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__) || \
128 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
129 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
130 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
132 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
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");
138 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
140 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
145 val = __elfN(pie_base);
146 error = sysctl_handle_long(oidp, &val, 0, req);
147 if (error != 0 || req->newptr == NULL)
149 if ((val & PAGE_MASK) != 0)
151 __elfN(pie_base) = val;
154 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
155 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
156 sysctl_pie_base, "LU",
157 "PIE load base without randomization");
159 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
161 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
163 static int __elfN(aslr_enabled) = 0;
164 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
165 &__elfN(aslr_enabled), 0,
166 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
167 ": enable address map randomization");
169 static int __elfN(pie_aslr_enabled) = 0;
170 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
171 &__elfN(pie_aslr_enabled), 0,
172 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
173 ": enable address map randomization for PIE binaries");
175 static int __elfN(aslr_honor_sbrk) = 1;
176 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
177 &__elfN(aslr_honor_sbrk), 0,
178 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
180 static int __elfN(aslr_stack_gap) = 3;
181 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
182 &__elfN(aslr_stack_gap), 0,
183 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
184 ": maximum percentage of main stack to waste on a random gap");
186 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
188 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
190 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
192 Elf_Brandnote __elfN(freebsd_brandnote) = {
193 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
194 .hdr.n_descsz = sizeof(int32_t),
195 .hdr.n_type = NT_FREEBSD_ABI_TAG,
196 .vendor = FREEBSD_ABI_VENDOR,
197 .flags = BN_TRANSLATE_OSREL,
198 .trans_osrel = __elfN(freebsd_trans_osrel)
202 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
206 p = (uintptr_t)(note + 1);
207 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
208 *osrel = *(const int32_t *)(p);
213 static const char GNU_ABI_VENDOR[] = "GNU";
214 static int GNU_KFREEBSD_ABI_DESC = 3;
216 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
217 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
218 .hdr.n_descsz = 16, /* XXX at least 16 */
220 .vendor = GNU_ABI_VENDOR,
221 .flags = BN_TRANSLATE_OSREL,
222 .trans_osrel = kfreebsd_trans_osrel
226 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
228 const Elf32_Word *desc;
231 p = (uintptr_t)(note + 1);
232 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
234 desc = (const Elf32_Word *)p;
235 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
239 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
240 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
242 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
248 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
252 for (i = 0; i < MAX_BRANDS; i++) {
253 if (elf_brand_list[i] == NULL) {
254 elf_brand_list[i] = entry;
258 if (i == MAX_BRANDS) {
259 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
267 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
271 for (i = 0; i < MAX_BRANDS; i++) {
272 if (elf_brand_list[i] == entry) {
273 elf_brand_list[i] = NULL;
283 __elfN(brand_inuse)(Elf_Brandinfo *entry)
288 sx_slock(&allproc_lock);
289 FOREACH_PROC_IN_SYSTEM(p) {
290 if (p->p_sysent == entry->sysvec) {
295 sx_sunlock(&allproc_lock);
300 static Elf_Brandinfo *
301 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
302 int32_t *osrel, uint32_t *fctl0)
304 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
305 Elf_Brandinfo *bi, *bi_m;
307 int i, interp_name_len;
309 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
312 * We support four types of branding -- (1) the ELF EI_OSABI field
313 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
314 * branding w/in the ELF header, (3) path of the `interp_path'
315 * field, and (4) the ".note.ABI-tag" ELF section.
318 /* Look for an ".note.ABI-tag" ELF section */
320 for (i = 0; i < MAX_BRANDS; i++) {
321 bi = elf_brand_list[i];
324 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
326 if (hdr->e_machine == bi->machine && (bi->flags &
327 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
328 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
330 /* Give brand a chance to veto check_note's guess */
331 if (ret && bi->header_supported)
332 ret = bi->header_supported(imgp);
334 * If note checker claimed the binary, but the
335 * interpreter path in the image does not
336 * match default one for the brand, try to
337 * search for other brands with the same
338 * interpreter. Either there is better brand
339 * with the right interpreter, or, failing
340 * this, we return first brand which accepted
341 * our note and, optionally, header.
343 if (ret && bi_m == NULL && interp != NULL &&
344 (bi->interp_path == NULL ||
345 (strlen(bi->interp_path) + 1 != interp_name_len ||
346 strncmp(interp, bi->interp_path, interp_name_len)
358 /* If the executable has a brand, search for it in the brand list. */
359 for (i = 0; i < MAX_BRANDS; i++) {
360 bi = elf_brand_list[i];
361 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
362 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
364 if (hdr->e_machine == bi->machine &&
365 (hdr->e_ident[EI_OSABI] == bi->brand ||
366 (bi->compat_3_brand != NULL &&
367 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
368 bi->compat_3_brand) == 0))) {
369 /* Looks good, but give brand a chance to veto */
370 if (bi->header_supported == NULL ||
371 bi->header_supported(imgp)) {
373 * Again, prefer strictly matching
376 if (interp_name_len == 0 &&
377 bi->interp_path == NULL)
379 if (bi->interp_path != NULL &&
380 strlen(bi->interp_path) + 1 ==
381 interp_name_len && strncmp(interp,
382 bi->interp_path, interp_name_len) == 0)
392 /* No known brand, see if the header is recognized by any brand */
393 for (i = 0; i < MAX_BRANDS; i++) {
394 bi = elf_brand_list[i];
395 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
396 bi->header_supported == NULL)
398 if (hdr->e_machine == bi->machine) {
399 ret = bi->header_supported(imgp);
405 /* Lacking a known brand, search for a recognized interpreter. */
406 if (interp != NULL) {
407 for (i = 0; i < MAX_BRANDS; i++) {
408 bi = elf_brand_list[i];
409 if (bi == NULL || (bi->flags &
410 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
413 if (hdr->e_machine == bi->machine &&
414 bi->interp_path != NULL &&
415 /* ELF image p_filesz includes terminating zero */
416 strlen(bi->interp_path) + 1 == interp_name_len &&
417 strncmp(interp, bi->interp_path, interp_name_len)
418 == 0 && (bi->header_supported == NULL ||
419 bi->header_supported(imgp)))
424 /* Lacking a recognized interpreter, try the default brand */
425 for (i = 0; i < MAX_BRANDS; i++) {
426 bi = elf_brand_list[i];
427 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
428 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
430 if (hdr->e_machine == bi->machine &&
431 __elfN(fallback_brand) == bi->brand &&
432 (bi->header_supported == NULL ||
433 bi->header_supported(imgp)))
440 __elfN(check_header)(const Elf_Ehdr *hdr)
446 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
447 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
448 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
449 hdr->e_phentsize != sizeof(Elf_Phdr) ||
450 hdr->e_version != ELF_TARG_VER)
454 * Make sure we have at least one brand for this machine.
457 for (i = 0; i < MAX_BRANDS; i++) {
458 bi = elf_brand_list[i];
459 if (bi != NULL && bi->machine == hdr->e_machine)
469 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
470 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
477 * Create the page if it doesn't exist yet. Ignore errors.
479 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
480 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
483 * Find the page from the underlying object.
485 if (object != NULL) {
486 sf = vm_imgact_map_page(object, offset);
488 return (KERN_FAILURE);
489 off = offset - trunc_page(offset);
490 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
492 vm_imgact_unmap_page(sf);
494 return (KERN_FAILURE);
497 return (KERN_SUCCESS);
501 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
502 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
508 int error, locked, rv;
510 if (start != trunc_page(start)) {
511 rv = __elfN(map_partial)(map, object, offset, start,
512 round_page(start), prot);
513 if (rv != KERN_SUCCESS)
515 offset += round_page(start) - start;
516 start = round_page(start);
518 if (end != round_page(end)) {
519 rv = __elfN(map_partial)(map, object, offset +
520 trunc_page(end) - start, trunc_page(end), end, prot);
521 if (rv != KERN_SUCCESS)
523 end = trunc_page(end);
526 return (KERN_SUCCESS);
527 if ((offset & PAGE_MASK) != 0) {
529 * The mapping is not page aligned. This means that we have
532 rv = vm_map_fixed(map, NULL, 0, start, end - start,
533 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
534 if (rv != KERN_SUCCESS)
537 return (KERN_SUCCESS);
538 for (; start < end; start += sz) {
539 sf = vm_imgact_map_page(object, offset);
541 return (KERN_FAILURE);
542 off = offset - trunc_page(offset);
544 if (sz > PAGE_SIZE - off)
545 sz = PAGE_SIZE - off;
546 error = copyout((caddr_t)sf_buf_kva(sf) + off,
548 vm_imgact_unmap_page(sf);
550 return (KERN_FAILURE);
554 vm_object_reference(object);
555 rv = vm_map_fixed(map, object, offset, start, end - start,
556 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
557 (object != NULL ? MAP_VN_EXEC : 0));
558 if (rv != KERN_SUCCESS) {
559 locked = VOP_ISLOCKED(imgp->vp);
560 VOP_UNLOCK(imgp->vp);
561 vm_object_deallocate(object);
562 vn_lock(imgp->vp, locked | LK_RETRY);
564 } else if (object != NULL) {
565 MPASS(imgp->vp->v_object == object);
566 VOP_SET_TEXT_CHECKED(imgp->vp);
569 return (KERN_SUCCESS);
573 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
574 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
580 vm_offset_t map_addr;
583 vm_ooffset_t file_addr;
586 * It's necessary to fail if the filsz + offset taken from the
587 * header is greater than the actual file pager object's size.
588 * If we were to allow this, then the vm_map_find() below would
589 * walk right off the end of the file object and into the ether.
591 * While I'm here, might as well check for something else that
592 * is invalid: filsz cannot be greater than memsz.
594 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
596 uprintf("elf_load_section: truncated ELF file\n");
600 object = imgp->object;
601 map = &imgp->proc->p_vmspace->vm_map;
602 map_addr = trunc_page((vm_offset_t)vmaddr);
603 file_addr = trunc_page(offset);
606 * We have two choices. We can either clear the data in the last page
607 * of an oversized mapping, or we can start the anon mapping a page
608 * early and copy the initialized data into that first page. We
613 else if (memsz > filsz)
614 map_len = trunc_page(offset + filsz) - file_addr;
616 map_len = round_page(offset + filsz) - file_addr;
619 /* cow flags: don't dump readonly sections in core */
620 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
621 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
623 rv = __elfN(map_insert)(imgp, map, object, file_addr,
624 map_addr, map_addr + map_len, prot, cow);
625 if (rv != KERN_SUCCESS)
628 /* we can stop now if we've covered it all */
635 * We have to get the remaining bit of the file into the first part
636 * of the oversized map segment. This is normally because the .data
637 * segment in the file is extended to provide bss. It's a neat idea
638 * to try and save a page, but it's a pain in the behind to implement.
640 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
642 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
643 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
645 /* This had damn well better be true! */
647 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
648 map_addr + map_len, prot, 0);
649 if (rv != KERN_SUCCESS)
654 sf = vm_imgact_map_page(object, offset + filsz);
658 /* send the page fragment to user space */
659 error = copyout((caddr_t)sf_buf_kva(sf), (caddr_t)map_addr,
661 vm_imgact_unmap_page(sf);
667 * Remove write access to the page if it was only granted by map_insert
670 if ((prot & VM_PROT_WRITE) == 0)
671 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
672 map_len), prot, FALSE);
678 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
679 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
686 ASSERT_VOP_LOCKED(imgp->vp, __func__);
691 for (i = 0; i < hdr->e_phnum; i++) {
692 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
695 /* Loadable segment */
696 prot = __elfN(trans_prot)(phdr[i].p_flags);
697 error = __elfN(load_section)(imgp, phdr[i].p_offset,
698 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
699 phdr[i].p_memsz, phdr[i].p_filesz, prot);
704 * Establish the base address if this is the first segment.
707 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
712 if (base_addrp != NULL)
713 *base_addrp = base_addr;
719 * Load the file "file" into memory. It may be either a shared object
722 * The "addr" reference parameter is in/out. On entry, it specifies
723 * the address where a shared object should be loaded. If the file is
724 * an executable, this value is ignored. On exit, "addr" specifies
725 * where the file was actually loaded.
727 * The "entry" reference parameter is out only. On exit, it specifies
728 * the entry point for the loaded file.
731 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
737 struct image_params image_params;
739 const Elf_Ehdr *hdr = NULL;
740 const Elf_Phdr *phdr = NULL;
741 struct nameidata *nd;
743 struct image_params *imgp;
745 u_long base_addr = 0;
748 #ifdef CAPABILITY_MODE
750 * XXXJA: This check can go away once we are sufficiently confident
751 * that the checks in namei() are correct.
753 if (IN_CAPABILITY_MODE(curthread))
757 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
759 attr = &tempdata->attr;
760 imgp = &tempdata->image_params;
763 * Initialize part of the common data
768 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
769 UIO_SYSSPACE, file, curthread);
770 if ((error = namei(nd)) != 0) {
774 NDFREE(nd, NDF_ONLY_PNBUF);
775 imgp->vp = nd->ni_vp;
778 * Check permissions, modes, uid, etc on the file, and "open" it.
780 error = exec_check_permissions(imgp);
784 error = exec_map_first_page(imgp);
788 imgp->object = nd->ni_vp->v_object;
790 hdr = (const Elf_Ehdr *)imgp->image_header;
791 if ((error = __elfN(check_header)(hdr)) != 0)
793 if (hdr->e_type == ET_DYN)
795 else if (hdr->e_type == ET_EXEC)
802 /* Only support headers that fit within first page for now */
803 if ((hdr->e_phoff > PAGE_SIZE) ||
804 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
809 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
810 if (!aligned(phdr, Elf_Addr)) {
815 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
820 *entry = (unsigned long)hdr->e_entry + rbase;
824 exec_unmap_first_page(imgp);
828 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
831 free(tempdata, M_TEMP);
837 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
842 MPASS(vm_map_min(map) <= minv);
843 MPASS(maxv <= vm_map_max(map));
845 MPASS(minv + align < maxv);
846 arc4rand(&rbase, sizeof(rbase), 0);
847 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
848 res &= ~((u_long)align - 1);
852 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
853 res, minv, maxv, rbase));
855 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
856 res, maxv, minv, rbase));
861 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
862 const Elf_Phdr *phdr, u_long et_dyn_addr)
864 struct vmspace *vmspace;
866 u_long text_size, data_size, total_size, text_addr, data_addr;
867 u_long seg_size, seg_addr;
871 text_size = data_size = total_size = text_addr = data_addr = 0;
873 for (i = 0; i < hdr->e_phnum; i++) {
874 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
877 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
878 seg_size = round_page(phdr[i].p_memsz +
879 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
882 * Make the largest executable segment the official
883 * text segment and all others data.
885 * Note that obreak() assumes that data_addr + data_size == end
886 * of data load area, and the ELF file format expects segments
887 * to be sorted by address. If multiple data segments exist,
888 * the last one will be used.
891 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
892 text_size = seg_size;
893 text_addr = seg_addr;
895 data_size = seg_size;
896 data_addr = seg_addr;
898 total_size += seg_size;
901 if (data_addr == 0 && data_size == 0) {
902 data_addr = text_addr;
903 data_size = text_size;
907 * Check limits. It should be safe to check the
908 * limits after loading the segments since we do
909 * not actually fault in all the segments pages.
911 PROC_LOCK(imgp->proc);
912 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
913 err_str = "Data segment size exceeds process limit";
914 else if (text_size > maxtsiz)
915 err_str = "Text segment size exceeds system limit";
916 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
917 err_str = "Total segment size exceeds process limit";
918 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
919 err_str = "Data segment size exceeds resource limit";
920 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
921 err_str = "Total segment size exceeds resource limit";
922 PROC_UNLOCK(imgp->proc);
923 if (err_str != NULL) {
924 uprintf("%s\n", err_str);
928 vmspace = imgp->proc->p_vmspace;
929 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
930 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
931 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
932 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
938 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
939 char **interpp, bool *free_interpp)
943 int error, interp_name_len;
945 KASSERT(phdr->p_type == PT_INTERP,
946 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
947 ASSERT_VOP_LOCKED(imgp->vp, __func__);
951 /* Path to interpreter */
952 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
953 uprintf("Invalid PT_INTERP\n");
957 interp_name_len = phdr->p_filesz;
958 if (phdr->p_offset > PAGE_SIZE ||
959 interp_name_len > PAGE_SIZE - phdr->p_offset) {
961 * The vnode lock might be needed by the pagedaemon to
962 * clean pages owned by the vnode. Do not allow sleep
963 * waiting for memory with the vnode locked, instead
964 * try non-sleepable allocation first, and if it
965 * fails, go to the slow path were we drop the lock
966 * and do M_WAITOK. A text reference prevents
967 * modifications to the vnode content.
969 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
970 if (interp == NULL) {
971 VOP_UNLOCK(imgp->vp);
972 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
973 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
976 error = vn_rdwr(UIO_READ, imgp->vp, interp,
977 interp_name_len, phdr->p_offset,
978 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
981 free(interp, M_TEMP);
982 uprintf("i/o error PT_INTERP %d\n", error);
985 interp[interp_name_len] = '\0';
988 *free_interpp = true;
992 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
993 if (interp[interp_name_len - 1] != '\0') {
994 uprintf("Invalid PT_INTERP\n");
999 *free_interpp = false;
1004 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1005 const char *interp, u_long *addr, u_long *entry)
1010 if (brand_info->emul_path != NULL &&
1011 brand_info->emul_path[0] != '\0') {
1012 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1013 snprintf(path, MAXPATHLEN, "%s%s",
1014 brand_info->emul_path, interp);
1015 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1021 if (brand_info->interp_newpath != NULL &&
1022 (brand_info->interp_path == NULL ||
1023 strcmp(interp, brand_info->interp_path) == 0)) {
1024 error = __elfN(load_file)(imgp->proc,
1025 brand_info->interp_newpath, addr, entry);
1030 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1034 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1039 * Impossible et_dyn_addr initial value indicating that the real base
1040 * must be calculated later with some randomization applied.
1042 #define ET_DYN_ADDR_RAND 1
1045 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1048 const Elf_Ehdr *hdr;
1049 const Elf_Phdr *phdr;
1050 Elf_Auxargs *elf_auxargs;
1051 struct vmspace *vmspace;
1054 Elf_Brandinfo *brand_info;
1055 struct sysentvec *sv;
1056 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1057 u_long maxalign, mapsz, maxv, maxv1;
1063 hdr = (const Elf_Ehdr *)imgp->image_header;
1066 * Do we have a valid ELF header ?
1068 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1069 * if particular brand doesn't support it.
1071 if (__elfN(check_header)(hdr) != 0 ||
1072 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1076 * From here on down, we return an errno, not -1, as we've
1077 * detected an ELF file.
1080 if ((hdr->e_phoff > PAGE_SIZE) ||
1081 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
1082 /* Only support headers in first page for now */
1083 uprintf("Program headers not in the first page\n");
1086 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1087 if (!aligned(phdr, Elf_Addr)) {
1088 uprintf("Unaligned program headers\n");
1096 entry = proghdr = 0;
1098 free_interp = false;
1100 maxalign = PAGE_SIZE;
1103 for (i = 0; i < hdr->e_phnum; i++) {
1104 switch (phdr[i].p_type) {
1107 baddr = phdr[i].p_vaddr;
1108 if (phdr[i].p_align > maxalign)
1109 maxalign = phdr[i].p_align;
1110 mapsz += phdr[i].p_memsz;
1114 * If this segment contains the program headers,
1115 * remember their virtual address for the AT_PHDR
1116 * aux entry. Static binaries don't usually include
1119 if (phdr[i].p_offset == 0 &&
1120 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1121 <= phdr[i].p_filesz)
1122 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1125 /* Path to interpreter */
1126 if (interp != NULL) {
1127 uprintf("Multiple PT_INTERP headers\n");
1131 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1137 if (__elfN(nxstack))
1139 __elfN(trans_prot)(phdr[i].p_flags);
1140 imgp->stack_sz = phdr[i].p_memsz;
1142 case PT_PHDR: /* Program header table info */
1143 proghdr = phdr[i].p_vaddr;
1148 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1149 if (brand_info == NULL) {
1150 uprintf("ELF binary type \"%u\" not known.\n",
1151 hdr->e_ident[EI_OSABI]);
1155 sv = brand_info->sysvec;
1157 if (hdr->e_type == ET_DYN) {
1158 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1159 uprintf("Cannot execute shared object\n");
1164 * Honour the base load address from the dso if it is
1165 * non-zero for some reason.
1168 if ((sv->sv_flags & SV_ASLR) == 0 ||
1169 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1170 et_dyn_addr = __elfN(pie_base);
1171 else if ((__elfN(pie_aslr_enabled) &&
1172 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1173 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1174 et_dyn_addr = ET_DYN_ADDR_RAND;
1176 et_dyn_addr = __elfN(pie_base);
1181 * Avoid a possible deadlock if the current address space is destroyed
1182 * and that address space maps the locked vnode. In the common case,
1183 * the locked vnode's v_usecount is decremented but remains greater
1184 * than zero. Consequently, the vnode lock is not needed by vrele().
1185 * However, in cases where the vnode lock is external, such as nullfs,
1186 * v_usecount may become zero.
1188 * The VV_TEXT flag prevents modifications to the executable while
1189 * the vnode is unlocked.
1191 VOP_UNLOCK(imgp->vp);
1194 * Decide whether to enable randomization of user mappings.
1195 * First, reset user preferences for the setid binaries.
1196 * Then, account for the support of the randomization by the
1197 * ABI, by user preferences, and make special treatment for
1200 if (imgp->credential_setid) {
1201 PROC_LOCK(imgp->proc);
1202 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1203 PROC_UNLOCK(imgp->proc);
1205 if ((sv->sv_flags & SV_ASLR) == 0 ||
1206 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1207 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1208 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1209 ("et_dyn_addr == RAND and !ASLR"));
1210 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1211 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1212 et_dyn_addr == ET_DYN_ADDR_RAND) {
1213 imgp->map_flags |= MAP_ASLR;
1215 * If user does not care about sbrk, utilize the bss
1216 * grow region for mappings as well. We can select
1217 * the base for the image anywere and still not suffer
1218 * from the fragmentation.
1220 if (!__elfN(aslr_honor_sbrk) ||
1221 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1222 imgp->map_flags |= MAP_ASLR_IGNSTART;
1225 error = exec_new_vmspace(imgp, sv);
1226 vmspace = imgp->proc->p_vmspace;
1227 map = &vmspace->vm_map;
1229 imgp->proc->p_sysent = sv;
1231 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1232 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1233 KASSERT((map->flags & MAP_ASLR) != 0,
1234 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1235 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1236 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1237 /* reserve half of the address space to interpreter */
1238 maxv / 2, 1UL << flsl(maxalign));
1241 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1245 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1249 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1253 entry = (u_long)hdr->e_entry + et_dyn_addr;
1256 * We load the dynamic linker where a userland call
1257 * to mmap(0, ...) would put it. The rationale behind this
1258 * calculation is that it leaves room for the heap to grow to
1259 * its maximum allowed size.
1261 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1263 if ((map->flags & MAP_ASLR) != 0) {
1264 maxv1 = maxv / 2 + addr / 2;
1265 MPASS(maxv1 >= addr); /* No overflow */
1266 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1267 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1269 map->anon_loc = addr;
1272 imgp->entry_addr = entry;
1274 if (interp != NULL) {
1275 VOP_UNLOCK(imgp->vp);
1276 if ((map->flags & MAP_ASLR) != 0) {
1277 /* Assume that interpeter fits into 1/4 of AS */
1278 maxv1 = maxv / 2 + addr / 2;
1279 MPASS(maxv1 >= addr); /* No overflow */
1280 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1283 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1285 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1292 * Construct auxargs table (used by the copyout_auxargs routine)
1294 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1295 if (elf_auxargs == NULL) {
1296 VOP_UNLOCK(imgp->vp);
1297 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1298 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1300 elf_auxargs->execfd = -1;
1301 elf_auxargs->phdr = proghdr + et_dyn_addr;
1302 elf_auxargs->phent = hdr->e_phentsize;
1303 elf_auxargs->phnum = hdr->e_phnum;
1304 elf_auxargs->pagesz = PAGE_SIZE;
1305 elf_auxargs->base = addr;
1306 elf_auxargs->flags = 0;
1307 elf_auxargs->entry = entry;
1308 elf_auxargs->hdr_eflags = hdr->e_flags;
1310 imgp->auxargs = elf_auxargs;
1311 imgp->interpreted = 0;
1312 imgp->reloc_base = addr;
1313 imgp->proc->p_osrel = osrel;
1314 imgp->proc->p_fctl0 = fctl0;
1315 imgp->proc->p_elf_machine = hdr->e_machine;
1316 imgp->proc->p_elf_flags = hdr->e_flags;
1320 free(interp, M_TEMP);
1324 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1327 __elfN(freebsd_copyout_auxargs)(struct image_params *imgp, uintptr_t base)
1329 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1330 Elf_Auxinfo *argarray, *pos;
1333 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1336 if (args->execfd != -1)
1337 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1338 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1339 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1340 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1341 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1342 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1343 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1344 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1345 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1346 if (imgp->execpathp != 0)
1347 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1348 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1349 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1350 if (imgp->canary != 0) {
1351 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1352 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1354 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1355 if (imgp->pagesizes != 0) {
1356 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1357 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1359 if (imgp->sysent->sv_timekeep_base != 0) {
1360 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1361 imgp->sysent->sv_timekeep_base);
1363 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1364 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1365 imgp->sysent->sv_stackprot);
1366 if (imgp->sysent->sv_hwcap != NULL)
1367 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1368 if (imgp->sysent->sv_hwcap2 != NULL)
1369 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1370 AUXARGS_ENTRY(pos, AT_NULL, 0);
1372 free(imgp->auxargs, M_TEMP);
1373 imgp->auxargs = NULL;
1374 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1376 error = copyout(argarray, (void *)base, sizeof(*argarray) * AT_COUNT);
1377 free(argarray, M_TEMP);
1382 __elfN(freebsd_fixup)(uintptr_t *stack_base, struct image_params *imgp)
1386 base = (Elf_Addr *)*stack_base;
1388 if (suword(base, imgp->args->argc) == -1)
1390 *stack_base = (uintptr_t)base;
1395 * Code for generating ELF core dumps.
1398 typedef void (*segment_callback)(vm_map_entry_t, void *);
1400 /* Closure for cb_put_phdr(). */
1401 struct phdr_closure {
1402 Elf_Phdr *phdr; /* Program header to fill in */
1403 Elf_Off offset; /* Offset of segment in core file */
1406 /* Closure for cb_size_segment(). */
1407 struct sseg_closure {
1408 int count; /* Count of writable segments. */
1409 size_t size; /* Total size of all writable segments. */
1412 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1415 int type; /* Note type. */
1416 outfunc_t outfunc; /* Output function. */
1417 void *outarg; /* Argument for the output function. */
1418 size_t outsize; /* Output size. */
1419 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1422 TAILQ_HEAD(note_info_list, note_info);
1424 /* Coredump output parameters. */
1425 struct coredump_params {
1427 struct ucred *active_cred;
1428 struct ucred *file_cred;
1431 struct compressor *comp;
1434 extern int compress_user_cores;
1435 extern int compress_user_cores_level;
1437 static void cb_put_phdr(vm_map_entry_t, void *);
1438 static void cb_size_segment(vm_map_entry_t, void *);
1439 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1441 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1442 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1443 struct note_info_list *, size_t);
1444 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1446 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1447 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1448 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1449 static int sbuf_drain_core_output(void *, const char *, int);
1451 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1452 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1453 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1454 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1455 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1456 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1457 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1458 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1459 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1460 static void note_procstat_files(void *, struct sbuf *, size_t *);
1461 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1462 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1463 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1464 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1465 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1468 * Write out a core segment to the compression stream.
1471 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1477 chunk_len = MIN(len, CORE_BUF_SIZE);
1480 * We can get EFAULT error here.
1481 * In that case zero out the current chunk of the segment.
1483 error = copyin(base, buf, chunk_len);
1485 bzero(buf, chunk_len);
1486 error = compressor_write(p->comp, buf, chunk_len);
1496 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1499 return (core_write((struct coredump_params *)arg, base, len, offset,
1504 core_write(struct coredump_params *p, const void *base, size_t len,
1505 off_t offset, enum uio_seg seg)
1508 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1509 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1510 p->active_cred, p->file_cred, NULL, p->td));
1514 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1519 if (p->comp != NULL)
1520 return (compress_chunk(p, base, tmpbuf, len));
1523 * EFAULT is a non-fatal error that we can get, for example,
1524 * if the segment is backed by a file but extends beyond its
1527 error = core_write(p, base, len, offset, UIO_USERSPACE);
1528 if (error == EFAULT) {
1529 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1530 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1531 "for process %s\n", base, len, offset, curproc->p_comm);
1534 * Write a "real" zero byte at the end of the target region
1535 * in the case this is the last segment.
1536 * The intermediate space will be implicitly zero-filled.
1538 error = core_write(p, zero_region, 1, offset + len - 1,
1545 * Drain into a core file.
1548 sbuf_drain_core_output(void *arg, const char *data, int len)
1550 struct coredump_params *p;
1553 p = (struct coredump_params *)arg;
1556 * Some kern_proc out routines that print to this sbuf may
1557 * call us with the process lock held. Draining with the
1558 * non-sleepable lock held is unsafe. The lock is needed for
1559 * those routines when dumping a live process. In our case we
1560 * can safely release the lock before draining and acquire
1563 locked = PROC_LOCKED(p->td->td_proc);
1565 PROC_UNLOCK(p->td->td_proc);
1566 if (p->comp != NULL)
1567 error = compressor_write(p->comp, __DECONST(char *, data), len);
1569 error = core_write(p, __DECONST(void *, data), len, p->offset,
1572 PROC_LOCK(p->td->td_proc);
1580 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1582 struct ucred *cred = td->td_ucred;
1584 struct sseg_closure seginfo;
1585 struct note_info_list notelst;
1586 struct coredump_params params;
1587 struct note_info *ninfo;
1589 size_t hdrsize, notesz, coresize;
1593 TAILQ_INIT(¬elst);
1595 /* Size the program segments. */
1598 each_dumpable_segment(td, cb_size_segment, &seginfo);
1601 * Collect info about the core file header area.
1603 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1604 if (seginfo.count + 1 >= PN_XNUM)
1605 hdrsize += sizeof(Elf_Shdr);
1606 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1607 coresize = round_page(hdrsize + notesz) + seginfo.size;
1609 /* Set up core dump parameters. */
1611 params.active_cred = cred;
1612 params.file_cred = NOCRED;
1619 PROC_LOCK(td->td_proc);
1620 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1621 PROC_UNLOCK(td->td_proc);
1628 if (coresize >= limit) {
1633 /* Create a compression stream if necessary. */
1634 if (compress_user_cores != 0) {
1635 params.comp = compressor_init(core_compressed_write,
1636 compress_user_cores, CORE_BUF_SIZE,
1637 compress_user_cores_level, ¶ms);
1638 if (params.comp == NULL) {
1642 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1646 * Allocate memory for building the header, fill it up,
1647 * and write it out following the notes.
1649 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1650 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1653 /* Write the contents of all of the writable segments. */
1659 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1660 offset = round_page(hdrsize + notesz);
1661 for (i = 0; i < seginfo.count; i++) {
1662 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1663 php->p_filesz, offset, ¶ms, tmpbuf);
1666 offset += php->p_filesz;
1669 if (error == 0 && params.comp != NULL)
1670 error = compressor_flush(params.comp);
1674 "Failed to write core file for process %s (error %d)\n",
1675 curproc->p_comm, error);
1679 free(tmpbuf, M_TEMP);
1680 if (params.comp != NULL)
1681 compressor_fini(params.comp);
1682 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1683 TAILQ_REMOVE(¬elst, ninfo, link);
1684 free(ninfo, M_TEMP);
1693 * A callback for each_dumpable_segment() to write out the segment's
1694 * program header entry.
1697 cb_put_phdr(vm_map_entry_t entry, void *closure)
1699 struct phdr_closure *phc = (struct phdr_closure *)closure;
1700 Elf_Phdr *phdr = phc->phdr;
1702 phc->offset = round_page(phc->offset);
1704 phdr->p_type = PT_LOAD;
1705 phdr->p_offset = phc->offset;
1706 phdr->p_vaddr = entry->start;
1708 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1709 phdr->p_align = PAGE_SIZE;
1710 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1712 phc->offset += phdr->p_filesz;
1717 * A callback for each_dumpable_segment() to gather information about
1718 * the number of segments and their total size.
1721 cb_size_segment(vm_map_entry_t entry, void *closure)
1723 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1726 ssc->size += entry->end - entry->start;
1730 * For each writable segment in the process's memory map, call the given
1731 * function with a pointer to the map entry and some arbitrary
1732 * caller-supplied data.
1735 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1737 struct proc *p = td->td_proc;
1738 vm_map_t map = &p->p_vmspace->vm_map;
1739 vm_map_entry_t entry;
1740 vm_object_t backing_object, object;
1741 boolean_t ignore_entry;
1743 vm_map_lock_read(map);
1744 VM_MAP_ENTRY_FOREACH(entry, map) {
1746 * Don't dump inaccessible mappings, deal with legacy
1749 * Note that read-only segments related to the elf binary
1750 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1751 * need to arbitrarily ignore such segments.
1753 if (elf_legacy_coredump) {
1754 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1757 if ((entry->protection & VM_PROT_ALL) == 0)
1762 * Dont include memory segment in the coredump if
1763 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1764 * madvise(2). Do not dump submaps (i.e. parts of the
1767 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1770 if ((object = entry->object.vm_object) == NULL)
1773 /* Ignore memory-mapped devices and such things. */
1774 VM_OBJECT_RLOCK(object);
1775 while ((backing_object = object->backing_object) != NULL) {
1776 VM_OBJECT_RLOCK(backing_object);
1777 VM_OBJECT_RUNLOCK(object);
1778 object = backing_object;
1780 ignore_entry = object->type != OBJT_DEFAULT &&
1781 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1782 object->type != OBJT_PHYS;
1783 VM_OBJECT_RUNLOCK(object);
1787 (*func)(entry, closure);
1789 vm_map_unlock_read(map);
1793 * Write the core file header to the file, including padding up to
1794 * the page boundary.
1797 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1798 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1800 struct note_info *ninfo;
1804 /* Fill in the header. */
1805 bzero(hdr, hdrsize);
1806 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1808 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1809 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1810 sbuf_start_section(sb, NULL);
1811 sbuf_bcat(sb, hdr, hdrsize);
1812 TAILQ_FOREACH(ninfo, notelst, link)
1813 __elfN(putnote)(ninfo, sb);
1814 /* Align up to a page boundary for the program segments. */
1815 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1816 error = sbuf_finish(sb);
1823 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1833 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1836 * To have the debugger select the right thread (LWP) as the initial
1837 * thread, we dump the state of the thread passed to us in td first.
1838 * This is the thread that causes the core dump and thus likely to
1839 * be the right thread one wants to have selected in the debugger.
1842 while (thr != NULL) {
1843 size += register_note(list, NT_PRSTATUS,
1844 __elfN(note_prstatus), thr);
1845 size += register_note(list, NT_FPREGSET,
1846 __elfN(note_fpregset), thr);
1847 size += register_note(list, NT_THRMISC,
1848 __elfN(note_thrmisc), thr);
1849 size += register_note(list, NT_PTLWPINFO,
1850 __elfN(note_ptlwpinfo), thr);
1851 size += register_note(list, -1,
1852 __elfN(note_threadmd), thr);
1854 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1855 TAILQ_NEXT(thr, td_plist);
1857 thr = TAILQ_NEXT(thr, td_plist);
1860 size += register_note(list, NT_PROCSTAT_PROC,
1861 __elfN(note_procstat_proc), p);
1862 size += register_note(list, NT_PROCSTAT_FILES,
1863 note_procstat_files, p);
1864 size += register_note(list, NT_PROCSTAT_VMMAP,
1865 note_procstat_vmmap, p);
1866 size += register_note(list, NT_PROCSTAT_GROUPS,
1867 note_procstat_groups, p);
1868 size += register_note(list, NT_PROCSTAT_UMASK,
1869 note_procstat_umask, p);
1870 size += register_note(list, NT_PROCSTAT_RLIMIT,
1871 note_procstat_rlimit, p);
1872 size += register_note(list, NT_PROCSTAT_OSREL,
1873 note_procstat_osrel, p);
1874 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1875 __elfN(note_procstat_psstrings), p);
1876 size += register_note(list, NT_PROCSTAT_AUXV,
1877 __elfN(note_procstat_auxv), p);
1883 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1889 struct phdr_closure phc;
1891 ehdr = (Elf_Ehdr *)hdr;
1893 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1894 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1895 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1896 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1897 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1898 ehdr->e_ident[EI_DATA] = ELF_DATA;
1899 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1900 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1901 ehdr->e_ident[EI_ABIVERSION] = 0;
1902 ehdr->e_ident[EI_PAD] = 0;
1903 ehdr->e_type = ET_CORE;
1904 ehdr->e_machine = td->td_proc->p_elf_machine;
1905 ehdr->e_version = EV_CURRENT;
1907 ehdr->e_phoff = sizeof(Elf_Ehdr);
1908 ehdr->e_flags = td->td_proc->p_elf_flags;
1909 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1910 ehdr->e_phentsize = sizeof(Elf_Phdr);
1911 ehdr->e_shentsize = sizeof(Elf_Shdr);
1912 ehdr->e_shstrndx = SHN_UNDEF;
1913 if (numsegs + 1 < PN_XNUM) {
1914 ehdr->e_phnum = numsegs + 1;
1917 ehdr->e_phnum = PN_XNUM;
1920 ehdr->e_shoff = ehdr->e_phoff +
1921 (numsegs + 1) * ehdr->e_phentsize;
1922 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1923 ("e_shoff: %zu, hdrsize - shdr: %zu",
1924 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1926 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1927 memset(shdr, 0, sizeof(*shdr));
1929 * A special first section is used to hold large segment and
1930 * section counts. This was proposed by Sun Microsystems in
1931 * Solaris and has been adopted by Linux; the standard ELF
1932 * tools are already familiar with the technique.
1934 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1935 * (or 12-7 depending on the version of the document) for more
1938 shdr->sh_type = SHT_NULL;
1939 shdr->sh_size = ehdr->e_shnum;
1940 shdr->sh_link = ehdr->e_shstrndx;
1941 shdr->sh_info = numsegs + 1;
1945 * Fill in the program header entries.
1947 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1949 /* The note segement. */
1950 phdr->p_type = PT_NOTE;
1951 phdr->p_offset = hdrsize;
1954 phdr->p_filesz = notesz;
1956 phdr->p_flags = PF_R;
1957 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1960 /* All the writable segments from the program. */
1962 phc.offset = round_page(hdrsize + notesz);
1963 each_dumpable_segment(td, cb_put_phdr, &phc);
1967 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1969 struct note_info *ninfo;
1970 size_t size, notesize;
1973 out(arg, NULL, &size);
1974 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1976 ninfo->outfunc = out;
1977 ninfo->outarg = arg;
1978 ninfo->outsize = size;
1979 TAILQ_INSERT_TAIL(list, ninfo, link);
1984 notesize = sizeof(Elf_Note) + /* note header */
1985 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1987 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1993 append_note_data(const void *src, void *dst, size_t len)
1997 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1999 bcopy(src, dst, len);
2000 bzero((char *)dst + len, padded_len - len);
2002 return (padded_len);
2006 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2014 note = (Elf_Note *)buf;
2015 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2016 note->n_descsz = size;
2017 note->n_type = type;
2018 buf += sizeof(*note);
2019 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2020 sizeof(FREEBSD_ABI_VENDOR));
2021 append_note_data(src, buf, size);
2026 notesize = sizeof(Elf_Note) + /* note header */
2027 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2029 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2035 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2038 ssize_t old_len, sect_len;
2039 size_t new_len, descsz, i;
2041 if (ninfo->type == -1) {
2042 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2046 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2047 note.n_descsz = ninfo->outsize;
2048 note.n_type = ninfo->type;
2050 sbuf_bcat(sb, ¬e, sizeof(note));
2051 sbuf_start_section(sb, &old_len);
2052 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2053 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2054 if (note.n_descsz == 0)
2056 sbuf_start_section(sb, &old_len);
2057 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2058 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2062 new_len = (size_t)sect_len;
2063 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2064 if (new_len < descsz) {
2066 * It is expected that individual note emitters will correctly
2067 * predict their expected output size and fill up to that size
2068 * themselves, padding in a format-specific way if needed.
2069 * However, in case they don't, just do it here with zeros.
2071 for (i = 0; i < descsz - new_len; i++)
2073 } else if (new_len > descsz) {
2075 * We can't always truncate sb -- we may have drained some
2078 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2079 "read it (%zu > %zu). Since it is longer than "
2080 "expected, this coredump's notes are corrupt. THIS "
2081 "IS A BUG in the note_procstat routine for type %u.\n",
2082 __func__, (unsigned)note.n_type, new_len, descsz,
2083 (unsigned)note.n_type));
2088 * Miscellaneous note out functions.
2091 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2092 #include <compat/freebsd32/freebsd32.h>
2093 #include <compat/freebsd32/freebsd32_signal.h>
2095 typedef struct prstatus32 elf_prstatus_t;
2096 typedef struct prpsinfo32 elf_prpsinfo_t;
2097 typedef struct fpreg32 elf_prfpregset_t;
2098 typedef struct fpreg32 elf_fpregset_t;
2099 typedef struct reg32 elf_gregset_t;
2100 typedef struct thrmisc32 elf_thrmisc_t;
2101 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2102 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2103 typedef uint32_t elf_ps_strings_t;
2105 typedef prstatus_t elf_prstatus_t;
2106 typedef prpsinfo_t elf_prpsinfo_t;
2107 typedef prfpregset_t elf_prfpregset_t;
2108 typedef prfpregset_t elf_fpregset_t;
2109 typedef gregset_t elf_gregset_t;
2110 typedef thrmisc_t elf_thrmisc_t;
2111 #define ELF_KERN_PROC_MASK 0
2112 typedef struct kinfo_proc elf_kinfo_proc_t;
2113 typedef vm_offset_t elf_ps_strings_t;
2117 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2123 elf_prpsinfo_t *psinfo;
2126 p = (struct proc *)arg;
2128 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2129 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2130 psinfo->pr_version = PRPSINFO_VERSION;
2131 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2132 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2134 if (p->p_args != NULL) {
2135 len = sizeof(psinfo->pr_psargs) - 1;
2136 if (len > p->p_args->ar_length)
2137 len = p->p_args->ar_length;
2138 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2144 sbuf_new(&sbarg, psinfo->pr_psargs,
2145 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2146 error = proc_getargv(curthread, p, &sbarg);
2148 if (sbuf_finish(&sbarg) == 0)
2149 len = sbuf_len(&sbarg) - 1;
2151 len = sizeof(psinfo->pr_psargs) - 1;
2152 sbuf_delete(&sbarg);
2154 if (error || len == 0)
2155 strlcpy(psinfo->pr_psargs, p->p_comm,
2156 sizeof(psinfo->pr_psargs));
2158 KASSERT(len < sizeof(psinfo->pr_psargs),
2159 ("len is too long: %zu vs %zu", len,
2160 sizeof(psinfo->pr_psargs)));
2161 cp = psinfo->pr_psargs;
2164 cp = memchr(cp, '\0', end - cp);
2170 psinfo->pr_pid = p->p_pid;
2171 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2172 free(psinfo, M_TEMP);
2174 *sizep = sizeof(*psinfo);
2178 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2181 elf_prstatus_t *status;
2183 td = (struct thread *)arg;
2185 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2186 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2187 status->pr_version = PRSTATUS_VERSION;
2188 status->pr_statussz = sizeof(elf_prstatus_t);
2189 status->pr_gregsetsz = sizeof(elf_gregset_t);
2190 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2191 status->pr_osreldate = osreldate;
2192 status->pr_cursig = td->td_proc->p_sig;
2193 status->pr_pid = td->td_tid;
2194 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2195 fill_regs32(td, &status->pr_reg);
2197 fill_regs(td, &status->pr_reg);
2199 sbuf_bcat(sb, status, sizeof(*status));
2200 free(status, M_TEMP);
2202 *sizep = sizeof(*status);
2206 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2209 elf_prfpregset_t *fpregset;
2211 td = (struct thread *)arg;
2213 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2214 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2215 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2216 fill_fpregs32(td, fpregset);
2218 fill_fpregs(td, fpregset);
2220 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2221 free(fpregset, M_TEMP);
2223 *sizep = sizeof(*fpregset);
2227 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2230 elf_thrmisc_t thrmisc;
2232 td = (struct thread *)arg;
2234 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2235 bzero(&thrmisc, sizeof(thrmisc));
2236 strcpy(thrmisc.pr_tname, td->td_name);
2237 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2239 *sizep = sizeof(thrmisc);
2243 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2248 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2249 struct ptrace_lwpinfo32 pl;
2251 struct ptrace_lwpinfo pl;
2254 td = (struct thread *)arg;
2255 size = sizeof(structsize) + sizeof(pl);
2257 KASSERT(*sizep == size, ("invalid size"));
2258 structsize = sizeof(pl);
2259 sbuf_bcat(sb, &structsize, sizeof(structsize));
2260 bzero(&pl, sizeof(pl));
2261 pl.pl_lwpid = td->td_tid;
2262 pl.pl_event = PL_EVENT_NONE;
2263 pl.pl_sigmask = td->td_sigmask;
2264 pl.pl_siglist = td->td_siglist;
2265 if (td->td_si.si_signo != 0) {
2266 pl.pl_event = PL_EVENT_SIGNAL;
2267 pl.pl_flags |= PL_FLAG_SI;
2268 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2269 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2271 pl.pl_siginfo = td->td_si;
2274 strcpy(pl.pl_tdname, td->td_name);
2275 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2276 sbuf_bcat(sb, &pl, sizeof(pl));
2282 * Allow for MD specific notes, as well as any MD
2283 * specific preparations for writing MI notes.
2286 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2292 td = (struct thread *)arg;
2294 if (size != 0 && sb != NULL)
2295 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2299 __elfN(dump_thread)(td, buf, &size);
2300 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2301 if (size != 0 && sb != NULL)
2302 sbuf_bcat(sb, buf, size);
2307 #ifdef KINFO_PROC_SIZE
2308 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2312 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2318 p = (struct proc *)arg;
2319 size = sizeof(structsize) + p->p_numthreads *
2320 sizeof(elf_kinfo_proc_t);
2323 KASSERT(*sizep == size, ("invalid size"));
2324 structsize = sizeof(elf_kinfo_proc_t);
2325 sbuf_bcat(sb, &structsize, sizeof(structsize));
2327 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2332 #ifdef KINFO_FILE_SIZE
2333 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2337 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2340 size_t size, sect_sz, i;
2341 ssize_t start_len, sect_len;
2342 int structsize, filedesc_flags;
2344 if (coredump_pack_fileinfo)
2345 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2349 p = (struct proc *)arg;
2350 structsize = sizeof(struct kinfo_file);
2353 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2354 sbuf_set_drain(sb, sbuf_count_drain, &size);
2355 sbuf_bcat(sb, &structsize, sizeof(structsize));
2357 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2362 sbuf_start_section(sb, &start_len);
2364 sbuf_bcat(sb, &structsize, sizeof(structsize));
2366 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2369 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2374 KASSERT(sect_sz <= *sizep,
2375 ("kern_proc_filedesc_out did not respect maxlen; "
2376 "requested %zu, got %zu", *sizep - sizeof(structsize),
2377 sect_sz - sizeof(structsize)));
2379 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2384 #ifdef KINFO_VMENTRY_SIZE
2385 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2389 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2393 int structsize, vmmap_flags;
2395 if (coredump_pack_vmmapinfo)
2396 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2400 p = (struct proc *)arg;
2401 structsize = sizeof(struct kinfo_vmentry);
2404 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2405 sbuf_set_drain(sb, sbuf_count_drain, &size);
2406 sbuf_bcat(sb, &structsize, sizeof(structsize));
2408 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2413 sbuf_bcat(sb, &structsize, sizeof(structsize));
2415 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2421 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2427 p = (struct proc *)arg;
2428 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2430 KASSERT(*sizep == size, ("invalid size"));
2431 structsize = sizeof(gid_t);
2432 sbuf_bcat(sb, &structsize, sizeof(structsize));
2433 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2440 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2446 p = (struct proc *)arg;
2447 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2449 KASSERT(*sizep == size, ("invalid size"));
2450 structsize = sizeof(p->p_fd->fd_cmask);
2451 sbuf_bcat(sb, &structsize, sizeof(structsize));
2452 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2458 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2461 struct rlimit rlim[RLIM_NLIMITS];
2465 p = (struct proc *)arg;
2466 size = sizeof(structsize) + sizeof(rlim);
2468 KASSERT(*sizep == size, ("invalid size"));
2469 structsize = sizeof(rlim);
2470 sbuf_bcat(sb, &structsize, sizeof(structsize));
2472 for (i = 0; i < RLIM_NLIMITS; i++)
2473 lim_rlimit_proc(p, i, &rlim[i]);
2475 sbuf_bcat(sb, rlim, sizeof(rlim));
2481 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2487 p = (struct proc *)arg;
2488 size = sizeof(structsize) + sizeof(p->p_osrel);
2490 KASSERT(*sizep == size, ("invalid size"));
2491 structsize = sizeof(p->p_osrel);
2492 sbuf_bcat(sb, &structsize, sizeof(structsize));
2493 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2499 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2502 elf_ps_strings_t ps_strings;
2506 p = (struct proc *)arg;
2507 size = sizeof(structsize) + sizeof(ps_strings);
2509 KASSERT(*sizep == size, ("invalid size"));
2510 structsize = sizeof(ps_strings);
2511 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2512 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2514 ps_strings = p->p_sysent->sv_psstrings;
2516 sbuf_bcat(sb, &structsize, sizeof(structsize));
2517 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2523 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2529 p = (struct proc *)arg;
2532 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2533 sbuf_set_drain(sb, sbuf_count_drain, &size);
2534 sbuf_bcat(sb, &structsize, sizeof(structsize));
2536 proc_getauxv(curthread, p, sb);
2542 structsize = sizeof(Elf_Auxinfo);
2543 sbuf_bcat(sb, &structsize, sizeof(structsize));
2545 proc_getauxv(curthread, p, sb);
2551 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2552 const char *note_vendor, const Elf_Phdr *pnote,
2553 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2555 const Elf_Note *note, *note0, *note_end;
2556 const char *note_name;
2561 /* We need some limit, might as well use PAGE_SIZE. */
2562 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2564 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2565 if (pnote->p_offset > PAGE_SIZE ||
2566 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2567 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2569 VOP_UNLOCK(imgp->vp);
2570 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2571 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2573 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2574 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2575 curthread->td_ucred, NOCRED, NULL, curthread);
2577 uprintf("i/o error PT_NOTE\n");
2580 note = note0 = (const Elf_Note *)buf;
2581 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2583 note = note0 = (const Elf_Note *)(imgp->image_header +
2585 note_end = (const Elf_Note *)(imgp->image_header +
2586 pnote->p_offset + pnote->p_filesz);
2589 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2590 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2591 (const char *)note < sizeof(Elf_Note)) {
2594 if (note->n_namesz != checknote->n_namesz ||
2595 note->n_descsz != checknote->n_descsz ||
2596 note->n_type != checknote->n_type)
2598 note_name = (const char *)(note + 1);
2599 if (note_name + checknote->n_namesz >=
2600 (const char *)note_end || strncmp(note_vendor,
2601 note_name, checknote->n_namesz) != 0)
2604 if (cb(note, cb_arg, &res))
2607 note = (const Elf_Note *)((const char *)(note + 1) +
2608 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2609 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2618 struct brandnote_cb_arg {
2619 Elf_Brandnote *brandnote;
2624 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2626 struct brandnote_cb_arg *arg;
2631 * Fetch the osreldate for binary from the ELF OSABI-note if
2634 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2635 arg->brandnote->trans_osrel != NULL ?
2636 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2641 static Elf_Note fctl_note = {
2642 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2643 .n_descsz = sizeof(uint32_t),
2644 .n_type = NT_FREEBSD_FEATURE_CTL,
2647 struct fctl_cb_arg {
2652 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2654 struct fctl_cb_arg *arg;
2655 const Elf32_Word *desc;
2659 p = (uintptr_t)(note + 1);
2660 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2661 desc = (const Elf32_Word *)p;
2662 *arg->fctl0 = desc[0];
2667 * Try to find the appropriate ABI-note section for checknote, fetch
2668 * the osreldate and feature control flags for binary from the ELF
2669 * OSABI-note. Only the first page of the image is searched, the same
2673 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2674 int32_t *osrel, uint32_t *fctl0)
2676 const Elf_Phdr *phdr;
2677 const Elf_Ehdr *hdr;
2678 struct brandnote_cb_arg b_arg;
2679 struct fctl_cb_arg f_arg;
2682 hdr = (const Elf_Ehdr *)imgp->image_header;
2683 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2684 b_arg.brandnote = brandnote;
2685 b_arg.osrel = osrel;
2686 f_arg.fctl0 = fctl0;
2688 for (i = 0; i < hdr->e_phnum; i++) {
2689 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2690 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2692 for (j = 0; j < hdr->e_phnum; j++) {
2693 if (phdr[j].p_type == PT_NOTE &&
2694 __elfN(parse_notes)(imgp, &fctl_note,
2695 FREEBSD_ABI_VENDOR, &phdr[j],
2696 note_fctl_cb, &f_arg))
2707 * Tell kern_execve.c about it, with a little help from the linker.
2709 static struct execsw __elfN(execsw) = {
2710 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2711 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2713 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2716 __elfN(trans_prot)(Elf_Word flags)
2722 prot |= VM_PROT_EXECUTE;
2724 prot |= VM_PROT_WRITE;
2726 prot |= VM_PROT_READ;
2727 #if __ELF_WORD_SIZE == 32 && (defined(__amd64__) || defined(__i386__))
2728 if (i386_read_exec && (flags & PF_R))
2729 prot |= VM_PROT_EXECUTE;
2735 __elfN(untrans_prot)(vm_prot_t prot)
2740 if (prot & VM_PROT_EXECUTE)
2742 if (prot & VM_PROT_READ)
2744 if (prot & VM_PROT_WRITE)
2750 __elfN(stackgap)(struct image_params *imgp, uintptr_t *stack_base)
2752 uintptr_t range, rbase, gap;
2755 if ((imgp->map_flags & MAP_ASLR) == 0)
2757 pct = __elfN(aslr_stack_gap);
2762 range = imgp->eff_stack_sz * pct / 100;
2763 arc4rand(&rbase, sizeof(rbase), 0);
2764 gap = rbase % range;
2765 gap &= ~(sizeof(u_long) - 1);