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__) || \
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, uint32_t *fctl0)
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,
285 /* Give brand a chance to veto check_note's guess */
286 if (ret && bi->header_supported)
287 ret = bi->header_supported(imgp);
289 * If note checker claimed the binary, but the
290 * interpreter path in the image does not
291 * match default one for the brand, try to
292 * search for other brands with the same
293 * interpreter. Either there is better brand
294 * with the right interpreter, or, failing
295 * this, we return first brand which accepted
296 * our note and, optionally, header.
298 if (ret && bi_m == NULL && interp != NULL &&
299 (bi->interp_path == NULL ||
300 (strlen(bi->interp_path) + 1 != interp_name_len ||
301 strncmp(interp, bi->interp_path, interp_name_len)
313 /* If the executable has a brand, search for it in the brand list. */
314 for (i = 0; i < MAX_BRANDS; i++) {
315 bi = elf_brand_list[i];
316 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
317 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
319 if (hdr->e_machine == bi->machine &&
320 (hdr->e_ident[EI_OSABI] == bi->brand ||
321 (bi->compat_3_brand != NULL &&
322 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
323 bi->compat_3_brand) == 0))) {
324 /* Looks good, but give brand a chance to veto */
325 if (bi->header_supported == NULL ||
326 bi->header_supported(imgp)) {
328 * Again, prefer strictly matching
331 if (interp_name_len == 0 &&
332 bi->interp_path == NULL)
334 if (bi->interp_path != NULL &&
335 strlen(bi->interp_path) + 1 ==
336 interp_name_len && strncmp(interp,
337 bi->interp_path, interp_name_len) == 0)
347 /* No known brand, see if the header is recognized by any brand */
348 for (i = 0; i < MAX_BRANDS; i++) {
349 bi = elf_brand_list[i];
350 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
351 bi->header_supported == NULL)
353 if (hdr->e_machine == bi->machine) {
354 ret = bi->header_supported(imgp);
360 /* Lacking a known brand, search for a recognized interpreter. */
361 if (interp != NULL) {
362 for (i = 0; i < MAX_BRANDS; i++) {
363 bi = elf_brand_list[i];
364 if (bi == NULL || (bi->flags &
365 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
368 if (hdr->e_machine == bi->machine &&
369 bi->interp_path != NULL &&
370 /* ELF image p_filesz includes terminating zero */
371 strlen(bi->interp_path) + 1 == interp_name_len &&
372 strncmp(interp, bi->interp_path, interp_name_len)
373 == 0 && (bi->header_supported == NULL ||
374 bi->header_supported(imgp)))
379 /* Lacking a recognized interpreter, try the default brand */
380 for (i = 0; i < MAX_BRANDS; i++) {
381 bi = elf_brand_list[i];
382 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
383 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
385 if (hdr->e_machine == bi->machine &&
386 __elfN(fallback_brand) == bi->brand &&
387 (bi->header_supported == NULL ||
388 bi->header_supported(imgp)))
395 __elfN(check_header)(const Elf_Ehdr *hdr)
401 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
402 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
403 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
404 hdr->e_phentsize != sizeof(Elf_Phdr) ||
405 hdr->e_version != ELF_TARG_VER)
409 * Make sure we have at least one brand for this machine.
412 for (i = 0; i < MAX_BRANDS; i++) {
413 bi = elf_brand_list[i];
414 if (bi != NULL && bi->machine == hdr->e_machine)
424 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
425 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
432 * Create the page if it doesn't exist yet. Ignore errors.
434 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
435 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
438 * Find the page from the underlying object.
440 if (object != NULL) {
441 sf = vm_imgact_map_page(object, offset);
443 return (KERN_FAILURE);
444 off = offset - trunc_page(offset);
445 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
447 vm_imgact_unmap_page(sf);
449 return (KERN_FAILURE);
452 return (KERN_SUCCESS);
456 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
457 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
463 int error, locked, rv;
465 if (start != trunc_page(start)) {
466 rv = __elfN(map_partial)(map, object, offset, start,
467 round_page(start), prot);
468 if (rv != KERN_SUCCESS)
470 offset += round_page(start) - start;
471 start = round_page(start);
473 if (end != round_page(end)) {
474 rv = __elfN(map_partial)(map, object, offset +
475 trunc_page(end) - start, trunc_page(end), end, prot);
476 if (rv != KERN_SUCCESS)
478 end = trunc_page(end);
481 return (KERN_SUCCESS);
482 if ((offset & PAGE_MASK) != 0) {
484 * The mapping is not page aligned. This means that we have
487 rv = vm_map_fixed(map, NULL, 0, start, end - start,
488 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
489 if (rv != KERN_SUCCESS)
492 return (KERN_SUCCESS);
493 for (; start < end; start += sz) {
494 sf = vm_imgact_map_page(object, offset);
496 return (KERN_FAILURE);
497 off = offset - trunc_page(offset);
499 if (sz > PAGE_SIZE - off)
500 sz = PAGE_SIZE - off;
501 error = copyout((caddr_t)sf_buf_kva(sf) + off,
503 vm_imgact_unmap_page(sf);
505 return (KERN_FAILURE);
509 vm_object_reference(object);
510 rv = vm_map_fixed(map, object, offset, start, end - start,
511 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
512 if (rv != KERN_SUCCESS) {
513 locked = VOP_ISLOCKED(imgp->vp);
514 VOP_UNLOCK(imgp->vp, 0);
515 vm_object_deallocate(object);
516 vn_lock(imgp->vp, locked | LK_RETRY);
520 return (KERN_SUCCESS);
524 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
525 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
532 vm_offset_t off, map_addr;
535 vm_ooffset_t file_addr;
538 * It's necessary to fail if the filsz + offset taken from the
539 * header is greater than the actual file pager object's size.
540 * If we were to allow this, then the vm_map_find() below would
541 * walk right off the end of the file object and into the ether.
543 * While I'm here, might as well check for something else that
544 * is invalid: filsz cannot be greater than memsz.
546 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
548 uprintf("elf_load_section: truncated ELF file\n");
552 object = imgp->object;
553 map = &imgp->proc->p_vmspace->vm_map;
554 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
555 file_addr = trunc_page_ps(offset, pagesize);
558 * We have two choices. We can either clear the data in the last page
559 * of an oversized mapping, or we can start the anon mapping a page
560 * early and copy the initialized data into that first page. We
565 else if (memsz > filsz)
566 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
568 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
571 /* cow flags: don't dump readonly sections in core */
572 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
573 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
575 rv = __elfN(map_insert)(imgp, map,
577 file_addr, /* file offset */
578 map_addr, /* virtual start */
579 map_addr + map_len,/* virtual end */
582 if (rv != KERN_SUCCESS)
585 /* we can stop now if we've covered it all */
592 * We have to get the remaining bit of the file into the first part
593 * of the oversized map segment. This is normally because the .data
594 * segment in the file is extended to provide bss. It's a neat idea
595 * to try and save a page, but it's a pain in the behind to implement.
597 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
599 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
600 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
603 /* This had damn well better be true! */
605 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
606 map_addr + map_len, prot, 0);
607 if (rv != KERN_SUCCESS)
612 sf = vm_imgact_map_page(object, offset + filsz);
616 /* send the page fragment to user space */
617 off = trunc_page_ps(offset + filsz, pagesize) -
618 trunc_page(offset + filsz);
619 error = copyout((caddr_t)sf_buf_kva(sf) + off,
620 (caddr_t)map_addr, copy_len);
621 vm_imgact_unmap_page(sf);
627 * Remove write access to the page if it was only granted by map_insert
630 if ((prot & VM_PROT_WRITE) == 0)
631 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
632 map_len), prot, FALSE);
638 * Load the file "file" into memory. It may be either a shared object
641 * The "addr" reference parameter is in/out. On entry, it specifies
642 * the address where a shared object should be loaded. If the file is
643 * an executable, this value is ignored. On exit, "addr" specifies
644 * where the file was actually loaded.
646 * The "entry" reference parameter is out only. On exit, it specifies
647 * the entry point for the loaded file.
650 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
651 u_long *entry, size_t pagesize)
656 struct image_params image_params;
658 const Elf_Ehdr *hdr = NULL;
659 const Elf_Phdr *phdr = NULL;
660 struct nameidata *nd;
662 struct image_params *imgp;
665 u_long base_addr = 0;
666 int error, i, numsegs;
668 #ifdef CAPABILITY_MODE
670 * XXXJA: This check can go away once we are sufficiently confident
671 * that the checks in namei() are correct.
673 if (IN_CAPABILITY_MODE(curthread))
677 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
679 attr = &tempdata->attr;
680 imgp = &tempdata->image_params;
683 * Initialize part of the common data
687 imgp->firstpage = NULL;
688 imgp->image_header = NULL;
690 imgp->execlabel = NULL;
692 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
693 if ((error = namei(nd)) != 0) {
697 NDFREE(nd, NDF_ONLY_PNBUF);
698 imgp->vp = nd->ni_vp;
701 * Check permissions, modes, uid, etc on the file, and "open" it.
703 error = exec_check_permissions(imgp);
707 error = exec_map_first_page(imgp);
712 * Also make certain that the interpreter stays the same, so set
713 * its VV_TEXT flag, too.
715 VOP_SET_TEXT(nd->ni_vp);
717 imgp->object = nd->ni_vp->v_object;
719 hdr = (const Elf_Ehdr *)imgp->image_header;
720 if ((error = __elfN(check_header)(hdr)) != 0)
722 if (hdr->e_type == ET_DYN)
724 else if (hdr->e_type == ET_EXEC)
731 /* Only support headers that fit within first page for now */
732 if ((hdr->e_phoff > PAGE_SIZE) ||
733 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
738 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
739 if (!aligned(phdr, Elf_Addr)) {
744 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
745 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
746 /* Loadable segment */
747 prot = __elfN(trans_prot)(phdr[i].p_flags);
748 error = __elfN(load_section)(imgp, phdr[i].p_offset,
749 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
750 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
754 * Establish the base address if this is the
758 base_addr = trunc_page(phdr[i].p_vaddr +
764 *entry = (unsigned long)hdr->e_entry + rbase;
768 exec_unmap_first_page(imgp);
773 free(tempdata, M_TEMP);
779 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
783 const Elf_Phdr *phdr;
784 Elf_Auxargs *elf_auxargs;
785 struct vmspace *vmspace;
786 const char *err_str, *newinterp;
787 char *interp, *interp_buf, *path;
788 Elf_Brandinfo *brand_info;
789 struct sysentvec *sv;
791 u_long text_size, data_size, total_size, text_addr, data_addr;
792 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
795 int error, i, n, interp_name_len, have_interp;
797 hdr = (const Elf_Ehdr *)imgp->image_header;
800 * Do we have a valid ELF header ?
802 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
803 * if particular brand doesn't support it.
805 if (__elfN(check_header)(hdr) != 0 ||
806 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
810 * From here on down, we return an errno, not -1, as we've
811 * detected an ELF file.
814 if ((hdr->e_phoff > PAGE_SIZE) ||
815 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
816 /* Only support headers in first page for now */
817 uprintf("Program headers not in the first page\n");
820 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
821 if (!aligned(phdr, Elf_Addr)) {
822 uprintf("Unaligned program headers\n");
830 text_size = data_size = total_size = text_addr = data_addr = 0;
833 err_str = newinterp = NULL;
834 interp = interp_buf = NULL;
837 for (i = 0; i < hdr->e_phnum; i++) {
838 switch (phdr[i].p_type) {
841 baddr = phdr[i].p_vaddr;
845 /* Path to interpreter */
846 if (phdr[i].p_filesz < 2 ||
847 phdr[i].p_filesz > MAXPATHLEN) {
848 uprintf("Invalid PT_INTERP\n");
852 if (interp != NULL) {
853 uprintf("Multiple PT_INTERP headers\n");
857 interp_name_len = phdr[i].p_filesz;
858 if (phdr[i].p_offset > PAGE_SIZE ||
859 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
860 VOP_UNLOCK(imgp->vp, 0);
861 interp_buf = malloc(interp_name_len + 1, M_TEMP,
863 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
864 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
865 interp_name_len, phdr[i].p_offset,
866 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
869 uprintf("i/o error PT_INTERP %d\n",
873 interp_buf[interp_name_len] = '\0';
876 interp = __DECONST(char *, imgp->image_header) +
878 if (interp[interp_name_len - 1] != '\0') {
879 uprintf("Invalid PT_INTERP\n");
888 __elfN(trans_prot)(phdr[i].p_flags);
889 imgp->stack_sz = phdr[i].p_memsz;
894 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
896 if (brand_info == NULL) {
897 uprintf("ELF binary type \"%u\" not known.\n",
898 hdr->e_ident[EI_OSABI]);
903 if (hdr->e_type == ET_DYN) {
904 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
905 uprintf("Cannot execute shared object\n");
910 * Honour the base load address from the dso if it is
911 * non-zero for some reason.
914 et_dyn_addr = ET_DYN_LOAD_ADDR;
916 sv = brand_info->sysvec;
917 if (interp != NULL && brand_info->interp_newpath != NULL)
918 newinterp = brand_info->interp_newpath;
921 * Avoid a possible deadlock if the current address space is destroyed
922 * and that address space maps the locked vnode. In the common case,
923 * the locked vnode's v_usecount is decremented but remains greater
924 * than zero. Consequently, the vnode lock is not needed by vrele().
925 * However, in cases where the vnode lock is external, such as nullfs,
926 * v_usecount may become zero.
928 * The VV_TEXT flag prevents modifications to the executable while
929 * the vnode is unlocked.
931 VOP_UNLOCK(imgp->vp, 0);
933 error = exec_new_vmspace(imgp, sv);
934 imgp->proc->p_sysent = sv;
936 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
940 for (i = 0; i < hdr->e_phnum; i++) {
941 switch (phdr[i].p_type) {
942 case PT_LOAD: /* Loadable segment */
943 if (phdr[i].p_memsz == 0)
945 prot = __elfN(trans_prot)(phdr[i].p_flags);
946 error = __elfN(load_section)(imgp, phdr[i].p_offset,
947 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
948 phdr[i].p_memsz, phdr[i].p_filesz, prot,
954 * If this segment contains the program headers,
955 * remember their virtual address for the AT_PHDR
956 * aux entry. Static binaries don't usually include
959 if (phdr[i].p_offset == 0 &&
960 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
962 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
965 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
966 seg_size = round_page(phdr[i].p_memsz +
967 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
970 * Make the largest executable segment the official
971 * text segment and all others data.
973 * Note that obreak() assumes that data_addr +
974 * data_size == end of data load area, and the ELF
975 * file format expects segments to be sorted by
976 * address. If multiple data segments exist, the
977 * last one will be used.
980 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
981 text_size = seg_size;
982 text_addr = seg_addr;
984 data_size = seg_size;
985 data_addr = seg_addr;
987 total_size += seg_size;
989 case PT_PHDR: /* Program header table info */
990 proghdr = phdr[i].p_vaddr + et_dyn_addr;
997 if (data_addr == 0 && data_size == 0) {
998 data_addr = text_addr;
999 data_size = text_size;
1002 entry = (u_long)hdr->e_entry + et_dyn_addr;
1005 * Check limits. It should be safe to check the
1006 * limits after loading the segments since we do
1007 * not actually fault in all the segments pages.
1009 PROC_LOCK(imgp->proc);
1010 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1011 err_str = "Data segment size exceeds process limit";
1012 else if (text_size > maxtsiz)
1013 err_str = "Text segment size exceeds system limit";
1014 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1015 err_str = "Total segment size exceeds process limit";
1016 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1017 err_str = "Data segment size exceeds resource limit";
1018 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1019 err_str = "Total segment size exceeds resource limit";
1020 if (err_str != NULL) {
1021 PROC_UNLOCK(imgp->proc);
1022 uprintf("%s\n", err_str);
1027 vmspace = imgp->proc->p_vmspace;
1028 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1029 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1030 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1031 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1034 * We load the dynamic linker where a userland call
1035 * to mmap(0, ...) would put it. The rationale behind this
1036 * calculation is that it leaves room for the heap to grow to
1037 * its maximum allowed size.
1039 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1041 PROC_UNLOCK(imgp->proc);
1043 imgp->entry_addr = entry;
1045 if (interp != NULL) {
1046 have_interp = FALSE;
1047 VOP_UNLOCK(imgp->vp, 0);
1048 if (brand_info->emul_path != NULL &&
1049 brand_info->emul_path[0] != '\0') {
1050 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1051 snprintf(path, MAXPATHLEN, "%s%s",
1052 brand_info->emul_path, interp);
1053 error = __elfN(load_file)(imgp->proc, path, &addr,
1054 &imgp->entry_addr, sv->sv_pagesize);
1059 if (!have_interp && newinterp != NULL &&
1060 (brand_info->interp_path == NULL ||
1061 strcmp(interp, brand_info->interp_path) == 0)) {
1062 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1063 &imgp->entry_addr, sv->sv_pagesize);
1068 error = __elfN(load_file)(imgp->proc, interp, &addr,
1069 &imgp->entry_addr, sv->sv_pagesize);
1071 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1073 uprintf("ELF interpreter %s not found, error %d\n",
1081 * Construct auxargs table (used by the fixup routine)
1083 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1084 elf_auxargs->execfd = -1;
1085 elf_auxargs->phdr = proghdr;
1086 elf_auxargs->phent = hdr->e_phentsize;
1087 elf_auxargs->phnum = hdr->e_phnum;
1088 elf_auxargs->pagesz = PAGE_SIZE;
1089 elf_auxargs->base = addr;
1090 elf_auxargs->flags = 0;
1091 elf_auxargs->entry = entry;
1092 elf_auxargs->hdr_eflags = hdr->e_flags;
1094 imgp->auxargs = elf_auxargs;
1095 imgp->interpreted = 0;
1096 imgp->reloc_base = addr;
1097 imgp->proc->p_osrel = osrel;
1098 imgp->proc->p_fctl0 = fctl0;
1099 imgp->proc->p_elf_machine = hdr->e_machine;
1100 imgp->proc->p_elf_flags = hdr->e_flags;
1103 free(interp_buf, M_TEMP);
1107 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1110 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1112 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1113 Elf_Auxinfo *argarray, *pos;
1114 Elf_Addr *base, *auxbase;
1117 base = (Elf_Addr *)*stack_base;
1118 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1119 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1122 if (args->execfd != -1)
1123 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1124 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1125 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1126 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1127 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1128 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1129 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1130 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1131 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1132 if (imgp->execpathp != 0)
1133 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1134 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1135 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1136 if (imgp->canary != 0) {
1137 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1138 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1140 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1141 if (imgp->pagesizes != 0) {
1142 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1143 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1145 if (imgp->sysent->sv_timekeep_base != 0) {
1146 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1147 imgp->sysent->sv_timekeep_base);
1149 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1150 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1151 imgp->sysent->sv_stackprot);
1152 if (imgp->sysent->sv_hwcap != NULL)
1153 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1154 if (imgp->sysent->sv_hwcap2 != NULL)
1155 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1156 AUXARGS_ENTRY(pos, AT_NULL, 0);
1158 free(imgp->auxargs, M_TEMP);
1159 imgp->auxargs = NULL;
1160 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1162 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1163 free(argarray, M_TEMP);
1168 if (suword(base, imgp->args->argc) == -1)
1170 *stack_base = (register_t *)base;
1175 * Code for generating ELF core dumps.
1178 typedef void (*segment_callback)(vm_map_entry_t, void *);
1180 /* Closure for cb_put_phdr(). */
1181 struct phdr_closure {
1182 Elf_Phdr *phdr; /* Program header to fill in */
1183 Elf_Off offset; /* Offset of segment in core file */
1186 /* Closure for cb_size_segment(). */
1187 struct sseg_closure {
1188 int count; /* Count of writable segments. */
1189 size_t size; /* Total size of all writable segments. */
1192 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1195 int type; /* Note type. */
1196 outfunc_t outfunc; /* Output function. */
1197 void *outarg; /* Argument for the output function. */
1198 size_t outsize; /* Output size. */
1199 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1202 TAILQ_HEAD(note_info_list, note_info);
1204 /* Coredump output parameters. */
1205 struct coredump_params {
1207 struct ucred *active_cred;
1208 struct ucred *file_cred;
1211 struct compressor *comp;
1214 extern int compress_user_cores;
1215 extern int compress_user_cores_level;
1217 static void cb_put_phdr(vm_map_entry_t, void *);
1218 static void cb_size_segment(vm_map_entry_t, void *);
1219 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1221 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1222 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1223 struct note_info_list *, size_t);
1224 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1226 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1227 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1228 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1229 static int sbuf_drain_core_output(void *, const char *, int);
1230 static int sbuf_drain_count(void *arg, const char *data, int len);
1232 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1233 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1234 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1235 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1236 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1237 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1238 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1239 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1240 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1241 static void note_procstat_files(void *, struct sbuf *, size_t *);
1242 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1243 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1244 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1245 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1246 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1249 * Write out a core segment to the compression stream.
1252 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1258 chunk_len = MIN(len, CORE_BUF_SIZE);
1261 * We can get EFAULT error here.
1262 * In that case zero out the current chunk of the segment.
1264 error = copyin(base, buf, chunk_len);
1266 bzero(buf, chunk_len);
1267 error = compressor_write(p->comp, buf, chunk_len);
1277 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1280 return (core_write((struct coredump_params *)arg, base, len, offset,
1285 core_write(struct coredump_params *p, const void *base, size_t len,
1286 off_t offset, enum uio_seg seg)
1289 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1290 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1291 p->active_cred, p->file_cred, NULL, p->td));
1295 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1300 if (p->comp != NULL)
1301 return (compress_chunk(p, base, tmpbuf, len));
1304 * EFAULT is a non-fatal error that we can get, for example,
1305 * if the segment is backed by a file but extends beyond its
1308 error = core_write(p, base, len, offset, UIO_USERSPACE);
1309 if (error == EFAULT) {
1310 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1311 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1312 "for process %s\n", base, len, offset, curproc->p_comm);
1315 * Write a "real" zero byte at the end of the target region
1316 * in the case this is the last segment.
1317 * The intermediate space will be implicitly zero-filled.
1319 error = core_write(p, zero_region, 1, offset + len - 1,
1326 * Drain into a core file.
1329 sbuf_drain_core_output(void *arg, const char *data, int len)
1331 struct coredump_params *p;
1334 p = (struct coredump_params *)arg;
1337 * Some kern_proc out routines that print to this sbuf may
1338 * call us with the process lock held. Draining with the
1339 * non-sleepable lock held is unsafe. The lock is needed for
1340 * those routines when dumping a live process. In our case we
1341 * can safely release the lock before draining and acquire
1344 locked = PROC_LOCKED(p->td->td_proc);
1346 PROC_UNLOCK(p->td->td_proc);
1347 if (p->comp != NULL)
1348 error = compressor_write(p->comp, __DECONST(char *, data), len);
1350 error = core_write(p, __DECONST(void *, data), len, p->offset,
1353 PROC_LOCK(p->td->td_proc);
1361 * Drain into a counter.
1364 sbuf_drain_count(void *arg, const char *data __unused, int len)
1368 sizep = (size_t *)arg;
1374 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1376 struct ucred *cred = td->td_ucred;
1378 struct sseg_closure seginfo;
1379 struct note_info_list notelst;
1380 struct coredump_params params;
1381 struct note_info *ninfo;
1383 size_t hdrsize, notesz, coresize;
1387 TAILQ_INIT(¬elst);
1389 /* Size the program segments. */
1392 each_dumpable_segment(td, cb_size_segment, &seginfo);
1395 * Collect info about the core file header area.
1397 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1398 if (seginfo.count + 1 >= PN_XNUM)
1399 hdrsize += sizeof(Elf_Shdr);
1400 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1401 coresize = round_page(hdrsize + notesz) + seginfo.size;
1403 /* Set up core dump parameters. */
1405 params.active_cred = cred;
1406 params.file_cred = NOCRED;
1413 PROC_LOCK(td->td_proc);
1414 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1415 PROC_UNLOCK(td->td_proc);
1422 if (coresize >= limit) {
1427 /* Create a compression stream if necessary. */
1428 if (compress_user_cores != 0) {
1429 params.comp = compressor_init(core_compressed_write,
1430 compress_user_cores, CORE_BUF_SIZE,
1431 compress_user_cores_level, ¶ms);
1432 if (params.comp == NULL) {
1436 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1440 * Allocate memory for building the header, fill it up,
1441 * and write it out following the notes.
1443 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1444 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1447 /* Write the contents of all of the writable segments. */
1453 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1454 offset = round_page(hdrsize + notesz);
1455 for (i = 0; i < seginfo.count; i++) {
1456 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1457 php->p_filesz, offset, ¶ms, tmpbuf);
1460 offset += php->p_filesz;
1463 if (error == 0 && params.comp != NULL)
1464 error = compressor_flush(params.comp);
1468 "Failed to write core file for process %s (error %d)\n",
1469 curproc->p_comm, error);
1473 free(tmpbuf, M_TEMP);
1474 if (params.comp != NULL)
1475 compressor_fini(params.comp);
1476 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1477 TAILQ_REMOVE(¬elst, ninfo, link);
1478 free(ninfo, M_TEMP);
1487 * A callback for each_dumpable_segment() to write out the segment's
1488 * program header entry.
1491 cb_put_phdr(vm_map_entry_t entry, void *closure)
1493 struct phdr_closure *phc = (struct phdr_closure *)closure;
1494 Elf_Phdr *phdr = phc->phdr;
1496 phc->offset = round_page(phc->offset);
1498 phdr->p_type = PT_LOAD;
1499 phdr->p_offset = phc->offset;
1500 phdr->p_vaddr = entry->start;
1502 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1503 phdr->p_align = PAGE_SIZE;
1504 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1506 phc->offset += phdr->p_filesz;
1511 * A callback for each_dumpable_segment() to gather information about
1512 * the number of segments and their total size.
1515 cb_size_segment(vm_map_entry_t entry, void *closure)
1517 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1520 ssc->size += entry->end - entry->start;
1524 * For each writable segment in the process's memory map, call the given
1525 * function with a pointer to the map entry and some arbitrary
1526 * caller-supplied data.
1529 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1531 struct proc *p = td->td_proc;
1532 vm_map_t map = &p->p_vmspace->vm_map;
1533 vm_map_entry_t entry;
1534 vm_object_t backing_object, object;
1535 boolean_t ignore_entry;
1537 vm_map_lock_read(map);
1538 for (entry = map->header.next; entry != &map->header;
1539 entry = entry->next) {
1541 * Don't dump inaccessible mappings, deal with legacy
1544 * Note that read-only segments related to the elf binary
1545 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1546 * need to arbitrarily ignore such segments.
1548 if (elf_legacy_coredump) {
1549 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1552 if ((entry->protection & VM_PROT_ALL) == 0)
1557 * Dont include memory segment in the coredump if
1558 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1559 * madvise(2). Do not dump submaps (i.e. parts of the
1562 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1565 if ((object = entry->object.vm_object) == NULL)
1568 /* Ignore memory-mapped devices and such things. */
1569 VM_OBJECT_RLOCK(object);
1570 while ((backing_object = object->backing_object) != NULL) {
1571 VM_OBJECT_RLOCK(backing_object);
1572 VM_OBJECT_RUNLOCK(object);
1573 object = backing_object;
1575 ignore_entry = object->type != OBJT_DEFAULT &&
1576 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1577 object->type != OBJT_PHYS;
1578 VM_OBJECT_RUNLOCK(object);
1582 (*func)(entry, closure);
1584 vm_map_unlock_read(map);
1588 * Write the core file header to the file, including padding up to
1589 * the page boundary.
1592 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1593 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1595 struct note_info *ninfo;
1599 /* Fill in the header. */
1600 bzero(hdr, hdrsize);
1601 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1603 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1604 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1605 sbuf_start_section(sb, NULL);
1606 sbuf_bcat(sb, hdr, hdrsize);
1607 TAILQ_FOREACH(ninfo, notelst, link)
1608 __elfN(putnote)(ninfo, sb);
1609 /* Align up to a page boundary for the program segments. */
1610 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1611 error = sbuf_finish(sb);
1618 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1628 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1631 * To have the debugger select the right thread (LWP) as the initial
1632 * thread, we dump the state of the thread passed to us in td first.
1633 * This is the thread that causes the core dump and thus likely to
1634 * be the right thread one wants to have selected in the debugger.
1637 while (thr != NULL) {
1638 size += register_note(list, NT_PRSTATUS,
1639 __elfN(note_prstatus), thr);
1640 size += register_note(list, NT_FPREGSET,
1641 __elfN(note_fpregset), thr);
1642 size += register_note(list, NT_THRMISC,
1643 __elfN(note_thrmisc), thr);
1644 size += register_note(list, NT_PTLWPINFO,
1645 __elfN(note_ptlwpinfo), thr);
1646 size += register_note(list, -1,
1647 __elfN(note_threadmd), thr);
1649 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1650 TAILQ_NEXT(thr, td_plist);
1652 thr = TAILQ_NEXT(thr, td_plist);
1655 size += register_note(list, NT_PROCSTAT_PROC,
1656 __elfN(note_procstat_proc), p);
1657 size += register_note(list, NT_PROCSTAT_FILES,
1658 note_procstat_files, p);
1659 size += register_note(list, NT_PROCSTAT_VMMAP,
1660 note_procstat_vmmap, p);
1661 size += register_note(list, NT_PROCSTAT_GROUPS,
1662 note_procstat_groups, p);
1663 size += register_note(list, NT_PROCSTAT_UMASK,
1664 note_procstat_umask, p);
1665 size += register_note(list, NT_PROCSTAT_RLIMIT,
1666 note_procstat_rlimit, p);
1667 size += register_note(list, NT_PROCSTAT_OSREL,
1668 note_procstat_osrel, p);
1669 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1670 __elfN(note_procstat_psstrings), p);
1671 size += register_note(list, NT_PROCSTAT_AUXV,
1672 __elfN(note_procstat_auxv), p);
1678 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1684 struct phdr_closure phc;
1686 ehdr = (Elf_Ehdr *)hdr;
1688 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1689 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1690 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1691 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1692 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1693 ehdr->e_ident[EI_DATA] = ELF_DATA;
1694 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1695 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1696 ehdr->e_ident[EI_ABIVERSION] = 0;
1697 ehdr->e_ident[EI_PAD] = 0;
1698 ehdr->e_type = ET_CORE;
1699 ehdr->e_machine = td->td_proc->p_elf_machine;
1700 ehdr->e_version = EV_CURRENT;
1702 ehdr->e_phoff = sizeof(Elf_Ehdr);
1703 ehdr->e_flags = td->td_proc->p_elf_flags;
1704 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1705 ehdr->e_phentsize = sizeof(Elf_Phdr);
1706 ehdr->e_shentsize = sizeof(Elf_Shdr);
1707 ehdr->e_shstrndx = SHN_UNDEF;
1708 if (numsegs + 1 < PN_XNUM) {
1709 ehdr->e_phnum = numsegs + 1;
1712 ehdr->e_phnum = PN_XNUM;
1715 ehdr->e_shoff = ehdr->e_phoff +
1716 (numsegs + 1) * ehdr->e_phentsize;
1717 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1718 ("e_shoff: %zu, hdrsize - shdr: %zu",
1719 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1721 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1722 memset(shdr, 0, sizeof(*shdr));
1724 * A special first section is used to hold large segment and
1725 * section counts. This was proposed by Sun Microsystems in
1726 * Solaris and has been adopted by Linux; the standard ELF
1727 * tools are already familiar with the technique.
1729 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1730 * (or 12-7 depending on the version of the document) for more
1733 shdr->sh_type = SHT_NULL;
1734 shdr->sh_size = ehdr->e_shnum;
1735 shdr->sh_link = ehdr->e_shstrndx;
1736 shdr->sh_info = numsegs + 1;
1740 * Fill in the program header entries.
1742 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1744 /* The note segement. */
1745 phdr->p_type = PT_NOTE;
1746 phdr->p_offset = hdrsize;
1749 phdr->p_filesz = notesz;
1751 phdr->p_flags = PF_R;
1752 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1755 /* All the writable segments from the program. */
1757 phc.offset = round_page(hdrsize + notesz);
1758 each_dumpable_segment(td, cb_put_phdr, &phc);
1762 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1764 struct note_info *ninfo;
1765 size_t size, notesize;
1768 out(arg, NULL, &size);
1769 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1771 ninfo->outfunc = out;
1772 ninfo->outarg = arg;
1773 ninfo->outsize = size;
1774 TAILQ_INSERT_TAIL(list, ninfo, link);
1779 notesize = sizeof(Elf_Note) + /* note header */
1780 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1782 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1788 append_note_data(const void *src, void *dst, size_t len)
1792 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1794 bcopy(src, dst, len);
1795 bzero((char *)dst + len, padded_len - len);
1797 return (padded_len);
1801 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1809 note = (Elf_Note *)buf;
1810 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1811 note->n_descsz = size;
1812 note->n_type = type;
1813 buf += sizeof(*note);
1814 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1815 sizeof(FREEBSD_ABI_VENDOR));
1816 append_note_data(src, buf, size);
1821 notesize = sizeof(Elf_Note) + /* note header */
1822 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1824 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1830 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1833 ssize_t old_len, sect_len;
1834 size_t new_len, descsz, i;
1836 if (ninfo->type == -1) {
1837 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1841 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1842 note.n_descsz = ninfo->outsize;
1843 note.n_type = ninfo->type;
1845 sbuf_bcat(sb, ¬e, sizeof(note));
1846 sbuf_start_section(sb, &old_len);
1847 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1848 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1849 if (note.n_descsz == 0)
1851 sbuf_start_section(sb, &old_len);
1852 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1853 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1857 new_len = (size_t)sect_len;
1858 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1859 if (new_len < descsz) {
1861 * It is expected that individual note emitters will correctly
1862 * predict their expected output size and fill up to that size
1863 * themselves, padding in a format-specific way if needed.
1864 * However, in case they don't, just do it here with zeros.
1866 for (i = 0; i < descsz - new_len; i++)
1868 } else if (new_len > descsz) {
1870 * We can't always truncate sb -- we may have drained some
1873 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1874 "read it (%zu > %zu). Since it is longer than "
1875 "expected, this coredump's notes are corrupt. THIS "
1876 "IS A BUG in the note_procstat routine for type %u.\n",
1877 __func__, (unsigned)note.n_type, new_len, descsz,
1878 (unsigned)note.n_type));
1883 * Miscellaneous note out functions.
1886 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1887 #include <compat/freebsd32/freebsd32.h>
1888 #include <compat/freebsd32/freebsd32_signal.h>
1890 typedef struct prstatus32 elf_prstatus_t;
1891 typedef struct prpsinfo32 elf_prpsinfo_t;
1892 typedef struct fpreg32 elf_prfpregset_t;
1893 typedef struct fpreg32 elf_fpregset_t;
1894 typedef struct reg32 elf_gregset_t;
1895 typedef struct thrmisc32 elf_thrmisc_t;
1896 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1897 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1898 typedef uint32_t elf_ps_strings_t;
1900 typedef prstatus_t elf_prstatus_t;
1901 typedef prpsinfo_t elf_prpsinfo_t;
1902 typedef prfpregset_t elf_prfpregset_t;
1903 typedef prfpregset_t elf_fpregset_t;
1904 typedef gregset_t elf_gregset_t;
1905 typedef thrmisc_t elf_thrmisc_t;
1906 #define ELF_KERN_PROC_MASK 0
1907 typedef struct kinfo_proc elf_kinfo_proc_t;
1908 typedef vm_offset_t elf_ps_strings_t;
1912 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1918 elf_prpsinfo_t *psinfo;
1921 p = (struct proc *)arg;
1923 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1924 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1925 psinfo->pr_version = PRPSINFO_VERSION;
1926 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1927 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1929 if (p->p_args != NULL) {
1930 len = sizeof(psinfo->pr_psargs) - 1;
1931 if (len > p->p_args->ar_length)
1932 len = p->p_args->ar_length;
1933 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1939 sbuf_new(&sbarg, psinfo->pr_psargs,
1940 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1941 error = proc_getargv(curthread, p, &sbarg);
1943 if (sbuf_finish(&sbarg) == 0)
1944 len = sbuf_len(&sbarg) - 1;
1946 len = sizeof(psinfo->pr_psargs) - 1;
1947 sbuf_delete(&sbarg);
1949 if (error || len == 0)
1950 strlcpy(psinfo->pr_psargs, p->p_comm,
1951 sizeof(psinfo->pr_psargs));
1953 KASSERT(len < sizeof(psinfo->pr_psargs),
1954 ("len is too long: %zu vs %zu", len,
1955 sizeof(psinfo->pr_psargs)));
1956 cp = psinfo->pr_psargs;
1959 cp = memchr(cp, '\0', end - cp);
1965 psinfo->pr_pid = p->p_pid;
1966 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1967 free(psinfo, M_TEMP);
1969 *sizep = sizeof(*psinfo);
1973 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1976 elf_prstatus_t *status;
1978 td = (struct thread *)arg;
1980 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1981 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1982 status->pr_version = PRSTATUS_VERSION;
1983 status->pr_statussz = sizeof(elf_prstatus_t);
1984 status->pr_gregsetsz = sizeof(elf_gregset_t);
1985 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1986 status->pr_osreldate = osreldate;
1987 status->pr_cursig = td->td_proc->p_sig;
1988 status->pr_pid = td->td_tid;
1989 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1990 fill_regs32(td, &status->pr_reg);
1992 fill_regs(td, &status->pr_reg);
1994 sbuf_bcat(sb, status, sizeof(*status));
1995 free(status, M_TEMP);
1997 *sizep = sizeof(*status);
2001 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2004 elf_prfpregset_t *fpregset;
2006 td = (struct thread *)arg;
2008 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2009 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2010 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2011 fill_fpregs32(td, fpregset);
2013 fill_fpregs(td, fpregset);
2015 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2016 free(fpregset, M_TEMP);
2018 *sizep = sizeof(*fpregset);
2022 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2025 elf_thrmisc_t thrmisc;
2027 td = (struct thread *)arg;
2029 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2030 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2031 strcpy(thrmisc.pr_tname, td->td_name);
2032 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2034 *sizep = sizeof(thrmisc);
2038 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2043 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2044 struct ptrace_lwpinfo32 pl;
2046 struct ptrace_lwpinfo pl;
2049 td = (struct thread *)arg;
2050 size = sizeof(structsize) + sizeof(pl);
2052 KASSERT(*sizep == size, ("invalid size"));
2053 structsize = sizeof(pl);
2054 sbuf_bcat(sb, &structsize, sizeof(structsize));
2055 bzero(&pl, sizeof(pl));
2056 pl.pl_lwpid = td->td_tid;
2057 pl.pl_event = PL_EVENT_NONE;
2058 pl.pl_sigmask = td->td_sigmask;
2059 pl.pl_siglist = td->td_siglist;
2060 if (td->td_si.si_signo != 0) {
2061 pl.pl_event = PL_EVENT_SIGNAL;
2062 pl.pl_flags |= PL_FLAG_SI;
2063 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2064 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2066 pl.pl_siginfo = td->td_si;
2069 strcpy(pl.pl_tdname, td->td_name);
2070 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2071 sbuf_bcat(sb, &pl, sizeof(pl));
2077 * Allow for MD specific notes, as well as any MD
2078 * specific preparations for writing MI notes.
2081 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2087 td = (struct thread *)arg;
2089 if (size != 0 && sb != NULL)
2090 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2094 __elfN(dump_thread)(td, buf, &size);
2095 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2096 if (size != 0 && sb != NULL)
2097 sbuf_bcat(sb, buf, size);
2102 #ifdef KINFO_PROC_SIZE
2103 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2107 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2113 p = (struct proc *)arg;
2114 size = sizeof(structsize) + p->p_numthreads *
2115 sizeof(elf_kinfo_proc_t);
2118 KASSERT(*sizep == size, ("invalid size"));
2119 structsize = sizeof(elf_kinfo_proc_t);
2120 sbuf_bcat(sb, &structsize, sizeof(structsize));
2121 sx_slock(&proctree_lock);
2123 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2124 sx_sunlock(&proctree_lock);
2129 #ifdef KINFO_FILE_SIZE
2130 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2134 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2137 size_t size, sect_sz, i;
2138 ssize_t start_len, sect_len;
2139 int structsize, filedesc_flags;
2141 if (coredump_pack_fileinfo)
2142 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2146 p = (struct proc *)arg;
2147 structsize = sizeof(struct kinfo_file);
2150 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2151 sbuf_set_drain(sb, sbuf_drain_count, &size);
2152 sbuf_bcat(sb, &structsize, sizeof(structsize));
2154 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2159 sbuf_start_section(sb, &start_len);
2161 sbuf_bcat(sb, &structsize, sizeof(structsize));
2163 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2166 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2171 KASSERT(sect_sz <= *sizep,
2172 ("kern_proc_filedesc_out did not respect maxlen; "
2173 "requested %zu, got %zu", *sizep - sizeof(structsize),
2174 sect_sz - sizeof(structsize)));
2176 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2181 #ifdef KINFO_VMENTRY_SIZE
2182 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2186 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2190 int structsize, vmmap_flags;
2192 if (coredump_pack_vmmapinfo)
2193 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2197 p = (struct proc *)arg;
2198 structsize = sizeof(struct kinfo_vmentry);
2201 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2202 sbuf_set_drain(sb, sbuf_drain_count, &size);
2203 sbuf_bcat(sb, &structsize, sizeof(structsize));
2205 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2210 sbuf_bcat(sb, &structsize, sizeof(structsize));
2212 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2218 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2224 p = (struct proc *)arg;
2225 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2227 KASSERT(*sizep == size, ("invalid size"));
2228 structsize = sizeof(gid_t);
2229 sbuf_bcat(sb, &structsize, sizeof(structsize));
2230 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2237 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2243 p = (struct proc *)arg;
2244 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2246 KASSERT(*sizep == size, ("invalid size"));
2247 structsize = sizeof(p->p_fd->fd_cmask);
2248 sbuf_bcat(sb, &structsize, sizeof(structsize));
2249 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2255 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2258 struct rlimit rlim[RLIM_NLIMITS];
2262 p = (struct proc *)arg;
2263 size = sizeof(structsize) + sizeof(rlim);
2265 KASSERT(*sizep == size, ("invalid size"));
2266 structsize = sizeof(rlim);
2267 sbuf_bcat(sb, &structsize, sizeof(structsize));
2269 for (i = 0; i < RLIM_NLIMITS; i++)
2270 lim_rlimit_proc(p, i, &rlim[i]);
2272 sbuf_bcat(sb, rlim, sizeof(rlim));
2278 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2284 p = (struct proc *)arg;
2285 size = sizeof(structsize) + sizeof(p->p_osrel);
2287 KASSERT(*sizep == size, ("invalid size"));
2288 structsize = sizeof(p->p_osrel);
2289 sbuf_bcat(sb, &structsize, sizeof(structsize));
2290 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2296 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2299 elf_ps_strings_t ps_strings;
2303 p = (struct proc *)arg;
2304 size = sizeof(structsize) + sizeof(ps_strings);
2306 KASSERT(*sizep == size, ("invalid size"));
2307 structsize = sizeof(ps_strings);
2308 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2309 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2311 ps_strings = p->p_sysent->sv_psstrings;
2313 sbuf_bcat(sb, &structsize, sizeof(structsize));
2314 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2320 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2326 p = (struct proc *)arg;
2329 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2330 sbuf_set_drain(sb, sbuf_drain_count, &size);
2331 sbuf_bcat(sb, &structsize, sizeof(structsize));
2333 proc_getauxv(curthread, p, sb);
2339 structsize = sizeof(Elf_Auxinfo);
2340 sbuf_bcat(sb, &structsize, sizeof(structsize));
2342 proc_getauxv(curthread, p, sb);
2348 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2349 const char *note_vendor, const Elf_Phdr *pnote,
2350 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2352 const Elf_Note *note, *note0, *note_end;
2353 const char *note_name;
2358 /* We need some limit, might as well use PAGE_SIZE. */
2359 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2361 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2362 if (pnote->p_offset > PAGE_SIZE ||
2363 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2364 VOP_UNLOCK(imgp->vp, 0);
2365 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2366 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2367 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2368 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2369 curthread->td_ucred, NOCRED, NULL, curthread);
2371 uprintf("i/o error PT_NOTE\n");
2374 note = note0 = (const Elf_Note *)buf;
2375 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2377 note = note0 = (const Elf_Note *)(imgp->image_header +
2379 note_end = (const Elf_Note *)(imgp->image_header +
2380 pnote->p_offset + pnote->p_filesz);
2383 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2384 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2385 (const char *)note < sizeof(Elf_Note)) {
2388 if (note->n_namesz != checknote->n_namesz ||
2389 note->n_descsz != checknote->n_descsz ||
2390 note->n_type != checknote->n_type)
2392 note_name = (const char *)(note + 1);
2393 if (note_name + checknote->n_namesz >=
2394 (const char *)note_end || strncmp(note_vendor,
2395 note_name, checknote->n_namesz) != 0)
2398 if (cb(note, cb_arg, &res))
2401 note = (const Elf_Note *)((const char *)(note + 1) +
2402 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2403 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2412 struct brandnote_cb_arg {
2413 Elf_Brandnote *brandnote;
2418 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2420 struct brandnote_cb_arg *arg;
2425 * Fetch the osreldate for binary from the ELF OSABI-note if
2428 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2429 arg->brandnote->trans_osrel != NULL ?
2430 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2435 static Elf_Note fctl_note = {
2436 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2437 .n_descsz = sizeof(uint32_t),
2438 .n_type = NT_FREEBSD_FEATURE_CTL,
2441 struct fctl_cb_arg {
2446 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2448 struct fctl_cb_arg *arg;
2449 const Elf32_Word *desc;
2453 p = (uintptr_t)(note + 1);
2454 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2455 desc = (const Elf32_Word *)p;
2456 *arg->fctl0 = desc[0];
2461 * Try to find the appropriate ABI-note section for checknote, fetch
2462 * the osreldate and feature control flags for binary from the ELF
2463 * OSABI-note. Only the first page of the image is searched, the same
2467 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2468 int32_t *osrel, uint32_t *fctl0)
2470 const Elf_Phdr *phdr;
2471 const Elf_Ehdr *hdr;
2472 struct brandnote_cb_arg b_arg;
2473 struct fctl_cb_arg f_arg;
2476 hdr = (const Elf_Ehdr *)imgp->image_header;
2477 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2478 b_arg.brandnote = brandnote;
2479 b_arg.osrel = osrel;
2480 f_arg.fctl0 = fctl0;
2482 for (i = 0; i < hdr->e_phnum; i++) {
2483 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2484 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2486 for (j = 0; j < hdr->e_phnum; j++) {
2487 if (phdr[j].p_type == PT_NOTE &&
2488 __elfN(parse_notes)(imgp, &fctl_note,
2489 FREEBSD_ABI_VENDOR, &phdr[j],
2490 note_fctl_cb, &f_arg))
2501 * Tell kern_execve.c about it, with a little help from the linker.
2503 static struct execsw __elfN(execsw) = {
2504 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2505 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2507 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2510 __elfN(trans_prot)(Elf_Word flags)
2516 prot |= VM_PROT_EXECUTE;
2518 prot |= VM_PROT_WRITE;
2520 prot |= VM_PROT_READ;
2521 #if __ELF_WORD_SIZE == 32
2522 #if defined(__amd64__)
2523 if (i386_read_exec && (flags & PF_R))
2524 prot |= VM_PROT_EXECUTE;
2531 __elfN(untrans_prot)(vm_prot_t prot)
2536 if (prot & VM_PROT_EXECUTE)
2538 if (prot & VM_PROT_READ)
2540 if (prot & VM_PROT_WRITE)