1 // SPDX-License-Identifier: GPL-2.0-or-later
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>
8 #define INCLUDE_VERMAGIC
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>
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/debugfs.h>
60 #include <uapi/linux/module.h>
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/module.h>
68 * 1) List of modules (also safely readable with preempt_disable),
69 * 2) module_use links,
70 * 3) mod_tree.addr_min/mod_tree.addr_max.
71 * (delete and add uses RCU list operations).
73 DEFINE_MUTEX(module_mutex);
76 /* Work queue for freeing init sections in success case */
77 static void do_free_init(struct work_struct *w);
78 static DECLARE_WORK(init_free_wq, do_free_init);
79 static LLIST_HEAD(init_free_list);
81 struct mod_tree_root mod_tree __cacheline_aligned = {
86 const struct kernel_symbol *start, *stop;
88 enum mod_license license;
92 * Bounds of module memory, for speeding up __module_address.
93 * Protected by module_mutex.
95 static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
96 unsigned int size, struct mod_tree_root *tree)
98 unsigned long min = (unsigned long)base;
99 unsigned long max = min + size;
101 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
102 if (mod_mem_type_is_core_data(type)) {
103 if (min < tree->data_addr_min)
104 tree->data_addr_min = min;
105 if (max > tree->data_addr_max)
106 tree->data_addr_max = max;
110 if (min < tree->addr_min)
111 tree->addr_min = min;
112 if (max > tree->addr_max)
113 tree->addr_max = max;
116 static void mod_update_bounds(struct module *mod)
118 for_each_mod_mem_type(type) {
119 struct module_memory *mod_mem = &mod->mem[type];
122 __mod_update_bounds(type, mod_mem->base, mod_mem->size, &mod_tree);
126 /* Block module loading/unloading? */
127 int modules_disabled;
128 core_param(nomodule, modules_disabled, bint, 0);
130 /* Waiting for a module to finish initializing? */
131 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
133 static BLOCKING_NOTIFIER_HEAD(module_notify_list);
135 int register_module_notifier(struct notifier_block *nb)
137 return blocking_notifier_chain_register(&module_notify_list, nb);
139 EXPORT_SYMBOL(register_module_notifier);
141 int unregister_module_notifier(struct notifier_block *nb)
143 return blocking_notifier_chain_unregister(&module_notify_list, nb);
145 EXPORT_SYMBOL(unregister_module_notifier);
148 * We require a truly strong try_module_get(): 0 means success.
149 * Otherwise an error is returned due to ongoing or failed
150 * initialization etc.
152 static inline int strong_try_module_get(struct module *mod)
154 BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
155 if (mod && mod->state == MODULE_STATE_COMING)
157 if (try_module_get(mod))
163 static inline void add_taint_module(struct module *mod, unsigned flag,
164 enum lockdep_ok lockdep_ok)
166 add_taint(flag, lockdep_ok);
167 set_bit(flag, &mod->taints);
171 * A thread that wants to hold a reference to a module only while it
172 * is running can call this to safely exit.
174 void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
179 EXPORT_SYMBOL(__module_put_and_kthread_exit);
181 /* Find a module section: 0 means not found. */
182 static unsigned int find_sec(const struct load_info *info, const char *name)
186 for (i = 1; i < info->hdr->e_shnum; i++) {
187 Elf_Shdr *shdr = &info->sechdrs[i];
188 /* Alloc bit cleared means "ignore it." */
189 if ((shdr->sh_flags & SHF_ALLOC)
190 && strcmp(info->secstrings + shdr->sh_name, name) == 0)
196 /* Find a module section, or NULL. */
197 static void *section_addr(const struct load_info *info, const char *name)
199 /* Section 0 has sh_addr 0. */
200 return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
203 /* Find a module section, or NULL. Fill in number of "objects" in section. */
204 static void *section_objs(const struct load_info *info,
209 unsigned int sec = find_sec(info, name);
211 /* Section 0 has sh_addr 0 and sh_size 0. */
212 *num = info->sechdrs[sec].sh_size / object_size;
213 return (void *)info->sechdrs[sec].sh_addr;
216 /* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
217 static unsigned int find_any_sec(const struct load_info *info, const char *name)
221 for (i = 1; i < info->hdr->e_shnum; i++) {
222 Elf_Shdr *shdr = &info->sechdrs[i];
223 if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
230 * Find a module section, or NULL. Fill in number of "objects" in section.
231 * Ignores SHF_ALLOC flag.
233 static __maybe_unused void *any_section_objs(const struct load_info *info,
238 unsigned int sec = find_any_sec(info, name);
240 /* Section 0 has sh_addr 0 and sh_size 0. */
241 *num = info->sechdrs[sec].sh_size / object_size;
242 return (void *)info->sechdrs[sec].sh_addr;
245 #ifndef CONFIG_MODVERSIONS
246 #define symversion(base, idx) NULL
248 #define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
251 static const char *kernel_symbol_name(const struct kernel_symbol *sym)
253 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
254 return offset_to_ptr(&sym->name_offset);
260 static const char *kernel_symbol_namespace(const struct kernel_symbol *sym)
262 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
263 if (!sym->namespace_offset)
265 return offset_to_ptr(&sym->namespace_offset);
267 return sym->namespace;
271 int cmp_name(const void *name, const void *sym)
273 return strcmp(name, kernel_symbol_name(sym));
276 static bool find_exported_symbol_in_section(const struct symsearch *syms,
277 struct module *owner,
278 struct find_symbol_arg *fsa)
280 struct kernel_symbol *sym;
282 if (!fsa->gplok && syms->license == GPL_ONLY)
285 sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
286 sizeof(struct kernel_symbol), cmp_name);
291 fsa->crc = symversion(syms->crcs, sym - syms->start);
293 fsa->license = syms->license;
299 * Find an exported symbol and return it, along with, (optional) crc and
300 * (optional) module which owns it. Needs preempt disabled or module_mutex.
302 bool find_symbol(struct find_symbol_arg *fsa)
304 static const struct symsearch arr[] = {
305 { __start___ksymtab, __stop___ksymtab, __start___kcrctab,
307 { __start___ksymtab_gpl, __stop___ksymtab_gpl,
308 __start___kcrctab_gpl,
314 module_assert_mutex_or_preempt();
316 for (i = 0; i < ARRAY_SIZE(arr); i++)
317 if (find_exported_symbol_in_section(&arr[i], NULL, fsa))
320 list_for_each_entry_rcu(mod, &modules, list,
321 lockdep_is_held(&module_mutex)) {
322 struct symsearch arr[] = {
323 { mod->syms, mod->syms + mod->num_syms, mod->crcs,
325 { mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
330 if (mod->state == MODULE_STATE_UNFORMED)
333 for (i = 0; i < ARRAY_SIZE(arr); i++)
334 if (find_exported_symbol_in_section(&arr[i], mod, fsa))
338 pr_debug("Failed to find symbol %s\n", fsa->name);
343 * Search for module by name: must hold module_mutex (or preempt disabled
344 * for read-only access).
346 struct module *find_module_all(const char *name, size_t len,
351 module_assert_mutex_or_preempt();
353 list_for_each_entry_rcu(mod, &modules, list,
354 lockdep_is_held(&module_mutex)) {
355 if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
357 if (strlen(mod->name) == len && !memcmp(mod->name, name, len))
363 struct module *find_module(const char *name)
365 return find_module_all(name, strlen(name), false);
370 static inline void __percpu *mod_percpu(struct module *mod)
375 static int percpu_modalloc(struct module *mod, struct load_info *info)
377 Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
378 unsigned long align = pcpusec->sh_addralign;
380 if (!pcpusec->sh_size)
383 if (align > PAGE_SIZE) {
384 pr_warn("%s: per-cpu alignment %li > %li\n",
385 mod->name, align, PAGE_SIZE);
389 mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
391 pr_warn("%s: Could not allocate %lu bytes percpu data\n",
392 mod->name, (unsigned long)pcpusec->sh_size);
395 mod->percpu_size = pcpusec->sh_size;
399 static void percpu_modfree(struct module *mod)
401 free_percpu(mod->percpu);
404 static unsigned int find_pcpusec(struct load_info *info)
406 return find_sec(info, ".data..percpu");
409 static void percpu_modcopy(struct module *mod,
410 const void *from, unsigned long size)
414 for_each_possible_cpu(cpu)
415 memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
418 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
425 list_for_each_entry_rcu(mod, &modules, list) {
426 if (mod->state == MODULE_STATE_UNFORMED)
428 if (!mod->percpu_size)
430 for_each_possible_cpu(cpu) {
431 void *start = per_cpu_ptr(mod->percpu, cpu);
432 void *va = (void *)addr;
434 if (va >= start && va < start + mod->percpu_size) {
436 *can_addr = (unsigned long) (va - start);
437 *can_addr += (unsigned long)
438 per_cpu_ptr(mod->percpu,
452 * is_module_percpu_address() - test whether address is from module static percpu
453 * @addr: address to test
455 * Test whether @addr belongs to module static percpu area.
457 * Return: %true if @addr is from module static percpu area
459 bool is_module_percpu_address(unsigned long addr)
461 return __is_module_percpu_address(addr, NULL);
464 #else /* ... !CONFIG_SMP */
466 static inline void __percpu *mod_percpu(struct module *mod)
470 static int percpu_modalloc(struct module *mod, struct load_info *info)
472 /* UP modules shouldn't have this section: ENOMEM isn't quite right */
473 if (info->sechdrs[info->index.pcpu].sh_size != 0)
477 static inline void percpu_modfree(struct module *mod)
480 static unsigned int find_pcpusec(struct load_info *info)
484 static inline void percpu_modcopy(struct module *mod,
485 const void *from, unsigned long size)
487 /* pcpusec should be 0, and size of that section should be 0. */
490 bool is_module_percpu_address(unsigned long addr)
495 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
500 #endif /* CONFIG_SMP */
502 #define MODINFO_ATTR(field) \
503 static void setup_modinfo_##field(struct module *mod, const char *s) \
505 mod->field = kstrdup(s, GFP_KERNEL); \
507 static ssize_t show_modinfo_##field(struct module_attribute *mattr, \
508 struct module_kobject *mk, char *buffer) \
510 return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field); \
512 static int modinfo_##field##_exists(struct module *mod) \
514 return mod->field != NULL; \
516 static void free_modinfo_##field(struct module *mod) \
521 static struct module_attribute modinfo_##field = { \
522 .attr = { .name = __stringify(field), .mode = 0444 }, \
523 .show = show_modinfo_##field, \
524 .setup = setup_modinfo_##field, \
525 .test = modinfo_##field##_exists, \
526 .free = free_modinfo_##field, \
529 MODINFO_ATTR(version);
530 MODINFO_ATTR(srcversion);
533 char name[MODULE_NAME_LEN + 1];
534 char taints[MODULE_FLAGS_BUF_SIZE];
535 } last_unloaded_module;
537 #ifdef CONFIG_MODULE_UNLOAD
539 EXPORT_TRACEPOINT_SYMBOL(module_get);
541 /* MODULE_REF_BASE is the base reference count by kmodule loader. */
542 #define MODULE_REF_BASE 1
544 /* Init the unload section of the module. */
545 static int module_unload_init(struct module *mod)
548 * Initialize reference counter to MODULE_REF_BASE.
549 * refcnt == 0 means module is going.
551 atomic_set(&mod->refcnt, MODULE_REF_BASE);
553 INIT_LIST_HEAD(&mod->source_list);
554 INIT_LIST_HEAD(&mod->target_list);
556 /* Hold reference count during initialization. */
557 atomic_inc(&mod->refcnt);
562 /* Does a already use b? */
563 static int already_uses(struct module *a, struct module *b)
565 struct module_use *use;
567 list_for_each_entry(use, &b->source_list, source_list) {
568 if (use->source == a)
571 pr_debug("%s does not use %s!\n", a->name, b->name);
577 * - we add 'a' as a "source", 'b' as a "target" of module use
578 * - the module_use is added to the list of 'b' sources (so
579 * 'b' can walk the list to see who sourced them), and of 'a'
580 * targets (so 'a' can see what modules it targets).
582 static int add_module_usage(struct module *a, struct module *b)
584 struct module_use *use;
586 pr_debug("Allocating new usage for %s.\n", a->name);
587 use = kmalloc(sizeof(*use), GFP_ATOMIC);
593 list_add(&use->source_list, &b->source_list);
594 list_add(&use->target_list, &a->target_list);
598 /* Module a uses b: caller needs module_mutex() */
599 static int ref_module(struct module *a, struct module *b)
603 if (b == NULL || already_uses(a, b))
606 /* If module isn't available, we fail. */
607 err = strong_try_module_get(b);
611 err = add_module_usage(a, b);
619 /* Clear the unload stuff of the module. */
620 static void module_unload_free(struct module *mod)
622 struct module_use *use, *tmp;
624 mutex_lock(&module_mutex);
625 list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
626 struct module *i = use->target;
627 pr_debug("%s unusing %s\n", mod->name, i->name);
629 list_del(&use->source_list);
630 list_del(&use->target_list);
633 mutex_unlock(&module_mutex);
636 #ifdef CONFIG_MODULE_FORCE_UNLOAD
637 static inline int try_force_unload(unsigned int flags)
639 int ret = (flags & O_TRUNC);
641 add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
645 static inline int try_force_unload(unsigned int flags)
649 #endif /* CONFIG_MODULE_FORCE_UNLOAD */
651 /* Try to release refcount of module, 0 means success. */
652 static int try_release_module_ref(struct module *mod)
656 /* Try to decrement refcnt which we set at loading */
657 ret = atomic_sub_return(MODULE_REF_BASE, &mod->refcnt);
660 /* Someone can put this right now, recover with checking */
661 ret = atomic_add_unless(&mod->refcnt, MODULE_REF_BASE, 0);
666 static int try_stop_module(struct module *mod, int flags, int *forced)
668 /* If it's not unused, quit unless we're forcing. */
669 if (try_release_module_ref(mod) != 0) {
670 *forced = try_force_unload(flags);
675 /* Mark it as dying. */
676 mod->state = MODULE_STATE_GOING;
682 * module_refcount() - return the refcount or -1 if unloading
683 * @mod: the module we're checking
686 * -1 if the module is in the process of unloading
687 * otherwise the number of references in the kernel to the module
689 int module_refcount(struct module *mod)
691 return atomic_read(&mod->refcnt) - MODULE_REF_BASE;
693 EXPORT_SYMBOL(module_refcount);
695 /* This exists whether we can unload or not */
696 static void free_module(struct module *mod);
698 SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
702 char name[MODULE_NAME_LEN];
703 char buf[MODULE_FLAGS_BUF_SIZE];
706 if (!capable(CAP_SYS_MODULE) || modules_disabled)
709 if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0)
711 name[MODULE_NAME_LEN-1] = '\0';
713 audit_log_kern_module(name);
715 if (mutex_lock_interruptible(&module_mutex) != 0)
718 mod = find_module(name);
724 if (!list_empty(&mod->source_list)) {
725 /* Other modules depend on us: get rid of them first. */
730 /* Doing init or already dying? */
731 if (mod->state != MODULE_STATE_LIVE) {
732 /* FIXME: if (force), slam module count damn the torpedoes */
733 pr_debug("%s already dying\n", mod->name);
738 /* If it has an init func, it must have an exit func to unload */
739 if (mod->init && !mod->exit) {
740 forced = try_force_unload(flags);
742 /* This module can't be removed */
748 ret = try_stop_module(mod, flags, &forced);
752 mutex_unlock(&module_mutex);
753 /* Final destruction now no one is using it. */
754 if (mod->exit != NULL)
756 blocking_notifier_call_chain(&module_notify_list,
757 MODULE_STATE_GOING, mod);
758 klp_module_going(mod);
759 ftrace_release_mod(mod);
761 async_synchronize_full();
763 /* Store the name and taints of the last unloaded module for diagnostic purposes */
764 strscpy(last_unloaded_module.name, mod->name, sizeof(last_unloaded_module.name));
765 strscpy(last_unloaded_module.taints, module_flags(mod, buf, false), sizeof(last_unloaded_module.taints));
768 /* someone could wait for the module in add_unformed_module() */
769 wake_up_all(&module_wq);
772 mutex_unlock(&module_mutex);
776 void __symbol_put(const char *symbol)
778 struct find_symbol_arg fsa = {
784 BUG_ON(!find_symbol(&fsa));
785 module_put(fsa.owner);
788 EXPORT_SYMBOL(__symbol_put);
790 /* Note this assumes addr is a function, which it currently always is. */
791 void symbol_put_addr(void *addr)
793 struct module *modaddr;
794 unsigned long a = (unsigned long)dereference_function_descriptor(addr);
796 if (core_kernel_text(a))
800 * Even though we hold a reference on the module; we still need to
801 * disable preemption in order to safely traverse the data structure.
804 modaddr = __module_text_address(a);
809 EXPORT_SYMBOL_GPL(symbol_put_addr);
811 static ssize_t show_refcnt(struct module_attribute *mattr,
812 struct module_kobject *mk, char *buffer)
814 return sprintf(buffer, "%i\n", module_refcount(mk->mod));
817 static struct module_attribute modinfo_refcnt =
818 __ATTR(refcnt, 0444, show_refcnt, NULL);
820 void __module_get(struct module *module)
823 atomic_inc(&module->refcnt);
824 trace_module_get(module, _RET_IP_);
827 EXPORT_SYMBOL(__module_get);
829 bool try_module_get(struct module *module)
834 /* Note: here, we can fail to get a reference */
835 if (likely(module_is_live(module) &&
836 atomic_inc_not_zero(&module->refcnt) != 0))
837 trace_module_get(module, _RET_IP_);
843 EXPORT_SYMBOL(try_module_get);
845 void module_put(struct module *module)
850 ret = atomic_dec_if_positive(&module->refcnt);
851 WARN_ON(ret < 0); /* Failed to put refcount */
852 trace_module_put(module, _RET_IP_);
855 EXPORT_SYMBOL(module_put);
857 #else /* !CONFIG_MODULE_UNLOAD */
858 static inline void module_unload_free(struct module *mod)
862 static int ref_module(struct module *a, struct module *b)
864 return strong_try_module_get(b);
867 static inline int module_unload_init(struct module *mod)
871 #endif /* CONFIG_MODULE_UNLOAD */
873 size_t module_flags_taint(unsigned long taints, char *buf)
878 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
879 if (taint_flags[i].module && test_bit(i, &taints))
880 buf[l++] = taint_flags[i].c_true;
886 static ssize_t show_initstate(struct module_attribute *mattr,
887 struct module_kobject *mk, char *buffer)
889 const char *state = "unknown";
891 switch (mk->mod->state) {
892 case MODULE_STATE_LIVE:
895 case MODULE_STATE_COMING:
898 case MODULE_STATE_GOING:
904 return sprintf(buffer, "%s\n", state);
907 static struct module_attribute modinfo_initstate =
908 __ATTR(initstate, 0444, show_initstate, NULL);
910 static ssize_t store_uevent(struct module_attribute *mattr,
911 struct module_kobject *mk,
912 const char *buffer, size_t count)
916 rc = kobject_synth_uevent(&mk->kobj, buffer, count);
917 return rc ? rc : count;
920 struct module_attribute module_uevent =
921 __ATTR(uevent, 0200, NULL, store_uevent);
923 static ssize_t show_coresize(struct module_attribute *mattr,
924 struct module_kobject *mk, char *buffer)
926 unsigned int size = mk->mod->mem[MOD_TEXT].size;
928 if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
929 for_class_mod_mem_type(type, core_data)
930 size += mk->mod->mem[type].size;
932 return sprintf(buffer, "%u\n", size);
935 static struct module_attribute modinfo_coresize =
936 __ATTR(coresize, 0444, show_coresize, NULL);
938 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
939 static ssize_t show_datasize(struct module_attribute *mattr,
940 struct module_kobject *mk, char *buffer)
942 unsigned int size = 0;
944 for_class_mod_mem_type(type, core_data)
945 size += mk->mod->mem[type].size;
946 return sprintf(buffer, "%u\n", size);
949 static struct module_attribute modinfo_datasize =
950 __ATTR(datasize, 0444, show_datasize, NULL);
953 static ssize_t show_initsize(struct module_attribute *mattr,
954 struct module_kobject *mk, char *buffer)
956 unsigned int size = 0;
958 for_class_mod_mem_type(type, init)
959 size += mk->mod->mem[type].size;
960 return sprintf(buffer, "%u\n", size);
963 static struct module_attribute modinfo_initsize =
964 __ATTR(initsize, 0444, show_initsize, NULL);
966 static ssize_t show_taint(struct module_attribute *mattr,
967 struct module_kobject *mk, char *buffer)
971 l = module_flags_taint(mk->mod->taints, buffer);
976 static struct module_attribute modinfo_taint =
977 __ATTR(taint, 0444, show_taint, NULL);
979 struct module_attribute *modinfo_attrs[] = {
985 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
990 #ifdef CONFIG_MODULE_UNLOAD
996 size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
998 static const char vermagic[] = VERMAGIC_STRING;
1000 int try_to_force_load(struct module *mod, const char *reason)
1002 #ifdef CONFIG_MODULE_FORCE_LOAD
1003 if (!test_taint(TAINT_FORCED_MODULE))
1004 pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1005 add_taint_module(mod, TAINT_FORCED_MODULE, LOCKDEP_NOW_UNRELIABLE);
1012 /* Parse tag=value strings from .modinfo section */
1013 char *module_next_tag_pair(char *string, unsigned long *secsize)
1015 /* Skip non-zero chars */
1018 if ((*secsize)-- <= 1)
1022 /* Skip any zero padding. */
1023 while (!string[0]) {
1025 if ((*secsize)-- <= 1)
1031 static char *get_next_modinfo(const struct load_info *info, const char *tag,
1035 unsigned int taglen = strlen(tag);
1036 Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1037 unsigned long size = infosec->sh_size;
1040 * get_modinfo() calls made before rewrite_section_headers()
1041 * must use sh_offset, as sh_addr isn't set!
1043 char *modinfo = (char *)info->hdr + infosec->sh_offset;
1046 size -= prev - modinfo;
1047 modinfo = module_next_tag_pair(prev, &size);
1050 for (p = modinfo; p; p = module_next_tag_pair(p, &size)) {
1051 if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1052 return p + taglen + 1;
1057 static char *get_modinfo(const struct load_info *info, const char *tag)
1059 return get_next_modinfo(info, tag, NULL);
1062 static int verify_namespace_is_imported(const struct load_info *info,
1063 const struct kernel_symbol *sym,
1066 const char *namespace;
1067 char *imported_namespace;
1069 namespace = kernel_symbol_namespace(sym);
1070 if (namespace && namespace[0]) {
1071 for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1072 if (strcmp(namespace, imported_namespace) == 0)
1075 #ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1080 "%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1081 mod->name, kernel_symbol_name(sym), namespace);
1082 #ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1089 static bool inherit_taint(struct module *mod, struct module *owner, const char *name)
1091 if (!owner || !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1094 if (mod->using_gplonly_symbols) {
1095 pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1096 mod->name, name, owner->name);
1100 if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1101 pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1102 mod->name, name, owner->name);
1103 set_bit(TAINT_PROPRIETARY_MODULE, &mod->taints);
1108 /* Resolve a symbol for this module. I.e. if we find one, record usage. */
1109 static const struct kernel_symbol *resolve_symbol(struct module *mod,
1110 const struct load_info *info,
1114 struct find_symbol_arg fsa = {
1116 .gplok = !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1122 * The module_mutex should not be a heavily contended lock;
1123 * if we get the occasional sleep here, we'll go an extra iteration
1124 * in the wait_event_interruptible(), which is harmless.
1126 sched_annotate_sleep();
1127 mutex_lock(&module_mutex);
1128 if (!find_symbol(&fsa))
1131 if (fsa.license == GPL_ONLY)
1132 mod->using_gplonly_symbols = true;
1134 if (!inherit_taint(mod, fsa.owner, name)) {
1139 if (!check_version(info, name, mod, fsa.crc)) {
1140 fsa.sym = ERR_PTR(-EINVAL);
1144 err = verify_namespace_is_imported(info, fsa.sym, mod);
1146 fsa.sym = ERR_PTR(err);
1150 err = ref_module(mod, fsa.owner);
1152 fsa.sym = ERR_PTR(err);
1157 /* We must make copy under the lock if we failed to get ref. */
1158 strncpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1160 mutex_unlock(&module_mutex);
1164 static const struct kernel_symbol *
1165 resolve_symbol_wait(struct module *mod,
1166 const struct load_info *info,
1169 const struct kernel_symbol *ksym;
1170 char owner[MODULE_NAME_LEN];
1172 if (wait_event_interruptible_timeout(module_wq,
1173 !IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1174 || PTR_ERR(ksym) != -EBUSY,
1176 pr_warn("%s: gave up waiting for init of module %s.\n",
1182 void __weak module_memfree(void *module_region)
1185 * This memory may be RO, and freeing RO memory in an interrupt is not
1186 * supported by vmalloc.
1188 WARN_ON(in_interrupt());
1189 vfree(module_region);
1192 void __weak module_arch_cleanup(struct module *mod)
1196 void __weak module_arch_freeing_init(struct module *mod)
1200 static bool mod_mem_use_vmalloc(enum mod_mem_type type)
1202 return IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC) &&
1203 mod_mem_type_is_core_data(type);
1206 static void *module_memory_alloc(unsigned int size, enum mod_mem_type type)
1208 if (mod_mem_use_vmalloc(type))
1209 return vzalloc(size);
1210 return module_alloc(size);
1213 static void module_memory_free(void *ptr, enum mod_mem_type type)
1215 if (mod_mem_use_vmalloc(type))
1218 module_memfree(ptr);
1221 static void free_mod_mem(struct module *mod)
1223 for_each_mod_mem_type(type) {
1224 struct module_memory *mod_mem = &mod->mem[type];
1226 if (type == MOD_DATA)
1229 /* Free lock-classes; relies on the preceding sync_rcu(). */
1230 lockdep_free_key_range(mod_mem->base, mod_mem->size);
1232 module_memory_free(mod_mem->base, type);
1235 /* MOD_DATA hosts mod, so free it at last */
1236 lockdep_free_key_range(mod->mem[MOD_DATA].base, mod->mem[MOD_DATA].size);
1237 module_memory_free(mod->mem[MOD_DATA].base, MOD_DATA);
1240 /* Free a module, remove from lists, etc. */
1241 static void free_module(struct module *mod)
1243 trace_module_free(mod);
1245 mod_sysfs_teardown(mod);
1248 * We leave it in list to prevent duplicate loads, but make sure
1249 * that noone uses it while it's being deconstructed.
1251 mutex_lock(&module_mutex);
1252 mod->state = MODULE_STATE_UNFORMED;
1253 mutex_unlock(&module_mutex);
1255 /* Arch-specific cleanup. */
1256 module_arch_cleanup(mod);
1258 /* Module unload stuff */
1259 module_unload_free(mod);
1261 /* Free any allocated parameters. */
1262 destroy_params(mod->kp, mod->num_kp);
1264 if (is_livepatch_module(mod))
1265 free_module_elf(mod);
1267 /* Now we can delete it from the lists */
1268 mutex_lock(&module_mutex);
1269 /* Unlink carefully: kallsyms could be walking list. */
1270 list_del_rcu(&mod->list);
1271 mod_tree_remove(mod);
1272 /* Remove this module from bug list, this uses list_del_rcu */
1273 module_bug_cleanup(mod);
1274 /* Wait for RCU-sched synchronizing before releasing mod->list and buglist. */
1276 if (try_add_tainted_module(mod))
1277 pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1279 mutex_unlock(&module_mutex);
1281 /* This may be empty, but that's OK */
1282 module_arch_freeing_init(mod);
1284 percpu_modfree(mod);
1289 void *__symbol_get(const char *symbol)
1291 struct find_symbol_arg fsa = {
1298 if (!find_symbol(&fsa))
1300 if (fsa.license != GPL_ONLY) {
1301 pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1305 if (strong_try_module_get(fsa.owner))
1308 return (void *)kernel_symbol_value(fsa.sym);
1313 EXPORT_SYMBOL_GPL(__symbol_get);
1316 * Ensure that an exported symbol [global namespace] does not already exist
1317 * in the kernel or in some other module's exported symbol table.
1319 * You must hold the module_mutex.
1321 static int verify_exported_symbols(struct module *mod)
1324 const struct kernel_symbol *s;
1326 const struct kernel_symbol *sym;
1329 { mod->syms, mod->num_syms },
1330 { mod->gpl_syms, mod->num_gpl_syms },
1333 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1334 for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1335 struct find_symbol_arg fsa = {
1336 .name = kernel_symbol_name(s),
1339 if (find_symbol(&fsa)) {
1340 pr_err("%s: exports duplicate symbol %s"
1342 mod->name, kernel_symbol_name(s),
1343 module_name(fsa.owner));
1351 static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1354 * On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1355 * before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1356 * i386 has a similar problem but may not deserve a fix.
1358 * If we ever have to ignore many symbols, consider refactoring the code to
1359 * only warn if referenced by a relocation.
1361 if (emachine == EM_386 || emachine == EM_X86_64)
1362 return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1366 /* Change all symbols so that st_value encodes the pointer directly. */
1367 static int simplify_symbols(struct module *mod, const struct load_info *info)
1369 Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1370 Elf_Sym *sym = (void *)symsec->sh_addr;
1371 unsigned long secbase;
1374 const struct kernel_symbol *ksym;
1376 for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1377 const char *name = info->strtab + sym[i].st_name;
1379 switch (sym[i].st_shndx) {
1381 /* Ignore common symbols */
1382 if (!strncmp(name, "__gnu_lto", 9))
1386 * We compiled with -fno-common. These are not
1387 * supposed to happen.
1389 pr_debug("Common symbol: %s\n", name);
1390 pr_warn("%s: please compile with -fno-common\n",
1396 /* Don't need to do anything */
1397 pr_debug("Absolute symbol: 0x%08lx %s\n",
1398 (long)sym[i].st_value, name);
1402 /* Livepatch symbols are resolved by livepatch */
1406 ksym = resolve_symbol_wait(mod, info, name);
1407 /* Ok if resolved. */
1408 if (ksym && !IS_ERR(ksym)) {
1409 sym[i].st_value = kernel_symbol_value(ksym);
1413 /* Ok if weak or ignored. */
1415 (ELF_ST_BIND(sym[i].st_info) == STB_WEAK ||
1416 ignore_undef_symbol(info->hdr->e_machine, name)))
1419 ret = PTR_ERR(ksym) ?: -ENOENT;
1420 pr_warn("%s: Unknown symbol %s (err %d)\n",
1421 mod->name, name, ret);
1425 /* Divert to percpu allocation if a percpu var. */
1426 if (sym[i].st_shndx == info->index.pcpu)
1427 secbase = (unsigned long)mod_percpu(mod);
1429 secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1430 sym[i].st_value += secbase;
1438 static int apply_relocations(struct module *mod, const struct load_info *info)
1443 /* Now do relocations. */
1444 for (i = 1; i < info->hdr->e_shnum; i++) {
1445 unsigned int infosec = info->sechdrs[i].sh_info;
1447 /* Not a valid relocation section? */
1448 if (infosec >= info->hdr->e_shnum)
1451 /* Don't bother with non-allocated sections */
1452 if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC))
1455 if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1456 err = klp_apply_section_relocs(mod, info->sechdrs,
1461 else if (info->sechdrs[i].sh_type == SHT_REL)
1462 err = apply_relocate(info->sechdrs, info->strtab,
1463 info->index.sym, i, mod);
1464 else if (info->sechdrs[i].sh_type == SHT_RELA)
1465 err = apply_relocate_add(info->sechdrs, info->strtab,
1466 info->index.sym, i, mod);
1473 /* Additional bytes needed by arch in front of individual sections */
1474 unsigned int __weak arch_mod_section_prepend(struct module *mod,
1475 unsigned int section)
1477 /* default implementation just returns zero */
1481 long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1482 Elf_Shdr *sechdr, unsigned int section)
1485 long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1487 mod->mem[type].size += arch_mod_section_prepend(mod, section);
1488 offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1489 mod->mem[type].size = offset + sechdr->sh_size;
1491 WARN_ON_ONCE(offset & mask);
1492 return offset | mask;
1495 bool module_init_layout_section(const char *sname)
1497 #ifndef CONFIG_MODULE_UNLOAD
1498 if (module_exit_section(sname))
1501 return module_init_section(sname);
1504 static void __layout_sections(struct module *mod, struct load_info *info, bool is_init)
1508 static const unsigned long masks[][2] = {
1510 * NOTE: all executable code must be the first section
1511 * in this array; otherwise modify the text_size
1512 * finder in the two loops below
1514 { SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1515 { SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1516 { SHF_RO_AFTER_INIT | SHF_ALLOC, ARCH_SHF_SMALL },
1517 { SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1518 { ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1520 static const int core_m_to_mem_type[] = {
1527 static const int init_m_to_mem_type[] = {
1535 for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1536 enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1538 for (i = 0; i < info->hdr->e_shnum; ++i) {
1539 Elf_Shdr *s = &info->sechdrs[i];
1540 const char *sname = info->secstrings + s->sh_name;
1542 if ((s->sh_flags & masks[m][0]) != masks[m][0]
1543 || (s->sh_flags & masks[m][1])
1544 || s->sh_entsize != ~0UL
1545 || is_init != module_init_layout_section(sname))
1548 if (WARN_ON_ONCE(type == MOD_INVALID))
1551 s->sh_entsize = module_get_offset_and_type(mod, type, s, i);
1552 pr_debug("\t%s\n", sname);
1558 * Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1559 * might -- code, read-only data, read-write data, small data. Tally
1560 * sizes, and place the offsets into sh_entsize fields: high bit means it
1563 static void layout_sections(struct module *mod, struct load_info *info)
1567 for (i = 0; i < info->hdr->e_shnum; i++)
1568 info->sechdrs[i].sh_entsize = ~0UL;
1570 pr_debug("Core section allocation order for %s:\n", mod->name);
1571 __layout_sections(mod, info, false);
1573 pr_debug("Init section allocation order for %s:\n", mod->name);
1574 __layout_sections(mod, info, true);
1577 static void module_license_taint_check(struct module *mod, const char *license)
1580 license = "unspecified";
1582 if (!license_is_gpl_compatible(license)) {
1583 if (!test_taint(TAINT_PROPRIETARY_MODULE))
1584 pr_warn("%s: module license '%s' taints kernel.\n",
1585 mod->name, license);
1586 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1587 LOCKDEP_NOW_UNRELIABLE);
1591 static void setup_modinfo(struct module *mod, struct load_info *info)
1593 struct module_attribute *attr;
1596 for (i = 0; (attr = modinfo_attrs[i]); i++) {
1598 attr->setup(mod, get_modinfo(info, attr->attr.name));
1602 static void free_modinfo(struct module *mod)
1604 struct module_attribute *attr;
1607 for (i = 0; (attr = modinfo_attrs[i]); i++) {
1613 void * __weak module_alloc(unsigned long size)
1615 return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END,
1616 GFP_KERNEL, PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS,
1617 NUMA_NO_NODE, __builtin_return_address(0));
1620 bool __weak module_init_section(const char *name)
1622 return strstarts(name, ".init");
1625 bool __weak module_exit_section(const char *name)
1627 return strstarts(name, ".exit");
1630 static int validate_section_offset(struct load_info *info, Elf_Shdr *shdr)
1632 #if defined(CONFIG_64BIT)
1633 unsigned long long secend;
1635 unsigned long secend;
1639 * Check for both overflow and offset/size being
1642 secend = shdr->sh_offset + shdr->sh_size;
1643 if (secend < shdr->sh_offset || secend > info->len)
1650 * Check userspace passed ELF module against our expectations, and cache
1651 * useful variables for further processing as we go.
1653 * This does basic validity checks against section offsets and sizes, the
1654 * section name string table, and the indices used for it (sh_name).
1656 * As a last step, since we're already checking the ELF sections we cache
1657 * useful variables which will be used later for our convenience:
1659 * o pointers to section headers
1660 * o cache the modinfo symbol section
1661 * o cache the string symbol section
1662 * o cache the module section
1664 * As a last step we set info->mod to the temporary copy of the module in
1665 * info->hdr. The final one will be allocated in move_module(). Any
1666 * modifications we make to our copy of the module will be carried over
1667 * to the final minted module.
1669 static int elf_validity_cache_copy(struct load_info *info, int flags)
1672 Elf_Shdr *shdr, *strhdr;
1674 unsigned int num_mod_secs = 0, mod_idx;
1675 unsigned int num_info_secs = 0, info_idx;
1676 unsigned int num_sym_secs = 0, sym_idx;
1678 if (info->len < sizeof(*(info->hdr))) {
1679 pr_err("Invalid ELF header len %lu\n", info->len);
1683 if (memcmp(info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1684 pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1687 if (info->hdr->e_type != ET_REL) {
1688 pr_err("Invalid ELF header type: %u != %u\n",
1689 info->hdr->e_type, ET_REL);
1692 if (!elf_check_arch(info->hdr)) {
1693 pr_err("Invalid architecture in ELF header: %u\n",
1694 info->hdr->e_machine);
1697 if (!module_elf_check_arch(info->hdr)) {
1698 pr_err("Invalid module architecture in ELF header: %u\n",
1699 info->hdr->e_machine);
1702 if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1703 pr_err("Invalid ELF section header size\n");
1708 * e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1709 * known and small. So e_shnum * sizeof(Elf_Shdr)
1710 * will not overflow unsigned long on any platform.
1712 if (info->hdr->e_shoff >= info->len
1713 || (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1714 info->len - info->hdr->e_shoff)) {
1715 pr_err("Invalid ELF section header overflow\n");
1719 info->sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1722 * Verify if the section name table index is valid.
1724 if (info->hdr->e_shstrndx == SHN_UNDEF
1725 || info->hdr->e_shstrndx >= info->hdr->e_shnum) {
1726 pr_err("Invalid ELF section name index: %d || e_shstrndx (%d) >= e_shnum (%d)\n",
1727 info->hdr->e_shstrndx, info->hdr->e_shstrndx,
1728 info->hdr->e_shnum);
1732 strhdr = &info->sechdrs[info->hdr->e_shstrndx];
1733 err = validate_section_offset(info, strhdr);
1735 pr_err("Invalid ELF section hdr(type %u)\n", strhdr->sh_type);
1740 * The section name table must be NUL-terminated, as required
1741 * by the spec. This makes strcmp and pr_* calls that access
1742 * strings in the section safe.
1744 info->secstrings = (void *)info->hdr + strhdr->sh_offset;
1745 if (strhdr->sh_size == 0) {
1746 pr_err("empty section name table\n");
1749 if (info->secstrings[strhdr->sh_size - 1] != '\0') {
1750 pr_err("ELF Spec violation: section name table isn't null terminated\n");
1755 * The code assumes that section 0 has a length of zero and
1756 * an addr of zero, so check for it.
1758 if (info->sechdrs[0].sh_type != SHT_NULL
1759 || info->sechdrs[0].sh_size != 0
1760 || info->sechdrs[0].sh_addr != 0) {
1761 pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1762 info->sechdrs[0].sh_type);
1766 for (i = 1; i < info->hdr->e_shnum; i++) {
1767 shdr = &info->sechdrs[i];
1768 switch (shdr->sh_type) {
1773 if (shdr->sh_link == SHN_UNDEF
1774 || shdr->sh_link >= info->hdr->e_shnum) {
1775 pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
1776 shdr->sh_link, shdr->sh_link,
1777 info->hdr->e_shnum);
1784 err = validate_section_offset(info, shdr);
1786 pr_err("Invalid ELF section in module (section %u type %u)\n",
1790 if (strcmp(info->secstrings + shdr->sh_name,
1791 ".gnu.linkonce.this_module") == 0) {
1794 } else if (strcmp(info->secstrings + shdr->sh_name,
1800 if (shdr->sh_flags & SHF_ALLOC) {
1801 if (shdr->sh_name >= strhdr->sh_size) {
1802 pr_err("Invalid ELF section name in module (section %u type %u)\n",
1811 if (num_info_secs > 1) {
1812 pr_err("Only one .modinfo section must exist.\n");
1814 } else if (num_info_secs == 1) {
1815 /* Try to find a name early so we can log errors with a module name */
1816 info->index.info = info_idx;
1817 info->name = get_modinfo(info, "name");
1820 if (num_sym_secs != 1) {
1821 pr_warn("%s: module has no symbols (stripped?)\n",
1822 info->name ?: "(missing .modinfo section or name field)");
1826 /* Sets internal symbols and strings. */
1827 info->index.sym = sym_idx;
1828 shdr = &info->sechdrs[sym_idx];
1829 info->index.str = shdr->sh_link;
1830 info->strtab = (char *)info->hdr + info->sechdrs[info->index.str].sh_offset;
1833 * The ".gnu.linkonce.this_module" ELF section is special. It is
1834 * what modpost uses to refer to __this_module and let's use rely
1835 * on THIS_MODULE to point to &__this_module properly. The kernel's
1836 * modpost declares it on each modules's *.mod.c file. If the struct
1837 * module of the kernel changes a full kernel rebuild is required.
1839 * We have a few expectaions for this special section, the following
1840 * code validates all this for us:
1842 * o Only one section must exist
1843 * o We expect the kernel to always have to allocate it: SHF_ALLOC
1844 * o The section size must match the kernel's run time's struct module
1847 if (num_mod_secs != 1) {
1848 pr_err("module %s: Only one .gnu.linkonce.this_module section must exist.\n",
1849 info->name ?: "(missing .modinfo section or name field)");
1853 shdr = &info->sechdrs[mod_idx];
1856 * This is already implied on the switch above, however let's be
1857 * pedantic about it.
1859 if (shdr->sh_type == SHT_NOBITS) {
1860 pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
1861 info->name ?: "(missing .modinfo section or name field)");
1865 if (!(shdr->sh_flags & SHF_ALLOC)) {
1866 pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
1867 info->name ?: "(missing .modinfo section or name field)");
1871 if (shdr->sh_size != sizeof(struct module)) {
1872 pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
1873 info->name ?: "(missing .modinfo section or name field)");
1877 info->index.mod = mod_idx;
1879 /* This is temporary: point mod into copy of data. */
1880 info->mod = (void *)info->hdr + shdr->sh_offset;
1883 * If we didn't load the .modinfo 'name' field earlier, fall back to
1884 * on-disk struct mod 'name' field.
1887 info->name = info->mod->name;
1889 if (flags & MODULE_INIT_IGNORE_MODVERSIONS)
1890 info->index.vers = 0; /* Pretend no __versions section! */
1892 info->index.vers = find_sec(info, "__versions");
1894 info->index.pcpu = find_pcpusec(info);
1902 #define COPY_CHUNK_SIZE (16*PAGE_SIZE)
1904 static int copy_chunked_from_user(void *dst, const void __user *usrc, unsigned long len)
1907 unsigned long n = min(len, COPY_CHUNK_SIZE);
1909 if (copy_from_user(dst, usrc, n) != 0)
1919 static int check_modinfo_livepatch(struct module *mod, struct load_info *info)
1921 if (!get_modinfo(info, "livepatch"))
1922 /* Nothing more to do */
1925 if (set_livepatch_module(mod))
1928 pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
1933 static void check_modinfo_retpoline(struct module *mod, struct load_info *info)
1935 if (retpoline_module_ok(get_modinfo(info, "retpoline")))
1938 pr_warn("%s: loading module not compiled with retpoline compiler.\n",
1942 /* Sets info->hdr and info->len. */
1943 static int copy_module_from_user(const void __user *umod, unsigned long len,
1944 struct load_info *info)
1949 if (info->len < sizeof(*(info->hdr)))
1952 err = security_kernel_load_data(LOADING_MODULE, true);
1956 /* Suck in entire file: we'll want most of it. */
1957 info->hdr = __vmalloc(info->len, GFP_KERNEL | __GFP_NOWARN);
1961 if (copy_chunked_from_user(info->hdr, umod, info->len) != 0) {
1966 err = security_kernel_post_load_data((char *)info->hdr, info->len,
1967 LOADING_MODULE, "init_module");
1975 static void free_copy(struct load_info *info, int flags)
1977 if (flags & MODULE_INIT_COMPRESSED_FILE)
1978 module_decompress_cleanup(info);
1983 static int rewrite_section_headers(struct load_info *info, int flags)
1987 /* This should always be true, but let's be sure. */
1988 info->sechdrs[0].sh_addr = 0;
1990 for (i = 1; i < info->hdr->e_shnum; i++) {
1991 Elf_Shdr *shdr = &info->sechdrs[i];
1994 * Mark all sections sh_addr with their address in the
1997 shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2001 /* Track but don't keep modinfo and version sections. */
2002 info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2003 info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2009 * These calls taint the kernel depending certain module circumstances */
2010 static void module_augment_kernel_taints(struct module *mod, struct load_info *info)
2012 int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2014 if (!get_modinfo(info, "intree")) {
2015 if (!test_taint(TAINT_OOT_MODULE))
2016 pr_warn("%s: loading out-of-tree module taints kernel.\n",
2018 add_taint_module(mod, TAINT_OOT_MODULE, LOCKDEP_STILL_OK);
2021 check_modinfo_retpoline(mod, info);
2023 if (get_modinfo(info, "staging")) {
2024 add_taint_module(mod, TAINT_CRAP, LOCKDEP_STILL_OK);
2025 pr_warn("%s: module is from the staging directory, the quality "
2026 "is unknown, you have been warned.\n", mod->name);
2029 if (is_livepatch_module(mod)) {
2030 add_taint_module(mod, TAINT_LIVEPATCH, LOCKDEP_STILL_OK);
2031 pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2035 module_license_taint_check(mod, get_modinfo(info, "license"));
2037 if (get_modinfo(info, "test")) {
2038 if (!test_taint(TAINT_TEST))
2039 pr_warn("%s: loading test module taints kernel.\n",
2041 add_taint_module(mod, TAINT_TEST, LOCKDEP_STILL_OK);
2043 #ifdef CONFIG_MODULE_SIG
2044 mod->sig_ok = info->sig_ok;
2046 pr_notice_once("%s: module verification failed: signature "
2047 "and/or required key missing - tainting "
2048 "kernel\n", mod->name);
2049 add_taint_module(mod, TAINT_UNSIGNED_MODULE, LOCKDEP_STILL_OK);
2054 * ndiswrapper is under GPL by itself, but loads proprietary modules.
2055 * Don't use add_taint_module(), as it would prevent ndiswrapper from
2056 * using GPL-only symbols it needs.
2058 if (strcmp(mod->name, "ndiswrapper") == 0)
2059 add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2061 /* driverloader was caught wrongly pretending to be under GPL */
2062 if (strcmp(mod->name, "driverloader") == 0)
2063 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2064 LOCKDEP_NOW_UNRELIABLE);
2066 /* lve claims to be GPL but upstream won't provide source */
2067 if (strcmp(mod->name, "lve") == 0)
2068 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2069 LOCKDEP_NOW_UNRELIABLE);
2071 if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2072 pr_warn("%s: module license taints kernel.\n", mod->name);
2076 static int check_modinfo(struct module *mod, struct load_info *info, int flags)
2078 const char *modmagic = get_modinfo(info, "vermagic");
2081 if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2084 /* This is allowed: modprobe --force will invalidate it. */
2086 err = try_to_force_load(mod, "bad vermagic");
2089 } else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2090 pr_err("%s: version magic '%s' should be '%s'\n",
2091 info->name, modmagic, vermagic);
2095 err = check_modinfo_livepatch(mod, info);
2102 static int find_module_sections(struct module *mod, struct load_info *info)
2104 mod->kp = section_objs(info, "__param",
2105 sizeof(*mod->kp), &mod->num_kp);
2106 mod->syms = section_objs(info, "__ksymtab",
2107 sizeof(*mod->syms), &mod->num_syms);
2108 mod->crcs = section_addr(info, "__kcrctab");
2109 mod->gpl_syms = section_objs(info, "__ksymtab_gpl",
2110 sizeof(*mod->gpl_syms),
2111 &mod->num_gpl_syms);
2112 mod->gpl_crcs = section_addr(info, "__kcrctab_gpl");
2114 #ifdef CONFIG_CONSTRUCTORS
2115 mod->ctors = section_objs(info, ".ctors",
2116 sizeof(*mod->ctors), &mod->num_ctors);
2118 mod->ctors = section_objs(info, ".init_array",
2119 sizeof(*mod->ctors), &mod->num_ctors);
2120 else if (find_sec(info, ".init_array")) {
2122 * This shouldn't happen with same compiler and binutils
2123 * building all parts of the module.
2125 pr_warn("%s: has both .ctors and .init_array.\n",
2131 mod->noinstr_text_start = section_objs(info, ".noinstr.text", 1,
2132 &mod->noinstr_text_size);
2134 #ifdef CONFIG_TRACEPOINTS
2135 mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2136 sizeof(*mod->tracepoints_ptrs),
2137 &mod->num_tracepoints);
2139 #ifdef CONFIG_TREE_SRCU
2140 mod->srcu_struct_ptrs = section_objs(info, "___srcu_struct_ptrs",
2141 sizeof(*mod->srcu_struct_ptrs),
2142 &mod->num_srcu_structs);
2144 #ifdef CONFIG_BPF_EVENTS
2145 mod->bpf_raw_events = section_objs(info, "__bpf_raw_tp_map",
2146 sizeof(*mod->bpf_raw_events),
2147 &mod->num_bpf_raw_events);
2149 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2150 mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2152 #ifdef CONFIG_JUMP_LABEL
2153 mod->jump_entries = section_objs(info, "__jump_table",
2154 sizeof(*mod->jump_entries),
2155 &mod->num_jump_entries);
2157 #ifdef CONFIG_EVENT_TRACING
2158 mod->trace_events = section_objs(info, "_ftrace_events",
2159 sizeof(*mod->trace_events),
2160 &mod->num_trace_events);
2161 mod->trace_evals = section_objs(info, "_ftrace_eval_map",
2162 sizeof(*mod->trace_evals),
2163 &mod->num_trace_evals);
2165 #ifdef CONFIG_TRACING
2166 mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2167 sizeof(*mod->trace_bprintk_fmt_start),
2168 &mod->num_trace_bprintk_fmt);
2170 #ifdef CONFIG_FTRACE_MCOUNT_RECORD
2171 /* sechdrs[0].sh_size is always zero */
2172 mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2173 sizeof(*mod->ftrace_callsites),
2174 &mod->num_ftrace_callsites);
2176 #ifdef CONFIG_FUNCTION_ERROR_INJECTION
2177 mod->ei_funcs = section_objs(info, "_error_injection_whitelist",
2178 sizeof(*mod->ei_funcs),
2179 &mod->num_ei_funcs);
2181 #ifdef CONFIG_KPROBES
2182 mod->kprobes_text_start = section_objs(info, ".kprobes.text", 1,
2183 &mod->kprobes_text_size);
2184 mod->kprobe_blacklist = section_objs(info, "_kprobe_blacklist",
2185 sizeof(unsigned long),
2186 &mod->num_kprobe_blacklist);
2188 #ifdef CONFIG_PRINTK_INDEX
2189 mod->printk_index_start = section_objs(info, ".printk_index",
2190 sizeof(*mod->printk_index_start),
2191 &mod->printk_index_size);
2193 #ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2194 mod->static_call_sites = section_objs(info, ".static_call_sites",
2195 sizeof(*mod->static_call_sites),
2196 &mod->num_static_call_sites);
2198 #if IS_ENABLED(CONFIG_KUNIT)
2199 mod->kunit_suites = section_objs(info, ".kunit_test_suites",
2200 sizeof(*mod->kunit_suites),
2201 &mod->num_kunit_suites);
2204 mod->extable = section_objs(info, "__ex_table",
2205 sizeof(*mod->extable), &mod->num_exentries);
2207 if (section_addr(info, "__obsparm"))
2208 pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2210 #ifdef CONFIG_DYNAMIC_DEBUG_CORE
2211 mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
2212 sizeof(*mod->dyndbg_info.descs),
2213 &mod->dyndbg_info.num_descs);
2214 mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
2215 sizeof(*mod->dyndbg_info.classes),
2216 &mod->dyndbg_info.num_classes);
2222 static int move_module(struct module *mod, struct load_info *info)
2226 enum mod_mem_type t = 0;
2229 for_each_mod_mem_type(type) {
2230 if (!mod->mem[type].size) {
2231 mod->mem[type].base = NULL;
2234 mod->mem[type].size = PAGE_ALIGN(mod->mem[type].size);
2235 ptr = module_memory_alloc(mod->mem[type].size, type);
2237 * The pointer to these blocks of memory are stored on the module
2238 * structure and we keep that around so long as the module is
2239 * around. We only free that memory when we unload the module.
2240 * Just mark them as not being a leak then. The .init* ELF
2241 * sections *do* get freed after boot so we *could* treat them
2242 * slightly differently with kmemleak_ignore() and only grey
2243 * them out as they work as typical memory allocations which
2244 * *do* eventually get freed, but let's just keep things simple
2245 * and avoid *any* false positives.
2247 kmemleak_not_leak(ptr);
2252 memset(ptr, 0, mod->mem[type].size);
2253 mod->mem[type].base = ptr;
2256 /* Transfer each section which specifies SHF_ALLOC */
2257 pr_debug("Final section addresses for %s:\n", mod->name);
2258 for (i = 0; i < info->hdr->e_shnum; i++) {
2260 Elf_Shdr *shdr = &info->sechdrs[i];
2261 enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2263 if (!(shdr->sh_flags & SHF_ALLOC))
2266 dest = mod->mem[type].base + (shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK);
2268 if (shdr->sh_type != SHT_NOBITS) {
2270 * Our ELF checker already validated this, but let's
2271 * be pedantic and make the goal clearer. We actually
2272 * end up copying over all modifications made to the
2273 * userspace copy of the entire struct module.
2275 if (i == info->index.mod &&
2276 (WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2280 memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2283 * Update the userspace copy's ELF section address to point to
2284 * our newly allocated memory as a pure convenience so that
2285 * users of info can keep taking advantage and using the newly
2286 * minted official memory area.
2288 shdr->sh_addr = (unsigned long)dest;
2289 pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2290 (long)shdr->sh_size, info->secstrings + shdr->sh_name);
2295 for (t--; t >= 0; t--)
2296 module_memory_free(mod->mem[t].base, t);
2300 static int check_export_symbol_versions(struct module *mod)
2302 #ifdef CONFIG_MODVERSIONS
2303 if ((mod->num_syms && !mod->crcs) ||
2304 (mod->num_gpl_syms && !mod->gpl_crcs)) {
2305 return try_to_force_load(mod,
2306 "no versions for exported symbols");
2312 static void flush_module_icache(const struct module *mod)
2315 * Flush the instruction cache, since we've played with text.
2316 * Do it before processing of module parameters, so the module
2317 * can provide parameter accessor functions of its own.
2319 for_each_mod_mem_type(type) {
2320 const struct module_memory *mod_mem = &mod->mem[type];
2322 if (mod_mem->size) {
2323 flush_icache_range((unsigned long)mod_mem->base,
2324 (unsigned long)mod_mem->base + mod_mem->size);
2329 bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2334 int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2342 /* module_blacklist is a comma-separated list of module names */
2343 static char *module_blacklist;
2344 static bool blacklisted(const char *module_name)
2349 if (!module_blacklist)
2352 for (p = module_blacklist; *p; p += len) {
2353 len = strcspn(p, ",");
2354 if (strlen(module_name) == len && !memcmp(module_name, p, len))
2361 core_param(module_blacklist, module_blacklist, charp, 0400);
2363 static struct module *layout_and_allocate(struct load_info *info, int flags)
2369 /* Allow arches to frob section contents and sizes. */
2370 err = module_frob_arch_sections(info->hdr, info->sechdrs,
2371 info->secstrings, info->mod);
2373 return ERR_PTR(err);
2375 err = module_enforce_rwx_sections(info->hdr, info->sechdrs,
2376 info->secstrings, info->mod);
2378 return ERR_PTR(err);
2380 /* We will do a special allocation for per-cpu sections later. */
2381 info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2384 * Mark ro_after_init section with SHF_RO_AFTER_INIT so that
2385 * layout_sections() can put it in the right place.
2386 * Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2388 ndx = find_sec(info, ".data..ro_after_init");
2390 info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2392 * Mark the __jump_table section as ro_after_init as well: these data
2393 * structures are never modified, with the exception of entries that
2394 * refer to code in the __init section, which are annotated as such
2395 * at module load time.
2397 ndx = find_sec(info, "__jump_table");
2399 info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2402 * Determine total sizes, and put offsets in sh_entsize. For now
2403 * this is done generically; there doesn't appear to be any
2404 * special cases for the architectures.
2406 layout_sections(info->mod, info);
2407 layout_symtab(info->mod, info);
2409 /* Allocate and move to the final place */
2410 err = move_module(info->mod, info);
2412 return ERR_PTR(err);
2414 /* Module has been copied to its final place now: return it. */
2415 mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2416 kmemleak_load_module(mod, info);
2420 /* mod is no longer valid after this! */
2421 static void module_deallocate(struct module *mod, struct load_info *info)
2423 percpu_modfree(mod);
2424 module_arch_freeing_init(mod);
2429 int __weak module_finalize(const Elf_Ehdr *hdr,
2430 const Elf_Shdr *sechdrs,
2436 static int post_relocation(struct module *mod, const struct load_info *info)
2438 /* Sort exception table now relocations are done. */
2439 sort_extable(mod->extable, mod->extable + mod->num_exentries);
2441 /* Copy relocated percpu area over. */
2442 percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
2443 info->sechdrs[info->index.pcpu].sh_size);
2445 /* Setup kallsyms-specific fields. */
2446 add_kallsyms(mod, info);
2448 /* Arch-specific module finalizing. */
2449 return module_finalize(info->hdr, info->sechdrs, mod);
2452 /* Call module constructors. */
2453 static void do_mod_ctors(struct module *mod)
2455 #ifdef CONFIG_CONSTRUCTORS
2458 for (i = 0; i < mod->num_ctors; i++)
2463 /* For freeing module_init on success, in case kallsyms traversing */
2464 struct mod_initfree {
2465 struct llist_node node;
2471 static void do_free_init(struct work_struct *w)
2473 struct llist_node *pos, *n, *list;
2474 struct mod_initfree *initfree;
2476 list = llist_del_all(&init_free_list);
2480 llist_for_each_safe(pos, n, list) {
2481 initfree = container_of(pos, struct mod_initfree, node);
2482 module_memfree(initfree->init_text);
2483 module_memfree(initfree->init_data);
2484 module_memfree(initfree->init_rodata);
2489 #undef MODULE_PARAM_PREFIX
2490 #define MODULE_PARAM_PREFIX "module."
2491 /* Default value for module->async_probe_requested */
2492 static bool async_probe;
2493 module_param(async_probe, bool, 0644);
2496 * This is where the real work happens.
2498 * Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
2499 * helper command 'lx-symbols'.
2501 static noinline int do_init_module(struct module *mod)
2504 struct mod_initfree *freeinit;
2505 #if defined(CONFIG_MODULE_STATS)
2506 unsigned int text_size = 0, total_size = 0;
2508 for_each_mod_mem_type(type) {
2509 const struct module_memory *mod_mem = &mod->mem[type];
2510 if (mod_mem->size) {
2511 total_size += mod_mem->size;
2512 if (type == MOD_TEXT || type == MOD_INIT_TEXT)
2513 text_size += mod_mem->size;
2518 freeinit = kmalloc(sizeof(*freeinit), GFP_KERNEL);
2523 freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
2524 freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
2525 freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
2528 /* Start the module */
2529 if (mod->init != NULL)
2530 ret = do_one_initcall(mod->init);
2532 goto fail_free_freeinit;
2535 pr_warn("%s: '%s'->init suspiciously returned %d, it should "
2536 "follow 0/-E convention\n"
2537 "%s: loading module anyway...\n",
2538 __func__, mod->name, ret, __func__);
2542 /* Now it's a first class citizen! */
2543 mod->state = MODULE_STATE_LIVE;
2544 blocking_notifier_call_chain(&module_notify_list,
2545 MODULE_STATE_LIVE, mod);
2547 /* Delay uevent until module has finished its init routine */
2548 kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
2551 * We need to finish all async code before the module init sequence
2552 * is done. This has potential to deadlock if synchronous module
2553 * loading is requested from async (which is not allowed!).
2555 * See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
2556 * request_module() from async workers") for more details.
2558 if (!mod->async_probe_requested)
2559 async_synchronize_full();
2561 ftrace_free_mem(mod, mod->mem[MOD_INIT_TEXT].base,
2562 mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
2563 mutex_lock(&module_mutex);
2564 /* Drop initial reference. */
2566 trim_init_extable(mod);
2567 #ifdef CONFIG_KALLSYMS
2568 /* Switch to core kallsyms now init is done: kallsyms may be walking! */
2569 rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
2571 module_enable_ro(mod, true);
2572 mod_tree_remove_init(mod);
2573 module_arch_freeing_init(mod);
2574 for_class_mod_mem_type(type, init) {
2575 mod->mem[type].base = NULL;
2576 mod->mem[type].size = 0;
2579 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2580 /* .BTF is not SHF_ALLOC and will get removed, so sanitize pointer */
2581 mod->btf_data = NULL;
2584 * We want to free module_init, but be aware that kallsyms may be
2585 * walking this with preempt disabled. In all the failure paths, we
2586 * call synchronize_rcu(), but we don't want to slow down the success
2587 * path. module_memfree() cannot be called in an interrupt, so do the
2588 * work and call synchronize_rcu() in a work queue.
2590 * Note that module_alloc() on most architectures creates W+X page
2591 * mappings which won't be cleaned up until do_free_init() runs. Any
2592 * code such as mark_rodata_ro() which depends on those mappings to
2593 * be cleaned up needs to sync with the queued work - ie
2596 if (llist_add(&freeinit->node, &init_free_list))
2597 schedule_work(&init_free_wq);
2599 mutex_unlock(&module_mutex);
2600 wake_up_all(&module_wq);
2602 mod_stat_add_long(text_size, &total_text_size);
2603 mod_stat_add_long(total_size, &total_mod_size);
2605 mod_stat_inc(&modcount);
2612 /* Try to protect us from buggy refcounters. */
2613 mod->state = MODULE_STATE_GOING;
2616 blocking_notifier_call_chain(&module_notify_list,
2617 MODULE_STATE_GOING, mod);
2618 klp_module_going(mod);
2619 ftrace_release_mod(mod);
2621 wake_up_all(&module_wq);
2626 static int may_init_module(void)
2628 if (!capable(CAP_SYS_MODULE) || modules_disabled)
2634 /* Is this module of this name done loading? No locks held. */
2635 static bool finished_loading(const char *name)
2641 * The module_mutex should not be a heavily contended lock;
2642 * if we get the occasional sleep here, we'll go an extra iteration
2643 * in the wait_event_interruptible(), which is harmless.
2645 sched_annotate_sleep();
2646 mutex_lock(&module_mutex);
2647 mod = find_module_all(name, strlen(name), true);
2648 ret = !mod || mod->state == MODULE_STATE_LIVE
2649 || mod->state == MODULE_STATE_GOING;
2650 mutex_unlock(&module_mutex);
2655 /* Must be called with module_mutex held */
2656 static int module_patient_check_exists(const char *name,
2657 enum fail_dup_mod_reason reason)
2662 old = find_module_all(name, strlen(name), true);
2666 if (old->state == MODULE_STATE_COMING ||
2667 old->state == MODULE_STATE_UNFORMED) {
2668 /* Wait in case it fails to load. */
2669 mutex_unlock(&module_mutex);
2670 err = wait_event_interruptible(module_wq,
2671 finished_loading(name));
2672 mutex_lock(&module_mutex);
2676 /* The module might have gone in the meantime. */
2677 old = find_module_all(name, strlen(name), true);
2680 if (try_add_failed_module(name, reason))
2681 pr_warn("Could not add fail-tracking for module: %s\n", name);
2684 * We are here only when the same module was being loaded. Do
2685 * not try to load it again right now. It prevents long delays
2686 * caused by serialized module load failures. It might happen
2687 * when more devices of the same type trigger load of
2688 * a particular module.
2690 if (old && old->state == MODULE_STATE_LIVE)
2696 * We try to place it in the list now to make sure it's unique before
2697 * we dedicate too many resources. In particular, temporary percpu
2698 * memory exhaustion.
2700 static int add_unformed_module(struct module *mod)
2704 mod->state = MODULE_STATE_UNFORMED;
2706 mutex_lock(&module_mutex);
2707 err = module_patient_check_exists(mod->name, FAIL_DUP_MOD_LOAD);
2711 mod_update_bounds(mod);
2712 list_add_rcu(&mod->list, &modules);
2713 mod_tree_insert(mod);
2717 mutex_unlock(&module_mutex);
2721 static int complete_formation(struct module *mod, struct load_info *info)
2725 mutex_lock(&module_mutex);
2727 /* Find duplicate symbols (must be called under lock). */
2728 err = verify_exported_symbols(mod);
2732 /* These rely on module_mutex for list integrity. */
2733 module_bug_finalize(info->hdr, info->sechdrs, mod);
2734 module_cfi_finalize(info->hdr, info->sechdrs, mod);
2736 module_enable_ro(mod, false);
2737 module_enable_nx(mod);
2738 module_enable_x(mod);
2741 * Mark state as coming so strong_try_module_get() ignores us,
2742 * but kallsyms etc. can see us.
2744 mod->state = MODULE_STATE_COMING;
2745 mutex_unlock(&module_mutex);
2750 mutex_unlock(&module_mutex);
2754 static int prepare_coming_module(struct module *mod)
2758 ftrace_module_enable(mod);
2759 err = klp_module_coming(mod);
2763 err = blocking_notifier_call_chain_robust(&module_notify_list,
2764 MODULE_STATE_COMING, MODULE_STATE_GOING, mod);
2765 err = notifier_to_errno(err);
2767 klp_module_going(mod);
2772 static int unknown_module_param_cb(char *param, char *val, const char *modname,
2775 struct module *mod = arg;
2778 if (strcmp(param, "async_probe") == 0) {
2779 if (kstrtobool(val, &mod->async_probe_requested))
2780 mod->async_probe_requested = true;
2784 /* Check for magic 'dyndbg' arg */
2785 ret = ddebug_dyndbg_module_param_cb(param, val, modname);
2787 pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
2791 /* Module within temporary copy, this doesn't do any allocation */
2792 static int early_mod_check(struct load_info *info, int flags)
2797 * Now that we know we have the correct module name, check
2798 * if it's blacklisted.
2800 if (blacklisted(info->name)) {
2801 pr_err("Module %s is blacklisted\n", info->name);
2805 err = rewrite_section_headers(info, flags);
2809 /* Check module struct version now, before we try to use module. */
2810 if (!check_modstruct_version(info, info->mod))
2813 err = check_modinfo(info->mod, info, flags);
2817 mutex_lock(&module_mutex);
2818 err = module_patient_check_exists(info->mod->name, FAIL_DUP_MOD_BECOMING);
2819 mutex_unlock(&module_mutex);
2825 * Allocate and load the module: note that size of section 0 is always
2826 * zero, and we rely on this for optional sections.
2828 static int load_module(struct load_info *info, const char __user *uargs,
2832 bool module_allocated = false;
2837 * Do the signature check (if any) first. All that
2838 * the signature check needs is info->len, it does
2839 * not need any of the section info. That can be
2840 * set up later. This will minimize the chances
2841 * of a corrupt module causing problems before
2842 * we even get to the signature check.
2844 * The check will also adjust info->len by stripping
2845 * off the sig length at the end of the module, making
2846 * checks against info->len more correct.
2848 err = module_sig_check(info, flags);
2853 * Do basic sanity checks against the ELF header and
2854 * sections. Cache useful sections and set the
2855 * info->mod to the userspace passed struct module.
2857 err = elf_validity_cache_copy(info, flags);
2861 err = early_mod_check(info, flags);
2865 /* Figure out module layout, and allocate all the memory. */
2866 mod = layout_and_allocate(info, flags);
2872 module_allocated = true;
2874 audit_log_kern_module(mod->name);
2876 /* Reserve our place in the list. */
2877 err = add_unformed_module(mod);
2882 * We are tainting your kernel if your module gets into
2883 * the modules linked list somehow.
2885 module_augment_kernel_taints(mod, info);
2887 /* To avoid stressing percpu allocator, do this once we're unique. */
2888 err = percpu_modalloc(mod, info);
2892 /* Now module is in final location, initialize linked lists, etc. */
2893 err = module_unload_init(mod);
2897 init_param_lock(mod);
2900 * Now we've got everything in the final locations, we can
2901 * find optional sections.
2903 err = find_module_sections(mod, info);
2907 err = check_export_symbol_versions(mod);
2911 /* Set up MODINFO_ATTR fields */
2912 setup_modinfo(mod, info);
2914 /* Fix up syms, so that st_value is a pointer to location. */
2915 err = simplify_symbols(mod, info);
2919 err = apply_relocations(mod, info);
2923 err = post_relocation(mod, info);
2927 flush_module_icache(mod);
2929 /* Now copy in args */
2930 mod->args = strndup_user(uargs, ~0UL >> 1);
2931 if (IS_ERR(mod->args)) {
2932 err = PTR_ERR(mod->args);
2933 goto free_arch_cleanup;
2936 init_build_id(mod, info);
2938 /* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
2939 ftrace_module_init(mod);
2941 /* Finally it's fully formed, ready to start executing. */
2942 err = complete_formation(mod, info);
2944 goto ddebug_cleanup;
2946 err = prepare_coming_module(mod);
2950 mod->async_probe_requested = async_probe;
2952 /* Module is ready to execute: parsing args may do that. */
2953 after_dashes = parse_args(mod->name, mod->args, mod->kp, mod->num_kp,
2955 unknown_module_param_cb);
2956 if (IS_ERR(after_dashes)) {
2957 err = PTR_ERR(after_dashes);
2958 goto coming_cleanup;
2959 } else if (after_dashes) {
2960 pr_warn("%s: parameters '%s' after `--' ignored\n",
2961 mod->name, after_dashes);
2964 /* Link in to sysfs. */
2965 err = mod_sysfs_setup(mod, info, mod->kp, mod->num_kp);
2967 goto coming_cleanup;
2969 if (is_livepatch_module(mod)) {
2970 err = copy_module_elf(mod, info);
2975 /* Get rid of temporary copy. */
2976 free_copy(info, flags);
2979 trace_module_load(mod);
2981 return do_init_module(mod);
2984 mod_sysfs_teardown(mod);
2986 mod->state = MODULE_STATE_GOING;
2987 destroy_params(mod->kp, mod->num_kp);
2988 blocking_notifier_call_chain(&module_notify_list,
2989 MODULE_STATE_GOING, mod);
2990 klp_module_going(mod);
2992 mod->state = MODULE_STATE_GOING;
2993 /* module_bug_cleanup needs module_mutex protection */
2994 mutex_lock(&module_mutex);
2995 module_bug_cleanup(mod);
2996 mutex_unlock(&module_mutex);
2999 ftrace_release_mod(mod);
3003 module_arch_cleanup(mod);
3007 module_unload_free(mod);
3009 mutex_lock(&module_mutex);
3010 /* Unlink carefully: kallsyms could be walking list. */
3011 list_del_rcu(&mod->list);
3012 mod_tree_remove(mod);
3013 wake_up_all(&module_wq);
3014 /* Wait for RCU-sched synchronizing before releasing mod->list. */
3016 mutex_unlock(&module_mutex);
3018 mod_stat_bump_invalid(info, flags);
3019 /* Free lock-classes; relies on the preceding sync_rcu() */
3020 for_class_mod_mem_type(type, core_data) {
3021 lockdep_free_key_range(mod->mem[type].base,
3022 mod->mem[type].size);
3025 module_deallocate(mod, info);
3028 * The info->len is always set. We distinguish between
3029 * failures once the proper module was allocated and
3032 if (!module_allocated)
3033 mod_stat_bump_becoming(info, flags);
3034 free_copy(info, flags);
3038 SYSCALL_DEFINE3(init_module, void __user *, umod,
3039 unsigned long, len, const char __user *, uargs)
3042 struct load_info info = { };
3044 err = may_init_module();
3048 pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3051 err = copy_module_from_user(umod, len, &info);
3053 mod_stat_inc(&failed_kreads);
3054 mod_stat_add_long(len, &invalid_kread_bytes);
3058 return load_module(&info, uargs, 0);
3063 struct hlist_node entry;
3064 struct completion complete;
3068 #define IDEM_HASH_BITS 8
3069 static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3070 static DEFINE_SPINLOCK(idem_lock);
3072 static bool idempotent(struct idempotent *u, const void *cookie)
3074 int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3075 struct hlist_head *head = idem_hash + hash;
3076 struct idempotent *existing;
3081 init_completion(&u->complete);
3083 spin_lock(&idem_lock);
3085 hlist_for_each_entry(existing, head, entry) {
3086 if (existing->cookie != cookie)
3091 hlist_add_head(&u->entry, idem_hash + hash);
3092 spin_unlock(&idem_lock);
3098 * We were the first one with 'cookie' on the list, and we ended
3099 * up completing the operation. We now need to walk the list,
3100 * remove everybody - which includes ourselves - fill in the return
3101 * value, and then complete the operation.
3103 static int idempotent_complete(struct idempotent *u, int ret)
3105 const void *cookie = u->cookie;
3106 int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3107 struct hlist_head *head = idem_hash + hash;
3108 struct hlist_node *next;
3109 struct idempotent *pos;
3111 spin_lock(&idem_lock);
3112 hlist_for_each_entry_safe(pos, next, head, entry) {
3113 if (pos->cookie != cookie)
3115 hlist_del(&pos->entry);
3117 complete(&pos->complete);
3119 spin_unlock(&idem_lock);
3123 static int init_module_from_file(struct file *f, const char __user * uargs, int flags)
3125 struct load_info info = { };
3129 len = kernel_read_file(f, 0, &buf, INT_MAX, NULL, READING_MODULE);
3131 mod_stat_inc(&failed_kreads);
3135 if (flags & MODULE_INIT_COMPRESSED_FILE) {
3136 int err = module_decompress(&info, buf, len);
3137 vfree(buf); /* compressed data is no longer needed */
3139 mod_stat_inc(&failed_decompress);
3140 mod_stat_add_long(len, &invalid_decompress_bytes);
3148 return load_module(&info, uargs, flags);
3151 static int idempotent_init_module(struct file *f, const char __user * uargs, int flags)
3153 struct idempotent idem;
3155 if (!f || !(f->f_mode & FMODE_READ))
3158 /* See if somebody else is doing the operation? */
3159 if (idempotent(&idem, file_inode(f))) {
3160 wait_for_completion(&idem.complete);
3164 /* Otherwise, we'll do it and complete others */
3165 return idempotent_complete(&idem,
3166 init_module_from_file(f, uargs, flags));
3169 SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3174 err = may_init_module();
3178 pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3180 if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3181 |MODULE_INIT_IGNORE_VERMAGIC
3182 |MODULE_INIT_COMPRESSED_FILE))
3186 err = idempotent_init_module(f.file, uargs, flags);
3191 /* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
3192 char *module_flags(struct module *mod, char *buf, bool show_state)
3196 BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3197 if (!mod->taints && !show_state)
3200 mod->state == MODULE_STATE_GOING ||
3201 mod->state == MODULE_STATE_COMING) {
3203 bx += module_flags_taint(mod->taints, buf + bx);
3204 /* Show a - for module-is-being-unloaded */
3205 if (mod->state == MODULE_STATE_GOING && show_state)
3207 /* Show a + for module-is-being-loaded */
3208 if (mod->state == MODULE_STATE_COMING && show_state)
3218 /* Given an address, look for it in the module exception tables. */
3219 const struct exception_table_entry *search_module_extables(unsigned long addr)
3221 const struct exception_table_entry *e = NULL;
3225 mod = __module_address(addr);
3229 if (!mod->num_exentries)
3232 e = search_extable(mod->extable,
3239 * Now, if we found one, we are running inside it now, hence
3240 * we cannot unload the module, hence no refcnt needed.
3246 * is_module_address() - is this address inside a module?
3247 * @addr: the address to check.
3249 * See is_module_text_address() if you simply want to see if the address
3250 * is code (not data).
3252 bool is_module_address(unsigned long addr)
3257 ret = __module_address(addr) != NULL;
3264 * __module_address() - get the module which contains an address.
3265 * @addr: the address.
3267 * Must be called with preempt disabled or module mutex held so that
3268 * module doesn't get freed during this.
3270 struct module *__module_address(unsigned long addr)
3274 if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3277 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3278 if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3285 module_assert_mutex_or_preempt();
3287 mod = mod_find(addr, &mod_tree);
3289 BUG_ON(!within_module(addr, mod));
3290 if (mod->state == MODULE_STATE_UNFORMED)
3297 * is_module_text_address() - is this address inside module code?
3298 * @addr: the address to check.
3300 * See is_module_address() if you simply want to see if the address is
3301 * anywhere in a module. See kernel_text_address() for testing if an
3302 * address corresponds to kernel or module code.
3304 bool is_module_text_address(unsigned long addr)
3309 ret = __module_text_address(addr) != NULL;
3316 * __module_text_address() - get the module whose code contains an address.
3317 * @addr: the address.
3319 * Must be called with preempt disabled or module mutex held so that
3320 * module doesn't get freed during this.
3322 struct module *__module_text_address(unsigned long addr)
3324 struct module *mod = __module_address(addr);
3326 /* Make sure it's within the text section. */
3327 if (!within_module_mem_type(addr, mod, MOD_TEXT) &&
3328 !within_module_mem_type(addr, mod, MOD_INIT_TEXT))
3334 /* Don't grab lock, we're oopsing. */
3335 void print_modules(void)
3338 char buf[MODULE_FLAGS_BUF_SIZE];
3340 printk(KERN_DEFAULT "Modules linked in:");
3341 /* Most callers should already have preempt disabled, but make sure */
3343 list_for_each_entry_rcu(mod, &modules, list) {
3344 if (mod->state == MODULE_STATE_UNFORMED)
3346 pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3349 print_unloaded_tainted_modules();
3351 if (last_unloaded_module.name[0])
3352 pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3353 last_unloaded_module.taints);
3357 #ifdef CONFIG_MODULE_DEBUGFS
3358 struct dentry *mod_debugfs_root;
3360 static int module_debugfs_init(void)
3362 mod_debugfs_root = debugfs_create_dir("modules", NULL);
3365 module_init(module_debugfs_init);