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("$FreeBSD$");
33
34 #include "opt_compat.h"
35
36 #include <sys/param.h>
37 #include <sys/exec.h>
38 #include <sys/fcntl.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mutex.h>
46 #include <sys/mman.h>
47 #include <sys/namei.h>
48 #include <sys/pioctl.h>
49 #include <sys/proc.h>
50 #include <sys/procfs.h>
51 #include <sys/resourcevar.h>
52 #include <sys/sf_buf.h>
53 #include <sys/systm.h>
54 #include <sys/signalvar.h>
55 #include <sys/stat.h>
56 #include <sys/sx.h>
57 #include <sys/syscall.h>
58 #include <sys/sysctl.h>
59 #include <sys/sysent.h>
60 #include <sys/vnode.h>
61
62 #include <vm/vm.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_param.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69
70 #include <machine/elf.h>
71 #include <machine/md_var.h>
72
73 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
74 #include <machine/fpu.h>
75 #include <compat/ia32/ia32_reg.h>
76 #endif
77
78 #define OLD_EI_BRAND 8
79
80 static int __elfN(check_header)(const Elf_Ehdr *hdr);
81 static Elf_Brandinfo *__elfN(get_brandinfo)(struct image_params *imgp,
82 const char *interp, int32_t *osrel);
83 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
84 u_long *entry, size_t pagesize);
85 static int __elfN(load_section)(struct vmspace *vmspace, vm_object_t object,
86 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
87 vm_prot_t prot, size_t pagesize);
88 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
89 static boolean_t __elfN(freebsd_trans_osrel)(const Elf_Note *note,
90 int32_t *osrel);
91 static boolean_t kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel);
92 static boolean_t __elfN(check_note)(struct image_params *imgp,
93 Elf_Brandnote *checknote, int32_t *osrel);
94
95 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
96 "");
97
98 int __elfN(fallback_brand) = -1;
99 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
100 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
101 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
102 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
103 &__elfN(fallback_brand));
104
105 static int elf_trace = 0;
106 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
107
108 static int elf_legacy_coredump = 0;
109 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
110 &elf_legacy_coredump, 0, "");
111
112 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
113
114 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
115 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
116 #define aligned(a, t) (trunc_page_ps((u_long)(a), sizeof(t)) == (u_long)(a))
117
118 static const char FREEBSD_ABI_VENDOR[] = "FreeBSD";
119
120 Elf_Brandnote __elfN(freebsd_brandnote) = {
121 .hdr.n_namesz = sizeof(FREEBSD_ABI_VENDOR),
122 .hdr.n_descsz = sizeof(int32_t),
123 .hdr.n_type = 1,
124 .vendor = FREEBSD_ABI_VENDOR,
125 .flags = BN_TRANSLATE_OSREL,
126 .trans_osrel = __elfN(freebsd_trans_osrel)
127 };
128
129 static boolean_t
130 __elfN(freebsd_trans_osrel)(const Elf_Note *note, int32_t *osrel)
131 {
132 uintptr_t p;
133
134 p = (uintptr_t)(note + 1);
135 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
136 *osrel = *(const int32_t *)(p);
137
138 return (TRUE);
139 }
140
141 static const char GNU_ABI_VENDOR[] = "GNU";
142 static int GNU_KFREEBSD_ABI_DESC = 3;
143
144 Elf_Brandnote __elfN(kfreebsd_brandnote) = {
145 .hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
146 .hdr.n_descsz = 16, /* XXX at least 16 */
147 .hdr.n_type = 1,
148 .vendor = GNU_ABI_VENDOR,
149 .flags = BN_TRANSLATE_OSREL,
150 .trans_osrel = kfreebsd_trans_osrel
151 };
152
153 static boolean_t
154 kfreebsd_trans_osrel(const Elf_Note *note, int32_t *osrel)
155 {
156 const Elf32_Word *desc;
157 uintptr_t p;
158
159 p = (uintptr_t)(note + 1);
160 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
161
162 desc = (const Elf32_Word *)p;
163 if (desc[0] != GNU_KFREEBSD_ABI_DESC)
164 return (FALSE);
165
166 /*
167 * Debian GNU/kFreeBSD embed the earliest compatible kernel version
168 * (__FreeBSD_version: <major><two digit minor>Rxx) in the LSB way.
169 */
170 *osrel = desc[1] * 100000 + desc[2] * 1000 + desc[3];
171
172 return (TRUE);
173 }
174
175 int
176 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
177 {
178 int i;
179
180 for (i = 0; i < MAX_BRANDS; i++) {
181 if (elf_brand_list[i] == NULL) {
182 elf_brand_list[i] = entry;
183 break;
184 }
185 }
186 if (i == MAX_BRANDS)
187 return (-1);
188 return (0);
189 }
190
191 int
192 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
193 {
194 int i;
195
196 for (i = 0; i < MAX_BRANDS; i++) {
197 if (elf_brand_list[i] == entry) {
198 elf_brand_list[i] = NULL;
199 break;
200 }
201 }
202 if (i == MAX_BRANDS)
203 return (-1);
204 return (0);
205 }
206
207 int
208 __elfN(brand_inuse)(Elf_Brandinfo *entry)
209 {
210 struct proc *p;
211 int rval = FALSE;
212
213 sx_slock(&allproc_lock);
214 FOREACH_PROC_IN_SYSTEM(p) {
215 if (p->p_sysent == entry->sysvec) {
216 rval = TRUE;
217 break;
218 }
219 }
220 sx_sunlock(&allproc_lock);
221
222 return (rval);
223 }
224
225 static Elf_Brandinfo *
226 __elfN(get_brandinfo)(struct image_params *imgp, const char *interp,
227 int32_t *osrel)
228 {
229 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
230 Elf_Brandinfo *bi;
231 boolean_t ret;
232 int i;
233
234 /*
235 * We support four types of branding -- (1) the ELF EI_OSABI field
236 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
237 * branding w/in the ELF header, (3) path of the `interp_path'
238 * field, and (4) the ".note.ABI-tag" ELF section.
239 */
240
241 /* Look for an ".note.ABI-tag" ELF section */
242 for (i = 0; i < MAX_BRANDS; i++) {
243 bi = elf_brand_list[i];
244 if (bi == NULL)
245 continue;
246 if (hdr->e_machine == bi->machine && (bi->flags &
247 (BI_BRAND_NOTE|BI_BRAND_NOTE_MANDATORY)) != 0) {
248 ret = __elfN(check_note)(imgp, bi->brand_note, osrel);
249 if (ret)
250 return (bi);
251 }
252 }
253
254 /* If the executable has a brand, search for it in the brand list. */
255 for (i = 0; i < MAX_BRANDS; i++) {
256 bi = elf_brand_list[i];
257 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
258 continue;
259 if (hdr->e_machine == bi->machine &&
260 (hdr->e_ident[EI_OSABI] == bi->brand ||
261 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
262 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
263 return (bi);
264 }
265
266 /* Lacking a known brand, search for a recognized interpreter. */
267 if (interp != NULL) {
268 for (i = 0; i < MAX_BRANDS; i++) {
269 bi = elf_brand_list[i];
270 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
271 continue;
272 if (hdr->e_machine == bi->machine &&
273 strcmp(interp, bi->interp_path) == 0)
274 return (bi);
275 }
276 }
277
278 /* Lacking a recognized interpreter, try the default brand */
279 for (i = 0; i < MAX_BRANDS; i++) {
280 bi = elf_brand_list[i];
281 if (bi == NULL || bi->flags & BI_BRAND_NOTE_MANDATORY)
282 continue;
283 if (hdr->e_machine == bi->machine &&
284 __elfN(fallback_brand) == bi->brand)
285 return (bi);
286 }
287 return (NULL);
288 }
289
290 static int
291 __elfN(check_header)(const Elf_Ehdr *hdr)
292 {
293 Elf_Brandinfo *bi;
294 int i;
295
296 if (!IS_ELF(*hdr) ||
297 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
298 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
299 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
300 hdr->e_phentsize != sizeof(Elf_Phdr) ||
301 hdr->e_version != ELF_TARG_VER)
302 return (ENOEXEC);
303
304 /*
305 * Make sure we have at least one brand for this machine.
306 */
307
308 for (i = 0; i < MAX_BRANDS; i++) {
309 bi = elf_brand_list[i];
310 if (bi != NULL && bi->machine == hdr->e_machine)
311 break;
312 }
313 if (i == MAX_BRANDS)
314 return (ENOEXEC);
315
316 return (0);
317 }
318
319 static int
320 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
321 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
322 {
323 struct sf_buf *sf;
324 int error;
325 vm_offset_t off;
326
327 /*
328 * Create the page if it doesn't exist yet. Ignore errors.
329 */
330 vm_map_lock(map);
331 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
332 VM_PROT_ALL, VM_PROT_ALL, 0);
333 vm_map_unlock(map);
334
335 /*
336 * Find the page from the underlying object.
337 */
338 if (object) {
339 sf = vm_imgact_map_page(object, offset);
340 if (sf == NULL)
341 return (KERN_FAILURE);
342 off = offset - trunc_page(offset);
343 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
344 end - start);
345 vm_imgact_unmap_page(sf);
346 if (error) {
347 return (KERN_FAILURE);
348 }
349 }
350
351 return (KERN_SUCCESS);
352 }
353
354 static int
355 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
356 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
357 {
358 struct sf_buf *sf;
359 vm_offset_t off;
360 vm_size_t sz;
361 int error, rv;
362
363 if (start != trunc_page(start)) {
364 rv = __elfN(map_partial)(map, object, offset, start,
365 round_page(start), prot);
366 if (rv)
367 return (rv);
368 offset += round_page(start) - start;
369 start = round_page(start);
370 }
371 if (end != round_page(end)) {
372 rv = __elfN(map_partial)(map, object, offset +
373 trunc_page(end) - start, trunc_page(end), end, prot);
374 if (rv)
375 return (rv);
376 end = trunc_page(end);
377 }
378 if (end > start) {
379 if (offset & PAGE_MASK) {
380 /*
381 * The mapping is not page aligned. This means we have
382 * to copy the data. Sigh.
383 */
384 rv = vm_map_find(map, NULL, 0, &start, end - start,
385 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
386 if (rv)
387 return (rv);
388 if (object == NULL)
389 return (KERN_SUCCESS);
390 for (; start < end; start += sz) {
391 sf = vm_imgact_map_page(object, offset);
392 if (sf == NULL)
393 return (KERN_FAILURE);
394 off = offset - trunc_page(offset);
395 sz = end - start;
396 if (sz > PAGE_SIZE - off)
397 sz = PAGE_SIZE - off;
398 error = copyout((caddr_t)sf_buf_kva(sf) + off,
399 (caddr_t)start, sz);
400 vm_imgact_unmap_page(sf);
401 if (error) {
402 return (KERN_FAILURE);
403 }
404 offset += sz;
405 }
406 rv = KERN_SUCCESS;
407 } else {
408 vm_object_reference(object);
409 vm_map_lock(map);
410 rv = vm_map_insert(map, object, offset, start, end,
411 prot, VM_PROT_ALL, cow);
412 vm_map_unlock(map);
413 if (rv != KERN_SUCCESS)
414 vm_object_deallocate(object);
415 }
416 return (rv);
417 } else {
418 return (KERN_SUCCESS);
419 }
420 }
421
422 static int
423 __elfN(load_section)(struct vmspace *vmspace,
424 vm_object_t object, vm_offset_t offset,
425 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
426 size_t pagesize)
427 {
428 struct sf_buf *sf;
429 size_t map_len;
430 vm_offset_t map_addr;
431 int error, rv, cow;
432 size_t copy_len;
433 vm_offset_t file_addr;
434
435 /*
436 * It's necessary to fail if the filsz + offset taken from the
437 * header is greater than the actual file pager object's size.
438 * If we were to allow this, then the vm_map_find() below would
439 * walk right off the end of the file object and into the ether.
440 *
441 * While I'm here, might as well check for something else that
442 * is invalid: filsz cannot be greater than memsz.
443 */
444 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
445 filsz > memsz) {
446 uprintf("elf_load_section: truncated ELF file\n");
447 return (ENOEXEC);
448 }
449
450 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
451 file_addr = trunc_page_ps(offset, pagesize);
452
453 /*
454 * We have two choices. We can either clear the data in the last page
455 * of an oversized mapping, or we can start the anon mapping a page
456 * early and copy the initialized data into that first page. We
457 * choose the second..
458 */
459 if (memsz > filsz)
460 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
461 else
462 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
463
464 if (map_len != 0) {
465 /* cow flags: don't dump readonly sections in core */
466 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
467 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
468
469 rv = __elfN(map_insert)(&vmspace->vm_map,
470 object,
471 file_addr, /* file offset */
472 map_addr, /* virtual start */
473 map_addr + map_len,/* virtual end */
474 prot,
475 cow);
476 if (rv != KERN_SUCCESS)
477 return (EINVAL);
478
479 /* we can stop now if we've covered it all */
480 if (memsz == filsz) {
481 return (0);
482 }
483 }
484
485
486 /*
487 * We have to get the remaining bit of the file into the first part
488 * of the oversized map segment. This is normally because the .data
489 * segment in the file is extended to provide bss. It's a neat idea
490 * to try and save a page, but it's a pain in the behind to implement.
491 */
492 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
493 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
494 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
495 map_addr;
496
497 /* This had damn well better be true! */
498 if (map_len != 0) {
499 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
500 map_addr + map_len, VM_PROT_ALL, 0);
501 if (rv != KERN_SUCCESS) {
502 return (EINVAL);
503 }
504 }
505
506 if (copy_len != 0) {
507 vm_offset_t off;
508
509 sf = vm_imgact_map_page(object, offset + filsz);
510 if (sf == NULL)
511 return (EIO);
512
513 /* send the page fragment to user space */
514 off = trunc_page_ps(offset + filsz, pagesize) -
515 trunc_page(offset + filsz);
516 error = copyout((caddr_t)sf_buf_kva(sf) + off,
517 (caddr_t)map_addr, copy_len);
518 vm_imgact_unmap_page(sf);
519 if (error) {
520 return (error);
521 }
522 }
523
524 /*
525 * set it to the specified protection.
526 * XXX had better undo the damage from pasting over the cracks here!
527 */
528 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
529 round_page(map_addr + map_len), prot, FALSE);
530
531 return (0);
532 }
533
534 /*
535 * Load the file "file" into memory. It may be either a shared object
536 * or an executable.
537 *
538 * The "addr" reference parameter is in/out. On entry, it specifies
539 * the address where a shared object should be loaded. If the file is
540 * an executable, this value is ignored. On exit, "addr" specifies
541 * where the file was actually loaded.
542 *
543 * The "entry" reference parameter is out only. On exit, it specifies
544 * the entry point for the loaded file.
545 */
546 static int
547 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
548 u_long *entry, size_t pagesize)
549 {
550 struct {
551 struct nameidata nd;
552 struct vattr attr;
553 struct image_params image_params;
554 } *tempdata;
555 const Elf_Ehdr *hdr = NULL;
556 const Elf_Phdr *phdr = NULL;
557 struct nameidata *nd;
558 struct vmspace *vmspace = p->p_vmspace;
559 struct vattr *attr;
560 struct image_params *imgp;
561 vm_prot_t prot;
562 u_long rbase;
563 u_long base_addr = 0;
564 int vfslocked, error, i, numsegs;
565
566 if (curthread->td_proc != p)
567 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
568
569 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
570 nd = &tempdata->nd;
571 attr = &tempdata->attr;
572 imgp = &tempdata->image_params;
573
574 /*
575 * Initialize part of the common data
576 */
577 imgp->proc = p;
578 imgp->attr = attr;
579 imgp->firstpage = NULL;
580 imgp->image_header = NULL;
581 imgp->object = NULL;
582 imgp->execlabel = NULL;
583
584 /* XXXKSE */
585 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
586 curthread);
587 vfslocked = 0;
588 if ((error = namei(nd)) != 0) {
589 nd->ni_vp = NULL;
590 goto fail;
591 }
592 vfslocked = NDHASGIANT(nd);
593 NDFREE(nd, NDF_ONLY_PNBUF);
594 imgp->vp = nd->ni_vp;
595
596 /*
597 * Check permissions, modes, uid, etc on the file, and "open" it.
598 */
599 error = exec_check_permissions(imgp);
600 if (error)
601 goto fail;
602
603 error = exec_map_first_page(imgp);
604 if (error)
605 goto fail;
606
607 /*
608 * Also make certain that the interpreter stays the same, so set
609 * its VV_TEXT flag, too.
610 */
611 nd->ni_vp->v_vflag |= VV_TEXT;
612
613 imgp->object = nd->ni_vp->v_object;
614
615 hdr = (const Elf_Ehdr *)imgp->image_header;
616 if ((error = __elfN(check_header)(hdr)) != 0)
617 goto fail;
618 if (hdr->e_type == ET_DYN)
619 rbase = *addr;
620 else if (hdr->e_type == ET_EXEC)
621 rbase = 0;
622 else {
623 error = ENOEXEC;
624 goto fail;
625 }
626
627 /* Only support headers that fit within first page for now */
628 /* (multiplication of two Elf_Half fields will not overflow) */
629 if ((hdr->e_phoff > PAGE_SIZE) ||
630 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
631 error = ENOEXEC;
632 goto fail;
633 }
634
635 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
636 if (!aligned(phdr, Elf_Addr)) {
637 error = ENOEXEC;
638 goto fail;
639 }
640
641 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
642 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
643 prot = 0;
644 if (phdr[i].p_flags & PF_X)
645 prot |= VM_PROT_EXECUTE;
646 if (phdr[i].p_flags & PF_W)
647 prot |= VM_PROT_WRITE;
648 if (phdr[i].p_flags & PF_R)
649 prot |= VM_PROT_READ;
650
651 if ((error = __elfN(load_section)(vmspace,
652 imgp->object, phdr[i].p_offset,
653 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
654 phdr[i].p_memsz, phdr[i].p_filesz, prot,
655 pagesize)) != 0)
656 goto fail;
657 /*
658 * Establish the base address if this is the
659 * first segment.
660 */
661 if (numsegs == 0)
662 base_addr = trunc_page(phdr[i].p_vaddr +
663 rbase);
664 numsegs++;
665 }
666 }
667 *addr = base_addr;
668 *entry = (unsigned long)hdr->e_entry + rbase;
669
670 fail:
671 if (imgp->firstpage)
672 exec_unmap_first_page(imgp);
673
674 if (nd->ni_vp)
675 vput(nd->ni_vp);
676
677 VFS_UNLOCK_GIANT(vfslocked);
678 free(tempdata, M_TEMP);
679
680 return (error);
681 }
682
683 static int
684 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
685 {
686 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
687 const Elf_Phdr *phdr;
688 Elf_Auxargs *elf_auxargs;
689 struct vmspace *vmspace;
690 vm_prot_t prot;
691 u_long text_size = 0, data_size = 0, total_size = 0;
692 u_long text_addr = 0, data_addr = 0;
693 u_long seg_size, seg_addr;
694 u_long addr, entry = 0, proghdr = 0;
695 int32_t osrel = 0;
696 int error = 0, i;
697 const char *interp = NULL, *newinterp = NULL;
698 Elf_Brandinfo *brand_info;
699 char *path;
700 struct thread *td = curthread;
701 struct sysentvec *sv;
702
703 /*
704 * Do we have a valid ELF header ?
705 *
706 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
707 * if particular brand doesn't support it.
708 */
709 if (__elfN(check_header)(hdr) != 0 ||
710 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
711 return (-1);
712
713 /*
714 * From here on down, we return an errno, not -1, as we've
715 * detected an ELF file.
716 */
717
718 if ((hdr->e_phoff > PAGE_SIZE) ||
719 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
720 /* Only support headers in first page for now */
721 return (ENOEXEC);
722 }
723 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
724 if (!aligned(phdr, Elf_Addr))
725 return (ENOEXEC);
726 for (i = 0; i < hdr->e_phnum; i++) {
727 if (phdr[i].p_type == PT_INTERP) {
728 /* Path to interpreter */
729 if (phdr[i].p_filesz > MAXPATHLEN ||
730 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
731 return (ENOEXEC);
732 interp = imgp->image_header + phdr[i].p_offset;
733 break;
734 }
735 }
736
737 brand_info = __elfN(get_brandinfo)(imgp, interp, &osrel);
738 if (brand_info == NULL) {
739 uprintf("ELF binary type \"%u\" not known.\n",
740 hdr->e_ident[EI_OSABI]);
741 return (ENOEXEC);
742 }
743 if (hdr->e_type == ET_DYN &&
744 (brand_info->flags & BI_CAN_EXEC_DYN) == 0)
745 return (ENOEXEC);
746 sv = brand_info->sysvec;
747 if (interp != NULL && brand_info->interp_newpath != NULL)
748 newinterp = brand_info->interp_newpath;
749
750 /*
751 * Avoid a possible deadlock if the current address space is destroyed
752 * and that address space maps the locked vnode. In the common case,
753 * the locked vnode's v_usecount is decremented but remains greater
754 * than zero. Consequently, the vnode lock is not needed by vrele().
755 * However, in cases where the vnode lock is external, such as nullfs,
756 * v_usecount may become zero.
757 */
758 VOP_UNLOCK(imgp->vp, 0, td);
759
760 error = exec_new_vmspace(imgp, sv);
761 imgp->proc->p_sysent = sv;
762
763 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
764 if (error)
765 return (error);
766
767 vmspace = imgp->proc->p_vmspace;
768
769 for (i = 0; i < hdr->e_phnum; i++) {
770 switch (phdr[i].p_type) {
771 case PT_LOAD: /* Loadable segment */
772 prot = 0;
773 if (phdr[i].p_flags & PF_X)
774 prot |= VM_PROT_EXECUTE;
775 if (phdr[i].p_flags & PF_W)
776 prot |= VM_PROT_WRITE;
777 if (phdr[i].p_flags & PF_R)
778 prot |= VM_PROT_READ;
779
780 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
781 /*
782 * Some x86 binaries assume read == executable,
783 * notably the M3 runtime and therefore cvsup
784 */
785 if (prot & VM_PROT_READ)
786 prot |= VM_PROT_EXECUTE;
787 #endif
788
789 if ((error = __elfN(load_section)(vmspace,
790 imgp->object, phdr[i].p_offset,
791 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
792 phdr[i].p_memsz, phdr[i].p_filesz, prot,
793 sv->sv_pagesize)) != 0)
794 return (error);
795
796 /*
797 * If this segment contains the program headers,
798 * remember their virtual address for the AT_PHDR
799 * aux entry. Static binaries don't usually include
800 * a PT_PHDR entry.
801 */
802 if (phdr[i].p_offset == 0 &&
803 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
804 <= phdr[i].p_filesz)
805 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
806
807 seg_addr = trunc_page(phdr[i].p_vaddr);
808 seg_size = round_page(phdr[i].p_memsz +
809 phdr[i].p_vaddr - seg_addr);
810
811 /*
812 * Is this .text or .data? We can't use
813 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
814 * alpha terribly and possibly does other bad
815 * things so we stick to the old way of figuring
816 * it out: If the segment contains the program
817 * entry point, it's a text segment, otherwise it
818 * is a data segment.
819 *
820 * Note that obreak() assumes that data_addr +
821 * data_size == end of data load area, and the ELF
822 * file format expects segments to be sorted by
823 * address. If multiple data segments exist, the
824 * last one will be used.
825 */
826 if (hdr->e_entry >= phdr[i].p_vaddr &&
827 hdr->e_entry < (phdr[i].p_vaddr +
828 phdr[i].p_memsz)) {
829 text_size = seg_size;
830 text_addr = seg_addr;
831 entry = (u_long)hdr->e_entry;
832 } else {
833 data_size = seg_size;
834 data_addr = seg_addr;
835 }
836 total_size += seg_size;
837 break;
838 case PT_PHDR: /* Program header table info */
839 proghdr = phdr[i].p_vaddr;
840 break;
841 default:
842 break;
843 }
844 }
845
846 if (data_addr == 0 && data_size == 0) {
847 data_addr = text_addr;
848 data_size = text_size;
849 }
850
851 /*
852 * Check limits. It should be safe to check the
853 * limits after loading the segments since we do
854 * not actually fault in all the segments pages.
855 */
856 PROC_LOCK(imgp->proc);
857 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
858 text_size > maxtsiz ||
859 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
860 PROC_UNLOCK(imgp->proc);
861 return (ENOMEM);
862 }
863
864 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
865 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
866 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
867 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
868
869 /*
870 * We load the dynamic linker where a userland call
871 * to mmap(0, ...) would put it. The rationale behind this
872 * calculation is that it leaves room for the heap to grow to
873 * its maximum allowed size.
874 */
875 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
876 lim_max(imgp->proc, RLIMIT_DATA));
877 PROC_UNLOCK(imgp->proc);
878
879 imgp->entry_addr = entry;
880
881 if (interp != NULL) {
882 int have_interp = FALSE;
883 VOP_UNLOCK(imgp->vp, 0, td);
884 if (brand_info->emul_path != NULL &&
885 brand_info->emul_path[0] != '\0') {
886 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
887 snprintf(path, MAXPATHLEN, "%s%s",
888 brand_info->emul_path, interp);
889 error = __elfN(load_file)(imgp->proc, path, &addr,
890 &imgp->entry_addr, sv->sv_pagesize);
891 free(path, M_TEMP);
892 if (error == 0)
893 have_interp = TRUE;
894 }
895 if (!have_interp && newinterp != NULL) {
896 error = __elfN(load_file)(imgp->proc, newinterp, &addr,
897 &imgp->entry_addr, sv->sv_pagesize);
898 if (error == 0)
899 have_interp = TRUE;
900 }
901 if (!have_interp) {
902 error = __elfN(load_file)(imgp->proc, interp, &addr,
903 &imgp->entry_addr, sv->sv_pagesize);
904 }
905 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
906 if (error != 0) {
907 uprintf("ELF interpreter %s not found\n", interp);
908 return (error);
909 }
910 } else
911 addr = 0;
912
913 /*
914 * Construct auxargs table (used by the fixup routine)
915 */
916 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
917 elf_auxargs->execfd = -1;
918 elf_auxargs->phdr = proghdr;
919 elf_auxargs->phent = hdr->e_phentsize;
920 elf_auxargs->phnum = hdr->e_phnum;
921 elf_auxargs->pagesz = PAGE_SIZE;
922 elf_auxargs->base = addr;
923 elf_auxargs->flags = 0;
924 elf_auxargs->entry = entry;
925 elf_auxargs->trace = elf_trace;
926
927 imgp->auxargs = elf_auxargs;
928 imgp->interpreted = 0;
929 imgp->proc->p_osrel = osrel;
930
931 return (error);
932 }
933
934 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
935
936 int
937 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
938 {
939 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
940 Elf_Addr *base;
941 Elf_Addr *pos;
942
943 base = (Elf_Addr *)*stack_base;
944 pos = base + (imgp->args->argc + imgp->args->envc + 2);
945
946 if (args->trace) {
947 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
948 }
949 if (args->execfd != -1) {
950 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
951 }
952 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
953 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
954 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
955 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
956 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
957 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
958 AUXARGS_ENTRY(pos, AT_BASE, args->base);
959 AUXARGS_ENTRY(pos, AT_NULL, 0);
960
961 free(imgp->auxargs, M_TEMP);
962 imgp->auxargs = NULL;
963
964 base--;
965 suword(base, (long)imgp->args->argc);
966 *stack_base = (register_t *)base;
967 return (0);
968 }
969
970 /*
971 * Code for generating ELF core dumps.
972 */
973
974 typedef void (*segment_callback)(vm_map_entry_t, void *);
975
976 /* Closure for cb_put_phdr(). */
977 struct phdr_closure {
978 Elf_Phdr *phdr; /* Program header to fill in */
979 Elf_Off offset; /* Offset of segment in core file */
980 };
981
982 /* Closure for cb_size_segment(). */
983 struct sseg_closure {
984 int count; /* Count of writable segments. */
985 size_t size; /* Total size of all writable segments. */
986 };
987
988 static void cb_put_phdr(vm_map_entry_t, void *);
989 static void cb_size_segment(vm_map_entry_t, void *);
990 static void each_writable_segment(struct thread *, segment_callback, void *);
991 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
992 int, void *, size_t);
993 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
994 static void __elfN(putnote)(void *, size_t *, const char *, int,
995 const void *, size_t);
996
997 int
998 __elfN(coredump)(td, vp, limit)
999 struct thread *td;
1000 struct vnode *vp;
1001 off_t limit;
1002 {
1003 struct ucred *cred = td->td_ucred;
1004 int error = 0;
1005 struct sseg_closure seginfo;
1006 void *hdr;
1007 size_t hdrsize;
1008
1009 /* Size the program segments. */
1010 seginfo.count = 0;
1011 seginfo.size = 0;
1012 each_writable_segment(td, cb_size_segment, &seginfo);
1013
1014 /*
1015 * Calculate the size of the core file header area by making
1016 * a dry run of generating it. Nothing is written, but the
1017 * size is calculated.
1018 */
1019 hdrsize = 0;
1020 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
1021
1022 if (hdrsize + seginfo.size >= limit)
1023 return (EFAULT);
1024
1025 /*
1026 * Allocate memory for building the header, fill it up,
1027 * and write it out.
1028 */
1029 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
1030 if (hdr == NULL) {
1031 return (EINVAL);
1032 }
1033 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
1034
1035 /* Write the contents of all of the writable segments. */
1036 if (error == 0) {
1037 Elf_Phdr *php;
1038 off_t offset;
1039 int i;
1040
1041 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
1042 offset = hdrsize;
1043 for (i = 0; i < seginfo.count; i++) {
1044 error = vn_rdwr_inchunks(UIO_WRITE, vp,
1045 (caddr_t)(uintptr_t)php->p_vaddr,
1046 php->p_filesz, offset, UIO_USERSPACE,
1047 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1048 curthread); /* XXXKSE */
1049 if (error != 0)
1050 break;
1051 offset += php->p_filesz;
1052 php++;
1053 }
1054 }
1055 free(hdr, M_TEMP);
1056
1057 return (error);
1058 }
1059
1060 /*
1061 * A callback for each_writable_segment() to write out the segment's
1062 * program header entry.
1063 */
1064 static void
1065 cb_put_phdr(entry, closure)
1066 vm_map_entry_t entry;
1067 void *closure;
1068 {
1069 struct phdr_closure *phc = (struct phdr_closure *)closure;
1070 Elf_Phdr *phdr = phc->phdr;
1071
1072 phc->offset = round_page(phc->offset);
1073
1074 phdr->p_type = PT_LOAD;
1075 phdr->p_offset = phc->offset;
1076 phdr->p_vaddr = entry->start;
1077 phdr->p_paddr = 0;
1078 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1079 phdr->p_align = PAGE_SIZE;
1080 phdr->p_flags = 0;
1081 if (entry->protection & VM_PROT_READ)
1082 phdr->p_flags |= PF_R;
1083 if (entry->protection & VM_PROT_WRITE)
1084 phdr->p_flags |= PF_W;
1085 if (entry->protection & VM_PROT_EXECUTE)
1086 phdr->p_flags |= PF_X;
1087
1088 phc->offset += phdr->p_filesz;
1089 phc->phdr++;
1090 }
1091
1092 /*
1093 * A callback for each_writable_segment() to gather information about
1094 * the number of segments and their total size.
1095 */
1096 static void
1097 cb_size_segment(entry, closure)
1098 vm_map_entry_t entry;
1099 void *closure;
1100 {
1101 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1102
1103 ssc->count++;
1104 ssc->size += entry->end - entry->start;
1105 }
1106
1107 /*
1108 * For each writable segment in the process's memory map, call the given
1109 * function with a pointer to the map entry and some arbitrary
1110 * caller-supplied data.
1111 */
1112 static void
1113 each_writable_segment(td, func, closure)
1114 struct thread *td;
1115 segment_callback func;
1116 void *closure;
1117 {
1118 struct proc *p = td->td_proc;
1119 vm_map_t map = &p->p_vmspace->vm_map;
1120 vm_map_entry_t entry;
1121 vm_object_t backing_object, object;
1122 boolean_t ignore_entry;
1123
1124 vm_map_lock_read(map);
1125 for (entry = map->header.next; entry != &map->header;
1126 entry = entry->next) {
1127 /*
1128 * Don't dump inaccessible mappings, deal with legacy
1129 * coredump mode.
1130 *
1131 * Note that read-only segments related to the elf binary
1132 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1133 * need to arbitrarily ignore such segments.
1134 */
1135 if (elf_legacy_coredump) {
1136 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1137 continue;
1138 } else {
1139 if ((entry->protection & VM_PROT_ALL) == 0)
1140 continue;
1141 }
1142
1143 /*
1144 * Dont include memory segment in the coredump if
1145 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1146 * madvise(2). Do not dump submaps (i.e. parts of the
1147 * kernel map).
1148 */
1149 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1150 continue;
1151
1152 if ((object = entry->object.vm_object) == NULL)
1153 continue;
1154
1155 /* Ignore memory-mapped devices and such things. */
1156 VM_OBJECT_LOCK(object);
1157 while ((backing_object = object->backing_object) != NULL) {
1158 VM_OBJECT_LOCK(backing_object);
1159 VM_OBJECT_UNLOCK(object);
1160 object = backing_object;
1161 }
1162 ignore_entry = object->type != OBJT_DEFAULT &&
1163 object->type != OBJT_SWAP && object->type != OBJT_VNODE;
1164 VM_OBJECT_UNLOCK(object);
1165 if (ignore_entry)
1166 continue;
1167
1168 (*func)(entry, closure);
1169 }
1170 vm_map_unlock_read(map);
1171 }
1172
1173 /*
1174 * Write the core file header to the file, including padding up to
1175 * the page boundary.
1176 */
1177 static int
1178 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1179 struct thread *td;
1180 struct vnode *vp;
1181 struct ucred *cred;
1182 int numsegs;
1183 size_t hdrsize;
1184 void *hdr;
1185 {
1186 size_t off;
1187
1188 /* Fill in the header. */
1189 bzero(hdr, hdrsize);
1190 off = 0;
1191 __elfN(puthdr)(td, hdr, &off, numsegs);
1192
1193 /* Write it to the core file. */
1194 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1195 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1196 td)); /* XXXKSE */
1197 }
1198
1199 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1200 typedef struct prstatus32 elf_prstatus_t;
1201 typedef struct prpsinfo32 elf_prpsinfo_t;
1202 typedef struct fpreg32 elf_prfpregset_t;
1203 typedef struct fpreg32 elf_fpregset_t;
1204 typedef struct reg32 elf_gregset_t;
1205 #else
1206 typedef prstatus_t elf_prstatus_t;
1207 typedef prpsinfo_t elf_prpsinfo_t;
1208 typedef prfpregset_t elf_prfpregset_t;
1209 typedef prfpregset_t elf_fpregset_t;
1210 typedef gregset_t elf_gregset_t;
1211 #endif
1212
1213 static void
1214 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1215 {
1216 struct {
1217 elf_prstatus_t status;
1218 elf_prfpregset_t fpregset;
1219 elf_prpsinfo_t psinfo;
1220 } *tempdata;
1221 elf_prstatus_t *status;
1222 elf_prfpregset_t *fpregset;
1223 elf_prpsinfo_t *psinfo;
1224 struct proc *p;
1225 struct thread *thr;
1226 size_t ehoff, noteoff, notesz, phoff;
1227
1228 p = td->td_proc;
1229
1230 ehoff = *off;
1231 *off += sizeof(Elf_Ehdr);
1232
1233 phoff = *off;
1234 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1235
1236 noteoff = *off;
1237 /*
1238 * Don't allocate space for the notes if we're just calculating
1239 * the size of the header. We also don't collect the data.
1240 */
1241 if (dst != NULL) {
1242 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1243 status = &tempdata->status;
1244 fpregset = &tempdata->fpregset;
1245 psinfo = &tempdata->psinfo;
1246 } else {
1247 tempdata = NULL;
1248 status = NULL;
1249 fpregset = NULL;
1250 psinfo = NULL;
1251 }
1252
1253 if (dst != NULL) {
1254 psinfo->pr_version = PRPSINFO_VERSION;
1255 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1256 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1257 /*
1258 * XXX - We don't fill in the command line arguments properly
1259 * yet.
1260 */
1261 strlcpy(psinfo->pr_psargs, p->p_comm,
1262 sizeof(psinfo->pr_psargs));
1263 }
1264 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1265 sizeof *psinfo);
1266
1267 /*
1268 * To have the debugger select the right thread (LWP) as the initial
1269 * thread, we dump the state of the thread passed to us in td first.
1270 * This is the thread that causes the core dump and thus likely to
1271 * be the right thread one wants to have selected in the debugger.
1272 */
1273 thr = td;
1274 while (thr != NULL) {
1275 if (dst != NULL) {
1276 status->pr_version = PRSTATUS_VERSION;
1277 status->pr_statussz = sizeof(elf_prstatus_t);
1278 status->pr_gregsetsz = sizeof(elf_gregset_t);
1279 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1280 status->pr_osreldate = osreldate;
1281 status->pr_cursig = p->p_sig;
1282 status->pr_pid = thr->td_tid;
1283 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1284 fill_regs32(thr, &status->pr_reg);
1285 fill_fpregs32(thr, fpregset);
1286 #else
1287 fill_regs(thr, &status->pr_reg);
1288 fill_fpregs(thr, fpregset);
1289 #endif
1290 }
1291 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1292 sizeof *status);
1293 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1294 sizeof *fpregset);
1295 /*
1296 * Allow for MD specific notes, as well as any MD
1297 * specific preparations for writing MI notes.
1298 */
1299 __elfN(dump_thread)(thr, dst, off);
1300
1301 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1302 TAILQ_NEXT(thr, td_plist);
1303 if (thr == td)
1304 thr = TAILQ_NEXT(thr, td_plist);
1305 }
1306
1307 notesz = *off - noteoff;
1308
1309 if (dst != NULL)
1310 free(tempdata, M_TEMP);
1311
1312 /* Align up to a page boundary for the program segments. */
1313 *off = round_page(*off);
1314
1315 if (dst != NULL) {
1316 Elf_Ehdr *ehdr;
1317 Elf_Phdr *phdr;
1318 struct phdr_closure phc;
1319
1320 /*
1321 * Fill in the ELF header.
1322 */
1323 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1324 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1325 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1326 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1327 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1328 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1329 ehdr->e_ident[EI_DATA] = ELF_DATA;
1330 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1331 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1332 ehdr->e_ident[EI_ABIVERSION] = 0;
1333 ehdr->e_ident[EI_PAD] = 0;
1334 ehdr->e_type = ET_CORE;
1335 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1336 ehdr->e_machine = EM_386;
1337 #else
1338 ehdr->e_machine = ELF_ARCH;
1339 #endif
1340 ehdr->e_version = EV_CURRENT;
1341 ehdr->e_entry = 0;
1342 ehdr->e_phoff = phoff;
1343 ehdr->e_flags = 0;
1344 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1345 ehdr->e_phentsize = sizeof(Elf_Phdr);
1346 ehdr->e_phnum = numsegs + 1;
1347 ehdr->e_shentsize = sizeof(Elf_Shdr);
1348 ehdr->e_shnum = 0;
1349 ehdr->e_shstrndx = SHN_UNDEF;
1350
1351 /*
1352 * Fill in the program header entries.
1353 */
1354 phdr = (Elf_Phdr *)((char *)dst + phoff);
1355
1356 /* The note segement. */
1357 phdr->p_type = PT_NOTE;
1358 phdr->p_offset = noteoff;
1359 phdr->p_vaddr = 0;
1360 phdr->p_paddr = 0;
1361 phdr->p_filesz = notesz;
1362 phdr->p_memsz = 0;
1363 phdr->p_flags = 0;
1364 phdr->p_align = 0;
1365 phdr++;
1366
1367 /* All the writable segments from the program. */
1368 phc.phdr = phdr;
1369 phc.offset = *off;
1370 each_writable_segment(td, cb_put_phdr, &phc);
1371 }
1372 }
1373
1374 static void
1375 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1376 const void *desc, size_t descsz)
1377 {
1378 Elf_Note note;
1379
1380 note.n_namesz = strlen(name) + 1;
1381 note.n_descsz = descsz;
1382 note.n_type = type;
1383 if (dst != NULL)
1384 bcopy(¬e, (char *)dst + *off, sizeof note);
1385 *off += sizeof note;
1386 if (dst != NULL)
1387 bcopy(name, (char *)dst + *off, note.n_namesz);
1388 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1389 if (dst != NULL)
1390 bcopy(desc, (char *)dst + *off, note.n_descsz);
1391 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1392 }
1393
1394 /*
1395 * Try to find the appropriate ABI-note section for checknote,
1396 * fetch the osreldate for binary from the ELF OSABI-note. Only the
1397 * first page of the image is searched, the same as for headers.
1398 */
1399 static boolean_t
1400 __elfN(check_note)(struct image_params *imgp, Elf_Brandnote *checknote,
1401 int32_t *osrel)
1402 {
1403 const Elf_Note *note, *note0, *note_end;
1404 const Elf_Phdr *phdr, *pnote;
1405 const Elf_Ehdr *hdr;
1406 const char *note_name;
1407 int i;
1408
1409 pnote = NULL;
1410 hdr = (const Elf_Ehdr *)imgp->image_header;
1411 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
1412
1413 for (i = 0; i < hdr->e_phnum; i++) {
1414 if (phdr[i].p_type == PT_NOTE) {
1415 pnote = &phdr[i];
1416 break;
1417 }
1418 }
1419
1420 if (pnote == NULL || pnote->p_offset >= PAGE_SIZE ||
1421 pnote->p_offset + pnote->p_filesz >= PAGE_SIZE)
1422 return (FALSE);
1423
1424 note = note0 = (const Elf_Note *)(imgp->image_header + pnote->p_offset);
1425 note_end = (const Elf_Note *)(imgp->image_header +
1426 pnote->p_offset + pnote->p_filesz);
1427 for (i = 0; i < 100 && note >= note0 && note < note_end; i++) {
1428 if (!aligned(note, Elf32_Addr))
1429 return (FALSE);
1430 if (note->n_namesz != checknote->hdr.n_namesz ||
1431 note->n_descsz != checknote->hdr.n_descsz ||
1432 note->n_type != checknote->hdr.n_type)
1433 goto nextnote;
1434 note_name = (const char *)(note + 1);
1435 if (strncmp(checknote->vendor, note_name,
1436 checknote->hdr.n_namesz) != 0)
1437 goto nextnote;
1438
1439 /*
1440 * Fetch the osreldate for binary
1441 * from the ELF OSABI-note if necessary.
1442 */
1443 if ((checknote->flags & BN_TRANSLATE_OSREL) != 0 &&
1444 checknote->trans_osrel != NULL)
1445 return (checknote->trans_osrel(note, osrel));
1446 return (TRUE);
1447
1448 nextnote:
1449 note = (const Elf_Note *)((const char *)(note + 1) +
1450 roundup2(note->n_namesz, sizeof(Elf32_Addr)) +
1451 roundup2(note->n_descsz, sizeof(Elf32_Addr)));
1452 }
1453
1454 return (FALSE);
1455 }
1456
1457 /*
1458 * Tell kern_execve.c about it, with a little help from the linker.
1459 */
1460 static struct execsw __elfN(execsw) = {
1461 __CONCAT(exec_, __elfN(imgact)),
1462 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1463 };
1464 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
Cache object: 4e6c91dda301e9b6a96556af0ba358f0
|