Home | History | Annotate | Line # | Download | only in kern
exec_elf.c revision 1.85.2.2
      1 /*	$NetBSD: exec_elf.c,v 1.85.2.2 2017/03/20 06:57:47 pgoyette Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1994, 2000, 2005, 2015 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 and Maxime Villard.
      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 1.85.2.2 2017/03/20 06:57:47 pgoyette 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 
     80 #include <sys/cpu.h>
     81 #include <machine/reg.h>
     82 
     83 #include <compat/common/compat_util.h>
     84 
     85 #include <sys/pax.h>
     86 #include <uvm/uvm_param.h>
     87 
     88 extern struct emul emul_netbsd;
     89 
     90 #define elf_check_header	ELFNAME(check_header)
     91 #define elf_copyargs		ELFNAME(copyargs)
     92 #define elf_load_interp		ELFNAME(load_interp)
     93 #define elf_load_psection	ELFNAME(load_psection)
     94 #define exec_elf_makecmds	ELFNAME2(exec,makecmds)
     95 #define netbsd_elf_signature	ELFNAME2(netbsd,signature)
     96 #define netbsd_elf_note       	ELFNAME2(netbsd,note)
     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_note(struct exec_package *, const Elf_Nhdr *, const char *,
    110 	    const char *);
    111 int	netbsd_elf_probe(struct lwp *, struct exec_package *, void *, char *,
    112 	    vaddr_t *);
    113 
    114 static void	elf_free_emul_arg(void *);
    115 
    116 #ifdef DEBUG_ELF
    117 #define DPRINTF(a, ...)	printf("%s: " a "\n", __func__, ##__VA_ARGS__)
    118 #else
    119 #define DPRINTF(a, ...)
    120 #endif
    121 
    122 /* round up and down to page boundaries. */
    123 #define	ELF_ROUND(a, b)		(((a) + (b) - 1) & ~((b) - 1))
    124 #define	ELF_TRUNC(a, b)		((a) & ~((b) - 1))
    125 
    126 static void
    127 elf_placedynexec(struct exec_package *epp, Elf_Ehdr *eh, Elf_Phdr *ph)
    128 {
    129 	Elf_Addr align, offset;
    130 	int i;
    131 
    132 	for (align = i = 0; i < eh->e_phnum; i++)
    133 		if (ph[i].p_type == PT_LOAD && ph[i].p_align > align)
    134 			align = ph[i].p_align;
    135 
    136 	offset = (Elf_Addr)pax_aslr_exec_offset(epp, align);
    137 	if (offset < epp->ep_vm_minaddr)
    138 		offset = roundup(epp->ep_vm_minaddr, align);
    139 	KASSERT((offset & (align - 1)) == 0);
    140 
    141 	for (i = 0; i < eh->e_phnum; i++)
    142 		ph[i].p_vaddr += offset;
    143 	epp->ep_entryoffset = offset;
    144 	eh->e_entry += offset;
    145 }
    146 
    147 /*
    148  * Copy arguments onto the stack in the normal way, but add some
    149  * extra information in case of dynamic binding.
    150  */
    151 int
    152 elf_copyargs(struct lwp *l, struct exec_package *pack,
    153     struct ps_strings *arginfo, char **stackp, void *argp)
    154 {
    155 	size_t len, vlen;
    156 	AuxInfo ai[ELF_AUX_ENTRIES], *a, *execname;
    157 	struct elf_args *ap;
    158 	int error;
    159 
    160 	if ((error = copyargs(l, pack, arginfo, stackp, argp)) != 0)
    161 		return error;
    162 
    163 	a = ai;
    164 	execname = NULL;
    165 
    166 	memset(ai, 0, sizeof(ai));
    167 
    168 	/*
    169 	 * Push extra arguments on the stack needed by dynamically
    170 	 * linked binaries
    171 	 */
    172 	if ((ap = (struct elf_args *)pack->ep_emul_arg)) {
    173 		struct vattr *vap = pack->ep_vap;
    174 
    175 		a->a_type = AT_PHDR;
    176 		a->a_v = ap->arg_phaddr;
    177 		a++;
    178 
    179 		a->a_type = AT_PHENT;
    180 		a->a_v = ap->arg_phentsize;
    181 		a++;
    182 
    183 		a->a_type = AT_PHNUM;
    184 		a->a_v = ap->arg_phnum;
    185 		a++;
    186 
    187 		a->a_type = AT_PAGESZ;
    188 		a->a_v = PAGE_SIZE;
    189 		a++;
    190 
    191 		a->a_type = AT_BASE;
    192 		a->a_v = ap->arg_interp;
    193 		a++;
    194 
    195 		a->a_type = AT_FLAGS;
    196 		a->a_v = 0;
    197 		a++;
    198 
    199 		a->a_type = AT_ENTRY;
    200 		a->a_v = ap->arg_entry;
    201 		a++;
    202 
    203 		a->a_type = AT_EUID;
    204 		if (vap->va_mode & S_ISUID)
    205 			a->a_v = vap->va_uid;
    206 		else
    207 			a->a_v = kauth_cred_geteuid(l->l_cred);
    208 		a++;
    209 
    210 		a->a_type = AT_RUID;
    211 		a->a_v = kauth_cred_getuid(l->l_cred);
    212 		a++;
    213 
    214 		a->a_type = AT_EGID;
    215 		if (vap->va_mode & S_ISGID)
    216 			a->a_v = vap->va_gid;
    217 		else
    218 			a->a_v = kauth_cred_getegid(l->l_cred);
    219 		a++;
    220 
    221 		a->a_type = AT_RGID;
    222 		a->a_v = kauth_cred_getgid(l->l_cred);
    223 		a++;
    224 
    225 		a->a_type = AT_STACKBASE;
    226 		a->a_v = l->l_proc->p_stackbase;
    227 		a++;
    228 
    229 		if (pack->ep_path) {
    230 			execname = a;
    231 			a->a_type = AT_SUN_EXECNAME;
    232 			a++;
    233 		}
    234 
    235 		exec_free_emul_arg(pack);
    236 	}
    237 
    238 	a->a_type = AT_NULL;
    239 	a->a_v = 0;
    240 	a++;
    241 
    242 	vlen = (a - ai) * sizeof(ai[0]);
    243 
    244 	KASSERT(vlen <= sizeof(ai));
    245 
    246 	if (execname) {
    247 		char *path = pack->ep_path;
    248 		execname->a_v = (uintptr_t)(*stackp + vlen);
    249 		len = strlen(path) + 1;
    250 		if ((error = copyout(path, (*stackp + vlen), len)) != 0)
    251 			return error;
    252 		len = ALIGN(len);
    253 	} else
    254 		len = 0;
    255 
    256 	if ((error = copyout(ai, *stackp, vlen)) != 0)
    257 		return error;
    258 	*stackp += vlen + len;
    259 
    260 	return 0;
    261 }
    262 
    263 /*
    264  * elf_check_header():
    265  *
    266  * Check header for validity; return 0 if ok, ENOEXEC if error
    267  */
    268 int
    269 elf_check_header(Elf_Ehdr *eh)
    270 {
    271 
    272 	if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 ||
    273 	    eh->e_ident[EI_CLASS] != ELFCLASS) {
    274 		DPRINTF("bad magic %#x%x%x", eh->e_ident[0], eh->e_ident[1],
    275 		    eh->e_ident[2]);
    276 		return ENOEXEC;
    277 	}
    278 
    279 	switch (eh->e_machine) {
    280 
    281 	ELFDEFNNAME(MACHDEP_ID_CASES)
    282 
    283 	default:
    284 		DPRINTF("bad machine %#x", eh->e_machine);
    285 		return ENOEXEC;
    286 	}
    287 
    288 	if (ELF_EHDR_FLAGS_OK(eh) == 0) {
    289 		DPRINTF("bad flags %#x", eh->e_flags);
    290 		return ENOEXEC;
    291 	}
    292 
    293 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM) {
    294 		DPRINTF("bad shnum/phnum %#x/%#x", eh->e_shnum, eh->e_phnum);
    295 		return ENOEXEC;
    296 	}
    297 
    298 	return 0;
    299 }
    300 
    301 /*
    302  * elf_load_psection():
    303  *
    304  * Load a psection at the appropriate address
    305  */
    306 static void
    307 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
    308     const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags)
    309 {
    310 	u_long msize, psize, rm, rf;
    311 	long diff, offset;
    312 	int vmprot = 0;
    313 
    314 	/*
    315 	 * If the user specified an address, then we load there.
    316 	 */
    317 	if (*addr == ELFDEFNNAME(NO_ADDR))
    318 		*addr = ph->p_vaddr;
    319 
    320 	if (ph->p_align > 1) {
    321 		/*
    322 		 * Make sure we are virtually aligned as we are supposed to be.
    323 		 */
    324 		diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
    325 		KASSERT(*addr - diff == ELF_TRUNC(*addr, ph->p_align));
    326 		/*
    327 		 * But make sure to not map any pages before the start of the
    328 		 * psection by limiting the difference to within a page.
    329 		 */
    330 		diff &= PAGE_MASK;
    331 	} else
    332 		diff = 0;
    333 
    334 	vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0;
    335 	vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0;
    336 	vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0;
    337 
    338 	/*
    339 	 * Adjust everything so it all starts on a page boundary.
    340 	 */
    341 	*addr -= diff;
    342 	offset = ph->p_offset - diff;
    343 	*size = ph->p_filesz + diff;
    344 	msize = ph->p_memsz + diff;
    345 
    346 	if (ph->p_align >= PAGE_SIZE) {
    347 		if ((ph->p_flags & PF_W) != 0) {
    348 			/*
    349 			 * Because the pagedvn pager can't handle zero fill
    350 			 * of the last data page if it's not page aligned we
    351 			 * map the last page readvn.
    352 			 */
    353 			psize = trunc_page(*size);
    354 		} else {
    355 			psize = round_page(*size);
    356 		}
    357 	} else {
    358 		psize = *size;
    359 	}
    360 
    361 	if (psize > 0) {
    362 		NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ?
    363 		    vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp,
    364 		    offset, vmprot, flags);
    365 		flags &= VMCMD_RELATIVE;
    366 	}
    367 	if (psize < *size) {
    368 		NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize,
    369 		    *addr + psize, vp, offset + psize, vmprot, flags);
    370 	}
    371 
    372 	/*
    373 	 * Check if we need to extend the size of the segment (does
    374 	 * bss extend page the next page boundary)?
    375 	 */
    376 	rm = round_page(*addr + msize);
    377 	rf = round_page(*addr + *size);
    378 
    379 	if (rm != rf) {
    380 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP,
    381 		    0, vmprot, flags & VMCMD_RELATIVE);
    382 		*size = msize;
    383 	}
    384 }
    385 
    386 /*
    387  * elf_load_interp():
    388  *
    389  * Load an interpreter pointed to by path.
    390  */
    391 static int
    392 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path,
    393     struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last)
    394 {
    395 	int error, i;
    396 	struct vnode *vp;
    397 	struct vattr attr;
    398 	Elf_Ehdr eh;
    399 	Elf_Phdr *ph = NULL;
    400 	const Elf_Phdr *base_ph;
    401 	const Elf_Phdr *last_ph;
    402 	u_long phsize;
    403 	Elf_Addr addr = *last;
    404 	struct proc *p;
    405 	bool use_topdown;
    406 
    407 	p = l->l_proc;
    408 
    409 	KASSERT(p->p_vmspace);
    410 	KASSERT(p->p_vmspace != proc0.p_vmspace);
    411 
    412 #ifdef __USE_TOPDOWN_VM
    413 	use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM;
    414 #else
    415 	use_topdown = false;
    416 #endif
    417 
    418 	/*
    419 	 * 1. open file
    420 	 * 2. read filehdr
    421 	 * 3. map text, data, and bss out of it using VM_*
    422 	 */
    423 	vp = epp->ep_interp;
    424 	if (vp == NULL) {
    425 		error = emul_find_interp(l, epp, path);
    426 		if (error != 0)
    427 			return error;
    428 		vp = epp->ep_interp;
    429 	}
    430 	/* We'll tidy this ourselves - otherwise we have locking issues */
    431 	epp->ep_interp = NULL;
    432 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    433 
    434 	/*
    435 	 * Similarly, if it's not marked as executable, or it's not a regular
    436 	 * file, we don't allow it to be used.
    437 	 */
    438 	if (vp->v_type != VREG) {
    439 		error = EACCES;
    440 		goto badunlock;
    441 	}
    442 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
    443 		goto badunlock;
    444 
    445 	/* get attributes */
    446 	if ((error = VOP_GETATTR(vp, &attr, l->l_cred)) != 0)
    447 		goto badunlock;
    448 
    449 	/*
    450 	 * Check mount point.  Though we're not trying to exec this binary,
    451 	 * we will be executing code from it, so if the mount point
    452 	 * disallows execution or set-id-ness, we punt or kill the set-id.
    453 	 */
    454 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
    455 		error = EACCES;
    456 		goto badunlock;
    457 	}
    458 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
    459 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
    460 
    461 	error = vn_marktext(vp);
    462 	if (error)
    463 		goto badunlock;
    464 
    465 	VOP_UNLOCK(vp);
    466 
    467 	if ((error = exec_read_from(l, vp, 0, &eh, sizeof(eh))) != 0)
    468 		goto bad;
    469 
    470 	if ((error = elf_check_header(&eh)) != 0)
    471 		goto bad;
    472 	if (eh.e_type != ET_DYN || eh.e_phnum == 0) {
    473 		DPRINTF("bad interpreter type %#x", eh.e_type);
    474 		error = ENOEXEC;
    475 		goto bad;
    476 	}
    477 
    478 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
    479 	ph = kmem_alloc(phsize, KM_SLEEP);
    480 
    481 	if ((error = exec_read_from(l, vp, eh.e_phoff, ph, phsize)) != 0)
    482 		goto bad;
    483 
    484 #ifdef ELF_INTERP_NON_RELOCATABLE
    485 	/*
    486 	 * Evil hack:  Only MIPS should be non-relocatable, and the
    487 	 * psections should have a high address (typically 0x5ffe0000).
    488 	 * If it's now relocatable, it should be linked at 0 and the
    489 	 * psections should have zeros in the upper part of the address.
    490 	 * Otherwise, force the load at the linked address.
    491 	 */
    492 	if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0)
    493 		*last = ELFDEFNNAME(NO_ADDR);
    494 #endif
    495 
    496 	/*
    497 	 * If no position to load the interpreter was set by a probe
    498 	 * function, pick the same address that a non-fixed mmap(0, ..)
    499 	 * would (i.e. something safely out of the way).
    500 	 */
    501 	if (*last == ELFDEFNNAME(NO_ADDR)) {
    502 		u_long limit = 0;
    503 		/*
    504 		 * Find the start and ending addresses of the psections to
    505 		 * be loaded.  This will give us the size.
    506 		 */
    507 		for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) {
    508 			if (ph[i].p_type == PT_LOAD) {
    509 				u_long psize = ph[i].p_vaddr + ph[i].p_memsz;
    510 				if (base_ph == NULL)
    511 					base_ph = &ph[i];
    512 				if (psize > limit)
    513 					limit = psize;
    514 			}
    515 		}
    516 
    517 		if (base_ph == NULL) {
    518 			DPRINTF("no interpreter loadable sections");
    519 			error = ENOEXEC;
    520 			goto bad;
    521 		}
    522 
    523 		/*
    524 		 * Now compute the size and load address.
    525 		 */
    526 		addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p,
    527 		    epp->ep_daddr,
    528 		    round_page(limit) - trunc_page(base_ph->p_vaddr),
    529 		    use_topdown);
    530 		addr += (Elf_Addr)pax_aslr_rtld_offset(epp, base_ph->p_align,
    531 		    use_topdown);
    532 	} else {
    533 		addr = *last; /* may be ELF_LINK_ADDR */
    534 	}
    535 
    536 	/*
    537 	 * Load all the necessary sections
    538 	 */
    539 	for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) {
    540 		switch (ph[i].p_type) {
    541 		case PT_LOAD: {
    542 			u_long size;
    543 			int flags;
    544 
    545 			if (base_ph == NULL) {
    546 				/*
    547 				 * First encountered psection is always the
    548 				 * base psection.  Make sure it's aligned
    549 				 * properly (align down for topdown and align
    550 				 * upwards for not topdown).
    551 				 */
    552 				base_ph = &ph[i];
    553 				flags = VMCMD_BASE;
    554 				if (addr == ELF_LINK_ADDR)
    555 					addr = ph[i].p_vaddr;
    556 				if (use_topdown)
    557 					addr = ELF_TRUNC(addr, ph[i].p_align);
    558 				else
    559 					addr = ELF_ROUND(addr, ph[i].p_align);
    560 			} else {
    561 				u_long limit = round_page(last_ph->p_vaddr
    562 				    + last_ph->p_memsz);
    563 				u_long base = trunc_page(ph[i].p_vaddr);
    564 
    565 				/*
    566 				 * If there is a gap in between the psections,
    567 				 * map it as inaccessible so nothing else
    568 				 * mmap'ed will be placed there.
    569 				 */
    570 				if (limit != base) {
    571 					NEW_VMCMD2(vcset, vmcmd_map_zero,
    572 					    base - limit,
    573 					    limit - base_ph->p_vaddr, NULLVP,
    574 					    0, VM_PROT_NONE, VMCMD_RELATIVE);
    575 				}
    576 
    577 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
    578 				flags = VMCMD_RELATIVE;
    579 			}
    580 			last_ph = &ph[i];
    581 			elf_load_psection(vcset, vp, &ph[i], &addr,
    582 			    &size, flags);
    583 			/*
    584 			 * If entry is within this psection then this
    585 			 * must contain the .text section.  *entryoff is
    586 			 * relative to the base psection.
    587 			 */
    588 			if (eh.e_entry >= ph[i].p_vaddr &&
    589 			    eh.e_entry < (ph[i].p_vaddr + size)) {
    590 				*entryoff = eh.e_entry - base_ph->p_vaddr;
    591 			}
    592 			addr += size;
    593 			break;
    594 		}
    595 
    596 		default:
    597 			break;
    598 		}
    599 	}
    600 
    601 	kmem_free(ph, phsize);
    602 	/*
    603 	 * This value is ignored if TOPDOWN.
    604 	 */
    605 	*last = addr;
    606 	vrele(vp);
    607 	return 0;
    608 
    609 badunlock:
    610 	VOP_UNLOCK(vp);
    611 
    612 bad:
    613 	if (ph != NULL)
    614 		kmem_free(ph, phsize);
    615 	vrele(vp);
    616 	return error;
    617 }
    618 
    619 /*
    620  * exec_elf_makecmds(): Prepare an Elf binary's exec package
    621  *
    622  * First, set of the various offsets/lengths in the exec package.
    623  *
    624  * Then, mark the text image busy (so it can be demand paged) or error
    625  * out if this is not possible.  Finally, set up vmcmds for the
    626  * text, data, bss, and stack segments.
    627  */
    628 int
    629 exec_elf_makecmds(struct lwp *l, struct exec_package *epp)
    630 {
    631 	Elf_Ehdr *eh = epp->ep_hdr;
    632 	Elf_Phdr *ph, *pp;
    633 	Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0;
    634 	int error, i;
    635 	char *interp = NULL;
    636 	u_long phsize;
    637 	struct elf_args *ap;
    638 	bool is_dyn = false;
    639 
    640 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) {
    641 		DPRINTF("small header %#x", epp->ep_hdrvalid);
    642 		return ENOEXEC;
    643 	}
    644 	if ((error = elf_check_header(eh)) != 0)
    645 		return error;
    646 
    647 	if (eh->e_type == ET_DYN)
    648 		/* PIE, and some libs have an entry point */
    649 		is_dyn = true;
    650 	else if (eh->e_type != ET_EXEC) {
    651 		DPRINTF("bad type %#x", eh->e_type);
    652 		return ENOEXEC;
    653 	}
    654 
    655 	if (eh->e_phnum == 0) {
    656 		DPRINTF("no program headers");
    657 		return ENOEXEC;
    658 	}
    659 
    660 	error = vn_marktext(epp->ep_vp);
    661 	if (error)
    662 		return error;
    663 
    664 	/*
    665 	 * Allocate space to hold all the program headers, and read them
    666 	 * from the file
    667 	 */
    668 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
    669 	ph = kmem_alloc(phsize, KM_SLEEP);
    670 
    671 	if ((error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize)) !=
    672 	    0)
    673 		goto bad;
    674 
    675 	epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
    676 	epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
    677 
    678 	for (i = 0; i < eh->e_phnum; i++) {
    679 		pp = &ph[i];
    680 		if (pp->p_type == PT_INTERP) {
    681 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) {
    682 				DPRINTF("bad interpreter namelen %#jx",
    683 				    (uintmax_t)pp->p_filesz);
    684 				error = ENOEXEC;
    685 				goto bad;
    686 			}
    687 			interp = PNBUF_GET();
    688 			if ((error = exec_read_from(l, epp->ep_vp,
    689 			    pp->p_offset, interp, pp->p_filesz)) != 0)
    690 				goto bad;
    691 			/* Ensure interp is NUL-terminated and of the expected length */
    692 			if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) {
    693 				DPRINTF("bad interpreter name");
    694 				error = ENOEXEC;
    695 				goto bad;
    696 			}
    697 			break;
    698 		}
    699 	}
    700 
    701 	/*
    702 	 * On the same architecture, we may be emulating different systems.
    703 	 * See which one will accept this executable.
    704 	 *
    705 	 * Probe functions would normally see if the interpreter (if any)
    706 	 * exists. Emulation packages may possibly replace the interpreter in
    707 	 * interp with a changed path (/emul/xxx/<path>).
    708 	 */
    709 	pos = ELFDEFNNAME(NO_ADDR);
    710 	if (epp->ep_esch->u.elf_probe_func) {
    711 		vaddr_t startp = (vaddr_t)pos;
    712 
    713 		error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp,
    714 							  &startp);
    715 		if (error)
    716 			goto bad;
    717 		pos = (Elf_Addr)startp;
    718 	}
    719 
    720 	if (is_dyn)
    721 		elf_placedynexec(epp, eh, ph);
    722 
    723 	/*
    724 	 * Load all the necessary sections
    725 	 */
    726 	for (i = 0; i < eh->e_phnum; i++) {
    727 		Elf_Addr addr = ELFDEFNNAME(NO_ADDR);
    728 		u_long size = 0;
    729 
    730 		switch (ph[i].p_type) {
    731 		case PT_LOAD:
    732 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
    733 			    &ph[i], &addr, &size, VMCMD_FIXED);
    734 
    735 			/*
    736 			 * Consider this as text segment, if it is executable.
    737 			 * If there is more than one text segment, pick the
    738 			 * largest.
    739 			 */
    740 			if (ph[i].p_flags & PF_X) {
    741 				if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) ||
    742 				    size > epp->ep_tsize) {
    743 					epp->ep_taddr = addr;
    744 					epp->ep_tsize = size;
    745 				}
    746 				end_text = addr + size;
    747 			} else {
    748 				epp->ep_daddr = addr;
    749 				epp->ep_dsize = size;
    750 			}
    751 			if (ph[i].p_offset == 0) {
    752 				computed_phdr = ph[i].p_vaddr + eh->e_phoff;
    753 			}
    754 			break;
    755 
    756 		case PT_SHLIB:
    757 			/* SCO has these sections. */
    758 		case PT_INTERP:
    759 			/* Already did this one. */
    760 		case PT_DYNAMIC:
    761 		case PT_NOTE:
    762 			break;
    763 		case PT_PHDR:
    764 			/* Note address of program headers (in text segment) */
    765 			phdr = ph[i].p_vaddr;
    766 			break;
    767 
    768 		default:
    769 			/*
    770 			 * Not fatal; we don't need to understand everything.
    771 			 */
    772 			break;
    773 		}
    774 	}
    775 
    776 	if (epp->ep_vmcmds.evs_used == 0) {
    777 		/* No VMCMD; there was no PT_LOAD section, or those
    778 		 * sections were empty */
    779 		DPRINTF("no vmcommands");
    780 		error = ENOEXEC;
    781 		goto bad;
    782 	}
    783 
    784 	if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) {
    785 		epp->ep_daddr = round_page(end_text);
    786 		epp->ep_dsize = 0;
    787 	}
    788 
    789 	/*
    790 	 * Check if we found a dynamically linked binary and arrange to load
    791 	 * its interpreter
    792 	 */
    793 	if (interp) {
    794 		u_int nused = epp->ep_vmcmds.evs_used;
    795 		u_long interp_offset = 0;
    796 
    797 		if ((error = elf_load_interp(l, epp, interp,
    798 		    &epp->ep_vmcmds, &interp_offset, &pos)) != 0) {
    799 			goto bad;
    800 		}
    801 		if (epp->ep_vmcmds.evs_used == nused) {
    802 			/* elf_load_interp() has not set up any new VMCMD */
    803 			DPRINTF("no vmcommands for interpreter");
    804 			error = ENOEXEC;
    805 			goto bad;
    806 		}
    807 
    808 		ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
    809 		ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr;
    810 		epp->ep_entryoffset = interp_offset;
    811 		epp->ep_entry = ap->arg_interp + interp_offset;
    812 		PNBUF_PUT(interp);
    813 		interp = NULL;
    814 	} else {
    815 		epp->ep_entry = eh->e_entry;
    816 		if (epp->ep_flags & EXEC_FORCEAUX) {
    817 			ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
    818 			ap->arg_interp = (vaddr_t)NULL;
    819 		} else
    820 			ap = NULL;
    821 	}
    822 
    823 	if (ap) {
    824 		ap->arg_phaddr = phdr ? phdr : computed_phdr;
    825 		ap->arg_phentsize = eh->e_phentsize;
    826 		ap->arg_phnum = eh->e_phnum;
    827 		ap->arg_entry = eh->e_entry;
    828 		epp->ep_emul_arg = ap;
    829 		epp->ep_emul_arg_free = elf_free_emul_arg;
    830 	}
    831 
    832 #ifdef ELF_MAP_PAGE_ZERO
    833 	/* Dell SVR4 maps page zero, yeuch! */
    834 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0,
    835 	    epp->ep_vp, 0, VM_PROT_READ);
    836 #endif
    837 
    838 	error = (*epp->ep_esch->es_setup_stack)(l, epp);
    839 	if (error)
    840 		goto bad;
    841 
    842 	kmem_free(ph, phsize);
    843 	return 0;
    844 
    845 bad:
    846 	if (interp)
    847 		PNBUF_PUT(interp);
    848 	exec_free_emul_arg(epp);
    849 	kmem_free(ph, phsize);
    850 	kill_vmcmds(&epp->ep_vmcmds);
    851 	return error;
    852 }
    853 
    854 int
    855 netbsd_elf_signature(struct lwp *l, struct exec_package *epp,
    856     Elf_Ehdr *eh)
    857 {
    858 	size_t i;
    859 	Elf_Phdr *ph;
    860 	size_t phsize;
    861 	char *nbuf;
    862 	int error;
    863 	int isnetbsd = 0;
    864 
    865 	epp->ep_pax_flags = 0;
    866 
    867 	if (eh->e_phnum > ELF_MAXPHNUM || eh->e_phnum == 0) {
    868 		DPRINTF("no signature %#x", eh->e_phnum);
    869 		return ENOEXEC;
    870 	}
    871 
    872 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
    873 	ph = kmem_alloc(phsize, KM_SLEEP);
    874 	error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize);
    875 	if (error)
    876 		goto out;
    877 
    878 	nbuf = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP);
    879 	for (i = 0; i < eh->e_phnum; i++) {
    880 		const char *nptr;
    881 		size_t nlen;
    882 
    883 		if (ph[i].p_type != PT_NOTE ||
    884 		    ph[i].p_filesz > ELF_MAXNOTESIZE)
    885 			continue;
    886 
    887 		nlen = ph[i].p_filesz;
    888 		error = exec_read_from(l, epp->ep_vp, ph[i].p_offset,
    889 				       nbuf, nlen);
    890 		if (error)
    891 			continue;
    892 
    893 		nptr = nbuf;
    894 		while (nlen > 0) {
    895 			const Elf_Nhdr *np;
    896 			const char *ndata, *ndesc;
    897 
    898 			/* note header */
    899 			np = (const Elf_Nhdr *)nptr;
    900 			if (nlen < sizeof(*np)) {
    901 				break;
    902 			}
    903 			nptr += sizeof(*np);
    904 			nlen -= sizeof(*np);
    905 
    906 			/* note name */
    907 			ndata = nptr;
    908 			if (nlen < roundup(np->n_namesz, 4)) {
    909 				break;
    910 			}
    911 			nptr += roundup(np->n_namesz, 4);
    912 			nlen -= roundup(np->n_namesz, 4);
    913 
    914 			/* note description */
    915 			ndesc = nptr;
    916 			if (nlen < roundup(np->n_descsz, 4)) {
    917 				break;
    918 			}
    919 			nptr += roundup(np->n_descsz, 4);
    920 			nlen -= roundup(np->n_descsz, 4);
    921 
    922 			isnetbsd |= netbsd_elf_note(epp, np, ndata, ndesc);
    923 		}
    924 	}
    925 	kmem_free(nbuf, ELF_MAXNOTESIZE);
    926 
    927 	error = isnetbsd ? 0 : ENOEXEC;
    928 #ifdef DEBUG_ELF
    929 	if (error)
    930 		DPRINTF("not netbsd");
    931 #endif
    932 out:
    933 	kmem_free(ph, phsize);
    934 	return error;
    935 }
    936 
    937 int
    938 netbsd_elf_note(struct exec_package *epp,
    939 		const Elf_Nhdr *np, const char *ndata, const char *ndesc)
    940 {
    941 	int isnetbsd = 0;
    942 
    943 #ifdef DIAGNOSTIC
    944 	const char *badnote;
    945 #define BADNOTE(n) badnote = (n)
    946 #else
    947 #define BADNOTE(n)
    948 #endif
    949 
    950 	switch (np->n_type) {
    951 	case ELF_NOTE_TYPE_NETBSD_TAG:
    952 		/* It is us */
    953 		if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ &&
    954 		    np->n_descsz == ELF_NOTE_NETBSD_DESCSZ &&
    955 		    memcmp(ndata, ELF_NOTE_NETBSD_NAME,
    956 		    ELF_NOTE_NETBSD_NAMESZ) == 0) {
    957 			memcpy(&epp->ep_osversion, ndesc,
    958 			    ELF_NOTE_NETBSD_DESCSZ);
    959 			isnetbsd = 1;
    960 			break;
    961 		}
    962 
    963 		/*
    964 		 * Ignore SuSE tags; SuSE's n_type is the same the
    965 		 * NetBSD one.
    966 		 */
    967 		if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ &&
    968 		    memcmp(ndata, ELF_NOTE_SUSE_NAME,
    969 		    ELF_NOTE_SUSE_NAMESZ) == 0)
    970 			break;
    971 		/*
    972 		 * Ignore old GCC
    973 		 */
    974 		if (np->n_namesz == ELF_NOTE_OGCC_NAMESZ &&
    975 		    memcmp(ndata, ELF_NOTE_OGCC_NAME,
    976 		    ELF_NOTE_OGCC_NAMESZ) == 0)
    977 			break;
    978 		BADNOTE("NetBSD tag");
    979 		goto bad;
    980 
    981 	case ELF_NOTE_TYPE_PAX_TAG:
    982 		if (np->n_namesz == ELF_NOTE_PAX_NAMESZ &&
    983 		    np->n_descsz == ELF_NOTE_PAX_DESCSZ &&
    984 		    memcmp(ndata, ELF_NOTE_PAX_NAME,
    985 		    ELF_NOTE_PAX_NAMESZ) == 0) {
    986 			uint32_t flags;
    987 			memcpy(&flags, ndesc, sizeof(flags));
    988 			/* Convert the flags and insert them into
    989 			 * the exec package. */
    990 			pax_setup_elf_flags(epp, flags);
    991 			break;
    992 		}
    993 		BADNOTE("PaX tag");
    994 		goto bad;
    995 
    996 	case ELF_NOTE_TYPE_MARCH_TAG:
    997 		/* Copy the machine arch into the package. */
    998 		if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ
    999 		    && memcmp(ndata, ELF_NOTE_MARCH_NAME,
   1000 			    ELF_NOTE_MARCH_NAMESZ) == 0) {
   1001 			/* Do not truncate the buffer */
   1002 			if (np->n_descsz > sizeof(epp->ep_machine_arch)) {
   1003 				BADNOTE("description size limit");
   1004 				goto bad;
   1005 			}
   1006 			/*
   1007 			 * Ensure ndesc is NUL-terminated and of the
   1008 			 * expected length.
   1009 			 */
   1010 			if (strnlen(ndesc, np->n_descsz) + 1 !=
   1011 			    np->n_descsz) {
   1012 				BADNOTE("description size");
   1013 				goto bad;
   1014 			}
   1015 			strlcpy(epp->ep_machine_arch, ndesc,
   1016 			    sizeof(epp->ep_machine_arch));
   1017 			break;
   1018 		}
   1019 		BADNOTE("march tag");
   1020 		goto bad;
   1021 
   1022 	case ELF_NOTE_TYPE_MCMODEL_TAG:
   1023 		/* arch specific check for code model */
   1024 #ifdef ELF_MD_MCMODEL_CHECK
   1025 		if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ
   1026 		    && memcmp(ndata, ELF_NOTE_MCMODEL_NAME,
   1027 			    ELF_NOTE_MCMODEL_NAMESZ) == 0) {
   1028 			ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz);
   1029 			break;
   1030 		}
   1031 		BADNOTE("mcmodel tag");
   1032 		goto bad;
   1033 #endif
   1034 		break;
   1035 
   1036 	case ELF_NOTE_TYPE_SUSE_VERSION_TAG:
   1037 		break;
   1038 
   1039 	case ELF_NOTE_TYPE_GO_BUILDID_TAG:
   1040 		break;
   1041 
   1042 	default:
   1043 		BADNOTE("unknown tag");
   1044 bad:
   1045 #ifdef DIAGNOSTIC
   1046 		/* Ignore GNU tags */
   1047 		if (np->n_namesz == ELF_NOTE_GNU_NAMESZ &&
   1048 		    memcmp(ndata, ELF_NOTE_GNU_NAME,
   1049 		    ELF_NOTE_GNU_NAMESZ) == 0)
   1050 		    break;
   1051 
   1052 		int ns = (int)np->n_namesz;
   1053 		printf("%s: Unknown elf note type %d (%s): "
   1054 		    "[namesz=%d, descsz=%d name=%-*.*s]\n",
   1055 		    epp->ep_kname, np->n_type, badnote, np->n_namesz,
   1056 		    np->n_descsz, ns, ns, ndata);
   1057 #endif
   1058 		break;
   1059 	}
   1060 
   1061 	return isnetbsd;
   1062 }
   1063 
   1064 int
   1065 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp,
   1066     vaddr_t *pos)
   1067 {
   1068 	int error;
   1069 
   1070 	if ((error = netbsd_elf_signature(l, epp, eh)) != 0)
   1071 		return error;
   1072 #ifdef ELF_MD_PROBE_FUNC
   1073 	if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0)
   1074 		return error;
   1075 #elif defined(ELF_INTERP_NON_RELOCATABLE)
   1076 	*pos = ELF_LINK_ADDR;
   1077 #endif
   1078 	epp->ep_flags |= EXEC_FORCEAUX;
   1079 	return 0;
   1080 }
   1081 
   1082 void
   1083 elf_free_emul_arg(void *arg)
   1084 {
   1085 	struct elf_args *ap = arg;
   1086 	KASSERT(ap != NULL);
   1087 	kmem_free(ap, sizeof(*ap));
   1088 }
   1089