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, int interp_name_len, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93 u_long *entry, size_t pagesize);
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,
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102 Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
109 #define CORE_BUF_SIZE (16 * 1024)
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118 &elf_legacy_coredump, 0,
119 "include all and only RW pages in core dumps");
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (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
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136 "enable execution from readable segments");
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 #define trunc_page_ps(va, ps) rounddown2(va, ps)
143 #define round_page_ps(va, ps) roundup2(va, ps)
144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
150 .hdr.n_descsz = sizeof(int32_t),
151 .hdr.n_type = NT_FREEBSD_ABI_TAG,
152 .vendor = FREEBSD_ABI_VENDOR,
153 .flags = BN_TRANSLATE_OSREL,
154 .trans_osrel = __elfN(freebsd_trans_osrel)
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 p = (uintptr_t)(note + 1);
163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 *osrel = *(const int32_t *)(p);
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
174 .hdr.n_descsz = 16, /* XXX at least 16 */
176 .vendor = GNU_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = kfreebsd_trans_osrel
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 const Elf32_Word *desc;
187 p = (uintptr_t)(note + 1);
188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 desc = (const Elf32_Word *)p;
191 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 for (i = 0; i < MAX_BRANDS; i++) {
209 if (elf_brand_list[i] == NULL) {
210 elf_brand_list[i] = entry;
214 if (i == MAX_BRANDS) {
215 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 for (i = 0; i < MAX_BRANDS; i++) {
228 if (elf_brand_list[i] == entry) {
229 elf_brand_list[i] = NULL;
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
244 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
246 if (p->p_sysent == entry->sysvec) {
251 sx_sunlock(&allproc_lock);
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258 int interp_name_len, int32_t *osrel, uint32_t *fctl0)
260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 Elf_Brandinfo *bi, *bi_m;
266 * We support four types of branding -- (1) the ELF EI_OSABI field
267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 * branding w/in the ELF header, (3) path of the `interp_path'
269 * field, and (4) the ".note.ABI-tag" ELF section.
272 /* Look for an ".note.ABI-tag" ELF section */
274 for (i = 0; i < MAX_BRANDS; i++) {
275 bi = elf_brand_list[i];
278 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
280 if (hdr->e_machine == bi->machine && (bi->flags &
281 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
282 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
284 /* Give brand a chance to veto check_note's guess */
285 if (ret && bi->header_supported)
286 ret = bi->header_supported(imgp);
288 * If note checker claimed the binary, but the
289 * interpreter path in the image does not
290 * match default one for the brand, try to
291 * search for other brands with the same
292 * interpreter. Either there is better brand
293 * with the right interpreter, or, failing
294 * this, we return first brand which accepted
295 * our note and, optionally, header.
297 if (ret && bi_m == NULL && interp != NULL &&
298 (bi->interp_path == NULL ||
299 (strlen(bi->interp_path) + 1 != interp_name_len ||
300 strncmp(interp, bi->interp_path, interp_name_len)
312 /* If the executable has a brand, search for it in the brand list. */
313 for (i = 0; i < MAX_BRANDS; i++) {
314 bi = elf_brand_list[i];
315 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
316 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
318 if (hdr->e_machine == bi->machine &&
319 (hdr->e_ident[EI_OSABI] == bi->brand ||
320 (bi->compat_3_brand != NULL &&
321 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
322 bi->compat_3_brand) == 0))) {
323 /* Looks good, but give brand a chance to veto */
324 if (bi->header_supported == NULL ||
325 bi->header_supported(imgp)) {
327 * Again, prefer strictly matching
330 if (interp_name_len == 0 &&
331 bi->interp_path == NULL)
333 if (bi->interp_path != NULL &&
334 strlen(bi->interp_path) + 1 ==
335 interp_name_len && strncmp(interp,
336 bi->interp_path, interp_name_len) == 0)
346 /* No known brand, see if the header is recognized by any brand */
347 for (i = 0; i < MAX_BRANDS; i++) {
348 bi = elf_brand_list[i];
349 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
350 bi->header_supported == NULL)
352 if (hdr->e_machine == bi->machine) {
353 ret = bi->header_supported(imgp);
359 /* Lacking a known brand, search for a recognized interpreter. */
360 if (interp != NULL) {
361 for (i = 0; i < MAX_BRANDS; i++) {
362 bi = elf_brand_list[i];
363 if (bi == NULL || (bi->flags &
364 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
367 if (hdr->e_machine == bi->machine &&
368 bi->interp_path != NULL &&
369 /* ELF image p_filesz includes terminating zero */
370 strlen(bi->interp_path) + 1 == interp_name_len &&
371 strncmp(interp, bi->interp_path, interp_name_len)
372 == 0 && (bi->header_supported == NULL ||
373 bi->header_supported(imgp)))
378 /* Lacking a recognized interpreter, try the default brand */
379 for (i = 0; i < MAX_BRANDS; i++) {
380 bi = elf_brand_list[i];
381 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
382 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
384 if (hdr->e_machine == bi->machine &&
385 __elfN(fallback_brand) == bi->brand &&
386 (bi->header_supported == NULL ||
387 bi->header_supported(imgp)))
394 __elfN(check_header)(const Elf_Ehdr *hdr)
400 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
401 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
402 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
403 hdr->e_phentsize != sizeof(Elf_Phdr) ||
404 hdr->e_version != ELF_TARG_VER)
408 * Make sure we have at least one brand for this machine.
411 for (i = 0; i < MAX_BRANDS; i++) {
412 bi = elf_brand_list[i];
413 if (bi != NULL && bi->machine == hdr->e_machine)
423 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
424 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
431 * Create the page if it doesn't exist yet. Ignore errors.
433 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
434 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
437 * Find the page from the underlying object.
439 if (object != NULL) {
440 sf = vm_imgact_map_page(object, offset);
442 return (KERN_FAILURE);
443 off = offset - trunc_page(offset);
444 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
446 vm_imgact_unmap_page(sf);
448 return (KERN_FAILURE);
451 return (KERN_SUCCESS);
455 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
456 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
462 int error, locked, rv;
464 if (start != trunc_page(start)) {
465 rv = __elfN(map_partial)(map, object, offset, start,
466 round_page(start), prot);
467 if (rv != KERN_SUCCESS)
469 offset += round_page(start) - start;
470 start = round_page(start);
472 if (end != round_page(end)) {
473 rv = __elfN(map_partial)(map, object, offset +
474 trunc_page(end) - start, trunc_page(end), end, prot);
475 if (rv != KERN_SUCCESS)
477 end = trunc_page(end);
480 return (KERN_SUCCESS);
481 if ((offset & PAGE_MASK) != 0) {
483 * The mapping is not page aligned. This means that we have
486 rv = vm_map_fixed(map, NULL, 0, start, end - start,
487 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
488 if (rv != KERN_SUCCESS)
491 return (KERN_SUCCESS);
492 for (; start < end; start += sz) {
493 sf = vm_imgact_map_page(object, offset);
495 return (KERN_FAILURE);
496 off = offset - trunc_page(offset);
498 if (sz > PAGE_SIZE - off)
499 sz = PAGE_SIZE - off;
500 error = copyout((caddr_t)sf_buf_kva(sf) + off,
502 vm_imgact_unmap_page(sf);
504 return (KERN_FAILURE);
508 vm_object_reference(object);
509 rv = vm_map_fixed(map, object, offset, start, end - start,
510 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
511 if (rv != KERN_SUCCESS) {
512 locked = VOP_ISLOCKED(imgp->vp);
513 VOP_UNLOCK(imgp->vp, 0);
514 vm_object_deallocate(object);
515 vn_lock(imgp->vp, locked | LK_RETRY);
519 return (KERN_SUCCESS);
523 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
524 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
531 vm_offset_t off, map_addr;
534 vm_ooffset_t file_addr;
537 * It's necessary to fail if the filsz + offset taken from the
538 * header is greater than the actual file pager object's size.
539 * If we were to allow this, then the vm_map_find() below would
540 * walk right off the end of the file object and into the ether.
542 * While I'm here, might as well check for something else that
543 * is invalid: filsz cannot be greater than memsz.
545 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
547 uprintf("elf_load_section: truncated ELF file\n");
551 object = imgp->object;
552 map = &imgp->proc->p_vmspace->vm_map;
553 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
554 file_addr = trunc_page_ps(offset, pagesize);
557 * We have two choices. We can either clear the data in the last page
558 * of an oversized mapping, or we can start the anon mapping a page
559 * early and copy the initialized data into that first page. We
564 else if (memsz > filsz)
565 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
567 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
570 /* cow flags: don't dump readonly sections in core */
571 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
572 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
574 rv = __elfN(map_insert)(imgp, map,
576 file_addr, /* file offset */
577 map_addr, /* virtual start */
578 map_addr + map_len,/* virtual end */
581 if (rv != KERN_SUCCESS)
584 /* we can stop now if we've covered it all */
591 * We have to get the remaining bit of the file into the first part
592 * of the oversized map segment. This is normally because the .data
593 * segment in the file is extended to provide bss. It's a neat idea
594 * to try and save a page, but it's a pain in the behind to implement.
596 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
598 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
599 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
602 /* This had damn well better be true! */
604 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
605 map_addr + map_len, prot, 0);
606 if (rv != KERN_SUCCESS)
611 sf = vm_imgact_map_page(object, offset + filsz);
615 /* send the page fragment to user space */
616 off = trunc_page_ps(offset + filsz, pagesize) -
617 trunc_page(offset + filsz);
618 error = copyout((caddr_t)sf_buf_kva(sf) + off,
619 (caddr_t)map_addr, copy_len);
620 vm_imgact_unmap_page(sf);
626 * Remove write access to the page if it was only granted by map_insert
629 if ((prot & VM_PROT_WRITE) == 0)
630 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
631 map_len), prot, FALSE);
637 * Load the file "file" into memory. It may be either a shared object
640 * The "addr" reference parameter is in/out. On entry, it specifies
641 * the address where a shared object should be loaded. If the file is
642 * an executable, this value is ignored. On exit, "addr" specifies
643 * where the file was actually loaded.
645 * The "entry" reference parameter is out only. On exit, it specifies
646 * the entry point for the loaded file.
649 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
650 u_long *entry, size_t pagesize)
655 struct image_params image_params;
657 const Elf_Ehdr *hdr = NULL;
658 const Elf_Phdr *phdr = NULL;
659 struct nameidata *nd;
661 struct image_params *imgp;
664 u_long base_addr = 0;
665 int error, i, numsegs;
667 #ifdef CAPABILITY_MODE
669 * XXXJA: This check can go away once we are sufficiently confident
670 * that the checks in namei() are correct.
672 if (IN_CAPABILITY_MODE(curthread))
676 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
678 attr = &tempdata->attr;
679 imgp = &tempdata->image_params;
682 * Initialize part of the common data
686 imgp->firstpage = NULL;
687 imgp->image_header = NULL;
689 imgp->execlabel = NULL;
691 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
692 if ((error = namei(nd)) != 0) {
696 NDFREE(nd, NDF_ONLY_PNBUF);
697 imgp->vp = nd->ni_vp;
700 * Check permissions, modes, uid, etc on the file, and "open" it.
702 error = exec_check_permissions(imgp);
706 error = exec_map_first_page(imgp);
711 * Also make certain that the interpreter stays the same, so set
712 * its VV_TEXT flag, too.
714 VOP_SET_TEXT(nd->ni_vp);
716 imgp->object = nd->ni_vp->v_object;
718 hdr = (const Elf_Ehdr *)imgp->image_header;
719 if ((error = __elfN(check_header)(hdr)) != 0)
721 if (hdr->e_type == ET_DYN)
723 else if (hdr->e_type == ET_EXEC)
730 /* Only support headers that fit within first page for now */
731 if ((hdr->e_phoff > PAGE_SIZE) ||
732 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
737 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
738 if (!aligned(phdr, Elf_Addr)) {
743 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
744 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
745 /* Loadable segment */
746 prot = __elfN(trans_prot)(phdr[i].p_flags);
747 error = __elfN(load_section)(imgp, phdr[i].p_offset,
748 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
749 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
753 * Establish the base address if this is the
757 base_addr = trunc_page(phdr[i].p_vaddr +
763 *entry = (unsigned long)hdr->e_entry + rbase;
767 exec_unmap_first_page(imgp);
772 free(tempdata, M_TEMP);
778 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
782 const Elf_Phdr *phdr;
783 Elf_Auxargs *elf_auxargs;
784 struct vmspace *vmspace;
785 const char *err_str, *newinterp;
786 char *interp, *interp_buf, *path;
787 Elf_Brandinfo *brand_info;
788 struct sysentvec *sv;
790 u_long text_size, data_size, total_size, text_addr, data_addr;
791 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
794 int error, i, n, interp_name_len, have_interp;
796 hdr = (const Elf_Ehdr *)imgp->image_header;
799 * Do we have a valid ELF header ?
801 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
802 * if particular brand doesn't support it.
804 if (__elfN(check_header)(hdr) != 0 ||
805 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
809 * From here on down, we return an errno, not -1, as we've
810 * detected an ELF file.
813 if ((hdr->e_phoff > PAGE_SIZE) ||
814 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
815 /* Only support headers in first page for now */
816 uprintf("Program headers not in the first page\n");
819 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
820 if (!aligned(phdr, Elf_Addr)) {
821 uprintf("Unaligned program headers\n");
829 text_size = data_size = total_size = text_addr = data_addr = 0;
832 err_str = newinterp = NULL;
833 interp = interp_buf = NULL;
836 for (i = 0; i < hdr->e_phnum; i++) {
837 switch (phdr[i].p_type) {
840 baddr = phdr[i].p_vaddr;
844 /* Path to interpreter */
845 if (phdr[i].p_filesz < 2 ||
846 phdr[i].p_filesz > MAXPATHLEN) {
847 uprintf("Invalid PT_INTERP\n");
851 if (interp != NULL) {
852 uprintf("Multiple PT_INTERP headers\n");
856 interp_name_len = phdr[i].p_filesz;
857 if (phdr[i].p_offset > PAGE_SIZE ||
858 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
859 VOP_UNLOCK(imgp->vp, 0);
860 interp_buf = malloc(interp_name_len + 1, M_TEMP,
862 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
863 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
864 interp_name_len, phdr[i].p_offset,
865 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
868 uprintf("i/o error PT_INTERP %d\n",
872 interp_buf[interp_name_len] = '\0';
875 interp = __DECONST(char *, imgp->image_header) +
877 if (interp[interp_name_len - 1] != '\0') {
878 uprintf("Invalid PT_INTERP\n");
887 __elfN(trans_prot)(phdr[i].p_flags);
888 imgp->stack_sz = phdr[i].p_memsz;
893 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
895 if (brand_info == NULL) {
896 uprintf("ELF binary type \"%u\" not known.\n",
897 hdr->e_ident[EI_OSABI]);
902 if (hdr->e_type == ET_DYN) {
903 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
904 uprintf("Cannot execute shared object\n");
909 * Honour the base load address from the dso if it is
910 * non-zero for some reason.
913 et_dyn_addr = ET_DYN_LOAD_ADDR;
915 sv = brand_info->sysvec;
916 if (interp != NULL && brand_info->interp_newpath != NULL)
917 newinterp = brand_info->interp_newpath;
920 * Avoid a possible deadlock if the current address space is destroyed
921 * and that address space maps the locked vnode. In the common case,
922 * the locked vnode's v_usecount is decremented but remains greater
923 * than zero. Consequently, the vnode lock is not needed by vrele().
924 * However, in cases where the vnode lock is external, such as nullfs,
925 * v_usecount may become zero.
927 * The VV_TEXT flag prevents modifications to the executable while
928 * the vnode is unlocked.
930 VOP_UNLOCK(imgp->vp, 0);
932 error = exec_new_vmspace(imgp, sv);
933 imgp->proc->p_sysent = sv;
935 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
939 for (i = 0; i < hdr->e_phnum; i++) {
940 switch (phdr[i].p_type) {
941 case PT_LOAD: /* Loadable segment */
942 if (phdr[i].p_memsz == 0)
944 prot = __elfN(trans_prot)(phdr[i].p_flags);
945 error = __elfN(load_section)(imgp, phdr[i].p_offset,
946 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
947 phdr[i].p_memsz, phdr[i].p_filesz, prot,
953 * If this segment contains the program headers,
954 * remember their virtual address for the AT_PHDR
955 * aux entry. Static binaries don't usually include
958 if (phdr[i].p_offset == 0 &&
959 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
961 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
964 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
965 seg_size = round_page(phdr[i].p_memsz +
966 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
969 * Make the largest executable segment the official
970 * text segment and all others data.
972 * Note that obreak() assumes that data_addr +
973 * data_size == end of data load area, and the ELF
974 * file format expects segments to be sorted by
975 * address. If multiple data segments exist, the
976 * last one will be used.
979 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
980 text_size = seg_size;
981 text_addr = seg_addr;
983 data_size = seg_size;
984 data_addr = seg_addr;
986 total_size += seg_size;
988 case PT_PHDR: /* Program header table info */
989 proghdr = phdr[i].p_vaddr + et_dyn_addr;
996 if (data_addr == 0 && data_size == 0) {
997 data_addr = text_addr;
998 data_size = text_size;
1001 entry = (u_long)hdr->e_entry + et_dyn_addr;
1004 * Check limits. It should be safe to check the
1005 * limits after loading the segments since we do
1006 * not actually fault in all the segments pages.
1008 PROC_LOCK(imgp->proc);
1009 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1010 err_str = "Data segment size exceeds process limit";
1011 else if (text_size > maxtsiz)
1012 err_str = "Text segment size exceeds system limit";
1013 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1014 err_str = "Total segment size exceeds process limit";
1015 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1016 err_str = "Data segment size exceeds resource limit";
1017 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1018 err_str = "Total segment size exceeds resource limit";
1019 if (err_str != NULL) {
1020 PROC_UNLOCK(imgp->proc);
1021 uprintf("%s\n", err_str);
1026 vmspace = imgp->proc->p_vmspace;
1027 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1028 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1029 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1030 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1033 * We load the dynamic linker where a userland call
1034 * to mmap(0, ...) would put it. The rationale behind this
1035 * calculation is that it leaves room for the heap to grow to
1036 * its maximum allowed size.
1038 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1040 PROC_UNLOCK(imgp->proc);
1042 imgp->entry_addr = entry;
1044 if (interp != NULL) {
1045 have_interp = FALSE;
1046 VOP_UNLOCK(imgp->vp, 0);
1047 if (brand_info->emul_path != NULL &&
1048 brand_info->emul_path[0] != '\0') {
1049 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1050 snprintf(path, MAXPATHLEN, "%s%s",
1051 brand_info->emul_path, interp);
1052 error = __elfN(load_file)(imgp->proc, path, &addr,
1053 &imgp->entry_addr, sv->sv_pagesize);
1058 if (!have_interp && newinterp != NULL &&
1059 (brand_info->interp_path == NULL ||
1060 strcmp(interp, brand_info->interp_path) == 0)) {
1061 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1062 &imgp->entry_addr, sv->sv_pagesize);
1067 error = __elfN(load_file)(imgp->proc, interp, &addr,
1068 &imgp->entry_addr, sv->sv_pagesize);
1070 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1072 uprintf("ELF interpreter %s not found, error %d\n",
1080 * Construct auxargs table (used by the fixup routine)
1082 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1083 elf_auxargs->execfd = -1;
1084 elf_auxargs->phdr = proghdr;
1085 elf_auxargs->phent = hdr->e_phentsize;
1086 elf_auxargs->phnum = hdr->e_phnum;
1087 elf_auxargs->pagesz = PAGE_SIZE;
1088 elf_auxargs->base = addr;
1089 elf_auxargs->flags = 0;
1090 elf_auxargs->entry = entry;
1091 elf_auxargs->hdr_eflags = hdr->e_flags;
1093 imgp->auxargs = elf_auxargs;
1094 imgp->interpreted = 0;
1095 imgp->reloc_base = addr;
1096 imgp->proc->p_osrel = osrel;
1097 imgp->proc->p_fctl0 = fctl0;
1098 imgp->proc->p_elf_machine = hdr->e_machine;
1099 imgp->proc->p_elf_flags = hdr->e_flags;
1102 free(interp_buf, M_TEMP);
1106 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1109 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1111 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1112 Elf_Auxinfo *argarray, *pos;
1113 Elf_Addr *base, *auxbase;
1116 base = (Elf_Addr *)*stack_base;
1117 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1118 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1121 if (args->execfd != -1)
1122 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1123 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1124 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1125 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1126 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1127 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1128 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1129 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1130 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1131 if (imgp->execpathp != 0)
1132 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1133 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1134 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1135 if (imgp->canary != 0) {
1136 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1137 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1139 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1140 if (imgp->pagesizes != 0) {
1141 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1142 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1144 if (imgp->sysent->sv_timekeep_base != 0) {
1145 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1146 imgp->sysent->sv_timekeep_base);
1148 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1149 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1150 imgp->sysent->sv_stackprot);
1151 if (imgp->sysent->sv_hwcap != NULL)
1152 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1153 if (imgp->sysent->sv_hwcap2 != NULL)
1154 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1155 AUXARGS_ENTRY(pos, AT_NULL, 0);
1157 free(imgp->auxargs, M_TEMP);
1158 imgp->auxargs = NULL;
1159 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1161 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1162 free(argarray, M_TEMP);
1167 if (suword(base, imgp->args->argc) == -1)
1169 *stack_base = (register_t *)base;
1174 * Code for generating ELF core dumps.
1177 typedef void (*segment_callback)(vm_map_entry_t, void *);
1179 /* Closure for cb_put_phdr(). */
1180 struct phdr_closure {
1181 Elf_Phdr *phdr; /* Program header to fill in */
1182 Elf_Off offset; /* Offset of segment in core file */
1185 /* Closure for cb_size_segment(). */
1186 struct sseg_closure {
1187 int count; /* Count of writable segments. */
1188 size_t size; /* Total size of all writable segments. */
1191 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1194 int type; /* Note type. */
1195 outfunc_t outfunc; /* Output function. */
1196 void *outarg; /* Argument for the output function. */
1197 size_t outsize; /* Output size. */
1198 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1201 TAILQ_HEAD(note_info_list, note_info);
1203 /* Coredump output parameters. */
1204 struct coredump_params {
1206 struct ucred *active_cred;
1207 struct ucred *file_cred;
1210 struct compressor *comp;
1213 extern int compress_user_cores;
1214 extern int compress_user_cores_level;
1216 static void cb_put_phdr(vm_map_entry_t, void *);
1217 static void cb_size_segment(vm_map_entry_t, void *);
1218 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1220 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1221 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1222 struct note_info_list *, size_t);
1223 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1225 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1226 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1227 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1228 static int sbuf_drain_core_output(void *, const char *, int);
1229 static int sbuf_drain_count(void *arg, const char *data, int len);
1231 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1232 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1233 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1234 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1235 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1236 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1237 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1238 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1239 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1240 static void note_procstat_files(void *, struct sbuf *, size_t *);
1241 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1242 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1243 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1244 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1245 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1248 * Write out a core segment to the compression stream.
1251 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1257 chunk_len = MIN(len, CORE_BUF_SIZE);
1260 * We can get EFAULT error here.
1261 * In that case zero out the current chunk of the segment.
1263 error = copyin(base, buf, chunk_len);
1265 bzero(buf, chunk_len);
1266 error = compressor_write(p->comp, buf, chunk_len);
1276 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1279 return (core_write((struct coredump_params *)arg, base, len, offset,
1284 core_write(struct coredump_params *p, const void *base, size_t len,
1285 off_t offset, enum uio_seg seg)
1288 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1289 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1290 p->active_cred, p->file_cred, NULL, p->td));
1294 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1299 if (p->comp != NULL)
1300 return (compress_chunk(p, base, tmpbuf, len));
1303 * EFAULT is a non-fatal error that we can get, for example,
1304 * if the segment is backed by a file but extends beyond its
1307 error = core_write(p, base, len, offset, UIO_USERSPACE);
1308 if (error == EFAULT) {
1309 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1310 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1311 "for process %s\n", base, len, offset, curproc->p_comm);
1314 * Write a "real" zero byte at the end of the target region
1315 * in the case this is the last segment.
1316 * The intermediate space will be implicitly zero-filled.
1318 error = core_write(p, zero_region, 1, offset + len - 1,
1325 * Drain into a core file.
1328 sbuf_drain_core_output(void *arg, const char *data, int len)
1330 struct coredump_params *p;
1333 p = (struct coredump_params *)arg;
1336 * Some kern_proc out routines that print to this sbuf may
1337 * call us with the process lock held. Draining with the
1338 * non-sleepable lock held is unsafe. The lock is needed for
1339 * those routines when dumping a live process. In our case we
1340 * can safely release the lock before draining and acquire
1343 locked = PROC_LOCKED(p->td->td_proc);
1345 PROC_UNLOCK(p->td->td_proc);
1346 if (p->comp != NULL)
1347 error = compressor_write(p->comp, __DECONST(char *, data), len);
1349 error = core_write(p, __DECONST(void *, data), len, p->offset,
1352 PROC_LOCK(p->td->td_proc);
1360 * Drain into a counter.
1363 sbuf_drain_count(void *arg, const char *data __unused, int len)
1367 sizep = (size_t *)arg;
1373 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1375 struct ucred *cred = td->td_ucred;
1377 struct sseg_closure seginfo;
1378 struct note_info_list notelst;
1379 struct coredump_params params;
1380 struct note_info *ninfo;
1382 size_t hdrsize, notesz, coresize;
1386 TAILQ_INIT(¬elst);
1388 /* Size the program segments. */
1391 each_dumpable_segment(td, cb_size_segment, &seginfo);
1394 * Collect info about the core file header area.
1396 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1397 if (seginfo.count + 1 >= PN_XNUM)
1398 hdrsize += sizeof(Elf_Shdr);
1399 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1400 coresize = round_page(hdrsize + notesz) + seginfo.size;
1402 /* Set up core dump parameters. */
1404 params.active_cred = cred;
1405 params.file_cred = NOCRED;
1412 PROC_LOCK(td->td_proc);
1413 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1414 PROC_UNLOCK(td->td_proc);
1421 if (coresize >= limit) {
1426 /* Create a compression stream if necessary. */
1427 if (compress_user_cores != 0) {
1428 params.comp = compressor_init(core_compressed_write,
1429 compress_user_cores, CORE_BUF_SIZE,
1430 compress_user_cores_level, ¶ms);
1431 if (params.comp == NULL) {
1435 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1439 * Allocate memory for building the header, fill it up,
1440 * and write it out following the notes.
1442 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1443 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1446 /* Write the contents of all of the writable segments. */
1452 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1453 offset = round_page(hdrsize + notesz);
1454 for (i = 0; i < seginfo.count; i++) {
1455 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1456 php->p_filesz, offset, ¶ms, tmpbuf);
1459 offset += php->p_filesz;
1462 if (error == 0 && params.comp != NULL)
1463 error = compressor_flush(params.comp);
1467 "Failed to write core file for process %s (error %d)\n",
1468 curproc->p_comm, error);
1472 free(tmpbuf, M_TEMP);
1473 if (params.comp != NULL)
1474 compressor_fini(params.comp);
1475 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1476 TAILQ_REMOVE(¬elst, ninfo, link);
1477 free(ninfo, M_TEMP);
1486 * A callback for each_dumpable_segment() to write out the segment's
1487 * program header entry.
1490 cb_put_phdr(vm_map_entry_t entry, void *closure)
1492 struct phdr_closure *phc = (struct phdr_closure *)closure;
1493 Elf_Phdr *phdr = phc->phdr;
1495 phc->offset = round_page(phc->offset);
1497 phdr->p_type = PT_LOAD;
1498 phdr->p_offset = phc->offset;
1499 phdr->p_vaddr = entry->start;
1501 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1502 phdr->p_align = PAGE_SIZE;
1503 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1505 phc->offset += phdr->p_filesz;
1510 * A callback for each_dumpable_segment() to gather information about
1511 * the number of segments and their total size.
1514 cb_size_segment(vm_map_entry_t entry, void *closure)
1516 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1519 ssc->size += entry->end - entry->start;
1523 * For each writable segment in the process's memory map, call the given
1524 * function with a pointer to the map entry and some arbitrary
1525 * caller-supplied data.
1528 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1530 struct proc *p = td->td_proc;
1531 vm_map_t map = &p->p_vmspace->vm_map;
1532 vm_map_entry_t entry;
1533 vm_object_t backing_object, object;
1534 boolean_t ignore_entry;
1536 vm_map_lock_read(map);
1537 for (entry = map->header.next; entry != &map->header;
1538 entry = entry->next) {
1540 * Don't dump inaccessible mappings, deal with legacy
1543 * Note that read-only segments related to the elf binary
1544 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1545 * need to arbitrarily ignore such segments.
1547 if (elf_legacy_coredump) {
1548 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1551 if ((entry->protection & VM_PROT_ALL) == 0)
1556 * Dont include memory segment in the coredump if
1557 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1558 * madvise(2). Do not dump submaps (i.e. parts of the
1561 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1564 if ((object = entry->object.vm_object) == NULL)
1567 /* Ignore memory-mapped devices and such things. */
1568 VM_OBJECT_RLOCK(object);
1569 while ((backing_object = object->backing_object) != NULL) {
1570 VM_OBJECT_RLOCK(backing_object);
1571 VM_OBJECT_RUNLOCK(object);
1572 object = backing_object;
1574 ignore_entry = object->type != OBJT_DEFAULT &&
1575 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1576 object->type != OBJT_PHYS;
1577 VM_OBJECT_RUNLOCK(object);
1581 (*func)(entry, closure);
1583 vm_map_unlock_read(map);
1587 * Write the core file header to the file, including padding up to
1588 * the page boundary.
1591 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1592 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1594 struct note_info *ninfo;
1598 /* Fill in the header. */
1599 bzero(hdr, hdrsize);
1600 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1602 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1603 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1604 sbuf_start_section(sb, NULL);
1605 sbuf_bcat(sb, hdr, hdrsize);
1606 TAILQ_FOREACH(ninfo, notelst, link)
1607 __elfN(putnote)(ninfo, sb);
1608 /* Align up to a page boundary for the program segments. */
1609 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1610 error = sbuf_finish(sb);
1617 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1627 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1630 * To have the debugger select the right thread (LWP) as the initial
1631 * thread, we dump the state of the thread passed to us in td first.
1632 * This is the thread that causes the core dump and thus likely to
1633 * be the right thread one wants to have selected in the debugger.
1636 while (thr != NULL) {
1637 size += register_note(list, NT_PRSTATUS,
1638 __elfN(note_prstatus), thr);
1639 size += register_note(list, NT_FPREGSET,
1640 __elfN(note_fpregset), thr);
1641 size += register_note(list, NT_THRMISC,
1642 __elfN(note_thrmisc), thr);
1643 size += register_note(list, NT_PTLWPINFO,
1644 __elfN(note_ptlwpinfo), thr);
1645 size += register_note(list, -1,
1646 __elfN(note_threadmd), thr);
1648 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1649 TAILQ_NEXT(thr, td_plist);
1651 thr = TAILQ_NEXT(thr, td_plist);
1654 size += register_note(list, NT_PROCSTAT_PROC,
1655 __elfN(note_procstat_proc), p);
1656 size += register_note(list, NT_PROCSTAT_FILES,
1657 note_procstat_files, p);
1658 size += register_note(list, NT_PROCSTAT_VMMAP,
1659 note_procstat_vmmap, p);
1660 size += register_note(list, NT_PROCSTAT_GROUPS,
1661 note_procstat_groups, p);
1662 size += register_note(list, NT_PROCSTAT_UMASK,
1663 note_procstat_umask, p);
1664 size += register_note(list, NT_PROCSTAT_RLIMIT,
1665 note_procstat_rlimit, p);
1666 size += register_note(list, NT_PROCSTAT_OSREL,
1667 note_procstat_osrel, p);
1668 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1669 __elfN(note_procstat_psstrings), p);
1670 size += register_note(list, NT_PROCSTAT_AUXV,
1671 __elfN(note_procstat_auxv), p);
1677 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1683 struct phdr_closure phc;
1685 ehdr = (Elf_Ehdr *)hdr;
1687 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1688 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1689 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1690 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1691 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1692 ehdr->e_ident[EI_DATA] = ELF_DATA;
1693 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1694 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1695 ehdr->e_ident[EI_ABIVERSION] = 0;
1696 ehdr->e_ident[EI_PAD] = 0;
1697 ehdr->e_type = ET_CORE;
1698 ehdr->e_machine = td->td_proc->p_elf_machine;
1699 ehdr->e_version = EV_CURRENT;
1701 ehdr->e_phoff = sizeof(Elf_Ehdr);
1702 ehdr->e_flags = td->td_proc->p_elf_flags;
1703 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1704 ehdr->e_phentsize = sizeof(Elf_Phdr);
1705 ehdr->e_shentsize = sizeof(Elf_Shdr);
1706 ehdr->e_shstrndx = SHN_UNDEF;
1707 if (numsegs + 1 < PN_XNUM) {
1708 ehdr->e_phnum = numsegs + 1;
1711 ehdr->e_phnum = PN_XNUM;
1714 ehdr->e_shoff = ehdr->e_phoff +
1715 (numsegs + 1) * ehdr->e_phentsize;
1716 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1717 ("e_shoff: %zu, hdrsize - shdr: %zu",
1718 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1720 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1721 memset(shdr, 0, sizeof(*shdr));
1723 * A special first section is used to hold large segment and
1724 * section counts. This was proposed by Sun Microsystems in
1725 * Solaris and has been adopted by Linux; the standard ELF
1726 * tools are already familiar with the technique.
1728 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1729 * (or 12-7 depending on the version of the document) for more
1732 shdr->sh_type = SHT_NULL;
1733 shdr->sh_size = ehdr->e_shnum;
1734 shdr->sh_link = ehdr->e_shstrndx;
1735 shdr->sh_info = numsegs + 1;
1739 * Fill in the program header entries.
1741 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1743 /* The note segement. */
1744 phdr->p_type = PT_NOTE;
1745 phdr->p_offset = hdrsize;
1748 phdr->p_filesz = notesz;
1750 phdr->p_flags = PF_R;
1751 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1754 /* All the writable segments from the program. */
1756 phc.offset = round_page(hdrsize + notesz);
1757 each_dumpable_segment(td, cb_put_phdr, &phc);
1761 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1763 struct note_info *ninfo;
1764 size_t size, notesize;
1767 out(arg, NULL, &size);
1768 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1770 ninfo->outfunc = out;
1771 ninfo->outarg = arg;
1772 ninfo->outsize = size;
1773 TAILQ_INSERT_TAIL(list, ninfo, link);
1778 notesize = sizeof(Elf_Note) + /* note header */
1779 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1781 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1787 append_note_data(const void *src, void *dst, size_t len)
1791 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1793 bcopy(src, dst, len);
1794 bzero((char *)dst + len, padded_len - len);
1796 return (padded_len);
1800 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1808 note = (Elf_Note *)buf;
1809 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1810 note->n_descsz = size;
1811 note->n_type = type;
1812 buf += sizeof(*note);
1813 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1814 sizeof(FREEBSD_ABI_VENDOR));
1815 append_note_data(src, buf, size);
1820 notesize = sizeof(Elf_Note) + /* note header */
1821 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1823 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1829 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1832 ssize_t old_len, sect_len;
1833 size_t new_len, descsz, i;
1835 if (ninfo->type == -1) {
1836 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1840 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1841 note.n_descsz = ninfo->outsize;
1842 note.n_type = ninfo->type;
1844 sbuf_bcat(sb, ¬e, sizeof(note));
1845 sbuf_start_section(sb, &old_len);
1846 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1847 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1848 if (note.n_descsz == 0)
1850 sbuf_start_section(sb, &old_len);
1851 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1852 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1856 new_len = (size_t)sect_len;
1857 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1858 if (new_len < descsz) {
1860 * It is expected that individual note emitters will correctly
1861 * predict their expected output size and fill up to that size
1862 * themselves, padding in a format-specific way if needed.
1863 * However, in case they don't, just do it here with zeros.
1865 for (i = 0; i < descsz - new_len; i++)
1867 } else if (new_len > descsz) {
1869 * We can't always truncate sb -- we may have drained some
1872 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1873 "read it (%zu > %zu). Since it is longer than "
1874 "expected, this coredump's notes are corrupt. THIS "
1875 "IS A BUG in the note_procstat routine for type %u.\n",
1876 __func__, (unsigned)note.n_type, new_len, descsz,
1877 (unsigned)note.n_type));
1882 * Miscellaneous note out functions.
1885 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1886 #include <compat/freebsd32/freebsd32.h>
1887 #include <compat/freebsd32/freebsd32_signal.h>
1889 typedef struct prstatus32 elf_prstatus_t;
1890 typedef struct prpsinfo32 elf_prpsinfo_t;
1891 typedef struct fpreg32 elf_prfpregset_t;
1892 typedef struct fpreg32 elf_fpregset_t;
1893 typedef struct reg32 elf_gregset_t;
1894 typedef struct thrmisc32 elf_thrmisc_t;
1895 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1896 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1897 typedef uint32_t elf_ps_strings_t;
1899 typedef prstatus_t elf_prstatus_t;
1900 typedef prpsinfo_t elf_prpsinfo_t;
1901 typedef prfpregset_t elf_prfpregset_t;
1902 typedef prfpregset_t elf_fpregset_t;
1903 typedef gregset_t elf_gregset_t;
1904 typedef thrmisc_t elf_thrmisc_t;
1905 #define ELF_KERN_PROC_MASK 0
1906 typedef struct kinfo_proc elf_kinfo_proc_t;
1907 typedef vm_offset_t elf_ps_strings_t;
1911 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1917 elf_prpsinfo_t *psinfo;
1920 p = (struct proc *)arg;
1922 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1923 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1924 psinfo->pr_version = PRPSINFO_VERSION;
1925 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1926 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1928 if (p->p_args != NULL) {
1929 len = sizeof(psinfo->pr_psargs) - 1;
1930 if (len > p->p_args->ar_length)
1931 len = p->p_args->ar_length;
1932 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1938 sbuf_new(&sbarg, psinfo->pr_psargs,
1939 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1940 error = proc_getargv(curthread, p, &sbarg);
1942 if (sbuf_finish(&sbarg) == 0)
1943 len = sbuf_len(&sbarg) - 1;
1945 len = sizeof(psinfo->pr_psargs) - 1;
1946 sbuf_delete(&sbarg);
1948 if (error || len == 0)
1949 strlcpy(psinfo->pr_psargs, p->p_comm,
1950 sizeof(psinfo->pr_psargs));
1952 KASSERT(len < sizeof(psinfo->pr_psargs),
1953 ("len is too long: %zu vs %zu", len,
1954 sizeof(psinfo->pr_psargs)));
1955 cp = psinfo->pr_psargs;
1958 cp = memchr(cp, '\0', end - cp);
1964 psinfo->pr_pid = p->p_pid;
1965 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1966 free(psinfo, M_TEMP);
1968 *sizep = sizeof(*psinfo);
1972 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1975 elf_prstatus_t *status;
1977 td = (struct thread *)arg;
1979 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1980 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1981 status->pr_version = PRSTATUS_VERSION;
1982 status->pr_statussz = sizeof(elf_prstatus_t);
1983 status->pr_gregsetsz = sizeof(elf_gregset_t);
1984 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1985 status->pr_osreldate = osreldate;
1986 status->pr_cursig = td->td_proc->p_sig;
1987 status->pr_pid = td->td_tid;
1988 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1989 fill_regs32(td, &status->pr_reg);
1991 fill_regs(td, &status->pr_reg);
1993 sbuf_bcat(sb, status, sizeof(*status));
1994 free(status, M_TEMP);
1996 *sizep = sizeof(*status);
2000 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2003 elf_prfpregset_t *fpregset;
2005 td = (struct thread *)arg;
2007 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2008 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2009 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2010 fill_fpregs32(td, fpregset);
2012 fill_fpregs(td, fpregset);
2014 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2015 free(fpregset, M_TEMP);
2017 *sizep = sizeof(*fpregset);
2021 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2024 elf_thrmisc_t thrmisc;
2026 td = (struct thread *)arg;
2028 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2029 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2030 strcpy(thrmisc.pr_tname, td->td_name);
2031 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2033 *sizep = sizeof(thrmisc);
2037 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2042 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2043 struct ptrace_lwpinfo32 pl;
2045 struct ptrace_lwpinfo pl;
2048 td = (struct thread *)arg;
2049 size = sizeof(structsize) + sizeof(pl);
2051 KASSERT(*sizep == size, ("invalid size"));
2052 structsize = sizeof(pl);
2053 sbuf_bcat(sb, &structsize, sizeof(structsize));
2054 bzero(&pl, sizeof(pl));
2055 pl.pl_lwpid = td->td_tid;
2056 pl.pl_event = PL_EVENT_NONE;
2057 pl.pl_sigmask = td->td_sigmask;
2058 pl.pl_siglist = td->td_siglist;
2059 if (td->td_si.si_signo != 0) {
2060 pl.pl_event = PL_EVENT_SIGNAL;
2061 pl.pl_flags |= PL_FLAG_SI;
2062 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2063 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2065 pl.pl_siginfo = td->td_si;
2068 strcpy(pl.pl_tdname, td->td_name);
2069 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2070 sbuf_bcat(sb, &pl, sizeof(pl));
2076 * Allow for MD specific notes, as well as any MD
2077 * specific preparations for writing MI notes.
2080 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2086 td = (struct thread *)arg;
2088 if (size != 0 && sb != NULL)
2089 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2093 __elfN(dump_thread)(td, buf, &size);
2094 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2095 if (size != 0 && sb != NULL)
2096 sbuf_bcat(sb, buf, size);
2101 #ifdef KINFO_PROC_SIZE
2102 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2106 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2112 p = (struct proc *)arg;
2113 size = sizeof(structsize) + p->p_numthreads *
2114 sizeof(elf_kinfo_proc_t);
2117 KASSERT(*sizep == size, ("invalid size"));
2118 structsize = sizeof(elf_kinfo_proc_t);
2119 sbuf_bcat(sb, &structsize, sizeof(structsize));
2121 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2126 #ifdef KINFO_FILE_SIZE
2127 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2131 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2134 size_t size, sect_sz, i;
2135 ssize_t start_len, sect_len;
2136 int structsize, filedesc_flags;
2138 if (coredump_pack_fileinfo)
2139 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2143 p = (struct proc *)arg;
2144 structsize = sizeof(struct kinfo_file);
2147 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2148 sbuf_set_drain(sb, sbuf_drain_count, &size);
2149 sbuf_bcat(sb, &structsize, sizeof(structsize));
2151 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2156 sbuf_start_section(sb, &start_len);
2158 sbuf_bcat(sb, &structsize, sizeof(structsize));
2160 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2163 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2168 KASSERT(sect_sz <= *sizep,
2169 ("kern_proc_filedesc_out did not respect maxlen; "
2170 "requested %zu, got %zu", *sizep - sizeof(structsize),
2171 sect_sz - sizeof(structsize)));
2173 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2178 #ifdef KINFO_VMENTRY_SIZE
2179 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2183 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2187 int structsize, vmmap_flags;
2189 if (coredump_pack_vmmapinfo)
2190 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2194 p = (struct proc *)arg;
2195 structsize = sizeof(struct kinfo_vmentry);
2198 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2199 sbuf_set_drain(sb, sbuf_drain_count, &size);
2200 sbuf_bcat(sb, &structsize, sizeof(structsize));
2202 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2207 sbuf_bcat(sb, &structsize, sizeof(structsize));
2209 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2215 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2221 p = (struct proc *)arg;
2222 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2224 KASSERT(*sizep == size, ("invalid size"));
2225 structsize = sizeof(gid_t);
2226 sbuf_bcat(sb, &structsize, sizeof(structsize));
2227 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2234 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2240 p = (struct proc *)arg;
2241 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2243 KASSERT(*sizep == size, ("invalid size"));
2244 structsize = sizeof(p->p_fd->fd_cmask);
2245 sbuf_bcat(sb, &structsize, sizeof(structsize));
2246 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2252 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2255 struct rlimit rlim[RLIM_NLIMITS];
2259 p = (struct proc *)arg;
2260 size = sizeof(structsize) + sizeof(rlim);
2262 KASSERT(*sizep == size, ("invalid size"));
2263 structsize = sizeof(rlim);
2264 sbuf_bcat(sb, &structsize, sizeof(structsize));
2266 for (i = 0; i < RLIM_NLIMITS; i++)
2267 lim_rlimit_proc(p, i, &rlim[i]);
2269 sbuf_bcat(sb, rlim, sizeof(rlim));
2275 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2281 p = (struct proc *)arg;
2282 size = sizeof(structsize) + sizeof(p->p_osrel);
2284 KASSERT(*sizep == size, ("invalid size"));
2285 structsize = sizeof(p->p_osrel);
2286 sbuf_bcat(sb, &structsize, sizeof(structsize));
2287 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2293 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2296 elf_ps_strings_t ps_strings;
2300 p = (struct proc *)arg;
2301 size = sizeof(structsize) + sizeof(ps_strings);
2303 KASSERT(*sizep == size, ("invalid size"));
2304 structsize = sizeof(ps_strings);
2305 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2306 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2308 ps_strings = p->p_sysent->sv_psstrings;
2310 sbuf_bcat(sb, &structsize, sizeof(structsize));
2311 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2317 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2323 p = (struct proc *)arg;
2326 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2327 sbuf_set_drain(sb, sbuf_drain_count, &size);
2328 sbuf_bcat(sb, &structsize, sizeof(structsize));
2330 proc_getauxv(curthread, p, sb);
2336 structsize = sizeof(Elf_Auxinfo);
2337 sbuf_bcat(sb, &structsize, sizeof(structsize));
2339 proc_getauxv(curthread, p, sb);
2345 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2346 const char *note_vendor, const Elf_Phdr *pnote,
2347 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2349 const Elf_Note *note, *note0, *note_end;
2350 const char *note_name;
2355 /* We need some limit, might as well use PAGE_SIZE. */
2356 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2358 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2359 if (pnote->p_offset > PAGE_SIZE ||
2360 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2361 VOP_UNLOCK(imgp->vp, 0);
2362 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2363 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2364 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2365 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2366 curthread->td_ucred, NOCRED, NULL, curthread);
2368 uprintf("i/o error PT_NOTE\n");
2371 note = note0 = (const Elf_Note *)buf;
2372 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2374 note = note0 = (const Elf_Note *)(imgp->image_header +
2376 note_end = (const Elf_Note *)(imgp->image_header +
2377 pnote->p_offset + pnote->p_filesz);
2380 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2381 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2382 (const char *)note < sizeof(Elf_Note)) {
2385 if (note->n_namesz != checknote->n_namesz ||
2386 note->n_descsz != checknote->n_descsz ||
2387 note->n_type != checknote->n_type)
2389 note_name = (const char *)(note + 1);
2390 if (note_name + checknote->n_namesz >=
2391 (const char *)note_end || strncmp(note_vendor,
2392 note_name, checknote->n_namesz) != 0)
2395 if (cb(note, cb_arg, &res))
2398 note = (const Elf_Note *)((const char *)(note + 1) +
2399 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2400 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2409 struct brandnote_cb_arg {
2410 Elf_Brandnote *brandnote;
2415 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2417 struct brandnote_cb_arg *arg;
2422 * Fetch the osreldate for binary from the ELF OSABI-note if
2425 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2426 arg->brandnote->trans_osrel != NULL ?
2427 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2432 static Elf_Note fctl_note = {
2433 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2434 .n_descsz = sizeof(uint32_t),
2435 .n_type = NT_FREEBSD_FEATURE_CTL,
2438 struct fctl_cb_arg {
2443 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2445 struct fctl_cb_arg *arg;
2446 const Elf32_Word *desc;
2450 p = (uintptr_t)(note + 1);
2451 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2452 desc = (const Elf32_Word *)p;
2453 *arg->fctl0 = desc[0];
2458 * Try to find the appropriate ABI-note section for checknote, fetch
2459 * the osreldate and feature control flags for binary from the ELF
2460 * OSABI-note. Only the first page of the image is searched, the same
2464 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2465 int32_t *osrel, uint32_t *fctl0)
2467 const Elf_Phdr *phdr;
2468 const Elf_Ehdr *hdr;
2469 struct brandnote_cb_arg b_arg;
2470 struct fctl_cb_arg f_arg;
2473 hdr = (const Elf_Ehdr *)imgp->image_header;
2474 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2475 b_arg.brandnote = brandnote;
2476 b_arg.osrel = osrel;
2477 f_arg.fctl0 = fctl0;
2479 for (i = 0; i < hdr->e_phnum; i++) {
2480 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2481 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2483 for (j = 0; j < hdr->e_phnum; j++) {
2484 if (phdr[j].p_type == PT_NOTE &&
2485 __elfN(parse_notes)(imgp, &fctl_note,
2486 FREEBSD_ABI_VENDOR, &phdr[j],
2487 note_fctl_cb, &f_arg))
2498 * Tell kern_execve.c about it, with a little help from the linker.
2500 static struct execsw __elfN(execsw) = {
2501 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2502 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2504 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2507 __elfN(trans_prot)(Elf_Word flags)
2513 prot |= VM_PROT_EXECUTE;
2515 prot |= VM_PROT_WRITE;
2517 prot |= VM_PROT_READ;
2518 #if __ELF_WORD_SIZE == 32
2519 #if defined(__amd64__)
2520 if (i386_read_exec && (flags & PF_R))
2521 prot |= VM_PROT_EXECUTE;
2528 __elfN(untrans_prot)(vm_prot_t prot)
2533 if (prot & VM_PROT_EXECUTE)
2535 if (prot & VM_PROT_READ)
2537 if (prot & VM_PROT_WRITE)