Deleted Added
full compact
imgact_elf.c (50477) imgact_elf.c (50717)
1/*-
2 * Copyright (c) 1995-1996 S�ren Schmidt
3 * Copyright (c) 1996 Peter Wemm
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer
11 * in this position and unchanged.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 * derived from this software withough specific prior written permission
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
1/*-
2 * Copyright (c) 1995-1996 S�ren Schmidt
3 * Copyright (c) 1996 Peter Wemm
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer
11 * in this position and unchanged.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote products
16 * derived from this software withough specific prior written permission
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
29 * $FreeBSD: head/sys/kern/imgact_elf.c 50477 1999-08-28 01:08:13Z peter $
29 * $FreeBSD: head/sys/kern/imgact_elf.c 50717 1999-09-01 00:29:56Z julian $
30 */
31
32#include "opt_rlimit.h"
33
34#include <sys/param.h>
35#include <sys/acct.h>
36#include <sys/exec.h>
37#include <sys/fcntl.h>
38#include <sys/imgact.h>
39#include <sys/imgact_elf.h>
40#include <sys/kernel.h>
41#include <sys/malloc.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/signalvar.h>
49#include <sys/stat.h>
50#include <sys/syscall.h>
51#include <sys/sysctl.h>
52#include <sys/sysent.h>
53#include <sys/systm.h>
54#include <sys/vnode.h>
55
56#include <vm/vm.h>
57#include <vm/vm_kern.h>
58#include <vm/vm_param.h>
59#include <vm/pmap.h>
60#include <sys/lock.h>
61#include <vm/vm_map.h>
62#include <vm/vm_object.h>
63#include <vm/vm_prot.h>
64#include <vm/vm_extern.h>
65
66#include <machine/elf.h>
67#include <machine/md_var.h>
68
69__ElfType(Brandinfo);
70__ElfType(Auxargs);
71
72static int elf_check_header __P((const Elf_Ehdr *hdr));
73static int elf_freebsd_fixup __P((long **stack_base,
74 struct image_params *imgp));
75static int elf_load_file __P((struct proc *p, const char *file, u_long *addr,
76 u_long *entry));
77static int elf_load_section __P((struct proc *p,
78 struct vmspace *vmspace, struct vnode *vp,
79 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
80 vm_prot_t prot));
81static int exec_elf_imgact __P((struct image_params *imgp));
82
83static int elf_trace = 0;
84SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
85
86/*
87 * XXX Maximum length of an ELF brand (sysctl wants a statically-allocated
88 * buffer).
89 */
90#define MAXBRANDLEN 16
91
92static struct sysentvec elf_freebsd_sysvec = {
93 SYS_MAXSYSCALL,
94 sysent,
95 0,
96 0,
97 0,
98 0,
99 0,
100 0,
101 elf_freebsd_fixup,
102 sendsig,
103 sigcode,
104 &szsigcode,
105 0,
106 "FreeBSD ELF",
107 elf_coredump
108};
109
110static Elf_Brandinfo freebsd_brand_info = {
111 "FreeBSD",
112 "",
113 "/usr/libexec/ld-elf.so.1",
114 &elf_freebsd_sysvec
115 };
116static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
117 &freebsd_brand_info,
118 NULL, NULL, NULL,
119 NULL, NULL, NULL, NULL
120 };
121
122int
123elf_insert_brand_entry(Elf_Brandinfo *entry)
124{
125 int i;
126
127 for (i=1; i<MAX_BRANDS; i++) {
128 if (elf_brand_list[i] == NULL) {
129 elf_brand_list[i] = entry;
130 break;
131 }
132 }
133 if (i == MAX_BRANDS)
134 return -1;
135 return 0;
136}
137
138int
139elf_remove_brand_entry(Elf_Brandinfo *entry)
140{
141 int i;
142
143 for (i=1; i<MAX_BRANDS; i++) {
144 if (elf_brand_list[i] == entry) {
145 elf_brand_list[i] = NULL;
146 break;
147 }
148 }
149 if (i == MAX_BRANDS)
150 return -1;
151 return 0;
152}
153
154int
155elf_brand_inuse(Elf_Brandinfo *entry)
156{
157 struct proc *p;
158
159 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
160 if (p->p_sysent == entry->sysvec)
161 return TRUE;
162 }
163
164 return FALSE;
165}
166
167static int
168elf_check_header(const Elf_Ehdr *hdr)
169{
170 if (!IS_ELF(*hdr) ||
171 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
172 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
173 hdr->e_ident[EI_VERSION] != EV_CURRENT)
174 return ENOEXEC;
175
176 if (!ELF_MACHINE_OK(hdr->e_machine))
177 return ENOEXEC;
178
179 if (hdr->e_version != ELF_TARG_VER)
180 return ENOEXEC;
181
182 return 0;
183}
184
185static int
186elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
187{
188 size_t map_len;
189 vm_offset_t map_addr;
190 int error, rv;
191 size_t copy_len;
192 vm_object_t object;
193 vm_offset_t file_addr;
194 vm_offset_t data_buf = 0;
195
196 object = vp->v_object;
197 error = 0;
198
199 map_addr = trunc_page((vm_offset_t)vmaddr);
200 file_addr = trunc_page(offset);
201
202 /*
203 * We have two choices. We can either clear the data in the last page
204 * of an oversized mapping, or we can start the anon mapping a page
205 * early and copy the initialized data into that first page. We
206 * choose the second..
207 */
208 if (memsz > filsz)
209 map_len = trunc_page(offset+filsz) - file_addr;
210 else
211 map_len = round_page(offset+filsz) - file_addr;
212
213 if (map_len != 0) {
214 vm_object_reference(object);
215 vm_map_lock(&vmspace->vm_map);
216 rv = vm_map_insert(&vmspace->vm_map,
217 object,
218 file_addr, /* file offset */
219 map_addr, /* virtual start */
220 map_addr + map_len,/* virtual end */
221 prot,
222 VM_PROT_ALL,
223 MAP_COPY_ON_WRITE | MAP_PREFAULT);
224 vm_map_unlock(&vmspace->vm_map);
225 if (rv != KERN_SUCCESS)
226 return EINVAL;
227
228 /* we can stop now if we've covered it all */
229 if (memsz == filsz)
230 return 0;
231 }
232
233
234 /*
235 * We have to get the remaining bit of the file into the first part
236 * of the oversized map segment. This is normally because the .data
237 * segment in the file is extended to provide bss. It's a neat idea
238 * to try and save a page, but it's a pain in the behind to implement.
239 */
240 copy_len = (offset + filsz) - trunc_page(offset + filsz);
241 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
242 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
243
244 /* This had damn well better be true! */
245 if (map_len != 0) {
246 vm_map_lock(&vmspace->vm_map);
247 rv = vm_map_insert(&vmspace->vm_map, NULL, 0,
248 map_addr, map_addr + map_len,
249 VM_PROT_ALL, VM_PROT_ALL, 0);
250 vm_map_unlock(&vmspace->vm_map);
251 if (rv != KERN_SUCCESS)
252 return EINVAL;
253 }
254
255 if (copy_len != 0) {
256 vm_object_reference(object);
257 rv = vm_map_find(exec_map,
258 object,
259 trunc_page(offset + filsz),
260 &data_buf,
261 PAGE_SIZE,
262 TRUE,
263 VM_PROT_READ,
264 VM_PROT_ALL,
265 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
266 if (rv != KERN_SUCCESS) {
267 vm_object_deallocate(object);
268 return EINVAL;
269 }
270
271 /* send the page fragment to user space */
272 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
273 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
274 if (error)
275 return (error);
276 }
277
278 /*
279 * set it to the specified protection
280 */
281 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot,
282 FALSE);
283
284 return error;
285}
286
287/*
288 * Load the file "file" into memory. It may be either a shared object
289 * or an executable.
290 *
291 * The "addr" reference parameter is in/out. On entry, it specifies
292 * the address where a shared object should be loaded. If the file is
293 * an executable, this value is ignored. On exit, "addr" specifies
294 * where the file was actually loaded.
295 *
296 * The "entry" reference parameter is out only. On exit, it specifies
297 * the entry point for the loaded file.
298 */
299static int
300elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
301{
302 const Elf_Ehdr *hdr = NULL;
303 const Elf_Phdr *phdr = NULL;
304 struct nameidata nd;
305 struct vmspace *vmspace = p->p_vmspace;
306 struct vattr attr;
307 struct image_params image_params, *imgp;
308 vm_prot_t prot;
309 u_long rbase;
310 u_long base_addr = 0;
311 int error, i, numsegs;
312
313 imgp = &image_params;
314 /*
315 * Initialize part of the common data
316 */
317 imgp->proc = p;
318 imgp->uap = NULL;
319 imgp->attr = &attr;
320 imgp->firstpage = NULL;
321 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
322
323 if (imgp->image_header == NULL) {
324 nd.ni_vp = NULL;
325 error = ENOMEM;
326 goto fail;
327 }
328
329 NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p);
330
331 if ((error = namei(&nd)) != 0) {
332 nd.ni_vp = NULL;
333 goto fail;
334 }
335
336 imgp->vp = nd.ni_vp;
337
338 /*
339 * Check permissions, modes, uid, etc on the file, and "open" it.
340 */
341 error = exec_check_permissions(imgp);
342 if (error) {
343 VOP_UNLOCK(nd.ni_vp, 0, p);
344 goto fail;
345 }
346
347 error = exec_map_first_page(imgp);
348 VOP_UNLOCK(nd.ni_vp, 0, p);
349 if (error)
350 goto fail;
351
352 hdr = (const Elf_Ehdr *)imgp->image_header;
353 if ((error = elf_check_header(hdr)) != 0)
354 goto fail;
355 if (hdr->e_type == ET_DYN)
356 rbase = *addr;
357 else if (hdr->e_type == ET_EXEC)
358 rbase = 0;
359 else {
360 error = ENOEXEC;
361 goto fail;
362 }
363
364 /* Only support headers that fit within first page for now */
365 if ((hdr->e_phoff > PAGE_SIZE) ||
366 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
367 error = ENOEXEC;
368 goto fail;
369 }
370
371 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
372
373 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
374 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
375 prot = 0;
376 if (phdr[i].p_flags & PF_X)
377 prot |= VM_PROT_EXECUTE;
378 if (phdr[i].p_flags & PF_W)
379 prot |= VM_PROT_WRITE;
380 if (phdr[i].p_flags & PF_R)
381 prot |= VM_PROT_READ;
382
383 if ((error = elf_load_section(p, vmspace, nd.ni_vp,
384 phdr[i].p_offset,
385 (caddr_t)phdr[i].p_vaddr +
386 rbase,
387 phdr[i].p_memsz,
388 phdr[i].p_filesz, prot)) != 0)
389 goto fail;
390 /*
391 * Establish the base address if this is the
392 * first segment.
393 */
394 if (numsegs == 0)
395 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
396 numsegs++;
397 }
398 }
399 *addr = base_addr;
400 *entry=(unsigned long)hdr->e_entry + rbase;
401
402fail:
403 if (imgp->firstpage)
404 exec_unmap_first_page(imgp);
405 if (imgp->image_header)
406 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
407 PAGE_SIZE);
408 if (nd.ni_vp)
409 vrele(nd.ni_vp);
410
411 return error;
412}
413
414static char fallback_elf_brand[MAXBRANDLEN+1] = { "none" };
415SYSCTL_STRING(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
416 fallback_elf_brand, sizeof(fallback_elf_brand),
417 "ELF brand of last resort");
418
419static int
420exec_elf_imgact(struct image_params *imgp)
421{
422 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
423 const Elf_Phdr *phdr;
424 Elf_Auxargs *elf_auxargs = NULL;
425 struct vmspace *vmspace;
426 vm_prot_t prot;
427 u_long text_size = 0, data_size = 0;
428 u_long text_addr = 0, data_addr = 0;
429 u_long addr, entry = 0, proghdr = 0;
430 int error, i;
431 const char *interp = NULL;
432 Elf_Brandinfo *brand_info;
433 const char *brand;
434 char path[MAXPATHLEN];
435
436 /*
437 * Do we have a valid ELF header ?
438 */
439 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
440 return -1;
441
442 /*
443 * From here on down, we return an errno, not -1, as we've
444 * detected an ELF file.
445 */
446
447 if ((hdr->e_phoff > PAGE_SIZE) ||
448 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
449 /* Only support headers in first page for now */
450 return ENOEXEC;
451 }
452 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
453
454 /*
455 * From this point on, we may have resources that need to be freed.
456 */
457 if ((error = exec_extract_strings(imgp)) != 0)
458 goto fail;
459
460 exec_new_vmspace(imgp);
461
462 vmspace = imgp->proc->p_vmspace;
463
464 for (i = 0; i < hdr->e_phnum; i++) {
465 switch(phdr[i].p_type) {
466
467 case PT_LOAD: /* Loadable segment */
468 prot = 0;
469 if (phdr[i].p_flags & PF_X)
470 prot |= VM_PROT_EXECUTE;
471 if (phdr[i].p_flags & PF_W)
472 prot |= VM_PROT_WRITE;
473 if (phdr[i].p_flags & PF_R)
474 prot |= VM_PROT_READ;
475
476 if ((error = elf_load_section(imgp->proc,
477 vmspace, imgp->vp,
478 phdr[i].p_offset,
479 (caddr_t)phdr[i].p_vaddr,
480 phdr[i].p_memsz,
481 phdr[i].p_filesz, prot)) != 0)
482 goto fail;
483
484 /*
485 * Is this .text or .data ??
486 *
487 * We only handle one each of those yet XXX
488 */
489 if (hdr->e_entry >= phdr[i].p_vaddr &&
490 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) {
491 text_addr = trunc_page(phdr[i].p_vaddr);
492 text_size = round_page(phdr[i].p_memsz +
493 phdr[i].p_vaddr -
494 text_addr);
495 entry = (u_long)hdr->e_entry;
496 } else {
497 data_addr = trunc_page(phdr[i].p_vaddr);
498 data_size = round_page(phdr[i].p_memsz +
499 phdr[i].p_vaddr -
500 data_addr);
501 }
502 break;
503 case PT_INTERP: /* Path to interpreter */
504 if (phdr[i].p_filesz > MAXPATHLEN ||
505 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
506 error = ENOEXEC;
507 goto fail;
508 }
509 interp = imgp->image_header + phdr[i].p_offset;
510 break;
511 case PT_PHDR: /* Program header table info */
512 proghdr = phdr[i].p_vaddr;
513 break;
514 default:
515 break;
516 }
517 }
518
519 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
520 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
521 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
522 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
523
524 addr = ELF_RTLD_ADDR(vmspace);
525
526 imgp->entry_addr = entry;
527
528 /* If the executable has a brand, search for it in the brand list. */
529 brand_info = NULL;
530 brand = (const char *)&hdr->e_ident[EI_BRAND];
531 if (brand[0] != '\0') {
532 for (i = 0; i < MAX_BRANDS; i++) {
533 Elf_Brandinfo *bi = elf_brand_list[i];
534
535 if (bi != NULL && strcmp(brand, bi->brand) == 0) {
536 brand_info = bi;
537 break;
538 }
539 }
540 }
541
542 /* Lacking a known brand, search for a recognized interpreter. */
543 if (brand_info == NULL && interp != NULL) {
544 for (i = 0; i < MAX_BRANDS; i++) {
545 Elf_Brandinfo *bi = elf_brand_list[i];
546
547 if (bi != NULL &&
548 strcmp(interp, bi->interp_path) == 0) {
549 brand_info = bi;
550 break;
551 }
552 }
553 }
554
555 /* Lacking a recognized interpreter, try the default brand */
556 if (brand_info == NULL && fallback_elf_brand[0] != '\0') {
557 for (i = 0; i < MAX_BRANDS; i++) {
558 Elf_Brandinfo *bi = elf_brand_list[i];
559
560 if (bi != NULL
561 && strcmp(fallback_elf_brand, bi->brand) == 0) {
562 brand_info = bi;
563 break;
564 }
565 }
566 }
567
568#ifdef __alpha__
569 /* XXX - Assume FreeBSD on the alpha. */
570 if (brand_info == NULL)
571 brand_info = &freebsd_brand_info;
572#endif
573
574 if (brand_info == NULL) {
575 if (brand[0] == 0)
576 uprintf("ELF binary type not known."
577 " Use \"brandelf\" to brand it.\n");
578 else
579 uprintf("ELF binary type \"%.*s\" not known.\n",
580 EI_NIDENT - EI_BRAND, brand);
581 error = ENOEXEC;
582 goto fail;
583 }
584
585 imgp->proc->p_sysent = brand_info->sysvec;
586 if (interp != NULL) {
587 snprintf(path, sizeof(path), "%s%s",
588 brand_info->emul_path, interp);
589 if ((error = elf_load_file(imgp->proc, path, &addr,
590 &imgp->entry_addr)) != 0) {
591 if ((error = elf_load_file(imgp->proc, interp, &addr,
592 &imgp->entry_addr)) != 0) {
593 uprintf("ELF interpreter %s not found\n", path);
594 goto fail;
595 }
596 }
597 }
598
599 /*
600 * Construct auxargs table (used by the fixup routine)
601 */
602 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
603 elf_auxargs->execfd = -1;
604 elf_auxargs->phdr = proghdr;
605 elf_auxargs->phent = hdr->e_phentsize;
606 elf_auxargs->phnum = hdr->e_phnum;
607 elf_auxargs->pagesz = PAGE_SIZE;
608 elf_auxargs->base = addr;
609 elf_auxargs->flags = 0;
610 elf_auxargs->entry = entry;
611 elf_auxargs->trace = elf_trace;
612
613 imgp->auxargs = elf_auxargs;
614 imgp->interpreted = 0;
615
616 /* don't allow modifying the file while we run it */
617 imgp->vp->v_flag |= VTEXT;
618
619fail:
620 return error;
621}
622
623static int
624elf_freebsd_fixup(long **stack_base, struct image_params *imgp)
625{
626 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
627 long *pos;
628
629 pos = *stack_base + (imgp->argc + imgp->envc + 2);
630
631 if (args->trace) {
632 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
633 }
634 if (args->execfd != -1) {
635 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
636 }
637 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
638 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
639 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
640 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
641 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
642 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
643 AUXARGS_ENTRY(pos, AT_BASE, args->base);
644 AUXARGS_ENTRY(pos, AT_NULL, 0);
645
646 free(imgp->auxargs, M_TEMP);
647 imgp->auxargs = NULL;
648
649 (*stack_base)--;
650 suword(*stack_base, (long) imgp->argc);
651 return 0;
652}
653
654/*
655 * Code for generating ELF core dumps.
656 */
657
658typedef void (*segment_callback) __P((vm_map_entry_t, void *));
659
660/* Closure for cb_put_phdr(). */
661struct phdr_closure {
662 Elf_Phdr *phdr; /* Program header to fill in */
663 Elf_Off offset; /* Offset of segment in core file */
664};
665
666/* Closure for cb_size_segment(). */
667struct sseg_closure {
668 int count; /* Count of writable segments. */
669 size_t size; /* Total size of all writable segments. */
670};
671
672static void cb_put_phdr __P((vm_map_entry_t, void *));
673static void cb_size_segment __P((vm_map_entry_t, void *));
674static void each_writable_segment __P((struct proc *, segment_callback,
675 void *));
676static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *,
677 int, void *, size_t));
678static void elf_puthdr __P((struct proc *, void *, size_t *,
679 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int));
680static void elf_putnote __P((void *, size_t *, const char *, int,
681 const void *, size_t));
682
683extern int osreldate;
684
685int
30 */
31
32#include "opt_rlimit.h"
33
34#include <sys/param.h>
35#include <sys/acct.h>
36#include <sys/exec.h>
37#include <sys/fcntl.h>
38#include <sys/imgact.h>
39#include <sys/imgact_elf.h>
40#include <sys/kernel.h>
41#include <sys/malloc.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/signalvar.h>
49#include <sys/stat.h>
50#include <sys/syscall.h>
51#include <sys/sysctl.h>
52#include <sys/sysent.h>
53#include <sys/systm.h>
54#include <sys/vnode.h>
55
56#include <vm/vm.h>
57#include <vm/vm_kern.h>
58#include <vm/vm_param.h>
59#include <vm/pmap.h>
60#include <sys/lock.h>
61#include <vm/vm_map.h>
62#include <vm/vm_object.h>
63#include <vm/vm_prot.h>
64#include <vm/vm_extern.h>
65
66#include <machine/elf.h>
67#include <machine/md_var.h>
68
69__ElfType(Brandinfo);
70__ElfType(Auxargs);
71
72static int elf_check_header __P((const Elf_Ehdr *hdr));
73static int elf_freebsd_fixup __P((long **stack_base,
74 struct image_params *imgp));
75static int elf_load_file __P((struct proc *p, const char *file, u_long *addr,
76 u_long *entry));
77static int elf_load_section __P((struct proc *p,
78 struct vmspace *vmspace, struct vnode *vp,
79 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
80 vm_prot_t prot));
81static int exec_elf_imgact __P((struct image_params *imgp));
82
83static int elf_trace = 0;
84SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
85
86/*
87 * XXX Maximum length of an ELF brand (sysctl wants a statically-allocated
88 * buffer).
89 */
90#define MAXBRANDLEN 16
91
92static struct sysentvec elf_freebsd_sysvec = {
93 SYS_MAXSYSCALL,
94 sysent,
95 0,
96 0,
97 0,
98 0,
99 0,
100 0,
101 elf_freebsd_fixup,
102 sendsig,
103 sigcode,
104 &szsigcode,
105 0,
106 "FreeBSD ELF",
107 elf_coredump
108};
109
110static Elf_Brandinfo freebsd_brand_info = {
111 "FreeBSD",
112 "",
113 "/usr/libexec/ld-elf.so.1",
114 &elf_freebsd_sysvec
115 };
116static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
117 &freebsd_brand_info,
118 NULL, NULL, NULL,
119 NULL, NULL, NULL, NULL
120 };
121
122int
123elf_insert_brand_entry(Elf_Brandinfo *entry)
124{
125 int i;
126
127 for (i=1; i<MAX_BRANDS; i++) {
128 if (elf_brand_list[i] == NULL) {
129 elf_brand_list[i] = entry;
130 break;
131 }
132 }
133 if (i == MAX_BRANDS)
134 return -1;
135 return 0;
136}
137
138int
139elf_remove_brand_entry(Elf_Brandinfo *entry)
140{
141 int i;
142
143 for (i=1; i<MAX_BRANDS; i++) {
144 if (elf_brand_list[i] == entry) {
145 elf_brand_list[i] = NULL;
146 break;
147 }
148 }
149 if (i == MAX_BRANDS)
150 return -1;
151 return 0;
152}
153
154int
155elf_brand_inuse(Elf_Brandinfo *entry)
156{
157 struct proc *p;
158
159 for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
160 if (p->p_sysent == entry->sysvec)
161 return TRUE;
162 }
163
164 return FALSE;
165}
166
167static int
168elf_check_header(const Elf_Ehdr *hdr)
169{
170 if (!IS_ELF(*hdr) ||
171 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
172 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
173 hdr->e_ident[EI_VERSION] != EV_CURRENT)
174 return ENOEXEC;
175
176 if (!ELF_MACHINE_OK(hdr->e_machine))
177 return ENOEXEC;
178
179 if (hdr->e_version != ELF_TARG_VER)
180 return ENOEXEC;
181
182 return 0;
183}
184
185static int
186elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp, vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz, vm_prot_t prot)
187{
188 size_t map_len;
189 vm_offset_t map_addr;
190 int error, rv;
191 size_t copy_len;
192 vm_object_t object;
193 vm_offset_t file_addr;
194 vm_offset_t data_buf = 0;
195
196 object = vp->v_object;
197 error = 0;
198
199 map_addr = trunc_page((vm_offset_t)vmaddr);
200 file_addr = trunc_page(offset);
201
202 /*
203 * We have two choices. We can either clear the data in the last page
204 * of an oversized mapping, or we can start the anon mapping a page
205 * early and copy the initialized data into that first page. We
206 * choose the second..
207 */
208 if (memsz > filsz)
209 map_len = trunc_page(offset+filsz) - file_addr;
210 else
211 map_len = round_page(offset+filsz) - file_addr;
212
213 if (map_len != 0) {
214 vm_object_reference(object);
215 vm_map_lock(&vmspace->vm_map);
216 rv = vm_map_insert(&vmspace->vm_map,
217 object,
218 file_addr, /* file offset */
219 map_addr, /* virtual start */
220 map_addr + map_len,/* virtual end */
221 prot,
222 VM_PROT_ALL,
223 MAP_COPY_ON_WRITE | MAP_PREFAULT);
224 vm_map_unlock(&vmspace->vm_map);
225 if (rv != KERN_SUCCESS)
226 return EINVAL;
227
228 /* we can stop now if we've covered it all */
229 if (memsz == filsz)
230 return 0;
231 }
232
233
234 /*
235 * We have to get the remaining bit of the file into the first part
236 * of the oversized map segment. This is normally because the .data
237 * segment in the file is extended to provide bss. It's a neat idea
238 * to try and save a page, but it's a pain in the behind to implement.
239 */
240 copy_len = (offset + filsz) - trunc_page(offset + filsz);
241 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
242 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
243
244 /* This had damn well better be true! */
245 if (map_len != 0) {
246 vm_map_lock(&vmspace->vm_map);
247 rv = vm_map_insert(&vmspace->vm_map, NULL, 0,
248 map_addr, map_addr + map_len,
249 VM_PROT_ALL, VM_PROT_ALL, 0);
250 vm_map_unlock(&vmspace->vm_map);
251 if (rv != KERN_SUCCESS)
252 return EINVAL;
253 }
254
255 if (copy_len != 0) {
256 vm_object_reference(object);
257 rv = vm_map_find(exec_map,
258 object,
259 trunc_page(offset + filsz),
260 &data_buf,
261 PAGE_SIZE,
262 TRUE,
263 VM_PROT_READ,
264 VM_PROT_ALL,
265 MAP_COPY_ON_WRITE | MAP_PREFAULT_PARTIAL);
266 if (rv != KERN_SUCCESS) {
267 vm_object_deallocate(object);
268 return EINVAL;
269 }
270
271 /* send the page fragment to user space */
272 error = copyout((caddr_t)data_buf, (caddr_t)map_addr, copy_len);
273 vm_map_remove(exec_map, data_buf, data_buf + PAGE_SIZE);
274 if (error)
275 return (error);
276 }
277
278 /*
279 * set it to the specified protection
280 */
281 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot,
282 FALSE);
283
284 return error;
285}
286
287/*
288 * Load the file "file" into memory. It may be either a shared object
289 * or an executable.
290 *
291 * The "addr" reference parameter is in/out. On entry, it specifies
292 * the address where a shared object should be loaded. If the file is
293 * an executable, this value is ignored. On exit, "addr" specifies
294 * where the file was actually loaded.
295 *
296 * The "entry" reference parameter is out only. On exit, it specifies
297 * the entry point for the loaded file.
298 */
299static int
300elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
301{
302 const Elf_Ehdr *hdr = NULL;
303 const Elf_Phdr *phdr = NULL;
304 struct nameidata nd;
305 struct vmspace *vmspace = p->p_vmspace;
306 struct vattr attr;
307 struct image_params image_params, *imgp;
308 vm_prot_t prot;
309 u_long rbase;
310 u_long base_addr = 0;
311 int error, i, numsegs;
312
313 imgp = &image_params;
314 /*
315 * Initialize part of the common data
316 */
317 imgp->proc = p;
318 imgp->uap = NULL;
319 imgp->attr = &attr;
320 imgp->firstpage = NULL;
321 imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
322
323 if (imgp->image_header == NULL) {
324 nd.ni_vp = NULL;
325 error = ENOMEM;
326 goto fail;
327 }
328
329 NDINIT(&nd, LOOKUP, LOCKLEAF|FOLLOW, UIO_SYSSPACE, file, p);
330
331 if ((error = namei(&nd)) != 0) {
332 nd.ni_vp = NULL;
333 goto fail;
334 }
335
336 imgp->vp = nd.ni_vp;
337
338 /*
339 * Check permissions, modes, uid, etc on the file, and "open" it.
340 */
341 error = exec_check_permissions(imgp);
342 if (error) {
343 VOP_UNLOCK(nd.ni_vp, 0, p);
344 goto fail;
345 }
346
347 error = exec_map_first_page(imgp);
348 VOP_UNLOCK(nd.ni_vp, 0, p);
349 if (error)
350 goto fail;
351
352 hdr = (const Elf_Ehdr *)imgp->image_header;
353 if ((error = elf_check_header(hdr)) != 0)
354 goto fail;
355 if (hdr->e_type == ET_DYN)
356 rbase = *addr;
357 else if (hdr->e_type == ET_EXEC)
358 rbase = 0;
359 else {
360 error = ENOEXEC;
361 goto fail;
362 }
363
364 /* Only support headers that fit within first page for now */
365 if ((hdr->e_phoff > PAGE_SIZE) ||
366 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
367 error = ENOEXEC;
368 goto fail;
369 }
370
371 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
372
373 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
374 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
375 prot = 0;
376 if (phdr[i].p_flags & PF_X)
377 prot |= VM_PROT_EXECUTE;
378 if (phdr[i].p_flags & PF_W)
379 prot |= VM_PROT_WRITE;
380 if (phdr[i].p_flags & PF_R)
381 prot |= VM_PROT_READ;
382
383 if ((error = elf_load_section(p, vmspace, nd.ni_vp,
384 phdr[i].p_offset,
385 (caddr_t)phdr[i].p_vaddr +
386 rbase,
387 phdr[i].p_memsz,
388 phdr[i].p_filesz, prot)) != 0)
389 goto fail;
390 /*
391 * Establish the base address if this is the
392 * first segment.
393 */
394 if (numsegs == 0)
395 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
396 numsegs++;
397 }
398 }
399 *addr = base_addr;
400 *entry=(unsigned long)hdr->e_entry + rbase;
401
402fail:
403 if (imgp->firstpage)
404 exec_unmap_first_page(imgp);
405 if (imgp->image_header)
406 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
407 PAGE_SIZE);
408 if (nd.ni_vp)
409 vrele(nd.ni_vp);
410
411 return error;
412}
413
414static char fallback_elf_brand[MAXBRANDLEN+1] = { "none" };
415SYSCTL_STRING(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
416 fallback_elf_brand, sizeof(fallback_elf_brand),
417 "ELF brand of last resort");
418
419static int
420exec_elf_imgact(struct image_params *imgp)
421{
422 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
423 const Elf_Phdr *phdr;
424 Elf_Auxargs *elf_auxargs = NULL;
425 struct vmspace *vmspace;
426 vm_prot_t prot;
427 u_long text_size = 0, data_size = 0;
428 u_long text_addr = 0, data_addr = 0;
429 u_long addr, entry = 0, proghdr = 0;
430 int error, i;
431 const char *interp = NULL;
432 Elf_Brandinfo *brand_info;
433 const char *brand;
434 char path[MAXPATHLEN];
435
436 /*
437 * Do we have a valid ELF header ?
438 */
439 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
440 return -1;
441
442 /*
443 * From here on down, we return an errno, not -1, as we've
444 * detected an ELF file.
445 */
446
447 if ((hdr->e_phoff > PAGE_SIZE) ||
448 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
449 /* Only support headers in first page for now */
450 return ENOEXEC;
451 }
452 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
453
454 /*
455 * From this point on, we may have resources that need to be freed.
456 */
457 if ((error = exec_extract_strings(imgp)) != 0)
458 goto fail;
459
460 exec_new_vmspace(imgp);
461
462 vmspace = imgp->proc->p_vmspace;
463
464 for (i = 0; i < hdr->e_phnum; i++) {
465 switch(phdr[i].p_type) {
466
467 case PT_LOAD: /* Loadable segment */
468 prot = 0;
469 if (phdr[i].p_flags & PF_X)
470 prot |= VM_PROT_EXECUTE;
471 if (phdr[i].p_flags & PF_W)
472 prot |= VM_PROT_WRITE;
473 if (phdr[i].p_flags & PF_R)
474 prot |= VM_PROT_READ;
475
476 if ((error = elf_load_section(imgp->proc,
477 vmspace, imgp->vp,
478 phdr[i].p_offset,
479 (caddr_t)phdr[i].p_vaddr,
480 phdr[i].p_memsz,
481 phdr[i].p_filesz, prot)) != 0)
482 goto fail;
483
484 /*
485 * Is this .text or .data ??
486 *
487 * We only handle one each of those yet XXX
488 */
489 if (hdr->e_entry >= phdr[i].p_vaddr &&
490 hdr->e_entry <(phdr[i].p_vaddr+phdr[i].p_memsz)) {
491 text_addr = trunc_page(phdr[i].p_vaddr);
492 text_size = round_page(phdr[i].p_memsz +
493 phdr[i].p_vaddr -
494 text_addr);
495 entry = (u_long)hdr->e_entry;
496 } else {
497 data_addr = trunc_page(phdr[i].p_vaddr);
498 data_size = round_page(phdr[i].p_memsz +
499 phdr[i].p_vaddr -
500 data_addr);
501 }
502 break;
503 case PT_INTERP: /* Path to interpreter */
504 if (phdr[i].p_filesz > MAXPATHLEN ||
505 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
506 error = ENOEXEC;
507 goto fail;
508 }
509 interp = imgp->image_header + phdr[i].p_offset;
510 break;
511 case PT_PHDR: /* Program header table info */
512 proghdr = phdr[i].p_vaddr;
513 break;
514 default:
515 break;
516 }
517 }
518
519 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
520 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
521 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
522 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
523
524 addr = ELF_RTLD_ADDR(vmspace);
525
526 imgp->entry_addr = entry;
527
528 /* If the executable has a brand, search for it in the brand list. */
529 brand_info = NULL;
530 brand = (const char *)&hdr->e_ident[EI_BRAND];
531 if (brand[0] != '\0') {
532 for (i = 0; i < MAX_BRANDS; i++) {
533 Elf_Brandinfo *bi = elf_brand_list[i];
534
535 if (bi != NULL && strcmp(brand, bi->brand) == 0) {
536 brand_info = bi;
537 break;
538 }
539 }
540 }
541
542 /* Lacking a known brand, search for a recognized interpreter. */
543 if (brand_info == NULL && interp != NULL) {
544 for (i = 0; i < MAX_BRANDS; i++) {
545 Elf_Brandinfo *bi = elf_brand_list[i];
546
547 if (bi != NULL &&
548 strcmp(interp, bi->interp_path) == 0) {
549 brand_info = bi;
550 break;
551 }
552 }
553 }
554
555 /* Lacking a recognized interpreter, try the default brand */
556 if (brand_info == NULL && fallback_elf_brand[0] != '\0') {
557 for (i = 0; i < MAX_BRANDS; i++) {
558 Elf_Brandinfo *bi = elf_brand_list[i];
559
560 if (bi != NULL
561 && strcmp(fallback_elf_brand, bi->brand) == 0) {
562 brand_info = bi;
563 break;
564 }
565 }
566 }
567
568#ifdef __alpha__
569 /* XXX - Assume FreeBSD on the alpha. */
570 if (brand_info == NULL)
571 brand_info = &freebsd_brand_info;
572#endif
573
574 if (brand_info == NULL) {
575 if (brand[0] == 0)
576 uprintf("ELF binary type not known."
577 " Use \"brandelf\" to brand it.\n");
578 else
579 uprintf("ELF binary type \"%.*s\" not known.\n",
580 EI_NIDENT - EI_BRAND, brand);
581 error = ENOEXEC;
582 goto fail;
583 }
584
585 imgp->proc->p_sysent = brand_info->sysvec;
586 if (interp != NULL) {
587 snprintf(path, sizeof(path), "%s%s",
588 brand_info->emul_path, interp);
589 if ((error = elf_load_file(imgp->proc, path, &addr,
590 &imgp->entry_addr)) != 0) {
591 if ((error = elf_load_file(imgp->proc, interp, &addr,
592 &imgp->entry_addr)) != 0) {
593 uprintf("ELF interpreter %s not found\n", path);
594 goto fail;
595 }
596 }
597 }
598
599 /*
600 * Construct auxargs table (used by the fixup routine)
601 */
602 elf_auxargs = malloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
603 elf_auxargs->execfd = -1;
604 elf_auxargs->phdr = proghdr;
605 elf_auxargs->phent = hdr->e_phentsize;
606 elf_auxargs->phnum = hdr->e_phnum;
607 elf_auxargs->pagesz = PAGE_SIZE;
608 elf_auxargs->base = addr;
609 elf_auxargs->flags = 0;
610 elf_auxargs->entry = entry;
611 elf_auxargs->trace = elf_trace;
612
613 imgp->auxargs = elf_auxargs;
614 imgp->interpreted = 0;
615
616 /* don't allow modifying the file while we run it */
617 imgp->vp->v_flag |= VTEXT;
618
619fail:
620 return error;
621}
622
623static int
624elf_freebsd_fixup(long **stack_base, struct image_params *imgp)
625{
626 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
627 long *pos;
628
629 pos = *stack_base + (imgp->argc + imgp->envc + 2);
630
631 if (args->trace) {
632 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
633 }
634 if (args->execfd != -1) {
635 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
636 }
637 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
638 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
639 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
640 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
641 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
642 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
643 AUXARGS_ENTRY(pos, AT_BASE, args->base);
644 AUXARGS_ENTRY(pos, AT_NULL, 0);
645
646 free(imgp->auxargs, M_TEMP);
647 imgp->auxargs = NULL;
648
649 (*stack_base)--;
650 suword(*stack_base, (long) imgp->argc);
651 return 0;
652}
653
654/*
655 * Code for generating ELF core dumps.
656 */
657
658typedef void (*segment_callback) __P((vm_map_entry_t, void *));
659
660/* Closure for cb_put_phdr(). */
661struct phdr_closure {
662 Elf_Phdr *phdr; /* Program header to fill in */
663 Elf_Off offset; /* Offset of segment in core file */
664};
665
666/* Closure for cb_size_segment(). */
667struct sseg_closure {
668 int count; /* Count of writable segments. */
669 size_t size; /* Total size of all writable segments. */
670};
671
672static void cb_put_phdr __P((vm_map_entry_t, void *));
673static void cb_size_segment __P((vm_map_entry_t, void *));
674static void each_writable_segment __P((struct proc *, segment_callback,
675 void *));
676static int elf_corehdr __P((struct proc *, struct vnode *, struct ucred *,
677 int, void *, size_t));
678static void elf_puthdr __P((struct proc *, void *, size_t *,
679 const prstatus_t *, const prfpregset_t *, const prpsinfo_t *, int));
680static void elf_putnote __P((void *, size_t *, const char *, int,
681 const void *, size_t));
682
683extern int osreldate;
684
685int
686elf_coredump(p)
686elf_coredump(p, vp, limit)
687 register struct proc *p;
687 register struct proc *p;
688{
689 register struct vnode *vp;
688 register struct vnode *vp;
689 off_t limit;
690{
690 register struct ucred *cred = p->p_cred->pc_ucred;
691 register struct ucred *cred = p->p_cred->pc_ucred;
691 struct nameidata nd;
692 struct vattr vattr;
693 int error, error1;
694 char *name; /* name of corefile */
692 int error = 0;
695 struct sseg_closure seginfo;
696 void *hdr;
697 size_t hdrsize;
698
693 struct sseg_closure seginfo;
694 void *hdr;
695 size_t hdrsize;
696
699 STOPEVENT(p, S_CORE, 0);
700
701 if (sugid_coredump == 0 && p->p_flag & P_SUGID)
702 return (EFAULT);
703
704 /* Size the program segments. */
705 seginfo.count = 0;
706 seginfo.size = 0;
707 each_writable_segment(p, cb_size_segment, &seginfo);
708
709 /*
710 * Calculate the size of the core file header area by making
711 * a dry run of generating it. Nothing is written, but the
712 * size is calculated.
713 */
714 hdrsize = 0;
715 elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
716 (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
717 (const prpsinfo_t *)NULL, seginfo.count);
718
697 /* Size the program segments. */
698 seginfo.count = 0;
699 seginfo.size = 0;
700 each_writable_segment(p, cb_size_segment, &seginfo);
701
702 /*
703 * Calculate the size of the core file header area by making
704 * a dry run of generating it. Nothing is written, but the
705 * size is calculated.
706 */
707 hdrsize = 0;
708 elf_puthdr((struct proc *)NULL, (void *)NULL, &hdrsize,
709 (const prstatus_t *)NULL, (const prfpregset_t *)NULL,
710 (const prpsinfo_t *)NULL, seginfo.count);
711
719 if (hdrsize + seginfo.size >= p->p_rlimit[RLIMIT_CORE].rlim_cur)
712 if (hdrsize + seginfo.size >= limit)
720 return (EFAULT);
713 return (EFAULT);
721 name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
722 if (name == NULL)
723 return (EFAULT); /* XXX -- not the best error */
724
725 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, p);
726 error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
727 free(name, M_TEMP);
728 if (error)
729 return (error);
730 vp = nd.ni_vp;
731
714
732 /* Don't dump to non-regular files or files with links. */
733 if (vp->v_type != VREG ||
734 VOP_GETATTR(vp, &vattr, cred, p) || vattr.va_nlink != 1) {
735 error = EFAULT;
736 goto out;
737 }
738 VATTR_NULL(&vattr);
739 vattr.va_size = 0;
740 VOP_LEASE(vp, p, cred, LEASE_WRITE);
741 VOP_SETATTR(vp, &vattr, cred, p);
742 p->p_acflag |= ACORE;
743
744
745 /*
746 * Allocate memory for building the header, fill it up,
747 * and write it out.
748 */
749 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
750 if (hdr == NULL) {
715 /*
716 * Allocate memory for building the header, fill it up,
717 * and write it out.
718 */
719 hdr = malloc(hdrsize, M_TEMP, M_WAITOK);
720 if (hdr == NULL) {
751 error = EINVAL;
752 goto out;
721 return EINVAL;
753 }
754 error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
755
756 /* Write the contents of all of the writable segments. */
757 if (error == 0) {
758 Elf_Phdr *php;
759 off_t offset;
760 int i;
761
762 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
763 offset = hdrsize;
764 for (i = 0; i < seginfo.count; i++) {
765 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr,
766 php->p_filesz, offset, UIO_USERSPACE,
767 IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p);
768 if (error != 0)
769 break;
770 offset += php->p_filesz;
771 php++;
772 }
773 }
774 free(hdr, M_TEMP);
722 }
723 error = elf_corehdr(p, vp, cred, seginfo.count, hdr, hdrsize);
724
725 /* Write the contents of all of the writable segments. */
726 if (error == 0) {
727 Elf_Phdr *php;
728 off_t offset;
729 int i;
730
731 php = (Elf_Phdr *)((char *)hdr + sizeof(Elf_Ehdr)) + 1;
732 offset = hdrsize;
733 for (i = 0; i < seginfo.count; i++) {
734 error = vn_rdwr(UIO_WRITE, vp, (caddr_t)php->p_vaddr,
735 php->p_filesz, offset, UIO_USERSPACE,
736 IO_NODELOCKED|IO_UNIT, cred, (int *)NULL, p);
737 if (error != 0)
738 break;
739 offset += php->p_filesz;
740 php++;
741 }
742 }
743 free(hdr, M_TEMP);
775
776out:
777 VOP_UNLOCK(vp, 0, p);
778 error1 = vn_close(vp, FWRITE, cred, p);
779 if (error == 0)
780 error = error1;
781 return (error);
744
745 return error;
782}
783
784/*
785 * A callback for each_writable_segment() to write out the segment's
786 * program header entry.
787 */
788static void
789cb_put_phdr(entry, closure)
790 vm_map_entry_t entry;
791 void *closure;
792{
793 struct phdr_closure *phc = (struct phdr_closure *)closure;
794 Elf_Phdr *phdr = phc->phdr;
795
796 phc->offset = round_page(phc->offset);
797
798 phdr->p_type = PT_LOAD;
799 phdr->p_offset = phc->offset;
800 phdr->p_vaddr = entry->start;
801 phdr->p_paddr = 0;
802 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
803 phdr->p_align = PAGE_SIZE;
804 phdr->p_flags = 0;
805 if (entry->protection & VM_PROT_READ)
806 phdr->p_flags |= PF_R;
807 if (entry->protection & VM_PROT_WRITE)
808 phdr->p_flags |= PF_W;
809 if (entry->protection & VM_PROT_EXECUTE)
810 phdr->p_flags |= PF_X;
811
812 phc->offset += phdr->p_filesz;
813 phc->phdr++;
814}
815
816/*
817 * A callback for each_writable_segment() to gather information about
818 * the number of segments and their total size.
819 */
820static void
821cb_size_segment(entry, closure)
822 vm_map_entry_t entry;
823 void *closure;
824{
825 struct sseg_closure *ssc = (struct sseg_closure *)closure;
826
827 ssc->count++;
828 ssc->size += entry->end - entry->start;
829}
830
831/*
832 * For each writable segment in the process's memory map, call the given
833 * function with a pointer to the map entry and some arbitrary
834 * caller-supplied data.
835 */
836static void
837each_writable_segment(p, func, closure)
838 struct proc *p;
839 segment_callback func;
840 void *closure;
841{
842 vm_map_t map = &p->p_vmspace->vm_map;
843 vm_map_entry_t entry;
844
845 for (entry = map->header.next; entry != &map->header;
846 entry = entry->next) {
847 vm_object_t obj;
848
849 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
850 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) !=
851 (VM_PROT_READ|VM_PROT_WRITE))
852 continue;
853
854 if ((obj = entry->object.vm_object) == NULL)
855 continue;
856
857 /* Find the deepest backing object. */
858 while (obj->backing_object != NULL)
859 obj = obj->backing_object;
860
861 /* Ignore memory-mapped devices and such things. */
862 if (obj->type != OBJT_DEFAULT &&
863 obj->type != OBJT_SWAP &&
864 obj->type != OBJT_VNODE)
865 continue;
866
867 (*func)(entry, closure);
868 }
869}
870
871/*
872 * Write the core file header to the file, including padding up to
873 * the page boundary.
874 */
875static int
876elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
877 struct proc *p;
878 struct vnode *vp;
879 struct ucred *cred;
880 int numsegs;
881 size_t hdrsize;
882 void *hdr;
883{
884 size_t off;
885 prstatus_t status;
886 prfpregset_t fpregset;
887 prpsinfo_t psinfo;
888
889 /* Gather the information for the header. */
890 bzero(&status, sizeof status);
891 status.pr_version = PRSTATUS_VERSION;
892 status.pr_statussz = sizeof(prstatus_t);
893 status.pr_gregsetsz = sizeof(gregset_t);
894 status.pr_fpregsetsz = sizeof(fpregset_t);
895 status.pr_osreldate = osreldate;
896 status.pr_cursig = p->p_sig;
897 status.pr_pid = p->p_pid;
898 fill_regs(p, &status.pr_reg);
899
900 fill_fpregs(p, &fpregset);
901
902 bzero(&psinfo, sizeof psinfo);
903 psinfo.pr_version = PRPSINFO_VERSION;
904 psinfo.pr_psinfosz = sizeof(prpsinfo_t);
905 strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN);
906 /* XXX - We don't fill in the command line arguments properly yet. */
907 strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ);
908
909 /* Fill in the header. */
910 bzero(hdr, hdrsize);
911 off = 0;
912 elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs);
913
914 /* Write it to the core file. */
915 return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
916 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p);
917}
918
919static void
920elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
921 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
922{
923 size_t ehoff;
924 size_t phoff;
925 size_t noteoff;
926 size_t notesz;
927
928 ehoff = *off;
929 *off += sizeof(Elf_Ehdr);
930
931 phoff = *off;
932 *off += (numsegs + 1) * sizeof(Elf_Phdr);
933
934 noteoff = *off;
935 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
936 sizeof *status);
937 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
938 sizeof *fpregset);
939 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
940 sizeof *psinfo);
941 notesz = *off - noteoff;
942
943 /* Align up to a page boundary for the program segments. */
944 *off = round_page(*off);
945
946 if (dst != NULL) {
947 Elf_Ehdr *ehdr;
948 Elf_Phdr *phdr;
949 struct phdr_closure phc;
950
951 /*
952 * Fill in the ELF header.
953 */
954 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
955 ehdr->e_ident[EI_MAG0] = ELFMAG0;
956 ehdr->e_ident[EI_MAG1] = ELFMAG1;
957 ehdr->e_ident[EI_MAG2] = ELFMAG2;
958 ehdr->e_ident[EI_MAG3] = ELFMAG3;
959 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
960 ehdr->e_ident[EI_DATA] = ELF_DATA;
961 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
962 ehdr->e_ident[EI_PAD] = 0;
963 strncpy(ehdr->e_ident + EI_BRAND, "FreeBSD",
964 EI_NIDENT - EI_BRAND);
965 ehdr->e_type = ET_CORE;
966 ehdr->e_machine = ELF_ARCH;
967 ehdr->e_version = EV_CURRENT;
968 ehdr->e_entry = 0;
969 ehdr->e_phoff = phoff;
970 ehdr->e_flags = 0;
971 ehdr->e_ehsize = sizeof(Elf_Ehdr);
972 ehdr->e_phentsize = sizeof(Elf_Phdr);
973 ehdr->e_phnum = numsegs + 1;
974 ehdr->e_shentsize = sizeof(Elf_Shdr);
975 ehdr->e_shnum = 0;
976 ehdr->e_shstrndx = SHN_UNDEF;
977
978 /*
979 * Fill in the program header entries.
980 */
981 phdr = (Elf_Phdr *)((char *)dst + phoff);
982
983 /* The note segement. */
984 phdr->p_type = PT_NOTE;
985 phdr->p_offset = noteoff;
986 phdr->p_vaddr = 0;
987 phdr->p_paddr = 0;
988 phdr->p_filesz = notesz;
989 phdr->p_memsz = 0;
990 phdr->p_flags = 0;
991 phdr->p_align = 0;
992 phdr++;
993
994 /* All the writable segments from the program. */
995 phc.phdr = phdr;
996 phc.offset = *off;
997 each_writable_segment(p, cb_put_phdr, &phc);
998 }
999}
1000
1001static void
1002elf_putnote(void *dst, size_t *off, const char *name, int type,
1003 const void *desc, size_t descsz)
1004{
1005 Elf_Note note;
1006
1007 note.n_namesz = strlen(name) + 1;
1008 note.n_descsz = descsz;
1009 note.n_type = type;
1010 if (dst != NULL)
1011 bcopy(&note, (char *)dst + *off, sizeof note);
1012 *off += sizeof note;
1013 if (dst != NULL)
1014 bcopy(name, (char *)dst + *off, note.n_namesz);
1015 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
1016 if (dst != NULL)
1017 bcopy(desc, (char *)dst + *off, note.n_descsz);
1018 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
1019}
1020
1021/*
1022 * Tell kern_execve.c about it, with a little help from the linker.
1023 */
1024static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1025EXEC_SET(elf, elf_execsw);
746}
747
748/*
749 * A callback for each_writable_segment() to write out the segment's
750 * program header entry.
751 */
752static void
753cb_put_phdr(entry, closure)
754 vm_map_entry_t entry;
755 void *closure;
756{
757 struct phdr_closure *phc = (struct phdr_closure *)closure;
758 Elf_Phdr *phdr = phc->phdr;
759
760 phc->offset = round_page(phc->offset);
761
762 phdr->p_type = PT_LOAD;
763 phdr->p_offset = phc->offset;
764 phdr->p_vaddr = entry->start;
765 phdr->p_paddr = 0;
766 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
767 phdr->p_align = PAGE_SIZE;
768 phdr->p_flags = 0;
769 if (entry->protection & VM_PROT_READ)
770 phdr->p_flags |= PF_R;
771 if (entry->protection & VM_PROT_WRITE)
772 phdr->p_flags |= PF_W;
773 if (entry->protection & VM_PROT_EXECUTE)
774 phdr->p_flags |= PF_X;
775
776 phc->offset += phdr->p_filesz;
777 phc->phdr++;
778}
779
780/*
781 * A callback for each_writable_segment() to gather information about
782 * the number of segments and their total size.
783 */
784static void
785cb_size_segment(entry, closure)
786 vm_map_entry_t entry;
787 void *closure;
788{
789 struct sseg_closure *ssc = (struct sseg_closure *)closure;
790
791 ssc->count++;
792 ssc->size += entry->end - entry->start;
793}
794
795/*
796 * For each writable segment in the process's memory map, call the given
797 * function with a pointer to the map entry and some arbitrary
798 * caller-supplied data.
799 */
800static void
801each_writable_segment(p, func, closure)
802 struct proc *p;
803 segment_callback func;
804 void *closure;
805{
806 vm_map_t map = &p->p_vmspace->vm_map;
807 vm_map_entry_t entry;
808
809 for (entry = map->header.next; entry != &map->header;
810 entry = entry->next) {
811 vm_object_t obj;
812
813 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
814 (entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) !=
815 (VM_PROT_READ|VM_PROT_WRITE))
816 continue;
817
818 if ((obj = entry->object.vm_object) == NULL)
819 continue;
820
821 /* Find the deepest backing object. */
822 while (obj->backing_object != NULL)
823 obj = obj->backing_object;
824
825 /* Ignore memory-mapped devices and such things. */
826 if (obj->type != OBJT_DEFAULT &&
827 obj->type != OBJT_SWAP &&
828 obj->type != OBJT_VNODE)
829 continue;
830
831 (*func)(entry, closure);
832 }
833}
834
835/*
836 * Write the core file header to the file, including padding up to
837 * the page boundary.
838 */
839static int
840elf_corehdr(p, vp, cred, numsegs, hdr, hdrsize)
841 struct proc *p;
842 struct vnode *vp;
843 struct ucred *cred;
844 int numsegs;
845 size_t hdrsize;
846 void *hdr;
847{
848 size_t off;
849 prstatus_t status;
850 prfpregset_t fpregset;
851 prpsinfo_t psinfo;
852
853 /* Gather the information for the header. */
854 bzero(&status, sizeof status);
855 status.pr_version = PRSTATUS_VERSION;
856 status.pr_statussz = sizeof(prstatus_t);
857 status.pr_gregsetsz = sizeof(gregset_t);
858 status.pr_fpregsetsz = sizeof(fpregset_t);
859 status.pr_osreldate = osreldate;
860 status.pr_cursig = p->p_sig;
861 status.pr_pid = p->p_pid;
862 fill_regs(p, &status.pr_reg);
863
864 fill_fpregs(p, &fpregset);
865
866 bzero(&psinfo, sizeof psinfo);
867 psinfo.pr_version = PRPSINFO_VERSION;
868 psinfo.pr_psinfosz = sizeof(prpsinfo_t);
869 strncpy(psinfo.pr_fname, p->p_comm, MAXCOMLEN);
870 /* XXX - We don't fill in the command line arguments properly yet. */
871 strncpy(psinfo.pr_psargs, p->p_comm, PRARGSZ);
872
873 /* Fill in the header. */
874 bzero(hdr, hdrsize);
875 off = 0;
876 elf_puthdr(p, hdr, &off, &status, &fpregset, &psinfo, numsegs);
877
878 /* Write it to the core file. */
879 return vn_rdwr(UIO_WRITE, vp, hdr, hdrsize, (off_t)0,
880 UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, cred, NULL, p);
881}
882
883static void
884elf_puthdr(struct proc *p, void *dst, size_t *off, const prstatus_t *status,
885 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs)
886{
887 size_t ehoff;
888 size_t phoff;
889 size_t noteoff;
890 size_t notesz;
891
892 ehoff = *off;
893 *off += sizeof(Elf_Ehdr);
894
895 phoff = *off;
896 *off += (numsegs + 1) * sizeof(Elf_Phdr);
897
898 noteoff = *off;
899 elf_putnote(dst, off, "FreeBSD", NT_PRSTATUS, status,
900 sizeof *status);
901 elf_putnote(dst, off, "FreeBSD", NT_FPREGSET, fpregset,
902 sizeof *fpregset);
903 elf_putnote(dst, off, "FreeBSD", NT_PRPSINFO, psinfo,
904 sizeof *psinfo);
905 notesz = *off - noteoff;
906
907 /* Align up to a page boundary for the program segments. */
908 *off = round_page(*off);
909
910 if (dst != NULL) {
911 Elf_Ehdr *ehdr;
912 Elf_Phdr *phdr;
913 struct phdr_closure phc;
914
915 /*
916 * Fill in the ELF header.
917 */
918 ehdr = (Elf_Ehdr *)((char *)dst + ehoff);
919 ehdr->e_ident[EI_MAG0] = ELFMAG0;
920 ehdr->e_ident[EI_MAG1] = ELFMAG1;
921 ehdr->e_ident[EI_MAG2] = ELFMAG2;
922 ehdr->e_ident[EI_MAG3] = ELFMAG3;
923 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
924 ehdr->e_ident[EI_DATA] = ELF_DATA;
925 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
926 ehdr->e_ident[EI_PAD] = 0;
927 strncpy(ehdr->e_ident + EI_BRAND, "FreeBSD",
928 EI_NIDENT - EI_BRAND);
929 ehdr->e_type = ET_CORE;
930 ehdr->e_machine = ELF_ARCH;
931 ehdr->e_version = EV_CURRENT;
932 ehdr->e_entry = 0;
933 ehdr->e_phoff = phoff;
934 ehdr->e_flags = 0;
935 ehdr->e_ehsize = sizeof(Elf_Ehdr);
936 ehdr->e_phentsize = sizeof(Elf_Phdr);
937 ehdr->e_phnum = numsegs + 1;
938 ehdr->e_shentsize = sizeof(Elf_Shdr);
939 ehdr->e_shnum = 0;
940 ehdr->e_shstrndx = SHN_UNDEF;
941
942 /*
943 * Fill in the program header entries.
944 */
945 phdr = (Elf_Phdr *)((char *)dst + phoff);
946
947 /* The note segement. */
948 phdr->p_type = PT_NOTE;
949 phdr->p_offset = noteoff;
950 phdr->p_vaddr = 0;
951 phdr->p_paddr = 0;
952 phdr->p_filesz = notesz;
953 phdr->p_memsz = 0;
954 phdr->p_flags = 0;
955 phdr->p_align = 0;
956 phdr++;
957
958 /* All the writable segments from the program. */
959 phc.phdr = phdr;
960 phc.offset = *off;
961 each_writable_segment(p, cb_put_phdr, &phc);
962 }
963}
964
965static void
966elf_putnote(void *dst, size_t *off, const char *name, int type,
967 const void *desc, size_t descsz)
968{
969 Elf_Note note;
970
971 note.n_namesz = strlen(name) + 1;
972 note.n_descsz = descsz;
973 note.n_type = type;
974 if (dst != NULL)
975 bcopy(&note, (char *)dst + *off, sizeof note);
976 *off += sizeof note;
977 if (dst != NULL)
978 bcopy(name, (char *)dst + *off, note.n_namesz);
979 *off += roundup2(note.n_namesz, sizeof(Elf_Size));
980 if (dst != NULL)
981 bcopy(desc, (char *)dst + *off, note.n_descsz);
982 *off += roundup2(note.n_descsz, sizeof(Elf_Size));
983}
984
985/*
986 * Tell kern_execve.c about it, with a little help from the linker.
987 */
988static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
989EXEC_SET(elf, elf_execsw);