2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
59 #include <sys/sf_buf.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
84 #include <machine/acle-compat.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_offset_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, "");
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136 "enable execution from readable segments");
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
143 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
150 .hdr.n_descsz = sizeof(int32_t),
152 .vendor = FREEBSD_ABI_VENDOR,
153 .flags = BN_TRANSLATE_OSREL,
154 .trans_osrel = __elfN(freebsd_trans_osrel)
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 p = (uintptr_t)(note + 1);
163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 *osrel = *(const int32_t *)(p);
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
174 .hdr.n_descsz = 16, /* XXX at least 16 */
176 .vendor = GNU_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = kfreebsd_trans_osrel
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 const Elf32_Word *desc;
187 p = (uintptr_t)(note + 1);
188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 desc = (const Elf32_Word *)p;
191 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 for (i = 0; i < MAX_BRANDS; i++) {
209 if (elf_brand_list[i] == NULL) {
210 elf_brand_list[i] = entry;
214 if (i == MAX_BRANDS) {
215 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 for (i = 0; i < MAX_BRANDS; i++) {
228 if (elf_brand_list[i] == entry) {
229 elf_brand_list[i] = NULL;
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
244 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
246 if (p->p_sysent == entry->sysvec) {
251 sx_sunlock(&allproc_lock);
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258 int interp_name_len, int32_t *osrel)
260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
266 * We support four types of branding -- (1) the ELF EI_OSABI field
267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 * branding w/in the ELF header, (3) path of the `interp_path'
269 * field, and (4) the ".note.ABI-tag" ELF section.
272 /* Look for an ".note.ABI-tag" ELF section */
273 for (i = 0; i < MAX_BRANDS; i++) {
274 bi = elf_brand_list[i];
277 if (hdr->e_machine == bi->machine && (bi->flags &
278 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
279 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
280 /* Give brand a chance to veto check_note's guess */
281 if (ret && bi->header_supported)
282 ret = bi->header_supported(imgp);
288 /* If the executable has a brand, search for it in the brand list. */
289 for (i = 0; i < MAX_BRANDS; i++) {
290 bi = elf_brand_list[i];
291 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
293 if (hdr->e_machine == bi->machine &&
294 (hdr->e_ident[EI_OSABI] == bi->brand ||
295 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
296 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
297 /* Looks good, but give brand a chance to veto */
298 if (!bi->header_supported || bi->header_supported(imgp))
303 /* No known brand, see if the header is recognized by any brand */
304 for (i = 0; i < MAX_BRANDS; i++) {
305 bi = elf_brand_list[i];
306 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
307 bi->header_supported == NULL)
309 if (hdr->e_machine == bi->machine) {
310 ret = bi->header_supported(imgp);
316 /* Lacking a known brand, search for a recognized interpreter. */
317 if (interp != NULL) {
318 for (i = 0; i < MAX_BRANDS; i++) {
319 bi = elf_brand_list[i];
320 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
322 if (hdr->e_machine == bi->machine &&
323 /* ELF image p_filesz includes terminating zero */
324 strlen(bi->interp_path) + 1 == interp_name_len &&
325 strncmp(interp, bi->interp_path, interp_name_len)
331 /* Lacking a recognized interpreter, try the default brand */
332 for (i = 0; i < MAX_BRANDS; i++) {
333 bi = elf_brand_list[i];
334 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
336 if (hdr->e_machine == bi->machine &&
337 __elfN(fallback_brand) == bi->brand)
344 __elfN(check_header)(const Elf_Ehdr *hdr)
350 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
351 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
352 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
353 hdr->e_phentsize != sizeof(Elf_Phdr) ||
354 hdr->e_version != ELF_TARG_VER)
358 * Make sure we have at least one brand for this machine.
361 for (i = 0; i < MAX_BRANDS; i++) {
362 bi = elf_brand_list[i];
363 if (bi != NULL && bi->machine == hdr->e_machine)
373 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
374 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
381 * Create the page if it doesn't exist yet. Ignore errors.
384 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
385 VM_PROT_ALL, VM_PROT_ALL, 0);
389 * Find the page from the underlying object.
392 sf = vm_imgact_map_page(object, offset);
394 return (KERN_FAILURE);
395 off = offset - trunc_page(offset);
396 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
398 vm_imgact_unmap_page(sf);
400 return (KERN_FAILURE);
404 return (KERN_SUCCESS);
408 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
409 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
416 if (start != trunc_page(start)) {
417 rv = __elfN(map_partial)(map, object, offset, start,
418 round_page(start), prot);
421 offset += round_page(start) - start;
422 start = round_page(start);
424 if (end != round_page(end)) {
425 rv = __elfN(map_partial)(map, object, offset +
426 trunc_page(end) - start, trunc_page(end), end, prot);
429 end = trunc_page(end);
432 if (offset & PAGE_MASK) {
434 * The mapping is not page aligned. This means we have
435 * to copy the data. Sigh.
437 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
438 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
443 return (KERN_SUCCESS);
444 for (; start < end; start += sz) {
445 sf = vm_imgact_map_page(object, offset);
447 return (KERN_FAILURE);
448 off = offset - trunc_page(offset);
450 if (sz > PAGE_SIZE - off)
451 sz = PAGE_SIZE - off;
452 error = copyout((caddr_t)sf_buf_kva(sf) + off,
454 vm_imgact_unmap_page(sf);
456 return (KERN_FAILURE);
462 vm_object_reference(object);
464 rv = vm_map_insert(map, object, offset, start, end,
465 prot, VM_PROT_ALL, cow);
467 if (rv != KERN_SUCCESS)
468 vm_object_deallocate(object);
472 return (KERN_SUCCESS);
477 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
478 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
485 vm_offset_t map_addr;
488 vm_offset_t file_addr;
491 * It's necessary to fail if the filsz + offset taken from the
492 * header is greater than the actual file pager object's size.
493 * If we were to allow this, then the vm_map_find() below would
494 * walk right off the end of the file object and into the ether.
496 * While I'm here, might as well check for something else that
497 * is invalid: filsz cannot be greater than memsz.
499 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
500 uprintf("elf_load_section: truncated ELF file\n");
504 object = imgp->object;
505 map = &imgp->proc->p_vmspace->vm_map;
506 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
507 file_addr = trunc_page_ps(offset, pagesize);
510 * We have two choices. We can either clear the data in the last page
511 * of an oversized mapping, or we can start the anon mapping a page
512 * early and copy the initialized data into that first page. We
513 * choose the second..
516 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
518 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
521 /* cow flags: don't dump readonly sections in core */
522 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
523 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
525 rv = __elfN(map_insert)(map,
527 file_addr, /* file offset */
528 map_addr, /* virtual start */
529 map_addr + map_len,/* virtual end */
532 if (rv != KERN_SUCCESS)
535 /* we can stop now if we've covered it all */
536 if (memsz == filsz) {
543 * We have to get the remaining bit of the file into the first part
544 * of the oversized map segment. This is normally because the .data
545 * segment in the file is extended to provide bss. It's a neat idea
546 * to try and save a page, but it's a pain in the behind to implement.
548 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
549 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
550 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
553 /* This had damn well better be true! */
555 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
556 map_len, VM_PROT_ALL, 0);
557 if (rv != KERN_SUCCESS) {
565 sf = vm_imgact_map_page(object, offset + filsz);
569 /* send the page fragment to user space */
570 off = trunc_page_ps(offset + filsz, pagesize) -
571 trunc_page(offset + filsz);
572 error = copyout((caddr_t)sf_buf_kva(sf) + off,
573 (caddr_t)map_addr, copy_len);
574 vm_imgact_unmap_page(sf);
581 * set it to the specified protection.
582 * XXX had better undo the damage from pasting over the cracks here!
584 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
585 map_len), prot, FALSE);
591 * Load the file "file" into memory. It may be either a shared object
594 * The "addr" reference parameter is in/out. On entry, it specifies
595 * the address where a shared object should be loaded. If the file is
596 * an executable, this value is ignored. On exit, "addr" specifies
597 * where the file was actually loaded.
599 * The "entry" reference parameter is out only. On exit, it specifies
600 * the entry point for the loaded file.
603 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
604 u_long *entry, size_t pagesize)
609 struct image_params image_params;
611 const Elf_Ehdr *hdr = NULL;
612 const Elf_Phdr *phdr = NULL;
613 struct nameidata *nd;
615 struct image_params *imgp;
618 u_long base_addr = 0;
619 int error, i, numsegs;
621 #ifdef CAPABILITY_MODE
623 * XXXJA: This check can go away once we are sufficiently confident
624 * that the checks in namei() are correct.
626 if (IN_CAPABILITY_MODE(curthread))
630 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
632 attr = &tempdata->attr;
633 imgp = &tempdata->image_params;
636 * Initialize part of the common data
640 imgp->firstpage = NULL;
641 imgp->image_header = NULL;
643 imgp->execlabel = NULL;
645 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
646 if ((error = namei(nd)) != 0) {
650 NDFREE(nd, NDF_ONLY_PNBUF);
651 imgp->vp = nd->ni_vp;
654 * Check permissions, modes, uid, etc on the file, and "open" it.
656 error = exec_check_permissions(imgp);
660 error = exec_map_first_page(imgp);
665 * Also make certain that the interpreter stays the same, so set
666 * its VV_TEXT flag, too.
668 VOP_SET_TEXT(nd->ni_vp);
670 imgp->object = nd->ni_vp->v_object;
672 hdr = (const Elf_Ehdr *)imgp->image_header;
673 if ((error = __elfN(check_header)(hdr)) != 0)
675 if (hdr->e_type == ET_DYN)
677 else if (hdr->e_type == ET_EXEC)
684 /* Only support headers that fit within first page for now */
685 if ((hdr->e_phoff > PAGE_SIZE) ||
686 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
691 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
692 if (!aligned(phdr, Elf_Addr)) {
697 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
698 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
699 /* Loadable segment */
700 prot = __elfN(trans_prot)(phdr[i].p_flags);
701 error = __elfN(load_section)(imgp, phdr[i].p_offset,
702 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
703 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
707 * Establish the base address if this is the
711 base_addr = trunc_page(phdr[i].p_vaddr +
717 *entry = (unsigned long)hdr->e_entry + rbase;
721 exec_unmap_first_page(imgp);
726 free(tempdata, M_TEMP);
732 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
736 const Elf_Phdr *phdr;
737 Elf_Auxargs *elf_auxargs;
738 struct vmspace *vmspace;
739 const char *err_str, *newinterp;
740 char *interp, *interp_buf, *path;
741 Elf_Brandinfo *brand_info;
742 struct sysentvec *sv;
744 u_long text_size, data_size, total_size, text_addr, data_addr;
745 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
747 int error, i, n, interp_name_len, have_interp;
749 hdr = (const Elf_Ehdr *)imgp->image_header;
752 * Do we have a valid ELF header ?
754 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
755 * if particular brand doesn't support it.
757 if (__elfN(check_header)(hdr) != 0 ||
758 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
762 * From here on down, we return an errno, not -1, as we've
763 * detected an ELF file.
766 if ((hdr->e_phoff > PAGE_SIZE) ||
767 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
768 /* Only support headers in first page for now */
769 uprintf("Program headers not in the first page\n");
772 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
773 if (!aligned(phdr, Elf_Addr)) {
774 uprintf("Unaligned program headers\n");
781 text_size = data_size = total_size = text_addr = data_addr = 0;
784 err_str = newinterp = NULL;
785 interp = interp_buf = NULL;
788 for (i = 0; i < hdr->e_phnum; i++) {
789 switch (phdr[i].p_type) {
792 baddr = phdr[i].p_vaddr;
796 /* Path to interpreter */
797 if (phdr[i].p_filesz > MAXPATHLEN) {
798 uprintf("Invalid PT_INTERP\n");
802 interp_name_len = phdr[i].p_filesz;
803 if (phdr[i].p_offset > PAGE_SIZE ||
804 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
805 VOP_UNLOCK(imgp->vp, 0);
806 interp_buf = malloc(interp_name_len + 1, M_TEMP,
808 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
809 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
810 interp_name_len, phdr[i].p_offset,
811 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
814 uprintf("i/o error PT_INTERP\n");
817 interp_buf[interp_name_len] = '\0';
820 interp = __DECONST(char *, imgp->image_header) +
827 __elfN(trans_prot)(phdr[i].p_flags);
828 imgp->stack_sz = phdr[i].p_memsz;
833 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
835 if (brand_info == NULL) {
836 uprintf("ELF binary type \"%u\" not known.\n",
837 hdr->e_ident[EI_OSABI]);
841 if (hdr->e_type == ET_DYN) {
842 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
843 uprintf("Cannot execute shared object\n");
848 * Honour the base load address from the dso if it is
849 * non-zero for some reason.
852 et_dyn_addr = ET_DYN_LOAD_ADDR;
857 sv = brand_info->sysvec;
858 if (interp != NULL && brand_info->interp_newpath != NULL)
859 newinterp = brand_info->interp_newpath;
862 * Avoid a possible deadlock if the current address space is destroyed
863 * and that address space maps the locked vnode. In the common case,
864 * the locked vnode's v_usecount is decremented but remains greater
865 * than zero. Consequently, the vnode lock is not needed by vrele().
866 * However, in cases where the vnode lock is external, such as nullfs,
867 * v_usecount may become zero.
869 * The VV_TEXT flag prevents modifications to the executable while
870 * the vnode is unlocked.
872 VOP_UNLOCK(imgp->vp, 0);
874 error = exec_new_vmspace(imgp, sv);
875 imgp->proc->p_sysent = sv;
877 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
881 for (i = 0; i < hdr->e_phnum; i++) {
882 switch (phdr[i].p_type) {
883 case PT_LOAD: /* Loadable segment */
884 if (phdr[i].p_memsz == 0)
886 prot = __elfN(trans_prot)(phdr[i].p_flags);
887 error = __elfN(load_section)(imgp, phdr[i].p_offset,
888 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
889 phdr[i].p_memsz, phdr[i].p_filesz, prot,
895 * If this segment contains the program headers,
896 * remember their virtual address for the AT_PHDR
897 * aux entry. Static binaries don't usually include
900 if (phdr[i].p_offset == 0 &&
901 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
903 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
906 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
907 seg_size = round_page(phdr[i].p_memsz +
908 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
911 * Make the largest executable segment the official
912 * text segment and all others data.
914 * Note that obreak() assumes that data_addr +
915 * data_size == end of data load area, and the ELF
916 * file format expects segments to be sorted by
917 * address. If multiple data segments exist, the
918 * last one will be used.
921 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
922 text_size = seg_size;
923 text_addr = seg_addr;
925 data_size = seg_size;
926 data_addr = seg_addr;
928 total_size += seg_size;
930 case PT_PHDR: /* Program header table info */
931 proghdr = phdr[i].p_vaddr + et_dyn_addr;
938 if (data_addr == 0 && data_size == 0) {
939 data_addr = text_addr;
940 data_size = text_size;
943 entry = (u_long)hdr->e_entry + et_dyn_addr;
946 * Check limits. It should be safe to check the
947 * limits after loading the segments since we do
948 * not actually fault in all the segments pages.
950 PROC_LOCK(imgp->proc);
951 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
952 err_str = "Data segment size exceeds process limit";
953 else if (text_size > maxtsiz)
954 err_str = "Text segment size exceeds system limit";
955 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
956 err_str = "Total segment size exceeds process limit";
957 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
958 err_str = "Data segment size exceeds resource limit";
959 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
960 err_str = "Total segment size exceeds resource limit";
961 if (err_str != NULL) {
962 PROC_UNLOCK(imgp->proc);
963 uprintf("%s\n", err_str);
968 vmspace = imgp->proc->p_vmspace;
969 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
970 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
971 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
972 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
975 * We load the dynamic linker where a userland call
976 * to mmap(0, ...) would put it. The rationale behind this
977 * calculation is that it leaves room for the heap to grow to
978 * its maximum allowed size.
980 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
982 PROC_UNLOCK(imgp->proc);
984 imgp->entry_addr = entry;
986 if (interp != NULL) {
988 VOP_UNLOCK(imgp->vp, 0);
989 if (brand_info->emul_path != NULL &&
990 brand_info->emul_path[0] != '\0') {
991 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
992 snprintf(path, MAXPATHLEN, "%s%s",
993 brand_info->emul_path, interp);
994 error = __elfN(load_file)(imgp->proc, path, &addr,
995 &imgp->entry_addr, sv->sv_pagesize);
1000 if (!have_interp && newinterp != NULL) {
1001 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1002 &imgp->entry_addr, sv->sv_pagesize);
1007 error = __elfN(load_file)(imgp->proc, interp, &addr,
1008 &imgp->entry_addr, sv->sv_pagesize);
1010 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1012 uprintf("ELF interpreter %s not found\n", interp);
1019 * Construct auxargs table (used by the fixup routine)
1021 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1022 elf_auxargs->execfd = -1;
1023 elf_auxargs->phdr = proghdr;
1024 elf_auxargs->phent = hdr->e_phentsize;
1025 elf_auxargs->phnum = hdr->e_phnum;
1026 elf_auxargs->pagesz = PAGE_SIZE;
1027 elf_auxargs->base = addr;
1028 elf_auxargs->flags = 0;
1029 elf_auxargs->entry = entry;
1030 elf_auxargs->hdr_eflags = hdr->e_flags;
1032 imgp->auxargs = elf_auxargs;
1033 imgp->interpreted = 0;
1034 imgp->reloc_base = addr;
1035 imgp->proc->p_osrel = osrel;
1038 free(interp_buf, M_TEMP);
1042 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1045 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1047 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1051 base = (Elf_Addr *)*stack_base;
1052 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1054 if (args->execfd != -1)
1055 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1056 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1057 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1058 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1059 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1060 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1061 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1062 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1064 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1066 if (imgp->execpathp != 0)
1067 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1068 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1069 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1070 if (imgp->canary != 0) {
1071 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1072 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1074 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1075 if (imgp->pagesizes != 0) {
1076 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1077 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1079 if (imgp->sysent->sv_timekeep_base != 0) {
1080 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1081 imgp->sysent->sv_timekeep_base);
1083 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1084 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1085 imgp->sysent->sv_stackprot);
1086 AUXARGS_ENTRY(pos, AT_NULL, 0);
1088 free(imgp->auxargs, M_TEMP);
1089 imgp->auxargs = NULL;
1092 suword(base, (long)imgp->args->argc);
1093 *stack_base = (register_t *)base;
1098 * Code for generating ELF core dumps.
1101 typedef void (*segment_callback)(vm_map_entry_t, void *);
1103 /* Closure for cb_put_phdr(). */
1104 struct phdr_closure {
1105 Elf_Phdr *phdr; /* Program header to fill in */
1106 Elf_Off offset; /* Offset of segment in core file */
1109 /* Closure for cb_size_segment(). */
1110 struct sseg_closure {
1111 int count; /* Count of writable segments. */
1112 size_t size; /* Total size of all writable segments. */
1115 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1118 int type; /* Note type. */
1119 outfunc_t outfunc; /* Output function. */
1120 void *outarg; /* Argument for the output function. */
1121 size_t outsize; /* Output size. */
1122 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1125 TAILQ_HEAD(note_info_list, note_info);
1127 /* Coredump output parameters. */
1128 struct coredump_params {
1130 struct ucred *active_cred;
1131 struct ucred *file_cred;
1134 struct gzio_stream *gzs;
1137 static void cb_put_phdr(vm_map_entry_t, void *);
1138 static void cb_size_segment(vm_map_entry_t, void *);
1139 static int core_write(struct coredump_params *, void *, size_t, off_t,
1141 static void each_writable_segment(struct thread *, segment_callback, void *);
1142 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1143 struct note_info_list *, size_t);
1144 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1146 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1147 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1148 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1149 static int sbuf_drain_core_output(void *, const char *, int);
1150 static int sbuf_drain_count(void *arg, const char *data, int len);
1152 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1154 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1155 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1156 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1157 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1158 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1159 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1160 static void note_procstat_files(void *, struct sbuf *, size_t *);
1161 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1162 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1163 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1164 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1165 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1168 extern int compress_user_cores_gzlevel;
1171 * Write out a core segment to the compression stream.
1174 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1180 chunk_len = MIN(len, CORE_BUF_SIZE);
1181 copyin(base, buf, chunk_len);
1182 error = gzio_write(p->gzs, buf, chunk_len);
1192 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1195 return (core_write((struct coredump_params *)arg, base, len, offset,
1201 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1205 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1206 seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1207 p->active_cred, p->file_cred, NULL, p->td));
1211 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1217 return (compress_chunk(p, base, tmpbuf, len));
1219 return (core_write(p, base, len, offset, UIO_USERSPACE));
1223 * Drain into a core file.
1226 sbuf_drain_core_output(void *arg, const char *data, int len)
1228 struct coredump_params *p;
1231 p = (struct coredump_params *)arg;
1234 * Some kern_proc out routines that print to this sbuf may
1235 * call us with the process lock held. Draining with the
1236 * non-sleepable lock held is unsafe. The lock is needed for
1237 * those routines when dumping a live process. In our case we
1238 * can safely release the lock before draining and acquire
1241 locked = PROC_LOCKED(p->td->td_proc);
1243 PROC_UNLOCK(p->td->td_proc);
1246 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1249 error = core_write(p, __DECONST(void *, data), len, p->offset,
1252 PROC_LOCK(p->td->td_proc);
1260 * Drain into a counter.
1263 sbuf_drain_count(void *arg, const char *data __unused, int len)
1267 sizep = (size_t *)arg;
1273 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1275 struct ucred *cred = td->td_ucred;
1277 struct sseg_closure seginfo;
1278 struct note_info_list notelst;
1279 struct coredump_params params;
1280 struct note_info *ninfo;
1282 size_t hdrsize, notesz, coresize;
1286 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1290 TAILQ_INIT(¬elst);
1292 /* Size the program segments. */
1295 each_writable_segment(td, cb_size_segment, &seginfo);
1298 * Collect info about the core file header area.
1300 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1301 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1302 coresize = round_page(hdrsize + notesz) + seginfo.size;
1304 /* Set up core dump parameters. */
1306 params.active_cred = cred;
1307 params.file_cred = NOCRED;
1314 PROC_LOCK(td->td_proc);
1315 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1316 PROC_UNLOCK(td->td_proc);
1323 if (coresize >= limit) {
1329 /* Create a compression stream if necessary. */
1331 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1332 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1333 if (params.gzs == NULL) {
1337 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1342 * Allocate memory for building the header, fill it up,
1343 * and write it out following the notes.
1345 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1350 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1353 /* Write the contents of all of the writable segments. */
1359 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1360 offset = round_page(hdrsize + notesz);
1361 for (i = 0; i < seginfo.count; i++) {
1362 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1363 php->p_filesz, offset, ¶ms, tmpbuf);
1366 offset += php->p_filesz;
1370 if (error == 0 && compress)
1371 error = gzio_flush(params.gzs);
1376 "Failed to write core file for process %s (error %d)\n",
1377 curproc->p_comm, error);
1383 free(tmpbuf, M_TEMP);
1384 if (params.gzs != NULL)
1385 gzio_fini(params.gzs);
1388 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1389 TAILQ_REMOVE(¬elst, ninfo, link);
1390 free(ninfo, M_TEMP);
1399 * A callback for each_writable_segment() to write out the segment's
1400 * program header entry.
1403 cb_put_phdr(entry, closure)
1404 vm_map_entry_t entry;
1407 struct phdr_closure *phc = (struct phdr_closure *)closure;
1408 Elf_Phdr *phdr = phc->phdr;
1410 phc->offset = round_page(phc->offset);
1412 phdr->p_type = PT_LOAD;
1413 phdr->p_offset = phc->offset;
1414 phdr->p_vaddr = entry->start;
1416 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1417 phdr->p_align = PAGE_SIZE;
1418 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1420 phc->offset += phdr->p_filesz;
1425 * A callback for each_writable_segment() to gather information about
1426 * the number of segments and their total size.
1429 cb_size_segment(entry, closure)
1430 vm_map_entry_t entry;
1433 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1436 ssc->size += entry->end - entry->start;
1440 * For each writable segment in the process's memory map, call the given
1441 * function with a pointer to the map entry and some arbitrary
1442 * caller-supplied data.
1445 each_writable_segment(td, func, closure)
1447 segment_callback func;
1450 struct proc *p = td->td_proc;
1451 vm_map_t map = &p->p_vmspace->vm_map;
1452 vm_map_entry_t entry;
1453 vm_object_t backing_object, object;
1454 boolean_t ignore_entry;
1456 vm_map_lock_read(map);
1457 for (entry = map->header.next; entry != &map->header;
1458 entry = entry->next) {
1460 * Don't dump inaccessible mappings, deal with legacy
1463 * Note that read-only segments related to the elf binary
1464 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1465 * need to arbitrarily ignore such segments.
1467 if (elf_legacy_coredump) {
1468 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1471 if ((entry->protection & VM_PROT_ALL) == 0)
1476 * Dont include memory segment in the coredump if
1477 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1478 * madvise(2). Do not dump submaps (i.e. parts of the
1481 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1484 if ((object = entry->object.vm_object) == NULL)
1487 /* Ignore memory-mapped devices and such things. */
1488 VM_OBJECT_RLOCK(object);
1489 while ((backing_object = object->backing_object) != NULL) {
1490 VM_OBJECT_RLOCK(backing_object);
1491 VM_OBJECT_RUNLOCK(object);
1492 object = backing_object;
1494 ignore_entry = object->type != OBJT_DEFAULT &&
1495 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1496 object->type != OBJT_PHYS;
1497 VM_OBJECT_RUNLOCK(object);
1501 (*func)(entry, closure);
1503 vm_map_unlock_read(map);
1507 * Write the core file header to the file, including padding up to
1508 * the page boundary.
1511 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1512 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1514 struct note_info *ninfo;
1518 /* Fill in the header. */
1519 bzero(hdr, hdrsize);
1520 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1522 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1523 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1524 sbuf_start_section(sb, NULL);
1525 sbuf_bcat(sb, hdr, hdrsize);
1526 TAILQ_FOREACH(ninfo, notelst, link)
1527 __elfN(putnote)(ninfo, sb);
1528 /* Align up to a page boundary for the program segments. */
1529 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1530 error = sbuf_finish(sb);
1537 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1547 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1550 * To have the debugger select the right thread (LWP) as the initial
1551 * thread, we dump the state of the thread passed to us in td first.
1552 * This is the thread that causes the core dump and thus likely to
1553 * be the right thread one wants to have selected in the debugger.
1556 while (thr != NULL) {
1557 size += register_note(list, NT_PRSTATUS,
1558 __elfN(note_prstatus), thr);
1559 size += register_note(list, NT_FPREGSET,
1560 __elfN(note_fpregset), thr);
1561 size += register_note(list, NT_THRMISC,
1562 __elfN(note_thrmisc), thr);
1563 size += register_note(list, -1,
1564 __elfN(note_threadmd), thr);
1566 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1567 TAILQ_NEXT(thr, td_plist);
1569 thr = TAILQ_NEXT(thr, td_plist);
1572 size += register_note(list, NT_PROCSTAT_PROC,
1573 __elfN(note_procstat_proc), p);
1574 size += register_note(list, NT_PROCSTAT_FILES,
1575 note_procstat_files, p);
1576 size += register_note(list, NT_PROCSTAT_VMMAP,
1577 note_procstat_vmmap, p);
1578 size += register_note(list, NT_PROCSTAT_GROUPS,
1579 note_procstat_groups, p);
1580 size += register_note(list, NT_PROCSTAT_UMASK,
1581 note_procstat_umask, p);
1582 size += register_note(list, NT_PROCSTAT_RLIMIT,
1583 note_procstat_rlimit, p);
1584 size += register_note(list, NT_PROCSTAT_OSREL,
1585 note_procstat_osrel, p);
1586 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1587 __elfN(note_procstat_psstrings), p);
1588 size += register_note(list, NT_PROCSTAT_AUXV,
1589 __elfN(note_procstat_auxv), p);
1595 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1600 struct phdr_closure phc;
1602 ehdr = (Elf_Ehdr *)hdr;
1603 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1605 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1606 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1607 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1608 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1609 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1610 ehdr->e_ident[EI_DATA] = ELF_DATA;
1611 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1612 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1613 ehdr->e_ident[EI_ABIVERSION] = 0;
1614 ehdr->e_ident[EI_PAD] = 0;
1615 ehdr->e_type = ET_CORE;
1616 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1617 ehdr->e_machine = ELF_ARCH32;
1619 ehdr->e_machine = ELF_ARCH;
1621 ehdr->e_version = EV_CURRENT;
1623 ehdr->e_phoff = sizeof(Elf_Ehdr);
1625 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1626 ehdr->e_phentsize = sizeof(Elf_Phdr);
1627 ehdr->e_phnum = numsegs + 1;
1628 ehdr->e_shentsize = sizeof(Elf_Shdr);
1630 ehdr->e_shstrndx = SHN_UNDEF;
1633 * Fill in the program header entries.
1636 /* The note segement. */
1637 phdr->p_type = PT_NOTE;
1638 phdr->p_offset = hdrsize;
1641 phdr->p_filesz = notesz;
1643 phdr->p_flags = PF_R;
1644 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1647 /* All the writable segments from the program. */
1649 phc.offset = round_page(hdrsize + notesz);
1650 each_writable_segment(td, cb_put_phdr, &phc);
1654 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1656 struct note_info *ninfo;
1657 size_t size, notesize;
1660 out(arg, NULL, &size);
1661 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1663 ninfo->outfunc = out;
1664 ninfo->outarg = arg;
1665 ninfo->outsize = size;
1666 TAILQ_INSERT_TAIL(list, ninfo, link);
1671 notesize = sizeof(Elf_Note) + /* note header */
1672 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1674 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1680 append_note_data(const void *src, void *dst, size_t len)
1684 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1686 bcopy(src, dst, len);
1687 bzero((char *)dst + len, padded_len - len);
1689 return (padded_len);
1693 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1701 note = (Elf_Note *)buf;
1702 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1703 note->n_descsz = size;
1704 note->n_type = type;
1705 buf += sizeof(*note);
1706 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1707 sizeof(FREEBSD_ABI_VENDOR));
1708 append_note_data(src, buf, size);
1713 notesize = sizeof(Elf_Note) + /* note header */
1714 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1716 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1722 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1725 ssize_t old_len, sect_len;
1726 size_t new_len, descsz, i;
1728 if (ninfo->type == -1) {
1729 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1733 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1734 note.n_descsz = ninfo->outsize;
1735 note.n_type = ninfo->type;
1737 sbuf_bcat(sb, ¬e, sizeof(note));
1738 sbuf_start_section(sb, &old_len);
1739 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1740 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1741 if (note.n_descsz == 0)
1743 sbuf_start_section(sb, &old_len);
1744 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1745 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1749 new_len = (size_t)sect_len;
1750 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1751 if (new_len < descsz) {
1753 * It is expected that individual note emitters will correctly
1754 * predict their expected output size and fill up to that size
1755 * themselves, padding in a format-specific way if needed.
1756 * However, in case they don't, just do it here with zeros.
1758 for (i = 0; i < descsz - new_len; i++)
1760 } else if (new_len > descsz) {
1762 * We can't always truncate sb -- we may have drained some
1765 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1766 "read it (%zu > %zu). Since it is longer than "
1767 "expected, this coredump's notes are corrupt. THIS "
1768 "IS A BUG in the note_procstat routine for type %u.\n",
1769 __func__, (unsigned)note.n_type, new_len, descsz,
1770 (unsigned)note.n_type));
1775 * Miscellaneous note out functions.
1778 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1779 #include <compat/freebsd32/freebsd32.h>
1781 typedef struct prstatus32 elf_prstatus_t;
1782 typedef struct prpsinfo32 elf_prpsinfo_t;
1783 typedef struct fpreg32 elf_prfpregset_t;
1784 typedef struct fpreg32 elf_fpregset_t;
1785 typedef struct reg32 elf_gregset_t;
1786 typedef struct thrmisc32 elf_thrmisc_t;
1787 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1788 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1789 typedef uint32_t elf_ps_strings_t;
1791 typedef prstatus_t elf_prstatus_t;
1792 typedef prpsinfo_t elf_prpsinfo_t;
1793 typedef prfpregset_t elf_prfpregset_t;
1794 typedef prfpregset_t elf_fpregset_t;
1795 typedef gregset_t elf_gregset_t;
1796 typedef thrmisc_t elf_thrmisc_t;
1797 #define ELF_KERN_PROC_MASK 0
1798 typedef struct kinfo_proc elf_kinfo_proc_t;
1799 typedef vm_offset_t elf_ps_strings_t;
1803 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1806 elf_prpsinfo_t *psinfo;
1808 p = (struct proc *)arg;
1810 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1811 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1812 psinfo->pr_version = PRPSINFO_VERSION;
1813 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1814 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1816 * XXX - We don't fill in the command line arguments properly
1819 strlcpy(psinfo->pr_psargs, p->p_comm,
1820 sizeof(psinfo->pr_psargs));
1822 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1823 free(psinfo, M_TEMP);
1825 *sizep = sizeof(*psinfo);
1829 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1832 elf_prstatus_t *status;
1834 td = (struct thread *)arg;
1836 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1837 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1838 status->pr_version = PRSTATUS_VERSION;
1839 status->pr_statussz = sizeof(elf_prstatus_t);
1840 status->pr_gregsetsz = sizeof(elf_gregset_t);
1841 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1842 status->pr_osreldate = osreldate;
1843 status->pr_cursig = td->td_proc->p_sig;
1844 status->pr_pid = td->td_tid;
1845 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1846 fill_regs32(td, &status->pr_reg);
1848 fill_regs(td, &status->pr_reg);
1850 sbuf_bcat(sb, status, sizeof(*status));
1851 free(status, M_TEMP);
1853 *sizep = sizeof(*status);
1857 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1860 elf_prfpregset_t *fpregset;
1862 td = (struct thread *)arg;
1864 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1865 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1866 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1867 fill_fpregs32(td, fpregset);
1869 fill_fpregs(td, fpregset);
1871 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1872 free(fpregset, M_TEMP);
1874 *sizep = sizeof(*fpregset);
1878 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1881 elf_thrmisc_t thrmisc;
1883 td = (struct thread *)arg;
1885 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1886 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1887 strcpy(thrmisc.pr_tname, td->td_name);
1888 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1890 *sizep = sizeof(thrmisc);
1894 * Allow for MD specific notes, as well as any MD
1895 * specific preparations for writing MI notes.
1898 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1904 td = (struct thread *)arg;
1906 if (size != 0 && sb != NULL)
1907 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1911 __elfN(dump_thread)(td, buf, &size);
1912 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1913 if (size != 0 && sb != NULL)
1914 sbuf_bcat(sb, buf, size);
1919 #ifdef KINFO_PROC_SIZE
1920 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1924 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1930 p = (struct proc *)arg;
1931 size = sizeof(structsize) + p->p_numthreads *
1932 sizeof(elf_kinfo_proc_t);
1935 KASSERT(*sizep == size, ("invalid size"));
1936 structsize = sizeof(elf_kinfo_proc_t);
1937 sbuf_bcat(sb, &structsize, sizeof(structsize));
1938 sx_slock(&proctree_lock);
1940 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1941 sx_sunlock(&proctree_lock);
1946 #ifdef KINFO_FILE_SIZE
1947 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1951 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1954 size_t size, sect_sz, i;
1955 ssize_t start_len, sect_len;
1956 int structsize, filedesc_flags;
1958 if (coredump_pack_fileinfo)
1959 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1963 p = (struct proc *)arg;
1964 structsize = sizeof(struct kinfo_file);
1967 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1968 sbuf_set_drain(sb, sbuf_drain_count, &size);
1969 sbuf_bcat(sb, &structsize, sizeof(structsize));
1971 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1976 sbuf_start_section(sb, &start_len);
1978 sbuf_bcat(sb, &structsize, sizeof(structsize));
1980 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1983 sect_len = sbuf_end_section(sb, start_len, 0, 0);
1988 KASSERT(sect_sz <= *sizep,
1989 ("kern_proc_filedesc_out did not respect maxlen; "
1990 "requested %zu, got %zu", *sizep - sizeof(structsize),
1991 sect_sz - sizeof(structsize)));
1993 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1998 #ifdef KINFO_VMENTRY_SIZE
1999 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2003 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2007 int structsize, vmmap_flags;
2009 if (coredump_pack_vmmapinfo)
2010 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2014 p = (struct proc *)arg;
2015 structsize = sizeof(struct kinfo_vmentry);
2018 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2019 sbuf_set_drain(sb, sbuf_drain_count, &size);
2020 sbuf_bcat(sb, &structsize, sizeof(structsize));
2022 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2027 sbuf_bcat(sb, &structsize, sizeof(structsize));
2029 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2035 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2041 p = (struct proc *)arg;
2042 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2044 KASSERT(*sizep == size, ("invalid size"));
2045 structsize = sizeof(gid_t);
2046 sbuf_bcat(sb, &structsize, sizeof(structsize));
2047 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2054 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2060 p = (struct proc *)arg;
2061 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2063 KASSERT(*sizep == size, ("invalid size"));
2064 structsize = sizeof(p->p_fd->fd_cmask);
2065 sbuf_bcat(sb, &structsize, sizeof(structsize));
2066 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2072 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2075 struct rlimit rlim[RLIM_NLIMITS];
2079 p = (struct proc *)arg;
2080 size = sizeof(structsize) + sizeof(rlim);
2082 KASSERT(*sizep == size, ("invalid size"));
2083 structsize = sizeof(rlim);
2084 sbuf_bcat(sb, &structsize, sizeof(structsize));
2086 for (i = 0; i < RLIM_NLIMITS; i++)
2087 lim_rlimit_proc(p, i, &rlim[i]);
2089 sbuf_bcat(sb, rlim, sizeof(rlim));
2095 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2101 p = (struct proc *)arg;
2102 size = sizeof(structsize) + sizeof(p->p_osrel);
2104 KASSERT(*sizep == size, ("invalid size"));
2105 structsize = sizeof(p->p_osrel);
2106 sbuf_bcat(sb, &structsize, sizeof(structsize));
2107 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2113 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2116 elf_ps_strings_t ps_strings;
2120 p = (struct proc *)arg;
2121 size = sizeof(structsize) + sizeof(ps_strings);
2123 KASSERT(*sizep == size, ("invalid size"));
2124 structsize = sizeof(ps_strings);
2125 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2126 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2128 ps_strings = p->p_sysent->sv_psstrings;
2130 sbuf_bcat(sb, &structsize, sizeof(structsize));
2131 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2137 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2143 p = (struct proc *)arg;
2146 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2147 sbuf_set_drain(sb, sbuf_drain_count, &size);
2148 sbuf_bcat(sb, &structsize, sizeof(structsize));
2150 proc_getauxv(curthread, p, sb);
2156 structsize = sizeof(Elf_Auxinfo);
2157 sbuf_bcat(sb, &structsize, sizeof(structsize));
2159 proc_getauxv(curthread, p, sb);
2165 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2166 int32_t *osrel, const Elf_Phdr *pnote)
2168 const Elf_Note *note, *note0, *note_end;
2169 const char *note_name;
2174 /* We need some limit, might as well use PAGE_SIZE. */
2175 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2177 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2178 if (pnote->p_offset > PAGE_SIZE ||
2179 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2180 VOP_UNLOCK(imgp->vp, 0);
2181 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2182 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2183 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2184 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2185 curthread->td_ucred, NOCRED, NULL, curthread);
2187 uprintf("i/o error PT_NOTE\n");
2191 note = note0 = (const Elf_Note *)buf;
2192 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2194 note = note0 = (const Elf_Note *)(imgp->image_header +
2196 note_end = (const Elf_Note *)(imgp->image_header +
2197 pnote->p_offset + pnote->p_filesz);
2200 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2201 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2202 (const char *)note < sizeof(Elf_Note)) {
2206 if (note->n_namesz != checknote->hdr.n_namesz ||
2207 note->n_descsz != checknote->hdr.n_descsz ||
2208 note->n_type != checknote->hdr.n_type)
2210 note_name = (const char *)(note + 1);
2211 if (note_name + checknote->hdr.n_namesz >=
2212 (const char *)note_end || strncmp(checknote->vendor,
2213 note_name, checknote->hdr.n_namesz) != 0)
2217 * Fetch the osreldate for binary
2218 * from the ELF OSABI-note if necessary.
2220 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2221 checknote->trans_osrel != NULL) {
2222 res = checknote->trans_osrel(note, osrel);
2228 note = (const Elf_Note *)((const char *)(note + 1) +
2229 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2230 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2239 * Try to find the appropriate ABI-note section for checknote,
2240 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2241 * first page of the image is searched, the same as for headers.
2244 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2247 const Elf_Phdr *phdr;
2248 const Elf_Ehdr *hdr;
2251 hdr = (const Elf_Ehdr *)imgp->image_header;
2252 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2254 for (i = 0; i < hdr->e_phnum; i++) {
2255 if (phdr[i].p_type == PT_NOTE &&
2256 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2264 * Tell kern_execve.c about it, with a little help from the linker.
2266 static struct execsw __elfN(execsw) = {
2267 __CONCAT(exec_, __elfN(imgact)),
2268 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2270 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2273 __elfN(trans_prot)(Elf_Word flags)
2279 prot |= VM_PROT_EXECUTE;
2281 prot |= VM_PROT_WRITE;
2283 prot |= VM_PROT_READ;
2284 #if __ELF_WORD_SIZE == 32
2285 #if defined(__amd64__)
2286 if (i386_read_exec && (flags & PF_R))
2287 prot |= VM_PROT_EXECUTE;
2294 __elfN(untrans_prot)(vm_prot_t prot)
2299 if (prot & VM_PROT_EXECUTE)
2301 if (prot & VM_PROT_READ)
2303 if (prot & VM_PROT_WRITE)