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>
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_RWTUN, &__elfN(fallback_brand), 0,
116 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
118 static int elf_legacy_coredump = 0;
119 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
120 &elf_legacy_coredump, 0, "");
122 int __elfN(nxstack) =
123 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __ELF_WORD_SIZE == 32
133 #if defined(__amd64__) || defined(__ia64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136 "enable execution from readable segments");
140 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
142 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
143 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
150 .hdr.n_descsz = sizeof(int32_t),
152 .vendor = FREEBSD_ABI_VENDOR,
153 .flags = BN_TRANSLATE_OSREL,
154 .trans_osrel = __elfN(freebsd_trans_osrel)
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 p = (uintptr_t)(note + 1);
163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 *osrel = *(const int32_t *)(p);
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
174 .hdr.n_descsz = 16, /* XXX at least 16 */
176 .vendor = GNU_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = kfreebsd_trans_osrel
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
184 const Elf32_Word *desc;
187 p = (uintptr_t)(note + 1);
188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
190 desc = (const Elf32_Word *)p;
191 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 for (i = 0; i < MAX_BRANDS; i++) {
209 if (elf_brand_list[i] == NULL) {
210 elf_brand_list[i] = entry;
214 if (i == MAX_BRANDS) {
215 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 for (i = 0; i < MAX_BRANDS; i++) {
228 if (elf_brand_list[i] == entry) {
229 elf_brand_list[i] = NULL;
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
244 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
246 if (p->p_sysent == entry->sysvec) {
251 sx_sunlock(&allproc_lock);
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258 int interp_name_len, int32_t *osrel)
260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
266 * We support four types of branding -- (1) the ELF EI_OSABI field
267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 * branding w/in the ELF header, (3) path of the `interp_path'
269 * field, and (4) the ".note.ABI-tag" ELF section.
272 /* Look for an ".note.ABI-tag" ELF section */
273 for (i = 0; i < MAX_BRANDS; i++) {
274 bi = elf_brand_list[i];
277 if (hdr->e_machine == bi->machine && (bi->flags &
278 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
279 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
285 /* If the executable has a brand, search for it in the brand list. */
286 for (i = 0; i < MAX_BRANDS; i++) {
287 bi = elf_brand_list[i];
288 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
290 if (hdr->e_machine == bi->machine &&
291 (hdr->e_ident[EI_OSABI] == bi->brand ||
292 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
293 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
297 /* Lacking a known brand, search for a recognized interpreter. */
298 if (interp != NULL) {
299 for (i = 0; i < MAX_BRANDS; i++) {
300 bi = elf_brand_list[i];
301 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
303 if (hdr->e_machine == bi->machine &&
304 /* ELF image p_filesz includes terminating zero */
305 strlen(bi->interp_path) + 1 == interp_name_len &&
306 strncmp(interp, bi->interp_path, interp_name_len)
312 /* Lacking a recognized interpreter, try the default brand */
313 for (i = 0; i < MAX_BRANDS; i++) {
314 bi = elf_brand_list[i];
315 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
317 if (hdr->e_machine == bi->machine &&
318 __elfN(fallback_brand) == bi->brand)
325 __elfN(check_header)(const Elf_Ehdr *hdr)
331 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
332 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
333 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
334 hdr->e_phentsize != sizeof(Elf_Phdr) ||
335 hdr->e_version != ELF_TARG_VER)
339 * Make sure we have at least one brand for this machine.
342 for (i = 0; i < MAX_BRANDS; i++) {
343 bi = elf_brand_list[i];
344 if (bi != NULL && bi->machine == hdr->e_machine)
354 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
355 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
362 * Create the page if it doesn't exist yet. Ignore errors.
365 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
366 VM_PROT_ALL, VM_PROT_ALL, 0);
370 * Find the page from the underlying object.
373 sf = vm_imgact_map_page(object, offset);
375 return (KERN_FAILURE);
376 off = offset - trunc_page(offset);
377 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
379 vm_imgact_unmap_page(sf);
381 return (KERN_FAILURE);
385 return (KERN_SUCCESS);
389 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
390 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
397 if (start != trunc_page(start)) {
398 rv = __elfN(map_partial)(map, object, offset, start,
399 round_page(start), prot);
402 offset += round_page(start) - start;
403 start = round_page(start);
405 if (end != round_page(end)) {
406 rv = __elfN(map_partial)(map, object, offset +
407 trunc_page(end) - start, trunc_page(end), end, prot);
410 end = trunc_page(end);
413 if (offset & PAGE_MASK) {
415 * The mapping is not page aligned. This means we have
416 * to copy the data. Sigh.
418 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
419 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
424 return (KERN_SUCCESS);
425 for (; start < end; start += sz) {
426 sf = vm_imgact_map_page(object, offset);
428 return (KERN_FAILURE);
429 off = offset - trunc_page(offset);
431 if (sz > PAGE_SIZE - off)
432 sz = PAGE_SIZE - off;
433 error = copyout((caddr_t)sf_buf_kva(sf) + off,
435 vm_imgact_unmap_page(sf);
437 return (KERN_FAILURE);
443 vm_object_reference(object);
445 rv = vm_map_insert(map, object, offset, start, end,
446 prot, VM_PROT_ALL, cow);
448 if (rv != KERN_SUCCESS)
449 vm_object_deallocate(object);
453 return (KERN_SUCCESS);
458 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
459 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
466 vm_offset_t map_addr;
469 vm_offset_t file_addr;
472 * It's necessary to fail if the filsz + offset taken from the
473 * header is greater than the actual file pager object's size.
474 * If we were to allow this, then the vm_map_find() below would
475 * walk right off the end of the file object and into the ether.
477 * While I'm here, might as well check for something else that
478 * is invalid: filsz cannot be greater than memsz.
480 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
481 uprintf("elf_load_section: truncated ELF file\n");
485 object = imgp->object;
486 map = &imgp->proc->p_vmspace->vm_map;
487 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
488 file_addr = trunc_page_ps(offset, pagesize);
491 * We have two choices. We can either clear the data in the last page
492 * of an oversized mapping, or we can start the anon mapping a page
493 * early and copy the initialized data into that first page. We
494 * choose the second..
497 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
499 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
502 /* cow flags: don't dump readonly sections in core */
503 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
504 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
506 rv = __elfN(map_insert)(map,
508 file_addr, /* file offset */
509 map_addr, /* virtual start */
510 map_addr + map_len,/* virtual end */
513 if (rv != KERN_SUCCESS)
516 /* we can stop now if we've covered it all */
517 if (memsz == filsz) {
524 * We have to get the remaining bit of the file into the first part
525 * of the oversized map segment. This is normally because the .data
526 * segment in the file is extended to provide bss. It's a neat idea
527 * to try and save a page, but it's a pain in the behind to implement.
529 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
530 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
531 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
534 /* This had damn well better be true! */
536 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
537 map_len, VM_PROT_ALL, 0);
538 if (rv != KERN_SUCCESS) {
546 sf = vm_imgact_map_page(object, offset + filsz);
550 /* send the page fragment to user space */
551 off = trunc_page_ps(offset + filsz, pagesize) -
552 trunc_page(offset + filsz);
553 error = copyout((caddr_t)sf_buf_kva(sf) + off,
554 (caddr_t)map_addr, copy_len);
555 vm_imgact_unmap_page(sf);
562 * set it to the specified protection.
563 * XXX had better undo the damage from pasting over the cracks here!
565 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
566 map_len), prot, FALSE);
572 * Load the file "file" into memory. It may be either a shared object
575 * The "addr" reference parameter is in/out. On entry, it specifies
576 * the address where a shared object should be loaded. If the file is
577 * an executable, this value is ignored. On exit, "addr" specifies
578 * where the file was actually loaded.
580 * The "entry" reference parameter is out only. On exit, it specifies
581 * the entry point for the loaded file.
584 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
585 u_long *entry, size_t pagesize)
590 struct image_params image_params;
592 const Elf_Ehdr *hdr = NULL;
593 const Elf_Phdr *phdr = NULL;
594 struct nameidata *nd;
596 struct image_params *imgp;
599 u_long base_addr = 0;
600 int error, i, numsegs;
602 #ifdef CAPABILITY_MODE
604 * XXXJA: This check can go away once we are sufficiently confident
605 * that the checks in namei() are correct.
607 if (IN_CAPABILITY_MODE(curthread))
611 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
613 attr = &tempdata->attr;
614 imgp = &tempdata->image_params;
617 * Initialize part of the common data
621 imgp->firstpage = NULL;
622 imgp->image_header = NULL;
624 imgp->execlabel = NULL;
626 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
627 if ((error = namei(nd)) != 0) {
631 NDFREE(nd, NDF_ONLY_PNBUF);
632 imgp->vp = nd->ni_vp;
635 * Check permissions, modes, uid, etc on the file, and "open" it.
637 error = exec_check_permissions(imgp);
641 error = exec_map_first_page(imgp);
646 * Also make certain that the interpreter stays the same, so set
647 * its VV_TEXT flag, too.
649 VOP_SET_TEXT(nd->ni_vp);
651 imgp->object = nd->ni_vp->v_object;
653 hdr = (const Elf_Ehdr *)imgp->image_header;
654 if ((error = __elfN(check_header)(hdr)) != 0)
656 if (hdr->e_type == ET_DYN)
658 else if (hdr->e_type == ET_EXEC)
665 /* Only support headers that fit within first page for now */
666 if ((hdr->e_phoff > PAGE_SIZE) ||
667 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
672 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
673 if (!aligned(phdr, Elf_Addr)) {
678 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
679 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
680 /* Loadable segment */
681 prot = __elfN(trans_prot)(phdr[i].p_flags);
682 error = __elfN(load_section)(imgp, phdr[i].p_offset,
683 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
684 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
688 * Establish the base address if this is the
692 base_addr = trunc_page(phdr[i].p_vaddr +
698 *entry = (unsigned long)hdr->e_entry + rbase;
702 exec_unmap_first_page(imgp);
707 free(tempdata, M_TEMP);
713 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
715 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
716 const Elf_Phdr *phdr;
717 Elf_Auxargs *elf_auxargs;
718 struct vmspace *vmspace;
720 u_long text_size = 0, data_size = 0, total_size = 0;
721 u_long text_addr = 0, data_addr = 0;
722 u_long seg_size, seg_addr;
723 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
725 int error = 0, i, n, interp_name_len = 0;
726 const char *interp = NULL, *newinterp = NULL;
727 Elf_Brandinfo *brand_info;
729 struct sysentvec *sv;
732 * Do we have a valid ELF header ?
734 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
735 * if particular brand doesn't support it.
737 if (__elfN(check_header)(hdr) != 0 ||
738 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
742 * From here on down, we return an errno, not -1, as we've
743 * detected an ELF file.
746 if ((hdr->e_phoff > PAGE_SIZE) ||
747 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
748 /* Only support headers in first page for now */
751 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
752 if (!aligned(phdr, Elf_Addr))
756 for (i = 0; i < hdr->e_phnum; i++) {
757 switch (phdr[i].p_type) {
760 baddr = phdr[i].p_vaddr;
764 /* Path to interpreter */
765 if (phdr[i].p_filesz > MAXPATHLEN ||
766 phdr[i].p_offset > PAGE_SIZE ||
767 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
769 interp = imgp->image_header + phdr[i].p_offset;
770 interp_name_len = phdr[i].p_filesz;
775 __elfN(trans_prot)(phdr[i].p_flags);
780 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
782 if (brand_info == NULL) {
783 uprintf("ELF binary type \"%u\" not known.\n",
784 hdr->e_ident[EI_OSABI]);
787 if (hdr->e_type == ET_DYN) {
788 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
791 * Honour the base load address from the dso if it is
792 * non-zero for some reason.
795 et_dyn_addr = ET_DYN_LOAD_ADDR;
800 sv = brand_info->sysvec;
801 if (interp != NULL && brand_info->interp_newpath != NULL)
802 newinterp = brand_info->interp_newpath;
805 * Avoid a possible deadlock if the current address space is destroyed
806 * and that address space maps the locked vnode. In the common case,
807 * the locked vnode's v_usecount is decremented but remains greater
808 * than zero. Consequently, the vnode lock is not needed by vrele().
809 * However, in cases where the vnode lock is external, such as nullfs,
810 * v_usecount may become zero.
812 * The VV_TEXT flag prevents modifications to the executable while
813 * the vnode is unlocked.
815 VOP_UNLOCK(imgp->vp, 0);
817 error = exec_new_vmspace(imgp, sv);
818 imgp->proc->p_sysent = sv;
820 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
824 for (i = 0; i < hdr->e_phnum; i++) {
825 switch (phdr[i].p_type) {
826 case PT_LOAD: /* Loadable segment */
827 if (phdr[i].p_memsz == 0)
829 prot = __elfN(trans_prot)(phdr[i].p_flags);
830 error = __elfN(load_section)(imgp, phdr[i].p_offset,
831 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
832 phdr[i].p_memsz, phdr[i].p_filesz, prot,
838 * If this segment contains the program headers,
839 * remember their virtual address for the AT_PHDR
840 * aux entry. Static binaries don't usually include
843 if (phdr[i].p_offset == 0 &&
844 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
846 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
849 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
850 seg_size = round_page(phdr[i].p_memsz +
851 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
854 * Make the largest executable segment the official
855 * text segment and all others data.
857 * Note that obreak() assumes that data_addr +
858 * data_size == end of data load area, and the ELF
859 * file format expects segments to be sorted by
860 * address. If multiple data segments exist, the
861 * last one will be used.
864 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
865 text_size = seg_size;
866 text_addr = seg_addr;
868 data_size = seg_size;
869 data_addr = seg_addr;
871 total_size += seg_size;
873 case PT_PHDR: /* Program header table info */
874 proghdr = phdr[i].p_vaddr + et_dyn_addr;
881 if (data_addr == 0 && data_size == 0) {
882 data_addr = text_addr;
883 data_size = text_size;
886 entry = (u_long)hdr->e_entry + et_dyn_addr;
889 * Check limits. It should be safe to check the
890 * limits after loading the segments since we do
891 * not actually fault in all the segments pages.
893 PROC_LOCK(imgp->proc);
894 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
895 text_size > maxtsiz ||
896 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
897 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
898 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
899 PROC_UNLOCK(imgp->proc);
903 vmspace = imgp->proc->p_vmspace;
904 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
905 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
906 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
907 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
910 * We load the dynamic linker where a userland call
911 * to mmap(0, ...) would put it. The rationale behind this
912 * calculation is that it leaves room for the heap to grow to
913 * its maximum allowed size.
915 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
917 PROC_UNLOCK(imgp->proc);
919 imgp->entry_addr = entry;
921 if (interp != NULL) {
922 int have_interp = FALSE;
923 VOP_UNLOCK(imgp->vp, 0);
924 if (brand_info->emul_path != NULL &&
925 brand_info->emul_path[0] != '\0') {
926 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
927 snprintf(path, MAXPATHLEN, "%s%s",
928 brand_info->emul_path, interp);
929 error = __elfN(load_file)(imgp->proc, path, &addr,
930 &imgp->entry_addr, sv->sv_pagesize);
935 if (!have_interp && newinterp != NULL) {
936 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
937 &imgp->entry_addr, sv->sv_pagesize);
942 error = __elfN(load_file)(imgp->proc, interp, &addr,
943 &imgp->entry_addr, sv->sv_pagesize);
945 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
947 uprintf("ELF interpreter %s not found\n", interp);
954 * Construct auxargs table (used by the fixup routine)
956 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
957 elf_auxargs->execfd = -1;
958 elf_auxargs->phdr = proghdr;
959 elf_auxargs->phent = hdr->e_phentsize;
960 elf_auxargs->phnum = hdr->e_phnum;
961 elf_auxargs->pagesz = PAGE_SIZE;
962 elf_auxargs->base = addr;
963 elf_auxargs->flags = 0;
964 elf_auxargs->entry = entry;
966 imgp->auxargs = elf_auxargs;
967 imgp->interpreted = 0;
968 imgp->reloc_base = addr;
969 imgp->proc->p_osrel = osrel;
974 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
977 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
979 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
983 base = (Elf_Addr *)*stack_base;
984 pos = base + (imgp->args->argc + imgp->args->envc + 2);
986 if (args->execfd != -1)
987 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
988 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
989 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
990 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
991 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
992 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
993 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
994 AUXARGS_ENTRY(pos, AT_BASE, args->base);
995 if (imgp->execpathp != 0)
996 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
997 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
998 if (imgp->canary != 0) {
999 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1000 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1002 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1003 if (imgp->pagesizes != 0) {
1004 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1005 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1007 if (imgp->sysent->sv_timekeep_base != 0) {
1008 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1009 imgp->sysent->sv_timekeep_base);
1011 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1012 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1013 imgp->sysent->sv_stackprot);
1014 AUXARGS_ENTRY(pos, AT_NULL, 0);
1016 free(imgp->auxargs, M_TEMP);
1017 imgp->auxargs = NULL;
1020 suword(base, (long)imgp->args->argc);
1021 *stack_base = (register_t *)base;
1026 * Code for generating ELF core dumps.
1029 typedef void (*segment_callback)(vm_map_entry_t, void *);
1031 /* Closure for cb_put_phdr(). */
1032 struct phdr_closure {
1033 Elf_Phdr *phdr; /* Program header to fill in */
1034 Elf_Off offset; /* Offset of segment in core file */
1037 /* Closure for cb_size_segment(). */
1038 struct sseg_closure {
1039 int count; /* Count of writable segments. */
1040 size_t size; /* Total size of all writable segments. */
1043 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1046 int type; /* Note type. */
1047 outfunc_t outfunc; /* Output function. */
1048 void *outarg; /* Argument for the output function. */
1049 size_t outsize; /* Output size. */
1050 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1053 TAILQ_HEAD(note_info_list, note_info);
1055 static void cb_put_phdr(vm_map_entry_t, void *);
1056 static void cb_size_segment(vm_map_entry_t, void *);
1057 static void each_writable_segment(struct thread *, segment_callback, void *);
1058 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1059 int, void *, size_t, struct note_info_list *, size_t, gzFile);
1060 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1062 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1063 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1064 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1065 static int sbuf_drain_core_output(void *, const char *, int);
1066 static int sbuf_drain_count(void *arg, const char *data, int len);
1068 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1069 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1070 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1071 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1072 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1073 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1074 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1075 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1076 static void note_procstat_files(void *, struct sbuf *, size_t *);
1077 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1078 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1079 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1080 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1081 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1083 #ifdef COMPRESS_USER_CORES
1084 extern int compress_user_cores;
1085 extern int compress_user_cores_gzlevel;
1089 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1090 struct ucred *active_cred, struct ucred *file_cred,
1091 struct thread *td, char *core_buf, gzFile gzfile) {
1095 #ifdef COMPRESS_USER_CORES
1096 error = compress_core(gzfile, base, core_buf, len, td);
1098 panic("shouldn't be here");
1101 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1102 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1108 /* Coredump output parameters for sbuf drain routine. */
1109 struct sbuf_drain_core_params {
1111 struct ucred *active_cred;
1112 struct ucred *file_cred;
1115 #ifdef COMPRESS_USER_CORES
1121 * Drain into a core file.
1124 sbuf_drain_core_output(void *arg, const char *data, int len)
1126 struct sbuf_drain_core_params *p;
1129 p = (struct sbuf_drain_core_params *)arg;
1132 * Some kern_proc out routines that print to this sbuf may
1133 * call us with the process lock held. Draining with the
1134 * non-sleepable lock held is unsafe. The lock is needed for
1135 * those routines when dumping a live process. In our case we
1136 * can safely release the lock before draining and acquire
1139 locked = PROC_LOCKED(p->td->td_proc);
1141 PROC_UNLOCK(p->td->td_proc);
1142 #ifdef COMPRESS_USER_CORES
1143 if (p->gzfile != Z_NULL)
1144 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1148 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1149 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1150 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1153 PROC_LOCK(p->td->td_proc);
1161 * Drain into a counter.
1164 sbuf_drain_count(void *arg, const char *data __unused, int len)
1168 sizep = (size_t *)arg;
1174 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1176 struct ucred *cred = td->td_ucred;
1178 struct sseg_closure seginfo;
1179 struct note_info_list notelst;
1180 struct note_info *ninfo;
1182 size_t hdrsize, notesz, coresize;
1184 gzFile gzfile = Z_NULL;
1185 char *core_buf = NULL;
1186 #ifdef COMPRESS_USER_CORES
1187 char gzopen_flags[8];
1189 int doing_compress = flags & IMGACT_CORE_COMPRESS;
1193 TAILQ_INIT(¬elst);
1195 #ifdef COMPRESS_USER_CORES
1196 if (doing_compress) {
1199 if (compress_user_cores_gzlevel >= 0 &&
1200 compress_user_cores_gzlevel <= 9)
1201 *p++ = '0' + compress_user_cores_gzlevel;
1203 gzfile = gz_open("", gzopen_flags, vp);
1204 if (gzfile == Z_NULL) {
1208 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1216 /* Size the program segments. */
1219 each_writable_segment(td, cb_size_segment, &seginfo);
1222 * Collect info about the core file header area.
1224 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1225 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1226 coresize = round_page(hdrsize + notesz) + seginfo.size;
1229 PROC_LOCK(td->td_proc);
1230 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1231 PROC_UNLOCK(td->td_proc);
1237 if (coresize >= limit) {
1243 * Allocate memory for building the header, fill it up,
1244 * and write it out following the notes.
1246 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1251 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1252 ¬elst, notesz, gzfile);
1254 /* Write the contents of all of the writable segments. */
1260 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1261 offset = round_page(hdrsize + notesz);
1262 for (i = 0; i < seginfo.count; i++) {
1263 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1264 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1267 offset += php->p_filesz;
1273 "Failed to write core file for process %s (error %d)\n",
1274 curproc->p_comm, error);
1278 #ifdef COMPRESS_USER_CORES
1280 free(core_buf, M_TEMP);
1284 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1285 TAILQ_REMOVE(¬elst, ninfo, link);
1286 free(ninfo, M_TEMP);
1295 * A callback for each_writable_segment() to write out the segment's
1296 * program header entry.
1299 cb_put_phdr(entry, closure)
1300 vm_map_entry_t entry;
1303 struct phdr_closure *phc = (struct phdr_closure *)closure;
1304 Elf_Phdr *phdr = phc->phdr;
1306 phc->offset = round_page(phc->offset);
1308 phdr->p_type = PT_LOAD;
1309 phdr->p_offset = phc->offset;
1310 phdr->p_vaddr = entry->start;
1312 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1313 phdr->p_align = PAGE_SIZE;
1314 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1316 phc->offset += phdr->p_filesz;
1321 * A callback for each_writable_segment() to gather information about
1322 * the number of segments and their total size.
1325 cb_size_segment(entry, closure)
1326 vm_map_entry_t entry;
1329 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1332 ssc->size += entry->end - entry->start;
1336 * For each writable segment in the process's memory map, call the given
1337 * function with a pointer to the map entry and some arbitrary
1338 * caller-supplied data.
1341 each_writable_segment(td, func, closure)
1343 segment_callback func;
1346 struct proc *p = td->td_proc;
1347 vm_map_t map = &p->p_vmspace->vm_map;
1348 vm_map_entry_t entry;
1349 vm_object_t backing_object, object;
1350 boolean_t ignore_entry;
1352 vm_map_lock_read(map);
1353 for (entry = map->header.next; entry != &map->header;
1354 entry = entry->next) {
1356 * Don't dump inaccessible mappings, deal with legacy
1359 * Note that read-only segments related to the elf binary
1360 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1361 * need to arbitrarily ignore such segments.
1363 if (elf_legacy_coredump) {
1364 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1367 if ((entry->protection & VM_PROT_ALL) == 0)
1372 * Dont include memory segment in the coredump if
1373 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1374 * madvise(2). Do not dump submaps (i.e. parts of the
1377 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1380 if ((object = entry->object.vm_object) == NULL)
1383 /* Ignore memory-mapped devices and such things. */
1384 VM_OBJECT_RLOCK(object);
1385 while ((backing_object = object->backing_object) != NULL) {
1386 VM_OBJECT_RLOCK(backing_object);
1387 VM_OBJECT_RUNLOCK(object);
1388 object = backing_object;
1390 ignore_entry = object->type != OBJT_DEFAULT &&
1391 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1392 VM_OBJECT_RUNLOCK(object);
1396 (*func)(entry, closure);
1398 vm_map_unlock_read(map);
1402 * Write the core file header to the file, including padding up to
1403 * the page boundary.
1406 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1407 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1408 size_t notesz, gzFile gzfile)
1410 struct sbuf_drain_core_params params;
1411 struct note_info *ninfo;
1415 /* Fill in the header. */
1416 bzero(hdr, hdrsize);
1417 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1420 params.active_cred = cred;
1421 params.file_cred = NOCRED;
1424 #ifdef COMPRESS_USER_CORES
1425 params.gzfile = gzfile;
1427 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1428 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms);
1429 sbuf_start_section(sb, NULL);
1430 sbuf_bcat(sb, hdr, hdrsize);
1431 TAILQ_FOREACH(ninfo, notelst, link)
1432 __elfN(putnote)(ninfo, sb);
1433 /* Align up to a page boundary for the program segments. */
1434 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1435 error = sbuf_finish(sb);
1442 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1452 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1455 * To have the debugger select the right thread (LWP) as the initial
1456 * thread, we dump the state of the thread passed to us in td first.
1457 * This is the thread that causes the core dump and thus likely to
1458 * be the right thread one wants to have selected in the debugger.
1461 while (thr != NULL) {
1462 size += register_note(list, NT_PRSTATUS,
1463 __elfN(note_prstatus), thr);
1464 size += register_note(list, NT_FPREGSET,
1465 __elfN(note_fpregset), thr);
1466 size += register_note(list, NT_THRMISC,
1467 __elfN(note_thrmisc), thr);
1468 size += register_note(list, -1,
1469 __elfN(note_threadmd), thr);
1471 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1472 TAILQ_NEXT(thr, td_plist);
1474 thr = TAILQ_NEXT(thr, td_plist);
1477 size += register_note(list, NT_PROCSTAT_PROC,
1478 __elfN(note_procstat_proc), p);
1479 size += register_note(list, NT_PROCSTAT_FILES,
1480 note_procstat_files, p);
1481 size += register_note(list, NT_PROCSTAT_VMMAP,
1482 note_procstat_vmmap, p);
1483 size += register_note(list, NT_PROCSTAT_GROUPS,
1484 note_procstat_groups, p);
1485 size += register_note(list, NT_PROCSTAT_UMASK,
1486 note_procstat_umask, p);
1487 size += register_note(list, NT_PROCSTAT_RLIMIT,
1488 note_procstat_rlimit, p);
1489 size += register_note(list, NT_PROCSTAT_OSREL,
1490 note_procstat_osrel, p);
1491 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1492 __elfN(note_procstat_psstrings), p);
1493 size += register_note(list, NT_PROCSTAT_AUXV,
1494 __elfN(note_procstat_auxv), p);
1500 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1505 struct phdr_closure phc;
1507 ehdr = (Elf_Ehdr *)hdr;
1508 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1510 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1511 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1512 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1513 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1514 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1515 ehdr->e_ident[EI_DATA] = ELF_DATA;
1516 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1517 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1518 ehdr->e_ident[EI_ABIVERSION] = 0;
1519 ehdr->e_ident[EI_PAD] = 0;
1520 ehdr->e_type = ET_CORE;
1521 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1522 ehdr->e_machine = ELF_ARCH32;
1524 ehdr->e_machine = ELF_ARCH;
1526 ehdr->e_version = EV_CURRENT;
1528 ehdr->e_phoff = sizeof(Elf_Ehdr);
1530 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1531 ehdr->e_phentsize = sizeof(Elf_Phdr);
1532 ehdr->e_phnum = numsegs + 1;
1533 ehdr->e_shentsize = sizeof(Elf_Shdr);
1535 ehdr->e_shstrndx = SHN_UNDEF;
1538 * Fill in the program header entries.
1541 /* The note segement. */
1542 phdr->p_type = PT_NOTE;
1543 phdr->p_offset = hdrsize;
1546 phdr->p_filesz = notesz;
1548 phdr->p_flags = PF_R;
1549 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1552 /* All the writable segments from the program. */
1554 phc.offset = round_page(hdrsize + notesz);
1555 each_writable_segment(td, cb_put_phdr, &phc);
1559 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1561 struct note_info *ninfo;
1562 size_t size, notesize;
1565 out(arg, NULL, &size);
1566 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1568 ninfo->outfunc = out;
1569 ninfo->outarg = arg;
1570 ninfo->outsize = size;
1571 TAILQ_INSERT_TAIL(list, ninfo, link);
1576 notesize = sizeof(Elf_Note) + /* note header */
1577 roundup2(8, ELF_NOTE_ROUNDSIZE) + /* note name ("FreeBSD") */
1578 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1584 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1589 if (ninfo->type == -1) {
1590 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1594 note.n_namesz = 8; /* strlen("FreeBSD") + 1 */
1595 note.n_descsz = ninfo->outsize;
1596 note.n_type = ninfo->type;
1598 sbuf_bcat(sb, ¬e, sizeof(note));
1599 sbuf_start_section(sb, &old_len);
1600 sbuf_bcat(sb, "FreeBSD", note.n_namesz);
1601 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1602 if (note.n_descsz == 0)
1604 sbuf_start_section(sb, &old_len);
1605 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1606 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1610 * Miscellaneous note out functions.
1613 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1614 #include <compat/freebsd32/freebsd32.h>
1616 typedef struct prstatus32 elf_prstatus_t;
1617 typedef struct prpsinfo32 elf_prpsinfo_t;
1618 typedef struct fpreg32 elf_prfpregset_t;
1619 typedef struct fpreg32 elf_fpregset_t;
1620 typedef struct reg32 elf_gregset_t;
1621 typedef struct thrmisc32 elf_thrmisc_t;
1622 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1623 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1624 typedef uint32_t elf_ps_strings_t;
1626 typedef prstatus_t elf_prstatus_t;
1627 typedef prpsinfo_t elf_prpsinfo_t;
1628 typedef prfpregset_t elf_prfpregset_t;
1629 typedef prfpregset_t elf_fpregset_t;
1630 typedef gregset_t elf_gregset_t;
1631 typedef thrmisc_t elf_thrmisc_t;
1632 #define ELF_KERN_PROC_MASK 0
1633 typedef struct kinfo_proc elf_kinfo_proc_t;
1634 typedef vm_offset_t elf_ps_strings_t;
1638 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1641 elf_prpsinfo_t *psinfo;
1643 p = (struct proc *)arg;
1645 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1646 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1647 psinfo->pr_version = PRPSINFO_VERSION;
1648 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1649 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1651 * XXX - We don't fill in the command line arguments properly
1654 strlcpy(psinfo->pr_psargs, p->p_comm,
1655 sizeof(psinfo->pr_psargs));
1657 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1658 free(psinfo, M_TEMP);
1660 *sizep = sizeof(*psinfo);
1664 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1667 elf_prstatus_t *status;
1669 td = (struct thread *)arg;
1671 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1672 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1673 status->pr_version = PRSTATUS_VERSION;
1674 status->pr_statussz = sizeof(elf_prstatus_t);
1675 status->pr_gregsetsz = sizeof(elf_gregset_t);
1676 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1677 status->pr_osreldate = osreldate;
1678 status->pr_cursig = td->td_proc->p_sig;
1679 status->pr_pid = td->td_tid;
1680 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1681 fill_regs32(td, &status->pr_reg);
1683 fill_regs(td, &status->pr_reg);
1685 sbuf_bcat(sb, status, sizeof(*status));
1686 free(status, M_TEMP);
1688 *sizep = sizeof(*status);
1692 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1695 elf_prfpregset_t *fpregset;
1697 td = (struct thread *)arg;
1699 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1700 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1701 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1702 fill_fpregs32(td, fpregset);
1704 fill_fpregs(td, fpregset);
1706 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1707 free(fpregset, M_TEMP);
1709 *sizep = sizeof(*fpregset);
1713 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1716 elf_thrmisc_t thrmisc;
1718 td = (struct thread *)arg;
1720 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1721 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1722 strcpy(thrmisc.pr_tname, td->td_name);
1723 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1725 *sizep = sizeof(thrmisc);
1729 * Allow for MD specific notes, as well as any MD
1730 * specific preparations for writing MI notes.
1733 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1739 td = (struct thread *)arg;
1741 if (size != 0 && sb != NULL)
1742 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);
1754 #ifdef KINFO_PROC_SIZE
1755 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1759 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1765 p = (struct proc *)arg;
1766 size = sizeof(structsize) + p->p_numthreads *
1767 sizeof(elf_kinfo_proc_t);
1770 KASSERT(*sizep == size, ("invalid size"));
1771 structsize = sizeof(elf_kinfo_proc_t);
1772 sbuf_bcat(sb, &structsize, sizeof(structsize));
1774 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1779 #ifdef KINFO_FILE_SIZE
1780 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1784 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1790 p = (struct proc *)arg;
1793 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1794 sbuf_set_drain(sb, sbuf_drain_count, &size);
1795 sbuf_bcat(sb, &structsize, sizeof(structsize));
1797 kern_proc_filedesc_out(p, sb, -1);
1802 structsize = sizeof(struct kinfo_file);
1803 sbuf_bcat(sb, &structsize, sizeof(structsize));
1805 kern_proc_filedesc_out(p, sb, -1);
1809 #ifdef KINFO_VMENTRY_SIZE
1810 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1814 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1820 p = (struct proc *)arg;
1823 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1824 sbuf_set_drain(sb, sbuf_drain_count, &size);
1825 sbuf_bcat(sb, &structsize, sizeof(structsize));
1827 kern_proc_vmmap_out(p, sb);
1832 structsize = sizeof(struct kinfo_vmentry);
1833 sbuf_bcat(sb, &structsize, sizeof(structsize));
1835 kern_proc_vmmap_out(p, sb);
1840 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1846 p = (struct proc *)arg;
1847 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1849 KASSERT(*sizep == size, ("invalid size"));
1850 structsize = sizeof(gid_t);
1851 sbuf_bcat(sb, &structsize, sizeof(structsize));
1852 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1859 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1865 p = (struct proc *)arg;
1866 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1868 KASSERT(*sizep == size, ("invalid size"));
1869 structsize = sizeof(p->p_fd->fd_cmask);
1870 sbuf_bcat(sb, &structsize, sizeof(structsize));
1871 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1877 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1880 struct rlimit rlim[RLIM_NLIMITS];
1884 p = (struct proc *)arg;
1885 size = sizeof(structsize) + sizeof(rlim);
1887 KASSERT(*sizep == size, ("invalid size"));
1888 structsize = sizeof(rlim);
1889 sbuf_bcat(sb, &structsize, sizeof(structsize));
1891 for (i = 0; i < RLIM_NLIMITS; i++)
1892 lim_rlimit(p, i, &rlim[i]);
1894 sbuf_bcat(sb, rlim, sizeof(rlim));
1900 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1906 p = (struct proc *)arg;
1907 size = sizeof(structsize) + sizeof(p->p_osrel);
1909 KASSERT(*sizep == size, ("invalid size"));
1910 structsize = sizeof(p->p_osrel);
1911 sbuf_bcat(sb, &structsize, sizeof(structsize));
1912 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1918 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1921 elf_ps_strings_t ps_strings;
1925 p = (struct proc *)arg;
1926 size = sizeof(structsize) + sizeof(ps_strings);
1928 KASSERT(*sizep == size, ("invalid size"));
1929 structsize = sizeof(ps_strings);
1930 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1931 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1933 ps_strings = p->p_sysent->sv_psstrings;
1935 sbuf_bcat(sb, &structsize, sizeof(structsize));
1936 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1942 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1948 p = (struct proc *)arg;
1951 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1952 sbuf_set_drain(sb, sbuf_drain_count, &size);
1953 sbuf_bcat(sb, &structsize, sizeof(structsize));
1955 proc_getauxv(curthread, p, sb);
1961 structsize = sizeof(Elf_Auxinfo);
1962 sbuf_bcat(sb, &structsize, sizeof(structsize));
1964 proc_getauxv(curthread, p, sb);
1970 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
1971 int32_t *osrel, const Elf_Phdr *pnote)
1973 const Elf_Note *note, *note0, *note_end;
1974 const char *note_name;
1977 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
1978 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
1981 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1982 note_end = (const Elf_Note *)(imgp->image_header +
1983 pnote->p_offset + pnote->p_filesz);
1984 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1985 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
1986 (const char *)note < sizeof(Elf_Note))
1988 if (note->n_namesz != checknote->hdr.n_namesz ||
1989 note->n_descsz != checknote->hdr.n_descsz ||
1990 note->n_type != checknote->hdr.n_type)
1992 note_name = (const char *)(note + 1);
1993 if (note_name + checknote->hdr.n_namesz >=
1994 (const char *)note_end || strncmp(checknote->vendor,
1995 note_name, checknote->hdr.n_namesz) != 0)
1999 * Fetch the osreldate for binary
2000 * from the ELF OSABI-note if necessary.
2002 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2003 checknote->trans_osrel != NULL)
2004 return (checknote->trans_osrel(note, osrel));
2008 note = (const Elf_Note *)((const char *)(note + 1) +
2009 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2010 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2017 * Try to find the appropriate ABI-note section for checknote,
2018 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2019 * first page of the image is searched, the same as for headers.
2022 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2025 const Elf_Phdr *phdr;
2026 const Elf_Ehdr *hdr;
2029 hdr = (const Elf_Ehdr *)imgp->image_header;
2030 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2032 for (i = 0; i < hdr->e_phnum; i++) {
2033 if (phdr[i].p_type == PT_NOTE &&
2034 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2042 * Tell kern_execve.c about it, with a little help from the linker.
2044 static struct execsw __elfN(execsw) = {
2045 __CONCAT(exec_, __elfN(imgact)),
2046 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2048 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2050 #ifdef COMPRESS_USER_CORES
2052 * Compress and write out a core segment for a user process.
2054 * 'inbuf' is the starting address of a VM segment in the process' address
2055 * space that is to be compressed and written out to the core file. 'dest_buf'
2056 * is a buffer in the kernel's address space. The segment is copied from
2057 * 'inbuf' to 'dest_buf' first before being processed by the compression
2058 * routine gzwrite(). This copying is necessary because the content of the VM
2059 * segment may change between the compression pass and the crc-computation pass
2060 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel.
2062 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2065 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2070 unsigned int chunk_len;
2073 if (inbuf != NULL) {
2074 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2075 copyin(inbuf, dest_buf, chunk_len);
2080 len_compressed = gzwrite(file, dest_buf, chunk_len);
2082 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2084 if ((unsigned int)len_compressed != chunk_len) {
2086 "compress_core: length mismatch (0x%x returned, "
2087 "0x%x expected)\n", len_compressed, chunk_len);
2088 EVENTHANDLER_INVOKE(app_coredump_error, td,
2089 "compress_core: length mismatch %x -> %x",
2090 chunk_len, len_compressed);
2100 #endif /* COMPRESS_USER_CORES */
2103 __elfN(trans_prot)(Elf_Word flags)
2109 prot |= VM_PROT_EXECUTE;
2111 prot |= VM_PROT_WRITE;
2113 prot |= VM_PROT_READ;
2114 #if __ELF_WORD_SIZE == 32
2115 #if defined(__amd64__) || defined(__ia64__)
2116 if (i386_read_exec && (flags & PF_R))
2117 prot |= VM_PROT_EXECUTE;
2124 __elfN(untrans_prot)(vm_prot_t prot)
2129 if (prot & VM_PROT_EXECUTE)
2131 if (prot & VM_PROT_READ)
2133 if (prot & VM_PROT_WRITE)