Home | History | Annotate | Line # | Download | only in common
linux_exec_aout.c revision 1.7
      1 /*	$NetBSD: linux_exec_aout.c,v 1.7 1995/06/11 15:15:09 fvdl Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1995 Frank van der Linden
      5  * Copyright (c) 1994 Christos Zoulas
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. The name of the author may not be used to endorse or promote products
     17  *    derived from this software without specific prior written permission
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     29  *
     30  * based on exec_aout.c, sunos_exec.c and svr4_exec.c
     31  */
     32 
     33 #include <sys/param.h>
     34 #include <sys/systm.h>
     35 #include <sys/filedesc.h>
     36 #include <sys/kernel.h>
     37 #include <sys/proc.h>
     38 #include <sys/mount.h>
     39 #include <sys/malloc.h>
     40 #include <sys/namei.h>
     41 #include <sys/vnode.h>
     42 #include <sys/file.h>
     43 #include <sys/resourcevar.h>
     44 #include <sys/wait.h>
     45 
     46 #include <sys/mman.h>
     47 #include <vm/vm.h>
     48 #include <vm/vm_param.h>
     49 #include <vm/vm_map.h>
     50 #include <vm/vm_kern.h>
     51 #include <vm/vm_pager.h>
     52 
     53 #include <machine/cpu.h>
     54 #include <machine/reg.h>
     55 #include <machine/exec.h>
     56 #include <machine/linux_machdep.h>
     57 
     58 #include <compat/linux/linux_types.h>
     59 #include <compat/linux/linux_syscall.h>
     60 #include <compat/linux/linux_syscallargs.h>
     61 #include <compat/linux/linux_util.h>
     62 #include <compat/linux/linux_exec.h>
     63 
     64 struct elf_args {
     65 	u_long  arg_entry;	/* progran entry point */
     66 	u_long  arg_interp;	/* Interpreter load address */
     67 	u_long  arg_phaddr;	/* program header address */
     68 	u_long  arg_phentsize;	/* Size of program header */
     69 	u_long  arg_phnum;	/* Number of program headers */
     70 };
     71 
     72 static void *linux_aout_copyargs __P((struct exec_package *,
     73 	struct ps_strings *, void *, void *));
     74 static void *linux_elf_copyargs __P((struct exec_package *, struct ps_strings *,
     75 	void *, void *));
     76 static int linux_elf_check_header __P((Elf32_Ehdr *, int));
     77 static void linux_elf_load_psection __P((struct exec_vmcmd_set *,
     78 	struct vnode *, Elf32_Phdr *, u_long *, u_long *, int *));
     79 static int linux_elf_set_segment __P((struct exec_package *, u_long, u_long,
     80 	int));
     81 static int linux_elf_read_from __P((struct vnode *, u_long, struct proc *,
     82 	caddr_t, int));
     83 static int linux_elf_load_file __P((struct proc *, char *,
     84 	struct exec_vmcmd_set *, u_long *, struct elf_args *, u_long *));
     85 
     86 #ifdef DEBUG_EXEC_LINUX_ELF
     87 #define DPRINTF(x) printf x
     88 #else
     89 #define DPRINTF(x)
     90 #endif
     91 
     92 #define LINUX_ELF_ALIGN(a, b) ((a) & ~((b) - 1))
     93 #define LINUX_ELF_AUX_ARGSIZ (sizeof(AuxInfo) * 8 / sizeof(char *))
     94 #define	LINUX_AOUT_AUX_ARGSIZ	2
     95 
     96 extern int linux_error[];
     97 extern struct sysent linux_sysent[];
     98 extern char *linux_syscallnames[];
     99 
    100 struct emul emul_linux_aout = {
    101 	"linux",
    102 	linux_error,
    103 	linux_sendsig,
    104 	LINUX_SYS_syscall,
    105 	LINUX_SYS_MAXSYSCALL,
    106 	linux_sysent,
    107 	linux_syscallnames,
    108 	LINUX_AOUT_AUX_ARGSIZ,
    109 	linux_aout_copyargs,
    110 	setregs,
    111 	linux_sigcode,
    112 	linux_esigcode,
    113 };
    114 
    115 struct emul emul_linux_elf = {
    116 	"linux",
    117 	linux_error,
    118 	linux_sendsig,
    119 	LINUX_SYS_syscall,
    120 	LINUX_SYS_MAXSYSCALL,
    121 	linux_sysent,
    122 	linux_syscallnames,
    123 	LINUX_ELF_AUX_ARGSIZ,
    124 	linux_elf_copyargs,
    125 	setregs,
    126 	linux_sigcode,
    127 	linux_esigcode,
    128 };
    129 
    130 
    131 static void *
    132 linux_aout_copyargs(pack, arginfo, stack, argp)
    133 	struct exec_package *pack;
    134 	struct ps_strings *arginfo;
    135 	void *stack;
    136 	void *argp;
    137 {
    138 	char **cpp = stack;
    139 	char **stk = stack;
    140 	char *dp, *sp;
    141 	size_t len;
    142 	void *nullp = NULL;
    143 	int argc = arginfo->ps_nargvstr;
    144 	int envc = arginfo->ps_nenvstr;
    145 
    146 	if (copyout(&argc, cpp++, sizeof(argc)))
    147 		return NULL;
    148 
    149 	/* leave room for envp and argv */
    150 	cpp += 2;
    151 	if (copyout(&cpp, &stk[1], sizeof (cpp)))
    152 		return NULL;
    153 
    154 	dp = (char *) (cpp + argc + envc + 2);
    155 	sp = argp;
    156 
    157 	/* XXX don't copy them out, remap them! */
    158 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
    159 
    160 	for (; --argc >= 0; sp += len, dp += len)
    161 		if (copyout(&dp, cpp++, sizeof(dp)) ||
    162 		    copyoutstr(sp, dp, ARG_MAX, &len))
    163 			return NULL;
    164 
    165 	if (copyout(&nullp, cpp++, sizeof(nullp)))
    166 		return NULL;
    167 
    168 	if (copyout(&cpp, &stk[2], sizeof (cpp)))
    169 		return NULL;
    170 
    171 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
    172 
    173 	for (; --envc >= 0; sp += len, dp += len)
    174 		if (copyout(&dp, cpp++, sizeof(dp)) ||
    175 		    copyoutstr(sp, dp, ARG_MAX, &len))
    176 			return NULL;
    177 
    178 	if (copyout(&nullp, cpp++, sizeof(nullp)))
    179 		return NULL;
    180 
    181 	return cpp;
    182 }
    183 
    184 static void *
    185 linux_elf_copyargs(pack, arginfo, stack, argp)
    186 	struct exec_package *pack;
    187 	struct ps_strings *arginfo;
    188 	void *stack;
    189 	void *argp;
    190 {
    191 	char **cpp = stack;
    192 	char *dp, *sp;
    193 	size_t len;
    194 	void *nullp = NULL;
    195 	int argc = arginfo->ps_nargvstr;
    196 	int envc = arginfo->ps_nenvstr;
    197 	AuxInfo *a;
    198 	struct elf_args *ap;
    199 
    200 	if (copyout(&argc, cpp++, sizeof(argc)))
    201 		return NULL;
    202 
    203 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
    204 	sp = argp;
    205 
    206 	/* XXX don't copy them out, remap them! */
    207 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
    208 
    209 	for (; --argc >= 0; sp += len, dp += len)
    210 		if (copyout(&dp, cpp++, sizeof(dp)) ||
    211 		    copyoutstr(sp, dp, ARG_MAX, &len))
    212 			return NULL;
    213 
    214 	if (copyout(&nullp, cpp++, sizeof(nullp)))
    215 		return NULL;
    216 
    217 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
    218 
    219 	for (; --envc >= 0; sp += len, dp += len)
    220 		if (copyout(&dp, cpp++, sizeof(dp)) ||
    221 		    copyoutstr(sp, dp, ARG_MAX, &len))
    222 			return NULL;
    223 
    224 	if (copyout(&nullp, cpp++, sizeof(nullp)))
    225 		return NULL;
    226 
    227 	/*
    228 	 * Push extra arguments on the stack needed by dynamically
    229 	 * linked binaries
    230 	 */
    231 	a = (AuxInfo *) cpp;
    232 	if ((ap = (struct elf_args *) pack->ep_emul_arg)) {
    233 
    234 		DPRINTF(("phaddr=0x%x, phsize=%d, phnum=%d, interp=0x%x, ",
    235 			 ap->arg_phaddr, ap->arg_phentsize, ap->arg_phnum,
    236 			 ap->arg_interp));
    237 		DPRINTF((" entry=0x%x\n", ap->arg_entry));
    238 
    239 		a->au_id = AUX_phdr;
    240 		a->au_v = ap->arg_phaddr;
    241 		a++;
    242 
    243 		a->au_id = AUX_phent;
    244 		a->au_v = ap->arg_phentsize;
    245 		a++;
    246 
    247 		a->au_id = AUX_phnum;
    248 		a->au_v = ap->arg_phnum;
    249 		a++;
    250 
    251 		a->au_id = AUX_pagesz;
    252 		a->au_v = NBPG;
    253 		a++;
    254 
    255 		a->au_id = AUX_base;
    256 		a->au_v = ap->arg_interp;
    257 		a++;
    258 
    259 		a->au_id = AUX_flags;
    260 		a->au_v = 0;
    261 		a++;
    262 
    263 		a->au_id = AUX_entry;
    264 		a->au_v = ap->arg_entry;
    265 		a++;
    266 
    267 		a->au_id = AUX_null;
    268 		a->au_v = 0;
    269 		a++;
    270 
    271 		free((char *) ap, M_TEMP);
    272 	}
    273 	return a;
    274 }
    275 
    276 #ifdef DEBUG_EXEC_LINUX_ELF
    277 static void
    278 print_Ehdr(e)
    279 	Elf32_Ehdr     *e;
    280 {
    281 	printf("e_ident %s, ", e->e_ident);
    282 	printf("e_type %d, ", e->e_type);
    283 	printf("e_machine %d, ", e->e_machine);
    284 	printf("e_version %ld, ", e->e_version);
    285 	printf("e_entry %lx, ", e->e_entry);
    286 	printf("e_phoff %lx, ", e->e_phoff);
    287 	printf("e_shoff %lx, ", e->e_shoff);
    288 	printf("e_flags %lx, ", e->e_flags);
    289 	printf("e_ehsize %d, ", e->e_ehsize);
    290 	printf("e_phentsize %d, ", e->e_phentsize);
    291 	printf("e_phnum %d, ", e->e_phnum);
    292 	printf("e_shentsize %d, ", e->e_shentsize);
    293 	printf("e_shnum %d, ", e->e_shnum);
    294 	printf("e_shstrndx %d\n", e->e_shstrndx);
    295 }
    296 
    297 
    298 static void
    299 print_Phdr(p)
    300 	Elf32_Phdr     *p;
    301 {
    302 	static char    *types[] =
    303 	{
    304 		"null", "load", "dynamic", "interp",
    305 		"note", "shlib", "phdr", "entry7"
    306 	};
    307 
    308 	printf("p_type %ld [%s], ", p->p_type, types[p->p_type & 7]);
    309 	printf("p_offset %lx, ", p->p_offset);
    310 	printf("p_vaddr %lx, ", p->p_vaddr);
    311 	printf("p_paddr %lx, ", p->p_paddr);
    312 	printf("p_filesz %ld, ", p->p_filesz);
    313 	printf("p_memsz %ld, ", p->p_memsz);
    314 	printf("p_flags %lx, ", p->p_flags);
    315 	printf("p_align %ld\n", p->p_align);
    316 }
    317 #endif
    318 
    319 int
    320 exec_linux_aout_makecmds(p, epp)
    321 	struct proc *p;
    322 	struct exec_package *epp;
    323 {
    324 	struct exec *linux_ep = epp->ep_hdr;
    325 	int machtype, magic;
    326 	int error = ENOEXEC;
    327 
    328 	magic = LINUX_N_MAGIC(linux_ep);
    329 	machtype = LINUX_N_MACHTYPE(linux_ep);
    330 
    331 
    332 	if (machtype != LINUX_MID_MACHINE)
    333 		return (ENOEXEC);
    334 
    335 	switch (magic) {
    336 	case QMAGIC:
    337 		error = exec_linux_aout_prep_qmagic(p, epp);
    338 		break;
    339 	case ZMAGIC:
    340 		error = exec_linux_aout_prep_zmagic(p, epp);
    341 		break;
    342 	case NMAGIC:
    343 		error = exec_linux_aout_prep_nmagic(p, epp);
    344 		break;
    345 	case OMAGIC:
    346 		error = exec_linux_aout_prep_omagic(p, epp);
    347 		break;
    348 	}
    349 	if (error == 0)
    350 		epp->ep_emul = &emul_linux_aout;
    351 	return error;
    352 }
    353 
    354 /*
    355  * Since text starts at 0x400 in Linux ZMAGIC executables, and 0x400
    356  * is very likely not page aligned on most architectures, it is treated
    357  * as an NMAGIC here. XXX
    358  */
    359 
    360 int
    361 exec_linux_aout_prep_zmagic(p, epp)
    362 	struct proc *p;
    363 	struct exec_package *epp;
    364 {
    365 	struct exec *execp = epp->ep_hdr;
    366 
    367 	epp->ep_taddr = LINUX_N_TXTADDR(*execp, ZMAGIC);
    368 	epp->ep_tsize = execp->a_text;
    369 	epp->ep_daddr = LINUX_N_DATADDR(*execp, ZMAGIC);
    370 	epp->ep_dsize = execp->a_data + execp->a_bss;
    371 	epp->ep_entry = execp->a_entry;
    372 
    373 	/* set up command for text segment */
    374 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_text,
    375 	    epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, ZMAGIC),
    376 	    VM_PROT_READ|VM_PROT_EXECUTE);
    377 
    378 	/* set up command for data segment */
    379 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_data,
    380 	    epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, ZMAGIC),
    381 	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    382 
    383 	/* set up command for bss segment */
    384 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, execp->a_bss,
    385 	    epp->ep_daddr + execp->a_data, NULLVP, 0,
    386 	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    387 
    388 	return exec_aout_setup_stack(p, epp);
    389 }
    390 
    391 /*
    392  * exec_aout_prep_nmagic(): Prepare Linux NMAGIC package.
    393  * Not different from the normal stuff.
    394  */
    395 
    396 int
    397 exec_linux_aout_prep_nmagic(p, epp)
    398 	struct proc *p;
    399 	struct exec_package *epp;
    400 {
    401 	struct exec *execp = epp->ep_hdr;
    402 	long bsize, baddr;
    403 
    404 	epp->ep_taddr = LINUX_N_TXTADDR(*execp, NMAGIC);
    405 	epp->ep_tsize = execp->a_text;
    406 	epp->ep_daddr = LINUX_N_DATADDR(*execp, NMAGIC);
    407 	epp->ep_dsize = execp->a_data + execp->a_bss;
    408 	epp->ep_entry = execp->a_entry;
    409 
    410 	/* set up command for text segment */
    411 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_text,
    412 	    epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, NMAGIC),
    413 	    VM_PROT_READ|VM_PROT_EXECUTE);
    414 
    415 	/* set up command for data segment */
    416 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, execp->a_data,
    417 	    epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, NMAGIC),
    418 	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    419 
    420 	/* set up command for bss segment */
    421 	baddr = roundup(epp->ep_daddr + execp->a_data, NBPG);
    422 	bsize = epp->ep_daddr + epp->ep_dsize - baddr;
    423 	if (bsize > 0)
    424 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, bsize, baddr,
    425 		    NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    426 
    427 	return exec_aout_setup_stack(p, epp);
    428 }
    429 
    430 /*
    431  * exec_aout_prep_omagic(): Prepare Linux OMAGIC package.
    432  * Business as usual.
    433  */
    434 
    435 int
    436 exec_linux_aout_prep_omagic(p, epp)
    437 	struct proc *p;
    438 	struct exec_package *epp;
    439 {
    440 	struct exec *execp = epp->ep_hdr;
    441 	long dsize, bsize, baddr;
    442 
    443 	epp->ep_taddr = LINUX_N_TXTADDR(*execp, OMAGIC);
    444 	epp->ep_tsize = execp->a_text;
    445 	epp->ep_daddr = LINUX_N_DATADDR(*execp, OMAGIC);
    446 	epp->ep_dsize = execp->a_data + execp->a_bss;
    447 	epp->ep_entry = execp->a_entry;
    448 
    449 	/* set up command for text and data segments */
    450 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn,
    451 	    execp->a_text + execp->a_data, epp->ep_taddr, epp->ep_vp,
    452 	    LINUX_N_TXTOFF(*execp, OMAGIC), VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    453 
    454 	/* set up command for bss segment */
    455 	baddr = roundup(epp->ep_daddr + execp->a_data, NBPG);
    456 	bsize = epp->ep_daddr + epp->ep_dsize - baddr;
    457 	if (bsize > 0)
    458 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, bsize, baddr,
    459 		    NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    460 
    461 	/*
    462 	 * Make sure (# of pages) mapped above equals (vm_tsize + vm_dsize);
    463 	 * obreak(2) relies on this fact. Both `vm_tsize' and `vm_dsize' are
    464 	 * computed (in execve(2)) by rounding *up* `ep_tsize' and `ep_dsize'
    465 	 * respectively to page boundaries.
    466 	 * Compensate `ep_dsize' for the amount of data covered by the last
    467 	 * text page.
    468 	 */
    469 	dsize = epp->ep_dsize + execp->a_text - roundup(execp->a_text, NBPG);
    470 	epp->ep_dsize = (dsize > 0) ? dsize : 0;
    471 	return exec_aout_setup_stack(p, epp);
    472 }
    473 
    474 int
    475 exec_linux_aout_prep_qmagic(p, epp)
    476 	struct proc *p;
    477 	struct exec_package *epp;
    478 {
    479 	struct exec *execp = epp->ep_hdr;
    480 
    481 	epp->ep_taddr = LINUX_N_TXTADDR(*execp, QMAGIC);
    482 	epp->ep_tsize = execp->a_text;
    483 	epp->ep_daddr = LINUX_N_DATADDR(*execp, QMAGIC);
    484 	epp->ep_dsize = execp->a_data + execp->a_bss;
    485 	epp->ep_entry = execp->a_entry;
    486 
    487 	/*
    488 	 * check if vnode is in open for writing, because we want to
    489 	 * demand-page out of it.  if it is, don't do it, for various
    490 	 * reasons
    491 	 */
    492 	if ((execp->a_text != 0 || execp->a_data != 0) &&
    493 	    epp->ep_vp->v_writecount != 0) {
    494 #ifdef DIAGNOSTIC
    495 		if (epp->ep_vp->v_flag & VTEXT)
    496 			panic("exec: a VTEXT vnode has writecount != 0\n");
    497 #endif
    498 		return ETXTBSY;
    499 	}
    500 	epp->ep_vp->v_flag |= VTEXT;
    501 
    502 	/* set up command for text segment */
    503 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_pagedvn, execp->a_text,
    504 	    epp->ep_taddr, epp->ep_vp, LINUX_N_TXTOFF(*execp, QMAGIC),
    505 	    VM_PROT_READ|VM_PROT_EXECUTE);
    506 
    507 	/* set up command for data segment */
    508 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_pagedvn, execp->a_data,
    509 	    epp->ep_daddr, epp->ep_vp, LINUX_N_DATOFF(*execp, QMAGIC),
    510 	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    511 
    512 	/* set up command for bss segment */
    513 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_zero, execp->a_bss,
    514 	    epp->ep_daddr + execp->a_data, NULLVP, 0,
    515 	    VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
    516 
    517 	return exec_aout_setup_stack(p, epp);
    518 }
    519 
    520 /*
    521  * linux_elf_check_header():
    522  *
    523  * Check header for validity; return 0 of ok ENOEXEC if error
    524  */
    525 static int
    526 linux_elf_check_header(eh, type)
    527 	Elf32_Ehdr     *eh;
    528 	int             type;
    529 {
    530 #ifdef sparc
    531   /* #$%@#$%@#$%! */
    532 # define memcmp bcmp
    533 #endif
    534 	if (memcmp(eh->e_ident, Elf32_e_ident, Elf32_e_siz) != 0) {
    535 		DPRINTF(("Not an elf file\n"));
    536 		return ENOEXEC;
    537 	}
    538 
    539 	switch (eh->e_machine) {
    540 #ifdef i386
    541 	case Elf32_em_386:
    542 	case Elf32_em_486:
    543 #endif
    544 #ifdef sparc
    545 	case Elf32_em_sparc:
    546 #endif
    547 		break;
    548 
    549 	default:
    550 		DPRINTF(("Unsupported elf machine type %d\n", eh->e_machine));
    551 		return ENOEXEC;
    552 	}
    553 
    554 	if (eh->e_type != type) {
    555 		DPRINTF(("Not an elf executable\n"));
    556 		return ENOEXEC;
    557 	}
    558 
    559 	return 0;
    560 }
    561 
    562 
    563 /*
    564  * linux_elf_load_psection():
    565  *
    566  * Load a psection at the appropriate address
    567  */
    568 static void
    569 linux_elf_load_psection(vcset, vp, ph, addr, size, prot)
    570 	struct exec_vmcmd_set   *vcset;
    571 	struct vnode		*vp;
    572 	Elf32_Phdr		*ph;
    573 	u_long			*addr;
    574 	u_long			*size;
    575 	int			*prot;
    576 {
    577 	u_long	uaddr;
    578 	long	diff;
    579 	long	offset;
    580 	u_long	msize;
    581 
    582 	/*
    583          * If the user specified an address, then we load there.
    584          */
    585 	if (*addr != ~0) {
    586 		uaddr = *addr + ph->p_align;
    587 		*addr = LINUX_ELF_ALIGN(uaddr, ph->p_align);
    588 		uaddr = LINUX_ELF_ALIGN(ph->p_vaddr, ph->p_align);
    589 		diff = ph->p_vaddr - uaddr;
    590 	} else {
    591 		uaddr = ph->p_vaddr;
    592 		*addr = LINUX_ELF_ALIGN(uaddr, ph->p_align);
    593 		diff = uaddr - *addr;
    594 	}
    595 
    596 	*prot |= (ph->p_flags & Elf32_pf_r) ? VM_PROT_READ : 0;
    597 	*prot |= (ph->p_flags & Elf32_pf_w) ? VM_PROT_WRITE : 0;
    598 	*prot |= (ph->p_flags & Elf32_pf_x) ? VM_PROT_EXECUTE : 0;
    599 
    600 	offset = ph->p_offset - diff;
    601 	*size = ph->p_filesz + diff;
    602 	msize = ph->p_memsz + diff;
    603 
    604 	DPRINTF(("Elf Seg@ 0x%x/0x%x sz %d/%d off 0x%x/0x%x[%d] algn 0x%x\n",
    605 		 ph->p_vaddr, *addr, *size, msize, ph->p_offset, offset,
    606 		 diff, ph->p_align));
    607 
    608 	NEW_VMCMD(vcset, vmcmd_map_readvn, *size,
    609 		  *addr, vp, offset, *prot);
    610 
    611 	/*
    612          * Check if we need to extend the size of the segment
    613          */
    614 	{
    615 		u_long	rm = round_page(*addr + msize);
    616 		u_long	rf = round_page(*addr + *size);
    617 		if (rm != rf) {
    618 			DPRINTF(("zeropad 0x%x-0x%x\n", rf, rm));
    619 			NEW_VMCMD(vcset, vmcmd_map_zero, rm - rf,
    620 				  rf, NULLVP, 0, *prot);
    621 			*size = msize;
    622 		}
    623 	}
    624 }
    625 
    626 
    627 /*
    628  * linux_elf_set_segment():
    629  *
    630  * Decide if the segment is text or data, depending on the protection
    631  * and set it appropriately
    632  */
    633 static int
    634 linux_elf_set_segment(epp, vaddr, size, prot)
    635 	struct exec_package	*epp;
    636 	u_long			 vaddr;
    637 	u_long			 size;
    638 	int			 prot;
    639 {
    640 	/*
    641          * Kludge: Unfortunately the current implementation of
    642          * exec package assumes a single text and data segment.
    643          * In Elf we can have more, but here we limit ourselves
    644          * to two and hope :-(
    645          * We also assume that the text is r-x, and data is rwx.
    646          */
    647 	switch (prot) {
    648 	case (VM_PROT_READ | VM_PROT_EXECUTE):
    649 		if (epp->ep_tsize != ~0) {
    650 			DPRINTF(("More than one text segment\n"));
    651 			return ENOEXEC;
    652 		}
    653 		epp->ep_taddr = vaddr;
    654 		epp->ep_tsize = size;
    655 		DPRINTF(("Elf Text@ 0x%x, size %d\n", vaddr, size));
    656 		break;
    657 
    658 	case (VM_PROT_READ | VM_PROT_WRITE):
    659 	case (VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE):
    660 		if (epp->ep_dsize != ~0) {
    661 			DPRINTF(("More than one data segment\n"));
    662 			return ENOEXEC;
    663 		}
    664 		epp->ep_daddr = vaddr;
    665 		epp->ep_dsize = size;
    666 
    667 		DPRINTF(("Elf Data@ 0x%x, size %d\n", vaddr, size));
    668 		break;
    669 
    670 	default:
    671 		DPRINTF(("Bad protection 0%o\n", prot));
    672 		return ENOEXEC;
    673 	}
    674 	return 0;
    675 }
    676 
    677 
    678 /*
    679  * linux_elf_read_from():
    680  *
    681  *	Read from vnode into buffer at offset.
    682  */
    683 static int
    684 linux_elf_read_from(vp, off, p, buf, size)
    685 	struct vnode	*vp;
    686 	u_long		 off;
    687 	struct proc	*p;
    688 	caddr_t		 buf;
    689 	int		 size;
    690 {
    691 	int	error;
    692 	int	resid;
    693 
    694 	DPRINTF(("read from 0x%x to 0x%x size %d\n",
    695 		 off, buf, size));
    696 	if ((error = vn_rdwr(UIO_READ, vp, buf, size,
    697 			     off, UIO_SYSSPACE, IO_NODELOCKED, p->p_ucred,
    698 			     &resid, p)) != 0) {
    699 		DPRINTF(("Bad read error %d\n", error));
    700 		return error;
    701 	}
    702 	/*
    703          * See if we got all of it
    704          */
    705 	if (resid != 0) {
    706 		DPRINTF(("Incomplete read for header ask=%d, rem=%d\n",
    707 			 size, resid));
    708 		return error;
    709 	}
    710 	return 0;
    711 }
    712 
    713 
    714 /*
    715  * linux_elf_load_file():
    716  *
    717  * Load a file (interpreter/library) pointed to by path
    718  * [stolen from coff_load_shlib()]. Made slightly more generic than
    719  * the svr4 version, for possible later use in linux_uselib().
    720  */
    721 static int
    722 linux_elf_load_file(p, path, vcset, entry, ap, last)
    723 	struct proc		*p;
    724 	char			*path;
    725 	struct exec_vmcmd_set   *vcset;
    726 	u_long			*entry;
    727 	struct elf_args		*ap;
    728 	u_long			*last;
    729 {
    730 	int			 error, i;
    731 	struct nameidata	 nd;
    732 	Elf32_Ehdr		 eh;
    733 	Elf32_Phdr		*ph = NULL;
    734 	u_long			 phsize;
    735 	char			*bp = NULL;
    736 	u_long			 addr = *last;
    737 
    738 	DPRINTF(("Loading file %s @ %x\n", path, addr));
    739 
    740 	if ((error = linux_emul_find(p, NULL, linux_emul_path, path, &bp, 0)) != 0)
    741 		bp = NULL;
    742 	else
    743 		path = bp;
    744 	/*
    745          * 1. open file
    746          * 2. read filehdr
    747          * 3. map text, data, and bss out of it using VM_*
    748          */
    749 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, path, p);
    750 	/* first get the vnode */
    751 	if ((error = namei(&nd)) != 0) {
    752 		if (bp != NULL)
    753 			free((char *) bp, M_TEMP);
    754 		return error;
    755 	}
    756 	if ((error = linux_elf_read_from(nd.ni_vp, 0, p, (caddr_t) &eh,
    757 				    sizeof(eh))) != 0)
    758 		goto bad;
    759 
    760 #ifdef DEBUG_EXEC_LINUX_ELF
    761 	print_Ehdr(&eh);
    762 #endif
    763 
    764 	if ((error = linux_elf_check_header(&eh, Elf32_et_dyn)) != 0)
    765 		goto bad;
    766 
    767 	phsize = eh.e_phnum * sizeof(Elf32_Phdr);
    768 	ph = (Elf32_Phdr *) malloc(phsize, M_TEMP, M_WAITOK);
    769 
    770 	if ((error = linux_elf_read_from(nd.ni_vp, eh.e_phoff, p,
    771 				    (caddr_t) ph, phsize)) != 0)
    772 		goto bad;
    773 
    774 	/*
    775          * Load all the necessary sections
    776          */
    777 	for (i = 0; i < eh.e_phnum; i++) {
    778 		u_long	size = 0;
    779 		int	prot = 0;
    780 #ifdef DEBUG_EXEC_LINUX_ELF
    781 		print_Phdr(&ph[i]);
    782 #endif
    783 
    784 		switch (ph[i].p_type) {
    785 		case Elf32_pt_load:
    786 			linux_elf_load_psection(vcset, nd.ni_vp, &ph[i], &addr,
    787 						&size, &prot);
    788 			/* Assume that the text segment is r-x only */
    789 			if ((prot & PROT_WRITE) == 0) {
    790 				*entry = addr + eh.e_entry;
    791 				ap->arg_interp = addr;
    792 				DPRINTF(("Interpreter@ 0x%x\n", addr));
    793 			}
    794 			addr += size;
    795 			break;
    796 
    797 		case Elf32_pt_dynamic:
    798 		case Elf32_pt_phdr:
    799 		case Elf32_pt_note:
    800 			break;
    801 
    802 		default:
    803 			DPRINTF(("interp: Unexpected program header type %d\n",
    804 				 ph[i].p_type));
    805 			break;
    806 		}
    807 	}
    808 
    809 bad:
    810 	if (ph != NULL)
    811 		free((char *) ph, M_TEMP);
    812 	if (bp != NULL)
    813 		free((char *) bp, M_TEMP);
    814 
    815 	*last = addr;
    816 	vrele(nd.ni_vp);
    817 	return error;
    818 }
    819 
    820 
    821 /*
    822  * exec_linux_elf_makecmds(): Prepare an Elf binary's exec package
    823  *
    824  * First, set of the various offsets/lengths in the exec package.
    825  *
    826  * Then, mark the text image busy (so it can be demand paged) or error
    827  * out if this is not possible.  Finally, set up vmcmds for the
    828  * text, data, bss, and stack segments.
    829  */
    830 int
    831 exec_linux_elf_makecmds(p, epp)
    832 	struct proc		*p;
    833 	struct exec_package	*epp;
    834 {
    835 	Elf32_Ehdr     *eh = epp->ep_hdr;
    836 	Elf32_Phdr     *ph, *pp;
    837 	int             error;
    838 	int             i;
    839 	char            interp[MAXPATHLEN];
    840 	u_long          pos = 0;
    841 	u_long          phsize;
    842 
    843 #ifdef DEBUG_EXEC_LINUX_ELF
    844 	print_Ehdr(eh);
    845 #endif
    846 	if (epp->ep_hdrvalid < sizeof(Elf32_Ehdr))
    847 		return ENOEXEC;
    848 
    849 	if (linux_elf_check_header(eh, Elf32_et_exec))
    850 		return ENOEXEC;
    851 
    852 	/*
    853          * check if vnode is in open for writing, because we want to
    854          * demand-page out of it.  if it is, don't do it, for various
    855          * reasons
    856          */
    857 	if (epp->ep_vp->v_writecount != 0) {
    858 #ifdef DIAGNOSTIC
    859 		if (epp->ep_vp->v_flag & VTEXT)
    860 			panic("exec: a VTEXT vnode has writecount != 0\n");
    861 #endif
    862 		return ETXTBSY;
    863 	}
    864 	/*
    865          * Allocate space to hold all the program headers, and read them
    866          * from the file
    867          */
    868 	phsize = eh->e_phnum * sizeof(Elf32_Phdr);
    869 	ph = (Elf32_Phdr *) malloc(phsize, M_TEMP, M_WAITOK);
    870 
    871 	if ((error = linux_elf_read_from(epp->ep_vp, eh->e_phoff, p,
    872 				    (caddr_t) ph, phsize)) != 0)
    873 		goto bad;
    874 
    875 	epp->ep_tsize = ~0;
    876 	epp->ep_dsize = ~0;
    877 
    878 	interp[0] = '\0';
    879 
    880 	/*
    881          * Load all the necessary sections
    882          */
    883 	for (i = 0; i < eh->e_phnum; i++) {
    884 		u_long          addr = ~0, size = 0;
    885 		int             prot = 0;
    886 
    887 		pp = &ph[i];
    888 #ifdef DEBUG_EXEC_LINUX_ELF
    889 		print_Phdr(pp);
    890 #endif
    891 
    892 		switch (ph[i].p_type) {
    893 		case Elf32_pt_load:
    894 			linux_elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
    895 				&ph[i], &addr, &size, &prot);
    896 			if ((error = linux_elf_set_segment(epp, addr, size,
    897 						      prot)) != 0)
    898 				goto bad;
    899 			break;
    900 
    901 		case Elf32_pt_shlib:
    902 			DPRINTF(("No support for COFF libraries (yet)\n"));
    903 			error = ENOEXEC;
    904 			goto bad;
    905 
    906 		case Elf32_pt_interp:
    907 			if (pp->p_filesz >= sizeof(interp)) {
    908 				DPRINTF(("Interpreter path too long %d\n",
    909 					 pp->p_filesz));
    910 				goto bad;
    911 			}
    912 			if ((error = linux_elf_read_from(epp->ep_vp, pp->p_offset, p,
    913 				      (caddr_t) interp, pp->p_filesz)) != 0)
    914 				goto bad;
    915 			break;
    916 
    917 		case Elf32_pt_dynamic:
    918 		case Elf32_pt_phdr:
    919 		case Elf32_pt_note:
    920 			break;
    921 
    922 		default:
    923 			/*
    924 			 * Not fatal, we don't need to understand everything
    925 			 * :-)
    926 			 */
    927 			DPRINTF(("Unsupported program header type %d\n",
    928 				 pp->p_type));
    929 			break;
    930 		}
    931 	}
    932 
    933 	/*
    934          * Check if we found a dynamically linked binary and arrange to load
    935          * it's interpreter
    936          */
    937 	if (interp[0]) {
    938 		struct elf_args *ap;
    939 		pos = ~0;
    940 
    941 		ap = (struct elf_args *) malloc(sizeof(struct elf_args),
    942 						 M_TEMP, M_WAITOK);
    943 		if ((error = linux_elf_load_file(p, interp, &epp->ep_vmcmds,
    944 				&epp->ep_entry, ap, &pos)) != 0) {
    945 			free((char *) ap, M_TEMP);
    946 			goto bad;
    947 		}
    948 		/* Arrange to load the program headers. */
    949 		pos = LINUX_ELF_ALIGN(pos + NBPG, NBPG);
    950 		DPRINTF(("Program header @0x%x\n", pos));
    951 		ap->arg_phaddr = pos;
    952 		NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, phsize,
    953 			  pos, epp->ep_vp, eh->e_phoff,
    954 			  VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE);
    955 		pos += phsize;
    956 
    957 		ap->arg_phentsize = eh->e_phentsize;
    958 		ap->arg_phnum = eh->e_phnum;
    959 		ap->arg_entry = eh->e_entry;
    960 
    961 		epp->ep_emul_arg = ap;
    962 	} else
    963 		epp->ep_entry = eh->e_entry;
    964 
    965 	DPRINTF(("taddr 0x%x tsize 0x%x daddr 0x%x dsize 0x%x\n",
    966 	       epp->ep_taddr, epp->ep_tsize, epp->ep_daddr, epp->ep_dsize));
    967 
    968 	free((char *) ph, M_TEMP);
    969 
    970 	DPRINTF(("Elf entry@ 0x%x\n", epp->ep_entry));
    971 	epp->ep_vp->v_flag |= VTEXT;
    972 
    973 	epp->ep_emul = &emul_linux_elf;
    974 
    975 	return exec_aout_setup_stack(p, epp);
    976 
    977 bad:
    978 	free((char *) ph, M_TEMP);
    979 	kill_vmcmds(&epp->ep_vmcmds);
    980 	return ENOEXEC;
    981 }
    982 /*
    983  * The Linux system call to load shared libraries, a.out version. The
    984  * a.out shared libs are just files that are mapped onto a fixed
    985  * address in the process' address space. The address is given in
    986  * a_entry. Read in the header, set up some VM commands and run them.
    987  *
    988  * Yes, both text and data are mapped at once, so we're left with
    989  * writeable text for the shared libs. The Linux crt0 seemed to break
    990  * sometimes when data was mapped seperately. It munmapped a uselib()
    991  * of ld.so by hand, which failed with shared text and data for ld.so
    992  * Yuck.
    993  *
    994  * Because of the problem with ZMAGIC executables (text starts
    995  * at 0x400 in the file, but needs to be mapped at 0), ZMAGIC
    996  * shared libs are not handled very efficiently :-(
    997  */
    998 
    999 int
   1000 linux_uselib(p, uap, retval)
   1001 	struct proc *p;
   1002 	struct linux_uselib_args /* {
   1003 		syscallarg(char *) path;
   1004 	} */ *uap;
   1005 	register_t *retval;
   1006 {
   1007 	caddr_t sg;
   1008 	long bsize, dsize, tsize, taddr, baddr, daddr;
   1009 	struct nameidata ni;
   1010 	struct vnode *vp;
   1011 	struct exec hdr;
   1012 	struct exec_vmcmd_set vcset;
   1013 	int rem, i, magic, error;
   1014 
   1015 	sg = stackgap_init();
   1016 	CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
   1017 
   1018 	NDINIT(&ni, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
   1019 
   1020 	if ((error = namei(&ni)))
   1021 		return error;
   1022 
   1023 	vp = ni.ni_vp;
   1024 
   1025 	if ((error = vn_rdwr(UIO_READ, vp, (caddr_t) &hdr, LINUX_AOUT_HDR_SIZE,
   1026 			     0, UIO_SYSSPACE, IO_NODELOCKED, p->p_ucred,
   1027 			     &rem, p))) {
   1028 		vrele(vp);
   1029 		return error;
   1030 	}
   1031 
   1032 	if (rem != 0) {
   1033 		vrele(vp);
   1034 		return ENOEXEC;
   1035 	}
   1036 
   1037 	if (LINUX_N_MACHTYPE(&hdr) != LINUX_MID_MACHINE)
   1038 		return ENOEXEC;
   1039 
   1040 	magic = LINUX_N_MAGIC(&hdr);
   1041 	taddr = hdr.a_entry & (~(NBPG - 1));
   1042 	tsize = hdr.a_text;
   1043 	daddr = taddr + tsize;
   1044 	dsize = hdr.a_data + hdr.a_bss;
   1045 
   1046 	if ((hdr.a_text != 0 || hdr.a_data != 0) && vp->v_writecount != 0) {
   1047 		vrele(vp);
   1048                 return ETXTBSY;
   1049         }
   1050 	vp->v_flag |= VTEXT;
   1051 
   1052 	vcset.evs_cnt = 0;
   1053 	vcset.evs_used = 0;
   1054 
   1055 	NEW_VMCMD(&vcset,
   1056 		  magic == ZMAGIC ? vmcmd_map_readvn : vmcmd_map_pagedvn,
   1057 		  hdr.a_text + hdr.a_data, taddr,
   1058 		  vp, LINUX_N_TXTOFF(hdr, magic),
   1059 		  VM_PROT_READ|VM_PROT_EXECUTE|VM_PROT_WRITE);
   1060 
   1061 	baddr = roundup(daddr + hdr.a_data, NBPG);
   1062 	bsize = daddr + dsize - baddr;
   1063         if (bsize > 0) {
   1064                 NEW_VMCMD(&vcset, vmcmd_map_zero, bsize, baddr,
   1065                     NULLVP, 0, VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
   1066 	}
   1067 
   1068 	for (i = 0; i < vcset.evs_used && !error; i++) {
   1069 		struct exec_vmcmd *vcp;
   1070 
   1071 		vcp = &vcset.evs_cmds[i];
   1072 		error = (*vcp->ev_proc)(p, vcp);
   1073 	}
   1074 
   1075 	kill_vmcmds(&vcset);
   1076 
   1077 	vrele(vp);
   1078 
   1079 	return error;
   1080 }
   1081 
   1082 /*
   1083  * Execve(2). Just check the alternate emulation path, and pass it on
   1084  * to the NetBSD execve().
   1085  */
   1086 int
   1087 linux_execve(p, uap, retval)
   1088 	struct proc *p;
   1089 	struct linux_execve_args /* {
   1090 		syscallarg(char *) path;
   1091 		syscallarg(char **) argv;
   1092 		syscallarg(char **) envp;
   1093 	} */ *uap;
   1094 	register_t *retval;
   1095 {
   1096 	caddr_t sg;
   1097 
   1098 	sg = stackgap_init();
   1099 	CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
   1100 
   1101 	return execve(p, uap, retval);
   1102 }
   1103