2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static bool __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 static size_t __elfN(prepare_register_notes)(struct thread *td,
106 struct note_info_list *list, struct thread *target_td);
108 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
109 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119 &elf_legacy_coredump, 0,
120 "include all and only RW pages in core dumps");
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
134 #if defined(__amd64__)
135 static int __elfN(vdso) = 1;
136 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
137 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
138 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
140 static int __elfN(vdso) = 0;
143 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
144 int i386_read_exec = 0;
145 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
146 "enable execution from readable segments");
149 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
151 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
156 val = __elfN(pie_base);
157 error = sysctl_handle_long(oidp, &val, 0, req);
158 if (error != 0 || req->newptr == NULL)
160 if ((val & PAGE_MASK) != 0)
162 __elfN(pie_base) = val;
165 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
166 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
167 sysctl_pie_base, "LU",
168 "PIE load base without randomization");
170 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
176 * While for 64-bit machines ASLR works properly, there are
177 * still some problems when using 32-bit architectures. For this
178 * reason ASLR is only enabled by default when running native
179 * 64-bit non-PIE executables.
181 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
182 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
183 &__elfN(aslr_enabled), 0,
184 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
185 ": enable address map randomization");
188 * Enable ASLR only for 64-bit PIE binaries by default.
190 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
191 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
192 &__elfN(pie_aslr_enabled), 0,
193 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
194 ": enable address map randomization for PIE binaries");
197 * Sbrk is now deprecated and it can be assumed, that in most
198 * cases it will not be used anyway. This setting is valid only
199 * for the ASLR enabled and allows for utilizing the bss grow region.
201 static int __elfN(aslr_honor_sbrk) = 0;
202 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
203 &__elfN(aslr_honor_sbrk), 0,
204 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
206 static int __elfN(aslr_stack) = 1;
207 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
208 &__elfN(aslr_stack), 0,
209 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
210 ": enable stack address randomization");
212 static int __elfN(sigfastblock) = 1;
213 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
214 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
215 "enable sigfastblock for new processes");
217 static bool __elfN(allow_wx) = true;
218 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
219 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
220 "Allow pages to be mapped simultaneously writable and executable");
222 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
224 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
226 Elf_Brandnote __elfN(freebsd_brandnote) = {
227 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
228 .hdr.n_descsz = sizeof(int32_t),
229 .hdr.n_type = NT_FREEBSD_ABI_TAG,
230 .vendor = FREEBSD_ABI_VENDOR,
231 .flags = BN_TRANSLATE_OSREL,
232 .trans_osrel = __elfN(freebsd_trans_osrel)
236 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
240 p = (uintptr_t)(note + 1);
241 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
242 *osrel = *(const int32_t *)(p);
247 static const char GNU_ABI_VENDOR[] = "GNU";
248 static int GNU_KFREEBSD_ABI_DESC = 3;
250 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
251 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
252 .hdr.n_descsz = 16, /* XXX at least 16 */
254 .vendor = GNU_ABI_VENDOR,
255 .flags = BN_TRANSLATE_OSREL,
256 .trans_osrel = kfreebsd_trans_osrel
260 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
262 const Elf32_Word *desc;
265 p = (uintptr_t)(note + 1);
266 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
268 desc = (const Elf32_Word *)p;
269 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
273 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
274 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
276 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
282 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
286 for (i = 0; i < MAX_BRANDS; i++) {
287 if (elf_brand_list[i] == NULL) {
288 elf_brand_list[i] = entry;
292 if (i == MAX_BRANDS) {
293 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
301 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
305 for (i = 0; i < MAX_BRANDS; i++) {
306 if (elf_brand_list[i] == entry) {
307 elf_brand_list[i] = NULL;
317 __elfN(brand_inuse)(Elf_Brandinfo *entry)
322 sx_slock(&allproc_lock);
323 FOREACH_PROC_IN_SYSTEM(p) {
324 if (p->p_sysent == entry->sysvec) {
329 sx_sunlock(&allproc_lock);
334 static Elf_Brandinfo *
335 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
336 int32_t *osrel, uint32_t *fctl0)
338 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
339 Elf_Brandinfo *bi, *bi_m;
341 int i, interp_name_len;
343 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
346 * We support four types of branding -- (1) the ELF EI_OSABI field
347 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
348 * branding w/in the ELF header, (3) path of the `interp_path'
349 * field, and (4) the ".note.ABI-tag" ELF section.
352 /* Look for an ".note.ABI-tag" ELF section */
354 for (i = 0; i < MAX_BRANDS; i++) {
355 bi = elf_brand_list[i];
358 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
360 if (hdr->e_machine == bi->machine && (bi->flags &
361 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
365 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
367 /* Give brand a chance to veto check_note's guess */
368 if (ret && bi->header_supported) {
369 ret = bi->header_supported(imgp, osrel,
370 has_fctl0 ? fctl0 : NULL);
373 * If note checker claimed the binary, but the
374 * interpreter path in the image does not
375 * match default one for the brand, try to
376 * search for other brands with the same
377 * interpreter. Either there is better brand
378 * with the right interpreter, or, failing
379 * this, we return first brand which accepted
380 * our note and, optionally, header.
382 if (ret && bi_m == NULL && interp != NULL &&
383 (bi->interp_path == NULL ||
384 (strlen(bi->interp_path) + 1 != interp_name_len ||
385 strncmp(interp, bi->interp_path, interp_name_len)
397 /* If the executable has a brand, search for it in the brand list. */
398 for (i = 0; i < MAX_BRANDS; i++) {
399 bi = elf_brand_list[i];
400 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
401 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
403 if (hdr->e_machine == bi->machine &&
404 (hdr->e_ident[EI_OSABI] == bi->brand ||
405 (bi->compat_3_brand != NULL &&
406 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
407 bi->compat_3_brand) == 0))) {
408 /* Looks good, but give brand a chance to veto */
409 if (bi->header_supported == NULL ||
410 bi->header_supported(imgp, NULL, NULL)) {
412 * Again, prefer strictly matching
415 if (interp_name_len == 0 &&
416 bi->interp_path == NULL)
418 if (bi->interp_path != NULL &&
419 strlen(bi->interp_path) + 1 ==
420 interp_name_len && strncmp(interp,
421 bi->interp_path, interp_name_len) == 0)
431 /* No known brand, see if the header is recognized by any brand */
432 for (i = 0; i < MAX_BRANDS; i++) {
433 bi = elf_brand_list[i];
434 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
435 bi->header_supported == NULL)
437 if (hdr->e_machine == bi->machine) {
438 ret = bi->header_supported(imgp, NULL, NULL);
444 /* Lacking a known brand, search for a recognized interpreter. */
445 if (interp != NULL) {
446 for (i = 0; i < MAX_BRANDS; i++) {
447 bi = elf_brand_list[i];
448 if (bi == NULL || (bi->flags &
449 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
452 if (hdr->e_machine == bi->machine &&
453 bi->interp_path != NULL &&
454 /* ELF image p_filesz includes terminating zero */
455 strlen(bi->interp_path) + 1 == interp_name_len &&
456 strncmp(interp, bi->interp_path, interp_name_len)
457 == 0 && (bi->header_supported == NULL ||
458 bi->header_supported(imgp, NULL, NULL)))
463 /* Lacking a recognized interpreter, try the default brand */
464 for (i = 0; i < MAX_BRANDS; i++) {
465 bi = elf_brand_list[i];
466 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
467 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
469 if (hdr->e_machine == bi->machine &&
470 __elfN(fallback_brand) == bi->brand &&
471 (bi->header_supported == NULL ||
472 bi->header_supported(imgp, NULL, NULL)))
479 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
481 return (hdr->e_phoff <= PAGE_SIZE &&
482 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
486 __elfN(check_header)(const Elf_Ehdr *hdr)
492 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
493 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
494 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
495 hdr->e_phentsize != sizeof(Elf_Phdr) ||
496 hdr->e_version != ELF_TARG_VER)
500 * Make sure we have at least one brand for this machine.
503 for (i = 0; i < MAX_BRANDS; i++) {
504 bi = elf_brand_list[i];
505 if (bi != NULL && bi->machine == hdr->e_machine)
515 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
516 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
523 * Create the page if it doesn't exist yet. Ignore errors.
525 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
526 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
529 * Find the page from the underlying object.
531 if (object != NULL) {
532 sf = vm_imgact_map_page(object, offset);
534 return (KERN_FAILURE);
535 off = offset - trunc_page(offset);
536 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
538 vm_imgact_unmap_page(sf);
540 return (KERN_FAILURE);
543 return (KERN_SUCCESS);
547 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
548 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
554 int error, locked, rv;
556 if (start != trunc_page(start)) {
557 rv = __elfN(map_partial)(map, object, offset, start,
558 round_page(start), prot);
559 if (rv != KERN_SUCCESS)
561 offset += round_page(start) - start;
562 start = round_page(start);
564 if (end != round_page(end)) {
565 rv = __elfN(map_partial)(map, object, offset +
566 trunc_page(end) - start, trunc_page(end), end, prot);
567 if (rv != KERN_SUCCESS)
569 end = trunc_page(end);
572 return (KERN_SUCCESS);
573 if ((offset & PAGE_MASK) != 0) {
575 * The mapping is not page aligned. This means that we have
578 rv = vm_map_fixed(map, NULL, 0, start, end - start,
579 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
580 if (rv != KERN_SUCCESS)
583 return (KERN_SUCCESS);
584 for (; start < end; start += sz) {
585 sf = vm_imgact_map_page(object, offset);
587 return (KERN_FAILURE);
588 off = offset - trunc_page(offset);
590 if (sz > PAGE_SIZE - off)
591 sz = PAGE_SIZE - off;
592 error = copyout((caddr_t)sf_buf_kva(sf) + off,
594 vm_imgact_unmap_page(sf);
596 return (KERN_FAILURE);
600 vm_object_reference(object);
601 rv = vm_map_fixed(map, object, offset, start, end - start,
602 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
603 (object != NULL ? MAP_VN_EXEC : 0));
604 if (rv != KERN_SUCCESS) {
605 locked = VOP_ISLOCKED(imgp->vp);
606 VOP_UNLOCK(imgp->vp);
607 vm_object_deallocate(object);
608 vn_lock(imgp->vp, locked | LK_RETRY);
610 } else if (object != NULL) {
611 MPASS(imgp->vp->v_object == object);
612 VOP_SET_TEXT_CHECKED(imgp->vp);
615 return (KERN_SUCCESS);
619 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
620 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
626 vm_offset_t map_addr;
629 vm_ooffset_t file_addr;
632 * It's necessary to fail if the filsz + offset taken from the
633 * header is greater than the actual file pager object's size.
634 * If we were to allow this, then the vm_map_find() below would
635 * walk right off the end of the file object and into the ether.
637 * While I'm here, might as well check for something else that
638 * is invalid: filsz cannot be greater than memsz.
640 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
642 uprintf("elf_load_section: truncated ELF file\n");
646 object = imgp->object;
647 map = &imgp->proc->p_vmspace->vm_map;
648 map_addr = trunc_page((vm_offset_t)vmaddr);
649 file_addr = trunc_page(offset);
652 * We have two choices. We can either clear the data in the last page
653 * of an oversized mapping, or we can start the anon mapping a page
654 * early and copy the initialized data into that first page. We
659 else if (memsz > filsz)
660 map_len = trunc_page(offset + filsz) - file_addr;
662 map_len = round_page(offset + filsz) - file_addr;
665 /* cow flags: don't dump readonly sections in core */
666 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
667 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
669 rv = __elfN(map_insert)(imgp, map, object, file_addr,
670 map_addr, map_addr + map_len, prot, cow);
671 if (rv != KERN_SUCCESS)
674 /* we can stop now if we've covered it all */
680 * We have to get the remaining bit of the file into the first part
681 * of the oversized map segment. This is normally because the .data
682 * segment in the file is extended to provide bss. It's a neat idea
683 * to try and save a page, but it's a pain in the behind to implement.
685 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
687 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
688 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
690 /* This had damn well better be true! */
692 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
693 map_addr + map_len, prot, 0);
694 if (rv != KERN_SUCCESS)
699 sf = vm_imgact_map_page(object, offset + filsz);
703 /* send the page fragment to user space */
704 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
706 vm_imgact_unmap_page(sf);
712 * Remove write access to the page if it was only granted by map_insert
715 if ((prot & VM_PROT_WRITE) == 0)
716 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
717 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
723 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
724 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
731 ASSERT_VOP_LOCKED(imgp->vp, __func__);
736 for (i = 0; i < hdr->e_phnum; i++) {
737 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
740 /* Loadable segment */
741 prot = __elfN(trans_prot)(phdr[i].p_flags);
742 error = __elfN(load_section)(imgp, phdr[i].p_offset,
743 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
744 phdr[i].p_memsz, phdr[i].p_filesz, prot);
749 * Establish the base address if this is the first segment.
752 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
757 if (base_addrp != NULL)
758 *base_addrp = base_addr;
764 * Load the file "file" into memory. It may be either a shared object
767 * The "addr" reference parameter is in/out. On entry, it specifies
768 * the address where a shared object should be loaded. If the file is
769 * an executable, this value is ignored. On exit, "addr" specifies
770 * where the file was actually loaded.
772 * The "entry" reference parameter is out only. On exit, it specifies
773 * the entry point for the loaded file.
776 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
782 struct image_params image_params;
784 const Elf_Ehdr *hdr = NULL;
785 const Elf_Phdr *phdr = NULL;
786 struct nameidata *nd;
788 struct image_params *imgp;
790 u_long base_addr = 0;
793 #ifdef CAPABILITY_MODE
795 * XXXJA: This check can go away once we are sufficiently confident
796 * that the checks in namei() are correct.
798 if (IN_CAPABILITY_MODE(curthread))
802 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
804 attr = &tempdata->attr;
805 imgp = &tempdata->image_params;
808 * Initialize part of the common data
813 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
815 if ((error = namei(nd)) != 0) {
820 imgp->vp = nd->ni_vp;
823 * Check permissions, modes, uid, etc on the file, and "open" it.
825 error = exec_check_permissions(imgp);
829 error = exec_map_first_page(imgp);
833 imgp->object = nd->ni_vp->v_object;
835 hdr = (const Elf_Ehdr *)imgp->image_header;
836 if ((error = __elfN(check_header)(hdr)) != 0)
838 if (hdr->e_type == ET_DYN)
840 else if (hdr->e_type == ET_EXEC)
847 /* Only support headers that fit within first page for now */
848 if (!__elfN(phdr_in_zero_page)(hdr)) {
853 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
854 if (!aligned(phdr, Elf_Addr)) {
859 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
864 *entry = (unsigned long)hdr->e_entry + rbase;
868 exec_unmap_first_page(imgp);
872 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
875 free(tempdata, M_TEMP);
881 * Select randomized valid address in the map map, between minv and
882 * maxv, with specified alignment. The [minv, maxv) range must belong
883 * to the map. Note that function only allocates the address, it is
884 * up to caller to clamp maxv in a way that the final allocation
885 * length fit into the map.
887 * Result is returned in *resp, error code indicates that arguments
888 * did not pass sanity checks for overflow and range correctness.
891 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
892 u_int align, u_long *resp)
896 MPASS(vm_map_min(map) <= minv);
898 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
899 uprintf("Invalid ELF segments layout\n");
903 arc4rand(&rbase, sizeof(rbase), 0);
904 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
905 res &= ~((u_long)align - 1);
910 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
911 res, minv, maxv, rbase));
913 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
914 res, maxv, minv, rbase));
921 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
922 const Elf_Phdr *phdr, u_long et_dyn_addr)
924 struct vmspace *vmspace;
926 u_long text_size, data_size, total_size, text_addr, data_addr;
927 u_long seg_size, seg_addr;
931 text_size = data_size = total_size = text_addr = data_addr = 0;
933 for (i = 0; i < hdr->e_phnum; i++) {
934 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
937 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
938 seg_size = round_page(phdr[i].p_memsz +
939 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
942 * Make the largest executable segment the official
943 * text segment and all others data.
945 * Note that obreak() assumes that data_addr + data_size == end
946 * of data load area, and the ELF file format expects segments
947 * to be sorted by address. If multiple data segments exist,
948 * the last one will be used.
951 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
952 text_size = seg_size;
953 text_addr = seg_addr;
955 data_size = seg_size;
956 data_addr = seg_addr;
958 total_size += seg_size;
961 if (data_addr == 0 && data_size == 0) {
962 data_addr = text_addr;
963 data_size = text_size;
967 * Check limits. It should be safe to check the
968 * limits after loading the segments since we do
969 * not actually fault in all the segments pages.
971 PROC_LOCK(imgp->proc);
972 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
973 err_str = "Data segment size exceeds process limit";
974 else if (text_size > maxtsiz)
975 err_str = "Text segment size exceeds system limit";
976 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
977 err_str = "Total segment size exceeds process limit";
978 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
979 err_str = "Data segment size exceeds resource limit";
980 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
981 err_str = "Total segment size exceeds resource limit";
982 PROC_UNLOCK(imgp->proc);
983 if (err_str != NULL) {
984 uprintf("%s\n", err_str);
988 vmspace = imgp->proc->p_vmspace;
989 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
990 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
991 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
992 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
998 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
999 char **interpp, bool *free_interpp)
1003 int error, interp_name_len;
1005 KASSERT(phdr->p_type == PT_INTERP,
1006 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
1007 ASSERT_VOP_LOCKED(imgp->vp, __func__);
1011 /* Path to interpreter */
1012 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1013 uprintf("Invalid PT_INTERP\n");
1017 interp_name_len = phdr->p_filesz;
1018 if (phdr->p_offset > PAGE_SIZE ||
1019 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1021 * The vnode lock might be needed by the pagedaemon to
1022 * clean pages owned by the vnode. Do not allow sleep
1023 * waiting for memory with the vnode locked, instead
1024 * try non-sleepable allocation first, and if it
1025 * fails, go to the slow path were we drop the lock
1026 * and do M_WAITOK. A text reference prevents
1027 * modifications to the vnode content.
1029 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1030 if (interp == NULL) {
1031 VOP_UNLOCK(imgp->vp);
1032 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1033 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1036 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1037 interp_name_len, phdr->p_offset,
1038 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1041 free(interp, M_TEMP);
1042 uprintf("i/o error PT_INTERP %d\n", error);
1045 interp[interp_name_len] = '\0';
1048 *free_interpp = true;
1052 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1053 if (interp[interp_name_len - 1] != '\0') {
1054 uprintf("Invalid PT_INTERP\n");
1059 *free_interpp = false;
1064 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1065 const char *interp, u_long *addr, u_long *entry)
1070 if (brand_info->emul_path != NULL &&
1071 brand_info->emul_path[0] != '\0') {
1072 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1073 snprintf(path, MAXPATHLEN, "%s%s",
1074 brand_info->emul_path, interp);
1075 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1081 if (brand_info->interp_newpath != NULL &&
1082 (brand_info->interp_path == NULL ||
1083 strcmp(interp, brand_info->interp_path) == 0)) {
1084 error = __elfN(load_file)(imgp->proc,
1085 brand_info->interp_newpath, addr, entry);
1090 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1094 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1099 * Impossible et_dyn_addr initial value indicating that the real base
1100 * must be calculated later with some randomization applied.
1102 #define ET_DYN_ADDR_RAND 1
1105 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1108 const Elf_Ehdr *hdr;
1109 const Elf_Phdr *phdr;
1110 Elf_Auxargs *elf_auxargs;
1111 struct vmspace *vmspace;
1114 Elf_Brandinfo *brand_info;
1115 struct sysentvec *sv;
1116 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1117 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1123 hdr = (const Elf_Ehdr *)imgp->image_header;
1126 * Do we have a valid ELF header ?
1128 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1129 * if particular brand doesn't support it.
1131 if (__elfN(check_header)(hdr) != 0 ||
1132 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1136 * From here on down, we return an errno, not -1, as we've
1137 * detected an ELF file.
1140 if (!__elfN(phdr_in_zero_page)(hdr)) {
1141 uprintf("Program headers not in the first page\n");
1144 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1145 if (!aligned(phdr, Elf_Addr)) {
1146 uprintf("Unaligned program headers\n");
1154 entry = proghdr = 0;
1156 free_interp = false;
1160 * Somewhat arbitrary, limit accepted max alignment for the
1161 * loadable segment to the max supported superpage size. Too
1162 * large alignment requests are not useful and are indicators
1163 * of corrupted or outright malicious binary.
1165 maxalign = PAGE_SIZE;
1166 maxsalign = PAGE_SIZE * 1024;
1167 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1168 if (pagesizes[i] > maxsalign)
1169 maxsalign = pagesizes[i];
1174 for (i = 0; i < hdr->e_phnum; i++) {
1175 switch (phdr[i].p_type) {
1178 baddr = phdr[i].p_vaddr;
1179 if (!powerof2(phdr[i].p_align) ||
1180 phdr[i].p_align > maxsalign) {
1181 uprintf("Invalid segment alignment\n");
1185 if (phdr[i].p_align > maxalign)
1186 maxalign = phdr[i].p_align;
1187 if (mapsz + phdr[i].p_memsz < mapsz) {
1188 uprintf("Mapsize overflow\n");
1192 mapsz += phdr[i].p_memsz;
1196 * If this segment contains the program headers,
1197 * remember their virtual address for the AT_PHDR
1198 * aux entry. Static binaries don't usually include
1201 if (phdr[i].p_offset == 0 &&
1202 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1204 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1207 /* Path to interpreter */
1208 if (interp != NULL) {
1209 uprintf("Multiple PT_INTERP headers\n");
1213 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1219 if (__elfN(nxstack))
1221 __elfN(trans_prot)(phdr[i].p_flags);
1222 imgp->stack_sz = phdr[i].p_memsz;
1224 case PT_PHDR: /* Program header table info */
1225 proghdr = phdr[i].p_vaddr;
1230 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1231 if (brand_info == NULL) {
1232 uprintf("ELF binary type \"%u\" not known.\n",
1233 hdr->e_ident[EI_OSABI]);
1237 sv = brand_info->sysvec;
1239 if (hdr->e_type == ET_DYN) {
1240 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1241 uprintf("Cannot execute shared object\n");
1246 * Honour the base load address from the dso if it is
1247 * non-zero for some reason.
1250 if ((sv->sv_flags & SV_ASLR) == 0 ||
1251 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1252 et_dyn_addr = __elfN(pie_base);
1253 else if ((__elfN(pie_aslr_enabled) &&
1254 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1255 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1256 et_dyn_addr = ET_DYN_ADDR_RAND;
1258 et_dyn_addr = __elfN(pie_base);
1263 * Avoid a possible deadlock if the current address space is destroyed
1264 * and that address space maps the locked vnode. In the common case,
1265 * the locked vnode's v_usecount is decremented but remains greater
1266 * than zero. Consequently, the vnode lock is not needed by vrele().
1267 * However, in cases where the vnode lock is external, such as nullfs,
1268 * v_usecount may become zero.
1270 * The VV_TEXT flag prevents modifications to the executable while
1271 * the vnode is unlocked.
1273 VOP_UNLOCK(imgp->vp);
1276 * Decide whether to enable randomization of user mappings.
1277 * First, reset user preferences for the setid binaries.
1278 * Then, account for the support of the randomization by the
1279 * ABI, by user preferences, and make special treatment for
1282 if (imgp->credential_setid) {
1283 PROC_LOCK(imgp->proc);
1284 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1285 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1286 PROC_UNLOCK(imgp->proc);
1288 if ((sv->sv_flags & SV_ASLR) == 0 ||
1289 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1290 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1291 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1292 ("et_dyn_addr == RAND and !ASLR"));
1293 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1294 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1295 et_dyn_addr == ET_DYN_ADDR_RAND) {
1296 imgp->map_flags |= MAP_ASLR;
1298 * If user does not care about sbrk, utilize the bss
1299 * grow region for mappings as well. We can select
1300 * the base for the image anywere and still not suffer
1301 * from the fragmentation.
1303 if (!__elfN(aslr_honor_sbrk) ||
1304 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1305 imgp->map_flags |= MAP_ASLR_IGNSTART;
1306 if (__elfN(aslr_stack))
1307 imgp->map_flags |= MAP_ASLR_STACK;
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 && 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, &et_dyn_addr);
1339 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1343 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1347 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
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 + et_dyn_addr;
1372 imgp->entry_addr = entry;
1374 if (interp != NULL) {
1375 VOP_UNLOCK(imgp->vp);
1376 if ((map->flags & MAP_ASLR) != 0) {
1377 /* Assume that interpreter fits into 1/4 of AS */
1378 maxv1 = maxv / 2 + addr / 2;
1379 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1380 maxv1, PAGE_SIZE, &addr);
1383 error = __elfN(load_interp)(imgp, brand_info, interp,
1384 &addr, &imgp->entry_addr);
1386 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1392 error = exec_map_stack(imgp);
1397 * Construct auxargs table (used by the copyout_auxargs routine)
1399 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1400 if (elf_auxargs == NULL) {
1401 VOP_UNLOCK(imgp->vp);
1402 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1403 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1405 elf_auxargs->execfd = -1;
1406 elf_auxargs->phdr = proghdr + et_dyn_addr;
1407 elf_auxargs->phent = hdr->e_phentsize;
1408 elf_auxargs->phnum = hdr->e_phnum;
1409 elf_auxargs->pagesz = PAGE_SIZE;
1410 elf_auxargs->base = addr;
1411 elf_auxargs->flags = 0;
1412 elf_auxargs->entry = entry;
1413 elf_auxargs->hdr_eflags = hdr->e_flags;
1415 imgp->auxargs = elf_auxargs;
1416 imgp->interpreted = 0;
1417 imgp->reloc_base = addr;
1418 imgp->proc->p_osrel = osrel;
1419 imgp->proc->p_fctl0 = fctl0;
1420 imgp->proc->p_elf_flags = hdr->e_flags;
1423 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1425 free(interp, M_TEMP);
1429 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1432 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1434 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1435 Elf_Auxinfo *argarray, *pos;
1438 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1441 if (args->execfd != -1)
1442 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1443 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1444 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1445 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1446 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1447 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1448 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1449 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1450 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1451 if (imgp->execpathp != 0)
1452 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1453 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1454 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1455 if (imgp->canary != 0) {
1456 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1457 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1459 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1460 if (imgp->pagesizes != 0) {
1461 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1462 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1464 if (imgp->sysent->sv_timekeep_base != 0) {
1465 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1466 imgp->sysent->sv_timekeep_base);
1468 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1469 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1470 imgp->sysent->sv_stackprot);
1471 if (imgp->sysent->sv_hwcap != NULL)
1472 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1473 if (imgp->sysent->sv_hwcap2 != NULL)
1474 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1475 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1476 ELF_BSDF_SIGFASTBLK : 0);
1477 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1478 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1479 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1480 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1481 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1482 if (imgp->sysent->sv_fxrng_gen_base != 0)
1483 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1484 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1485 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1486 AUXARGS_ENTRY(pos, AT_NULL, 0);
1488 free(imgp->auxargs, M_TEMP);
1489 imgp->auxargs = NULL;
1490 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1492 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1493 free(argarray, M_TEMP);
1498 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1502 base = (Elf_Addr *)*stack_base;
1504 if (elf_suword(base, imgp->args->argc) == -1)
1506 *stack_base = (uintptr_t)base;
1511 * Code for generating ELF core dumps.
1514 typedef void (*segment_callback)(vm_map_entry_t, void *);
1516 /* Closure for cb_put_phdr(). */
1517 struct phdr_closure {
1518 Elf_Phdr *phdr; /* Program header to fill in */
1519 Elf_Off offset; /* Offset of segment in core file */
1523 int type; /* Note type. */
1524 struct regset *regset; /* Register set. */
1525 outfunc_t outfunc; /* Output function. */
1526 void *outarg; /* Argument for the output function. */
1527 size_t outsize; /* Output size. */
1528 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1531 TAILQ_HEAD(note_info_list, note_info);
1533 extern int compress_user_cores;
1534 extern int compress_user_cores_level;
1536 static void cb_put_phdr(vm_map_entry_t, void *);
1537 static void cb_size_segment(vm_map_entry_t, void *);
1538 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1540 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1541 struct note_info_list *, size_t, int);
1542 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1544 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1545 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1546 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1547 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1548 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1549 static void note_procstat_files(void *, struct sbuf *, size_t *);
1550 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1551 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1552 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1553 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1554 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1557 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1560 return (core_write((struct coredump_params *)arg, base, len, offset,
1561 UIO_SYSSPACE, NULL));
1565 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1567 struct ucred *cred = td->td_ucred;
1568 int compm, error = 0;
1569 struct sseg_closure seginfo;
1570 struct note_info_list notelst;
1571 struct coredump_params params;
1572 struct note_info *ninfo;
1574 size_t hdrsize, notesz, coresize;
1578 TAILQ_INIT(¬elst);
1580 /* Size the program segments. */
1581 __elfN(size_segments)(td, &seginfo, flags);
1584 * Collect info about the core file header area.
1586 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1587 if (seginfo.count + 1 >= PN_XNUM)
1588 hdrsize += sizeof(Elf_Shdr);
1589 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1590 coresize = round_page(hdrsize + notesz) + seginfo.size;
1592 /* Set up core dump parameters. */
1594 params.active_cred = cred;
1595 params.file_cred = NOCRED;
1602 PROC_LOCK(td->td_proc);
1603 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1604 PROC_UNLOCK(td->td_proc);
1611 if (coresize >= limit) {
1616 /* Create a compression stream if necessary. */
1617 compm = compress_user_cores;
1618 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1620 compm = COMPRESS_GZIP;
1622 params.comp = compressor_init(core_compressed_write,
1623 compm, CORE_BUF_SIZE,
1624 compress_user_cores_level, ¶ms);
1625 if (params.comp == NULL) {
1629 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1633 * Allocate memory for building the header, fill it up,
1634 * and write it out following the notes.
1636 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1637 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1640 /* Write the contents of all of the writable segments. */
1646 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1647 offset = round_page(hdrsize + notesz);
1648 for (i = 0; i < seginfo.count; i++) {
1649 error = core_output((char *)(uintptr_t)php->p_vaddr,
1650 php->p_filesz, offset, ¶ms, tmpbuf);
1653 offset += php->p_filesz;
1656 if (error == 0 && params.comp != NULL)
1657 error = compressor_flush(params.comp);
1661 "Failed to write core file for process %s (error %d)\n",
1662 curproc->p_comm, error);
1666 free(tmpbuf, M_TEMP);
1667 if (params.comp != NULL)
1668 compressor_fini(params.comp);
1669 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1670 TAILQ_REMOVE(¬elst, ninfo, link);
1671 free(ninfo, M_TEMP);
1680 * A callback for each_dumpable_segment() to write out the segment's
1681 * program header entry.
1684 cb_put_phdr(vm_map_entry_t entry, void *closure)
1686 struct phdr_closure *phc = (struct phdr_closure *)closure;
1687 Elf_Phdr *phdr = phc->phdr;
1689 phc->offset = round_page(phc->offset);
1691 phdr->p_type = PT_LOAD;
1692 phdr->p_offset = phc->offset;
1693 phdr->p_vaddr = entry->start;
1695 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1696 phdr->p_align = PAGE_SIZE;
1697 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1699 phc->offset += phdr->p_filesz;
1704 * A callback for each_dumpable_segment() to gather information about
1705 * the number of segments and their total size.
1708 cb_size_segment(vm_map_entry_t entry, void *closure)
1710 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1713 ssc->size += entry->end - entry->start;
1717 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1723 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1727 * For each writable segment in the process's memory map, call the given
1728 * function with a pointer to the map entry and some arbitrary
1729 * caller-supplied data.
1732 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1735 struct proc *p = td->td_proc;
1736 vm_map_t map = &p->p_vmspace->vm_map;
1737 vm_map_entry_t entry;
1738 vm_object_t backing_object, object;
1741 vm_map_lock_read(map);
1742 VM_MAP_ENTRY_FOREACH(entry, map) {
1744 * Don't dump inaccessible mappings, deal with legacy
1747 * Note that read-only segments related to the elf binary
1748 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1749 * need to arbitrarily ignore such segments.
1751 if ((flags & SVC_ALL) == 0) {
1752 if (elf_legacy_coredump) {
1753 if ((entry->protection & VM_PROT_RW) !=
1757 if ((entry->protection & VM_PROT_ALL) == 0)
1763 * Dont include memory segment in the coredump if
1764 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1765 * madvise(2). Do not dump submaps (i.e. parts of the
1768 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1770 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1771 (flags & SVC_ALL) == 0)
1773 if ((object = entry->object.vm_object) == NULL)
1776 /* Ignore memory-mapped devices and such things. */
1777 VM_OBJECT_RLOCK(object);
1778 while ((backing_object = object->backing_object) != NULL) {
1779 VM_OBJECT_RLOCK(backing_object);
1780 VM_OBJECT_RUNLOCK(object);
1781 object = backing_object;
1783 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1784 VM_OBJECT_RUNLOCK(object);
1788 (*func)(entry, closure);
1790 vm_map_unlock_read(map);
1794 * Write the core file header to the file, including padding up to
1795 * the page boundary.
1798 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1799 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1802 struct note_info *ninfo;
1806 /* Fill in the header. */
1807 bzero(hdr, hdrsize);
1808 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1810 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1811 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1812 sbuf_start_section(sb, NULL);
1813 sbuf_bcat(sb, hdr, hdrsize);
1814 TAILQ_FOREACH(ninfo, notelst, link)
1815 __elfN(putnote)(p->td, ninfo, sb);
1816 /* Align up to a page boundary for the program segments. */
1817 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1818 error = sbuf_finish(sb);
1825 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1835 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1836 __elfN(note_prpsinfo), p);
1839 * To have the debugger select the right thread (LWP) as the initial
1840 * thread, we dump the state of the thread passed to us in td first.
1841 * This is the thread that causes the core dump and thus likely to
1842 * be the right thread one wants to have selected in the debugger.
1845 while (thr != NULL) {
1846 size += __elfN(prepare_register_notes)(td, list, thr);
1847 size += __elfN(register_note)(td, list, -1,
1848 __elfN(note_threadmd), thr);
1850 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1851 TAILQ_NEXT(thr, td_plist);
1853 thr = TAILQ_NEXT(thr, td_plist);
1856 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1857 __elfN(note_procstat_proc), p);
1858 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1859 note_procstat_files, p);
1860 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1861 note_procstat_vmmap, p);
1862 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1863 note_procstat_groups, p);
1864 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1865 note_procstat_umask, p);
1866 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1867 note_procstat_rlimit, p);
1868 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1869 note_procstat_osrel, p);
1870 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1871 __elfN(note_procstat_psstrings), p);
1872 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1873 __elfN(note_procstat_auxv), p);
1879 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1880 size_t notesz, int flags)
1885 struct phdr_closure phc;
1888 ehdr = (Elf_Ehdr *)hdr;
1889 bi = td->td_proc->p_elf_brandinfo;
1891 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1892 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1893 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1894 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1895 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1896 ehdr->e_ident[EI_DATA] = ELF_DATA;
1897 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1898 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1899 ehdr->e_ident[EI_ABIVERSION] = 0;
1900 ehdr->e_ident[EI_PAD] = 0;
1901 ehdr->e_type = ET_CORE;
1902 ehdr->e_machine = bi->machine;
1903 ehdr->e_version = EV_CURRENT;
1905 ehdr->e_phoff = sizeof(Elf_Ehdr);
1906 ehdr->e_flags = td->td_proc->p_elf_flags;
1907 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1908 ehdr->e_phentsize = sizeof(Elf_Phdr);
1909 ehdr->e_shentsize = sizeof(Elf_Shdr);
1910 ehdr->e_shstrndx = SHN_UNDEF;
1911 if (numsegs + 1 < PN_XNUM) {
1912 ehdr->e_phnum = numsegs + 1;
1915 ehdr->e_phnum = PN_XNUM;
1918 ehdr->e_shoff = ehdr->e_phoff +
1919 (numsegs + 1) * ehdr->e_phentsize;
1920 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1921 ("e_shoff: %zu, hdrsize - shdr: %zu",
1922 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1924 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1925 memset(shdr, 0, sizeof(*shdr));
1927 * A special first section is used to hold large segment and
1928 * section counts. This was proposed by Sun Microsystems in
1929 * Solaris and has been adopted by Linux; the standard ELF
1930 * tools are already familiar with the technique.
1932 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1933 * (or 12-7 depending on the version of the document) for more
1936 shdr->sh_type = SHT_NULL;
1937 shdr->sh_size = ehdr->e_shnum;
1938 shdr->sh_link = ehdr->e_shstrndx;
1939 shdr->sh_info = numsegs + 1;
1943 * Fill in the program header entries.
1945 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1947 /* The note segement. */
1948 phdr->p_type = PT_NOTE;
1949 phdr->p_offset = hdrsize;
1952 phdr->p_filesz = notesz;
1954 phdr->p_flags = PF_R;
1955 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1958 /* All the writable segments from the program. */
1960 phc.offset = round_page(hdrsize + notesz);
1961 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1965 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1966 struct regset *regset, struct thread *target_td)
1968 const struct sysentvec *sv;
1969 struct note_info *ninfo;
1970 size_t size, notesize;
1973 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1976 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1977 ninfo->type = regset->note;
1978 ninfo->regset = regset;
1979 ninfo->outarg = target_td;
1980 ninfo->outsize = size;
1981 TAILQ_INSERT_TAIL(list, ninfo, link);
1983 sv = td->td_proc->p_sysent;
1984 notesize = sizeof(Elf_Note) + /* note header */
1985 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1987 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1993 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1994 int type, outfunc_t out, void *arg)
1996 const struct sysentvec *sv;
1997 struct note_info *ninfo;
1998 size_t size, notesize;
2000 sv = td->td_proc->p_sysent;
2002 out(arg, NULL, &size);
2003 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2005 ninfo->outfunc = out;
2006 ninfo->outarg = arg;
2007 ninfo->outsize = size;
2008 TAILQ_INSERT_TAIL(list, ninfo, link);
2013 notesize = sizeof(Elf_Note) + /* note header */
2014 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2016 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2022 append_note_data(const void *src, void *dst, size_t len)
2026 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2028 bcopy(src, dst, len);
2029 bzero((char *)dst + len, padded_len - len);
2031 return (padded_len);
2035 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2043 note = (Elf_Note *)buf;
2044 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2045 note->n_descsz = size;
2046 note->n_type = type;
2047 buf += sizeof(*note);
2048 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2049 sizeof(FREEBSD_ABI_VENDOR));
2050 append_note_data(src, buf, size);
2055 notesize = sizeof(Elf_Note) + /* note header */
2056 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2058 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2064 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2067 const struct sysentvec *sv;
2068 ssize_t old_len, sect_len;
2069 size_t new_len, descsz, i;
2071 if (ninfo->type == -1) {
2072 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2076 sv = td->td_proc->p_sysent;
2078 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2079 note.n_descsz = ninfo->outsize;
2080 note.n_type = ninfo->type;
2082 sbuf_bcat(sb, ¬e, sizeof(note));
2083 sbuf_start_section(sb, &old_len);
2084 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2085 strlen(sv->sv_elf_core_abi_vendor) + 1);
2086 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2087 if (note.n_descsz == 0)
2089 sbuf_start_section(sb, &old_len);
2090 if (ninfo->regset != NULL) {
2091 struct regset *regset = ninfo->regset;
2094 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2095 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2096 sbuf_bcat(sb, buf, ninfo->outsize);
2099 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2100 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2104 new_len = (size_t)sect_len;
2105 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2106 if (new_len < descsz) {
2108 * It is expected that individual note emitters will correctly
2109 * predict their expected output size and fill up to that size
2110 * themselves, padding in a format-specific way if needed.
2111 * However, in case they don't, just do it here with zeros.
2113 for (i = 0; i < descsz - new_len; i++)
2115 } else if (new_len > descsz) {
2117 * We can't always truncate sb -- we may have drained some
2120 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2121 "read it (%zu > %zu). Since it is longer than "
2122 "expected, this coredump's notes are corrupt. THIS "
2123 "IS A BUG in the note_procstat routine for type %u.\n",
2124 __func__, (unsigned)note.n_type, new_len, descsz,
2125 (unsigned)note.n_type));
2130 * Miscellaneous note out functions.
2133 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2134 #include <compat/freebsd32/freebsd32.h>
2135 #include <compat/freebsd32/freebsd32_signal.h>
2137 typedef struct prstatus32 elf_prstatus_t;
2138 typedef struct prpsinfo32 elf_prpsinfo_t;
2139 typedef struct fpreg32 elf_prfpregset_t;
2140 typedef struct fpreg32 elf_fpregset_t;
2141 typedef struct reg32 elf_gregset_t;
2142 typedef struct thrmisc32 elf_thrmisc_t;
2143 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2144 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2145 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2146 typedef uint32_t elf_ps_strings_t;
2148 typedef prstatus_t elf_prstatus_t;
2149 typedef prpsinfo_t elf_prpsinfo_t;
2150 typedef prfpregset_t elf_prfpregset_t;
2151 typedef prfpregset_t elf_fpregset_t;
2152 typedef gregset_t elf_gregset_t;
2153 typedef thrmisc_t elf_thrmisc_t;
2154 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2155 #define ELF_KERN_PROC_MASK 0
2156 typedef struct kinfo_proc elf_kinfo_proc_t;
2157 typedef vm_offset_t elf_ps_strings_t;
2161 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2167 elf_prpsinfo_t *psinfo;
2172 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2173 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2174 psinfo->pr_version = PRPSINFO_VERSION;
2175 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2176 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2178 if (p->p_args != NULL) {
2179 len = sizeof(psinfo->pr_psargs) - 1;
2180 if (len > p->p_args->ar_length)
2181 len = p->p_args->ar_length;
2182 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2188 sbuf_new(&sbarg, psinfo->pr_psargs,
2189 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2190 error = proc_getargv(curthread, p, &sbarg);
2192 if (sbuf_finish(&sbarg) == 0)
2193 len = sbuf_len(&sbarg) - 1;
2195 len = sizeof(psinfo->pr_psargs) - 1;
2196 sbuf_delete(&sbarg);
2198 if (error || len == 0)
2199 strlcpy(psinfo->pr_psargs, p->p_comm,
2200 sizeof(psinfo->pr_psargs));
2202 KASSERT(len < sizeof(psinfo->pr_psargs),
2203 ("len is too long: %zu vs %zu", len,
2204 sizeof(psinfo->pr_psargs)));
2205 cp = psinfo->pr_psargs;
2208 cp = memchr(cp, '\0', end - cp);
2214 psinfo->pr_pid = p->p_pid;
2215 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2216 free(psinfo, M_TEMP);
2218 *sizep = sizeof(*psinfo);
2222 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2225 elf_prstatus_t *status;
2228 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2231 memset(status, 0, *sizep);
2232 status->pr_version = PRSTATUS_VERSION;
2233 status->pr_statussz = sizeof(elf_prstatus_t);
2234 status->pr_gregsetsz = sizeof(elf_gregset_t);
2235 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2236 status->pr_osreldate = osreldate;
2237 status->pr_cursig = td->td_proc->p_sig;
2238 status->pr_pid = td->td_tid;
2239 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2240 fill_regs32(td, &status->pr_reg);
2242 fill_regs(td, &status->pr_reg);
2245 *sizep = sizeof(*status);
2250 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2253 elf_prstatus_t *status;
2255 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2257 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2258 set_regs32(td, &status->pr_reg);
2260 set_regs(td, &status->pr_reg);
2265 static struct regset __elfN(regset_prstatus) = {
2266 .note = NT_PRSTATUS,
2267 .size = sizeof(elf_prstatus_t),
2268 .get = __elfN(get_prstatus),
2269 .set = __elfN(set_prstatus),
2271 ELF_REGSET(__elfN(regset_prstatus));
2274 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2277 elf_prfpregset_t *fpregset;
2280 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2283 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2284 fill_fpregs32(td, fpregset);
2286 fill_fpregs(td, fpregset);
2289 *sizep = sizeof(*fpregset);
2294 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2297 elf_prfpregset_t *fpregset;
2300 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2301 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2302 set_fpregs32(td, fpregset);
2304 set_fpregs(td, fpregset);
2309 static struct regset __elfN(regset_fpregset) = {
2310 .note = NT_FPREGSET,
2311 .size = sizeof(elf_prfpregset_t),
2312 .get = __elfN(get_fpregset),
2313 .set = __elfN(set_fpregset),
2315 ELF_REGSET(__elfN(regset_fpregset));
2318 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2321 elf_thrmisc_t *thrmisc;
2324 KASSERT(*sizep == sizeof(*thrmisc),
2325 ("%s: invalid size", __func__));
2327 bzero(thrmisc, sizeof(*thrmisc));
2328 strcpy(thrmisc->pr_tname, td->td_name);
2330 *sizep = sizeof(*thrmisc);
2334 static struct regset __elfN(regset_thrmisc) = {
2336 .size = sizeof(elf_thrmisc_t),
2337 .get = __elfN(get_thrmisc),
2339 ELF_REGSET(__elfN(regset_thrmisc));
2342 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2349 size = sizeof(structsize) + sizeof(pl);
2351 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2352 structsize = sizeof(pl);
2353 memcpy(buf, &structsize, sizeof(structsize));
2354 bzero(&pl, sizeof(pl));
2355 pl.pl_lwpid = td->td_tid;
2356 pl.pl_event = PL_EVENT_NONE;
2357 pl.pl_sigmask = td->td_sigmask;
2358 pl.pl_siglist = td->td_siglist;
2359 if (td->td_si.si_signo != 0) {
2360 pl.pl_event = PL_EVENT_SIGNAL;
2361 pl.pl_flags |= PL_FLAG_SI;
2362 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2363 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2365 pl.pl_siginfo = td->td_si;
2368 strcpy(pl.pl_tdname, td->td_name);
2369 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2370 memcpy((int *)buf + 1, &pl, sizeof(pl));
2376 static struct regset __elfN(regset_lwpinfo) = {
2377 .note = NT_PTLWPINFO,
2378 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2379 .get = __elfN(get_lwpinfo),
2381 ELF_REGSET(__elfN(regset_lwpinfo));
2384 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2385 struct thread *target_td)
2387 struct sysentvec *sv = td->td_proc->p_sysent;
2388 struct regset **regsetp, **regset_end, *regset;
2393 /* NT_PRSTATUS must be the first register set note. */
2394 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2397 regsetp = sv->sv_regset_begin;
2398 if (regsetp == NULL) {
2399 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2400 size += __elfN(register_regset_note)(td, list,
2401 &__elfN(regset_fpregset), target_td);
2404 regset_end = sv->sv_regset_end;
2405 MPASS(regset_end != NULL);
2406 for (; regsetp < regset_end; regsetp++) {
2408 if (regset->note == NT_PRSTATUS)
2410 size += __elfN(register_regset_note)(td, list, regset,
2417 * Allow for MD specific notes, as well as any MD
2418 * specific preparations for writing MI notes.
2421 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2427 td = (struct thread *)arg;
2429 if (size != 0 && sb != NULL)
2430 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2434 __elfN(dump_thread)(td, buf, &size);
2435 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2436 if (size != 0 && sb != NULL)
2437 sbuf_bcat(sb, buf, size);
2442 #ifdef KINFO_PROC_SIZE
2443 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2447 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2454 size = sizeof(structsize) + p->p_numthreads *
2455 sizeof(elf_kinfo_proc_t);
2458 KASSERT(*sizep == size, ("invalid size"));
2459 structsize = sizeof(elf_kinfo_proc_t);
2460 sbuf_bcat(sb, &structsize, sizeof(structsize));
2461 sx_slock(&proctree_lock);
2463 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2464 sx_sunlock(&proctree_lock);
2469 #ifdef KINFO_FILE_SIZE
2470 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2474 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2477 size_t size, sect_sz, i;
2478 ssize_t start_len, sect_len;
2479 int structsize, filedesc_flags;
2481 if (coredump_pack_fileinfo)
2482 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2487 structsize = sizeof(struct kinfo_file);
2490 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2491 sbuf_set_drain(sb, sbuf_count_drain, &size);
2492 sbuf_bcat(sb, &structsize, sizeof(structsize));
2494 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2499 sbuf_start_section(sb, &start_len);
2501 sbuf_bcat(sb, &structsize, sizeof(structsize));
2503 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2506 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2511 KASSERT(sect_sz <= *sizep,
2512 ("kern_proc_filedesc_out did not respect maxlen; "
2513 "requested %zu, got %zu", *sizep - sizeof(structsize),
2514 sect_sz - sizeof(structsize)));
2516 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2521 #ifdef KINFO_VMENTRY_SIZE
2522 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2526 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2530 int structsize, vmmap_flags;
2532 if (coredump_pack_vmmapinfo)
2533 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2538 structsize = sizeof(struct kinfo_vmentry);
2541 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2542 sbuf_set_drain(sb, sbuf_count_drain, &size);
2543 sbuf_bcat(sb, &structsize, sizeof(structsize));
2545 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2550 sbuf_bcat(sb, &structsize, sizeof(structsize));
2552 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2558 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2565 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2567 KASSERT(*sizep == size, ("invalid size"));
2568 structsize = sizeof(gid_t);
2569 sbuf_bcat(sb, &structsize, sizeof(structsize));
2570 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2577 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2584 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2586 KASSERT(*sizep == size, ("invalid size"));
2587 structsize = sizeof(p->p_pd->pd_cmask);
2588 sbuf_bcat(sb, &structsize, sizeof(structsize));
2589 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2595 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2598 struct rlimit rlim[RLIM_NLIMITS];
2603 size = sizeof(structsize) + sizeof(rlim);
2605 KASSERT(*sizep == size, ("invalid size"));
2606 structsize = sizeof(rlim);
2607 sbuf_bcat(sb, &structsize, sizeof(structsize));
2609 for (i = 0; i < RLIM_NLIMITS; i++)
2610 lim_rlimit_proc(p, i, &rlim[i]);
2612 sbuf_bcat(sb, rlim, sizeof(rlim));
2618 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2625 size = sizeof(structsize) + sizeof(p->p_osrel);
2627 KASSERT(*sizep == size, ("invalid size"));
2628 structsize = sizeof(p->p_osrel);
2629 sbuf_bcat(sb, &structsize, sizeof(structsize));
2630 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2636 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2639 elf_ps_strings_t ps_strings;
2644 size = sizeof(structsize) + sizeof(ps_strings);
2646 KASSERT(*sizep == size, ("invalid size"));
2647 structsize = sizeof(ps_strings);
2648 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2649 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2651 ps_strings = PROC_PS_STRINGS(p);
2653 sbuf_bcat(sb, &structsize, sizeof(structsize));
2654 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2660 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2669 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2671 sbuf_set_drain(sb, sbuf_count_drain, &size);
2672 sbuf_bcat(sb, &structsize, sizeof(structsize));
2674 proc_getauxv(curthread, p, sb);
2680 structsize = sizeof(Elf_Auxinfo);
2681 sbuf_bcat(sb, &structsize, sizeof(structsize));
2683 proc_getauxv(curthread, p, sb);
2689 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2690 const char *note_vendor, const Elf_Phdr *pnote,
2691 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2693 const Elf_Note *note, *note0, *note_end;
2694 const char *note_name;
2699 /* We need some limit, might as well use PAGE_SIZE. */
2700 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2702 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2703 if (pnote->p_offset > PAGE_SIZE ||
2704 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2705 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2707 VOP_UNLOCK(imgp->vp);
2708 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2709 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2711 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2712 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2713 curthread->td_ucred, NOCRED, NULL, curthread);
2715 uprintf("i/o error PT_NOTE\n");
2718 note = note0 = (const Elf_Note *)buf;
2719 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2721 note = note0 = (const Elf_Note *)(imgp->image_header +
2723 note_end = (const Elf_Note *)(imgp->image_header +
2724 pnote->p_offset + pnote->p_filesz);
2727 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2728 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2729 (const char *)note < sizeof(Elf_Note)) {
2732 if (note->n_namesz != checknote->n_namesz ||
2733 note->n_descsz != checknote->n_descsz ||
2734 note->n_type != checknote->n_type)
2736 note_name = (const char *)(note + 1);
2737 if (note_name + checknote->n_namesz >=
2738 (const char *)note_end || strncmp(note_vendor,
2739 note_name, checknote->n_namesz) != 0)
2742 if (cb(note, cb_arg, &res))
2745 note = (const Elf_Note *)((const char *)(note + 1) +
2746 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2747 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2756 struct brandnote_cb_arg {
2757 Elf_Brandnote *brandnote;
2762 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2764 struct brandnote_cb_arg *arg;
2769 * Fetch the osreldate for binary from the ELF OSABI-note if
2772 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2773 arg->brandnote->trans_osrel != NULL ?
2774 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2779 static Elf_Note fctl_note = {
2780 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2781 .n_descsz = sizeof(uint32_t),
2782 .n_type = NT_FREEBSD_FEATURE_CTL,
2785 struct fctl_cb_arg {
2791 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2793 struct fctl_cb_arg *arg;
2794 const Elf32_Word *desc;
2798 p = (uintptr_t)(note + 1);
2799 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2800 desc = (const Elf32_Word *)p;
2801 *arg->has_fctl0 = true;
2802 *arg->fctl0 = desc[0];
2808 * Try to find the appropriate ABI-note section for checknote, fetch
2809 * the osreldate and feature control flags for binary from the ELF
2810 * OSABI-note. Only the first page of the image is searched, the same
2814 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2815 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2817 const Elf_Phdr *phdr;
2818 const Elf_Ehdr *hdr;
2819 struct brandnote_cb_arg b_arg;
2820 struct fctl_cb_arg f_arg;
2823 hdr = (const Elf_Ehdr *)imgp->image_header;
2824 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2825 b_arg.brandnote = brandnote;
2826 b_arg.osrel = osrel;
2827 f_arg.has_fctl0 = has_fctl0;
2828 f_arg.fctl0 = fctl0;
2830 for (i = 0; i < hdr->e_phnum; i++) {
2831 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2832 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2834 for (j = 0; j < hdr->e_phnum; j++) {
2835 if (phdr[j].p_type == PT_NOTE &&
2836 __elfN(parse_notes)(imgp, &fctl_note,
2837 FREEBSD_ABI_VENDOR, &phdr[j],
2838 note_fctl_cb, &f_arg))
2849 * Tell kern_execve.c about it, with a little help from the linker.
2851 static struct execsw __elfN(execsw) = {
2852 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2853 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2855 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2858 __elfN(trans_prot)(Elf_Word flags)
2864 prot |= VM_PROT_EXECUTE;
2866 prot |= VM_PROT_WRITE;
2868 prot |= VM_PROT_READ;
2869 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2870 if (i386_read_exec && (flags & PF_R))
2871 prot |= VM_PROT_EXECUTE;
2877 __elfN(untrans_prot)(vm_prot_t prot)
2882 if (prot & VM_PROT_EXECUTE)
2884 if (prot & VM_PROT_READ)
2886 if (prot & VM_PROT_WRITE)