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>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
60 #include <sys/sf_buf.h>
62 #include <sys/systm.h>
63 #include <sys/signalvar.h>
66 #include <sys/syscall.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/vnode.h>
70 #include <sys/syslog.h>
71 #include <sys/eventhandler.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
85 #define ELF_NOTE_ROUNDSIZE 4
86 #define OLD_EI_BRAND 8
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90 const char *interp, int32_t *osrel, uint32_t *fctl0);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
94 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
106 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 #if defined(__amd64__)
132 static int __elfN(vdso) = 1;
133 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
134 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
135 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
137 static int __elfN(vdso) = 0;
140 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
141 int i386_read_exec = 0;
142 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
143 "enable execution from readable segments");
146 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
148 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
153 val = __elfN(pie_base);
154 error = sysctl_handle_long(oidp, &val, 0, req);
155 if (error != 0 || req->newptr == NULL)
157 if ((val & PAGE_MASK) != 0)
159 __elfN(pie_base) = val;
162 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
163 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
164 sysctl_pie_base, "LU",
165 "PIE load base without randomization");
167 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
172 static int __elfN(aslr_enabled) = 0;
173 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
174 &__elfN(aslr_enabled), 0,
175 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
176 ": enable address map randomization");
178 static int __elfN(pie_aslr_enabled) = 0;
179 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
180 &__elfN(pie_aslr_enabled), 0,
181 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
182 ": enable address map randomization for PIE binaries");
184 static int __elfN(aslr_honor_sbrk) = 1;
185 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
186 &__elfN(aslr_honor_sbrk), 0,
187 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
189 static int __elfN(aslr_stack) = 1;
190 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
191 &__elfN(aslr_stack), 0,
192 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
193 ": enable stack address randomization");
195 static int __elfN(sigfastblock) = 1;
196 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
197 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
198 "enable sigfastblock for new processes");
200 static bool __elfN(allow_wx) = true;
201 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
202 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
203 "Allow pages to be mapped simultaneously writable and executable");
205 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
207 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
209 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
211 Elf_Brandnote __elfN(freebsd_brandnote) = {
212 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
213 .hdr.n_descsz = sizeof(int32_t),
214 .hdr.n_type = NT_FREEBSD_ABI_TAG,
215 .vendor = FREEBSD_ABI_VENDOR,
216 .flags = BN_TRANSLATE_OSREL,
217 .trans_osrel = __elfN(freebsd_trans_osrel)
221 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
225 p = (uintptr_t)(note + 1);
226 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
227 *osrel = *(const int32_t *)(p);
232 static const char GNU_ABI_VENDOR[] = "GNU";
233 static int GNU_KFREEBSD_ABI_DESC = 3;
235 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
236 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
237 .hdr.n_descsz = 16, /* XXX at least 16 */
239 .vendor = GNU_ABI_VENDOR,
240 .flags = BN_TRANSLATE_OSREL,
241 .trans_osrel = kfreebsd_trans_osrel
245 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
247 const Elf32_Word *desc;
250 p = (uintptr_t)(note + 1);
251 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
253 desc = (const Elf32_Word *)p;
254 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
258 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
259 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
261 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
267 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
271 for (i = 0; i < MAX_BRANDS; i++) {
272 if (elf_brand_list[i] == NULL) {
273 elf_brand_list[i] = entry;
277 if (i == MAX_BRANDS) {
278 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
286 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
290 for (i = 0; i < MAX_BRANDS; i++) {
291 if (elf_brand_list[i] == entry) {
292 elf_brand_list[i] = NULL;
302 __elfN(brand_inuse)(Elf_Brandinfo *entry)
307 sx_slock(&allproc_lock);
308 FOREACH_PROC_IN_SYSTEM(p) {
309 if (p->p_sysent == entry->sysvec) {
314 sx_sunlock(&allproc_lock);
319 static Elf_Brandinfo *
320 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
321 int32_t *osrel, uint32_t *fctl0)
323 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
324 Elf_Brandinfo *bi, *bi_m;
325 boolean_t ret, has_fctl0;
326 int i, interp_name_len;
328 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
331 * We support four types of branding -- (1) the ELF EI_OSABI field
332 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
333 * branding w/in the ELF header, (3) path of the `interp_path'
334 * field, and (4) the ".note.ABI-tag" ELF section.
337 /* Look for an ".note.ABI-tag" ELF section */
339 for (i = 0; i < MAX_BRANDS; i++) {
340 bi = elf_brand_list[i];
343 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
345 if (hdr->e_machine == bi->machine && (bi->flags &
346 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
350 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
352 /* Give brand a chance to veto check_note's guess */
353 if (ret && bi->header_supported) {
354 ret = bi->header_supported(imgp, osrel,
355 has_fctl0 ? fctl0 : NULL);
358 * If note checker claimed the binary, but the
359 * interpreter path in the image does not
360 * match default one for the brand, try to
361 * search for other brands with the same
362 * interpreter. Either there is better brand
363 * with the right interpreter, or, failing
364 * this, we return first brand which accepted
365 * our note and, optionally, header.
367 if (ret && bi_m == NULL && interp != NULL &&
368 (bi->interp_path == NULL ||
369 (strlen(bi->interp_path) + 1 != interp_name_len ||
370 strncmp(interp, bi->interp_path, interp_name_len)
382 /* If the executable has a brand, search for it in the brand list. */
383 for (i = 0; i < MAX_BRANDS; i++) {
384 bi = elf_brand_list[i];
385 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
386 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
388 if (hdr->e_machine == bi->machine &&
389 (hdr->e_ident[EI_OSABI] == bi->brand ||
390 (bi->compat_3_brand != NULL &&
391 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
392 bi->compat_3_brand) == 0))) {
393 /* Looks good, but give brand a chance to veto */
394 if (bi->header_supported == NULL ||
395 bi->header_supported(imgp, NULL, NULL)) {
397 * Again, prefer strictly matching
400 if (interp_name_len == 0 &&
401 bi->interp_path == NULL)
403 if (bi->interp_path != NULL &&
404 strlen(bi->interp_path) + 1 ==
405 interp_name_len && strncmp(interp,
406 bi->interp_path, interp_name_len) == 0)
416 /* No known brand, see if the header is recognized by any brand */
417 for (i = 0; i < MAX_BRANDS; i++) {
418 bi = elf_brand_list[i];
419 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
420 bi->header_supported == NULL)
422 if (hdr->e_machine == bi->machine) {
423 ret = bi->header_supported(imgp, NULL, NULL);
429 /* Lacking a known brand, search for a recognized interpreter. */
430 if (interp != NULL) {
431 for (i = 0; i < MAX_BRANDS; i++) {
432 bi = elf_brand_list[i];
433 if (bi == NULL || (bi->flags &
434 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
437 if (hdr->e_machine == bi->machine &&
438 bi->interp_path != NULL &&
439 /* ELF image p_filesz includes terminating zero */
440 strlen(bi->interp_path) + 1 == interp_name_len &&
441 strncmp(interp, bi->interp_path, interp_name_len)
442 == 0 && (bi->header_supported == NULL ||
443 bi->header_supported(imgp, NULL, NULL)))
448 /* Lacking a recognized interpreter, try the default brand */
449 for (i = 0; i < MAX_BRANDS; i++) {
450 bi = elf_brand_list[i];
451 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
452 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
454 if (hdr->e_machine == bi->machine &&
455 __elfN(fallback_brand) == bi->brand &&
456 (bi->header_supported == NULL ||
457 bi->header_supported(imgp, NULL, NULL)))
464 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
466 return (hdr->e_phoff <= PAGE_SIZE &&
467 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
471 __elfN(check_header)(const Elf_Ehdr *hdr)
477 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
478 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
479 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
480 hdr->e_phentsize != sizeof(Elf_Phdr) ||
481 hdr->e_version != ELF_TARG_VER)
485 * Make sure we have at least one brand for this machine.
488 for (i = 0; i < MAX_BRANDS; i++) {
489 bi = elf_brand_list[i];
490 if (bi != NULL && bi->machine == hdr->e_machine)
500 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
501 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
508 * Create the page if it doesn't exist yet. Ignore errors.
510 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
511 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
514 * Find the page from the underlying object.
516 if (object != NULL) {
517 sf = vm_imgact_map_page(object, offset);
519 return (KERN_FAILURE);
520 off = offset - trunc_page(offset);
521 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
523 vm_imgact_unmap_page(sf);
525 return (KERN_FAILURE);
528 return (KERN_SUCCESS);
532 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
533 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
539 int error, locked, rv;
541 if (start != trunc_page(start)) {
542 rv = __elfN(map_partial)(map, object, offset, start,
543 round_page(start), prot);
544 if (rv != KERN_SUCCESS)
546 offset += round_page(start) - start;
547 start = round_page(start);
549 if (end != round_page(end)) {
550 rv = __elfN(map_partial)(map, object, offset +
551 trunc_page(end) - start, trunc_page(end), end, prot);
552 if (rv != KERN_SUCCESS)
554 end = trunc_page(end);
557 return (KERN_SUCCESS);
558 if ((offset & PAGE_MASK) != 0) {
560 * The mapping is not page aligned. This means that we have
563 rv = vm_map_fixed(map, NULL, 0, start, end - start,
564 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
565 if (rv != KERN_SUCCESS)
568 return (KERN_SUCCESS);
569 for (; start < end; start += sz) {
570 sf = vm_imgact_map_page(object, offset);
572 return (KERN_FAILURE);
573 off = offset - trunc_page(offset);
575 if (sz > PAGE_SIZE - off)
576 sz = PAGE_SIZE - off;
577 error = copyout((caddr_t)sf_buf_kva(sf) + off,
579 vm_imgact_unmap_page(sf);
581 return (KERN_FAILURE);
585 vm_object_reference(object);
586 rv = vm_map_fixed(map, object, offset, start, end - start,
587 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
588 (object != NULL ? MAP_VN_EXEC : 0));
589 if (rv != KERN_SUCCESS) {
590 locked = VOP_ISLOCKED(imgp->vp);
591 VOP_UNLOCK(imgp->vp);
592 vm_object_deallocate(object);
593 vn_lock(imgp->vp, locked | LK_RETRY);
595 } else if (object != NULL) {
596 MPASS(imgp->vp->v_object == object);
597 VOP_SET_TEXT_CHECKED(imgp->vp);
600 return (KERN_SUCCESS);
604 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
605 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
611 vm_offset_t map_addr;
614 vm_ooffset_t file_addr;
617 * It's necessary to fail if the filsz + offset taken from the
618 * header is greater than the actual file pager object's size.
619 * If we were to allow this, then the vm_map_find() below would
620 * walk right off the end of the file object and into the ether.
622 * While I'm here, might as well check for something else that
623 * is invalid: filsz cannot be greater than memsz.
625 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
627 uprintf("elf_load_section: truncated ELF file\n");
631 object = imgp->object;
632 map = &imgp->proc->p_vmspace->vm_map;
633 map_addr = trunc_page((vm_offset_t)vmaddr);
634 file_addr = trunc_page(offset);
637 * We have two choices. We can either clear the data in the last page
638 * of an oversized mapping, or we can start the anon mapping a page
639 * early and copy the initialized data into that first page. We
644 else if (memsz > filsz)
645 map_len = trunc_page(offset + filsz) - file_addr;
647 map_len = round_page(offset + filsz) - file_addr;
650 /* cow flags: don't dump readonly sections in core */
651 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
652 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
654 rv = __elfN(map_insert)(imgp, map, object, file_addr,
655 map_addr, map_addr + map_len, prot, cow);
656 if (rv != KERN_SUCCESS)
659 /* we can stop now if we've covered it all */
665 * We have to get the remaining bit of the file into the first part
666 * of the oversized map segment. This is normally because the .data
667 * segment in the file is extended to provide bss. It's a neat idea
668 * to try and save a page, but it's a pain in the behind to implement.
670 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
672 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
673 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
675 /* This had damn well better be true! */
677 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
678 map_addr + map_len, prot, 0);
679 if (rv != KERN_SUCCESS)
684 sf = vm_imgact_map_page(object, offset + filsz);
688 /* send the page fragment to user space */
689 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
691 vm_imgact_unmap_page(sf);
697 * Remove write access to the page if it was only granted by map_insert
700 if ((prot & VM_PROT_WRITE) == 0)
701 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
702 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
708 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
709 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
716 ASSERT_VOP_LOCKED(imgp->vp, __func__);
721 for (i = 0; i < hdr->e_phnum; i++) {
722 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
725 /* Loadable segment */
726 prot = __elfN(trans_prot)(phdr[i].p_flags);
727 error = __elfN(load_section)(imgp, phdr[i].p_offset,
728 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
729 phdr[i].p_memsz, phdr[i].p_filesz, prot);
734 * Establish the base address if this is the first segment.
737 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
742 if (base_addrp != NULL)
743 *base_addrp = base_addr;
749 * Load the file "file" into memory. It may be either a shared object
752 * The "addr" reference parameter is in/out. On entry, it specifies
753 * the address where a shared object should be loaded. If the file is
754 * an executable, this value is ignored. On exit, "addr" specifies
755 * where the file was actually loaded.
757 * The "entry" reference parameter is out only. On exit, it specifies
758 * the entry point for the loaded file.
761 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
767 struct image_params image_params;
769 const Elf_Ehdr *hdr = NULL;
770 const Elf_Phdr *phdr = NULL;
771 struct nameidata *nd;
773 struct image_params *imgp;
775 u_long base_addr = 0;
778 #ifdef CAPABILITY_MODE
780 * XXXJA: This check can go away once we are sufficiently confident
781 * that the checks in namei() are correct.
783 if (IN_CAPABILITY_MODE(curthread))
787 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
789 attr = &tempdata->attr;
790 imgp = &tempdata->image_params;
793 * Initialize part of the common data
798 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
799 UIO_SYSSPACE, file, curthread);
800 if ((error = namei(nd)) != 0) {
804 NDFREE(nd, NDF_ONLY_PNBUF);
805 imgp->vp = nd->ni_vp;
808 * Check permissions, modes, uid, etc on the file, and "open" it.
810 error = exec_check_permissions(imgp);
814 error = exec_map_first_page(imgp);
818 imgp->object = nd->ni_vp->v_object;
820 hdr = (const Elf_Ehdr *)imgp->image_header;
821 if ((error = __elfN(check_header)(hdr)) != 0)
823 if (hdr->e_type == ET_DYN)
825 else if (hdr->e_type == ET_EXEC)
832 /* Only support headers that fit within first page for now */
833 if (!__elfN(phdr_in_zero_page)(hdr)) {
838 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
839 if (!aligned(phdr, Elf_Addr)) {
844 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
849 *entry = (unsigned long)hdr->e_entry + rbase;
853 exec_unmap_first_page(imgp);
857 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
860 free(tempdata, M_TEMP);
866 * Select randomized valid address in the map map, between minv and
867 * maxv, with specified alignment. The [minv, maxv) range must belong
868 * to the map. Note that function only allocates the address, it is
869 * up to caller to clamp maxv in a way that the final allocation
870 * length fit into the map.
872 * Result is returned in *resp, error code indicates that arguments
873 * did not pass sanity checks for overflow and range correctness.
876 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
877 u_int align, u_long *resp)
881 MPASS(vm_map_min(map) <= minv);
883 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
884 uprintf("Invalid ELF segments layout\n");
888 arc4rand(&rbase, sizeof(rbase), 0);
889 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
890 res &= ~((u_long)align - 1);
895 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
896 res, minv, maxv, rbase));
898 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
899 res, maxv, minv, rbase));
906 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
907 const Elf_Phdr *phdr, u_long et_dyn_addr)
909 struct vmspace *vmspace;
911 u_long text_size, data_size, total_size, text_addr, data_addr;
912 u_long seg_size, seg_addr;
916 text_size = data_size = total_size = text_addr = data_addr = 0;
918 for (i = 0; i < hdr->e_phnum; i++) {
919 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
922 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
923 seg_size = round_page(phdr[i].p_memsz +
924 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
927 * Make the largest executable segment the official
928 * text segment and all others data.
930 * Note that obreak() assumes that data_addr + data_size == end
931 * of data load area, and the ELF file format expects segments
932 * to be sorted by address. If multiple data segments exist,
933 * the last one will be used.
936 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
937 text_size = seg_size;
938 text_addr = seg_addr;
940 data_size = seg_size;
941 data_addr = seg_addr;
943 total_size += seg_size;
946 if (data_addr == 0 && data_size == 0) {
947 data_addr = text_addr;
948 data_size = text_size;
952 * Check limits. It should be safe to check the
953 * limits after loading the segments since we do
954 * not actually fault in all the segments pages.
956 PROC_LOCK(imgp->proc);
957 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
958 err_str = "Data segment size exceeds process limit";
959 else if (text_size > maxtsiz)
960 err_str = "Text segment size exceeds system limit";
961 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
962 err_str = "Total segment size exceeds process limit";
963 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
964 err_str = "Data segment size exceeds resource limit";
965 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
966 err_str = "Total segment size exceeds resource limit";
967 PROC_UNLOCK(imgp->proc);
968 if (err_str != NULL) {
969 uprintf("%s\n", err_str);
973 vmspace = imgp->proc->p_vmspace;
974 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
975 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
976 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
977 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
983 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
984 char **interpp, bool *free_interpp)
988 int error, interp_name_len;
990 KASSERT(phdr->p_type == PT_INTERP,
991 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
992 ASSERT_VOP_LOCKED(imgp->vp, __func__);
996 /* Path to interpreter */
997 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
998 uprintf("Invalid PT_INTERP\n");
1002 interp_name_len = phdr->p_filesz;
1003 if (phdr->p_offset > PAGE_SIZE ||
1004 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1006 * The vnode lock might be needed by the pagedaemon to
1007 * clean pages owned by the vnode. Do not allow sleep
1008 * waiting for memory with the vnode locked, instead
1009 * try non-sleepable allocation first, and if it
1010 * fails, go to the slow path were we drop the lock
1011 * and do M_WAITOK. A text reference prevents
1012 * modifications to the vnode content.
1014 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1015 if (interp == NULL) {
1016 VOP_UNLOCK(imgp->vp);
1017 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1018 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1021 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1022 interp_name_len, phdr->p_offset,
1023 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1026 free(interp, M_TEMP);
1027 uprintf("i/o error PT_INTERP %d\n", error);
1030 interp[interp_name_len] = '\0';
1033 *free_interpp = true;
1037 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1038 if (interp[interp_name_len - 1] != '\0') {
1039 uprintf("Invalid PT_INTERP\n");
1044 *free_interpp = false;
1049 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1050 const char *interp, u_long *addr, u_long *entry)
1055 if (brand_info->emul_path != NULL &&
1056 brand_info->emul_path[0] != '\0') {
1057 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1058 snprintf(path, MAXPATHLEN, "%s%s",
1059 brand_info->emul_path, interp);
1060 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1066 if (brand_info->interp_newpath != NULL &&
1067 (brand_info->interp_path == NULL ||
1068 strcmp(interp, brand_info->interp_path) == 0)) {
1069 error = __elfN(load_file)(imgp->proc,
1070 brand_info->interp_newpath, addr, entry);
1075 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1079 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1084 * Impossible et_dyn_addr initial value indicating that the real base
1085 * must be calculated later with some randomization applied.
1087 #define ET_DYN_ADDR_RAND 1
1090 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1093 const Elf_Ehdr *hdr;
1094 const Elf_Phdr *phdr;
1095 Elf_Auxargs *elf_auxargs;
1096 struct vmspace *vmspace;
1099 Elf_Brandinfo *brand_info;
1100 struct sysentvec *sv;
1101 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1102 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1108 hdr = (const Elf_Ehdr *)imgp->image_header;
1111 * Do we have a valid ELF header ?
1113 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1114 * if particular brand doesn't support it.
1116 if (__elfN(check_header)(hdr) != 0 ||
1117 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1121 * From here on down, we return an errno, not -1, as we've
1122 * detected an ELF file.
1125 if (!__elfN(phdr_in_zero_page)(hdr)) {
1126 uprintf("Program headers not in the first page\n");
1129 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1130 if (!aligned(phdr, Elf_Addr)) {
1131 uprintf("Unaligned program headers\n");
1139 entry = proghdr = 0;
1141 free_interp = false;
1145 * Somewhat arbitrary, limit accepted max alignment for the
1146 * loadable segment to the max supported superpage size. Too
1147 * large alignment requests are not useful and are indicators
1148 * of corrupted or outright malicious binary.
1150 maxalign = PAGE_SIZE;
1151 maxsalign = PAGE_SIZE * 1024;
1152 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1153 if (pagesizes[i] > maxsalign)
1154 maxsalign = pagesizes[i];
1159 for (i = 0; i < hdr->e_phnum; i++) {
1160 switch (phdr[i].p_type) {
1163 baddr = phdr[i].p_vaddr;
1164 if (!powerof2(phdr[i].p_align) ||
1165 phdr[i].p_align > maxsalign) {
1166 uprintf("Invalid segment alignment\n");
1170 if (phdr[i].p_align > maxalign)
1171 maxalign = phdr[i].p_align;
1172 if (mapsz + phdr[i].p_memsz < mapsz) {
1173 uprintf("Mapsize overflow\n");
1177 mapsz += phdr[i].p_memsz;
1181 * If this segment contains the program headers,
1182 * remember their virtual address for the AT_PHDR
1183 * aux entry. Static binaries don't usually include
1186 if (phdr[i].p_offset == 0 &&
1187 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1189 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1192 /* Path to interpreter */
1193 if (interp != NULL) {
1194 uprintf("Multiple PT_INTERP headers\n");
1198 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1204 if (__elfN(nxstack))
1206 __elfN(trans_prot)(phdr[i].p_flags);
1207 imgp->stack_sz = phdr[i].p_memsz;
1209 case PT_PHDR: /* Program header table info */
1210 proghdr = phdr[i].p_vaddr;
1215 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1216 if (brand_info == NULL) {
1217 uprintf("ELF binary type \"%u\" not known.\n",
1218 hdr->e_ident[EI_OSABI]);
1222 sv = brand_info->sysvec;
1224 if (hdr->e_type == ET_DYN) {
1225 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1226 uprintf("Cannot execute shared object\n");
1231 * Honour the base load address from the dso if it is
1232 * non-zero for some reason.
1235 if ((sv->sv_flags & SV_ASLR) == 0 ||
1236 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1237 et_dyn_addr = __elfN(pie_base);
1238 else if ((__elfN(pie_aslr_enabled) &&
1239 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1240 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1241 et_dyn_addr = ET_DYN_ADDR_RAND;
1243 et_dyn_addr = __elfN(pie_base);
1248 * Avoid a possible deadlock if the current address space is destroyed
1249 * and that address space maps the locked vnode. In the common case,
1250 * the locked vnode's v_usecount is decremented but remains greater
1251 * than zero. Consequently, the vnode lock is not needed by vrele().
1252 * However, in cases where the vnode lock is external, such as nullfs,
1253 * v_usecount may become zero.
1255 * The VV_TEXT flag prevents modifications to the executable while
1256 * the vnode is unlocked.
1258 VOP_UNLOCK(imgp->vp);
1261 * Decide whether to enable randomization of user mappings.
1262 * First, reset user preferences for the setid binaries.
1263 * Then, account for the support of the randomization by the
1264 * ABI, by user preferences, and make special treatment for
1267 if (imgp->credential_setid) {
1268 PROC_LOCK(imgp->proc);
1269 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1270 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1271 PROC_UNLOCK(imgp->proc);
1273 if ((sv->sv_flags & SV_ASLR) == 0 ||
1274 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1275 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1276 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1277 ("et_dyn_addr == RAND and !ASLR"));
1278 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1279 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1280 et_dyn_addr == ET_DYN_ADDR_RAND) {
1281 imgp->map_flags |= MAP_ASLR;
1283 * If user does not care about sbrk, utilize the bss
1284 * grow region for mappings as well. We can select
1285 * the base for the image anywere and still not suffer
1286 * from the fragmentation.
1288 if (!__elfN(aslr_honor_sbrk) ||
1289 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1290 imgp->map_flags |= MAP_ASLR_IGNSTART;
1291 if (__elfN(aslr_stack))
1292 imgp->map_flags |= MAP_ASLR_STACK;
1295 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1296 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1297 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1298 imgp->map_flags |= MAP_WXORX;
1300 error = exec_new_vmspace(imgp, sv);
1302 imgp->proc->p_sysent = sv;
1304 vmspace = imgp->proc->p_vmspace;
1305 map = &vmspace->vm_map;
1306 maxv = sv->sv_usrstack;
1307 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1308 maxv -= lim_max(td, RLIMIT_STACK);
1309 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1310 uprintf("Excessive mapping size\n");
1314 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1315 KASSERT((map->flags & MAP_ASLR) != 0,
1316 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1317 error = __CONCAT(rnd_, __elfN(base))(map,
1318 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1319 /* reserve half of the address space to interpreter */
1320 maxv / 2, maxalign, &et_dyn_addr);
1323 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1327 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1331 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1336 * We load the dynamic linker where a userland call
1337 * to mmap(0, ...) would put it. The rationale behind this
1338 * calculation is that it leaves room for the heap to grow to
1339 * its maximum allowed size.
1341 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1343 if ((map->flags & MAP_ASLR) != 0) {
1344 maxv1 = maxv / 2 + addr / 2;
1345 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1346 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1347 pagesizes[1] : pagesizes[0], &anon_loc);
1350 map->anon_loc = anon_loc;
1352 map->anon_loc = addr;
1355 entry = (u_long)hdr->e_entry + et_dyn_addr;
1356 imgp->entry_addr = entry;
1358 if (interp != NULL) {
1359 VOP_UNLOCK(imgp->vp);
1360 if ((map->flags & MAP_ASLR) != 0) {
1361 /* Assume that interpreter fits into 1/4 of AS */
1362 maxv1 = maxv / 2 + addr / 2;
1363 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1364 maxv1, PAGE_SIZE, &addr);
1367 error = __elfN(load_interp)(imgp, brand_info, interp,
1368 &addr, &imgp->entry_addr);
1370 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1376 error = exec_map_stack(imgp);
1381 * Construct auxargs table (used by the copyout_auxargs routine)
1383 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1384 if (elf_auxargs == NULL) {
1385 VOP_UNLOCK(imgp->vp);
1386 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1387 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1389 elf_auxargs->execfd = -1;
1390 elf_auxargs->phdr = proghdr + et_dyn_addr;
1391 elf_auxargs->phent = hdr->e_phentsize;
1392 elf_auxargs->phnum = hdr->e_phnum;
1393 elf_auxargs->pagesz = PAGE_SIZE;
1394 elf_auxargs->base = addr;
1395 elf_auxargs->flags = 0;
1396 elf_auxargs->entry = entry;
1397 elf_auxargs->hdr_eflags = hdr->e_flags;
1399 imgp->auxargs = elf_auxargs;
1400 imgp->interpreted = 0;
1401 imgp->reloc_base = addr;
1402 imgp->proc->p_osrel = osrel;
1403 imgp->proc->p_fctl0 = fctl0;
1404 imgp->proc->p_elf_machine = hdr->e_machine;
1405 imgp->proc->p_elf_flags = hdr->e_flags;
1408 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1410 free(interp, M_TEMP);
1414 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1417 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1419 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1420 Elf_Auxinfo *argarray, *pos;
1423 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1426 if (args->execfd != -1)
1427 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1428 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1429 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1430 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1431 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1432 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1433 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1434 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1435 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1436 if (imgp->execpathp != 0)
1437 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1438 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1439 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1440 if (imgp->canary != 0) {
1441 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1442 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1444 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1445 if (imgp->pagesizes != 0) {
1446 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1447 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1449 if (imgp->sysent->sv_timekeep_base != 0) {
1450 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1451 imgp->sysent->sv_timekeep_base);
1453 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1454 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1455 imgp->sysent->sv_stackprot);
1456 if (imgp->sysent->sv_hwcap != NULL)
1457 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1458 if (imgp->sysent->sv_hwcap2 != NULL)
1459 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1460 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1461 ELF_BSDF_SIGFASTBLK : 0);
1462 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1463 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1464 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1465 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1466 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1467 if (imgp->sysent->sv_fxrng_gen_base != 0)
1468 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1469 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1470 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1471 AUXARGS_ENTRY(pos, AT_NULL, 0);
1473 free(imgp->auxargs, M_TEMP);
1474 imgp->auxargs = NULL;
1475 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1477 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1478 free(argarray, M_TEMP);
1483 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1487 base = (Elf_Addr *)*stack_base;
1489 if (elf_suword(base, imgp->args->argc) == -1)
1491 *stack_base = (uintptr_t)base;
1496 * Code for generating ELF core dumps.
1499 typedef void (*segment_callback)(vm_map_entry_t, void *);
1501 /* Closure for cb_put_phdr(). */
1502 struct phdr_closure {
1503 Elf_Phdr *phdr; /* Program header to fill in */
1504 Elf_Off offset; /* Offset of segment in core file */
1507 /* Closure for cb_size_segment(). */
1508 struct sseg_closure {
1509 int count; /* Count of writable segments. */
1510 size_t size; /* Total size of all writable segments. */
1513 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1516 int type; /* Note type. */
1517 outfunc_t outfunc; /* Output function. */
1518 void *outarg; /* Argument for the output function. */
1519 size_t outsize; /* Output size. */
1520 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1523 TAILQ_HEAD(note_info_list, note_info);
1525 extern int compress_user_cores;
1526 extern int compress_user_cores_level;
1528 static void cb_put_phdr(vm_map_entry_t, void *);
1529 static void cb_size_segment(vm_map_entry_t, void *);
1530 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1532 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1533 struct note_info_list *, size_t, int);
1534 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1536 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t, int);
1537 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1538 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1540 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1541 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1542 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1543 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1544 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1545 static void __elfN(note_ptlwpinfo)(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. */
1583 each_dumpable_segment(td, cb_size_segment, &seginfo, flags);
1586 * Collect info about the core file header area.
1588 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1589 if (seginfo.count + 1 >= PN_XNUM)
1590 hdrsize += sizeof(Elf_Shdr);
1591 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1592 coresize = round_page(hdrsize + notesz) + seginfo.size;
1594 /* Set up core dump parameters. */
1596 params.active_cred = cred;
1597 params.file_cred = NOCRED;
1604 PROC_LOCK(td->td_proc);
1605 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1606 PROC_UNLOCK(td->td_proc);
1613 if (coresize >= limit) {
1618 /* Create a compression stream if necessary. */
1619 compm = compress_user_cores;
1620 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1622 compm = COMPRESS_GZIP;
1624 params.comp = compressor_init(core_compressed_write,
1625 compm, CORE_BUF_SIZE,
1626 compress_user_cores_level, ¶ms);
1627 if (params.comp == NULL) {
1631 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1635 * Allocate memory for building the header, fill it up,
1636 * and write it out following the notes.
1638 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1639 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1642 /* Write the contents of all of the writable segments. */
1648 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1649 offset = round_page(hdrsize + notesz);
1650 for (i = 0; i < seginfo.count; i++) {
1651 error = core_output((char *)(uintptr_t)php->p_vaddr,
1652 php->p_filesz, offset, ¶ms, tmpbuf);
1655 offset += php->p_filesz;
1658 if (error == 0 && params.comp != NULL)
1659 error = compressor_flush(params.comp);
1663 "Failed to write core file for process %s (error %d)\n",
1664 curproc->p_comm, error);
1668 free(tmpbuf, M_TEMP);
1669 if (params.comp != NULL)
1670 compressor_fini(params.comp);
1671 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1672 TAILQ_REMOVE(¬elst, ninfo, link);
1673 free(ninfo, M_TEMP);
1682 * A callback for each_dumpable_segment() to write out the segment's
1683 * program header entry.
1686 cb_put_phdr(vm_map_entry_t entry, void *closure)
1688 struct phdr_closure *phc = (struct phdr_closure *)closure;
1689 Elf_Phdr *phdr = phc->phdr;
1691 phc->offset = round_page(phc->offset);
1693 phdr->p_type = PT_LOAD;
1694 phdr->p_offset = phc->offset;
1695 phdr->p_vaddr = entry->start;
1697 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1698 phdr->p_align = PAGE_SIZE;
1699 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1701 phc->offset += phdr->p_filesz;
1706 * A callback for each_dumpable_segment() to gather information about
1707 * the number of segments and their total size.
1710 cb_size_segment(vm_map_entry_t entry, void *closure)
1712 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1715 ssc->size += entry->end - entry->start;
1719 * For each writable segment in the process's memory map, call the given
1720 * function with a pointer to the map entry and some arbitrary
1721 * caller-supplied data.
1724 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1727 struct proc *p = td->td_proc;
1728 vm_map_t map = &p->p_vmspace->vm_map;
1729 vm_map_entry_t entry;
1730 vm_object_t backing_object, object;
1733 vm_map_lock_read(map);
1734 VM_MAP_ENTRY_FOREACH(entry, map) {
1736 * Don't dump inaccessible mappings, deal with legacy
1739 * Note that read-only segments related to the elf binary
1740 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1741 * need to arbitrarily ignore such segments.
1743 if ((flags & SVC_ALL) == 0) {
1744 if (elf_legacy_coredump) {
1745 if ((entry->protection & VM_PROT_RW) !=
1749 if ((entry->protection & VM_PROT_ALL) == 0)
1755 * Dont include memory segment in the coredump if
1756 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1757 * madvise(2). Do not dump submaps (i.e. parts of the
1760 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1762 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1763 (flags & SVC_ALL) == 0)
1765 if ((object = entry->object.vm_object) == NULL)
1768 /* Ignore memory-mapped devices and such things. */
1769 VM_OBJECT_RLOCK(object);
1770 while ((backing_object = object->backing_object) != NULL) {
1771 VM_OBJECT_RLOCK(backing_object);
1772 VM_OBJECT_RUNLOCK(object);
1773 object = backing_object;
1775 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1776 VM_OBJECT_RUNLOCK(object);
1780 (*func)(entry, closure);
1782 vm_map_unlock_read(map);
1786 * Write the core file header to the file, including padding up to
1787 * the page boundary.
1790 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1791 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1794 struct note_info *ninfo;
1798 /* Fill in the header. */
1799 bzero(hdr, hdrsize);
1800 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1802 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1803 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1804 sbuf_start_section(sb, NULL);
1805 sbuf_bcat(sb, hdr, hdrsize);
1806 TAILQ_FOREACH(ninfo, notelst, link)
1807 __elfN(putnote)(ninfo, sb);
1808 /* Align up to a page boundary for the program segments. */
1809 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1810 error = sbuf_finish(sb);
1817 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1827 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1830 * To have the debugger select the right thread (LWP) as the initial
1831 * thread, we dump the state of the thread passed to us in td first.
1832 * This is the thread that causes the core dump and thus likely to
1833 * be the right thread one wants to have selected in the debugger.
1836 while (thr != NULL) {
1837 size += register_note(list, NT_PRSTATUS,
1838 __elfN(note_prstatus), thr);
1839 size += register_note(list, NT_FPREGSET,
1840 __elfN(note_fpregset), thr);
1841 size += register_note(list, NT_THRMISC,
1842 __elfN(note_thrmisc), thr);
1843 size += register_note(list, NT_PTLWPINFO,
1844 __elfN(note_ptlwpinfo), thr);
1845 size += register_note(list, -1,
1846 __elfN(note_threadmd), thr);
1848 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1849 TAILQ_NEXT(thr, td_plist);
1851 thr = TAILQ_NEXT(thr, td_plist);
1854 size += register_note(list, NT_PROCSTAT_PROC,
1855 __elfN(note_procstat_proc), p);
1856 size += register_note(list, NT_PROCSTAT_FILES,
1857 note_procstat_files, p);
1858 size += register_note(list, NT_PROCSTAT_VMMAP,
1859 note_procstat_vmmap, p);
1860 size += register_note(list, NT_PROCSTAT_GROUPS,
1861 note_procstat_groups, p);
1862 size += register_note(list, NT_PROCSTAT_UMASK,
1863 note_procstat_umask, p);
1864 size += register_note(list, NT_PROCSTAT_RLIMIT,
1865 note_procstat_rlimit, p);
1866 size += register_note(list, NT_PROCSTAT_OSREL,
1867 note_procstat_osrel, p);
1868 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1869 __elfN(note_procstat_psstrings), p);
1870 size += register_note(list, NT_PROCSTAT_AUXV,
1871 __elfN(note_procstat_auxv), p);
1877 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1878 size_t notesz, int flags)
1883 struct phdr_closure phc;
1885 ehdr = (Elf_Ehdr *)hdr;
1887 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1888 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1889 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1890 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1891 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1892 ehdr->e_ident[EI_DATA] = ELF_DATA;
1893 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1894 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1895 ehdr->e_ident[EI_ABIVERSION] = 0;
1896 ehdr->e_ident[EI_PAD] = 0;
1897 ehdr->e_type = ET_CORE;
1898 ehdr->e_machine = td->td_proc->p_elf_machine;
1899 ehdr->e_version = EV_CURRENT;
1901 ehdr->e_phoff = sizeof(Elf_Ehdr);
1902 ehdr->e_flags = td->td_proc->p_elf_flags;
1903 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1904 ehdr->e_phentsize = sizeof(Elf_Phdr);
1905 ehdr->e_shentsize = sizeof(Elf_Shdr);
1906 ehdr->e_shstrndx = SHN_UNDEF;
1907 if (numsegs + 1 < PN_XNUM) {
1908 ehdr->e_phnum = numsegs + 1;
1911 ehdr->e_phnum = PN_XNUM;
1914 ehdr->e_shoff = ehdr->e_phoff +
1915 (numsegs + 1) * ehdr->e_phentsize;
1916 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1917 ("e_shoff: %zu, hdrsize - shdr: %zu",
1918 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1920 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1921 memset(shdr, 0, sizeof(*shdr));
1923 * A special first section is used to hold large segment and
1924 * section counts. This was proposed by Sun Microsystems in
1925 * Solaris and has been adopted by Linux; the standard ELF
1926 * tools are already familiar with the technique.
1928 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1929 * (or 12-7 depending on the version of the document) for more
1932 shdr->sh_type = SHT_NULL;
1933 shdr->sh_size = ehdr->e_shnum;
1934 shdr->sh_link = ehdr->e_shstrndx;
1935 shdr->sh_info = numsegs + 1;
1939 * Fill in the program header entries.
1941 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1943 /* The note segement. */
1944 phdr->p_type = PT_NOTE;
1945 phdr->p_offset = hdrsize;
1948 phdr->p_filesz = notesz;
1950 phdr->p_flags = PF_R;
1951 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1954 /* All the writable segments from the program. */
1956 phc.offset = round_page(hdrsize + notesz);
1957 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1961 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1963 struct note_info *ninfo;
1964 size_t size, notesize;
1967 out(arg, NULL, &size);
1968 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1970 ninfo->outfunc = out;
1971 ninfo->outarg = arg;
1972 ninfo->outsize = size;
1973 TAILQ_INSERT_TAIL(list, ninfo, link);
1978 notesize = sizeof(Elf_Note) + /* note header */
1979 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1981 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1987 append_note_data(const void *src, void *dst, size_t len)
1991 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1993 bcopy(src, dst, len);
1994 bzero((char *)dst + len, padded_len - len);
1996 return (padded_len);
2000 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2008 note = (Elf_Note *)buf;
2009 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2010 note->n_descsz = size;
2011 note->n_type = type;
2012 buf += sizeof(*note);
2013 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2014 sizeof(FREEBSD_ABI_VENDOR));
2015 append_note_data(src, buf, size);
2020 notesize = sizeof(Elf_Note) + /* note header */
2021 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2023 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2029 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2032 ssize_t old_len, sect_len;
2033 size_t new_len, descsz, i;
2035 if (ninfo->type == -1) {
2036 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2040 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2041 note.n_descsz = ninfo->outsize;
2042 note.n_type = ninfo->type;
2044 sbuf_bcat(sb, ¬e, sizeof(note));
2045 sbuf_start_section(sb, &old_len);
2046 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2047 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2048 if (note.n_descsz == 0)
2050 sbuf_start_section(sb, &old_len);
2051 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2052 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2056 new_len = (size_t)sect_len;
2057 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2058 if (new_len < descsz) {
2060 * It is expected that individual note emitters will correctly
2061 * predict their expected output size and fill up to that size
2062 * themselves, padding in a format-specific way if needed.
2063 * However, in case they don't, just do it here with zeros.
2065 for (i = 0; i < descsz - new_len; i++)
2067 } else if (new_len > descsz) {
2069 * We can't always truncate sb -- we may have drained some
2072 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2073 "read it (%zu > %zu). Since it is longer than "
2074 "expected, this coredump's notes are corrupt. THIS "
2075 "IS A BUG in the note_procstat routine for type %u.\n",
2076 __func__, (unsigned)note.n_type, new_len, descsz,
2077 (unsigned)note.n_type));
2082 * Miscellaneous note out functions.
2085 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2086 #include <compat/freebsd32/freebsd32.h>
2087 #include <compat/freebsd32/freebsd32_signal.h>
2089 typedef struct prstatus32 elf_prstatus_t;
2090 typedef struct prpsinfo32 elf_prpsinfo_t;
2091 typedef struct fpreg32 elf_prfpregset_t;
2092 typedef struct fpreg32 elf_fpregset_t;
2093 typedef struct reg32 elf_gregset_t;
2094 typedef struct thrmisc32 elf_thrmisc_t;
2095 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2096 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2097 typedef uint32_t elf_ps_strings_t;
2099 typedef prstatus_t elf_prstatus_t;
2100 typedef prpsinfo_t elf_prpsinfo_t;
2101 typedef prfpregset_t elf_prfpregset_t;
2102 typedef prfpregset_t elf_fpregset_t;
2103 typedef gregset_t elf_gregset_t;
2104 typedef thrmisc_t elf_thrmisc_t;
2105 #define ELF_KERN_PROC_MASK 0
2106 typedef struct kinfo_proc elf_kinfo_proc_t;
2107 typedef vm_offset_t elf_ps_strings_t;
2111 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2117 elf_prpsinfo_t *psinfo;
2120 p = (struct proc *)arg;
2122 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2123 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2124 psinfo->pr_version = PRPSINFO_VERSION;
2125 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2126 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2128 if (p->p_args != NULL) {
2129 len = sizeof(psinfo->pr_psargs) - 1;
2130 if (len > p->p_args->ar_length)
2131 len = p->p_args->ar_length;
2132 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2138 sbuf_new(&sbarg, psinfo->pr_psargs,
2139 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2140 error = proc_getargv(curthread, p, &sbarg);
2142 if (sbuf_finish(&sbarg) == 0)
2143 len = sbuf_len(&sbarg) - 1;
2145 len = sizeof(psinfo->pr_psargs) - 1;
2146 sbuf_delete(&sbarg);
2148 if (error || len == 0)
2149 strlcpy(psinfo->pr_psargs, p->p_comm,
2150 sizeof(psinfo->pr_psargs));
2152 KASSERT(len < sizeof(psinfo->pr_psargs),
2153 ("len is too long: %zu vs %zu", len,
2154 sizeof(psinfo->pr_psargs)));
2155 cp = psinfo->pr_psargs;
2158 cp = memchr(cp, '\0', end - cp);
2164 psinfo->pr_pid = p->p_pid;
2165 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2166 free(psinfo, M_TEMP);
2168 *sizep = sizeof(*psinfo);
2172 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2175 elf_prstatus_t *status;
2177 td = (struct thread *)arg;
2179 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2180 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2181 status->pr_version = PRSTATUS_VERSION;
2182 status->pr_statussz = sizeof(elf_prstatus_t);
2183 status->pr_gregsetsz = sizeof(elf_gregset_t);
2184 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2185 status->pr_osreldate = osreldate;
2186 status->pr_cursig = td->td_proc->p_sig;
2187 status->pr_pid = td->td_tid;
2188 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2189 fill_regs32(td, &status->pr_reg);
2191 fill_regs(td, &status->pr_reg);
2193 sbuf_bcat(sb, status, sizeof(*status));
2194 free(status, M_TEMP);
2196 *sizep = sizeof(*status);
2200 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2203 elf_prfpregset_t *fpregset;
2205 td = (struct thread *)arg;
2207 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2208 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2209 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2210 fill_fpregs32(td, fpregset);
2212 fill_fpregs(td, fpregset);
2214 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2215 free(fpregset, M_TEMP);
2217 *sizep = sizeof(*fpregset);
2221 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2224 elf_thrmisc_t thrmisc;
2226 td = (struct thread *)arg;
2228 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2229 bzero(&thrmisc, sizeof(thrmisc));
2230 strcpy(thrmisc.pr_tname, td->td_name);
2231 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2233 *sizep = sizeof(thrmisc);
2237 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2242 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2243 struct ptrace_lwpinfo32 pl;
2245 struct ptrace_lwpinfo pl;
2248 td = (struct thread *)arg;
2249 size = sizeof(structsize) + sizeof(pl);
2251 KASSERT(*sizep == size, ("invalid size"));
2252 structsize = sizeof(pl);
2253 sbuf_bcat(sb, &structsize, sizeof(structsize));
2254 bzero(&pl, sizeof(pl));
2255 pl.pl_lwpid = td->td_tid;
2256 pl.pl_event = PL_EVENT_NONE;
2257 pl.pl_sigmask = td->td_sigmask;
2258 pl.pl_siglist = td->td_siglist;
2259 if (td->td_si.si_signo != 0) {
2260 pl.pl_event = PL_EVENT_SIGNAL;
2261 pl.pl_flags |= PL_FLAG_SI;
2262 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2263 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2265 pl.pl_siginfo = td->td_si;
2268 strcpy(pl.pl_tdname, td->td_name);
2269 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2270 sbuf_bcat(sb, &pl, sizeof(pl));
2276 * Allow for MD specific notes, as well as any MD
2277 * specific preparations for writing MI notes.
2280 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2286 td = (struct thread *)arg;
2288 if (size != 0 && sb != NULL)
2289 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2293 __elfN(dump_thread)(td, buf, &size);
2294 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2295 if (size != 0 && sb != NULL)
2296 sbuf_bcat(sb, buf, size);
2301 #ifdef KINFO_PROC_SIZE
2302 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2306 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2312 p = (struct proc *)arg;
2313 size = sizeof(structsize) + p->p_numthreads *
2314 sizeof(elf_kinfo_proc_t);
2317 KASSERT(*sizep == size, ("invalid size"));
2318 structsize = sizeof(elf_kinfo_proc_t);
2319 sbuf_bcat(sb, &structsize, sizeof(structsize));
2320 sx_slock(&proctree_lock);
2322 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2323 sx_sunlock(&proctree_lock);
2328 #ifdef KINFO_FILE_SIZE
2329 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2333 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2336 size_t size, sect_sz, i;
2337 ssize_t start_len, sect_len;
2338 int structsize, filedesc_flags;
2340 if (coredump_pack_fileinfo)
2341 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2345 p = (struct proc *)arg;
2346 structsize = sizeof(struct kinfo_file);
2349 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2350 sbuf_set_drain(sb, sbuf_count_drain, &size);
2351 sbuf_bcat(sb, &structsize, sizeof(structsize));
2353 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2358 sbuf_start_section(sb, &start_len);
2360 sbuf_bcat(sb, &structsize, sizeof(structsize));
2362 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2365 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2370 KASSERT(sect_sz <= *sizep,
2371 ("kern_proc_filedesc_out did not respect maxlen; "
2372 "requested %zu, got %zu", *sizep - sizeof(structsize),
2373 sect_sz - sizeof(structsize)));
2375 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2380 #ifdef KINFO_VMENTRY_SIZE
2381 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2385 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2389 int structsize, vmmap_flags;
2391 if (coredump_pack_vmmapinfo)
2392 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2396 p = (struct proc *)arg;
2397 structsize = sizeof(struct kinfo_vmentry);
2400 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2401 sbuf_set_drain(sb, sbuf_count_drain, &size);
2402 sbuf_bcat(sb, &structsize, sizeof(structsize));
2404 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2409 sbuf_bcat(sb, &structsize, sizeof(structsize));
2411 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2417 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2423 p = (struct proc *)arg;
2424 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2426 KASSERT(*sizep == size, ("invalid size"));
2427 structsize = sizeof(gid_t);
2428 sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2436 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2442 p = (struct proc *)arg;
2443 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2445 KASSERT(*sizep == size, ("invalid size"));
2446 structsize = sizeof(p->p_pd->pd_cmask);
2447 sbuf_bcat(sb, &structsize, sizeof(structsize));
2448 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2454 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2457 struct rlimit rlim[RLIM_NLIMITS];
2461 p = (struct proc *)arg;
2462 size = sizeof(structsize) + sizeof(rlim);
2464 KASSERT(*sizep == size, ("invalid size"));
2465 structsize = sizeof(rlim);
2466 sbuf_bcat(sb, &structsize, sizeof(structsize));
2468 for (i = 0; i < RLIM_NLIMITS; i++)
2469 lim_rlimit_proc(p, i, &rlim[i]);
2471 sbuf_bcat(sb, rlim, sizeof(rlim));
2477 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2483 p = (struct proc *)arg;
2484 size = sizeof(structsize) + sizeof(p->p_osrel);
2486 KASSERT(*sizep == size, ("invalid size"));
2487 structsize = sizeof(p->p_osrel);
2488 sbuf_bcat(sb, &structsize, sizeof(structsize));
2489 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2495 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2498 elf_ps_strings_t ps_strings;
2502 p = (struct proc *)arg;
2503 size = sizeof(structsize) + sizeof(ps_strings);
2505 KASSERT(*sizep == size, ("invalid size"));
2506 structsize = sizeof(ps_strings);
2507 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2508 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2510 ps_strings = PROC_PS_STRINGS(p);
2512 sbuf_bcat(sb, &structsize, sizeof(structsize));
2513 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2519 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2525 p = (struct proc *)arg;
2528 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2529 sbuf_set_drain(sb, sbuf_count_drain, &size);
2530 sbuf_bcat(sb, &structsize, sizeof(structsize));
2532 proc_getauxv(curthread, p, sb);
2538 structsize = sizeof(Elf_Auxinfo);
2539 sbuf_bcat(sb, &structsize, sizeof(structsize));
2541 proc_getauxv(curthread, p, sb);
2547 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2548 const char *note_vendor, const Elf_Phdr *pnote,
2549 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2551 const Elf_Note *note, *note0, *note_end;
2552 const char *note_name;
2557 /* We need some limit, might as well use PAGE_SIZE. */
2558 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2560 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2561 if (pnote->p_offset > PAGE_SIZE ||
2562 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2563 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2565 VOP_UNLOCK(imgp->vp);
2566 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2567 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2569 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2570 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2571 curthread->td_ucred, NOCRED, NULL, curthread);
2573 uprintf("i/o error PT_NOTE\n");
2576 note = note0 = (const Elf_Note *)buf;
2577 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2579 note = note0 = (const Elf_Note *)(imgp->image_header +
2581 note_end = (const Elf_Note *)(imgp->image_header +
2582 pnote->p_offset + pnote->p_filesz);
2585 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2586 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2587 (const char *)note < sizeof(Elf_Note)) {
2590 if (note->n_namesz != checknote->n_namesz ||
2591 note->n_descsz != checknote->n_descsz ||
2592 note->n_type != checknote->n_type)
2594 note_name = (const char *)(note + 1);
2595 if (note_name + checknote->n_namesz >=
2596 (const char *)note_end || strncmp(note_vendor,
2597 note_name, checknote->n_namesz) != 0)
2600 if (cb(note, cb_arg, &res))
2603 note = (const Elf_Note *)((const char *)(note + 1) +
2604 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2605 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2614 struct brandnote_cb_arg {
2615 Elf_Brandnote *brandnote;
2620 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2622 struct brandnote_cb_arg *arg;
2627 * Fetch the osreldate for binary from the ELF OSABI-note if
2630 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2631 arg->brandnote->trans_osrel != NULL ?
2632 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2637 static Elf_Note fctl_note = {
2638 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2639 .n_descsz = sizeof(uint32_t),
2640 .n_type = NT_FREEBSD_FEATURE_CTL,
2643 struct fctl_cb_arg {
2644 boolean_t *has_fctl0;
2649 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2651 struct fctl_cb_arg *arg;
2652 const Elf32_Word *desc;
2656 p = (uintptr_t)(note + 1);
2657 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2658 desc = (const Elf32_Word *)p;
2659 *arg->has_fctl0 = TRUE;
2660 *arg->fctl0 = desc[0];
2666 * Try to find the appropriate ABI-note section for checknote, fetch
2667 * the osreldate and feature control flags for binary from the ELF
2668 * OSABI-note. Only the first page of the image is searched, the same
2672 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2673 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2675 const Elf_Phdr *phdr;
2676 const Elf_Ehdr *hdr;
2677 struct brandnote_cb_arg b_arg;
2678 struct fctl_cb_arg f_arg;
2681 hdr = (const Elf_Ehdr *)imgp->image_header;
2682 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2683 b_arg.brandnote = brandnote;
2684 b_arg.osrel = osrel;
2685 f_arg.has_fctl0 = has_fctl0;
2686 f_arg.fctl0 = fctl0;
2688 for (i = 0; i < hdr->e_phnum; i++) {
2689 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2690 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2692 for (j = 0; j < hdr->e_phnum; j++) {
2693 if (phdr[j].p_type == PT_NOTE &&
2694 __elfN(parse_notes)(imgp, &fctl_note,
2695 FREEBSD_ABI_VENDOR, &phdr[j],
2696 note_fctl_cb, &f_arg))
2707 * Tell kern_execve.c about it, with a little help from the linker.
2709 static struct execsw __elfN(execsw) = {
2710 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2711 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2713 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2716 __elfN(trans_prot)(Elf_Word flags)
2722 prot |= VM_PROT_EXECUTE;
2724 prot |= VM_PROT_WRITE;
2726 prot |= VM_PROT_READ;
2727 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2728 if (i386_read_exec && (flags & PF_R))
2729 prot |= VM_PROT_EXECUTE;
2735 __elfN(untrans_prot)(vm_prot_t prot)
2740 if (prot & VM_PROT_EXECUTE)
2742 if (prot & VM_PROT_READ)
2744 if (prot & VM_PROT_WRITE)