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 #define ELF_NOTE_ROUNDSIZE 4
85 #define OLD_EI_BRAND 8
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89 const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
93 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107 #define CORE_BUF_SIZE (16 * 1024)
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0, "");
118 int __elfN(nxstack) =
119 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
124 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
125 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
126 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
128 #if __ELF_WORD_SIZE == 32
129 #if defined(__amd64__)
130 int i386_read_exec = 0;
131 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
132 "enable execution from readable segments");
136 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
138 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
139 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
140 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
142 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
144 Elf_Brandnote __elfN(freebsd_brandnote) = {
145 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
146 .hdr.n_descsz = sizeof(int32_t),
148 .vendor = FREEBSD_ABI_VENDOR,
149 .flags = BN_TRANSLATE_OSREL,
150 .trans_osrel = __elfN(freebsd_trans_osrel)
154 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
158 p = (uintptr_t)(note + 1);
159 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
160 *osrel = *(const int32_t *)(p);
165 static const char GNU_ABI_VENDOR[] = "GNU";
166 static int GNU_KFREEBSD_ABI_DESC = 3;
168 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
169 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
170 .hdr.n_descsz = 16, /* XXX at least 16 */
172 .vendor = GNU_ABI_VENDOR,
173 .flags = BN_TRANSLATE_OSREL,
174 .trans_osrel = kfreebsd_trans_osrel
178 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
180 const Elf32_Word *desc;
183 p = (uintptr_t)(note + 1);
184 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
186 desc = (const Elf32_Word *)p;
187 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
191 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
192 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
194 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
200 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
204 for (i = 0; i < MAX_BRANDS; i++) {
205 if (elf_brand_list[i] == NULL) {
206 elf_brand_list[i] = entry;
210 if (i == MAX_BRANDS) {
211 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
219 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
223 for (i = 0; i < MAX_BRANDS; i++) {
224 if (elf_brand_list[i] == entry) {
225 elf_brand_list[i] = NULL;
235 __elfN(brand_inuse)(Elf_Brandinfo *entry)
240 sx_slock(&allproc_lock);
241 FOREACH_PROC_IN_SYSTEM(p) {
242 if (p->p_sysent == entry->sysvec) {
247 sx_sunlock(&allproc_lock);
252 static Elf_Brandinfo *
253 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
254 int interp_name_len, int32_t *osrel)
256 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
262 * We support four types of branding -- (1) the ELF EI_OSABI field
263 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
264 * branding w/in the ELF header, (3) path of the `interp_path'
265 * field, and (4) the ".note.ABI-tag" ELF section.
268 /* Look for an ".note.ABI-tag" ELF section */
269 for (i = 0; i < MAX_BRANDS; i++) {
270 bi = elf_brand_list[i];
273 if (hdr->e_machine == bi->machine && (bi->flags &
274 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
275 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
276 /* Give brand a chance to veto check_note's guess */
277 if (ret && bi->header_supported)
278 ret = bi->header_supported(imgp);
284 /* If the executable has a brand, search for it in the brand list. */
285 for (i = 0; i < MAX_BRANDS; i++) {
286 bi = elf_brand_list[i];
287 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
289 if (hdr->e_machine == bi->machine &&
290 (hdr->e_ident[EI_OSABI] == bi->brand ||
291 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
292 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
293 /* Looks good, but give brand a chance to veto */
294 if (!bi->header_supported || bi->header_supported(imgp))
299 /* No known brand, see if the header is recognized by any brand */
300 for (i = 0; i < MAX_BRANDS; i++) {
301 bi = elf_brand_list[i];
302 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
303 bi->header_supported == NULL)
305 if (hdr->e_machine == bi->machine) {
306 ret = bi->header_supported(imgp);
312 /* Lacking a known brand, search for a recognized interpreter. */
313 if (interp != NULL) {
314 for (i = 0; i < MAX_BRANDS; i++) {
315 bi = elf_brand_list[i];
316 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
318 if (hdr->e_machine == bi->machine &&
319 /* ELF image p_filesz includes terminating zero */
320 strlen(bi->interp_path) + 1 == interp_name_len &&
321 strncmp(interp, bi->interp_path, interp_name_len)
327 /* Lacking a recognized interpreter, try the default brand */
328 for (i = 0; i < MAX_BRANDS; i++) {
329 bi = elf_brand_list[i];
330 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
332 if (hdr->e_machine == bi->machine &&
333 __elfN(fallback_brand) == bi->brand)
340 __elfN(check_header)(const Elf_Ehdr *hdr)
346 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
347 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
348 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
349 hdr->e_phentsize != sizeof(Elf_Phdr) ||
350 hdr->e_version != ELF_TARG_VER)
354 * Make sure we have at least one brand for this machine.
357 for (i = 0; i < MAX_BRANDS; i++) {
358 bi = elf_brand_list[i];
359 if (bi != NULL && bi->machine == hdr->e_machine)
369 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
370 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
377 * Create the page if it doesn't exist yet. Ignore errors.
380 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
381 VM_PROT_ALL, VM_PROT_ALL, 0);
385 * Find the page from the underlying object.
388 sf = vm_imgact_map_page(object, offset);
390 return (KERN_FAILURE);
391 off = offset - trunc_page(offset);
392 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
394 vm_imgact_unmap_page(sf);
396 return (KERN_FAILURE);
400 return (KERN_SUCCESS);
404 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
405 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
412 if (start != trunc_page(start)) {
413 rv = __elfN(map_partial)(map, object, offset, start,
414 round_page(start), prot);
417 offset += round_page(start) - start;
418 start = round_page(start);
420 if (end != round_page(end)) {
421 rv = __elfN(map_partial)(map, object, offset +
422 trunc_page(end) - start, trunc_page(end), end, prot);
425 end = trunc_page(end);
428 if (offset & PAGE_MASK) {
430 * The mapping is not page aligned. This means we have
431 * to copy the data. Sigh.
433 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
434 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
439 return (KERN_SUCCESS);
440 for (; start < end; start += sz) {
441 sf = vm_imgact_map_page(object, offset);
443 return (KERN_FAILURE);
444 off = offset - trunc_page(offset);
446 if (sz > PAGE_SIZE - off)
447 sz = PAGE_SIZE - off;
448 error = copyout((caddr_t)sf_buf_kva(sf) + off,
450 vm_imgact_unmap_page(sf);
452 return (KERN_FAILURE);
458 vm_object_reference(object);
460 rv = vm_map_insert(map, object, offset, start, end,
461 prot, VM_PROT_ALL, cow);
463 if (rv != KERN_SUCCESS)
464 vm_object_deallocate(object);
468 return (KERN_SUCCESS);
473 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
474 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
481 vm_offset_t map_addr;
484 vm_offset_t file_addr;
487 * It's necessary to fail if the filsz + offset taken from the
488 * header is greater than the actual file pager object's size.
489 * If we were to allow this, then the vm_map_find() below would
490 * walk right off the end of the file object and into the ether.
492 * While I'm here, might as well check for something else that
493 * is invalid: filsz cannot be greater than memsz.
495 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
496 uprintf("elf_load_section: truncated ELF file\n");
500 object = imgp->object;
501 map = &imgp->proc->p_vmspace->vm_map;
502 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
503 file_addr = trunc_page_ps(offset, pagesize);
506 * We have two choices. We can either clear the data in the last page
507 * of an oversized mapping, or we can start the anon mapping a page
508 * early and copy the initialized data into that first page. We
509 * choose the second..
512 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
514 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
517 /* cow flags: don't dump readonly sections in core */
518 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
519 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
521 rv = __elfN(map_insert)(map,
523 file_addr, /* file offset */
524 map_addr, /* virtual start */
525 map_addr + map_len,/* virtual end */
528 if (rv != KERN_SUCCESS)
531 /* we can stop now if we've covered it all */
532 if (memsz == filsz) {
539 * We have to get the remaining bit of the file into the first part
540 * of the oversized map segment. This is normally because the .data
541 * segment in the file is extended to provide bss. It's a neat idea
542 * to try and save a page, but it's a pain in the behind to implement.
544 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
545 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
546 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
549 /* This had damn well better be true! */
551 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
552 map_len, VM_PROT_ALL, 0);
553 if (rv != KERN_SUCCESS) {
561 sf = vm_imgact_map_page(object, offset + filsz);
565 /* send the page fragment to user space */
566 off = trunc_page_ps(offset + filsz, pagesize) -
567 trunc_page(offset + filsz);
568 error = copyout((caddr_t)sf_buf_kva(sf) + off,
569 (caddr_t)map_addr, copy_len);
570 vm_imgact_unmap_page(sf);
577 * set it to the specified protection.
578 * XXX had better undo the damage from pasting over the cracks here!
580 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
581 map_len), prot, FALSE);
587 * Load the file "file" into memory. It may be either a shared object
590 * The "addr" reference parameter is in/out. On entry, it specifies
591 * the address where a shared object should be loaded. If the file is
592 * an executable, this value is ignored. On exit, "addr" specifies
593 * where the file was actually loaded.
595 * The "entry" reference parameter is out only. On exit, it specifies
596 * the entry point for the loaded file.
599 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
600 u_long *entry, size_t pagesize)
605 struct image_params image_params;
607 const Elf_Ehdr *hdr = NULL;
608 const Elf_Phdr *phdr = NULL;
609 struct nameidata *nd;
611 struct image_params *imgp;
614 u_long base_addr = 0;
615 int error, i, numsegs;
617 #ifdef CAPABILITY_MODE
619 * XXXJA: This check can go away once we are sufficiently confident
620 * that the checks in namei() are correct.
622 if (IN_CAPABILITY_MODE(curthread))
626 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
628 attr = &tempdata->attr;
629 imgp = &tempdata->image_params;
632 * Initialize part of the common data
636 imgp->firstpage = NULL;
637 imgp->image_header = NULL;
639 imgp->execlabel = NULL;
641 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
642 if ((error = namei(nd)) != 0) {
646 NDFREE(nd, NDF_ONLY_PNBUF);
647 imgp->vp = nd->ni_vp;
650 * Check permissions, modes, uid, etc on the file, and "open" it.
652 error = exec_check_permissions(imgp);
656 error = exec_map_first_page(imgp);
661 * Also make certain that the interpreter stays the same, so set
662 * its VV_TEXT flag, too.
664 VOP_SET_TEXT(nd->ni_vp);
666 imgp->object = nd->ni_vp->v_object;
668 hdr = (const Elf_Ehdr *)imgp->image_header;
669 if ((error = __elfN(check_header)(hdr)) != 0)
671 if (hdr->e_type == ET_DYN)
673 else if (hdr->e_type == ET_EXEC)
680 /* Only support headers that fit within first page for now */
681 if ((hdr->e_phoff > PAGE_SIZE) ||
682 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
687 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
688 if (!aligned(phdr, Elf_Addr)) {
693 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
694 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
695 /* Loadable segment */
696 prot = __elfN(trans_prot)(phdr[i].p_flags);
697 error = __elfN(load_section)(imgp, phdr[i].p_offset,
698 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
699 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
703 * Establish the base address if this is the
707 base_addr = trunc_page(phdr[i].p_vaddr +
713 *entry = (unsigned long)hdr->e_entry + rbase;
717 exec_unmap_first_page(imgp);
722 free(tempdata, M_TEMP);
728 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
732 const Elf_Phdr *phdr;
733 Elf_Auxargs *elf_auxargs;
734 struct vmspace *vmspace;
735 const char *err_str, *newinterp;
736 char *interp, *interp_buf, *path;
737 Elf_Brandinfo *brand_info;
738 struct sysentvec *sv;
740 u_long text_size, data_size, total_size, text_addr, data_addr;
741 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
743 int error, i, n, interp_name_len, have_interp;
745 hdr = (const Elf_Ehdr *)imgp->image_header;
748 * Do we have a valid ELF header ?
750 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
751 * if particular brand doesn't support it.
753 if (__elfN(check_header)(hdr) != 0 ||
754 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
758 * From here on down, we return an errno, not -1, as we've
759 * detected an ELF file.
762 if ((hdr->e_phoff > PAGE_SIZE) ||
763 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
764 /* Only support headers in first page for now */
765 uprintf("Program headers not in the first page\n");
768 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
769 if (!aligned(phdr, Elf_Addr)) {
770 uprintf("Unaligned program headers\n");
777 text_size = data_size = total_size = text_addr = data_addr = 0;
780 err_str = newinterp = NULL;
781 interp = interp_buf = NULL;
784 for (i = 0; i < hdr->e_phnum; i++) {
785 switch (phdr[i].p_type) {
788 baddr = phdr[i].p_vaddr;
792 /* Path to interpreter */
793 if (phdr[i].p_filesz > MAXPATHLEN) {
794 uprintf("Invalid PT_INTERP\n");
798 interp_name_len = phdr[i].p_filesz;
799 if (phdr[i].p_offset > PAGE_SIZE ||
800 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
801 VOP_UNLOCK(imgp->vp, 0);
802 interp_buf = malloc(interp_name_len + 1, M_TEMP,
804 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
805 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
806 interp_name_len, phdr[i].p_offset,
807 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
810 uprintf("i/o error PT_INTERP\n");
813 interp_buf[interp_name_len] = '\0';
816 interp = __DECONST(char *, imgp->image_header) +
823 __elfN(trans_prot)(phdr[i].p_flags);
824 imgp->stack_sz = phdr[i].p_memsz;
829 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
831 if (brand_info == NULL) {
832 uprintf("ELF binary type \"%u\" not known.\n",
833 hdr->e_ident[EI_OSABI]);
837 if (hdr->e_type == ET_DYN) {
838 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
839 uprintf("Cannot execute shared object\n");
844 * Honour the base load address from the dso if it is
845 * non-zero for some reason.
848 et_dyn_addr = ET_DYN_LOAD_ADDR;
853 sv = brand_info->sysvec;
854 if (interp != NULL && brand_info->interp_newpath != NULL)
855 newinterp = brand_info->interp_newpath;
858 * Avoid a possible deadlock if the current address space is destroyed
859 * and that address space maps the locked vnode. In the common case,
860 * the locked vnode's v_usecount is decremented but remains greater
861 * than zero. Consequently, the vnode lock is not needed by vrele().
862 * However, in cases where the vnode lock is external, such as nullfs,
863 * v_usecount may become zero.
865 * The VV_TEXT flag prevents modifications to the executable while
866 * the vnode is unlocked.
868 VOP_UNLOCK(imgp->vp, 0);
870 error = exec_new_vmspace(imgp, sv);
871 imgp->proc->p_sysent = sv;
873 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
877 for (i = 0; i < hdr->e_phnum; i++) {
878 switch (phdr[i].p_type) {
879 case PT_LOAD: /* Loadable segment */
880 if (phdr[i].p_memsz == 0)
882 prot = __elfN(trans_prot)(phdr[i].p_flags);
883 error = __elfN(load_section)(imgp, phdr[i].p_offset,
884 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
885 phdr[i].p_memsz, phdr[i].p_filesz, prot,
891 * If this segment contains the program headers,
892 * remember their virtual address for the AT_PHDR
893 * aux entry. Static binaries don't usually include
896 if (phdr[i].p_offset == 0 &&
897 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
899 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
902 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
903 seg_size = round_page(phdr[i].p_memsz +
904 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
907 * Make the largest executable segment the official
908 * text segment and all others data.
910 * Note that obreak() assumes that data_addr +
911 * data_size == end of data load area, and the ELF
912 * file format expects segments to be sorted by
913 * address. If multiple data segments exist, the
914 * last one will be used.
917 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
918 text_size = seg_size;
919 text_addr = seg_addr;
921 data_size = seg_size;
922 data_addr = seg_addr;
924 total_size += seg_size;
926 case PT_PHDR: /* Program header table info */
927 proghdr = phdr[i].p_vaddr + et_dyn_addr;
934 if (data_addr == 0 && data_size == 0) {
935 data_addr = text_addr;
936 data_size = text_size;
939 entry = (u_long)hdr->e_entry + et_dyn_addr;
942 * Check limits. It should be safe to check the
943 * limits after loading the segments since we do
944 * not actually fault in all the segments pages.
946 PROC_LOCK(imgp->proc);
947 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
948 err_str = "Data segment size exceeds process limit";
949 else if (text_size > maxtsiz)
950 err_str = "Text segment size exceeds system limit";
951 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
952 err_str = "Total segment size exceeds process limit";
953 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
954 err_str = "Data segment size exceeds resource limit";
955 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
956 err_str = "Total segment size exceeds resource limit";
957 if (err_str != NULL) {
958 PROC_UNLOCK(imgp->proc);
959 uprintf("%s\n", err_str);
964 vmspace = imgp->proc->p_vmspace;
965 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
966 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
967 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
968 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
971 * We load the dynamic linker where a userland call
972 * to mmap(0, ...) would put it. The rationale behind this
973 * calculation is that it leaves room for the heap to grow to
974 * its maximum allowed size.
976 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
978 PROC_UNLOCK(imgp->proc);
980 imgp->entry_addr = entry;
982 if (interp != NULL) {
984 VOP_UNLOCK(imgp->vp, 0);
985 if (brand_info->emul_path != NULL &&
986 brand_info->emul_path[0] != '\0') {
987 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
988 snprintf(path, MAXPATHLEN, "%s%s",
989 brand_info->emul_path, interp);
990 error = __elfN(load_file)(imgp->proc, path, &addr,
991 &imgp->entry_addr, sv->sv_pagesize);
996 if (!have_interp && newinterp != NULL) {
997 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
998 &imgp->entry_addr, sv->sv_pagesize);
1003 error = __elfN(load_file)(imgp->proc, interp, &addr,
1004 &imgp->entry_addr, sv->sv_pagesize);
1006 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1008 uprintf("ELF interpreter %s not found\n", interp);
1015 * Construct auxargs table (used by the fixup routine)
1017 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1018 elf_auxargs->execfd = -1;
1019 elf_auxargs->phdr = proghdr;
1020 elf_auxargs->phent = hdr->e_phentsize;
1021 elf_auxargs->phnum = hdr->e_phnum;
1022 elf_auxargs->pagesz = PAGE_SIZE;
1023 elf_auxargs->base = addr;
1024 elf_auxargs->flags = 0;
1025 elf_auxargs->entry = entry;
1026 elf_auxargs->hdr_eflags = hdr->e_flags;
1028 imgp->auxargs = elf_auxargs;
1029 imgp->interpreted = 0;
1030 imgp->reloc_base = addr;
1031 imgp->proc->p_osrel = osrel;
1034 free(interp_buf, M_TEMP);
1038 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1041 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1043 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1047 base = (Elf_Addr *)*stack_base;
1048 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1050 if (args->execfd != -1)
1051 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1052 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1053 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1054 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1055 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1056 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1057 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1058 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1060 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1062 if (imgp->execpathp != 0)
1063 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1064 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1065 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1066 if (imgp->canary != 0) {
1067 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1068 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1070 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1071 if (imgp->pagesizes != 0) {
1072 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1073 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1075 if (imgp->sysent->sv_timekeep_base != 0) {
1076 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1077 imgp->sysent->sv_timekeep_base);
1079 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1080 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1081 imgp->sysent->sv_stackprot);
1082 AUXARGS_ENTRY(pos, AT_NULL, 0);
1084 free(imgp->auxargs, M_TEMP);
1085 imgp->auxargs = NULL;
1088 suword(base, (long)imgp->args->argc);
1089 *stack_base = (register_t *)base;
1094 * Code for generating ELF core dumps.
1097 typedef void (*segment_callback)(vm_map_entry_t, void *);
1099 /* Closure for cb_put_phdr(). */
1100 struct phdr_closure {
1101 Elf_Phdr *phdr; /* Program header to fill in */
1102 Elf_Off offset; /* Offset of segment in core file */
1105 /* Closure for cb_size_segment(). */
1106 struct sseg_closure {
1107 int count; /* Count of writable segments. */
1108 size_t size; /* Total size of all writable segments. */
1111 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1114 int type; /* Note type. */
1115 outfunc_t outfunc; /* Output function. */
1116 void *outarg; /* Argument for the output function. */
1117 size_t outsize; /* Output size. */
1118 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1121 TAILQ_HEAD(note_info_list, note_info);
1123 /* Coredump output parameters. */
1124 struct coredump_params {
1126 struct ucred *active_cred;
1127 struct ucred *file_cred;
1130 struct gzio_stream *gzs;
1133 static void cb_put_phdr(vm_map_entry_t, void *);
1134 static void cb_size_segment(vm_map_entry_t, void *);
1135 static int core_write(struct coredump_params *, void *, size_t, off_t,
1137 static void each_writable_segment(struct thread *, segment_callback, void *);
1138 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1139 struct note_info_list *, size_t);
1140 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1142 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1143 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1144 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1145 static int sbuf_drain_core_output(void *, const char *, int);
1146 static int sbuf_drain_count(void *arg, const char *data, int len);
1148 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1149 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1150 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1151 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1152 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1153 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1154 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1155 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1156 static void note_procstat_files(void *, struct sbuf *, size_t *);
1157 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1158 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1159 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1160 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1161 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1164 extern int compress_user_cores_gzlevel;
1167 * Write out a core segment to the compression stream.
1170 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1176 chunk_len = MIN(len, CORE_BUF_SIZE);
1177 copyin(base, buf, chunk_len);
1178 error = gzio_write(p->gzs, buf, chunk_len);
1188 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1191 return (core_write((struct coredump_params *)arg, base, len, offset,
1197 core_write(struct coredump_params *p, void *base, size_t len, off_t offset,
1201 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, base, len, offset,
1202 seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1203 p->active_cred, p->file_cred, NULL, p->td));
1207 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1213 return (compress_chunk(p, base, tmpbuf, len));
1215 return (core_write(p, base, len, offset, UIO_USERSPACE));
1219 * Drain into a core file.
1222 sbuf_drain_core_output(void *arg, const char *data, int len)
1224 struct coredump_params *p;
1227 p = (struct coredump_params *)arg;
1230 * Some kern_proc out routines that print to this sbuf may
1231 * call us with the process lock held. Draining with the
1232 * non-sleepable lock held is unsafe. The lock is needed for
1233 * those routines when dumping a live process. In our case we
1234 * can safely release the lock before draining and acquire
1237 locked = PROC_LOCKED(p->td->td_proc);
1239 PROC_UNLOCK(p->td->td_proc);
1242 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1245 error = core_write(p, __DECONST(void *, data), len, p->offset,
1248 PROC_LOCK(p->td->td_proc);
1256 * Drain into a counter.
1259 sbuf_drain_count(void *arg, const char *data __unused, int len)
1263 sizep = (size_t *)arg;
1269 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1271 struct ucred *cred = td->td_ucred;
1273 struct sseg_closure seginfo;
1274 struct note_info_list notelst;
1275 struct coredump_params params;
1276 struct note_info *ninfo;
1278 size_t hdrsize, notesz, coresize;
1282 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1286 TAILQ_INIT(¬elst);
1288 /* Size the program segments. */
1291 each_writable_segment(td, cb_size_segment, &seginfo);
1294 * Collect info about the core file header area.
1296 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1297 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1298 coresize = round_page(hdrsize + notesz) + seginfo.size;
1300 /* Set up core dump parameters. */
1302 params.active_cred = cred;
1303 params.file_cred = NOCRED;
1310 PROC_LOCK(td->td_proc);
1311 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1312 PROC_UNLOCK(td->td_proc);
1319 if (coresize >= limit) {
1325 /* Create a compression stream if necessary. */
1327 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1328 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1329 if (params.gzs == NULL) {
1333 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1338 * Allocate memory for building the header, fill it up,
1339 * and write it out following the notes.
1341 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1346 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1349 /* Write the contents of all of the writable segments. */
1355 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1356 offset = round_page(hdrsize + notesz);
1357 for (i = 0; i < seginfo.count; i++) {
1358 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1359 php->p_filesz, offset, ¶ms, tmpbuf);
1362 offset += php->p_filesz;
1366 if (error == 0 && compress)
1367 error = gzio_flush(params.gzs);
1372 "Failed to write core file for process %s (error %d)\n",
1373 curproc->p_comm, error);
1379 free(tmpbuf, M_TEMP);
1380 if (params.gzs != NULL)
1381 gzio_fini(params.gzs);
1384 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1385 TAILQ_REMOVE(¬elst, ninfo, link);
1386 free(ninfo, M_TEMP);
1395 * A callback for each_writable_segment() to write out the segment's
1396 * program header entry.
1399 cb_put_phdr(entry, closure)
1400 vm_map_entry_t entry;
1403 struct phdr_closure *phc = (struct phdr_closure *)closure;
1404 Elf_Phdr *phdr = phc->phdr;
1406 phc->offset = round_page(phc->offset);
1408 phdr->p_type = PT_LOAD;
1409 phdr->p_offset = phc->offset;
1410 phdr->p_vaddr = entry->start;
1412 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1413 phdr->p_align = PAGE_SIZE;
1414 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1416 phc->offset += phdr->p_filesz;
1421 * A callback for each_writable_segment() to gather information about
1422 * the number of segments and their total size.
1425 cb_size_segment(entry, closure)
1426 vm_map_entry_t entry;
1429 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1432 ssc->size += entry->end - entry->start;
1436 * For each writable segment in the process's memory map, call the given
1437 * function with a pointer to the map entry and some arbitrary
1438 * caller-supplied data.
1441 each_writable_segment(td, func, closure)
1443 segment_callback func;
1446 struct proc *p = td->td_proc;
1447 vm_map_t map = &p->p_vmspace->vm_map;
1448 vm_map_entry_t entry;
1449 vm_object_t backing_object, object;
1450 boolean_t ignore_entry;
1452 vm_map_lock_read(map);
1453 for (entry = map->header.next; entry != &map->header;
1454 entry = entry->next) {
1456 * Don't dump inaccessible mappings, deal with legacy
1459 * Note that read-only segments related to the elf binary
1460 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1461 * need to arbitrarily ignore such segments.
1463 if (elf_legacy_coredump) {
1464 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1467 if ((entry->protection & VM_PROT_ALL) == 0)
1472 * Dont include memory segment in the coredump if
1473 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1474 * madvise(2). Do not dump submaps (i.e. parts of the
1477 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1480 if ((object = entry->object.vm_object) == NULL)
1483 /* Ignore memory-mapped devices and such things. */
1484 VM_OBJECT_RLOCK(object);
1485 while ((backing_object = object->backing_object) != NULL) {
1486 VM_OBJECT_RLOCK(backing_object);
1487 VM_OBJECT_RUNLOCK(object);
1488 object = backing_object;
1490 ignore_entry = object->type != OBJT_DEFAULT &&
1491 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1492 object->type != OBJT_PHYS;
1493 VM_OBJECT_RUNLOCK(object);
1497 (*func)(entry, closure);
1499 vm_map_unlock_read(map);
1503 * Write the core file header to the file, including padding up to
1504 * the page boundary.
1507 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1508 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1510 struct note_info *ninfo;
1514 /* Fill in the header. */
1515 bzero(hdr, hdrsize);
1516 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1518 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1519 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1520 sbuf_start_section(sb, NULL);
1521 sbuf_bcat(sb, hdr, hdrsize);
1522 TAILQ_FOREACH(ninfo, notelst, link)
1523 __elfN(putnote)(ninfo, sb);
1524 /* Align up to a page boundary for the program segments. */
1525 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1526 error = sbuf_finish(sb);
1533 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1543 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1546 * To have the debugger select the right thread (LWP) as the initial
1547 * thread, we dump the state of the thread passed to us in td first.
1548 * This is the thread that causes the core dump and thus likely to
1549 * be the right thread one wants to have selected in the debugger.
1552 while (thr != NULL) {
1553 size += register_note(list, NT_PRSTATUS,
1554 __elfN(note_prstatus), thr);
1555 size += register_note(list, NT_FPREGSET,
1556 __elfN(note_fpregset), thr);
1557 size += register_note(list, NT_THRMISC,
1558 __elfN(note_thrmisc), thr);
1559 size += register_note(list, -1,
1560 __elfN(note_threadmd), thr);
1562 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1563 TAILQ_NEXT(thr, td_plist);
1565 thr = TAILQ_NEXT(thr, td_plist);
1568 size += register_note(list, NT_PROCSTAT_PROC,
1569 __elfN(note_procstat_proc), p);
1570 size += register_note(list, NT_PROCSTAT_FILES,
1571 note_procstat_files, p);
1572 size += register_note(list, NT_PROCSTAT_VMMAP,
1573 note_procstat_vmmap, p);
1574 size += register_note(list, NT_PROCSTAT_GROUPS,
1575 note_procstat_groups, p);
1576 size += register_note(list, NT_PROCSTAT_UMASK,
1577 note_procstat_umask, p);
1578 size += register_note(list, NT_PROCSTAT_RLIMIT,
1579 note_procstat_rlimit, p);
1580 size += register_note(list, NT_PROCSTAT_OSREL,
1581 note_procstat_osrel, p);
1582 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1583 __elfN(note_procstat_psstrings), p);
1584 size += register_note(list, NT_PROCSTAT_AUXV,
1585 __elfN(note_procstat_auxv), p);
1591 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1596 struct phdr_closure phc;
1598 ehdr = (Elf_Ehdr *)hdr;
1599 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1601 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1602 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1603 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1604 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1605 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1606 ehdr->e_ident[EI_DATA] = ELF_DATA;
1607 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1608 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1609 ehdr->e_ident[EI_ABIVERSION] = 0;
1610 ehdr->e_ident[EI_PAD] = 0;
1611 ehdr->e_type = ET_CORE;
1612 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1613 ehdr->e_machine = ELF_ARCH32;
1615 ehdr->e_machine = ELF_ARCH;
1617 ehdr->e_version = EV_CURRENT;
1619 ehdr->e_phoff = sizeof(Elf_Ehdr);
1621 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1622 ehdr->e_phentsize = sizeof(Elf_Phdr);
1623 ehdr->e_phnum = numsegs + 1;
1624 ehdr->e_shentsize = sizeof(Elf_Shdr);
1626 ehdr->e_shstrndx = SHN_UNDEF;
1629 * Fill in the program header entries.
1632 /* The note segement. */
1633 phdr->p_type = PT_NOTE;
1634 phdr->p_offset = hdrsize;
1637 phdr->p_filesz = notesz;
1639 phdr->p_flags = PF_R;
1640 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1643 /* All the writable segments from the program. */
1645 phc.offset = round_page(hdrsize + notesz);
1646 each_writable_segment(td, cb_put_phdr, &phc);
1650 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1652 struct note_info *ninfo;
1653 size_t size, notesize;
1656 out(arg, NULL, &size);
1657 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1659 ninfo->outfunc = out;
1660 ninfo->outarg = arg;
1661 ninfo->outsize = size;
1662 TAILQ_INSERT_TAIL(list, ninfo, link);
1667 notesize = sizeof(Elf_Note) + /* note header */
1668 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1670 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1676 append_note_data(const void *src, void *dst, size_t len)
1680 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1682 bcopy(src, dst, len);
1683 bzero((char *)dst + len, padded_len - len);
1685 return (padded_len);
1689 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1697 note = (Elf_Note *)buf;
1698 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1699 note->n_descsz = size;
1700 note->n_type = type;
1701 buf += sizeof(*note);
1702 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1703 sizeof(FREEBSD_ABI_VENDOR));
1704 append_note_data(src, buf, size);
1709 notesize = sizeof(Elf_Note) + /* note header */
1710 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1712 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1718 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1721 ssize_t old_len, sect_len;
1722 size_t new_len, descsz, i;
1724 if (ninfo->type == -1) {
1725 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1729 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1730 note.n_descsz = ninfo->outsize;
1731 note.n_type = ninfo->type;
1733 sbuf_bcat(sb, ¬e, sizeof(note));
1734 sbuf_start_section(sb, &old_len);
1735 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1736 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1737 if (note.n_descsz == 0)
1739 sbuf_start_section(sb, &old_len);
1740 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1741 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1745 new_len = (size_t)sect_len;
1746 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1747 if (new_len < descsz) {
1749 * It is expected that individual note emitters will correctly
1750 * predict their expected output size and fill up to that size
1751 * themselves, padding in a format-specific way if needed.
1752 * However, in case they don't, just do it here with zeros.
1754 for (i = 0; i < descsz - new_len; i++)
1756 } else if (new_len > descsz) {
1758 * We can't always truncate sb -- we may have drained some
1761 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1762 "read it (%zu > %zu). Since it is longer than "
1763 "expected, this coredump's notes are corrupt. THIS "
1764 "IS A BUG in the note_procstat routine for type %u.\n",
1765 __func__, (unsigned)note.n_type, new_len, descsz,
1766 (unsigned)note.n_type));
1771 * Miscellaneous note out functions.
1774 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1775 #include <compat/freebsd32/freebsd32.h>
1777 typedef struct prstatus32 elf_prstatus_t;
1778 typedef struct prpsinfo32 elf_prpsinfo_t;
1779 typedef struct fpreg32 elf_prfpregset_t;
1780 typedef struct fpreg32 elf_fpregset_t;
1781 typedef struct reg32 elf_gregset_t;
1782 typedef struct thrmisc32 elf_thrmisc_t;
1783 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1784 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1785 typedef uint32_t elf_ps_strings_t;
1787 typedef prstatus_t elf_prstatus_t;
1788 typedef prpsinfo_t elf_prpsinfo_t;
1789 typedef prfpregset_t elf_prfpregset_t;
1790 typedef prfpregset_t elf_fpregset_t;
1791 typedef gregset_t elf_gregset_t;
1792 typedef thrmisc_t elf_thrmisc_t;
1793 #define ELF_KERN_PROC_MASK 0
1794 typedef struct kinfo_proc elf_kinfo_proc_t;
1795 typedef vm_offset_t elf_ps_strings_t;
1799 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1802 elf_prpsinfo_t *psinfo;
1804 p = (struct proc *)arg;
1806 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1807 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1808 psinfo->pr_version = PRPSINFO_VERSION;
1809 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1810 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1812 * XXX - We don't fill in the command line arguments properly
1815 strlcpy(psinfo->pr_psargs, p->p_comm,
1816 sizeof(psinfo->pr_psargs));
1818 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1819 free(psinfo, M_TEMP);
1821 *sizep = sizeof(*psinfo);
1825 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1828 elf_prstatus_t *status;
1830 td = (struct thread *)arg;
1832 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1833 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1834 status->pr_version = PRSTATUS_VERSION;
1835 status->pr_statussz = sizeof(elf_prstatus_t);
1836 status->pr_gregsetsz = sizeof(elf_gregset_t);
1837 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1838 status->pr_osreldate = osreldate;
1839 status->pr_cursig = td->td_proc->p_sig;
1840 status->pr_pid = td->td_tid;
1841 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1842 fill_regs32(td, &status->pr_reg);
1844 fill_regs(td, &status->pr_reg);
1846 sbuf_bcat(sb, status, sizeof(*status));
1847 free(status, M_TEMP);
1849 *sizep = sizeof(*status);
1853 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1856 elf_prfpregset_t *fpregset;
1858 td = (struct thread *)arg;
1860 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1861 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1862 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1863 fill_fpregs32(td, fpregset);
1865 fill_fpregs(td, fpregset);
1867 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1868 free(fpregset, M_TEMP);
1870 *sizep = sizeof(*fpregset);
1874 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1877 elf_thrmisc_t thrmisc;
1879 td = (struct thread *)arg;
1881 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1882 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1883 strcpy(thrmisc.pr_tname, td->td_name);
1884 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1886 *sizep = sizeof(thrmisc);
1890 * Allow for MD specific notes, as well as any MD
1891 * specific preparations for writing MI notes.
1894 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1900 td = (struct thread *)arg;
1902 if (size != 0 && sb != NULL)
1903 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1907 __elfN(dump_thread)(td, buf, &size);
1908 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1909 if (size != 0 && sb != NULL)
1910 sbuf_bcat(sb, buf, size);
1915 #ifdef KINFO_PROC_SIZE
1916 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1920 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1926 p = (struct proc *)arg;
1927 size = sizeof(structsize) + p->p_numthreads *
1928 sizeof(elf_kinfo_proc_t);
1931 KASSERT(*sizep == size, ("invalid size"));
1932 structsize = sizeof(elf_kinfo_proc_t);
1933 sbuf_bcat(sb, &structsize, sizeof(structsize));
1934 sx_slock(&proctree_lock);
1936 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1937 sx_sunlock(&proctree_lock);
1942 #ifdef KINFO_FILE_SIZE
1943 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1947 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1950 size_t size, sect_sz, i;
1951 ssize_t start_len, sect_len;
1952 int structsize, filedesc_flags;
1954 if (coredump_pack_fileinfo)
1955 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
1959 p = (struct proc *)arg;
1960 structsize = sizeof(struct kinfo_file);
1963 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1964 sbuf_set_drain(sb, sbuf_drain_count, &size);
1965 sbuf_bcat(sb, &structsize, sizeof(structsize));
1967 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
1972 sbuf_start_section(sb, &start_len);
1974 sbuf_bcat(sb, &structsize, sizeof(structsize));
1976 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
1979 sect_len = sbuf_end_section(sb, start_len, 0, 0);
1984 KASSERT(sect_sz <= *sizep,
1985 ("kern_proc_filedesc_out did not respect maxlen; "
1986 "requested %zu, got %zu", *sizep - sizeof(structsize),
1987 sect_sz - sizeof(structsize)));
1989 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
1994 #ifdef KINFO_VMENTRY_SIZE
1995 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1999 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2003 int structsize, vmmap_flags;
2005 if (coredump_pack_vmmapinfo)
2006 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2010 p = (struct proc *)arg;
2011 structsize = sizeof(struct kinfo_vmentry);
2014 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2015 sbuf_set_drain(sb, sbuf_drain_count, &size);
2016 sbuf_bcat(sb, &structsize, sizeof(structsize));
2018 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2023 sbuf_bcat(sb, &structsize, sizeof(structsize));
2025 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2031 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2037 p = (struct proc *)arg;
2038 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2040 KASSERT(*sizep == size, ("invalid size"));
2041 structsize = sizeof(gid_t);
2042 sbuf_bcat(sb, &structsize, sizeof(structsize));
2043 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2050 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2056 p = (struct proc *)arg;
2057 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2059 KASSERT(*sizep == size, ("invalid size"));
2060 structsize = sizeof(p->p_fd->fd_cmask);
2061 sbuf_bcat(sb, &structsize, sizeof(structsize));
2062 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2068 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2071 struct rlimit rlim[RLIM_NLIMITS];
2075 p = (struct proc *)arg;
2076 size = sizeof(structsize) + sizeof(rlim);
2078 KASSERT(*sizep == size, ("invalid size"));
2079 structsize = sizeof(rlim);
2080 sbuf_bcat(sb, &structsize, sizeof(structsize));
2082 for (i = 0; i < RLIM_NLIMITS; i++)
2083 lim_rlimit_proc(p, i, &rlim[i]);
2085 sbuf_bcat(sb, rlim, sizeof(rlim));
2091 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2097 p = (struct proc *)arg;
2098 size = sizeof(structsize) + sizeof(p->p_osrel);
2100 KASSERT(*sizep == size, ("invalid size"));
2101 structsize = sizeof(p->p_osrel);
2102 sbuf_bcat(sb, &structsize, sizeof(structsize));
2103 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2109 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2112 elf_ps_strings_t ps_strings;
2116 p = (struct proc *)arg;
2117 size = sizeof(structsize) + sizeof(ps_strings);
2119 KASSERT(*sizep == size, ("invalid size"));
2120 structsize = sizeof(ps_strings);
2121 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2122 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2124 ps_strings = p->p_sysent->sv_psstrings;
2126 sbuf_bcat(sb, &structsize, sizeof(structsize));
2127 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2133 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2139 p = (struct proc *)arg;
2142 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2143 sbuf_set_drain(sb, sbuf_drain_count, &size);
2144 sbuf_bcat(sb, &structsize, sizeof(structsize));
2146 proc_getauxv(curthread, p, sb);
2152 structsize = sizeof(Elf_Auxinfo);
2153 sbuf_bcat(sb, &structsize, sizeof(structsize));
2155 proc_getauxv(curthread, p, sb);
2161 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2162 int32_t *osrel, const Elf_Phdr *pnote)
2164 const Elf_Note *note, *note0, *note_end;
2165 const char *note_name;
2170 /* We need some limit, might as well use PAGE_SIZE. */
2171 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2173 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2174 if (pnote->p_offset > PAGE_SIZE ||
2175 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2176 VOP_UNLOCK(imgp->vp, 0);
2177 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2178 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2179 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2180 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2181 curthread->td_ucred, NOCRED, NULL, curthread);
2183 uprintf("i/o error PT_NOTE\n");
2187 note = note0 = (const Elf_Note *)buf;
2188 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2190 note = note0 = (const Elf_Note *)(imgp->image_header +
2192 note_end = (const Elf_Note *)(imgp->image_header +
2193 pnote->p_offset + pnote->p_filesz);
2196 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2197 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2198 (const char *)note < sizeof(Elf_Note)) {
2202 if (note->n_namesz != checknote->hdr.n_namesz ||
2203 note->n_descsz != checknote->hdr.n_descsz ||
2204 note->n_type != checknote->hdr.n_type)
2206 note_name = (const char *)(note + 1);
2207 if (note_name + checknote->hdr.n_namesz >=
2208 (const char *)note_end || strncmp(checknote->vendor,
2209 note_name, checknote->hdr.n_namesz) != 0)
2213 * Fetch the osreldate for binary
2214 * from the ELF OSABI-note if necessary.
2216 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2217 checknote->trans_osrel != NULL) {
2218 res = checknote->trans_osrel(note, osrel);
2224 note = (const Elf_Note *)((const char *)(note + 1) +
2225 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2226 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2235 * Try to find the appropriate ABI-note section for checknote,
2236 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2237 * first page of the image is searched, the same as for headers.
2240 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2243 const Elf_Phdr *phdr;
2244 const Elf_Ehdr *hdr;
2247 hdr = (const Elf_Ehdr *)imgp->image_header;
2248 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2250 for (i = 0; i < hdr->e_phnum; i++) {
2251 if (phdr[i].p_type == PT_NOTE &&
2252 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2260 * Tell kern_execve.c about it, with a little help from the linker.
2262 static struct execsw __elfN(execsw) = {
2263 __CONCAT(exec_, __elfN(imgact)),
2264 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2266 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2269 __elfN(trans_prot)(Elf_Word flags)
2275 prot |= VM_PROT_EXECUTE;
2277 prot |= VM_PROT_WRITE;
2279 prot |= VM_PROT_READ;
2280 #if __ELF_WORD_SIZE == 32
2281 #if defined(__amd64__)
2282 if (i386_read_exec && (flags & PF_R))
2283 prot |= VM_PROT_EXECUTE;
2290 __elfN(untrans_prot)(vm_prot_t prot)
2295 if (prot & VM_PROT_EXECUTE)
2297 if (prot & VM_PROT_READ)
2299 if (prot & VM_PROT_WRITE)