FreeBSD/Linux Kernel Cross Reference
sys/kern/imgact_elf.c
1 /*-
2 * Copyright (c) 2000 David O'Brien
3 * Copyright (c) 1995-1996 Søren Schmidt
4 * Copyright (c) 1996 Peter Wemm
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer
12 * in this position and unchanged.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote products
17 * derived from this software without specific prior written permission
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD: releng/6.2/sys/kern/imgact_elf.c 164286 2006-11-14 20:42:41Z cvs2svn $");
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)(const Elf_Ehdr *hdr,
82 const char *interp);
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
90 SYSCTL_NODE(_kern, OID_AUTO, __CONCAT(elf, __ELF_WORD_SIZE), CTLFLAG_RW, 0,
91 "");
92
93 int __elfN(fallback_brand) = -1;
94 SYSCTL_INT(__CONCAT(_kern_elf, __ELF_WORD_SIZE), OID_AUTO,
95 fallback_brand, CTLFLAG_RW, &__elfN(fallback_brand), 0,
96 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE)) " brand of last resort");
97 TUNABLE_INT("kern.elf" __XSTRING(__ELF_WORD_SIZE) ".fallback_brand",
98 &__elfN(fallback_brand));
99
100 static int elf_trace = 0;
101 SYSCTL_INT(_debug, OID_AUTO, __elfN(trace), CTLFLAG_RW, &elf_trace, 0, "");
102
103 static int elf_legacy_coredump = 0;
104 SYSCTL_INT(_debug, OID_AUTO, __elfN(legacy_coredump), CTLFLAG_RW,
105 &elf_legacy_coredump, 0, "");
106
107 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
108
109 int
110 __elfN(insert_brand_entry)(Elf_Brandinfo *entry)
111 {
112 int i;
113
114 for (i = 0; i < MAX_BRANDS; i++) {
115 if (elf_brand_list[i] == NULL) {
116 elf_brand_list[i] = entry;
117 break;
118 }
119 }
120 if (i == MAX_BRANDS)
121 return (-1);
122 return (0);
123 }
124
125 int
126 __elfN(remove_brand_entry)(Elf_Brandinfo *entry)
127 {
128 int i;
129
130 for (i = 0; i < MAX_BRANDS; i++) {
131 if (elf_brand_list[i] == entry) {
132 elf_brand_list[i] = NULL;
133 break;
134 }
135 }
136 if (i == MAX_BRANDS)
137 return (-1);
138 return (0);
139 }
140
141 int
142 __elfN(brand_inuse)(Elf_Brandinfo *entry)
143 {
144 struct proc *p;
145 int rval = FALSE;
146
147 sx_slock(&allproc_lock);
148 LIST_FOREACH(p, &allproc, p_list) {
149 if (p->p_sysent == entry->sysvec) {
150 rval = TRUE;
151 break;
152 }
153 }
154 sx_sunlock(&allproc_lock);
155
156 return (rval);
157 }
158
159 static Elf_Brandinfo *
160 __elfN(get_brandinfo)(const Elf_Ehdr *hdr, const char *interp)
161 {
162 Elf_Brandinfo *bi;
163 int i;
164
165 /*
166 * We support three types of branding -- (1) the ELF EI_OSABI field
167 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
168 * branding w/in the ELF header, and (3) path of the `interp_path'
169 * field. We should also look for an ".note.ABI-tag" ELF section now
170 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
171 */
172
173 /* If the executable has a brand, search for it in the brand list. */
174 for (i = 0; i < MAX_BRANDS; i++) {
175 bi = elf_brand_list[i];
176 if (bi != NULL && hdr->e_machine == bi->machine &&
177 (hdr->e_ident[EI_OSABI] == bi->brand ||
178 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
179 bi->compat_3_brand, strlen(bi->compat_3_brand)) == 0))
180 return (bi);
181 }
182
183 /* Lacking a known brand, search for a recognized interpreter. */
184 if (interp != NULL) {
185 for (i = 0; i < MAX_BRANDS; i++) {
186 bi = elf_brand_list[i];
187 if (bi != NULL && hdr->e_machine == bi->machine &&
188 strcmp(interp, bi->interp_path) == 0)
189 return (bi);
190 }
191 }
192
193 /* Lacking a recognized interpreter, try the default brand */
194 for (i = 0; i < MAX_BRANDS; i++) {
195 bi = elf_brand_list[i];
196 if (bi != NULL && hdr->e_machine == bi->machine &&
197 __elfN(fallback_brand) == bi->brand)
198 return (bi);
199 }
200 return (NULL);
201 }
202
203 static int
204 __elfN(check_header)(const Elf_Ehdr *hdr)
205 {
206 Elf_Brandinfo *bi;
207 int i;
208
209 if (!IS_ELF(*hdr) ||
210 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
211 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
212 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
213 hdr->e_phentsize != sizeof(Elf_Phdr) ||
214 hdr->e_version != ELF_TARG_VER)
215 return (ENOEXEC);
216
217 /*
218 * Make sure we have at least one brand for this machine.
219 */
220
221 for (i = 0; i < MAX_BRANDS; i++) {
222 bi = elf_brand_list[i];
223 if (bi != NULL && bi->machine == hdr->e_machine)
224 break;
225 }
226 if (i == MAX_BRANDS)
227 return (ENOEXEC);
228
229 return (0);
230 }
231
232 static int
233 __elfN(map_partial)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
234 vm_offset_t start, vm_offset_t end, vm_prot_t prot)
235 {
236 struct sf_buf *sf;
237 int error;
238 vm_offset_t off;
239
240 /*
241 * Create the page if it doesn't exist yet. Ignore errors.
242 */
243 vm_map_lock(map);
244 vm_map_insert(map, NULL, 0, trunc_page(start), round_page(end),
245 VM_PROT_ALL, VM_PROT_ALL, 0);
246 vm_map_unlock(map);
247
248 /*
249 * Find the page from the underlying object.
250 */
251 if (object) {
252 sf = vm_imgact_map_page(object, offset);
253 if (sf == NULL)
254 return (KERN_FAILURE);
255 off = offset - trunc_page(offset);
256 error = copyout((caddr_t)sf_buf_kva(sf) + off, (caddr_t)start,
257 end - start);
258 vm_imgact_unmap_page(sf);
259 if (error) {
260 return (KERN_FAILURE);
261 }
262 }
263
264 return (KERN_SUCCESS);
265 }
266
267 static int
268 __elfN(map_insert)(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
269 vm_offset_t start, vm_offset_t end, vm_prot_t prot, int cow)
270 {
271 struct sf_buf *sf;
272 vm_offset_t off;
273 vm_size_t sz;
274 int error, rv;
275
276 if (start != trunc_page(start)) {
277 rv = __elfN(map_partial)(map, object, offset, start,
278 round_page(start), prot);
279 if (rv)
280 return (rv);
281 offset += round_page(start) - start;
282 start = round_page(start);
283 }
284 if (end != round_page(end)) {
285 rv = __elfN(map_partial)(map, object, offset +
286 trunc_page(end) - start, trunc_page(end), end, prot);
287 if (rv)
288 return (rv);
289 end = trunc_page(end);
290 }
291 if (end > start) {
292 if (offset & PAGE_MASK) {
293 /*
294 * The mapping is not page aligned. This means we have
295 * to copy the data. Sigh.
296 */
297 rv = vm_map_find(map, NULL, 0, &start, end - start,
298 FALSE, prot | VM_PROT_WRITE, VM_PROT_ALL, 0);
299 if (rv)
300 return (rv);
301 if (object == NULL)
302 return (KERN_SUCCESS);
303 for (; start < end; start += sz) {
304 sf = vm_imgact_map_page(object, offset);
305 if (sf == NULL)
306 return (KERN_FAILURE);
307 off = offset - trunc_page(offset);
308 sz = end - start;
309 if (sz > PAGE_SIZE - off)
310 sz = PAGE_SIZE - off;
311 error = copyout((caddr_t)sf_buf_kva(sf) + off,
312 (caddr_t)start, sz);
313 vm_imgact_unmap_page(sf);
314 if (error) {
315 return (KERN_FAILURE);
316 }
317 offset += sz;
318 }
319 rv = KERN_SUCCESS;
320 } else {
321 vm_object_reference(object);
322 vm_map_lock(map);
323 rv = vm_map_insert(map, object, offset, start, end,
324 prot, VM_PROT_ALL, cow);
325 vm_map_unlock(map);
326 if (rv != KERN_SUCCESS)
327 vm_object_deallocate(object);
328 }
329 return (rv);
330 } else {
331 return (KERN_SUCCESS);
332 }
333 }
334
335 static int
336 __elfN(load_section)(struct vmspace *vmspace,
337 vm_object_t object, vm_offset_t offset,
338 caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot,
339 size_t pagesize)
340 {
341 struct sf_buf *sf;
342 size_t map_len;
343 vm_offset_t map_addr;
344 int error, rv, cow;
345 size_t copy_len;
346 vm_offset_t file_addr;
347
348 /*
349 * It's necessary to fail if the filsz + offset taken from the
350 * header is greater than the actual file pager object's size.
351 * If we were to allow this, then the vm_map_find() below would
352 * walk right off the end of the file object and into the ether.
353 *
354 * While I'm here, might as well check for something else that
355 * is invalid: filsz cannot be greater than memsz.
356 */
357 if ((off_t)filsz + offset > object->un_pager.vnp.vnp_size ||
358 filsz > memsz) {
359 uprintf("elf_load_section: truncated ELF file\n");
360 return (ENOEXEC);
361 }
362
363 #define trunc_page_ps(va, ps) ((va) & ~(ps - 1))
364 #define round_page_ps(va, ps) (((va) + (ps - 1)) & ~(ps - 1))
365
366 map_addr = trunc_page_ps((vm_offset_t)vmaddr, pagesize);
367 file_addr = trunc_page_ps(offset, pagesize);
368
369 /*
370 * We have two choices. We can either clear the data in the last page
371 * of an oversized mapping, or we can start the anon mapping a page
372 * early and copy the initialized data into that first page. We
373 * choose the second..
374 */
375 if (memsz > filsz)
376 map_len = trunc_page_ps(offset + filsz, pagesize) - file_addr;
377 else
378 map_len = round_page_ps(offset + filsz, pagesize) - file_addr;
379
380 if (map_len != 0) {
381 /* cow flags: don't dump readonly sections in core */
382 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
383 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
384
385 rv = __elfN(map_insert)(&vmspace->vm_map,
386 object,
387 file_addr, /* file offset */
388 map_addr, /* virtual start */
389 map_addr + map_len,/* virtual end */
390 prot,
391 cow);
392 if (rv != KERN_SUCCESS)
393 return (EINVAL);
394
395 /* we can stop now if we've covered it all */
396 if (memsz == filsz) {
397 return (0);
398 }
399 }
400
401
402 /*
403 * We have to get the remaining bit of the file into the first part
404 * of the oversized map segment. This is normally because the .data
405 * segment in the file is extended to provide bss. It's a neat idea
406 * to try and save a page, but it's a pain in the behind to implement.
407 */
408 copy_len = (offset + filsz) - trunc_page_ps(offset + filsz, pagesize);
409 map_addr = trunc_page_ps((vm_offset_t)vmaddr + filsz, pagesize);
410 map_len = round_page_ps((vm_offset_t)vmaddr + memsz, pagesize) -
411 map_addr;
412
413 /* This had damn well better be true! */
414 if (map_len != 0) {
415 rv = __elfN(map_insert)(&vmspace->vm_map, NULL, 0, map_addr,
416 map_addr + map_len, VM_PROT_ALL, 0);
417 if (rv != KERN_SUCCESS) {
418 return (EINVAL);
419 }
420 }
421
422 if (copy_len != 0) {
423 vm_offset_t off;
424
425 sf = vm_imgact_map_page(object, offset + filsz);
426 if (sf == NULL)
427 return (EIO);
428
429 /* send the page fragment to user space */
430 off = trunc_page_ps(offset + filsz, pagesize) -
431 trunc_page(offset + filsz);
432 error = copyout((caddr_t)sf_buf_kva(sf) + off,
433 (caddr_t)map_addr, copy_len);
434 vm_imgact_unmap_page(sf);
435 if (error) {
436 return (error);
437 }
438 }
439
440 /*
441 * set it to the specified protection.
442 * XXX had better undo the damage from pasting over the cracks here!
443 */
444 vm_map_protect(&vmspace->vm_map, trunc_page(map_addr),
445 round_page(map_addr + map_len), prot, FALSE);
446
447 return (0);
448 }
449
450 /*
451 * Load the file "file" into memory. It may be either a shared object
452 * or an executable.
453 *
454 * The "addr" reference parameter is in/out. On entry, it specifies
455 * the address where a shared object should be loaded. If the file is
456 * an executable, this value is ignored. On exit, "addr" specifies
457 * where the file was actually loaded.
458 *
459 * The "entry" reference parameter is out only. On exit, it specifies
460 * the entry point for the loaded file.
461 */
462 static int
463 __elfN(load_file)(struct proc *p, const char *file, u_long *addr,
464 u_long *entry, size_t pagesize)
465 {
466 struct {
467 struct nameidata nd;
468 struct vattr attr;
469 struct image_params image_params;
470 } *tempdata;
471 const Elf_Ehdr *hdr = NULL;
472 const Elf_Phdr *phdr = NULL;
473 struct nameidata *nd;
474 struct vmspace *vmspace = p->p_vmspace;
475 struct vattr *attr;
476 struct image_params *imgp;
477 vm_prot_t prot;
478 u_long rbase;
479 u_long base_addr = 0;
480 int vfslocked, error, i, numsegs;
481
482 if (curthread->td_proc != p)
483 panic("elf_load_file - thread"); /* XXXKSE DIAGNOSTIC */
484
485 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
486 nd = &tempdata->nd;
487 attr = &tempdata->attr;
488 imgp = &tempdata->image_params;
489
490 /*
491 * Initialize part of the common data
492 */
493 imgp->proc = p;
494 imgp->attr = attr;
495 imgp->firstpage = NULL;
496 imgp->image_header = NULL;
497 imgp->object = NULL;
498 imgp->execlabel = NULL;
499
500 /* XXXKSE */
501 NDINIT(nd, LOOKUP, MPSAFE|LOCKLEAF|FOLLOW, UIO_SYSSPACE, file,
502 curthread);
503 vfslocked = 0;
504 if ((error = namei(nd)) != 0) {
505 nd->ni_vp = NULL;
506 goto fail;
507 }
508 vfslocked = NDHASGIANT(nd);
509 NDFREE(nd, NDF_ONLY_PNBUF);
510 imgp->vp = nd->ni_vp;
511
512 /*
513 * Check permissions, modes, uid, etc on the file, and "open" it.
514 */
515 error = exec_check_permissions(imgp);
516 if (error)
517 goto fail;
518
519 error = exec_map_first_page(imgp);
520 if (error)
521 goto fail;
522
523 /*
524 * Also make certain that the interpreter stays the same, so set
525 * its VV_TEXT flag, too.
526 */
527 nd->ni_vp->v_vflag |= VV_TEXT;
528
529 imgp->object = nd->ni_vp->v_object;
530
531 hdr = (const Elf_Ehdr *)imgp->image_header;
532 if ((error = __elfN(check_header)(hdr)) != 0)
533 goto fail;
534 if (hdr->e_type == ET_DYN)
535 rbase = *addr;
536 else if (hdr->e_type == ET_EXEC)
537 rbase = 0;
538 else {
539 error = ENOEXEC;
540 goto fail;
541 }
542
543 /* Only support headers that fit within first page for now */
544 /* (multiplication of two Elf_Half fields will not overflow) */
545 if ((hdr->e_phoff > PAGE_SIZE) ||
546 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
547 error = ENOEXEC;
548 goto fail;
549 }
550
551 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
552
553 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
554 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
555 prot = 0;
556 if (phdr[i].p_flags & PF_X)
557 prot |= VM_PROT_EXECUTE;
558 if (phdr[i].p_flags & PF_W)
559 prot |= VM_PROT_WRITE;
560 if (phdr[i].p_flags & PF_R)
561 prot |= VM_PROT_READ;
562
563 if ((error = __elfN(load_section)(vmspace,
564 imgp->object, phdr[i].p_offset,
565 (caddr_t)(uintptr_t)phdr[i].p_vaddr + rbase,
566 phdr[i].p_memsz, phdr[i].p_filesz, prot,
567 pagesize)) != 0)
568 goto fail;
569 /*
570 * Establish the base address if this is the
571 * first segment.
572 */
573 if (numsegs == 0)
574 base_addr = trunc_page(phdr[i].p_vaddr +
575 rbase);
576 numsegs++;
577 }
578 }
579 *addr = base_addr;
580 *entry = (unsigned long)hdr->e_entry + rbase;
581
582 fail:
583 if (imgp->firstpage)
584 exec_unmap_first_page(imgp);
585
586 if (nd->ni_vp)
587 vput(nd->ni_vp);
588
589 VFS_UNLOCK_GIANT(vfslocked);
590 free(tempdata, M_TEMP);
591
592 return (error);
593 }
594
595 static int
596 __CONCAT(exec_, __elfN(imgact))(struct image_params *imgp)
597 {
598 const Elf_Ehdr *hdr = (const Elf_Ehdr *)imgp->image_header;
599 const Elf_Phdr *phdr;
600 Elf_Auxargs *elf_auxargs;
601 struct vmspace *vmspace;
602 vm_prot_t prot;
603 u_long text_size = 0, data_size = 0, total_size = 0;
604 u_long text_addr = 0, data_addr = 0;
605 u_long seg_size, seg_addr;
606 u_long addr, entry = 0, proghdr = 0;
607 int error = 0, i;
608 const char *interp = NULL;
609 Elf_Brandinfo *brand_info;
610 char *path;
611 struct thread *td = curthread;
612 struct sysentvec *sv;
613
614 /*
615 * Do we have a valid ELF header ?
616 *
617 * Only allow ET_EXEC & ET_DYN here, reject ET_DYN later
618 * if particular brand doesn't support it.
619 */
620 if (__elfN(check_header)(hdr) != 0 ||
621 (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN))
622 return (-1);
623
624 /*
625 * From here on down, we return an errno, not -1, as we've
626 * detected an ELF file.
627 */
628
629 if ((hdr->e_phoff > PAGE_SIZE) ||
630 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
631 /* Only support headers in first page for now */
632 return (ENOEXEC);
633 }
634 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
635 for (i = 0; i < hdr->e_phnum; i++) {
636 if (phdr[i].p_type == PT_INTERP) {
637 /* Path to interpreter */
638 if (phdr[i].p_filesz > MAXPATHLEN ||
639 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
640 return (ENOEXEC);
641 interp = imgp->image_header + phdr[i].p_offset;
642 break;
643 }
644 }
645
646 brand_info = __elfN(get_brandinfo)(hdr, interp);
647 if (brand_info == NULL) {
648 uprintf("ELF binary type \"%u\" not known.\n",
649 hdr->e_ident[EI_OSABI]);
650 return (ENOEXEC);
651 }
652 if (hdr->e_type == ET_DYN && brand_info->brand != ELFOSABI_LINUX)
653 return (ENOEXEC);
654 sv = brand_info->sysvec;
655 if (interp != NULL && brand_info->interp_newpath != NULL)
656 interp = brand_info->interp_newpath;
657
658 /*
659 * Avoid a possible deadlock if the current address space is destroyed
660 * and that address space maps the locked vnode. In the common case,
661 * the locked vnode's v_usecount is decremented but remains greater
662 * than zero. Consequently, the vnode lock is not needed by vrele().
663 * However, in cases where the vnode lock is external, such as nullfs,
664 * v_usecount may become zero.
665 */
666 VOP_UNLOCK(imgp->vp, 0, td);
667
668 exec_new_vmspace(imgp, sv);
669
670 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
671
672 vmspace = imgp->proc->p_vmspace;
673
674 for (i = 0; i < hdr->e_phnum; i++) {
675 switch (phdr[i].p_type) {
676 case PT_LOAD: /* Loadable segment */
677 prot = 0;
678 if (phdr[i].p_flags & PF_X)
679 prot |= VM_PROT_EXECUTE;
680 if (phdr[i].p_flags & PF_W)
681 prot |= VM_PROT_WRITE;
682 if (phdr[i].p_flags & PF_R)
683 prot |= VM_PROT_READ;
684
685 #if defined(__ia64__) && __ELF_WORD_SIZE == 32 && defined(IA32_ME_HARDER)
686 /*
687 * Some x86 binaries assume read == executable,
688 * notably the M3 runtime and therefore cvsup
689 */
690 if (prot & VM_PROT_READ)
691 prot |= VM_PROT_EXECUTE;
692 #endif
693
694 if ((error = __elfN(load_section)(vmspace,
695 imgp->object, phdr[i].p_offset,
696 (caddr_t)(uintptr_t)phdr[i].p_vaddr,
697 phdr[i].p_memsz, phdr[i].p_filesz, prot,
698 sv->sv_pagesize)) != 0)
699 return (error);
700
701 /*
702 * If this segment contains the program headers,
703 * remember their virtual address for the AT_PHDR
704 * aux entry. Static binaries don't usually include
705 * a PT_PHDR entry.
706 */
707 if (phdr[i].p_offset == 0 &&
708 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
709 <= phdr[i].p_filesz)
710 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
711
712 seg_addr = trunc_page(phdr[i].p_vaddr);
713 seg_size = round_page(phdr[i].p_memsz +
714 phdr[i].p_vaddr - seg_addr);
715
716 /*
717 * Is this .text or .data? We can't use
718 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
719 * alpha terribly and possibly does other bad
720 * things so we stick to the old way of figuring
721 * it out: If the segment contains the program
722 * entry point, it's a text segment, otherwise it
723 * is a data segment.
724 *
725 * Note that obreak() assumes that data_addr +
726 * data_size == end of data load area, and the ELF
727 * file format expects segments to be sorted by
728 * address. If multiple data segments exist, the
729 * last one will be used.
730 */
731 if (hdr->e_entry >= phdr[i].p_vaddr &&
732 hdr->e_entry < (phdr[i].p_vaddr +
733 phdr[i].p_memsz)) {
734 text_size = seg_size;
735 text_addr = seg_addr;
736 entry = (u_long)hdr->e_entry;
737 } else {
738 data_size = seg_size;
739 data_addr = seg_addr;
740 }
741 total_size += seg_size;
742 break;
743 case PT_PHDR: /* Program header table info */
744 proghdr = phdr[i].p_vaddr;
745 break;
746 default:
747 break;
748 }
749 }
750
751 if (data_addr == 0 && data_size == 0) {
752 data_addr = text_addr;
753 data_size = text_size;
754 }
755
756 /*
757 * Check limits. It should be safe to check the
758 * limits after loading the segments since we do
759 * not actually fault in all the segments pages.
760 */
761 PROC_LOCK(imgp->proc);
762 if (data_size > lim_cur(imgp->proc, RLIMIT_DATA) ||
763 text_size > maxtsiz ||
764 total_size > lim_cur(imgp->proc, RLIMIT_VMEM)) {
765 PROC_UNLOCK(imgp->proc);
766 return (ENOMEM);
767 }
768
769 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
770 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
771 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
772 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
773
774 /*
775 * We load the dynamic linker where a userland call
776 * to mmap(0, ...) would put it. The rationale behind this
777 * calculation is that it leaves room for the heap to grow to
778 * its maximum allowed size.
779 */
780 addr = round_page((vm_offset_t)imgp->proc->p_vmspace->vm_daddr +
781 lim_max(imgp->proc, RLIMIT_DATA));
782 PROC_UNLOCK(imgp->proc);
783
784 imgp->entry_addr = entry;
785
786 imgp->proc->p_sysent = sv;
787 if (interp != NULL) {
788 VOP_UNLOCK(imgp->vp, 0, td);
789 if (brand_info->emul_path != NULL &&
790 brand_info->emul_path[0] != '\0') {
791 path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
792 snprintf(path, MAXPATHLEN, "%s%s",
793 brand_info->emul_path, interp);
794 error = __elfN(load_file)(imgp->proc, path, &addr,
795 &imgp->entry_addr, sv->sv_pagesize);
796 free(path, M_TEMP);
797 if (error == 0)
798 interp = NULL;
799 }
800 if (interp != NULL) {
801 error = __elfN(load_file)(imgp->proc, interp, &addr,
802 &imgp->entry_addr, sv->sv_pagesize);
803 }
804 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
805 if (error != 0) {
806 uprintf("ELF interpreter %s not found\n", interp);
807 return (error);
808 }
809 }
810
811 /*
812 * Construct auxargs table (used by the fixup routine)
813 */
814 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
815 elf_auxargs->execfd = -1;
816 elf_auxargs->phdr = proghdr;
817 elf_auxargs->phent = hdr->e_phentsize;
818 elf_auxargs->phnum = hdr->e_phnum;
819 elf_auxargs->pagesz = PAGE_SIZE;
820 elf_auxargs->base = addr;
821 elf_auxargs->flags = 0;
822 elf_auxargs->entry = entry;
823 elf_auxargs->trace = elf_trace;
824
825 imgp->auxargs = elf_auxargs;
826 imgp->interpreted = 0;
827
828 return (error);
829 }
830
831 #define suword __CONCAT(suword, __ELF_WORD_SIZE)
832
833 int
834 __elfN(freebsd_fixup)(register_t **stack_base, struct image_params *imgp)
835 {
836 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
837 Elf_Addr *base;
838 Elf_Addr *pos;
839
840 base = (Elf_Addr *)*stack_base;
841 pos = base + (imgp->args->argc + imgp->args->envc + 2);
842
843 if (args->trace) {
844 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
845 }
846 if (args->execfd != -1) {
847 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
848 }
849 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
850 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
851 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
852 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
853 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
854 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
855 AUXARGS_ENTRY(pos, AT_BASE, args->base);
856 AUXARGS_ENTRY(pos, AT_NULL, 0);
857
858 free(imgp->auxargs, M_TEMP);
859 imgp->auxargs = NULL;
860
861 base--;
862 suword(base, (long)imgp->args->argc);
863 *stack_base = (register_t *)base;
864 return (0);
865 }
866
867 /*
868 * Code for generating ELF core dumps.
869 */
870
871 typedef void (*segment_callback)(vm_map_entry_t, void *);
872
873 /* Closure for cb_put_phdr(). */
874 struct phdr_closure {
875 Elf_Phdr *phdr; /* Program header to fill in */
876 Elf_Off offset; /* Offset of segment in core file */
877 };
878
879 /* Closure for cb_size_segment(). */
880 struct sseg_closure {
881 int count; /* Count of writable segments. */
882 size_t size; /* Total size of all writable segments. */
883 };
884
885 static void cb_put_phdr(vm_map_entry_t, void *);
886 static void cb_size_segment(vm_map_entry_t, void *);
887 static void each_writable_segment(struct thread *, segment_callback, void *);
888 static int __elfN(corehdr)(struct thread *, struct vnode *, struct ucred *,
889 int, void *, size_t);
890 static void __elfN(puthdr)(struct thread *, void *, size_t *, int);
891 static void __elfN(putnote)(void *, size_t *, const char *, int,
892 const void *, size_t);
893
894 extern int osreldate;
895
896 int
897 __elfN(coredump)(td, vp, limit)
898 struct thread *td;
899 struct vnode *vp;
900 off_t limit;
901 {
902 struct ucred *cred = td->td_ucred;
903 int error = 0;
904 struct sseg_closure seginfo;
905 void *hdr;
906 size_t hdrsize;
907
908 /* Size the program segments. */
909 seginfo.count = 0;
910 seginfo.size = 0;
911 each_writable_segment(td, cb_size_segment, &seginfo);
912
913 /*
914 * Calculate the size of the core file header area by making
915 * a dry run of generating it. Nothing is written, but the
916 * size is calculated.
917 */
918 hdrsize = 0;
919 __elfN(puthdr)(td, (void *)NULL, &hdrsize, seginfo.count);
920
921 if (hdrsize + seginfo.size >= limit)
922 return (EFAULT);
923
924 /*
925 * Allocate memory for building the header, fill it up,
926 * and write it out.
927 */
928 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
929 if (hdr == NULL) {
930 return (EINVAL);
931 }
932 error = __elfN(corehdr)(td, vp, cred, seginfo.count, hdr, hdrsize);
933
934 /* Write the contents of all of the writable segments. */
935 if (error == 0) {
936 Elf_Phdr *php;
937 off_t offset;
938 int i;
939
940 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
941 offset = hdrsize;
942 for (i = 0; i < seginfo.count; i++) {
943 error = vn_rdwr_inchunks(UIO_WRITE, vp,
944 (caddr_t)(uintptr_t)php->p_vaddr,
945 php->p_filesz, offset, UIO_USERSPACE,
946 IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
947 curthread); /* XXXKSE */
948 if (error != 0)
949 break;
950 offset += php->p_filesz;
951 php++;
952 }
953 }
954 free(hdr, M_TEMP);
955
956 return (error);
957 }
958
959 /*
960 * A callback for each_writable_segment() to write out the segment's
961 * program header entry.
962 */
963 static void
964 cb_put_phdr(entry, closure)
965 vm_map_entry_t entry;
966 void *closure;
967 {
968 struct phdr_closure *phc = (struct phdr_closure *)closure;
969 Elf_Phdr *phdr = phc->phdr;
970
971 phc->offset = round_page(phc->offset);
972
973 phdr->p_type = PT_LOAD;
974 phdr->p_offset = phc->offset;
975 phdr->p_vaddr = entry->start;
976 phdr->p_paddr = 0;
977 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
978 phdr->p_align = PAGE_SIZE;
979 phdr->p_flags = 0;
980 if (entry->protection & VM_PROT_READ)
981 phdr->p_flags |= PF_R;
982 if (entry->protection & VM_PROT_WRITE)
983 phdr->p_flags |= PF_W;
984 if (entry->protection & VM_PROT_EXECUTE)
985 phdr->p_flags |= PF_X;
986
987 phc->offset += phdr->p_filesz;
988 phc->phdr++;
989 }
990
991 /*
992 * A callback for each_writable_segment() to gather information about
993 * the number of segments and their total size.
994 */
995 static void
996 cb_size_segment(entry, closure)
997 vm_map_entry_t entry;
998 void *closure;
999 {
1000 struct sseg_closure *ssc = (struct sseg_closure *)closure;
1001
1002 ssc->count++;
1003 ssc->size += entry->end - entry->start;
1004 }
1005
1006 /*
1007 * For each writable segment in the process's memory map, call the given
1008 * function with a pointer to the map entry and some arbitrary
1009 * caller-supplied data.
1010 */
1011 static void
1012 each_writable_segment(td, func, closure)
1013 struct thread *td;
1014 segment_callback func;
1015 void *closure;
1016 {
1017 struct proc *p = td->td_proc;
1018 vm_map_t map = &p->p_vmspace->vm_map;
1019 vm_map_entry_t entry;
1020
1021 for (entry = map->header.next; entry != &map->header;
1022 entry = entry->next) {
1023 vm_object_t obj;
1024
1025 /*
1026 * Don't dump inaccessible mappings, deal with legacy
1027 * coredump mode.
1028 *
1029 * Note that read-only segments related to the elf binary
1030 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1031 * need to arbitrarily ignore such segments.
1032 */
1033 if (elf_legacy_coredump) {
1034 if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
1035 continue;
1036 } else {
1037 if ((entry->protection & VM_PROT_ALL) == 0)
1038 continue;
1039 }
1040
1041 /*
1042 * Dont include memory segment in the coredump if
1043 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1044 * madvise(2). Do not dump submaps (i.e. parts of the
1045 * kernel map).
1046 */
1047 if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
1048 continue;
1049
1050 if ((obj = entry->object.vm_object) == NULL)
1051 continue;
1052
1053 /* Find the deepest backing object. */
1054 while (obj->backing_object != NULL)
1055 obj = obj->backing_object;
1056
1057 /* Ignore memory-mapped devices and such things. */
1058 if (obj->type != OBJT_DEFAULT &&
1059 obj->type != OBJT_SWAP &&
1060 obj->type != OBJT_VNODE)
1061 continue;
1062
1063 (*func)(entry, closure);
1064 }
1065 }
1066
1067 /*
1068 * Write the core file header to the file, including padding up to
1069 * the page boundary.
1070 */
1071 static int
1072 __elfN(corehdr)(td, vp, cred, numsegs, hdr, hdrsize)
1073 struct thread *td;
1074 struct vnode *vp;
1075 struct ucred *cred;
1076 int numsegs;
1077 size_t hdrsize;
1078 void *hdr;
1079 {
1080 size_t off;
1081
1082 /* Fill in the header. */
1083 bzero(hdr, hdrsize);
1084 off = 0;
1085 __elfN(puthdr)(td, hdr, &off, numsegs);
1086
1087 /* Write it to the core file. */
1088 return (vn_rdwr_inchunks(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
1089 UIO_SYSSPACE, IO_UNIT | IO_DIRECT, cred, NOCRED, NULL,
1090 td)); /* XXXKSE */
1091 }
1092
1093 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1094 typedef struct prstatus32 elf_prstatus_t;
1095 typedef struct prpsinfo32 elf_prpsinfo_t;
1096 typedef struct fpreg32 elf_prfpregset_t;
1097 typedef struct fpreg32 elf_fpregset_t;
1098 typedef struct reg32 elf_gregset_t;
1099 #else
1100 typedef prstatus_t elf_prstatus_t;
1101 typedef prpsinfo_t elf_prpsinfo_t;
1102 typedef prfpregset_t elf_prfpregset_t;
1103 typedef prfpregset_t elf_fpregset_t;
1104 typedef gregset_t elf_gregset_t;
1105 #endif
1106
1107 static void
1108 __elfN(puthdr)(struct thread *td, void *dst, size_t *off, int numsegs)
1109 {
1110 struct {
1111 elf_prstatus_t status;
1112 elf_prfpregset_t fpregset;
1113 elf_prpsinfo_t psinfo;
1114 } *tempdata;
1115 elf_prstatus_t *status;
1116 elf_prfpregset_t *fpregset;
1117 elf_prpsinfo_t *psinfo;
1118 struct proc *p;
1119 struct thread *thr;
1120 size_t ehoff, noteoff, notesz, phoff;
1121
1122 p = td->td_proc;
1123
1124 ehoff = *off;
1125 *off += sizeof(Elf_Ehdr);
1126
1127 phoff = *off;
1128 *off += (numsegs + 1) * sizeof(Elf_Phdr);
1129
1130 noteoff = *off;
1131 /*
1132 * Don't allocate space for the notes if we're just calculating
1133 * the size of the header. We also don't collect the data.
1134 */
1135 if (dst != NULL) {
1136 tempdata = malloc(sizeof(*tempdata), M_TEMP, M_ZERO|M_WAITOK);
1137 status = &tempdata->status;
1138 fpregset = &tempdata->fpregset;
1139 psinfo = &tempdata->psinfo;
1140 } else {
1141 tempdata = NULL;
1142 status = NULL;
1143 fpregset = NULL;
1144 psinfo = NULL;
1145 }
1146
1147 if (dst != NULL) {
1148 psinfo->pr_version = PRPSINFO_VERSION;
1149 psinfo->pr_psinfosz = sizeof(elf_prpsinfo_t);
1150 strlcpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname));
1151 /*
1152 * XXX - We don't fill in the command line arguments properly
1153 * yet.
1154 */
1155 strlcpy(psinfo->pr_psargs, p->p_comm,
1156 sizeof(psinfo->pr_psargs));
1157 }
1158 __elfN(putnote)(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
1159 sizeof *psinfo);
1160
1161 /*
1162 * To have the debugger select the right thread (LWP) as the initial
1163 * thread, we dump the state of the thread passed to us in td first.
1164 * This is the thread that causes the core dump and thus likely to
1165 * be the right thread one wants to have selected in the debugger.
1166 */
1167 thr = td;
1168 while (thr != NULL) {
1169 if (dst != NULL) {
1170 status->pr_version = PRSTATUS_VERSION;
1171 status->pr_statussz = sizeof(elf_prstatus_t);
1172 status->pr_gregsetsz = sizeof(elf_gregset_t);
1173 status->pr_fpregsetsz = sizeof(elf_fpregset_t);
1174 status->pr_osreldate = osreldate;
1175 status->pr_cursig = p->p_sig;
1176 status->pr_pid = thr->td_tid;
1177 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1178 fill_regs32(thr, &status->pr_reg);
1179 fill_fpregs32(thr, fpregset);
1180 #else
1181 fill_regs(thr, &status->pr_reg);
1182 fill_fpregs(thr, fpregset);
1183 #endif
1184 }
1185 __elfN(putnote)(dst, off, "FreeBSD", NT_PRSTATUS, status,
1186 sizeof *status);
1187 __elfN(putnote)(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
1188 sizeof *fpregset);
1189 /*
1190 * Allow for MD specific notes, as well as any MD
1191 * specific preparations for writing MI notes.
1192 */
1193 __elfN(dump_thread)(thr, dst, off);
1194
1195 thr = (thr == td) ? TAILQ_FIRST(&p->p_threads) :
1196 TAILQ_NEXT(thr, td_plist);
1197 if (thr == td)
1198 thr = TAILQ_NEXT(thr, td_plist);
1199 }
1200
1201 notesz = *off - noteoff;
1202
1203 if (dst != NULL)
1204 free(tempdata, M_TEMP);
1205
1206 /* Align up to a page boundary for the program segments. */
1207 *off = round_page(*off);
1208
1209 if (dst != NULL) {
1210 Elf_Ehdr *ehdr;
1211 Elf_Phdr *phdr;
1212 struct phdr_closure phc;
1213
1214 /*
1215 * Fill in the ELF header.
1216 */
1217 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
1218 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1219 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1220 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1221 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1222 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1223 ehdr->e_ident[EI_DATA] = ELF_DATA;
1224 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1225 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1226 ehdr->e_ident[EI_ABIVERSION] = 0;
1227 ehdr->e_ident[EI_PAD] = 0;
1228 ehdr->e_type = ET_CORE;
1229 #if defined(COMPAT_IA32) && __ELF_WORD_SIZE == 32
1230 ehdr->e_machine = EM_386;
1231 #else
1232 ehdr->e_machine = ELF_ARCH;
1233 #endif
1234 ehdr->e_version = EV_CURRENT;
1235 ehdr->e_entry = 0;
1236 ehdr->e_phoff = phoff;
1237 ehdr->e_flags = 0;
1238 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1239 ehdr->e_phentsize = sizeof(Elf_Phdr);
1240 ehdr->e_phnum = numsegs + 1;
1241 ehdr->e_shentsize = sizeof(Elf_Shdr);
1242 ehdr->e_shnum = 0;
1243 ehdr->e_shstrndx = SHN_UNDEF;
1244
1245 /*
1246 * Fill in the program header entries.
1247 */
1248 phdr = (Elf_Phdr *)((char *)dst + phoff);
1249
1250 /* The note segement. */
1251 phdr->p_type = PT_NOTE;
1252 phdr->p_offset = noteoff;
1253 phdr->p_vaddr = 0;
1254 phdr->p_paddr = 0;
1255 phdr->p_filesz = notesz;
1256 phdr->p_memsz = 0;
1257 phdr->p_flags = 0;
1258 phdr->p_align = 0;
1259 phdr++;
1260
1261 /* All the writable segments from the program. */
1262 phc.phdr = phdr;
1263 phc.offset = *off;
1264 each_writable_segment(td, cb_put_phdr, &phc);
1265 }
1266 }
1267
1268 static void
1269 __elfN(putnote)(void *dst, size_t *off, const char *name, int type,
1270 const void *desc, size_t descsz)
1271 {
1272 Elf_Note note;
1273
1274 note.n_namesz = strlen(name) + 1;
1275 note.n_descsz = descsz;
1276 note.n_type = type;
1277 if (dst != NULL)
1278 bcopy(¬e, (char *)dst + *off, sizeof note);
1279 *off += sizeof note;
1280 if (dst != NULL)
1281 bcopy(name, (char *)dst + *off, note.n_namesz);
1282 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1283 if (dst != NULL)
1284 bcopy(desc, (char *)dst + *off, note.n_descsz);
1285 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1286 }
1287
1288 /*
1289 * Tell kern_execve.c about it, with a little help from the linker.
1290 */
1291 static struct execsw __elfN(execsw) = {
1292 __CONCAT(exec_, __elfN(imgact)),
1293 __XSTRING(__CONCAT(ELF, __ELF_WORD_SIZE))
1294 };
1295 EXEC_SET(__CONCAT(elf, __ELF_WORD_SIZE), __elfN(execsw));
Cache object: 0a610262706312e10de2447e4b5a34e9
|