);
1799 while ((backing_object = object->backing_object) != NULL) {
1800 VM_OBJECT_RLOCK(backing_object);
1801 VM_OBJECT_RUNLOCK(object);
1802 object = backing_object;
1804 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1805 VM_OBJECT_RUNLOCK(object);
1809 (*func)(entry, closure);
1811 vm_map_unlock_read(map);
1815 * Write the core file header to the file, including padding up to
1816 * the page boundary.
1819 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1820 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1823 struct note_info *ninfo;
1827 /* Fill in the header. */
1828 bzero(hdr, hdrsize);
1829 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1831 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1832 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1833 sbuf_start_section(sb, NULL);
1834 sbuf_bcat(sb, hdr, hdrsize);
1835 TAILQ_FOREACH(ninfo, notelst, link)
1836 __elfN(putnote)(p->td, ninfo, sb);
1837 /* Align up to a page boundary for the program segments. */
1838 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1839 error = sbuf_finish(sb);
1846 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1856 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1857 __elfN(note_prpsinfo), p);
1860 * To have the debugger select the right thread (LWP) as the initial
1861 * thread, we dump the state of the thread passed to us in td first.
1862 * This is the thread that causes the core dump and thus likely to
1863 * be the right thread one wants to have selected in the debugger.
1866 while (thr != NULL) {
1867 size += __elfN(prepare_register_notes)(td, list, thr);
1868 size += __elfN(register_note)(td, list, -1,
1869 __elfN(note_threadmd), thr);
1871 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1872 TAILQ_NEXT(thr, td_plist);
1874 thr = TAILQ_NEXT(thr, td_plist);
1877 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1878 __elfN(note_procstat_proc), p);
1879 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1880 note_procstat_files, p);
1881 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1882 note_procstat_vmmap, p);
1883 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1884 note_procstat_groups, p);
1885 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1886 note_procstat_umask, p);
1887 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1888 note_procstat_rlimit, p);
1889 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1890 note_procstat_osrel, p);
1891 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1892 __elfN(note_procstat_psstrings), p);
1893 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1894 __elfN(note_procstat_auxv), p);
1900 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1901 size_t notesz, int flags)
1906 struct phdr_closure phc;
1909 ehdr = (Elf_Ehdr *)hdr;
1910 bi = td->td_proc->p_elf_brandinfo;
1912 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1913 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1914 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1915 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1916 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1917 ehdr->e_ident[EI_DATA] = ELF_DATA;
1918 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1919 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1920 ehdr->e_ident[EI_ABIVERSION] = 0;
1921 ehdr->e_ident[EI_PAD] = 0;
1922 ehdr->e_type = ET_CORE;
1923 ehdr->e_machine = bi->machine;
1924 ehdr->e_version = EV_CURRENT;
1926 ehdr->e_phoff = sizeof(Elf_Ehdr);
1927 ehdr->e_flags = td->td_proc->p_elf_flags;
1928 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1929 ehdr->e_phentsize = sizeof(Elf_Phdr);
1930 ehdr->e_shentsize = sizeof(Elf_Shdr);
1931 ehdr->e_shstrndx = SHN_UNDEF;
1932 if (numsegs + 1 < PN_XNUM) {
1933 ehdr->e_phnum = numsegs + 1;
1936 ehdr->e_phnum = PN_XNUM;
1939 ehdr->e_shoff = ehdr->e_phoff +
1940 (numsegs + 1) * ehdr->e_phentsize;
1941 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1942 ("e_shoff: %zu, hdrsize - shdr: %zu",
1943 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1945 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1946 memset(shdr, 0, sizeof(*shdr));
1948 * A special first section is used to hold large segment and
1949 * section counts. This was proposed by Sun Microsystems in
1950 * Solaris and has been adopted by Linux; the standard ELF
1951 * tools are already familiar with the technique.
1953 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1954 * (or 12-7 depending on the version of the document) for more
1957 shdr->sh_type = SHT_NULL;
1958 shdr->sh_size = ehdr->e_shnum;
1959 shdr->sh_link = ehdr->e_shstrndx;
1960 shdr->sh_info = numsegs + 1;
1964 * Fill in the program header entries.
1966 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1968 /* The note segement. */
1969 phdr->p_type = PT_NOTE;
1970 phdr->p_offset = hdrsize;
1973 phdr->p_filesz = notesz;
1975 phdr->p_flags = PF_R;
1976 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1979 /* All the writable segments from the program. */
1981 phc.offset = round_page(hdrsize + notesz);
1982 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1986 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1987 struct regset *regset, struct thread *target_td)
1989 const struct sysentvec *sv;
1990 struct note_info *ninfo;
1991 size_t size, notesize;
1994 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1997 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1998 ninfo->type = regset->note;
1999 ninfo->regset = regset;
2000 ninfo->outarg = target_td;
2001 ninfo->outsize = size;
2002 TAILQ_INSERT_TAIL(list, ninfo, link);
2004 sv = td->td_proc->p_sysent;
2005 notesize = sizeof(Elf_Note) + /* note header */
2006 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2008 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2014 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2015 int type, outfunc_t out, void *arg)
2017 const struct sysentvec *sv;
2018 struct note_info *ninfo;
2019 size_t size, notesize;
2021 sv = td->td_proc->p_sysent;
2023 out(arg, NULL, &size);
2024 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2026 ninfo->outfunc = out;
2027 ninfo->outarg = arg;
2028 ninfo->outsize = size;
2029 TAILQ_INSERT_TAIL(list, ninfo, link);
2034 notesize = sizeof(Elf_Note) + /* note header */
2035 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2037 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2043 append_note_data(const void *src, void *dst, size_t len)
2047 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2049 bcopy(src, dst, len);
2050 bzero((char *)dst + len, padded_len - len);
2052 return (padded_len);
2056 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2064 note = (Elf_Note *)buf;
2065 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2066 note->n_descsz = size;
2067 note->n_type = type;
2068 buf += sizeof(*note);
2069 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2070 sizeof(FREEBSD_ABI_VENDOR));
2071 append_note_data(src, buf, size);
2076 notesize = sizeof(Elf_Note) + /* note header */
2077 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2079 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2085 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2088 const struct sysentvec *sv;
2089 ssize_t old_len, sect_len;
2090 size_t new_len, descsz, i;
2092 if (ninfo->type == -1) {
2093 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2097 sv = td->td_proc->p_sysent;
2099 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2100 note.n_descsz = ninfo->outsize;
2101 note.n_type = ninfo->type;
2103 sbuf_bcat(sb, ¬e, sizeof(note));
2104 sbuf_start_section(sb, &old_len);
2105 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2106 strlen(sv->sv_elf_core_abi_vendor) + 1);
2107 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2108 if (note.n_descsz == 0)
2110 sbuf_start_section(sb, &old_len);
2111 if (ninfo->regset != NULL) {
2112 struct regset *regset = ninfo->regset;
2115 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2116 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2117 sbuf_bcat(sb, buf, ninfo->outsize);
2120 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2121 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2125 new_len = (size_t)sect_len;
2126 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2127 if (new_len < descsz) {
2129 * It is expected that individual note emitters will correctly
2130 * predict their expected output size and fill up to that size
2131 * themselves, padding in a format-specific way if needed.
2132 * However, in case they don't, just do it here with zeros.
2134 for (i = 0; i < descsz - new_len; i++)
2136 } else if (new_len > descsz) {
2138 * We can't always truncate sb -- we may have drained some
2141 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2142 "read it (%zu > %zu). Since it is longer than "
2143 "expected, this coredump's notes are corrupt. THIS "
2144 "IS A BUG in the note_procstat routine for type %u.\n",
2145 __func__, (unsigned)note.n_type, new_len, descsz,
2146 (unsigned)note.n_type));
2151 * Miscellaneous note out functions.
2154 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2155 #include <compat/freebsd32/freebsd32.h>
2156 #include <compat/freebsd32/freebsd32_signal.h>
2158 typedef struct prstatus32 elf_prstatus_t;
2159 typedef struct prpsinfo32 elf_prpsinfo_t;
2160 typedef struct fpreg32 elf_prfpregset_t;
2161 typedef struct fpreg32 elf_fpregset_t;
2162 typedef struct reg32 elf_gregset_t;
2163 typedef struct thrmisc32 elf_thrmisc_t;
2164 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2165 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2166 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2167 typedef uint32_t elf_ps_strings_t;
2169 typedef prstatus_t elf_prstatus_t;
2170 typedef prpsinfo_t elf_prpsinfo_t;
2171 typedef prfpregset_t elf_prfpregset_t;
2172 typedef prfpregset_t elf_fpregset_t;
2173 typedef gregset_t elf_gregset_t;
2174 typedef thrmisc_t elf_thrmisc_t;
2175 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2176 #define ELF_KERN_PROC_MASK 0
2177 typedef struct kinfo_proc elf_kinfo_proc_t;
2178 typedef vm_offset_t elf_ps_strings_t;
2182 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2188 elf_prpsinfo_t *psinfo;
2193 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2194 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2195 psinfo->pr_version = PRPSINFO_VERSION;
2196 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2197 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2199 if (p->p_args != NULL) {
2200 len = sizeof(psinfo->pr_psargs) - 1;
2201 if (len > p->p_args->ar_length)
2202 len = p->p_args->ar_length;
2203 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2209 sbuf_new(&sbarg, psinfo->pr_psargs,
2210 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2211 error = proc_getargv(curthread, p, &sbarg);
2213 if (sbuf_finish(&sbarg) == 0) {
2214 len = sbuf_len(&sbarg);
2218 len = sizeof(psinfo->pr_psargs) - 1;
2220 sbuf_delete(&sbarg);
2222 if (error != 0 || len == 0 || (ssize_t)len == -1)
2223 strlcpy(psinfo->pr_psargs, p->p_comm,
2224 sizeof(psinfo->pr_psargs));
2226 KASSERT(len < sizeof(psinfo->pr_psargs),
2227 ("len is too long: %zu vs %zu", len,
2228 sizeof(psinfo->pr_psargs)));
2229 cp = psinfo->pr_psargs;
2232 cp = memchr(cp, '\0', end - cp);
2238 psinfo->pr_pid = p->p_pid;
2239 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2240 free(psinfo, M_TEMP);
2242 *sizep = sizeof(*psinfo);
2246 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2249 elf_prstatus_t *status;
2252 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2255 memset(status, 0, *sizep);
2256 status->pr_version = PRSTATUS_VERSION;
2257 status->pr_statussz = sizeof(elf_prstatus_t);
2258 status->pr_gregsetsz = sizeof(elf_gregset_t);
2259 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2260 status->pr_osreldate = osreldate;
2261 status->pr_cursig = td->td_proc->p_sig;
2262 status->pr_pid = td->td_tid;
2263 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2264 fill_regs32(td, &status->pr_reg);
2266 fill_regs(td, &status->pr_reg);
2269 *sizep = sizeof(*status);
2274 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2277 elf_prstatus_t *status;
2279 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2281 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2282 set_regs32(td, &status->pr_reg);
2284 set_regs(td, &status->pr_reg);
2289 static struct regset __elfN(regset_prstatus) = {
2290 .note = NT_PRSTATUS,
2291 .size = sizeof(elf_prstatus_t),
2292 .get = __elfN(get_prstatus),
2293 .set = __elfN(set_prstatus),
2295 ELF_REGSET(__elfN(regset_prstatus));
2298 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2301 elf_prfpregset_t *fpregset;
2304 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2307 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2308 fill_fpregs32(td, fpregset);
2310 fill_fpregs(td, fpregset);
2313 *sizep = sizeof(*fpregset);
2318 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2321 elf_prfpregset_t *fpregset;
2324 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2325 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2326 set_fpregs32(td, fpregset);
2328 set_fpregs(td, fpregset);
2333 static struct regset __elfN(regset_fpregset) = {
2334 .note = NT_FPREGSET,
2335 .size = sizeof(elf_prfpregset_t),
2336 .get = __elfN(get_fpregset),
2337 .set = __elfN(set_fpregset),
2339 ELF_REGSET(__elfN(regset_fpregset));
2342 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2345 elf_thrmisc_t *thrmisc;
2348 KASSERT(*sizep == sizeof(*thrmisc),
2349 ("%s: invalid size", __func__));
2351 bzero(thrmisc, sizeof(*thrmisc));
2352 strcpy(thrmisc->pr_tname, td->td_name);
2354 *sizep = sizeof(*thrmisc);
2358 static struct regset __elfN(regset_thrmisc) = {
2360 .size = sizeof(elf_thrmisc_t),
2361 .get = __elfN(get_thrmisc),
2363 ELF_REGSET(__elfN(regset_thrmisc));
2366 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2373 size = sizeof(structsize) + sizeof(pl);
2375 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2376 structsize = sizeof(pl);
2377 memcpy(buf, &structsize, sizeof(structsize));
2378 bzero(&pl, sizeof(pl));
2379 pl.pl_lwpid = td->td_tid;
2380 pl.pl_event = PL_EVENT_NONE;
2381 pl.pl_sigmask = td->td_sigmask;
2382 pl.pl_siglist = td->td_siglist;
2383 if (td->td_si.si_signo != 0) {
2384 pl.pl_event = PL_EVENT_SIGNAL;
2385 pl.pl_flags |= PL_FLAG_SI;
2386 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2387 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2389 pl.pl_siginfo = td->td_si;
2392 strcpy(pl.pl_tdname, td->td_name);
2393 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2394 memcpy((int *)buf + 1, &pl, sizeof(pl));
2400 static struct regset __elfN(regset_lwpinfo) = {
2401 .note = NT_PTLWPINFO,
2402 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2403 .get = __elfN(get_lwpinfo),
2405 ELF_REGSET(__elfN(regset_lwpinfo));
2408 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2409 struct thread *target_td)
2411 struct sysentvec *sv = td->td_proc->p_sysent;
2412 struct regset **regsetp, **regset_end, *regset;
2417 /* NT_PRSTATUS must be the first register set note. */
2418 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2421 regsetp = sv->sv_regset_begin;
2422 if (regsetp == NULL) {
2423 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2424 size += __elfN(register_regset_note)(td, list,
2425 &__elfN(regset_fpregset), target_td);
2428 regset_end = sv->sv_regset_end;
2429 MPASS(regset_end != NULL);
2430 for (; regsetp < regset_end; regsetp++) {
2432 if (regset->note == NT_PRSTATUS)
2434 size += __elfN(register_regset_note)(td, list, regset,
2441 * Allow for MD specific notes, as well as any MD
2442 * specific preparations for writing MI notes.
2445 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2451 td = (struct thread *)arg;
2453 if (size != 0 && sb != NULL)
2454 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2458 __elfN(dump_thread)(td, buf, &size);
2459 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2460 if (size != 0 && sb != NULL)
2461 sbuf_bcat(sb, buf, size);
2466 #ifdef KINFO_PROC_SIZE
2467 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2471 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2478 size = sizeof(structsize) + p->p_numthreads *
2479 sizeof(elf_kinfo_proc_t);
2482 KASSERT(*sizep == size, ("invalid size"));
2483 structsize = sizeof(elf_kinfo_proc_t);
2484 sbuf_bcat(sb, &structsize, sizeof(structsize));
2485 sx_slock(&proctree_lock);
2487 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2488 sx_sunlock(&proctree_lock);
2493 #ifdef KINFO_FILE_SIZE
2494 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2498 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2501 size_t size, sect_sz, i;
2502 ssize_t start_len, sect_len;
2503 int structsize, filedesc_flags;
2505 if (coredump_pack_fileinfo)
2506 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2511 structsize = sizeof(struct kinfo_file);
2514 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2515 sbuf_set_drain(sb, sbuf_count_drain, &size);
2516 sbuf_bcat(sb, &structsize, sizeof(structsize));
2518 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2523 sbuf_start_section(sb, &start_len);
2525 sbuf_bcat(sb, &structsize, sizeof(structsize));
2527 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2530 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2535 KASSERT(sect_sz <= *sizep,
2536 ("kern_proc_filedesc_out did not respect maxlen; "
2537 "requested %zu, got %zu", *sizep - sizeof(structsize),
2538 sect_sz - sizeof(structsize)));
2540 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2545 #ifdef KINFO_VMENTRY_SIZE
2546 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2550 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2554 int structsize, vmmap_flags;
2556 if (coredump_pack_vmmapinfo)
2557 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2562 structsize = sizeof(struct kinfo_vmentry);
2565 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2566 sbuf_set_drain(sb, sbuf_count_drain, &size);
2567 sbuf_bcat(sb, &structsize, sizeof(structsize));
2569 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2574 sbuf_bcat(sb, &structsize, sizeof(structsize));
2576 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2582 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2589 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2591 KASSERT(*sizep == size, ("invalid size"));
2592 structsize = sizeof(gid_t);
2593 sbuf_bcat(sb, &structsize, sizeof(structsize));
2594 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2601 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2608 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2610 KASSERT(*sizep == size, ("invalid size"));
2611 structsize = sizeof(p->p_pd->pd_cmask);
2612 sbuf_bcat(sb, &structsize, sizeof(structsize));
2613 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2619 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2622 struct rlimit rlim[RLIM_NLIMITS];
2627 size = sizeof(structsize) + sizeof(rlim);
2629 KASSERT(*sizep == size, ("invalid size"));
2630 structsize = sizeof(rlim);
2631 sbuf_bcat(sb, &structsize, sizeof(structsize));
2633 for (i = 0; i < RLIM_NLIMITS; i++)
2634 lim_rlimit_proc(p, i, &rlim[i]);
2636 sbuf_bcat(sb, rlim, sizeof(rlim));
2642 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2649 size = sizeof(structsize) + sizeof(p->p_osrel);
2651 KASSERT(*sizep == size, ("invalid size"));
2652 structsize = sizeof(p->p_osrel);
2653 sbuf_bcat(sb, &structsize, sizeof(structsize));
2654 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2660 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2663 elf_ps_strings_t ps_strings;
2668 size = sizeof(structsize) + sizeof(ps_strings);
2670 KASSERT(*sizep == size, ("invalid size"));
2671 structsize = sizeof(ps_strings);
2672 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2673 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2675 ps_strings = PROC_PS_STRINGS(p);
2677 sbuf_bcat(sb, &structsize, sizeof(structsize));
2678 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2684 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2693 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2695 sbuf_set_drain(sb, sbuf_count_drain, &size);
2696 sbuf_bcat(sb, &structsize, sizeof(structsize));
2698 proc_getauxv(curthread, p, sb);
2704 structsize = sizeof(Elf_Auxinfo);
2705 sbuf_bcat(sb, &structsize, sizeof(structsize));
2707 proc_getauxv(curthread, p, sb);
2712 #define MAX_NOTES_LOOP 4096
2714 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2715 const char *note_vendor, const Elf_Phdr *pnote,
2716 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2718 const Elf_Note *note, *note0, *note_end;
2719 const char *note_name;
2724 /* We need some limit, might as well use PAGE_SIZE. */
2725 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2727 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2728 if (pnote->p_offset > PAGE_SIZE ||
2729 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2730 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2732 VOP_UNLOCK(imgp->vp);
2733 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2734 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2736 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2737 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2738 curthread->td_ucred, NOCRED, NULL, curthread);
2740 uprintf("i/o error PT_NOTE\n");
2743 note = note0 = (const Elf_Note *)buf;
2744 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2746 note = note0 = (const Elf_Note *)(imgp->image_header +
2748 note_end = (const Elf_Note *)(imgp->image_header +
2749 pnote->p_offset + pnote->p_filesz);
2752 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end;
2754 if (!aligned(note, Elf32_Addr)) {
2755 uprintf("Unaligned ELF note\n");
2758 if ((const char *)note_end - (const char *)note <
2760 uprintf("ELF note to short\n");
2763 if (note->n_namesz != checknote->n_namesz ||
2764 note->n_descsz != checknote->n_descsz ||
2765 note->n_type != checknote->n_type)
2767 note_name = (const char *)(note + 1);
2768 if (note_name + checknote->n_namesz >=
2769 (const char *)note_end || strncmp(note_vendor,
2770 note_name, checknote->n_namesz) != 0)
2773 if (cb(note, cb_arg, &res))
2776 note = (const Elf_Note *)((const char *)(note + 1) +
2777 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2778 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2780 if (i >= MAX_NOTES_LOOP)
2781 uprintf("ELF note parser reached %d notes\n", i);
2789 struct brandnote_cb_arg {
2790 Elf_Brandnote *brandnote;
2795 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2797 struct brandnote_cb_arg *arg;
2802 * Fetch the osreldate for binary from the ELF OSABI-note if
2805 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2806 arg->brandnote->trans_osrel != NULL ?
2807 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2812 static Elf_Note fctl_note = {
2813 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2814 .n_descsz = sizeof(uint32_t),
2815 .n_type = NT_FREEBSD_FEATURE_CTL,
2818 struct fctl_cb_arg {
2824 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2826 struct fctl_cb_arg *arg;
2827 const Elf32_Word *desc;
2831 p = (uintptr_t)(note + 1);
2832 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2833 desc = (const Elf32_Word *)p;
2834 *arg->has_fctl0 = true;
2835 *arg->fctl0 = desc[0];
2841 * Try to find the appropriate ABI-note section for checknote, fetch
2842 * the osreldate and feature control flags for binary from the ELF
2843 * OSABI-note. Only the first page of the image is searched, the same
2847 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2848 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2850 const Elf_Phdr *phdr;
2851 const Elf_Ehdr *hdr;
2852 struct brandnote_cb_arg b_arg;
2853 struct fctl_cb_arg f_arg;
2856 hdr = (const Elf_Ehdr *)imgp->image_header;
2857 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2858 b_arg.brandnote = brandnote;
2859 b_arg.osrel = osrel;
2860 f_arg.has_fctl0 = has_fctl0;
2861 f_arg.fctl0 = fctl0;
2863 for (i = 0; i < hdr->e_phnum; i++) {
2864 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2865 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2867 for (j = 0; j < hdr->e_phnum; j++) {
2868 if (phdr[j].p_type == PT_NOTE &&
2869 __elfN(parse_notes)(imgp, &fctl_note,
2870 FREEBSD_ABI_VENDOR, &phdr[j],
2871 note_fctl_cb, &f_arg))
2882 * Tell kern_execve.c about it, with a little help from the linker.
2884 static struct execsw __elfN(execsw) = {
2885 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2886 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2888 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2891 __elfN(trans_prot)(Elf_Word flags)
2897 prot |= VM_PROT_EXECUTE;
2899 prot |= VM_PROT_WRITE;
2901 prot |= VM_PROT_READ;
2902 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2903 if (i386_read_exec && (flags & PF_R))
2904 prot |= VM_PROT_EXECUTE;
2910 __elfN(untrans_prot)(vm_prot_t prot)
2915 if (prot & VM_PROT_EXECUTE)
2917 if (prot & VM_PROT_READ)
2919 if (prot & VM_PROT_WRITE)