2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
59 #include <sys/sf_buf.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
84 #define ELF_NOTE_ROUNDSIZE 4
85 #define OLD_EI_BRAND 8
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89 const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
93 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107 #define CORE_BUF_SIZE (16 * 1024)
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
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 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
313 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0)) {
314 /* Looks good, but give brand a chance to veto */
315 if (!bi->header_supported || bi->header_supported(imgp))
320 /* No known brand, see if the header is recognized by any brand */
321 for (i = 0; i < MAX_BRANDS; i++) {
322 bi = elf_brand_list[i];
323 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
324 bi->header_supported == NULL)
326 if (hdr->e_machine == bi->machine) {
327 ret = bi->header_supported(imgp);
333 /* Lacking a known brand, search for a recognized interpreter. */
334 if (interp != NULL) {
335 for (i = 0; i < MAX_BRANDS; i++) {
336 bi = elf_brand_list[i];
337 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
339 if (hdr->e_machine == bi->machine &&
340 /* ELF image p_filesz includes terminating zero */
341 strlen(bi->interp_path) + 1 == interp_name_len &&
342 strncmp(interp, bi->interp_path, interp_name_len)
348 /* Lacking a recognized interpreter, try the default brand */
349 for (i = 0; i < MAX_BRANDS; i++) {
350 bi = elf_brand_list[i];
351 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
353 if (hdr->e_machine == bi->machine &&
354 __elfN(fallback_brand) == bi->brand)
361 __elfN(check_header)(const Elf_Ehdr *hdr)
367 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
368 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
369 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
370 hdr->e_phentsize != sizeof(Elf_Phdr) ||
371 hdr->e_version != ELF_TARG_VER)
375 * Make sure we have at least one brand for this machine.
378 for (i = 0; i < MAX_BRANDS; i++) {
379 bi = elf_brand_list[i];
380 if (bi != NULL && bi->machine == hdr->e_machine)
390 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
391 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
398 * Create the page if it doesn't exist yet. Ignore errors.
401 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
402 VM_PROT_ALL, VM_PROT_ALL, 0);
406 * Find the page from the underlying object.
409 sf = vm_imgact_map_page(object, offset);
411 return (KERN_FAILURE);
412 off = offset - trunc_page(offset);
413 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
415 vm_imgact_unmap_page(sf);
417 return (KERN_FAILURE);
421 return (KERN_SUCCESS);
425 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
426 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
433 if (start != trunc_page(start)) {
434 rv = __elfN(map_partial)(map, object, offset, start,
435 round_page(start), prot);
438 offset += round_page(start) - start;
439 start = round_page(start);
441 if (end != round_page(end)) {
442 rv = __elfN(map_partial)(map, object, offset +
443 trunc_page(end) - start, trunc_page(end), end, prot);
446 end = trunc_page(end);
449 if (offset & PAGE_MASK) {
451 * The mapping is not page aligned. This means we have
452 * to copy the data. Sigh.
454 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
455 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
457 if (rv != KERN_SUCCESS)
460 return (KERN_SUCCESS);
461 for (; start < end; start += sz) {
462 sf = vm_imgact_map_page(object, offset);
464 return (KERN_FAILURE);
465 off = offset - trunc_page(offset);
467 if (sz > PAGE_SIZE - off)
468 sz = PAGE_SIZE - off;
469 error = copyout((caddr_t)sf_buf_kva(sf) + off,
471 vm_imgact_unmap_page(sf);
473 return (KERN_FAILURE);
478 vm_object_reference(object);
480 rv = vm_map_insert(map, object, offset, start, end,
481 prot, VM_PROT_ALL, cow);
483 if (rv != KERN_SUCCESS)
484 vm_object_deallocate(object);
488 return (KERN_SUCCESS);
493 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
494 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
501 vm_offset_t map_addr;
504 vm_offset_t file_addr;
507 * It's necessary to fail if the filsz + offset taken from the
508 * header is greater than the actual file pager object's size.
509 * If we were to allow this, then the vm_map_find() below would
510 * walk right off the end of the file object and into the ether.
512 * While I'm here, might as well check for something else that
513 * is invalid: filsz cannot be greater than memsz.
515 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
516 uprintf("elf_load_section: truncated ELF file\n");
520 object = imgp->object;
521 map = &imgp->proc->p_vmspace->vm_map;
522 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
523 file_addr = trunc_page_ps(offset, pagesize);
526 * We have two choices. We can either clear the data in the last page
527 * of an oversized mapping, or we can start the anon mapping a page
528 * early and copy the initialized data into that first page. We
529 * choose the second..
532 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
534 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
537 /* cow flags: don't dump readonly sections in core */
538 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
539 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
541 rv = __elfN(map_insert)(map,
543 file_addr, /* file offset */
544 map_addr, /* virtual start */
545 map_addr + map_len,/* virtual end */
548 if (rv != KERN_SUCCESS)
551 /* we can stop now if we've covered it all */
552 if (memsz == filsz) {
559 * We have to get the remaining bit of the file into the first part
560 * of the oversized map segment. This is normally because the .data
561 * segment in the file is extended to provide bss. It's a neat idea
562 * to try and save a page, but it's a pain in the behind to implement.
564 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
565 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
566 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
569 /* This had damn well better be true! */
571 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
572 map_len, VM_PROT_ALL, 0);
573 if (rv != KERN_SUCCESS) {
581 sf = vm_imgact_map_page(object, offset + filsz);
585 /* send the page fragment to user space */
586 off = trunc_page_ps(offset + filsz, pagesize) -
587 trunc_page(offset + filsz);
588 error = copyout((caddr_t)sf_buf_kva(sf) + off,
589 (caddr_t)map_addr, copy_len);
590 vm_imgact_unmap_page(sf);
597 * set it to the specified protection.
598 * XXX had better undo the damage from pasting over the cracks here!
600 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
601 map_len), prot, FALSE);
607 * Load the file "file" into memory. It may be either a shared object
610 * The "addr" reference parameter is in/out. On entry, it specifies
611 * the address where a shared object should be loaded. If the file is
612 * an executable, this value is ignored. On exit, "addr" specifies
613 * where the file was actually loaded.
615 * The "entry" reference parameter is out only. On exit, it specifies
616 * the entry point for the loaded file.
619 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
620 u_long *entry, size_t pagesize)
625 struct image_params image_params;
627 const Elf_Ehdr *hdr = NULL;
628 const Elf_Phdr *phdr = NULL;
629 struct nameidata *nd;
631 struct image_params *imgp;
634 u_long base_addr = 0;
635 int error, i, numsegs;
637 #ifdef CAPABILITY_MODE
639 * XXXJA: This check can go away once we are sufficiently confident
640 * that the checks in namei() are correct.
642 if (IN_CAPABILITY_MODE(curthread))
646 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
648 attr = &tempdata->attr;
649 imgp = &tempdata->image_params;
652 * Initialize part of the common data
656 imgp->firstpage = NULL;
657 imgp->image_header = NULL;
659 imgp->execlabel = NULL;
661 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
662 if ((error = namei(nd)) != 0) {
666 NDFREE(nd, NDF_ONLY_PNBUF);
667 imgp->vp = nd->ni_vp;
670 * Check permissions, modes, uid, etc on the file, and "open" it.
672 error = exec_check_permissions(imgp);
676 error = exec_map_first_page(imgp);
681 * Also make certain that the interpreter stays the same, so set
682 * its VV_TEXT flag, too.
684 VOP_SET_TEXT(nd->ni_vp);
686 imgp->object = nd->ni_vp->v_object;
688 hdr = (const Elf_Ehdr *)imgp->image_header;
689 if ((error = __elfN(check_header)(hdr)) != 0)
691 if (hdr->e_type == ET_DYN)
693 else if (hdr->e_type == ET_EXEC)
700 /* Only support headers that fit within first page for now */
701 if ((hdr->e_phoff > PAGE_SIZE) ||
702 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
707 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
708 if (!aligned(phdr, Elf_Addr)) {
713 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
714 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
715 /* Loadable segment */
716 prot = __elfN(trans_prot)(phdr[i].p_flags);
717 error = __elfN(load_section)(imgp, phdr[i].p_offset,
718 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
719 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
723 * Establish the base address if this is the
727 base_addr = trunc_page(phdr[i].p_vaddr +
733 *entry = (unsigned long)hdr->e_entry + rbase;
737 exec_unmap_first_page(imgp);
742 free(tempdata, M_TEMP);
748 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
752 const Elf_Phdr *phdr;
753 Elf_Auxargs *elf_auxargs;
754 struct vmspace *vmspace;
755 const char *err_str, *newinterp;
756 char *interp, *interp_buf, *path;
757 Elf_Brandinfo *brand_info;
758 struct sysentvec *sv;
760 u_long text_size, data_size, total_size, text_addr, data_addr;
761 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
763 int error, i, n, interp_name_len, have_interp;
765 hdr = (const Elf_Ehdr *)imgp->image_header;
768 * Do we have a valid ELF header ?
770 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
771 * if particular brand doesn't support it.
773 if (__elfN(check_header)(hdr) != 0 ||
774 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
778 * From here on down, we return an errno, not -1, as we've
779 * detected an ELF file.
782 if ((hdr->e_phoff > PAGE_SIZE) ||
783 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
784 /* Only support headers in first page for now */
785 uprintf("Program headers not in the first page\n");
788 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
789 if (!aligned(phdr, Elf_Addr)) {
790 uprintf("Unaligned program headers\n");
797 text_size = data_size = total_size = text_addr = data_addr = 0;
800 err_str = newinterp = NULL;
801 interp = interp_buf = NULL;
804 for (i = 0; i < hdr->e_phnum; i++) {
805 switch (phdr[i].p_type) {
808 baddr = phdr[i].p_vaddr;
812 /* Path to interpreter */
813 if (phdr[i].p_filesz > MAXPATHLEN) {
814 uprintf("Invalid PT_INTERP\n");
818 if (interp != NULL) {
819 uprintf("Multiple PT_INTERP headers\n");
823 interp_name_len = phdr[i].p_filesz;
824 if (phdr[i].p_offset > PAGE_SIZE ||
825 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
826 VOP_UNLOCK(imgp->vp, 0);
827 interp_buf = malloc(interp_name_len + 1, M_TEMP,
829 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
830 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
831 interp_name_len, phdr[i].p_offset,
832 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
835 uprintf("i/o error PT_INTERP\n");
838 interp_buf[interp_name_len] = '\0';
841 interp = __DECONST(char *, imgp->image_header) +
848 __elfN(trans_prot)(phdr[i].p_flags);
849 imgp->stack_sz = phdr[i].p_memsz;
854 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
856 if (brand_info == NULL) {
857 uprintf("ELF binary type \"%u\" not known.\n",
858 hdr->e_ident[EI_OSABI]);
862 if (hdr->e_type == ET_DYN) {
863 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
864 uprintf("Cannot execute shared object\n");
869 * Honour the base load address from the dso if it is
870 * non-zero for some reason.
873 et_dyn_addr = ET_DYN_LOAD_ADDR;
878 sv = brand_info->sysvec;
879 if (interp != NULL && brand_info->interp_newpath != NULL)
880 newinterp = brand_info->interp_newpath;
883 * Avoid a possible deadlock if the current address space is destroyed
884 * and that address space maps the locked vnode. In the common case,
885 * the locked vnode's v_usecount is decremented but remains greater
886 * than zero. Consequently, the vnode lock is not needed by vrele().
887 * However, in cases where the vnode lock is external, such as nullfs,
888 * v_usecount may become zero.
890 * The VV_TEXT flag prevents modifications to the executable while
891 * the vnode is unlocked.
893 VOP_UNLOCK(imgp->vp, 0);
895 error = exec_new_vmspace(imgp, sv);
896 imgp->proc->p_sysent = sv;
898 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
902 for (i = 0; i < hdr->e_phnum; i++) {
903 switch (phdr[i].p_type) {
904 case PT_LOAD: /* Loadable segment */
905 if (phdr[i].p_memsz == 0)
907 prot = __elfN(trans_prot)(phdr[i].p_flags);
908 error = __elfN(load_section)(imgp, phdr[i].p_offset,
909 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
910 phdr[i].p_memsz, phdr[i].p_filesz, prot,
916 * If this segment contains the program headers,
917 * remember their virtual address for the AT_PHDR
918 * aux entry. Static binaries don't usually include
921 if (phdr[i].p_offset == 0 &&
922 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
924 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
927 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
928 seg_size = round_page(phdr[i].p_memsz +
929 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
932 * Make the largest executable segment the official
933 * text segment and all others data.
935 * Note that obreak() assumes that data_addr +
936 * data_size == end of data load area, and the ELF
937 * file format expects segments to be sorted by
938 * address. If multiple data segments exist, the
939 * last one will be used.
942 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
943 text_size = seg_size;
944 text_addr = seg_addr;
946 data_size = seg_size;
947 data_addr = seg_addr;
949 total_size += seg_size;
951 case PT_PHDR: /* Program header table info */
952 proghdr = phdr[i].p_vaddr + et_dyn_addr;
959 if (data_addr == 0 && data_size == 0) {
960 data_addr = text_addr;
961 data_size = text_size;
964 entry = (u_long)hdr->e_entry + et_dyn_addr;
967 * Check limits. It should be safe to check the
968 * limits after loading the segments since we do
969 * not actually fault in all the segments pages.
971 PROC_LOCK(imgp->proc);
972 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
973 err_str = "Data segment size exceeds process limit";
974 else if (text_size > maxtsiz)
975 err_str = "Text segment size exceeds system limit";
976 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
977 err_str = "Total segment size exceeds process limit";
978 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
979 err_str = "Data segment size exceeds resource limit";
980 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
981 err_str = "Total segment size exceeds resource limit";
982 if (err_str != NULL) {
983 PROC_UNLOCK(imgp->proc);
984 uprintf("%s\n", err_str);
989 vmspace = imgp->proc->p_vmspace;
990 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
991 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
992 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
993 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
996 * We load the dynamic linker where a userland call
997 * to mmap(0, ...) would put it. The rationale behind this
998 * calculation is that it leaves room for the heap to grow to
999 * its maximum allowed size.
1001 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1003 PROC_UNLOCK(imgp->proc);
1005 imgp->entry_addr = entry;
1007 if (interp != NULL) {
1008 have_interp = FALSE;
1009 VOP_UNLOCK(imgp->vp, 0);
1010 if (brand_info->emul_path != NULL &&
1011 brand_info->emul_path[0] != '\0') {
1012 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1013 snprintf(path, MAXPATHLEN, "%s%s",
1014 brand_info->emul_path, interp);
1015 error = __elfN(load_file)(imgp->proc, path, &addr,
1016 &imgp->entry_addr, sv->sv_pagesize);
1021 if (!have_interp && newinterp != NULL &&
1022 (brand_info->interp_path == NULL ||
1023 strcmp(interp, brand_info->interp_path) == 0)) {
1024 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1025 &imgp->entry_addr, sv->sv_pagesize);
1030 error = __elfN(load_file)(imgp->proc, interp, &addr,
1031 &imgp->entry_addr, sv->sv_pagesize);
1033 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1035 uprintf("ELF interpreter %s not found, error %d\n",
1043 * Construct auxargs table (used by the fixup routine)
1045 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1046 elf_auxargs->execfd = -1;
1047 elf_auxargs->phdr = proghdr;
1048 elf_auxargs->phent = hdr->e_phentsize;
1049 elf_auxargs->phnum = hdr->e_phnum;
1050 elf_auxargs->pagesz = PAGE_SIZE;
1051 elf_auxargs->base = addr;
1052 elf_auxargs->flags = 0;
1053 elf_auxargs->entry = entry;
1054 elf_auxargs->hdr_eflags = hdr->e_flags;
1056 imgp->auxargs = elf_auxargs;
1057 imgp->interpreted = 0;
1058 imgp->reloc_base = addr;
1059 imgp->proc->p_osrel = osrel;
1062 free(interp_buf, M_TEMP);
1066 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1069 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1071 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1075 base = (Elf_Addr *)*stack_base;
1076 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1078 if (args->execfd != -1)
1079 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1080 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1081 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1082 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1083 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1084 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1085 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1086 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1088 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1090 if (imgp->execpathp != 0)
1091 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1092 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1093 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1094 if (imgp->canary != 0) {
1095 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1096 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1098 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1099 if (imgp->pagesizes != 0) {
1100 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1101 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1103 if (imgp->sysent->sv_timekeep_base != 0) {
1104 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1105 imgp->sysent->sv_timekeep_base);
1107 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1108 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1109 imgp->sysent->sv_stackprot);
1110 AUXARGS_ENTRY(pos, AT_NULL, 0);
1112 free(imgp->auxargs, M_TEMP);
1113 imgp->auxargs = NULL;
1116 suword(base, (long)imgp->args->argc);
1117 *stack_base = (register_t *)base;
1122 * Code for generating ELF core dumps.
1125 typedef void (*segment_callback)(vm_map_entry_t, void *);
1127 /* Closure for cb_put_phdr(). */
1128 struct phdr_closure {
1129 Elf_Phdr *phdr; /* Program header to fill in */
1130 Elf_Off offset; /* Offset of segment in core file */
1133 /* Closure for cb_size_segment(). */
1134 struct sseg_closure {
1135 int count; /* Count of writable segments. */
1136 size_t size; /* Total size of all writable segments. */
1139 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1142 int type; /* Note type. */
1143 outfunc_t outfunc; /* Output function. */
1144 void *outarg; /* Argument for the output function. */
1145 size_t outsize; /* Output size. */
1146 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1149 TAILQ_HEAD(note_info_list, note_info);
1151 /* Coredump output parameters. */
1152 struct coredump_params {
1154 struct ucred *active_cred;
1155 struct ucred *file_cred;
1158 struct gzio_stream *gzs;
1161 static void cb_put_phdr(vm_map_entry_t, void *);
1162 static void cb_size_segment(vm_map_entry_t, void *);
1163 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1165 static void each_writable_segment(struct thread *, segment_callback, void *);
1166 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1167 struct note_info_list *, size_t);
1168 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1170 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1171 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1172 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1173 static int sbuf_drain_core_output(void *, const char *, int);
1174 static int sbuf_drain_count(void *arg, const char *data, int len);
1176 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1177 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1178 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1179 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1180 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1181 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1182 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1183 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1184 static void note_procstat_files(void *, struct sbuf *, size_t *);
1185 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1186 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1187 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1188 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1189 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1192 extern int compress_user_cores_gzlevel;
1195 * Write out a core segment to the compression stream.
1198 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1204 chunk_len = MIN(len, CORE_BUF_SIZE);
1207 * We can get EFAULT error here.
1208 * In that case zero out the current chunk of the segment.
1210 error = copyin(base, buf, chunk_len);
1212 bzero(buf, chunk_len);
1213 error = gzio_write(p->gzs, buf, chunk_len);
1223 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1226 return (core_write((struct coredump_params *)arg, base, len, offset,
1232 core_write(struct coredump_params *p, const void *base, size_t len,
1233 off_t offset, enum uio_seg seg)
1236 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1237 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1238 p->active_cred, p->file_cred, NULL, p->td));
1242 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1249 return (compress_chunk(p, base, tmpbuf, len));
1252 * EFAULT is a non-fatal error that we can get, for example,
1253 * if the segment is backed by a file but extends beyond its
1256 error = core_write(p, base, len, offset, UIO_USERSPACE);
1257 if (error == EFAULT) {
1258 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1259 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1260 "for process %s\n", base, len, offset, curproc->p_comm);
1263 * Write a "real" zero byte at the end of the target region
1264 * in the case this is the last segment.
1265 * The intermediate space will be implicitly zero-filled.
1267 error = core_write(p, zero_region, 1, offset + len - 1,
1274 * Drain into a core file.
1277 sbuf_drain_core_output(void *arg, const char *data, int len)
1279 struct coredump_params *p;
1282 p = (struct coredump_params *)arg;
1285 * Some kern_proc out routines that print to this sbuf may
1286 * call us with the process lock held. Draining with the
1287 * non-sleepable lock held is unsafe. The lock is needed for
1288 * those routines when dumping a live process. In our case we
1289 * can safely release the lock before draining and acquire
1292 locked = PROC_LOCKED(p->td->td_proc);
1294 PROC_UNLOCK(p->td->td_proc);
1297 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1300 error = core_write(p, __DECONST(void *, data), len, p->offset,
1303 PROC_LOCK(p->td->td_proc);
1311 * Drain into a counter.
1314 sbuf_drain_count(void *arg, const char *data __unused, int len)
1318 sizep = (size_t *)arg;
1324 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1326 struct ucred *cred = td->td_ucred;
1328 struct sseg_closure seginfo;
1329 struct note_info_list notelst;
1330 struct coredump_params params;
1331 struct note_info *ninfo;
1333 size_t hdrsize, notesz, coresize;
1337 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1341 TAILQ_INIT(¬elst);
1343 /* Size the program segments. */
1346 each_writable_segment(td, cb_size_segment, &seginfo);
1349 * Collect info about the core file header area.
1351 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1352 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1353 coresize = round_page(hdrsize + notesz) + seginfo.size;
1355 /* Set up core dump parameters. */
1357 params.active_cred = cred;
1358 params.file_cred = NOCRED;
1365 PROC_LOCK(td->td_proc);
1366 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1367 PROC_UNLOCK(td->td_proc);
1374 if (coresize >= limit) {
1380 /* Create a compression stream if necessary. */
1382 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1383 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1384 if (params.gzs == NULL) {
1388 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1393 * Allocate memory for building the header, fill it up,
1394 * and write it out following the notes.
1396 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1397 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1400 /* Write the contents of all of the writable segments. */
1406 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1407 offset = round_page(hdrsize + notesz);
1408 for (i = 0; i < seginfo.count; i++) {
1409 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1410 php->p_filesz, offset, ¶ms, tmpbuf);
1413 offset += php->p_filesz;
1417 if (error == 0 && compress)
1418 error = gzio_flush(params.gzs);
1423 "Failed to write core file for process %s (error %d)\n",
1424 curproc->p_comm, error);
1430 free(tmpbuf, M_TEMP);
1431 if (params.gzs != NULL)
1432 gzio_fini(params.gzs);
1435 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1436 TAILQ_REMOVE(¬elst, ninfo, link);
1437 free(ninfo, M_TEMP);
1446 * A callback for each_writable_segment() to write out the segment's
1447 * program header entry.
1450 cb_put_phdr(entry, closure)
1451 vm_map_entry_t entry;
1454 struct phdr_closure *phc = (struct phdr_closure *)closure;
1455 Elf_Phdr *phdr = phc->phdr;
1457 phc->offset = round_page(phc->offset);
1459 phdr->p_type = PT_LOAD;
1460 phdr->p_offset = phc->offset;
1461 phdr->p_vaddr = entry->start;
1463 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1464 phdr->p_align = PAGE_SIZE;
1465 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1467 phc->offset += phdr->p_filesz;
1472 * A callback for each_writable_segment() to gather information about
1473 * the number of segments and their total size.
1476 cb_size_segment(entry, closure)
1477 vm_map_entry_t entry;
1480 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1483 ssc->size += entry->end - entry->start;
1487 * For each writable segment in the process's memory map, call the given
1488 * function with a pointer to the map entry and some arbitrary
1489 * caller-supplied data.
1492 each_writable_segment(td, func, closure)
1494 segment_callback func;
1497 struct proc *p = td->td_proc;
1498 vm_map_t map = &p->p_vmspace->vm_map;
1499 vm_map_entry_t entry;
1500 vm_object_t backing_object, object;
1501 boolean_t ignore_entry;
1503 vm_map_lock_read(map);
1504 for (entry = map->header.next; entry != &map->header;
1505 entry = entry->next) {
1507 * Don't dump inaccessible mappings, deal with legacy
1510 * Note that read-only segments related to the elf binary
1511 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1512 * need to arbitrarily ignore such segments.
1514 if (elf_legacy_coredump) {
1515 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1518 if ((entry->protection & VM_PROT_ALL) == 0)
1523 * Dont include memory segment in the coredump if
1524 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1525 * madvise(2). Do not dump submaps (i.e. parts of the
1528 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1531 if ((object = entry->object.vm_object) == NULL)
1534 /* Ignore memory-mapped devices and such things. */
1535 VM_OBJECT_RLOCK(object);
1536 while ((backing_object = object->backing_object) != NULL) {
1537 VM_OBJECT_RLOCK(backing_object);
1538 VM_OBJECT_RUNLOCK(object);
1539 object = backing_object;
1541 ignore_entry = object->type != OBJT_DEFAULT &&
1542 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1543 object->type != OBJT_PHYS;
1544 VM_OBJECT_RUNLOCK(object);
1548 (*func)(entry, closure);
1550 vm_map_unlock_read(map);
1554 * Write the core file header to the file, including padding up to
1555 * the page boundary.
1558 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1559 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1561 struct note_info *ninfo;
1565 /* Fill in the header. */
1566 bzero(hdr, hdrsize);
1567 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1569 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1570 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1571 sbuf_start_section(sb, NULL);
1572 sbuf_bcat(sb, hdr, hdrsize);
1573 TAILQ_FOREACH(ninfo, notelst, link)
1574 __elfN(putnote)(ninfo, sb);
1575 /* Align up to a page boundary for the program segments. */
1576 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1577 error = sbuf_finish(sb);
1584 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1594 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1597 * To have the debugger select the right thread (LWP) as the initial
1598 * thread, we dump the state of the thread passed to us in td first.
1599 * This is the thread that causes the core dump and thus likely to
1600 * be the right thread one wants to have selected in the debugger.
1603 while (thr != NULL) {
1604 size += register_note(list, NT_PRSTATUS,
1605 __elfN(note_prstatus), thr);
1606 size += register_note(list, NT_FPREGSET,
1607 __elfN(note_fpregset), thr);
1608 size += register_note(list, NT_THRMISC,
1609 __elfN(note_thrmisc), thr);
1610 size += register_note(list, -1,
1611 __elfN(note_threadmd), thr);
1613 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1614 TAILQ_NEXT(thr, td_plist);
1616 thr = TAILQ_NEXT(thr, td_plist);
1619 size += register_note(list, NT_PROCSTAT_PROC,
1620 __elfN(note_procstat_proc), p);
1621 size += register_note(list, NT_PROCSTAT_FILES,
1622 note_procstat_files, p);
1623 size += register_note(list, NT_PROCSTAT_VMMAP,
1624 note_procstat_vmmap, p);
1625 size += register_note(list, NT_PROCSTAT_GROUPS,
1626 note_procstat_groups, p);
1627 size += register_note(list, NT_PROCSTAT_UMASK,
1628 note_procstat_umask, p);
1629 size += register_note(list, NT_PROCSTAT_RLIMIT,
1630 note_procstat_rlimit, p);
1631 size += register_note(list, NT_PROCSTAT_OSREL,
1632 note_procstat_osrel, p);
1633 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1634 __elfN(note_procstat_psstrings), p);
1635 size += register_note(list, NT_PROCSTAT_AUXV,
1636 __elfN(note_procstat_auxv), p);
1642 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1647 struct phdr_closure phc;
1649 ehdr = (Elf_Ehdr *)hdr;
1650 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1652 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1653 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1654 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1655 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1656 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1657 ehdr->e_ident[EI_DATA] = ELF_DATA;
1658 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1659 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1660 ehdr->e_ident[EI_ABIVERSION] = 0;
1661 ehdr->e_ident[EI_PAD] = 0;
1662 ehdr->e_type = ET_CORE;
1663 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1664 ehdr->e_machine = ELF_ARCH32;
1666 ehdr->e_machine = ELF_ARCH;
1668 ehdr->e_version = EV_CURRENT;
1670 ehdr->e_phoff = sizeof(Elf_Ehdr);
1672 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1673 ehdr->e_phentsize = sizeof(Elf_Phdr);
1674 ehdr->e_phnum = numsegs + 1;
1675 ehdr->e_shentsize = sizeof(Elf_Shdr);
1677 ehdr->e_shstrndx = SHN_UNDEF;
1680 * Fill in the program header entries.
1683 /* The note segement. */
1684 phdr->p_type = PT_NOTE;
1685 phdr->p_offset = hdrsize;
1688 phdr->p_filesz = notesz;
1690 phdr->p_flags = PF_R;
1691 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1694 /* All the writable segments from the program. */
1696 phc.offset = round_page(hdrsize + notesz);
1697 each_writable_segment(td, cb_put_phdr, &phc);
1701 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1703 struct note_info *ninfo;
1704 size_t size, notesize;
1707 out(arg, NULL, &size);
1708 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1710 ninfo->outfunc = out;
1711 ninfo->outarg = arg;
1712 ninfo->outsize = size;
1713 TAILQ_INSERT_TAIL(list, ninfo, link);
1718 notesize = sizeof(Elf_Note) + /* note header */
1719 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1721 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1727 append_note_data(const void *src, void *dst, size_t len)
1731 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1733 bcopy(src, dst, len);
1734 bzero((char *)dst + len, padded_len - len);
1736 return (padded_len);
1740 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1748 note = (Elf_Note *)buf;
1749 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1750 note->n_descsz = size;
1751 note->n_type = type;
1752 buf += sizeof(*note);
1753 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1754 sizeof(FREEBSD_ABI_VENDOR));
1755 append_note_data(src, buf, size);
1760 notesize = sizeof(Elf_Note) + /* note header */
1761 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1763 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1769 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1772 ssize_t old_len, sect_len;
1773 size_t new_len, descsz, i;
1775 if (ninfo->type == -1) {
1776 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1780 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1781 note.n_descsz = ninfo->outsize;
1782 note.n_type = ninfo->type;
1784 sbuf_bcat(sb, ¬e, sizeof(note));
1785 sbuf_start_section(sb, &old_len);
1786 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1787 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1788 if (note.n_descsz == 0)
1790 sbuf_start_section(sb, &old_len);
1791 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1792 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1796 new_len = (size_t)sect_len;
1797 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1798 if (new_len < descsz) {
1800 * It is expected that individual note emitters will correctly
1801 * predict their expected output size and fill up to that size
1802 * themselves, padding in a format-specific way if needed.
1803 * However, in case they don't, just do it here with zeros.
1805 for (i = 0; i < descsz - new_len; i++)
1807 } else if (new_len > descsz) {
1809 * We can't always truncate sb -- we may have drained some
1812 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1813 "read it (%zu > %zu). Since it is longer than "
1814 "expected, this coredump's notes are corrupt. THIS "
1815 "IS A BUG in the note_procstat routine for type %u.\n",
1816 __func__, (unsigned)note.n_type, new_len, descsz,
1817 (unsigned)note.n_type));
1822 * Miscellaneous note out functions.
1825 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1826 #include <compat/freebsd32/freebsd32.h>
1828 typedef struct prstatus32 elf_prstatus_t;
1829 typedef struct prpsinfo32 elf_prpsinfo_t;
1830 typedef struct fpreg32 elf_prfpregset_t;
1831 typedef struct fpreg32 elf_fpregset_t;
1832 typedef struct reg32 elf_gregset_t;
1833 typedef struct thrmisc32 elf_thrmisc_t;
1834 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1835 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1836 typedef uint32_t elf_ps_strings_t;
1838 typedef prstatus_t elf_prstatus_t;
1839 typedef prpsinfo_t elf_prpsinfo_t;
1840 typedef prfpregset_t elf_prfpregset_t;
1841 typedef prfpregset_t elf_fpregset_t;
1842 typedef gregset_t elf_gregset_t;
1843 typedef thrmisc_t elf_thrmisc_t;
1844 #define ELF_KERN_PROC_MASK 0
1845 typedef struct kinfo_proc elf_kinfo_proc_t;
1846 typedef vm_offset_t elf_ps_strings_t;
1850 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1856 elf_prpsinfo_t *psinfo;
1859 p = (struct proc *)arg;
1861 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1862 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1863 psinfo->pr_version = PRPSINFO_VERSION;
1864 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1865 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1867 if (p->p_args != NULL) {
1868 len = sizeof(psinfo->pr_psargs) - 1;
1869 if (len > p->p_args->ar_length)
1870 len = p->p_args->ar_length;
1871 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1877 sbuf_new(&sbarg, psinfo->pr_psargs,
1878 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1879 error = proc_getargv(curthread, p, &sbarg);
1881 if (sbuf_finish(&sbarg) == 0)
1882 len = sbuf_len(&sbarg) - 1;
1884 len = sizeof(psinfo->pr_psargs) - 1;
1885 sbuf_delete(&sbarg);
1887 if (error || len == 0)
1888 strlcpy(psinfo->pr_psargs, p->p_comm,
1889 sizeof(psinfo->pr_psargs));
1891 KASSERT(len < sizeof(psinfo->pr_psargs),
1892 ("len is too long: %zu vs %zu", len,
1893 sizeof(psinfo->pr_psargs)));
1894 cp = psinfo->pr_psargs;
1897 cp = memchr(cp, '\0', end - cp);
1903 psinfo->pr_pid = p->p_pid;
1904 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1905 free(psinfo, M_TEMP);
1907 *sizep = sizeof(*psinfo);
1911 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1914 elf_prstatus_t *status;
1916 td = (struct thread *)arg;
1918 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1919 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1920 status->pr_version = PRSTATUS_VERSION;
1921 status->pr_statussz = sizeof(elf_prstatus_t);
1922 status->pr_gregsetsz = sizeof(elf_gregset_t);
1923 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1924 status->pr_osreldate = osreldate;
1925 status->pr_cursig = td->td_proc->p_sig;
1926 status->pr_pid = td->td_tid;
1927 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1928 fill_regs32(td, &status->pr_reg);
1930 fill_regs(td, &status->pr_reg);
1932 sbuf_bcat(sb, status, sizeof(*status));
1933 free(status, M_TEMP);
1935 *sizep = sizeof(*status);
1939 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1942 elf_prfpregset_t *fpregset;
1944 td = (struct thread *)arg;
1946 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1947 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1948 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1949 fill_fpregs32(td, fpregset);
1951 fill_fpregs(td, fpregset);
1953 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1954 free(fpregset, M_TEMP);
1956 *sizep = sizeof(*fpregset);
1960 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1963 elf_thrmisc_t thrmisc;
1965 td = (struct thread *)arg;
1967 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1968 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1969 strcpy(thrmisc.pr_tname, td->td_name);
1970 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1972 *sizep = sizeof(thrmisc);
1976 * Allow for MD specific notes, as well as any MD
1977 * specific preparations for writing MI notes.
1980 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1986 td = (struct thread *)arg;
1988 if (size != 0 && sb != NULL)
1989 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1993 __elfN(dump_thread)(td, buf, &size);
1994 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1995 if (size != 0 && sb != NULL)
1996 sbuf_bcat(sb, buf, size);
2001 #ifdef KINFO_PROC_SIZE
2002 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2006 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2012 p = (struct proc *)arg;
2013 size = sizeof(structsize) + p->p_numthreads *
2014 sizeof(elf_kinfo_proc_t);
2017 KASSERT(*sizep == size, ("invalid size"));
2018 structsize = sizeof(elf_kinfo_proc_t);
2019 sbuf_bcat(sb, &structsize, sizeof(structsize));
2020 sx_slock(&proctree_lock);
2022 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2023 sx_sunlock(&proctree_lock);
2028 #ifdef KINFO_FILE_SIZE
2029 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2033 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2036 size_t size, sect_sz, i;
2037 ssize_t start_len, sect_len;
2038 int structsize, filedesc_flags;
2040 if (coredump_pack_fileinfo)
2041 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2045 p = (struct proc *)arg;
2046 structsize = sizeof(struct kinfo_file);
2049 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2050 sbuf_set_drain(sb, sbuf_drain_count, &size);
2051 sbuf_bcat(sb, &structsize, sizeof(structsize));
2053 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2058 sbuf_start_section(sb, &start_len);
2060 sbuf_bcat(sb, &structsize, sizeof(structsize));
2062 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2065 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2070 KASSERT(sect_sz <= *sizep,
2071 ("kern_proc_filedesc_out did not respect maxlen; "
2072 "requested %zu, got %zu", *sizep - sizeof(structsize),
2073 sect_sz - sizeof(structsize)));
2075 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2080 #ifdef KINFO_VMENTRY_SIZE
2081 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2085 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2089 int structsize, vmmap_flags;
2091 if (coredump_pack_vmmapinfo)
2092 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2096 p = (struct proc *)arg;
2097 structsize = sizeof(struct kinfo_vmentry);
2100 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2101 sbuf_set_drain(sb, sbuf_drain_count, &size);
2102 sbuf_bcat(sb, &structsize, sizeof(structsize));
2104 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2109 sbuf_bcat(sb, &structsize, sizeof(structsize));
2111 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2117 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2123 p = (struct proc *)arg;
2124 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2126 KASSERT(*sizep == size, ("invalid size"));
2127 structsize = sizeof(gid_t);
2128 sbuf_bcat(sb, &structsize, sizeof(structsize));
2129 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2136 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2142 p = (struct proc *)arg;
2143 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2145 KASSERT(*sizep == size, ("invalid size"));
2146 structsize = sizeof(p->p_fd->fd_cmask);
2147 sbuf_bcat(sb, &structsize, sizeof(structsize));
2148 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2154 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2157 struct rlimit rlim[RLIM_NLIMITS];
2161 p = (struct proc *)arg;
2162 size = sizeof(structsize) + sizeof(rlim);
2164 KASSERT(*sizep == size, ("invalid size"));
2165 structsize = sizeof(rlim);
2166 sbuf_bcat(sb, &structsize, sizeof(structsize));
2168 for (i = 0; i < RLIM_NLIMITS; i++)
2169 lim_rlimit_proc(p, i, &rlim[i]);
2171 sbuf_bcat(sb, rlim, sizeof(rlim));
2177 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2183 p = (struct proc *)arg;
2184 size = sizeof(structsize) + sizeof(p->p_osrel);
2186 KASSERT(*sizep == size, ("invalid size"));
2187 structsize = sizeof(p->p_osrel);
2188 sbuf_bcat(sb, &structsize, sizeof(structsize));
2189 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2195 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2198 elf_ps_strings_t ps_strings;
2202 p = (struct proc *)arg;
2203 size = sizeof(structsize) + sizeof(ps_strings);
2205 KASSERT(*sizep == size, ("invalid size"));
2206 structsize = sizeof(ps_strings);
2207 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2208 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2210 ps_strings = p->p_sysent->sv_psstrings;
2212 sbuf_bcat(sb, &structsize, sizeof(structsize));
2213 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2219 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2225 p = (struct proc *)arg;
2228 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2229 sbuf_set_drain(sb, sbuf_drain_count, &size);
2230 sbuf_bcat(sb, &structsize, sizeof(structsize));
2232 proc_getauxv(curthread, p, sb);
2238 structsize = sizeof(Elf_Auxinfo);
2239 sbuf_bcat(sb, &structsize, sizeof(structsize));
2241 proc_getauxv(curthread, p, sb);
2247 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2248 int32_t *osrel, const Elf_Phdr *pnote)
2250 const Elf_Note *note, *note0, *note_end;
2251 const char *note_name;
2256 /* We need some limit, might as well use PAGE_SIZE. */
2257 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2259 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2260 if (pnote->p_offset > PAGE_SIZE ||
2261 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2262 VOP_UNLOCK(imgp->vp, 0);
2263 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2264 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2265 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2266 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2267 curthread->td_ucred, NOCRED, NULL, curthread);
2269 uprintf("i/o error PT_NOTE\n");
2273 note = note0 = (const Elf_Note *)buf;
2274 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2276 note = note0 = (const Elf_Note *)(imgp->image_header +
2278 note_end = (const Elf_Note *)(imgp->image_header +
2279 pnote->p_offset + pnote->p_filesz);
2282 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2283 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2284 (const char *)note < sizeof(Elf_Note)) {
2288 if (note->n_namesz != checknote->hdr.n_namesz ||
2289 note->n_descsz != checknote->hdr.n_descsz ||
2290 note->n_type != checknote->hdr.n_type)
2292 note_name = (const char *)(note + 1);
2293 if (note_name + checknote->hdr.n_namesz >=
2294 (const char *)note_end || strncmp(checknote->vendor,
2295 note_name, checknote->hdr.n_namesz) != 0)
2299 * Fetch the osreldate for binary
2300 * from the ELF OSABI-note if necessary.
2302 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2303 checknote->trans_osrel != NULL) {
2304 res = checknote->trans_osrel(note, osrel);
2310 note = (const Elf_Note *)((const char *)(note + 1) +
2311 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2312 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2321 * Try to find the appropriate ABI-note section for checknote,
2322 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2323 * first page of the image is searched, the same as for headers.
2326 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2329 const Elf_Phdr *phdr;
2330 const Elf_Ehdr *hdr;
2333 hdr = (const Elf_Ehdr *)imgp->image_header;
2334 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2336 for (i = 0; i < hdr->e_phnum; i++) {
2337 if (phdr[i].p_type == PT_NOTE &&
2338 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2346 * Tell kern_execve.c about it, with a little help from the linker.
2348 static struct execsw __elfN(execsw) = {
2349 __CONCAT(exec_, __elfN(imgact)),
2350 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2352 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2355 __elfN(trans_prot)(Elf_Word flags)
2361 prot |= VM_PROT_EXECUTE;
2363 prot |= VM_PROT_WRITE;
2365 prot |= VM_PROT_READ;
2366 #if __ELF_WORD_SIZE == 32
2367 #if defined(__amd64__)
2368 if (i386_read_exec && (flags & PF_R))
2369 prot |= VM_PROT_EXECUTE;
2376 __elfN(untrans_prot)(vm_prot_t prot)
2381 if (prot & VM_PROT_EXECUTE)
2383 if (prot & VM_PROT_READ)
2385 if (prot & VM_PROT_WRITE)