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 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures
177 * have limited address space (which can cause issues for applications with
178 * high memory use) so we leave it off there.
180 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
182 &__elfN(aslr_enabled), 0,
183 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
184 ": enable address map randomization");
187 * Enable ASLR by default for 64-bit PIE binaries.
189 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
190 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
191 &__elfN(pie_aslr_enabled), 0,
192 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
193 ": enable address map randomization for PIE binaries");
196 * Sbrk is deprecated and it can be assumed that in most cases it will not be
197 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR
198 * to use the bss grow region.
200 static int __elfN(aslr_honor_sbrk) = 0;
201 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
202 &__elfN(aslr_honor_sbrk), 0,
203 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
205 static int __elfN(aslr_stack) = 1;
206 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
207 &__elfN(aslr_stack), 0,
208 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
209 ": enable stack address randomization");
211 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64;
212 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN,
213 &__elfN(aslr_shared_page), 0,
214 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
215 ": enable shared page address randomization");
217 static int __elfN(sigfastblock) = 1;
218 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
219 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
220 "enable sigfastblock for new processes");
222 static bool __elfN(allow_wx) = true;
223 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
224 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
225 "Allow pages to be mapped simultaneously writable and executable");
227 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
229 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
231 Elf_Brandnote __elfN(freebsd_brandnote) = {
232 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
233 .hdr.n_descsz = sizeof(int32_t),
234 .hdr.n_type = NT_FREEBSD_ABI_TAG,
235 .vendor = FREEBSD_ABI_VENDOR,
236 .flags = BN_TRANSLATE_OSREL,
237 .trans_osrel = __elfN(freebsd_trans_osrel)
241 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
245 p = (uintptr_t)(note + 1);
246 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
247 *osrel = *(const int32_t *)(p);
252 static const char GNU_ABI_VENDOR[] = "GNU";
253 static int GNU_KFREEBSD_ABI_DESC = 3;
255 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
256 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
257 .hdr.n_descsz = 16, /* XXX at least 16 */
259 .vendor = GNU_ABI_VENDOR,
260 .flags = BN_TRANSLATE_OSREL,
261 .trans_osrel = kfreebsd_trans_osrel
265 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
267 const Elf32_Word *desc;
270 p = (uintptr_t)(note + 1);
271 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
273 desc = (const Elf32_Word *)p;
274 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
278 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
279 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
281 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
287 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
291 for (i = 0; i < MAX_BRANDS; i++) {
292 if (elf_brand_list[i] == NULL) {
293 elf_brand_list[i] = entry;
297 if (i == MAX_BRANDS) {
298 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
306 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
310 for (i = 0; i < MAX_BRANDS; i++) {
311 if (elf_brand_list[i] == entry) {
312 elf_brand_list[i] = NULL;
322 __elfN(brand_inuse)(Elf_Brandinfo *entry)
327 sx_slock(&allproc_lock);
328 FOREACH_PROC_IN_SYSTEM(p) {
329 if (p->p_sysent == entry->sysvec) {
334 sx_sunlock(&allproc_lock);
339 static Elf_Brandinfo *
340 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
341 int32_t *osrel, uint32_t *fctl0)
343 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
344 Elf_Brandinfo *bi, *bi_m;
346 int i, interp_name_len;
348 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
351 * We support four types of branding -- (1) the ELF EI_OSABI field
352 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
353 * branding w/in the ELF header, (3) path of the `interp_path'
354 * field, and (4) the ".note.ABI-tag" ELF section.
357 /* Look for an ".note.ABI-tag" ELF section */
359 for (i = 0; i < MAX_BRANDS; i++) {
360 bi = elf_brand_list[i];
363 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
365 if (hdr->e_machine == bi->machine && (bi->flags &
366 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
370 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
372 /* Give brand a chance to veto check_note's guess */
373 if (ret && bi->header_supported) {
374 ret = bi->header_supported(imgp, osrel,
375 has_fctl0 ? fctl0 : NULL);
378 * If note checker claimed the binary, but the
379 * interpreter path in the image does not
380 * match default one for the brand, try to
381 * search for other brands with the same
382 * interpreter. Either there is better brand
383 * with the right interpreter, or, failing
384 * this, we return first brand which accepted
385 * our note and, optionally, header.
387 if (ret && bi_m == NULL && interp != NULL &&
388 (bi->interp_path == NULL ||
389 (strlen(bi->interp_path) + 1 != interp_name_len ||
390 strncmp(interp, bi->interp_path, interp_name_len)
402 /* If the executable has a brand, search for it in the brand list. */
403 for (i = 0; i < MAX_BRANDS; i++) {
404 bi = elf_brand_list[i];
405 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
406 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
408 if (hdr->e_machine == bi->machine &&
409 (hdr->e_ident[EI_OSABI] == bi->brand ||
410 (bi->compat_3_brand != NULL &&
411 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
412 bi->compat_3_brand) == 0))) {
413 /* Looks good, but give brand a chance to veto */
414 if (bi->header_supported == NULL ||
415 bi->header_supported(imgp, NULL, NULL)) {
417 * Again, prefer strictly matching
420 if (interp_name_len == 0 &&
421 bi->interp_path == NULL)
423 if (bi->interp_path != NULL &&
424 strlen(bi->interp_path) + 1 ==
425 interp_name_len && strncmp(interp,
426 bi->interp_path, interp_name_len) == 0)
436 /* No known brand, see if the header is recognized by any brand */
437 for (i = 0; i < MAX_BRANDS; i++) {
438 bi = elf_brand_list[i];
439 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
440 bi->header_supported == NULL)
442 if (hdr->e_machine == bi->machine) {
443 ret = bi->header_supported(imgp, NULL, NULL);
449 /* Lacking a known brand, search for a recognized interpreter. */
450 if (interp != NULL) {
451 for (i = 0; i < MAX_BRANDS; i++) {
452 bi = elf_brand_list[i];
453 if (bi == NULL || (bi->flags &
454 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
457 if (hdr->e_machine == bi->machine &&
458 bi->interp_path != NULL &&
459 /* ELF image p_filesz includes terminating zero */
460 strlen(bi->interp_path) + 1 == interp_name_len &&
461 strncmp(interp, bi->interp_path, interp_name_len)
462 == 0 && (bi->header_supported == NULL ||
463 bi->header_supported(imgp, NULL, NULL)))
468 /* Lacking a recognized interpreter, try the default brand */
469 for (i = 0; i < MAX_BRANDS; i++) {
470 bi = elf_brand_list[i];
471 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
472 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
474 if (hdr->e_machine == bi->machine &&
475 __elfN(fallback_brand) == bi->brand &&
476 (bi->header_supported == NULL ||
477 bi->header_supported(imgp, NULL, NULL)))
484 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
486 return (hdr->e_phoff <= PAGE_SIZE &&
487 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
491 __elfN(check_header)(const Elf_Ehdr *hdr)
497 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
498 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
499 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
500 hdr->e_phentsize != sizeof(Elf_Phdr) ||
501 hdr->e_version != ELF_TARG_VER)
505 * Make sure we have at least one brand for this machine.
508 for (i = 0; i < MAX_BRANDS; i++) {
509 bi = elf_brand_list[i];
510 if (bi != NULL && bi->machine == hdr->e_machine)
520 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
521 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
528 * Create the page if it doesn't exist yet. Ignore errors.
530 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
531 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
534 * Find the page from the underlying object.
536 if (object != NULL) {
537 sf = vm_imgact_map_page(object, offset);
539 return (KERN_FAILURE);
540 off = offset - trunc_page(offset);
541 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
543 vm_imgact_unmap_page(sf);
545 return (KERN_FAILURE);
548 return (KERN_SUCCESS);
552 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
553 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
559 int error, locked, rv;
561 if (start != trunc_page(start)) {
562 rv = __elfN(map_partial)(map, object, offset, start,
563 round_page(start), prot);
564 if (rv != KERN_SUCCESS)
566 offset += round_page(start) - start;
567 start = round_page(start);
569 if (end != round_page(end)) {
570 rv = __elfN(map_partial)(map, object, offset +
571 trunc_page(end) - start, trunc_page(end), end, prot);
572 if (rv != KERN_SUCCESS)
574 end = trunc_page(end);
577 return (KERN_SUCCESS);
578 if ((offset & PAGE_MASK) != 0) {
580 * The mapping is not page aligned. This means that we have
583 rv = vm_map_fixed(map, NULL, 0, start, end - start,
584 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
585 if (rv != KERN_SUCCESS)
588 return (KERN_SUCCESS);
589 for (; start < end; start += sz) {
590 sf = vm_imgact_map_page(object, offset);
592 return (KERN_FAILURE);
593 off = offset - trunc_page(offset);
595 if (sz > PAGE_SIZE - off)
596 sz = PAGE_SIZE - off;
597 error = copyout((caddr_t)sf_buf_kva(sf) + off,
599 vm_imgact_unmap_page(sf);
601 return (KERN_FAILURE);
605 vm_object_reference(object);
606 rv = vm_map_fixed(map, object, offset, start, end - start,
607 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
608 (object != NULL ? MAP_VN_EXEC : 0));
609 if (rv != KERN_SUCCESS) {
610 locked = VOP_ISLOCKED(imgp->vp);
611 VOP_UNLOCK(imgp->vp);
612 vm_object_deallocate(object);
613 vn_lock(imgp->vp, locked | LK_RETRY);
615 } else if (object != NULL) {
616 MPASS(imgp->vp->v_object == object);
617 VOP_SET_TEXT_CHECKED(imgp->vp);
620 return (KERN_SUCCESS);
624 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
625 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
631 vm_offset_t map_addr;
634 vm_ooffset_t file_addr;
637 * It's necessary to fail if the filsz + offset taken from the
638 * header is greater than the actual file pager object's size.
639 * If we were to allow this, then the vm_map_find() below would
640 * walk right off the end of the file object and into the ether.
642 * While I'm here, might as well check for something else that
643 * is invalid: filsz cannot be greater than memsz.
645 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
647 uprintf("elf_load_section: truncated ELF file\n");
651 object = imgp->object;
652 map = &imgp->proc->p_vmspace->vm_map;
653 map_addr = trunc_page((vm_offset_t)vmaddr);
654 file_addr = trunc_page(offset);
657 * We have two choices. We can either clear the data in the last page
658 * of an oversized mapping, or we can start the anon mapping a page
659 * early and copy the initialized data into that first page. We
664 else if (memsz > filsz)
665 map_len = trunc_page(offset + filsz) - file_addr;
667 map_len = round_page(offset + filsz) - file_addr;
670 /* cow flags: don't dump readonly sections in core */
671 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
672 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
674 rv = __elfN(map_insert)(imgp, map, object, file_addr,
675 map_addr, map_addr + map_len, prot, cow);
676 if (rv != KERN_SUCCESS)
679 /* we can stop now if we've covered it all */
685 * We have to get the remaining bit of the file into the first part
686 * of the oversized map segment. This is normally because the .data
687 * segment in the file is extended to provide bss. It's a neat idea
688 * to try and save a page, but it's a pain in the behind to implement.
690 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
692 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
693 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
695 /* This had damn well better be true! */
697 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
698 map_addr + map_len, prot, 0);
699 if (rv != KERN_SUCCESS)
704 sf = vm_imgact_map_page(object, offset + filsz);
708 /* send the page fragment to user space */
709 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
711 vm_imgact_unmap_page(sf);
717 * Remove write access to the page if it was only granted by map_insert
720 if ((prot & VM_PROT_WRITE) == 0)
721 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
722 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
728 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
729 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
736 ASSERT_VOP_LOCKED(imgp->vp, __func__);
741 for (i = 0; i < hdr->e_phnum; i++) {
742 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
745 /* Loadable segment */
746 prot = __elfN(trans_prot)(phdr[i].p_flags);
747 error = __elfN(load_section)(imgp, phdr[i].p_offset,
748 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
749 phdr[i].p_memsz, phdr[i].p_filesz, prot);
754 * Establish the base address if this is the first segment.
757 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
762 if (base_addrp != NULL)
763 *base_addrp = base_addr;
769 * Load the file "file" into memory. It may be either a shared object
772 * The "addr" reference parameter is in/out. On entry, it specifies
773 * the address where a shared object should be loaded. If the file is
774 * an executable, this value is ignored. On exit, "addr" specifies
775 * where the file was actually loaded.
777 * The "entry" reference parameter is out only. On exit, it specifies
778 * the entry point for the loaded file.
781 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
787 struct image_params image_params;
789 const Elf_Ehdr *hdr = NULL;
790 const Elf_Phdr *phdr = NULL;
791 struct nameidata *nd;
793 struct image_params *imgp;
795 u_long base_addr = 0;
798 #ifdef CAPABILITY_MODE
800 * XXXJA: This check can go away once we are sufficiently confident
801 * that the checks in namei() are correct.
803 if (IN_CAPABILITY_MODE(curthread))
807 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
809 attr = &tempdata->attr;
810 imgp = &tempdata->image_params;
813 * Initialize part of the common data
818 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
820 if ((error = namei(nd)) != 0) {
825 imgp->vp = nd->ni_vp;
828 * Check permissions, modes, uid, etc on the file, and "open" it.
830 error = exec_check_permissions(imgp);
834 error = exec_map_first_page(imgp);
838 imgp->object = nd->ni_vp->v_object;
840 hdr = (const Elf_Ehdr *)imgp->image_header;
841 if ((error = __elfN(check_header)(hdr)) != 0)
843 if (hdr->e_type == ET_DYN)
845 else if (hdr->e_type == ET_EXEC)
852 /* Only support headers that fit within first page for now */
853 if (!__elfN(phdr_in_zero_page)(hdr)) {
858 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
859 if (!aligned(phdr, Elf_Addr)) {
864 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
869 *entry = (unsigned long)hdr->e_entry + rbase;
873 exec_unmap_first_page(imgp);
877 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
880 free(tempdata, M_TEMP);
886 * Select randomized valid address in the map map, between minv and
887 * maxv, with specified alignment. The [minv, maxv) range must belong
888 * to the map. Note that function only allocates the address, it is
889 * up to caller to clamp maxv in a way that the final allocation
890 * length fit into the map.
892 * Result is returned in *resp, error code indicates that arguments
893 * did not pass sanity checks for overflow and range correctness.
896 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
897 u_int align, u_long *resp)
901 MPASS(vm_map_min(map) <= minv);
903 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
904 uprintf("Invalid ELF segments layout\n");
908 arc4rand(&rbase, sizeof(rbase), 0);
909 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
910 res &= ~((u_long)align - 1);
915 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
916 res, minv, maxv, rbase));
918 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
919 res, maxv, minv, rbase));
926 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
927 const Elf_Phdr *phdr, u_long et_dyn_addr)
929 struct vmspace *vmspace;
931 u_long text_size, data_size, total_size, text_addr, data_addr;
932 u_long seg_size, seg_addr;
936 text_size = data_size = total_size = text_addr = data_addr = 0;
938 for (i = 0; i < hdr->e_phnum; i++) {
939 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
942 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
943 seg_size = round_page(phdr[i].p_memsz +
944 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
947 * Make the largest executable segment the official
948 * text segment and all others data.
950 * Note that obreak() assumes that data_addr + data_size == end
951 * of data load area, and the ELF file format expects segments
952 * to be sorted by address. If multiple data segments exist,
953 * the last one will be used.
956 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
957 text_size = seg_size;
958 text_addr = seg_addr;
960 data_size = seg_size;
961 data_addr = seg_addr;
963 total_size += seg_size;
966 if (data_addr == 0 && data_size == 0) {
967 data_addr = text_addr;
968 data_size = text_size;
972 * Check limits. It should be safe to check the
973 * limits after loading the segments since we do
974 * not actually fault in all the segments pages.
976 PROC_LOCK(imgp->proc);
977 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
978 err_str = "Data segment size exceeds process limit";
979 else if (text_size > maxtsiz)
980 err_str = "Text segment size exceeds system limit";
981 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
982 err_str = "Total segment size exceeds process limit";
983 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
984 err_str = "Data segment size exceeds resource limit";
985 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
986 err_str = "Total segment size exceeds resource limit";
987 PROC_UNLOCK(imgp->proc);
988 if (err_str != NULL) {
989 uprintf("%s\n", err_str);
993 vmspace = imgp->proc->p_vmspace;
994 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
995 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
996 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
997 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1003 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
1004 char **interpp, bool *free_interpp)
1008 int error, interp_name_len;
1010 KASSERT(phdr->p_type == PT_INTERP,
1011 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1012 ASSERT_VOP_LOCKED(imgp->vp, __func__);
1016 /* Path to interpreter */
1017 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1018 uprintf("Invalid PT_INTERP\n");
1022 interp_name_len = phdr->p_filesz;
1023 if (phdr->p_offset > PAGE_SIZE ||
1024 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1026 * The vnode lock might be needed by the pagedaemon to
1027 * clean pages owned by the vnode. Do not allow sleep
1028 * waiting for memory with the vnode locked, instead
1029 * try non-sleepable allocation first, and if it
1030 * fails, go to the slow path were we drop the lock
1031 * and do M_WAITOK. A text reference prevents
1032 * modifications to the vnode content.
1034 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1035 if (interp == NULL) {
1036 VOP_UNLOCK(imgp->vp);
1037 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1038 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1041 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1042 interp_name_len, phdr->p_offset,
1043 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1046 free(interp, M_TEMP);
1047 uprintf("i/o error PT_INTERP %d\n", error);
1050 interp[interp_name_len] = '\0';
1053 *free_interpp = true;
1057 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1058 if (interp[interp_name_len - 1] != '\0') {
1059 uprintf("Invalid PT_INTERP\n");
1064 *free_interpp = false;
1069 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1070 const char *interp, u_long *addr, u_long *entry)
1075 if (brand_info->emul_path != NULL &&
1076 brand_info->emul_path[0] != '\0') {
1077 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1078 snprintf(path, MAXPATHLEN, "%s%s",
1079 brand_info->emul_path, interp);
1080 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1086 if (brand_info->interp_newpath != NULL &&
1087 (brand_info->interp_path == NULL ||
1088 strcmp(interp, brand_info->interp_path) == 0)) {
1089 error = __elfN(load_file)(imgp->proc,
1090 brand_info->interp_newpath, addr, entry);
1095 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1099 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1104 * Impossible et_dyn_addr initial value indicating that the real base
1105 * must be calculated later with some randomization applied.
1107 #define ET_DYN_ADDR_RAND 1
1110 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1113 const Elf_Ehdr *hdr;
1114 const Elf_Phdr *phdr;
1115 Elf_Auxargs *elf_auxargs;
1116 struct vmspace *vmspace;
1119 Elf_Brandinfo *brand_info;
1120 struct sysentvec *sv;
1121 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1122 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1128 hdr = (const Elf_Ehdr *)imgp->image_header;
1131 * Do we have a valid ELF header ?
1133 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1134 * if particular brand doesn't support it.
1136 if (__elfN(check_header)(hdr) != 0 ||
1137 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1141 * From here on down, we return an errno, not -1, as we've
1142 * detected an ELF file.
1145 if (!__elfN(phdr_in_zero_page)(hdr)) {
1146 uprintf("Program headers not in the first page\n");
1149 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1150 if (!aligned(phdr, Elf_Addr)) {
1151 uprintf("Unaligned program headers\n");
1159 entry = proghdr = 0;
1161 free_interp = false;
1165 * Somewhat arbitrary, limit accepted max alignment for the
1166 * loadable segment to the max supported superpage size. Too
1167 * large alignment requests are not useful and are indicators
1168 * of corrupted or outright malicious binary.
1170 maxalign = PAGE_SIZE;
1171 maxsalign = PAGE_SIZE * 1024;
1172 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1173 if (pagesizes[i] > maxsalign)
1174 maxsalign = pagesizes[i];
1179 for (i = 0; i < hdr->e_phnum; i++) {
1180 switch (phdr[i].p_type) {
1183 baddr = phdr[i].p_vaddr;
1184 if (!powerof2(phdr[i].p_align) ||
1185 phdr[i].p_align > maxsalign) {
1186 uprintf("Invalid segment alignment\n");
1190 if (phdr[i].p_align > maxalign)
1191 maxalign = phdr[i].p_align;
1192 if (mapsz + phdr[i].p_memsz < mapsz) {
1193 uprintf("Mapsize overflow\n");
1197 mapsz += phdr[i].p_memsz;
1201 * If this segment contains the program headers,
1202 * remember their virtual address for the AT_PHDR
1203 * aux entry. Static binaries don't usually include
1206 if (phdr[i].p_offset == 0 &&
1207 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1209 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1212 /* Path to interpreter */
1213 if (interp != NULL) {
1214 uprintf("Multiple PT_INTERP headers\n");
1218 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1224 if (__elfN(nxstack))
1226 __elfN(trans_prot)(phdr[i].p_flags);
1227 imgp->stack_sz = phdr[i].p_memsz;
1229 case PT_PHDR: /* Program header table info */
1230 proghdr = phdr[i].p_vaddr;
1235 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1236 if (brand_info == NULL) {
1237 uprintf("ELF binary type \"%u\" not known.\n",
1238 hdr->e_ident[EI_OSABI]);
1242 sv = brand_info->sysvec;
1244 if (hdr->e_type == ET_DYN) {
1245 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1246 uprintf("Cannot execute shared object\n");
1251 * Honour the base load address from the dso if it is
1252 * non-zero for some reason.
1255 if ((sv->sv_flags & SV_ASLR) == 0 ||
1256 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1257 et_dyn_addr = __elfN(pie_base);
1258 else if ((__elfN(pie_aslr_enabled) &&
1259 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1260 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1261 et_dyn_addr = ET_DYN_ADDR_RAND;
1263 et_dyn_addr = __elfN(pie_base);
1268 * Avoid a possible deadlock if the current address space is destroyed
1269 * and that address space maps the locked vnode. In the common case,
1270 * the locked vnode's v_usecount is decremented but remains greater
1271 * than zero. Consequently, the vnode lock is not needed by vrele().
1272 * However, in cases where the vnode lock is external, such as nullfs,
1273 * v_usecount may become zero.
1275 * The VV_TEXT flag prevents modifications to the executable while
1276 * the vnode is unlocked.
1278 VOP_UNLOCK(imgp->vp);
1281 * Decide whether to enable randomization of user mappings.
1282 * First, reset user preferences for the setid binaries.
1283 * Then, account for the support of the randomization by the
1284 * ABI, by user preferences, and make special treatment for
1287 if (imgp->credential_setid) {
1288 PROC_LOCK(imgp->proc);
1289 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1290 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1291 PROC_UNLOCK(imgp->proc);
1293 if ((sv->sv_flags & SV_ASLR) == 0 ||
1294 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1295 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1296 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1297 ("et_dyn_addr == RAND and !ASLR"));
1298 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1299 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1300 et_dyn_addr == ET_DYN_ADDR_RAND) {
1301 imgp->map_flags |= MAP_ASLR;
1303 * If user does not care about sbrk, utilize the bss
1304 * grow region for mappings as well. We can select
1305 * the base for the image anywere and still not suffer
1306 * from the fragmentation.
1308 if (!__elfN(aslr_honor_sbrk) ||
1309 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1310 imgp->map_flags |= MAP_ASLR_IGNSTART;
1311 if (__elfN(aslr_stack))
1312 imgp->map_flags |= MAP_ASLR_STACK;
1313 if (__elfN(aslr_shared_page))
1314 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE;
1317 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1318 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1319 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1320 imgp->map_flags |= MAP_WXORX;
1322 error = exec_new_vmspace(imgp, sv);
1324 imgp->proc->p_sysent = sv;
1325 imgp->proc->p_elf_brandinfo = brand_info;
1327 vmspace = imgp->proc->p_vmspace;
1328 map = &vmspace->vm_map;
1329 maxv = sv->sv_usrstack;
1330 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1331 maxv -= lim_max(td, RLIMIT_STACK);
1332 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1333 uprintf("Excessive mapping size\n");
1337 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1338 KASSERT((map->flags & MAP_ASLR) != 0,
1339 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1340 error = __CONCAT(rnd_, __elfN(base))(map,
1341 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1342 /* reserve half of the address space to interpreter */
1343 maxv / 2, maxalign, &et_dyn_addr);
1346 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1350 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1354 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1359 * We load the dynamic linker where a userland call
1360 * to mmap(0, ...) would put it. The rationale behind this
1361 * calculation is that it leaves room for the heap to grow to
1362 * its maximum allowed size.
1364 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1366 if ((map->flags & MAP_ASLR) != 0) {
1367 maxv1 = maxv / 2 + addr / 2;
1368 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1369 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1370 pagesizes[1] : pagesizes[0], &anon_loc);
1373 map->anon_loc = anon_loc;
1375 map->anon_loc = addr;
1378 entry = (u_long)hdr->e_entry + et_dyn_addr;
1379 imgp->entry_addr = entry;
1381 if (interp != NULL) {
1382 VOP_UNLOCK(imgp->vp);
1383 if ((map->flags & MAP_ASLR) != 0) {
1384 /* Assume that interpreter fits into 1/4 of AS */
1385 maxv1 = maxv / 2 + addr / 2;
1386 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1387 maxv1, PAGE_SIZE, &addr);
1390 error = __elfN(load_interp)(imgp, brand_info, interp,
1391 &addr, &imgp->entry_addr);
1393 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1399 error = exec_map_stack(imgp);
1404 * Construct auxargs table (used by the copyout_auxargs routine)
1406 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1407 if (elf_auxargs == NULL) {
1408 VOP_UNLOCK(imgp->vp);
1409 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1410 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1412 elf_auxargs->execfd = -1;
1413 elf_auxargs->phdr = proghdr + et_dyn_addr;
1414 elf_auxargs->phent = hdr->e_phentsize;
1415 elf_auxargs->phnum = hdr->e_phnum;
1416 elf_auxargs->pagesz = PAGE_SIZE;
1417 elf_auxargs->base = addr;
1418 elf_auxargs->flags = 0;
1419 elf_auxargs->entry = entry;
1420 elf_auxargs->hdr_eflags = hdr->e_flags;
1422 imgp->auxargs = elf_auxargs;
1423 imgp->interpreted = 0;
1424 imgp->reloc_base = addr;
1425 imgp->proc->p_osrel = osrel;
1426 imgp->proc->p_fctl0 = fctl0;
1427 imgp->proc->p_elf_flags = hdr->e_flags;
1430 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1432 free(interp, M_TEMP);
1436 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1439 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1441 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1442 Elf_Auxinfo *argarray, *pos;
1443 struct vmspace *vmspace;
1446 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1449 vmspace = imgp->proc->p_vmspace;
1451 if (args->execfd != -1)
1452 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1453 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1454 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1455 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1456 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1457 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1458 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1459 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1460 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1461 if (imgp->execpathp != 0)
1462 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1463 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1464 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1465 if (imgp->canary != 0) {
1466 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1467 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1469 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1470 if (imgp->pagesizes != 0) {
1471 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1472 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1474 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) {
1475 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1476 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset);
1478 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1479 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1480 imgp->sysent->sv_stackprot);
1481 if (imgp->sysent->sv_hwcap != NULL)
1482 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1483 if (imgp->sysent->sv_hwcap2 != NULL)
1484 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1485 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1486 ELF_BSDF_SIGFASTBLK : 0);
1487 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1488 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1489 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1490 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1491 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1492 #ifdef RANDOM_FENESTRASX
1493 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) {
1494 AUXARGS_ENTRY(pos, AT_FXRNG,
1495 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset);
1498 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) {
1499 AUXARGS_ENTRY(pos, AT_KPRELOAD,
1500 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset);
1502 AUXARGS_ENTRY(pos, AT_NULL, 0);
1504 free(imgp->auxargs, M_TEMP);
1505 imgp->auxargs = NULL;
1506 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1508 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1509 free(argarray, M_TEMP);
1514 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1518 base = (Elf_Addr *)*stack_base;
1520 if (elf_suword(base, imgp->args->argc) == -1)
1522 *stack_base = (uintptr_t)base;
1527 * Code for generating ELF core dumps.
1530 typedef void (*segment_callback)(vm_map_entry_t, void *);
1532 /* Closure for cb_put_phdr(). */
1533 struct phdr_closure {
1534 Elf_Phdr *phdr; /* Program header to fill in */
1535 Elf_Off offset; /* Offset of segment in core file */
1539 int type; /* Note type. */
1540 struct regset *regset; /* Register set. */
1541 outfunc_t outfunc; /* Output function. */
1542 void *outarg; /* Argument for the output function. */
1543 size_t outsize; /* Output size. */
1544 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1547 TAILQ_HEAD(note_info_list, note_info);
1549 extern int compress_user_cores;
1550 extern int compress_user_cores_level;
1552 static void cb_put_phdr(vm_map_entry_t, void *);
1553 static void cb_size_segment(vm_map_entry_t, void *);
1554 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1556 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1557 struct note_info_list *, size_t, int);
1558 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1560 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1561 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1562 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1563 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1564 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1565 static void note_procstat_files(void *, struct sbuf *, size_t *);
1566 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1567 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1568 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1569 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1570 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1573 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1576 return (core_write((struct coredump_params *)arg, base, len, offset,
1577 UIO_SYSSPACE, NULL));
1581 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1583 struct ucred *cred = td->td_ucred;
1584 int compm, error = 0;
1585 struct sseg_closure seginfo;
1586 struct note_info_list notelst;
1587 struct coredump_params params;
1588 struct note_info *ninfo;
1590 size_t hdrsize, notesz, coresize;
1594 TAILQ_INIT(¬elst);
1596 /* Size the program segments. */
1597 __elfN(size_segments)(td, &seginfo, flags);
1600 * Collect info about the core file header area.
1602 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1603 if (seginfo.count + 1 >= PN_XNUM)
1604 hdrsize += sizeof(Elf_Shdr);
1605 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1606 coresize = round_page(hdrsize + notesz) + seginfo.size;
1608 /* Set up core dump parameters. */
1610 params.active_cred = cred;
1611 params.file_cred = NOCRED;
1618 PROC_LOCK(td->td_proc);
1619 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1620 PROC_UNLOCK(td->td_proc);
1627 if (coresize >= limit) {
1632 /* Create a compression stream if necessary. */
1633 compm = compress_user_cores;
1634 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1636 compm = COMPRESS_GZIP;
1638 params.comp = compressor_init(core_compressed_write,
1639 compm, CORE_BUF_SIZE,
1640 compress_user_cores_level, ¶ms);
1641 if (params.comp == NULL) {
1645 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1649 * Allocate memory for building the header, fill it up,
1650 * and write it out following the notes.
1652 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1653 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1656 /* Write the contents of all of the writable segments. */
1662 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1663 offset = round_page(hdrsize + notesz);
1664 for (i = 0; i < seginfo.count; i++) {
1665 error = core_output((char *)(uintptr_t)php->p_vaddr,
1666 php->p_filesz, offset, ¶ms, tmpbuf);
1669 offset += php->p_filesz;
1672 if (error == 0 && params.comp != NULL)
1673 error = compressor_flush(params.comp);
1677 "Failed to write core file for process %s (error %d)\n",
1678 curproc->p_comm, error);
1682 free(tmpbuf, M_TEMP);
1683 if (params.comp != NULL)
1684 compressor_fini(params.comp);
1685 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1686 TAILQ_REMOVE(¬elst, ninfo, link);
1687 free(ninfo, M_TEMP);
1696 * A callback for each_dumpable_segment() to write out the segment's
1697 * program header entry.
1700 cb_put_phdr(vm_map_entry_t entry, void *closure)
1702 struct phdr_closure *phc = (struct phdr_closure *)closure;
1703 Elf_Phdr *phdr = phc->phdr;
1705 phc->offset = round_page(phc->offset);
1707 phdr->p_type = PT_LOAD;
1708 phdr->p_offset = phc->offset;
1709 phdr->p_vaddr = entry->start;
1711 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1712 phdr->p_align = PAGE_SIZE;
1713 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1715 phc->offset += phdr->p_filesz;
1720 * A callback for each_dumpable_segment() to gather information about
1721 * the number of segments and their total size.
1724 cb_size_segment(vm_map_entry_t entry, void *closure)
1726 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1729 ssc->size += entry->end - entry->start;
1733 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1739 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1743 * For each writable segment in the process's memory map, call the given
1744 * function with a pointer to the map entry and some arbitrary
1745 * caller-supplied data.
1748 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1751 struct proc *p = td->td_proc;
1752 vm_map_t map = &p->p_vmspace->vm_map;
1753 vm_map_entry_t entry;
1754 vm_object_t backing_object, object;
1757 vm_map_lock_read(map);
1758 VM_MAP_ENTRY_FOREACH(entry, map) {
1760 * Don't dump inaccessible mappings, deal with legacy
1763 * Note that read-only segments related to the elf binary
1764 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1765 * need to arbitrarily ignore such segments.
1767 if ((flags & SVC_ALL) == 0) {
1768 if (elf_legacy_coredump) {
1769 if ((entry->protection & VM_PROT_RW) !=
1773 if ((entry->protection & VM_PROT_ALL) == 0)
1779 * Dont include memory segment in the coredump if
1780 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1781 * madvise(2). Do not dump submaps (i.e. parts of the
1784 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1786 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1787 (flags & SVC_ALL) == 0)
1789 if ((object = entry->object.vm_object) == NULL)
1792 /* Ignore memory-mapped devices and such things. */
1793 VM_OBJECT_RLOCK(object);
1794 while ((backing_object = object->backing_object) != NULL) {
1795 VM_OBJECT_RLOCK(backing_object);
1796 VM_OBJECT_RUNLOCK(object);
1797 object = backing_object;
1799 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1800 VM_OBJECT_RUNLOCK(object);
1804 (*func)(entry, closure);
1806 vm_map_unlock_read(map);
1810 * Write the core file header to the file, including padding up to
1811 * the page boundary.
1814 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1815 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1818 struct note_info *ninfo;
1822 /* Fill in the header. */
1823 bzero(hdr, hdrsize);
1824 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1826 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1827 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1828 sbuf_start_section(sb, NULL);
1829 sbuf_bcat(sb, hdr, hdrsize);
1830 TAILQ_FOREACH(ninfo, notelst, link)
1831 __elfN(putnote)(p->td, ninfo, sb);
1832 /* Align up to a page boundary for the program segments. */
1833 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1834 error = sbuf_finish(sb);
1841 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1851 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1852 __elfN(note_prpsinfo), p);
1855 * To have the debugger select the right thread (LWP) as the initial
1856 * thread, we dump the state of the thread passed to us in td first.
1857 * This is the thread that causes the core dump and thus likely to
1858 * be the right thread one wants to have selected in the debugger.
1861 while (thr != NULL) {
1862 size += __elfN(prepare_register_notes)(td, list, thr);
1863 size += __elfN(register_note)(td, list, -1,
1864 __elfN(note_threadmd), thr);
1866 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1867 TAILQ_NEXT(thr, td_plist);
1869 thr = TAILQ_NEXT(thr, td_plist);
1872 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1873 __elfN(note_procstat_proc), p);
1874 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1875 note_procstat_files, p);
1876 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1877 note_procstat_vmmap, p);
1878 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1879 note_procstat_groups, p);
1880 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1881 note_procstat_umask, p);
1882 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1883 note_procstat_rlimit, p);
1884 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1885 note_procstat_osrel, p);
1886 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1887 __elfN(note_procstat_psstrings), p);
1888 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1889 __elfN(note_procstat_auxv), p);
1895 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1896 size_t notesz, int flags)
1901 struct phdr_closure phc;
1904 ehdr = (Elf_Ehdr *)hdr;
1905 bi = td->td_proc->p_elf_brandinfo;
1907 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1908 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1909 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1910 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1911 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1912 ehdr->e_ident[EI_DATA] = ELF_DATA;
1913 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1914 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1915 ehdr->e_ident[EI_ABIVERSION] = 0;
1916 ehdr->e_ident[EI_PAD] = 0;
1917 ehdr->e_type = ET_CORE;
1918 ehdr->e_machine = bi->machine;
1919 ehdr->e_version = EV_CURRENT;
1921 ehdr->e_phoff = sizeof(Elf_Ehdr);
1922 ehdr->e_flags = td->td_proc->p_elf_flags;
1923 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1924 ehdr->e_phentsize = sizeof(Elf_Phdr);
1925 ehdr->e_shentsize = sizeof(Elf_Shdr);
1926 ehdr->e_shstrndx = SHN_UNDEF;
1927 if (numsegs + 1 < PN_XNUM) {
1928 ehdr->e_phnum = numsegs + 1;
1931 ehdr->e_phnum = PN_XNUM;
1934 ehdr->e_shoff = ehdr->e_phoff +
1935 (numsegs + 1) * ehdr->e_phentsize;
1936 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1937 ("e_shoff: %zu, hdrsize - shdr: %zu",
1938 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1940 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1941 memset(shdr, 0, sizeof(*shdr));
1943 * A special first section is used to hold large segment and
1944 * section counts. This was proposed by Sun Microsystems in
1945 * Solaris and has been adopted by Linux; the standard ELF
1946 * tools are already familiar with the technique.
1948 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1949 * (or 12-7 depending on the version of the document) for more
1952 shdr->sh_type = SHT_NULL;
1953 shdr->sh_size = ehdr->e_shnum;
1954 shdr->sh_link = ehdr->e_shstrndx;
1955 shdr->sh_info = numsegs + 1;
1959 * Fill in the program header entries.
1961 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1963 /* The note segement. */
1964 phdr->p_type = PT_NOTE;
1965 phdr->p_offset = hdrsize;
1968 phdr->p_filesz = notesz;
1970 phdr->p_flags = PF_R;
1971 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1974 /* All the writable segments from the program. */
1976 phc.offset = round_page(hdrsize + notesz);
1977 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1981 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1982 struct regset *regset, struct thread *target_td)
1984 const struct sysentvec *sv;
1985 struct note_info *ninfo;
1986 size_t size, notesize;
1989 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1992 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1993 ninfo->type = regset->note;
1994 ninfo->regset = regset;
1995 ninfo->outarg = target_td;
1996 ninfo->outsize = size;
1997 TAILQ_INSERT_TAIL(list, ninfo, link);
1999 sv = td->td_proc->p_sysent;
2000 notesize = sizeof(Elf_Note) + /* note header */
2001 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2003 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2009 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2010 int type, outfunc_t out, void *arg)
2012 const struct sysentvec *sv;
2013 struct note_info *ninfo;
2014 size_t size, notesize;
2016 sv = td->td_proc->p_sysent;
2018 out(arg, NULL, &size);
2019 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2021 ninfo->outfunc = out;
2022 ninfo->outarg = arg;
2023 ninfo->outsize = size;
2024 TAILQ_INSERT_TAIL(list, ninfo, link);
2029 notesize = sizeof(Elf_Note) + /* note header */
2030 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2032 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2038 append_note_data(const void *src, void *dst, size_t len)
2042 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2044 bcopy(src, dst, len);
2045 bzero((char *)dst + len, padded_len - len);
2047 return (padded_len);
2051 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2059 note = (Elf_Note *)buf;
2060 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2061 note->n_descsz = size;
2062 note->n_type = type;
2063 buf += sizeof(*note);
2064 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2065 sizeof(FREEBSD_ABI_VENDOR));
2066 append_note_data(src, buf, size);
2071 notesize = sizeof(Elf_Note) + /* note header */
2072 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2074 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2080 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2083 const struct sysentvec *sv;
2084 ssize_t old_len, sect_len;
2085 size_t new_len, descsz, i;
2087 if (ninfo->type == -1) {
2088 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2092 sv = td->td_proc->p_sysent;
2094 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2095 note.n_descsz = ninfo->outsize;
2096 note.n_type = ninfo->type;
2098 sbuf_bcat(sb, ¬e, sizeof(note));
2099 sbuf_start_section(sb, &old_len);
2100 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2101 strlen(sv->sv_elf_core_abi_vendor) + 1);
2102 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2103 if (note.n_descsz == 0)
2105 sbuf_start_section(sb, &old_len);
2106 if (ninfo->regset != NULL) {
2107 struct regset *regset = ninfo->regset;
2110 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2111 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2112 sbuf_bcat(sb, buf, ninfo->outsize);
2115 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2116 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2120 new_len = (size_t)sect_len;
2121 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2122 if (new_len < descsz) {
2124 * It is expected that individual note emitters will correctly
2125 * predict their expected output size and fill up to that size
2126 * themselves, padding in a format-specific way if needed.
2127 * However, in case they don't, just do it here with zeros.
2129 for (i = 0; i < descsz - new_len; i++)
2131 } else if (new_len > descsz) {
2133 * We can't always truncate sb -- we may have drained some
2136 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2137 "read it (%zu > %zu). Since it is longer than "
2138 "expected, this coredump's notes are corrupt. THIS "
2139 "IS A BUG in the note_procstat routine for type %u.\n",
2140 __func__, (unsigned)note.n_type, new_len, descsz,
2141 (unsigned)note.n_type));
2146 * Miscellaneous note out functions.
2149 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2150 #include <compat/freebsd32/freebsd32.h>
2151 #include <compat/freebsd32/freebsd32_signal.h>
2153 typedef struct prstatus32 elf_prstatus_t;
2154 typedef struct prpsinfo32 elf_prpsinfo_t;
2155 typedef struct fpreg32 elf_prfpregset_t;
2156 typedef struct fpreg32 elf_fpregset_t;
2157 typedef struct reg32 elf_gregset_t;
2158 typedef struct thrmisc32 elf_thrmisc_t;
2159 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2160 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2161 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2162 typedef uint32_t elf_ps_strings_t;
2164 typedef prstatus_t elf_prstatus_t;
2165 typedef prpsinfo_t elf_prpsinfo_t;
2166 typedef prfpregset_t elf_prfpregset_t;
2167 typedef prfpregset_t elf_fpregset_t;
2168 typedef gregset_t elf_gregset_t;
2169 typedef thrmisc_t elf_thrmisc_t;
2170 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2171 #define ELF_KERN_PROC_MASK 0
2172 typedef struct kinfo_proc elf_kinfo_proc_t;
2173 typedef vm_offset_t elf_ps_strings_t;
2177 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2183 elf_prpsinfo_t *psinfo;
2188 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2189 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2190 psinfo->pr_version = PRPSINFO_VERSION;
2191 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2192 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2194 if (p->p_args != NULL) {
2195 len = sizeof(psinfo->pr_psargs) - 1;
2196 if (len > p->p_args->ar_length)
2197 len = p->p_args->ar_length;
2198 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2204 sbuf_new(&sbarg, psinfo->pr_psargs,
2205 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2206 error = proc_getargv(curthread, p, &sbarg);
2208 if (sbuf_finish(&sbarg) == 0) {
2209 len = sbuf_len(&sbarg);
2213 len = sizeof(psinfo->pr_psargs) - 1;
2215 sbuf_delete(&sbarg);
2217 if (error != 0 || len == 0 || (ssize_t)len == -1)
2218 strlcpy(psinfo->pr_psargs, p->p_comm,
2219 sizeof(psinfo->pr_psargs));
2221 KASSERT(len < sizeof(psinfo->pr_psargs),
2222 ("len is too long: %zu vs %zu", len,
2223 sizeof(psinfo->pr_psargs)));
2224 cp = psinfo->pr_psargs;
2227 cp = memchr(cp, '\0', end - cp);
2233 psinfo->pr_pid = p->p_pid;
2234 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2235 free(psinfo, M_TEMP);
2237 *sizep = sizeof(*psinfo);
2241 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2244 elf_prstatus_t *status;
2247 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2250 memset(status, 0, *sizep);
2251 status->pr_version = PRSTATUS_VERSION;
2252 status->pr_statussz = sizeof(elf_prstatus_t);
2253 status->pr_gregsetsz = sizeof(elf_gregset_t);
2254 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2255 status->pr_osreldate = osreldate;
2256 status->pr_cursig = td->td_proc->p_sig;
2257 status->pr_pid = td->td_tid;
2258 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2259 fill_regs32(td, &status->pr_reg);
2261 fill_regs(td, &status->pr_reg);
2264 *sizep = sizeof(*status);
2269 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2272 elf_prstatus_t *status;
2274 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2276 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2277 set_regs32(td, &status->pr_reg);
2279 set_regs(td, &status->pr_reg);
2284 static struct regset __elfN(regset_prstatus) = {
2285 .note = NT_PRSTATUS,
2286 .size = sizeof(elf_prstatus_t),
2287 .get = __elfN(get_prstatus),
2288 .set = __elfN(set_prstatus),
2290 ELF_REGSET(__elfN(regset_prstatus));
2293 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2296 elf_prfpregset_t *fpregset;
2299 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2302 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2303 fill_fpregs32(td, fpregset);
2305 fill_fpregs(td, fpregset);
2308 *sizep = sizeof(*fpregset);
2313 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2316 elf_prfpregset_t *fpregset;
2319 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2320 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2321 set_fpregs32(td, fpregset);
2323 set_fpregs(td, fpregset);
2328 static struct regset __elfN(regset_fpregset) = {
2329 .note = NT_FPREGSET,
2330 .size = sizeof(elf_prfpregset_t),
2331 .get = __elfN(get_fpregset),
2332 .set = __elfN(set_fpregset),
2334 ELF_REGSET(__elfN(regset_fpregset));
2337 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2340 elf_thrmisc_t *thrmisc;
2343 KASSERT(*sizep == sizeof(*thrmisc),
2344 ("%s: invalid size", __func__));
2346 bzero(thrmisc, sizeof(*thrmisc));
2347 strcpy(thrmisc->pr_tname, td->td_name);
2349 *sizep = sizeof(*thrmisc);
2353 static struct regset __elfN(regset_thrmisc) = {
2355 .size = sizeof(elf_thrmisc_t),
2356 .get = __elfN(get_thrmisc),
2358 ELF_REGSET(__elfN(regset_thrmisc));
2361 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2368 size = sizeof(structsize) + sizeof(pl);
2370 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2371 structsize = sizeof(pl);
2372 memcpy(buf, &structsize, sizeof(structsize));
2373 bzero(&pl, sizeof(pl));
2374 pl.pl_lwpid = td->td_tid;
2375 pl.pl_event = PL_EVENT_NONE;
2376 pl.pl_sigmask = td->td_sigmask;
2377 pl.pl_siglist = td->td_siglist;
2378 if (td->td_si.si_signo != 0) {
2379 pl.pl_event = PL_EVENT_SIGNAL;
2380 pl.pl_flags |= PL_FLAG_SI;
2381 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2382 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2384 pl.pl_siginfo = td->td_si;
2387 strcpy(pl.pl_tdname, td->td_name);
2388 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2389 memcpy((int *)buf + 1, &pl, sizeof(pl));
2395 static struct regset __elfN(regset_lwpinfo) = {
2396 .note = NT_PTLWPINFO,
2397 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2398 .get = __elfN(get_lwpinfo),
2400 ELF_REGSET(__elfN(regset_lwpinfo));
2403 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2404 struct thread *target_td)
2406 struct sysentvec *sv = td->td_proc->p_sysent;
2407 struct regset **regsetp, **regset_end, *regset;
2412 /* NT_PRSTATUS must be the first register set note. */
2413 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2416 regsetp = sv->sv_regset_begin;
2417 if (regsetp == NULL) {
2418 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2419 size += __elfN(register_regset_note)(td, list,
2420 &__elfN(regset_fpregset), target_td);
2423 regset_end = sv->sv_regset_end;
2424 MPASS(regset_end != NULL);
2425 for (; regsetp < regset_end; regsetp++) {
2427 if (regset->note == NT_PRSTATUS)
2429 size += __elfN(register_regset_note)(td, list, regset,
2436 * Allow for MD specific notes, as well as any MD
2437 * specific preparations for writing MI notes.
2440 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2446 td = (struct thread *)arg;
2448 if (size != 0 && sb != NULL)
2449 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2453 __elfN(dump_thread)(td, buf, &size);
2454 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2455 if (size != 0 && sb != NULL)
2456 sbuf_bcat(sb, buf, size);
2461 #ifdef KINFO_PROC_SIZE
2462 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2466 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2473 size = sizeof(structsize) + p->p_numthreads *
2474 sizeof(elf_kinfo_proc_t);
2477 KASSERT(*sizep == size, ("invalid size"));
2478 structsize = sizeof(elf_kinfo_proc_t);
2479 sbuf_bcat(sb, &structsize, sizeof(structsize));
2480 sx_slock(&proctree_lock);
2482 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2483 sx_sunlock(&proctree_lock);
2488 #ifdef KINFO_FILE_SIZE
2489 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2493 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2496 size_t size, sect_sz, i;
2497 ssize_t start_len, sect_len;
2498 int structsize, filedesc_flags;
2500 if (coredump_pack_fileinfo)
2501 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2506 structsize = sizeof(struct kinfo_file);
2509 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2510 sbuf_set_drain(sb, sbuf_count_drain, &size);
2511 sbuf_bcat(sb, &structsize, sizeof(structsize));
2513 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2518 sbuf_start_section(sb, &start_len);
2520 sbuf_bcat(sb, &structsize, sizeof(structsize));
2522 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2525 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2530 KASSERT(sect_sz <= *sizep,
2531 ("kern_proc_filedesc_out did not respect maxlen; "
2532 "requested %zu, got %zu", *sizep - sizeof(structsize),
2533 sect_sz - sizeof(structsize)));
2535 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2540 #ifdef KINFO_VMENTRY_SIZE
2541 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2545 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2549 int structsize, vmmap_flags;
2551 if (coredump_pack_vmmapinfo)
2552 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2557 structsize = sizeof(struct kinfo_vmentry);
2560 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2561 sbuf_set_drain(sb, sbuf_count_drain, &size);
2562 sbuf_bcat(sb, &structsize, sizeof(structsize));
2564 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2569 sbuf_bcat(sb, &structsize, sizeof(structsize));
2571 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2577 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2584 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2586 KASSERT(*sizep == size, ("invalid size"));
2587 structsize = sizeof(gid_t);
2588 sbuf_bcat(sb, &structsize, sizeof(structsize));
2589 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2596 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2603 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2605 KASSERT(*sizep == size, ("invalid size"));
2606 structsize = sizeof(p->p_pd->pd_cmask);
2607 sbuf_bcat(sb, &structsize, sizeof(structsize));
2608 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2614 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2617 struct rlimit rlim[RLIM_NLIMITS];
2622 size = sizeof(structsize) + sizeof(rlim);
2624 KASSERT(*sizep == size, ("invalid size"));
2625 structsize = sizeof(rlim);
2626 sbuf_bcat(sb, &structsize, sizeof(structsize));
2628 for (i = 0; i < RLIM_NLIMITS; i++)
2629 lim_rlimit_proc(p, i, &rlim[i]);
2631 sbuf_bcat(sb, rlim, sizeof(rlim));
2637 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2644 size = sizeof(structsize) + sizeof(p->p_osrel);
2646 KASSERT(*sizep == size, ("invalid size"));
2647 structsize = sizeof(p->p_osrel);
2648 sbuf_bcat(sb, &structsize, sizeof(structsize));
2649 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2655 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2658 elf_ps_strings_t ps_strings;
2663 size = sizeof(structsize) + sizeof(ps_strings);
2665 KASSERT(*sizep == size, ("invalid size"));
2666 structsize = sizeof(ps_strings);
2667 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2668 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2670 ps_strings = PROC_PS_STRINGS(p);
2672 sbuf_bcat(sb, &structsize, sizeof(structsize));
2673 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2679 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2688 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2690 sbuf_set_drain(sb, sbuf_count_drain, &size);
2691 sbuf_bcat(sb, &structsize, sizeof(structsize));
2693 proc_getauxv(curthread, p, sb);
2699 structsize = sizeof(Elf_Auxinfo);
2700 sbuf_bcat(sb, &structsize, sizeof(structsize));
2702 proc_getauxv(curthread, p, sb);
2708 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2709 const char *note_vendor, const Elf_Phdr *pnote,
2710 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2712 const Elf_Note *note, *note0, *note_end;
2713 const char *note_name;
2718 /* We need some limit, might as well use PAGE_SIZE. */
2719 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2721 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2722 if (pnote->p_offset > PAGE_SIZE ||
2723 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2724 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2726 VOP_UNLOCK(imgp->vp);
2727 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2728 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2730 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2731 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2732 curthread->td_ucred, NOCRED, NULL, curthread);
2734 uprintf("i/o error PT_NOTE\n");
2737 note = note0 = (const Elf_Note *)buf;
2738 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2740 note = note0 = (const Elf_Note *)(imgp->image_header +
2742 note_end = (const Elf_Note *)(imgp->image_header +
2743 pnote->p_offset + pnote->p_filesz);
2746 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2747 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2748 (const char *)note < sizeof(Elf_Note)) {
2751 if (note->n_namesz != checknote->n_namesz ||
2752 note->n_descsz != checknote->n_descsz ||
2753 note->n_type != checknote->n_type)
2755 note_name = (const char *)(note + 1);
2756 if (note_name + checknote->n_namesz >=
2757 (const char *)note_end || strncmp(note_vendor,
2758 note_name, checknote->n_namesz) != 0)
2761 if (cb(note, cb_arg, &res))
2764 note = (const Elf_Note *)((const char *)(note + 1) +
2765 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2766 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2775 struct brandnote_cb_arg {
2776 Elf_Brandnote *brandnote;
2781 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2783 struct brandnote_cb_arg *arg;
2788 * Fetch the osreldate for binary from the ELF OSABI-note if
2791 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2792 arg->brandnote->trans_osrel != NULL ?
2793 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2798 static Elf_Note fctl_note = {
2799 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2800 .n_descsz = sizeof(uint32_t),
2801 .n_type = NT_FREEBSD_FEATURE_CTL,
2804 struct fctl_cb_arg {
2810 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2812 struct fctl_cb_arg *arg;
2813 const Elf32_Word *desc;
2817 p = (uintptr_t)(note + 1);
2818 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2819 desc = (const Elf32_Word *)p;
2820 *arg->has_fctl0 = true;
2821 *arg->fctl0 = desc[0];
2827 * Try to find the appropriate ABI-note section for checknote, fetch
2828 * the osreldate and feature control flags for binary from the ELF
2829 * OSABI-note. Only the first page of the image is searched, the same
2833 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2834 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2836 const Elf_Phdr *phdr;
2837 const Elf_Ehdr *hdr;
2838 struct brandnote_cb_arg b_arg;
2839 struct fctl_cb_arg f_arg;
2842 hdr = (const Elf_Ehdr *)imgp->image_header;
2843 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2844 b_arg.brandnote = brandnote;
2845 b_arg.osrel = osrel;
2846 f_arg.has_fctl0 = has_fctl0;
2847 f_arg.fctl0 = fctl0;
2849 for (i = 0; i < hdr->e_phnum; i++) {
2850 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2851 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2853 for (j = 0; j < hdr->e_phnum; j++) {
2854 if (phdr[j].p_type == PT_NOTE &&
2855 __elfN(parse_notes)(imgp, &fctl_note,
2856 FREEBSD_ABI_VENDOR, &phdr[j],
2857 note_fctl_cb, &f_arg))
2868 * Tell kern_execve.c about it, with a little help from the linker.
2870 static struct execsw __elfN(execsw) = {
2871 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2872 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2874 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2877 __elfN(trans_prot)(Elf_Word flags)
2883 prot |= VM_PROT_EXECUTE;
2885 prot |= VM_PROT_WRITE;
2887 prot |= VM_PROT_READ;
2888 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2889 if (i386_read_exec && (flags & PF_R))
2890 prot |= VM_PROT_EXECUTE;
2896 __elfN(untrans_prot)(vm_prot_t prot)
2901 if (prot & VM_PROT_EXECUTE)
2903 if (prot & VM_PROT_READ)
2905 if (prot & VM_PROT_WRITE)