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__)
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 /* No known brand, see if the header is recognized by any brand */
298 for (i = 0; i < MAX_BRANDS; i++) {
299 bi = elf_brand_list[i];
300 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
301 bi->header_supported == NULL)
303 if (hdr->e_machine == bi->machine) {
304 ret = bi->header_supported(imgp);
310 /* Lacking a known brand, search for a recognized interpreter. */
311 if (interp != NULL) {
312 for (i = 0; i < MAX_BRANDS; i++) {
313 bi = elf_brand_list[i];
314 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
316 if (hdr->e_machine == bi->machine &&
317 /* ELF image p_filesz includes terminating zero */
318 strlen(bi->interp_path) + 1 == interp_name_len &&
319 strncmp(interp, bi->interp_path, interp_name_len)
325 /* Lacking a recognized interpreter, try the default brand */
326 for (i = 0; i < MAX_BRANDS; i++) {
327 bi = elf_brand_list[i];
328 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
330 if (hdr->e_machine == bi->machine &&
331 __elfN(fallback_brand) == bi->brand)
338 __elfN(check_header)(const Elf_Ehdr *hdr)
344 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
345 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
346 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
347 hdr->e_phentsize != sizeof(Elf_Phdr) ||
348 hdr->e_version != ELF_TARG_VER)
352 * Make sure we have at least one brand for this machine.
355 for (i = 0; i < MAX_BRANDS; i++) {
356 bi = elf_brand_list[i];
357 if (bi != NULL && bi->machine == hdr->e_machine)
367 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
368 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
375 * Create the page if it doesn't exist yet. Ignore errors.
378 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
379 VM_PROT_ALL, VM_PROT_ALL, 0);
383 * Find the page from the underlying object.
386 sf = vm_imgact_map_page(object, offset);
388 return (KERN_FAILURE);
389 off = offset - trunc_page(offset);
390 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
392 vm_imgact_unmap_page(sf);
394 return (KERN_FAILURE);
398 return (KERN_SUCCESS);
402 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
403 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
410 if (start != trunc_page(start)) {
411 rv = __elfN(map_partial)(map, object, offset, start,
412 round_page(start), prot);
415 offset += round_page(start) - start;
416 start = round_page(start);
418 if (end != round_page(end)) {
419 rv = __elfN(map_partial)(map, object, offset +
420 trunc_page(end) - start, trunc_page(end), end, prot);
423 end = trunc_page(end);
426 if (offset & PAGE_MASK) {
428 * The mapping is not page aligned. This means we have
429 * to copy the data. Sigh.
431 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
432 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
437 return (KERN_SUCCESS);
438 for (; start < end; start += sz) {
439 sf = vm_imgact_map_page(object, offset);
441 return (KERN_FAILURE);
442 off = offset - trunc_page(offset);
444 if (sz > PAGE_SIZE - off)
445 sz = PAGE_SIZE - off;
446 error = copyout((caddr_t)sf_buf_kva(sf) + off,
448 vm_imgact_unmap_page(sf);
450 return (KERN_FAILURE);
456 vm_object_reference(object);
458 rv = vm_map_insert(map, object, offset, start, end,
459 prot, VM_PROT_ALL, cow);
461 if (rv != KERN_SUCCESS)
462 vm_object_deallocate(object);
466 return (KERN_SUCCESS);
471 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
472 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
479 vm_offset_t map_addr;
482 vm_offset_t file_addr;
485 * It's necessary to fail if the filsz + offset taken from the
486 * header is greater than the actual file pager object's size.
487 * If we were to allow this, then the vm_map_find() below would
488 * walk right off the end of the file object and into the ether.
490 * While I'm here, might as well check for something else that
491 * is invalid: filsz cannot be greater than memsz.
493 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
494 uprintf("elf_load_section: truncated ELF file\n");
498 object = imgp->object;
499 map = &imgp->proc->p_vmspace->vm_map;
500 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
501 file_addr = trunc_page_ps(offset, pagesize);
504 * We have two choices. We can either clear the data in the last page
505 * of an oversized mapping, or we can start the anon mapping a page
506 * early and copy the initialized data into that first page. We
507 * choose the second..
510 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
512 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
515 /* cow flags: don't dump readonly sections in core */
516 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
517 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
519 rv = __elfN(map_insert)(map,
521 file_addr, /* file offset */
522 map_addr, /* virtual start */
523 map_addr + map_len,/* virtual end */
526 if (rv != KERN_SUCCESS)
529 /* we can stop now if we've covered it all */
530 if (memsz == filsz) {
537 * We have to get the remaining bit of the file into the first part
538 * of the oversized map segment. This is normally because the .data
539 * segment in the file is extended to provide bss. It's a neat idea
540 * to try and save a page, but it's a pain in the behind to implement.
542 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
543 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
544 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
547 /* This had damn well better be true! */
549 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
550 map_len, VM_PROT_ALL, 0);
551 if (rv != KERN_SUCCESS) {
559 sf = vm_imgact_map_page(object, offset + filsz);
563 /* send the page fragment to user space */
564 off = trunc_page_ps(offset + filsz, pagesize) -
565 trunc_page(offset + filsz);
566 error = copyout((caddr_t)sf_buf_kva(sf) + off,
567 (caddr_t)map_addr, copy_len);
568 vm_imgact_unmap_page(sf);
575 * set it to the specified protection.
576 * XXX had better undo the damage from pasting over the cracks here!
578 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
579 map_len), prot, FALSE);
585 * Load the file "file" into memory. It may be either a shared object
588 * The "addr" reference parameter is in/out. On entry, it specifies
589 * the address where a shared object should be loaded. If the file is
590 * an executable, this value is ignored. On exit, "addr" specifies
591 * where the file was actually loaded.
593 * The "entry" reference parameter is out only. On exit, it specifies
594 * the entry point for the loaded file.
597 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
598 u_long *entry, size_t pagesize)
603 struct image_params image_params;
605 const Elf_Ehdr *hdr = NULL;
606 const Elf_Phdr *phdr = NULL;
607 struct nameidata *nd;
609 struct image_params *imgp;
612 u_long base_addr = 0;
613 int error, i, numsegs;
615 #ifdef CAPABILITY_MODE
617 * XXXJA: This check can go away once we are sufficiently confident
618 * that the checks in namei() are correct.
620 if (IN_CAPABILITY_MODE(curthread))
624 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
626 attr = &tempdata->attr;
627 imgp = &tempdata->image_params;
630 * Initialize part of the common data
634 imgp->firstpage = NULL;
635 imgp->image_header = NULL;
637 imgp->execlabel = NULL;
639 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
640 if ((error = namei(nd)) != 0) {
644 NDFREE(nd, NDF_ONLY_PNBUF);
645 imgp->vp = nd->ni_vp;
648 * Check permissions, modes, uid, etc on the file, and "open" it.
650 error = exec_check_permissions(imgp);
654 error = exec_map_first_page(imgp);
659 * Also make certain that the interpreter stays the same, so set
660 * its VV_TEXT flag, too.
662 VOP_SET_TEXT(nd->ni_vp);
664 imgp->object = nd->ni_vp->v_object;
666 hdr = (const Elf_Ehdr *)imgp->image_header;
667 if ((error = __elfN(check_header)(hdr)) != 0)
669 if (hdr->e_type == ET_DYN)
671 else if (hdr->e_type == ET_EXEC)
678 /* Only support headers that fit within first page for now */
679 if ((hdr->e_phoff > PAGE_SIZE) ||
680 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
685 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
686 if (!aligned(phdr, Elf_Addr)) {
691 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
692 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
693 /* Loadable segment */
694 prot = __elfN(trans_prot)(phdr[i].p_flags);
695 error = __elfN(load_section)(imgp, phdr[i].p_offset,
696 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
697 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
701 * Establish the base address if this is the
705 base_addr = trunc_page(phdr[i].p_vaddr +
711 *entry = (unsigned long)hdr->e_entry + rbase;
715 exec_unmap_first_page(imgp);
720 free(tempdata, M_TEMP);
726 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
728 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
729 const Elf_Phdr *phdr;
730 Elf_Auxargs *elf_auxargs;
731 struct vmspace *vmspace;
733 u_long text_size = 0, data_size = 0, total_size = 0;
734 u_long text_addr = 0, data_addr = 0;
735 u_long seg_size, seg_addr;
736 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
738 int error = 0, i, n, interp_name_len = 0;
739 const char *interp = NULL, *newinterp = NULL;
740 Elf_Brandinfo *brand_info;
742 struct sysentvec *sv;
745 * Do we have a valid ELF header ?
747 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
748 * if particular brand doesn't support it.
750 if (__elfN(check_header)(hdr) != 0 ||
751 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
755 * From here on down, we return an errno, not -1, as we've
756 * detected an ELF file.
759 if ((hdr->e_phoff > PAGE_SIZE) ||
760 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
761 /* Only support headers in first page for now */
764 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
765 if (!aligned(phdr, Elf_Addr))
769 for (i = 0; i < hdr->e_phnum; i++) {
770 switch (phdr[i].p_type) {
773 baddr = phdr[i].p_vaddr;
777 /* Path to interpreter */
778 if (phdr[i].p_filesz > MAXPATHLEN ||
779 phdr[i].p_offset > PAGE_SIZE ||
780 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
782 interp = imgp->image_header + phdr[i].p_offset;
783 interp_name_len = phdr[i].p_filesz;
788 __elfN(trans_prot)(phdr[i].p_flags);
793 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
795 if (brand_info == NULL) {
796 uprintf("ELF binary type \"%u\" not known.\n",
797 hdr->e_ident[EI_OSABI]);
800 if (hdr->e_type == ET_DYN) {
801 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
804 * Honour the base load address from the dso if it is
805 * non-zero for some reason.
808 et_dyn_addr = ET_DYN_LOAD_ADDR;
813 sv = brand_info->sysvec;
814 if (interp != NULL && brand_info->interp_newpath != NULL)
815 newinterp = brand_info->interp_newpath;
818 * Avoid a possible deadlock if the current address space is destroyed
819 * and that address space maps the locked vnode. In the common case,
820 * the locked vnode's v_usecount is decremented but remains greater
821 * than zero. Consequently, the vnode lock is not needed by vrele().
822 * However, in cases where the vnode lock is external, such as nullfs,
823 * v_usecount may become zero.
825 * The VV_TEXT flag prevents modifications to the executable while
826 * the vnode is unlocked.
828 VOP_UNLOCK(imgp->vp, 0);
830 error = exec_new_vmspace(imgp, sv);
831 imgp->proc->p_sysent = sv;
833 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
837 for (i = 0; i < hdr->e_phnum; i++) {
838 switch (phdr[i].p_type) {
839 case PT_LOAD: /* Loadable segment */
840 if (phdr[i].p_memsz == 0)
842 prot = __elfN(trans_prot)(phdr[i].p_flags);
843 error = __elfN(load_section)(imgp, phdr[i].p_offset,
844 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
845 phdr[i].p_memsz, phdr[i].p_filesz, prot,
851 * If this segment contains the program headers,
852 * remember their virtual address for the AT_PHDR
853 * aux entry. Static binaries don't usually include
856 if (phdr[i].p_offset == 0 &&
857 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
859 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
862 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
863 seg_size = round_page(phdr[i].p_memsz +
864 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
867 * Make the largest executable segment the official
868 * text segment and all others data.
870 * Note that obreak() assumes that data_addr +
871 * data_size == end of data load area, and the ELF
872 * file format expects segments to be sorted by
873 * address. If multiple data segments exist, the
874 * last one will be used.
877 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
878 text_size = seg_size;
879 text_addr = seg_addr;
881 data_size = seg_size;
882 data_addr = seg_addr;
884 total_size += seg_size;
886 case PT_PHDR: /* Program header table info */
887 proghdr = phdr[i].p_vaddr + et_dyn_addr;
894 if (data_addr == 0 && data_size == 0) {
895 data_addr = text_addr;
896 data_size = text_size;
899 entry = (u_long)hdr->e_entry + et_dyn_addr;
902 * Check limits. It should be safe to check the
903 * limits after loading the segments since we do
904 * not actually fault in all the segments pages.
906 PROC_LOCK(imgp->proc);
907 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
908 text_size > maxtsiz ||
909 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
910 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
911 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
912 PROC_UNLOCK(imgp->proc);
916 vmspace = imgp->proc->p_vmspace;
917 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
918 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
919 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
920 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
923 * We load the dynamic linker where a userland call
924 * to mmap(0, ...) would put it. The rationale behind this
925 * calculation is that it leaves room for the heap to grow to
926 * its maximum allowed size.
928 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
930 PROC_UNLOCK(imgp->proc);
932 imgp->entry_addr = entry;
934 if (interp != NULL) {
935 int have_interp = FALSE;
936 VOP_UNLOCK(imgp->vp, 0);
937 if (brand_info->emul_path != NULL &&
938 brand_info->emul_path[0] != '\0') {
939 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
940 snprintf(path, MAXPATHLEN, "%s%s",
941 brand_info->emul_path, interp);
942 error = __elfN(load_file)(imgp->proc, path, &addr,
943 &imgp->entry_addr, sv->sv_pagesize);
948 if (!have_interp && newinterp != NULL) {
949 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
950 &imgp->entry_addr, sv->sv_pagesize);
955 error = __elfN(load_file)(imgp->proc, interp, &addr,
956 &imgp->entry_addr, sv->sv_pagesize);
958 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
960 uprintf("ELF interpreter %s not found\n", interp);
967 * Construct auxargs table (used by the fixup routine)
969 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
970 elf_auxargs->execfd = -1;
971 elf_auxargs->phdr = proghdr;
972 elf_auxargs->phent = hdr->e_phentsize;
973 elf_auxargs->phnum = hdr->e_phnum;
974 elf_auxargs->pagesz = PAGE_SIZE;
975 elf_auxargs->base = addr;
976 elf_auxargs->flags = 0;
977 elf_auxargs->entry = entry;
979 imgp->auxargs = elf_auxargs;
980 imgp->interpreted = 0;
981 imgp->reloc_base = addr;
982 imgp->proc->p_osrel = osrel;
987 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
990 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
992 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
996 base = (Elf_Addr *)*stack_base;
997 pos = base + (imgp->args->argc + imgp->args->envc + 2);
999 if (args->execfd != -1)
1000 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1001 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1002 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1003 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1004 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1005 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1006 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1007 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1008 if (imgp->execpathp != 0)
1009 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1010 AUXARGS_ENTRY(pos, AT_OSRELDATE, osreldate);
1011 if (imgp->canary != 0) {
1012 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1013 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1015 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1016 if (imgp->pagesizes != 0) {
1017 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1018 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1020 if (imgp->sysent->sv_timekeep_base != 0) {
1021 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1022 imgp->sysent->sv_timekeep_base);
1024 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1025 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1026 imgp->sysent->sv_stackprot);
1027 AUXARGS_ENTRY(pos, AT_NULL, 0);
1029 free(imgp->auxargs, M_TEMP);
1030 imgp->auxargs = NULL;
1033 suword(base, (long)imgp->args->argc);
1034 *stack_base = (register_t *)base;
1039 * Code for generating ELF core dumps.
1042 typedef void (*segment_callback)(vm_map_entry_t, void *);
1044 /* Closure for cb_put_phdr(). */
1045 struct phdr_closure {
1046 Elf_Phdr *phdr; /* Program header to fill in */
1047 Elf_Off offset; /* Offset of segment in core file */
1050 /* Closure for cb_size_segment(). */
1051 struct sseg_closure {
1052 int count; /* Count of writable segments. */
1053 size_t size; /* Total size of all writable segments. */
1056 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1059 int type; /* Note type. */
1060 outfunc_t outfunc; /* Output function. */
1061 void *outarg; /* Argument for the output function. */
1062 size_t outsize; /* Output size. */
1063 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1066 TAILQ_HEAD(note_info_list, note_info);
1068 static void cb_put_phdr(vm_map_entry_t, void *);
1069 static void cb_size_segment(vm_map_entry_t, void *);
1070 static void each_writable_segment(struct thread *, segment_callback, void *);
1071 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1072 int, void *, size_t, struct note_info_list *, size_t, gzFile);
1073 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1075 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1076 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1077 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1078 static int sbuf_drain_core_output(void *, const char *, int);
1079 static int sbuf_drain_count(void *arg, const char *data, int len);
1081 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1082 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1083 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1084 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1085 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1086 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1087 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1088 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1089 static void note_procstat_files(void *, struct sbuf *, size_t *);
1090 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1091 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1092 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1093 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1094 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1096 #ifdef COMPRESS_USER_CORES
1097 extern int compress_user_cores;
1098 extern int compress_user_cores_gzlevel;
1102 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1103 struct ucred *active_cred, struct ucred *file_cred,
1104 struct thread *td, char *core_buf, gzFile gzfile) {
1108 #ifdef COMPRESS_USER_CORES
1109 error = compress_core(gzfile, base, core_buf, len, td);
1111 panic("shouldn't be here");
1114 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1115 UIO_USERSPACE, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1116 active_cred, file_cred, NULL, td);
1121 /* Coredump output parameters for sbuf drain routine. */
1122 struct sbuf_drain_core_params {
1124 struct ucred *active_cred;
1125 struct ucred *file_cred;
1128 #ifdef COMPRESS_USER_CORES
1134 * Drain into a core file.
1137 sbuf_drain_core_output(void *arg, const char *data, int len)
1139 struct sbuf_drain_core_params *p;
1142 p = (struct sbuf_drain_core_params *)arg;
1145 * Some kern_proc out routines that print to this sbuf may
1146 * call us with the process lock held. Draining with the
1147 * non-sleepable lock held is unsafe. The lock is needed for
1148 * those routines when dumping a live process. In our case we
1149 * can safely release the lock before draining and acquire
1152 locked = PROC_LOCKED(p->td->td_proc);
1154 PROC_UNLOCK(p->td->td_proc);
1155 #ifdef COMPRESS_USER_CORES
1156 if (p->gzfile != Z_NULL)
1157 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1161 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1162 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1163 IO_UNIT | IO_DIRECT | IO_RANGELOCKED, p->active_cred,
1164 p->file_cred, NULL, p->td);
1166 PROC_LOCK(p->td->td_proc);
1174 * Drain into a counter.
1177 sbuf_drain_count(void *arg, const char *data __unused, int len)
1181 sizep = (size_t *)arg;
1187 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1189 struct ucred *cred = td->td_ucred;
1191 struct sseg_closure seginfo;
1192 struct note_info_list notelst;
1193 struct note_info *ninfo;
1195 size_t hdrsize, notesz, coresize;
1197 gzFile gzfile = Z_NULL;
1198 char *core_buf = NULL;
1199 #ifdef COMPRESS_USER_CORES
1200 char gzopen_flags[8];
1202 int doing_compress = flags & IMGACT_CORE_COMPRESS;
1206 TAILQ_INIT(¬elst);
1208 #ifdef COMPRESS_USER_CORES
1209 if (doing_compress) {
1212 if (compress_user_cores_gzlevel >= 0 &&
1213 compress_user_cores_gzlevel <= 9)
1214 *p++ = '0' + compress_user_cores_gzlevel;
1216 gzfile = gz_open("", gzopen_flags, vp);
1217 if (gzfile == Z_NULL) {
1221 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1229 /* Size the program segments. */
1232 each_writable_segment(td, cb_size_segment, &seginfo);
1235 * Collect info about the core file header area.
1237 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1238 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1239 coresize = round_page(hdrsize + notesz) + seginfo.size;
1242 PROC_LOCK(td->td_proc);
1243 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1244 PROC_UNLOCK(td->td_proc);
1250 if (coresize >= limit) {
1256 * Allocate memory for building the header, fill it up,
1257 * and write it out following the notes.
1259 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1264 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1265 ¬elst, notesz, gzfile);
1267 /* Write the contents of all of the writable segments. */
1273 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1274 offset = round_page(hdrsize + notesz);
1275 for (i = 0; i < seginfo.count; i++) {
1276 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1277 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1280 offset += php->p_filesz;
1286 "Failed to write core file for process %s (error %d)\n",
1287 curproc->p_comm, error);
1291 #ifdef COMPRESS_USER_CORES
1293 free(core_buf, M_TEMP);
1297 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1298 TAILQ_REMOVE(¬elst, ninfo, link);
1299 free(ninfo, M_TEMP);
1308 * A callback for each_writable_segment() to write out the segment's
1309 * program header entry.
1312 cb_put_phdr(entry, closure)
1313 vm_map_entry_t entry;
1316 struct phdr_closure *phc = (struct phdr_closure *)closure;
1317 Elf_Phdr *phdr = phc->phdr;
1319 phc->offset = round_page(phc->offset);
1321 phdr->p_type = PT_LOAD;
1322 phdr->p_offset = phc->offset;
1323 phdr->p_vaddr = entry->start;
1325 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1326 phdr->p_align = PAGE_SIZE;
1327 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1329 phc->offset += phdr->p_filesz;
1334 * A callback for each_writable_segment() to gather information about
1335 * the number of segments and their total size.
1338 cb_size_segment(entry, closure)
1339 vm_map_entry_t entry;
1342 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1345 ssc->size += entry->end - entry->start;
1349 * For each writable segment in the process's memory map, call the given
1350 * function with a pointer to the map entry and some arbitrary
1351 * caller-supplied data.
1354 each_writable_segment(td, func, closure)
1356 segment_callback func;
1359 struct proc *p = td->td_proc;
1360 vm_map_t map = &p->p_vmspace->vm_map;
1361 vm_map_entry_t entry;
1362 vm_object_t backing_object, object;
1363 boolean_t ignore_entry;
1365 vm_map_lock_read(map);
1366 for (entry = map->header.next; entry != &map->header;
1367 entry = entry->next) {
1369 * Don't dump inaccessible mappings, deal with legacy
1372 * Note that read-only segments related to the elf binary
1373 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1374 * need to arbitrarily ignore such segments.
1376 if (elf_legacy_coredump) {
1377 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1380 if ((entry->protection & VM_PROT_ALL) == 0)
1385 * Dont include memory segment in the coredump if
1386 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1387 * madvise(2). Do not dump submaps (i.e. parts of the
1390 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1393 if ((object = entry->object.vm_object) == NULL)
1396 /* Ignore memory-mapped devices and such things. */
1397 VM_OBJECT_RLOCK(object);
1398 while ((backing_object = object->backing_object) != NULL) {
1399 VM_OBJECT_RLOCK(backing_object);
1400 VM_OBJECT_RUNLOCK(object);
1401 object = backing_object;
1403 ignore_entry = object->type != OBJT_DEFAULT &&
1404 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1405 VM_OBJECT_RUNLOCK(object);
1409 (*func)(entry, closure);
1411 vm_map_unlock_read(map);
1415 * Write the core file header to the file, including padding up to
1416 * the page boundary.
1419 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1420 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1421 size_t notesz, gzFile gzfile)
1423 struct sbuf_drain_core_params params;
1424 struct note_info *ninfo;
1428 /* Fill in the header. */
1429 bzero(hdr, hdrsize);
1430 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1433 params.active_cred = cred;
1434 params.file_cred = NOCRED;
1437 #ifdef COMPRESS_USER_CORES
1438 params.gzfile = gzfile;
1440 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1441 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms);
1442 sbuf_start_section(sb, NULL);
1443 sbuf_bcat(sb, hdr, hdrsize);
1444 TAILQ_FOREACH(ninfo, notelst, link)
1445 __elfN(putnote)(ninfo, sb);
1446 /* Align up to a page boundary for the program segments. */
1447 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1448 error = sbuf_finish(sb);
1455 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1465 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1468 * To have the debugger select the right thread (LWP) as the initial
1469 * thread, we dump the state of the thread passed to us in td first.
1470 * This is the thread that causes the core dump and thus likely to
1471 * be the right thread one wants to have selected in the debugger.
1474 while (thr != NULL) {
1475 size += register_note(list, NT_PRSTATUS,
1476 __elfN(note_prstatus), thr);
1477 size += register_note(list, NT_FPREGSET,
1478 __elfN(note_fpregset), thr);
1479 size += register_note(list, NT_THRMISC,
1480 __elfN(note_thrmisc), thr);
1481 size += register_note(list, -1,
1482 __elfN(note_threadmd), thr);
1484 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1485 TAILQ_NEXT(thr, td_plist);
1487 thr = TAILQ_NEXT(thr, td_plist);
1490 size += register_note(list, NT_PROCSTAT_PROC,
1491 __elfN(note_procstat_proc), p);
1492 size += register_note(list, NT_PROCSTAT_FILES,
1493 note_procstat_files, p);
1494 size += register_note(list, NT_PROCSTAT_VMMAP,
1495 note_procstat_vmmap, p);
1496 size += register_note(list, NT_PROCSTAT_GROUPS,
1497 note_procstat_groups, p);
1498 size += register_note(list, NT_PROCSTAT_UMASK,
1499 note_procstat_umask, p);
1500 size += register_note(list, NT_PROCSTAT_RLIMIT,
1501 note_procstat_rlimit, p);
1502 size += register_note(list, NT_PROCSTAT_OSREL,
1503 note_procstat_osrel, p);
1504 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1505 __elfN(note_procstat_psstrings), p);
1506 size += register_note(list, NT_PROCSTAT_AUXV,
1507 __elfN(note_procstat_auxv), p);
1513 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1518 struct phdr_closure phc;
1520 ehdr = (Elf_Ehdr *)hdr;
1521 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1523 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1524 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1525 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1526 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1527 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1528 ehdr->e_ident[EI_DATA] = ELF_DATA;
1529 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1530 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1531 ehdr->e_ident[EI_ABIVERSION] = 0;
1532 ehdr->e_ident[EI_PAD] = 0;
1533 ehdr->e_type = ET_CORE;
1534 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1535 ehdr->e_machine = ELF_ARCH32;
1537 ehdr->e_machine = ELF_ARCH;
1539 ehdr->e_version = EV_CURRENT;
1541 ehdr->e_phoff = sizeof(Elf_Ehdr);
1543 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1544 ehdr->e_phentsize = sizeof(Elf_Phdr);
1545 ehdr->e_phnum = numsegs + 1;
1546 ehdr->e_shentsize = sizeof(Elf_Shdr);
1548 ehdr->e_shstrndx = SHN_UNDEF;
1551 * Fill in the program header entries.
1554 /* The note segement. */
1555 phdr->p_type = PT_NOTE;
1556 phdr->p_offset = hdrsize;
1559 phdr->p_filesz = notesz;
1561 phdr->p_flags = PF_R;
1562 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1565 /* All the writable segments from the program. */
1567 phc.offset = round_page(hdrsize + notesz);
1568 each_writable_segment(td, cb_put_phdr, &phc);
1572 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1574 struct note_info *ninfo;
1575 size_t size, notesize;
1578 out(arg, NULL, &size);
1579 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1581 ninfo->outfunc = out;
1582 ninfo->outarg = arg;
1583 ninfo->outsize = size;
1584 TAILQ_INSERT_TAIL(list, ninfo, link);
1589 notesize = sizeof(Elf_Note) + /* note header */
1590 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1592 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1598 append_note_data(const void *src, void *dst, size_t len)
1602 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1604 bcopy(src, dst, len);
1605 bzero((char *)dst + len, padded_len - len);
1607 return (padded_len);
1611 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1619 note = (Elf_Note *)buf;
1620 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1621 note->n_descsz = size;
1622 note->n_type = type;
1623 buf += sizeof(*note);
1624 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1625 sizeof(FREEBSD_ABI_VENDOR));
1626 append_note_data(src, buf, size);
1631 notesize = sizeof(Elf_Note) + /* note header */
1632 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1634 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1640 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1645 if (ninfo->type == -1) {
1646 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1650 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1651 note.n_descsz = ninfo->outsize;
1652 note.n_type = ninfo->type;
1654 sbuf_bcat(sb, ¬e, sizeof(note));
1655 sbuf_start_section(sb, &old_len);
1656 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1657 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1658 if (note.n_descsz == 0)
1660 sbuf_start_section(sb, &old_len);
1661 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1662 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1666 * Miscellaneous note out functions.
1669 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1670 #include <compat/freebsd32/freebsd32.h>
1672 typedef struct prstatus32 elf_prstatus_t;
1673 typedef struct prpsinfo32 elf_prpsinfo_t;
1674 typedef struct fpreg32 elf_prfpregset_t;
1675 typedef struct fpreg32 elf_fpregset_t;
1676 typedef struct reg32 elf_gregset_t;
1677 typedef struct thrmisc32 elf_thrmisc_t;
1678 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1679 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1680 typedef uint32_t elf_ps_strings_t;
1682 typedef prstatus_t elf_prstatus_t;
1683 typedef prpsinfo_t elf_prpsinfo_t;
1684 typedef prfpregset_t elf_prfpregset_t;
1685 typedef prfpregset_t elf_fpregset_t;
1686 typedef gregset_t elf_gregset_t;
1687 typedef thrmisc_t elf_thrmisc_t;
1688 #define ELF_KERN_PROC_MASK 0
1689 typedef struct kinfo_proc elf_kinfo_proc_t;
1690 typedef vm_offset_t elf_ps_strings_t;
1694 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1697 elf_prpsinfo_t *psinfo;
1699 p = (struct proc *)arg;
1701 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1702 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1703 psinfo->pr_version = PRPSINFO_VERSION;
1704 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1705 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1707 * XXX - We don't fill in the command line arguments properly
1710 strlcpy(psinfo->pr_psargs, p->p_comm,
1711 sizeof(psinfo->pr_psargs));
1713 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1714 free(psinfo, M_TEMP);
1716 *sizep = sizeof(*psinfo);
1720 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1723 elf_prstatus_t *status;
1725 td = (struct thread *)arg;
1727 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1728 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1729 status->pr_version = PRSTATUS_VERSION;
1730 status->pr_statussz = sizeof(elf_prstatus_t);
1731 status->pr_gregsetsz = sizeof(elf_gregset_t);
1732 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1733 status->pr_osreldate = osreldate;
1734 status->pr_cursig = td->td_proc->p_sig;
1735 status->pr_pid = td->td_tid;
1736 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1737 fill_regs32(td, &status->pr_reg);
1739 fill_regs(td, &status->pr_reg);
1741 sbuf_bcat(sb, status, sizeof(*status));
1742 free(status, M_TEMP);
1744 *sizep = sizeof(*status);
1748 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1751 elf_prfpregset_t *fpregset;
1753 td = (struct thread *)arg;
1755 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1756 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1757 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1758 fill_fpregs32(td, fpregset);
1760 fill_fpregs(td, fpregset);
1762 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1763 free(fpregset, M_TEMP);
1765 *sizep = sizeof(*fpregset);
1769 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1772 elf_thrmisc_t thrmisc;
1774 td = (struct thread *)arg;
1776 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1777 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1778 strcpy(thrmisc.pr_tname, td->td_name);
1779 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1781 *sizep = sizeof(thrmisc);
1785 * Allow for MD specific notes, as well as any MD
1786 * specific preparations for writing MI notes.
1789 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1795 td = (struct thread *)arg;
1797 if (size != 0 && sb != NULL)
1798 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1802 __elfN(dump_thread)(td, buf, &size);
1803 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1804 if (size != 0 && sb != NULL)
1805 sbuf_bcat(sb, buf, size);
1810 #ifdef KINFO_PROC_SIZE
1811 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1815 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1821 p = (struct proc *)arg;
1822 size = sizeof(structsize) + p->p_numthreads *
1823 sizeof(elf_kinfo_proc_t);
1826 KASSERT(*sizep == size, ("invalid size"));
1827 structsize = sizeof(elf_kinfo_proc_t);
1828 sbuf_bcat(sb, &structsize, sizeof(structsize));
1829 sx_slock(&proctree_lock);
1831 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1832 sx_sunlock(&proctree_lock);
1837 #ifdef KINFO_FILE_SIZE
1838 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1842 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1848 p = (struct proc *)arg;
1851 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1852 sbuf_set_drain(sb, sbuf_drain_count, &size);
1853 sbuf_bcat(sb, &structsize, sizeof(structsize));
1855 kern_proc_filedesc_out(p, sb, -1);
1860 structsize = sizeof(struct kinfo_file);
1861 sbuf_bcat(sb, &structsize, sizeof(structsize));
1863 kern_proc_filedesc_out(p, sb, -1);
1867 #ifdef KINFO_VMENTRY_SIZE
1868 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1872 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1878 p = (struct proc *)arg;
1881 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1882 sbuf_set_drain(sb, sbuf_drain_count, &size);
1883 sbuf_bcat(sb, &structsize, sizeof(structsize));
1885 kern_proc_vmmap_out(p, sb);
1890 structsize = sizeof(struct kinfo_vmentry);
1891 sbuf_bcat(sb, &structsize, sizeof(structsize));
1893 kern_proc_vmmap_out(p, sb);
1898 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1904 p = (struct proc *)arg;
1905 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1907 KASSERT(*sizep == size, ("invalid size"));
1908 structsize = sizeof(gid_t);
1909 sbuf_bcat(sb, &structsize, sizeof(structsize));
1910 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1917 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1923 p = (struct proc *)arg;
1924 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1926 KASSERT(*sizep == size, ("invalid size"));
1927 structsize = sizeof(p->p_fd->fd_cmask);
1928 sbuf_bcat(sb, &structsize, sizeof(structsize));
1929 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1935 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1938 struct rlimit rlim[RLIM_NLIMITS];
1942 p = (struct proc *)arg;
1943 size = sizeof(structsize) + sizeof(rlim);
1945 KASSERT(*sizep == size, ("invalid size"));
1946 structsize = sizeof(rlim);
1947 sbuf_bcat(sb, &structsize, sizeof(structsize));
1949 for (i = 0; i < RLIM_NLIMITS; i++)
1950 lim_rlimit(p, i, &rlim[i]);
1952 sbuf_bcat(sb, rlim, sizeof(rlim));
1958 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1964 p = (struct proc *)arg;
1965 size = sizeof(structsize) + sizeof(p->p_osrel);
1967 KASSERT(*sizep == size, ("invalid size"));
1968 structsize = sizeof(p->p_osrel);
1969 sbuf_bcat(sb, &structsize, sizeof(structsize));
1970 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1976 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1979 elf_ps_strings_t ps_strings;
1983 p = (struct proc *)arg;
1984 size = sizeof(structsize) + sizeof(ps_strings);
1986 KASSERT(*sizep == size, ("invalid size"));
1987 structsize = sizeof(ps_strings);
1988 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1989 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1991 ps_strings = p->p_sysent->sv_psstrings;
1993 sbuf_bcat(sb, &structsize, sizeof(structsize));
1994 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2000 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2006 p = (struct proc *)arg;
2009 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2010 sbuf_set_drain(sb, sbuf_drain_count, &size);
2011 sbuf_bcat(sb, &structsize, sizeof(structsize));
2013 proc_getauxv(curthread, p, sb);
2019 structsize = sizeof(Elf_Auxinfo);
2020 sbuf_bcat(sb, &structsize, sizeof(structsize));
2022 proc_getauxv(curthread, p, sb);
2028 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2029 int32_t *osrel, const Elf_Phdr *pnote)
2031 const Elf_Note *note, *note0, *note_end;
2032 const char *note_name;
2035 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2036 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2039 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2040 note_end = (const Elf_Note *)(imgp->image_header +
2041 pnote->p_offset + pnote->p_filesz);
2042 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2043 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2044 (const char *)note < sizeof(Elf_Note))
2046 if (note->n_namesz != checknote->hdr.n_namesz ||
2047 note->n_descsz != checknote->hdr.n_descsz ||
2048 note->n_type != checknote->hdr.n_type)
2050 note_name = (const char *)(note + 1);
2051 if (note_name + checknote->hdr.n_namesz >=
2052 (const char *)note_end || strncmp(checknote->vendor,
2053 note_name, checknote->hdr.n_namesz) != 0)
2057 * Fetch the osreldate for binary
2058 * from the ELF OSABI-note if necessary.
2060 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2061 checknote->trans_osrel != NULL)
2062 return (checknote->trans_osrel(note, osrel));
2066 note = (const Elf_Note *)((const char *)(note + 1) +
2067 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2068 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2075 * Try to find the appropriate ABI-note section for checknote,
2076 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2077 * first page of the image is searched, the same as for headers.
2080 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2083 const Elf_Phdr *phdr;
2084 const Elf_Ehdr *hdr;
2087 hdr = (const Elf_Ehdr *)imgp->image_header;
2088 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2090 for (i = 0; i < hdr->e_phnum; i++) {
2091 if (phdr[i].p_type == PT_NOTE &&
2092 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2100 * Tell kern_execve.c about it, with a little help from the linker.
2102 static struct execsw __elfN(execsw) = {
2103 __CONCAT(exec_, __elfN(imgact)),
2104 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2106 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2108 #ifdef COMPRESS_USER_CORES
2110 * Compress and write out a core segment for a user process.
2112 * 'inbuf' is the starting address of a VM segment in the process' address
2113 * space that is to be compressed and written out to the core file. 'dest_buf'
2114 * is a buffer in the kernel's address space. The segment is copied from
2115 * 'inbuf' to 'dest_buf' first before being processed by the compression
2116 * routine gzwrite(). This copying is necessary because the content of the VM
2117 * segment may change between the compression pass and the crc-computation pass
2118 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel.
2120 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2123 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2128 unsigned int chunk_len;
2131 if (inbuf != NULL) {
2132 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2133 copyin(inbuf, dest_buf, chunk_len);
2138 len_compressed = gzwrite(file, dest_buf, chunk_len);
2140 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2142 if ((unsigned int)len_compressed != chunk_len) {
2144 "compress_core: length mismatch (0x%x returned, "
2145 "0x%x expected)\n", len_compressed, chunk_len);
2146 EVENTHANDLER_INVOKE(app_coredump_error, td,
2147 "compress_core: length mismatch %x -> %x",
2148 chunk_len, len_compressed);
2158 #endif /* COMPRESS_USER_CORES */
2161 __elfN(trans_prot)(Elf_Word flags)
2167 prot |= VM_PROT_EXECUTE;
2169 prot |= VM_PROT_WRITE;
2171 prot |= VM_PROT_READ;
2172 #if __ELF_WORD_SIZE == 32
2173 #if defined(__amd64__)
2174 if (i386_read_exec && (flags & PF_R))
2175 prot |= VM_PROT_EXECUTE;
2182 __elfN(untrans_prot)(vm_prot_t prot)
2187 if (prot & VM_PROT_EXECUTE)
2189 if (prot & VM_PROT_READ)
2191 if (prot & VM_PROT_WRITE)