xref: /freebsd/sys/kern/link_elf.c (revision f01edb6f5d55fe093f8ab9c5da2b1707af698a56)
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 	{ 0, 0 }
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 static int
740 link_elf_link_preload(linker_class_t cls,
741     const char* filename, linker_file_t *result)
742 {
743 	Elf_Addr *ctors_addrp;
744 	Elf_Size *ctors_sizep;
745 	caddr_t modptr, baseptr, sizeptr, dynptr;
746 	char *type;
747 	elf_file_t ef;
748 	linker_file_t lf;
749 	int error;
750 	vm_offset_t dp;
751 
752 	/* Look to see if we have the file preloaded */
753 	modptr = preload_search_by_name(filename);
754 	if (modptr == NULL)
755 		return (ENOENT);
756 
757 	type = (char *)preload_search_info(modptr, MODINFO_TYPE);
758 	baseptr = preload_search_info(modptr, MODINFO_ADDR);
759 	sizeptr = preload_search_info(modptr, MODINFO_SIZE);
760 	dynptr = preload_search_info(modptr,
761 	    MODINFO_METADATA | MODINFOMD_DYNAMIC);
762 	if (type == NULL ||
763 	    (strcmp(type, "elf" __XSTRING(__ELF_WORD_SIZE) " module") != 0 &&
764 	     strcmp(type, "elf module") != 0))
765 		return (EFTYPE);
766 	if (baseptr == NULL || sizeptr == NULL || dynptr == NULL)
767 		return (EINVAL);
768 
769 	lf = linker_make_file(filename, &link_elf_class);
770 	if (lf == NULL)
771 		return (ENOMEM);
772 
773 	ef = (elf_file_t) lf;
774 	ef->preloaded = 1;
775 	ef->modptr = modptr;
776 	ef->address = *(caddr_t *)baseptr;
777 #ifdef SPARSE_MAPPING
778 	ef->object = NULL;
779 #endif
780 	dp = (vm_offset_t)ef->address + *(vm_offset_t *)dynptr;
781 	ef->dynamic = (Elf_Dyn *)dp;
782 	lf->address = ef->address;
783 	lf->size = *(size_t *)sizeptr;
784 
785 	ctors_addrp = (Elf_Addr *)preload_search_info(modptr,
786 	    MODINFO_METADATA | MODINFOMD_CTORS_ADDR);
787 	ctors_sizep = (Elf_Size *)preload_search_info(modptr,
788 	    MODINFO_METADATA | MODINFOMD_CTORS_SIZE);
789 	if (ctors_addrp != NULL && ctors_sizep != NULL) {
790 		lf->ctors_addr = ef->address + *ctors_addrp;
791 		lf->ctors_size = *ctors_sizep;
792 	}
793 
794 	error = parse_dynamic(ef);
795 	if (error == 0)
796 		error = parse_dpcpu(ef);
797 #ifdef VIMAGE
798 	if (error == 0)
799 		error = parse_vnet(ef);
800 #endif
801 	if (error != 0) {
802 		linker_file_unload(lf, LINKER_UNLOAD_FORCE);
803 		return (error);
804 	}
805 	link_elf_reloc_local(lf);
806 	*result = lf;
807 	return (0);
808 }
809 
810 static int
811 link_elf_link_preload_finish(linker_file_t lf)
812 {
813 	elf_file_t ef;
814 	int error;
815 
816 	ef = (elf_file_t) lf;
817 	error = relocate_file(ef);
818 	if (error != 0)
819 		return (error);
820 	(void)link_elf_preload_parse_symbols(ef);
821 
822 	return (link_elf_link_common_finish(lf));
823 }
824 
825 static int
826 link_elf_load_file(linker_class_t cls, const char* filename,
827     linker_file_t* result)
828 {
829 	struct nameidata nd;
830 	struct thread* td = curthread;	/* XXX */
831 	Elf_Ehdr *hdr;
832 	caddr_t firstpage, segbase;
833 	int nbytes, i;
834 	Elf_Phdr *phdr;
835 	Elf_Phdr *phlimit;
836 	Elf_Phdr *segs[MAXSEGS];
837 	int nsegs;
838 	Elf_Phdr *phdyn;
839 	caddr_t mapbase;
840 	size_t mapsize;
841 	Elf_Addr base_vaddr;
842 	Elf_Addr base_vlimit;
843 	int error = 0;
844 	ssize_t resid;
845 	int flags;
846 	elf_file_t ef;
847 	linker_file_t lf;
848 	Elf_Shdr *shdr;
849 	int symtabindex;
850 	int symstrindex;
851 	int shstrindex;
852 	int symcnt;
853 	int strcnt;
854 	char *shstrs;
855 
856 	shdr = NULL;
857 	lf = NULL;
858 	shstrs = NULL;
859 
860 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, filename, td);
861 	flags = FREAD;
862 	error = vn_open(&nd, &flags, 0, NULL);
863 	if (error != 0)
864 		return (error);
865 	NDFREE(&nd, NDF_ONLY_PNBUF);
866 	if (nd.ni_vp->v_type != VREG) {
867 		error = ENOEXEC;
868 		firstpage = NULL;
869 		goto out;
870 	}
871 #ifdef MAC
872 	error = mac_kld_check_load(curthread->td_ucred, nd.ni_vp);
873 	if (error != 0) {
874 		firstpage = NULL;
875 		goto out;
876 	}
877 #endif
878 
879 	/*
880 	 * Read the elf header from the file.
881 	 */
882 	firstpage = malloc(PAGE_SIZE, M_LINKER, M_WAITOK);
883 	hdr = (Elf_Ehdr *)firstpage;
884 	error = vn_rdwr(UIO_READ, nd.ni_vp, firstpage, PAGE_SIZE, 0,
885 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
886 	    &resid, td);
887 	nbytes = PAGE_SIZE - resid;
888 	if (error != 0)
889 		goto out;
890 
891 	if (!IS_ELF(*hdr)) {
892 		error = ENOEXEC;
893 		goto out;
894 	}
895 
896 	if (hdr->e_ident[EI_CLASS] != ELF_TARG_CLASS ||
897 	    hdr->e_ident[EI_DATA] != ELF_TARG_DATA) {
898 		link_elf_error(filename, "Unsupported file layout");
899 		error = ENOEXEC;
900 		goto out;
901 	}
902 	if (hdr->e_ident[EI_VERSION] != EV_CURRENT ||
903 	    hdr->e_version != EV_CURRENT) {
904 		link_elf_error(filename, "Unsupported file version");
905 		error = ENOEXEC;
906 		goto out;
907 	}
908 	if (hdr->e_type != ET_EXEC && hdr->e_type != ET_DYN) {
909 		error = ENOSYS;
910 		goto out;
911 	}
912 	if (hdr->e_machine != ELF_TARG_MACH) {
913 		link_elf_error(filename, "Unsupported machine");
914 		error = ENOEXEC;
915 		goto out;
916 	}
917 
918 	/*
919 	 * We rely on the program header being in the first page.
920 	 * This is not strictly required by the ABI specification, but
921 	 * it seems to always true in practice.  And, it simplifies
922 	 * things considerably.
923 	 */
924 	if (!((hdr->e_phentsize == sizeof(Elf_Phdr)) &&
925 	      (hdr->e_phoff + hdr->e_phnum*sizeof(Elf_Phdr) <= PAGE_SIZE) &&
926 	      (hdr->e_phoff + hdr->e_phnum*sizeof(Elf_Phdr) <= nbytes)))
927 		link_elf_error(filename, "Unreadable program headers");
928 
929 	/*
930 	 * Scan the program header entries, and save key information.
931 	 *
932 	 * We rely on there being exactly two load segments, text and data,
933 	 * in that order.
934 	 */
935 	phdr = (Elf_Phdr *) (firstpage + hdr->e_phoff);
936 	phlimit = phdr + hdr->e_phnum;
937 	nsegs = 0;
938 	phdyn = NULL;
939 	while (phdr < phlimit) {
940 		switch (phdr->p_type) {
941 		case PT_LOAD:
942 			if (nsegs == MAXSEGS) {
943 				link_elf_error(filename, "Too many sections");
944 				error = ENOEXEC;
945 				goto out;
946 			}
947 			/*
948 			 * XXX: We just trust they come in right order ??
949 			 */
950 			segs[nsegs] = phdr;
951 			++nsegs;
952 			break;
953 
954 		case PT_DYNAMIC:
955 			phdyn = phdr;
956 			break;
957 
958 		case PT_INTERP:
959 			error = ENOSYS;
960 			goto out;
961 		}
962 
963 		++phdr;
964 	}
965 	if (phdyn == NULL) {
966 		link_elf_error(filename, "Object is not dynamically-linked");
967 		error = ENOEXEC;
968 		goto out;
969 	}
970 	if (nsegs == 0) {
971 		link_elf_error(filename, "No sections");
972 		error = ENOEXEC;
973 		goto out;
974 	}
975 
976 	/*
977 	 * Allocate the entire address space of the object, to stake
978 	 * out our contiguous region, and to establish the base
979 	 * address for relocation.
980 	 */
981 	base_vaddr = trunc_page(segs[0]->p_vaddr);
982 	base_vlimit = round_page(segs[nsegs - 1]->p_vaddr +
983 	    segs[nsegs - 1]->p_memsz);
984 	mapsize = base_vlimit - base_vaddr;
985 
986 	lf = linker_make_file(filename, &link_elf_class);
987 	if (lf == NULL) {
988 		error = ENOMEM;
989 		goto out;
990 	}
991 
992 	ef = (elf_file_t) lf;
993 #ifdef SPARSE_MAPPING
994 	ef->object = vm_object_allocate(OBJT_PHYS, atop(mapsize));
995 	if (ef->object == NULL) {
996 		error = ENOMEM;
997 		goto out;
998 	}
999 #ifdef __amd64__
1000 	mapbase = (caddr_t)KERNBASE;
1001 #else
1002 	mapbase = (caddr_t)vm_map_min(kernel_map);
1003 #endif
1004 	/*
1005 	 * Mapping protections are downgraded after relocation processing.
1006 	 */
1007 	error = vm_map_find(kernel_map, ef->object, 0,
1008 	    (vm_offset_t *)&mapbase, mapsize, 0, VMFS_OPTIMAL_SPACE,
1009 	    VM_PROT_ALL, VM_PROT_ALL, 0);
1010 	if (error != 0) {
1011 		vm_object_deallocate(ef->object);
1012 		ef->object = NULL;
1013 		goto out;
1014 	}
1015 #else
1016 	mapbase = malloc(mapsize, M_LINKER, M_EXEC | M_WAITOK);
1017 #endif
1018 	ef->address = mapbase;
1019 
1020 	/*
1021 	 * Read the text and data sections and zero the bss.
1022 	 */
1023 	for (i = 0; i < nsegs; i++) {
1024 		segbase = mapbase + segs[i]->p_vaddr - base_vaddr;
1025 
1026 #ifdef SPARSE_MAPPING
1027 		/*
1028 		 * Consecutive segments may have different mapping permissions,
1029 		 * so be strict and verify that their mappings do not overlap.
1030 		 */
1031 		if (((vm_offset_t)segbase & PAGE_MASK) != 0) {
1032 			error = EINVAL;
1033 			goto out;
1034 		}
1035 
1036 		error = vm_map_wire(kernel_map,
1037 		    (vm_offset_t)segbase,
1038 		    (vm_offset_t)segbase + round_page(segs[i]->p_memsz),
1039 		    VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
1040 		if (error != KERN_SUCCESS) {
1041 			error = ENOMEM;
1042 			goto out;
1043 		}
1044 #endif
1045 
1046 		error = vn_rdwr(UIO_READ, nd.ni_vp,
1047 		    segbase, segs[i]->p_filesz, segs[i]->p_offset,
1048 		    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1049 		    &resid, td);
1050 		if (error != 0)
1051 			goto out;
1052 		bzero(segbase + segs[i]->p_filesz,
1053 		    segs[i]->p_memsz - segs[i]->p_filesz);
1054 	}
1055 
1056 #ifdef GPROF
1057 	/* Update profiling information with the new text segment. */
1058 	mtx_lock(&Giant);
1059 	kmupetext((uintfptr_t)(mapbase + segs[0]->p_vaddr - base_vaddr +
1060 	    segs[0]->p_memsz));
1061 	mtx_unlock(&Giant);
1062 #endif
1063 
1064 	ef->dynamic = (Elf_Dyn *) (mapbase + phdyn->p_vaddr - base_vaddr);
1065 
1066 	lf->address = ef->address;
1067 	lf->size = mapsize;
1068 
1069 	error = parse_dynamic(ef);
1070 	if (error != 0)
1071 		goto out;
1072 	error = parse_dpcpu(ef);
1073 	if (error != 0)
1074 		goto out;
1075 #ifdef VIMAGE
1076 	error = parse_vnet(ef);
1077 	if (error != 0)
1078 		goto out;
1079 #endif
1080 	link_elf_reloc_local(lf);
1081 
1082 	VOP_UNLOCK(nd.ni_vp, 0);
1083 	error = linker_load_dependencies(lf);
1084 	vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1085 	if (error != 0)
1086 		goto out;
1087 	error = relocate_file(ef);
1088 	if (error != 0)
1089 		goto out;
1090 
1091 #ifdef SPARSE_MAPPING
1092 	/*
1093 	 * Downgrade permissions on text segment mappings now that relocation
1094 	 * processing is complete.  Restrict permissions on read-only segments.
1095 	 */
1096 	for (i = 0; i < nsegs; i++) {
1097 		vm_prot_t prot;
1098 
1099 		if (segs[i]->p_type != PT_LOAD)
1100 			continue;
1101 
1102 		prot = VM_PROT_READ;
1103 		if ((segs[i]->p_flags & PF_W) != 0)
1104 			prot |= VM_PROT_WRITE;
1105 		if ((segs[i]->p_flags & PF_X) != 0)
1106 			prot |= VM_PROT_EXECUTE;
1107 		segbase = mapbase + segs[i]->p_vaddr - base_vaddr;
1108 		error = vm_map_protect(kernel_map,
1109 		    (vm_offset_t)segbase,
1110 		    (vm_offset_t)segbase + round_page(segs[i]->p_memsz),
1111 		    prot, FALSE);
1112 		if (error != KERN_SUCCESS) {
1113 			error = ENOMEM;
1114 			goto out;
1115 		}
1116 	}
1117 #endif
1118 
1119 	/*
1120 	 * Try and load the symbol table if it's present.  (you can
1121 	 * strip it!)
1122 	 */
1123 	nbytes = hdr->e_shnum * hdr->e_shentsize;
1124 	if (nbytes == 0 || hdr->e_shoff == 0)
1125 		goto nosyms;
1126 	shdr = malloc(nbytes, M_LINKER, M_WAITOK | M_ZERO);
1127 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1128 	    (caddr_t)shdr, nbytes, hdr->e_shoff,
1129 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1130 	    &resid, td);
1131 	if (error != 0)
1132 		goto out;
1133 
1134 	/* Read section string table */
1135 	shstrindex = hdr->e_shstrndx;
1136 	if (shstrindex != 0 && shdr[shstrindex].sh_type == SHT_STRTAB &&
1137 	    shdr[shstrindex].sh_size != 0) {
1138 		nbytes = shdr[shstrindex].sh_size;
1139 		shstrs = malloc(nbytes, M_LINKER, M_WAITOK | M_ZERO);
1140 		error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)shstrs, nbytes,
1141 		    shdr[shstrindex].sh_offset, UIO_SYSSPACE, IO_NODELOCKED,
1142 		    td->td_ucred, NOCRED, &resid, td);
1143 		if (error)
1144 			goto out;
1145 	}
1146 
1147 	symtabindex = -1;
1148 	symstrindex = -1;
1149 	for (i = 0; i < hdr->e_shnum; i++) {
1150 		if (shdr[i].sh_type == SHT_SYMTAB) {
1151 			symtabindex = i;
1152 			symstrindex = shdr[i].sh_link;
1153 		} else if (shstrs != NULL && shdr[i].sh_name != 0 &&
1154 		    strcmp(shstrs + shdr[i].sh_name, ".ctors") == 0) {
1155 			/* Record relocated address and size of .ctors. */
1156 			lf->ctors_addr = mapbase + shdr[i].sh_addr - base_vaddr;
1157 			lf->ctors_size = shdr[i].sh_size;
1158 		}
1159 	}
1160 	if (symtabindex < 0 || symstrindex < 0)
1161 		goto nosyms;
1162 
1163 	symcnt = shdr[symtabindex].sh_size;
1164 	ef->symbase = malloc(symcnt, M_LINKER, M_WAITOK);
1165 	strcnt = shdr[symstrindex].sh_size;
1166 	ef->strbase = malloc(strcnt, M_LINKER, M_WAITOK);
1167 
1168 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1169 	    ef->symbase, symcnt, shdr[symtabindex].sh_offset,
1170 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1171 	    &resid, td);
1172 	if (error != 0)
1173 		goto out;
1174 	error = vn_rdwr(UIO_READ, nd.ni_vp,
1175 	    ef->strbase, strcnt, shdr[symstrindex].sh_offset,
1176 	    UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED,
1177 	    &resid, td);
1178 	if (error != 0)
1179 		goto out;
1180 
1181 	ef->ddbsymcnt = symcnt / sizeof(Elf_Sym);
1182 	ef->ddbsymtab = (const Elf_Sym *)ef->symbase;
1183 	ef->ddbstrcnt = strcnt;
1184 	ef->ddbstrtab = ef->strbase;
1185 
1186 nosyms:
1187 	error = link_elf_link_common_finish(lf);
1188 	if (error != 0)
1189 		goto out;
1190 
1191 	*result = lf;
1192 
1193 out:
1194 	VOP_UNLOCK(nd.ni_vp, 0);
1195 	vn_close(nd.ni_vp, FREAD, td->td_ucred, td);
1196 	if (error != 0 && lf != NULL)
1197 		linker_file_unload(lf, LINKER_UNLOAD_FORCE);
1198 	free(shdr, M_LINKER);
1199 	free(firstpage, M_LINKER);
1200 	free(shstrs, M_LINKER);
1201 
1202 	return (error);
1203 }
1204 
1205 Elf_Addr
1206 elf_relocaddr(linker_file_t lf, Elf_Addr x)
1207 {
1208 	elf_file_t ef;
1209 
1210 	KASSERT(lf->ops->cls == (kobj_class_t)&link_elf_class,
1211 	    ("elf_relocaddr: unexpected linker file %p", lf));
1212 
1213 	ef = (elf_file_t)lf;
1214 	if (x >= ef->pcpu_start && x < ef->pcpu_stop)
1215 		return ((x - ef->pcpu_start) + ef->pcpu_base);
1216 #ifdef VIMAGE
1217 	if (x >= ef->vnet_start && x < ef->vnet_stop)
1218 		return ((x - ef->vnet_start) + ef->vnet_base);
1219 #endif
1220 	return (x);
1221 }
1222 
1223 
1224 static void
1225 link_elf_unload_file(linker_file_t file)
1226 {
1227 	elf_file_t ef = (elf_file_t) file;
1228 
1229 	if (ef->pcpu_base != 0) {
1230 		dpcpu_free((void *)ef->pcpu_base,
1231 		    ef->pcpu_stop - ef->pcpu_start);
1232 		elf_set_delete(&set_pcpu_list, ef->pcpu_start);
1233 	}
1234 #ifdef VIMAGE
1235 	if (ef->vnet_base != 0) {
1236 		vnet_data_free((void *)ef->vnet_base,
1237 		    ef->vnet_stop - ef->vnet_start);
1238 		elf_set_delete(&set_vnet_list, ef->vnet_start);
1239 	}
1240 #endif
1241 #ifdef GDB
1242 	if (ef->gdb.l_ld != NULL) {
1243 		GDB_STATE(RT_DELETE);
1244 		free((void *)(uintptr_t)ef->gdb.l_name, M_LINKER);
1245 		link_elf_delete_gdb(&ef->gdb);
1246 		GDB_STATE(RT_CONSISTENT);
1247 	}
1248 #endif
1249 
1250 	/* Notify MD code that a module is being unloaded. */
1251 	elf_cpu_unload_file(file);
1252 
1253 	if (ef->preloaded) {
1254 		link_elf_unload_preload(file);
1255 		return;
1256 	}
1257 
1258 #ifdef SPARSE_MAPPING
1259 	if (ef->object != NULL) {
1260 		vm_map_remove(kernel_map, (vm_offset_t) ef->address,
1261 		    (vm_offset_t) ef->address
1262 		    + (ef->object->size << PAGE_SHIFT));
1263 	}
1264 #else
1265 	free(ef->address, M_LINKER);
1266 #endif
1267 	free(ef->symbase, M_LINKER);
1268 	free(ef->strbase, M_LINKER);
1269 	free(ef->ctftab, M_LINKER);
1270 	free(ef->ctfoff, M_LINKER);
1271 	free(ef->typoff, M_LINKER);
1272 }
1273 
1274 static void
1275 link_elf_unload_preload(linker_file_t file)
1276 {
1277 	if (file->pathname != NULL)
1278 		preload_delete_name(file->pathname);
1279 }
1280 
1281 static const char *
1282 symbol_name(elf_file_t ef, Elf_Size r_info)
1283 {
1284 	const Elf_Sym *ref;
1285 
1286 	if (ELF_R_SYM(r_info)) {
1287 		ref = ef->symtab + ELF_R_SYM(r_info);
1288 		return (ef->strtab + ref->st_name);
1289 	}
1290 	return (NULL);
1291 }
1292 
1293 static int
1294 symbol_type(elf_file_t ef, Elf_Size r_info)
1295 {
1296 	const Elf_Sym *ref;
1297 
1298 	if (ELF_R_SYM(r_info)) {
1299 		ref = ef->symtab + ELF_R_SYM(r_info);
1300 		return (ELF_ST_TYPE(ref->st_info));
1301 	}
1302 	return (STT_NOTYPE);
1303 }
1304 
1305 static int
1306 relocate_file1(elf_file_t ef, elf_lookup_fn lookup, elf_reloc_fn reloc,
1307     bool ifuncs)
1308 {
1309 	const Elf_Rel *rel;
1310 	const Elf_Rela *rela;
1311 	const char *symname;
1312 
1313 #define	APPLY_RELOCS(iter, tbl, tblsize, type) do {			\
1314 	for ((iter) = (tbl); (iter) != NULL &&				\
1315 	    (iter) < (tbl) + (tblsize) / sizeof(*(iter)); (iter)++) {	\
1316 		if ((symbol_type(ef, (iter)->r_info) ==			\
1317 		    STT_GNU_IFUNC ||					\
1318 		    elf_is_ifunc_reloc((iter)->r_info)) != ifuncs)	\
1319 			continue;					\
1320 		if (reloc(&ef->lf, (Elf_Addr)ef->address,		\
1321 		    (iter), (type), lookup)) {				\
1322 			symname = symbol_name(ef, (iter)->r_info);	\
1323 			printf("link_elf: symbol %s undefined\n",	\
1324 			    symname);					\
1325 			return (ENOENT);				\
1326 		}							\
1327 	}								\
1328 } while (0)
1329 
1330 	APPLY_RELOCS(rel, ef->rel, ef->relsize, ELF_RELOC_REL);
1331 	APPLY_RELOCS(rela, ef->rela, ef->relasize, ELF_RELOC_RELA);
1332 	APPLY_RELOCS(rel, ef->pltrel, ef->pltrelsize, ELF_RELOC_REL);
1333 	APPLY_RELOCS(rela, ef->pltrela, ef->pltrelasize, ELF_RELOC_RELA);
1334 
1335 #undef APPLY_RELOCS
1336 
1337 	return (0);
1338 }
1339 
1340 static int
1341 relocate_file(elf_file_t ef)
1342 {
1343 	int error;
1344 
1345 	error = relocate_file1(ef, elf_lookup, elf_reloc, false);
1346 	if (error == 0)
1347 		error = relocate_file1(ef, elf_lookup, elf_reloc, true);
1348 	return (error);
1349 }
1350 
1351 /*
1352  * Hash function for symbol table lookup.  Don't even think about changing
1353  * this.  It is specified by the System V ABI.
1354  */
1355 static unsigned long
1356 elf_hash(const char *name)
1357 {
1358 	const unsigned char *p = (const unsigned char *) name;
1359 	unsigned long h = 0;
1360 	unsigned long g;
1361 
1362 	while (*p != '\0') {
1363 		h = (h << 4) + *p++;
1364 		if ((g = h & 0xf0000000) != 0)
1365 			h ^= g >> 24;
1366 		h &= ~g;
1367 	}
1368 	return (h);
1369 }
1370 
1371 static int
1372 link_elf_lookup_symbol(linker_file_t lf, const char *name, c_linker_sym_t *sym)
1373 {
1374 	elf_file_t ef = (elf_file_t) lf;
1375 	unsigned long symnum;
1376 	const Elf_Sym* symp;
1377 	const char *strp;
1378 	unsigned long hash;
1379 	int i;
1380 
1381 	/* If we don't have a hash, bail. */
1382 	if (ef->buckets == NULL || ef->nbuckets == 0) {
1383 		printf("link_elf_lookup_symbol: missing symbol hash table\n");
1384 		return (ENOENT);
1385 	}
1386 
1387 	/* First, search hashed global symbols */
1388 	hash = elf_hash(name);
1389 	symnum = ef->buckets[hash % ef->nbuckets];
1390 
1391 	while (symnum != STN_UNDEF) {
1392 		if (symnum >= ef->nchains) {
1393 			printf("%s: corrupt symbol table\n", __func__);
1394 			return (ENOENT);
1395 		}
1396 
1397 		symp = ef->symtab + symnum;
1398 		if (symp->st_name == 0) {
1399 			printf("%s: corrupt symbol table\n", __func__);
1400 			return (ENOENT);
1401 		}
1402 
1403 		strp = ef->strtab + symp->st_name;
1404 
1405 		if (strcmp(name, strp) == 0) {
1406 			if (symp->st_shndx != SHN_UNDEF ||
1407 			    (symp->st_value != 0 &&
1408 			    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1409 			    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC))) {
1410 				*sym = (c_linker_sym_t) symp;
1411 				return (0);
1412 			}
1413 			return (ENOENT);
1414 		}
1415 
1416 		symnum = ef->chains[symnum];
1417 	}
1418 
1419 	/* If we have not found it, look at the full table (if loaded) */
1420 	if (ef->symtab == ef->ddbsymtab)
1421 		return (ENOENT);
1422 
1423 	/* Exhaustive search */
1424 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1425 		strp = ef->ddbstrtab + symp->st_name;
1426 		if (strcmp(name, strp) == 0) {
1427 			if (symp->st_shndx != SHN_UNDEF ||
1428 			    (symp->st_value != 0 &&
1429 			    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1430 			    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC))) {
1431 				*sym = (c_linker_sym_t) symp;
1432 				return (0);
1433 			}
1434 			return (ENOENT);
1435 		}
1436 	}
1437 
1438 	return (ENOENT);
1439 }
1440 
1441 static int
1442 link_elf_symbol_values(linker_file_t lf, c_linker_sym_t sym,
1443     linker_symval_t *symval)
1444 {
1445 	elf_file_t ef;
1446 	const Elf_Sym *es;
1447 	caddr_t val;
1448 
1449 	ef = (elf_file_t)lf;
1450 	es = (const Elf_Sym *)sym;
1451 	if (es >= ef->symtab && es < (ef->symtab + ef->nchains)) {
1452 		symval->name = ef->strtab + es->st_name;
1453 		val = (caddr_t)ef->address + es->st_value;
1454 		if (ELF_ST_TYPE(es->st_info) == STT_GNU_IFUNC)
1455 			val = ((caddr_t (*)(void))val)();
1456 		symval->value = val;
1457 		symval->size = es->st_size;
1458 		return (0);
1459 	}
1460 	if (ef->symtab == ef->ddbsymtab)
1461 		return (ENOENT);
1462 	if (es >= ef->ddbsymtab && es < (ef->ddbsymtab + ef->ddbsymcnt)) {
1463 		symval->name = ef->ddbstrtab + es->st_name;
1464 		val = (caddr_t)ef->address + es->st_value;
1465 		if (ELF_ST_TYPE(es->st_info) == STT_GNU_IFUNC)
1466 			val = ((caddr_t (*)(void))val)();
1467 		symval->value = val;
1468 		symval->size = es->st_size;
1469 		return (0);
1470 	}
1471 	return (ENOENT);
1472 }
1473 
1474 static int
1475 link_elf_search_symbol(linker_file_t lf, caddr_t value,
1476     c_linker_sym_t *sym, long *diffp)
1477 {
1478 	elf_file_t ef = (elf_file_t) lf;
1479 	u_long off = (uintptr_t) (void *) value;
1480 	u_long diff = off;
1481 	u_long st_value;
1482 	const Elf_Sym* es;
1483 	const Elf_Sym* best = NULL;
1484 	int i;
1485 
1486 	for (i = 0, es = ef->ddbsymtab; i < ef->ddbsymcnt; i++, es++) {
1487 		if (es->st_name == 0)
1488 			continue;
1489 		st_value = es->st_value + (uintptr_t) (void *) ef->address;
1490 		if (off >= st_value) {
1491 			if (off - st_value < diff) {
1492 				diff = off - st_value;
1493 				best = es;
1494 				if (diff == 0)
1495 					break;
1496 			} else if (off - st_value == diff) {
1497 				best = es;
1498 			}
1499 		}
1500 	}
1501 	if (best == NULL)
1502 		*diffp = off;
1503 	else
1504 		*diffp = diff;
1505 	*sym = (c_linker_sym_t) best;
1506 
1507 	return (0);
1508 }
1509 
1510 /*
1511  * Look up a linker set on an ELF system.
1512  */
1513 static int
1514 link_elf_lookup_set(linker_file_t lf, const char *name,
1515     void ***startp, void ***stopp, int *countp)
1516 {
1517 	c_linker_sym_t sym;
1518 	linker_symval_t symval;
1519 	char *setsym;
1520 	void **start, **stop;
1521 	int len, error = 0, count;
1522 
1523 	len = strlen(name) + sizeof("__start_set_"); /* sizeof includes \0 */
1524 	setsym = malloc(len, M_LINKER, M_WAITOK);
1525 
1526 	/* get address of first entry */
1527 	snprintf(setsym, len, "%s%s", "__start_set_", name);
1528 	error = link_elf_lookup_symbol(lf, setsym, &sym);
1529 	if (error != 0)
1530 		goto out;
1531 	link_elf_symbol_values(lf, sym, &symval);
1532 	if (symval.value == 0) {
1533 		error = ESRCH;
1534 		goto out;
1535 	}
1536 	start = (void **)symval.value;
1537 
1538 	/* get address of last entry */
1539 	snprintf(setsym, len, "%s%s", "__stop_set_", name);
1540 	error = link_elf_lookup_symbol(lf, setsym, &sym);
1541 	if (error != 0)
1542 		goto out;
1543 	link_elf_symbol_values(lf, sym, &symval);
1544 	if (symval.value == 0) {
1545 		error = ESRCH;
1546 		goto out;
1547 	}
1548 	stop = (void **)symval.value;
1549 
1550 	/* and the number of entries */
1551 	count = stop - start;
1552 
1553 	/* and copy out */
1554 	if (startp != NULL)
1555 		*startp = start;
1556 	if (stopp != NULL)
1557 		*stopp = stop;
1558 	if (countp != NULL)
1559 		*countp = count;
1560 
1561 out:
1562 	free(setsym, M_LINKER);
1563 	return (error);
1564 }
1565 
1566 static int
1567 link_elf_each_function_name(linker_file_t file,
1568   int (*callback)(const char *, void *), void *opaque)
1569 {
1570 	elf_file_t ef = (elf_file_t)file;
1571 	const Elf_Sym *symp;
1572 	int i, error;
1573 
1574 	/* Exhaustive search */
1575 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1576 		if (symp->st_value != 0 &&
1577 		    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1578 		    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
1579 			error = callback(ef->ddbstrtab + symp->st_name, opaque);
1580 			if (error != 0)
1581 				return (error);
1582 		}
1583 	}
1584 	return (0);
1585 }
1586 
1587 static int
1588 link_elf_each_function_nameval(linker_file_t file,
1589     linker_function_nameval_callback_t callback, void *opaque)
1590 {
1591 	linker_symval_t symval;
1592 	elf_file_t ef = (elf_file_t)file;
1593 	const Elf_Sym* symp;
1594 	int i, error;
1595 
1596 	/* Exhaustive search */
1597 	for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) {
1598 		if (symp->st_value != 0 &&
1599 		    (ELF_ST_TYPE(symp->st_info) == STT_FUNC ||
1600 		    ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC)) {
1601 			error = link_elf_symbol_values(file,
1602 			    (c_linker_sym_t) symp, &symval);
1603 			if (error != 0)
1604 				return (error);
1605 			error = callback(file, i, &symval, opaque);
1606 			if (error != 0)
1607 				return (error);
1608 		}
1609 	}
1610 	return (0);
1611 }
1612 
1613 const Elf_Sym *
1614 elf_get_sym(linker_file_t lf, Elf_Size symidx)
1615 {
1616 	elf_file_t ef = (elf_file_t)lf;
1617 
1618 	if (symidx >= ef->nchains)
1619 		return (NULL);
1620 	return (ef->symtab + symidx);
1621 }
1622 
1623 const char *
1624 elf_get_symname(linker_file_t lf, Elf_Size symidx)
1625 {
1626 	elf_file_t ef = (elf_file_t)lf;
1627 	const Elf_Sym *sym;
1628 
1629 	if (symidx >= ef->nchains)
1630 		return (NULL);
1631 	sym = ef->symtab + symidx;
1632 	return (ef->strtab + sym->st_name);
1633 }
1634 
1635 /*
1636  * Symbol lookup function that can be used when the symbol index is known (ie
1637  * in relocations). It uses the symbol index instead of doing a fully fledged
1638  * hash table based lookup when such is valid. For example for local symbols.
1639  * This is not only more efficient, it's also more correct. It's not always
1640  * the case that the symbol can be found through the hash table.
1641  */
1642 static int
1643 elf_lookup(linker_file_t lf, Elf_Size symidx, int deps, Elf_Addr *res)
1644 {
1645 	elf_file_t ef = (elf_file_t)lf;
1646 	const Elf_Sym *sym;
1647 	const char *symbol;
1648 	Elf_Addr addr, start, base;
1649 
1650 	/* Don't even try to lookup the symbol if the index is bogus. */
1651 	if (symidx >= ef->nchains) {
1652 		*res = 0;
1653 		return (EINVAL);
1654 	}
1655 
1656 	sym = ef->symtab + symidx;
1657 
1658 	/*
1659 	 * Don't do a full lookup when the symbol is local. It may even
1660 	 * fail because it may not be found through the hash table.
1661 	 */
1662 	if (ELF_ST_BIND(sym->st_info) == STB_LOCAL) {
1663 		/* Force lookup failure when we have an insanity. */
1664 		if (sym->st_shndx == SHN_UNDEF || sym->st_value == 0) {
1665 			*res = 0;
1666 			return (EINVAL);
1667 		}
1668 		*res = ((Elf_Addr)ef->address + sym->st_value);
1669 		return (0);
1670 	}
1671 
1672 	/*
1673 	 * XXX we can avoid doing a hash table based lookup for global
1674 	 * symbols as well. This however is not always valid, so we'll
1675 	 * just do it the hard way for now. Performance tweaks can
1676 	 * always be added.
1677 	 */
1678 
1679 	symbol = ef->strtab + sym->st_name;
1680 
1681 	/* Force a lookup failure if the symbol name is bogus. */
1682 	if (*symbol == 0) {
1683 		*res = 0;
1684 		return (EINVAL);
1685 	}
1686 
1687 	addr = ((Elf_Addr)linker_file_lookup_symbol(lf, symbol, deps));
1688 	if (addr == 0 && ELF_ST_BIND(sym->st_info) != STB_WEAK) {
1689 		*res = 0;
1690 		return (EINVAL);
1691 	}
1692 
1693 	if (elf_set_find(&set_pcpu_list, addr, &start, &base))
1694 		addr = addr - start + base;
1695 #ifdef VIMAGE
1696 	else if (elf_set_find(&set_vnet_list, addr, &start, &base))
1697 		addr = addr - start + base;
1698 #endif
1699 	*res = addr;
1700 	return (0);
1701 }
1702 
1703 static void
1704 link_elf_reloc_local(linker_file_t lf)
1705 {
1706 	const Elf_Rel *rellim;
1707 	const Elf_Rel *rel;
1708 	const Elf_Rela *relalim;
1709 	const Elf_Rela *rela;
1710 	elf_file_t ef = (elf_file_t)lf;
1711 
1712 	/* Perform relocations without addend if there are any: */
1713 	if ((rel = ef->rel) != NULL) {
1714 		rellim = (const Elf_Rel *)((const char *)ef->rel + ef->relsize);
1715 		while (rel < rellim) {
1716 			elf_reloc_local(lf, (Elf_Addr)ef->address, rel,
1717 			    ELF_RELOC_REL, elf_lookup);
1718 			rel++;
1719 		}
1720 	}
1721 
1722 	/* Perform relocations with addend if there are any: */
1723 	if ((rela = ef->rela) != NULL) {
1724 		relalim = (const Elf_Rela *)
1725 		    ((const char *)ef->rela + ef->relasize);
1726 		while (rela < relalim) {
1727 			elf_reloc_local(lf, (Elf_Addr)ef->address, rela,
1728 			    ELF_RELOC_RELA, elf_lookup);
1729 			rela++;
1730 		}
1731 	}
1732 }
1733 
1734 static long
1735 link_elf_symtab_get(linker_file_t lf, const Elf_Sym **symtab)
1736 {
1737 	elf_file_t ef = (elf_file_t)lf;
1738 
1739 	*symtab = ef->ddbsymtab;
1740 
1741 	if (*symtab == NULL)
1742 		return (0);
1743 
1744 	return (ef->ddbsymcnt);
1745 }
1746 
1747 static long
1748 link_elf_strtab_get(linker_file_t lf, caddr_t *strtab)
1749 {
1750 	elf_file_t ef = (elf_file_t)lf;
1751 
1752 	*strtab = ef->ddbstrtab;
1753 
1754 	if (*strtab == NULL)
1755 		return (0);
1756 
1757 	return (ef->ddbstrcnt);
1758 }
1759 
1760 #if defined(__i386__) || defined(__amd64__) || defined(__aarch64__)
1761 /*
1762  * Use this lookup routine when performing relocations early during boot.
1763  * The generic lookup routine depends on kobj, which is not initialized
1764  * at that point.
1765  */
1766 static int
1767 elf_lookup_ifunc(linker_file_t lf, Elf_Size symidx, int deps __unused,
1768     Elf_Addr *res)
1769 {
1770 	elf_file_t ef;
1771 	const Elf_Sym *symp;
1772 	caddr_t val;
1773 
1774 	ef = (elf_file_t)lf;
1775 	symp = ef->symtab + symidx;
1776 	if (ELF_ST_TYPE(symp->st_info) == STT_GNU_IFUNC) {
1777 		val = (caddr_t)ef->address + symp->st_value;
1778 		*res = ((Elf_Addr (*)(void))val)();
1779 		return (0);
1780 	}
1781 	return (ENOENT);
1782 }
1783 
1784 void
1785 link_elf_ireloc(caddr_t kmdp)
1786 {
1787 	struct elf_file eff;
1788 	elf_file_t ef;
1789 
1790 	ef = &eff;
1791 
1792 	bzero_early(ef, sizeof(*ef));
1793 
1794 	ef->modptr = kmdp;
1795 	ef->dynamic = (Elf_Dyn *)&_DYNAMIC;
1796 	parse_dynamic(ef);
1797 	ef->address = 0;
1798 	link_elf_preload_parse_symbols(ef);
1799 	relocate_file1(ef, elf_lookup_ifunc, elf_reloc, true);
1800 }
1801 #endif
1802