2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000 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"
38 #include "opt_compat.h"
40 #include <sys/param.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
48 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
56 #include <sys/procfs.h>
57 #include <sys/ptrace.h>
58 #include <sys/racct.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
62 #include <sys/sf_buf.h>
64 #include <sys/systm.h>
65 #include <sys/signalvar.h>
68 #include <sys/syscall.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/vnode.h>
72 #include <sys/syslog.h>
73 #include <sys/eventhandler.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
84 #include <machine/elf.h>
85 #include <machine/md_var.h>
87 #define ELF_NOTE_ROUNDSIZE 4
88 #define OLD_EI_BRAND 8
90 static int __elfN(check_header)(const Elf_Ehdr *hdr);
91 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
92 const char *interp, int interp_name_len, int32_t *osrel);
93 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 u_long *entry, size_t pagesize);
95 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
96 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
98 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
99 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
101 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
102 static boolean_t __elfN(check_note)(struct image_params *imgp,
103 Elf_Brandnote *checknote, int32_t *osrel);
104 static vm_prot_t __elfN(trans_prot)(Elf_Word);
105 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
107 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
110 #define CORE_BUF_SIZE (16 * 1024)
112 int __elfN(fallback_brand) = -1;
113 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
114 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
115 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 static int elf_legacy_coredump = 0;
118 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
119 &elf_legacy_coredump, 0,
120 "include all and only RW pages in core dumps");
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
124 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
129 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
130 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
131 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
133 #if __ELF_WORD_SIZE == 32
134 #if defined(__amd64__)
135 int i386_read_exec = 0;
136 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
137 "enable execution from readable segments");
141 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
143 #define trunc_page_ps(va, ps) rounddown2(va, ps)
144 #define round_page_ps(va, ps) roundup2(va, ps)
145 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
147 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
149 Elf_Brandnote __elfN(freebsd_brandnote) = {
150 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
151 .hdr.n_descsz = sizeof(int32_t),
152 .hdr.n_type = NT_FREEBSD_ABI_TAG,
153 .vendor = FREEBSD_ABI_VENDOR,
154 .flags = BN_TRANSLATE_OSREL,
155 .trans_osrel = __elfN(freebsd_trans_osrel)
159 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
163 p = (uintptr_t)(note + 1);
164 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
165 *osrel = *(const int32_t *)(p);
170 static const char GNU_ABI_VENDOR[] = "GNU";
171 static int GNU_KFREEBSD_ABI_DESC = 3;
173 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
174 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
175 .hdr.n_descsz = 16, /* XXX at least 16 */
177 .vendor = GNU_ABI_VENDOR,
178 .flags = BN_TRANSLATE_OSREL,
179 .trans_osrel = kfreebsd_trans_osrel
183 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
185 const Elf32_Word *desc;
188 p = (uintptr_t)(note + 1);
189 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
191 desc = (const Elf32_Word *)p;
192 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
196 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
197 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
199 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
205 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
209 for (i = 0; i < MAX_BRANDS; i++) {
210 if (elf_brand_list[i] == NULL) {
211 elf_brand_list[i] = entry;
215 if (i == MAX_BRANDS) {
216 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
224 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
228 for (i = 0; i < MAX_BRANDS; i++) {
229 if (elf_brand_list[i] == entry) {
230 elf_brand_list[i] = NULL;
240 __elfN(brand_inuse)(Elf_Brandinfo *entry)
245 sx_slock(&allproc_lock);
246 FOREACH_PROC_IN_SYSTEM(p) {
247 if (p->p_sysent == entry->sysvec) {
252 sx_sunlock(&allproc_lock);
257 static Elf_Brandinfo *
258 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
259 int interp_name_len, int32_t *osrel)
261 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
262 Elf_Brandinfo *bi, *bi_m;
267 * We support four types of branding -- (1) the ELF EI_OSABI field
268 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
269 * branding w/in the ELF header, (3) path of the `interp_path'
270 * field, and (4) the ".note.ABI-tag" ELF section.
273 /* Look for an ".note.ABI-tag" ELF section */
275 for (i = 0; i < MAX_BRANDS; i++) {
276 bi = elf_brand_list[i];
279 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
281 if (hdr->e_machine == bi->machine && (bi->flags &
282 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
283 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;
793 int error, i, n, interp_name_len, have_interp;
795 hdr = (const Elf_Ehdr *)imgp->image_header;
798 * Do we have a valid ELF header ?
800 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
801 * if particular brand doesn't support it.
803 if (__elfN(check_header)(hdr) != 0 ||
804 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
808 * From here on down, we return an errno, not -1, as we've
809 * detected an ELF file.
812 if ((hdr->e_phoff > PAGE_SIZE) ||
813 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
814 /* Only support headers in first page for now */
815 uprintf("Program headers not in the first page\n");
818 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
819 if (!aligned(phdr, Elf_Addr)) {
820 uprintf("Unaligned program headers\n");
827 text_size = data_size = total_size = text_addr = data_addr = 0;
830 err_str = newinterp = NULL;
831 interp = interp_buf = NULL;
834 for (i = 0; i < hdr->e_phnum; i++) {
835 switch (phdr[i].p_type) {
838 baddr = phdr[i].p_vaddr;
842 /* Path to interpreter */
843 if (phdr[i].p_filesz > MAXPATHLEN) {
844 uprintf("Invalid PT_INTERP\n");
848 if (interp != NULL) {
849 uprintf("Multiple PT_INTERP headers\n");
853 interp_name_len = phdr[i].p_filesz;
854 if (phdr[i].p_offset > PAGE_SIZE ||
855 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
856 VOP_UNLOCK(imgp->vp, 0);
857 interp_buf = malloc(interp_name_len + 1, M_TEMP,
859 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
860 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
861 interp_name_len, phdr[i].p_offset,
862 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
865 uprintf("i/o error PT_INTERP\n");
868 interp_buf[interp_name_len] = '\0';
871 interp = __DECONST(char *, imgp->image_header) +
878 __elfN(trans_prot)(phdr[i].p_flags);
879 imgp->stack_sz = phdr[i].p_memsz;
884 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
886 if (brand_info == NULL) {
887 uprintf("ELF binary type \"%u\" not known.\n",
888 hdr->e_ident[EI_OSABI]);
893 if (hdr->e_type == ET_DYN) {
894 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
895 uprintf("Cannot execute shared object\n");
900 * Honour the base load address from the dso if it is
901 * non-zero for some reason.
904 et_dyn_addr = ET_DYN_LOAD_ADDR;
906 sv = brand_info->sysvec;
907 if (interp != NULL && brand_info->interp_newpath != NULL)
908 newinterp = brand_info->interp_newpath;
911 * Avoid a possible deadlock if the current address space is destroyed
912 * and that address space maps the locked vnode. In the common case,
913 * the locked vnode's v_usecount is decremented but remains greater
914 * than zero. Consequently, the vnode lock is not needed by vrele().
915 * However, in cases where the vnode lock is external, such as nullfs,
916 * v_usecount may become zero.
918 * The VV_TEXT flag prevents modifications to the executable while
919 * the vnode is unlocked.
921 VOP_UNLOCK(imgp->vp, 0);
923 error = exec_new_vmspace(imgp, sv);
924 imgp->proc->p_sysent = sv;
926 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
930 for (i = 0; i < hdr->e_phnum; i++) {
931 switch (phdr[i].p_type) {
932 case PT_LOAD: /* Loadable segment */
933 if (phdr[i].p_memsz == 0)
935 prot = __elfN(trans_prot)(phdr[i].p_flags);
936 error = __elfN(load_section)(imgp, phdr[i].p_offset,
937 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
938 phdr[i].p_memsz, phdr[i].p_filesz, prot,
944 * If this segment contains the program headers,
945 * remember their virtual address for the AT_PHDR
946 * aux entry. Static binaries don't usually include
949 if (phdr[i].p_offset == 0 &&
950 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
952 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
955 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
956 seg_size = round_page(phdr[i].p_memsz +
957 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
960 * Make the largest executable segment the official
961 * text segment and all others data.
963 * Note that obreak() assumes that data_addr +
964 * data_size == end of data load area, and the ELF
965 * file format expects segments to be sorted by
966 * address. If multiple data segments exist, the
967 * last one will be used.
970 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
971 text_size = seg_size;
972 text_addr = seg_addr;
974 data_size = seg_size;
975 data_addr = seg_addr;
977 total_size += seg_size;
979 case PT_PHDR: /* Program header table info */
980 proghdr = phdr[i].p_vaddr + et_dyn_addr;
987 if (data_addr == 0 && data_size == 0) {
988 data_addr = text_addr;
989 data_size = text_size;
992 entry = (u_long)hdr->e_entry + et_dyn_addr;
995 * Check limits. It should be safe to check the
996 * limits after loading the segments since we do
997 * not actually fault in all the segments pages.
999 PROC_LOCK(imgp->proc);
1000 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1001 err_str = "Data segment size exceeds process limit";
1002 else if (text_size > maxtsiz)
1003 err_str = "Text segment size exceeds system limit";
1004 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1005 err_str = "Total segment size exceeds process limit";
1006 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1007 err_str = "Data segment size exceeds resource limit";
1008 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1009 err_str = "Total segment size exceeds resource limit";
1010 if (err_str != NULL) {
1011 PROC_UNLOCK(imgp->proc);
1012 uprintf("%s\n", err_str);
1017 vmspace = imgp->proc->p_vmspace;
1018 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1019 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1020 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1021 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1024 * We load the dynamic linker where a userland call
1025 * to mmap(0, ...) would put it. The rationale behind this
1026 * calculation is that it leaves room for the heap to grow to
1027 * its maximum allowed size.
1029 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1031 PROC_UNLOCK(imgp->proc);
1033 imgp->entry_addr = entry;
1035 if (interp != NULL) {
1036 have_interp = FALSE;
1037 VOP_UNLOCK(imgp->vp, 0);
1038 if (brand_info->emul_path != NULL &&
1039 brand_info->emul_path[0] != '\0') {
1040 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1041 snprintf(path, MAXPATHLEN, "%s%s",
1042 brand_info->emul_path, interp);
1043 error = __elfN(load_file)(imgp->proc, path, &addr,
1044 &imgp->entry_addr, sv->sv_pagesize);
1049 if (!have_interp && newinterp != NULL &&
1050 (brand_info->interp_path == NULL ||
1051 strcmp(interp, brand_info->interp_path) == 0)) {
1052 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1053 &imgp->entry_addr, sv->sv_pagesize);
1058 error = __elfN(load_file)(imgp->proc, interp, &addr,
1059 &imgp->entry_addr, sv->sv_pagesize);
1061 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1063 uprintf("ELF interpreter %s not found, error %d\n",
1071 * Construct auxargs table (used by the fixup routine)
1073 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1074 elf_auxargs->execfd = -1;
1075 elf_auxargs->phdr = proghdr;
1076 elf_auxargs->phent = hdr->e_phentsize;
1077 elf_auxargs->phnum = hdr->e_phnum;
1078 elf_auxargs->pagesz = PAGE_SIZE;
1079 elf_auxargs->base = addr;
1080 elf_auxargs->flags = 0;
1081 elf_auxargs->entry = entry;
1082 elf_auxargs->hdr_eflags = hdr->e_flags;
1084 imgp->auxargs = elf_auxargs;
1085 imgp->interpreted = 0;
1086 imgp->reloc_base = addr;
1087 imgp->proc->p_osrel = osrel;
1088 imgp->proc->p_elf_machine = hdr->e_machine;
1089 imgp->proc->p_elf_flags = hdr->e_flags;
1092 free(interp_buf, M_TEMP);
1096 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1099 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1101 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1105 base = (Elf_Addr *)*stack_base;
1106 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1108 if (args->execfd != -1)
1109 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1110 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1111 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1112 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1113 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1114 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1115 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1116 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1117 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1118 if (imgp->execpathp != 0)
1119 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1120 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1121 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1122 if (imgp->canary != 0) {
1123 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1124 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1126 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1127 if (imgp->pagesizes != 0) {
1128 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1129 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1131 if (imgp->sysent->sv_timekeep_base != 0) {
1132 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1133 imgp->sysent->sv_timekeep_base);
1135 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1136 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1137 imgp->sysent->sv_stackprot);
1138 if (imgp->sysent->sv_hwcap != NULL)
1139 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1140 if (imgp->sysent->sv_hwcap2 != NULL)
1141 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1142 AUXARGS_ENTRY(pos, AT_NULL, 0);
1144 free(imgp->auxargs, M_TEMP);
1145 imgp->auxargs = NULL;
1148 suword(base, (long)imgp->args->argc);
1149 *stack_base = (register_t *)base;
1154 * Code for generating ELF core dumps.
1157 typedef void (*segment_callback)(vm_map_entry_t, void *);
1159 /* Closure for cb_put_phdr(). */
1160 struct phdr_closure {
1161 Elf_Phdr *phdr; /* Program header to fill in */
1162 Elf_Off offset; /* Offset of segment in core file */
1165 /* Closure for cb_size_segment(). */
1166 struct sseg_closure {
1167 int count; /* Count of writable segments. */
1168 size_t size; /* Total size of all writable segments. */
1171 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1174 int type; /* Note type. */
1175 outfunc_t outfunc; /* Output function. */
1176 void *outarg; /* Argument for the output function. */
1177 size_t outsize; /* Output size. */
1178 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1181 TAILQ_HEAD(note_info_list, note_info);
1183 /* Coredump output parameters. */
1184 struct coredump_params {
1186 struct ucred *active_cred;
1187 struct ucred *file_cred;
1190 struct compressor *comp;
1193 extern int compress_user_cores;
1194 extern int compress_user_cores_level;
1196 static void cb_put_phdr(vm_map_entry_t, void *);
1197 static void cb_size_segment(vm_map_entry_t, void *);
1198 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1200 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1201 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1202 struct note_info_list *, size_t);
1203 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1205 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1206 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1207 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1208 static int sbuf_drain_core_output(void *, const char *, int);
1209 static int sbuf_drain_count(void *arg, const char *data, int len);
1211 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1215 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1216 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1217 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1218 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1219 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1220 static void note_procstat_files(void *, struct sbuf *, size_t *);
1221 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1222 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1223 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1224 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1225 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1228 * Write out a core segment to the compression stream.
1231 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1237 chunk_len = MIN(len, CORE_BUF_SIZE);
1240 * We can get EFAULT error here.
1241 * In that case zero out the current chunk of the segment.
1243 error = copyin(base, buf, chunk_len);
1245 bzero(buf, chunk_len);
1246 error = compressor_write(p->comp, buf, chunk_len);
1256 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1259 return (core_write((struct coredump_params *)arg, base, len, offset,
1264 core_write(struct coredump_params *p, const void *base, size_t len,
1265 off_t offset, enum uio_seg seg)
1268 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1269 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1270 p->active_cred, p->file_cred, NULL, p->td));
1274 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1279 if (p->comp != NULL)
1280 return (compress_chunk(p, base, tmpbuf, len));
1283 * EFAULT is a non-fatal error that we can get, for example,
1284 * if the segment is backed by a file but extends beyond its
1287 error = core_write(p, base, len, offset, UIO_USERSPACE);
1288 if (error == EFAULT) {
1289 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1290 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1291 "for process %s\n", base, len, offset, curproc->p_comm);
1294 * Write a "real" zero byte at the end of the target region
1295 * in the case this is the last segment.
1296 * The intermediate space will be implicitly zero-filled.
1298 error = core_write(p, zero_region, 1, offset + len - 1,
1305 * Drain into a core file.
1308 sbuf_drain_core_output(void *arg, const char *data, int len)
1310 struct coredump_params *p;
1313 p = (struct coredump_params *)arg;
1316 * Some kern_proc out routines that print to this sbuf may
1317 * call us with the process lock held. Draining with the
1318 * non-sleepable lock held is unsafe. The lock is needed for
1319 * those routines when dumping a live process. In our case we
1320 * can safely release the lock before draining and acquire
1323 locked = PROC_LOCKED(p->td->td_proc);
1325 PROC_UNLOCK(p->td->td_proc);
1326 if (p->comp != NULL)
1327 error = compressor_write(p->comp, __DECONST(char *, data), len);
1329 error = core_write(p, __DECONST(void *, data), len, p->offset,
1332 PROC_LOCK(p->td->td_proc);
1340 * Drain into a counter.
1343 sbuf_drain_count(void *arg, const char *data __unused, int len)
1347 sizep = (size_t *)arg;
1353 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1355 struct ucred *cred = td->td_ucred;
1357 struct sseg_closure seginfo;
1358 struct note_info_list notelst;
1359 struct coredump_params params;
1360 struct note_info *ninfo;
1362 size_t hdrsize, notesz, coresize;
1366 TAILQ_INIT(¬elst);
1368 /* Size the program segments. */
1371 each_dumpable_segment(td, cb_size_segment, &seginfo);
1374 * Collect info about the core file header area.
1376 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1377 if (seginfo.count + 1 >= PN_XNUM)
1378 hdrsize += sizeof(Elf_Shdr);
1379 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1380 coresize = round_page(hdrsize + notesz) + seginfo.size;
1382 /* Set up core dump parameters. */
1384 params.active_cred = cred;
1385 params.file_cred = NOCRED;
1392 PROC_LOCK(td->td_proc);
1393 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1394 PROC_UNLOCK(td->td_proc);
1401 if (coresize >= limit) {
1406 /* Create a compression stream if necessary. */
1407 if (compress_user_cores != 0) {
1408 params.comp = compressor_init(core_compressed_write,
1409 compress_user_cores, CORE_BUF_SIZE,
1410 compress_user_cores_level, ¶ms);
1411 if (params.comp == NULL) {
1415 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1419 * Allocate memory for building the header, fill it up,
1420 * and write it out following the notes.
1422 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1423 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1426 /* Write the contents of all of the writable segments. */
1432 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1433 offset = round_page(hdrsize + notesz);
1434 for (i = 0; i < seginfo.count; i++) {
1435 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1436 php->p_filesz, offset, ¶ms, tmpbuf);
1439 offset += php->p_filesz;
1442 if (error == 0 && params.comp != NULL)
1443 error = compressor_flush(params.comp);
1447 "Failed to write core file for process %s (error %d)\n",
1448 curproc->p_comm, error);
1452 free(tmpbuf, M_TEMP);
1453 if (params.comp != NULL)
1454 compressor_fini(params.comp);
1455 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1456 TAILQ_REMOVE(¬elst, ninfo, link);
1457 free(ninfo, M_TEMP);
1466 * A callback for each_dumpable_segment() to write out the segment's
1467 * program header entry.
1470 cb_put_phdr(entry, closure)
1471 vm_map_entry_t entry;
1474 struct phdr_closure *phc = (struct phdr_closure *)closure;
1475 Elf_Phdr *phdr = phc->phdr;
1477 phc->offset = round_page(phc->offset);
1479 phdr->p_type = PT_LOAD;
1480 phdr->p_offset = phc->offset;
1481 phdr->p_vaddr = entry->start;
1483 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1484 phdr->p_align = PAGE_SIZE;
1485 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1487 phc->offset += phdr->p_filesz;
1492 * A callback for each_dumpable_segment() to gather information about
1493 * the number of segments and their total size.
1496 cb_size_segment(vm_map_entry_t entry, void *closure)
1498 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1501 ssc->size += entry->end - entry->start;
1505 * For each writable segment in the process's memory map, call the given
1506 * function with a pointer to the map entry and some arbitrary
1507 * caller-supplied data.
1510 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1512 struct proc *p = td->td_proc;
1513 vm_map_t map = &p->p_vmspace->vm_map;
1514 vm_map_entry_t entry;
1515 vm_object_t backing_object, object;
1516 boolean_t ignore_entry;
1518 vm_map_lock_read(map);
1519 for (entry = map->header.next; entry != &map->header;
1520 entry = entry->next) {
1522 * Don't dump inaccessible mappings, deal with legacy
1525 * Note that read-only segments related to the elf binary
1526 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1527 * need to arbitrarily ignore such segments.
1529 if (elf_legacy_coredump) {
1530 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1533 if ((entry->protection & VM_PROT_ALL) == 0)
1538 * Dont include memory segment in the coredump if
1539 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1540 * madvise(2). Do not dump submaps (i.e. parts of the
1543 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1546 if ((object = entry->object.vm_object) == NULL)
1549 /* Ignore memory-mapped devices and such things. */
1550 VM_OBJECT_RLOCK(object);
1551 while ((backing_object = object->backing_object) != NULL) {
1552 VM_OBJECT_RLOCK(backing_object);
1553 VM_OBJECT_RUNLOCK(object);
1554 object = backing_object;
1556 ignore_entry = object->type != OBJT_DEFAULT &&
1557 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1558 object->type != OBJT_PHYS;
1559 VM_OBJECT_RUNLOCK(object);
1563 (*func)(entry, closure);
1565 vm_map_unlock_read(map);
1569 * Write the core file header to the file, including padding up to
1570 * the page boundary.
1573 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1574 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1576 struct note_info *ninfo;
1580 /* Fill in the header. */
1581 bzero(hdr, hdrsize);
1582 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1584 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1585 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1586 sbuf_start_section(sb, NULL);
1587 sbuf_bcat(sb, hdr, hdrsize);
1588 TAILQ_FOREACH(ninfo, notelst, link)
1589 __elfN(putnote)(ninfo, sb);
1590 /* Align up to a page boundary for the program segments. */
1591 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1592 error = sbuf_finish(sb);
1599 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1609 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1612 * To have the debugger select the right thread (LWP) as the initial
1613 * thread, we dump the state of the thread passed to us in td first.
1614 * This is the thread that causes the core dump and thus likely to
1615 * be the right thread one wants to have selected in the debugger.
1618 while (thr != NULL) {
1619 size += register_note(list, NT_PRSTATUS,
1620 __elfN(note_prstatus), thr);
1621 size += register_note(list, NT_FPREGSET,
1622 __elfN(note_fpregset), thr);
1623 size += register_note(list, NT_THRMISC,
1624 __elfN(note_thrmisc), thr);
1625 size += register_note(list, NT_PTLWPINFO,
1626 __elfN(note_ptlwpinfo), thr);
1627 size += register_note(list, -1,
1628 __elfN(note_threadmd), thr);
1630 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1631 TAILQ_NEXT(thr, td_plist);
1633 thr = TAILQ_NEXT(thr, td_plist);
1636 size += register_note(list, NT_PROCSTAT_PROC,
1637 __elfN(note_procstat_proc), p);
1638 size += register_note(list, NT_PROCSTAT_FILES,
1639 note_procstat_files, p);
1640 size += register_note(list, NT_PROCSTAT_VMMAP,
1641 note_procstat_vmmap, p);
1642 size += register_note(list, NT_PROCSTAT_GROUPS,
1643 note_procstat_groups, p);
1644 size += register_note(list, NT_PROCSTAT_UMASK,
1645 note_procstat_umask, p);
1646 size += register_note(list, NT_PROCSTAT_RLIMIT,
1647 note_procstat_rlimit, p);
1648 size += register_note(list, NT_PROCSTAT_OSREL,
1649 note_procstat_osrel, p);
1650 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1651 __elfN(note_procstat_psstrings), p);
1652 size += register_note(list, NT_PROCSTAT_AUXV,
1653 __elfN(note_procstat_auxv), p);
1659 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1665 struct phdr_closure phc;
1667 ehdr = (Elf_Ehdr *)hdr;
1669 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1670 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1671 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1672 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1673 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1674 ehdr->e_ident[EI_DATA] = ELF_DATA;
1675 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1676 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1677 ehdr->e_ident[EI_ABIVERSION] = 0;
1678 ehdr->e_ident[EI_PAD] = 0;
1679 ehdr->e_type = ET_CORE;
1680 ehdr->e_machine = td->td_proc->p_elf_machine;
1681 ehdr->e_version = EV_CURRENT;
1683 ehdr->e_phoff = sizeof(Elf_Ehdr);
1684 ehdr->e_flags = td->td_proc->p_elf_flags;
1685 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1686 ehdr->e_phentsize = sizeof(Elf_Phdr);
1687 ehdr->e_shentsize = sizeof(Elf_Shdr);
1688 ehdr->e_shstrndx = SHN_UNDEF;
1689 if (numsegs + 1 < PN_XNUM) {
1690 ehdr->e_phnum = numsegs + 1;
1693 ehdr->e_phnum = PN_XNUM;
1696 ehdr->e_shoff = ehdr->e_phoff +
1697 (numsegs + 1) * ehdr->e_phentsize;
1698 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1699 ("e_shoff: %zu, hdrsize - shdr: %zu",
1700 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1702 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1703 memset(shdr, 0, sizeof(*shdr));
1705 * A special first section is used to hold large segment and
1706 * section counts. This was proposed by Sun Microsystems in
1707 * Solaris and has been adopted by Linux; the standard ELF
1708 * tools are already familiar with the technique.
1710 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1711 * (or 12-7 depending on the version of the document) for more
1714 shdr->sh_type = SHT_NULL;
1715 shdr->sh_size = ehdr->e_shnum;
1716 shdr->sh_link = ehdr->e_shstrndx;
1717 shdr->sh_info = numsegs + 1;
1721 * Fill in the program header entries.
1723 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1725 /* The note segement. */
1726 phdr->p_type = PT_NOTE;
1727 phdr->p_offset = hdrsize;
1730 phdr->p_filesz = notesz;
1732 phdr->p_flags = PF_R;
1733 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1736 /* All the writable segments from the program. */
1738 phc.offset = round_page(hdrsize + notesz);
1739 each_dumpable_segment(td, cb_put_phdr, &phc);
1743 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1745 struct note_info *ninfo;
1746 size_t size, notesize;
1749 out(arg, NULL, &size);
1750 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1752 ninfo->outfunc = out;
1753 ninfo->outarg = arg;
1754 ninfo->outsize = size;
1755 TAILQ_INSERT_TAIL(list, ninfo, link);
1760 notesize = sizeof(Elf_Note) + /* note header */
1761 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1763 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1769 append_note_data(const void *src, void *dst, size_t len)
1773 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1775 bcopy(src, dst, len);
1776 bzero((char *)dst + len, padded_len - len);
1778 return (padded_len);
1782 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1790 note = (Elf_Note *)buf;
1791 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1792 note->n_descsz = size;
1793 note->n_type = type;
1794 buf += sizeof(*note);
1795 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1796 sizeof(FREEBSD_ABI_VENDOR));
1797 append_note_data(src, buf, size);
1802 notesize = sizeof(Elf_Note) + /* note header */
1803 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1805 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1811 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1814 ssize_t old_len, sect_len;
1815 size_t new_len, descsz, i;
1817 if (ninfo->type == -1) {
1818 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1822 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1823 note.n_descsz = ninfo->outsize;
1824 note.n_type = ninfo->type;
1826 sbuf_bcat(sb, ¬e, sizeof(note));
1827 sbuf_start_section(sb, &old_len);
1828 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1829 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1830 if (note.n_descsz == 0)
1832 sbuf_start_section(sb, &old_len);
1833 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1834 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1838 new_len = (size_t)sect_len;
1839 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1840 if (new_len < descsz) {
1842 * It is expected that individual note emitters will correctly
1843 * predict their expected output size and fill up to that size
1844 * themselves, padding in a format-specific way if needed.
1845 * However, in case they don't, just do it here with zeros.
1847 for (i = 0; i < descsz - new_len; i++)
1849 } else if (new_len > descsz) {
1851 * We can't always truncate sb -- we may have drained some
1854 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1855 "read it (%zu > %zu). Since it is longer than "
1856 "expected, this coredump's notes are corrupt. THIS "
1857 "IS A BUG in the note_procstat routine for type %u.\n",
1858 __func__, (unsigned)note.n_type, new_len, descsz,
1859 (unsigned)note.n_type));
1864 * Miscellaneous note out functions.
1867 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1868 #include <compat/freebsd32/freebsd32.h>
1869 #include <compat/freebsd32/freebsd32_signal.h>
1871 typedef struct prstatus32 elf_prstatus_t;
1872 typedef struct prpsinfo32 elf_prpsinfo_t;
1873 typedef struct fpreg32 elf_prfpregset_t;
1874 typedef struct fpreg32 elf_fpregset_t;
1875 typedef struct reg32 elf_gregset_t;
1876 typedef struct thrmisc32 elf_thrmisc_t;
1877 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1878 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1879 typedef uint32_t elf_ps_strings_t;
1881 typedef prstatus_t elf_prstatus_t;
1882 typedef prpsinfo_t elf_prpsinfo_t;
1883 typedef prfpregset_t elf_prfpregset_t;
1884 typedef prfpregset_t elf_fpregset_t;
1885 typedef gregset_t elf_gregset_t;
1886 typedef thrmisc_t elf_thrmisc_t;
1887 #define ELF_KERN_PROC_MASK 0
1888 typedef struct kinfo_proc elf_kinfo_proc_t;
1889 typedef vm_offset_t elf_ps_strings_t;
1893 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1899 elf_prpsinfo_t *psinfo;
1902 p = (struct proc *)arg;
1904 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1905 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1906 psinfo->pr_version = PRPSINFO_VERSION;
1907 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1908 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1910 if (p->p_args != NULL) {
1911 len = sizeof(psinfo->pr_psargs) - 1;
1912 if (len > p->p_args->ar_length)
1913 len = p->p_args->ar_length;
1914 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1920 sbuf_new(&sbarg, psinfo->pr_psargs,
1921 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1922 error = proc_getargv(curthread, p, &sbarg);
1924 if (sbuf_finish(&sbarg) == 0)
1925 len = sbuf_len(&sbarg) - 1;
1927 len = sizeof(psinfo->pr_psargs) - 1;
1928 sbuf_delete(&sbarg);
1930 if (error || len == 0)
1931 strlcpy(psinfo->pr_psargs, p->p_comm,
1932 sizeof(psinfo->pr_psargs));
1934 KASSERT(len < sizeof(psinfo->pr_psargs),
1935 ("len is too long: %zu vs %zu", len,
1936 sizeof(psinfo->pr_psargs)));
1937 cp = psinfo->pr_psargs;
1940 cp = memchr(cp, '\0', end - cp);
1946 psinfo->pr_pid = p->p_pid;
1947 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1948 free(psinfo, M_TEMP);
1950 *sizep = sizeof(*psinfo);
1954 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1957 elf_prstatus_t *status;
1959 td = (struct thread *)arg;
1961 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1962 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1963 status->pr_version = PRSTATUS_VERSION;
1964 status->pr_statussz = sizeof(elf_prstatus_t);
1965 status->pr_gregsetsz = sizeof(elf_gregset_t);
1966 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1967 status->pr_osreldate = osreldate;
1968 status->pr_cursig = td->td_proc->p_sig;
1969 status->pr_pid = td->td_tid;
1970 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1971 fill_regs32(td, &status->pr_reg);
1973 fill_regs(td, &status->pr_reg);
1975 sbuf_bcat(sb, status, sizeof(*status));
1976 free(status, M_TEMP);
1978 *sizep = sizeof(*status);
1982 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1985 elf_prfpregset_t *fpregset;
1987 td = (struct thread *)arg;
1989 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1990 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1991 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1992 fill_fpregs32(td, fpregset);
1994 fill_fpregs(td, fpregset);
1996 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1997 free(fpregset, M_TEMP);
1999 *sizep = sizeof(*fpregset);
2003 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2006 elf_thrmisc_t thrmisc;
2008 td = (struct thread *)arg;
2010 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2011 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2012 strcpy(thrmisc.pr_tname, td->td_name);
2013 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2015 *sizep = sizeof(thrmisc);
2019 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2024 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2025 struct ptrace_lwpinfo32 pl;
2027 struct ptrace_lwpinfo pl;
2030 td = (struct thread *)arg;
2031 size = sizeof(structsize) + sizeof(pl);
2033 KASSERT(*sizep == size, ("invalid size"));
2034 structsize = sizeof(pl);
2035 sbuf_bcat(sb, &structsize, sizeof(structsize));
2036 bzero(&pl, sizeof(pl));
2037 pl.pl_lwpid = td->td_tid;
2038 pl.pl_event = PL_EVENT_NONE;
2039 pl.pl_sigmask = td->td_sigmask;
2040 pl.pl_siglist = td->td_siglist;
2041 if (td->td_si.si_signo != 0) {
2042 pl.pl_event = PL_EVENT_SIGNAL;
2043 pl.pl_flags |= PL_FLAG_SI;
2044 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2045 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2047 pl.pl_siginfo = td->td_si;
2050 strcpy(pl.pl_tdname, td->td_name);
2051 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2052 sbuf_bcat(sb, &pl, sizeof(pl));
2058 * Allow for MD specific notes, as well as any MD
2059 * specific preparations for writing MI notes.
2062 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2068 td = (struct thread *)arg;
2070 if (size != 0 && sb != NULL)
2071 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2075 __elfN(dump_thread)(td, buf, &size);
2076 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2077 if (size != 0 && sb != NULL)
2078 sbuf_bcat(sb, buf, size);
2083 #ifdef KINFO_PROC_SIZE
2084 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2088 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2094 p = (struct proc *)arg;
2095 size = sizeof(structsize) + p->p_numthreads *
2096 sizeof(elf_kinfo_proc_t);
2099 KASSERT(*sizep == size, ("invalid size"));
2100 structsize = sizeof(elf_kinfo_proc_t);
2101 sbuf_bcat(sb, &structsize, sizeof(structsize));
2102 sx_slock(&proctree_lock);
2104 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2105 sx_sunlock(&proctree_lock);
2110 #ifdef KINFO_FILE_SIZE
2111 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2115 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2118 size_t size, sect_sz, i;
2119 ssize_t start_len, sect_len;
2120 int structsize, filedesc_flags;
2122 if (coredump_pack_fileinfo)
2123 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2127 p = (struct proc *)arg;
2128 structsize = sizeof(struct kinfo_file);
2131 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2132 sbuf_set_drain(sb, sbuf_drain_count, &size);
2133 sbuf_bcat(sb, &structsize, sizeof(structsize));
2135 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2140 sbuf_start_section(sb, &start_len);
2142 sbuf_bcat(sb, &structsize, sizeof(structsize));
2144 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2147 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2152 KASSERT(sect_sz <= *sizep,
2153 ("kern_proc_filedesc_out did not respect maxlen; "
2154 "requested %zu, got %zu", *sizep - sizeof(structsize),
2155 sect_sz - sizeof(structsize)));
2157 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2162 #ifdef KINFO_VMENTRY_SIZE
2163 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2167 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2171 int structsize, vmmap_flags;
2173 if (coredump_pack_vmmapinfo)
2174 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2178 p = (struct proc *)arg;
2179 structsize = sizeof(struct kinfo_vmentry);
2182 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2183 sbuf_set_drain(sb, sbuf_drain_count, &size);
2184 sbuf_bcat(sb, &structsize, sizeof(structsize));
2186 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2191 sbuf_bcat(sb, &structsize, sizeof(structsize));
2193 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2199 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2205 p = (struct proc *)arg;
2206 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2208 KASSERT(*sizep == size, ("invalid size"));
2209 structsize = sizeof(gid_t);
2210 sbuf_bcat(sb, &structsize, sizeof(structsize));
2211 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2218 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2224 p = (struct proc *)arg;
2225 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2227 KASSERT(*sizep == size, ("invalid size"));
2228 structsize = sizeof(p->p_fd->fd_cmask);
2229 sbuf_bcat(sb, &structsize, sizeof(structsize));
2230 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2236 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2239 struct rlimit rlim[RLIM_NLIMITS];
2243 p = (struct proc *)arg;
2244 size = sizeof(structsize) + sizeof(rlim);
2246 KASSERT(*sizep == size, ("invalid size"));
2247 structsize = sizeof(rlim);
2248 sbuf_bcat(sb, &structsize, sizeof(structsize));
2250 for (i = 0; i < RLIM_NLIMITS; i++)
2251 lim_rlimit_proc(p, i, &rlim[i]);
2253 sbuf_bcat(sb, rlim, sizeof(rlim));
2259 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2265 p = (struct proc *)arg;
2266 size = sizeof(structsize) + sizeof(p->p_osrel);
2268 KASSERT(*sizep == size, ("invalid size"));
2269 structsize = sizeof(p->p_osrel);
2270 sbuf_bcat(sb, &structsize, sizeof(structsize));
2271 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2277 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2280 elf_ps_strings_t ps_strings;
2284 p = (struct proc *)arg;
2285 size = sizeof(structsize) + sizeof(ps_strings);
2287 KASSERT(*sizep == size, ("invalid size"));
2288 structsize = sizeof(ps_strings);
2289 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2290 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2292 ps_strings = p->p_sysent->sv_psstrings;
2294 sbuf_bcat(sb, &structsize, sizeof(structsize));
2295 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2301 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2307 p = (struct proc *)arg;
2310 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2311 sbuf_set_drain(sb, sbuf_drain_count, &size);
2312 sbuf_bcat(sb, &structsize, sizeof(structsize));
2314 proc_getauxv(curthread, p, sb);
2320 structsize = sizeof(Elf_Auxinfo);
2321 sbuf_bcat(sb, &structsize, sizeof(structsize));
2323 proc_getauxv(curthread, p, sb);
2329 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2330 int32_t *osrel, const Elf_Phdr *pnote)
2332 const Elf_Note *note, *note0, *note_end;
2333 const char *note_name;
2338 /* We need some limit, might as well use PAGE_SIZE. */
2339 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2341 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2342 if (pnote->p_offset > PAGE_SIZE ||
2343 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2344 VOP_UNLOCK(imgp->vp, 0);
2345 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2346 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2347 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2348 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2349 curthread->td_ucred, NOCRED, NULL, curthread);
2351 uprintf("i/o error PT_NOTE\n");
2355 note = note0 = (const Elf_Note *)buf;
2356 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2358 note = note0 = (const Elf_Note *)(imgp->image_header +
2360 note_end = (const Elf_Note *)(imgp->image_header +
2361 pnote->p_offset + pnote->p_filesz);
2364 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2365 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2366 (const char *)note < sizeof(Elf_Note)) {
2370 if (note->n_namesz != checknote->hdr.n_namesz ||
2371 note->n_descsz != checknote->hdr.n_descsz ||
2372 note->n_type != checknote->hdr.n_type)
2374 note_name = (const char *)(note + 1);
2375 if (note_name + checknote->hdr.n_namesz >=
2376 (const char *)note_end || strncmp(checknote->vendor,
2377 note_name, checknote->hdr.n_namesz) != 0)
2381 * Fetch the osreldate for binary
2382 * from the ELF OSABI-note if necessary.
2384 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2385 checknote->trans_osrel != NULL) {
2386 res = checknote->trans_osrel(note, osrel);
2392 note = (const Elf_Note *)((const char *)(note + 1) +
2393 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2394 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2403 * Try to find the appropriate ABI-note section for checknote,
2404 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2405 * first page of the image is searched, the same as for headers.
2408 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2411 const Elf_Phdr *phdr;
2412 const Elf_Ehdr *hdr;
2415 hdr = (const Elf_Ehdr *)imgp->image_header;
2416 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2418 for (i = 0; i < hdr->e_phnum; i++) {
2419 if (phdr[i].p_type == PT_NOTE &&
2420 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2428 * Tell kern_execve.c about it, with a little help from the linker.
2430 static struct execsw __elfN(execsw) = {
2431 __CONCAT(exec_, __elfN(imgact)),
2432 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2434 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2437 __elfN(trans_prot)(Elf_Word flags)
2443 prot |= VM_PROT_EXECUTE;
2445 prot |= VM_PROT_WRITE;
2447 prot |= VM_PROT_READ;
2448 #if __ELF_WORD_SIZE == 32
2449 #if defined(__amd64__)
2450 if (i386_read_exec && (flags & PF_R))
2451 prot |= VM_PROT_EXECUTE;
2458 __elfN(untrans_prot)(vm_prot_t prot)
2463 if (prot & VM_PROT_EXECUTE)
2465 if (prot & VM_PROT_READ)
2467 if (prot & VM_PROT_WRITE)