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);
738 attr = &tempdata->attr;
739 imgp = &tempdata->image_params;
742 * Initialize part of the common data
746 imgp->firstpage = NULL;
747 imgp->image_header = NULL;
749 imgp->execlabel = NULL;
751 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
752 UIO_SYSSPACE, file, curthread);
753 if ((error = namei(nd)) != 0) {
757 NDFREE(nd, NDF_ONLY_PNBUF);
758 imgp->vp = nd->ni_vp;
761 * Check permissions, modes, uid, etc on the file, and "open" it.
763 error = exec_check_permissions(imgp);
767 error = exec_map_first_page(imgp);
771 imgp->object = nd->ni_vp->v_object;
773 hdr = (const Elf_Ehdr *)imgp->image_header;
774 if ((error = __elfN(check_header)(hdr)) != 0)
776 if (hdr->e_type == ET_DYN)
778 else if (hdr->e_type == ET_EXEC)
785 /* Only support headers that fit within first page for now */
786 if ((hdr->e_phoff > PAGE_SIZE) ||
787 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
792 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
793 if (!aligned(phdr, Elf_Addr)) {
798 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
803 *entry = (unsigned long)hdr->e_entry + rbase;
807 exec_unmap_first_page(imgp);
811 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
814 free(tempdata, M_TEMP);
820 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
825 MPASS(vm_map_min(map) <= minv);
826 MPASS(maxv <= vm_map_max(map));
828 MPASS(minv + align < maxv);
829 arc4rand(&rbase, sizeof(rbase), 0);
830 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
831 res &= ~((u_long)align - 1);
835 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
836 res, minv, maxv, rbase));
838 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
839 res, maxv, minv, rbase));
844 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
845 const Elf_Phdr *phdr, u_long et_dyn_addr)
847 struct vmspace *vmspace;
849 u_long text_size, data_size, total_size, text_addr, data_addr;
850 u_long seg_size, seg_addr;
854 text_size = data_size = total_size = text_addr = data_addr = 0;
856 for (i = 0; i < hdr->e_phnum; i++) {
857 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
860 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
861 seg_size = round_page(phdr[i].p_memsz +
862 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
865 * Make the largest executable segment the official
866 * text segment and all others data.
868 * Note that obreak() assumes that data_addr + data_size == end
869 * of data load area, and the ELF file format expects segments
870 * to be sorted by address. If multiple data segments exist,
871 * the last one will be used.
874 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
875 text_size = seg_size;
876 text_addr = seg_addr;
878 data_size = seg_size;
879 data_addr = seg_addr;
881 total_size += seg_size;
884 if (data_addr == 0 && data_size == 0) {
885 data_addr = text_addr;
886 data_size = text_size;
890 * Check limits. It should be safe to check the
891 * limits after loading the segments since we do
892 * not actually fault in all the segments pages.
894 PROC_LOCK(imgp->proc);
895 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
896 err_str = "Data segment size exceeds process limit";
897 else if (text_size > maxtsiz)
898 err_str = "Text segment size exceeds system limit";
899 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
900 err_str = "Total segment size exceeds process limit";
901 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
902 err_str = "Data segment size exceeds resource limit";
903 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
904 err_str = "Total segment size exceeds resource limit";
905 PROC_UNLOCK(imgp->proc);
906 if (err_str != NULL) {
907 uprintf("%s\n", err_str);
911 vmspace = imgp->proc->p_vmspace;
912 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
913 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
914 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
915 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
921 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
922 char **interpp, bool *free_interpp)
926 int error, interp_name_len;
928 KASSERT(phdr->p_type == PT_INTERP,
929 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
930 ASSERT_VOP_LOCKED(imgp->vp, __func__);
934 /* Path to interpreter */
935 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
936 uprintf("Invalid PT_INTERP\n");
940 interp_name_len = phdr->p_filesz;
941 if (phdr->p_offset > PAGE_SIZE ||
942 interp_name_len > PAGE_SIZE - phdr->p_offset) {
944 * The vnode lock might be needed by the pagedaemon to
945 * clean pages owned by the vnode. Do not allow sleep
946 * waiting for memory with the vnode locked, instead
947 * try non-sleepable allocation first, and if it
948 * fails, go to the slow path were we drop the lock
949 * and do M_WAITOK. A text reference prevents
950 * modifications to the vnode content.
952 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
953 if (interp == NULL) {
954 VOP_UNLOCK(imgp->vp, 0);
955 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
956 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
959 error = vn_rdwr(UIO_READ, imgp->vp, interp,
960 interp_name_len, phdr->p_offset,
961 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
964 free(interp, M_TEMP);
965 uprintf("i/o error PT_INTERP %d\n", error);
968 interp[interp_name_len] = '\0';
971 *free_interpp = true;
975 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
976 if (interp[interp_name_len - 1] != '\0') {
977 uprintf("Invalid PT_INTERP\n");
982 *free_interpp = false;
987 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
988 const char *interp, u_long *addr, u_long *entry)
993 if (brand_info->emul_path != NULL &&
994 brand_info->emul_path[0] != '\0') {
995 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
996 snprintf(path, MAXPATHLEN, "%s%s",
997 brand_info->emul_path, interp);
998 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1004 if (brand_info->interp_newpath != NULL &&
1005 (brand_info->interp_path == NULL ||
1006 strcmp(interp, brand_info->interp_path) == 0)) {
1007 error = __elfN(load_file)(imgp->proc,
1008 brand_info->interp_newpath, addr, entry);
1013 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1017 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1022 * Impossible et_dyn_addr initial value indicating that the real base
1023 * must be calculated later with some randomization applied.
1025 #define ET_DYN_ADDR_RAND 1
1028 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1031 const Elf_Ehdr *hdr;
1032 const Elf_Phdr *phdr;
1033 Elf_Auxargs *elf_auxargs;
1034 struct vmspace *vmspace;
1037 Elf_Brandinfo *brand_info;
1038 struct sysentvec *sv;
1039 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1040 u_long maxalign, mapsz, maxv, maxv1;
1046 hdr = (const Elf_Ehdr *)imgp->image_header;
1049 * Do we have a valid ELF header ?
1051 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1052 * if particular brand doesn't support it.
1054 if (__elfN(check_header)(hdr) != 0 ||
1055 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1059 * From here on down, we return an errno, not -1, as we've
1060 * detected an ELF file.
1063 if ((hdr->e_phoff > PAGE_SIZE) ||
1064 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1065 /* Only support headers in first page for now */
1066 uprintf("Program headers not in the first page\n");
1069 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1070 if (!aligned(phdr, Elf_Addr)) {
1071 uprintf("Unaligned program headers\n");
1079 entry = proghdr = 0;
1081 free_interp = false;
1083 maxalign = PAGE_SIZE;
1086 for (i = 0; i < hdr->e_phnum; i++) {
1087 switch (phdr[i].p_type) {
1090 baddr = phdr[i].p_vaddr;
1091 if (phdr[i].p_align > maxalign)
1092 maxalign = phdr[i].p_align;
1093 mapsz += phdr[i].p_memsz;
1097 * If this segment contains the program headers,
1098 * remember their virtual address for the AT_PHDR
1099 * aux entry. Static binaries don't usually include
1102 if (phdr[i].p_offset == 0 &&
1103 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1104 <= phdr[i].p_filesz)
1105 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1108 /* Path to interpreter */
1109 if (interp != NULL) {
1110 uprintf("Multiple PT_INTERP headers\n");
1114 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1120 if (__elfN(nxstack))
1122 __elfN(trans_prot)(phdr[i].p_flags);
1123 imgp->stack_sz = phdr[i].p_memsz;
1125 case PT_PHDR: /* Program header table info */
1126 proghdr = phdr[i].p_vaddr;
1131 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1132 if (brand_info == NULL) {
1133 uprintf("ELF binary type \"%u\" not known.\n",
1134 hdr->e_ident[EI_OSABI]);
1138 sv = brand_info->sysvec;
1140 if (hdr->e_type == ET_DYN) {
1141 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1142 uprintf("Cannot execute shared object\n");
1147 * Honour the base load address from the dso if it is
1148 * non-zero for some reason.
1151 if ((sv->sv_flags & SV_ASLR) == 0 ||
1152 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1153 et_dyn_addr = ET_DYN_LOAD_ADDR;
1154 else if ((__elfN(pie_aslr_enabled) &&
1155 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1156 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1157 et_dyn_addr = ET_DYN_ADDR_RAND;
1159 et_dyn_addr = ET_DYN_LOAD_ADDR;
1164 * Avoid a possible deadlock if the current address space is destroyed
1165 * and that address space maps the locked vnode. In the common case,
1166 * the locked vnode's v_usecount is decremented but remains greater
1167 * than zero. Consequently, the vnode lock is not needed by vrele().
1168 * However, in cases where the vnode lock is external, such as nullfs,
1169 * v_usecount may become zero.
1171 * The VV_TEXT flag prevents modifications to the executable while
1172 * the vnode is unlocked.
1174 VOP_UNLOCK(imgp->vp, 0);
1177 * Decide whether to enable randomization of user mappings.
1178 * First, reset user preferences for the setid binaries.
1179 * Then, account for the support of the randomization by the
1180 * ABI, by user preferences, and make special treatment for
1183 if (imgp->credential_setid) {
1184 PROC_LOCK(imgp->proc);
1185 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1186 PROC_UNLOCK(imgp->proc);
1188 if ((sv->sv_flags & SV_ASLR) == 0 ||
1189 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1190 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1191 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1192 ("et_dyn_addr == RAND and !ASLR"));
1193 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1194 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1195 et_dyn_addr == ET_DYN_ADDR_RAND) {
1196 imgp->map_flags |= MAP_ASLR;
1198 * If user does not care about sbrk, utilize the bss
1199 * grow region for mappings as well. We can select
1200 * the base for the image anywere and still not suffer
1201 * from the fragmentation.
1203 if (!__elfN(aslr_honor_sbrk) ||
1204 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1205 imgp->map_flags |= MAP_ASLR_IGNSTART;
1208 error = exec_new_vmspace(imgp, sv);
1209 vmspace = imgp->proc->p_vmspace;
1210 map = &vmspace->vm_map;
1212 imgp->proc->p_sysent = sv;
1214 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1215 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1216 KASSERT((map->flags & MAP_ASLR) != 0,
1217 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1218 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1219 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1220 /* reserve half of the address space to interpreter */
1221 maxv / 2, 1UL << flsl(maxalign));
1224 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1228 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1232 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1236 entry = (u_long)hdr->e_entry + et_dyn_addr;
1239 * We load the dynamic linker where a userland call
1240 * to mmap(0, ...) would put it. The rationale behind this
1241 * calculation is that it leaves room for the heap to grow to
1242 * its maximum allowed size.
1244 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1246 if ((map->flags & MAP_ASLR) != 0) {
1247 maxv1 = maxv / 2 + addr / 2;
1248 MPASS(maxv1 >= addr); /* No overflow */
1249 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1250 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1252 map->anon_loc = addr;
1255 imgp->entry_addr = entry;
1257 if (interp != NULL) {
1258 VOP_UNLOCK(imgp->vp, 0);
1259 if ((map->flags & MAP_ASLR) != 0) {
1260 /* Assume that interpeter fits into 1/4 of AS */
1261 maxv1 = maxv / 2 + addr / 2;
1262 MPASS(maxv1 >= addr); /* No overflow */
1263 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1266 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1268 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1275 * Construct auxargs table (used by the fixup routine)
1277 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1278 if (elf_auxargs == NULL) {
1279 VOP_UNLOCK(imgp->vp, 0);
1280 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1281 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1283 elf_auxargs->execfd = -1;
1284 elf_auxargs->phdr = proghdr + et_dyn_addr;
1285 elf_auxargs->phent = hdr->e_phentsize;
1286 elf_auxargs->phnum = hdr->e_phnum;
1287 elf_auxargs->pagesz = PAGE_SIZE;
1288 elf_auxargs->base = addr;
1289 elf_auxargs->flags = 0;
1290 elf_auxargs->entry = entry;
1291 elf_auxargs->hdr_eflags = hdr->e_flags;
1293 imgp->auxargs = elf_auxargs;
1294 imgp->interpreted = 0;
1295 imgp->reloc_base = addr;
1296 imgp->proc->p_osrel = osrel;
1297 imgp->proc->p_fctl0 = fctl0;
1298 imgp->proc->p_elf_machine = hdr->e_machine;
1299 imgp->proc->p_elf_flags = hdr->e_flags;
1303 free(interp, M_TEMP);
1307 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1310 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1312 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1313 Elf_Auxinfo *argarray, *pos;
1314 Elf_Addr *base, *auxbase;
1317 base = (Elf_Addr *)*stack_base;
1318 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1319 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1322 if (args->execfd != -1)
1323 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1324 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1325 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1326 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1327 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1328 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1329 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1330 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1331 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1332 if (imgp->execpathp != 0)
1333 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1334 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1335 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1336 if (imgp->canary != 0) {
1337 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1338 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1340 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1341 if (imgp->pagesizes != 0) {
1342 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1343 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1345 if (imgp->sysent->sv_timekeep_base != 0) {
1346 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1347 imgp->sysent->sv_timekeep_base);
1349 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1350 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1351 imgp->sysent->sv_stackprot);
1352 if (imgp->sysent->sv_hwcap != NULL)
1353 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1354 if (imgp->sysent->sv_hwcap2 != NULL)
1355 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1356 AUXARGS_ENTRY(pos, AT_NULL, 0);
1358 free(imgp->auxargs, M_TEMP);
1359 imgp->auxargs = NULL;
1360 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1362 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1363 free(argarray, M_TEMP);
1368 if (suword(base, imgp->args->argc) == -1)
1370 *stack_base = (register_t *)base;
1375 * Code for generating ELF core dumps.
1378 typedef void (*segment_callback)(vm_map_entry_t, void *);
1380 /* Closure for cb_put_phdr(). */
1381 struct phdr_closure {
1382 Elf_Phdr *phdr; /* Program header to fill in */
1383 Elf_Off offset; /* Offset of segment in core file */
1386 /* Closure for cb_size_segment(). */
1387 struct sseg_closure {
1388 int count; /* Count of writable segments. */
1389 size_t size; /* Total size of all writable segments. */
1392 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1395 int type; /* Note type. */
1396 outfunc_t outfunc; /* Output function. */
1397 void *outarg; /* Argument for the output function. */
1398 size_t outsize; /* Output size. */
1399 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1402 TAILQ_HEAD(note_info_list, note_info);
1404 /* Coredump output parameters. */
1405 struct coredump_params {
1407 struct ucred *active_cred;
1408 struct ucred *file_cred;
1411 struct compressor *comp;
1414 extern int compress_user_cores;
1415 extern int compress_user_cores_level;
1417 static void cb_put_phdr(vm_map_entry_t, void *);
1418 static void cb_size_segment(vm_map_entry_t, void *);
1419 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1421 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1422 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1423 struct note_info_list *, size_t);
1424 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1426 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1427 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1428 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1429 static int sbuf_drain_core_output(void *, const char *, int);
1431 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1434 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1435 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1436 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1437 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1438 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1439 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1440 static void note_procstat_files(void *, struct sbuf *, size_t *);
1441 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1442 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1443 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1444 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1445 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1448 * Write out a core segment to the compression stream.
1451 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1457 chunk_len = MIN(len, CORE_BUF_SIZE);
1460 * We can get EFAULT error here.
1461 * In that case zero out the current chunk of the segment.
1463 error = copyin(base, buf, chunk_len);
1465 bzero(buf, chunk_len);
1466 error = compressor_write(p->comp, buf, chunk_len);
1476 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1479 return (core_write((struct coredump_params *)arg, base, len, offset,
1484 core_write(struct coredump_params *p, const void *base, size_t len,
1485 off_t offset, enum uio_seg seg)
1488 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1489 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1490 p->active_cred, p->file_cred, NULL, p->td));
1494 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1499 if (p->comp != NULL)
1500 return (compress_chunk(p, base, tmpbuf, len));
1503 * EFAULT is a non-fatal error that we can get, for example,
1504 * if the segment is backed by a file but extends beyond its
1507 error = core_write(p, base, len, offset, UIO_USERSPACE);
1508 if (error == EFAULT) {
1509 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1510 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1511 "for process %s\n", base, len, offset, curproc->p_comm);
1514 * Write a "real" zero byte at the end of the target region
1515 * in the case this is the last segment.
1516 * The intermediate space will be implicitly zero-filled.
1518 error = core_write(p, zero_region, 1, offset + len - 1,
1525 * Drain into a core file.
1528 sbuf_drain_core_output(void *arg, const char *data, int len)
1530 struct coredump_params *p;
1533 p = (struct coredump_params *)arg;
1536 * Some kern_proc out routines that print to this sbuf may
1537 * call us with the process lock held. Draining with the
1538 * non-sleepable lock held is unsafe. The lock is needed for
1539 * those routines when dumping a live process. In our case we
1540 * can safely release the lock before draining and acquire
1543 locked = PROC_LOCKED(p->td->td_proc);
1545 PROC_UNLOCK(p->td->td_proc);
1546 if (p->comp != NULL)
1547 error = compressor_write(p->comp, __DECONST(char *, data), len);
1549 error = core_write(p, __DECONST(void *, data), len, p->offset,
1552 PROC_LOCK(p->td->td_proc);
1560 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1562 struct ucred *cred = td->td_ucred;
1564 struct sseg_closure seginfo;
1565 struct note_info_list notelst;
1566 struct coredump_params params;
1567 struct note_info *ninfo;
1569 size_t hdrsize, notesz, coresize;
1573 TAILQ_INIT(¬elst);
1575 /* Size the program segments. */
1578 each_dumpable_segment(td, cb_size_segment, &seginfo);
1581 * Collect info about the core file header area.
1583 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1584 if (seginfo.count + 1 >= PN_XNUM)
1585 hdrsize += sizeof(Elf_Shdr);
1586 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1587 coresize = round_page(hdrsize + notesz) + seginfo.size;
1589 /* Set up core dump parameters. */
1591 params.active_cred = cred;
1592 params.file_cred = NOCRED;
1599 PROC_LOCK(td->td_proc);
1600 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1601 PROC_UNLOCK(td->td_proc);
1608 if (coresize >= limit) {
1613 /* Create a compression stream if necessary. */
1614 if (compress_user_cores != 0) {
1615 params.comp = compressor_init(core_compressed_write,
1616 compress_user_cores, CORE_BUF_SIZE,
1617 compress_user_cores_level, ¶ms);
1618 if (params.comp == NULL) {
1622 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1626 * Allocate memory for building the header, fill it up,
1627 * and write it out following the notes.
1629 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1630 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1633 /* Write the contents of all of the writable segments. */
1639 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1640 offset = round_page(hdrsize + notesz);
1641 for (i = 0; i < seginfo.count; i++) {
1642 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1643 php->p_filesz, offset, ¶ms, tmpbuf);
1646 offset += php->p_filesz;
1649 if (error == 0 && params.comp != NULL)
1650 error = compressor_flush(params.comp);
1654 "Failed to write core file for process %s (error %d)\n",
1655 curproc->p_comm, error);
1659 free(tmpbuf, M_TEMP);
1660 if (params.comp != NULL)
1661 compressor_fini(params.comp);
1662 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1663 TAILQ_REMOVE(¬elst, ninfo, link);
1664 free(ninfo, M_TEMP);
1673 * A callback for each_dumpable_segment() to write out the segment's
1674 * program header entry.
1677 cb_put_phdr(vm_map_entry_t entry, void *closure)
1679 struct phdr_closure *phc = (struct phdr_closure *)closure;
1680 Elf_Phdr *phdr = phc->phdr;
1682 phc->offset = round_page(phc->offset);
1684 phdr->p_type = PT_LOAD;
1685 phdr->p_offset = phc->offset;
1686 phdr->p_vaddr = entry->start;
1688 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1689 phdr->p_align = PAGE_SIZE;
1690 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1692 phc->offset += phdr->p_filesz;
1697 * A callback for each_dumpable_segment() to gather information about
1698 * the number of segments and their total size.
1701 cb_size_segment(vm_map_entry_t entry, void *closure)
1703 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1706 ssc->size += entry->end - entry->start;
1710 * For each writable segment in the process's memory map, call the given
1711 * function with a pointer to the map entry and some arbitrary
1712 * caller-supplied data.
1715 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1717 struct proc *p = td->td_proc;
1718 vm_map_t map = &p->p_vmspace->vm_map;
1719 vm_map_entry_t entry;
1720 vm_object_t backing_object, object;
1721 boolean_t ignore_entry;
1723 vm_map_lock_read(map);
1724 for (entry = map->header.next; entry != &map->header;
1725 entry = entry->next) {
1727 * Don't dump inaccessible mappings, deal with legacy
1730 * Note that read-only segments related to the elf binary
1731 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1732 * need to arbitrarily ignore such segments.
1734 if (elf_legacy_coredump) {
1735 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1738 if ((entry->protection & VM_PROT_ALL) == 0)
1743 * Dont include memory segment in the coredump if
1744 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1745 * madvise(2). Do not dump submaps (i.e. parts of the
1748 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1751 if ((object = entry->object.vm_object) == NULL)
1754 /* Ignore memory-mapped devices and such things. */
1755 VM_OBJECT_RLOCK(object);
1756 while ((backing_object = object->backing_object) != NULL) {
1757 VM_OBJECT_RLOCK(backing_object);
1758 VM_OBJECT_RUNLOCK(object);
1759 object = backing_object;
1761 ignore_entry = object->type != OBJT_DEFAULT &&
1762 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1763 object->type != OBJT_PHYS;
1764 VM_OBJECT_RUNLOCK(object);
1768 (*func)(entry, closure);
1770 vm_map_unlock_read(map);
1774 * Write the core file header to the file, including padding up to
1775 * the page boundary.
1778 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1779 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1781 struct note_info *ninfo;
1785 /* Fill in the header. */
1786 bzero(hdr, hdrsize);
1787 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1789 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1790 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1791 sbuf_start_section(sb, NULL);
1792 sbuf_bcat(sb, hdr, hdrsize);
1793 TAILQ_FOREACH(ninfo, notelst, link)
1794 __elfN(putnote)(ninfo, sb);
1795 /* Align up to a page boundary for the program segments. */
1796 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1797 error = sbuf_finish(sb);
1804 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1814 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1817 * To have the debugger select the right thread (LWP) as the initial
1818 * thread, we dump the state of the thread passed to us in td first.
1819 * This is the thread that causes the core dump and thus likely to
1820 * be the right thread one wants to have selected in the debugger.
1823 while (thr != NULL) {
1824 size += register_note(list, NT_PRSTATUS,
1825 __elfN(note_prstatus), thr);
1826 size += register_note(list, NT_FPREGSET,
1827 __elfN(note_fpregset), thr);
1828 size += register_note(list, NT_THRMISC,
1829 __elfN(note_thrmisc), thr);
1830 size += register_note(list, NT_PTLWPINFO,
1831 __elfN(note_ptlwpinfo), thr);
1832 size += register_note(list, -1,
1833 __elfN(note_threadmd), thr);
1835 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1836 TAILQ_NEXT(thr, td_plist);
1838 thr = TAILQ_NEXT(thr, td_plist);
1841 size += register_note(list, NT_PROCSTAT_PROC,
1842 __elfN(note_procstat_proc), p);
1843 size += register_note(list, NT_PROCSTAT_FILES,
1844 note_procstat_files, p);
1845 size += register_note(list, NT_PROCSTAT_VMMAP,
1846 note_procstat_vmmap, p);
1847 size += register_note(list, NT_PROCSTAT_GROUPS,
1848 note_procstat_groups, p);
1849 size += register_note(list, NT_PROCSTAT_UMASK,
1850 note_procstat_umask, p);
1851 size += register_note(list, NT_PROCSTAT_RLIMIT,
1852 note_procstat_rlimit, p);
1853 size += register_note(list, NT_PROCSTAT_OSREL,
1854 note_procstat_osrel, p);
1855 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1856 __elfN(note_procstat_psstrings), p);
1857 size += register_note(list, NT_PROCSTAT_AUXV,
1858 __elfN(note_procstat_auxv), p);
1864 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1870 struct phdr_closure phc;
1872 ehdr = (Elf_Ehdr *)hdr;
1874 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1875 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1876 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1877 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1878 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1879 ehdr->e_ident[EI_DATA] = ELF_DATA;
1880 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1881 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1882 ehdr->e_ident[EI_ABIVERSION] = 0;
1883 ehdr->e_ident[EI_PAD] = 0;
1884 ehdr->e_type = ET_CORE;
1885 ehdr->e_machine = td->td_proc->p_elf_machine;
1886 ehdr->e_version = EV_CURRENT;
1888 ehdr->e_phoff = sizeof(Elf_Ehdr);
1889 ehdr->e_flags = td->td_proc->p_elf_flags;
1890 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1891 ehdr->e_phentsize = sizeof(Elf_Phdr);
1892 ehdr->e_shentsize = sizeof(Elf_Shdr);
1893 ehdr->e_shstrndx = SHN_UNDEF;
1894 if (numsegs + 1 < PN_XNUM) {
1895 ehdr->e_phnum = numsegs + 1;
1898 ehdr->e_phnum = PN_XNUM;
1901 ehdr->e_shoff = ehdr->e_phoff +
1902 (numsegs + 1) * ehdr->e_phentsize;
1903 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1904 ("e_shoff: %zu, hdrsize - shdr: %zu",
1905 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1907 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1908 memset(shdr, 0, sizeof(*shdr));
1910 * A special first section is used to hold large segment and
1911 * section counts. This was proposed by Sun Microsystems in
1912 * Solaris and has been adopted by Linux; the standard ELF
1913 * tools are already familiar with the technique.
1915 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1916 * (or 12-7 depending on the version of the document) for more
1919 shdr->sh_type = SHT_NULL;
1920 shdr->sh_size = ehdr->e_shnum;
1921 shdr->sh_link = ehdr->e_shstrndx;
1922 shdr->sh_info = numsegs + 1;
1926 * Fill in the program header entries.
1928 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1930 /* The note segement. */
1931 phdr->p_type = PT_NOTE;
1932 phdr->p_offset = hdrsize;
1935 phdr->p_filesz = notesz;
1937 phdr->p_flags = PF_R;
1938 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1941 /* All the writable segments from the program. */
1943 phc.offset = round_page(hdrsize + notesz);
1944 each_dumpable_segment(td, cb_put_phdr, &phc);
1948 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1950 struct note_info *ninfo;
1951 size_t size, notesize;
1954 out(arg, NULL, &size);
1955 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1957 ninfo->outfunc = out;
1958 ninfo->outarg = arg;
1959 ninfo->outsize = size;
1960 TAILQ_INSERT_TAIL(list, ninfo, link);
1965 notesize = sizeof(Elf_Note) + /* note header */
1966 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1968 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1974 append_note_data(const void *src, void *dst, size_t len)
1978 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1980 bcopy(src, dst, len);
1981 bzero((char *)dst + len, padded_len - len);
1983 return (padded_len);
1987 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1995 note = (Elf_Note *)buf;
1996 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1997 note->n_descsz = size;
1998 note->n_type = type;
1999 buf += sizeof(*note);
2000 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2001 sizeof(FREEBSD_ABI_VENDOR));
2002 append_note_data(src, buf, size);
2007 notesize = sizeof(Elf_Note) + /* note header */
2008 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2010 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2016 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2019 ssize_t old_len, sect_len;
2020 size_t new_len, descsz, i;
2022 if (ninfo->type == -1) {
2023 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2027 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2028 note.n_descsz = ninfo->outsize;
2029 note.n_type = ninfo->type;
2031 sbuf_bcat(sb, ¬e, sizeof(note));
2032 sbuf_start_section(sb, &old_len);
2033 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2034 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2035 if (note.n_descsz == 0)
2037 sbuf_start_section(sb, &old_len);
2038 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2039 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2043 new_len = (size_t)sect_len;
2044 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2045 if (new_len < descsz) {
2047 * It is expected that individual note emitters will correctly
2048 * predict their expected output size and fill up to that size
2049 * themselves, padding in a format-specific way if needed.
2050 * However, in case they don't, just do it here with zeros.
2052 for (i = 0; i < descsz - new_len; i++)
2054 } else if (new_len > descsz) {
2056 * We can't always truncate sb -- we may have drained some
2059 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2060 "read it (%zu > %zu). Since it is longer than "
2061 "expected, this coredump's notes are corrupt. THIS "
2062 "IS A BUG in the note_procstat routine for type %u.\n",
2063 __func__, (unsigned)note.n_type, new_len, descsz,
2064 (unsigned)note.n_type));
2069 * Miscellaneous note out functions.
2072 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2073 #include <compat/freebsd32/freebsd32.h>
2074 #include <compat/freebsd32/freebsd32_signal.h>
2076 typedef struct prstatus32 elf_prstatus_t;
2077 typedef struct prpsinfo32 elf_prpsinfo_t;
2078 typedef struct fpreg32 elf_prfpregset_t;
2079 typedef struct fpreg32 elf_fpregset_t;
2080 typedef struct reg32 elf_gregset_t;
2081 typedef struct thrmisc32 elf_thrmisc_t;
2082 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2083 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2084 typedef uint32_t elf_ps_strings_t;
2086 typedef prstatus_t elf_prstatus_t;
2087 typedef prpsinfo_t elf_prpsinfo_t;
2088 typedef prfpregset_t elf_prfpregset_t;
2089 typedef prfpregset_t elf_fpregset_t;
2090 typedef gregset_t elf_gregset_t;
2091 typedef thrmisc_t elf_thrmisc_t;
2092 #define ELF_KERN_PROC_MASK 0
2093 typedef struct kinfo_proc elf_kinfo_proc_t;
2094 typedef vm_offset_t elf_ps_strings_t;
2098 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2104 elf_prpsinfo_t *psinfo;
2107 p = (struct proc *)arg;
2109 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2110 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2111 psinfo->pr_version = PRPSINFO_VERSION;
2112 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2113 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2115 if (p->p_args != NULL) {
2116 len = sizeof(psinfo->pr_psargs) - 1;
2117 if (len > p->p_args->ar_length)
2118 len = p->p_args->ar_length;
2119 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2125 sbuf_new(&sbarg, psinfo->pr_psargs,
2126 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2127 error = proc_getargv(curthread, p, &sbarg);
2129 if (sbuf_finish(&sbarg) == 0)
2130 len = sbuf_len(&sbarg) - 1;
2132 len = sizeof(psinfo->pr_psargs) - 1;
2133 sbuf_delete(&sbarg);
2135 if (error || len == 0)
2136 strlcpy(psinfo->pr_psargs, p->p_comm,
2137 sizeof(psinfo->pr_psargs));
2139 KASSERT(len < sizeof(psinfo->pr_psargs),
2140 ("len is too long: %zu vs %zu", len,
2141 sizeof(psinfo->pr_psargs)));
2142 cp = psinfo->pr_psargs;
2145 cp = memchr(cp, '\0', end - cp);
2151 psinfo->pr_pid = p->p_pid;
2152 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2153 free(psinfo, M_TEMP);
2155 *sizep = sizeof(*psinfo);
2159 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2162 elf_prstatus_t *status;
2164 td = (struct thread *)arg;
2166 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2167 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2168 status->pr_version = PRSTATUS_VERSION;
2169 status->pr_statussz = sizeof(elf_prstatus_t);
2170 status->pr_gregsetsz = sizeof(elf_gregset_t);
2171 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2172 status->pr_osreldate = osreldate;
2173 status->pr_cursig = td->td_proc->p_sig;
2174 status->pr_pid = td->td_tid;
2175 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2176 fill_regs32(td, &status->pr_reg);
2178 fill_regs(td, &status->pr_reg);
2180 sbuf_bcat(sb, status, sizeof(*status));
2181 free(status, M_TEMP);
2183 *sizep = sizeof(*status);
2187 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2190 elf_prfpregset_t *fpregset;
2192 td = (struct thread *)arg;
2194 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2195 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2196 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2197 fill_fpregs32(td, fpregset);
2199 fill_fpregs(td, fpregset);
2201 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2202 free(fpregset, M_TEMP);
2204 *sizep = sizeof(*fpregset);
2208 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2211 elf_thrmisc_t thrmisc;
2213 td = (struct thread *)arg;
2215 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2216 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2217 strcpy(thrmisc.pr_tname, td->td_name);
2218 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2220 *sizep = sizeof(thrmisc);
2224 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2229 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2230 struct ptrace_lwpinfo32 pl;
2232 struct ptrace_lwpinfo pl;
2235 td = (struct thread *)arg;
2236 size = sizeof(structsize) + sizeof(pl);
2238 KASSERT(*sizep == size, ("invalid size"));
2239 structsize = sizeof(pl);
2240 sbuf_bcat(sb, &structsize, sizeof(structsize));
2241 bzero(&pl, sizeof(pl));
2242 pl.pl_lwpid = td->td_tid;
2243 pl.pl_event = PL_EVENT_NONE;
2244 pl.pl_sigmask = td->td_sigmask;
2245 pl.pl_siglist = td->td_siglist;
2246 if (td->td_si.si_signo != 0) {
2247 pl.pl_event = PL_EVENT_SIGNAL;
2248 pl.pl_flags |= PL_FLAG_SI;
2249 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2250 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2252 pl.pl_siginfo = td->td_si;
2255 strcpy(pl.pl_tdname, td->td_name);
2256 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2257 sbuf_bcat(sb, &pl, sizeof(pl));
2263 * Allow for MD specific notes, as well as any MD
2264 * specific preparations for writing MI notes.
2267 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2273 td = (struct thread *)arg;
2275 if (size != 0 && sb != NULL)
2276 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2280 __elfN(dump_thread)(td, buf, &size);
2281 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2282 if (size != 0 && sb != NULL)
2283 sbuf_bcat(sb, buf, size);
2288 #ifdef KINFO_PROC_SIZE
2289 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2293 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2299 p = (struct proc *)arg;
2300 size = sizeof(structsize) + p->p_numthreads *
2301 sizeof(elf_kinfo_proc_t);
2304 KASSERT(*sizep == size, ("invalid size"));
2305 structsize = sizeof(elf_kinfo_proc_t);
2306 sbuf_bcat(sb, &structsize, sizeof(structsize));
2308 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2313 #ifdef KINFO_FILE_SIZE
2314 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2318 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2321 size_t size, sect_sz, i;
2322 ssize_t start_len, sect_len;
2323 int structsize, filedesc_flags;
2325 if (coredump_pack_fileinfo)
2326 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2330 p = (struct proc *)arg;
2331 structsize = sizeof(struct kinfo_file);
2334 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2335 sbuf_set_drain(sb, sbuf_count_drain, &size);
2336 sbuf_bcat(sb, &structsize, sizeof(structsize));
2338 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2343 sbuf_start_section(sb, &start_len);
2345 sbuf_bcat(sb, &structsize, sizeof(structsize));
2347 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2350 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2355 KASSERT(sect_sz <= *sizep,
2356 ("kern_proc_filedesc_out did not respect maxlen; "
2357 "requested %zu, got %zu", *sizep - sizeof(structsize),
2358 sect_sz - sizeof(structsize)));
2360 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2365 #ifdef KINFO_VMENTRY_SIZE
2366 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2370 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2374 int structsize, vmmap_flags;
2376 if (coredump_pack_vmmapinfo)
2377 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2381 p = (struct proc *)arg;
2382 structsize = sizeof(struct kinfo_vmentry);
2385 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2386 sbuf_set_drain(sb, sbuf_count_drain, &size);
2387 sbuf_bcat(sb, &structsize, sizeof(structsize));
2389 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2394 sbuf_bcat(sb, &structsize, sizeof(structsize));
2396 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2402 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2408 p = (struct proc *)arg;
2409 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2411 KASSERT(*sizep == size, ("invalid size"));
2412 structsize = sizeof(gid_t);
2413 sbuf_bcat(sb, &structsize, sizeof(structsize));
2414 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2421 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2427 p = (struct proc *)arg;
2428 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2430 KASSERT(*sizep == size, ("invalid size"));
2431 structsize = sizeof(p->p_fd->fd_cmask);
2432 sbuf_bcat(sb, &structsize, sizeof(structsize));
2433 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2439 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2442 struct rlimit rlim[RLIM_NLIMITS];
2446 p = (struct proc *)arg;
2447 size = sizeof(structsize) + sizeof(rlim);
2449 KASSERT(*sizep == size, ("invalid size"));
2450 structsize = sizeof(rlim);
2451 sbuf_bcat(sb, &structsize, sizeof(structsize));
2453 for (i = 0; i < RLIM_NLIMITS; i++)
2454 lim_rlimit_proc(p, i, &rlim[i]);
2456 sbuf_bcat(sb, rlim, sizeof(rlim));
2462 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2468 p = (struct proc *)arg;
2469 size = sizeof(structsize) + sizeof(p->p_osrel);
2471 KASSERT(*sizep == size, ("invalid size"));
2472 structsize = sizeof(p->p_osrel);
2473 sbuf_bcat(sb, &structsize, sizeof(structsize));
2474 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2480 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2483 elf_ps_strings_t ps_strings;
2487 p = (struct proc *)arg;
2488 size = sizeof(structsize) + sizeof(ps_strings);
2490 KASSERT(*sizep == size, ("invalid size"));
2491 structsize = sizeof(ps_strings);
2492 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2493 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2495 ps_strings = p->p_sysent->sv_psstrings;
2497 sbuf_bcat(sb, &structsize, sizeof(structsize));
2498 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2504 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2510 p = (struct proc *)arg;
2513 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2514 sbuf_set_drain(sb, sbuf_count_drain, &size);
2515 sbuf_bcat(sb, &structsize, sizeof(structsize));
2517 proc_getauxv(curthread, p, sb);
2523 structsize = sizeof(Elf_Auxinfo);
2524 sbuf_bcat(sb, &structsize, sizeof(structsize));
2526 proc_getauxv(curthread, p, sb);
2532 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2533 const char *note_vendor, const Elf_Phdr *pnote,
2534 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2536 const Elf_Note *note, *note0, *note_end;
2537 const char *note_name;
2542 /* We need some limit, might as well use PAGE_SIZE. */
2543 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2545 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2546 if (pnote->p_offset > PAGE_SIZE ||
2547 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2548 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2550 VOP_UNLOCK(imgp->vp, 0);
2551 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2552 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2554 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2555 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2556 curthread->td_ucred, NOCRED, NULL, curthread);
2558 uprintf("i/o error PT_NOTE\n");
2561 note = note0 = (const Elf_Note *)buf;
2562 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2564 note = note0 = (const Elf_Note *)(imgp->image_header +
2566 note_end = (const Elf_Note *)(imgp->image_header +
2567 pnote->p_offset + pnote->p_filesz);
2570 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2571 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2572 (const char *)note < sizeof(Elf_Note)) {
2575 if (note->n_namesz != checknote->n_namesz ||
2576 note->n_descsz != checknote->n_descsz ||
2577 note->n_type != checknote->n_type)
2579 note_name = (const char *)(note + 1);
2580 if (note_name + checknote->n_namesz >=
2581 (const char *)note_end || strncmp(note_vendor,
2582 note_name, checknote->n_namesz) != 0)
2585 if (cb(note, cb_arg, &res))
2588 note = (const Elf_Note *)((const char *)(note + 1) +
2589 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2590 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2599 struct brandnote_cb_arg {
2600 Elf_Brandnote *brandnote;
2605 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2607 struct brandnote_cb_arg *arg;
2612 * Fetch the osreldate for binary from the ELF OSABI-note if
2615 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2616 arg->brandnote->trans_osrel != NULL ?
2617 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2622 static Elf_Note fctl_note = {
2623 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2624 .n_descsz = sizeof(uint32_t),
2625 .n_type = NT_FREEBSD_FEATURE_CTL,
2628 struct fctl_cb_arg {
2633 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2635 struct fctl_cb_arg *arg;
2636 const Elf32_Word *desc;
2640 p = (uintptr_t)(note + 1);
2641 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2642 desc = (const Elf32_Word *)p;
2643 *arg->fctl0 = desc[0];
2648 * Try to find the appropriate ABI-note section for checknote, fetch
2649 * the osreldate and feature control flags for binary from the ELF
2650 * OSABI-note. Only the first page of the image is searched, the same
2654 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2655 int32_t *osrel, uint32_t *fctl0)
2657 const Elf_Phdr *phdr;
2658 const Elf_Ehdr *hdr;
2659 struct brandnote_cb_arg b_arg;
2660 struct fctl_cb_arg f_arg;
2663 hdr = (const Elf_Ehdr *)imgp->image_header;
2664 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2665 b_arg.brandnote = brandnote;
2666 b_arg.osrel = osrel;
2667 f_arg.fctl0 = fctl0;
2669 for (i = 0; i < hdr->e_phnum; i++) {
2670 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2671 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2673 for (j = 0; j < hdr->e_phnum; j++) {
2674 if (phdr[j].p_type == PT_NOTE &&
2675 __elfN(parse_notes)(imgp, &fctl_note,
2676 FREEBSD_ABI_VENDOR, &phdr[j],
2677 note_fctl_cb, &f_arg))
2688 * Tell kern_execve.c about it, with a little help from the linker.
2690 static struct execsw __elfN(execsw) = {
2691 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2692 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2694 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2697 __elfN(trans_prot)(Elf_Word flags)
2703 prot |= VM_PROT_EXECUTE;
2705 prot |= VM_PROT_WRITE;
2707 prot |= VM_PROT_READ;
2708 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2709 if (i386_read_exec && (flags & PF_R))
2710 prot |= VM_PROT_EXECUTE;
2716 __elfN(untrans_prot)(vm_prot_t prot)
2721 if (prot & VM_PROT_EXECUTE)
2723 if (prot & VM_PROT_READ)
2725 if (prot & VM_PROT_WRITE)
2731 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2733 u_long range, rbase, gap;
2736 if ((imgp->map_flags & MAP_ASLR) == 0)
2738 pct = __elfN(aslr_stack_gap);
2743 range = imgp->eff_stack_sz * pct / 100;
2744 arc4rand(&rbase, sizeof(rbase), 0);
2745 gap = rbase % range;
2746 gap &= ~(sizeof(u_long) - 1);