Home | History | Annotate | Line # | Download | only in kern
exec_elf.c revision 1.93.2.1
      1 /*	$NetBSD: exec_elf.c,v 1.93.2.1 2018/03/06 10:37:41 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 1.93.2.1 2018/03/06 10:37:41 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 		execname = a;
    230 		a->a_type = AT_SUN_EXECNAME;
    231 		a++;
    232 
    233 		exec_free_emul_arg(pack);
    234 	}
    235 
    236 	a->a_type = AT_NULL;
    237 	a->a_v = 0;
    238 	a++;
    239 
    240 	vlen = (a - ai) * sizeof(ai[0]);
    241 
    242 	KASSERT(vlen <= sizeof(ai));
    243 
    244 	char *path = l->l_proc->p_path;
    245 	execname->a_v = (uintptr_t)(*stackp + vlen);
    246 	len = strlen(path) + 1;
    247 	if ((error = copyout(path, (*stackp + vlen), len)) != 0)
    248 		return error;
    249 	len = ALIGN(len);
    250 
    251 	if ((error = copyout(ai, *stackp, vlen)) != 0)
    252 		return error;
    253 	*stackp += vlen + len;
    254 
    255 	return 0;
    256 }
    257 
    258 /*
    259  * elf_check_header():
    260  *
    261  * Check header for validity; return 0 if ok, ENOEXEC if error
    262  */
    263 int
    264 elf_check_header(Elf_Ehdr *eh)
    265 {
    266 
    267 	if (memcmp(eh->e_ident, ELFMAG, SELFMAG) != 0 ||
    268 	    eh->e_ident[EI_CLASS] != ELFCLASS) {
    269 		DPRINTF("bad magic e_ident[EI_MAG0,EI_MAG3] %#x%x%x%x, "
    270 		    "e_ident[EI_CLASS] %#x", eh->e_ident[EI_MAG0],
    271 		    eh->e_ident[EI_MAG1], eh->e_ident[EI_MAG2],
    272 		    eh->e_ident[EI_MAG3], eh->e_ident[EI_CLASS]);
    273 		return ENOEXEC;
    274 	}
    275 
    276 	switch (eh->e_machine) {
    277 
    278 	ELFDEFNNAME(MACHDEP_ID_CASES)
    279 
    280 	default:
    281 		DPRINTF("bad machine %#x", eh->e_machine);
    282 		return ENOEXEC;
    283 	}
    284 
    285 	if (ELF_EHDR_FLAGS_OK(eh) == 0) {
    286 		DPRINTF("bad flags %#x", eh->e_flags);
    287 		return ENOEXEC;
    288 	}
    289 
    290 	if (eh->e_shnum > ELF_MAXSHNUM || eh->e_phnum > ELF_MAXPHNUM) {
    291 		DPRINTF("bad shnum/phnum %#x/%#x", eh->e_shnum, eh->e_phnum);
    292 		return ENOEXEC;
    293 	}
    294 
    295 	return 0;
    296 }
    297 
    298 /*
    299  * elf_load_psection():
    300  *
    301  * Load a psection at the appropriate address
    302  */
    303 static void
    304 elf_load_psection(struct exec_vmcmd_set *vcset, struct vnode *vp,
    305     const Elf_Phdr *ph, Elf_Addr *addr, u_long *size, int flags)
    306 {
    307 	u_long msize, psize, rm, rf;
    308 	long diff, offset;
    309 	int vmprot = 0;
    310 
    311 	/*
    312 	 * If the user specified an address, then we load there.
    313 	 */
    314 	if (*addr == ELFDEFNNAME(NO_ADDR))
    315 		*addr = ph->p_vaddr;
    316 
    317 	if (ph->p_align > 1) {
    318 		/*
    319 		 * Make sure we are virtually aligned as we are supposed to be.
    320 		 */
    321 		diff = ph->p_vaddr - ELF_TRUNC(ph->p_vaddr, ph->p_align);
    322 		KASSERT(*addr - diff == ELF_TRUNC(*addr, ph->p_align));
    323 		/*
    324 		 * But make sure to not map any pages before the start of the
    325 		 * psection by limiting the difference to within a page.
    326 		 */
    327 		diff &= PAGE_MASK;
    328 	} else
    329 		diff = 0;
    330 
    331 	vmprot |= (ph->p_flags & PF_R) ? VM_PROT_READ : 0;
    332 	vmprot |= (ph->p_flags & PF_W) ? VM_PROT_WRITE : 0;
    333 	vmprot |= (ph->p_flags & PF_X) ? VM_PROT_EXECUTE : 0;
    334 
    335 	/*
    336 	 * Adjust everything so it all starts on a page boundary.
    337 	 */
    338 	*addr -= diff;
    339 	offset = ph->p_offset - diff;
    340 	*size = ph->p_filesz + diff;
    341 	msize = ph->p_memsz + diff;
    342 
    343 	if (ph->p_align >= PAGE_SIZE) {
    344 		if ((ph->p_flags & PF_W) != 0) {
    345 			/*
    346 			 * Because the pagedvn pager can't handle zero fill
    347 			 * of the last data page if it's not page aligned we
    348 			 * map the last page readvn.
    349 			 */
    350 			psize = trunc_page(*size);
    351 		} else {
    352 			psize = round_page(*size);
    353 		}
    354 	} else {
    355 		psize = *size;
    356 	}
    357 
    358 	if (psize > 0) {
    359 		NEW_VMCMD2(vcset, ph->p_align < PAGE_SIZE ?
    360 		    vmcmd_map_readvn : vmcmd_map_pagedvn, psize, *addr, vp,
    361 		    offset, vmprot, flags);
    362 		flags &= VMCMD_RELATIVE;
    363 	}
    364 	if (psize < *size) {
    365 		NEW_VMCMD2(vcset, vmcmd_map_readvn, *size - psize,
    366 		    *addr + psize, vp, offset + psize, vmprot, flags);
    367 	}
    368 
    369 	/*
    370 	 * Check if we need to extend the size of the segment (does
    371 	 * bss extend page the next page boundary)?
    372 	 */
    373 	rm = round_page(*addr + msize);
    374 	rf = round_page(*addr + *size);
    375 
    376 	if (rm != rf) {
    377 		NEW_VMCMD2(vcset, vmcmd_map_zero, rm - rf, rf, NULLVP,
    378 		    0, vmprot, flags & VMCMD_RELATIVE);
    379 		*size = msize;
    380 	}
    381 }
    382 
    383 /*
    384  * elf_load_interp():
    385  *
    386  * Load an interpreter pointed to by path.
    387  */
    388 static int
    389 elf_load_interp(struct lwp *l, struct exec_package *epp, char *path,
    390     struct exec_vmcmd_set *vcset, u_long *entryoff, Elf_Addr *last)
    391 {
    392 	int error, i;
    393 	struct vnode *vp;
    394 	struct vattr attr;
    395 	Elf_Ehdr eh;
    396 	Elf_Phdr *ph = NULL;
    397 	const Elf_Phdr *base_ph;
    398 	const Elf_Phdr *last_ph;
    399 	u_long phsize;
    400 	Elf_Addr addr = *last;
    401 	struct proc *p;
    402 	bool use_topdown;
    403 
    404 	p = l->l_proc;
    405 
    406 	KASSERT(p->p_vmspace);
    407 	KASSERT(p->p_vmspace != proc0.p_vmspace);
    408 
    409 #ifdef __USE_TOPDOWN_VM
    410 	use_topdown = epp->ep_flags & EXEC_TOPDOWN_VM;
    411 #else
    412 	use_topdown = false;
    413 #endif
    414 
    415 	/*
    416 	 * 1. open file
    417 	 * 2. read filehdr
    418 	 * 3. map text, data, and bss out of it using VM_*
    419 	 */
    420 	vp = epp->ep_interp;
    421 	if (vp == NULL) {
    422 		error = emul_find_interp(l, epp, path);
    423 		if (error != 0)
    424 			return error;
    425 		vp = epp->ep_interp;
    426 	}
    427 	/* We'll tidy this ourselves - otherwise we have locking issues */
    428 	epp->ep_interp = NULL;
    429 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    430 
    431 	/*
    432 	 * Similarly, if it's not marked as executable, or it's not a regular
    433 	 * file, we don't allow it to be used.
    434 	 */
    435 	if (vp->v_type != VREG) {
    436 		error = EACCES;
    437 		goto badunlock;
    438 	}
    439 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
    440 		goto badunlock;
    441 
    442 	/* get attributes */
    443 	if ((error = VOP_GETATTR(vp, &attr, l->l_cred)) != 0)
    444 		goto badunlock;
    445 
    446 	/*
    447 	 * Check mount point.  Though we're not trying to exec this binary,
    448 	 * we will be executing code from it, so if the mount point
    449 	 * disallows execution or set-id-ness, we punt or kill the set-id.
    450 	 */
    451 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
    452 		error = EACCES;
    453 		goto badunlock;
    454 	}
    455 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
    456 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
    457 
    458 	error = vn_marktext(vp);
    459 	if (error)
    460 		goto badunlock;
    461 
    462 	VOP_UNLOCK(vp);
    463 
    464 	if ((error = exec_read_from(l, vp, 0, &eh, sizeof(eh))) != 0)
    465 		goto bad;
    466 
    467 	if ((error = elf_check_header(&eh)) != 0)
    468 		goto bad;
    469 	if (eh.e_type != ET_DYN || eh.e_phnum == 0) {
    470 		DPRINTF("bad interpreter type %#x", eh.e_type);
    471 		error = ENOEXEC;
    472 		goto bad;
    473 	}
    474 
    475 	phsize = eh.e_phnum * sizeof(Elf_Phdr);
    476 	ph = kmem_alloc(phsize, KM_SLEEP);
    477 
    478 	if ((error = exec_read_from(l, vp, eh.e_phoff, ph, phsize)) != 0)
    479 		goto bad;
    480 
    481 #ifdef ELF_INTERP_NON_RELOCATABLE
    482 	/*
    483 	 * Evil hack:  Only MIPS should be non-relocatable, and the
    484 	 * psections should have a high address (typically 0x5ffe0000).
    485 	 * If it's now relocatable, it should be linked at 0 and the
    486 	 * psections should have zeros in the upper part of the address.
    487 	 * Otherwise, force the load at the linked address.
    488 	 */
    489 	if (*last == ELF_LINK_ADDR && (ph->p_vaddr & 0xffff0000) == 0)
    490 		*last = ELFDEFNNAME(NO_ADDR);
    491 #endif
    492 
    493 	/*
    494 	 * If no position to load the interpreter was set by a probe
    495 	 * function, pick the same address that a non-fixed mmap(0, ..)
    496 	 * would (i.e. something safely out of the way).
    497 	 */
    498 	if (*last == ELFDEFNNAME(NO_ADDR)) {
    499 		u_long limit = 0;
    500 		/*
    501 		 * Find the start and ending addresses of the psections to
    502 		 * be loaded.  This will give us the size.
    503 		 */
    504 		for (i = 0, base_ph = NULL; i < eh.e_phnum; i++) {
    505 			if (ph[i].p_type == PT_LOAD) {
    506 				u_long psize = ph[i].p_vaddr + ph[i].p_memsz;
    507 				if (base_ph == NULL)
    508 					base_ph = &ph[i];
    509 				if (psize > limit)
    510 					limit = psize;
    511 			}
    512 		}
    513 
    514 		if (base_ph == NULL) {
    515 			DPRINTF("no interpreter loadable sections");
    516 			error = ENOEXEC;
    517 			goto bad;
    518 		}
    519 
    520 		/*
    521 		 * Now compute the size and load address.
    522 		 */
    523 		addr = (*epp->ep_esch->es_emul->e_vm_default_addr)(p,
    524 		    epp->ep_daddr,
    525 		    round_page(limit) - trunc_page(base_ph->p_vaddr),
    526 		    use_topdown);
    527 		addr += (Elf_Addr)pax_aslr_rtld_offset(epp, base_ph->p_align,
    528 		    use_topdown);
    529 	} else {
    530 		addr = *last; /* may be ELF_LINK_ADDR */
    531 	}
    532 
    533 	/*
    534 	 * Load all the necessary sections
    535 	 */
    536 	for (i = 0, base_ph = NULL, last_ph = NULL; i < eh.e_phnum; i++) {
    537 		switch (ph[i].p_type) {
    538 		case PT_LOAD: {
    539 			u_long size;
    540 			int flags;
    541 
    542 			if (base_ph == NULL) {
    543 				/*
    544 				 * First encountered psection is always the
    545 				 * base psection.  Make sure it's aligned
    546 				 * properly (align down for topdown and align
    547 				 * upwards for not topdown).
    548 				 */
    549 				base_ph = &ph[i];
    550 				flags = VMCMD_BASE;
    551 				if (addr == ELF_LINK_ADDR)
    552 					addr = ph[i].p_vaddr;
    553 				if (use_topdown)
    554 					addr = ELF_TRUNC(addr, ph[i].p_align);
    555 				else
    556 					addr = ELF_ROUND(addr, ph[i].p_align);
    557 			} else {
    558 				u_long limit = round_page(last_ph->p_vaddr
    559 				    + last_ph->p_memsz);
    560 				u_long base = trunc_page(ph[i].p_vaddr);
    561 
    562 				/*
    563 				 * If there is a gap in between the psections,
    564 				 * map it as inaccessible so nothing else
    565 				 * mmap'ed will be placed there.
    566 				 */
    567 				if (limit != base) {
    568 					NEW_VMCMD2(vcset, vmcmd_map_zero,
    569 					    base - limit,
    570 					    limit - base_ph->p_vaddr, NULLVP,
    571 					    0, VM_PROT_NONE, VMCMD_RELATIVE);
    572 				}
    573 
    574 				addr = ph[i].p_vaddr - base_ph->p_vaddr;
    575 				flags = VMCMD_RELATIVE;
    576 			}
    577 			last_ph = &ph[i];
    578 			elf_load_psection(vcset, vp, &ph[i], &addr,
    579 			    &size, flags);
    580 			/*
    581 			 * If entry is within this psection then this
    582 			 * must contain the .text section.  *entryoff is
    583 			 * relative to the base psection.
    584 			 */
    585 			if (eh.e_entry >= ph[i].p_vaddr &&
    586 			    eh.e_entry < (ph[i].p_vaddr + size)) {
    587 				*entryoff = eh.e_entry - base_ph->p_vaddr;
    588 			}
    589 			addr += size;
    590 			break;
    591 		}
    592 
    593 		default:
    594 			break;
    595 		}
    596 	}
    597 
    598 	kmem_free(ph, phsize);
    599 	/*
    600 	 * This value is ignored if TOPDOWN.
    601 	 */
    602 	*last = addr;
    603 	vrele(vp);
    604 	return 0;
    605 
    606 badunlock:
    607 	VOP_UNLOCK(vp);
    608 
    609 bad:
    610 	if (ph != NULL)
    611 		kmem_free(ph, phsize);
    612 	vrele(vp);
    613 	return error;
    614 }
    615 
    616 /*
    617  * exec_elf_makecmds(): Prepare an Elf binary's exec package
    618  *
    619  * First, set of the various offsets/lengths in the exec package.
    620  *
    621  * Then, mark the text image busy (so it can be demand paged) or error
    622  * out if this is not possible.  Finally, set up vmcmds for the
    623  * text, data, bss, and stack segments.
    624  */
    625 int
    626 exec_elf_makecmds(struct lwp *l, struct exec_package *epp)
    627 {
    628 	Elf_Ehdr *eh = epp->ep_hdr;
    629 	Elf_Phdr *ph, *pp;
    630 	Elf_Addr phdr = 0, computed_phdr = 0, pos = 0, end_text = 0;
    631 	int error, i;
    632 	char *interp = NULL;
    633 	u_long phsize;
    634 	struct elf_args *ap;
    635 	bool is_dyn = false;
    636 
    637 	if (epp->ep_hdrvalid < sizeof(Elf_Ehdr)) {
    638 		DPRINTF("small header %#x", epp->ep_hdrvalid);
    639 		return ENOEXEC;
    640 	}
    641 	if ((error = elf_check_header(eh)) != 0)
    642 		return error;
    643 
    644 	if (eh->e_type == ET_DYN)
    645 		/* PIE, and some libs have an entry point */
    646 		is_dyn = true;
    647 	else if (eh->e_type != ET_EXEC) {
    648 		DPRINTF("bad type %#x", eh->e_type);
    649 		return ENOEXEC;
    650 	}
    651 
    652 	if (eh->e_phnum == 0) {
    653 		DPRINTF("no program headers");
    654 		return ENOEXEC;
    655 	}
    656 
    657 	error = vn_marktext(epp->ep_vp);
    658 	if (error)
    659 		return error;
    660 
    661 	/*
    662 	 * Allocate space to hold all the program headers, and read them
    663 	 * from the file
    664 	 */
    665 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
    666 	ph = kmem_alloc(phsize, KM_SLEEP);
    667 
    668 	if ((error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize)) !=
    669 	    0)
    670 		goto bad;
    671 
    672 	epp->ep_taddr = epp->ep_tsize = ELFDEFNNAME(NO_ADDR);
    673 	epp->ep_daddr = epp->ep_dsize = ELFDEFNNAME(NO_ADDR);
    674 
    675 	for (i = 0; i < eh->e_phnum; i++) {
    676 		pp = &ph[i];
    677 		if (pp->p_type == PT_INTERP) {
    678 			if (pp->p_filesz < 2 || pp->p_filesz > MAXPATHLEN) {
    679 				DPRINTF("bad interpreter namelen %#jx",
    680 				    (uintmax_t)pp->p_filesz);
    681 				error = ENOEXEC;
    682 				goto bad;
    683 			}
    684 			interp = PNBUF_GET();
    685 			if ((error = exec_read_from(l, epp->ep_vp,
    686 			    pp->p_offset, interp, pp->p_filesz)) != 0)
    687 				goto bad;
    688 			/* Ensure interp is NUL-terminated and of the expected length */
    689 			if (strnlen(interp, pp->p_filesz) != pp->p_filesz - 1) {
    690 				DPRINTF("bad interpreter name");
    691 				error = ENOEXEC;
    692 				goto bad;
    693 			}
    694 			break;
    695 		}
    696 	}
    697 
    698 	/*
    699 	 * On the same architecture, we may be emulating different systems.
    700 	 * See which one will accept this executable.
    701 	 *
    702 	 * Probe functions would normally see if the interpreter (if any)
    703 	 * exists. Emulation packages may possibly replace the interpreter in
    704 	 * interp with a changed path (/emul/xxx/<path>).
    705 	 */
    706 	pos = ELFDEFNNAME(NO_ADDR);
    707 	if (epp->ep_esch->u.elf_probe_func) {
    708 		vaddr_t startp = (vaddr_t)pos;
    709 
    710 		error = (*epp->ep_esch->u.elf_probe_func)(l, epp, eh, interp,
    711 							  &startp);
    712 		if (error)
    713 			goto bad;
    714 		pos = (Elf_Addr)startp;
    715 	}
    716 
    717 	if (is_dyn)
    718 		elf_placedynexec(epp, eh, ph);
    719 
    720 	/*
    721 	 * Load all the necessary sections
    722 	 */
    723 	for (i = 0; i < eh->e_phnum; i++) {
    724 		Elf_Addr addr = ELFDEFNNAME(NO_ADDR);
    725 		u_long size = 0;
    726 
    727 		switch (ph[i].p_type) {
    728 		case PT_LOAD:
    729 			elf_load_psection(&epp->ep_vmcmds, epp->ep_vp,
    730 			    &ph[i], &addr, &size, VMCMD_FIXED);
    731 
    732 			/*
    733 			 * Consider this as text segment, if it is executable.
    734 			 * If there is more than one text segment, pick the
    735 			 * largest.
    736 			 */
    737 			if (ph[i].p_flags & PF_X) {
    738 				if (epp->ep_taddr == ELFDEFNNAME(NO_ADDR) ||
    739 				    size > epp->ep_tsize) {
    740 					epp->ep_taddr = addr;
    741 					epp->ep_tsize = size;
    742 				}
    743 				end_text = addr + size;
    744 			} else {
    745 				epp->ep_daddr = addr;
    746 				epp->ep_dsize = size;
    747 			}
    748 			if (ph[i].p_offset == 0) {
    749 				computed_phdr = ph[i].p_vaddr + eh->e_phoff;
    750 			}
    751 			break;
    752 
    753 		case PT_SHLIB:
    754 			/* SCO has these sections. */
    755 		case PT_INTERP:
    756 			/* Already did this one. */
    757 		case PT_DYNAMIC:
    758 		case PT_NOTE:
    759 			break;
    760 		case PT_PHDR:
    761 			/* Note address of program headers (in text segment) */
    762 			phdr = ph[i].p_vaddr;
    763 			break;
    764 
    765 		default:
    766 			/*
    767 			 * Not fatal; we don't need to understand everything.
    768 			 */
    769 			break;
    770 		}
    771 	}
    772 
    773 	if (epp->ep_vmcmds.evs_used == 0) {
    774 		/* No VMCMD; there was no PT_LOAD section, or those
    775 		 * sections were empty */
    776 		DPRINTF("no vmcommands");
    777 		error = ENOEXEC;
    778 		goto bad;
    779 	}
    780 
    781 	if (epp->ep_daddr == ELFDEFNNAME(NO_ADDR)) {
    782 		epp->ep_daddr = round_page(end_text);
    783 		epp->ep_dsize = 0;
    784 	}
    785 
    786 	/*
    787 	 * Check if we found a dynamically linked binary and arrange to load
    788 	 * its interpreter
    789 	 */
    790 	if (interp) {
    791 		u_int nused = epp->ep_vmcmds.evs_used;
    792 		u_long interp_offset = 0;
    793 
    794 		if ((error = elf_load_interp(l, epp, interp,
    795 		    &epp->ep_vmcmds, &interp_offset, &pos)) != 0) {
    796 			goto bad;
    797 		}
    798 		if (epp->ep_vmcmds.evs_used == nused) {
    799 			/* elf_load_interp() has not set up any new VMCMD */
    800 			DPRINTF("no vmcommands for interpreter");
    801 			error = ENOEXEC;
    802 			goto bad;
    803 		}
    804 
    805 		ap = kmem_alloc(sizeof(*ap), KM_SLEEP);
    806 		ap->arg_interp = epp->ep_vmcmds.evs_cmds[nused].ev_addr;
    807 		epp->ep_entryoffset = interp_offset;
    808 		epp->ep_entry = ap->arg_interp + interp_offset;
    809 		PNBUF_PUT(interp);
    810 		interp = NULL;
    811 	} else {
    812 		epp->ep_entry = eh->e_entry;
    813 		if (epp->ep_flags & EXEC_FORCEAUX) {
    814 			ap = kmem_zalloc(sizeof(*ap), KM_SLEEP);
    815 			ap->arg_interp = (vaddr_t)NULL;
    816 		} else {
    817 			ap = NULL;
    818 		}
    819 	}
    820 
    821 	if (ap) {
    822 		ap->arg_phaddr = phdr ? phdr : computed_phdr;
    823 		ap->arg_phentsize = eh->e_phentsize;
    824 		ap->arg_phnum = eh->e_phnum;
    825 		ap->arg_entry = eh->e_entry;
    826 		epp->ep_emul_arg = ap;
    827 		epp->ep_emul_arg_free = elf_free_emul_arg;
    828 	}
    829 
    830 #ifdef ELF_MAP_PAGE_ZERO
    831 	/* Dell SVR4 maps page zero, yeuch! */
    832 	NEW_VMCMD(&epp->ep_vmcmds, vmcmd_map_readvn, PAGE_SIZE, 0,
    833 	    epp->ep_vp, 0, VM_PROT_READ);
    834 #endif
    835 
    836 	error = (*epp->ep_esch->es_setup_stack)(l, epp);
    837 	if (error)
    838 		goto bad;
    839 
    840 	kmem_free(ph, phsize);
    841 	return 0;
    842 
    843 bad:
    844 	if (interp)
    845 		PNBUF_PUT(interp);
    846 	exec_free_emul_arg(epp);
    847 	kmem_free(ph, phsize);
    848 	kill_vmcmds(&epp->ep_vmcmds);
    849 	return error;
    850 }
    851 
    852 int
    853 netbsd_elf_signature(struct lwp *l, struct exec_package *epp,
    854     Elf_Ehdr *eh)
    855 {
    856 	size_t i;
    857 	Elf_Phdr *ph;
    858 	size_t phsize;
    859 	char *nbuf;
    860 	int error;
    861 	int isnetbsd = 0;
    862 
    863 	epp->ep_pax_flags = 0;
    864 
    865 	if (eh->e_phnum > ELF_MAXPHNUM || eh->e_phnum == 0) {
    866 		DPRINTF("no signature %#x", eh->e_phnum);
    867 		return ENOEXEC;
    868 	}
    869 
    870 	phsize = eh->e_phnum * sizeof(Elf_Phdr);
    871 	ph = kmem_alloc(phsize, KM_SLEEP);
    872 	error = exec_read_from(l, epp->ep_vp, eh->e_phoff, ph, phsize);
    873 	if (error)
    874 		goto out;
    875 
    876 	nbuf = kmem_alloc(ELF_MAXNOTESIZE, KM_SLEEP);
    877 	for (i = 0; i < eh->e_phnum; i++) {
    878 		const char *nptr;
    879 		size_t nlen;
    880 
    881 		if (ph[i].p_type != PT_NOTE ||
    882 		    ph[i].p_filesz > ELF_MAXNOTESIZE)
    883 			continue;
    884 
    885 		nlen = ph[i].p_filesz;
    886 		error = exec_read_from(l, epp->ep_vp, ph[i].p_offset,
    887 				       nbuf, nlen);
    888 		if (error)
    889 			continue;
    890 
    891 		nptr = nbuf;
    892 		while (nlen > 0) {
    893 			const Elf_Nhdr *np;
    894 			const char *ndata, *ndesc;
    895 
    896 			/* note header */
    897 			np = (const Elf_Nhdr *)nptr;
    898 			if (nlen < sizeof(*np)) {
    899 				break;
    900 			}
    901 			nptr += sizeof(*np);
    902 			nlen -= sizeof(*np);
    903 
    904 			/* note name */
    905 			ndata = nptr;
    906 			if (nlen < roundup(np->n_namesz, 4)) {
    907 				break;
    908 			}
    909 			nptr += roundup(np->n_namesz, 4);
    910 			nlen -= roundup(np->n_namesz, 4);
    911 
    912 			/* note description */
    913 			ndesc = nptr;
    914 			if (nlen < roundup(np->n_descsz, 4)) {
    915 				break;
    916 			}
    917 			nptr += roundup(np->n_descsz, 4);
    918 			nlen -= roundup(np->n_descsz, 4);
    919 
    920 			isnetbsd |= netbsd_elf_note(epp, np, ndata, ndesc);
    921 		}
    922 	}
    923 	kmem_free(nbuf, ELF_MAXNOTESIZE);
    924 
    925 	error = isnetbsd ? 0 : ENOEXEC;
    926 #ifdef DEBUG_ELF
    927 	if (error)
    928 		DPRINTF("not netbsd");
    929 #endif
    930 out:
    931 	kmem_free(ph, phsize);
    932 	return error;
    933 }
    934 
    935 int
    936 netbsd_elf_note(struct exec_package *epp,
    937 		const Elf_Nhdr *np, const char *ndata, const char *ndesc)
    938 {
    939 	int isnetbsd = 0;
    940 
    941 #ifdef DIAGNOSTIC
    942 	const char *badnote;
    943 #define BADNOTE(n) badnote = (n)
    944 #else
    945 #define BADNOTE(n)
    946 #endif
    947 
    948 	switch (np->n_type) {
    949 	case ELF_NOTE_TYPE_NETBSD_TAG:
    950 		/* It is us */
    951 		if (np->n_namesz == ELF_NOTE_NETBSD_NAMESZ &&
    952 		    np->n_descsz == ELF_NOTE_NETBSD_DESCSZ &&
    953 		    memcmp(ndata, ELF_NOTE_NETBSD_NAME,
    954 		    ELF_NOTE_NETBSD_NAMESZ) == 0) {
    955 			memcpy(&epp->ep_osversion, ndesc,
    956 			    ELF_NOTE_NETBSD_DESCSZ);
    957 			isnetbsd = 1;
    958 			break;
    959 		}
    960 
    961 		/*
    962 		 * Ignore SuSE tags; SuSE's n_type is the same the
    963 		 * NetBSD one.
    964 		 */
    965 		if (np->n_namesz == ELF_NOTE_SUSE_NAMESZ &&
    966 		    memcmp(ndata, ELF_NOTE_SUSE_NAME,
    967 		    ELF_NOTE_SUSE_NAMESZ) == 0)
    968 			break;
    969 		/*
    970 		 * Ignore old GCC
    971 		 */
    972 		if (np->n_namesz == ELF_NOTE_OGCC_NAMESZ &&
    973 		    memcmp(ndata, ELF_NOTE_OGCC_NAME,
    974 		    ELF_NOTE_OGCC_NAMESZ) == 0)
    975 			break;
    976 		BADNOTE("NetBSD tag");
    977 		goto bad;
    978 
    979 	case ELF_NOTE_TYPE_PAX_TAG:
    980 		if (np->n_namesz == ELF_NOTE_PAX_NAMESZ &&
    981 		    np->n_descsz == ELF_NOTE_PAX_DESCSZ &&
    982 		    memcmp(ndata, ELF_NOTE_PAX_NAME,
    983 		    ELF_NOTE_PAX_NAMESZ) == 0) {
    984 			uint32_t flags;
    985 			memcpy(&flags, ndesc, sizeof(flags));
    986 			/* Convert the flags and insert them into
    987 			 * the exec package. */
    988 			pax_setup_elf_flags(epp, flags);
    989 			break;
    990 		}
    991 		BADNOTE("PaX tag");
    992 		goto bad;
    993 
    994 	case ELF_NOTE_TYPE_MARCH_TAG:
    995 		/* Copy the machine arch into the package. */
    996 		if (np->n_namesz == ELF_NOTE_MARCH_NAMESZ
    997 		    && memcmp(ndata, ELF_NOTE_MARCH_NAME,
    998 			    ELF_NOTE_MARCH_NAMESZ) == 0) {
    999 			/* Do not truncate the buffer */
   1000 			if (np->n_descsz > sizeof(epp->ep_machine_arch)) {
   1001 				BADNOTE("description size limit");
   1002 				goto bad;
   1003 			}
   1004 			/*
   1005 			 * Ensure ndesc is NUL-terminated and of the
   1006 			 * expected length.
   1007 			 */
   1008 			if (strnlen(ndesc, np->n_descsz) + 1 !=
   1009 			    np->n_descsz) {
   1010 				BADNOTE("description size");
   1011 				goto bad;
   1012 			}
   1013 			strlcpy(epp->ep_machine_arch, ndesc,
   1014 			    sizeof(epp->ep_machine_arch));
   1015 			break;
   1016 		}
   1017 		BADNOTE("march tag");
   1018 		goto bad;
   1019 
   1020 	case ELF_NOTE_TYPE_MCMODEL_TAG:
   1021 		/* arch specific check for code model */
   1022 #ifdef ELF_MD_MCMODEL_CHECK
   1023 		if (np->n_namesz == ELF_NOTE_MCMODEL_NAMESZ
   1024 		    && memcmp(ndata, ELF_NOTE_MCMODEL_NAME,
   1025 			    ELF_NOTE_MCMODEL_NAMESZ) == 0) {
   1026 			ELF_MD_MCMODEL_CHECK(epp, ndesc, np->n_descsz);
   1027 			break;
   1028 		}
   1029 		BADNOTE("mcmodel tag");
   1030 		goto bad;
   1031 #endif
   1032 		break;
   1033 
   1034 	case ELF_NOTE_TYPE_SUSE_VERSION_TAG:
   1035 		break;
   1036 
   1037 	case ELF_NOTE_TYPE_GO_BUILDID_TAG:
   1038 		break;
   1039 
   1040 	default:
   1041 		BADNOTE("unknown tag");
   1042 bad:
   1043 #ifdef DIAGNOSTIC
   1044 		/* Ignore GNU tags */
   1045 		if (np->n_namesz == ELF_NOTE_GNU_NAMESZ &&
   1046 		    memcmp(ndata, ELF_NOTE_GNU_NAME,
   1047 		    ELF_NOTE_GNU_NAMESZ) == 0)
   1048 		    break;
   1049 
   1050 		int ns = (int)np->n_namesz;
   1051 		printf("%s: Unknown elf note type %d (%s): "
   1052 		    "[namesz=%d, descsz=%d name=%-*.*s]\n",
   1053 		    epp->ep_kname, np->n_type, badnote, np->n_namesz,
   1054 		    np->n_descsz, ns, ns, ndata);
   1055 #endif
   1056 		break;
   1057 	}
   1058 
   1059 	return isnetbsd;
   1060 }
   1061 
   1062 int
   1063 netbsd_elf_probe(struct lwp *l, struct exec_package *epp, void *eh, char *itp,
   1064     vaddr_t *pos)
   1065 {
   1066 	int error;
   1067 
   1068 	if ((error = netbsd_elf_signature(l, epp, eh)) != 0)
   1069 		return error;
   1070 #ifdef ELF_MD_PROBE_FUNC
   1071 	if ((error = ELF_MD_PROBE_FUNC(l, epp, eh, itp, pos)) != 0)
   1072 		return error;
   1073 #elif defined(ELF_INTERP_NON_RELOCATABLE)
   1074 	*pos = ELF_LINK_ADDR;
   1075 #endif
   1076 	epp->ep_flags |= EXEC_FORCEAUX;
   1077 	return 0;
   1078 }
   1079 
   1080 void
   1081 elf_free_emul_arg(void *arg)
   1082 {
   1083 	struct elf_args *ap = arg;
   1084 	KASSERT(ap != NULL);
   1085 	kmem_free(ap, sizeof(*ap));
   1086 }
   1087 
   1088 void
   1089 emul_find_root(struct lwp *l, struct exec_package *epp)
   1090 {
   1091 	struct vnode *vp;
   1092 	const char *emul_path;
   1093 
   1094 	if (epp->ep_emul_root != NULL)
   1095 		/* We've already found it */
   1096 		return;
   1097 
   1098 	emul_path = epp->ep_esch->es_emul->e_path;
   1099 	if (emul_path == NULL)
   1100 		/* Emulation doesn't have a root */
   1101 		return;
   1102 
   1103 	if (namei_simple_kernel(emul_path, NSM_FOLLOW_NOEMULROOT, &vp) != 0)
   1104 		/* emulation root doesn't exist */
   1105 		return;
   1106 
   1107 	epp->ep_emul_root = vp;
   1108 }
   1109 
   1110 /*
   1111  * Search the alternate path for dynamic binary interpreter. If not found
   1112  * there, check if the interpreter exists in within 'proper' tree.
   1113  */
   1114 int
   1115 emul_find_interp(struct lwp *l, struct exec_package *epp, const char *itp)
   1116 {
   1117 	int error;
   1118 	struct pathbuf *pb;
   1119 	struct nameidata nd;
   1120 	unsigned int flags;
   1121 
   1122 	pb = pathbuf_create(itp);
   1123 	if (pb == NULL) {
   1124 		return ENOMEM;
   1125 	}
   1126 
   1127 	/* If we haven't found the emulation root already, do so now */
   1128 	/* Maybe we should remember failures somehow ? */
   1129 	if (epp->ep_esch->es_emul->e_path != 0 && epp->ep_emul_root == NULL)
   1130 		emul_find_root(l, epp);
   1131 
   1132 	if (epp->ep_interp != NULL)
   1133 		vrele(epp->ep_interp);
   1134 
   1135 	/* We need to use the emulation root for the new program,
   1136 	 * not the one for the current process. */
   1137 	if (epp->ep_emul_root == NULL)
   1138 		flags = FOLLOW;
   1139 	else {
   1140 		nd.ni_erootdir = epp->ep_emul_root;
   1141 		/* hack: Pass in the emulation path for ktrace calls */
   1142 		nd.ni_next = epp->ep_esch->es_emul->e_path;
   1143 		flags = FOLLOW | TRYEMULROOT | EMULROOTSET;
   1144 	}
   1145 
   1146 	NDINIT(&nd, LOOKUP, flags, pb);
   1147 	error = namei(&nd);
   1148 	if (error != 0) {
   1149 		epp->ep_interp = NULL;
   1150 		pathbuf_destroy(pb);
   1151 		return error;
   1152 	}
   1153 
   1154 	/* Save interpreter in case we actually need to load it */
   1155 	epp->ep_interp = nd.ni_vp;
   1156 
   1157 	pathbuf_destroy(pb);
   1158 
   1159 	return 0;
   1160 }
   1161