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/capability.h>
41 #include <sys/fcntl.h>
42 #include <sys/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/pioctl.h>
52 #include <sys/procfs.h>
53 #include <sys/racct.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
57 #include <sys/sf_buf.h>
59 #include <sys/systm.h>
60 #include <sys/signalvar.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/vnode.h>
67 #include <sys/syslog.h>
68 #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 #ifdef COMPRESS_USER_CORES
108 static int compress_core(gzFile, char *, char *, unsigned int,
111 #define CORE_BUF_SIZE (16 * 1024)
113 int __elfN(fallback_brand) = -1;
114 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
115 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
116 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
117 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
118 &__elfN(fallback_brand));
120 static int elf_legacy_coredump = 0;
121 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
122 &elf_legacy_coredump, 0, "");
124 int __elfN(nxstack) =
125 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
130 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
131 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
132 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
134 #if __ELF_WORD_SIZE == 32
135 #if defined(__amd64__) || defined(__ia64__)
136 int i386_read_exec = 0;
137 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
138 "enable execution from readable segments");
142 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
144 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
145 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
146 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
148 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
150 Elf_Brandnote __elfN(freebsd_brandnote) = {
151 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
152 .hdr.n_descsz = sizeof(int32_t),
154 .vendor = FREEBSD_ABI_VENDOR,
155 .flags = BN_TRANSLATE_OSREL,
156 .trans_osrel = __elfN(freebsd_trans_osrel)
160 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
164 p = (uintptr_t)(note + 1);
165 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
166 *osrel = *(const int32_t *)(p);
171 static const char GNU_ABI_VENDOR[] = "GNU";
172 static int GNU_KFREEBSD_ABI_DESC = 3;
174 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
175 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
176 .hdr.n_descsz = 16, /* XXX at least 16 */
178 .vendor = GNU_ABI_VENDOR,
179 .flags = BN_TRANSLATE_OSREL,
180 .trans_osrel = kfreebsd_trans_osrel
184 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
186 const Elf32_Word *desc;
189 p = (uintptr_t)(note + 1);
190 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
192 desc = (const Elf32_Word *)p;
193 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
197 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
198 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
200 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
206 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
210 for (i = 0; i < MAX_BRANDS; i++) {
211 if (elf_brand_list[i] == NULL) {
212 elf_brand_list[i] = entry;
216 if (i == MAX_BRANDS) {
217 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
225 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
229 for (i = 0; i < MAX_BRANDS; i++) {
230 if (elf_brand_list[i] == entry) {
231 elf_brand_list[i] = NULL;
241 __elfN(brand_inuse)(Elf_Brandinfo *entry)
246 sx_slock(&allproc_lock);
247 FOREACH_PROC_IN_SYSTEM(p) {
248 if (p->p_sysent == entry->sysvec) {
253 sx_sunlock(&allproc_lock);
258 static Elf_Brandinfo *
259 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
260 int interp_name_len, int32_t *osrel)
262 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
268 * We support four types of branding -- (1) the ELF EI_OSABI field
269 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
270 * branding w/in the ELF header, (3) path of the `interp_path'
271 * field, and (4) the ".note.ABI-tag" ELF section.
274 /* Look for an ".note.ABI-tag" ELF section */
275 for (i = 0; i < MAX_BRANDS; i++) {
276 bi = elf_brand_list[i];
279 if (hdr->e_machine == bi->machine && (bi->flags &
280 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
281 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
287 /* If the executable has a brand, search for it in the brand list. */
288 for (i = 0; i < MAX_BRANDS; i++) {
289 bi = elf_brand_list[i];
290 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
292 if (hdr->e_machine == bi->machine &&
293 (hdr->e_ident[EI_OSABI] == bi->brand ||
294 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
295 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
299 /* Lacking a known brand, search for a recognized interpreter. */
300 if (interp != NULL) {
301 for (i = 0; i < MAX_BRANDS; i++) {
302 bi = elf_brand_list[i];
303 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
305 if (hdr->e_machine == bi->machine &&
306 /* ELF image p_filesz includes terminating zero */
307 strlen(bi->interp_path) + 1 == interp_name_len &&
308 strncmp(interp, bi->interp_path, interp_name_len)
314 /* Lacking a recognized interpreter, try the default brand */
315 for (i = 0; i < MAX_BRANDS; i++) {
316 bi = elf_brand_list[i];
317 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
319 if (hdr->e_machine == bi->machine &&
320 __elfN(fallback_brand) == bi->brand)
327 __elfN(check_header)(const Elf_Ehdr *hdr)
333 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
334 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
335 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
336 hdr->e_phentsize != sizeof(Elf_Phdr) ||
337 hdr->e_version != ELF_TARG_VER)
341 * Make sure we have at least one brand for this machine.
344 for (i = 0; i < MAX_BRANDS; i++) {
345 bi = elf_brand_list[i];
346 if (bi != NULL && bi->machine == hdr->e_machine)
356 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
357 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
364 * Create the page if it doesn't exist yet. Ignore errors.
367 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
368 VM_PROT_ALL, VM_PROT_ALL, 0);
372 * Find the page from the underlying object.
375 sf = vm_imgact_map_page(object, offset);
377 return (KERN_FAILURE);
378 off = offset - trunc_page(offset);
379 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
381 vm_imgact_unmap_page(sf);
383 return (KERN_FAILURE);
387 return (KERN_SUCCESS);
391 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
392 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
399 if (start != trunc_page(start)) {
400 rv = __elfN(map_partial)(map, object, offset, start,
401 round_page(start), prot);
404 offset += round_page(start) - start;
405 start = round_page(start);
407 if (end != round_page(end)) {
408 rv = __elfN(map_partial)(map, object, offset +
409 trunc_page(end) - start, trunc_page(end), end, prot);
412 end = trunc_page(end);
415 if (offset & PAGE_MASK) {
417 * The mapping is not page aligned. This means we have
418 * to copy the data. Sigh.
420 rv = vm_map_find(map, NULL, 0, &start, end - start,
421 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
425 return (KERN_SUCCESS);
426 for (; start < end; start += sz) {
427 sf = vm_imgact_map_page(object, offset);
429 return (KERN_FAILURE);
430 off = offset - trunc_page(offset);
432 if (sz > PAGE_SIZE - off)
433 sz = PAGE_SIZE - off;
434 error = copyout((caddr_t)sf_buf_kva(sf) + off,
436 vm_imgact_unmap_page(sf);
438 return (KERN_FAILURE);
444 vm_object_reference(object);
446 rv = vm_map_insert(map, object, offset, start, end,
447 prot, VM_PROT_ALL, cow);
449 if (rv != KERN_SUCCESS)
450 vm_object_deallocate(object);
454 return (KERN_SUCCESS);
459 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
460 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
467 vm_offset_t map_addr;
470 vm_offset_t file_addr;
473 * It's necessary to fail if the filsz + offset taken from the
474 * header is greater than the actual file pager object's size.
475 * If we were to allow this, then the vm_map_find() below would
476 * walk right off the end of the file object and into the ether.
478 * While I'm here, might as well check for something else that
479 * is invalid: filsz cannot be greater than memsz.
481 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
482 uprintf("elf_load_section: truncated ELF file\n");
486 object = imgp->object;
487 map = &imgp->proc->p_vmspace->vm_map;
488 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
489 file_addr = trunc_page_ps(offset, pagesize);
492 * We have two choices. We can either clear the data in the last page
493 * of an oversized mapping, or we can start the anon mapping a page
494 * early and copy the initialized data into that first page. We
495 * choose the second..
498 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
500 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
503 /* cow flags: don't dump readonly sections in core */
504 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
505 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
507 rv = __elfN(map_insert)(map,
509 file_addr, /* file offset */
510 map_addr, /* virtual start */
511 map_addr + map_len,/* virtual end */
514 if (rv != KERN_SUCCESS)
517 /* we can stop now if we've covered it all */
518 if (memsz == filsz) {
525 * We have to get the remaining bit of the file into the first part
526 * of the oversized map segment. This is normally because the .data
527 * segment in the file is extended to provide bss. It's a neat idea
528 * to try and save a page, but it's a pain in the behind to implement.
530 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
531 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
532 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
535 /* This had damn well better be true! */
537 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
538 map_len, VM_PROT_ALL, 0);
539 if (rv != KERN_SUCCESS) {
547 sf = vm_imgact_map_page(object, offset + filsz);
551 /* send the page fragment to user space */
552 off = trunc_page_ps(offset + filsz, pagesize) -
553 trunc_page(offset + filsz);
554 error = copyout((caddr_t)sf_buf_kva(sf) + off,
555 (caddr_t)map_addr, copy_len);
556 vm_imgact_unmap_page(sf);
563 * set it to the specified protection.
564 * XXX had better undo the damage from pasting over the cracks here!
566 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
567 map_len), prot, FALSE);
573 * Load the file "file" into memory. It may be either a shared object
576 * The "addr" reference parameter is in/out. On entry, it specifies
577 * the address where a shared object should be loaded. If the file is
578 * an executable, this value is ignored. On exit, "addr" specifies
579 * where the file was actually loaded.
581 * The "entry" reference parameter is out only. On exit, it specifies
582 * the entry point for the loaded file.
585 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
586 u_long *entry, size_t pagesize)
591 struct image_params image_params;
593 const Elf_Ehdr *hdr = NULL;
594 const Elf_Phdr *phdr = NULL;
595 struct nameidata *nd;
597 struct image_params *imgp;
600 u_long base_addr = 0;
601 int error, i, numsegs;
603 #ifdef CAPABILITY_MODE
605 * XXXJA: This check can go away once we are sufficiently confident
606 * that the checks in namei() are correct.
608 if (IN_CAPABILITY_MODE(curthread))
612 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
614 attr = &tempdata->attr;
615 imgp = &tempdata->image_params;
618 * Initialize part of the common data
622 imgp->firstpage = NULL;
623 imgp->image_header = NULL;
625 imgp->execlabel = NULL;
627 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
628 if ((error = namei(nd)) != 0) {
632 NDFREE(nd, NDF_ONLY_PNBUF);
633 imgp->vp = nd->ni_vp;
636 * Check permissions, modes, uid, etc on the file, and "open" it.
638 error = exec_check_permissions(imgp);
642 error = exec_map_first_page(imgp);
647 * Also make certain that the interpreter stays the same, so set
648 * its VV_TEXT flag, too.
650 VOP_SET_TEXT(nd->ni_vp);
652 imgp->object = nd->ni_vp->v_object;
654 hdr = (const Elf_Ehdr *)imgp->image_header;
655 if ((error = __elfN(check_header)(hdr)) != 0)
657 if (hdr->e_type == ET_DYN)
659 else if (hdr->e_type == ET_EXEC)
666 /* Only support headers that fit within first page for now */
667 if ((hdr->e_phoff > PAGE_SIZE) ||
668 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
673 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
674 if (!aligned(phdr, Elf_Addr)) {
679 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
680 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
681 /* Loadable segment */
682 prot = __elfN(trans_prot)(phdr[i].p_flags);
683 error = __elfN(load_section)(imgp, phdr[i].p_offset,
684 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
685 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
689 * Establish the base address if this is the
693 base_addr = trunc_page(phdr[i].p_vaddr +
699 *entry = (unsigned long)hdr->e_entry + rbase;
703 exec_unmap_first_page(imgp);
708 free(tempdata, M_TEMP);
714 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
716 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
717 const Elf_Phdr *phdr;
718 Elf_Auxargs *elf_auxargs;
719 struct vmspace *vmspace;
721 u_long text_size = 0, data_size = 0, total_size = 0;
722 u_long text_addr = 0, data_addr = 0;
723 u_long seg_size, seg_addr;
724 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
726 int error = 0, i, n, interp_name_len = 0;
727 const char *interp = NULL, *newinterp = NULL;
728 Elf_Brandinfo *brand_info;
730 struct sysentvec *sv;
733 * Do we have a valid ELF header ?
735 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
736 * if particular brand doesn't support it.
738 if (__elfN(check_header)(hdr) != 0 ||
739 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
743 * From here on down, we return an errno, not -1, as we've
744 * detected an ELF file.
747 if ((hdr->e_phoff > PAGE_SIZE) ||
748 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
749 /* Only support headers in first page for now */
752 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
753 if (!aligned(phdr, Elf_Addr))
757 for (i = 0; i < hdr->e_phnum; i++) {
758 switch (phdr[i].p_type) {
761 baddr = phdr[i].p_vaddr;
765 /* Path to interpreter */
766 if (phdr[i].p_filesz > MAXPATHLEN ||
767 phdr[i].p_offset > PAGE_SIZE ||
768 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
770 interp = imgp->image_header + phdr[i].p_offset;
771 interp_name_len = phdr[i].p_filesz;
776 __elfN(trans_prot)(phdr[i].p_flags);
781 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
783 if (brand_info == NULL) {
784 uprintf("ELF binary type \"%u\" not known.\n",
785 hdr->e_ident[EI_OSABI]);
788 if (hdr->e_type == ET_DYN) {
789 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
792 * Honour the base load address from the dso if it is
793 * non-zero for some reason.
796 et_dyn_addr = ET_DYN_LOAD_ADDR;
801 sv = brand_info->sysvec;
802 if (interp != NULL && brand_info->interp_newpath != NULL)
803 newinterp = brand_info->interp_newpath;
806 * Avoid a possible deadlock if the current address space is destroyed
807 * and that address space maps the locked vnode. In the common case,
808 * the locked vnode's v_usecount is decremented but remains greater
809 * than zero. Consequently, the vnode lock is not needed by vrele().
810 * However, in cases where the vnode lock is external, such as nullfs,
811 * v_usecount may become zero.
813 * The VV_TEXT flag prevents modifications to the executable while
814 * the vnode is unlocked.
816 VOP_UNLOCK(imgp->vp, 0);
818 error = exec_new_vmspace(imgp, sv);
819 imgp->proc->p_sysent = sv;
821 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
825 for (i = 0; i < hdr->e_phnum; i++) {
826 switch (phdr[i].p_type) {
827 case PT_LOAD: /* Loadable segment */
828 if (phdr[i].p_memsz == 0)
830 prot = __elfN(trans_prot)(phdr[i].p_flags);
831 error = __elfN(load_section)(imgp, phdr[i].p_offset,
832 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
833 phdr[i].p_memsz, phdr[i].p_filesz, prot,
839 * If this segment contains the program headers,
840 * remember their virtual address for the AT_PHDR
841 * aux entry. Static binaries don't usually include
844 if (phdr[i].p_offset == 0 &&
845 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
847 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
850 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
851 seg_size = round_page(phdr[i].p_memsz +
852 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
855 * Make the largest executable segment the official
856 * text segment and all others data.
858 * Note that obreak() assumes that data_addr +
859 * data_size == end of data load area, and the ELF
860 * file format expects segments to be sorted by
861 * address. If multiple data segments exist, the
862 * last one will be used.
865 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
866 text_size = seg_size;
867 text_addr = seg_addr;
869 data_size = seg_size;
870 data_addr = seg_addr;
872 total_size += seg_size;
874 case PT_PHDR: /* Program header table info */
875 proghdr = phdr[i].p_vaddr + et_dyn_addr;
882 if (data_addr == 0 && data_size == 0) {
883 data_addr = text_addr;
884 data_size = text_size;
887 entry = (u_long)hdr->e_entry + et_dyn_addr;
890 * Check limits. It should be safe to check the
891 * limits after loading the segments since we do
892 * not actually fault in all the segments pages.
894 PROC_LOCK(imgp->proc);
895 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
896 text_size > maxtsiz ||
897 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
898 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
899 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
900 PROC_UNLOCK(imgp->proc);
904 vmspace = imgp->proc->p_vmspace;
905 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
906 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
907 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
908 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
911 * We load the dynamic linker where a userland call
912 * to mmap(0, ...) would put it. The rationale behind this
913 * calculation is that it leaves room for the heap to grow to
914 * its maximum allowed size.
916 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
918 PROC_UNLOCK(imgp->proc);
920 imgp->entry_addr = entry;
922 if (interp != NULL) {
923 int have_interp = FALSE;
924 VOP_UNLOCK(imgp->vp, 0);
925 if (brand_info->emul_path != NULL &&
926 brand_info->emul_path[0] != '\0') {
927 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
928 snprintf(path, MAXPATHLEN, "%s%s",
929 brand_info->emul_path, interp);
930 error = __elfN(load_file)(imgp->proc, path, &addr,
931 &imgp->entry_addr, sv->sv_pagesize);
936 if (!have_interp && newinterp != NULL) {
937 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
938 &imgp->entry_addr, sv->sv_pagesize);
943 error = __elfN(load_file)(imgp->proc, interp, &addr,
944 &imgp->entry_addr, sv->sv_pagesize);
946 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
948 uprintf("ELF interpreter %s not found\n", interp);
955 * Construct auxargs table (used by the fixup routine)
957 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
958 elf_auxargs->execfd = -1;
959 elf_auxargs->phdr = proghdr;
960 elf_auxargs->phent = hdr->e_phentsize;
961 elf_auxargs->phnum = hdr->e_phnum;
962 elf_auxargs->pagesz = PAGE_SIZE;
963 elf_auxargs->base = addr;
964 elf_auxargs->flags = 0;
965 elf_auxargs->entry = entry;
967 imgp->auxargs = elf_auxargs;
968 imgp->interpreted = 0;
969 imgp->reloc_base = addr;
970 imgp->proc->p_osrel = osrel;
975 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
978 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
980 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
984 base = (Elf_Addr *)*stack_base;
985 pos = base + (imgp->args->argc + imgp->args->envc + 2);
987 if (args->execfd != -1)
988 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
989 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
990 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
991 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
992 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
993 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
994 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
995 AUXARGS_ENTRY(pos, AT_BASE, args->base);
996 if (imgp->execpathp != 0)
997 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
998 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
999 if (imgp->canary != 0) {
1000 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1001 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1003 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1004 if (imgp->pagesizes != 0) {
1005 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1006 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1008 if (imgp->sysent->sv_timekeep_base != 0) {
1009 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1010 imgp->sysent->sv_timekeep_base);
1012 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1013 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1014 imgp->sysent->sv_stackprot);
1015 AUXARGS_ENTRY(pos, AT_NULL, 0);
1017 free(imgp->auxargs, M_TEMP);
1018 imgp->auxargs = NULL;
1021 suword(base, (long)imgp->args->argc);
1022 *stack_base = (register_t *)base;
1027 * Code for generating ELF core dumps.
1030 typedef void (*segment_callback)(vm_map_entry_t, void *);
1032 /* Closure for cb_put_phdr(). */
1033 struct phdr_closure {
1034 Elf_Phdr *phdr; /* Program header to fill in */
1035 Elf_Off offset; /* Offset of segment in core file */
1038 /* Closure for cb_size_segment(). */
1039 struct sseg_closure {
1040 int count; /* Count of writable segments. */
1041 size_t size; /* Total size of all writable segments. */
1044 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1047 int type; /* Note type. */
1048 outfunc_t outfunc; /* Output function. */
1049 void *outarg; /* Argument for the output function. */
1050 size_t outsize; /* Output size. */
1051 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1054 TAILQ_HEAD(note_info_list, note_info);
1056 static void cb_put_phdr(vm_map_entry_t, void *);
1057 static void cb_size_segment(vm_map_entry_t, void *);
1058 static void each_writable_segment(struct thread *, segment_callback, void *);
1059 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1060 int, void *, size_t, struct note_info_list *, size_t, gzFile);
1061 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1063 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1064 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1065 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1066 static int sbuf_drain_core_output(void *, const char *, int);
1067 static int sbuf_drain_count(void *arg, const char *data, int len);
1069 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1070 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1071 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1072 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1073 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1074 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1075 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1076 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1077 static void note_procstat_files(void *, struct sbuf *, size_t *);
1078 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1079 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1080 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1081 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1082 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1084 #ifdef COMPRESS_USER_CORES
1085 extern int compress_user_cores;
1086 extern int compress_user_cores_gzlevel;
1090 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1091 struct ucred *active_cred, struct ucred *file_cred,
1092 struct thread *td, char *core_buf, gzFile gzfile) {
1096 #ifdef COMPRESS_USER_CORES
1097 error = compress_core(gzfile, base, core_buf, len, td);
1099 panic("shouldn't be here");
1102 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1103 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1109 /* Coredump output parameters for sbuf drain routine. */
1110 struct sbuf_drain_core_params {
1112 struct ucred *active_cred;
1113 struct ucred *file_cred;
1116 #ifdef COMPRESS_USER_CORES
1122 * Drain into a core file.
1125 sbuf_drain_core_output(void *arg, const char *data, int len)
1127 struct sbuf_drain_core_params *p;
1130 p = (struct sbuf_drain_core_params *)arg;
1133 * Some kern_proc out routines that print to this sbuf may
1134 * call us with the process lock held. Draining with the
1135 * non-sleepable lock held is unsafe. The lock is needed for
1136 * those routines when dumping a live process. In our case we
1137 * can safely release the lock before draining and acquire
1140 locked = PROC_LOCKED(p->td->td_proc);
1142 PROC_UNLOCK(p->td->td_proc);
1143 #ifdef COMPRESS_USER_CORES
1144 if (p->gzfile != Z_NULL)
1145 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1149 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1150 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1151 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1154 PROC_LOCK(p->td->td_proc);
1162 * Drain into a counter.
1165 sbuf_drain_count(void *arg, const char *data __unused, int len)
1169 sizep = (size_t *)arg;
1175 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1177 struct ucred *cred = td->td_ucred;
1179 struct sseg_closure seginfo;
1180 struct note_info_list notelst;
1181 struct note_info *ninfo;
1183 size_t hdrsize, notesz, coresize;
1185 gzFile gzfile = Z_NULL;
1186 char *core_buf = NULL;
1187 #ifdef COMPRESS_USER_CORES
1188 char gzopen_flags[8];
1190 int doing_compress = flags & IMGACT_CORE_COMPRESS;
1194 TAILQ_INIT(¬elst);
1196 #ifdef COMPRESS_USER_CORES
1197 if (doing_compress) {
1200 if (compress_user_cores_gzlevel >= 0 &&
1201 compress_user_cores_gzlevel <= 9)
1202 *p++ = '0' + compress_user_cores_gzlevel;
1204 gzfile = gz_open("", gzopen_flags, vp);
1205 if (gzfile == Z_NULL) {
1209 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1217 /* Size the program segments. */
1220 each_writable_segment(td, cb_size_segment, &seginfo);
1223 * Collect info about the core file header area.
1225 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1226 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1227 coresize = round_page(hdrsize + notesz) + seginfo.size;
1230 PROC_LOCK(td->td_proc);
1231 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1232 PROC_UNLOCK(td->td_proc);
1238 if (coresize >= limit) {
1244 * Allocate memory for building the header, fill it up,
1245 * and write it out following the notes.
1247 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1252 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1253 ¬elst, notesz, gzfile);
1255 /* Write the contents of all of the writable segments. */
1261 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1262 offset = round_page(hdrsize + notesz);
1263 for (i = 0; i < seginfo.count; i++) {
1264 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1265 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1268 offset += php->p_filesz;
1274 "Failed to write core file for process %s (error %d)\n",
1275 curproc->p_comm, error);
1279 #ifdef COMPRESS_USER_CORES
1281 free(core_buf, M_TEMP);
1285 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1286 TAILQ_REMOVE(¬elst, ninfo, link);
1287 free(ninfo, M_TEMP);
1296 * A callback for each_writable_segment() to write out the segment's
1297 * program header entry.
1300 cb_put_phdr(entry, closure)
1301 vm_map_entry_t entry;
1304 struct phdr_closure *phc = (struct phdr_closure *)closure;
1305 Elf_Phdr *phdr = phc->phdr;
1307 phc->offset = round_page(phc->offset);
1309 phdr->p_type = PT_LOAD;
1310 phdr->p_offset = phc->offset;
1311 phdr->p_vaddr = entry->start;
1313 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1314 phdr->p_align = PAGE_SIZE;
1315 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1317 phc->offset += phdr->p_filesz;
1322 * A callback for each_writable_segment() to gather information about
1323 * the number of segments and their total size.
1326 cb_size_segment(entry, closure)
1327 vm_map_entry_t entry;
1330 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1333 ssc->size += entry->end - entry->start;
1337 * For each writable segment in the process's memory map, call the given
1338 * function with a pointer to the map entry and some arbitrary
1339 * caller-supplied data.
1342 each_writable_segment(td, func, closure)
1344 segment_callback func;
1347 struct proc *p = td->td_proc;
1348 vm_map_t map = &p->p_vmspace->vm_map;
1349 vm_map_entry_t entry;
1350 vm_object_t backing_object, object;
1351 boolean_t ignore_entry;
1353 vm_map_lock_read(map);
1354 for (entry = map->header.next; entry != &map->header;
1355 entry = entry->next) {
1357 * Don't dump inaccessible mappings, deal with legacy
1360 * Note that read-only segments related to the elf binary
1361 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1362 * need to arbitrarily ignore such segments.
1364 if (elf_legacy_coredump) {
1365 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1368 if ((entry->protection & VM_PROT_ALL) == 0)
1373 * Dont include memory segment in the coredump if
1374 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1375 * madvise(2). Do not dump submaps (i.e. parts of the
1378 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1381 if ((object = entry->object.vm_object) == NULL)
1384 /* Ignore memory-mapped devices and such things. */
1385 VM_OBJECT_RLOCK(object);
1386 while ((backing_object = object->backing_object) != NULL) {
1387 VM_OBJECT_RLOCK(backing_object);
1388 VM_OBJECT_RUNLOCK(object);
1389 object = backing_object;
1391 ignore_entry = object->type != OBJT_DEFAULT &&
1392 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1393 VM_OBJECT_RUNLOCK(object);
1397 (*func)(entry, closure);
1399 vm_map_unlock_read(map);
1403 * Write the core file header to the file, including padding up to
1404 * the page boundary.
1407 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1408 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1409 size_t notesz, gzFile gzfile)
1411 struct sbuf_drain_core_params params;
1412 struct note_info *ninfo;
1416 /* Fill in the header. */
1417 bzero(hdr, hdrsize);
1418 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1421 params.active_cred = cred;
1422 params.file_cred = NOCRED;
1425 #ifdef COMPRESS_USER_CORES
1426 params.gzfile = gzfile;
1428 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1429 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms);
1430 sbuf_start_section(sb, NULL);
1431 sbuf_bcat(sb, hdr, hdrsize);
1432 TAILQ_FOREACH(ninfo, notelst, link)
1433 __elfN(putnote)(ninfo, sb);
1434 /* Align up to a page boundary for the program segments. */
1435 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1436 error = sbuf_finish(sb);
1443 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1453 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1456 * To have the debugger select the right thread (LWP) as the initial
1457 * thread, we dump the state of the thread passed to us in td first.
1458 * This is the thread that causes the core dump and thus likely to
1459 * be the right thread one wants to have selected in the debugger.
1462 while (thr != NULL) {
1463 size += register_note(list, NT_PRSTATUS,
1464 __elfN(note_prstatus), thr);
1465 size += register_note(list, NT_FPREGSET,
1466 __elfN(note_fpregset), thr);
1467 size += register_note(list, NT_THRMISC,
1468 __elfN(note_thrmisc), thr);
1469 size += register_note(list, -1,
1470 __elfN(note_threadmd), thr);
1472 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1473 TAILQ_NEXT(thr, td_plist);
1475 thr = TAILQ_NEXT(thr, td_plist);
1478 size += register_note(list, NT_PROCSTAT_PROC,
1479 __elfN(note_procstat_proc), p);
1480 size += register_note(list, NT_PROCSTAT_FILES,
1481 note_procstat_files, p);
1482 size += register_note(list, NT_PROCSTAT_VMMAP,
1483 note_procstat_vmmap, p);
1484 size += register_note(list, NT_PROCSTAT_GROUPS,
1485 note_procstat_groups, p);
1486 size += register_note(list, NT_PROCSTAT_UMASK,
1487 note_procstat_umask, p);
1488 size += register_note(list, NT_PROCSTAT_RLIMIT,
1489 note_procstat_rlimit, p);
1490 size += register_note(list, NT_PROCSTAT_OSREL,
1491 note_procstat_osrel, p);
1492 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1493 __elfN(note_procstat_psstrings), p);
1494 size += register_note(list, NT_PROCSTAT_AUXV,
1495 __elfN(note_procstat_auxv), p);
1501 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1506 struct phdr_closure phc;
1508 ehdr = (Elf_Ehdr *)hdr;
1509 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1511 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1512 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1513 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1514 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1515 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1516 ehdr->e_ident[EI_DATA] = ELF_DATA;
1517 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1518 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1519 ehdr->e_ident[EI_ABIVERSION] = 0;
1520 ehdr->e_ident[EI_PAD] = 0;
1521 ehdr->e_type = ET_CORE;
1522 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1523 ehdr->e_machine = ELF_ARCH32;
1525 ehdr->e_machine = ELF_ARCH;
1527 ehdr->e_version = EV_CURRENT;
1529 ehdr->e_phoff = sizeof(Elf_Ehdr);
1531 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1532 ehdr->e_phentsize = sizeof(Elf_Phdr);
1533 ehdr->e_phnum = numsegs + 1;
1534 ehdr->e_shentsize = sizeof(Elf_Shdr);
1536 ehdr->e_shstrndx = SHN_UNDEF;
1539 * Fill in the program header entries.
1542 /* The note segement. */
1543 phdr->p_type = PT_NOTE;
1544 phdr->p_offset = hdrsize;
1547 phdr->p_filesz = notesz;
1549 phdr->p_flags = PF_R;
1550 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1553 /* All the writable segments from the program. */
1555 phc.offset = round_page(hdrsize + notesz);
1556 each_writable_segment(td, cb_put_phdr, &phc);
1560 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1562 struct note_info *ninfo;
1563 size_t size, notesize;
1566 out(arg, NULL, &size);
1567 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1569 ninfo->outfunc = out;
1570 ninfo->outarg = arg;
1571 ninfo->outsize = size;
1572 TAILQ_INSERT_TAIL(list, ninfo, link);
1577 notesize = sizeof(Elf_Note) + /* note header */
1578 roundup2(8, ELF_NOTE_ROUNDSIZE) + /* note name ("FreeBSD") */
1579 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1585 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1590 if (ninfo->type == -1) {
1591 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1595 note.n_namesz = 8; /* strlen("FreeBSD") + 1 */
1596 note.n_descsz = ninfo->outsize;
1597 note.n_type = ninfo->type;
1599 sbuf_bcat(sb, ¬e, sizeof(note));
1600 sbuf_start_section(sb, &old_len);
1601 sbuf_bcat(sb, "FreeBSD", note.n_namesz);
1602 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1603 if (note.n_descsz == 0)
1605 sbuf_start_section(sb, &old_len);
1606 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1607 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1611 * Miscellaneous note out functions.
1614 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1615 #include <compat/freebsd32/freebsd32.h>
1617 typedef struct prstatus32 elf_prstatus_t;
1618 typedef struct prpsinfo32 elf_prpsinfo_t;
1619 typedef struct fpreg32 elf_prfpregset_t;
1620 typedef struct fpreg32 elf_fpregset_t;
1621 typedef struct reg32 elf_gregset_t;
1622 typedef struct thrmisc32 elf_thrmisc_t;
1623 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1624 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1625 typedef uint32_t elf_ps_strings_t;
1627 typedef prstatus_t elf_prstatus_t;
1628 typedef prpsinfo_t elf_prpsinfo_t;
1629 typedef prfpregset_t elf_prfpregset_t;
1630 typedef prfpregset_t elf_fpregset_t;
1631 typedef gregset_t elf_gregset_t;
1632 typedef thrmisc_t elf_thrmisc_t;
1633 #define ELF_KERN_PROC_MASK 0
1634 typedef struct kinfo_proc elf_kinfo_proc_t;
1635 typedef vm_offset_t elf_ps_strings_t;
1639 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1642 elf_prpsinfo_t *psinfo;
1644 p = (struct proc *)arg;
1646 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1647 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1648 psinfo->pr_version = PRPSINFO_VERSION;
1649 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1650 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1652 * XXX - We don't fill in the command line arguments properly
1655 strlcpy(psinfo->pr_psargs, p->p_comm,
1656 sizeof(psinfo->pr_psargs));
1658 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1659 free(psinfo, M_TEMP);
1661 *sizep = sizeof(*psinfo);
1665 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1668 elf_prstatus_t *status;
1670 td = (struct thread *)arg;
1672 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1673 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1674 status->pr_version = PRSTATUS_VERSION;
1675 status->pr_statussz = sizeof(elf_prstatus_t);
1676 status->pr_gregsetsz = sizeof(elf_gregset_t);
1677 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1678 status->pr_osreldate = osreldate;
1679 status->pr_cursig = td->td_proc->p_sig;
1680 status->pr_pid = td->td_tid;
1681 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1682 fill_regs32(td, &status->pr_reg);
1684 fill_regs(td, &status->pr_reg);
1686 sbuf_bcat(sb, status, sizeof(*status));
1687 free(status, M_TEMP);
1689 *sizep = sizeof(*status);
1693 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1696 elf_prfpregset_t *fpregset;
1698 td = (struct thread *)arg;
1700 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1701 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1702 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1703 fill_fpregs32(td, fpregset);
1705 fill_fpregs(td, fpregset);
1707 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1708 free(fpregset, M_TEMP);
1710 *sizep = sizeof(*fpregset);
1714 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1717 elf_thrmisc_t thrmisc;
1719 td = (struct thread *)arg;
1721 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1722 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1723 strcpy(thrmisc.pr_tname, td->td_name);
1724 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1726 *sizep = sizeof(thrmisc);
1730 * Allow for MD specific notes, as well as any MD
1731 * specific preparations for writing MI notes.
1734 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1740 td = (struct thread *)arg;
1743 if (size != 0 && sb != NULL)
1744 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1746 __elfN(dump_thread)(td, buf, &size);
1747 KASSERT(*sizep == size, ("invalid size"));
1748 if (size != 0 && sb != NULL)
1749 sbuf_bcat(sb, buf, size);
1753 #ifdef KINFO_PROC_SIZE
1754 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1758 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1764 p = (struct proc *)arg;
1765 size = sizeof(structsize) + p->p_numthreads *
1766 sizeof(elf_kinfo_proc_t);
1769 KASSERT(*sizep == size, ("invalid size"));
1770 structsize = sizeof(elf_kinfo_proc_t);
1771 sbuf_bcat(sb, &structsize, sizeof(structsize));
1773 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1778 #ifdef KINFO_FILE_SIZE
1779 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1783 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1789 p = (struct proc *)arg;
1792 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1793 sbuf_set_drain(sb, sbuf_drain_count, &size);
1794 sbuf_bcat(sb, &structsize, sizeof(structsize));
1796 kern_proc_filedesc_out(p, sb, -1);
1801 structsize = sizeof(struct kinfo_file);
1802 sbuf_bcat(sb, &structsize, sizeof(structsize));
1804 kern_proc_filedesc_out(p, sb, -1);
1808 #ifdef KINFO_VMENTRY_SIZE
1809 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1813 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1819 p = (struct proc *)arg;
1822 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1823 sbuf_set_drain(sb, sbuf_drain_count, &size);
1824 sbuf_bcat(sb, &structsize, sizeof(structsize));
1826 kern_proc_vmmap_out(p, sb);
1831 structsize = sizeof(struct kinfo_vmentry);
1832 sbuf_bcat(sb, &structsize, sizeof(structsize));
1834 kern_proc_vmmap_out(p, sb);
1839 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1845 p = (struct proc *)arg;
1846 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1848 KASSERT(*sizep == size, ("invalid size"));
1849 structsize = sizeof(gid_t);
1850 sbuf_bcat(sb, &structsize, sizeof(structsize));
1851 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1858 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1864 p = (struct proc *)arg;
1865 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1867 KASSERT(*sizep == size, ("invalid size"));
1868 structsize = sizeof(p->p_fd->fd_cmask);
1869 sbuf_bcat(sb, &structsize, sizeof(structsize));
1870 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1876 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1879 struct rlimit rlim[RLIM_NLIMITS];
1883 p = (struct proc *)arg;
1884 size = sizeof(structsize) + sizeof(rlim);
1886 KASSERT(*sizep == size, ("invalid size"));
1887 structsize = sizeof(rlim);
1888 sbuf_bcat(sb, &structsize, sizeof(structsize));
1890 for (i = 0; i < RLIM_NLIMITS; i++)
1891 lim_rlimit(p, i, &rlim[i]);
1893 sbuf_bcat(sb, rlim, sizeof(rlim));
1899 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1905 p = (struct proc *)arg;
1906 size = sizeof(structsize) + sizeof(p->p_osrel);
1908 KASSERT(*sizep == size, ("invalid size"));
1909 structsize = sizeof(p->p_osrel);
1910 sbuf_bcat(sb, &structsize, sizeof(structsize));
1911 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1917 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1920 elf_ps_strings_t ps_strings;
1924 p = (struct proc *)arg;
1925 size = sizeof(structsize) + sizeof(ps_strings);
1927 KASSERT(*sizep == size, ("invalid size"));
1928 structsize = sizeof(ps_strings);
1929 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1930 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1932 ps_strings = p->p_sysent->sv_psstrings;
1934 sbuf_bcat(sb, &structsize, sizeof(structsize));
1935 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1941 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1947 p = (struct proc *)arg;
1950 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1951 sbuf_set_drain(sb, sbuf_drain_count, &size);
1952 sbuf_bcat(sb, &structsize, sizeof(structsize));
1954 proc_getauxv(curthread, p, sb);
1960 structsize = sizeof(Elf_Auxinfo);
1961 sbuf_bcat(sb, &structsize, sizeof(structsize));
1963 proc_getauxv(curthread, p, sb);
1969 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1970 int32_t *osrel, const Elf_Phdr *pnote)
1972 const Elf_Note *note, *note0, *note_end;
1973 const char *note_name;
1976 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
1977 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
1980 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1981 note_end = (const Elf_Note *)(imgp->image_header +
1982 pnote->p_offset + pnote->p_filesz);
1983 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1984 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
1985 (const char *)note < sizeof(Elf_Note))
1987 if (note->n_namesz != checknote->hdr.n_namesz ||
1988 note->n_descsz != checknote->hdr.n_descsz ||
1989 note->n_type != checknote->hdr.n_type)
1991 note_name = (const char *)(note + 1);
1992 if (note_name + checknote->hdr.n_namesz >=
1993 (const char *)note_end || strncmp(checknote->vendor,
1994 note_name, checknote->hdr.n_namesz) != 0)
1998 * Fetch the osreldate for binary
1999 * from the ELF OSABI-note if necessary.
2001 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2002 checknote->trans_osrel != NULL)
2003 return (checknote->trans_osrel(note, osrel));
2007 note = (const Elf_Note *)((const char *)(note + 1) +
2008 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2009 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2016 * Try to find the appropriate ABI-note section for checknote,
2017 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2018 * first page of the image is searched, the same as for headers.
2021 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2024 const Elf_Phdr *phdr;
2025 const Elf_Ehdr *hdr;
2028 hdr = (const Elf_Ehdr *)imgp->image_header;
2029 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2031 for (i = 0; i < hdr->e_phnum; i++) {
2032 if (phdr[i].p_type == PT_NOTE &&
2033 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2041 * Tell kern_execve.c about it, with a little help from the linker.
2043 static struct execsw __elfN(execsw) = {
2044 __CONCAT(exec_, __elfN(imgact)),
2045 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2047 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2049 #ifdef COMPRESS_USER_CORES
2051 * Compress and write out a core segment for a user process.
2053 * 'inbuf' is the starting address of a VM segment in the process' address
2054 * space that is to be compressed and written out to the core file. 'dest_buf'
2055 * is a buffer in the kernel's address space. The segment is copied from
2056 * 'inbuf' to 'dest_buf' first before being processed by the compression
2057 * routine gzwrite(). This copying is necessary because the content of the VM
2058 * segment may change between the compression pass and the crc-computation pass
2059 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel.
2061 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2064 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2069 unsigned int chunk_len;
2072 if (inbuf != NULL) {
2073 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2074 copyin(inbuf, dest_buf, chunk_len);
2079 len_compressed = gzwrite(file, dest_buf, chunk_len);
2081 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2083 if ((unsigned int)len_compressed != chunk_len) {
2085 "compress_core: length mismatch (0x%x returned, "
2086 "0x%x expected)\n", len_compressed, chunk_len);
2087 EVENTHANDLER_INVOKE(app_coredump_error, td,
2088 "compress_core: length mismatch %x -> %x",
2089 chunk_len, len_compressed);
2099 #endif /* COMPRESS_USER_CORES */
2102 __elfN(trans_prot)(Elf_Word flags)
2108 prot |= VM_PROT_EXECUTE;
2110 prot |= VM_PROT_WRITE;
2112 prot |= VM_PROT_READ;
2113 #if __ELF_WORD_SIZE == 32
2114 #if defined(__amd64__) || defined(__ia64__)
2115 if (i386_read_exec && (flags & PF_R))
2116 prot |= VM_PROT_EXECUTE;
2123 __elfN(untrans_prot)(vm_prot_t prot)
2128 if (prot & VM_PROT_EXECUTE)
2130 if (prot & VM_PROT_READ)
2132 if (prot & VM_PROT_WRITE)