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("$FreeBSD$");
33
34 #include "opt_capsicum.h"
35 #include "opt_compat.h"
36 #include "opt_core.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/imgact.h>
43 #include <sys/imgact_elf.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mman.h>
50 #include <sys/namei.h>
51 #include <sys/pioctl.h>
52 #include <sys/proc.h>
53 #include <sys/procfs.h>
54 #include <sys/racct.h>
55 #include <sys/resourcevar.h>
56 #include <sys/rwlock.h>
57 #include <sys/sbuf.h>
58 #include <sys/sf_buf.h>
59 #include <sys/smp.h>
60 #include <sys/systm.h>
61 #include <sys/signalvar.h>
62 #include <sys/stat.h>
63 #include <sys/sx.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/vnode.h>
68 #include <sys/syslog.h>
69 #include <sys/eventhandler.h>
70 #include <sys/user.h>
71
72 #include <net/zlib.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
81
82 #include <machine/elf.h>
83 #include <machine/md_var.h>
84
85 #define ELF_NOTE_ROUNDSIZE 4
86 #define OLD_EI_BRAND 8
87
88 static int __elfN(check_header)(const Elf_Ehdr *hdr);
89 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
90 const char *interp, int interp_name_len, int32_t *osrel);
91 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
92 u_long *entry, size_t pagesize);
93 static int __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
94 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
95 size_t pagesize);
96 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
97 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
98 int32_t *osrel);
99 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
100 static boolean_t __elfN(check_note)(struct image_params *imgp,
101 Elf_Brandnote *checknote, int32_t *osrel);
102 static vm_prot_t __elfN(trans_prot)(Elf_Word);
103 static Elf_Word __elfN(untrans_prot)(vm_prot_t);
104
105 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
106 "");
107
108 #ifdef COMPRESS_USER_CORES
109 static int compress_core(gzFile, char *, char *, unsigned int,
110 struct thread * td);
111 #endif
112 #define CORE_BUF_SIZE (16 * 1024)
113
114 int __elfN(fallback_brand) = -1;
115 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
116 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
117 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
118 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
119 &__elfN(fallback_brand));
120
121 static int elf_legacy_coredump = 0;
122 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
123 &elf_legacy_coredump, 0, "");
124
125 int __elfN(nxstack) =
126 #if defined(__amd64__) || defined(__powerpc64__) /* both 64 and 32 bit */
127 1;
128 #else
129 0;
130 #endif
131 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
132 nxstack, CTLFLAG_RW, &__elfN(nxstack), 0,
133 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) ": enable non-executable stack");
134
135 #if __ELF_WORD_SIZE == 32
136 #if defined(__amd64__) || defined(__ia64__)
137 int i386_read_exec = 0;
138 SYSCTL_INT(_kern_elf32, OID_AUTO, read_exec, CTLFLAG_RW, &i386_read_exec, 0,
139 "enable execution from readable segments");
140 #endif
141 #endif
142
143 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
144
145 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
146 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
147 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
148
149 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
150
151 Elf_Brandnote __elfN(freebsd_brandnote) = {
152 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
153 .hdr.n_descsz = sizeof(int32_t),
154 .hdr.n_type = 1,
155 .vendor = FREEBSD_ABI_VENDOR,
156 .flags = BN_TRANSLATE_OSREL,
157 .trans_osrel = __elfN(freebsd_trans_osrel)
158 };
159
160 static boolean_t
__elfN(freebsd_trans_osrel)161 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
162 {
163 uintptr_t p;
164
165 p = (uintptr_t)(note + 1);
166 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
167 *osrel = *(const int32_t *)(p);
168
169 return (TRUE);
170 }
171
172 static const char GNU_ABI_VENDOR[] = "GNU";
173 static int GNU_KFREEBSD_ABI_DESC = 3;
174
175 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
176 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
177 .hdr.n_descsz = 16, /* XXX at least 16 */
178 .hdr.n_type = 1,
179 .vendor = GNU_ABI_VENDOR,
180 .flags = BN_TRANSLATE_OSREL,
181 .trans_osrel = kfreebsd_trans_osrel
182 };
183
184 static boolean_t
kfreebsd_trans_osrel(const Elf_Note * note,int32_t * osrel)185 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
186 {
187 const Elf32_Word *desc;
188 uintptr_t p;
189
190 p = (uintptr_t)(note + 1);
191 p += roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE);
192
193 desc = (const Elf32_Word *)p;
194 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
195 return (FALSE);
196
197 /*
198 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
199 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
200 */
201 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
202
203 return (TRUE);
204 }
205
206 int
__elfN(insert_brand_entry)207 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
208 {
209 int i;
210
211 for (i = 0; i < MAX_BRANDS; i++) {
212 if (elf_brand_list[i] == NULL) {
213 elf_brand_list[i] = entry;
214 break;
215 }
216 }
217 if (i == MAX_BRANDS) {
218 printf("WARNING: %s: could not insert brandinfo entry: %p\n",
219 __func__, entry);
220 return (-1);
221 }
222 return (0);
223 }
224
225 int
__elfN(remove_brand_entry)226 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
227 {
228 int i;
229
230 for (i = 0; i < MAX_BRANDS; i++) {
231 if (elf_brand_list[i] == entry) {
232 elf_brand_list[i] = NULL;
233 break;
234 }
235 }
236 if (i == MAX_BRANDS)
237 return (-1);
238 return (0);
239 }
240
241 int
__elfN(brand_inuse)242 __elfN(brand_inuse)(Elf_Brandinfo *entry)
243 {
244 struct proc *p;
245 int rval = FALSE;
246
247 sx_slock(&allproc_lock);
248 FOREACH_PROC_IN_SYSTEM(p) {
249 if (p->p_sysent == entry->sysvec) {
250 rval = TRUE;
251 break;
252 }
253 }
254 sx_sunlock(&allproc_lock);
255
256 return (rval);
257 }
258
259 static Elf_Brandinfo *
__elfN(get_brandinfo)260 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
261 int interp_name_len, int32_t *osrel)
262 {
263 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
264 Elf_Brandinfo *bi;
265 boolean_t ret;
266 int i;
267
268 /*
269 * We support four types of branding -- (1) the ELF EI_OSABI field
270 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
271 * branding w/in the ELF header, (3) path of the `interp_path'
272 * field, and (4) the ".note.ABI-tag" ELF section.
273 */
274
275 /* Look for an ".note.ABI-tag" ELF section */
276 for (i = 0; i < MAX_BRANDS; i++) {
277 bi = elf_brand_list[i];
278 if (bi == NULL)
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 if (ret)
284 return (bi);
285 }
286 }
287
288 /* If the executable has a brand, search for it in the brand list. */
289 for (i = 0; i < MAX_BRANDS; i++) {
290 bi = elf_brand_list[i];
291 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
292 continue;
293 if (hdr->e_machine == bi->machine &&
294 (hdr->e_ident[EI_OSABI] == bi->brand ||
295 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
296 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
297 return (bi);
298 }
299
300 /* Lacking a known brand, search for a recognized interpreter. */
301 if (interp != NULL) {
302 for (i = 0; i < MAX_BRANDS; i++) {
303 bi = elf_brand_list[i];
304 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
305 continue;
306 if (hdr->e_machine == bi->machine &&
307 /* ELF image p_filesz includes terminating zero */
308 strlen(bi->interp_path) + 1 == interp_name_len &&
309 strncmp(interp, bi->interp_path, interp_name_len)
310 == 0)
311 return (bi);
312 }
313 }
314
315 /* Lacking a recognized interpreter, try the default brand */
316 for (i = 0; i < MAX_BRANDS; i++) {
317 bi = elf_brand_list[i];
318 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
319 continue;
320 if (hdr->e_machine == bi->machine &&
321 __elfN(fallback_brand) == bi->brand)
322 return (bi);
323 }
324 return (NULL);
325 }
326
327 static int
__elfN(check_header)328 __elfN(check_header)(const Elf_Ehdr *hdr)
329 {
330 Elf_Brandinfo *bi;
331 int i;
332
333 if (!IS_ELF(*hdr) ||
334 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
335 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
336 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
337 hdr->e_phentsize != sizeof(Elf_Phdr) ||
338 hdr->e_version != ELF_TARG_VER)
339 return (ENOEXEC);
340
341 /*
342 * Make sure we have at least one brand for this machine.
343 */
344
345 for (i = 0; i < MAX_BRANDS; i++) {
346 bi = elf_brand_list[i];
347 if (bi != NULL && bi->machine == hdr->e_machine)
348 break;
349 }
350 if (i == MAX_BRANDS)
351 return (ENOEXEC);
352
353 return (0);
354 }
355
356 static int
__elfN(map_partial)357 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
358 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
359 {
360 struct sf_buf *sf;
361 int error;
362 vm_offset_t off;
363
364 /*
365 * Create the page if it doesn't exist yet. Ignore errors.
366 */
367 vm_map_lock(map);
368 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
369 VM_PROT_ALL, VM_PROT_ALL, 0);
370 vm_map_unlock(map);
371
372 /*
373 * Find the page from the underlying object.
374 */
375 if (object) {
376 sf = vm_imgact_map_page(object, offset);
377 if (sf == NULL)
378 return (KERN_FAILURE);
379 off = offset - trunc_page(offset);
380 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
381 end - start);
382 vm_imgact_unmap_page(sf);
383 if (error) {
384 return (KERN_FAILURE);
385 }
386 }
387
388 return (KERN_SUCCESS);
389 }
390
391 static int
__elfN(map_insert)392 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
393 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
394 {
395 struct sf_buf *sf;
396 vm_offset_t off;
397 vm_size_t sz;
398 int error, rv;
399
400 if (start != trunc_page(start)) {
401 rv = __elfN(map_partial)(map, object, offset, start,
402 round_page(start), prot);
403 if (rv)
404 return (rv);
405 offset += round_page(start) - start;
406 start = round_page(start);
407 }
408 if (end != round_page(end)) {
409 rv = __elfN(map_partial)(map, object, offset +
410 trunc_page(end) - start, trunc_page(end), end, prot);
411 if (rv)
412 return (rv);
413 end = trunc_page(end);
414 }
415 if (end > start) {
416 if (offset & PAGE_MASK) {
417 /*
418 * The mapping is not page aligned. This means we have
419 * to copy the data. Sigh.
420 */
421 rv = vm_map_find(map, NULL, 0, &start, end - start, 0,
422 VMFS_NO_SPACE, prot | VM_PROT_WRITE, VM_PROT_ALL,
423 0);
424 if (rv)
425 return (rv);
426 if (object == NULL)
427 return (KERN_SUCCESS);
428 for (; start < end; start += sz) {
429 sf = vm_imgact_map_page(object, offset);
430 if (sf == NULL)
431 return (KERN_FAILURE);
432 off = offset - trunc_page(offset);
433 sz = end - start;
434 if (sz > PAGE_SIZE - off)
435 sz = PAGE_SIZE - off;
436 error = copyout((caddr_t)sf_buf_kva(sf) + off,
437 (caddr_t)start, sz);
438 vm_imgact_unmap_page(sf);
439 if (error) {
440 return (KERN_FAILURE);
441 }
442 offset += sz;
443 }
444 rv = KERN_SUCCESS;
445 } else {
446 vm_object_reference(object);
447 vm_map_lock(map);
448 rv = vm_map_insert(map, object, offset, start, end,
449 prot, VM_PROT_ALL, cow);
450 vm_map_unlock(map);
451 if (rv != KERN_SUCCESS)
452 vm_object_deallocate(object);
453 }
454 return (rv);
455 } else {
456 return (KERN_SUCCESS);
457 }
458 }
459
460 static int
__elfN(load_section)461 __elfN(load_section)(struct image_params *imgp, vm_offset_t offset,
462 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
463 size_t pagesize)
464 {
465 struct sf_buf *sf;
466 size_t map_len;
467 vm_map_t map;
468 vm_object_t object;
469 vm_offset_t map_addr;
470 int error, rv, cow;
471 size_t copy_len;
472 vm_offset_t file_addr;
473
474 /*
475 * It's necessary to fail if the filsz + offset taken from the
476 * header is greater than the actual file pager object's size.
477 * If we were to allow this, then the vm_map_find() below would
478 * walk right off the end of the file object and into the ether.
479 *
480 * While I'm here, might as well check for something else that
481 * is invalid: filsz cannot be greater than memsz.
482 */
483 if ((off_t)filsz + offset > imgp->attr->va_size || filsz > memsz) {
484 uprintf("elf_load_section: truncated ELF file\n");
485 return (ENOEXEC);
486 }
487
488 object = imgp->object;
489 map = &imgp->proc->p_vmspace->vm_map;
490 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
491 file_addr = trunc_page_ps(offset, pagesize);
492
493 /*
494 * We have two choices. We can either clear the data in the last page
495 * of an oversized mapping, or we can start the anon mapping a page
496 * early and copy the initialized data into that first page. We
497 * choose the second..
498 */
499 if (memsz > filsz)
500 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
501 else
502 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
503
504 if (map_len != 0) {
505 /* cow flags: don't dump readonly sections in core */
506 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
507 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
508
509 rv = __elfN(map_insert)(map,
510 object,
511 file_addr, /* file offset */
512 map_addr, /* virtual start */
513 map_addr + map_len,/* virtual end */
514 prot,
515 cow);
516 if (rv != KERN_SUCCESS)
517 return (EINVAL);
518
519 /* we can stop now if we've covered it all */
520 if (memsz == filsz) {
521 return (0);
522 }
523 }
524
525
526 /*
527 * We have to get the remaining bit of the file into the first part
528 * of the oversized map segment. This is normally because the .data
529 * segment in the file is extended to provide bss. It's a neat idea
530 * to try and save a page, but it's a pain in the behind to implement.
531 */
532 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
533 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
534 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
535 map_addr;
536
537 /* This had damn well better be true! */
538 if (map_len != 0) {
539 rv = __elfN(map_insert)(map, NULL, 0, map_addr, map_addr +
540 map_len, VM_PROT_ALL, 0);
541 if (rv != KERN_SUCCESS) {
542 return (EINVAL);
543 }
544 }
545
546 if (copy_len != 0) {
547 vm_offset_t off;
548
549 sf = vm_imgact_map_page(object, offset + filsz);
550 if (sf == NULL)
551 return (EIO);
552
553 /* send the page fragment to user space */
554 off = trunc_page_ps(offset + filsz, pagesize) -
555 trunc_page(offset + filsz);
556 error = copyout((caddr_t)sf_buf_kva(sf) + off,
557 (caddr_t)map_addr, copy_len);
558 vm_imgact_unmap_page(sf);
559 if (error) {
560 return (error);
561 }
562 }
563
564 /*
565 * set it to the specified protection.
566 * XXX had better undo the damage from pasting over the cracks here!
567 */
568 vm_map_protect(map, trunc_page(map_addr), round_page(map_addr +
569 map_len), prot, FALSE);
570
571 return (0);
572 }
573
574 /*
575 * Load the file "file" into memory. It may be either a shared object
576 * or an executable.
577 *
578 * The "addr" reference parameter is in/out. On entry, it specifies
579 * the address where a shared object should be loaded. If the file is
580 * an executable, this value is ignored. On exit, "addr" specifies
581 * where the file was actually loaded.
582 *
583 * The "entry" reference parameter is out only. On exit, it specifies
584 * the entry point for the loaded file.
585 */
586 static int
__elfN(load_file)587 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
588 u_long *entry, size_t pagesize)
589 {
590 struct {
591 struct nameidata nd;
592 struct vattr attr;
593 struct image_params image_params;
594 } *tempdata;
595 const Elf_Ehdr *hdr = NULL;
596 const Elf_Phdr *phdr = NULL;
597 struct nameidata *nd;
598 struct vattr *attr;
599 struct image_params *imgp;
600 vm_prot_t prot;
601 u_long rbase;
602 u_long base_addr = 0;
603 int error, i, numsegs;
604
605 #ifdef CAPABILITY_MODE
606 /*
607 * XXXJA: This check can go away once we are sufficiently confident
608 * that the checks in namei() are correct.
609 */
610 if (IN_CAPABILITY_MODE(curthread))
611 return (ECAPMODE);
612 #endif
613
614 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
615 nd = &tempdata->nd;
616 attr = &tempdata->attr;
617 imgp = &tempdata->image_params;
618
619 /*
620 * Initialize part of the common data
621 */
622 imgp->proc = p;
623 imgp->attr = attr;
624 imgp->firstpage = NULL;
625 imgp->image_header = NULL;
626 imgp->object = NULL;
627 imgp->execlabel = NULL;
628
629 NDINIT(nd, LOOKUP, LOCKLEAF | FOLLOW, UIO_SYSSPACE, file, curthread);
630 if ((error = namei(nd)) != 0) {
631 nd->ni_vp = NULL;
632 goto fail;
633 }
634 NDFREE(nd, NDF_ONLY_PNBUF);
635 imgp->vp = nd->ni_vp;
636
637 /*
638 * Check permissions, modes, uid, etc on the file, and "open" it.
639 */
640 error = exec_check_permissions(imgp);
641 if (error)
642 goto fail;
643
644 error = exec_map_first_page(imgp);
645 if (error)
646 goto fail;
647
648 /*
649 * Also make certain that the interpreter stays the same, so set
650 * its VV_TEXT flag, too.
651 */
652 VOP_SET_TEXT(nd->ni_vp);
653
654 imgp->object = nd->ni_vp->v_object;
655
656 hdr = (const Elf_Ehdr *)imgp->image_header;
657 if ((error = __elfN(check_header)(hdr)) != 0)
658 goto fail;
659 if (hdr->e_type == ET_DYN)
660 rbase = *addr;
661 else if (hdr->e_type == ET_EXEC)
662 rbase = 0;
663 else {
664 error = ENOEXEC;
665 goto fail;
666 }
667
668 /* Only support headers that fit within first page for now */
669 if ((hdr->e_phoff > PAGE_SIZE) ||
670 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
671 error = ENOEXEC;
672 goto fail;
673 }
674
675 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
676 if (!aligned(phdr, Elf_Addr)) {
677 error = ENOEXEC;
678 goto fail;
679 }
680
681 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
682 if (phdr[i].p_type == PT_LOAD && phdr[i].p_memsz != 0) {
683 /* Loadable segment */
684 prot = __elfN(trans_prot)(phdr[i].p_flags);
685 error = __elfN(load_section)(imgp, phdr[i].p_offset,
686 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
687 phdr[i].p_memsz, phdr[i].p_filesz, prot, pagesize);
688 if (error != 0)
689 goto fail;
690 /*
691 * Establish the base address if this is the
692 * first segment.
693 */
694 if (numsegs == 0)
695 base_addr = trunc_page(phdr[i].p_vaddr +
696 rbase);
697 numsegs++;
698 }
699 }
700 *addr = base_addr;
701 *entry = (unsigned long)hdr->e_entry + rbase;
702
703 fail:
704 if (imgp->firstpage)
705 exec_unmap_first_page(imgp);
706
707 if (nd->ni_vp)
708 vput(nd->ni_vp);
709
710 free(tempdata, M_TEMP);
711
712 return (error);
713 }
714
715 static int
__CONCAT(exec_,__elfN (imgact))716 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
717 {
718 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
719 const Elf_Phdr *phdr;
720 Elf_Auxargs *elf_auxargs;
721 struct vmspace *vmspace;
722 vm_prot_t prot;
723 u_long text_size = 0, data_size = 0, total_size = 0;
724 u_long text_addr = 0, data_addr = 0;
725 u_long seg_size, seg_addr;
726 u_long addr, baddr, et_dyn_addr, entry = 0, proghdr = 0;
727 int32_t osrel = 0;
728 int error = 0, i, n, interp_name_len = 0;
729 const char *interp = NULL, *newinterp = NULL;
730 Elf_Brandinfo *brand_info;
731 char *path;
732 struct sysentvec *sv;
733
734 /*
735 * Do we have a valid ELF header ?
736 *
737 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
738 * if particular brand doesn't support it.
739 */
740 if (__elfN(check_header)(hdr) != 0 ||
741 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
742 return (-1);
743
744 /*
745 * From here on down, we return an errno, not -1, as we've
746 * detected an ELF file.
747 */
748
749 if ((hdr->e_phoff > PAGE_SIZE) ||
750 (u_int)hdr->e_phentsize * hdr->e_phnum > PAGE_SIZE - hdr->e_phoff) {
751 /* Only support headers in first page for now */
752 return (ENOEXEC);
753 }
754 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
755 if (!aligned(phdr, Elf_Addr))
756 return (ENOEXEC);
757 n = 0;
758 baddr = 0;
759 for (i = 0; i < hdr->e_phnum; i++) {
760 switch (phdr[i].p_type) {
761 case PT_LOAD:
762 if (n == 0)
763 baddr = phdr[i].p_vaddr;
764 n++;
765 break;
766 case PT_INTERP:
767 /* Path to interpreter */
768 if (phdr[i].p_filesz > MAXPATHLEN ||
769 phdr[i].p_offset > PAGE_SIZE ||
770 phdr[i].p_filesz > PAGE_SIZE - phdr[i].p_offset)
771 return (ENOEXEC);
772 interp = imgp->image_header + phdr[i].p_offset;
773 interp_name_len = phdr[i].p_filesz;
774 break;
775 case PT_GNU_STACK:
776 if (__elfN(nxstack))
777 imgp->stack_prot =
778 __elfN(trans_prot)(phdr[i].p_flags);
779 imgp->stack_sz = phdr[i].p_memsz;
780 break;
781 }
782 }
783
784 brand_info = __elfN(get_brandinfo)(imgp, interp, interp_name_len,
785 &osrel);
786 if (brand_info == NULL) {
787 uprintf("ELF binary type \"%u\" not known.\n",
788 hdr->e_ident[EI_OSABI]);
789 return (ENOEXEC);
790 }
791 if (hdr->e_type == ET_DYN) {
792 if ((brand_info->flags & BI_CAN_EXEC_DYN) == 0)
793 return (ENOEXEC);
794 /*
795 * Honour the base load address from the dso if it is
796 * non-zero for some reason.
797 */
798 if (baddr == 0)
799 et_dyn_addr = ET_DYN_LOAD_ADDR;
800 else
801 et_dyn_addr = 0;
802 } else
803 et_dyn_addr = 0;
804 sv = brand_info->sysvec;
805 if (interp != NULL && brand_info->interp_newpath != NULL)
806 newinterp = brand_info->interp_newpath;
807
808 /*
809 * Avoid a possible deadlock if the current address space is destroyed
810 * and that address space maps the locked vnode. In the common case,
811 * the locked vnode's v_usecount is decremented but remains greater
812 * than zero. Consequently, the vnode lock is not needed by vrele().
813 * However, in cases where the vnode lock is external, such as nullfs,
814 * v_usecount may become zero.
815 *
816 * The VV_TEXT flag prevents modifications to the executable while
817 * the vnode is unlocked.
818 */
819 VOP_UNLOCK(imgp->vp, 0);
820
821 error = exec_new_vmspace(imgp, sv);
822 imgp->proc->p_sysent = sv;
823
824 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
825 if (error)
826 return (error);
827
828 for (i = 0; i < hdr->e_phnum; i++) {
829 switch (phdr[i].p_type) {
830 case PT_LOAD: /* Loadable segment */
831 if (phdr[i].p_memsz == 0)
832 break;
833 prot = __elfN(trans_prot)(phdr[i].p_flags);
834 error = __elfN(load_section)(imgp, phdr[i].p_offset,
835 (caddr_t)(uintptr_t)phdr[i].p_vaddr + et_dyn_addr,
836 phdr[i].p_memsz, phdr[i].p_filesz, prot,
837 sv->sv_pagesize);
838 if (error != 0)
839 return (error);
840
841 /*
842 * If this segment contains the program headers,
843 * remember their virtual address for the AT_PHDR
844 * aux entry. Static binaries don't usually include
845 * a PT_PHDR entry.
846 */
847 if (phdr[i].p_offset == 0 &&
848 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
849 <= phdr[i].p_filesz)
850 proghdr = phdr[i].p_vaddr + hdr->e_phoff +
851 et_dyn_addr;
852
853 seg_addr = trunc_page(phdr[i].p_vaddr + et_dyn_addr);
854 seg_size = round_page(phdr[i].p_memsz +
855 phdr[i].p_vaddr + et_dyn_addr - seg_addr);
856
857 /*
858 * Make the largest executable segment the official
859 * text segment and all others data.
860 *
861 * Note that obreak() assumes that data_addr +
862 * data_size == end of data load area, and the ELF
863 * file format expects segments to be sorted by
864 * address. If multiple data segments exist, the
865 * last one will be used.
866 */
867
868 if (phdr[i].p_flags & PF_X && text_size < seg_size) {
869 text_size = seg_size;
870 text_addr = seg_addr;
871 } else {
872 data_size = seg_size;
873 data_addr = seg_addr;
874 }
875 total_size += seg_size;
876 break;
877 case PT_PHDR: /* Program header table info */
878 proghdr = phdr[i].p_vaddr + et_dyn_addr;
879 break;
880 default:
881 break;
882 }
883 }
884
885 if (data_addr == 0 && data_size == 0) {
886 data_addr = text_addr;
887 data_size = text_size;
888 }
889
890 entry = (u_long)hdr->e_entry + et_dyn_addr;
891
892 /*
893 * Check limits. It should be safe to check the
894 * limits after loading the segments since we do
895 * not actually fault in all the segments pages.
896 */
897 PROC_LOCK(imgp->proc);
898 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
899 text_size > maxtsiz ||
900 total_size > lim_cur(imgp->proc, RLIMIT_VMEM) ||
901 racct_set(imgp->proc, RACCT_DATA, data_size) != 0 ||
902 racct_set(imgp->proc, RACCT_VMEM, total_size) != 0) {
903 PROC_UNLOCK(imgp->proc);
904 return (ENOMEM);
905 }
906
907 vmspace = imgp->proc->p_vmspace;
908 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
909 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
910 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
911 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
912
913 /*
914 * We load the dynamic linker where a userland call
915 * to mmap(0, ...) would put it. The rationale behind this
916 * calculation is that it leaves room for the heap to grow to
917 * its maximum allowed size.
918 */
919 addr = round_page((vm_offset_t)vmspace->vm_daddr + lim_max(imgp->proc,
920 RLIMIT_DATA));
921 PROC_UNLOCK(imgp->proc);
922
923 imgp->entry_addr = entry;
924
925 if (interp != NULL) {
926 int have_interp = FALSE;
927 VOP_UNLOCK(imgp->vp, 0);
928 if (brand_info->emul_path != NULL &&
929 brand_info->emul_path[0] != '\0') {
930 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
931 snprintf(path, MAXPATHLEN, "%s%s",
932 brand_info->emul_path, interp);
933 error = __elfN(load_file)(imgp->proc, path, &addr,
934 &imgp->entry_addr, sv->sv_pagesize);
935 free(path, M_TEMP);
936 if (error == 0)
937 have_interp = TRUE;
938 }
939 if (!have_interp && newinterp != NULL) {
940 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
941 &imgp->entry_addr, sv->sv_pagesize);
942 if (error == 0)
943 have_interp = TRUE;
944 }
945 if (!have_interp) {
946 error = __elfN(load_file)(imgp->proc, interp, &addr,
947 &imgp->entry_addr, sv->sv_pagesize);
948 }
949 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
950 if (error != 0) {
951 uprintf("ELF interpreter %s not found\n", interp);
952 return (error);
953 }
954 } else
955 addr = et_dyn_addr;
956
957 /*
958 * Construct auxargs table (used by the fixup routine)
959 */
960 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
961 elf_auxargs->execfd = -1;
962 elf_auxargs->phdr = proghdr;
963 elf_auxargs->phent = hdr->e_phentsize;
964 elf_auxargs->phnum = hdr->e_phnum;
965 elf_auxargs->pagesz = PAGE_SIZE;
966 elf_auxargs->base = addr;
967 elf_auxargs->flags = 0;
968 elf_auxargs->entry = entry;
969
970 imgp->auxargs = elf_auxargs;
971 imgp->interpreted = 0;
972 imgp->reloc_base = addr;
973 imgp->proc->p_osrel = osrel;
974
975 return (error);
976 }
977
978 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
979
980 int
__elfN(freebsd_fixup)981 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
982 {
983 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
984 Elf_Addr *base;
985 Elf_Addr *pos;
986
987 base = (Elf_Addr *)*stack_base;
988 pos = base + (imgp->args->argc + imgp->args->envc + 2);
989
990 if (args->execfd != -1)
991 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
992 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
993 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
994 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
995 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
996 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
997 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
998 AUXARGS_ENTRY(pos, AT_BASE, args->base);
999 if (imgp->execpathp != 0)
1000 AUXARGS_ENTRY(pos, AT_EXECPATH, imgp->execpathp);
1001 AUXARGS_ENTRY(pos, AT_OSRELDATE,
1002 imgp->proc->p_ucred->cr_prison->pr_osreldate);
1003 if (imgp->canary != 0) {
1004 AUXARGS_ENTRY(pos, AT_CANARY, imgp->canary);
1005 AUXARGS_ENTRY(pos, AT_CANARYLEN, imgp->canarylen);
1006 }
1007 AUXARGS_ENTRY(pos, AT_NCPUS, mp_ncpus);
1008 if (imgp->pagesizes != 0) {
1009 AUXARGS_ENTRY(pos, AT_PAGESIZES, imgp->pagesizes);
1010 AUXARGS_ENTRY(pos, AT_PAGESIZESLEN, imgp->pagesizeslen);
1011 }
1012 if (imgp->sysent->sv_timekeep_base != 0) {
1013 AUXARGS_ENTRY(pos, AT_TIMEKEEP,
1014 imgp->sysent->sv_timekeep_base);
1015 }
1016 AUXARGS_ENTRY(pos, AT_STACKPROT, imgp->sysent->sv_shared_page_obj
1017 != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1018 imgp->sysent->sv_stackprot);
1019 AUXARGS_ENTRY(pos, AT_NULL, 0);
1020
1021 free(imgp->auxargs, M_TEMP);
1022 imgp->auxargs = NULL;
1023
1024 base--;
1025 suword(base, (long)imgp->args->argc);
1026 *stack_base = (register_t *)base;
1027 return (0);
1028 }
1029
1030 /*
1031 * Code for generating ELF core dumps.
1032 */
1033
1034 typedef void (*segment_callback)(vm_map_entry_t, void *);
1035
1036 /* Closure for cb_put_phdr(). */
1037 struct phdr_closure {
1038 Elf_Phdr *phdr; /* Program header to fill in */
1039 Elf_Off offset; /* Offset of segment in core file */
1040 };
1041
1042 /* Closure for cb_size_segment(). */
1043 struct sseg_closure {
1044 int count; /* Count of writable segments. */
1045 size_t size; /* Total size of all writable segments. */
1046 };
1047
1048 typedef void (*outfunc_t)(void *, struct sbuf *, size_t *);
1049
1050 struct note_info {
1051 int type; /* Note type. */
1052 outfunc_t outfunc; /* Output function. */
1053 void *outarg; /* Argument for the output function. */
1054 size_t outsize; /* Output size. */
1055 TAILQ_ENTRY(note_info) link; /* Link to the next note info. */
1056 };
1057
1058 TAILQ_HEAD(note_info_list, note_info);
1059
1060 static void cb_put_phdr(vm_map_entry_t, void *);
1061 static void cb_size_segment(vm_map_entry_t, void *);
1062 static void each_writable_segment(struct thread *, segment_callback, void *);
1063 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
1064 int, void *, size_t, struct note_info_list *, size_t, gzFile);
1065 static void __elfN(prepare_notes)(struct thread *, struct note_info_list *,
1066 size_t *);
1067 static void __elfN(puthdr)(struct thread *, void *, size_t, int, size_t);
1068 static void __elfN(putnote)(struct note_info *, struct sbuf *);
1069 static size_t register_note(struct note_info_list *, int, outfunc_t, void *);
1070 static int sbuf_drain_core_output(void *, const char *, int);
1071 static int sbuf_drain_count(void *arg, const char *data, int len);
1072
1073 static void __elfN(note_fpregset)(void *, struct sbuf *, size_t *);
1074 static void __elfN(note_prpsinfo)(void *, struct sbuf *, size_t *);
1075 static void __elfN(note_prstatus)(void *, struct sbuf *, size_t *);
1076 static void __elfN(note_threadmd)(void *, struct sbuf *, size_t *);
1077 static void __elfN(note_thrmisc)(void *, struct sbuf *, size_t *);
1078 static void __elfN(note_procstat_auxv)(void *, struct sbuf *, size_t *);
1079 static void __elfN(note_procstat_proc)(void *, struct sbuf *, size_t *);
1080 static void __elfN(note_procstat_psstrings)(void *, struct sbuf *, size_t *);
1081 static void note_procstat_files(void *, struct sbuf *, size_t *);
1082 static void note_procstat_groups(void *, struct sbuf *, size_t *);
1083 static void note_procstat_osrel(void *, struct sbuf *, size_t *);
1084 static void note_procstat_rlimit(void *, struct sbuf *, size_t *);
1085 static void note_procstat_umask(void *, struct sbuf *, size_t *);
1086 static void note_procstat_vmmap(void *, struct sbuf *, size_t *);
1087
1088 #ifdef COMPRESS_USER_CORES
1089 extern int compress_user_cores;
1090 extern int compress_user_cores_gzlevel;
1091 #endif
1092
1093 static int
core_output(struct vnode * vp,void * base,size_t len,off_t offset,struct ucred * active_cred,struct ucred * file_cred,struct thread * td,char * core_buf,gzFile gzfile)1094 core_output(struct vnode *vp, void *base, size_t len, off_t offset,
1095 struct ucred *active_cred, struct ucred *file_cred,
1096 struct thread *td, char *core_buf, gzFile gzfile) {
1097
1098 int error;
1099 if (gzfile) {
1100 #ifdef COMPRESS_USER_CORES
1101 error = compress_core(gzfile, base, core_buf, len, td);
1102 #else
1103 panic("shouldn't be here");
1104 #endif
1105 } else {
1106 error = vn_rdwr_inchunks(UIO_WRITE, vp, base, len, offset,
1107 UIO_USERSPACE, IO_UNIT | IO_DIRECT, active_cred, file_cred,
1108 NULL, td);
1109 }
1110 return (error);
1111 }
1112
1113 /* Coredump output parameters for sbuf drain routine. */
1114 struct sbuf_drain_core_params {
1115 off_t offset;
1116 struct ucred *active_cred;
1117 struct ucred *file_cred;
1118 struct thread *td;
1119 struct vnode *vp;
1120 #ifdef COMPRESS_USER_CORES
1121 gzFile gzfile;
1122 #endif
1123 };
1124
1125 /*
1126 * Drain into a core file.
1127 */
1128 static int
sbuf_drain_core_output(void * arg,const char * data,int len)1129 sbuf_drain_core_output(void *arg, const char *data, int len)
1130 {
1131 struct sbuf_drain_core_params *p;
1132 int error, locked;
1133
1134 p = (struct sbuf_drain_core_params *)arg;
1135
1136 /*
1137 * Some kern_proc out routines that print to this sbuf may
1138 * call us with the process lock held. Draining with the
1139 * non-sleepable lock held is unsafe. The lock is needed for
1140 * those routines when dumping a live process. In our case we
1141 * can safely release the lock before draining and acquire
1142 * again after.
1143 */
1144 locked = PROC_LOCKED(p->td->td_proc);
1145 if (locked)
1146 PROC_UNLOCK(p->td->td_proc);
1147 #ifdef COMPRESS_USER_CORES
1148 if (p->gzfile != Z_NULL)
1149 error = compress_core(p->gzfile, NULL, __DECONST(char *, data),
1150 len, p->td);
1151 else
1152 #endif
1153 error = vn_rdwr_inchunks(UIO_WRITE, p->vp,
1154 __DECONST(void *, data), len, p->offset, UIO_SYSSPACE,
1155 IO_UNIT | IO_DIRECT, p->active_cred, p->file_cred, NULL,
1156 p->td);
1157 if (locked)
1158 PROC_LOCK(p->td->td_proc);
1159 if (error != 0)
1160 return (-error);
1161 p->offset += len;
1162 return (len);
1163 }
1164
1165 /*
1166 * Drain into a counter.
1167 */
1168 static int
sbuf_drain_count(void * arg,const char * data __unused,int len)1169 sbuf_drain_count(void *arg, const char *data __unused, int len)
1170 {
1171 size_t *sizep;
1172
1173 sizep = (size_t *)arg;
1174 *sizep += len;
1175 return (len);
1176 }
1177
1178 int
__elfN(coredump)1179 __elfN(coredump)(struct thread *td, struct vnode *vp, off_t limit, int flags)
1180 {
1181 struct ucred *cred = td->td_ucred;
1182 int error = 0;
1183 struct sseg_closure seginfo;
1184 struct note_info_list notelst;
1185 struct note_info *ninfo;
1186 void *hdr;
1187 size_t hdrsize, notesz, coresize;
1188
1189 gzFile gzfile = Z_NULL;
1190 char *core_buf = NULL;
1191 #ifdef COMPRESS_USER_CORES
1192 char gzopen_flags[8];
1193 char *p;
1194 int doing_compress = flags & IMGACT_CORE_COMPRESS;
1195 #endif
1196
1197 hdr = NULL;
1198 TAILQ_INIT(¬elst);
1199
1200 #ifdef COMPRESS_USER_CORES
1201 if (doing_compress) {
1202 p = gzopen_flags;
1203 *p++ = 'w';
1204 if (compress_user_cores_gzlevel >= 0 &&
1205 compress_user_cores_gzlevel <= 9)
1206 *p++ = '0' + compress_user_cores_gzlevel;
1207 *p = 0;
1208 gzfile = gz_open("", gzopen_flags, vp);
1209 if (gzfile == Z_NULL) {
1210 error = EFAULT;
1211 goto done;
1212 }
1213 core_buf = malloc(CORE_BUF_SIZE, M_TEMP, M_WAITOK | M_ZERO);
1214 if (!core_buf) {
1215 error = ENOMEM;
1216 goto done;
1217 }
1218 }
1219 #endif
1220
1221 /* Size the program segments. */
1222 seginfo.count = 0;
1223 seginfo.size = 0;
1224 each_writable_segment(td, cb_size_segment, &seginfo);
1225
1226 /*
1227 * Collect info about the core file header area.
1228 */
1229 hdrsize = sizeof(Elf_Ehdr) + sizeof(Elf_Phdr) * (1 + seginfo.count);
1230 __elfN(prepare_notes)(td, ¬elst, ¬esz);
1231 coresize = round_page(hdrsize + notesz) + seginfo.size;
1232
1233 #ifdef RACCT
1234 PROC_LOCK(td->td_proc);
1235 error = racct_add(td->td_proc, RACCT_CORE, coresize);
1236 PROC_UNLOCK(td->td_proc);
1237 if (error != 0) {
1238 error = EFAULT;
1239 goto done;
1240 }
1241 #endif
1242 if (coresize >= limit) {
1243 error = EFAULT;
1244 goto done;
1245 }
1246
1247 /*
1248 * Allocate memory for building the header, fill it up,
1249 * and write it out following the notes.
1250 */
1251 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1252 if (hdr == NULL) {
1253 error = EINVAL;
1254 goto done;
1255 }
1256 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize,
1257 ¬elst, notesz, gzfile);
1258
1259 /* Write the contents of all of the writable segments. */
1260 if (error == 0) {
1261 Elf_Phdr *php;
1262 off_t offset;
1263 int i;
1264
1265 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1266 offset = round_page(hdrsize + notesz);
1267 for (i = 0; i < seginfo.count; i++) {
1268 error = core_output(vp, (caddr_t)(uintptr_t)php->p_vaddr,
1269 php->p_filesz, offset, cred, NOCRED, curthread, core_buf, gzfile);
1270 if (error != 0)
1271 break;
1272 offset += php->p_filesz;
1273 php++;
1274 }
1275 }
1276 if (error) {
1277 log(LOG_WARNING,
1278 "Failed to write core file for process %s (error %d)\n",
1279 curproc->p_comm, error);
1280 }
1281
1282 done:
1283 #ifdef COMPRESS_USER_CORES
1284 if (core_buf)
1285 free(core_buf, M_TEMP);
1286 if (gzfile)
1287 gzclose(gzfile);
1288 #endif
1289 while ((ninfo = TAILQ_FIRST(¬elst)) != NULL) {
1290 TAILQ_REMOVE(¬elst, ninfo, link);
1291 free(ninfo, M_TEMP);
1292 }
1293 if (hdr != NULL)
1294 free(hdr, M_TEMP);
1295
1296 return (error);
1297 }
1298
1299 /*
1300 * A callback for each_writable_segment() to write out the segment's
1301 * program header entry.
1302 */
1303 static void
cb_put_phdr(entry,closure)1304 cb_put_phdr(entry, closure)
1305 vm_map_entry_t entry;
1306 void *closure;
1307 {
1308 struct phdr_closure *phc = (struct phdr_closure *)closure;
1309 Elf_Phdr *phdr = phc->phdr;
1310
1311 phc->offset = round_page(phc->offset);
1312
1313 phdr->p_type = PT_LOAD;
1314 phdr->p_offset = phc->offset;
1315 phdr->p_vaddr = entry->start;
1316 phdr->p_paddr = 0;
1317 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1318 phdr->p_align = PAGE_SIZE;
1319 phdr->p_flags = __elfN(untrans_prot)(entry->protection);
1320
1321 phc->offset += phdr->p_filesz;
1322 phc->phdr++;
1323 }
1324
1325 /*
1326 * A callback for each_writable_segment() to gather information about
1327 * the number of segments and their total size.
1328 */
1329 static void
cb_size_segment(entry,closure)1330 cb_size_segment(entry, closure)
1331 vm_map_entry_t entry;
1332 void *closure;
1333 {
1334 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1335
1336 ssc->count++;
1337 ssc->size += entry->end - entry->start;
1338 }
1339
1340 /*
1341 * For each writable segment in the process's memory map, call the given
1342 * function with a pointer to the map entry and some arbitrary
1343 * caller-supplied data.
1344 */
1345 static void
each_writable_segment(td,func,closure)1346 each_writable_segment(td, func, closure)
1347 struct thread *td;
1348 segment_callback func;
1349 void *closure;
1350 {
1351 struct proc *p = td->td_proc;
1352 vm_map_t map = &p->p_vmspace->vm_map;
1353 vm_map_entry_t entry;
1354 vm_object_t backing_object, object;
1355 boolean_t ignore_entry;
1356
1357 vm_map_lock_read(map);
1358 for (entry = map->header.next; entry != &map->header;
1359 entry = entry->next) {
1360 /*
1361 * Don't dump inaccessible mappings, deal with legacy
1362 * coredump mode.
1363 *
1364 * Note that read-only segments related to the elf binary
1365 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1366 * need to arbitrarily ignore such segments.
1367 */
1368 if (elf_legacy_coredump) {
1369 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1370 continue;
1371 } else {
1372 if ((entry->protection & VM_PROT_ALL) == 0)
1373 continue;
1374 }
1375
1376 /*
1377 * Dont include memory segment in the coredump if
1378 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1379 * madvise(2). Do not dump submaps (i.e. parts of the
1380 * kernel map).
1381 */
1382 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1383 continue;
1384
1385 if ((object = entry->object.vm_object) == NULL)
1386 continue;
1387
1388 /* Ignore memory-mapped devices and such things. */
1389 VM_OBJECT_RLOCK(object);
1390 while ((backing_object = object->backing_object) != NULL) {
1391 VM_OBJECT_RLOCK(backing_object);
1392 VM_OBJECT_RUNLOCK(object);
1393 object = backing_object;
1394 }
1395 ignore_entry = object->type != OBJT_DEFAULT &&
1396 object->type != OBJT_SWAP && object->type != OBJT_VNODE &&
1397 object->type != OBJT_PHYS;
1398 VM_OBJECT_RUNLOCK(object);
1399 if (ignore_entry)
1400 continue;
1401
1402 (*func)(entry, closure);
1403 }
1404 vm_map_unlock_read(map);
1405 }
1406
1407 /*
1408 * Write the core file header to the file, including padding up to
1409 * the page boundary.
1410 */
1411 static int
__elfN(corehdr)1412 __elfN(corehdr)(struct thread *td, struct vnode *vp, struct ucred *cred,
1413 int numsegs, void *hdr, size_t hdrsize, struct note_info_list *notelst,
1414 size_t notesz, gzFile gzfile)
1415 {
1416 struct sbuf_drain_core_params params;
1417 struct note_info *ninfo;
1418 struct sbuf *sb;
1419 int error;
1420
1421 /* Fill in the header. */
1422 bzero(hdr, hdrsize);
1423 __elfN(puthdr)(td, hdr, hdrsize, numsegs, notesz);
1424
1425 params.offset = 0;
1426 params.active_cred = cred;
1427 params.file_cred = NOCRED;
1428 params.td = td;
1429 params.vp = vp;
1430 #ifdef COMPRESS_USER_CORES
1431 params.gzfile = gzfile;
1432 #endif
1433 sb = sbuf_new(NULL, NULL, CORE_BUF_SIZE, SBUF_FIXEDLEN);
1434 sbuf_set_drain(sb, sbuf_drain_core_output, ¶ms);
1435 sbuf_start_section(sb, NULL);
1436 sbuf_bcat(sb, hdr, hdrsize);
1437 TAILQ_FOREACH(ninfo, notelst, link)
1438 __elfN(putnote)(ninfo, sb);
1439 /* Align up to a page boundary for the program segments. */
1440 sbuf_end_section(sb, -1, PAGE_SIZE, 0);
1441 error = sbuf_finish(sb);
1442 sbuf_delete(sb);
1443
1444 return (error);
1445 }
1446
1447 static void
__elfN(prepare_notes)1448 __elfN(prepare_notes)(struct thread *td, struct note_info_list *list,
1449 size_t *sizep)
1450 {
1451 struct proc *p;
1452 struct thread *thr;
1453 size_t size;
1454
1455 p = td->td_proc;
1456 size = 0;
1457
1458 size += register_note(list, NT_PRPSINFO, __elfN(note_prpsinfo), p);
1459
1460 /*
1461 * To have the debugger select the right thread (LWP) as the initial
1462 * thread, we dump the state of the thread passed to us in td first.
1463 * This is the thread that causes the core dump and thus likely to
1464 * be the right thread one wants to have selected in the debugger.
1465 */
1466 thr = td;
1467 while (thr != NULL) {
1468 size += register_note(list, NT_PRSTATUS,
1469 __elfN(note_prstatus), thr);
1470 size += register_note(list, NT_FPREGSET,
1471 __elfN(note_fpregset), thr);
1472 size += register_note(list, NT_THRMISC,
1473 __elfN(note_thrmisc), thr);
1474 size += register_note(list, -1,
1475 __elfN(note_threadmd), thr);
1476
1477 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1478 TAILQ_NEXT(thr, td_plist);
1479 if (thr == td)
1480 thr = TAILQ_NEXT(thr, td_plist);
1481 }
1482
1483 size += register_note(list, NT_PROCSTAT_PROC,
1484 __elfN(note_procstat_proc), p);
1485 size += register_note(list, NT_PROCSTAT_FILES,
1486 note_procstat_files, p);
1487 size += register_note(list, NT_PROCSTAT_VMMAP,
1488 note_procstat_vmmap, p);
1489 size += register_note(list, NT_PROCSTAT_GROUPS,
1490 note_procstat_groups, p);
1491 size += register_note(list, NT_PROCSTAT_UMASK,
1492 note_procstat_umask, p);
1493 size += register_note(list, NT_PROCSTAT_RLIMIT,
1494 note_procstat_rlimit, p);
1495 size += register_note(list, NT_PROCSTAT_OSREL,
1496 note_procstat_osrel, p);
1497 size += register_note(list, NT_PROCSTAT_PSSTRINGS,
1498 __elfN(note_procstat_psstrings), p);
1499 size += register_note(list, NT_PROCSTAT_AUXV,
1500 __elfN(note_procstat_auxv), p);
1501
1502 *sizep = size;
1503 }
1504
1505 static void
__elfN(puthdr)1506 __elfN(puthdr)(struct thread *td, void *hdr, size_t hdrsize, int numsegs,
1507 size_t notesz)
1508 {
1509 Elf_Ehdr *ehdr;
1510 Elf_Phdr *phdr;
1511 struct phdr_closure phc;
1512
1513 ehdr = (Elf_Ehdr *)hdr;
1514 phdr = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr));
1515
1516 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1517 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1518 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1519 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1520 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1521 ehdr->e_ident[EI_DATA] = ELF_DATA;
1522 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1523 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1524 ehdr->e_ident[EI_ABIVERSION] = 0;
1525 ehdr->e_ident[EI_PAD] = 0;
1526 ehdr->e_type = ET_CORE;
1527 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1528 ehdr->e_machine = ELF_ARCH32;
1529 #else
1530 ehdr->e_machine = ELF_ARCH;
1531 #endif
1532 ehdr->e_version = EV_CURRENT;
1533 ehdr->e_entry = 0;
1534 ehdr->e_phoff = sizeof(Elf_Ehdr);
1535 ehdr->e_flags = 0;
1536 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1537 ehdr->e_phentsize = sizeof(Elf_Phdr);
1538 ehdr->e_phnum = numsegs + 1;
1539 ehdr->e_shentsize = sizeof(Elf_Shdr);
1540 ehdr->e_shnum = 0;
1541 ehdr->e_shstrndx = SHN_UNDEF;
1542
1543 /*
1544 * Fill in the program header entries.
1545 */
1546
1547 /* The note segement. */
1548 phdr->p_type = PT_NOTE;
1549 phdr->p_offset = hdrsize;
1550 phdr->p_vaddr = 0;
1551 phdr->p_paddr = 0;
1552 phdr->p_filesz = notesz;
1553 phdr->p_memsz = 0;
1554 phdr->p_flags = PF_R;
1555 phdr->p_align = ELF_NOTE_ROUNDSIZE;
1556 phdr++;
1557
1558 /* All the writable segments from the program. */
1559 phc.phdr = phdr;
1560 phc.offset = round_page(hdrsize + notesz);
1561 each_writable_segment(td, cb_put_phdr, &phc);
1562 }
1563
1564 static size_t
register_note(struct note_info_list * list,int type,outfunc_t out,void * arg)1565 register_note(struct note_info_list *list, int type, outfunc_t out, void *arg)
1566 {
1567 struct note_info *ninfo;
1568 size_t size, notesize;
1569
1570 size = 0;
1571 out(arg, NULL, &size);
1572 ninfo = malloc(sizeof(*ninfo), M_TEMP, M_ZERO | M_WAITOK);
1573 ninfo->type = type;
1574 ninfo->outfunc = out;
1575 ninfo->outarg = arg;
1576 ninfo->outsize = size;
1577 TAILQ_INSERT_TAIL(list, ninfo, link);
1578
1579 if (type == -1)
1580 return (size);
1581
1582 notesize = sizeof(Elf_Note) + /* note header */
1583 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1584 /* note name */
1585 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1586
1587 return (notesize);
1588 }
1589
1590 static size_t
append_note_data(const void * src,void * dst,size_t len)1591 append_note_data(const void *src, void *dst, size_t len)
1592 {
1593 size_t padded_len;
1594
1595 padded_len = roundup2(len, ELF_NOTE_ROUNDSIZE);
1596 if (dst != NULL) {
1597 bcopy(src, dst, len);
1598 bzero((char *)dst + len, padded_len - len);
1599 }
1600 return (padded_len);
1601 }
1602
1603 size_t
__elfN(populate_note)1604 __elfN(populate_note)(int type, void *src, void *dst, size_t size, void **descp)
1605 {
1606 Elf_Note *note;
1607 char *buf;
1608 size_t notesize;
1609
1610 buf = dst;
1611 if (buf != NULL) {
1612 note = (Elf_Note *)buf;
1613 note->n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1614 note->n_descsz = size;
1615 note->n_type = type;
1616 buf += sizeof(*note);
1617 buf += append_note_data(FREEBSD_ABI_VENDOR, buf,
1618 sizeof(FREEBSD_ABI_VENDOR));
1619 append_note_data(src, buf, size);
1620 if (descp != NULL)
1621 *descp = buf;
1622 }
1623
1624 notesize = sizeof(Elf_Note) + /* note header */
1625 roundup2(sizeof(FREEBSD_ABI_VENDOR), ELF_NOTE_ROUNDSIZE) +
1626 /* note name */
1627 roundup2(size, ELF_NOTE_ROUNDSIZE); /* note description */
1628
1629 return (notesize);
1630 }
1631
1632 static void
__elfN(putnote)1633 __elfN(putnote)(struct note_info *ninfo, struct sbuf *sb)
1634 {
1635 Elf_Note note;
1636 ssize_t old_len;
1637
1638 if (ninfo->type == -1) {
1639 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1640 return;
1641 }
1642
1643 note.n_namesz = sizeof(FREEBSD_ABI_VENDOR);
1644 note.n_descsz = ninfo->outsize;
1645 note.n_type = ninfo->type;
1646
1647 sbuf_bcat(sb, ¬e, sizeof(note));
1648 sbuf_start_section(sb, &old_len);
1649 sbuf_bcat(sb, FREEBSD_ABI_VENDOR, sizeof(FREEBSD_ABI_VENDOR));
1650 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1651 if (note.n_descsz == 0)
1652 return;
1653 sbuf_start_section(sb, &old_len);
1654 ninfo->outfunc(ninfo->outarg, sb, &ninfo->outsize);
1655 sbuf_end_section(sb, old_len, ELF_NOTE_ROUNDSIZE, 0);
1656 }
1657
1658 /*
1659 * Miscellaneous note out functions.
1660 */
1661
1662 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1663 #include <compat/freebsd32/freebsd32.h>
1664
1665 typedef struct prstatus32 elf_prstatus_t;
1666 typedef struct prpsinfo32 elf_prpsinfo_t;
1667 typedef struct fpreg32 elf_prfpregset_t;
1668 typedef struct fpreg32 elf_fpregset_t;
1669 typedef struct reg32 elf_gregset_t;
1670 typedef struct thrmisc32 elf_thrmisc_t;
1671 #define ELF_KERN_PROC_MASK KERN_PROC_MASK32
1672 typedef struct kinfo_proc32 elf_kinfo_proc_t;
1673 typedef uint32_t elf_ps_strings_t;
1674 #else
1675 typedef prstatus_t elf_prstatus_t;
1676 typedef prpsinfo_t elf_prpsinfo_t;
1677 typedef prfpregset_t elf_prfpregset_t;
1678 typedef prfpregset_t elf_fpregset_t;
1679 typedef gregset_t elf_gregset_t;
1680 typedef thrmisc_t elf_thrmisc_t;
1681 #define ELF_KERN_PROC_MASK 0
1682 typedef struct kinfo_proc elf_kinfo_proc_t;
1683 typedef vm_offset_t elf_ps_strings_t;
1684 #endif
1685
1686 static void
__elfN(note_prpsinfo)1687 __elfN(note_prpsinfo)(void *arg, struct sbuf *sb, size_t *sizep)
1688 {
1689 struct proc *p;
1690 elf_prpsinfo_t *psinfo;
1691
1692 p = (struct proc *)arg;
1693 if (sb != NULL) {
1694 KASSERT(*sizep == sizeof(*psinfo), ("invalid size"));
1695 psinfo = malloc(sizeof(*psinfo), M_TEMP, M_ZERO | M_WAITOK);
1696 psinfo->pr_version = PRPSINFO_VERSION;
1697 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1698 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1699 /*
1700 * XXX - We don't fill in the command line arguments properly
1701 * yet.
1702 */
1703 strlcpy(psinfo->pr_psargs, p->p_comm,
1704 sizeof(psinfo->pr_psargs));
1705
1706 sbuf_bcat(sb, psinfo, sizeof(*psinfo));
1707 free(psinfo, M_TEMP);
1708 }
1709 *sizep = sizeof(*psinfo);
1710 }
1711
1712 static void
__elfN(note_prstatus)1713 __elfN(note_prstatus)(void *arg, struct sbuf *sb, size_t *sizep)
1714 {
1715 struct thread *td;
1716 elf_prstatus_t *status;
1717
1718 td = (struct thread *)arg;
1719 if (sb != NULL) {
1720 KASSERT(*sizep == sizeof(*status), ("invalid size"));
1721 status = malloc(sizeof(*status), M_TEMP, M_ZERO | M_WAITOK);
1722 status->pr_version = PRSTATUS_VERSION;
1723 status->pr_statussz = sizeof(elf_prstatus_t);
1724 status->pr_gregsetsz = sizeof(elf_gregset_t);
1725 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1726 status->pr_osreldate = osreldate;
1727 status->pr_cursig = td->td_proc->p_sig;
1728 status->pr_pid = td->td_tid;
1729 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1730 fill_regs32(td, &status->pr_reg);
1731 #else
1732 fill_regs(td, &status->pr_reg);
1733 #endif
1734 sbuf_bcat(sb, status, sizeof(*status));
1735 free(status, M_TEMP);
1736 }
1737 *sizep = sizeof(*status);
1738 }
1739
1740 static void
__elfN(note_fpregset)1741 __elfN(note_fpregset)(void *arg, struct sbuf *sb, size_t *sizep)
1742 {
1743 struct thread *td;
1744 elf_prfpregset_t *fpregset;
1745
1746 td = (struct thread *)arg;
1747 if (sb != NULL) {
1748 KASSERT(*sizep == sizeof(*fpregset), ("invalid size"));
1749 fpregset = malloc(sizeof(*fpregset), M_TEMP, M_ZERO | M_WAITOK);
1750 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1751 fill_fpregs32(td, fpregset);
1752 #else
1753 fill_fpregs(td, fpregset);
1754 #endif
1755 sbuf_bcat(sb, fpregset, sizeof(*fpregset));
1756 free(fpregset, M_TEMP);
1757 }
1758 *sizep = sizeof(*fpregset);
1759 }
1760
1761 static void
__elfN(note_thrmisc)1762 __elfN(note_thrmisc)(void *arg, struct sbuf *sb, size_t *sizep)
1763 {
1764 struct thread *td;
1765 elf_thrmisc_t thrmisc;
1766
1767 td = (struct thread *)arg;
1768 if (sb != NULL) {
1769 KASSERT(*sizep == sizeof(thrmisc), ("invalid size"));
1770 bzero(&thrmisc._pad, sizeof(thrmisc._pad));
1771 strcpy(thrmisc.pr_tname, td->td_name);
1772 sbuf_bcat(sb, &thrmisc, sizeof(thrmisc));
1773 }
1774 *sizep = sizeof(thrmisc);
1775 }
1776
1777 /*
1778 * Allow for MD specific notes, as well as any MD
1779 * specific preparations for writing MI notes.
1780 */
1781 static void
__elfN(note_threadmd)1782 __elfN(note_threadmd)(void *arg, struct sbuf *sb, size_t *sizep)
1783 {
1784 struct thread *td;
1785 void *buf;
1786 size_t size;
1787
1788 td = (struct thread *)arg;
1789 size = *sizep;
1790 if (size != 0 && sb != NULL)
1791 buf = malloc(size, M_TEMP, M_ZERO | M_WAITOK);
1792 else
1793 buf = NULL;
1794 size = 0;
1795 __elfN(dump_thread)(td, buf, &size);
1796 KASSERT(sb == NULL || *sizep == size, ("invalid size"));
1797 if (size != 0 && sb != NULL)
1798 sbuf_bcat(sb, buf, size);
1799 free(buf, M_TEMP);
1800 *sizep = size;
1801 }
1802
1803 #ifdef KINFO_PROC_SIZE
1804 CTASSERT(sizeof(struct kinfo_proc) == KINFO_PROC_SIZE);
1805 #endif
1806
1807 static void
__elfN(note_procstat_proc)1808 __elfN(note_procstat_proc)(void *arg, struct sbuf *sb, size_t *sizep)
1809 {
1810 struct proc *p;
1811 size_t size;
1812 int structsize;
1813
1814 p = (struct proc *)arg;
1815 size = sizeof(structsize) + p->p_numthreads *
1816 sizeof(elf_kinfo_proc_t);
1817
1818 if (sb != NULL) {
1819 KASSERT(*sizep == size, ("invalid size"));
1820 structsize = sizeof(elf_kinfo_proc_t);
1821 sbuf_bcat(sb, &structsize, sizeof(structsize));
1822 PROC_LOCK(p);
1823 kern_proc_out(p, sb, ELF_KERN_PROC_MASK);
1824 }
1825 *sizep = size;
1826 }
1827
1828 #ifdef KINFO_FILE_SIZE
1829 CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE);
1830 #endif
1831
1832 static void
note_procstat_files(void * arg,struct sbuf * sb,size_t * sizep)1833 note_procstat_files(void *arg, struct sbuf *sb, size_t *sizep)
1834 {
1835 struct proc *p;
1836 size_t size;
1837 int structsize;
1838
1839 p = (struct proc *)arg;
1840 if (sb == NULL) {
1841 size = 0;
1842 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1843 sbuf_set_drain(sb, sbuf_drain_count, &size);
1844 sbuf_bcat(sb, &structsize, sizeof(structsize));
1845 PROC_LOCK(p);
1846 kern_proc_filedesc_out(p, sb, -1);
1847 sbuf_finish(sb);
1848 sbuf_delete(sb);
1849 *sizep = size;
1850 } else {
1851 structsize = sizeof(struct kinfo_file);
1852 sbuf_bcat(sb, &structsize, sizeof(structsize));
1853 PROC_LOCK(p);
1854 kern_proc_filedesc_out(p, sb, -1);
1855 }
1856 }
1857
1858 #ifdef KINFO_VMENTRY_SIZE
1859 CTASSERT(sizeof(struct kinfo_vmentry) == KINFO_VMENTRY_SIZE);
1860 #endif
1861
1862 static void
note_procstat_vmmap(void * arg,struct sbuf * sb,size_t * sizep)1863 note_procstat_vmmap(void *arg, struct sbuf *sb, size_t *sizep)
1864 {
1865 struct proc *p;
1866 size_t size;
1867 int structsize;
1868
1869 p = (struct proc *)arg;
1870 if (sb == NULL) {
1871 size = 0;
1872 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
1873 sbuf_set_drain(sb, sbuf_drain_count, &size);
1874 sbuf_bcat(sb, &structsize, sizeof(structsize));
1875 PROC_LOCK(p);
1876 kern_proc_vmmap_out(p, sb);
1877 sbuf_finish(sb);
1878 sbuf_delete(sb);
1879 *sizep = size;
1880 } else {
1881 structsize = sizeof(struct kinfo_vmentry);
1882 sbuf_bcat(sb, &structsize, sizeof(structsize));
1883 PROC_LOCK(p);
1884 kern_proc_vmmap_out(p, sb);
1885 }
1886 }
1887
1888 static void
note_procstat_groups(void * arg,struct sbuf * sb,size_t * sizep)1889 note_procstat_groups(void *arg, struct sbuf *sb, size_t *sizep)
1890 {
1891 struct proc *p;
1892 size_t size;
1893 int structsize;
1894
1895 p = (struct proc *)arg;
1896 size = sizeof(structsize) + p->p_ucred->cr_ngroups * sizeof(gid_t);
1897 if (sb != NULL) {
1898 KASSERT(*sizep == size, ("invalid size"));
1899 structsize = sizeof(gid_t);
1900 sbuf_bcat(sb, &structsize, sizeof(structsize));
1901 sbuf_bcat(sb, p->p_ucred->cr_groups, p->p_ucred->cr_ngroups *
1902 sizeof(gid_t));
1903 }
1904 *sizep = size;
1905 }
1906
1907 static void
note_procstat_umask(void * arg,struct sbuf * sb,size_t * sizep)1908 note_procstat_umask(void *arg, struct sbuf *sb, size_t *sizep)
1909 {
1910 struct proc *p;
1911 size_t size;
1912 int structsize;
1913
1914 p = (struct proc *)arg;
1915 size = sizeof(structsize) + sizeof(p->p_fd->fd_cmask);
1916 if (sb != NULL) {
1917 KASSERT(*sizep == size, ("invalid size"));
1918 structsize = sizeof(p->p_fd->fd_cmask);
1919 sbuf_bcat(sb, &structsize, sizeof(structsize));
1920 sbuf_bcat(sb, &p->p_fd->fd_cmask, sizeof(p->p_fd->fd_cmask));
1921 }
1922 *sizep = size;
1923 }
1924
1925 static void
note_procstat_rlimit(void * arg,struct sbuf * sb,size_t * sizep)1926 note_procstat_rlimit(void *arg, struct sbuf *sb, size_t *sizep)
1927 {
1928 struct proc *p;
1929 struct rlimit rlim[RLIM_NLIMITS];
1930 size_t size;
1931 int structsize, i;
1932
1933 p = (struct proc *)arg;
1934 size = sizeof(structsize) + sizeof(rlim);
1935 if (sb != NULL) {
1936 KASSERT(*sizep == size, ("invalid size"));
1937 structsize = sizeof(rlim);
1938 sbuf_bcat(sb, &structsize, sizeof(structsize));
1939 PROC_LOCK(p);
1940 for (i = 0; i < RLIM_NLIMITS; i++)
1941 lim_rlimit(p, i, &rlim[i]);
1942 PROC_UNLOCK(p);
1943 sbuf_bcat(sb, rlim, sizeof(rlim));
1944 }
1945 *sizep = size;
1946 }
1947
1948 static void
note_procstat_osrel(void * arg,struct sbuf * sb,size_t * sizep)1949 note_procstat_osrel(void *arg, struct sbuf *sb, size_t *sizep)
1950 {
1951 struct proc *p;
1952 size_t size;
1953 int structsize;
1954
1955 p = (struct proc *)arg;
1956 size = sizeof(structsize) + sizeof(p->p_osrel);
1957 if (sb != NULL) {
1958 KASSERT(*sizep == size, ("invalid size"));
1959 structsize = sizeof(p->p_osrel);
1960 sbuf_bcat(sb, &structsize, sizeof(structsize));
1961 sbuf_bcat(sb, &p->p_osrel, sizeof(p->p_osrel));
1962 }
1963 *sizep = size;
1964 }
1965
1966 static void
__elfN(note_procstat_psstrings)1967 __elfN(note_procstat_psstrings)(void *arg, struct sbuf *sb, size_t *sizep)
1968 {
1969 struct proc *p;
1970 elf_ps_strings_t ps_strings;
1971 size_t size;
1972 int structsize;
1973
1974 p = (struct proc *)arg;
1975 size = sizeof(structsize) + sizeof(ps_strings);
1976 if (sb != NULL) {
1977 KASSERT(*sizep == size, ("invalid size"));
1978 structsize = sizeof(ps_strings);
1979 #if defined(COMPAT_FREEBSD32) && __ELF_WORD_SIZE == 32
1980 ps_strings = PTROUT(p->p_sysent->sv_psstrings);
1981 #else
1982 ps_strings = p->p_sysent->sv_psstrings;
1983 #endif
1984 sbuf_bcat(sb, &structsize, sizeof(structsize));
1985 sbuf_bcat(sb, &ps_strings, sizeof(ps_strings));
1986 }
1987 *sizep = size;
1988 }
1989
1990 static void
__elfN(note_procstat_auxv)1991 __elfN(note_procstat_auxv)(void *arg, struct sbuf *sb, size_t *sizep)
1992 {
1993 struct proc *p;
1994 size_t size;
1995 int structsize;
1996
1997 p = (struct proc *)arg;
1998 if (sb == NULL) {
1999 size = 0;
2000 sb = sbuf_new(NULL, NULL, 128, SBUF_FIXEDLEN);
2001 sbuf_set_drain(sb, sbuf_drain_count, &size);
2002 sbuf_bcat(sb, &structsize, sizeof(structsize));
2003 PHOLD(p);
2004 proc_getauxv(curthread, p, sb);
2005 PRELE(p);
2006 sbuf_finish(sb);
2007 sbuf_delete(sb);
2008 *sizep = size;
2009 } else {
2010 structsize = sizeof(Elf_Auxinfo);
2011 sbuf_bcat(sb, &structsize, sizeof(structsize));
2012 PHOLD(p);
2013 proc_getauxv(curthread, p, sb);
2014 PRELE(p);
2015 }
2016 }
2017
2018 static boolean_t
__elfN(parse_notes)2019 __elfN(parse_notes)(struct image_params *imgp, Elf_Brandnote *checknote,
2020 int32_t *osrel, const Elf_Phdr *pnote)
2021 {
2022 const Elf_Note *note, *note0, *note_end;
2023 const char *note_name;
2024 int i;
2025
2026 if (pnote == NULL || pnote->p_offset > PAGE_SIZE ||
2027 pnote->p_filesz > PAGE_SIZE - pnote->p_offset)
2028 return (FALSE);
2029
2030 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
2031 note_end = (const Elf_Note *)(imgp->image_header +
2032 pnote->p_offset + pnote->p_filesz);
2033 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
2034 if (!aligned(note, Elf32_Addr) || (const char *)note_end -
2035 (const char *)note < sizeof(Elf_Note))
2036 return (FALSE);
2037 if (note->n_namesz != checknote->hdr.n_namesz ||
2038 note->n_descsz != checknote->hdr.n_descsz ||
2039 note->n_type != checknote->hdr.n_type)
2040 goto nextnote;
2041 note_name = (const char *)(note + 1);
2042 if (note_name + checknote->hdr.n_namesz >=
2043 (const char *)note_end || strncmp(checknote->vendor,
2044 note_name, checknote->hdr.n_namesz) != 0)
2045 goto nextnote;
2046
2047 /*
2048 * Fetch the osreldate for binary
2049 * from the ELF OSABI-note if necessary.
2050 */
2051 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
2052 checknote->trans_osrel != NULL)
2053 return (checknote->trans_osrel(note, osrel));
2054 return (TRUE);
2055
2056 nextnote:
2057 note = (const Elf_Note *)((const char *)(note + 1) +
2058 roundup2(note->n_namesz, ELF_NOTE_ROUNDSIZE) +
2059 roundup2(note->n_descsz, ELF_NOTE_ROUNDSIZE));
2060 }
2061
2062 return (FALSE);
2063 }
2064
2065 /*
2066 * Try to find the appropriate ABI-note section for checknote,
2067 * fetch the osreldate for binary from the ELF OSABI-note. Only the
2068 * first page of the image is searched, the same as for headers.
2069 */
2070 static boolean_t
__elfN(check_note)2071 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
2072 int32_t *osrel)
2073 {
2074 const Elf_Phdr *phdr;
2075 const Elf_Ehdr *hdr;
2076 int i;
2077
2078 hdr = (const Elf_Ehdr *)imgp->image_header;
2079 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
2080
2081 for (i = 0; i < hdr->e_phnum; i++) {
2082 if (phdr[i].p_type == PT_NOTE &&
2083 __elfN(parse_notes)(imgp, checknote, osrel, &phdr[i]))
2084 return (TRUE);
2085 }
2086 return (FALSE);
2087
2088 }
2089
2090 /*
2091 * Tell kern_execve.c about it, with a little help from the linker.
2092 */
2093 static struct execsw __elfN(execsw) = {
2094 __CONCAT(exec_, __elfN(imgact)),
2095 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
2096 };
2097 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
2098
2099 #ifdef COMPRESS_USER_CORES
2100 /*
2101 * Compress and write out a core segment for a user process.
2102 *
2103 * 'inbuf' is the starting address of a VM segment in the process' address
2104 * space that is to be compressed and written out to the core file. 'dest_buf'
2105 * is a buffer in the kernel's address space. The segment is copied from
2106 * 'inbuf' to 'dest_buf' first before being processed by the compression
2107 * routine gzwrite(). This copying is necessary because the content of the VM
2108 * segment may change between the compression pass and the crc-computation pass
2109 * in gzwrite(). This is because realtime threads may preempt the UNIX kernel.
2110 *
2111 * If inbuf is NULL it is assumed that data is already copied to 'dest_buf'.
2112 */
2113 static int
compress_core(gzFile file,char * inbuf,char * dest_buf,unsigned int len,struct thread * td)2114 compress_core (gzFile file, char *inbuf, char *dest_buf, unsigned int len,
2115 struct thread *td)
2116 {
2117 int len_compressed;
2118 int error = 0;
2119 unsigned int chunk_len;
2120
2121 while (len) {
2122 if (inbuf != NULL) {
2123 chunk_len = (len > CORE_BUF_SIZE) ? CORE_BUF_SIZE : len;
2124 copyin(inbuf, dest_buf, chunk_len);
2125 inbuf += chunk_len;
2126 } else {
2127 chunk_len = len;
2128 }
2129 len_compressed = gzwrite(file, dest_buf, chunk_len);
2130
2131 EVENTHANDLER_INVOKE(app_coredump_progress, td, len_compressed);
2132
2133 if ((unsigned int)len_compressed != chunk_len) {
2134 log(LOG_WARNING,
2135 "compress_core: length mismatch (0x%x returned, "
2136 "0x%x expected)\n", len_compressed, chunk_len);
2137 EVENTHANDLER_INVOKE(app_coredump_error, td,
2138 "compress_core: length mismatch %x -> %x",
2139 chunk_len, len_compressed);
2140 error = EFAULT;
2141 break;
2142 }
2143 len -= chunk_len;
2144 maybe_yield();
2145 }
2146
2147 return (error);
2148 }
2149 #endif /* COMPRESS_USER_CORES */
2150
2151 static vm_prot_t
__elfN(trans_prot)2152 __elfN(trans_prot)(Elf_Word flags)
2153 {
2154 vm_prot_t prot;
2155
2156 prot = 0;
2157 if (flags & PF_X)
2158 prot |= VM_PROT_EXECUTE;
2159 if (flags & PF_W)
2160 prot |= VM_PROT_WRITE;
2161 if (flags & PF_R)
2162 prot |= VM_PROT_READ;
2163 #if __ELF_WORD_SIZE == 32
2164 #if defined(__amd64__) || defined(__ia64__)
2165 if (i386_read_exec && (flags & PF_R))
2166 prot |= VM_PROT_EXECUTE;
2167 #endif
2168 #endif
2169 return (prot);
2170 }
2171
2172 static Elf_Word
__elfN(untrans_prot)2173 __elfN(untrans_prot)(vm_prot_t prot)
2174 {
2175 Elf_Word flags;
2176
2177 flags = 0;
2178 if (prot & VM_PROT_EXECUTE)
2179 flags |= PF_X;
2180 if (prot & VM_PROT_READ)
2181 flags |= PF_R;
2182 if (prot & VM_PROT_WRITE)
2183 flags |= PF_W;
2184 return (flags);
2185 }
2186