kernel - Optimize the x86-64 lwbuf API
[dragonfly.git] / sys / kern / imgact_elf.c
CommitLineData
984263bc 1/*-
08d72226 2 * Copyright (c) 1995-1996 Søren Schmidt
984263bc
MD
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
4d9022e3 16 * derived from this software without specific prior written permission
984263bc
MD
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 $
d913b2fa 30 * $DragonFly: src/sys/kern/imgact_elf.c,v 1.55 2008/08/17 17:21:36 nth Exp $
984263bc
MD
31 */
32
33#include <sys/param.h>
34#include <sys/exec.h>
35#include <sys/fcntl.h>
731100e5 36#include <sys/file.h>
984263bc
MD
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>
dadab5e9
MD
42#include <sys/systm.h>
43#include <sys/proc.h>
fad57d0e 44#include <sys/nlookup.h>
984263bc 45#include <sys/pioctl.h>
984263bc
MD
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>
984263bc 53#include <sys/vnode.h>
8ba5f7ef 54#include <sys/eventhandler.h>
984263bc 55
5c5185ae
SG
56#include <cpu/lwbuf.h>
57
984263bc
MD
58#include <vm/vm.h>
59#include <vm/vm_kern.h>
60#include <vm/vm_param.h>
61#include <vm/pmap.h>
62#include <sys/lock.h>
63#include <vm/vm_map.h>
64#include <vm/vm_object.h>
65#include <vm/vm_extern.h>
66
67#include <machine/elf.h>
68#include <machine/md_var.h>
731100e5
MD
69#include <sys/mount.h>
70#include <sys/ckpt.h>
984263bc
MD
71#define OLD_EI_BRAND 8
72
73__ElfType(Brandinfo);
74__ElfType(Auxargs);
75
402ed7e1
RG
76static int elf_check_header (const Elf_Ehdr *hdr);
77static int elf_freebsd_fixup (register_t **stack_base,
78 struct image_params *imgp);
79static int elf_load_file (struct proc *p, const char *file, u_long *addr,
80 u_long *entry);
81static int elf_load_section (struct proc *p,
984263bc
MD
82 struct vmspace *vmspace, struct vnode *vp,
83 vm_offset_t offset, caddr_t vmaddr, size_t memsz, size_t filsz,
402ed7e1
RG
84 vm_prot_t prot);
85static int exec_elf_imgact (struct image_params *imgp);
984263bc
MD
86
87static int elf_trace = 0;
88SYSCTL_INT(_debug, OID_AUTO, elf_trace, CTLFLAG_RW, &elf_trace, 0, "");
89static int elf_legacy_coredump = 0;
90SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
0c52fa62
SG
91 &elf_legacy_coredump, 0,
92 "Do not dump inaccessible mappings, legacy coredump mode");
984263bc 93
8d0415e1
JS
94static int dragonfly_match_abi_note(const Elf_Note *);
95static int freebsd_match_abi_note(const Elf_Note *);
96
984263bc
MD
97static struct sysentvec elf_freebsd_sysvec = {
98 SYS_MAXSYSCALL,
99 sysent,
8ec60c3f 100 -1,
984263bc
MD
101 0,
102 0,
103 0,
104 0,
105 0,
106 elf_freebsd_fixup,
107 sendsig,
108 sigcode,
109 &szsigcode,
110 0,
1799f7b1
AHJ
111#if defined(__x86_64__)
112 "FreeBSD ELF64",
113#else
114 "FreeBSD ELF32",
115#endif
984263bc
MD
116 elf_coredump,
117 NULL,
118 MINSIGSTKSZ
119};
120
121static Elf_Brandinfo freebsd_brand_info = {
122 ELFOSABI_FREEBSD,
123 "FreeBSD",
8d0415e1 124 freebsd_match_abi_note,
984263bc
MD
125 "",
126 "/usr/libexec/ld-elf.so.1",
127 &elf_freebsd_sysvec
128 };
8d0415e1
JS
129
130static Elf_Brandinfo dragonfly_brand_info = {
131 ELFOSABI_NONE,
132 "DragonFly",
133 dragonfly_match_abi_note,
134 "",
135 "/usr/libexec/ld-elf.so.2",
136 &elf_freebsd_sysvec
137 };
138
984263bc 139static Elf_Brandinfo *elf_brand_list[MAX_BRANDS] = {
8d0415e1 140 &dragonfly_brand_info,
984263bc
MD
141 &freebsd_brand_info,
142 NULL, NULL, NULL,
8d0415e1 143 NULL, NULL, NULL
984263bc
MD
144 };
145
8d0415e1
JS
146static int
147freebsd_match_abi_note(const Elf_Note *abi_note)
148{
149 const char *abi_name = (const char *)
150 ((const uint8_t *)abi_note + sizeof(*abi_note));
151
152 if (abi_note->n_namesz != sizeof("FreeBSD"))
153 return(FALSE);
154 if (memcmp(abi_name, "FreeBSD", sizeof("FreeBSD")))
155 return(FALSE);
156 return(TRUE);
157}
158
159static int
160dragonfly_match_abi_note(const Elf_Note *abi_note)
161{
162 const char *abi_name = (const char *)
163 ((const uint8_t *)abi_note + sizeof(*abi_note));
164
165 if (abi_note->n_namesz != sizeof("DragonFly"))
166 return(FALSE);
167 if (memcmp(abi_name, "DragonFly", sizeof("DragonFly")))
168 return(FALSE);
169 return(TRUE);
170}
171
984263bc
MD
172int
173elf_insert_brand_entry(Elf_Brandinfo *entry)
174{
175 int i;
176
177 for (i=1; i<MAX_BRANDS; i++) {
178 if (elf_brand_list[i] == NULL) {
179 elf_brand_list[i] = entry;
180 break;
181 }
182 }
183 if (i == MAX_BRANDS)
184 return -1;
185 return 0;
186}
187
188int
189elf_remove_brand_entry(Elf_Brandinfo *entry)
190{
191 int i;
192
193 for (i=1; i<MAX_BRANDS; i++) {
194 if (elf_brand_list[i] == entry) {
195 elf_brand_list[i] = NULL;
196 break;
197 }
198 }
199 if (i == MAX_BRANDS)
200 return -1;
201 return 0;
202}
203
0e5797fe
MD
204/*
205 * Check if an elf brand is being used anywhere in the system.
206 *
7bd34050 207 * Used by the linux emulation module unloader. This isn't safe from
0e5797fe
MD
208 * races.
209 */
210struct elf_brand_inuse_info {
211 int rval;
212 Elf_Brandinfo *entry;
213};
214
215static int elf_brand_inuse_callback(struct proc *p, void *data);
216
984263bc
MD
217int
218elf_brand_inuse(Elf_Brandinfo *entry)
219{
0e5797fe 220 struct elf_brand_inuse_info info;
984263bc 221
0e5797fe
MD
222 info.rval = FALSE;
223 info.entry = entry;
224 allproc_scan(elf_brand_inuse_callback, entry);
225 return (info.rval);
226}
227
228static
229int
230elf_brand_inuse_callback(struct proc *p, void *data)
231{
232 struct elf_brand_inuse_info *info = data;
984263bc 233
0e5797fe
MD
234 if (p->p_sysent == info->entry->sysvec) {
235 info->rval = TRUE;
236 return(-1);
237 }
238 return(0);
984263bc
MD
239}
240
241static int
242elf_check_header(const Elf_Ehdr *hdr)
243{
244 if (!IS_ELF(*hdr) ||
245 hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
246 hdr->e_ident[EI_DATA] != ELF_TARG_DATA ||
2cc7c579
DR
247 hdr->e_ident[EI_VERSION] != EV_CURRENT ||
248 hdr->e_phentsize != sizeof(Elf_Phdr) ||
249 hdr->e_ehsize != sizeof(Elf_Ehdr) ||
250 hdr->e_version != ELF_TARG_VER)
984263bc
MD
251 return ENOEXEC;
252
253 if (!ELF_MACHINE_OK(hdr->e_machine))
254 return ENOEXEC;
255
984263bc
MD
256 return 0;
257}
258
7c34d798
SS
259static Elf_Brandinfo *
260elf_check_abi_note(struct image_params *imgp, const Elf_Phdr *ph)
261{
262 Elf_Brandinfo *match = NULL;
263 const Elf_Note *tmp_note;
5c5185ae 264 struct lwbuf *lwb;
7a683a24 265 struct lwbuf lwb_cache;
7c34d798
SS
266 const char *page;
267 char *data = NULL;
268 Elf_Off off;
269 size_t firstoff;
270 size_t len;
271 size_t firstlen;
272
273 len = ph->p_filesz;
274 off = ph->p_offset;
275
276 firstoff = off & PAGE_MASK;
277 firstlen = PAGE_SIZE - firstoff;
278
279 if (len < sizeof(Elf_Note) || len > PAGE_SIZE)
280 return NULL; /* ENOEXEC? */
281
7a683a24 282 lwb = &lwb_cache;
5c5185ae 283 if (exec_map_page(imgp, off >> PAGE_SHIFT, &lwb, &page))
7c34d798
SS
284 return NULL;
285
286 /*
287 * Crosses page boundary? Is that allowed?
288 */
289 if (firstlen < len) {
290 data = kmalloc(len, M_TEMP, M_WAITOK);
291
292 bcopy(page + firstoff, data, firstlen);
293
5c5185ae 294 exec_unmap_page(lwb);
7a683a24 295 lwb = &lwb_cache;
5c5185ae 296 if (exec_map_page(imgp, (off >> PAGE_SHIFT) + 1, &lwb, &page)) {
7c34d798
SS
297 kfree(data, M_TEMP);
298 return NULL;
299 }
43297079 300 bcopy(page, data + firstlen, len - firstlen);
7c34d798
SS
301 tmp_note = (void *)data;
302 } else {
303 tmp_note = (const void *)(page + firstoff);
304 }
305
306 while (len >= sizeof(Elf_Note)) {
307 int i;
308 size_t nlen = roundup(tmp_note->n_namesz, sizeof(Elf_Word)) +
309 roundup(tmp_note->n_descsz, sizeof(Elf_Word)) +
310 sizeof(Elf_Note);
311
312 if (nlen > len)
313 break;
314
315 if (tmp_note->n_type != 1)
316 goto next;
317
318 for (i = 0; i < MAX_BRANDS; i++) {
319 Elf_Brandinfo *bi = elf_brand_list[i];
320
321 if (bi != NULL && bi->match_abi_note != NULL &&
322 bi->match_abi_note(tmp_note)) {
323 match = bi;
324 break;
325 }
326 }
327
328 if (match != NULL)
329 break;
330
331next:
332 len -= nlen;
333 tmp_note += nlen;
334 }
335
336 if (data != NULL)
337 kfree(data, M_TEMP);
5c5185ae 338 exec_unmap_page(lwb);
7c34d798
SS
339
340 return (match);
341}
342
984263bc 343static int
731100e5 344elf_load_section(struct proc *p, struct vmspace *vmspace, struct vnode *vp,
57f7b636
MD
345 vm_offset_t offset, caddr_t vmaddr, size_t memsz,
346 size_t filsz, vm_prot_t prot)
984263bc
MD
347{
348 size_t map_len;
349 vm_offset_t map_addr;
350 int error, rv, cow;
a108bf71 351 int count;
984263bc
MD
352 size_t copy_len;
353 vm_object_t object;
354 vm_offset_t file_addr;
984263bc 355
7540ab49 356 object = vp->v_object;
984263bc
MD
357 error = 0;
358
359 /*
360 * It's necessary to fail if the filsz + offset taken from the
361 * header is greater than the actual file pager object's size.
362 * If we were to allow this, then the vm_map_find() below would
363 * walk right off the end of the file object and into the ether.
364 *
365 * While I'm here, might as well check for something else that
366 * is invalid: filsz cannot be greater than memsz.
367 */
57f7b636 368 if ((off_t)filsz + offset > vp->v_filesize || filsz > memsz) {
984263bc
MD
369 uprintf("elf_load_section: truncated ELF file\n");
370 return (ENOEXEC);
371 }
372
373 map_addr = trunc_page((vm_offset_t)vmaddr);
374 file_addr = trunc_page(offset);
375
376 /*
377 * We have two choices. We can either clear the data in the last page
378 * of an oversized mapping, or we can start the anon mapping a page
379 * early and copy the initialized data into that first page. We
380 * choose the second..
381 */
382 if (memsz > filsz)
383 map_len = trunc_page(offset+filsz) - file_addr;
384 else
385 map_len = round_page(offset+filsz) - file_addr;
386
387 if (map_len != 0) {
388 vm_object_reference(object);
389
390 /* cow flags: don't dump readonly sections in core */
391 cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
392 (prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
393
a108bf71 394 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 395 vm_map_lock(&vmspace->vm_map);
a108bf71 396 rv = vm_map_insert(&vmspace->vm_map, &count,
984263bc
MD
397 object,
398 file_addr, /* file offset */
399 map_addr, /* virtual start */
400 map_addr + map_len,/* virtual end */
1b874851
MD
401 VM_MAPTYPE_NORMAL,
402 prot, VM_PROT_ALL,
984263bc
MD
403 cow);
404 vm_map_unlock(&vmspace->vm_map);
a108bf71 405 vm_map_entry_release(count);
984263bc
MD
406 if (rv != KERN_SUCCESS) {
407 vm_object_deallocate(object);
408 return EINVAL;
409 }
410
411 /* we can stop now if we've covered it all */
412 if (memsz == filsz) {
413 return 0;
414 }
415 }
416
417
418 /*
419 * We have to get the remaining bit of the file into the first part
420 * of the oversized map segment. This is normally because the .data
421 * segment in the file is extended to provide bss. It's a neat idea
422 * to try and save a page, but it's a pain in the behind to implement.
423 */
424 copy_len = (offset + filsz) - trunc_page(offset + filsz);
425 map_addr = trunc_page((vm_offset_t)vmaddr + filsz);
426 map_len = round_page((vm_offset_t)vmaddr + memsz) - map_addr;
427
428 /* This had damn well better be true! */
429 if (map_len != 0) {
a108bf71 430 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 431 vm_map_lock(&vmspace->vm_map);
a108bf71
MD
432 rv = vm_map_insert(&vmspace->vm_map, &count,
433 NULL, 0,
984263bc 434 map_addr, map_addr + map_len,
1b874851
MD
435 VM_MAPTYPE_NORMAL,
436 VM_PROT_ALL, VM_PROT_ALL,
437 0);
984263bc 438 vm_map_unlock(&vmspace->vm_map);
a108bf71 439 vm_map_entry_release(count);
984263bc
MD
440 if (rv != KERN_SUCCESS) {
441 return EINVAL;
442 }
443 }
444
445 if (copy_len != 0) {
18b3f457 446 vm_page_t m;
5c5185ae 447 struct lwbuf *lwb;
7a683a24 448 struct lwbuf lwb_cache;
18b3f457
MD
449
450 m = vm_fault_object_page(object, trunc_page(offset + filsz),
451 VM_PROT_READ, 0, &error);
452 if (m) {
7a683a24 453 lwb = lwbuf_alloc(m, &lwb_cache);
5c5185ae 454 error = copyout((caddr_t)lwbuf_kva(lwb),
18b3f457 455 (caddr_t)map_addr, copy_len);
5c5185ae 456 lwbuf_free(lwb);
18b3f457 457 vm_page_unhold(m);
984263bc 458 }
984263bc
MD
459 if (error) {
460 return (error);
461 }
462 }
463
464 /*
465 * set it to the specified protection
466 */
467 vm_map_protect(&vmspace->vm_map, map_addr, map_addr + map_len, prot,
468 FALSE);
469
470 return error;
471}
472
473/*
474 * Load the file "file" into memory. It may be either a shared object
475 * or an executable.
476 *
477 * The "addr" reference parameter is in/out. On entry, it specifies
478 * the address where a shared object should be loaded. If the file is
479 * an executable, this value is ignored. On exit, "addr" specifies
480 * where the file was actually loaded.
481 *
482 * The "entry" reference parameter is out only. On exit, it specifies
483 * the entry point for the loaded file.
484 */
485static int
486elf_load_file(struct proc *p, const char *file, u_long *addr, u_long *entry)
487{
488 struct {
fad57d0e 489 struct nlookupdata nd;
984263bc
MD
490 struct vattr attr;
491 struct image_params image_params;
492 } *tempdata;
493 const Elf_Ehdr *hdr = NULL;
494 const Elf_Phdr *phdr = NULL;
fad57d0e 495 struct nlookupdata *nd;
984263bc
MD
496 struct vmspace *vmspace = p->p_vmspace;
497 struct vattr *attr;
498 struct image_params *imgp;
246693ac 499 struct mount *topmnt;
984263bc
MD
500 vm_prot_t prot;
501 u_long rbase;
502 u_long base_addr = 0;
503 int error, i, numsegs;
504
efda3bd0 505 tempdata = kmalloc(sizeof(*tempdata), M_TEMP, M_WAITOK);
984263bc
MD
506 nd = &tempdata->nd;
507 attr = &tempdata->attr;
508 imgp = &tempdata->image_params;
509
510 /*
511 * Initialize part of the common data
512 */
513 imgp->proc = p;
984263bc
MD
514 imgp->attr = attr;
515 imgp->firstpage = NULL;
18f40545 516 imgp->image_header = NULL;
fad57d0e 517 imgp->vp = NULL;
984263bc 518
fad57d0e
MD
519 error = nlookup_init(nd, file, UIO_SYSSPACE, NLC_FOLLOW);
520 if (error == 0)
521 error = nlookup(nd);
522 if (error == 0)
28623bf9 523 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
246693ac 524 topmnt = nd->nl_nch.mount;
fad57d0e
MD
525 nlookup_done(nd);
526 if (error)
984263bc 527 goto fail;
984263bc
MD
528
529 /*
530 * Check permissions, modes, uid, etc on the file, and "open" it.
531 */
246693ac 532 error = exec_check_permissions(imgp, topmnt);
984263bc 533 if (error) {
a11aaa81 534 vn_unlock(imgp->vp);
984263bc
MD
535 goto fail;
536 }
537
538 error = exec_map_first_page(imgp);
539 /*
540 * Also make certain that the interpreter stays the same, so set
541 * its VTEXT flag, too.
542 */
543 if (error == 0)
2247fe02 544 vsetflags(imgp->vp, VTEXT);
a11aaa81 545 vn_unlock(imgp->vp);
984263bc
MD
546 if (error)
547 goto fail;
548
549 hdr = (const Elf_Ehdr *)imgp->image_header;
550 if ((error = elf_check_header(hdr)) != 0)
551 goto fail;
552 if (hdr->e_type == ET_DYN)
553 rbase = *addr;
554 else if (hdr->e_type == ET_EXEC)
555 rbase = 0;
556 else {
557 error = ENOEXEC;
558 goto fail;
559 }
560
2cc7c579
DR
561 /* Only support headers that fit within first page for now
562 * (multiplication of two Elf_Half fields will not overflow) */
984263bc 563 if ((hdr->e_phoff > PAGE_SIZE) ||
2cc7c579 564 (hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE - hdr->e_phoff) {
984263bc
MD
565 error = ENOEXEC;
566 goto fail;
567 }
568
569 phdr = (const Elf_Phdr *)(imgp->image_header + hdr->e_phoff);
570
571 for (i = 0, numsegs = 0; i < hdr->e_phnum; i++) {
572 if (phdr[i].p_type == PT_LOAD) { /* Loadable segment */
573 prot = 0;
574 if (phdr[i].p_flags & PF_X)
575 prot |= VM_PROT_EXECUTE;
576 if (phdr[i].p_flags & PF_W)
577 prot |= VM_PROT_WRITE;
578 if (phdr[i].p_flags & PF_R)
579 prot |= VM_PROT_READ;
580
dadab5e9 581 error = elf_load_section(
fad57d0e 582 p, vmspace, imgp->vp,
dadab5e9
MD
583 phdr[i].p_offset,
584 (caddr_t)phdr[i].p_vaddr +
585 rbase,
586 phdr[i].p_memsz,
587 phdr[i].p_filesz, prot);
588 if (error != 0)
984263bc
MD
589 goto fail;
590 /*
591 * Establish the base address if this is the
592 * first segment.
593 */
594 if (numsegs == 0)
595 base_addr = trunc_page(phdr[i].p_vaddr + rbase);
596 numsegs++;
597 }
598 }
599 *addr = base_addr;
600 *entry=(unsigned long)hdr->e_entry + rbase;
601
602fail:
603 if (imgp->firstpage)
604 exec_unmap_first_page(imgp);
fad57d0e
MD
605 if (imgp->vp) {
606 vrele(imgp->vp);
607 imgp->vp = NULL;
608 }
efda3bd0 609 kfree(tempdata, M_TEMP);
984263bc
MD
610
611 return error;
612}
613
614/*
615 * non static, as it can be overridden by start_init()
616 */
617int fallback_elf_brand = -1;
618SYSCTL_INT(_kern, OID_AUTO, fallback_elf_brand, CTLFLAG_RW,
619 &fallback_elf_brand, -1,
620 "ELF brand of last resort");
621
08d72226
AH
622static int can_exec_dyn = 1;
623SYSCTL_INT(_kern, OID_AUTO, elf_exec_dyn, CTLFLAG_RW,
624 &can_exec_dyn, 1,
625 "ELF: can exec shared libraries");
626
984263bc
MD
627static int
628exec_elf_imgact(struct image_params *imgp)
629{
630 const Elf_Ehdr *hdr = (const Elf_Ehdr *) imgp->image_header;
631 const Elf_Phdr *phdr;
632 Elf_Auxargs *elf_auxargs = NULL;
633 struct vmspace *vmspace;
634 vm_prot_t prot;
635 u_long text_size = 0, data_size = 0, total_size = 0;
636 u_long text_addr = 0, data_addr = 0;
637 u_long seg_size, seg_addr;
638 u_long addr, entry = 0, proghdr = 0;
639 int error, i;
640 const char *interp = NULL;
8d0415e1 641 const Elf_Note *abi_note = NULL;
7c34d798 642 Elf_Brandinfo *brand_info = NULL;
984263bc
MD
643 char *path;
644
2bd9d75c
DRJ
645 error = 0;
646
984263bc
MD
647 /*
648 * Do we have a valid ELF header ?
08d72226 649 * We allow execution of ET_EXEC and, if kern.elf_exec_dyn is 1, ET_DYN.
984263bc 650 */
08d72226
AH
651 if (elf_check_header(hdr) != 0 ||
652 (hdr->e_type != ET_EXEC && (!can_exec_dyn || hdr->e_type != ET_DYN)))
984263bc
MD
653 return -1;
654
655 /*
656 * From here on down, we return an errno, not -1, as we've
657 * detected an ELF file.
658 */
659
660 if ((hdr->e_phoff > PAGE_SIZE) ||
661 (hdr->e_phoff + hdr->e_phentsize * hdr->e_phnum) > PAGE_SIZE) {
662 /* Only support headers in first page for now */
663 return ENOEXEC;
664 }
665 phdr = (const Elf_Phdr*)(imgp->image_header + hdr->e_phoff);
666
667 /*
668 * From this point on, we may have resources that need to be freed.
669 */
670
29802dbb 671 exec_new_vmspace(imgp, NULL);
984263bc
MD
672
673 /*
674 * Yeah, I'm paranoid. There is every reason in the world to get
675 * VTEXT now since from here on out, there are places we can have
676 * a context switch. Better safe than sorry; I really don't want
677 * the file to change while it's being loaded.
678 */
5fd012e0 679 vsetflags(imgp->vp, VTEXT);
984263bc
MD
680
681 vmspace = imgp->proc->p_vmspace;
682
683 for (i = 0; i < hdr->e_phnum; i++) {
684 switch(phdr[i].p_type) {
685
686 case PT_LOAD: /* Loadable segment */
687 prot = 0;
688 if (phdr[i].p_flags & PF_X)
689 prot |= VM_PROT_EXECUTE;
690 if (phdr[i].p_flags & PF_W)
691 prot |= VM_PROT_WRITE;
692 if (phdr[i].p_flags & PF_R)
693 prot |= VM_PROT_READ;
694
695 if ((error = elf_load_section(imgp->proc,
696 vmspace, imgp->vp,
697 phdr[i].p_offset,
698 (caddr_t)phdr[i].p_vaddr,
699 phdr[i].p_memsz,
700 phdr[i].p_filesz, prot)) != 0)
701 goto fail;
702
22cc9971
DX
703 /*
704 * If this segment contains the program headers,
705 * remember their virtual address for the AT_PHDR
706 * aux entry. Static binaries don't usually include
707 * a PT_PHDR entry.
708 */
709 if (phdr[i].p_offset == 0 &&
710 hdr->e_phoff + hdr->e_phnum * hdr->e_phentsize
711 <= phdr[i].p_filesz)
712 proghdr = phdr[i].p_vaddr + hdr->e_phoff;
713
984263bc
MD
714 seg_addr = trunc_page(phdr[i].p_vaddr);
715 seg_size = round_page(phdr[i].p_memsz +
716 phdr[i].p_vaddr - seg_addr);
717
718 /*
719 * Is this .text or .data? We can't use
720 * VM_PROT_WRITE or VM_PROT_EXEC, it breaks the
721 * alpha terribly and possibly does other bad
722 * things so we stick to the old way of figuring
723 * it out: If the segment contains the program
724 * entry point, it's a text segment, otherwise it
725 * is a data segment.
726 *
727 * Note that obreak() assumes that data_addr +
728 * data_size == end of data load area, and the ELF
729 * file format expects segments to be sorted by
730 * address. If multiple data segments exist, the
731 * last one will be used.
732 */
733 if (hdr->e_entry >= phdr[i].p_vaddr &&
734 hdr->e_entry < (phdr[i].p_vaddr +
735 phdr[i].p_memsz)) {
736 text_size = seg_size;
737 text_addr = seg_addr;
738 entry = (u_long)hdr->e_entry;
739 } else {
740 data_size = seg_size;
741 data_addr = seg_addr;
742 }
743 total_size += seg_size;
744
745 /*
746 * Check limits. It should be safe to check the
747 * limits after loading the segment since we do
748 * not actually fault in all the segment's pages.
749 */
750 if (data_size >
751 imgp->proc->p_rlimit[RLIMIT_DATA].rlim_cur ||
752 text_size > maxtsiz ||
753 total_size >
754 imgp->proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
755 error = ENOMEM;
756 goto fail;
757 }
758 break;
759 case PT_INTERP: /* Path to interpreter */
760 if (phdr[i].p_filesz > MAXPATHLEN ||
761 phdr[i].p_offset + phdr[i].p_filesz > PAGE_SIZE) {
762 error = ENOEXEC;
763 goto fail;
764 }
765 interp = imgp->image_header + phdr[i].p_offset;
766 break;
8d0415e1 767 case PT_NOTE: /* Check for .note.ABI-tag */
7c34d798
SS
768 if (brand_info == NULL)
769 brand_info = elf_check_abi_note(imgp, &phdr[i]);
8d0415e1 770 break;
984263bc
MD
771 case PT_PHDR: /* Program header table info */
772 proghdr = phdr[i].p_vaddr;
773 break;
774 default:
775 break;
776 }
777 }
778
779 vmspace->vm_tsize = text_size >> PAGE_SHIFT;
780 vmspace->vm_taddr = (caddr_t)(uintptr_t)text_addr;
781 vmspace->vm_dsize = data_size >> PAGE_SHIFT;
782 vmspace->vm_daddr = (caddr_t)(uintptr_t)data_addr;
783
784 addr = ELF_RTLD_ADDR(vmspace);
785
786 imgp->entry_addr = entry;
787
984263bc
MD
788 /* We support three types of branding -- (1) the ELF EI_OSABI field
789 * that SCO added to the ELF spec, (2) FreeBSD 3.x's traditional string
790 * branding w/in the ELF header, and (3) path of the `interp_path'
791 * field. We should also look for an ".note.ABI-tag" ELF section now
792 * in all Linux ELF binaries, FreeBSD 4.1+, and some NetBSD ones.
793 */
794
795 /* If the executable has a brand, search for it in the brand list. */
8d0415e1 796 if (brand_info == NULL && hdr->e_ident[EI_OSABI] != ELFOSABI_NONE) {
984263bc
MD
797 for (i = 0; i < MAX_BRANDS; i++) {
798 Elf_Brandinfo *bi = elf_brand_list[i];
799
800 if (bi != NULL &&
801 (hdr->e_ident[EI_OSABI] == bi->brand
802 || 0 ==
803 strncmp((const char *)&hdr->e_ident[OLD_EI_BRAND],
804 bi->compat_3_brand, strlen(bi->compat_3_brand)))) {
805 brand_info = bi;
806 break;
807 }
808 }
809 }
810
8d0415e1
JS
811 /* Search for a recognized ABI. */
812 if (brand_info == NULL && abi_note != NULL) {
237770a0
JS
813 }
814
815 /*
816 * ELFOSABI_NONE == ELFOSABI_SYSV, so a SYSV binary misses all
817 * checks so far, since it is neither branded nor does it have
818 * an ABI note. If the EI_OSABI field is ELFOSABI_NONE, assume
819 * it is svr4 and look for an entry in the elf_brand_list with
820 * match_abi_note == NULL.
821 */
822 if (brand_info == NULL && hdr->e_ident[EI_OSABI] == ELFOSABI_NONE) {
823 for (i = 0; i < MAX_BRANDS; i++) {
824 Elf_Brandinfo *bi = elf_brand_list[i];
825
826 if (bi != NULL && bi->match_abi_note == NULL &&
827 ELFOSABI_SYSV == bi->brand) {
828 brand_info = bi;
8d0415e1
JS
829 break;
830 }
831 }
832 }
833
834 /* Lacking a recognized ABI, search for a recognized interpreter. */
984263bc
MD
835 if (brand_info == NULL && interp != NULL) {
836 for (i = 0; i < MAX_BRANDS; i++) {
837 Elf_Brandinfo *bi = elf_brand_list[i];
838
839 if (bi != NULL &&
840 strcmp(interp, bi->interp_path) == 0) {
841 brand_info = bi;
842 break;
843 }
844 }
845 }
846
847 /* Lacking a recognized interpreter, try the default brand */
848 if (brand_info == NULL) {
849 for (i = 0; i < MAX_BRANDS; i++) {
850 Elf_Brandinfo *bi = elf_brand_list[i];
851
852 if (bi != NULL && fallback_elf_brand == bi->brand) {
853 brand_info = bi;
854 break;
855 }
856 }
857 }
858
859 if (brand_info == NULL) {
860 uprintf("ELF binary type \"%u\" not known.\n",
861 hdr->e_ident[EI_OSABI]);
862 error = ENOEXEC;
863 goto fail;
7c03f43a 864 }
984263bc
MD
865
866 imgp->proc->p_sysent = brand_info->sysvec;
8ba5f7ef
AH
867 EVENTHANDLER_INVOKE(process_exec, imgp);
868
984263bc 869 if (interp != NULL) {
efda3bd0 870 path = kmalloc(MAXPATHLEN, M_TEMP, M_WAITOK);
f8c7a42d 871 ksnprintf(path, MAXPATHLEN, "%s%s",
984263bc
MD
872 brand_info->emul_path, interp);
873 if ((error = elf_load_file(imgp->proc, path, &addr,
874 &imgp->entry_addr)) != 0) {
875 if ((error = elf_load_file(imgp->proc, interp, &addr,
876 &imgp->entry_addr)) != 0) {
877 uprintf("ELF interpreter %s not found\n", path);
efda3bd0 878 kfree(path, M_TEMP);
984263bc
MD
879 goto fail;
880 }
881 }
efda3bd0 882 kfree(path, M_TEMP);
5347affc
MD
883 } else {
884 addr = 0;
984263bc
MD
885 }
886
887 /*
888 * Construct auxargs table (used by the fixup routine)
889 */
efda3bd0 890 elf_auxargs = kmalloc(sizeof(Elf_Auxargs), M_TEMP, M_WAITOK);
984263bc
MD
891 elf_auxargs->execfd = -1;
892 elf_auxargs->phdr = proghdr;
893 elf_auxargs->phent = hdr->e_phentsize;
894 elf_auxargs->phnum = hdr->e_phnum;
895 elf_auxargs->pagesz = PAGE_SIZE;
896 elf_auxargs->base = addr;
897 elf_auxargs->flags = 0;
898 elf_auxargs->entry = entry;
899 elf_auxargs->trace = elf_trace;
900
901 imgp->auxargs = elf_auxargs;
902 imgp->interpreted = 0;
903
904fail:
905 return error;
906}
907
908static int
909elf_freebsd_fixup(register_t **stack_base, struct image_params *imgp)
910{
911 Elf_Auxargs *args = (Elf_Auxargs *)imgp->auxargs;
912 register_t *pos;
913
2bd9d75c 914 pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2);
984263bc
MD
915
916 if (args->trace) {
917 AUXARGS_ENTRY(pos, AT_DEBUG, 1);
918 }
919 if (args->execfd != -1) {
920 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
921 }
922 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
923 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
924 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
925 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
926 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
927 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
928 AUXARGS_ENTRY(pos, AT_BASE, args->base);
929 AUXARGS_ENTRY(pos, AT_NULL, 0);
930
efda3bd0 931 kfree(imgp->auxargs, M_TEMP);
984263bc
MD
932 imgp->auxargs = NULL;
933
934 (*stack_base)--;
2bd9d75c 935 suword(*stack_base, (long) imgp->args->argc);
984263bc
MD
936 return 0;
937}
938
939/*
940 * Code for generating ELF core dumps.
941 */
942
55e44363 943typedef int (*segment_callback) (vm_map_entry_t, void *);
984263bc
MD
944
945/* Closure for cb_put_phdr(). */
946struct phdr_closure {
55e44363
MD
947 Elf_Phdr *phdr; /* Program header to fill in (incremented) */
948 Elf_Phdr *phdr_max; /* Pointer bound for error check */
984263bc
MD
949 Elf_Off offset; /* Offset of segment in core file */
950};
951
952/* Closure for cb_size_segment(). */
953struct sseg_closure {
954 int count; /* Count of writable segments. */
55e44363 955 size_t vsize; /* Total size of all writable segments. */
984263bc
MD
956};
957
55e44363 958/* Closure for cb_put_fp(). */
731100e5
MD
959struct fp_closure {
960 struct vn_hdr *vnh;
55e44363 961 struct vn_hdr *vnh_max;
731100e5
MD
962 int count;
963 struct stat *sb;
964};
965
55e44363
MD
966typedef struct elf_buf {
967 char *buf;
968 size_t off;
969 size_t off_max;
970} *elf_buf_t;
731100e5 971
55e44363 972static void *target_reserve(elf_buf_t target, size_t bytes, int *error);
731100e5 973
55e44363
MD
974static int cb_put_phdr (vm_map_entry_t, void *);
975static int cb_size_segment (vm_map_entry_t, void *);
976static int cb_fpcount_segment(vm_map_entry_t, void *);
977static int cb_put_fp(vm_map_entry_t, void *);
984263bc 978
731100e5 979
55e44363 980static int each_segment (struct proc *, segment_callback, void *, int);
7d20a8ff 981static int elf_corehdr (struct lwp *, int, struct file *, struct ucred *,
55e44363 982 int, elf_buf_t);
d913b2fa
NT
983enum putmode { WRITE, DRYRUN };
984static int elf_puthdr (struct lwp *, elf_buf_t, int sig, enum putmode,
985 int, struct file *);
986static int elf_putallnotes(struct lwp *, elf_buf_t, int, enum putmode);
55e44363 987static int elf_putnote (elf_buf_t, const char *, int, const void *, size_t);
731100e5 988
7d20a8ff 989static int elf_putsigs(struct lwp *, elf_buf_t);
55e44363 990static int elf_puttextvp(struct proc *, elf_buf_t);
ff7a3478 991static int elf_putfiles(struct proc *, elf_buf_t, struct file *);
731100e5 992
984263bc
MD
993extern int osreldate;
994
995int
7d20a8ff 996elf_coredump(struct lwp *lp, int sig, struct vnode *vp, off_t limit)
984263bc 997{
731100e5
MD
998 struct file *fp;
999 int error;
1000
1001 if ((error = falloc(NULL, &fp, NULL)) != 0)
1002 return (error);
7d20a8ff 1003 fsetcred(fp, lp->lwp_proc->p_ucred);
731100e5 1004
fad57d0e
MD
1005 /*
1006 * XXX fixme.
1007 */
fbb4eeab 1008 fp->f_type = DTYPE_VNODE;
731100e5 1009 fp->f_flag = O_CREAT|O_WRONLY|O_NOFOLLOW;
fad57d0e 1010 fp->f_ops = &vnode_fileops;
fbb4eeab 1011 fp->f_data = vp;
a11aaa81 1012 vn_unlock(vp);
731100e5 1013
7d20a8ff 1014 error = generic_elf_coredump(lp, sig, fp, limit);
731100e5 1015
fbb4eeab 1016 fp->f_type = 0;
731100e5
MD
1017 fp->f_flag = 0;
1018 fp->f_ops = &badfileops;
fbb4eeab 1019 fp->f_data = NULL;
9f87144f 1020 fdrop(fp);
731100e5
MD
1021 return (error);
1022}
1023
1024int
7d20a8ff 1025generic_elf_coredump(struct lwp *lp, int sig, struct file *fp, off_t limit)
731100e5 1026{
7d20a8ff 1027 struct proc *p = lp->lwp_proc;
dadab5e9 1028 struct ucred *cred = p->p_ucred;
984263bc
MD
1029 int error = 0;
1030 struct sseg_closure seginfo;
55e44363 1031 struct elf_buf target;
984263bc 1032
731100e5 1033 if (!fp)
6ea70f76 1034 kprintf("can't dump core - null fp\n");
55e44363
MD
1035
1036 /*
1037 * Size the program segments
1038 */
984263bc 1039 seginfo.count = 0;
55e44363 1040 seginfo.vsize = 0;
731100e5 1041 each_segment(p, cb_size_segment, &seginfo, 1);
984263bc
MD
1042
1043 /*
1044 * Calculate the size of the core file header area by making
1045 * a dry run of generating it. Nothing is written, but the
1046 * size is calculated.
1047 */
55e44363 1048 bzero(&target, sizeof(target));
d913b2fa 1049 elf_puthdr(lp, &target, sig, DRYRUN, seginfo.count, fp);
731100e5 1050
55e44363 1051 if (target.off + seginfo.vsize >= limit)
984263bc
MD
1052 return (EFAULT);
1053
1054 /*
1055 * Allocate memory for building the header, fill it up,
1056 * and write it out.
1057 */
55e44363
MD
1058 target.off_max = target.off;
1059 target.off = 0;
efda3bd0 1060 target.buf = kmalloc(target.off_max, M_TEMP, M_WAITOK|M_ZERO);
55e44363 1061
7d20a8ff 1062 error = elf_corehdr(lp, sig, fp, cred, seginfo.count, &target);
984263bc
MD
1063
1064 /* Write the contents of all of the writable segments. */
1065 if (error == 0) {
1066 Elf_Phdr *php;
984263bc 1067 int i;
a63bf9b6 1068 ssize_t nbytes;
984263bc 1069
55e44363
MD
1070 php = (Elf_Phdr *)(target.buf + sizeof(Elf_Ehdr)) + 1;
1071 for (i = 0; i < seginfo.count; i++) {
be7d8f4f 1072 error = fp_write(fp, (caddr_t)php->p_vaddr,
e7440b28 1073 php->p_filesz, &nbytes, UIO_USERSPACE);
984263bc
MD
1074 if (error != 0)
1075 break;
984263bc
MD
1076 php++;
1077 }
1078 }
efda3bd0 1079 kfree(target.buf, M_TEMP);
984263bc
MD
1080
1081 return error;
1082}
1083
1084/*
731100e5 1085 * A callback for each_segment() to write out the segment's
984263bc
MD
1086 * program header entry.
1087 */
55e44363 1088static int
731100e5 1089cb_put_phdr(vm_map_entry_t entry, void *closure)
984263bc 1090{
55e44363 1091 struct phdr_closure *phc = closure;
984263bc
MD
1092 Elf_Phdr *phdr = phc->phdr;
1093
55e44363
MD
1094 if (phc->phdr == phc->phdr_max)
1095 return EINVAL;
1096
984263bc
MD
1097 phc->offset = round_page(phc->offset);
1098
1099 phdr->p_type = PT_LOAD;
1100 phdr->p_offset = phc->offset;
1101 phdr->p_vaddr = entry->start;
1102 phdr->p_paddr = 0;
1103 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1104 phdr->p_align = PAGE_SIZE;
1105 phdr->p_flags = 0;
1106 if (entry->protection & VM_PROT_READ)
1107 phdr->p_flags |= PF_R;
1108 if (entry->protection & VM_PROT_WRITE)
1109 phdr->p_flags |= PF_W;
1110 if (entry->protection & VM_PROT_EXECUTE)
1111 phdr->p_flags |= PF_X;
1112
1113 phc->offset += phdr->p_filesz;
55e44363
MD
1114 ++phc->phdr;
1115 return 0;
984263bc
MD
1116}
1117
1118/*
1119 * A callback for each_writable_segment() to gather information about
1120 * the number of segments and their total size.
1121 */
55e44363 1122static int
731100e5 1123cb_size_segment(vm_map_entry_t entry, void *closure)
984263bc 1124{
55e44363 1125 struct sseg_closure *ssc = closure;
984263bc 1126
55e44363
MD
1127 ++ssc->count;
1128 ssc->vsize += entry->end - entry->start;
1129 return 0;
984263bc
MD
1130}
1131
1132/*
731100e5
MD
1133 * A callback for each_segment() to gather information about
1134 * the number of text segments.
1135 */
55e44363 1136static int
731100e5
MD
1137cb_fpcount_segment(vm_map_entry_t entry, void *closure)
1138{
55e44363 1139 int *count = closure;
12693083
MD
1140 struct vnode *vp;
1141
1142 if (entry->object.vm_object->type == OBJT_VNODE) {
1143 vp = (struct vnode *)entry->object.vm_object->handle;
1144 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1145 return 0;
731100e5 1146 ++*count;
12693083 1147 }
55e44363 1148 return 0;
731100e5
MD
1149}
1150
55e44363 1151static int
731100e5
MD
1152cb_put_fp(vm_map_entry_t entry, void *closure)
1153{
55e44363 1154 struct fp_closure *fpc = closure;
731100e5
MD
1155 struct vn_hdr *vnh = fpc->vnh;
1156 Elf_Phdr *phdr = &vnh->vnh_phdr;
1157 struct vnode *vp;
1158 int error;
1159
12693083
MD
1160 /*
1161 * If an entry represents a vnode then write out a file handle.
1162 *
1163 * If we are checkpointing a checkpoint-restored program we do
1164 * NOT record the filehandle for the old checkpoint vnode (which
1165 * is mapped all over the place). Instead we rely on the fact
1166 * that a checkpoint-restored program does not mmap() the checkpt
1167 * vnode NOCORE, so its contents will be written out to the
4f12bfd3
MD
1168 * new checkpoint file. This is necessary because the 'old'
1169 * checkpoint file is typically destroyed when a new one is created
1170 * and thus cannot be used to restore the new checkpoint.
1171 *
1172 * Theoretically we could create a chain of checkpoint files and
1173 * operate the checkpointing operation kinda like an incremental
1174 * checkpoint, but a checkpoint restore would then likely wind up
1175 * referencing many prior checkpoint files and that is a bit over
1176 * the top for the purpose of the checkpoint API.
12693083 1177 */
731100e5 1178 if (entry->object.vm_object->type == OBJT_VNODE) {
12693083
MD
1179 vp = (struct vnode *)entry->object.vm_object->handle;
1180 if ((vp->v_flag & VCKPT) && curproc->p_textvp == vp)
1181 return 0;
55e44363
MD
1182 if (vnh == fpc->vnh_max)
1183 return EINVAL;
55e44363
MD
1184
1185 if (vp->v_mount)
1186 vnh->vnh_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
731100e5 1187 error = VFS_VPTOFH(vp, &vnh->vnh_fh.fh_fid);
4fe2217f
MD
1188 if (error) {
1189 char *freepath, *fullpath;
1190
5b4cfb7e 1191 if (vn_fullpath(curproc, vp, &fullpath, &freepath, 0)) {
6ea70f76 1192 kprintf("Warning: coredump, error %d: cannot store file handle for vnode %p\n", error, vp);
4fe2217f 1193 } else {
6ea70f76 1194 kprintf("Warning: coredump, error %d: cannot store file handle for %s\n", error, fullpath);
efda3bd0 1195 kfree(freepath, M_TEMP);
4fe2217f
MD
1196 }
1197 error = 0;
1198 }
55e44363 1199
731100e5
MD
1200 phdr->p_type = PT_LOAD;
1201 phdr->p_offset = 0; /* not written to core */
1202 phdr->p_vaddr = entry->start;
1203 phdr->p_paddr = 0;
1204 phdr->p_filesz = phdr->p_memsz = entry->end - entry->start;
1205 phdr->p_align = PAGE_SIZE;
1206 phdr->p_flags = 0;
1207 if (entry->protection & VM_PROT_READ)
1208 phdr->p_flags |= PF_R;
1209 if (entry->protection & VM_PROT_WRITE)
1210 phdr->p_flags |= PF_W;
1211 if (entry->protection & VM_PROT_EXECUTE)
1212 phdr->p_flags |= PF_X;
1213 ++fpc->vnh;
1214 ++fpc->count;
1215 }
55e44363 1216 return 0;
731100e5
MD
1217}
1218
731100e5 1219/*
984263bc
MD
1220 * For each writable segment in the process's memory map, call the given
1221 * function with a pointer to the map entry and some arbitrary
1222 * caller-supplied data.
1223 */
55e44363 1224static int
731100e5 1225each_segment(struct proc *p, segment_callback func, void *closure, int writable)
984263bc 1226{
55e44363 1227 int error = 0;
984263bc
MD
1228 vm_map_t map = &p->p_vmspace->vm_map;
1229 vm_map_entry_t entry;
1230
55e44363 1231 for (entry = map->header.next; error == 0 && entry != &map->header;
984263bc
MD
1232 entry = entry->next) {
1233 vm_object_t obj;
1234
1235 /*
1236 * Don't dump inaccessible mappings, deal with legacy
1237 * coredump mode.
1238 *
1239 * Note that read-only segments related to the elf binary
1240 * are marked MAP_ENTRY_NOCOREDUMP now so we no longer
1241 * need to arbitrarily ignore such segments.
1242 */
1243 if (elf_legacy_coredump) {
731100e5 1244 if (writable && (entry->protection & VM_PROT_RW) != VM_PROT_RW)
984263bc
MD
1245 continue;
1246 } else {
731100e5 1247 if (writable && (entry->protection & VM_PROT_ALL) == 0)
984263bc
MD
1248 continue;
1249 }
1250
1251 /*
1252 * Dont include memory segment in the coredump if
1253 * MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
1b874851
MD
1254 * madvise(2).
1255 *
1256 * Currently we only dump normal VM object maps. We do
1257 * not dump submaps or virtual page tables.
984263bc 1258 */
1b874851
MD
1259 if (writable && (entry->eflags & MAP_ENTRY_NOCOREDUMP))
1260 continue;
1261 if (entry->maptype != VM_MAPTYPE_NORMAL)
984263bc 1262 continue;
984263bc
MD
1263 if ((obj = entry->object.vm_object) == NULL)
1264 continue;
1265
1266 /* Find the deepest backing object. */
1267 while (obj->backing_object != NULL)
1268 obj = obj->backing_object;
1269
1270 /* Ignore memory-mapped devices and such things. */
1271 if (obj->type != OBJT_DEFAULT &&
1272 obj->type != OBJT_SWAP &&
1273 obj->type != OBJT_VNODE)
1274 continue;
1275
55e44363 1276 error = (*func)(entry, closure);
984263bc 1277 }
55e44363
MD
1278 return error;
1279}
1280
1281static
1282void *
1283target_reserve(elf_buf_t target, size_t bytes, int *error)
1284{
1285 void *res = NULL;
1286
1287 if (target->buf) {
1288 if (target->off + bytes > target->off_max)
1289 *error = EINVAL;
1290 else
1291 res = target->buf + target->off;
1292 }
1293 target->off += bytes;
1294 return (res);
984263bc
MD
1295}
1296
1297/*
1298 * Write the core file header to the file, including padding up to
1299 * the page boundary.
1300 */
1301static int
d913b2fa
NT
1302elf_corehdr(struct lwp *lp, int sig, struct file *fp, struct ucred *cred,
1303 int numsegs, elf_buf_t target)
984263bc 1304{
55e44363 1305 int error;
a63bf9b6 1306 ssize_t nbytes;
7d20a8ff 1307
ff7a3478
MD
1308 /*
1309 * Fill in the header. The fp is passed so we can detect and flag
1310 * a checkpoint file pointer within the core file itself, because
1311 * it may not be restored from the same file handle.
1312 */
d913b2fa 1313 error = elf_puthdr(lp, target, sig, WRITE, numsegs, fp);
984263bc
MD
1314
1315 /* Write it to the core file. */
e7440b28
MD
1316 if (error == 0) {
1317 error = fp_write(fp, target->buf, target->off, &nbytes,
1318 UIO_SYSSPACE);
1319 }
55e44363 1320 return error;
984263bc
MD
1321}
1322
55e44363 1323static int
d913b2fa
NT
1324elf_puthdr(struct lwp *lp, elf_buf_t target, int sig, enum putmode mode,
1325 int numsegs, struct file *fp)
984263bc 1326{
7d20a8ff 1327 struct proc *p = lp->lwp_proc;
55e44363 1328 int error = 0;
984263bc
MD
1329 size_t phoff;
1330 size_t noteoff;
1331 size_t notesz;
55e44363
MD
1332 Elf_Ehdr *ehdr;
1333 Elf_Phdr *phdr;
984263bc 1334
55e44363
MD
1335 ehdr = target_reserve(target, sizeof(Elf_Ehdr), &error);
1336
1337 phoff = target->off;
1338 phdr = target_reserve(target, (numsegs + 1) * sizeof(Elf_Phdr), &error);
1339
1340 noteoff = target->off;
d913b2fa
NT
1341 if (error == 0)
1342 elf_putallnotes(lp, target, sig, mode);
55e44363 1343 notesz = target->off - noteoff;
984263bc 1344
55e44363
MD
1345 /*
1346 * put extra cruft for dumping process state here
1347 * - we really want it be before all the program
1348 * mappings
1349 * - we just need to update the offset accordingly
1350 * and GDB will be none the wiser.
1351 */
1352 if (error == 0)
1353 error = elf_puttextvp(p, target);
1354 if (error == 0)
7d20a8ff 1355 error = elf_putsigs(lp, target);
55e44363 1356 if (error == 0)
ff7a3478 1357 error = elf_putfiles(p, target, fp);
55e44363
MD
1358
1359 /*
1360 * Align up to a page boundary for the program segments. The
1361 * actual data will be written to the outptu file, not to elf_buf_t,
1362 * so we do not have to do any further bounds checking.
1363 */
1364 target->off = round_page(target->off);
1365 if (error == 0 && ehdr != NULL) {
984263bc
MD
1366 /*
1367 * Fill in the ELF header.
1368 */
984263bc
MD
1369 ehdr->e_ident[EI_MAG0] = ELFMAG0;
1370 ehdr->e_ident[EI_MAG1] = ELFMAG1;
1371 ehdr->e_ident[EI_MAG2] = ELFMAG2;
1372 ehdr->e_ident[EI_MAG3] = ELFMAG3;
1373 ehdr->e_ident[EI_CLASS] = ELF_CLASS;
1374 ehdr->e_ident[EI_DATA] = ELF_DATA;
1375 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
64fce751 1376 ehdr->e_ident[EI_OSABI] = ELFOSABI_NONE;
984263bc
MD
1377 ehdr->e_ident[EI_ABIVERSION] = 0;
1378 ehdr->e_ident[EI_PAD] = 0;
1379 ehdr->e_type = ET_CORE;
1380 ehdr->e_machine = ELF_ARCH;
1381 ehdr->e_version = EV_CURRENT;
1382 ehdr->e_entry = 0;
1383 ehdr->e_phoff = phoff;
1384 ehdr->e_flags = 0;
1385 ehdr->e_ehsize = sizeof(Elf_Ehdr);
1386 ehdr->e_phentsize = sizeof(Elf_Phdr);
1387 ehdr->e_phnum = numsegs + 1;
1388 ehdr->e_shentsize = sizeof(Elf_Shdr);
1389 ehdr->e_shnum = 0;
1390 ehdr->e_shstrndx = SHN_UNDEF;
55e44363
MD
1391 }
1392 if (error == 0 && phdr != NULL) {
984263bc
MD
1393 /*
1394 * Fill in the program header entries.
1395 */
55e44363 1396 struct phdr_closure phc;
984263bc
MD
1397
1398 /* The note segement. */
1399 phdr->p_type = PT_NOTE;
1400 phdr->p_offset = noteoff;
1401 phdr->p_vaddr = 0;
1402 phdr->p_paddr = 0;
1403 phdr->p_filesz = notesz;
1404 phdr->p_memsz = 0;
1405 phdr->p_flags = 0;
1406 phdr->p_align = 0;
55e44363 1407 ++phdr;
984263bc
MD
1408
1409 /* All the writable segments from the program. */
1410 phc.phdr = phdr;
55e44363
MD
1411 phc.phdr_max = phdr + numsegs;
1412 phc.offset = target->off;
731100e5 1413 each_segment(p, cb_put_phdr, &phc, 1);
984263bc 1414 }
55e44363 1415 return (error);
984263bc
MD
1416}
1417
d913b2fa
NT
1418/*
1419 * Append core dump notes to target ELF buffer or simply update target size
1420 * if dryrun selected.
1421 */
1422static int
1423elf_putallnotes(struct lwp *corelp, elf_buf_t target, int sig,
1424 enum putmode mode)
1425{
1426 struct proc *p = corelp->lwp_proc;
1427 int error;
1428 struct {
1429 prstatus_t status;
1430 prfpregset_t fpregs;
1431 prpsinfo_t psinfo;
1432 } *tmpdata;
1433 prstatus_t *status;
1434 prfpregset_t *fpregs;
1435 prpsinfo_t *psinfo;
1436 struct lwp *lp;
1437
1438 /*
1439 * Allocate temporary storage for notes on heap to avoid stack overflow.
1440 */
1441 if (mode != DRYRUN) {
1442 tmpdata = kmalloc(sizeof(*tmpdata), M_TEMP, M_ZERO | M_WAITOK);
1443 status = &tmpdata->status;
1444 fpregs = &tmpdata->fpregs;
1445 psinfo = &tmpdata->psinfo;
1446 } else {
1447 tmpdata = NULL;
1448 status = NULL;
1449 fpregs = NULL;
1450 psinfo = NULL;
1451 }
1452
1453 /*
1454 * Append LWP-agnostic note.
1455 */
1456 if (mode != DRYRUN) {
1457 psinfo->pr_version = PRPSINFO_VERSION;
1458 psinfo->pr_psinfosz = sizeof(prpsinfo_t);
1459 strncpy(psinfo->pr_fname, p->p_comm,
1460 sizeof(psinfo->pr_fname) - 1);
1461 /*
1462 * XXX - We don't fill in the command line arguments
1463 * properly yet.
1464 */
1465 strncpy(psinfo->pr_psargs, p->p_comm, PRARGSZ);
1466 }
1467 error =
64fce751 1468 elf_putnote(target, "CORE", NT_PRPSINFO, psinfo, sizeof *psinfo);
d913b2fa
NT
1469 if (error)
1470 goto exit;
1471
1472 /*
1473 * Append first note for LWP that triggered core so that it is
1474 * the selected one when the debugger starts.
1475 */
1476 if (mode != DRYRUN) {
1477 status->pr_version = PRSTATUS_VERSION;
1478 status->pr_statussz = sizeof(prstatus_t);
1479 status->pr_gregsetsz = sizeof(gregset_t);
1480 status->pr_fpregsetsz = sizeof(fpregset_t);
1481 status->pr_osreldate = osreldate;
1482 status->pr_cursig = sig;
1483 /*
1484 * XXX GDB needs unique pr_pid for each LWP and does not
1485 * not support pr_pid==0 but lwp_tid can be 0, so hack unique
1486 * value.
1487 */
64fce751 1488 status->pr_pid = corelp->lwp_tid;
d913b2fa
NT
1489 fill_regs(corelp, &status->pr_reg);
1490 fill_fpregs(corelp, fpregs);
1491 }
1492 error =
64fce751 1493 elf_putnote(target, "CORE", NT_PRSTATUS, status, sizeof *status);
d913b2fa
NT
1494 if (error)
1495 goto exit;
1496 error =
64fce751 1497 elf_putnote(target, "CORE", NT_FPREGSET, fpregs, sizeof *fpregs);
d913b2fa
NT
1498 if (error)
1499 goto exit;
1500
1501 /*
1502 * Then append notes for other LWPs.
1503 */
1504 FOREACH_LWP_IN_PROC(lp, p) {
1505 if (lp == corelp)
1506 continue;
1507 /* skip lwps being created */
1508 if (lp->lwp_thread == NULL)
1509 continue;
1510 if (mode != DRYRUN) {
64fce751 1511 status->pr_pid = lp->lwp_tid;
d913b2fa
NT
1512 fill_regs(lp, &status->pr_reg);
1513 fill_fpregs(lp, fpregs);
1514 }
64fce751 1515 error = elf_putnote(target, "CORE", NT_PRSTATUS,
d913b2fa
NT
1516 status, sizeof *status);
1517 if (error)
1518 goto exit;
64fce751 1519 error = elf_putnote(target, "CORE", NT_FPREGSET,
d913b2fa
NT
1520 fpregs, sizeof *fpregs);
1521 if (error)
1522 goto exit;
1523 }
1524
1525exit:
1526 if (tmpdata != NULL)
1527 kfree(tmpdata, M_TEMP);
1528 return (error);
1529}
1530
f606e7b4
MD
1531/*
1532 * Generate a note sub-structure.
1533 *
1534 * NOTE: 4-byte alignment.
1535 */
55e44363
MD
1536static int
1537elf_putnote(elf_buf_t target, const char *name, int type,
1538 const void *desc, size_t descsz)
984263bc 1539{
55e44363
MD
1540 int error = 0;
1541 char *dst;
984263bc
MD
1542 Elf_Note note;
1543
1544 note.n_namesz = strlen(name) + 1;
1545 note.n_descsz = descsz;
1546 note.n_type = type;
55e44363 1547 dst = target_reserve(target, sizeof(note), &error);
984263bc 1548 if (dst != NULL)
55e44363
MD
1549 bcopy(&note, dst, sizeof note);
1550 dst = target_reserve(target, note.n_namesz, &error);
984263bc 1551 if (dst != NULL)
55e44363 1552 bcopy(name, dst, note.n_namesz);
f606e7b4 1553 target->off = roundup2(target->off, sizeof(Elf_Word));
55e44363 1554 dst = target_reserve(target, note.n_descsz, &error);
984263bc 1555 if (dst != NULL)
55e44363 1556 bcopy(desc, dst, note.n_descsz);
f606e7b4 1557 target->off = roundup2(target->off, sizeof(Elf_Word));
55e44363 1558 return(error);
984263bc
MD
1559}
1560
731100e5 1561
55e44363 1562static int
7d20a8ff 1563elf_putsigs(struct lwp *lp, elf_buf_t target)
731100e5 1564{
7d20a8ff
SS
1565 /* XXX lwp handle more than one lwp */
1566 struct proc *p = lp->lwp_proc;
55e44363 1567 int error = 0;
731100e5 1568 struct ckpt_siginfo *csi;
55e44363
MD
1569
1570 csi = target_reserve(target, sizeof(struct ckpt_siginfo), &error);
1571 if (csi) {
1572 csi->csi_ckptpisz = sizeof(struct ckpt_siginfo);
b1b4e5a6 1573 bcopy(p->p_sigacts, &csi->csi_sigacts, sizeof(*p->p_sigacts));
55e44363 1574 bcopy(&p->p_realtimer, &csi->csi_itimerval, sizeof(struct itimerval));
7d20a8ff 1575 bcopy(&lp->lwp_sigmask, &csi->csi_sigmask,
08f2f1bb 1576 sizeof(sigset_t));
55e44363 1577 csi->csi_sigparent = p->p_sigparent;
731100e5 1578 }
55e44363 1579 return(error);
731100e5
MD
1580}
1581
55e44363 1582static int
ff7a3478 1583elf_putfiles(struct proc *p, elf_buf_t target, struct file *ckfp)
731100e5 1584{
55e44363
MD
1585 int error = 0;
1586 int i;
be7d8f4f 1587 struct ckpt_filehdr *cfh = NULL;
731100e5
MD
1588 struct ckpt_fileinfo *cfi;
1589 struct file *fp;
1590 struct vnode *vp;
1591 /*
1592 * the duplicated loop is gross, but it was the only way
1593 * to eliminate uninitialized variable warnings
1594 */
55e44363
MD
1595 cfh = target_reserve(target, sizeof(struct ckpt_filehdr), &error);
1596 if (cfh) {
731100e5 1597 cfh->cfh_nfiles = 0;
be7d8f4f 1598 }
be7d8f4f
MD
1599
1600 /*
4f12bfd3 1601 * ignore STDIN/STDERR/STDOUT.
be7d8f4f 1602 */
55e44363 1603 for (i = 3; error == 0 && i < p->p_fd->fd_nfiles; i++) {
228b401d
MD
1604 fp = holdfp(p->p_fd, i, -1);
1605 if (fp == NULL)
be7d8f4f 1606 continue;
4f12bfd3
MD
1607 /*
1608 * XXX Only checkpoint vnodes for now.
1609 */
228b401d
MD
1610 if (fp->f_type != DTYPE_VNODE) {
1611 fdrop(fp);
be7d8f4f 1612 continue;
228b401d 1613 }
4f12bfd3
MD
1614 cfi = target_reserve(target, sizeof(struct ckpt_fileinfo),
1615 &error);
228b401d
MD
1616 if (cfi == NULL) {
1617 fdrop(fp);
4f12bfd3 1618 continue;
228b401d 1619 }
4f12bfd3
MD
1620 cfi->cfi_index = -1;
1621 cfi->cfi_type = fp->f_type;
1622 cfi->cfi_flags = fp->f_flag;
1623 cfi->cfi_offset = fp->f_offset;
ff7a3478
MD
1624 cfi->cfi_ckflags = 0;
1625
1626 if (fp == ckfp)
1627 cfi->cfi_ckflags |= CKFIF_ISCKPTFD;
4f12bfd3
MD
1628 /* f_count and f_msgcount should not be saved/restored */
1629 /* XXX save cred info */
1630
1631 switch(fp->f_type) {
1632 case DTYPE_VNODE:
9ce25870 1633 vp = (struct vnode *)fp->f_data;
55e44363
MD
1634 /*
1635 * it looks like a bug in ptrace is marking
9ce25870
MD
1636 * a non-vnode as a vnode - until we find the
1637 * root cause this will at least prevent
1638 * further panics from truss
1639 */
55e44363 1640 if (vp == NULL || vp->v_mount == NULL)
4f12bfd3 1641 break;
be7d8f4f 1642 cfh->cfh_nfiles++;
be7d8f4f 1643 cfi->cfi_index = i;
be7d8f4f
MD
1644 cfi->cfi_fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1645 error = VFS_VPTOFH(vp, &cfi->cfi_fh.fh_fid);
4f12bfd3
MD
1646 break;
1647 default:
1648 break;
731100e5 1649 }
228b401d 1650 fdrop(fp);
731100e5 1651 }
55e44363 1652 return(error);
731100e5
MD
1653}
1654
55e44363
MD
1655static int
1656elf_puttextvp(struct proc *p, elf_buf_t target)
731100e5 1657{
55e44363 1658 int error = 0;
731100e5 1659 int *vn_count;
5a5c6af6
MD
1660 struct fp_closure fpc;
1661 struct ckpt_vminfo *vminfo;
1662
55e44363
MD
1663 vminfo = target_reserve(target, sizeof(struct ckpt_vminfo), &error);
1664 if (vminfo != NULL) {
5a5c6af6
MD
1665 vminfo->cvm_dsize = p->p_vmspace->vm_dsize;
1666 vminfo->cvm_tsize = p->p_vmspace->vm_tsize;
1667 vminfo->cvm_daddr = p->p_vmspace->vm_daddr;
1668 vminfo->cvm_taddr = p->p_vmspace->vm_taddr;
731100e5 1669 }
5a5c6af6 1670
731100e5 1671 fpc.count = 0;
55e44363
MD
1672 vn_count = target_reserve(target, sizeof(int), &error);
1673 if (target->buf != NULL) {
1674 fpc.vnh = (struct vn_hdr *)(target->buf + target->off);
1675 fpc.vnh_max = fpc.vnh +
1676 (target->off_max - target->off) / sizeof(struct vn_hdr);
1677 error = each_segment(p, cb_put_fp, &fpc, 0);
1678 if (vn_count)
1679 *vn_count = fpc.count;
5a5c6af6 1680 } else {
55e44363 1681 error = each_segment(p, cb_fpcount_segment, &fpc.count, 0);
5a5c6af6 1682 }
55e44363
MD
1683 target->off += fpc.count * sizeof(struct vn_hdr);
1684 return(error);
731100e5
MD
1685}
1686
1687
984263bc
MD
1688/*
1689 * Tell kern_execve.c about it, with a little help from the linker.
1690 */
1691static struct execsw elf_execsw = {exec_elf_imgact, "ELF"};
1843e3f2 1692EXEC_SET_ORDERED(elf, elf_execsw, SI_ORDER_FIRST);