FreeBSD/Linux Kernel Cross Reference
sys/kern/imgact_elf.c
1 /*-
2 * Copyright (c) 2017 Dell EMC
3 * Copyright (c) 2000 David O'Brien
4 * Copyright (c) 1995-1996 Søren Schmidt
5 * Copyright (c) 1996 Peter Wemm
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer
13 * in this position and unchanged.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. The name of the author may not be used to endorse or promote products
18 * derived from this software without specific prior written permission
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/11.2/sys/kern/imgact_elf.c 338606 2018-09-12 05:07:35Z gordon $");
34
35 #include "opt_capsicum.h"
36 #include "opt_compat.h"
37 #include "opt_gzio.h"
38
39 #include <sys/param.h>
40 #include <sys/capsicum.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/gzio.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, int interp_name_len, int32_t *osrel);
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,
96 size_t pagesize);
97 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
98 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
99 int32_t *osrel);
100 static boolean_t 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);
103 static vm_prot_t __elfN(trans_prot)(Elf_Word);
104 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
105
106 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
107 "");
108
109 #define CORE_BUF_SIZE (16 * 1024)
110
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");
115
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");
120
121 int __elfN(nxstack) =
122 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
123 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
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 Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
141
142 #define trunc_page_ps(va, ps) rounddown2(va, ps)
143 #define round_page_ps(va, ps) roundup2(va, ps)
144 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
145
146 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
147
148 Elf_Brandnote __elfN(freebsd_brandnote) = {
149 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
150 .hdr.n_descsz = sizeof(int32_t),
151 .hdr.n_type = NT_FREEBSD_ABI_TAG,
152 .vendor = FREEBSD_ABI_VENDOR,
153 .flags = BN_TRANSLATE_OSREL,
154 .trans_osrel = __elfN(freebsd_trans_osrel)
155 };
156
157 static boolean_t
158 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
159 {
160 uintptr_t p;
161
162 p = (uintptr_t)(note + 1);
163 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
164 *osrel = *(const int32_t *)(p);
165
166 return (TRUE);
167 }
168
169 static const char GNU_ABI_VENDOR[] = "GNU";
170 static int GNU_KFREEBSD_ABI_DESC = 3;
171
172 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
173 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
174 .hdr.n_descsz = 16, /* XXX at least 16 */
175 .hdr.n_type = 1,
176 .vendor = GNU_ABI_VENDOR,
177 .flags = BN_TRANSLATE_OSREL,
178 .trans_osrel = kfreebsd_trans_osrel
179 };
180
181 static boolean_t
182 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
183 {
184 const Elf32_Word *desc;
185 uintptr_t p;
186
187 p = (uintptr_t)(note + 1);
188 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
189
190 desc = (const Elf32_Word *)p;
191 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
192 return (FALSE);
193
194 /*
195 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
196 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
197 */
198 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
199
200 return (TRUE);
201 }
202
203 int
204 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
205 {
206 int i;
207
208 for (i = 0; i < MAX_BRANDS; i++) {
209 if (elf_brand_list[i] == NULL) {
210 elf_brand_list[i] = entry;
211 break;
212 }
213 }
214 if (i == MAX_BRANDS) {
215 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
216 __func__, entry);
217 return (-1);
218 }
219 return (0);
220 }
221
222 int
223 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
224 {
225 int i;
226
227 for (i = 0; i < MAX_BRANDS; i++) {
228 if (elf_brand_list[i] == entry) {
229 elf_brand_list[i] = NULL;
230 break;
231 }
232 }
233 if (i == MAX_BRANDS)
234 return (-1);
235 return (0);
236 }
237
238 int
239 __elfN(brand_inuse)(Elf_Brandinfo *entry)
240 {
241 struct proc *p;
242 int rval = FALSE;
243
244 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
246 if (p->p_sysent == entry->sysvec) {
247 rval = TRUE;
248 break;
249 }
250 }
251 sx_sunlock(&allproc_lock);
252
253 return (rval);
254 }
255
256 static Elf_Brandinfo *
257 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
258 int interp_name_len, int32_t *osrel)
259 {
260 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
261 Elf_Brandinfo *bi, *bi_m;
262 boolean_t ret;
263 int i;
264
265 /*
266 * We support four types of branding -- (1) the ELF EI_OSABI field
267 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
268 * branding w/in the ELF header, (3) path of the `interp_path'
269 * field, and (4) the ".note.ABI-tag" ELF section.
270 */
271
272 /* Look for an ".note.ABI-tag" ELF section */
273 bi_m = NULL;
274 for (i = 0; i < MAX_BRANDS; i++) {
275 bi = elf_brand_list[i];
276 if (bi == NULL)
277 continue;
278 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
279 continue;
280 if (hdr->e_machine == bi->machine && (bi->flags &
281 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
282 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
283 /* Give brand a chance to veto check_note's guess */
284 if (ret && bi->header_supported)
285 ret = bi->header_supported(imgp);
286 /*
287 * If note checker claimed the binary, but the
288 * interpreter path in the image does not
289 * match default one for the brand, try to
290 * search for other brands with the same
291 * interpreter. Either there is better brand
292 * with the right interpreter, or, failing
293 * this, we return first brand which accepted
294 * our note and, optionally, header.
295 */
296 if (ret && bi_m == NULL && interp != NULL &&
297 (bi->interp_path == NULL ||
298 (strlen(bi->interp_path) + 1 != interp_name_len ||
299 strncmp(interp, bi->interp_path, interp_name_len)
300 != 0))) {
301 bi_m = bi;
302 ret = 0;
303 }
304 if (ret)
305 return (bi);
306 }
307 }
308 if (bi_m != NULL)
309 return (bi_m);
310
311 /* If the executable has a brand, search for it in the brand list. */
312 for (i = 0; i < MAX_BRANDS; i++) {
313 bi = elf_brand_list[i];
314 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
315 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
316 continue;
317 if (hdr->e_machine == bi->machine &&
318 (hdr->e_ident[EI_OSABI] == bi->brand ||
319 (bi->compat_3_brand != NULL &&
320 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
321 bi->compat_3_brand) == 0))) {
322 /* Looks good, but give brand a chance to veto */
323 if (bi->header_supported == NULL ||
324 bi->header_supported(imgp)) {
325 /*
326 * Again, prefer strictly matching
327 * interpreter path.
328 */
329 if (interp_name_len == 0 &&
330 bi->interp_path == NULL)
331 return (bi);
332 if (bi->interp_path != NULL &&
333 strlen(bi->interp_path) + 1 ==
334 interp_name_len && strncmp(interp,
335 bi->interp_path, interp_name_len) == 0)
336 return (bi);
337 if (bi_m == NULL)
338 bi_m = bi;
339 }
340 }
341 }
342 if (bi_m != NULL)
343 return (bi_m);
344
345 /* No known brand, see if the header is recognized by any brand */
346 for (i = 0; i < MAX_BRANDS; i++) {
347 bi = elf_brand_list[i];
348 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
349 bi->header_supported == NULL)
350 continue;
351 if (hdr->e_machine == bi->machine) {
352 ret = bi->header_supported(imgp);
353 if (ret)
354 return (bi);
355 }
356 }
357
358 /* Lacking a known brand, search for a recognized interpreter. */
359 if (interp != NULL) {
360 for (i = 0; i < MAX_BRANDS; i++) {
361 bi = elf_brand_list[i];
362 if (bi == NULL || (bi->flags &
363 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
364 != 0)
365 continue;
366 if (hdr->e_machine == bi->machine &&
367 bi->interp_path != NULL &&
368 /* ELF image p_filesz includes terminating zero */
369 strlen(bi->interp_path) + 1 == interp_name_len &&
370 strncmp(interp, bi->interp_path, interp_name_len)
371 == 0 && (bi->header_supported == NULL ||
372 bi->header_supported(imgp)))
373 return (bi);
374 }
375 }
376
377 /* Lacking a recognized interpreter, try the default brand */
378 for (i = 0; i < MAX_BRANDS; i++) {
379 bi = elf_brand_list[i];
380 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
381 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
382 continue;
383 if (hdr->e_machine == bi->machine &&
384 __elfN(fallback_brand) == bi->brand &&
385 (bi->header_supported == NULL ||
386 bi->header_supported(imgp)))
387 return (bi);
388 }
389 return (NULL);
390 }
391
392 static int
393 __elfN(check_header)(const Elf_Ehdr *hdr)
394 {
395 Elf_Brandinfo *bi;
396 int i;
397
398 if (!IS_ELF(*hdr) ||
399 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
400 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
401 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
402 hdr->e_phentsize != sizeof(Elf_Phdr) ||
403 hdr->e_version != ELF_TARG_VER)
404 return (ENOEXEC);
405
406 /*
407 * Make sure we have at least one brand for this machine.
408 */
409
410 for (i = 0; i < MAX_BRANDS; i++) {
411 bi = elf_brand_list[i];
412 if (bi != NULL && bi->machine == hdr->e_machine)
413 break;
414 }
415 if (i == MAX_BRANDS)
416 return (ENOEXEC);
417
418 return (0);
419 }
420
421 static int
422 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
423 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
424 {
425 struct sf_buf *sf;
426 int error;
427 vm_offset_t off;
428
429 /*
430 * Create the page if it doesn't exist yet. Ignore errors.
431 */
432 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
433 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
434
435 /*
436 * Find the page from the underlying object.
437 */
438 if (object != NULL) {
439 sf = vm_imgact_map_page(object, offset);
440 if (sf == NULL)
441 return (KERN_FAILURE);
442 off = offset - trunc_page(offset);
443 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
444 end - start);
445 vm_imgact_unmap_page(sf);
446 if (error != 0)
447 return (KERN_FAILURE);
448 }
449
450 return (KERN_SUCCESS);
451 }
452
453 static int
454 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
455 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
456 int cow)
457 {
458 struct sf_buf *sf;
459 vm_offset_t off;
460 vm_size_t sz;
461 int error, locked, rv;
462
463 if (start != trunc_page(start)) {
464 rv = __elfN(map_partial)(map, object, offset, start,
465 round_page(start), prot);
466 if (rv != KERN_SUCCESS)
467 return (rv);
468 offset += round_page(start) - start;
469 start = round_page(start);
470 }
471 if (end != round_page(end)) {
472 rv = __elfN(map_partial)(map, object, offset +
473 trunc_page(end) - start, trunc_page(end), end, prot);
474 if (rv != KERN_SUCCESS)
475 return (rv);
476 end = trunc_page(end);
477 }
478 if (start >= end)
479 return (KERN_SUCCESS);
480 if ((offset & PAGE_MASK) != 0) {
481 /*
482 * The mapping is not page aligned. This means that we have
483 * to copy the data.
484 */
485 rv = vm_map_fixed(map, NULL, 0, start, end - start,
486 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
487 if (rv != KERN_SUCCESS)
488 return (rv);
489 if (object == NULL)
490 return (KERN_SUCCESS);
491 for (; start < end; start += sz) {
492 sf = vm_imgact_map_page(object, offset);
493 if (sf == NULL)
494 return (KERN_FAILURE);
495 off = offset - trunc_page(offset);
496 sz = end - start;
497 if (sz > PAGE_SIZE - off)
498 sz = PAGE_SIZE - off;
499 error = copyout((caddr_t)sf_buf_kva(sf) + off,
500 (caddr_t)start, sz);
501 vm_imgact_unmap_page(sf);
502 if (error != 0)
503 return (KERN_FAILURE);
504 offset += sz;
505 }
506 } else {
507 vm_object_reference(object);
508 rv = vm_map_fixed(map, object, offset, start, end - start,
509 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
510 if (rv != KERN_SUCCESS) {
511 locked = VOP_ISLOCKED(imgp->vp);
512 VOP_UNLOCK(imgp->vp, 0);
513 vm_object_deallocate(object);
514 vn_lock(imgp->vp, locked | LK_RETRY);
515 return (rv);
516 }
517 }
518 return (KERN_SUCCESS);
519 }
520
521 static int
522 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
523 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
524 size_t pagesize)
525 {
526 struct sf_buf *sf;
527 size_t map_len;
528 vm_map_t map;
529 vm_object_t object;
530 vm_offset_t off, map_addr;
531 int error, rv, cow;
532 size_t copy_len;
533 vm_ooffset_t file_addr;
534
535 /*
536 * It's necessary to fail if the filsz + offset taken from the
537 * header is greater than the actual file pager object's size.
538 * If we were to allow this, then the vm_map_find() below would
539 * walk right off the end of the file object and into the ether.
540 *
541 * While I'm here, might as well check for something else that
542 * is invalid: filsz cannot be greater than memsz.
543 */
544 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
545 filsz > memsz) {
546 uprintf("elf_load_section: truncated ELF file\n");
547 return (ENOEXEC);
548 }
549
550 object = imgp->object;
551 map = &imgp->proc->p_vmspace->vm_map;
552 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
553 file_addr = trunc_page_ps(offset, pagesize);
554
555 /*
556 * We have two choices. We can either clear the data in the last page
557 * of an oversized mapping, or we can start the anon mapping a page
558 * early and copy the initialized data into that first page. We
559 * choose the second.
560 */
561 if (filsz == 0)
562 map_len = 0;
563 else if (memsz > filsz)
564 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
565 else
566 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
567
568 if (map_len != 0) {
569 /* cow flags: don't dump readonly sections in core */
570 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
571 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
572
573 rv = __elfN(map_insert)(imgp, map,
574 object,
575 file_addr, /* file offset */
576 map_addr, /* virtual start */
577 map_addr + map_len,/* virtual end */
578 prot,
579 cow);
580 if (rv != KERN_SUCCESS)
581 return (EINVAL);
582
583 /* we can stop now if we've covered it all */
584 if (memsz == filsz)
585 return (0);
586 }
587
588
589 /*
590 * We have to get the remaining bit of the file into the first part
591 * of the oversized map segment. This is normally because the .data
592 * segment in the file is extended to provide bss. It's a neat idea
593 * to try and save a page, but it's a pain in the behind to implement.
594 */
595 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
596 filsz, pagesize);
597 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
598 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
599 map_addr;
600
601 /* This had damn well better be true! */
602 if (map_len != 0) {
603 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
604 map_addr + map_len, prot, 0);
605 if (rv != KERN_SUCCESS)
606 return (EINVAL);
607 }
608
609 if (copy_len != 0) {
610 sf = vm_imgact_map_page(object, offset + filsz);
611 if (sf == NULL)
612 return (EIO);
613
614 /* send the page fragment to user space */
615 off = trunc_page_ps(offset + filsz, pagesize) -
616 trunc_page(offset + filsz);
617 error = copyout((caddr_t)sf_buf_kva(sf) + off,
618 (caddr_t)map_addr, copy_len);
619 vm_imgact_unmap_page(sf);
620 if (error != 0)
621 return (error);
622 }
623
624 /*
625 * Remove write access to the page if it was only granted by map_insert
626 * to allow copyout.
627 */
628 if ((prot & VM_PROT_WRITE) == 0)
629 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
630 map_len), prot, FALSE);
631
632 return (0);
633 }
634
635 /*
636 * Load the file "file" into memory. It may be either a shared object
637 * or an executable.
638 *
639 * The "addr" reference parameter is in/out. On entry, it specifies
640 * the address where a shared object should be loaded. If the file is
641 * an executable, this value is ignored. On exit, "addr" specifies
642 * where the file was actually loaded.
643 *
644 * The "entry" reference parameter is out only. On exit, it specifies
645 * the entry point for the loaded file.
646 */
647 static int
648 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
649 u_long *entry, size_t pagesize)
650 {
651 struct {
652 struct nameidata nd;
653 struct vattr attr;
654 struct image_params image_params;
655 } *tempdata;
656 const Elf_Ehdr *hdr = NULL;
657 const Elf_Phdr *phdr = NULL;
658 struct nameidata *nd;
659 struct vattr *attr;
660 struct image_params *imgp;
661 vm_prot_t prot;
662 u_long rbase;
663 u_long base_addr = 0;
664 int error, i, numsegs;
665
666 #ifdef CAPABILITY_MODE
667 /*
668 * XXXJA: This check can go away once we are sufficiently confident
669 * that the checks in namei() are correct.
670 */
671 if (IN_CAPABILITY_MODE(curthread))
672 return (ECAPMODE);
673 #endif
674
675 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
676 nd = &tempdata->nd;
677 attr = &tempdata->attr;
678 imgp = &tempdata->image_params;
679
680 /*
681 * Initialize part of the common data
682 */
683 imgp->proc = p;
684 imgp->attr = attr;
685 imgp->firstpage = NULL;
686 imgp->image_header = NULL;
687 imgp->object = NULL;
688 imgp->execlabel = NULL;
689
690 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
691 if ((error = namei(nd)) != 0) {
692 nd->ni_vp = NULL;
693 goto fail;
694 }
695 NDFREE(nd, NDF_ONLY_PNBUF);
696 imgp->vp = nd->ni_vp;
697
698 /*
699 * Check permissions, modes, uid, etc on the file, and "open" it.
700 */
701 error = exec_check_permissions(imgp);
702 if (error)
703 goto fail;
704
705 error = exec_map_first_page(imgp);
706 if (error)
707 goto fail;
708
709 /*
710 * Also make certain that the interpreter stays the same, so set
711 * its VV_TEXT flag, too.
712 */
713 VOP_SET_TEXT(nd->ni_vp);
714
715 imgp->object = nd->ni_vp->v_object;
716
717 hdr = (const Elf_Ehdr *)imgp->image_header;
718 if ((error = __elfN(check_header)(hdr)) != 0)
719 goto fail;
720 if (hdr->e_type == ET_DYN)
721 rbase = *addr;
722 else if (hdr->e_type == ET_EXEC)
723 rbase = 0;
724 else {
725 error = ENOEXEC;
726 goto fail;
727 }
728
729 /* Only support headers that fit within first page for now */
730 if ((hdr->e_phoff > PAGE_SIZE) ||
731 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
732 error = ENOEXEC;
733 goto fail;
734 }
735
736 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
737 if (!aligned(phdr, Elf_Addr)) {
738 error = ENOEXEC;
739 goto fail;
740 }
741
742 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
743 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
744 /* Loadable segment */
745 prot = __elfN(trans_prot)(phdr[i].p_flags);
746 error = __elfN(load_section)(imgp, phdr[i].p_offset,
747 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
748 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
749 if (error != 0)
750 goto fail;
751 /*
752 * Establish the base address if this is the
753 * first segment.
754 */
755 if (numsegs == 0)
756 base_addr = trunc_page(phdr[i].p_vaddr +
757 rbase);
758 numsegs++;
759 }
760 }
761 *addr = base_addr;
762 *entry = (unsigned long)hdr->e_entry + rbase;
763
764 fail:
765 if (imgp->firstpage)
766 exec_unmap_first_page(imgp);
767
768 if (nd->ni_vp)
769 vput(nd->ni_vp);
770
771 free(tempdata, M_TEMP);
772
773 return (error);
774 }
775
776 static int
777 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
778 {
779 struct thread *td;
780 const Elf_Ehdr *hdr;
781 const Elf_Phdr *phdr;
782 Elf_Auxargs *elf_auxargs;
783 struct vmspace *vmspace;
784 const char *err_str, *newinterp;
785 char *interp, *interp_buf, *path;
786 Elf_Brandinfo *brand_info;
787 struct sysentvec *sv;
788 vm_prot_t prot;
789 u_long text_size, data_size, total_size, text_addr, data_addr;
790 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
791 int32_t osrel;
792 int error, i, n, interp_name_len, have_interp;
793
794 hdr = (const Elf_Ehdr *)imgp->image_header;
795
796 /*
797 * Do we have a valid ELF header ?
798 *
799 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
800 * if particular brand doesn't support it.
801 */
802 if (__elfN(check_header)(hdr) != 0 ||
803 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
804 return (-1);
805
806 /*
807 * From here on down, we return an errno, not -1, as we've
808 * detected an ELF file.
809 */
810
811 if ((hdr->e_phoff > PAGE_SIZE) ||
812 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
813 /* Only support headers in first page for now */
814 uprintf("Program headers not in the first page\n");
815 return (ENOEXEC);
816 }
817 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
818 if (!aligned(phdr, Elf_Addr)) {
819 uprintf("Unaligned program headers\n");
820 return (ENOEXEC);
821 }
822
823 n = error = 0;
824 baddr = 0;
825 osrel = 0;
826 text_size = data_size = total_size = text_addr = data_addr = 0;
827 entry = proghdr = 0;
828 interp_name_len = 0;
829 err_str = newinterp = NULL;
830 interp = interp_buf = NULL;
831 td = curthread;
832
833 for (i = 0; i < hdr->e_phnum; i++) {
834 switch (phdr[i].p_type) {
835 case PT_LOAD:
836 if (n == 0)
837 baddr = phdr[i].p_vaddr;
838 n++;
839 break;
840 case PT_INTERP:
841 /* Path to interpreter */
842 if (phdr[i].p_filesz < 2 ||
843 phdr[i].p_filesz > MAXPATHLEN) {
844 uprintf("Invalid PT_INTERP\n");
845 error = ENOEXEC;
846 goto ret;
847 }
848 if (interp != NULL) {
849 uprintf("Multiple PT_INTERP headers\n");
850 error = ENOEXEC;
851 goto ret;
852 }
853 interp_name_len = phdr[i].p_filesz;
854 if (phdr[i].p_offset > PAGE_SIZE ||
855 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
856 VOP_UNLOCK(imgp->vp, 0);
857 interp_buf = malloc(interp_name_len + 1, M_TEMP,
858 M_WAITOK);
859 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
860 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
861 interp_name_len, phdr[i].p_offset,
862 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
863 NOCRED, NULL, td);
864 if (error != 0) {
865 uprintf("i/o error PT_INTERP\n");
866 goto ret;
867 }
868 interp_buf[interp_name_len] = '\0';
869 interp = interp_buf;
870 } else {
871 interp = __DECONST(char *, imgp->image_header) +
872 phdr[i].p_offset;
873 if (interp[interp_name_len - 1] != '\0') {
874 uprintf("Invalid PT_INTERP\n");
875 error = ENOEXEC;
876 goto ret;
877 }
878 }
879 break;
880 case PT_GNU_STACK:
881 if (__elfN(nxstack))
882 imgp->stack_prot =
883 __elfN(trans_prot)(phdr[i].p_flags);
884 imgp->stack_sz = phdr[i].p_memsz;
885 break;
886 }
887 }
888
889 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
890 &osrel);
891 if (brand_info == NULL) {
892 uprintf("ELF binary type \"%u\" not known.\n",
893 hdr->e_ident[EI_OSABI]);
894 error = ENOEXEC;
895 goto ret;
896 }
897 if (hdr->e_type == ET_DYN) {
898 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
899 uprintf("Cannot execute shared object\n");
900 error = ENOEXEC;
901 goto ret;
902 }
903 /*
904 * Honour the base load address from the dso if it is
905 * non-zero for some reason.
906 */
907 if (baddr == 0)
908 et_dyn_addr = ET_DYN_LOAD_ADDR;
909 else
910 et_dyn_addr = 0;
911 } else
912 et_dyn_addr = 0;
913 sv = brand_info->sysvec;
914 if (interp != NULL && brand_info->interp_newpath != NULL)
915 newinterp = brand_info->interp_newpath;
916
917 /*
918 * Avoid a possible deadlock if the current address space is destroyed
919 * and that address space maps the locked vnode. In the common case,
920 * the locked vnode's v_usecount is decremented but remains greater
921 * than zero. Consequently, the vnode lock is not needed by vrele().
922 * However, in cases where the vnode lock is external, such as nullfs,
923 * v_usecount may become zero.
924 *
925 * The VV_TEXT flag prevents modifications to the executable while
926 * the vnode is unlocked.
927 */
928 VOP_UNLOCK(imgp->vp, 0);
929
930 error = exec_new_vmspace(imgp, sv);
931 imgp->proc->p_sysent = sv;
932
933 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
934 if (error != 0)
935 goto ret;
936
937 for (i = 0; i < hdr->e_phnum; i++) {
938 switch (phdr[i].p_type) {
939 case PT_LOAD: /* Loadable segment */
940 if (phdr[i].p_memsz == 0)
941 break;
942 prot = __elfN(trans_prot)(phdr[i].p_flags);
943 error = __elfN(load_section)(imgp, phdr[i].p_offset,
944 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
945 phdr[i].p_memsz, phdr[i].p_filesz, prot,
946 sv->sv_pagesize);
947 if (error != 0)
948 goto ret;
949
950 /*
951 * If this segment contains the program headers,
952 * remember their virtual address for the AT_PHDR
953 * aux entry. Static binaries don't usually include
954 * a PT_PHDR entry.
955 */
956 if (phdr[i].p_offset == 0 &&
957 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
958 <= phdr[i].p_filesz)
959 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
960 et_dyn_addr;
961
962 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
963 seg_size = round_page(phdr[i].p_memsz +
964 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
965
966 /*
967 * Make the largest executable segment the official
968 * text segment and all others data.
969 *
970 * Note that obreak() assumes that data_addr +
971 * data_size == end of data load area, and the ELF
972 * file format expects segments to be sorted by
973 * address. If multiple data segments exist, the
974 * last one will be used.
975 */
976
977 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
978 text_size = seg_size;
979 text_addr = seg_addr;
980 } else {
981 data_size = seg_size;
982 data_addr = seg_addr;
983 }
984 total_size += seg_size;
985 break;
986 case PT_PHDR: /* Program header table info */
987 proghdr = phdr[i].p_vaddr + et_dyn_addr;
988 break;
989 default:
990 break;
991 }
992 }
993
994 if (data_addr == 0 && data_size == 0) {
995 data_addr = text_addr;
996 data_size = text_size;
997 }
998
999 entry = (u_long)hdr->e_entry + et_dyn_addr;
1000
1001 /*
1002 * Check limits. It should be safe to check the
1003 * limits after loading the segments since we do
1004 * not actually fault in all the segments pages.
1005 */
1006 PROC_LOCK(imgp->proc);
1007 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1008 err_str = "Data segment size exceeds process limit";
1009 else if (text_size > maxtsiz)
1010 err_str = "Text segment size exceeds system limit";
1011 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1012 err_str = "Total segment size exceeds process limit";
1013 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1014 err_str = "Data segment size exceeds resource limit";
1015 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1016 err_str = "Total segment size exceeds resource limit";
1017 if (err_str != NULL) {
1018 PROC_UNLOCK(imgp->proc);
1019 uprintf("%s\n", err_str);
1020 error = ENOMEM;
1021 goto ret;
1022 }
1023
1024 vmspace = imgp->proc->p_vmspace;
1025 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1026 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1027 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1028 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1029
1030 /*
1031 * We load the dynamic linker where a userland call
1032 * to mmap(0, ...) would put it. The rationale behind this
1033 * calculation is that it leaves room for the heap to grow to
1034 * its maximum allowed size.
1035 */
1036 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1037 RLIMIT_DATA));
1038 PROC_UNLOCK(imgp->proc);
1039
1040 imgp->entry_addr = entry;
1041
1042 if (interp != NULL) {
1043 have_interp = FALSE;
1044 VOP_UNLOCK(imgp->vp, 0);
1045 if (brand_info->emul_path != NULL &&
1046 brand_info->emul_path[0] != '\0') {
1047 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1048 snprintf(path, MAXPATHLEN, "%s%s",
1049 brand_info->emul_path, interp);
1050 error = __elfN(load_file)(imgp->proc, path, &addr,
1051 &imgp->entry_addr, sv->sv_pagesize);
1052 free(path, M_TEMP);
1053 if (error == 0)
1054 have_interp = TRUE;
1055 }
1056 if (!have_interp && newinterp != NULL &&
1057 (brand_info->interp_path == NULL ||
1058 strcmp(interp, brand_info->interp_path) == 0)) {
1059 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1060 &imgp->entry_addr, sv->sv_pagesize);
1061 if (error == 0)
1062 have_interp = TRUE;
1063 }
1064 if (!have_interp) {
1065 error = __elfN(load_file)(imgp->proc, interp, &addr,
1066 &imgp->entry_addr, sv->sv_pagesize);
1067 }
1068 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1069 if (error != 0) {
1070 uprintf("ELF interpreter %s not found, error %d\n",
1071 interp, error);
1072 goto ret;
1073 }
1074 } else
1075 addr = et_dyn_addr;
1076
1077 /*
1078 * Construct auxargs table (used by the fixup routine)
1079 */
1080 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1081 elf_auxargs->execfd = -1;
1082 elf_auxargs->phdr = proghdr;
1083 elf_auxargs->phent = hdr->e_phentsize;
1084 elf_auxargs->phnum = hdr->e_phnum;
1085 elf_auxargs->pagesz = PAGE_SIZE;
1086 elf_auxargs->base = addr;
1087 elf_auxargs->flags = 0;
1088 elf_auxargs->entry = entry;
1089 elf_auxargs->hdr_eflags = hdr->e_flags;
1090
1091 imgp->auxargs = elf_auxargs;
1092 imgp->interpreted = 0;
1093 imgp->reloc_base = addr;
1094 imgp->proc->p_osrel = osrel;
1095 imgp->proc->p_elf_machine = hdr->e_machine;
1096 imgp->proc->p_elf_flags = hdr->e_flags;
1097
1098 ret:
1099 free(interp_buf, M_TEMP);
1100 return (error);
1101 }
1102
1103 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1104
1105 int
1106 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1107 {
1108 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1109 Elf_Addr *base;
1110 Elf_Addr *pos;
1111
1112 base = (Elf_Addr *)*stack_base;
1113 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1114
1115 if (args->execfd != -1)
1116 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1117 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1118 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1119 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1120 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1121 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1122 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1123 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1124 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1125 if (imgp->execpathp != 0)
1126 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1127 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1128 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1129 if (imgp->canary != 0) {
1130 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1131 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1132 }
1133 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1134 if (imgp->pagesizes != 0) {
1135 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1136 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1137 }
1138 if (imgp->sysent->sv_timekeep_base != 0) {
1139 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1140 imgp->sysent->sv_timekeep_base);
1141 }
1142 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1143 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1144 imgp->sysent->sv_stackprot);
1145 if ((imgp->sysent->sv_flags & SV_HWCAP) != 0) {
1146 if (imgp->sysent->sv_hwcap != NULL)
1147 AUXARGS_ENTRY(pos, AT_HWCAP, *imgp->sysent->sv_hwcap);
1148 if (imgp->sysent->sv_hwcap2 != NULL)
1149 AUXARGS_ENTRY(pos, AT_HWCAP2, *imgp->sysent->sv_hwcap2);
1150 }
1151 AUXARGS_ENTRY(pos, AT_NULL, 0);
1152
1153 free(imgp->auxargs, M_TEMP);
1154 imgp->auxargs = NULL;
1155
1156 base--;
1157 suword(base, (long)imgp->args->argc);
1158 *stack_base = (register_t *)base;
1159 return (0);
1160 }
1161
1162 /*
1163 * Code for generating ELF core dumps.
1164 */
1165
1166 typedef void (*segment_callback)(vm_map_entry_t, void *);
1167
1168 /* Closure for cb_put_phdr(). */
1169 struct phdr_closure {
1170 Elf_Phdr *phdr; /* Program header to fill in */
1171 Elf_Off offset; /* Offset of segment in core file */
1172 };
1173
1174 /* Closure for cb_size_segment(). */
1175 struct sseg_closure {
1176 int count; /* Count of writable segments. */
1177 size_t size; /* Total size of all writable segments. */
1178 };
1179
1180 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1181
1182 struct note_info {
1183 int type; /* Note type. */
1184 outfunc_t outfunc; /* Output function. */
1185 void *outarg; /* Argument for the output function. */
1186 size_t outsize; /* Output size. */
1187 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1188 };
1189
1190 TAILQ_HEAD(note_info_list, note_info);
1191
1192 /* Coredump output parameters. */
1193 struct coredump_params {
1194 off_t offset;
1195 struct ucred *active_cred;
1196 struct ucred *file_cred;
1197 struct thread *td;
1198 struct vnode *vp;
1199 struct gzio_stream *gzs;
1200 };
1201
1202 static void cb_put_phdr(vm_map_entry_t, void *);
1203 static void cb_size_segment(vm_map_entry_t, void *);
1204 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1205 enum uio_seg);
1206 static void each_writable_segment(struct thread *, segment_callback, void *);
1207 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1208 struct note_info_list *, size_t);
1209 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1210 size_t *);
1211 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1212 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1213 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1214 static int sbuf_drain_core_output(void *, const char *, int);
1215 static int sbuf_drain_count(void *arg, const char *data, int len);
1216
1217 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1218 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1219 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1220 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1221 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1222 static void __elfN(note_ptlwpinfo)(void *, struct sbuf *, size_t *);
1223 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1224 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1225 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1226 static void note_procstat_files(void *, struct sbuf *, size_t *);
1227 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1228 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1229 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1230 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1231 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1232
1233 #ifdef GZIO
1234 extern int compress_user_cores_gzlevel;
1235
1236 /*
1237 * Write out a core segment to the compression stream.
1238 */
1239 static int
1240 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1241 {
1242 u_int chunk_len;
1243 int error;
1244
1245 while (len > 0) {
1246 chunk_len = MIN(len, CORE_BUF_SIZE);
1247
1248 /*
1249 * We can get EFAULT error here.
1250 * In that case zero out the current chunk of the segment.
1251 */
1252 error = copyin(base, buf, chunk_len);
1253 if (error != 0)
1254 bzero(buf, chunk_len);
1255 error = gzio_write(p->gzs, buf, chunk_len);
1256 if (error != 0)
1257 break;
1258 base += chunk_len;
1259 len -= chunk_len;
1260 }
1261 return (error);
1262 }
1263
1264 static int
1265 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1266 {
1267
1268 return (core_write((struct coredump_params *)arg, base, len, offset,
1269 UIO_SYSSPACE));
1270 }
1271 #endif /* GZIO */
1272
1273 static int
1274 core_write(struct coredump_params *p, const void *base, size_t len,
1275 off_t offset, enum uio_seg seg)
1276 {
1277
1278 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1279 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1280 p->active_cred, p->file_cred, NULL, p->td));
1281 }
1282
1283 static int
1284 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1285 void *tmpbuf)
1286 {
1287 int error;
1288
1289 #ifdef GZIO
1290 if (p->gzs != NULL)
1291 return (compress_chunk(p, base, tmpbuf, len));
1292 #endif
1293 /*
1294 * EFAULT is a non-fatal error that we can get, for example,
1295 * if the segment is backed by a file but extends beyond its
1296 * end.
1297 */
1298 error = core_write(p, base, len, offset, UIO_USERSPACE);
1299 if (error == EFAULT) {
1300 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1301 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1302 "for process %s\n", base, len, offset, curproc->p_comm);
1303
1304 /*
1305 * Write a "real" zero byte at the end of the target region
1306 * in the case this is the last segment.
1307 * The intermediate space will be implicitly zero-filled.
1308 */
1309 error = core_write(p, zero_region, 1, offset + len - 1,
1310 UIO_SYSSPACE);
1311 }
1312 return (error);
1313 }
1314
1315 /*
1316 * Drain into a core file.
1317 */
1318 static int
1319 sbuf_drain_core_output(void *arg, const char *data, int len)
1320 {
1321 struct coredump_params *p;
1322 int error, locked;
1323
1324 p = (struct coredump_params *)arg;
1325
1326 /*
1327 * Some kern_proc out routines that print to this sbuf may
1328 * call us with the process lock held. Draining with the
1329 * non-sleepable lock held is unsafe. The lock is needed for
1330 * those routines when dumping a live process. In our case we
1331 * can safely release the lock before draining and acquire
1332 * again after.
1333 */
1334 locked = PROC_LOCKED(p->td->td_proc);
1335 if (locked)
1336 PROC_UNLOCK(p->td->td_proc);
1337 #ifdef GZIO
1338 if (p->gzs != NULL)
1339 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1340 else
1341 #endif
1342 error = core_write(p, __DECONST(void *, data), len, p->offset,
1343 UIO_SYSSPACE);
1344 if (locked)
1345 PROC_LOCK(p->td->td_proc);
1346 if (error != 0)
1347 return (-error);
1348 p->offset += len;
1349 return (len);
1350 }
1351
1352 /*
1353 * Drain into a counter.
1354 */
1355 static int
1356 sbuf_drain_count(void *arg, const char *data __unused, int len)
1357 {
1358 size_t *sizep;
1359
1360 sizep = (size_t *)arg;
1361 *sizep += len;
1362 return (len);
1363 }
1364
1365 int
1366 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1367 {
1368 struct ucred *cred = td->td_ucred;
1369 int error = 0;
1370 struct sseg_closure seginfo;
1371 struct note_info_list notelst;
1372 struct coredump_params params;
1373 struct note_info *ninfo;
1374 void *hdr, *tmpbuf;
1375 size_t hdrsize, notesz, coresize;
1376 #ifdef GZIO
1377 boolean_t compress;
1378
1379 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1380 #endif
1381 hdr = NULL;
1382 tmpbuf = NULL;
1383 TAILQ_INIT(¬elst);
1384
1385 /* Size the program segments. */
1386 seginfo.count = 0;
1387 seginfo.size = 0;
1388 each_writable_segment(td, cb_size_segment, &seginfo);
1389
1390 /*
1391 * Collect info about the core file header area.
1392 */
1393 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1394 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1395 coresize = round_page(hdrsize + notesz) + seginfo.size;
1396
1397 /* Set up core dump parameters. */
1398 params.offset = 0;
1399 params.active_cred = cred;
1400 params.file_cred = NOCRED;
1401 params.td = td;
1402 params.vp = vp;
1403 params.gzs = NULL;
1404
1405 #ifdef RACCT
1406 if (racct_enable) {
1407 PROC_LOCK(td->td_proc);
1408 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1409 PROC_UNLOCK(td->td_proc);
1410 if (error != 0) {
1411 error = EFAULT;
1412 goto done;
1413 }
1414 }
1415 #endif
1416 if (coresize >= limit) {
1417 error = EFAULT;
1418 goto done;
1419 }
1420
1421 #ifdef GZIO
1422 /* Create a compression stream if necessary. */
1423 if (compress) {
1424 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1425 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1426 if (params.gzs == NULL) {
1427 error = EFAULT;
1428 goto done;
1429 }
1430 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1431 }
1432 #endif
1433
1434 /*
1435 * Allocate memory for building the header, fill it up,
1436 * and write it out following the notes.
1437 */
1438 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1439 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1440 notesz);
1441
1442 /* Write the contents of all of the writable segments. */
1443 if (error == 0) {
1444 Elf_Phdr *php;
1445 off_t offset;
1446 int i;
1447
1448 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1449 offset = round_page(hdrsize + notesz);
1450 for (i = 0; i < seginfo.count; i++) {
1451 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1452 php->p_filesz, offset, ¶ms, tmpbuf);
1453 if (error != 0)
1454 break;
1455 offset += php->p_filesz;
1456 php++;
1457 }
1458 #ifdef GZIO
1459 if (error == 0 && compress)
1460 error = gzio_flush(params.gzs);
1461 #endif
1462 }
1463 if (error) {
1464 log(LOG_WARNING,
1465 "Failed to write core file for process %s (error %d)\n",
1466 curproc->p_comm, error);
1467 }
1468
1469 done:
1470 #ifdef GZIO
1471 if (compress) {
1472 free(tmpbuf, M_TEMP);
1473 if (params.gzs != NULL)
1474 gzio_fini(params.gzs);
1475 }
1476 #endif
1477 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1478 TAILQ_REMOVE(¬elst, ninfo, link);
1479 free(ninfo, M_TEMP);
1480 }
1481 if (hdr != NULL)
1482 free(hdr, M_TEMP);
1483
1484 return (error);
1485 }
1486
1487 /*
1488 * A callback for each_writable_segment() to write out the segment's
1489 * program header entry.
1490 */
1491 static void
1492 cb_put_phdr(entry, closure)
1493 vm_map_entry_t entry;
1494 void *closure;
1495 {
1496 struct phdr_closure *phc = (struct phdr_closure *)closure;
1497 Elf_Phdr *phdr = phc->phdr;
1498
1499 phc->offset = round_page(phc->offset);
1500
1501 phdr->p_type = PT_LOAD;
1502 phdr->p_offset = phc->offset;
1503 phdr->p_vaddr = entry->start;
1504 phdr->p_paddr = 0;
1505 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1506 phdr->p_align = PAGE_SIZE;
1507 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1508
1509 phc->offset += phdr->p_filesz;
1510 phc->phdr++;
1511 }
1512
1513 /*
1514 * A callback for each_writable_segment() to gather information about
1515 * the number of segments and their total size.
1516 */
1517 static void
1518 cb_size_segment(entry, closure)
1519 vm_map_entry_t entry;
1520 void *closure;
1521 {
1522 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1523
1524 ssc->count++;
1525 ssc->size += entry->end - entry->start;
1526 }
1527
1528 /*
1529 * For each writable segment in the process's memory map, call the given
1530 * function with a pointer to the map entry and some arbitrary
1531 * caller-supplied data.
1532 */
1533 static void
1534 each_writable_segment(td, func, closure)
1535 struct thread *td;
1536 segment_callback func;
1537 void *closure;
1538 {
1539 struct proc *p = td->td_proc;
1540 vm_map_t map = &p->p_vmspace->vm_map;
1541 vm_map_entry_t entry;
1542 vm_object_t backing_object, object;
1543 boolean_t ignore_entry;
1544
1545 vm_map_lock_read(map);
1546 for (entry = map->header.next; entry != &map->header;
1547 entry = entry->next) {
1548 /*
1549 * Don't dump inaccessible mappings, deal with legacy
1550 * coredump mode.
1551 *
1552 * Note that read-only segments related to the elf binary
1553 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1554 * need to arbitrarily ignore such segments.
1555 */
1556 if (elf_legacy_coredump) {
1557 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1558 continue;
1559 } else {
1560 if ((entry->protection & VM_PROT_ALL) == 0)
1561 continue;
1562 }
1563
1564 /*
1565 * Dont include memory segment in the coredump if
1566 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1567 * madvise(2). Do not dump submaps (i.e. parts of the
1568 * kernel map).
1569 */
1570 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1571 continue;
1572
1573 if ((object = entry->object.vm_object) == NULL)
1574 continue;
1575
1576 /* Ignore memory-mapped devices and such things. */
1577 VM_OBJECT_RLOCK(object);
1578 while ((backing_object = object->backing_object) != NULL) {
1579 VM_OBJECT_RLOCK(backing_object);
1580 VM_OBJECT_RUNLOCK(object);
1581 object = backing_object;
1582 }
1583 ignore_entry = object->type != OBJT_DEFAULT &&
1584 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1585 object->type != OBJT_PHYS;
1586 VM_OBJECT_RUNLOCK(object);
1587 if (ignore_entry)
1588 continue;
1589
1590 (*func)(entry, closure);
1591 }
1592 vm_map_unlock_read(map);
1593 }
1594
1595 /*
1596 * Write the core file header to the file, including padding up to
1597 * the page boundary.
1598 */
1599 static int
1600 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1601 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1602 {
1603 struct note_info *ninfo;
1604 struct sbuf *sb;
1605 int error;
1606
1607 /* Fill in the header. */
1608 bzero(hdr, hdrsize);
1609 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1610
1611 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1612 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1613 sbuf_start_section(sb, NULL);
1614 sbuf_bcat(sb, hdr, hdrsize);
1615 TAILQ_FOREACH(ninfo, notelst, link)
1616 __elfN(putnote)(ninfo, sb);
1617 /* Align up to a page boundary for the program segments. */
1618 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1619 error = sbuf_finish(sb);
1620 sbuf_delete(sb);
1621
1622 return (error);
1623 }
1624
1625 static void
1626 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1627 size_t *sizep)
1628 {
1629 struct proc *p;
1630 struct thread *thr;
1631 size_t size;
1632
1633 p = td->td_proc;
1634 size = 0;
1635
1636 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1637
1638 /*
1639 * To have the debugger select the right thread (LWP) as the initial
1640 * thread, we dump the state of the thread passed to us in td first.
1641 * This is the thread that causes the core dump and thus likely to
1642 * be the right thread one wants to have selected in the debugger.
1643 */
1644 thr = td;
1645 while (thr != NULL) {
1646 size += register_note(list, NT_PRSTATUS,
1647 __elfN(note_prstatus), thr);
1648 size += register_note(list, NT_FPREGSET,
1649 __elfN(note_fpregset), thr);
1650 size += register_note(list, NT_THRMISC,
1651 __elfN(note_thrmisc), thr);
1652 size += register_note(list, NT_PTLWPINFO,
1653 __elfN(note_ptlwpinfo), thr);
1654 size += register_note(list, -1,
1655 __elfN(note_threadmd), thr);
1656
1657 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1658 TAILQ_NEXT(thr, td_plist);
1659 if (thr == td)
1660 thr = TAILQ_NEXT(thr, td_plist);
1661 }
1662
1663 size += register_note(list, NT_PROCSTAT_PROC,
1664 __elfN(note_procstat_proc), p);
1665 size += register_note(list, NT_PROCSTAT_FILES,
1666 note_procstat_files, p);
1667 size += register_note(list, NT_PROCSTAT_VMMAP,
1668 note_procstat_vmmap, p);
1669 size += register_note(list, NT_PROCSTAT_GROUPS,
1670 note_procstat_groups, p);
1671 size += register_note(list, NT_PROCSTAT_UMASK,
1672 note_procstat_umask, p);
1673 size += register_note(list, NT_PROCSTAT_RLIMIT,
1674 note_procstat_rlimit, p);
1675 size += register_note(list, NT_PROCSTAT_OSREL,
1676 note_procstat_osrel, p);
1677 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1678 __elfN(note_procstat_psstrings), p);
1679 size += register_note(list, NT_PROCSTAT_AUXV,
1680 __elfN(note_procstat_auxv), p);
1681
1682 *sizep = size;
1683 }
1684
1685 static void
1686 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1687 size_t notesz)
1688 {
1689 Elf_Ehdr *ehdr;
1690 Elf_Phdr *phdr;
1691 struct phdr_closure phc;
1692
1693 ehdr = (Elf_Ehdr *)hdr;
1694 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1695
1696 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1697 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1698 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1699 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1700 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1701 ehdr->e_ident[EI_DATA] = ELF_DATA;
1702 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1703 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1704 ehdr->e_ident[EI_ABIVERSION] = 0;
1705 ehdr->e_ident[EI_PAD] = 0;
1706 ehdr->e_type = ET_CORE;
1707 ehdr->e_machine = td->td_proc->p_elf_machine;
1708 ehdr->e_version = EV_CURRENT;
1709 ehdr->e_entry = 0;
1710 ehdr->e_phoff = sizeof(Elf_Ehdr);
1711 ehdr->e_flags = td->td_proc->p_elf_flags;
1712 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1713 ehdr->e_phentsize = sizeof(Elf_Phdr);
1714 ehdr->e_phnum = numsegs + 1;
1715 ehdr->e_shentsize = sizeof(Elf_Shdr);
1716 ehdr->e_shnum = 0;
1717 ehdr->e_shstrndx = SHN_UNDEF;
1718
1719 /*
1720 * Fill in the program header entries.
1721 */
1722
1723 /* The note segement. */
1724 phdr->p_type = PT_NOTE;
1725 phdr->p_offset = hdrsize;
1726 phdr->p_vaddr = 0;
1727 phdr->p_paddr = 0;
1728 phdr->p_filesz = notesz;
1729 phdr->p_memsz = 0;
1730 phdr->p_flags = PF_R;
1731 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1732 phdr++;
1733
1734 /* All the writable segments from the program. */
1735 phc.phdr = phdr;
1736 phc.offset = round_page(hdrsize + notesz);
1737 each_writable_segment(td, cb_put_phdr, &phc);
1738 }
1739
1740 static size_t
1741 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1742 {
1743 struct note_info *ninfo;
1744 size_t size, notesize;
1745
1746 size = 0;
1747 out(arg, NULL, &size);
1748 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1749 ninfo->type = type;
1750 ninfo->outfunc = out;
1751 ninfo->outarg = arg;
1752 ninfo->outsize = size;
1753 TAILQ_INSERT_TAIL(list, ninfo, link);
1754
1755 if (type == -1)
1756 return (size);
1757
1758 notesize = sizeof(Elf_Note) + /* note header */
1759 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1760 /* note name */
1761 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1762
1763 return (notesize);
1764 }
1765
1766 static size_t
1767 append_note_data(const void *src, void *dst, size_t len)
1768 {
1769 size_t padded_len;
1770
1771 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1772 if (dst != NULL) {
1773 bcopy(src, dst, len);
1774 bzero((char *)dst + len, padded_len - len);
1775 }
1776 return (padded_len);
1777 }
1778
1779 size_t
1780 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1781 {
1782 Elf_Note *note;
1783 char *buf;
1784 size_t notesize;
1785
1786 buf = dst;
1787 if (buf != NULL) {
1788 note = (Elf_Note *)buf;
1789 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1790 note->n_descsz = size;
1791 note->n_type = type;
1792 buf += sizeof(*note);
1793 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1794 sizeof(FREEBSD_ABI_VENDOR));
1795 append_note_data(src, buf, size);
1796 if (descp != NULL)
1797 *descp = buf;
1798 }
1799
1800 notesize = sizeof(Elf_Note) + /* note header */
1801 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1802 /* note name */
1803 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1804
1805 return (notesize);
1806 }
1807
1808 static void
1809 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1810 {
1811 Elf_Note note;
1812 ssize_t old_len, sect_len;
1813 size_t new_len, descsz, i;
1814
1815 if (ninfo->type == -1) {
1816 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1817 return;
1818 }
1819
1820 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1821 note.n_descsz = ninfo->outsize;
1822 note.n_type = ninfo->type;
1823
1824 sbuf_bcat(sb, ¬e, sizeof(note));
1825 sbuf_start_section(sb, &old_len);
1826 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1827 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1828 if (note.n_descsz == 0)
1829 return;
1830 sbuf_start_section(sb, &old_len);
1831 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1832 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1833 if (sect_len < 0)
1834 return;
1835
1836 new_len = (size_t)sect_len;
1837 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1838 if (new_len < descsz) {
1839 /*
1840 * It is expected that individual note emitters will correctly
1841 * predict their expected output size and fill up to that size
1842 * themselves, padding in a format-specific way if needed.
1843 * However, in case they don't, just do it here with zeros.
1844 */
1845 for (i = 0; i < descsz - new_len; i++)
1846 sbuf_putc(sb, 0);
1847 } else if (new_len > descsz) {
1848 /*
1849 * We can't always truncate sb -- we may have drained some
1850 * of it already.
1851 */
1852 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1853 "read it (%zu > %zu). Since it is longer than "
1854 "expected, this coredump's notes are corrupt. THIS "
1855 "IS A BUG in the note_procstat routine for type %u.\n",
1856 __func__, (unsigned)note.n_type, new_len, descsz,
1857 (unsigned)note.n_type));
1858 }
1859 }
1860
1861 /*
1862 * Miscellaneous note out functions.
1863 */
1864
1865 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1866 #include <compat/freebsd32/freebsd32.h>
1867 #include <compat/freebsd32/freebsd32_signal.h>
1868
1869 typedef struct prstatus32 elf_prstatus_t;
1870 typedef struct prpsinfo32 elf_prpsinfo_t;
1871 typedef struct fpreg32 elf_prfpregset_t;
1872 typedef struct fpreg32 elf_fpregset_t;
1873 typedef struct reg32 elf_gregset_t;
1874 typedef struct thrmisc32 elf_thrmisc_t;
1875 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1876 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1877 typedef uint32_t elf_ps_strings_t;
1878 #else
1879 typedef prstatus_t elf_prstatus_t;
1880 typedef prpsinfo_t elf_prpsinfo_t;
1881 typedef prfpregset_t elf_prfpregset_t;
1882 typedef prfpregset_t elf_fpregset_t;
1883 typedef gregset_t elf_gregset_t;
1884 typedef thrmisc_t elf_thrmisc_t;
1885 #define ELF_KERN_PROC_MASK 0
1886 typedef struct kinfo_proc elf_kinfo_proc_t;
1887 typedef vm_offset_t elf_ps_strings_t;
1888 #endif
1889
1890 static void
1891 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1892 {
1893 struct sbuf sbarg;
1894 size_t len;
1895 char *cp, *end;
1896 struct proc *p;
1897 elf_prpsinfo_t *psinfo;
1898 int error;
1899
1900 p = (struct proc *)arg;
1901 if (sb != NULL) {
1902 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1903 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1904 psinfo->pr_version = PRPSINFO_VERSION;
1905 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1906 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1907 PROC_LOCK(p);
1908 if (p->p_args != NULL) {
1909 len = sizeof(psinfo->pr_psargs) - 1;
1910 if (len > p->p_args->ar_length)
1911 len = p->p_args->ar_length;
1912 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1913 PROC_UNLOCK(p);
1914 error = 0;
1915 } else {
1916 _PHOLD(p);
1917 PROC_UNLOCK(p);
1918 sbuf_new(&sbarg, psinfo->pr_psargs,
1919 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1920 error = proc_getargv(curthread, p, &sbarg);
1921 PRELE(p);
1922 if (sbuf_finish(&sbarg) == 0)
1923 len = sbuf_len(&sbarg) - 1;
1924 else
1925 len = sizeof(psinfo->pr_psargs) - 1;
1926 sbuf_delete(&sbarg);
1927 }
1928 if (error || len == 0)
1929 strlcpy(psinfo->pr_psargs, p->p_comm,
1930 sizeof(psinfo->pr_psargs));
1931 else {
1932 KASSERT(len < sizeof(psinfo->pr_psargs),
1933 ("len is too long: %zu vs %zu", len,
1934 sizeof(psinfo->pr_psargs)));
1935 cp = psinfo->pr_psargs;
1936 end = cp + len - 1;
1937 for (;;) {
1938 cp = memchr(cp, '\0', end - cp);
1939 if (cp == NULL)
1940 break;
1941 *cp = ' ';
1942 }
1943 }
1944 psinfo->pr_pid = p->p_pid;
1945 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1946 free(psinfo, M_TEMP);
1947 }
1948 *sizep = sizeof(*psinfo);
1949 }
1950
1951 static void
1952 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1953 {
1954 struct thread *td;
1955 elf_prstatus_t *status;
1956
1957 td = (struct thread *)arg;
1958 if (sb != NULL) {
1959 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1960 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1961 status->pr_version = PRSTATUS_VERSION;
1962 status->pr_statussz = sizeof(elf_prstatus_t);
1963 status->pr_gregsetsz = sizeof(elf_gregset_t);
1964 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1965 status->pr_osreldate = osreldate;
1966 status->pr_cursig = td->td_proc->p_sig;
1967 status->pr_pid = td->td_tid;
1968 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1969 fill_regs32(td, &status->pr_reg);
1970 #else
1971 fill_regs(td, &status->pr_reg);
1972 #endif
1973 sbuf_bcat(sb, status, sizeof(*status));
1974 free(status, M_TEMP);
1975 }
1976 *sizep = sizeof(*status);
1977 }
1978
1979 static void
1980 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1981 {
1982 struct thread *td;
1983 elf_prfpregset_t *fpregset;
1984
1985 td = (struct thread *)arg;
1986 if (sb != NULL) {
1987 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1988 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1989 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1990 fill_fpregs32(td, fpregset);
1991 #else
1992 fill_fpregs(td, fpregset);
1993 #endif
1994 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1995 free(fpregset, M_TEMP);
1996 }
1997 *sizep = sizeof(*fpregset);
1998 }
1999
2000 static void
2001 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
2002 {
2003 struct thread *td;
2004 elf_thrmisc_t thrmisc;
2005
2006 td = (struct thread *)arg;
2007 if (sb != NULL) {
2008 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
2009 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
2010 strcpy(thrmisc.pr_tname, td->td_name);
2011 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
2012 }
2013 *sizep = sizeof(thrmisc);
2014 }
2015
2016 static void
2017 __elfN(note_ptlwpinfo)(void *arg, struct sbuf *sb, size_t *sizep)
2018 {
2019 struct thread *td;
2020 size_t size;
2021 int structsize;
2022 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2023 struct ptrace_lwpinfo32 pl;
2024 #else
2025 struct ptrace_lwpinfo pl;
2026 #endif
2027
2028 td = (struct thread *)arg;
2029 size = sizeof(structsize) + sizeof(pl);
2030 if (sb != NULL) {
2031 KASSERT(*sizep == size, ("invalid size"));
2032 structsize = sizeof(pl);
2033 sbuf_bcat(sb, &structsize, sizeof(structsize));
2034 bzero(&pl, sizeof(pl));
2035 pl.pl_lwpid = td->td_tid;
2036 pl.pl_event = PL_EVENT_NONE;
2037 pl.pl_sigmask = td->td_sigmask;
2038 pl.pl_siglist = td->td_siglist;
2039 if (td->td_si.si_signo != 0) {
2040 pl.pl_event = PL_EVENT_SIGNAL;
2041 pl.pl_flags |= PL_FLAG_SI;
2042 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2043 siginfo_to_siginfo32(&td->td_si, &pl.pl_siginfo);
2044 #else
2045 pl.pl_siginfo = td->td_si;
2046 #endif
2047 }
2048 strcpy(pl.pl_tdname, td->td_name);
2049 /* XXX TODO: supply more information in struct ptrace_lwpinfo*/
2050 sbuf_bcat(sb, &pl, sizeof(pl));
2051 }
2052 *sizep = size;
2053 }
2054
2055 /*
2056 * Allow for MD specific notes, as well as any MD
2057 * specific preparations for writing MI notes.
2058 */
2059 static void
2060 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2061 {
2062 struct thread *td;
2063 void *buf;
2064 size_t size;
2065
2066 td = (struct thread *)arg;
2067 size = *sizep;
2068 if (size != 0 && sb != NULL)
2069 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2070 else
2071 buf = NULL;
2072 size = 0;
2073 __elfN(dump_thread)(td, buf, &size);
2074 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2075 if (size != 0 && sb != NULL)
2076 sbuf_bcat(sb, buf, size);
2077 free(buf, M_TEMP);
2078 *sizep = size;
2079 }
2080
2081 #ifdef KINFO_PROC_SIZE
2082 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2083 #endif
2084
2085 static void
2086 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2087 {
2088 struct proc *p;
2089 size_t size;
2090 int structsize;
2091
2092 p = (struct proc *)arg;
2093 size = sizeof(structsize) + p->p_numthreads *
2094 sizeof(elf_kinfo_proc_t);
2095
2096 if (sb != NULL) {
2097 KASSERT(*sizep == size, ("invalid size"));
2098 structsize = sizeof(elf_kinfo_proc_t);
2099 sbuf_bcat(sb, &structsize, sizeof(structsize));
2100 sx_slock(&proctree_lock);
2101 PROC_LOCK(p);
2102 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2103 sx_sunlock(&proctree_lock);
2104 }
2105 *sizep = size;
2106 }
2107
2108 #ifdef KINFO_FILE_SIZE
2109 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2110 #endif
2111
2112 static void
2113 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2114 {
2115 struct proc *p;
2116 size_t size, sect_sz, i;
2117 ssize_t start_len, sect_len;
2118 int structsize, filedesc_flags;
2119
2120 if (coredump_pack_fileinfo)
2121 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2122 else
2123 filedesc_flags = 0;
2124
2125 p = (struct proc *)arg;
2126 structsize = sizeof(struct kinfo_file);
2127 if (sb == NULL) {
2128 size = 0;
2129 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2130 sbuf_set_drain(sb, sbuf_drain_count, &size);
2131 sbuf_bcat(sb, &structsize, sizeof(structsize));
2132 PROC_LOCK(p);
2133 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2134 sbuf_finish(sb);
2135 sbuf_delete(sb);
2136 *sizep = size;
2137 } else {
2138 sbuf_start_section(sb, &start_len);
2139
2140 sbuf_bcat(sb, &structsize, sizeof(structsize));
2141 PROC_LOCK(p);
2142 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2143 filedesc_flags);
2144
2145 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2146 if (sect_len < 0)
2147 return;
2148 sect_sz = sect_len;
2149
2150 KASSERT(sect_sz <= *sizep,
2151 ("kern_proc_filedesc_out did not respect maxlen; "
2152 "requested %zu, got %zu", *sizep - sizeof(structsize),
2153 sect_sz - sizeof(structsize)));
2154
2155 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2156 sbuf_putc(sb, 0);
2157 }
2158 }
2159
2160 #ifdef KINFO_VMENTRY_SIZE
2161 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2162 #endif
2163
2164 static void
2165 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2166 {
2167 struct proc *p;
2168 size_t size;
2169 int structsize, vmmap_flags;
2170
2171 if (coredump_pack_vmmapinfo)
2172 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2173 else
2174 vmmap_flags = 0;
2175
2176 p = (struct proc *)arg;
2177 structsize = sizeof(struct kinfo_vmentry);
2178 if (sb == NULL) {
2179 size = 0;
2180 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2181 sbuf_set_drain(sb, sbuf_drain_count, &size);
2182 sbuf_bcat(sb, &structsize, sizeof(structsize));
2183 PROC_LOCK(p);
2184 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2185 sbuf_finish(sb);
2186 sbuf_delete(sb);
2187 *sizep = size;
2188 } else {
2189 sbuf_bcat(sb, &structsize, sizeof(structsize));
2190 PROC_LOCK(p);
2191 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2192 vmmap_flags);
2193 }
2194 }
2195
2196 static void
2197 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2198 {
2199 struct proc *p;
2200 size_t size;
2201 int structsize;
2202
2203 p = (struct proc *)arg;
2204 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2205 if (sb != NULL) {
2206 KASSERT(*sizep == size, ("invalid size"));
2207 structsize = sizeof(gid_t);
2208 sbuf_bcat(sb, &structsize, sizeof(structsize));
2209 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2210 sizeof(gid_t));
2211 }
2212 *sizep = size;
2213 }
2214
2215 static void
2216 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2217 {
2218 struct proc *p;
2219 size_t size;
2220 int structsize;
2221
2222 p = (struct proc *)arg;
2223 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2224 if (sb != NULL) {
2225 KASSERT(*sizep == size, ("invalid size"));
2226 structsize = sizeof(p->p_fd->fd_cmask);
2227 sbuf_bcat(sb, &structsize, sizeof(structsize));
2228 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2229 }
2230 *sizep = size;
2231 }
2232
2233 static void
2234 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2235 {
2236 struct proc *p;
2237 struct rlimit rlim[RLIM_NLIMITS];
2238 size_t size;
2239 int structsize, i;
2240
2241 p = (struct proc *)arg;
2242 size = sizeof(structsize) + sizeof(rlim);
2243 if (sb != NULL) {
2244 KASSERT(*sizep == size, ("invalid size"));
2245 structsize = sizeof(rlim);
2246 sbuf_bcat(sb, &structsize, sizeof(structsize));
2247 PROC_LOCK(p);
2248 for (i = 0; i < RLIM_NLIMITS; i++)
2249 lim_rlimit_proc(p, i, &rlim[i]);
2250 PROC_UNLOCK(p);
2251 sbuf_bcat(sb, rlim, sizeof(rlim));
2252 }
2253 *sizep = size;
2254 }
2255
2256 static void
2257 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2258 {
2259 struct proc *p;
2260 size_t size;
2261 int structsize;
2262
2263 p = (struct proc *)arg;
2264 size = sizeof(structsize) + sizeof(p->p_osrel);
2265 if (sb != NULL) {
2266 KASSERT(*sizep == size, ("invalid size"));
2267 structsize = sizeof(p->p_osrel);
2268 sbuf_bcat(sb, &structsize, sizeof(structsize));
2269 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2270 }
2271 *sizep = size;
2272 }
2273
2274 static void
2275 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2276 {
2277 struct proc *p;
2278 elf_ps_strings_t ps_strings;
2279 size_t size;
2280 int structsize;
2281
2282 p = (struct proc *)arg;
2283 size = sizeof(structsize) + sizeof(ps_strings);
2284 if (sb != NULL) {
2285 KASSERT(*sizep == size, ("invalid size"));
2286 structsize = sizeof(ps_strings);
2287 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2288 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2289 #else
2290 ps_strings = p->p_sysent->sv_psstrings;
2291 #endif
2292 sbuf_bcat(sb, &structsize, sizeof(structsize));
2293 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2294 }
2295 *sizep = size;
2296 }
2297
2298 static void
2299 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2300 {
2301 struct proc *p;
2302 size_t size;
2303 int structsize;
2304
2305 p = (struct proc *)arg;
2306 if (sb == NULL) {
2307 size = 0;
2308 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2309 sbuf_set_drain(sb, sbuf_drain_count, &size);
2310 sbuf_bcat(sb, &structsize, sizeof(structsize));
2311 PHOLD(p);
2312 proc_getauxv(curthread, p, sb);
2313 PRELE(p);
2314 sbuf_finish(sb);
2315 sbuf_delete(sb);
2316 *sizep = size;
2317 } else {
2318 structsize = sizeof(Elf_Auxinfo);
2319 sbuf_bcat(sb, &structsize, sizeof(structsize));
2320 PHOLD(p);
2321 proc_getauxv(curthread, p, sb);
2322 PRELE(p);
2323 }
2324 }
2325
2326 static boolean_t
2327 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2328 int32_t *osrel, const Elf_Phdr *pnote)
2329 {
2330 const Elf_Note *note, *note0, *note_end;
2331 const char *note_name;
2332 char *buf;
2333 int i, error;
2334 boolean_t res;
2335
2336 /* We need some limit, might as well use PAGE_SIZE. */
2337 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2338 return (FALSE);
2339 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2340 if (pnote->p_offset > PAGE_SIZE ||
2341 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2342 VOP_UNLOCK(imgp->vp, 0);
2343 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2344 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2345 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2346 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2347 curthread->td_ucred, NOCRED, NULL, curthread);
2348 if (error != 0) {
2349 uprintf("i/o error PT_NOTE\n");
2350 res = FALSE;
2351 goto ret;
2352 }
2353 note = note0 = (const Elf_Note *)buf;
2354 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2355 } else {
2356 note = note0 = (const Elf_Note *)(imgp->image_header +
2357 pnote->p_offset);
2358 note_end = (const Elf_Note *)(imgp->image_header +
2359 pnote->p_offset + pnote->p_filesz);
2360 buf = NULL;
2361 }
2362 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2363 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2364 (const char *)note < sizeof(Elf_Note)) {
2365 res = FALSE;
2366 goto ret;
2367 }
2368 if (note->n_namesz != checknote->hdr.n_namesz ||
2369 note->n_descsz != checknote->hdr.n_descsz ||
2370 note->n_type != checknote->hdr.n_type)
2371 goto nextnote;
2372 note_name = (const char *)(note + 1);
2373 if (note_name + checknote->hdr.n_namesz >=
2374 (const char *)note_end || strncmp(checknote->vendor,
2375 note_name, checknote->hdr.n_namesz) != 0)
2376 goto nextnote;
2377
2378 /*
2379 * Fetch the osreldate for binary
2380 * from the ELF OSABI-note if necessary.
2381 */
2382 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2383 checknote->trans_osrel != NULL) {
2384 res = checknote->trans_osrel(note, osrel);
2385 goto ret;
2386 }
2387 res = TRUE;
2388 goto ret;
2389 nextnote:
2390 note = (const Elf_Note *)((const char *)(note + 1) +
2391 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2392 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2393 }
2394 res = FALSE;
2395 ret:
2396 free(buf, M_TEMP);
2397 return (res);
2398 }
2399
2400 /*
2401 * Try to find the appropriate ABI-note section for checknote,
2402 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2403 * first page of the image is searched, the same as for headers.
2404 */
2405 static boolean_t
2406 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2407 int32_t *osrel)
2408 {
2409 const Elf_Phdr *phdr;
2410 const Elf_Ehdr *hdr;
2411 int i;
2412
2413 hdr = (const Elf_Ehdr *)imgp->image_header;
2414 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2415
2416 for (i = 0; i < hdr->e_phnum; i++) {
2417 if (phdr[i].p_type == PT_NOTE &&
2418 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2419 return (TRUE);
2420 }
2421 return (FALSE);
2422
2423 }
2424
2425 /*
2426 * Tell kern_execve.c about it, with a little help from the linker.
2427 */
2428 static struct execsw __elfN(execsw) = {
2429 __CONCAT(exec_, __elfN(imgact)),
2430 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2431 };
2432 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2433
2434 static vm_prot_t
2435 __elfN(trans_prot)(Elf_Word flags)
2436 {
2437 vm_prot_t prot;
2438
2439 prot = 0;
2440 if (flags & PF_X)
2441 prot |= VM_PROT_EXECUTE;
2442 if (flags & PF_W)
2443 prot |= VM_PROT_WRITE;
2444 if (flags & PF_R)
2445 prot |= VM_PROT_READ;
2446 #if __ELF_WORD_SIZE == 32
2447 #if defined(__amd64__)
2448 if (i386_read_exec && (flags & PF_R))
2449 prot |= VM_PROT_EXECUTE;
2450 #endif
2451 #endif
2452 return (prot);
2453 }
2454
2455 static Elf_Word
2456 __elfN(untrans_prot)(vm_prot_t prot)
2457 {
2458 Elf_Word flags;
2459
2460 flags = 0;
2461 if (prot & VM_PROT_EXECUTE)
2462 flags |= PF_X;
2463 if (prot & VM_PROT_READ)
2464 flags |= PF_R;
2465 if (prot & VM_PROT_WRITE)
2466 flags |= PF_W;
2467 return (flags);
2468 }
Cache object: 33c09ebe29b1de1595644e1eddb21bf0
|