FreeBSD/Linux Kernel Cross Reference
sys/kern/imgact_elf.c
1 /*-
2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD: releng/11.1/sys/kern/imgact_elf.c 338606 2018-09-12 05:07:35Z gordon $");
33
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_gzio.h"
37
38 #include <sys/param.h>
39 #include <sys/capsicum.h>
40 #include <sys/exec.h>
41 #include <sys/fcntl.h>
42 #include <sys/gzio.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/jail.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mman.h>
51 #include <sys/namei.h>
52 #include <sys/pioctl.h>
53 #include <sys/proc.h>
54 #include <sys/procfs.h>
55 #include <sys/racct.h>
56 #include <sys/resourcevar.h>
57 #include <sys/rwlock.h>
58 #include <sys/sbuf.h>
59 #include <sys/sf_buf.h>
60 #include <sys/smp.h>
61 #include <sys/systm.h>
62 #include <sys/signalvar.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/syscall.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/vnode.h>
69 #include <sys/syslog.h>
70 #include <sys/eventhandler.h>
71 #include <sys/user.h>
72
73 #include <vm/vm.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_extern.h>
80
81 #include <machine/elf.h>
82 #include <machine/md_var.h>
83
84 #define ELF_NOTE_ROUNDSIZE 4
85 #define OLD_EI_BRAND 8
86
87 static int __elfN(check_header)(const Elf_Ehdr *hdr);
88 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
89 const char *interp, int interp_name_len, int32_t *osrel);
90 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
91 u_long *entry, size_t pagesize);
92 static int __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
93 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
94 size_t pagesize);
95 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
96 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
97 int32_t *osrel);
98 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
99 static boolean_t __elfN(check_note)(struct image_params *imgp,
100 Elf_Brandnote *checknote, int32_t *osrel);
101 static vm_prot_t __elfN(trans_prot)(Elf_Word);
102 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
103
104 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
105 "");
106
107 #define CORE_BUF_SIZE (16 * 1024)
108
109 int __elfN(fallback_brand) = -1;
110 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
111 fallback_brand, CTLFLAG_RWTUN, &__elfN(fallback_brand), 0,
112 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
113
114 static int elf_legacy_coredump = 0;
115 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
116 &elf_legacy_coredump, 0,
117 "include all and only RW pages in core dumps");
118
119 int __elfN(nxstack) =
120 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */ || \
121 (defined(__arm__) && __ARM_ARCH >= 7) || defined(__aarch64__)
122 1;
123 #else
124 0;
125 #endif
126 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
127 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
128 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
129
130 #if __ELF_WORD_SIZE == 32
131 #if defined(__amd64__)
132 int i386_read_exec = 0;
133 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
134 "enable execution from readable segments");
135 #endif
136 #endif
137
138 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
139
140 #define trunc_page_ps(va, ps) rounddown2(va, ps)
141 #define round_page_ps(va, ps) roundup2(va, ps)
142 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
143
144 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
145
146 Elf_Brandnote __elfN(freebsd_brandnote) = {
147 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
148 .hdr.n_descsz = sizeof(int32_t),
149 .hdr.n_type = NT_FREEBSD_ABI_TAG,
150 .vendor = FREEBSD_ABI_VENDOR,
151 .flags = BN_TRANSLATE_OSREL,
152 .trans_osrel = __elfN(freebsd_trans_osrel)
153 };
154
155 static boolean_t
156 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
157 {
158 uintptr_t p;
159
160 p = (uintptr_t)(note + 1);
161 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
162 *osrel = *(const int32_t *)(p);
163
164 return (TRUE);
165 }
166
167 static const char GNU_ABI_VENDOR[] = "GNU";
168 static int GNU_KFREEBSD_ABI_DESC = 3;
169
170 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
171 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
172 .hdr.n_descsz = 16, /* XXX at least 16 */
173 .hdr.n_type = 1,
174 .vendor = GNU_ABI_VENDOR,
175 .flags = BN_TRANSLATE_OSREL,
176 .trans_osrel = kfreebsd_trans_osrel
177 };
178
179 static boolean_t
180 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
181 {
182 const Elf32_Word *desc;
183 uintptr_t p;
184
185 p = (uintptr_t)(note + 1);
186 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
187
188 desc = (const Elf32_Word *)p;
189 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
190 return (FALSE);
191
192 /*
193 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
194 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
195 */
196 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
197
198 return (TRUE);
199 }
200
201 int
202 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
203 {
204 int i;
205
206 for (i = 0; i < MAX_BRANDS; i++) {
207 if (elf_brand_list[i] == NULL) {
208 elf_brand_list[i] = entry;
209 break;
210 }
211 }
212 if (i == MAX_BRANDS) {
213 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
214 __func__, entry);
215 return (-1);
216 }
217 return (0);
218 }
219
220 int
221 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
222 {
223 int i;
224
225 for (i = 0; i < MAX_BRANDS; i++) {
226 if (elf_brand_list[i] == entry) {
227 elf_brand_list[i] = NULL;
228 break;
229 }
230 }
231 if (i == MAX_BRANDS)
232 return (-1);
233 return (0);
234 }
235
236 int
237 __elfN(brand_inuse)(Elf_Brandinfo *entry)
238 {
239 struct proc *p;
240 int rval = FALSE;
241
242 sx_slock(&allproc_lock);
243 FOREACH_PROC_IN_SYSTEM(p) {
244 if (p->p_sysent == entry->sysvec) {
245 rval = TRUE;
246 break;
247 }
248 }
249 sx_sunlock(&allproc_lock);
250
251 return (rval);
252 }
253
254 static Elf_Brandinfo *
255 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
256 int interp_name_len, int32_t *osrel)
257 {
258 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
259 Elf_Brandinfo *bi, *bi_m;
260 boolean_t ret;
261 int i;
262
263 /*
264 * We support four types of branding -- (1) the ELF EI_OSABI field
265 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
266 * branding w/in the ELF header, (3) path of the `interp_path'
267 * field, and (4) the ".note.ABI-tag" ELF section.
268 */
269
270 /* Look for an ".note.ABI-tag" ELF section */
271 bi_m = NULL;
272 for (i = 0; i < MAX_BRANDS; i++) {
273 bi = elf_brand_list[i];
274 if (bi == NULL)
275 continue;
276 if (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0)
277 continue;
278 if (hdr->e_machine == bi->machine && (bi->flags &
279 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
280 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
281 /* Give brand a chance to veto check_note's guess */
282 if (ret && bi->header_supported)
283 ret = bi->header_supported(imgp);
284 /*
285 * If note checker claimed the binary, but the
286 * interpreter path in the image does not
287 * match default one for the brand, try to
288 * search for other brands with the same
289 * interpreter. Either there is better brand
290 * with the right interpreter, or, failing
291 * this, we return first brand which accepted
292 * our note and, optionally, header.
293 */
294 if (ret && bi_m == NULL && interp != NULL &&
295 (bi->interp_path == NULL ||
296 (strlen(bi->interp_path) + 1 != interp_name_len ||
297 strncmp(interp, bi->interp_path, interp_name_len)
298 != 0))) {
299 bi_m = bi;
300 ret = 0;
301 }
302 if (ret)
303 return (bi);
304 }
305 }
306 if (bi_m != NULL)
307 return (bi_m);
308
309 /* If the executable has a brand, search for it in the brand list. */
310 for (i = 0; i < MAX_BRANDS; i++) {
311 bi = elf_brand_list[i];
312 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
313 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
314 continue;
315 if (hdr->e_machine == bi->machine &&
316 (hdr->e_ident[EI_OSABI] == bi->brand ||
317 (bi->compat_3_brand != NULL &&
318 strcmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
319 bi->compat_3_brand) == 0))) {
320 /* Looks good, but give brand a chance to veto */
321 if (!bi->header_supported ||
322 bi->header_supported(imgp)) {
323 /*
324 * Again, prefer strictly matching
325 * interpreter path.
326 */
327 if (interp_name_len == 0 &&
328 bi->interp_path == NULL)
329 return (bi);
330 if (bi->interp_path != NULL &&
331 strlen(bi->interp_path) + 1 ==
332 interp_name_len && strncmp(interp,
333 bi->interp_path, interp_name_len) == 0)
334 return (bi);
335 if (bi_m == NULL)
336 bi_m = bi;
337 }
338 }
339 }
340 if (bi_m != NULL)
341 return (bi_m);
342
343 /* No known brand, see if the header is recognized by any brand */
344 for (i = 0; i < MAX_BRANDS; i++) {
345 bi = elf_brand_list[i];
346 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY ||
347 bi->header_supported == NULL)
348 continue;
349 if (hdr->e_machine == bi->machine) {
350 ret = bi->header_supported(imgp);
351 if (ret)
352 return (bi);
353 }
354 }
355
356 /* Lacking a known brand, search for a recognized interpreter. */
357 if (interp != NULL) {
358 for (i = 0; i < MAX_BRANDS; i++) {
359 bi = elf_brand_list[i];
360 if (bi == NULL || (bi->flags &
361 (BI_BRAND_NOTE_MANDATORY | BI_BRAND_ONLY_STATIC))
362 != 0)
363 continue;
364 if (hdr->e_machine == bi->machine &&
365 bi->interp_path != NULL &&
366 /* ELF image p_filesz includes terminating zero */
367 strlen(bi->interp_path) + 1 == interp_name_len &&
368 strncmp(interp, bi->interp_path, interp_name_len)
369 == 0)
370 return (bi);
371 }
372 }
373
374 /* Lacking a recognized interpreter, try the default brand */
375 for (i = 0; i < MAX_BRANDS; i++) {
376 bi = elf_brand_list[i];
377 if (bi == NULL || (bi->flags & BI_BRAND_NOTE_MANDATORY) != 0 ||
378 (interp != NULL && (bi->flags & BI_BRAND_ONLY_STATIC) != 0))
379 continue;
380 if (hdr->e_machine == bi->machine &&
381 __elfN(fallback_brand) == bi->brand)
382 return (bi);
383 }
384 return (NULL);
385 }
386
387 static int
388 __elfN(check_header)(const Elf_Ehdr *hdr)
389 {
390 Elf_Brandinfo *bi;
391 int i;
392
393 if (!IS_ELF(*hdr) ||
394 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
395 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
396 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
397 hdr->e_phentsize != sizeof(Elf_Phdr) ||
398 hdr->e_version != ELF_TARG_VER)
399 return (ENOEXEC);
400
401 /*
402 * Make sure we have at least one brand for this machine.
403 */
404
405 for (i = 0; i < MAX_BRANDS; i++) {
406 bi = elf_brand_list[i];
407 if (bi != NULL && bi->machine == hdr->e_machine)
408 break;
409 }
410 if (i == MAX_BRANDS)
411 return (ENOEXEC);
412
413 return (0);
414 }
415
416 static int
417 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
418 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
419 {
420 struct sf_buf *sf;
421 int error;
422 vm_offset_t off;
423
424 /*
425 * Create the page if it doesn't exist yet. Ignore errors.
426 */
427 vm_map_fixed(map, NULL, 0, trunc_page(start), round_page(end) -
428 trunc_page(start), VM_PROT_ALL, VM_PROT_ALL, MAP_CHECK_EXCL);
429
430 /*
431 * Find the page from the underlying object.
432 */
433 if (object != NULL) {
434 sf = vm_imgact_map_page(object, offset);
435 if (sf == NULL)
436 return (KERN_FAILURE);
437 off = offset - trunc_page(offset);
438 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
439 end - start);
440 vm_imgact_unmap_page(sf);
441 if (error != 0)
442 return (KERN_FAILURE);
443 }
444
445 return (KERN_SUCCESS);
446 }
447
448 static int
449 __elfN(map_insert)(struct image_params *imgp, vm_map_t map, vm_object_t object,
450 vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot,
451 int cow)
452 {
453 struct sf_buf *sf;
454 vm_offset_t off;
455 vm_size_t sz;
456 int error, locked, rv;
457
458 if (start != trunc_page(start)) {
459 rv = __elfN(map_partial)(map, object, offset, start,
460 round_page(start), prot);
461 if (rv != KERN_SUCCESS)
462 return (rv);
463 offset += round_page(start) - start;
464 start = round_page(start);
465 }
466 if (end != round_page(end)) {
467 rv = __elfN(map_partial)(map, object, offset +
468 trunc_page(end) - start, trunc_page(end), end, prot);
469 if (rv != KERN_SUCCESS)
470 return (rv);
471 end = trunc_page(end);
472 }
473 if (start >= end)
474 return (KERN_SUCCESS);
475 if ((offset & PAGE_MASK) != 0) {
476 /*
477 * The mapping is not page aligned. This means that we have
478 * to copy the data.
479 */
480 rv = vm_map_fixed(map, NULL, 0, start, end - start,
481 prot | VM_PROT_WRITE, VM_PROT_ALL, MAP_CHECK_EXCL);
482 if (rv != KERN_SUCCESS)
483 return (rv);
484 if (object == NULL)
485 return (KERN_SUCCESS);
486 for (; start < end; start += sz) {
487 sf = vm_imgact_map_page(object, offset);
488 if (sf == NULL)
489 return (KERN_FAILURE);
490 off = offset - trunc_page(offset);
491 sz = end - start;
492 if (sz > PAGE_SIZE - off)
493 sz = PAGE_SIZE - off;
494 error = copyout((caddr_t)sf_buf_kva(sf) + off,
495 (caddr_t)start, sz);
496 vm_imgact_unmap_page(sf);
497 if (error != 0)
498 return (KERN_FAILURE);
499 offset += sz;
500 }
501 } else {
502 vm_object_reference(object);
503 rv = vm_map_fixed(map, object, offset, start, end - start,
504 prot, VM_PROT_ALL, cow | MAP_CHECK_EXCL);
505 if (rv != KERN_SUCCESS) {
506 locked = VOP_ISLOCKED(imgp->vp);
507 VOP_UNLOCK(imgp->vp, 0);
508 vm_object_deallocate(object);
509 vn_lock(imgp->vp, locked | LK_RETRY);
510 return (rv);
511 }
512 }
513 return (KERN_SUCCESS);
514 }
515
516 static int
517 __elfN(load_section)(struct image_params *imgp, vm_ooffset_t offset,
518 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
519 size_t pagesize)
520 {
521 struct sf_buf *sf;
522 size_t map_len;
523 vm_map_t map;
524 vm_object_t object;
525 vm_offset_t off, map_addr;
526 int error, rv, cow;
527 size_t copy_len;
528 vm_ooffset_t file_addr;
529
530 /*
531 * It's necessary to fail if the filsz + offset taken from the
532 * header is greater than the actual file pager object's size.
533 * If we were to allow this, then the vm_map_find() below would
534 * walk right off the end of the file object and into the ether.
535 *
536 * While I'm here, might as well check for something else that
537 * is invalid: filsz cannot be greater than memsz.
538 */
539 if ((filsz != 0 && (off_t)filsz + offset > imgp->attr->va_size) ||
540 filsz > memsz) {
541 uprintf("elf_load_section: truncated ELF file\n");
542 return (ENOEXEC);
543 }
544
545 object = imgp->object;
546 map = &imgp->proc->p_vmspace->vm_map;
547 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
548 file_addr = trunc_page_ps(offset, pagesize);
549
550 /*
551 * We have two choices. We can either clear the data in the last page
552 * of an oversized mapping, or we can start the anon mapping a page
553 * early and copy the initialized data into that first page. We
554 * choose the second.
555 */
556 if (filsz == 0)
557 map_len = 0;
558 else if (memsz > filsz)
559 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
560 else
561 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
562
563 if (map_len != 0) {
564 /* cow flags: don't dump readonly sections in core */
565 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
566 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
567
568 rv = __elfN(map_insert)(imgp, map,
569 object,
570 file_addr, /* file offset */
571 map_addr, /* virtual start */
572 map_addr + map_len,/* virtual end */
573 prot,
574 cow);
575 if (rv != KERN_SUCCESS)
576 return (EINVAL);
577
578 /* we can stop now if we've covered it all */
579 if (memsz == filsz)
580 return (0);
581 }
582
583
584 /*
585 * We have to get the remaining bit of the file into the first part
586 * of the oversized map segment. This is normally because the .data
587 * segment in the file is extended to provide bss. It's a neat idea
588 * to try and save a page, but it's a pain in the behind to implement.
589 */
590 copy_len = filsz == 0 ? 0 : (offset + filsz) - trunc_page_ps(offset +
591 filsz, pagesize);
592 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
593 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
594 map_addr;
595
596 /* This had damn well better be true! */
597 if (map_len != 0) {
598 rv = __elfN(map_insert)(imgp, map, NULL, 0, map_addr,
599 map_addr + map_len, prot, 0);
600 if (rv != KERN_SUCCESS)
601 return (EINVAL);
602 }
603
604 if (copy_len != 0) {
605 sf = vm_imgact_map_page(object, offset + filsz);
606 if (sf == NULL)
607 return (EIO);
608
609 /* send the page fragment to user space */
610 off = trunc_page_ps(offset + filsz, pagesize) -
611 trunc_page(offset + filsz);
612 error = copyout((caddr_t)sf_buf_kva(sf) + off,
613 (caddr_t)map_addr, copy_len);
614 vm_imgact_unmap_page(sf);
615 if (error != 0)
616 return (error);
617 }
618
619 /*
620 * Remove write access to the page if it was only granted by map_insert
621 * to allow copyout.
622 */
623 if ((prot & VM_PROT_WRITE) == 0)
624 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
625 map_len), prot, FALSE);
626
627 return (0);
628 }
629
630 /*
631 * Load the file "file" into memory. It may be either a shared object
632 * or an executable.
633 *
634 * The "addr" reference parameter is in/out. On entry, it specifies
635 * the address where a shared object should be loaded. If the file is
636 * an executable, this value is ignored. On exit, "addr" specifies
637 * where the file was actually loaded.
638 *
639 * The "entry" reference parameter is out only. On exit, it specifies
640 * the entry point for the loaded file.
641 */
642 static int
643 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
644 u_long *entry, size_t pagesize)
645 {
646 struct {
647 struct nameidata nd;
648 struct vattr attr;
649 struct image_params image_params;
650 } *tempdata;
651 const Elf_Ehdr *hdr = NULL;
652 const Elf_Phdr *phdr = NULL;
653 struct nameidata *nd;
654 struct vattr *attr;
655 struct image_params *imgp;
656 vm_prot_t prot;
657 u_long rbase;
658 u_long base_addr = 0;
659 int error, i, numsegs;
660
661 #ifdef CAPABILITY_MODE
662 /*
663 * XXXJA: This check can go away once we are sufficiently confident
664 * that the checks in namei() are correct.
665 */
666 if (IN_CAPABILITY_MODE(curthread))
667 return (ECAPMODE);
668 #endif
669
670 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
671 nd = &tempdata->nd;
672 attr = &tempdata->attr;
673 imgp = &tempdata->image_params;
674
675 /*
676 * Initialize part of the common data
677 */
678 imgp->proc = p;
679 imgp->attr = attr;
680 imgp->firstpage = NULL;
681 imgp->image_header = NULL;
682 imgp->object = NULL;
683 imgp->execlabel = NULL;
684
685 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
686 if ((error = namei(nd)) != 0) {
687 nd->ni_vp = NULL;
688 goto fail;
689 }
690 NDFREE(nd, NDF_ONLY_PNBUF);
691 imgp->vp = nd->ni_vp;
692
693 /*
694 * Check permissions, modes, uid, etc on the file, and "open" it.
695 */
696 error = exec_check_permissions(imgp);
697 if (error)
698 goto fail;
699
700 error = exec_map_first_page(imgp);
701 if (error)
702 goto fail;
703
704 /*
705 * Also make certain that the interpreter stays the same, so set
706 * its VV_TEXT flag, too.
707 */
708 VOP_SET_TEXT(nd->ni_vp);
709
710 imgp->object = nd->ni_vp->v_object;
711
712 hdr = (const Elf_Ehdr *)imgp->image_header;
713 if ((error = __elfN(check_header)(hdr)) != 0)
714 goto fail;
715 if (hdr->e_type == ET_DYN)
716 rbase = *addr;
717 else if (hdr->e_type == ET_EXEC)
718 rbase = 0;
719 else {
720 error = ENOEXEC;
721 goto fail;
722 }
723
724 /* Only support headers that fit within first page for now */
725 if ((hdr->e_phoff > PAGE_SIZE) ||
726 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
727 error = ENOEXEC;
728 goto fail;
729 }
730
731 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
732 if (!aligned(phdr, Elf_Addr)) {
733 error = ENOEXEC;
734 goto fail;
735 }
736
737 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
738 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
739 /* Loadable segment */
740 prot = __elfN(trans_prot)(phdr[i].p_flags);
741 error = __elfN(load_section)(imgp, phdr[i].p_offset,
742 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
743 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
744 if (error != 0)
745 goto fail;
746 /*
747 * Establish the base address if this is the
748 * first segment.
749 */
750 if (numsegs == 0)
751 base_addr = trunc_page(phdr[i].p_vaddr +
752 rbase);
753 numsegs++;
754 }
755 }
756 *addr = base_addr;
757 *entry = (unsigned long)hdr->e_entry + rbase;
758
759 fail:
760 if (imgp->firstpage)
761 exec_unmap_first_page(imgp);
762
763 if (nd->ni_vp)
764 vput(nd->ni_vp);
765
766 free(tempdata, M_TEMP);
767
768 return (error);
769 }
770
771 static int
772 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
773 {
774 struct thread *td;
775 const Elf_Ehdr *hdr;
776 const Elf_Phdr *phdr;
777 Elf_Auxargs *elf_auxargs;
778 struct vmspace *vmspace;
779 const char *err_str, *newinterp;
780 char *interp, *interp_buf, *path;
781 Elf_Brandinfo *brand_info;
782 struct sysentvec *sv;
783 vm_prot_t prot;
784 u_long text_size, data_size, total_size, text_addr, data_addr;
785 u_long seg_size, seg_addr, addr, baddr, et_dyn_addr, entry, proghdr;
786 int32_t osrel;
787 int error, i, n, interp_name_len, have_interp;
788
789 hdr = (const Elf_Ehdr *)imgp->image_header;
790
791 /*
792 * Do we have a valid ELF header ?
793 *
794 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
795 * if particular brand doesn't support it.
796 */
797 if (__elfN(check_header)(hdr) != 0 ||
798 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
799 return (-1);
800
801 /*
802 * From here on down, we return an errno, not -1, as we've
803 * detected an ELF file.
804 */
805
806 if ((hdr->e_phoff > PAGE_SIZE) ||
807 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
808 /* Only support headers in first page for now */
809 uprintf("Program headers not in the first page\n");
810 return (ENOEXEC);
811 }
812 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
813 if (!aligned(phdr, Elf_Addr)) {
814 uprintf("Unaligned program headers\n");
815 return (ENOEXEC);
816 }
817
818 n = error = 0;
819 baddr = 0;
820 osrel = 0;
821 text_size = data_size = total_size = text_addr = data_addr = 0;
822 entry = proghdr = 0;
823 interp_name_len = 0;
824 err_str = newinterp = NULL;
825 interp = interp_buf = NULL;
826 td = curthread;
827
828 for (i = 0; i < hdr->e_phnum; i++) {
829 switch (phdr[i].p_type) {
830 case PT_LOAD:
831 if (n == 0)
832 baddr = phdr[i].p_vaddr;
833 n++;
834 break;
835 case PT_INTERP:
836 /* Path to interpreter */
837 if (phdr[i].p_filesz < 2 ||
838 phdr[i].p_filesz > MAXPATHLEN) {
839 uprintf("Invalid PT_INTERP\n");
840 error = ENOEXEC;
841 goto ret;
842 }
843 if (interp != NULL) {
844 uprintf("Multiple PT_INTERP headers\n");
845 error = ENOEXEC;
846 goto ret;
847 }
848 interp_name_len = phdr[i].p_filesz;
849 if (phdr[i].p_offset > PAGE_SIZE ||
850 interp_name_len > PAGE_SIZE - phdr[i].p_offset) {
851 VOP_UNLOCK(imgp->vp, 0);
852 interp_buf = malloc(interp_name_len + 1, M_TEMP,
853 M_WAITOK);
854 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
855 error = vn_rdwr(UIO_READ, imgp->vp, interp_buf,
856 interp_name_len, phdr[i].p_offset,
857 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred,
858 NOCRED, NULL, td);
859 if (error != 0) {
860 uprintf("i/o error PT_INTERP\n");
861 goto ret;
862 }
863 interp_buf[interp_name_len] = '\0';
864 interp = interp_buf;
865 } else {
866 interp = __DECONST(char *, imgp->image_header) +
867 phdr[i].p_offset;
868 if (interp[interp_name_len - 1] != '\0') {
869 uprintf("Invalid PT_INTERP\n");
870 error = ENOEXEC;
871 goto ret;
872 }
873 }
874 break;
875 case PT_GNU_STACK:
876 if (__elfN(nxstack))
877 imgp->stack_prot =
878 __elfN(trans_prot)(phdr[i].p_flags);
879 imgp->stack_sz = phdr[i].p_memsz;
880 break;
881 }
882 }
883
884 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
885 &osrel);
886 if (brand_info == NULL) {
887 uprintf("ELF binary type \"%u\" not known.\n",
888 hdr->e_ident[EI_OSABI]);
889 error = ENOEXEC;
890 goto ret;
891 }
892 if (hdr->e_type == ET_DYN) {
893 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0) {
894 uprintf("Cannot execute shared object\n");
895 error = ENOEXEC;
896 goto ret;
897 }
898 /*
899 * Honour the base load address from the dso if it is
900 * non-zero for some reason.
901 */
902 if (baddr == 0)
903 et_dyn_addr = ET_DYN_LOAD_ADDR;
904 else
905 et_dyn_addr = 0;
906 } else
907 et_dyn_addr = 0;
908 sv = brand_info->sysvec;
909 if (interp != NULL && brand_info->interp_newpath != NULL)
910 newinterp = brand_info->interp_newpath;
911
912 /*
913 * Avoid a possible deadlock if the current address space is destroyed
914 * and that address space maps the locked vnode. In the common case,
915 * the locked vnode's v_usecount is decremented but remains greater
916 * than zero. Consequently, the vnode lock is not needed by vrele().
917 * However, in cases where the vnode lock is external, such as nullfs,
918 * v_usecount may become zero.
919 *
920 * The VV_TEXT flag prevents modifications to the executable while
921 * the vnode is unlocked.
922 */
923 VOP_UNLOCK(imgp->vp, 0);
924
925 error = exec_new_vmspace(imgp, sv);
926 imgp->proc->p_sysent = sv;
927
928 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
929 if (error != 0)
930 goto ret;
931
932 for (i = 0; i < hdr->e_phnum; i++) {
933 switch (phdr[i].p_type) {
934 case PT_LOAD: /* Loadable segment */
935 if (phdr[i].p_memsz == 0)
936 break;
937 prot = __elfN(trans_prot)(phdr[i].p_flags);
938 error = __elfN(load_section)(imgp, phdr[i].p_offset,
939 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
940 phdr[i].p_memsz, phdr[i].p_filesz, prot,
941 sv->sv_pagesize);
942 if (error != 0)
943 goto ret;
944
945 /*
946 * If this segment contains the program headers,
947 * remember their virtual address for the AT_PHDR
948 * aux entry. Static binaries don't usually include
949 * a PT_PHDR entry.
950 */
951 if (phdr[i].p_offset == 0 &&
952 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
953 <= phdr[i].p_filesz)
954 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
955 et_dyn_addr;
956
957 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
958 seg_size = round_page(phdr[i].p_memsz +
959 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
960
961 /*
962 * Make the largest executable segment the official
963 * text segment and all others data.
964 *
965 * Note that obreak() assumes that data_addr +
966 * data_size == end of data load area, and the ELF
967 * file format expects segments to be sorted by
968 * address. If multiple data segments exist, the
969 * last one will be used.
970 */
971
972 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
973 text_size = seg_size;
974 text_addr = seg_addr;
975 } else {
976 data_size = seg_size;
977 data_addr = seg_addr;
978 }
979 total_size += seg_size;
980 break;
981 case PT_PHDR: /* Program header table info */
982 proghdr = phdr[i].p_vaddr + et_dyn_addr;
983 break;
984 default:
985 break;
986 }
987 }
988
989 if (data_addr == 0 && data_size == 0) {
990 data_addr = text_addr;
991 data_size = text_size;
992 }
993
994 entry = (u_long)hdr->e_entry + et_dyn_addr;
995
996 /*
997 * Check limits. It should be safe to check the
998 * limits after loading the segments since we do
999 * not actually fault in all the segments pages.
1000 */
1001 PROC_LOCK(imgp->proc);
1002 if (data_size > lim_cur_proc(imgp->proc, RLIMIT_DATA))
1003 err_str = "Data segment size exceeds process limit";
1004 else if (text_size > maxtsiz)
1005 err_str = "Text segment size exceeds system limit";
1006 else if (total_size > lim_cur_proc(imgp->proc, RLIMIT_VMEM))
1007 err_str = "Total segment size exceeds process limit";
1008 else if (racct_set(imgp->proc, RACCT_DATA, data_size) != 0)
1009 err_str = "Data segment size exceeds resource limit";
1010 else if (racct_set(imgp->proc, RACCT_VMEM, total_size) != 0)
1011 err_str = "Total segment size exceeds resource limit";
1012 if (err_str != NULL) {
1013 PROC_UNLOCK(imgp->proc);
1014 uprintf("%s\n", err_str);
1015 error = ENOMEM;
1016 goto ret;
1017 }
1018
1019 vmspace = imgp->proc->p_vmspace;
1020 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
1021 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
1022 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
1023 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
1024
1025 /*
1026 * We load the dynamic linker where a userland call
1027 * to mmap(0, ...) would put it. The rationale behind this
1028 * calculation is that it leaves room for the heap to grow to
1029 * its maximum allowed size.
1030 */
1031 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(td,
1032 RLIMIT_DATA));
1033 PROC_UNLOCK(imgp->proc);
1034
1035 imgp->entry_addr = entry;
1036
1037 if (interp != NULL) {
1038 have_interp = FALSE;
1039 VOP_UNLOCK(imgp->vp, 0);
1040 if (brand_info->emul_path != NULL &&
1041 brand_info->emul_path[0] != '\0') {
1042 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1043 snprintf(path, MAXPATHLEN, "%s%s",
1044 brand_info->emul_path, interp);
1045 error = __elfN(load_file)(imgp->proc, path, &addr,
1046 &imgp->entry_addr, sv->sv_pagesize);
1047 free(path, M_TEMP);
1048 if (error == 0)
1049 have_interp = TRUE;
1050 }
1051 if (!have_interp && newinterp != NULL &&
1052 (brand_info->interp_path == NULL ||
1053 strcmp(interp, brand_info->interp_path) == 0)) {
1054 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
1055 &imgp->entry_addr, sv->sv_pagesize);
1056 if (error == 0)
1057 have_interp = TRUE;
1058 }
1059 if (!have_interp) {
1060 error = __elfN(load_file)(imgp->proc, interp, &addr,
1061 &imgp->entry_addr, sv->sv_pagesize);
1062 }
1063 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
1064 if (error != 0) {
1065 uprintf("ELF interpreter %s not found, error %d\n",
1066 interp, error);
1067 goto ret;
1068 }
1069 } else
1070 addr = et_dyn_addr;
1071
1072 /*
1073 * Construct auxargs table (used by the fixup routine)
1074 */
1075 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
1076 elf_auxargs->execfd = -1;
1077 elf_auxargs->phdr = proghdr;
1078 elf_auxargs->phent = hdr->e_phentsize;
1079 elf_auxargs->phnum = hdr->e_phnum;
1080 elf_auxargs->pagesz = PAGE_SIZE;
1081 elf_auxargs->base = addr;
1082 elf_auxargs->flags = 0;
1083 elf_auxargs->entry = entry;
1084 elf_auxargs->hdr_eflags = hdr->e_flags;
1085
1086 imgp->auxargs = elf_auxargs;
1087 imgp->interpreted = 0;
1088 imgp->reloc_base = addr;
1089 imgp->proc->p_osrel = osrel;
1090 imgp->proc->p_elf_machine = hdr->e_machine;
1091 imgp->proc->p_elf_flags = hdr->e_flags;
1092
1093 ret:
1094 free(interp_buf, M_TEMP);
1095 return (error);
1096 }
1097
1098 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
1099
1100 int
1101 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
1102 {
1103 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
1104 Elf_Addr *base;
1105 Elf_Addr *pos;
1106
1107 base = (Elf_Addr *)*stack_base;
1108 pos = base + (imgp->args->argc + imgp->args->envc + 2);
1109
1110 if (args->execfd != -1)
1111 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
1112 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
1113 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
1114 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
1115 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
1116 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
1117 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
1118 AUXARGS_ENTRY(pos, AT_BASE, args->base);
1119 #ifdef AT_EHDRFLAGS
1120 AUXARGS_ENTRY(pos, AT_EHDRFLAGS, args->hdr_eflags);
1121 #endif
1122 if (imgp->execpathp != 0)
1123 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1124 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1125 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1126 if (imgp->canary != 0) {
1127 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1128 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1129 }
1130 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1131 if (imgp->pagesizes != 0) {
1132 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1133 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1134 }
1135 if (imgp->sysent->sv_timekeep_base != 0) {
1136 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1137 imgp->sysent->sv_timekeep_base);
1138 }
1139 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1140 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1141 imgp->sysent->sv_stackprot);
1142 AUXARGS_ENTRY(pos, AT_NULL, 0);
1143
1144 free(imgp->auxargs, M_TEMP);
1145 imgp->auxargs = NULL;
1146
1147 base--;
1148 suword(base, (long)imgp->args->argc);
1149 *stack_base = (register_t *)base;
1150 return (0);
1151 }
1152
1153 /*
1154 * Code for generating ELF core dumps.
1155 */
1156
1157 typedef void (*segment_callback)(vm_map_entry_t, void *);
1158
1159 /* Closure for cb_put_phdr(). */
1160 struct phdr_closure {
1161 Elf_Phdr *phdr; /* Program header to fill in */
1162 Elf_Off offset; /* Offset of segment in core file */
1163 };
1164
1165 /* Closure for cb_size_segment(). */
1166 struct sseg_closure {
1167 int count; /* Count of writable segments. */
1168 size_t size; /* Total size of all writable segments. */
1169 };
1170
1171 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1172
1173 struct note_info {
1174 int type; /* Note type. */
1175 outfunc_t outfunc; /* Output function. */
1176 void *outarg; /* Argument for the output function. */
1177 size_t outsize; /* Output size. */
1178 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1179 };
1180
1181 TAILQ_HEAD(note_info_list, note_info);
1182
1183 /* Coredump output parameters. */
1184 struct coredump_params {
1185 off_t offset;
1186 struct ucred *active_cred;
1187 struct ucred *file_cred;
1188 struct thread *td;
1189 struct vnode *vp;
1190 struct gzio_stream *gzs;
1191 };
1192
1193 static void cb_put_phdr(vm_map_entry_t, void *);
1194 static void cb_size_segment(vm_map_entry_t, void *);
1195 static int core_write(struct coredump_params *, const void *, size_t, off_t,
1196 enum uio_seg);
1197 static void each_writable_segment(struct thread *, segment_callback, void *);
1198 static int __elfN(corehdr)(struct coredump_params *, int, void *, size_t,
1199 struct note_info_list *, size_t);
1200 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1201 size_t *);
1202 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1203 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1204 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1205 static int sbuf_drain_core_output(void *, const char *, int);
1206 static int sbuf_drain_count(void *arg, const char *data, int len);
1207
1208 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1209 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1210 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1211 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1212 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1213 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1214 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1215 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1216 static void note_procstat_files(void *, struct sbuf *, size_t *);
1217 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1218 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1219 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1220 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1221 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1222
1223 #ifdef GZIO
1224 extern int compress_user_cores_gzlevel;
1225
1226 /*
1227 * Write out a core segment to the compression stream.
1228 */
1229 static int
1230 compress_chunk(struct coredump_params *p, char *base, char *buf, u_int len)
1231 {
1232 u_int chunk_len;
1233 int error;
1234
1235 while (len > 0) {
1236 chunk_len = MIN(len, CORE_BUF_SIZE);
1237
1238 /*
1239 * We can get EFAULT error here.
1240 * In that case zero out the current chunk of the segment.
1241 */
1242 error = copyin(base, buf, chunk_len);
1243 if (error != 0)
1244 bzero(buf, chunk_len);
1245 error = gzio_write(p->gzs, buf, chunk_len);
1246 if (error != 0)
1247 break;
1248 base += chunk_len;
1249 len -= chunk_len;
1250 }
1251 return (error);
1252 }
1253
1254 static int
1255 core_gz_write(void *base, size_t len, off_t offset, void *arg)
1256 {
1257
1258 return (core_write((struct coredump_params *)arg, base, len, offset,
1259 UIO_SYSSPACE));
1260 }
1261 #endif /* GZIO */
1262
1263 static int
1264 core_write(struct coredump_params *p, const void *base, size_t len,
1265 off_t offset, enum uio_seg seg)
1266 {
1267
1268 return (vn_rdwr_inchunks(UIO_WRITE, p->vp, __DECONST(void *, base),
1269 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1270 p->active_cred, p->file_cred, NULL, p->td));
1271 }
1272
1273 static int
1274 core_output(void *base, size_t len, off_t offset, struct coredump_params *p,
1275 void *tmpbuf)
1276 {
1277 int error;
1278
1279 #ifdef GZIO
1280 if (p->gzs != NULL)
1281 return (compress_chunk(p, base, tmpbuf, len));
1282 #endif
1283 /*
1284 * EFAULT is a non-fatal error that we can get, for example,
1285 * if the segment is backed by a file but extends beyond its
1286 * end.
1287 */
1288 error = core_write(p, base, len, offset, UIO_USERSPACE);
1289 if (error == EFAULT) {
1290 log(LOG_WARNING, "Failed to fully fault in a core file segment "
1291 "at VA %p with size 0x%zx to be written at offset 0x%jx "
1292 "for process %s\n", base, len, offset, curproc->p_comm);
1293
1294 /*
1295 * Write a "real" zero byte at the end of the target region
1296 * in the case this is the last segment.
1297 * The intermediate space will be implicitly zero-filled.
1298 */
1299 error = core_write(p, zero_region, 1, offset + len - 1,
1300 UIO_SYSSPACE);
1301 }
1302 return (error);
1303 }
1304
1305 /*
1306 * Drain into a core file.
1307 */
1308 static int
1309 sbuf_drain_core_output(void *arg, const char *data, int len)
1310 {
1311 struct coredump_params *p;
1312 int error, locked;
1313
1314 p = (struct coredump_params *)arg;
1315
1316 /*
1317 * Some kern_proc out routines that print to this sbuf may
1318 * call us with the process lock held. Draining with the
1319 * non-sleepable lock held is unsafe. The lock is needed for
1320 * those routines when dumping a live process. In our case we
1321 * can safely release the lock before draining and acquire
1322 * again after.
1323 */
1324 locked = PROC_LOCKED(p->td->td_proc);
1325 if (locked)
1326 PROC_UNLOCK(p->td->td_proc);
1327 #ifdef GZIO
1328 if (p->gzs != NULL)
1329 error = gzio_write(p->gzs, __DECONST(char *, data), len);
1330 else
1331 #endif
1332 error = core_write(p, __DECONST(void *, data), len, p->offset,
1333 UIO_SYSSPACE);
1334 if (locked)
1335 PROC_LOCK(p->td->td_proc);
1336 if (error != 0)
1337 return (-error);
1338 p->offset += len;
1339 return (len);
1340 }
1341
1342 /*
1343 * Drain into a counter.
1344 */
1345 static int
1346 sbuf_drain_count(void *arg, const char *data __unused, int len)
1347 {
1348 size_t *sizep;
1349
1350 sizep = (size_t *)arg;
1351 *sizep += len;
1352 return (len);
1353 }
1354
1355 int
1356 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1357 {
1358 struct ucred *cred = td->td_ucred;
1359 int error = 0;
1360 struct sseg_closure seginfo;
1361 struct note_info_list notelst;
1362 struct coredump_params params;
1363 struct note_info *ninfo;
1364 void *hdr, *tmpbuf;
1365 size_t hdrsize, notesz, coresize;
1366 #ifdef GZIO
1367 boolean_t compress;
1368
1369 compress = (flags & IMGACT_CORE_COMPRESS) != 0;
1370 #endif
1371 hdr = NULL;
1372 tmpbuf = NULL;
1373 TAILQ_INIT(¬elst);
1374
1375 /* Size the program segments. */
1376 seginfo.count = 0;
1377 seginfo.size = 0;
1378 each_writable_segment(td, cb_size_segment, &seginfo);
1379
1380 /*
1381 * Collect info about the core file header area.
1382 */
1383 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1384 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1385 coresize = round_page(hdrsize + notesz) + seginfo.size;
1386
1387 /* Set up core dump parameters. */
1388 params.offset = 0;
1389 params.active_cred = cred;
1390 params.file_cred = NOCRED;
1391 params.td = td;
1392 params.vp = vp;
1393 params.gzs = NULL;
1394
1395 #ifdef RACCT
1396 if (racct_enable) {
1397 PROC_LOCK(td->td_proc);
1398 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1399 PROC_UNLOCK(td->td_proc);
1400 if (error != 0) {
1401 error = EFAULT;
1402 goto done;
1403 }
1404 }
1405 #endif
1406 if (coresize >= limit) {
1407 error = EFAULT;
1408 goto done;
1409 }
1410
1411 #ifdef GZIO
1412 /* Create a compression stream if necessary. */
1413 if (compress) {
1414 params.gzs = gzio_init(core_gz_write, GZIO_DEFLATE,
1415 CORE_BUF_SIZE, compress_user_cores_gzlevel, ¶ms);
1416 if (params.gzs == NULL) {
1417 error = EFAULT;
1418 goto done;
1419 }
1420 tmpbuf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1421 }
1422 #endif
1423
1424 /*
1425 * Allocate memory for building the header, fill it up,
1426 * and write it out following the notes.
1427 */
1428 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1429 error = __elfN(corehdr)(¶ms, seginfo.count, hdr, hdrsize, ¬elst,
1430 notesz);
1431
1432 /* Write the contents of all of the writable segments. */
1433 if (error == 0) {
1434 Elf_Phdr *php;
1435 off_t offset;
1436 int i;
1437
1438 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1439 offset = round_page(hdrsize + notesz);
1440 for (i = 0; i < seginfo.count; i++) {
1441 error = core_output((caddr_t)(uintptr_t)php->p_vaddr,
1442 php->p_filesz, offset, ¶ms, tmpbuf);
1443 if (error != 0)
1444 break;
1445 offset += php->p_filesz;
1446 php++;
1447 }
1448 #ifdef GZIO
1449 if (error == 0 && compress)
1450 error = gzio_flush(params.gzs);
1451 #endif
1452 }
1453 if (error) {
1454 log(LOG_WARNING,
1455 "Failed to write core file for process %s (error %d)\n",
1456 curproc->p_comm, error);
1457 }
1458
1459 done:
1460 #ifdef GZIO
1461 if (compress) {
1462 free(tmpbuf, M_TEMP);
1463 if (params.gzs != NULL)
1464 gzio_fini(params.gzs);
1465 }
1466 #endif
1467 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1468 TAILQ_REMOVE(¬elst, ninfo, link);
1469 free(ninfo, M_TEMP);
1470 }
1471 if (hdr != NULL)
1472 free(hdr, M_TEMP);
1473
1474 return (error);
1475 }
1476
1477 /*
1478 * A callback for each_writable_segment() to write out the segment's
1479 * program header entry.
1480 */
1481 static void
1482 cb_put_phdr(entry, closure)
1483 vm_map_entry_t entry;
1484 void *closure;
1485 {
1486 struct phdr_closure *phc = (struct phdr_closure *)closure;
1487 Elf_Phdr *phdr = phc->phdr;
1488
1489 phc->offset = round_page(phc->offset);
1490
1491 phdr->p_type = PT_LOAD;
1492 phdr->p_offset = phc->offset;
1493 phdr->p_vaddr = entry->start;
1494 phdr->p_paddr = 0;
1495 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1496 phdr->p_align = PAGE_SIZE;
1497 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1498
1499 phc->offset += phdr->p_filesz;
1500 phc->phdr++;
1501 }
1502
1503 /*
1504 * A callback for each_writable_segment() to gather information about
1505 * the number of segments and their total size.
1506 */
1507 static void
1508 cb_size_segment(entry, closure)
1509 vm_map_entry_t entry;
1510 void *closure;
1511 {
1512 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1513
1514 ssc->count++;
1515 ssc->size += entry->end - entry->start;
1516 }
1517
1518 /*
1519 * For each writable segment in the process's memory map, call the given
1520 * function with a pointer to the map entry and some arbitrary
1521 * caller-supplied data.
1522 */
1523 static void
1524 each_writable_segment(td, func, closure)
1525 struct thread *td;
1526 segment_callback func;
1527 void *closure;
1528 {
1529 struct proc *p = td->td_proc;
1530 vm_map_t map = &p->p_vmspace->vm_map;
1531 vm_map_entry_t entry;
1532 vm_object_t backing_object, object;
1533 boolean_t ignore_entry;
1534
1535 vm_map_lock_read(map);
1536 for (entry = map->header.next; entry != &map->header;
1537 entry = entry->next) {
1538 /*
1539 * Don't dump inaccessible mappings, deal with legacy
1540 * coredump mode.
1541 *
1542 * Note that read-only segments related to the elf binary
1543 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1544 * need to arbitrarily ignore such segments.
1545 */
1546 if (elf_legacy_coredump) {
1547 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1548 continue;
1549 } else {
1550 if ((entry->protection & VM_PROT_ALL) == 0)
1551 continue;
1552 }
1553
1554 /*
1555 * Dont include memory segment in the coredump if
1556 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1557 * madvise(2). Do not dump submaps (i.e. parts of the
1558 * kernel map).
1559 */
1560 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1561 continue;
1562
1563 if ((object = entry->object.vm_object) == NULL)
1564 continue;
1565
1566 /* Ignore memory-mapped devices and such things. */
1567 VM_OBJECT_RLOCK(object);
1568 while ((backing_object = object->backing_object) != NULL) {
1569 VM_OBJECT_RLOCK(backing_object);
1570 VM_OBJECT_RUNLOCK(object);
1571 object = backing_object;
1572 }
1573 ignore_entry = object->type != OBJT_DEFAULT &&
1574 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1575 object->type != OBJT_PHYS;
1576 VM_OBJECT_RUNLOCK(object);
1577 if (ignore_entry)
1578 continue;
1579
1580 (*func)(entry, closure);
1581 }
1582 vm_map_unlock_read(map);
1583 }
1584
1585 /*
1586 * Write the core file header to the file, including padding up to
1587 * the page boundary.
1588 */
1589 static int
1590 __elfN(corehdr)(struct coredump_params *p, int numsegs, void *hdr,
1591 size_t hdrsize, struct note_info_list *notelst, size_t notesz)
1592 {
1593 struct note_info *ninfo;
1594 struct sbuf *sb;
1595 int error;
1596
1597 /* Fill in the header. */
1598 bzero(hdr, hdrsize);
1599 __elfN(puthdr)(p->td, hdr, hdrsize, numsegs, notesz);
1600
1601 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1602 sbuf_set_drain(sb, sbuf_drain_core_output, p);
1603 sbuf_start_section(sb, NULL);
1604 sbuf_bcat(sb, hdr, hdrsize);
1605 TAILQ_FOREACH(ninfo, notelst, link)
1606 __elfN(putnote)(ninfo, sb);
1607 /* Align up to a page boundary for the program segments. */
1608 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1609 error = sbuf_finish(sb);
1610 sbuf_delete(sb);
1611
1612 return (error);
1613 }
1614
1615 static void
1616 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1617 size_t *sizep)
1618 {
1619 struct proc *p;
1620 struct thread *thr;
1621 size_t size;
1622
1623 p = td->td_proc;
1624 size = 0;
1625
1626 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1627
1628 /*
1629 * To have the debugger select the right thread (LWP) as the initial
1630 * thread, we dump the state of the thread passed to us in td first.
1631 * This is the thread that causes the core dump and thus likely to
1632 * be the right thread one wants to have selected in the debugger.
1633 */
1634 thr = td;
1635 while (thr != NULL) {
1636 size += register_note(list, NT_PRSTATUS,
1637 __elfN(note_prstatus), thr);
1638 size += register_note(list, NT_FPREGSET,
1639 __elfN(note_fpregset), thr);
1640 size += register_note(list, NT_THRMISC,
1641 __elfN(note_thrmisc), thr);
1642 size += register_note(list, -1,
1643 __elfN(note_threadmd), thr);
1644
1645 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1646 TAILQ_NEXT(thr, td_plist);
1647 if (thr == td)
1648 thr = TAILQ_NEXT(thr, td_plist);
1649 }
1650
1651 size += register_note(list, NT_PROCSTAT_PROC,
1652 __elfN(note_procstat_proc), p);
1653 size += register_note(list, NT_PROCSTAT_FILES,
1654 note_procstat_files, p);
1655 size += register_note(list, NT_PROCSTAT_VMMAP,
1656 note_procstat_vmmap, p);
1657 size += register_note(list, NT_PROCSTAT_GROUPS,
1658 note_procstat_groups, p);
1659 size += register_note(list, NT_PROCSTAT_UMASK,
1660 note_procstat_umask, p);
1661 size += register_note(list, NT_PROCSTAT_RLIMIT,
1662 note_procstat_rlimit, p);
1663 size += register_note(list, NT_PROCSTAT_OSREL,
1664 note_procstat_osrel, p);
1665 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1666 __elfN(note_procstat_psstrings), p);
1667 size += register_note(list, NT_PROCSTAT_AUXV,
1668 __elfN(note_procstat_auxv), p);
1669
1670 *sizep = size;
1671 }
1672
1673 static void
1674 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1675 size_t notesz)
1676 {
1677 Elf_Ehdr *ehdr;
1678 Elf_Phdr *phdr;
1679 struct phdr_closure phc;
1680
1681 ehdr = (Elf_Ehdr *)hdr;
1682 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1683
1684 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1685 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1686 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1687 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1688 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1689 ehdr->e_ident[EI_DATA] = ELF_DATA;
1690 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1691 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1692 ehdr->e_ident[EI_ABIVERSION] = 0;
1693 ehdr->e_ident[EI_PAD] = 0;
1694 ehdr->e_type = ET_CORE;
1695 ehdr->e_machine = td->td_proc->p_elf_machine;
1696 ehdr->e_version = EV_CURRENT;
1697 ehdr->e_entry = 0;
1698 ehdr->e_phoff = sizeof(Elf_Ehdr);
1699 ehdr->e_flags = td->td_proc->p_elf_flags;
1700 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1701 ehdr->e_phentsize = sizeof(Elf_Phdr);
1702 ehdr->e_phnum = numsegs + 1;
1703 ehdr->e_shentsize = sizeof(Elf_Shdr);
1704 ehdr->e_shnum = 0;
1705 ehdr->e_shstrndx = SHN_UNDEF;
1706
1707 /*
1708 * Fill in the program header entries.
1709 */
1710
1711 /* The note segement. */
1712 phdr->p_type = PT_NOTE;
1713 phdr->p_offset = hdrsize;
1714 phdr->p_vaddr = 0;
1715 phdr->p_paddr = 0;
1716 phdr->p_filesz = notesz;
1717 phdr->p_memsz = 0;
1718 phdr->p_flags = PF_R;
1719 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1720 phdr++;
1721
1722 /* All the writable segments from the program. */
1723 phc.phdr = phdr;
1724 phc.offset = round_page(hdrsize + notesz);
1725 each_writable_segment(td, cb_put_phdr, &phc);
1726 }
1727
1728 static size_t
1729 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1730 {
1731 struct note_info *ninfo;
1732 size_t size, notesize;
1733
1734 size = 0;
1735 out(arg, NULL, &size);
1736 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1737 ninfo->type = type;
1738 ninfo->outfunc = out;
1739 ninfo->outarg = arg;
1740 ninfo->outsize = size;
1741 TAILQ_INSERT_TAIL(list, ninfo, link);
1742
1743 if (type == -1)
1744 return (size);
1745
1746 notesize = sizeof(Elf_Note) + /* note header */
1747 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1748 /* note name */
1749 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1750
1751 return (notesize);
1752 }
1753
1754 static size_t
1755 append_note_data(const void *src, void *dst, size_t len)
1756 {
1757 size_t padded_len;
1758
1759 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1760 if (dst != NULL) {
1761 bcopy(src, dst, len);
1762 bzero((char *)dst + len, padded_len - len);
1763 }
1764 return (padded_len);
1765 }
1766
1767 size_t
1768 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1769 {
1770 Elf_Note *note;
1771 char *buf;
1772 size_t notesize;
1773
1774 buf = dst;
1775 if (buf != NULL) {
1776 note = (Elf_Note *)buf;
1777 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1778 note->n_descsz = size;
1779 note->n_type = type;
1780 buf += sizeof(*note);
1781 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1782 sizeof(FREEBSD_ABI_VENDOR));
1783 append_note_data(src, buf, size);
1784 if (descp != NULL)
1785 *descp = buf;
1786 }
1787
1788 notesize = sizeof(Elf_Note) + /* note header */
1789 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1790 /* note name */
1791 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1792
1793 return (notesize);
1794 }
1795
1796 static void
1797 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1798 {
1799 Elf_Note note;
1800 ssize_t old_len, sect_len;
1801 size_t new_len, descsz, i;
1802
1803 if (ninfo->type == -1) {
1804 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1805 return;
1806 }
1807
1808 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1809 note.n_descsz = ninfo->outsize;
1810 note.n_type = ninfo->type;
1811
1812 sbuf_bcat(sb, ¬e, sizeof(note));
1813 sbuf_start_section(sb, &old_len);
1814 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1815 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1816 if (note.n_descsz == 0)
1817 return;
1818 sbuf_start_section(sb, &old_len);
1819 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1820 sect_len = sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1821 if (sect_len < 0)
1822 return;
1823
1824 new_len = (size_t)sect_len;
1825 descsz = roundup(note.n_descsz, ELF_NOTE_ROUNDSIZE);
1826 if (new_len < descsz) {
1827 /*
1828 * It is expected that individual note emitters will correctly
1829 * predict their expected output size and fill up to that size
1830 * themselves, padding in a format-specific way if needed.
1831 * However, in case they don't, just do it here with zeros.
1832 */
1833 for (i = 0; i < descsz - new_len; i++)
1834 sbuf_putc(sb, 0);
1835 } else if (new_len > descsz) {
1836 /*
1837 * We can't always truncate sb -- we may have drained some
1838 * of it already.
1839 */
1840 KASSERT(new_len == descsz, ("%s: Note type %u changed as we "
1841 "read it (%zu > %zu). Since it is longer than "
1842 "expected, this coredump's notes are corrupt. THIS "
1843 "IS A BUG in the note_procstat routine for type %u.\n",
1844 __func__, (unsigned)note.n_type, new_len, descsz,
1845 (unsigned)note.n_type));
1846 }
1847 }
1848
1849 /*
1850 * Miscellaneous note out functions.
1851 */
1852
1853 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1854 #include <compat/freebsd32/freebsd32.h>
1855
1856 typedef struct prstatus32 elf_prstatus_t;
1857 typedef struct prpsinfo32 elf_prpsinfo_t;
1858 typedef struct fpreg32 elf_prfpregset_t;
1859 typedef struct fpreg32 elf_fpregset_t;
1860 typedef struct reg32 elf_gregset_t;
1861 typedef struct thrmisc32 elf_thrmisc_t;
1862 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1863 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1864 typedef uint32_t elf_ps_strings_t;
1865 #else
1866 typedef prstatus_t elf_prstatus_t;
1867 typedef prpsinfo_t elf_prpsinfo_t;
1868 typedef prfpregset_t elf_prfpregset_t;
1869 typedef prfpregset_t elf_fpregset_t;
1870 typedef gregset_t elf_gregset_t;
1871 typedef thrmisc_t elf_thrmisc_t;
1872 #define ELF_KERN_PROC_MASK 0
1873 typedef struct kinfo_proc elf_kinfo_proc_t;
1874 typedef vm_offset_t elf_ps_strings_t;
1875 #endif
1876
1877 static void
1878 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1879 {
1880 struct sbuf sbarg;
1881 size_t len;
1882 char *cp, *end;
1883 struct proc *p;
1884 elf_prpsinfo_t *psinfo;
1885 int error;
1886
1887 p = (struct proc *)arg;
1888 if (sb != NULL) {
1889 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1890 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1891 psinfo->pr_version = PRPSINFO_VERSION;
1892 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1893 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1894 PROC_LOCK(p);
1895 if (p->p_args != NULL) {
1896 len = sizeof(psinfo->pr_psargs) - 1;
1897 if (len > p->p_args->ar_length)
1898 len = p->p_args->ar_length;
1899 memcpy(psinfo->pr_psargs, p->p_args->ar_args, len);
1900 PROC_UNLOCK(p);
1901 error = 0;
1902 } else {
1903 _PHOLD(p);
1904 PROC_UNLOCK(p);
1905 sbuf_new(&sbarg, psinfo->pr_psargs,
1906 sizeof(psinfo->pr_psargs), SBUF_FIXEDLEN);
1907 error = proc_getargv(curthread, p, &sbarg);
1908 PRELE(p);
1909 if (sbuf_finish(&sbarg) == 0)
1910 len = sbuf_len(&sbarg) - 1;
1911 else
1912 len = sizeof(psinfo->pr_psargs) - 1;
1913 sbuf_delete(&sbarg);
1914 }
1915 if (error || len == 0)
1916 strlcpy(psinfo->pr_psargs, p->p_comm,
1917 sizeof(psinfo->pr_psargs));
1918 else {
1919 KASSERT(len < sizeof(psinfo->pr_psargs),
1920 ("len is too long: %zu vs %zu", len,
1921 sizeof(psinfo->pr_psargs)));
1922 cp = psinfo->pr_psargs;
1923 end = cp + len - 1;
1924 for (;;) {
1925 cp = memchr(cp, '\0', end - cp);
1926 if (cp == NULL)
1927 break;
1928 *cp = ' ';
1929 }
1930 }
1931 psinfo->pr_pid = p->p_pid;
1932 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1933 free(psinfo, M_TEMP);
1934 }
1935 *sizep = sizeof(*psinfo);
1936 }
1937
1938 static void
1939 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1940 {
1941 struct thread *td;
1942 elf_prstatus_t *status;
1943
1944 td = (struct thread *)arg;
1945 if (sb != NULL) {
1946 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1947 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1948 status->pr_version = PRSTATUS_VERSION;
1949 status->pr_statussz = sizeof(elf_prstatus_t);
1950 status->pr_gregsetsz = sizeof(elf_gregset_t);
1951 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1952 status->pr_osreldate = osreldate;
1953 status->pr_cursig = td->td_proc->p_sig;
1954 status->pr_pid = td->td_tid;
1955 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1956 fill_regs32(td, &status->pr_reg);
1957 #else
1958 fill_regs(td, &status->pr_reg);
1959 #endif
1960 sbuf_bcat(sb, status, sizeof(*status));
1961 free(status, M_TEMP);
1962 }
1963 *sizep = sizeof(*status);
1964 }
1965
1966 static void
1967 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1968 {
1969 struct thread *td;
1970 elf_prfpregset_t *fpregset;
1971
1972 td = (struct thread *)arg;
1973 if (sb != NULL) {
1974 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1975 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1976 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1977 fill_fpregs32(td, fpregset);
1978 #else
1979 fill_fpregs(td, fpregset);
1980 #endif
1981 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1982 free(fpregset, M_TEMP);
1983 }
1984 *sizep = sizeof(*fpregset);
1985 }
1986
1987 static void
1988 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1989 {
1990 struct thread *td;
1991 elf_thrmisc_t thrmisc;
1992
1993 td = (struct thread *)arg;
1994 if (sb != NULL) {
1995 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1996 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1997 strcpy(thrmisc.pr_tname, td->td_name);
1998 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1999 }
2000 *sizep = sizeof(thrmisc);
2001 }
2002
2003 /*
2004 * Allow for MD specific notes, as well as any MD
2005 * specific preparations for writing MI notes.
2006 */
2007 static void
2008 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
2009 {
2010 struct thread *td;
2011 void *buf;
2012 size_t size;
2013
2014 td = (struct thread *)arg;
2015 size = *sizep;
2016 if (size != 0 && sb != NULL)
2017 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
2018 else
2019 buf = NULL;
2020 size = 0;
2021 __elfN(dump_thread)(td, buf, &size);
2022 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
2023 if (size != 0 && sb != NULL)
2024 sbuf_bcat(sb, buf, size);
2025 free(buf, M_TEMP);
2026 *sizep = size;
2027 }
2028
2029 #ifdef KINFO_PROC_SIZE
2030 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
2031 #endif
2032
2033 static void
2034 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
2035 {
2036 struct proc *p;
2037 size_t size;
2038 int structsize;
2039
2040 p = (struct proc *)arg;
2041 size = sizeof(structsize) + p->p_numthreads *
2042 sizeof(elf_kinfo_proc_t);
2043
2044 if (sb != NULL) {
2045 KASSERT(*sizep == size, ("invalid size"));
2046 structsize = sizeof(elf_kinfo_proc_t);
2047 sbuf_bcat(sb, &structsize, sizeof(structsize));
2048 sx_slock(&proctree_lock);
2049 PROC_LOCK(p);
2050 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
2051 sx_sunlock(&proctree_lock);
2052 }
2053 *sizep = size;
2054 }
2055
2056 #ifdef KINFO_FILE_SIZE
2057 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
2058 #endif
2059
2060 static void
2061 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
2062 {
2063 struct proc *p;
2064 size_t size, sect_sz, i;
2065 ssize_t start_len, sect_len;
2066 int structsize, filedesc_flags;
2067
2068 if (coredump_pack_fileinfo)
2069 filedesc_flags = KERN_FILEDESC_PACK_KINFO;
2070 else
2071 filedesc_flags = 0;
2072
2073 p = (struct proc *)arg;
2074 structsize = sizeof(struct kinfo_file);
2075 if (sb == NULL) {
2076 size = 0;
2077 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2078 sbuf_set_drain(sb, sbuf_drain_count, &size);
2079 sbuf_bcat(sb, &structsize, sizeof(structsize));
2080 PROC_LOCK(p);
2081 kern_proc_filedesc_out(p, sb, -1, filedesc_flags);
2082 sbuf_finish(sb);
2083 sbuf_delete(sb);
2084 *sizep = size;
2085 } else {
2086 sbuf_start_section(sb, &start_len);
2087
2088 sbuf_bcat(sb, &structsize, sizeof(structsize));
2089 PROC_LOCK(p);
2090 kern_proc_filedesc_out(p, sb, *sizep - sizeof(structsize),
2091 filedesc_flags);
2092
2093 sect_len = sbuf_end_section(sb, start_len, 0, 0);
2094 if (sect_len < 0)
2095 return;
2096 sect_sz = sect_len;
2097
2098 KASSERT(sect_sz <= *sizep,
2099 ("kern_proc_filedesc_out did not respect maxlen; "
2100 "requested %zu, got %zu", *sizep - sizeof(structsize),
2101 sect_sz - sizeof(structsize)));
2102
2103 for (i = 0; i < *sizep - sect_sz && sb->s_error == 0; i++)
2104 sbuf_putc(sb, 0);
2105 }
2106 }
2107
2108 #ifdef KINFO_VMENTRY_SIZE
2109 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
2110 #endif
2111
2112 static void
2113 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
2114 {
2115 struct proc *p;
2116 size_t size;
2117 int structsize, vmmap_flags;
2118
2119 if (coredump_pack_vmmapinfo)
2120 vmmap_flags = KERN_VMMAP_PACK_KINFO;
2121 else
2122 vmmap_flags = 0;
2123
2124 p = (struct proc *)arg;
2125 structsize = sizeof(struct kinfo_vmentry);
2126 if (sb == NULL) {
2127 size = 0;
2128 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2129 sbuf_set_drain(sb, sbuf_drain_count, &size);
2130 sbuf_bcat(sb, &structsize, sizeof(structsize));
2131 PROC_LOCK(p);
2132 kern_proc_vmmap_out(p, sb, -1, vmmap_flags);
2133 sbuf_finish(sb);
2134 sbuf_delete(sb);
2135 *sizep = size;
2136 } else {
2137 sbuf_bcat(sb, &structsize, sizeof(structsize));
2138 PROC_LOCK(p);
2139 kern_proc_vmmap_out(p, sb, *sizep - sizeof(structsize),
2140 vmmap_flags);
2141 }
2142 }
2143
2144 static void
2145 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
2146 {
2147 struct proc *p;
2148 size_t size;
2149 int structsize;
2150
2151 p = (struct proc *)arg;
2152 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
2153 if (sb != NULL) {
2154 KASSERT(*sizep == size, ("invalid size"));
2155 structsize = sizeof(gid_t);
2156 sbuf_bcat(sb, &structsize, sizeof(structsize));
2157 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
2158 sizeof(gid_t));
2159 }
2160 *sizep = size;
2161 }
2162
2163 static void
2164 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
2165 {
2166 struct proc *p;
2167 size_t size;
2168 int structsize;
2169
2170 p = (struct proc *)arg;
2171 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
2172 if (sb != NULL) {
2173 KASSERT(*sizep == size, ("invalid size"));
2174 structsize = sizeof(p->p_fd->fd_cmask);
2175 sbuf_bcat(sb, &structsize, sizeof(structsize));
2176 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
2177 }
2178 *sizep = size;
2179 }
2180
2181 static void
2182 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
2183 {
2184 struct proc *p;
2185 struct rlimit rlim[RLIM_NLIMITS];
2186 size_t size;
2187 int structsize, i;
2188
2189 p = (struct proc *)arg;
2190 size = sizeof(structsize) + sizeof(rlim);
2191 if (sb != NULL) {
2192 KASSERT(*sizep == size, ("invalid size"));
2193 structsize = sizeof(rlim);
2194 sbuf_bcat(sb, &structsize, sizeof(structsize));
2195 PROC_LOCK(p);
2196 for (i = 0; i < RLIM_NLIMITS; i++)
2197 lim_rlimit_proc(p, i, &rlim[i]);
2198 PROC_UNLOCK(p);
2199 sbuf_bcat(sb, rlim, sizeof(rlim));
2200 }
2201 *sizep = size;
2202 }
2203
2204 static void
2205 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
2206 {
2207 struct proc *p;
2208 size_t size;
2209 int structsize;
2210
2211 p = (struct proc *)arg;
2212 size = sizeof(structsize) + sizeof(p->p_osrel);
2213 if (sb != NULL) {
2214 KASSERT(*sizep == size, ("invalid size"));
2215 structsize = sizeof(p->p_osrel);
2216 sbuf_bcat(sb, &structsize, sizeof(structsize));
2217 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
2218 }
2219 *sizep = size;
2220 }
2221
2222 static void
2223 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
2224 {
2225 struct proc *p;
2226 elf_ps_strings_t ps_strings;
2227 size_t size;
2228 int structsize;
2229
2230 p = (struct proc *)arg;
2231 size = sizeof(structsize) + sizeof(ps_strings);
2232 if (sb != NULL) {
2233 KASSERT(*sizep == size, ("invalid size"));
2234 structsize = sizeof(ps_strings);
2235 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
2236 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
2237 #else
2238 ps_strings = p->p_sysent->sv_psstrings;
2239 #endif
2240 sbuf_bcat(sb, &structsize, sizeof(structsize));
2241 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
2242 }
2243 *sizep = size;
2244 }
2245
2246 static void
2247 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
2248 {
2249 struct proc *p;
2250 size_t size;
2251 int structsize;
2252
2253 p = (struct proc *)arg;
2254 if (sb == NULL) {
2255 size = 0;
2256 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2257 sbuf_set_drain(sb, sbuf_drain_count, &size);
2258 sbuf_bcat(sb, &structsize, sizeof(structsize));
2259 PHOLD(p);
2260 proc_getauxv(curthread, p, sb);
2261 PRELE(p);
2262 sbuf_finish(sb);
2263 sbuf_delete(sb);
2264 *sizep = size;
2265 } else {
2266 structsize = sizeof(Elf_Auxinfo);
2267 sbuf_bcat(sb, &structsize, sizeof(structsize));
2268 PHOLD(p);
2269 proc_getauxv(curthread, p, sb);
2270 PRELE(p);
2271 }
2272 }
2273
2274 static boolean_t
2275 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2276 int32_t *osrel, const Elf_Phdr *pnote)
2277 {
2278 const Elf_Note *note, *note0, *note_end;
2279 const char *note_name;
2280 char *buf;
2281 int i, error;
2282 boolean_t res;
2283
2284 /* We need some limit, might as well use PAGE_SIZE. */
2285 if (pnote == NULL || pnote->p_filesz > PAGE_SIZE)
2286 return (FALSE);
2287 ASSERT_VOP_LOCKED(imgp->vp, "parse_notes");
2288 if (pnote->p_offset > PAGE_SIZE ||
2289 pnote->p_filesz > PAGE_SIZE - pnote->p_offset) {
2290 VOP_UNLOCK(imgp->vp, 0);
2291 buf = malloc(pnote->p_filesz, M_TEMP, M_WAITOK);
2292 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
2293 error = vn_rdwr(UIO_READ, imgp->vp, buf, pnote->p_filesz,
2294 pnote->p_offset, UIO_SYSSPACE, IO_NODELOCKED,
2295 curthread->td_ucred, NOCRED, NULL, curthread);
2296 if (error != 0) {
2297 uprintf("i/o error PT_NOTE\n");
2298 res = FALSE;
2299 goto ret;
2300 }
2301 note = note0 = (const Elf_Note *)buf;
2302 note_end = (const Elf_Note *)(buf + pnote->p_filesz);
2303 } else {
2304 note = note0 = (const Elf_Note *)(imgp->image_header +
2305 pnote->p_offset);
2306 note_end = (const Elf_Note *)(imgp->image_header +
2307 pnote->p_offset + pnote->p_filesz);
2308 buf = NULL;
2309 }
2310 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2311 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2312 (const char *)note < sizeof(Elf_Note)) {
2313 res = FALSE;
2314 goto ret;
2315 }
2316 if (note->n_namesz != checknote->hdr.n_namesz ||
2317 note->n_descsz != checknote->hdr.n_descsz ||
2318 note->n_type != checknote->hdr.n_type)
2319 goto nextnote;
2320 note_name = (const char *)(note + 1);
2321 if (note_name + checknote->hdr.n_namesz >=
2322 (const char *)note_end || strncmp(checknote->vendor,
2323 note_name, checknote->hdr.n_namesz) != 0)
2324 goto nextnote;
2325
2326 /*
2327 * Fetch the osreldate for binary
2328 * from the ELF OSABI-note if necessary.
2329 */
2330 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2331 checknote->trans_osrel != NULL) {
2332 res = checknote->trans_osrel(note, osrel);
2333 goto ret;
2334 }
2335 res = TRUE;
2336 goto ret;
2337 nextnote:
2338 note = (const Elf_Note *)((const char *)(note + 1) +
2339 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2340 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2341 }
2342 res = FALSE;
2343 ret:
2344 free(buf, M_TEMP);
2345 return (res);
2346 }
2347
2348 /*
2349 * Try to find the appropriate ABI-note section for checknote,
2350 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2351 * first page of the image is searched, the same as for headers.
2352 */
2353 static boolean_t
2354 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2355 int32_t *osrel)
2356 {
2357 const Elf_Phdr *phdr;
2358 const Elf_Ehdr *hdr;
2359 int i;
2360
2361 hdr = (const Elf_Ehdr *)imgp->image_header;
2362 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2363
2364 for (i = 0; i < hdr->e_phnum; i++) {
2365 if (phdr[i].p_type == PT_NOTE &&
2366 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2367 return (TRUE);
2368 }
2369 return (FALSE);
2370
2371 }
2372
2373 /*
2374 * Tell kern_execve.c about it, with a little help from the linker.
2375 */
2376 static struct execsw __elfN(execsw) = {
2377 __CONCAT(exec_, __elfN(imgact)),
2378 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2379 };
2380 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2381
2382 static vm_prot_t
2383 __elfN(trans_prot)(Elf_Word flags)
2384 {
2385 vm_prot_t prot;
2386
2387 prot = 0;
2388 if (flags & PF_X)
2389 prot |= VM_PROT_EXECUTE;
2390 if (flags & PF_W)
2391 prot |= VM_PROT_WRITE;
2392 if (flags & PF_R)
2393 prot |= VM_PROT_READ;
2394 #if __ELF_WORD_SIZE == 32
2395 #if defined(__amd64__)
2396 if (i386_read_exec && (flags & PF_R))
2397 prot |= VM_PROT_EXECUTE;
2398 #endif
2399 #endif
2400 return (prot);
2401 }
2402
2403 static Elf_Word
2404 __elfN(untrans_prot)(vm_prot_t prot)
2405 {
2406 Elf_Word flags;
2407
2408 flags = 0;
2409 if (prot & VM_PROT_EXECUTE)
2410 flags |= PF_X;
2411 if (prot & VM_PROT_READ)
2412 flags |= PF_R;
2413 if (prot & VM_PROT_WRITE)
2414 flags |= PF_W;
2415 return (flags);
2416 }
Cache object: 231b8ed29e0a28cdd1e27d71259a50ea
|