FreeBSD/Linux Kernel Cross Reference
sys/kern/imgact_elf.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
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
8 * All rights reserved.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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
21 *
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.
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_capsicum.h"
38
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mman.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
54 #include <sys/proc.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>
60 #include <sys/sbuf.h>
61 #include <sys/sf_buf.h>
62 #include <sys/smp.h>
63 #include <sys/systm.h>
64 #include <sys/signalvar.h>
65 #include <sys/stat.h>
66 #include <sys/sx.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>
73 #include <sys/user.h>
74
75 #include <vm/vm.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82
83 #include <machine/elf.h>
84 #include <machine/md_var.h>
85
86 #define ELF_NOTE_ROUNDSIZE 4
87 #define OLD_EI_BRAND 8
88
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,
93 u_long *entry);
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,
98 int32_t *osrel);
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);
104
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106 "");
107
108 #define CORE_BUF_SIZE (16 * 1024)
109
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");
114
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");
119
120 int __elfN(nxstack) =
121 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
122 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__) || \
123 defined(__riscv)
124 1;
125 #else
126 0;
127 #endif
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");
131
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");
137 #endif
138 #endif
139
140 static u_long __elfN(pie_base) = ET_DYN_LOAD_ADDR;
141 static int
142 sysctl_pie_base(SYSCTL_HANDLER_ARGS)
143 {
144 u_long val;
145 int error;
146
147 val = __elfN(pie_base);
148 error = sysctl_handle_long(oidp, &val, 0, req);
149 if (error != 0 || req->newptr == NULL)
150 return (error);
151 if ((val & PAGE_MASK) != 0)
152 return (EINVAL);
153 __elfN(pie_base) = val;
154 return (0);
155 }
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");
160
161 SYSCTL_NODE(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO, aslr, CTLFLAG_RW, 0,
162 "");
163 #define ASLR_NODE_OID __CONCAT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), _aslr)
164
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");
170
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");
176
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");
181
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");
187
188 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
189
190 #define aligned(a, t) (rounddown2((u_long)(a), sizeof(t)) == (u_long)(a))
191
192 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
193
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)
201 };
202
203 static bool
204 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
205 {
206 uintptr_t p;
207
208 p = (uintptr_t)(note + 1);
209 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
210 *osrel = *(const int32_t *)(p);
211
212 return (true);
213 }
214
215 static const char GNU_ABI_VENDOR[] = "GNU";
216 static int GNU_KFREEBSD_ABI_DESC = 3;
217
218 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
219 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
220 .hdr.n_descsz = 16, /* XXX at least 16 */
221 .hdr.n_type = 1,
222 .vendor = GNU_ABI_VENDOR,
223 .flags = BN_TRANSLATE_OSREL,
224 .trans_osrel = kfreebsd_trans_osrel
225 };
226
227 static bool
228 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
229 {
230 const Elf32_Word *desc;
231 uintptr_t p;
232
233 p = (uintptr_t)(note + 1);
234 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
235
236 desc = (const Elf32_Word *)p;
237 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
238 return (false);
239
240 /*
241 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
242 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
243 */
244 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
245
246 return (true);
247 }
248
249 int
250 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
251 {
252 int i;
253
254 for (i = 0; i < MAX_BRANDS; i++) {
255 if (elf_brand_list[i] == NULL) {
256 elf_brand_list[i] = entry;
257 break;
258 }
259 }
260 if (i == MAX_BRANDS) {
261 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
262 __func__, entry);
263 return (-1);
264 }
265 return (0);
266 }
267
268 int
269 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
270 {
271 int i;
272
273 for (i = 0; i < MAX_BRANDS; i++) {
274 if (elf_brand_list[i] == entry) {
275 elf_brand_list[i] = NULL;
276 break;
277 }
278 }
279 if (i == MAX_BRANDS)
280 return (-1);
281 return (0);
282 }
283
284 int
285 __elfN(brand_inuse)(Elf_Brandinfo *entry)
286 {
287 struct proc *p;
288 int rval = FALSE;
289
290 sx_slock(&allproc_lock);
291 FOREACH_PROC_IN_SYSTEM(p) {
292 if (p->p_sysent == entry->sysvec) {
293 rval = TRUE;
294 break;
295 }
296 }
297 sx_sunlock(&allproc_lock);
298
299 return (rval);
300 }
301
302 static Elf_Brandinfo *
303 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
304 int32_t *osrel, uint32_t *fctl0)
305 {
306 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
307 Elf_Brandinfo *bi, *bi_m;
308 boolean_t ret;
309 int i, interp_name_len;
310
311 interp_name_len = interp != NULL ? strlen(interp) + 1 : 0;
312
313 /*
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.
318 */
319
320 /* Look for an ".note.ABI-tag" ELF section */
321 bi_m = NULL;
322 for (i = 0; i < MAX_BRANDS; i++) {
323 bi = elf_brand_list[i];
324 if (bi == NULL)
325 continue;
326 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
327 continue;
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,
331 fctl0);
332 /* Give brand a chance to veto check_note's guess */
333 if (ret && bi->header_supported)
334 ret = bi->header_supported(imgp);
335 /*
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.
344 */
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)
349 != 0))) {
350 bi_m = bi;
351 ret = 0;
352 }
353 if (ret)
354 return (bi);
355 }
356 }
357 if (bi_m != NULL)
358 return (bi_m);
359
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))
365 continue;
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)) {
374 /*
375 * Again, prefer strictly matching
376 * interpreter path.
377 */
378 if (interp_name_len == 0 &&
379 bi->interp_path == NULL)
380 return (bi);
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)
385 return (bi);
386 if (bi_m == NULL)
387 bi_m = bi;
388 }
389 }
390 }
391 if (bi_m != NULL)
392 return (bi_m);
393
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)
399 continue;
400 if (hdr->e_machine == bi->machine) {
401 ret = bi->header_supported(imgp);
402 if (ret)
403 return (bi);
404 }
405 }
406
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))
413 != 0)
414 continue;
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)))
422 return (bi);
423 }
424 }
425
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))
431 continue;
432 if (hdr->e_machine == bi->machine &&
433 __elfN(fallback_brand) == bi->brand &&
434 (bi->header_supported == NULL ||
435 bi->header_supported(imgp)))
436 return (bi);
437 }
438 return (NULL);
439 }
440
441 static bool
442 __elfN(phdr_in_zero_page)(const Elf_Ehdr *hdr)
443 {
444 return (hdr->e_phoff <= PAGE_SIZE &&
445 (u_int)hdr->e_phentsize * hdr->e_phnum <= PAGE_SIZE - hdr->e_phoff);
446 }
447
448 static int
449 __elfN(check_header)(const Elf_Ehdr *hdr)
450 {
451 Elf_Brandinfo *bi;
452 int i;
453
454 if (!IS_ELF(*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)
460 return (ENOEXEC);
461
462 /*
463 * Make sure we have at least one brand for this machine.
464 */
465
466 for (i = 0; i < MAX_BRANDS; i++) {
467 bi = elf_brand_list[i];
468 if (bi != NULL && bi->machine == hdr->e_machine)
469 break;
470 }
471 if (i == MAX_BRANDS)
472 return (ENOEXEC);
473
474 return (0);
475 }
476
477 static int
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)
480 {
481 struct sf_buf *sf;
482 int error;
483 vm_offset_t off;
484
485 /*
486 * Create the page if it doesn't exist yet. Ignore errors.
487 */
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);
490
491 /*
492 * Find the page from the underlying object.
493 */
494 if (object != NULL) {
495 sf = vm_imgact_map_page(object, offset);
496 if (sf == NULL)
497 return (KERN_FAILURE);
498 off = offset - trunc_page(offset);
499 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
500 end - start);
501 vm_imgact_unmap_page(sf);
502 if (error != 0)
503 return (KERN_FAILURE);
504 }
505
506 return (KERN_SUCCESS);
507 }
508
509 static int
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,
512 int cow)
513 {
514 struct sf_buf *sf;
515 vm_offset_t off;
516 vm_size_t sz;
517 int error, locked, rv;
518
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)
523 return (rv);
524 offset += round_page(start) - start;
525 start = round_page(start);
526 }
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)
531 return (rv);
532 end = trunc_page(end);
533 }
534 if (start >= end)
535 return (KERN_SUCCESS);
536 if ((offset & PAGE_MASK) != 0) {
537 /*
538 * The mapping is not page aligned. This means that we have
539 * to copy the data.
540 */
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)
544 return (rv);
545 if (object == NULL)
546 return (KERN_SUCCESS);
547 for (; start < end; start += sz) {
548 sf = vm_imgact_map_page(object, offset);
549 if (sf == NULL)
550 return (KERN_FAILURE);
551 off = offset - trunc_page(offset);
552 sz = end - start;
553 if (sz > PAGE_SIZE - off)
554 sz = PAGE_SIZE - off;
555 error = copyout((caddr_t)sf_buf_kva(sf) + off,
556 (caddr_t)start, sz);
557 vm_imgact_unmap_page(sf);
558 if (error != 0)
559 return (KERN_FAILURE);
560 offset += sz;
561 }
562 } else {
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);
572 return (rv);
573 } else if (object != NULL) {
574 MPASS(imgp->vp->v_object == object);
575 VOP_SET_TEXT_CHECKED(imgp->vp);
576 }
577 }
578 return (KERN_SUCCESS);
579 }
580
581 static int
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)
584 {
585 struct sf_buf *sf;
586 size_t map_len;
587 vm_map_t map;
588 vm_object_t object;
589 vm_offset_t off, map_addr;
590 int error, rv, cow;
591 size_t copy_len;
592 vm_ooffset_t file_addr;
593
594 /*
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.
599 *
600 * While I'm here, might as well check for something else that
601 * is invalid: filsz cannot be greater than memsz.
602 */
603 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
604 filsz > memsz) {
605 uprintf("elf_load_section: truncated ELF file\n");
606 return (ENOEXEC);
607 }
608
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);
613
614 /*
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
618 * choose the second.
619 */
620 if (filsz == 0)
621 map_len = 0;
622 else if (memsz > filsz)
623 map_len = trunc_page(offset + filsz) - file_addr;
624 else
625 map_len = round_page(offset + filsz) - file_addr;
626
627 if (map_len != 0) {
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);
631
632 rv = __elfN(map_insert)(imgp, map, object, file_addr,
633 map_addr, map_addr + map_len, prot, cow);
634 if (rv != KERN_SUCCESS)
635 return (EINVAL);
636
637 /* we can stop now if we've covered it all */
638 if (memsz == filsz)
639 return (0);
640 }
641
642
643 /*
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.
648 */
649 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page(offset +
650 filsz);
651 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
652 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
653
654 /* This had damn well better be true! */
655 if (map_len != 0) {
656 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
657 map_addr + map_len, prot, 0);
658 if (rv != KERN_SUCCESS)
659 return (EINVAL);
660 }
661
662 if (copy_len != 0) {
663 sf = vm_imgact_map_page(object, offset + filsz);
664 if (sf == NULL)
665 return (EIO);
666
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);
672 if (error != 0)
673 return (error);
674 }
675
676 /*
677 * Remove write access to the page if it was only granted by map_insert
678 * to allow copyout.
679 */
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);
683
684 return (0);
685 }
686
687 static int
688 __elfN(load_sections)(struct image_params *imgp, const Elf_Ehdr *hdr,
689 const Elf_Phdr *phdr, u_long rbase, u_long *base_addrp)
690 {
691 vm_prot_t prot;
692 u_long base_addr;
693 bool first;
694 int error, i;
695
696 ASSERT_VOP_LOCKED(imgp->vp, __func__);
697
698 base_addr = 0;
699 first = true;
700
701 for (i = 0; i < hdr->e_phnum; i++) {
702 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
703 continue;
704
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);
710 if (error != 0)
711 return (error);
712
713 /*
714 * Establish the base address if this is the first segment.
715 */
716 if (first) {
717 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
718 first = false;
719 }
720 }
721
722 if (base_addrp != NULL)
723 *base_addrp = base_addr;
724
725 return (0);
726 }
727
728 /*
729 * Load the file "file" into memory. It may be either a shared object
730 * or an executable.
731 *
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.
736 *
737 * The "entry" reference parameter is out only. On exit, it specifies
738 * the entry point for the loaded file.
739 */
740 static int
741 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
742 u_long *entry)
743 {
744 struct {
745 struct nameidata nd;
746 struct vattr attr;
747 struct image_params image_params;
748 } *tempdata;
749 const Elf_Ehdr *hdr = NULL;
750 const Elf_Phdr *phdr = NULL;
751 struct nameidata *nd;
752 struct vattr *attr;
753 struct image_params *imgp;
754 u_long rbase;
755 u_long base_addr = 0;
756 int error;
757
758 #ifdef CAPABILITY_MODE
759 /*
760 * XXXJA: This check can go away once we are sufficiently confident
761 * that the checks in namei() are correct.
762 */
763 if (IN_CAPABILITY_MODE(curthread))
764 return (ECAPMODE);
765 #endif
766
767 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK | M_ZERO);
768 nd = &tempdata->nd;
769 attr = &tempdata->attr;
770 imgp = &tempdata->image_params;
771
772 /*
773 * Initialize part of the common data
774 */
775 imgp->proc = p;
776 imgp->attr = attr;
777
778 NDINIT(nd, LOOKUP, ISOPEN | FOLLOW | LOCKSHARED | LOCKLEAF,
779 UIO_SYSSPACE, file, curthread);
780 if ((error = namei(nd)) != 0) {
781 nd->ni_vp = NULL;
782 goto fail;
783 }
784 NDFREE(nd, NDF_ONLY_PNBUF);
785 imgp->vp = nd->ni_vp;
786
787 /*
788 * Check permissions, modes, uid, etc on the file, and "open" it.
789 */
790 error = exec_check_permissions(imgp);
791 if (error)
792 goto fail;
793
794 error = exec_map_first_page(imgp);
795 if (error)
796 goto fail;
797
798 imgp->object = nd->ni_vp->v_object;
799
800 hdr = (const Elf_Ehdr *)imgp->image_header;
801 if ((error = __elfN(check_header)(hdr)) != 0)
802 goto fail;
803 if (hdr->e_type == ET_DYN)
804 rbase = *addr;
805 else if (hdr->e_type == ET_EXEC)
806 rbase = 0;
807 else {
808 error = ENOEXEC;
809 goto fail;
810 }
811
812 /* Only support headers that fit within first page for now */
813 if (!__elfN(phdr_in_zero_page)(hdr)) {
814 error = ENOEXEC;
815 goto fail;
816 }
817
818 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
819 if (!aligned(phdr, Elf_Addr)) {
820 error = ENOEXEC;
821 goto fail;
822 }
823
824 error = __elfN(load_sections)(imgp, hdr, phdr, rbase, &base_addr);
825 if (error != 0)
826 goto fail;
827
828 *addr = base_addr;
829 *entry = (unsigned long)hdr->e_entry + rbase;
830
831 fail:
832 if (imgp->firstpage)
833 exec_unmap_first_page(imgp);
834
835 if (nd->ni_vp) {
836 if (imgp->textset)
837 VOP_UNSET_TEXT_CHECKED(nd->ni_vp);
838 vput(nd->ni_vp);
839 }
840 free(tempdata, M_TEMP);
841
842 return (error);
843 }
844
845 static u_long
846 __CONCAT(rnd_, __elfN(base))(vm_map_t map __unused, u_long minv, u_long maxv,
847 u_int align)
848 {
849 u_long rbase, res;
850
851 MPASS(vm_map_min(map) <= minv);
852 MPASS(maxv <= vm_map_max(map));
853 MPASS(minv < maxv);
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);
858 if (res >= maxv)
859 res -= align;
860 KASSERT(res >= minv,
861 ("res %#lx < minv %#lx, maxv %#lx rbase %#lx",
862 res, minv, maxv, rbase));
863 KASSERT(res < maxv,
864 ("res %#lx > maxv %#lx, minv %#lx rbase %#lx",
865 res, maxv, minv, rbase));
866 return (res);
867 }
868
869 static int
870 __elfN(enforce_limits)(struct image_params *imgp, const Elf_Ehdr *hdr,
871 const Elf_Phdr *phdr, u_long et_dyn_addr)
872 {
873 struct vmspace *vmspace;
874 const char *err_str;
875 u_long text_size, data_size, total_size, text_addr, data_addr;
876 u_long seg_size, seg_addr;
877 int i;
878
879 err_str = NULL;
880 text_size = data_size = total_size = text_addr = data_addr = 0;
881
882 for (i = 0; i < hdr->e_phnum; i++) {
883 if (phdr[i].p_type != PT_LOAD || phdr[i].p_memsz == 0)
884 continue;
885
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);
889
890 /*
891 * Make the largest executable segment the official
892 * text segment and all others data.
893 *
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.
898 */
899
900 if ((phdr[i].p_flags & PF_X) != 0 && text_size < seg_size) {
901 text_size = seg_size;
902 text_addr = seg_addr;
903 } else {
904 data_size = seg_size;
905 data_addr = seg_addr;
906 }
907 total_size += seg_size;
908 }
909
910 if (data_addr == 0 && data_size == 0) {
911 data_addr = text_addr;
912 data_size = text_size;
913 }
914
915 /*
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.
919 */
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);
934 return (ENOMEM);
935 }
936
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;
942
943 return (0);
944 }
945
946 static int
947 __elfN(get_interp)(struct image_params *imgp, const Elf_Phdr *phdr,
948 char **interpp, bool *free_interpp)
949 {
950 struct thread *td;
951 char *interp;
952 int error, interp_name_len;
953
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__);
957
958 td = curthread;
959
960 /* Path to interpreter */
961 if (phdr->p_filesz < 2 || phdr->p_filesz > MAXPATHLEN) {
962 uprintf("Invalid PT_INTERP\n");
963 return (ENOEXEC);
964 }
965
966 interp_name_len = phdr->p_filesz;
967 if (phdr->p_offset > PAGE_SIZE ||
968 interp_name_len > PAGE_SIZE - phdr->p_offset) {
969 /*
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.
977 */
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);
983 }
984
985 error = vn_rdwr(UIO_READ, imgp->vp, interp,
986 interp_name_len, phdr->p_offset,
987 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
988 NOCRED, NULL, td);
989 if (error != 0) {
990 free(interp, M_TEMP);
991 uprintf("i/o error PT_INTERP %d\n", error);
992 return (error);
993 }
994 interp[interp_name_len] = '\0';
995
996 *interpp = interp;
997 *free_interpp = true;
998 return (0);
999 }
1000
1001 interp = __DECONST(char *, imgp->image_header) + phdr->p_offset;
1002 if (interp[interp_name_len - 1] != '\0') {
1003 uprintf("Invalid PT_INTERP\n");
1004 return (ENOEXEC);
1005 }
1006
1007 *interpp = interp;
1008 *free_interpp = false;
1009 return (0);
1010 }
1011
1012 static int
1013 __elfN(load_interp)(struct image_params *imgp, const Elf_Brandinfo *brand_info,
1014 const char *interp, u_long *addr, u_long *entry)
1015 {
1016 char *path;
1017 int error;
1018
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);
1025 free(path, M_TEMP);
1026 if (error == 0)
1027 return (0);
1028 }
1029
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);
1035 if (error == 0)
1036 return (0);
1037 }
1038
1039 error = __elfN(load_file)(imgp->proc, interp, addr, entry);
1040 if (error == 0)
1041 return (0);
1042
1043 uprintf("ELF interpreter %s not found, error %d\n", interp, error);
1044 return (error);
1045 }
1046
1047 /*
1048 * Impossible et_dyn_addr initial value indicating that the real base
1049 * must be calculated later with some randomization applied.
1050 */
1051 #define ET_DYN_ADDR_RAND 1
1052
1053 static int
1054 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
1055 {
1056 struct thread *td;
1057 const Elf_Ehdr *hdr;
1058 const Elf_Phdr *phdr;
1059 Elf_Auxargs *elf_auxargs;
1060 struct vmspace *vmspace;
1061 vm_map_t map;
1062 char *interp;
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;
1067 uint32_t fctl0;
1068 int32_t osrel;
1069 bool free_interp;
1070 int error, i, n;
1071
1072 hdr = (const Elf_Ehdr *)imgp->image_header;
1073
1074 /*
1075 * Do we have a valid ELF header ?
1076 *
1077 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
1078 * if particular brand doesn't support it.
1079 */
1080 if (__elfN(check_header)(hdr) != 0 ||
1081 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
1082 return (-1);
1083
1084 /*
1085 * From here on down, we return an errno, not -1, as we've
1086 * detected an ELF file.
1087 */
1088
1089 if (!__elfN(phdr_in_zero_page)(hdr)) {
1090 uprintf("Program headers not in the first page\n");
1091 return (ENOEXEC);
1092 }
1093 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1094 if (!aligned(phdr, Elf_Addr)) {
1095 uprintf("Unaligned program headers\n");
1096 return (ENOEXEC);
1097 }
1098
1099 n = error = 0;
1100 baddr = 0;
1101 osrel = 0;
1102 fctl0 = 0;
1103 entry = proghdr = 0;
1104 interp = NULL;
1105 free_interp = false;
1106 td = curthread;
1107 maxalign = PAGE_SIZE;
1108 mapsz = 0;
1109
1110 for (i = 0; i < hdr->e_phnum; i++) {
1111 switch (phdr[i].p_type) {
1112 case PT_LOAD:
1113 if (n == 0)
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;
1118 n++;
1119
1120 /*
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
1124 * a PT_PHDR entry.
1125 */
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;
1130 break;
1131 case PT_INTERP:
1132 /* Path to interpreter */
1133 if (interp != NULL) {
1134 uprintf("Multiple PT_INTERP headers\n");
1135 error = ENOEXEC;
1136 goto ret;
1137 }
1138 error = __elfN(get_interp)(imgp, &phdr[i], &interp,
1139 &free_interp);
1140 if (error != 0)
1141 goto ret;
1142 break;
1143 case PT_GNU_STACK:
1144 if (__elfN(nxstack)) {
1145 imgp->stack_prot =
1146 __elfN(trans_prot)(phdr[i].p_flags);
1147 if ((imgp->stack_prot & VM_PROT_RW) !=
1148 VM_PROT_RW) {
1149 uprintf("Invalid PT_GNU_STACK\n");
1150 error = ENOEXEC;
1151 goto ret;
1152 }
1153 }
1154 imgp->stack_sz = phdr[i].p_memsz;
1155 break;
1156 case PT_PHDR: /* Program header table info */
1157 proghdr = phdr[i].p_vaddr;
1158 break;
1159 }
1160 }
1161
1162 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel, &fctl0);
1163 if (brand_info == NULL) {
1164 uprintf("ELF binary type \"%u\" not known.\n",
1165 hdr->e_ident[EI_OSABI]);
1166 error = ENOEXEC;
1167 goto ret;
1168 }
1169 sv = brand_info->sysvec;
1170 et_dyn_addr = 0;
1171 if (hdr->e_type == ET_DYN) {
1172 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
1173 uprintf("Cannot execute shared object\n");
1174 error = ENOEXEC;
1175 goto ret;
1176 }
1177 /*
1178 * Honour the base load address from the dso if it is
1179 * non-zero for some reason.
1180 */
1181 if (baddr == 0) {
1182 if ((sv->sv_flags & SV_ASLR) == 0 ||
1183 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0)
1184 et_dyn_addr = __elfN(pie_base);
1185 else if ((__elfN(pie_aslr_enabled) &&
1186 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) == 0) ||
1187 (imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0)
1188 et_dyn_addr = ET_DYN_ADDR_RAND;
1189 else
1190 et_dyn_addr = __elfN(pie_base);
1191 }
1192 }
1193
1194 /*
1195 * Avoid a possible deadlock if the current address space is destroyed
1196 * and that address space maps the locked vnode. In the common case,
1197 * the locked vnode's v_usecount is decremented but remains greater
1198 * than zero. Consequently, the vnode lock is not needed by vrele().
1199 * However, in cases where the vnode lock is external, such as nullfs,
1200 * v_usecount may become zero.
1201 *
1202 * The VV_TEXT flag prevents modifications to the executable while
1203 * the vnode is unlocked.
1204 */
1205 VOP_UNLOCK(imgp->vp, 0);
1206
1207 /*
1208 * Decide whether to enable randomization of user mappings.
1209 * First, reset user preferences for the setid binaries.
1210 * Then, account for the support of the randomization by the
1211 * ABI, by user preferences, and make special treatment for
1212 * PIE binaries.
1213 */
1214 if (imgp->credential_setid) {
1215 PROC_LOCK(imgp->proc);
1216 imgp->proc->p_flag2 &= ~(P2_ASLR_ENABLE | P2_ASLR_DISABLE);
1217 PROC_UNLOCK(imgp->proc);
1218 }
1219 if ((sv->sv_flags & SV_ASLR) == 0 ||
1220 (imgp->proc->p_flag2 & P2_ASLR_DISABLE) != 0 ||
1221 (fctl0 & NT_FREEBSD_FCTL_ASLR_DISABLE) != 0) {
1222 KASSERT(et_dyn_addr != ET_DYN_ADDR_RAND,
1223 ("et_dyn_addr == RAND and !ASLR"));
1224 } else if ((imgp->proc->p_flag2 & P2_ASLR_ENABLE) != 0 ||
1225 (__elfN(aslr_enabled) && hdr->e_type == ET_EXEC) ||
1226 et_dyn_addr == ET_DYN_ADDR_RAND) {
1227 imgp->map_flags |= MAP_ASLR;
1228 /*
1229 * If user does not care about sbrk, utilize the bss
1230 * grow region for mappings as well. We can select
1231 * the base for the image anywere and still not suffer
1232 * from the fragmentation.
1233 */
1234 if (!__elfN(aslr_honor_sbrk) ||
1235 (imgp->proc->p_flag2 & P2_ASLR_IGNSTART) != 0)
1236 imgp->map_flags |= MAP_ASLR_IGNSTART;
1237 }
1238
1239 error = exec_new_vmspace(imgp, sv);
1240 vmspace = imgp->proc->p_vmspace;
1241 map = &vmspace->vm_map;
1242
1243 imgp->proc->p_sysent = sv;
1244
1245 maxv = vm_map_max(map) - lim_max(td, RLIMIT_STACK);
1246 if (et_dyn_addr == ET_DYN_ADDR_RAND) {
1247 KASSERT((map->flags & MAP_ASLR) != 0,
1248 ("ET_DYN_ADDR_RAND but !MAP_ASLR"));
1249 et_dyn_addr = __CONCAT(rnd_, __elfN(base))(map,
1250 vm_map_min(map) + mapsz + lim_max(td, RLIMIT_DATA),
1251 /* reserve half of the address space to interpreter */
1252 maxv / 2, 1UL << flsl(maxalign));
1253 }
1254
1255 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1256 if (error != 0)
1257 goto ret;
1258
1259 error = __elfN(load_sections)(imgp, hdr, phdr, et_dyn_addr, NULL);
1260 if (error != 0)
1261 goto ret;
1262
1263 error = __elfN(enforce_limits)(imgp, hdr, phdr, et_dyn_addr);
1264 if (error != 0)
1265 goto ret;
1266
1267 entry = (u_long)hdr->e_entry + et_dyn_addr;
1268
1269 /*
1270 * We load the dynamic linker where a userland call
1271 * to mmap(0, ...) would put it. The rationale behind this
1272 * calculation is that it leaves room for the heap to grow to
1273 * its maximum allowed size.
1274 */
1275 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1276 RLIMIT_DATA));
1277 if ((map->flags & MAP_ASLR) != 0) {
1278 maxv1 = maxv / 2 + addr / 2;
1279 MPASS(maxv1 >= addr); /* No overflow */
1280 map->anon_loc = __CONCAT(rnd_, __elfN(base))(map, addr, maxv1,
1281 (MAXPAGESIZES > 1 && pagesizes[1] != 0) ?
1282 pagesizes[1] : pagesizes[0]);
1283 } else {
1284 map->anon_loc = addr;
1285 }
1286
1287 imgp->entry_addr = entry;
1288
1289 if (interp != NULL) {
1290 VOP_UNLOCK(imgp->vp, 0);
1291 if ((map->flags & MAP_ASLR) != 0) {
1292 /* Assume that interpreter fits into 1/4 of AS */
1293 maxv1 = maxv / 2 + addr / 2;
1294 MPASS(maxv1 >= addr); /* No overflow */
1295 addr = __CONCAT(rnd_, __elfN(base))(map, addr,
1296 maxv1, PAGE_SIZE);
1297 }
1298 error = __elfN(load_interp)(imgp, brand_info, interp, &addr,
1299 &imgp->entry_addr);
1300 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1301 if (error != 0)
1302 goto ret;
1303 } else
1304 addr = et_dyn_addr;
1305
1306 /*
1307 * Construct auxargs table (used by the fixup routine)
1308 */
1309 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_NOWAIT);
1310 if (elf_auxargs == NULL) {
1311 VOP_UNLOCK(imgp->vp, 0);
1312 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1313 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
1314 }
1315 elf_auxargs->execfd = -1;
1316 elf_auxargs->phdr = proghdr + et_dyn_addr;
1317 elf_auxargs->phent = hdr->e_phentsize;
1318 elf_auxargs->phnum = hdr->e_phnum;
1319 elf_auxargs->pagesz = PAGE_SIZE;
1320 elf_auxargs->base = addr;
1321 elf_auxargs->flags = 0;
1322 elf_auxargs->entry = entry;
1323 elf_auxargs->hdr_eflags = hdr->e_flags;
1324
1325 imgp->auxargs = elf_auxargs;
1326 imgp->interpreted = 0;
1327 imgp->reloc_base = addr;
1328 imgp->proc->p_osrel = osrel;
1329 imgp->proc->p_fctl0 = fctl0;
1330 imgp->proc->p_elf_machine = hdr->e_machine;
1331 imgp->proc->p_elf_flags = hdr->e_flags;
1332
1333 ret:
1334 if (free_interp)
1335 free(interp, M_TEMP);
1336 return (error);
1337 }
1338
1339 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1340
1341 int
1342 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1343 {
1344 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1345 Elf_Auxinfo *argarray, *pos;
1346 Elf_Addr *base, *auxbase;
1347 int error;
1348
1349 base = (Elf_Addr *)*stack_base;
1350 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
1351 argarray = pos = malloc(AT_COUNT * sizeof(*pos), M_TEMP,
1352 M_WAITOK | M_ZERO);
1353
1354 if (args->execfd != -1)
1355 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1356 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1357 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1358 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1359 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1360 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1361 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1362 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1363 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1364 if (imgp->execpathp != 0)
1365 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1366 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1367 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1368 if (imgp->canary != 0) {
1369 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1370 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1371 }
1372 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1373 if (imgp->pagesizes != 0) {
1374 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1375 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1376 }
1377 if (imgp->sysent->sv_timekeep_base != 0) {
1378 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1379 imgp->sysent->sv_timekeep_base);
1380 }
1381 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1382 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1383 imgp->sysent->sv_stackprot);
1384 if (imgp->sysent->sv_hwcap != NULL)
1385 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1386 if (imgp->sysent->sv_hwcap2 != NULL)
1387 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1388 AUXARGS_ENTRY(pos, AT_NULL, 0);
1389
1390 free(imgp->auxargs, M_TEMP);
1391 imgp->auxargs = NULL;
1392 KASSERT(pos - argarray <= AT_COUNT, ("Too many auxargs"));
1393
1394 error = copyout(argarray, auxbase, sizeof(*argarray) * AT_COUNT);
1395 free(argarray, M_TEMP);
1396 if (error != 0)
1397 return (error);
1398
1399 base--;
1400 if (suword(base, imgp->args->argc) == -1)
1401 return (EFAULT);
1402 *stack_base = (register_t *)base;
1403 return (0);
1404 }
1405
1406 /*
1407 * Code for generating ELF core dumps.
1408 */
1409
1410 typedef void (*segment_callback)(vm_map_entry_t, void *);
1411
1412 /* Closure for cb_put_phdr(). */
1413 struct phdr_closure {
1414 Elf_Phdr *phdr; /* Program header to fill in */
1415 Elf_Off offset; /* Offset of segment in core file */
1416 };
1417
1418 /* Closure for cb_size_segment(). */
1419 struct sseg_closure {
1420 int count; /* Count of writable segments. */
1421 size_t size; /* Total size of all writable segments. */
1422 };
1423
1424 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1425
1426 struct note_info {
1427 int type; /* Note type. */
1428 outfunc_t outfunc; /* Output function. */
1429 void *outarg; /* Argument for the output function. */
1430 size_t outsize; /* Output size. */
1431 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1432 };
1433
1434 TAILQ_HEAD(note_info_list, note_info);
1435
1436 /* Coredump output parameters. */
1437 struct coredump_params {
1438 off_t offset;
1439 struct ucred *active_cred;
1440 struct ucred *file_cred;
1441 struct thread *td;
1442 struct vnode *vp;
1443 struct compressor *comp;
1444 };
1445
1446 extern int compress_user_cores;
1447 extern int compress_user_cores_level;
1448
1449 static void cb_put_phdr(vm_map_entry_t, void *);
1450 static void cb_size_segment(vm_map_entry_t, void *);
1451 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1452 enum uio_seg);
1453 static void each_dumpable_segment(struct thread *, segment_callback, void *);
1454 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1455 struct note_info_list *, size_t);
1456 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1457 size_t *);
1458 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1459 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1460 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1461 static int sbuf_drain_core_output(void *, const char *, int);
1462
1463 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1464 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1465 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1466 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1467 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1468 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1469 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1470 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1471 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1472 static void note_procstat_files(void *, struct sbuf *, size_t *);
1473 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1474 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1475 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1476 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1477 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1478
1479 /*
1480 * Write out a core segment to the compression stream.
1481 */
1482 static int
1483 compress_chunk(struct coredump_params *p, char *base, char *buf, size_t len)
1484 {
1485 size_t chunk_len;
1486 int error;
1487
1488 while (len > 0) {
1489 chunk_len = MIN(len, CORE_BUF_SIZE);
1490
1491 /*
1492 * We can get EFAULT error here.
1493 * In that case zero out the current chunk of the segment.
1494 */
1495 error = copyin(base, buf, chunk_len);
1496 if (error != 0)
1497 bzero(buf, chunk_len);
1498 error = compressor_write(p->comp, buf, chunk_len);
1499 if (error != 0)
1500 break;
1501 base += chunk_len;
1502 len -= chunk_len;
1503 }
1504 return (error);
1505 }
1506
1507 static int
1508 core_compressed_write(void *base, size_t len, off_t offset, void *arg)
1509 {
1510
1511 return (core_write((struct coredump_params *)arg, base, len, offset,
1512 UIO_SYSSPACE));
1513 }
1514
1515 static int
1516 core_write(struct coredump_params *p, const void *base, size_t len,
1517 off_t offset, enum uio_seg seg)
1518 {
1519
1520 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1521 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1522 p->active_cred, p->file_cred, NULL, p->td));
1523 }
1524
1525 static int
1526 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1527 void *tmpbuf)
1528 {
1529 int error;
1530
1531 if (p->comp != NULL)
1532 return (compress_chunk(p, base, tmpbuf, len));
1533
1534 /*
1535 * EFAULT is a non-fatal error that we can get, for example,
1536 * if the segment is backed by a file but extends beyond its
1537 * end.
1538 */
1539 error = core_write(p, base, len, offset, UIO_USERSPACE);
1540 if (error == EFAULT) {
1541 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1542 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1543 "for process %s\n", base, len, offset, curproc->p_comm);
1544
1545 /*
1546 * Write a "real" zero byte at the end of the target region
1547 * in the case this is the last segment.
1548 * The intermediate space will be implicitly zero-filled.
1549 */
1550 error = core_write(p, zero_region, 1, offset + len - 1,
1551 UIO_SYSSPACE);
1552 }
1553 return (error);
1554 }
1555
1556 /*
1557 * Drain into a core file.
1558 */
1559 static int
1560 sbuf_drain_core_output(void *arg, const char *data, int len)
1561 {
1562 struct coredump_params *p;
1563 int error, locked;
1564
1565 p = (struct coredump_params *)arg;
1566
1567 /*
1568 * Some kern_proc out routines that print to this sbuf may
1569 * call us with the process lock held. Draining with the
1570 * non-sleepable lock held is unsafe. The lock is needed for
1571 * those routines when dumping a live process. In our case we
1572 * can safely release the lock before draining and acquire
1573 * again after.
1574 */
1575 locked = PROC_LOCKED(p->td->td_proc);
1576 if (locked)
1577 PROC_UNLOCK(p->td->td_proc);
1578 if (p->comp != NULL)
1579 error = compressor_write(p->comp, __DECONST(char *, data), len);
1580 else
1581 error = core_write(p, __DECONST(void *, data), len, p->offset,
1582 UIO_SYSSPACE);
1583 if (locked)
1584 PROC_LOCK(p->td->td_proc);
1585 if (error != 0)
1586 return (-error);
1587 p->offset += len;
1588 return (len);
1589 }
1590
1591 int
1592 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1593 {
1594 struct ucred *cred = td->td_ucred;
1595 int error = 0;
1596 struct sseg_closure seginfo;
1597 struct note_info_list notelst;
1598 struct coredump_params params;
1599 struct note_info *ninfo;
1600 void *hdr, *tmpbuf;
1601 size_t hdrsize, notesz, coresize;
1602
1603 hdr = NULL;
1604 tmpbuf = NULL;
1605 TAILQ_INIT(¬elst);
1606
1607 /* Size the program segments. */
1608 seginfo.count = 0;
1609 seginfo.size = 0;
1610 each_dumpable_segment(td, cb_size_segment, &seginfo);
1611
1612 /*
1613 * Collect info about the core file header area.
1614 */
1615 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1616 if (seginfo.count + 1 >= PN_XNUM)
1617 hdrsize += sizeof(Elf_Shdr);
1618 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1619 coresize = round_page(hdrsize + notesz) + seginfo.size;
1620
1621 /* Set up core dump parameters. */
1622 params.offset = 0;
1623 params.active_cred = cred;
1624 params.file_cred = NOCRED;
1625 params.td = td;
1626 params.vp = vp;
1627 params.comp = NULL;
1628
1629 #ifdef RACCT
1630 if (racct_enable) {
1631 PROC_LOCK(td->td_proc);
1632 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1633 PROC_UNLOCK(td->td_proc);
1634 if (error != 0) {
1635 error = EFAULT;
1636 goto done;
1637 }
1638 }
1639 #endif
1640 if (coresize >= limit) {
1641 error = EFAULT;
1642 goto done;
1643 }
1644
1645 /* Create a compression stream if necessary. */
1646 if (compress_user_cores != 0) {
1647 params.comp = compressor_init(core_compressed_write,
1648 compress_user_cores, CORE_BUF_SIZE,
1649 compress_user_cores_level, ¶ms);
1650 if (params.comp == NULL) {
1651 error = EFAULT;
1652 goto done;
1653 }
1654 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1655 }
1656
1657 /*
1658 * Allocate memory for building the header, fill it up,
1659 * and write it out following the notes.
1660 */
1661 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1662 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1663 notesz);
1664
1665 /* Write the contents of all of the writable segments. */
1666 if (error == 0) {
1667 Elf_Phdr *php;
1668 off_t offset;
1669 int i;
1670
1671 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1672 offset = round_page(hdrsize + notesz);
1673 for (i = 0; i < seginfo.count; i++) {
1674 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1675 php->p_filesz, offset, ¶ms, tmpbuf);
1676 if (error != 0)
1677 break;
1678 offset += php->p_filesz;
1679 php++;
1680 }
1681 if (error == 0 && params.comp != NULL)
1682 error = compressor_flush(params.comp);
1683 }
1684 if (error) {
1685 log(LOG_WARNING,
1686 "Failed to write core file for process %s (error %d)\n",
1687 curproc->p_comm, error);
1688 }
1689
1690 done:
1691 free(tmpbuf, M_TEMP);
1692 if (params.comp != NULL)
1693 compressor_fini(params.comp);
1694 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1695 TAILQ_REMOVE(¬elst, ninfo, link);
1696 free(ninfo, M_TEMP);
1697 }
1698 if (hdr != NULL)
1699 free(hdr, M_TEMP);
1700
1701 return (error);
1702 }
1703
1704 /*
1705 * A callback for each_dumpable_segment() to write out the segment's
1706 * program header entry.
1707 */
1708 static void
1709 cb_put_phdr(vm_map_entry_t entry, void *closure)
1710 {
1711 struct phdr_closure *phc = (struct phdr_closure *)closure;
1712 Elf_Phdr *phdr = phc->phdr;
1713
1714 phc->offset = round_page(phc->offset);
1715
1716 phdr->p_type = PT_LOAD;
1717 phdr->p_offset = phc->offset;
1718 phdr->p_vaddr = entry->start;
1719 phdr->p_paddr = 0;
1720 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1721 phdr->p_align = PAGE_SIZE;
1722 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1723
1724 phc->offset += phdr->p_filesz;
1725 phc->phdr++;
1726 }
1727
1728 /*
1729 * A callback for each_dumpable_segment() to gather information about
1730 * the number of segments and their total size.
1731 */
1732 static void
1733 cb_size_segment(vm_map_entry_t entry, void *closure)
1734 {
1735 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1736
1737 ssc->count++;
1738 ssc->size += entry->end - entry->start;
1739 }
1740
1741 /*
1742 * For each writable segment in the process's memory map, call the given
1743 * function with a pointer to the map entry and some arbitrary
1744 * caller-supplied data.
1745 */
1746 static void
1747 each_dumpable_segment(struct thread *td, segment_callback func, void *closure)
1748 {
1749 struct proc *p = td->td_proc;
1750 vm_map_t map = &p->p_vmspace->vm_map;
1751 vm_map_entry_t entry;
1752 vm_object_t backing_object, object;
1753 bool ignore_entry;
1754
1755 vm_map_lock_read(map);
1756 for (entry = map->header.next; entry != &map->header;
1757 entry = entry->next) {
1758 /*
1759 * Don't dump inaccessible mappings, deal with legacy
1760 * coredump mode.
1761 *
1762 * Note that read-only segments related to the elf binary
1763 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1764 * need to arbitrarily ignore such segments.
1765 */
1766 if (elf_legacy_coredump) {
1767 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1768 continue;
1769 } else {
1770 if ((entry->protection & VM_PROT_ALL) == 0)
1771 continue;
1772 }
1773
1774 /*
1775 * Dont include memory segment in the coredump if
1776 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1777 * madvise(2). Do not dump submaps (i.e. parts of the
1778 * kernel map).
1779 */
1780 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1781 continue;
1782
1783 if ((object = entry->object.vm_object) == NULL)
1784 continue;
1785
1786 /* Ignore memory-mapped devices and such things. */
1787 VM_OBJECT_RLOCK(object);
1788 while ((backing_object = object->backing_object) != NULL) {
1789 VM_OBJECT_RLOCK(backing_object);
1790 VM_OBJECT_RUNLOCK(object);
1791 object = backing_object;
1792 }
1793 ignore_entry = (object->flags & OBJ_FICTITIOUS) != 0;
1794 VM_OBJECT_RUNLOCK(object);
1795 if (ignore_entry)
1796 continue;
1797
1798 (*func)(entry, closure);
1799 }
1800 vm_map_unlock_read(map);
1801 }
1802
1803 /*
1804 * Write the core file header to the file, including padding up to
1805 * the page boundary.
1806 */
1807 static int
1808 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1809 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1810 {
1811 struct note_info *ninfo;
1812 struct sbuf *sb;
1813 int error;
1814
1815 /* Fill in the header. */
1816 bzero(hdr, hdrsize);
1817 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1818
1819 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1820 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1821 sbuf_start_section(sb, NULL);
1822 sbuf_bcat(sb, hdr, hdrsize);
1823 TAILQ_FOREACH(ninfo, notelst, link)
1824 __elfN(putnote)(ninfo, sb);
1825 /* Align up to a page boundary for the program segments. */
1826 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1827 error = sbuf_finish(sb);
1828 sbuf_delete(sb);
1829
1830 return (error);
1831 }
1832
1833 static void
1834 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1835 size_t *sizep)
1836 {
1837 struct proc *p;
1838 struct thread *thr;
1839 size_t size;
1840
1841 p = td->td_proc;
1842 size = 0;
1843
1844 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1845
1846 /*
1847 * To have the debugger select the right thread (LWP) as the initial
1848 * thread, we dump the state of the thread passed to us in td first.
1849 * This is the thread that causes the core dump and thus likely to
1850 * be the right thread one wants to have selected in the debugger.
1851 */
1852 thr = td;
1853 while (thr != NULL) {
1854 size += register_note(list, NT_PRSTATUS,
1855 __elfN(note_prstatus), thr);
1856 size += register_note(list, NT_FPREGSET,
1857 __elfN(note_fpregset), thr);
1858 size += register_note(list, NT_THRMISC,
1859 __elfN(note_thrmisc), thr);
1860 size += register_note(list, NT_PTLWPINFO,
1861 __elfN(note_ptlwpinfo), thr);
1862 size += register_note(list, -1,
1863 __elfN(note_threadmd), thr);
1864
1865 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1866 TAILQ_NEXT(thr, td_plist);
1867 if (thr == td)
1868 thr = TAILQ_NEXT(thr, td_plist);
1869 }
1870
1871 size += register_note(list, NT_PROCSTAT_PROC,
1872 __elfN(note_procstat_proc), p);
1873 size += register_note(list, NT_PROCSTAT_FILES,
1874 note_procstat_files, p);
1875 size += register_note(list, NT_PROCSTAT_VMMAP,
1876 note_procstat_vmmap, p);
1877 size += register_note(list, NT_PROCSTAT_GROUPS,
1878 note_procstat_groups, p);
1879 size += register_note(list, NT_PROCSTAT_UMASK,
1880 note_procstat_umask, p);
1881 size += register_note(list, NT_PROCSTAT_RLIMIT,
1882 note_procstat_rlimit, p);
1883 size += register_note(list, NT_PROCSTAT_OSREL,
1884 note_procstat_osrel, p);
1885 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1886 __elfN(note_procstat_psstrings), p);
1887 size += register_note(list, NT_PROCSTAT_AUXV,
1888 __elfN(note_procstat_auxv), p);
1889
1890 *sizep = size;
1891 }
1892
1893 static void
1894 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1895 size_t notesz)
1896 {
1897 Elf_Ehdr *ehdr;
1898 Elf_Phdr *phdr;
1899 Elf_Shdr *shdr;
1900 struct phdr_closure phc;
1901
1902 ehdr = (Elf_Ehdr *)hdr;
1903
1904 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1905 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1906 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1907 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1908 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1909 ehdr->e_ident[EI_DATA] = ELF_DATA;
1910 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1911 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1912 ehdr->e_ident[EI_ABIVERSION] = 0;
1913 ehdr->e_ident[EI_PAD] = 0;
1914 ehdr->e_type = ET_CORE;
1915 ehdr->e_machine = td->td_proc->p_elf_machine;
1916 ehdr->e_version = EV_CURRENT;
1917 ehdr->e_entry = 0;
1918 ehdr->e_phoff = sizeof(Elf_Ehdr);
1919 ehdr->e_flags = td->td_proc->p_elf_flags;
1920 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1921 ehdr->e_phentsize = sizeof(Elf_Phdr);
1922 ehdr->e_shentsize = sizeof(Elf_Shdr);
1923 ehdr->e_shstrndx = SHN_UNDEF;
1924 if (numsegs + 1 < PN_XNUM) {
1925 ehdr->e_phnum = numsegs + 1;
1926 ehdr->e_shnum = 0;
1927 } else {
1928 ehdr->e_phnum = PN_XNUM;
1929 ehdr->e_shnum = 1;
1930
1931 ehdr->e_shoff = ehdr->e_phoff +
1932 (numsegs + 1) * ehdr->e_phentsize;
1933 KASSERT(ehdr->e_shoff == hdrsize - sizeof(Elf_Shdr),
1934 ("e_shoff: %zu, hdrsize - shdr: %zu",
1935 (size_t)ehdr->e_shoff, hdrsize - sizeof(Elf_Shdr)));
1936
1937 shdr = (Elf_Shdr *)((char *)hdr + ehdr->e_shoff);
1938 memset(shdr, 0, sizeof(*shdr));
1939 /*
1940 * A special first section is used to hold large segment and
1941 * section counts. This was proposed by Sun Microsystems in
1942 * Solaris and has been adopted by Linux; the standard ELF
1943 * tools are already familiar with the technique.
1944 *
1945 * See table 7-7 of the Solaris "Linker and Libraries Guide"
1946 * (or 12-7 depending on the version of the document) for more
1947 * details.
1948 */
1949 shdr->sh_type = SHT_NULL;
1950 shdr->sh_size = ehdr->e_shnum;
1951 shdr->sh_link = ehdr->e_shstrndx;
1952 shdr->sh_info = numsegs + 1;
1953 }
1954
1955 /*
1956 * Fill in the program header entries.
1957 */
1958 phdr = (Elf_Phdr *)((char *)hdr + ehdr->e_phoff);
1959
1960 /* The note segement. */
1961 phdr->p_type = PT_NOTE;
1962 phdr->p_offset = hdrsize;
1963 phdr->p_vaddr = 0;
1964 phdr->p_paddr = 0;
1965 phdr->p_filesz = notesz;
1966 phdr->p_memsz = 0;
1967 phdr->p_flags = PF_R;
1968 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1969 phdr++;
1970
1971 /* All the writable segments from the program. */
1972 phc.phdr = phdr;
1973 phc.offset = round_page(hdrsize + notesz);
1974 each_dumpable_segment(td, cb_put_phdr, &phc);
1975 }
1976
1977 static size_t
1978 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1979 {
1980 struct note_info *ninfo;
1981 size_t size, notesize;
1982
1983 size = 0;
1984 out(arg, NULL, &size);
1985 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1986 ninfo->type = type;
1987 ninfo->outfunc = out;
1988 ninfo->outarg = arg;
1989 ninfo->outsize = size;
1990 TAILQ_INSERT_TAIL(list, ninfo, link);
1991
1992 if (type == -1)
1993 return (size);
1994
1995 notesize = sizeof(Elf_Note) + /* note header */
1996 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1997 /* note name */
1998 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1999
2000 return (notesize);
2001 }
2002
2003 static size_t
2004 append_note_data(const void *src, void *dst, size_t len)
2005 {
2006 size_t padded_len;
2007
2008 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
2009 if (dst != NULL) {
2010 bcopy(src, dst, len);
2011 bzero((char *)dst + len, padded_len - len);
2012 }
2013 return (padded_len);
2014 }
2015
2016 size_t
2017 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
2018 {
2019 Elf_Note *note;
2020 char *buf;
2021 size_t notesize;
2022
2023 buf = dst;
2024 if (buf != NULL) {
2025 note = (Elf_Note *)buf;
2026 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2027 note->n_descsz = size;
2028 note->n_type = type;
2029 buf += sizeof(*note);
2030 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
2031 sizeof(FREEBSD_ABI_VENDOR));
2032 append_note_data(src, buf, size);
2033 if (descp != NULL)
2034 *descp = buf;
2035 }
2036
2037 notesize = sizeof(Elf_Note) + /* note header */
2038 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
2039 /* note name */
2040 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
2041
2042 return (notesize);
2043 }
2044
2045 static void
2046 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
2047 {
2048 Elf_Note note;
2049 ssize_t old_len, sect_len;
2050 size_t new_len, descsz, i;
2051
2052 if (ninfo->type == -1) {
2053 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2054 return;
2055 }
2056
2057 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
2058 note.n_descsz = ninfo->outsize;
2059 note.n_type = ninfo->type;
2060
2061 sbuf_bcat(sb, ¬e, sizeof(note));
2062 sbuf_start_section(sb, &old_len);
2063 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
2064 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2065 if (note.n_descsz == 0)
2066 return;
2067 sbuf_start_section(sb, &old_len);
2068 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
2069 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
2070 if (sect_len < 0)
2071 return;
2072
2073 new_len = (size_t)sect_len;
2074 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
2075 if (new_len < descsz) {
2076 /*
2077 * It is expected that individual note emitters will correctly
2078 * predict their expected output size and fill up to that size
2079 * themselves, padding in a format-specific way if needed.
2080 * However, in case they don't, just do it here with zeros.
2081 */
2082 for (i = 0; i < descsz - new_len; i++)
2083 sbuf_putc(sb, 0);
2084 } else if (new_len > descsz) {
2085 /*
2086 * We can't always truncate sb -- we may have drained some
2087 * of it already.
2088 */
2089 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
2090 "read it (%zu > %zu). Since it is longer than "
2091 "expected, this coredump's notes are corrupt. THIS "
2092 "IS A BUG in the note_procstat routine for type %u.\n",
2093 __func__, (unsigned)note.n_type, new_len, descsz,
2094 (unsigned)note.n_type));
2095 }
2096 }
2097
2098 /*
2099 * Miscellaneous note out functions.
2100 */
2101
2102 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2103 #include <compat/freebsd32/freebsd32.h>
2104 #include <compat/freebsd32/freebsd32_signal.h>
2105
2106 typedef struct prstatus32 elf_prstatus_t;
2107 typedef struct prpsinfo32 elf_prpsinfo_t;
2108 typedef struct fpreg32 elf_prfpregset_t;
2109 typedef struct fpreg32 elf_fpregset_t;
2110 typedef struct reg32 elf_gregset_t;
2111 typedef struct thrmisc32 elf_thrmisc_t;
2112 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
2113 typedef struct kinfo_proc32 elf_kinfo_proc_t;
2114 typedef uint32_t elf_ps_strings_t;
2115 #else
2116 typedef prstatus_t elf_prstatus_t;
2117 typedef prpsinfo_t elf_prpsinfo_t;
2118 typedef prfpregset_t elf_prfpregset_t;
2119 typedef prfpregset_t elf_fpregset_t;
2120 typedef gregset_t elf_gregset_t;
2121 typedef thrmisc_t elf_thrmisc_t;
2122 #define ELF_KERN_PROC_MASK 0
2123 typedef struct kinfo_proc elf_kinfo_proc_t;
2124 typedef vm_offset_t elf_ps_strings_t;
2125 #endif
2126
2127 static void
2128 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2129 {
2130 struct sbuf sbarg;
2131 size_t len;
2132 char *cp, *end;
2133 struct proc *p;
2134 elf_prpsinfo_t *psinfo;
2135 int error;
2136
2137 p = (struct proc *)arg;
2138 if (sb != NULL) {
2139 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
2140 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
2141 psinfo->pr_version = PRPSINFO_VERSION;
2142 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
2143 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
2144 PROC_LOCK(p);
2145 if (p->p_args != NULL) {
2146 len = sizeof(psinfo->pr_psargs) - 1;
2147 if (len > p->p_args->ar_length)
2148 len = p->p_args->ar_length;
2149 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
2150 PROC_UNLOCK(p);
2151 error = 0;
2152 } else {
2153 _PHOLD(p);
2154 PROC_UNLOCK(p);
2155 sbuf_new(&sbarg, psinfo->pr_psargs,
2156 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
2157 error = proc_getargv(curthread, p, &sbarg);
2158 PRELE(p);
2159 if (sbuf_finish(&sbarg) == 0) {
2160 len = sbuf_len(&sbarg);
2161 if (len > 0)
2162 len--;
2163 } else {
2164 len = sizeof(psinfo->pr_psargs) - 1;
2165 }
2166 sbuf_delete(&sbarg);
2167 }
2168 if (error != 0 || len == 0 || (ssize_t)len == -1)
2169 strlcpy(psinfo->pr_psargs, p->p_comm,
2170 sizeof(psinfo->pr_psargs));
2171 else {
2172 KASSERT(len < sizeof(psinfo->pr_psargs),
2173 ("len is too long: %zu vs %zu", len,
2174 sizeof(psinfo->pr_psargs)));
2175 cp = psinfo->pr_psargs;
2176 end = cp + len - 1;
2177 for (;;) {
2178 cp = memchr(cp, '\0', end - cp);
2179 if (cp == NULL)
2180 break;
2181 *cp = ' ';
2182 }
2183 }
2184 psinfo->pr_pid = p->p_pid;
2185 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
2186 free(psinfo, M_TEMP);
2187 }
2188 *sizep = sizeof(*psinfo);
2189 }
2190
2191 static void
2192 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
2193 {
2194 struct thread *td;
2195 elf_prstatus_t *status;
2196
2197 td = (struct thread *)arg;
2198 if (sb != NULL) {
2199 KASSERT(*sizep == sizeof(*status), ("invalid size"));
2200 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
2201 status->pr_version = PRSTATUS_VERSION;
2202 status->pr_statussz = sizeof(elf_prstatus_t);
2203 status->pr_gregsetsz = sizeof(elf_gregset_t);
2204 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
2205 status->pr_osreldate = osreldate;
2206 status->pr_cursig = td->td_proc->p_sig;
2207 status->pr_pid = td->td_tid;
2208 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2209 fill_regs32(td, &status->pr_reg);
2210 #else
2211 fill_regs(td, &status->pr_reg);
2212 #endif
2213 sbuf_bcat(sb, status, sizeof(*status));
2214 free(status, M_TEMP);
2215 }
2216 *sizep = sizeof(*status);
2217 }
2218
2219 static void
2220 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
2221 {
2222 struct thread *td;
2223 elf_prfpregset_t *fpregset;
2224
2225 td = (struct thread *)arg;
2226 if (sb != NULL) {
2227 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
2228 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
2229 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2230 fill_fpregs32(td, fpregset);
2231 #else
2232 fill_fpregs(td, fpregset);
2233 #endif
2234 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
2235 free(fpregset, M_TEMP);
2236 }
2237 *sizep = sizeof(*fpregset);
2238 }
2239
2240 static void
2241 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2242 {
2243 struct thread *td;
2244 elf_thrmisc_t thrmisc;
2245
2246 td = (struct thread *)arg;
2247 if (sb != NULL) {
2248 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2249 bzero(&thrmisc, sizeof(thrmisc));
2250 strcpy(thrmisc.pr_tname, td->td_name);
2251 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2252 }
2253 *sizep = sizeof(thrmisc);
2254 }
2255
2256 static void
2257 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2258 {
2259 struct thread *td;
2260 size_t size;
2261 int structsize;
2262 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2263 struct ptrace_lwpinfo32 pl;
2264 #else
2265 struct ptrace_lwpinfo pl;
2266 #endif
2267
2268 td = (struct thread *)arg;
2269 size = sizeof(structsize) + sizeof(pl);
2270 if (sb != NULL) {
2271 KASSERT(*sizep == size, ("invalid size"));
2272 structsize = sizeof(pl);
2273 sbuf_bcat(sb, &structsize, sizeof(structsize));
2274 bzero(&pl, sizeof(pl));
2275 pl.pl_lwpid = td->td_tid;
2276 pl.pl_event = PL_EVENT_NONE;
2277 pl.pl_sigmask = td->td_sigmask;
2278 pl.pl_siglist = td->td_siglist;
2279 if (td->td_si.si_signo != 0) {
2280 pl.pl_event = PL_EVENT_SIGNAL;
2281 pl.pl_flags |= PL_FLAG_SI;
2282 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2283 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2284 #else
2285 pl.pl_siginfo = td->td_si;
2286 #endif
2287 }
2288 strcpy(pl.pl_tdname, td->td_name);
2289 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2290 sbuf_bcat(sb, &pl, sizeof(pl));
2291 }
2292 *sizep = size;
2293 }
2294
2295 /*
2296 * Allow for MD specific notes, as well as any MD
2297 * specific preparations for writing MI notes.
2298 */
2299 static void
2300 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2301 {
2302 struct thread *td;
2303 void *buf;
2304 size_t size;
2305
2306 td = (struct thread *)arg;
2307 size = *sizep;
2308 if (size != 0 && sb != NULL)
2309 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2310 else
2311 buf = NULL;
2312 size = 0;
2313 __elfN(dump_thread)(td, buf, &size);
2314 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2315 if (size != 0 && sb != NULL)
2316 sbuf_bcat(sb, buf, size);
2317 free(buf, M_TEMP);
2318 *sizep = size;
2319 }
2320
2321 #ifdef KINFO_PROC_SIZE
2322 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2323 #endif
2324
2325 static void
2326 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2327 {
2328 struct proc *p;
2329 size_t size;
2330 int structsize;
2331
2332 p = (struct proc *)arg;
2333 size = sizeof(structsize) + p->p_numthreads *
2334 sizeof(elf_kinfo_proc_t);
2335
2336 if (sb != NULL) {
2337 KASSERT(*sizep == size, ("invalid size"));
2338 structsize = sizeof(elf_kinfo_proc_t);
2339 sbuf_bcat(sb, &structsize, sizeof(structsize));
2340 sx_slock(&proctree_lock);
2341 PROC_LOCK(p);
2342 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2343 sx_sunlock(&proctree_lock);
2344 }
2345 *sizep = size;
2346 }
2347
2348 #ifdef KINFO_FILE_SIZE
2349 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2350 #endif
2351
2352 static void
2353 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2354 {
2355 struct proc *p;
2356 size_t size, sect_sz, i;
2357 ssize_t start_len, sect_len;
2358 int structsize, filedesc_flags;
2359
2360 if (coredump_pack_fileinfo)
2361 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2362 else
2363 filedesc_flags = 0;
2364
2365 p = (struct proc *)arg;
2366 structsize = sizeof(struct kinfo_file);
2367 if (sb == NULL) {
2368 size = 0;
2369 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2370 sbuf_set_drain(sb, sbuf_count_drain, &size);
2371 sbuf_bcat(sb, &structsize, sizeof(structsize));
2372 PROC_LOCK(p);
2373 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2374 sbuf_finish(sb);
2375 sbuf_delete(sb);
2376 *sizep = size;
2377 } else {
2378 sbuf_start_section(sb, &start_len);
2379
2380 sbuf_bcat(sb, &structsize, sizeof(structsize));
2381 PROC_LOCK(p);
2382 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2383 filedesc_flags);
2384
2385 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2386 if (sect_len < 0)
2387 return;
2388 sect_sz = sect_len;
2389
2390 KASSERT(sect_sz <= *sizep,
2391 ("kern_proc_filedesc_out did not respect maxlen; "
2392 "requested %zu, got %zu", *sizep - sizeof(structsize),
2393 sect_sz - sizeof(structsize)));
2394
2395 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2396 sbuf_putc(sb, 0);
2397 }
2398 }
2399
2400 #ifdef KINFO_VMENTRY_SIZE
2401 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2402 #endif
2403
2404 static void
2405 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2406 {
2407 struct proc *p;
2408 size_t size;
2409 int structsize, vmmap_flags;
2410
2411 if (coredump_pack_vmmapinfo)
2412 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2413 else
2414 vmmap_flags = 0;
2415
2416 p = (struct proc *)arg;
2417 structsize = sizeof(struct kinfo_vmentry);
2418 if (sb == NULL) {
2419 size = 0;
2420 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2421 sbuf_set_drain(sb, sbuf_count_drain, &size);
2422 sbuf_bcat(sb, &structsize, sizeof(structsize));
2423 PROC_LOCK(p);
2424 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2425 sbuf_finish(sb);
2426 sbuf_delete(sb);
2427 *sizep = size;
2428 } else {
2429 sbuf_bcat(sb, &structsize, sizeof(structsize));
2430 PROC_LOCK(p);
2431 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2432 vmmap_flags);
2433 }
2434 }
2435
2436 static void
2437 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2438 {
2439 struct proc *p;
2440 size_t size;
2441 int structsize;
2442
2443 p = (struct proc *)arg;
2444 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2445 if (sb != NULL) {
2446 KASSERT(*sizep == size, ("invalid size"));
2447 structsize = sizeof(gid_t);
2448 sbuf_bcat(sb, &structsize, sizeof(structsize));
2449 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2450 sizeof(gid_t));
2451 }
2452 *sizep = size;
2453 }
2454
2455 static void
2456 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2457 {
2458 struct proc *p;
2459 size_t size;
2460 int structsize;
2461
2462 p = (struct proc *)arg;
2463 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2464 if (sb != NULL) {
2465 KASSERT(*sizep == size, ("invalid size"));
2466 structsize = sizeof(p->p_fd->fd_cmask);
2467 sbuf_bcat(sb, &structsize, sizeof(structsize));
2468 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2469 }
2470 *sizep = size;
2471 }
2472
2473 static void
2474 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2475 {
2476 struct proc *p;
2477 struct rlimit rlim[RLIM_NLIMITS];
2478 size_t size;
2479 int structsize, i;
2480
2481 p = (struct proc *)arg;
2482 size = sizeof(structsize) + sizeof(rlim);
2483 if (sb != NULL) {
2484 KASSERT(*sizep == size, ("invalid size"));
2485 structsize = sizeof(rlim);
2486 sbuf_bcat(sb, &structsize, sizeof(structsize));
2487 PROC_LOCK(p);
2488 for (i = 0; i < RLIM_NLIMITS; i++)
2489 lim_rlimit_proc(p, i, &rlim[i]);
2490 PROC_UNLOCK(p);
2491 sbuf_bcat(sb, rlim, sizeof(rlim));
2492 }
2493 *sizep = size;
2494 }
2495
2496 static void
2497 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2498 {
2499 struct proc *p;
2500 size_t size;
2501 int structsize;
2502
2503 p = (struct proc *)arg;
2504 size = sizeof(structsize) + sizeof(p->p_osrel);
2505 if (sb != NULL) {
2506 KASSERT(*sizep == size, ("invalid size"));
2507 structsize = sizeof(p->p_osrel);
2508 sbuf_bcat(sb, &structsize, sizeof(structsize));
2509 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2510 }
2511 *sizep = size;
2512 }
2513
2514 static void
2515 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2516 {
2517 struct proc *p;
2518 elf_ps_strings_t ps_strings;
2519 size_t size;
2520 int structsize;
2521
2522 p = (struct proc *)arg;
2523 size = sizeof(structsize) + sizeof(ps_strings);
2524 if (sb != NULL) {
2525 KASSERT(*sizep == size, ("invalid size"));
2526 structsize = sizeof(ps_strings);
2527 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2528 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2529 #else
2530 ps_strings = p->p_sysent->sv_psstrings;
2531 #endif
2532 sbuf_bcat(sb, &structsize, sizeof(structsize));
2533 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2534 }
2535 *sizep = size;
2536 }
2537
2538 static void
2539 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2540 {
2541 struct proc *p;
2542 size_t size;
2543 int structsize;
2544
2545 p = (struct proc *)arg;
2546 if (sb == NULL) {
2547 size = 0;
2548 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2549 sbuf_set_drain(sb, sbuf_count_drain, &size);
2550 sbuf_bcat(sb, &structsize, sizeof(structsize));
2551 PHOLD(p);
2552 proc_getauxv(curthread, p, sb);
2553 PRELE(p);
2554 sbuf_finish(sb);
2555 sbuf_delete(sb);
2556 *sizep = size;
2557 } else {
2558 structsize = sizeof(Elf_Auxinfo);
2559 sbuf_bcat(sb, &structsize, sizeof(structsize));
2560 PHOLD(p);
2561 proc_getauxv(curthread, p, sb);
2562 PRELE(p);
2563 }
2564 }
2565
2566 static boolean_t
2567 __elfN(parse_notes)(struct image_params *imgp, Elf_Note *checknote,
2568 const char *note_vendor, const Elf_Phdr *pnote,
2569 boolean_t (*cb)(const Elf_Note *, void *, boolean_t *), void *cb_arg)
2570 {
2571 const Elf_Note *note, *note0, *note_end;
2572 const char *note_name;
2573 char *buf;
2574 int i, error;
2575 boolean_t res;
2576
2577 /* We need some limit, might as well use PAGE_SIZE. */
2578 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2579 return (FALSE);
2580 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2581 if (pnote->p_offset > PAGE_SIZE ||
2582 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2583 buf = malloc(pnote->p_filesz, M_TEMP, M_NOWAIT);
2584 if (buf == NULL) {
2585 VOP_UNLOCK(imgp->vp, 0);
2586 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2587 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
2588 }
2589 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2590 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2591 curthread->td_ucred, NOCRED, NULL, curthread);
2592 if (error != 0) {
2593 uprintf("i/o error PT_NOTE\n");
2594 goto retf;
2595 }
2596 note = note0 = (const Elf_Note *)buf;
2597 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2598 } else {
2599 note = note0 = (const Elf_Note *)(imgp->image_header +
2600 pnote->p_offset);
2601 note_end = (const Elf_Note *)(imgp->image_header +
2602 pnote->p_offset + pnote->p_filesz);
2603 buf = NULL;
2604 }
2605 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2606 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2607 (const char *)note < sizeof(Elf_Note)) {
2608 goto retf;
2609 }
2610 if (note->n_namesz != checknote->n_namesz ||
2611 note->n_descsz != checknote->n_descsz ||
2612 note->n_type != checknote->n_type)
2613 goto nextnote;
2614 note_name = (const char *)(note + 1);
2615 if (note_name + checknote->n_namesz >=
2616 (const char *)note_end || strncmp(note_vendor,
2617 note_name, checknote->n_namesz) != 0)
2618 goto nextnote;
2619
2620 if (cb(note, cb_arg, &res))
2621 goto ret;
2622 nextnote:
2623 note = (const Elf_Note *)((const char *)(note + 1) +
2624 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2625 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2626 }
2627 retf:
2628 res = FALSE;
2629 ret:
2630 free(buf, M_TEMP);
2631 return (res);
2632 }
2633
2634 struct brandnote_cb_arg {
2635 Elf_Brandnote *brandnote;
2636 int32_t *osrel;
2637 };
2638
2639 static boolean_t
2640 brandnote_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2641 {
2642 struct brandnote_cb_arg *arg;
2643
2644 arg = arg0;
2645
2646 /*
2647 * Fetch the osreldate for binary from the ELF OSABI-note if
2648 * necessary.
2649 */
2650 *res = (arg->brandnote->flags & BN_TRANSLATE_OSREL) != 0 &&
2651 arg->brandnote->trans_osrel != NULL ?
2652 arg->brandnote->trans_osrel(note, arg->osrel) : TRUE;
2653
2654 return (TRUE);
2655 }
2656
2657 static Elf_Note fctl_note = {
2658 .n_namesz = sizeof(FREEBSD_ABI_VENDOR),
2659 .n_descsz = sizeof(uint32_t),
2660 .n_type = NT_FREEBSD_FEATURE_CTL,
2661 };
2662
2663 struct fctl_cb_arg {
2664 uint32_t *fctl0;
2665 };
2666
2667 static boolean_t
2668 note_fctl_cb(const Elf_Note *note, void *arg0, boolean_t *res)
2669 {
2670 struct fctl_cb_arg *arg;
2671 const Elf32_Word *desc;
2672 uintptr_t p;
2673
2674 arg = arg0;
2675 p = (uintptr_t)(note + 1);
2676 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
2677 desc = (const Elf32_Word *)p;
2678 *arg->fctl0 = desc[0];
2679 *res = TRUE;
2680 return (TRUE);
2681 }
2682
2683 /*
2684 * Try to find the appropriate ABI-note section for checknote, fetch
2685 * the osreldate and feature control flags for binary from the ELF
2686 * OSABI-note. Only the first page of the image is searched, the same
2687 * as for headers.
2688 */
2689 static boolean_t
2690 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *brandnote,
2691 int32_t *osrel, uint32_t *fctl0)
2692 {
2693 const Elf_Phdr *phdr;
2694 const Elf_Ehdr *hdr;
2695 struct brandnote_cb_arg b_arg;
2696 struct fctl_cb_arg f_arg;
2697 int i, j;
2698
2699 hdr = (const Elf_Ehdr *)imgp->image_header;
2700 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2701 b_arg.brandnote = brandnote;
2702 b_arg.osrel = osrel;
2703 f_arg.fctl0 = fctl0;
2704
2705 for (i = 0; i < hdr->e_phnum; i++) {
2706 if (phdr[i].p_type == PT_NOTE && __elfN(parse_notes)(imgp,
2707 &brandnote->hdr, brandnote->vendor, &phdr[i], brandnote_cb,
2708 &b_arg)) {
2709 for (j = 0; j < hdr->e_phnum; j++) {
2710 if (phdr[j].p_type == PT_NOTE &&
2711 __elfN(parse_notes)(imgp, &fctl_note,
2712 FREEBSD_ABI_VENDOR, &phdr[j],
2713 note_fctl_cb, &f_arg))
2714 break;
2715 }
2716 return (TRUE);
2717 }
2718 }
2719 return (FALSE);
2720
2721 }
2722
2723 /*
2724 * Tell kern_execve.c about it, with a little help from the linker.
2725 */
2726 static struct execsw __elfN(execsw) = {
2727 .ex_imgact = __CONCAT(exec_, __elfN(imgact)),
2728 .ex_name = __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2729 };
2730 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2731
2732 static vm_prot_t
2733 __elfN(trans_prot)(Elf_Word flags)
2734 {
2735 vm_prot_t prot;
2736
2737 prot = 0;
2738 if (flags & PF_X)
2739 prot |= VM_PROT_EXECUTE;
2740 if (flags & PF_W)
2741 prot |= VM_PROT_WRITE;
2742 if (flags & PF_R)
2743 prot |= VM_PROT_READ;
2744 #if __ELF_WORD_SIZE == 32
2745 #if defined(__amd64__)
2746 if (i386_read_exec && (flags & PF_R))
2747 prot |= VM_PROT_EXECUTE;
2748 #endif
2749 #endif
2750 return (prot);
2751 }
2752
2753 static Elf_Word
2754 __elfN(untrans_prot)(vm_prot_t prot)
2755 {
2756 Elf_Word flags;
2757
2758 flags = 0;
2759 if (prot & VM_PROT_EXECUTE)
2760 flags |= PF_X;
2761 if (prot & VM_PROT_READ)
2762 flags |= PF_R;
2763 if (prot & VM_PROT_WRITE)
2764 flags |= PF_W;
2765 return (flags);
2766 }
2767
2768 void
2769 __elfN(stackgap)(struct image_params *imgp, u_long *stack_base)
2770 {
2771 u_long range, rbase, gap;
2772 int pct;
2773
2774 pct = __elfN(aslr_stack_gap);
2775 if (pct == 0)
2776 return;
2777 if (pct > 50)
2778 pct = 50;
2779 range = imgp->eff_stack_sz * pct / 100;
2780 arc4rand(&rbase, sizeof(rbase), 0);
2781 gap = rbase % range;
2782 gap &= ~(sizeof(u_long) - 1);
2783 *stack_base -= gap;
2784 }
Cache object: 5c7cd53f5490bd49d00f099efb5b1d9b
|