Home | History | Annotate | Line # | Download | only in kern
exec_elf.c revision 1.43.2.2
      1 /*	$NetBSD: exec_elf.c,v 1.43.2.2 2014/08/20 00:04:28 tls Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1994, 2000, 2005 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Christos Zoulas.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 1996 Christopher G. Demetriou
     34  * All rights reserved.
     35  *
     36  * Redistribution and use in source and binary forms, with or without
     37  * modification, are permitted provided that the following conditions
     38  * are met:
     39  * 1. Redistributions of source code must retain the above copyright
     40  *    notice, this list of conditions and the following disclaimer.
     41  * 2. Redistributions in binary form must reproduce the above copyright
     42  *    notice, this list of conditions and the following disclaimer in the
     43  *    documentation and/or other materials provided with the distribution.
     44  * 3. The name of the author may not be used to endorse or promote products
     45  *    derived from this software without specific prior written permission
     46  *
     47  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     48  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     49  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     50  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     51  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     52  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     53  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     54  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     55  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     56  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     57  */
     58 
     59 #include <sys/cdefs.h>
     60 __KERNEL_RCSID(1, "$NetBSD: exec_elf.c,v 1.43.2.2 2014/08/20 00:04:28 tls Exp $");
     61 
     62 #ifdef _KERNEL_OPT
     63 #include "opt_pax.h"
     64 #endif /* _KERNEL_OPT */
     65 
     66 #include <sys/param.h>
     67 #include <sys/proc.h>
     68 #include <sys/kmem.h>
     69 #include <sys/namei.h>
     70 #include <sys/vnode.h>
     71 #include <sys/exec.h>
     72 #include <sys/exec_elf.h>
     73 #include <sys/syscall.h>
     74 #include <sys/signalvar.h>
     75 #include <sys/mount.h>
     76 #include <sys/stat.h>
     77 #include <sys/kauth.h>
     78 #include <sys/bitops.h>
     79 #include <sys/cprng.h>
     80 
     81 #include <sys/cpu.h>
     82 #include <machine/reg.h>
     83 
     84 #include <compat/common/compat_util.h>
     85 
     86 #include <sys/pax.h>
     87 #include <uvm/uvm_param.h>
     88 
     89 extern struct emul emul_netbsd;
     90 
     91 #define elf_check_header	ELFNAME(check_header)
     92 #define elf_copyargs		ELFNAME(copyargs)
     93 #define elf_load_interp		ELFNAME(load_interp)
     94 #define elf_load_psection	ELFNAME(load_psection)
     95 #define exec_elf_makecmds	ELFNAME2(exec,makecmds)
     96 #define netbsd_elf_signature	ELFNAME2(netbsd,signature)
     97 #define netbsd_elf_probe	ELFNAME2(netbsd,probe)
     98 #define	coredump		ELFNAMEEND(coredump)
     99 #define	elf_free_emul_arg	ELFNAME(free_emul_arg)
    100 
    101 static int
    102 elf_load_interp(struct lwp *, struct exec_package *, char *,
    103     struct exec_vmcmd_set *, u_long *, Elf_Addr *);
    104 static void
    105 elf_load_psection(struct exec_vmcmd_set *, struct vnode *, const Elf_Phdr *,
    106     Elf_Addr *, u_long *, int);
    107 
    108 int	netbsd_elf_signature(struct lwp *, struct exec_package *, Elf_Ehdr *);
    109 int	netbsd_elf_probe(struct lwp *, struct exec_package *, void *, char *,
    110 	    vaddr_t *);
    111 
    112 static void	elf_free_emul_arg(void *);
    113 
    114 /* round up and down to page boundaries. */
    115 #define	ELF_ROUND(a, b)		(((a) + (b) - 1) & ~((b) - 1))
    116 #define	ELF_TRUNC(a, b)		((a) & ~((b) - 1))
    117 
    118 static void
    119 elf_placedynexec(struct lwp *l, struct exec_package *epp, Elf_Ehdr *eh,
    120     Elf_Phdr *ph)
    121 {
    122 	Elf_Addr align, offset;
    123 	int i;
    124 
    125 	for (align = i = 0; i < eh->e_phnum; i++)
    126 		if (ph[i].p_type == PT_LOAD && ph[i].p_align > align)
    127 			align = ph[i].p_align;
    128 
    129 #ifdef PAX_ASLR
    130 	if (pax_aslr_active(l)) {
    131 		size_t pax_align, l2, delta;
    132 		uint32_t r;
    133 
    134 		pax_align = align;
    135 
    136 		r = cprng_fast32();
    137 
    138 		if (pax_align == 0)
    139 			pax_align = PGSHIFT;
    140 		l2 = ilog2(pax_align);
    141 		delta = PAX_ASLR_DELTA(r, l2, PAX_ASLR_DELTA_EXEC_LEN);
    142 		offset = ELF_TRUNC(delta, pax_align) + PAGE_SIZE;
    143 #ifdef PAX_ASLR_DEBUG
    144 		uprintf("r=0x%x l2=0x%zx PGSHIFT=0x%x Delta=0x%zx\n", r, l2,
    145 		    PGSHIFT, delta);
    146 		uprintf("pax offset=0x%llx entry=0x%llx\n",
    147 		    (unsigned long long)offset,
    148 		    (unsigned long long)eh->e_entry);
    149 #endif /* PAX_ASLR_DEBUG */
    150 	} else
    151 #endif /* PAX_ASLR */
    152 		offset = MAX(align, PAGE_SIZE);
    153 
    154 	offset += epp->ep_vm_minaddr;
    155 
    156 	for (i = 0; i < eh->e_phnum; i++)
    157 		ph[i].p_vaddr += offset;
    158 	epp->ep_entryoffset = offset;
    159 	eh->e_entry += offset;
    160 }
    161 
    162 /*
    163  * Copy arguments onto the stack in the normal way, but add some
    164  * extra information in case of dynamic binding.
    165  */
    166 int
    167 elf_copyargs(struct lwp *l, struct exec_package *pack,
    168     struct ps_strings *arginfo, char **stackp, void *argp)
    169 {
    170 	size_t len, vlen;
    171 	AuxInfo ai[ELF_AUX_ENTRIES], *a, *execname;
    172 	struct elf_args *ap;
    173 	int error;
    174 
    175 	if ((error = copyargs(l, pack, arginfo, stackp, argp)) != 0)
    176 		return error;
    177 
    178 	a = ai;
    179 	execname = NULL;
    180 
    181 	/*
    182 	 * Push extra arguments on the stack needed by dynamically
    183 	 * linked binaries
    184 	 */
    185 	if ((ap = (struct elf_args *)pack->ep_emul_arg)) {
    186 		struct vattr *vap = pack->ep_vap;
    187 
    188 		a->a_type = AT_PHDR;
    189 		a->a_v = ap->arg_phaddr;
    190 		a++;
    191 
    192 		a->a_type = AT_PHENT;
    193 		a->a_v = ap->arg_phentsize;
    194 		a++;
    195 
    196 		a->a_type = AT_PHNUM;
    197 		a->a_v = ap->arg_phnum;
    198 		a++;
    199 
    200 		a->a_type = AT_PAGESZ;
    201 		a->a_v = PAGE_SIZE;
    202 		a++;
    203 
    204 		a->a_type = AT_BASE;
    205 		a->a_v = ap->arg_interp;
    206 		a++;
    207 
    208 		a->a_type = AT_FLAGS;
    209 		a->a_v = 0;
    210 		a++;
    211 
    212 		a->a_type = AT_ENTRY;
    213 		a->a_v = ap->arg_entry;
    214 		a++;
    215 
    216 		a->a_type = AT_EUID;
    217 		if (vap->va_mode & S_ISUID)
    218 			a->a_v = vap->va_uid;
    219 		else
    220 			a->a_v = kauth_cred_geteuid(l->l_cred);
    221 		a++;
    222 
    223 		a->a_type = AT_RUID;
    224 		a->a_v = kauth_cred_getuid(l->l_cred);
    225 		a++;
    226 
    227 		a->a_type = AT_EGID;
    228 		if (vap->va_mode & S_ISGID)
    229 			a->a_v = vap->va_gid;
    230 		else
    231 			a->a_v = kauth_cred_getegid(l->l_cred);
    232 		a++;
    233 
    234 		a->a_type = AT_RGID;
    235 		a->a_v = kauth_cred_getgid(l->l_cred);
    236 		a++;
    237 
    238 		a->a_type = AT_STACKBASE;
    239 		a->a_v = l->l_proc->p_stackbase;
    240 		a++;
    241 
    242 		if (pack->ep_path) {
    243 			execname = a;
    244 			a->a_type = AT_SUN_EXECNAME;
    245 			a++;
    246 		}
    247 
    248 		exec_free_emul_arg(pack);
    249 	}
    250 
    251 	a->a_type = AT_NULL;
    252 	a->a_v = 0;
    253 	a++;
    254 
    255 	vlen = (a - ai) * sizeof(ai[0]);
    256 
    257 	KASSERT(vlen <= sizeof(ai));
    258 
    259 	if (execname) {
    260 		char *path = pack->ep_path;
    261 		execname->a_v = (uintptr_t)(*stackp + vlen);
    262 		len = strlen(path) + 1;
    263 		if ((error = copyout(path, (*stackp + vlen), len)) != 0)
    264 			return error;
    265 		len = ALIGN(len);
    266 	} else
    267 		len = 0;
    268 
    269 	if ((error = copyout(ai, *stackp, vlen)) != 0)
    270 		return error;
    271 	*stackp += vlen + len;
    272 
    273 	return 0;
    274 }
    275 
    276 /*
    277  * elf_check_header():
    278  *
    279  * Check header for validity; return 0 if ok, ENOEXEC if error
    280  */
    281 int
    282 elf_check_header(Elf_Ehdr *eh)
    283 {
    284 
    285 	if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 ||
    286 	    eh->e_ident[EI_CLASS] != ELFCLASS)
    287 		return ENOEXEC;
    288 
    289 	switch (eh->e_machine) {
    290 
    291 	ELFDEFNNAME(MACHDEP_ID_CASES)
    292 
    293 	default:
    294 		return ENOEXEC;
    295 	}
    296 
    297 	if (ELF_EHDR_FLAGS_OK(eh) == 0)
    298 		return ENOEXEC;
    299 
    300 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM)
    301 		return ENOEXEC;
    302 
    303 	return 0;
    304 }
    305 
    306 /*
    307  * elf_load_psection():
    308  *
    309  * Load a psection at the appropriate address
    310  */
    311 static void
    312 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
    313     const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags)
    314 {
    315 	u_long msize, psize, rm, rf;
    316 	long diff, offset;
    317 	int vmprot = 0;
    318 
    319 	/*
    320 	 * If the user specified an address, then we load there.
    321 	 */
    322 	if (*addr == ELFDEFNNAME(NO_ADDR))
    323 		*addr = ph->p_vaddr;
    324 
    325 	if (ph->p_align > 1) {
    326 		/*
    327 		 * Make sure we are virtually aligned as we are supposed to be.
    328 		 */
    329 		diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
    330 		KASSERT(*addr - diff == ELF_TRUNC(*addr, ph->p_align));
    331 		/*
    332 		 * But make sure to not map any pages before the start of the
    333 		 * psection by limiting the difference to within a page.
    334 		 */
    335 		diff &= PAGE_MASK;
    336 	} else
    337 		diff = 0;
    338 
    339 	vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0;
    340 	vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0;
    341 	vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0;
    342 
    343 	/*
    344 	 * Adjust everything so it all starts on a page boundary.
    345 	 */
    346 	*addr -= diff;
    347 	offset = ph->p_offset - diff;
    348 	*size = ph->p_filesz + diff;
    349 	msize = ph->p_memsz + diff;
    350 
    351 	if (ph->p_align >= PAGE_SIZE) {
    352 		if ((ph->p_flags & PF_W) != 0) {
    353 			/*
    354 			 * Because the pagedvn pager can't handle zero fill
    355 			 * of the last data page if it's not page aligned we
    356 			 * map the last page readvn.
    357 			 */
    358 			psize = trunc_page(*size);
    359 		} else {
    360 			psize = round_page(*size);
    361 		}
    362 	} else {
    363 		psize = *size;
    364 	}
    365 
    366 	if (psize > 0) {
    367 		NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ?
    368 		    vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp,
    369 		    offset, vmprot, flags);
    370 		flags &= VMCMD_RELATIVE;
    371 	}
    372 	if (psize < *size) {
    373 		NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize,
    374 		    *addr + psize, vp, offset + psize, vmprot, flags);
    375 	}
    376 
    377 	/*
    378 	 * Check if we need to extend the size of the segment (does
    379 	 * bss extend page the next page boundary)?
    380 	 */
    381 	rm = round_page(*addr + msize);
    382 	rf = round_page(*addr + *size);
    383 
    384 	if (rm != rf) {
    385 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP,
    386 		    0, vmprot, flags & VMCMD_RELATIVE);
    387 		*size = msize;
    388 	}
    389 }
    390 
    391 /*
    392  * elf_load_interp():
    393  *
    394  * Load an interpreter pointed to by path.
    395  */
    396 static int
    397 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path,
    398     struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last)
    399 {
    400 	int error, i;
    401 	struct vnode *vp;
    402 	struct vattr attr;
    403 	Elf_Ehdr eh;
    404 	Elf_Phdr *ph = NULL;
    405 	const Elf_Phdr *base_ph;
    406 	const Elf_Phdr *last_ph;
    407 	u_long phsize;
    408 	Elf_Addr addr = *last;
    409 	struct proc *p;
    410 	bool use_topdown;
    411 
    412 	p = l->l_proc;
    413 
    414 	KASSERT(p->p_vmspace);
    415 	if (__predict_true(p->p_vmspace != proc0.p_vmspace)) {
    416 		use_topdown = p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN;
    417 	} else {
    418 #ifdef __USE_TOPDOWN_VM
    419 		use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM;
    420 #else
    421 		use_topdown = false;
    422 #endif
    423 	}
    424 
    425 	/*
    426 	 * 1. open file
    427 	 * 2. read filehdr
    428 	 * 3. map text, data, and bss out of it using VM_*
    429 	 */
    430 	vp = epp->ep_interp;
    431 	if (vp == NULL) {
    432 		error = emul_find_interp(l, epp, path);
    433 		if (error != 0)
    434 			return error;
    435 		vp = epp->ep_interp;
    436 	}
    437 	/* We'll tidy this ourselves - otherwise we have locking issues */
    438 	epp->ep_interp = NULL;
    439 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    440 
    441 	/*
    442 	 * Similarly, if it's not marked as executable, or it's not a regular
    443 	 * file, we don't allow it to be used.
    444 	 */
    445 	if (vp->v_type != VREG) {
    446 		error = EACCES;
    447 		goto badunlock;
    448 	}
    449 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
    450 		goto badunlock;
    451 
    452 	/* get attributes */
    453 	if ((error = VOP_GETATTR(vp, &attr, l->l_cred)) != 0)
    454 		goto badunlock;
    455 
    456 	/*
    457 	 * Check mount point.  Though we're not trying to exec this binary,
    458 	 * we will be executing code from it, so if the mount point
    459 	 * disallows execution or set-id-ness, we punt or kill the set-id.
    460 	 */
    461 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
    462 		error = EACCES;
    463 		goto badunlock;
    464 	}
    465 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
    466 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
    467 
    468 #ifdef notyet /* XXX cgd 960926 */
    469 	XXX cgd 960926: (maybe) VOP_OPEN it (and VOP_CLOSE in copyargs?)
    470 
    471 	XXXps: this problem will make it impossible to use an interpreter
    472 	from a file system which actually does something in VOP_OPEN
    473 #endif
    474 
    475 	error = vn_marktext(vp);
    476 	if (error)
    477 		goto badunlock;
    478 
    479 	VOP_UNLOCK(vp);
    480 
    481 	if ((error = exec_read_from(l, vp, 0, &eh, sizeof(eh))) != 0)
    482 		goto bad;
    483 
    484 	if ((error = elf_check_header(&eh)) != 0)
    485 		goto bad;
    486 	if (eh.e_type != ET_DYN || eh.e_phnum == 0) {
    487 		error = ENOEXEC;
    488 		goto bad;
    489 	}
    490 
    491 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
    492 	ph = kmem_alloc(phsize, KM_SLEEP);
    493 
    494 	if ((error = exec_read_from(l, vp, eh.e_phoff, ph, phsize)) != 0)
    495 		goto bad;
    496 
    497 #ifdef ELF_INTERP_NON_RELOCATABLE
    498 	/*
    499 	 * Evil hack:  Only MIPS should be non-relocatable, and the
    500 	 * psections should have a high address (typically 0x5ffe0000).
    501 	 * If it's now relocatable, it should be linked at 0 and the
    502 	 * psections should have zeros in the upper part of the address.
    503 	 * Otherwise, force the load at the linked address.
    504 	 */
    505 	if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0)
    506 		*last = ELFDEFNNAME(NO_ADDR);
    507 #endif
    508 
    509 	/*
    510 	 * If no position to load the interpreter was set by a probe
    511 	 * function, pick the same address that a non-fixed mmap(0, ..)
    512 	 * would (i.e. something safely out of the way).
    513 	 */
    514 	if (*last == ELFDEFNNAME(NO_ADDR)) {
    515 		u_long limit = 0;
    516 		/*
    517 		 * Find the start and ending addresses of the psections to
    518 		 * be loaded.  This will give us the size.
    519 		 */
    520 		for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) {
    521 			if (ph[i].p_type == PT_LOAD) {
    522 				u_long psize = ph[i].p_vaddr + ph[i].p_memsz;
    523 				if (base_ph == NULL)
    524 					base_ph = &ph[i];
    525 				if (psize > limit)
    526 					limit = psize;
    527 			}
    528 		}
    529 
    530 		if (base_ph == NULL) {
    531 			error = ENOEXEC;
    532 			goto bad;
    533 		}
    534 
    535 		/*
    536 		 * Now compute the size and load address.
    537 		 */
    538 		addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p,
    539 		    epp->ep_daddr,
    540 		    round_page(limit) - trunc_page(base_ph->p_vaddr));
    541 	} else
    542 		addr = *last; /* may be ELF_LINK_ADDR */
    543 
    544 	/*
    545 	 * Load all the necessary sections
    546 	 */
    547 	for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) {
    548 		switch (ph[i].p_type) {
    549 		case PT_LOAD: {
    550 			u_long size;
    551 			int flags;
    552 
    553 			if (base_ph == NULL) {
    554 				/*
    555 				 * First encountered psection is always the
    556 				 * base psection.  Make sure it's aligned
    557 				 * properly (align down for topdown and align
    558 				 * upwards for not topdown).
    559 				 */
    560 				base_ph = &ph[i];
    561 				flags = VMCMD_BASE;
    562 				if (addr == ELF_LINK_ADDR)
    563 					addr = ph[i].p_vaddr;
    564 				if (use_topdown)
    565 					addr = ELF_TRUNC(addr, ph[i].p_align);
    566 				else
    567 					addr = ELF_ROUND(addr, ph[i].p_align);
    568 			} else {
    569 				u_long limit = round_page(last_ph->p_vaddr
    570 				    + last_ph->p_memsz);
    571 				u_long base = trunc_page(ph[i].p_vaddr);
    572 
    573 				/*
    574 				 * If there is a gap in between the psections,
    575 				 * map it as inaccessible so nothing else
    576 				 * mmap'ed will be placed there.
    577 				 */
    578 				if (limit != base) {
    579 					NEW_VMCMD2(vcset, vmcmd_map_zero,
    580 					    base - limit,
    581 					    limit - base_ph->p_vaddr, NULLVP,
    582 					    0, VM_PROT_NONE, VMCMD_RELATIVE);
    583 				}
    584 
    585 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
    586 				flags = VMCMD_RELATIVE;
    587 			}
    588 			last_ph = &ph[i];
    589 			elf_load_psection(vcset, vp, &ph[i], &addr,
    590 			    &size, flags);
    591 			/*
    592 			 * If entry is within this psection then this
    593 			 * must contain the .text section.  *entryoff is
    594 			 * relative to the base psection.
    595 			 */
    596 			if (eh.e_entry >= ph[i].p_vaddr &&
    597 			    eh.e_entry < (ph[i].p_vaddr + size)) {
    598 				*entryoff = eh.e_entry - base_ph->p_vaddr;
    599 			}
    600 			addr += size;
    601 			break;
    602 		}
    603 
    604 		default:
    605 			break;
    606 		}
    607 	}
    608 
    609 	kmem_free(ph, phsize);
    610 	/*
    611 	 * This value is ignored if TOPDOWN.
    612 	 */
    613 	*last = addr;
    614 	vrele(vp);
    615 	return 0;
    616 
    617 badunlock:
    618 	VOP_UNLOCK(vp);
    619 
    620 bad:
    621 	if (ph != NULL)
    622 		kmem_free(ph, phsize);
    623 #ifdef notyet /* XXX cgd 960926 */
    624 	(maybe) VOP_CLOSE it
    625 #endif
    626 	vrele(vp);
    627 	return error;
    628 }
    629 
    630 /*
    631  * exec_elf_makecmds(): Prepare an Elf binary's exec package
    632  *
    633  * First, set of the various offsets/lengths in the exec package.
    634  *
    635  * Then, mark the text image busy (so it can be demand paged) or error
    636  * out if this is not possible.  Finally, set up vmcmds for the
    637  * text, data, bss, and stack segments.
    638  */
    639 int
    640 exec_elf_makecmds(struct lwp *l, struct exec_package *epp)
    641 {
    642 	Elf_Ehdr *eh = epp->ep_hdr;
    643 	Elf_Phdr *ph, *pp;
    644 	Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0;
    645 	int error, i;
    646 	char *interp = NULL;
    647 	u_long phsize;
    648 	struct elf_args *ap;
    649 	bool is_dyn = false;
    650 
    651 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr))
    652 		return ENOEXEC;
    653 	if ((error = elf_check_header(eh)) != 0)
    654 		return error;
    655 
    656 	if (eh->e_type == ET_DYN)
    657 		/*
    658 		 * XXX allow for executing shared objects. It seems silly
    659 		 * but other ELF-based systems allow it as well.
    660 		 */
    661 		is_dyn = true;
    662 	else if (eh->e_type != ET_EXEC)
    663 		return ENOEXEC;
    664 
    665 	if (eh->e_phnum == 0)
    666 		return ENOEXEC;
    667 
    668 	error = vn_marktext(epp->ep_vp);
    669 	if (error)
    670 		return error;
    671 
    672 	/*
    673 	 * Allocate space to hold all the program headers, and read them
    674 	 * from the file
    675 	 */
    676 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
    677 	ph = kmem_alloc(phsize, KM_SLEEP);
    678 
    679 	if ((error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize)) !=
    680 	    0)
    681 		goto bad;
    682 
    683 	epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
    684 	epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
    685 
    686 	for (i = 0; i < eh->e_phnum; i++) {
    687 		pp = &ph[i];
    688 		if (pp->p_type == PT_INTERP) {
    689 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) {
    690 				error = ENOEXEC;
    691 				goto bad;
    692 			}
    693 			interp = PNBUF_GET();
    694 			if ((error = exec_read_from(l, epp->ep_vp,
    695 			    pp->p_offset, interp, pp->p_filesz)) != 0)
    696 				goto bad;
    697 			/* Ensure interp is NUL-terminated and of the expected length */
    698 			if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) {
    699 				error = ENOEXEC;
    700 				goto bad;
    701 			}
    702 			break;
    703 		}
    704 	}
    705 
    706 	/*
    707 	 * On the same architecture, we may be emulating different systems.
    708 	 * See which one will accept this executable.
    709 	 *
    710 	 * Probe functions would normally see if the interpreter (if any)
    711 	 * exists. Emulation packages may possibly replace the interpreter in
    712 	 * interp[] with a changed path (/emul/xxx/<path>).
    713 	 */
    714 	pos = ELFDEFNNAME(NO_ADDR);
    715 	if (epp->ep_esch->u.elf_probe_func) {
    716 		vaddr_t startp = (vaddr_t)pos;
    717 
    718 		error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp,
    719 							  &startp);
    720 		if (error)
    721 			goto bad;
    722 		pos = (Elf_Addr)startp;
    723 	}
    724 
    725 #if defined(PAX_MPROTECT) || defined(PAX_SEGVGUARD) || defined(PAX_ASLR)
    726 	l->l_proc->p_pax = epp->ep_pax_flags;
    727 #endif /* PAX_MPROTECT || PAX_SEGVGUARD || PAX_ASLR */
    728 
    729 	if (is_dyn)
    730 		elf_placedynexec(l, epp, eh, ph);
    731 
    732 	/*
    733 	 * Load all the necessary sections
    734 	 */
    735 	for (i = 0; i < eh->e_phnum; i++) {
    736 		Elf_Addr addr = ELFDEFNNAME(NO_ADDR);
    737 		u_long size = 0;
    738 
    739 		switch (ph[i].p_type) {
    740 		case PT_LOAD:
    741 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
    742 			    &ph[i], &addr, &size, VMCMD_FIXED);
    743 
    744 			/*
    745 			 * Consider this as text segment, if it is executable.
    746 			 * If there is more than one text segment, pick the
    747 			 * largest.
    748 			 */
    749 			if (ph[i].p_flags & PF_X) {
    750 				if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) ||
    751 				    size > epp->ep_tsize) {
    752 					epp->ep_taddr = addr;
    753 					epp->ep_tsize = size;
    754 				}
    755 				end_text = addr + size;
    756 			} else {
    757 				epp->ep_daddr = addr;
    758 				epp->ep_dsize = size;
    759 			}
    760 			if (ph[i].p_offset == 0) {
    761 				computed_phdr = ph[i].p_vaddr + eh->e_phoff;
    762 			}
    763 			break;
    764 
    765 		case PT_SHLIB:
    766 			/* SCO has these sections. */
    767 		case PT_INTERP:
    768 			/* Already did this one. */
    769 		case PT_DYNAMIC:
    770 		case PT_NOTE:
    771 			break;
    772 		case PT_PHDR:
    773 			/* Note address of program headers (in text segment) */
    774 			phdr = ph[i].p_vaddr;
    775 			break;
    776 
    777 		default:
    778 			/*
    779 			 * Not fatal; we don't need to understand everything.
    780 			 */
    781 			break;
    782 		}
    783 	}
    784 
    785 	if (epp->ep_vmcmds.evs_used == 0) {
    786 		/* No VMCMD; there was no PT_LOAD section, or those
    787 		 * sections were empty */
    788 		error = ENOEXEC;
    789 		goto bad;
    790 	}
    791 
    792 	if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) {
    793 		epp->ep_daddr = round_page(end_text);
    794 		epp->ep_dsize = 0;
    795 	}
    796 
    797 	/*
    798 	 * Check if we found a dynamically linked binary and arrange to load
    799 	 * its interpreter
    800 	 */
    801 	if (interp) {
    802 		u_int nused = epp->ep_vmcmds.evs_used;
    803 		u_long interp_offset = 0;
    804 
    805 		if ((error = elf_load_interp(l, epp, interp,
    806 		    &epp->ep_vmcmds, &interp_offset, &pos)) != 0) {
    807 			goto bad;
    808 		}
    809 		if (epp->ep_vmcmds.evs_used == nused) {
    810 			/* elf_load_interp() has not set up any new VMCMD */
    811 			error = ENOEXEC;
    812 			goto bad;
    813 		}
    814 
    815 		ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
    816 		ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr;
    817 		epp->ep_entryoffset = interp_offset;
    818 		epp->ep_entry = ap->arg_interp + interp_offset;
    819 		PNBUF_PUT(interp);
    820 	} else {
    821 		epp->ep_entry = eh->e_entry;
    822 		if (epp->ep_flags & EXEC_FORCEAUX) {
    823 			ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
    824 			ap->arg_interp = (vaddr_t)NULL;
    825 		} else
    826 			ap = NULL;
    827 	}
    828 
    829 	if (ap) {
    830 		ap->arg_phaddr = phdr ? phdr : computed_phdr;
    831 		ap->arg_phentsize = eh->e_phentsize;
    832 		ap->arg_phnum = eh->e_phnum;
    833 		ap->arg_entry = eh->e_entry;
    834 		epp->ep_emul_arg = ap;
    835 		epp->ep_emul_arg_free = elf_free_emul_arg;
    836 	}
    837 
    838 #ifdef ELF_MAP_PAGE_ZERO
    839 	/* Dell SVR4 maps page zero, yeuch! */
    840 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0,
    841 	    epp->ep_vp, 0, VM_PROT_READ);
    842 #endif
    843 	kmem_free(ph, phsize);
    844 	return (*epp->ep_esch->es_setup_stack)(l, epp);
    845 
    846 bad:
    847 	if (interp)
    848 		PNBUF_PUT(interp);
    849 	exec_free_emul_arg(epp);
    850 	kmem_free(ph, phsize);
    851 	kill_vmcmds(&epp->ep_vmcmds);
    852 	return error;
    853 }
    854 
    855 int
    856 netbsd_elf_signature(struct lwp *l, struct exec_package *epp,
    857     Elf_Ehdr *eh)
    858 {
    859 	size_t i;
    860 	Elf_Shdr *sh;
    861 	Elf_Nhdr *np;
    862 	size_t shsize, nsize;
    863 	int error;
    864 	int isnetbsd = 0;
    865 	char *ndata, *ndesc;
    866 #ifdef COMPAT_OLDNOTE
    867 	int compat_oldnote = -1;
    868 #endif
    869 
    870 #ifdef DIAGNOSTIC
    871 	const char *badnote;
    872 #define BADNOTE(n) badnote = (n)
    873 #else
    874 #define BADNOTE(n)
    875 #endif
    876 
    877 	epp->ep_pax_flags = 0;
    878 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_shnum == 0)
    879 		return ENOEXEC;
    880 
    881 	shsize = eh->e_shnum * sizeof(Elf_Shdr);
    882 	sh = kmem_alloc(shsize, KM_SLEEP);
    883 	error = exec_read_from(l, epp->ep_vp, eh->e_shoff, sh, shsize);
    884 	if (error)
    885 		goto out;
    886 
    887 	np = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP);
    888 	for (i = 0; i < eh->e_shnum; i++) {
    889 		Elf_Shdr *shp = &sh[i];
    890 
    891 		if (shp->sh_type != SHT_NOTE ||
    892 		    shp->sh_size > ELF_MAXNOTESIZE ||
    893 		    shp->sh_size < sizeof(Elf_Nhdr) + ELF_NOTE_NETBSD_NAMESZ)
    894 			continue;
    895 
    896 		error = exec_read_from(l, epp->ep_vp, shp->sh_offset, np,
    897 		    shp->sh_size);
    898 		if (error)
    899 			continue;
    900 
    901 		/* Point to the note, skip the header */
    902 		ndata = (char *)(np + 1);
    903 
    904 		/*
    905 		 * Padding is present if necessary to ensure 4-byte alignment.
    906 		 * The actual section size is therefore:
    907 		 *    header size + 4-byte aligned name + 4-byte aligned desc
    908 		 * Ensure this size is consistent with what is indicated
    909 		 * in sh_size. The first check avoids integer overflows.
    910 		 */
    911 		if (np->n_namesz > shp->sh_size || np->n_descsz > shp->sh_size) {
    912 			BADNOTE("note size limit");
    913 			goto bad;
    914 		}
    915 		nsize = sizeof(*np) + roundup(np->n_namesz, 4) +
    916 		    roundup(np->n_descsz, 4);
    917 		if (nsize != shp->sh_size) {
    918 			BADNOTE("note size");
    919 #ifdef COMPAT_OLDNOTE
    920 			if (nsize > shp->sh_size || compat_oldnote == 0)
    921 #endif
    922 			goto bad;
    923 		}
    924 		ndesc = ndata + roundup(np->n_namesz, 4);
    925 
    926 		switch (np->n_type) {
    927 		case ELF_NOTE_TYPE_NETBSD_TAG:
    928 			/* It is us */
    929 			if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ &&
    930 			    np->n_descsz == ELF_NOTE_NETBSD_DESCSZ &&
    931 			    memcmp(ndata, ELF_NOTE_NETBSD_NAME,
    932 			    ELF_NOTE_NETBSD_NAMESZ) == 0) {
    933 				memcpy(&epp->ep_osversion, ndesc,
    934 				    ELF_NOTE_NETBSD_DESCSZ);
    935 				isnetbsd = 1;
    936 #ifdef COMPAT_OLDNOTE
    937 				compat_oldnote = epp->ep_osversion == 199905;
    938 #endif
    939 				break;
    940 			}
    941 
    942 			/*
    943 			 * Ignore SuSE tags; SuSE's n_type is the same as NetBSD's
    944 			 * one.
    945 			 */
    946 			if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ &&
    947 			    memcmp(ndata, ELF_NOTE_SUSE_NAME,
    948 			    ELF_NOTE_SUSE_NAMESZ) == 0)
    949 				break;
    950 #ifdef COMPAT_OLDNOTE
    951 			if (compat_oldnote == 1)
    952 				break;
    953 #endif
    954 			BADNOTE("NetBSD tag");
    955 			goto bad;
    956 
    957 		case ELF_NOTE_TYPE_PAX_TAG:
    958 			if (np->n_namesz == ELF_NOTE_PAX_NAMESZ &&
    959 			    np->n_descsz == ELF_NOTE_PAX_DESCSZ &&
    960 			    memcmp(ndata, ELF_NOTE_PAX_NAME,
    961 			    ELF_NOTE_PAX_NAMESZ) == 0) {
    962 				memcpy(&epp->ep_pax_flags, ndesc,
    963 				    sizeof(epp->ep_pax_flags));
    964 				break;
    965 			}
    966 			BADNOTE("PaX tag");
    967 			goto bad;
    968 
    969 		case ELF_NOTE_TYPE_MARCH_TAG:
    970 			/* Copy the machine arch into the package. */
    971 			if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ
    972 			    && memcmp(ndata, ELF_NOTE_MARCH_NAME,
    973 				    ELF_NOTE_MARCH_NAMESZ) == 0) {
    974 				/* Do not truncate the buffer */
    975 				if (np->n_descsz > sizeof(epp->ep_machine_arch)) {
    976 					BADNOTE("description size limit");
    977 					goto bad;
    978 				}
    979 				/*
    980 				 * Ensure ndesc is NUL-terminated and of the
    981 				 * expected length.
    982 				 */
    983 				if (strnlen(ndesc, np->n_descsz) + 1 !=
    984 				    np->n_descsz) {
    985 					BADNOTE("description size");
    986 					goto bad;
    987 				}
    988 				strlcpy(epp->ep_machine_arch, ndesc,
    989 				    sizeof(epp->ep_machine_arch));
    990 				break;
    991 			}
    992 			BADNOTE("march tag");
    993 			goto bad;
    994 
    995 		case ELF_NOTE_TYPE_MCMODEL_TAG:
    996 			/* arch specific check for code model */
    997 #ifdef ELF_MD_MCMODEL_CHECK
    998 			if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ
    999 			    && memcmp(ndata, ELF_NOTE_MCMODEL_NAME,
   1000 				    ELF_NOTE_MCMODEL_NAMESZ) == 0) {
   1001 				ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz);
   1002 				break;
   1003 			}
   1004 			BADNOTE("mcmodel tag");
   1005 			goto bad;
   1006 #endif
   1007 			break;
   1008 
   1009 		case ELF_NOTE_TYPE_SUSE_VERSION_TAG:
   1010 			break;
   1011 
   1012 		default:
   1013 			BADNOTE("unknown tag");
   1014 bad:
   1015 #ifdef DIAGNOSTIC
   1016 			/* Ignore GNU tags */
   1017 			if (np->n_namesz == ELF_NOTE_GNU_NAMESZ &&
   1018 			    memcmp(ndata, ELF_NOTE_GNU_NAME,
   1019 			    ELF_NOTE_GNU_NAMESZ) == 0)
   1020 			    break;
   1021 
   1022 			int ns = MIN(np->n_namesz, shp->sh_size - sizeof(*np));
   1023 			printf("%s: Unknown elf note type %d (%s): "
   1024 			    "[namesz=%d, descsz=%d name=%-*.*s]\n",
   1025 			    epp->ep_kname, np->n_type, badnote, np->n_namesz,
   1026 			    np->n_descsz, ns, ns, ndata);
   1027 #endif
   1028 			break;
   1029 		}
   1030 	}
   1031 	kmem_free(np, ELF_MAXNOTESIZE);
   1032 
   1033 	error = isnetbsd ? 0 : ENOEXEC;
   1034 out:
   1035 	kmem_free(sh, shsize);
   1036 	return error;
   1037 }
   1038 
   1039 int
   1040 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp,
   1041     vaddr_t *pos)
   1042 {
   1043 	int error;
   1044 
   1045 	if ((error = netbsd_elf_signature(l, epp, eh)) != 0)
   1046 		return error;
   1047 #ifdef ELF_MD_PROBE_FUNC
   1048 	if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0)
   1049 		return error;
   1050 #elif defined(ELF_INTERP_NON_RELOCATABLE)
   1051 	*pos = ELF_LINK_ADDR;
   1052 #endif
   1053 	epp->ep_flags |= EXEC_FORCEAUX;
   1054 	return 0;
   1055 }
   1056 
   1057 void
   1058 elf_free_emul_arg(void *arg)
   1059 {
   1060 	struct elf_args *ap = arg;
   1061 	KASSERT(ap != NULL);
   1062 	kmem_free(ap, sizeof(*ap));
   1063 }
   1064