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 * $FreeBSD: releng/5.1/sys/kern/imgact_elf.c 111119 2003-02-19 05:47:46Z imp $
31 */
32
33 #include <sys/param.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/imgact.h>
37 #include <sys/imgact_elf.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/mutex.h>
42 #include <sys/mman.h>
43 #include <sys/namei.h>
44 #include <sys/pioctl.h>
45 #include <sys/proc.h>
46 #include <sys/procfs.h>
47 #include <sys/resourcevar.h>
48 #include <sys/systm.h>
49 #include <sys/signalvar.h>
50 #include <sys/stat.h>
51 #include <sys/sx.h>
52 #include <sys/syscall.h>
53 #include <sys/sysctl.h>
54 #include <sys/sysent.h>
55 #include <sys/vnode.h>
56
57 #include <vm/vm.h>
58 #include <vm/vm_kern.h>
59 #include <vm/vm_param.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_extern.h>
64
65 #include <machine/elf.h>
66 #include <machine/md_var.h>
67
68 #define OLD_EI_BRAND 8
69
70 static int __elfN(check_header)(const Elf_Ehdr *hdr);
71 static Elf_Brandinfo *__elfN(get_brandinfo)(const Elf_Ehdr *hdr,
72 const char *interp);
73 static int __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
74 u_long *entry, size_t pagesize);
75 static int __elfN(load_section)(struct proc *p,
76 struct vmspace *vmspace, struct vnode *vp, vm_object_t object,
77 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
78 vm_prot_t prot, size_t pagesize);
79 static int __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp);
80
81 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
82 "");
83
84 int __elfN(fallback_brand) = -1;
85 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
86 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
87 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
88 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
89 &__elfN(fallback_brand));
90
91 static int elf_trace = 0;
92 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
93
94 static int elf_legacy_coredump = 0;
95 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
96 &elf_legacy_coredump, 0, "");
97
98 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
99
100 int
101 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
102 {
103 int i;
104
105 for (i = 0; i < MAX_BRANDS; i++) {
106 if (elf_brand_list[i] == NULL) {
107 elf_brand_list[i] = entry;
108 break;
109 }
110 }
111 if (i == MAX_BRANDS)
112 return (-1);
113 return (0);
114 }
115
116 int
117 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
118 {
119 int i;
120
121 for (i = 0; i < MAX_BRANDS; i++) {
122 if (elf_brand_list[i] == entry) {
123 elf_brand_list[i] = NULL;
124 break;
125 }
126 }
127 if (i == MAX_BRANDS)
128 return (-1);
129 return (0);
130 }
131
132 int
133 __elfN(brand_inuse)(Elf_Brandinfo *entry)
134 {
135 struct proc *p;
136 int rval = FALSE;
137
138 sx_slock(&allproc_lock);
139 LIST_FOREACH(p, &allproc, p_list) {
140 if (p->p_sysent == entry->sysvec) {
141 rval = TRUE;
142 break;
143 }
144 }
145 sx_sunlock(&allproc_lock);
146
147 return (rval);
148 }
149
150 static Elf_Brandinfo *
151 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
152 {
153 Elf_Brandinfo *bi;
154 int i;
155
156 /*
157 * We support three types of branding -- (1) the ELF EI_OSABI field
158 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
159 * branding w/in the ELF header, and (3) path of the `interp_path'
160 * field. We should also look for an ".note.ABI-tag" ELF section now
161 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
162 */
163
164 /* If the executable has a brand, search for it in the brand list. */
165 for (i = 0; i < MAX_BRANDS; i++) {
166 bi = elf_brand_list[i];
167 if (bi != NULL && hdr->e_machine == bi->machine &&
168 (hdr->e_ident[EI_OSABI] == bi->brand ||
169 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
170 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
171 return (bi);
172 }
173
174 /* Lacking a known brand, search for a recognized interpreter. */
175 if (interp != NULL) {
176 for (i = 0; i < MAX_BRANDS; i++) {
177 bi = elf_brand_list[i];
178 if (bi != NULL && hdr->e_machine == bi->machine &&
179 strcmp(interp, bi->interp_path) == 0)
180 return (bi);
181 }
182 }
183
184 /* Lacking a recognized interpreter, try the default brand */
185 for (i = 0; i < MAX_BRANDS; i++) {
186 bi = elf_brand_list[i];
187 if (bi != NULL && hdr->e_machine == bi->machine &&
188 __elfN(fallback_brand) == bi->brand)
189 return (bi);
190 }
191 return (NULL);
192 }
193
194 static int
195 __elfN(check_header)(const Elf_Ehdr *hdr)
196 {
197 Elf_Brandinfo *bi;
198 int i;
199
200 if (!IS_ELF(*hdr) ||
201 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
202 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
203 hdr->e_ident[EI_VERSION] != EV_CURRENT)
204 return (ENOEXEC);
205
206 /*
207 * Make sure we have at least one brand for this machine.
208 */
209
210 for (i = 0; i < MAX_BRANDS; i++) {
211 bi = elf_brand_list[i];
212 if (bi != NULL && bi->machine == hdr->e_machine)
213 break;
214 }
215 if (i == MAX_BRANDS)
216 return (ENOEXEC);
217
218 if (hdr->e_version != ELF_TARG_VER)
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 int rv;
278
279 if (start != trunc_page(start)) {
280 rv = __elfN(map_partial)(map, object, offset, start,
281 round_page(start), prot, max);
282 if (rv)
283 return (rv);
284 offset += round_page(start) - start;
285 start = round_page(start);
286 }
287 if (end != round_page(end)) {
288 rv = __elfN(map_partial)(map, object, offset +
289 trunc_page(end) - start, trunc_page(end), end, prot, max);
290 if (rv)
291 return (rv);
292 end = trunc_page(end);
293 }
294 if (end > start) {
295 if (offset & PAGE_MASK) {
296 vm_offset_t data_buf, off;
297 vm_size_t sz;
298 int error;
299
300 /*
301 * The mapping is not page aligned. This means we have
302 * to copy the data. Sigh.
303 */
304 rv = vm_map_find(map, 0, 0, &start, end - start,
305 FALSE, prot, max, 0);
306 if (rv)
307 return (rv);
308 while (start < end) {
309 vm_object_reference(object);
310 rv = vm_map_find(exec_map,
311 object,
312 trunc_page(offset),
313 &data_buf,
314 2 * PAGE_SIZE,
315 TRUE,
316 VM_PROT_READ,
317 VM_PROT_ALL,
318 (MAP_COPY_ON_WRITE
319 | MAP_PREFAULT_PARTIAL));
320 if (rv != KERN_SUCCESS) {
321 vm_object_deallocate(object);
322 return (rv);
323 }
324 off = offset - trunc_page(offset);
325 sz = end - start;
326 if (sz > PAGE_SIZE)
327 sz = PAGE_SIZE;
328 error = copyout((caddr_t)data_buf + off,
329 (caddr_t)start, sz);
330 vm_map_remove(exec_map, data_buf,
331 data_buf + 2 * PAGE_SIZE);
332 if (error) {
333 return (KERN_FAILURE);
334 }
335 start += sz;
336 }
337 rv = KERN_SUCCESS;
338 } else {
339 vm_map_lock(map);
340 rv = vm_map_insert(map, object, offset, start, end,
341 prot, max, cow);
342 vm_map_unlock(map);
343 }
344 return (rv);
345 } else {
346 return (KERN_SUCCESS);
347 }
348 }
349
350 static int
351 __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
352 struct vnode *vp, vm_object_t object, vm_offset_t offset,
353 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
354 size_t pagesize)
355 {
356 size_t map_len;
357 vm_offset_t map_addr;
358 int error, rv, cow;
359 size_t copy_len;
360 vm_offset_t file_addr;
361 vm_offset_t data_buf = 0;
362
363 GIANT_REQUIRED;
364
365 error = 0;
366
367 /*
368 * It's necessary to fail if the filsz + offset taken from the
369 * header is greater than the actual file pager object's size.
370 * If we were to allow this, then the vm_map_find() below would
371 * walk right off the end of the file object and into the ether.
372 *
373 * While I'm here, might as well check for something else that
374 * is invalid: filsz cannot be greater than memsz.
375 */
376 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
377 filsz > memsz) {
378 uprintf("elf_load_section: truncated ELF file\n");
379 return (ENOEXEC);
380 }
381
382 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
383 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
384
385 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
386 file_addr = trunc_page_ps(offset, pagesize);
387
388 /*
389 * We have two choices. We can either clear the data in the last page
390 * of an oversized mapping, or we can start the anon mapping a page
391 * early and copy the initialized data into that first page. We
392 * choose the second..
393 */
394 if (memsz > filsz)
395 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
396 else
397 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
398
399 if (map_len != 0) {
400 vm_object_reference(object);
401
402 /* cow flags: don't dump readonly sections in core */
403 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
404 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
405
406 rv = __elfN(map_insert)(&vmspace->vm_map,
407 object,
408 file_addr, /* file offset */
409 map_addr, /* virtual start */
410 map_addr + map_len,/* virtual end */
411 prot,
412 VM_PROT_ALL,
413 cow);
414 if (rv != KERN_SUCCESS) {
415 vm_object_deallocate(object);
416 return (EINVAL);
417 }
418
419 /* we can stop now if we've covered it all */
420 if (memsz == filsz) {
421 return (0);
422 }
423 }
424
425
426 /*
427 * We have to get the remaining bit of the file into the first part
428 * of the oversized map segment. This is normally because the .data
429 * segment in the file is extended to provide bss. It's a neat idea
430 * to try and save a page, but it's a pain in the behind to implement.
431 */
432 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
433 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
434 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
435 map_addr;
436
437 /* This had damn well better be true! */
438 if (map_len != 0) {
439 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
440 map_addr + map_len, VM_PROT_ALL, VM_PROT_ALL, 0);
441 if (rv != KERN_SUCCESS) {
442 return (EINVAL);
443 }
444 }
445
446 if (copy_len != 0) {
447 vm_offset_t off;
448 vm_object_reference(object);
449 rv = vm_map_find(exec_map,
450 object,
451 trunc_page(offset + filsz),
452 &data_buf,
453 PAGE_SIZE,
454 TRUE,
455 VM_PROT_READ,
456 VM_PROT_ALL,
457 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
458 if (rv != KERN_SUCCESS) {
459 vm_object_deallocate(object);
460 return (EINVAL);
461 }
462
463 /* send the page fragment to user space */
464 off = trunc_page_ps(offset + filsz, pagesize) -
465 trunc_page(offset + filsz);
466 error = copyout((caddr_t)data_buf + off, (caddr_t)map_addr,
467 copy_len);
468 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
469 if (error) {
470 return (error);
471 }
472 }
473
474 /*
475 * set it to the specified protection.
476 * XXX had better undo the damage from pasting over the cracks here!
477 */
478 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
479 round_page(map_addr + map_len), prot, FALSE);
480
481 return (error);
482 }
483
484 /*
485 * Load the file "file" into memory. It may be either a shared object
486 * or an executable.
487 *
488 * The "addr" reference parameter is in/out. On entry, it specifies
489 * the address where a shared object should be loaded. If the file is
490 * an executable, this value is ignored. On exit, "addr" specifies
491 * where the file was actually loaded.
492 *
493 * The "entry" reference parameter is out only. On exit, it specifies
494 * the entry point for the loaded file.
495 */
496 static int
497 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
498 u_long *entry, size_t pagesize)
499 {
500 struct {
501 struct nameidata nd;
502 struct vattr attr;
503 struct image_params image_params;
504 } *tempdata;
505 const Elf_Ehdr *hdr = NULL;
506 const Elf_Phdr *phdr = NULL;
507 struct nameidata *nd;
508 struct vmspace *vmspace = p->p_vmspace;
509 struct vattr *attr;
510 struct image_params *imgp;
511 vm_prot_t prot;
512 u_long rbase;
513 u_long base_addr = 0;
514 int error, i, numsegs;
515
516 if (curthread->td_proc != p)
517 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
518
519 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
520 nd = &tempdata->nd;
521 attr = &tempdata->attr;
522 imgp = &tempdata->image_params;
523
524 /*
525 * Initialize part of the common data
526 */
527 imgp->proc = p;
528 imgp->userspace_argv = NULL;
529 imgp->userspace_envv = NULL;
530 imgp->attr = attr;
531 imgp->firstpage = NULL;
532 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
533 imgp->object = NULL;
534 imgp->execlabel = NULL;
535
536 if (imgp->image_header == NULL) {
537 nd->ni_vp = NULL;
538 error = ENOMEM;
539 goto fail;
540 }
541
542 /* XXXKSE */
543 NDINIT(nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, curthread);
544
545 if ((error = namei(nd)) != 0) {
546 nd->ni_vp = NULL;
547 goto fail;
548 }
549 NDFREE(nd, NDF_ONLY_PNBUF);
550 imgp->vp = nd->ni_vp;
551
552 /*
553 * Check permissions, modes, uid, etc on the file, and "open" it.
554 */
555 error = exec_check_permissions(imgp);
556 if (error) {
557 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
558 goto fail;
559 }
560
561 error = exec_map_first_page(imgp);
562 /*
563 * Also make certain that the interpreter stays the same, so set
564 * its VV_TEXT flag, too.
565 */
566 if (error == 0)
567 nd->ni_vp->v_vflag |= VV_TEXT;
568
569 VOP_GETVOBJECT(nd->ni_vp, &imgp->object);
570 vm_object_reference(imgp->object);
571
572 VOP_UNLOCK(nd->ni_vp, 0, curthread); /* XXXKSE */
573 if (error)
574 goto fail;
575
576 hdr = (const Elf_Ehdr *)imgp->image_header;
577 if ((error = __elfN(check_header)(hdr)) != 0)
578 goto fail;
579 if (hdr->e_type == ET_DYN)
580 rbase = *addr;
581 else if (hdr->e_type == ET_EXEC)
582 rbase = 0;
583 else {
584 error = ENOEXEC;
585 goto fail;
586 }
587
588 /* Only support headers that fit within first page for now */
589 if ((hdr->e_phoff > PAGE_SIZE) ||
590 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
591 error = ENOEXEC;
592 goto fail;
593 }
594
595 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
596
597 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
598 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
599 prot = 0;
600 if (phdr[i].p_flags & PF_X)
601 prot |= VM_PROT_EXECUTE;
602 if (phdr[i].p_flags & PF_W)
603 prot |= VM_PROT_WRITE;
604 if (phdr[i].p_flags & PF_R)
605 prot |= VM_PROT_READ;
606
607 if ((error = __elfN(load_section)(p, vmspace,
608 nd->ni_vp, imgp->object, phdr[i].p_offset,
609 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
610 phdr[i].p_memsz, phdr[i].p_filesz, prot,
611 pagesize)) != 0)
612 goto fail;
613 /*
614 * Establish the base address if this is the
615 * first segment.
616 */
617 if (numsegs == 0)
618 base_addr = trunc_page(phdr[i].p_vaddr +
619 rbase);
620 numsegs++;
621 }
622 }
623 *addr = base_addr;
624 *entry = (unsigned long)hdr->e_entry + rbase;
625
626 fail:
627 if (imgp->firstpage)
628 exec_unmap_first_page(imgp);
629 if (imgp->image_header)
630 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
631 PAGE_SIZE);
632 if (imgp->object)
633 vm_object_deallocate(imgp->object);
634
635 if (nd->ni_vp)
636 vrele(nd->ni_vp);
637
638 free(tempdata, M_TEMP);
639
640 return (error);
641 }
642
643 static int
644 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
645 {
646 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
647 const Elf_Phdr *phdr;
648 Elf_Auxargs *elf_auxargs = NULL;
649 struct vmspace *vmspace;
650 vm_prot_t prot;
651 u_long text_size = 0, data_size = 0, total_size = 0;
652 u_long text_addr = 0, data_addr = 0;
653 u_long seg_size, seg_addr;
654 u_long addr, entry = 0, proghdr = 0;
655 int error, i;
656 const char *interp = NULL;
657 Elf_Brandinfo *brand_info;
658 char *path;
659 struct thread *td = curthread;
660 struct sysentvec *sv;
661
662 GIANT_REQUIRED;
663
664 /*
665 * Do we have a valid ELF header ?
666 */
667 if (__elfN(check_header)(hdr) != 0 || hdr->e_type != ET_EXEC)
668 return (-1);
669
670 /*
671 * From here on down, we return an errno, not -1, as we've
672 * detected an ELF file.
673 */
674
675 if ((hdr->e_phoff > PAGE_SIZE) ||
676 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
677 /* Only support headers in first page for now */
678 return (ENOEXEC);
679 }
680 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
681
682 /*
683 * From this point on, we may have resources that need to be freed.
684 */
685
686 VOP_UNLOCK(imgp->vp, 0, td);
687
688 for (i = 0; i < hdr->e_phnum; i++) {
689 switch (phdr[i].p_type) {
690 case PT_INTERP: /* Path to interpreter */
691 if (phdr[i].p_filesz > MAXPATHLEN ||
692 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
693 error = ENOEXEC;
694 goto fail;
695 }
696 interp = imgp->image_header + phdr[i].p_offset;
697 break;
698 default:
699 break;
700 }
701 }
702
703 brand_info = __elfN(get_brandinfo)(hdr, interp);
704 if (brand_info == NULL) {
705 uprintf("ELF binary type \"%u\" not known.\n",
706 hdr->e_ident[EI_OSABI]);
707 error = ENOEXEC;
708 goto fail;
709 }
710 sv = brand_info->sysvec;
711
712 if ((error = exec_extract_strings(imgp)) != 0)
713 goto fail;
714
715 exec_new_vmspace(imgp, sv);
716
717 vmspace = imgp->proc->p_vmspace;
718
719 for (i = 0; i < hdr->e_phnum; i++) {
720 switch (phdr[i].p_type) {
721 case PT_LOAD: /* Loadable segment */
722 prot = 0;
723 if (phdr[i].p_flags & PF_X)
724 prot |= VM_PROT_EXECUTE;
725 if (phdr[i].p_flags & PF_W)
726 prot |= VM_PROT_WRITE;
727 if (phdr[i].p_flags & PF_R)
728 prot |= VM_PROT_READ;
729
730 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
731 /*
732 * Some x86 binaries assume read == executable,
733 * notably the M3 runtime and therefore cvsup
734 */
735 if (prot & VM_PROT_READ)
736 prot |= VM_PROT_EXECUTE;
737 #endif
738
739 if ((error = __elfN(load_section)(imgp->proc, vmspace,
740 imgp->vp, imgp->object, phdr[i].p_offset,
741 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
742 phdr[i].p_memsz, phdr[i].p_filesz, prot,
743 sv->sv_pagesize)) != 0)
744 goto fail;
745
746 seg_addr = trunc_page(phdr[i].p_vaddr);
747 seg_size = round_page(phdr[i].p_memsz +
748 phdr[i].p_vaddr - seg_addr);
749
750 /*
751 * Is this .text or .data? We can't use
752 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
753 * alpha terribly and possibly does other bad
754 * things so we stick to the old way of figuring
755 * it out: If the segment contains the program
756 * entry point, it's a text segment, otherwise it
757 * is a data segment.
758 *
759 * Note that obreak() assumes that data_addr +
760 * data_size == end of data load area, and the ELF
761 * file format expects segments to be sorted by
762 * address. If multiple data segments exist, the
763 * last one will be used.
764 */
765 if (hdr->e_entry >= phdr[i].p_vaddr &&
766 hdr->e_entry < (phdr[i].p_vaddr +
767 phdr[i].p_memsz)) {
768 text_size = seg_size;
769 text_addr = seg_addr;
770 entry = (u_long)hdr->e_entry;
771 } else {
772 data_size = seg_size;
773 data_addr = seg_addr;
774 }
775 total_size += seg_size;
776 break;
777 case PT_PHDR: /* Program header table info */
778 proghdr = phdr[i].p_vaddr;
779 break;
780 default:
781 break;
782 }
783 }
784
785 if (data_addr == 0 && data_size == 0) {
786 data_addr = text_addr;
787 data_size = text_size;
788 }
789
790 /*
791 * Check limits. It should be safe to check the
792 * limits after loading the segments since we do
793 * not actually fault in all the segments pages.
794 */
795 if (data_size >
796 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
797 text_size > maxtsiz ||
798 total_size >
799 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
800 error = ENOMEM;
801 goto fail;
802 }
803
804 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
805 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
806 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
807 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
808
809 addr = ELF_RTLD_ADDR(vmspace);
810
811 imgp->entry_addr = entry;
812
813 imgp->proc->p_sysent = sv;
814 if (interp != NULL) {
815 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
816 snprintf(path, MAXPATHLEN, "%s%s", brand_info->emul_path,
817 interp);
818 if ((error = __elfN(load_file)(imgp->proc, path, &addr,
819 &imgp->entry_addr, sv->sv_pagesize)) != 0) {
820 if ((error = __elfN(load_file)(imgp->proc, interp,
821 &addr, &imgp->entry_addr, sv->sv_pagesize)) != 0) {
822 uprintf("ELF interpreter %s not found\n",
823 path);
824 free(path, M_TEMP);
825 goto fail;
826 }
827 }
828 free(path, M_TEMP);
829 }
830
831 /*
832 * Construct auxargs table (used by the fixup routine)
833 */
834 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
835 elf_auxargs->execfd = -1;
836 elf_auxargs->phdr = proghdr;
837 elf_auxargs->phent = hdr->e_phentsize;
838 elf_auxargs->phnum = hdr->e_phnum;
839 elf_auxargs->pagesz = PAGE_SIZE;
840 elf_auxargs->base = addr;
841 elf_auxargs->flags = 0;
842 elf_auxargs->entry = entry;
843 elf_auxargs->trace = elf_trace;
844
845 imgp->auxargs = elf_auxargs;
846 imgp->interpreted = 0;
847
848 fail:
849 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
850 return (error);
851 }
852
853 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
854
855 int
856 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
857 {
858 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
859 Elf_Addr *base;
860 Elf_Addr *pos;
861
862 base = (Elf_Addr *)*stack_base;
863 pos = base + (imgp->argc + imgp->envc + 2);
864
865 if (args->trace) {
866 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
867 }
868 if (args->execfd != -1) {
869 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
870 }
871 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
872 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
873 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
874 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
875 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
876 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
877 AUXARGS_ENTRY(pos, AT_BASE, args->base);
878 AUXARGS_ENTRY(pos, AT_NULL, 0);
879
880 free(imgp->auxargs, M_TEMP);
881 imgp->auxargs = NULL;
882
883 base--;
884 suword(base, (long)imgp->argc);
885 *stack_base = (register_t *)base;
886 return (0);
887 }
888
889 /*
890 * Code for generating ELF core dumps.
891 */
892
893 typedef void (*segment_callback)(vm_map_entry_t, void *);
894
895 /* Closure for cb_put_phdr(). */
896 struct phdr_closure {
897 Elf_Phdr *phdr; /* Program header to fill in */
898 Elf_Off offset; /* Offset of segment in core file */
899 };
900
901 /* Closure for cb_size_segment(). */
902 struct sseg_closure {
903 int count; /* Count of writable segments. */
904 size_t size; /* Total size of all writable segments. */
905 };
906
907 static void cb_put_phdr(vm_map_entry_t, void *);
908 static void cb_size_segment(vm_map_entry_t, void *);
909 static void each_writable_segment(struct proc *, segment_callback, void *);
910 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
911 int, void *, size_t);
912 static void __elfN(puthdr)(struct proc *, void *, size_t *,
913 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int);
914 static void __elfN(putnote)(void *, size_t *, const char *, int,
915 const void *, size_t);
916
917 extern int osreldate;
918
919 int
920 __elfN(coredump)(td, vp, limit)
921 struct thread *td;
922 register struct vnode *vp;
923 off_t limit;
924 {
925 register struct proc *p = td->td_proc;
926 register struct ucred *cred = td->td_ucred;
927 int error = 0;
928 struct sseg_closure seginfo;
929 void *hdr;
930 size_t hdrsize;
931
932 /* Size the program segments. */
933 seginfo.count = 0;
934 seginfo.size = 0;
935 each_writable_segment(p, cb_size_segment, &seginfo);
936
937 /*
938 * Calculate the size of the core file header area by making
939 * a dry run of generating it. Nothing is written, but the
940 * size is calculated.
941 */
942 hdrsize = 0;
943 __elfN(puthdr)((struct proc *)NULL, (void *)NULL, &hdrsize,
944 (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
945 (const prpsinfo_t *)NULL, seginfo.count);
946
947 if (hdrsize + seginfo.size >= limit)
948 return (EFAULT);
949
950 /*
951 * Allocate memory for building the header, fill it up,
952 * and write it out.
953 */
954 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
955 if (hdr == NULL) {
956 return (EINVAL);
957 }
958 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
959
960 /* Write the contents of all of the writable segments. */
961 if (error == 0) {
962 Elf_Phdr *php;
963 off_t offset;
964 int i;
965
966 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
967 offset = hdrsize;
968 for (i = 0; i < seginfo.count; i++) {
969 error = vn_rdwr_inchunks(UIO_WRITE, vp,
970 (caddr_t)(uintptr_t)php->p_vaddr,
971 php->p_filesz, offset, UIO_USERSPACE,
972 IO_UNIT | IO_DIRECT, cred, NOCRED, (int *)NULL,
973 curthread); /* XXXKSE */
974 if (error != 0)
975 break;
976 offset += php->p_filesz;
977 php++;
978 }
979 }
980 free(hdr, M_TEMP);
981
982 return (error);
983 }
984
985 /*
986 * A callback for each_writable_segment() to write out the segment's
987 * program header entry.
988 */
989 static void
990 cb_put_phdr(entry, closure)
991 vm_map_entry_t entry;
992 void *closure;
993 {
994 struct phdr_closure *phc = (struct phdr_closure *)closure;
995 Elf_Phdr *phdr = phc->phdr;
996
997 phc->offset = round_page(phc->offset);
998
999 phdr->p_type = PT_LOAD;
1000 phdr->p_offset = phc->offset;
1001 phdr->p_vaddr = entry->start;
1002 phdr->p_paddr = 0;
1003 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1004 phdr->p_align = PAGE_SIZE;
1005 phdr->p_flags = 0;
1006 if (entry->protection & VM_PROT_READ)
1007 phdr->p_flags |= PF_R;
1008 if (entry->protection & VM_PROT_WRITE)
1009 phdr->p_flags |= PF_W;
1010 if (entry->protection & VM_PROT_EXECUTE)
1011 phdr->p_flags |= PF_X;
1012
1013 phc->offset += phdr->p_filesz;
1014 phc->phdr++;
1015 }
1016
1017 /*
1018 * A callback for each_writable_segment() to gather information about
1019 * the number of segments and their total size.
1020 */
1021 static void
1022 cb_size_segment(entry, closure)
1023 vm_map_entry_t entry;
1024 void *closure;
1025 {
1026 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1027
1028 ssc->count++;
1029 ssc->size += entry->end - entry->start;
1030 }
1031
1032 /*
1033 * For each writable segment in the process's memory map, call the given
1034 * function with a pointer to the map entry and some arbitrary
1035 * caller-supplied data.
1036 */
1037 static void
1038 each_writable_segment(p, func, closure)
1039 struct proc *p;
1040 segment_callback func;
1041 void *closure;
1042 {
1043 vm_map_t map = &p->p_vmspace->vm_map;
1044 vm_map_entry_t entry;
1045
1046 for (entry = map->header.next; entry != &map->header;
1047 entry = entry->next) {
1048 vm_object_t obj;
1049
1050 /*
1051 * Don't dump inaccessible mappings, deal with legacy
1052 * coredump mode.
1053 *
1054 * Note that read-only segments related to the elf binary
1055 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1056 * need to arbitrarily ignore such segments.
1057 */
1058 if (elf_legacy_coredump) {
1059 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1060 continue;
1061 } else {
1062 if ((entry->protection & VM_PROT_ALL) == 0)
1063 continue;
1064 }
1065
1066 /*
1067 * Dont include memory segment in the coredump if
1068 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1069 * madvise(2). Do not dump submaps (i.e. parts of the
1070 * kernel map).
1071 */
1072 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1073 continue;
1074
1075 if ((obj = entry->object.vm_object) == NULL)
1076 continue;
1077
1078 /* Find the deepest backing object. */
1079 while (obj->backing_object != NULL)
1080 obj = obj->backing_object;
1081
1082 /* Ignore memory-mapped devices and such things. */
1083 if (obj->type != OBJT_DEFAULT &&
1084 obj->type != OBJT_SWAP &&
1085 obj->type != OBJT_VNODE)
1086 continue;
1087
1088 (*func)(entry, closure);
1089 }
1090 }
1091
1092 /*
1093 * Write the core file header to the file, including padding up to
1094 * the page boundary.
1095 */
1096 static int
1097 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1098 struct thread *td;
1099 struct vnode *vp;
1100 struct ucred *cred;
1101 int numsegs;
1102 size_t hdrsize;
1103 void *hdr;
1104 {
1105 struct {
1106 prstatus_t status;
1107 prfpregset_t fpregset;
1108 prpsinfo_t psinfo;
1109 } *tempdata;
1110 struct proc *p = td->td_proc;
1111 size_t off;
1112 prstatus_t *status;
1113 prfpregset_t *fpregset;
1114 prpsinfo_t *psinfo;
1115
1116 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
1117 status = &tempdata->status;
1118 fpregset = &tempdata->fpregset;
1119 psinfo = &tempdata->psinfo;
1120
1121 /* Gather the information for the header. */
1122 status->pr_version = PRSTATUS_VERSION;
1123 status->pr_statussz = sizeof(prstatus_t);
1124 status->pr_gregsetsz = sizeof(gregset_t);
1125 status->pr_fpregsetsz = sizeof(fpregset_t);
1126 status->pr_osreldate = osreldate;
1127 status->pr_cursig = p->p_sig;
1128 status->pr_pid = p->p_pid;
1129 fill_regs(td, &status->pr_reg);
1130
1131 fill_fpregs(td, fpregset);
1132
1133 psinfo->pr_version = PRPSINFO_VERSION;
1134 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1135 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1136
1137 /* XXX - We don't fill in the command line arguments properly yet. */
1138 strlcpy(psinfo->pr_psargs, p->p_comm, sizeof(psinfo->pr_psargs));
1139
1140 /* Fill in the header. */
1141 bzero(hdr, hdrsize);
1142 off = 0;
1143 __elfN(puthdr)(p, hdr, &off, status, fpregset, psinfo, numsegs);
1144
1145 free(tempdata, M_TEMP);
1146
1147 /* Write it to the core file. */
1148 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1149 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1150 td)); /* XXXKSE */
1151 }
1152
1153 static void
1154 __elfN(puthdr)(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
1155 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
1156 {
1157 size_t ehoff;
1158 size_t phoff;
1159 size_t noteoff;
1160 size_t notesz;
1161
1162 ehoff = *off;
1163 *off += sizeof(Elf_Ehdr);
1164
1165 phoff = *off;
1166 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1167
1168 noteoff = *off;
1169 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1170 sizeof *status);
1171 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1172 sizeof *fpregset);
1173 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1174 sizeof *psinfo);
1175 notesz = *off - noteoff;
1176
1177 /* Align up to a page boundary for the program segments. */
1178 *off = round_page(*off);
1179
1180 if (dst != NULL) {
1181 Elf_Ehdr *ehdr;
1182 Elf_Phdr *phdr;
1183 struct phdr_closure phc;
1184
1185 /*
1186 * Fill in the ELF header.
1187 */
1188 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1189 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1190 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1191 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1192 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1193 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1194 ehdr->e_ident[EI_DATA] = ELF_DATA;
1195 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1196 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1197 ehdr->e_ident[EI_ABIVERSION] = 0;
1198 ehdr->e_ident[EI_PAD] = 0;
1199 ehdr->e_type = ET_CORE;
1200 ehdr->e_machine = ELF_ARCH;
1201 ehdr->e_version = EV_CURRENT;
1202 ehdr->e_entry = 0;
1203 ehdr->e_phoff = phoff;
1204 ehdr->e_flags = 0;
1205 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1206 ehdr->e_phentsize = sizeof(Elf_Phdr);
1207 ehdr->e_phnum = numsegs + 1;
1208 ehdr->e_shentsize = sizeof(Elf_Shdr);
1209 ehdr->e_shnum = 0;
1210 ehdr->e_shstrndx = SHN_UNDEF;
1211
1212 /*
1213 * Fill in the program header entries.
1214 */
1215 phdr = (Elf_Phdr *)((char *)dst + phoff);
1216
1217 /* The note segement. */
1218 phdr->p_type = PT_NOTE;
1219 phdr->p_offset = noteoff;
1220 phdr->p_vaddr = 0;
1221 phdr->p_paddr = 0;
1222 phdr->p_filesz = notesz;
1223 phdr->p_memsz = 0;
1224 phdr->p_flags = 0;
1225 phdr->p_align = 0;
1226 phdr++;
1227
1228 /* All the writable segments from the program. */
1229 phc.phdr = phdr;
1230 phc.offset = *off;
1231 each_writable_segment(p, cb_put_phdr, &phc);
1232 }
1233 }
1234
1235 static void
1236 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1237 const void *desc, size_t descsz)
1238 {
1239 Elf_Note note;
1240
1241 note.n_namesz = strlen(name) + 1;
1242 note.n_descsz = descsz;
1243 note.n_type = type;
1244 if (dst != NULL)
1245 bcopy(¬e, (char *)dst + *off, sizeof note);
1246 *off += sizeof note;
1247 if (dst != NULL)
1248 bcopy(name, (char *)dst + *off, note.n_namesz);
1249 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1250 if (dst != NULL)
1251 bcopy(desc, (char *)dst + *off, note.n_descsz);
1252 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1253 }
1254
1255 /*
1256 * Tell kern_execve.c about it, with a little help from the linker.
1257 */
1258 static struct execsw __elfN(execsw) = {
1259 __CONCAT(exec_, __elfN(imgact)),
1260 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1261 };
1262 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
Cache object: 889a6d030b299e11a85cbfd643ba21ae
|