2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2017 Dell EMC
5 * Copyright (c) 2000-2001, 2003 David O'Brien
6 * Copyright (c) 1995-1996 Søren Schmidt
7 * Copyright (c) 1996 Peter Wemm
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_capsicum.h"
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
55 #include <sys/procfs.h>
56 #include <sys/ptrace.h>
57 #include <sys/racct.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
61 #include <sys/sf_buf.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscall.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/vnode.h>
71 #include <sys/syslog.h>
72 #include <sys/eventhandler.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
89 static int __elfN(check_header)(const Elf_Ehdr *hdr);
90 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
91 const char *interp, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
94 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
95 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
108 #define CORE_BUF_SIZE (16 * 1024)
110 int __elfN(fallback_brand) = -1;
111 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
112 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
113 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
115 static int elf_legacy_coredump = 0;
116 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
117 &elf_legacy_coredump, 0,
118 "include all and only RW pages in core dumps");
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
127 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
128 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
129 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
131 #if __ELF_WORD_SIZE == 32
132 #if defined(__amd64__)
133 int i386_read_exec = 0;
134 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
135 "enable execution from readable segments");
139 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
141 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
143 static int __elfN(aslr_enabled) = 0;
144 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
145 &__elfN(aslr_enabled), 0,
146 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
147 ": enable address map randomization");
149 static int __elfN(pie_aslr_enabled) = 0;
150 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
151 &__elfN(pie_aslr_enabled), 0,
152 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
153 ": enable address map randomization for PIE binaries");
155 static int __elfN(aslr_honor_sbrk) = 1;
156 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
157 &__elfN(aslr_honor_sbrk), 0,
158 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
160 static int __elfN(aslr_stack_gap) = 3;
161 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
162 &__elfN(aslr_stack_gap), 0,
163 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
164 ": maximum percentage of main stack to waste on a random gap");
166 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
168 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
170 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
172 Elf_Brandnote __elfN(freebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
174 .hdr.n_descsz = sizeof(int32_t),
175 .hdr.n_type = NT_FREEBSD_ABI_TAG,
176 .vendor = FREEBSD_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = __elfN(freebsd_trans_osrel)
182 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
186 p = (uintptr_t)(note + 1);
187 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
188 *osrel = *(const int32_t *)(p);
193 static const char GNU_ABI_VENDOR[] = "GNU";
194 static int GNU_KFREEBSD_ABI_DESC = 3;
196 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
197 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
198 .hdr.n_descsz = 16, /* XXX at least 16 */
200 .vendor = GNU_ABI_VENDOR,
201 .flags = BN_TRANSLATE_OSREL,
202 .trans_osrel = kfreebsd_trans_osrel
206 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
208 const Elf32_Word *desc;
211 p = (uintptr_t)(note + 1);
212 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
214 desc = (const Elf32_Word *)p;
215 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
219 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
220 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
222 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
228 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
232 for (i = 0; i < MAX_BRANDS; i++) {
233 if (elf_brand_list[i] == NULL) {
234 elf_brand_list[i] = entry;
238 if (i == MAX_BRANDS) {
239 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
247 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
251 for (i = 0; i < MAX_BRANDS; i++) {
252 if (elf_brand_list[i] == entry) {
253 elf_brand_list[i] = NULL;
263 __elfN(brand_inuse)(Elf_Brandinfo *entry)
268 sx_slock(&allproc_lock);
269 FOREACH_PROC_IN_SYSTEM(p) {
270 if (p->p_sysent == entry->sysvec) {
275 sx_sunlock(&allproc_lock);
280 static Elf_Brandinfo *
281 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
282 int32_t *osrel, uint32_t *fctl0)
284 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
285 Elf_Brandinfo *bi, *bi_m;
287 int i, interp_name_len;
289 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
292 * We support four types of branding -- (1) the ELF EI_OSABI field
293 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
294 * branding w/in the ELF header, (3) path of the `interp_path'
295 * field, and (4) the ".note.ABI-tag" ELF section.
298 /* Look for an ".note.ABI-tag" ELF section */
300 for (i = 0; i < MAX_BRANDS; i++) {
301 bi = elf_brand_list[i];
304 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
306 if (hdr->e_machine == bi->machine && (bi->flags &
307 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
308 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
310 /* Give brand a chance to veto check_note's guess */
311 if (ret && bi->header_supported)
312 ret = bi->header_supported(imgp);
314 * If note checker claimed the binary, but the
315 * interpreter path in the image does not
316 * match default one for the brand, try to
317 * search for other brands with the same
318 * interpreter. Either there is better brand
319 * with the right interpreter, or, failing
320 * this, we return first brand which accepted
321 * our note and, optionally, header.
323 if (ret && bi_m == NULL && interp != NULL &&
324 (bi->interp_path == NULL ||
325 (strlen(bi->interp_path) + 1 != interp_name_len ||
326 strncmp(interp, bi->interp_path, interp_name_len)
338 /* If the executable has a brand, search for it in the brand list. */
339 for (i = 0; i < MAX_BRANDS; i++) {
340 bi = elf_brand_list[i];
341 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
342 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
344 if (hdr->e_machine == bi->machine &&
345 (hdr->e_ident[EI_OSABI] == bi->brand ||
346 (bi->compat_3_brand != NULL &&
347 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
348 bi->compat_3_brand) == 0))) {
349 /* Looks good, but give brand a chance to veto */
350 if (bi->header_supported == NULL ||
351 bi->header_supported(imgp)) {
353 * Again, prefer strictly matching
356 if (interp_name_len == 0 &&
357 bi->interp_path == NULL)
359 if (bi->interp_path != NULL &&
360 strlen(bi->interp_path) + 1 ==
361 interp_name_len && strncmp(interp,
362 bi->interp_path, interp_name_len) == 0)
372 /* No known brand, see if the header is recognized by any brand */
373 for (i = 0; i < MAX_BRANDS; i++) {
374 bi = elf_brand_list[i];
375 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
376 bi->header_supported == NULL)
378 if (hdr->e_machine == bi->machine) {
379 ret = bi->header_supported(imgp);
385 /* Lacking a known brand, search for a recognized interpreter. */
386 if (interp != NULL) {
387 for (i = 0; i < MAX_BRANDS; i++) {
388 bi = elf_brand_list[i];
389 if (bi == NULL || (bi->flags &
390 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
393 if (hdr->e_machine == bi->machine &&
394 bi->interp_path != NULL &&
395 /* ELF image p_filesz includes terminating zero */
396 strlen(bi->interp_path) + 1 == interp_name_len &&
397 strncmp(interp, bi->interp_path, interp_name_len)
398 == 0 && (bi->header_supported == NULL ||
399 bi->header_supported(imgp)))
404 /* Lacking a recognized interpreter, try the default brand */
405 for (i = 0; i < MAX_BRANDS; i++) {
406 bi = elf_brand_list[i];
407 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
408 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
410 if (hdr->e_machine == bi->machine &&
411 __elfN(fallback_brand) == bi->brand &&
412 (bi->header_supported == NULL ||
413 bi->header_supported(imgp)))
420 __elfN(check_header)(const Elf_Ehdr *hdr)
426 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
427 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
428 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
429 hdr->e_phentsize != sizeof(Elf_Phdr) ||
430 hdr->e_version != ELF_TARG_VER)
434 * Make sure we have at least one brand for this machine.
437 for (i = 0; i < MAX_BRANDS; i++) {
438 bi = elf_brand_list[i];
439 if (bi != NULL && bi->machine == hdr->e_machine)
449 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
450 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
457 * Create the page if it doesn't exist yet. Ignore errors.
459 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
460 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
463 * Find the page from the underlying object.
465 if (object != NULL) {
466 sf = vm_imgact_map_page(object, offset);
468 return (KERN_FAILURE);
469 off = offset - trunc_page(offset);
470 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
472 vm_imgact_unmap_page(sf);
474 return (KERN_FAILURE);
477 return (KERN_SUCCESS);
481 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
482 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
488 int error, locked, rv;
490 if (start != trunc_page(start)) {
491 rv = __elfN(map_partial)(map, object, offset, start,
492 round_page(start), prot);
493 if (rv != KERN_SUCCESS)
495 offset += round_page(start) - start;
496 start = round_page(start);
498 if (end != round_page(end)) {
499 rv = __elfN(map_partial)(map, object, offset +
500 trunc_page(end) - start, trunc_page(end), end, prot);
501 if (rv != KERN_SUCCESS)
503 end = trunc_page(end);
506 return (KERN_SUCCESS);
507 if ((offset & PAGE_MASK) != 0) {
509 * The mapping is not page aligned. This means that we have
512 rv = vm_map_fixed(map, NULL, 0, start, end - start,
513 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
514 if (rv != KERN_SUCCESS)
517 return (KERN_SUCCESS);
518 for (; start < end; start += sz) {
519 sf = vm_imgact_map_page(object, offset);
521 return (KERN_FAILURE);
522 off = offset - trunc_page(offset);
524 if (sz > PAGE_SIZE - off)
525 sz = PAGE_SIZE - off;
526 error = copyout((caddr_t)sf_buf_kva(sf) + off,
528 vm_imgact_unmap_page(sf);
530 return (KERN_FAILURE);
534 vm_object_reference(object);
535 rv = vm_map_fixed(map, object, offset, start, end - start,
536 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
537 (object != NULL ? MAP_VN_EXEC : 0));
538 if (rv != KERN_SUCCESS) {
539 locked = VOP_ISLOCKED(imgp->vp);
540 VOP_UNLOCK(imgp->vp, 0);
541 vm_object_deallocate(object);
542 vn_lock(imgp->vp, locked | LK_RETRY);
544 } else if (object != NULL) {
545 MPASS(imgp->vp->v_object == object);
546 VOP_SET_TEXT_CHECKED(imgp->vp);
549 return (KERN_SUCCESS);
553 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
554 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
560 vm_offset_t off, map_addr;
563 vm_ooffset_t file_addr;
566 * It's necessary to fail if the filsz + offset taken from the
567 * header is greater than the actual file pager object's size.
568 * If we were to allow this, then the vm_map_find() below would
569 * walk right off the end of the file object and into the ether.
571 * While I'm here, might as well check for something else that
572 * is invalid: filsz cannot be greater than memsz.
574 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
576 uprintf("elf_load_section: truncated ELF file\n");
580 object = imgp->object;
581 map = &imgp->proc->p_vmspace->vm_map;
582 map_addr = trunc_page((vm_offset_t)vmaddr);
583 file_addr = trunc_page(offset);
586 * We have two choices. We can either clear the data in the last page
587 * of an oversized mapping, or we can start the anon mapping a page
588 * early and copy the initialized data into that first page. We
593 else if (memsz > filsz)
594 map_len = trunc_page(offset + filsz) - file_addr;
596 map_len = round_page(offset + filsz) - file_addr;
599 /* cow flags: don't dump readonly sections in core */
600 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
601 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
603 rv = __elfN(map_insert)(imgp, map, object, file_addr,
604 map_addr, map_addr + map_len, prot, cow);
605 if (rv != KERN_SUCCESS)
608 /* we can stop now if we've covered it all */
615 * We have to get the remaining bit of the file into the first part
616 * of the oversized map segment. This is normally because the .data
617 * segment in the file is extended to provide bss. It's a neat idea
618 * to try and save a page, but it's a pain in the behind to implement.
620 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
622 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
623 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
625 /* This had damn well better be true! */
627 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
628 map_addr + map_len, prot, 0);
629 if (rv != KERN_SUCCESS)
634 sf = vm_imgact_map_page(object, offset + filsz);
638 /* send the page fragment to user space */
639 off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
640 error = copyout((caddr_t)sf_buf_kva(sf) + off,
641 (caddr_t)map_addr, copy_len);
642 vm_imgact_unmap_page(sf);
648 * Remove write access to the page if it was only granted by map_insert
651 if ((prot & VM_PROT_WRITE) == 0)
652 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
653 map_len), prot, FALSE);
659 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
660 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
667 ASSERT_VOP_LOCKED(imgp->vp, __func__);
672 for (i = 0; i < hdr->e_phnum; i++) {
673 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
676 /* Loadable segment */
677 prot = __elfN(trans_prot)(phdr[i].p_flags);
678 error = __elfN(load_section)(imgp, phdr[i].p_offset,
679 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
680 phdr[i].p_memsz, phdr[i].p_filesz, prot);
685 * Establish the base address if this is the first segment.
688 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
693 if (base_addrp != NULL)
694 *base_addrp = base_addr;
700 * Load the file "file" into memory. It may be either a shared object
703 * The "addr" reference parameter is in/out. On entry, it specifies
704 * the address where a shared object should be loaded. If the file is
705 * an executable, this value is ignored. On exit, "addr" specifies
706 * where the file was actually loaded.
708 * The "entry" reference parameter is out only. On exit, it specifies
709 * the entry point for the loaded file.
712 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
718 struct image_params image_params;
720 const Elf_Ehdr *hdr = NULL;
721 const Elf_Phdr *phdr = NULL;
722 struct nameidata *nd;
724 struct image_params *imgp;
726 u_long base_addr = 0;
729 #ifdef CAPABILITY_MODE
731 * XXXJA: This check can go away once we are sufficiently confident
732 * that the checks in namei() are correct.
734 if (IN_CAPABILITY_MODE(curthread))
738 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
740 attr = &tempdata->attr;
741 imgp = &tempdata->image_params;
744 * Initialize part of the common data
749 NDINIT(nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, file,
751 if ((error = namei(nd)) != 0) {
755 NDFREE(nd, NDF_ONLY_PNBUF);
756 imgp->vp = nd->ni_vp;
759 * Check permissions, modes, uid, etc on the file, and "open" it.
761 error = exec_check_permissions(imgp);
765 error = exec_map_first_page(imgp);
769 imgp->object = nd->ni_vp->v_object;
771 hdr = (const Elf_Ehdr *)imgp->image_header;
772 if ((error = __elfN(check_header)(hdr)) != 0)
774 if (hdr->e_type == ET_DYN)
776 else if (hdr->e_type == ET_EXEC)
783 /* Only support headers that fit within first page for now */
784 if ((hdr->e_phoff > PAGE_SIZE) ||
785 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
790 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
791 if (!aligned(phdr, Elf_Addr)) {
796 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
801 *entry = (unsigned long)hdr->e_entry + rbase;
805 exec_unmap_first_page(imgp);
809 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
812 free(tempdata, M_TEMP);
818 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
823 MPASS(vm_map_min(map) <= minv);
824 MPASS(maxv <= vm_map_max(map));
826 MPASS(minv + align < maxv);
827 arc4rand(&rbase, sizeof(rbase), 0);
828 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
829 res &= ~((u_long)align - 1);
833 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
834 res, minv, maxv, rbase));
836 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
837 res, maxv, minv, rbase));
842 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
843 const Elf_Phdr *phdr, u_long et_dyn_addr)
845 struct vmspace *vmspace;
847 u_long text_size, data_size, total_size, text_addr, data_addr;
848 u_long seg_size, seg_addr;
852 text_size = data_size = total_size = text_addr = data_addr = 0;
854 for (i = 0; i < hdr->e_phnum; i++) {
855 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
858 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
859 seg_size = round_page(phdr[i].p_memsz +
860 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
863 * Make the largest executable segment the official
864 * text segment and all others data.
866 * Note that obreak() assumes that data_addr + data_size == end
867 * of data load area, and the ELF file format expects segments
868 * to be sorted by address. If multiple data segments exist,
869 * the last one will be used.
872 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
873 text_size = seg_size;
874 text_addr = seg_addr;
876 data_size = seg_size;
877 data_addr = seg_addr;
879 total_size += seg_size;
882 if (data_addr == 0 && data_size == 0) {
883 data_addr = text_addr;
884 data_size = text_size;
888 * Check limits. It should be safe to check the
889 * limits after loading the segments since we do
890 * not actually fault in all the segments pages.
892 PROC_LOCK(imgp->proc);
893 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
894 err_str = "Data segment size exceeds process limit";
895 else if (text_size > maxtsiz)
896 err_str = "Text segment size exceeds system limit";
897 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
898 err_str = "Total segment size exceeds process limit";
899 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
900 err_str = "Data segment size exceeds resource limit";
901 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
902 err_str = "Total segment size exceeds resource limit";
903 PROC_UNLOCK(imgp->proc);
904 if (err_str != NULL) {
905 uprintf("%s\n", err_str);
909 vmspace = imgp->proc->p_vmspace;
910 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
911 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
912 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
913 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
919 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
920 char **interpp, bool *free_interpp)
924 int error, interp_name_len;
926 KASSERT(phdr->p_type == PT_INTERP,
927 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
928 ASSERT_VOP_LOCKED(imgp->vp, __func__);
932 /* Path to interpreter */
933 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
934 uprintf("Invalid PT_INTERP\n");
938 interp_name_len = phdr->p_filesz;
939 if (phdr->p_offset > PAGE_SIZE ||
940 interp_name_len > PAGE_SIZE - phdr->p_offset) {
942 * The vnode lock might be needed by the pagedaemon to
943 * clean pages owned by the vnode. Do not allow sleep
944 * waiting for memory with the vnode locked, instead
945 * try non-sleepable allocation first, and if it
946 * fails, go to the slow path were we drop the lock
947 * and do M_WAITOK. A text reference prevents
948 * modifications to the vnode content.
950 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
951 if (interp == NULL) {
952 VOP_UNLOCK(imgp->vp, 0);
953 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
954 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
957 error = vn_rdwr(UIO_READ, imgp->vp, interp,
958 interp_name_len, phdr->p_offset,
959 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
962 free(interp, M_TEMP);
963 uprintf("i/o error PT_INTERP %d\n", error);
966 interp[interp_name_len] = '\0';
969 *free_interpp = true;
973 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
974 if (interp[interp_name_len - 1] != '\0') {
975 uprintf("Invalid PT_INTERP\n");
980 *free_interpp = false;
985 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
986 const char *interp, u_long *addr, u_long *entry)
991 if (brand_info->emul_path != NULL &&
992 brand_info->emul_path[0] != '\0') {
993 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
994 snprintf(path, MAXPATHLEN, "%s%s",
995 brand_info->emul_path, interp);
996 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1002 if (brand_info->interp_newpath != NULL &&
1003 (brand_info->interp_path == NULL ||
1004 strcmp(interp, brand_info->interp_path) == 0)) {
1005 error = __elfN(load_file)(imgp->proc,
1006 brand_info->interp_newpath, addr, entry);
1011 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1015 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1020 * Impossible et_dyn_addr initial value indicating that the real base
1021 * must be calculated later with some randomization applied.
1023 #define ET_DYN_ADDR_RAND 1
1026 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1029 const Elf_Ehdr *hdr;
1030 const Elf_Phdr *phdr;
1031 Elf_Auxargs *elf_auxargs;
1032 struct vmspace *vmspace;
1035 Elf_Brandinfo *brand_info;
1036 struct sysentvec *sv;
1037 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1038 u_long maxalign, mapsz, maxv, maxv1;
1044 hdr = (const Elf_Ehdr *)imgp->image_header;
1047 * Do we have a valid ELF header ?
1049 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1050 * if particular brand doesn't support it.
1052 if (__elfN(check_header)(hdr) != 0 ||
1053 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1057 * From here on down, we return an errno, not -1, as we've
1058 * detected an ELF file.
1061 if ((hdr->e_phoff > PAGE_SIZE) ||
1062 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1063 /* Only support headers in first page for now */
1064 uprintf("Program headers not in the first page\n");
1067 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1068 if (!aligned(phdr, Elf_Addr)) {
1069 uprintf("Unaligned program headers\n");
1077 entry = proghdr = 0;
1079 free_interp = false;
1081 maxalign = PAGE_SIZE;
1084 for (i = 0; i < hdr->e_phnum; i++) {
1085 switch (phdr[i].p_type) {
1088 baddr = phdr[i].p_vaddr;
1089 if (phdr[i].p_align > maxalign)
1090 maxalign = phdr[i].p_align;
1091 mapsz += phdr[i].p_memsz;
1095 * If this segment contains the program headers,
1096 * remember their virtual address for the AT_PHDR
1097 * aux entry. Static binaries don't usually include
1100 if (phdr[i].p_offset == 0 &&
1101 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1102 <= phdr[i].p_filesz)
1103 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1106 /* Path to interpreter */
1107 if (interp != NULL) {
1108 uprintf("Multiple PT_INTERP headers\n");
1112 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1118 if (__elfN(nxstack))
1120 __elfN(trans_prot)(phdr[i].p_flags);
1121 imgp->stack_sz = phdr[i].p_memsz;
1123 case PT_PHDR: /* Program header table info */
1124 proghdr = phdr[i].p_vaddr;
1129 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1130 if (brand_info == NULL) {
1131 uprintf("ELF binary type \"%u\" not known.\n",
1132 hdr->e_ident[EI_OSABI]);
1136 sv = brand_info->sysvec;
1138 if (hdr->e_type == ET_DYN) {
1139 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1140 uprintf("Cannot execute shared object\n");
1145 * Honour the base load address from the dso if it is
1146 * non-zero for some reason.
1149 if ((sv->sv_flags & SV_ASLR) == 0 ||
1150 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1151 et_dyn_addr = ET_DYN_LOAD_ADDR;
1152 else if ((__elfN(pie_aslr_enabled) &&
1153 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1154 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1155 et_dyn_addr = ET_DYN_ADDR_RAND;
1157 et_dyn_addr = ET_DYN_LOAD_ADDR;
1162 * Avoid a possible deadlock if the current address space is destroyed
1163 * and that address space maps the locked vnode. In the common case,
1164 * the locked vnode's v_usecount is decremented but remains greater
1165 * than zero. Consequently, the vnode lock is not needed by vrele().
1166 * However, in cases where the vnode lock is external, such as nullfs,
1167 * v_usecount may become zero.
1169 * The VV_TEXT flag prevents modifications to the executable while
1170 * the vnode is unlocked.
1172 VOP_UNLOCK(imgp->vp, 0);
1175 * Decide whether to enable randomization of user mappings.
1176 * First, reset user preferences for the setid binaries.
1177 * Then, account for the support of the randomization by the
1178 * ABI, by user preferences, and make special treatment for
1181 if (imgp->credential_setid) {
1182 PROC_LOCK(imgp->proc);
1183 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1184 PROC_UNLOCK(imgp->proc);
1186 if ((sv->sv_flags & SV_ASLR) == 0 ||
1187 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1188 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1189 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1190 ("et_dyn_addr == RAND and !ASLR"));
1191 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1192 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1193 et_dyn_addr == ET_DYN_ADDR_RAND) {
1194 imgp->map_flags |= MAP_ASLR;
1196 * If user does not care about sbrk, utilize the bss
1197 * grow region for mappings as well. We can select
1198 * the base for the image anywere and still not suffer
1199 * from the fragmentation.
1201 if (!__elfN(aslr_honor_sbrk) ||
1202 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1203 imgp->map_flags |= MAP_ASLR_IGNSTART;
1206 error = exec_new_vmspace(imgp, sv);
1207 vmspace = imgp->proc->p_vmspace;
1208 map = &vmspace->vm_map;
1210 imgp->proc->p_sysent = sv;
1212 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1213 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1214 KASSERT((map->flags & MAP_ASLR) != 0,
1215 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1216 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1217 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1218 /* reserve half of the address space to interpreter */
1219 maxv / 2, 1UL << flsl(maxalign));
1222 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1226 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1230 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1234 entry = (u_long)hdr->e_entry + et_dyn_addr;
1237 * We load the dynamic linker where a userland call
1238 * to mmap(0, ...) would put it. The rationale behind this
1239 * calculation is that it leaves room for the heap to grow to
1240 * its maximum allowed size.
1242 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1244 if ((map->flags & MAP_ASLR) != 0) {
1245 maxv1 = maxv / 2 + addr / 2;
1246 MPASS(maxv1 >= addr); /* No overflow */
1247 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1248 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1250 map->anon_loc = addr;
1253 imgp->entry_addr = entry;
1255 if (interp != NULL) {
1256 VOP_UNLOCK(imgp->vp, 0);
1257 if ((map->flags & MAP_ASLR) != 0) {
1258 /* Assume that interpeter fits into 1/4 of AS */
1259 maxv1 = maxv / 2 + addr / 2;
1260 MPASS(maxv1 >= addr); /* No overflow */
1261 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1264 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1266 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1273 * Construct auxargs table (used by the fixup routine)
1275 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1276 if (elf_auxargs == NULL) {
1277 VOP_UNLOCK(imgp->vp, 0);
1278 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1279 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1281 elf_auxargs->execfd = -1;
1282 elf_auxargs->phdr = proghdr + et_dyn_addr;
1283 elf_auxargs->phent = hdr->e_phentsize;
1284 elf_auxargs->phnum = hdr->e_phnum;
1285 elf_auxargs->pagesz = PAGE_SIZE;
1286 elf_auxargs->base = addr;
1287 elf_auxargs->flags = 0;
1288 elf_auxargs->entry = entry;
1289 elf_auxargs->hdr_eflags = hdr->e_flags;
1291 imgp->auxargs = elf_auxargs;
1292 imgp->interpreted = 0;
1293 imgp->reloc_base = addr;
1294 imgp->proc->p_osrel = osrel;
1295 imgp->proc->p_fctl0 = fctl0;
1296 imgp->proc->p_elf_machine = hdr->e_machine;
1297 imgp->proc->p_elf_flags = hdr->e_flags;
1301 free(interp, M_TEMP);
1305 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1308 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1310 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1311 Elf_Auxinfo *argarray, *pos;
1312 Elf_Addr *base, *auxbase;
1315 base = (Elf_Addr *)*stack_base;
1316 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1317 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1320 if (args->execfd != -1)
1321 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1322 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1323 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1324 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1325 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1326 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1327 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1328 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1329 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1330 if (imgp->execpathp != 0)
1331 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1332 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1333 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1334 if (imgp->canary != 0) {
1335 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1336 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1338 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1339 if (imgp->pagesizes != 0) {
1340 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1341 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1343 if (imgp->sysent->sv_timekeep_base != 0) {
1344 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1345 imgp->sysent->sv_timekeep_base);
1347 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1348 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1349 imgp->sysent->sv_stackprot);
1350 if (imgp->sysent->sv_hwcap != NULL)
1351 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1352 if (imgp->sysent->sv_hwcap2 != NULL)
1353 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1354 AUXARGS_ENTRY(pos, AT_NULL, 0);
1356 free(imgp->auxargs, M_TEMP);
1357 imgp->auxargs = NULL;
1358 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1360 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1361 free(argarray, M_TEMP);
1366 if (suword(base, imgp->args->argc) == -1)
1368 *stack_base = (register_t *)base;
1373 * Code for generating ELF core dumps.
1376 typedef void (*segment_callback)(vm_map_entry_t, void *);
1378 /* Closure for cb_put_phdr(). */
1379 struct phdr_closure {
1380 Elf_Phdr *phdr; /* Program header to fill in */
1381 Elf_Off offset; /* Offset of segment in core file */
1384 /* Closure for cb_size_segment(). */
1385 struct sseg_closure {
1386 int count; /* Count of writable segments. */
1387 size_t size; /* Total size of all writable segments. */
1390 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1393 int type; /* Note type. */
1394 outfunc_t outfunc; /* Output function. */
1395 void *outarg; /* Argument for the output function. */
1396 size_t outsize; /* Output size. */
1397 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1400 TAILQ_HEAD(note_info_list, note_info);
1402 /* Coredump output parameters. */
1403 struct coredump_params {
1405 struct ucred *active_cred;
1406 struct ucred *file_cred;
1409 struct compressor *comp;
1412 extern int compress_user_cores;
1413 extern int compress_user_cores_level;
1415 static void cb_put_phdr(vm_map_entry_t, void *);
1416 static void cb_size_segment(vm_map_entry_t, void *);
1417 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1419 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1420 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1421 struct note_info_list *, size_t);
1422 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1424 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1425 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1426 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1427 static int sbuf_drain_core_output(void *, const char *, int);
1429 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1430 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1431 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1432 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1433 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1434 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1435 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1436 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1437 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1438 static void note_procstat_files(void *, struct sbuf *, size_t *);
1439 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1440 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1441 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1442 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1443 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1446 * Write out a core segment to the compression stream.
1449 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1455 chunk_len = MIN(len, CORE_BUF_SIZE);
1458 * We can get EFAULT error here.
1459 * In that case zero out the current chunk of the segment.
1461 error = copyin(base, buf, chunk_len);
1463 bzero(buf, chunk_len);
1464 error = compressor_write(p->comp, buf, chunk_len);
1474 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1477 return (core_write((struct coredump_params *)arg, base, len, offset,
1482 core_write(struct coredump_params *p, const void *base, size_t len,
1483 off_t offset, enum uio_seg seg)
1486 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1487 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1488 p->active_cred, p->file_cred, NULL, p->td));
1492 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1497 if (p->comp != NULL)
1498 return (compress_chunk(p, base, tmpbuf, len));
1501 * EFAULT is a non-fatal error that we can get, for example,
1502 * if the segment is backed by a file but extends beyond its
1505 error = core_write(p, base, len, offset, UIO_USERSPACE);
1506 if (error == EFAULT) {
1507 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1508 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1509 "for process %s\n", base, len, offset, curproc->p_comm);
1512 * Write a "real" zero byte at the end of the target region
1513 * in the case this is the last segment.
1514 * The intermediate space will be implicitly zero-filled.
1516 error = core_write(p, zero_region, 1, offset + len - 1,
1523 * Drain into a core file.
1526 sbuf_drain_core_output(void *arg, const char *data, int len)
1528 struct coredump_params *p;
1531 p = (struct coredump_params *)arg;
1534 * Some kern_proc out routines that print to this sbuf may
1535 * call us with the process lock held. Draining with the
1536 * non-sleepable lock held is unsafe. The lock is needed for
1537 * those routines when dumping a live process. In our case we
1538 * can safely release the lock before draining and acquire
1541 locked = PROC_LOCKED(p->td->td_proc);
1543 PROC_UNLOCK(p->td->td_proc);
1544 if (p->comp != NULL)
1545 error = compressor_write(p->comp, __DECONST(char *, data), len);
1547 error = core_write(p, __DECONST(void *, data), len, p->offset,
1550 PROC_LOCK(p->td->td_proc);
1558 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1560 struct ucred *cred = td->td_ucred;
1562 struct sseg_closure seginfo;
1563 struct note_info_list notelst;
1564 struct coredump_params params;
1565 struct note_info *ninfo;
1567 size_t hdrsize, notesz, coresize;
1571 TAILQ_INIT(¬elst);
1573 /* Size the program segments. */
1576 each_dumpable_segment(td, cb_size_segment, &seginfo);
1579 * Collect info about the core file header area.
1581 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1582 if (seginfo.count + 1 >= PN_XNUM)
1583 hdrsize += sizeof(Elf_Shdr);
1584 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1585 coresize = round_page(hdrsize + notesz) + seginfo.size;
1587 /* Set up core dump parameters. */
1589 params.active_cred = cred;
1590 params.file_cred = NOCRED;
1597 PROC_LOCK(td->td_proc);
1598 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1599 PROC_UNLOCK(td->td_proc);
1606 if (coresize >= limit) {
1611 /* Create a compression stream if necessary. */
1612 if (compress_user_cores != 0) {
1613 params.comp = compressor_init(core_compressed_write,
1614 compress_user_cores, CORE_BUF_SIZE,
1615 compress_user_cores_level, ¶ms);
1616 if (params.comp == NULL) {
1620 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1624 * Allocate memory for building the header, fill it up,
1625 * and write it out following the notes.
1627 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1628 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1631 /* Write the contents of all of the writable segments. */
1637 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1638 offset = round_page(hdrsize + notesz);
1639 for (i = 0; i < seginfo.count; i++) {
1640 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1641 php->p_filesz, offset, ¶ms, tmpbuf);
1644 offset += php->p_filesz;
1647 if (error == 0 && params.comp != NULL)
1648 error = compressor_flush(params.comp);
1652 "Failed to write core file for process %s (error %d)\n",
1653 curproc->p_comm, error);
1657 free(tmpbuf, M_TEMP);
1658 if (params.comp != NULL)
1659 compressor_fini(params.comp);
1660 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1661 TAILQ_REMOVE(¬elst, ninfo, link);
1662 free(ninfo, M_TEMP);
1671 * A callback for each_dumpable_segment() to write out the segment's
1672 * program header entry.
1675 cb_put_phdr(vm_map_entry_t entry, void *closure)
1677 struct phdr_closure *phc = (struct phdr_closure *)closure;
1678 Elf_Phdr *phdr = phc->phdr;
1680 phc->offset = round_page(phc->offset);
1682 phdr->p_type = PT_LOAD;
1683 phdr->p_offset = phc->offset;
1684 phdr->p_vaddr = entry->start;
1686 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1687 phdr->p_align = PAGE_SIZE;
1688 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1690 phc->offset += phdr->p_filesz;
1695 * A callback for each_dumpable_segment() to gather information about
1696 * the number of segments and their total size.
1699 cb_size_segment(vm_map_entry_t entry, void *closure)
1701 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1704 ssc->size += entry->end - entry->start;
1708 * For each writable segment in the process's memory map, call the given
1709 * function with a pointer to the map entry and some arbitrary
1710 * caller-supplied data.
1713 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1715 struct proc *p = td->td_proc;
1716 vm_map_t map = &p->p_vmspace->vm_map;
1717 vm_map_entry_t entry;
1718 vm_object_t backing_object, object;
1719 boolean_t ignore_entry;
1721 vm_map_lock_read(map);
1722 for (entry = map->header.next; entry != &map->header;
1723 entry = entry->next) {
1725 * Don't dump inaccessible mappings, deal with legacy
1728 * Note that read-only segments related to the elf binary
1729 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1730 * need to arbitrarily ignore such segments.
1732 if (elf_legacy_coredump) {
1733 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1736 if ((entry->protection & VM_PROT_ALL) == 0)
1741 * Dont include memory segment in the coredump if
1742 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1743 * madvise(2). Do not dump submaps (i.e. parts of the
1746 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1749 if ((object = entry->object.vm_object) == NULL)
1752 /* Ignore memory-mapped devices and such things. */
1753 VM_OBJECT_RLOCK(object);
1754 while ((backing_object = object->backing_object) != NULL) {
1755 VM_OBJECT_RLOCK(backing_object);
1756 VM_OBJECT_RUNLOCK(object);
1757 object = backing_object;
1759 ignore_entry = object->type != OBJT_DEFAULT &&
1760 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1761 object->type != OBJT_PHYS;
1762 VM_OBJECT_RUNLOCK(object);
1766 (*func)(entry, closure);
1768 vm_map_unlock_read(map);
1772 * Write the core file header to the file, including padding up to
1773 * the page boundary.
1776 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1777 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1779 struct note_info *ninfo;
1783 /* Fill in the header. */
1784 bzero(hdr, hdrsize);
1785 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1787 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1788 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1789 sbuf_start_section(sb, NULL);
1790 sbuf_bcat(sb, hdr, hdrsize);
1791 TAILQ_FOREACH(ninfo, notelst, link)
1792 __elfN(putnote)(ninfo, sb);
1793 /* Align up to a page boundary for the program segments. */
1794 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1795 error = sbuf_finish(sb);
1802 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1812 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1815 * To have the debugger select the right thread (LWP) as the initial
1816 * thread, we dump the state of the thread passed to us in td first.
1817 * This is the thread that causes the core dump and thus likely to
1818 * be the right thread one wants to have selected in the debugger.
1821 while (thr != NULL) {
1822 size += register_note(list, NT_PRSTATUS,
1823 __elfN(note_prstatus), thr);
1824 size += register_note(list, NT_FPREGSET,
1825 __elfN(note_fpregset), thr);
1826 size += register_note(list, NT_THRMISC,
1827 __elfN(note_thrmisc), thr);
1828 size += register_note(list, NT_PTLWPINFO,
1829 __elfN(note_ptlwpinfo), thr);
1830 size += register_note(list, -1,
1831 __elfN(note_threadmd), thr);
1833 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1834 TAILQ_NEXT(thr, td_plist);
1836 thr = TAILQ_NEXT(thr, td_plist);
1839 size += register_note(list, NT_PROCSTAT_PROC,
1840 __elfN(note_procstat_proc), p);
1841 size += register_note(list, NT_PROCSTAT_FILES,
1842 note_procstat_files, p);
1843 size += register_note(list, NT_PROCSTAT_VMMAP,
1844 note_procstat_vmmap, p);
1845 size += register_note(list, NT_PROCSTAT_GROUPS,
1846 note_procstat_groups, p);
1847 size += register_note(list, NT_PROCSTAT_UMASK,
1848 note_procstat_umask, p);
1849 size += register_note(list, NT_PROCSTAT_RLIMIT,
1850 note_procstat_rlimit, p);
1851 size += register_note(list, NT_PROCSTAT_OSREL,
1852 note_procstat_osrel, p);
1853 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1854 __elfN(note_procstat_psstrings), p);
1855 size += register_note(list, NT_PROCSTAT_AUXV,
1856 __elfN(note_procstat_auxv), p);
1862 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1868 struct phdr_closure phc;
1870 ehdr = (Elf_Ehdr *)hdr;
1872 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1873 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1874 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1875 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1876 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1877 ehdr->e_ident[EI_DATA] = ELF_DATA;
1878 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1879 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1880 ehdr->e_ident[EI_ABIVERSION] = 0;
1881 ehdr->e_ident[EI_PAD] = 0;
1882 ehdr->e_type = ET_CORE;
1883 ehdr->e_machine = td->td_proc->p_elf_machine;
1884 ehdr->e_version = EV_CURRENT;
1886 ehdr->e_phoff = sizeof(Elf_Ehdr);
1887 ehdr->e_flags = td->td_proc->p_elf_flags;
1888 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1889 ehdr->e_phentsize = sizeof(Elf_Phdr);
1890 ehdr->e_shentsize = sizeof(Elf_Shdr);
1891 ehdr->e_shstrndx = SHN_UNDEF;
1892 if (numsegs + 1 < PN_XNUM) {
1893 ehdr->e_phnum = numsegs + 1;
1896 ehdr->e_phnum = PN_XNUM;
1899 ehdr->e_shoff = ehdr->e_phoff +
1900 (numsegs + 1) * ehdr->e_phentsize;
1901 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1902 ("e_shoff: %zu, hdrsize - shdr: %zu",
1903 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1905 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1906 memset(shdr, 0, sizeof(*shdr));
1908 * A special first section is used to hold large segment and
1909 * section counts. This was proposed by Sun Microsystems in
1910 * Solaris and has been adopted by Linux; the standard ELF
1911 * tools are already familiar with the technique.
1913 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1914 * (or 12-7 depending on the version of the document) for more
1917 shdr->sh_type = SHT_NULL;
1918 shdr->sh_size = ehdr->e_shnum;
1919 shdr->sh_link = ehdr->e_shstrndx;
1920 shdr->sh_info = numsegs + 1;
1924 * Fill in the program header entries.
1926 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1928 /* The note segement. */
1929 phdr->p_type = PT_NOTE;
1930 phdr->p_offset = hdrsize;
1933 phdr->p_filesz = notesz;
1935 phdr->p_flags = PF_R;
1936 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1939 /* All the writable segments from the program. */
1941 phc.offset = round_page(hdrsize + notesz);
1942 each_dumpable_segment(td, cb_put_phdr, &phc);
1946 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1948 struct note_info *ninfo;
1949 size_t size, notesize;
1952 out(arg, NULL, &size);
1953 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1955 ninfo->outfunc = out;
1956 ninfo->outarg = arg;
1957 ninfo->outsize = size;
1958 TAILQ_INSERT_TAIL(list, ninfo, link);
1963 notesize = sizeof(Elf_Note) + /* note header */
1964 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1966 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1972 append_note_data(const void *src, void *dst, size_t len)
1976 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1978 bcopy(src, dst, len);
1979 bzero((char *)dst + len, padded_len - len);
1981 return (padded_len);
1985 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1993 note = (Elf_Note *)buf;
1994 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1995 note->n_descsz = size;
1996 note->n_type = type;
1997 buf += sizeof(*note);
1998 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1999 sizeof(FREEBSD_ABI_VENDOR));
2000 append_note_data(src, buf, size);
2005 notesize = sizeof(Elf_Note) + /* note header */
2006 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2008 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2014 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2017 ssize_t old_len, sect_len;
2018 size_t new_len, descsz, i;
2020 if (ninfo->type == -1) {
2021 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2025 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2026 note.n_descsz = ninfo->outsize;
2027 note.n_type = ninfo->type;
2029 sbuf_bcat(sb, ¬e, sizeof(note));
2030 sbuf_start_section(sb, &old_len);
2031 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2032 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2033 if (note.n_descsz == 0)
2035 sbuf_start_section(sb, &old_len);
2036 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2037 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2041 new_len = (size_t)sect_len;
2042 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2043 if (new_len < descsz) {
2045 * It is expected that individual note emitters will correctly
2046 * predict their expected output size and fill up to that size
2047 * themselves, padding in a format-specific way if needed.
2048 * However, in case they don't, just do it here with zeros.
2050 for (i = 0; i < descsz - new_len; i++)
2052 } else if (new_len > descsz) {
2054 * We can't always truncate sb -- we may have drained some
2057 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2058 "read it (%zu > %zu). Since it is longer than "
2059 "expected, this coredump's notes are corrupt. THIS "
2060 "IS A BUG in the note_procstat routine for type %u.\n",
2061 __func__, (unsigned)note.n_type, new_len, descsz,
2062 (unsigned)note.n_type));
2067 * Miscellaneous note out functions.
2070 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2071 #include <compat/freebsd32/freebsd32.h>
2072 #include <compat/freebsd32/freebsd32_signal.h>
2074 typedef struct prstatus32 elf_prstatus_t;
2075 typedef struct prpsinfo32 elf_prpsinfo_t;
2076 typedef struct fpreg32 elf_prfpregset_t;
2077 typedef struct fpreg32 elf_fpregset_t;
2078 typedef struct reg32 elf_gregset_t;
2079 typedef struct thrmisc32 elf_thrmisc_t;
2080 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2081 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2082 typedef uint32_t elf_ps_strings_t;
2084 typedef prstatus_t elf_prstatus_t;
2085 typedef prpsinfo_t elf_prpsinfo_t;
2086 typedef prfpregset_t elf_prfpregset_t;
2087 typedef prfpregset_t elf_fpregset_t;
2088 typedef gregset_t elf_gregset_t;
2089 typedef thrmisc_t elf_thrmisc_t;
2090 #define ELF_KERN_PROC_MASK 0
2091 typedef struct kinfo_proc elf_kinfo_proc_t;
2092 typedef vm_offset_t elf_ps_strings_t;
2096 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2102 elf_prpsinfo_t *psinfo;
2105 p = (struct proc *)arg;
2107 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2108 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2109 psinfo->pr_version = PRPSINFO_VERSION;
2110 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2111 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2113 if (p->p_args != NULL) {
2114 len = sizeof(psinfo->pr_psargs) - 1;
2115 if (len > p->p_args->ar_length)
2116 len = p->p_args->ar_length;
2117 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2123 sbuf_new(&sbarg, psinfo->pr_psargs,
2124 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2125 error = proc_getargv(curthread, p, &sbarg);
2127 if (sbuf_finish(&sbarg) == 0)
2128 len = sbuf_len(&sbarg) - 1;
2130 len = sizeof(psinfo->pr_psargs) - 1;
2131 sbuf_delete(&sbarg);
2133 if (error || len == 0)
2134 strlcpy(psinfo->pr_psargs, p->p_comm,
2135 sizeof(psinfo->pr_psargs));
2137 KASSERT(len < sizeof(psinfo->pr_psargs),
2138 ("len is too long: %zu vs %zu", len,
2139 sizeof(psinfo->pr_psargs)));
2140 cp = psinfo->pr_psargs;
2143 cp = memchr(cp, '\0', end - cp);
2149 psinfo->pr_pid = p->p_pid;
2150 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2151 free(psinfo, M_TEMP);
2153 *sizep = sizeof(*psinfo);
2157 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2160 elf_prstatus_t *status;
2162 td = (struct thread *)arg;
2164 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2165 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2166 status->pr_version = PRSTATUS_VERSION;
2167 status->pr_statussz = sizeof(elf_prstatus_t);
2168 status->pr_gregsetsz = sizeof(elf_gregset_t);
2169 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2170 status->pr_osreldate = osreldate;
2171 status->pr_cursig = td->td_proc->p_sig;
2172 status->pr_pid = td->td_tid;
2173 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2174 fill_regs32(td, &status->pr_reg);
2176 fill_regs(td, &status->pr_reg);
2178 sbuf_bcat(sb, status, sizeof(*status));
2179 free(status, M_TEMP);
2181 *sizep = sizeof(*status);
2185 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2188 elf_prfpregset_t *fpregset;
2190 td = (struct thread *)arg;
2192 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2193 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2194 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2195 fill_fpregs32(td, fpregset);
2197 fill_fpregs(td, fpregset);
2199 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2200 free(fpregset, M_TEMP);
2202 *sizep = sizeof(*fpregset);
2206 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2209 elf_thrmisc_t thrmisc;
2211 td = (struct thread *)arg;
2213 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2214 bzero(&thrmisc, sizeof(thrmisc));
2215 strcpy(thrmisc.pr_tname, td->td_name);
2216 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2218 *sizep = sizeof(thrmisc);
2222 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2227 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2228 struct ptrace_lwpinfo32 pl;
2230 struct ptrace_lwpinfo pl;
2233 td = (struct thread *)arg;
2234 size = sizeof(structsize) + sizeof(pl);
2236 KASSERT(*sizep == size, ("invalid size"));
2237 structsize = sizeof(pl);
2238 sbuf_bcat(sb, &structsize, sizeof(structsize));
2239 bzero(&pl, sizeof(pl));
2240 pl.pl_lwpid = td->td_tid;
2241 pl.pl_event = PL_EVENT_NONE;
2242 pl.pl_sigmask = td->td_sigmask;
2243 pl.pl_siglist = td->td_siglist;
2244 if (td->td_si.si_signo != 0) {
2245 pl.pl_event = PL_EVENT_SIGNAL;
2246 pl.pl_flags |= PL_FLAG_SI;
2247 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2248 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2250 pl.pl_siginfo = td->td_si;
2253 strcpy(pl.pl_tdname, td->td_name);
2254 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2255 sbuf_bcat(sb, &pl, sizeof(pl));
2261 * Allow for MD specific notes, as well as any MD
2262 * specific preparations for writing MI notes.
2265 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2271 td = (struct thread *)arg;
2273 if (size != 0 && sb != NULL)
2274 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2278 __elfN(dump_thread)(td, buf, &size);
2279 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2280 if (size != 0 && sb != NULL)
2281 sbuf_bcat(sb, buf, size);
2286 #ifdef KINFO_PROC_SIZE
2287 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2291 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2297 p = (struct proc *)arg;
2298 size = sizeof(structsize) + p->p_numthreads *
2299 sizeof(elf_kinfo_proc_t);
2302 KASSERT(*sizep == size, ("invalid size"));
2303 structsize = sizeof(elf_kinfo_proc_t);
2304 sbuf_bcat(sb, &structsize, sizeof(structsize));
2306 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2311 #ifdef KINFO_FILE_SIZE
2312 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2316 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2319 size_t size, sect_sz, i;
2320 ssize_t start_len, sect_len;
2321 int structsize, filedesc_flags;
2323 if (coredump_pack_fileinfo)
2324 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2328 p = (struct proc *)arg;
2329 structsize = sizeof(struct kinfo_file);
2332 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2333 sbuf_set_drain(sb, sbuf_count_drain, &size);
2334 sbuf_bcat(sb, &structsize, sizeof(structsize));
2336 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2341 sbuf_start_section(sb, &start_len);
2343 sbuf_bcat(sb, &structsize, sizeof(structsize));
2345 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2348 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2353 KASSERT(sect_sz <= *sizep,
2354 ("kern_proc_filedesc_out did not respect maxlen; "
2355 "requested %zu, got %zu", *sizep - sizeof(structsize),
2356 sect_sz - sizeof(structsize)));
2358 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2363 #ifdef KINFO_VMENTRY_SIZE
2364 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2368 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2372 int structsize, vmmap_flags;
2374 if (coredump_pack_vmmapinfo)
2375 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2379 p = (struct proc *)arg;
2380 structsize = sizeof(struct kinfo_vmentry);
2383 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2384 sbuf_set_drain(sb, sbuf_count_drain, &size);
2385 sbuf_bcat(sb, &structsize, sizeof(structsize));
2387 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2392 sbuf_bcat(sb, &structsize, sizeof(structsize));
2394 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2400 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2406 p = (struct proc *)arg;
2407 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2409 KASSERT(*sizep == size, ("invalid size"));
2410 structsize = sizeof(gid_t);
2411 sbuf_bcat(sb, &structsize, sizeof(structsize));
2412 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2419 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2425 p = (struct proc *)arg;
2426 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2428 KASSERT(*sizep == size, ("invalid size"));
2429 structsize = sizeof(p->p_fd->fd_cmask);
2430 sbuf_bcat(sb, &structsize, sizeof(structsize));
2431 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2437 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2440 struct rlimit rlim[RLIM_NLIMITS];
2444 p = (struct proc *)arg;
2445 size = sizeof(structsize) + sizeof(rlim);
2447 KASSERT(*sizep == size, ("invalid size"));
2448 structsize = sizeof(rlim);
2449 sbuf_bcat(sb, &structsize, sizeof(structsize));
2451 for (i = 0; i < RLIM_NLIMITS; i++)
2452 lim_rlimit_proc(p, i, &rlim[i]);
2454 sbuf_bcat(sb, rlim, sizeof(rlim));
2460 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2466 p = (struct proc *)arg;
2467 size = sizeof(structsize) + sizeof(p->p_osrel);
2469 KASSERT(*sizep == size, ("invalid size"));
2470 structsize = sizeof(p->p_osrel);
2471 sbuf_bcat(sb, &structsize, sizeof(structsize));
2472 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2478 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2481 elf_ps_strings_t ps_strings;
2485 p = (struct proc *)arg;
2486 size = sizeof(structsize) + sizeof(ps_strings);
2488 KASSERT(*sizep == size, ("invalid size"));
2489 structsize = sizeof(ps_strings);
2490 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2491 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2493 ps_strings = p->p_sysent->sv_psstrings;
2495 sbuf_bcat(sb, &structsize, sizeof(structsize));
2496 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2502 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2508 p = (struct proc *)arg;
2511 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2512 sbuf_set_drain(sb, sbuf_count_drain, &size);
2513 sbuf_bcat(sb, &structsize, sizeof(structsize));
2515 proc_getauxv(curthread, p, sb);
2521 structsize = sizeof(Elf_Auxinfo);
2522 sbuf_bcat(sb, &structsize, sizeof(structsize));
2524 proc_getauxv(curthread, p, sb);
2530 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2531 const char *note_vendor, const Elf_Phdr *pnote,
2532 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2534 const Elf_Note *note, *note0, *note_end;
2535 const char *note_name;
2540 /* We need some limit, might as well use PAGE_SIZE. */
2541 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2543 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2544 if (pnote->p_offset > PAGE_SIZE ||
2545 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2546 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2548 VOP_UNLOCK(imgp->vp, 0);
2549 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2550 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2552 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2553 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2554 curthread->td_ucred, NOCRED, NULL, curthread);
2556 uprintf("i/o error PT_NOTE\n");
2559 note = note0 = (const Elf_Note *)buf;
2560 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2562 note = note0 = (const Elf_Note *)(imgp->image_header +
2564 note_end = (const Elf_Note *)(imgp->image_header +
2565 pnote->p_offset + pnote->p_filesz);
2568 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2569 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2570 (const char *)note < sizeof(Elf_Note)) {
2573 if (note->n_namesz != checknote->n_namesz ||
2574 note->n_descsz != checknote->n_descsz ||
2575 note->n_type != checknote->n_type)
2577 note_name = (const char *)(note + 1);
2578 if (note_name + checknote->n_namesz >=
2579 (const char *)note_end || strncmp(note_vendor,
2580 note_name, checknote->n_namesz) != 0)
2583 if (cb(note, cb_arg, &res))
2586 note = (const Elf_Note *)((const char *)(note + 1) +
2587 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2588 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2597 struct brandnote_cb_arg {
2598 Elf_Brandnote *brandnote;
2603 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2605 struct brandnote_cb_arg *arg;
2610 * Fetch the osreldate for binary from the ELF OSABI-note if
2613 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2614 arg->brandnote->trans_osrel != NULL ?
2615 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2620 static Elf_Note fctl_note = {
2621 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2622 .n_descsz = sizeof(uint32_t),
2623 .n_type = NT_FREEBSD_FEATURE_CTL,
2626 struct fctl_cb_arg {
2631 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2633 struct fctl_cb_arg *arg;
2634 const Elf32_Word *desc;
2638 p = (uintptr_t)(note + 1);
2639 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2640 desc = (const Elf32_Word *)p;
2641 *arg->fctl0 = desc[0];
2646 * Try to find the appropriate ABI-note section for checknote, fetch
2647 * the osreldate and feature control flags for binary from the ELF
2648 * OSABI-note. Only the first page of the image is searched, the same
2652 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2653 int32_t *osrel, uint32_t *fctl0)
2655 const Elf_Phdr *phdr;
2656 const Elf_Ehdr *hdr;
2657 struct brandnote_cb_arg b_arg;
2658 struct fctl_cb_arg f_arg;
2661 hdr = (const Elf_Ehdr *)imgp->image_header;
2662 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2663 b_arg.brandnote = brandnote;
2664 b_arg.osrel = osrel;
2665 f_arg.fctl0 = fctl0;
2667 for (i = 0; i < hdr->e_phnum; i++) {
2668 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2669 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2671 for (j = 0; j < hdr->e_phnum; j++) {
2672 if (phdr[j].p_type == PT_NOTE &&
2673 __elfN(parse_notes)(imgp, &fctl_note,
2674 FREEBSD_ABI_VENDOR, &phdr[j],
2675 note_fctl_cb, &f_arg))
2686 * Tell kern_execve.c about it, with a little help from the linker.
2688 static struct execsw __elfN(execsw) = {
2689 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2690 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2692 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2695 __elfN(trans_prot)(Elf_Word flags)
2701 prot |= VM_PROT_EXECUTE;
2703 prot |= VM_PROT_WRITE;
2705 prot |= VM_PROT_READ;
2706 #if __ELF_WORD_SIZE == 32
2707 #if defined(__amd64__)
2708 if (i386_read_exec && (flags & PF_R))
2709 prot |= VM_PROT_EXECUTE;
2716 __elfN(untrans_prot)(vm_prot_t prot)
2721 if (prot & VM_PROT_EXECUTE)
2723 if (prot & VM_PROT_READ)
2725 if (prot & VM_PROT_WRITE)
2731 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2733 u_long range, rbase, gap;
2736 if ((imgp->map_flags & MAP_ASLR) == 0)
2738 pct = __elfN(aslr_stack_gap);
2743 range = imgp->eff_stack_sz * pct / 100;
2744 arc4rand(&rbase, sizeof(rbase), 0);
2745 gap = rbase % range;
2746 gap &= ~(sizeof(u_long) - 1);