2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
60 #include <sys/sf_buf.h>
62 #include <sys/systm.h>
63 #include <sys/signalvar.h>
66 #include <sys/syscall.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/vnode.h>
70 #include <sys/syslog.h>
71 #include <sys/eventhandler.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
85 #define ELF_NOTE_ROUNDSIZE 4
86 #define OLD_EI_BRAND 8
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90 const char *interp, int32_t *osrel, uint32_t *fctl0);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
94 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
106 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
132 int i386_read_exec = 0;
133 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
134 "enable execution from readable segments");
137 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
139 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
144 val = __elfN(pie_base);
145 error = sysctl_handle_long(oidp, &val, 0, req);
146 if (error != 0 || req->newptr == NULL)
148 if ((val & PAGE_MASK) != 0)
150 __elfN(pie_base) = val;
153 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
154 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
155 sysctl_pie_base, "LU",
156 "PIE load base without randomization");
158 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
159 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
161 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
163 static int __elfN(aslr_enabled) = 0;
164 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
165 &__elfN(aslr_enabled), 0,
166 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
167 ": enable address map randomization");
169 static int __elfN(pie_aslr_enabled) = 0;
170 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
171 &__elfN(pie_aslr_enabled), 0,
172 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
173 ": enable address map randomization for PIE binaries");
175 static int __elfN(aslr_honor_sbrk) = 1;
176 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
177 &__elfN(aslr_honor_sbrk), 0,
178 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
180 static int __elfN(aslr_stack_gap) = 3;
181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
182 &__elfN(aslr_stack_gap), 0,
183 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
184 ": maximum percentage of main stack to waste on a random gap");
186 static int __elfN(sigfastblock) = 1;
187 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
188 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
189 "enable sigfastblock for new processes");
191 static bool __elfN(allow_wx) = true;
192 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
193 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
194 "Allow pages to be mapped simultaneously writable and executable");
196 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
198 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
200 Elf_Brandnote __elfN(freebsd_brandnote) = {
201 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
202 .hdr.n_descsz = sizeof(int32_t),
203 .hdr.n_type = NT_FREEBSD_ABI_TAG,
204 .vendor = FREEBSD_ABI_VENDOR,
205 .flags = BN_TRANSLATE_OSREL,
206 .trans_osrel = __elfN(freebsd_trans_osrel)
210 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
214 p = (uintptr_t)(note + 1);
215 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
216 *osrel = *(const int32_t *)(p);
221 static const char GNU_ABI_VENDOR[] = "GNU";
222 static int GNU_KFREEBSD_ABI_DESC = 3;
224 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
225 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
226 .hdr.n_descsz = 16, /* XXX at least 16 */
228 .vendor = GNU_ABI_VENDOR,
229 .flags = BN_TRANSLATE_OSREL,
230 .trans_osrel = kfreebsd_trans_osrel
234 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
236 const Elf32_Word *desc;
239 p = (uintptr_t)(note + 1);
240 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
242 desc = (const Elf32_Word *)p;
243 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
247 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
248 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
250 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
256 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
260 for (i = 0; i < MAX_BRANDS; i++) {
261 if (elf_brand_list[i] == NULL) {
262 elf_brand_list[i] = entry;
266 if (i == MAX_BRANDS) {
267 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
275 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
279 for (i = 0; i < MAX_BRANDS; i++) {
280 if (elf_brand_list[i] == entry) {
281 elf_brand_list[i] = NULL;
291 __elfN(brand_inuse)(Elf_Brandinfo *entry)
296 sx_slock(&allproc_lock);
297 FOREACH_PROC_IN_SYSTEM(p) {
298 if (p->p_sysent == entry->sysvec) {
303 sx_sunlock(&allproc_lock);
308 static Elf_Brandinfo *
309 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
310 int32_t *osrel, uint32_t *fctl0)
312 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
313 Elf_Brandinfo *bi, *bi_m;
314 boolean_t ret, has_fctl0;
315 int i, interp_name_len;
317 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
320 * We support four types of branding -- (1) the ELF EI_OSABI field
321 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
322 * branding w/in the ELF header, (3) path of the `interp_path'
323 * field, and (4) the ".note.ABI-tag" ELF section.
326 /* Look for an ".note.ABI-tag" ELF section */
328 for (i = 0; i < MAX_BRANDS; i++) {
329 bi = elf_brand_list[i];
332 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
334 if (hdr->e_machine == bi->machine && (bi->flags &
335 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
339 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
341 /* Give brand a chance to veto check_note's guess */
342 if (ret && bi->header_supported) {
343 ret = bi->header_supported(imgp, osrel,
344 has_fctl0 ? fctl0 : NULL);
347 * If note checker claimed the binary, but the
348 * interpreter path in the image does not
349 * match default one for the brand, try to
350 * search for other brands with the same
351 * interpreter. Either there is better brand
352 * with the right interpreter, or, failing
353 * this, we return first brand which accepted
354 * our note and, optionally, header.
356 if (ret && bi_m == NULL && interp != NULL &&
357 (bi->interp_path == NULL ||
358 (strlen(bi->interp_path) + 1 != interp_name_len ||
359 strncmp(interp, bi->interp_path, interp_name_len)
371 /* If the executable has a brand, search for it in the brand list. */
372 for (i = 0; i < MAX_BRANDS; i++) {
373 bi = elf_brand_list[i];
374 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
375 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
377 if (hdr->e_machine == bi->machine &&
378 (hdr->e_ident[EI_OSABI] == bi->brand ||
379 (bi->compat_3_brand != NULL &&
380 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
381 bi->compat_3_brand) == 0))) {
382 /* Looks good, but give brand a chance to veto */
383 if (bi->header_supported == NULL ||
384 bi->header_supported(imgp, NULL, NULL)) {
386 * Again, prefer strictly matching
389 if (interp_name_len == 0 &&
390 bi->interp_path == NULL)
392 if (bi->interp_path != NULL &&
393 strlen(bi->interp_path) + 1 ==
394 interp_name_len && strncmp(interp,
395 bi->interp_path, interp_name_len) == 0)
405 /* No known brand, see if the header is recognized by any brand */
406 for (i = 0; i < MAX_BRANDS; i++) {
407 bi = elf_brand_list[i];
408 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
409 bi->header_supported == NULL)
411 if (hdr->e_machine == bi->machine) {
412 ret = bi->header_supported(imgp, NULL, NULL);
418 /* Lacking a known brand, search for a recognized interpreter. */
419 if (interp != NULL) {
420 for (i = 0; i < MAX_BRANDS; i++) {
421 bi = elf_brand_list[i];
422 if (bi == NULL || (bi->flags &
423 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
426 if (hdr->e_machine == bi->machine &&
427 bi->interp_path != NULL &&
428 /* ELF image p_filesz includes terminating zero */
429 strlen(bi->interp_path) + 1 == interp_name_len &&
430 strncmp(interp, bi->interp_path, interp_name_len)
431 == 0 && (bi->header_supported == NULL ||
432 bi->header_supported(imgp, NULL, NULL)))
437 /* Lacking a recognized interpreter, try the default brand */
438 for (i = 0; i < MAX_BRANDS; i++) {
439 bi = elf_brand_list[i];
440 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
441 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
443 if (hdr->e_machine == bi->machine &&
444 __elfN(fallback_brand) == bi->brand &&
445 (bi->header_supported == NULL ||
446 bi->header_supported(imgp, NULL, NULL)))
453 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
455 return (hdr->e_phoff <= PAGE_SIZE &&
456 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
460 __elfN(check_header)(const Elf_Ehdr *hdr)
466 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
467 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
468 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
469 hdr->e_phentsize != sizeof(Elf_Phdr) ||
470 hdr->e_version != ELF_TARG_VER)
474 * Make sure we have at least one brand for this machine.
477 for (i = 0; i < MAX_BRANDS; i++) {
478 bi = elf_brand_list[i];
479 if (bi != NULL && bi->machine == hdr->e_machine)
489 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
490 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
497 * Create the page if it doesn't exist yet. Ignore errors.
499 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
500 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
503 * Find the page from the underlying object.
505 if (object != NULL) {
506 sf = vm_imgact_map_page(object, offset);
508 return (KERN_FAILURE);
509 off = offset - trunc_page(offset);
510 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
512 vm_imgact_unmap_page(sf);
514 return (KERN_FAILURE);
517 return (KERN_SUCCESS);
521 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
522 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
528 int error, locked, rv;
530 if (start != trunc_page(start)) {
531 rv = __elfN(map_partial)(map, object, offset, start,
532 round_page(start), prot);
533 if (rv != KERN_SUCCESS)
535 offset += round_page(start) - start;
536 start = round_page(start);
538 if (end != round_page(end)) {
539 rv = __elfN(map_partial)(map, object, offset +
540 trunc_page(end) - start, trunc_page(end), end, prot);
541 if (rv != KERN_SUCCESS)
543 end = trunc_page(end);
546 return (KERN_SUCCESS);
547 if ((offset & PAGE_MASK) != 0) {
549 * The mapping is not page aligned. This means that we have
552 rv = vm_map_fixed(map, NULL, 0, start, end - start,
553 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
554 if (rv != KERN_SUCCESS)
557 return (KERN_SUCCESS);
558 for (; start < end; start += sz) {
559 sf = vm_imgact_map_page(object, offset);
561 return (KERN_FAILURE);
562 off = offset - trunc_page(offset);
564 if (sz > PAGE_SIZE - off)
565 sz = PAGE_SIZE - off;
566 error = copyout((caddr_t)sf_buf_kva(sf) + off,
568 vm_imgact_unmap_page(sf);
570 return (KERN_FAILURE);
574 vm_object_reference(object);
575 rv = vm_map_fixed(map, object, offset, start, end - start,
576 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
577 (object != NULL ? MAP_VN_EXEC : 0));
578 if (rv != KERN_SUCCESS) {
579 locked = VOP_ISLOCKED(imgp->vp);
580 VOP_UNLOCK(imgp->vp);
581 vm_object_deallocate(object);
582 vn_lock(imgp->vp, locked | LK_RETRY);
584 } else if (object != NULL) {
585 MPASS(imgp->vp->v_object == object);
586 VOP_SET_TEXT_CHECKED(imgp->vp);
589 return (KERN_SUCCESS);
593 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
594 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
600 vm_offset_t map_addr;
603 vm_ooffset_t file_addr;
606 * It's necessary to fail if the filsz + offset taken from the
607 * header is greater than the actual file pager object's size.
608 * If we were to allow this, then the vm_map_find() below would
609 * walk right off the end of the file object and into the ether.
611 * While I'm here, might as well check for something else that
612 * is invalid: filsz cannot be greater than memsz.
614 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
616 uprintf("elf_load_section: truncated ELF file\n");
620 object = imgp->object;
621 map = &imgp->proc->p_vmspace->vm_map;
622 map_addr = trunc_page((vm_offset_t)vmaddr);
623 file_addr = trunc_page(offset);
626 * We have two choices. We can either clear the data in the last page
627 * of an oversized mapping, or we can start the anon mapping a page
628 * early and copy the initialized data into that first page. We
633 else if (memsz > filsz)
634 map_len = trunc_page(offset + filsz) - file_addr;
636 map_len = round_page(offset + filsz) - file_addr;
639 /* cow flags: don't dump readonly sections in core */
640 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
641 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
643 rv = __elfN(map_insert)(imgp, map, object, file_addr,
644 map_addr, map_addr + map_len, prot, cow);
645 if (rv != KERN_SUCCESS)
648 /* we can stop now if we've covered it all */
654 * We have to get the remaining bit of the file into the first part
655 * of the oversized map segment. This is normally because the .data
656 * segment in the file is extended to provide bss. It's a neat idea
657 * to try and save a page, but it's a pain in the behind to implement.
659 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
661 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
662 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
664 /* This had damn well better be true! */
666 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
667 map_addr + map_len, prot, 0);
668 if (rv != KERN_SUCCESS)
673 sf = vm_imgact_map_page(object, offset + filsz);
677 /* send the page fragment to user space */
678 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
680 vm_imgact_unmap_page(sf);
686 * Remove write access to the page if it was only granted by map_insert
689 if ((prot & VM_PROT_WRITE) == 0)
690 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
691 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
697 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
698 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
705 ASSERT_VOP_LOCKED(imgp->vp, __func__);
710 for (i = 0; i < hdr->e_phnum; i++) {
711 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
714 /* Loadable segment */
715 prot = __elfN(trans_prot)(phdr[i].p_flags);
716 error = __elfN(load_section)(imgp, phdr[i].p_offset,
717 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
718 phdr[i].p_memsz, phdr[i].p_filesz, prot);
723 * Establish the base address if this is the first segment.
726 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
731 if (base_addrp != NULL)
732 *base_addrp = base_addr;
738 * Load the file "file" into memory. It may be either a shared object
741 * The "addr" reference parameter is in/out. On entry, it specifies
742 * the address where a shared object should be loaded. If the file is
743 * an executable, this value is ignored. On exit, "addr" specifies
744 * where the file was actually loaded.
746 * The "entry" reference parameter is out only. On exit, it specifies
747 * the entry point for the loaded file.
750 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
756 struct image_params image_params;
758 const Elf_Ehdr *hdr = NULL;
759 const Elf_Phdr *phdr = NULL;
760 struct nameidata *nd;
762 struct image_params *imgp;
764 u_long base_addr = 0;
767 #ifdef CAPABILITY_MODE
769 * XXXJA: This check can go away once we are sufficiently confident
770 * that the checks in namei() are correct.
772 if (IN_CAPABILITY_MODE(curthread))
776 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
778 attr = &tempdata->attr;
779 imgp = &tempdata->image_params;
782 * Initialize part of the common data
787 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
788 UIO_SYSSPACE, file, curthread);
789 if ((error = namei(nd)) != 0) {
793 NDFREE(nd, NDF_ONLY_PNBUF);
794 imgp->vp = nd->ni_vp;
797 * Check permissions, modes, uid, etc on the file, and "open" it.
799 error = exec_check_permissions(imgp);
803 error = exec_map_first_page(imgp);
807 imgp->object = nd->ni_vp->v_object;
809 hdr = (const Elf_Ehdr *)imgp->image_header;
810 if ((error = __elfN(check_header)(hdr)) != 0)
812 if (hdr->e_type == ET_DYN)
814 else if (hdr->e_type == ET_EXEC)
821 /* Only support headers that fit within first page for now */
822 if (!__elfN(phdr_in_zero_page)(hdr)) {
827 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
828 if (!aligned(phdr, Elf_Addr)) {
833 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
838 *entry = (unsigned long)hdr->e_entry + rbase;
842 exec_unmap_first_page(imgp);
846 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
849 free(tempdata, M_TEMP);
855 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
860 MPASS(vm_map_min(map) <= minv);
861 MPASS(maxv <= vm_map_max(map));
863 MPASS(minv + align < maxv);
864 arc4rand(&rbase, sizeof(rbase), 0);
865 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
866 res &= ~((u_long)align - 1);
870 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
871 res, minv, maxv, rbase));
873 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
874 res, maxv, minv, rbase));
879 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
880 const Elf_Phdr *phdr, u_long et_dyn_addr)
882 struct vmspace *vmspace;
884 u_long text_size, data_size, total_size, text_addr, data_addr;
885 u_long seg_size, seg_addr;
889 text_size = data_size = total_size = text_addr = data_addr = 0;
891 for (i = 0; i < hdr->e_phnum; i++) {
892 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
895 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
896 seg_size = round_page(phdr[i].p_memsz +
897 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
900 * Make the largest executable segment the official
901 * text segment and all others data.
903 * Note that obreak() assumes that data_addr + data_size == end
904 * of data load area, and the ELF file format expects segments
905 * to be sorted by address. If multiple data segments exist,
906 * the last one will be used.
909 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
910 text_size = seg_size;
911 text_addr = seg_addr;
913 data_size = seg_size;
914 data_addr = seg_addr;
916 total_size += seg_size;
919 if (data_addr == 0 && data_size == 0) {
920 data_addr = text_addr;
921 data_size = text_size;
925 * Check limits. It should be safe to check the
926 * limits after loading the segments since we do
927 * not actually fault in all the segments pages.
929 PROC_LOCK(imgp->proc);
930 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
931 err_str = "Data segment size exceeds process limit";
932 else if (text_size > maxtsiz)
933 err_str = "Text segment size exceeds system limit";
934 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
935 err_str = "Total segment size exceeds process limit";
936 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
937 err_str = "Data segment size exceeds resource limit";
938 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
939 err_str = "Total segment size exceeds resource limit";
940 PROC_UNLOCK(imgp->proc);
941 if (err_str != NULL) {
942 uprintf("%s\n", err_str);
946 vmspace = imgp->proc->p_vmspace;
947 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
948 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
949 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
950 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
956 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
957 char **interpp, bool *free_interpp)
961 int error, interp_name_len;
963 KASSERT(phdr->p_type == PT_INTERP,
964 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
965 ASSERT_VOP_LOCKED(imgp->vp, __func__);
969 /* Path to interpreter */
970 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
971 uprintf("Invalid PT_INTERP\n");
975 interp_name_len = phdr->p_filesz;
976 if (phdr->p_offset > PAGE_SIZE ||
977 interp_name_len > PAGE_SIZE - phdr->p_offset) {
979 * The vnode lock might be needed by the pagedaemon to
980 * clean pages owned by the vnode. Do not allow sleep
981 * waiting for memory with the vnode locked, instead
982 * try non-sleepable allocation first, and if it
983 * fails, go to the slow path were we drop the lock
984 * and do M_WAITOK. A text reference prevents
985 * modifications to the vnode content.
987 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
988 if (interp == NULL) {
989 VOP_UNLOCK(imgp->vp);
990 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
991 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
994 error = vn_rdwr(UIO_READ, imgp->vp, interp,
995 interp_name_len, phdr->p_offset,
996 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
999 free(interp, M_TEMP);
1000 uprintf("i/o error PT_INTERP %d\n", error);
1003 interp[interp_name_len] = '\0';
1006 *free_interpp = true;
1010 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1011 if (interp[interp_name_len - 1] != '\0') {
1012 uprintf("Invalid PT_INTERP\n");
1017 *free_interpp = false;
1022 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1023 const char *interp, u_long *addr, u_long *entry)
1028 if (brand_info->emul_path != NULL &&
1029 brand_info->emul_path[0] != '\0') {
1030 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1031 snprintf(path, MAXPATHLEN, "%s%s",
1032 brand_info->emul_path, interp);
1033 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1039 if (brand_info->interp_newpath != NULL &&
1040 (brand_info->interp_path == NULL ||
1041 strcmp(interp, brand_info->interp_path) == 0)) {
1042 error = __elfN(load_file)(imgp->proc,
1043 brand_info->interp_newpath, addr, entry);
1048 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1052 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1057 * Impossible et_dyn_addr initial value indicating that the real base
1058 * must be calculated later with some randomization applied.
1060 #define ET_DYN_ADDR_RAND 1
1063 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1066 const Elf_Ehdr *hdr;
1067 const Elf_Phdr *phdr;
1068 Elf_Auxargs *elf_auxargs;
1069 struct vmspace *vmspace;
1072 Elf_Brandinfo *brand_info;
1073 struct sysentvec *sv;
1074 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1075 u_long maxalign, mapsz, maxv, maxv1;
1081 hdr = (const Elf_Ehdr *)imgp->image_header;
1084 * Do we have a valid ELF header ?
1086 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1087 * if particular brand doesn't support it.
1089 if (__elfN(check_header)(hdr) != 0 ||
1090 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1094 * From here on down, we return an errno, not -1, as we've
1095 * detected an ELF file.
1098 if (!__elfN(phdr_in_zero_page)(hdr)) {
1099 uprintf("Program headers not in the first page\n");
1102 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1103 if (!aligned(phdr, Elf_Addr)) {
1104 uprintf("Unaligned program headers\n");
1112 entry = proghdr = 0;
1114 free_interp = false;
1116 maxalign = PAGE_SIZE;
1119 for (i = 0; i < hdr->e_phnum; i++) {
1120 switch (phdr[i].p_type) {
1123 baddr = phdr[i].p_vaddr;
1124 if (phdr[i].p_align > maxalign)
1125 maxalign = phdr[i].p_align;
1126 mapsz += phdr[i].p_memsz;
1130 * If this segment contains the program headers,
1131 * remember their virtual address for the AT_PHDR
1132 * aux entry. Static binaries don't usually include
1135 if (phdr[i].p_offset == 0 &&
1136 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1137 <= phdr[i].p_filesz)
1138 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1141 /* Path to interpreter */
1142 if (interp != NULL) {
1143 uprintf("Multiple PT_INTERP headers\n");
1147 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1153 if (__elfN(nxstack))
1155 __elfN(trans_prot)(phdr[i].p_flags);
1156 imgp->stack_sz = phdr[i].p_memsz;
1158 case PT_PHDR: /* Program header table info */
1159 proghdr = phdr[i].p_vaddr;
1164 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1165 if (brand_info == NULL) {
1166 uprintf("ELF binary type \"%u\" not known.\n",
1167 hdr->e_ident[EI_OSABI]);
1171 sv = brand_info->sysvec;
1173 if (hdr->e_type == ET_DYN) {
1174 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1175 uprintf("Cannot execute shared object\n");
1180 * Honour the base load address from the dso if it is
1181 * non-zero for some reason.
1184 if ((sv->sv_flags & SV_ASLR) == 0 ||
1185 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1186 et_dyn_addr = __elfN(pie_base);
1187 else if ((__elfN(pie_aslr_enabled) &&
1188 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1189 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1190 et_dyn_addr = ET_DYN_ADDR_RAND;
1192 et_dyn_addr = __elfN(pie_base);
1197 * Avoid a possible deadlock if the current address space is destroyed
1198 * and that address space maps the locked vnode. In the common case,
1199 * the locked vnode's v_usecount is decremented but remains greater
1200 * than zero. Consequently, the vnode lock is not needed by vrele().
1201 * However, in cases where the vnode lock is external, such as nullfs,
1202 * v_usecount may become zero.
1204 * The VV_TEXT flag prevents modifications to the executable while
1205 * the vnode is unlocked.
1207 VOP_UNLOCK(imgp->vp);
1210 * Decide whether to enable randomization of user mappings.
1211 * First, reset user preferences for the setid binaries.
1212 * Then, account for the support of the randomization by the
1213 * ABI, by user preferences, and make special treatment for
1216 if (imgp->credential_setid) {
1217 PROC_LOCK(imgp->proc);
1218 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1219 PROC_UNLOCK(imgp->proc);
1221 if ((sv->sv_flags & SV_ASLR) == 0 ||
1222 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1223 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1224 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1225 ("et_dyn_addr == RAND and !ASLR"));
1226 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1227 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1228 et_dyn_addr == ET_DYN_ADDR_RAND) {
1229 imgp->map_flags |= MAP_ASLR;
1231 * If user does not care about sbrk, utilize the bss
1232 * grow region for mappings as well. We can select
1233 * the base for the image anywere and still not suffer
1234 * from the fragmentation.
1236 if (!__elfN(aslr_honor_sbrk) ||
1237 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1238 imgp->map_flags |= MAP_ASLR_IGNSTART;
1241 if (!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0)
1242 imgp->map_flags |= MAP_WXORX;
1244 error = exec_new_vmspace(imgp, sv);
1245 vmspace = imgp->proc->p_vmspace;
1246 map = &vmspace->vm_map;
1248 imgp->proc->p_sysent = sv;
1249 imgp->proc->p_elf_brandinfo = brand_info;
1251 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1252 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1253 KASSERT((map->flags & MAP_ASLR) != 0,
1254 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1255 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1256 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1257 /* reserve half of the address space to interpreter */
1258 maxv / 2, 1UL << flsl(maxalign));
1261 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1265 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1269 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1273 entry = (u_long)hdr->e_entry + et_dyn_addr;
1276 * We load the dynamic linker where a userland call
1277 * to mmap(0, ...) would put it. The rationale behind this
1278 * calculation is that it leaves room for the heap to grow to
1279 * its maximum allowed size.
1281 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1283 if ((map->flags & MAP_ASLR) != 0) {
1284 maxv1 = maxv / 2 + addr / 2;
1285 MPASS(maxv1 >= addr); /* No overflow */
1286 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1287 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1288 pagesizes[1] : pagesizes[0]);
1290 map->anon_loc = addr;
1293 imgp->entry_addr = entry;
1295 if (interp != NULL) {
1296 VOP_UNLOCK(imgp->vp);
1297 if ((map->flags & MAP_ASLR) != 0) {
1298 /* Assume that interpreter fits into 1/4 of AS */
1299 maxv1 = maxv / 2 + addr / 2;
1300 MPASS(maxv1 >= addr); /* No overflow */
1301 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1304 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1306 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1313 * Construct auxargs table (used by the copyout_auxargs routine)
1315 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1316 if (elf_auxargs == NULL) {
1317 VOP_UNLOCK(imgp->vp);
1318 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1319 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1321 elf_auxargs->execfd = -1;
1322 elf_auxargs->phdr = proghdr + et_dyn_addr;
1323 elf_auxargs->phent = hdr->e_phentsize;
1324 elf_auxargs->phnum = hdr->e_phnum;
1325 elf_auxargs->pagesz = PAGE_SIZE;
1326 elf_auxargs->base = addr;
1327 elf_auxargs->flags = 0;
1328 elf_auxargs->entry = entry;
1329 elf_auxargs->hdr_eflags = hdr->e_flags;
1331 imgp->auxargs = elf_auxargs;
1332 imgp->interpreted = 0;
1333 imgp->reloc_base = addr;
1334 imgp->proc->p_osrel = osrel;
1335 imgp->proc->p_fctl0 = fctl0;
1336 imgp->proc->p_elf_flags = hdr->e_flags;
1340 free(interp, M_TEMP);
1344 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1347 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1349 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1350 Elf_Auxinfo *argarray, *pos;
1353 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1356 if (args->execfd != -1)
1357 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1358 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1359 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1360 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1361 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1362 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1363 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1364 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1365 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1366 if (imgp->execpathp != 0)
1367 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1368 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1369 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1370 if (imgp->canary != 0) {
1371 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1372 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1374 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1375 if (imgp->pagesizes != 0) {
1376 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1377 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1379 if (imgp->sysent->sv_timekeep_base != 0) {
1380 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1381 imgp->sysent->sv_timekeep_base);
1383 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1384 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1385 imgp->sysent->sv_stackprot);
1386 if (imgp->sysent->sv_hwcap != NULL)
1387 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1388 if (imgp->sysent->sv_hwcap2 != NULL)
1389 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1390 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1391 ELF_BSDF_SIGFASTBLK : 0);
1392 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1393 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1394 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1395 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1396 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1397 if (imgp->sysent->sv_fxrng_gen_base != 0)
1398 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1399 AUXARGS_ENTRY(pos, AT_NULL, 0);
1401 free(imgp->auxargs, M_TEMP);
1402 imgp->auxargs = NULL;
1403 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1405 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1406 free(argarray, M_TEMP);
1411 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1415 base = (Elf_Addr *)*stack_base;
1417 if (suword(base, imgp->args->argc) == -1)
1419 *stack_base = (uintptr_t)base;
1424 * Code for generating ELF core dumps.
1427 typedef void (*segment_callback)(vm_map_entry_t, void *);
1429 /* Closure for cb_put_phdr(). */
1430 struct phdr_closure {
1431 Elf_Phdr *phdr; /* Program header to fill in */
1432 Elf_Off offset; /* Offset of segment in core file */
1436 int type; /* Note type. */
1437 outfunc_t outfunc; /* Output function. */
1438 void *outarg; /* Argument for the output function. */
1439 size_t outsize; /* Output size. */
1440 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1443 TAILQ_HEAD(note_info_list, note_info);
1445 extern int compress_user_cores;
1446 extern int compress_user_cores_level;
1448 static void cb_put_phdr(vm_map_entry_t, void *);
1449 static void cb_size_segment(vm_map_entry_t, void *);
1450 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1452 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1453 struct note_info_list *, size_t, int);
1454 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1456 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1457 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1458 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1459 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1460 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1461 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1462 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1463 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1465 static void note_procstat_files(void *, struct sbuf *, size_t *);
1466 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1467 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1468 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1469 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1470 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1473 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1476 return (core_write((struct coredump_params *)arg, base, len, offset,
1477 UIO_SYSSPACE, NULL));
1481 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1483 struct ucred *cred = td->td_ucred;
1484 int compm, error = 0;
1485 struct sseg_closure seginfo;
1486 struct note_info_list notelst;
1487 struct coredump_params params;
1488 struct note_info *ninfo;
1490 size_t hdrsize, notesz, coresize;
1494 TAILQ_INIT(¬elst);
1496 /* Size the program segments. */
1497 __elfN(size_segments)(td, &seginfo, flags);
1500 * Collect info about the core file header area.
1502 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1503 if (seginfo.count + 1 >= PN_XNUM)
1504 hdrsize += sizeof(Elf_Shdr);
1505 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1506 coresize = round_page(hdrsize + notesz) + seginfo.size;
1508 /* Set up core dump parameters. */
1510 params.active_cred = cred;
1511 params.file_cred = NOCRED;
1518 PROC_LOCK(td->td_proc);
1519 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1520 PROC_UNLOCK(td->td_proc);
1527 if (coresize >= limit) {
1532 /* Create a compression stream if necessary. */
1533 compm = compress_user_cores;
1534 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1536 compm = COMPRESS_GZIP;
1538 params.comp = compressor_init(core_compressed_write,
1539 compm, CORE_BUF_SIZE,
1540 compress_user_cores_level, ¶ms);
1541 if (params.comp == NULL) {
1545 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1549 * Allocate memory for building the header, fill it up,
1550 * and write it out following the notes.
1552 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1553 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1556 /* Write the contents of all of the writable segments. */
1562 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1563 offset = round_page(hdrsize + notesz);
1564 for (i = 0; i < seginfo.count; i++) {
1565 error = core_output((char *)(uintptr_t)php->p_vaddr,
1566 php->p_filesz, offset, ¶ms, tmpbuf);
1569 offset += php->p_filesz;
1572 if (error == 0 && params.comp != NULL)
1573 error = compressor_flush(params.comp);
1577 "Failed to write core file for process %s (error %d)\n",
1578 curproc->p_comm, error);
1582 free(tmpbuf, M_TEMP);
1583 if (params.comp != NULL)
1584 compressor_fini(params.comp);
1585 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1586 TAILQ_REMOVE(¬elst, ninfo, link);
1587 free(ninfo, M_TEMP);
1596 * A callback for each_dumpable_segment() to write out the segment's
1597 * program header entry.
1600 cb_put_phdr(vm_map_entry_t entry, void *closure)
1602 struct phdr_closure *phc = (struct phdr_closure *)closure;
1603 Elf_Phdr *phdr = phc->phdr;
1605 phc->offset = round_page(phc->offset);
1607 phdr->p_type = PT_LOAD;
1608 phdr->p_offset = phc->offset;
1609 phdr->p_vaddr = entry->start;
1611 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1612 phdr->p_align = PAGE_SIZE;
1613 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1615 phc->offset += phdr->p_filesz;
1620 * A callback for each_dumpable_segment() to gather information about
1621 * the number of segments and their total size.
1624 cb_size_segment(vm_map_entry_t entry, void *closure)
1626 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1629 ssc->size += entry->end - entry->start;
1633 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1639 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1643 * For each writable segment in the process's memory map, call the given
1644 * function with a pointer to the map entry and some arbitrary
1645 * caller-supplied data.
1648 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1651 struct proc *p = td->td_proc;
1652 vm_map_t map = &p->p_vmspace->vm_map;
1653 vm_map_entry_t entry;
1654 vm_object_t backing_object, object;
1657 vm_map_lock_read(map);
1658 VM_MAP_ENTRY_FOREACH(entry, map) {
1660 * Don't dump inaccessible mappings, deal with legacy
1663 * Note that read-only segments related to the elf binary
1664 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1665 * need to arbitrarily ignore such segments.
1667 if ((flags & SVC_ALL) == 0) {
1668 if (elf_legacy_coredump) {
1669 if ((entry->protection & VM_PROT_RW) !=
1673 if ((entry->protection & VM_PROT_ALL) == 0)
1679 * Dont include memory segment in the coredump if
1680 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1681 * madvise(2). Do not dump submaps (i.e. parts of the
1684 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1686 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1687 (flags & SVC_ALL) == 0)
1689 if ((object = entry->object.vm_object) == NULL)
1692 /* Ignore memory-mapped devices and such things. */
1693 VM_OBJECT_RLOCK(object);
1694 while ((backing_object = object->backing_object) != NULL) {
1695 VM_OBJECT_RLOCK(backing_object);
1696 VM_OBJECT_RUNLOCK(object);
1697 object = backing_object;
1699 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1700 VM_OBJECT_RUNLOCK(object);
1704 (*func)(entry, closure);
1706 vm_map_unlock_read(map);
1710 * Write the core file header to the file, including padding up to
1711 * the page boundary.
1714 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1715 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1718 struct note_info *ninfo;
1722 /* Fill in the header. */
1723 bzero(hdr, hdrsize);
1724 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1726 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1727 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1728 sbuf_start_section(sb, NULL);
1729 sbuf_bcat(sb, hdr, hdrsize);
1730 TAILQ_FOREACH(ninfo, notelst, link)
1731 __elfN(putnote)(p->td, ninfo, sb);
1732 /* Align up to a page boundary for the program segments. */
1733 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1734 error = sbuf_finish(sb);
1741 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1751 size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1754 * To have the debugger select the right thread (LWP) as the initial
1755 * thread, we dump the state of the thread passed to us in td first.
1756 * This is the thread that causes the core dump and thus likely to
1757 * be the right thread one wants to have selected in the debugger.
1760 while (thr != NULL) {
1761 size += __elfN(register_note)(td, list, NT_PRSTATUS,
1762 __elfN(note_prstatus), thr);
1763 size += __elfN(register_note)(td, list, NT_FPREGSET,
1764 __elfN(note_fpregset), thr);
1765 size += __elfN(register_note)(td, list, NT_THRMISC,
1766 __elfN(note_thrmisc), thr);
1767 size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1768 __elfN(note_ptlwpinfo), thr);
1769 size += __elfN(register_note)(td, list, -1,
1770 __elfN(note_threadmd), thr);
1772 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1773 TAILQ_NEXT(thr, td_plist);
1775 thr = TAILQ_NEXT(thr, td_plist);
1778 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1779 __elfN(note_procstat_proc), p);
1780 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1781 note_procstat_files, p);
1782 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1783 note_procstat_vmmap, p);
1784 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1785 note_procstat_groups, p);
1786 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1787 note_procstat_umask, p);
1788 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1789 note_procstat_rlimit, p);
1790 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1791 note_procstat_osrel, p);
1792 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1793 __elfN(note_procstat_psstrings), p);
1794 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1795 __elfN(note_procstat_auxv), p);
1801 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1802 size_t notesz, int flags)
1807 struct phdr_closure phc;
1810 ehdr = (Elf_Ehdr *)hdr;
1811 bi = td->td_proc->p_elf_brandinfo;
1813 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1814 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1815 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1816 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1817 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1818 ehdr->e_ident[EI_DATA] = ELF_DATA;
1819 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1820 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1821 ehdr->e_ident[EI_ABIVERSION] = 0;
1822 ehdr->e_ident[EI_PAD] = 0;
1823 ehdr->e_type = ET_CORE;
1824 ehdr->e_machine = bi->machine;
1825 ehdr->e_version = EV_CURRENT;
1827 ehdr->e_phoff = sizeof(Elf_Ehdr);
1828 ehdr->e_flags = td->td_proc->p_elf_flags;
1829 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1830 ehdr->e_phentsize = sizeof(Elf_Phdr);
1831 ehdr->e_shentsize = sizeof(Elf_Shdr);
1832 ehdr->e_shstrndx = SHN_UNDEF;
1833 if (numsegs + 1 < PN_XNUM) {
1834 ehdr->e_phnum = numsegs + 1;
1837 ehdr->e_phnum = PN_XNUM;
1840 ehdr->e_shoff = ehdr->e_phoff +
1841 (numsegs + 1) * ehdr->e_phentsize;
1842 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1843 ("e_shoff: %zu, hdrsize - shdr: %zu",
1844 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1846 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1847 memset(shdr, 0, sizeof(*shdr));
1849 * A special first section is used to hold large segment and
1850 * section counts. This was proposed by Sun Microsystems in
1851 * Solaris and has been adopted by Linux; the standard ELF
1852 * tools are already familiar with the technique.
1854 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1855 * (or 12-7 depending on the version of the document) for more
1858 shdr->sh_type = SHT_NULL;
1859 shdr->sh_size = ehdr->e_shnum;
1860 shdr->sh_link = ehdr->e_shstrndx;
1861 shdr->sh_info = numsegs + 1;
1865 * Fill in the program header entries.
1867 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1869 /* The note segement. */
1870 phdr->p_type = PT_NOTE;
1871 phdr->p_offset = hdrsize;
1874 phdr->p_filesz = notesz;
1876 phdr->p_flags = PF_R;
1877 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1880 /* All the writable segments from the program. */
1882 phc.offset = round_page(hdrsize + notesz);
1883 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1887 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1888 int type, outfunc_t out, void *arg)
1890 const struct sysentvec *sv;
1891 struct note_info *ninfo;
1892 size_t size, notesize;
1894 sv = td->td_proc->p_sysent;
1896 out(arg, NULL, &size);
1897 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1899 ninfo->outfunc = out;
1900 ninfo->outarg = arg;
1901 ninfo->outsize = size;
1902 TAILQ_INSERT_TAIL(list, ninfo, link);
1907 notesize = sizeof(Elf_Note) + /* note header */
1908 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1910 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1916 append_note_data(const void *src, void *dst, size_t len)
1920 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1922 bcopy(src, dst, len);
1923 bzero((char *)dst + len, padded_len - len);
1925 return (padded_len);
1929 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1937 note = (Elf_Note *)buf;
1938 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1939 note->n_descsz = size;
1940 note->n_type = type;
1941 buf += sizeof(*note);
1942 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1943 sizeof(FREEBSD_ABI_VENDOR));
1944 append_note_data(src, buf, size);
1949 notesize = sizeof(Elf_Note) + /* note header */
1950 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1952 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1958 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1961 const struct sysentvec *sv;
1962 ssize_t old_len, sect_len;
1963 size_t new_len, descsz, i;
1965 if (ninfo->type == -1) {
1966 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1970 sv = td->td_proc->p_sysent;
1972 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
1973 note.n_descsz = ninfo->outsize;
1974 note.n_type = ninfo->type;
1976 sbuf_bcat(sb, ¬e, sizeof(note));
1977 sbuf_start_section(sb, &old_len);
1978 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
1979 strlen(sv->sv_elf_core_abi_vendor) + 1);
1980 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1981 if (note.n_descsz == 0)
1983 sbuf_start_section(sb, &old_len);
1984 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1985 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1989 new_len = (size_t)sect_len;
1990 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1991 if (new_len < descsz) {
1993 * It is expected that individual note emitters will correctly
1994 * predict their expected output size and fill up to that size
1995 * themselves, padding in a format-specific way if needed.
1996 * However, in case they don't, just do it here with zeros.
1998 for (i = 0; i < descsz - new_len; i++)
2000 } else if (new_len > descsz) {
2002 * We can't always truncate sb -- we may have drained some
2005 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2006 "read it (%zu > %zu). Since it is longer than "
2007 "expected, this coredump's notes are corrupt. THIS "
2008 "IS A BUG in the note_procstat routine for type %u.\n",
2009 __func__, (unsigned)note.n_type, new_len, descsz,
2010 (unsigned)note.n_type));
2015 * Miscellaneous note out functions.
2018 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2019 #include <compat/freebsd32/freebsd32.h>
2020 #include <compat/freebsd32/freebsd32_signal.h>
2022 typedef struct prstatus32 elf_prstatus_t;
2023 typedef struct prpsinfo32 elf_prpsinfo_t;
2024 typedef struct fpreg32 elf_prfpregset_t;
2025 typedef struct fpreg32 elf_fpregset_t;
2026 typedef struct reg32 elf_gregset_t;
2027 typedef struct thrmisc32 elf_thrmisc_t;
2028 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2029 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2030 typedef uint32_t elf_ps_strings_t;
2032 typedef prstatus_t elf_prstatus_t;
2033 typedef prpsinfo_t elf_prpsinfo_t;
2034 typedef prfpregset_t elf_prfpregset_t;
2035 typedef prfpregset_t elf_fpregset_t;
2036 typedef gregset_t elf_gregset_t;
2037 typedef thrmisc_t elf_thrmisc_t;
2038 #define ELF_KERN_PROC_MASK 0
2039 typedef struct kinfo_proc elf_kinfo_proc_t;
2040 typedef vm_offset_t elf_ps_strings_t;
2044 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2050 elf_prpsinfo_t *psinfo;
2055 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2056 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2057 psinfo->pr_version = PRPSINFO_VERSION;
2058 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2059 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2061 if (p->p_args != NULL) {
2062 len = sizeof(psinfo->pr_psargs) - 1;
2063 if (len > p->p_args->ar_length)
2064 len = p->p_args->ar_length;
2065 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2071 sbuf_new(&sbarg, psinfo->pr_psargs,
2072 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2073 error = proc_getargv(curthread, p, &sbarg);
2075 if (sbuf_finish(&sbarg) == 0)
2076 len = sbuf_len(&sbarg) - 1;
2078 len = sizeof(psinfo->pr_psargs) - 1;
2079 sbuf_delete(&sbarg);
2081 if (error || len == 0)
2082 strlcpy(psinfo->pr_psargs, p->p_comm,
2083 sizeof(psinfo->pr_psargs));
2085 KASSERT(len < sizeof(psinfo->pr_psargs),
2086 ("len is too long: %zu vs %zu", len,
2087 sizeof(psinfo->pr_psargs)));
2088 cp = psinfo->pr_psargs;
2091 cp = memchr(cp, '\0', end - cp);
2097 psinfo->pr_pid = p->p_pid;
2098 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2099 free(psinfo, M_TEMP);
2101 *sizep = sizeof(*psinfo);
2105 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2108 elf_prstatus_t *status;
2112 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2113 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2114 status->pr_version = PRSTATUS_VERSION;
2115 status->pr_statussz = sizeof(elf_prstatus_t);
2116 status->pr_gregsetsz = sizeof(elf_gregset_t);
2117 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2118 status->pr_osreldate = osreldate;
2119 status->pr_cursig = td->td_proc->p_sig;
2120 status->pr_pid = td->td_tid;
2121 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2122 fill_regs32(td, &status->pr_reg);
2124 fill_regs(td, &status->pr_reg);
2126 sbuf_bcat(sb, status, sizeof(*status));
2127 free(status, M_TEMP);
2129 *sizep = sizeof(*status);
2133 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2136 elf_prfpregset_t *fpregset;
2140 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2141 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2142 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2143 fill_fpregs32(td, fpregset);
2145 fill_fpregs(td, fpregset);
2147 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2148 free(fpregset, M_TEMP);
2150 *sizep = sizeof(*fpregset);
2154 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2157 elf_thrmisc_t thrmisc;
2161 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2162 bzero(&thrmisc, sizeof(thrmisc));
2163 strcpy(thrmisc.pr_tname, td->td_name);
2164 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2166 *sizep = sizeof(thrmisc);
2170 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2175 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2176 struct ptrace_lwpinfo32 pl;
2178 struct ptrace_lwpinfo pl;
2182 size = sizeof(structsize) + sizeof(pl);
2184 KASSERT(*sizep == size, ("invalid size"));
2185 structsize = sizeof(pl);
2186 sbuf_bcat(sb, &structsize, sizeof(structsize));
2187 bzero(&pl, sizeof(pl));
2188 pl.pl_lwpid = td->td_tid;
2189 pl.pl_event = PL_EVENT_NONE;
2190 pl.pl_sigmask = td->td_sigmask;
2191 pl.pl_siglist = td->td_siglist;
2192 if (td->td_si.si_signo != 0) {
2193 pl.pl_event = PL_EVENT_SIGNAL;
2194 pl.pl_flags |= PL_FLAG_SI;
2195 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2196 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2198 pl.pl_siginfo = td->td_si;
2201 strcpy(pl.pl_tdname, td->td_name);
2202 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2203 sbuf_bcat(sb, &pl, sizeof(pl));
2209 * Allow for MD specific notes, as well as any MD
2210 * specific preparations for writing MI notes.
2213 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2219 td = (struct thread *)arg;
2221 if (size != 0 && sb != NULL)
2222 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2226 __elfN(dump_thread)(td, buf, &size);
2227 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2228 if (size != 0 && sb != NULL)
2229 sbuf_bcat(sb, buf, size);
2234 #ifdef KINFO_PROC_SIZE
2235 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2239 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2246 size = sizeof(structsize) + p->p_numthreads *
2247 sizeof(elf_kinfo_proc_t);
2250 KASSERT(*sizep == size, ("invalid size"));
2251 structsize = sizeof(elf_kinfo_proc_t);
2252 sbuf_bcat(sb, &structsize, sizeof(structsize));
2253 sx_slock(&proctree_lock);
2255 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2256 sx_sunlock(&proctree_lock);
2261 #ifdef KINFO_FILE_SIZE
2262 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2266 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2269 size_t size, sect_sz, i;
2270 ssize_t start_len, sect_len;
2271 int structsize, filedesc_flags;
2273 if (coredump_pack_fileinfo)
2274 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2279 structsize = sizeof(struct kinfo_file);
2282 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2283 sbuf_set_drain(sb, sbuf_count_drain, &size);
2284 sbuf_bcat(sb, &structsize, sizeof(structsize));
2286 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2291 sbuf_start_section(sb, &start_len);
2293 sbuf_bcat(sb, &structsize, sizeof(structsize));
2295 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2298 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2303 KASSERT(sect_sz <= *sizep,
2304 ("kern_proc_filedesc_out did not respect maxlen; "
2305 "requested %zu, got %zu", *sizep - sizeof(structsize),
2306 sect_sz - sizeof(structsize)));
2308 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2313 #ifdef KINFO_VMENTRY_SIZE
2314 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2318 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2322 int structsize, vmmap_flags;
2324 if (coredump_pack_vmmapinfo)
2325 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2330 structsize = sizeof(struct kinfo_vmentry);
2333 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2334 sbuf_set_drain(sb, sbuf_count_drain, &size);
2335 sbuf_bcat(sb, &structsize, sizeof(structsize));
2337 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2342 sbuf_bcat(sb, &structsize, sizeof(structsize));
2344 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2350 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2357 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2359 KASSERT(*sizep == size, ("invalid size"));
2360 structsize = sizeof(gid_t);
2361 sbuf_bcat(sb, &structsize, sizeof(structsize));
2362 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2369 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2376 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2378 KASSERT(*sizep == size, ("invalid size"));
2379 structsize = sizeof(p->p_pd->pd_cmask);
2380 sbuf_bcat(sb, &structsize, sizeof(structsize));
2381 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2387 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2390 struct rlimit rlim[RLIM_NLIMITS];
2395 size = sizeof(structsize) + sizeof(rlim);
2397 KASSERT(*sizep == size, ("invalid size"));
2398 structsize = sizeof(rlim);
2399 sbuf_bcat(sb, &structsize, sizeof(structsize));
2401 for (i = 0; i < RLIM_NLIMITS; i++)
2402 lim_rlimit_proc(p, i, &rlim[i]);
2404 sbuf_bcat(sb, rlim, sizeof(rlim));
2410 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2417 size = sizeof(structsize) + sizeof(p->p_osrel);
2419 KASSERT(*sizep == size, ("invalid size"));
2420 structsize = sizeof(p->p_osrel);
2421 sbuf_bcat(sb, &structsize, sizeof(structsize));
2422 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2428 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2431 elf_ps_strings_t ps_strings;
2436 size = sizeof(structsize) + sizeof(ps_strings);
2438 KASSERT(*sizep == size, ("invalid size"));
2439 structsize = sizeof(ps_strings);
2440 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2441 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2443 ps_strings = p->p_sysent->sv_psstrings;
2445 sbuf_bcat(sb, &structsize, sizeof(structsize));
2446 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2452 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2461 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2463 sbuf_set_drain(sb, sbuf_count_drain, &size);
2464 sbuf_bcat(sb, &structsize, sizeof(structsize));
2466 proc_getauxv(curthread, p, sb);
2472 structsize = sizeof(Elf_Auxinfo);
2473 sbuf_bcat(sb, &structsize, sizeof(structsize));
2475 proc_getauxv(curthread, p, sb);
2481 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2482 const char *note_vendor, const Elf_Phdr *pnote,
2483 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2485 const Elf_Note *note, *note0, *note_end;
2486 const char *note_name;
2491 /* We need some limit, might as well use PAGE_SIZE. */
2492 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2494 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2495 if (pnote->p_offset > PAGE_SIZE ||
2496 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2497 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2499 VOP_UNLOCK(imgp->vp);
2500 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2501 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2503 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2504 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2505 curthread->td_ucred, NOCRED, NULL, curthread);
2507 uprintf("i/o error PT_NOTE\n");
2510 note = note0 = (const Elf_Note *)buf;
2511 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2513 note = note0 = (const Elf_Note *)(imgp->image_header +
2515 note_end = (const Elf_Note *)(imgp->image_header +
2516 pnote->p_offset + pnote->p_filesz);
2519 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2520 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2521 (const char *)note < sizeof(Elf_Note)) {
2524 if (note->n_namesz != checknote->n_namesz ||
2525 note->n_descsz != checknote->n_descsz ||
2526 note->n_type != checknote->n_type)
2528 note_name = (const char *)(note + 1);
2529 if (note_name + checknote->n_namesz >=
2530 (const char *)note_end || strncmp(note_vendor,
2531 note_name, checknote->n_namesz) != 0)
2534 if (cb(note, cb_arg, &res))
2537 note = (const Elf_Note *)((const char *)(note + 1) +
2538 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2539 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2548 struct brandnote_cb_arg {
2549 Elf_Brandnote *brandnote;
2554 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2556 struct brandnote_cb_arg *arg;
2561 * Fetch the osreldate for binary from the ELF OSABI-note if
2564 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2565 arg->brandnote->trans_osrel != NULL ?
2566 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2571 static Elf_Note fctl_note = {
2572 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2573 .n_descsz = sizeof(uint32_t),
2574 .n_type = NT_FREEBSD_FEATURE_CTL,
2577 struct fctl_cb_arg {
2578 boolean_t *has_fctl0;
2583 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2585 struct fctl_cb_arg *arg;
2586 const Elf32_Word *desc;
2590 p = (uintptr_t)(note + 1);
2591 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2592 desc = (const Elf32_Word *)p;
2593 *arg->has_fctl0 = TRUE;
2594 *arg->fctl0 = desc[0];
2600 * Try to find the appropriate ABI-note section for checknote, fetch
2601 * the osreldate and feature control flags for binary from the ELF
2602 * OSABI-note. Only the first page of the image is searched, the same
2606 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2607 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2609 const Elf_Phdr *phdr;
2610 const Elf_Ehdr *hdr;
2611 struct brandnote_cb_arg b_arg;
2612 struct fctl_cb_arg f_arg;
2615 hdr = (const Elf_Ehdr *)imgp->image_header;
2616 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2617 b_arg.brandnote = brandnote;
2618 b_arg.osrel = osrel;
2619 f_arg.has_fctl0 = has_fctl0;
2620 f_arg.fctl0 = fctl0;
2622 for (i = 0; i < hdr->e_phnum; i++) {
2623 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2624 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2626 for (j = 0; j < hdr->e_phnum; j++) {
2627 if (phdr[j].p_type == PT_NOTE &&
2628 __elfN(parse_notes)(imgp, &fctl_note,
2629 FREEBSD_ABI_VENDOR, &phdr[j],
2630 note_fctl_cb, &f_arg))
2641 * Tell kern_execve.c about it, with a little help from the linker.
2643 static struct execsw __elfN(execsw) = {
2644 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2645 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2647 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2650 __elfN(trans_prot)(Elf_Word flags)
2656 prot |= VM_PROT_EXECUTE;
2658 prot |= VM_PROT_WRITE;
2660 prot |= VM_PROT_READ;
2661 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2662 if (i386_read_exec && (flags & PF_R))
2663 prot |= VM_PROT_EXECUTE;
2669 __elfN(untrans_prot)(vm_prot_t prot)
2674 if (prot & VM_PROT_EXECUTE)
2676 if (prot & VM_PROT_READ)
2678 if (prot & VM_PROT_WRITE)
2684 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2686 uintptr_t range, rbase, gap;
2689 pct = __elfN(aslr_stack_gap);
2694 range = imgp->eff_stack_sz * pct / 100;
2695 arc4rand(&rbase, sizeof(rbase), 0);
2696 gap = rbase % range;
2697 gap &= ~(sizeof(u_long) - 1);