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: src/sys/kern/imgact_elf.c,v 1.155.2.1 2006/01/10 02:12:27 sobomax Exp $");
33
34 #include <sys/param.h>
35 #include <sys/exec.h>
36 #include <sys/fcntl.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/mman.h>
44 #include <sys/namei.h>
45 #include <sys/pioctl.h>
46 #include <sys/proc.h>
47 #include <sys/procfs.h>
48 #include <sys/resourcevar.h>
49 #include <sys/systm.h>
50 #include <sys/signalvar.h>
51 #include <sys/stat.h>
52 #include <sys/sx.h>
53 #include <sys/syscall.h>
54 #include <sys/sysctl.h>
55 #include <sys/sysent.h>
56 #include <sys/vnode.h>
57
58 #include <vm/vm.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vm_param.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_extern.h>
65
66 #include <machine/elf.h>
67 #include <machine/md_var.h>
68
69 #define OLD_EI_BRAND 8
70
71 static int __elfN(check_header)(const Elf_Ehdr *hdr);
72 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
73 const char *interp);
74 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
75 u_long *entry, size_t pagesize);
76 static int __elfN(load_section)(struct proc *p,
77 struct vmspace *vmspace, struct vnode *vp, vm_object_t object,
78 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
79 vm_prot_t prot, size_t pagesize);
80 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
81
82 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
83 "");
84
85 int __elfN(fallback_brand) = -1;
86 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
87 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
88 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
89 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
90 &__elfN(fallback_brand));
91
92 static int elf_trace = 0;
93 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
94
95 static int elf_legacy_coredump = 0;
96 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
97 &elf_legacy_coredump, 0, "");
98
99 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
100
101 int
102 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
103 {
104 int i;
105
106 for (i = 0; i < MAX_BRANDS; i++) {
107 if (elf_brand_list[i] == NULL) {
108 elf_brand_list[i] = entry;
109 break;
110 }
111 }
112 if (i == MAX_BRANDS)
113 return (-1);
114 return (0);
115 }
116
117 int
118 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
119 {
120 int i;
121
122 for (i = 0; i < MAX_BRANDS; i++) {
123 if (elf_brand_list[i] == entry) {
124 elf_brand_list[i] = NULL;
125 break;
126 }
127 }
128 if (i == MAX_BRANDS)
129 return (-1);
130 return (0);
131 }
132
133 int
134 __elfN(brand_inuse)(Elf_Brandinfo *entry)
135 {
136 struct proc *p;
137 int rval = FALSE;
138
139 sx_slock(&allproc_lock);
140 LIST_FOREACH(p, &allproc, p_list) {
141 if (p->p_sysent == entry->sysvec) {
142 rval = TRUE;
143 break;
144 }
145 }
146 sx_sunlock(&allproc_lock);
147
148 return (rval);
149 }
150
151 static Elf_Brandinfo *
152 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
153 {
154 Elf_Brandinfo *bi;
155 int i;
156
157 /*
158 * We support three types of branding -- (1) the ELF EI_OSABI field
159 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
160 * branding w/in the ELF header, and (3) path of the `interp_path'
161 * field. We should also look for an ".note.ABI-tag" ELF section now
162 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
163 */
164
165 /* If the executable has a brand, search for it in the brand list. */
166 for (i = 0; i < MAX_BRANDS; i++) {
167 bi = elf_brand_list[i];
168 if (bi != NULL && hdr->e_machine == bi->machine &&
169 (hdr->e_ident[EI_OSABI] == bi->brand ||
170 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
171 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
172 return (bi);
173 }
174
175 /* Lacking a known brand, search for a recognized interpreter. */
176 if (interp != NULL) {
177 for (i = 0; i < MAX_BRANDS; i++) {
178 bi = elf_brand_list[i];
179 if (bi != NULL && hdr->e_machine == bi->machine &&
180 strcmp(interp, bi->interp_path) == 0)
181 return (bi);
182 }
183 }
184
185 /* Lacking a recognized interpreter, try the default brand */
186 for (i = 0; i < MAX_BRANDS; i++) {
187 bi = elf_brand_list[i];
188 if (bi != NULL && hdr->e_machine == bi->machine &&
189 __elfN(fallback_brand) == bi->brand)
190 return (bi);
191 }
192 return (NULL);
193 }
194
195 static int
196 __elfN(check_header)(const Elf_Ehdr *hdr)
197 {
198 Elf_Brandinfo *bi;
199 int i;
200
201 if (!IS_ELF(*hdr) ||
202 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
203 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
204 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
205 hdr->e_phentsize != sizeof(Elf_Phdr) ||
206 hdr->e_version != ELF_TARG_VER)
207 return (ENOEXEC);
208
209 /*
210 * Make sure we have at least one brand for this machine.
211 */
212
213 for (i = 0; i < MAX_BRANDS; i++) {
214 bi = elf_brand_list[i];
215 if (bi != NULL && bi->machine == hdr->e_machine)
216 break;
217 }
218 if (i == MAX_BRANDS)
219 return (ENOEXEC);
220
221 return (0);
222 }
223
224 static int
225 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
226 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
227 vm_prot_t max)
228 {
229 int error, rv;
230 vm_offset_t off;
231 vm_offset_t data_buf = 0;
232
233 /*
234 * Create the page if it doesn't exist yet. Ignore errors.
235 */
236 vm_map_lock(map);
237 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end), max,
238 max, 0);
239 vm_map_unlock(map);
240
241 /*
242 * Find the page from the underlying object.
243 */
244 if (object) {
245 vm_object_reference(object);
246 rv = vm_map_find(exec_map,
247 object,
248 trunc_page(offset),
249 &data_buf,
250 PAGE_SIZE,
251 TRUE,
252 VM_PROT_READ,
253 VM_PROT_ALL,
254 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
255 if (rv != KERN_SUCCESS) {
256 vm_object_deallocate(object);
257 return (rv);
258 }
259
260 off = offset - trunc_page(offset);
261 error = copyout((caddr_t)data_buf + off, (caddr_t)start,
262 end - start);
263 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
264 if (error) {
265 return (KERN_FAILURE);
266 }
267 }
268
269 return (KERN_SUCCESS);
270 }
271
272 static int
273 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
274 vm_offset_t start, vm_offset_t end, vm_prot_t prot,
275 vm_prot_t max, int cow)
276 {
277 vm_offset_t data_buf, off;
278 vm_size_t sz;
279 int error, rv;
280
281 if (start != trunc_page(start)) {
282 rv = __elfN(map_partial)(map, object, offset, start,
283 round_page(start), prot, max);
284 if (rv)
285 return (rv);
286 offset += round_page(start) - start;
287 start = round_page(start);
288 }
289 if (end != round_page(end)) {
290 rv = __elfN(map_partial)(map, object, offset +
291 trunc_page(end) - start, trunc_page(end), end, prot, max);
292 if (rv)
293 return (rv);
294 end = trunc_page(end);
295 }
296 if (end > start) {
297 if (offset & PAGE_MASK) {
298 /*
299 * The mapping is not page aligned. This means we have
300 * to copy the data. Sigh.
301 */
302 rv = vm_map_find(map, 0, 0, &start, end - start,
303 FALSE, prot, max, 0);
304 if (rv)
305 return (rv);
306 data_buf = 0;
307 while (start < end) {
308 vm_object_reference(object);
309 rv = vm_map_find(exec_map,
310 object,
311 trunc_page(offset),
312 &data_buf,
313 2 * PAGE_SIZE,
314 TRUE,
315 VM_PROT_READ,
316 VM_PROT_ALL,
317 (MAP_COPY_ON_WRITE
318 | MAP_PREFAULT_PARTIAL));
319 if (rv != KERN_SUCCESS) {
320 vm_object_deallocate(object);
321 return (rv);
322 }
323 off = offset - trunc_page(offset);
324 sz = end - start;
325 if (sz > PAGE_SIZE)
326 sz = PAGE_SIZE;
327 error = copyout((caddr_t)data_buf + off,
328 (caddr_t)start, sz);
329 vm_map_remove(exec_map, data_buf,
330 data_buf + 2 * PAGE_SIZE);
331 if (error) {
332 return (KERN_FAILURE);
333 }
334 start += sz;
335 }
336 rv = KERN_SUCCESS;
337 } else {
338 vm_map_lock(map);
339 rv = vm_map_insert(map, object, offset, start, end,
340 prot, max, cow);
341 vm_map_unlock(map);
342 }
343 return (rv);
344 } else {
345 return (KERN_SUCCESS);
346 }
347 }
348
349 static int
350 __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
351 struct vnode *vp, vm_object_t object, vm_offset_t offset,
352 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
353 size_t pagesize)
354 {
355 size_t map_len;
356 vm_offset_t map_addr;
357 int error, rv, cow;
358 size_t copy_len;
359 vm_offset_t file_addr;
360 vm_offset_t data_buf = 0;
361
362 GIANT_REQUIRED;
363
364 error = 0;
365
366 /*
367 * It's necessary to fail if the filsz + offset taken from the
368 * header is greater than the actual file pager object's size.
369 * If we were to allow this, then the vm_map_find() below would
370 * walk right off the end of the file object and into the ether.
371 *
372 * While I'm here, might as well check for something else that
373 * is invalid: filsz cannot be greater than memsz.
374 */
375 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
376 filsz > memsz) {
377 uprintf("elf_load_section: truncated ELF file\n");
378 return (ENOEXEC);
379 }
380
381 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
382 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
383
384 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
385 file_addr = trunc_page_ps(offset, pagesize);
386
387 /*
388 * We have two choices. We can either clear the data in the last page
389 * of an oversized mapping, or we can start the anon mapping a page
390 * early and copy the initialized data into that first page. We
391 * choose the second..
392 */
393 if (memsz > filsz)
394 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
395 else
396 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
397
398 if (map_len != 0) {
399 vm_object_reference(object);
400
401 /* cow flags: don't dump readonly sections in core */
402 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
403 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
404
405 rv = __elfN(map_insert)(&vmspace->vm_map,
406 object,
407 file_addr, /* file offset */
408 map_addr, /* virtual start */
409 map_addr + map_len,/* virtual end */
410 prot,
411 VM_PROT_ALL,
412 cow);
413 if (rv != KERN_SUCCESS) {
414 vm_object_deallocate(object);
415 return (EINVAL);
416 }
417
418 /* we can stop now if we've covered it all */
419 if (memsz == filsz) {
420 return (0);
421 }
422 }
423
424
425 /*
426 * We have to get the remaining bit of the file into the first part
427 * of the oversized map segment. This is normally because the .data
428 * segment in the file is extended to provide bss. It's a neat idea
429 * to try and save a page, but it's a pain in the behind to implement.
430 */
431 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
432 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
433 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
434 map_addr;
435
436 /* This had damn well better be true! */
437 if (map_len != 0) {
438 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
439 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0);
440 if (rv != KERN_SUCCESS) {
441 return (EINVAL);
442 }
443 }
444
445 if (copy_len != 0) {
446 vm_offset_t off;
447 vm_object_reference(object);
448 rv = vm_map_find(exec_map,
449 object,
450 trunc_page(offset + filsz),
451 &data_buf,
452 PAGE_SIZE,
453 TRUE,
454 VM_PROT_READ,
455 VM_PROT_ALL,
456 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
457 if (rv != KERN_SUCCESS) {
458 vm_object_deallocate(object);
459 return (EINVAL);
460 }
461
462 /* send the page fragment to user space */
463 off = trunc_page_ps(offset + filsz, pagesize) -
464 trunc_page(offset + filsz);
465 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr,
466 copy_len);
467 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
468 if (error) {
469 return (error);
470 }
471 }
472
473 /*
474 * set it to the specified protection.
475 * XXX had better undo the damage from pasting over the cracks here!
476 */
477 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
478 round_page(map_addr + map_len), prot, FALSE);
479
480 return (error);
481 }
482
483 /*
484 * Load the file "file" into memory. It may be either a shared object
485 * or an executable.
486 *
487 * The "addr" reference parameter is in/out. On entry, it specifies
488 * the address where a shared object should be loaded. If the file is
489 * an executable, this value is ignored. On exit, "addr" specifies
490 * where the file was actually loaded.
491 *
492 * The "entry" reference parameter is out only. On exit, it specifies
493 * the entry point for the loaded file.
494 */
495 static int
496 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
497 u_long *entry, size_t pagesize)
498 {
499 struct {
500 struct nameidata nd;
501 struct vattr attr;
502 struct image_params image_params;
503 } *tempdata;
504 const Elf_Ehdr *hdr = NULL;
505 const Elf_Phdr *phdr = NULL;
506 struct nameidata *nd;
507 struct vmspace *vmspace = p->p_vmspace;
508 struct vattr *attr;
509 struct image_params *imgp;
510 vm_prot_t prot;
511 u_long rbase;
512 u_long base_addr = 0;
513 int error, i, numsegs;
514
515 if (curthread->td_proc != p)
516 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
517
518 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
519 nd = &tempdata->nd;
520 attr = &tempdata->attr;
521 imgp = &tempdata->image_params;
522
523 /*
524 * Initialize part of the common data
525 */
526 imgp->proc = p;
527 imgp->userspace_argv = NULL;
528 imgp->userspace_envv = NULL;
529 imgp->attr = attr;
530 imgp->firstpage = NULL;
531 imgp->image_header = NULL;
532 imgp->object = NULL;
533 imgp->execlabel = NULL;
534
535 /* XXXKSE */
536 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread);
537
538 if ((error = namei(nd)) != 0) {
539 nd->ni_vp = NULL;
540 goto fail;
541 }
542 NDFREE(nd, NDF_ONLY_PNBUF);
543 imgp->vp = nd->ni_vp;
544
545 /*
546 * Check permissions, modes, uid, etc on the file, and "open" it.
547 */
548 error = exec_check_permissions(imgp);
549 if (error) {
550 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
551 goto fail;
552 }
553
554 error = exec_map_first_page(imgp);
555 /*
556 * Also make certain that the interpreter stays the same, so set
557 * its VV_TEXT flag, too.
558 */
559 if (error == 0)
560 nd->ni_vp->v_vflag |= VV_TEXT;
561
562 VOP_GETVOBJECT(nd->ni_vp, &imgp->object);
563 vm_object_reference(imgp->object);
564
565 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
566 if (error)
567 goto fail;
568
569 hdr = (const Elf_Ehdr *)imgp->image_header;
570 if ((error = __elfN(check_header)(hdr)) != 0)
571 goto fail;
572 if (hdr->e_type == ET_DYN)
573 rbase = *addr;
574 else if (hdr->e_type == ET_EXEC)
575 rbase = 0;
576 else {
577 error = ENOEXEC;
578 goto fail;
579 }
580
581 /* Only support headers that fit within first page for now */
582 /* (multiplication of two Elf_Half fields will not overflow) */
583 if ((hdr->e_phoff > PAGE_SIZE) ||
584 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
585 error = ENOEXEC;
586 goto fail;
587 }
588
589 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
590
591 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
592 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
593 prot = 0;
594 if (phdr[i].p_flags & PF_X)
595 prot |= VM_PROT_EXECUTE;
596 if (phdr[i].p_flags & PF_W)
597 prot |= VM_PROT_WRITE;
598 if (phdr[i].p_flags & PF_R)
599 prot |= VM_PROT_READ;
600
601 if ((error = __elfN(load_section)(p, vmspace,
602 nd->ni_vp, imgp->object, phdr[i].p_offset,
603 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
604 phdr[i].p_memsz, phdr[i].p_filesz, prot,
605 pagesize)) != 0)
606 goto fail;
607 /*
608 * Establish the base address if this is the
609 * first segment.
610 */
611 if (numsegs == 0)
612 base_addr = trunc_page(phdr[i].p_vaddr +
613 rbase);
614 numsegs++;
615 }
616 }
617 *addr = base_addr;
618 *entry = (unsigned long)hdr->e_entry + rbase;
619
620 fail:
621 if (imgp->firstpage)
622 exec_unmap_first_page(imgp);
623 if (imgp->object)
624 vm_object_deallocate(imgp->object);
625
626 if (nd->ni_vp)
627 vrele(nd->ni_vp);
628
629 free(tempdata, M_TEMP);
630
631 return (error);
632 }
633
634 static int
635 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
636 {
637 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
638 const Elf_Phdr *phdr;
639 Elf_Auxargs *elf_auxargs = NULL;
640 struct vmspace *vmspace;
641 vm_prot_t prot;
642 u_long text_size = 0, data_size = 0, total_size = 0;
643 u_long text_addr = 0, data_addr = 0;
644 u_long seg_size, seg_addr;
645 u_long addr, entry = 0, proghdr = 0;
646 int error, i;
647 const char *interp = NULL;
648 Elf_Brandinfo *brand_info;
649 char *path;
650 struct thread *td = curthread;
651 struct sysentvec *sv;
652
653 GIANT_REQUIRED;
654
655 /*
656 * Do we have a valid ELF header ?
657 *
658 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
659 * if particular brand doesn't support it.
660 */
661 if (__elfN(check_header)(hdr) != 0 ||
662 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
663 return (-1);
664
665 /*
666 * From here on down, we return an errno, not -1, as we've
667 * detected an ELF file.
668 */
669
670 if ((hdr->e_phoff > PAGE_SIZE) ||
671 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
672 /* Only support headers in first page for now */
673 return (ENOEXEC);
674 }
675 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
676
677 /*
678 * From this point on, we may have resources that need to be freed.
679 */
680
681 VOP_UNLOCK(imgp->vp, 0, td);
682
683 for (i = 0; i < hdr->e_phnum; i++) {
684 switch (phdr[i].p_type) {
685 case PT_INTERP: /* Path to interpreter */
686 if (phdr[i].p_filesz > MAXPATHLEN ||
687 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
688 error = ENOEXEC;
689 goto fail;
690 }
691 interp = imgp->image_header + phdr[i].p_offset;
692 break;
693 default:
694 break;
695 }
696 }
697
698 brand_info = __elfN(get_brandinfo)(hdr, interp);
699 if (brand_info == NULL) {
700 uprintf("ELF binary type \"%u\" not known.\n",
701 hdr->e_ident[EI_OSABI]);
702 error = ENOEXEC;
703 goto fail;
704 }
705 if (hdr->e_type == ET_DYN && brand_info->brand != ELFOSABI_LINUX) {
706 error = ENOEXEC;
707 goto fail;
708 }
709 sv = brand_info->sysvec;
710 if (interp != NULL && brand_info->interp_newpath != NULL)
711 interp = brand_info->interp_newpath;
712
713 if ((error = exec_extract_strings(imgp)) != 0)
714 goto fail;
715
716 exec_new_vmspace(imgp, sv);
717
718 vmspace = imgp->proc->p_vmspace;
719
720 for (i = 0; i < hdr->e_phnum; i++) {
721 switch (phdr[i].p_type) {
722 case PT_LOAD: /* Loadable segment */
723 prot = 0;
724 if (phdr[i].p_flags & PF_X)
725 prot |= VM_PROT_EXECUTE;
726 if (phdr[i].p_flags & PF_W)
727 prot |= VM_PROT_WRITE;
728 if (phdr[i].p_flags & PF_R)
729 prot |= VM_PROT_READ;
730
731 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
732 /*
733 * Some x86 binaries assume read == executable,
734 * notably the M3 runtime and therefore cvsup
735 */
736 if (prot & VM_PROT_READ)
737 prot |= VM_PROT_EXECUTE;
738 #endif
739
740 if ((error = __elfN(load_section)(imgp->proc, vmspace,
741 imgp->vp, imgp->object, phdr[i].p_offset,
742 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
743 phdr[i].p_memsz, phdr[i].p_filesz, prot,
744 sv->sv_pagesize)) != 0)
745 goto fail;
746
747 /*
748 * If this segment contains the program headers,
749 * remember their virtual address for the AT_PHDR
750 * aux entry. Static binaries don't usually include
751 * a PT_PHDR entry.
752 */
753 if (phdr[i].p_offset == 0 &&
754 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
755 <= phdr[i].p_filesz)
756 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
757
758 seg_addr = trunc_page(phdr[i].p_vaddr);
759 seg_size = round_page(phdr[i].p_memsz +
760 phdr[i].p_vaddr - seg_addr);
761
762 /*
763 * Is this .text or .data? We can't use
764 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
765 * alpha terribly and possibly does other bad
766 * things so we stick to the old way of figuring
767 * it out: If the segment contains the program
768 * entry point, it's a text segment, otherwise it
769 * is a data segment.
770 *
771 * Note that obreak() assumes that data_addr +
772 * data_size == end of data load area, and the ELF
773 * file format expects segments to be sorted by
774 * address. If multiple data segments exist, the
775 * last one will be used.
776 */
777 if (hdr->e_entry >= phdr[i].p_vaddr &&
778 hdr->e_entry < (phdr[i].p_vaddr +
779 phdr[i].p_memsz)) {
780 text_size = seg_size;
781 text_addr = seg_addr;
782 entry = (u_long)hdr->e_entry;
783 } else {
784 data_size = seg_size;
785 data_addr = seg_addr;
786 }
787 total_size += seg_size;
788 break;
789 case PT_PHDR: /* Program header table info */
790 proghdr = phdr[i].p_vaddr;
791 break;
792 default:
793 break;
794 }
795 }
796
797 if (data_addr == 0 && data_size == 0) {
798 data_addr = text_addr;
799 data_size = text_size;
800 }
801
802 /*
803 * Check limits. It should be safe to check the
804 * limits after loading the segments since we do
805 * not actually fault in all the segments pages.
806 */
807 PROC_LOCK(imgp->proc);
808 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
809 text_size > maxtsiz ||
810 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
811 PROC_UNLOCK(imgp->proc);
812 error = ENOMEM;
813 goto fail;
814 }
815
816 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
817 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
818 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
819 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
820
821 /*
822 * We load the dynamic linker where a userland call
823 * to mmap(0, ...) would put it. The rationale behind this
824 * calculation is that it leaves room for the heap to grow to
825 * its maximum allowed size.
826 */
827 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
828 lim_max(imgp->proc, RLIMIT_DATA));
829 PROC_UNLOCK(imgp->proc);
830
831 imgp->entry_addr = entry;
832
833 imgp->proc->p_sysent = sv;
834 if (interp != NULL && brand_info->emul_path != NULL &&
835 brand_info->emul_path[0] != '\0') {
836 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
837 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path,
838 interp);
839 error = __elfN(load_file)(imgp->proc, path, &addr,
840 &imgp->entry_addr, sv->sv_pagesize);
841 free(path, M_TEMP);
842 if (error == 0)
843 interp = NULL;
844 }
845 if (interp != NULL) {
846 error = __elfN(load_file)(imgp->proc, interp, &addr,
847 &imgp->entry_addr, sv->sv_pagesize);
848 if (error != 0) {
849 uprintf("ELF interpreter %s not found\n", interp);
850 goto fail;
851 }
852 }
853
854 /*
855 * Construct auxargs table (used by the fixup routine)
856 */
857 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
858 elf_auxargs->execfd = -1;
859 elf_auxargs->phdr = proghdr;
860 elf_auxargs->phent = hdr->e_phentsize;
861 elf_auxargs->phnum = hdr->e_phnum;
862 elf_auxargs->pagesz = PAGE_SIZE;
863 elf_auxargs->base = addr;
864 elf_auxargs->flags = 0;
865 elf_auxargs->entry = entry;
866 elf_auxargs->trace = elf_trace;
867
868 imgp->auxargs = elf_auxargs;
869 imgp->interpreted = 0;
870
871 fail:
872 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
873 return (error);
874 }
875
876 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
877
878 int
879 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
880 {
881 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
882 Elf_Addr *base;
883 Elf_Addr *pos;
884
885 base = (Elf_Addr *)*stack_base;
886 pos = base + (imgp->argc + imgp->envc + 2);
887
888 if (args->trace) {
889 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
890 }
891 if (args->execfd != -1) {
892 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
893 }
894 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
895 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
896 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
897 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
898 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
899 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
900 AUXARGS_ENTRY(pos, AT_BASE, args->base);
901 AUXARGS_ENTRY(pos, AT_NULL, 0);
902
903 free(imgp->auxargs, M_TEMP);
904 imgp->auxargs = NULL;
905
906 base--;
907 suword(base, (long)imgp->argc);
908 *stack_base = (register_t *)base;
909 return (0);
910 }
911
912 /*
913 * Code for generating ELF core dumps.
914 */
915
916 typedef void (*segment_callback)(vm_map_entry_t, void *);
917
918 /* Closure for cb_put_phdr(). */
919 struct phdr_closure {
920 Elf_Phdr *phdr; /* Program header to fill in */
921 Elf_Off offset; /* Offset of segment in core file */
922 };
923
924 /* Closure for cb_size_segment(). */
925 struct sseg_closure {
926 int count; /* Count of writable segments. */
927 size_t size; /* Total size of all writable segments. */
928 };
929
930 static void cb_put_phdr(vm_map_entry_t, void *);
931 static void cb_size_segment(vm_map_entry_t, void *);
932 static void each_writable_segment(struct thread *, segment_callback, void *);
933 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
934 int, void *, size_t);
935 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
936 static void __elfN(putnote)(void *, size_t *, const char *, int,
937 const void *, size_t);
938
939 extern int osreldate;
940
941 int
942 __elfN(coredump)(td, vp, limit)
943 struct thread *td;
944 struct vnode *vp;
945 off_t limit;
946 {
947 struct ucred *cred = td->td_ucred;
948 int error = 0;
949 struct sseg_closure seginfo;
950 void *hdr;
951 size_t hdrsize;
952
953 /* Size the program segments. */
954 seginfo.count = 0;
955 seginfo.size = 0;
956 each_writable_segment(td, cb_size_segment, &seginfo);
957
958 /*
959 * Calculate the size of the core file header area by making
960 * a dry run of generating it. Nothing is written, but the
961 * size is calculated.
962 */
963 hdrsize = 0;
964 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
965
966 if (hdrsize + seginfo.size >= limit)
967 return (EFAULT);
968
969 /*
970 * Allocate memory for building the header, fill it up,
971 * and write it out.
972 */
973 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
974 if (hdr == NULL) {
975 return (EINVAL);
976 }
977 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
978
979 /* Write the contents of all of the writable segments. */
980 if (error == 0) {
981 Elf_Phdr *php;
982 off_t offset;
983 int i;
984
985 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
986 offset = hdrsize;
987 for (i = 0; i < seginfo.count; i++) {
988 error = vn_rdwr_inchunks(UIO_WRITE, vp,
989 (caddr_t)(uintptr_t)php->p_vaddr,
990 php->p_filesz, offset, UIO_USERSPACE,
991 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
992 curthread); /* XXXKSE */
993 if (error != 0)
994 break;
995 offset += php->p_filesz;
996 php++;
997 }
998 }
999 free(hdr, M_TEMP);
1000
1001 return (error);
1002 }
1003
1004 /*
1005 * A callback for each_writable_segment() to write out the segment's
1006 * program header entry.
1007 */
1008 static void
1009 cb_put_phdr(entry, closure)
1010 vm_map_entry_t entry;
1011 void *closure;
1012 {
1013 struct phdr_closure *phc = (struct phdr_closure *)closure;
1014 Elf_Phdr *phdr = phc->phdr;
1015
1016 phc->offset = round_page(phc->offset);
1017
1018 phdr->p_type = PT_LOAD;
1019 phdr->p_offset = phc->offset;
1020 phdr->p_vaddr = entry->start;
1021 phdr->p_paddr = 0;
1022 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1023 phdr->p_align = PAGE_SIZE;
1024 phdr->p_flags = 0;
1025 if (entry->protection & VM_PROT_READ)
1026 phdr->p_flags |= PF_R;
1027 if (entry->protection & VM_PROT_WRITE)
1028 phdr->p_flags |= PF_W;
1029 if (entry->protection & VM_PROT_EXECUTE)
1030 phdr->p_flags |= PF_X;
1031
1032 phc->offset += phdr->p_filesz;
1033 phc->phdr++;
1034 }
1035
1036 /*
1037 * A callback for each_writable_segment() to gather information about
1038 * the number of segments and their total size.
1039 */
1040 static void
1041 cb_size_segment(entry, closure)
1042 vm_map_entry_t entry;
1043 void *closure;
1044 {
1045 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1046
1047 ssc->count++;
1048 ssc->size += entry->end - entry->start;
1049 }
1050
1051 /*
1052 * For each writable segment in the process's memory map, call the given
1053 * function with a pointer to the map entry and some arbitrary
1054 * caller-supplied data.
1055 */
1056 static void
1057 each_writable_segment(td, func, closure)
1058 struct thread *td;
1059 segment_callback func;
1060 void *closure;
1061 {
1062 struct proc *p = td->td_proc;
1063 vm_map_t map = &p->p_vmspace->vm_map;
1064 vm_map_entry_t entry;
1065
1066 for (entry = map->header.next; entry != &map->header;
1067 entry = entry->next) {
1068 vm_object_t obj;
1069
1070 /*
1071 * Don't dump inaccessible mappings, deal with legacy
1072 * coredump mode.
1073 *
1074 * Note that read-only segments related to the elf binary
1075 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1076 * need to arbitrarily ignore such segments.
1077 */
1078 if (elf_legacy_coredump) {
1079 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1080 continue;
1081 } else {
1082 if ((entry->protection & VM_PROT_ALL) == 0)
1083 continue;
1084 }
1085
1086 /*
1087 * Dont include memory segment in the coredump if
1088 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1089 * madvise(2). Do not dump submaps (i.e. parts of the
1090 * kernel map).
1091 */
1092 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1093 continue;
1094
1095 if ((obj = entry->object.vm_object) == NULL)
1096 continue;
1097
1098 /* Find the deepest backing object. */
1099 while (obj->backing_object != NULL)
1100 obj = obj->backing_object;
1101
1102 /* Ignore memory-mapped devices and such things. */
1103 if (obj->type != OBJT_DEFAULT &&
1104 obj->type != OBJT_SWAP &&
1105 obj->type != OBJT_VNODE)
1106 continue;
1107
1108 (*func)(entry, closure);
1109 }
1110 }
1111
1112 /*
1113 * Write the core file header to the file, including padding up to
1114 * the page boundary.
1115 */
1116 static int
1117 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1118 struct thread *td;
1119 struct vnode *vp;
1120 struct ucred *cred;
1121 int numsegs;
1122 size_t hdrsize;
1123 void *hdr;
1124 {
1125 size_t off;
1126
1127 /* Fill in the header. */
1128 bzero(hdr, hdrsize);
1129 off = 0;
1130 __elfN(puthdr)(td, hdr, &off, numsegs);
1131
1132 /* Write it to the core file. */
1133 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1134 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1135 td)); /* XXXKSE */
1136 }
1137
1138 static void
1139 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1140 {
1141 struct {
1142 prstatus_t status;
1143 prfpregset_t fpregset;
1144 prpsinfo_t psinfo;
1145 } *tempdata;
1146 prstatus_t *status;
1147 prfpregset_t *fpregset;
1148 prpsinfo_t *psinfo;
1149 struct proc *p;
1150 struct thread *thr;
1151 size_t ehoff, noteoff, notesz, phoff;
1152
1153 p = td->td_proc;
1154
1155 ehoff = *off;
1156 *off += sizeof(Elf_Ehdr);
1157
1158 phoff = *off;
1159 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1160
1161 noteoff = *off;
1162 /*
1163 * Don't allocate space for the notes if we're just calculating
1164 * the size of the header. We also don't collect the data.
1165 */
1166 if (dst != NULL) {
1167 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1168 status = &tempdata->status;
1169 fpregset = &tempdata->fpregset;
1170 psinfo = &tempdata->psinfo;
1171 } else {
1172 tempdata = NULL;
1173 status = NULL;
1174 fpregset = NULL;
1175 psinfo = NULL;
1176 }
1177
1178 if (dst != NULL) {
1179 psinfo->pr_version = PRPSINFO_VERSION;
1180 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1181 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1182 /*
1183 * XXX - We don't fill in the command line arguments properly
1184 * yet.
1185 */
1186 strlcpy(psinfo->pr_psargs, p->p_comm,
1187 sizeof(psinfo->pr_psargs));
1188 }
1189 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1190 sizeof *psinfo);
1191
1192 /*
1193 * To have the debugger select the right thread (LWP) as the initial
1194 * thread, we dump the state of the thread passed to us in td first.
1195 * This is the thread that causes the core dump and thus likely to
1196 * be the right thread one wants to have selected in the debugger.
1197 */
1198 thr = td;
1199 while (thr != NULL) {
1200 if (dst != NULL) {
1201 status->pr_version = PRSTATUS_VERSION;
1202 status->pr_statussz = sizeof(prstatus_t);
1203 status->pr_gregsetsz = sizeof(gregset_t);
1204 status->pr_fpregsetsz = sizeof(fpregset_t);
1205 status->pr_osreldate = osreldate;
1206 status->pr_cursig = p->p_sig;
1207 status->pr_pid = thr->td_tid;
1208 fill_regs(thr, &status->pr_reg);
1209 fill_fpregs(thr, fpregset);
1210 }
1211 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1212 sizeof *status);
1213 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1214 sizeof *fpregset);
1215 /*
1216 * Allow for MD specific notes, as well as any MD
1217 * specific preparations for writing MI notes.
1218 */
1219 __elfN(dump_thread)(thr, dst, off);
1220
1221 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1222 TAILQ_NEXT(thr, td_plist);
1223 if (thr == td)
1224 thr = TAILQ_NEXT(thr, td_plist);
1225 }
1226
1227 notesz = *off - noteoff;
1228
1229 if (dst != NULL)
1230 free(tempdata, M_TEMP);
1231
1232 /* Align up to a page boundary for the program segments. */
1233 *off = round_page(*off);
1234
1235 if (dst != NULL) {
1236 Elf_Ehdr *ehdr;
1237 Elf_Phdr *phdr;
1238 struct phdr_closure phc;
1239
1240 /*
1241 * Fill in the ELF header.
1242 */
1243 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1244 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1245 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1246 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1247 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1248 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1249 ehdr->e_ident[EI_DATA] = ELF_DATA;
1250 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1251 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1252 ehdr->e_ident[EI_ABIVERSION] = 0;
1253 ehdr->e_ident[EI_PAD] = 0;
1254 ehdr->e_type = ET_CORE;
1255 ehdr->e_machine = ELF_ARCH;
1256 ehdr->e_version = EV_CURRENT;
1257 ehdr->e_entry = 0;
1258 ehdr->e_phoff = phoff;
1259 ehdr->e_flags = 0;
1260 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1261 ehdr->e_phentsize = sizeof(Elf_Phdr);
1262 ehdr->e_phnum = numsegs + 1;
1263 ehdr->e_shentsize = sizeof(Elf_Shdr);
1264 ehdr->e_shnum = 0;
1265 ehdr->e_shstrndx = SHN_UNDEF;
1266
1267 /*
1268 * Fill in the program header entries.
1269 */
1270 phdr = (Elf_Phdr *)((char *)dst + phoff);
1271
1272 /* The note segement. */
1273 phdr->p_type = PT_NOTE;
1274 phdr->p_offset = noteoff;
1275 phdr->p_vaddr = 0;
1276 phdr->p_paddr = 0;
1277 phdr->p_filesz = notesz;
1278 phdr->p_memsz = 0;
1279 phdr->p_flags = 0;
1280 phdr->p_align = 0;
1281 phdr++;
1282
1283 /* All the writable segments from the program. */
1284 phc.phdr = phdr;
1285 phc.offset = *off;
1286 each_writable_segment(td, cb_put_phdr, &phc);
1287 }
1288 }
1289
1290 static void
1291 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1292 const void *desc, size_t descsz)
1293 {
1294 Elf_Note note;
1295
1296 note.n_namesz = strlen(name) + 1;
1297 note.n_descsz = descsz;
1298 note.n_type = type;
1299 if (dst != NULL)
1300 bcopy(¬e, (char *)dst + *off, sizeof note);
1301 *off += sizeof note;
1302 if (dst != NULL)
1303 bcopy(name, (char *)dst + *off, note.n_namesz);
1304 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1305 if (dst != NULL)
1306 bcopy(desc, (char *)dst + *off, note.n_descsz);
1307 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1308 }
1309
1310 /*
1311 * Tell kern_execve.c about it, with a little help from the linker.
1312 */
1313 static struct execsw __elfN(execsw) = {
1314 __CONCAT(exec_, __elfN(imgact)),
1315 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1316 };
1317 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
Cache object: c198f638b1388735ab25c9c22f9970b6
|