?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
HAMMER Utilities: Sync with recent work.
[dragonfly.git] / sys / kern / imgact_elf.c
blobf7e811beeb958a4476a5471f37c2f09a9bbc0d1a
1 /*-
2 * Copyright (c) 1995-1996 Søren 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: src/sys/kern/imgact_elf.c,v 1.73.2.13 2002/12/28 19:49:41 dillon Exp $
30 * $DragonFly: src/sys/kern/imgact_elf.c,v 1.53 2008/06/07 11:37:23 mneumann Exp $
31 */
32
33 #include <sys/param.h>
34 #include <sys/exec.h>
35 #include <sys/fcntl.h>
36 #include <sys/file.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
41 #include <sys/mman.h>
42 #include <sys/systm.h>
43 #include <sys/proc.h>
44 #include <sys/nlookup.h>
45 #include <sys/pioctl.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/vnode.h>
54 #include <sys/sfbuf.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_extern.h>
64
65 #include <machine/elf.h>
66 #include <machine/md_var.h>
67 #include <sys/mount.h>
68 #include <sys/ckpt.h>
69 #define OLD_EI_BRAND 8
70
71 __ElfType(Brandinfo);
72 __ElfType(Auxargs);
73
74 static int elf_check_header (const Elf_Ehdr *hdr);
75 static int elf_freebsd_fixup (register_t **stack_base,
76 struct image_params *imgp);
77 static int elf_load_file (struct proc *p, const char *file, u_long *addr,
78 u_long *entry);
79 static int elf_load_section (struct proc *p,
80 struct vmspace *vmspace, struct vnode *vp,
81 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
82 vm_prot_t prot);
83 static int exec_elf_imgact (struct image_params *imgp);
84
85 static int elf_trace = 0;
86 SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
87 static int elf_legacy_coredump = 0;
88 SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
89 &elf_legacy_coredump, 0, "");
90
91 static int dragonfly_match_abi_note(const Elf_Note *);
92 static int freebsd_match_abi_note(const Elf_Note *);
93
94 static struct sysentvec elf_freebsd_sysvec = {
95 SYS_MAXSYSCALL,
96 sysent,
97 -1,
98 0,
99 0,
100 0,
101 0,
102 0,
103 elf_freebsd_fixup,
104 sendsig,
105 sigcode,
106 &szsigcode,
107 0,
108 "FreeBSD ELF",
109 elf_coredump,
110 NULL,
111 MINSIGSTKSZ
112 };
113
114 static Elf_Brandinfo freebsd_brand_info = {
115 ELFOSABI_FREEBSD,
116 "FreeBSD",
117 freebsd_match_abi_note,
118 "",
119 "/usr/libexec/ld-elf.so.1",
120 &elf_freebsd_sysvec
121 };
122
123 static Elf_Brandinfo dragonfly_brand_info = {
124 ELFOSABI_NONE,
125 "DragonFly",
126 dragonfly_match_abi_note,
127 "",
128 "/usr/libexec/ld-elf.so.2",
129 &elf_freebsd_sysvec
130 };
131
132 static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
133 &dragonfly_brand_info,
134 &freebsd_brand_info,
135 NULL, NULL, NULL,
136 NULL, NULL, NULL
137 };
138
139 static int
140 freebsd_match_abi_note(const Elf_Note *abi_note)
141 {
142 const char *abi_name = (const char *)
143 ((const uint8_t *)abi_note + sizeof(*abi_note));
144
145 if (abi_note->n_namesz != sizeof("FreeBSD"))
146 return(FALSE);
147 if (memcmp(abi_name, "FreeBSD", sizeof("FreeBSD")))
148 return(FALSE);
149 return(TRUE);
150 }
151
152 static int
153 dragonfly_match_abi_note(const Elf_Note *abi_note)
154 {
155 const char *abi_name = (const char *)
156 ((const uint8_t *)abi_note + sizeof(*abi_note));
157
158 if (abi_note->n_namesz != sizeof("DragonFly"))
159 return(FALSE);
160 if (memcmp(abi_name, "DragonFly", sizeof("DragonFly")))
161 return(FALSE);
162 return(TRUE);
163 }
164
165 int
166 elf_insert_brand_entry(Elf_Brandinfo *entry)
167 {
168 int i;
169
170 for (i=1; i<MAX_BRANDS; i++) {
171 if (elf_brand_list[i] == NULL) {
172 elf_brand_list[i] = entry;
173 break;
174 }
175 }
176 if (i == MAX_BRANDS)
177 return -1;
178 return 0;
179 }
180
181 int
182 elf_remove_brand_entry(Elf_Brandinfo *entry)
183 {
184 int i;
185
186 for (i=1; i<MAX_BRANDS; i++) {
187 if (elf_brand_list[i] == entry) {
188 elf_brand_list[i] = NULL;
189 break;
190 }
191 }
192 if (i == MAX_BRANDS)
193 return -1;
194 return 0;
195 }
196
197 /*
198 * Check if an elf brand is being used anywhere in the system.
199 *
200 * Used by the linux emulation module unloader. This isn't safe from
201 * races.
202 */
203 struct elf_brand_inuse_info {
204 int rval;
205 Elf_Brandinfo *entry;
206 };
207
208 static int elf_brand_inuse_callback(struct proc *p, void *data);
209
210 int
211 elf_brand_inuse(Elf_Brandinfo *entry)
212 {
213 struct elf_brand_inuse_info info;
214
215 info.rval = FALSE;
216 info.entry = entry;
217 allproc_scan(elf_brand_inuse_callback, entry);
218 return (info.rval);
219 }
220
221 static
222 int
223 elf_brand_inuse_callback(struct proc *p, void *data)
224 {
225 struct elf_brand_inuse_info *info = data;
226
227 if (p->p_sysent == info->entry->sysvec) {
228 info->rval = TRUE;
229 return(-1);
230 }
231 return(0);
232 }
233
234 static int
235 elf_check_header(const Elf_Ehdr *hdr)
236 {
237 if (!IS_ELF(*hdr) ||
238 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
239 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
240 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
241 hdr->e_phentsize != sizeof(Elf_Phdr) ||
242 hdr->e_ehsize != sizeof(Elf_Ehdr) ||
243 hdr->e_version != ELF_TARG_VER)
244 return ENOEXEC;
245
246 if (!ELF_MACHINE_OK(hdr->e_machine))
247 return ENOEXEC;
248
249 return 0;
250 }
251
252 static int
253 elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp,
254 vm_offset_t offset, caddr_t vmaddr, size_t memsz,
255 size_t filsz, vm_prot_t prot)
256 {
257 size_t map_len;
258 vm_offset_t map_addr;
259 int error, rv, cow;
260 int count;
261 size_t copy_len;
262 vm_object_t object;
263 vm_offset_t file_addr;
264
265 object = vp->v_object;
266 error = 0;
267
268 /*
269 * It's necessary to fail if the filsz + offset taken from the
270 * header is greater than the actual file pager object's size.
271 * If we were to allow this, then the vm_map_find() below would
272 * walk right off the end of the file object and into the ether.
273 *
274 * While I'm here, might as well check for something else that
275 * is invalid: filsz cannot be greater than memsz.
276 */
277 if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) {
278 uprintf("elf_load_section: truncated ELF file\n");
279 return (ENOEXEC);
280 }
281
282 map_addr = trunc_page((vm_offset_t)vmaddr);
283 file_addr = trunc_page(offset);
284
285 /*
286 * We have two choices. We can either clear the data in the last page
287 * of an oversized mapping, or we can start the anon mapping a page
288 * early and copy the initialized data into that first page. We
289 * choose the second..
290 */
291 if (memsz > filsz)
292 map_len = trunc_page(offset+filsz) - file_addr;
293 else
294 map_len = round_page(offset+filsz) - file_addr;
295
296 if (map_len != 0) {
297 vm_object_reference(object);
298
299 /* cow flags: don't dump readonly sections in core */
300 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
301 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
302
303 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
304 vm_map_lock(&vmspace->vm_map);
305 rv = vm_map_insert(&vmspace->vm_map, &count,
306 object,
307 file_addr, /* file offset */
308 map_addr, /* virtual start */
309 map_addr + map_len,/* virtual end */
310 VM_MAPTYPE_NORMAL,
311 prot, VM_PROT_ALL,
312 cow);
313 vm_map_unlock(&vmspace->vm_map);
314 vm_map_entry_release(count);
315 if (rv != KERN_SUCCESS) {
316 vm_object_deallocate(object);
317 return EINVAL;
318 }
319
320 /* we can stop now if we've covered it all */
321 if (memsz == filsz) {
322 return 0;
323 }
324 }
325
326
327 /*
328 * We have to get the remaining bit of the file into the first part
329 * of the oversized map segment. This is normally because the .data
330 * segment in the file is extended to provide bss. It's a neat idea
331 * to try and save a page, but it's a pain in the behind to implement.
332 */
333 copy_len = (offset + filsz) - trunc_page(offset + filsz);
334 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
335 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
336
337 /* This had damn well better be true! */
338 if (map_len != 0) {
339 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
340 vm_map_lock(&vmspace->vm_map);
341 rv = vm_map_insert(&vmspace->vm_map, &count,
342 NULL, 0,
343 map_addr, map_addr + map_len,
344 VM_MAPTYPE_NORMAL,
345 VM_PROT_ALL, VM_PROT_ALL,
346 0);
347 vm_map_unlock(&vmspace->vm_map);
348 vm_map_entry_release(count);
349 if (rv != KERN_SUCCESS) {
350 return EINVAL;
351 }
352 }
353
354 if (copy_len != 0) {
355 vm_page_t m;
356 struct sf_buf *sf;
357
358 m = vm_fault_object_page(object, trunc_page(offset + filsz),
359 VM_PROT_READ, 0, &error);
360 if (m) {
361 sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
362 error = copyout((caddr_t)sf_buf_kva(sf),
363 (caddr_t)map_addr, copy_len);
364 sf_buf_free(sf);
365 vm_page_unhold(m);
366 }
367 if (error) {
368 return (error);
369 }
370 }
371
372 /*
373 * set it to the specified protection
374 */
375 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot,
376 FALSE);
377
378 return error;
379 }
380
381 /*
382 * Load the file "file" into memory. It may be either a shared object
383 * or an executable.
384 *
385 * The "addr" reference parameter is in/out. On entry, it specifies
386 * the address where a shared object should be loaded. If the file is
387 * an executable, this value is ignored. On exit, "addr" specifies
388 * where the file was actually loaded.
389 *
390 * The "entry" reference parameter is out only. On exit, it specifies
391 * the entry point for the loaded file.
392 */
393 static int
394 elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
395 {
396 struct {
397 struct nlookupdata nd;
398 struct vattr attr;
399 struct image_params image_params;
400 } *tempdata;
401 const Elf_Ehdr *hdr = NULL;
402 const Elf_Phdr *phdr = NULL;
403 struct nlookupdata *nd;
404 struct vmspace *vmspace = p->p_vmspace;
405 struct vattr *attr;
406 struct image_params *imgp;
407 vm_prot_t prot;
408 u_long rbase;
409 u_long base_addr = 0;
410 int error, i, numsegs;
411
412 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
413 nd = &tempdata->nd;
414 attr = &tempdata->attr;
415 imgp = &tempdata->image_params;
416
417 /*
418 * Initialize part of the common data
419 */
420 imgp->proc = p;
421 imgp->attr = attr;
422 imgp->firstpage = NULL;
423 imgp->image_header = NULL;
424 imgp->vp = NULL;
425
426 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW);
427 if (error == 0)
428 error = nlookup(nd);
429 if (error == 0)
430 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
431 nlookup_done(nd);
432 if (error)
433 goto fail;
434
435 /*
436 * Check permissions, modes, uid, etc on the file, and "open" it.
437 */
438 error = exec_check_permissions(imgp);
439 if (error) {
440 vn_unlock(imgp->vp);
441 goto fail;
442 }
443
444 error = exec_map_first_page(imgp);
445 /*
446 * Also make certain that the interpreter stays the same, so set
447 * its VTEXT flag, too.
448 */
449 if (error == 0)
450 imgp->vp->v_flag |= VTEXT;
451 vn_unlock(imgp->vp);
452 if (error)
453 goto fail;
454
455 hdr = (const Elf_Ehdr *)imgp->image_header;
456 if ((error = elf_check_header(hdr)) != 0)
457 goto fail;
458 if (hdr->e_type == ET_DYN)
459 rbase = *addr;
460 else if (hdr->e_type == ET_EXEC)
461 rbase = 0;
462 else {
463 error = ENOEXEC;
464 goto fail;
465 }
466
467 /* Only support headers that fit within first page for now
468 * (multiplication of two Elf_Half fields will not overflow) */
469 if ((hdr->e_phoff > PAGE_SIZE) ||
470 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
471 error = ENOEXEC;
472 goto fail;
473 }
474
475 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
476
477 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
478 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
479 prot = 0;
480 if (phdr[i].p_flags & PF_X)
481 prot |= VM_PROT_EXECUTE;
482 if (phdr[i].p_flags & PF_W)
483 prot |= VM_PROT_WRITE;
484 if (phdr[i].p_flags & PF_R)
485 prot |= VM_PROT_READ;
486
487 error = elf_load_section(
488 p, vmspace, imgp->vp,
489 phdr[i].p_offset,
490 (caddr_t)phdr[i].p_vaddr +
491 rbase,
492 phdr[i].p_memsz,
493 phdr[i].p_filesz, prot);
494 if (error != 0)
495 goto fail;
496 /*
497 * Establish the base address if this is the
498 * first segment.
499 */
500 if (numsegs == 0)
501 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
502 numsegs++;
503 }
504 }
505 *addr = base_addr;
506 *entry=(unsigned long)hdr->e_entry + rbase;
507
508 fail:
509 if (imgp->firstpage)
510 exec_unmap_first_page(imgp);
511 if (imgp->vp) {
512 vrele(imgp->vp);
513 imgp->vp = NULL;
514 }
515 kfree(tempdata, M_TEMP);
516
517 return error;
518 }
519
520 /*
521 * non static, as it can be overridden by start_init()
522 */
523 int fallback_elf_brand = -1;
524 SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
525 &fallback_elf_brand, -1,
526 "ELF brand of last resort");
527
528 static int
529 exec_elf_imgact(struct image_params *imgp)
530 {
531 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
532 const Elf_Phdr *phdr;
533 Elf_Auxargs *elf_auxargs = NULL;
534 struct vmspace *vmspace;
535 vm_prot_t prot;
536 u_long text_size = 0, data_size = 0, total_size = 0;
537 u_long text_addr = 0, data_addr = 0;
538 u_long seg_size, seg_addr;
539 u_long addr, entry = 0, proghdr = 0;
540 int error, i;
541 const char *interp = NULL;
542 const Elf_Note *abi_note = NULL;
543 Elf_Brandinfo *brand_info;
544 char *path;
545
546 error = 0;
547
548 /*
549 * Do we have a valid ELF header ?
550 */
551 if (elf_check_header(hdr) != 0 || hdr->e_type != ET_EXEC)
552 return -1;
553
554 /*
555 * From here on down, we return an errno, not -1, as we've
556 * detected an ELF file.
557 */
558
559 if ((hdr->e_phoff > PAGE_SIZE) ||
560 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
561 /* Only support headers in first page for now */
562 return ENOEXEC;
563 }
564 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
565
566 /*
567 * From this point on, we may have resources that need to be freed.
568 */
569
570 exec_new_vmspace(imgp, NULL);
571
572 /*
573 * Yeah, I'm paranoid. There is every reason in the world to get
574 * VTEXT now since from here on out, there are places we can have
575 * a context switch. Better safe than sorry; I really don't want
576 * the file to change while it's being loaded.
577 */
578 vsetflags(imgp->vp, VTEXT);
579
580 vmspace = imgp->proc->p_vmspace;
581
582 for (i = 0; i < hdr->e_phnum; i++) {
583 switch(phdr[i].p_type) {
584
585 case PT_LOAD: /* Loadable segment */
586 prot = 0;
587 if (phdr[i].p_flags & PF_X)
588 prot |= VM_PROT_EXECUTE;
589 if (phdr[i].p_flags & PF_W)
590 prot |= VM_PROT_WRITE;
591 if (phdr[i].p_flags & PF_R)
592 prot |= VM_PROT_READ;
593
594 if ((error = elf_load_section(imgp->proc,
595 vmspace, imgp->vp,
596 phdr[i].p_offset,
597 (caddr_t)phdr[i].p_vaddr,
598 phdr[i].p_memsz,
599 phdr[i].p_filesz, prot)) != 0)
600 goto fail;
601
602 /*
603 * If this segment contains the program headers,
604 * remember their virtual address for the AT_PHDR
605 * aux entry. Static binaries don't usually include
606 * a PT_PHDR entry.
607 */
608 if (phdr[i].p_offset == 0 &&
609 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
610 <= phdr[i].p_filesz)
611 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
612
613 seg_addr = trunc_page(phdr[i].p_vaddr);
614 seg_size = round_page(phdr[i].p_memsz +
615 phdr[i].p_vaddr - seg_addr);
616
617 /*
618 * Is this .text or .data? We can't use
619 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
620 * alpha terribly and possibly does other bad
621 * things so we stick to the old way of figuring
622 * it out: If the segment contains the program
623 * entry point, it's a text segment, otherwise it
624 * is a data segment.
625 *
626 * Note that obreak() assumes that data_addr +
627 * data_size == end of data load area, and the ELF
628 * file format expects segments to be sorted by
629 * address. If multiple data segments exist, the
630 * last one will be used.
631 */
632 if (hdr->e_entry >= phdr[i].p_vaddr &&
633 hdr->e_entry < (phdr[i].p_vaddr +
634 phdr[i].p_memsz)) {
635 text_size = seg_size;
636 text_addr = seg_addr;
637 entry = (u_long)hdr->e_entry;
638 } else {
639 data_size = seg_size;
640 data_addr = seg_addr;
641 }
642 total_size += seg_size;
643
644 /*
645 * Check limits. It should be safe to check the
646 * limits after loading the segment since we do
647 * not actually fault in all the segment's pages.
648 */
649 if (data_size >
650 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
651 text_size > maxtsiz ||
652 total_size >
653 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
654 error = ENOMEM;
655 goto fail;
656 }
657 break;
658 case PT_INTERP: /* Path to interpreter */
659 if (phdr[i].p_filesz > MAXPATHLEN ||
660 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
661 error = ENOEXEC;
662 goto fail;
663 }
664 interp = imgp->image_header + phdr[i].p_offset;
665 break;
666 case PT_NOTE: /* Check for .note.ABI-tag */
667 {
668 const Elf_Note *tmp_note;
669 /* XXX handle anything outside the first page */
670 if (phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE)
671 continue;
672 if (phdr[i].p_filesz < sizeof(Elf_Note))
673 continue; /* ENOEXEC? */
674 tmp_note = (const Elf_Note *)(imgp->image_header + phdr[i].p_offset);
675 if (tmp_note->n_type != 1)
676 continue;
677 if (tmp_note->n_namesz + sizeof(Elf_Note) +
678 tmp_note->n_descsz > phdr[i].p_filesz)
679 continue; /* ENOEXEC? */
680 abi_note = tmp_note;
681 }
682 break;
683 case PT_PHDR: /* Program header table info */
684 proghdr = phdr[i].p_vaddr;
685 break;
686 default:
687 break;
688 }
689 }
690
691 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
692 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
693 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
694 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
695
696 addr = ELF_RTLD_ADDR(vmspace);
697
698 imgp->entry_addr = entry;
699
700 brand_info = NULL;
701
702 /* We support three types of branding -- (1) the ELF EI_OSABI field
703 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
704 * branding w/in the ELF header, and (3) path of the `interp_path'
705 * field. We should also look for an ".note.ABI-tag" ELF section now
706 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
707 */
708
709 /* If the executable has a brand, search for it in the brand list. */
710 if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) {
711 for (i = 0; i < MAX_BRANDS; i++) {
712 Elf_Brandinfo *bi = elf_brand_list[i];
713
714 if (bi != NULL &&
715 (hdr->e_ident[EI_OSABI] == bi->brand
716 || 0 ==
717 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
718 bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
719 brand_info = bi;
720 break;
721 }
722 }
723 }
724
725 /* Search for a recognized ABI. */
726 if (brand_info == NULL && abi_note != NULL) {
727 for (i = 0; i < MAX_BRANDS; i++) {
728 Elf_Brandinfo *bi = elf_brand_list[i];
729
730 if (bi != NULL && bi->match_abi_note != NULL &&
731 (*bi->match_abi_note)(abi_note)) {
732 brand_info = bi;
733 break;
734 }
735 }
736 }
737
738 /*
739 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all
740 * checks so far, since it is neither branded nor does it have
741 * an ABI note. If the EI_OSABI field is ELFOSABI_NONE, assume
742 * it is svr4 and look for an entry in the elf_brand_list with
743 * match_abi_note == NULL.
744 */
745 if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) {
746 for (i = 0; i < MAX_BRANDS; i++) {
747 Elf_Brandinfo *bi = elf_brand_list[i];
748
749 if (bi != NULL && bi->match_abi_note == NULL &&
750 ELFOSABI_SYSV == bi->brand) {
751 brand_info = bi;
752 break;
753 }
754 }
755 }
756
757 /* Lacking a recognized ABI, search for a recognized interpreter. */
758 if (brand_info == NULL && interp != NULL) {
759 for (i = 0; i < MAX_BRANDS; i++) {
760 Elf_Brandinfo *bi = elf_brand_list[i];
761
762 if (bi != NULL &&
763 strcmp(interp, bi->interp_path) == 0) {
764 brand_info = bi;
765 break;
766 }
767 }
768 }
769
770 /* Lacking a recognized interpreter, try the default brand */
771 if (brand_info == NULL) {
772 for (i = 0; i < MAX_BRANDS; i++) {
773 Elf_Brandinfo *bi = elf_brand_list[i];
774
775 if (bi != NULL && fallback_elf_brand == bi->brand) {
776 brand_info = bi;
777 break;
778 }
779 }
780 }
781
782 if (brand_info == NULL) {
783 uprintf("ELF binary type \"%u\" not known.\n",
784 hdr->e_ident[EI_OSABI]);
785 error = ENOEXEC;
786 goto fail;
787 }
788
789 imgp->proc->p_sysent = brand_info->sysvec;
790 if (interp != NULL) {
791 path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK);
792 ksnprintf(path, MAXPATHLEN, "%s%s",
793 brand_info->emul_path, interp);
794 if ((error = elf_load_file(imgp->proc, path, &addr,
795 &imgp->entry_addr)) != 0) {
796 if ((error = elf_load_file(imgp->proc, interp, &addr,
797 &imgp->entry_addr)) != 0) {
798 uprintf("ELF interpreter %s not found\n", path);
799 kfree(path, M_TEMP);
800 goto fail;
801 }
802 }
803 kfree(path, M_TEMP);
804 }
805
806 /*
807 * Construct auxargs table (used by the fixup routine)
808 */
809 elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
810 elf_auxargs->execfd = -1;
811 elf_auxargs->phdr = proghdr;
812 elf_auxargs->phent = hdr->e_phentsize;
813 elf_auxargs->phnum = hdr->e_phnum;
814 elf_auxargs->pagesz = PAGE_SIZE;
815 elf_auxargs->base = addr;
816 elf_auxargs->flags = 0;
817 elf_auxargs->entry = entry;
818 elf_auxargs->trace = elf_trace;
819
820 imgp->auxargs = elf_auxargs;
821 imgp->interpreted = 0;
822
823 fail:
824 return error;
825 }
826
827 static int
828 elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
829 {
830 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
831 register_t *pos;
832
833 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
834
835 if (args->trace) {
836 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
837 }
838 if (args->execfd != -1) {
839 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
840 }
841 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
842 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
843 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
844 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
845 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
846 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
847 AUXARGS_ENTRY(pos, AT_BASE, args->base);
848 AUXARGS_ENTRY(pos, AT_NULL, 0);
849
850 kfree(imgp->auxargs, M_TEMP);
851 imgp->auxargs = NULL;
852
853 (*stack_base)--;
854 suword(*stack_base, (long) imgp->args->argc);
855 return 0;
856 }
857
858 /*
859 * Code for generating ELF core dumps.
860 */
861
862 typedef int (*segment_callback) (vm_map_entry_t, void *);
863
864 /* Closure for cb_put_phdr(). */
865 struct phdr_closure {
866 Elf_Phdr *phdr; /* Program header to fill in (incremented) */
867 Elf_Phdr *phdr_max; /* Pointer bound for error check */
868 Elf_Off offset; /* Offset of segment in core file */
869 };
870
871 /* Closure for cb_size_segment(). */
872 struct sseg_closure {
873 int count; /* Count of writable segments. */
874 size_t vsize; /* Total size of all writable segments. */
875 };
876
877 /* Closure for cb_put_fp(). */
878 struct fp_closure {
879 struct vn_hdr *vnh;
880 struct vn_hdr *vnh_max;
881 int count;
882 struct stat *sb;
883 };
884
885 typedef struct elf_buf {
886 char *buf;
887 size_t off;
888 size_t off_max;
889 } *elf_buf_t;
890
891 static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
892
893 static int cb_put_phdr (vm_map_entry_t, void *);
894 static int cb_size_segment (vm_map_entry_t, void *);
895 static int cb_fpcount_segment(vm_map_entry_t, void *);
896 static int cb_put_fp(vm_map_entry_t, void *);
897
898
899 static int each_segment (struct proc *, segment_callback, void *, int);
900 static int elf_corehdr (struct lwp *, int, struct file *, struct ucred *,
901 int, elf_buf_t);
902 static int elf_puthdr (struct lwp *, elf_buf_t, const prstatus_t *,
903 const prfpregset_t *, const prpsinfo_t *, int,
904 struct file *);
905 static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t);
906
907 static int elf_putsigs(struct lwp *, elf_buf_t);
908 static int elf_puttextvp(struct proc *, elf_buf_t);
909 static int elf_putfiles(struct proc *, elf_buf_t, struct file *);
910
911 extern int osreldate;
912
913 int
914 elf_coredump(struct lwp *lp, int sig, struct vnode *vp, off_t limit)
915 {
916 struct file *fp;
917 int error;
918
919 if ((error = falloc(NULL, &fp, NULL)) != 0)
920 return (error);
921 fsetcred(fp, lp->lwp_proc->p_ucred);
922
923 /*
924 * XXX fixme.
925 */
926 fp->f_type = DTYPE_VNODE;
927 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
928 fp->f_ops = &vnode_fileops;
929 fp->f_data = vp;
930 vn_unlock(vp);
931
932 error = generic_elf_coredump(lp, sig, fp, limit);
933
934 fp->f_type = 0;
935 fp->f_flag = 0;
936 fp->f_ops = &badfileops;
937 fp->f_data = NULL;
938 fdrop(fp);
939 return (error);
940 }
941
942 int
943 generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit)
944 {
945 struct proc *p = lp->lwp_proc;
946 struct ucred *cred = p->p_ucred;
947 int error = 0;
948 struct sseg_closure seginfo;
949 struct elf_buf target;
950
951 if (!fp)
952 kprintf("can't dump core - null fp\n");
953
954 /*
955 * Size the program segments
956 */
957 seginfo.count = 0;
958 seginfo.vsize = 0;
959 each_segment(p, cb_size_segment, &seginfo, 1);
960
961 /*
962 * Calculate the size of the core file header area by making
963 * a dry run of generating it. Nothing is written, but the
964 * size is calculated.
965 */
966 bzero(&target, sizeof(target));
967 elf_puthdr(lp, &target, NULL, NULL, NULL, seginfo.count, fp);
968
969 if (target.off + seginfo.vsize >= limit)
970 return (EFAULT);
971
972 /*
973 * Allocate memory for building the header, fill it up,
974 * and write it out.
975 */
976 target.off_max = target.off;
977 target.off = 0;
978 target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
979
980 if (target.buf == NULL)
981 return EINVAL;
982 error = elf_corehdr(lp, sig, fp, cred, seginfo.count, &target);
983
984 /* Write the contents of all of the writable segments. */
985 if (error == 0) {
986 Elf_Phdr *php;
987 int i;
988 int nbytes;
989
990 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
991 for (i = 0; i < seginfo.count; i++) {
992 error = fp_write(fp, (caddr_t)php->p_vaddr,
993 php->p_filesz, &nbytes, UIO_USERSPACE);
994 if (error != 0)
995 break;
996 php++;
997 }
998 }
999 kfree(target.buf, M_TEMP);
1000
1001 return error;
1002 }
1003
1004 /*
1005 * A callback for each_segment() to write out the segment's
1006 * program header entry.
1007 */
1008 static int
1009 cb_put_phdr(vm_map_entry_t entry, void *closure)
1010 {
1011 struct phdr_closure *phc = closure;
1012 Elf_Phdr *phdr = phc->phdr;
1013
1014 if (phc->phdr == phc->phdr_max)
1015 return EINVAL;
1016
1017 phc->offset = round_page(phc->offset);
1018
1019 phdr->p_type = PT_LOAD;
1020 phdr->p_offset = phc->offset;
1021 phdr->p_vaddr = entry->start;
1022 phdr->p_paddr = 0;
1023 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1024 phdr->p_align = PAGE_SIZE;
1025 phdr->p_flags = 0;
1026 if (entry->protection & VM_PROT_READ)
1027 phdr->p_flags |= PF_R;
1028 if (entry->protection & VM_PROT_WRITE)
1029 phdr->p_flags |= PF_W;
1030 if (entry->protection & VM_PROT_EXECUTE)
1031 phdr->p_flags |= PF_X;
1032
1033 phc->offset += phdr->p_filesz;
1034 ++phc->phdr;
1035 return 0;
1036 }
1037
1038 /*
1039 * A callback for each_writable_segment() to gather information about
1040 * the number of segments and their total size.
1041 */
1042 static int
1043 cb_size_segment(vm_map_entry_t entry, void *closure)
1044 {
1045 struct sseg_closure *ssc = closure;
1046
1047 ++ssc->count;
1048 ssc->vsize += entry->end - entry->start;
1049 return 0;
1050 }
1051
1052 /*
1053 * A callback for each_segment() to gather information about
1054 * the number of text segments.
1055 */
1056 static int
1057 cb_fpcount_segment(vm_map_entry_t entry, void *closure)
1058 {
1059 int *count = closure;
1060 struct vnode *vp;
1061
1062 if (entry->object.vm_object->type == OBJT_VNODE) {
1063 vp = (struct vnode *)entry->object.vm_object->handle;
1064 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1065 return 0;
1066 ++*count;
1067 }
1068 return 0;
1069 }
1070
1071 static int
1072 cb_put_fp(vm_map_entry_t entry, void *closure)
1073 {
1074 struct fp_closure *fpc = closure;
1075 struct vn_hdr *vnh = fpc->vnh;
1076 Elf_Phdr *phdr = &vnh->vnh_phdr;
1077 struct vnode *vp;
1078 int error;
1079
1080 /*
1081 * If an entry represents a vnode then write out a file handle.
1082 *
1083 * If we are checkpointing a checkpoint-restored program we do
1084 * NOT record the filehandle for the old checkpoint vnode (which
1085 * is mapped all over the place). Instead we rely on the fact
1086 * that a checkpoint-restored program does not mmap() the checkpt
1087 * vnode NOCORE, so its contents will be written out to the
1088 * new checkpoint file. This is necessary because the 'old'
1089 * checkpoint file is typically destroyed when a new one is created
1090 * and thus cannot be used to restore the new checkpoint.
1091 *
1092 * Theoretically we could create a chain of checkpoint files and
1093 * operate the checkpointing operation kinda like an incremental
1094 * checkpoint, but a checkpoint restore would then likely wind up
1095 * referencing many prior checkpoint files and that is a bit over
1096 * the top for the purpose of the checkpoint API.
1097 */
1098 if (entry->object.vm_object->type == OBJT_VNODE) {
1099 vp = (struct vnode *)entry->object.vm_object->handle;
1100 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1101 return 0;
1102 if (vnh == fpc->vnh_max)
1103 return EINVAL;
1104
1105 if (vp->v_mount)
1106 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1107 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
1108 if (error) {
1109 char *freepath, *fullpath;
1110
1111 if (vn_fullpath(curproc, vp, &fullpath, &freepath)) {
1112 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp);
1113 } else {
1114 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath);
1115 kfree(freepath, M_TEMP);
1116 }
1117 error = 0;
1118 }
1119
1120 phdr->p_type = PT_LOAD;
1121 phdr->p_offset = 0; /* not written to core */
1122 phdr->p_vaddr = entry->start;
1123 phdr->p_paddr = 0;
1124 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1125 phdr->p_align = PAGE_SIZE;
1126 phdr->p_flags = 0;
1127 if (entry->protection & VM_PROT_READ)
1128 phdr->p_flags |= PF_R;
1129 if (entry->protection & VM_PROT_WRITE)
1130 phdr->p_flags |= PF_W;
1131 if (entry->protection & VM_PROT_EXECUTE)
1132 phdr->p_flags |= PF_X;
1133 ++fpc->vnh;
1134 ++fpc->count;
1135 }
1136 return 0;
1137 }
1138
1139 /*
1140 * For each writable segment in the process's memory map, call the given
1141 * function with a pointer to the map entry and some arbitrary
1142 * caller-supplied data.
1143 */
1144 static int
1145 each_segment(struct proc *p, segment_callback func, void *closure, int writable)
1146 {
1147 int error = 0;
1148 vm_map_t map = &p->p_vmspace->vm_map;
1149 vm_map_entry_t entry;
1150
1151 for (entry = map->header.next; error == 0 && entry != &map->header;
1152 entry = entry->next) {
1153 vm_object_t obj;
1154
1155 /*
1156 * Don't dump inaccessible mappings, deal with legacy
1157 * coredump mode.
1158 *
1159 * Note that read-only segments related to the elf binary
1160 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1161 * need to arbitrarily ignore such segments.
1162 */
1163 if (elf_legacy_coredump) {
1164 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
1165 continue;
1166 } else {
1167 if (writable && (entry->protection & VM_PROT_ALL) == 0)
1168 continue;
1169 }
1170
1171 /*
1172 * Dont include memory segment in the coredump if
1173 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1174 * madvise(2).
1175 *
1176 * Currently we only dump normal VM object maps. We do
1177 * not dump submaps or virtual page tables.
1178 */
1179 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP))
1180 continue;
1181 if (entry->maptype != VM_MAPTYPE_NORMAL)
1182 continue;
1183 if ((obj = entry->object.vm_object) == NULL)
1184 continue;
1185
1186 /* Find the deepest backing object. */
1187 while (obj->backing_object != NULL)
1188 obj = obj->backing_object;
1189
1190 /* Ignore memory-mapped devices and such things. */
1191 if (obj->type != OBJT_DEFAULT &&
1192 obj->type != OBJT_SWAP &&
1193 obj->type != OBJT_VNODE)
1194 continue;
1195
1196 error = (*func)(entry, closure);
1197 }
1198 return error;
1199 }
1200
1201 static
1202 void *
1203 target_reserve(elf_buf_t target, size_t bytes, int *error)
1204 {
1205 void *res = NULL;
1206
1207 if (target->buf) {
1208 if (target->off + bytes > target->off_max)
1209 *error = EINVAL;
1210 else
1211 res = target->buf + target->off;
1212 }
1213 target->off += bytes;
1214 return (res);
1215 }
1216
1217 /*
1218 * Write the core file header to the file, including padding up to
1219 * the page boundary.
1220 */
1221 static int
1222 elf_corehdr(struct lwp *lp, int sig, struct file *fp, struct ucred *cred, int numsegs,
1223 elf_buf_t target)
1224 {
1225 /* XXX lwp handle more than one lwp */
1226 struct proc *p = lp->lwp_proc;
1227 struct {
1228 prstatus_t status;
1229 prfpregset_t fpregset;
1230 prpsinfo_t psinfo;
1231 } *tempdata;
1232 int error;
1233 prstatus_t *status;
1234 prfpregset_t *fpregset;
1235 prpsinfo_t *psinfo;
1236 int nbytes;
1237
1238 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_ZERO | M_WAITOK);
1239 status = &tempdata->status;
1240 fpregset = &tempdata->fpregset;
1241 psinfo = &tempdata->psinfo;
1242
1243 /* Gather the information for the header. */
1244 status->pr_version = PRSTATUS_VERSION;
1245 status->pr_statussz = sizeof(prstatus_t);
1246 status->pr_gregsetsz = sizeof(gregset_t);
1247 status->pr_fpregsetsz = sizeof(fpregset_t);
1248 status->pr_osreldate = osreldate;
1249 status->pr_cursig = sig;
1250 status->pr_pid = p->p_pid;
1251 fill_regs(lp, &status->pr_reg);
1252
1253 fill_fpregs(lp, fpregset);
1254
1255 psinfo->pr_version = PRPSINFO_VERSION;
1256 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1257 strncpy(psinfo->pr_fname, p->p_comm, sizeof(psinfo->pr_fname) - 1);
1258
1259 /* XXX - We don't fill in the command line arguments properly yet. */
1260 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1261
1262 /*
1263 * Fill in the header. The fp is passed so we can detect and flag
1264 * a checkpoint file pointer within the core file itself, because
1265 * it may not be restored from the same file handle.
1266 */
1267 error = elf_puthdr(lp, target, status, fpregset, psinfo, numsegs, fp);
1268
1269 kfree(tempdata, M_TEMP);
1270
1271 /* Write it to the core file. */
1272 if (error == 0) {
1273 error = fp_write(fp, target->buf, target->off, &nbytes,
1274 UIO_SYSSPACE);
1275 }
1276 return error;
1277 }
1278
1279 static int
1280 elf_puthdr(struct lwp *lp, elf_buf_t target, const prstatus_t *status,
1281 const prfpregset_t *fpregset, const prpsinfo_t *psinfo, int numsegs,
1282 struct file *fp)
1283 {
1284 struct proc *p = lp->lwp_proc;
1285 int error = 0;
1286 size_t phoff;
1287 size_t noteoff;
1288 size_t notesz;
1289 Elf_Ehdr *ehdr;
1290 Elf_Phdr *phdr;
1291
1292 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1293
1294 phoff = target->off;
1295 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1296
1297 noteoff = target->off;
1298 if (error == 0) {
1299 error = elf_putnote(target, "FreeBSD", NT_PRSTATUS,
1300 status, sizeof *status);
1301 }
1302 if (error == 0) {
1303 error = elf_putnote(target, "FreeBSD", NT_FPREGSET,
1304 fpregset, sizeof *fpregset);
1305 }
1306 if (error == 0) {
1307 error = elf_putnote(target, "FreeBSD", NT_PRPSINFO,
1308 psinfo, sizeof *psinfo);
1309 }
1310 notesz = target->off - noteoff;
1311
1312 /*
1313 * put extra cruft for dumping process state here
1314 * - we really want it be before all the program
1315 * mappings
1316 * - we just need to update the offset accordingly
1317 * and GDB will be none the wiser.
1318 */
1319 if (error == 0)
1320 error = elf_puttextvp(p, target);
1321 if (error == 0)
1322 error = elf_putsigs(lp, target);
1323 if (error == 0)
1324 error = elf_putfiles(p, target, fp);
1325
1326 /*
1327 * Align up to a page boundary for the program segments. The
1328 * actual data will be written to the outptu file, not to elf_buf_t,
1329 * so we do not have to do any further bounds checking.
1330 */
1331 target->off = round_page(target->off);
1332 if (error == 0 && ehdr != NULL) {
1333 /*
1334 * Fill in the ELF header.
1335 */
1336 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1337 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1338 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1339 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1340 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1341 ehdr->e_ident[EI_DATA] = ELF_DATA;
1342 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1343 ehdr->e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
1344 ehdr->e_ident[EI_ABIVERSION] = 0;
1345 ehdr->e_ident[EI_PAD] = 0;
1346 ehdr->e_type = ET_CORE;
1347 ehdr->e_machine = ELF_ARCH;
1348 ehdr->e_version = EV_CURRENT;
1349 ehdr->e_entry = 0;
1350 ehdr->e_phoff = phoff;
1351 ehdr->e_flags = 0;
1352 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1353 ehdr->e_phentsize = sizeof(Elf_Phdr);
1354 ehdr->e_phnum = numsegs + 1;
1355 ehdr->e_shentsize = sizeof(Elf_Shdr);
1356 ehdr->e_shnum = 0;
1357 ehdr->e_shstrndx = SHN_UNDEF;
1358 }
1359 if (error == 0 && phdr != NULL) {
1360 /*
1361 * Fill in the program header entries.
1362 */
1363 struct phdr_closure phc;
1364
1365 /* The note segement. */
1366 phdr->p_type = PT_NOTE;
1367 phdr->p_offset = noteoff;
1368 phdr->p_vaddr = 0;
1369 phdr->p_paddr = 0;
1370 phdr->p_filesz = notesz;
1371 phdr->p_memsz = 0;
1372 phdr->p_flags = 0;
1373 phdr->p_align = 0;
1374 ++phdr;
1375
1376 /* All the writable segments from the program. */
1377 phc.phdr = phdr;
1378 phc.phdr_max = phdr + numsegs;
1379 phc.offset = target->off;
1380 each_segment(p, cb_put_phdr, &phc, 1);
1381 }
1382 return (error);
1383 }
1384
1385 static int
1386 elf_putnote(elf_buf_t target, const char *name, int type,
1387 const void *desc, size_t descsz)
1388 {
1389 int error = 0;
1390 char *dst;
1391 Elf_Note note;
1392
1393 note.n_namesz = strlen(name) + 1;
1394 note.n_descsz = descsz;
1395 note.n_type = type;
1396 dst = target_reserve(target, sizeof(note), &error);
1397 if (dst != NULL)
1398 bcopy(&note, dst, sizeof note);
1399 dst = target_reserve(target, note.n_namesz, &error);
1400 if (dst != NULL)
1401 bcopy(name, dst, note.n_namesz);
1402 target->off = roundup2(target->off, sizeof(Elf_Size));
1403 dst = target_reserve(target, note.n_descsz, &error);
1404 if (dst != NULL)
1405 bcopy(desc, dst, note.n_descsz);
1406 target->off = roundup2(target->off, sizeof(Elf_Size));
1407 return(error);
1408 }
1409
1410
1411 static int
1412 elf_putsigs(struct lwp *lp, elf_buf_t target)
1413 {
1414 /* XXX lwp handle more than one lwp */
1415 struct proc *p = lp->lwp_proc;
1416 int error = 0;
1417 struct ckpt_siginfo *csi;
1418
1419 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1420 if (csi) {
1421 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
1422 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts));
1423 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
1424 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask,
1425 sizeof(sigset_t));
1426 csi->csi_sigparent = p->p_sigparent;
1427 }
1428 return(error);
1429 }
1430
1431 static int
1432 elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp)
1433 {
1434 int error = 0;
1435 int i;
1436 struct ckpt_filehdr *cfh = NULL;
1437 struct ckpt_fileinfo *cfi;
1438 struct file *fp;
1439 struct vnode *vp;
1440 /*
1441 * the duplicated loop is gross, but it was the only way
1442 * to eliminate uninitialized variable warnings
1443 */
1444 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1445 if (cfh) {
1446 cfh->cfh_nfiles = 0;
1447 }
1448
1449 /*
1450 * ignore STDIN/STDERR/STDOUT.
1451 */
1452 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
1453 fp = holdfp(p->p_fd, i, -1);
1454 if (fp == NULL)
1455 continue;
1456 /*
1457 * XXX Only checkpoint vnodes for now.
1458 */
1459 if (fp->f_type != DTYPE_VNODE) {
1460 fdrop(fp);
1461 continue;
1462 }
1463 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo),
1464 &error);
1465 if (cfi == NULL) {
1466 fdrop(fp);
1467 continue;
1468 }
1469 cfi->cfi_index = -1;
1470 cfi->cfi_type = fp->f_type;
1471 cfi->cfi_flags = fp->f_flag;
1472 cfi->cfi_offset = fp->f_offset;
1473 cfi->cfi_ckflags = 0;
1474
1475 if (fp == ckfp)
1476 cfi->cfi_ckflags |= CKFIF_ISCKPTFD;
1477 /* f_count and f_msgcount should not be saved/restored */
1478 /* XXX save cred info */
1479
1480 switch(fp->f_type) {
1481 case DTYPE_VNODE:
1482 vp = (struct vnode *)fp->f_data;
1483 /*
1484 * it looks like a bug in ptrace is marking
1485 * a non-vnode as a vnode - until we find the
1486 * root cause this will at least prevent
1487 * further panics from truss
1488 */
1489 if (vp == NULL || vp->v_mount == NULL)
1490 break;
1491 cfh->cfh_nfiles++;
1492 cfi->cfi_index = i;
1493 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1494 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
1495 break;
1496 default:
1497 break;
1498 }
1499 fdrop(fp);
1500 }
1501 return(error);
1502 }
1503
1504 static int
1505 elf_puttextvp(struct proc *p, elf_buf_t target)
1506 {
1507 int error = 0;
1508 int *vn_count;
1509 struct fp_closure fpc;
1510 struct ckpt_vminfo *vminfo;
1511
1512 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1513 if (vminfo != NULL) {
1514 vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1515 vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1516 vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1517 vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
1518 }
1519
1520 fpc.count = 0;
1521 vn_count = target_reserve(target, sizeof(int), &error);
1522 if (target->buf != NULL) {
1523 fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1524 fpc.vnh_max = fpc.vnh +
1525 (target->off_max - target->off) / sizeof(struct vn_hdr);
1526 error = each_segment(p, cb_put_fp, &fpc, 0);
1527 if (vn_count)
1528 *vn_count = fpc.count;
1529 } else {
1530 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
1531 }
1532 target->off += fpc.count * sizeof(struct vn_hdr);
1533 return(error);
1534 }
1535
1536
1537 /*
1538 * Tell kern_execve.c about it, with a little help from the linker.
1539 */
1540 static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1541 EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST);
DragonFly BSD