xref: /freebsd/sys/kern/link_elf.c (revision adc56f5a383771f594829b7db9c263b6f0dcf1bd)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1998-2000 Doug Rabson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_ddb.h"
33 #include "opt_gdb.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #ifdef GPROF
38 #include <sys/gmon.h>
39 #endif
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #ifdef SPARSE_MAPPING
44 #include <sys/mman.h>
45 #endif
46 #include <sys/mutex.h>
47 #include <sys/mount.h>
48 #include <sys/pcpu.h>
49 #include <sys/proc.h>
50 #include <sys/namei.h>
51 #include <sys/fcntl.h>
52 #include <sys/vnode.h>
53 #include <sys/linker.h>
54 
55 #include <machine/elf.h>
56 
57 #include <net/vnet.h>
58 
59 #include <security/mac/mac_framework.h>
60 
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #ifdef SPARSE_MAPPING
64 #include <vm/vm_object.h>
65 #include <vm/vm_kern.h>
66 #include <vm/vm_extern.h>
67 #endif
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 
71 #include <sys/link_elf.h>
72 
73 #include "linker_if.h"
74 
75 #define MAXSEGS 4
76 
77 typedef struct elf_file {
78 	struct linker_file lf;		/* Common fields */
79 	int		preloaded;	/* Was file pre-loaded */
80 	caddr_t		address;	/* Relocation address */
81 #ifdef SPARSE_MAPPING
82 	vm_object_t	object;		/* VM object to hold file pages */
83 #endif
84 	Elf_Dyn		*dynamic;	/* Symbol table etc. */
85 	Elf_Hashelt	nbuckets;	/* DT_HASH info */
86 	Elf_Hashelt	nchains;
87 	const Elf_Hashelt *buckets;
88 	const Elf_Hashelt *chains;
89 	caddr_t		hash;
90 	caddr_t		strtab;		/* DT_STRTAB */
91 	int		strsz;		/* DT_STRSZ */
92 	const Elf_Sym	*symtab;		/* DT_SYMTAB */
93 	Elf_Addr	*got;		/* DT_PLTGOT */
94 	const Elf_Rel	*pltrel;	/* DT_JMPREL */
95 	int		pltrelsize;	/* DT_PLTRELSZ */
96 	const Elf_Rela	*pltrela;	/* DT_JMPREL */
97 	int		pltrelasize;	/* DT_PLTRELSZ */
98 	const Elf_Rel	*rel;		/* DT_REL */
99 	int		relsize;	/* DT_RELSZ */
100 	const Elf_Rela	*rela;		/* DT_RELA */
101 	int		relasize;	/* DT_RELASZ */
102 	caddr_t		modptr;
103 	const Elf_Sym	*ddbsymtab;	/* The symbol table we are using */
104 	long		ddbsymcnt;	/* Number of symbols */
105 	caddr_t		ddbstrtab;	/* String table */
106 	long		ddbstrcnt;	/* number of bytes in string table */
107 	caddr_t		symbase;	/* malloc'ed symbold base */
108 	caddr_t		strbase;	/* malloc'ed string base */
109 	caddr_t		ctftab;		/* CTF table */
110 	long		ctfcnt;		/* number of bytes in CTF table */
111 	caddr_t		ctfoff;		/* CTF offset table */
112 	caddr_t		typoff;		/* Type offset table */
113 	long		typlen;		/* Number of type entries. */
114 	Elf_Addr	pcpu_start;	/* Pre-relocation pcpu set start. */
115 	Elf_Addr	pcpu_stop;	/* Pre-relocation pcpu set stop. */
116 	Elf_Addr	pcpu_base;	/* Relocated pcpu set address. */
117 #ifdef VIMAGE
118 	Elf_Addr	vnet_start;	/* Pre-relocation vnet set start. */
119 	Elf_Addr	vnet_stop;	/* Pre-relocation vnet set stop. */
120 	Elf_Addr	vnet_base;	/* Relocated vnet set address. */
121 #endif
122 #ifdef GDB
123 	struct link_map	gdb;		/* hooks for gdb */
124 #endif
125 } *elf_file_t;
126 
127 struct elf_set {
128 	Elf_Addr	es_start;
129 	Elf_Addr	es_stop;
130 	Elf_Addr	es_base;
131 	TAILQ_ENTRY(elf_set)	es_link;
132 };
133 
134 TAILQ_HEAD(elf_set_head, elf_set);
135 
136 #include <kern/kern_ctf.c>
137 
138 static int	link_elf_link_common_finish(linker_file_t);
139 static int	link_elf_link_preload(linker_class_t cls,
140 				      const char *, linker_file_t *);
141 static int	link_elf_link_preload_finish(linker_file_t);
142 static int	link_elf_load_file(linker_class_t, const char *,
143 		    linker_file_t *);
144 static int	link_elf_lookup_symbol(linker_file_t, const char *,
145 		    c_linker_sym_t *);
146 static int	link_elf_symbol_values(linker_file_t, c_linker_sym_t,
147 		    linker_symval_t *);
148 static int	link_elf_search_symbol(linker_file_t, caddr_t,
149 		    c_linker_sym_t *, long *);
150 
151 static void	link_elf_unload_file(linker_file_t);
152 static void	link_elf_unload_preload(linker_file_t);
153 static int	link_elf_lookup_set(linker_file_t, const char *,
154 		    void ***, void ***, int *);
155 static int	link_elf_each_function_name(linker_file_t,
156 		    int (*)(const char *, void *), void *);
157 static int	link_elf_each_function_nameval(linker_file_t,
158 		    linker_function_nameval_callback_t, void *);
159 static void	link_elf_reloc_local(linker_file_t);
160 static long	link_elf_symtab_get(linker_file_t, const Elf_Sym **);
161 static long	link_elf_strtab_get(linker_file_t, caddr_t *);
162 static int	elf_lookup(linker_file_t, Elf_Size, int, Elf_Addr *);
163 
164 static kobj_method_t link_elf_methods[] = {
165 	KOBJMETHOD(linker_lookup_symbol,	link_elf_lookup_symbol),
166 	KOBJMETHOD(linker_symbol_values,	link_elf_symbol_values),
167 	KOBJMETHOD(linker_search_symbol,	link_elf_search_symbol),
168 	KOBJMETHOD(linker_unload,		link_elf_unload_file),
169 	KOBJMETHOD(linker_load_file,		link_elf_load_file),
170 	KOBJMETHOD(linker_link_preload,		link_elf_link_preload),
171 	KOBJMETHOD(linker_link_preload_finish,	link_elf_link_preload_finish),
172 	KOBJMETHOD(linker_lookup_set,		link_elf_lookup_set),
173 	KOBJMETHOD(linker_each_function_name,	link_elf_each_function_name),
174 	KOBJMETHOD(linker_each_function_nameval, link_elf_each_function_nameval),
175 	KOBJMETHOD(linker_ctf_get,		link_elf_ctf_get),
176 	KOBJMETHOD(linker_symtab_get,		link_elf_symtab_get),
177 	KOBJMETHOD(linker_strtab_get,		link_elf_strtab_get),
178 	KOBJMETHOD_END
179 };
180 
181 static struct linker_class link_elf_class = {
182 #if ELF_TARG_CLASS == ELFCLASS32
183 	"elf32",
184 #else
185 	"elf64",
186 #endif
187 	link_elf_methods, sizeof(struct elf_file)
188 };
189 
190 typedef int (*elf_reloc_fn)(linker_file_t lf, Elf_Addr relocbase,
191     const void *data, int type, elf_lookup_fn lookup);
192 
193 static int	parse_dynamic(elf_file_t);
194 static int	relocate_file(elf_file_t);
195 static int	relocate_file1(elf_file_t ef, elf_lookup_fn lookup,
196 		    elf_reloc_fn reloc, bool ifuncs);
197 static int	link_elf_preload_parse_symbols(elf_file_t);
198 
199 static struct elf_set_head set_pcpu_list;
200 #ifdef VIMAGE
201 static struct elf_set_head set_vnet_list;
202 #endif
203 
204 static void
205 elf_set_add(struct elf_set_head *list, Elf_Addr start, Elf_Addr stop, Elf_Addr base)
206 {
207 	struct elf_set *set, *iter;
208 
209 	set = malloc(sizeof(*set), M_LINKER, M_WAITOK);
210 	set->es_start = start;
211 	set->es_stop = stop;
212 	set->es_base = base;
213 
214 	TAILQ_FOREACH(iter, list, es_link) {
215 
216 		KASSERT((set->es_start < iter->es_start && set->es_stop < iter->es_stop) ||
217 		    (set->es_start > iter->es_start && set->es_stop > iter->es_stop),
218 		    ("linker sets intersection: to insert: 0x%jx-0x%jx; inserted: 0x%jx-0x%jx",
219 		    (uintmax_t)set->es_start, (uintmax_t)set->es_stop,
220 		    (uintmax_t)iter->es_start, (uintmax_t)iter->es_stop));
221 
222 		if (iter->es_start > set->es_start) {
223 			TAILQ_INSERT_BEFORE(iter, set, es_link);
224 			break;
225 		}
226 	}
227 
228 	if (iter == NULL)
229 		TAILQ_INSERT_TAIL(list, set, es_link);
230 }
231 
232 static int
233 elf_set_find(struct elf_set_head *list, Elf_Addr addr, Elf_Addr *start, Elf_Addr *base)
234 {
235 	struct elf_set *set;
236 
237 	TAILQ_FOREACH(set, list, es_link) {
238 		if (addr < set->es_start)
239 			return (0);
240 		if (addr < set->es_stop) {
241 			*start = set->es_start;
242 			*base = set->es_base;
243 			return (1);
244 		}
245 	}
246 
247 	return (0);
248 }
249 
250 static void
251 elf_set_delete(struct elf_set_head *list, Elf_Addr start)
252 {
253 	struct elf_set *set;
254 
255 	TAILQ_FOREACH(set, list, es_link) {
256 		if (start < set->es_start)
257 			break;
258 		if (start == set->es_start) {
259 			TAILQ_REMOVE(list, set, es_link);
260 			free(set, M_LINKER);
261 			return;
262 		}
263 	}
264 	KASSERT(0, ("deleting unknown linker set (start = 0x%jx)",
265 	    (uintmax_t)start));
266 }
267 
268 #ifdef GDB
269 static void	r_debug_state(struct r_debug *, struct link_map *);
270 
271 /*
272  * A list of loaded modules for GDB to use for loading symbols.
273  */
274 struct r_debug r_debug;
275 
276 #define GDB_STATE(s) do {				\
277 	r_debug.r_state = s; r_debug_state(NULL, NULL);	\
278 } while (0)
279 
280 /*
281  * Function for the debugger to set a breakpoint on to gain control.
282  */
283 static void
284 r_debug_state(struct r_debug *dummy_one __unused,
285 	      struct link_map *dummy_two __unused)
286 {
287 }
288 
289 static void
290 link_elf_add_gdb(struct link_map *l)
291 {
292 	struct link_map *prev;
293 
294 	l->l_next = NULL;
295 
296 	if (r_debug.r_map == NULL) {
297 		/* Add first. */
298 		l->l_prev = NULL;
299 		r_debug.r_map = l;
300 	} else {
301 		/* Append to list. */
302 		for (prev = r_debug.r_map;
303 		    prev->l_next != NULL;
304 		    prev = prev->l_next)
305 			;
306 		l->l_prev = prev;
307 		prev->l_next = l;
308 	}
309 }
310 
311 static void
312 link_elf_delete_gdb(struct link_map *l)
313 {
314 	if (l->l_prev == NULL) {
315 		/* Remove first. */
316 		if ((r_debug.r_map = l->l_next) != NULL)
317 			l->l_next->l_prev = NULL;
318 	} else {
319 		/* Remove any but first. */
320 		if ((l->l_prev->l_next = l->l_next) != NULL)
321 			l->l_next->l_prev = l->l_prev;
322 	}
323 }
324 #endif /* GDB */
325 
326 /*
327  * The kernel symbol table starts here.
328  */
329 extern struct _dynamic _DYNAMIC;
330 
331 static void
332 link_elf_error(const char *filename, const char *s)
333 {
334 	if (filename == NULL)
335 		printf("kldload: %s\n", s);
336 	else
337 		printf("kldload: %s: %s\n", filename, s);
338 }
339 
340 static void
341 link_elf_invoke_ctors(caddr_t addr, size_t size)
342 {
343 	void (**ctor)(void);
344 	size_t i, cnt;
345 
346 	if (addr == NULL || size == 0)
347 		return;
348 	cnt = size / sizeof(*ctor);
349 	ctor = (void *)addr;
350 	for (i = 0; i < cnt; i++) {
351 		if (ctor[i] != NULL)
352 			(*ctor[i])();
353 	}
354 }
355 
356 /*
357  * Actions performed after linking/loading both the preloaded kernel and any
358  * modules; whether preloaded or dynamicly loaded.
359  */
360 static int
361 link_elf_link_common_finish(linker_file_t lf)
362 {
363 #ifdef GDB
364 	elf_file_t ef = (elf_file_t)lf;
365 	char *newfilename;
366 #endif
367 	int error;
368 
369 	/* Notify MD code that a module is being loaded. */
370 	error = elf_cpu_load_file(lf);
371 	if (error != 0)
372 		return (error);
373 
374 #ifdef GDB
375 	GDB_STATE(RT_ADD);
376 	ef->gdb.l_addr = lf->address;
377 	newfilename = malloc(strlen(lf->filename) + 1, M_LINKER, M_WAITOK);
378 	strcpy(newfilename, lf->filename);
379 	ef->gdb.l_name = newfilename;
380 	ef->gdb.l_ld = ef->dynamic;
381 	link_elf_add_gdb(&ef->gdb);
382 	GDB_STATE(RT_CONSISTENT);
383 #endif
384 
385 	/* Invoke .ctors */
386 	link_elf_invoke_ctors(lf->ctors_addr, lf->ctors_size);
387 	return (0);
388 }
389 
390 extern vm_offset_t __startkernel, __endkernel;
391 
392 static void
393 link_elf_init(void* arg)
394 {
395 	Elf_Dyn *dp;
396 	Elf_Addr *ctors_addrp;
397 	Elf_Size *ctors_sizep;
398 	caddr_t modptr, baseptr, sizeptr;
399 	elf_file_t ef;
400 	char *modname;
401 
402 	linker_add_class(&link_elf_class);
403 
404 	dp = (Elf_Dyn *)&_DYNAMIC;
405 	modname = NULL;
406 	modptr = preload_search_by_type("elf" __XSTRING(__ELF_WORD_SIZE) " kernel");
407 	if (modptr == NULL)
408 		modptr = preload_search_by_type("elf kernel");
409 	modname = (char *)preload_search_info(modptr, MODINFO_NAME);
410 	if (modname == NULL)
411 		modname = "kernel";
412 	linker_kernel_file = linker_make_file(modname, &link_elf_class);
413 	if (linker_kernel_file == NULL)
414 		panic("%s: Can't create linker structures for kernel",
415 		    __func__);
416 
417 	ef = (elf_file_t) linker_kernel_file;
418 	ef->preloaded = 1;
419 #ifdef __powerpc__
420 	ef->address = (caddr_t) (__startkernel - KERNBASE);
421 #else
422 	ef->address = 0;
423 #endif
424 #ifdef SPARSE_MAPPING
425 	ef->object = NULL;
426 #endif
427 	ef->dynamic = dp;
428 
429 	if (dp != NULL)
430 		parse_dynamic(ef);
431 #ifdef __powerpc__
432 	linker_kernel_file->address = (caddr_t)__startkernel;
433 	linker_kernel_file->size = (intptr_t)(__endkernel - __startkernel);
434 #else
435 	linker_kernel_file->address += KERNBASE;
436 	linker_kernel_file->size = -(intptr_t)linker_kernel_file->address;
437 #endif
438 
439 	if (modptr != NULL) {
440 		ef->modptr = modptr;
441 		baseptr = preload_search_info(modptr, MODINFO_ADDR);
442 		if (baseptr != NULL)
443 			linker_kernel_file->address = *(caddr_t *)baseptr;
444 		sizeptr = preload_search_info(modptr, MODINFO_SIZE);
445 		if (sizeptr != NULL)
446 			linker_kernel_file->size = *(size_t *)sizeptr;
447 		ctors_addrp = (Elf_Addr *)preload_search_info(modptr,
448 			MODINFO_METADATA | MODINFOMD_CTORS_ADDR);
449 		ctors_sizep = (Elf_Size *)preload_search_info(modptr,
450 			MODINFO_METADATA | MODINFOMD_CTORS_SIZE);
451 		if (ctors_addrp != NULL && ctors_sizep != NULL) {
452 			linker_kernel_file->ctors_addr = ef->address +
453 			    *ctors_addrp;
454 			linker_kernel_file->ctors_size = *ctors_sizep;
455 		}
456 	}
457 	(void)link_elf_preload_parse_symbols(ef);
458 
459 #ifdef GDB
460 	r_debug.r_map = NULL;
461 	r_debug.r_brk = r_debug_state;
462 	r_debug.r_state = RT_CONSISTENT;
463 #endif
464 
465 	(void)link_elf_link_common_finish(linker_kernel_file);
466 	linker_kernel_file->flags |= LINKER_FILE_LINKED;
467 	TAILQ_INIT(&set_pcpu_list);
468 #ifdef VIMAGE
469 	TAILQ_INIT(&set_vnet_list);
470 #endif
471 }
472 
473 SYSINIT(link_elf, SI_SUB_KLD, SI_ORDER_THIRD, link_elf_init, NULL);
474 
475 static int
476 link_elf_preload_parse_symbols(elf_file_t ef)
477 {
478 	caddr_t pointer;
479 	caddr_t ssym, esym, base;
480 	caddr_t strtab;
481 	int strcnt;
482 	Elf_Sym *symtab;
483 	int symcnt;
484 
485 	if (ef->modptr == NULL)
486 		return (0);
487 	pointer = preload_search_info(ef->modptr,
488 	    MODINFO_METADATA | MODINFOMD_SSYM);
489 	if (pointer == NULL)
490 		return (0);
491 	ssym = *(caddr_t *)pointer;
492 	pointer = preload_search_info(ef->modptr,
493 	    MODINFO_METADATA | MODINFOMD_ESYM);
494 	if (pointer == NULL)
495 		return (0);
496 	esym = *(caddr_t *)pointer;
497 
498 	base = ssym;
499 
500 	symcnt = *(long *)base;
501 	base += sizeof(long);
502 	symtab = (Elf_Sym *)base;
503 	base += roundup(symcnt, sizeof(long));
504 
505 	if (base > esym || base < ssym) {
506 		printf("Symbols are corrupt!\n");
507 		return (EINVAL);
508 	}
509 
510 	strcnt = *(long *)base;
511 	base += sizeof(long);
512 	strtab = base;
513 	base += roundup(strcnt, sizeof(long));
514 
515 	if (base > esym || base < ssym) {
516 		printf("Symbols are corrupt!\n");
517 		return (EINVAL);
518 	}
519 
520 	ef->ddbsymtab = symtab;
521 	ef->ddbsymcnt = symcnt / sizeof(Elf_Sym);
522 	ef->ddbstrtab = strtab;
523 	ef->ddbstrcnt = strcnt;
524 
525 	return (0);
526 }
527 
528 static int
529 parse_dynamic(elf_file_t ef)
530 {
531 	Elf_Dyn *dp;
532 	int plttype = DT_REL;
533 
534 	for (dp = ef->dynamic; dp->d_tag != DT_NULL; dp++) {
535 		switch (dp->d_tag) {
536 		case DT_HASH:
537 		{
538 			/* From src/libexec/rtld-elf/rtld.c */
539 			const Elf_Hashelt *hashtab = (const Elf_Hashelt *)
540 			    (ef->address + dp->d_un.d_ptr);
541 			ef->nbuckets = hashtab[0];
542 			ef->nchains = hashtab[1];
543 			ef->buckets = hashtab + 2;
544 			ef->chains = ef->buckets + ef->nbuckets;
545 			break;
546 		}
547 		case DT_STRTAB:
548 			ef->strtab = (caddr_t) (ef->address + dp->d_un.d_ptr);
549 			break;
550 		case DT_STRSZ:
551 			ef->strsz = dp->d_un.d_val;
552 			break;
553 		case DT_SYMTAB:
554 			ef->symtab = (Elf_Sym*) (ef->address + dp->d_un.d_ptr);
555 			break;
556 		case DT_SYMENT:
557 			if (dp->d_un.d_val != sizeof(Elf_Sym))
558 				return (ENOEXEC);
559 			break;
560 		case DT_PLTGOT:
561 			ef->got = (Elf_Addr *) (ef->address + dp->d_un.d_ptr);
562 			break;
563 		case DT_REL:
564 			ef->rel = (const Elf_Rel *) (ef->address + dp->d_un.d_ptr);
565 			break;
566 		case DT_RELSZ:
567 			ef->relsize = dp->d_un.d_val;
568 			break;
569 		case DT_RELENT:
570 			if (dp->d_un.d_val != sizeof(Elf_Rel))
571 				return (ENOEXEC);
572 			break;
573 		case DT_JMPREL:
574 			ef->pltrel = (const Elf_Rel *) (ef->address + dp->d_un.d_ptr);
575 			break;
576 		case DT_PLTRELSZ:
577 			ef->pltrelsize = dp->d_un.d_val;
578 			break;
579 		case DT_RELA:
580 			ef->rela = (const Elf_Rela *) (ef->address + dp->d_un.d_ptr);
581 			break;
582 		case DT_RELASZ:
583 			ef->relasize = dp->d_un.d_val;
584 			break;
585 		case DT_RELAENT:
586 			if (dp->d_un.d_val != sizeof(Elf_Rela))
587 				return (ENOEXEC);
588 			break;
589 		case DT_PLTREL:
590 			plttype = dp->d_un.d_val;
591 			if (plttype != DT_REL && plttype != DT_RELA)
592 				return (ENOEXEC);
593 			break;
594 #ifdef GDB
595 		case DT_DEBUG:
596 			dp->d_un.d_ptr = (Elf_Addr)&r_debug;
597 			break;
598 #endif
599 		}
600 	}
601 
602 	if (plttype == DT_RELA) {
603 		ef->pltrela = (const Elf_Rela *)ef->pltrel;
604 		ef->pltrel = NULL;
605 		ef->pltrelasize = ef->pltrelsize;
606 		ef->pltrelsize = 0;
607 	}
608 
609 	ef->ddbsymtab = ef->symtab;
610 	ef->ddbsymcnt = ef->nchains;
611 	ef->ddbstrtab = ef->strtab;
612 	ef->ddbstrcnt = ef->strsz;
613 
614 	return (0);
615 }
616 
617 #define	LS_PADDING	0x90909090
618 static int
619 parse_dpcpu(elf_file_t ef)
620 {
621 	int error, size;
622 #if defined(__i386__)
623 	uint32_t pad;
624 #endif
625 
626 	ef->pcpu_start = 0;
627 	ef->pcpu_stop = 0;
628 	error = link_elf_lookup_set(&ef->lf, "pcpu", (void ***)&ef->pcpu_start,
629 	    (void ***)&ef->pcpu_stop, NULL);
630 	/* Error just means there is no pcpu set to relocate. */
631 	if (error != 0)
632 		return (0);
633 	size = (uintptr_t)ef->pcpu_stop - (uintptr_t)ef->pcpu_start;
634 	/* Empty set? */
635 	if (size < 1)
636 		return (0);
637 #if defined(__i386__)
638 	/* In case we do find __start/stop_set_ symbols double-check. */
639 	if (size < 4) {
640 		uprintf("Kernel module '%s' must be recompiled with "
641 		    "linker script\n", ef->lf.pathname);
642 		return (ENOEXEC);
643 	}
644 
645 	/* Padding from linker-script correct? */
646 	pad = *(uint32_t *)((uintptr_t)ef->pcpu_stop - sizeof(pad));
647 	if (pad != LS_PADDING) {
648 		uprintf("Kernel module '%s' must be recompiled with "
649 		    "linker script, invalid padding %#04x (%#04x)\n",
650 		    ef->lf.pathname, pad, LS_PADDING);
651 		return (ENOEXEC);
652 	}
653 	/* If we only have valid padding, nothing to do. */
654 	if (size == 4)
655 		return (0);
656 #endif
657 	/*
658 	 * Allocate space in the primary pcpu area.  Copy in our
659 	 * initialization from the data section and then initialize
660 	 * all per-cpu storage from that.
661 	 */
662 	ef->pcpu_base = (Elf_Addr)(uintptr_t)dpcpu_alloc(size);
663 	if (ef->pcpu_base == 0) {
664 		printf("%s: pcpu module space is out of space; "
665 		    "cannot allocate %d for %s\n",
666 		    __func__, size, ef->lf.pathname);
667 		return (ENOSPC);
668 	}
669 	memcpy((void *)ef->pcpu_base, (void *)ef->pcpu_start, size);
670 	dpcpu_copy((void *)ef->pcpu_base, size);
671 	elf_set_add(&set_pcpu_list, ef->pcpu_start, ef->pcpu_stop,
672 	    ef->pcpu_base);
673 
674 	return (0);
675 }
676 
677 #ifdef VIMAGE
678 static int
679 parse_vnet(elf_file_t ef)
680 {
681 	int error, size;
682 #if defined(__i386__)
683 	uint32_t pad;
684 #endif
685 
686 	ef->vnet_start = 0;
687 	ef->vnet_stop = 0;
688 	error = link_elf_lookup_set(&ef->lf, "vnet", (void ***)&ef->vnet_start,
689 	    (void ***)&ef->vnet_stop, NULL);
690 	/* Error just means there is no vnet data set to relocate. */
691 	if (error != 0)
692 		return (0);
693 	size = (uintptr_t)ef->vnet_stop - (uintptr_t)ef->vnet_start;
694 	/* Empty set? */
695 	if (size < 1)
696 		return (0);
697 #if defined(__i386__)
698 	/* In case we do find __start/stop_set_ symbols double-check. */
699 	if (size < 4) {
700 		uprintf("Kernel module '%s' must be recompiled with "
701 		    "linker script\n", ef->lf.pathname);
702 		return (ENOEXEC);
703 	}
704 
705 	/* Padding from linker-script correct? */
706 	pad = *(uint32_t *)((uintptr_t)ef->vnet_stop - sizeof(pad));
707 	if (pad != LS_PADDING) {
708 		uprintf("Kernel module '%s' must be recompiled with "
709 		    "linker script, invalid padding %#04x (%#04x)\n",
710 		    ef->lf.pathname, pad, LS_PADDING);
711 		return (ENOEXEC);
712 	}
713 	/* If we only have valid padding, nothing to do. */
714 	if (size == 4)
715 		return (0);
716 #endif
717 	/*
718 	 * Allocate space in the primary vnet area.  Copy in our
719 	 * initialization from the data section and then initialize
720 	 * all per-vnet storage from that.
721 	 */
722 	ef->vnet_base = (Elf_Addr)(uintptr_t)vnet_data_alloc(size);
723 	if (ef->vnet_base == 0) {
724 		printf("%s: vnet module space is out of space; "
725 		    "cannot allocate %d for %s\n",
726 		    __func__, size, ef->lf.pathname);
727 		return (ENOSPC);
728 	}
729 	memcpy((void *)ef->vnet_base, (void *)ef->vnet_start, size);
730 	vnet_data_copy((void *)ef->vnet_base, size);
731 	elf_set_add(&set_vnet_list, ef->vnet_start, ef->vnet_stop,
732 	    ef->vnet_base);
733 
734 	return (0);
735 }
736 #endif
737 #undef LS_PADDING
738 
739 /*
740  * Apply the specified protection to the loadable segments of a preloaded linker
741  * file.
742  */
743 static int
744 preload_protect(elf_file_t ef, vm_prot_t prot)
745 {
746 #ifdef __amd64__
747 	Elf_Ehdr *hdr;
748 	Elf_Phdr *phdr, *phlimit;
749 	vm_prot_t nprot;
750 	int error;
751 
752 	error = 0;
753 	hdr = (Elf_Ehdr *)ef->address;
754 	phdr = (Elf_Phdr *)(ef->address + hdr->e_phoff);
755 	phlimit = phdr + hdr->e_phnum;
756 	for (; phdr < phlimit; phdr++) {
757 		if (phdr->p_type != PT_LOAD)
758 			continue;
759 
760 		nprot = prot | VM_PROT_READ;
761 		if ((phdr->p_flags & PF_W) != 0)
762 			nprot |= VM_PROT_WRITE;
763 		if ((phdr->p_flags & PF_X) != 0)
764 			nprot |= VM_PROT_EXECUTE;
765 		error = pmap_change_prot((vm_offset_t)ef->address +
766 		    phdr->p_vaddr, round_page(phdr->p_memsz), nprot);
767 		if (error != 0)
768 			break;
769 	}
770 	return (error);
771 #else
772 	return (0);
773 #endif
774 }
775 
776 static int
777 link_elf_link_preload(linker_class_t cls, const char *filename,
778     linker_file_t *result)
779 {
780 	Elf_Addr *ctors_addrp;
781 	Elf_Size *ctors_sizep;
782 	caddr_t modptr, baseptr, sizeptr, dynptr;
783 	char *type;
784 	elf_file_t ef;
785 	linker_file_t lf;
786 	int error;
787 	vm_offset_t dp;
788 
789 	/* Look to see if we have the file preloaded */
790 	modptr = preload_search_by_name(filename);
791 	if (modptr == NULL)
792 		return (ENOENT);
793 
794 	type = (char *)preload_search_info(modptr, MODINFO_TYPE);
795 	baseptr = preload_search_info(modptr, MODINFO_ADDR);
796 	sizeptr = preload_search_info(modptr, MODINFO_SIZE);
797 	dynptr = preload_search_info(modptr,
798 	    MODINFO_METADATA | MODINFOMD_DYNAMIC);
799 	if (type == NULL ||
800 	    (strcmp(type, "elf" __XSTRING(__ELF_WORD_SIZE) " module") != 0 &&
801 	     strcmp(type, "elf module") != 0))
802 		return (EFTYPE);
803 	if (baseptr == NULL || sizeptr == NULL || dynptr == NULL)
804 		return (EINVAL);
805 
806 	lf = linker_make_file(filename, &link_elf_class);
807 	if (lf == NULL)
808 		return (ENOMEM);
809 
810 	ef = (elf_file_t) lf;
811 	ef->preloaded = 1;
812 	ef->modptr = modptr;
813 	ef->address = *(caddr_t *)baseptr;
814 #ifdef SPARSE_MAPPING
815 	ef->object = NULL;
816 #endif
817 	dp = (vm_offset_t)ef->address + *(vm_offset_t *)dynptr;
818 	ef->dynamic = (Elf_Dyn *)dp;
819 	lf->address = ef->address;
820 	lf->size = *(size_t *)sizeptr;
821 
822 	ctors_addrp = (Elf_Addr *)preload_search_info(modptr,
823 	    MODINFO_METADATA | MODINFOMD_CTORS_ADDR);
824 	ctors_sizep = (Elf_Size *)preload_search_info(modptr,
825 	    MODINFO_METADATA | MODINFOMD_CTORS_SIZE);
826 	if (ctors_addrp != NULL && ctors_sizep != NULL) {
827 		lf->ctors_addr = ef->address + *ctors_addrp;
828 		lf->ctors_size = *ctors_sizep;
829 	}
830 
831 	error = parse_dynamic(ef);
832 	if (error == 0)
833 		error = parse_dpcpu(ef);
834 #ifdef VIMAGE
835 	if (error == 0)
836 		error = parse_vnet(ef);
837 #endif
838 	if (error == 0)
839 		error = preload_protect(ef, VM_PROT_ALL);
840 	if (error != 0) {
841 		linker_file_unload(lf, LINKER_UNLOAD_FORCE);
842 		return (error);
843 	}
844 	link_elf_reloc_local(lf);
845 	*result = lf;
846 	return (0);
847 }
848 
849 static int
850 link_elf_link_preload_finish(linker_file_t lf)
851 {
852 	elf_file_t ef;
853 	int error;
854 
855 	ef = (elf_file_t) lf;
856 	error = relocate_file(ef);
857 	if (error == 0)
858 		error = preload_protect(ef, VM_PROT_NONE);
859 	if (error != 0)
860 		return (error);
861 	(void)link_elf_preload_parse_symbols(ef);
862 
863 	return (link_elf_link_common_finish(lf));
864 }
865 
866 static int
867 link_elf_load_file(linker_class_t cls, const char* filename,
868     linker_file_t* result)
869 {
870 	struct nameidata nd;
871 	struct thread* td = curthread;	/* XXX */
872 	Elf_Ehdr *hdr;
873 	caddr_t firstpage, segbase;
874 	int nbytes, i;
875 	Elf_Phdr *phdr;
876 	Elf_Phdr *phlimit;
877 	Elf_Phdr *segs[MAXSEGS];
878 	int nsegs;
879 	Elf_Phdr *phdyn;
880 	caddr_t mapbase;
881 	size_t mapsize;
882 	Elf_Addr base_vaddr;
883 	Elf_Addr base_vlimit;
884 	int error = 0;
885 	ssize_t resid;
886 	int flags;
887 	elf_file_t ef;
888 	linker_file_t lf;
889 	Elf_Shdr *shdr;
890 	int symtabindex;
891 	int symstrindex;
892 	int shstrindex;
893 	int symcnt;
894 	int strcnt;
895 	char *shstrs;
896 
897 	shdr = NULL;
898 	lf = NULL;
899 	shstrs = NULL;
900 
901 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, filename, td);
902 	flags = FREAD;
903 	error = vn_open(&nd, &flags, 0, NULL);
904 	if (error != 0)
905 		return (error);
906 	NDFREE(&nd, NDF_ONLY_PNBUF);
907 	if (nd.ni_vp->v_type != VREG) {
908 		error = ENOEXEC;
909 		firstpage = NULL;
910 		goto out;
911 	}
912 #ifdef MAC
913 	error = mac_kld_check_load(curthread->td_ucred, nd.ni_vp);
914 	if (error != 0) {
915 		firstpage = NULL;
916 		goto out;
917 	}
918 #endif
919 
920 	/*
921 	 * Read the elf header from the file.
922 	 */
923 	firstpage = malloc(PAGE_SIZE, M_LINKER, M_WAITOK);
924 	hdr = (Elf_Ehdr *)firstpage;
925 	error = vn_rdwr(UIO_READ, nd.ni_vp, firstpage, PAGE_SIZE, 0,
926 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
927 	    &resid, td);
928 	nbytes = PAGE_SIZE - resid;
929 	if (error != 0)
930 		goto out;
931 
932 	if (!IS_ELF(*hdr)) {
933 		error = ENOEXEC;
934 		goto out;
935 	}
936 
937 	if (hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
938 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA) {
939 		link_elf_error(filename, "Unsupported file layout");
940 		error = ENOEXEC;
941 		goto out;
942 	}
943 	if (hdr->e_ident[EI_VERSION] != EV_CURRENT ||
944 	    hdr->e_version != EV_CURRENT) {
945 		link_elf_error(filename, "Unsupported file version");
946 		error = ENOEXEC;
947 		goto out;
948 	}
949 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN) {
950 		error = ENOSYS;
951 		goto out;
952 	}
953 	if (hdr->e_machine != ELF_TARG_MACH) {
954 		link_elf_error(filename, "Unsupported machine");
955 		error = ENOEXEC;
956 		goto out;
957 	}
958 
959 	/*
960 	 * We rely on the program header being in the first page.
961 	 * This is not strictly required by the ABI specification, but
962 	 * it seems to always true in practice.  And, it simplifies
963 	 * things considerably.
964 	 */
965 	if (!((hdr->e_phentsize == sizeof(Elf_Phdr)) &&
966 	      (hdr->e_phoff + hdr->e_phnum*sizeof(Elf_Phdr) <= PAGE_SIZE) &&
967 	      (hdr->e_phoff + hdr->e_phnum*sizeof(Elf_Phdr) <= nbytes)))
968 		link_elf_error(filename, "Unreadable program headers");
969 
970 	/*
971 	 * Scan the program header entries, and save key information.
972 	 *
973 	 * We rely on there being exactly two load segments, text and data,
974 	 * in that order.
975 	 */
976 	phdr = (Elf_Phdr *) (firstpage + hdr->e_phoff);
977 	phlimit = phdr + hdr->e_phnum;
978 	nsegs = 0;
979 	phdyn = NULL;
980 	while (phdr < phlimit) {
981 		switch (phdr->p_type) {
982 		case PT_LOAD:
983 			if (nsegs == MAXSEGS) {
984 				link_elf_error(filename, "Too many sections");
985 				error = ENOEXEC;
986 				goto out;
987 			}
988 			/*
989 			 * XXX: We just trust they come in right order ??
990 			 */
991 			segs[nsegs] = phdr;
992 			++nsegs;
993 			break;
994 
995 		case PT_DYNAMIC:
996 			phdyn = phdr;
997 			break;
998 
999 		case PT_INTERP:
1000 			error = ENOSYS;
1001 			goto out;
1002 		}
1003 
1004 		++phdr;
1005 	}
1006 	if (phdyn == NULL) {
1007 		link_elf_error(filename, "Object is not dynamically-linked");
1008 		error = ENOEXEC;
1009 		goto out;
1010 	}
1011 	if (nsegs == 0) {
1012 		link_elf_error(filename, "No sections");
1013 		error = ENOEXEC;
1014 		goto out;
1015 	}
1016 
1017 	/*
1018 	 * Allocate the entire address space of the object, to stake
1019 	 * out our contiguous region, and to establish the base
1020 	 * address for relocation.
1021 	 */
1022 	base_vaddr = trunc_page(segs[0]->p_vaddr);
1023 	base_vlimit = round_page(segs[nsegs - 1]->p_vaddr +
1024 	    segs[nsegs - 1]->p_memsz);
1025 	mapsize = base_vlimit - base_vaddr;
1026 
1027 	lf = linker_make_file(filename, &link_elf_class);
1028 	if (lf == NULL) {
1029 		error = ENOMEM;
1030 		goto out;
1031 	}
1032 
1033 	ef = (elf_file_t) lf;
1034 #ifdef SPARSE_MAPPING
1035 	ef->object = vm_object_allocate(OBJT_PHYS, atop(mapsize));
1036 	if (ef->object == NULL) {
1037 		error = ENOMEM;
1038 		goto out;
1039 	}
1040 #ifdef __amd64__
1041 	mapbase = (caddr_t)KERNBASE;
1042 #else
1043 	mapbase = (caddr_t)vm_map_min(kernel_map);
1044 #endif
1045 	/*
1046 	 * Mapping protections are downgraded after relocation processing.
1047 	 */
1048 	error = vm_map_find(kernel_map, ef->object, 0,
1049 	    (vm_offset_t *)&mapbase, mapsize, 0, VMFS_OPTIMAL_SPACE,
1050 	    VM_PROT_ALL, VM_PROT_ALL, 0);
1051 	if (error != 0) {
1052 		vm_object_deallocate(ef->object);
1053 		ef->object = NULL;
1054 		goto out;
1055 	}
1056 #else
1057 	mapbase = malloc(mapsize, M_LINKER, M_EXEC | M_WAITOK);
1058 #endif
1059 	ef->address = mapbase;
1060 
1061 	/*
1062 	 * Read the text and data sections and zero the bss.
1063 	 */
1064 	for (i = 0; i < nsegs; i++) {
1065 		segbase = mapbase + segs[i]->p_vaddr - base_vaddr;
1066 
1067 #ifdef SPARSE_MAPPING
1068 		/*
1069 		 * Consecutive segments may have different mapping permissions,
1070 		 * so be strict and verify that their mappings do not overlap.
1071 		 */
1072 		if (((vm_offset_t)segbase & PAGE_MASK) != 0) {
1073 			error = EINVAL;
1074 			goto out;
1075 		}
1076 
1077 		error = vm_map_wire(kernel_map,
1078 		    (vm_offset_t)segbase,
1079 		    (vm_offset_t)segbase + round_page(segs[i]->p_memsz),
1080 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1081 		if (error != KERN_SUCCESS) {
1082 			error = ENOMEM;
1083 			goto out;
1084 		}
1085 #endif
1086 
1087 		error = vn_rdwr(UIO_READ, nd.ni_vp,
1088 		    segbase, segs[i]->p_filesz, segs[i]->p_offset,
1089 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1090 		    &resid, td);
1091 		if (error != 0)
1092 			goto out;
1093 		bzero(segbase + segs[i]->p_filesz,
1094 		    segs[i]->p_memsz - segs[i]->p_filesz);
1095 	}
1096 
1097 #ifdef GPROF
1098 	/* Update profiling information with the new text segment. */
1099 	mtx_lock(&Giant);
1100 	kmupetext((uintfptr_t)(mapbase + segs[0]->p_vaddr - base_vaddr +
1101 	    segs[0]->p_memsz));
1102 	mtx_unlock(&Giant);
1103 #endif
1104 
1105 	ef->dynamic = (Elf_Dyn *) (mapbase + phdyn->p_vaddr - base_vaddr);
1106 
1107 	lf->address = ef->address;
1108 	lf->size = mapsize;
1109 
1110 	error = parse_dynamic(ef);
1111 	if (error != 0)
1112 		goto out;
1113 	error = parse_dpcpu(ef);
1114 	if (error != 0)
1115 		goto out;
1116 #ifdef VIMAGE
1117 	error = parse_vnet(ef);
1118 	if (error != 0)
1119 		goto out;
1120 #endif
1121 	link_elf_reloc_local(lf);
1122 
1123 	VOP_UNLOCK(nd.ni_vp, 0);
1124 	error = linker_load_dependencies(lf);
1125 	vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1126 	if (error != 0)
1127 		goto out;
1128 	error = relocate_file(ef);
1129 	if (error != 0)
1130 		goto out;
1131 
1132 #ifdef SPARSE_MAPPING
1133 	/*
1134 	 * Downgrade permissions on text segment mappings now that relocation
1135 	 * processing is complete.  Restrict permissions on read-only segments.
1136 	 */
1137 	for (i = 0; i < nsegs; i++) {
1138 		vm_prot_t prot;
1139 
1140 		if (segs[i]->p_type != PT_LOAD)
1141 			continue;
1142 
1143 		prot = VM_PROT_READ;
1144 		if ((segs[i]->p_flags & PF_W) != 0)
1145 			prot |= VM_PROT_WRITE;
1146 		if ((segs[i]->p_flags & PF_X) != 0)
1147 			prot |= VM_PROT_EXECUTE;
1148 		segbase = mapbase + segs[i]->p_vaddr - base_vaddr;
1149 		error = vm_map_protect(kernel_map,
1150 		    (vm_offset_t)segbase,
1151 		    (vm_offset_t)segbase + round_page(segs[i]->p_memsz),
1152 		    prot, FALSE);
1153 		if (error != KERN_SUCCESS) {
1154 			error = ENOMEM;
1155 			goto out;
1156 		}
1157 	}
1158 #endif
1159 
1160 	/*
1161 	 * Try and load the symbol table if it's present.  (you can
1162 	 * strip it!)
1163 	 */
1164 	nbytes = hdr->e_shnum * hdr->e_shentsize;
1165 	if (nbytes == 0 || hdr->e_shoff == 0)
1166 		goto nosyms;
1167 	shdr = malloc(nbytes, M_LINKER, M_WAITOK | M_ZERO);
1168 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1169 	    (caddr_t)shdr, nbytes, hdr->e_shoff,
1170 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1171 	    &resid, td);
1172 	if (error != 0)
1173 		goto out;
1174 
1175 	/* Read section string table */
1176 	shstrindex = hdr->e_shstrndx;
1177 	if (shstrindex != 0 && shdr[shstrindex].sh_type == SHT_STRTAB &&
1178 	    shdr[shstrindex].sh_size != 0) {
1179 		nbytes = shdr[shstrindex].sh_size;
1180 		shstrs = malloc(nbytes, M_LINKER, M_WAITOK | M_ZERO);
1181 		error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)shstrs, nbytes,
1182 		    shdr[shstrindex].sh_offset, UIO_SYSSPACE, IO_NODELOCKED,
1183 		    td->td_ucred, NOCRED, &resid, td);
1184 		if (error)
1185 			goto out;
1186 	}
1187 
1188 	symtabindex = -1;
1189 	symstrindex = -1;
1190 	for (i = 0; i < hdr->e_shnum; i++) {
1191 		if (shdr[i].sh_type == SHT_SYMTAB) {
1192 			symtabindex = i;
1193 			symstrindex = shdr[i].sh_link;
1194 		} else if (shstrs != NULL && shdr[i].sh_name != 0 &&
1195 		    strcmp(shstrs + shdr[i].sh_name, ".ctors") == 0) {
1196 			/* Record relocated address and size of .ctors. */
1197 			lf->ctors_addr = mapbase + shdr[i].sh_addr - base_vaddr;
1198 			lf->ctors_size = shdr[i].sh_size;
1199 		}
1200 	}
1201 	if (symtabindex < 0 || symstrindex < 0)
1202 		goto nosyms;
1203 
1204 	symcnt = shdr[symtabindex].sh_size;
1205 	ef->symbase = malloc(symcnt, M_LINKER, M_WAITOK);
1206 	strcnt = shdr[symstrindex].sh_size;
1207 	ef->strbase = malloc(strcnt, M_LINKER, M_WAITOK);
1208 
1209 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1210 	    ef->symbase, symcnt, shdr[symtabindex].sh_offset,
1211 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1212 	    &resid, td);
1213 	if (error != 0)
1214 		goto out;
1215 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1216 	    ef->strbase, strcnt, shdr[symstrindex].sh_offset,
1217 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1218 	    &resid, td);
1219 	if (error != 0)
1220 		goto out;
1221 
1222 	ef->ddbsymcnt = symcnt / sizeof(Elf_Sym);
1223 	ef->ddbsymtab = (const Elf_Sym *)ef->symbase;
1224 	ef->ddbstrcnt = strcnt;
1225 	ef->ddbstrtab = ef->strbase;
1226 
1227 nosyms:
1228 	error = link_elf_link_common_finish(lf);
1229 	if (error != 0)
1230 		goto out;
1231 
1232 	*result = lf;
1233 
1234 out:
1235 	VOP_UNLOCK(nd.ni_vp, 0);
1236 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
1237 	if (error != 0 && lf != NULL)
1238 		linker_file_unload(lf, LINKER_UNLOAD_FORCE);
1239 	free(shdr, M_LINKER);
1240 	free(firstpage, M_LINKER);
1241 	free(shstrs, M_LINKER);
1242 
1243 	return (error);
1244 }
1245 
1246 Elf_Addr
1247 elf_relocaddr(linker_file_t lf, Elf_Addr x)
1248 {
1249 	elf_file_t ef;
1250 
1251 	KASSERT(lf->ops->cls == (kobj_class_t)&link_elf_class,
1252 	    ("elf_relocaddr: unexpected linker file %p", lf));
1253 
1254 	ef = (elf_file_t)lf;
1255 	if (x >= ef->pcpu_start && x < ef->pcpu_stop)
1256 		return ((x - ef->pcpu_start) + ef->pcpu_base);
1257 #ifdef VIMAGE
1258 	if (x >= ef->vnet_start && x < ef->vnet_stop)
1259 		return ((x - ef->vnet_start) + ef->vnet_base);
1260 #endif
1261 	return (x);
1262 }
1263 
1264 
1265 static void
1266 link_elf_unload_file(linker_file_t file)
1267 {
1268 	elf_file_t ef = (elf_file_t) file;
1269 
1270 	if (ef->pcpu_base != 0) {
1271 		dpcpu_free((void *)ef->pcpu_base,
1272 		    ef->pcpu_stop - ef->pcpu_start);
1273 		elf_set_delete(&set_pcpu_list, ef->pcpu_start);
1274 	}
1275 #ifdef VIMAGE
1276 	if (ef->vnet_base != 0) {
1277 		vnet_data_free((void *)ef->vnet_base,
1278 		    ef->vnet_stop - ef->vnet_start);
1279 		elf_set_delete(&set_vnet_list, ef->vnet_start);
1280 	}
1281 #endif
1282 #ifdef GDB
1283 	if (ef->gdb.l_ld != NULL) {
1284 		GDB_STATE(RT_DELETE);
1285 		free((void *)(uintptr_t)ef->gdb.l_name, M_LINKER);
1286 		link_elf_delete_gdb(&ef->gdb);
1287 		GDB_STATE(RT_CONSISTENT);
1288 	}
1289 #endif
1290 
1291 	/* Notify MD code that a module is being unloaded. */
1292 	elf_cpu_unload_file(file);
1293 
1294 	if (ef->preloaded) {
1295 		link_elf_unload_preload(file);
1296 		return;
1297 	}
1298 
1299 #ifdef SPARSE_MAPPING
1300 	if (ef->object != NULL) {
1301 		vm_map_remove(kernel_map, (vm_offset_t) ef->address,
1302 		    (vm_offset_t) ef->address
1303 		    + (ef->object->size << PAGE_SHIFT));
1304 	}
1305 #else
1306 	free(ef->address, M_LINKER);
1307 #endif
1308 	free(ef->symbase, M_LINKER);
1309 	free(ef->strbase, M_LINKER);
1310 	free(ef->ctftab, M_LINKER);
1311 	free(ef->ctfoff, M_LINKER);
1312 	free(ef->typoff, M_LINKER);
1313 }
1314 
1315 static void
1316 link_elf_unload_preload(linker_file_t file)
1317 {
1318 
1319 	if (file->pathname != NULL)
1320 		preload_delete_name(file->pathname);
1321 }
1322 
1323 static const char *
1324 symbol_name(elf_file_t ef, Elf_Size r_info)
1325 {
1326 	const Elf_Sym *ref;
1327 
1328 	if (ELF_R_SYM(r_info)) {
1329 		ref = ef->symtab + ELF_R_SYM(r_info);
1330 		return (ef->strtab + ref->st_name);
1331 	}
1332 	return (NULL);
1333 }
1334 
1335 static int
1336 symbol_type(elf_file_t ef, Elf_Size r_info)
1337 {
1338 	const Elf_Sym *ref;
1339 
1340 	if (ELF_R_SYM(r_info)) {
1341 		ref = ef->symtab + ELF_R_SYM(r_info);
1342 		return (ELF_ST_TYPE(ref->st_info));
1343 	}
1344 	return (STT_NOTYPE);
1345 }
1346 
1347 static int
1348 relocate_file1(elf_file_t ef, elf_lookup_fn lookup, elf_reloc_fn reloc,
1349     bool ifuncs)
1350 {
1351 	const Elf_Rel *rel;
1352 	const Elf_Rela *rela;
1353 	const char *symname;
1354 
1355 #define	APPLY_RELOCS(iter, tbl, tblsize, type) do {			\
1356 	for ((iter) = (tbl); (iter) != NULL &&				\
1357 	    (iter) < (tbl) + (tblsize) / sizeof(*(iter)); (iter)++) {	\
1358 		if ((symbol_type(ef, (iter)->r_info) ==			\
1359 		    STT_GNU_IFUNC ||					\
1360 		    elf_is_ifunc_reloc((iter)->r_info)) != ifuncs)	\
1361 			continue;					\
1362 		if (reloc(&ef->lf, (Elf_Addr)ef->address,		\
1363 		    (iter), (type), lookup)) {				\
1364 			symname = symbol_name(ef, (iter)->r_info);	\
1365 			printf("link_elf: symbol %s undefined\n",	\
1366 			    symname);					\
1367 			return (ENOENT);				\
1368 		}							\
1369 	}								\
1370 } while (0)
1371 
1372 	APPLY_RELOCS(rel, ef->rel, ef->relsize, ELF_RELOC_REL);
1373 	APPLY_RELOCS(rela, ef->rela, ef->relasize, ELF_RELOC_RELA);
1374 	APPLY_RELOCS(rel, ef->pltrel, ef->pltrelsize, ELF_RELOC_REL);
1375 	APPLY_RELOCS(rela, ef->pltrela, ef->pltrelasize, ELF_RELOC_RELA);
1376 
1377 #undef APPLY_RELOCS
1378 
1379 	return (0);
1380 }
1381 
1382 static int
1383 relocate_file(elf_file_t ef)
1384 {
1385 	int error;
1386 
1387 	error = relocate_file1(ef, elf_lookup, elf_reloc, false);
1388 	if (error == 0)
1389 		error = relocate_file1(ef, elf_lookup, elf_reloc, true);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * Hash function for symbol table lookup.  Don't even think about changing
1395  * this.  It is specified by the System V ABI.
1396  */
1397 static unsigned long
1398 elf_hash(const char *name)
1399 {
1400 	const unsigned char *p = (const unsigned char *) name;
1401 	unsigned long h = 0;
1402 	unsigned long g;
1403 
1404 	while (*p != '\0') {
1405 		h = (h << 4) + *p++;
1406 		if ((g = h & 0xf0000000) != 0)
1407 			h ^= g >> 24;
1408 		h &= ~g;
1409 	}
1410 	return (h);
1411 }
1412 
1413 static int
1414 link_elf_lookup_symbol(linker_file_t lf, const char *name, c_linker_sym_t *sym)
1415 {
1416 	elf_file_t ef = (elf_file_t) lf;
1417 	unsigned long symnum;
1418 	const Elf_Sym* symp;
1419 	const char *strp;
1420 	unsigned long hash;
1421 	int i;
1422 
1423 	/* If we don't have a hash, bail. */
1424 	if (ef->buckets == NULL || ef->nbuckets == 0) {
1425 		printf("link_elf_lookup_symbol: missing symbol hash table\n");
1426 		return (ENOENT);
1427 	}
1428 
1429 	/* First, search hashed global symbols */
1430 	hash = elf_hash(name);
1431 	symnum = ef->buckets[hash % ef->nbuckets];
1432 
1433 	while (symnum != STN_UNDEF) {
1434 		if (symnum >= ef->nchains) {
1435 			printf("%s: corrupt symbol table\n", __func__);
1436 			return (ENOENT);
1437 		}
1438 
1439 		symp = ef->symtab + symnum;
1440 		if (symp->st_name == 0) {
1441 			printf("%s: corrupt symbol table\n", __func__);
1442 			return (ENOENT);
1443 		}
1444 
1445 		strp = ef->strtab + symp->st_name;
1446 
1447 		if (strcmp(name, strp) == 0) {
1448 			if (symp->st_shndx != SHN_UNDEF ||
1449 			    (symp->st_value != 0 &&
1450 			    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1451 			    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC))) {
1452 				*sym = (c_linker_sym_t) symp;
1453 				return (0);
1454 			}
1455 			return (ENOENT);
1456 		}
1457 
1458 		symnum = ef->chains[symnum];
1459 	}
1460 
1461 	/* If we have not found it, look at the full table (if loaded) */
1462 	if (ef->symtab == ef->ddbsymtab)
1463 		return (ENOENT);
1464 
1465 	/* Exhaustive search */
1466 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1467 		strp = ef->ddbstrtab + symp->st_name;
1468 		if (strcmp(name, strp) == 0) {
1469 			if (symp->st_shndx != SHN_UNDEF ||
1470 			    (symp->st_value != 0 &&
1471 			    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1472 			    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC))) {
1473 				*sym = (c_linker_sym_t) symp;
1474 				return (0);
1475 			}
1476 			return (ENOENT);
1477 		}
1478 	}
1479 
1480 	return (ENOENT);
1481 }
1482 
1483 static int
1484 link_elf_symbol_values(linker_file_t lf, c_linker_sym_t sym,
1485     linker_symval_t *symval)
1486 {
1487 	elf_file_t ef;
1488 	const Elf_Sym *es;
1489 	caddr_t val;
1490 
1491 	ef = (elf_file_t)lf;
1492 	es = (const Elf_Sym *)sym;
1493 	if (es >= ef->symtab && es < (ef->symtab + ef->nchains)) {
1494 		symval->name = ef->strtab + es->st_name;
1495 		val = (caddr_t)ef->address + es->st_value;
1496 		if (ELF_ST_TYPE(es->st_info) == STT_GNU_IFUNC)
1497 			val = ((caddr_t (*)(void))val)();
1498 		symval->value = val;
1499 		symval->size = es->st_size;
1500 		return (0);
1501 	}
1502 	if (ef->symtab == ef->ddbsymtab)
1503 		return (ENOENT);
1504 	if (es >= ef->ddbsymtab && es < (ef->ddbsymtab + ef->ddbsymcnt)) {
1505 		symval->name = ef->ddbstrtab + es->st_name;
1506 		val = (caddr_t)ef->address + es->st_value;
1507 		if (ELF_ST_TYPE(es->st_info) == STT_GNU_IFUNC)
1508 			val = ((caddr_t (*)(void))val)();
1509 		symval->value = val;
1510 		symval->size = es->st_size;
1511 		return (0);
1512 	}
1513 	return (ENOENT);
1514 }
1515 
1516 static int
1517 link_elf_search_symbol(linker_file_t lf, caddr_t value,
1518     c_linker_sym_t *sym, long *diffp)
1519 {
1520 	elf_file_t ef = (elf_file_t) lf;
1521 	u_long off = (uintptr_t) (void *) value;
1522 	u_long diff = off;
1523 	u_long st_value;
1524 	const Elf_Sym* es;
1525 	const Elf_Sym* best = NULL;
1526 	int i;
1527 
1528 	for (i = 0, es = ef->ddbsymtab; i < ef->ddbsymcnt; i++, es++) {
1529 		if (es->st_name == 0)
1530 			continue;
1531 		st_value = es->st_value + (uintptr_t) (void *) ef->address;
1532 		if (off >= st_value) {
1533 			if (off - st_value < diff) {
1534 				diff = off - st_value;
1535 				best = es;
1536 				if (diff == 0)
1537 					break;
1538 			} else if (off - st_value == diff) {
1539 				best = es;
1540 			}
1541 		}
1542 	}
1543 	if (best == NULL)
1544 		*diffp = off;
1545 	else
1546 		*diffp = diff;
1547 	*sym = (c_linker_sym_t) best;
1548 
1549 	return (0);
1550 }
1551 
1552 /*
1553  * Look up a linker set on an ELF system.
1554  */
1555 static int
1556 link_elf_lookup_set(linker_file_t lf, const char *name,
1557     void ***startp, void ***stopp, int *countp)
1558 {
1559 	c_linker_sym_t sym;
1560 	linker_symval_t symval;
1561 	char *setsym;
1562 	void **start, **stop;
1563 	int len, error = 0, count;
1564 
1565 	len = strlen(name) + sizeof("__start_set_"); /* sizeof includes \0 */
1566 	setsym = malloc(len, M_LINKER, M_WAITOK);
1567 
1568 	/* get address of first entry */
1569 	snprintf(setsym, len, "%s%s", "__start_set_", name);
1570 	error = link_elf_lookup_symbol(lf, setsym, &sym);
1571 	if (error != 0)
1572 		goto out;
1573 	link_elf_symbol_values(lf, sym, &symval);
1574 	if (symval.value == 0) {
1575 		error = ESRCH;
1576 		goto out;
1577 	}
1578 	start = (void **)symval.value;
1579 
1580 	/* get address of last entry */
1581 	snprintf(setsym, len, "%s%s", "__stop_set_", name);
1582 	error = link_elf_lookup_symbol(lf, setsym, &sym);
1583 	if (error != 0)
1584 		goto out;
1585 	link_elf_symbol_values(lf, sym, &symval);
1586 	if (symval.value == 0) {
1587 		error = ESRCH;
1588 		goto out;
1589 	}
1590 	stop = (void **)symval.value;
1591 
1592 	/* and the number of entries */
1593 	count = stop - start;
1594 
1595 	/* and copy out */
1596 	if (startp != NULL)
1597 		*startp = start;
1598 	if (stopp != NULL)
1599 		*stopp = stop;
1600 	if (countp != NULL)
1601 		*countp = count;
1602 
1603 out:
1604 	free(setsym, M_LINKER);
1605 	return (error);
1606 }
1607 
1608 static int
1609 link_elf_each_function_name(linker_file_t file,
1610   int (*callback)(const char *, void *), void *opaque)
1611 {
1612 	elf_file_t ef = (elf_file_t)file;
1613 	const Elf_Sym *symp;
1614 	int i, error;
1615 
1616 	/* Exhaustive search */
1617 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1618 		if (symp->st_value != 0 &&
1619 		    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1620 		    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
1621 			error = callback(ef->ddbstrtab + symp->st_name, opaque);
1622 			if (error != 0)
1623 				return (error);
1624 		}
1625 	}
1626 	return (0);
1627 }
1628 
1629 static int
1630 link_elf_each_function_nameval(linker_file_t file,
1631     linker_function_nameval_callback_t callback, void *opaque)
1632 {
1633 	linker_symval_t symval;
1634 	elf_file_t ef = (elf_file_t)file;
1635 	const Elf_Sym* symp;
1636 	int i, error;
1637 
1638 	/* Exhaustive search */
1639 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1640 		if (symp->st_value != 0 &&
1641 		    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1642 		    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
1643 			error = link_elf_symbol_values(file,
1644 			    (c_linker_sym_t) symp, &symval);
1645 			if (error != 0)
1646 				return (error);
1647 			error = callback(file, i, &symval, opaque);
1648 			if (error != 0)
1649 				return (error);
1650 		}
1651 	}
1652 	return (0);
1653 }
1654 
1655 const Elf_Sym *
1656 elf_get_sym(linker_file_t lf, Elf_Size symidx)
1657 {
1658 	elf_file_t ef = (elf_file_t)lf;
1659 
1660 	if (symidx >= ef->nchains)
1661 		return (NULL);
1662 	return (ef->symtab + symidx);
1663 }
1664 
1665 const char *
1666 elf_get_symname(linker_file_t lf, Elf_Size symidx)
1667 {
1668 	elf_file_t ef = (elf_file_t)lf;
1669 	const Elf_Sym *sym;
1670 
1671 	if (symidx >= ef->nchains)
1672 		return (NULL);
1673 	sym = ef->symtab + symidx;
1674 	return (ef->strtab + sym->st_name);
1675 }
1676 
1677 /*
1678  * Symbol lookup function that can be used when the symbol index is known (ie
1679  * in relocations). It uses the symbol index instead of doing a fully fledged
1680  * hash table based lookup when such is valid. For example for local symbols.
1681  * This is not only more efficient, it's also more correct. It's not always
1682  * the case that the symbol can be found through the hash table.
1683  */
1684 static int
1685 elf_lookup(linker_file_t lf, Elf_Size symidx, int deps, Elf_Addr *res)
1686 {
1687 	elf_file_t ef = (elf_file_t)lf;
1688 	const Elf_Sym *sym;
1689 	const char *symbol;
1690 	Elf_Addr addr, start, base;
1691 
1692 	/* Don't even try to lookup the symbol if the index is bogus. */
1693 	if (symidx >= ef->nchains) {
1694 		*res = 0;
1695 		return (EINVAL);
1696 	}
1697 
1698 	sym = ef->symtab + symidx;
1699 
1700 	/*
1701 	 * Don't do a full lookup when the symbol is local. It may even
1702 	 * fail because it may not be found through the hash table.
1703 	 */
1704 	if (ELF_ST_BIND(sym->st_info) == STB_LOCAL) {
1705 		/* Force lookup failure when we have an insanity. */
1706 		if (sym->st_shndx == SHN_UNDEF || sym->st_value == 0) {
1707 			*res = 0;
1708 			return (EINVAL);
1709 		}
1710 		*res = ((Elf_Addr)ef->address + sym->st_value);
1711 		return (0);
1712 	}
1713 
1714 	/*
1715 	 * XXX we can avoid doing a hash table based lookup for global
1716 	 * symbols as well. This however is not always valid, so we'll
1717 	 * just do it the hard way for now. Performance tweaks can
1718 	 * always be added.
1719 	 */
1720 
1721 	symbol = ef->strtab + sym->st_name;
1722 
1723 	/* Force a lookup failure if the symbol name is bogus. */
1724 	if (*symbol == 0) {
1725 		*res = 0;
1726 		return (EINVAL);
1727 	}
1728 
1729 	addr = ((Elf_Addr)linker_file_lookup_symbol(lf, symbol, deps));
1730 	if (addr == 0 && ELF_ST_BIND(sym->st_info) != STB_WEAK) {
1731 		*res = 0;
1732 		return (EINVAL);
1733 	}
1734 
1735 	if (elf_set_find(&set_pcpu_list, addr, &start, &base))
1736 		addr = addr - start + base;
1737 #ifdef VIMAGE
1738 	else if (elf_set_find(&set_vnet_list, addr, &start, &base))
1739 		addr = addr - start + base;
1740 #endif
1741 	*res = addr;
1742 	return (0);
1743 }
1744 
1745 static void
1746 link_elf_reloc_local(linker_file_t lf)
1747 {
1748 	const Elf_Rel *rellim;
1749 	const Elf_Rel *rel;
1750 	const Elf_Rela *relalim;
1751 	const Elf_Rela *rela;
1752 	elf_file_t ef = (elf_file_t)lf;
1753 
1754 	/* Perform relocations without addend if there are any: */
1755 	if ((rel = ef->rel) != NULL) {
1756 		rellim = (const Elf_Rel *)((const char *)ef->rel + ef->relsize);
1757 		while (rel < rellim) {
1758 			elf_reloc_local(lf, (Elf_Addr)ef->address, rel,
1759 			    ELF_RELOC_REL, elf_lookup);
1760 			rel++;
1761 		}
1762 	}
1763 
1764 	/* Perform relocations with addend if there are any: */
1765 	if ((rela = ef->rela) != NULL) {
1766 		relalim = (const Elf_Rela *)
1767 		    ((const char *)ef->rela + ef->relasize);
1768 		while (rela < relalim) {
1769 			elf_reloc_local(lf, (Elf_Addr)ef->address, rela,
1770 			    ELF_RELOC_RELA, elf_lookup);
1771 			rela++;
1772 		}
1773 	}
1774 }
1775 
1776 static long
1777 link_elf_symtab_get(linker_file_t lf, const Elf_Sym **symtab)
1778 {
1779 	elf_file_t ef = (elf_file_t)lf;
1780 
1781 	*symtab = ef->ddbsymtab;
1782 
1783 	if (*symtab == NULL)
1784 		return (0);
1785 
1786 	return (ef->ddbsymcnt);
1787 }
1788 
1789 static long
1790 link_elf_strtab_get(linker_file_t lf, caddr_t *strtab)
1791 {
1792 	elf_file_t ef = (elf_file_t)lf;
1793 
1794 	*strtab = ef->ddbstrtab;
1795 
1796 	if (*strtab == NULL)
1797 		return (0);
1798 
1799 	return (ef->ddbstrcnt);
1800 }
1801 
1802 #if defined(__i386__) || defined(__amd64__) || defined(__aarch64__)
1803 /*
1804  * Use this lookup routine when performing relocations early during boot.
1805  * The generic lookup routine depends on kobj, which is not initialized
1806  * at that point.
1807  */
1808 static int
1809 elf_lookup_ifunc(linker_file_t lf, Elf_Size symidx, int deps __unused,
1810     Elf_Addr *res)
1811 {
1812 	elf_file_t ef;
1813 	const Elf_Sym *symp;
1814 	caddr_t val;
1815 
1816 	ef = (elf_file_t)lf;
1817 	symp = ef->symtab + symidx;
1818 	if (ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC) {
1819 		val = (caddr_t)ef->address + symp->st_value;
1820 		*res = ((Elf_Addr (*)(void))val)();
1821 		return (0);
1822 	}
1823 	return (ENOENT);
1824 }
1825 
1826 void
1827 link_elf_ireloc(caddr_t kmdp)
1828 {
1829 	struct elf_file eff;
1830 	elf_file_t ef;
1831 
1832 	ef = &eff;
1833 
1834 	bzero_early(ef, sizeof(*ef));
1835 
1836 	ef->modptr = kmdp;
1837 	ef->dynamic = (Elf_Dyn *)&_DYNAMIC;
1838 	parse_dynamic(ef);
1839 	ef->address = 0;
1840 	link_elf_preload_parse_symbols(ef);
1841 	relocate_file1(ef, elf_lookup_ifunc, elf_reloc, true);
1842 }
1843 #endif
1844