xref: /linux/kernel/module/main.c (revision 42b16d3ac371a2fac9b6f08fd75f23f34ba3955a)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2002 Richard Henderson
4  * Copyright (C) 2001 Rusty Russell, 2002, 2010 Rusty Russell IBM.
5  * Copyright (C) 2023 Luis Chamberlain <mcgrof@kernel.org>
6  */
7 
8 #define INCLUDE_VERMAGIC
9 
10 #include <linux/export.h>
11 #include <linux/extable.h>
12 #include <linux/moduleloader.h>
13 #include <linux/module_signature.h>
14 #include <linux/trace_events.h>
15 #include <linux/init.h>
16 #include <linux/kallsyms.h>
17 #include <linux/buildid.h>
18 #include <linux/fs.h>
19 #include <linux/kernel.h>
20 #include <linux/kernel_read_file.h>
21 #include <linux/kstrtox.h>
22 #include <linux/slab.h>
23 #include <linux/vmalloc.h>
24 #include <linux/elf.h>
25 #include <linux/seq_file.h>
26 #include <linux/syscalls.h>
27 #include <linux/fcntl.h>
28 #include <linux/rcupdate.h>
29 #include <linux/capability.h>
30 #include <linux/cpu.h>
31 #include <linux/moduleparam.h>
32 #include <linux/errno.h>
33 #include <linux/err.h>
34 #include <linux/vermagic.h>
35 #include <linux/notifier.h>
36 #include <linux/sched.h>
37 #include <linux/device.h>
38 #include <linux/string.h>
39 #include <linux/mutex.h>
40 #include <linux/rculist.h>
41 #include <linux/uaccess.h>
42 #include <asm/cacheflush.h>
43 #include <linux/set_memory.h>
44 #include <asm/mmu_context.h>
45 #include <linux/license.h>
46 #include <asm/sections.h>
47 #include <linux/tracepoint.h>
48 #include <linux/ftrace.h>
49 #include <linux/livepatch.h>
50 #include <linux/async.h>
51 #include <linux/percpu.h>
52 #include <linux/kmemleak.h>
53 #include <linux/jump_label.h>
54 #include <linux/pfn.h>
55 #include <linux/bsearch.h>
56 #include <linux/dynamic_debug.h>
57 #include <linux/audit.h>
58 #include <linux/cfi.h>
59 #include <linux/codetag.h>
60 #include <linux/debugfs.h>
61 #include <linux/execmem.h>
62 #include <uapi/linux/module.h>
63 #include "internal.h"
64 
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/module.h>
67 
68 /*
69  * Mutex protects:
70  * 1) List of modules (also safely readable with preempt_disable),
71  * 2) module_use links,
72  * 3) mod_tree.addr_min/mod_tree.addr_max.
73  * (delete and add uses RCU list operations).
74  */
75 DEFINE_MUTEX(module_mutex);
76 LIST_HEAD(modules);
77 
78 /* Work queue for freeing init sections in success case */
79 static void do_free_init(struct work_struct *w);
80 static DECLARE_WORK(init_free_wq, do_free_init);
81 static LLIST_HEAD(init_free_list);
82 
83 struct mod_tree_root mod_tree __cacheline_aligned = {
84 	.addr_min = -1UL,
85 };
86 
87 struct symsearch {
88 	const struct kernel_symbol *start, *stop;
89 	const s32 *crcs;
90 	enum mod_license license;
91 };
92 
93 /*
94  * Bounds of module memory, for speeding up __module_address.
95  * Protected by module_mutex.
96  */
__mod_update_bounds(enum mod_mem_type type __maybe_unused,void * base,unsigned int size,struct mod_tree_root * tree)97 static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
98 				unsigned int size, struct mod_tree_root *tree)
99 {
100 	unsigned long min = (unsigned long)base;
101 	unsigned long max = min + size;
102 
103 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
104 	if (mod_mem_type_is_core_data(type)) {
105 		if (min < tree->data_addr_min)
106 			tree->data_addr_min = min;
107 		if (max > tree->data_addr_max)
108 			tree->data_addr_max = max;
109 		return;
110 	}
111 #endif
112 	if (min < tree->addr_min)
113 		tree->addr_min = min;
114 	if (max > tree->addr_max)
115 		tree->addr_max = max;
116 }
117 
mod_update_bounds(struct module * mod)118 static void mod_update_bounds(struct module *mod)
119 {
120 	for_each_mod_mem_type(type) {
121 		struct module_memory *mod_mem = &mod->mem[type];
122 
123 		if (mod_mem->size)
124 			__mod_update_bounds(type, mod_mem->base, mod_mem->size, &mod_tree);
125 	}
126 }
127 
128 /* Block module loading/unloading? */
129 int modules_disabled;
130 core_param(nomodule, modules_disabled, bint, 0);
131 
132 /* Waiting for a module to finish initializing? */
133 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
134 
135 static BLOCKING_NOTIFIER_HEAD(module_notify_list);
136 
register_module_notifier(struct notifier_block * nb)137 int register_module_notifier(struct notifier_block *nb)
138 {
139 	return blocking_notifier_chain_register(&module_notify_list, nb);
140 }
141 EXPORT_SYMBOL(register_module_notifier);
142 
unregister_module_notifier(struct notifier_block * nb)143 int unregister_module_notifier(struct notifier_block *nb)
144 {
145 	return blocking_notifier_chain_unregister(&module_notify_list, nb);
146 }
147 EXPORT_SYMBOL(unregister_module_notifier);
148 
149 /*
150  * We require a truly strong try_module_get(): 0 means success.
151  * Otherwise an error is returned due to ongoing or failed
152  * initialization etc.
153  */
strong_try_module_get(struct module * mod)154 static inline int strong_try_module_get(struct module *mod)
155 {
156 	BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
157 	if (mod && mod->state == MODULE_STATE_COMING)
158 		return -EBUSY;
159 	if (try_module_get(mod))
160 		return 0;
161 	else
162 		return -ENOENT;
163 }
164 
add_taint_module(struct module * mod,unsigned flag,enum lockdep_ok lockdep_ok)165 static inline void add_taint_module(struct module *mod, unsigned flag,
166 				    enum lockdep_ok lockdep_ok)
167 {
168 	add_taint(flag, lockdep_ok);
169 	set_bit(flag, &mod->taints);
170 }
171 
172 /*
173  * A thread that wants to hold a reference to a module only while it
174  * is running can call this to safely exit.
175  */
__module_put_and_kthread_exit(struct module * mod,long code)176 void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
177 {
178 	module_put(mod);
179 	kthread_exit(code);
180 }
181 EXPORT_SYMBOL(__module_put_and_kthread_exit);
182 
183 /* Find a module section: 0 means not found. */
find_sec(const struct load_info * info,const char * name)184 static unsigned int find_sec(const struct load_info *info, const char *name)
185 {
186 	unsigned int i;
187 
188 	for (i = 1; i < info->hdr->e_shnum; i++) {
189 		Elf_Shdr *shdr = &info->sechdrs[i];
190 		/* Alloc bit cleared means "ignore it." */
191 		if ((shdr->sh_flags & SHF_ALLOC)
192 		    && strcmp(info->secstrings + shdr->sh_name, name) == 0)
193 			return i;
194 	}
195 	return 0;
196 }
197 
198 /* Find a module section, or NULL. */
section_addr(const struct load_info * info,const char * name)199 static void *section_addr(const struct load_info *info, const char *name)
200 {
201 	/* Section 0 has sh_addr 0. */
202 	return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
203 }
204 
205 /* Find a module section, or NULL.  Fill in number of "objects" in section. */
section_objs(const struct load_info * info,const char * name,size_t object_size,unsigned int * num)206 static void *section_objs(const struct load_info *info,
207 			  const char *name,
208 			  size_t object_size,
209 			  unsigned int *num)
210 {
211 	unsigned int sec = find_sec(info, name);
212 
213 	/* Section 0 has sh_addr 0 and sh_size 0. */
214 	*num = info->sechdrs[sec].sh_size / object_size;
215 	return (void *)info->sechdrs[sec].sh_addr;
216 }
217 
218 /* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
find_any_sec(const struct load_info * info,const char * name)219 static unsigned int find_any_sec(const struct load_info *info, const char *name)
220 {
221 	unsigned int i;
222 
223 	for (i = 1; i < info->hdr->e_shnum; i++) {
224 		Elf_Shdr *shdr = &info->sechdrs[i];
225 		if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
226 			return i;
227 	}
228 	return 0;
229 }
230 
231 /*
232  * Find a module section, or NULL. Fill in number of "objects" in section.
233  * Ignores SHF_ALLOC flag.
234  */
any_section_objs(const struct load_info * info,const char * name,size_t object_size,unsigned int * num)235 static __maybe_unused void *any_section_objs(const struct load_info *info,
236 					     const char *name,
237 					     size_t object_size,
238 					     unsigned int *num)
239 {
240 	unsigned int sec = find_any_sec(info, name);
241 
242 	/* Section 0 has sh_addr 0 and sh_size 0. */
243 	*num = info->sechdrs[sec].sh_size / object_size;
244 	return (void *)info->sechdrs[sec].sh_addr;
245 }
246 
247 #ifndef CONFIG_MODVERSIONS
248 #define symversion(base, idx) NULL
249 #else
250 #define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
251 #endif
252 
kernel_symbol_name(const struct kernel_symbol * sym)253 static const char *kernel_symbol_name(const struct kernel_symbol *sym)
254 {
255 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
256 	return offset_to_ptr(&sym->name_offset);
257 #else
258 	return sym->name;
259 #endif
260 }
261 
kernel_symbol_namespace(const struct kernel_symbol * sym)262 static const char *kernel_symbol_namespace(const struct kernel_symbol *sym)
263 {
264 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
265 	if (!sym->namespace_offset)
266 		return NULL;
267 	return offset_to_ptr(&sym->namespace_offset);
268 #else
269 	return sym->namespace;
270 #endif
271 }
272 
cmp_name(const void * name,const void * sym)273 int cmp_name(const void *name, const void *sym)
274 {
275 	return strcmp(name, kernel_symbol_name(sym));
276 }
277 
find_exported_symbol_in_section(const struct symsearch * syms,struct module * owner,struct find_symbol_arg * fsa)278 static bool find_exported_symbol_in_section(const struct symsearch *syms,
279 					    struct module *owner,
280 					    struct find_symbol_arg *fsa)
281 {
282 	struct kernel_symbol *sym;
283 
284 	if (!fsa->gplok && syms->license == GPL_ONLY)
285 		return false;
286 
287 	sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
288 			sizeof(struct kernel_symbol), cmp_name);
289 	if (!sym)
290 		return false;
291 
292 	fsa->owner = owner;
293 	fsa->crc = symversion(syms->crcs, sym - syms->start);
294 	fsa->sym = sym;
295 	fsa->license = syms->license;
296 
297 	return true;
298 }
299 
300 /*
301  * Find an exported symbol and return it, along with, (optional) crc and
302  * (optional) module which owns it.  Needs preempt disabled or module_mutex.
303  */
find_symbol(struct find_symbol_arg * fsa)304 bool find_symbol(struct find_symbol_arg *fsa)
305 {
306 	static const struct symsearch arr[] = {
307 		{ __start___ksymtab, __stop___ksymtab, __start___kcrctab,
308 		  NOT_GPL_ONLY },
309 		{ __start___ksymtab_gpl, __stop___ksymtab_gpl,
310 		  __start___kcrctab_gpl,
311 		  GPL_ONLY },
312 	};
313 	struct module *mod;
314 	unsigned int i;
315 
316 	module_assert_mutex_or_preempt();
317 
318 	for (i = 0; i < ARRAY_SIZE(arr); i++)
319 		if (find_exported_symbol_in_section(&arr[i], NULL, fsa))
320 			return true;
321 
322 	list_for_each_entry_rcu(mod, &modules, list,
323 				lockdep_is_held(&module_mutex)) {
324 		struct symsearch arr[] = {
325 			{ mod->syms, mod->syms + mod->num_syms, mod->crcs,
326 			  NOT_GPL_ONLY },
327 			{ mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
328 			  mod->gpl_crcs,
329 			  GPL_ONLY },
330 		};
331 
332 		if (mod->state == MODULE_STATE_UNFORMED)
333 			continue;
334 
335 		for (i = 0; i < ARRAY_SIZE(arr); i++)
336 			if (find_exported_symbol_in_section(&arr[i], mod, fsa))
337 				return true;
338 	}
339 
340 	pr_debug("Failed to find symbol %s\n", fsa->name);
341 	return false;
342 }
343 
344 /*
345  * Search for module by name: must hold module_mutex (or preempt disabled
346  * for read-only access).
347  */
find_module_all(const char * name,size_t len,bool even_unformed)348 struct module *find_module_all(const char *name, size_t len,
349 			       bool even_unformed)
350 {
351 	struct module *mod;
352 
353 	module_assert_mutex_or_preempt();
354 
355 	list_for_each_entry_rcu(mod, &modules, list,
356 				lockdep_is_held(&module_mutex)) {
357 		if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
358 			continue;
359 		if (strlen(mod->name) == len && !memcmp(mod->name, name, len))
360 			return mod;
361 	}
362 	return NULL;
363 }
364 
find_module(const char * name)365 struct module *find_module(const char *name)
366 {
367 	return find_module_all(name, strlen(name), false);
368 }
369 
370 #ifdef CONFIG_SMP
371 
mod_percpu(struct module * mod)372 static inline void __percpu *mod_percpu(struct module *mod)
373 {
374 	return mod->percpu;
375 }
376 
percpu_modalloc(struct module * mod,struct load_info * info)377 static int percpu_modalloc(struct module *mod, struct load_info *info)
378 {
379 	Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
380 	unsigned long align = pcpusec->sh_addralign;
381 
382 	if (!pcpusec->sh_size)
383 		return 0;
384 
385 	if (align > PAGE_SIZE) {
386 		pr_warn("%s: per-cpu alignment %li > %li\n",
387 			mod->name, align, PAGE_SIZE);
388 		align = PAGE_SIZE;
389 	}
390 
391 	mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
392 	if (!mod->percpu) {
393 		pr_warn("%s: Could not allocate %lu bytes percpu data\n",
394 			mod->name, (unsigned long)pcpusec->sh_size);
395 		return -ENOMEM;
396 	}
397 	mod->percpu_size = pcpusec->sh_size;
398 	return 0;
399 }
400 
percpu_modfree(struct module * mod)401 static void percpu_modfree(struct module *mod)
402 {
403 	free_percpu(mod->percpu);
404 }
405 
find_pcpusec(struct load_info * info)406 static unsigned int find_pcpusec(struct load_info *info)
407 {
408 	return find_sec(info, ".data..percpu");
409 }
410 
percpu_modcopy(struct module * mod,const void * from,unsigned long size)411 static void percpu_modcopy(struct module *mod,
412 			   const void *from, unsigned long size)
413 {
414 	int cpu;
415 
416 	for_each_possible_cpu(cpu)
417 		memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
418 }
419 
__is_module_percpu_address(unsigned long addr,unsigned long * can_addr)420 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
421 {
422 	struct module *mod;
423 	unsigned int cpu;
424 
425 	preempt_disable();
426 
427 	list_for_each_entry_rcu(mod, &modules, list) {
428 		if (mod->state == MODULE_STATE_UNFORMED)
429 			continue;
430 		if (!mod->percpu_size)
431 			continue;
432 		for_each_possible_cpu(cpu) {
433 			void *start = per_cpu_ptr(mod->percpu, cpu);
434 			void *va = (void *)addr;
435 
436 			if (va >= start && va < start + mod->percpu_size) {
437 				if (can_addr) {
438 					*can_addr = (unsigned long) (va - start);
439 					*can_addr += (unsigned long)
440 						per_cpu_ptr(mod->percpu,
441 							    get_boot_cpu_id());
442 				}
443 				preempt_enable();
444 				return true;
445 			}
446 		}
447 	}
448 
449 	preempt_enable();
450 	return false;
451 }
452 
453 /**
454  * is_module_percpu_address() - test whether address is from module static percpu
455  * @addr: address to test
456  *
457  * Test whether @addr belongs to module static percpu area.
458  *
459  * Return: %true if @addr is from module static percpu area
460  */
is_module_percpu_address(unsigned long addr)461 bool is_module_percpu_address(unsigned long addr)
462 {
463 	return __is_module_percpu_address(addr, NULL);
464 }
465 
466 #else /* ... !CONFIG_SMP */
467 
mod_percpu(struct module * mod)468 static inline void __percpu *mod_percpu(struct module *mod)
469 {
470 	return NULL;
471 }
percpu_modalloc(struct module * mod,struct load_info * info)472 static int percpu_modalloc(struct module *mod, struct load_info *info)
473 {
474 	/* UP modules shouldn't have this section: ENOMEM isn't quite right */
475 	if (info->sechdrs[info->index.pcpu].sh_size != 0)
476 		return -ENOMEM;
477 	return 0;
478 }
percpu_modfree(struct module * mod)479 static inline void percpu_modfree(struct module *mod)
480 {
481 }
find_pcpusec(struct load_info * info)482 static unsigned int find_pcpusec(struct load_info *info)
483 {
484 	return 0;
485 }
percpu_modcopy(struct module * mod,const void * from,unsigned long size)486 static inline void percpu_modcopy(struct module *mod,
487 				  const void *from, unsigned long size)
488 {
489 	/* pcpusec should be 0, and size of that section should be 0. */
490 	BUG_ON(size != 0);
491 }
is_module_percpu_address(unsigned long addr)492 bool is_module_percpu_address(unsigned long addr)
493 {
494 	return false;
495 }
496 
__is_module_percpu_address(unsigned long addr,unsigned long * can_addr)497 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
498 {
499 	return false;
500 }
501 
502 #endif /* CONFIG_SMP */
503 
504 #define MODINFO_ATTR(field)	\
505 static void setup_modinfo_##field(struct module *mod, const char *s)  \
506 {                                                                     \
507 	mod->field = kstrdup(s, GFP_KERNEL);                          \
508 }                                                                     \
509 static ssize_t show_modinfo_##field(struct module_attribute *mattr,   \
510 			struct module_kobject *mk, char *buffer)      \
511 {                                                                     \
512 	return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field);  \
513 }                                                                     \
514 static int modinfo_##field##_exists(struct module *mod)               \
515 {                                                                     \
516 	return mod->field != NULL;                                    \
517 }                                                                     \
518 static void free_modinfo_##field(struct module *mod)                  \
519 {                                                                     \
520 	kfree(mod->field);                                            \
521 	mod->field = NULL;                                            \
522 }                                                                     \
523 static struct module_attribute modinfo_##field = {                    \
524 	.attr = { .name = __stringify(field), .mode = 0444 },         \
525 	.show = show_modinfo_##field,                                 \
526 	.setup = setup_modinfo_##field,                               \
527 	.test = modinfo_##field##_exists,                             \
528 	.free = free_modinfo_##field,                                 \
529 };
530 
531 MODINFO_ATTR(version);
532 MODINFO_ATTR(srcversion);
533 
534 static struct {
535 	char name[MODULE_NAME_LEN + 1];
536 	char taints[MODULE_FLAGS_BUF_SIZE];
537 } last_unloaded_module;
538 
539 #ifdef CONFIG_MODULE_UNLOAD
540 
541 EXPORT_TRACEPOINT_SYMBOL(module_get);
542 
543 /* MODULE_REF_BASE is the base reference count by kmodule loader. */
544 #define MODULE_REF_BASE	1
545 
546 /* Init the unload section of the module. */
module_unload_init(struct module * mod)547 static int module_unload_init(struct module *mod)
548 {
549 	/*
550 	 * Initialize reference counter to MODULE_REF_BASE.
551 	 * refcnt == 0 means module is going.
552 	 */
553 	atomic_set(&mod->refcnt, MODULE_REF_BASE);
554 
555 	INIT_LIST_HEAD(&mod->source_list);
556 	INIT_LIST_HEAD(&mod->target_list);
557 
558 	/* Hold reference count during initialization. */
559 	atomic_inc(&mod->refcnt);
560 
561 	return 0;
562 }
563 
564 /* Does a already use b? */
already_uses(struct module * a,struct module * b)565 static int already_uses(struct module *a, struct module *b)
566 {
567 	struct module_use *use;
568 
569 	list_for_each_entry(use, &b->source_list, source_list) {
570 		if (use->source == a)
571 			return 1;
572 	}
573 	pr_debug("%s does not use %s!\n", a->name, b->name);
574 	return 0;
575 }
576 
577 /*
578  * Module a uses b
579  *  - we add 'a' as a "source", 'b' as a "target" of module use
580  *  - the module_use is added to the list of 'b' sources (so
581  *    'b' can walk the list to see who sourced them), and of 'a'
582  *    targets (so 'a' can see what modules it targets).
583  */
add_module_usage(struct module * a,struct module * b)584 static int add_module_usage(struct module *a, struct module *b)
585 {
586 	struct module_use *use;
587 
588 	pr_debug("Allocating new usage for %s.\n", a->name);
589 	use = kmalloc(sizeof(*use), GFP_ATOMIC);
590 	if (!use)
591 		return -ENOMEM;
592 
593 	use->source = a;
594 	use->target = b;
595 	list_add(&use->source_list, &b->source_list);
596 	list_add(&use->target_list, &a->target_list);
597 	return 0;
598 }
599 
600 /* Module a uses b: caller needs module_mutex() */
ref_module(struct module * a,struct module * b)601 static int ref_module(struct module *a, struct module *b)
602 {
603 	int err;
604 
605 	if (b == NULL || already_uses(a, b))
606 		return 0;
607 
608 	/* If module isn't available, we fail. */
609 	err = strong_try_module_get(b);
610 	if (err)
611 		return err;
612 
613 	err = add_module_usage(a, b);
614 	if (err) {
615 		module_put(b);
616 		return err;
617 	}
618 	return 0;
619 }
620 
621 /* Clear the unload stuff of the module. */
module_unload_free(struct module * mod)622 static void module_unload_free(struct module *mod)
623 {
624 	struct module_use *use, *tmp;
625 
626 	mutex_lock(&module_mutex);
627 	list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
628 		struct module *i = use->target;
629 		pr_debug("%s unusing %s\n", mod->name, i->name);
630 		module_put(i);
631 		list_del(&use->source_list);
632 		list_del(&use->target_list);
633 		kfree(use);
634 	}
635 	mutex_unlock(&module_mutex);
636 }
637 
638 #ifdef CONFIG_MODULE_FORCE_UNLOAD
try_force_unload(unsigned int flags)639 static inline int try_force_unload(unsigned int flags)
640 {
641 	int ret = (flags & O_TRUNC);
642 	if (ret)
643 		add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
644 	return ret;
645 }
646 #else
try_force_unload(unsigned int flags)647 static inline int try_force_unload(unsigned int flags)
648 {
649 	return 0;
650 }
651 #endif /* CONFIG_MODULE_FORCE_UNLOAD */
652 
653 /* Try to release refcount of module, 0 means success. */
try_release_module_ref(struct module * mod)654 static int try_release_module_ref(struct module *mod)
655 {
656 	int ret;
657 
658 	/* Try to decrement refcnt which we set at loading */
659 	ret = atomic_sub_return(MODULE_REF_BASE, &mod->refcnt);
660 	BUG_ON(ret < 0);
661 	if (ret)
662 		/* Someone can put this right now, recover with checking */
663 		ret = atomic_add_unless(&mod->refcnt, MODULE_REF_BASE, 0);
664 
665 	return ret;
666 }
667 
try_stop_module(struct module * mod,int flags,int * forced)668 static int try_stop_module(struct module *mod, int flags, int *forced)
669 {
670 	/* If it's not unused, quit unless we're forcing. */
671 	if (try_release_module_ref(mod) != 0) {
672 		*forced = try_force_unload(flags);
673 		if (!(*forced))
674 			return -EWOULDBLOCK;
675 	}
676 
677 	/* Mark it as dying. */
678 	mod->state = MODULE_STATE_GOING;
679 
680 	return 0;
681 }
682 
683 /**
684  * module_refcount() - return the refcount or -1 if unloading
685  * @mod:	the module we're checking
686  *
687  * Return:
688  *	-1 if the module is in the process of unloading
689  *	otherwise the number of references in the kernel to the module
690  */
module_refcount(struct module * mod)691 int module_refcount(struct module *mod)
692 {
693 	return atomic_read(&mod->refcnt) - MODULE_REF_BASE;
694 }
695 EXPORT_SYMBOL(module_refcount);
696 
697 /* This exists whether we can unload or not */
698 static void free_module(struct module *mod);
699 
SYSCALL_DEFINE2(delete_module,const char __user *,name_user,unsigned int,flags)700 SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
701 		unsigned int, flags)
702 {
703 	struct module *mod;
704 	char name[MODULE_NAME_LEN];
705 	char buf[MODULE_FLAGS_BUF_SIZE];
706 	int ret, forced = 0;
707 
708 	if (!capable(CAP_SYS_MODULE) || modules_disabled)
709 		return -EPERM;
710 
711 	if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0)
712 		return -EFAULT;
713 	name[MODULE_NAME_LEN-1] = '\0';
714 
715 	audit_log_kern_module(name);
716 
717 	if (mutex_lock_interruptible(&module_mutex) != 0)
718 		return -EINTR;
719 
720 	mod = find_module(name);
721 	if (!mod) {
722 		ret = -ENOENT;
723 		goto out;
724 	}
725 
726 	if (!list_empty(&mod->source_list)) {
727 		/* Other modules depend on us: get rid of them first. */
728 		ret = -EWOULDBLOCK;
729 		goto out;
730 	}
731 
732 	/* Doing init or already dying? */
733 	if (mod->state != MODULE_STATE_LIVE) {
734 		/* FIXME: if (force), slam module count damn the torpedoes */
735 		pr_debug("%s already dying\n", mod->name);
736 		ret = -EBUSY;
737 		goto out;
738 	}
739 
740 	/* If it has an init func, it must have an exit func to unload */
741 	if (mod->init && !mod->exit) {
742 		forced = try_force_unload(flags);
743 		if (!forced) {
744 			/* This module can't be removed */
745 			ret = -EBUSY;
746 			goto out;
747 		}
748 	}
749 
750 	ret = try_stop_module(mod, flags, &forced);
751 	if (ret != 0)
752 		goto out;
753 
754 	mutex_unlock(&module_mutex);
755 	/* Final destruction now no one is using it. */
756 	if (mod->exit != NULL)
757 		mod->exit();
758 	blocking_notifier_call_chain(&module_notify_list,
759 				     MODULE_STATE_GOING, mod);
760 	klp_module_going(mod);
761 	ftrace_release_mod(mod);
762 
763 	async_synchronize_full();
764 
765 	/* Store the name and taints of the last unloaded module for diagnostic purposes */
766 	strscpy(last_unloaded_module.name, mod->name, sizeof(last_unloaded_module.name));
767 	strscpy(last_unloaded_module.taints, module_flags(mod, buf, false), sizeof(last_unloaded_module.taints));
768 
769 	free_module(mod);
770 	/* someone could wait for the module in add_unformed_module() */
771 	wake_up_all(&module_wq);
772 	return 0;
773 out:
774 	mutex_unlock(&module_mutex);
775 	return ret;
776 }
777 
__symbol_put(const char * symbol)778 void __symbol_put(const char *symbol)
779 {
780 	struct find_symbol_arg fsa = {
781 		.name	= symbol,
782 		.gplok	= true,
783 	};
784 
785 	preempt_disable();
786 	BUG_ON(!find_symbol(&fsa));
787 	module_put(fsa.owner);
788 	preempt_enable();
789 }
790 EXPORT_SYMBOL(__symbol_put);
791 
792 /* Note this assumes addr is a function, which it currently always is. */
symbol_put_addr(void * addr)793 void symbol_put_addr(void *addr)
794 {
795 	struct module *modaddr;
796 	unsigned long a = (unsigned long)dereference_function_descriptor(addr);
797 
798 	if (core_kernel_text(a))
799 		return;
800 
801 	/*
802 	 * Even though we hold a reference on the module; we still need to
803 	 * disable preemption in order to safely traverse the data structure.
804 	 */
805 	preempt_disable();
806 	modaddr = __module_text_address(a);
807 	BUG_ON(!modaddr);
808 	module_put(modaddr);
809 	preempt_enable();
810 }
811 EXPORT_SYMBOL_GPL(symbol_put_addr);
812 
show_refcnt(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)813 static ssize_t show_refcnt(struct module_attribute *mattr,
814 			   struct module_kobject *mk, char *buffer)
815 {
816 	return sprintf(buffer, "%i\n", module_refcount(mk->mod));
817 }
818 
819 static struct module_attribute modinfo_refcnt =
820 	__ATTR(refcnt, 0444, show_refcnt, NULL);
821 
__module_get(struct module * module)822 void __module_get(struct module *module)
823 {
824 	if (module) {
825 		atomic_inc(&module->refcnt);
826 		trace_module_get(module, _RET_IP_);
827 	}
828 }
829 EXPORT_SYMBOL(__module_get);
830 
try_module_get(struct module * module)831 bool try_module_get(struct module *module)
832 {
833 	bool ret = true;
834 
835 	if (module) {
836 		/* Note: here, we can fail to get a reference */
837 		if (likely(module_is_live(module) &&
838 			   atomic_inc_not_zero(&module->refcnt) != 0))
839 			trace_module_get(module, _RET_IP_);
840 		else
841 			ret = false;
842 	}
843 	return ret;
844 }
845 EXPORT_SYMBOL(try_module_get);
846 
module_put(struct module * module)847 void module_put(struct module *module)
848 {
849 	int ret;
850 
851 	if (module) {
852 		ret = atomic_dec_if_positive(&module->refcnt);
853 		WARN_ON(ret < 0);	/* Failed to put refcount */
854 		trace_module_put(module, _RET_IP_);
855 	}
856 }
857 EXPORT_SYMBOL(module_put);
858 
859 #else /* !CONFIG_MODULE_UNLOAD */
module_unload_free(struct module * mod)860 static inline void module_unload_free(struct module *mod)
861 {
862 }
863 
ref_module(struct module * a,struct module * b)864 static int ref_module(struct module *a, struct module *b)
865 {
866 	return strong_try_module_get(b);
867 }
868 
module_unload_init(struct module * mod)869 static inline int module_unload_init(struct module *mod)
870 {
871 	return 0;
872 }
873 #endif /* CONFIG_MODULE_UNLOAD */
874 
module_flags_taint(unsigned long taints,char * buf)875 size_t module_flags_taint(unsigned long taints, char *buf)
876 {
877 	size_t l = 0;
878 	int i;
879 
880 	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
881 		if (taint_flags[i].module && test_bit(i, &taints))
882 			buf[l++] = taint_flags[i].c_true;
883 	}
884 
885 	return l;
886 }
887 
show_initstate(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)888 static ssize_t show_initstate(struct module_attribute *mattr,
889 			      struct module_kobject *mk, char *buffer)
890 {
891 	const char *state = "unknown";
892 
893 	switch (mk->mod->state) {
894 	case MODULE_STATE_LIVE:
895 		state = "live";
896 		break;
897 	case MODULE_STATE_COMING:
898 		state = "coming";
899 		break;
900 	case MODULE_STATE_GOING:
901 		state = "going";
902 		break;
903 	default:
904 		BUG();
905 	}
906 	return sprintf(buffer, "%s\n", state);
907 }
908 
909 static struct module_attribute modinfo_initstate =
910 	__ATTR(initstate, 0444, show_initstate, NULL);
911 
store_uevent(struct module_attribute * mattr,struct module_kobject * mk,const char * buffer,size_t count)912 static ssize_t store_uevent(struct module_attribute *mattr,
913 			    struct module_kobject *mk,
914 			    const char *buffer, size_t count)
915 {
916 	int rc;
917 
918 	rc = kobject_synth_uevent(&mk->kobj, buffer, count);
919 	return rc ? rc : count;
920 }
921 
922 struct module_attribute module_uevent =
923 	__ATTR(uevent, 0200, NULL, store_uevent);
924 
show_coresize(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)925 static ssize_t show_coresize(struct module_attribute *mattr,
926 			     struct module_kobject *mk, char *buffer)
927 {
928 	unsigned int size = mk->mod->mem[MOD_TEXT].size;
929 
930 	if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
931 		for_class_mod_mem_type(type, core_data)
932 			size += mk->mod->mem[type].size;
933 	}
934 	return sprintf(buffer, "%u\n", size);
935 }
936 
937 static struct module_attribute modinfo_coresize =
938 	__ATTR(coresize, 0444, show_coresize, NULL);
939 
940 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
show_datasize(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)941 static ssize_t show_datasize(struct module_attribute *mattr,
942 			     struct module_kobject *mk, char *buffer)
943 {
944 	unsigned int size = 0;
945 
946 	for_class_mod_mem_type(type, core_data)
947 		size += mk->mod->mem[type].size;
948 	return sprintf(buffer, "%u\n", size);
949 }
950 
951 static struct module_attribute modinfo_datasize =
952 	__ATTR(datasize, 0444, show_datasize, NULL);
953 #endif
954 
show_initsize(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)955 static ssize_t show_initsize(struct module_attribute *mattr,
956 			     struct module_kobject *mk, char *buffer)
957 {
958 	unsigned int size = 0;
959 
960 	for_class_mod_mem_type(type, init)
961 		size += mk->mod->mem[type].size;
962 	return sprintf(buffer, "%u\n", size);
963 }
964 
965 static struct module_attribute modinfo_initsize =
966 	__ATTR(initsize, 0444, show_initsize, NULL);
967 
show_taint(struct module_attribute * mattr,struct module_kobject * mk,char * buffer)968 static ssize_t show_taint(struct module_attribute *mattr,
969 			  struct module_kobject *mk, char *buffer)
970 {
971 	size_t l;
972 
973 	l = module_flags_taint(mk->mod->taints, buffer);
974 	buffer[l++] = '\n';
975 	return l;
976 }
977 
978 static struct module_attribute modinfo_taint =
979 	__ATTR(taint, 0444, show_taint, NULL);
980 
981 struct module_attribute *modinfo_attrs[] = {
982 	&module_uevent,
983 	&modinfo_version,
984 	&modinfo_srcversion,
985 	&modinfo_initstate,
986 	&modinfo_coresize,
987 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
988 	&modinfo_datasize,
989 #endif
990 	&modinfo_initsize,
991 	&modinfo_taint,
992 #ifdef CONFIG_MODULE_UNLOAD
993 	&modinfo_refcnt,
994 #endif
995 	NULL,
996 };
997 
998 size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
999 
1000 static const char vermagic[] = VERMAGIC_STRING;
1001 
try_to_force_load(struct module * mod,const char * reason)1002 int try_to_force_load(struct module *mod, const char *reason)
1003 {
1004 #ifdef CONFIG_MODULE_FORCE_LOAD
1005 	if (!test_taint(TAINT_FORCED_MODULE))
1006 		pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1007 	add_taint_module(mod, TAINT_FORCED_MODULE, LOCKDEP_NOW_UNRELIABLE);
1008 	return 0;
1009 #else
1010 	return -ENOEXEC;
1011 #endif
1012 }
1013 
1014 /* Parse tag=value strings from .modinfo section */
module_next_tag_pair(char * string,unsigned long * secsize)1015 char *module_next_tag_pair(char *string, unsigned long *secsize)
1016 {
1017 	/* Skip non-zero chars */
1018 	while (string[0]) {
1019 		string++;
1020 		if ((*secsize)-- <= 1)
1021 			return NULL;
1022 	}
1023 
1024 	/* Skip any zero padding. */
1025 	while (!string[0]) {
1026 		string++;
1027 		if ((*secsize)-- <= 1)
1028 			return NULL;
1029 	}
1030 	return string;
1031 }
1032 
get_next_modinfo(const struct load_info * info,const char * tag,char * prev)1033 static char *get_next_modinfo(const struct load_info *info, const char *tag,
1034 			      char *prev)
1035 {
1036 	char *p;
1037 	unsigned int taglen = strlen(tag);
1038 	Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1039 	unsigned long size = infosec->sh_size;
1040 
1041 	/*
1042 	 * get_modinfo() calls made before rewrite_section_headers()
1043 	 * must use sh_offset, as sh_addr isn't set!
1044 	 */
1045 	char *modinfo = (char *)info->hdr + infosec->sh_offset;
1046 
1047 	if (prev) {
1048 		size -= prev - modinfo;
1049 		modinfo = module_next_tag_pair(prev, &size);
1050 	}
1051 
1052 	for (p = modinfo; p; p = module_next_tag_pair(p, &size)) {
1053 		if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1054 			return p + taglen + 1;
1055 	}
1056 	return NULL;
1057 }
1058 
get_modinfo(const struct load_info * info,const char * tag)1059 static char *get_modinfo(const struct load_info *info, const char *tag)
1060 {
1061 	return get_next_modinfo(info, tag, NULL);
1062 }
1063 
verify_namespace_is_imported(const struct load_info * info,const struct kernel_symbol * sym,struct module * mod)1064 static int verify_namespace_is_imported(const struct load_info *info,
1065 					const struct kernel_symbol *sym,
1066 					struct module *mod)
1067 {
1068 	const char *namespace;
1069 	char *imported_namespace;
1070 
1071 	namespace = kernel_symbol_namespace(sym);
1072 	if (namespace && namespace[0]) {
1073 		for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1074 			if (strcmp(namespace, imported_namespace) == 0)
1075 				return 0;
1076 		}
1077 #ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1078 		pr_warn(
1079 #else
1080 		pr_err(
1081 #endif
1082 			"%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1083 			mod->name, kernel_symbol_name(sym), namespace);
1084 #ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1085 		return -EINVAL;
1086 #endif
1087 	}
1088 	return 0;
1089 }
1090 
inherit_taint(struct module * mod,struct module * owner,const char * name)1091 static bool inherit_taint(struct module *mod, struct module *owner, const char *name)
1092 {
1093 	if (!owner || !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1094 		return true;
1095 
1096 	if (mod->using_gplonly_symbols) {
1097 		pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1098 			mod->name, name, owner->name);
1099 		return false;
1100 	}
1101 
1102 	if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1103 		pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1104 			mod->name, name, owner->name);
1105 		set_bit(TAINT_PROPRIETARY_MODULE, &mod->taints);
1106 	}
1107 	return true;
1108 }
1109 
1110 /* Resolve a symbol for this module.  I.e. if we find one, record usage. */
resolve_symbol(struct module * mod,const struct load_info * info,const char * name,char ownername[])1111 static const struct kernel_symbol *resolve_symbol(struct module *mod,
1112 						  const struct load_info *info,
1113 						  const char *name,
1114 						  char ownername[])
1115 {
1116 	struct find_symbol_arg fsa = {
1117 		.name	= name,
1118 		.gplok	= !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1119 		.warn	= true,
1120 	};
1121 	int err;
1122 
1123 	/*
1124 	 * The module_mutex should not be a heavily contended lock;
1125 	 * if we get the occasional sleep here, we'll go an extra iteration
1126 	 * in the wait_event_interruptible(), which is harmless.
1127 	 */
1128 	sched_annotate_sleep();
1129 	mutex_lock(&module_mutex);
1130 	if (!find_symbol(&fsa))
1131 		goto unlock;
1132 
1133 	if (fsa.license == GPL_ONLY)
1134 		mod->using_gplonly_symbols = true;
1135 
1136 	if (!inherit_taint(mod, fsa.owner, name)) {
1137 		fsa.sym = NULL;
1138 		goto getname;
1139 	}
1140 
1141 	if (!check_version(info, name, mod, fsa.crc)) {
1142 		fsa.sym = ERR_PTR(-EINVAL);
1143 		goto getname;
1144 	}
1145 
1146 	err = verify_namespace_is_imported(info, fsa.sym, mod);
1147 	if (err) {
1148 		fsa.sym = ERR_PTR(err);
1149 		goto getname;
1150 	}
1151 
1152 	err = ref_module(mod, fsa.owner);
1153 	if (err) {
1154 		fsa.sym = ERR_PTR(err);
1155 		goto getname;
1156 	}
1157 
1158 getname:
1159 	/* We must make copy under the lock if we failed to get ref. */
1160 	strncpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1161 unlock:
1162 	mutex_unlock(&module_mutex);
1163 	return fsa.sym;
1164 }
1165 
1166 static const struct kernel_symbol *
resolve_symbol_wait(struct module * mod,const struct load_info * info,const char * name)1167 resolve_symbol_wait(struct module *mod,
1168 		    const struct load_info *info,
1169 		    const char *name)
1170 {
1171 	const struct kernel_symbol *ksym;
1172 	char owner[MODULE_NAME_LEN];
1173 
1174 	if (wait_event_interruptible_timeout(module_wq,
1175 			!IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1176 			|| PTR_ERR(ksym) != -EBUSY,
1177 					     30 * HZ) <= 0) {
1178 		pr_warn("%s: gave up waiting for init of module %s.\n",
1179 			mod->name, owner);
1180 	}
1181 	return ksym;
1182 }
1183 
module_arch_cleanup(struct module * mod)1184 void __weak module_arch_cleanup(struct module *mod)
1185 {
1186 }
1187 
module_arch_freeing_init(struct module * mod)1188 void __weak module_arch_freeing_init(struct module *mod)
1189 {
1190 }
1191 
module_memory_alloc(struct module * mod,enum mod_mem_type type)1192 static int module_memory_alloc(struct module *mod, enum mod_mem_type type)
1193 {
1194 	unsigned int size = PAGE_ALIGN(mod->mem[type].size);
1195 	enum execmem_type execmem_type;
1196 	void *ptr;
1197 
1198 	mod->mem[type].size = size;
1199 
1200 	if (mod_mem_type_is_data(type))
1201 		execmem_type = EXECMEM_MODULE_DATA;
1202 	else
1203 		execmem_type = EXECMEM_MODULE_TEXT;
1204 
1205 	ptr = execmem_alloc(execmem_type, size);
1206 	if (!ptr)
1207 		return -ENOMEM;
1208 
1209 	/*
1210 	 * The pointer to these blocks of memory are stored on the module
1211 	 * structure and we keep that around so long as the module is
1212 	 * around. We only free that memory when we unload the module.
1213 	 * Just mark them as not being a leak then. The .init* ELF
1214 	 * sections *do* get freed after boot so we *could* treat them
1215 	 * slightly differently with kmemleak_ignore() and only grey
1216 	 * them out as they work as typical memory allocations which
1217 	 * *do* eventually get freed, but let's just keep things simple
1218 	 * and avoid *any* false positives.
1219 	 */
1220 	kmemleak_not_leak(ptr);
1221 
1222 	memset(ptr, 0, size);
1223 	mod->mem[type].base = ptr;
1224 
1225 	return 0;
1226 }
1227 
module_memory_free(struct module * mod,enum mod_mem_type type,bool unload_codetags)1228 static void module_memory_free(struct module *mod, enum mod_mem_type type,
1229 			       bool unload_codetags)
1230 {
1231 	void *ptr = mod->mem[type].base;
1232 
1233 	if (!unload_codetags && mod_mem_type_is_core_data(type))
1234 		return;
1235 
1236 	execmem_free(ptr);
1237 }
1238 
free_mod_mem(struct module * mod,bool unload_codetags)1239 static void free_mod_mem(struct module *mod, bool unload_codetags)
1240 {
1241 	for_each_mod_mem_type(type) {
1242 		struct module_memory *mod_mem = &mod->mem[type];
1243 
1244 		if (type == MOD_DATA)
1245 			continue;
1246 
1247 		/* Free lock-classes; relies on the preceding sync_rcu(). */
1248 		lockdep_free_key_range(mod_mem->base, mod_mem->size);
1249 		if (mod_mem->size)
1250 			module_memory_free(mod, type, unload_codetags);
1251 	}
1252 
1253 	/* MOD_DATA hosts mod, so free it at last */
1254 	lockdep_free_key_range(mod->mem[MOD_DATA].base, mod->mem[MOD_DATA].size);
1255 	module_memory_free(mod, MOD_DATA, unload_codetags);
1256 }
1257 
1258 /* Free a module, remove from lists, etc. */
free_module(struct module * mod)1259 static void free_module(struct module *mod)
1260 {
1261 	bool unload_codetags;
1262 
1263 	trace_module_free(mod);
1264 
1265 	unload_codetags = codetag_unload_module(mod);
1266 	if (!unload_codetags)
1267 		pr_warn("%s: memory allocation(s) from the module still alive, cannot unload cleanly\n",
1268 			mod->name);
1269 
1270 	mod_sysfs_teardown(mod);
1271 
1272 	/*
1273 	 * We leave it in list to prevent duplicate loads, but make sure
1274 	 * that noone uses it while it's being deconstructed.
1275 	 */
1276 	mutex_lock(&module_mutex);
1277 	mod->state = MODULE_STATE_UNFORMED;
1278 	mutex_unlock(&module_mutex);
1279 
1280 	/* Arch-specific cleanup. */
1281 	module_arch_cleanup(mod);
1282 
1283 	/* Module unload stuff */
1284 	module_unload_free(mod);
1285 
1286 	/* Free any allocated parameters. */
1287 	destroy_params(mod->kp, mod->num_kp);
1288 
1289 	if (is_livepatch_module(mod))
1290 		free_module_elf(mod);
1291 
1292 	/* Now we can delete it from the lists */
1293 	mutex_lock(&module_mutex);
1294 	/* Unlink carefully: kallsyms could be walking list. */
1295 	list_del_rcu(&mod->list);
1296 	mod_tree_remove(mod);
1297 	/* Remove this module from bug list, this uses list_del_rcu */
1298 	module_bug_cleanup(mod);
1299 	/* Wait for RCU-sched synchronizing before releasing mod->list and buglist. */
1300 	synchronize_rcu();
1301 	if (try_add_tainted_module(mod))
1302 		pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1303 		       mod->name);
1304 	mutex_unlock(&module_mutex);
1305 
1306 	/* This may be empty, but that's OK */
1307 	module_arch_freeing_init(mod);
1308 	kfree(mod->args);
1309 	percpu_modfree(mod);
1310 
1311 	free_mod_mem(mod, unload_codetags);
1312 }
1313 
__symbol_get(const char * symbol)1314 void *__symbol_get(const char *symbol)
1315 {
1316 	struct find_symbol_arg fsa = {
1317 		.name	= symbol,
1318 		.gplok	= true,
1319 		.warn	= true,
1320 	};
1321 
1322 	preempt_disable();
1323 	if (!find_symbol(&fsa))
1324 		goto fail;
1325 	if (fsa.license != GPL_ONLY) {
1326 		pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1327 			symbol);
1328 		goto fail;
1329 	}
1330 	if (strong_try_module_get(fsa.owner))
1331 		goto fail;
1332 	preempt_enable();
1333 	return (void *)kernel_symbol_value(fsa.sym);
1334 fail:
1335 	preempt_enable();
1336 	return NULL;
1337 }
1338 EXPORT_SYMBOL_GPL(__symbol_get);
1339 
1340 /*
1341  * Ensure that an exported symbol [global namespace] does not already exist
1342  * in the kernel or in some other module's exported symbol table.
1343  *
1344  * You must hold the module_mutex.
1345  */
verify_exported_symbols(struct module * mod)1346 static int verify_exported_symbols(struct module *mod)
1347 {
1348 	unsigned int i;
1349 	const struct kernel_symbol *s;
1350 	struct {
1351 		const struct kernel_symbol *sym;
1352 		unsigned int num;
1353 	} arr[] = {
1354 		{ mod->syms, mod->num_syms },
1355 		{ mod->gpl_syms, mod->num_gpl_syms },
1356 	};
1357 
1358 	for (i = 0; i < ARRAY_SIZE(arr); i++) {
1359 		for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1360 			struct find_symbol_arg fsa = {
1361 				.name	= kernel_symbol_name(s),
1362 				.gplok	= true,
1363 			};
1364 			if (find_symbol(&fsa)) {
1365 				pr_err("%s: exports duplicate symbol %s"
1366 				       " (owned by %s)\n",
1367 				       mod->name, kernel_symbol_name(s),
1368 				       module_name(fsa.owner));
1369 				return -ENOEXEC;
1370 			}
1371 		}
1372 	}
1373 	return 0;
1374 }
1375 
ignore_undef_symbol(Elf_Half emachine,const char * name)1376 static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1377 {
1378 	/*
1379 	 * On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1380 	 * before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1381 	 * i386 has a similar problem but may not deserve a fix.
1382 	 *
1383 	 * If we ever have to ignore many symbols, consider refactoring the code to
1384 	 * only warn if referenced by a relocation.
1385 	 */
1386 	if (emachine == EM_386 || emachine == EM_X86_64)
1387 		return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1388 	return false;
1389 }
1390 
1391 /* Change all symbols so that st_value encodes the pointer directly. */
simplify_symbols(struct module * mod,const struct load_info * info)1392 static int simplify_symbols(struct module *mod, const struct load_info *info)
1393 {
1394 	Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1395 	Elf_Sym *sym = (void *)symsec->sh_addr;
1396 	unsigned long secbase;
1397 	unsigned int i;
1398 	int ret = 0;
1399 	const struct kernel_symbol *ksym;
1400 
1401 	for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1402 		const char *name = info->strtab + sym[i].st_name;
1403 
1404 		switch (sym[i].st_shndx) {
1405 		case SHN_COMMON:
1406 			/* Ignore common symbols */
1407 			if (!strncmp(name, "__gnu_lto", 9))
1408 				break;
1409 
1410 			/*
1411 			 * We compiled with -fno-common.  These are not
1412 			 * supposed to happen.
1413 			 */
1414 			pr_debug("Common symbol: %s\n", name);
1415 			pr_warn("%s: please compile with -fno-common\n",
1416 			       mod->name);
1417 			ret = -ENOEXEC;
1418 			break;
1419 
1420 		case SHN_ABS:
1421 			/* Don't need to do anything */
1422 			pr_debug("Absolute symbol: 0x%08lx %s\n",
1423 				 (long)sym[i].st_value, name);
1424 			break;
1425 
1426 		case SHN_LIVEPATCH:
1427 			/* Livepatch symbols are resolved by livepatch */
1428 			break;
1429 
1430 		case SHN_UNDEF:
1431 			ksym = resolve_symbol_wait(mod, info, name);
1432 			/* Ok if resolved.  */
1433 			if (ksym && !IS_ERR(ksym)) {
1434 				sym[i].st_value = kernel_symbol_value(ksym);
1435 				break;
1436 			}
1437 
1438 			/* Ok if weak or ignored.  */
1439 			if (!ksym &&
1440 			    (ELF_ST_BIND(sym[i].st_info) == STB_WEAK ||
1441 			     ignore_undef_symbol(info->hdr->e_machine, name)))
1442 				break;
1443 
1444 			ret = PTR_ERR(ksym) ?: -ENOENT;
1445 			pr_warn("%s: Unknown symbol %s (err %d)\n",
1446 				mod->name, name, ret);
1447 			break;
1448 
1449 		default:
1450 			/* Divert to percpu allocation if a percpu var. */
1451 			if (sym[i].st_shndx == info->index.pcpu)
1452 				secbase = (unsigned long)mod_percpu(mod);
1453 			else
1454 				secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1455 			sym[i].st_value += secbase;
1456 			break;
1457 		}
1458 	}
1459 
1460 	return ret;
1461 }
1462 
apply_relocations(struct module * mod,const struct load_info * info)1463 static int apply_relocations(struct module *mod, const struct load_info *info)
1464 {
1465 	unsigned int i;
1466 	int err = 0;
1467 
1468 	/* Now do relocations. */
1469 	for (i = 1; i < info->hdr->e_shnum; i++) {
1470 		unsigned int infosec = info->sechdrs[i].sh_info;
1471 
1472 		/* Not a valid relocation section? */
1473 		if (infosec >= info->hdr->e_shnum)
1474 			continue;
1475 
1476 		/* Don't bother with non-allocated sections */
1477 		if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC))
1478 			continue;
1479 
1480 		if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1481 			err = klp_apply_section_relocs(mod, info->sechdrs,
1482 						       info->secstrings,
1483 						       info->strtab,
1484 						       info->index.sym, i,
1485 						       NULL);
1486 		else if (info->sechdrs[i].sh_type == SHT_REL)
1487 			err = apply_relocate(info->sechdrs, info->strtab,
1488 					     info->index.sym, i, mod);
1489 		else if (info->sechdrs[i].sh_type == SHT_RELA)
1490 			err = apply_relocate_add(info->sechdrs, info->strtab,
1491 						 info->index.sym, i, mod);
1492 		if (err < 0)
1493 			break;
1494 	}
1495 	return err;
1496 }
1497 
1498 /* Additional bytes needed by arch in front of individual sections */
arch_mod_section_prepend(struct module * mod,unsigned int section)1499 unsigned int __weak arch_mod_section_prepend(struct module *mod,
1500 					     unsigned int section)
1501 {
1502 	/* default implementation just returns zero */
1503 	return 0;
1504 }
1505 
module_get_offset_and_type(struct module * mod,enum mod_mem_type type,Elf_Shdr * sechdr,unsigned int section)1506 long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1507 				Elf_Shdr *sechdr, unsigned int section)
1508 {
1509 	long offset;
1510 	long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1511 
1512 	mod->mem[type].size += arch_mod_section_prepend(mod, section);
1513 	offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1514 	mod->mem[type].size = offset + sechdr->sh_size;
1515 
1516 	WARN_ON_ONCE(offset & mask);
1517 	return offset | mask;
1518 }
1519 
module_init_layout_section(const char * sname)1520 bool module_init_layout_section(const char *sname)
1521 {
1522 #ifndef CONFIG_MODULE_UNLOAD
1523 	if (module_exit_section(sname))
1524 		return true;
1525 #endif
1526 	return module_init_section(sname);
1527 }
1528 
__layout_sections(struct module * mod,struct load_info * info,bool is_init)1529 static void __layout_sections(struct module *mod, struct load_info *info, bool is_init)
1530 {
1531 	unsigned int m, i;
1532 
1533 	static const unsigned long masks[][2] = {
1534 		/*
1535 		 * NOTE: all executable code must be the first section
1536 		 * in this array; otherwise modify the text_size
1537 		 * finder in the two loops below
1538 		 */
1539 		{ SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1540 		{ SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1541 		{ SHF_RO_AFTER_INIT | SHF_ALLOC, ARCH_SHF_SMALL },
1542 		{ SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1543 		{ ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1544 	};
1545 	static const int core_m_to_mem_type[] = {
1546 		MOD_TEXT,
1547 		MOD_RODATA,
1548 		MOD_RO_AFTER_INIT,
1549 		MOD_DATA,
1550 		MOD_DATA,
1551 	};
1552 	static const int init_m_to_mem_type[] = {
1553 		MOD_INIT_TEXT,
1554 		MOD_INIT_RODATA,
1555 		MOD_INVALID,
1556 		MOD_INIT_DATA,
1557 		MOD_INIT_DATA,
1558 	};
1559 
1560 	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1561 		enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1562 
1563 		for (i = 0; i < info->hdr->e_shnum; ++i) {
1564 			Elf_Shdr *s = &info->sechdrs[i];
1565 			const char *sname = info->secstrings + s->sh_name;
1566 
1567 			if ((s->sh_flags & masks[m][0]) != masks[m][0]
1568 			    || (s->sh_flags & masks[m][1])
1569 			    || s->sh_entsize != ~0UL
1570 			    || is_init != module_init_layout_section(sname))
1571 				continue;
1572 
1573 			if (WARN_ON_ONCE(type == MOD_INVALID))
1574 				continue;
1575 
1576 			s->sh_entsize = module_get_offset_and_type(mod, type, s, i);
1577 			pr_debug("\t%s\n", sname);
1578 		}
1579 	}
1580 }
1581 
1582 /*
1583  * Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1584  * might -- code, read-only data, read-write data, small data.  Tally
1585  * sizes, and place the offsets into sh_entsize fields: high bit means it
1586  * belongs in init.
1587  */
layout_sections(struct module * mod,struct load_info * info)1588 static void layout_sections(struct module *mod, struct load_info *info)
1589 {
1590 	unsigned int i;
1591 
1592 	for (i = 0; i < info->hdr->e_shnum; i++)
1593 		info->sechdrs[i].sh_entsize = ~0UL;
1594 
1595 	pr_debug("Core section allocation order for %s:\n", mod->name);
1596 	__layout_sections(mod, info, false);
1597 
1598 	pr_debug("Init section allocation order for %s:\n", mod->name);
1599 	__layout_sections(mod, info, true);
1600 }
1601 
module_license_taint_check(struct module * mod,const char * license)1602 static void module_license_taint_check(struct module *mod, const char *license)
1603 {
1604 	if (!license)
1605 		license = "unspecified";
1606 
1607 	if (!license_is_gpl_compatible(license)) {
1608 		if (!test_taint(TAINT_PROPRIETARY_MODULE))
1609 			pr_warn("%s: module license '%s' taints kernel.\n",
1610 				mod->name, license);
1611 		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1612 				 LOCKDEP_NOW_UNRELIABLE);
1613 	}
1614 }
1615 
setup_modinfo(struct module * mod,struct load_info * info)1616 static void setup_modinfo(struct module *mod, struct load_info *info)
1617 {
1618 	struct module_attribute *attr;
1619 	int i;
1620 
1621 	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1622 		if (attr->setup)
1623 			attr->setup(mod, get_modinfo(info, attr->attr.name));
1624 	}
1625 }
1626 
free_modinfo(struct module * mod)1627 static void free_modinfo(struct module *mod)
1628 {
1629 	struct module_attribute *attr;
1630 	int i;
1631 
1632 	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1633 		if (attr->free)
1634 			attr->free(mod);
1635 	}
1636 }
1637 
module_init_section(const char * name)1638 bool __weak module_init_section(const char *name)
1639 {
1640 	return strstarts(name, ".init");
1641 }
1642 
module_exit_section(const char * name)1643 bool __weak module_exit_section(const char *name)
1644 {
1645 	return strstarts(name, ".exit");
1646 }
1647 
validate_section_offset(struct load_info * info,Elf_Shdr * shdr)1648 static int validate_section_offset(struct load_info *info, Elf_Shdr *shdr)
1649 {
1650 #if defined(CONFIG_64BIT)
1651 	unsigned long long secend;
1652 #else
1653 	unsigned long secend;
1654 #endif
1655 
1656 	/*
1657 	 * Check for both overflow and offset/size being
1658 	 * too large.
1659 	 */
1660 	secend = shdr->sh_offset + shdr->sh_size;
1661 	if (secend < shdr->sh_offset || secend > info->len)
1662 		return -ENOEXEC;
1663 
1664 	return 0;
1665 }
1666 
1667 /*
1668  * Check userspace passed ELF module against our expectations, and cache
1669  * useful variables for further processing as we go.
1670  *
1671  * This does basic validity checks against section offsets and sizes, the
1672  * section name string table, and the indices used for it (sh_name).
1673  *
1674  * As a last step, since we're already checking the ELF sections we cache
1675  * useful variables which will be used later for our convenience:
1676  *
1677  * 	o pointers to section headers
1678  * 	o cache the modinfo symbol section
1679  * 	o cache the string symbol section
1680  * 	o cache the module section
1681  *
1682  * As a last step we set info->mod to the temporary copy of the module in
1683  * info->hdr. The final one will be allocated in move_module(). Any
1684  * modifications we make to our copy of the module will be carried over
1685  * to the final minted module.
1686  */
elf_validity_cache_copy(struct load_info * info,int flags)1687 static int elf_validity_cache_copy(struct load_info *info, int flags)
1688 {
1689 	unsigned int i;
1690 	Elf_Shdr *shdr, *strhdr;
1691 	int err;
1692 	unsigned int num_mod_secs = 0, mod_idx;
1693 	unsigned int num_info_secs = 0, info_idx;
1694 	unsigned int num_sym_secs = 0, sym_idx;
1695 
1696 	if (info->len < sizeof(*(info->hdr))) {
1697 		pr_err("Invalid ELF header len %lu\n", info->len);
1698 		goto no_exec;
1699 	}
1700 
1701 	if (memcmp(info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1702 		pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1703 		goto no_exec;
1704 	}
1705 	if (info->hdr->e_type != ET_REL) {
1706 		pr_err("Invalid ELF header type: %u != %u\n",
1707 		       info->hdr->e_type, ET_REL);
1708 		goto no_exec;
1709 	}
1710 	if (!elf_check_arch(info->hdr)) {
1711 		pr_err("Invalid architecture in ELF header: %u\n",
1712 		       info->hdr->e_machine);
1713 		goto no_exec;
1714 	}
1715 	if (!module_elf_check_arch(info->hdr)) {
1716 		pr_err("Invalid module architecture in ELF header: %u\n",
1717 		       info->hdr->e_machine);
1718 		goto no_exec;
1719 	}
1720 	if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1721 		pr_err("Invalid ELF section header size\n");
1722 		goto no_exec;
1723 	}
1724 
1725 	/*
1726 	 * e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1727 	 * known and small. So e_shnum * sizeof(Elf_Shdr)
1728 	 * will not overflow unsigned long on any platform.
1729 	 */
1730 	if (info->hdr->e_shoff >= info->len
1731 	    || (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1732 		info->len - info->hdr->e_shoff)) {
1733 		pr_err("Invalid ELF section header overflow\n");
1734 		goto no_exec;
1735 	}
1736 
1737 	info->sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1738 
1739 	/*
1740 	 * Verify if the section name table index is valid.
1741 	 */
1742 	if (info->hdr->e_shstrndx == SHN_UNDEF
1743 	    || info->hdr->e_shstrndx >= info->hdr->e_shnum) {
1744 		pr_err("Invalid ELF section name index: %d || e_shstrndx (%d) >= e_shnum (%d)\n",
1745 		       info->hdr->e_shstrndx, info->hdr->e_shstrndx,
1746 		       info->hdr->e_shnum);
1747 		goto no_exec;
1748 	}
1749 
1750 	strhdr = &info->sechdrs[info->hdr->e_shstrndx];
1751 	err = validate_section_offset(info, strhdr);
1752 	if (err < 0) {
1753 		pr_err("Invalid ELF section hdr(type %u)\n", strhdr->sh_type);
1754 		return err;
1755 	}
1756 
1757 	/*
1758 	 * The section name table must be NUL-terminated, as required
1759 	 * by the spec. This makes strcmp and pr_* calls that access
1760 	 * strings in the section safe.
1761 	 */
1762 	info->secstrings = (void *)info->hdr + strhdr->sh_offset;
1763 	if (strhdr->sh_size == 0) {
1764 		pr_err("empty section name table\n");
1765 		goto no_exec;
1766 	}
1767 	if (info->secstrings[strhdr->sh_size - 1] != '\0') {
1768 		pr_err("ELF Spec violation: section name table isn't null terminated\n");
1769 		goto no_exec;
1770 	}
1771 
1772 	/*
1773 	 * The code assumes that section 0 has a length of zero and
1774 	 * an addr of zero, so check for it.
1775 	 */
1776 	if (info->sechdrs[0].sh_type != SHT_NULL
1777 	    || info->sechdrs[0].sh_size != 0
1778 	    || info->sechdrs[0].sh_addr != 0) {
1779 		pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1780 		       info->sechdrs[0].sh_type);
1781 		goto no_exec;
1782 	}
1783 
1784 	for (i = 1; i < info->hdr->e_shnum; i++) {
1785 		shdr = &info->sechdrs[i];
1786 		switch (shdr->sh_type) {
1787 		case SHT_NULL:
1788 		case SHT_NOBITS:
1789 			continue;
1790 		case SHT_SYMTAB:
1791 			if (shdr->sh_link == SHN_UNDEF
1792 			    || shdr->sh_link >= info->hdr->e_shnum) {
1793 				pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
1794 				       shdr->sh_link, shdr->sh_link,
1795 				       info->hdr->e_shnum);
1796 				goto no_exec;
1797 			}
1798 			num_sym_secs++;
1799 			sym_idx = i;
1800 			fallthrough;
1801 		default:
1802 			err = validate_section_offset(info, shdr);
1803 			if (err < 0) {
1804 				pr_err("Invalid ELF section in module (section %u type %u)\n",
1805 					i, shdr->sh_type);
1806 				return err;
1807 			}
1808 			if (strcmp(info->secstrings + shdr->sh_name,
1809 				   ".gnu.linkonce.this_module") == 0) {
1810 				num_mod_secs++;
1811 				mod_idx = i;
1812 			} else if (strcmp(info->secstrings + shdr->sh_name,
1813 				   ".modinfo") == 0) {
1814 				num_info_secs++;
1815 				info_idx = i;
1816 			}
1817 
1818 			if (shdr->sh_flags & SHF_ALLOC) {
1819 				if (shdr->sh_name >= strhdr->sh_size) {
1820 					pr_err("Invalid ELF section name in module (section %u type %u)\n",
1821 					       i, shdr->sh_type);
1822 					return -ENOEXEC;
1823 				}
1824 			}
1825 			break;
1826 		}
1827 	}
1828 
1829 	if (num_info_secs > 1) {
1830 		pr_err("Only one .modinfo section must exist.\n");
1831 		goto no_exec;
1832 	} else if (num_info_secs == 1) {
1833 		/* Try to find a name early so we can log errors with a module name */
1834 		info->index.info = info_idx;
1835 		info->name = get_modinfo(info, "name");
1836 	}
1837 
1838 	if (num_sym_secs != 1) {
1839 		pr_warn("%s: module has no symbols (stripped?)\n",
1840 			info->name ?: "(missing .modinfo section or name field)");
1841 		goto no_exec;
1842 	}
1843 
1844 	/* Sets internal symbols and strings. */
1845 	info->index.sym = sym_idx;
1846 	shdr = &info->sechdrs[sym_idx];
1847 	info->index.str = shdr->sh_link;
1848 	info->strtab = (char *)info->hdr + info->sechdrs[info->index.str].sh_offset;
1849 
1850 	/*
1851 	 * The ".gnu.linkonce.this_module" ELF section is special. It is
1852 	 * what modpost uses to refer to __this_module and let's use rely
1853 	 * on THIS_MODULE to point to &__this_module properly. The kernel's
1854 	 * modpost declares it on each modules's *.mod.c file. If the struct
1855 	 * module of the kernel changes a full kernel rebuild is required.
1856 	 *
1857 	 * We have a few expectaions for this special section, the following
1858 	 * code validates all this for us:
1859 	 *
1860 	 *   o Only one section must exist
1861 	 *   o We expect the kernel to always have to allocate it: SHF_ALLOC
1862 	 *   o The section size must match the kernel's run time's struct module
1863 	 *     size
1864 	 */
1865 	if (num_mod_secs != 1) {
1866 		pr_err("module %s: Only one .gnu.linkonce.this_module section must exist.\n",
1867 		       info->name ?: "(missing .modinfo section or name field)");
1868 		goto no_exec;
1869 	}
1870 
1871 	shdr = &info->sechdrs[mod_idx];
1872 
1873 	/*
1874 	 * This is already implied on the switch above, however let's be
1875 	 * pedantic about it.
1876 	 */
1877 	if (shdr->sh_type == SHT_NOBITS) {
1878 		pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
1879 		       info->name ?: "(missing .modinfo section or name field)");
1880 		goto no_exec;
1881 	}
1882 
1883 	if (!(shdr->sh_flags & SHF_ALLOC)) {
1884 		pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
1885 		       info->name ?: "(missing .modinfo section or name field)");
1886 		goto no_exec;
1887 	}
1888 
1889 	if (shdr->sh_size != sizeof(struct module)) {
1890 		pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
1891 		       info->name ?: "(missing .modinfo section or name field)");
1892 		goto no_exec;
1893 	}
1894 
1895 	info->index.mod = mod_idx;
1896 
1897 	/* This is temporary: point mod into copy of data. */
1898 	info->mod = (void *)info->hdr + shdr->sh_offset;
1899 
1900 	/*
1901 	 * If we didn't load the .modinfo 'name' field earlier, fall back to
1902 	 * on-disk struct mod 'name' field.
1903 	 */
1904 	if (!info->name)
1905 		info->name = info->mod->name;
1906 
1907 	if (flags & MODULE_INIT_IGNORE_MODVERSIONS)
1908 		info->index.vers = 0; /* Pretend no __versions section! */
1909 	else
1910 		info->index.vers = find_sec(info, "__versions");
1911 
1912 	info->index.pcpu = find_pcpusec(info);
1913 
1914 	return 0;
1915 
1916 no_exec:
1917 	return -ENOEXEC;
1918 }
1919 
1920 #define COPY_CHUNK_SIZE (16*PAGE_SIZE)
1921 
copy_chunked_from_user(void * dst,const void __user * usrc,unsigned long len)1922 static int copy_chunked_from_user(void *dst, const void __user *usrc, unsigned long len)
1923 {
1924 	do {
1925 		unsigned long n = min(len, COPY_CHUNK_SIZE);
1926 
1927 		if (copy_from_user(dst, usrc, n) != 0)
1928 			return -EFAULT;
1929 		cond_resched();
1930 		dst += n;
1931 		usrc += n;
1932 		len -= n;
1933 	} while (len);
1934 	return 0;
1935 }
1936 
check_modinfo_livepatch(struct module * mod,struct load_info * info)1937 static int check_modinfo_livepatch(struct module *mod, struct load_info *info)
1938 {
1939 	if (!get_modinfo(info, "livepatch"))
1940 		/* Nothing more to do */
1941 		return 0;
1942 
1943 	if (set_livepatch_module(mod))
1944 		return 0;
1945 
1946 	pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
1947 	       mod->name);
1948 	return -ENOEXEC;
1949 }
1950 
check_modinfo_retpoline(struct module * mod,struct load_info * info)1951 static void check_modinfo_retpoline(struct module *mod, struct load_info *info)
1952 {
1953 	if (retpoline_module_ok(get_modinfo(info, "retpoline")))
1954 		return;
1955 
1956 	pr_warn("%s: loading module not compiled with retpoline compiler.\n",
1957 		mod->name);
1958 }
1959 
1960 /* Sets info->hdr and info->len. */
copy_module_from_user(const void __user * umod,unsigned long len,struct load_info * info)1961 static int copy_module_from_user(const void __user *umod, unsigned long len,
1962 				  struct load_info *info)
1963 {
1964 	int err;
1965 
1966 	info->len = len;
1967 	if (info->len < sizeof(*(info->hdr)))
1968 		return -ENOEXEC;
1969 
1970 	err = security_kernel_load_data(LOADING_MODULE, true);
1971 	if (err)
1972 		return err;
1973 
1974 	/* Suck in entire file: we'll want most of it. */
1975 	info->hdr = __vmalloc(info->len, GFP_KERNEL | __GFP_NOWARN);
1976 	if (!info->hdr)
1977 		return -ENOMEM;
1978 
1979 	if (copy_chunked_from_user(info->hdr, umod, info->len) != 0) {
1980 		err = -EFAULT;
1981 		goto out;
1982 	}
1983 
1984 	err = security_kernel_post_load_data((char *)info->hdr, info->len,
1985 					     LOADING_MODULE, "init_module");
1986 out:
1987 	if (err)
1988 		vfree(info->hdr);
1989 
1990 	return err;
1991 }
1992 
free_copy(struct load_info * info,int flags)1993 static void free_copy(struct load_info *info, int flags)
1994 {
1995 	if (flags & MODULE_INIT_COMPRESSED_FILE)
1996 		module_decompress_cleanup(info);
1997 	else
1998 		vfree(info->hdr);
1999 }
2000 
rewrite_section_headers(struct load_info * info,int flags)2001 static int rewrite_section_headers(struct load_info *info, int flags)
2002 {
2003 	unsigned int i;
2004 
2005 	/* This should always be true, but let's be sure. */
2006 	info->sechdrs[0].sh_addr = 0;
2007 
2008 	for (i = 1; i < info->hdr->e_shnum; i++) {
2009 		Elf_Shdr *shdr = &info->sechdrs[i];
2010 
2011 		/*
2012 		 * Mark all sections sh_addr with their address in the
2013 		 * temporary image.
2014 		 */
2015 		shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2016 
2017 	}
2018 
2019 	/* Track but don't keep modinfo and version sections. */
2020 	info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2021 	info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2022 
2023 	return 0;
2024 }
2025 
2026 /*
2027  * These calls taint the kernel depending certain module circumstances */
module_augment_kernel_taints(struct module * mod,struct load_info * info)2028 static void module_augment_kernel_taints(struct module *mod, struct load_info *info)
2029 {
2030 	int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2031 
2032 	if (!get_modinfo(info, "intree")) {
2033 		if (!test_taint(TAINT_OOT_MODULE))
2034 			pr_warn("%s: loading out-of-tree module taints kernel.\n",
2035 				mod->name);
2036 		add_taint_module(mod, TAINT_OOT_MODULE, LOCKDEP_STILL_OK);
2037 	}
2038 
2039 	check_modinfo_retpoline(mod, info);
2040 
2041 	if (get_modinfo(info, "staging")) {
2042 		add_taint_module(mod, TAINT_CRAP, LOCKDEP_STILL_OK);
2043 		pr_warn("%s: module is from the staging directory, the quality "
2044 			"is unknown, you have been warned.\n", mod->name);
2045 	}
2046 
2047 	if (is_livepatch_module(mod)) {
2048 		add_taint_module(mod, TAINT_LIVEPATCH, LOCKDEP_STILL_OK);
2049 		pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2050 				mod->name);
2051 	}
2052 
2053 	module_license_taint_check(mod, get_modinfo(info, "license"));
2054 
2055 	if (get_modinfo(info, "test")) {
2056 		if (!test_taint(TAINT_TEST))
2057 			pr_warn("%s: loading test module taints kernel.\n",
2058 				mod->name);
2059 		add_taint_module(mod, TAINT_TEST, LOCKDEP_STILL_OK);
2060 	}
2061 #ifdef CONFIG_MODULE_SIG
2062 	mod->sig_ok = info->sig_ok;
2063 	if (!mod->sig_ok) {
2064 		pr_notice_once("%s: module verification failed: signature "
2065 			       "and/or required key missing - tainting "
2066 			       "kernel\n", mod->name);
2067 		add_taint_module(mod, TAINT_UNSIGNED_MODULE, LOCKDEP_STILL_OK);
2068 	}
2069 #endif
2070 
2071 	/*
2072 	 * ndiswrapper is under GPL by itself, but loads proprietary modules.
2073 	 * Don't use add_taint_module(), as it would prevent ndiswrapper from
2074 	 * using GPL-only symbols it needs.
2075 	 */
2076 	if (strcmp(mod->name, "ndiswrapper") == 0)
2077 		add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2078 
2079 	/* driverloader was caught wrongly pretending to be under GPL */
2080 	if (strcmp(mod->name, "driverloader") == 0)
2081 		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2082 				 LOCKDEP_NOW_UNRELIABLE);
2083 
2084 	/* lve claims to be GPL but upstream won't provide source */
2085 	if (strcmp(mod->name, "lve") == 0)
2086 		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2087 				 LOCKDEP_NOW_UNRELIABLE);
2088 
2089 	if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2090 		pr_warn("%s: module license taints kernel.\n", mod->name);
2091 
2092 }
2093 
check_modinfo(struct module * mod,struct load_info * info,int flags)2094 static int check_modinfo(struct module *mod, struct load_info *info, int flags)
2095 {
2096 	const char *modmagic = get_modinfo(info, "vermagic");
2097 	int err;
2098 
2099 	if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2100 		modmagic = NULL;
2101 
2102 	/* This is allowed: modprobe --force will invalidate it. */
2103 	if (!modmagic) {
2104 		err = try_to_force_load(mod, "bad vermagic");
2105 		if (err)
2106 			return err;
2107 	} else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2108 		pr_err("%s: version magic '%s' should be '%s'\n",
2109 		       info->name, modmagic, vermagic);
2110 		return -ENOEXEC;
2111 	}
2112 
2113 	err = check_modinfo_livepatch(mod, info);
2114 	if (err)
2115 		return err;
2116 
2117 	return 0;
2118 }
2119 
find_module_sections(struct module * mod,struct load_info * info)2120 static int find_module_sections(struct module *mod, struct load_info *info)
2121 {
2122 	mod->kp = section_objs(info, "__param",
2123 			       sizeof(*mod->kp), &mod->num_kp);
2124 	mod->syms = section_objs(info, "__ksymtab",
2125 				 sizeof(*mod->syms), &mod->num_syms);
2126 	mod->crcs = section_addr(info, "__kcrctab");
2127 	mod->gpl_syms = section_objs(info, "__ksymtab_gpl",
2128 				     sizeof(*mod->gpl_syms),
2129 				     &mod->num_gpl_syms);
2130 	mod->gpl_crcs = section_addr(info, "__kcrctab_gpl");
2131 
2132 #ifdef CONFIG_CONSTRUCTORS
2133 	mod->ctors = section_objs(info, ".ctors",
2134 				  sizeof(*mod->ctors), &mod->num_ctors);
2135 	if (!mod->ctors)
2136 		mod->ctors = section_objs(info, ".init_array",
2137 				sizeof(*mod->ctors), &mod->num_ctors);
2138 	else if (find_sec(info, ".init_array")) {
2139 		/*
2140 		 * This shouldn't happen with same compiler and binutils
2141 		 * building all parts of the module.
2142 		 */
2143 		pr_warn("%s: has both .ctors and .init_array.\n",
2144 		       mod->name);
2145 		return -EINVAL;
2146 	}
2147 #endif
2148 
2149 	mod->noinstr_text_start = section_objs(info, ".noinstr.text", 1,
2150 						&mod->noinstr_text_size);
2151 
2152 #ifdef CONFIG_TRACEPOINTS
2153 	mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2154 					     sizeof(*mod->tracepoints_ptrs),
2155 					     &mod->num_tracepoints);
2156 #endif
2157 #ifdef CONFIG_TREE_SRCU
2158 	mod->srcu_struct_ptrs = section_objs(info, "___srcu_struct_ptrs",
2159 					     sizeof(*mod->srcu_struct_ptrs),
2160 					     &mod->num_srcu_structs);
2161 #endif
2162 #ifdef CONFIG_BPF_EVENTS
2163 	mod->bpf_raw_events = section_objs(info, "__bpf_raw_tp_map",
2164 					   sizeof(*mod->bpf_raw_events),
2165 					   &mod->num_bpf_raw_events);
2166 #endif
2167 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2168 	mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2169 	mod->btf_base_data = any_section_objs(info, ".BTF.base", 1,
2170 					      &mod->btf_base_data_size);
2171 #endif
2172 #ifdef CONFIG_JUMP_LABEL
2173 	mod->jump_entries = section_objs(info, "__jump_table",
2174 					sizeof(*mod->jump_entries),
2175 					&mod->num_jump_entries);
2176 #endif
2177 #ifdef CONFIG_EVENT_TRACING
2178 	mod->trace_events = section_objs(info, "_ftrace_events",
2179 					 sizeof(*mod->trace_events),
2180 					 &mod->num_trace_events);
2181 	mod->trace_evals = section_objs(info, "_ftrace_eval_map",
2182 					sizeof(*mod->trace_evals),
2183 					&mod->num_trace_evals);
2184 #endif
2185 #ifdef CONFIG_TRACING
2186 	mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2187 					 sizeof(*mod->trace_bprintk_fmt_start),
2188 					 &mod->num_trace_bprintk_fmt);
2189 #endif
2190 #ifdef CONFIG_FTRACE_MCOUNT_RECORD
2191 	/* sechdrs[0].sh_size is always zero */
2192 	mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2193 					     sizeof(*mod->ftrace_callsites),
2194 					     &mod->num_ftrace_callsites);
2195 #endif
2196 #ifdef CONFIG_FUNCTION_ERROR_INJECTION
2197 	mod->ei_funcs = section_objs(info, "_error_injection_whitelist",
2198 					    sizeof(*mod->ei_funcs),
2199 					    &mod->num_ei_funcs);
2200 #endif
2201 #ifdef CONFIG_KPROBES
2202 	mod->kprobes_text_start = section_objs(info, ".kprobes.text", 1,
2203 						&mod->kprobes_text_size);
2204 	mod->kprobe_blacklist = section_objs(info, "_kprobe_blacklist",
2205 						sizeof(unsigned long),
2206 						&mod->num_kprobe_blacklist);
2207 #endif
2208 #ifdef CONFIG_PRINTK_INDEX
2209 	mod->printk_index_start = section_objs(info, ".printk_index",
2210 					       sizeof(*mod->printk_index_start),
2211 					       &mod->printk_index_size);
2212 #endif
2213 #ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2214 	mod->static_call_sites = section_objs(info, ".static_call_sites",
2215 					      sizeof(*mod->static_call_sites),
2216 					      &mod->num_static_call_sites);
2217 #endif
2218 #if IS_ENABLED(CONFIG_KUNIT)
2219 	mod->kunit_suites = section_objs(info, ".kunit_test_suites",
2220 					      sizeof(*mod->kunit_suites),
2221 					      &mod->num_kunit_suites);
2222 	mod->kunit_init_suites = section_objs(info, ".kunit_init_test_suites",
2223 					      sizeof(*mod->kunit_init_suites),
2224 					      &mod->num_kunit_init_suites);
2225 #endif
2226 
2227 	mod->extable = section_objs(info, "__ex_table",
2228 				    sizeof(*mod->extable), &mod->num_exentries);
2229 
2230 	if (section_addr(info, "__obsparm"))
2231 		pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2232 
2233 #ifdef CONFIG_DYNAMIC_DEBUG_CORE
2234 	mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
2235 					      sizeof(*mod->dyndbg_info.descs),
2236 					      &mod->dyndbg_info.num_descs);
2237 	mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
2238 						sizeof(*mod->dyndbg_info.classes),
2239 						&mod->dyndbg_info.num_classes);
2240 #endif
2241 
2242 	return 0;
2243 }
2244 
move_module(struct module * mod,struct load_info * info)2245 static int move_module(struct module *mod, struct load_info *info)
2246 {
2247 	int i;
2248 	enum mod_mem_type t = 0;
2249 	int ret = -ENOMEM;
2250 
2251 	for_each_mod_mem_type(type) {
2252 		if (!mod->mem[type].size) {
2253 			mod->mem[type].base = NULL;
2254 			continue;
2255 		}
2256 
2257 		ret = module_memory_alloc(mod, type);
2258 		if (ret) {
2259 			t = type;
2260 			goto out_enomem;
2261 		}
2262 	}
2263 
2264 	/* Transfer each section which specifies SHF_ALLOC */
2265 	pr_debug("Final section addresses for %s:\n", mod->name);
2266 	for (i = 0; i < info->hdr->e_shnum; i++) {
2267 		void *dest;
2268 		Elf_Shdr *shdr = &info->sechdrs[i];
2269 		enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2270 
2271 		if (!(shdr->sh_flags & SHF_ALLOC))
2272 			continue;
2273 
2274 		dest = mod->mem[type].base + (shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK);
2275 
2276 		if (shdr->sh_type != SHT_NOBITS) {
2277 			/*
2278 			 * Our ELF checker already validated this, but let's
2279 			 * be pedantic and make the goal clearer. We actually
2280 			 * end up copying over all modifications made to the
2281 			 * userspace copy of the entire struct module.
2282 			 */
2283 			if (i == info->index.mod &&
2284 			   (WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2285 				ret = -ENOEXEC;
2286 				goto out_enomem;
2287 			}
2288 			memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2289 		}
2290 		/*
2291 		 * Update the userspace copy's ELF section address to point to
2292 		 * our newly allocated memory as a pure convenience so that
2293 		 * users of info can keep taking advantage and using the newly
2294 		 * minted official memory area.
2295 		 */
2296 		shdr->sh_addr = (unsigned long)dest;
2297 		pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2298 			 (long)shdr->sh_size, info->secstrings + shdr->sh_name);
2299 	}
2300 
2301 	return 0;
2302 out_enomem:
2303 	for (t--; t >= 0; t--)
2304 		module_memory_free(mod, t, true);
2305 	return ret;
2306 }
2307 
check_export_symbol_versions(struct module * mod)2308 static int check_export_symbol_versions(struct module *mod)
2309 {
2310 #ifdef CONFIG_MODVERSIONS
2311 	if ((mod->num_syms && !mod->crcs) ||
2312 	    (mod->num_gpl_syms && !mod->gpl_crcs)) {
2313 		return try_to_force_load(mod,
2314 					 "no versions for exported symbols");
2315 	}
2316 #endif
2317 	return 0;
2318 }
2319 
flush_module_icache(const struct module * mod)2320 static void flush_module_icache(const struct module *mod)
2321 {
2322 	/*
2323 	 * Flush the instruction cache, since we've played with text.
2324 	 * Do it before processing of module parameters, so the module
2325 	 * can provide parameter accessor functions of its own.
2326 	 */
2327 	for_each_mod_mem_type(type) {
2328 		const struct module_memory *mod_mem = &mod->mem[type];
2329 
2330 		if (mod_mem->size) {
2331 			flush_icache_range((unsigned long)mod_mem->base,
2332 					   (unsigned long)mod_mem->base + mod_mem->size);
2333 		}
2334 	}
2335 }
2336 
module_elf_check_arch(Elf_Ehdr * hdr)2337 bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2338 {
2339 	return true;
2340 }
2341 
module_frob_arch_sections(Elf_Ehdr * hdr,Elf_Shdr * sechdrs,char * secstrings,struct module * mod)2342 int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2343 				     Elf_Shdr *sechdrs,
2344 				     char *secstrings,
2345 				     struct module *mod)
2346 {
2347 	return 0;
2348 }
2349 
2350 /* module_blacklist is a comma-separated list of module names */
2351 static char *module_blacklist;
blacklisted(const char * module_name)2352 static bool blacklisted(const char *module_name)
2353 {
2354 	const char *p;
2355 	size_t len;
2356 
2357 	if (!module_blacklist)
2358 		return false;
2359 
2360 	for (p = module_blacklist; *p; p += len) {
2361 		len = strcspn(p, ",");
2362 		if (strlen(module_name) == len && !memcmp(module_name, p, len))
2363 			return true;
2364 		if (p[len] == ',')
2365 			len++;
2366 	}
2367 	return false;
2368 }
2369 core_param(module_blacklist, module_blacklist, charp, 0400);
2370 
layout_and_allocate(struct load_info * info,int flags)2371 static struct module *layout_and_allocate(struct load_info *info, int flags)
2372 {
2373 	struct module *mod;
2374 	unsigned int ndx;
2375 	int err;
2376 
2377 	/* Allow arches to frob section contents and sizes.  */
2378 	err = module_frob_arch_sections(info->hdr, info->sechdrs,
2379 					info->secstrings, info->mod);
2380 	if (err < 0)
2381 		return ERR_PTR(err);
2382 
2383 	err = module_enforce_rwx_sections(info->hdr, info->sechdrs,
2384 					  info->secstrings, info->mod);
2385 	if (err < 0)
2386 		return ERR_PTR(err);
2387 
2388 	/* We will do a special allocation for per-cpu sections later. */
2389 	info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2390 
2391 	/*
2392 	 * Mark ro_after_init section with SHF_RO_AFTER_INIT so that
2393 	 * layout_sections() can put it in the right place.
2394 	 * Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2395 	 */
2396 	ndx = find_sec(info, ".data..ro_after_init");
2397 	if (ndx)
2398 		info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2399 	/*
2400 	 * Mark the __jump_table section as ro_after_init as well: these data
2401 	 * structures are never modified, with the exception of entries that
2402 	 * refer to code in the __init section, which are annotated as such
2403 	 * at module load time.
2404 	 */
2405 	ndx = find_sec(info, "__jump_table");
2406 	if (ndx)
2407 		info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2408 
2409 	/*
2410 	 * Determine total sizes, and put offsets in sh_entsize.  For now
2411 	 * this is done generically; there doesn't appear to be any
2412 	 * special cases for the architectures.
2413 	 */
2414 	layout_sections(info->mod, info);
2415 	layout_symtab(info->mod, info);
2416 
2417 	/* Allocate and move to the final place */
2418 	err = move_module(info->mod, info);
2419 	if (err)
2420 		return ERR_PTR(err);
2421 
2422 	/* Module has been copied to its final place now: return it. */
2423 	mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2424 	kmemleak_load_module(mod, info);
2425 	return mod;
2426 }
2427 
2428 /* mod is no longer valid after this! */
module_deallocate(struct module * mod,struct load_info * info)2429 static void module_deallocate(struct module *mod, struct load_info *info)
2430 {
2431 	percpu_modfree(mod);
2432 	module_arch_freeing_init(mod);
2433 
2434 	free_mod_mem(mod, true);
2435 }
2436 
module_finalize(const Elf_Ehdr * hdr,const Elf_Shdr * sechdrs,struct module * me)2437 int __weak module_finalize(const Elf_Ehdr *hdr,
2438 			   const Elf_Shdr *sechdrs,
2439 			   struct module *me)
2440 {
2441 	return 0;
2442 }
2443 
post_relocation(struct module * mod,const struct load_info * info)2444 static int post_relocation(struct module *mod, const struct load_info *info)
2445 {
2446 	/* Sort exception table now relocations are done. */
2447 	sort_extable(mod->extable, mod->extable + mod->num_exentries);
2448 
2449 	/* Copy relocated percpu area over. */
2450 	percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
2451 		       info->sechdrs[info->index.pcpu].sh_size);
2452 
2453 	/* Setup kallsyms-specific fields. */
2454 	add_kallsyms(mod, info);
2455 
2456 	/* Arch-specific module finalizing. */
2457 	return module_finalize(info->hdr, info->sechdrs, mod);
2458 }
2459 
2460 /* Call module constructors. */
do_mod_ctors(struct module * mod)2461 static void do_mod_ctors(struct module *mod)
2462 {
2463 #ifdef CONFIG_CONSTRUCTORS
2464 	unsigned long i;
2465 
2466 	for (i = 0; i < mod->num_ctors; i++)
2467 		mod->ctors[i]();
2468 #endif
2469 }
2470 
2471 /* For freeing module_init on success, in case kallsyms traversing */
2472 struct mod_initfree {
2473 	struct llist_node node;
2474 	void *init_text;
2475 	void *init_data;
2476 	void *init_rodata;
2477 };
2478 
do_free_init(struct work_struct * w)2479 static void do_free_init(struct work_struct *w)
2480 {
2481 	struct llist_node *pos, *n, *list;
2482 	struct mod_initfree *initfree;
2483 
2484 	list = llist_del_all(&init_free_list);
2485 
2486 	synchronize_rcu();
2487 
2488 	llist_for_each_safe(pos, n, list) {
2489 		initfree = container_of(pos, struct mod_initfree, node);
2490 		execmem_free(initfree->init_text);
2491 		execmem_free(initfree->init_data);
2492 		execmem_free(initfree->init_rodata);
2493 		kfree(initfree);
2494 	}
2495 }
2496 
flush_module_init_free_work(void)2497 void flush_module_init_free_work(void)
2498 {
2499 	flush_work(&init_free_wq);
2500 }
2501 
2502 #undef MODULE_PARAM_PREFIX
2503 #define MODULE_PARAM_PREFIX "module."
2504 /* Default value for module->async_probe_requested */
2505 static bool async_probe;
2506 module_param(async_probe, bool, 0644);
2507 
2508 /*
2509  * This is where the real work happens.
2510  *
2511  * Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
2512  * helper command 'lx-symbols'.
2513  */
do_init_module(struct module * mod)2514 static noinline int do_init_module(struct module *mod)
2515 {
2516 	int ret = 0;
2517 	struct mod_initfree *freeinit;
2518 #if defined(CONFIG_MODULE_STATS)
2519 	unsigned int text_size = 0, total_size = 0;
2520 
2521 	for_each_mod_mem_type(type) {
2522 		const struct module_memory *mod_mem = &mod->mem[type];
2523 		if (mod_mem->size) {
2524 			total_size += mod_mem->size;
2525 			if (type == MOD_TEXT || type == MOD_INIT_TEXT)
2526 				text_size += mod_mem->size;
2527 		}
2528 	}
2529 #endif
2530 
2531 	freeinit = kmalloc(sizeof(*freeinit), GFP_KERNEL);
2532 	if (!freeinit) {
2533 		ret = -ENOMEM;
2534 		goto fail;
2535 	}
2536 	freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
2537 	freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
2538 	freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
2539 
2540 	do_mod_ctors(mod);
2541 	/* Start the module */
2542 	if (mod->init != NULL)
2543 		ret = do_one_initcall(mod->init);
2544 	if (ret < 0) {
2545 		goto fail_free_freeinit;
2546 	}
2547 	if (ret > 0) {
2548 		pr_warn("%s: '%s'->init suspiciously returned %d, it should "
2549 			"follow 0/-E convention\n"
2550 			"%s: loading module anyway...\n",
2551 			__func__, mod->name, ret, __func__);
2552 		dump_stack();
2553 	}
2554 
2555 	/* Now it's a first class citizen! */
2556 	mod->state = MODULE_STATE_LIVE;
2557 	blocking_notifier_call_chain(&module_notify_list,
2558 				     MODULE_STATE_LIVE, mod);
2559 
2560 	/* Delay uevent until module has finished its init routine */
2561 	kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
2562 
2563 	/*
2564 	 * We need to finish all async code before the module init sequence
2565 	 * is done. This has potential to deadlock if synchronous module
2566 	 * loading is requested from async (which is not allowed!).
2567 	 *
2568 	 * See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
2569 	 * request_module() from async workers") for more details.
2570 	 */
2571 	if (!mod->async_probe_requested)
2572 		async_synchronize_full();
2573 
2574 	ftrace_free_mem(mod, mod->mem[MOD_INIT_TEXT].base,
2575 			mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
2576 	mutex_lock(&module_mutex);
2577 	/* Drop initial reference. */
2578 	module_put(mod);
2579 	trim_init_extable(mod);
2580 #ifdef CONFIG_KALLSYMS
2581 	/* Switch to core kallsyms now init is done: kallsyms may be walking! */
2582 	rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
2583 #endif
2584 	ret = module_enable_rodata_ro(mod, true);
2585 	if (ret)
2586 		goto fail_mutex_unlock;
2587 	mod_tree_remove_init(mod);
2588 	module_arch_freeing_init(mod);
2589 	for_class_mod_mem_type(type, init) {
2590 		mod->mem[type].base = NULL;
2591 		mod->mem[type].size = 0;
2592 	}
2593 
2594 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2595 	/* .BTF is not SHF_ALLOC and will get removed, so sanitize pointers */
2596 	mod->btf_data = NULL;
2597 	mod->btf_base_data = NULL;
2598 #endif
2599 	/*
2600 	 * We want to free module_init, but be aware that kallsyms may be
2601 	 * walking this with preempt disabled.  In all the failure paths, we
2602 	 * call synchronize_rcu(), but we don't want to slow down the success
2603 	 * path. execmem_free() cannot be called in an interrupt, so do the
2604 	 * work and call synchronize_rcu() in a work queue.
2605 	 *
2606 	 * Note that execmem_alloc() on most architectures creates W+X page
2607 	 * mappings which won't be cleaned up until do_free_init() runs.  Any
2608 	 * code such as mark_rodata_ro() which depends on those mappings to
2609 	 * be cleaned up needs to sync with the queued work by invoking
2610 	 * flush_module_init_free_work().
2611 	 */
2612 	if (llist_add(&freeinit->node, &init_free_list))
2613 		schedule_work(&init_free_wq);
2614 
2615 	mutex_unlock(&module_mutex);
2616 	wake_up_all(&module_wq);
2617 
2618 	mod_stat_add_long(text_size, &total_text_size);
2619 	mod_stat_add_long(total_size, &total_mod_size);
2620 
2621 	mod_stat_inc(&modcount);
2622 
2623 	return 0;
2624 
2625 fail_mutex_unlock:
2626 	mutex_unlock(&module_mutex);
2627 fail_free_freeinit:
2628 	kfree(freeinit);
2629 fail:
2630 	/* Try to protect us from buggy refcounters. */
2631 	mod->state = MODULE_STATE_GOING;
2632 	synchronize_rcu();
2633 	module_put(mod);
2634 	blocking_notifier_call_chain(&module_notify_list,
2635 				     MODULE_STATE_GOING, mod);
2636 	klp_module_going(mod);
2637 	ftrace_release_mod(mod);
2638 	free_module(mod);
2639 	wake_up_all(&module_wq);
2640 
2641 	return ret;
2642 }
2643 
may_init_module(void)2644 static int may_init_module(void)
2645 {
2646 	if (!capable(CAP_SYS_MODULE) || modules_disabled)
2647 		return -EPERM;
2648 
2649 	return 0;
2650 }
2651 
2652 /* Is this module of this name done loading?  No locks held. */
finished_loading(const char * name)2653 static bool finished_loading(const char *name)
2654 {
2655 	struct module *mod;
2656 	bool ret;
2657 
2658 	/*
2659 	 * The module_mutex should not be a heavily contended lock;
2660 	 * if we get the occasional sleep here, we'll go an extra iteration
2661 	 * in the wait_event_interruptible(), which is harmless.
2662 	 */
2663 	sched_annotate_sleep();
2664 	mutex_lock(&module_mutex);
2665 	mod = find_module_all(name, strlen(name), true);
2666 	ret = !mod || mod->state == MODULE_STATE_LIVE
2667 		|| mod->state == MODULE_STATE_GOING;
2668 	mutex_unlock(&module_mutex);
2669 
2670 	return ret;
2671 }
2672 
2673 /* Must be called with module_mutex held */
module_patient_check_exists(const char * name,enum fail_dup_mod_reason reason)2674 static int module_patient_check_exists(const char *name,
2675 				       enum fail_dup_mod_reason reason)
2676 {
2677 	struct module *old;
2678 	int err = 0;
2679 
2680 	old = find_module_all(name, strlen(name), true);
2681 	if (old == NULL)
2682 		return 0;
2683 
2684 	if (old->state == MODULE_STATE_COMING ||
2685 	    old->state == MODULE_STATE_UNFORMED) {
2686 		/* Wait in case it fails to load. */
2687 		mutex_unlock(&module_mutex);
2688 		err = wait_event_interruptible(module_wq,
2689 				       finished_loading(name));
2690 		mutex_lock(&module_mutex);
2691 		if (err)
2692 			return err;
2693 
2694 		/* The module might have gone in the meantime. */
2695 		old = find_module_all(name, strlen(name), true);
2696 	}
2697 
2698 	if (try_add_failed_module(name, reason))
2699 		pr_warn("Could not add fail-tracking for module: %s\n", name);
2700 
2701 	/*
2702 	 * We are here only when the same module was being loaded. Do
2703 	 * not try to load it again right now. It prevents long delays
2704 	 * caused by serialized module load failures. It might happen
2705 	 * when more devices of the same type trigger load of
2706 	 * a particular module.
2707 	 */
2708 	if (old && old->state == MODULE_STATE_LIVE)
2709 		return -EEXIST;
2710 	return -EBUSY;
2711 }
2712 
2713 /*
2714  * We try to place it in the list now to make sure it's unique before
2715  * we dedicate too many resources.  In particular, temporary percpu
2716  * memory exhaustion.
2717  */
add_unformed_module(struct module * mod)2718 static int add_unformed_module(struct module *mod)
2719 {
2720 	int err;
2721 
2722 	mod->state = MODULE_STATE_UNFORMED;
2723 
2724 	mutex_lock(&module_mutex);
2725 	err = module_patient_check_exists(mod->name, FAIL_DUP_MOD_LOAD);
2726 	if (err)
2727 		goto out;
2728 
2729 	mod_update_bounds(mod);
2730 	list_add_rcu(&mod->list, &modules);
2731 	mod_tree_insert(mod);
2732 	err = 0;
2733 
2734 out:
2735 	mutex_unlock(&module_mutex);
2736 	return err;
2737 }
2738 
complete_formation(struct module * mod,struct load_info * info)2739 static int complete_formation(struct module *mod, struct load_info *info)
2740 {
2741 	int err;
2742 
2743 	mutex_lock(&module_mutex);
2744 
2745 	/* Find duplicate symbols (must be called under lock). */
2746 	err = verify_exported_symbols(mod);
2747 	if (err < 0)
2748 		goto out;
2749 
2750 	/* These rely on module_mutex for list integrity. */
2751 	module_bug_finalize(info->hdr, info->sechdrs, mod);
2752 	module_cfi_finalize(info->hdr, info->sechdrs, mod);
2753 
2754 	err = module_enable_rodata_ro(mod, false);
2755 	if (err)
2756 		goto out_strict_rwx;
2757 	err = module_enable_data_nx(mod);
2758 	if (err)
2759 		goto out_strict_rwx;
2760 	err = module_enable_text_rox(mod);
2761 	if (err)
2762 		goto out_strict_rwx;
2763 
2764 	/*
2765 	 * Mark state as coming so strong_try_module_get() ignores us,
2766 	 * but kallsyms etc. can see us.
2767 	 */
2768 	mod->state = MODULE_STATE_COMING;
2769 	mutex_unlock(&module_mutex);
2770 
2771 	return 0;
2772 
2773 out_strict_rwx:
2774 	module_bug_cleanup(mod);
2775 out:
2776 	mutex_unlock(&module_mutex);
2777 	return err;
2778 }
2779 
prepare_coming_module(struct module * mod)2780 static int prepare_coming_module(struct module *mod)
2781 {
2782 	int err;
2783 
2784 	ftrace_module_enable(mod);
2785 	err = klp_module_coming(mod);
2786 	if (err)
2787 		return err;
2788 
2789 	err = blocking_notifier_call_chain_robust(&module_notify_list,
2790 			MODULE_STATE_COMING, MODULE_STATE_GOING, mod);
2791 	err = notifier_to_errno(err);
2792 	if (err)
2793 		klp_module_going(mod);
2794 
2795 	return err;
2796 }
2797 
unknown_module_param_cb(char * param,char * val,const char * modname,void * arg)2798 static int unknown_module_param_cb(char *param, char *val, const char *modname,
2799 				   void *arg)
2800 {
2801 	struct module *mod = arg;
2802 	int ret;
2803 
2804 	if (strcmp(param, "async_probe") == 0) {
2805 		if (kstrtobool(val, &mod->async_probe_requested))
2806 			mod->async_probe_requested = true;
2807 		return 0;
2808 	}
2809 
2810 	/* Check for magic 'dyndbg' arg */
2811 	ret = ddebug_dyndbg_module_param_cb(param, val, modname);
2812 	if (ret != 0)
2813 		pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
2814 	return 0;
2815 }
2816 
2817 /* Module within temporary copy, this doesn't do any allocation  */
early_mod_check(struct load_info * info,int flags)2818 static int early_mod_check(struct load_info *info, int flags)
2819 {
2820 	int err;
2821 
2822 	/*
2823 	 * Now that we know we have the correct module name, check
2824 	 * if it's blacklisted.
2825 	 */
2826 	if (blacklisted(info->name)) {
2827 		pr_err("Module %s is blacklisted\n", info->name);
2828 		return -EPERM;
2829 	}
2830 
2831 	err = rewrite_section_headers(info, flags);
2832 	if (err)
2833 		return err;
2834 
2835 	/* Check module struct version now, before we try to use module. */
2836 	if (!check_modstruct_version(info, info->mod))
2837 		return -ENOEXEC;
2838 
2839 	err = check_modinfo(info->mod, info, flags);
2840 	if (err)
2841 		return err;
2842 
2843 	mutex_lock(&module_mutex);
2844 	err = module_patient_check_exists(info->mod->name, FAIL_DUP_MOD_BECOMING);
2845 	mutex_unlock(&module_mutex);
2846 
2847 	return err;
2848 }
2849 
2850 /*
2851  * Allocate and load the module: note that size of section 0 is always
2852  * zero, and we rely on this for optional sections.
2853  */
load_module(struct load_info * info,const char __user * uargs,int flags)2854 static int load_module(struct load_info *info, const char __user *uargs,
2855 		       int flags)
2856 {
2857 	struct module *mod;
2858 	bool module_allocated = false;
2859 	long err = 0;
2860 	char *after_dashes;
2861 
2862 	/*
2863 	 * Do the signature check (if any) first. All that
2864 	 * the signature check needs is info->len, it does
2865 	 * not need any of the section info. That can be
2866 	 * set up later. This will minimize the chances
2867 	 * of a corrupt module causing problems before
2868 	 * we even get to the signature check.
2869 	 *
2870 	 * The check will also adjust info->len by stripping
2871 	 * off the sig length at the end of the module, making
2872 	 * checks against info->len more correct.
2873 	 */
2874 	err = module_sig_check(info, flags);
2875 	if (err)
2876 		goto free_copy;
2877 
2878 	/*
2879 	 * Do basic sanity checks against the ELF header and
2880 	 * sections. Cache useful sections and set the
2881 	 * info->mod to the userspace passed struct module.
2882 	 */
2883 	err = elf_validity_cache_copy(info, flags);
2884 	if (err)
2885 		goto free_copy;
2886 
2887 	err = early_mod_check(info, flags);
2888 	if (err)
2889 		goto free_copy;
2890 
2891 	/* Figure out module layout, and allocate all the memory. */
2892 	mod = layout_and_allocate(info, flags);
2893 	if (IS_ERR(mod)) {
2894 		err = PTR_ERR(mod);
2895 		goto free_copy;
2896 	}
2897 
2898 	module_allocated = true;
2899 
2900 	audit_log_kern_module(mod->name);
2901 
2902 	/* Reserve our place in the list. */
2903 	err = add_unformed_module(mod);
2904 	if (err)
2905 		goto free_module;
2906 
2907 	/*
2908 	 * We are tainting your kernel if your module gets into
2909 	 * the modules linked list somehow.
2910 	 */
2911 	module_augment_kernel_taints(mod, info);
2912 
2913 	/* To avoid stressing percpu allocator, do this once we're unique. */
2914 	err = percpu_modalloc(mod, info);
2915 	if (err)
2916 		goto unlink_mod;
2917 
2918 	/* Now module is in final location, initialize linked lists, etc. */
2919 	err = module_unload_init(mod);
2920 	if (err)
2921 		goto unlink_mod;
2922 
2923 	init_param_lock(mod);
2924 
2925 	/*
2926 	 * Now we've got everything in the final locations, we can
2927 	 * find optional sections.
2928 	 */
2929 	err = find_module_sections(mod, info);
2930 	if (err)
2931 		goto free_unload;
2932 
2933 	err = check_export_symbol_versions(mod);
2934 	if (err)
2935 		goto free_unload;
2936 
2937 	/* Set up MODINFO_ATTR fields */
2938 	setup_modinfo(mod, info);
2939 
2940 	/* Fix up syms, so that st_value is a pointer to location. */
2941 	err = simplify_symbols(mod, info);
2942 	if (err < 0)
2943 		goto free_modinfo;
2944 
2945 	err = apply_relocations(mod, info);
2946 	if (err < 0)
2947 		goto free_modinfo;
2948 
2949 	err = post_relocation(mod, info);
2950 	if (err < 0)
2951 		goto free_modinfo;
2952 
2953 	flush_module_icache(mod);
2954 
2955 	/* Now copy in args */
2956 	mod->args = strndup_user(uargs, ~0UL >> 1);
2957 	if (IS_ERR(mod->args)) {
2958 		err = PTR_ERR(mod->args);
2959 		goto free_arch_cleanup;
2960 	}
2961 
2962 	init_build_id(mod, info);
2963 
2964 	/* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
2965 	ftrace_module_init(mod);
2966 
2967 	/* Finally it's fully formed, ready to start executing. */
2968 	err = complete_formation(mod, info);
2969 	if (err)
2970 		goto ddebug_cleanup;
2971 
2972 	err = prepare_coming_module(mod);
2973 	if (err)
2974 		goto bug_cleanup;
2975 
2976 	mod->async_probe_requested = async_probe;
2977 
2978 	/* Module is ready to execute: parsing args may do that. */
2979 	after_dashes = parse_args(mod->name, mod->args, mod->kp, mod->num_kp,
2980 				  -32768, 32767, mod,
2981 				  unknown_module_param_cb);
2982 	if (IS_ERR(after_dashes)) {
2983 		err = PTR_ERR(after_dashes);
2984 		goto coming_cleanup;
2985 	} else if (after_dashes) {
2986 		pr_warn("%s: parameters '%s' after `--' ignored\n",
2987 		       mod->name, after_dashes);
2988 	}
2989 
2990 	/* Link in to sysfs. */
2991 	err = mod_sysfs_setup(mod, info, mod->kp, mod->num_kp);
2992 	if (err < 0)
2993 		goto coming_cleanup;
2994 
2995 	if (is_livepatch_module(mod)) {
2996 		err = copy_module_elf(mod, info);
2997 		if (err < 0)
2998 			goto sysfs_cleanup;
2999 	}
3000 
3001 	/* Get rid of temporary copy. */
3002 	free_copy(info, flags);
3003 
3004 	codetag_load_module(mod);
3005 
3006 	/* Done! */
3007 	trace_module_load(mod);
3008 
3009 	return do_init_module(mod);
3010 
3011  sysfs_cleanup:
3012 	mod_sysfs_teardown(mod);
3013  coming_cleanup:
3014 	mod->state = MODULE_STATE_GOING;
3015 	destroy_params(mod->kp, mod->num_kp);
3016 	blocking_notifier_call_chain(&module_notify_list,
3017 				     MODULE_STATE_GOING, mod);
3018 	klp_module_going(mod);
3019  bug_cleanup:
3020 	mod->state = MODULE_STATE_GOING;
3021 	/* module_bug_cleanup needs module_mutex protection */
3022 	mutex_lock(&module_mutex);
3023 	module_bug_cleanup(mod);
3024 	mutex_unlock(&module_mutex);
3025 
3026  ddebug_cleanup:
3027 	ftrace_release_mod(mod);
3028 	synchronize_rcu();
3029 	kfree(mod->args);
3030  free_arch_cleanup:
3031 	module_arch_cleanup(mod);
3032  free_modinfo:
3033 	free_modinfo(mod);
3034  free_unload:
3035 	module_unload_free(mod);
3036  unlink_mod:
3037 	mutex_lock(&module_mutex);
3038 	/* Unlink carefully: kallsyms could be walking list. */
3039 	list_del_rcu(&mod->list);
3040 	mod_tree_remove(mod);
3041 	wake_up_all(&module_wq);
3042 	/* Wait for RCU-sched synchronizing before releasing mod->list. */
3043 	synchronize_rcu();
3044 	mutex_unlock(&module_mutex);
3045  free_module:
3046 	mod_stat_bump_invalid(info, flags);
3047 	/* Free lock-classes; relies on the preceding sync_rcu() */
3048 	for_class_mod_mem_type(type, core_data) {
3049 		lockdep_free_key_range(mod->mem[type].base,
3050 				       mod->mem[type].size);
3051 	}
3052 
3053 	module_deallocate(mod, info);
3054  free_copy:
3055 	/*
3056 	 * The info->len is always set. We distinguish between
3057 	 * failures once the proper module was allocated and
3058 	 * before that.
3059 	 */
3060 	if (!module_allocated)
3061 		mod_stat_bump_becoming(info, flags);
3062 	free_copy(info, flags);
3063 	return err;
3064 }
3065 
SYSCALL_DEFINE3(init_module,void __user *,umod,unsigned long,len,const char __user *,uargs)3066 SYSCALL_DEFINE3(init_module, void __user *, umod,
3067 		unsigned long, len, const char __user *, uargs)
3068 {
3069 	int err;
3070 	struct load_info info = { };
3071 
3072 	err = may_init_module();
3073 	if (err)
3074 		return err;
3075 
3076 	pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3077 	       umod, len, uargs);
3078 
3079 	err = copy_module_from_user(umod, len, &info);
3080 	if (err) {
3081 		mod_stat_inc(&failed_kreads);
3082 		mod_stat_add_long(len, &invalid_kread_bytes);
3083 		return err;
3084 	}
3085 
3086 	return load_module(&info, uargs, 0);
3087 }
3088 
3089 struct idempotent {
3090 	const void *cookie;
3091 	struct hlist_node entry;
3092 	struct completion complete;
3093 	int ret;
3094 };
3095 
3096 #define IDEM_HASH_BITS 8
3097 static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3098 static DEFINE_SPINLOCK(idem_lock);
3099 
idempotent(struct idempotent * u,const void * cookie)3100 static bool idempotent(struct idempotent *u, const void *cookie)
3101 {
3102 	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3103 	struct hlist_head *head = idem_hash + hash;
3104 	struct idempotent *existing;
3105 	bool first;
3106 
3107 	u->ret = -EINTR;
3108 	u->cookie = cookie;
3109 	init_completion(&u->complete);
3110 
3111 	spin_lock(&idem_lock);
3112 	first = true;
3113 	hlist_for_each_entry(existing, head, entry) {
3114 		if (existing->cookie != cookie)
3115 			continue;
3116 		first = false;
3117 		break;
3118 	}
3119 	hlist_add_head(&u->entry, idem_hash + hash);
3120 	spin_unlock(&idem_lock);
3121 
3122 	return !first;
3123 }
3124 
3125 /*
3126  * We were the first one with 'cookie' on the list, and we ended
3127  * up completing the operation. We now need to walk the list,
3128  * remove everybody - which includes ourselves - fill in the return
3129  * value, and then complete the operation.
3130  */
idempotent_complete(struct idempotent * u,int ret)3131 static int idempotent_complete(struct idempotent *u, int ret)
3132 {
3133 	const void *cookie = u->cookie;
3134 	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3135 	struct hlist_head *head = idem_hash + hash;
3136 	struct hlist_node *next;
3137 	struct idempotent *pos;
3138 
3139 	spin_lock(&idem_lock);
3140 	hlist_for_each_entry_safe(pos, next, head, entry) {
3141 		if (pos->cookie != cookie)
3142 			continue;
3143 		hlist_del_init(&pos->entry);
3144 		pos->ret = ret;
3145 		complete(&pos->complete);
3146 	}
3147 	spin_unlock(&idem_lock);
3148 	return ret;
3149 }
3150 
3151 /*
3152  * Wait for the idempotent worker.
3153  *
3154  * If we get interrupted, we need to remove ourselves from the
3155  * the idempotent list, and the completion may still come in.
3156  *
3157  * The 'idem_lock' protects against the race, and 'idem.ret' was
3158  * initialized to -EINTR and is thus always the right return
3159  * value even if the idempotent work then completes between
3160  * the wait_for_completion and the cleanup.
3161  */
idempotent_wait_for_completion(struct idempotent * u)3162 static int idempotent_wait_for_completion(struct idempotent *u)
3163 {
3164 	if (wait_for_completion_interruptible(&u->complete)) {
3165 		spin_lock(&idem_lock);
3166 		if (!hlist_unhashed(&u->entry))
3167 			hlist_del(&u->entry);
3168 		spin_unlock(&idem_lock);
3169 	}
3170 	return u->ret;
3171 }
3172 
init_module_from_file(struct file * f,const char __user * uargs,int flags)3173 static int init_module_from_file(struct file *f, const char __user * uargs, int flags)
3174 {
3175 	struct load_info info = { };
3176 	void *buf = NULL;
3177 	int len;
3178 
3179 	len = kernel_read_file(f, 0, &buf, INT_MAX, NULL, READING_MODULE);
3180 	if (len < 0) {
3181 		mod_stat_inc(&failed_kreads);
3182 		return len;
3183 	}
3184 
3185 	if (flags & MODULE_INIT_COMPRESSED_FILE) {
3186 		int err = module_decompress(&info, buf, len);
3187 		vfree(buf); /* compressed data is no longer needed */
3188 		if (err) {
3189 			mod_stat_inc(&failed_decompress);
3190 			mod_stat_add_long(len, &invalid_decompress_bytes);
3191 			return err;
3192 		}
3193 	} else {
3194 		info.hdr = buf;
3195 		info.len = len;
3196 	}
3197 
3198 	return load_module(&info, uargs, flags);
3199 }
3200 
idempotent_init_module(struct file * f,const char __user * uargs,int flags)3201 static int idempotent_init_module(struct file *f, const char __user * uargs, int flags)
3202 {
3203 	struct idempotent idem;
3204 
3205 	if (!f || !(f->f_mode & FMODE_READ))
3206 		return -EBADF;
3207 
3208 	/* Are we the winners of the race and get to do this? */
3209 	if (!idempotent(&idem, file_inode(f))) {
3210 		int ret = init_module_from_file(f, uargs, flags);
3211 		return idempotent_complete(&idem, ret);
3212 	}
3213 
3214 	/*
3215 	 * Somebody else won the race and is loading the module.
3216 	 */
3217 	return idempotent_wait_for_completion(&idem);
3218 }
3219 
SYSCALL_DEFINE3(finit_module,int,fd,const char __user *,uargs,int,flags)3220 SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3221 {
3222 	int err;
3223 	struct fd f;
3224 
3225 	err = may_init_module();
3226 	if (err)
3227 		return err;
3228 
3229 	pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3230 
3231 	if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3232 		      |MODULE_INIT_IGNORE_VERMAGIC
3233 		      |MODULE_INIT_COMPRESSED_FILE))
3234 		return -EINVAL;
3235 
3236 	f = fdget(fd);
3237 	err = idempotent_init_module(fd_file(f), uargs, flags);
3238 	fdput(f);
3239 	return err;
3240 }
3241 
3242 /* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
module_flags(struct module * mod,char * buf,bool show_state)3243 char *module_flags(struct module *mod, char *buf, bool show_state)
3244 {
3245 	int bx = 0;
3246 
3247 	BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3248 	if (!mod->taints && !show_state)
3249 		goto out;
3250 	if (mod->taints ||
3251 	    mod->state == MODULE_STATE_GOING ||
3252 	    mod->state == MODULE_STATE_COMING) {
3253 		buf[bx++] = '(';
3254 		bx += module_flags_taint(mod->taints, buf + bx);
3255 		/* Show a - for module-is-being-unloaded */
3256 		if (mod->state == MODULE_STATE_GOING && show_state)
3257 			buf[bx++] = '-';
3258 		/* Show a + for module-is-being-loaded */
3259 		if (mod->state == MODULE_STATE_COMING && show_state)
3260 			buf[bx++] = '+';
3261 		buf[bx++] = ')';
3262 	}
3263 out:
3264 	buf[bx] = '\0';
3265 
3266 	return buf;
3267 }
3268 
3269 /* Given an address, look for it in the module exception tables. */
search_module_extables(unsigned long addr)3270 const struct exception_table_entry *search_module_extables(unsigned long addr)
3271 {
3272 	const struct exception_table_entry *e = NULL;
3273 	struct module *mod;
3274 
3275 	preempt_disable();
3276 	mod = __module_address(addr);
3277 	if (!mod)
3278 		goto out;
3279 
3280 	if (!mod->num_exentries)
3281 		goto out;
3282 
3283 	e = search_extable(mod->extable,
3284 			   mod->num_exentries,
3285 			   addr);
3286 out:
3287 	preempt_enable();
3288 
3289 	/*
3290 	 * Now, if we found one, we are running inside it now, hence
3291 	 * we cannot unload the module, hence no refcnt needed.
3292 	 */
3293 	return e;
3294 }
3295 
3296 /**
3297  * is_module_address() - is this address inside a module?
3298  * @addr: the address to check.
3299  *
3300  * See is_module_text_address() if you simply want to see if the address
3301  * is code (not data).
3302  */
is_module_address(unsigned long addr)3303 bool is_module_address(unsigned long addr)
3304 {
3305 	bool ret;
3306 
3307 	preempt_disable();
3308 	ret = __module_address(addr) != NULL;
3309 	preempt_enable();
3310 
3311 	return ret;
3312 }
3313 
3314 /**
3315  * __module_address() - get the module which contains an address.
3316  * @addr: the address.
3317  *
3318  * Must be called with preempt disabled or module mutex held so that
3319  * module doesn't get freed during this.
3320  */
__module_address(unsigned long addr)3321 struct module *__module_address(unsigned long addr)
3322 {
3323 	struct module *mod;
3324 
3325 	if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3326 		goto lookup;
3327 
3328 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3329 	if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3330 		goto lookup;
3331 #endif
3332 
3333 	return NULL;
3334 
3335 lookup:
3336 	module_assert_mutex_or_preempt();
3337 
3338 	mod = mod_find(addr, &mod_tree);
3339 	if (mod) {
3340 		BUG_ON(!within_module(addr, mod));
3341 		if (mod->state == MODULE_STATE_UNFORMED)
3342 			mod = NULL;
3343 	}
3344 	return mod;
3345 }
3346 
3347 /**
3348  * is_module_text_address() - is this address inside module code?
3349  * @addr: the address to check.
3350  *
3351  * See is_module_address() if you simply want to see if the address is
3352  * anywhere in a module.  See kernel_text_address() for testing if an
3353  * address corresponds to kernel or module code.
3354  */
is_module_text_address(unsigned long addr)3355 bool is_module_text_address(unsigned long addr)
3356 {
3357 	bool ret;
3358 
3359 	preempt_disable();
3360 	ret = __module_text_address(addr) != NULL;
3361 	preempt_enable();
3362 
3363 	return ret;
3364 }
3365 
3366 /**
3367  * __module_text_address() - get the module whose code contains an address.
3368  * @addr: the address.
3369  *
3370  * Must be called with preempt disabled or module mutex held so that
3371  * module doesn't get freed during this.
3372  */
__module_text_address(unsigned long addr)3373 struct module *__module_text_address(unsigned long addr)
3374 {
3375 	struct module *mod = __module_address(addr);
3376 	if (mod) {
3377 		/* Make sure it's within the text section. */
3378 		if (!within_module_mem_type(addr, mod, MOD_TEXT) &&
3379 		    !within_module_mem_type(addr, mod, MOD_INIT_TEXT))
3380 			mod = NULL;
3381 	}
3382 	return mod;
3383 }
3384 
3385 /* Don't grab lock, we're oopsing. */
print_modules(void)3386 void print_modules(void)
3387 {
3388 	struct module *mod;
3389 	char buf[MODULE_FLAGS_BUF_SIZE];
3390 
3391 	printk(KERN_DEFAULT "Modules linked in:");
3392 	/* Most callers should already have preempt disabled, but make sure */
3393 	preempt_disable();
3394 	list_for_each_entry_rcu(mod, &modules, list) {
3395 		if (mod->state == MODULE_STATE_UNFORMED)
3396 			continue;
3397 		pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3398 	}
3399 
3400 	print_unloaded_tainted_modules();
3401 	preempt_enable();
3402 	if (last_unloaded_module.name[0])
3403 		pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3404 			last_unloaded_module.taints);
3405 	pr_cont("\n");
3406 }
3407 
3408 #ifdef CONFIG_MODULE_DEBUGFS
3409 struct dentry *mod_debugfs_root;
3410 
module_debugfs_init(void)3411 static int module_debugfs_init(void)
3412 {
3413 	mod_debugfs_root = debugfs_create_dir("modules", NULL);
3414 	return 0;
3415 }
3416 module_init(module_debugfs_init);
3417 #endif
3418