linux - kernel/module/main.c

4005 lines · 101.9 KiB

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 <[email protected]>
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/module_symbol.h>
15	#include <linux/trace_events.h>
16	#include <linux/init.h>
17	#include <linux/kallsyms.h>
18	#include <linux/buildid.h>
19	#include <linux/fs.h>
20	#include <linux/kernel.h>
21	#include <linux/kernel_read_file.h>
22	#include <linux/kstrtox.h>
23	#include <linux/slab.h>
24	#include <linux/vmalloc.h>
25	#include <linux/elf.h>
26	#include <linux/seq_file.h>
27	#include <linux/syscalls.h>
28	#include <linux/fcntl.h>
29	#include <linux/rcupdate.h>
30	#include <linux/capability.h>
31	#include <linux/cpu.h>
32	#include <linux/moduleparam.h>
33	#include <linux/errno.h>
34	#include <linux/err.h>
35	#include <linux/vermagic.h>
36	#include <linux/notifier.h>
37	#include <linux/sched.h>
38	#include <linux/device.h>
39	#include <linux/string.h>
40	#include <linux/mutex.h>
41	#include <linux/rculist.h>
42	#include <linux/uaccess.h>
43	#include <asm/cacheflush.h>
44	#include <linux/set_memory.h>
45	#include <asm/mmu_context.h>
46	#include <linux/license.h>
47	#include <asm/sections.h>
48	#include <linux/tracepoint.h>
49	#include <linux/ftrace.h>
50	#include <linux/livepatch.h>
51	#include <linux/async.h>
52	#include <linux/percpu.h>
53	#include <linux/kmemleak.h>
54	#include <linux/jump_label.h>
55	#include <linux/pfn.h>
56	#include <linux/bsearch.h>
57	#include <linux/dynamic_debug.h>
58	#include <linux/audit.h>
59	#include <linux/cfi.h>
60	#include <linux/codetag.h>
61	#include <linux/debugfs.h>
62	#include <linux/execmem.h>
63	#include <uapi/linux/module.h>
64	#include "internal.h"
65
66	#define CREATE_TRACE_POINTS
67	#include <trace/events/module.h>
68
69	/*
70	* Mutex protects:
71	* 1) List of modules (also safely readable within RCU read section),
72	* 2) module_use links,
73	* 3) mod_tree.addr_min/mod_tree.addr_max.
74	* (delete and add uses RCU list operations).
75	*/
76	DEFINE_MUTEX(module_mutex);
77	LIST_HEAD(modules);
78
79	/* Work queue for freeing init sections in success case */
80	static void do_free_init(struct work_struct *w);
81	static DECLARE_WORK(init_free_wq, do_free_init);
82	static LLIST_HEAD(init_free_list);
83
84	struct mod_tree_root mod_tree __cacheline_aligned = {
85	.addr_min = -1UL,
86	};
87
88	struct symsearch {
89	const struct kernel_symbol start, stop;
90	const u32 *crcs;
91	const u8 *flagstab;
92	};
93
94	/*
95	* Bounds of module memory, for speeding up __module_address.
96	* Protected by module_mutex.
97	*/
98	static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
99	unsigned int size, struct mod_tree_root *tree)
100	{
101	unsigned long min = (unsigned long)base;
102	unsigned long max = min + size;
103
104	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
105	if (mod_mem_type_is_core_data(type)) {
106	if (min < tree->data_addr_min)
107	tree->data_addr_min = min;
108	if (max > tree->data_addr_max)
109	tree->data_addr_max = max;
110	return;
111	}
112	#endif
113	if (min < tree->addr_min)
114	tree->addr_min = min;
115	if (max > tree->addr_max)
116	tree->addr_max = max;
117	}
118
119	static void mod_update_bounds(struct module *mod)
120	{
121	for_each_mod_mem_type(type) {
122	struct module_memory *mod_mem = &mod->mem[type];
123
124	if (mod_mem->size)
125	__mod_update_bounds(type, mod_mem->base, mod_mem->size, &mod_tree);
126	}
127	}
128
129	/* Block module loading/unloading? */
130	static int modules_disabled;
131	core_param(nomodule, modules_disabled, bint, 0);
132
133	static const struct ctl_table module_sysctl_table[] = {
134	{
135	.procname = "modprobe",
136	.data = &modprobe_path,
137	.maxlen = KMOD_PATH_LEN,
138	.mode = 0644,
139	.proc_handler = proc_dostring,
140	},
141	{
142	.procname = "modules_disabled",
143	.data = &modules_disabled,
144	.maxlen = sizeof(int),
145	.mode = 0644,
146	/* only handle a transition from default "0" to "1" */
147	.proc_handler = proc_dointvec_minmax,
148	.extra1 = SYSCTL_ONE,
149	.extra2 = SYSCTL_ONE,
150	},
151	};
152
153	static int __init init_module_sysctl(void)
154	{
155	register_sysctl_init("kernel", module_sysctl_table);
156	return 0;
157	}
158
159	subsys_initcall(init_module_sysctl);
160
161	/* Waiting for a module to finish initializing? */
162	static DECLARE_WAIT_QUEUE_HEAD(module_wq);
163
164	static BLOCKING_NOTIFIER_HEAD(module_notify_list);
165
166	int register_module_notifier(struct notifier_block *nb)
167	{
168	return blocking_notifier_chain_register(&module_notify_list, nb);
169	}
170	EXPORT_SYMBOL(register_module_notifier);
171
172	int unregister_module_notifier(struct notifier_block *nb)
173	{
174	return blocking_notifier_chain_unregister(&module_notify_list, nb);
175	}
176	EXPORT_SYMBOL(unregister_module_notifier);
177
178	/*
179	* We require a truly strong try_module_get(): 0 means success.
180	* Otherwise an error is returned due to ongoing or failed
181	* initialization etc.
182	*/
183	static inline int strong_try_module_get(struct module *mod)
184	{
185	BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
186	if (mod && mod->state == MODULE_STATE_COMING)
187	return -EBUSY;
188	if (try_module_get(mod))
189	return 0;
190	else
191	return -ENOENT;
192	}
193
194	static inline void add_taint_module(struct module *mod, unsigned flag,
195	enum lockdep_ok lockdep_ok)
196	{
197	add_taint(flag, lockdep_ok);
198	set_bit(flag, &mod->taints);
199	}
200
201	/*
202	* Like strncmp(), except s/-/_/g as per scripts/Makefile.lib:name-fix-token rule.
203	*/
204	static int mod_strncmp(const char str_a, const char str_b, size_t n)
205	{
206	for (int i = 0; i < n; i++) {
207	char a = str_a[i];
208	char b = str_b[i];
209	int d;
210
211	if (a == '-') a = '_';
212	if (b == '-') b = '_';
213
214	d = a - b;
215	if (d)
216	return d;
217
218	if (!a)
219	break;
220	}
221
222	return 0;
223	}
224
225	/*
226	* A thread that wants to hold a reference to a module only while it
227	* is running can call this to safely exit.
228	*/
229	void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
230	{
231	module_put(mod);
232	kthread_exit(code);
233	}
234	EXPORT_SYMBOL(__module_put_and_kthread_exit);
235
236	/* Find a module section: 0 means not found. */
237	static unsigned int find_sec(const struct load_info info, const char name)
238	{
239	unsigned int i;
240
241	for (i = 1; i < info->hdr->e_shnum; i++) {
242	Elf_Shdr *shdr = &info->sechdrs[i];
243	/* Alloc bit cleared means "ignore it." */
244	if ((shdr->sh_flags & SHF_ALLOC)
245	&& strcmp(info->secstrings + shdr->sh_name, name) == 0)
246	return i;
247	}
248	return 0;
249	}
250
251	/**
252	* find_any_unique_sec() - Find a unique section index by name
253	* @info: Load info for the module to scan
254	* @name: Name of the section we're looking for
255	*
256	* Locates a unique section by name. Ignores SHF_ALLOC.
257	*
258	* Return: Section index if found uniquely, zero if absent, negative count
259	* of total instances if multiple were found.
260	*/
261	static int find_any_unique_sec(const struct load_info info, const char name)
262	{
263	unsigned int idx;
264	unsigned int count = 0;
265	int i;
266
267	for (i = 1; i < info->hdr->e_shnum; i++) {
268	if (strcmp(info->secstrings + info->sechdrs[i].sh_name,
269	name) == 0) {
270	count++;
271	idx = i;
272	}
273	}
274	if (count == 1) {
275	return idx;
276	} else if (count == 0) {
277	return 0;
278	} else {
279	return -count;
280	}
281	}
282
283	/* Find a module section, or NULL. */
284	static void section_addr(const struct load_info info, const char *name)
285	{
286	/* Section 0 has sh_addr 0. */
287	return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
288	}
289
290	/* Find a module section, or NULL. Fill in number of "objects" in section. */
291	static void section_objs(const struct load_info info,
292	const char *name,
293	size_t object_size,
294	unsigned int *num)
295	{
296	unsigned int sec = find_sec(info, name);
297
298	/* Section 0 has sh_addr 0 and sh_size 0. */
299	*num = info->sechdrs[sec].sh_size / object_size;
300	return (void *)info->sechdrs[sec].sh_addr;
301	}
302
303	/* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
304	static unsigned int find_any_sec(const struct load_info info, const char name)
305	{
306	unsigned int i;
307
308	for (i = 1; i < info->hdr->e_shnum; i++) {
309	Elf_Shdr *shdr = &info->sechdrs[i];
310	if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
311	return i;
312	}
313	return 0;
314	}
315
316	/*
317	* Find a module section, or NULL. Fill in number of "objects" in section.
318	* Ignores SHF_ALLOC flag.
319	*/
320	static __maybe_unused void any_section_objs(const struct load_info info,
321	const char *name,
322	size_t object_size,
323	unsigned int *num)
324	{
325	unsigned int sec = find_any_sec(info, name);
326
327	/* Section 0 has sh_addr 0 and sh_size 0. */
328	*num = info->sechdrs[sec].sh_size / object_size;
329	return (void *)info->sechdrs[sec].sh_addr;
330	}
331
332	#ifndef CONFIG_MODVERSIONS
333	#define symversion(base, idx) NULL
334	#else
335	#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
336	#endif
337
338	static const char kernel_symbol_name(const struct kernel_symbol sym)
339	{
340	#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
341	return offset_to_ptr(&sym->name_offset);
342	#else
343	return sym->name;
344	#endif
345	}
346
347	static const char kernel_symbol_namespace(const struct kernel_symbol sym)
348	{
349	#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
350	if (!sym->namespace_offset)
351	return NULL;
352	return offset_to_ptr(&sym->namespace_offset);
353	#else
354	return sym->namespace;
355	#endif
356	}
357
358	int cmp_name(const void name, const void sym)
359	{
360	return strcmp(name, kernel_symbol_name(sym));
361	}
362
363	static bool find_exported_symbol_in_section(const struct symsearch *syms,
364	struct module *owner,
365	struct find_symbol_arg *fsa)
366	{
367	struct kernel_symbol *sym;
368	u8 sym_flags;
369
370	sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
371	sizeof(struct kernel_symbol), cmp_name);
372	if (!sym)
373	return false;
374
375	sym_flags = *(syms->flagstab + (sym - syms->start));
376	if (!fsa->gplok && (sym_flags & KSYM_FLAG_GPL_ONLY))
377	return false;
378
379	fsa->owner = owner;
380	fsa->crc = symversion(syms->crcs, sym - syms->start);
381	fsa->sym = sym;
382	fsa->license = (sym_flags & KSYM_FLAG_GPL_ONLY) ? GPL_ONLY : NOT_GPL_ONLY;
383
384	return true;
385	}
386
387	/*
388	* Find an exported symbol and return it, along with, (optional) crc and
389	* (optional) module which owns it. Needs RCU or module_mutex.
390	*/
391	bool find_symbol(struct find_symbol_arg *fsa)
392	{
393	const struct symsearch syms = {
394	.start = __start___ksymtab,
395	.stop = __stop___ksymtab,
396	.crcs = __start___kcrctab,
397	.flagstab = __start___kflagstab,
398	};
399	struct module *mod;
400
401	if (find_exported_symbol_in_section(&syms, NULL, fsa))
402	return true;
403
404	list_for_each_entry_rcu(mod, &modules, list,
405	lockdep_is_held(&module_mutex)) {
406	const struct symsearch syms = {
407	.start = mod->syms,
408	.stop = mod->syms + mod->num_syms,
409	.crcs = mod->crcs,
410	.flagstab = mod->flagstab,
411	};
412
413	if (mod->state == MODULE_STATE_UNFORMED)
414	continue;
415
416	if (find_exported_symbol_in_section(&syms, mod, fsa))
417	return true;
418	}
419
420	pr_debug("Failed to find symbol %s\n", fsa->name);
421	return false;
422	}
423
424	/*
425	* Search for module by name: must hold module_mutex (or RCU for read-only
426	* access).
427	*/
428	struct module find_module_all(const char name, size_t len,
429	bool even_unformed)
430	{
431	struct module *mod;
432
433	list_for_each_entry_rcu(mod, &modules, list,
434	lockdep_is_held(&module_mutex)) {
435	if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
436	continue;
437	if (strlen(mod->name) == len && !memcmp(mod->name, name, len))
438	return mod;
439	}
440	return NULL;
441	}
442
443	struct module find_module(const char name)
444	{
445	return find_module_all(name, strlen(name), false);
446	}
447
448	#ifdef CONFIG_SMP
449
450	static inline void __percpu mod_percpu(struct module mod)
451	{
452	return mod->percpu;
453	}
454
455	static int percpu_modalloc(struct module mod, struct load_info info)
456	{
457	Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
458	unsigned long align = pcpusec->sh_addralign;
459
460	if (!pcpusec->sh_size)
461	return 0;
462
463	if (align > PAGE_SIZE) {
464	pr_warn("%s: per-cpu alignment %li > %li\n",
465	mod->name, align, PAGE_SIZE);
466	align = PAGE_SIZE;
467	}
468
469	mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
470	if (!mod->percpu) {
471	pr_warn("%s: Could not allocate %lu bytes percpu data\n",
472	mod->name, (unsigned long)pcpusec->sh_size);
473	return -ENOMEM;
474	}
475	mod->percpu_size = pcpusec->sh_size;
476	return 0;
477	}
478
479	static void percpu_modfree(struct module *mod)
480	{
481	free_percpu(mod->percpu);
482	}
483
484	static unsigned int find_pcpusec(struct load_info *info)
485	{
486	return find_sec(info, ".data..percpu");
487	}
488
489	static void percpu_modcopy(struct module *mod,
490	const void *from, unsigned long size)
491	{
492	int cpu;
493
494	for_each_possible_cpu(cpu)
495	memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
496	}
497
498	bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
499	{
500	struct module *mod;
501	unsigned int cpu;
502
503	guard(rcu)();
504	list_for_each_entry_rcu(mod, &modules, list) {
505	if (mod->state == MODULE_STATE_UNFORMED)
506	continue;
507	if (!mod->percpu_size)
508	continue;
509	for_each_possible_cpu(cpu) {
510	void *start = per_cpu_ptr(mod->percpu, cpu);
511	void va = (void )addr;
512
513	if (va >= start && va < start + mod->percpu_size) {
514	if (can_addr) {
515	*can_addr = (unsigned long) (va - start);
516	*can_addr += (unsigned long)
517	per_cpu_ptr(mod->percpu,
518	get_boot_cpu_id());
519	}
520	return true;
521	}
522	}
523	}
524	return false;
525	}
526
527	/**
528	* is_module_percpu_address() - test whether address is from module static percpu
529	* @addr: address to test
530	*
531	* Test whether @addr belongs to module static percpu area.
532	*
533	* Return: %true if @addr is from module static percpu area
534	*/
535	bool is_module_percpu_address(unsigned long addr)
536	{
537	return __is_module_percpu_address(addr, NULL);
538	}
539
540	#else /* ... !CONFIG_SMP */
541
542	static inline void __percpu mod_percpu(struct module mod)
543	{
544	return NULL;
545	}
546	static int percpu_modalloc(struct module mod, struct load_info info)
547	{
548	/* UP modules shouldn't have this section: ENOMEM isn't quite right */
549	if (info->sechdrs[info->index.pcpu].sh_size != 0)
550	return -ENOMEM;
551	return 0;
552	}
553	static inline void percpu_modfree(struct module *mod)
554	{
555	}
556	static unsigned int find_pcpusec(struct load_info *info)
557	{
558	return 0;
559	}
560	static inline void percpu_modcopy(struct module *mod,
561	const void *from, unsigned long size)
562	{
563	/* pcpusec should be 0, and size of that section should be 0. */
564	BUG_ON(size != 0);
565	}
566	bool is_module_percpu_address(unsigned long addr)
567	{
568	return false;
569	}
570
571	bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
572	{
573	return false;
574	}
575
576	#endif /* CONFIG_SMP */
577
578	#define MODINFO_ATTR(field) \
579	static void setup_modinfo_##field(struct module mod, const char s) \
580	{ \
581	mod->field = kstrdup(s, GFP_KERNEL); \
582	} \
583	static ssize_t show_modinfo_##field(const struct module_attribute *mattr, \
584	struct module_kobject mk, char buffer) \
585	{ \
586	return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field); \
587	} \
588	static int modinfo_##field##_exists(struct module *mod) \
589	{ \
590	return mod->field != NULL; \
591	} \
592	static void free_modinfo_##field(struct module *mod) \
593	{ \
594	kfree(mod->field); \
595	mod->field = NULL; \
596	} \
597	static const struct module_attribute modinfo_##field = { \
598	.attr = { .name = __stringify(field), .mode = 0444 }, \
599	.show = show_modinfo_##field, \
600	.setup = setup_modinfo_##field, \
601	.test = modinfo_##field##_exists, \
602	.free = free_modinfo_##field, \
603	};
604
605	MODINFO_ATTR(version);
606	MODINFO_ATTR(srcversion);
607
608	static void setup_modinfo_import_ns(struct module mod, const char s)
609	{
610	mod->imported_namespaces = NULL;
611	}
612
613	static ssize_t show_modinfo_import_ns(const struct module_attribute *mattr,
614	struct module_kobject mk, char buffer)
615	{
616	return sysfs_emit(buffer, "%s\n", mk->mod->imported_namespaces);
617	}
618
619	static int modinfo_import_ns_exists(struct module *mod)
620	{
621	return mod->imported_namespaces != NULL;
622	}
623
624	static void free_modinfo_import_ns(struct module *mod)
625	{
626	kfree(mod->imported_namespaces);
627	mod->imported_namespaces = NULL;
628	}
629
630	static const struct module_attribute modinfo_import_ns = {
631	.attr = { .name = "import_ns", .mode = 0444 },
632	.show = show_modinfo_import_ns,
633	.setup = setup_modinfo_import_ns,
634	.test = modinfo_import_ns_exists,
635	.free = free_modinfo_import_ns,
636	};
637
638	static struct {
639	char name[MODULE_NAME_LEN];
640	char taints[MODULE_FLAGS_BUF_SIZE];
641	} last_unloaded_module;
642
643	#ifdef CONFIG_MODULE_UNLOAD
644
645	EXPORT_TRACEPOINT_SYMBOL(module_get);
646
647	/* MODULE_REF_BASE is the base reference count by kmodule loader. */
648	#define MODULE_REF_BASE 1
649
650	/* Init the unload section of the module. */
651	static int module_unload_init(struct module *mod)
652	{
653	/*
654	* Initialize reference counter to MODULE_REF_BASE.
655	* refcnt == 0 means module is going.
656	*/
657	atomic_set(&mod->refcnt, MODULE_REF_BASE);
658
659	INIT_LIST_HEAD(&mod->source_list);
660	INIT_LIST_HEAD(&mod->target_list);
661
662	/* Hold reference count during initialization. */
663	atomic_inc(&mod->refcnt);
664
665	return 0;
666	}
667
668	/* Does a already use b? */
669	static int already_uses(struct module a, struct module b)
670	{
671	struct module_use *use;
672
673	list_for_each_entry(use, &b->source_list, source_list) {
674	if (use->source == a)
675	return 1;
676	}
677	pr_debug("%s does not use %s!\n", a->name, b->name);
678	return 0;
679	}
680
681	/*
682	* Module a uses b
683	* - we add 'a' as a "source", 'b' as a "target" of module use
684	* - the module_use is added to the list of 'b' sources (so
685	* 'b' can walk the list to see who sourced them), and of 'a'
686	* targets (so 'a' can see what modules it targets).
687	*/
688	static int add_module_usage(struct module a, struct module b)
689	{
690	struct module_use *use;
691
692	pr_debug("Allocating new usage for %s.\n", a->name);
693	use = kmalloc_obj(*use, GFP_ATOMIC);
694	if (!use)
695	return -ENOMEM;
696
697	use->source = a;
698	use->target = b;
699	list_add(&use->source_list, &b->source_list);
700	list_add(&use->target_list, &a->target_list);
701	return 0;
702	}
703
704	/* Module a uses b: caller needs module_mutex() */
705	static int ref_module(struct module a, struct module b)
706	{
707	int err;
708
709	if (b == NULL \|\| already_uses(a, b))
710	return 0;
711
712	/* If module isn't available, we fail. */
713	err = strong_try_module_get(b);
714	if (err)
715	return err;
716
717	err = add_module_usage(a, b);
718	if (err) {
719	module_put(b);
720	return err;
721	}
722	return 0;
723	}
724
725	/* Clear the unload stuff of the module. */
726	static void module_unload_free(struct module *mod)
727	{
728	struct module_use use, tmp;
729
730	mutex_lock(&module_mutex);
731	list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
732	struct module *i = use->target;
733	pr_debug("%s unusing %s\n", mod->name, i->name);
734	module_put(i);
735	list_del(&use->source_list);
736	list_del(&use->target_list);
737	kfree(use);
738	}
739	mutex_unlock(&module_mutex);
740	}
741
742	#ifdef CONFIG_MODULE_FORCE_UNLOAD
743	static inline int try_force_unload(unsigned int flags)
744	{
745	int ret = (flags & O_TRUNC);
746	if (ret)
747	add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
748	return ret;
749	}
750	#else
751	static inline int try_force_unload(unsigned int flags)
752	{
753	return 0;
754	}
755	#endif /* CONFIG_MODULE_FORCE_UNLOAD */
756
757	/* Try to release refcount of module, 0 means success. */
758	static int try_release_module_ref(struct module *mod)
759	{
760	int ret;
761
762	/* Try to decrement refcnt which we set at loading */
763	ret = atomic_sub_return(MODULE_REF_BASE, &mod->refcnt);
764	BUG_ON(ret < 0);
765	if (ret)
766	/* Someone can put this right now, recover with checking */
767	ret = atomic_add_unless(&mod->refcnt, MODULE_REF_BASE, 0);
768
769	return ret;
770	}
771
772	static int try_stop_module(struct module mod, int flags, int forced)
773	{
774	/* If it's not unused, quit unless we're forcing. */
775	if (try_release_module_ref(mod) != 0) {
776	*forced = try_force_unload(flags);
777	if (!(*forced))
778	return -EWOULDBLOCK;
779	}
780
781	/* Mark it as dying. */
782	mod->state = MODULE_STATE_GOING;
783
784	return 0;
785	}
786
787	/**
788	* module_refcount() - return the refcount or -1 if unloading
789	* @mod: the module we're checking
790	*
791	* Return:
792	* -1 if the module is in the process of unloading
793	* otherwise the number of references in the kernel to the module
794	*/
795	int module_refcount(struct module *mod)
796	{
797	return atomic_read(&mod->refcnt) - MODULE_REF_BASE;
798	}
799	EXPORT_SYMBOL(module_refcount);
800
801	/* This exists whether we can unload or not */
802	static void free_module(struct module *mod);
803
804	SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
805	unsigned int, flags)
806	{
807	struct module *mod;
808	char name[MODULE_NAME_LEN];
809	char buf[MODULE_FLAGS_BUF_SIZE];
810	int ret, len, forced = 0;
811
812	if (!capable(CAP_SYS_MODULE) \|\| modules_disabled)
813	return -EPERM;
814
815	len = strncpy_from_user(name, name_user, MODULE_NAME_LEN);
816	if (len == 0 \|\| len == MODULE_NAME_LEN)
817	return -ENOENT;
818	if (len < 0)
819	return len;
820
821	audit_log_kern_module(name);
822
823	if (mutex_lock_interruptible(&module_mutex) != 0)
824	return -EINTR;
825
826	mod = find_module(name);
827	if (!mod) {
828	ret = -ENOENT;
829	goto out;
830	}
831
832	if (!list_empty(&mod->source_list)) {
833	/* Other modules depend on us: get rid of them first. */
834	ret = -EWOULDBLOCK;
835	goto out;
836	}
837
838	/* Doing init or already dying? */
839	if (mod->state != MODULE_STATE_LIVE) {
840	/* FIXME: if (force), slam module count damn the torpedoes */
841	pr_debug("%s already dying\n", mod->name);
842	ret = -EBUSY;
843	goto out;
844	}
845
846	/* If it has an init func, it must have an exit func to unload */
847	if (mod->init && !mod->exit) {
848	forced = try_force_unload(flags);
849	if (!forced) {
850	/* This module can't be removed */
851	ret = -EBUSY;
852	goto out;
853	}
854	}
855
856	ret = try_stop_module(mod, flags, &forced);
857	if (ret != 0)
858	goto out;
859
860	mutex_unlock(&module_mutex);
861	/* Final destruction now no one is using it. */
862	if (mod->exit != NULL)
863	mod->exit();
864	blocking_notifier_call_chain(&module_notify_list,
865	MODULE_STATE_GOING, mod);
866	klp_module_going(mod);
867	ftrace_release_mod(mod);
868
869	async_synchronize_full();
870
871	/* Store the name and taints of the last unloaded module for diagnostic purposes */
872	strscpy(last_unloaded_module.name, mod->name);
873	strscpy(last_unloaded_module.taints, module_flags(mod, buf, false));
874
875	free_module(mod);
876	/* someone could wait for the module in add_unformed_module() */
877	wake_up_all(&module_wq);
878	return 0;
879	out:
880	mutex_unlock(&module_mutex);
881	return ret;
882	}
883
884	void __symbol_put(const char *symbol)
885	{
886	struct find_symbol_arg fsa = {
887	.name = symbol,
888	.gplok = true,
889	};
890
891	guard(rcu)();
892	BUG_ON(!find_symbol(&fsa));
893	module_put(fsa.owner);
894	}
895	EXPORT_SYMBOL(__symbol_put);
896
897	/* Note this assumes addr is a function, which it currently always is. */
898	void symbol_put_addr(void *addr)
899	{
900	struct module *modaddr;
901	unsigned long a = (unsigned long)dereference_function_descriptor(addr);
902
903	if (core_kernel_text(a))
904	return;
905
906	/*
907	* Even though we hold a reference on the module; we still need to
908	* RCU read section in order to safely traverse the data structure.
909	*/
910	guard(rcu)();
911	modaddr = __module_text_address(a);
912	BUG_ON(!modaddr);
913	module_put(modaddr);
914	}
915	EXPORT_SYMBOL_GPL(symbol_put_addr);
916
917	static ssize_t show_refcnt(const struct module_attribute *mattr,
918	struct module_kobject mk, char buffer)
919	{
920	return sprintf(buffer, "%i\n", module_refcount(mk->mod));
921	}
922
923	static const struct module_attribute modinfo_refcnt =
924	__ATTR(refcnt, 0444, show_refcnt, NULL);
925
926	void __module_get(struct module *module)
927	{
928	if (module) {
929	atomic_inc(&module->refcnt);
930	trace_module_get(module, _RET_IP_);
931	}
932	}
933	EXPORT_SYMBOL(__module_get);
934
935	bool try_module_get(struct module *module)
936	{
937	bool ret = true;
938
939	if (module) {
940	/* Note: here, we can fail to get a reference */
941	if (likely(module_is_live(module) &&
942	atomic_inc_not_zero(&module->refcnt) != 0))
943	trace_module_get(module, _RET_IP_);
944	else
945	ret = false;
946	}
947	return ret;
948	}
949	EXPORT_SYMBOL(try_module_get);
950
951	void module_put(struct module *module)
952	{
953	int ret;
954
955	if (module) {
956	ret = atomic_dec_if_positive(&module->refcnt);
957	WARN_ON(ret < 0); /* Failed to put refcount */
958	trace_module_put(module, _RET_IP_);
959	}
960	}
961	EXPORT_SYMBOL(module_put);
962
963	#else /* !CONFIG_MODULE_UNLOAD */
964	static inline void module_unload_free(struct module *mod)
965	{
966	}
967
968	static int ref_module(struct module a, struct module b)
969	{
970	return strong_try_module_get(b);
971	}
972
973	static inline int module_unload_init(struct module *mod)
974	{
975	return 0;
976	}
977	#endif /* CONFIG_MODULE_UNLOAD */
978
979	size_t module_flags_taint(unsigned long taints, char *buf)
980	{
981	size_t l = 0;
982	int i;
983
984	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
985	if (test_bit(i, &taints))
986	buf[l++] = taint_flags[i].c_true;
987	}
988
989	return l;
990	}
991
992	static ssize_t show_initstate(const struct module_attribute *mattr,
993	struct module_kobject mk, char buffer)
994	{
995	const char *state = "unknown";
996
997	switch (mk->mod->state) {
998	case MODULE_STATE_LIVE:
999	state = "live";
1000	break;
1001	case MODULE_STATE_COMING:
1002	state = "coming";
1003	break;
1004	case MODULE_STATE_GOING:
1005	state = "going";
1006	break;
1007	default:
1008	BUG();
1009	}
1010	return sprintf(buffer, "%s\n", state);
1011	}
1012
1013	static const struct module_attribute modinfo_initstate =
1014	__ATTR(initstate, 0444, show_initstate, NULL);
1015
1016	static ssize_t store_uevent(const struct module_attribute *mattr,
1017	struct module_kobject *mk,
1018	const char *buffer, size_t count)
1019	{
1020	int rc;
1021
1022	rc = kobject_synth_uevent(&mk->kobj, buffer, count);
1023	return rc ? rc : count;
1024	}
1025
1026	const struct module_attribute module_uevent =
1027	__ATTR(uevent, 0200, NULL, store_uevent);
1028
1029	static ssize_t show_coresize(const struct module_attribute *mattr,
1030	struct module_kobject mk, char buffer)
1031	{
1032	unsigned int size = mk->mod->mem[MOD_TEXT].size;
1033
1034	if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
1035	for_class_mod_mem_type(type, core_data)
1036	size += mk->mod->mem[type].size;
1037	}
1038	return sprintf(buffer, "%u\n", size);
1039	}
1040
1041	static const struct module_attribute modinfo_coresize =
1042	__ATTR(coresize, 0444, show_coresize, NULL);
1043
1044	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
1045	static ssize_t show_datasize(const struct module_attribute *mattr,
1046	struct module_kobject mk, char buffer)
1047	{
1048	unsigned int size = 0;
1049
1050	for_class_mod_mem_type(type, core_data)
1051	size += mk->mod->mem[type].size;
1052	return sprintf(buffer, "%u\n", size);
1053	}
1054
1055	static const struct module_attribute modinfo_datasize =
1056	__ATTR(datasize, 0444, show_datasize, NULL);
1057	#endif
1058
1059	static ssize_t show_initsize(const struct module_attribute *mattr,
1060	struct module_kobject mk, char buffer)
1061	{
1062	unsigned int size = 0;
1063
1064	for_class_mod_mem_type(type, init)
1065	size += mk->mod->mem[type].size;
1066	return sprintf(buffer, "%u\n", size);
1067	}
1068
1069	static const struct module_attribute modinfo_initsize =
1070	__ATTR(initsize, 0444, show_initsize, NULL);
1071
1072	static ssize_t show_taint(const struct module_attribute *mattr,
1073	struct module_kobject mk, char buffer)
1074	{
1075	size_t l;
1076
1077	l = module_flags_taint(mk->mod->taints, buffer);
1078	buffer[l++] = '\n';
1079	return l;
1080	}
1081
1082	static const struct module_attribute modinfo_taint =
1083	__ATTR(taint, 0444, show_taint, NULL);
1084
1085	const struct module_attribute *const modinfo_attrs[] = {
1086	&module_uevent,
1087	&modinfo_version,
1088	&modinfo_srcversion,
1089	&modinfo_import_ns,
1090	&modinfo_initstate,
1091	&modinfo_coresize,
1092	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
1093	&modinfo_datasize,
1094	#endif
1095	&modinfo_initsize,
1096	&modinfo_taint,
1097	#ifdef CONFIG_MODULE_UNLOAD
1098	&modinfo_refcnt,
1099	#endif
1100	NULL,
1101	};
1102
1103	const size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
1104
1105	static const char vermagic[] = VERMAGIC_STRING;
1106
1107	int try_to_force_load(struct module mod, const char reason)
1108	{
1109	#ifdef CONFIG_MODULE_FORCE_LOAD
1110	if (!test_taint(TAINT_FORCED_MODULE))
1111	pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1112	add_taint_module(mod, TAINT_FORCED_MODULE, LOCKDEP_NOW_UNRELIABLE);
1113	return 0;
1114	#else
1115	return -ENOEXEC;
1116	#endif
1117	}
1118
1119	/* Parse tag=value strings from .modinfo section */
1120	char module_next_tag_pair(char string, unsigned long *secsize)
1121	{
1122	/* Skip non-zero chars */
1123	while (string[0]) {
1124	string++;
1125	if ((*secsize)-- <= 1)
1126	return NULL;
1127	}
1128
1129	/* Skip any zero padding. */
1130	while (!string[0]) {
1131	string++;
1132	if ((*secsize)-- <= 1)
1133	return NULL;
1134	}
1135	return string;
1136	}
1137
1138	static char get_next_modinfo(const struct load_info info, const char *tag,
1139	char *prev)
1140	{
1141	char *p;
1142	unsigned int taglen = strlen(tag);
1143	Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1144	unsigned long size = infosec->sh_size;
1145
1146	/*
1147	* get_modinfo() calls made before rewrite_section_headers()
1148	* must use sh_offset, as sh_addr isn't set!
1149	*/
1150	char modinfo = (char )info->hdr + infosec->sh_offset;
1151
1152	if (prev) {
1153	size -= prev - modinfo;
1154	modinfo = module_next_tag_pair(prev, &size);
1155	}
1156
1157	for (p = modinfo; p; p = module_next_tag_pair(p, &size)) {
1158	if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1159	return p + taglen + 1;
1160	}
1161	return NULL;
1162	}
1163
1164	static char get_modinfo(const struct load_info info, const char *tag)
1165	{
1166	return get_next_modinfo(info, tag, NULL);
1167	}
1168
1169	/**
1170	* verify_module_namespace() - does @modname have access to this symbol's @namespace
1171	* @namespace: export symbol namespace
1172	* @modname: module name
1173	*
1174	* If @namespace is prefixed with "module:" to indicate it is a module namespace
1175	* then test if @modname matches any of the comma separated patterns.
1176	*
1177	* The patterns only support tail-glob.
1178	*/
1179	static bool verify_module_namespace(const char namespace, const char modname)
1180	{
1181	size_t len, modlen = strlen(modname);
1182	const char *prefix = "module:";
1183	const char *sep;
1184	bool glob;
1185
1186	if (!strstarts(namespace, prefix))
1187	return false;
1188
1189	for (namespace += strlen(prefix); *namespace; namespace = sep) {
1190	sep = strchrnul(namespace, ',');
1191	len = sep - namespace;
1192
1193	glob = false;
1194	if (sep[-1] == '*') {
1195	len--;
1196	glob = true;
1197	}
1198
1199	if (*sep)
1200	sep++;
1201
1202	if (mod_strncmp(namespace, modname, len) == 0 && (glob \|\| len == modlen))
1203	return true;
1204	}
1205
1206	return false;
1207	}
1208
1209	static int verify_namespace_is_imported(const struct load_info *info,
1210	const struct kernel_symbol *sym,
1211	struct module *mod)
1212	{
1213	const char *namespace;
1214	char *imported_namespace;
1215
1216	namespace = kernel_symbol_namespace(sym);
1217	if (namespace && namespace[0]) {
1218
1219	if (verify_module_namespace(namespace, mod->name))
1220	return 0;
1221
1222	for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1223	if (strcmp(namespace, imported_namespace) == 0)
1224	return 0;
1225	}
1226	#ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1227	pr_warn(
1228	#else
1229	pr_err(
1230	#endif
1231	"%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1232	mod->name, kernel_symbol_name(sym), namespace);
1233	#ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1234	return -EINVAL;
1235	#endif
1236	}
1237	return 0;
1238	}
1239
1240	static bool inherit_taint(struct module mod, struct module owner, const char *name)
1241	{
1242	if (!owner \|\| !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1243	return true;
1244
1245	if (mod->using_gplonly_symbols) {
1246	pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1247	mod->name, name, owner->name);
1248	return false;
1249	}
1250
1251	if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1252	pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1253	mod->name, name, owner->name);
1254	set_bit(TAINT_PROPRIETARY_MODULE, &mod->taints);
1255	}
1256	return true;
1257	}
1258
1259	/* Resolve a symbol for this module. I.e. if we find one, record usage. */
1260	static const struct kernel_symbol resolve_symbol(struct module mod,
1261	const struct load_info *info,
1262	const char *name,
1263	char ownername[])
1264	{
1265	struct find_symbol_arg fsa = {
1266	.name = name,
1267	.gplok = !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1268	.warn = true,
1269	};
1270	int err;
1271
1272	/*
1273	* The module_mutex should not be a heavily contended lock;
1274	* if we get the occasional sleep here, we'll go an extra iteration
1275	* in the wait_event_interruptible(), which is harmless.
1276	*/
1277	sched_annotate_sleep();
1278	mutex_lock(&module_mutex);
1279	if (!find_symbol(&fsa))
1280	goto unlock;
1281
1282	if (fsa.license == GPL_ONLY)
1283	mod->using_gplonly_symbols = true;
1284
1285	if (!inherit_taint(mod, fsa.owner, name)) {
1286	fsa.sym = NULL;
1287	goto getname;
1288	}
1289
1290	if (!check_version(info, name, mod, fsa.crc)) {
1291	fsa.sym = ERR_PTR(-EINVAL);
1292	goto getname;
1293	}
1294
1295	err = verify_namespace_is_imported(info, fsa.sym, mod);
1296	if (err) {
1297	fsa.sym = ERR_PTR(err);
1298	goto getname;
1299	}
1300
1301	err = ref_module(mod, fsa.owner);
1302	if (err) {
1303	fsa.sym = ERR_PTR(err);
1304	goto getname;
1305	}
1306
1307	getname:
1308	/* We must make copy under the lock if we failed to get ref. */
1309	strscpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1310	unlock:
1311	mutex_unlock(&module_mutex);
1312	return fsa.sym;
1313	}
1314
1315	static const struct kernel_symbol *
1316	resolve_symbol_wait(struct module *mod,
1317	const struct load_info *info,
1318	const char *name)
1319	{
1320	const struct kernel_symbol *ksym;
1321	char owner[MODULE_NAME_LEN];
1322
1323	if (wait_event_interruptible_timeout(module_wq,
1324	!IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1325	\|\| PTR_ERR(ksym) != -EBUSY,
1326	30 * HZ) <= 0) {
1327	pr_warn("%s: gave up waiting for init of module %s.\n",
1328	mod->name, owner);
1329	}
1330	return ksym;
1331	}
1332
1333	void __weak module_arch_cleanup(struct module *mod)
1334	{
1335	}
1336
1337	void __weak module_arch_freeing_init(struct module *mod)
1338	{
1339	}
1340
1341	static int module_memory_alloc(struct module *mod, enum mod_mem_type type)
1342	{
1343	unsigned int size = PAGE_ALIGN(mod->mem[type].size);
1344	enum execmem_type execmem_type;
1345	void *ptr;
1346
1347	mod->mem[type].size = size;
1348
1349	if (mod_mem_type_is_data(type))
1350	execmem_type = EXECMEM_MODULE_DATA;
1351	else
1352	execmem_type = EXECMEM_MODULE_TEXT;
1353
1354	ptr = execmem_alloc_rw(execmem_type, size);
1355	if (!ptr)
1356	return -ENOMEM;
1357
1358	mod->mem[type].is_rox = execmem_is_rox(execmem_type);
1359
1360	/*
1361	* The pointer to these blocks of memory are stored on the module
1362	* structure and we keep that around so long as the module is
1363	* around. We only free that memory when we unload the module.
1364	* Just mark them as not being a leak then. The .init* ELF
1365	* sections do get freed after boot so we could treat them
1366	* slightly differently with kmemleak_ignore() and only grey
1367	* them out as they work as typical memory allocations which
1368	* do eventually get freed, but let's just keep things simple
1369	* and avoid any false positives.
1370	*/
1371	if (!mod->mem[type].is_rox)
1372	kmemleak_not_leak(ptr);
1373
1374	memset(ptr, 0, size);
1375	mod->mem[type].base = ptr;
1376
1377	return 0;
1378	}
1379
1380	static void module_memory_restore_rox(struct module *mod)
1381	{
1382	for_class_mod_mem_type(type, text) {
1383	struct module_memory *mem = &mod->mem[type];
1384
1385	if (mem->is_rox)
1386	execmem_restore_rox(mem->base, mem->size);
1387	}
1388	}
1389
1390	static void module_memory_free(struct module *mod, enum mod_mem_type type)
1391	{
1392	struct module_memory *mem = &mod->mem[type];
1393
1394	execmem_free(mem->base);
1395	}
1396
1397	static void free_mod_mem(struct module *mod)
1398	{
1399	for_each_mod_mem_type(type) {
1400	struct module_memory *mod_mem = &mod->mem[type];
1401
1402	if (type == MOD_DATA)
1403	continue;
1404
1405	/* Free lock-classes; relies on the preceding sync_rcu(). */
1406	lockdep_free_key_range(mod_mem->base, mod_mem->size);
1407	if (mod_mem->size)
1408	module_memory_free(mod, type);
1409	}
1410
1411	/* MOD_DATA hosts mod, so free it at last */
1412	lockdep_free_key_range(mod->mem[MOD_DATA].base, mod->mem[MOD_DATA].size);
1413	module_memory_free(mod, MOD_DATA);
1414	}
1415
1416	/* Free a module, remove from lists, etc. */
1417	static void free_module(struct module *mod)
1418	{
1419	trace_module_free(mod);
1420
1421	codetag_unload_module(mod);
1422
1423	mod_sysfs_teardown(mod);
1424
1425	/*
1426	* We leave it in list to prevent duplicate loads, but make sure
1427	* that noone uses it while it's being deconstructed.
1428	*/
1429	mutex_lock(&module_mutex);
1430	mod->state = MODULE_STATE_UNFORMED;
1431	mutex_unlock(&module_mutex);
1432
1433	/* Arch-specific cleanup. */
1434	module_arch_cleanup(mod);
1435
1436	/* Module unload stuff */
1437	module_unload_free(mod);
1438
1439	/* Free any allocated parameters. */
1440	module_destroy_params(mod->kp, mod->num_kp);
1441
1442	if (is_livepatch_module(mod))
1443	free_module_elf(mod);
1444
1445	/* Now we can delete it from the lists */
1446	mutex_lock(&module_mutex);
1447	/* Unlink carefully: kallsyms could be walking list. */
1448	list_del_rcu(&mod->list);
1449	mod_tree_remove(mod);
1450	/* Remove this module from bug list, this uses list_del_rcu */
1451	module_bug_cleanup(mod);
1452	/* Wait for RCU synchronizing before releasing mod->list and buglist. */
1453	synchronize_rcu();
1454	if (try_add_tainted_module(mod))
1455	pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1456	mod->name);
1457	mutex_unlock(&module_mutex);
1458
1459	/* This may be empty, but that's OK */
1460	module_arch_freeing_init(mod);
1461	percpu_modfree(mod);
1462
1463	free_mod_mem(mod);
1464	}
1465
1466	void __symbol_get(const char symbol)
1467	{
1468	struct find_symbol_arg fsa = {
1469	.name = symbol,
1470	.gplok = true,
1471	.warn = true,
1472	};
1473
1474	scoped_guard(rcu) {
1475	if (!find_symbol(&fsa))
1476	return NULL;
1477	if (fsa.license != GPL_ONLY) {
1478	pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1479	symbol);
1480	return NULL;
1481	}
1482	if (strong_try_module_get(fsa.owner))
1483	return NULL;
1484	}
1485	return (void *)kernel_symbol_value(fsa.sym);
1486	}
1487	EXPORT_SYMBOL_GPL(__symbol_get);
1488
1489	/*
1490	* Ensure that an exported symbol [global namespace] does not already exist
1491	* in the kernel or in some other module's exported symbol table.
1492	*
1493	* You must hold the module_mutex.
1494	*/
1495	static int verify_exported_symbols(struct module *mod)
1496	{
1497	const struct kernel_symbol *s;
1498	for (s = mod->syms; s < mod->syms + mod->num_syms; s++) {
1499	struct find_symbol_arg fsa = {
1500	.name = kernel_symbol_name(s),
1501	.gplok = true,
1502	};
1503	if (find_symbol(&fsa)) {
1504	pr_err("%s: exports duplicate symbol %s (owned by %s)\n",
1505	mod->name, kernel_symbol_name(s),
1506	module_name(fsa.owner));
1507	return -ENOEXEC;
1508	}
1509	}
1510	return 0;
1511	}
1512
1513	static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1514	{
1515	/*
1516	* On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1517	* before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1518	* i386 has a similar problem but may not deserve a fix.
1519	*
1520	* If we ever have to ignore many symbols, consider refactoring the code to
1521	* only warn if referenced by a relocation.
1522	*/
1523	if (emachine == EM_386 \|\| emachine == EM_X86_64)
1524	return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1525	return false;
1526	}
1527
1528	/* Change all symbols so that st_value encodes the pointer directly. */
1529	static int simplify_symbols(struct module mod, const struct load_info info)
1530	{
1531	Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1532	Elf_Sym sym = (void )symsec->sh_addr;
1533	unsigned long secbase;
1534	unsigned int i;
1535	int ret = 0;
1536	const struct kernel_symbol *ksym;
1537
1538	for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1539	const char *name = info->strtab + sym[i].st_name;
1540
1541	switch (sym[i].st_shndx) {
1542	case SHN_COMMON:
1543	/* Ignore common symbols */
1544	if (!strncmp(name, "__gnu_lto", 9))
1545	break;
1546
1547	/*
1548	* We compiled with -fno-common. These are not
1549	* supposed to happen.
1550	*/
1551	pr_debug("Common symbol: %s\n", name);
1552	pr_warn("%s: please compile with -fno-common\n",
1553	mod->name);
1554	ret = -ENOEXEC;
1555	break;
1556
1557	case SHN_ABS:
1558	/* Don't need to do anything */
1559	pr_debug("Absolute symbol: 0x%08lx %s\n",
1560	(long)sym[i].st_value, name);
1561	break;
1562
1563	case SHN_LIVEPATCH:
1564	/* Livepatch symbols are resolved by livepatch */
1565	break;
1566
1567	case SHN_UNDEF:
1568	ksym = resolve_symbol_wait(mod, info, name);
1569	/* Ok if resolved. */
1570	if (ksym && !IS_ERR(ksym)) {
1571	sym[i].st_value = kernel_symbol_value(ksym);
1572	break;
1573	}
1574
1575	/* Ok if weak or ignored. */
1576	if (!ksym &&
1577	(ELF_ST_BIND(sym[i].st_info) == STB_WEAK \|\|
1578	ignore_undef_symbol(info->hdr->e_machine, name)))
1579	break;
1580
1581	ret = PTR_ERR(ksym) ?: -ENOENT;
1582	pr_warn("%s: Unknown symbol %s (err %d)\n",
1583	mod->name, name, ret);
1584	break;
1585
1586	default:
1587	if (sym[i].st_shndx >= info->hdr->e_shnum) {
1588	pr_err("%s: Symbol %s has an invalid section index %u (max %u)\n",
1589	mod->name, name, sym[i].st_shndx, info->hdr->e_shnum - 1);
1590	ret = -ENOEXEC;
1591	break;
1592	}
1593
1594	/* Divert to percpu allocation if a percpu var. */
1595	if (sym[i].st_shndx == info->index.pcpu)
1596	secbase = (unsigned long)mod_percpu(mod);
1597	else
1598	secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1599	sym[i].st_value += secbase;
1600	break;
1601	}
1602	}
1603
1604	return ret;
1605	}
1606
1607	static int apply_relocations(struct module mod, const struct load_info info)
1608	{
1609	unsigned int i;
1610	int err = 0;
1611
1612	/* Now do relocations. */
1613	for (i = 1; i < info->hdr->e_shnum; i++) {
1614	unsigned int infosec = info->sechdrs[i].sh_info;
1615
1616	/* Not a valid relocation section? */
1617	if (infosec >= info->hdr->e_shnum)
1618	continue;
1619
1620	/*
1621	* Don't bother with non-allocated sections.
1622	* An exception is the percpu section, which has separate allocations
1623	* for individual CPUs. We relocate the percpu section in the initial
1624	* ELF template and subsequently copy it to the per-CPU destinations.
1625	*/
1626	if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC) &&
1627	(!infosec \|\| infosec != info->index.pcpu))
1628	continue;
1629
1630	if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1631	err = klp_apply_section_relocs(mod, info->sechdrs,
1632	info->secstrings,
1633	info->strtab,
1634	info->index.sym, i,
1635	NULL);
1636	else if (info->sechdrs[i].sh_type == SHT_REL)
1637	err = apply_relocate(info->sechdrs, info->strtab,
1638	info->index.sym, i, mod);
1639	else if (info->sechdrs[i].sh_type == SHT_RELA)
1640	err = apply_relocate_add(info->sechdrs, info->strtab,
1641	info->index.sym, i, mod);
1642	if (err < 0)
1643	break;
1644	}
1645	return err;
1646	}
1647
1648	/* Additional bytes needed by arch in front of individual sections */
1649	unsigned int __weak arch_mod_section_prepend(struct module *mod,
1650	unsigned int section)
1651	{
1652	/* default implementation just returns zero */
1653	return 0;
1654	}
1655
1656	long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1657	Elf_Shdr *sechdr, unsigned int section)
1658	{
1659	long offset;
1660	long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1661
1662	mod->mem[type].size += arch_mod_section_prepend(mod, section);
1663	offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1664	mod->mem[type].size = offset + sechdr->sh_size;
1665
1666	WARN_ON_ONCE(offset & mask);
1667	return offset \| mask;
1668	}
1669
1670	bool module_init_layout_section(const char *sname)
1671	{
1672	#ifndef CONFIG_MODULE_UNLOAD
1673	if (module_exit_section(sname))
1674	return true;
1675	#endif
1676	return module_init_section(sname);
1677	}
1678
1679	static void __layout_sections(struct module mod, struct load_info info, bool is_init)
1680	{
1681	unsigned int m, i;
1682
1683	/*
1684	* { Mask of required section header flags,
1685	* Mask of excluded section header flags }
1686	*/
1687	static const unsigned long masks[][2] = {
1688	{ SHF_EXECINSTR \| SHF_ALLOC, ARCH_SHF_SMALL },
1689	{ SHF_ALLOC, SHF_WRITE \| ARCH_SHF_SMALL },
1690	{ SHF_RO_AFTER_INIT \| SHF_ALLOC, ARCH_SHF_SMALL },
1691	{ SHF_WRITE \| SHF_ALLOC, ARCH_SHF_SMALL },
1692	{ ARCH_SHF_SMALL \| SHF_ALLOC, 0 }
1693	};
1694	static const int core_m_to_mem_type[] = {
1695	MOD_TEXT,
1696	MOD_RODATA,
1697	MOD_RO_AFTER_INIT,
1698	MOD_DATA,
1699	MOD_DATA,
1700	};
1701	static const int init_m_to_mem_type[] = {
1702	MOD_INIT_TEXT,
1703	MOD_INIT_RODATA,
1704	MOD_INVALID,
1705	MOD_INIT_DATA,
1706	MOD_INIT_DATA,
1707	};
1708
1709	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1710	enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1711
1712	for (i = 0; i < info->hdr->e_shnum; ++i) {
1713	Elf_Shdr *s = &info->sechdrs[i];
1714	const char *sname = info->secstrings + s->sh_name;
1715
1716	if ((s->sh_flags & masks[m][0]) != masks[m][0]
1717	\|\| (s->sh_flags & masks[m][1])
1718	\|\| s->sh_entsize != ~0UL
1719	\|\| is_init != module_init_layout_section(sname))
1720	continue;
1721
1722	if (WARN_ON_ONCE(type == MOD_INVALID))
1723	continue;
1724
1725	/*
1726	* Do not allocate codetag memory as we load it into
1727	* preallocated contiguous memory.
1728	*/
1729	if (codetag_needs_module_section(mod, sname, s->sh_size)) {
1730	/*
1731	* s->sh_entsize won't be used but populate the
1732	* type field to avoid confusion.
1733	*/
1734	s->sh_entsize = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK)
1735	<< SH_ENTSIZE_TYPE_SHIFT;
1736	continue;
1737	}
1738
1739	s->sh_entsize = module_get_offset_and_type(mod, type, s, i);
1740	pr_debug("\t%s\n", sname);
1741	}
1742	}
1743	}
1744
1745	/*
1746	* Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1747	* might -- code, read-only data, read-write data, small data. Tally
1748	* sizes, and place the offsets into sh_entsize fields: high bit means it
1749	* belongs in init.
1750	*/
1751	static void layout_sections(struct module mod, struct load_info info)
1752	{
1753	unsigned int i;
1754
1755	for (i = 0; i < info->hdr->e_shnum; i++)
1756	info->sechdrs[i].sh_entsize = ~0UL;
1757
1758	pr_debug("Core section allocation order for %s:\n", mod->name);
1759	__layout_sections(mod, info, false);
1760
1761	pr_debug("Init section allocation order for %s:\n", mod->name);
1762	__layout_sections(mod, info, true);
1763	}
1764
1765	static void module_license_taint_check(struct module mod, const char license)
1766	{
1767	if (!license)
1768	license = "unspecified";
1769
1770	if (!license_is_gpl_compatible(license)) {
1771	if (!test_taint(TAINT_PROPRIETARY_MODULE))
1772	pr_warn("%s: module license '%s' taints kernel.\n",
1773	mod->name, license);
1774	add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1775	LOCKDEP_NOW_UNRELIABLE);
1776	}
1777	}
1778
1779	static int copy_modinfo_import_ns(struct module mod, struct load_info info)
1780	{
1781	char *ns;
1782	size_t len, total_len = 0;
1783	char buf, p;
1784
1785	for_each_modinfo_entry(ns, info, "import_ns")
1786	total_len += strlen(ns) + 1;
1787
1788	if (!total_len) {
1789	mod->imported_namespaces = NULL;
1790	return 0;
1791	}
1792
1793	buf = kmalloc(total_len, GFP_KERNEL);
1794	if (!buf)
1795	return -ENOMEM;
1796
1797	p = buf;
1798	for_each_modinfo_entry(ns, info, "import_ns") {
1799	len = strlen(ns);
1800	memcpy(p, ns, len);
1801	p += len;
1802	*p++ = '\n';
1803	}
1804	/* Replace trailing newline with null terminator. */
1805	*(p - 1) = '\0';
1806
1807	mod->imported_namespaces = buf;
1808	return 0;
1809	}
1810
1811	static int setup_modinfo(struct module mod, struct load_info info)
1812	{
1813	const struct module_attribute *attr;
1814	char *imported_namespace;
1815	int i, err;
1816
1817	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1818	if (attr->setup)
1819	attr->setup(mod, get_modinfo(info, attr->attr.name));
1820	}
1821
1822	for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1823	/*
1824	* 'module:' prefixed namespaces are implicit, disallow
1825	* explicit imports.
1826	*/
1827	if (strstarts(imported_namespace, "module:")) {
1828	pr_err("%s: module tries to import module namespace: %s\n",
1829	mod->name, imported_namespace);
1830	return -EPERM;
1831	}
1832	}
1833
1834	err = copy_modinfo_import_ns(mod, info);
1835	if (err)
1836	return err;
1837
1838	return 0;
1839	}
1840
1841	static void free_modinfo(struct module *mod)
1842	{
1843	const struct module_attribute *attr;
1844	int i;
1845
1846	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1847	if (attr->free)
1848	attr->free(mod);
1849	}
1850	}
1851
1852	bool __weak module_init_section(const char *name)
1853	{
1854	return strstarts(name, ".init");
1855	}
1856
1857	bool __weak module_exit_section(const char *name)
1858	{
1859	return strstarts(name, ".exit");
1860	}
1861
1862	static int validate_section_offset(const struct load_info info, Elf_Shdr shdr)
1863	{
1864	#if defined(CONFIG_64BIT)
1865	unsigned long long secend;
1866	#else
1867	unsigned long secend;
1868	#endif
1869
1870	/*
1871	* Check for both overflow and offset/size being
1872	* too large.
1873	*/
1874	secend = shdr->sh_offset + shdr->sh_size;
1875	if (secend < shdr->sh_offset \|\| secend > info->len)
1876	return -ENOEXEC;
1877
1878	return 0;
1879	}
1880
1881	/**
1882	* elf_validity_ehdr() - Checks an ELF header for module validity
1883	* @info: Load info containing the ELF header to check
1884	*
1885	* Checks whether an ELF header could belong to a valid module. Checks:
1886	*
1887	* * ELF header is within the data the user provided
1888	* * ELF magic is present
1889	* * It is relocatable (not final linked, not core file, etc.)
1890	* * The header's machine type matches what the architecture expects.
1891	* * Optional arch-specific hook for other properties
1892	* - module_elf_check_arch() is currently only used by PPC to check
1893	* ELF ABI version, but may be used by others in the future.
1894	*
1895	* Return: %0 if valid, %-ENOEXEC on failure.
1896	*/
1897	static int elf_validity_ehdr(const struct load_info *info)
1898	{
1899	if (info->len < sizeof(*(info->hdr))) {
1900	pr_err("Invalid ELF header len %lu\n", info->len);
1901	return -ENOEXEC;
1902	}
1903	if (memcmp(info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1904	pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1905	return -ENOEXEC;
1906	}
1907	if (info->hdr->e_type != ET_REL) {
1908	pr_err("Invalid ELF header type: %u != %u\n",
1909	info->hdr->e_type, ET_REL);
1910	return -ENOEXEC;
1911	}
1912	if (!elf_check_arch(info->hdr)) {
1913	pr_err("Invalid architecture in ELF header: %u\n",
1914	info->hdr->e_machine);
1915	return -ENOEXEC;
1916	}
1917	if (!module_elf_check_arch(info->hdr)) {
1918	pr_err("Invalid module architecture in ELF header: %u\n",
1919	info->hdr->e_machine);
1920	return -ENOEXEC;
1921	}
1922	return 0;
1923	}
1924
1925	/**
1926	* elf_validity_cache_sechdrs() - Cache section headers if valid
1927	* @info: Load info to compute section headers from
1928	*
1929	* Checks:
1930	*
1931	* * ELF header is valid (see elf_validity_ehdr())
1932	* * Section headers are the size we expect
1933	* * Section array fits in the user provided data
1934	* * Section index 0 is NULL
1935	* * Section contents are inbounds
1936	*
1937	* Then updates @info with a &load_info->sechdrs pointer if valid.
1938	*
1939	* Return: %0 if valid, negative error code if validation failed.
1940	*/
1941	static int elf_validity_cache_sechdrs(struct load_info *info)
1942	{
1943	Elf_Shdr *sechdrs;
1944	Elf_Shdr *shdr;
1945	int i;
1946	int err;
1947
1948	err = elf_validity_ehdr(info);
1949	if (err < 0)
1950	return err;
1951
1952	if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1953	pr_err("Invalid ELF section header size\n");
1954	return -ENOEXEC;
1955	}
1956
1957	/*
1958	* e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1959	* known and small. So e_shnum * sizeof(Elf_Shdr)
1960	* will not overflow unsigned long on any platform.
1961	*/
1962	if (info->hdr->e_shoff >= info->len
1963	\|\| (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1964	info->len - info->hdr->e_shoff)) {
1965	pr_err("Invalid ELF section header overflow\n");
1966	return -ENOEXEC;
1967	}
1968
1969	sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1970
1971	/*
1972	* The code assumes that section 0 has a length of zero and
1973	* an addr of zero, so check for it.
1974	*/
1975	if (sechdrs[0].sh_type != SHT_NULL
1976	\|\| sechdrs[0].sh_size != 0
1977	\|\| sechdrs[0].sh_addr != 0) {
1978	pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1979	sechdrs[0].sh_type);
1980	return -ENOEXEC;
1981	}
1982
1983	/* Validate contents are inbounds */
1984	for (i = 1; i < info->hdr->e_shnum; i++) {
1985	shdr = &sechdrs[i];
1986	switch (shdr->sh_type) {
1987	case SHT_NULL:
1988	case SHT_NOBITS:
1989	/* No contents, offset/size don't mean anything */
1990	continue;
1991	default:
1992	err = validate_section_offset(info, shdr);
1993	if (err < 0) {
1994	pr_err("Invalid ELF section in module (section %u type %u)\n",
1995	i, shdr->sh_type);
1996	return err;
1997	}
1998	}
1999	}
2000
2001	info->sechdrs = sechdrs;
2002
2003	return 0;
2004	}
2005
2006	/**
2007	* elf_validity_cache_secstrings() - Caches section names if valid
2008	* @info: Load info to cache section names from. Must have valid sechdrs.
2009	*
2010	* Specifically checks:
2011	*
2012	* * Section name table index is inbounds of section headers
2013	* * Section name table type is SHT_STRTAB
2014	* * Section name table is not empty
2015	* * Section name table is NUL terminated
2016	* * All section name offsets are inbounds of the section
2017	*
2018	* Then updates @info with a &load_info->secstrings pointer if valid.
2019	*
2020	* Return: %0 if valid, negative error code if validation failed.
2021	*/
2022	static int elf_validity_cache_secstrings(struct load_info *info)
2023	{
2024	Elf_Shdr strhdr, shdr;
2025	char *secstrings;
2026	int i;
2027
2028	/*
2029	* Verify if the section name table index is valid.
2030	*/
2031	if (info->hdr->e_shstrndx == SHN_UNDEF
2032	\|\| info->hdr->e_shstrndx >= info->hdr->e_shnum) {
2033	pr_err("Invalid ELF section name index: %d \|\| e_shstrndx (%d) >= e_shnum (%d)\n",
2034	info->hdr->e_shstrndx, info->hdr->e_shstrndx,
2035	info->hdr->e_shnum);
2036	return -ENOEXEC;
2037	}
2038
2039	strhdr = &info->sechdrs[info->hdr->e_shstrndx];
2040
2041	if (strhdr->sh_type != SHT_STRTAB) {
2042	pr_err("Invalid ELF section name table type: %u\n", strhdr->sh_type);
2043	return -ENOEXEC;
2044	}
2045
2046	/*
2047	* The section name table must be NUL-terminated, as required
2048	* by the spec. This makes strcmp and pr_* calls that access
2049	* strings in the section safe.
2050	*/
2051	secstrings = (void *)info->hdr + strhdr->sh_offset;
2052	if (strhdr->sh_size == 0) {
2053	pr_err("empty section name table\n");
2054	return -ENOEXEC;
2055	}
2056	if (secstrings[strhdr->sh_size - 1] != '\0') {
2057	pr_err("ELF Spec violation: section name table isn't null terminated\n");
2058	return -ENOEXEC;
2059	}
2060
2061	for (i = 0; i < info->hdr->e_shnum; i++) {
2062	shdr = &info->sechdrs[i];
2063	/* SHT_NULL means sh_name has an undefined value */
2064	if (shdr->sh_type == SHT_NULL)
2065	continue;
2066	if (shdr->sh_name >= strhdr->sh_size) {
2067	pr_err("Invalid ELF section name in module (section %u type %u)\n",
2068	i, shdr->sh_type);
2069	return -ENOEXEC;
2070	}
2071	}
2072
2073	info->secstrings = secstrings;
2074	return 0;
2075	}
2076
2077	/**
2078	* elf_validity_cache_index_info() - Validate and cache modinfo section
2079	* @info: Load info to populate the modinfo index on.
2080	* Must have &load_info->sechdrs and &load_info->secstrings populated
2081	*
2082	* Checks that if there is a .modinfo section, it is unique.
2083	* Then, it caches its index in &load_info->index.info.
2084	* Finally, it tries to populate the name to improve error messages.
2085	*
2086	* Return: %0 if valid, %-ENOEXEC if multiple modinfo sections were found.
2087	*/
2088	static int elf_validity_cache_index_info(struct load_info *info)
2089	{
2090	int info_idx;
2091
2092	info_idx = find_any_unique_sec(info, ".modinfo");
2093
2094	if (info_idx == 0)
2095	/* Early return, no .modinfo */
2096	return 0;
2097
2098	if (info_idx < 0) {
2099	pr_err("Only one .modinfo section must exist.\n");
2100	return -ENOEXEC;
2101	}
2102
2103	info->index.info = info_idx;
2104	/* Try to find a name early so we can log errors with a module name */
2105	info->name = get_modinfo(info, "name");
2106
2107	return 0;
2108	}
2109
2110	/**
2111	* elf_validity_cache_index_mod() - Validates and caches this_module section
2112	* @info: Load info to cache this_module on.
2113	* Must have &load_info->sechdrs and &load_info->secstrings populated
2114	*
2115	* The ".gnu.linkonce.this_module" ELF section is special. It is what modpost
2116	* uses to refer to __this_module and let's use rely on THIS_MODULE to point
2117	* to &__this_module properly. The kernel's modpost declares it on each
2118	* modules's *.mod.c file. If the struct module of the kernel changes a full
2119	* kernel rebuild is required.
2120	*
2121	* We have a few expectations for this special section, this function
2122	* validates all this for us:
2123	*
2124	* * The section has contents
2125	* * The section is unique
2126	* * We expect the kernel to always have to allocate it: SHF_ALLOC
2127	* * The section size must match the kernel's run time's struct module
2128	* size
2129	*
2130	* If all checks pass, the index will be cached in &load_info->index.mod
2131	*
2132	* Return: %0 on validation success, %-ENOEXEC on failure
2133	*/
2134	static int elf_validity_cache_index_mod(struct load_info *info)
2135	{
2136	Elf_Shdr *shdr;
2137	int mod_idx;
2138
2139	mod_idx = find_any_unique_sec(info, ".gnu.linkonce.this_module");
2140	if (mod_idx <= 0) {
2141	pr_err("module %s: Exactly one .gnu.linkonce.this_module section must exist.\n",
2142	info->name ?: "(missing .modinfo section or name field)");
2143	return -ENOEXEC;
2144	}
2145
2146	shdr = &info->sechdrs[mod_idx];
2147
2148	if (shdr->sh_type == SHT_NOBITS) {
2149	pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
2150	info->name ?: "(missing .modinfo section or name field)");
2151	return -ENOEXEC;
2152	}
2153
2154	if (!(shdr->sh_flags & SHF_ALLOC)) {
2155	pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
2156	info->name ?: "(missing .modinfo section or name field)");
2157	return -ENOEXEC;
2158	}
2159
2160	if (shdr->sh_size != sizeof(struct module)) {
2161	pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
2162	info->name ?: "(missing .modinfo section or name field)");
2163	return -ENOEXEC;
2164	}
2165
2166	info->index.mod = mod_idx;
2167
2168	return 0;
2169	}
2170
2171	/**
2172	* elf_validity_cache_index_sym() - Validate and cache symtab index
2173	* @info: Load info to cache symtab index in.
2174	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2175	*
2176	* Checks that there is exactly one symbol table, then caches its index in
2177	* &load_info->index.sym.
2178	*
2179	* Return: %0 if valid, %-ENOEXEC on failure.
2180	*/
2181	static int elf_validity_cache_index_sym(struct load_info *info)
2182	{
2183	unsigned int sym_idx;
2184	unsigned int num_sym_secs = 0;
2185	int i;
2186
2187	for (i = 1; i < info->hdr->e_shnum; i++) {
2188	if (info->sechdrs[i].sh_type == SHT_SYMTAB) {
2189	num_sym_secs++;
2190	sym_idx = i;
2191	}
2192	}
2193
2194	if (num_sym_secs != 1) {
2195	pr_warn("%s: module has no symbols (stripped?)\n",
2196	info->name ?: "(missing .modinfo section or name field)");
2197	return -ENOEXEC;
2198	}
2199
2200	info->index.sym = sym_idx;
2201
2202	return 0;
2203	}
2204
2205	/**
2206	* elf_validity_cache_index_str() - Validate and cache strtab index
2207	* @info: Load info to cache strtab index in.
2208	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2209	* Must have &load_info->index.sym populated.
2210	*
2211	* Looks at the symbol table's associated string table, makes sure it is
2212	* in-bounds and of type SHT_STRTAB, and caches it.
2213	*
2214	* Return: %0 if valid, %-ENOEXEC on failure.
2215	*/
2216	static int elf_validity_cache_index_str(struct load_info *info)
2217	{
2218	unsigned int str_idx = info->sechdrs[info->index.sym].sh_link;
2219
2220	if (str_idx == SHN_UNDEF \|\| str_idx >= info->hdr->e_shnum) {
2221	pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
2222	str_idx, str_idx, info->hdr->e_shnum);
2223	return -ENOEXEC;
2224	}
2225
2226	if (info->sechdrs[str_idx].sh_type != SHT_STRTAB) {
2227	pr_err("Invalid ELF symbol string table type: %u\n",
2228	info->sechdrs[str_idx].sh_type);
2229	return -ENOEXEC;
2230	}
2231
2232	info->index.str = str_idx;
2233	return 0;
2234	}
2235
2236	/**
2237	* elf_validity_cache_index_versions() - Validate and cache version indices
2238	* @info: Load info to cache version indices in.
2239	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2240	* @flags: Load flags, relevant to suppress version loading, see
2241	* uapi/linux/module.h
2242	*
2243	* If we're ignoring modversions based on @flags, zero all version indices
2244	* and return validity. Othewrise check:
2245	*
2246	* * If "__version_ext_crcs" is present, "__version_ext_names" is present
2247	* * There is a name present for every crc
2248	*
2249	* Then populate:
2250	*
2251	* * &load_info->index.vers
2252	* * &load_info->index.vers_ext_crc
2253	* * &load_info->index.vers_ext_names
2254	*
2255	* if present.
2256	*
2257	* Return: %0 if valid, %-ENOEXEC on failure.
2258	*/
2259	static int elf_validity_cache_index_versions(struct load_info *info, int flags)
2260	{
2261	unsigned int vers_ext_crc;
2262	unsigned int vers_ext_name;
2263	size_t crc_count;
2264	size_t remaining_len;
2265	size_t name_size;
2266	char *name;
2267
2268	/* If modversions were suppressed, pretend we didn't find any */
2269	if (flags & MODULE_INIT_IGNORE_MODVERSIONS) {
2270	info->index.vers = 0;
2271	info->index.vers_ext_crc = 0;
2272	info->index.vers_ext_name = 0;
2273	return 0;
2274	}
2275
2276	vers_ext_crc = find_sec(info, "__version_ext_crcs");
2277	vers_ext_name = find_sec(info, "__version_ext_names");
2278
2279	/* If we have one field, we must have the other */
2280	if (!!vers_ext_crc != !!vers_ext_name) {
2281	pr_err("extended version crc+name presence does not match");
2282	return -ENOEXEC;
2283	}
2284
2285	/*
2286	* If we have extended version information, we should have the same
2287	* number of entries in every section.
2288	*/
2289	if (vers_ext_crc) {
2290	crc_count = info->sechdrs[vers_ext_crc].sh_size / sizeof(u32);
2291	name = (void *)info->hdr +
2292	info->sechdrs[vers_ext_name].sh_offset;
2293	remaining_len = info->sechdrs[vers_ext_name].sh_size;
2294
2295	while (crc_count--) {
2296	name_size = strnlen(name, remaining_len) + 1;
2297	if (name_size > remaining_len) {
2298	pr_err("more extended version crcs than names");
2299	return -ENOEXEC;
2300	}
2301	remaining_len -= name_size;
2302	name += name_size;
2303	}
2304	}
2305
2306	info->index.vers = find_sec(info, "__versions");
2307	info->index.vers_ext_crc = vers_ext_crc;
2308	info->index.vers_ext_name = vers_ext_name;
2309	return 0;
2310	}
2311
2312	/**
2313	* elf_validity_cache_index() - Resolve, validate, cache section indices
2314	* @info: Load info to read from and update.
2315	* &load_info->sechdrs and &load_info->secstrings must be populated.
2316	* @flags: Load flags, relevant to suppress version loading, see
2317	* uapi/linux/module.h
2318	*
2319	* Populates &load_info->index, validating as it goes.
2320	* See child functions for per-field validation:
2321	*
2322	* * elf_validity_cache_index_info()
2323	* * elf_validity_cache_index_mod()
2324	* * elf_validity_cache_index_sym()
2325	* * elf_validity_cache_index_str()
2326	* * elf_validity_cache_index_versions()
2327	*
2328	* If CONFIG_SMP is enabled, load the percpu section by name with no
2329	* validation.
2330	*
2331	* Return: 0 on success, negative error code if an index failed validation.
2332	*/
2333	static int elf_validity_cache_index(struct load_info *info, int flags)
2334	{
2335	int err;
2336
2337	err = elf_validity_cache_index_info(info);
2338	if (err < 0)
2339	return err;
2340	err = elf_validity_cache_index_mod(info);
2341	if (err < 0)
2342	return err;
2343	err = elf_validity_cache_index_sym(info);
2344	if (err < 0)
2345	return err;
2346	err = elf_validity_cache_index_str(info);
2347	if (err < 0)
2348	return err;
2349	err = elf_validity_cache_index_versions(info, flags);
2350	if (err < 0)
2351	return err;
2352
2353	info->index.pcpu = find_pcpusec(info);
2354
2355	return 0;
2356	}
2357
2358	/**
2359	* elf_validity_cache_strtab() - Validate and cache symbol string table
2360	* @info: Load info to read from and update.
2361	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2362	* Must have &load_info->index populated.
2363	*
2364	* Checks:
2365	*
2366	* * The string table is not empty.
2367	* * The string table starts and ends with NUL (required by ELF spec).
2368	* * Every &Elf_Sym->st_name offset in the symbol table is inbounds of the
2369	* string table.
2370	*
2371	* And caches the pointer as &load_info->strtab in @info.
2372	*
2373	* Return: 0 on success, negative error code if a check failed.
2374	*/
2375	static int elf_validity_cache_strtab(struct load_info *info)
2376	{
2377	Elf_Shdr *str_shdr = &info->sechdrs[info->index.str];
2378	Elf_Shdr *sym_shdr = &info->sechdrs[info->index.sym];
2379	char strtab = (char )info->hdr + str_shdr->sh_offset;
2380	Elf_Sym syms = (void )info->hdr + sym_shdr->sh_offset;
2381	int i;
2382
2383	if (str_shdr->sh_size == 0) {
2384	pr_err("empty symbol string table\n");
2385	return -ENOEXEC;
2386	}
2387	if (strtab[0] != '\0') {
2388	pr_err("symbol string table missing leading NUL\n");
2389	return -ENOEXEC;
2390	}
2391	if (strtab[str_shdr->sh_size - 1] != '\0') {
2392	pr_err("symbol string table isn't NUL terminated\n");
2393	return -ENOEXEC;
2394	}
2395
2396	/*
2397	* Now that we know strtab is correctly structured, check symbol
2398	* starts are inbounds before they're used later.
2399	*/
2400	for (i = 0; i < sym_shdr->sh_size / sizeof(*syms); i++) {
2401	if (syms[i].st_name >= str_shdr->sh_size) {
2402	pr_err("symbol name out of bounds in string table");
2403	return -ENOEXEC;
2404	}
2405	}
2406
2407	info->strtab = strtab;
2408	return 0;
2409	}
2410
2411	/*
2412	* Check userspace passed ELF module against our expectations, and cache
2413	* useful variables for further processing as we go.
2414	*
2415	* This does basic validity checks against section offsets and sizes, the
2416	* section name string table, and the indices used for it (sh_name).
2417	*
2418	* As a last step, since we're already checking the ELF sections we cache
2419	* useful variables which will be used later for our convenience:
2420	*
2421	* o pointers to section headers
2422	* o cache the modinfo symbol section
2423	* o cache the string symbol section
2424	* o cache the module section
2425	*
2426	* As a last step we set info->mod to the temporary copy of the module in
2427	* info->hdr. The final one will be allocated in move_module(). Any
2428	* modifications we make to our copy of the module will be carried over
2429	* to the final minted module.
2430	*/
2431	static int elf_validity_cache_copy(struct load_info *info, int flags)
2432	{
2433	int err;
2434
2435	err = elf_validity_cache_sechdrs(info);
2436	if (err < 0)
2437	return err;
2438	err = elf_validity_cache_secstrings(info);
2439	if (err < 0)
2440	return err;
2441	err = elf_validity_cache_index(info, flags);
2442	if (err < 0)
2443	return err;
2444	err = elf_validity_cache_strtab(info);
2445	if (err < 0)
2446	return err;
2447
2448	/* This is temporary: point mod into copy of data. */
2449	info->mod = (void *)info->hdr + info->sechdrs[info->index.mod].sh_offset;
2450
2451	/*
2452	* If we didn't load the .modinfo 'name' field earlier, fall back to
2453	* on-disk struct mod 'name' field.
2454	*/
2455	if (!info->name)
2456	info->name = info->mod->name;
2457
2458	return 0;
2459	}
2460
2461	#define COPY_CHUNK_SIZE (16*PAGE_SIZE)
2462
2463	static int copy_chunked_from_user(void dst, const void __user usrc, unsigned long len)
2464	{
2465	do {
2466	unsigned long n = min(len, COPY_CHUNK_SIZE);
2467
2468	if (copy_from_user(dst, usrc, n) != 0)
2469	return -EFAULT;
2470	cond_resched();
2471	dst += n;
2472	usrc += n;
2473	len -= n;
2474	} while (len);
2475	return 0;
2476	}
2477
2478	static int check_modinfo_livepatch(struct module mod, struct load_info info)
2479	{
2480	if (!get_modinfo(info, "livepatch"))
2481	/* Nothing more to do */
2482	return 0;
2483
2484	if (set_livepatch_module(mod))
2485	return 0;
2486
2487	pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
2488	mod->name);
2489	return -ENOEXEC;
2490	}
2491
2492	static void check_modinfo_retpoline(struct module mod, struct load_info info)
2493	{
2494	if (retpoline_module_ok(get_modinfo(info, "retpoline")))
2495	return;
2496
2497	pr_warn("%s: loading module not compiled with retpoline compiler.\n",
2498	mod->name);
2499	}
2500
2501	/* Sets info->hdr and info->len. */
2502	static int copy_module_from_user(const void __user *umod, unsigned long len,
2503	struct load_info *info)
2504	{
2505	int err;
2506
2507	info->len = len;
2508	if (info->len < sizeof(*(info->hdr)))
2509	return -ENOEXEC;
2510
2511	err = security_kernel_load_data(LOADING_MODULE, true);
2512	if (err)
2513	return err;
2514
2515	/* Suck in entire file: we'll want most of it. */
2516	info->hdr = __vmalloc(info->len, GFP_KERNEL \| __GFP_NOWARN);
2517	if (!info->hdr)
2518	return -ENOMEM;
2519
2520	if (copy_chunked_from_user(info->hdr, umod, info->len) != 0) {
2521	err = -EFAULT;
2522	goto out;
2523	}
2524
2525	err = security_kernel_post_load_data((char *)info->hdr, info->len,
2526	LOADING_MODULE, "init_module");
2527	out:
2528	if (err)
2529	vfree(info->hdr);
2530
2531	return err;
2532	}
2533
2534	static void free_copy(struct load_info *info, int flags)
2535	{
2536	if (flags & MODULE_INIT_COMPRESSED_FILE)
2537	module_decompress_cleanup(info);
2538	else
2539	vfree(info->hdr);
2540	}
2541
2542	static int rewrite_section_headers(struct load_info *info, int flags)
2543	{
2544	unsigned int i;
2545
2546	/* This should always be true, but let's be sure. */
2547	info->sechdrs[0].sh_addr = 0;
2548
2549	for (i = 1; i < info->hdr->e_shnum; i++) {
2550	Elf_Shdr *shdr = &info->sechdrs[i];
2551
2552	/*
2553	* Mark all sections sh_addr with their address in the
2554	* temporary image.
2555	*/
2556	shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2557
2558	}
2559
2560	/* Track but don't keep modinfo and version sections. */
2561	info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2562	info->sechdrs[info->index.vers_ext_crc].sh_flags &=
2563	~(unsigned long)SHF_ALLOC;
2564	info->sechdrs[info->index.vers_ext_name].sh_flags &=
2565	~(unsigned long)SHF_ALLOC;
2566	info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2567
2568	return 0;
2569	}
2570
2571	static const char *const module_license_offenders[] = {
2572	/* driverloader was caught wrongly pretending to be under GPL */
2573	"driverloader",
2574
2575	/* lve claims to be GPL but upstream won't provide source */
2576	"lve",
2577	};
2578
2579	/*
2580	* These calls taint the kernel depending certain module circumstances */
2581	static void module_augment_kernel_taints(struct module mod, struct load_info info)
2582	{
2583	int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2584	size_t i;
2585
2586	if (!get_modinfo(info, "intree")) {
2587	if (!test_taint(TAINT_OOT_MODULE))
2588	pr_warn("%s: loading out-of-tree module taints kernel.\n",
2589	mod->name);
2590	add_taint_module(mod, TAINT_OOT_MODULE, LOCKDEP_STILL_OK);
2591	}
2592
2593	check_modinfo_retpoline(mod, info);
2594
2595	if (get_modinfo(info, "staging")) {
2596	add_taint_module(mod, TAINT_CRAP, LOCKDEP_STILL_OK);
2597	pr_warn("%s: module is from the staging directory, the quality "
2598	"is unknown, you have been warned.\n", mod->name);
2599	}
2600
2601	if (is_livepatch_module(mod)) {
2602	add_taint_module(mod, TAINT_LIVEPATCH, LOCKDEP_STILL_OK);
2603	pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2604	mod->name);
2605	}
2606
2607	module_license_taint_check(mod, get_modinfo(info, "license"));
2608
2609	if (get_modinfo(info, "test")) {
2610	if (!test_taint(TAINT_TEST))
2611	pr_warn("%s: loading test module taints kernel.\n",
2612	mod->name);
2613	add_taint_module(mod, TAINT_TEST, LOCKDEP_STILL_OK);
2614	}
2615	#ifdef CONFIG_MODULE_SIG
2616	mod->sig_ok = info->sig_ok;
2617	if (!mod->sig_ok) {
2618	pr_notice_once("%s: module verification failed: signature "
2619	"and/or required key missing - tainting "
2620	"kernel\n", mod->name);
2621	add_taint_module(mod, TAINT_UNSIGNED_MODULE, LOCKDEP_STILL_OK);
2622	}
2623	#endif
2624
2625	/*
2626	* ndiswrapper is under GPL by itself, but loads proprietary modules.
2627	* Don't use add_taint_module(), as it would prevent ndiswrapper from
2628	* using GPL-only symbols it needs.
2629	*/
2630	if (strcmp(mod->name, "ndiswrapper") == 0)
2631	add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2632
2633	for (i = 0; i < ARRAY_SIZE(module_license_offenders); ++i) {
2634	if (strcmp(mod->name, module_license_offenders[i]) == 0)
2635	add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2636	LOCKDEP_NOW_UNRELIABLE);
2637	}
2638
2639	if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2640	pr_warn("%s: module license taints kernel.\n", mod->name);
2641
2642	}
2643
2644	static int check_modinfo(struct module mod, struct load_info info, int flags)
2645	{
2646	const char *modmagic = get_modinfo(info, "vermagic");
2647	int err;
2648
2649	if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2650	modmagic = NULL;
2651
2652	/* This is allowed: modprobe --force will invalidate it. */
2653	if (!modmagic) {
2654	err = try_to_force_load(mod, "bad vermagic");
2655	if (err)
2656	return err;
2657	} else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2658	pr_err("%s: version magic '%s' should be '%s'\n",
2659	info->name, modmagic, vermagic);
2660	return -ENOEXEC;
2661	}
2662
2663	err = check_modinfo_livepatch(mod, info);
2664	if (err)
2665	return err;
2666
2667	return 0;
2668	}
2669
2670	static int find_module_sections(struct module mod, struct load_info info)
2671	{
2672	mod->kp = section_objs(info, "__param",
2673	sizeof(*mod->kp), &mod->num_kp);
2674	mod->syms = section_objs(info, "__ksymtab",
2675	sizeof(*mod->syms), &mod->num_syms);
2676	mod->crcs = section_addr(info, "__kcrctab");
2677	mod->flagstab = section_addr(info, "__kflagstab");
2678
2679	if (section_addr(info, "__ksymtab_gpl"))
2680	pr_warn("%s: ignoring obsolete section __ksymtab_gpl\n",
2681	mod->name);
2682	if (section_addr(info, "__kcrctab_gpl"))
2683	pr_warn("%s: ignoring obsolete section __kcrctab_gpl\n",
2684	mod->name);
2685
2686	#ifdef CONFIG_CONSTRUCTORS
2687	mod->ctors = section_objs(info, ".ctors",
2688	sizeof(*mod->ctors), &mod->num_ctors);
2689	if (!mod->ctors)
2690	mod->ctors = section_objs(info, ".init_array",
2691	sizeof(*mod->ctors), &mod->num_ctors);
2692	else if (find_sec(info, ".init_array")) {
2693	/*
2694	* This shouldn't happen with same compiler and binutils
2695	* building all parts of the module.
2696	*/
2697	pr_warn("%s: has both .ctors and .init_array.\n",
2698	mod->name);
2699	return -EINVAL;
2700	}
2701	#endif
2702
2703	mod->noinstr_text_start = section_objs(info, ".noinstr.text", 1,
2704	&mod->noinstr_text_size);
2705
2706	#ifdef CONFIG_TRACEPOINTS
2707	mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2708	sizeof(*mod->tracepoints_ptrs),
2709	&mod->num_tracepoints);
2710	#endif
2711	#ifdef CONFIG_TREE_SRCU
2712	mod->srcu_struct_ptrs = section_objs(info, "___srcu_struct_ptrs",
2713	sizeof(*mod->srcu_struct_ptrs),
2714	&mod->num_srcu_structs);
2715	#endif
2716	#ifdef CONFIG_BPF_EVENTS
2717	mod->bpf_raw_events = section_objs(info, "__bpf_raw_tp_map",
2718	sizeof(*mod->bpf_raw_events),
2719	&mod->num_bpf_raw_events);
2720	#endif
2721	#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2722	mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2723	mod->btf_base_data = any_section_objs(info, ".BTF.base", 1,
2724	&mod->btf_base_data_size);
2725	#endif
2726	#ifdef CONFIG_JUMP_LABEL
2727	mod->jump_entries = section_objs(info, "__jump_table",
2728	sizeof(*mod->jump_entries),
2729	&mod->num_jump_entries);
2730	#endif
2731	#ifdef CONFIG_EVENT_TRACING
2732	mod->trace_events = section_objs(info, "_ftrace_events",
2733	sizeof(*mod->trace_events),
2734	&mod->num_trace_events);
2735	mod->trace_evals = section_objs(info, "_ftrace_eval_map",
2736	sizeof(*mod->trace_evals),
2737	&mod->num_trace_evals);
2738	#endif
2739	#ifdef CONFIG_TRACING
2740	mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2741	sizeof(*mod->trace_bprintk_fmt_start),
2742	&mod->num_trace_bprintk_fmt);
2743	#endif
2744	#ifdef CONFIG_DYNAMIC_FTRACE
2745	/* sechdrs[0].sh_size is always zero */
2746	mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2747	sizeof(*mod->ftrace_callsites),
2748	&mod->num_ftrace_callsites);
2749	#endif
2750	#ifdef CONFIG_FUNCTION_ERROR_INJECTION
2751	mod->ei_funcs = section_objs(info, "_error_injection_whitelist",
2752	sizeof(*mod->ei_funcs),
2753	&mod->num_ei_funcs);
2754	#endif
2755	#ifdef CONFIG_KPROBES
2756	mod->kprobes_text_start = section_objs(info, ".kprobes.text", 1,
2757	&mod->kprobes_text_size);
2758	mod->kprobe_blacklist = section_objs(info, "_kprobe_blacklist",
2759	sizeof(unsigned long),
2760	&mod->num_kprobe_blacklist);
2761	#endif
2762	#ifdef CONFIG_PRINTK_INDEX
2763	mod->printk_index_start = section_objs(info, ".printk_index",
2764	sizeof(*mod->printk_index_start),
2765	&mod->printk_index_size);
2766	#endif
2767	#ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2768	mod->static_call_sites = section_objs(info, ".static_call_sites",
2769	sizeof(*mod->static_call_sites),
2770	&mod->num_static_call_sites);
2771	#endif
2772	#if IS_ENABLED(CONFIG_KUNIT)
2773	mod->kunit_suites = section_objs(info, ".kunit_test_suites",
2774	sizeof(*mod->kunit_suites),
2775	&mod->num_kunit_suites);
2776	mod->kunit_init_suites = section_objs(info, ".kunit_init_test_suites",
2777	sizeof(*mod->kunit_init_suites),
2778	&mod->num_kunit_init_suites);
2779	#endif
2780
2781	mod->extable = section_objs(info, "__ex_table",
2782	sizeof(*mod->extable), &mod->num_exentries);
2783
2784	if (section_addr(info, "__obsparm"))
2785	pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2786
2787	#ifdef CONFIG_DYNAMIC_DEBUG_CORE
2788	mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
2789	sizeof(*mod->dyndbg_info.descs),
2790	&mod->dyndbg_info.num_descs);
2791	mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
2792	sizeof(*mod->dyndbg_info.classes),
2793	&mod->dyndbg_info.num_classes);
2794	#endif
2795
2796	return 0;
2797	}
2798
2799	static int move_module(struct module mod, struct load_info info)
2800	{
2801	int i, ret;
2802	enum mod_mem_type t = MOD_MEM_NUM_TYPES;
2803	bool codetag_section_found = false;
2804
2805	for_each_mod_mem_type(type) {
2806	if (!mod->mem[type].size) {
2807	mod->mem[type].base = NULL;
2808	continue;
2809	}
2810
2811	ret = module_memory_alloc(mod, type);
2812	if (ret) {
2813	t = type;
2814	goto out_err;
2815	}
2816	}
2817
2818	/* Transfer each section which specifies SHF_ALLOC */
2819	pr_debug("Final section addresses for %s:\n", mod->name);
2820	for (i = 0; i < info->hdr->e_shnum; i++) {
2821	void *dest;
2822	Elf_Shdr *shdr = &info->sechdrs[i];
2823	const char *sname;
2824
2825	if (!(shdr->sh_flags & SHF_ALLOC))
2826	continue;
2827
2828	sname = info->secstrings + shdr->sh_name;
2829	/*
2830	* Load codetag sections separately as they might still be used
2831	* after module unload.
2832	*/
2833	if (codetag_needs_module_section(mod, sname, shdr->sh_size)) {
2834	dest = codetag_alloc_module_section(mod, sname, shdr->sh_size,
2835	arch_mod_section_prepend(mod, i), shdr->sh_addralign);
2836	if (WARN_ON(!dest)) {
2837	ret = -EINVAL;
2838	goto out_err;
2839	}
2840	if (IS_ERR(dest)) {
2841	ret = PTR_ERR(dest);
2842	goto out_err;
2843	}
2844	codetag_section_found = true;
2845	} else {
2846	enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2847	unsigned long offset = shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK;
2848
2849	dest = mod->mem[type].base + offset;
2850	}
2851
2852	if (shdr->sh_type != SHT_NOBITS) {
2853	/*
2854	* Our ELF checker already validated this, but let's
2855	* be pedantic and make the goal clearer. We actually
2856	* end up copying over all modifications made to the
2857	* userspace copy of the entire struct module.
2858	*/
2859	if (i == info->index.mod &&
2860	(WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2861	ret = -ENOEXEC;
2862	goto out_err;
2863	}
2864	memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2865	}
2866	/*
2867	* Update the userspace copy's ELF section address to point to
2868	* our newly allocated memory as a pure convenience so that
2869	* users of info can keep taking advantage and using the newly
2870	* minted official memory area.
2871	*/
2872	shdr->sh_addr = (unsigned long)dest;
2873	pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2874	(long)shdr->sh_size, info->secstrings + shdr->sh_name);
2875	}
2876
2877	return 0;
2878	out_err:
2879	module_memory_restore_rox(mod);
2880	while (t--)
2881	module_memory_free(mod, t);
2882	if (codetag_section_found)
2883	codetag_free_module_sections(mod);
2884
2885	return ret;
2886	}
2887
2888	static int check_export_symbol_sections(struct module *mod)
2889	{
2890	if (mod->num_syms && !mod->flagstab) {
2891	pr_err("%s: no flags for exported symbols\n", mod->name);
2892	return -ENOEXEC;
2893	}
2894	#ifdef CONFIG_MODVERSIONS
2895	if (mod->num_syms && !mod->crcs) {
2896	return try_to_force_load(mod,
2897	"no versions for exported symbols");
2898	}
2899	#endif
2900	return 0;
2901	}
2902
2903	static void flush_module_icache(const struct module *mod)
2904	{
2905	/*
2906	* Flush the instruction cache, since we've played with text.
2907	* Do it before processing of module parameters, so the module
2908	* can provide parameter accessor functions of its own.
2909	*/
2910	for_each_mod_mem_type(type) {
2911	const struct module_memory *mod_mem = &mod->mem[type];
2912
2913	if (mod_mem->size) {
2914	flush_icache_range((unsigned long)mod_mem->base,
2915	(unsigned long)mod_mem->base + mod_mem->size);
2916	}
2917	}
2918	}
2919
2920	bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2921	{
2922	return true;
2923	}
2924
2925	int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2926	Elf_Shdr *sechdrs,
2927	char *secstrings,
2928	struct module *mod)
2929	{
2930	return 0;
2931	}
2932
2933	/* module_blacklist is a comma-separated list of module names */
2934	static char *module_blacklist;
2935	static bool blacklisted(const char *module_name)
2936	{
2937	const char *p;
2938	size_t len;
2939
2940	if (!module_blacklist)
2941	return false;
2942
2943	for (p = module_blacklist; *p; p += len) {
2944	len = strcspn(p, ",");
2945	if (strlen(module_name) == len && !memcmp(module_name, p, len))
2946	return true;
2947	if (p[len] == ',')
2948	len++;
2949	}
2950	return false;
2951	}
2952	core_param(module_blacklist, module_blacklist, charp, 0400);
2953
2954	static struct module layout_and_allocate(struct load_info info, int flags)
2955	{
2956	struct module *mod;
2957	int err;
2958
2959	/* Allow arches to frob section contents and sizes. */
2960	err = module_frob_arch_sections(info->hdr, info->sechdrs,
2961	info->secstrings, info->mod);
2962	if (err < 0)
2963	return ERR_PTR(err);
2964
2965	err = module_enforce_rwx_sections(info->hdr, info->sechdrs,
2966	info->secstrings, info->mod);
2967	if (err < 0)
2968	return ERR_PTR(err);
2969
2970	/* We will do a special allocation for per-cpu sections later. */
2971	info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2972
2973	/*
2974	* Mark relevant sections as SHF_RO_AFTER_INIT so layout_sections() can
2975	* put them in the right place.
2976	* Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2977	*/
2978	module_mark_ro_after_init(info->hdr, info->sechdrs, info->secstrings);
2979
2980	/*
2981	* Determine total sizes, and put offsets in sh_entsize. For now
2982	* this is done generically; there doesn't appear to be any
2983	* special cases for the architectures.
2984	*/
2985	layout_sections(info->mod, info);
2986	layout_symtab(info->mod, info);
2987
2988	/* Allocate and move to the final place */
2989	err = move_module(info->mod, info);
2990	if (err)
2991	return ERR_PTR(err);
2992
2993	/* Module has been copied to its final place now: return it. */
2994	mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2995	kmemleak_load_module(mod, info);
2996	codetag_module_replaced(info->mod, mod);
2997
2998	return mod;
2999	}
3000
3001	/* mod is no longer valid after this! */
3002	static void module_deallocate(struct module mod, struct load_info info)
3003	{
3004	percpu_modfree(mod);
3005	module_arch_freeing_init(mod);
3006	codetag_free_module_sections(mod);
3007
3008	free_mod_mem(mod);
3009	}
3010
3011	int __weak module_finalize(const Elf_Ehdr *hdr,
3012	const Elf_Shdr *sechdrs,
3013	struct module *me)
3014	{
3015	return 0;
3016	}
3017
3018	static int post_relocation(struct module mod, const struct load_info info)
3019	{
3020	/* Sort exception table now relocations are done. */
3021	sort_extable(mod->extable, mod->extable + mod->num_exentries);
3022
3023	/* Copy relocated percpu area over. */
3024	percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
3025	info->sechdrs[info->index.pcpu].sh_size);
3026
3027	/* Setup kallsyms-specific fields. */
3028	add_kallsyms(mod, info);
3029
3030	/* Arch-specific module finalizing. */
3031	return module_finalize(info->hdr, info->sechdrs, mod);
3032	}
3033
3034	/* Call module constructors. */
3035	static void do_mod_ctors(struct module *mod)
3036	{
3037	#ifdef CONFIG_CONSTRUCTORS
3038	unsigned long i;
3039
3040	for (i = 0; i < mod->num_ctors; i++)
3041	mod->ctors[i]();
3042	#endif
3043	}
3044
3045	/* For freeing module_init on success, in case kallsyms traversing */
3046	struct mod_initfree {
3047	struct llist_node node;
3048	void *init_text;
3049	void *init_data;
3050	void *init_rodata;
3051	};
3052
3053	static void do_free_init(struct work_struct *w)
3054	{
3055	struct llist_node pos, n, *list;
3056	struct mod_initfree *initfree;
3057
3058	list = llist_del_all(&init_free_list);
3059
3060	synchronize_rcu();
3061
3062	llist_for_each_safe(pos, n, list) {
3063	initfree = container_of(pos, struct mod_initfree, node);
3064	execmem_free(initfree->init_text);
3065	execmem_free(initfree->init_data);
3066	execmem_free(initfree->init_rodata);
3067	kfree(initfree);
3068	}
3069	}
3070
3071	void flush_module_init_free_work(void)
3072	{
3073	flush_work(&init_free_wq);
3074	}
3075
3076	#undef MODULE_PARAM_PREFIX
3077	#define MODULE_PARAM_PREFIX "module."
3078	/* Default value for module->async_probe_requested */
3079	static bool async_probe;
3080	module_param(async_probe, bool, 0644);
3081
3082	/*
3083	* This is where the real work happens.
3084	*
3085	* Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
3086	* helper command 'lx-symbols'.
3087	*/
3088	static noinline int do_init_module(struct module *mod)
3089	{
3090	int ret = 0;
3091	struct mod_initfree *freeinit;
3092	#if defined(CONFIG_MODULE_STATS)
3093	unsigned int text_size = 0, total_size = 0;
3094
3095	for_each_mod_mem_type(type) {
3096	const struct module_memory *mod_mem = &mod->mem[type];
3097	if (mod_mem->size) {
3098	total_size += mod_mem->size;
3099	if (type == MOD_TEXT \|\| type == MOD_INIT_TEXT)
3100	text_size += mod_mem->size;
3101	}
3102	}
3103	#endif
3104
3105	freeinit = kmalloc_obj(*freeinit);
3106	if (!freeinit) {
3107	ret = -ENOMEM;
3108	goto fail;
3109	}
3110	freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
3111	freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
3112	freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
3113
3114	do_mod_ctors(mod);
3115	/* Start the module */
3116	if (mod->init != NULL)
3117	ret = do_one_initcall(mod->init);
3118	if (ret < 0) {
3119	/*
3120	* -EEXIST is reserved by [f]init_module() to signal to userspace that
3121	* a module with this name is already loaded. Use something else if the
3122	* module itself is returning that.
3123	*/
3124	if (ret == -EEXIST)
3125	ret = -EBUSY;
3126
3127	goto fail_free_freeinit;
3128	}
3129	if (ret > 0)
3130	pr_warn("%s: init suspiciously returned %d, it should follow 0/-E convention\n",
3131	mod->name, ret);
3132
3133	/* Now it's a first class citizen! */
3134	mod->state = MODULE_STATE_LIVE;
3135	blocking_notifier_call_chain(&module_notify_list,
3136	MODULE_STATE_LIVE, mod);
3137
3138	/* Delay uevent until module has finished its init routine */
3139	kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
3140
3141	/*
3142	* We need to finish all async code before the module init sequence
3143	* is done. This has potential to deadlock if synchronous module
3144	* loading is requested from async (which is not allowed!).
3145	*
3146	* See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
3147	* request_module() from async workers") for more details.
3148	*/
3149	if (!mod->async_probe_requested)
3150	async_synchronize_full();
3151
3152	ftrace_free_mem(mod, mod->mem[MOD_INIT_TEXT].base,
3153	mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
3154	mutex_lock(&module_mutex);
3155	/* Drop initial reference. */
3156	module_put(mod);
3157	trim_init_extable(mod);
3158	#ifdef CONFIG_KALLSYMS
3159	/* Switch to core kallsyms now init is done: kallsyms may be walking! */
3160	rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
3161	#endif
3162	ret = module_enable_rodata_ro_after_init(mod);
3163	if (ret)
3164	pr_warn("%s: module_enable_rodata_ro_after_init() returned %d, "
3165	"ro_after_init data might still be writable\n",
3166	mod->name, ret);
3167
3168	mod_tree_remove_init(mod);
3169	module_arch_freeing_init(mod);
3170	for_class_mod_mem_type(type, init) {
3171	mod->mem[type].base = NULL;
3172	mod->mem[type].size = 0;
3173	}
3174
3175	#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
3176	/* .BTF is not SHF_ALLOC and will get removed, so sanitize pointers */
3177	mod->btf_data = NULL;
3178	mod->btf_base_data = NULL;
3179	#endif
3180	/*
3181	* We want to free module_init, but be aware that kallsyms may be
3182	* walking this within an RCU read section. In all the failure paths, we
3183	* call synchronize_rcu(), but we don't want to slow down the success
3184	* path. execmem_free() cannot be called in an interrupt, so do the
3185	* work and call synchronize_rcu() in a work queue.
3186	*
3187	* Note that execmem_alloc() on most architectures creates W+X page
3188	* mappings which won't be cleaned up until do_free_init() runs. Any
3189	* code such as mark_rodata_ro() which depends on those mappings to
3190	* be cleaned up needs to sync with the queued work by invoking
3191	* flush_module_init_free_work().
3192	*/
3193	if (llist_add(&freeinit->node, &init_free_list))
3194	schedule_work(&init_free_wq);
3195
3196	mutex_unlock(&module_mutex);
3197	wake_up_all(&module_wq);
3198
3199	mod_stat_add_long(text_size, &total_text_size);
3200	mod_stat_add_long(total_size, &total_mod_size);
3201
3202	mod_stat_inc(&modcount);
3203
3204	return 0;
3205
3206	fail_free_freeinit:
3207	kfree(freeinit);
3208	fail:
3209	/* Try to protect us from buggy refcounters. */
3210	mod->state = MODULE_STATE_GOING;
3211	synchronize_rcu();
3212	module_put(mod);
3213	blocking_notifier_call_chain(&module_notify_list,
3214	MODULE_STATE_GOING, mod);
3215	klp_module_going(mod);
3216	ftrace_release_mod(mod);
3217	free_module(mod);
3218	wake_up_all(&module_wq);
3219
3220	return ret;
3221	}
3222
3223	static int may_init_module(void)
3224	{
3225	if (!capable(CAP_SYS_MODULE) \|\| modules_disabled)
3226	return -EPERM;
3227
3228	return 0;
3229	}
3230
3231	/* Is this module of this name done loading? No locks held. */
3232	static bool finished_loading(const char *name)
3233	{
3234	struct module *mod;
3235	bool ret;
3236
3237	/*
3238	* The module_mutex should not be a heavily contended lock;
3239	* if we get the occasional sleep here, we'll go an extra iteration
3240	* in the wait_event_interruptible(), which is harmless.
3241	*/
3242	sched_annotate_sleep();
3243	mutex_lock(&module_mutex);
3244	mod = find_module_all(name, strlen(name), true);
3245	ret = !mod \|\| mod->state == MODULE_STATE_LIVE
3246	\|\| mod->state == MODULE_STATE_GOING;
3247	mutex_unlock(&module_mutex);
3248
3249	return ret;
3250	}
3251
3252	/* Must be called with module_mutex held */
3253	static int module_patient_check_exists(const char *name,
3254	enum fail_dup_mod_reason reason)
3255	{
3256	struct module *old;
3257	int err = 0;
3258
3259	old = find_module_all(name, strlen(name), true);
3260	if (old == NULL)
3261	return 0;
3262
3263	if (old->state == MODULE_STATE_COMING \|\|
3264	old->state == MODULE_STATE_UNFORMED) {
3265	/* Wait in case it fails to load. */
3266	mutex_unlock(&module_mutex);
3267	err = wait_event_interruptible(module_wq,
3268	finished_loading(name));
3269	mutex_lock(&module_mutex);
3270	if (err)
3271	return err;
3272
3273	/* The module might have gone in the meantime. */
3274	old = find_module_all(name, strlen(name), true);
3275	}
3276
3277	if (try_add_failed_module(name, reason))
3278	pr_warn("Could not add fail-tracking for module: %s\n", name);
3279
3280	/*
3281	* We are here only when the same module was being loaded. Do
3282	* not try to load it again right now. It prevents long delays
3283	* caused by serialized module load failures. It might happen
3284	* when more devices of the same type trigger load of
3285	* a particular module.
3286	*/
3287	if (old && old->state == MODULE_STATE_LIVE)
3288	return -EEXIST;
3289	return -EBUSY;
3290	}
3291
3292	/*
3293	* We try to place it in the list now to make sure it's unique before
3294	* we dedicate too many resources. In particular, temporary percpu
3295	* memory exhaustion.
3296	*/
3297	static int add_unformed_module(struct module *mod)
3298	{
3299	int err;
3300
3301	mod->state = MODULE_STATE_UNFORMED;
3302
3303	mutex_lock(&module_mutex);
3304	err = module_patient_check_exists(mod->name, FAIL_DUP_MOD_LOAD);
3305	if (err)
3306	goto out;
3307
3308	mod_update_bounds(mod);
3309	list_add_rcu(&mod->list, &modules);
3310	mod_tree_insert(mod);
3311	err = 0;
3312
3313	out:
3314	mutex_unlock(&module_mutex);
3315	return err;
3316	}
3317
3318	static int complete_formation(struct module mod, struct load_info info)
3319	{
3320	int err;
3321
3322	mutex_lock(&module_mutex);
3323
3324	/* Find duplicate symbols (must be called under lock). */
3325	err = verify_exported_symbols(mod);
3326	if (err < 0)
3327	goto out;
3328
3329	/* These rely on module_mutex for list integrity. */
3330	module_bug_finalize(info->hdr, info->sechdrs, mod);
3331	module_cfi_finalize(info->hdr, info->sechdrs, mod);
3332
3333	err = module_enable_rodata_ro(mod);
3334	if (err)
3335	goto out_strict_rwx;
3336	err = module_enable_data_nx(mod);
3337	if (err)
3338	goto out_strict_rwx;
3339	err = module_enable_text_rox(mod);
3340	if (err)
3341	goto out_strict_rwx;
3342
3343	/*
3344	* Mark state as coming so strong_try_module_get() ignores us,
3345	* but kallsyms etc. can see us.
3346	*/
3347	mod->state = MODULE_STATE_COMING;
3348	mutex_unlock(&module_mutex);
3349
3350	return 0;
3351
3352	out_strict_rwx:
3353	module_bug_cleanup(mod);
3354	out:
3355	mutex_unlock(&module_mutex);
3356	return err;
3357	}
3358
3359	static int prepare_coming_module(struct module *mod)
3360	{
3361	int err;
3362
3363	ftrace_module_enable(mod);
3364	err = klp_module_coming(mod);
3365	if (err)
3366	return err;
3367
3368	err = blocking_notifier_call_chain_robust(&module_notify_list,
3369	MODULE_STATE_COMING, MODULE_STATE_GOING, mod);
3370	err = notifier_to_errno(err);
3371	if (err)
3372	klp_module_going(mod);
3373
3374	return err;
3375	}
3376
3377	static int unknown_module_param_cb(char param, char val, const char *modname,
3378	void *arg)
3379	{
3380	struct module *mod = arg;
3381	int ret;
3382
3383	if (strcmp(param, "async_probe") == 0) {
3384	if (kstrtobool(val, &mod->async_probe_requested))
3385	mod->async_probe_requested = true;
3386	return 0;
3387	}
3388
3389	/* Check for magic 'dyndbg' arg */
3390	ret = ddebug_dyndbg_module_param_cb(param, val, modname);
3391	if (ret != 0)
3392	pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
3393	return 0;
3394	}
3395
3396	/* Module within temporary copy, this doesn't do any allocation */
3397	static int early_mod_check(struct load_info *info, int flags)
3398	{
3399	int err;
3400
3401	/*
3402	* Now that we know we have the correct module name, check
3403	* if it's blacklisted.
3404	*/
3405	if (blacklisted(info->name)) {
3406	pr_err("Module %s is blacklisted\n", info->name);
3407	return -EPERM;
3408	}
3409
3410	err = rewrite_section_headers(info, flags);
3411	if (err)
3412	return err;
3413
3414	/* Check module struct version now, before we try to use module. */
3415	if (!check_modstruct_version(info, info->mod))
3416	return -ENOEXEC;
3417
3418	err = check_modinfo(info->mod, info, flags);
3419	if (err)
3420	return err;
3421
3422	mutex_lock(&module_mutex);
3423	err = module_patient_check_exists(info->mod->name, FAIL_DUP_MOD_BECOMING);
3424	mutex_unlock(&module_mutex);
3425
3426	return err;
3427	}
3428
3429	/*
3430	* Allocate and load the module: note that size of section 0 is always
3431	* zero, and we rely on this for optional sections.
3432	*/
3433	static int load_module(struct load_info info, const char __user uargs,
3434	int flags)
3435	{
3436	struct module *mod;
3437	bool module_allocated = false;
3438	long err = 0;
3439	char args = NULL, after_dashes;
3440
3441	/*
3442	* Do the signature check (if any) first. All that
3443	* the signature check needs is info->len, it does
3444	* not need any of the section info. That can be
3445	* set up later. This will minimize the chances
3446	* of a corrupt module causing problems before
3447	* we even get to the signature check.
3448	*
3449	* The check will also adjust info->len by stripping
3450	* off the sig length at the end of the module, making
3451	* checks against info->len more correct.
3452	*/
3453	err = module_sig_check(info, flags);
3454	if (err)
3455	goto free_copy;
3456
3457	/*
3458	* Do basic sanity checks against the ELF header and
3459	* sections. Cache useful sections and set the
3460	* info->mod to the userspace passed struct module.
3461	*/
3462	err = elf_validity_cache_copy(info, flags);
3463	if (err)
3464	goto free_copy;
3465
3466	err = early_mod_check(info, flags);
3467	if (err)
3468	goto free_copy;
3469
3470	/* Figure out module layout, and allocate all the memory. */
3471	mod = layout_and_allocate(info, flags);
3472	if (IS_ERR(mod)) {
3473	err = PTR_ERR(mod);
3474	goto free_copy;
3475	}
3476
3477	module_allocated = true;
3478
3479	audit_log_kern_module(info->name);
3480
3481	/* Reserve our place in the list. */
3482	err = add_unformed_module(mod);
3483	if (err)
3484	goto free_module;
3485
3486	/*
3487	* We are tainting your kernel if your module gets into
3488	* the modules linked list somehow.
3489	*/
3490	module_augment_kernel_taints(mod, info);
3491
3492	/* To avoid stressing percpu allocator, do this once we're unique. */
3493	err = percpu_modalloc(mod, info);
3494	if (err)
3495	goto unlink_mod;
3496
3497	/* Now module is in final location, initialize linked lists, etc. */
3498	err = module_unload_init(mod);
3499	if (err)
3500	goto unlink_mod;
3501
3502	init_param_lock(mod);
3503
3504	/*
3505	* Now we've got everything in the final locations, we can
3506	* find optional sections.
3507	*/
3508	err = find_module_sections(mod, info);
3509	if (err)
3510	goto free_unload;
3511
3512	err = check_export_symbol_sections(mod);
3513	if (err)
3514	goto free_unload;
3515
3516	/* Set up MODINFO_ATTR fields */
3517	err = setup_modinfo(mod, info);
3518	if (err)
3519	goto free_modinfo;
3520
3521	/* Fix up syms, so that st_value is a pointer to location. */
3522	err = simplify_symbols(mod, info);
3523	if (err < 0)
3524	goto free_modinfo;
3525
3526	err = apply_relocations(mod, info);
3527	if (err < 0)
3528	goto free_modinfo;
3529
3530	err = post_relocation(mod, info);
3531	if (err < 0)
3532	goto free_modinfo;
3533
3534	flush_module_icache(mod);
3535
3536	/* Now copy in args */
3537	args = strndup_user(uargs, ~0UL >> 1);
3538	if (IS_ERR(args)) {
3539	err = PTR_ERR(args);
3540	goto free_arch_cleanup;
3541	}
3542
3543	init_build_id(mod, info);
3544
3545	/* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
3546	ftrace_module_init(mod);
3547
3548	/* Finally it's fully formed, ready to start executing. */
3549	err = complete_formation(mod, info);
3550	if (err)
3551	goto ddebug_cleanup;
3552
3553	err = prepare_coming_module(mod);
3554	if (err)
3555	goto bug_cleanup;
3556
3557	mod->async_probe_requested = async_probe;
3558
3559	/* Module is ready to execute: parsing args may do that. */
3560	after_dashes = parse_args(mod->name, args, mod->kp, mod->num_kp,
3561	-32768, 32767, mod,
3562	unknown_module_param_cb);
3563	if (IS_ERR(after_dashes)) {
3564	err = PTR_ERR(after_dashes);
3565	goto coming_cleanup;
3566	} else if (after_dashes) {
3567	pr_warn("%s: parameters '%s' after `--' ignored\n",
3568	mod->name, after_dashes);
3569	}
3570	kfree(args);
3571	args = NULL;
3572
3573	/* Link in to sysfs. */
3574	err = mod_sysfs_setup(mod, info, mod->kp, mod->num_kp);
3575	if (err < 0)
3576	goto coming_cleanup;
3577
3578	if (is_livepatch_module(mod)) {
3579	err = copy_module_elf(mod, info);
3580	if (err < 0)
3581	goto sysfs_cleanup;
3582	}
3583
3584	err = codetag_load_module(mod);
3585	if (err)
3586	goto livepatch_cleanup;
3587
3588	/* Get rid of temporary copy. */
3589	free_copy(info, flags);
3590
3591	/* Done! */
3592	trace_module_load(mod);
3593
3594	return do_init_module(mod);
3595
3596	livepatch_cleanup:
3597	if (is_livepatch_module(mod))
3598	free_module_elf(mod);
3599	sysfs_cleanup:
3600	mod_sysfs_teardown(mod);
3601	coming_cleanup:
3602	mod->state = MODULE_STATE_GOING;
3603	module_destroy_params(mod->kp, mod->num_kp);
3604	blocking_notifier_call_chain(&module_notify_list,
3605	MODULE_STATE_GOING, mod);
3606	klp_module_going(mod);
3607	bug_cleanup:
3608	mod->state = MODULE_STATE_GOING;
3609	/* module_bug_cleanup needs module_mutex protection */
3610	mutex_lock(&module_mutex);
3611	module_bug_cleanup(mod);
3612	mutex_unlock(&module_mutex);
3613
3614	ddebug_cleanup:
3615	ftrace_release_mod(mod);
3616	synchronize_rcu();
3617	kfree(args);
3618	free_arch_cleanup:
3619	module_arch_cleanup(mod);
3620	free_modinfo:
3621	free_modinfo(mod);
3622	free_unload:
3623	module_unload_free(mod);
3624	unlink_mod:
3625	mutex_lock(&module_mutex);
3626	/* Unlink carefully: kallsyms could be walking list. */
3627	list_del_rcu(&mod->list);
3628	mod_tree_remove(mod);
3629	wake_up_all(&module_wq);
3630	/* Wait for RCU-sched synchronizing before releasing mod->list. */
3631	synchronize_rcu();
3632	mutex_unlock(&module_mutex);
3633	free_module:
3634	mod_stat_bump_invalid(info, flags);
3635	module_memory_restore_rox(mod);
3636	module_deallocate(mod, info);
3637	free_copy:
3638	/*
3639	* The info->len is always set. We distinguish between
3640	* failures once the proper module was allocated and
3641	* before that.
3642	*/
3643	if (!module_allocated) {
3644	audit_log_kern_module(info->name ? info->name : "?");
3645	mod_stat_bump_becoming(info, flags);
3646	}
3647	free_copy(info, flags);
3648	return err;
3649	}
3650
3651	SYSCALL_DEFINE3(init_module, void __user *, umod,
3652	unsigned long, len, const char __user *, uargs)
3653	{
3654	int err;
3655	struct load_info info = { };
3656
3657	err = may_init_module();
3658	if (err)
3659	return err;
3660
3661	pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3662	umod, len, uargs);
3663
3664	err = copy_module_from_user(umod, len, &info);
3665	if (err) {
3666	mod_stat_inc(&failed_kreads);
3667	mod_stat_add_long(len, &invalid_kread_bytes);
3668	return err;
3669	}
3670
3671	return load_module(&info, uargs, 0);
3672	}
3673
3674	struct idempotent {
3675	const void *cookie;
3676	struct hlist_node entry;
3677	struct completion complete;
3678	int ret;
3679	};
3680
3681	#define IDEM_HASH_BITS 8
3682	static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3683	static DEFINE_SPINLOCK(idem_lock);
3684
3685	static bool idempotent(struct idempotent u, const void cookie)
3686	{
3687	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3688	struct hlist_head *head = idem_hash + hash;
3689	struct idempotent *existing;
3690	bool first;
3691
3692	u->ret = -EINTR;
3693	u->cookie = cookie;
3694	init_completion(&u->complete);
3695
3696	spin_lock(&idem_lock);
3697	first = true;
3698	hlist_for_each_entry(existing, head, entry) {
3699	if (existing->cookie != cookie)
3700	continue;
3701	first = false;
3702	break;
3703	}
3704	hlist_add_head(&u->entry, idem_hash + hash);
3705	spin_unlock(&idem_lock);
3706
3707	return !first;
3708	}
3709
3710	/*
3711	* We were the first one with 'cookie' on the list, and we ended
3712	* up completing the operation. We now need to walk the list,
3713	* remove everybody - which includes ourselves - fill in the return
3714	* value, and then complete the operation.
3715	*/
3716	static int idempotent_complete(struct idempotent *u, int ret)
3717	{
3718	const void *cookie = u->cookie;
3719	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3720	struct hlist_head *head = idem_hash + hash;
3721	struct hlist_node *next;
3722	struct idempotent *pos;
3723
3724	spin_lock(&idem_lock);
3725	hlist_for_each_entry_safe(pos, next, head, entry) {
3726	if (pos->cookie != cookie)
3727	continue;
3728	hlist_del_init(&pos->entry);
3729	pos->ret = ret;
3730	complete(&pos->complete);
3731	}
3732	spin_unlock(&idem_lock);
3733	return ret;
3734	}
3735
3736	/*
3737	* Wait for the idempotent worker.
3738	*
3739	* If we get interrupted, we need to remove ourselves from the
3740	* the idempotent list, and the completion may still come in.
3741	*
3742	* The 'idem_lock' protects against the race, and 'idem.ret' was
3743	* initialized to -EINTR and is thus always the right return
3744	* value even if the idempotent work then completes between
3745	* the wait_for_completion and the cleanup.
3746	*/
3747	static int idempotent_wait_for_completion(struct idempotent *u)
3748	{
3749	if (wait_for_completion_interruptible(&u->complete)) {
3750	spin_lock(&idem_lock);
3751	if (!hlist_unhashed(&u->entry))
3752	hlist_del(&u->entry);
3753	spin_unlock(&idem_lock);
3754	}
3755	return u->ret;
3756	}
3757
3758	static int init_module_from_file(struct file f, const char __user uargs, int flags)
3759	{
3760	bool compressed = !!(flags & MODULE_INIT_COMPRESSED_FILE);
3761	struct load_info info = { };
3762	void *buf = NULL;
3763	int len;
3764	int err;
3765
3766	len = kernel_read_file(f, 0, &buf, INT_MAX, NULL,
3767	compressed ? READING_MODULE_COMPRESSED :
3768	READING_MODULE);
3769	if (len < 0) {
3770	mod_stat_inc(&failed_kreads);
3771	return len;
3772	}
3773
3774	if (compressed) {
3775	err = module_decompress(&info, buf, len);
3776	vfree(buf); /* compressed data is no longer needed */
3777	if (err) {
3778	mod_stat_inc(&failed_decompress);
3779	mod_stat_add_long(len, &invalid_decompress_bytes);
3780	return err;
3781	}
3782	err = security_kernel_post_read_file(f, (char *)info.hdr, info.len,
3783	READING_MODULE);
3784	if (err) {
3785	mod_stat_inc(&failed_kreads);
3786	free_copy(&info, flags);
3787	return err;
3788	}
3789	} else {
3790	info.hdr = buf;
3791	info.len = len;
3792	}
3793
3794	return load_module(&info, uargs, flags);
3795	}
3796
3797	static int idempotent_init_module(struct file f, const char __user uargs, int flags)
3798	{
3799	struct idempotent idem;
3800
3801	if (!(f->f_mode & FMODE_READ))
3802	return -EBADF;
3803
3804	/* Are we the winners of the race and get to do this? */
3805	if (!idempotent(&idem, file_inode(f))) {
3806	int ret = init_module_from_file(f, uargs, flags);
3807	return idempotent_complete(&idem, ret);
3808	}
3809
3810	/*
3811	* Somebody else won the race and is loading the module.
3812	*/
3813	return idempotent_wait_for_completion(&idem);
3814	}
3815
3816	SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3817	{
3818	int err = may_init_module();
3819	if (err)
3820	return err;
3821
3822	pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3823
3824	if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3825	\|MODULE_INIT_IGNORE_VERMAGIC
3826	\|MODULE_INIT_COMPRESSED_FILE))
3827	return -EINVAL;
3828
3829	CLASS(fd, f)(fd);
3830	if (fd_empty(f))
3831	return -EBADF;
3832	return idempotent_init_module(fd_file(f), uargs, flags);
3833	}
3834
3835	/* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
3836	char module_flags(struct module mod, char *buf, bool show_state)
3837	{
3838	int bx = 0;
3839
3840	BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3841	if (!mod->taints && !show_state)
3842	goto out;
3843	if (mod->taints \|\|
3844	mod->state == MODULE_STATE_GOING \|\|
3845	mod->state == MODULE_STATE_COMING) {
3846	buf[bx++] = '(';
3847	bx += module_flags_taint(mod->taints, buf + bx);
3848	/* Show a - for module-is-being-unloaded */
3849	if (mod->state == MODULE_STATE_GOING && show_state)
3850	buf[bx++] = '-';
3851	/* Show a + for module-is-being-loaded */
3852	if (mod->state == MODULE_STATE_COMING && show_state)
3853	buf[bx++] = '+';
3854	buf[bx++] = ')';
3855	}
3856	out:
3857	buf[bx] = '\0';
3858
3859	return buf;
3860	}
3861
3862	/* Given an address, look for it in the module exception tables. */
3863	const struct exception_table_entry *search_module_extables(unsigned long addr)
3864	{
3865	struct module *mod;
3866
3867	guard(rcu)();
3868	mod = __module_address(addr);
3869	if (!mod)
3870	return NULL;
3871
3872	if (!mod->num_exentries)
3873	return NULL;
3874	/*
3875	* The address passed here belongs to a module that is currently
3876	* invoked (we are running inside it). Therefore its module::refcnt
3877	* needs already be >0 to ensure that it is not removed at this stage.
3878	* All other user need to invoke this function within a RCU read
3879	* section.
3880	*/
3881	return search_extable(mod->extable, mod->num_exentries, addr);
3882	}
3883
3884	/**
3885	* is_module_address() - is this address inside a module?
3886	* @addr: the address to check.
3887	*
3888	* See is_module_text_address() if you simply want to see if the address
3889	* is code (not data).
3890	*/
3891	bool is_module_address(unsigned long addr)
3892	{
3893	guard(rcu)();
3894	return __module_address(addr) != NULL;
3895	}
3896
3897	/**
3898	* __module_address() - get the module which contains an address.
3899	* @addr: the address.
3900	*
3901	* Must be called within RCU read section or module mutex held so that
3902	* module doesn't get freed during this.
3903	*/
3904	struct module *__module_address(unsigned long addr)
3905	{
3906	struct module *mod;
3907
3908	if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3909	goto lookup;
3910
3911	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3912	if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3913	goto lookup;
3914	#endif
3915
3916	return NULL;
3917
3918	lookup:
3919	mod = mod_find(addr, &mod_tree);
3920	if (mod) {
3921	BUG_ON(!within_module(addr, mod));
3922	if (mod->state == MODULE_STATE_UNFORMED)
3923	mod = NULL;
3924	}
3925	return mod;
3926	}
3927
3928	/**
3929	* is_module_text_address() - is this address inside module code?
3930	* @addr: the address to check.
3931	*
3932	* See is_module_address() if you simply want to see if the address is
3933	* anywhere in a module. See kernel_text_address() for testing if an
3934	* address corresponds to kernel or module code.
3935	*/
3936	bool is_module_text_address(unsigned long addr)
3937	{
3938	guard(rcu)();
3939	return __module_text_address(addr) != NULL;
3940	}
3941
3942	void module_for_each_mod(int(func)(struct module mod, void data), void data)
3943	{
3944	struct module *mod;
3945
3946	guard(rcu)();
3947	list_for_each_entry_rcu(mod, &modules, list) {
3948	if (mod->state == MODULE_STATE_UNFORMED)
3949	continue;
3950	if (func(mod, data))
3951	break;
3952	}
3953	}
3954
3955	/**
3956	* __module_text_address() - get the module whose code contains an address.
3957	* @addr: the address.
3958	*
3959	* Must be called within RCU read section or module mutex held so that
3960	* module doesn't get freed during this.
3961	*/
3962	struct module *__module_text_address(unsigned long addr)
3963	{
3964	struct module *mod = __module_address(addr);
3965	if (mod) {
3966	/* Make sure it's within the text section. */
3967	if (!within_module_mem_type(addr, mod, MOD_TEXT) &&
3968	!within_module_mem_type(addr, mod, MOD_INIT_TEXT))
3969	mod = NULL;
3970	}
3971	return mod;
3972	}
3973
3974	/* Don't grab lock, we're oopsing. */
3975	void print_modules(void)
3976	{
3977	struct module *mod;
3978	char buf[MODULE_FLAGS_BUF_SIZE];
3979
3980	printk(KERN_DEFAULT "Modules linked in:");
3981	/* Most callers should already have preempt disabled, but make sure */
3982	guard(rcu)();
3983	list_for_each_entry_rcu(mod, &modules, list) {
3984	if (mod->state == MODULE_STATE_UNFORMED)
3985	continue;
3986	pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3987	}
3988
3989	print_unloaded_tainted_modules();
3990	if (last_unloaded_module.name[0])
3991	pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3992	last_unloaded_module.taints);
3993	pr_cont("\n");
3994	}
3995
3996	#ifdef CONFIG_MODULE_DEBUGFS
3997	struct dentry *mod_debugfs_root;
3998
3999	static int module_debugfs_init(void)
4000	{
4001	mod_debugfs_root = debugfs_create_dir("modules", NULL);
4002	return 0;
4003	}
4004	module_init(module_debugfs_init);
4005	#endif