2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108 #define CORE_BUF_SIZE (16 * 1024)
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
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
132 #if defined(__amd64__)
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");
139 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
141 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
143 static int __elfN(aslr_enabled) = 0;
144 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
145 &__elfN(aslr_enabled), 0,
146 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
147 ": enable address map randomization");
149 static int __elfN(pie_aslr_enabled) = 0;
150 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
151 &__elfN(pie_aslr_enabled), 0,
152 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
153 ": enable address map randomization for PIE binaries");
155 static int __elfN(aslr_honor_sbrk) = 1;
156 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
157 &__elfN(aslr_honor_sbrk), 0,
158 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
160 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
162 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
164 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
166 Elf_Brandnote __elfN(freebsd_brandnote) = {
167 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
168 .hdr.n_descsz = sizeof(int32_t),
169 .hdr.n_type = NT_FREEBSD_ABI_TAG,
170 .vendor = FREEBSD_ABI_VENDOR,
171 .flags = BN_TRANSLATE_OSREL,
172 .trans_osrel = __elfN(freebsd_trans_osrel)
176 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
180 p = (uintptr_t)(note + 1);
181 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
182 *osrel = *(const int32_t *)(p);
187 static const char GNU_ABI_VENDOR[] = "GNU";
188 static int GNU_KFREEBSD_ABI_DESC = 3;
190 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
191 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
192 .hdr.n_descsz = 16, /* XXX at least 16 */
194 .vendor = GNU_ABI_VENDOR,
195 .flags = BN_TRANSLATE_OSREL,
196 .trans_osrel = kfreebsd_trans_osrel
200 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
202 const Elf32_Word *desc;
205 p = (uintptr_t)(note + 1);
206 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
208 desc = (const Elf32_Word *)p;
209 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
213 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
214 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
216 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
222 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
226 for (i = 0; i < MAX_BRANDS; i++) {
227 if (elf_brand_list[i] == NULL) {
228 elf_brand_list[i] = entry;
232 if (i == MAX_BRANDS) {
233 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
241 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
245 for (i = 0; i < MAX_BRANDS; i++) {
246 if (elf_brand_list[i] == entry) {
247 elf_brand_list[i] = NULL;
257 __elfN(brand_inuse)(Elf_Brandinfo *entry)
262 sx_slock(&allproc_lock);
263 FOREACH_PROC_IN_SYSTEM(p) {
264 if (p->p_sysent == entry->sysvec) {
269 sx_sunlock(&allproc_lock);
274 static Elf_Brandinfo *
275 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
276 int32_t *osrel, uint32_t *fctl0)
278 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
279 Elf_Brandinfo *bi, *bi_m;
281 int i, interp_name_len;
283 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
286 * We support four types of branding -- (1) the ELF EI_OSABI field
287 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
288 * branding w/in the ELF header, (3) path of the `interp_path'
289 * field, and (4) the ".note.ABI-tag" ELF section.
292 /* Look for an ".note.ABI-tag" ELF section */
294 for (i = 0; i < MAX_BRANDS; i++) {
295 bi = elf_brand_list[i];
298 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
300 if (hdr->e_machine == bi->machine && (bi->flags &
301 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
302 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
304 /* Give brand a chance to veto check_note's guess */
305 if (ret && bi->header_supported)
306 ret = bi->header_supported(imgp);
308 * If note checker claimed the binary, but the
309 * interpreter path in the image does not
310 * match default one for the brand, try to
311 * search for other brands with the same
312 * interpreter. Either there is better brand
313 * with the right interpreter, or, failing
314 * this, we return first brand which accepted
315 * our note and, optionally, header.
317 if (ret && bi_m == NULL && interp != NULL &&
318 (bi->interp_path == NULL ||
319 (strlen(bi->interp_path) + 1 != interp_name_len ||
320 strncmp(interp, bi->interp_path, interp_name_len)
332 /* If the executable has a brand, search for it in the brand list. */
333 for (i = 0; i < MAX_BRANDS; i++) {
334 bi = elf_brand_list[i];
335 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
336 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
338 if (hdr->e_machine == bi->machine &&
339 (hdr->e_ident[EI_OSABI] == bi->brand ||
340 (bi->compat_3_brand != NULL &&
341 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
342 bi->compat_3_brand) == 0))) {
343 /* Looks good, but give brand a chance to veto */
344 if (bi->header_supported == NULL ||
345 bi->header_supported(imgp)) {
347 * Again, prefer strictly matching
350 if (interp_name_len == 0 &&
351 bi->interp_path == NULL)
353 if (bi->interp_path != NULL &&
354 strlen(bi->interp_path) + 1 ==
355 interp_name_len && strncmp(interp,
356 bi->interp_path, interp_name_len) == 0)
366 /* No known brand, see if the header is recognized by any brand */
367 for (i = 0; i < MAX_BRANDS; i++) {
368 bi = elf_brand_list[i];
369 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
370 bi->header_supported == NULL)
372 if (hdr->e_machine == bi->machine) {
373 ret = bi->header_supported(imgp);
379 /* Lacking a known brand, search for a recognized interpreter. */
380 if (interp != NULL) {
381 for (i = 0; i < MAX_BRANDS; i++) {
382 bi = elf_brand_list[i];
383 if (bi == NULL || (bi->flags &
384 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
387 if (hdr->e_machine == bi->machine &&
388 bi->interp_path != NULL &&
389 /* ELF image p_filesz includes terminating zero */
390 strlen(bi->interp_path) + 1 == interp_name_len &&
391 strncmp(interp, bi->interp_path, interp_name_len)
392 == 0 && (bi->header_supported == NULL ||
393 bi->header_supported(imgp)))
398 /* Lacking a recognized interpreter, try the default brand */
399 for (i = 0; i < MAX_BRANDS; i++) {
400 bi = elf_brand_list[i];
401 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
402 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
404 if (hdr->e_machine == bi->machine &&
405 __elfN(fallback_brand) == bi->brand &&
406 (bi->header_supported == NULL ||
407 bi->header_supported(imgp)))
414 __elfN(check_header)(const Elf_Ehdr *hdr)
420 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
421 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
422 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
423 hdr->e_phentsize != sizeof(Elf_Phdr) ||
424 hdr->e_version != ELF_TARG_VER)
428 * Make sure we have at least one brand for this machine.
431 for (i = 0; i < MAX_BRANDS; i++) {
432 bi = elf_brand_list[i];
433 if (bi != NULL && bi->machine == hdr->e_machine)
443 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
444 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
451 * Create the page if it doesn't exist yet. Ignore errors.
453 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
454 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
457 * Find the page from the underlying object.
459 if (object != NULL) {
460 sf = vm_imgact_map_page(object, offset);
462 return (KERN_FAILURE);
463 off = offset - trunc_page(offset);
464 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
466 vm_imgact_unmap_page(sf);
468 return (KERN_FAILURE);
471 return (KERN_SUCCESS);
475 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
476 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
482 int error, locked, rv;
484 if (start != trunc_page(start)) {
485 rv = __elfN(map_partial)(map, object, offset, start,
486 round_page(start), prot);
487 if (rv != KERN_SUCCESS)
489 offset += round_page(start) - start;
490 start = round_page(start);
492 if (end != round_page(end)) {
493 rv = __elfN(map_partial)(map, object, offset +
494 trunc_page(end) - start, trunc_page(end), end, prot);
495 if (rv != KERN_SUCCESS)
497 end = trunc_page(end);
500 return (KERN_SUCCESS);
501 if ((offset & PAGE_MASK) != 0) {
503 * The mapping is not page aligned. This means that we have
506 rv = vm_map_fixed(map, NULL, 0, start, end - start,
507 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
508 if (rv != KERN_SUCCESS)
511 return (KERN_SUCCESS);
512 for (; start < end; start += sz) {
513 sf = vm_imgact_map_page(object, offset);
515 return (KERN_FAILURE);
516 off = offset - trunc_page(offset);
518 if (sz > PAGE_SIZE - off)
519 sz = PAGE_SIZE - off;
520 error = copyout((caddr_t)sf_buf_kva(sf) + off,
522 vm_imgact_unmap_page(sf);
524 return (KERN_FAILURE);
528 vm_object_reference(object);
529 rv = vm_map_fixed(map, object, offset, start, end - start,
530 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
531 if (rv != KERN_SUCCESS) {
532 locked = VOP_ISLOCKED(imgp->vp);
533 VOP_UNLOCK(imgp->vp, 0);
534 vm_object_deallocate(object);
535 vn_lock(imgp->vp, locked | LK_RETRY);
539 return (KERN_SUCCESS);
543 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
544 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
550 vm_offset_t off, map_addr;
553 vm_ooffset_t file_addr;
556 * It's necessary to fail if the filsz + offset taken from the
557 * header is greater than the actual file pager object's size.
558 * If we were to allow this, then the vm_map_find() below would
559 * walk right off the end of the file object and into the ether.
561 * While I'm here, might as well check for something else that
562 * is invalid: filsz cannot be greater than memsz.
564 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
566 uprintf("elf_load_section: truncated ELF file\n");
570 object = imgp->object;
571 map = &imgp->proc->p_vmspace->vm_map;
572 map_addr = trunc_page((vm_offset_t)vmaddr);
573 file_addr = trunc_page(offset);
576 * We have two choices. We can either clear the data in the last page
577 * of an oversized mapping, or we can start the anon mapping a page
578 * early and copy the initialized data into that first page. We
583 else if (memsz > filsz)
584 map_len = trunc_page(offset + filsz) - file_addr;
586 map_len = round_page(offset + filsz) - file_addr;
589 /* cow flags: don't dump readonly sections in core */
590 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
591 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
593 rv = __elfN(map_insert)(imgp, map,
595 file_addr, /* file offset */
596 map_addr, /* virtual start */
597 map_addr + map_len,/* virtual end */
600 if (rv != KERN_SUCCESS)
603 /* we can stop now if we've covered it all */
610 * We have to get the remaining bit of the file into the first part
611 * of the oversized map segment. This is normally because the .data
612 * segment in the file is extended to provide bss. It's a neat idea
613 * to try and save a page, but it's a pain in the behind to implement.
615 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
617 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
618 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
620 /* This had damn well better be true! */
622 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
623 map_addr + map_len, prot, 0);
624 if (rv != KERN_SUCCESS)
629 sf = vm_imgact_map_page(object, offset + filsz);
633 /* send the page fragment to user space */
634 off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
635 error = copyout((caddr_t)sf_buf_kva(sf) + off,
636 (caddr_t)map_addr, copy_len);
637 vm_imgact_unmap_page(sf);
643 * Remove write access to the page if it was only granted by map_insert
646 if ((prot & VM_PROT_WRITE) == 0)
647 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
648 map_len), prot, FALSE);
654 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
655 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
665 for (i = 0; i < hdr->e_phnum; i++) {
666 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
669 /* Loadable segment */
670 prot = __elfN(trans_prot)(phdr[i].p_flags);
671 error = __elfN(load_section)(imgp, phdr[i].p_offset,
672 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
673 phdr[i].p_memsz, phdr[i].p_filesz, prot);
678 * Establish the base address if this is the first segment.
681 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
686 if (base_addrp != NULL)
687 *base_addrp = base_addr;
693 * Load the file "file" into memory. It may be either a shared object
696 * The "addr" reference parameter is in/out. On entry, it specifies
697 * the address where a shared object should be loaded. If the file is
698 * an executable, this value is ignored. On exit, "addr" specifies
699 * where the file was actually loaded.
701 * The "entry" reference parameter is out only. On exit, it specifies
702 * the entry point for the loaded file.
705 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
711 struct image_params image_params;
713 const Elf_Ehdr *hdr = NULL;
714 const Elf_Phdr *phdr = NULL;
715 struct nameidata *nd;
717 struct image_params *imgp;
719 u_long base_addr = 0;
722 #ifdef CAPABILITY_MODE
724 * XXXJA: This check can go away once we are sufficiently confident
725 * that the checks in namei() are correct.
727 if (IN_CAPABILITY_MODE(curthread))
731 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
733 attr = &tempdata->attr;
734 imgp = &tempdata->image_params;
737 * Initialize part of the common data
741 imgp->firstpage = NULL;
742 imgp->image_header = NULL;
744 imgp->execlabel = NULL;
746 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
747 if ((error = namei(nd)) != 0) {
751 NDFREE(nd, NDF_ONLY_PNBUF);
752 imgp->vp = nd->ni_vp;
755 * Check permissions, modes, uid, etc on the file, and "open" it.
757 error = exec_check_permissions(imgp);
761 error = exec_map_first_page(imgp);
766 * Also make certain that the interpreter stays the same, so set
767 * its VV_TEXT flag, too.
769 VOP_SET_TEXT(nd->ni_vp);
771 imgp->object = nd->ni_vp->v_object;
773 hdr = (const Elf_Ehdr *)imgp->image_header;
774 if ((error = __elfN(check_header)(hdr)) != 0)
776 if (hdr->e_type == ET_DYN)
778 else if (hdr->e_type == ET_EXEC)
785 /* Only support headers that fit within first page for now */
786 if ((hdr->e_phoff > PAGE_SIZE) ||
787 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
792 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
793 if (!aligned(phdr, Elf_Addr)) {
798 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
803 *entry = (unsigned long)hdr->e_entry + rbase;
807 exec_unmap_first_page(imgp);
812 free(tempdata, M_TEMP);
818 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
823 MPASS(vm_map_min(map) <= minv);
824 MPASS(maxv <= vm_map_max(map));
826 MPASS(minv + align < maxv);
827 arc4rand(&rbase, sizeof(rbase), 0);
828 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
829 res &= ~((u_long)align - 1);
833 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
834 res, minv, maxv, rbase));
836 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
837 res, maxv, minv, rbase));
842 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
843 const Elf_Phdr *phdr, u_long et_dyn_addr)
845 struct vmspace *vmspace;
847 u_long text_size, data_size, total_size, text_addr, data_addr;
848 u_long seg_size, seg_addr;
852 text_size = data_size = total_size = text_addr = data_addr = 0;
854 for (i = 0; i < hdr->e_phnum; i++) {
855 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
858 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
859 seg_size = round_page(phdr[i].p_memsz +
860 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
863 * Make the largest executable segment the official
864 * text segment and all others data.
866 * Note that obreak() assumes that data_addr + data_size == end
867 * of data load area, and the ELF file format expects segments
868 * to be sorted by address. If multiple data segments exist,
869 * the last one will be used.
872 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
873 text_size = seg_size;
874 text_addr = seg_addr;
876 data_size = seg_size;
877 data_addr = seg_addr;
879 total_size += seg_size;
882 if (data_addr == 0 && data_size == 0) {
883 data_addr = text_addr;
884 data_size = text_size;
888 * Check limits. It should be safe to check the
889 * limits after loading the segments since we do
890 * not actually fault in all the segments pages.
892 PROC_LOCK(imgp->proc);
893 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
894 err_str = "Data segment size exceeds process limit";
895 else if (text_size > maxtsiz)
896 err_str = "Text segment size exceeds system limit";
897 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
898 err_str = "Total segment size exceeds process limit";
899 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
900 err_str = "Data segment size exceeds resource limit";
901 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
902 err_str = "Total segment size exceeds resource limit";
903 PROC_UNLOCK(imgp->proc);
904 if (err_str != NULL) {
905 uprintf("%s\n", err_str);
909 vmspace = imgp->proc->p_vmspace;
910 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
911 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
912 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
913 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
919 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
920 char **interpp, bool *free_interpp)
924 int error, interp_name_len;
926 KASSERT(phdr->p_type == PT_INTERP,
927 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
928 KASSERT(VOP_ISLOCKED(imgp->vp),
929 ("%s: vp %p is not locked", __func__, imgp->vp));
933 /* Path to interpreter */
934 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
935 uprintf("Invalid PT_INTERP\n");
939 interp_name_len = phdr->p_filesz;
940 if (phdr->p_offset > PAGE_SIZE ||
941 interp_name_len > PAGE_SIZE - phdr->p_offset) {
942 VOP_UNLOCK(imgp->vp, 0);
943 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
944 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
945 error = vn_rdwr(UIO_READ, imgp->vp, interp,
946 interp_name_len, phdr->p_offset,
947 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
950 free(interp, M_TEMP);
951 uprintf("i/o error PT_INTERP %d\n", error);
954 interp[interp_name_len] = '\0';
957 *free_interpp = true;
961 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
962 if (interp[interp_name_len - 1] != '\0') {
963 uprintf("Invalid PT_INTERP\n");
968 *free_interpp = false;
973 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
974 const char *interp, u_long *addr, u_long *entry)
979 if (brand_info->emul_path != NULL &&
980 brand_info->emul_path[0] != '\0') {
981 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
982 snprintf(path, MAXPATHLEN, "%s%s",
983 brand_info->emul_path, interp);
984 error = __elfN(load_file)(imgp->proc, path, addr, entry);
990 if (brand_info->interp_newpath != NULL &&
991 (brand_info->interp_path == NULL ||
992 strcmp(interp, brand_info->interp_path) == 0)) {
993 error = __elfN(load_file)(imgp->proc,
994 brand_info->interp_newpath, addr, entry);
999 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1003 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1008 * Impossible et_dyn_addr initial value indicating that the real base
1009 * must be calculated later with some randomization applied.
1011 #define ET_DYN_ADDR_RAND 1
1014 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1017 const Elf_Ehdr *hdr;
1018 const Elf_Phdr *phdr;
1019 Elf_Auxargs *elf_auxargs;
1020 struct vmspace *vmspace;
1023 Elf_Brandinfo *brand_info;
1024 struct sysentvec *sv;
1025 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1026 u_long maxalign, mapsz, maxv, maxv1;
1032 hdr = (const Elf_Ehdr *)imgp->image_header;
1035 * Do we have a valid ELF header ?
1037 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1038 * if particular brand doesn't support it.
1040 if (__elfN(check_header)(hdr) != 0 ||
1041 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1045 * From here on down, we return an errno, not -1, as we've
1046 * detected an ELF file.
1049 if ((hdr->e_phoff > PAGE_SIZE) ||
1050 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1051 /* Only support headers in first page for now */
1052 uprintf("Program headers not in the first page\n");
1055 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1056 if (!aligned(phdr, Elf_Addr)) {
1057 uprintf("Unaligned program headers\n");
1065 entry = proghdr = 0;
1067 free_interp = false;
1069 maxalign = PAGE_SIZE;
1072 for (i = 0; i < hdr->e_phnum; i++) {
1073 switch (phdr[i].p_type) {
1076 baddr = phdr[i].p_vaddr;
1077 if (phdr[i].p_align > maxalign)
1078 maxalign = phdr[i].p_align;
1079 mapsz += phdr[i].p_memsz;
1083 * If this segment contains the program headers,
1084 * remember their virtual address for the AT_PHDR
1085 * aux entry. Static binaries don't usually include
1088 if (phdr[i].p_offset == 0 &&
1089 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1090 <= phdr[i].p_filesz)
1091 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1094 /* Path to interpreter */
1095 if (interp != NULL) {
1096 uprintf("Multiple PT_INTERP headers\n");
1100 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1106 if (__elfN(nxstack))
1108 __elfN(trans_prot)(phdr[i].p_flags);
1109 imgp->stack_sz = phdr[i].p_memsz;
1111 case PT_PHDR: /* Program header table info */
1112 proghdr = phdr[i].p_vaddr;
1117 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1118 if (brand_info == NULL) {
1119 uprintf("ELF binary type \"%u\" not known.\n",
1120 hdr->e_ident[EI_OSABI]);
1124 sv = brand_info->sysvec;
1126 if (hdr->e_type == ET_DYN) {
1127 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1128 uprintf("Cannot execute shared object\n");
1133 * Honour the base load address from the dso if it is
1134 * non-zero for some reason.
1137 if ((sv->sv_flags & SV_ASLR) == 0 ||
1138 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1139 et_dyn_addr = ET_DYN_LOAD_ADDR;
1140 else if ((__elfN(pie_aslr_enabled) &&
1141 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1142 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1143 et_dyn_addr = ET_DYN_ADDR_RAND;
1145 et_dyn_addr = ET_DYN_LOAD_ADDR;
1150 * Avoid a possible deadlock if the current address space is destroyed
1151 * and that address space maps the locked vnode. In the common case,
1152 * the locked vnode's v_usecount is decremented but remains greater
1153 * than zero. Consequently, the vnode lock is not needed by vrele().
1154 * However, in cases where the vnode lock is external, such as nullfs,
1155 * v_usecount may become zero.
1157 * The VV_TEXT flag prevents modifications to the executable while
1158 * the vnode is unlocked.
1160 VOP_UNLOCK(imgp->vp, 0);
1163 * Decide whether to enable randomization of user mappings.
1164 * First, reset user preferences for the setid binaries.
1165 * Then, account for the support of the randomization by the
1166 * ABI, by user preferences, and make special treatment for
1169 if (imgp->credential_setid) {
1170 PROC_LOCK(imgp->proc);
1171 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1172 PROC_UNLOCK(imgp->proc);
1174 if ((sv->sv_flags & SV_ASLR) == 0 ||
1175 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1176 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1177 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1178 ("et_dyn_addr == RAND and !ASLR"));
1179 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1180 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1181 et_dyn_addr == ET_DYN_ADDR_RAND) {
1182 imgp->map_flags |= MAP_ASLR;
1184 * If user does not care about sbrk, utilize the bss
1185 * grow region for mappings as well. We can select
1186 * the base for the image anywere and still not suffer
1187 * from the fragmentation.
1189 if (!__elfN(aslr_honor_sbrk) ||
1190 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1191 imgp->map_flags |= MAP_ASLR_IGNSTART;
1194 error = exec_new_vmspace(imgp, sv);
1195 vmspace = imgp->proc->p_vmspace;
1196 map = &vmspace->vm_map;
1198 imgp->proc->p_sysent = sv;
1200 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1201 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1202 KASSERT((map->flags & MAP_ASLR) != 0,
1203 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1204 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1205 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1206 /* reserve half of the address space to interpreter */
1207 maxv / 2, 1UL << flsl(maxalign));
1210 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1214 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1218 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1222 entry = (u_long)hdr->e_entry + et_dyn_addr;
1225 * We load the dynamic linker where a userland call
1226 * to mmap(0, ...) would put it. The rationale behind this
1227 * calculation is that it leaves room for the heap to grow to
1228 * its maximum allowed size.
1230 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1232 if ((map->flags & MAP_ASLR) != 0) {
1233 maxv1 = maxv / 2 + addr / 2;
1234 MPASS(maxv1 >= addr); /* No overflow */
1235 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1236 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1238 map->anon_loc = addr;
1241 imgp->entry_addr = entry;
1243 if (interp != NULL) {
1244 VOP_UNLOCK(imgp->vp, 0);
1245 if ((map->flags & MAP_ASLR) != 0) {
1246 /* Assume that interpeter fits into 1/4 of AS */
1247 maxv1 = maxv / 2 + addr / 2;
1248 MPASS(maxv1 >= addr); /* No overflow */
1249 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1252 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1254 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1261 * Construct auxargs table (used by the fixup routine)
1263 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1264 elf_auxargs->execfd = -1;
1265 elf_auxargs->phdr = proghdr + et_dyn_addr;
1266 elf_auxargs->phent = hdr->e_phentsize;
1267 elf_auxargs->phnum = hdr->e_phnum;
1268 elf_auxargs->pagesz = PAGE_SIZE;
1269 elf_auxargs->base = addr;
1270 elf_auxargs->flags = 0;
1271 elf_auxargs->entry = entry;
1272 elf_auxargs->hdr_eflags = hdr->e_flags;
1274 imgp->auxargs = elf_auxargs;
1275 imgp->interpreted = 0;
1276 imgp->reloc_base = addr;
1277 imgp->proc->p_osrel = osrel;
1278 imgp->proc->p_fctl0 = fctl0;
1279 imgp->proc->p_elf_machine = hdr->e_machine;
1280 imgp->proc->p_elf_flags = hdr->e_flags;
1284 free(interp, M_TEMP);
1288 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1291 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1293 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1294 Elf_Auxinfo *argarray, *pos;
1295 Elf_Addr *base, *auxbase;
1298 base = (Elf_Addr *)*stack_base;
1299 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1300 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1303 if (args->execfd != -1)
1304 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1305 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1306 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1307 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1308 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1309 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1310 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1311 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1312 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1313 if (imgp->execpathp != 0)
1314 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1315 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1316 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1317 if (imgp->canary != 0) {
1318 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1319 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1321 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1322 if (imgp->pagesizes != 0) {
1323 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1324 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1326 if (imgp->sysent->sv_timekeep_base != 0) {
1327 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1328 imgp->sysent->sv_timekeep_base);
1330 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1331 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1332 imgp->sysent->sv_stackprot);
1333 if (imgp->sysent->sv_hwcap != NULL)
1334 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1335 if (imgp->sysent->sv_hwcap2 != NULL)
1336 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1337 AUXARGS_ENTRY(pos, AT_NULL, 0);
1339 free(imgp->auxargs, M_TEMP);
1340 imgp->auxargs = NULL;
1341 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1343 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1344 free(argarray, M_TEMP);
1349 if (suword(base, imgp->args->argc) == -1)
1351 *stack_base = (register_t *)base;
1356 * Code for generating ELF core dumps.
1359 typedef void (*segment_callback)(vm_map_entry_t, void *);
1361 /* Closure for cb_put_phdr(). */
1362 struct phdr_closure {
1363 Elf_Phdr *phdr; /* Program header to fill in */
1364 Elf_Off offset; /* Offset of segment in core file */
1367 /* Closure for cb_size_segment(). */
1368 struct sseg_closure {
1369 int count; /* Count of writable segments. */
1370 size_t size; /* Total size of all writable segments. */
1373 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1376 int type; /* Note type. */
1377 outfunc_t outfunc; /* Output function. */
1378 void *outarg; /* Argument for the output function. */
1379 size_t outsize; /* Output size. */
1380 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1383 TAILQ_HEAD(note_info_list, note_info);
1385 /* Coredump output parameters. */
1386 struct coredump_params {
1388 struct ucred *active_cred;
1389 struct ucred *file_cred;
1392 struct compressor *comp;
1395 extern int compress_user_cores;
1396 extern int compress_user_cores_level;
1398 static void cb_put_phdr(vm_map_entry_t, void *);
1399 static void cb_size_segment(vm_map_entry_t, void *);
1400 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1402 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1403 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1404 struct note_info_list *, size_t);
1405 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1407 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1408 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1409 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1410 static int sbuf_drain_core_output(void *, const char *, int);
1411 static int sbuf_drain_count(void *arg, const char *data, int len);
1413 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1414 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1415 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1416 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1417 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1418 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1419 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1420 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1421 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1422 static void note_procstat_files(void *, struct sbuf *, size_t *);
1423 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1424 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1425 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1426 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1427 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1430 * Write out a core segment to the compression stream.
1433 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1439 chunk_len = MIN(len, CORE_BUF_SIZE);
1442 * We can get EFAULT error here.
1443 * In that case zero out the current chunk of the segment.
1445 error = copyin(base, buf, chunk_len);
1447 bzero(buf, chunk_len);
1448 error = compressor_write(p->comp, buf, chunk_len);
1458 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1461 return (core_write((struct coredump_params *)arg, base, len, offset,
1466 core_write(struct coredump_params *p, const void *base, size_t len,
1467 off_t offset, enum uio_seg seg)
1470 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1471 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1472 p->active_cred, p->file_cred, NULL, p->td));
1476 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1481 if (p->comp != NULL)
1482 return (compress_chunk(p, base, tmpbuf, len));
1485 * EFAULT is a non-fatal error that we can get, for example,
1486 * if the segment is backed by a file but extends beyond its
1489 error = core_write(p, base, len, offset, UIO_USERSPACE);
1490 if (error == EFAULT) {
1491 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1492 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1493 "for process %s\n", base, len, offset, curproc->p_comm);
1496 * Write a "real" zero byte at the end of the target region
1497 * in the case this is the last segment.
1498 * The intermediate space will be implicitly zero-filled.
1500 error = core_write(p, zero_region, 1, offset + len - 1,
1507 * Drain into a core file.
1510 sbuf_drain_core_output(void *arg, const char *data, int len)
1512 struct coredump_params *p;
1515 p = (struct coredump_params *)arg;
1518 * Some kern_proc out routines that print to this sbuf may
1519 * call us with the process lock held. Draining with the
1520 * non-sleepable lock held is unsafe. The lock is needed for
1521 * those routines when dumping a live process. In our case we
1522 * can safely release the lock before draining and acquire
1525 locked = PROC_LOCKED(p->td->td_proc);
1527 PROC_UNLOCK(p->td->td_proc);
1528 if (p->comp != NULL)
1529 error = compressor_write(p->comp, __DECONST(char *, data), len);
1531 error = core_write(p, __DECONST(void *, data), len, p->offset,
1534 PROC_LOCK(p->td->td_proc);
1542 * Drain into a counter.
1545 sbuf_drain_count(void *arg, const char *data __unused, int len)
1549 sizep = (size_t *)arg;
1555 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1557 struct ucred *cred = td->td_ucred;
1559 struct sseg_closure seginfo;
1560 struct note_info_list notelst;
1561 struct coredump_params params;
1562 struct note_info *ninfo;
1564 size_t hdrsize, notesz, coresize;
1568 TAILQ_INIT(¬elst);
1570 /* Size the program segments. */
1573 each_dumpable_segment(td, cb_size_segment, &seginfo);
1576 * Collect info about the core file header area.
1578 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1579 if (seginfo.count + 1 >= PN_XNUM)
1580 hdrsize += sizeof(Elf_Shdr);
1581 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1582 coresize = round_page(hdrsize + notesz) + seginfo.size;
1584 /* Set up core dump parameters. */
1586 params.active_cred = cred;
1587 params.file_cred = NOCRED;
1594 PROC_LOCK(td->td_proc);
1595 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1596 PROC_UNLOCK(td->td_proc);
1603 if (coresize >= limit) {
1608 /* Create a compression stream if necessary. */
1609 if (compress_user_cores != 0) {
1610 params.comp = compressor_init(core_compressed_write,
1611 compress_user_cores, CORE_BUF_SIZE,
1612 compress_user_cores_level, ¶ms);
1613 if (params.comp == NULL) {
1617 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1621 * Allocate memory for building the header, fill it up,
1622 * and write it out following the notes.
1624 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1625 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1628 /* Write the contents of all of the writable segments. */
1634 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1635 offset = round_page(hdrsize + notesz);
1636 for (i = 0; i < seginfo.count; i++) {
1637 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1638 php->p_filesz, offset, ¶ms, tmpbuf);
1641 offset += php->p_filesz;
1644 if (error == 0 && params.comp != NULL)
1645 error = compressor_flush(params.comp);
1649 "Failed to write core file for process %s (error %d)\n",
1650 curproc->p_comm, error);
1654 free(tmpbuf, M_TEMP);
1655 if (params.comp != NULL)
1656 compressor_fini(params.comp);
1657 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1658 TAILQ_REMOVE(¬elst, ninfo, link);
1659 free(ninfo, M_TEMP);
1668 * A callback for each_dumpable_segment() to write out the segment's
1669 * program header entry.
1672 cb_put_phdr(vm_map_entry_t entry, void *closure)
1674 struct phdr_closure *phc = (struct phdr_closure *)closure;
1675 Elf_Phdr *phdr = phc->phdr;
1677 phc->offset = round_page(phc->offset);
1679 phdr->p_type = PT_LOAD;
1680 phdr->p_offset = phc->offset;
1681 phdr->p_vaddr = entry->start;
1683 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1684 phdr->p_align = PAGE_SIZE;
1685 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1687 phc->offset += phdr->p_filesz;
1692 * A callback for each_dumpable_segment() to gather information about
1693 * the number of segments and their total size.
1696 cb_size_segment(vm_map_entry_t entry, void *closure)
1698 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1701 ssc->size += entry->end - entry->start;
1705 * For each writable segment in the process's memory map, call the given
1706 * function with a pointer to the map entry and some arbitrary
1707 * caller-supplied data.
1710 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1712 struct proc *p = td->td_proc;
1713 vm_map_t map = &p->p_vmspace->vm_map;
1714 vm_map_entry_t entry;
1715 vm_object_t backing_object, object;
1716 boolean_t ignore_entry;
1718 vm_map_lock_read(map);
1719 for (entry = map->header.next; entry != &map->header;
1720 entry = entry->next) {
1722 * Don't dump inaccessible mappings, deal with legacy
1725 * Note that read-only segments related to the elf binary
1726 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1727 * need to arbitrarily ignore such segments.
1729 if (elf_legacy_coredump) {
1730 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1733 if ((entry->protection & VM_PROT_ALL) == 0)
1738 * Dont include memory segment in the coredump if
1739 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1740 * madvise(2). Do not dump submaps (i.e. parts of the
1743 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1746 if ((object = entry->object.vm_object) == NULL)
1749 /* Ignore memory-mapped devices and such things. */
1750 VM_OBJECT_RLOCK(object);
1751 while ((backing_object = object->backing_object) != NULL) {
1752 VM_OBJECT_RLOCK(backing_object);
1753 VM_OBJECT_RUNLOCK(object);
1754 object = backing_object;
1756 ignore_entry = object->type != OBJT_DEFAULT &&
1757 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1758 object->type != OBJT_PHYS;
1759 VM_OBJECT_RUNLOCK(object);
1763 (*func)(entry, closure);
1765 vm_map_unlock_read(map);
1769 * Write the core file header to the file, including padding up to
1770 * the page boundary.
1773 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1774 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1776 struct note_info *ninfo;
1780 /* Fill in the header. */
1781 bzero(hdr, hdrsize);
1782 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1784 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1785 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1786 sbuf_start_section(sb, NULL);
1787 sbuf_bcat(sb, hdr, hdrsize);
1788 TAILQ_FOREACH(ninfo, notelst, link)
1789 __elfN(putnote)(ninfo, sb);
1790 /* Align up to a page boundary for the program segments. */
1791 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1792 error = sbuf_finish(sb);
1799 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1809 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1812 * To have the debugger select the right thread (LWP) as the initial
1813 * thread, we dump the state of the thread passed to us in td first.
1814 * This is the thread that causes the core dump and thus likely to
1815 * be the right thread one wants to have selected in the debugger.
1818 while (thr != NULL) {
1819 size += register_note(list, NT_PRSTATUS,
1820 __elfN(note_prstatus), thr);
1821 size += register_note(list, NT_FPREGSET,
1822 __elfN(note_fpregset), thr);
1823 size += register_note(list, NT_THRMISC,
1824 __elfN(note_thrmisc), thr);
1825 size += register_note(list, NT_PTLWPINFO,
1826 __elfN(note_ptlwpinfo), thr);
1827 size += register_note(list, -1,
1828 __elfN(note_threadmd), thr);
1830 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1831 TAILQ_NEXT(thr, td_plist);
1833 thr = TAILQ_NEXT(thr, td_plist);
1836 size += register_note(list, NT_PROCSTAT_PROC,
1837 __elfN(note_procstat_proc), p);
1838 size += register_note(list, NT_PROCSTAT_FILES,
1839 note_procstat_files, p);
1840 size += register_note(list, NT_PROCSTAT_VMMAP,
1841 note_procstat_vmmap, p);
1842 size += register_note(list, NT_PROCSTAT_GROUPS,
1843 note_procstat_groups, p);
1844 size += register_note(list, NT_PROCSTAT_UMASK,
1845 note_procstat_umask, p);
1846 size += register_note(list, NT_PROCSTAT_RLIMIT,
1847 note_procstat_rlimit, p);
1848 size += register_note(list, NT_PROCSTAT_OSREL,
1849 note_procstat_osrel, p);
1850 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1851 __elfN(note_procstat_psstrings), p);
1852 size += register_note(list, NT_PROCSTAT_AUXV,
1853 __elfN(note_procstat_auxv), p);
1859 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1865 struct phdr_closure phc;
1867 ehdr = (Elf_Ehdr *)hdr;
1869 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1870 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1871 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1872 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1873 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1874 ehdr->e_ident[EI_DATA] = ELF_DATA;
1875 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1876 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1877 ehdr->e_ident[EI_ABIVERSION] = 0;
1878 ehdr->e_ident[EI_PAD] = 0;
1879 ehdr->e_type = ET_CORE;
1880 ehdr->e_machine = td->td_proc->p_elf_machine;
1881 ehdr->e_version = EV_CURRENT;
1883 ehdr->e_phoff = sizeof(Elf_Ehdr);
1884 ehdr->e_flags = td->td_proc->p_elf_flags;
1885 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1886 ehdr->e_phentsize = sizeof(Elf_Phdr);
1887 ehdr->e_shentsize = sizeof(Elf_Shdr);
1888 ehdr->e_shstrndx = SHN_UNDEF;
1889 if (numsegs + 1 < PN_XNUM) {
1890 ehdr->e_phnum = numsegs + 1;
1893 ehdr->e_phnum = PN_XNUM;
1896 ehdr->e_shoff = ehdr->e_phoff +
1897 (numsegs + 1) * ehdr->e_phentsize;
1898 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1899 ("e_shoff: %zu, hdrsize - shdr: %zu",
1900 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1902 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1903 memset(shdr, 0, sizeof(*shdr));
1905 * A special first section is used to hold large segment and
1906 * section counts. This was proposed by Sun Microsystems in
1907 * Solaris and has been adopted by Linux; the standard ELF
1908 * tools are already familiar with the technique.
1910 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1911 * (or 12-7 depending on the version of the document) for more
1914 shdr->sh_type = SHT_NULL;
1915 shdr->sh_size = ehdr->e_shnum;
1916 shdr->sh_link = ehdr->e_shstrndx;
1917 shdr->sh_info = numsegs + 1;
1921 * Fill in the program header entries.
1923 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1925 /* The note segement. */
1926 phdr->p_type = PT_NOTE;
1927 phdr->p_offset = hdrsize;
1930 phdr->p_filesz = notesz;
1932 phdr->p_flags = PF_R;
1933 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1936 /* All the writable segments from the program. */
1938 phc.offset = round_page(hdrsize + notesz);
1939 each_dumpable_segment(td, cb_put_phdr, &phc);
1943 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1945 struct note_info *ninfo;
1946 size_t size, notesize;
1949 out(arg, NULL, &size);
1950 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1952 ninfo->outfunc = out;
1953 ninfo->outarg = arg;
1954 ninfo->outsize = size;
1955 TAILQ_INSERT_TAIL(list, ninfo, link);
1960 notesize = sizeof(Elf_Note) + /* note header */
1961 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1963 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1969 append_note_data(const void *src, void *dst, size_t len)
1973 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1975 bcopy(src, dst, len);
1976 bzero((char *)dst + len, padded_len - len);
1978 return (padded_len);
1982 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1990 note = (Elf_Note *)buf;
1991 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1992 note->n_descsz = size;
1993 note->n_type = type;
1994 buf += sizeof(*note);
1995 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1996 sizeof(FREEBSD_ABI_VENDOR));
1997 append_note_data(src, buf, size);
2002 notesize = sizeof(Elf_Note) + /* note header */
2003 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2005 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2011 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2014 ssize_t old_len, sect_len;
2015 size_t new_len, descsz, i;
2017 if (ninfo->type == -1) {
2018 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2022 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2023 note.n_descsz = ninfo->outsize;
2024 note.n_type = ninfo->type;
2026 sbuf_bcat(sb, ¬e, sizeof(note));
2027 sbuf_start_section(sb, &old_len);
2028 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2029 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2030 if (note.n_descsz == 0)
2032 sbuf_start_section(sb, &old_len);
2033 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2034 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2038 new_len = (size_t)sect_len;
2039 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2040 if (new_len < descsz) {
2042 * It is expected that individual note emitters will correctly
2043 * predict their expected output size and fill up to that size
2044 * themselves, padding in a format-specific way if needed.
2045 * However, in case they don't, just do it here with zeros.
2047 for (i = 0; i < descsz - new_len; i++)
2049 } else if (new_len > descsz) {
2051 * We can't always truncate sb -- we may have drained some
2054 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2055 "read it (%zu > %zu). Since it is longer than "
2056 "expected, this coredump's notes are corrupt. THIS "
2057 "IS A BUG in the note_procstat routine for type %u.\n",
2058 __func__, (unsigned)note.n_type, new_len, descsz,
2059 (unsigned)note.n_type));
2064 * Miscellaneous note out functions.
2067 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2068 #include <compat/freebsd32/freebsd32.h>
2069 #include <compat/freebsd32/freebsd32_signal.h>
2071 typedef struct prstatus32 elf_prstatus_t;
2072 typedef struct prpsinfo32 elf_prpsinfo_t;
2073 typedef struct fpreg32 elf_prfpregset_t;
2074 typedef struct fpreg32 elf_fpregset_t;
2075 typedef struct reg32 elf_gregset_t;
2076 typedef struct thrmisc32 elf_thrmisc_t;
2077 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2078 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2079 typedef uint32_t elf_ps_strings_t;
2081 typedef prstatus_t elf_prstatus_t;
2082 typedef prpsinfo_t elf_prpsinfo_t;
2083 typedef prfpregset_t elf_prfpregset_t;
2084 typedef prfpregset_t elf_fpregset_t;
2085 typedef gregset_t elf_gregset_t;
2086 typedef thrmisc_t elf_thrmisc_t;
2087 #define ELF_KERN_PROC_MASK 0
2088 typedef struct kinfo_proc elf_kinfo_proc_t;
2089 typedef vm_offset_t elf_ps_strings_t;
2093 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2099 elf_prpsinfo_t *psinfo;
2102 p = (struct proc *)arg;
2104 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2105 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2106 psinfo->pr_version = PRPSINFO_VERSION;
2107 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2108 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2110 if (p->p_args != NULL) {
2111 len = sizeof(psinfo->pr_psargs) - 1;
2112 if (len > p->p_args->ar_length)
2113 len = p->p_args->ar_length;
2114 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2120 sbuf_new(&sbarg, psinfo->pr_psargs,
2121 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2122 error = proc_getargv(curthread, p, &sbarg);
2124 if (sbuf_finish(&sbarg) == 0)
2125 len = sbuf_len(&sbarg) - 1;
2127 len = sizeof(psinfo->pr_psargs) - 1;
2128 sbuf_delete(&sbarg);
2130 if (error || len == 0)
2131 strlcpy(psinfo->pr_psargs, p->p_comm,
2132 sizeof(psinfo->pr_psargs));
2134 KASSERT(len < sizeof(psinfo->pr_psargs),
2135 ("len is too long: %zu vs %zu", len,
2136 sizeof(psinfo->pr_psargs)));
2137 cp = psinfo->pr_psargs;
2140 cp = memchr(cp, '\0', end - cp);
2146 psinfo->pr_pid = p->p_pid;
2147 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2148 free(psinfo, M_TEMP);
2150 *sizep = sizeof(*psinfo);
2154 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2157 elf_prstatus_t *status;
2159 td = (struct thread *)arg;
2161 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2162 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2163 status->pr_version = PRSTATUS_VERSION;
2164 status->pr_statussz = sizeof(elf_prstatus_t);
2165 status->pr_gregsetsz = sizeof(elf_gregset_t);
2166 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2167 status->pr_osreldate = osreldate;
2168 status->pr_cursig = td->td_proc->p_sig;
2169 status->pr_pid = td->td_tid;
2170 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2171 fill_regs32(td, &status->pr_reg);
2173 fill_regs(td, &status->pr_reg);
2175 sbuf_bcat(sb, status, sizeof(*status));
2176 free(status, M_TEMP);
2178 *sizep = sizeof(*status);
2182 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2185 elf_prfpregset_t *fpregset;
2187 td = (struct thread *)arg;
2189 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2190 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2191 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2192 fill_fpregs32(td, fpregset);
2194 fill_fpregs(td, fpregset);
2196 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2197 free(fpregset, M_TEMP);
2199 *sizep = sizeof(*fpregset);
2203 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2206 elf_thrmisc_t thrmisc;
2208 td = (struct thread *)arg;
2210 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2211 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2212 strcpy(thrmisc.pr_tname, td->td_name);
2213 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2215 *sizep = sizeof(thrmisc);
2219 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2224 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2225 struct ptrace_lwpinfo32 pl;
2227 struct ptrace_lwpinfo pl;
2230 td = (struct thread *)arg;
2231 size = sizeof(structsize) + sizeof(pl);
2233 KASSERT(*sizep == size, ("invalid size"));
2234 structsize = sizeof(pl);
2235 sbuf_bcat(sb, &structsize, sizeof(structsize));
2236 bzero(&pl, sizeof(pl));
2237 pl.pl_lwpid = td->td_tid;
2238 pl.pl_event = PL_EVENT_NONE;
2239 pl.pl_sigmask = td->td_sigmask;
2240 pl.pl_siglist = td->td_siglist;
2241 if (td->td_si.si_signo != 0) {
2242 pl.pl_event = PL_EVENT_SIGNAL;
2243 pl.pl_flags |= PL_FLAG_SI;
2244 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2245 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2247 pl.pl_siginfo = td->td_si;
2250 strcpy(pl.pl_tdname, td->td_name);
2251 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2252 sbuf_bcat(sb, &pl, sizeof(pl));
2258 * Allow for MD specific notes, as well as any MD
2259 * specific preparations for writing MI notes.
2262 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2268 td = (struct thread *)arg;
2270 if (size != 0 && sb != NULL)
2271 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2275 __elfN(dump_thread)(td, buf, &size);
2276 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2277 if (size != 0 && sb != NULL)
2278 sbuf_bcat(sb, buf, size);
2283 #ifdef KINFO_PROC_SIZE
2284 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2288 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2294 p = (struct proc *)arg;
2295 size = sizeof(structsize) + p->p_numthreads *
2296 sizeof(elf_kinfo_proc_t);
2299 KASSERT(*sizep == size, ("invalid size"));
2300 structsize = sizeof(elf_kinfo_proc_t);
2301 sbuf_bcat(sb, &structsize, sizeof(structsize));
2303 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2308 #ifdef KINFO_FILE_SIZE
2309 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2313 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2316 size_t size, sect_sz, i;
2317 ssize_t start_len, sect_len;
2318 int structsize, filedesc_flags;
2320 if (coredump_pack_fileinfo)
2321 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2325 p = (struct proc *)arg;
2326 structsize = sizeof(struct kinfo_file);
2329 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2330 sbuf_set_drain(sb, sbuf_drain_count, &size);
2331 sbuf_bcat(sb, &structsize, sizeof(structsize));
2333 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2338 sbuf_start_section(sb, &start_len);
2340 sbuf_bcat(sb, &structsize, sizeof(structsize));
2342 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2345 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2350 KASSERT(sect_sz <= *sizep,
2351 ("kern_proc_filedesc_out did not respect maxlen; "
2352 "requested %zu, got %zu", *sizep - sizeof(structsize),
2353 sect_sz - sizeof(structsize)));
2355 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2360 #ifdef KINFO_VMENTRY_SIZE
2361 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2365 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2369 int structsize, vmmap_flags;
2371 if (coredump_pack_vmmapinfo)
2372 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2376 p = (struct proc *)arg;
2377 structsize = sizeof(struct kinfo_vmentry);
2380 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2381 sbuf_set_drain(sb, sbuf_drain_count, &size);
2382 sbuf_bcat(sb, &structsize, sizeof(structsize));
2384 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2389 sbuf_bcat(sb, &structsize, sizeof(structsize));
2391 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2397 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2403 p = (struct proc *)arg;
2404 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2406 KASSERT(*sizep == size, ("invalid size"));
2407 structsize = sizeof(gid_t);
2408 sbuf_bcat(sb, &structsize, sizeof(structsize));
2409 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2416 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2422 p = (struct proc *)arg;
2423 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2425 KASSERT(*sizep == size, ("invalid size"));
2426 structsize = sizeof(p->p_fd->fd_cmask);
2427 sbuf_bcat(sb, &structsize, sizeof(structsize));
2428 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2434 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2437 struct rlimit rlim[RLIM_NLIMITS];
2441 p = (struct proc *)arg;
2442 size = sizeof(structsize) + sizeof(rlim);
2444 KASSERT(*sizep == size, ("invalid size"));
2445 structsize = sizeof(rlim);
2446 sbuf_bcat(sb, &structsize, sizeof(structsize));
2448 for (i = 0; i < RLIM_NLIMITS; i++)
2449 lim_rlimit_proc(p, i, &rlim[i]);
2451 sbuf_bcat(sb, rlim, sizeof(rlim));
2457 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2463 p = (struct proc *)arg;
2464 size = sizeof(structsize) + sizeof(p->p_osrel);
2466 KASSERT(*sizep == size, ("invalid size"));
2467 structsize = sizeof(p->p_osrel);
2468 sbuf_bcat(sb, &structsize, sizeof(structsize));
2469 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2475 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2478 elf_ps_strings_t ps_strings;
2482 p = (struct proc *)arg;
2483 size = sizeof(structsize) + sizeof(ps_strings);
2485 KASSERT(*sizep == size, ("invalid size"));
2486 structsize = sizeof(ps_strings);
2487 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2488 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2490 ps_strings = p->p_sysent->sv_psstrings;
2492 sbuf_bcat(sb, &structsize, sizeof(structsize));
2493 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2499 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2505 p = (struct proc *)arg;
2508 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2509 sbuf_set_drain(sb, sbuf_drain_count, &size);
2510 sbuf_bcat(sb, &structsize, sizeof(structsize));
2512 proc_getauxv(curthread, p, sb);
2518 structsize = sizeof(Elf_Auxinfo);
2519 sbuf_bcat(sb, &structsize, sizeof(structsize));
2521 proc_getauxv(curthread, p, sb);
2527 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2528 const char *note_vendor, const Elf_Phdr *pnote,
2529 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2531 const Elf_Note *note, *note0, *note_end;
2532 const char *note_name;
2537 /* We need some limit, might as well use PAGE_SIZE. */
2538 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2540 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2541 if (pnote->p_offset > PAGE_SIZE ||
2542 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2543 VOP_UNLOCK(imgp->vp, 0);
2544 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2545 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2546 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2547 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2548 curthread->td_ucred, NOCRED, NULL, curthread);
2550 uprintf("i/o error PT_NOTE\n");
2553 note = note0 = (const Elf_Note *)buf;
2554 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2556 note = note0 = (const Elf_Note *)(imgp->image_header +
2558 note_end = (const Elf_Note *)(imgp->image_header +
2559 pnote->p_offset + pnote->p_filesz);
2562 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2563 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2564 (const char *)note < sizeof(Elf_Note)) {
2567 if (note->n_namesz != checknote->n_namesz ||
2568 note->n_descsz != checknote->n_descsz ||
2569 note->n_type != checknote->n_type)
2571 note_name = (const char *)(note + 1);
2572 if (note_name + checknote->n_namesz >=
2573 (const char *)note_end || strncmp(note_vendor,
2574 note_name, checknote->n_namesz) != 0)
2577 if (cb(note, cb_arg, &res))
2580 note = (const Elf_Note *)((const char *)(note + 1) +
2581 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2582 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2591 struct brandnote_cb_arg {
2592 Elf_Brandnote *brandnote;
2597 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2599 struct brandnote_cb_arg *arg;
2604 * Fetch the osreldate for binary from the ELF OSABI-note if
2607 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2608 arg->brandnote->trans_osrel != NULL ?
2609 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2614 static Elf_Note fctl_note = {
2615 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2616 .n_descsz = sizeof(uint32_t),
2617 .n_type = NT_FREEBSD_FEATURE_CTL,
2620 struct fctl_cb_arg {
2625 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2627 struct fctl_cb_arg *arg;
2628 const Elf32_Word *desc;
2632 p = (uintptr_t)(note + 1);
2633 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2634 desc = (const Elf32_Word *)p;
2635 *arg->fctl0 = desc[0];
2640 * Try to find the appropriate ABI-note section for checknote, fetch
2641 * the osreldate and feature control flags for binary from the ELF
2642 * OSABI-note. Only the first page of the image is searched, the same
2646 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2647 int32_t *osrel, uint32_t *fctl0)
2649 const Elf_Phdr *phdr;
2650 const Elf_Ehdr *hdr;
2651 struct brandnote_cb_arg b_arg;
2652 struct fctl_cb_arg f_arg;
2655 hdr = (const Elf_Ehdr *)imgp->image_header;
2656 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2657 b_arg.brandnote = brandnote;
2658 b_arg.osrel = osrel;
2659 f_arg.fctl0 = fctl0;
2661 for (i = 0; i < hdr->e_phnum; i++) {
2662 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2663 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2665 for (j = 0; j < hdr->e_phnum; j++) {
2666 if (phdr[j].p_type == PT_NOTE &&
2667 __elfN(parse_notes)(imgp, &fctl_note,
2668 FREEBSD_ABI_VENDOR, &phdr[j],
2669 note_fctl_cb, &f_arg))
2680 * Tell kern_execve.c about it, with a little help from the linker.
2682 static struct execsw __elfN(execsw) = {
2683 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2684 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2686 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2689 __elfN(trans_prot)(Elf_Word flags)
2695 prot |= VM_PROT_EXECUTE;
2697 prot |= VM_PROT_WRITE;
2699 prot |= VM_PROT_READ;
2700 #if __ELF_WORD_SIZE == 32
2701 #if defined(__amd64__)
2702 if (i386_read_exec && (flags & PF_R))
2703 prot |= VM_PROT_EXECUTE;
2710 __elfN(untrans_prot)(vm_prot_t prot)
2715 if (prot & VM_PROT_EXECUTE)
2717 if (prot & VM_PROT_READ)
2719 if (prot & VM_PROT_WRITE)