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 boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, boolean_t *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)
173 static int __elfN(aslr_enabled) = 0;
174 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
175 &__elfN(aslr_enabled), 0,
176 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
177 ": enable address map randomization");
179 static int __elfN(pie_aslr_enabled) = 0;
180 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
181 &__elfN(pie_aslr_enabled), 0,
182 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
183 ": enable address map randomization for PIE binaries");
185 static int __elfN(aslr_honor_sbrk) = 1;
186 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
187 &__elfN(aslr_honor_sbrk), 0,
188 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
190 static int __elfN(aslr_stack) = 1;
191 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack, CTLFLAG_RWTUN,
192 &__elfN(aslr_stack), 0,
193 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
194 ": enable stack address randomization");
196 static int __elfN(sigfastblock) = 1;
197 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
198 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
199 "enable sigfastblock for new processes");
201 static bool __elfN(allow_wx) = true;
202 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
203 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
204 "Allow pages to be mapped simultaneously writable and executable");
206 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
208 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
210 Elf_Brandnote __elfN(freebsd_brandnote) = {
211 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
212 .hdr.n_descsz = sizeof(int32_t),
213 .hdr.n_type = NT_FREEBSD_ABI_TAG,
214 .vendor = FREEBSD_ABI_VENDOR,
215 .flags = BN_TRANSLATE_OSREL,
216 .trans_osrel = __elfN(freebsd_trans_osrel)
220 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
224 p = (uintptr_t)(note + 1);
225 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
226 *osrel = *(const int32_t *)(p);
231 static const char GNU_ABI_VENDOR[] = "GNU";
232 static int GNU_KFREEBSD_ABI_DESC = 3;
234 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
235 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
236 .hdr.n_descsz = 16, /* XXX at least 16 */
238 .vendor = GNU_ABI_VENDOR,
239 .flags = BN_TRANSLATE_OSREL,
240 .trans_osrel = kfreebsd_trans_osrel
244 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
246 const Elf32_Word *desc;
249 p = (uintptr_t)(note + 1);
250 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
252 desc = (const Elf32_Word *)p;
253 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
257 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
258 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
260 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
266 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
270 for (i = 0; i < MAX_BRANDS; i++) {
271 if (elf_brand_list[i] == NULL) {
272 elf_brand_list[i] = entry;
276 if (i == MAX_BRANDS) {
277 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
285 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
289 for (i = 0; i < MAX_BRANDS; i++) {
290 if (elf_brand_list[i] == entry) {
291 elf_brand_list[i] = NULL;
301 __elfN(brand_inuse)(Elf_Brandinfo *entry)
306 sx_slock(&allproc_lock);
307 FOREACH_PROC_IN_SYSTEM(p) {
308 if (p->p_sysent == entry->sysvec) {
313 sx_sunlock(&allproc_lock);
318 static Elf_Brandinfo *
319 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
320 int32_t *osrel, uint32_t *fctl0)
322 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
323 Elf_Brandinfo *bi, *bi_m;
324 boolean_t ret, has_fctl0;
325 int i, interp_name_len;
327 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
330 * We support four types of branding -- (1) the ELF EI_OSABI field
331 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
332 * branding w/in the ELF header, (3) path of the `interp_path'
333 * field, and (4) the ".note.ABI-tag" ELF section.
336 /* Look for an ".note.ABI-tag" ELF section */
338 for (i = 0; i < MAX_BRANDS; i++) {
339 bi = elf_brand_list[i];
342 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
344 if (hdr->e_machine == bi->machine && (bi->flags &
345 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
349 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
351 /* Give brand a chance to veto check_note's guess */
352 if (ret && bi->header_supported) {
353 ret = bi->header_supported(imgp, osrel,
354 has_fctl0 ? fctl0 : NULL);
357 * If note checker claimed the binary, but the
358 * interpreter path in the image does not
359 * match default one for the brand, try to
360 * search for other brands with the same
361 * interpreter. Either there is better brand
362 * with the right interpreter, or, failing
363 * this, we return first brand which accepted
364 * our note and, optionally, header.
366 if (ret && bi_m == NULL && interp != NULL &&
367 (bi->interp_path == NULL ||
368 (strlen(bi->interp_path) + 1 != interp_name_len ||
369 strncmp(interp, bi->interp_path, interp_name_len)
381 /* If the executable has a brand, search for it in the brand list. */
382 for (i = 0; i < MAX_BRANDS; i++) {
383 bi = elf_brand_list[i];
384 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
385 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
387 if (hdr->e_machine == bi->machine &&
388 (hdr->e_ident[EI_OSABI] == bi->brand ||
389 (bi->compat_3_brand != NULL &&
390 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
391 bi->compat_3_brand) == 0))) {
392 /* Looks good, but give brand a chance to veto */
393 if (bi->header_supported == NULL ||
394 bi->header_supported(imgp, NULL, NULL)) {
396 * Again, prefer strictly matching
399 if (interp_name_len == 0 &&
400 bi->interp_path == NULL)
402 if (bi->interp_path != NULL &&
403 strlen(bi->interp_path) + 1 ==
404 interp_name_len && strncmp(interp,
405 bi->interp_path, interp_name_len) == 0)
415 /* No known brand, see if the header is recognized by any brand */
416 for (i = 0; i < MAX_BRANDS; i++) {
417 bi = elf_brand_list[i];
418 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
419 bi->header_supported == NULL)
421 if (hdr->e_machine == bi->machine) {
422 ret = bi->header_supported(imgp, NULL, NULL);
428 /* Lacking a known brand, search for a recognized interpreter. */
429 if (interp != NULL) {
430 for (i = 0; i < MAX_BRANDS; i++) {
431 bi = elf_brand_list[i];
432 if (bi == NULL || (bi->flags &
433 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
436 if (hdr->e_machine == bi->machine &&
437 bi->interp_path != NULL &&
438 /* ELF image p_filesz includes terminating zero */
439 strlen(bi->interp_path) + 1 == interp_name_len &&
440 strncmp(interp, bi->interp_path, interp_name_len)
441 == 0 && (bi->header_supported == NULL ||
442 bi->header_supported(imgp, NULL, NULL)))
447 /* Lacking a recognized interpreter, try the default brand */
448 for (i = 0; i < MAX_BRANDS; i++) {
449 bi = elf_brand_list[i];
450 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
451 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
453 if (hdr->e_machine == bi->machine &&
454 __elfN(fallback_brand) == bi->brand &&
455 (bi->header_supported == NULL ||
456 bi->header_supported(imgp, NULL, NULL)))
463 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
465 return (hdr->e_phoff <= PAGE_SIZE &&
466 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
470 __elfN(check_header)(const Elf_Ehdr *hdr)
476 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
477 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
478 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
479 hdr->e_phentsize != sizeof(Elf_Phdr) ||
480 hdr->e_version != ELF_TARG_VER)
484 * Make sure we have at least one brand for this machine.
487 for (i = 0; i < MAX_BRANDS; i++) {
488 bi = elf_brand_list[i];
489 if (bi != NULL && bi->machine == hdr->e_machine)
499 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
500 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
507 * Create the page if it doesn't exist yet. Ignore errors.
509 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
510 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
513 * Find the page from the underlying object.
515 if (object != NULL) {
516 sf = vm_imgact_map_page(object, offset);
518 return (KERN_FAILURE);
519 off = offset - trunc_page(offset);
520 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
522 vm_imgact_unmap_page(sf);
524 return (KERN_FAILURE);
527 return (KERN_SUCCESS);
531 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
532 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
538 int error, locked, rv;
540 if (start != trunc_page(start)) {
541 rv = __elfN(map_partial)(map, object, offset, start,
542 round_page(start), prot);
543 if (rv != KERN_SUCCESS)
545 offset += round_page(start) - start;
546 start = round_page(start);
548 if (end != round_page(end)) {
549 rv = __elfN(map_partial)(map, object, offset +
550 trunc_page(end) - start, trunc_page(end), end, prot);
551 if (rv != KERN_SUCCESS)
553 end = trunc_page(end);
556 return (KERN_SUCCESS);
557 if ((offset & PAGE_MASK) != 0) {
559 * The mapping is not page aligned. This means that we have
562 rv = vm_map_fixed(map, NULL, 0, start, end - start,
563 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
564 if (rv != KERN_SUCCESS)
567 return (KERN_SUCCESS);
568 for (; start < end; start += sz) {
569 sf = vm_imgact_map_page(object, offset);
571 return (KERN_FAILURE);
572 off = offset - trunc_page(offset);
574 if (sz > PAGE_SIZE - off)
575 sz = PAGE_SIZE - off;
576 error = copyout((caddr_t)sf_buf_kva(sf) + off,
578 vm_imgact_unmap_page(sf);
580 return (KERN_FAILURE);
584 vm_object_reference(object);
585 rv = vm_map_fixed(map, object, offset, start, end - start,
586 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
587 (object != NULL ? MAP_VN_EXEC : 0));
588 if (rv != KERN_SUCCESS) {
589 locked = VOP_ISLOCKED(imgp->vp);
590 VOP_UNLOCK(imgp->vp);
591 vm_object_deallocate(object);
592 vn_lock(imgp->vp, locked | LK_RETRY);
594 } else if (object != NULL) {
595 MPASS(imgp->vp->v_object == object);
596 VOP_SET_TEXT_CHECKED(imgp->vp);
599 return (KERN_SUCCESS);
603 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
604 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
610 vm_offset_t map_addr;
613 vm_ooffset_t file_addr;
616 * It's necessary to fail if the filsz + offset taken from the
617 * header is greater than the actual file pager object's size.
618 * If we were to allow this, then the vm_map_find() below would
619 * walk right off the end of the file object and into the ether.
621 * While I'm here, might as well check for something else that
622 * is invalid: filsz cannot be greater than memsz.
624 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
626 uprintf("elf_load_section: truncated ELF file\n");
630 object = imgp->object;
631 map = &imgp->proc->p_vmspace->vm_map;
632 map_addr = trunc_page((vm_offset_t)vmaddr);
633 file_addr = trunc_page(offset);
636 * We have two choices. We can either clear the data in the last page
637 * of an oversized mapping, or we can start the anon mapping a page
638 * early and copy the initialized data into that first page. We
643 else if (memsz > filsz)
644 map_len = trunc_page(offset + filsz) - file_addr;
646 map_len = round_page(offset + filsz) - file_addr;
649 /* cow flags: don't dump readonly sections in core */
650 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
651 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
653 rv = __elfN(map_insert)(imgp, map, object, file_addr,
654 map_addr, map_addr + map_len, prot, cow);
655 if (rv != KERN_SUCCESS)
658 /* we can stop now if we've covered it all */
664 * We have to get the remaining bit of the file into the first part
665 * of the oversized map segment. This is normally because the .data
666 * segment in the file is extended to provide bss. It's a neat idea
667 * to try and save a page, but it's a pain in the behind to implement.
669 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
671 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
672 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
674 /* This had damn well better be true! */
676 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
677 map_addr + map_len, prot, 0);
678 if (rv != KERN_SUCCESS)
683 sf = vm_imgact_map_page(object, offset + filsz);
687 /* send the page fragment to user space */
688 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
690 vm_imgact_unmap_page(sf);
696 * Remove write access to the page if it was only granted by map_insert
699 if ((prot & VM_PROT_WRITE) == 0)
700 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
701 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
707 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
708 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
715 ASSERT_VOP_LOCKED(imgp->vp, __func__);
720 for (i = 0; i < hdr->e_phnum; i++) {
721 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
724 /* Loadable segment */
725 prot = __elfN(trans_prot)(phdr[i].p_flags);
726 error = __elfN(load_section)(imgp, phdr[i].p_offset,
727 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
728 phdr[i].p_memsz, phdr[i].p_filesz, prot);
733 * Establish the base address if this is the first segment.
736 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
741 if (base_addrp != NULL)
742 *base_addrp = base_addr;
748 * Load the file "file" into memory. It may be either a shared object
751 * The "addr" reference parameter is in/out. On entry, it specifies
752 * the address where a shared object should be loaded. If the file is
753 * an executable, this value is ignored. On exit, "addr" specifies
754 * where the file was actually loaded.
756 * The "entry" reference parameter is out only. On exit, it specifies
757 * the entry point for the loaded file.
760 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
766 struct image_params image_params;
768 const Elf_Ehdr *hdr = NULL;
769 const Elf_Phdr *phdr = NULL;
770 struct nameidata *nd;
772 struct image_params *imgp;
774 u_long base_addr = 0;
777 #ifdef CAPABILITY_MODE
779 * XXXJA: This check can go away once we are sufficiently confident
780 * that the checks in namei() are correct.
782 if (IN_CAPABILITY_MODE(curthread))
786 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
788 attr = &tempdata->attr;
789 imgp = &tempdata->image_params;
792 * Initialize part of the common data
797 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
798 UIO_SYSSPACE, file, curthread);
799 if ((error = namei(nd)) != 0) {
803 NDFREE(nd, NDF_ONLY_PNBUF);
804 imgp->vp = nd->ni_vp;
807 * Check permissions, modes, uid, etc on the file, and "open" it.
809 error = exec_check_permissions(imgp);
813 error = exec_map_first_page(imgp);
817 imgp->object = nd->ni_vp->v_object;
819 hdr = (const Elf_Ehdr *)imgp->image_header;
820 if ((error = __elfN(check_header)(hdr)) != 0)
822 if (hdr->e_type == ET_DYN)
824 else if (hdr->e_type == ET_EXEC)
831 /* Only support headers that fit within first page for now */
832 if (!__elfN(phdr_in_zero_page)(hdr)) {
837 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
838 if (!aligned(phdr, Elf_Addr)) {
843 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
848 *entry = (unsigned long)hdr->e_entry + rbase;
852 exec_unmap_first_page(imgp);
856 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
859 free(tempdata, M_TEMP);
865 * Select randomized valid address in the map map, between minv and
866 * maxv, with specified alignment. The [minv, maxv) range must belong
867 * to the map. Note that function only allocates the address, it is
868 * up to caller to clamp maxv in a way that the final allocation
869 * length fit into the map.
871 * Result is returned in *resp, error code indicates that arguments
872 * did not pass sanity checks for overflow and range correctness.
875 __CONCAT(rnd_, __elfN(base))(vm_map_t map, u_long minv, u_long maxv,
876 u_int align, u_long *resp)
880 MPASS(vm_map_min(map) <= minv);
882 if (minv >= maxv || minv + align >= maxv || maxv > vm_map_max(map)) {
883 uprintf("Invalid ELF segments layout\n");
887 arc4rand(&rbase, sizeof(rbase), 0);
888 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
889 res &= ~((u_long)align - 1);
894 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
895 res, minv, maxv, rbase));
897 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
898 res, maxv, minv, rbase));
905 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
906 const Elf_Phdr *phdr, u_long et_dyn_addr)
908 struct vmspace *vmspace;
910 u_long text_size, data_size, total_size, text_addr, data_addr;
911 u_long seg_size, seg_addr;
915 text_size = data_size = total_size = text_addr = data_addr = 0;
917 for (i = 0; i < hdr->e_phnum; i++) {
918 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
921 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
922 seg_size = round_page(phdr[i].p_memsz +
923 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
926 * Make the largest executable segment the official
927 * text segment and all others data.
929 * Note that obreak() assumes that data_addr + data_size == end
930 * of data load area, and the ELF file format expects segments
931 * to be sorted by address. If multiple data segments exist,
932 * the last one will be used.
935 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
936 text_size = seg_size;
937 text_addr = seg_addr;
939 data_size = seg_size;
940 data_addr = seg_addr;
942 total_size += seg_size;
945 if (data_addr == 0 && data_size == 0) {
946 data_addr = text_addr;
947 data_size = text_size;
951 * Check limits. It should be safe to check the
952 * limits after loading the segments since we do
953 * not actually fault in all the segments pages.
955 PROC_LOCK(imgp->proc);
956 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
957 err_str = "Data segment size exceeds process limit";
958 else if (text_size > maxtsiz)
959 err_str = "Text segment size exceeds system limit";
960 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
961 err_str = "Total segment size exceeds process limit";
962 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
963 err_str = "Data segment size exceeds resource limit";
964 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
965 err_str = "Total segment size exceeds resource limit";
966 PROC_UNLOCK(imgp->proc);
967 if (err_str != NULL) {
968 uprintf("%s\n", err_str);
972 vmspace = imgp->proc->p_vmspace;
973 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
974 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
975 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
976 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
982 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
983 char **interpp, bool *free_interpp)
987 int error, interp_name_len;
989 KASSERT(phdr->p_type == PT_INTERP,
990 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
991 ASSERT_VOP_LOCKED(imgp->vp, __func__);
995 /* Path to interpreter */
996 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
997 uprintf("Invalid PT_INTERP\n");
1001 interp_name_len = phdr->p_filesz;
1002 if (phdr->p_offset > PAGE_SIZE ||
1003 interp_name_len > PAGE_SIZE - phdr->p_offset) {
1005 * The vnode lock might be needed by the pagedaemon to
1006 * clean pages owned by the vnode. Do not allow sleep
1007 * waiting for memory with the vnode locked, instead
1008 * try non-sleepable allocation first, and if it
1009 * fails, go to the slow path were we drop the lock
1010 * and do M_WAITOK. A text reference prevents
1011 * modifications to the vnode content.
1013 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
1014 if (interp == NULL) {
1015 VOP_UNLOCK(imgp->vp);
1016 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
1017 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1020 error = vn_rdwr(UIO_READ, imgp->vp, interp,
1021 interp_name_len, phdr->p_offset,
1022 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1025 free(interp, M_TEMP);
1026 uprintf("i/o error PT_INTERP %d\n", error);
1029 interp[interp_name_len] = '\0';
1032 *free_interpp = true;
1036 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1037 if (interp[interp_name_len - 1] != '\0') {
1038 uprintf("Invalid PT_INTERP\n");
1043 *free_interpp = false;
1048 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1049 const char *interp, u_long *addr, u_long *entry)
1054 if (brand_info->emul_path != NULL &&
1055 brand_info->emul_path[0] != '\0') {
1056 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1057 snprintf(path, MAXPATHLEN, "%s%s",
1058 brand_info->emul_path, interp);
1059 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1065 if (brand_info->interp_newpath != NULL &&
1066 (brand_info->interp_path == NULL ||
1067 strcmp(interp, brand_info->interp_path) == 0)) {
1068 error = __elfN(load_file)(imgp->proc,
1069 brand_info->interp_newpath, addr, entry);
1074 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1078 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1083 * Impossible et_dyn_addr initial value indicating that the real base
1084 * must be calculated later with some randomization applied.
1086 #define ET_DYN_ADDR_RAND 1
1089 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1092 const Elf_Ehdr *hdr;
1093 const Elf_Phdr *phdr;
1094 Elf_Auxargs *elf_auxargs;
1095 struct vmspace *vmspace;
1098 Elf_Brandinfo *brand_info;
1099 struct sysentvec *sv;
1100 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1101 u_long maxalign, maxsalign, mapsz, maxv, maxv1, anon_loc;
1107 hdr = (const Elf_Ehdr *)imgp->image_header;
1110 * Do we have a valid ELF header ?
1112 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1113 * if particular brand doesn't support it.
1115 if (__elfN(check_header)(hdr) != 0 ||
1116 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1120 * From here on down, we return an errno, not -1, as we've
1121 * detected an ELF file.
1124 if (!__elfN(phdr_in_zero_page)(hdr)) {
1125 uprintf("Program headers not in the first page\n");
1128 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1129 if (!aligned(phdr, Elf_Addr)) {
1130 uprintf("Unaligned program headers\n");
1138 entry = proghdr = 0;
1140 free_interp = false;
1144 * Somewhat arbitrary, limit accepted max alignment for the
1145 * loadable segment to the max supported superpage size. Too
1146 * large alignment requests are not useful and are indicators
1147 * of corrupted or outright malicious binary.
1149 maxalign = PAGE_SIZE;
1150 maxsalign = PAGE_SIZE * 1024;
1151 for (i = MAXPAGESIZES - 1; i > 0; i--) {
1152 if (pagesizes[i] > maxsalign)
1153 maxsalign = pagesizes[i];
1158 for (i = 0; i < hdr->e_phnum; i++) {
1159 switch (phdr[i].p_type) {
1162 baddr = phdr[i].p_vaddr;
1163 if (!powerof2(phdr[i].p_align) ||
1164 phdr[i].p_align > maxsalign) {
1165 uprintf("Invalid segment alignment\n");
1169 if (phdr[i].p_align > maxalign)
1170 maxalign = phdr[i].p_align;
1171 if (mapsz + phdr[i].p_memsz < mapsz) {
1172 uprintf("Mapsize overflow\n");
1176 mapsz += phdr[i].p_memsz;
1180 * If this segment contains the program headers,
1181 * remember their virtual address for the AT_PHDR
1182 * aux entry. Static binaries don't usually include
1185 if (phdr[i].p_offset == 0 &&
1186 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize <=
1188 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1191 /* Path to interpreter */
1192 if (interp != NULL) {
1193 uprintf("Multiple PT_INTERP headers\n");
1197 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1203 if (__elfN(nxstack))
1205 __elfN(trans_prot)(phdr[i].p_flags);
1206 imgp->stack_sz = phdr[i].p_memsz;
1208 case PT_PHDR: /* Program header table info */
1209 proghdr = phdr[i].p_vaddr;
1214 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1215 if (brand_info == NULL) {
1216 uprintf("ELF binary type \"%u\" not known.\n",
1217 hdr->e_ident[EI_OSABI]);
1221 sv = brand_info->sysvec;
1223 if (hdr->e_type == ET_DYN) {
1224 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1225 uprintf("Cannot execute shared object\n");
1230 * Honour the base load address from the dso if it is
1231 * non-zero for some reason.
1234 if ((sv->sv_flags & SV_ASLR) == 0 ||
1235 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1236 et_dyn_addr = __elfN(pie_base);
1237 else if ((__elfN(pie_aslr_enabled) &&
1238 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1239 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1240 et_dyn_addr = ET_DYN_ADDR_RAND;
1242 et_dyn_addr = __elfN(pie_base);
1247 * Avoid a possible deadlock if the current address space is destroyed
1248 * and that address space maps the locked vnode. In the common case,
1249 * the locked vnode's v_usecount is decremented but remains greater
1250 * than zero. Consequently, the vnode lock is not needed by vrele().
1251 * However, in cases where the vnode lock is external, such as nullfs,
1252 * v_usecount may become zero.
1254 * The VV_TEXT flag prevents modifications to the executable while
1255 * the vnode is unlocked.
1257 VOP_UNLOCK(imgp->vp);
1260 * Decide whether to enable randomization of user mappings.
1261 * First, reset user preferences for the setid binaries.
1262 * Then, account for the support of the randomization by the
1263 * ABI, by user preferences, and make special treatment for
1266 if (imgp->credential_setid) {
1267 PROC_LOCK(imgp->proc);
1268 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1269 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1270 PROC_UNLOCK(imgp->proc);
1272 if ((sv->sv_flags & SV_ASLR) == 0 ||
1273 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1274 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1275 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1276 ("et_dyn_addr == RAND and !ASLR"));
1277 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1278 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1279 et_dyn_addr == ET_DYN_ADDR_RAND) {
1280 imgp->map_flags |= MAP_ASLR;
1282 * If user does not care about sbrk, utilize the bss
1283 * grow region for mappings as well. We can select
1284 * the base for the image anywere and still not suffer
1285 * from the fragmentation.
1287 if (!__elfN(aslr_honor_sbrk) ||
1288 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1289 imgp->map_flags |= MAP_ASLR_IGNSTART;
1290 if (__elfN(aslr_stack))
1291 imgp->map_flags |= MAP_ASLR_STACK;
1294 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1295 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1296 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1297 imgp->map_flags |= MAP_WXORX;
1299 error = exec_new_vmspace(imgp, sv);
1301 imgp->proc->p_sysent = sv;
1303 vmspace = imgp->proc->p_vmspace;
1304 map = &vmspace->vm_map;
1305 maxv = sv->sv_usrstack;
1306 if ((imgp->map_flags & MAP_ASLR_STACK) == 0)
1307 maxv -= lim_max(td, RLIMIT_STACK);
1308 if (error == 0 && mapsz >= maxv - vm_map_min(map)) {
1309 uprintf("Excessive mapping size\n");
1313 if (error == 0 && et_dyn_addr == ET_DYN_ADDR_RAND) {
1314 KASSERT((map->flags & MAP_ASLR) != 0,
1315 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1316 error = __CONCAT(rnd_, __elfN(base))(map,
1317 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1318 /* reserve half of the address space to interpreter */
1319 maxv / 2, maxalign, &et_dyn_addr);
1322 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1326 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1330 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1335 * We load the dynamic linker where a userland call
1336 * to mmap(0, ...) would put it. The rationale behind this
1337 * calculation is that it leaves room for the heap to grow to
1338 * its maximum allowed size.
1340 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1342 if ((map->flags & MAP_ASLR) != 0) {
1343 maxv1 = maxv / 2 + addr / 2;
1344 error = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1345 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1346 pagesizes[1] : pagesizes[0], &anon_loc);
1349 map->anon_loc = anon_loc;
1351 map->anon_loc = addr;
1354 entry = (u_long)hdr->e_entry + et_dyn_addr;
1355 imgp->entry_addr = entry;
1357 if (interp != NULL) {
1358 VOP_UNLOCK(imgp->vp);
1359 if ((map->flags & MAP_ASLR) != 0) {
1360 /* Assume that interpreter fits into 1/4 of AS */
1361 maxv1 = maxv / 2 + addr / 2;
1362 error = __CONCAT(rnd_, __elfN(base))(map, addr,
1363 maxv1, PAGE_SIZE, &addr);
1366 error = __elfN(load_interp)(imgp, brand_info, interp,
1367 &addr, &imgp->entry_addr);
1369 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1375 error = exec_map_stack(imgp);
1380 * Construct auxargs table (used by the copyout_auxargs routine)
1382 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1383 if (elf_auxargs == NULL) {
1384 VOP_UNLOCK(imgp->vp);
1385 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1386 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1388 elf_auxargs->execfd = -1;
1389 elf_auxargs->phdr = proghdr + et_dyn_addr;
1390 elf_auxargs->phent = hdr->e_phentsize;
1391 elf_auxargs->phnum = hdr->e_phnum;
1392 elf_auxargs->pagesz = PAGE_SIZE;
1393 elf_auxargs->base = addr;
1394 elf_auxargs->flags = 0;
1395 elf_auxargs->entry = entry;
1396 elf_auxargs->hdr_eflags = hdr->e_flags;
1398 imgp->auxargs = elf_auxargs;
1399 imgp->interpreted = 0;
1400 imgp->reloc_base = addr;
1401 imgp->proc->p_osrel = osrel;
1402 imgp->proc->p_fctl0 = fctl0;
1403 imgp->proc->p_elf_machine = hdr->e_machine;
1404 imgp->proc->p_elf_flags = hdr->e_flags;
1407 ASSERT_VOP_LOCKED(imgp->vp, "skipped relock");
1409 free(interp, M_TEMP);
1413 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1416 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1418 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1419 Elf_Auxinfo *argarray, *pos;
1422 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1425 if (args->execfd != -1)
1426 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1427 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1428 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1429 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1430 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1431 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1432 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1433 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1434 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1435 if (imgp->execpathp != 0)
1436 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1437 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1438 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1439 if (imgp->canary != 0) {
1440 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1441 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1443 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1444 if (imgp->pagesizes != 0) {
1445 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1446 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1448 if (imgp->sysent->sv_timekeep_base != 0) {
1449 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1450 imgp->sysent->sv_timekeep_base);
1452 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1453 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1454 imgp->sysent->sv_stackprot);
1455 if (imgp->sysent->sv_hwcap != NULL)
1456 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1457 if (imgp->sysent->sv_hwcap2 != NULL)
1458 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1459 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1460 ELF_BSDF_SIGFASTBLK : 0);
1461 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1462 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1463 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1464 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1465 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1466 if (imgp->sysent->sv_fxrng_gen_base != 0)
1467 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1468 if (imgp->sysent->sv_vdso_base != 0 && __elfN(vdso) != 0)
1469 AUXARGS_ENTRY(pos, AT_KPRELOAD, imgp->sysent->sv_vdso_base);
1470 AUXARGS_ENTRY(pos, AT_NULL, 0);
1472 free(imgp->auxargs, M_TEMP);
1473 imgp->auxargs = NULL;
1474 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1476 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1477 free(argarray, M_TEMP);
1482 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1486 base = (Elf_Addr *)*stack_base;
1488 if (elf_suword(base, imgp->args->argc) == -1)
1490 *stack_base = (uintptr_t)base;
1495 * Code for generating ELF core dumps.
1498 typedef void (*segment_callback)(vm_map_entry_t, void *);
1500 /* Closure for cb_put_phdr(). */
1501 struct phdr_closure {
1502 Elf_Phdr *phdr; /* Program header to fill in */
1503 Elf_Off offset; /* Offset of segment in core file */
1507 int type; /* Note type. */
1508 outfunc_t outfunc; /* Output function. */
1509 void *outarg; /* Argument for the output function. */
1510 size_t outsize; /* Output size. */
1511 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1514 TAILQ_HEAD(note_info_list, note_info);
1516 extern int compress_user_cores;
1517 extern int compress_user_cores_level;
1519 static void cb_put_phdr(vm_map_entry_t, void *);
1520 static void cb_size_segment(vm_map_entry_t, void *);
1521 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1523 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1524 struct note_info_list *, size_t, int);
1525 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1527 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1528 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1529 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1530 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1531 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1532 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1533 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1534 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1535 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1536 static void note_procstat_files(void *, struct sbuf *, size_t *);
1537 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1538 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1539 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1540 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1541 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1544 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1547 return (core_write((struct coredump_params *)arg, base, len, offset,
1548 UIO_SYSSPACE, NULL));
1552 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1554 struct ucred *cred = td->td_ucred;
1555 int compm, error = 0;
1556 struct sseg_closure seginfo;
1557 struct note_info_list notelst;
1558 struct coredump_params params;
1559 struct note_info *ninfo;
1561 size_t hdrsize, notesz, coresize;
1565 TAILQ_INIT(¬elst);
1567 /* Size the program segments. */
1568 __elfN(size_segments)(td, &seginfo, flags);
1571 * Collect info about the core file header area.
1573 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1574 if (seginfo.count + 1 >= PN_XNUM)
1575 hdrsize += sizeof(Elf_Shdr);
1576 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1577 coresize = round_page(hdrsize + notesz) + seginfo.size;
1579 /* Set up core dump parameters. */
1581 params.active_cred = cred;
1582 params.file_cred = NOCRED;
1589 PROC_LOCK(td->td_proc);
1590 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1591 PROC_UNLOCK(td->td_proc);
1598 if (coresize >= limit) {
1603 /* Create a compression stream if necessary. */
1604 compm = compress_user_cores;
1605 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1607 compm = COMPRESS_GZIP;
1609 params.comp = compressor_init(core_compressed_write,
1610 compm, CORE_BUF_SIZE,
1611 compress_user_cores_level, ¶ms);
1612 if (params.comp == NULL) {
1616 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1620 * Allocate memory for building the header, fill it up,
1621 * and write it out following the notes.
1623 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1624 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1627 /* Write the contents of all of the writable segments. */
1633 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1634 offset = round_page(hdrsize + notesz);
1635 for (i = 0; i < seginfo.count; i++) {
1636 error = core_output((char *)(uintptr_t)php->p_vaddr,
1637 php->p_filesz, offset, ¶ms, tmpbuf);
1640 offset += php->p_filesz;
1643 if (error == 0 && params.comp != NULL)
1644 error = compressor_flush(params.comp);
1648 "Failed to write core file for process %s (error %d)\n",
1649 curproc->p_comm, error);
1653 free(tmpbuf, M_TEMP);
1654 if (params.comp != NULL)
1655 compressor_fini(params.comp);
1656 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1657 TAILQ_REMOVE(¬elst, ninfo, link);
1658 free(ninfo, M_TEMP);
1667 * A callback for each_dumpable_segment() to write out the segment's
1668 * program header entry.
1671 cb_put_phdr(vm_map_entry_t entry, void *closure)
1673 struct phdr_closure *phc = (struct phdr_closure *)closure;
1674 Elf_Phdr *phdr = phc->phdr;
1676 phc->offset = round_page(phc->offset);
1678 phdr->p_type = PT_LOAD;
1679 phdr->p_offset = phc->offset;
1680 phdr->p_vaddr = entry->start;
1682 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1683 phdr->p_align = PAGE_SIZE;
1684 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1686 phc->offset += phdr->p_filesz;
1691 * A callback for each_dumpable_segment() to gather information about
1692 * the number of segments and their total size.
1695 cb_size_segment(vm_map_entry_t entry, void *closure)
1697 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1700 ssc->size += entry->end - entry->start;
1704 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1710 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1714 * For each writable segment in the process's memory map, call the given
1715 * function with a pointer to the map entry and some arbitrary
1716 * caller-supplied data.
1719 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1722 struct proc *p = td->td_proc;
1723 vm_map_t map = &p->p_vmspace->vm_map;
1724 vm_map_entry_t entry;
1725 vm_object_t backing_object, object;
1728 vm_map_lock_read(map);
1729 VM_MAP_ENTRY_FOREACH(entry, map) {
1731 * Don't dump inaccessible mappings, deal with legacy
1734 * Note that read-only segments related to the elf binary
1735 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1736 * need to arbitrarily ignore such segments.
1738 if ((flags & SVC_ALL) == 0) {
1739 if (elf_legacy_coredump) {
1740 if ((entry->protection & VM_PROT_RW) !=
1744 if ((entry->protection & VM_PROT_ALL) == 0)
1750 * Dont include memory segment in the coredump if
1751 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1752 * madvise(2). Do not dump submaps (i.e. parts of the
1755 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1757 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1758 (flags & SVC_ALL) == 0)
1760 if ((object = entry->object.vm_object) == NULL)
1763 /* Ignore memory-mapped devices and such things. */
1764 VM_OBJECT_RLOCK(object);
1765 while ((backing_object = object->backing_object) != NULL) {
1766 VM_OBJECT_RLOCK(backing_object);
1767 VM_OBJECT_RUNLOCK(object);
1768 object = backing_object;
1770 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1771 VM_OBJECT_RUNLOCK(object);
1775 (*func)(entry, closure);
1777 vm_map_unlock_read(map);
1781 * Write the core file header to the file, including padding up to
1782 * the page boundary.
1785 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1786 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1789 struct note_info *ninfo;
1793 /* Fill in the header. */
1794 bzero(hdr, hdrsize);
1795 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1797 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1798 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1799 sbuf_start_section(sb, NULL);
1800 sbuf_bcat(sb, hdr, hdrsize);
1801 TAILQ_FOREACH(ninfo, notelst, link)
1802 __elfN(putnote)(p->td, ninfo, sb);
1803 /* Align up to a page boundary for the program segments. */
1804 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1805 error = sbuf_finish(sb);
1812 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1822 size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1825 * To have the debugger select the right thread (LWP) as the initial
1826 * thread, we dump the state of the thread passed to us in td first.
1827 * This is the thread that causes the core dump and thus likely to
1828 * be the right thread one wants to have selected in the debugger.
1831 while (thr != NULL) {
1832 size += __elfN(register_note)(td, list, NT_PRSTATUS,
1833 __elfN(note_prstatus), thr);
1834 size += __elfN(register_note)(td, list, NT_FPREGSET,
1835 __elfN(note_fpregset), thr);
1836 size += __elfN(register_note)(td, list, NT_THRMISC,
1837 __elfN(note_thrmisc), thr);
1838 size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1839 __elfN(note_ptlwpinfo), thr);
1840 size += __elfN(register_note)(td, list, -1,
1841 __elfN(note_threadmd), thr);
1843 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1844 TAILQ_NEXT(thr, td_plist);
1846 thr = TAILQ_NEXT(thr, td_plist);
1849 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1850 __elfN(note_procstat_proc), p);
1851 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1852 note_procstat_files, p);
1853 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1854 note_procstat_vmmap, p);
1855 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1856 note_procstat_groups, p);
1857 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1858 note_procstat_umask, p);
1859 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1860 note_procstat_rlimit, p);
1861 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1862 note_procstat_osrel, p);
1863 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1864 __elfN(note_procstat_psstrings), p);
1865 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1866 __elfN(note_procstat_auxv), p);
1872 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1873 size_t notesz, int flags)
1878 struct phdr_closure phc;
1880 ehdr = (Elf_Ehdr *)hdr;
1882 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1883 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1884 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1885 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1886 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1887 ehdr->e_ident[EI_DATA] = ELF_DATA;
1888 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1889 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1890 ehdr->e_ident[EI_ABIVERSION] = 0;
1891 ehdr->e_ident[EI_PAD] = 0;
1892 ehdr->e_type = ET_CORE;
1893 ehdr->e_machine = td->td_proc->p_elf_machine;
1894 ehdr->e_version = EV_CURRENT;
1896 ehdr->e_phoff = sizeof(Elf_Ehdr);
1897 ehdr->e_flags = td->td_proc->p_elf_flags;
1898 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1899 ehdr->e_phentsize = sizeof(Elf_Phdr);
1900 ehdr->e_shentsize = sizeof(Elf_Shdr);
1901 ehdr->e_shstrndx = SHN_UNDEF;
1902 if (numsegs + 1 < PN_XNUM) {
1903 ehdr->e_phnum = numsegs + 1;
1906 ehdr->e_phnum = PN_XNUM;
1909 ehdr->e_shoff = ehdr->e_phoff +
1910 (numsegs + 1) * ehdr->e_phentsize;
1911 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1912 ("e_shoff: %zu, hdrsize - shdr: %zu",
1913 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1915 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1916 memset(shdr, 0, sizeof(*shdr));
1918 * A special first section is used to hold large segment and
1919 * section counts. This was proposed by Sun Microsystems in
1920 * Solaris and has been adopted by Linux; the standard ELF
1921 * tools are already familiar with the technique.
1923 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1924 * (or 12-7 depending on the version of the document) for more
1927 shdr->sh_type = SHT_NULL;
1928 shdr->sh_size = ehdr->e_shnum;
1929 shdr->sh_link = ehdr->e_shstrndx;
1930 shdr->sh_info = numsegs + 1;
1934 * Fill in the program header entries.
1936 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1938 /* The note segement. */
1939 phdr->p_type = PT_NOTE;
1940 phdr->p_offset = hdrsize;
1943 phdr->p_filesz = notesz;
1945 phdr->p_flags = PF_R;
1946 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1949 /* All the writable segments from the program. */
1951 phc.offset = round_page(hdrsize + notesz);
1952 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1956 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1957 int type, outfunc_t out, void *arg)
1959 const struct sysentvec *sv;
1960 struct note_info *ninfo;
1961 size_t size, notesize;
1963 sv = td->td_proc->p_sysent;
1965 out(arg, NULL, &size);
1966 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1968 ninfo->outfunc = out;
1969 ninfo->outarg = arg;
1970 ninfo->outsize = size;
1971 TAILQ_INSERT_TAIL(list, ninfo, link);
1976 notesize = sizeof(Elf_Note) + /* note header */
1977 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1979 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1985 append_note_data(const void *src, void *dst, size_t len)
1989 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1991 bcopy(src, dst, len);
1992 bzero((char *)dst + len, padded_len - len);
1994 return (padded_len);
1998 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2006 note = (Elf_Note *)buf;
2007 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2008 note->n_descsz = size;
2009 note->n_type = type;
2010 buf += sizeof(*note);
2011 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2012 sizeof(FREEBSD_ABI_VENDOR));
2013 append_note_data(src, buf, size);
2018 notesize = sizeof(Elf_Note) + /* note header */
2019 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2021 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2027 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
2030 const struct sysentvec *sv;
2031 ssize_t old_len, sect_len;
2032 size_t new_len, descsz, i;
2034 if (ninfo->type == -1) {
2035 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2039 sv = td->td_proc->p_sysent;
2041 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
2042 note.n_descsz = ninfo->outsize;
2043 note.n_type = ninfo->type;
2045 sbuf_bcat(sb, ¬e, sizeof(note));
2046 sbuf_start_section(sb, &old_len);
2047 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
2048 strlen(sv->sv_elf_core_abi_vendor) + 1);
2049 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2050 if (note.n_descsz == 0)
2052 sbuf_start_section(sb, &old_len);
2053 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2054 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2058 new_len = (size_t)sect_len;
2059 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2060 if (new_len < descsz) {
2062 * It is expected that individual note emitters will correctly
2063 * predict their expected output size and fill up to that size
2064 * themselves, padding in a format-specific way if needed.
2065 * However, in case they don't, just do it here with zeros.
2067 for (i = 0; i < descsz - new_len; i++)
2069 } else if (new_len > descsz) {
2071 * We can't always truncate sb -- we may have drained some
2074 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2075 "read it (%zu > %zu). Since it is longer than "
2076 "expected, this coredump's notes are corrupt. THIS "
2077 "IS A BUG in the note_procstat routine for type %u.\n",
2078 __func__, (unsigned)note.n_type, new_len, descsz,
2079 (unsigned)note.n_type));
2084 * Miscellaneous note out functions.
2087 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2088 #include <compat/freebsd32/freebsd32.h>
2089 #include <compat/freebsd32/freebsd32_signal.h>
2091 typedef struct prstatus32 elf_prstatus_t;
2092 typedef struct prpsinfo32 elf_prpsinfo_t;
2093 typedef struct fpreg32 elf_prfpregset_t;
2094 typedef struct fpreg32 elf_fpregset_t;
2095 typedef struct reg32 elf_gregset_t;
2096 typedef struct thrmisc32 elf_thrmisc_t;
2097 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2098 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2099 typedef uint32_t elf_ps_strings_t;
2101 typedef prstatus_t elf_prstatus_t;
2102 typedef prpsinfo_t elf_prpsinfo_t;
2103 typedef prfpregset_t elf_prfpregset_t;
2104 typedef prfpregset_t elf_fpregset_t;
2105 typedef gregset_t elf_gregset_t;
2106 typedef thrmisc_t elf_thrmisc_t;
2107 #define ELF_KERN_PROC_MASK 0
2108 typedef struct kinfo_proc elf_kinfo_proc_t;
2109 typedef vm_offset_t elf_ps_strings_t;
2113 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2119 elf_prpsinfo_t *psinfo;
2124 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2125 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2126 psinfo->pr_version = PRPSINFO_VERSION;
2127 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2128 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2130 if (p->p_args != NULL) {
2131 len = sizeof(psinfo->pr_psargs) - 1;
2132 if (len > p->p_args->ar_length)
2133 len = p->p_args->ar_length;
2134 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2140 sbuf_new(&sbarg, psinfo->pr_psargs,
2141 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2142 error = proc_getargv(curthread, p, &sbarg);
2144 if (sbuf_finish(&sbarg) == 0)
2145 len = sbuf_len(&sbarg) - 1;
2147 len = sizeof(psinfo->pr_psargs) - 1;
2148 sbuf_delete(&sbarg);
2150 if (error || len == 0)
2151 strlcpy(psinfo->pr_psargs, p->p_comm,
2152 sizeof(psinfo->pr_psargs));
2154 KASSERT(len < sizeof(psinfo->pr_psargs),
2155 ("len is too long: %zu vs %zu", len,
2156 sizeof(psinfo->pr_psargs)));
2157 cp = psinfo->pr_psargs;
2160 cp = memchr(cp, '\0', end - cp);
2166 psinfo->pr_pid = p->p_pid;
2167 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2168 free(psinfo, M_TEMP);
2170 *sizep = sizeof(*psinfo);
2174 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2177 elf_prstatus_t *status;
2181 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2182 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2183 status->pr_version = PRSTATUS_VERSION;
2184 status->pr_statussz = sizeof(elf_prstatus_t);
2185 status->pr_gregsetsz = sizeof(elf_gregset_t);
2186 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2187 status->pr_osreldate = osreldate;
2188 status->pr_cursig = td->td_proc->p_sig;
2189 status->pr_pid = td->td_tid;
2190 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2191 fill_regs32(td, &status->pr_reg);
2193 fill_regs(td, &status->pr_reg);
2195 sbuf_bcat(sb, status, sizeof(*status));
2196 free(status, M_TEMP);
2198 *sizep = sizeof(*status);
2202 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2205 elf_prfpregset_t *fpregset;
2209 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2210 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2211 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2212 fill_fpregs32(td, fpregset);
2214 fill_fpregs(td, fpregset);
2216 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2217 free(fpregset, M_TEMP);
2219 *sizep = sizeof(*fpregset);
2223 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2226 elf_thrmisc_t thrmisc;
2230 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2231 bzero(&thrmisc, sizeof(thrmisc));
2232 strcpy(thrmisc.pr_tname, td->td_name);
2233 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2235 *sizep = sizeof(thrmisc);
2239 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2244 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2245 struct ptrace_lwpinfo32 pl;
2247 struct ptrace_lwpinfo pl;
2251 size = sizeof(structsize) + sizeof(pl);
2253 KASSERT(*sizep == size, ("invalid size"));
2254 structsize = sizeof(pl);
2255 sbuf_bcat(sb, &structsize, sizeof(structsize));
2256 bzero(&pl, sizeof(pl));
2257 pl.pl_lwpid = td->td_tid;
2258 pl.pl_event = PL_EVENT_NONE;
2259 pl.pl_sigmask = td->td_sigmask;
2260 pl.pl_siglist = td->td_siglist;
2261 if (td->td_si.si_signo != 0) {
2262 pl.pl_event = PL_EVENT_SIGNAL;
2263 pl.pl_flags |= PL_FLAG_SI;
2264 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2265 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2267 pl.pl_siginfo = td->td_si;
2270 strcpy(pl.pl_tdname, td->td_name);
2271 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2272 sbuf_bcat(sb, &pl, sizeof(pl));
2278 * Allow for MD specific notes, as well as any MD
2279 * specific preparations for writing MI notes.
2282 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2288 td = (struct thread *)arg;
2290 if (size != 0 && sb != NULL)
2291 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2295 __elfN(dump_thread)(td, buf, &size);
2296 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2297 if (size != 0 && sb != NULL)
2298 sbuf_bcat(sb, buf, size);
2303 #ifdef KINFO_PROC_SIZE
2304 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2308 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2315 size = sizeof(structsize) + p->p_numthreads *
2316 sizeof(elf_kinfo_proc_t);
2319 KASSERT(*sizep == size, ("invalid size"));
2320 structsize = sizeof(elf_kinfo_proc_t);
2321 sbuf_bcat(sb, &structsize, sizeof(structsize));
2322 sx_slock(&proctree_lock);
2324 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2325 sx_sunlock(&proctree_lock);
2330 #ifdef KINFO_FILE_SIZE
2331 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2335 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2338 size_t size, sect_sz, i;
2339 ssize_t start_len, sect_len;
2340 int structsize, filedesc_flags;
2342 if (coredump_pack_fileinfo)
2343 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2348 structsize = sizeof(struct kinfo_file);
2351 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2352 sbuf_set_drain(sb, sbuf_count_drain, &size);
2353 sbuf_bcat(sb, &structsize, sizeof(structsize));
2355 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2360 sbuf_start_section(sb, &start_len);
2362 sbuf_bcat(sb, &structsize, sizeof(structsize));
2364 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2367 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2372 KASSERT(sect_sz <= *sizep,
2373 ("kern_proc_filedesc_out did not respect maxlen; "
2374 "requested %zu, got %zu", *sizep - sizeof(structsize),
2375 sect_sz - sizeof(structsize)));
2377 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2382 #ifdef KINFO_VMENTRY_SIZE
2383 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2387 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2391 int structsize, vmmap_flags;
2393 if (coredump_pack_vmmapinfo)
2394 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2399 structsize = sizeof(struct kinfo_vmentry);
2402 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2403 sbuf_set_drain(sb, sbuf_count_drain, &size);
2404 sbuf_bcat(sb, &structsize, sizeof(structsize));
2406 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2411 sbuf_bcat(sb, &structsize, sizeof(structsize));
2413 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2419 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2426 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2428 KASSERT(*sizep == size, ("invalid size"));
2429 structsize = sizeof(gid_t);
2430 sbuf_bcat(sb, &structsize, sizeof(structsize));
2431 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2438 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2445 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2447 KASSERT(*sizep == size, ("invalid size"));
2448 structsize = sizeof(p->p_pd->pd_cmask);
2449 sbuf_bcat(sb, &structsize, sizeof(structsize));
2450 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2456 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2459 struct rlimit rlim[RLIM_NLIMITS];
2464 size = sizeof(structsize) + sizeof(rlim);
2466 KASSERT(*sizep == size, ("invalid size"));
2467 structsize = sizeof(rlim);
2468 sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 for (i = 0; i < RLIM_NLIMITS; i++)
2471 lim_rlimit_proc(p, i, &rlim[i]);
2473 sbuf_bcat(sb, rlim, sizeof(rlim));
2479 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2486 size = sizeof(structsize) + sizeof(p->p_osrel);
2488 KASSERT(*sizep == size, ("invalid size"));
2489 structsize = sizeof(p->p_osrel);
2490 sbuf_bcat(sb, &structsize, sizeof(structsize));
2491 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2497 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2500 elf_ps_strings_t ps_strings;
2505 size = sizeof(structsize) + sizeof(ps_strings);
2507 KASSERT(*sizep == size, ("invalid size"));
2508 structsize = sizeof(ps_strings);
2509 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2510 ps_strings = PTROUT(PROC_PS_STRINGS(p));
2512 ps_strings = PROC_PS_STRINGS(p);
2514 sbuf_bcat(sb, &structsize, sizeof(structsize));
2515 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2521 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2530 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2532 sbuf_set_drain(sb, sbuf_count_drain, &size);
2533 sbuf_bcat(sb, &structsize, sizeof(structsize));
2535 proc_getauxv(curthread, p, sb);
2541 structsize = sizeof(Elf_Auxinfo);
2542 sbuf_bcat(sb, &structsize, sizeof(structsize));
2544 proc_getauxv(curthread, p, sb);
2550 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2551 const char *note_vendor, const Elf_Phdr *pnote,
2552 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2554 const Elf_Note *note, *note0, *note_end;
2555 const char *note_name;
2560 /* We need some limit, might as well use PAGE_SIZE. */
2561 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2563 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2564 if (pnote->p_offset > PAGE_SIZE ||
2565 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2566 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2568 VOP_UNLOCK(imgp->vp);
2569 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2570 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2572 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2573 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2574 curthread->td_ucred, NOCRED, NULL, curthread);
2576 uprintf("i/o error PT_NOTE\n");
2579 note = note0 = (const Elf_Note *)buf;
2580 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2582 note = note0 = (const Elf_Note *)(imgp->image_header +
2584 note_end = (const Elf_Note *)(imgp->image_header +
2585 pnote->p_offset + pnote->p_filesz);
2588 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2589 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2590 (const char *)note < sizeof(Elf_Note)) {
2593 if (note->n_namesz != checknote->n_namesz ||
2594 note->n_descsz != checknote->n_descsz ||
2595 note->n_type != checknote->n_type)
2597 note_name = (const char *)(note + 1);
2598 if (note_name + checknote->n_namesz >=
2599 (const char *)note_end || strncmp(note_vendor,
2600 note_name, checknote->n_namesz) != 0)
2603 if (cb(note, cb_arg, &res))
2606 note = (const Elf_Note *)((const char *)(note + 1) +
2607 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2608 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2617 struct brandnote_cb_arg {
2618 Elf_Brandnote *brandnote;
2623 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2625 struct brandnote_cb_arg *arg;
2630 * Fetch the osreldate for binary from the ELF OSABI-note if
2633 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2634 arg->brandnote->trans_osrel != NULL ?
2635 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2640 static Elf_Note fctl_note = {
2641 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2642 .n_descsz = sizeof(uint32_t),
2643 .n_type = NT_FREEBSD_FEATURE_CTL,
2646 struct fctl_cb_arg {
2647 boolean_t *has_fctl0;
2652 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2654 struct fctl_cb_arg *arg;
2655 const Elf32_Word *desc;
2659 p = (uintptr_t)(note + 1);
2660 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2661 desc = (const Elf32_Word *)p;
2662 *arg->has_fctl0 = TRUE;
2663 *arg->fctl0 = desc[0];
2669 * Try to find the appropriate ABI-note section for checknote, fetch
2670 * the osreldate and feature control flags for binary from the ELF
2671 * OSABI-note. Only the first page of the image is searched, the same
2675 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2676 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2678 const Elf_Phdr *phdr;
2679 const Elf_Ehdr *hdr;
2680 struct brandnote_cb_arg b_arg;
2681 struct fctl_cb_arg f_arg;
2684 hdr = (const Elf_Ehdr *)imgp->image_header;
2685 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2686 b_arg.brandnote = brandnote;
2687 b_arg.osrel = osrel;
2688 f_arg.has_fctl0 = has_fctl0;
2689 f_arg.fctl0 = fctl0;
2691 for (i = 0; i < hdr->e_phnum; i++) {
2692 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2693 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2695 for (j = 0; j < hdr->e_phnum; j++) {
2696 if (phdr[j].p_type == PT_NOTE &&
2697 __elfN(parse_notes)(imgp, &fctl_note,
2698 FREEBSD_ABI_VENDOR, &phdr[j],
2699 note_fctl_cb, &f_arg))
2710 * Tell kern_execve.c about it, with a little help from the linker.
2712 static struct execsw __elfN(execsw) = {
2713 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2714 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2716 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2719 __elfN(trans_prot)(Elf_Word flags)
2725 prot |= VM_PROT_EXECUTE;
2727 prot |= VM_PROT_WRITE;
2729 prot |= VM_PROT_READ;
2730 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2731 if (i386_read_exec && (flags & PF_R))
2732 prot |= VM_PROT_EXECUTE;
2738 __elfN(untrans_prot)(vm_prot_t prot)
2743 if (prot & VM_PROT_EXECUTE)
2745 if (prot & VM_PROT_READ)
2747 if (prot & VM_PROT_WRITE)