2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include "opt_capsicum.h"
36 #include <sys/param.h>
37 #include <sys/capsicum.h>
38 #include <sys/compressor.h>
40 #include <sys/fcntl.h>
41 #include <sys/imgact.h>
42 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
49 #include <sys/namei.h>
51 #include <sys/procfs.h>
52 #include <sys/ptrace.h>
53 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
58 #include <sys/sf_buf.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/syslog.h>
69 #include <sys/eventhandler.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_extern.h>
80 #include <machine/elf.h>
81 #include <machine/md_var.h>
83 #define ELF_NOTE_ROUNDSIZE 4
84 #define OLD_EI_BRAND 8
86 static int __elfN(check_header)(const Elf_Ehdr *hdr);
87 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
88 const char *interp, int32_t *osrel, uint32_t *fctl0);
89 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 static int __elfN(load_section)(const struct image_params *imgp,
92 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
94 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
95 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
98 static bool __elfN(check_note)(struct image_params *imgp,
99 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103 static size_t __elfN(prepare_register_notes)(struct thread *td,
104 struct note_info_list *list, struct thread *target_td);
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if defined(__amd64__)
133 static int __elfN(vdso) = 1;
134 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
135 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
136 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
138 static int __elfN(vdso) = 0;
141 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
142 int i386_read_exec = 0;
143 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
144 "enable execution from readable segments");
147 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
149 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
154 val = __elfN(pie_base);
155 error = sysctl_handle_long(oidp, &val, 0, req);
156 if (error != 0 || req->newptr == NULL)
158 if ((val & PAGE_MASK) != 0)
160 __elfN(pie_base) = val;
163 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
164 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
165 sysctl_pie_base, "LU",
166 "PIE load base without randomization");
168 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
174 * Enable ASLR by default for 64-bit non-PIE binaries. 32-bit architectures
175 * have limited address space (which can cause issues for applications with
176 * high memory use) so we leave it off there.
178 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
179 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
180 &__elfN(aslr_enabled), 0,
181 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
182 ": enable address map randomization");
185 * Enable ASLR by default for 64-bit PIE binaries.
187 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
188 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
189 &__elfN(pie_aslr_enabled), 0,
190 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
191 ": enable address map randomization for PIE binaries");
194 * Sbrk is deprecated and it can be assumed that in most cases it will not be
195 * used anyway. This setting is valid only with ASLR enabled, and allows ASLR
196 * to use the bss grow region.
198 static int __elfN(aslr_honor_sbrk) = 0;
199 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
200 &__elfN(aslr_honor_sbrk), 0,
201 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
203 static int __elfN(aslr_stack) = __ELF_WORD_SIZE == 64;
204 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
205 &__elfN(aslr_stack), 0,
206 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
207 ": enable stack address randomization");
209 static int __elfN(aslr_shared_page) = __ELF_WORD_SIZE == 64;
210 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, shared_page, CTLFLAG_RWTUN,
211 &__elfN(aslr_shared_page), 0,
212 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
213 ": enable shared page address randomization");
215 static int __elfN(sigfastblock) = 1;
216 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
217 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
218 "enable sigfastblock for new processes");
220 static bool __elfN(allow_wx) = true;
221 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
222 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
223 "Allow pages to be mapped simultaneously writable and executable");
225 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
227 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
229 Elf_Brandnote __elfN(freebsd_brandnote) = {
230 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
231 .hdr.n_descsz = sizeof(int32_t),
232 .hdr.n_type = NT_FREEBSD_ABI_TAG,
233 .vendor = FREEBSD_ABI_VENDOR,
234 .flags = BN_TRANSLATE_OSREL,
235 .trans_osrel = __elfN(freebsd_trans_osrel)
239 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
243 p = (uintptr_t)(note + 1);
244 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
245 *osrel = *(const int32_t *)(p);
250 static int GNU_KFREEBSD_ABI_DESC = 3;
252 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
253 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
254 .hdr.n_descsz = 16, /* XXX at least 16 */
256 .vendor = GNU_ABI_VENDOR,
257 .flags = BN_TRANSLATE_OSREL,
258 .trans_osrel = kfreebsd_trans_osrel
262 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
264 const Elf32_Word *desc;
267 p = (uintptr_t)(note + 1);
268 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
270 desc = (const Elf32_Word *)p;
271 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
275 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
276 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
278 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
284 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
288 for (i = 0; i < MAX_BRANDS; i++) {
289 if (elf_brand_list[i] == NULL) {
290 elf_brand_list[i] = entry;
294 if (i == MAX_BRANDS) {
295 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
303 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
307 for (i = 0; i < MAX_BRANDS; i++) {
308 if (elf_brand_list[i] == entry) {
309 elf_brand_list[i] = NULL;
319 __elfN(brand_inuse)(Elf_Brandinfo *entry)
324 sx_slock(&allproc_lock);
325 FOREACH_PROC_IN_SYSTEM(p) {
326 if (p->p_sysent == entry->sysvec) {
331 sx_sunlock(&allproc_lock);
336 static Elf_Brandinfo *
337 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
338 int32_t *osrel, uint32_t *fctl0)
340 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
341 Elf_Brandinfo *bi, *bi_m;
343 int i, interp_name_len;
345 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
348 * We support four types of branding -- (1) the ELF EI_OSABI field
349 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
350 * branding w/in the ELF header, (3) path of the `interp_path'
351 * field, and (4) the ".note.ABI-tag" ELF section.
354 /* Look for an ".note.ABI-tag" ELF section */
356 for (i = 0; i < MAX_BRANDS; i++) {
357 bi = elf_brand_list[i];
360 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
362 if (hdr->e_machine == bi->machine && (bi->flags &
363 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
367 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
369 /* Give brand a chance to veto check_note's guess */
370 if (ret && bi->header_supported) {
371 ret = bi->header_supported(imgp, osrel,
372 has_fctl0 ? fctl0 : NULL);
375 * If note checker claimed the binary, but the
376 * interpreter path in the image does not
377 * match default one for the brand, try to
378 * search for other brands with the same
379 * interpreter. Either there is better brand
380 * with the right interpreter, or, failing
381 * this, we return first brand which accepted
382 * our note and, optionally, header.
384 if (ret && bi_m == NULL && interp != NULL &&
385 (bi->interp_path == NULL ||
386 (strlen(bi->interp_path) + 1 != interp_name_len ||
387 strncmp(interp, bi->interp_path, interp_name_len)
399 /* If the executable has a brand, search for it in the brand list. */
400 for (i = 0; i < MAX_BRANDS; i++) {
401 bi = elf_brand_list[i];
402 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
403 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
405 if (hdr->e_machine == bi->machine &&
406 (hdr->e_ident[EI_OSABI] == bi->brand ||
407 (bi->compat_3_brand != NULL &&
408 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
409 bi->compat_3_brand) == 0))) {
410 /* Looks good, but give brand a chance to veto */
411 if (bi->header_supported == NULL ||
412 bi->header_supported(imgp, NULL, NULL)) {
414 * Again, prefer strictly matching
417 if (interp_name_len == 0 &&
418 bi->interp_path == NULL)
420 if (bi->interp_path != NULL &&
421 strlen(bi->interp_path) + 1 ==
422 interp_name_len && strncmp(interp,
423 bi->interp_path, interp_name_len) == 0)
433 /* No known brand, see if the header is recognized by any brand */
434 for (i = 0; i < MAX_BRANDS; i++) {
435 bi = elf_brand_list[i];
436 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
437 bi->header_supported == NULL)
439 if (hdr->e_machine == bi->machine) {
440 ret = bi->header_supported(imgp, NULL, NULL);
446 /* Lacking a known brand, search for a recognized interpreter. */
447 if (interp != NULL) {
448 for (i = 0; i < MAX_BRANDS; i++) {
449 bi = elf_brand_list[i];
450 if (bi == NULL || (bi->flags &
451 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
454 if (hdr->e_machine == bi->machine &&
455 bi->interp_path != NULL &&
456 /* ELF image p_filesz includes terminating zero */
457 strlen(bi->interp_path) + 1 == interp_name_len &&
458 strncmp(interp, bi->interp_path, interp_name_len)
459 == 0 && (bi->header_supported == NULL ||
460 bi->header_supported(imgp, NULL, NULL)))
465 /* Lacking a recognized interpreter, try the default brand */
466 for (i = 0; i < MAX_BRANDS; i++) {
467 bi = elf_brand_list[i];
468 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
469 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
471 if (hdr->e_machine == bi->machine &&
472 __elfN(fallback_brand) == bi->brand &&
473 (bi->header_supported == NULL ||
474 bi->header_supported(imgp, NULL, NULL)))
481 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
483 return (hdr->e_phoff <= PAGE_SIZE &&
484 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
488 __elfN(check_header)(const Elf_Ehdr *hdr)
494 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
495 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
496 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
497 hdr->e_phentsize != sizeof(Elf_Phdr) ||
498 hdr->e_version != ELF_TARG_VER)
502 * Make sure we have at least one brand for this machine.
505 for (i = 0; i < MAX_BRANDS; i++) {
506 bi = elf_brand_list[i];
507 if (bi != NULL && bi->machine == hdr->e_machine)
517 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
518 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
525 * Create the page if it doesn't exist yet. Ignore errors.
527 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
528 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
531 * Find the page from the underlying object.
533 if (object != NULL) {
534 sf = vm_imgact_map_page(object, offset);
536 return (KERN_FAILURE);
537 off = offset - trunc_page(offset);
538 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
540 vm_imgact_unmap_page(sf);
542 return (KERN_FAILURE);
545 return (KERN_SUCCESS);
549 __elfN(map_insert)(const struct image_params *imgp, vm_map_t map,
550 vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end,
551 vm_prot_t prot, int cow)
556 int error, locked, rv;
558 if (start != trunc_page(start)) {
559 rv = __elfN(map_partial)(map, object, offset, start,
560 round_page(start), prot);
561 if (rv != KERN_SUCCESS)
563 offset += round_page(start) - start;
564 start = round_page(start);
566 if (end != round_page(end)) {
567 rv = __elfN(map_partial)(map, object, offset +
568 trunc_page(end) - start, trunc_page(end), end, prot);
569 if (rv != KERN_SUCCESS)
571 end = trunc_page(end);
574 return (KERN_SUCCESS);
575 if ((offset & PAGE_MASK) != 0) {
577 * The mapping is not page aligned. This means that we have
580 rv = vm_map_fixed(map, NULL, 0, start, end - start,
581 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
582 if (rv != KERN_SUCCESS)
585 return (KERN_SUCCESS);
586 for (; start < end; start += sz) {
587 sf = vm_imgact_map_page(object, offset);
589 return (KERN_FAILURE);
590 off = offset - trunc_page(offset);
592 if (sz > PAGE_SIZE - off)
593 sz = PAGE_SIZE - off;
594 error = copyout((caddr_t)sf_buf_kva(sf) + off,
596 vm_imgact_unmap_page(sf);
598 return (KERN_FAILURE);
602 vm_object_reference(object);
603 rv = vm_map_fixed(map, object, offset, start, end - start,
604 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
605 (object != NULL ? MAP_VN_EXEC : 0));
606 if (rv != KERN_SUCCESS) {
607 locked = VOP_ISLOCKED(imgp->vp);
608 VOP_UNLOCK(imgp->vp);
609 vm_object_deallocate(object);
610 vn_lock(imgp->vp, locked | LK_RETRY);
612 } else if (object != NULL) {
613 MPASS(imgp->vp->v_object == object);
614 VOP_SET_TEXT_CHECKED(imgp->vp);
617 return (KERN_SUCCESS);
620 static int __elfN(load_section)(const struct image_params *imgp,
621 vm_ooffset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
628 vm_offset_t map_addr;
631 vm_ooffset_t file_addr;
634 * It's necessary to fail if the filsz + offset taken from the
635 * header is greater than the actual file pager object's size.
636 * If we were to allow this, then the vm_map_find() below would
637 * walk right off the end of the file object and into the ether.
639 * While I'm here, might as well check for something else that
640 * is invalid: filsz cannot be greater than memsz.
642 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
644 uprintf("elf_load_section: truncated ELF file\n");
648 object = imgp->object;
649 map = &imgp->proc->p_vmspace->vm_map;
650 map_addr = trunc_page((vm_offset_t)vmaddr);
651 file_addr = trunc_page(offset);
654 * We have two choices. We can either clear the data in the last page
655 * of an oversized mapping, or we can start the anon mapping a page
656 * early and copy the initialized data into that first page. We
661 else if (memsz > filsz)
662 map_len = trunc_page(offset + filsz) - file_addr;
664 map_len = round_page(offset + filsz) - file_addr;
667 /* cow flags: don't dump readonly sections in core */
668 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
669 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
671 rv = __elfN(map_insert)(imgp, map, object, file_addr,
672 map_addr, map_addr + map_len, prot, cow);
673 if (rv != KERN_SUCCESS)
676 /* we can stop now if we've covered it all */
682 * We have to get the remaining bit of the file into the first part
683 * of the oversized map segment. This is normally because the .data
684 * segment in the file is extended to provide bss. It's a neat idea
685 * to try and save a page, but it's a pain in the behind to implement.
687 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
689 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
690 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
692 /* This had damn well better be true! */
694 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
695 map_addr + map_len, prot, 0);
696 if (rv != KERN_SUCCESS)
701 sf = vm_imgact_map_page(object, offset + filsz);
705 /* send the page fragment to user space */
706 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
708 vm_imgact_unmap_page(sf);
714 * Remove write access to the page if it was only granted by map_insert
717 if ((prot & VM_PROT_WRITE) == 0)
718 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
719 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
725 __elfN(load_sections)(const struct image_params *imgp, const Elf_Ehdr *hdr,
726 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
733 ASSERT_VOP_LOCKED(imgp->vp, __func__);
738 for (i = 0; i < hdr->e_phnum; i++) {
739 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
742 /* Loadable segment */
743 prot = __elfN(trans_prot)(phdr[i].p_flags);
744 error = __elfN(load_section)(imgp, phdr[i].p_offset,
745 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
746 phdr[i].p_memsz, phdr[i].p_filesz, prot);
751 * Establish the base address if this is the first segment.
754 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
759 if (base_addrp != NULL)
760 *base_addrp = base_addr;
766 * Load the file "file" into memory. It may be either a shared object
769 * The "addr" reference parameter is in/out. On entry, it specifies
770 * the address where a shared object should be loaded. If the file is
771 * an executable, this value is ignored. On exit, "addr" specifies
772 * where the file was actually loaded.
774 * The "entry" reference parameter is out only. On exit, it specifies
775 * the entry point for the loaded file.
778 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
784 struct image_params image_params;
786 const Elf_Ehdr *hdr = NULL;
787 const Elf_Phdr *phdr = NULL;
788 struct nameidata *nd;
790 struct image_params *imgp;
792 u_long base_addr = 0;
795 #ifdef CAPABILITY_MODE
797 * XXXJA: This check can go away once we are sufficiently confident
798 * that the checks in namei() are correct.
800 if (IN_CAPABILITY_MODE(curthread))
804 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
806 attr = &tempdata->attr;
807 imgp = &tempdata->image_params;
810 * Initialize part of the common data
815 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
817 if ((error = namei(nd)) != 0) {
822 imgp->vp = nd->ni_vp;
825 * Check permissions, modes, uid, etc on the file, and "open" it.
827 error = exec_check_permissions(imgp);
831 error = exec_map_first_page(imgp);
835 imgp->object = nd->ni_vp->v_object;
837 hdr = (const Elf_Ehdr *)imgp->image_header;
838 if ((error = __elfN(check_header)(hdr)) != 0)
840 if (hdr->e_type == ET_DYN)
842 else if (hdr->e_type == ET_EXEC)
849 /* Only support headers that fit within first page for now */
850 if (!__elfN(phdr_in_zero_page)(hdr)) {
855 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
856 if (!aligned(phdr, Elf_Addr)) {
861 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
865 if (p->p_sysent->sv_protect != NULL)
866 p->p_sysent->sv_protect(imgp, SVP_INTERP);
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)
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 + imgp->et_dyn_addr);
943 seg_size = round_page(phdr[i].p_memsz +
944 phdr[i].p_vaddr + imgp->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)
1074 if (brand_info->interp_newpath != NULL &&
1075 (brand_info->interp_path == NULL ||
1076 strcmp(interp, brand_info->interp_path) == 0)) {
1077 error = __elfN(load_file)(imgp->proc,
1078 brand_info->interp_newpath, addr, entry);
1083 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1087 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1092 * Impossible et_dyn_addr initial value indicating that the real base
1093 * must be calculated later with some randomization applied.
1095 #define ET_DYN_ADDR_RAND 1
1098 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1101 const Elf_Ehdr *hdr;
1102 const Elf_Phdr *phdr;
1103 Elf_Auxargs *elf_auxargs;
1104 struct vmspace *vmspace;
1107 Elf_Brandinfo *brand_info;
1108 struct sysentvec *sv;
1109 u_long addr, baddr, entry, proghdr;
1110 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1116 hdr = (const Elf_Ehdr *)imgp->image_header;
1119 * Do we have a valid ELF header ?
1121 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1122 * if particular brand doesn't support it.
1124 if (__elfN(check_header)(hdr) != 0 ||
1125 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1129 * From here on down, we return an errno, not -1, as we've
1130 * detected an ELF file.
1133 if (!__elfN(phdr_in_zero_page)(hdr)) {
1134 uprintf("Program headers not in the first page\n");
1137 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1138 if (!aligned(phdr, Elf_Addr)) {
1139 uprintf("Unaligned program headers\n");
1147 entry = proghdr = 0;
1149 free_interp = false;
1153 * Somewhat arbitrary, limit accepted max alignment for the
1154 * loadable segment to the max supported superpage size. Too
1155 * large alignment requests are not useful and are indicators
1156 * of corrupted or outright malicious binary.
1158 maxalign = PAGE_SIZE;
1159 maxsalign = PAGE_SIZE * 1024;
1160 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1161 if (pagesizes[i] > maxsalign)
1162 maxsalign = pagesizes[i];
1167 for (i = 0; i < hdr->e_phnum; i++) {
1168 switch (phdr[i].p_type) {
1171 baddr = phdr[i].p_vaddr;
1172 if (!powerof2(phdr[i].p_align) ||
1173 phdr[i].p_align > maxsalign) {
1174 uprintf("Invalid segment alignment\n");
1178 if (phdr[i].p_align > maxalign)
1179 maxalign = phdr[i].p_align;
1180 if (mapsz + phdr[i].p_memsz < mapsz) {
1181 uprintf("Mapsize overflow\n");
1185 mapsz += phdr[i].p_memsz;
1189 * If this segment contains the program headers,
1190 * remember their virtual address for the AT_PHDR
1191 * aux entry. Static binaries don't usually include
1194 if (phdr[i].p_offset == 0 &&
1195 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1197 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1200 /* Path to interpreter */
1201 if (interp != NULL) {
1202 uprintf("Multiple PT_INTERP headers\n");
1206 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1212 if (__elfN(nxstack)) {
1214 __elfN(trans_prot)(phdr[i].p_flags);
1215 if ((imgp->stack_prot & VM_PROT_RW) !=
1217 uprintf("Invalid PT_GNU_STACK\n");
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;
1238 if (hdr->e_type == ET_DYN) {
1239 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1240 uprintf("Cannot execute shared object\n");
1245 * Honour the base load address from the dso if it is
1246 * non-zero for some reason.
1249 if ((sv->sv_flags & SV_ASLR) == 0 ||
1250 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1251 imgp->et_dyn_addr = __elfN(pie_base);
1252 else if ((__elfN(pie_aslr_enabled) &&
1253 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1254 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1255 imgp->et_dyn_addr = ET_DYN_ADDR_RAND;
1257 imgp->et_dyn_addr = __elfN(pie_base);
1262 * Avoid a possible deadlock if the current address space is destroyed
1263 * and that address space maps the locked vnode. In the common case,
1264 * the locked vnode's v_usecount is decremented but remains greater
1265 * than zero. Consequently, the vnode lock is not needed by vrele().
1266 * However, in cases where the vnode lock is external, such as nullfs,
1267 * v_usecount may become zero.
1269 * The VV_TEXT flag prevents modifications to the executable while
1270 * the vnode is unlocked.
1272 VOP_UNLOCK(imgp->vp);
1275 * Decide whether to enable randomization of user mappings.
1276 * First, reset user preferences for the setid binaries.
1277 * Then, account for the support of the randomization by the
1278 * ABI, by user preferences, and make special treatment for
1281 if (imgp->credential_setid) {
1282 PROC_LOCK(imgp->proc);
1283 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1284 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1285 PROC_UNLOCK(imgp->proc);
1287 if ((sv->sv_flags & SV_ASLR) == 0 ||
1288 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1289 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1290 KASSERT(imgp->et_dyn_addr != ET_DYN_ADDR_RAND,
1291 ("imgp->et_dyn_addr == RAND and !ASLR"));
1292 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1293 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1294 imgp->et_dyn_addr == ET_DYN_ADDR_RAND) {
1295 imgp->map_flags |= MAP_ASLR;
1297 * If user does not care about sbrk, utilize the bss
1298 * grow region for mappings as well. We can select
1299 * the base for the image anywere and still not suffer
1300 * from the fragmentation.
1302 if (!__elfN(aslr_honor_sbrk) ||
1303 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1304 imgp->map_flags |= MAP_ASLR_IGNSTART;
1305 if (__elfN(aslr_stack))
1306 imgp->map_flags |= MAP_ASLR_STACK;
1307 if (__elfN(aslr_shared_page))
1308 imgp->imgp_flags |= IMGP_ASLR_SHARED_PAGE;
1311 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1312 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1313 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1314 imgp->map_flags |= MAP_WXORX;
1316 error = exec_new_vmspace(imgp, sv);
1318 imgp->proc->p_sysent = sv;
1319 imgp->proc->p_elf_brandinfo = brand_info;
1321 vmspace = imgp->proc->p_vmspace;
1322 map = &vmspace->vm_map;
1323 maxv = sv->sv_usrstack;
1324 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1325 maxv -= lim_max(td, RLIMIT_STACK);
1326 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1327 uprintf("Excessive mapping size\n");
1331 if (error == 0 && imgp->et_dyn_addr == ET_DYN_ADDR_RAND) {
1332 KASSERT((map->flags & MAP_ASLR) != 0,
1333 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1334 error = __CONCAT(rnd_, __elfN(base))(map,
1335 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1336 /* reserve half of the address space to interpreter */
1337 maxv / 2, maxalign, &imgp->et_dyn_addr);
1340 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1344 error = __elfN(load_sections)(imgp, hdr, phdr, imgp->et_dyn_addr, NULL);
1348 error = __elfN(enforce_limits)(imgp, hdr, phdr);
1353 * We load the dynamic linker where a userland call
1354 * to mmap(0, ...) would put it. The rationale behind this
1355 * calculation is that it leaves room for the heap to grow to
1356 * its maximum allowed size.
1358 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1360 if ((map->flags & MAP_ASLR) != 0) {
1361 maxv1 = maxv / 2 + addr / 2;
1362 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1363 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1364 pagesizes[1] : pagesizes[0], &anon_loc);
1367 map->anon_loc = anon_loc;
1369 map->anon_loc = addr;
1372 entry = (u_long)hdr->e_entry + imgp->et_dyn_addr;
1373 imgp->entry_addr = entry;
1375 if (sv->sv_protect != NULL)
1376 sv->sv_protect(imgp, SVP_IMAGE);
1378 if (interp != NULL) {
1379 VOP_UNLOCK(imgp->vp);
1380 if ((map->flags & MAP_ASLR) != 0) {
1381 /* Assume that interpreter fits into 1/4 of AS */
1382 maxv1 = maxv / 2 + addr / 2;
1383 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1384 maxv1, PAGE_SIZE, &addr);
1387 error = __elfN(load_interp)(imgp, brand_info, interp,
1388 &addr, &imgp->entry_addr);
1390 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1394 addr = imgp->et_dyn_addr;
1396 error = exec_map_stack(imgp);
1401 * Construct auxargs table (used by the copyout_auxargs routine)
1403 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1404 if (elf_auxargs == NULL) {
1405 VOP_UNLOCK(imgp->vp);
1406 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1407 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1409 elf_auxargs->execfd = -1;
1410 elf_auxargs->phdr = proghdr + imgp->et_dyn_addr;
1411 elf_auxargs->phent = hdr->e_phentsize;
1412 elf_auxargs->phnum = hdr->e_phnum;
1413 elf_auxargs->pagesz = PAGE_SIZE;
1414 elf_auxargs->base = addr;
1415 elf_auxargs->flags = 0;
1416 elf_auxargs->entry = entry;
1417 elf_auxargs->hdr_eflags = hdr->e_flags;
1419 imgp->auxargs = elf_auxargs;
1420 imgp->interpreted = 0;
1421 imgp->reloc_base = addr;
1422 imgp->proc->p_osrel = osrel;
1423 imgp->proc->p_fctl0 = fctl0;
1424 imgp->proc->p_elf_flags = hdr->e_flags;
1427 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1429 free(interp, M_TEMP);
1433 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1436 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1438 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1439 Elf_Auxinfo *argarray, *pos;
1440 struct vmspace *vmspace;
1445 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1448 vmspace = imgp->proc->p_vmspace;
1450 if (args->execfd != -1)
1451 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1452 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1453 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1454 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1455 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1456 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1457 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1458 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1459 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1460 if (imgp->execpathp != 0)
1461 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1462 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1463 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1464 if (imgp->canary != 0) {
1465 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1466 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1468 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1469 if (imgp->pagesizes != 0) {
1470 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1471 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1473 if ((imgp->sysent->sv_flags & SV_TIMEKEEP) != 0) {
1474 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1475 vmspace->vm_shp_base + imgp->sysent->sv_timekeep_offset);
1477 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1478 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1479 imgp->sysent->sv_stackprot);
1480 if (imgp->sysent->sv_hwcap != NULL)
1481 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1482 if (imgp->sysent->sv_hwcap2 != NULL)
1483 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1485 bsdflags |= __elfN(sigfastblock) ? ELF_BSDF_SIGFASTBLK : 0;
1486 oc = atomic_load_int(&vm_overcommit);
1487 bsdflags |= (oc & (SWAP_RESERVE_FORCE_ON | SWAP_RESERVE_RLIMIT_ON)) !=
1488 0 ? ELF_BSDF_VMNOOVERCOMMIT : 0;
1489 AUXARGS_ENTRY(pos, AT_BSDFLAGS, bsdflags);
1490 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1491 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1492 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1493 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1494 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1495 #ifdef RANDOM_FENESTRASX
1496 if ((imgp->sysent->sv_flags & SV_RNG_SEED_VER) != 0) {
1497 AUXARGS_ENTRY(pos, AT_FXRNG,
1498 vmspace->vm_shp_base + imgp->sysent->sv_fxrng_gen_offset);
1501 if ((imgp->sysent->sv_flags & SV_DSO_SIG) != 0 && __elfN(vdso) != 0) {
1502 AUXARGS_ENTRY(pos, AT_KPRELOAD,
1503 vmspace->vm_shp_base + imgp->sysent->sv_vdso_offset);
1505 AUXARGS_ENTRY(pos, AT_USRSTACKBASE, round_page(vmspace->vm_stacktop));
1506 stacksz = imgp->proc->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur;
1507 AUXARGS_ENTRY(pos, AT_USRSTACKLIM, stacksz);
1508 AUXARGS_ENTRY(pos, AT_NULL, 0);
1510 free(imgp->auxargs, M_TEMP);
1511 imgp->auxargs = NULL;
1512 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1514 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1515 free(argarray, M_TEMP);
1520 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1524 base = (Elf_Addr *)*stack_base;
1526 if (elf_suword(base, imgp->args->argc) == -1)
1528 *stack_base = (uintptr_t)base;
1533 * Code for generating ELF core dumps.
1536 typedef void (*segment_callback)(vm_map_entry_t, void *);
1538 /* Closure for cb_put_phdr(). */
1539 struct phdr_closure {
1540 Elf_Phdr *phdr; /* Program header to fill in */
1541 Elf_Off offset; /* Offset of segment in core file */
1545 int type; /* Note type. */
1546 struct regset *regset; /* Register set. */
1547 outfunc_t outfunc; /* Output function. */
1548 void *outarg; /* Argument for the output function. */
1549 size_t outsize; /* Output size. */
1550 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1553 TAILQ_HEAD(note_info_list, note_info);
1555 extern int compress_user_cores;
1556 extern int compress_user_cores_level;
1558 static void cb_put_phdr(vm_map_entry_t, void *);
1559 static void cb_size_segment(vm_map_entry_t, void *);
1560 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1562 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1563 struct note_info_list *, size_t, int);
1564 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1566 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1567 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1568 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1569 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1570 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1571 static void note_procstat_files(void *, struct sbuf *, size_t *);
1572 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1573 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1574 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1575 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1576 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1579 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1582 return (core_write((struct coredump_params *)arg, base, len, offset,
1583 UIO_SYSSPACE, NULL));
1587 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1589 struct ucred *cred = td->td_ucred;
1590 int compm, error = 0;
1591 struct sseg_closure seginfo;
1592 struct note_info_list notelst;
1593 struct coredump_params params;
1594 struct note_info *ninfo;
1596 size_t hdrsize, notesz, coresize;
1600 TAILQ_INIT(¬elst);
1602 /* Size the program segments. */
1603 __elfN(size_segments)(td, &seginfo, flags);
1606 * Collect info about the core file header area.
1608 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1609 if (seginfo.count + 1 >= PN_XNUM)
1610 hdrsize += sizeof(Elf_Shdr);
1611 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1612 coresize = round_page(hdrsize + notesz) + seginfo.size;
1614 /* Set up core dump parameters. */
1616 params.active_cred = cred;
1617 params.file_cred = NOCRED;
1624 PROC_LOCK(td->td_proc);
1625 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1626 PROC_UNLOCK(td->td_proc);
1633 if (coresize >= limit) {
1638 /* Create a compression stream if necessary. */
1639 compm = compress_user_cores;
1640 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1642 compm = COMPRESS_GZIP;
1644 params.comp = compressor_init(core_compressed_write,
1645 compm, CORE_BUF_SIZE,
1646 compress_user_cores_level, ¶ms);
1647 if (params.comp == NULL) {
1651 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1655 * Allocate memory for building the header, fill it up,
1656 * and write it out following the notes.
1658 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1659 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1662 /* Write the contents of all of the writable segments. */
1668 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1669 offset = round_page(hdrsize + notesz);
1670 for (i = 0; i < seginfo.count; i++) {
1671 error = core_output((char *)(uintptr_t)php->p_vaddr,
1672 php->p_filesz, offset, ¶ms, tmpbuf);
1675 offset += php->p_filesz;
1678 if (error == 0 && params.comp != NULL)
1679 error = compressor_flush(params.comp);
1683 "Failed to write core file for process %s (error %d)\n",
1684 curproc->p_comm, error);
1688 free(tmpbuf, M_TEMP);
1689 if (params.comp != NULL)
1690 compressor_fini(params.comp);
1691 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1692 TAILQ_REMOVE(¬elst, ninfo, link);
1693 free(ninfo, M_TEMP);
1702 * A callback for each_dumpable_segment() to write out the segment's
1703 * program header entry.
1706 cb_put_phdr(vm_map_entry_t entry, void *closure)
1708 struct phdr_closure *phc = (struct phdr_closure *)closure;
1709 Elf_Phdr *phdr = phc->phdr;
1711 phc->offset = round_page(phc->offset);
1713 phdr->p_type = PT_LOAD;
1714 phdr->p_offset = phc->offset;
1715 phdr->p_vaddr = entry->start;
1717 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1718 phdr->p_align = PAGE_SIZE;
1719 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1721 phc->offset += phdr->p_filesz;
1726 * A callback for each_dumpable_segment() to gather information about
1727 * the number of segments and their total size.
1730 cb_size_segment(vm_map_entry_t entry, void *closure)
1732 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1735 ssc->size += entry->end - entry->start;
1739 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1745 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1749 * For each writable segment in the process's memory map, call the given
1750 * function with a pointer to the map entry and some arbitrary
1751 * caller-supplied data.
1754 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1757 struct proc *p = td->td_proc;
1758 vm_map_t map = &p->p_vmspace->vm_map;
1759 vm_map_entry_t entry;
1760 vm_object_t backing_object, object;
1763 vm_map_lock_read(map);
1764 VM_MAP_ENTRY_FOREACH(entry, map) {
1766 * Don't dump inaccessible mappings, deal with legacy
1769 * Note that read-only segments related to the elf binary
1770 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1771 * need to arbitrarily ignore such segments.
1773 if ((flags & SVC_ALL) == 0) {
1774 if (elf_legacy_coredump) {
1775 if ((entry->protection & VM_PROT_RW) !=
1779 if ((entry->protection & VM_PROT_ALL) == 0)
1785 * Dont include memory segment in the coredump if
1786 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1787 * madvise(2). Do not dump submaps (i.e. parts of the
1790 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1792 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1793 (flags & SVC_ALL) == 0)
1795 if ((object = entry->object.vm_object) == NULL)
1798 /* Ignore memory-mapped devices and such things. */
1799 VM_OBJECT_RLOCK(object);
1800 while ((backing_object = object->backing_object) != NULL) {
1801 VM_OBJECT_RLOCK(backing_object);
1802 VM_OBJECT_RUNLOCK(object);
1803 object = backing_object;
1805 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1806 VM_OBJECT_RUNLOCK(object);
1810 (*func)(entry, closure);
1812 vm_map_unlock_read(map);
1816 * Write the core file header to the file, including padding up to
1817 * the page boundary.
1820 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1821 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1824 struct note_info *ninfo;
1828 /* Fill in the header. */
1829 bzero(hdr, hdrsize);
1830 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1832 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1833 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1834 sbuf_start_section(sb, NULL);
1835 sbuf_bcat(sb, hdr, hdrsize);
1836 TAILQ_FOREACH(ninfo, notelst, link)
1837 __elfN(putnote)(p->td, ninfo, sb);
1838 /* Align up to a page boundary for the program segments. */
1839 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1840 error = sbuf_finish(sb);
1847 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1857 size += __elfN(register_note)(td, list, NT_PRPSINFO,
1858 __elfN(note_prpsinfo), p);
1861 * To have the debugger select the right thread (LWP) as the initial
1862 * thread, we dump the state of the thread passed to us in td first.
1863 * This is the thread that causes the core dump and thus likely to
1864 * be the right thread one wants to have selected in the debugger.
1867 while (thr != NULL) {
1868 size += __elfN(prepare_register_notes)(td, list, thr);
1869 size += __elfN(register_note)(td, list, -1,
1870 __elfN(note_threadmd), thr);
1872 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1873 TAILQ_NEXT(thr, td_plist);
1875 thr = TAILQ_NEXT(thr, td_plist);
1878 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1879 __elfN(note_procstat_proc), p);
1880 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1881 note_procstat_files, p);
1882 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1883 note_procstat_vmmap, p);
1884 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1885 note_procstat_groups, p);
1886 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1887 note_procstat_umask, p);
1888 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1889 note_procstat_rlimit, p);
1890 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1891 note_procstat_osrel, p);
1892 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1893 __elfN(note_procstat_psstrings), p);
1894 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1895 __elfN(note_procstat_auxv), p);
1901 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1902 size_t notesz, int flags)
1907 struct phdr_closure phc;
1910 ehdr = (Elf_Ehdr *)hdr;
1911 bi = td->td_proc->p_elf_brandinfo;
1913 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1914 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1915 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1916 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1917 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1918 ehdr->e_ident[EI_DATA] = ELF_DATA;
1919 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1920 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1921 ehdr->e_ident[EI_ABIVERSION] = 0;
1922 ehdr->e_ident[EI_PAD] = 0;
1923 ehdr->e_type = ET_CORE;
1924 ehdr->e_machine = bi->machine;
1925 ehdr->e_version = EV_CURRENT;
1927 ehdr->e_phoff = sizeof(Elf_Ehdr);
1928 ehdr->e_flags = td->td_proc->p_elf_flags;
1929 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1930 ehdr->e_phentsize = sizeof(Elf_Phdr);
1931 ehdr->e_shentsize = sizeof(Elf_Shdr);
1932 ehdr->e_shstrndx = SHN_UNDEF;
1933 if (numsegs + 1 < PN_XNUM) {
1934 ehdr->e_phnum = numsegs + 1;
1937 ehdr->e_phnum = PN_XNUM;
1940 ehdr->e_shoff = ehdr->e_phoff +
1941 (numsegs + 1) * ehdr->e_phentsize;
1942 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1943 ("e_shoff: %zu, hdrsize - shdr: %zu",
1944 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1946 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1947 memset(shdr, 0, sizeof(*shdr));
1949 * A special first section is used to hold large segment and
1950 * section counts. This was proposed by Sun Microsystems in
1951 * Solaris and has been adopted by Linux; the standard ELF
1952 * tools are already familiar with the technique.
1954 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1955 * (or 12-7 depending on the version of the document) for more
1958 shdr->sh_type = SHT_NULL;
1959 shdr->sh_size = ehdr->e_shnum;
1960 shdr->sh_link = ehdr->e_shstrndx;
1961 shdr->sh_info = numsegs + 1;
1965 * Fill in the program header entries.
1967 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1969 /* The note segement. */
1970 phdr->p_type = PT_NOTE;
1971 phdr->p_offset = hdrsize;
1974 phdr->p_filesz = notesz;
1976 phdr->p_flags = PF_R;
1977 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1980 /* All the writable segments from the program. */
1982 phc.offset = round_page(hdrsize + notesz);
1983 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1987 __elfN(register_regset_note)(struct thread *td, struct note_info_list *list,
1988 struct regset *regset, struct thread *target_td)
1990 const struct sysentvec *sv;
1991 struct note_info *ninfo;
1992 size_t size, notesize;
1995 if (!regset->get(regset, target_td, NULL, &size) || size == 0)
1998 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1999 ninfo->type = regset->note;
2000 ninfo->regset = regset;
2001 ninfo->outarg = target_td;
2002 ninfo->outsize = size;
2003 TAILQ_INSERT_TAIL(list, ninfo, link);
2005 sv = td->td_proc->p_sysent;
2006 notesize = sizeof(Elf_Note) + /* note header */
2007 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2009 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2015 __elfN(register_note)(struct thread *td, struct note_info_list *list,
2016 int type, outfunc_t out, void *arg)
2018 const struct sysentvec *sv;
2019 struct note_info *ninfo;
2020 size_t size, notesize;
2022 sv = td->td_proc->p_sysent;
2024 out(arg, NULL, &size);
2025 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2027 ninfo->outfunc = out;
2028 ninfo->outarg = arg;
2029 ninfo->outsize = size;
2030 TAILQ_INSERT_TAIL(list, ninfo, link);
2035 notesize = sizeof(Elf_Note) + /* note header */
2036 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
2038 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2044 append_note_data(const void *src, void *dst, size_t len)
2048 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2050 bcopy(src, dst, len);
2051 bzero((char *)dst + len, padded_len - len);
2053 return (padded_len);
2057 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2065 note = (Elf_Note *)buf;
2066 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2067 note->n_descsz = size;
2068 note->n_type = type;
2069 buf += sizeof(*note);
2070 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2071 sizeof(FREEBSD_ABI_VENDOR));
2072 append_note_data(src, buf, size);
2077 notesize = sizeof(Elf_Note) + /* note header */
2078 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2080 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2086 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2089 const struct sysentvec *sv;
2090 ssize_t old_len, sect_len;
2091 size_t new_len, descsz, i;
2093 if (ninfo->type == -1) {
2094 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2098 sv = td->td_proc->p_sysent;
2100 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2101 note.n_descsz = ninfo->outsize;
2102 note.n_type = ninfo->type;
2104 sbuf_bcat(sb, ¬e, sizeof(note));
2105 sbuf_start_section(sb, &old_len);
2106 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2107 strlen(sv->sv_elf_core_abi_vendor) + 1);
2108 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2109 if (note.n_descsz == 0)
2111 sbuf_start_section(sb, &old_len);
2112 if (ninfo->regset != NULL) {
2113 struct regset *regset = ninfo->regset;
2116 buf = malloc(ninfo->outsize, M_TEMP, M_ZERO | M_WAITOK);
2117 (void)regset->get(regset, ninfo->outarg, buf, &ninfo->outsize);
2118 sbuf_bcat(sb, buf, ninfo->outsize);
2121 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2122 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2126 new_len = (size_t)sect_len;
2127 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2128 if (new_len < descsz) {
2130 * It is expected that individual note emitters will correctly
2131 * predict their expected output size and fill up to that size
2132 * themselves, padding in a format-specific way if needed.
2133 * However, in case they don't, just do it here with zeros.
2135 for (i = 0; i < descsz - new_len; i++)
2137 } else if (new_len > descsz) {
2139 * We can't always truncate sb -- we may have drained some
2142 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2143 "read it (%zu > %zu). Since it is longer than "
2144 "expected, this coredump's notes are corrupt. THIS "
2145 "IS A BUG in the note_procstat routine for type %u.\n",
2146 __func__, (unsigned)note.n_type, new_len, descsz,
2147 (unsigned)note.n_type));
2152 * Miscellaneous note out functions.
2155 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2156 #include <compat/freebsd32/freebsd32.h>
2157 #include <compat/freebsd32/freebsd32_signal.h>
2159 typedef struct prstatus32 elf_prstatus_t;
2160 typedef struct prpsinfo32 elf_prpsinfo_t;
2161 typedef struct fpreg32 elf_prfpregset_t;
2162 typedef struct fpreg32 elf_fpregset_t;
2163 typedef struct reg32 elf_gregset_t;
2164 typedef struct thrmisc32 elf_thrmisc_t;
2165 typedef struct ptrace_lwpinfo32 elf_lwpinfo_t;
2166 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2167 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2168 typedef uint32_t elf_ps_strings_t;
2170 typedef prstatus_t elf_prstatus_t;
2171 typedef prpsinfo_t elf_prpsinfo_t;
2172 typedef prfpregset_t elf_prfpregset_t;
2173 typedef prfpregset_t elf_fpregset_t;
2174 typedef gregset_t elf_gregset_t;
2175 typedef thrmisc_t elf_thrmisc_t;
2176 typedef struct ptrace_lwpinfo elf_lwpinfo_t;
2177 #define ELF_KERN_PROC_MASK 0
2178 typedef struct kinfo_proc elf_kinfo_proc_t;
2179 typedef vm_offset_t elf_ps_strings_t;
2183 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2189 elf_prpsinfo_t *psinfo;
2194 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2195 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2196 psinfo->pr_version = PRPSINFO_VERSION;
2197 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2198 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2200 if (p->p_args != NULL) {
2201 len = sizeof(psinfo->pr_psargs) - 1;
2202 if (len > p->p_args->ar_length)
2203 len = p->p_args->ar_length;
2204 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2210 sbuf_new(&sbarg, psinfo->pr_psargs,
2211 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2212 error = proc_getargv(curthread, p, &sbarg);
2214 if (sbuf_finish(&sbarg) == 0) {
2215 len = sbuf_len(&sbarg);
2219 len = sizeof(psinfo->pr_psargs) - 1;
2221 sbuf_delete(&sbarg);
2223 if (error != 0 || len == 0 || (ssize_t)len == -1)
2224 strlcpy(psinfo->pr_psargs, p->p_comm,
2225 sizeof(psinfo->pr_psargs));
2227 KASSERT(len < sizeof(psinfo->pr_psargs),
2228 ("len is too long: %zu vs %zu", len,
2229 sizeof(psinfo->pr_psargs)));
2230 cp = psinfo->pr_psargs;
2233 cp = memchr(cp, '\0', end - cp);
2239 psinfo->pr_pid = p->p_pid;
2240 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2241 free(psinfo, M_TEMP);
2243 *sizep = sizeof(*psinfo);
2247 __elfN(get_prstatus)(struct regset *rs, struct thread *td, void *buf,
2250 elf_prstatus_t *status;
2253 KASSERT(*sizep == sizeof(*status), ("%s: invalid size",
2256 memset(status, 0, *sizep);
2257 status->pr_version = PRSTATUS_VERSION;
2258 status->pr_statussz = sizeof(elf_prstatus_t);
2259 status->pr_gregsetsz = sizeof(elf_gregset_t);
2260 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2261 status->pr_osreldate = osreldate;
2262 status->pr_cursig = td->td_proc->p_sig;
2263 status->pr_pid = td->td_tid;
2264 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2265 fill_regs32(td, &status->pr_reg);
2267 fill_regs(td, &status->pr_reg);
2270 *sizep = sizeof(*status);
2275 __elfN(set_prstatus)(struct regset *rs, struct thread *td, void *buf,
2278 elf_prstatus_t *status;
2280 KASSERT(size == sizeof(*status), ("%s: invalid size", __func__));
2282 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2283 set_regs32(td, &status->pr_reg);
2285 set_regs(td, &status->pr_reg);
2290 static struct regset __elfN(regset_prstatus) = {
2291 .note = NT_PRSTATUS,
2292 .size = sizeof(elf_prstatus_t),
2293 .get = __elfN(get_prstatus),
2294 .set = __elfN(set_prstatus),
2296 ELF_REGSET(__elfN(regset_prstatus));
2299 __elfN(get_fpregset)(struct regset *rs, struct thread *td, void *buf,
2302 elf_prfpregset_t *fpregset;
2305 KASSERT(*sizep == sizeof(*fpregset), ("%s: invalid size",
2308 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2309 fill_fpregs32(td, fpregset);
2311 fill_fpregs(td, fpregset);
2314 *sizep = sizeof(*fpregset);
2319 __elfN(set_fpregset)(struct regset *rs, struct thread *td, void *buf,
2322 elf_prfpregset_t *fpregset;
2325 KASSERT(size == sizeof(*fpregset), ("%s: invalid size", __func__));
2326 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2327 set_fpregs32(td, fpregset);
2329 set_fpregs(td, fpregset);
2334 static struct regset __elfN(regset_fpregset) = {
2335 .note = NT_FPREGSET,
2336 .size = sizeof(elf_prfpregset_t),
2337 .get = __elfN(get_fpregset),
2338 .set = __elfN(set_fpregset),
2340 ELF_REGSET(__elfN(regset_fpregset));
2343 __elfN(get_thrmisc)(struct regset *rs, struct thread *td, void *buf,
2346 elf_thrmisc_t *thrmisc;
2349 KASSERT(*sizep == sizeof(*thrmisc),
2350 ("%s: invalid size", __func__));
2352 bzero(thrmisc, sizeof(*thrmisc));
2353 strcpy(thrmisc->pr_tname, td->td_name);
2355 *sizep = sizeof(*thrmisc);
2359 static struct regset __elfN(regset_thrmisc) = {
2361 .size = sizeof(elf_thrmisc_t),
2362 .get = __elfN(get_thrmisc),
2364 ELF_REGSET(__elfN(regset_thrmisc));
2367 __elfN(get_lwpinfo)(struct regset *rs, struct thread *td, void *buf,
2374 size = sizeof(structsize) + sizeof(pl);
2376 KASSERT(*sizep == size, ("%s: invalid size", __func__));
2377 structsize = sizeof(pl);
2378 memcpy(buf, &structsize, sizeof(structsize));
2379 bzero(&pl, sizeof(pl));
2380 pl.pl_lwpid = td->td_tid;
2381 pl.pl_event = PL_EVENT_NONE;
2382 pl.pl_sigmask = td->td_sigmask;
2383 pl.pl_siglist = td->td_siglist;
2384 if (td->td_si.si_signo != 0) {
2385 pl.pl_event = PL_EVENT_SIGNAL;
2386 pl.pl_flags |= PL_FLAG_SI;
2387 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2388 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2390 pl.pl_siginfo = td->td_si;
2393 strcpy(pl.pl_tdname, td->td_name);
2394 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2395 memcpy((int *)buf + 1, &pl, sizeof(pl));
2401 static struct regset __elfN(regset_lwpinfo) = {
2402 .note = NT_PTLWPINFO,
2403 .size = sizeof(int) + sizeof(elf_lwpinfo_t),
2404 .get = __elfN(get_lwpinfo),
2406 ELF_REGSET(__elfN(regset_lwpinfo));
2409 __elfN(prepare_register_notes)(struct thread *td, struct note_info_list *list,
2410 struct thread *target_td)
2412 struct sysentvec *sv = td->td_proc->p_sysent;
2413 struct regset **regsetp, **regset_end, *regset;
2418 /* NT_PRSTATUS must be the first register set note. */
2419 size += __elfN(register_regset_note)(td, list, &__elfN(regset_prstatus),
2422 regsetp = sv->sv_regset_begin;
2423 if (regsetp == NULL) {
2424 /* XXX: This shouldn't be true for any FreeBSD ABIs. */
2425 size += __elfN(register_regset_note)(td, list,
2426 &__elfN(regset_fpregset), target_td);
2429 regset_end = sv->sv_regset_end;
2430 MPASS(regset_end != NULL);
2431 for (; regsetp < regset_end; regsetp++) {
2433 if (regset->note == NT_PRSTATUS)
2435 size += __elfN(register_regset_note)(td, list, regset,
2442 * Allow for MD specific notes, as well as any MD
2443 * specific preparations for writing MI notes.
2446 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2452 td = (struct thread *)arg;
2454 if (size != 0 && sb != NULL)
2455 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2459 __elfN(dump_thread)(td, buf, &size);
2460 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2461 if (size != 0 && sb != NULL)
2462 sbuf_bcat(sb, buf, size);
2467 #ifdef KINFO_PROC_SIZE
2468 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2472 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2479 size = sizeof(structsize) + p->p_numthreads *
2480 sizeof(elf_kinfo_proc_t);
2483 KASSERT(*sizep == size, ("invalid size"));
2484 structsize = sizeof(elf_kinfo_proc_t);
2485 sbuf_bcat(sb, &structsize, sizeof(structsize));
2486 sx_slock(&proctree_lock);
2488 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2489 sx_sunlock(&proctree_lock);
2494 #ifdef KINFO_FILE_SIZE
2495 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2499 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2502 size_t size, sect_sz, i;
2503 ssize_t start_len, sect_len;
2504 int structsize, filedesc_flags;
2506 if (coredump_pack_fileinfo)
2507 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2512 structsize = sizeof(struct kinfo_file);
2515 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2516 sbuf_set_drain(sb, sbuf_count_drain, &size);
2517 sbuf_bcat(sb, &structsize, sizeof(structsize));
2519 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2524 sbuf_start_section(sb, &start_len);
2526 sbuf_bcat(sb, &structsize, sizeof(structsize));
2528 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2531 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2536 KASSERT(sect_sz <= *sizep,
2537 ("kern_proc_filedesc_out did not respect maxlen; "
2538 "requested %zu, got %zu", *sizep - sizeof(structsize),
2539 sect_sz - sizeof(structsize)));
2541 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2546 #ifdef KINFO_VMENTRY_SIZE
2547 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2551 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2555 int structsize, vmmap_flags;
2557 if (coredump_pack_vmmapinfo)
2558 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2563 structsize = sizeof(struct kinfo_vmentry);
2566 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2567 sbuf_set_drain(sb, sbuf_count_drain, &size);
2568 sbuf_bcat(sb, &structsize, sizeof(structsize));
2570 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2575 sbuf_bcat(sb, &structsize, sizeof(structsize));
2577 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2583 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2590 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2592 KASSERT(*sizep == size, ("invalid size"));
2593 structsize = sizeof(gid_t);
2594 sbuf_bcat(sb, &structsize, sizeof(structsize));
2595 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2602 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2609 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2611 KASSERT(*sizep == size, ("invalid size"));
2612 structsize = sizeof(p->p_pd->pd_cmask);
2613 sbuf_bcat(sb, &structsize, sizeof(structsize));
2614 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2620 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2623 struct rlimit rlim[RLIM_NLIMITS];
2628 size = sizeof(structsize) + sizeof(rlim);
2630 KASSERT(*sizep == size, ("invalid size"));
2631 structsize = sizeof(rlim);
2632 sbuf_bcat(sb, &structsize, sizeof(structsize));
2634 for (i = 0; i < RLIM_NLIMITS; i++)
2635 lim_rlimit_proc(p, i, &rlim[i]);
2637 sbuf_bcat(sb, rlim, sizeof(rlim));
2643 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2650 size = sizeof(structsize) + sizeof(p->p_osrel);
2652 KASSERT(*sizep == size, ("invalid size"));
2653 structsize = sizeof(p->p_osrel);
2654 sbuf_bcat(sb, &structsize, sizeof(structsize));
2655 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2661 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2664 elf_ps_strings_t ps_strings;
2669 size = sizeof(structsize) + sizeof(ps_strings);
2671 KASSERT(*sizep == size, ("invalid size"));
2672 structsize = sizeof(ps_strings);
2673 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2674 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2676 ps_strings = PROC_PS_STRINGS(p);
2678 sbuf_bcat(sb, &structsize, sizeof(structsize));
2679 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2685 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2694 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2696 sbuf_set_drain(sb, sbuf_count_drain, &size);
2697 sbuf_bcat(sb, &structsize, sizeof(structsize));
2699 proc_getauxv(curthread, p, sb);
2705 structsize = sizeof(Elf_Auxinfo);
2706 sbuf_bcat(sb, &structsize, sizeof(structsize));
2708 proc_getauxv(curthread, p, sb);
2713 #define MAX_NOTES_LOOP 4096
2715 __elfN(parse_notes)(const struct image_params *imgp, const Elf_Note *checknote,
2716 const char *note_vendor, const Elf_Phdr *pnote,
2717 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2719 const Elf_Note *note, *note0, *note_end;
2720 const char *note_name;
2725 /* We need some limit, might as well use PAGE_SIZE. */
2726 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2728 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2729 if (pnote->p_offset > PAGE_SIZE ||
2730 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2731 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2733 VOP_UNLOCK(imgp->vp);
2734 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2735 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2737 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2738 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2739 curthread->td_ucred, NOCRED, NULL, curthread);
2741 uprintf("i/o error PT_NOTE\n");
2744 note = note0 = (const Elf_Note *)buf;
2745 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2747 note = note0 = (const Elf_Note *)(imgp->image_header +
2749 note_end = (const Elf_Note *)(imgp->image_header +
2750 pnote->p_offset + pnote->p_filesz);
2753 for (i = 0; i < MAX_NOTES_LOOP && note >= note0 && note < note_end;
2755 if (!aligned(note, Elf32_Addr)) {
2756 uprintf("Unaligned ELF note\n");
2759 if ((const char *)note_end - (const char *)note <
2761 uprintf("ELF note to short\n");
2764 if (note->n_namesz != checknote->n_namesz ||
2765 note->n_descsz != checknote->n_descsz ||
2766 note->n_type != checknote->n_type)
2768 note_name = (const char *)(note + 1);
2769 if (note_name + checknote->n_namesz >=
2770 (const char *)note_end || strncmp(note_vendor,
2771 note_name, checknote->n_namesz) != 0)
2774 if (cb(note, cb_arg, &res))
2777 note = (const Elf_Note *)((const char *)(note + 1) +
2778 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2779 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2781 if (i >= MAX_NOTES_LOOP)
2782 uprintf("ELF note parser reached %d notes\n", i);
2790 struct brandnote_cb_arg {
2791 Elf_Brandnote *brandnote;
2796 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2798 struct brandnote_cb_arg *arg;
2803 * Fetch the osreldate for binary from the ELF OSABI-note if
2806 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2807 arg->brandnote->trans_osrel != NULL ?
2808 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2813 static Elf_Note fctl_note = {
2814 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2815 .n_descsz = sizeof(uint32_t),
2816 .n_type = NT_FREEBSD_FEATURE_CTL,
2819 struct fctl_cb_arg {
2825 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2827 struct fctl_cb_arg *arg;
2828 const Elf32_Word *desc;
2832 p = (uintptr_t)(note + 1);
2833 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2834 desc = (const Elf32_Word *)p;
2835 *arg->has_fctl0 = true;
2836 *arg->fctl0 = desc[0];
2842 * Try to find the appropriate ABI-note section for checknote, fetch
2843 * the osreldate and feature control flags for binary from the ELF
2844 * OSABI-note. Only the first page of the image is searched, the same
2848 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2849 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2851 const Elf_Phdr *phdr;
2852 const Elf_Ehdr *hdr;
2853 struct brandnote_cb_arg b_arg;
2854 struct fctl_cb_arg f_arg;
2857 hdr = (const Elf_Ehdr *)imgp->image_header;
2858 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2859 b_arg.brandnote = brandnote;
2860 b_arg.osrel = osrel;
2861 f_arg.has_fctl0 = has_fctl0;
2862 f_arg.fctl0 = fctl0;
2864 for (i = 0; i < hdr->e_phnum; i++) {
2865 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2866 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2868 for (j = 0; j < hdr->e_phnum; j++) {
2869 if (phdr[j].p_type == PT_NOTE &&
2870 __elfN(parse_notes)(imgp, &fctl_note,
2871 FREEBSD_ABI_VENDOR, &phdr[j],
2872 note_fctl_cb, &f_arg))
2883 * Tell kern_execve.c about it, with a little help from the linker.
2885 static struct execsw __elfN(execsw) = {
2886 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2887 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2889 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2892 __elfN(trans_prot)(Elf_Word flags)
2898 prot |= VM_PROT_EXECUTE;
2900 prot |= VM_PROT_WRITE;
2902 prot |= VM_PROT_READ;
2903 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2904 if (i386_read_exec && (flags & PF_R))
2905 prot |= VM_PROT_EXECUTE;
2911 __elfN(untrans_prot)(vm_prot_t prot)
2916 if (prot & VM_PROT_EXECUTE)
2918 if (prot & VM_PROT_READ)
2920 if (prot & VM_PROT_WRITE)