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, int interp_name_len, int32_t *osrel, uint32_t *fctl0);
92 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
93 u_long *entry, size_t pagesize);
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,
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static bool __elfN(freebsd_trans_osrel)(const Elf_Note *note,
100 static bool kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
101 static boolean_t __elfN(check_note)(struct image_params *imgp,
102 Elf_Brandnote *checknote, int32_t *osrel, uint32_t *fctl0);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
109 #define CORE_BUF_SIZE (16 * 1024)
111 int __elfN(fallback_brand) = -1;
112 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
113 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
114 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
116 static int elf_legacy_coredump = 0;
117 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
118 &elf_legacy_coredump, 0,
119 "include all and only RW pages in core dumps");
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (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 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
142 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
144 static int __elfN(aslr_enabled) = 0;
145 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, enable, CTLFLAG_RWTUN,
146 &__elfN(aslr_enabled), 0,
147 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
148 ": enable address map randomization");
150 static int __elfN(pie_aslr_enabled) = 0;
151 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, pie_enable, CTLFLAG_RWTUN,
152 &__elfN(pie_aslr_enabled), 0,
153 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
154 ": enable address map randomization for PIE binaries");
156 static int __elfN(aslr_honor_sbrk) = 1;
157 SYSCTL_INT(ASLR_NODE_OID, OID_AUTO, honor_sbrk, CTLFLAG_RW,
158 &__elfN(aslr_honor_sbrk), 0,
159 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": assume sbrk is used");
161 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
163 #define trunc_page_ps(va, ps) rounddown2(va, ps)
164 #define round_page_ps(va, ps) roundup2(va, ps)
165 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
167 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
169 Elf_Brandnote __elfN(freebsd_brandnote) = {
170 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
171 .hdr.n_descsz = sizeof(int32_t),
172 .hdr.n_type = NT_FREEBSD_ABI_TAG,
173 .vendor = FREEBSD_ABI_VENDOR,
174 .flags = BN_TRANSLATE_OSREL,
175 .trans_osrel = __elfN(freebsd_trans_osrel)
179 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
183 p = (uintptr_t)(note + 1);
184 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
185 *osrel = *(const int32_t *)(p);
190 static const char GNU_ABI_VENDOR[] = "GNU";
191 static int GNU_KFREEBSD_ABI_DESC = 3;
193 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
194 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
195 .hdr.n_descsz = 16, /* XXX at least 16 */
197 .vendor = GNU_ABI_VENDOR,
198 .flags = BN_TRANSLATE_OSREL,
199 .trans_osrel = kfreebsd_trans_osrel
203 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
205 const Elf32_Word *desc;
208 p = (uintptr_t)(note + 1);
209 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
211 desc = (const Elf32_Word *)p;
212 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
216 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
217 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
219 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
225 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
229 for (i = 0; i < MAX_BRANDS; i++) {
230 if (elf_brand_list[i] == NULL) {
231 elf_brand_list[i] = entry;
235 if (i == MAX_BRANDS) {
236 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
244 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
248 for (i = 0; i < MAX_BRANDS; i++) {
249 if (elf_brand_list[i] == entry) {
250 elf_brand_list[i] = NULL;
260 __elfN(brand_inuse)(Elf_Brandinfo *entry)
265 sx_slock(&allproc_lock);
266 FOREACH_PROC_IN_SYSTEM(p) {
267 if (p->p_sysent == entry->sysvec) {
272 sx_sunlock(&allproc_lock);
277 static Elf_Brandinfo *
278 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
279 int interp_name_len, int32_t *osrel, uint32_t *fctl0)
281 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
282 Elf_Brandinfo *bi, *bi_m;
287 * We support four types of branding -- (1) the ELF EI_OSABI field
288 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
289 * branding w/in the ELF header, (3) path of the `interp_path'
290 * field, and (4) the ".note.ABI-tag" ELF section.
293 /* Look for an ".note.ABI-tag" ELF section */
295 for (i = 0; i < MAX_BRANDS; i++) {
296 bi = elf_brand_list[i];
299 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
301 if (hdr->e_machine == bi->machine && (bi->flags &
302 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
303 ret = __elfN(check_note)(imgp, bi->brand_note, osrel,
305 /* Give brand a chance to veto check_note's guess */
306 if (ret && bi->header_supported)
307 ret = bi->header_supported(imgp);
309 * If note checker claimed the binary, but the
310 * interpreter path in the image does not
311 * match default one for the brand, try to
312 * search for other brands with the same
313 * interpreter. Either there is better brand
314 * with the right interpreter, or, failing
315 * this, we return first brand which accepted
316 * our note and, optionally, header.
318 if (ret && bi_m == NULL && interp != NULL &&
319 (bi->interp_path == NULL ||
320 (strlen(bi->interp_path) + 1 != interp_name_len ||
321 strncmp(interp, bi->interp_path, interp_name_len)
333 /* If the executable has a brand, search for it in the brand list. */
334 for (i = 0; i < MAX_BRANDS; i++) {
335 bi = elf_brand_list[i];
336 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
337 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
339 if (hdr->e_machine == bi->machine &&
340 (hdr->e_ident[EI_OSABI] == bi->brand ||
341 (bi->compat_3_brand != NULL &&
342 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
343 bi->compat_3_brand) == 0))) {
344 /* Looks good, but give brand a chance to veto */
345 if (bi->header_supported == NULL ||
346 bi->header_supported(imgp)) {
348 * Again, prefer strictly matching
351 if (interp_name_len == 0 &&
352 bi->interp_path == NULL)
354 if (bi->interp_path != NULL &&
355 strlen(bi->interp_path) + 1 ==
356 interp_name_len && strncmp(interp,
357 bi->interp_path, interp_name_len) == 0)
367 /* No known brand, see if the header is recognized by any brand */
368 for (i = 0; i < MAX_BRANDS; i++) {
369 bi = elf_brand_list[i];
370 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
371 bi->header_supported == NULL)
373 if (hdr->e_machine == bi->machine) {
374 ret = bi->header_supported(imgp);
380 /* Lacking a known brand, search for a recognized interpreter. */
381 if (interp != NULL) {
382 for (i = 0; i < MAX_BRANDS; i++) {
383 bi = elf_brand_list[i];
384 if (bi == NULL || (bi->flags &
385 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
388 if (hdr->e_machine == bi->machine &&
389 bi->interp_path != NULL &&
390 /* ELF image p_filesz includes terminating zero */
391 strlen(bi->interp_path) + 1 == interp_name_len &&
392 strncmp(interp, bi->interp_path, interp_name_len)
393 == 0 && (bi->header_supported == NULL ||
394 bi->header_supported(imgp)))
399 /* Lacking a recognized interpreter, try the default brand */
400 for (i = 0; i < MAX_BRANDS; i++) {
401 bi = elf_brand_list[i];
402 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
403 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
405 if (hdr->e_machine == bi->machine &&
406 __elfN(fallback_brand) == bi->brand &&
407 (bi->header_supported == NULL ||
408 bi->header_supported(imgp)))
415 __elfN(check_header)(const Elf_Ehdr *hdr)
421 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
422 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
423 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
424 hdr->e_phentsize != sizeof(Elf_Phdr) ||
425 hdr->e_version != ELF_TARG_VER)
429 * Make sure we have at least one brand for this machine.
432 for (i = 0; i < MAX_BRANDS; i++) {
433 bi = elf_brand_list[i];
434 if (bi != NULL && bi->machine == hdr->e_machine)
444 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
445 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
452 * Create the page if it doesn't exist yet. Ignore errors.
454 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
455 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
458 * Find the page from the underlying object.
460 if (object != NULL) {
461 sf = vm_imgact_map_page(object, offset);
463 return (KERN_FAILURE);
464 off = offset - trunc_page(offset);
465 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
467 vm_imgact_unmap_page(sf);
469 return (KERN_FAILURE);
472 return (KERN_SUCCESS);
476 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
477 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
483 int error, locked, rv;
485 if (start != trunc_page(start)) {
486 rv = __elfN(map_partial)(map, object, offset, start,
487 round_page(start), prot);
488 if (rv != KERN_SUCCESS)
490 offset += round_page(start) - start;
491 start = round_page(start);
493 if (end != round_page(end)) {
494 rv = __elfN(map_partial)(map, object, offset +
495 trunc_page(end) - start, trunc_page(end), end, prot);
496 if (rv != KERN_SUCCESS)
498 end = trunc_page(end);
501 return (KERN_SUCCESS);
502 if ((offset & PAGE_MASK) != 0) {
504 * The mapping is not page aligned. This means that we have
507 rv = vm_map_fixed(map, NULL, 0, start, end - start,
508 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
509 if (rv != KERN_SUCCESS)
512 return (KERN_SUCCESS);
513 for (; start < end; start += sz) {
514 sf = vm_imgact_map_page(object, offset);
516 return (KERN_FAILURE);
517 off = offset - trunc_page(offset);
519 if (sz > PAGE_SIZE - off)
520 sz = PAGE_SIZE - off;
521 error = copyout((caddr_t)sf_buf_kva(sf) + off,
523 vm_imgact_unmap_page(sf);
525 return (KERN_FAILURE);
529 vm_object_reference(object);
530 rv = vm_map_fixed(map, object, offset, start, end - start,
531 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
532 if (rv != KERN_SUCCESS) {
533 locked = VOP_ISLOCKED(imgp->vp);
534 VOP_UNLOCK(imgp->vp, 0);
535 vm_object_deallocate(object);
536 vn_lock(imgp->vp, locked | LK_RETRY);
540 return (KERN_SUCCESS);
544 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
545 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
552 vm_offset_t off, map_addr;
555 vm_ooffset_t file_addr;
558 * It's necessary to fail if the filsz + offset taken from the
559 * header is greater than the actual file pager object's size.
560 * If we were to allow this, then the vm_map_find() below would
561 * walk right off the end of the file object and into the ether.
563 * While I'm here, might as well check for something else that
564 * is invalid: filsz cannot be greater than memsz.
566 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
568 uprintf("elf_load_section: truncated ELF file\n");
572 object = imgp->object;
573 map = &imgp->proc->p_vmspace->vm_map;
574 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
575 file_addr = trunc_page_ps(offset, pagesize);
578 * We have two choices. We can either clear the data in the last page
579 * of an oversized mapping, or we can start the anon mapping a page
580 * early and copy the initialized data into that first page. We
585 else if (memsz > filsz)
586 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
588 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
591 /* cow flags: don't dump readonly sections in core */
592 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
593 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
595 rv = __elfN(map_insert)(imgp, map,
597 file_addr, /* file offset */
598 map_addr, /* virtual start */
599 map_addr + map_len,/* virtual end */
602 if (rv != KERN_SUCCESS)
605 /* we can stop now if we've covered it all */
612 * We have to get the remaining bit of the file into the first part
613 * of the oversized map segment. This is normally because the .data
614 * segment in the file is extended to provide bss. It's a neat idea
615 * to try and save a page, but it's a pain in the behind to implement.
617 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
619 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
620 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
623 /* This had damn well better be true! */
625 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
626 map_addr + map_len, prot, 0);
627 if (rv != KERN_SUCCESS)
632 sf = vm_imgact_map_page(object, offset + filsz);
636 /* send the page fragment to user space */
637 off = trunc_page_ps(offset + filsz, pagesize) -
638 trunc_page(offset + filsz);
639 error = copyout((caddr_t)sf_buf_kva(sf) + off,
640 (caddr_t)map_addr, copy_len);
641 vm_imgact_unmap_page(sf);
647 * Remove write access to the page if it was only granted by map_insert
650 if ((prot & VM_PROT_WRITE) == 0)
651 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
652 map_len), prot, FALSE);
658 * Load the file "file" into memory. It may be either a shared object
661 * The "addr" reference parameter is in/out. On entry, it specifies
662 * the address where a shared object should be loaded. If the file is
663 * an executable, this value is ignored. On exit, "addr" specifies
664 * where the file was actually loaded.
666 * The "entry" reference parameter is out only. On exit, it specifies
667 * the entry point for the loaded file.
670 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
671 u_long *entry, size_t pagesize)
676 struct image_params image_params;
678 const Elf_Ehdr *hdr = NULL;
679 const Elf_Phdr *phdr = NULL;
680 struct nameidata *nd;
682 struct image_params *imgp;
685 u_long base_addr = 0;
686 int error, i, numsegs;
688 #ifdef CAPABILITY_MODE
690 * XXXJA: This check can go away once we are sufficiently confident
691 * that the checks in namei() are correct.
693 if (IN_CAPABILITY_MODE(curthread))
697 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
699 attr = &tempdata->attr;
700 imgp = &tempdata->image_params;
703 * Initialize part of the common data
707 imgp->firstpage = NULL;
708 imgp->image_header = NULL;
710 imgp->execlabel = NULL;
712 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
713 if ((error = namei(nd)) != 0) {
717 NDFREE(nd, NDF_ONLY_PNBUF);
718 imgp->vp = nd->ni_vp;
721 * Check permissions, modes, uid, etc on the file, and "open" it.
723 error = exec_check_permissions(imgp);
727 error = exec_map_first_page(imgp);
732 * Also make certain that the interpreter stays the same, so set
733 * its VV_TEXT flag, too.
735 VOP_SET_TEXT(nd->ni_vp);
737 imgp->object = nd->ni_vp->v_object;
739 hdr = (const Elf_Ehdr *)imgp->image_header;
740 if ((error = __elfN(check_header)(hdr)) != 0)
742 if (hdr->e_type == ET_DYN)
744 else if (hdr->e_type == ET_EXEC)
751 /* Only support headers that fit within first page for now */
752 if ((hdr->e_phoff > PAGE_SIZE) ||
753 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
758 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
759 if (!aligned(phdr, Elf_Addr)) {
764 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
765 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
766 /* Loadable segment */
767 prot = __elfN(trans_prot)(phdr[i].p_flags);
768 error = __elfN(load_section)(imgp, phdr[i].p_offset,
769 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
770 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
774 * Establish the base address if this is the
778 base_addr = trunc_page(phdr[i].p_vaddr +
784 *entry = (unsigned long)hdr->e_entry + rbase;
788 exec_unmap_first_page(imgp);
793 free(tempdata, M_TEMP);
799 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
804 MPASS(vm_map_min(map) <= minv);
805 MPASS(maxv <= vm_map_max(map));
807 MPASS(minv + align < maxv);
808 arc4rand(&rbase, sizeof(rbase), 0);
809 res = roundup(minv, (u_long)align) + rbase % (maxv - minv);
810 res &= ~((u_long)align - 1);
814 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
815 res, minv, maxv, rbase));
817 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
818 res, maxv, minv, rbase));
823 * Impossible et_dyn_addr initial value indicating that the real base
824 * must be calculated later with some randomization applied.
826 #define ET_DYN_ADDR_RAND 1
829 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
833 const Elf_Phdr *phdr;
834 Elf_Auxargs *elf_auxargs;
835 struct vmspace *vmspace;
837 const char *err_str, *newinterp;
838 char *interp, *interp_buf, *path;
839 Elf_Brandinfo *brand_info;
840 struct sysentvec *sv;
842 u_long text_size, data_size, total_size, text_addr, data_addr;
843 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
844 u_long maxalign, mapsz, maxv, maxv1;
847 int error, i, n, interp_name_len, have_interp;
849 hdr = (const Elf_Ehdr *)imgp->image_header;
852 * Do we have a valid ELF header ?
854 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
855 * if particular brand doesn't support it.
857 if (__elfN(check_header)(hdr) != 0 ||
858 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
862 * From here on down, we return an errno, not -1, as we've
863 * detected an ELF file.
866 if ((hdr->e_phoff > PAGE_SIZE) ||
867 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
868 /* Only support headers in first page for now */
869 uprintf("Program headers not in the first page\n");
872 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
873 if (!aligned(phdr, Elf_Addr)) {
874 uprintf("Unaligned program headers\n");
882 text_size = data_size = total_size = text_addr = data_addr = 0;
885 err_str = newinterp = NULL;
886 interp = interp_buf = NULL;
888 maxalign = PAGE_SIZE;
891 for (i = 0; i < hdr->e_phnum; i++) {
892 switch (phdr[i].p_type) {
895 baddr = phdr[i].p_vaddr;
896 if (phdr[i].p_align > maxalign)
897 maxalign = phdr[i].p_align;
898 mapsz += phdr[i].p_memsz;
902 /* Path to interpreter */
903 if (phdr[i].p_filesz < 2 ||
904 phdr[i].p_filesz > MAXPATHLEN) {
905 uprintf("Invalid PT_INTERP\n");
909 if (interp != NULL) {
910 uprintf("Multiple PT_INTERP headers\n");
914 interp_name_len = phdr[i].p_filesz;
915 if (phdr[i].p_offset > PAGE_SIZE ||
916 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
917 VOP_UNLOCK(imgp->vp, 0);
918 interp_buf = malloc(interp_name_len + 1, M_TEMP,
920 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
921 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
922 interp_name_len, phdr[i].p_offset,
923 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
926 uprintf("i/o error PT_INTERP %d\n",
930 interp_buf[interp_name_len] = '\0';
933 interp = __DECONST(char *, imgp->image_header) +
935 if (interp[interp_name_len - 1] != '\0') {
936 uprintf("Invalid PT_INTERP\n");
945 __elfN(trans_prot)(phdr[i].p_flags);
946 imgp->stack_sz = phdr[i].p_memsz;
951 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
953 if (brand_info == NULL) {
954 uprintf("ELF binary type \"%u\" not known.\n",
955 hdr->e_ident[EI_OSABI]);
959 sv = brand_info->sysvec;
961 if (hdr->e_type == ET_DYN) {
962 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
963 uprintf("Cannot execute shared object\n");
968 * Honour the base load address from the dso if it is
969 * non-zero for some reason.
972 if ((sv->sv_flags & SV_ASLR) == 0 ||
973 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
974 et_dyn_addr = ET_DYN_LOAD_ADDR;
975 else if ((__elfN(pie_aslr_enabled) &&
976 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
977 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
978 et_dyn_addr = ET_DYN_ADDR_RAND;
980 et_dyn_addr = ET_DYN_LOAD_ADDR;
983 if (interp != NULL && brand_info->interp_newpath != NULL)
984 newinterp = brand_info->interp_newpath;
987 * Avoid a possible deadlock if the current address space is destroyed
988 * and that address space maps the locked vnode. In the common case,
989 * the locked vnode's v_usecount is decremented but remains greater
990 * than zero. Consequently, the vnode lock is not needed by vrele().
991 * However, in cases where the vnode lock is external, such as nullfs,
992 * v_usecount may become zero.
994 * The VV_TEXT flag prevents modifications to the executable while
995 * the vnode is unlocked.
997 VOP_UNLOCK(imgp->vp, 0);
1000 * Decide whether to enable randomization of user mappings.
1001 * First, reset user preferences for the setid binaries.
1002 * Then, account for the support of the randomization by the
1003 * ABI, by user preferences, and make special treatment for
1006 if (imgp->credential_setid) {
1007 PROC_LOCK(imgp->proc);
1008 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1009 PROC_UNLOCK(imgp->proc);
1011 if ((sv->sv_flags & SV_ASLR) == 0 ||
1012 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1013 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1014 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1015 ("et_dyn_addr == RAND and !ASLR"));
1016 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1017 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1018 et_dyn_addr == ET_DYN_ADDR_RAND) {
1019 imgp->map_flags |= MAP_ASLR;
1021 * If user does not care about sbrk, utilize the bss
1022 * grow region for mappings as well. We can select
1023 * the base for the image anywere and still not suffer
1024 * from the fragmentation.
1026 if (!__elfN(aslr_honor_sbrk) ||
1027 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1028 imgp->map_flags |= MAP_ASLR_IGNSTART;
1031 error = exec_new_vmspace(imgp, sv);
1032 vmspace = imgp->proc->p_vmspace;
1033 map = &vmspace->vm_map;
1035 imgp->proc->p_sysent = sv;
1037 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1038 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1039 KASSERT((map->flags & MAP_ASLR) != 0,
1040 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1041 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1042 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1043 /* reserve half of the address space to interpreter */
1044 maxv / 2, 1UL << flsl(maxalign));
1047 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1051 for (i = 0; i < hdr->e_phnum; i++) {
1052 switch (phdr[i].p_type) {
1053 case PT_LOAD: /* Loadable segment */
1054 if (phdr[i].p_memsz == 0)
1056 prot = __elfN(trans_prot)(phdr[i].p_flags);
1057 error = __elfN(load_section)(imgp, phdr[i].p_offset,
1058 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
1059 phdr[i].p_memsz, phdr[i].p_filesz, prot,
1065 * If this segment contains the program headers,
1066 * remember their virtual address for the AT_PHDR
1067 * aux entry. Static binaries don't usually include
1070 if (phdr[i].p_offset == 0 &&
1071 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
1072 <= phdr[i].p_filesz)
1073 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
1076 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
1077 seg_size = round_page(phdr[i].p_memsz +
1078 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
1081 * Make the largest executable segment the official
1082 * text segment and all others data.
1084 * Note that obreak() assumes that data_addr +
1085 * data_size == end of data load area, and the ELF
1086 * file format expects segments to be sorted by
1087 * address. If multiple data segments exist, the
1088 * last one will be used.
1091 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
1092 text_size = seg_size;
1093 text_addr = seg_addr;
1095 data_size = seg_size;
1096 data_addr = seg_addr;
1098 total_size += seg_size;
1100 case PT_PHDR: /* Program header table info */
1101 proghdr = phdr[i].p_vaddr + et_dyn_addr;
1108 if (data_addr == 0 && data_size == 0) {
1109 data_addr = text_addr;
1110 data_size = text_size;
1113 entry = (u_long)hdr->e_entry + et_dyn_addr;
1116 * Check limits. It should be safe to check the
1117 * limits after loading the segments since we do
1118 * not actually fault in all the segments pages.
1120 PROC_LOCK(imgp->proc);
1121 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1122 err_str = "Data segment size exceeds process limit";
1123 else if (text_size > maxtsiz)
1124 err_str = "Text segment size exceeds system limit";
1125 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1126 err_str = "Total segment size exceeds process limit";
1127 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1128 err_str = "Data segment size exceeds resource limit";
1129 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1130 err_str = "Total segment size exceeds resource limit";
1131 if (err_str != NULL) {
1132 PROC_UNLOCK(imgp->proc);
1133 uprintf("%s\n", err_str);
1138 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1139 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1140 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1141 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1144 * We load the dynamic linker where a userland call
1145 * to mmap(0, ...) would put it. The rationale behind this
1146 * calculation is that it leaves room for the heap to grow to
1147 * its maximum allowed size.
1149 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1151 if ((map->flags & MAP_ASLR) != 0) {
1152 maxv1 = maxv / 2 + addr / 2;
1153 MPASS(maxv1 >= addr); /* No overflow */
1154 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1155 MAXPAGESIZES > 1 ? pagesizes[1] : pagesizes[0]);
1157 map->anon_loc = addr;
1159 PROC_UNLOCK(imgp->proc);
1161 imgp->entry_addr = entry;
1163 if (interp != NULL) {
1164 have_interp = FALSE;
1165 VOP_UNLOCK(imgp->vp, 0);
1166 if ((map->flags & MAP_ASLR) != 0) {
1167 /* Assume that interpeter fits into 1/4 of AS */
1168 maxv1 = maxv / 2 + addr / 2;
1169 MPASS(maxv1 >= addr); /* No overflow */
1170 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1173 if (brand_info->emul_path != NULL &&
1174 brand_info->emul_path[0] != '\0') {
1175 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1176 snprintf(path, MAXPATHLEN, "%s%s",
1177 brand_info->emul_path, interp);
1178 error = __elfN(load_file)(imgp->proc, path, &addr,
1179 &imgp->entry_addr, sv->sv_pagesize);
1184 if (!have_interp && newinterp != NULL &&
1185 (brand_info->interp_path == NULL ||
1186 strcmp(interp, brand_info->interp_path) == 0)) {
1187 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1188 &imgp->entry_addr, sv->sv_pagesize);
1193 error = __elfN(load_file)(imgp->proc, interp, &addr,
1194 &imgp->entry_addr, sv->sv_pagesize);
1196 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1198 uprintf("ELF interpreter %s not found, error %d\n",
1206 * Construct auxargs table (used by the fixup routine)
1208 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1209 elf_auxargs->execfd = -1;
1210 elf_auxargs->phdr = proghdr;
1211 elf_auxargs->phent = hdr->e_phentsize;
1212 elf_auxargs->phnum = hdr->e_phnum;
1213 elf_auxargs->pagesz = PAGE_SIZE;
1214 elf_auxargs->base = addr;
1215 elf_auxargs->flags = 0;
1216 elf_auxargs->entry = entry;
1217 elf_auxargs->hdr_eflags = hdr->e_flags;
1219 imgp->auxargs = elf_auxargs;
1220 imgp->interpreted = 0;
1221 imgp->reloc_base = addr;
1222 imgp->proc->p_osrel = osrel;
1223 imgp->proc->p_fctl0 = fctl0;
1224 imgp->proc->p_elf_machine = hdr->e_machine;
1225 imgp->proc->p_elf_flags = hdr->e_flags;
1228 free(interp_buf, M_TEMP);
1232 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1235 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1237 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1238 Elf_Auxinfo *argarray, *pos;
1239 Elf_Addr *base, *auxbase;
1242 base = (Elf_Addr *)*stack_base;
1243 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1244 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1247 if (args->execfd != -1)
1248 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1249 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1250 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1251 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1252 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1253 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1254 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1255 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1256 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1257 if (imgp->execpathp != 0)
1258 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1259 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1260 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1261 if (imgp->canary != 0) {
1262 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1263 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1265 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1266 if (imgp->pagesizes != 0) {
1267 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1268 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1270 if (imgp->sysent->sv_timekeep_base != 0) {
1271 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1272 imgp->sysent->sv_timekeep_base);
1274 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1275 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1276 imgp->sysent->sv_stackprot);
1277 if (imgp->sysent->sv_hwcap != NULL)
1278 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1279 if (imgp->sysent->sv_hwcap2 != NULL)
1280 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1281 AUXARGS_ENTRY(pos, AT_NULL, 0);
1283 free(imgp->auxargs, M_TEMP);
1284 imgp->auxargs = NULL;
1285 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1287 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1288 free(argarray, M_TEMP);
1293 if (suword(base, imgp->args->argc) == -1)
1295 *stack_base = (register_t *)base;
1300 * Code for generating ELF core dumps.
1303 typedef void (*segment_callback)(vm_map_entry_t, void *);
1305 /* Closure for cb_put_phdr(). */
1306 struct phdr_closure {
1307 Elf_Phdr *phdr; /* Program header to fill in */
1308 Elf_Off offset; /* Offset of segment in core file */
1311 /* Closure for cb_size_segment(). */
1312 struct sseg_closure {
1313 int count; /* Count of writable segments. */
1314 size_t size; /* Total size of all writable segments. */
1317 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1320 int type; /* Note type. */
1321 outfunc_t outfunc; /* Output function. */
1322 void *outarg; /* Argument for the output function. */
1323 size_t outsize; /* Output size. */
1324 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1327 TAILQ_HEAD(note_info_list, note_info);
1329 /* Coredump output parameters. */
1330 struct coredump_params {
1332 struct ucred *active_cred;
1333 struct ucred *file_cred;
1336 struct compressor *comp;
1339 extern int compress_user_cores;
1340 extern int compress_user_cores_level;
1342 static void cb_put_phdr(vm_map_entry_t, void *);
1343 static void cb_size_segment(vm_map_entry_t, void *);
1344 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1346 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1347 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1348 struct note_info_list *, size_t);
1349 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1351 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1352 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1353 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1354 static int sbuf_drain_core_output(void *, const char *, int);
1355 static int sbuf_drain_count(void *arg, const char *data, int len);
1357 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1358 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1359 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1360 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1361 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1362 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1363 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1364 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1365 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1366 static void note_procstat_files(void *, struct sbuf *, size_t *);
1367 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1368 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1369 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1370 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1371 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1374 * Write out a core segment to the compression stream.
1377 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1383 chunk_len = MIN(len, CORE_BUF_SIZE);
1386 * We can get EFAULT error here.
1387 * In that case zero out the current chunk of the segment.
1389 error = copyin(base, buf, chunk_len);
1391 bzero(buf, chunk_len);
1392 error = compressor_write(p->comp, buf, chunk_len);
1402 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1405 return (core_write((struct coredump_params *)arg, base, len, offset,
1410 core_write(struct coredump_params *p, const void *base, size_t len,
1411 off_t offset, enum uio_seg seg)
1414 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1415 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1416 p->active_cred, p->file_cred, NULL, p->td));
1420 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1425 if (p->comp != NULL)
1426 return (compress_chunk(p, base, tmpbuf, len));
1429 * EFAULT is a non-fatal error that we can get, for example,
1430 * if the segment is backed by a file but extends beyond its
1433 error = core_write(p, base, len, offset, UIO_USERSPACE);
1434 if (error == EFAULT) {
1435 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1436 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1437 "for process %s\n", base, len, offset, curproc->p_comm);
1440 * Write a "real" zero byte at the end of the target region
1441 * in the case this is the last segment.
1442 * The intermediate space will be implicitly zero-filled.
1444 error = core_write(p, zero_region, 1, offset + len - 1,
1451 * Drain into a core file.
1454 sbuf_drain_core_output(void *arg, const char *data, int len)
1456 struct coredump_params *p;
1459 p = (struct coredump_params *)arg;
1462 * Some kern_proc out routines that print to this sbuf may
1463 * call us with the process lock held. Draining with the
1464 * non-sleepable lock held is unsafe. The lock is needed for
1465 * those routines when dumping a live process. In our case we
1466 * can safely release the lock before draining and acquire
1469 locked = PROC_LOCKED(p->td->td_proc);
1471 PROC_UNLOCK(p->td->td_proc);
1472 if (p->comp != NULL)
1473 error = compressor_write(p->comp, __DECONST(char *, data), len);
1475 error = core_write(p, __DECONST(void *, data), len, p->offset,
1478 PROC_LOCK(p->td->td_proc);
1486 * Drain into a counter.
1489 sbuf_drain_count(void *arg, const char *data __unused, int len)
1493 sizep = (size_t *)arg;
1499 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1501 struct ucred *cred = td->td_ucred;
1503 struct sseg_closure seginfo;
1504 struct note_info_list notelst;
1505 struct coredump_params params;
1506 struct note_info *ninfo;
1508 size_t hdrsize, notesz, coresize;
1512 TAILQ_INIT(¬elst);
1514 /* Size the program segments. */
1517 each_dumpable_segment(td, cb_size_segment, &seginfo);
1520 * Collect info about the core file header area.
1522 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1523 if (seginfo.count + 1 >= PN_XNUM)
1524 hdrsize += sizeof(Elf_Shdr);
1525 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1526 coresize = round_page(hdrsize + notesz) + seginfo.size;
1528 /* Set up core dump parameters. */
1530 params.active_cred = cred;
1531 params.file_cred = NOCRED;
1538 PROC_LOCK(td->td_proc);
1539 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1540 PROC_UNLOCK(td->td_proc);
1547 if (coresize >= limit) {
1552 /* Create a compression stream if necessary. */
1553 if (compress_user_cores != 0) {
1554 params.comp = compressor_init(core_compressed_write,
1555 compress_user_cores, CORE_BUF_SIZE,
1556 compress_user_cores_level, ¶ms);
1557 if (params.comp == NULL) {
1561 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1565 * Allocate memory for building the header, fill it up,
1566 * and write it out following the notes.
1568 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1569 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1572 /* Write the contents of all of the writable segments. */
1578 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1579 offset = round_page(hdrsize + notesz);
1580 for (i = 0; i < seginfo.count; i++) {
1581 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1582 php->p_filesz, offset, ¶ms, tmpbuf);
1585 offset += php->p_filesz;
1588 if (error == 0 && params.comp != NULL)
1589 error = compressor_flush(params.comp);
1593 "Failed to write core file for process %s (error %d)\n",
1594 curproc->p_comm, error);
1598 free(tmpbuf, M_TEMP);
1599 if (params.comp != NULL)
1600 compressor_fini(params.comp);
1601 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1602 TAILQ_REMOVE(¬elst, ninfo, link);
1603 free(ninfo, M_TEMP);
1612 * A callback for each_dumpable_segment() to write out the segment's
1613 * program header entry.
1616 cb_put_phdr(vm_map_entry_t entry, void *closure)
1618 struct phdr_closure *phc = (struct phdr_closure *)closure;
1619 Elf_Phdr *phdr = phc->phdr;
1621 phc->offset = round_page(phc->offset);
1623 phdr->p_type = PT_LOAD;
1624 phdr->p_offset = phc->offset;
1625 phdr->p_vaddr = entry->start;
1627 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1628 phdr->p_align = PAGE_SIZE;
1629 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1631 phc->offset += phdr->p_filesz;
1636 * A callback for each_dumpable_segment() to gather information about
1637 * the number of segments and their total size.
1640 cb_size_segment(vm_map_entry_t entry, void *closure)
1642 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1645 ssc->size += entry->end - entry->start;
1649 * For each writable segment in the process's memory map, call the given
1650 * function with a pointer to the map entry and some arbitrary
1651 * caller-supplied data.
1654 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1656 struct proc *p = td->td_proc;
1657 vm_map_t map = &p->p_vmspace->vm_map;
1658 vm_map_entry_t entry;
1659 vm_object_t backing_object, object;
1660 boolean_t ignore_entry;
1662 vm_map_lock_read(map);
1663 for (entry = map->header.next; entry != &map->header;
1664 entry = entry->next) {
1666 * Don't dump inaccessible mappings, deal with legacy
1669 * Note that read-only segments related to the elf binary
1670 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1671 * need to arbitrarily ignore such segments.
1673 if (elf_legacy_coredump) {
1674 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1677 if ((entry->protection & VM_PROT_ALL) == 0)
1682 * Dont include memory segment in the coredump if
1683 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1684 * madvise(2). Do not dump submaps (i.e. parts of the
1687 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1690 if ((object = entry->object.vm_object) == NULL)
1693 /* Ignore memory-mapped devices and such things. */
1694 VM_OBJECT_RLOCK(object);
1695 while ((backing_object = object->backing_object) != NULL) {
1696 VM_OBJECT_RLOCK(backing_object);
1697 VM_OBJECT_RUNLOCK(object);
1698 object = backing_object;
1700 ignore_entry = object->type != OBJT_DEFAULT &&
1701 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1702 object->type != OBJT_PHYS;
1703 VM_OBJECT_RUNLOCK(object);
1707 (*func)(entry, closure);
1709 vm_map_unlock_read(map);
1713 * Write the core file header to the file, including padding up to
1714 * the page boundary.
1717 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1718 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1720 struct note_info *ninfo;
1724 /* Fill in the header. */
1725 bzero(hdr, hdrsize);
1726 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1728 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1729 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1730 sbuf_start_section(sb, NULL);
1731 sbuf_bcat(sb, hdr, hdrsize);
1732 TAILQ_FOREACH(ninfo, notelst, link)
1733 __elfN(putnote)(ninfo, sb);
1734 /* Align up to a page boundary for the program segments. */
1735 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1736 error = sbuf_finish(sb);
1743 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1753 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1756 * To have the debugger select the right thread (LWP) as the initial
1757 * thread, we dump the state of the thread passed to us in td first.
1758 * This is the thread that causes the core dump and thus likely to
1759 * be the right thread one wants to have selected in the debugger.
1762 while (thr != NULL) {
1763 size += register_note(list, NT_PRSTATUS,
1764 __elfN(note_prstatus), thr);
1765 size += register_note(list, NT_FPREGSET,
1766 __elfN(note_fpregset), thr);
1767 size += register_note(list, NT_THRMISC,
1768 __elfN(note_thrmisc), thr);
1769 size += register_note(list, NT_PTLWPINFO,
1770 __elfN(note_ptlwpinfo), thr);
1771 size += register_note(list, -1,
1772 __elfN(note_threadmd), thr);
1774 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1775 TAILQ_NEXT(thr, td_plist);
1777 thr = TAILQ_NEXT(thr, td_plist);
1780 size += register_note(list, NT_PROCSTAT_PROC,
1781 __elfN(note_procstat_proc), p);
1782 size += register_note(list, NT_PROCSTAT_FILES,
1783 note_procstat_files, p);
1784 size += register_note(list, NT_PROCSTAT_VMMAP,
1785 note_procstat_vmmap, p);
1786 size += register_note(list, NT_PROCSTAT_GROUPS,
1787 note_procstat_groups, p);
1788 size += register_note(list, NT_PROCSTAT_UMASK,
1789 note_procstat_umask, p);
1790 size += register_note(list, NT_PROCSTAT_RLIMIT,
1791 note_procstat_rlimit, p);
1792 size += register_note(list, NT_PROCSTAT_OSREL,
1793 note_procstat_osrel, p);
1794 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1795 __elfN(note_procstat_psstrings), p);
1796 size += register_note(list, NT_PROCSTAT_AUXV,
1797 __elfN(note_procstat_auxv), p);
1803 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1809 struct phdr_closure phc;
1811 ehdr = (Elf_Ehdr *)hdr;
1813 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1814 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1815 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1816 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1817 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1818 ehdr->e_ident[EI_DATA] = ELF_DATA;
1819 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1820 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1821 ehdr->e_ident[EI_ABIVERSION] = 0;
1822 ehdr->e_ident[EI_PAD] = 0;
1823 ehdr->e_type = ET_CORE;
1824 ehdr->e_machine = td->td_proc->p_elf_machine;
1825 ehdr->e_version = EV_CURRENT;
1827 ehdr->e_phoff = sizeof(Elf_Ehdr);
1828 ehdr->e_flags = td->td_proc->p_elf_flags;
1829 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1830 ehdr->e_phentsize = sizeof(Elf_Phdr);
1831 ehdr->e_shentsize = sizeof(Elf_Shdr);
1832 ehdr->e_shstrndx = SHN_UNDEF;
1833 if (numsegs + 1 < PN_XNUM) {
1834 ehdr->e_phnum = numsegs + 1;
1837 ehdr->e_phnum = PN_XNUM;
1840 ehdr->e_shoff = ehdr->e_phoff +
1841 (numsegs + 1) * ehdr->e_phentsize;
1842 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1843 ("e_shoff: %zu, hdrsize - shdr: %zu",
1844 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1846 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1847 memset(shdr, 0, sizeof(*shdr));
1849 * A special first section is used to hold large segment and
1850 * section counts. This was proposed by Sun Microsystems in
1851 * Solaris and has been adopted by Linux; the standard ELF
1852 * tools are already familiar with the technique.
1854 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1855 * (or 12-7 depending on the version of the document) for more
1858 shdr->sh_type = SHT_NULL;
1859 shdr->sh_size = ehdr->e_shnum;
1860 shdr->sh_link = ehdr->e_shstrndx;
1861 shdr->sh_info = numsegs + 1;
1865 * Fill in the program header entries.
1867 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1869 /* The note segement. */
1870 phdr->p_type = PT_NOTE;
1871 phdr->p_offset = hdrsize;
1874 phdr->p_filesz = notesz;
1876 phdr->p_flags = PF_R;
1877 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1880 /* All the writable segments from the program. */
1882 phc.offset = round_page(hdrsize + notesz);
1883 each_dumpable_segment(td, cb_put_phdr, &phc);
1887 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1889 struct note_info *ninfo;
1890 size_t size, notesize;
1893 out(arg, NULL, &size);
1894 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1896 ninfo->outfunc = out;
1897 ninfo->outarg = arg;
1898 ninfo->outsize = size;
1899 TAILQ_INSERT_TAIL(list, ninfo, link);
1904 notesize = sizeof(Elf_Note) + /* note header */
1905 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1907 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1913 append_note_data(const void *src, void *dst, size_t len)
1917 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1919 bcopy(src, dst, len);
1920 bzero((char *)dst + len, padded_len - len);
1922 return (padded_len);
1926 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1934 note = (Elf_Note *)buf;
1935 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1936 note->n_descsz = size;
1937 note->n_type = type;
1938 buf += sizeof(*note);
1939 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1940 sizeof(FREEBSD_ABI_VENDOR));
1941 append_note_data(src, buf, size);
1946 notesize = sizeof(Elf_Note) + /* note header */
1947 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1949 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1955 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1958 ssize_t old_len, sect_len;
1959 size_t new_len, descsz, i;
1961 if (ninfo->type == -1) {
1962 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1966 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1967 note.n_descsz = ninfo->outsize;
1968 note.n_type = ninfo->type;
1970 sbuf_bcat(sb, ¬e, sizeof(note));
1971 sbuf_start_section(sb, &old_len);
1972 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1973 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1974 if (note.n_descsz == 0)
1976 sbuf_start_section(sb, &old_len);
1977 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1978 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1982 new_len = (size_t)sect_len;
1983 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1984 if (new_len < descsz) {
1986 * It is expected that individual note emitters will correctly
1987 * predict their expected output size and fill up to that size
1988 * themselves, padding in a format-specific way if needed.
1989 * However, in case they don't, just do it here with zeros.
1991 for (i = 0; i < descsz - new_len; i++)
1993 } else if (new_len > descsz) {
1995 * We can't always truncate sb -- we may have drained some
1998 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1999 "read it (%zu > %zu). Since it is longer than "
2000 "expected, this coredump's notes are corrupt. THIS "
2001 "IS A BUG in the note_procstat routine for type %u.\n",
2002 __func__, (unsigned)note.n_type, new_len, descsz,
2003 (unsigned)note.n_type));
2008 * Miscellaneous note out functions.
2011 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2012 #include <compat/freebsd32/freebsd32.h>
2013 #include <compat/freebsd32/freebsd32_signal.h>
2015 typedef struct prstatus32 elf_prstatus_t;
2016 typedef struct prpsinfo32 elf_prpsinfo_t;
2017 typedef struct fpreg32 elf_prfpregset_t;
2018 typedef struct fpreg32 elf_fpregset_t;
2019 typedef struct reg32 elf_gregset_t;
2020 typedef struct thrmisc32 elf_thrmisc_t;
2021 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2022 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2023 typedef uint32_t elf_ps_strings_t;
2025 typedef prstatus_t elf_prstatus_t;
2026 typedef prpsinfo_t elf_prpsinfo_t;
2027 typedef prfpregset_t elf_prfpregset_t;
2028 typedef prfpregset_t elf_fpregset_t;
2029 typedef gregset_t elf_gregset_t;
2030 typedef thrmisc_t elf_thrmisc_t;
2031 #define ELF_KERN_PROC_MASK 0
2032 typedef struct kinfo_proc elf_kinfo_proc_t;
2033 typedef vm_offset_t elf_ps_strings_t;
2037 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2043 elf_prpsinfo_t *psinfo;
2046 p = (struct proc *)arg;
2048 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2049 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2050 psinfo->pr_version = PRPSINFO_VERSION;
2051 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2052 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2054 if (p->p_args != NULL) {
2055 len = sizeof(psinfo->pr_psargs) - 1;
2056 if (len > p->p_args->ar_length)
2057 len = p->p_args->ar_length;
2058 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2064 sbuf_new(&sbarg, psinfo->pr_psargs,
2065 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2066 error = proc_getargv(curthread, p, &sbarg);
2068 if (sbuf_finish(&sbarg) == 0)
2069 len = sbuf_len(&sbarg) - 1;
2071 len = sizeof(psinfo->pr_psargs) - 1;
2072 sbuf_delete(&sbarg);
2074 if (error || len == 0)
2075 strlcpy(psinfo->pr_psargs, p->p_comm,
2076 sizeof(psinfo->pr_psargs));
2078 KASSERT(len < sizeof(psinfo->pr_psargs),
2079 ("len is too long: %zu vs %zu", len,
2080 sizeof(psinfo->pr_psargs)));
2081 cp = psinfo->pr_psargs;
2084 cp = memchr(cp, '\0', end - cp);
2090 psinfo->pr_pid = p->p_pid;
2091 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2092 free(psinfo, M_TEMP);
2094 *sizep = sizeof(*psinfo);
2098 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2101 elf_prstatus_t *status;
2103 td = (struct thread *)arg;
2105 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2106 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2107 status->pr_version = PRSTATUS_VERSION;
2108 status->pr_statussz = sizeof(elf_prstatus_t);
2109 status->pr_gregsetsz = sizeof(elf_gregset_t);
2110 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2111 status->pr_osreldate = osreldate;
2112 status->pr_cursig = td->td_proc->p_sig;
2113 status->pr_pid = td->td_tid;
2114 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2115 fill_regs32(td, &status->pr_reg);
2117 fill_regs(td, &status->pr_reg);
2119 sbuf_bcat(sb, status, sizeof(*status));
2120 free(status, M_TEMP);
2122 *sizep = sizeof(*status);
2126 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2129 elf_prfpregset_t *fpregset;
2131 td = (struct thread *)arg;
2133 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2134 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2135 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2136 fill_fpregs32(td, fpregset);
2138 fill_fpregs(td, fpregset);
2140 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2141 free(fpregset, M_TEMP);
2143 *sizep = sizeof(*fpregset);
2147 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2150 elf_thrmisc_t thrmisc;
2152 td = (struct thread *)arg;
2154 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2155 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2156 strcpy(thrmisc.pr_tname, td->td_name);
2157 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2159 *sizep = sizeof(thrmisc);
2163 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2168 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2169 struct ptrace_lwpinfo32 pl;
2171 struct ptrace_lwpinfo pl;
2174 td = (struct thread *)arg;
2175 size = sizeof(structsize) + sizeof(pl);
2177 KASSERT(*sizep == size, ("invalid size"));
2178 structsize = sizeof(pl);
2179 sbuf_bcat(sb, &structsize, sizeof(structsize));
2180 bzero(&pl, sizeof(pl));
2181 pl.pl_lwpid = td->td_tid;
2182 pl.pl_event = PL_EVENT_NONE;
2183 pl.pl_sigmask = td->td_sigmask;
2184 pl.pl_siglist = td->td_siglist;
2185 if (td->td_si.si_signo != 0) {
2186 pl.pl_event = PL_EVENT_SIGNAL;
2187 pl.pl_flags |= PL_FLAG_SI;
2188 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2189 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2191 pl.pl_siginfo = td->td_si;
2194 strcpy(pl.pl_tdname, td->td_name);
2195 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2196 sbuf_bcat(sb, &pl, sizeof(pl));
2202 * Allow for MD specific notes, as well as any MD
2203 * specific preparations for writing MI notes.
2206 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2212 td = (struct thread *)arg;
2214 if (size != 0 && sb != NULL)
2215 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2219 __elfN(dump_thread)(td, buf, &size);
2220 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2221 if (size != 0 && sb != NULL)
2222 sbuf_bcat(sb, buf, size);
2227 #ifdef KINFO_PROC_SIZE
2228 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2232 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2238 p = (struct proc *)arg;
2239 size = sizeof(structsize) + p->p_numthreads *
2240 sizeof(elf_kinfo_proc_t);
2243 KASSERT(*sizep == size, ("invalid size"));
2244 structsize = sizeof(elf_kinfo_proc_t);
2245 sbuf_bcat(sb, &structsize, sizeof(structsize));
2247 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2252 #ifdef KINFO_FILE_SIZE
2253 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2257 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2260 size_t size, sect_sz, i;
2261 ssize_t start_len, sect_len;
2262 int structsize, filedesc_flags;
2264 if (coredump_pack_fileinfo)
2265 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2269 p = (struct proc *)arg;
2270 structsize = sizeof(struct kinfo_file);
2273 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2274 sbuf_set_drain(sb, sbuf_drain_count, &size);
2275 sbuf_bcat(sb, &structsize, sizeof(structsize));
2277 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2282 sbuf_start_section(sb, &start_len);
2284 sbuf_bcat(sb, &structsize, sizeof(structsize));
2286 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2289 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2294 KASSERT(sect_sz <= *sizep,
2295 ("kern_proc_filedesc_out did not respect maxlen; "
2296 "requested %zu, got %zu", *sizep - sizeof(structsize),
2297 sect_sz - sizeof(structsize)));
2299 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2304 #ifdef KINFO_VMENTRY_SIZE
2305 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2309 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2313 int structsize, vmmap_flags;
2315 if (coredump_pack_vmmapinfo)
2316 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2320 p = (struct proc *)arg;
2321 structsize = sizeof(struct kinfo_vmentry);
2324 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2325 sbuf_set_drain(sb, sbuf_drain_count, &size);
2326 sbuf_bcat(sb, &structsize, sizeof(structsize));
2328 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2333 sbuf_bcat(sb, &structsize, sizeof(structsize));
2335 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2341 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2347 p = (struct proc *)arg;
2348 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2350 KASSERT(*sizep == size, ("invalid size"));
2351 structsize = sizeof(gid_t);
2352 sbuf_bcat(sb, &structsize, sizeof(structsize));
2353 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2360 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2366 p = (struct proc *)arg;
2367 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2369 KASSERT(*sizep == size, ("invalid size"));
2370 structsize = sizeof(p->p_fd->fd_cmask);
2371 sbuf_bcat(sb, &structsize, sizeof(structsize));
2372 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2378 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2381 struct rlimit rlim[RLIM_NLIMITS];
2385 p = (struct proc *)arg;
2386 size = sizeof(structsize) + sizeof(rlim);
2388 KASSERT(*sizep == size, ("invalid size"));
2389 structsize = sizeof(rlim);
2390 sbuf_bcat(sb, &structsize, sizeof(structsize));
2392 for (i = 0; i < RLIM_NLIMITS; i++)
2393 lim_rlimit_proc(p, i, &rlim[i]);
2395 sbuf_bcat(sb, rlim, sizeof(rlim));
2401 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2407 p = (struct proc *)arg;
2408 size = sizeof(structsize) + sizeof(p->p_osrel);
2410 KASSERT(*sizep == size, ("invalid size"));
2411 structsize = sizeof(p->p_osrel);
2412 sbuf_bcat(sb, &structsize, sizeof(structsize));
2413 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2419 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2422 elf_ps_strings_t ps_strings;
2426 p = (struct proc *)arg;
2427 size = sizeof(structsize) + sizeof(ps_strings);
2429 KASSERT(*sizep == size, ("invalid size"));
2430 structsize = sizeof(ps_strings);
2431 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2432 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2434 ps_strings = p->p_sysent->sv_psstrings;
2436 sbuf_bcat(sb, &structsize, sizeof(structsize));
2437 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2443 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2449 p = (struct proc *)arg;
2452 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2453 sbuf_set_drain(sb, sbuf_drain_count, &size);
2454 sbuf_bcat(sb, &structsize, sizeof(structsize));
2456 proc_getauxv(curthread, p, sb);
2462 structsize = sizeof(Elf_Auxinfo);
2463 sbuf_bcat(sb, &structsize, sizeof(structsize));
2465 proc_getauxv(curthread, p, sb);
2471 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2472 const char *note_vendor, const Elf_Phdr *pnote,
2473 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2475 const Elf_Note *note, *note0, *note_end;
2476 const char *note_name;
2481 /* We need some limit, might as well use PAGE_SIZE. */
2482 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2484 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2485 if (pnote->p_offset > PAGE_SIZE ||
2486 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2487 VOP_UNLOCK(imgp->vp, 0);
2488 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2489 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2490 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2491 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2492 curthread->td_ucred, NOCRED, NULL, curthread);
2494 uprintf("i/o error PT_NOTE\n");
2497 note = note0 = (const Elf_Note *)buf;
2498 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2500 note = note0 = (const Elf_Note *)(imgp->image_header +
2502 note_end = (const Elf_Note *)(imgp->image_header +
2503 pnote->p_offset + pnote->p_filesz);
2506 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2507 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2508 (const char *)note < sizeof(Elf_Note)) {
2511 if (note->n_namesz != checknote->n_namesz ||
2512 note->n_descsz != checknote->n_descsz ||
2513 note->n_type != checknote->n_type)
2515 note_name = (const char *)(note + 1);
2516 if (note_name + checknote->n_namesz >=
2517 (const char *)note_end || strncmp(note_vendor,
2518 note_name, checknote->n_namesz) != 0)
2521 if (cb(note, cb_arg, &res))
2524 note = (const Elf_Note *)((const char *)(note + 1) +
2525 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2526 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2535 struct brandnote_cb_arg {
2536 Elf_Brandnote *brandnote;
2541 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2543 struct brandnote_cb_arg *arg;
2548 * Fetch the osreldate for binary from the ELF OSABI-note if
2551 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2552 arg->brandnote->trans_osrel != NULL ?
2553 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2558 static Elf_Note fctl_note = {
2559 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2560 .n_descsz = sizeof(uint32_t),
2561 .n_type = NT_FREEBSD_FEATURE_CTL,
2564 struct fctl_cb_arg {
2569 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2571 struct fctl_cb_arg *arg;
2572 const Elf32_Word *desc;
2576 p = (uintptr_t)(note + 1);
2577 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2578 desc = (const Elf32_Word *)p;
2579 *arg->fctl0 = desc[0];
2584 * Try to find the appropriate ABI-note section for checknote, fetch
2585 * the osreldate and feature control flags for binary from the ELF
2586 * OSABI-note. Only the first page of the image is searched, the same
2590 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2591 int32_t *osrel, uint32_t *fctl0)
2593 const Elf_Phdr *phdr;
2594 const Elf_Ehdr *hdr;
2595 struct brandnote_cb_arg b_arg;
2596 struct fctl_cb_arg f_arg;
2599 hdr = (const Elf_Ehdr *)imgp->image_header;
2600 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2601 b_arg.brandnote = brandnote;
2602 b_arg.osrel = osrel;
2603 f_arg.fctl0 = fctl0;
2605 for (i = 0; i < hdr->e_phnum; i++) {
2606 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2607 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2609 for (j = 0; j < hdr->e_phnum; j++) {
2610 if (phdr[j].p_type == PT_NOTE &&
2611 __elfN(parse_notes)(imgp, &fctl_note,
2612 FREEBSD_ABI_VENDOR, &phdr[j],
2613 note_fctl_cb, &f_arg))
2624 * Tell kern_execve.c about it, with a little help from the linker.
2626 static struct execsw __elfN(execsw) = {
2627 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2628 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2630 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2633 __elfN(trans_prot)(Elf_Word flags)
2639 prot |= VM_PROT_EXECUTE;
2641 prot |= VM_PROT_WRITE;
2643 prot |= VM_PROT_READ;
2644 #if __ELF_WORD_SIZE == 32
2645 #if defined(__amd64__)
2646 if (i386_read_exec && (flags & PF_R))
2647 prot |= VM_PROT_EXECUTE;
2654 __elfN(untrans_prot)(vm_prot_t prot)
2659 if (prot & VM_PROT_EXECUTE)
2661 if (prot & VM_PROT_READ)
2663 if (prot & VM_PROT_WRITE)