1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* Copyright (C) 2002 Richard Henderson
4	* Copyright (C) 2001 Rusty Russell, 2002, 2010 Rusty Russell IBM.
5	* Copyright (C) 2023 Luis Chamberlain <mcgrof@kernel.org>
6	*/
7
8	#define INCLUDE_VERMAGIC
9
10	#include <linux/export.h>
11	#include <linux/extable.h>
12	#include <linux/moduleloader.h>
13	#include <linux/module_signature.h>
14	#include <linux/trace_events.h>
15	#include <linux/init.h>
16	#include <linux/kallsyms.h>
17	#include <linux/buildid.h>
18	#include <linux/fs.h>
19	#include <linux/kernel.h>
20	#include <linux/kernel_read_file.h>
21	#include <linux/kstrtox.h>
22	#include <linux/slab.h>
23	#include <linux/vmalloc.h>
24	#include <linux/elf.h>
25	#include <linux/seq_file.h>
26	#include <linux/syscalls.h>
27	#include <linux/fcntl.h>
28	#include <linux/rcupdate.h>
29	#include <linux/capability.h>
30	#include <linux/cpu.h>
31	#include <linux/moduleparam.h>
32	#include <linux/errno.h>
33	#include <linux/err.h>
34	#include <linux/vermagic.h>
35	#include <linux/notifier.h>
36	#include <linux/sched.h>
37	#include <linux/device.h>
38	#include <linux/string.h>
39	#include <linux/mutex.h>
40	#include <linux/rculist.h>
41	#include <linux/uaccess.h>
42	#include <asm/cacheflush.h>
43	#include <linux/set_memory.h>
44	#include <asm/mmu_context.h>
45	#include <linux/license.h>
46	#include <asm/sections.h>
47	#include <linux/tracepoint.h>
48	#include <linux/ftrace.h>
49	#include <linux/livepatch.h>
50	#include <linux/async.h>
51	#include <linux/percpu.h>
52	#include <linux/kmemleak.h>
53	#include <linux/jump_label.h>
54	#include <linux/pfn.h>
55	#include <linux/bsearch.h>
56	#include <linux/dynamic_debug.h>
57	#include <linux/audit.h>
58	#include <linux/cfi.h>
59	#include <linux/codetag.h>
60	#include <linux/debugfs.h>
61	#include <linux/execmem.h>
62	#include <uapi/linux/module.h>
63	#include "internal.h"
64
65	#define CREATE_TRACE_POINTS
66	#include <trace/events/module.h>
67
68	/*
69	* Mutex protects:
70	* 1) List of modules (also safely readable within RCU read section),
71	* 2) module_use links,
72	* 3) mod_tree.addr_min/mod_tree.addr_max.
73	* (delete and add uses RCU list operations).
74	*/
75	DEFINE_MUTEX(module_mutex);
76	LIST_HEAD(modules);
77
78	/* Work queue for freeing init sections in success case */
79	static void do_free_init(struct work_struct *w);
80	static DECLARE_WORK(init_free_wq, do_free_init);
81	static LLIST_HEAD(init_free_list);
82
83	struct mod_tree_root mod_tree __cacheline_aligned = {
84	.addr_min = -1UL,
85	};
86
87	struct symsearch {
88	const struct kernel_symbol *start, *stop;
89	const u32 *crcs;
90	enum mod_license license;
91	};
92
93	/*
94	* Bounds of module memory, for speeding up __module_address.
95	* Protected by module_mutex.
96	*/
97	static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
98	unsigned int size, struct mod_tree_root *tree)
99	{
100	unsigned long min = (unsigned long)base;
101	unsigned long max = min + size;
102
103	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
104	if (mod_mem_type_is_core_data(type)) {
105	if (min < tree->data_addr_min)
106	tree->data_addr_min = min;
107	if (max > tree->data_addr_max)
108	tree->data_addr_max = max;
109	return;
110	}
111	#endif
112	if (min < tree->addr_min)
113	tree->addr_min = min;
114	if (max > tree->addr_max)
115	tree->addr_max = max;
116	}
117
118	static void mod_update_bounds(struct module *mod)
119	{
120	for_each_mod_mem_type(type) {
121	struct module_memory *mod_mem = &mod->mem[type];
122
123	if (mod_mem->size)
124	__mod_update_bounds(type, base: mod_mem->base, size: mod_mem->size, tree: &mod_tree);
125	}
126	}
127
128	/* Block module loading/unloading? */
129	static int modules_disabled;
130	core_param(nomodule, modules_disabled, bint, 0);
131
132	static const struct ctl_table module_sysctl_table[] = {
133	{
134	.procname = "modprobe",
135	.data = &modprobe_path,
136	.maxlen = KMOD_PATH_LEN,
137	.mode = 0644,
138	.proc_handler = proc_dostring,
139	},
140	{
141	.procname = "modules_disabled",
142	.data = &modules_disabled,
143	.maxlen = sizeof(int),
144	.mode = 0644,
145	/* only handle a transition from default "0" to "1" */
146	.proc_handler = proc_dointvec_minmax,
147	.extra1 = SYSCTL_ONE,
148	.extra2 = SYSCTL_ONE,
149	},
150	};
151
152	static int __init init_module_sysctl(void)
153	{
154	register_sysctl_init("kernel", module_sysctl_table);
155	return 0;
156	}
157
158	subsys_initcall(init_module_sysctl);
159
160	/* Waiting for a module to finish initializing? */
161	static DECLARE_WAIT_QUEUE_HEAD(module_wq);
162
163	static BLOCKING_NOTIFIER_HEAD(module_notify_list);
164
165	int register_module_notifier(struct notifier_block *nb)
166	{
167	return blocking_notifier_chain_register(nh: &module_notify_list, nb);
168	}
169	EXPORT_SYMBOL(register_module_notifier);
170
171	int unregister_module_notifier(struct notifier_block *nb)
172	{
173	return blocking_notifier_chain_unregister(nh: &module_notify_list, nb);
174	}
175	EXPORT_SYMBOL(unregister_module_notifier);
176
177	/*
178	* We require a truly strong try_module_get(): 0 means success.
179	* Otherwise an error is returned due to ongoing or failed
180	* initialization etc.
181	*/
182	static inline int strong_try_module_get(struct module *mod)
183	{
184	BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
185	if (mod && mod->state == MODULE_STATE_COMING)
186	return -EBUSY;
187	if (try_module_get(module: mod))
188	return 0;
189	else
190	return -ENOENT;
191	}
192
193	static inline void add_taint_module(struct module *mod, unsigned flag,
194	enum lockdep_ok lockdep_ok)
195	{
196	add_taint(flag, lockdep_ok);
197	set_bit(nr: flag, addr: &mod->taints);
198	}
199
200	/*
201	* Like strncmp(), except s/-/_/g as per scripts/Makefile.lib:name-fix-token rule.
202	*/
203	static int mod_strncmp(const char *str_a, const char *str_b, size_t n)
204	{
205	for (int i = 0; i < n; i++) {
206	char a = str_a[i];
207	char b = str_b[i];
208	int d;
209
210	if (a == '-') a = '_';
211	if (b == '-') b = '_';
212
213	d = a - b;
214	if (d)
215	return d;
216
217	if (!a)
218	break;
219	}
220
221	return 0;
222	}
223
224	/*
225	* A thread that wants to hold a reference to a module only while it
226	* is running can call this to safely exit.
227	*/
228	void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
229	{
230	module_put(module: mod);
231	kthread_exit(result: code);
232	}
233	EXPORT_SYMBOL(__module_put_and_kthread_exit);
234
235	/* Find a module section: 0 means not found. */
236	static unsigned int find_sec(const struct load_info *info, const char *name)
237	{
238	unsigned int i;
239
240	for (i = 1; i < info->hdr->e_shnum; i++) {
241	Elf_Shdr *shdr = &info->sechdrs[i];
242	/* Alloc bit cleared means "ignore it." */
243	if ((shdr->sh_flags & SHF_ALLOC)
244	&& strcmp(info->secstrings + shdr->sh_name, name) == 0)
245	return i;
246	}
247	return 0;
248	}
249
250	/**
251	* find_any_unique_sec() - Find a unique section index by name
252	* @info: Load info for the module to scan
253	* @name: Name of the section we're looking for
254	*
255	* Locates a unique section by name. Ignores SHF_ALLOC.
256	*
257	* Return: Section index if found uniquely, zero if absent, negative count
258	* of total instances if multiple were found.
259	*/
260	static int find_any_unique_sec(const struct load_info *info, const char *name)
261	{
262	unsigned int idx;
263	unsigned int count = 0;
264	int i;
265
266	for (i = 1; i < info->hdr->e_shnum; i++) {
267	if (strcmp(info->secstrings + info->sechdrs[i].sh_name,
268	name) == 0) {
269	count++;
270	idx = i;
271	}
272	}
273	if (count == 1) {
274	return idx;
275	} else if (count == 0) {
276	return 0;
277	} else {
278	return -count;
279	}
280	}
281
282	/* Find a module section, or NULL. */
283	static void *section_addr(const struct load_info *info, const char *name)
284	{
285	/* Section 0 has sh_addr 0. */
286	return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
287	}
288
289	/* Find a module section, or NULL. Fill in number of "objects" in section. */
290	static void *section_objs(const struct load_info *info,
291	const char *name,
292	size_t object_size,
293	unsigned int *num)
294	{
295	unsigned int sec = find_sec(info, name);
296
297	/* Section 0 has sh_addr 0 and sh_size 0. */
298	*num = info->sechdrs[sec].sh_size / object_size;
299	return (void *)info->sechdrs[sec].sh_addr;
300	}
301
302	/* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
303	static unsigned int find_any_sec(const struct load_info *info, const char *name)
304	{
305	unsigned int i;
306
307	for (i = 1; i < info->hdr->e_shnum; i++) {
308	Elf_Shdr *shdr = &info->sechdrs[i];
309	if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
310	return i;
311	}
312	return 0;
313	}
314
315	/*
316	* Find a module section, or NULL. Fill in number of "objects" in section.
317	* Ignores SHF_ALLOC flag.
318	*/
319	static __maybe_unused void *any_section_objs(const struct load_info *info,
320	const char *name,
321	size_t object_size,
322	unsigned int *num)
323	{
324	unsigned int sec = find_any_sec(info, name);
325
326	/* Section 0 has sh_addr 0 and sh_size 0. */
327	*num = info->sechdrs[sec].sh_size / object_size;
328	return (void *)info->sechdrs[sec].sh_addr;
329	}
330
331	#ifndef CONFIG_MODVERSIONS
332	#define symversion(base, idx) NULL
333	#else
334	#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
335	#endif
336
337	static const char *kernel_symbol_name(const struct kernel_symbol *sym)
338	{
339	#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
340	return offset_to_ptr(off: &sym->name_offset);
341	#else
342	return sym->name;
343	#endif
344	}
345
346	static const char *kernel_symbol_namespace(const struct kernel_symbol *sym)
347	{
348	#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
349	if (!sym->namespace_offset)
350	return NULL;
351	return offset_to_ptr(off: &sym->namespace_offset);
352	#else
353	return sym->namespace;
354	#endif
355	}
356
357	int cmp_name(const void *name, const void *sym)
358	{
359	return strcmp(name, kernel_symbol_name(sym));
360	}
361
362	static bool find_exported_symbol_in_section(const struct symsearch *syms,
363	struct module *owner,
364	struct find_symbol_arg *fsa)
365	{
366	struct kernel_symbol *sym;
367
368	if (!fsa->gplok && syms->license == GPL_ONLY)
369	return false;
370
371	sym = bsearch(key: fsa->name, base: syms->start, num: syms->stop - syms->start,
372	size: sizeof(struct kernel_symbol), cmp: cmp_name);
373	if (!sym)
374	return false;
375
376	fsa->owner = owner;
377	fsa->crc = symversion(syms->crcs, sym - syms->start);
378	fsa->sym = sym;
379	fsa->license = syms->license;
380
381	return true;
382	}
383
384	/*
385	* Find an exported symbol and return it, along with, (optional) crc and
386	* (optional) module which owns it. Needs RCU or module_mutex.
387	*/
388	bool find_symbol(struct find_symbol_arg *fsa)
389	{
390	static const struct symsearch arr[] = {
391	{ __start___ksymtab, __stop___ksymtab, __start___kcrctab,
392	NOT_GPL_ONLY },
393	{ __start___ksymtab_gpl, __stop___ksymtab_gpl,
394	__start___kcrctab_gpl,
395	GPL_ONLY },
396	};
397	struct module *mod;
398	unsigned int i;
399
400	for (i = 0; i < ARRAY_SIZE(arr); i++)
401	if (find_exported_symbol_in_section(syms: &arr[i], NULL, fsa))
402	return true;
403
404	list_for_each_entry_rcu(mod, &modules, list,
405	lockdep_is_held(&module_mutex)) {
406	struct symsearch arr[] = {
407	{ mod->syms, mod->syms + mod->num_syms, mod->crcs,
408	NOT_GPL_ONLY },
409	{ mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
410	mod->gpl_crcs,
411	GPL_ONLY },
412	};
413
414	if (mod->state == MODULE_STATE_UNFORMED)
415	continue;
416
417	for (i = 0; i < ARRAY_SIZE(arr); i++)
418	if (find_exported_symbol_in_section(syms: &arr[i], owner: mod, fsa))
419	return true;
420	}
421
422	pr_debug("Failed to find symbol %s\n", fsa->name);
423	return false;
424	}
425
426	/*
427	* Search for module by name: must hold module_mutex (or RCU for read-only
428	* access).
429	*/
430	struct module *find_module_all(const char *name, size_t len,
431	bool even_unformed)
432	{
433	struct module *mod;
434
435	list_for_each_entry_rcu(mod, &modules, list,
436	lockdep_is_held(&module_mutex)) {
437	if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
438	continue;
439	if (strlen(mod->name) == len && !memcmp(p: mod->name, q: name, size: len))
440	return mod;
441	}
442	return NULL;
443	}
444
445	struct module *find_module(const char *name)
446	{
447	return find_module_all(name, strlen(name), even_unformed: false);
448	}
449
450	#ifdef CONFIG_SMP
451
452	static inline void __percpu *mod_percpu(struct module *mod)
453	{
454	return mod->percpu;
455	}
456
457	static int percpu_modalloc(struct module *mod, struct load_info *info)
458	{
459	Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
460	unsigned long align = pcpusec->sh_addralign;
461
462	if (!pcpusec->sh_size)
463	return 0;
464
465	if (align > PAGE_SIZE) {
466	pr_warn("%s: per-cpu alignment %li > %li\n",
467	mod->name, align, PAGE_SIZE);
468	align = PAGE_SIZE;
469	}
470
471	mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
472	if (!mod->percpu) {
473	pr_warn("%s: Could not allocate %lu bytes percpu data\n",
474	mod->name, (unsigned long)pcpusec->sh_size);
475	return -ENOMEM;
476	}
477	mod->percpu_size = pcpusec->sh_size;
478	return 0;
479	}
480
481	static void percpu_modfree(struct module *mod)
482	{
483	free_percpu(pdata: mod->percpu);
484	}
485
486	static unsigned int find_pcpusec(struct load_info *info)
487	{
488	return find_sec(info, name: ".data..percpu");
489	}
490
491	static void percpu_modcopy(struct module *mod,
492	const void *from, unsigned long size)
493	{
494	int cpu;
495
496	for_each_possible_cpu(cpu)
497	memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
498	}
499
500	bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
501	{
502	struct module *mod;
503	unsigned int cpu;
504
505	guard(rcu)();
506	list_for_each_entry_rcu(mod, &modules, list) {
507	if (mod->state == MODULE_STATE_UNFORMED)
508	continue;
509	if (!mod->percpu_size)
510	continue;
511	for_each_possible_cpu(cpu) {
512	void *start = per_cpu_ptr(mod->percpu, cpu);
513	void *va = (void *)addr;
514
515	if (va >= start && va < start + mod->percpu_size) {
516	if (can_addr) {
517	*can_addr = (unsigned long) (va - start);
518	*can_addr += (unsigned long)
519	per_cpu_ptr(mod->percpu,
520	get_boot_cpu_id());
521	}
522	return true;
523	}
524	}
525	}
526	return false;
527	}
528
529	/**
530	* is_module_percpu_address() - test whether address is from module static percpu
531	* @addr: address to test
532	*
533	* Test whether @addr belongs to module static percpu area.
534	*
535	* Return: %true if @addr is from module static percpu area
536	*/
537	bool is_module_percpu_address(unsigned long addr)
538	{
539	return __is_module_percpu_address(addr, NULL);
540	}
541
542	#else /* ... !CONFIG_SMP */
543
544	static inline void __percpu *mod_percpu(struct module *mod)
545	{
546	return NULL;
547	}
548	static int percpu_modalloc(struct module *mod, struct load_info *info)
549	{
550	/* UP modules shouldn't have this section: ENOMEM isn't quite right */
551	if (info->sechdrs[info->index.pcpu].sh_size != 0)
552	return -ENOMEM;
553	return 0;
554	}
555	static inline void percpu_modfree(struct module *mod)
556	{
557	}
558	static unsigned int find_pcpusec(struct load_info *info)
559	{
560	return 0;
561	}
562	static inline void percpu_modcopy(struct module *mod,
563	const void *from, unsigned long size)
564	{
565	/* pcpusec should be 0, and size of that section should be 0. */
566	BUG_ON(size != 0);
567	}
568	bool is_module_percpu_address(unsigned long addr)
569	{
570	return false;
571	}
572
573	bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
574	{
575	return false;
576	}
577
578	#endif /* CONFIG_SMP */
579
580	#define MODINFO_ATTR(field) \
581	static void setup_modinfo_##field(struct module *mod, const char *s) \
582	{ \
583	mod->field = kstrdup(s, GFP_KERNEL); \
584	} \
585	static ssize_t show_modinfo_##field(const struct module_attribute *mattr, \
586	struct module_kobject *mk, char *buffer) \
587	{ \
588	return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field); \
589	} \
590	static int modinfo_##field##_exists(struct module *mod) \
591	{ \
592	return mod->field != NULL; \
593	} \
594	static void free_modinfo_##field(struct module *mod) \
595	{ \
596	kfree(mod->field); \
597	mod->field = NULL; \
598	} \
599	static const struct module_attribute modinfo_##field = { \
600	.attr = { .name = __stringify(field), .mode = 0444 }, \
601	.show = show_modinfo_##field, \
602	.setup = setup_modinfo_##field, \
603	.test = modinfo_##field##_exists, \
604	.free = free_modinfo_##field, \
605	};
606
607	MODINFO_ATTR(version);
608	MODINFO_ATTR(srcversion);
609
610	static struct {
611	char name[MODULE_NAME_LEN];
612	char taints[MODULE_FLAGS_BUF_SIZE];
613	} last_unloaded_module;
614
615	#ifdef CONFIG_MODULE_UNLOAD
616
617	EXPORT_TRACEPOINT_SYMBOL(module_get);
618
619	/* MODULE_REF_BASE is the base reference count by kmodule loader. */
620	#define MODULE_REF_BASE 1
621
622	/* Init the unload section of the module. */
623	static int module_unload_init(struct module *mod)
624	{
625	/*
626	* Initialize reference counter to MODULE_REF_BASE.
627	* refcnt == 0 means module is going.
628	*/
629	atomic_set(v: &mod->refcnt, MODULE_REF_BASE);
630
631	INIT_LIST_HEAD(list: &mod->source_list);
632	INIT_LIST_HEAD(list: &mod->target_list);
633
634	/* Hold reference count during initialization. */
635	atomic_inc(v: &mod->refcnt);
636
637	return 0;
638	}
639
640	/* Does a already use b? */
641	static int already_uses(struct module *a, struct module *b)
642	{
643	struct module_use *use;
644
645	list_for_each_entry(use, &b->source_list, source_list) {
646	if (use->source == a)
647	return 1;
648	}
649	pr_debug("%s does not use %s!\n", a->name, b->name);
650	return 0;
651	}
652
653	/*
654	* Module a uses b
655	* - we add 'a' as a "source", 'b' as a "target" of module use
656	* - the module_use is added to the list of 'b' sources (so
657	* 'b' can walk the list to see who sourced them), and of 'a'
658	* targets (so 'a' can see what modules it targets).
659	*/
660	static int add_module_usage(struct module *a, struct module *b)
661	{
662	struct module_use *use;
663
664	pr_debug("Allocating new usage for %s.\n", a->name);
665	use = kmalloc(sizeof(*use), GFP_ATOMIC);
666	if (!use)
667	return -ENOMEM;
668
669	use->source = a;
670	use->target = b;
671	list_add(new: &use->source_list, head: &b->source_list);
672	list_add(new: &use->target_list, head: &a->target_list);
673	return 0;
674	}
675
676	/* Module a uses b: caller needs module_mutex() */
677	static int ref_module(struct module *a, struct module *b)
678	{
679	int err;
680
681	if (b == NULL || already_uses(a, b))
682	return 0;
683
684	/* If module isn't available, we fail. */
685	err = strong_try_module_get(mod: b);
686	if (err)
687	return err;
688
689	err = add_module_usage(a, b);
690	if (err) {
691	module_put(module: b);
692	return err;
693	}
694	return 0;
695	}
696
697	/* Clear the unload stuff of the module. */
698	static void module_unload_free(struct module *mod)
699	{
700	struct module_use *use, *tmp;
701
702	mutex_lock(&module_mutex);
703	list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
704	struct module *i = use->target;
705	pr_debug("%s unusing %s\n", mod->name, i->name);
706	module_put(module: i);
707	list_del(entry: &use->source_list);
708	list_del(entry: &use->target_list);
709	kfree(objp: use);
710	}
711	mutex_unlock(lock: &module_mutex);
712	}
713
714	#ifdef CONFIG_MODULE_FORCE_UNLOAD
715	static inline int try_force_unload(unsigned int flags)
716	{
717	int ret = (flags & O_TRUNC);
718	if (ret)
719	add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
720	return ret;
721	}
722	#else
723	static inline int try_force_unload(unsigned int flags)
724	{
725	return 0;
726	}
727	#endif /* CONFIG_MODULE_FORCE_UNLOAD */
728
729	/* Try to release refcount of module, 0 means success. */
730	static int try_release_module_ref(struct module *mod)
731	{
732	int ret;
733
734	/* Try to decrement refcnt which we set at loading */
735	ret = atomic_sub_return(MODULE_REF_BASE, v: &mod->refcnt);
736	BUG_ON(ret < 0);
737	if (ret)
738	/* Someone can put this right now, recover with checking */
739	ret = atomic_add_unless(v: &mod->refcnt, MODULE_REF_BASE, u: 0);
740
741	return ret;
742	}
743
744	static int try_stop_module(struct module *mod, int flags, int *forced)
745	{
746	/* If it's not unused, quit unless we're forcing. */
747	if (try_release_module_ref(mod) != 0) {
748	*forced = try_force_unload(flags);
749	if (!(*forced))
750	return -EWOULDBLOCK;
751	}
752
753	/* Mark it as dying. */
754	mod->state = MODULE_STATE_GOING;
755
756	return 0;
757	}
758
759	/**
760	* module_refcount() - return the refcount or -1 if unloading
761	* @mod: the module we're checking
762	*
763	* Return:
764	* -1 if the module is in the process of unloading
765	* otherwise the number of references in the kernel to the module
766	*/
767	int module_refcount(struct module *mod)
768	{
769	return atomic_read(v: &mod->refcnt) - MODULE_REF_BASE;
770	}
771	EXPORT_SYMBOL(module_refcount);
772
773	/* This exists whether we can unload or not */
774	static void free_module(struct module *mod);
775
776	SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
777	unsigned int, flags)
778	{
779	struct module *mod;
780	char name[MODULE_NAME_LEN];
781	char buf[MODULE_FLAGS_BUF_SIZE];
782	int ret, len, forced = 0;
783
784	if (!capable(CAP_SYS_MODULE) || modules_disabled)
785	return -EPERM;
786
787	len = strncpy_from_user(dst: name, src: name_user, MODULE_NAME_LEN);
788	if (len == 0 || len == MODULE_NAME_LEN)
789	return -ENOENT;
790	if (len < 0)
791	return len;
792
793	audit_log_kern_module(name);
794
795	if (mutex_lock_interruptible(&module_mutex) != 0)
796	return -EINTR;
797
798	mod = find_module(name);
799	if (!mod) {
800	ret = -ENOENT;
801	goto out;
802	}
803
804	if (!list_empty(head: &mod->source_list)) {
805	/* Other modules depend on us: get rid of them first. */
806	ret = -EWOULDBLOCK;
807	goto out;
808	}
809
810	/* Doing init or already dying? */
811	if (mod->state != MODULE_STATE_LIVE) {
812	/* FIXME: if (force), slam module count damn the torpedoes */
813	pr_debug("%s already dying\n", mod->name);
814	ret = -EBUSY;
815	goto out;
816	}
817
818	/* If it has an init func, it must have an exit func to unload */
819	if (mod->init && !mod->exit) {
820	forced = try_force_unload(flags);
821	if (!forced) {
822	/* This module can't be removed */
823	ret = -EBUSY;
824	goto out;
825	}
826	}
827
828	ret = try_stop_module(mod, flags, forced: &forced);
829	if (ret != 0)
830	goto out;
831
832	mutex_unlock(lock: &module_mutex);
833	/* Final destruction now no one is using it. */
834	if (mod->exit != NULL)
835	mod->exit();
836	blocking_notifier_call_chain(nh: &module_notify_list,
837	val: MODULE_STATE_GOING, v: mod);
838	klp_module_going(mod);
839	ftrace_release_mod(mod);
840
841	async_synchronize_full();
842
843	/* Store the name and taints of the last unloaded module for diagnostic purposes */
844	strscpy(last_unloaded_module.name, mod->name);
845	strscpy(last_unloaded_module.taints, module_flags(mod, buf, false));
846
847	free_module(mod);
848	/* someone could wait for the module in add_unformed_module() */
849	wake_up_all(&module_wq);
850	return 0;
851	out:
852	mutex_unlock(lock: &module_mutex);
853	return ret;
854	}
855
856	void __symbol_put(const char *symbol)
857	{
858	struct find_symbol_arg fsa = {
859	.name = symbol,
860	.gplok = true,
861	};
862
863	guard(rcu)();
864	BUG_ON(!find_symbol(&fsa));
865	module_put(module: fsa.owner);
866	}
867	EXPORT_SYMBOL(__symbol_put);
868
869	/* Note this assumes addr is a function, which it currently always is. */
870	void symbol_put_addr(void *addr)
871	{
872	struct module *modaddr;
873	unsigned long a = (unsigned long)dereference_function_descriptor(addr);
874
875	if (core_kernel_text(addr: a))
876	return;
877
878	/*
879	* Even though we hold a reference on the module; we still need to
880	* RCU read section in order to safely traverse the data structure.
881	*/
882	guard(rcu)();
883	modaddr = __module_text_address(addr: a);
884	BUG_ON(!modaddr);
885	module_put(module: modaddr);
886	}
887	EXPORT_SYMBOL_GPL(symbol_put_addr);
888
889	static ssize_t show_refcnt(const struct module_attribute *mattr,
890	struct module_kobject *mk, char *buffer)
891	{
892	return sprintf(buf: buffer, fmt: "%i\n", module_refcount(mk->mod));
893	}
894
895	static const struct module_attribute modinfo_refcnt =
896	__ATTR(refcnt, 0444, show_refcnt, NULL);
897
898	void __module_get(struct module *module)
899	{
900	if (module) {
901	atomic_inc(v: &module->refcnt);
902	trace_module_get(mod: module, _RET_IP_);
903	}
904	}
905	EXPORT_SYMBOL(__module_get);
906
907	bool try_module_get(struct module *module)
908	{
909	bool ret = true;
910
911	if (module) {
912	/* Note: here, we can fail to get a reference */
913	if (likely(module_is_live(module) &&
914	atomic_inc_not_zero(&module->refcnt) != 0))
915	trace_module_get(mod: module, _RET_IP_);
916	else
917	ret = false;
918	}
919	return ret;
920	}
921	EXPORT_SYMBOL(try_module_get);
922
923	void module_put(struct module *module)
924	{
925	int ret;
926
927	if (module) {
928	ret = atomic_dec_if_positive(v: &module->refcnt);
929	WARN_ON(ret < 0); /* Failed to put refcount */
930	trace_module_put(mod: module, _RET_IP_);
931	}
932	}
933	EXPORT_SYMBOL(module_put);
934
935	#else /* !CONFIG_MODULE_UNLOAD */
936	static inline void module_unload_free(struct module *mod)
937	{
938	}
939
940	static int ref_module(struct module *a, struct module *b)
941	{
942	return strong_try_module_get(b);
943	}
944
945	static inline int module_unload_init(struct module *mod)
946	{
947	return 0;
948	}
949	#endif /* CONFIG_MODULE_UNLOAD */
950
951	size_t module_flags_taint(unsigned long taints, char *buf)
952	{
953	size_t l = 0;
954	int i;
955
956	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
957	if (test_bit(i, &taints))
958	buf[l++] = taint_flags[i].c_true;
959	}
960
961	return l;
962	}
963
964	static ssize_t show_initstate(const struct module_attribute *mattr,
965	struct module_kobject *mk, char *buffer)
966	{
967	const char *state = "unknown";
968
969	switch (mk->mod->state) {
970	case MODULE_STATE_LIVE:
971	state = "live";
972	break;
973	case MODULE_STATE_COMING:
974	state = "coming";
975	break;
976	case MODULE_STATE_GOING:
977	state = "going";
978	break;
979	default:
980	BUG();
981	}
982	return sprintf(buf: buffer, fmt: "%s\n", state);
983	}
984
985	static const struct module_attribute modinfo_initstate =
986	__ATTR(initstate, 0444, show_initstate, NULL);
987
988	static ssize_t store_uevent(const struct module_attribute *mattr,
989	struct module_kobject *mk,
990	const char *buffer, size_t count)
991	{
992	int rc;
993
994	rc = kobject_synth_uevent(kobj: &mk->kobj, buf: buffer, count);
995	return rc ? rc : count;
996	}
997
998	const struct module_attribute module_uevent =
999	__ATTR(uevent, 0200, NULL, store_uevent);
1000
1001	static ssize_t show_coresize(const struct module_attribute *mattr,
1002	struct module_kobject *mk, char *buffer)
1003	{
1004	unsigned int size = mk->mod->mem[MOD_TEXT].size;
1005
1006	if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
1007	for_class_mod_mem_type(type, core_data)
1008	size += mk->mod->mem[type].size;
1009	}
1010	return sprintf(buf: buffer, fmt: "%u\n", size);
1011	}
1012
1013	static const struct module_attribute modinfo_coresize =
1014	__ATTR(coresize, 0444, show_coresize, NULL);
1015
1016	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
1017	static ssize_t show_datasize(const struct module_attribute *mattr,
1018	struct module_kobject *mk, char *buffer)
1019	{
1020	unsigned int size = 0;
1021
1022	for_class_mod_mem_type(type, core_data)
1023	size += mk->mod->mem[type].size;
1024	return sprintf(buffer, "%u\n", size);
1025	}
1026
1027	static const struct module_attribute modinfo_datasize =
1028	__ATTR(datasize, 0444, show_datasize, NULL);
1029	#endif
1030
1031	static ssize_t show_initsize(const struct module_attribute *mattr,
1032	struct module_kobject *mk, char *buffer)
1033	{
1034	unsigned int size = 0;
1035
1036	for_class_mod_mem_type(type, init)
1037	size += mk->mod->mem[type].size;
1038	return sprintf(buf: buffer, fmt: "%u\n", size);
1039	}
1040
1041	static const struct module_attribute modinfo_initsize =
1042	__ATTR(initsize, 0444, show_initsize, NULL);
1043
1044	static ssize_t show_taint(const struct module_attribute *mattr,
1045	struct module_kobject *mk, char *buffer)
1046	{
1047	size_t l;
1048
1049	l = module_flags_taint(taints: mk->mod->taints, buf: buffer);
1050	buffer[l++] = '\n';
1051	return l;
1052	}
1053
1054	static const struct module_attribute modinfo_taint =
1055	__ATTR(taint, 0444, show_taint, NULL);
1056
1057	const struct module_attribute *const modinfo_attrs[] = {
1058	&module_uevent,
1059	&modinfo_version,
1060	&modinfo_srcversion,
1061	&modinfo_initstate,
1062	&modinfo_coresize,
1063	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
1064	&modinfo_datasize,
1065	#endif
1066	&modinfo_initsize,
1067	&modinfo_taint,
1068	#ifdef CONFIG_MODULE_UNLOAD
1069	&modinfo_refcnt,
1070	#endif
1071	NULL,
1072	};
1073
1074	const size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
1075
1076	static const char vermagic[] = VERMAGIC_STRING;
1077
1078	int try_to_force_load(struct module *mod, const char *reason)
1079	{
1080	#ifdef CONFIG_MODULE_FORCE_LOAD
1081	if (!test_taint(TAINT_FORCED_MODULE))
1082	pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1083	add_taint_module(mod, TAINT_FORCED_MODULE, lockdep_ok: LOCKDEP_NOW_UNRELIABLE);
1084	return 0;
1085	#else
1086	return -ENOEXEC;
1087	#endif
1088	}
1089
1090	/* Parse tag=value strings from .modinfo section */
1091	char *module_next_tag_pair(char *string, unsigned long *secsize)
1092	{
1093	/* Skip non-zero chars */
1094	while (string[0]) {
1095	string++;
1096	if ((*secsize)-- <= 1)
1097	return NULL;
1098	}
1099
1100	/* Skip any zero padding. */
1101	while (!string[0]) {
1102	string++;
1103	if ((*secsize)-- <= 1)
1104	return NULL;
1105	}
1106	return string;
1107	}
1108
1109	static char *get_next_modinfo(const struct load_info *info, const char *tag,
1110	char *prev)
1111	{
1112	char *p;
1113	unsigned int taglen = strlen(tag);
1114	Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1115	unsigned long size = infosec->sh_size;
1116
1117	/*
1118	* get_modinfo() calls made before rewrite_section_headers()
1119	* must use sh_offset, as sh_addr isn't set!
1120	*/
1121	char *modinfo = (char *)info->hdr + infosec->sh_offset;
1122
1123	if (prev) {
1124	size -= prev - modinfo;
1125	modinfo = module_next_tag_pair(string: prev, secsize: &size);
1126	}
1127
1128	for (p = modinfo; p; p = module_next_tag_pair(string: p, secsize: &size)) {
1129	if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1130	return p + taglen + 1;
1131	}
1132	return NULL;
1133	}
1134
1135	static char *get_modinfo(const struct load_info *info, const char *tag)
1136	{
1137	return get_next_modinfo(info, tag, NULL);
1138	}
1139
1140	/**
1141	* verify_module_namespace() - does @modname have access to this symbol's @namespace
1142	* @namespace: export symbol namespace
1143	* @modname: module name
1144	*
1145	* If @namespace is prefixed with "module:" to indicate it is a module namespace
1146	* then test if @modname matches any of the comma separated patterns.
1147	*
1148	* The patterns only support tail-glob.
1149	*/
1150	static bool verify_module_namespace(const char *namespace, const char *modname)
1151	{
1152	size_t len, modlen = strlen(modname);
1153	const char *prefix = "module:";
1154	const char *sep;
1155	bool glob;
1156
1157	if (!strstarts(str: namespace, prefix))
1158	return false;
1159
1160	for (namespace += strlen(prefix); *namespace; namespace = sep) {
1161	sep = strchrnul(namespace, ',');
1162	len = sep - namespace;
1163
1164	glob = false;
1165	if (sep[-1] == '*') {
1166	len--;
1167	glob = true;
1168	}
1169
1170	if (*sep)
1171	sep++;
1172
1173	if (mod_strncmp(str_a: namespace, str_b: modname, n: len) == 0 && (glob || len == modlen))
1174	return true;
1175	}
1176
1177	return false;
1178	}
1179
1180	static int verify_namespace_is_imported(const struct load_info *info,
1181	const struct kernel_symbol *sym,
1182	struct module *mod)
1183	{
1184	const char *namespace;
1185	char *imported_namespace;
1186
1187	namespace = kernel_symbol_namespace(sym);
1188	if (namespace && namespace[0]) {
1189
1190	if (verify_module_namespace(namespace, modname: mod->name))
1191	return 0;
1192
1193	for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1194	if (strcmp(namespace, imported_namespace) == 0)
1195	return 0;
1196	}
1197	#ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1198	pr_warn(
1199	#else
1200	pr_err(
1201	#endif
1202	"%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1203	mod->name, kernel_symbol_name(sym), namespace);
1204	#ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1205	return -EINVAL;
1206	#endif
1207	}
1208	return 0;
1209	}
1210
1211	static bool inherit_taint(struct module *mod, struct module *owner, const char *name)
1212	{
1213	if (!owner || !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1214	return true;
1215
1216	if (mod->using_gplonly_symbols) {
1217	pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1218	mod->name, name, owner->name);
1219	return false;
1220	}
1221
1222	if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1223	pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1224	mod->name, name, owner->name);
1225	set_bit(TAINT_PROPRIETARY_MODULE, addr: &mod->taints);
1226	}
1227	return true;
1228	}
1229
1230	/* Resolve a symbol for this module. I.e. if we find one, record usage. */
1231	static const struct kernel_symbol *resolve_symbol(struct module *mod,
1232	const struct load_info *info,
1233	const char *name,
1234	char ownername[])
1235	{
1236	struct find_symbol_arg fsa = {
1237	.name = name,
1238	.gplok = !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1239	.warn = true,
1240	};
1241	int err;
1242
1243	/*
1244	* The module_mutex should not be a heavily contended lock;
1245	* if we get the occasional sleep here, we'll go an extra iteration
1246	* in the wait_event_interruptible(), which is harmless.
1247	*/
1248	sched_annotate_sleep();
1249	mutex_lock(&module_mutex);
1250	if (!find_symbol(fsa: &fsa))
1251	goto unlock;
1252
1253	if (fsa.license == GPL_ONLY)
1254	mod->using_gplonly_symbols = true;
1255
1256	if (!inherit_taint(mod, owner: fsa.owner, name)) {
1257	fsa.sym = NULL;
1258	goto getname;
1259	}
1260
1261	if (!check_version(info, symname: name, mod, crc: fsa.crc)) {
1262	fsa.sym = ERR_PTR(error: -EINVAL);
1263	goto getname;
1264	}
1265
1266	err = verify_namespace_is_imported(info, sym: fsa.sym, mod);
1267	if (err) {
1268	fsa.sym = ERR_PTR(error: err);
1269	goto getname;
1270	}
1271
1272	err = ref_module(a: mod, b: fsa.owner);
1273	if (err) {
1274	fsa.sym = ERR_PTR(error: err);
1275	goto getname;
1276	}
1277
1278	getname:
1279	/* We must make copy under the lock if we failed to get ref. */
1280	strscpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1281	unlock:
1282	mutex_unlock(lock: &module_mutex);
1283	return fsa.sym;
1284	}
1285
1286	static const struct kernel_symbol *
1287	resolve_symbol_wait(struct module *mod,
1288	const struct load_info *info,
1289	const char *name)
1290	{
1291	const struct kernel_symbol *ksym;
1292	char owner[MODULE_NAME_LEN];
1293
1294	if (wait_event_interruptible_timeout(module_wq,
1295	!IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1296	|| PTR_ERR(ksym) != -EBUSY,
1297	30 * HZ) <= 0) {
1298	pr_warn("%s: gave up waiting for init of module %s.\n",
1299	mod->name, owner);
1300	}
1301	return ksym;
1302	}
1303
1304	void __weak module_arch_cleanup(struct module *mod)
1305	{
1306	}
1307
1308	void __weak module_arch_freeing_init(struct module *mod)
1309	{
1310	}
1311
1312	static int module_memory_alloc(struct module *mod, enum mod_mem_type type)
1313	{
1314	unsigned int size = PAGE_ALIGN(mod->mem[type].size);
1315	enum execmem_type execmem_type;
1316	void *ptr;
1317
1318	mod->mem[type].size = size;
1319
1320	if (mod_mem_type_is_data(type))
1321	execmem_type = EXECMEM_MODULE_DATA;
1322	else
1323	execmem_type = EXECMEM_MODULE_TEXT;
1324
1325	ptr = execmem_alloc_rw(type: execmem_type, size);
1326	if (!ptr)
1327	return -ENOMEM;
1328
1329	mod->mem[type].is_rox = execmem_is_rox(type: execmem_type);
1330
1331	/*
1332	* The pointer to these blocks of memory are stored on the module
1333	* structure and we keep that around so long as the module is
1334	* around. We only free that memory when we unload the module.
1335	* Just mark them as not being a leak then. The .init* ELF
1336	* sections *do* get freed after boot so we *could* treat them
1337	* slightly differently with kmemleak_ignore() and only grey
1338	* them out as they work as typical memory allocations which
1339	* *do* eventually get freed, but let's just keep things simple
1340	* and avoid *any* false positives.
1341	*/
1342	if (!mod->mem[type].is_rox)
1343	kmemleak_not_leak(ptr);
1344
1345	memset(ptr, 0, size);
1346	mod->mem[type].base = ptr;
1347
1348	return 0;
1349	}
1350
1351	static void module_memory_restore_rox(struct module *mod)
1352	{
1353	for_class_mod_mem_type(type, text) {
1354	struct module_memory *mem = &mod->mem[type];
1355
1356	if (mem->is_rox)
1357	execmem_restore_rox(ptr: mem->base, size: mem->size);
1358	}
1359	}
1360
1361	static void module_memory_free(struct module *mod, enum mod_mem_type type)
1362	{
1363	struct module_memory *mem = &mod->mem[type];
1364
1365	execmem_free(ptr: mem->base);
1366	}
1367
1368	static void free_mod_mem(struct module *mod)
1369	{
1370	for_each_mod_mem_type(type) {
1371	struct module_memory *mod_mem = &mod->mem[type];
1372
1373	if (type == MOD_DATA)
1374	continue;
1375
1376	/* Free lock-classes; relies on the preceding sync_rcu(). */
1377	lockdep_free_key_range(start: mod_mem->base, size: mod_mem->size);
1378	if (mod_mem->size)
1379	module_memory_free(mod, type);
1380	}
1381
1382	/* MOD_DATA hosts mod, so free it at last */
1383	lockdep_free_key_range(start: mod->mem[MOD_DATA].base, size: mod->mem[MOD_DATA].size);
1384	module_memory_free(mod, type: MOD_DATA);
1385	}
1386
1387	/* Free a module, remove from lists, etc. */
1388	static void free_module(struct module *mod)
1389	{
1390	trace_module_free(mod);
1391
1392	codetag_unload_module(mod);
1393
1394	mod_sysfs_teardown(mod);
1395
1396	/*
1397	* We leave it in list to prevent duplicate loads, but make sure
1398	* that noone uses it while it's being deconstructed.
1399	*/
1400	mutex_lock(&module_mutex);
1401	mod->state = MODULE_STATE_UNFORMED;
1402	mutex_unlock(lock: &module_mutex);
1403
1404	/* Arch-specific cleanup. */
1405	module_arch_cleanup(mod);
1406
1407	/* Module unload stuff */
1408	module_unload_free(mod);
1409
1410	/* Free any allocated parameters. */
1411	destroy_params(params: mod->kp, num: mod->num_kp);
1412
1413	if (is_livepatch_module(mod))
1414	free_module_elf(mod);
1415
1416	/* Now we can delete it from the lists */
1417	mutex_lock(&module_mutex);
1418	/* Unlink carefully: kallsyms could be walking list. */
1419	list_del_rcu(entry: &mod->list);
1420	mod_tree_remove(mod);
1421	/* Remove this module from bug list, this uses list_del_rcu */
1422	module_bug_cleanup(mod);
1423	/* Wait for RCU synchronizing before releasing mod->list and buglist. */
1424	synchronize_rcu();
1425	if (try_add_tainted_module(mod))
1426	pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1427	mod->name);
1428	mutex_unlock(lock: &module_mutex);
1429
1430	/* This may be empty, but that's OK */
1431	module_arch_freeing_init(mod);
1432	kfree(objp: mod->args);
1433	percpu_modfree(mod);
1434
1435	free_mod_mem(mod);
1436	}
1437
1438	void *__symbol_get(const char *symbol)
1439	{
1440	struct find_symbol_arg fsa = {
1441	.name = symbol,
1442	.gplok = true,
1443	.warn = true,
1444	};
1445
1446	scoped_guard(rcu) {
1447	if (!find_symbol(fsa: &fsa))
1448	return NULL;
1449	if (fsa.license != GPL_ONLY) {
1450	pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1451	symbol);
1452	return NULL;
1453	}
1454	if (strong_try_module_get(mod: fsa.owner))
1455	return NULL;
1456	}
1457	return (void *)kernel_symbol_value(sym: fsa.sym);
1458	}
1459	EXPORT_SYMBOL_GPL(__symbol_get);
1460
1461	/*
1462	* Ensure that an exported symbol [global namespace] does not already exist
1463	* in the kernel or in some other module's exported symbol table.
1464	*
1465	* You must hold the module_mutex.
1466	*/
1467	static int verify_exported_symbols(struct module *mod)
1468	{
1469	unsigned int i;
1470	const struct kernel_symbol *s;
1471	struct {
1472	const struct kernel_symbol *sym;
1473	unsigned int num;
1474	} arr[] = {
1475	{ mod->syms, mod->num_syms },
1476	{ mod->gpl_syms, mod->num_gpl_syms },
1477	};
1478
1479	for (i = 0; i < ARRAY_SIZE(arr); i++) {
1480	for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1481	struct find_symbol_arg fsa = {
1482	.name = kernel_symbol_name(sym: s),
1483	.gplok = true,
1484	};
1485	if (find_symbol(fsa: &fsa)) {
1486	pr_err("%s: exports duplicate symbol %s"
1487	" (owned by %s)\n",
1488	mod->name, kernel_symbol_name(s),
1489	module_name(fsa.owner));
1490	return -ENOEXEC;
1491	}
1492	}
1493	}
1494	return 0;
1495	}
1496
1497	static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1498	{
1499	/*
1500	* On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1501	* before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1502	* i386 has a similar problem but may not deserve a fix.
1503	*
1504	* If we ever have to ignore many symbols, consider refactoring the code to
1505	* only warn if referenced by a relocation.
1506	*/
1507	if (emachine == EM_386 || emachine == EM_X86_64)
1508	return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1509	return false;
1510	}
1511
1512	/* Change all symbols so that st_value encodes the pointer directly. */
1513	static int simplify_symbols(struct module *mod, const struct load_info *info)
1514	{
1515	Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1516	Elf_Sym *sym = (void *)symsec->sh_addr;
1517	unsigned long secbase;
1518	unsigned int i;
1519	int ret = 0;
1520	const struct kernel_symbol *ksym;
1521
1522	for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1523	const char *name = info->strtab + sym[i].st_name;
1524
1525	switch (sym[i].st_shndx) {
1526	case SHN_COMMON:
1527	/* Ignore common symbols */
1528	if (!strncmp(name, "__gnu_lto", 9))
1529	break;
1530
1531	/*
1532	* We compiled with -fno-common. These are not
1533	* supposed to happen.
1534	*/
1535	pr_debug("Common symbol: %s\n", name);
1536	pr_warn("%s: please compile with -fno-common\n",
1537	mod->name);
1538	ret = -ENOEXEC;
1539	break;
1540
1541	case SHN_ABS:
1542	/* Don't need to do anything */
1543	pr_debug("Absolute symbol: 0x%08lx %s\n",
1544	(long)sym[i].st_value, name);
1545	break;
1546
1547	case SHN_LIVEPATCH:
1548	/* Livepatch symbols are resolved by livepatch */
1549	break;
1550
1551	case SHN_UNDEF:
1552	ksym = resolve_symbol_wait(mod, info, name);
1553	/* Ok if resolved. */
1554	if (ksym && !IS_ERR(ptr: ksym)) {
1555	sym[i].st_value = kernel_symbol_value(sym: ksym);
1556	break;
1557	}
1558
1559	/* Ok if weak or ignored. */
1560	if (!ksym &&
1561	(ELF_ST_BIND(sym[i].st_info) == STB_WEAK ||
1562	ignore_undef_symbol(emachine: info->hdr->e_machine, name)))
1563	break;
1564
1565	ret = PTR_ERR(ptr: ksym) ?: -ENOENT;
1566	pr_warn("%s: Unknown symbol %s (err %d)\n",
1567	mod->name, name, ret);
1568	break;
1569
1570	default:
1571	/* Divert to percpu allocation if a percpu var. */
1572	if (sym[i].st_shndx == info->index.pcpu)
1573	secbase = (unsigned long)mod_percpu(mod);
1574	else
1575	secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1576	sym[i].st_value += secbase;
1577	break;
1578	}
1579	}
1580
1581	return ret;
1582	}
1583
1584	static int apply_relocations(struct module *mod, const struct load_info *info)
1585	{
1586	unsigned int i;
1587	int err = 0;
1588
1589	/* Now do relocations. */
1590	for (i = 1; i < info->hdr->e_shnum; i++) {
1591	unsigned int infosec = info->sechdrs[i].sh_info;
1592
1593	/* Not a valid relocation section? */
1594	if (infosec >= info->hdr->e_shnum)
1595	continue;
1596
1597	/*
1598	* Don't bother with non-allocated sections.
1599	* An exception is the percpu section, which has separate allocations
1600	* for individual CPUs. We relocate the percpu section in the initial
1601	* ELF template and subsequently copy it to the per-CPU destinations.
1602	*/
1603	if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC) &&
1604	(!infosec || infosec != info->index.pcpu))
1605	continue;
1606
1607	if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1608	err = klp_apply_section_relocs(pmod: mod, sechdrs: info->sechdrs,
1609	shstrtab: info->secstrings,
1610	strtab: info->strtab,
1611	symindex: info->index.sym, secindex: i,
1612	NULL);
1613	else if (info->sechdrs[i].sh_type == SHT_REL)
1614	err = apply_relocate(sechdrs: info->sechdrs, strtab: info->strtab,
1615	symindex: info->index.sym, relsec: i, me: mod);
1616	else if (info->sechdrs[i].sh_type == SHT_RELA)
1617	err = apply_relocate_add(sechdrs: info->sechdrs, strtab: info->strtab,
1618	symindex: info->index.sym, relsec: i, mod);
1619	if (err < 0)
1620	break;
1621	}
1622	return err;
1623	}
1624
1625	/* Additional bytes needed by arch in front of individual sections */
1626	unsigned int __weak arch_mod_section_prepend(struct module *mod,
1627	unsigned int section)
1628	{
1629	/* default implementation just returns zero */
1630	return 0;
1631	}
1632
1633	long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1634	Elf_Shdr *sechdr, unsigned int section)
1635	{
1636	long offset;
1637	long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1638
1639	mod->mem[type].size += arch_mod_section_prepend(mod, section);
1640	offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1641	mod->mem[type].size = offset + sechdr->sh_size;
1642
1643	WARN_ON_ONCE(offset & mask);
1644	return offset | mask;
1645	}
1646
1647	bool module_init_layout_section(const char *sname)
1648	{
1649	#ifndef CONFIG_MODULE_UNLOAD
1650	if (module_exit_section(sname))
1651	return true;
1652	#endif
1653	return module_init_section(name: sname);
1654	}
1655
1656	static void __layout_sections(struct module *mod, struct load_info *info, bool is_init)
1657	{
1658	unsigned int m, i;
1659
1660	/*
1661	* { Mask of required section header flags,
1662	* Mask of excluded section header flags }
1663	*/
1664	static const unsigned long masks[][2] = {
1665	{ SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1666	{ SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1667	{ SHF_RO_AFTER_INIT | SHF_ALLOC, ARCH_SHF_SMALL },
1668	{ SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1669	{ ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1670	};
1671	static const int core_m_to_mem_type[] = {
1672	MOD_TEXT,
1673	MOD_RODATA,
1674	MOD_RO_AFTER_INIT,
1675	MOD_DATA,
1676	MOD_DATA,
1677	};
1678	static const int init_m_to_mem_type[] = {
1679	MOD_INIT_TEXT,
1680	MOD_INIT_RODATA,
1681	MOD_INVALID,
1682	MOD_INIT_DATA,
1683	MOD_INIT_DATA,
1684	};
1685
1686	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1687	enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1688
1689	for (i = 0; i < info->hdr->e_shnum; ++i) {
1690	Elf_Shdr *s = &info->sechdrs[i];
1691	const char *sname = info->secstrings + s->sh_name;
1692
1693	if ((s->sh_flags & masks[m][0]) != masks[m][0]
1694	|| (s->sh_flags & masks[m][1])
1695	|| s->sh_entsize != ~0UL
1696	|| is_init != module_init_layout_section(sname))
1697	continue;
1698
1699	if (WARN_ON_ONCE(type == MOD_INVALID))
1700	continue;
1701
1702	/*
1703	* Do not allocate codetag memory as we load it into
1704	* preallocated contiguous memory.
1705	*/
1706	if (codetag_needs_module_section(mod, name: sname, size: s->sh_size)) {
1707	/*
1708	* s->sh_entsize won't be used but populate the
1709	* type field to avoid confusion.
1710	*/
1711	s->sh_entsize = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK)
1712	<< SH_ENTSIZE_TYPE_SHIFT;
1713	continue;
1714	}
1715
1716	s->sh_entsize = module_get_offset_and_type(mod, type, sechdr: s, section: i);
1717	pr_debug("\t%s\n", sname);
1718	}
1719	}
1720	}
1721
1722	/*
1723	* Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1724	* might -- code, read-only data, read-write data, small data. Tally
1725	* sizes, and place the offsets into sh_entsize fields: high bit means it
1726	* belongs in init.
1727	*/
1728	static void layout_sections(struct module *mod, struct load_info *info)
1729	{
1730	unsigned int i;
1731
1732	for (i = 0; i < info->hdr->e_shnum; i++)
1733	info->sechdrs[i].sh_entsize = ~0UL;
1734
1735	pr_debug("Core section allocation order for %s:\n", mod->name);
1736	__layout_sections(mod, info, is_init: false);
1737
1738	pr_debug("Init section allocation order for %s:\n", mod->name);
1739	__layout_sections(mod, info, is_init: true);
1740	}
1741
1742	static void module_license_taint_check(struct module *mod, const char *license)
1743	{
1744	if (!license)
1745	license = "unspecified";
1746
1747	if (!license_is_gpl_compatible(license)) {
1748	if (!test_taint(TAINT_PROPRIETARY_MODULE))
1749	pr_warn("%s: module license '%s' taints kernel.\n",
1750	mod->name, license);
1751	add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1752	lockdep_ok: LOCKDEP_NOW_UNRELIABLE);
1753	}
1754	}
1755
1756	static int setup_modinfo(struct module *mod, struct load_info *info)
1757	{
1758	const struct module_attribute *attr;
1759	char *imported_namespace;
1760	int i;
1761
1762	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1763	if (attr->setup)
1764	attr->setup(mod, get_modinfo(info, tag: attr->attr.name));
1765	}
1766
1767	for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1768	/*
1769	* 'module:' prefixed namespaces are implicit, disallow
1770	* explicit imports.
1771	*/
1772	if (strstarts(str: imported_namespace, prefix: "module:")) {
1773	pr_err("%s: module tries to import module namespace: %s\n",
1774	mod->name, imported_namespace);
1775	return -EPERM;
1776	}
1777	}
1778
1779	return 0;
1780	}
1781
1782	static void free_modinfo(struct module *mod)
1783	{
1784	const struct module_attribute *attr;
1785	int i;
1786
1787	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1788	if (attr->free)
1789	attr->free(mod);
1790	}
1791	}
1792
1793	bool __weak module_init_section(const char *name)
1794	{
1795	return strstarts(str: name, prefix: ".init");
1796	}
1797
1798	bool __weak module_exit_section(const char *name)
1799	{
1800	return strstarts(str: name, prefix: ".exit");
1801	}
1802
1803	static int validate_section_offset(const struct load_info *info, Elf_Shdr *shdr)
1804	{
1805	#if defined(CONFIG_64BIT)
1806	unsigned long long secend;
1807	#else
1808	unsigned long secend;
1809	#endif
1810
1811	/*
1812	* Check for both overflow and offset/size being
1813	* too large.
1814	*/
1815	secend = shdr->sh_offset + shdr->sh_size;
1816	if (secend < shdr->sh_offset || secend > info->len)
1817	return -ENOEXEC;
1818
1819	return 0;
1820	}
1821
1822	/**
1823	* elf_validity_ehdr() - Checks an ELF header for module validity
1824	* @info: Load info containing the ELF header to check
1825	*
1826	* Checks whether an ELF header could belong to a valid module. Checks:
1827	*
1828	* * ELF header is within the data the user provided
1829	* * ELF magic is present
1830	* * It is relocatable (not final linked, not core file, etc.)
1831	* * The header's machine type matches what the architecture expects.
1832	* * Optional arch-specific hook for other properties
1833	* - module_elf_check_arch() is currently only used by PPC to check
1834	* ELF ABI version, but may be used by others in the future.
1835	*
1836	* Return: %0 if valid, %-ENOEXEC on failure.
1837	*/
1838	static int elf_validity_ehdr(const struct load_info *info)
1839	{
1840	if (info->len < sizeof(*(info->hdr))) {
1841	pr_err("Invalid ELF header len %lu\n", info->len);
1842	return -ENOEXEC;
1843	}
1844	if (memcmp(p: info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1845	pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1846	return -ENOEXEC;
1847	}
1848	if (info->hdr->e_type != ET_REL) {
1849	pr_err("Invalid ELF header type: %u != %u\n",
1850	info->hdr->e_type, ET_REL);
1851	return -ENOEXEC;
1852	}
1853	if (!elf_check_arch(info->hdr)) {
1854	pr_err("Invalid architecture in ELF header: %u\n",
1855	info->hdr->e_machine);
1856	return -ENOEXEC;
1857	}
1858	if (!module_elf_check_arch(hdr: info->hdr)) {
1859	pr_err("Invalid module architecture in ELF header: %u\n",
1860	info->hdr->e_machine);
1861	return -ENOEXEC;
1862	}
1863	return 0;
1864	}
1865
1866	/**
1867	* elf_validity_cache_sechdrs() - Cache section headers if valid
1868	* @info: Load info to compute section headers from
1869	*
1870	* Checks:
1871	*
1872	* * ELF header is valid (see elf_validity_ehdr())
1873	* * Section headers are the size we expect
1874	* * Section array fits in the user provided data
1875	* * Section index 0 is NULL
1876	* * Section contents are inbounds
1877	*
1878	* Then updates @info with a &load_info->sechdrs pointer if valid.
1879	*
1880	* Return: %0 if valid, negative error code if validation failed.
1881	*/
1882	static int elf_validity_cache_sechdrs(struct load_info *info)
1883	{
1884	Elf_Shdr *sechdrs;
1885	Elf_Shdr *shdr;
1886	int i;
1887	int err;
1888
1889	err = elf_validity_ehdr(info);
1890	if (err < 0)
1891	return err;
1892
1893	if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1894	pr_err("Invalid ELF section header size\n");
1895	return -ENOEXEC;
1896	}
1897
1898	/*
1899	* e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1900	* known and small. So e_shnum * sizeof(Elf_Shdr)
1901	* will not overflow unsigned long on any platform.
1902	*/
1903	if (info->hdr->e_shoff >= info->len
1904	|| (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1905	info->len - info->hdr->e_shoff)) {
1906	pr_err("Invalid ELF section header overflow\n");
1907	return -ENOEXEC;
1908	}
1909
1910	sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1911
1912	/*
1913	* The code assumes that section 0 has a length of zero and
1914	* an addr of zero, so check for it.
1915	*/
1916	if (sechdrs[0].sh_type != SHT_NULL
1917	|| sechdrs[0].sh_size != 0
1918	|| sechdrs[0].sh_addr != 0) {
1919	pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1920	sechdrs[0].sh_type);
1921	return -ENOEXEC;
1922	}
1923
1924	/* Validate contents are inbounds */
1925	for (i = 1; i < info->hdr->e_shnum; i++) {
1926	shdr = &sechdrs[i];
1927	switch (shdr->sh_type) {
1928	case SHT_NULL:
1929	case SHT_NOBITS:
1930	/* No contents, offset/size don't mean anything */
1931	continue;
1932	default:
1933	err = validate_section_offset(info, shdr);
1934	if (err < 0) {
1935	pr_err("Invalid ELF section in module (section %u type %u)\n",
1936	i, shdr->sh_type);
1937	return err;
1938	}
1939	}
1940	}
1941
1942	info->sechdrs = sechdrs;
1943
1944	return 0;
1945	}
1946
1947	/**
1948	* elf_validity_cache_secstrings() - Caches section names if valid
1949	* @info: Load info to cache section names from. Must have valid sechdrs.
1950	*
1951	* Specifically checks:
1952	*
1953	* * Section name table index is inbounds of section headers
1954	* * Section name table is not empty
1955	* * Section name table is NUL terminated
1956	* * All section name offsets are inbounds of the section
1957	*
1958	* Then updates @info with a &load_info->secstrings pointer if valid.
1959	*
1960	* Return: %0 if valid, negative error code if validation failed.
1961	*/
1962	static int elf_validity_cache_secstrings(struct load_info *info)
1963	{
1964	Elf_Shdr *strhdr, *shdr;
1965	char *secstrings;
1966	int i;
1967
1968	/*
1969	* Verify if the section name table index is valid.
1970	*/
1971	if (info->hdr->e_shstrndx == SHN_UNDEF
1972	|| info->hdr->e_shstrndx >= info->hdr->e_shnum) {
1973	pr_err("Invalid ELF section name index: %d || e_shstrndx (%d) >= e_shnum (%d)\n",
1974	info->hdr->e_shstrndx, info->hdr->e_shstrndx,
1975	info->hdr->e_shnum);
1976	return -ENOEXEC;
1977	}
1978
1979	strhdr = &info->sechdrs[info->hdr->e_shstrndx];
1980
1981	/*
1982	* The section name table must be NUL-terminated, as required
1983	* by the spec. This makes strcmp and pr_* calls that access
1984	* strings in the section safe.
1985	*/
1986	secstrings = (void *)info->hdr + strhdr->sh_offset;
1987	if (strhdr->sh_size == 0) {
1988	pr_err("empty section name table\n");
1989	return -ENOEXEC;
1990	}
1991	if (secstrings[strhdr->sh_size - 1] != '\0') {
1992	pr_err("ELF Spec violation: section name table isn't null terminated\n");
1993	return -ENOEXEC;
1994	}
1995
1996	for (i = 0; i < info->hdr->e_shnum; i++) {
1997	shdr = &info->sechdrs[i];
1998	/* SHT_NULL means sh_name has an undefined value */
1999	if (shdr->sh_type == SHT_NULL)
2000	continue;
2001	if (shdr->sh_name >= strhdr->sh_size) {
2002	pr_err("Invalid ELF section name in module (section %u type %u)\n",
2003	i, shdr->sh_type);
2004	return -ENOEXEC;
2005	}
2006	}
2007
2008	info->secstrings = secstrings;
2009	return 0;
2010	}
2011
2012	/**
2013	* elf_validity_cache_index_info() - Validate and cache modinfo section
2014	* @info: Load info to populate the modinfo index on.
2015	* Must have &load_info->sechdrs and &load_info->secstrings populated
2016	*
2017	* Checks that if there is a .modinfo section, it is unique.
2018	* Then, it caches its index in &load_info->index.info.
2019	* Finally, it tries to populate the name to improve error messages.
2020	*
2021	* Return: %0 if valid, %-ENOEXEC if multiple modinfo sections were found.
2022	*/
2023	static int elf_validity_cache_index_info(struct load_info *info)
2024	{
2025	int info_idx;
2026
2027	info_idx = find_any_unique_sec(info, name: ".modinfo");
2028
2029	if (info_idx == 0)
2030	/* Early return, no .modinfo */
2031	return 0;
2032
2033	if (info_idx < 0) {
2034	pr_err("Only one .modinfo section must exist.\n");
2035	return -ENOEXEC;
2036	}
2037
2038	info->index.info = info_idx;
2039	/* Try to find a name early so we can log errors with a module name */
2040	info->name = get_modinfo(info, tag: "name");
2041
2042	return 0;
2043	}
2044
2045	/**
2046	* elf_validity_cache_index_mod() - Validates and caches this_module section
2047	* @info: Load info to cache this_module on.
2048	* Must have &load_info->sechdrs and &load_info->secstrings populated
2049	*
2050	* The ".gnu.linkonce.this_module" ELF section is special. It is what modpost
2051	* uses to refer to __this_module and let's use rely on THIS_MODULE to point
2052	* to &__this_module properly. The kernel's modpost declares it on each
2053	* modules's *.mod.c file. If the struct module of the kernel changes a full
2054	* kernel rebuild is required.
2055	*
2056	* We have a few expectations for this special section, this function
2057	* validates all this for us:
2058	*
2059	* * The section has contents
2060	* * The section is unique
2061	* * We expect the kernel to always have to allocate it: SHF_ALLOC
2062	* * The section size must match the kernel's run time's struct module
2063	* size
2064	*
2065	* If all checks pass, the index will be cached in &load_info->index.mod
2066	*
2067	* Return: %0 on validation success, %-ENOEXEC on failure
2068	*/
2069	static int elf_validity_cache_index_mod(struct load_info *info)
2070	{
2071	Elf_Shdr *shdr;
2072	int mod_idx;
2073
2074	mod_idx = find_any_unique_sec(info, name: ".gnu.linkonce.this_module");
2075	if (mod_idx <= 0) {
2076	pr_err("module %s: Exactly one .gnu.linkonce.this_module section must exist.\n",
2077	info->name ?: "(missing .modinfo section or name field)");
2078	return -ENOEXEC;
2079	}
2080
2081	shdr = &info->sechdrs[mod_idx];
2082
2083	if (shdr->sh_type == SHT_NOBITS) {
2084	pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
2085	info->name ?: "(missing .modinfo section or name field)");
2086	return -ENOEXEC;
2087	}
2088
2089	if (!(shdr->sh_flags & SHF_ALLOC)) {
2090	pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
2091	info->name ?: "(missing .modinfo section or name field)");
2092	return -ENOEXEC;
2093	}
2094
2095	if (shdr->sh_size != sizeof(struct module)) {
2096	pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
2097	info->name ?: "(missing .modinfo section or name field)");
2098	return -ENOEXEC;
2099	}
2100
2101	info->index.mod = mod_idx;
2102
2103	return 0;
2104	}
2105
2106	/**
2107	* elf_validity_cache_index_sym() - Validate and cache symtab index
2108	* @info: Load info to cache symtab index in.
2109	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2110	*
2111	* Checks that there is exactly one symbol table, then caches its index in
2112	* &load_info->index.sym.
2113	*
2114	* Return: %0 if valid, %-ENOEXEC on failure.
2115	*/
2116	static int elf_validity_cache_index_sym(struct load_info *info)
2117	{
2118	unsigned int sym_idx;
2119	unsigned int num_sym_secs = 0;
2120	int i;
2121
2122	for (i = 1; i < info->hdr->e_shnum; i++) {
2123	if (info->sechdrs[i].sh_type == SHT_SYMTAB) {
2124	num_sym_secs++;
2125	sym_idx = i;
2126	}
2127	}
2128
2129	if (num_sym_secs != 1) {
2130	pr_warn("%s: module has no symbols (stripped?)\n",
2131	info->name ?: "(missing .modinfo section or name field)");
2132	return -ENOEXEC;
2133	}
2134
2135	info->index.sym = sym_idx;
2136
2137	return 0;
2138	}
2139
2140	/**
2141	* elf_validity_cache_index_str() - Validate and cache strtab index
2142	* @info: Load info to cache strtab index in.
2143	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2144	* Must have &load_info->index.sym populated.
2145	*
2146	* Looks at the symbol table's associated string table, makes sure it is
2147	* in-bounds, and caches it.
2148	*
2149	* Return: %0 if valid, %-ENOEXEC on failure.
2150	*/
2151	static int elf_validity_cache_index_str(struct load_info *info)
2152	{
2153	unsigned int str_idx = info->sechdrs[info->index.sym].sh_link;
2154
2155	if (str_idx == SHN_UNDEF || str_idx >= info->hdr->e_shnum) {
2156	pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
2157	str_idx, str_idx, info->hdr->e_shnum);
2158	return -ENOEXEC;
2159	}
2160
2161	info->index.str = str_idx;
2162	return 0;
2163	}
2164
2165	/**
2166	* elf_validity_cache_index_versions() - Validate and cache version indices
2167	* @info: Load info to cache version indices in.
2168	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2169	* @flags: Load flags, relevant to suppress version loading, see
2170	* uapi/linux/module.h
2171	*
2172	* If we're ignoring modversions based on @flags, zero all version indices
2173	* and return validity. Othewrise check:
2174	*
2175	* * If "__version_ext_crcs" is present, "__version_ext_names" is present
2176	* * There is a name present for every crc
2177	*
2178	* Then populate:
2179	*
2180	* * &load_info->index.vers
2181	* * &load_info->index.vers_ext_crc
2182	* * &load_info->index.vers_ext_names
2183	*
2184	* if present.
2185	*
2186	* Return: %0 if valid, %-ENOEXEC on failure.
2187	*/
2188	static int elf_validity_cache_index_versions(struct load_info *info, int flags)
2189	{
2190	unsigned int vers_ext_crc;
2191	unsigned int vers_ext_name;
2192	size_t crc_count;
2193	size_t remaining_len;
2194	size_t name_size;
2195	char *name;
2196
2197	/* If modversions were suppressed, pretend we didn't find any */
2198	if (flags & MODULE_INIT_IGNORE_MODVERSIONS) {
2199	info->index.vers = 0;
2200	info->index.vers_ext_crc = 0;
2201	info->index.vers_ext_name = 0;
2202	return 0;
2203	}
2204
2205	vers_ext_crc = find_sec(info, name: "__version_ext_crcs");
2206	vers_ext_name = find_sec(info, name: "__version_ext_names");
2207
2208	/* If we have one field, we must have the other */
2209	if (!!vers_ext_crc != !!vers_ext_name) {
2210	pr_err("extended version crc+name presence does not match");
2211	return -ENOEXEC;
2212	}
2213
2214	/*
2215	* If we have extended version information, we should have the same
2216	* number of entries in every section.
2217	*/
2218	if (vers_ext_crc) {
2219	crc_count = info->sechdrs[vers_ext_crc].sh_size / sizeof(u32);
2220	name = (void *)info->hdr +
2221	info->sechdrs[vers_ext_name].sh_offset;
2222	remaining_len = info->sechdrs[vers_ext_name].sh_size;
2223
2224	while (crc_count--) {
2225	name_size = strnlen(p: name, maxlen: remaining_len) + 1;
2226	if (name_size > remaining_len) {
2227	pr_err("more extended version crcs than names");
2228	return -ENOEXEC;
2229	}
2230	remaining_len -= name_size;
2231	name += name_size;
2232	}
2233	}
2234
2235	info->index.vers = find_sec(info, name: "__versions");
2236	info->index.vers_ext_crc = vers_ext_crc;
2237	info->index.vers_ext_name = vers_ext_name;
2238	return 0;
2239	}
2240
2241	/**
2242	* elf_validity_cache_index() - Resolve, validate, cache section indices
2243	* @info: Load info to read from and update.
2244	* &load_info->sechdrs and &load_info->secstrings must be populated.
2245	* @flags: Load flags, relevant to suppress version loading, see
2246	* uapi/linux/module.h
2247	*
2248	* Populates &load_info->index, validating as it goes.
2249	* See child functions for per-field validation:
2250	*
2251	* * elf_validity_cache_index_info()
2252	* * elf_validity_cache_index_mod()
2253	* * elf_validity_cache_index_sym()
2254	* * elf_validity_cache_index_str()
2255	* * elf_validity_cache_index_versions()
2256	*
2257	* If CONFIG_SMP is enabled, load the percpu section by name with no
2258	* validation.
2259	*
2260	* Return: 0 on success, negative error code if an index failed validation.
2261	*/
2262	static int elf_validity_cache_index(struct load_info *info, int flags)
2263	{
2264	int err;
2265
2266	err = elf_validity_cache_index_info(info);
2267	if (err < 0)
2268	return err;
2269	err = elf_validity_cache_index_mod(info);
2270	if (err < 0)
2271	return err;
2272	err = elf_validity_cache_index_sym(info);
2273	if (err < 0)
2274	return err;
2275	err = elf_validity_cache_index_str(info);
2276	if (err < 0)
2277	return err;
2278	err = elf_validity_cache_index_versions(info, flags);
2279	if (err < 0)
2280	return err;
2281
2282	info->index.pcpu = find_pcpusec(info);
2283
2284	return 0;
2285	}
2286
2287	/**
2288	* elf_validity_cache_strtab() - Validate and cache symbol string table
2289	* @info: Load info to read from and update.
2290	* Must have &load_info->sechdrs and &load_info->secstrings populated.
2291	* Must have &load_info->index populated.
2292	*
2293	* Checks:
2294	*
2295	* * The string table is not empty.
2296	* * The string table starts and ends with NUL (required by ELF spec).
2297	* * Every &Elf_Sym->st_name offset in the symbol table is inbounds of the
2298	* string table.
2299	*
2300	* And caches the pointer as &load_info->strtab in @info.
2301	*
2302	* Return: 0 on success, negative error code if a check failed.
2303	*/
2304	static int elf_validity_cache_strtab(struct load_info *info)
2305	{
2306	Elf_Shdr *str_shdr = &info->sechdrs[info->index.str];
2307	Elf_Shdr *sym_shdr = &info->sechdrs[info->index.sym];
2308	char *strtab = (char *)info->hdr + str_shdr->sh_offset;
2309	Elf_Sym *syms = (void *)info->hdr + sym_shdr->sh_offset;
2310	int i;
2311
2312	if (str_shdr->sh_size == 0) {
2313	pr_err("empty symbol string table\n");
2314	return -ENOEXEC;
2315	}
2316	if (strtab[0] != '\0') {
2317	pr_err("symbol string table missing leading NUL\n");
2318	return -ENOEXEC;
2319	}
2320	if (strtab[str_shdr->sh_size - 1] != '\0') {
2321	pr_err("symbol string table isn't NUL terminated\n");
2322	return -ENOEXEC;
2323	}
2324
2325	/*
2326	* Now that we know strtab is correctly structured, check symbol
2327	* starts are inbounds before they're used later.
2328	*/
2329	for (i = 0; i < sym_shdr->sh_size / sizeof(*syms); i++) {
2330	if (syms[i].st_name >= str_shdr->sh_size) {
2331	pr_err("symbol name out of bounds in string table");
2332	return -ENOEXEC;
2333	}
2334	}
2335
2336	info->strtab = strtab;
2337	return 0;
2338	}
2339
2340	/*
2341	* Check userspace passed ELF module against our expectations, and cache
2342	* useful variables for further processing as we go.
2343	*
2344	* This does basic validity checks against section offsets and sizes, the
2345	* section name string table, and the indices used for it (sh_name).
2346	*
2347	* As a last step, since we're already checking the ELF sections we cache
2348	* useful variables which will be used later for our convenience:
2349	*
2350	* o pointers to section headers
2351	* o cache the modinfo symbol section
2352	* o cache the string symbol section
2353	* o cache the module section
2354	*
2355	* As a last step we set info->mod to the temporary copy of the module in
2356	* info->hdr. The final one will be allocated in move_module(). Any
2357	* modifications we make to our copy of the module will be carried over
2358	* to the final minted module.
2359	*/
2360	static int elf_validity_cache_copy(struct load_info *info, int flags)
2361	{
2362	int err;
2363
2364	err = elf_validity_cache_sechdrs(info);
2365	if (err < 0)
2366	return err;
2367	err = elf_validity_cache_secstrings(info);
2368	if (err < 0)
2369	return err;
2370	err = elf_validity_cache_index(info, flags);
2371	if (err < 0)
2372	return err;
2373	err = elf_validity_cache_strtab(info);
2374	if (err < 0)
2375	return err;
2376
2377	/* This is temporary: point mod into copy of data. */
2378	info->mod = (void *)info->hdr + info->sechdrs[info->index.mod].sh_offset;
2379
2380	/*
2381	* If we didn't load the .modinfo 'name' field earlier, fall back to
2382	* on-disk struct mod 'name' field.
2383	*/
2384	if (!info->name)
2385	info->name = info->mod->name;
2386
2387	return 0;
2388	}
2389
2390	#define COPY_CHUNK_SIZE (16*PAGE_SIZE)
2391
2392	static int copy_chunked_from_user(void *dst, const void __user *usrc, unsigned long len)
2393	{
2394	do {
2395	unsigned long n = min(len, COPY_CHUNK_SIZE);
2396
2397	if (copy_from_user(to: dst, from: usrc, n) != 0)
2398	return -EFAULT;
2399	cond_resched();
2400	dst += n;
2401	usrc += n;
2402	len -= n;
2403	} while (len);
2404	return 0;
2405	}
2406
2407	static int check_modinfo_livepatch(struct module *mod, struct load_info *info)
2408	{
2409	if (!get_modinfo(info, tag: "livepatch"))
2410	/* Nothing more to do */
2411	return 0;
2412
2413	if (set_livepatch_module(mod))
2414	return 0;
2415
2416	pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
2417	mod->name);
2418	return -ENOEXEC;
2419	}
2420
2421	static void check_modinfo_retpoline(struct module *mod, struct load_info *info)
2422	{
2423	if (retpoline_module_ok(has_retpoline: get_modinfo(info, tag: "retpoline")))
2424	return;
2425
2426	pr_warn("%s: loading module not compiled with retpoline compiler.\n",
2427	mod->name);
2428	}
2429
2430	/* Sets info->hdr and info->len. */
2431	static int copy_module_from_user(const void __user *umod, unsigned long len,
2432	struct load_info *info)
2433	{
2434	int err;
2435
2436	info->len = len;
2437	if (info->len < sizeof(*(info->hdr)))
2438	return -ENOEXEC;
2439
2440	err = security_kernel_load_data(id: LOADING_MODULE, contents: true);
2441	if (err)
2442	return err;
2443
2444	/* Suck in entire file: we'll want most of it. */
2445	info->hdr = __vmalloc(info->len, GFP_KERNEL | __GFP_NOWARN);
2446	if (!info->hdr)
2447	return -ENOMEM;
2448
2449	if (copy_chunked_from_user(dst: info->hdr, usrc: umod, len: info->len) != 0) {
2450	err = -EFAULT;
2451	goto out;
2452	}
2453
2454	err = security_kernel_post_load_data(buf: (char *)info->hdr, size: info->len,
2455	id: LOADING_MODULE, description: "init_module");
2456	out:
2457	if (err)
2458	vfree(addr: info->hdr);
2459
2460	return err;
2461	}
2462
2463	static void free_copy(struct load_info *info, int flags)
2464	{
2465	if (flags & MODULE_INIT_COMPRESSED_FILE)
2466	module_decompress_cleanup(info);
2467	else
2468	vfree(addr: info->hdr);
2469	}
2470
2471	static int rewrite_section_headers(struct load_info *info, int flags)
2472	{
2473	unsigned int i;
2474
2475	/* This should always be true, but let's be sure. */
2476	info->sechdrs[0].sh_addr = 0;
2477
2478	for (i = 1; i < info->hdr->e_shnum; i++) {
2479	Elf_Shdr *shdr = &info->sechdrs[i];
2480
2481	/*
2482	* Mark all sections sh_addr with their address in the
2483	* temporary image.
2484	*/
2485	shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2486
2487	}
2488
2489	/* Track but don't keep modinfo and version sections. */
2490	info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2491	info->sechdrs[info->index.vers_ext_crc].sh_flags &=
2492	~(unsigned long)SHF_ALLOC;
2493	info->sechdrs[info->index.vers_ext_name].sh_flags &=
2494	~(unsigned long)SHF_ALLOC;
2495	info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2496
2497	return 0;
2498	}
2499
2500	static const char *const module_license_offenders[] = {
2501	/* driverloader was caught wrongly pretending to be under GPL */
2502	"driverloader",
2503
2504	/* lve claims to be GPL but upstream won't provide source */
2505	"lve",
2506	};
2507
2508	/*
2509	* These calls taint the kernel depending certain module circumstances */
2510	static void module_augment_kernel_taints(struct module *mod, struct load_info *info)
2511	{
2512	int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2513	size_t i;
2514
2515	if (!get_modinfo(info, tag: "intree")) {
2516	if (!test_taint(TAINT_OOT_MODULE))
2517	pr_warn("%s: loading out-of-tree module taints kernel.\n",
2518	mod->name);
2519	add_taint_module(mod, TAINT_OOT_MODULE, lockdep_ok: LOCKDEP_STILL_OK);
2520	}
2521
2522	check_modinfo_retpoline(mod, info);
2523
2524	if (get_modinfo(info, tag: "staging")) {
2525	add_taint_module(mod, TAINT_CRAP, lockdep_ok: LOCKDEP_STILL_OK);
2526	pr_warn("%s: module is from the staging directory, the quality "
2527	"is unknown, you have been warned.\n", mod->name);
2528	}
2529
2530	if (is_livepatch_module(mod)) {
2531	add_taint_module(mod, TAINT_LIVEPATCH, lockdep_ok: LOCKDEP_STILL_OK);
2532	pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2533	mod->name);
2534	}
2535
2536	module_license_taint_check(mod, license: get_modinfo(info, tag: "license"));
2537
2538	if (get_modinfo(info, tag: "test")) {
2539	if (!test_taint(TAINT_TEST))
2540	pr_warn("%s: loading test module taints kernel.\n",
2541	mod->name);
2542	add_taint_module(mod, TAINT_TEST, lockdep_ok: LOCKDEP_STILL_OK);
2543	}
2544	#ifdef CONFIG_MODULE_SIG
2545	mod->sig_ok = info->sig_ok;
2546	if (!mod->sig_ok) {
2547	pr_notice_once("%s: module verification failed: signature "
2548	"and/or required key missing - tainting "
2549	"kernel\n", mod->name);
2550	add_taint_module(mod, TAINT_UNSIGNED_MODULE, lockdep_ok: LOCKDEP_STILL_OK);
2551	}
2552	#endif
2553
2554	/*
2555	* ndiswrapper is under GPL by itself, but loads proprietary modules.
2556	* Don't use add_taint_module(), as it would prevent ndiswrapper from
2557	* using GPL-only symbols it needs.
2558	*/
2559	if (strcmp(mod->name, "ndiswrapper") == 0)
2560	add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2561
2562	for (i = 0; i < ARRAY_SIZE(module_license_offenders); ++i) {
2563	if (strcmp(mod->name, module_license_offenders[i]) == 0)
2564	add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2565	lockdep_ok: LOCKDEP_NOW_UNRELIABLE);
2566	}
2567
2568	if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2569	pr_warn("%s: module license taints kernel.\n", mod->name);
2570
2571	}
2572
2573	static int check_modinfo(struct module *mod, struct load_info *info, int flags)
2574	{
2575	const char *modmagic = get_modinfo(info, tag: "vermagic");
2576	int err;
2577
2578	if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2579	modmagic = NULL;
2580
2581	/* This is allowed: modprobe --force will invalidate it. */
2582	if (!modmagic) {
2583	err = try_to_force_load(mod, reason: "bad vermagic");
2584	if (err)
2585	return err;
2586	} else if (!same_magic(amagic: modmagic, bmagic: vermagic, has_crcs: info->index.vers)) {
2587	pr_err("%s: version magic '%s' should be '%s'\n",
2588	info->name, modmagic, vermagic);
2589	return -ENOEXEC;
2590	}
2591
2592	err = check_modinfo_livepatch(mod, info);
2593	if (err)
2594	return err;
2595
2596	return 0;
2597	}
2598
2599	static int find_module_sections(struct module *mod, struct load_info *info)
2600	{
2601	mod->kp = section_objs(info, name: "__param",
2602	object_size: sizeof(*mod->kp), num: &mod->num_kp);
2603	mod->syms = section_objs(info, name: "__ksymtab",
2604	object_size: sizeof(*mod->syms), num: &mod->num_syms);
2605	mod->crcs = section_addr(info, name: "__kcrctab");
2606	mod->gpl_syms = section_objs(info, name: "__ksymtab_gpl",
2607	object_size: sizeof(*mod->gpl_syms),
2608	num: &mod->num_gpl_syms);
2609	mod->gpl_crcs = section_addr(info, name: "__kcrctab_gpl");
2610
2611	#ifdef CONFIG_CONSTRUCTORS
2612	mod->ctors = section_objs(info, name: ".ctors",
2613	object_size: sizeof(*mod->ctors), num: &mod->num_ctors);
2614	if (!mod->ctors)
2615	mod->ctors = section_objs(info, name: ".init_array",
2616	object_size: sizeof(*mod->ctors), num: &mod->num_ctors);
2617	else if (find_sec(info, name: ".init_array")) {
2618	/*
2619	* This shouldn't happen with same compiler and binutils
2620	* building all parts of the module.
2621	*/
2622	pr_warn("%s: has both .ctors and .init_array.\n",
2623	mod->name);
2624	return -EINVAL;
2625	}
2626	#endif
2627
2628	mod->noinstr_text_start = section_objs(info, name: ".noinstr.text", object_size: 1,
2629	num: &mod->noinstr_text_size);
2630
2631	#ifdef CONFIG_TRACEPOINTS
2632	mod->tracepoints_ptrs = section_objs(info, name: "__tracepoints_ptrs",
2633	object_size: sizeof(*mod->tracepoints_ptrs),
2634	num: &mod->num_tracepoints);
2635	#endif
2636	#ifdef CONFIG_TREE_SRCU
2637	mod->srcu_struct_ptrs = section_objs(info, name: "___srcu_struct_ptrs",
2638	object_size: sizeof(*mod->srcu_struct_ptrs),
2639	num: &mod->num_srcu_structs);
2640	#endif
2641	#ifdef CONFIG_BPF_EVENTS
2642	mod->bpf_raw_events = section_objs(info, name: "__bpf_raw_tp_map",
2643	object_size: sizeof(*mod->bpf_raw_events),
2644	num: &mod->num_bpf_raw_events);
2645	#endif
2646	#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2647	mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2648	mod->btf_base_data = any_section_objs(info, ".BTF.base", 1,
2649	&mod->btf_base_data_size);
2650	#endif
2651	#ifdef CONFIG_JUMP_LABEL
2652	mod->jump_entries = section_objs(info, name: "__jump_table",
2653	object_size: sizeof(*mod->jump_entries),
2654	num: &mod->num_jump_entries);
2655	#endif
2656	#ifdef CONFIG_EVENT_TRACING
2657	mod->trace_events = section_objs(info, name: "_ftrace_events",
2658	object_size: sizeof(*mod->trace_events),
2659	num: &mod->num_trace_events);
2660	mod->trace_evals = section_objs(info, name: "_ftrace_eval_map",
2661	object_size: sizeof(*mod->trace_evals),
2662	num: &mod->num_trace_evals);
2663	#endif
2664	#ifdef CONFIG_TRACING
2665	mod->trace_bprintk_fmt_start = section_objs(info, name: "__trace_printk_fmt",
2666	object_size: sizeof(*mod->trace_bprintk_fmt_start),
2667	num: &mod->num_trace_bprintk_fmt);
2668	#endif
2669	#ifdef CONFIG_DYNAMIC_FTRACE
2670	/* sechdrs[0].sh_size is always zero */
2671	mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2672	object_size: sizeof(*mod->ftrace_callsites),
2673	num: &mod->num_ftrace_callsites);
2674	#endif
2675	#ifdef CONFIG_FUNCTION_ERROR_INJECTION
2676	mod->ei_funcs = section_objs(info, name: "_error_injection_whitelist",
2677	object_size: sizeof(*mod->ei_funcs),
2678	num: &mod->num_ei_funcs);
2679	#endif
2680	#ifdef CONFIG_KPROBES
2681	mod->kprobes_text_start = section_objs(info, name: ".kprobes.text", object_size: 1,
2682	num: &mod->kprobes_text_size);
2683	mod->kprobe_blacklist = section_objs(info, name: "_kprobe_blacklist",
2684	object_size: sizeof(unsigned long),
2685	num: &mod->num_kprobe_blacklist);
2686	#endif
2687	#ifdef CONFIG_PRINTK_INDEX
2688	mod->printk_index_start = section_objs(info, name: ".printk_index",
2689	object_size: sizeof(*mod->printk_index_start),
2690	num: &mod->printk_index_size);
2691	#endif
2692	#ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2693	mod->static_call_sites = section_objs(info, name: ".static_call_sites",
2694	object_size: sizeof(*mod->static_call_sites),
2695	num: &mod->num_static_call_sites);
2696	#endif
2697	#if IS_ENABLED(CONFIG_KUNIT)
2698	mod->kunit_suites = section_objs(info, name: ".kunit_test_suites",
2699	object_size: sizeof(*mod->kunit_suites),
2700	num: &mod->num_kunit_suites);
2701	mod->kunit_init_suites = section_objs(info, name: ".kunit_init_test_suites",
2702	object_size: sizeof(*mod->kunit_init_suites),
2703	num: &mod->num_kunit_init_suites);
2704	#endif
2705
2706	mod->extable = section_objs(info, name: "__ex_table",
2707	object_size: sizeof(*mod->extable), num: &mod->num_exentries);
2708
2709	if (section_addr(info, name: "__obsparm"))
2710	pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2711
2712	#ifdef CONFIG_DYNAMIC_DEBUG_CORE
2713	mod->dyndbg_info.descs = section_objs(info, name: "__dyndbg",
2714	object_size: sizeof(*mod->dyndbg_info.descs),
2715	num: &mod->dyndbg_info.num_descs);
2716	mod->dyndbg_info.classes = section_objs(info, name: "__dyndbg_classes",
2717	object_size: sizeof(*mod->dyndbg_info.classes),
2718	num: &mod->dyndbg_info.num_classes);
2719	#endif
2720
2721	return 0;
2722	}
2723
2724	static int move_module(struct module *mod, struct load_info *info)
2725	{
2726	int i, ret;
2727	enum mod_mem_type t = MOD_MEM_NUM_TYPES;
2728	bool codetag_section_found = false;
2729
2730	for_each_mod_mem_type(type) {
2731	if (!mod->mem[type].size) {
2732	mod->mem[type].base = NULL;
2733	continue;
2734	}
2735
2736	ret = module_memory_alloc(mod, type);
2737	if (ret) {
2738	t = type;
2739	goto out_err;
2740	}
2741	}
2742
2743	/* Transfer each section which specifies SHF_ALLOC */
2744	pr_debug("Final section addresses for %s:\n", mod->name);
2745	for (i = 0; i < info->hdr->e_shnum; i++) {
2746	void *dest;
2747	Elf_Shdr *shdr = &info->sechdrs[i];
2748	const char *sname;
2749
2750	if (!(shdr->sh_flags & SHF_ALLOC))
2751	continue;
2752
2753	sname = info->secstrings + shdr->sh_name;
2754	/*
2755	* Load codetag sections separately as they might still be used
2756	* after module unload.
2757	*/
2758	if (codetag_needs_module_section(mod, name: sname, size: shdr->sh_size)) {
2759	dest = codetag_alloc_module_section(mod, name: sname, size: shdr->sh_size,
2760	prepend: arch_mod_section_prepend(mod, section: i), align: shdr->sh_addralign);
2761	if (WARN_ON(!dest)) {
2762	ret = -EINVAL;
2763	goto out_err;
2764	}
2765	if (IS_ERR(ptr: dest)) {
2766	ret = PTR_ERR(ptr: dest);
2767	goto out_err;
2768	}
2769	codetag_section_found = true;
2770	} else {
2771	enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2772	unsigned long offset = shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK;
2773
2774	dest = mod->mem[type].base + offset;
2775	}
2776
2777	if (shdr->sh_type != SHT_NOBITS) {
2778	/*
2779	* Our ELF checker already validated this, but let's
2780	* be pedantic and make the goal clearer. We actually
2781	* end up copying over all modifications made to the
2782	* userspace copy of the entire struct module.
2783	*/
2784	if (i == info->index.mod &&
2785	(WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2786	ret = -ENOEXEC;
2787	goto out_err;
2788	}
2789	memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2790	}
2791	/*
2792	* Update the userspace copy's ELF section address to point to
2793	* our newly allocated memory as a pure convenience so that
2794	* users of info can keep taking advantage and using the newly
2795	* minted official memory area.
2796	*/
2797	shdr->sh_addr = (unsigned long)dest;
2798	pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2799	(long)shdr->sh_size, info->secstrings + shdr->sh_name);
2800	}
2801
2802	return 0;
2803	out_err:
2804	module_memory_restore_rox(mod);
2805	while (t--)
2806	module_memory_free(mod, type: t);
2807	if (codetag_section_found)
2808	codetag_free_module_sections(mod);
2809
2810	return ret;
2811	}
2812
2813	static int check_export_symbol_versions(struct module *mod)
2814	{
2815	#ifdef CONFIG_MODVERSIONS
2816	if ((mod->num_syms && !mod->crcs) ||
2817	(mod->num_gpl_syms && !mod->gpl_crcs)) {
2818	return try_to_force_load(mod,
2819	"no versions for exported symbols");
2820	}
2821	#endif
2822	return 0;
2823	}
2824
2825	static void flush_module_icache(const struct module *mod)
2826	{
2827	/*
2828	* Flush the instruction cache, since we've played with text.
2829	* Do it before processing of module parameters, so the module
2830	* can provide parameter accessor functions of its own.
2831	*/
2832	for_each_mod_mem_type(type) {
2833	const struct module_memory *mod_mem = &mod->mem[type];
2834
2835	if (mod_mem->size) {
2836	flush_icache_range(start: (unsigned long)mod_mem->base,
2837	end: (unsigned long)mod_mem->base + mod_mem->size);
2838	}
2839	}
2840	}
2841
2842	bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2843	{
2844	return true;
2845	}
2846
2847	int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2848	Elf_Shdr *sechdrs,
2849	char *secstrings,
2850	struct module *mod)
2851	{
2852	return 0;
2853	}
2854
2855	/* module_blacklist is a comma-separated list of module names */
2856	static char *module_blacklist;
2857	static bool blacklisted(const char *module_name)
2858	{
2859	const char *p;
2860	size_t len;
2861
2862	if (!module_blacklist)
2863	return false;
2864
2865	for (p = module_blacklist; *p; p += len) {
2866	len = strcspn(p, ",");
2867	if (strlen(module_name) == len && !memcmp(p: module_name, q: p, size: len))
2868	return true;
2869	if (p[len] == ',')
2870	len++;
2871	}
2872	return false;
2873	}
2874	core_param(module_blacklist, module_blacklist, charp, 0400);
2875
2876	static struct module *layout_and_allocate(struct load_info *info, int flags)
2877	{
2878	struct module *mod;
2879	int err;
2880
2881	/* Allow arches to frob section contents and sizes. */
2882	err = module_frob_arch_sections(hdr: info->hdr, sechdrs: info->sechdrs,
2883	secstrings: info->secstrings, mod: info->mod);
2884	if (err < 0)
2885	return ERR_PTR(error: err);
2886
2887	err = module_enforce_rwx_sections(hdr: info->hdr, sechdrs: info->sechdrs,
2888	secstrings: info->secstrings, mod: info->mod);
2889	if (err < 0)
2890	return ERR_PTR(error: err);
2891
2892	/* We will do a special allocation for per-cpu sections later. */
2893	info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2894
2895	/*
2896	* Mark relevant sections as SHF_RO_AFTER_INIT so layout_sections() can
2897	* put them in the right place.
2898	* Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2899	*/
2900	module_mark_ro_after_init(hdr: info->hdr, sechdrs: info->sechdrs, secstrings: info->secstrings);
2901
2902	/*
2903	* Determine total sizes, and put offsets in sh_entsize. For now
2904	* this is done generically; there doesn't appear to be any
2905	* special cases for the architectures.
2906	*/
2907	layout_sections(mod: info->mod, info);
2908	layout_symtab(mod: info->mod, info);
2909
2910	/* Allocate and move to the final place */
2911	err = move_module(mod: info->mod, info);
2912	if (err)
2913	return ERR_PTR(error: err);
2914
2915	/* Module has been copied to its final place now: return it. */
2916	mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2917	kmemleak_load_module(mod, info);
2918	codetag_module_replaced(mod: info->mod, new_mod: mod);
2919
2920	return mod;
2921	}
2922
2923	/* mod is no longer valid after this! */
2924	static void module_deallocate(struct module *mod, struct load_info *info)
2925	{
2926	percpu_modfree(mod);
2927	module_arch_freeing_init(mod);
2928	codetag_free_module_sections(mod);
2929
2930	free_mod_mem(mod);
2931	}
2932
2933	int __weak module_finalize(const Elf_Ehdr *hdr,
2934	const Elf_Shdr *sechdrs,
2935	struct module *me)
2936	{
2937	return 0;
2938	}
2939
2940	static int post_relocation(struct module *mod, const struct load_info *info)
2941	{
2942	/* Sort exception table now relocations are done. */
2943	sort_extable(start: mod->extable, finish: mod->extable + mod->num_exentries);
2944
2945	/* Copy relocated percpu area over. */
2946	percpu_modcopy(mod, from: (void *)info->sechdrs[info->index.pcpu].sh_addr,
2947	size: info->sechdrs[info->index.pcpu].sh_size);
2948
2949	/* Setup kallsyms-specific fields. */
2950	add_kallsyms(mod, info);
2951
2952	/* Arch-specific module finalizing. */
2953	return module_finalize(hdr: info->hdr, sechdrs: info->sechdrs, me: mod);
2954	}
2955
2956	/* Call module constructors. */
2957	static void do_mod_ctors(struct module *mod)
2958	{
2959	#ifdef CONFIG_CONSTRUCTORS
2960	unsigned long i;
2961
2962	for (i = 0; i < mod->num_ctors; i++)
2963	mod->ctors[i]();
2964	#endif
2965	}
2966
2967	/* For freeing module_init on success, in case kallsyms traversing */
2968	struct mod_initfree {
2969	struct llist_node node;
2970	void *init_text;
2971	void *init_data;
2972	void *init_rodata;
2973	};
2974
2975	static void do_free_init(struct work_struct *w)
2976	{
2977	struct llist_node *pos, *n, *list;
2978	struct mod_initfree *initfree;
2979
2980	list = llist_del_all(head: &init_free_list);
2981
2982	synchronize_rcu();
2983
2984	llist_for_each_safe(pos, n, list) {
2985	initfree = container_of(pos, struct mod_initfree, node);
2986	execmem_free(ptr: initfree->init_text);
2987	execmem_free(ptr: initfree->init_data);
2988	execmem_free(ptr: initfree->init_rodata);
2989	kfree(objp: initfree);
2990	}
2991	}
2992
2993	void flush_module_init_free_work(void)
2994	{
2995	flush_work(work: &init_free_wq);
2996	}
2997
2998	#undef MODULE_PARAM_PREFIX
2999	#define MODULE_PARAM_PREFIX "module."
3000	/* Default value for module->async_probe_requested */
3001	static bool async_probe;
3002	module_param(async_probe, bool, 0644);
3003
3004	/*
3005	* This is where the real work happens.
3006	*
3007	* Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
3008	* helper command 'lx-symbols'.
3009	*/
3010	static noinline int do_init_module(struct module *mod)
3011	{
3012	int ret = 0;
3013	struct mod_initfree *freeinit;
3014	#if defined(CONFIG_MODULE_STATS)
3015	unsigned int text_size = 0, total_size = 0;
3016
3017	for_each_mod_mem_type(type) {
3018	const struct module_memory *mod_mem = &mod->mem[type];
3019	if (mod_mem->size) {
3020	total_size += mod_mem->size;
3021	if (type == MOD_TEXT || type == MOD_INIT_TEXT)
3022	text_size += mod_mem->size;
3023	}
3024	}
3025	#endif
3026
3027	freeinit = kmalloc(sizeof(*freeinit), GFP_KERNEL);
3028	if (!freeinit) {
3029	ret = -ENOMEM;
3030	goto fail;
3031	}
3032	freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
3033	freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
3034	freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
3035
3036	do_mod_ctors(mod);
3037	/* Start the module */
3038	if (mod->init != NULL)
3039	ret = do_one_initcall(fn: mod->init);
3040	if (ret < 0) {
3041	goto fail_free_freeinit;
3042	}
3043	if (ret > 0) {
3044	pr_warn("%s: '%s'->init suspiciously returned %d, it should "
3045	"follow 0/-E convention\n"
3046	"%s: loading module anyway...\n",
3047	__func__, mod->name, ret, __func__);
3048	dump_stack();
3049	}
3050
3051	/* Now it's a first class citizen! */
3052	mod->state = MODULE_STATE_LIVE;
3053	blocking_notifier_call_chain(nh: &module_notify_list,
3054	val: MODULE_STATE_LIVE, v: mod);
3055
3056	/* Delay uevent until module has finished its init routine */
3057	kobject_uevent(kobj: &mod->mkobj.kobj, action: KOBJ_ADD);
3058
3059	/*
3060	* We need to finish all async code before the module init sequence
3061	* is done. This has potential to deadlock if synchronous module
3062	* loading is requested from async (which is not allowed!).
3063	*
3064	* See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
3065	* request_module() from async workers") for more details.
3066	*/
3067	if (!mod->async_probe_requested)
3068	async_synchronize_full();
3069
3070	ftrace_free_mem(mod, start: mod->mem[MOD_INIT_TEXT].base,
3071	end: mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
3072	mutex_lock(&module_mutex);
3073	/* Drop initial reference. */
3074	module_put(mod);
3075	trim_init_extable(m: mod);
3076	#ifdef CONFIG_KALLSYMS
3077	/* Switch to core kallsyms now init is done: kallsyms may be walking! */
3078	rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
3079	#endif
3080	ret = module_enable_rodata_ro_after_init(mod);
3081	if (ret)
3082	pr_warn("%s: module_enable_rodata_ro_after_init() returned %d, "
3083	"ro_after_init data might still be writable\n",
3084	mod->name, ret);
3085
3086	mod_tree_remove_init(mod);
3087	module_arch_freeing_init(mod);
3088	for_class_mod_mem_type(type, init) {
3089	mod->mem[type].base = NULL;
3090	mod->mem[type].size = 0;
3091	}
3092
3093	#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
3094	/* .BTF is not SHF_ALLOC and will get removed, so sanitize pointers */
3095	mod->btf_data = NULL;
3096	mod->btf_base_data = NULL;
3097	#endif
3098	/*
3099	* We want to free module_init, but be aware that kallsyms may be
3100	* walking this within an RCU read section. In all the failure paths, we
3101	* call synchronize_rcu(), but we don't want to slow down the success
3102	* path. execmem_free() cannot be called in an interrupt, so do the
3103	* work and call synchronize_rcu() in a work queue.
3104	*
3105	* Note that execmem_alloc() on most architectures creates W+X page
3106	* mappings which won't be cleaned up until do_free_init() runs. Any
3107	* code such as mark_rodata_ro() which depends on those mappings to
3108	* be cleaned up needs to sync with the queued work by invoking
3109	* flush_module_init_free_work().
3110	*/
3111	if (llist_add(new: &freeinit->node, head: &init_free_list))
3112	schedule_work(work: &init_free_wq);
3113
3114	mutex_unlock(lock: &module_mutex);
3115	wake_up_all(&module_wq);
3116
3117	mod_stat_add_long(text_size, &total_text_size);
3118	mod_stat_add_long(total_size, &total_mod_size);
3119
3120	mod_stat_inc(&modcount);
3121
3122	return 0;
3123
3124	fail_free_freeinit:
3125	kfree(objp: freeinit);
3126	fail:
3127	/* Try to protect us from buggy refcounters. */
3128	mod->state = MODULE_STATE_GOING;
3129	synchronize_rcu();
3130	module_put(mod);
3131	blocking_notifier_call_chain(nh: &module_notify_list,
3132	val: MODULE_STATE_GOING, v: mod);
3133	klp_module_going(mod);
3134	ftrace_release_mod(mod);
3135	free_module(mod);
3136	wake_up_all(&module_wq);
3137
3138	return ret;
3139	}
3140
3141	static int may_init_module(void)
3142	{
3143	if (!capable(CAP_SYS_MODULE) || modules_disabled)
3144	return -EPERM;
3145
3146	return 0;
3147	}
3148
3149	/* Is this module of this name done loading? No locks held. */
3150	static bool finished_loading(const char *name)
3151	{
3152	struct module *mod;
3153	bool ret;
3154
3155	/*
3156	* The module_mutex should not be a heavily contended lock;
3157	* if we get the occasional sleep here, we'll go an extra iteration
3158	* in the wait_event_interruptible(), which is harmless.
3159	*/
3160	sched_annotate_sleep();
3161	mutex_lock(&module_mutex);
3162	mod = find_module_all(name, strlen(name), even_unformed: true);
3163	ret = !mod || mod->state == MODULE_STATE_LIVE
3164	|| mod->state == MODULE_STATE_GOING;
3165	mutex_unlock(lock: &module_mutex);
3166
3167	return ret;
3168	}
3169
3170	/* Must be called with module_mutex held */
3171	static int module_patient_check_exists(const char *name,
3172	enum fail_dup_mod_reason reason)
3173	{
3174	struct module *old;
3175	int err = 0;
3176
3177	old = find_module_all(name, strlen(name), even_unformed: true);
3178	if (old == NULL)
3179	return 0;
3180
3181	if (old->state == MODULE_STATE_COMING ||
3182	old->state == MODULE_STATE_UNFORMED) {
3183	/* Wait in case it fails to load. */
3184	mutex_unlock(lock: &module_mutex);
3185	err = wait_event_interruptible(module_wq,
3186	finished_loading(name));
3187	mutex_lock(&module_mutex);
3188	if (err)
3189	return err;
3190
3191	/* The module might have gone in the meantime. */
3192	old = find_module_all(name, strlen(name), even_unformed: true);
3193	}
3194
3195	if (try_add_failed_module(name, reason))
3196	pr_warn("Could not add fail-tracking for module: %s\n", name);
3197
3198	/*
3199	* We are here only when the same module was being loaded. Do
3200	* not try to load it again right now. It prevents long delays
3201	* caused by serialized module load failures. It might happen
3202	* when more devices of the same type trigger load of
3203	* a particular module.
3204	*/
3205	if (old && old->state == MODULE_STATE_LIVE)
3206	return -EEXIST;
3207	return -EBUSY;
3208	}
3209
3210	/*
3211	* We try to place it in the list now to make sure it's unique before
3212	* we dedicate too many resources. In particular, temporary percpu
3213	* memory exhaustion.
3214	*/
3215	static int add_unformed_module(struct module *mod)
3216	{
3217	int err;
3218
3219	mod->state = MODULE_STATE_UNFORMED;
3220
3221	mutex_lock(&module_mutex);
3222	err = module_patient_check_exists(name: mod->name, reason: FAIL_DUP_MOD_LOAD);
3223	if (err)
3224	goto out;
3225
3226	mod_update_bounds(mod);
3227	list_add_rcu(new: &mod->list, head: &modules);
3228	mod_tree_insert(mod);
3229	err = 0;
3230
3231	out:
3232	mutex_unlock(lock: &module_mutex);
3233	return err;
3234	}
3235
3236	static int complete_formation(struct module *mod, struct load_info *info)
3237	{
3238	int err;
3239
3240	mutex_lock(&module_mutex);
3241
3242	/* Find duplicate symbols (must be called under lock). */
3243	err = verify_exported_symbols(mod);
3244	if (err < 0)
3245	goto out;
3246
3247	/* These rely on module_mutex for list integrity. */
3248	module_bug_finalize(info->hdr, info->sechdrs, mod);
3249	module_cfi_finalize(hdr: info->hdr, sechdrs: info->sechdrs, mod);
3250
3251	err = module_enable_rodata_ro(mod);
3252	if (err)
3253	goto out_strict_rwx;
3254	err = module_enable_data_nx(mod);
3255	if (err)
3256	goto out_strict_rwx;
3257	err = module_enable_text_rox(mod);
3258	if (err)
3259	goto out_strict_rwx;
3260
3261	/*
3262	* Mark state as coming so strong_try_module_get() ignores us,
3263	* but kallsyms etc. can see us.
3264	*/
3265	mod->state = MODULE_STATE_COMING;
3266	mutex_unlock(lock: &module_mutex);
3267
3268	return 0;
3269
3270	out_strict_rwx:
3271	module_bug_cleanup(mod);
3272	out:
3273	mutex_unlock(lock: &module_mutex);
3274	return err;
3275	}
3276
3277	static int prepare_coming_module(struct module *mod)
3278	{
3279	int err;
3280
3281	ftrace_module_enable(mod);
3282	err = klp_module_coming(mod);
3283	if (err)
3284	return err;
3285
3286	err = blocking_notifier_call_chain_robust(nh: &module_notify_list,
3287	val_up: MODULE_STATE_COMING, val_down: MODULE_STATE_GOING, v: mod);
3288	err = notifier_to_errno(ret: err);
3289	if (err)
3290	klp_module_going(mod);
3291
3292	return err;
3293	}
3294
3295	static int unknown_module_param_cb(char *param, char *val, const char *modname,
3296	void *arg)
3297	{
3298	struct module *mod = arg;
3299	int ret;
3300
3301	if (strcmp(param, "async_probe") == 0) {
3302	if (kstrtobool(s: val, res: &mod->async_probe_requested))
3303	mod->async_probe_requested = true;
3304	return 0;
3305	}
3306
3307	/* Check for magic 'dyndbg' arg */
3308	ret = ddebug_dyndbg_module_param_cb(param, val, modname);
3309	if (ret != 0)
3310	pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
3311	return 0;
3312	}
3313
3314	/* Module within temporary copy, this doesn't do any allocation */
3315	static int early_mod_check(struct load_info *info, int flags)
3316	{
3317	int err;
3318
3319	/*
3320	* Now that we know we have the correct module name, check
3321	* if it's blacklisted.
3322	*/
3323	if (blacklisted(module_name: info->name)) {
3324	pr_err("Module %s is blacklisted\n", info->name);
3325	return -EPERM;
3326	}
3327
3328	err = rewrite_section_headers(info, flags);
3329	if (err)
3330	return err;
3331
3332	/* Check module struct version now, before we try to use module. */
3333	if (!check_modstruct_version(info, mod: info->mod))
3334	return -ENOEXEC;
3335
3336	err = check_modinfo(mod: info->mod, info, flags);
3337	if (err)
3338	return err;
3339
3340	mutex_lock(&module_mutex);
3341	err = module_patient_check_exists(name: info->mod->name, reason: FAIL_DUP_MOD_BECOMING);
3342	mutex_unlock(lock: &module_mutex);
3343
3344	return err;
3345	}
3346
3347	/*
3348	* Allocate and load the module: note that size of section 0 is always
3349	* zero, and we rely on this for optional sections.
3350	*/
3351	static int load_module(struct load_info *info, const char __user *uargs,
3352	int flags)
3353	{
3354	struct module *mod;
3355	bool module_allocated = false;
3356	long err = 0;
3357	char *after_dashes;
3358
3359	/*
3360	* Do the signature check (if any) first. All that
3361	* the signature check needs is info->len, it does
3362	* not need any of the section info. That can be
3363	* set up later. This will minimize the chances
3364	* of a corrupt module causing problems before
3365	* we even get to the signature check.
3366	*
3367	* The check will also adjust info->len by stripping
3368	* off the sig length at the end of the module, making
3369	* checks against info->len more correct.
3370	*/
3371	err = module_sig_check(info, flags);
3372	if (err)
3373	goto free_copy;
3374
3375	/*
3376	* Do basic sanity checks against the ELF header and
3377	* sections. Cache useful sections and set the
3378	* info->mod to the userspace passed struct module.
3379	*/
3380	err = elf_validity_cache_copy(info, flags);
3381	if (err)
3382	goto free_copy;
3383
3384	err = early_mod_check(info, flags);
3385	if (err)
3386	goto free_copy;
3387
3388	/* Figure out module layout, and allocate all the memory. */
3389	mod = layout_and_allocate(info, flags);
3390	if (IS_ERR(ptr: mod)) {
3391	err = PTR_ERR(ptr: mod);
3392	goto free_copy;
3393	}
3394
3395	module_allocated = true;
3396
3397	audit_log_kern_module(name: info->name);
3398
3399	/* Reserve our place in the list. */
3400	err = add_unformed_module(mod);
3401	if (err)
3402	goto free_module;
3403
3404	/*
3405	* We are tainting your kernel if your module gets into
3406	* the modules linked list somehow.
3407	*/
3408	module_augment_kernel_taints(mod, info);
3409
3410	/* To avoid stressing percpu allocator, do this once we're unique. */
3411	err = percpu_modalloc(mod, info);
3412	if (err)
3413	goto unlink_mod;
3414
3415	/* Now module is in final location, initialize linked lists, etc. */
3416	err = module_unload_init(mod);
3417	if (err)
3418	goto unlink_mod;
3419
3420	init_param_lock(mod);
3421
3422	/*
3423	* Now we've got everything in the final locations, we can
3424	* find optional sections.
3425	*/
3426	err = find_module_sections(mod, info);
3427	if (err)
3428	goto free_unload;
3429
3430	err = check_export_symbol_versions(mod);
3431	if (err)
3432	goto free_unload;
3433
3434	/* Set up MODINFO_ATTR fields */
3435	err = setup_modinfo(mod, info);
3436	if (err)
3437	goto free_modinfo;
3438
3439	/* Fix up syms, so that st_value is a pointer to location. */
3440	err = simplify_symbols(mod, info);
3441	if (err < 0)
3442	goto free_modinfo;
3443
3444	err = apply_relocations(mod, info);
3445	if (err < 0)
3446	goto free_modinfo;
3447
3448	err = post_relocation(mod, info);
3449	if (err < 0)
3450	goto free_modinfo;
3451
3452	flush_module_icache(mod);
3453
3454	/* Now copy in args */
3455	mod->args = strndup_user(uargs, ~0UL >> 1);
3456	if (IS_ERR(ptr: mod->args)) {
3457	err = PTR_ERR(ptr: mod->args);
3458	goto free_arch_cleanup;
3459	}
3460
3461	init_build_id(mod, info);
3462
3463	/* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
3464	ftrace_module_init(mod);
3465
3466	/* Finally it's fully formed, ready to start executing. */
3467	err = complete_formation(mod, info);
3468	if (err)
3469	goto ddebug_cleanup;
3470
3471	err = prepare_coming_module(mod);
3472	if (err)
3473	goto bug_cleanup;
3474
3475	mod->async_probe_requested = async_probe;
3476
3477	/* Module is ready to execute: parsing args may do that. */
3478	after_dashes = parse_args(name: mod->name, args: mod->args, params: mod->kp, num: mod->num_kp,
3479	level_min: -32768, level_max: 32767, arg: mod,
3480	unknown: unknown_module_param_cb);
3481	if (IS_ERR(ptr: after_dashes)) {
3482	err = PTR_ERR(ptr: after_dashes);
3483	goto coming_cleanup;
3484	} else if (after_dashes) {
3485	pr_warn("%s: parameters '%s' after `--' ignored\n",
3486	mod->name, after_dashes);
3487	}
3488
3489	/* Link in to sysfs. */
3490	err = mod_sysfs_setup(mod, info, kparam: mod->kp, num_params: mod->num_kp);
3491	if (err < 0)
3492	goto coming_cleanup;
3493
3494	if (is_livepatch_module(mod)) {
3495	err = copy_module_elf(mod, info);
3496	if (err < 0)
3497	goto sysfs_cleanup;
3498	}
3499
3500	if (codetag_load_module(mod))
3501	goto sysfs_cleanup;
3502
3503	/* Get rid of temporary copy. */
3504	free_copy(info, flags);
3505
3506	/* Done! */
3507	trace_module_load(mod);
3508
3509	return do_init_module(mod);
3510
3511	sysfs_cleanup:
3512	mod_sysfs_teardown(mod);
3513	coming_cleanup:
3514	mod->state = MODULE_STATE_GOING;
3515	destroy_params(params: mod->kp, num: mod->num_kp);
3516	blocking_notifier_call_chain(nh: &module_notify_list,
3517	val: MODULE_STATE_GOING, v: mod);
3518	klp_module_going(mod);
3519	bug_cleanup:
3520	mod->state = MODULE_STATE_GOING;
3521	/* module_bug_cleanup needs module_mutex protection */
3522	mutex_lock(&module_mutex);
3523	module_bug_cleanup(mod);
3524	mutex_unlock(lock: &module_mutex);
3525
3526	ddebug_cleanup:
3527	ftrace_release_mod(mod);
3528	synchronize_rcu();
3529	kfree(objp: mod->args);
3530	free_arch_cleanup:
3531	module_arch_cleanup(mod);
3532	free_modinfo:
3533	free_modinfo(mod);
3534	free_unload:
3535	module_unload_free(mod);
3536	unlink_mod:
3537	mutex_lock(&module_mutex);
3538	/* Unlink carefully: kallsyms could be walking list. */
3539	list_del_rcu(entry: &mod->list);
3540	mod_tree_remove(mod);
3541	wake_up_all(&module_wq);
3542	/* Wait for RCU-sched synchronizing before releasing mod->list. */
3543	synchronize_rcu();
3544	mutex_unlock(lock: &module_mutex);
3545	free_module:
3546	mod_stat_bump_invalid(info, flags);
3547	/* Free lock-classes; relies on the preceding sync_rcu() */
3548	for_class_mod_mem_type(type, core_data) {
3549	lockdep_free_key_range(start: mod->mem[type].base,
3550	size: mod->mem[type].size);
3551	}
3552
3553	module_memory_restore_rox(mod);
3554	module_deallocate(mod, info);
3555	free_copy:
3556	/*
3557	* The info->len is always set. We distinguish between
3558	* failures once the proper module was allocated and
3559	* before that.
3560	*/
3561	if (!module_allocated) {
3562	audit_log_kern_module(name: info->name ? info->name : "?");
3563	mod_stat_bump_becoming(info, flags);
3564	}
3565	free_copy(info, flags);
3566	return err;
3567	}
3568
3569	SYSCALL_DEFINE3(init_module, void __user *, umod,
3570	unsigned long, len, const char __user *, uargs)
3571	{
3572	int err;
3573	struct load_info info = { };
3574
3575	err = may_init_module();
3576	if (err)
3577	return err;
3578
3579	pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3580	umod, len, uargs);
3581
3582	err = copy_module_from_user(umod, len, info: &info);
3583	if (err) {
3584	mod_stat_inc(&failed_kreads);
3585	mod_stat_add_long(len, &invalid_kread_bytes);
3586	return err;
3587	}
3588
3589	return load_module(info: &info, uargs, flags: 0);
3590	}
3591
3592	struct idempotent {
3593	const void *cookie;
3594	struct hlist_node entry;
3595	struct completion complete;
3596	int ret;
3597	};
3598
3599	#define IDEM_HASH_BITS 8
3600	static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3601	static DEFINE_SPINLOCK(idem_lock);
3602
3603	static bool idempotent(struct idempotent *u, const void *cookie)
3604	{
3605	int hash = hash_ptr(ptr: cookie, IDEM_HASH_BITS);
3606	struct hlist_head *head = idem_hash + hash;
3607	struct idempotent *existing;
3608	bool first;
3609
3610	u->ret = -EINTR;
3611	u->cookie = cookie;
3612	init_completion(x: &u->complete);
3613
3614	spin_lock(lock: &idem_lock);
3615	first = true;
3616	hlist_for_each_entry(existing, head, entry) {
3617	if (existing->cookie != cookie)
3618	continue;
3619	first = false;
3620	break;
3621	}
3622	hlist_add_head(n: &u->entry, h: idem_hash + hash);
3623	spin_unlock(lock: &idem_lock);
3624
3625	return !first;
3626	}
3627
3628	/*
3629	* We were the first one with 'cookie' on the list, and we ended
3630	* up completing the operation. We now need to walk the list,
3631	* remove everybody - which includes ourselves - fill in the return
3632	* value, and then complete the operation.
3633	*/
3634	static int idempotent_complete(struct idempotent *u, int ret)
3635	{
3636	const void *cookie = u->cookie;
3637	int hash = hash_ptr(ptr: cookie, IDEM_HASH_BITS);
3638	struct hlist_head *head = idem_hash + hash;
3639	struct hlist_node *next;
3640	struct idempotent *pos;
3641
3642	spin_lock(lock: &idem_lock);
3643	hlist_for_each_entry_safe(pos, next, head, entry) {
3644	if (pos->cookie != cookie)
3645	continue;
3646	hlist_del_init(n: &pos->entry);
3647	pos->ret = ret;
3648	complete(&pos->complete);
3649	}
3650	spin_unlock(lock: &idem_lock);
3651	return ret;
3652	}
3653
3654	/*
3655	* Wait for the idempotent worker.
3656	*
3657	* If we get interrupted, we need to remove ourselves from the
3658	* the idempotent list, and the completion may still come in.
3659	*
3660	* The 'idem_lock' protects against the race, and 'idem.ret' was
3661	* initialized to -EINTR and is thus always the right return
3662	* value even if the idempotent work then completes between
3663	* the wait_for_completion and the cleanup.
3664	*/
3665	static int idempotent_wait_for_completion(struct idempotent *u)
3666	{
3667	if (wait_for_completion_interruptible(x: &u->complete)) {
3668	spin_lock(lock: &idem_lock);
3669	if (!hlist_unhashed(h: &u->entry))
3670	hlist_del(n: &u->entry);
3671	spin_unlock(lock: &idem_lock);
3672	}
3673	return u->ret;
3674	}
3675
3676	static int init_module_from_file(struct file *f, const char __user * uargs, int flags)
3677	{
3678	bool compressed = !!(flags & MODULE_INIT_COMPRESSED_FILE);
3679	struct load_info info = { };
3680	void *buf = NULL;
3681	int len;
3682	int err;
3683
3684	len = kernel_read_file(file: f, offset: 0, buf: &buf, INT_MAX, NULL,
3685	id: compressed ? READING_MODULE_COMPRESSED :
3686	READING_MODULE);
3687	if (len < 0) {
3688	mod_stat_inc(&failed_kreads);
3689	return len;
3690	}
3691
3692	if (compressed) {
3693	err = module_decompress(info: &info, buf, size: len);
3694	vfree(addr: buf); /* compressed data is no longer needed */
3695	if (err) {
3696	mod_stat_inc(&failed_decompress);
3697	mod_stat_add_long(len, &invalid_decompress_bytes);
3698	return err;
3699	}
3700	err = security_kernel_post_read_file(file: f, buf: (char *)info.hdr, size: info.len,
3701	id: READING_MODULE);
3702	if (err) {
3703	mod_stat_inc(&failed_kreads);
3704	free_copy(info: &info, flags);
3705	return err;
3706	}
3707	} else {
3708	info.hdr = buf;
3709	info.len = len;
3710	}
3711
3712	return load_module(info: &info, uargs, flags);
3713	}
3714
3715	static int idempotent_init_module(struct file *f, const char __user * uargs, int flags)
3716	{
3717	struct idempotent idem;
3718
3719	if (!(f->f_mode & FMODE_READ))
3720	return -EBADF;
3721
3722	/* Are we the winners of the race and get to do this? */
3723	if (!idempotent(u: &idem, cookie: file_inode(f))) {
3724	int ret = init_module_from_file(f, uargs, flags);
3725	return idempotent_complete(u: &idem, ret);
3726	}
3727
3728	/*
3729	* Somebody else won the race and is loading the module.
3730	*/
3731	return idempotent_wait_for_completion(u: &idem);
3732	}
3733
3734	SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3735	{
3736	int err = may_init_module();
3737	if (err)
3738	return err;
3739
3740	pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3741
3742	if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3743	|MODULE_INIT_IGNORE_VERMAGIC
3744	|MODULE_INIT_COMPRESSED_FILE))
3745	return -EINVAL;
3746
3747	CLASS(fd, f)(fd);
3748	if (fd_empty(f))
3749	return -EBADF;
3750	return idempotent_init_module(fd_file(f), uargs, flags);
3751	}
3752
3753	/* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
3754	char *module_flags(struct module *mod, char *buf, bool show_state)
3755	{
3756	int bx = 0;
3757
3758	BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3759	if (!mod->taints && !show_state)
3760	goto out;
3761	if (mod->taints ||
3762	mod->state == MODULE_STATE_GOING ||
3763	mod->state == MODULE_STATE_COMING) {
3764	buf[bx++] = '(';
3765	bx += module_flags_taint(taints: mod->taints, buf: buf + bx);
3766	/* Show a - for module-is-being-unloaded */
3767	if (mod->state == MODULE_STATE_GOING && show_state)
3768	buf[bx++] = '-';
3769	/* Show a + for module-is-being-loaded */
3770	if (mod->state == MODULE_STATE_COMING && show_state)
3771	buf[bx++] = '+';
3772	buf[bx++] = ')';
3773	}
3774	out:
3775	buf[bx] = '\0';
3776
3777	return buf;
3778	}
3779
3780	/* Given an address, look for it in the module exception tables. */
3781	const struct exception_table_entry *search_module_extables(unsigned long addr)
3782	{
3783	struct module *mod;
3784
3785	guard(rcu)();
3786	mod = __module_address(addr);
3787	if (!mod)
3788	return NULL;
3789
3790	if (!mod->num_exentries)
3791	return NULL;
3792	/*
3793	* The address passed here belongs to a module that is currently
3794	* invoked (we are running inside it). Therefore its module::refcnt
3795	* needs already be >0 to ensure that it is not removed at this stage.
3796	* All other user need to invoke this function within a RCU read
3797	* section.
3798	*/
3799	return search_extable(base: mod->extable, num: mod->num_exentries, value: addr);
3800	}
3801
3802	/**
3803	* is_module_address() - is this address inside a module?
3804	* @addr: the address to check.
3805	*
3806	* See is_module_text_address() if you simply want to see if the address
3807	* is code (not data).
3808	*/
3809	bool is_module_address(unsigned long addr)
3810	{
3811	guard(rcu)();
3812	return __module_address(addr) != NULL;
3813	}
3814
3815	/**
3816	* __module_address() - get the module which contains an address.
3817	* @addr: the address.
3818	*
3819	* Must be called within RCU read section or module mutex held so that
3820	* module doesn't get freed during this.
3821	*/
3822	struct module *__module_address(unsigned long addr)
3823	{
3824	struct module *mod;
3825
3826	if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3827	goto lookup;
3828
3829	#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3830	if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3831	goto lookup;
3832	#endif
3833
3834	return NULL;
3835
3836	lookup:
3837	mod = mod_find(addr, tree: &mod_tree);
3838	if (mod) {
3839	BUG_ON(!within_module(addr, mod));
3840	if (mod->state == MODULE_STATE_UNFORMED)
3841	mod = NULL;
3842	}
3843	return mod;
3844	}
3845
3846	/**
3847	* is_module_text_address() - is this address inside module code?
3848	* @addr: the address to check.
3849	*
3850	* See is_module_address() if you simply want to see if the address is
3851	* anywhere in a module. See kernel_text_address() for testing if an
3852	* address corresponds to kernel or module code.
3853	*/
3854	bool is_module_text_address(unsigned long addr)
3855	{
3856	guard(rcu)();
3857	return __module_text_address(addr) != NULL;
3858	}
3859
3860	void module_for_each_mod(int(*func)(struct module *mod, void *data), void *data)
3861	{
3862	struct module *mod;
3863
3864	guard(rcu)();
3865	list_for_each_entry_rcu(mod, &modules, list) {
3866	if (mod->state == MODULE_STATE_UNFORMED)
3867	continue;
3868	if (func(mod, data))
3869	break;
3870	}
3871	}
3872
3873	/**
3874	* __module_text_address() - get the module whose code contains an address.
3875	* @addr: the address.
3876	*
3877	* Must be called within RCU read section or module mutex held so that
3878	* module doesn't get freed during this.
3879	*/
3880	struct module *__module_text_address(unsigned long addr)
3881	{
3882	struct module *mod = __module_address(addr);
3883	if (mod) {
3884	/* Make sure it's within the text section. */
3885	if (!within_module_mem_type(addr, mod, type: MOD_TEXT) &&
3886	!within_module_mem_type(addr, mod, type: MOD_INIT_TEXT))
3887	mod = NULL;
3888	}
3889	return mod;
3890	}
3891
3892	/* Don't grab lock, we're oopsing. */
3893	void print_modules(void)
3894	{
3895	struct module *mod;
3896	char buf[MODULE_FLAGS_BUF_SIZE];
3897
3898	printk(KERN_DEFAULT "Modules linked in:");
3899	/* Most callers should already have preempt disabled, but make sure */
3900	guard(rcu)();
3901	list_for_each_entry_rcu(mod, &modules, list) {
3902	if (mod->state == MODULE_STATE_UNFORMED)
3903	continue;
3904	pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3905	}
3906
3907	print_unloaded_tainted_modules();
3908	if (last_unloaded_module.name[0])
3909	pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3910	last_unloaded_module.taints);
3911	pr_cont("\n");
3912	}
3913
3914	#ifdef CONFIG_MODULE_DEBUGFS
3915	struct dentry *mod_debugfs_root;
3916
3917	static int module_debugfs_init(void)
3918	{
3919	mod_debugfs_root = debugfs_create_dir(name: "modules", NULL);
3920	return 0;
3921	}
3922	module_init(module_debugfs_init);
3923	#endif
3924