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>
53 #include <sys/pioctl.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #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, uint32_t *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), CTLFLAG_RW, 0,
108 #define CORE_BUF_SIZE (16 * 1024)
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 __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135 "enable execution from readable segments");
138 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
140 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
142 static int __elfN(aslr_enabled) = 0;
143 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
144 &__elfN(aslr_enabled), 0,
145 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
146 ": enable address map randomization");
148 static int __elfN(pie_aslr_enabled) = 0;
149 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
150 &__elfN(pie_aslr_enabled), 0,
151 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
152 ": enable address map randomization for PIE binaries");
154 static int __elfN(aslr_honor_sbrk) = 1;
155 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
156 &__elfN(aslr_honor_sbrk), 0,
157 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
159 static int __elfN(aslr_stack_gap) = 3;
160 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
161 &__elfN(aslr_stack_gap), 0,
162 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
163 ": maximum percentage of main stack to waste on a random gap");
165 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
167 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
169 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
171 Elf_Brandnote __elfN(freebsd_brandnote) = {
172 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
173 .hdr.n_descsz = sizeof(int32_t),
174 .hdr.n_type = NT_FREEBSD_ABI_TAG,
175 .vendor = FREEBSD_ABI_VENDOR,
176 .flags = BN_TRANSLATE_OSREL,
177 .trans_osrel = __elfN(freebsd_trans_osrel)
181 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
185 p = (uintptr_t)(note + 1);
186 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
187 *osrel = *(const int32_t *)(p);
192 static const char GNU_ABI_VENDOR[] = "GNU";
193 static int GNU_KFREEBSD_ABI_DESC = 3;
195 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
196 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
197 .hdr.n_descsz = 16, /* XXX at least 16 */
199 .vendor = GNU_ABI_VENDOR,
200 .flags = BN_TRANSLATE_OSREL,
201 .trans_osrel = kfreebsd_trans_osrel
205 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
207 const Elf32_Word *desc;
210 p = (uintptr_t)(note + 1);
211 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
213 desc = (const Elf32_Word *)p;
214 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
218 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
219 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
221 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
227 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
231 for (i = 0; i < MAX_BRANDS; i++) {
232 if (elf_brand_list[i] == NULL) {
233 elf_brand_list[i] = entry;
237 if (i == MAX_BRANDS) {
238 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
246 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
250 for (i = 0; i < MAX_BRANDS; i++) {
251 if (elf_brand_list[i] == entry) {
252 elf_brand_list[i] = NULL;
262 __elfN(brand_inuse)(Elf_Brandinfo *entry)
267 sx_slock(&allproc_lock);
268 FOREACH_PROC_IN_SYSTEM(p) {
269 if (p->p_sysent == entry->sysvec) {
274 sx_sunlock(&allproc_lock);
279 static Elf_Brandinfo *
280 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
281 int32_t *osrel, uint32_t *fctl0)
283 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
284 Elf_Brandinfo *bi, *bi_m;
286 int i, interp_name_len;
288 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
291 * We support four types of branding -- (1) the ELF EI_OSABI field
292 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
293 * branding w/in the ELF header, (3) path of the `interp_path'
294 * field, and (4) the ".note.ABI-tag" ELF section.
297 /* Look for an ".note.ABI-tag" ELF section */
299 for (i = 0; i < MAX_BRANDS; i++) {
300 bi = elf_brand_list[i];
303 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
305 if (hdr->e_machine == bi->machine && (bi->flags &
306 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
307 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
309 /* Give brand a chance to veto check_note's guess */
310 if (ret && bi->header_supported)
311 ret = bi->header_supported(imgp);
313 * If note checker claimed the binary, but the
314 * interpreter path in the image does not
315 * match default one for the brand, try to
316 * search for other brands with the same
317 * interpreter. Either there is better brand
318 * with the right interpreter, or, failing
319 * this, we return first brand which accepted
320 * our note and, optionally, header.
322 if (ret && bi_m == NULL && interp != NULL &&
323 (bi->interp_path == NULL ||
324 (strlen(bi->interp_path) + 1 != interp_name_len ||
325 strncmp(interp, bi->interp_path, interp_name_len)
337 /* If the executable has a brand, search for it in the brand list. */
338 for (i = 0; i < MAX_BRANDS; i++) {
339 bi = elf_brand_list[i];
340 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
341 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
343 if (hdr->e_machine == bi->machine &&
344 (hdr->e_ident[EI_OSABI] == bi->brand ||
345 (bi->compat_3_brand != NULL &&
346 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
347 bi->compat_3_brand) == 0))) {
348 /* Looks good, but give brand a chance to veto */
349 if (bi->header_supported == NULL ||
350 bi->header_supported(imgp)) {
352 * Again, prefer strictly matching
355 if (interp_name_len == 0 &&
356 bi->interp_path == NULL)
358 if (bi->interp_path != NULL &&
359 strlen(bi->interp_path) + 1 ==
360 interp_name_len && strncmp(interp,
361 bi->interp_path, interp_name_len) == 0)
371 /* No known brand, see if the header is recognized by any brand */
372 for (i = 0; i < MAX_BRANDS; i++) {
373 bi = elf_brand_list[i];
374 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
375 bi->header_supported == NULL)
377 if (hdr->e_machine == bi->machine) {
378 ret = bi->header_supported(imgp);
384 /* Lacking a known brand, search for a recognized interpreter. */
385 if (interp != NULL) {
386 for (i = 0; i < MAX_BRANDS; i++) {
387 bi = elf_brand_list[i];
388 if (bi == NULL || (bi->flags &
389 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
392 if (hdr->e_machine == bi->machine &&
393 bi->interp_path != NULL &&
394 /* ELF image p_filesz includes terminating zero */
395 strlen(bi->interp_path) + 1 == interp_name_len &&
396 strncmp(interp, bi->interp_path, interp_name_len)
397 == 0 && (bi->header_supported == NULL ||
398 bi->header_supported(imgp)))
403 /* Lacking a recognized interpreter, try the default brand */
404 for (i = 0; i < MAX_BRANDS; i++) {
405 bi = elf_brand_list[i];
406 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
407 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
409 if (hdr->e_machine == bi->machine &&
410 __elfN(fallback_brand) == bi->brand &&
411 (bi->header_supported == NULL ||
412 bi->header_supported(imgp)))
419 __elfN(check_header)(const Elf_Ehdr *hdr)
425 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
426 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
427 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
428 hdr->e_phentsize != sizeof(Elf_Phdr) ||
429 hdr->e_version != ELF_TARG_VER)
433 * Make sure we have at least one brand for this machine.
436 for (i = 0; i < MAX_BRANDS; i++) {
437 bi = elf_brand_list[i];
438 if (bi != NULL && bi->machine == hdr->e_machine)
448 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
449 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
456 * Create the page if it doesn't exist yet. Ignore errors.
458 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
459 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
462 * Find the page from the underlying object.
464 if (object != NULL) {
465 sf = vm_imgact_map_page(object, offset);
467 return (KERN_FAILURE);
468 off = offset - trunc_page(offset);
469 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
471 vm_imgact_unmap_page(sf);
473 return (KERN_FAILURE);
476 return (KERN_SUCCESS);
480 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
481 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
487 int error, locked, rv;
489 if (start != trunc_page(start)) {
490 rv = __elfN(map_partial)(map, object, offset, start,
491 round_page(start), prot);
492 if (rv != KERN_SUCCESS)
494 offset += round_page(start) - start;
495 start = round_page(start);
497 if (end != round_page(end)) {
498 rv = __elfN(map_partial)(map, object, offset +
499 trunc_page(end) - start, trunc_page(end), end, prot);
500 if (rv != KERN_SUCCESS)
502 end = trunc_page(end);
505 return (KERN_SUCCESS);
506 if ((offset & PAGE_MASK) != 0) {
508 * The mapping is not page aligned. This means that we have
511 rv = vm_map_fixed(map, NULL, 0, start, end - start,
512 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
513 if (rv != KERN_SUCCESS)
516 return (KERN_SUCCESS);
517 for (; start < end; start += sz) {
518 sf = vm_imgact_map_page(object, offset);
520 return (KERN_FAILURE);
521 off = offset - trunc_page(offset);
523 if (sz > PAGE_SIZE - off)
524 sz = PAGE_SIZE - off;
525 error = copyout((caddr_t)sf_buf_kva(sf) + off,
527 vm_imgact_unmap_page(sf);
529 return (KERN_FAILURE);
533 vm_object_reference(object);
534 rv = vm_map_fixed(map, object, offset, start, end - start,
535 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
536 (object != NULL ? MAP_VN_EXEC : 0));
537 if (rv != KERN_SUCCESS) {
538 locked = VOP_ISLOCKED(imgp->vp);
539 VOP_UNLOCK(imgp->vp, 0);
540 vm_object_deallocate(object);
541 vn_lock(imgp->vp, locked | LK_RETRY);
543 } else if (object != NULL) {
544 MPASS(imgp->vp->v_object == object);
545 VOP_SET_TEXT_CHECKED(imgp->vp);
548 return (KERN_SUCCESS);
552 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
553 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
559 vm_offset_t map_addr;
562 vm_ooffset_t file_addr;
565 * It's necessary to fail if the filsz + offset taken from the
566 * header is greater than the actual file pager object's size.
567 * If we were to allow this, then the vm_map_find() below would
568 * walk right off the end of the file object and into the ether.
570 * While I'm here, might as well check for something else that
571 * is invalid: filsz cannot be greater than memsz.
573 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
575 uprintf("elf_load_section: truncated ELF file\n");
579 object = imgp->object;
580 map = &imgp->proc->p_vmspace->vm_map;
581 map_addr = trunc_page((vm_offset_t)vmaddr);
582 file_addr = trunc_page(offset);
585 * We have two choices. We can either clear the data in the last page
586 * of an oversized mapping, or we can start the anon mapping a page
587 * early and copy the initialized data into that first page. We
592 else if (memsz > filsz)
593 map_len = trunc_page(offset + filsz) - file_addr;
595 map_len = round_page(offset + filsz) - file_addr;
598 /* cow flags: don't dump readonly sections in core */
599 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
600 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
602 rv = __elfN(map_insert)(imgp, map, object, file_addr,
603 map_addr, map_addr + map_len, prot, cow);
604 if (rv != KERN_SUCCESS)
607 /* we can stop now if we've covered it all */
614 * We have to get the remaining bit of the file into the first part
615 * of the oversized map segment. This is normally because the .data
616 * segment in the file is extended to provide bss. It's a neat idea
617 * to try and save a page, but it's a pain in the behind to implement.
619 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
621 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
622 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
624 /* This had damn well better be true! */
626 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
627 map_addr + map_len, prot, 0);
628 if (rv != KERN_SUCCESS)
633 sf = vm_imgact_map_page(object, offset + filsz);
637 /* send the page fragment to user space */
638 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
640 vm_imgact_unmap_page(sf);
646 * Remove write access to the page if it was only granted by map_insert
649 if ((prot & VM_PROT_WRITE) == 0)
650 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
651 map_len), prot, FALSE);
657 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
658 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
665 ASSERT_VOP_LOCKED(imgp->vp, __func__);
670 for (i = 0; i < hdr->e_phnum; i++) {
671 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
674 /* Loadable segment */
675 prot = __elfN(trans_prot)(phdr[i].p_flags);
676 error = __elfN(load_section)(imgp, phdr[i].p_offset,
677 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
678 phdr[i].p_memsz, phdr[i].p_filesz, prot);
683 * Establish the base address if this is the first segment.
686 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
691 if (base_addrp != NULL)
692 *base_addrp = base_addr;
698 * Load the file "file" into memory. It may be either a shared object
701 * The "addr" reference parameter is in/out. On entry, it specifies
702 * the address where a shared object should be loaded. If the file is
703 * an executable, this value is ignored. On exit, "addr" specifies
704 * where the file was actually loaded.
706 * The "entry" reference parameter is out only. On exit, it specifies
707 * the entry point for the loaded file.
710 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
716 struct image_params image_params;
718 const Elf_Ehdr *hdr = NULL;
719 const Elf_Phdr *phdr = NULL;
720 struct nameidata *nd;
722 struct image_params *imgp;
724 u_long base_addr = 0;
727 #ifdef CAPABILITY_MODE
729 * XXXJA: This check can go away once we are sufficiently confident
730 * that the checks in namei() are correct.
732 if (IN_CAPABILITY_MODE(curthread))
736 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
738 attr = &tempdata->attr;
739 imgp = &tempdata->image_params;
742 * Initialize part of the common data
747 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
748 UIO_SYSSPACE, file, curthread);
749 if ((error = namei(nd)) != 0) {
753 NDFREE(nd, NDF_ONLY_PNBUF);
754 imgp->vp = nd->ni_vp;
757 * Check permissions, modes, uid, etc on the file, and "open" it.
759 error = exec_check_permissions(imgp);
763 error = exec_map_first_page(imgp);
767 imgp->object = nd->ni_vp->v_object;
769 hdr = (const Elf_Ehdr *)imgp->image_header;
770 if ((error = __elfN(check_header)(hdr)) != 0)
772 if (hdr->e_type == ET_DYN)
774 else if (hdr->e_type == ET_EXEC)
781 /* Only support headers that fit within first page for now */
782 if ((hdr->e_phoff > PAGE_SIZE) ||
783 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
788 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
789 if (!aligned(phdr, Elf_Addr)) {
794 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
799 *entry = (unsigned long)hdr->e_entry + rbase;
803 exec_unmap_first_page(imgp);
807 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
810 free(tempdata, M_TEMP);
816 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
821 MPASS(vm_map_min(map) <= minv);
822 MPASS(maxv <= vm_map_max(map));
824 MPASS(minv + align < maxv);
825 arc4rand(&rbase, sizeof(rbase), 0);
826 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
827 res &= ~((u_long)align - 1);
831 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
832 res, minv, maxv, rbase));
834 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
835 res, maxv, minv, rbase));
840 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
841 const Elf_Phdr *phdr, u_long et_dyn_addr)
843 struct vmspace *vmspace;
845 u_long text_size, data_size, total_size, text_addr, data_addr;
846 u_long seg_size, seg_addr;
850 text_size = data_size = total_size = text_addr = data_addr = 0;
852 for (i = 0; i < hdr->e_phnum; i++) {
853 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
856 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
857 seg_size = round_page(phdr[i].p_memsz +
858 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
861 * Make the largest executable segment the official
862 * text segment and all others data.
864 * Note that obreak() assumes that data_addr + data_size == end
865 * of data load area, and the ELF file format expects segments
866 * to be sorted by address. If multiple data segments exist,
867 * the last one will be used.
870 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
871 text_size = seg_size;
872 text_addr = seg_addr;
874 data_size = seg_size;
875 data_addr = seg_addr;
877 total_size += seg_size;
880 if (data_addr == 0 && data_size == 0) {
881 data_addr = text_addr;
882 data_size = text_size;
886 * Check limits. It should be safe to check the
887 * limits after loading the segments since we do
888 * not actually fault in all the segments pages.
890 PROC_LOCK(imgp->proc);
891 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
892 err_str = "Data segment size exceeds process limit";
893 else if (text_size > maxtsiz)
894 err_str = "Text segment size exceeds system limit";
895 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
896 err_str = "Total segment size exceeds process limit";
897 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
898 err_str = "Data segment size exceeds resource limit";
899 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
900 err_str = "Total segment size exceeds resource limit";
901 PROC_UNLOCK(imgp->proc);
902 if (err_str != NULL) {
903 uprintf("%s\n", err_str);
907 vmspace = imgp->proc->p_vmspace;
908 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
909 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
910 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
911 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
917 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
918 char **interpp, bool *free_interpp)
922 int error, interp_name_len;
924 KASSERT(phdr->p_type == PT_INTERP,
925 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
926 ASSERT_VOP_LOCKED(imgp->vp, __func__);
930 /* Path to interpreter */
931 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
932 uprintf("Invalid PT_INTERP\n");
936 interp_name_len = phdr->p_filesz;
937 if (phdr->p_offset > PAGE_SIZE ||
938 interp_name_len > PAGE_SIZE - phdr->p_offset) {
940 * The vnode lock might be needed by the pagedaemon to
941 * clean pages owned by the vnode. Do not allow sleep
942 * waiting for memory with the vnode locked, instead
943 * try non-sleepable allocation first, and if it
944 * fails, go to the slow path were we drop the lock
945 * and do M_WAITOK. A text reference prevents
946 * modifications to the vnode content.
948 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
949 if (interp == NULL) {
950 VOP_UNLOCK(imgp->vp, 0);
951 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
952 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
955 error = vn_rdwr(UIO_READ, imgp->vp, interp,
956 interp_name_len, phdr->p_offset,
957 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
960 free(interp, M_TEMP);
961 uprintf("i/o error PT_INTERP %d\n", error);
964 interp[interp_name_len] = '\0';
967 *free_interpp = true;
971 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
972 if (interp[interp_name_len - 1] != '\0') {
973 uprintf("Invalid PT_INTERP\n");
978 *free_interpp = false;
983 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
984 const char *interp, u_long *addr, u_long *entry)
989 if (brand_info->emul_path != NULL &&
990 brand_info->emul_path[0] != '\0') {
991 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
992 snprintf(path, MAXPATHLEN, "%s%s",
993 brand_info->emul_path, interp);
994 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1000 if (brand_info->interp_newpath != NULL &&
1001 (brand_info->interp_path == NULL ||
1002 strcmp(interp, brand_info->interp_path) == 0)) {
1003 error = __elfN(load_file)(imgp->proc,
1004 brand_info->interp_newpath, addr, entry);
1009 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1013 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1018 * Impossible et_dyn_addr initial value indicating that the real base
1019 * must be calculated later with some randomization applied.
1021 #define ET_DYN_ADDR_RAND 1
1024 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1027 const Elf_Ehdr *hdr;
1028 const Elf_Phdr *phdr;
1029 Elf_Auxargs *elf_auxargs;
1030 struct vmspace *vmspace;
1033 Elf_Brandinfo *brand_info;
1034 struct sysentvec *sv;
1035 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1036 u_long maxalign, mapsz, maxv, maxv1;
1042 hdr = (const Elf_Ehdr *)imgp->image_header;
1045 * Do we have a valid ELF header ?
1047 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1048 * if particular brand doesn't support it.
1050 if (__elfN(check_header)(hdr) != 0 ||
1051 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1055 * From here on down, we return an errno, not -1, as we've
1056 * detected an ELF file.
1059 if ((hdr->e_phoff > PAGE_SIZE) ||
1060 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1061 /* Only support headers in first page for now */
1062 uprintf("Program headers not in the first page\n");
1065 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1066 if (!aligned(phdr, Elf_Addr)) {
1067 uprintf("Unaligned program headers\n");
1075 entry = proghdr = 0;
1077 free_interp = false;
1079 maxalign = PAGE_SIZE;
1082 for (i = 0; i < hdr->e_phnum; i++) {
1083 switch (phdr[i].p_type) {
1086 baddr = phdr[i].p_vaddr;
1087 if (phdr[i].p_align > maxalign)
1088 maxalign = phdr[i].p_align;
1089 mapsz += phdr[i].p_memsz;
1093 * If this segment contains the program headers,
1094 * remember their virtual address for the AT_PHDR
1095 * aux entry. Static binaries don't usually include
1098 if (phdr[i].p_offset == 0 &&
1099 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1100 <= phdr[i].p_filesz)
1101 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1104 /* Path to interpreter */
1105 if (interp != NULL) {
1106 uprintf("Multiple PT_INTERP headers\n");
1110 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1116 if (__elfN(nxstack))
1118 __elfN(trans_prot)(phdr[i].p_flags);
1119 imgp->stack_sz = phdr[i].p_memsz;
1121 case PT_PHDR: /* Program header table info */
1122 proghdr = phdr[i].p_vaddr;
1127 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1128 if (brand_info == NULL) {
1129 uprintf("ELF binary type \"%u\" not known.\n",
1130 hdr->e_ident[EI_OSABI]);
1134 sv = brand_info->sysvec;
1136 if (hdr->e_type == ET_DYN) {
1137 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1138 uprintf("Cannot execute shared object\n");
1143 * Honour the base load address from the dso if it is
1144 * non-zero for some reason.
1147 if ((sv->sv_flags & SV_ASLR) == 0 ||
1148 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1149 et_dyn_addr = ET_DYN_LOAD_ADDR;
1150 else if ((__elfN(pie_aslr_enabled) &&
1151 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1152 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1153 et_dyn_addr = ET_DYN_ADDR_RAND;
1155 et_dyn_addr = ET_DYN_LOAD_ADDR;
1160 * Avoid a possible deadlock if the current address space is destroyed
1161 * and that address space maps the locked vnode. In the common case,
1162 * the locked vnode's v_usecount is decremented but remains greater
1163 * than zero. Consequently, the vnode lock is not needed by vrele().
1164 * However, in cases where the vnode lock is external, such as nullfs,
1165 * v_usecount may become zero.
1167 * The VV_TEXT flag prevents modifications to the executable while
1168 * the vnode is unlocked.
1170 VOP_UNLOCK(imgp->vp, 0);
1173 * Decide whether to enable randomization of user mappings.
1174 * First, reset user preferences for the setid binaries.
1175 * Then, account for the support of the randomization by the
1176 * ABI, by user preferences, and make special treatment for
1179 if (imgp->credential_setid) {
1180 PROC_LOCK(imgp->proc);
1181 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1182 PROC_UNLOCK(imgp->proc);
1184 if ((sv->sv_flags & SV_ASLR) == 0 ||
1185 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1186 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1187 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1188 ("et_dyn_addr == RAND and !ASLR"));
1189 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1190 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1191 et_dyn_addr == ET_DYN_ADDR_RAND) {
1192 imgp->map_flags |= MAP_ASLR;
1194 * If user does not care about sbrk, utilize the bss
1195 * grow region for mappings as well. We can select
1196 * the base for the image anywere and still not suffer
1197 * from the fragmentation.
1199 if (!__elfN(aslr_honor_sbrk) ||
1200 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1201 imgp->map_flags |= MAP_ASLR_IGNSTART;
1204 error = exec_new_vmspace(imgp, sv);
1205 vmspace = imgp->proc->p_vmspace;
1206 map = &vmspace->vm_map;
1208 imgp->proc->p_sysent = sv;
1210 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1211 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1212 KASSERT((map->flags & MAP_ASLR) != 0,
1213 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1214 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1215 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1216 /* reserve half of the address space to interpreter */
1217 maxv / 2, 1UL << flsl(maxalign));
1220 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1224 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1228 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1232 entry = (u_long)hdr->e_entry + et_dyn_addr;
1235 * We load the dynamic linker where a userland call
1236 * to mmap(0, ...) would put it. The rationale behind this
1237 * calculation is that it leaves room for the heap to grow to
1238 * its maximum allowed size.
1240 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1242 if ((map->flags & MAP_ASLR) != 0) {
1243 maxv1 = maxv / 2 + addr / 2;
1244 MPASS(maxv1 >= addr); /* No overflow */
1245 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1246 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1248 map->anon_loc = addr;
1251 imgp->entry_addr = entry;
1253 if (interp != NULL) {
1254 VOP_UNLOCK(imgp->vp, 0);
1255 if ((map->flags & MAP_ASLR) != 0) {
1256 /* Assume that interpeter fits into 1/4 of AS */
1257 maxv1 = maxv / 2 + addr / 2;
1258 MPASS(maxv1 >= addr); /* No overflow */
1259 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1262 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1264 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1271 * Construct auxargs table (used by the fixup routine)
1273 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1274 if (elf_auxargs == NULL) {
1275 VOP_UNLOCK(imgp->vp, 0);
1276 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1277 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1279 elf_auxargs->execfd = -1;
1280 elf_auxargs->phdr = proghdr + et_dyn_addr;
1281 elf_auxargs->phent = hdr->e_phentsize;
1282 elf_auxargs->phnum = hdr->e_phnum;
1283 elf_auxargs->pagesz = PAGE_SIZE;
1284 elf_auxargs->base = addr;
1285 elf_auxargs->flags = 0;
1286 elf_auxargs->entry = entry;
1287 elf_auxargs->hdr_eflags = hdr->e_flags;
1289 imgp->auxargs = elf_auxargs;
1290 imgp->interpreted = 0;
1291 imgp->reloc_base = addr;
1292 imgp->proc->p_osrel = osrel;
1293 imgp->proc->p_fctl0 = fctl0;
1294 imgp->proc->p_elf_machine = hdr->e_machine;
1295 imgp->proc->p_elf_flags = hdr->e_flags;
1299 free(interp, M_TEMP);
1303 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1306 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1308 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1309 Elf_Auxinfo *argarray, *pos;
1310 Elf_Addr *base, *auxbase;
1313 base = (Elf_Addr *)*stack_base;
1314 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1315 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1318 if (args->execfd != -1)
1319 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1320 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1321 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1322 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1323 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1324 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1325 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1326 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1327 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1328 if (imgp->execpathp != 0)
1329 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1330 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1331 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1332 if (imgp->canary != 0) {
1333 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1334 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1336 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1337 if (imgp->pagesizes != 0) {
1338 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1339 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1341 if (imgp->sysent->sv_timekeep_base != 0) {
1342 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1343 imgp->sysent->sv_timekeep_base);
1345 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1346 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1347 imgp->sysent->sv_stackprot);
1348 if (imgp->sysent->sv_hwcap != NULL)
1349 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1350 if (imgp->sysent->sv_hwcap2 != NULL)
1351 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1352 AUXARGS_ENTRY(pos, AT_NULL, 0);
1354 free(imgp->auxargs, M_TEMP);
1355 imgp->auxargs = NULL;
1356 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1358 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1359 free(argarray, M_TEMP);
1364 if (suword(base, imgp->args->argc) == -1)
1366 *stack_base = (register_t *)base;
1371 * Code for generating ELF core dumps.
1374 typedef void (*segment_callback)(vm_map_entry_t, void *);
1376 /* Closure for cb_put_phdr(). */
1377 struct phdr_closure {
1378 Elf_Phdr *phdr; /* Program header to fill in */
1379 Elf_Off offset; /* Offset of segment in core file */
1382 /* Closure for cb_size_segment(). */
1383 struct sseg_closure {
1384 int count; /* Count of writable segments. */
1385 size_t size; /* Total size of all writable segments. */
1388 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1391 int type; /* Note type. */
1392 outfunc_t outfunc; /* Output function. */
1393 void *outarg; /* Argument for the output function. */
1394 size_t outsize; /* Output size. */
1395 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1398 TAILQ_HEAD(note_info_list, note_info);
1400 /* Coredump output parameters. */
1401 struct coredump_params {
1403 struct ucred *active_cred;
1404 struct ucred *file_cred;
1407 struct compressor *comp;
1410 extern int compress_user_cores;
1411 extern int compress_user_cores_level;
1413 static void cb_put_phdr(vm_map_entry_t, void *);
1414 static void cb_size_segment(vm_map_entry_t, void *);
1415 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1417 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1418 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1419 struct note_info_list *, size_t);
1420 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1422 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1423 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1424 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1425 static int sbuf_drain_core_output(void *, const char *, int);
1427 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1428 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1429 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1430 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1431 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1434 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1435 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1436 static void note_procstat_files(void *, struct sbuf *, size_t *);
1437 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1438 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1439 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1440 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1441 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1444 * Write out a core segment to the compression stream.
1447 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1453 chunk_len = MIN(len, CORE_BUF_SIZE);
1456 * We can get EFAULT error here.
1457 * In that case zero out the current chunk of the segment.
1459 error = copyin(base, buf, chunk_len);
1461 bzero(buf, chunk_len);
1462 error = compressor_write(p->comp, buf, chunk_len);
1472 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1475 return (core_write((struct coredump_params *)arg, base, len, offset,
1480 core_write(struct coredump_params *p, const void *base, size_t len,
1481 off_t offset, enum uio_seg seg)
1484 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1485 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1486 p->active_cred, p->file_cred, NULL, p->td));
1490 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1495 if (p->comp != NULL)
1496 return (compress_chunk(p, base, tmpbuf, len));
1499 * EFAULT is a non-fatal error that we can get, for example,
1500 * if the segment is backed by a file but extends beyond its
1503 error = core_write(p, base, len, offset, UIO_USERSPACE);
1504 if (error == EFAULT) {
1505 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1506 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1507 "for process %s\n", base, len, offset, curproc->p_comm);
1510 * Write a "real" zero byte at the end of the target region
1511 * in the case this is the last segment.
1512 * The intermediate space will be implicitly zero-filled.
1514 error = core_write(p, zero_region, 1, offset + len - 1,
1521 * Drain into a core file.
1524 sbuf_drain_core_output(void *arg, const char *data, int len)
1526 struct coredump_params *p;
1529 p = (struct coredump_params *)arg;
1532 * Some kern_proc out routines that print to this sbuf may
1533 * call us with the process lock held. Draining with the
1534 * non-sleepable lock held is unsafe. The lock is needed for
1535 * those routines when dumping a live process. In our case we
1536 * can safely release the lock before draining and acquire
1539 locked = PROC_LOCKED(p->td->td_proc);
1541 PROC_UNLOCK(p->td->td_proc);
1542 if (p->comp != NULL)
1543 error = compressor_write(p->comp, __DECONST(char *, data), len);
1545 error = core_write(p, __DECONST(void *, data), len, p->offset,
1548 PROC_LOCK(p->td->td_proc);
1556 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1558 struct ucred *cred = td->td_ucred;
1560 struct sseg_closure seginfo;
1561 struct note_info_list notelst;
1562 struct coredump_params params;
1563 struct note_info *ninfo;
1565 size_t hdrsize, notesz, coresize;
1569 TAILQ_INIT(¬elst);
1571 /* Size the program segments. */
1574 each_dumpable_segment(td, cb_size_segment, &seginfo);
1577 * Collect info about the core file header area.
1579 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1580 if (seginfo.count + 1 >= PN_XNUM)
1581 hdrsize += sizeof(Elf_Shdr);
1582 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1583 coresize = round_page(hdrsize + notesz) + seginfo.size;
1585 /* Set up core dump parameters. */
1587 params.active_cred = cred;
1588 params.file_cred = NOCRED;
1595 PROC_LOCK(td->td_proc);
1596 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1597 PROC_UNLOCK(td->td_proc);
1604 if (coresize >= limit) {
1609 /* Create a compression stream if necessary. */
1610 if (compress_user_cores != 0) {
1611 params.comp = compressor_init(core_compressed_write,
1612 compress_user_cores, CORE_BUF_SIZE,
1613 compress_user_cores_level, ¶ms);
1614 if (params.comp == NULL) {
1618 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1622 * Allocate memory for building the header, fill it up,
1623 * and write it out following the notes.
1625 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1626 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1629 /* Write the contents of all of the writable segments. */
1635 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1636 offset = round_page(hdrsize + notesz);
1637 for (i = 0; i < seginfo.count; i++) {
1638 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1639 php->p_filesz, offset, ¶ms, tmpbuf);
1642 offset += php->p_filesz;
1645 if (error == 0 && params.comp != NULL)
1646 error = compressor_flush(params.comp);
1650 "Failed to write core file for process %s (error %d)\n",
1651 curproc->p_comm, error);
1655 free(tmpbuf, M_TEMP);
1656 if (params.comp != NULL)
1657 compressor_fini(params.comp);
1658 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1659 TAILQ_REMOVE(¬elst, ninfo, link);
1660 free(ninfo, M_TEMP);
1669 * A callback for each_dumpable_segment() to write out the segment's
1670 * program header entry.
1673 cb_put_phdr(vm_map_entry_t entry, void *closure)
1675 struct phdr_closure *phc = (struct phdr_closure *)closure;
1676 Elf_Phdr *phdr = phc->phdr;
1678 phc->offset = round_page(phc->offset);
1680 phdr->p_type = PT_LOAD;
1681 phdr->p_offset = phc->offset;
1682 phdr->p_vaddr = entry->start;
1684 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1685 phdr->p_align = PAGE_SIZE;
1686 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1688 phc->offset += phdr->p_filesz;
1693 * A callback for each_dumpable_segment() to gather information about
1694 * the number of segments and their total size.
1697 cb_size_segment(vm_map_entry_t entry, void *closure)
1699 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1702 ssc->size += entry->end - entry->start;
1706 * For each writable segment in the process's memory map, call the given
1707 * function with a pointer to the map entry and some arbitrary
1708 * caller-supplied data.
1711 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1713 struct proc *p = td->td_proc;
1714 vm_map_t map = &p->p_vmspace->vm_map;
1715 vm_map_entry_t entry;
1716 vm_object_t backing_object, object;
1717 boolean_t ignore_entry;
1719 vm_map_lock_read(map);
1720 for (entry = map->header.next; entry != &map->header;
1721 entry = entry->next) {
1723 * Don't dump inaccessible mappings, deal with legacy
1726 * Note that read-only segments related to the elf binary
1727 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1728 * need to arbitrarily ignore such segments.
1730 if (elf_legacy_coredump) {
1731 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1734 if ((entry->protection & VM_PROT_ALL) == 0)
1739 * Dont include memory segment in the coredump if
1740 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1741 * madvise(2). Do not dump submaps (i.e. parts of the
1744 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1747 if ((object = entry->object.vm_object) == NULL)
1750 /* Ignore memory-mapped devices and such things. */
1751 VM_OBJECT_RLOCK(object);
1752 while ((backing_object = object->backing_object) != NULL) {
1753 VM_OBJECT_RLOCK(backing_object);
1754 VM_OBJECT_RUNLOCK(object);
1755 object = backing_object;
1757 ignore_entry = object->type != OBJT_DEFAULT &&
1758 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1759 object->type != OBJT_PHYS;
1760 VM_OBJECT_RUNLOCK(object);
1764 (*func)(entry, closure);
1766 vm_map_unlock_read(map);
1770 * Write the core file header to the file, including padding up to
1771 * the page boundary.
1774 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1775 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1777 struct note_info *ninfo;
1781 /* Fill in the header. */
1782 bzero(hdr, hdrsize);
1783 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1785 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1786 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1787 sbuf_start_section(sb, NULL);
1788 sbuf_bcat(sb, hdr, hdrsize);
1789 TAILQ_FOREACH(ninfo, notelst, link)
1790 __elfN(putnote)(ninfo, sb);
1791 /* Align up to a page boundary for the program segments. */
1792 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1793 error = sbuf_finish(sb);
1800 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1810 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1813 * To have the debugger select the right thread (LWP) as the initial
1814 * thread, we dump the state of the thread passed to us in td first.
1815 * This is the thread that causes the core dump and thus likely to
1816 * be the right thread one wants to have selected in the debugger.
1819 while (thr != NULL) {
1820 size += register_note(list, NT_PRSTATUS,
1821 __elfN(note_prstatus), thr);
1822 size += register_note(list, NT_FPREGSET,
1823 __elfN(note_fpregset), thr);
1824 size += register_note(list, NT_THRMISC,
1825 __elfN(note_thrmisc), thr);
1826 size += register_note(list, NT_PTLWPINFO,
1827 __elfN(note_ptlwpinfo), thr);
1828 size += register_note(list, -1,
1829 __elfN(note_threadmd), thr);
1831 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1832 TAILQ_NEXT(thr, td_plist);
1834 thr = TAILQ_NEXT(thr, td_plist);
1837 size += register_note(list, NT_PROCSTAT_PROC,
1838 __elfN(note_procstat_proc), p);
1839 size += register_note(list, NT_PROCSTAT_FILES,
1840 note_procstat_files, p);
1841 size += register_note(list, NT_PROCSTAT_VMMAP,
1842 note_procstat_vmmap, p);
1843 size += register_note(list, NT_PROCSTAT_GROUPS,
1844 note_procstat_groups, p);
1845 size += register_note(list, NT_PROCSTAT_UMASK,
1846 note_procstat_umask, p);
1847 size += register_note(list, NT_PROCSTAT_RLIMIT,
1848 note_procstat_rlimit, p);
1849 size += register_note(list, NT_PROCSTAT_OSREL,
1850 note_procstat_osrel, p);
1851 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1852 __elfN(note_procstat_psstrings), p);
1853 size += register_note(list, NT_PROCSTAT_AUXV,
1854 __elfN(note_procstat_auxv), p);
1860 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1866 struct phdr_closure phc;
1868 ehdr = (Elf_Ehdr *)hdr;
1870 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1871 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1872 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1873 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1874 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1875 ehdr->e_ident[EI_DATA] = ELF_DATA;
1876 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1877 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1878 ehdr->e_ident[EI_ABIVERSION] = 0;
1879 ehdr->e_ident[EI_PAD] = 0;
1880 ehdr->e_type = ET_CORE;
1881 ehdr->e_machine = td->td_proc->p_elf_machine;
1882 ehdr->e_version = EV_CURRENT;
1884 ehdr->e_phoff = sizeof(Elf_Ehdr);
1885 ehdr->e_flags = td->td_proc->p_elf_flags;
1886 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1887 ehdr->e_phentsize = sizeof(Elf_Phdr);
1888 ehdr->e_shentsize = sizeof(Elf_Shdr);
1889 ehdr->e_shstrndx = SHN_UNDEF;
1890 if (numsegs + 1 < PN_XNUM) {
1891 ehdr->e_phnum = numsegs + 1;
1894 ehdr->e_phnum = PN_XNUM;
1897 ehdr->e_shoff = ehdr->e_phoff +
1898 (numsegs + 1) * ehdr->e_phentsize;
1899 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1900 ("e_shoff: %zu, hdrsize - shdr: %zu",
1901 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1903 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1904 memset(shdr, 0, sizeof(*shdr));
1906 * A special first section is used to hold large segment and
1907 * section counts. This was proposed by Sun Microsystems in
1908 * Solaris and has been adopted by Linux; the standard ELF
1909 * tools are already familiar with the technique.
1911 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1912 * (or 12-7 depending on the version of the document) for more
1915 shdr->sh_type = SHT_NULL;
1916 shdr->sh_size = ehdr->e_shnum;
1917 shdr->sh_link = ehdr->e_shstrndx;
1918 shdr->sh_info = numsegs + 1;
1922 * Fill in the program header entries.
1924 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1926 /* The note segement. */
1927 phdr->p_type = PT_NOTE;
1928 phdr->p_offset = hdrsize;
1931 phdr->p_filesz = notesz;
1933 phdr->p_flags = PF_R;
1934 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1937 /* All the writable segments from the program. */
1939 phc.offset = round_page(hdrsize + notesz);
1940 each_dumpable_segment(td, cb_put_phdr, &phc);
1944 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1946 struct note_info *ninfo;
1947 size_t size, notesize;
1950 out(arg, NULL, &size);
1951 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1953 ninfo->outfunc = out;
1954 ninfo->outarg = arg;
1955 ninfo->outsize = size;
1956 TAILQ_INSERT_TAIL(list, ninfo, link);
1961 notesize = sizeof(Elf_Note) + /* note header */
1962 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1964 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1970 append_note_data(const void *src, void *dst, size_t len)
1974 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1976 bcopy(src, dst, len);
1977 bzero((char *)dst + len, padded_len - len);
1979 return (padded_len);
1983 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1991 note = (Elf_Note *)buf;
1992 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1993 note->n_descsz = size;
1994 note->n_type = type;
1995 buf += sizeof(*note);
1996 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1997 sizeof(FREEBSD_ABI_VENDOR));
1998 append_note_data(src, buf, size);
2003 notesize = sizeof(Elf_Note) + /* note header */
2004 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2006 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2012 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2015 ssize_t old_len, sect_len;
2016 size_t new_len, descsz, i;
2018 if (ninfo->type == -1) {
2019 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2023 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2024 note.n_descsz = ninfo->outsize;
2025 note.n_type = ninfo->type;
2027 sbuf_bcat(sb, ¬e, sizeof(note));
2028 sbuf_start_section(sb, &old_len);
2029 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2030 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2031 if (note.n_descsz == 0)
2033 sbuf_start_section(sb, &old_len);
2034 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2035 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2039 new_len = (size_t)sect_len;
2040 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2041 if (new_len < descsz) {
2043 * It is expected that individual note emitters will correctly
2044 * predict their expected output size and fill up to that size
2045 * themselves, padding in a format-specific way if needed.
2046 * However, in case they don't, just do it here with zeros.
2048 for (i = 0; i < descsz - new_len; i++)
2050 } else if (new_len > descsz) {
2052 * We can't always truncate sb -- we may have drained some
2055 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2056 "read it (%zu > %zu). Since it is longer than "
2057 "expected, this coredump's notes are corrupt. THIS "
2058 "IS A BUG in the note_procstat routine for type %u.\n",
2059 __func__, (unsigned)note.n_type, new_len, descsz,
2060 (unsigned)note.n_type));
2065 * Miscellaneous note out functions.
2068 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2069 #include <compat/freebsd32/freebsd32.h>
2070 #include <compat/freebsd32/freebsd32_signal.h>
2072 typedef struct prstatus32 elf_prstatus_t;
2073 typedef struct prpsinfo32 elf_prpsinfo_t;
2074 typedef struct fpreg32 elf_prfpregset_t;
2075 typedef struct fpreg32 elf_fpregset_t;
2076 typedef struct reg32 elf_gregset_t;
2077 typedef struct thrmisc32 elf_thrmisc_t;
2078 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2079 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2080 typedef uint32_t elf_ps_strings_t;
2082 typedef prstatus_t elf_prstatus_t;
2083 typedef prpsinfo_t elf_prpsinfo_t;
2084 typedef prfpregset_t elf_prfpregset_t;
2085 typedef prfpregset_t elf_fpregset_t;
2086 typedef gregset_t elf_gregset_t;
2087 typedef thrmisc_t elf_thrmisc_t;
2088 #define ELF_KERN_PROC_MASK 0
2089 typedef struct kinfo_proc elf_kinfo_proc_t;
2090 typedef vm_offset_t elf_ps_strings_t;
2094 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2100 elf_prpsinfo_t *psinfo;
2103 p = (struct proc *)arg;
2105 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2106 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2107 psinfo->pr_version = PRPSINFO_VERSION;
2108 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2109 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2111 if (p->p_args != NULL) {
2112 len = sizeof(psinfo->pr_psargs) - 1;
2113 if (len > p->p_args->ar_length)
2114 len = p->p_args->ar_length;
2115 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2121 sbuf_new(&sbarg, psinfo->pr_psargs,
2122 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2123 error = proc_getargv(curthread, p, &sbarg);
2125 if (sbuf_finish(&sbarg) == 0)
2126 len = sbuf_len(&sbarg) - 1;
2128 len = sizeof(psinfo->pr_psargs) - 1;
2129 sbuf_delete(&sbarg);
2131 if (error || len == 0)
2132 strlcpy(psinfo->pr_psargs, p->p_comm,
2133 sizeof(psinfo->pr_psargs));
2135 KASSERT(len < sizeof(psinfo->pr_psargs),
2136 ("len is too long: %zu vs %zu", len,
2137 sizeof(psinfo->pr_psargs)));
2138 cp = psinfo->pr_psargs;
2141 cp = memchr(cp, '\0', end - cp);
2147 psinfo->pr_pid = p->p_pid;
2148 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2149 free(psinfo, M_TEMP);
2151 *sizep = sizeof(*psinfo);
2155 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2158 elf_prstatus_t *status;
2160 td = (struct thread *)arg;
2162 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2163 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2164 status->pr_version = PRSTATUS_VERSION;
2165 status->pr_statussz = sizeof(elf_prstatus_t);
2166 status->pr_gregsetsz = sizeof(elf_gregset_t);
2167 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2168 status->pr_osreldate = osreldate;
2169 status->pr_cursig = td->td_proc->p_sig;
2170 status->pr_pid = td->td_tid;
2171 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2172 fill_regs32(td, &status->pr_reg);
2174 fill_regs(td, &status->pr_reg);
2176 sbuf_bcat(sb, status, sizeof(*status));
2177 free(status, M_TEMP);
2179 *sizep = sizeof(*status);
2183 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2186 elf_prfpregset_t *fpregset;
2188 td = (struct thread *)arg;
2190 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2191 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2192 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2193 fill_fpregs32(td, fpregset);
2195 fill_fpregs(td, fpregset);
2197 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2198 free(fpregset, M_TEMP);
2200 *sizep = sizeof(*fpregset);
2204 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2207 elf_thrmisc_t thrmisc;
2209 td = (struct thread *)arg;
2211 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2212 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2213 strcpy(thrmisc.pr_tname, td->td_name);
2214 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2216 *sizep = sizeof(thrmisc);
2220 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2225 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2226 struct ptrace_lwpinfo32 pl;
2228 struct ptrace_lwpinfo pl;
2231 td = (struct thread *)arg;
2232 size = sizeof(structsize) + sizeof(pl);
2234 KASSERT(*sizep == size, ("invalid size"));
2235 structsize = sizeof(pl);
2236 sbuf_bcat(sb, &structsize, sizeof(structsize));
2237 bzero(&pl, sizeof(pl));
2238 pl.pl_lwpid = td->td_tid;
2239 pl.pl_event = PL_EVENT_NONE;
2240 pl.pl_sigmask = td->td_sigmask;
2241 pl.pl_siglist = td->td_siglist;
2242 if (td->td_si.si_signo != 0) {
2243 pl.pl_event = PL_EVENT_SIGNAL;
2244 pl.pl_flags |= PL_FLAG_SI;
2245 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2246 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2248 pl.pl_siginfo = td->td_si;
2251 strcpy(pl.pl_tdname, td->td_name);
2252 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2253 sbuf_bcat(sb, &pl, sizeof(pl));
2259 * Allow for MD specific notes, as well as any MD
2260 * specific preparations for writing MI notes.
2263 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2269 td = (struct thread *)arg;
2271 if (size != 0 && sb != NULL)
2272 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2276 __elfN(dump_thread)(td, buf, &size);
2277 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2278 if (size != 0 && sb != NULL)
2279 sbuf_bcat(sb, buf, size);
2284 #ifdef KINFO_PROC_SIZE
2285 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2289 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2295 p = (struct proc *)arg;
2296 size = sizeof(structsize) + p->p_numthreads *
2297 sizeof(elf_kinfo_proc_t);
2300 KASSERT(*sizep == size, ("invalid size"));
2301 structsize = sizeof(elf_kinfo_proc_t);
2302 sbuf_bcat(sb, &structsize, sizeof(structsize));
2304 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2309 #ifdef KINFO_FILE_SIZE
2310 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2314 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2317 size_t size, sect_sz, i;
2318 ssize_t start_len, sect_len;
2319 int structsize, filedesc_flags;
2321 if (coredump_pack_fileinfo)
2322 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2326 p = (struct proc *)arg;
2327 structsize = sizeof(struct kinfo_file);
2330 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2331 sbuf_set_drain(sb, sbuf_count_drain, &size);
2332 sbuf_bcat(sb, &structsize, sizeof(structsize));
2334 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2339 sbuf_start_section(sb, &start_len);
2341 sbuf_bcat(sb, &structsize, sizeof(structsize));
2343 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2346 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2351 KASSERT(sect_sz <= *sizep,
2352 ("kern_proc_filedesc_out did not respect maxlen; "
2353 "requested %zu, got %zu", *sizep - sizeof(structsize),
2354 sect_sz - sizeof(structsize)));
2356 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2361 #ifdef KINFO_VMENTRY_SIZE
2362 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2366 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2370 int structsize, vmmap_flags;
2372 if (coredump_pack_vmmapinfo)
2373 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2377 p = (struct proc *)arg;
2378 structsize = sizeof(struct kinfo_vmentry);
2381 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2382 sbuf_set_drain(sb, sbuf_count_drain, &size);
2383 sbuf_bcat(sb, &structsize, sizeof(structsize));
2385 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2390 sbuf_bcat(sb, &structsize, sizeof(structsize));
2392 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2398 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2404 p = (struct proc *)arg;
2405 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2407 KASSERT(*sizep == size, ("invalid size"));
2408 structsize = sizeof(gid_t);
2409 sbuf_bcat(sb, &structsize, sizeof(structsize));
2410 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2417 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2423 p = (struct proc *)arg;
2424 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2426 KASSERT(*sizep == size, ("invalid size"));
2427 structsize = sizeof(p->p_fd->fd_cmask);
2428 sbuf_bcat(sb, &structsize, sizeof(structsize));
2429 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2435 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2438 struct rlimit rlim[RLIM_NLIMITS];
2442 p = (struct proc *)arg;
2443 size = sizeof(structsize) + sizeof(rlim);
2445 KASSERT(*sizep == size, ("invalid size"));
2446 structsize = sizeof(rlim);
2447 sbuf_bcat(sb, &structsize, sizeof(structsize));
2449 for (i = 0; i < RLIM_NLIMITS; i++)
2450 lim_rlimit_proc(p, i, &rlim[i]);
2452 sbuf_bcat(sb, rlim, sizeof(rlim));
2458 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2464 p = (struct proc *)arg;
2465 size = sizeof(structsize) + sizeof(p->p_osrel);
2467 KASSERT(*sizep == size, ("invalid size"));
2468 structsize = sizeof(p->p_osrel);
2469 sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2476 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2479 elf_ps_strings_t ps_strings;
2483 p = (struct proc *)arg;
2484 size = sizeof(structsize) + sizeof(ps_strings);
2486 KASSERT(*sizep == size, ("invalid size"));
2487 structsize = sizeof(ps_strings);
2488 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2489 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2491 ps_strings = p->p_sysent->sv_psstrings;
2493 sbuf_bcat(sb, &structsize, sizeof(structsize));
2494 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2500 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2506 p = (struct proc *)arg;
2509 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2510 sbuf_set_drain(sb, sbuf_count_drain, &size);
2511 sbuf_bcat(sb, &structsize, sizeof(structsize));
2513 proc_getauxv(curthread, p, sb);
2519 structsize = sizeof(Elf_Auxinfo);
2520 sbuf_bcat(sb, &structsize, sizeof(structsize));
2522 proc_getauxv(curthread, p, sb);
2528 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2529 const char *note_vendor, const Elf_Phdr *pnote,
2530 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2532 const Elf_Note *note, *note0, *note_end;
2533 const char *note_name;
2538 /* We need some limit, might as well use PAGE_SIZE. */
2539 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2541 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2542 if (pnote->p_offset > PAGE_SIZE ||
2543 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2544 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2546 VOP_UNLOCK(imgp->vp, 0);
2547 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2548 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2550 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2551 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2552 curthread->td_ucred, NOCRED, NULL, curthread);
2554 uprintf("i/o error PT_NOTE\n");
2557 note = note0 = (const Elf_Note *)buf;
2558 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2560 note = note0 = (const Elf_Note *)(imgp->image_header +
2562 note_end = (const Elf_Note *)(imgp->image_header +
2563 pnote->p_offset + pnote->p_filesz);
2566 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2567 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2568 (const char *)note < sizeof(Elf_Note)) {
2571 if (note->n_namesz != checknote->n_namesz ||
2572 note->n_descsz != checknote->n_descsz ||
2573 note->n_type != checknote->n_type)
2575 note_name = (const char *)(note + 1);
2576 if (note_name + checknote->n_namesz >=
2577 (const char *)note_end || strncmp(note_vendor,
2578 note_name, checknote->n_namesz) != 0)
2581 if (cb(note, cb_arg, &res))
2584 note = (const Elf_Note *)((const char *)(note + 1) +
2585 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2586 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2595 struct brandnote_cb_arg {
2596 Elf_Brandnote *brandnote;
2601 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2603 struct brandnote_cb_arg *arg;
2608 * Fetch the osreldate for binary from the ELF OSABI-note if
2611 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2612 arg->brandnote->trans_osrel != NULL ?
2613 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2618 static Elf_Note fctl_note = {
2619 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2620 .n_descsz = sizeof(uint32_t),
2621 .n_type = NT_FREEBSD_FEATURE_CTL,
2624 struct fctl_cb_arg {
2629 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2631 struct fctl_cb_arg *arg;
2632 const Elf32_Word *desc;
2636 p = (uintptr_t)(note + 1);
2637 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2638 desc = (const Elf32_Word *)p;
2639 *arg->fctl0 = desc[0];
2644 * Try to find the appropriate ABI-note section for checknote, fetch
2645 * the osreldate and feature control flags for binary from the ELF
2646 * OSABI-note. Only the first page of the image is searched, the same
2650 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2651 int32_t *osrel, uint32_t *fctl0)
2653 const Elf_Phdr *phdr;
2654 const Elf_Ehdr *hdr;
2655 struct brandnote_cb_arg b_arg;
2656 struct fctl_cb_arg f_arg;
2659 hdr = (const Elf_Ehdr *)imgp->image_header;
2660 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2661 b_arg.brandnote = brandnote;
2662 b_arg.osrel = osrel;
2663 f_arg.fctl0 = fctl0;
2665 for (i = 0; i < hdr->e_phnum; i++) {
2666 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2667 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2669 for (j = 0; j < hdr->e_phnum; j++) {
2670 if (phdr[j].p_type == PT_NOTE &&
2671 __elfN(parse_notes)(imgp, &fctl_note,
2672 FREEBSD_ABI_VENDOR, &phdr[j],
2673 note_fctl_cb, &f_arg))
2684 * Tell kern_execve.c about it, with a little help from the linker.
2686 static struct execsw __elfN(execsw) = {
2687 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2688 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2690 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2693 __elfN(trans_prot)(Elf_Word flags)
2699 prot |= VM_PROT_EXECUTE;
2701 prot |= VM_PROT_WRITE;
2703 prot |= VM_PROT_READ;
2704 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2705 if (i386_read_exec && (flags & PF_R))
2706 prot |= VM_PROT_EXECUTE;
2712 __elfN(untrans_prot)(vm_prot_t prot)
2717 if (prot & VM_PROT_EXECUTE)
2719 if (prot & VM_PROT_READ)
2721 if (prot & VM_PROT_WRITE)
2727 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2729 u_long range, rbase, gap;
2732 if ((imgp->map_flags & MAP_ASLR) == 0)
2734 pct = __elfN(aslr_stack_gap);
2739 range = imgp->eff_stack_sz * pct / 100;
2740 arc4rand(&rbase, sizeof(rbase), 0);
2741 gap = rbase % range;
2742 gap &= ~(sizeof(u_long) - 1);