2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108 #define CORE_BUF_SIZE (16 * 1024)
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135 "enable execution from readable segments");
138 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
140 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
142 static int __elfN(aslr_enabled) = 0;
143 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
144 &__elfN(aslr_enabled), 0,
145 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
146 ": enable address map randomization");
148 static int __elfN(pie_aslr_enabled) = 0;
149 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
150 &__elfN(pie_aslr_enabled), 0,
151 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
152 ": enable address map randomization for PIE binaries");
154 static int __elfN(aslr_honor_sbrk) = 1;
155 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
156 &__elfN(aslr_honor_sbrk), 0,
157 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
159 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
161 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
163 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
165 Elf_Brandnote __elfN(freebsd_brandnote) = {
166 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
167 .hdr.n_descsz = sizeof(int32_t),
168 .hdr.n_type = NT_FREEBSD_ABI_TAG,
169 .vendor = FREEBSD_ABI_VENDOR,
170 .flags = BN_TRANSLATE_OSREL,
171 .trans_osrel = __elfN(freebsd_trans_osrel)
175 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
179 p = (uintptr_t)(note + 1);
180 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
181 *osrel = *(const int32_t *)(p);
186 static const char GNU_ABI_VENDOR[] = "GNU";
187 static int GNU_KFREEBSD_ABI_DESC = 3;
189 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
190 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
191 .hdr.n_descsz = 16, /* XXX at least 16 */
193 .vendor = GNU_ABI_VENDOR,
194 .flags = BN_TRANSLATE_OSREL,
195 .trans_osrel = kfreebsd_trans_osrel
199 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
201 const Elf32_Word *desc;
204 p = (uintptr_t)(note + 1);
205 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
207 desc = (const Elf32_Word *)p;
208 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
212 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
213 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
215 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
221 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
225 for (i = 0; i < MAX_BRANDS; i++) {
226 if (elf_brand_list[i] == NULL) {
227 elf_brand_list[i] = entry;
231 if (i == MAX_BRANDS) {
232 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
240 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
244 for (i = 0; i < MAX_BRANDS; i++) {
245 if (elf_brand_list[i] == entry) {
246 elf_brand_list[i] = NULL;
256 __elfN(brand_inuse)(Elf_Brandinfo *entry)
261 sx_slock(&allproc_lock);
262 FOREACH_PROC_IN_SYSTEM(p) {
263 if (p->p_sysent == entry->sysvec) {
268 sx_sunlock(&allproc_lock);
273 static Elf_Brandinfo *
274 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
275 int32_t *osrel, uint32_t *fctl0)
277 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
278 Elf_Brandinfo *bi, *bi_m;
280 int i, interp_name_len;
282 interp_name_len = interp != NULL ? strlen(interp) : 0;
285 * We support four types of branding -- (1) the ELF EI_OSABI field
286 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
287 * branding w/in the ELF header, (3) path of the `interp_path'
288 * field, and (4) the ".note.ABI-tag" ELF section.
291 /* Look for an ".note.ABI-tag" ELF section */
293 for (i = 0; i < MAX_BRANDS; i++) {
294 bi = elf_brand_list[i];
297 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
299 if (hdr->e_machine == bi->machine && (bi->flags &
300 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
301 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
303 /* Give brand a chance to veto check_note's guess */
304 if (ret && bi->header_supported)
305 ret = bi->header_supported(imgp);
307 * If note checker claimed the binary, but the
308 * interpreter path in the image does not
309 * match default one for the brand, try to
310 * search for other brands with the same
311 * interpreter. Either there is better brand
312 * with the right interpreter, or, failing
313 * this, we return first brand which accepted
314 * our note and, optionally, header.
316 if (ret && bi_m == NULL && interp != NULL &&
317 (bi->interp_path == NULL ||
318 (strlen(bi->interp_path) + 1 != interp_name_len ||
319 strncmp(interp, bi->interp_path, interp_name_len)
331 /* If the executable has a brand, search for it in the brand list. */
332 for (i = 0; i < MAX_BRANDS; i++) {
333 bi = elf_brand_list[i];
334 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
335 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
337 if (hdr->e_machine == bi->machine &&
338 (hdr->e_ident[EI_OSABI] == bi->brand ||
339 (bi->compat_3_brand != NULL &&
340 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
341 bi->compat_3_brand) == 0))) {
342 /* Looks good, but give brand a chance to veto */
343 if (bi->header_supported == NULL ||
344 bi->header_supported(imgp)) {
346 * Again, prefer strictly matching
349 if (interp_name_len == 0 &&
350 bi->interp_path == NULL)
352 if (bi->interp_path != NULL &&
353 strlen(bi->interp_path) + 1 ==
354 interp_name_len && strncmp(interp,
355 bi->interp_path, interp_name_len) == 0)
365 /* No known brand, see if the header is recognized by any brand */
366 for (i = 0; i < MAX_BRANDS; i++) {
367 bi = elf_brand_list[i];
368 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
369 bi->header_supported == NULL)
371 if (hdr->e_machine == bi->machine) {
372 ret = bi->header_supported(imgp);
378 /* Lacking a known brand, search for a recognized interpreter. */
379 if (interp != NULL) {
380 for (i = 0; i < MAX_BRANDS; i++) {
381 bi = elf_brand_list[i];
382 if (bi == NULL || (bi->flags &
383 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
386 if (hdr->e_machine == bi->machine &&
387 bi->interp_path != NULL &&
388 /* ELF image p_filesz includes terminating zero */
389 strlen(bi->interp_path) + 1 == interp_name_len &&
390 strncmp(interp, bi->interp_path, interp_name_len)
391 == 0 && (bi->header_supported == NULL ||
392 bi->header_supported(imgp)))
397 /* Lacking a recognized interpreter, try the default brand */
398 for (i = 0; i < MAX_BRANDS; i++) {
399 bi = elf_brand_list[i];
400 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
401 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
403 if (hdr->e_machine == bi->machine &&
404 __elfN(fallback_brand) == bi->brand &&
405 (bi->header_supported == NULL ||
406 bi->header_supported(imgp)))
413 __elfN(check_header)(const Elf_Ehdr *hdr)
419 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
420 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
421 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
422 hdr->e_phentsize != sizeof(Elf_Phdr) ||
423 hdr->e_version != ELF_TARG_VER)
427 * Make sure we have at least one brand for this machine.
430 for (i = 0; i < MAX_BRANDS; i++) {
431 bi = elf_brand_list[i];
432 if (bi != NULL && bi->machine == hdr->e_machine)
442 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
443 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
450 * Create the page if it doesn't exist yet. Ignore errors.
452 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
453 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
456 * Find the page from the underlying object.
458 if (object != NULL) {
459 sf = vm_imgact_map_page(object, offset);
461 return (KERN_FAILURE);
462 off = offset - trunc_page(offset);
463 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
465 vm_imgact_unmap_page(sf);
467 return (KERN_FAILURE);
470 return (KERN_SUCCESS);
474 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
475 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
481 int error, locked, rv;
483 if (start != trunc_page(start)) {
484 rv = __elfN(map_partial)(map, object, offset, start,
485 round_page(start), prot);
486 if (rv != KERN_SUCCESS)
488 offset += round_page(start) - start;
489 start = round_page(start);
491 if (end != round_page(end)) {
492 rv = __elfN(map_partial)(map, object, offset +
493 trunc_page(end) - start, trunc_page(end), end, prot);
494 if (rv != KERN_SUCCESS)
496 end = trunc_page(end);
499 return (KERN_SUCCESS);
500 if ((offset & PAGE_MASK) != 0) {
502 * The mapping is not page aligned. This means that we have
505 rv = vm_map_fixed(map, NULL, 0, start, end - start,
506 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
507 if (rv != KERN_SUCCESS)
510 return (KERN_SUCCESS);
511 for (; start < end; start += sz) {
512 sf = vm_imgact_map_page(object, offset);
514 return (KERN_FAILURE);
515 off = offset - trunc_page(offset);
517 if (sz > PAGE_SIZE - off)
518 sz = PAGE_SIZE - off;
519 error = copyout((caddr_t)sf_buf_kva(sf) + off,
521 vm_imgact_unmap_page(sf);
523 return (KERN_FAILURE);
527 vm_object_reference(object);
528 rv = vm_map_fixed(map, object, offset, start, end - start,
529 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
530 if (rv != KERN_SUCCESS) {
531 locked = VOP_ISLOCKED(imgp->vp);
532 VOP_UNLOCK(imgp->vp, 0);
533 vm_object_deallocate(object);
534 vn_lock(imgp->vp, locked | LK_RETRY);
538 return (KERN_SUCCESS);
542 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
543 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
549 vm_offset_t off, map_addr;
552 vm_ooffset_t file_addr;
555 * It's necessary to fail if the filsz + offset taken from the
556 * header is greater than the actual file pager object's size.
557 * If we were to allow this, then the vm_map_find() below would
558 * walk right off the end of the file object and into the ether.
560 * While I'm here, might as well check for something else that
561 * is invalid: filsz cannot be greater than memsz.
563 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
565 uprintf("elf_load_section: truncated ELF file\n");
569 object = imgp->object;
570 map = &imgp->proc->p_vmspace->vm_map;
571 map_addr = trunc_page((vm_offset_t)vmaddr);
572 file_addr = trunc_page(offset);
575 * We have two choices. We can either clear the data in the last page
576 * of an oversized mapping, or we can start the anon mapping a page
577 * early and copy the initialized data into that first page. We
582 else if (memsz > filsz)
583 map_len = trunc_page(offset + filsz) - file_addr;
585 map_len = round_page(offset + filsz) - file_addr;
588 /* cow flags: don't dump readonly sections in core */
589 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
590 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
592 rv = __elfN(map_insert)(imgp, map,
594 file_addr, /* file offset */
595 map_addr, /* virtual start */
596 map_addr + map_len,/* virtual end */
599 if (rv != KERN_SUCCESS)
602 /* we can stop now if we've covered it all */
609 * We have to get the remaining bit of the file into the first part
610 * of the oversized map segment. This is normally because the .data
611 * segment in the file is extended to provide bss. It's a neat idea
612 * to try and save a page, but it's a pain in the behind to implement.
614 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
616 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
617 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
619 /* This had damn well better be true! */
621 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
622 map_addr + map_len, prot, 0);
623 if (rv != KERN_SUCCESS)
628 sf = vm_imgact_map_page(object, offset + filsz);
632 /* send the page fragment to user space */
633 off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
634 error = copyout((caddr_t)sf_buf_kva(sf) + off,
635 (caddr_t)map_addr, copy_len);
636 vm_imgact_unmap_page(sf);
642 * Remove write access to the page if it was only granted by map_insert
645 if ((prot & VM_PROT_WRITE) == 0)
646 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
647 map_len), prot, FALSE);
653 * Load the file "file" into memory. It may be either a shared object
656 * The "addr" reference parameter is in/out. On entry, it specifies
657 * the address where a shared object should be loaded. If the file is
658 * an executable, this value is ignored. On exit, "addr" specifies
659 * where the file was actually loaded.
661 * The "entry" reference parameter is out only. On exit, it specifies
662 * the entry point for the loaded file.
665 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
671 struct image_params image_params;
673 const Elf_Ehdr *hdr = NULL;
674 const Elf_Phdr *phdr = NULL;
675 struct nameidata *nd;
677 struct image_params *imgp;
680 u_long base_addr = 0;
681 int error, i, numsegs;
683 #ifdef CAPABILITY_MODE
685 * XXXJA: This check can go away once we are sufficiently confident
686 * that the checks in namei() are correct.
688 if (IN_CAPABILITY_MODE(curthread))
692 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
694 attr = &tempdata->attr;
695 imgp = &tempdata->image_params;
698 * Initialize part of the common data
702 imgp->firstpage = NULL;
703 imgp->image_header = NULL;
705 imgp->execlabel = NULL;
707 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
708 if ((error = namei(nd)) != 0) {
712 NDFREE(nd, NDF_ONLY_PNBUF);
713 imgp->vp = nd->ni_vp;
716 * Check permissions, modes, uid, etc on the file, and "open" it.
718 error = exec_check_permissions(imgp);
722 error = exec_map_first_page(imgp);
727 * Also make certain that the interpreter stays the same, so set
728 * its VV_TEXT flag, too.
730 VOP_SET_TEXT(nd->ni_vp);
732 imgp->object = nd->ni_vp->v_object;
734 hdr = (const Elf_Ehdr *)imgp->image_header;
735 if ((error = __elfN(check_header)(hdr)) != 0)
737 if (hdr->e_type == ET_DYN)
739 else if (hdr->e_type == ET_EXEC)
746 /* Only support headers that fit within first page for now */
747 if ((hdr->e_phoff > PAGE_SIZE) ||
748 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
753 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
754 if (!aligned(phdr, Elf_Addr)) {
759 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
760 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
761 /* Loadable segment */
762 prot = __elfN(trans_prot)(phdr[i].p_flags);
763 error = __elfN(load_section)(imgp, phdr[i].p_offset,
764 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
765 phdr[i].p_memsz, phdr[i].p_filesz, prot);
769 * Establish the base address if this is the
773 base_addr = trunc_page(phdr[i].p_vaddr +
779 *entry = (unsigned long)hdr->e_entry + rbase;
783 exec_unmap_first_page(imgp);
788 free(tempdata, M_TEMP);
794 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
799 MPASS(vm_map_min(map) <= minv);
800 MPASS(maxv <= vm_map_max(map));
802 MPASS(minv + align < maxv);
803 arc4rand(&rbase, sizeof(rbase), 0);
804 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
805 res &= ~((u_long)align - 1);
809 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
810 res, minv, maxv, rbase));
812 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
813 res, maxv, minv, rbase));
818 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
819 const Elf_Phdr *phdr, u_long et_dyn_addr)
821 struct vmspace *vmspace;
823 u_long text_size, data_size, total_size, text_addr, data_addr;
824 u_long seg_size, seg_addr;
828 text_size = data_size = total_size = text_addr = data_addr = 0;
830 for (i = 0; i < hdr->e_phnum; i++) {
831 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
834 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
835 seg_size = round_page(phdr[i].p_memsz +
836 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
839 * Make the largest executable segment the official
840 * text segment and all others data.
842 * Note that obreak() assumes that data_addr + data_size == end
843 * of data load area, and the ELF file format expects segments
844 * to be sorted by address. If multiple data segments exist,
845 * the last one will be used.
848 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
849 text_size = seg_size;
850 text_addr = seg_addr;
852 data_size = seg_size;
853 data_addr = seg_addr;
855 total_size += seg_size;
858 if (data_addr == 0 && data_size == 0) {
859 data_addr = text_addr;
860 data_size = text_size;
864 * Check limits. It should be safe to check the
865 * limits after loading the segments since we do
866 * not actually fault in all the segments pages.
868 PROC_LOCK(imgp->proc);
869 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
870 err_str = "Data segment size exceeds process limit";
871 else if (text_size > maxtsiz)
872 err_str = "Text segment size exceeds system limit";
873 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
874 err_str = "Total segment size exceeds process limit";
875 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
876 err_str = "Data segment size exceeds resource limit";
877 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
878 err_str = "Total segment size exceeds resource limit";
879 PROC_UNLOCK(imgp->proc);
880 if (err_str != NULL) {
881 uprintf("%s\n", err_str);
885 vmspace = imgp->proc->p_vmspace;
886 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
887 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
888 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
889 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
895 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
896 char **interpp, bool *free_interpp)
900 int error, interp_name_len;
902 KASSERT(phdr->p_type == PT_INTERP,
903 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
904 KASSERT(VOP_ISLOCKED(imgp->vp),
905 ("%s: vp %p is not locked", __func__, imgp->vp));
909 /* Path to interpreter */
910 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
911 uprintf("Invalid PT_INTERP\n");
915 interp_name_len = phdr->p_filesz;
916 if (phdr->p_offset > PAGE_SIZE ||
917 interp_name_len > PAGE_SIZE - phdr->p_offset) {
918 VOP_UNLOCK(imgp->vp, 0);
919 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
920 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
921 error = vn_rdwr(UIO_READ, imgp->vp, interp,
922 interp_name_len, phdr->p_offset,
923 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
926 free(interp, M_TEMP);
927 uprintf("i/o error PT_INTERP %d\n", error);
930 interp[interp_name_len] = '\0';
933 *free_interpp = true;
937 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
938 if (interp[interp_name_len - 1] != '\0') {
939 uprintf("Invalid PT_INTERP\n");
944 *free_interpp = false;
949 * Impossible et_dyn_addr initial value indicating that the real base
950 * must be calculated later with some randomization applied.
952 #define ET_DYN_ADDR_RAND 1
955 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
959 const Elf_Phdr *phdr;
960 Elf_Auxargs *elf_auxargs;
961 struct vmspace *vmspace;
963 const char *newinterp;
965 Elf_Brandinfo *brand_info;
966 struct sysentvec *sv;
968 u_long addr, baddr, et_dyn_addr, entry, proghdr;
969 u_long maxalign, mapsz, maxv, maxv1;
973 int error, i, n, have_interp;
975 hdr = (const Elf_Ehdr *)imgp->image_header;
978 * Do we have a valid ELF header ?
980 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
981 * if particular brand doesn't support it.
983 if (__elfN(check_header)(hdr) != 0 ||
984 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
988 * From here on down, we return an errno, not -1, as we've
989 * detected an ELF file.
992 if ((hdr->e_phoff > PAGE_SIZE) ||
993 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
994 /* Only support headers in first page for now */
995 uprintf("Program headers not in the first page\n");
998 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
999 if (!aligned(phdr, Elf_Addr)) {
1000 uprintf("Unaligned program headers\n");
1008 entry = proghdr = 0;
1009 newinterp = interp = NULL;
1010 free_interp = false;
1012 maxalign = PAGE_SIZE;
1015 for (i = 0; i < hdr->e_phnum; i++) {
1016 switch (phdr[i].p_type) {
1019 baddr = phdr[i].p_vaddr;
1020 if (phdr[i].p_align > maxalign)
1021 maxalign = phdr[i].p_align;
1022 mapsz += phdr[i].p_memsz;
1026 /* Path to interpreter */
1027 if (interp != NULL) {
1028 uprintf("Multiple PT_INTERP headers\n");
1032 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1038 if (__elfN(nxstack))
1040 __elfN(trans_prot)(phdr[i].p_flags);
1041 imgp->stack_sz = phdr[i].p_memsz;
1046 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1047 if (brand_info == NULL) {
1048 uprintf("ELF binary type \"%u\" not known.\n",
1049 hdr->e_ident[EI_OSABI]);
1053 sv = brand_info->sysvec;
1055 if (hdr->e_type == ET_DYN) {
1056 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1057 uprintf("Cannot execute shared object\n");
1062 * Honour the base load address from the dso if it is
1063 * non-zero for some reason.
1066 if ((sv->sv_flags & SV_ASLR) == 0 ||
1067 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1068 et_dyn_addr = ET_DYN_LOAD_ADDR;
1069 else if ((__elfN(pie_aslr_enabled) &&
1070 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1071 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1072 et_dyn_addr = ET_DYN_ADDR_RAND;
1074 et_dyn_addr = ET_DYN_LOAD_ADDR;
1077 if (interp != NULL && brand_info->interp_newpath != NULL)
1078 newinterp = brand_info->interp_newpath;
1081 * Avoid a possible deadlock if the current address space is destroyed
1082 * and that address space maps the locked vnode. In the common case,
1083 * the locked vnode's v_usecount is decremented but remains greater
1084 * than zero. Consequently, the vnode lock is not needed by vrele().
1085 * However, in cases where the vnode lock is external, such as nullfs,
1086 * v_usecount may become zero.
1088 * The VV_TEXT flag prevents modifications to the executable while
1089 * the vnode is unlocked.
1091 VOP_UNLOCK(imgp->vp, 0);
1094 * Decide whether to enable randomization of user mappings.
1095 * First, reset user preferences for the setid binaries.
1096 * Then, account for the support of the randomization by the
1097 * ABI, by user preferences, and make special treatment for
1100 if (imgp->credential_setid) {
1101 PROC_LOCK(imgp->proc);
1102 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1103 PROC_UNLOCK(imgp->proc);
1105 if ((sv->sv_flags & SV_ASLR) == 0 ||
1106 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1107 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1108 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1109 ("et_dyn_addr == RAND and !ASLR"));
1110 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1111 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1112 et_dyn_addr == ET_DYN_ADDR_RAND) {
1113 imgp->map_flags |= MAP_ASLR;
1115 * If user does not care about sbrk, utilize the bss
1116 * grow region for mappings as well. We can select
1117 * the base for the image anywere and still not suffer
1118 * from the fragmentation.
1120 if (!__elfN(aslr_honor_sbrk) ||
1121 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1122 imgp->map_flags |= MAP_ASLR_IGNSTART;
1125 error = exec_new_vmspace(imgp, sv);
1126 vmspace = imgp->proc->p_vmspace;
1127 map = &vmspace->vm_map;
1129 imgp->proc->p_sysent = sv;
1131 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1132 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1133 KASSERT((map->flags & MAP_ASLR) != 0,
1134 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1135 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1136 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1137 /* reserve half of the address space to interpreter */
1138 maxv / 2, 1UL << flsl(maxalign));
1141 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1145 for (i = 0; i < hdr->e_phnum; i++) {
1146 switch (phdr[i].p_type) {
1147 case PT_LOAD: /* Loadable segment */
1148 if (phdr[i].p_memsz == 0)
1150 prot = __elfN(trans_prot)(phdr[i].p_flags);
1151 error = __elfN(load_section)(imgp, phdr[i].p_offset,
1152 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
1153 phdr[i].p_memsz, phdr[i].p_filesz, prot);
1158 * If this segment contains the program headers,
1159 * remember their virtual address for the AT_PHDR
1160 * aux entry. Static binaries don't usually include
1163 if (phdr[i].p_offset == 0 &&
1164 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1165 <= phdr[i].p_filesz)
1166 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
1169 case PT_PHDR: /* Program header table info */
1170 proghdr = phdr[i].p_vaddr + et_dyn_addr;
1177 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1181 entry = (u_long)hdr->e_entry + et_dyn_addr;
1184 * We load the dynamic linker where a userland call
1185 * to mmap(0, ...) would put it. The rationale behind this
1186 * calculation is that it leaves room for the heap to grow to
1187 * its maximum allowed size.
1189 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1191 if ((map->flags & MAP_ASLR) != 0) {
1192 maxv1 = maxv / 2 + addr / 2;
1193 MPASS(maxv1 >= addr); /* No overflow */
1194 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1195 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1197 map->anon_loc = addr;
1200 imgp->entry_addr = entry;
1202 if (interp != NULL) {
1203 have_interp = FALSE;
1204 VOP_UNLOCK(imgp->vp, 0);
1205 if ((map->flags & MAP_ASLR) != 0) {
1206 /* Assume that interpeter fits into 1/4 of AS */
1207 maxv1 = maxv / 2 + addr / 2;
1208 MPASS(maxv1 >= addr); /* No overflow */
1209 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1212 if (brand_info->emul_path != NULL &&
1213 brand_info->emul_path[0] != '\0') {
1214 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1215 snprintf(path, MAXPATHLEN, "%s%s",
1216 brand_info->emul_path, interp);
1217 error = __elfN(load_file)(imgp->proc, path, &addr,
1223 if (!have_interp && newinterp != NULL &&
1224 (brand_info->interp_path == NULL ||
1225 strcmp(interp, brand_info->interp_path) == 0)) {
1226 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1232 error = __elfN(load_file)(imgp->proc, interp, &addr,
1235 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1237 uprintf("ELF interpreter %s not found, error %d\n",
1245 * Construct auxargs table (used by the fixup routine)
1247 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1248 elf_auxargs->execfd = -1;
1249 elf_auxargs->phdr = proghdr;
1250 elf_auxargs->phent = hdr->e_phentsize;
1251 elf_auxargs->phnum = hdr->e_phnum;
1252 elf_auxargs->pagesz = PAGE_SIZE;
1253 elf_auxargs->base = addr;
1254 elf_auxargs->flags = 0;
1255 elf_auxargs->entry = entry;
1256 elf_auxargs->hdr_eflags = hdr->e_flags;
1258 imgp->auxargs = elf_auxargs;
1259 imgp->interpreted = 0;
1260 imgp->reloc_base = addr;
1261 imgp->proc->p_osrel = osrel;
1262 imgp->proc->p_fctl0 = fctl0;
1263 imgp->proc->p_elf_machine = hdr->e_machine;
1264 imgp->proc->p_elf_flags = hdr->e_flags;
1268 free(interp, M_TEMP);
1272 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1275 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1277 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1278 Elf_Auxinfo *argarray, *pos;
1279 Elf_Addr *base, *auxbase;
1282 base = (Elf_Addr *)*stack_base;
1283 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1284 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1287 if (args->execfd != -1)
1288 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1289 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1290 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1291 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1292 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1293 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1294 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1295 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1296 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1297 if (imgp->execpathp != 0)
1298 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1299 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1300 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1301 if (imgp->canary != 0) {
1302 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1303 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1305 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1306 if (imgp->pagesizes != 0) {
1307 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1308 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1310 if (imgp->sysent->sv_timekeep_base != 0) {
1311 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1312 imgp->sysent->sv_timekeep_base);
1314 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1315 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1316 imgp->sysent->sv_stackprot);
1317 if (imgp->sysent->sv_hwcap != NULL)
1318 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1319 if (imgp->sysent->sv_hwcap2 != NULL)
1320 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1321 AUXARGS_ENTRY(pos, AT_NULL, 0);
1323 free(imgp->auxargs, M_TEMP);
1324 imgp->auxargs = NULL;
1325 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1327 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1328 free(argarray, M_TEMP);
1333 if (suword(base, imgp->args->argc) == -1)
1335 *stack_base = (register_t *)base;
1340 * Code for generating ELF core dumps.
1343 typedef void (*segment_callback)(vm_map_entry_t, void *);
1345 /* Closure for cb_put_phdr(). */
1346 struct phdr_closure {
1347 Elf_Phdr *phdr; /* Program header to fill in */
1348 Elf_Off offset; /* Offset of segment in core file */
1351 /* Closure for cb_size_segment(). */
1352 struct sseg_closure {
1353 int count; /* Count of writable segments. */
1354 size_t size; /* Total size of all writable segments. */
1357 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1360 int type; /* Note type. */
1361 outfunc_t outfunc; /* Output function. */
1362 void *outarg; /* Argument for the output function. */
1363 size_t outsize; /* Output size. */
1364 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1367 TAILQ_HEAD(note_info_list, note_info);
1369 /* Coredump output parameters. */
1370 struct coredump_params {
1372 struct ucred *active_cred;
1373 struct ucred *file_cred;
1376 struct compressor *comp;
1379 extern int compress_user_cores;
1380 extern int compress_user_cores_level;
1382 static void cb_put_phdr(vm_map_entry_t, void *);
1383 static void cb_size_segment(vm_map_entry_t, void *);
1384 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1386 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1387 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1388 struct note_info_list *, size_t);
1389 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1391 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1392 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1393 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1394 static int sbuf_drain_core_output(void *, const char *, int);
1395 static int sbuf_drain_count(void *arg, const char *data, int len);
1397 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1398 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1399 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1400 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1401 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1402 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1403 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1404 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1405 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1406 static void note_procstat_files(void *, struct sbuf *, size_t *);
1407 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1408 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1409 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1410 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1411 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1414 * Write out a core segment to the compression stream.
1417 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1423 chunk_len = MIN(len, CORE_BUF_SIZE);
1426 * We can get EFAULT error here.
1427 * In that case zero out the current chunk of the segment.
1429 error = copyin(base, buf, chunk_len);
1431 bzero(buf, chunk_len);
1432 error = compressor_write(p->comp, buf, chunk_len);
1442 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1445 return (core_write((struct coredump_params *)arg, base, len, offset,
1450 core_write(struct coredump_params *p, const void *base, size_t len,
1451 off_t offset, enum uio_seg seg)
1454 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1455 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1456 p->active_cred, p->file_cred, NULL, p->td));
1460 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1465 if (p->comp != NULL)
1466 return (compress_chunk(p, base, tmpbuf, len));
1469 * EFAULT is a non-fatal error that we can get, for example,
1470 * if the segment is backed by a file but extends beyond its
1473 error = core_write(p, base, len, offset, UIO_USERSPACE);
1474 if (error == EFAULT) {
1475 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1476 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1477 "for process %s\n", base, len, offset, curproc->p_comm);
1480 * Write a "real" zero byte at the end of the target region
1481 * in the case this is the last segment.
1482 * The intermediate space will be implicitly zero-filled.
1484 error = core_write(p, zero_region, 1, offset + len - 1,
1491 * Drain into a core file.
1494 sbuf_drain_core_output(void *arg, const char *data, int len)
1496 struct coredump_params *p;
1499 p = (struct coredump_params *)arg;
1502 * Some kern_proc out routines that print to this sbuf may
1503 * call us with the process lock held. Draining with the
1504 * non-sleepable lock held is unsafe. The lock is needed for
1505 * those routines when dumping a live process. In our case we
1506 * can safely release the lock before draining and acquire
1509 locked = PROC_LOCKED(p->td->td_proc);
1511 PROC_UNLOCK(p->td->td_proc);
1512 if (p->comp != NULL)
1513 error = compressor_write(p->comp, __DECONST(char *, data), len);
1515 error = core_write(p, __DECONST(void *, data), len, p->offset,
1518 PROC_LOCK(p->td->td_proc);
1526 * Drain into a counter.
1529 sbuf_drain_count(void *arg, const char *data __unused, int len)
1533 sizep = (size_t *)arg;
1539 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1541 struct ucred *cred = td->td_ucred;
1543 struct sseg_closure seginfo;
1544 struct note_info_list notelst;
1545 struct coredump_params params;
1546 struct note_info *ninfo;
1548 size_t hdrsize, notesz, coresize;
1552 TAILQ_INIT(¬elst);
1554 /* Size the program segments. */
1557 each_dumpable_segment(td, cb_size_segment, &seginfo);
1560 * Collect info about the core file header area.
1562 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1563 if (seginfo.count + 1 >= PN_XNUM)
1564 hdrsize += sizeof(Elf_Shdr);
1565 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1566 coresize = round_page(hdrsize + notesz) + seginfo.size;
1568 /* Set up core dump parameters. */
1570 params.active_cred = cred;
1571 params.file_cred = NOCRED;
1578 PROC_LOCK(td->td_proc);
1579 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1580 PROC_UNLOCK(td->td_proc);
1587 if (coresize >= limit) {
1592 /* Create a compression stream if necessary. */
1593 if (compress_user_cores != 0) {
1594 params.comp = compressor_init(core_compressed_write,
1595 compress_user_cores, CORE_BUF_SIZE,
1596 compress_user_cores_level, ¶ms);
1597 if (params.comp == NULL) {
1601 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1605 * Allocate memory for building the header, fill it up,
1606 * and write it out following the notes.
1608 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1609 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1612 /* Write the contents of all of the writable segments. */
1618 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1619 offset = round_page(hdrsize + notesz);
1620 for (i = 0; i < seginfo.count; i++) {
1621 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1622 php->p_filesz, offset, ¶ms, tmpbuf);
1625 offset += php->p_filesz;
1628 if (error == 0 && params.comp != NULL)
1629 error = compressor_flush(params.comp);
1633 "Failed to write core file for process %s (error %d)\n",
1634 curproc->p_comm, error);
1638 free(tmpbuf, M_TEMP);
1639 if (params.comp != NULL)
1640 compressor_fini(params.comp);
1641 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1642 TAILQ_REMOVE(¬elst, ninfo, link);
1643 free(ninfo, M_TEMP);
1652 * A callback for each_dumpable_segment() to write out the segment's
1653 * program header entry.
1656 cb_put_phdr(vm_map_entry_t entry, void *closure)
1658 struct phdr_closure *phc = (struct phdr_closure *)closure;
1659 Elf_Phdr *phdr = phc->phdr;
1661 phc->offset = round_page(phc->offset);
1663 phdr->p_type = PT_LOAD;
1664 phdr->p_offset = phc->offset;
1665 phdr->p_vaddr = entry->start;
1667 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1668 phdr->p_align = PAGE_SIZE;
1669 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1671 phc->offset += phdr->p_filesz;
1676 * A callback for each_dumpable_segment() to gather information about
1677 * the number of segments and their total size.
1680 cb_size_segment(vm_map_entry_t entry, void *closure)
1682 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1685 ssc->size += entry->end - entry->start;
1689 * For each writable segment in the process's memory map, call the given
1690 * function with a pointer to the map entry and some arbitrary
1691 * caller-supplied data.
1694 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1696 struct proc *p = td->td_proc;
1697 vm_map_t map = &p->p_vmspace->vm_map;
1698 vm_map_entry_t entry;
1699 vm_object_t backing_object, object;
1700 boolean_t ignore_entry;
1702 vm_map_lock_read(map);
1703 for (entry = map->header.next; entry != &map->header;
1704 entry = entry->next) {
1706 * Don't dump inaccessible mappings, deal with legacy
1709 * Note that read-only segments related to the elf binary
1710 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1711 * need to arbitrarily ignore such segments.
1713 if (elf_legacy_coredump) {
1714 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1717 if ((entry->protection & VM_PROT_ALL) == 0)
1722 * Dont include memory segment in the coredump if
1723 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1724 * madvise(2). Do not dump submaps (i.e. parts of the
1727 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1730 if ((object = entry->object.vm_object) == NULL)
1733 /* Ignore memory-mapped devices and such things. */
1734 VM_OBJECT_RLOCK(object);
1735 while ((backing_object = object->backing_object) != NULL) {
1736 VM_OBJECT_RLOCK(backing_object);
1737 VM_OBJECT_RUNLOCK(object);
1738 object = backing_object;
1740 ignore_entry = object->type != OBJT_DEFAULT &&
1741 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1742 object->type != OBJT_PHYS;
1743 VM_OBJECT_RUNLOCK(object);
1747 (*func)(entry, closure);
1749 vm_map_unlock_read(map);
1753 * Write the core file header to the file, including padding up to
1754 * the page boundary.
1757 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1758 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1760 struct note_info *ninfo;
1764 /* Fill in the header. */
1765 bzero(hdr, hdrsize);
1766 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1768 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1769 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1770 sbuf_start_section(sb, NULL);
1771 sbuf_bcat(sb, hdr, hdrsize);
1772 TAILQ_FOREACH(ninfo, notelst, link)
1773 __elfN(putnote)(ninfo, sb);
1774 /* Align up to a page boundary for the program segments. */
1775 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1776 error = sbuf_finish(sb);
1783 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1793 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1796 * To have the debugger select the right thread (LWP) as the initial
1797 * thread, we dump the state of the thread passed to us in td first.
1798 * This is the thread that causes the core dump and thus likely to
1799 * be the right thread one wants to have selected in the debugger.
1802 while (thr != NULL) {
1803 size += register_note(list, NT_PRSTATUS,
1804 __elfN(note_prstatus), thr);
1805 size += register_note(list, NT_FPREGSET,
1806 __elfN(note_fpregset), thr);
1807 size += register_note(list, NT_THRMISC,
1808 __elfN(note_thrmisc), thr);
1809 size += register_note(list, NT_PTLWPINFO,
1810 __elfN(note_ptlwpinfo), thr);
1811 size += register_note(list, -1,
1812 __elfN(note_threadmd), thr);
1814 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1815 TAILQ_NEXT(thr, td_plist);
1817 thr = TAILQ_NEXT(thr, td_plist);
1820 size += register_note(list, NT_PROCSTAT_PROC,
1821 __elfN(note_procstat_proc), p);
1822 size += register_note(list, NT_PROCSTAT_FILES,
1823 note_procstat_files, p);
1824 size += register_note(list, NT_PROCSTAT_VMMAP,
1825 note_procstat_vmmap, p);
1826 size += register_note(list, NT_PROCSTAT_GROUPS,
1827 note_procstat_groups, p);
1828 size += register_note(list, NT_PROCSTAT_UMASK,
1829 note_procstat_umask, p);
1830 size += register_note(list, NT_PROCSTAT_RLIMIT,
1831 note_procstat_rlimit, p);
1832 size += register_note(list, NT_PROCSTAT_OSREL,
1833 note_procstat_osrel, p);
1834 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1835 __elfN(note_procstat_psstrings), p);
1836 size += register_note(list, NT_PROCSTAT_AUXV,
1837 __elfN(note_procstat_auxv), p);
1843 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1849 struct phdr_closure phc;
1851 ehdr = (Elf_Ehdr *)hdr;
1853 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1854 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1855 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1856 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1857 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1858 ehdr->e_ident[EI_DATA] = ELF_DATA;
1859 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1860 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1861 ehdr->e_ident[EI_ABIVERSION] = 0;
1862 ehdr->e_ident[EI_PAD] = 0;
1863 ehdr->e_type = ET_CORE;
1864 ehdr->e_machine = td->td_proc->p_elf_machine;
1865 ehdr->e_version = EV_CURRENT;
1867 ehdr->e_phoff = sizeof(Elf_Ehdr);
1868 ehdr->e_flags = td->td_proc->p_elf_flags;
1869 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1870 ehdr->e_phentsize = sizeof(Elf_Phdr);
1871 ehdr->e_shentsize = sizeof(Elf_Shdr);
1872 ehdr->e_shstrndx = SHN_UNDEF;
1873 if (numsegs + 1 < PN_XNUM) {
1874 ehdr->e_phnum = numsegs + 1;
1877 ehdr->e_phnum = PN_XNUM;
1880 ehdr->e_shoff = ehdr->e_phoff +
1881 (numsegs + 1) * ehdr->e_phentsize;
1882 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1883 ("e_shoff: %zu, hdrsize - shdr: %zu",
1884 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1886 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1887 memset(shdr, 0, sizeof(*shdr));
1889 * A special first section is used to hold large segment and
1890 * section counts. This was proposed by Sun Microsystems in
1891 * Solaris and has been adopted by Linux; the standard ELF
1892 * tools are already familiar with the technique.
1894 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1895 * (or 12-7 depending on the version of the document) for more
1898 shdr->sh_type = SHT_NULL;
1899 shdr->sh_size = ehdr->e_shnum;
1900 shdr->sh_link = ehdr->e_shstrndx;
1901 shdr->sh_info = numsegs + 1;
1905 * Fill in the program header entries.
1907 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1909 /* The note segement. */
1910 phdr->p_type = PT_NOTE;
1911 phdr->p_offset = hdrsize;
1914 phdr->p_filesz = notesz;
1916 phdr->p_flags = PF_R;
1917 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1920 /* All the writable segments from the program. */
1922 phc.offset = round_page(hdrsize + notesz);
1923 each_dumpable_segment(td, cb_put_phdr, &phc);
1927 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1929 struct note_info *ninfo;
1930 size_t size, notesize;
1933 out(arg, NULL, &size);
1934 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1936 ninfo->outfunc = out;
1937 ninfo->outarg = arg;
1938 ninfo->outsize = size;
1939 TAILQ_INSERT_TAIL(list, ninfo, link);
1944 notesize = sizeof(Elf_Note) + /* note header */
1945 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1947 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1953 append_note_data(const void *src, void *dst, size_t len)
1957 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1959 bcopy(src, dst, len);
1960 bzero((char *)dst + len, padded_len - len);
1962 return (padded_len);
1966 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1974 note = (Elf_Note *)buf;
1975 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1976 note->n_descsz = size;
1977 note->n_type = type;
1978 buf += sizeof(*note);
1979 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1980 sizeof(FREEBSD_ABI_VENDOR));
1981 append_note_data(src, buf, size);
1986 notesize = sizeof(Elf_Note) + /* note header */
1987 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1989 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1995 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1998 ssize_t old_len, sect_len;
1999 size_t new_len, descsz, i;
2001 if (ninfo->type == -1) {
2002 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2006 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2007 note.n_descsz = ninfo->outsize;
2008 note.n_type = ninfo->type;
2010 sbuf_bcat(sb, ¬e, sizeof(note));
2011 sbuf_start_section(sb, &old_len);
2012 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2013 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2014 if (note.n_descsz == 0)
2016 sbuf_start_section(sb, &old_len);
2017 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2018 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2022 new_len = (size_t)sect_len;
2023 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2024 if (new_len < descsz) {
2026 * It is expected that individual note emitters will correctly
2027 * predict their expected output size and fill up to that size
2028 * themselves, padding in a format-specific way if needed.
2029 * However, in case they don't, just do it here with zeros.
2031 for (i = 0; i < descsz - new_len; i++)
2033 } else if (new_len > descsz) {
2035 * We can't always truncate sb -- we may have drained some
2038 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2039 "read it (%zu > %zu). Since it is longer than "
2040 "expected, this coredump's notes are corrupt. THIS "
2041 "IS A BUG in the note_procstat routine for type %u.\n",
2042 __func__, (unsigned)note.n_type, new_len, descsz,
2043 (unsigned)note.n_type));
2048 * Miscellaneous note out functions.
2051 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2052 #include <compat/freebsd32/freebsd32.h>
2053 #include <compat/freebsd32/freebsd32_signal.h>
2055 typedef struct prstatus32 elf_prstatus_t;
2056 typedef struct prpsinfo32 elf_prpsinfo_t;
2057 typedef struct fpreg32 elf_prfpregset_t;
2058 typedef struct fpreg32 elf_fpregset_t;
2059 typedef struct reg32 elf_gregset_t;
2060 typedef struct thrmisc32 elf_thrmisc_t;
2061 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2062 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2063 typedef uint32_t elf_ps_strings_t;
2065 typedef prstatus_t elf_prstatus_t;
2066 typedef prpsinfo_t elf_prpsinfo_t;
2067 typedef prfpregset_t elf_prfpregset_t;
2068 typedef prfpregset_t elf_fpregset_t;
2069 typedef gregset_t elf_gregset_t;
2070 typedef thrmisc_t elf_thrmisc_t;
2071 #define ELF_KERN_PROC_MASK 0
2072 typedef struct kinfo_proc elf_kinfo_proc_t;
2073 typedef vm_offset_t elf_ps_strings_t;
2077 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2083 elf_prpsinfo_t *psinfo;
2086 p = (struct proc *)arg;
2088 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2089 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2090 psinfo->pr_version = PRPSINFO_VERSION;
2091 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2092 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2094 if (p->p_args != NULL) {
2095 len = sizeof(psinfo->pr_psargs) - 1;
2096 if (len > p->p_args->ar_length)
2097 len = p->p_args->ar_length;
2098 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2104 sbuf_new(&sbarg, psinfo->pr_psargs,
2105 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2106 error = proc_getargv(curthread, p, &sbarg);
2108 if (sbuf_finish(&sbarg) == 0)
2109 len = sbuf_len(&sbarg) - 1;
2111 len = sizeof(psinfo->pr_psargs) - 1;
2112 sbuf_delete(&sbarg);
2114 if (error || len == 0)
2115 strlcpy(psinfo->pr_psargs, p->p_comm,
2116 sizeof(psinfo->pr_psargs));
2118 KASSERT(len < sizeof(psinfo->pr_psargs),
2119 ("len is too long: %zu vs %zu", len,
2120 sizeof(psinfo->pr_psargs)));
2121 cp = psinfo->pr_psargs;
2124 cp = memchr(cp, '\0', end - cp);
2130 psinfo->pr_pid = p->p_pid;
2131 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2132 free(psinfo, M_TEMP);
2134 *sizep = sizeof(*psinfo);
2138 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2141 elf_prstatus_t *status;
2143 td = (struct thread *)arg;
2145 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2146 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2147 status->pr_version = PRSTATUS_VERSION;
2148 status->pr_statussz = sizeof(elf_prstatus_t);
2149 status->pr_gregsetsz = sizeof(elf_gregset_t);
2150 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2151 status->pr_osreldate = osreldate;
2152 status->pr_cursig = td->td_proc->p_sig;
2153 status->pr_pid = td->td_tid;
2154 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2155 fill_regs32(td, &status->pr_reg);
2157 fill_regs(td, &status->pr_reg);
2159 sbuf_bcat(sb, status, sizeof(*status));
2160 free(status, M_TEMP);
2162 *sizep = sizeof(*status);
2166 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2169 elf_prfpregset_t *fpregset;
2171 td = (struct thread *)arg;
2173 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2174 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2175 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2176 fill_fpregs32(td, fpregset);
2178 fill_fpregs(td, fpregset);
2180 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2181 free(fpregset, M_TEMP);
2183 *sizep = sizeof(*fpregset);
2187 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2190 elf_thrmisc_t thrmisc;
2192 td = (struct thread *)arg;
2194 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2195 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2196 strcpy(thrmisc.pr_tname, td->td_name);
2197 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2199 *sizep = sizeof(thrmisc);
2203 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2208 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2209 struct ptrace_lwpinfo32 pl;
2211 struct ptrace_lwpinfo pl;
2214 td = (struct thread *)arg;
2215 size = sizeof(structsize) + sizeof(pl);
2217 KASSERT(*sizep == size, ("invalid size"));
2218 structsize = sizeof(pl);
2219 sbuf_bcat(sb, &structsize, sizeof(structsize));
2220 bzero(&pl, sizeof(pl));
2221 pl.pl_lwpid = td->td_tid;
2222 pl.pl_event = PL_EVENT_NONE;
2223 pl.pl_sigmask = td->td_sigmask;
2224 pl.pl_siglist = td->td_siglist;
2225 if (td->td_si.si_signo != 0) {
2226 pl.pl_event = PL_EVENT_SIGNAL;
2227 pl.pl_flags |= PL_FLAG_SI;
2228 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2229 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2231 pl.pl_siginfo = td->td_si;
2234 strcpy(pl.pl_tdname, td->td_name);
2235 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2236 sbuf_bcat(sb, &pl, sizeof(pl));
2242 * Allow for MD specific notes, as well as any MD
2243 * specific preparations for writing MI notes.
2246 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2252 td = (struct thread *)arg;
2254 if (size != 0 && sb != NULL)
2255 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2259 __elfN(dump_thread)(td, buf, &size);
2260 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2261 if (size != 0 && sb != NULL)
2262 sbuf_bcat(sb, buf, size);
2267 #ifdef KINFO_PROC_SIZE
2268 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2272 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2278 p = (struct proc *)arg;
2279 size = sizeof(structsize) + p->p_numthreads *
2280 sizeof(elf_kinfo_proc_t);
2283 KASSERT(*sizep == size, ("invalid size"));
2284 structsize = sizeof(elf_kinfo_proc_t);
2285 sbuf_bcat(sb, &structsize, sizeof(structsize));
2287 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2292 #ifdef KINFO_FILE_SIZE
2293 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2297 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2300 size_t size, sect_sz, i;
2301 ssize_t start_len, sect_len;
2302 int structsize, filedesc_flags;
2304 if (coredump_pack_fileinfo)
2305 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2309 p = (struct proc *)arg;
2310 structsize = sizeof(struct kinfo_file);
2313 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2314 sbuf_set_drain(sb, sbuf_drain_count, &size);
2315 sbuf_bcat(sb, &structsize, sizeof(structsize));
2317 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2322 sbuf_start_section(sb, &start_len);
2324 sbuf_bcat(sb, &structsize, sizeof(structsize));
2326 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2329 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2334 KASSERT(sect_sz <= *sizep,
2335 ("kern_proc_filedesc_out did not respect maxlen; "
2336 "requested %zu, got %zu", *sizep - sizeof(structsize),
2337 sect_sz - sizeof(structsize)));
2339 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2344 #ifdef KINFO_VMENTRY_SIZE
2345 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2349 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2353 int structsize, vmmap_flags;
2355 if (coredump_pack_vmmapinfo)
2356 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2360 p = (struct proc *)arg;
2361 structsize = sizeof(struct kinfo_vmentry);
2364 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2365 sbuf_set_drain(sb, sbuf_drain_count, &size);
2366 sbuf_bcat(sb, &structsize, sizeof(structsize));
2368 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2373 sbuf_bcat(sb, &structsize, sizeof(structsize));
2375 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2381 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2387 p = (struct proc *)arg;
2388 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2390 KASSERT(*sizep == size, ("invalid size"));
2391 structsize = sizeof(gid_t);
2392 sbuf_bcat(sb, &structsize, sizeof(structsize));
2393 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2400 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2406 p = (struct proc *)arg;
2407 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2409 KASSERT(*sizep == size, ("invalid size"));
2410 structsize = sizeof(p->p_fd->fd_cmask);
2411 sbuf_bcat(sb, &structsize, sizeof(structsize));
2412 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2418 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2421 struct rlimit rlim[RLIM_NLIMITS];
2425 p = (struct proc *)arg;
2426 size = sizeof(structsize) + sizeof(rlim);
2428 KASSERT(*sizep == size, ("invalid size"));
2429 structsize = sizeof(rlim);
2430 sbuf_bcat(sb, &structsize, sizeof(structsize));
2432 for (i = 0; i < RLIM_NLIMITS; i++)
2433 lim_rlimit_proc(p, i, &rlim[i]);
2435 sbuf_bcat(sb, rlim, sizeof(rlim));
2441 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2447 p = (struct proc *)arg;
2448 size = sizeof(structsize) + sizeof(p->p_osrel);
2450 KASSERT(*sizep == size, ("invalid size"));
2451 structsize = sizeof(p->p_osrel);
2452 sbuf_bcat(sb, &structsize, sizeof(structsize));
2453 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2459 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2462 elf_ps_strings_t ps_strings;
2466 p = (struct proc *)arg;
2467 size = sizeof(structsize) + sizeof(ps_strings);
2469 KASSERT(*sizep == size, ("invalid size"));
2470 structsize = sizeof(ps_strings);
2471 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2472 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2474 ps_strings = p->p_sysent->sv_psstrings;
2476 sbuf_bcat(sb, &structsize, sizeof(structsize));
2477 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2483 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2489 p = (struct proc *)arg;
2492 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2493 sbuf_set_drain(sb, sbuf_drain_count, &size);
2494 sbuf_bcat(sb, &structsize, sizeof(structsize));
2496 proc_getauxv(curthread, p, sb);
2502 structsize = sizeof(Elf_Auxinfo);
2503 sbuf_bcat(sb, &structsize, sizeof(structsize));
2505 proc_getauxv(curthread, p, sb);
2511 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2512 const char *note_vendor, const Elf_Phdr *pnote,
2513 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2515 const Elf_Note *note, *note0, *note_end;
2516 const char *note_name;
2521 /* We need some limit, might as well use PAGE_SIZE. */
2522 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2524 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2525 if (pnote->p_offset > PAGE_SIZE ||
2526 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2527 VOP_UNLOCK(imgp->vp, 0);
2528 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2529 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2530 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2531 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2532 curthread->td_ucred, NOCRED, NULL, curthread);
2534 uprintf("i/o error PT_NOTE\n");
2537 note = note0 = (const Elf_Note *)buf;
2538 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2540 note = note0 = (const Elf_Note *)(imgp->image_header +
2542 note_end = (const Elf_Note *)(imgp->image_header +
2543 pnote->p_offset + pnote->p_filesz);
2546 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2547 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2548 (const char *)note < sizeof(Elf_Note)) {
2551 if (note->n_namesz != checknote->n_namesz ||
2552 note->n_descsz != checknote->n_descsz ||
2553 note->n_type != checknote->n_type)
2555 note_name = (const char *)(note + 1);
2556 if (note_name + checknote->n_namesz >=
2557 (const char *)note_end || strncmp(note_vendor,
2558 note_name, checknote->n_namesz) != 0)
2561 if (cb(note, cb_arg, &res))
2564 note = (const Elf_Note *)((const char *)(note + 1) +
2565 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2566 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2575 struct brandnote_cb_arg {
2576 Elf_Brandnote *brandnote;
2581 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2583 struct brandnote_cb_arg *arg;
2588 * Fetch the osreldate for binary from the ELF OSABI-note if
2591 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2592 arg->brandnote->trans_osrel != NULL ?
2593 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2598 static Elf_Note fctl_note = {
2599 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2600 .n_descsz = sizeof(uint32_t),
2601 .n_type = NT_FREEBSD_FEATURE_CTL,
2604 struct fctl_cb_arg {
2609 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2611 struct fctl_cb_arg *arg;
2612 const Elf32_Word *desc;
2616 p = (uintptr_t)(note + 1);
2617 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2618 desc = (const Elf32_Word *)p;
2619 *arg->fctl0 = desc[0];
2624 * Try to find the appropriate ABI-note section for checknote, fetch
2625 * the osreldate and feature control flags for binary from the ELF
2626 * OSABI-note. Only the first page of the image is searched, the same
2630 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2631 int32_t *osrel, uint32_t *fctl0)
2633 const Elf_Phdr *phdr;
2634 const Elf_Ehdr *hdr;
2635 struct brandnote_cb_arg b_arg;
2636 struct fctl_cb_arg f_arg;
2639 hdr = (const Elf_Ehdr *)imgp->image_header;
2640 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2641 b_arg.brandnote = brandnote;
2642 b_arg.osrel = osrel;
2643 f_arg.fctl0 = fctl0;
2645 for (i = 0; i < hdr->e_phnum; i++) {
2646 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2647 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2649 for (j = 0; j < hdr->e_phnum; j++) {
2650 if (phdr[j].p_type == PT_NOTE &&
2651 __elfN(parse_notes)(imgp, &fctl_note,
2652 FREEBSD_ABI_VENDOR, &phdr[j],
2653 note_fctl_cb, &f_arg))
2664 * Tell kern_execve.c about it, with a little help from the linker.
2666 static struct execsw __elfN(execsw) = {
2667 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2668 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2670 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2673 __elfN(trans_prot)(Elf_Word flags)
2679 prot |= VM_PROT_EXECUTE;
2681 prot |= VM_PROT_WRITE;
2683 prot |= VM_PROT_READ;
2684 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2685 if (i386_read_exec && (flags & PF_R))
2686 prot |= VM_PROT_EXECUTE;
2692 __elfN(untrans_prot)(vm_prot_t prot)
2697 if (prot & VM_PROT_EXECUTE)
2699 if (prot & VM_PROT_READ)
2701 if (prot & VM_PROT_WRITE)