2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
54 #include <sys/procfs.h>
55 #include <sys/ptrace.h>
56 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, boolean_t *has_fctl0,
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE),
107 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __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 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
140 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
145 val = __elfN(pie_base);
146 error = sysctl_handle_long(oidp, &val, 0, req);
147 if (error != 0 || req->newptr == NULL)
149 if ((val & PAGE_MASK) != 0)
151 __elfN(pie_base) = val;
154 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
155 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
156 sysctl_pie_base, "LU",
157 "PIE load base without randomization");
159 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr,
160 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
162 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
164 static int __elfN(aslr_enabled) = 0;
165 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
166 &__elfN(aslr_enabled), 0,
167 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
168 ": enable address map randomization");
170 static int __elfN(pie_aslr_enabled) = 0;
171 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
172 &__elfN(pie_aslr_enabled), 0,
173 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
174 ": enable address map randomization for PIE binaries");
176 static int __elfN(aslr_honor_sbrk) = 1;
177 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
178 &__elfN(aslr_honor_sbrk), 0,
179 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
181 static int __elfN(aslr_stack_gap) = 3;
182 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
183 &__elfN(aslr_stack_gap), 0,
184 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
185 ": maximum percentage of main stack to waste on a random gap");
187 static int __elfN(sigfastblock) = 1;
188 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, sigfastblock,
189 CTLFLAG_RWTUN, &__elfN(sigfastblock), 0,
190 "enable sigfastblock for new processes");
192 static bool __elfN(allow_wx) = true;
193 SYSCTL_BOOL(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, allow_wx,
194 CTLFLAG_RWTUN, &__elfN(allow_wx), 0,
195 "Allow pages to be mapped simultaneously writable and executable");
197 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
199 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
201 Elf_Brandnote __elfN(freebsd_brandnote) = {
202 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
203 .hdr.n_descsz = sizeof(int32_t),
204 .hdr.n_type = NT_FREEBSD_ABI_TAG,
205 .vendor = FREEBSD_ABI_VENDOR,
206 .flags = BN_TRANSLATE_OSREL,
207 .trans_osrel = __elfN(freebsd_trans_osrel)
211 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
215 p = (uintptr_t)(note + 1);
216 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
217 *osrel = *(const int32_t *)(p);
222 static const char GNU_ABI_VENDOR[] = "GNU";
223 static int GNU_KFREEBSD_ABI_DESC = 3;
225 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
226 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
227 .hdr.n_descsz = 16, /* XXX at least 16 */
229 .vendor = GNU_ABI_VENDOR,
230 .flags = BN_TRANSLATE_OSREL,
231 .trans_osrel = kfreebsd_trans_osrel
235 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
237 const Elf32_Word *desc;
240 p = (uintptr_t)(note + 1);
241 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
243 desc = (const Elf32_Word *)p;
244 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
248 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
249 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
251 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
257 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
261 for (i = 0; i < MAX_BRANDS; i++) {
262 if (elf_brand_list[i] == NULL) {
263 elf_brand_list[i] = entry;
267 if (i == MAX_BRANDS) {
268 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
276 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
280 for (i = 0; i < MAX_BRANDS; i++) {
281 if (elf_brand_list[i] == entry) {
282 elf_brand_list[i] = NULL;
292 __elfN(brand_inuse)(Elf_Brandinfo *entry)
297 sx_slock(&allproc_lock);
298 FOREACH_PROC_IN_SYSTEM(p) {
299 if (p->p_sysent == entry->sysvec) {
304 sx_sunlock(&allproc_lock);
309 static Elf_Brandinfo *
310 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
311 int32_t *osrel, uint32_t *fctl0)
313 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
314 Elf_Brandinfo *bi, *bi_m;
315 boolean_t ret, has_fctl0;
316 int i, interp_name_len;
318 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
321 * We support four types of branding -- (1) the ELF EI_OSABI field
322 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
323 * branding w/in the ELF header, (3) path of the `interp_path'
324 * field, and (4) the ".note.ABI-tag" ELF section.
327 /* Look for an ".note.ABI-tag" ELF section */
329 for (i = 0; i < MAX_BRANDS; i++) {
330 bi = elf_brand_list[i];
333 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
335 if (hdr->e_machine == bi->machine && (bi->flags &
336 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
340 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
342 /* Give brand a chance to veto check_note's guess */
343 if (ret && bi->header_supported) {
344 ret = bi->header_supported(imgp, osrel,
345 has_fctl0 ? fctl0 : NULL);
348 * If note checker claimed the binary, but the
349 * interpreter path in the image does not
350 * match default one for the brand, try to
351 * search for other brands with the same
352 * interpreter. Either there is better brand
353 * with the right interpreter, or, failing
354 * this, we return first brand which accepted
355 * our note and, optionally, header.
357 if (ret && bi_m == NULL && interp != NULL &&
358 (bi->interp_path == NULL ||
359 (strlen(bi->interp_path) + 1 != interp_name_len ||
360 strncmp(interp, bi->interp_path, interp_name_len)
372 /* If the executable has a brand, search for it in the brand list. */
373 for (i = 0; i < MAX_BRANDS; i++) {
374 bi = elf_brand_list[i];
375 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
376 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
378 if (hdr->e_machine == bi->machine &&
379 (hdr->e_ident[EI_OSABI] == bi->brand ||
380 (bi->compat_3_brand != NULL &&
381 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
382 bi->compat_3_brand) == 0))) {
383 /* Looks good, but give brand a chance to veto */
384 if (bi->header_supported == NULL ||
385 bi->header_supported(imgp, NULL, NULL)) {
387 * Again, prefer strictly matching
390 if (interp_name_len == 0 &&
391 bi->interp_path == NULL)
393 if (bi->interp_path != NULL &&
394 strlen(bi->interp_path) + 1 ==
395 interp_name_len && strncmp(interp,
396 bi->interp_path, interp_name_len) == 0)
406 /* No known brand, see if the header is recognized by any brand */
407 for (i = 0; i < MAX_BRANDS; i++) {
408 bi = elf_brand_list[i];
409 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
410 bi->header_supported == NULL)
412 if (hdr->e_machine == bi->machine) {
413 ret = bi->header_supported(imgp, NULL, NULL);
419 /* Lacking a known brand, search for a recognized interpreter. */
420 if (interp != NULL) {
421 for (i = 0; i < MAX_BRANDS; i++) {
422 bi = elf_brand_list[i];
423 if (bi == NULL || (bi->flags &
424 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
427 if (hdr->e_machine == bi->machine &&
428 bi->interp_path != NULL &&
429 /* ELF image p_filesz includes terminating zero */
430 strlen(bi->interp_path) + 1 == interp_name_len &&
431 strncmp(interp, bi->interp_path, interp_name_len)
432 == 0 && (bi->header_supported == NULL ||
433 bi->header_supported(imgp, NULL, NULL)))
438 /* Lacking a recognized interpreter, try the default brand */
439 for (i = 0; i < MAX_BRANDS; i++) {
440 bi = elf_brand_list[i];
441 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
442 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
444 if (hdr->e_machine == bi->machine &&
445 __elfN(fallback_brand) == bi->brand &&
446 (bi->header_supported == NULL ||
447 bi->header_supported(imgp, NULL, NULL)))
454 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
456 return (hdr->e_phoff <= PAGE_SIZE &&
457 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
461 __elfN(check_header)(const Elf_Ehdr *hdr)
467 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
468 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
469 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
470 hdr->e_phentsize != sizeof(Elf_Phdr) ||
471 hdr->e_version != ELF_TARG_VER)
475 * Make sure we have at least one brand for this machine.
478 for (i = 0; i < MAX_BRANDS; i++) {
479 bi = elf_brand_list[i];
480 if (bi != NULL && bi->machine == hdr->e_machine)
490 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
491 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
498 * Create the page if it doesn't exist yet. Ignore errors.
500 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
501 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
504 * Find the page from the underlying object.
506 if (object != NULL) {
507 sf = vm_imgact_map_page(object, offset);
509 return (KERN_FAILURE);
510 off = offset - trunc_page(offset);
511 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
513 vm_imgact_unmap_page(sf);
515 return (KERN_FAILURE);
518 return (KERN_SUCCESS);
522 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
523 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
529 int error, locked, rv;
531 if (start != trunc_page(start)) {
532 rv = __elfN(map_partial)(map, object, offset, start,
533 round_page(start), prot);
534 if (rv != KERN_SUCCESS)
536 offset += round_page(start) - start;
537 start = round_page(start);
539 if (end != round_page(end)) {
540 rv = __elfN(map_partial)(map, object, offset +
541 trunc_page(end) - start, trunc_page(end), end, prot);
542 if (rv != KERN_SUCCESS)
544 end = trunc_page(end);
547 return (KERN_SUCCESS);
548 if ((offset & PAGE_MASK) != 0) {
550 * The mapping is not page aligned. This means that we have
553 rv = vm_map_fixed(map, NULL, 0, start, end - start,
554 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
555 if (rv != KERN_SUCCESS)
558 return (KERN_SUCCESS);
559 for (; start < end; start += sz) {
560 sf = vm_imgact_map_page(object, offset);
562 return (KERN_FAILURE);
563 off = offset - trunc_page(offset);
565 if (sz > PAGE_SIZE - off)
566 sz = PAGE_SIZE - off;
567 error = copyout((caddr_t)sf_buf_kva(sf) + off,
569 vm_imgact_unmap_page(sf);
571 return (KERN_FAILURE);
575 vm_object_reference(object);
576 rv = vm_map_fixed(map, object, offset, start, end - start,
577 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
578 (object != NULL ? MAP_VN_EXEC : 0));
579 if (rv != KERN_SUCCESS) {
580 locked = VOP_ISLOCKED(imgp->vp);
581 VOP_UNLOCK(imgp->vp);
582 vm_object_deallocate(object);
583 vn_lock(imgp->vp, locked | LK_RETRY);
585 } else if (object != NULL) {
586 MPASS(imgp->vp->v_object == object);
587 VOP_SET_TEXT_CHECKED(imgp->vp);
590 return (KERN_SUCCESS);
594 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
595 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
601 vm_offset_t map_addr;
604 vm_ooffset_t file_addr;
607 * It's necessary to fail if the filsz + offset taken from the
608 * header is greater than the actual file pager object's size.
609 * If we were to allow this, then the vm_map_find() below would
610 * walk right off the end of the file object and into the ether.
612 * While I'm here, might as well check for something else that
613 * is invalid: filsz cannot be greater than memsz.
615 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
617 uprintf("elf_load_section: truncated ELF file\n");
621 object = imgp->object;
622 map = &imgp->proc->p_vmspace->vm_map;
623 map_addr = trunc_page((vm_offset_t)vmaddr);
624 file_addr = trunc_page(offset);
627 * We have two choices. We can either clear the data in the last page
628 * of an oversized mapping, or we can start the anon mapping a page
629 * early and copy the initialized data into that first page. We
634 else if (memsz > filsz)
635 map_len = trunc_page(offset + filsz) - file_addr;
637 map_len = round_page(offset + filsz) - file_addr;
640 /* cow flags: don't dump readonly sections in core */
641 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
642 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
644 rv = __elfN(map_insert)(imgp, map, object, file_addr,
645 map_addr, map_addr + map_len, prot, cow);
646 if (rv != KERN_SUCCESS)
649 /* we can stop now if we've covered it all */
655 * We have to get the remaining bit of the file into the first part
656 * of the oversized map segment. This is normally because the .data
657 * segment in the file is extended to provide bss. It's a neat idea
658 * to try and save a page, but it's a pain in the behind to implement.
660 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
662 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
663 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
665 /* This had damn well better be true! */
667 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
668 map_addr + map_len, prot, 0);
669 if (rv != KERN_SUCCESS)
674 sf = vm_imgact_map_page(object, offset + filsz);
678 /* send the page fragment to user space */
679 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
681 vm_imgact_unmap_page(sf);
687 * Remove write access to the page if it was only granted by map_insert
690 if ((prot & VM_PROT_WRITE) == 0)
691 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
692 map_len), prot, 0, VM_MAP_PROTECT_SET_PROT);
698 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
699 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
706 ASSERT_VOP_LOCKED(imgp->vp, __func__);
711 for (i = 0; i < hdr->e_phnum; i++) {
712 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
715 /* Loadable segment */
716 prot = __elfN(trans_prot)(phdr[i].p_flags);
717 error = __elfN(load_section)(imgp, phdr[i].p_offset,
718 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
719 phdr[i].p_memsz, phdr[i].p_filesz, prot);
724 * Establish the base address if this is the first segment.
727 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
732 if (base_addrp != NULL)
733 *base_addrp = base_addr;
739 * Load the file "file" into memory. It may be either a shared object
742 * The "addr" reference parameter is in/out. On entry, it specifies
743 * the address where a shared object should be loaded. If the file is
744 * an executable, this value is ignored. On exit, "addr" specifies
745 * where the file was actually loaded.
747 * The "entry" reference parameter is out only. On exit, it specifies
748 * the entry point for the loaded file.
751 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
757 struct image_params image_params;
759 const Elf_Ehdr *hdr = NULL;
760 const Elf_Phdr *phdr = NULL;
761 struct nameidata *nd;
763 struct image_params *imgp;
765 u_long base_addr = 0;
768 #ifdef CAPABILITY_MODE
770 * XXXJA: This check can go away once we are sufficiently confident
771 * that the checks in namei() are correct.
773 if (IN_CAPABILITY_MODE(curthread))
777 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
779 attr = &tempdata->attr;
780 imgp = &tempdata->image_params;
783 * Initialize part of the common data
788 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
789 UIO_SYSSPACE, file, curthread);
790 if ((error = namei(nd)) != 0) {
794 NDFREE(nd, NDF_ONLY_PNBUF);
795 imgp->vp = nd->ni_vp;
798 * Check permissions, modes, uid, etc on the file, and "open" it.
800 error = exec_check_permissions(imgp);
804 error = exec_map_first_page(imgp);
808 imgp->object = nd->ni_vp->v_object;
810 hdr = (const Elf_Ehdr *)imgp->image_header;
811 if ((error = __elfN(check_header)(hdr)) != 0)
813 if (hdr->e_type == ET_DYN)
815 else if (hdr->e_type == ET_EXEC)
822 /* Only support headers that fit within first page for now */
823 if (!__elfN(phdr_in_zero_page)(hdr)) {
828 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
829 if (!aligned(phdr, Elf_Addr)) {
834 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
839 *entry = (unsigned long)hdr->e_entry + rbase;
843 exec_unmap_first_page(imgp);
847 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
850 free(tempdata, M_TEMP);
856 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
861 MPASS(vm_map_min(map) <= minv);
862 MPASS(maxv <= vm_map_max(map));
864 MPASS(minv + align < maxv);
865 arc4rand(&rbase, sizeof(rbase), 0);
866 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
867 res &= ~((u_long)align - 1);
871 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
872 res, minv, maxv, rbase));
874 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
875 res, maxv, minv, rbase));
880 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
881 const Elf_Phdr *phdr, u_long et_dyn_addr)
883 struct vmspace *vmspace;
885 u_long text_size, data_size, total_size, text_addr, data_addr;
886 u_long seg_size, seg_addr;
890 text_size = data_size = total_size = text_addr = data_addr = 0;
892 for (i = 0; i < hdr->e_phnum; i++) {
893 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
896 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
897 seg_size = round_page(phdr[i].p_memsz +
898 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
901 * Make the largest executable segment the official
902 * text segment and all others data.
904 * Note that obreak() assumes that data_addr + data_size == end
905 * of data load area, and the ELF file format expects segments
906 * to be sorted by address. If multiple data segments exist,
907 * the last one will be used.
910 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
911 text_size = seg_size;
912 text_addr = seg_addr;
914 data_size = seg_size;
915 data_addr = seg_addr;
917 total_size += seg_size;
920 if (data_addr == 0 && data_size == 0) {
921 data_addr = text_addr;
922 data_size = text_size;
926 * Check limits. It should be safe to check the
927 * limits after loading the segments since we do
928 * not actually fault in all the segments pages.
930 PROC_LOCK(imgp->proc);
931 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
932 err_str = "Data segment size exceeds process limit";
933 else if (text_size > maxtsiz)
934 err_str = "Text segment size exceeds system limit";
935 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
936 err_str = "Total segment size exceeds process limit";
937 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
938 err_str = "Data segment size exceeds resource limit";
939 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
940 err_str = "Total segment size exceeds resource limit";
941 PROC_UNLOCK(imgp->proc);
942 if (err_str != NULL) {
943 uprintf("%s\n", err_str);
947 vmspace = imgp->proc->p_vmspace;
948 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
949 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
950 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
951 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
957 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
958 char **interpp, bool *free_interpp)
962 int error, interp_name_len;
964 KASSERT(phdr->p_type == PT_INTERP,
965 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
966 ASSERT_VOP_LOCKED(imgp->vp, __func__);
970 /* Path to interpreter */
971 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
972 uprintf("Invalid PT_INTERP\n");
976 interp_name_len = phdr->p_filesz;
977 if (phdr->p_offset > PAGE_SIZE ||
978 interp_name_len > PAGE_SIZE - phdr->p_offset) {
980 * The vnode lock might be needed by the pagedaemon to
981 * clean pages owned by the vnode. Do not allow sleep
982 * waiting for memory with the vnode locked, instead
983 * try non-sleepable allocation first, and if it
984 * fails, go to the slow path were we drop the lock
985 * and do M_WAITOK. A text reference prevents
986 * modifications to the vnode content.
988 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
989 if (interp == NULL) {
990 VOP_UNLOCK(imgp->vp);
991 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
992 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
995 error = vn_rdwr(UIO_READ, imgp->vp, interp,
996 interp_name_len, phdr->p_offset,
997 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
1000 free(interp, M_TEMP);
1001 uprintf("i/o error PT_INTERP %d\n", error);
1004 interp[interp_name_len] = '\0';
1007 *free_interpp = true;
1011 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1012 if (interp[interp_name_len - 1] != '\0') {
1013 uprintf("Invalid PT_INTERP\n");
1018 *free_interpp = false;
1023 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1024 const char *interp, u_long *addr, u_long *entry)
1029 if (brand_info->emul_path != NULL &&
1030 brand_info->emul_path[0] != '\0') {
1031 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1032 snprintf(path, MAXPATHLEN, "%s%s",
1033 brand_info->emul_path, interp);
1034 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1040 if (brand_info->interp_newpath != NULL &&
1041 (brand_info->interp_path == NULL ||
1042 strcmp(interp, brand_info->interp_path) == 0)) {
1043 error = __elfN(load_file)(imgp->proc,
1044 brand_info->interp_newpath, addr, entry);
1049 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1053 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1058 * Impossible et_dyn_addr initial value indicating that the real base
1059 * must be calculated later with some randomization applied.
1061 #define ET_DYN_ADDR_RAND 1
1064 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1067 const Elf_Ehdr *hdr;
1068 const Elf_Phdr *phdr;
1069 Elf_Auxargs *elf_auxargs;
1070 struct vmspace *vmspace;
1073 Elf_Brandinfo *brand_info;
1074 struct sysentvec *sv;
1075 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1076 u_long maxalign, mapsz, maxv, maxv1;
1082 hdr = (const Elf_Ehdr *)imgp->image_header;
1085 * Do we have a valid ELF header ?
1087 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1088 * if particular brand doesn't support it.
1090 if (__elfN(check_header)(hdr) != 0 ||
1091 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1095 * From here on down, we return an errno, not -1, as we've
1096 * detected an ELF file.
1099 if (!__elfN(phdr_in_zero_page)(hdr)) {
1100 uprintf("Program headers not in the first page\n");
1103 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1104 if (!aligned(phdr, Elf_Addr)) {
1105 uprintf("Unaligned program headers\n");
1113 entry = proghdr = 0;
1115 free_interp = false;
1117 maxalign = PAGE_SIZE;
1120 for (i = 0; i < hdr->e_phnum; i++) {
1121 switch (phdr[i].p_type) {
1124 baddr = phdr[i].p_vaddr;
1125 if (phdr[i].p_align > maxalign)
1126 maxalign = phdr[i].p_align;
1127 mapsz += phdr[i].p_memsz;
1131 * If this segment contains the program headers,
1132 * remember their virtual address for the AT_PHDR
1133 * aux entry. Static binaries don't usually include
1136 if (phdr[i].p_offset == 0 &&
1137 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1138 <= phdr[i].p_filesz)
1139 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1142 /* Path to interpreter */
1143 if (interp != NULL) {
1144 uprintf("Multiple PT_INTERP headers\n");
1148 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1154 if (__elfN(nxstack))
1156 __elfN(trans_prot)(phdr[i].p_flags);
1157 imgp->stack_sz = phdr[i].p_memsz;
1159 case PT_PHDR: /* Program header table info */
1160 proghdr = phdr[i].p_vaddr;
1165 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1166 if (brand_info == NULL) {
1167 uprintf("ELF binary type \"%u\" not known.\n",
1168 hdr->e_ident[EI_OSABI]);
1172 sv = brand_info->sysvec;
1174 if (hdr->e_type == ET_DYN) {
1175 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1176 uprintf("Cannot execute shared object\n");
1181 * Honour the base load address from the dso if it is
1182 * non-zero for some reason.
1185 if ((sv->sv_flags & SV_ASLR) == 0 ||
1186 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1187 et_dyn_addr = __elfN(pie_base);
1188 else if ((__elfN(pie_aslr_enabled) &&
1189 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1190 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1191 et_dyn_addr = ET_DYN_ADDR_RAND;
1193 et_dyn_addr = __elfN(pie_base);
1198 * Avoid a possible deadlock if the current address space is destroyed
1199 * and that address space maps the locked vnode. In the common case,
1200 * the locked vnode's v_usecount is decremented but remains greater
1201 * than zero. Consequently, the vnode lock is not needed by vrele().
1202 * However, in cases where the vnode lock is external, such as nullfs,
1203 * v_usecount may become zero.
1205 * The VV_TEXT flag prevents modifications to the executable while
1206 * the vnode is unlocked.
1208 VOP_UNLOCK(imgp->vp);
1211 * Decide whether to enable randomization of user mappings.
1212 * First, reset user preferences for the setid binaries.
1213 * Then, account for the support of the randomization by the
1214 * ABI, by user preferences, and make special treatment for
1217 if (imgp->credential_setid) {
1218 PROC_LOCK(imgp->proc);
1219 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE |
1220 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
1221 PROC_UNLOCK(imgp->proc);
1223 if ((sv->sv_flags & SV_ASLR) == 0 ||
1224 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1225 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1226 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1227 ("et_dyn_addr == RAND and !ASLR"));
1228 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1229 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1230 et_dyn_addr == ET_DYN_ADDR_RAND) {
1231 imgp->map_flags |= MAP_ASLR;
1233 * If user does not care about sbrk, utilize the bss
1234 * grow region for mappings as well. We can select
1235 * the base for the image anywere and still not suffer
1236 * from the fragmentation.
1238 if (!__elfN(aslr_honor_sbrk) ||
1239 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1240 imgp->map_flags |= MAP_ASLR_IGNSTART;
1243 if ((!__elfN(allow_wx) && (fctl0 & NT_FREEBSD_FCTL_WXNEEDED) == 0 &&
1244 (imgp->proc->p_flag2 & P2_WXORX_DISABLE) == 0) ||
1245 (imgp->proc->p_flag2 & P2_WXORX_ENABLE_EXEC) != 0)
1246 imgp->map_flags |= MAP_WXORX;
1248 error = exec_new_vmspace(imgp, sv);
1249 vmspace = imgp->proc->p_vmspace;
1250 map = &vmspace->vm_map;
1252 imgp->proc->p_sysent = sv;
1253 imgp->proc->p_elf_brandinfo = brand_info;
1255 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1256 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1257 KASSERT((map->flags & MAP_ASLR) != 0,
1258 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1259 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1260 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1261 /* reserve half of the address space to interpreter */
1262 maxv / 2, 1UL << flsl(maxalign));
1265 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1269 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1273 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1277 entry = (u_long)hdr->e_entry + et_dyn_addr;
1280 * We load the dynamic linker where a userland call
1281 * to mmap(0, ...) would put it. The rationale behind this
1282 * calculation is that it leaves room for the heap to grow to
1283 * its maximum allowed size.
1285 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1287 if ((map->flags & MAP_ASLR) != 0) {
1288 maxv1 = maxv / 2 + addr / 2;
1289 MPASS(maxv1 >= addr); /* No overflow */
1290 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1291 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1292 pagesizes[1] : pagesizes[0]);
1294 map->anon_loc = addr;
1297 imgp->entry_addr = entry;
1299 if (interp != NULL) {
1300 VOP_UNLOCK(imgp->vp);
1301 if ((map->flags & MAP_ASLR) != 0) {
1302 /* Assume that interpreter fits into 1/4 of AS */
1303 maxv1 = maxv / 2 + addr / 2;
1304 MPASS(maxv1 >= addr); /* No overflow */
1305 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1308 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1310 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1317 * Construct auxargs table (used by the copyout_auxargs routine)
1319 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1320 if (elf_auxargs == NULL) {
1321 VOP_UNLOCK(imgp->vp);
1322 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1323 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1325 elf_auxargs->execfd = -1;
1326 elf_auxargs->phdr = proghdr + et_dyn_addr;
1327 elf_auxargs->phent = hdr->e_phentsize;
1328 elf_auxargs->phnum = hdr->e_phnum;
1329 elf_auxargs->pagesz = PAGE_SIZE;
1330 elf_auxargs->base = addr;
1331 elf_auxargs->flags = 0;
1332 elf_auxargs->entry = entry;
1333 elf_auxargs->hdr_eflags = hdr->e_flags;
1335 imgp->auxargs = elf_auxargs;
1336 imgp->interpreted = 0;
1337 imgp->reloc_base = addr;
1338 imgp->proc->p_osrel = osrel;
1339 imgp->proc->p_fctl0 = fctl0;
1340 imgp->proc->p_elf_flags = hdr->e_flags;
1344 free(interp, M_TEMP);
1348 #define elf_suword __CONCAT(suword, __ELF_WORD_SIZE)
1351 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1353 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1354 Elf_Auxinfo *argarray, *pos;
1357 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1360 if (args->execfd != -1)
1361 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1362 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1363 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1364 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1365 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1366 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1367 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1368 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1369 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1370 if (imgp->execpathp != 0)
1371 AUXARGS_ENTRY_PTR(pos, AT_EXECPATH, imgp->execpathp);
1372 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1373 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1374 if (imgp->canary != 0) {
1375 AUXARGS_ENTRY_PTR(pos, AT_CANARY, imgp->canary);
1376 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1378 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1379 if (imgp->pagesizes != 0) {
1380 AUXARGS_ENTRY_PTR(pos, AT_PAGESIZES, imgp->pagesizes);
1381 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1383 if (imgp->sysent->sv_timekeep_base != 0) {
1384 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1385 imgp->sysent->sv_timekeep_base);
1387 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1388 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1389 imgp->sysent->sv_stackprot);
1390 if (imgp->sysent->sv_hwcap != NULL)
1391 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1392 if (imgp->sysent->sv_hwcap2 != NULL)
1393 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1394 AUXARGS_ENTRY(pos, AT_BSDFLAGS, __elfN(sigfastblock) ?
1395 ELF_BSDF_SIGFASTBLK : 0);
1396 AUXARGS_ENTRY(pos, AT_ARGC, imgp->args->argc);
1397 AUXARGS_ENTRY_PTR(pos, AT_ARGV, imgp->argv);
1398 AUXARGS_ENTRY(pos, AT_ENVC, imgp->args->envc);
1399 AUXARGS_ENTRY_PTR(pos, AT_ENVV, imgp->envv);
1400 AUXARGS_ENTRY_PTR(pos, AT_PS_STRINGS, imgp->ps_strings);
1401 if (imgp->sysent->sv_fxrng_gen_base != 0)
1402 AUXARGS_ENTRY(pos, AT_FXRNG, imgp->sysent->sv_fxrng_gen_base);
1403 AUXARGS_ENTRY(pos, AT_NULL, 0);
1405 free(imgp->auxargs, M_TEMP);
1406 imgp->auxargs = NULL;
1407 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1409 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1410 free(argarray, M_TEMP);
1415 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1419 base = (Elf_Addr *)*stack_base;
1421 if (elf_suword(base, imgp->args->argc) == -1)
1423 *stack_base = (uintptr_t)base;
1428 * Code for generating ELF core dumps.
1431 typedef void (*segment_callback)(vm_map_entry_t, void *);
1433 /* Closure for cb_put_phdr(). */
1434 struct phdr_closure {
1435 Elf_Phdr *phdr; /* Program header to fill in */
1436 Elf_Off offset; /* Offset of segment in core file */
1440 int type; /* Note type. */
1441 outfunc_t outfunc; /* Output function. */
1442 void *outarg; /* Argument for the output function. */
1443 size_t outsize; /* Output size. */
1444 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1447 TAILQ_HEAD(note_info_list, note_info);
1449 extern int compress_user_cores;
1450 extern int compress_user_cores_level;
1452 static void cb_put_phdr(vm_map_entry_t, void *);
1453 static void cb_size_segment(vm_map_entry_t, void *);
1454 static void each_dumpable_segment(struct thread *, segment_callback, void *,
1456 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1457 struct note_info_list *, size_t, int);
1458 static void __elfN(putnote)(struct thread *td, struct note_info *, struct sbuf *);
1460 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1461 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1462 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1463 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1465 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1466 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1467 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1468 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1469 static void note_procstat_files(void *, struct sbuf *, size_t *);
1470 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1471 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1472 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1473 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1474 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1477 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1480 return (core_write((struct coredump_params *)arg, base, len, offset,
1481 UIO_SYSSPACE, NULL));
1485 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1487 struct ucred *cred = td->td_ucred;
1488 int compm, error = 0;
1489 struct sseg_closure seginfo;
1490 struct note_info_list notelst;
1491 struct coredump_params params;
1492 struct note_info *ninfo;
1494 size_t hdrsize, notesz, coresize;
1498 TAILQ_INIT(¬elst);
1500 /* Size the program segments. */
1501 __elfN(size_segments)(td, &seginfo, flags);
1504 * Collect info about the core file header area.
1506 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1507 if (seginfo.count + 1 >= PN_XNUM)
1508 hdrsize += sizeof(Elf_Shdr);
1509 td->td_proc->p_sysent->sv_elf_core_prepare_notes(td, ¬elst, ¬esz);
1510 coresize = round_page(hdrsize + notesz) + seginfo.size;
1512 /* Set up core dump parameters. */
1514 params.active_cred = cred;
1515 params.file_cred = NOCRED;
1522 PROC_LOCK(td->td_proc);
1523 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1524 PROC_UNLOCK(td->td_proc);
1531 if (coresize >= limit) {
1536 /* Create a compression stream if necessary. */
1537 compm = compress_user_cores;
1538 if ((flags & (SVC_PT_COREDUMP | SVC_NOCOMPRESS)) == SVC_PT_COREDUMP &&
1540 compm = COMPRESS_GZIP;
1542 params.comp = compressor_init(core_compressed_write,
1543 compm, CORE_BUF_SIZE,
1544 compress_user_cores_level, ¶ms);
1545 if (params.comp == NULL) {
1549 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1553 * Allocate memory for building the header, fill it up,
1554 * and write it out following the notes.
1556 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1557 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1560 /* Write the contents of all of the writable segments. */
1566 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1567 offset = round_page(hdrsize + notesz);
1568 for (i = 0; i < seginfo.count; i++) {
1569 error = core_output((char *)(uintptr_t)php->p_vaddr,
1570 php->p_filesz, offset, ¶ms, tmpbuf);
1573 offset += php->p_filesz;
1576 if (error == 0 && params.comp != NULL)
1577 error = compressor_flush(params.comp);
1581 "Failed to write core file for process %s (error %d)\n",
1582 curproc->p_comm, error);
1586 free(tmpbuf, M_TEMP);
1587 if (params.comp != NULL)
1588 compressor_fini(params.comp);
1589 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1590 TAILQ_REMOVE(¬elst, ninfo, link);
1591 free(ninfo, M_TEMP);
1600 * A callback for each_dumpable_segment() to write out the segment's
1601 * program header entry.
1604 cb_put_phdr(vm_map_entry_t entry, void *closure)
1606 struct phdr_closure *phc = (struct phdr_closure *)closure;
1607 Elf_Phdr *phdr = phc->phdr;
1609 phc->offset = round_page(phc->offset);
1611 phdr->p_type = PT_LOAD;
1612 phdr->p_offset = phc->offset;
1613 phdr->p_vaddr = entry->start;
1615 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1616 phdr->p_align = PAGE_SIZE;
1617 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1619 phc->offset += phdr->p_filesz;
1624 * A callback for each_dumpable_segment() to gather information about
1625 * the number of segments and their total size.
1628 cb_size_segment(vm_map_entry_t entry, void *closure)
1630 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1633 ssc->size += entry->end - entry->start;
1637 __elfN(size_segments)(struct thread *td, struct sseg_closure *seginfo,
1643 each_dumpable_segment(td, cb_size_segment, seginfo, flags);
1647 * For each writable segment in the process's memory map, call the given
1648 * function with a pointer to the map entry and some arbitrary
1649 * caller-supplied data.
1652 each_dumpable_segment(struct thread *td, segment_callback func, void *closure,
1655 struct proc *p = td->td_proc;
1656 vm_map_t map = &p->p_vmspace->vm_map;
1657 vm_map_entry_t entry;
1658 vm_object_t backing_object, object;
1661 vm_map_lock_read(map);
1662 VM_MAP_ENTRY_FOREACH(entry, map) {
1664 * Don't dump inaccessible mappings, deal with legacy
1667 * Note that read-only segments related to the elf binary
1668 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1669 * need to arbitrarily ignore such segments.
1671 if ((flags & SVC_ALL) == 0) {
1672 if (elf_legacy_coredump) {
1673 if ((entry->protection & VM_PROT_RW) !=
1677 if ((entry->protection & VM_PROT_ALL) == 0)
1683 * Dont include memory segment in the coredump if
1684 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1685 * madvise(2). Do not dump submaps (i.e. parts of the
1688 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
1690 if ((entry->eflags & MAP_ENTRY_NOCOREDUMP) != 0 &&
1691 (flags & SVC_ALL) == 0)
1693 if ((object = entry->object.vm_object) == NULL)
1696 /* Ignore memory-mapped devices and such things. */
1697 VM_OBJECT_RLOCK(object);
1698 while ((backing_object = object->backing_object) != NULL) {
1699 VM_OBJECT_RLOCK(backing_object);
1700 VM_OBJECT_RUNLOCK(object);
1701 object = backing_object;
1703 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1704 VM_OBJECT_RUNLOCK(object);
1708 (*func)(entry, closure);
1710 vm_map_unlock_read(map);
1714 * Write the core file header to the file, including padding up to
1715 * the page boundary.
1718 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1719 size_t hdrsize, struct note_info_list *notelst, size_t notesz,
1722 struct note_info *ninfo;
1726 /* Fill in the header. */
1727 bzero(hdr, hdrsize);
1728 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz, flags);
1730 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1731 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1732 sbuf_start_section(sb, NULL);
1733 sbuf_bcat(sb, hdr, hdrsize);
1734 TAILQ_FOREACH(ninfo, notelst, link)
1735 __elfN(putnote)(p->td, ninfo, sb);
1736 /* Align up to a page boundary for the program segments. */
1737 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1738 error = sbuf_finish(sb);
1745 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1755 size += __elfN(register_note)(td, list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1758 * To have the debugger select the right thread (LWP) as the initial
1759 * thread, we dump the state of the thread passed to us in td first.
1760 * This is the thread that causes the core dump and thus likely to
1761 * be the right thread one wants to have selected in the debugger.
1764 while (thr != NULL) {
1765 size += __elfN(register_note)(td, list, NT_PRSTATUS,
1766 __elfN(note_prstatus), thr);
1767 size += __elfN(register_note)(td, list, NT_FPREGSET,
1768 __elfN(note_fpregset), thr);
1769 size += __elfN(register_note)(td, list, NT_THRMISC,
1770 __elfN(note_thrmisc), thr);
1771 size += __elfN(register_note)(td, list, NT_PTLWPINFO,
1772 __elfN(note_ptlwpinfo), thr);
1773 size += __elfN(register_note)(td, list, -1,
1774 __elfN(note_threadmd), thr);
1776 thr = thr == td ? TAILQ_FIRST(&p->p_threads) :
1777 TAILQ_NEXT(thr, td_plist);
1779 thr = TAILQ_NEXT(thr, td_plist);
1782 size += __elfN(register_note)(td, list, NT_PROCSTAT_PROC,
1783 __elfN(note_procstat_proc), p);
1784 size += __elfN(register_note)(td, list, NT_PROCSTAT_FILES,
1785 note_procstat_files, p);
1786 size += __elfN(register_note)(td, list, NT_PROCSTAT_VMMAP,
1787 note_procstat_vmmap, p);
1788 size += __elfN(register_note)(td, list, NT_PROCSTAT_GROUPS,
1789 note_procstat_groups, p);
1790 size += __elfN(register_note)(td, list, NT_PROCSTAT_UMASK,
1791 note_procstat_umask, p);
1792 size += __elfN(register_note)(td, list, NT_PROCSTAT_RLIMIT,
1793 note_procstat_rlimit, p);
1794 size += __elfN(register_note)(td, list, NT_PROCSTAT_OSREL,
1795 note_procstat_osrel, p);
1796 size += __elfN(register_note)(td, list, NT_PROCSTAT_PSSTRINGS,
1797 __elfN(note_procstat_psstrings), p);
1798 size += __elfN(register_note)(td, list, NT_PROCSTAT_AUXV,
1799 __elfN(note_procstat_auxv), p);
1805 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1806 size_t notesz, int flags)
1811 struct phdr_closure phc;
1814 ehdr = (Elf_Ehdr *)hdr;
1815 bi = td->td_proc->p_elf_brandinfo;
1817 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1818 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1819 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1820 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1821 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1822 ehdr->e_ident[EI_DATA] = ELF_DATA;
1823 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1824 ehdr->e_ident[EI_OSABI] = td->td_proc->p_sysent->sv_elf_core_osabi;
1825 ehdr->e_ident[EI_ABIVERSION] = 0;
1826 ehdr->e_ident[EI_PAD] = 0;
1827 ehdr->e_type = ET_CORE;
1828 ehdr->e_machine = bi->machine;
1829 ehdr->e_version = EV_CURRENT;
1831 ehdr->e_phoff = sizeof(Elf_Ehdr);
1832 ehdr->e_flags = td->td_proc->p_elf_flags;
1833 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1834 ehdr->e_phentsize = sizeof(Elf_Phdr);
1835 ehdr->e_shentsize = sizeof(Elf_Shdr);
1836 ehdr->e_shstrndx = SHN_UNDEF;
1837 if (numsegs + 1 < PN_XNUM) {
1838 ehdr->e_phnum = numsegs + 1;
1841 ehdr->e_phnum = PN_XNUM;
1844 ehdr->e_shoff = ehdr->e_phoff +
1845 (numsegs + 1) * ehdr->e_phentsize;
1846 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1847 ("e_shoff: %zu, hdrsize - shdr: %zu",
1848 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1850 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1851 memset(shdr, 0, sizeof(*shdr));
1853 * A special first section is used to hold large segment and
1854 * section counts. This was proposed by Sun Microsystems in
1855 * Solaris and has been adopted by Linux; the standard ELF
1856 * tools are already familiar with the technique.
1858 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1859 * (or 12-7 depending on the version of the document) for more
1862 shdr->sh_type = SHT_NULL;
1863 shdr->sh_size = ehdr->e_shnum;
1864 shdr->sh_link = ehdr->e_shstrndx;
1865 shdr->sh_info = numsegs + 1;
1869 * Fill in the program header entries.
1871 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1873 /* The note segement. */
1874 phdr->p_type = PT_NOTE;
1875 phdr->p_offset = hdrsize;
1878 phdr->p_filesz = notesz;
1880 phdr->p_flags = PF_R;
1881 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1884 /* All the writable segments from the program. */
1886 phc.offset = round_page(hdrsize + notesz);
1887 each_dumpable_segment(td, cb_put_phdr, &phc, flags);
1891 __elfN(register_note)(struct thread *td, struct note_info_list *list,
1892 int type, outfunc_t out, void *arg)
1894 const struct sysentvec *sv;
1895 struct note_info *ninfo;
1896 size_t size, notesize;
1898 sv = td->td_proc->p_sysent;
1900 out(arg, NULL, &size);
1901 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1903 ninfo->outfunc = out;
1904 ninfo->outarg = arg;
1905 ninfo->outsize = size;
1906 TAILQ_INSERT_TAIL(list, ninfo, link);
1911 notesize = sizeof(Elf_Note) + /* note header */
1912 roundup2(strlen(sv->sv_elf_core_abi_vendor) + 1, ELF_NOTE_ROUNDSIZE) +
1914 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1920 append_note_data(const void *src, void *dst, size_t len)
1924 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1926 bcopy(src, dst, len);
1927 bzero((char *)dst + len, padded_len - len);
1929 return (padded_len);
1933 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1941 note = (Elf_Note *)buf;
1942 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1943 note->n_descsz = size;
1944 note->n_type = type;
1945 buf += sizeof(*note);
1946 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1947 sizeof(FREEBSD_ABI_VENDOR));
1948 append_note_data(src, buf, size);
1953 notesize = sizeof(Elf_Note) + /* note header */
1954 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1956 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1962 __elfN(putnote)(struct thread *td, struct note_info *ninfo, struct sbuf *sb)
1965 const struct sysentvec *sv;
1966 ssize_t old_len, sect_len;
1967 size_t new_len, descsz, i;
1969 if (ninfo->type == -1) {
1970 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1974 sv = td->td_proc->p_sysent;
1976 note.n_namesz = strlen(sv->sv_elf_core_abi_vendor) + 1;
1977 note.n_descsz = ninfo->outsize;
1978 note.n_type = ninfo->type;
1980 sbuf_bcat(sb, ¬e, sizeof(note));
1981 sbuf_start_section(sb, &old_len);
1982 sbuf_bcat(sb, sv->sv_elf_core_abi_vendor,
1983 strlen(sv->sv_elf_core_abi_vendor) + 1);
1984 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1985 if (note.n_descsz == 0)
1987 sbuf_start_section(sb, &old_len);
1988 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1989 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1993 new_len = (size_t)sect_len;
1994 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1995 if (new_len < descsz) {
1997 * It is expected that individual note emitters will correctly
1998 * predict their expected output size and fill up to that size
1999 * themselves, padding in a format-specific way if needed.
2000 * However, in case they don't, just do it here with zeros.
2002 for (i = 0; i < descsz - new_len; i++)
2004 } else if (new_len > descsz) {
2006 * We can't always truncate sb -- we may have drained some
2009 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2010 "read it (%zu > %zu). Since it is longer than "
2011 "expected, this coredump's notes are corrupt. THIS "
2012 "IS A BUG in the note_procstat routine for type %u.\n",
2013 __func__, (unsigned)note.n_type, new_len, descsz,
2014 (unsigned)note.n_type));
2019 * Miscellaneous note out functions.
2022 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2023 #include <compat/freebsd32/freebsd32.h>
2024 #include <compat/freebsd32/freebsd32_signal.h>
2026 typedef struct prstatus32 elf_prstatus_t;
2027 typedef struct prpsinfo32 elf_prpsinfo_t;
2028 typedef struct fpreg32 elf_prfpregset_t;
2029 typedef struct fpreg32 elf_fpregset_t;
2030 typedef struct reg32 elf_gregset_t;
2031 typedef struct thrmisc32 elf_thrmisc_t;
2032 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2033 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2034 typedef uint32_t elf_ps_strings_t;
2036 typedef prstatus_t elf_prstatus_t;
2037 typedef prpsinfo_t elf_prpsinfo_t;
2038 typedef prfpregset_t elf_prfpregset_t;
2039 typedef prfpregset_t elf_fpregset_t;
2040 typedef gregset_t elf_gregset_t;
2041 typedef thrmisc_t elf_thrmisc_t;
2042 #define ELF_KERN_PROC_MASK 0
2043 typedef struct kinfo_proc elf_kinfo_proc_t;
2044 typedef vm_offset_t elf_ps_strings_t;
2048 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2054 elf_prpsinfo_t *psinfo;
2059 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2060 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2061 psinfo->pr_version = PRPSINFO_VERSION;
2062 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2063 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2065 if (p->p_args != NULL) {
2066 len = sizeof(psinfo->pr_psargs) - 1;
2067 if (len > p->p_args->ar_length)
2068 len = p->p_args->ar_length;
2069 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2075 sbuf_new(&sbarg, psinfo->pr_psargs,
2076 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2077 error = proc_getargv(curthread, p, &sbarg);
2079 if (sbuf_finish(&sbarg) == 0)
2080 len = sbuf_len(&sbarg) - 1;
2082 len = sizeof(psinfo->pr_psargs) - 1;
2083 sbuf_delete(&sbarg);
2085 if (error || len == 0)
2086 strlcpy(psinfo->pr_psargs, p->p_comm,
2087 sizeof(psinfo->pr_psargs));
2089 KASSERT(len < sizeof(psinfo->pr_psargs),
2090 ("len is too long: %zu vs %zu", len,
2091 sizeof(psinfo->pr_psargs)));
2092 cp = psinfo->pr_psargs;
2095 cp = memchr(cp, '\0', end - cp);
2101 psinfo->pr_pid = p->p_pid;
2102 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2103 free(psinfo, M_TEMP);
2105 *sizep = sizeof(*psinfo);
2109 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2112 elf_prstatus_t *status;
2116 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2117 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2118 status->pr_version = PRSTATUS_VERSION;
2119 status->pr_statussz = sizeof(elf_prstatus_t);
2120 status->pr_gregsetsz = sizeof(elf_gregset_t);
2121 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2122 status->pr_osreldate = osreldate;
2123 status->pr_cursig = td->td_proc->p_sig;
2124 status->pr_pid = td->td_tid;
2125 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2126 fill_regs32(td, &status->pr_reg);
2128 fill_regs(td, &status->pr_reg);
2130 sbuf_bcat(sb, status, sizeof(*status));
2131 free(status, M_TEMP);
2133 *sizep = sizeof(*status);
2137 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2140 elf_prfpregset_t *fpregset;
2144 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2145 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2146 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2147 fill_fpregs32(td, fpregset);
2149 fill_fpregs(td, fpregset);
2151 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2152 free(fpregset, M_TEMP);
2154 *sizep = sizeof(*fpregset);
2158 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2161 elf_thrmisc_t thrmisc;
2165 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2166 bzero(&thrmisc, sizeof(thrmisc));
2167 strcpy(thrmisc.pr_tname, td->td_name);
2168 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2170 *sizep = sizeof(thrmisc);
2174 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2179 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2180 struct ptrace_lwpinfo32 pl;
2182 struct ptrace_lwpinfo pl;
2186 size = sizeof(structsize) + sizeof(pl);
2188 KASSERT(*sizep == size, ("invalid size"));
2189 structsize = sizeof(pl);
2190 sbuf_bcat(sb, &structsize, sizeof(structsize));
2191 bzero(&pl, sizeof(pl));
2192 pl.pl_lwpid = td->td_tid;
2193 pl.pl_event = PL_EVENT_NONE;
2194 pl.pl_sigmask = td->td_sigmask;
2195 pl.pl_siglist = td->td_siglist;
2196 if (td->td_si.si_signo != 0) {
2197 pl.pl_event = PL_EVENT_SIGNAL;
2198 pl.pl_flags |= PL_FLAG_SI;
2199 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2200 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2202 pl.pl_siginfo = td->td_si;
2205 strcpy(pl.pl_tdname, td->td_name);
2206 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2207 sbuf_bcat(sb, &pl, sizeof(pl));
2213 * Allow for MD specific notes, as well as any MD
2214 * specific preparations for writing MI notes.
2217 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2223 td = (struct thread *)arg;
2225 if (size != 0 && sb != NULL)
2226 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2230 __elfN(dump_thread)(td, buf, &size);
2231 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2232 if (size != 0 && sb != NULL)
2233 sbuf_bcat(sb, buf, size);
2238 #ifdef KINFO_PROC_SIZE
2239 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2243 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2250 size = sizeof(structsize) + p->p_numthreads *
2251 sizeof(elf_kinfo_proc_t);
2254 KASSERT(*sizep == size, ("invalid size"));
2255 structsize = sizeof(elf_kinfo_proc_t);
2256 sbuf_bcat(sb, &structsize, sizeof(structsize));
2257 sx_slock(&proctree_lock);
2259 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2260 sx_sunlock(&proctree_lock);
2265 #ifdef KINFO_FILE_SIZE
2266 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2270 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2273 size_t size, sect_sz, i;
2274 ssize_t start_len, sect_len;
2275 int structsize, filedesc_flags;
2277 if (coredump_pack_fileinfo)
2278 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2283 structsize = sizeof(struct kinfo_file);
2286 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2287 sbuf_set_drain(sb, sbuf_count_drain, &size);
2288 sbuf_bcat(sb, &structsize, sizeof(structsize));
2290 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2295 sbuf_start_section(sb, &start_len);
2297 sbuf_bcat(sb, &structsize, sizeof(structsize));
2299 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2302 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2307 KASSERT(sect_sz <= *sizep,
2308 ("kern_proc_filedesc_out did not respect maxlen; "
2309 "requested %zu, got %zu", *sizep - sizeof(structsize),
2310 sect_sz - sizeof(structsize)));
2312 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2317 #ifdef KINFO_VMENTRY_SIZE
2318 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2322 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2326 int structsize, vmmap_flags;
2328 if (coredump_pack_vmmapinfo)
2329 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2334 structsize = sizeof(struct kinfo_vmentry);
2337 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2338 sbuf_set_drain(sb, sbuf_count_drain, &size);
2339 sbuf_bcat(sb, &structsize, sizeof(structsize));
2341 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2346 sbuf_bcat(sb, &structsize, sizeof(structsize));
2348 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2354 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2361 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2363 KASSERT(*sizep == size, ("invalid size"));
2364 structsize = sizeof(gid_t);
2365 sbuf_bcat(sb, &structsize, sizeof(structsize));
2366 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2373 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2380 size = sizeof(structsize) + sizeof(p->p_pd->pd_cmask);
2382 KASSERT(*sizep == size, ("invalid size"));
2383 structsize = sizeof(p->p_pd->pd_cmask);
2384 sbuf_bcat(sb, &structsize, sizeof(structsize));
2385 sbuf_bcat(sb, &p->p_pd->pd_cmask, sizeof(p->p_pd->pd_cmask));
2391 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2394 struct rlimit rlim[RLIM_NLIMITS];
2399 size = sizeof(structsize) + sizeof(rlim);
2401 KASSERT(*sizep == size, ("invalid size"));
2402 structsize = sizeof(rlim);
2403 sbuf_bcat(sb, &structsize, sizeof(structsize));
2405 for (i = 0; i < RLIM_NLIMITS; i++)
2406 lim_rlimit_proc(p, i, &rlim[i]);
2408 sbuf_bcat(sb, rlim, sizeof(rlim));
2414 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2421 size = sizeof(structsize) + sizeof(p->p_osrel);
2423 KASSERT(*sizep == size, ("invalid size"));
2424 structsize = sizeof(p->p_osrel);
2425 sbuf_bcat(sb, &structsize, sizeof(structsize));
2426 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2432 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2435 elf_ps_strings_t ps_strings;
2440 size = sizeof(structsize) + sizeof(ps_strings);
2442 KASSERT(*sizep == size, ("invalid size"));
2443 structsize = sizeof(ps_strings);
2444 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2445 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2447 ps_strings = p->p_sysent->sv_psstrings;
2449 sbuf_bcat(sb, &structsize, sizeof(structsize));
2450 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2456 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2465 sb = sbuf_new(NULL, NULL, AT_COUNT * sizeof(Elf_Auxinfo),
2467 sbuf_set_drain(sb, sbuf_count_drain, &size);
2468 sbuf_bcat(sb, &structsize, sizeof(structsize));
2470 proc_getauxv(curthread, p, sb);
2476 structsize = sizeof(Elf_Auxinfo);
2477 sbuf_bcat(sb, &structsize, sizeof(structsize));
2479 proc_getauxv(curthread, p, sb);
2485 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2486 const char *note_vendor, const Elf_Phdr *pnote,
2487 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2489 const Elf_Note *note, *note0, *note_end;
2490 const char *note_name;
2495 /* We need some limit, might as well use PAGE_SIZE. */
2496 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2498 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2499 if (pnote->p_offset > PAGE_SIZE ||
2500 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2501 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2503 VOP_UNLOCK(imgp->vp);
2504 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2505 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2507 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2508 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2509 curthread->td_ucred, NOCRED, NULL, curthread);
2511 uprintf("i/o error PT_NOTE\n");
2514 note = note0 = (const Elf_Note *)buf;
2515 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2517 note = note0 = (const Elf_Note *)(imgp->image_header +
2519 note_end = (const Elf_Note *)(imgp->image_header +
2520 pnote->p_offset + pnote->p_filesz);
2523 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2524 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2525 (const char *)note < sizeof(Elf_Note)) {
2528 if (note->n_namesz != checknote->n_namesz ||
2529 note->n_descsz != checknote->n_descsz ||
2530 note->n_type != checknote->n_type)
2532 note_name = (const char *)(note + 1);
2533 if (note_name + checknote->n_namesz >=
2534 (const char *)note_end || strncmp(note_vendor,
2535 note_name, checknote->n_namesz) != 0)
2538 if (cb(note, cb_arg, &res))
2541 note = (const Elf_Note *)((const char *)(note + 1) +
2542 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2543 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2552 struct brandnote_cb_arg {
2553 Elf_Brandnote *brandnote;
2558 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2560 struct brandnote_cb_arg *arg;
2565 * Fetch the osreldate for binary from the ELF OSABI-note if
2568 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2569 arg->brandnote->trans_osrel != NULL ?
2570 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2575 static Elf_Note fctl_note = {
2576 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2577 .n_descsz = sizeof(uint32_t),
2578 .n_type = NT_FREEBSD_FEATURE_CTL,
2581 struct fctl_cb_arg {
2582 boolean_t *has_fctl0;
2587 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2589 struct fctl_cb_arg *arg;
2590 const Elf32_Word *desc;
2594 p = (uintptr_t)(note + 1);
2595 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2596 desc = (const Elf32_Word *)p;
2597 *arg->has_fctl0 = TRUE;
2598 *arg->fctl0 = desc[0];
2604 * Try to find the appropriate ABI-note section for checknote, fetch
2605 * the osreldate and feature control flags for binary from the ELF
2606 * OSABI-note. Only the first page of the image is searched, the same
2610 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2611 int32_t *osrel, boolean_t *has_fctl0, uint32_t *fctl0)
2613 const Elf_Phdr *phdr;
2614 const Elf_Ehdr *hdr;
2615 struct brandnote_cb_arg b_arg;
2616 struct fctl_cb_arg f_arg;
2619 hdr = (const Elf_Ehdr *)imgp->image_header;
2620 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2621 b_arg.brandnote = brandnote;
2622 b_arg.osrel = osrel;
2623 f_arg.has_fctl0 = has_fctl0;
2624 f_arg.fctl0 = fctl0;
2626 for (i = 0; i < hdr->e_phnum; i++) {
2627 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2628 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2630 for (j = 0; j < hdr->e_phnum; j++) {
2631 if (phdr[j].p_type == PT_NOTE &&
2632 __elfN(parse_notes)(imgp, &fctl_note,
2633 FREEBSD_ABI_VENDOR, &phdr[j],
2634 note_fctl_cb, &f_arg))
2645 * Tell kern_execve.c about it, with a little help from the linker.
2647 static struct execsw __elfN(execsw) = {
2648 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2649 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2651 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2654 __elfN(trans_prot)(Elf_Word flags)
2660 prot |= VM_PROT_EXECUTE;
2662 prot |= VM_PROT_WRITE;
2664 prot |= VM_PROT_READ;
2665 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2666 if (i386_read_exec && (flags & PF_R))
2667 prot |= VM_PROT_EXECUTE;
2673 __elfN(untrans_prot)(vm_prot_t prot)
2678 if (prot & VM_PROT_EXECUTE)
2680 if (prot & VM_PROT_READ)
2682 if (prot & VM_PROT_WRITE)
2688 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2690 uintptr_t range, rbase, gap;
2693 pct = __elfN(aslr_stack_gap);
2698 range = imgp->eff_stack_sz * pct / 100;
2699 arc4rand(&rbase, sizeof(rbase), 0);
2700 gap = rbase % range;
2701 gap &= ~(sizeof(u_long) - 1);