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 #define CORE_BUF_SIZE (16 * 1024)
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118 &elf_legacy_coredump, 0,
119 "include all and only RW pages in core dumps");
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
133 #if defined(__amd64__)
134 static int __elfN(vdso) = 1;
135 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
136 vdso, CTLFLAG_RWTUN, &__elfN(vdso), 0,
137 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable vdso preloading");
139 static int __elfN(vdso) = 0;
142 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
143 int i386_read_exec = 0;
144 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
145 "enable execution from readable segments");
148 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
150 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
155 val = __elfN(pie_base);
156 error = sysctl_handle_long(oidp, &val, 0, req);
157 if (error != 0 || req->newptr == NULL)
159 if ((val & PAGE_MASK) != 0)
161 __elfN(pie_base) = val;
164 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
165 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
166 sysctl_pie_base, "LU",
167 "PIE load base without randomization");
169 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
172 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
174 static int __elfN(aslr_enabled) = 0;
175 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
176 &__elfN(aslr_enabled), 0,
177 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
178 ": enable address map randomization");
180 static int __elfN(pie_aslr_enabled) = 0;
181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
182 &__elfN(pie_aslr_enabled), 0,
183 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
184 ": enable address map randomization for PIE binaries");
186 static int __elfN(aslr_honor_sbrk) = 1;
187 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
188 &__elfN(aslr_honor_sbrk), 0,
189 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
191 static int __elfN(aslr_stack) = 1;
192 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
193 &__elfN(aslr_stack), 0,
194 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
195 ": enable stack address randomization");
197 static int __elfN(sigfastblock) = 1;
198 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
199 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
200 "enable sigfastblock for new processes");
202 static bool __elfN(allow_wx) = true;
203 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
204 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
205 "Allow pages to be mapped simultaneously writable and executable");
207 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
209 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
211 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
213 Elf_Brandnote __elfN(freebsd_brandnote) = {
214 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
215 .hdr.n_descsz = sizeof(int32_t),
216 .hdr.n_type = NT_FREEBSD_ABI_TAG,
217 .vendor = FREEBSD_ABI_VENDOR,
218 .flags = BN_TRANSLATE_OSREL,
219 .trans_osrel = __elfN(freebsd_trans_osrel)
223 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
227 p = (uintptr_t)(note + 1);
228 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
229 *osrel = *(const int32_t *)(p);
234 static const char GNU_ABI_VENDOR[] = "GNU";
235 static int GNU_KFREEBSD_ABI_DESC = 3;
237 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
238 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
239 .hdr.n_descsz = 16, /* XXX at least 16 */
241 .vendor = GNU_ABI_VENDOR,
242 .flags = BN_TRANSLATE_OSREL,
243 .trans_osrel = kfreebsd_trans_osrel
247 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
249 const Elf32_Word *desc;
252 p = (uintptr_t)(note + 1);
253 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
255 desc = (const Elf32_Word *)p;
256 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
260 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
261 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
263 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
269 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
273 for (i = 0; i < MAX_BRANDS; i++) {
274 if (elf_brand_list[i] == NULL) {
275 elf_brand_list[i] = entry;
279 if (i == MAX_BRANDS) {
280 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
288 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
292 for (i = 0; i < MAX_BRANDS; i++) {
293 if (elf_brand_list[i] == entry) {
294 elf_brand_list[i] = NULL;
304 __elfN(brand_inuse)(Elf_Brandinfo *entry)
309 sx_slock(&allproc_lock);
310 FOREACH_PROC_IN_SYSTEM(p) {
311 if (p->p_sysent == entry->sysvec) {
316 sx_sunlock(&allproc_lock);
321 static Elf_Brandinfo *
322 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
323 int32_t *osrel, uint32_t *fctl0)
325 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
326 Elf_Brandinfo *bi, *bi_m;
327 boolean_t ret, has_fctl0;
328 int i, interp_name_len;
330 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
333 * We support four types of branding -- (1) the ELF EI_OSABI field
334 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
335 * branding w/in the ELF header, (3) path of the `interp_path'
336 * field, and (4) the ".note.ABI-tag" ELF section.
339 /* Look for an ".note.ABI-tag" ELF section */
341 for (i = 0; i < MAX_BRANDS; i++) {
342 bi = elf_brand_list[i];
345 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
347 if (hdr->e_machine == bi->machine && (bi->flags &
348 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
352 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
354 /* Give brand a chance to veto check_note's guess */
355 if (ret && bi->header_supported) {
356 ret = bi->header_supported(imgp, osrel,
357 has_fctl0 ? fctl0 : NULL);
360 * If note checker claimed the binary, but the
361 * interpreter path in the image does not
362 * match default one for the brand, try to
363 * search for other brands with the same
364 * interpreter. Either there is better brand
365 * with the right interpreter, or, failing
366 * this, we return first brand which accepted
367 * our note and, optionally, header.
369 if (ret && bi_m == NULL && interp != NULL &&
370 (bi->interp_path == NULL ||
371 (strlen(bi->interp_path) + 1 != interp_name_len ||
372 strncmp(interp, bi->interp_path, interp_name_len)
384 /* If the executable has a brand, search for it in the brand list. */
385 for (i = 0; i < MAX_BRANDS; i++) {
386 bi = elf_brand_list[i];
387 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
388 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
390 if (hdr->e_machine == bi->machine &&
391 (hdr->e_ident[EI_OSABI] == bi->brand ||
392 (bi->compat_3_brand != NULL &&
393 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
394 bi->compat_3_brand) == 0))) {
395 /* Looks good, but give brand a chance to veto */
396 if (bi->header_supported == NULL ||
397 bi->header_supported(imgp, NULL, NULL)) {
399 * Again, prefer strictly matching
402 if (interp_name_len == 0 &&
403 bi->interp_path == NULL)
405 if (bi->interp_path != NULL &&
406 strlen(bi->interp_path) + 1 ==
407 interp_name_len && strncmp(interp,
408 bi->interp_path, interp_name_len) == 0)
418 /* No known brand, see if the header is recognized by any brand */
419 for (i = 0; i < MAX_BRANDS; i++) {
420 bi = elf_brand_list[i];
421 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
422 bi->header_supported == NULL)
424 if (hdr->e_machine == bi->machine) {
425 ret = bi->header_supported(imgp, NULL, NULL);
431 /* Lacking a known brand, search for a recognized interpreter. */
432 if (interp != NULL) {
433 for (i = 0; i < MAX_BRANDS; i++) {
434 bi = elf_brand_list[i];
435 if (bi == NULL || (bi->flags &
436 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
439 if (hdr->e_machine == bi->machine &&
440 bi->interp_path != NULL &&
441 /* ELF image p_filesz includes terminating zero */
442 strlen(bi->interp_path) + 1 == interp_name_len &&
443 strncmp(interp, bi->interp_path, interp_name_len)
444 == 0 && (bi->header_supported == NULL ||
445 bi->header_supported(imgp, NULL, NULL)))
450 /* Lacking a recognized interpreter, try the default brand */
451 for (i = 0; i < MAX_BRANDS; i++) {
452 bi = elf_brand_list[i];
453 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
454 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
456 if (hdr->e_machine == bi->machine &&
457 __elfN(fallback_brand) == bi->brand &&
458 (bi->header_supported == NULL ||
459 bi->header_supported(imgp, NULL, NULL)))
466 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
468 return (hdr->e_phoff <= PAGE_SIZE &&
469 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
473 __elfN(check_header)(const Elf_Ehdr *hdr)
479 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
480 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
481 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
482 hdr->e_phentsize != sizeof(Elf_Phdr) ||
483 hdr->e_version != ELF_TARG_VER)
487 * Make sure we have at least one brand for this machine.
490 for (i = 0; i < MAX_BRANDS; i++) {
491 bi = elf_brand_list[i];
492 if (bi != NULL && bi->machine == hdr->e_machine)
502 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
503 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
510 * Create the page if it doesn't exist yet. Ignore errors.
512 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
513 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
516 * Find the page from the underlying object.
518 if (object != NULL) {
519 sf = vm_imgact_map_page(object, offset);
521 return (KERN_FAILURE);
522 off = offset - trunc_page(offset);
523 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
525 vm_imgact_unmap_page(sf);
527 return (KERN_FAILURE);
530 return (KERN_SUCCESS);
534 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
535 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
541 int error, locked, rv;
543 if (start != trunc_page(start)) {
544 rv = __elfN(map_partial)(map, object, offset, start,
545 round_page(start), prot);
546 if (rv != KERN_SUCCESS)
548 offset += round_page(start) - start;
549 start = round_page(start);
551 if (end != round_page(end)) {
552 rv = __elfN(map_partial)(map, object, offset +
553 trunc_page(end) - start, trunc_page(end), end, prot);
554 if (rv != KERN_SUCCESS)
556 end = trunc_page(end);
559 return (KERN_SUCCESS);
560 if ((offset & PAGE_MASK) != 0) {
562 * The mapping is not page aligned. This means that we have
565 rv = vm_map_fixed(map, NULL, 0, start, end - start,
566 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
567 if (rv != KERN_SUCCESS)
570 return (KERN_SUCCESS);
571 for (; start < end; start += sz) {
572 sf = vm_imgact_map_page(object, offset);
574 return (KERN_FAILURE);
575 off = offset - trunc_page(offset);
577 if (sz > PAGE_SIZE - off)
578 sz = PAGE_SIZE - off;
579 error = copyout((caddr_t)sf_buf_kva(sf) + off,
581 vm_imgact_unmap_page(sf);
583 return (KERN_FAILURE);
587 vm_object_reference(object);
588 rv = vm_map_fixed(map, object, offset, start, end - start,
589 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
590 (object != NULL ? MAP_VN_EXEC : 0));
591 if (rv != KERN_SUCCESS) {
592 locked = VOP_ISLOCKED(imgp->vp);
593 VOP_UNLOCK(imgp->vp);
594 vm_object_deallocate(object);
595 vn_lock(imgp->vp, locked | LK_RETRY);
597 } else if (object != NULL) {
598 MPASS(imgp->vp->v_object == object);
599 VOP_SET_TEXT_CHECKED(imgp->vp);
602 return (KERN_SUCCESS);
606 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
607 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
613 vm_offset_t map_addr;
616 vm_ooffset_t file_addr;
619 * It's necessary to fail if the filsz + offset taken from the
620 * header is greater than the actual file pager object's size.
621 * If we were to allow this, then the vm_map_find() below would
622 * walk right off the end of the file object and into the ether.
624 * While I'm here, might as well check for something else that
625 * is invalid: filsz cannot be greater than memsz.
627 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
629 uprintf("elf_load_section: truncated ELF file\n");
633 object = imgp->object;
634 map = &imgp->proc->p_vmspace->vm_map;
635 map_addr = trunc_page((vm_offset_t)vmaddr);
636 file_addr = trunc_page(offset);
639 * We have two choices. We can either clear the data in the last page
640 * of an oversized mapping, or we can start the anon mapping a page
641 * early and copy the initialized data into that first page. We
646 else if (memsz > filsz)
647 map_len = trunc_page(offset + filsz) - file_addr;
649 map_len = round_page(offset + filsz) - file_addr;
652 /* cow flags: don't dump readonly sections in core */
653 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
654 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
656 rv = __elfN(map_insert)(imgp, map, object, file_addr,
657 map_addr, map_addr + map_len, prot, cow);
658 if (rv != KERN_SUCCESS)
661 /* we can stop now if we've covered it all */
667 * We have to get the remaining bit of the file into the first part
668 * of the oversized map segment. This is normally because the .data
669 * segment in the file is extended to provide bss. It's a neat idea
670 * to try and save a page, but it's a pain in the behind to implement.
672 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
674 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
675 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
677 /* This had damn well better be true! */
679 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
680 map_addr + map_len, prot, 0);
681 if (rv != KERN_SUCCESS)
686 sf = vm_imgact_map_page(object, offset + filsz);
690 /* send the page fragment to user space */
691 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
693 vm_imgact_unmap_page(sf);
699 * Remove write access to the page if it was only granted by map_insert
702 if ((prot & VM_PROT_WRITE) == 0)
703 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
704 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
710 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
711 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
718 ASSERT_VOP_LOCKED(imgp->vp, __func__);
723 for (i = 0; i < hdr->e_phnum; i++) {
724 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
727 /* Loadable segment */
728 prot = __elfN(trans_prot)(phdr[i].p_flags);
729 error = __elfN(load_section)(imgp, phdr[i].p_offset,
730 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
731 phdr[i].p_memsz, phdr[i].p_filesz, prot);
736 * Establish the base address if this is the first segment.
739 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
744 if (base_addrp != NULL)
745 *base_addrp = base_addr;
751 * Load the file "file" into memory. It may be either a shared object
754 * The "addr" reference parameter is in/out. On entry, it specifies
755 * the address where a shared object should be loaded. If the file is
756 * an executable, this value is ignored. On exit, "addr" specifies
757 * where the file was actually loaded.
759 * The "entry" reference parameter is out only. On exit, it specifies
760 * the entry point for the loaded file.
763 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
769 struct image_params image_params;
771 const Elf_Ehdr *hdr = NULL;
772 const Elf_Phdr *phdr = NULL;
773 struct nameidata *nd;
775 struct image_params *imgp;
777 u_long base_addr = 0;
780 #ifdef CAPABILITY_MODE
782 * XXXJA: This check can go away once we are sufficiently confident
783 * that the checks in namei() are correct.
785 if (IN_CAPABILITY_MODE(curthread))
789 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
791 attr = &tempdata->attr;
792 imgp = &tempdata->image_params;
795 * Initialize part of the common data
800 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
801 UIO_SYSSPACE, file, curthread);
802 if ((error = namei(nd)) != 0) {
806 NDFREE(nd, NDF_ONLY_PNBUF);
807 imgp->vp = nd->ni_vp;
810 * Check permissions, modes, uid, etc on the file, and "open" it.
812 error = exec_check_permissions(imgp);
816 error = exec_map_first_page(imgp);
820 imgp->object = nd->ni_vp->v_object;
822 hdr = (const Elf_Ehdr *)imgp->image_header;
823 if ((error = __elfN(check_header)(hdr)) != 0)
825 if (hdr->e_type == ET_DYN)
827 else if (hdr->e_type == ET_EXEC)
834 /* Only support headers that fit within first page for now */
835 if (!__elfN(phdr_in_zero_page)(hdr)) {
840 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
841 if (!aligned(phdr, Elf_Addr)) {
846 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
851 *entry = (unsigned long)hdr->e_entry + rbase;
855 exec_unmap_first_page(imgp);
859 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
862 free(tempdata, M_TEMP);
868 * Select randomized valid address in the map map, between minv and
869 * maxv, with specified alignment. The [minv, maxv) range must belong
870 * to the map. Note that function only allocates the address, it is
871 * up to caller to clamp maxv in a way that the final allocation
872 * length fit into the map.
874 * Result is returned in *resp, error code indicates that arguments
875 * did not pass sanity checks for overflow and range correctness.
878 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
879 u_int align, u_long *resp)
883 MPASS(vm_map_min(map) <= minv);
885 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
886 uprintf("Invalid ELF segments layout\n");
890 arc4rand(&rbase, sizeof(rbase), 0);
891 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
892 res &= ~((u_long)align - 1);
897 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
898 res, minv, maxv, rbase));
900 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
901 res, maxv, minv, rbase));
908 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
909 const Elf_Phdr *phdr, u_long et_dyn_addr)
911 struct vmspace *vmspace;
913 u_long text_size, data_size, total_size, text_addr, data_addr;
914 u_long seg_size, seg_addr;
918 text_size = data_size = total_size = text_addr = data_addr = 0;
920 for (i = 0; i < hdr->e_phnum; i++) {
921 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
924 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
925 seg_size = round_page(phdr[i].p_memsz +
926 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
929 * Make the largest executable segment the official
930 * text segment and all others data.
932 * Note that obreak() assumes that data_addr + data_size == end
933 * of data load area, and the ELF file format expects segments
934 * to be sorted by address. If multiple data segments exist,
935 * the last one will be used.
938 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
939 text_size = seg_size;
940 text_addr = seg_addr;
942 data_size = seg_size;
943 data_addr = seg_addr;
945 total_size += seg_size;
948 if (data_addr == 0 && data_size == 0) {
949 data_addr = text_addr;
950 data_size = text_size;
954 * Check limits. It should be safe to check the
955 * limits after loading the segments since we do
956 * not actually fault in all the segments pages.
958 PROC_LOCK(imgp->proc);
959 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
960 err_str = "Data segment size exceeds process limit";
961 else if (text_size > maxtsiz)
962 err_str = "Text segment size exceeds system limit";
963 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
964 err_str = "Total segment size exceeds process limit";
965 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
966 err_str = "Data segment size exceeds resource limit";
967 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
968 err_str = "Total segment size exceeds resource limit";
969 PROC_UNLOCK(imgp->proc);
970 if (err_str != NULL) {
971 uprintf("%s\n", err_str);
975 vmspace = imgp->proc->p_vmspace;
976 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
977 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
978 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
979 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
985 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
986 char **interpp, bool *free_interpp)
990 int error, interp_name_len;
992 KASSERT(phdr->p_type == PT_INTERP,
993 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
994 ASSERT_VOP_LOCKED(imgp->vp, __func__);
998 /* Path to interpreter */
999 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
1000 uprintf("Invalid PT_INTERP\n");
1004 interp_name_len = phdr->p_filesz;
1005 if (phdr->p_offset > PAGE_SIZE ||
1006 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1008 * The vnode lock might be needed by the pagedaemon to
1009 * clean pages owned by the vnode. Do not allow sleep
1010 * waiting for memory with the vnode locked, instead
1011 * try non-sleepable allocation first, and if it
1012 * fails, go to the slow path were we drop the lock
1013 * and do M_WAITOK. A text reference prevents
1014 * modifications to the vnode content.
1016 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1017 if (interp == NULL) {
1018 VOP_UNLOCK(imgp->vp);
1019 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1020 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1023 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1024 interp_name_len, phdr->p_offset,
1025 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1028 free(interp, M_TEMP);
1029 uprintf("i/o error PT_INTERP %d\n", error);
1032 interp[interp_name_len] = '\0';
1035 *free_interpp = true;
1039 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1040 if (interp[interp_name_len - 1] != '\0') {
1041 uprintf("Invalid PT_INTERP\n");
1046 *free_interpp = false;
1051 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1052 const char *interp, u_long *addr, u_long *entry)
1057 if (brand_info->emul_path != NULL &&
1058 brand_info->emul_path[0] != '\0') {
1059 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1060 snprintf(path, MAXPATHLEN, "%s%s",
1061 brand_info->emul_path, interp);
1062 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1068 if (brand_info->interp_newpath != NULL &&
1069 (brand_info->interp_path == NULL ||
1070 strcmp(interp, brand_info->interp_path) == 0)) {
1071 error = __elfN(load_file)(imgp->proc,
1072 brand_info->interp_newpath, addr, entry);
1077 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1081 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1086 * Impossible et_dyn_addr initial value indicating that the real base
1087 * must be calculated later with some randomization applied.
1089 #define ET_DYN_ADDR_RAND 1
1092 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1095 const Elf_Ehdr *hdr;
1096 const Elf_Phdr *phdr;
1097 Elf_Auxargs *elf_auxargs;
1098 struct vmspace *vmspace;
1101 Elf_Brandinfo *brand_info;
1102 struct sysentvec *sv;
1103 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1104 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1110 hdr = (const Elf_Ehdr *)imgp->image_header;
1113 * Do we have a valid ELF header ?
1115 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1116 * if particular brand doesn't support it.
1118 if (__elfN(check_header)(hdr) != 0 ||
1119 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1123 * From here on down, we return an errno, not -1, as we've
1124 * detected an ELF file.
1127 if (!__elfN(phdr_in_zero_page)(hdr)) {
1128 uprintf("Program headers not in the first page\n");
1131 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1132 if (!aligned(phdr, Elf_Addr)) {
1133 uprintf("Unaligned program headers\n");
1141 entry = proghdr = 0;
1143 free_interp = false;
1147 * Somewhat arbitrary, limit accepted max alignment for the
1148 * loadable segment to the max supported superpage size. Too
1149 * large alignment requests are not useful and are indicators
1150 * of corrupted or outright malicious binary.
1152 maxalign = PAGE_SIZE;
1153 maxsalign = PAGE_SIZE * 1024;
1154 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1155 if (pagesizes[i] > maxsalign)
1156 maxsalign = pagesizes[i];
1161 for (i = 0; i < hdr->e_phnum; i++) {
1162 switch (phdr[i].p_type) {
1165 baddr = phdr[i].p_vaddr;
1166 if (!powerof2(phdr[i].p_align) ||
1167 phdr[i].p_align > maxsalign) {
1168 uprintf("Invalid segment alignment\n");
1172 if (phdr[i].p_align > maxalign)
1173 maxalign = phdr[i].p_align;
1174 if (mapsz + phdr[i].p_memsz < mapsz) {
1175 uprintf("Mapsize overflow\n");
1179 mapsz += phdr[i].p_memsz;
1183 * If this segment contains the program headers,
1184 * remember their virtual address for the AT_PHDR
1185 * aux entry. Static binaries don't usually include
1188 if (phdr[i].p_offset == 0 &&
1189 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1191 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1194 /* Path to interpreter */
1195 if (interp != NULL) {
1196 uprintf("Multiple PT_INTERP headers\n");
1200 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1206 if (__elfN(nxstack))
1208 __elfN(trans_prot)(phdr[i].p_flags);
1209 imgp->stack_sz = phdr[i].p_memsz;
1211 case PT_PHDR: /* Program header table info */
1212 proghdr = phdr[i].p_vaddr;
1217 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1218 if (brand_info == NULL) {
1219 uprintf("ELF binary type \"%u\" not known.\n",
1220 hdr->e_ident[EI_OSABI]);
1224 sv = brand_info->sysvec;
1226 if (hdr->e_type == ET_DYN) {
1227 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1228 uprintf("Cannot execute shared object\n");
1233 * Honour the base load address from the dso if it is
1234 * non-zero for some reason.
1237 if ((sv->sv_flags & SV_ASLR) == 0 ||
1238 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1239 et_dyn_addr = __elfN(pie_base);
1240 else if ((__elfN(pie_aslr_enabled) &&
1241 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1242 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1243 et_dyn_addr = ET_DYN_ADDR_RAND;
1245 et_dyn_addr = __elfN(pie_base);
1250 * Avoid a possible deadlock if the current address space is destroyed
1251 * and that address space maps the locked vnode. In the common case,
1252 * the locked vnode's v_usecount is decremented but remains greater
1253 * than zero. Consequently, the vnode lock is not needed by vrele().
1254 * However, in cases where the vnode lock is external, such as nullfs,
1255 * v_usecount may become zero.
1257 * The VV_TEXT flag prevents modifications to the executable while
1258 * the vnode is unlocked.
1260 VOP_UNLOCK(imgp->vp);
1263 * Decide whether to enable randomization of user mappings.
1264 * First, reset user preferences for the setid binaries.
1265 * Then, account for the support of the randomization by the
1266 * ABI, by user preferences, and make special treatment for
1269 if (imgp->credential_setid) {
1270 PROC_LOCK(imgp->proc);
1271 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1272 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1273 PROC_UNLOCK(imgp->proc);
1275 if ((sv->sv_flags & SV_ASLR) == 0 ||
1276 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1277 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1278 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1279 ("et_dyn_addr == RAND and !ASLR"));
1280 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1281 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1282 et_dyn_addr == ET_DYN_ADDR_RAND) {
1283 imgp->map_flags |= MAP_ASLR;
1285 * If user does not care about sbrk, utilize the bss
1286 * grow region for mappings as well. We can select
1287 * the base for the image anywere and still not suffer
1288 * from the fragmentation.
1290 if (!__elfN(aslr_honor_sbrk) ||
1291 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1292 imgp->map_flags |= MAP_ASLR_IGNSTART;
1293 if (__elfN(aslr_stack))
1294 imgp->map_flags |= MAP_ASLR_STACK;
1297 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1298 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1299 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1300 imgp->map_flags |= MAP_WXORX;
1302 error = exec_new_vmspace(imgp, sv);
1304 imgp->proc->p_sysent = sv;
1306 vmspace = imgp->proc->p_vmspace;
1307 map = &vmspace->vm_map;
1308 maxv = sv->sv_usrstack;
1309 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1310 maxv -= lim_max(td, RLIMIT_STACK);
1311 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1312 uprintf("Excessive mapping size\n");
1316 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1317 KASSERT((map->flags & MAP_ASLR) != 0,
1318 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1319 error = __CONCAT(rnd_, __elfN(base))(map,
1320 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1321 /* reserve half of the address space to interpreter */
1322 maxv / 2, maxalign, &et_dyn_addr);
1325 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1329 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1333 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1338 * We load the dynamic linker where a userland call
1339 * to mmap(0, ...) would put it. The rationale behind this
1340 * calculation is that it leaves room for the heap to grow to
1341 * its maximum allowed size.
1343 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1345 if ((map->flags & MAP_ASLR) != 0) {
1346 maxv1 = maxv / 2 + addr / 2;
1347 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1348 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1349 pagesizes[1] : pagesizes[0], &anon_loc);
1352 map->anon_loc = anon_loc;
1354 map->anon_loc = addr;
1357 entry = (u_long)hdr->e_entry + et_dyn_addr;
1358 imgp->entry_addr = entry;
1360 if (interp != NULL) {
1361 VOP_UNLOCK(imgp->vp);
1362 if ((map->flags & MAP_ASLR) != 0) {
1363 /* Assume that interpreter fits into 1/4 of AS */
1364 maxv1 = maxv / 2 + addr / 2;
1365 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1366 maxv1, PAGE_SIZE, &addr);
1369 error = __elfN(load_interp)(imgp, brand_info, interp,
1370 &addr, &imgp->entry_addr);
1372 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1378 error = exec_map_stack(imgp);
1383 * Construct auxargs table (used by the copyout_auxargs routine)
1385 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1386 if (elf_auxargs == NULL) {
1387 VOP_UNLOCK(imgp->vp);
1388 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1389 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1391 elf_auxargs->execfd = -1;
1392 elf_auxargs->phdr = proghdr + et_dyn_addr;
1393 elf_auxargs->phent = hdr->e_phentsize;
1394 elf_auxargs->phnum = hdr->e_phnum;
1395 elf_auxargs->pagesz = PAGE_SIZE;
1396 elf_auxargs->base = addr;
1397 elf_auxargs->flags = 0;
1398 elf_auxargs->entry = entry;
1399 elf_auxargs->hdr_eflags = hdr->e_flags;
1401 imgp->auxargs = elf_auxargs;
1402 imgp->interpreted = 0;
1403 imgp->reloc_base = addr;
1404 imgp->proc->p_osrel = osrel;
1405 imgp->proc->p_fctl0 = fctl0;
1406 imgp->proc->p_elf_machine = hdr->e_machine;
1407 imgp->proc->p_elf_flags = hdr->e_flags;
1410 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1412 free(interp, M_TEMP);
1416 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1419 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1421 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1422 Elf_Auxinfo *argarray, *pos;
1425 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1428 if (args->execfd != -1)
1429 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1430 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1431 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1432 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1433 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1434 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1435 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1436 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1437 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1438 if (imgp->execpathp != 0)
1439 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1440 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1441 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1442 if (imgp->canary != 0) {
1443 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1444 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1446 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1447 if (imgp->pagesizes != 0) {
1448 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1449 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1451 if (imgp->sysent->sv_timekeep_base != 0) {
1452 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1453 imgp->sysent->sv_timekeep_base);
1455 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1456 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1457 imgp->sysent->sv_stackprot);
1458 if (imgp->sysent->sv_hwcap != NULL)
1459 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1460 if (imgp->sysent->sv_hwcap2 != NULL)
1461 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1462 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1463 ELF_BSDF_SIGFASTBLK : 0);
1464 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1465 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1466 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1467 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1468 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1469 if (imgp->sysent->sv_fxrng_gen_base != 0)
1470 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1471 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1472 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1473 AUXARGS_ENTRY(pos, AT_NULL, 0);
1475 free(imgp->auxargs, M_TEMP);
1476 imgp->auxargs = NULL;
1477 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1479 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1480 free(argarray, M_TEMP);
1485 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1489 base = (Elf_Addr *)*stack_base;
1491 if (elf_suword(base, imgp->args->argc) == -1)
1493 *stack_base = (uintptr_t)base;
1498 * Code for generating ELF core dumps.
1501 typedef void (*segment_callback)(vm_map_entry_t, void *);
1503 /* Closure for cb_put_phdr(). */
1504 struct phdr_closure {
1505 Elf_Phdr *phdr; /* Program header to fill in */
1506 Elf_Off offset; /* Offset of segment in core file */
1509 /* Closure for cb_size_segment(). */
1510 struct sseg_closure {
1511 int count; /* Count of writable segments. */
1512 size_t size; /* Total size of all writable segments. */
1515 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1518 int type; /* Note type. */
1519 outfunc_t outfunc; /* Output function. */
1520 void *outarg; /* Argument for the output function. */
1521 size_t outsize; /* Output size. */
1522 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1525 TAILQ_HEAD(note_info_list, note_info);
1527 /* Coredump output parameters. */
1528 struct coredump_params {
1530 struct ucred *active_cred;
1531 struct ucred *file_cred;
1534 struct compressor *comp;
1537 extern int compress_user_cores;
1538 extern int compress_user_cores_level;
1540 static void cb_put_phdr(vm_map_entry_t, void *);
1541 static void cb_size_segment(vm_map_entry_t, void *);
1542 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1543 enum uio_seg, size_t *);
1544 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1546 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1547 struct note_info_list *, size_t, int);
1548 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1550 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t, int);
1551 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1552 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1553 static int sbuf_drain_core_output(void *, const char *, int);
1555 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1556 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1557 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1558 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1559 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1560 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1561 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1562 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1563 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1564 static void note_procstat_files(void *, struct sbuf *, size_t *);
1565 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1566 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1567 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1568 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1569 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1572 * Write out a core segment to the compression stream.
1575 compress_chunk(struct coredump_params *p, char *base, char *buf, size_t len)
1581 chunk_len = MIN(len, CORE_BUF_SIZE);
1584 * We can get EFAULT error here.
1585 * In that case zero out the current chunk of the segment.
1587 error = copyin(base, buf, chunk_len);
1589 bzero(buf, chunk_len);
1590 error = compressor_write(p->comp, buf, chunk_len);
1600 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1603 return (core_write((struct coredump_params *)arg, base, len, offset,
1604 UIO_SYSSPACE, NULL));
1608 core_write(struct coredump_params *p, const void *base, size_t len,
1609 off_t offset, enum uio_seg seg, size_t *resid)
1612 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1613 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1614 p->active_cred, p->file_cred, resid, p->td));
1617 extern int core_dump_can_intr;
1620 core_output(char *base, size_t len, off_t offset, struct coredump_params *p,
1625 size_t resid, runlen;
1629 KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1630 ("%s: user address %p is not page-aligned", __func__, base));
1632 if (p->comp != NULL)
1633 return (compress_chunk(p, base, tmpbuf, len));
1635 map = &p->td->td_proc->p_vmspace->vm_map;
1636 for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1638 * Attempt to page in all virtual pages in the range. If a
1639 * virtual page is not backed by the pager, it is represented as
1640 * a hole in the file. This can occur with zero-filled
1641 * anonymous memory or truncated files, for example.
1643 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1644 if (core_dump_can_intr && curproc_sigkilled())
1646 error = vm_fault(map, (uintptr_t)base + runlen,
1647 VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1649 success = error == KERN_SUCCESS;
1650 else if ((error == KERN_SUCCESS) != success)
1655 error = core_write(p, base, runlen, offset,
1656 UIO_USERSPACE, &resid);
1658 if (error != EFAULT)
1662 * EFAULT may be returned if the user mapping
1663 * could not be accessed, e.g., because a mapped
1664 * file has been truncated. Skip the page if no
1665 * progress was made, to protect against a
1666 * hypothetical scenario where vm_fault() was
1667 * successful but core_write() returns EFAULT
1678 error = vn_start_write(p->vp, &mp, V_WAIT);
1681 vn_lock(p->vp, LK_EXCLUSIVE | LK_RETRY);
1682 error = vn_truncate_locked(p->vp, offset + runlen,
1683 false, p->td->td_ucred);
1685 vn_finished_write(mp);
1694 * Drain into a core file.
1697 sbuf_drain_core_output(void *arg, const char *data, int len)
1699 struct coredump_params *p;
1702 p = (struct coredump_params *)arg;
1705 * Some kern_proc out routines that print to this sbuf may
1706 * call us with the process lock held. Draining with the
1707 * non-sleepable lock held is unsafe. The lock is needed for
1708 * those routines when dumping a live process. In our case we
1709 * can safely release the lock before draining and acquire
1712 locked = PROC_LOCKED(p->td->td_proc);
1714 PROC_UNLOCK(p->td->td_proc);
1715 if (p->comp != NULL)
1716 error = compressor_write(p->comp, __DECONST(char *, data), len);
1718 error = core_write(p, __DECONST(void *, data), len, p->offset,
1719 UIO_SYSSPACE, NULL);
1721 PROC_LOCK(p->td->td_proc);
1729 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1731 struct ucred *cred = td->td_ucred;
1732 int compm, error = 0;
1733 struct sseg_closure seginfo;
1734 struct note_info_list notelst;
1735 struct coredump_params params;
1736 struct note_info *ninfo;
1738 size_t hdrsize, notesz, coresize;
1742 TAILQ_INIT(¬elst);
1744 /* Size the program segments. */
1747 each_dumpable_segment(td, cb_size_segment, &seginfo, flags);
1750 * Collect info about the core file header area.
1752 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1753 if (seginfo.count + 1 >= PN_XNUM)
1754 hdrsize += sizeof(Elf_Shdr);
1755 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1756 coresize = round_page(hdrsize + notesz) + seginfo.size;
1758 /* Set up core dump parameters. */
1760 params.active_cred = cred;
1761 params.file_cred = NOCRED;
1768 PROC_LOCK(td->td_proc);
1769 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1770 PROC_UNLOCK(td->td_proc);
1777 if (coresize >= limit) {
1782 /* Create a compression stream if necessary. */
1783 compm = compress_user_cores;
1784 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1786 compm = COMPRESS_GZIP;
1788 params.comp = compressor_init(core_compressed_write,
1789 compm, CORE_BUF_SIZE,
1790 compress_user_cores_level, ¶ms);
1791 if (params.comp == NULL) {
1795 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1799 * Allocate memory for building the header, fill it up,
1800 * and write it out following the notes.
1802 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1803 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1806 /* Write the contents of all of the writable segments. */
1812 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1813 offset = round_page(hdrsize + notesz);
1814 for (i = 0; i < seginfo.count; i++) {
1815 error = core_output((char *)(uintptr_t)php->p_vaddr,
1816 php->p_filesz, offset, ¶ms, tmpbuf);
1819 offset += php->p_filesz;
1822 if (error == 0 && params.comp != NULL)
1823 error = compressor_flush(params.comp);
1827 "Failed to write core file for process %s (error %d)\n",
1828 curproc->p_comm, error);
1832 free(tmpbuf, M_TEMP);
1833 if (params.comp != NULL)
1834 compressor_fini(params.comp);
1835 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1836 TAILQ_REMOVE(¬elst, ninfo, link);
1837 free(ninfo, M_TEMP);
1846 * A callback for each_dumpable_segment() to write out the segment's
1847 * program header entry.
1850 cb_put_phdr(vm_map_entry_t entry, void *closure)
1852 struct phdr_closure *phc = (struct phdr_closure *)closure;
1853 Elf_Phdr *phdr = phc->phdr;
1855 phc->offset = round_page(phc->offset);
1857 phdr->p_type = PT_LOAD;
1858 phdr->p_offset = phc->offset;
1859 phdr->p_vaddr = entry->start;
1861 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1862 phdr->p_align = PAGE_SIZE;
1863 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1865 phc->offset += phdr->p_filesz;
1870 * A callback for each_dumpable_segment() to gather information about
1871 * the number of segments and their total size.
1874 cb_size_segment(vm_map_entry_t entry, void *closure)
1876 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1879 ssc->size += entry->end - entry->start;
1883 * For each writable segment in the process's memory map, call the given
1884 * function with a pointer to the map entry and some arbitrary
1885 * caller-supplied data.
1888 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1891 struct proc *p = td->td_proc;
1892 vm_map_t map = &p->p_vmspace->vm_map;
1893 vm_map_entry_t entry;
1894 vm_object_t backing_object, object;
1897 vm_map_lock_read(map);
1898 VM_MAP_ENTRY_FOREACH(entry, map) {
1900 * Don't dump inaccessible mappings, deal with legacy
1903 * Note that read-only segments related to the elf binary
1904 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1905 * need to arbitrarily ignore such segments.
1907 if ((flags & SVC_ALL) == 0) {
1908 if (elf_legacy_coredump) {
1909 if ((entry->protection & VM_PROT_RW) !=
1913 if ((entry->protection & VM_PROT_ALL) == 0)
1919 * Dont include memory segment in the coredump if
1920 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1921 * madvise(2). Do not dump submaps (i.e. parts of the
1924 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1926 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1927 (flags & SVC_ALL) == 0)
1929 if ((object = entry->object.vm_object) == NULL)
1932 /* Ignore memory-mapped devices and such things. */
1933 VM_OBJECT_RLOCK(object);
1934 while ((backing_object = object->backing_object) != NULL) {
1935 VM_OBJECT_RLOCK(backing_object);
1936 VM_OBJECT_RUNLOCK(object);
1937 object = backing_object;
1939 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1940 VM_OBJECT_RUNLOCK(object);
1944 (*func)(entry, closure);
1946 vm_map_unlock_read(map);
1950 * Write the core file header to the file, including padding up to
1951 * the page boundary.
1954 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1955 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1958 struct note_info *ninfo;
1962 /* Fill in the header. */
1963 bzero(hdr, hdrsize);
1964 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1966 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1967 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1968 sbuf_start_section(sb, NULL);
1969 sbuf_bcat(sb, hdr, hdrsize);
1970 TAILQ_FOREACH(ninfo, notelst, link)
1971 __elfN(putnote)(ninfo, sb);
1972 /* Align up to a page boundary for the program segments. */
1973 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1974 error = sbuf_finish(sb);
1981 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1991 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1994 * To have the debugger select the right thread (LWP) as the initial
1995 * thread, we dump the state of the thread passed to us in td first.
1996 * This is the thread that causes the core dump and thus likely to
1997 * be the right thread one wants to have selected in the debugger.
2000 while (thr != NULL) {
2001 size += register_note(list, NT_PRSTATUS,
2002 __elfN(note_prstatus), thr);
2003 size += register_note(list, NT_FPREGSET,
2004 __elfN(note_fpregset), thr);
2005 size += register_note(list, NT_THRMISC,
2006 __elfN(note_thrmisc), thr);
2007 size += register_note(list, NT_PTLWPINFO,
2008 __elfN(note_ptlwpinfo), thr);
2009 size += register_note(list, -1,
2010 __elfN(note_threadmd), thr);
2012 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
2013 TAILQ_NEXT(thr, td_plist);
2015 thr = TAILQ_NEXT(thr, td_plist);
2018 size += register_note(list, NT_PROCSTAT_PROC,
2019 __elfN(note_procstat_proc), p);
2020 size += register_note(list, NT_PROCSTAT_FILES,
2021 note_procstat_files, p);
2022 size += register_note(list, NT_PROCSTAT_VMMAP,
2023 note_procstat_vmmap, p);
2024 size += register_note(list, NT_PROCSTAT_GROUPS,
2025 note_procstat_groups, p);
2026 size += register_note(list, NT_PROCSTAT_UMASK,
2027 note_procstat_umask, p);
2028 size += register_note(list, NT_PROCSTAT_RLIMIT,
2029 note_procstat_rlimit, p);
2030 size += register_note(list, NT_PROCSTAT_OSREL,
2031 note_procstat_osrel, p);
2032 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
2033 __elfN(note_procstat_psstrings), p);
2034 size += register_note(list, NT_PROCSTAT_AUXV,
2035 __elfN(note_procstat_auxv), p);
2041 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
2042 size_t notesz, int flags)
2047 struct phdr_closure phc;
2049 ehdr = (Elf_Ehdr *)hdr;
2051 ehdr->e_ident[EI_MAG0] = ELFMAG0;
2052 ehdr->e_ident[EI_MAG1] = ELFMAG1;
2053 ehdr->e_ident[EI_MAG2] = ELFMAG2;
2054 ehdr->e_ident[EI_MAG3] = ELFMAG3;
2055 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
2056 ehdr->e_ident[EI_DATA] = ELF_DATA;
2057 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
2058 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2059 ehdr->e_ident[EI_ABIVERSION] = 0;
2060 ehdr->e_ident[EI_PAD] = 0;
2061 ehdr->e_type = ET_CORE;
2062 ehdr->e_machine = td->td_proc->p_elf_machine;
2063 ehdr->e_version = EV_CURRENT;
2065 ehdr->e_phoff = sizeof(Elf_Ehdr);
2066 ehdr->e_flags = td->td_proc->p_elf_flags;
2067 ehdr->e_ehsize = sizeof(Elf_Ehdr);
2068 ehdr->e_phentsize = sizeof(Elf_Phdr);
2069 ehdr->e_shentsize = sizeof(Elf_Shdr);
2070 ehdr->e_shstrndx = SHN_UNDEF;
2071 if (numsegs + 1 < PN_XNUM) {
2072 ehdr->e_phnum = numsegs + 1;
2075 ehdr->e_phnum = PN_XNUM;
2078 ehdr->e_shoff = ehdr->e_phoff +
2079 (numsegs + 1) * ehdr->e_phentsize;
2080 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
2081 ("e_shoff: %zu, hdrsize - shdr: %zu",
2082 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
2084 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
2085 memset(shdr, 0, sizeof(*shdr));
2087 * A special first section is used to hold large segment and
2088 * section counts. This was proposed by Sun Microsystems in
2089 * Solaris and has been adopted by Linux; the standard ELF
2090 * tools are already familiar with the technique.
2092 * See table 7-7 of the Solaris "Linker and Libraries Guide"
2093 * (or 12-7 depending on the version of the document) for more
2096 shdr->sh_type = SHT_NULL;
2097 shdr->sh_size = ehdr->e_shnum;
2098 shdr->sh_link = ehdr->e_shstrndx;
2099 shdr->sh_info = numsegs + 1;
2103 * Fill in the program header entries.
2105 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
2107 /* The note segement. */
2108 phdr->p_type = PT_NOTE;
2109 phdr->p_offset = hdrsize;
2112 phdr->p_filesz = notesz;
2114 phdr->p_flags = PF_R;
2115 phdr->p_align = ELF_NOTE_ROUNDSIZE;
2118 /* All the writable segments from the program. */
2120 phc.offset = round_page(hdrsize + notesz);
2121 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
2125 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
2127 struct note_info *ninfo;
2128 size_t size, notesize;
2131 out(arg, NULL, &size);
2132 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
2134 ninfo->outfunc = out;
2135 ninfo->outarg = arg;
2136 ninfo->outsize = size;
2137 TAILQ_INSERT_TAIL(list, ninfo, link);
2142 notesize = sizeof(Elf_Note) + /* note header */
2143 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2145 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2151 append_note_data(const void *src, void *dst, size_t len)
2155 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2157 bcopy(src, dst, len);
2158 bzero((char *)dst + len, padded_len - len);
2160 return (padded_len);
2164 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2172 note = (Elf_Note *)buf;
2173 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2174 note->n_descsz = size;
2175 note->n_type = type;
2176 buf += sizeof(*note);
2177 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2178 sizeof(FREEBSD_ABI_VENDOR));
2179 append_note_data(src, buf, size);
2184 notesize = sizeof(Elf_Note) + /* note header */
2185 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2187 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2193 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2196 ssize_t old_len, sect_len;
2197 size_t new_len, descsz, i;
2199 if (ninfo->type == -1) {
2200 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2204 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2205 note.n_descsz = ninfo->outsize;
2206 note.n_type = ninfo->type;
2208 sbuf_bcat(sb, ¬e, sizeof(note));
2209 sbuf_start_section(sb, &old_len);
2210 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2211 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2212 if (note.n_descsz == 0)
2214 sbuf_start_section(sb, &old_len);
2215 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2216 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2220 new_len = (size_t)sect_len;
2221 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2222 if (new_len < descsz) {
2224 * It is expected that individual note emitters will correctly
2225 * predict their expected output size and fill up to that size
2226 * themselves, padding in a format-specific way if needed.
2227 * However, in case they don't, just do it here with zeros.
2229 for (i = 0; i < descsz - new_len; i++)
2231 } else if (new_len > descsz) {
2233 * We can't always truncate sb -- we may have drained some
2236 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2237 "read it (%zu > %zu). Since it is longer than "
2238 "expected, this coredump's notes are corrupt. THIS "
2239 "IS A BUG in the note_procstat routine for type %u.\n",
2240 __func__, (unsigned)note.n_type, new_len, descsz,
2241 (unsigned)note.n_type));
2246 * Miscellaneous note out functions.
2249 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2250 #include <compat/freebsd32/freebsd32.h>
2251 #include <compat/freebsd32/freebsd32_signal.h>
2253 typedef struct prstatus32 elf_prstatus_t;
2254 typedef struct prpsinfo32 elf_prpsinfo_t;
2255 typedef struct fpreg32 elf_prfpregset_t;
2256 typedef struct fpreg32 elf_fpregset_t;
2257 typedef struct reg32 elf_gregset_t;
2258 typedef struct thrmisc32 elf_thrmisc_t;
2259 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2260 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2261 typedef uint32_t elf_ps_strings_t;
2263 typedef prstatus_t elf_prstatus_t;
2264 typedef prpsinfo_t elf_prpsinfo_t;
2265 typedef prfpregset_t elf_prfpregset_t;
2266 typedef prfpregset_t elf_fpregset_t;
2267 typedef gregset_t elf_gregset_t;
2268 typedef thrmisc_t elf_thrmisc_t;
2269 #define ELF_KERN_PROC_MASK 0
2270 typedef struct kinfo_proc elf_kinfo_proc_t;
2271 typedef vm_offset_t elf_ps_strings_t;
2275 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2281 elf_prpsinfo_t *psinfo;
2284 p = (struct proc *)arg;
2286 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2287 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2288 psinfo->pr_version = PRPSINFO_VERSION;
2289 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2290 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2292 if (p->p_args != NULL) {
2293 len = sizeof(psinfo->pr_psargs) - 1;
2294 if (len > p->p_args->ar_length)
2295 len = p->p_args->ar_length;
2296 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2302 sbuf_new(&sbarg, psinfo->pr_psargs,
2303 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2304 error = proc_getargv(curthread, p, &sbarg);
2306 if (sbuf_finish(&sbarg) == 0)
2307 len = sbuf_len(&sbarg) - 1;
2309 len = sizeof(psinfo->pr_psargs) - 1;
2310 sbuf_delete(&sbarg);
2312 if (error || len == 0)
2313 strlcpy(psinfo->pr_psargs, p->p_comm,
2314 sizeof(psinfo->pr_psargs));
2316 KASSERT(len < sizeof(psinfo->pr_psargs),
2317 ("len is too long: %zu vs %zu", len,
2318 sizeof(psinfo->pr_psargs)));
2319 cp = psinfo->pr_psargs;
2322 cp = memchr(cp, '\0', end - cp);
2328 psinfo->pr_pid = p->p_pid;
2329 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2330 free(psinfo, M_TEMP);
2332 *sizep = sizeof(*psinfo);
2336 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2339 elf_prstatus_t *status;
2341 td = (struct thread *)arg;
2343 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2344 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2345 status->pr_version = PRSTATUS_VERSION;
2346 status->pr_statussz = sizeof(elf_prstatus_t);
2347 status->pr_gregsetsz = sizeof(elf_gregset_t);
2348 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2349 status->pr_osreldate = osreldate;
2350 status->pr_cursig = td->td_proc->p_sig;
2351 status->pr_pid = td->td_tid;
2352 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2353 fill_regs32(td, &status->pr_reg);
2355 fill_regs(td, &status->pr_reg);
2357 sbuf_bcat(sb, status, sizeof(*status));
2358 free(status, M_TEMP);
2360 *sizep = sizeof(*status);
2364 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2367 elf_prfpregset_t *fpregset;
2369 td = (struct thread *)arg;
2371 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2372 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2373 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2374 fill_fpregs32(td, fpregset);
2376 fill_fpregs(td, fpregset);
2378 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2379 free(fpregset, M_TEMP);
2381 *sizep = sizeof(*fpregset);
2385 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2388 elf_thrmisc_t thrmisc;
2390 td = (struct thread *)arg;
2392 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2393 bzero(&thrmisc, sizeof(thrmisc));
2394 strcpy(thrmisc.pr_tname, td->td_name);
2395 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2397 *sizep = sizeof(thrmisc);
2401 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2406 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2407 struct ptrace_lwpinfo32 pl;
2409 struct ptrace_lwpinfo pl;
2412 td = (struct thread *)arg;
2413 size = sizeof(structsize) + sizeof(pl);
2415 KASSERT(*sizep == size, ("invalid size"));
2416 structsize = sizeof(pl);
2417 sbuf_bcat(sb, &structsize, sizeof(structsize));
2418 bzero(&pl, sizeof(pl));
2419 pl.pl_lwpid = td->td_tid;
2420 pl.pl_event = PL_EVENT_NONE;
2421 pl.pl_sigmask = td->td_sigmask;
2422 pl.pl_siglist = td->td_siglist;
2423 if (td->td_si.si_signo != 0) {
2424 pl.pl_event = PL_EVENT_SIGNAL;
2425 pl.pl_flags |= PL_FLAG_SI;
2426 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2427 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2429 pl.pl_siginfo = td->td_si;
2432 strcpy(pl.pl_tdname, td->td_name);
2433 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2434 sbuf_bcat(sb, &pl, sizeof(pl));
2440 * Allow for MD specific notes, as well as any MD
2441 * specific preparations for writing MI notes.
2444 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2450 td = (struct thread *)arg;
2452 if (size != 0 && sb != NULL)
2453 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2457 __elfN(dump_thread)(td, buf, &size);
2458 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2459 if (size != 0 && sb != NULL)
2460 sbuf_bcat(sb, buf, size);
2465 #ifdef KINFO_PROC_SIZE
2466 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2470 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2476 p = (struct proc *)arg;
2477 size = sizeof(structsize) + p->p_numthreads *
2478 sizeof(elf_kinfo_proc_t);
2481 KASSERT(*sizep == size, ("invalid size"));
2482 structsize = sizeof(elf_kinfo_proc_t);
2483 sbuf_bcat(sb, &structsize, sizeof(structsize));
2484 sx_slock(&proctree_lock);
2486 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2487 sx_sunlock(&proctree_lock);
2492 #ifdef KINFO_FILE_SIZE
2493 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2497 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2500 size_t size, sect_sz, i;
2501 ssize_t start_len, sect_len;
2502 int structsize, filedesc_flags;
2504 if (coredump_pack_fileinfo)
2505 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2509 p = (struct proc *)arg;
2510 structsize = sizeof(struct kinfo_file);
2513 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2514 sbuf_set_drain(sb, sbuf_count_drain, &size);
2515 sbuf_bcat(sb, &structsize, sizeof(structsize));
2517 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2522 sbuf_start_section(sb, &start_len);
2524 sbuf_bcat(sb, &structsize, sizeof(structsize));
2526 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2529 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2534 KASSERT(sect_sz <= *sizep,
2535 ("kern_proc_filedesc_out did not respect maxlen; "
2536 "requested %zu, got %zu", *sizep - sizeof(structsize),
2537 sect_sz - sizeof(structsize)));
2539 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2544 #ifdef KINFO_VMENTRY_SIZE
2545 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2549 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2553 int structsize, vmmap_flags;
2555 if (coredump_pack_vmmapinfo)
2556 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2560 p = (struct proc *)arg;
2561 structsize = sizeof(struct kinfo_vmentry);
2564 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2565 sbuf_set_drain(sb, sbuf_count_drain, &size);
2566 sbuf_bcat(sb, &structsize, sizeof(structsize));
2568 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2573 sbuf_bcat(sb, &structsize, sizeof(structsize));
2575 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2581 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2587 p = (struct proc *)arg;
2588 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2590 KASSERT(*sizep == size, ("invalid size"));
2591 structsize = sizeof(gid_t);
2592 sbuf_bcat(sb, &structsize, sizeof(structsize));
2593 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2600 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2606 p = (struct proc *)arg;
2607 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2609 KASSERT(*sizep == size, ("invalid size"));
2610 structsize = sizeof(p->p_pd->pd_cmask);
2611 sbuf_bcat(sb, &structsize, sizeof(structsize));
2612 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2618 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2621 struct rlimit rlim[RLIM_NLIMITS];
2625 p = (struct proc *)arg;
2626 size = sizeof(structsize) + sizeof(rlim);
2628 KASSERT(*sizep == size, ("invalid size"));
2629 structsize = sizeof(rlim);
2630 sbuf_bcat(sb, &structsize, sizeof(structsize));
2632 for (i = 0; i < RLIM_NLIMITS; i++)
2633 lim_rlimit_proc(p, i, &rlim[i]);
2635 sbuf_bcat(sb, rlim, sizeof(rlim));
2641 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2647 p = (struct proc *)arg;
2648 size = sizeof(structsize) + sizeof(p->p_osrel);
2650 KASSERT(*sizep == size, ("invalid size"));
2651 structsize = sizeof(p->p_osrel);
2652 sbuf_bcat(sb, &structsize, sizeof(structsize));
2653 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2659 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2662 elf_ps_strings_t ps_strings;
2666 p = (struct proc *)arg;
2667 size = sizeof(structsize) + sizeof(ps_strings);
2669 KASSERT(*sizep == size, ("invalid size"));
2670 structsize = sizeof(ps_strings);
2671 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2672 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2674 ps_strings = PROC_PS_STRINGS(p);
2676 sbuf_bcat(sb, &structsize, sizeof(structsize));
2677 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2683 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2689 p = (struct proc *)arg;
2692 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2693 sbuf_set_drain(sb, sbuf_count_drain, &size);
2694 sbuf_bcat(sb, &structsize, sizeof(structsize));
2696 proc_getauxv(curthread, p, sb);
2702 structsize = sizeof(Elf_Auxinfo);
2703 sbuf_bcat(sb, &structsize, sizeof(structsize));
2705 proc_getauxv(curthread, p, sb);
2711 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2712 const char *note_vendor, const Elf_Phdr *pnote,
2713 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2715 const Elf_Note *note, *note0, *note_end;
2716 const char *note_name;
2721 /* We need some limit, might as well use PAGE_SIZE. */
2722 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2724 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2725 if (pnote->p_offset > PAGE_SIZE ||
2726 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2727 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2729 VOP_UNLOCK(imgp->vp);
2730 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2731 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2733 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2734 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2735 curthread->td_ucred, NOCRED, NULL, curthread);
2737 uprintf("i/o error PT_NOTE\n");
2740 note = note0 = (const Elf_Note *)buf;
2741 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2743 note = note0 = (const Elf_Note *)(imgp->image_header +
2745 note_end = (const Elf_Note *)(imgp->image_header +
2746 pnote->p_offset + pnote->p_filesz);
2749 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2750 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2751 (const char *)note < sizeof(Elf_Note)) {
2754 if (note->n_namesz != checknote->n_namesz ||
2755 note->n_descsz != checknote->n_descsz ||
2756 note->n_type != checknote->n_type)
2758 note_name = (const char *)(note + 1);
2759 if (note_name + checknote->n_namesz >=
2760 (const char *)note_end || strncmp(note_vendor,
2761 note_name, checknote->n_namesz) != 0)
2764 if (cb(note, cb_arg, &res))
2767 note = (const Elf_Note *)((const char *)(note + 1) +
2768 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2769 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2778 struct brandnote_cb_arg {
2779 Elf_Brandnote *brandnote;
2784 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2786 struct brandnote_cb_arg *arg;
2791 * Fetch the osreldate for binary from the ELF OSABI-note if
2794 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2795 arg->brandnote->trans_osrel != NULL ?
2796 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2801 static Elf_Note fctl_note = {
2802 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2803 .n_descsz = sizeof(uint32_t),
2804 .n_type = NT_FREEBSD_FEATURE_CTL,
2807 struct fctl_cb_arg {
2808 boolean_t *has_fctl0;
2813 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2815 struct fctl_cb_arg *arg;
2816 const Elf32_Word *desc;
2820 p = (uintptr_t)(note + 1);
2821 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2822 desc = (const Elf32_Word *)p;
2823 *arg->has_fctl0 = TRUE;
2824 *arg->fctl0 = desc[0];
2830 * Try to find the appropriate ABI-note section for checknote, fetch
2831 * the osreldate and feature control flags for binary from the ELF
2832 * OSABI-note. Only the first page of the image is searched, the same
2836 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2837 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2839 const Elf_Phdr *phdr;
2840 const Elf_Ehdr *hdr;
2841 struct brandnote_cb_arg b_arg;
2842 struct fctl_cb_arg f_arg;
2845 hdr = (const Elf_Ehdr *)imgp->image_header;
2846 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2847 b_arg.brandnote = brandnote;
2848 b_arg.osrel = osrel;
2849 f_arg.has_fctl0 = has_fctl0;
2850 f_arg.fctl0 = fctl0;
2852 for (i = 0; i < hdr->e_phnum; i++) {
2853 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2854 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2856 for (j = 0; j < hdr->e_phnum; j++) {
2857 if (phdr[j].p_type == PT_NOTE &&
2858 __elfN(parse_notes)(imgp, &fctl_note,
2859 FREEBSD_ABI_VENDOR, &phdr[j],
2860 note_fctl_cb, &f_arg))
2871 * Tell kern_execve.c about it, with a little help from the linker.
2873 static struct execsw __elfN(execsw) = {
2874 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2875 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2877 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2880 __elfN(trans_prot)(Elf_Word flags)
2886 prot |= VM_PROT_EXECUTE;
2888 prot |= VM_PROT_WRITE;
2890 prot |= VM_PROT_READ;
2891 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2892 if (i386_read_exec && (flags & PF_R))
2893 prot |= VM_PROT_EXECUTE;
2899 __elfN(untrans_prot)(vm_prot_t prot)
2904 if (prot & VM_PROT_EXECUTE)
2906 if (prot & VM_PROT_READ)
2908 if (prot & VM_PROT_WRITE)