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
133 #if defined(__amd64__)
134 int i386_read_exec = 0;
135 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
136 "enable execution from readable segments");
140 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
142 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
147 val = __elfN(pie_base);
148 error = sysctl_handle_long(oidp, &val, 0, req);
149 if (error != 0 || req->newptr == NULL)
151 if ((val & PAGE_MASK) != 0)
153 __elfN(pie_base) = val;
156 SYSCTL_PROC(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, pie_base,
157 CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
158 sysctl_pie_base, "LU",
159 "PIE load base without randomization");
161 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
163 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
165 static int __elfN(aslr_enabled) = 0;
166 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
167 &__elfN(aslr_enabled), 0,
168 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
169 ": enable address map randomization");
171 static int __elfN(pie_aslr_enabled) = 0;
172 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
173 &__elfN(pie_aslr_enabled), 0,
174 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
175 ": enable address map randomization for PIE binaries");
177 static int __elfN(aslr_honor_sbrk) = 1;
178 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
179 &__elfN(aslr_honor_sbrk), 0,
180 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
182 static int __elfN(aslr_stack_gap) = 3;
183 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, stack_gap, CTLFLAG_RW,
184 &__elfN(aslr_stack_gap), 0,
185 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
186 ": maximum percentage of main stack to waste on a random gap");
188 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
190 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
192 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
194 Elf_Brandnote __elfN(freebsd_brandnote) = {
195 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
196 .hdr.n_descsz = sizeof(int32_t),
197 .hdr.n_type = NT_FREEBSD_ABI_TAG,
198 .vendor = FREEBSD_ABI_VENDOR,
199 .flags = BN_TRANSLATE_OSREL,
200 .trans_osrel = __elfN(freebsd_trans_osrel)
204 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
208 p = (uintptr_t)(note + 1);
209 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
210 *osrel = *(const int32_t *)(p);
215 static const char GNU_ABI_VENDOR[] = "GNU";
216 static int GNU_KFREEBSD_ABI_DESC = 3;
218 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
219 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
220 .hdr.n_descsz = 16, /* XXX at least 16 */
222 .vendor = GNU_ABI_VENDOR,
223 .flags = BN_TRANSLATE_OSREL,
224 .trans_osrel = kfreebsd_trans_osrel
228 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
230 const Elf32_Word *desc;
233 p = (uintptr_t)(note + 1);
234 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
236 desc = (const Elf32_Word *)p;
237 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
241 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
242 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
244 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
250 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
254 for (i = 0; i < MAX_BRANDS; i++) {
255 if (elf_brand_list[i] == NULL) {
256 elf_brand_list[i] = entry;
260 if (i == MAX_BRANDS) {
261 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
269 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
273 for (i = 0; i < MAX_BRANDS; i++) {
274 if (elf_brand_list[i] == entry) {
275 elf_brand_list[i] = NULL;
285 __elfN(brand_inuse)(Elf_Brandinfo *entry)
290 sx_slock(&allproc_lock);
291 FOREACH_PROC_IN_SYSTEM(p) {
292 if (p->p_sysent == entry->sysvec) {
297 sx_sunlock(&allproc_lock);
302 static Elf_Brandinfo *
303 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
304 int32_t *osrel, uint32_t *fctl0)
306 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
307 Elf_Brandinfo *bi, *bi_m;
309 int i, interp_name_len;
311 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
314 * We support four types of branding -- (1) the ELF EI_OSABI field
315 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
316 * branding w/in the ELF header, (3) path of the `interp_path'
317 * field, and (4) the ".note.ABI-tag" ELF section.
320 /* Look for an ".note.ABI-tag" ELF section */
322 for (i = 0; i < MAX_BRANDS; i++) {
323 bi = elf_brand_list[i];
326 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
328 if (hdr->e_machine == bi->machine && (bi->flags &
329 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
330 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
332 /* Give brand a chance to veto check_note's guess */
333 if (ret && bi->header_supported)
334 ret = bi->header_supported(imgp);
336 * If note checker claimed the binary, but the
337 * interpreter path in the image does not
338 * match default one for the brand, try to
339 * search for other brands with the same
340 * interpreter. Either there is better brand
341 * with the right interpreter, or, failing
342 * this, we return first brand which accepted
343 * our note and, optionally, header.
345 if (ret && bi_m == NULL && interp != NULL &&
346 (bi->interp_path == NULL ||
347 (strlen(bi->interp_path) + 1 != interp_name_len ||
348 strncmp(interp, bi->interp_path, interp_name_len)
360 /* If the executable has a brand, search for it in the brand list. */
361 for (i = 0; i < MAX_BRANDS; i++) {
362 bi = elf_brand_list[i];
363 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
364 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
366 if (hdr->e_machine == bi->machine &&
367 (hdr->e_ident[EI_OSABI] == bi->brand ||
368 (bi->compat_3_brand != NULL &&
369 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
370 bi->compat_3_brand) == 0))) {
371 /* Looks good, but give brand a chance to veto */
372 if (bi->header_supported == NULL ||
373 bi->header_supported(imgp)) {
375 * Again, prefer strictly matching
378 if (interp_name_len == 0 &&
379 bi->interp_path == NULL)
381 if (bi->interp_path != NULL &&
382 strlen(bi->interp_path) + 1 ==
383 interp_name_len && strncmp(interp,
384 bi->interp_path, interp_name_len) == 0)
394 /* No known brand, see if the header is recognized by any brand */
395 for (i = 0; i < MAX_BRANDS; i++) {
396 bi = elf_brand_list[i];
397 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
398 bi->header_supported == NULL)
400 if (hdr->e_machine == bi->machine) {
401 ret = bi->header_supported(imgp);
407 /* Lacking a known brand, search for a recognized interpreter. */
408 if (interp != NULL) {
409 for (i = 0; i < MAX_BRANDS; i++) {
410 bi = elf_brand_list[i];
411 if (bi == NULL || (bi->flags &
412 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
415 if (hdr->e_machine == bi->machine &&
416 bi->interp_path != NULL &&
417 /* ELF image p_filesz includes terminating zero */
418 strlen(bi->interp_path) + 1 == interp_name_len &&
419 strncmp(interp, bi->interp_path, interp_name_len)
420 == 0 && (bi->header_supported == NULL ||
421 bi->header_supported(imgp)))
426 /* Lacking a recognized interpreter, try the default brand */
427 for (i = 0; i < MAX_BRANDS; i++) {
428 bi = elf_brand_list[i];
429 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
430 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
432 if (hdr->e_machine == bi->machine &&
433 __elfN(fallback_brand) == bi->brand &&
434 (bi->header_supported == NULL ||
435 bi->header_supported(imgp)))
442 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
444 return (hdr->e_phoff <= PAGE_SIZE &&
445 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
449 __elfN(check_header)(const Elf_Ehdr *hdr)
455 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
456 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
457 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
458 hdr->e_phentsize != sizeof(Elf_Phdr) ||
459 hdr->e_version != ELF_TARG_VER)
463 * Make sure we have at least one brand for this machine.
466 for (i = 0; i < MAX_BRANDS; i++) {
467 bi = elf_brand_list[i];
468 if (bi != NULL && bi->machine == hdr->e_machine)
478 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
479 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
486 * Create the page if it doesn't exist yet. Ignore errors.
488 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
489 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
492 * Find the page from the underlying object.
494 if (object != NULL) {
495 sf = vm_imgact_map_page(object, offset);
497 return (KERN_FAILURE);
498 off = offset - trunc_page(offset);
499 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
501 vm_imgact_unmap_page(sf);
503 return (KERN_FAILURE);
506 return (KERN_SUCCESS);
510 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
511 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
517 int error, locked, rv;
519 if (start != trunc_page(start)) {
520 rv = __elfN(map_partial)(map, object, offset, start,
521 round_page(start), prot);
522 if (rv != KERN_SUCCESS)
524 offset += round_page(start) - start;
525 start = round_page(start);
527 if (end != round_page(end)) {
528 rv = __elfN(map_partial)(map, object, offset +
529 trunc_page(end) - start, trunc_page(end), end, prot);
530 if (rv != KERN_SUCCESS)
532 end = trunc_page(end);
535 return (KERN_SUCCESS);
536 if ((offset & PAGE_MASK) != 0) {
538 * The mapping is not page aligned. This means that we have
541 rv = vm_map_fixed(map, NULL, 0, start, end - start,
542 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
543 if (rv != KERN_SUCCESS)
546 return (KERN_SUCCESS);
547 for (; start < end; start += sz) {
548 sf = vm_imgact_map_page(object, offset);
550 return (KERN_FAILURE);
551 off = offset - trunc_page(offset);
553 if (sz > PAGE_SIZE - off)
554 sz = PAGE_SIZE - off;
555 error = copyout((caddr_t)sf_buf_kva(sf) + off,
557 vm_imgact_unmap_page(sf);
559 return (KERN_FAILURE);
563 vm_object_reference(object);
564 rv = vm_map_fixed(map, object, offset, start, end - start,
565 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL |
566 (object != NULL ? MAP_VN_EXEC : 0));
567 if (rv != KERN_SUCCESS) {
568 locked = VOP_ISLOCKED(imgp->vp);
569 VOP_UNLOCK(imgp->vp, 0);
570 vm_object_deallocate(object);
571 vn_lock(imgp->vp, locked | LK_RETRY);
573 } else if (object != NULL) {
574 MPASS(imgp->vp->v_object == object);
575 VOP_SET_TEXT_CHECKED(imgp->vp);
578 return (KERN_SUCCESS);
582 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
583 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
589 vm_offset_t off, map_addr;
592 vm_ooffset_t file_addr;
595 * It's necessary to fail if the filsz + offset taken from the
596 * header is greater than the actual file pager object's size.
597 * If we were to allow this, then the vm_map_find() below would
598 * walk right off the end of the file object and into the ether.
600 * While I'm here, might as well check for something else that
601 * is invalid: filsz cannot be greater than memsz.
603 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
605 uprintf("elf_load_section: truncated ELF file\n");
609 object = imgp->object;
610 map = &imgp->proc->p_vmspace->vm_map;
611 map_addr = trunc_page((vm_offset_t)vmaddr);
612 file_addr = trunc_page(offset);
615 * We have two choices. We can either clear the data in the last page
616 * of an oversized mapping, or we can start the anon mapping a page
617 * early and copy the initialized data into that first page. We
622 else if (memsz > filsz)
623 map_len = trunc_page(offset + filsz) - file_addr;
625 map_len = round_page(offset + filsz) - file_addr;
628 /* cow flags: don't dump readonly sections in core */
629 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
630 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
632 rv = __elfN(map_insert)(imgp, map, object, file_addr,
633 map_addr, map_addr + map_len, prot, cow);
634 if (rv != KERN_SUCCESS)
637 /* we can stop now if we've covered it all */
644 * We have to get the remaining bit of the file into the first part
645 * of the oversized map segment. This is normally because the .data
646 * segment in the file is extended to provide bss. It's a neat idea
647 * to try and save a page, but it's a pain in the behind to implement.
649 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
651 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
652 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
654 /* This had damn well better be true! */
656 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
657 map_addr + map_len, prot, 0);
658 if (rv != KERN_SUCCESS)
663 sf = vm_imgact_map_page(object, offset + filsz);
667 /* send the page fragment to user space */
668 off = trunc_page(offset + filsz) - trunc_page(offset + filsz);
669 error = copyout((caddr_t)sf_buf_kva(sf) + off,
670 (caddr_t)map_addr, copy_len);
671 vm_imgact_unmap_page(sf);
677 * Remove write access to the page if it was only granted by map_insert
680 if ((prot & VM_PROT_WRITE) == 0)
681 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
682 map_len), prot, FALSE);
688 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
689 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
696 ASSERT_VOP_LOCKED(imgp->vp, __func__);
701 for (i = 0; i < hdr->e_phnum; i++) {
702 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
705 /* Loadable segment */
706 prot = __elfN(trans_prot)(phdr[i].p_flags);
707 error = __elfN(load_section)(imgp, phdr[i].p_offset,
708 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
709 phdr[i].p_memsz, phdr[i].p_filesz, prot);
714 * Establish the base address if this is the first segment.
717 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
722 if (base_addrp != NULL)
723 *base_addrp = base_addr;
729 * Load the file "file" into memory. It may be either a shared object
732 * The "addr" reference parameter is in/out. On entry, it specifies
733 * the address where a shared object should be loaded. If the file is
734 * an executable, this value is ignored. On exit, "addr" specifies
735 * where the file was actually loaded.
737 * The "entry" reference parameter is out only. On exit, it specifies
738 * the entry point for the loaded file.
741 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
747 struct image_params image_params;
749 const Elf_Ehdr *hdr = NULL;
750 const Elf_Phdr *phdr = NULL;
751 struct nameidata *nd;
753 struct image_params *imgp;
755 u_long base_addr = 0;
758 #ifdef CAPABILITY_MODE
760 * XXXJA: This check can go away once we are sufficiently confident
761 * that the checks in namei() are correct.
763 if (IN_CAPABILITY_MODE(curthread))
767 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
769 attr = &tempdata->attr;
770 imgp = &tempdata->image_params;
773 * Initialize part of the common data
778 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
779 UIO_SYSSPACE, file, curthread);
780 if ((error = namei(nd)) != 0) {
784 NDFREE(nd, NDF_ONLY_PNBUF);
785 imgp->vp = nd->ni_vp;
788 * Check permissions, modes, uid, etc on the file, and "open" it.
790 error = exec_check_permissions(imgp);
794 error = exec_map_first_page(imgp);
798 imgp->object = nd->ni_vp->v_object;
800 hdr = (const Elf_Ehdr *)imgp->image_header;
801 if ((error = __elfN(check_header)(hdr)) != 0)
803 if (hdr->e_type == ET_DYN)
805 else if (hdr->e_type == ET_EXEC)
812 /* Only support headers that fit within first page for now */
813 if (!__elfN(phdr_in_zero_page)(hdr)) {
818 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
819 if (!aligned(phdr, Elf_Addr)) {
824 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
829 *entry = (unsigned long)hdr->e_entry + rbase;
833 exec_unmap_first_page(imgp);
837 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
840 free(tempdata, M_TEMP);
846 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
851 MPASS(vm_map_min(map) <= minv);
852 MPASS(maxv <= vm_map_max(map));
854 MPASS(minv + align < maxv);
855 arc4rand(&rbase, sizeof(rbase), 0);
856 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
857 res &= ~((u_long)align - 1);
861 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
862 res, minv, maxv, rbase));
864 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
865 res, maxv, minv, rbase));
870 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
871 const Elf_Phdr *phdr, u_long et_dyn_addr)
873 struct vmspace *vmspace;
875 u_long text_size, data_size, total_size, text_addr, data_addr;
876 u_long seg_size, seg_addr;
880 text_size = data_size = total_size = text_addr = data_addr = 0;
882 for (i = 0; i < hdr->e_phnum; i++) {
883 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
886 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
887 seg_size = round_page(phdr[i].p_memsz +
888 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
891 * Make the largest executable segment the official
892 * text segment and all others data.
894 * Note that obreak() assumes that data_addr + data_size == end
895 * of data load area, and the ELF file format expects segments
896 * to be sorted by address. If multiple data segments exist,
897 * the last one will be used.
900 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
901 text_size = seg_size;
902 text_addr = seg_addr;
904 data_size = seg_size;
905 data_addr = seg_addr;
907 total_size += seg_size;
910 if (data_addr == 0 && data_size == 0) {
911 data_addr = text_addr;
912 data_size = text_size;
916 * Check limits. It should be safe to check the
917 * limits after loading the segments since we do
918 * not actually fault in all the segments pages.
920 PROC_LOCK(imgp->proc);
921 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
922 err_str = "Data segment size exceeds process limit";
923 else if (text_size > maxtsiz)
924 err_str = "Text segment size exceeds system limit";
925 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
926 err_str = "Total segment size exceeds process limit";
927 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
928 err_str = "Data segment size exceeds resource limit";
929 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
930 err_str = "Total segment size exceeds resource limit";
931 PROC_UNLOCK(imgp->proc);
932 if (err_str != NULL) {
933 uprintf("%s\n", err_str);
937 vmspace = imgp->proc->p_vmspace;
938 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
939 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
940 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
941 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
947 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
948 char **interpp, bool *free_interpp)
952 int error, interp_name_len;
954 KASSERT(phdr->p_type == PT_INTERP,
955 ("%s: p_type %u != PT_INTERP", __func__, phdr->p_type));
956 ASSERT_VOP_LOCKED(imgp->vp, __func__);
960 /* Path to interpreter */
961 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
962 uprintf("Invalid PT_INTERP\n");
966 interp_name_len = phdr->p_filesz;
967 if (phdr->p_offset > PAGE_SIZE ||
968 interp_name_len > PAGE_SIZE - phdr->p_offset) {
970 * The vnode lock might be needed by the pagedaemon to
971 * clean pages owned by the vnode. Do not allow sleep
972 * waiting for memory with the vnode locked, instead
973 * try non-sleepable allocation first, and if it
974 * fails, go to the slow path were we drop the lock
975 * and do M_WAITOK. A text reference prevents
976 * modifications to the vnode content.
978 interp = malloc(interp_name_len + 1, M_TEMP, M_NOWAIT);
979 if (interp == NULL) {
980 VOP_UNLOCK(imgp->vp, 0);
981 interp = malloc(interp_name_len + 1, M_TEMP, M_WAITOK);
982 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
985 error = vn_rdwr(UIO_READ, imgp->vp, interp,
986 interp_name_len, phdr->p_offset,
987 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
990 free(interp, M_TEMP);
991 uprintf("i/o error PT_INTERP %d\n", error);
994 interp[interp_name_len] = '\0';
997 *free_interpp = true;
1001 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1002 if (interp[interp_name_len - 1] != '\0') {
1003 uprintf("Invalid PT_INTERP\n");
1008 *free_interpp = false;
1013 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1014 const char *interp, u_long *addr, u_long *entry)
1019 if (brand_info->emul_path != NULL &&
1020 brand_info->emul_path[0] != '\0') {
1021 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1022 snprintf(path, MAXPATHLEN, "%s%s",
1023 brand_info->emul_path, interp);
1024 error = __elfN(load_file)(imgp->proc, path, addr, entry);
1030 if (brand_info->interp_newpath != NULL &&
1031 (brand_info->interp_path == NULL ||
1032 strcmp(interp, brand_info->interp_path) == 0)) {
1033 error = __elfN(load_file)(imgp->proc,
1034 brand_info->interp_newpath, addr, entry);
1039 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1043 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1048 * Impossible et_dyn_addr initial value indicating that the real base
1049 * must be calculated later with some randomization applied.
1051 #define ET_DYN_ADDR_RAND 1
1054 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1057 const Elf_Ehdr *hdr;
1058 const Elf_Phdr *phdr;
1059 Elf_Auxargs *elf_auxargs;
1060 struct vmspace *vmspace;
1063 Elf_Brandinfo *brand_info;
1064 struct sysentvec *sv;
1065 u_long addr, baddr, et_dyn_addr, entry, proghdr;
1066 u_long maxalign, mapsz, maxv, maxv1;
1072 hdr = (const Elf_Ehdr *)imgp->image_header;
1075 * Do we have a valid ELF header ?
1077 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1078 * if particular brand doesn't support it.
1080 if (__elfN(check_header)(hdr) != 0 ||
1081 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1085 * From here on down, we return an errno, not -1, as we've
1086 * detected an ELF file.
1089 if (!__elfN(phdr_in_zero_page)(hdr)) {
1090 uprintf("Program headers not in the first page\n");
1093 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1094 if (!aligned(phdr, Elf_Addr)) {
1095 uprintf("Unaligned program headers\n");
1103 entry = proghdr = 0;
1105 free_interp = false;
1107 maxalign = PAGE_SIZE;
1110 for (i = 0; i < hdr->e_phnum; i++) {
1111 switch (phdr[i].p_type) {
1114 baddr = phdr[i].p_vaddr;
1115 if (phdr[i].p_align > maxalign)
1116 maxalign = phdr[i].p_align;
1117 mapsz += phdr[i].p_memsz;
1121 * If this segment contains the program headers,
1122 * remember their virtual address for the AT_PHDR
1123 * aux entry. Static binaries don't usually include
1126 if (phdr[i].p_offset == 0 &&
1127 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1128 <= phdr[i].p_filesz)
1129 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
1132 /* Path to interpreter */
1133 if (interp != NULL) {
1134 uprintf("Multiple PT_INTERP headers\n");
1138 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1144 if (__elfN(nxstack))
1146 __elfN(trans_prot)(phdr[i].p_flags);
1147 imgp->stack_sz = phdr[i].p_memsz;
1149 case PT_PHDR: /* Program header table info */
1150 proghdr = phdr[i].p_vaddr;
1155 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1156 if (brand_info == NULL) {
1157 uprintf("ELF binary type \"%u\" not known.\n",
1158 hdr->e_ident[EI_OSABI]);
1162 sv = brand_info->sysvec;
1164 if (hdr->e_type == ET_DYN) {
1165 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1166 uprintf("Cannot execute shared object\n");
1171 * Honour the base load address from the dso if it is
1172 * non-zero for some reason.
1175 if ((sv->sv_flags & SV_ASLR) == 0 ||
1176 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1177 et_dyn_addr = __elfN(pie_base);
1178 else if ((__elfN(pie_aslr_enabled) &&
1179 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1180 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1181 et_dyn_addr = ET_DYN_ADDR_RAND;
1183 et_dyn_addr = __elfN(pie_base);
1188 * Avoid a possible deadlock if the current address space is destroyed
1189 * and that address space maps the locked vnode. In the common case,
1190 * the locked vnode's v_usecount is decremented but remains greater
1191 * than zero. Consequently, the vnode lock is not needed by vrele().
1192 * However, in cases where the vnode lock is external, such as nullfs,
1193 * v_usecount may become zero.
1195 * The VV_TEXT flag prevents modifications to the executable while
1196 * the vnode is unlocked.
1198 VOP_UNLOCK(imgp->vp, 0);
1201 * Decide whether to enable randomization of user mappings.
1202 * First, reset user preferences for the setid binaries.
1203 * Then, account for the support of the randomization by the
1204 * ABI, by user preferences, and make special treatment for
1207 if (imgp->credential_setid) {
1208 PROC_LOCK(imgp->proc);
1209 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1210 PROC_UNLOCK(imgp->proc);
1212 if ((sv->sv_flags & SV_ASLR) == 0 ||
1213 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1214 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1215 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1216 ("et_dyn_addr == RAND and !ASLR"));
1217 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1218 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1219 et_dyn_addr == ET_DYN_ADDR_RAND) {
1220 imgp->map_flags |= MAP_ASLR;
1222 * If user does not care about sbrk, utilize the bss
1223 * grow region for mappings as well. We can select
1224 * the base for the image anywere and still not suffer
1225 * from the fragmentation.
1227 if (!__elfN(aslr_honor_sbrk) ||
1228 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1229 imgp->map_flags |= MAP_ASLR_IGNSTART;
1232 error = exec_new_vmspace(imgp, sv);
1233 vmspace = imgp->proc->p_vmspace;
1234 map = &vmspace->vm_map;
1236 imgp->proc->p_sysent = sv;
1238 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1239 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1240 KASSERT((map->flags & MAP_ASLR) != 0,
1241 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1242 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1243 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1244 /* reserve half of the address space to interpreter */
1245 maxv / 2, 1UL << flsl(maxalign));
1248 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1252 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1256 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1260 entry = (u_long)hdr->e_entry + et_dyn_addr;
1263 * We load the dynamic linker where a userland call
1264 * to mmap(0, ...) would put it. The rationale behind this
1265 * calculation is that it leaves room for the heap to grow to
1266 * its maximum allowed size.
1268 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1270 if ((map->flags & MAP_ASLR) != 0) {
1271 maxv1 = maxv / 2 + addr / 2;
1272 MPASS(maxv1 >= addr); /* No overflow */
1273 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1274 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1275 pagesizes[1] : pagesizes[0]);
1277 map->anon_loc = addr;
1280 imgp->entry_addr = entry;
1282 if (interp != NULL) {
1283 VOP_UNLOCK(imgp->vp, 0);
1284 if ((map->flags & MAP_ASLR) != 0) {
1285 /* Assume that interpreter fits into 1/4 of AS */
1286 maxv1 = maxv / 2 + addr / 2;
1287 MPASS(maxv1 >= addr); /* No overflow */
1288 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1291 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1293 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1300 * Construct auxargs table (used by the fixup routine)
1302 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1303 if (elf_auxargs == NULL) {
1304 VOP_UNLOCK(imgp->vp, 0);
1305 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1306 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1308 elf_auxargs->execfd = -1;
1309 elf_auxargs->phdr = proghdr + et_dyn_addr;
1310 elf_auxargs->phent = hdr->e_phentsize;
1311 elf_auxargs->phnum = hdr->e_phnum;
1312 elf_auxargs->pagesz = PAGE_SIZE;
1313 elf_auxargs->base = addr;
1314 elf_auxargs->flags = 0;
1315 elf_auxargs->entry = entry;
1316 elf_auxargs->hdr_eflags = hdr->e_flags;
1318 imgp->auxargs = elf_auxargs;
1319 imgp->interpreted = 0;
1320 imgp->reloc_base = addr;
1321 imgp->proc->p_osrel = osrel;
1322 imgp->proc->p_fctl0 = fctl0;
1323 imgp->proc->p_elf_machine = hdr->e_machine;
1324 imgp->proc->p_elf_flags = hdr->e_flags;
1328 free(interp, M_TEMP);
1332 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1335 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1337 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1338 Elf_Auxinfo *argarray, *pos;
1339 Elf_Addr *base, *auxbase;
1342 base = (Elf_Addr *)*stack_base;
1343 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1344 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1347 if (args->execfd != -1)
1348 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1349 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1350 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1351 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1352 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1353 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1354 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1355 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1356 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1357 if (imgp->execpathp != 0)
1358 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1359 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1360 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1361 if (imgp->canary != 0) {
1362 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1363 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1365 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1366 if (imgp->pagesizes != 0) {
1367 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1368 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1370 if (imgp->sysent->sv_timekeep_base != 0) {
1371 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1372 imgp->sysent->sv_timekeep_base);
1374 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1375 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1376 imgp->sysent->sv_stackprot);
1377 if (imgp->sysent->sv_hwcap != NULL)
1378 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1379 if (imgp->sysent->sv_hwcap2 != NULL)
1380 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1381 AUXARGS_ENTRY(pos, AT_NULL, 0);
1383 free(imgp->auxargs, M_TEMP);
1384 imgp->auxargs = NULL;
1385 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1387 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1388 free(argarray, M_TEMP);
1393 if (suword(base, imgp->args->argc) == -1)
1395 *stack_base = (register_t *)base;
1400 * Code for generating ELF core dumps.
1403 typedef void (*segment_callback)(vm_map_entry_t, void *);
1405 /* Closure for cb_put_phdr(). */
1406 struct phdr_closure {
1407 Elf_Phdr *phdr; /* Program header to fill in */
1408 Elf_Off offset; /* Offset of segment in core file */
1411 /* Closure for cb_size_segment(). */
1412 struct sseg_closure {
1413 int count; /* Count of writable segments. */
1414 size_t size; /* Total size of all writable segments. */
1417 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1420 int type; /* Note type. */
1421 outfunc_t outfunc; /* Output function. */
1422 void *outarg; /* Argument for the output function. */
1423 size_t outsize; /* Output size. */
1424 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1427 TAILQ_HEAD(note_info_list, note_info);
1429 /* Coredump output parameters. */
1430 struct coredump_params {
1432 struct ucred *active_cred;
1433 struct ucred *file_cred;
1436 struct compressor *comp;
1439 extern int compress_user_cores;
1440 extern int compress_user_cores_level;
1442 static void cb_put_phdr(vm_map_entry_t, void *);
1443 static void cb_size_segment(vm_map_entry_t, void *);
1444 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1446 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1447 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1448 struct note_info_list *, size_t);
1449 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1451 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1452 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1453 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1454 static int sbuf_drain_core_output(void *, const char *, int);
1456 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1457 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1458 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1459 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1460 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1461 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1462 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1463 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1465 static void note_procstat_files(void *, struct sbuf *, size_t *);
1466 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1467 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1468 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1469 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1470 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1473 * Write out a core segment to the compression stream.
1476 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1482 chunk_len = MIN(len, CORE_BUF_SIZE);
1485 * We can get EFAULT error here.
1486 * In that case zero out the current chunk of the segment.
1488 error = copyin(base, buf, chunk_len);
1490 bzero(buf, chunk_len);
1491 error = compressor_write(p->comp, buf, chunk_len);
1501 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1504 return (core_write((struct coredump_params *)arg, base, len, offset,
1509 core_write(struct coredump_params *p, const void *base, size_t len,
1510 off_t offset, enum uio_seg seg)
1513 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1514 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1515 p->active_cred, p->file_cred, NULL, p->td));
1519 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1524 if (p->comp != NULL)
1525 return (compress_chunk(p, base, tmpbuf, len));
1528 * EFAULT is a non-fatal error that we can get, for example,
1529 * if the segment is backed by a file but extends beyond its
1532 error = core_write(p, base, len, offset, UIO_USERSPACE);
1533 if (error == EFAULT) {
1534 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1535 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1536 "for process %s\n", base, len, offset, curproc->p_comm);
1539 * Write a "real" zero byte at the end of the target region
1540 * in the case this is the last segment.
1541 * The intermediate space will be implicitly zero-filled.
1543 error = core_write(p, zero_region, 1, offset + len - 1,
1550 * Drain into a core file.
1553 sbuf_drain_core_output(void *arg, const char *data, int len)
1555 struct coredump_params *p;
1558 p = (struct coredump_params *)arg;
1561 * Some kern_proc out routines that print to this sbuf may
1562 * call us with the process lock held. Draining with the
1563 * non-sleepable lock held is unsafe. The lock is needed for
1564 * those routines when dumping a live process. In our case we
1565 * can safely release the lock before draining and acquire
1568 locked = PROC_LOCKED(p->td->td_proc);
1570 PROC_UNLOCK(p->td->td_proc);
1571 if (p->comp != NULL)
1572 error = compressor_write(p->comp, __DECONST(char *, data), len);
1574 error = core_write(p, __DECONST(void *, data), len, p->offset,
1577 PROC_LOCK(p->td->td_proc);
1585 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1587 struct ucred *cred = td->td_ucred;
1589 struct sseg_closure seginfo;
1590 struct note_info_list notelst;
1591 struct coredump_params params;
1592 struct note_info *ninfo;
1594 size_t hdrsize, notesz, coresize;
1598 TAILQ_INIT(¬elst);
1600 /* Size the program segments. */
1603 each_dumpable_segment(td, cb_size_segment, &seginfo);
1606 * Collect info about the core file header area.
1608 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1609 if (seginfo.count + 1 >= PN_XNUM)
1610 hdrsize += sizeof(Elf_Shdr);
1611 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1612 coresize = round_page(hdrsize + notesz) + seginfo.size;
1614 /* Set up core dump parameters. */
1616 params.active_cred = cred;
1617 params.file_cred = NOCRED;
1624 PROC_LOCK(td->td_proc);
1625 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1626 PROC_UNLOCK(td->td_proc);
1633 if (coresize >= limit) {
1638 /* Create a compression stream if necessary. */
1639 if (compress_user_cores != 0) {
1640 params.comp = compressor_init(core_compressed_write,
1641 compress_user_cores, CORE_BUF_SIZE,
1642 compress_user_cores_level, ¶ms);
1643 if (params.comp == NULL) {
1647 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1651 * Allocate memory for building the header, fill it up,
1652 * and write it out following the notes.
1654 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1655 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1658 /* Write the contents of all of the writable segments. */
1664 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1665 offset = round_page(hdrsize + notesz);
1666 for (i = 0; i < seginfo.count; i++) {
1667 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1668 php->p_filesz, offset, ¶ms, tmpbuf);
1671 offset += php->p_filesz;
1674 if (error == 0 && params.comp != NULL)
1675 error = compressor_flush(params.comp);
1679 "Failed to write core file for process %s (error %d)\n",
1680 curproc->p_comm, error);
1684 free(tmpbuf, M_TEMP);
1685 if (params.comp != NULL)
1686 compressor_fini(params.comp);
1687 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1688 TAILQ_REMOVE(¬elst, ninfo, link);
1689 free(ninfo, M_TEMP);
1698 * A callback for each_dumpable_segment() to write out the segment's
1699 * program header entry.
1702 cb_put_phdr(vm_map_entry_t entry, void *closure)
1704 struct phdr_closure *phc = (struct phdr_closure *)closure;
1705 Elf_Phdr *phdr = phc->phdr;
1707 phc->offset = round_page(phc->offset);
1709 phdr->p_type = PT_LOAD;
1710 phdr->p_offset = phc->offset;
1711 phdr->p_vaddr = entry->start;
1713 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1714 phdr->p_align = PAGE_SIZE;
1715 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1717 phc->offset += phdr->p_filesz;
1722 * A callback for each_dumpable_segment() to gather information about
1723 * the number of segments and their total size.
1726 cb_size_segment(vm_map_entry_t entry, void *closure)
1728 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1731 ssc->size += entry->end - entry->start;
1735 * For each writable segment in the process's memory map, call the given
1736 * function with a pointer to the map entry and some arbitrary
1737 * caller-supplied data.
1740 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1742 struct proc *p = td->td_proc;
1743 vm_map_t map = &p->p_vmspace->vm_map;
1744 vm_map_entry_t entry;
1745 vm_object_t backing_object, object;
1748 vm_map_lock_read(map);
1749 for (entry = map->header.next; entry != &map->header;
1750 entry = entry->next) {
1752 * Don't dump inaccessible mappings, deal with legacy
1755 * Note that read-only segments related to the elf binary
1756 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1757 * need to arbitrarily ignore such segments.
1759 if (elf_legacy_coredump) {
1760 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1763 if ((entry->protection & VM_PROT_ALL) == 0)
1768 * Dont include memory segment in the coredump if
1769 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1770 * madvise(2). Do not dump submaps (i.e. parts of the
1773 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1776 if ((object = entry->object.vm_object) == NULL)
1779 /* Ignore memory-mapped devices and such things. */
1780 VM_OBJECT_RLOCK(object);
1781 while ((backing_object = object->backing_object) != NULL) {
1782 VM_OBJECT_RLOCK(backing_object);
1783 VM_OBJECT_RUNLOCK(object);
1784 object = backing_object;
1786 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1787 VM_OBJECT_RUNLOCK(object);
1791 (*func)(entry, closure);
1793 vm_map_unlock_read(map);
1797 * Write the core file header to the file, including padding up to
1798 * the page boundary.
1801 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1802 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1804 struct note_info *ninfo;
1808 /* Fill in the header. */
1809 bzero(hdr, hdrsize);
1810 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1812 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1813 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1814 sbuf_start_section(sb, NULL);
1815 sbuf_bcat(sb, hdr, hdrsize);
1816 TAILQ_FOREACH(ninfo, notelst, link)
1817 __elfN(putnote)(ninfo, sb);
1818 /* Align up to a page boundary for the program segments. */
1819 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1820 error = sbuf_finish(sb);
1827 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1837 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1840 * To have the debugger select the right thread (LWP) as the initial
1841 * thread, we dump the state of the thread passed to us in td first.
1842 * This is the thread that causes the core dump and thus likely to
1843 * be the right thread one wants to have selected in the debugger.
1846 while (thr != NULL) {
1847 size += register_note(list, NT_PRSTATUS,
1848 __elfN(note_prstatus), thr);
1849 size += register_note(list, NT_FPREGSET,
1850 __elfN(note_fpregset), thr);
1851 size += register_note(list, NT_THRMISC,
1852 __elfN(note_thrmisc), thr);
1853 size += register_note(list, NT_PTLWPINFO,
1854 __elfN(note_ptlwpinfo), thr);
1855 size += register_note(list, -1,
1856 __elfN(note_threadmd), thr);
1858 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1859 TAILQ_NEXT(thr, td_plist);
1861 thr = TAILQ_NEXT(thr, td_plist);
1864 size += register_note(list, NT_PROCSTAT_PROC,
1865 __elfN(note_procstat_proc), p);
1866 size += register_note(list, NT_PROCSTAT_FILES,
1867 note_procstat_files, p);
1868 size += register_note(list, NT_PROCSTAT_VMMAP,
1869 note_procstat_vmmap, p);
1870 size += register_note(list, NT_PROCSTAT_GROUPS,
1871 note_procstat_groups, p);
1872 size += register_note(list, NT_PROCSTAT_UMASK,
1873 note_procstat_umask, p);
1874 size += register_note(list, NT_PROCSTAT_RLIMIT,
1875 note_procstat_rlimit, p);
1876 size += register_note(list, NT_PROCSTAT_OSREL,
1877 note_procstat_osrel, p);
1878 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1879 __elfN(note_procstat_psstrings), p);
1880 size += register_note(list, NT_PROCSTAT_AUXV,
1881 __elfN(note_procstat_auxv), p);
1887 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1893 struct phdr_closure phc;
1895 ehdr = (Elf_Ehdr *)hdr;
1897 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1898 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1899 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1900 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1901 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1902 ehdr->e_ident[EI_DATA] = ELF_DATA;
1903 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1904 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1905 ehdr->e_ident[EI_ABIVERSION] = 0;
1906 ehdr->e_ident[EI_PAD] = 0;
1907 ehdr->e_type = ET_CORE;
1908 ehdr->e_machine = td->td_proc->p_elf_machine;
1909 ehdr->e_version = EV_CURRENT;
1911 ehdr->e_phoff = sizeof(Elf_Ehdr);
1912 ehdr->e_flags = td->td_proc->p_elf_flags;
1913 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1914 ehdr->e_phentsize = sizeof(Elf_Phdr);
1915 ehdr->e_shentsize = sizeof(Elf_Shdr);
1916 ehdr->e_shstrndx = SHN_UNDEF;
1917 if (numsegs + 1 < PN_XNUM) {
1918 ehdr->e_phnum = numsegs + 1;
1921 ehdr->e_phnum = PN_XNUM;
1924 ehdr->e_shoff = ehdr->e_phoff +
1925 (numsegs + 1) * ehdr->e_phentsize;
1926 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1927 ("e_shoff: %zu, hdrsize - shdr: %zu",
1928 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1930 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1931 memset(shdr, 0, sizeof(*shdr));
1933 * A special first section is used to hold large segment and
1934 * section counts. This was proposed by Sun Microsystems in
1935 * Solaris and has been adopted by Linux; the standard ELF
1936 * tools are already familiar with the technique.
1938 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1939 * (or 12-7 depending on the version of the document) for more
1942 shdr->sh_type = SHT_NULL;
1943 shdr->sh_size = ehdr->e_shnum;
1944 shdr->sh_link = ehdr->e_shstrndx;
1945 shdr->sh_info = numsegs + 1;
1949 * Fill in the program header entries.
1951 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1953 /* The note segement. */
1954 phdr->p_type = PT_NOTE;
1955 phdr->p_offset = hdrsize;
1958 phdr->p_filesz = notesz;
1960 phdr->p_flags = PF_R;
1961 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1964 /* All the writable segments from the program. */
1966 phc.offset = round_page(hdrsize + notesz);
1967 each_dumpable_segment(td, cb_put_phdr, &phc);
1971 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1973 struct note_info *ninfo;
1974 size_t size, notesize;
1977 out(arg, NULL, &size);
1978 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1980 ninfo->outfunc = out;
1981 ninfo->outarg = arg;
1982 ninfo->outsize = size;
1983 TAILQ_INSERT_TAIL(list, ninfo, link);
1988 notesize = sizeof(Elf_Note) + /* note header */
1989 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1991 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1997 append_note_data(const void *src, void *dst, size_t len)
2001 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2003 bcopy(src, dst, len);
2004 bzero((char *)dst + len, padded_len - len);
2006 return (padded_len);
2010 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2018 note = (Elf_Note *)buf;
2019 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2020 note->n_descsz = size;
2021 note->n_type = type;
2022 buf += sizeof(*note);
2023 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2024 sizeof(FREEBSD_ABI_VENDOR));
2025 append_note_data(src, buf, size);
2030 notesize = sizeof(Elf_Note) + /* note header */
2031 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2033 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2039 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2042 ssize_t old_len, sect_len;
2043 size_t new_len, descsz, i;
2045 if (ninfo->type == -1) {
2046 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2050 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2051 note.n_descsz = ninfo->outsize;
2052 note.n_type = ninfo->type;
2054 sbuf_bcat(sb, ¬e, sizeof(note));
2055 sbuf_start_section(sb, &old_len);
2056 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2057 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2058 if (note.n_descsz == 0)
2060 sbuf_start_section(sb, &old_len);
2061 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2062 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2066 new_len = (size_t)sect_len;
2067 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2068 if (new_len < descsz) {
2070 * It is expected that individual note emitters will correctly
2071 * predict their expected output size and fill up to that size
2072 * themselves, padding in a format-specific way if needed.
2073 * However, in case they don't, just do it here with zeros.
2075 for (i = 0; i < descsz - new_len; i++)
2077 } else if (new_len > descsz) {
2079 * We can't always truncate sb -- we may have drained some
2082 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2083 "read it (%zu > %zu). Since it is longer than "
2084 "expected, this coredump's notes are corrupt. THIS "
2085 "IS A BUG in the note_procstat routine for type %u.\n",
2086 __func__, (unsigned)note.n_type, new_len, descsz,
2087 (unsigned)note.n_type));
2092 * Miscellaneous note out functions.
2095 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2096 #include <compat/freebsd32/freebsd32.h>
2097 #include <compat/freebsd32/freebsd32_signal.h>
2099 typedef struct prstatus32 elf_prstatus_t;
2100 typedef struct prpsinfo32 elf_prpsinfo_t;
2101 typedef struct fpreg32 elf_prfpregset_t;
2102 typedef struct fpreg32 elf_fpregset_t;
2103 typedef struct reg32 elf_gregset_t;
2104 typedef struct thrmisc32 elf_thrmisc_t;
2105 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2106 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2107 typedef uint32_t elf_ps_strings_t;
2109 typedef prstatus_t elf_prstatus_t;
2110 typedef prpsinfo_t elf_prpsinfo_t;
2111 typedef prfpregset_t elf_prfpregset_t;
2112 typedef prfpregset_t elf_fpregset_t;
2113 typedef gregset_t elf_gregset_t;
2114 typedef thrmisc_t elf_thrmisc_t;
2115 #define ELF_KERN_PROC_MASK 0
2116 typedef struct kinfo_proc elf_kinfo_proc_t;
2117 typedef vm_offset_t elf_ps_strings_t;
2121 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2127 elf_prpsinfo_t *psinfo;
2130 p = (struct proc *)arg;
2132 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2133 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2134 psinfo->pr_version = PRPSINFO_VERSION;
2135 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2136 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2138 if (p->p_args != NULL) {
2139 len = sizeof(psinfo->pr_psargs) - 1;
2140 if (len > p->p_args->ar_length)
2141 len = p->p_args->ar_length;
2142 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2148 sbuf_new(&sbarg, psinfo->pr_psargs,
2149 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2150 error = proc_getargv(curthread, p, &sbarg);
2152 if (sbuf_finish(&sbarg) == 0)
2153 len = sbuf_len(&sbarg) - 1;
2155 len = sizeof(psinfo->pr_psargs) - 1;
2156 sbuf_delete(&sbarg);
2158 if (error || len == 0)
2159 strlcpy(psinfo->pr_psargs, p->p_comm,
2160 sizeof(psinfo->pr_psargs));
2162 KASSERT(len < sizeof(psinfo->pr_psargs),
2163 ("len is too long: %zu vs %zu", len,
2164 sizeof(psinfo->pr_psargs)));
2165 cp = psinfo->pr_psargs;
2168 cp = memchr(cp, '\0', end - cp);
2174 psinfo->pr_pid = p->p_pid;
2175 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2176 free(psinfo, M_TEMP);
2178 *sizep = sizeof(*psinfo);
2182 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2185 elf_prstatus_t *status;
2187 td = (struct thread *)arg;
2189 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2190 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2191 status->pr_version = PRSTATUS_VERSION;
2192 status->pr_statussz = sizeof(elf_prstatus_t);
2193 status->pr_gregsetsz = sizeof(elf_gregset_t);
2194 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2195 status->pr_osreldate = osreldate;
2196 status->pr_cursig = td->td_proc->p_sig;
2197 status->pr_pid = td->td_tid;
2198 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2199 fill_regs32(td, &status->pr_reg);
2201 fill_regs(td, &status->pr_reg);
2203 sbuf_bcat(sb, status, sizeof(*status));
2204 free(status, M_TEMP);
2206 *sizep = sizeof(*status);
2210 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2213 elf_prfpregset_t *fpregset;
2215 td = (struct thread *)arg;
2217 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2218 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2219 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2220 fill_fpregs32(td, fpregset);
2222 fill_fpregs(td, fpregset);
2224 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2225 free(fpregset, M_TEMP);
2227 *sizep = sizeof(*fpregset);
2231 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2234 elf_thrmisc_t thrmisc;
2236 td = (struct thread *)arg;
2238 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2239 bzero(&thrmisc, sizeof(thrmisc));
2240 strcpy(thrmisc.pr_tname, td->td_name);
2241 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2243 *sizep = sizeof(thrmisc);
2247 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2252 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2253 struct ptrace_lwpinfo32 pl;
2255 struct ptrace_lwpinfo pl;
2258 td = (struct thread *)arg;
2259 size = sizeof(structsize) + sizeof(pl);
2261 KASSERT(*sizep == size, ("invalid size"));
2262 structsize = sizeof(pl);
2263 sbuf_bcat(sb, &structsize, sizeof(structsize));
2264 bzero(&pl, sizeof(pl));
2265 pl.pl_lwpid = td->td_tid;
2266 pl.pl_event = PL_EVENT_NONE;
2267 pl.pl_sigmask = td->td_sigmask;
2268 pl.pl_siglist = td->td_siglist;
2269 if (td->td_si.si_signo != 0) {
2270 pl.pl_event = PL_EVENT_SIGNAL;
2271 pl.pl_flags |= PL_FLAG_SI;
2272 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2273 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2275 pl.pl_siginfo = td->td_si;
2278 strcpy(pl.pl_tdname, td->td_name);
2279 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2280 sbuf_bcat(sb, &pl, sizeof(pl));
2286 * Allow for MD specific notes, as well as any MD
2287 * specific preparations for writing MI notes.
2290 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2296 td = (struct thread *)arg;
2298 if (size != 0 && sb != NULL)
2299 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2303 __elfN(dump_thread)(td, buf, &size);
2304 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2305 if (size != 0 && sb != NULL)
2306 sbuf_bcat(sb, buf, size);
2311 #ifdef KINFO_PROC_SIZE
2312 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2316 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2322 p = (struct proc *)arg;
2323 size = sizeof(structsize) + p->p_numthreads *
2324 sizeof(elf_kinfo_proc_t);
2327 KASSERT(*sizep == size, ("invalid size"));
2328 structsize = sizeof(elf_kinfo_proc_t);
2329 sbuf_bcat(sb, &structsize, sizeof(structsize));
2330 sx_slock(&proctree_lock);
2332 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2333 sx_sunlock(&proctree_lock);
2338 #ifdef KINFO_FILE_SIZE
2339 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2343 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2346 size_t size, sect_sz, i;
2347 ssize_t start_len, sect_len;
2348 int structsize, filedesc_flags;
2350 if (coredump_pack_fileinfo)
2351 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2355 p = (struct proc *)arg;
2356 structsize = sizeof(struct kinfo_file);
2359 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2360 sbuf_set_drain(sb, sbuf_count_drain, &size);
2361 sbuf_bcat(sb, &structsize, sizeof(structsize));
2363 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2368 sbuf_start_section(sb, &start_len);
2370 sbuf_bcat(sb, &structsize, sizeof(structsize));
2372 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2375 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2380 KASSERT(sect_sz <= *sizep,
2381 ("kern_proc_filedesc_out did not respect maxlen; "
2382 "requested %zu, got %zu", *sizep - sizeof(structsize),
2383 sect_sz - sizeof(structsize)));
2385 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2390 #ifdef KINFO_VMENTRY_SIZE
2391 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2395 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2399 int structsize, vmmap_flags;
2401 if (coredump_pack_vmmapinfo)
2402 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2406 p = (struct proc *)arg;
2407 structsize = sizeof(struct kinfo_vmentry);
2410 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2411 sbuf_set_drain(sb, sbuf_count_drain, &size);
2412 sbuf_bcat(sb, &structsize, sizeof(structsize));
2414 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2419 sbuf_bcat(sb, &structsize, sizeof(structsize));
2421 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2427 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2433 p = (struct proc *)arg;
2434 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2436 KASSERT(*sizep == size, ("invalid size"));
2437 structsize = sizeof(gid_t);
2438 sbuf_bcat(sb, &structsize, sizeof(structsize));
2439 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2446 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2452 p = (struct proc *)arg;
2453 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2455 KASSERT(*sizep == size, ("invalid size"));
2456 structsize = sizeof(p->p_fd->fd_cmask);
2457 sbuf_bcat(sb, &structsize, sizeof(structsize));
2458 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2464 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2467 struct rlimit rlim[RLIM_NLIMITS];
2471 p = (struct proc *)arg;
2472 size = sizeof(structsize) + sizeof(rlim);
2474 KASSERT(*sizep == size, ("invalid size"));
2475 structsize = sizeof(rlim);
2476 sbuf_bcat(sb, &structsize, sizeof(structsize));
2478 for (i = 0; i < RLIM_NLIMITS; i++)
2479 lim_rlimit_proc(p, i, &rlim[i]);
2481 sbuf_bcat(sb, rlim, sizeof(rlim));
2487 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2493 p = (struct proc *)arg;
2494 size = sizeof(structsize) + sizeof(p->p_osrel);
2496 KASSERT(*sizep == size, ("invalid size"));
2497 structsize = sizeof(p->p_osrel);
2498 sbuf_bcat(sb, &structsize, sizeof(structsize));
2499 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2505 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2508 elf_ps_strings_t ps_strings;
2512 p = (struct proc *)arg;
2513 size = sizeof(structsize) + sizeof(ps_strings);
2515 KASSERT(*sizep == size, ("invalid size"));
2516 structsize = sizeof(ps_strings);
2517 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2518 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2520 ps_strings = p->p_sysent->sv_psstrings;
2522 sbuf_bcat(sb, &structsize, sizeof(structsize));
2523 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2529 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2535 p = (struct proc *)arg;
2538 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2539 sbuf_set_drain(sb, sbuf_count_drain, &size);
2540 sbuf_bcat(sb, &structsize, sizeof(structsize));
2542 proc_getauxv(curthread, p, sb);
2548 structsize = sizeof(Elf_Auxinfo);
2549 sbuf_bcat(sb, &structsize, sizeof(structsize));
2551 proc_getauxv(curthread, p, sb);
2557 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2558 const char *note_vendor, const Elf_Phdr *pnote,
2559 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2561 const Elf_Note *note, *note0, *note_end;
2562 const char *note_name;
2567 /* We need some limit, might as well use PAGE_SIZE. */
2568 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2570 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2571 if (pnote->p_offset > PAGE_SIZE ||
2572 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2573 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2575 VOP_UNLOCK(imgp->vp, 0);
2576 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2577 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2579 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2580 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2581 curthread->td_ucred, NOCRED, NULL, curthread);
2583 uprintf("i/o error PT_NOTE\n");
2586 note = note0 = (const Elf_Note *)buf;
2587 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2589 note = note0 = (const Elf_Note *)(imgp->image_header +
2591 note_end = (const Elf_Note *)(imgp->image_header +
2592 pnote->p_offset + pnote->p_filesz);
2595 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2596 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2597 (const char *)note < sizeof(Elf_Note)) {
2600 if (note->n_namesz != checknote->n_namesz ||
2601 note->n_descsz != checknote->n_descsz ||
2602 note->n_type != checknote->n_type)
2604 note_name = (const char *)(note + 1);
2605 if (note_name + checknote->n_namesz >=
2606 (const char *)note_end || strncmp(note_vendor,
2607 note_name, checknote->n_namesz) != 0)
2610 if (cb(note, cb_arg, &res))
2613 note = (const Elf_Note *)((const char *)(note + 1) +
2614 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2615 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2624 struct brandnote_cb_arg {
2625 Elf_Brandnote *brandnote;
2630 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2632 struct brandnote_cb_arg *arg;
2637 * Fetch the osreldate for binary from the ELF OSABI-note if
2640 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2641 arg->brandnote->trans_osrel != NULL ?
2642 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2647 static Elf_Note fctl_note = {
2648 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2649 .n_descsz = sizeof(uint32_t),
2650 .n_type = NT_FREEBSD_FEATURE_CTL,
2653 struct fctl_cb_arg {
2658 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2660 struct fctl_cb_arg *arg;
2661 const Elf32_Word *desc;
2665 p = (uintptr_t)(note + 1);
2666 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2667 desc = (const Elf32_Word *)p;
2668 *arg->fctl0 = desc[0];
2674 * Try to find the appropriate ABI-note section for checknote, fetch
2675 * the osreldate and feature control flags for binary from the ELF
2676 * OSABI-note. Only the first page of the image is searched, the same
2680 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2681 int32_t *osrel, uint32_t *fctl0)
2683 const Elf_Phdr *phdr;
2684 const Elf_Ehdr *hdr;
2685 struct brandnote_cb_arg b_arg;
2686 struct fctl_cb_arg f_arg;
2689 hdr = (const Elf_Ehdr *)imgp->image_header;
2690 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2691 b_arg.brandnote = brandnote;
2692 b_arg.osrel = osrel;
2693 f_arg.fctl0 = fctl0;
2695 for (i = 0; i < hdr->e_phnum; i++) {
2696 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2697 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2699 for (j = 0; j < hdr->e_phnum; j++) {
2700 if (phdr[j].p_type == PT_NOTE &&
2701 __elfN(parse_notes)(imgp, &fctl_note,
2702 FREEBSD_ABI_VENDOR, &phdr[j],
2703 note_fctl_cb, &f_arg))
2714 * Tell kern_execve.c about it, with a little help from the linker.
2716 static struct execsw __elfN(execsw) = {
2717 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2718 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2720 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2723 __elfN(trans_prot)(Elf_Word flags)
2729 prot |= VM_PROT_EXECUTE;
2731 prot |= VM_PROT_WRITE;
2733 prot |= VM_PROT_READ;
2734 #if __ELF_WORD_SIZE == 32
2735 #if defined(__amd64__)
2736 if (i386_read_exec && (flags & PF_R))
2737 prot |= VM_PROT_EXECUTE;
2744 __elfN(untrans_prot)(vm_prot_t prot)
2749 if (prot & VM_PROT_EXECUTE)
2751 if (prot & VM_PROT_READ)
2753 if (prot & VM_PROT_WRITE)
2759 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2761 u_long range, rbase, gap;
2764 pct = __elfN(aslr_stack_gap);
2769 range = imgp->eff_stack_sz * pct / 100;
2770 arc4rand(&rbase, sizeof(rbase), 0);
2771 gap = rbase % range;
2772 gap &= ~(sizeof(u_long) - 1);