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);
420 return (KERN_SUCCESS);
424 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
425 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
431 int error, locked, rv;
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 locked = VOP_ISLOCKED(imgp->vp);
485 VOP_UNLOCK(imgp->vp, 0);
486 vm_object_deallocate(object);
487 vn_lock(imgp->vp, locked | LK_RETRY);
492 return (KERN_SUCCESS);
497 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
498 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
505 vm_offset_t map_addr;
508 vm_offset_t file_addr;
511 * It's necessary to fail if the filsz + offset taken from the
512 * header is greater than the actual file pager object's size.
513 * If we were to allow this, then the vm_map_find() below would
514 * walk right off the end of the file object and into the ether.
516 * While I'm here, might as well check for something else that
517 * is invalid: filsz cannot be greater than memsz.
519 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
520 uprintf("elf_load_section: truncated ELF file\n");
524 object = imgp->object;
525 map = &imgp->proc->p_vmspace->vm_map;
526 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
527 file_addr = trunc_page_ps(offset, pagesize);
530 * We have two choices. We can either clear the data in the last page
531 * of an oversized mapping, or we can start the anon mapping a page
532 * early and copy the initialized data into that first page. We
533 * choose the second..
536 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
538 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
541 /* cow flags: don't dump readonly sections in core */
542 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
543 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
545 rv = __elfN(map_insert)(imgp, map,
547 file_addr, /* file offset */
548 map_addr, /* virtual start */
549 map_addr + map_len,/* virtual end */
552 if (rv != KERN_SUCCESS)
555 /* we can stop now if we've covered it all */
556 if (memsz == filsz) {
563 * We have to get the remaining bit of the file into the first part
564 * of the oversized map segment. This is normally because the .data
565 * segment in the file is extended to provide bss. It's a neat idea
566 * to try and save a page, but it's a pain in the behind to implement.
568 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
569 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
570 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
573 /* This had damn well better be true! */
575 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
576 map_addr + map_len, VM_PROT_ALL, 0);
577 if (rv != KERN_SUCCESS) {
585 sf = vm_imgact_map_page(object, offset + filsz);
589 /* send the page fragment to user space */
590 off = trunc_page_ps(offset + filsz, pagesize) -
591 trunc_page(offset + filsz);
592 error = copyout((caddr_t)sf_buf_kva(sf) + off,
593 (caddr_t)map_addr, copy_len);
594 vm_imgact_unmap_page(sf);
601 * set it to the specified protection.
602 * XXX had better undo the damage from pasting over the cracks here!
604 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
605 map_len), prot, FALSE);
611 * Load the file "file" into memory. It may be either a shared object
614 * The "addr" reference parameter is in/out. On entry, it specifies
615 * the address where a shared object should be loaded. If the file is
616 * an executable, this value is ignored. On exit, "addr" specifies
617 * where the file was actually loaded.
619 * The "entry" reference parameter is out only. On exit, it specifies
620 * the entry point for the loaded file.
623 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
624 u_long *entry, size_t pagesize)
629 struct image_params image_params;
631 const Elf_Ehdr *hdr = NULL;
632 const Elf_Phdr *phdr = NULL;
633 struct nameidata *nd;
635 struct image_params *imgp;
638 u_long base_addr = 0;
639 int error, i, numsegs;
641 #ifdef CAPABILITY_MODE
643 * XXXJA: This check can go away once we are sufficiently confident
644 * that the checks in namei() are correct.
646 if (IN_CAPABILITY_MODE(curthread))
650 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
652 attr = &tempdata->attr;
653 imgp = &tempdata->image_params;
656 * Initialize part of the common data
660 imgp->firstpage = NULL;
661 imgp->image_header = NULL;
663 imgp->execlabel = NULL;
665 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
666 if ((error = namei(nd)) != 0) {
670 NDFREE(nd, NDF_ONLY_PNBUF);
671 imgp->vp = nd->ni_vp;
674 * Check permissions, modes, uid, etc on the file, and "open" it.
676 error = exec_check_permissions(imgp);
680 error = exec_map_first_page(imgp);
685 * Also make certain that the interpreter stays the same, so set
686 * its VV_TEXT flag, too.
688 VOP_SET_TEXT(nd->ni_vp);
690 imgp->object = nd->ni_vp->v_object;
692 hdr = (const Elf_Ehdr *)imgp->image_header;
693 if ((error = __elfN(check_header)(hdr)) != 0)
695 if (hdr->e_type == ET_DYN)
697 else if (hdr->e_type == ET_EXEC)
704 /* Only support headers that fit within first page for now */
705 if ((hdr->e_phoff > PAGE_SIZE) ||
706 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
711 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
712 if (!aligned(phdr, Elf_Addr)) {
717 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
718 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
719 /* Loadable segment */
720 prot = __elfN(trans_prot)(phdr[i].p_flags);
721 error = __elfN(load_section)(imgp, phdr[i].p_offset,
722 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
723 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
727 * Establish the base address if this is the
731 base_addr = trunc_page(phdr[i].p_vaddr +
737 *entry = (unsigned long)hdr->e_entry + rbase;
741 exec_unmap_first_page(imgp);
746 free(tempdata, M_TEMP);
752 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
756 const Elf_Phdr *phdr;
757 Elf_Auxargs *elf_auxargs;
758 struct vmspace *vmspace;
759 const char *err_str, *newinterp;
760 char *interp, *interp_buf, *path;
761 Elf_Brandinfo *brand_info;
762 struct sysentvec *sv;
764 u_long text_size, data_size, total_size, text_addr, data_addr;
765 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
767 int error, i, n, interp_name_len, have_interp;
769 hdr = (const Elf_Ehdr *)imgp->image_header;
772 * Do we have a valid ELF header ?
774 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
775 * if particular brand doesn't support it.
777 if (__elfN(check_header)(hdr) != 0 ||
778 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
782 * From here on down, we return an errno, not -1, as we've
783 * detected an ELF file.
786 if ((hdr->e_phoff > PAGE_SIZE) ||
787 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
788 /* Only support headers in first page for now */
789 uprintf("Program headers not in the first page\n");
792 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
793 if (!aligned(phdr, Elf_Addr)) {
794 uprintf("Unaligned program headers\n");
801 text_size = data_size = total_size = text_addr = data_addr = 0;
804 err_str = newinterp = NULL;
805 interp = interp_buf = NULL;
808 for (i = 0; i < hdr->e_phnum; i++) {
809 switch (phdr[i].p_type) {
812 baddr = phdr[i].p_vaddr;
816 /* Path to interpreter */
817 if (phdr[i].p_filesz > MAXPATHLEN) {
818 uprintf("Invalid PT_INTERP\n");
822 if (interp != NULL) {
823 uprintf("Multiple PT_INTERP headers\n");
827 interp_name_len = phdr[i].p_filesz;
828 if (phdr[i].p_offset > PAGE_SIZE ||
829 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
830 VOP_UNLOCK(imgp->vp, 0);
831 interp_buf = malloc(interp_name_len + 1, M_TEMP,
833 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
834 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
835 interp_name_len, phdr[i].p_offset,
836 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
839 uprintf("i/o error PT_INTERP\n");
842 interp_buf[interp_name_len] = '\0';
845 interp = __DECONST(char *, imgp->image_header) +
852 __elfN(trans_prot)(phdr[i].p_flags);
853 imgp->stack_sz = phdr[i].p_memsz;
858 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
860 if (brand_info == NULL) {
861 uprintf("ELF binary type \"%u\" not known.\n",
862 hdr->e_ident[EI_OSABI]);
866 if (hdr->e_type == ET_DYN) {
867 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
868 uprintf("Cannot execute shared object\n");
873 * Honour the base load address from the dso if it is
874 * non-zero for some reason.
877 et_dyn_addr = ET_DYN_LOAD_ADDR;
882 sv = brand_info->sysvec;
883 if (interp != NULL && brand_info->interp_newpath != NULL)
884 newinterp = brand_info->interp_newpath;
887 * Avoid a possible deadlock if the current address space is destroyed
888 * and that address space maps the locked vnode. In the common case,
889 * the locked vnode's v_usecount is decremented but remains greater
890 * than zero. Consequently, the vnode lock is not needed by vrele().
891 * However, in cases where the vnode lock is external, such as nullfs,
892 * v_usecount may become zero.
894 * The VV_TEXT flag prevents modifications to the executable while
895 * the vnode is unlocked.
897 VOP_UNLOCK(imgp->vp, 0);
899 error = exec_new_vmspace(imgp, sv);
900 imgp->proc->p_sysent = sv;
902 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
906 for (i = 0; i < hdr->e_phnum; i++) {
907 switch (phdr[i].p_type) {
908 case PT_LOAD: /* Loadable segment */
909 if (phdr[i].p_memsz == 0)
911 prot = __elfN(trans_prot)(phdr[i].p_flags);
912 error = __elfN(load_section)(imgp, phdr[i].p_offset,
913 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
914 phdr[i].p_memsz, phdr[i].p_filesz, prot,
920 * If this segment contains the program headers,
921 * remember their virtual address for the AT_PHDR
922 * aux entry. Static binaries don't usually include
925 if (phdr[i].p_offset == 0 &&
926 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
928 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
931 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
932 seg_size = round_page(phdr[i].p_memsz +
933 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
936 * Make the largest executable segment the official
937 * text segment and all others data.
939 * Note that obreak() assumes that data_addr +
940 * data_size == end of data load area, and the ELF
941 * file format expects segments to be sorted by
942 * address. If multiple data segments exist, the
943 * last one will be used.
946 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
947 text_size = seg_size;
948 text_addr = seg_addr;
950 data_size = seg_size;
951 data_addr = seg_addr;
953 total_size += seg_size;
955 case PT_PHDR: /* Program header table info */
956 proghdr = phdr[i].p_vaddr + et_dyn_addr;
963 if (data_addr == 0 && data_size == 0) {
964 data_addr = text_addr;
965 data_size = text_size;
968 entry = (u_long)hdr->e_entry + et_dyn_addr;
971 * Check limits. It should be safe to check the
972 * limits after loading the segments since we do
973 * not actually fault in all the segments pages.
975 PROC_LOCK(imgp->proc);
976 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
977 err_str = "Data segment size exceeds process limit";
978 else if (text_size > maxtsiz)
979 err_str = "Text segment size exceeds system limit";
980 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
981 err_str = "Total segment size exceeds process limit";
982 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
983 err_str = "Data segment size exceeds resource limit";
984 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
985 err_str = "Total segment size exceeds resource limit";
986 if (err_str != NULL) {
987 PROC_UNLOCK(imgp->proc);
988 uprintf("%s\n", err_str);
993 vmspace = imgp->proc->p_vmspace;
994 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
995 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
996 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
997 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1000 * We load the dynamic linker where a userland call
1001 * to mmap(0, ...) would put it. The rationale behind this
1002 * calculation is that it leaves room for the heap to grow to
1003 * its maximum allowed size.
1005 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1007 PROC_UNLOCK(imgp->proc);
1009 imgp->entry_addr = entry;
1011 if (interp != NULL) {
1012 have_interp = FALSE;
1013 VOP_UNLOCK(imgp->vp, 0);
1014 if (brand_info->emul_path != NULL &&
1015 brand_info->emul_path[0] != '\0') {
1016 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1017 snprintf(path, MAXPATHLEN, "%s%s",
1018 brand_info->emul_path, interp);
1019 error = __elfN(load_file)(imgp->proc, path, &addr,
1020 &imgp->entry_addr, sv->sv_pagesize);
1025 if (!have_interp && newinterp != NULL &&
1026 (brand_info->interp_path == NULL ||
1027 strcmp(interp, brand_info->interp_path) == 0)) {
1028 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1029 &imgp->entry_addr, sv->sv_pagesize);
1034 error = __elfN(load_file)(imgp->proc, interp, &addr,
1035 &imgp->entry_addr, sv->sv_pagesize);
1037 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1039 uprintf("ELF interpreter %s not found, error %d\n",
1047 * Construct auxargs table (used by the fixup routine)
1049 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1050 elf_auxargs->execfd = -1;
1051 elf_auxargs->phdr = proghdr;
1052 elf_auxargs->phent = hdr->e_phentsize;
1053 elf_auxargs->phnum = hdr->e_phnum;
1054 elf_auxargs->pagesz = PAGE_SIZE;
1055 elf_auxargs->base = addr;
1056 elf_auxargs->flags = 0;
1057 elf_auxargs->entry = entry;
1058 elf_auxargs->hdr_eflags = hdr->e_flags;
1060 imgp->auxargs = elf_auxargs;
1061 imgp->interpreted = 0;
1062 imgp->reloc_base = addr;
1063 imgp->proc->p_osrel = osrel;
1066 free(interp_buf, M_TEMP);
1070 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1073 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1075 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1079 base = (Elf_Addr *)*stack_base;
1080 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1082 if (args->execfd != -1)
1083 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1084 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1085 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1086 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1087 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1088 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1089 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1090 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1092 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1094 if (imgp->execpathp != 0)
1095 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1096 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1097 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1098 if (imgp->canary != 0) {
1099 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1100 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1102 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1103 if (imgp->pagesizes != 0) {
1104 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1105 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1107 if (imgp->sysent->sv_timekeep_base != 0) {
1108 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1109 imgp->sysent->sv_timekeep_base);
1111 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1112 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1113 imgp->sysent->sv_stackprot);
1114 AUXARGS_ENTRY(pos, AT_NULL, 0);
1116 free(imgp->auxargs, M_TEMP);
1117 imgp->auxargs = NULL;
1120 suword(base, (long)imgp->args->argc);
1121 *stack_base = (register_t *)base;
1126 * Code for generating ELF core dumps.
1129 typedef void (*segment_callback)(vm_map_entry_t, void *);
1131 /* Closure for cb_put_phdr(). */
1132 struct phdr_closure {
1133 Elf_Phdr *phdr; /* Program header to fill in */
1134 Elf_Off offset; /* Offset of segment in core file */
1137 /* Closure for cb_size_segment(). */
1138 struct sseg_closure {
1139 int count; /* Count of writable segments. */
1140 size_t size; /* Total size of all writable segments. */
1143 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1146 int type; /* Note type. */
1147 outfunc_t outfunc; /* Output function. */
1148 void *outarg; /* Argument for the output function. */
1149 size_t outsize; /* Output size. */
1150 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1153 TAILQ_HEAD(note_info_list, note_info);
1155 /* Coredump output parameters. */
1156 struct coredump_params {
1158 struct ucred *active_cred;
1159 struct ucred *file_cred;
1162 struct gzio_stream *gzs;
1165 static void cb_put_phdr(vm_map_entry_t, void *);
1166 static void cb_size_segment(vm_map_entry_t, void *);
1167 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1169 static void each_writable_segment(struct thread *, segment_callback, void *);
1170 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1171 struct note_info_list *, size_t);
1172 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1174 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1175 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1176 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1177 static int sbuf_drain_core_output(void *, const char *, int);
1178 static int sbuf_drain_count(void *arg, const char *data, int len);
1180 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1181 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1182 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1183 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1184 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1185 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1186 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1187 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1188 static void note_procstat_files(void *, struct sbuf *, size_t *);
1189 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1190 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1191 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1192 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1193 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1196 extern int compress_user_cores_gzlevel;
1199 * Write out a core segment to the compression stream.
1202 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1208 chunk_len = MIN(len, CORE_BUF_SIZE);
1211 * We can get EFAULT error here.
1212 * In that case zero out the current chunk of the segment.
1214 error = copyin(base, buf, chunk_len);
1216 bzero(buf, chunk_len);
1217 error = gzio_write(p->gzs, buf, chunk_len);
1227 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1230 return (core_write((struct coredump_params *)arg, base, len, offset,
1236 core_write(struct coredump_params *p, const void *base, size_t len,
1237 off_t offset, enum uio_seg seg)
1240 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1241 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1242 p->active_cred, p->file_cred, NULL, p->td));
1246 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1253 return (compress_chunk(p, base, tmpbuf, len));
1256 * EFAULT is a non-fatal error that we can get, for example,
1257 * if the segment is backed by a file but extends beyond its
1260 error = core_write(p, base, len, offset, UIO_USERSPACE);
1261 if (error == EFAULT) {
1262 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1263 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1264 "for process %s\n", base, len, offset, curproc->p_comm);
1267 * Write a "real" zero byte at the end of the target region
1268 * in the case this is the last segment.
1269 * The intermediate space will be implicitly zero-filled.
1271 error = core_write(p, zero_region, 1, offset + len - 1,
1278 * Drain into a core file.
1281 sbuf_drain_core_output(void *arg, const char *data, int len)
1283 struct coredump_params *p;
1286 p = (struct coredump_params *)arg;
1289 * Some kern_proc out routines that print to this sbuf may
1290 * call us with the process lock held. Draining with the
1291 * non-sleepable lock held is unsafe. The lock is needed for
1292 * those routines when dumping a live process. In our case we
1293 * can safely release the lock before draining and acquire
1296 locked = PROC_LOCKED(p->td->td_proc);
1298 PROC_UNLOCK(p->td->td_proc);
1301 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1304 error = core_write(p, __DECONST(void *, data), len, p->offset,
1307 PROC_LOCK(p->td->td_proc);
1315 * Drain into a counter.
1318 sbuf_drain_count(void *arg, const char *data __unused, int len)
1322 sizep = (size_t *)arg;
1328 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1330 struct ucred *cred = td->td_ucred;
1332 struct sseg_closure seginfo;
1333 struct note_info_list notelst;
1334 struct coredump_params params;
1335 struct note_info *ninfo;
1337 size_t hdrsize, notesz, coresize;
1341 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1345 TAILQ_INIT(¬elst);
1347 /* Size the program segments. */
1350 each_writable_segment(td, cb_size_segment, &seginfo);
1353 * Collect info about the core file header area.
1355 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1356 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1357 coresize = round_page(hdrsize + notesz) + seginfo.size;
1359 /* Set up core dump parameters. */
1361 params.active_cred = cred;
1362 params.file_cred = NOCRED;
1369 PROC_LOCK(td->td_proc);
1370 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1371 PROC_UNLOCK(td->td_proc);
1378 if (coresize >= limit) {
1384 /* Create a compression stream if necessary. */
1386 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1387 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1388 if (params.gzs == NULL) {
1392 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1397 * Allocate memory for building the header, fill it up,
1398 * and write it out following the notes.
1400 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1401 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1404 /* Write the contents of all of the writable segments. */
1410 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1411 offset = round_page(hdrsize + notesz);
1412 for (i = 0; i < seginfo.count; i++) {
1413 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1414 php->p_filesz, offset, ¶ms, tmpbuf);
1417 offset += php->p_filesz;
1421 if (error == 0 && compress)
1422 error = gzio_flush(params.gzs);
1427 "Failed to write core file for process %s (error %d)\n",
1428 curproc->p_comm, error);
1434 free(tmpbuf, M_TEMP);
1435 if (params.gzs != NULL)
1436 gzio_fini(params.gzs);
1439 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1440 TAILQ_REMOVE(¬elst, ninfo, link);
1441 free(ninfo, M_TEMP);
1450 * A callback for each_writable_segment() to write out the segment's
1451 * program header entry.
1454 cb_put_phdr(entry, closure)
1455 vm_map_entry_t entry;
1458 struct phdr_closure *phc = (struct phdr_closure *)closure;
1459 Elf_Phdr *phdr = phc->phdr;
1461 phc->offset = round_page(phc->offset);
1463 phdr->p_type = PT_LOAD;
1464 phdr->p_offset = phc->offset;
1465 phdr->p_vaddr = entry->start;
1467 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1468 phdr->p_align = PAGE_SIZE;
1469 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1471 phc->offset += phdr->p_filesz;
1476 * A callback for each_writable_segment() to gather information about
1477 * the number of segments and their total size.
1480 cb_size_segment(entry, closure)
1481 vm_map_entry_t entry;
1484 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1487 ssc->size += entry->end - entry->start;
1491 * For each writable segment in the process's memory map, call the given
1492 * function with a pointer to the map entry and some arbitrary
1493 * caller-supplied data.
1496 each_writable_segment(td, func, closure)
1498 segment_callback func;
1501 struct proc *p = td->td_proc;
1502 vm_map_t map = &p->p_vmspace->vm_map;
1503 vm_map_entry_t entry;
1504 vm_object_t backing_object, object;
1505 boolean_t ignore_entry;
1507 vm_map_lock_read(map);
1508 for (entry = map->header.next; entry != &map->header;
1509 entry = entry->next) {
1511 * Don't dump inaccessible mappings, deal with legacy
1514 * Note that read-only segments related to the elf binary
1515 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1516 * need to arbitrarily ignore such segments.
1518 if (elf_legacy_coredump) {
1519 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1522 if ((entry->protection & VM_PROT_ALL) == 0)
1527 * Dont include memory segment in the coredump if
1528 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1529 * madvise(2). Do not dump submaps (i.e. parts of the
1532 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1535 if ((object = entry->object.vm_object) == NULL)
1538 /* Ignore memory-mapped devices and such things. */
1539 VM_OBJECT_RLOCK(object);
1540 while ((backing_object = object->backing_object) != NULL) {
1541 VM_OBJECT_RLOCK(backing_object);
1542 VM_OBJECT_RUNLOCK(object);
1543 object = backing_object;
1545 ignore_entry = object->type != OBJT_DEFAULT &&
1546 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1547 object->type != OBJT_PHYS;
1548 VM_OBJECT_RUNLOCK(object);
1552 (*func)(entry, closure);
1554 vm_map_unlock_read(map);
1558 * Write the core file header to the file, including padding up to
1559 * the page boundary.
1562 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1563 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1565 struct note_info *ninfo;
1569 /* Fill in the header. */
1570 bzero(hdr, hdrsize);
1571 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1573 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1574 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1575 sbuf_start_section(sb, NULL);
1576 sbuf_bcat(sb, hdr, hdrsize);
1577 TAILQ_FOREACH(ninfo, notelst, link)
1578 __elfN(putnote)(ninfo, sb);
1579 /* Align up to a page boundary for the program segments. */
1580 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1581 error = sbuf_finish(sb);
1588 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1598 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1601 * To have the debugger select the right thread (LWP) as the initial
1602 * thread, we dump the state of the thread passed to us in td first.
1603 * This is the thread that causes the core dump and thus likely to
1604 * be the right thread one wants to have selected in the debugger.
1607 while (thr != NULL) {
1608 size += register_note(list, NT_PRSTATUS,
1609 __elfN(note_prstatus), thr);
1610 size += register_note(list, NT_FPREGSET,
1611 __elfN(note_fpregset), thr);
1612 size += register_note(list, NT_THRMISC,
1613 __elfN(note_thrmisc), thr);
1614 size += register_note(list, -1,
1615 __elfN(note_threadmd), thr);
1617 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1618 TAILQ_NEXT(thr, td_plist);
1620 thr = TAILQ_NEXT(thr, td_plist);
1623 size += register_note(list, NT_PROCSTAT_PROC,
1624 __elfN(note_procstat_proc), p);
1625 size += register_note(list, NT_PROCSTAT_FILES,
1626 note_procstat_files, p);
1627 size += register_note(list, NT_PROCSTAT_VMMAP,
1628 note_procstat_vmmap, p);
1629 size += register_note(list, NT_PROCSTAT_GROUPS,
1630 note_procstat_groups, p);
1631 size += register_note(list, NT_PROCSTAT_UMASK,
1632 note_procstat_umask, p);
1633 size += register_note(list, NT_PROCSTAT_RLIMIT,
1634 note_procstat_rlimit, p);
1635 size += register_note(list, NT_PROCSTAT_OSREL,
1636 note_procstat_osrel, p);
1637 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1638 __elfN(note_procstat_psstrings), p);
1639 size += register_note(list, NT_PROCSTAT_AUXV,
1640 __elfN(note_procstat_auxv), p);
1646 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1651 struct phdr_closure phc;
1653 ehdr = (Elf_Ehdr *)hdr;
1654 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1656 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1657 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1658 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1659 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1660 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1661 ehdr->e_ident[EI_DATA] = ELF_DATA;
1662 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1663 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1664 ehdr->e_ident[EI_ABIVERSION] = 0;
1665 ehdr->e_ident[EI_PAD] = 0;
1666 ehdr->e_type = ET_CORE;
1667 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1668 ehdr->e_machine = ELF_ARCH32;
1670 ehdr->e_machine = ELF_ARCH;
1672 ehdr->e_version = EV_CURRENT;
1674 ehdr->e_phoff = sizeof(Elf_Ehdr);
1676 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1677 ehdr->e_phentsize = sizeof(Elf_Phdr);
1678 ehdr->e_phnum = numsegs + 1;
1679 ehdr->e_shentsize = sizeof(Elf_Shdr);
1681 ehdr->e_shstrndx = SHN_UNDEF;
1684 * Fill in the program header entries.
1687 /* The note segement. */
1688 phdr->p_type = PT_NOTE;
1689 phdr->p_offset = hdrsize;
1692 phdr->p_filesz = notesz;
1694 phdr->p_flags = PF_R;
1695 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1698 /* All the writable segments from the program. */
1700 phc.offset = round_page(hdrsize + notesz);
1701 each_writable_segment(td, cb_put_phdr, &phc);
1705 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1707 struct note_info *ninfo;
1708 size_t size, notesize;
1711 out(arg, NULL, &size);
1712 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1714 ninfo->outfunc = out;
1715 ninfo->outarg = arg;
1716 ninfo->outsize = size;
1717 TAILQ_INSERT_TAIL(list, ninfo, link);
1722 notesize = sizeof(Elf_Note) + /* note header */
1723 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1725 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1731 append_note_data(const void *src, void *dst, size_t len)
1735 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1737 bcopy(src, dst, len);
1738 bzero((char *)dst + len, padded_len - len);
1740 return (padded_len);
1744 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1752 note = (Elf_Note *)buf;
1753 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1754 note->n_descsz = size;
1755 note->n_type = type;
1756 buf += sizeof(*note);
1757 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1758 sizeof(FREEBSD_ABI_VENDOR));
1759 append_note_data(src, buf, size);
1764 notesize = sizeof(Elf_Note) + /* note header */
1765 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1767 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1773 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1776 ssize_t old_len, sect_len;
1777 size_t new_len, descsz, i;
1779 if (ninfo->type == -1) {
1780 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1784 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1785 note.n_descsz = ninfo->outsize;
1786 note.n_type = ninfo->type;
1788 sbuf_bcat(sb, ¬e, sizeof(note));
1789 sbuf_start_section(sb, &old_len);
1790 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1791 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1792 if (note.n_descsz == 0)
1794 sbuf_start_section(sb, &old_len);
1795 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1796 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1800 new_len = (size_t)sect_len;
1801 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1802 if (new_len < descsz) {
1804 * It is expected that individual note emitters will correctly
1805 * predict their expected output size and fill up to that size
1806 * themselves, padding in a format-specific way if needed.
1807 * However, in case they don't, just do it here with zeros.
1809 for (i = 0; i < descsz - new_len; i++)
1811 } else if (new_len > descsz) {
1813 * We can't always truncate sb -- we may have drained some
1816 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1817 "read it (%zu > %zu). Since it is longer than "
1818 "expected, this coredump's notes are corrupt. THIS "
1819 "IS A BUG in the note_procstat routine for type %u.\n",
1820 __func__, (unsigned)note.n_type, new_len, descsz,
1821 (unsigned)note.n_type));
1826 * Miscellaneous note out functions.
1829 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1830 #include <compat/freebsd32/freebsd32.h>
1832 typedef struct prstatus32 elf_prstatus_t;
1833 typedef struct prpsinfo32 elf_prpsinfo_t;
1834 typedef struct fpreg32 elf_prfpregset_t;
1835 typedef struct fpreg32 elf_fpregset_t;
1836 typedef struct reg32 elf_gregset_t;
1837 typedef struct thrmisc32 elf_thrmisc_t;
1838 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1839 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1840 typedef uint32_t elf_ps_strings_t;
1842 typedef prstatus_t elf_prstatus_t;
1843 typedef prpsinfo_t elf_prpsinfo_t;
1844 typedef prfpregset_t elf_prfpregset_t;
1845 typedef prfpregset_t elf_fpregset_t;
1846 typedef gregset_t elf_gregset_t;
1847 typedef thrmisc_t elf_thrmisc_t;
1848 #define ELF_KERN_PROC_MASK 0
1849 typedef struct kinfo_proc elf_kinfo_proc_t;
1850 typedef vm_offset_t elf_ps_strings_t;
1854 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1860 elf_prpsinfo_t *psinfo;
1863 p = (struct proc *)arg;
1865 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1866 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1867 psinfo->pr_version = PRPSINFO_VERSION;
1868 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1869 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1871 if (p->p_args != NULL) {
1872 len = sizeof(psinfo->pr_psargs) - 1;
1873 if (len > p->p_args->ar_length)
1874 len = p->p_args->ar_length;
1875 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1881 sbuf_new(&sbarg, psinfo->pr_psargs,
1882 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1883 error = proc_getargv(curthread, p, &sbarg);
1885 if (sbuf_finish(&sbarg) == 0)
1886 len = sbuf_len(&sbarg) - 1;
1888 len = sizeof(psinfo->pr_psargs) - 1;
1889 sbuf_delete(&sbarg);
1891 if (error || len == 0)
1892 strlcpy(psinfo->pr_psargs, p->p_comm,
1893 sizeof(psinfo->pr_psargs));
1895 KASSERT(len < sizeof(psinfo->pr_psargs),
1896 ("len is too long: %zu vs %zu", len,
1897 sizeof(psinfo->pr_psargs)));
1898 cp = psinfo->pr_psargs;
1901 cp = memchr(cp, '\0', end - cp);
1907 psinfo->pr_pid = p->p_pid;
1908 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1909 free(psinfo, M_TEMP);
1911 *sizep = sizeof(*psinfo);
1915 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1918 elf_prstatus_t *status;
1920 td = (struct thread *)arg;
1922 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1923 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1924 status->pr_version = PRSTATUS_VERSION;
1925 status->pr_statussz = sizeof(elf_prstatus_t);
1926 status->pr_gregsetsz = sizeof(elf_gregset_t);
1927 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1928 status->pr_osreldate = osreldate;
1929 status->pr_cursig = td->td_proc->p_sig;
1930 status->pr_pid = td->td_tid;
1931 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1932 fill_regs32(td, &status->pr_reg);
1934 fill_regs(td, &status->pr_reg);
1936 sbuf_bcat(sb, status, sizeof(*status));
1937 free(status, M_TEMP);
1939 *sizep = sizeof(*status);
1943 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1946 elf_prfpregset_t *fpregset;
1948 td = (struct thread *)arg;
1950 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1951 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1952 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1953 fill_fpregs32(td, fpregset);
1955 fill_fpregs(td, fpregset);
1957 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1958 free(fpregset, M_TEMP);
1960 *sizep = sizeof(*fpregset);
1964 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1967 elf_thrmisc_t thrmisc;
1969 td = (struct thread *)arg;
1971 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1972 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1973 strcpy(thrmisc.pr_tname, td->td_name);
1974 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1976 *sizep = sizeof(thrmisc);
1980 * Allow for MD specific notes, as well as any MD
1981 * specific preparations for writing MI notes.
1984 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1990 td = (struct thread *)arg;
1992 if (size != 0 && sb != NULL)
1993 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1997 __elfN(dump_thread)(td, buf, &size);
1998 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1999 if (size != 0 && sb != NULL)
2000 sbuf_bcat(sb, buf, size);
2005 #ifdef KINFO_PROC_SIZE
2006 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2010 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2016 p = (struct proc *)arg;
2017 size = sizeof(structsize) + p->p_numthreads *
2018 sizeof(elf_kinfo_proc_t);
2021 KASSERT(*sizep == size, ("invalid size"));
2022 structsize = sizeof(elf_kinfo_proc_t);
2023 sbuf_bcat(sb, &structsize, sizeof(structsize));
2024 sx_slock(&proctree_lock);
2026 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2027 sx_sunlock(&proctree_lock);
2032 #ifdef KINFO_FILE_SIZE
2033 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2037 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2040 size_t size, sect_sz, i;
2041 ssize_t start_len, sect_len;
2042 int structsize, filedesc_flags;
2044 if (coredump_pack_fileinfo)
2045 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2049 p = (struct proc *)arg;
2050 structsize = sizeof(struct kinfo_file);
2053 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2054 sbuf_set_drain(sb, sbuf_drain_count, &size);
2055 sbuf_bcat(sb, &structsize, sizeof(structsize));
2057 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2062 sbuf_start_section(sb, &start_len);
2064 sbuf_bcat(sb, &structsize, sizeof(structsize));
2066 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2069 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2074 KASSERT(sect_sz <= *sizep,
2075 ("kern_proc_filedesc_out did not respect maxlen; "
2076 "requested %zu, got %zu", *sizep - sizeof(structsize),
2077 sect_sz - sizeof(structsize)));
2079 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2084 #ifdef KINFO_VMENTRY_SIZE
2085 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2089 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2093 int structsize, vmmap_flags;
2095 if (coredump_pack_vmmapinfo)
2096 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2100 p = (struct proc *)arg;
2101 structsize = sizeof(struct kinfo_vmentry);
2104 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2105 sbuf_set_drain(sb, sbuf_drain_count, &size);
2106 sbuf_bcat(sb, &structsize, sizeof(structsize));
2108 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2113 sbuf_bcat(sb, &structsize, sizeof(structsize));
2115 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2121 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2127 p = (struct proc *)arg;
2128 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2130 KASSERT(*sizep == size, ("invalid size"));
2131 structsize = sizeof(gid_t);
2132 sbuf_bcat(sb, &structsize, sizeof(structsize));
2133 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2140 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2146 p = (struct proc *)arg;
2147 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2149 KASSERT(*sizep == size, ("invalid size"));
2150 structsize = sizeof(p->p_fd->fd_cmask);
2151 sbuf_bcat(sb, &structsize, sizeof(structsize));
2152 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2158 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2161 struct rlimit rlim[RLIM_NLIMITS];
2165 p = (struct proc *)arg;
2166 size = sizeof(structsize) + sizeof(rlim);
2168 KASSERT(*sizep == size, ("invalid size"));
2169 structsize = sizeof(rlim);
2170 sbuf_bcat(sb, &structsize, sizeof(structsize));
2172 for (i = 0; i < RLIM_NLIMITS; i++)
2173 lim_rlimit_proc(p, i, &rlim[i]);
2175 sbuf_bcat(sb, rlim, sizeof(rlim));
2181 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2187 p = (struct proc *)arg;
2188 size = sizeof(structsize) + sizeof(p->p_osrel);
2190 KASSERT(*sizep == size, ("invalid size"));
2191 structsize = sizeof(p->p_osrel);
2192 sbuf_bcat(sb, &structsize, sizeof(structsize));
2193 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2199 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2202 elf_ps_strings_t ps_strings;
2206 p = (struct proc *)arg;
2207 size = sizeof(structsize) + sizeof(ps_strings);
2209 KASSERT(*sizep == size, ("invalid size"));
2210 structsize = sizeof(ps_strings);
2211 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2212 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2214 ps_strings = p->p_sysent->sv_psstrings;
2216 sbuf_bcat(sb, &structsize, sizeof(structsize));
2217 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2223 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2229 p = (struct proc *)arg;
2232 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2233 sbuf_set_drain(sb, sbuf_drain_count, &size);
2234 sbuf_bcat(sb, &structsize, sizeof(structsize));
2236 proc_getauxv(curthread, p, sb);
2242 structsize = sizeof(Elf_Auxinfo);
2243 sbuf_bcat(sb, &structsize, sizeof(structsize));
2245 proc_getauxv(curthread, p, sb);
2251 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2252 int32_t *osrel, const Elf_Phdr *pnote)
2254 const Elf_Note *note, *note0, *note_end;
2255 const char *note_name;
2260 /* We need some limit, might as well use PAGE_SIZE. */
2261 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2263 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2264 if (pnote->p_offset > PAGE_SIZE ||
2265 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2266 VOP_UNLOCK(imgp->vp, 0);
2267 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2268 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2269 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2270 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2271 curthread->td_ucred, NOCRED, NULL, curthread);
2273 uprintf("i/o error PT_NOTE\n");
2277 note = note0 = (const Elf_Note *)buf;
2278 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2280 note = note0 = (const Elf_Note *)(imgp->image_header +
2282 note_end = (const Elf_Note *)(imgp->image_header +
2283 pnote->p_offset + pnote->p_filesz);
2286 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2287 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2288 (const char *)note < sizeof(Elf_Note)) {
2292 if (note->n_namesz != checknote->hdr.n_namesz ||
2293 note->n_descsz != checknote->hdr.n_descsz ||
2294 note->n_type != checknote->hdr.n_type)
2296 note_name = (const char *)(note + 1);
2297 if (note_name + checknote->hdr.n_namesz >=
2298 (const char *)note_end || strncmp(checknote->vendor,
2299 note_name, checknote->hdr.n_namesz) != 0)
2303 * Fetch the osreldate for binary
2304 * from the ELF OSABI-note if necessary.
2306 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2307 checknote->trans_osrel != NULL) {
2308 res = checknote->trans_osrel(note, osrel);
2314 note = (const Elf_Note *)((const char *)(note + 1) +
2315 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2316 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2325 * Try to find the appropriate ABI-note section for checknote,
2326 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2327 * first page of the image is searched, the same as for headers.
2330 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2333 const Elf_Phdr *phdr;
2334 const Elf_Ehdr *hdr;
2337 hdr = (const Elf_Ehdr *)imgp->image_header;
2338 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2340 for (i = 0; i < hdr->e_phnum; i++) {
2341 if (phdr[i].p_type == PT_NOTE &&
2342 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2350 * Tell kern_execve.c about it, with a little help from the linker.
2352 static struct execsw __elfN(execsw) = {
2353 __CONCAT(exec_, __elfN(imgact)),
2354 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2356 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2359 __elfN(trans_prot)(Elf_Word flags)
2365 prot |= VM_PROT_EXECUTE;
2367 prot |= VM_PROT_WRITE;
2369 prot |= VM_PROT_READ;
2370 #if __ELF_WORD_SIZE == 32
2371 #if defined(__amd64__)
2372 if (i386_read_exec && (flags & PF_R))
2373 prot |= VM_PROT_EXECUTE;
2380 __elfN(untrans_prot)(vm_prot_t prot)
2385 if (prot & VM_PROT_EXECUTE)
2387 if (prot & VM_PROT_READ)
2389 if (prot & VM_PROT_WRITE)