2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static bool __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, bool *has_fctl0,
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
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 * While for 64-bit machines ASLR works properly, there are
175 * still some problems when using 32-bit architectures. For this
176 * reason ASLR is only enabled by default when running native
177 * 64-bit non-PIE executables.
179 static int __elfN(aslr_enabled) = __ELF_WORD_SIZE == 64;
180 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
181 &__elfN(aslr_enabled), 0,
182 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
183 ": enable address map randomization");
186 * Enable ASLR only for 64-bit PIE binaries by default.
188 static int __elfN(pie_aslr_enabled) = __ELF_WORD_SIZE == 64;
189 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
190 &__elfN(pie_aslr_enabled), 0,
191 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
192 ": enable address map randomization for PIE binaries");
195 * Sbrk is now deprecated and it can be assumed, that in most
196 * cases it will not be used anyway. This setting is valid only
197 * for the ASLR enabled and allows for utilizing the bss grow region.
199 static int __elfN(aslr_honor_sbrk) = 0;
200 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
201 &__elfN(aslr_honor_sbrk), 0,
202 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
204 static int __elfN(aslr_stack_gap) = 3;
205 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
206 &__elfN(aslr_stack_gap), 0,
207 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
208 ": maximum percentage of main stack to waste on a random gap");
210 static int __elfN(sigfastblock) = 1;
211 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
212 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
213 "enable sigfastblock for new processes");
215 static bool __elfN(allow_wx) = true;
216 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
217 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
218 "Allow pages to be mapped simultaneously writable and executable");
220 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
222 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
224 Elf_Brandnote __elfN(freebsd_brandnote) = {
225 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
226 .hdr.n_descsz = sizeof(int32_t),
227 .hdr.n_type = NT_FREEBSD_ABI_TAG,
228 .vendor = FREEBSD_ABI_VENDOR,
229 .flags = BN_TRANSLATE_OSREL,
230 .trans_osrel = __elfN(freebsd_trans_osrel)
234 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
238 p = (uintptr_t)(note + 1);
239 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
240 *osrel = *(const int32_t *)(p);
245 static const char GNU_ABI_VENDOR[] = "GNU";
246 static int GNU_KFREEBSD_ABI_DESC = 3;
248 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
249 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
250 .hdr.n_descsz = 16, /* XXX at least 16 */
252 .vendor = GNU_ABI_VENDOR,
253 .flags = BN_TRANSLATE_OSREL,
254 .trans_osrel = kfreebsd_trans_osrel
258 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
260 const Elf32_Word *desc;
263 p = (uintptr_t)(note + 1);
264 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
266 desc = (const Elf32_Word *)p;
267 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
271 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
272 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
274 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
280 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
284 for (i = 0; i < MAX_BRANDS; i++) {
285 if (elf_brand_list[i] == NULL) {
286 elf_brand_list[i] = entry;
290 if (i == MAX_BRANDS) {
291 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
299 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
303 for (i = 0; i < MAX_BRANDS; i++) {
304 if (elf_brand_list[i] == entry) {
305 elf_brand_list[i] = NULL;
315 __elfN(brand_inuse)(Elf_Brandinfo *entry)
320 sx_slock(&allproc_lock);
321 FOREACH_PROC_IN_SYSTEM(p) {
322 if (p->p_sysent == entry->sysvec) {
327 sx_sunlock(&allproc_lock);
332 static Elf_Brandinfo *
333 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
334 int32_t *osrel, uint32_t *fctl0)
336 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
337 Elf_Brandinfo *bi, *bi_m;
339 int i, interp_name_len;
341 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
344 * We support four types of branding -- (1) the ELF EI_OSABI field
345 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
346 * branding w/in the ELF header, (3) path of the `interp_path'
347 * field, and (4) the ".note.ABI-tag" ELF section.
350 /* Look for an ".note.ABI-tag" ELF section */
352 for (i = 0; i < MAX_BRANDS; i++) {
353 bi = elf_brand_list[i];
356 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
358 if (hdr->e_machine == bi->machine && (bi->flags &
359 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
363 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
365 /* Give brand a chance to veto check_note's guess */
366 if (ret && bi->header_supported) {
367 ret = bi->header_supported(imgp, osrel,
368 has_fctl0 ? fctl0 : NULL);
371 * If note checker claimed the binary, but the
372 * interpreter path in the image does not
373 * match default one for the brand, try to
374 * search for other brands with the same
375 * interpreter. Either there is better brand
376 * with the right interpreter, or, failing
377 * this, we return first brand which accepted
378 * our note and, optionally, header.
380 if (ret && bi_m == NULL && interp != NULL &&
381 (bi->interp_path == NULL ||
382 (strlen(bi->interp_path) + 1 != interp_name_len ||
383 strncmp(interp, bi->interp_path, interp_name_len)
395 /* If the executable has a brand, search for it in the brand list. */
396 for (i = 0; i < MAX_BRANDS; i++) {
397 bi = elf_brand_list[i];
398 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
399 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
401 if (hdr->e_machine == bi->machine &&
402 (hdr->e_ident[EI_OSABI] == bi->brand ||
403 (bi->compat_3_brand != NULL &&
404 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
405 bi->compat_3_brand) == 0))) {
406 /* Looks good, but give brand a chance to veto */
407 if (bi->header_supported == NULL ||
408 bi->header_supported(imgp, NULL, NULL)) {
410 * Again, prefer strictly matching
413 if (interp_name_len == 0 &&
414 bi->interp_path == NULL)
416 if (bi->interp_path != NULL &&
417 strlen(bi->interp_path) + 1 ==
418 interp_name_len && strncmp(interp,
419 bi->interp_path, interp_name_len) == 0)
429 /* No known brand, see if the header is recognized by any brand */
430 for (i = 0; i < MAX_BRANDS; i++) {
431 bi = elf_brand_list[i];
432 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
433 bi->header_supported == NULL)
435 if (hdr->e_machine == bi->machine) {
436 ret = bi->header_supported(imgp, NULL, NULL);
442 /* Lacking a known brand, search for a recognized interpreter. */
443 if (interp != NULL) {
444 for (i = 0; i < MAX_BRANDS; i++) {
445 bi = elf_brand_list[i];
446 if (bi == NULL || (bi->flags &
447 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
450 if (hdr->e_machine == bi->machine &&
451 bi->interp_path != NULL &&
452 /* ELF image p_filesz includes terminating zero */
453 strlen(bi->interp_path) + 1 == interp_name_len &&
454 strncmp(interp, bi->interp_path, interp_name_len)
455 == 0 && (bi->header_supported == NULL ||
456 bi->header_supported(imgp, NULL, NULL)))
461 /* Lacking a recognized interpreter, try the default brand */
462 for (i = 0; i < MAX_BRANDS; i++) {
463 bi = elf_brand_list[i];
464 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
465 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
467 if (hdr->e_machine == bi->machine &&
468 __elfN(fallback_brand) == bi->brand &&
469 (bi->header_supported == NULL ||
470 bi->header_supported(imgp, NULL, NULL)))
477 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
479 return (hdr->e_phoff <= PAGE_SIZE &&
480 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
484 __elfN(check_header)(const Elf_Ehdr *hdr)
490 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
491 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
492 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
493 hdr->e_phentsize != sizeof(Elf_Phdr) ||
494 hdr->e_version != ELF_TARG_VER)
498 * Make sure we have at least one brand for this machine.
501 for (i = 0; i < MAX_BRANDS; i++) {
502 bi = elf_brand_list[i];
503 if (bi != NULL && bi->machine == hdr->e_machine)
513 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
514 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
521 * Create the page if it doesn't exist yet. Ignore errors.
523 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
524 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
527 * Find the page from the underlying object.
529 if (object != NULL) {
530 sf = vm_imgact_map_page(object, offset);
532 return (KERN_FAILURE);
533 off = offset - trunc_page(offset);
534 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
536 vm_imgact_unmap_page(sf);
538 return (KERN_FAILURE);
541 return (KERN_SUCCESS);
545 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
546 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
552 int error, locked, rv;
554 if (start != trunc_page(start)) {
555 rv = __elfN(map_partial)(map, object, offset, start,
556 round_page(start), prot);
557 if (rv != KERN_SUCCESS)
559 offset += round_page(start) - start;
560 start = round_page(start);
562 if (end != round_page(end)) {
563 rv = __elfN(map_partial)(map, object, offset +
564 trunc_page(end) - start, trunc_page(end), end, prot);
565 if (rv != KERN_SUCCESS)
567 end = trunc_page(end);
570 return (KERN_SUCCESS);
571 if ((offset & PAGE_MASK) != 0) {
573 * The mapping is not page aligned. This means that we have
576 rv = vm_map_fixed(map, NULL, 0, start, end - start,
577 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
578 if (rv != KERN_SUCCESS)
581 return (KERN_SUCCESS);
582 for (; start < end; start += sz) {
583 sf = vm_imgact_map_page(object, offset);
585 return (KERN_FAILURE);
586 off = offset - trunc_page(offset);
588 if (sz > PAGE_SIZE - off)
589 sz = PAGE_SIZE - off;
590 error = copyout((caddr_t)sf_buf_kva(sf) + off,
592 vm_imgact_unmap_page(sf);
594 return (KERN_FAILURE);
598 vm_object_reference(object);
599 rv = vm_map_fixed(map, object, offset, start, end - start,
600 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
601 (object != NULL ? MAP_VN_EXEC : 0));
602 if (rv != KERN_SUCCESS) {
603 locked = VOP_ISLOCKED(imgp->vp);
604 VOP_UNLOCK(imgp->vp);
605 vm_object_deallocate(object);
606 vn_lock(imgp->vp, locked | LK_RETRY);
608 } else if (object != NULL) {
609 MPASS(imgp->vp->v_object == object);
610 VOP_SET_TEXT_CHECKED(imgp->vp);
613 return (KERN_SUCCESS);
617 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
618 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
624 vm_offset_t map_addr;
627 vm_ooffset_t file_addr;
630 * It's necessary to fail if the filsz + offset taken from the
631 * header is greater than the actual file pager object's size.
632 * If we were to allow this, then the vm_map_find() below would
633 * walk right off the end of the file object and into the ether.
635 * While I'm here, might as well check for something else that
636 * is invalid: filsz cannot be greater than memsz.
638 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
640 uprintf("elf_load_section: truncated ELF file\n");
644 object = imgp->object;
645 map = &imgp->proc->p_vmspace->vm_map;
646 map_addr = trunc_page((vm_offset_t)vmaddr);
647 file_addr = trunc_page(offset);
650 * We have two choices. We can either clear the data in the last page
651 * of an oversized mapping, or we can start the anon mapping a page
652 * early and copy the initialized data into that first page. We
657 else if (memsz > filsz)
658 map_len = trunc_page(offset + filsz) - file_addr;
660 map_len = round_page(offset + filsz) - file_addr;
663 /* cow flags: don't dump readonly sections in core */
664 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
665 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
667 rv = __elfN(map_insert)(imgp, map, object, file_addr,
668 map_addr, map_addr + map_len, prot, cow);
669 if (rv != KERN_SUCCESS)
672 /* we can stop now if we've covered it all */
678 * We have to get the remaining bit of the file into the first part
679 * of the oversized map segment. This is normally because the .data
680 * segment in the file is extended to provide bss. It's a neat idea
681 * to try and save a page, but it's a pain in the behind to implement.
683 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
685 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
686 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
688 /* This had damn well better be true! */
690 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
691 map_addr + map_len, prot, 0);
692 if (rv != KERN_SUCCESS)
697 sf = vm_imgact_map_page(object, offset + filsz);
701 /* send the page fragment to user space */
702 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
704 vm_imgact_unmap_page(sf);
710 * Remove write access to the page if it was only granted by map_insert
713 if ((prot & VM_PROT_WRITE) == 0)
714 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
715 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
721 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
722 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
729 ASSERT_VOP_LOCKED(imgp->vp, __func__);
734 for (i = 0; i < hdr->e_phnum; i++) {
735 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
738 /* Loadable segment */
739 prot = __elfN(trans_prot)(phdr[i].p_flags);
740 error = __elfN(load_section)(imgp, phdr[i].p_offset,
741 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
742 phdr[i].p_memsz, phdr[i].p_filesz, prot);
747 * Establish the base address if this is the first segment.
750 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
755 if (base_addrp != NULL)
756 *base_addrp = base_addr;
762 * Load the file "file" into memory. It may be either a shared object
765 * The "addr" reference parameter is in/out. On entry, it specifies
766 * the address where a shared object should be loaded. If the file is
767 * an executable, this value is ignored. On exit, "addr" specifies
768 * where the file was actually loaded.
770 * The "entry" reference parameter is out only. On exit, it specifies
771 * the entry point for the loaded file.
774 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
780 struct image_params image_params;
782 const Elf_Ehdr *hdr = NULL;
783 const Elf_Phdr *phdr = NULL;
784 struct nameidata *nd;
786 struct image_params *imgp;
788 u_long base_addr = 0;
791 #ifdef CAPABILITY_MODE
793 * XXXJA: This check can go away once we are sufficiently confident
794 * that the checks in namei() are correct.
796 if (IN_CAPABILITY_MODE(curthread))
800 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
802 attr = &tempdata->attr;
803 imgp = &tempdata->image_params;
806 * Initialize part of the common data
811 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
813 if ((error = namei(nd)) != 0) {
817 NDFREE(nd, NDF_ONLY_PNBUF);
818 imgp->vp = nd->ni_vp;
821 * Check permissions, modes, uid, etc on the file, and "open" it.
823 error = exec_check_permissions(imgp);
827 error = exec_map_first_page(imgp);
831 imgp->object = nd->ni_vp->v_object;
833 hdr = (const Elf_Ehdr *)imgp->image_header;
834 if ((error = __elfN(check_header)(hdr)) != 0)
836 if (hdr->e_type == ET_DYN)
838 else if (hdr->e_type == ET_EXEC)
845 /* Only support headers that fit within first page for now */
846 if (!__elfN(phdr_in_zero_page)(hdr)) {
851 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
852 if (!aligned(phdr, Elf_Addr)) {
857 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
862 *entry = (unsigned long)hdr->e_entry + rbase;
866 exec_unmap_first_page(imgp);
870 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
873 free(tempdata, M_TEMP);
879 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
880 u_int align, u_long *resp)
884 MPASS(vm_map_min(map) <= minv);
886 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
887 uprintf("Invalid ELF segments layout\n");
891 arc4rand(&rbase, sizeof(rbase), 0);
892 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
893 res &= ~((u_long)align - 1);
898 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
899 res, minv, maxv, rbase));
901 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
902 res, maxv, minv, rbase));
909 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
910 const Elf_Phdr *phdr, u_long et_dyn_addr)
912 struct vmspace *vmspace;
914 u_long text_size, data_size, total_size, text_addr, data_addr;
915 u_long seg_size, seg_addr;
919 text_size = data_size = total_size = text_addr = data_addr = 0;
921 for (i = 0; i < hdr->e_phnum; i++) {
922 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
925 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
926 seg_size = round_page(phdr[i].p_memsz +
927 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
930 * Make the largest executable segment the official
931 * text segment and all others data.
933 * Note that obreak() assumes that data_addr + data_size == end
934 * of data load area, and the ELF file format expects segments
935 * to be sorted by address. If multiple data segments exist,
936 * the last one will be used.
939 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
940 text_size = seg_size;
941 text_addr = seg_addr;
943 data_size = seg_size;
944 data_addr = seg_addr;
946 total_size += seg_size;
949 if (data_addr == 0 && data_size == 0) {
950 data_addr = text_addr;
951 data_size = text_size;
955 * Check limits. It should be safe to check the
956 * limits after loading the segments since we do
957 * not actually fault in all the segments pages.
959 PROC_LOCK(imgp->proc);
960 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
961 err_str = "Data segment size exceeds process limit";
962 else if (text_size > maxtsiz)
963 err_str = "Text segment size exceeds system limit";
964 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
965 err_str = "Total segment size exceeds process limit";
966 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
967 err_str = "Data segment size exceeds resource limit";
968 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
969 err_str = "Total segment size exceeds resource limit";
970 PROC_UNLOCK(imgp->proc);
971 if (err_str != NULL) {
972 uprintf("%s\n", err_str);
976 vmspace = imgp->proc->p_vmspace;
977 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
978 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
979 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
980 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
986 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
987 char **interpp, bool *free_interpp)
991 int error, interp_name_len;
993 KASSERT(phdr->p_type == PT_INTERP,
994 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
995 ASSERT_VOP_LOCKED(imgp->vp, __func__);
999 /* Path to interpreter */
1000 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1001 uprintf("Invalid PT_INTERP\n");
1005 interp_name_len = phdr->p_filesz;
1006 if (phdr->p_offset > PAGE_SIZE ||
1007 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1009 * The vnode lock might be needed by the pagedaemon to
1010 * clean pages owned by the vnode. Do not allow sleep
1011 * waiting for memory with the vnode locked, instead
1012 * try non-sleepable allocation first, and if it
1013 * fails, go to the slow path were we drop the lock
1014 * and do M_WAITOK. A text reference prevents
1015 * modifications to the vnode content.
1017 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1018 if (interp == NULL) {
1019 VOP_UNLOCK(imgp->vp);
1020 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1021 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1024 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1025 interp_name_len, phdr->p_offset,
1026 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1029 free(interp, M_TEMP);
1030 uprintf("i/o error PT_INTERP %d\n", error);
1033 interp[interp_name_len] = '\0';
1036 *free_interpp = true;
1040 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1041 if (interp[interp_name_len - 1] != '\0') {
1042 uprintf("Invalid PT_INTERP\n");
1047 *free_interpp = false;
1052 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1053 const char *interp, u_long *addr, u_long *entry)
1058 if (brand_info->emul_path != NULL &&
1059 brand_info->emul_path[0] != '\0') {
1060 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1061 snprintf(path, MAXPATHLEN, "%s%s",
1062 brand_info->emul_path, interp);
1063 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1069 if (brand_info->interp_newpath != NULL &&
1070 (brand_info->interp_path == NULL ||
1071 strcmp(interp, brand_info->interp_path) == 0)) {
1072 error = __elfN(load_file)(imgp->proc,
1073 brand_info->interp_newpath, addr, entry);
1078 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1082 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1087 * Impossible et_dyn_addr initial value indicating that the real base
1088 * must be calculated later with some randomization applied.
1090 #define ET_DYN_ADDR_RAND 1
1093 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1096 const Elf_Ehdr *hdr;
1097 const Elf_Phdr *phdr;
1098 Elf_Auxargs *elf_auxargs;
1099 struct vmspace *vmspace;
1102 Elf_Brandinfo *brand_info;
1103 struct sysentvec *sv;
1104 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1105 u_long maxalign, mapsz, maxv, maxv1;
1111 hdr = (const Elf_Ehdr *)imgp->image_header;
1114 * Do we have a valid ELF header ?
1116 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1117 * if particular brand doesn't support it.
1119 if (__elfN(check_header)(hdr) != 0 ||
1120 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1124 * From here on down, we return an errno, not -1, as we've
1125 * detected an ELF file.
1128 if (!__elfN(phdr_in_zero_page)(hdr)) {
1129 uprintf("Program headers not in the first page\n");
1132 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1133 if (!aligned(phdr, Elf_Addr)) {
1134 uprintf("Unaligned program headers\n");
1142 entry = proghdr = 0;
1144 free_interp = false;
1146 maxalign = PAGE_SIZE;
1149 for (i = 0; i < hdr->e_phnum; i++) {
1150 switch (phdr[i].p_type) {
1153 baddr = phdr[i].p_vaddr;
1154 if (phdr[i].p_align > maxalign)
1155 maxalign = phdr[i].p_align;
1156 mapsz += phdr[i].p_memsz;
1160 * If this segment contains the program headers,
1161 * remember their virtual address for the AT_PHDR
1162 * aux entry. Static binaries don't usually include
1165 if (phdr[i].p_offset == 0 &&
1166 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1168 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1171 /* Path to interpreter */
1172 if (interp != NULL) {
1173 uprintf("Multiple PT_INTERP headers\n");
1177 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1183 if (__elfN(nxstack))
1185 __elfN(trans_prot)(phdr[i].p_flags);
1186 imgp->stack_sz = phdr[i].p_memsz;
1188 case PT_PHDR: /* Program header table info */
1189 proghdr = phdr[i].p_vaddr;
1194 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1195 if (brand_info == NULL) {
1196 uprintf("ELF binary type \"%u\" not known.\n",
1197 hdr->e_ident[EI_OSABI]);
1201 sv = brand_info->sysvec;
1203 if (hdr->e_type == ET_DYN) {
1204 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1205 uprintf("Cannot execute shared object\n");
1210 * Honour the base load address from the dso if it is
1211 * non-zero for some reason.
1214 if ((sv->sv_flags & SV_ASLR) == 0 ||
1215 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1216 et_dyn_addr = __elfN(pie_base);
1217 else if ((__elfN(pie_aslr_enabled) &&
1218 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1219 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1220 et_dyn_addr = ET_DYN_ADDR_RAND;
1222 et_dyn_addr = __elfN(pie_base);
1227 * Avoid a possible deadlock if the current address space is destroyed
1228 * and that address space maps the locked vnode. In the common case,
1229 * the locked vnode's v_usecount is decremented but remains greater
1230 * than zero. Consequently, the vnode lock is not needed by vrele().
1231 * However, in cases where the vnode lock is external, such as nullfs,
1232 * v_usecount may become zero.
1234 * The VV_TEXT flag prevents modifications to the executable while
1235 * the vnode is unlocked.
1237 VOP_UNLOCK(imgp->vp);
1240 * Decide whether to enable randomization of user mappings.
1241 * First, reset user preferences for the setid binaries.
1242 * Then, account for the support of the randomization by the
1243 * ABI, by user preferences, and make special treatment for
1246 if (imgp->credential_setid) {
1247 PROC_LOCK(imgp->proc);
1248 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1249 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1250 PROC_UNLOCK(imgp->proc);
1252 if ((sv->sv_flags & SV_ASLR) == 0 ||
1253 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1254 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1255 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1256 ("et_dyn_addr == RAND and !ASLR"));
1257 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1258 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1259 et_dyn_addr == ET_DYN_ADDR_RAND) {
1260 imgp->map_flags |= MAP_ASLR;
1262 * If user does not care about sbrk, utilize the bss
1263 * grow region for mappings as well. We can select
1264 * the base for the image anywere and still not suffer
1265 * from the fragmentation.
1267 if (!__elfN(aslr_honor_sbrk) ||
1268 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1269 imgp->map_flags |= MAP_ASLR_IGNSTART;
1272 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1273 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1274 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1275 imgp->map_flags |= MAP_WXORX;
1277 error = exec_new_vmspace(imgp, sv);
1278 vmspace = imgp->proc->p_vmspace;
1279 map = &vmspace->vm_map;
1281 imgp->proc->p_sysent = sv;
1282 imgp->proc->p_elf_brandinfo = brand_info;
1284 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1285 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1286 KASSERT((map->flags & MAP_ASLR) != 0,
1287 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1288 error = __CONCAT(rnd_, __elfN(base))(map,
1289 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1290 /* reserve half of the address space to interpreter */
1291 maxv / 2, 1UL << flsl(maxalign), &et_dyn_addr);
1294 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1298 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1302 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1306 entry = (u_long)hdr->e_entry + et_dyn_addr;
1309 * We load the dynamic linker where a userland call
1310 * to mmap(0, ...) would put it. The rationale behind this
1311 * calculation is that it leaves room for the heap to grow to
1312 * its maximum allowed size.
1314 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1316 if ((map->flags & MAP_ASLR) != 0) {
1317 maxv1 = maxv / 2 + addr / 2;
1318 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1319 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1320 pagesizes[1] : pagesizes[0], &map->anon_loc);
1324 map->anon_loc = addr;
1327 imgp->entry_addr = entry;
1329 if (interp != NULL) {
1330 VOP_UNLOCK(imgp->vp);
1331 if ((map->flags & MAP_ASLR) != 0) {
1332 /* Assume that interpreter fits into 1/4 of AS */
1333 maxv1 = maxv / 2 + addr / 2;
1334 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1335 maxv1, PAGE_SIZE, &addr);
1338 error = __elfN(load_interp)(imgp, brand_info, interp,
1339 &addr, &imgp->entry_addr);
1341 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1348 * Construct auxargs table (used by the copyout_auxargs routine)
1350 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1351 if (elf_auxargs == NULL) {
1352 VOP_UNLOCK(imgp->vp);
1353 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1354 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1356 elf_auxargs->execfd = -1;
1357 elf_auxargs->phdr = proghdr + et_dyn_addr;
1358 elf_auxargs->phent = hdr->e_phentsize;
1359 elf_auxargs->phnum = hdr->e_phnum;
1360 elf_auxargs->pagesz = PAGE_SIZE;
1361 elf_auxargs->base = addr;
1362 elf_auxargs->flags = 0;
1363 elf_auxargs->entry = entry;
1364 elf_auxargs->hdr_eflags = hdr->e_flags;
1366 imgp->auxargs = elf_auxargs;
1367 imgp->interpreted = 0;
1368 imgp->reloc_base = addr;
1369 imgp->proc->p_osrel = osrel;
1370 imgp->proc->p_fctl0 = fctl0;
1371 imgp->proc->p_elf_flags = hdr->e_flags;
1374 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1376 free(interp, M_TEMP);
1380 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1383 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1385 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1386 Elf_Auxinfo *argarray, *pos;
1389 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1392 if (args->execfd != -1)
1393 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1394 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1395 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1396 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1397 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1398 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1399 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1400 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1401 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1402 if (imgp->execpathp != 0)
1403 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1404 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1405 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1406 if (imgp->canary != 0) {
1407 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1408 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1410 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1411 if (imgp->pagesizes != 0) {
1412 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1413 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1415 if (imgp->sysent->sv_timekeep_base != 0) {
1416 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1417 imgp->sysent->sv_timekeep_base);
1419 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1420 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1421 imgp->sysent->sv_stackprot);
1422 if (imgp->sysent->sv_hwcap != NULL)
1423 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1424 if (imgp->sysent->sv_hwcap2 != NULL)
1425 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1426 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1427 ELF_BSDF_SIGFASTBLK : 0);
1428 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1429 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1430 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1431 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1432 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1433 if (imgp->sysent->sv_fxrng_gen_base != 0)
1434 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1435 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1436 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1437 AUXARGS_ENTRY(pos, AT_NULL, 0);
1439 free(imgp->auxargs, M_TEMP);
1440 imgp->auxargs = NULL;
1441 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1443 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1444 free(argarray, M_TEMP);
1449 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1453 base = (Elf_Addr *)*stack_base;
1455 if (elf_suword(base, imgp->args->argc) == -1)
1457 *stack_base = (uintptr_t)base;
1462 * Code for generating ELF core dumps.
1465 typedef void (*segment_callback)(vm_map_entry_t, void *);
1467 /* Closure for cb_put_phdr(). */
1468 struct phdr_closure {
1469 Elf_Phdr *phdr; /* Program header to fill in */
1470 Elf_Off offset; /* Offset of segment in core file */
1474 int type; /* Note type. */
1475 outfunc_t outfunc; /* Output function. */
1476 void *outarg; /* Argument for the output function. */
1477 size_t outsize; /* Output size. */
1478 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1481 TAILQ_HEAD(note_info_list, note_info);
1483 extern int compress_user_cores;
1484 extern int compress_user_cores_level;
1486 static void cb_put_phdr(vm_map_entry_t, void *);
1487 static void cb_size_segment(vm_map_entry_t, void *);
1488 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1490 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1491 struct note_info_list *, size_t, int);
1492 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1494 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1495 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1496 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1497 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1498 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1499 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1500 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1501 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1502 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1503 static void note_procstat_files(void *, struct sbuf *, size_t *);
1504 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1505 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1506 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1507 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1508 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1511 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1514 return (core_write((struct coredump_params *)arg, base, len, offset,
1515 UIO_SYSSPACE, NULL));
1519 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1521 struct ucred *cred = td->td_ucred;
1522 int compm, error = 0;
1523 struct sseg_closure seginfo;
1524 struct note_info_list notelst;
1525 struct coredump_params params;
1526 struct note_info *ninfo;
1528 size_t hdrsize, notesz, coresize;
1532 TAILQ_INIT(¬elst);
1534 /* Size the program segments. */
1535 __elfN(size_segments)(td, &seginfo, flags);
1538 * Collect info about the core file header area.
1540 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1541 if (seginfo.count + 1 >= PN_XNUM)
1542 hdrsize += sizeof(Elf_Shdr);
1543 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1544 coresize = round_page(hdrsize + notesz) + seginfo.size;
1546 /* Set up core dump parameters. */
1548 params.active_cred = cred;
1549 params.file_cred = NOCRED;
1556 PROC_LOCK(td->td_proc);
1557 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1558 PROC_UNLOCK(td->td_proc);
1565 if (coresize >= limit) {
1570 /* Create a compression stream if necessary. */
1571 compm = compress_user_cores;
1572 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1574 compm = COMPRESS_GZIP;
1576 params.comp = compressor_init(core_compressed_write,
1577 compm, CORE_BUF_SIZE,
1578 compress_user_cores_level, ¶ms);
1579 if (params.comp == NULL) {
1583 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1587 * Allocate memory for building the header, fill it up,
1588 * and write it out following the notes.
1590 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1591 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1594 /* Write the contents of all of the writable segments. */
1600 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1601 offset = round_page(hdrsize + notesz);
1602 for (i = 0; i < seginfo.count; i++) {
1603 error = core_output((char *)(uintptr_t)php->p_vaddr,
1604 php->p_filesz, offset, ¶ms, tmpbuf);
1607 offset += php->p_filesz;
1610 if (error == 0 && params.comp != NULL)
1611 error = compressor_flush(params.comp);
1615 "Failed to write core file for process %s (error %d)\n",
1616 curproc->p_comm, error);
1620 free(tmpbuf, M_TEMP);
1621 if (params.comp != NULL)
1622 compressor_fini(params.comp);
1623 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1624 TAILQ_REMOVE(¬elst, ninfo, link);
1625 free(ninfo, M_TEMP);
1634 * A callback for each_dumpable_segment() to write out the segment's
1635 * program header entry.
1638 cb_put_phdr(vm_map_entry_t entry, void *closure)
1640 struct phdr_closure *phc = (struct phdr_closure *)closure;
1641 Elf_Phdr *phdr = phc->phdr;
1643 phc->offset = round_page(phc->offset);
1645 phdr->p_type = PT_LOAD;
1646 phdr->p_offset = phc->offset;
1647 phdr->p_vaddr = entry->start;
1649 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1650 phdr->p_align = PAGE_SIZE;
1651 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1653 phc->offset += phdr->p_filesz;
1658 * A callback for each_dumpable_segment() to gather information about
1659 * the number of segments and their total size.
1662 cb_size_segment(vm_map_entry_t entry, void *closure)
1664 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1667 ssc->size += entry->end - entry->start;
1671 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1677 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1681 * For each writable segment in the process's memory map, call the given
1682 * function with a pointer to the map entry and some arbitrary
1683 * caller-supplied data.
1686 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1689 struct proc *p = td->td_proc;
1690 vm_map_t map = &p->p_vmspace->vm_map;
1691 vm_map_entry_t entry;
1692 vm_object_t backing_object, object;
1695 vm_map_lock_read(map);
1696 VM_MAP_ENTRY_FOREACH(entry, map) {
1698 * Don't dump inaccessible mappings, deal with legacy
1701 * Note that read-only segments related to the elf binary
1702 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1703 * need to arbitrarily ignore such segments.
1705 if ((flags & SVC_ALL) == 0) {
1706 if (elf_legacy_coredump) {
1707 if ((entry->protection & VM_PROT_RW) !=
1711 if ((entry->protection & VM_PROT_ALL) == 0)
1717 * Dont include memory segment in the coredump if
1718 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1719 * madvise(2). Do not dump submaps (i.e. parts of the
1722 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1724 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1725 (flags & SVC_ALL) == 0)
1727 if ((object = entry->object.vm_object) == NULL)
1730 /* Ignore memory-mapped devices and such things. */
1731 VM_OBJECT_RLOCK(object);
1732 while ((backing_object = object->backing_object) != NULL) {
1733 VM_OBJECT_RLOCK(backing_object);
1734 VM_OBJECT_RUNLOCK(object);
1735 object = backing_object;
1737 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1738 VM_OBJECT_RUNLOCK(object);
1742 (*func)(entry, closure);
1744 vm_map_unlock_read(map);
1748 * Write the core file header to the file, including padding up to
1749 * the page boundary.
1752 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1753 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1756 struct note_info *ninfo;
1760 /* Fill in the header. */
1761 bzero(hdr, hdrsize);
1762 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1764 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1765 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1766 sbuf_start_section(sb, NULL);
1767 sbuf_bcat(sb, hdr, hdrsize);
1768 TAILQ_FOREACH(ninfo, notelst, link)
1769 __elfN(putnote)(p->td, ninfo, sb);
1770 /* Align up to a page boundary for the program segments. */
1771 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1772 error = sbuf_finish(sb);
1779 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1789 size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1792 * To have the debugger select the right thread (LWP) as the initial
1793 * thread, we dump the state of the thread passed to us in td first.
1794 * This is the thread that causes the core dump and thus likely to
1795 * be the right thread one wants to have selected in the debugger.
1798 while (thr != NULL) {
1799 size += __elfN(register_note)(td, list, NT_PRSTATUS,
1800 __elfN(note_prstatus), thr);
1801 size += __elfN(register_note)(td, list, NT_FPREGSET,
1802 __elfN(note_fpregset), thr);
1803 size += __elfN(register_note)(td, list, NT_THRMISC,
1804 __elfN(note_thrmisc), thr);
1805 size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1806 __elfN(note_ptlwpinfo), thr);
1807 size += __elfN(register_note)(td, list, -1,
1808 __elfN(note_threadmd), thr);
1810 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1811 TAILQ_NEXT(thr, td_plist);
1813 thr = TAILQ_NEXT(thr, td_plist);
1816 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1817 __elfN(note_procstat_proc), p);
1818 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1819 note_procstat_files, p);
1820 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1821 note_procstat_vmmap, p);
1822 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1823 note_procstat_groups, p);
1824 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1825 note_procstat_umask, p);
1826 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1827 note_procstat_rlimit, p);
1828 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1829 note_procstat_osrel, p);
1830 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1831 __elfN(note_procstat_psstrings), p);
1832 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1833 __elfN(note_procstat_auxv), p);
1839 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1840 size_t notesz, int flags)
1845 struct phdr_closure phc;
1848 ehdr = (Elf_Ehdr *)hdr;
1849 bi = td->td_proc->p_elf_brandinfo;
1851 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1852 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1853 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1854 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1855 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1856 ehdr->e_ident[EI_DATA] = ELF_DATA;
1857 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1858 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1859 ehdr->e_ident[EI_ABIVERSION] = 0;
1860 ehdr->e_ident[EI_PAD] = 0;
1861 ehdr->e_type = ET_CORE;
1862 ehdr->e_machine = bi->machine;
1863 ehdr->e_version = EV_CURRENT;
1865 ehdr->e_phoff = sizeof(Elf_Ehdr);
1866 ehdr->e_flags = td->td_proc->p_elf_flags;
1867 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1868 ehdr->e_phentsize = sizeof(Elf_Phdr);
1869 ehdr->e_shentsize = sizeof(Elf_Shdr);
1870 ehdr->e_shstrndx = SHN_UNDEF;
1871 if (numsegs + 1 < PN_XNUM) {
1872 ehdr->e_phnum = numsegs + 1;
1875 ehdr->e_phnum = PN_XNUM;
1878 ehdr->e_shoff = ehdr->e_phoff +
1879 (numsegs + 1) * ehdr->e_phentsize;
1880 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1881 ("e_shoff: %zu, hdrsize - shdr: %zu",
1882 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1884 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1885 memset(shdr, 0, sizeof(*shdr));
1887 * A special first section is used to hold large segment and
1888 * section counts. This was proposed by Sun Microsystems in
1889 * Solaris and has been adopted by Linux; the standard ELF
1890 * tools are already familiar with the technique.
1892 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1893 * (or 12-7 depending on the version of the document) for more
1896 shdr->sh_type = SHT_NULL;
1897 shdr->sh_size = ehdr->e_shnum;
1898 shdr->sh_link = ehdr->e_shstrndx;
1899 shdr->sh_info = numsegs + 1;
1903 * Fill in the program header entries.
1905 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1907 /* The note segement. */
1908 phdr->p_type = PT_NOTE;
1909 phdr->p_offset = hdrsize;
1912 phdr->p_filesz = notesz;
1914 phdr->p_flags = PF_R;
1915 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1918 /* All the writable segments from the program. */
1920 phc.offset = round_page(hdrsize + notesz);
1921 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1925 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1926 int type, outfunc_t out, void *arg)
1928 const struct sysentvec *sv;
1929 struct note_info *ninfo;
1930 size_t size, notesize;
1932 sv = td->td_proc->p_sysent;
1934 out(arg, NULL, &size);
1935 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1937 ninfo->outfunc = out;
1938 ninfo->outarg = arg;
1939 ninfo->outsize = size;
1940 TAILQ_INSERT_TAIL(list, ninfo, link);
1945 notesize = sizeof(Elf_Note) + /* note header */
1946 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1948 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1954 append_note_data(const void *src, void *dst, size_t len)
1958 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1960 bcopy(src, dst, len);
1961 bzero((char *)dst + len, padded_len - len);
1963 return (padded_len);
1967 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1975 note = (Elf_Note *)buf;
1976 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1977 note->n_descsz = size;
1978 note->n_type = type;
1979 buf += sizeof(*note);
1980 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1981 sizeof(FREEBSD_ABI_VENDOR));
1982 append_note_data(src, buf, size);
1987 notesize = sizeof(Elf_Note) + /* note header */
1988 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1990 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1996 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1999 const struct sysentvec *sv;
2000 ssize_t old_len, sect_len;
2001 size_t new_len, descsz, i;
2003 if (ninfo->type == -1) {
2004 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2008 sv = td->td_proc->p_sysent;
2010 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2011 note.n_descsz = ninfo->outsize;
2012 note.n_type = ninfo->type;
2014 sbuf_bcat(sb, ¬e, sizeof(note));
2015 sbuf_start_section(sb, &old_len);
2016 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2017 strlen(sv->sv_elf_core_abi_vendor) + 1);
2018 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2019 if (note.n_descsz == 0)
2021 sbuf_start_section(sb, &old_len);
2022 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2023 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2027 new_len = (size_t)sect_len;
2028 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2029 if (new_len < descsz) {
2031 * It is expected that individual note emitters will correctly
2032 * predict their expected output size and fill up to that size
2033 * themselves, padding in a format-specific way if needed.
2034 * However, in case they don't, just do it here with zeros.
2036 for (i = 0; i < descsz - new_len; i++)
2038 } else if (new_len > descsz) {
2040 * We can't always truncate sb -- we may have drained some
2043 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2044 "read it (%zu > %zu). Since it is longer than "
2045 "expected, this coredump's notes are corrupt. THIS "
2046 "IS A BUG in the note_procstat routine for type %u.\n",
2047 __func__, (unsigned)note.n_type, new_len, descsz,
2048 (unsigned)note.n_type));
2053 * Miscellaneous note out functions.
2056 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2057 #include <compat/freebsd32/freebsd32.h>
2058 #include <compat/freebsd32/freebsd32_signal.h>
2060 typedef struct prstatus32 elf_prstatus_t;
2061 typedef struct prpsinfo32 elf_prpsinfo_t;
2062 typedef struct fpreg32 elf_prfpregset_t;
2063 typedef struct fpreg32 elf_fpregset_t;
2064 typedef struct reg32 elf_gregset_t;
2065 typedef struct thrmisc32 elf_thrmisc_t;
2066 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2067 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2068 typedef uint32_t elf_ps_strings_t;
2070 typedef prstatus_t elf_prstatus_t;
2071 typedef prpsinfo_t elf_prpsinfo_t;
2072 typedef prfpregset_t elf_prfpregset_t;
2073 typedef prfpregset_t elf_fpregset_t;
2074 typedef gregset_t elf_gregset_t;
2075 typedef thrmisc_t elf_thrmisc_t;
2076 #define ELF_KERN_PROC_MASK 0
2077 typedef struct kinfo_proc elf_kinfo_proc_t;
2078 typedef vm_offset_t elf_ps_strings_t;
2082 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2088 elf_prpsinfo_t *psinfo;
2093 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2094 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2095 psinfo->pr_version = PRPSINFO_VERSION;
2096 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2097 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2099 if (p->p_args != NULL) {
2100 len = sizeof(psinfo->pr_psargs) - 1;
2101 if (len > p->p_args->ar_length)
2102 len = p->p_args->ar_length;
2103 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2109 sbuf_new(&sbarg, psinfo->pr_psargs,
2110 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2111 error = proc_getargv(curthread, p, &sbarg);
2113 if (sbuf_finish(&sbarg) == 0)
2114 len = sbuf_len(&sbarg) - 1;
2116 len = sizeof(psinfo->pr_psargs) - 1;
2117 sbuf_delete(&sbarg);
2119 if (error || len == 0)
2120 strlcpy(psinfo->pr_psargs, p->p_comm,
2121 sizeof(psinfo->pr_psargs));
2123 KASSERT(len < sizeof(psinfo->pr_psargs),
2124 ("len is too long: %zu vs %zu", len,
2125 sizeof(psinfo->pr_psargs)));
2126 cp = psinfo->pr_psargs;
2129 cp = memchr(cp, '\0', end - cp);
2135 psinfo->pr_pid = p->p_pid;
2136 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2137 free(psinfo, M_TEMP);
2139 *sizep = sizeof(*psinfo);
2143 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2146 elf_prstatus_t *status;
2150 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2151 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2152 status->pr_version = PRSTATUS_VERSION;
2153 status->pr_statussz = sizeof(elf_prstatus_t);
2154 status->pr_gregsetsz = sizeof(elf_gregset_t);
2155 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2156 status->pr_osreldate = osreldate;
2157 status->pr_cursig = td->td_proc->p_sig;
2158 status->pr_pid = td->td_tid;
2159 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2160 fill_regs32(td, &status->pr_reg);
2162 fill_regs(td, &status->pr_reg);
2164 sbuf_bcat(sb, status, sizeof(*status));
2165 free(status, M_TEMP);
2167 *sizep = sizeof(*status);
2171 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2174 elf_prfpregset_t *fpregset;
2178 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2179 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2180 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2181 fill_fpregs32(td, fpregset);
2183 fill_fpregs(td, fpregset);
2185 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2186 free(fpregset, M_TEMP);
2188 *sizep = sizeof(*fpregset);
2192 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2195 elf_thrmisc_t thrmisc;
2199 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2200 bzero(&thrmisc, sizeof(thrmisc));
2201 strcpy(thrmisc.pr_tname, td->td_name);
2202 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2204 *sizep = sizeof(thrmisc);
2208 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2213 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2214 struct ptrace_lwpinfo32 pl;
2216 struct ptrace_lwpinfo pl;
2220 size = sizeof(structsize) + sizeof(pl);
2222 KASSERT(*sizep == size, ("invalid size"));
2223 structsize = sizeof(pl);
2224 sbuf_bcat(sb, &structsize, sizeof(structsize));
2225 bzero(&pl, sizeof(pl));
2226 pl.pl_lwpid = td->td_tid;
2227 pl.pl_event = PL_EVENT_NONE;
2228 pl.pl_sigmask = td->td_sigmask;
2229 pl.pl_siglist = td->td_siglist;
2230 if (td->td_si.si_signo != 0) {
2231 pl.pl_event = PL_EVENT_SIGNAL;
2232 pl.pl_flags |= PL_FLAG_SI;
2233 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2234 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2236 pl.pl_siginfo = td->td_si;
2239 strcpy(pl.pl_tdname, td->td_name);
2240 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2241 sbuf_bcat(sb, &pl, sizeof(pl));
2247 * Allow for MD specific notes, as well as any MD
2248 * specific preparations for writing MI notes.
2251 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2257 td = (struct thread *)arg;
2259 if (size != 0 && sb != NULL)
2260 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2264 __elfN(dump_thread)(td, buf, &size);
2265 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2266 if (size != 0 && sb != NULL)
2267 sbuf_bcat(sb, buf, size);
2272 #ifdef KINFO_PROC_SIZE
2273 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2277 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2284 size = sizeof(structsize) + p->p_numthreads *
2285 sizeof(elf_kinfo_proc_t);
2288 KASSERT(*sizep == size, ("invalid size"));
2289 structsize = sizeof(elf_kinfo_proc_t);
2290 sbuf_bcat(sb, &structsize, sizeof(structsize));
2291 sx_slock(&proctree_lock);
2293 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2294 sx_sunlock(&proctree_lock);
2299 #ifdef KINFO_FILE_SIZE
2300 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2304 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2307 size_t size, sect_sz, i;
2308 ssize_t start_len, sect_len;
2309 int structsize, filedesc_flags;
2311 if (coredump_pack_fileinfo)
2312 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2317 structsize = sizeof(struct kinfo_file);
2320 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2321 sbuf_set_drain(sb, sbuf_count_drain, &size);
2322 sbuf_bcat(sb, &structsize, sizeof(structsize));
2324 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2329 sbuf_start_section(sb, &start_len);
2331 sbuf_bcat(sb, &structsize, sizeof(structsize));
2333 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2336 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2341 KASSERT(sect_sz <= *sizep,
2342 ("kern_proc_filedesc_out did not respect maxlen; "
2343 "requested %zu, got %zu", *sizep - sizeof(structsize),
2344 sect_sz - sizeof(structsize)));
2346 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2351 #ifdef KINFO_VMENTRY_SIZE
2352 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2356 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2360 int structsize, vmmap_flags;
2362 if (coredump_pack_vmmapinfo)
2363 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2368 structsize = sizeof(struct kinfo_vmentry);
2371 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2372 sbuf_set_drain(sb, sbuf_count_drain, &size);
2373 sbuf_bcat(sb, &structsize, sizeof(structsize));
2375 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2380 sbuf_bcat(sb, &structsize, sizeof(structsize));
2382 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2388 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2395 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2397 KASSERT(*sizep == size, ("invalid size"));
2398 structsize = sizeof(gid_t);
2399 sbuf_bcat(sb, &structsize, sizeof(structsize));
2400 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2407 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2414 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2416 KASSERT(*sizep == size, ("invalid size"));
2417 structsize = sizeof(p->p_pd->pd_cmask);
2418 sbuf_bcat(sb, &structsize, sizeof(structsize));
2419 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2425 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2428 struct rlimit rlim[RLIM_NLIMITS];
2433 size = sizeof(structsize) + sizeof(rlim);
2435 KASSERT(*sizep == size, ("invalid size"));
2436 structsize = sizeof(rlim);
2437 sbuf_bcat(sb, &structsize, sizeof(structsize));
2439 for (i = 0; i < RLIM_NLIMITS; i++)
2440 lim_rlimit_proc(p, i, &rlim[i]);
2442 sbuf_bcat(sb, rlim, sizeof(rlim));
2448 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2455 size = sizeof(structsize) + sizeof(p->p_osrel);
2457 KASSERT(*sizep == size, ("invalid size"));
2458 structsize = sizeof(p->p_osrel);
2459 sbuf_bcat(sb, &structsize, sizeof(structsize));
2460 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2466 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2469 elf_ps_strings_t ps_strings;
2474 size = sizeof(structsize) + sizeof(ps_strings);
2476 KASSERT(*sizep == size, ("invalid size"));
2477 structsize = sizeof(ps_strings);
2478 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2479 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2481 ps_strings = p->p_sysent->sv_psstrings;
2483 sbuf_bcat(sb, &structsize, sizeof(structsize));
2484 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2490 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2499 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2501 sbuf_set_drain(sb, sbuf_count_drain, &size);
2502 sbuf_bcat(sb, &structsize, sizeof(structsize));
2504 proc_getauxv(curthread, p, sb);
2510 structsize = sizeof(Elf_Auxinfo);
2511 sbuf_bcat(sb, &structsize, sizeof(structsize));
2513 proc_getauxv(curthread, p, sb);
2519 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2520 const char *note_vendor, const Elf_Phdr *pnote,
2521 bool (*cb)(const Elf_Note *, void *, bool *), void *cb_arg)
2523 const Elf_Note *note, *note0, *note_end;
2524 const char *note_name;
2529 /* We need some limit, might as well use PAGE_SIZE. */
2530 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2532 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2533 if (pnote->p_offset > PAGE_SIZE ||
2534 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2535 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2537 VOP_UNLOCK(imgp->vp);
2538 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2539 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2541 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2542 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2543 curthread->td_ucred, NOCRED, NULL, curthread);
2545 uprintf("i/o error PT_NOTE\n");
2548 note = note0 = (const Elf_Note *)buf;
2549 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2551 note = note0 = (const Elf_Note *)(imgp->image_header +
2553 note_end = (const Elf_Note *)(imgp->image_header +
2554 pnote->p_offset + pnote->p_filesz);
2557 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2558 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2559 (const char *)note < sizeof(Elf_Note)) {
2562 if (note->n_namesz != checknote->n_namesz ||
2563 note->n_descsz != checknote->n_descsz ||
2564 note->n_type != checknote->n_type)
2566 note_name = (const char *)(note + 1);
2567 if (note_name + checknote->n_namesz >=
2568 (const char *)note_end || strncmp(note_vendor,
2569 note_name, checknote->n_namesz) != 0)
2572 if (cb(note, cb_arg, &res))
2575 note = (const Elf_Note *)((const char *)(note + 1) +
2576 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2577 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2586 struct brandnote_cb_arg {
2587 Elf_Brandnote *brandnote;
2592 brandnote_cb(const Elf_Note *note, void *arg0, bool *res)
2594 struct brandnote_cb_arg *arg;
2599 * Fetch the osreldate for binary from the ELF OSABI-note if
2602 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2603 arg->brandnote->trans_osrel != NULL ?
2604 arg->brandnote->trans_osrel(note, arg->osrel) : true;
2609 static Elf_Note fctl_note = {
2610 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2611 .n_descsz = sizeof(uint32_t),
2612 .n_type = NT_FREEBSD_FEATURE_CTL,
2615 struct fctl_cb_arg {
2621 note_fctl_cb(const Elf_Note *note, void *arg0, bool *res)
2623 struct fctl_cb_arg *arg;
2624 const Elf32_Word *desc;
2628 p = (uintptr_t)(note + 1);
2629 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2630 desc = (const Elf32_Word *)p;
2631 *arg->has_fctl0 = true;
2632 *arg->fctl0 = desc[0];
2638 * Try to find the appropriate ABI-note section for checknote, fetch
2639 * the osreldate and feature control flags for binary from the ELF
2640 * OSABI-note. Only the first page of the image is searched, the same
2644 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2645 int32_t *osrel, bool *has_fctl0, uint32_t *fctl0)
2647 const Elf_Phdr *phdr;
2648 const Elf_Ehdr *hdr;
2649 struct brandnote_cb_arg b_arg;
2650 struct fctl_cb_arg f_arg;
2653 hdr = (const Elf_Ehdr *)imgp->image_header;
2654 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2655 b_arg.brandnote = brandnote;
2656 b_arg.osrel = osrel;
2657 f_arg.has_fctl0 = has_fctl0;
2658 f_arg.fctl0 = fctl0;
2660 for (i = 0; i < hdr->e_phnum; i++) {
2661 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2662 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2664 for (j = 0; j < hdr->e_phnum; j++) {
2665 if (phdr[j].p_type == PT_NOTE &&
2666 __elfN(parse_notes)(imgp, &fctl_note,
2667 FREEBSD_ABI_VENDOR, &phdr[j],
2668 note_fctl_cb, &f_arg))
2679 * Tell kern_execve.c about it, with a little help from the linker.
2681 static struct execsw __elfN(execsw) = {
2682 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2683 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2685 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2688 __elfN(trans_prot)(Elf_Word flags)
2694 prot |= VM_PROT_EXECUTE;
2696 prot |= VM_PROT_WRITE;
2698 prot |= VM_PROT_READ;
2699 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2700 if (i386_read_exec && (flags & PF_R))
2701 prot |= VM_PROT_EXECUTE;
2707 __elfN(untrans_prot)(vm_prot_t prot)
2712 if (prot & VM_PROT_EXECUTE)
2714 if (prot & VM_PROT_READ)
2716 if (prot & VM_PROT_WRITE)
2722 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2724 uintptr_t range, rbase, gap;
2727 pct = __elfN(aslr_stack_gap);
2732 range = imgp->eff_stack_sz * pct / 100;
2733 arc4rand(&rbase, sizeof(rbase), 0);
2734 gap = rbase % range;
2735 gap &= ~(sizeof(u_long) - 1);