2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
59 #include <sys/sf_buf.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
84 #define ELF_NOTE_ROUNDSIZE 4
85 #define OLD_EI_BRAND 8
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89 const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
93 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107 #define CORE_BUF_SIZE (16 * 1024)
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
126 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
127 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
128 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
130 #if __ELF_WORD_SIZE == 32
131 #if defined(__amd64__)
132 int i386_read_exec = 0;
133 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
134 "enable execution from readable segments");
138 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
140 #define trunc_page_ps(va, ps) rounddown2(va, ps)
141 #define round_page_ps(va, ps) roundup2(va, ps)
142 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
144 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
146 Elf_Brandnote __elfN(freebsd_brandnote) = {
147 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
148 .hdr.n_descsz = sizeof(int32_t),
149 .hdr.n_type = NT_FREEBSD_ABI_TAG,
150 .vendor = FREEBSD_ABI_VENDOR,
151 .flags = BN_TRANSLATE_OSREL,
152 .trans_osrel = __elfN(freebsd_trans_osrel)
156 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
160 p = (uintptr_t)(note + 1);
161 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
162 *osrel = *(const int32_t *)(p);
167 static const char GNU_ABI_VENDOR[] = "GNU";
168 static int GNU_KFREEBSD_ABI_DESC = 3;
170 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
171 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
172 .hdr.n_descsz = 16, /* XXX at least 16 */
174 .vendor = GNU_ABI_VENDOR,
175 .flags = BN_TRANSLATE_OSREL,
176 .trans_osrel = kfreebsd_trans_osrel
180 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
182 const Elf32_Word *desc;
185 p = (uintptr_t)(note + 1);
186 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
188 desc = (const Elf32_Word *)p;
189 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
193 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
194 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
196 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
202 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
206 for (i = 0; i < MAX_BRANDS; i++) {
207 if (elf_brand_list[i] == NULL) {
208 elf_brand_list[i] = entry;
212 if (i == MAX_BRANDS) {
213 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
221 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
225 for (i = 0; i < MAX_BRANDS; i++) {
226 if (elf_brand_list[i] == entry) {
227 elf_brand_list[i] = NULL;
237 __elfN(brand_inuse)(Elf_Brandinfo *entry)
242 sx_slock(&allproc_lock);
243 FOREACH_PROC_IN_SYSTEM(p) {
244 if (p->p_sysent == entry->sysvec) {
249 sx_sunlock(&allproc_lock);
254 static Elf_Brandinfo *
255 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
256 int interp_name_len, int32_t *osrel)
258 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
259 Elf_Brandinfo *bi, *bi_m;
264 * We support four types of branding -- (1) the ELF EI_OSABI field
265 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
266 * branding w/in the ELF header, (3) path of the `interp_path'
267 * field, and (4) the ".note.ABI-tag" ELF section.
270 /* Look for an ".note.ABI-tag" ELF section */
272 for (i = 0; i < MAX_BRANDS; i++) {
273 bi = elf_brand_list[i];
276 if (hdr->e_machine == bi->machine && (bi->flags &
277 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
278 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
279 /* Give brand a chance to veto check_note's guess */
280 if (ret && bi->header_supported)
281 ret = bi->header_supported(imgp);
283 * If note checker claimed the binary, but the
284 * interpreter path in the image does not
285 * match default one for the brand, try to
286 * search for other brands with the same
287 * interpreter. Either there is better brand
288 * with the right interpreter, or, failing
289 * this, we return first brand which accepted
290 * our note and, optionally, header.
292 if (ret && bi_m == NULL && (strlen(bi->interp_path) +
293 1 != interp_name_len || strncmp(interp,
294 bi->interp_path, interp_name_len) != 0)) {
305 /* If the executable has a brand, search for it in the brand list. */
306 for (i = 0; i < MAX_BRANDS; i++) {
307 bi = elf_brand_list[i];
308 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
310 if (hdr->e_machine == bi->machine &&
311 (hdr->e_ident[EI_OSABI] == bi->brand ||
312 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
313 bi->compat_3_brand) == 0)) {
314 /* Looks good, but give brand a chance to veto */
315 if (!bi->header_supported ||
316 bi->header_supported(imgp)) {
318 * Again, prefer strictly matching
321 if (strlen(bi->interp_path) + 1 ==
322 interp_name_len && strncmp(interp,
323 bi->interp_path, interp_name_len) == 0)
333 /* No known brand, see if the header is recognized by any brand */
334 for (i = 0; i < MAX_BRANDS; i++) {
335 bi = elf_brand_list[i];
336 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
337 bi->header_supported == NULL)
339 if (hdr->e_machine == bi->machine) {
340 ret = bi->header_supported(imgp);
346 /* Lacking a known brand, search for a recognized interpreter. */
347 if (interp != NULL) {
348 for (i = 0; i < MAX_BRANDS; i++) {
349 bi = elf_brand_list[i];
350 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
352 if (hdr->e_machine == bi->machine &&
353 /* ELF image p_filesz includes terminating zero */
354 strlen(bi->interp_path) + 1 == interp_name_len &&
355 strncmp(interp, bi->interp_path, interp_name_len)
361 /* Lacking a recognized interpreter, try the default brand */
362 for (i = 0; i < MAX_BRANDS; i++) {
363 bi = elf_brand_list[i];
364 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
366 if (hdr->e_machine == bi->machine &&
367 __elfN(fallback_brand) == bi->brand)
374 __elfN(check_header)(const Elf_Ehdr *hdr)
380 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
381 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
382 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
383 hdr->e_phentsize != sizeof(Elf_Phdr) ||
384 hdr->e_version != ELF_TARG_VER)
388 * Make sure we have at least one brand for this machine.
391 for (i = 0; i < MAX_BRANDS; i++) {
392 bi = elf_brand_list[i];
393 if (bi != NULL && bi->machine == hdr->e_machine)
403 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
404 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
411 * Create the page if it doesn't exist yet. Ignore errors.
413 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
414 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
417 * Find the page from the underlying object.
419 if (object != NULL) {
420 sf = vm_imgact_map_page(object, offset);
422 return (KERN_FAILURE);
423 off = offset - trunc_page(offset);
424 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
426 vm_imgact_unmap_page(sf);
428 return (KERN_FAILURE);
431 return (KERN_SUCCESS);
435 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
436 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
442 int error, locked, rv;
444 if (start != trunc_page(start)) {
445 rv = __elfN(map_partial)(map, object, offset, start,
446 round_page(start), prot);
447 if (rv != KERN_SUCCESS)
449 offset += round_page(start) - start;
450 start = round_page(start);
452 if (end != round_page(end)) {
453 rv = __elfN(map_partial)(map, object, offset +
454 trunc_page(end) - start, trunc_page(end), end, prot);
455 if (rv != KERN_SUCCESS)
457 end = trunc_page(end);
460 return (KERN_SUCCESS);
461 if ((offset & PAGE_MASK) != 0) {
463 * The mapping is not page aligned. This means that we have
466 rv = vm_map_fixed(map, NULL, 0, start, end - start,
467 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
468 if (rv != KERN_SUCCESS)
471 return (KERN_SUCCESS);
472 for (; start < end; start += sz) {
473 sf = vm_imgact_map_page(object, offset);
475 return (KERN_FAILURE);
476 off = offset - trunc_page(offset);
478 if (sz > PAGE_SIZE - off)
479 sz = PAGE_SIZE - off;
480 error = copyout((caddr_t)sf_buf_kva(sf) + off,
482 vm_imgact_unmap_page(sf);
484 return (KERN_FAILURE);
488 vm_object_reference(object);
489 rv = vm_map_fixed(map, object, offset, start, end - start,
490 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
491 if (rv != KERN_SUCCESS) {
492 locked = VOP_ISLOCKED(imgp->vp);
493 VOP_UNLOCK(imgp->vp, 0);
494 vm_object_deallocate(object);
495 vn_lock(imgp->vp, locked | LK_RETRY);
499 return (KERN_SUCCESS);
503 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
504 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
511 vm_offset_t map_addr;
514 vm_ooffset_t file_addr;
517 * It's necessary to fail if the filsz + offset taken from the
518 * header is greater than the actual file pager object's size.
519 * If we were to allow this, then the vm_map_find() below would
520 * walk right off the end of the file object and into the ether.
522 * While I'm here, might as well check for something else that
523 * is invalid: filsz cannot be greater than memsz.
525 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
526 uprintf("elf_load_section: truncated ELF file\n");
530 object = imgp->object;
531 map = &imgp->proc->p_vmspace->vm_map;
532 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
533 file_addr = trunc_page_ps(offset, pagesize);
536 * We have two choices. We can either clear the data in the last page
537 * of an oversized mapping, or we can start the anon mapping a page
538 * early and copy the initialized data into that first page. We
542 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
544 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
547 /* cow flags: don't dump readonly sections in core */
548 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
549 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
551 rv = __elfN(map_insert)(imgp, map,
553 file_addr, /* file offset */
554 map_addr, /* virtual start */
555 map_addr + map_len,/* virtual end */
558 if (rv != KERN_SUCCESS)
561 /* we can stop now if we've covered it all */
562 if (memsz == filsz) {
569 * We have to get the remaining bit of the file into the first part
570 * of the oversized map segment. This is normally because the .data
571 * segment in the file is extended to provide bss. It's a neat idea
572 * to try and save a page, but it's a pain in the behind to implement.
574 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
575 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
576 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
579 /* This had damn well better be true! */
581 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
582 map_addr + map_len, VM_PROT_ALL, 0);
583 if (rv != KERN_SUCCESS) {
591 sf = vm_imgact_map_page(object, offset + filsz);
595 /* send the page fragment to user space */
596 off = trunc_page_ps(offset + filsz, pagesize) -
597 trunc_page(offset + filsz);
598 error = copyout((caddr_t)sf_buf_kva(sf) + off,
599 (caddr_t)map_addr, copy_len);
600 vm_imgact_unmap_page(sf);
607 * set it to the specified protection.
608 * XXX had better undo the damage from pasting over the cracks here!
610 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
611 map_len), prot, FALSE);
617 * Load the file "file" into memory. It may be either a shared object
620 * The "addr" reference parameter is in/out. On entry, it specifies
621 * the address where a shared object should be loaded. If the file is
622 * an executable, this value is ignored. On exit, "addr" specifies
623 * where the file was actually loaded.
625 * The "entry" reference parameter is out only. On exit, it specifies
626 * the entry point for the loaded file.
629 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
630 u_long *entry, size_t pagesize)
635 struct image_params image_params;
637 const Elf_Ehdr *hdr = NULL;
638 const Elf_Phdr *phdr = NULL;
639 struct nameidata *nd;
641 struct image_params *imgp;
644 u_long base_addr = 0;
645 int error, i, numsegs;
647 #ifdef CAPABILITY_MODE
649 * XXXJA: This check can go away once we are sufficiently confident
650 * that the checks in namei() are correct.
652 if (IN_CAPABILITY_MODE(curthread))
656 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
658 attr = &tempdata->attr;
659 imgp = &tempdata->image_params;
662 * Initialize part of the common data
666 imgp->firstpage = NULL;
667 imgp->image_header = NULL;
669 imgp->execlabel = NULL;
671 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
672 if ((error = namei(nd)) != 0) {
676 NDFREE(nd, NDF_ONLY_PNBUF);
677 imgp->vp = nd->ni_vp;
680 * Check permissions, modes, uid, etc on the file, and "open" it.
682 error = exec_check_permissions(imgp);
686 error = exec_map_first_page(imgp);
691 * Also make certain that the interpreter stays the same, so set
692 * its VV_TEXT flag, too.
694 VOP_SET_TEXT(nd->ni_vp);
696 imgp->object = nd->ni_vp->v_object;
698 hdr = (const Elf_Ehdr *)imgp->image_header;
699 if ((error = __elfN(check_header)(hdr)) != 0)
701 if (hdr->e_type == ET_DYN)
703 else if (hdr->e_type == ET_EXEC)
710 /* Only support headers that fit within first page for now */
711 if ((hdr->e_phoff > PAGE_SIZE) ||
712 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
717 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
718 if (!aligned(phdr, Elf_Addr)) {
723 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
724 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
725 /* Loadable segment */
726 prot = __elfN(trans_prot)(phdr[i].p_flags);
727 error = __elfN(load_section)(imgp, phdr[i].p_offset,
728 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
729 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
733 * Establish the base address if this is the
737 base_addr = trunc_page(phdr[i].p_vaddr +
743 *entry = (unsigned long)hdr->e_entry + rbase;
747 exec_unmap_first_page(imgp);
752 free(tempdata, M_TEMP);
758 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
762 const Elf_Phdr *phdr;
763 Elf_Auxargs *elf_auxargs;
764 struct vmspace *vmspace;
765 const char *err_str, *newinterp;
766 char *interp, *interp_buf, *path;
767 Elf_Brandinfo *brand_info;
768 struct sysentvec *sv;
770 u_long text_size, data_size, total_size, text_addr, data_addr;
771 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
773 int error, i, n, interp_name_len, have_interp;
775 hdr = (const Elf_Ehdr *)imgp->image_header;
778 * Do we have a valid ELF header ?
780 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
781 * if particular brand doesn't support it.
783 if (__elfN(check_header)(hdr) != 0 ||
784 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
788 * From here on down, we return an errno, not -1, as we've
789 * detected an ELF file.
792 if ((hdr->e_phoff > PAGE_SIZE) ||
793 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
794 /* Only support headers in first page for now */
795 uprintf("Program headers not in the first page\n");
798 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
799 if (!aligned(phdr, Elf_Addr)) {
800 uprintf("Unaligned program headers\n");
807 text_size = data_size = total_size = text_addr = data_addr = 0;
810 err_str = newinterp = NULL;
811 interp = interp_buf = NULL;
814 for (i = 0; i < hdr->e_phnum; i++) {
815 switch (phdr[i].p_type) {
818 baddr = phdr[i].p_vaddr;
822 /* Path to interpreter */
823 if (phdr[i].p_filesz > MAXPATHLEN) {
824 uprintf("Invalid PT_INTERP\n");
828 if (interp != NULL) {
829 uprintf("Multiple PT_INTERP headers\n");
833 interp_name_len = phdr[i].p_filesz;
834 if (phdr[i].p_offset > PAGE_SIZE ||
835 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
836 VOP_UNLOCK(imgp->vp, 0);
837 interp_buf = malloc(interp_name_len + 1, M_TEMP,
839 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
840 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
841 interp_name_len, phdr[i].p_offset,
842 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
845 uprintf("i/o error PT_INTERP\n");
848 interp_buf[interp_name_len] = '\0';
851 interp = __DECONST(char *, imgp->image_header) +
858 __elfN(trans_prot)(phdr[i].p_flags);
859 imgp->stack_sz = phdr[i].p_memsz;
864 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
866 if (brand_info == NULL) {
867 uprintf("ELF binary type \"%u\" not known.\n",
868 hdr->e_ident[EI_OSABI]);
872 if (hdr->e_type == ET_DYN) {
873 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
874 uprintf("Cannot execute shared object\n");
879 * Honour the base load address from the dso if it is
880 * non-zero for some reason.
883 et_dyn_addr = ET_DYN_LOAD_ADDR;
888 sv = brand_info->sysvec;
889 if (interp != NULL && brand_info->interp_newpath != NULL)
890 newinterp = brand_info->interp_newpath;
893 * Avoid a possible deadlock if the current address space is destroyed
894 * and that address space maps the locked vnode. In the common case,
895 * the locked vnode's v_usecount is decremented but remains greater
896 * than zero. Consequently, the vnode lock is not needed by vrele().
897 * However, in cases where the vnode lock is external, such as nullfs,
898 * v_usecount may become zero.
900 * The VV_TEXT flag prevents modifications to the executable while
901 * the vnode is unlocked.
903 VOP_UNLOCK(imgp->vp, 0);
905 error = exec_new_vmspace(imgp, sv);
906 imgp->proc->p_sysent = sv;
908 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
912 for (i = 0; i < hdr->e_phnum; i++) {
913 switch (phdr[i].p_type) {
914 case PT_LOAD: /* Loadable segment */
915 if (phdr[i].p_memsz == 0)
917 prot = __elfN(trans_prot)(phdr[i].p_flags);
918 error = __elfN(load_section)(imgp, phdr[i].p_offset,
919 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
920 phdr[i].p_memsz, phdr[i].p_filesz, prot,
926 * If this segment contains the program headers,
927 * remember their virtual address for the AT_PHDR
928 * aux entry. Static binaries don't usually include
931 if (phdr[i].p_offset == 0 &&
932 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
934 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
937 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
938 seg_size = round_page(phdr[i].p_memsz +
939 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
942 * Make the largest executable segment the official
943 * text segment and all others data.
945 * Note that obreak() assumes that data_addr +
946 * data_size == end of data load area, and the ELF
947 * file format expects segments to be sorted by
948 * address. If multiple data segments exist, the
949 * last one will be used.
952 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
953 text_size = seg_size;
954 text_addr = seg_addr;
956 data_size = seg_size;
957 data_addr = seg_addr;
959 total_size += seg_size;
961 case PT_PHDR: /* Program header table info */
962 proghdr = phdr[i].p_vaddr + et_dyn_addr;
969 if (data_addr == 0 && data_size == 0) {
970 data_addr = text_addr;
971 data_size = text_size;
974 entry = (u_long)hdr->e_entry + et_dyn_addr;
977 * Check limits. It should be safe to check the
978 * limits after loading the segments since we do
979 * not actually fault in all the segments pages.
981 PROC_LOCK(imgp->proc);
982 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
983 err_str = "Data segment size exceeds process limit";
984 else if (text_size > maxtsiz)
985 err_str = "Text segment size exceeds system limit";
986 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
987 err_str = "Total segment size exceeds process limit";
988 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
989 err_str = "Data segment size exceeds resource limit";
990 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
991 err_str = "Total segment size exceeds resource limit";
992 if (err_str != NULL) {
993 PROC_UNLOCK(imgp->proc);
994 uprintf("%s\n", err_str);
999 vmspace = imgp->proc->p_vmspace;
1000 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1001 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1002 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1003 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1006 * We load the dynamic linker where a userland call
1007 * to mmap(0, ...) would put it. The rationale behind this
1008 * calculation is that it leaves room for the heap to grow to
1009 * its maximum allowed size.
1011 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1013 PROC_UNLOCK(imgp->proc);
1015 imgp->entry_addr = entry;
1017 if (interp != NULL) {
1018 have_interp = FALSE;
1019 VOP_UNLOCK(imgp->vp, 0);
1020 if (brand_info->emul_path != NULL &&
1021 brand_info->emul_path[0] != '\0') {
1022 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1023 snprintf(path, MAXPATHLEN, "%s%s",
1024 brand_info->emul_path, interp);
1025 error = __elfN(load_file)(imgp->proc, path, &addr,
1026 &imgp->entry_addr, sv->sv_pagesize);
1031 if (!have_interp && newinterp != NULL &&
1032 (brand_info->interp_path == NULL ||
1033 strcmp(interp, brand_info->interp_path) == 0)) {
1034 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1035 &imgp->entry_addr, sv->sv_pagesize);
1040 error = __elfN(load_file)(imgp->proc, interp, &addr,
1041 &imgp->entry_addr, sv->sv_pagesize);
1043 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1045 uprintf("ELF interpreter %s not found, error %d\n",
1053 * Construct auxargs table (used by the fixup routine)
1055 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1056 elf_auxargs->execfd = -1;
1057 elf_auxargs->phdr = proghdr;
1058 elf_auxargs->phent = hdr->e_phentsize;
1059 elf_auxargs->phnum = hdr->e_phnum;
1060 elf_auxargs->pagesz = PAGE_SIZE;
1061 elf_auxargs->base = addr;
1062 elf_auxargs->flags = 0;
1063 elf_auxargs->entry = entry;
1064 elf_auxargs->hdr_eflags = hdr->e_flags;
1066 imgp->auxargs = elf_auxargs;
1067 imgp->interpreted = 0;
1068 imgp->reloc_base = addr;
1069 imgp->proc->p_osrel = osrel;
1072 free(interp_buf, M_TEMP);
1076 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1079 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1081 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1085 base = (Elf_Addr *)*stack_base;
1086 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1088 if (args->execfd != -1)
1089 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1090 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1091 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1092 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1093 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1094 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1095 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1096 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1098 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1100 if (imgp->execpathp != 0)
1101 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1102 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1103 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1104 if (imgp->canary != 0) {
1105 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1106 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1108 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1109 if (imgp->pagesizes != 0) {
1110 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1111 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1113 if (imgp->sysent->sv_timekeep_base != 0) {
1114 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1115 imgp->sysent->sv_timekeep_base);
1117 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1118 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1119 imgp->sysent->sv_stackprot);
1120 AUXARGS_ENTRY(pos, AT_NULL, 0);
1122 free(imgp->auxargs, M_TEMP);
1123 imgp->auxargs = NULL;
1126 suword(base, (long)imgp->args->argc);
1127 *stack_base = (register_t *)base;
1132 * Code for generating ELF core dumps.
1135 typedef void (*segment_callback)(vm_map_entry_t, void *);
1137 /* Closure for cb_put_phdr(). */
1138 struct phdr_closure {
1139 Elf_Phdr *phdr; /* Program header to fill in */
1140 Elf_Off offset; /* Offset of segment in core file */
1143 /* Closure for cb_size_segment(). */
1144 struct sseg_closure {
1145 int count; /* Count of writable segments. */
1146 size_t size; /* Total size of all writable segments. */
1149 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1152 int type; /* Note type. */
1153 outfunc_t outfunc; /* Output function. */
1154 void *outarg; /* Argument for the output function. */
1155 size_t outsize; /* Output size. */
1156 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1159 TAILQ_HEAD(note_info_list, note_info);
1161 /* Coredump output parameters. */
1162 struct coredump_params {
1164 struct ucred *active_cred;
1165 struct ucred *file_cred;
1168 struct gzio_stream *gzs;
1171 static void cb_put_phdr(vm_map_entry_t, void *);
1172 static void cb_size_segment(vm_map_entry_t, void *);
1173 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1175 static void each_writable_segment(struct thread *, segment_callback, void *);
1176 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1177 struct note_info_list *, size_t);
1178 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1180 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1181 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1182 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1183 static int sbuf_drain_core_output(void *, const char *, int);
1184 static int sbuf_drain_count(void *arg, const char *data, int len);
1186 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1187 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1188 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1189 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1190 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1191 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1192 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1193 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1194 static void note_procstat_files(void *, struct sbuf *, size_t *);
1195 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1196 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1197 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1198 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1199 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1202 extern int compress_user_cores_gzlevel;
1205 * Write out a core segment to the compression stream.
1208 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1214 chunk_len = MIN(len, CORE_BUF_SIZE);
1217 * We can get EFAULT error here.
1218 * In that case zero out the current chunk of the segment.
1220 error = copyin(base, buf, chunk_len);
1222 bzero(buf, chunk_len);
1223 error = gzio_write(p->gzs, buf, chunk_len);
1233 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1236 return (core_write((struct coredump_params *)arg, base, len, offset,
1242 core_write(struct coredump_params *p, const void *base, size_t len,
1243 off_t offset, enum uio_seg seg)
1246 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1247 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1248 p->active_cred, p->file_cred, NULL, p->td));
1252 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1259 return (compress_chunk(p, base, tmpbuf, len));
1262 * EFAULT is a non-fatal error that we can get, for example,
1263 * if the segment is backed by a file but extends beyond its
1266 error = core_write(p, base, len, offset, UIO_USERSPACE);
1267 if (error == EFAULT) {
1268 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1269 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1270 "for process %s\n", base, len, offset, curproc->p_comm);
1273 * Write a "real" zero byte at the end of the target region
1274 * in the case this is the last segment.
1275 * The intermediate space will be implicitly zero-filled.
1277 error = core_write(p, zero_region, 1, offset + len - 1,
1284 * Drain into a core file.
1287 sbuf_drain_core_output(void *arg, const char *data, int len)
1289 struct coredump_params *p;
1292 p = (struct coredump_params *)arg;
1295 * Some kern_proc out routines that print to this sbuf may
1296 * call us with the process lock held. Draining with the
1297 * non-sleepable lock held is unsafe. The lock is needed for
1298 * those routines when dumping a live process. In our case we
1299 * can safely release the lock before draining and acquire
1302 locked = PROC_LOCKED(p->td->td_proc);
1304 PROC_UNLOCK(p->td->td_proc);
1307 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1310 error = core_write(p, __DECONST(void *, data), len, p->offset,
1313 PROC_LOCK(p->td->td_proc);
1321 * Drain into a counter.
1324 sbuf_drain_count(void *arg, const char *data __unused, int len)
1328 sizep = (size_t *)arg;
1334 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1336 struct ucred *cred = td->td_ucred;
1338 struct sseg_closure seginfo;
1339 struct note_info_list notelst;
1340 struct coredump_params params;
1341 struct note_info *ninfo;
1343 size_t hdrsize, notesz, coresize;
1347 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1351 TAILQ_INIT(¬elst);
1353 /* Size the program segments. */
1356 each_writable_segment(td, cb_size_segment, &seginfo);
1359 * Collect info about the core file header area.
1361 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1362 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1363 coresize = round_page(hdrsize + notesz) + seginfo.size;
1365 /* Set up core dump parameters. */
1367 params.active_cred = cred;
1368 params.file_cred = NOCRED;
1375 PROC_LOCK(td->td_proc);
1376 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1377 PROC_UNLOCK(td->td_proc);
1384 if (coresize >= limit) {
1390 /* Create a compression stream if necessary. */
1392 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1393 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1394 if (params.gzs == NULL) {
1398 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1403 * Allocate memory for building the header, fill it up,
1404 * and write it out following the notes.
1406 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1407 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1410 /* Write the contents of all of the writable segments. */
1416 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1417 offset = round_page(hdrsize + notesz);
1418 for (i = 0; i < seginfo.count; i++) {
1419 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1420 php->p_filesz, offset, ¶ms, tmpbuf);
1423 offset += php->p_filesz;
1427 if (error == 0 && compress)
1428 error = gzio_flush(params.gzs);
1433 "Failed to write core file for process %s (error %d)\n",
1434 curproc->p_comm, error);
1440 free(tmpbuf, M_TEMP);
1441 if (params.gzs != NULL)
1442 gzio_fini(params.gzs);
1445 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1446 TAILQ_REMOVE(¬elst, ninfo, link);
1447 free(ninfo, M_TEMP);
1456 * A callback for each_writable_segment() to write out the segment's
1457 * program header entry.
1460 cb_put_phdr(entry, closure)
1461 vm_map_entry_t entry;
1464 struct phdr_closure *phc = (struct phdr_closure *)closure;
1465 Elf_Phdr *phdr = phc->phdr;
1467 phc->offset = round_page(phc->offset);
1469 phdr->p_type = PT_LOAD;
1470 phdr->p_offset = phc->offset;
1471 phdr->p_vaddr = entry->start;
1473 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1474 phdr->p_align = PAGE_SIZE;
1475 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1477 phc->offset += phdr->p_filesz;
1482 * A callback for each_writable_segment() to gather information about
1483 * the number of segments and their total size.
1486 cb_size_segment(entry, closure)
1487 vm_map_entry_t entry;
1490 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1493 ssc->size += entry->end - entry->start;
1497 * For each writable segment in the process's memory map, call the given
1498 * function with a pointer to the map entry and some arbitrary
1499 * caller-supplied data.
1502 each_writable_segment(td, func, closure)
1504 segment_callback func;
1507 struct proc *p = td->td_proc;
1508 vm_map_t map = &p->p_vmspace->vm_map;
1509 vm_map_entry_t entry;
1510 vm_object_t backing_object, object;
1511 boolean_t ignore_entry;
1513 vm_map_lock_read(map);
1514 for (entry = map->header.next; entry != &map->header;
1515 entry = entry->next) {
1517 * Don't dump inaccessible mappings, deal with legacy
1520 * Note that read-only segments related to the elf binary
1521 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1522 * need to arbitrarily ignore such segments.
1524 if (elf_legacy_coredump) {
1525 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1528 if ((entry->protection & VM_PROT_ALL) == 0)
1533 * Dont include memory segment in the coredump if
1534 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1535 * madvise(2). Do not dump submaps (i.e. parts of the
1538 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1541 if ((object = entry->object.vm_object) == NULL)
1544 /* Ignore memory-mapped devices and such things. */
1545 VM_OBJECT_RLOCK(object);
1546 while ((backing_object = object->backing_object) != NULL) {
1547 VM_OBJECT_RLOCK(backing_object);
1548 VM_OBJECT_RUNLOCK(object);
1549 object = backing_object;
1551 ignore_entry = object->type != OBJT_DEFAULT &&
1552 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1553 object->type != OBJT_PHYS;
1554 VM_OBJECT_RUNLOCK(object);
1558 (*func)(entry, closure);
1560 vm_map_unlock_read(map);
1564 * Write the core file header to the file, including padding up to
1565 * the page boundary.
1568 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1569 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1571 struct note_info *ninfo;
1575 /* Fill in the header. */
1576 bzero(hdr, hdrsize);
1577 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1579 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1580 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1581 sbuf_start_section(sb, NULL);
1582 sbuf_bcat(sb, hdr, hdrsize);
1583 TAILQ_FOREACH(ninfo, notelst, link)
1584 __elfN(putnote)(ninfo, sb);
1585 /* Align up to a page boundary for the program segments. */
1586 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1587 error = sbuf_finish(sb);
1594 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1604 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1607 * To have the debugger select the right thread (LWP) as the initial
1608 * thread, we dump the state of the thread passed to us in td first.
1609 * This is the thread that causes the core dump and thus likely to
1610 * be the right thread one wants to have selected in the debugger.
1613 while (thr != NULL) {
1614 size += register_note(list, NT_PRSTATUS,
1615 __elfN(note_prstatus), thr);
1616 size += register_note(list, NT_FPREGSET,
1617 __elfN(note_fpregset), thr);
1618 size += register_note(list, NT_THRMISC,
1619 __elfN(note_thrmisc), thr);
1620 size += register_note(list, -1,
1621 __elfN(note_threadmd), thr);
1623 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1624 TAILQ_NEXT(thr, td_plist);
1626 thr = TAILQ_NEXT(thr, td_plist);
1629 size += register_note(list, NT_PROCSTAT_PROC,
1630 __elfN(note_procstat_proc), p);
1631 size += register_note(list, NT_PROCSTAT_FILES,
1632 note_procstat_files, p);
1633 size += register_note(list, NT_PROCSTAT_VMMAP,
1634 note_procstat_vmmap, p);
1635 size += register_note(list, NT_PROCSTAT_GROUPS,
1636 note_procstat_groups, p);
1637 size += register_note(list, NT_PROCSTAT_UMASK,
1638 note_procstat_umask, p);
1639 size += register_note(list, NT_PROCSTAT_RLIMIT,
1640 note_procstat_rlimit, p);
1641 size += register_note(list, NT_PROCSTAT_OSREL,
1642 note_procstat_osrel, p);
1643 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1644 __elfN(note_procstat_psstrings), p);
1645 size += register_note(list, NT_PROCSTAT_AUXV,
1646 __elfN(note_procstat_auxv), p);
1652 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1657 struct phdr_closure phc;
1659 ehdr = (Elf_Ehdr *)hdr;
1660 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1662 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1663 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1664 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1665 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1666 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1667 ehdr->e_ident[EI_DATA] = ELF_DATA;
1668 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1669 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1670 ehdr->e_ident[EI_ABIVERSION] = 0;
1671 ehdr->e_ident[EI_PAD] = 0;
1672 ehdr->e_type = ET_CORE;
1673 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1674 ehdr->e_machine = ELF_ARCH32;
1676 ehdr->e_machine = ELF_ARCH;
1678 ehdr->e_version = EV_CURRENT;
1680 ehdr->e_phoff = sizeof(Elf_Ehdr);
1682 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1683 ehdr->e_phentsize = sizeof(Elf_Phdr);
1684 ehdr->e_phnum = numsegs + 1;
1685 ehdr->e_shentsize = sizeof(Elf_Shdr);
1687 ehdr->e_shstrndx = SHN_UNDEF;
1690 * Fill in the program header entries.
1693 /* The note segement. */
1694 phdr->p_type = PT_NOTE;
1695 phdr->p_offset = hdrsize;
1698 phdr->p_filesz = notesz;
1700 phdr->p_flags = PF_R;
1701 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1704 /* All the writable segments from the program. */
1706 phc.offset = round_page(hdrsize + notesz);
1707 each_writable_segment(td, cb_put_phdr, &phc);
1711 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1713 struct note_info *ninfo;
1714 size_t size, notesize;
1717 out(arg, NULL, &size);
1718 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1720 ninfo->outfunc = out;
1721 ninfo->outarg = arg;
1722 ninfo->outsize = size;
1723 TAILQ_INSERT_TAIL(list, ninfo, link);
1728 notesize = sizeof(Elf_Note) + /* note header */
1729 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1731 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1737 append_note_data(const void *src, void *dst, size_t len)
1741 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1743 bcopy(src, dst, len);
1744 bzero((char *)dst + len, padded_len - len);
1746 return (padded_len);
1750 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1758 note = (Elf_Note *)buf;
1759 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1760 note->n_descsz = size;
1761 note->n_type = type;
1762 buf += sizeof(*note);
1763 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1764 sizeof(FREEBSD_ABI_VENDOR));
1765 append_note_data(src, buf, size);
1770 notesize = sizeof(Elf_Note) + /* note header */
1771 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1773 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1779 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1782 ssize_t old_len, sect_len;
1783 size_t new_len, descsz, i;
1785 if (ninfo->type == -1) {
1786 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1790 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1791 note.n_descsz = ninfo->outsize;
1792 note.n_type = ninfo->type;
1794 sbuf_bcat(sb, ¬e, sizeof(note));
1795 sbuf_start_section(sb, &old_len);
1796 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1797 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1798 if (note.n_descsz == 0)
1800 sbuf_start_section(sb, &old_len);
1801 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1802 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1806 new_len = (size_t)sect_len;
1807 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1808 if (new_len < descsz) {
1810 * It is expected that individual note emitters will correctly
1811 * predict their expected output size and fill up to that size
1812 * themselves, padding in a format-specific way if needed.
1813 * However, in case they don't, just do it here with zeros.
1815 for (i = 0; i < descsz - new_len; i++)
1817 } else if (new_len > descsz) {
1819 * We can't always truncate sb -- we may have drained some
1822 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1823 "read it (%zu > %zu). Since it is longer than "
1824 "expected, this coredump's notes are corrupt. THIS "
1825 "IS A BUG in the note_procstat routine for type %u.\n",
1826 __func__, (unsigned)note.n_type, new_len, descsz,
1827 (unsigned)note.n_type));
1832 * Miscellaneous note out functions.
1835 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1836 #include <compat/freebsd32/freebsd32.h>
1838 typedef struct prstatus32 elf_prstatus_t;
1839 typedef struct prpsinfo32 elf_prpsinfo_t;
1840 typedef struct fpreg32 elf_prfpregset_t;
1841 typedef struct fpreg32 elf_fpregset_t;
1842 typedef struct reg32 elf_gregset_t;
1843 typedef struct thrmisc32 elf_thrmisc_t;
1844 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1845 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1846 typedef uint32_t elf_ps_strings_t;
1848 typedef prstatus_t elf_prstatus_t;
1849 typedef prpsinfo_t elf_prpsinfo_t;
1850 typedef prfpregset_t elf_prfpregset_t;
1851 typedef prfpregset_t elf_fpregset_t;
1852 typedef gregset_t elf_gregset_t;
1853 typedef thrmisc_t elf_thrmisc_t;
1854 #define ELF_KERN_PROC_MASK 0
1855 typedef struct kinfo_proc elf_kinfo_proc_t;
1856 typedef vm_offset_t elf_ps_strings_t;
1860 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1866 elf_prpsinfo_t *psinfo;
1869 p = (struct proc *)arg;
1871 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1872 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1873 psinfo->pr_version = PRPSINFO_VERSION;
1874 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1875 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1877 if (p->p_args != NULL) {
1878 len = sizeof(psinfo->pr_psargs) - 1;
1879 if (len > p->p_args->ar_length)
1880 len = p->p_args->ar_length;
1881 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1887 sbuf_new(&sbarg, psinfo->pr_psargs,
1888 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1889 error = proc_getargv(curthread, p, &sbarg);
1891 if (sbuf_finish(&sbarg) == 0)
1892 len = sbuf_len(&sbarg) - 1;
1894 len = sizeof(psinfo->pr_psargs) - 1;
1895 sbuf_delete(&sbarg);
1897 if (error || len == 0)
1898 strlcpy(psinfo->pr_psargs, p->p_comm,
1899 sizeof(psinfo->pr_psargs));
1901 KASSERT(len < sizeof(psinfo->pr_psargs),
1902 ("len is too long: %zu vs %zu", len,
1903 sizeof(psinfo->pr_psargs)));
1904 cp = psinfo->pr_psargs;
1907 cp = memchr(cp, '\0', end - cp);
1913 psinfo->pr_pid = p->p_pid;
1914 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1915 free(psinfo, M_TEMP);
1917 *sizep = sizeof(*psinfo);
1921 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1924 elf_prstatus_t *status;
1926 td = (struct thread *)arg;
1928 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1929 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1930 status->pr_version = PRSTATUS_VERSION;
1931 status->pr_statussz = sizeof(elf_prstatus_t);
1932 status->pr_gregsetsz = sizeof(elf_gregset_t);
1933 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1934 status->pr_osreldate = osreldate;
1935 status->pr_cursig = td->td_proc->p_sig;
1936 status->pr_pid = td->td_tid;
1937 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1938 fill_regs32(td, &status->pr_reg);
1940 fill_regs(td, &status->pr_reg);
1942 sbuf_bcat(sb, status, sizeof(*status));
1943 free(status, M_TEMP);
1945 *sizep = sizeof(*status);
1949 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1952 elf_prfpregset_t *fpregset;
1954 td = (struct thread *)arg;
1956 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1957 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1958 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1959 fill_fpregs32(td, fpregset);
1961 fill_fpregs(td, fpregset);
1963 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1964 free(fpregset, M_TEMP);
1966 *sizep = sizeof(*fpregset);
1970 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1973 elf_thrmisc_t thrmisc;
1975 td = (struct thread *)arg;
1977 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1978 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1979 strcpy(thrmisc.pr_tname, td->td_name);
1980 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1982 *sizep = sizeof(thrmisc);
1986 * Allow for MD specific notes, as well as any MD
1987 * specific preparations for writing MI notes.
1990 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1996 td = (struct thread *)arg;
1998 if (size != 0 && sb != NULL)
1999 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2003 __elfN(dump_thread)(td, buf, &size);
2004 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2005 if (size != 0 && sb != NULL)
2006 sbuf_bcat(sb, buf, size);
2011 #ifdef KINFO_PROC_SIZE
2012 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2016 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2022 p = (struct proc *)arg;
2023 size = sizeof(structsize) + p->p_numthreads *
2024 sizeof(elf_kinfo_proc_t);
2027 KASSERT(*sizep == size, ("invalid size"));
2028 structsize = sizeof(elf_kinfo_proc_t);
2029 sbuf_bcat(sb, &structsize, sizeof(structsize));
2030 sx_slock(&proctree_lock);
2032 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2033 sx_sunlock(&proctree_lock);
2038 #ifdef KINFO_FILE_SIZE
2039 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2043 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2046 size_t size, sect_sz, i;
2047 ssize_t start_len, sect_len;
2048 int structsize, filedesc_flags;
2050 if (coredump_pack_fileinfo)
2051 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2055 p = (struct proc *)arg;
2056 structsize = sizeof(struct kinfo_file);
2059 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2060 sbuf_set_drain(sb, sbuf_drain_count, &size);
2061 sbuf_bcat(sb, &structsize, sizeof(structsize));
2063 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2068 sbuf_start_section(sb, &start_len);
2070 sbuf_bcat(sb, &structsize, sizeof(structsize));
2072 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2075 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2080 KASSERT(sect_sz <= *sizep,
2081 ("kern_proc_filedesc_out did not respect maxlen; "
2082 "requested %zu, got %zu", *sizep - sizeof(structsize),
2083 sect_sz - sizeof(structsize)));
2085 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2090 #ifdef KINFO_VMENTRY_SIZE
2091 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2095 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2099 int structsize, vmmap_flags;
2101 if (coredump_pack_vmmapinfo)
2102 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2106 p = (struct proc *)arg;
2107 structsize = sizeof(struct kinfo_vmentry);
2110 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2111 sbuf_set_drain(sb, sbuf_drain_count, &size);
2112 sbuf_bcat(sb, &structsize, sizeof(structsize));
2114 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2119 sbuf_bcat(sb, &structsize, sizeof(structsize));
2121 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2127 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2133 p = (struct proc *)arg;
2134 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2136 KASSERT(*sizep == size, ("invalid size"));
2137 structsize = sizeof(gid_t);
2138 sbuf_bcat(sb, &structsize, sizeof(structsize));
2139 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2146 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2152 p = (struct proc *)arg;
2153 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2155 KASSERT(*sizep == size, ("invalid size"));
2156 structsize = sizeof(p->p_fd->fd_cmask);
2157 sbuf_bcat(sb, &structsize, sizeof(structsize));
2158 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2164 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2167 struct rlimit rlim[RLIM_NLIMITS];
2171 p = (struct proc *)arg;
2172 size = sizeof(structsize) + sizeof(rlim);
2174 KASSERT(*sizep == size, ("invalid size"));
2175 structsize = sizeof(rlim);
2176 sbuf_bcat(sb, &structsize, sizeof(structsize));
2178 for (i = 0; i < RLIM_NLIMITS; i++)
2179 lim_rlimit_proc(p, i, &rlim[i]);
2181 sbuf_bcat(sb, rlim, sizeof(rlim));
2187 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2193 p = (struct proc *)arg;
2194 size = sizeof(structsize) + sizeof(p->p_osrel);
2196 KASSERT(*sizep == size, ("invalid size"));
2197 structsize = sizeof(p->p_osrel);
2198 sbuf_bcat(sb, &structsize, sizeof(structsize));
2199 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2205 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2208 elf_ps_strings_t ps_strings;
2212 p = (struct proc *)arg;
2213 size = sizeof(structsize) + sizeof(ps_strings);
2215 KASSERT(*sizep == size, ("invalid size"));
2216 structsize = sizeof(ps_strings);
2217 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2218 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2220 ps_strings = p->p_sysent->sv_psstrings;
2222 sbuf_bcat(sb, &structsize, sizeof(structsize));
2223 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2229 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2235 p = (struct proc *)arg;
2238 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2239 sbuf_set_drain(sb, sbuf_drain_count, &size);
2240 sbuf_bcat(sb, &structsize, sizeof(structsize));
2242 proc_getauxv(curthread, p, sb);
2248 structsize = sizeof(Elf_Auxinfo);
2249 sbuf_bcat(sb, &structsize, sizeof(structsize));
2251 proc_getauxv(curthread, p, sb);
2257 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2258 int32_t *osrel, const Elf_Phdr *pnote)
2260 const Elf_Note *note, *note0, *note_end;
2261 const char *note_name;
2266 /* We need some limit, might as well use PAGE_SIZE. */
2267 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2269 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2270 if (pnote->p_offset > PAGE_SIZE ||
2271 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2272 VOP_UNLOCK(imgp->vp, 0);
2273 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2274 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2275 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2276 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2277 curthread->td_ucred, NOCRED, NULL, curthread);
2279 uprintf("i/o error PT_NOTE\n");
2283 note = note0 = (const Elf_Note *)buf;
2284 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2286 note = note0 = (const Elf_Note *)(imgp->image_header +
2288 note_end = (const Elf_Note *)(imgp->image_header +
2289 pnote->p_offset + pnote->p_filesz);
2292 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2293 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2294 (const char *)note < sizeof(Elf_Note)) {
2298 if (note->n_namesz != checknote->hdr.n_namesz ||
2299 note->n_descsz != checknote->hdr.n_descsz ||
2300 note->n_type != checknote->hdr.n_type)
2302 note_name = (const char *)(note + 1);
2303 if (note_name + checknote->hdr.n_namesz >=
2304 (const char *)note_end || strncmp(checknote->vendor,
2305 note_name, checknote->hdr.n_namesz) != 0)
2309 * Fetch the osreldate for binary
2310 * from the ELF OSABI-note if necessary.
2312 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2313 checknote->trans_osrel != NULL) {
2314 res = checknote->trans_osrel(note, osrel);
2320 note = (const Elf_Note *)((const char *)(note + 1) +
2321 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2322 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2331 * Try to find the appropriate ABI-note section for checknote,
2332 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2333 * first page of the image is searched, the same as for headers.
2336 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2339 const Elf_Phdr *phdr;
2340 const Elf_Ehdr *hdr;
2343 hdr = (const Elf_Ehdr *)imgp->image_header;
2344 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2346 for (i = 0; i < hdr->e_phnum; i++) {
2347 if (phdr[i].p_type == PT_NOTE &&
2348 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2356 * Tell kern_execve.c about it, with a little help from the linker.
2358 static struct execsw __elfN(execsw) = {
2359 __CONCAT(exec_, __elfN(imgact)),
2360 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2362 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2365 __elfN(trans_prot)(Elf_Word flags)
2371 prot |= VM_PROT_EXECUTE;
2373 prot |= VM_PROT_WRITE;
2375 prot |= VM_PROT_READ;
2376 #if __ELF_WORD_SIZE == 32
2377 #if defined(__amd64__)
2378 if (i386_read_exec && (flags & PF_R))
2379 prot |= VM_PROT_EXECUTE;
2386 __elfN(untrans_prot)(vm_prot_t prot)
2391 if (prot & VM_PROT_EXECUTE)
2393 if (prot & VM_PROT_READ)
2395 if (prot & VM_PROT_WRITE)