linux_exec_aout.c revision 1.6 1 /* $NetBSD: linux_exec_aout.c,v 1.6 1995/06/11 14:56:47 fvdl Exp $ */
2
3 /*
4 * Copyright (c) 1995 Frank van der Linden
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 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. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed for the NetBSD Project
18 * by Frank van der Linden
19 * 4. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 * based on kern/exec_aout.c and compat/sunos/sunos_exec.c
34 */
35
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/filedesc.h>
39 #include <sys/kernel.h>
40 #include <sys/proc.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/namei.h>
44 #include <sys/vnode.h>
45 #include <sys/file.h>
46 #include <sys/resourcevar.h>
47 #include <sys/wait.h>
48
49 #include <sys/mman.h>
50 #include <vm/vm.h>
51 #include <vm/vm_param.h>
52 #include <vm/vm_map.h>
53 #include <vm/vm_kern.h>
54 #include <vm/vm_pager.h>
55
56 #include <machine/cpu.h>
57 #include <machine/reg.h>
58 #include <machine/exec.h>
59 #include <machine/linux_machdep.h>
60
61 #include <compat/linux/linux_types.h>
62 #include <compat/linux/linux_syscall.h>
63 #include <compat/linux/linux_syscallargs.h>
64 #include <compat/linux/linux_util.h>
65 #include <compat/linux/linux_exec.h>
66
67 struct elf_args {
68 u_long arg_entry; /* progran entry point */
69 u_long arg_interp; /* Interpreter load address */
70 u_long arg_phaddr; /* program header address */
71 u_long arg_phentsize; /* Size of program header */
72 u_long arg_phnum; /* Number of program headers */
73 };
74
75 static void *linux_aout_copyargs __P((struct exec_package *,
76 struct ps_strings *, void *, void *));
77 static void *linux_elf_copyargs __P((struct exec_package *, struct ps_strings *,
78 void *, void *));
79 static int linux_elf_check_header __P((Elf32_Ehdr *, int));
80 static void linux_elf_load_psection __P((struct exec_vmcmd_set *,
81 struct vnode *, Elf32_Phdr *, u_long *, u_long *, int *));
82 static int linux_elf_set_segment __P((struct exec_package *, u_long, u_long,
83 int));
84 static int linux_elf_read_from __P((struct vnode *, u_long, struct proc *,
85 caddr_t, int));
86 static int linux_elf_load_file __P((struct proc *, char *,
87 struct exec_vmcmd_set *, u_long *, struct elf_args *, u_long *));
88
89 #ifdef DEBUG_EXEC_LINUX_ELF
90 #define DPRINTF(x) printf x
91 #else
92 #define DPRINTF(x)
93 #endif
94
95 #define LINUX_ELF_ALIGN(a, b) ((a) & ~((b) - 1))
96 #define LINUX_ELF_AUX_ARGSIZ (sizeof(AuxInfo) * 8 / sizeof(char *))
97 #define LINUX_AOUT_AUX_ARGSIZ 2
98
99 extern int linux_error[];
100 extern struct sysent linux_sysent[];
101 extern char *linux_syscallnames[];
102
103 struct emul emul_linux_aout = {
104 "linux",
105 linux_error,
106 linux_sendsig,
107 LINUX_SYS_syscall,
108 LINUX_SYS_MAXSYSCALL,
109 linux_sysent,
110 linux_syscallnames,
111 LINUX_AOUT_AUX_ARGSIZ,
112 linux_aout_copyargs,
113 setregs,
114 linux_sigcode,
115 linux_esigcode,
116 };
117
118 struct emul emul_linux_elf = {
119 "linux",
120 linux_error,
121 linux_sendsig,
122 LINUX_SYS_syscall,
123 LINUX_SYS_MAXSYSCALL,
124 linux_sysent,
125 linux_syscallnames,
126 LINUX_ELF_AUX_ARGSIZ,
127 linux_elf_copyargs,
128 setregs,
129 linux_sigcode,
130 linux_esigcode,
131 };
132
133
134 static void *
135 linux_aout_copyargs(pack, arginfo, stack, argp)
136 struct exec_package *pack;
137 struct ps_strings *arginfo;
138 void *stack;
139 void *argp;
140 {
141 char **cpp = stack;
142 char **stk = stack;
143 char *dp, *sp;
144 size_t len;
145 void *nullp = NULL;
146 int argc = arginfo->ps_nargvstr;
147 int envc = arginfo->ps_nenvstr;
148
149 if (copyout(&argc, cpp++, sizeof(argc)))
150 return NULL;
151
152 /* leave room for envp and argv */
153 cpp += 2;
154 if (copyout(&cpp, &stk[1], sizeof (cpp)))
155 return NULL;
156
157 dp = (char *) (cpp + argc + envc + 2);
158 sp = argp;
159
160 /* XXX don't copy them out, remap them! */
161 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
162
163 for (; --argc >= 0; sp += len, dp += len)
164 if (copyout(&dp, cpp++, sizeof(dp)) ||
165 copyoutstr(sp, dp, ARG_MAX, &len))
166 return NULL;
167
168 if (copyout(&nullp, cpp++, sizeof(nullp)))
169 return NULL;
170
171 if (copyout(&cpp, &stk[2], sizeof (cpp)))
172 return NULL;
173
174 arginfo->ps_envstr = cpp; /* remember location of envp for later */
175
176 for (; --envc >= 0; sp += len, dp += len)
177 if (copyout(&dp, cpp++, sizeof(dp)) ||
178 copyoutstr(sp, dp, ARG_MAX, &len))
179 return NULL;
180
181 if (copyout(&nullp, cpp++, sizeof(nullp)))
182 return NULL;
183
184 return cpp;
185 }
186
187 static void *
188 linux_elf_copyargs(pack, arginfo, stack, argp)
189 struct exec_package *pack;
190 struct ps_strings *arginfo;
191 void *stack;
192 void *argp;
193 {
194 char **cpp = stack;
195 char *dp, *sp;
196 size_t len;
197 void *nullp = NULL;
198 int argc = arginfo->ps_nargvstr;
199 int envc = arginfo->ps_nenvstr;
200 AuxInfo *a;
201 struct elf_args *ap;
202
203 if (copyout(&argc, cpp++, sizeof(argc)))
204 return NULL;
205
206 dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
207 sp = argp;
208
209 /* XXX don't copy them out, remap them! */
210 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
211
212 for (; --argc >= 0; sp += len, dp += len)
213 if (copyout(&dp, cpp++, sizeof(dp)) ||
214 copyoutstr(sp, dp, ARG_MAX, &len))
215 return NULL;
216
217 if (copyout(&nullp, cpp++, sizeof(nullp)))
218 return NULL;
219
220 arginfo->ps_envstr = cpp; /* remember location of envp for later */
221
222 for (; --envc >= 0; sp += len, dp += len)
223 if (copyout(&dp, cpp++, sizeof(dp)) ||
224 copyoutstr(sp, dp, ARG_MAX, &len))
225 return NULL;
226
227 if (copyout(&nullp, cpp++, sizeof(nullp)))
228 return NULL;
229
230 /*
231 * Push extra arguments on the stack needed by dynamically
232 * linked binaries
233 */
234 a = (AuxInfo *) cpp;
235 if ((ap = (struct elf_args *) pack->ep_emul_arg)) {
236
237 DPRINTF(("phaddr=0x%x, phsize=%d, phnum=%d, interp=0x%x, ",
238 ap->arg_phaddr, ap->arg_phentsize, ap->arg_phnum,
239 ap->arg_interp));
240 DPRINTF((" entry=0x%x\n", ap->arg_entry));
241
242 a->au_id = AUX_phdr;
243 a->au_v = ap->arg_phaddr;
244 a++;
245
246 a->au_id = AUX_phent;
247 a->au_v = ap->arg_phentsize;
248 a++;
249
250 a->au_id = AUX_phnum;
251 a->au_v = ap->arg_phnum;
252 a++;
253
254 a->au_id = AUX_pagesz;
255 a->au_v = NBPG;
256 a++;
257
258 a->au_id = AUX_base;
259 a->au_v = ap->arg_interp;
260 a++;
261
262 a->au_id = AUX_flags;
263 a->au_v = 0;
264 a++;
265
266 a->au_id = AUX_entry;
267 a->au_v = ap->arg_entry;
268 a++;
269
270 a->au_id = AUX_null;
271 a->au_v = 0;
272 a++;
273
274 free((char *) ap, M_TEMP);
275 }
276 return a;
277 }
278
279 #ifdef DEBUG_EXEC_LINUX_ELF
280 static void
281 print_Ehdr(e)
282 Elf32_Ehdr *e;
283 {
284 printf("e_ident %s, ", e->e_ident);
285 printf("e_type %d, ", e->e_type);
286 printf("e_machine %d, ", e->e_machine);
287 printf("e_version %ld, ", e->e_version);
288 printf("e_entry %lx, ", e->e_entry);
289 printf("e_phoff %lx, ", e->e_phoff);
290 printf("e_shoff %lx, ", e->e_shoff);
291 printf("e_flags %lx, ", e->e_flags);
292 printf("e_ehsize %d, ", e->e_ehsize);
293 printf("e_phentsize %d, ", e->e_phentsize);
294 printf("e_phnum %d, ", e->e_phnum);
295 printf("e_shentsize %d, ", e->e_shentsize);
296 printf("e_shnum %d, ", e->e_shnum);
297 printf("e_shstrndx %d\n", e->e_shstrndx);
298 }
299
300
301 static void
302 print_Phdr(p)
303 Elf32_Phdr *p;
304 {
305 static char *types[] =
306 {
307 "null", "load", "dynamic", "interp",
308 "note", "shlib", "phdr", "entry7"
309 };
310
311 printf("p_type %ld [%s], ", p->p_type, types[p->p_type & 7]);
312 printf("p_offset %lx, ", p->p_offset);
313 printf("p_vaddr %lx, ", p->p_vaddr);
314 printf("p_paddr %lx, ", p->p_paddr);
315 printf("p_filesz %ld, ", p->p_filesz);
316 printf("p_memsz %ld, ", p->p_memsz);
317 printf("p_flags %lx, ", p->p_flags);
318 printf("p_align %ld\n", p->p_align);
319 }
320 #endif
321
322 int
323 exec_linux_aout_makecmds(p, epp)
324 struct proc *p;
325 struct exec_package *epp;
326 {
327 struct exec *linux_ep = epp->ep_hdr;
328 int machtype, magic;
329 int error = ENOEXEC;
330
331 magic = LINUX_N_MAGIC(linux_ep);
332 machtype = LINUX_N_MACHTYPE(linux_ep);
333
334
335 if (machtype != LINUX_MID_MACHINE)
336 return (ENOEXEC);
337
338 switch (magic) {
339 case QMAGIC:
340 error = exec_linux_aout_prep_qmagic(p, epp);
341 break;
342 case ZMAGIC:
343 error = exec_linux_aout_prep_zmagic(p, epp);
344 break;
345 case NMAGIC:
346 error = exec_linux_aout_prep_nmagic(p, epp);
347 break;
348 case OMAGIC:
349 error = exec_linux_aout_prep_omagic(p, epp);
350 break;
351 }
352 if (error == 0)
353 epp->ep_emul = &emul_linux_aout;
354 return error;
355 }
356
357 /*
358 * Since text starts at 0x400 in Linux ZMAGIC executables, and 0x400
359 * is very likely not page aligned on most architectures, it is treated
360 * as an NMAGIC here. XXX
361 */
362
363 int
364 exec_linux_aout_prep_zmagic(p, epp)
365 struct proc *p;
366 struct exec_package *epp;
367 {
368 struct exec *execp = epp->ep_hdr;
369
370 epp->ep_taddr = LINUX_N_TXTADDR(*execp, ZMAGIC);
371 epp->ep_tsize = execp->a_text;
372 epp->ep_daddr = LINUX_N_DATADDR(*execp, ZMAGIC);
373 epp->ep_dsize = execp->a_data + execp->a_bss;
374 epp->ep_entry = execp->a_entry;
375
376 /* set up command for text segment */
377 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_text,
378 epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, ZMAGIC),
379 VM_PROT_READ|VM_PROT_EXECUTE);
380
381 /* set up command for data segment */
382 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_data,
383 epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, ZMAGIC),
384 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
385
386 /* set up command for bss segment */
387 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, execp->a_bss,
388 epp->ep_daddr + execp->a_data, NULLVP, 0,
389 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
390
391 return exec_aout_setup_stack(p, epp);
392 }
393
394 /*
395 * exec_aout_prep_nmagic(): Prepare Linux NMAGIC package.
396 * Not different from the normal stuff.
397 */
398
399 int
400 exec_linux_aout_prep_nmagic(p, epp)
401 struct proc *p;
402 struct exec_package *epp;
403 {
404 struct exec *execp = epp->ep_hdr;
405 long bsize, baddr;
406
407 epp->ep_taddr = LINUX_N_TXTADDR(*execp, NMAGIC);
408 epp->ep_tsize = execp->a_text;
409 epp->ep_daddr = LINUX_N_DATADDR(*execp, NMAGIC);
410 epp->ep_dsize = execp->a_data + execp->a_bss;
411 epp->ep_entry = execp->a_entry;
412
413 /* set up command for text segment */
414 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_text,
415 epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, NMAGIC),
416 VM_PROT_READ|VM_PROT_EXECUTE);
417
418 /* set up command for data segment */
419 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_data,
420 epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, NMAGIC),
421 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
422
423 /* set up command for bss segment */
424 baddr = roundup(epp->ep_daddr + execp->a_data, NBPG);
425 bsize = epp->ep_daddr + epp->ep_dsize - baddr;
426 if (bsize > 0)
427 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, bsize, baddr,
428 NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
429
430 return exec_aout_setup_stack(p, epp);
431 }
432
433 /*
434 * exec_aout_prep_omagic(): Prepare Linux OMAGIC package.
435 * Business as usual.
436 */
437
438 int
439 exec_linux_aout_prep_omagic(p, epp)
440 struct proc *p;
441 struct exec_package *epp;
442 {
443 struct exec *execp = epp->ep_hdr;
444 long dsize, bsize, baddr;
445
446 epp->ep_taddr = LINUX_N_TXTADDR(*execp, OMAGIC);
447 epp->ep_tsize = execp->a_text;
448 epp->ep_daddr = LINUX_N_DATADDR(*execp, OMAGIC);
449 epp->ep_dsize = execp->a_data + execp->a_bss;
450 epp->ep_entry = execp->a_entry;
451
452 /* set up command for text and data segments */
453 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn,
454 execp->a_text + execp->a_data, epp->ep_taddr, epp->ep_vp,
455 LINUX_N_TXTOFF(*execp, OMAGIC), VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
456
457 /* set up command for bss segment */
458 baddr = roundup(epp->ep_daddr + execp->a_data, NBPG);
459 bsize = epp->ep_daddr + epp->ep_dsize - baddr;
460 if (bsize > 0)
461 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, bsize, baddr,
462 NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
463
464 /*
465 * Make sure (# of pages) mapped above equals (vm_tsize + vm_dsize);
466 * obreak(2) relies on this fact. Both `vm_tsize' and `vm_dsize' are
467 * computed (in execve(2)) by rounding *up* `ep_tsize' and `ep_dsize'
468 * respectively to page boundaries.
469 * Compensate `ep_dsize' for the amount of data covered by the last
470 * text page.
471 */
472 dsize = epp->ep_dsize + execp->a_text - roundup(execp->a_text, NBPG);
473 epp->ep_dsize = (dsize > 0) ? dsize : 0;
474 return exec_aout_setup_stack(p, epp);
475 }
476
477 int
478 exec_linux_aout_prep_qmagic(p, epp)
479 struct proc *p;
480 struct exec_package *epp;
481 {
482 struct exec *execp = epp->ep_hdr;
483
484 epp->ep_taddr = LINUX_N_TXTADDR(*execp, QMAGIC);
485 epp->ep_tsize = execp->a_text;
486 epp->ep_daddr = LINUX_N_DATADDR(*execp, QMAGIC);
487 epp->ep_dsize = execp->a_data + execp->a_bss;
488 epp->ep_entry = execp->a_entry;
489
490 /*
491 * check if vnode is in open for writing, because we want to
492 * demand-page out of it. if it is, don't do it, for various
493 * reasons
494 */
495 if ((execp->a_text != 0 || execp->a_data != 0) &&
496 epp->ep_vp->v_writecount != 0) {
497 #ifdef DIAGNOSTIC
498 if (epp->ep_vp->v_flag & VTEXT)
499 panic("exec: a VTEXT vnode has writecount != 0\n");
500 #endif
501 return ETXTBSY;
502 }
503 epp->ep_vp->v_flag |= VTEXT;
504
505 /* set up command for text segment */
506 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_pagedvn, execp->a_text,
507 epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, QMAGIC),
508 VM_PROT_READ|VM_PROT_EXECUTE);
509
510 /* set up command for data segment */
511 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_pagedvn, execp->a_data,
512 epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, QMAGIC),
513 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
514
515 /* set up command for bss segment */
516 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, execp->a_bss,
517 epp->ep_daddr + execp->a_data, NULLVP, 0,
518 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
519
520 return exec_aout_setup_stack(p, epp);
521 }
522
523 /*
524 * linux_elf_check_header():
525 *
526 * Check header for validity; return 0 of ok ENOEXEC if error
527 */
528 static int
529 linux_elf_check_header(eh, type)
530 Elf32_Ehdr *eh;
531 int type;
532 {
533 #ifdef sparc
534 /* #$%@#$%@#$%! */
535 # define memcmp bcmp
536 #endif
537 if (memcmp(eh->e_ident, Elf32_e_ident, Elf32_e_siz) != 0) {
538 DPRINTF(("Not an elf file\n"));
539 return ENOEXEC;
540 }
541
542 switch (eh->e_machine) {
543 #ifdef i386
544 case Elf32_em_386:
545 case Elf32_em_486:
546 #endif
547 #ifdef sparc
548 case Elf32_em_sparc:
549 #endif
550 break;
551
552 default:
553 DPRINTF(("Unsupported elf machine type %d\n", eh->e_machine));
554 return ENOEXEC;
555 }
556
557 if (eh->e_type != type) {
558 DPRINTF(("Not an elf executable\n"));
559 return ENOEXEC;
560 }
561
562 return 0;
563 }
564
565
566 /*
567 * linux_elf_load_psection():
568 *
569 * Load a psection at the appropriate address
570 */
571 static void
572 linux_elf_load_psection(vcset, vp, ph, addr, size, prot)
573 struct exec_vmcmd_set *vcset;
574 struct vnode *vp;
575 Elf32_Phdr *ph;
576 u_long *addr;
577 u_long *size;
578 int *prot;
579 {
580 u_long uaddr;
581 long diff;
582 long offset;
583 u_long msize;
584
585 /*
586 * If the user specified an address, then we load there.
587 */
588 if (*addr != ~0) {
589 uaddr = *addr + ph->p_align;
590 *addr = LINUX_ELF_ALIGN(uaddr, ph->p_align);
591 uaddr = LINUX_ELF_ALIGN(ph->p_vaddr, ph->p_align);
592 diff = ph->p_vaddr - uaddr;
593 } else {
594 uaddr = ph->p_vaddr;
595 *addr = LINUX_ELF_ALIGN(uaddr, ph->p_align);
596 diff = uaddr - *addr;
597 }
598
599 *prot |= (ph->p_flags & Elf32_pf_r) ? VM_PROT_READ : 0;
600 *prot |= (ph->p_flags & Elf32_pf_w) ? VM_PROT_WRITE : 0;
601 *prot |= (ph->p_flags & Elf32_pf_x) ? VM_PROT_EXECUTE : 0;
602
603 offset = ph->p_offset - diff;
604 *size = ph->p_filesz + diff;
605 msize = ph->p_memsz + diff;
606
607 DPRINTF(("Elf Seg@ 0x%x/0x%x sz %d/%d off 0x%x/0x%x[%d] algn 0x%x\n",
608 ph->p_vaddr, *addr, *size, msize, ph->p_offset, offset,
609 diff, ph->p_align));
610
611 NEW_VMCMD(vcset, vmcmd_map_readvn, *size,
612 *addr, vp, offset, *prot);
613
614 /*
615 * Check if we need to extend the size of the segment
616 */
617 {
618 u_long rm = round_page(*addr + msize);
619 u_long rf = round_page(*addr + *size);
620 if (rm != rf) {
621 DPRINTF(("zeropad 0x%x-0x%x\n", rf, rm));
622 NEW_VMCMD(vcset, vmcmd_map_zero, rm - rf,
623 rf, NULLVP, 0, *prot);
624 *size = msize;
625 }
626 }
627 }
628
629
630 /*
631 * linux_elf_set_segment():
632 *
633 * Decide if the segment is text or data, depending on the protection
634 * and set it appropriately
635 */
636 static int
637 linux_elf_set_segment(epp, vaddr, size, prot)
638 struct exec_package *epp;
639 u_long vaddr;
640 u_long size;
641 int prot;
642 {
643 /*
644 * Kludge: Unfortunately the current implementation of
645 * exec package assumes a single text and data segment.
646 * In Elf we can have more, but here we limit ourselves
647 * to two and hope :-(
648 * We also assume that the text is r-x, and data is rwx.
649 */
650 switch (prot) {
651 case (VM_PROT_READ | VM_PROT_EXECUTE):
652 if (epp->ep_tsize != ~0) {
653 DPRINTF(("More than one text segment\n"));
654 return ENOEXEC;
655 }
656 epp->ep_taddr = vaddr;
657 epp->ep_tsize = size;
658 DPRINTF(("Elf Text@ 0x%x, size %d\n", vaddr, size));
659 break;
660
661 case (VM_PROT_READ | VM_PROT_WRITE):
662 case (VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE):
663 if (epp->ep_dsize != ~0) {
664 DPRINTF(("More than one data segment\n"));
665 return ENOEXEC;
666 }
667 epp->ep_daddr = vaddr;
668 epp->ep_dsize = size;
669
670 DPRINTF(("Elf Data@ 0x%x, size %d\n", vaddr, size));
671 break;
672
673 default:
674 DPRINTF(("Bad protection 0%o\n", prot));
675 return ENOEXEC;
676 }
677 return 0;
678 }
679
680
681 /*
682 * linux_elf_read_from():
683 *
684 * Read from vnode into buffer at offset.
685 */
686 static int
687 linux_elf_read_from(vp, off, p, buf, size)
688 struct vnode *vp;
689 u_long off;
690 struct proc *p;
691 caddr_t buf;
692 int size;
693 {
694 int error;
695 int resid;
696
697 DPRINTF(("read from 0x%x to 0x%x size %d\n",
698 off, buf, size));
699 if ((error = vn_rdwr(UIO_READ, vp, buf, size,
700 off, UIO_SYSSPACE, IO_NODELOCKED, p->p_ucred,
701 &resid, p)) != 0) {
702 DPRINTF(("Bad read error %d\n", error));
703 return error;
704 }
705 /*
706 * See if we got all of it
707 */
708 if (resid != 0) {
709 DPRINTF(("Incomplete read for header ask=%d, rem=%d\n",
710 size, resid));
711 return error;
712 }
713 return 0;
714 }
715
716
717 /*
718 * linux_elf_load_file():
719 *
720 * Load a file (interpreter/library) pointed to by path
721 * [stolen from coff_load_shlib()]. Made slightly more generic than
722 * the svr4 version, for possible later use in linux_uselib().
723 */
724 static int
725 linux_elf_load_file(p, path, vcset, entry, ap, last)
726 struct proc *p;
727 char *path;
728 struct exec_vmcmd_set *vcset;
729 u_long *entry;
730 struct elf_args *ap;
731 u_long *last;
732 {
733 int error, i;
734 struct nameidata nd;
735 Elf32_Ehdr eh;
736 Elf32_Phdr *ph = NULL;
737 u_long phsize;
738 char *bp = NULL;
739 u_long addr = *last;
740
741 DPRINTF(("Loading file %s @ %x\n", path, addr));
742
743 if ((error = linux_emul_find(p, NULL, linux_emul_path, path, &bp, 0)) != 0)
744 bp = NULL;
745 else
746 path = bp;
747 /*
748 * 1. open file
749 * 2. read filehdr
750 * 3. map text, data, and bss out of it using VM_*
751 */
752 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
753 /* first get the vnode */
754 if ((error = namei(&nd)) != 0) {
755 if (bp != NULL)
756 free((char *) bp, M_TEMP);
757 return error;
758 }
759 if ((error = linux_elf_read_from(nd.ni_vp, 0, p, (caddr_t) &eh,
760 sizeof(eh))) != 0)
761 goto bad;
762
763 #ifdef DEBUG_EXEC_LINUX_ELF
764 print_Ehdr(&eh);
765 #endif
766
767 if ((error = linux_elf_check_header(&eh, Elf32_et_dyn)) != 0)
768 goto bad;
769
770 phsize = eh.e_phnum * sizeof(Elf32_Phdr);
771 ph = (Elf32_Phdr *) malloc(phsize, M_TEMP, M_WAITOK);
772
773 if ((error = linux_elf_read_from(nd.ni_vp, eh.e_phoff, p,
774 (caddr_t) ph, phsize)) != 0)
775 goto bad;
776
777 /*
778 * Load all the necessary sections
779 */
780 for (i = 0; i < eh.e_phnum; i++) {
781 u_long size = 0;
782 int prot = 0;
783 #ifdef DEBUG_EXEC_LINUX_ELF
784 print_Phdr(&ph[i]);
785 #endif
786
787 switch (ph[i].p_type) {
788 case Elf32_pt_load:
789 linux_elf_load_psection(vcset, nd.ni_vp, &ph[i], &addr,
790 &size, &prot);
791 /* Assume that the text segment is r-x only */
792 if ((prot & PROT_WRITE) == 0) {
793 *entry = addr + eh.e_entry;
794 ap->arg_interp = addr;
795 DPRINTF(("Interpreter@ 0x%x\n", addr));
796 }
797 addr += size;
798 break;
799
800 case Elf32_pt_dynamic:
801 case Elf32_pt_phdr:
802 case Elf32_pt_note:
803 break;
804
805 default:
806 DPRINTF(("interp: Unexpected program header type %d\n",
807 ph[i].p_type));
808 break;
809 }
810 }
811
812 bad:
813 if (ph != NULL)
814 free((char *) ph, M_TEMP);
815 if (bp != NULL)
816 free((char *) bp, M_TEMP);
817
818 *last = addr;
819 vrele(nd.ni_vp);
820 return error;
821 }
822
823
824 /*
825 * exec_linux_elf_makecmds(): Prepare an Elf binary's exec package
826 *
827 * First, set of the various offsets/lengths in the exec package.
828 *
829 * Then, mark the text image busy (so it can be demand paged) or error
830 * out if this is not possible. Finally, set up vmcmds for the
831 * text, data, bss, and stack segments.
832 */
833 int
834 exec_linux_elf_makecmds(p, epp)
835 struct proc *p;
836 struct exec_package *epp;
837 {
838 Elf32_Ehdr *eh = epp->ep_hdr;
839 Elf32_Phdr *ph, *pp;
840 int error;
841 int i;
842 char interp[MAXPATHLEN];
843 u_long pos = 0;
844 u_long phsize;
845
846 #ifdef DEBUG_EXEC_LINUX_ELF
847 print_Ehdr(eh);
848 #endif
849 if (epp->ep_hdrvalid < sizeof(Elf32_Ehdr))
850 return ENOEXEC;
851
852 if (linux_elf_check_header(eh, Elf32_et_exec))
853 return ENOEXEC;
854
855 /*
856 * check if vnode is in open for writing, because we want to
857 * demand-page out of it. if it is, don't do it, for various
858 * reasons
859 */
860 if (epp->ep_vp->v_writecount != 0) {
861 #ifdef DIAGNOSTIC
862 if (epp->ep_vp->v_flag & VTEXT)
863 panic("exec: a VTEXT vnode has writecount != 0\n");
864 #endif
865 return ETXTBSY;
866 }
867 /*
868 * Allocate space to hold all the program headers, and read them
869 * from the file
870 */
871 phsize = eh->e_phnum * sizeof(Elf32_Phdr);
872 ph = (Elf32_Phdr *) malloc(phsize, M_TEMP, M_WAITOK);
873
874 if ((error = linux_elf_read_from(epp->ep_vp, eh->e_phoff, p,
875 (caddr_t) ph, phsize)) != 0)
876 goto bad;
877
878 epp->ep_tsize = ~0;
879 epp->ep_dsize = ~0;
880
881 interp[0] = '\0';
882
883 /*
884 * Load all the necessary sections
885 */
886 for (i = 0; i < eh->e_phnum; i++) {
887 u_long addr = ~0, size = 0;
888 int prot = 0;
889
890 pp = &ph[i];
891 #ifdef DEBUG_EXEC_LINUX_ELF
892 print_Phdr(pp);
893 #endif
894
895 switch (ph[i].p_type) {
896 case Elf32_pt_load:
897 linux_elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
898 &ph[i], &addr, &size, &prot);
899 if ((error = linux_elf_set_segment(epp, addr, size,
900 prot)) != 0)
901 goto bad;
902 break;
903
904 case Elf32_pt_shlib:
905 DPRINTF(("No support for COFF libraries (yet)\n"));
906 error = ENOEXEC;
907 goto bad;
908
909 case Elf32_pt_interp:
910 if (pp->p_filesz >= sizeof(interp)) {
911 DPRINTF(("Interpreter path too long %d\n",
912 pp->p_filesz));
913 goto bad;
914 }
915 if ((error = linux_elf_read_from(epp->ep_vp, pp->p_offset, p,
916 (caddr_t) interp, pp->p_filesz)) != 0)
917 goto bad;
918 break;
919
920 case Elf32_pt_dynamic:
921 case Elf32_pt_phdr:
922 case Elf32_pt_note:
923 break;
924
925 default:
926 /*
927 * Not fatal, we don't need to understand everything
928 * :-)
929 */
930 DPRINTF(("Unsupported program header type %d\n",
931 pp->p_type));
932 break;
933 }
934 }
935
936 /*
937 * Check if we found a dynamically linked binary and arrange to load
938 * it's interpreter
939 */
940 if (interp[0]) {
941 struct elf_args *ap;
942 pos = ~0;
943
944 ap = (struct elf_args *) malloc(sizeof(struct elf_args),
945 M_TEMP, M_WAITOK);
946 if ((error = linux_elf_load_file(p, interp, &epp->ep_vmcmds,
947 &epp->ep_entry, ap, &pos)) != 0) {
948 free((char *) ap, M_TEMP);
949 goto bad;
950 }
951 /* Arrange to load the program headers. */
952 pos = LINUX_ELF_ALIGN(pos + NBPG, NBPG);
953 DPRINTF(("Program header @0x%x\n", pos));
954 ap->arg_phaddr = pos;
955 NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, phsize,
956 pos, epp->ep_vp, eh->e_phoff,
957 VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE);
958 pos += phsize;
959
960 ap->arg_phentsize = eh->e_phentsize;
961 ap->arg_phnum = eh->e_phnum;
962 ap->arg_entry = eh->e_entry;
963
964 epp->ep_emul_arg = ap;
965 } else
966 epp->ep_entry = eh->e_entry;
967
968 DPRINTF(("taddr 0x%x tsize 0x%x daddr 0x%x dsize 0x%x\n",
969 epp->ep_taddr, epp->ep_tsize, epp->ep_daddr, epp->ep_dsize));
970
971 free((char *) ph, M_TEMP);
972
973 DPRINTF(("Elf entry@ 0x%x\n", epp->ep_entry));
974 epp->ep_vp->v_flag |= VTEXT;
975
976 epp->ep_emul = &emul_linux_elf;
977
978 return exec_aout_setup_stack(p, epp);
979
980 bad:
981 free((char *) ph, M_TEMP);
982 kill_vmcmds(&epp->ep_vmcmds);
983 return ENOEXEC;
984 }
985 /*
986 * The Linux system call to load shared libraries, a.out version. The
987 * a.out shared libs are just files that are mapped onto a fixed
988 * address in the process' address space. The address is given in
989 * a_entry. Read in the header, set up some VM commands and run them.
990 *
991 * Yes, both text and data are mapped at once, so we're left with
992 * writeable text for the shared libs. The Linux crt0 seemed to break
993 * sometimes when data was mapped seperately. It munmapped a uselib()
994 * of ld.so by hand, which failed with shared text and data for ld.so
995 * Yuck.
996 *
997 * Because of the problem with ZMAGIC executables (text starts
998 * at 0x400 in the file, but needs to be mapped at 0), ZMAGIC
999 * shared libs are not handled very efficiently :-(
1000 */
1001
1002 int
1003 linux_uselib(p, uap, retval)
1004 struct proc *p;
1005 struct linux_uselib_args /* {
1006 syscallarg(char *) path;
1007 } */ *uap;
1008 register_t *retval;
1009 {
1010 caddr_t sg;
1011 long bsize, dsize, tsize, taddr, baddr, daddr;
1012 struct nameidata ni;
1013 struct vnode *vp;
1014 struct exec hdr;
1015 struct exec_vmcmd_set vcset;
1016 int rem, i, magic, error;
1017
1018 sg = stackgap_init();
1019 CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
1020
1021 NDINIT(&ni, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
1022
1023 if ((error = namei(&ni)))
1024 return error;
1025
1026 vp = ni.ni_vp;
1027
1028 if ((error = vn_rdwr(UIO_READ, vp, (caddr_t) &hdr, LINUX_AOUT_HDR_SIZE,
1029 0, UIO_SYSSPACE, IO_NODELOCKED, p->p_ucred,
1030 &rem, p))) {
1031 vrele(vp);
1032 return error;
1033 }
1034
1035 if (rem != 0) {
1036 vrele(vp);
1037 return ENOEXEC;
1038 }
1039
1040 if (LINUX_N_MACHTYPE(&hdr) != LINUX_MID_MACHINE)
1041 return ENOEXEC;
1042
1043 magic = LINUX_N_MAGIC(&hdr);
1044 taddr = hdr.a_entry & (~(NBPG - 1));
1045 tsize = hdr.a_text;
1046 daddr = taddr + tsize;
1047 dsize = hdr.a_data + hdr.a_bss;
1048
1049 if ((hdr.a_text != 0 || hdr.a_data != 0) && vp->v_writecount != 0) {
1050 vrele(vp);
1051 return ETXTBSY;
1052 }
1053 vp->v_flag |= VTEXT;
1054
1055 vcset.evs_cnt = 0;
1056 vcset.evs_used = 0;
1057
1058 NEW_VMCMD(&vcset,
1059 magic == ZMAGIC ? vmcmd_map_readvn : vmcmd_map_pagedvn,
1060 hdr.a_text + hdr.a_data, taddr,
1061 vp, LINUX_N_TXTOFF(hdr, magic),
1062 VM_PROT_READ|VM_PROT_EXECUTE|VM_PROT_WRITE);
1063
1064 baddr = roundup(daddr + hdr.a_data, NBPG);
1065 bsize = daddr + dsize - baddr;
1066 if (bsize > 0) {
1067 NEW_VMCMD(&vcset, vmcmd_map_zero, bsize, baddr,
1068 NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
1069 }
1070
1071 for (i = 0; i < vcset.evs_used && !error; i++) {
1072 struct exec_vmcmd *vcp;
1073
1074 vcp = &vcset.evs_cmds[i];
1075 error = (*vcp->ev_proc)(p, vcp);
1076 }
1077
1078 kill_vmcmds(&vcset);
1079
1080 vrele(vp);
1081
1082 return error;
1083 }
1084
1085 /*
1086 * Execve(2). Just check the alternate emulation path, and pass it on
1087 * to the NetBSD execve().
1088 */
1089 int
1090 linux_execve(p, uap, retval)
1091 struct proc *p;
1092 struct linux_execve_args /* {
1093 syscallarg(char *) path;
1094 syscallarg(char **) argv;
1095 syscallarg(char **) envp;
1096 } */ *uap;
1097 register_t *retval;
1098 {
1099 caddr_t sg;
1100
1101 sg = stackgap_init();
1102 CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
1103
1104 return execve(p, uap, retval);
1105 }
1106