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/codetag.h>
60 #include <linux/debugfs.h>
61 #include <linux/execmem.h>
62 #include <uapi/linux/module.h>
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/module.h>
70 * 1) List of modules (also safely readable with preempt_disable),
71 * 2) module_use links,
72 * 3) mod_tree.addr_min/mod_tree.addr_max.
73 * (delete and add uses RCU list operations).
75 DEFINE_MUTEX(module_mutex);
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);
83 struct mod_tree_root mod_tree __cacheline_aligned = {
88 const struct kernel_symbol *start, *stop;
90 enum mod_license license;
94 * Bounds of module memory, for speeding up __module_address.
95 * Protected by module_mutex.
97 static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
98 unsigned int size, struct mod_tree_root *tree)
100 unsigned long min = (unsigned long)base;
101 unsigned long max = min + size;
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;
112 if (min < tree->addr_min)
113 tree->addr_min = min;
114 if (max > tree->addr_max)
115 tree->addr_max = max;
118 static void mod_update_bounds(struct module *mod)
120 for_each_mod_mem_type(type) {
121 struct module_memory *mod_mem = &mod->mem[type];
124 __mod_update_bounds(type, mod_mem->base, mod_mem->size, &mod_tree);
128 /* Block module loading/unloading? */
129 int modules_disabled;
130 core_param(nomodule, modules_disabled, bint, 0);
132 /* Waiting for a module to finish initializing? */
133 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
135 static BLOCKING_NOTIFIER_HEAD(module_notify_list);
137 int register_module_notifier(struct notifier_block *nb)
139 return blocking_notifier_chain_register(&module_notify_list, nb);
141 EXPORT_SYMBOL(register_module_notifier);
143 int unregister_module_notifier(struct notifier_block *nb)
145 return blocking_notifier_chain_unregister(&module_notify_list, nb);
147 EXPORT_SYMBOL(unregister_module_notifier);
150 * We require a truly strong try_module_get(): 0 means success.
151 * Otherwise an error is returned due to ongoing or failed
152 * initialization etc.
154 static inline int strong_try_module_get(struct module *mod)
156 BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
157 if (mod && mod->state == MODULE_STATE_COMING)
159 if (try_module_get(mod))
165 static inline void add_taint_module(struct module *mod, unsigned flag,
166 enum lockdep_ok lockdep_ok)
168 add_taint(flag, lockdep_ok);
169 set_bit(flag, &mod->taints);
173 * A thread that wants to hold a reference to a module only while it
174 * is running can call this to safely exit.
176 void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
181 EXPORT_SYMBOL(__module_put_and_kthread_exit);
183 /* Find a module section: 0 means not found. */
184 static unsigned int find_sec(const struct load_info *info, const char *name)
188 for (i = 1; i < info->hdr->e_shnum; i++) {
189 Elf_Shdr *shdr = &info->sechdrs[i];
190 /* Alloc bit cleared means "ignore it." */
191 if ((shdr->sh_flags & SHF_ALLOC)
192 && strcmp(info->secstrings + shdr->sh_name, name) == 0)
198 /* Find a module section, or NULL. */
199 static void *section_addr(const struct load_info *info, const char *name)
201 /* Section 0 has sh_addr 0. */
202 return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
205 /* Find a module section, or NULL. Fill in number of "objects" in section. */
206 static void *section_objs(const struct load_info *info,
211 unsigned int sec = find_sec(info, name);
213 /* Section 0 has sh_addr 0 and sh_size 0. */
214 *num = info->sechdrs[sec].sh_size / object_size;
215 return (void *)info->sechdrs[sec].sh_addr;
218 /* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
219 static unsigned int find_any_sec(const struct load_info *info, const char *name)
223 for (i = 1; i < info->hdr->e_shnum; i++) {
224 Elf_Shdr *shdr = &info->sechdrs[i];
225 if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
232 * Find a module section, or NULL. Fill in number of "objects" in section.
233 * Ignores SHF_ALLOC flag.
235 static __maybe_unused void *any_section_objs(const struct load_info *info,
240 unsigned int sec = find_any_sec(info, name);
242 /* Section 0 has sh_addr 0 and sh_size 0. */
243 *num = info->sechdrs[sec].sh_size / object_size;
244 return (void *)info->sechdrs[sec].sh_addr;
247 #ifndef CONFIG_MODVERSIONS
248 #define symversion(base, idx) NULL
250 #define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
253 static const char *kernel_symbol_name(const struct kernel_symbol *sym)
255 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
256 return offset_to_ptr(&sym->name_offset);
262 static const char *kernel_symbol_namespace(const struct kernel_symbol *sym)
264 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
265 if (!sym->namespace_offset)
267 return offset_to_ptr(&sym->namespace_offset);
269 return sym->namespace;
273 int cmp_name(const void *name, const void *sym)
275 return strcmp(name, kernel_symbol_name(sym));
278 static bool find_exported_symbol_in_section(const struct symsearch *syms,
279 struct module *owner,
280 struct find_symbol_arg *fsa)
282 struct kernel_symbol *sym;
284 if (!fsa->gplok && syms->license == GPL_ONLY)
287 sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
288 sizeof(struct kernel_symbol), cmp_name);
293 fsa->crc = symversion(syms->crcs, sym - syms->start);
295 fsa->license = syms->license;
301 * Find an exported symbol and return it, along with, (optional) crc and
302 * (optional) module which owns it. Needs preempt disabled or module_mutex.
304 bool find_symbol(struct find_symbol_arg *fsa)
306 static const struct symsearch arr[] = {
307 { __start___ksymtab, __stop___ksymtab, __start___kcrctab,
309 { __start___ksymtab_gpl, __stop___ksymtab_gpl,
310 __start___kcrctab_gpl,
316 module_assert_mutex_or_preempt();
318 for (i = 0; i < ARRAY_SIZE(arr); i++)
319 if (find_exported_symbol_in_section(&arr[i], NULL, fsa))
322 list_for_each_entry_rcu(mod, &modules, list,
323 lockdep_is_held(&module_mutex)) {
324 struct symsearch arr[] = {
325 { mod->syms, mod->syms + mod->num_syms, mod->crcs,
327 { mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
332 if (mod->state == MODULE_STATE_UNFORMED)
335 for (i = 0; i < ARRAY_SIZE(arr); i++)
336 if (find_exported_symbol_in_section(&arr[i], mod, fsa))
340 pr_debug("Failed to find symbol %s\n", fsa->name);
345 * Search for module by name: must hold module_mutex (or preempt disabled
346 * for read-only access).
348 struct module *find_module_all(const char *name, size_t len,
353 module_assert_mutex_or_preempt();
355 list_for_each_entry_rcu(mod, &modules, list,
356 lockdep_is_held(&module_mutex)) {
357 if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
359 if (strlen(mod->name) == len && !memcmp(mod->name, name, len))
365 struct module *find_module(const char *name)
367 return find_module_all(name, strlen(name), false);
372 static inline void __percpu *mod_percpu(struct module *mod)
377 static int percpu_modalloc(struct module *mod, struct load_info *info)
379 Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
380 unsigned long align = pcpusec->sh_addralign;
382 if (!pcpusec->sh_size)
385 if (align > PAGE_SIZE) {
386 pr_warn("%s: per-cpu alignment %li > %li\n",
387 mod->name, align, PAGE_SIZE);
391 mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
393 pr_warn("%s: Could not allocate %lu bytes percpu data\n",
394 mod->name, (unsigned long)pcpusec->sh_size);
397 mod->percpu_size = pcpusec->sh_size;
401 static void percpu_modfree(struct module *mod)
403 free_percpu(mod->percpu);
406 static unsigned int find_pcpusec(struct load_info *info)
408 return find_sec(info, ".data..percpu");
411 static void percpu_modcopy(struct module *mod,
412 const void *from, unsigned long size)
416 for_each_possible_cpu(cpu)
417 memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
420 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
427 list_for_each_entry_rcu(mod, &modules, list) {
428 if (mod->state == MODULE_STATE_UNFORMED)
430 if (!mod->percpu_size)
432 for_each_possible_cpu(cpu) {
433 void *start = per_cpu_ptr(mod->percpu, cpu);
434 void *va = (void *)addr;
436 if (va >= start && va < start + mod->percpu_size) {
438 *can_addr = (unsigned long) (va - start);
439 *can_addr += (unsigned long)
440 per_cpu_ptr(mod->percpu,
454 * is_module_percpu_address() - test whether address is from module static percpu
455 * @addr: address to test
457 * Test whether @addr belongs to module static percpu area.
459 * Return: %true if @addr is from module static percpu area
461 bool is_module_percpu_address(unsigned long addr)
463 return __is_module_percpu_address(addr, NULL);
466 #else /* ... !CONFIG_SMP */
468 static inline void __percpu *mod_percpu(struct module *mod)
472 static int percpu_modalloc(struct module *mod, struct load_info *info)
474 /* UP modules shouldn't have this section: ENOMEM isn't quite right */
475 if (info->sechdrs[info->index.pcpu].sh_size != 0)
479 static inline void percpu_modfree(struct module *mod)
482 static unsigned int find_pcpusec(struct load_info *info)
486 static inline void percpu_modcopy(struct module *mod,
487 const void *from, unsigned long size)
489 /* pcpusec should be 0, and size of that section should be 0. */
492 bool is_module_percpu_address(unsigned long addr)
497 bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
502 #endif /* CONFIG_SMP */
504 #define MODINFO_ATTR(field) \
505 static void setup_modinfo_##field(struct module *mod, const char *s) \
507 mod->field = kstrdup(s, GFP_KERNEL); \
509 static ssize_t show_modinfo_##field(struct module_attribute *mattr, \
510 struct module_kobject *mk, char *buffer) \
512 return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field); \
514 static int modinfo_##field##_exists(struct module *mod) \
516 return mod->field != NULL; \
518 static void free_modinfo_##field(struct module *mod) \
523 static struct module_attribute modinfo_##field = { \
524 .attr = { .name = __stringify(field), .mode = 0444 }, \
525 .show = show_modinfo_##field, \
526 .setup = setup_modinfo_##field, \
527 .test = modinfo_##field##_exists, \
528 .free = free_modinfo_##field, \
531 MODINFO_ATTR(version);
532 MODINFO_ATTR(srcversion);
535 char name[MODULE_NAME_LEN + 1];
536 char taints[MODULE_FLAGS_BUF_SIZE];
537 } last_unloaded_module;
539 #ifdef CONFIG_MODULE_UNLOAD
541 EXPORT_TRACEPOINT_SYMBOL(module_get);
543 /* MODULE_REF_BASE is the base reference count by kmodule loader. */
544 #define MODULE_REF_BASE 1
546 /* Init the unload section of the module. */
547 static int module_unload_init(struct module *mod)
550 * Initialize reference counter to MODULE_REF_BASE.
551 * refcnt == 0 means module is going.
553 atomic_set(&mod->refcnt, MODULE_REF_BASE);
555 INIT_LIST_HEAD(&mod->source_list);
556 INIT_LIST_HEAD(&mod->target_list);
558 /* Hold reference count during initialization. */
559 atomic_inc(&mod->refcnt);
564 /* Does a already use b? */
565 static int already_uses(struct module *a, struct module *b)
567 struct module_use *use;
569 list_for_each_entry(use, &b->source_list, source_list) {
570 if (use->source == a)
573 pr_debug("%s does not use %s!\n", a->name, b->name);
579 * - we add 'a' as a "source", 'b' as a "target" of module use
580 * - the module_use is added to the list of 'b' sources (so
581 * 'b' can walk the list to see who sourced them), and of 'a'
582 * targets (so 'a' can see what modules it targets).
584 static int add_module_usage(struct module *a, struct module *b)
586 struct module_use *use;
588 pr_debug("Allocating new usage for %s.\n", a->name);
589 use = kmalloc(sizeof(*use), GFP_ATOMIC);
595 list_add(&use->source_list, &b->source_list);
596 list_add(&use->target_list, &a->target_list);
600 /* Module a uses b: caller needs module_mutex() */
601 static int ref_module(struct module *a, struct module *b)
605 if (b == NULL || already_uses(a, b))
608 /* If module isn't available, we fail. */
609 err = strong_try_module_get(b);
613 err = add_module_usage(a, b);
621 /* Clear the unload stuff of the module. */
622 static void module_unload_free(struct module *mod)
624 struct module_use *use, *tmp;
626 mutex_lock(&module_mutex);
627 list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
628 struct module *i = use->target;
629 pr_debug("%s unusing %s\n", mod->name, i->name);
631 list_del(&use->source_list);
632 list_del(&use->target_list);
635 mutex_unlock(&module_mutex);
638 #ifdef CONFIG_MODULE_FORCE_UNLOAD
639 static inline int try_force_unload(unsigned int flags)
641 int ret = (flags & O_TRUNC);
643 add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
647 static inline int try_force_unload(unsigned int flags)
651 #endif /* CONFIG_MODULE_FORCE_UNLOAD */
653 /* Try to release refcount of module, 0 means success. */
654 static int try_release_module_ref(struct module *mod)
658 /* Try to decrement refcnt which we set at loading */
659 ret = atomic_sub_return(MODULE_REF_BASE, &mod->refcnt);
662 /* Someone can put this right now, recover with checking */
663 ret = atomic_add_unless(&mod->refcnt, MODULE_REF_BASE, 0);
668 static int try_stop_module(struct module *mod, int flags, int *forced)
670 /* If it's not unused, quit unless we're forcing. */
671 if (try_release_module_ref(mod) != 0) {
672 *forced = try_force_unload(flags);
677 /* Mark it as dying. */
678 mod->state = MODULE_STATE_GOING;
684 * module_refcount() - return the refcount or -1 if unloading
685 * @mod: the module we're checking
688 * -1 if the module is in the process of unloading
689 * otherwise the number of references in the kernel to the module
691 int module_refcount(struct module *mod)
693 return atomic_read(&mod->refcnt) - MODULE_REF_BASE;
695 EXPORT_SYMBOL(module_refcount);
697 /* This exists whether we can unload or not */
698 static void free_module(struct module *mod);
700 SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
704 char name[MODULE_NAME_LEN];
705 char buf[MODULE_FLAGS_BUF_SIZE];
708 if (!capable(CAP_SYS_MODULE) || modules_disabled)
711 if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0)
713 name[MODULE_NAME_LEN-1] = '\0';
715 audit_log_kern_module(name);
717 if (mutex_lock_interruptible(&module_mutex) != 0)
720 mod = find_module(name);
726 if (!list_empty(&mod->source_list)) {
727 /* Other modules depend on us: get rid of them first. */
732 /* Doing init or already dying? */
733 if (mod->state != MODULE_STATE_LIVE) {
734 /* FIXME: if (force), slam module count damn the torpedoes */
735 pr_debug("%s already dying\n", mod->name);
740 /* If it has an init func, it must have an exit func to unload */
741 if (mod->init && !mod->exit) {
742 forced = try_force_unload(flags);
744 /* This module can't be removed */
750 ret = try_stop_module(mod, flags, &forced);
754 mutex_unlock(&module_mutex);
755 /* Final destruction now no one is using it. */
756 if (mod->exit != NULL)
758 blocking_notifier_call_chain(&module_notify_list,
759 MODULE_STATE_GOING, mod);
760 klp_module_going(mod);
761 ftrace_release_mod(mod);
763 async_synchronize_full();
765 /* Store the name and taints of the last unloaded module for diagnostic purposes */
766 strscpy(last_unloaded_module.name, mod->name, sizeof(last_unloaded_module.name));
767 strscpy(last_unloaded_module.taints, module_flags(mod, buf, false), sizeof(last_unloaded_module.taints));
770 /* someone could wait for the module in add_unformed_module() */
771 wake_up_all(&module_wq);
774 mutex_unlock(&module_mutex);
778 void __symbol_put(const char *symbol)
780 struct find_symbol_arg fsa = {
786 BUG_ON(!find_symbol(&fsa));
787 module_put(fsa.owner);
790 EXPORT_SYMBOL(__symbol_put);
792 /* Note this assumes addr is a function, which it currently always is. */
793 void symbol_put_addr(void *addr)
795 struct module *modaddr;
796 unsigned long a = (unsigned long)dereference_function_descriptor(addr);
798 if (core_kernel_text(a))
802 * Even though we hold a reference on the module; we still need to
803 * disable preemption in order to safely traverse the data structure.
806 modaddr = __module_text_address(a);
811 EXPORT_SYMBOL_GPL(symbol_put_addr);
813 static ssize_t show_refcnt(struct module_attribute *mattr,
814 struct module_kobject *mk, char *buffer)
816 return sprintf(buffer, "%i\n", module_refcount(mk->mod));
819 static struct module_attribute modinfo_refcnt =
820 __ATTR(refcnt, 0444, show_refcnt, NULL);
822 void __module_get(struct module *module)
825 atomic_inc(&module->refcnt);
826 trace_module_get(module, _RET_IP_);
829 EXPORT_SYMBOL(__module_get);
831 bool try_module_get(struct module *module)
836 /* Note: here, we can fail to get a reference */
837 if (likely(module_is_live(module) &&
838 atomic_inc_not_zero(&module->refcnt) != 0))
839 trace_module_get(module, _RET_IP_);
845 EXPORT_SYMBOL(try_module_get);
847 void module_put(struct module *module)
852 ret = atomic_dec_if_positive(&module->refcnt);
853 WARN_ON(ret < 0); /* Failed to put refcount */
854 trace_module_put(module, _RET_IP_);
857 EXPORT_SYMBOL(module_put);
859 #else /* !CONFIG_MODULE_UNLOAD */
860 static inline void module_unload_free(struct module *mod)
864 static int ref_module(struct module *a, struct module *b)
866 return strong_try_module_get(b);
869 static inline int module_unload_init(struct module *mod)
873 #endif /* CONFIG_MODULE_UNLOAD */
875 size_t module_flags_taint(unsigned long taints, char *buf)
880 for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
881 if (taint_flags[i].module && test_bit(i, &taints))
882 buf[l++] = taint_flags[i].c_true;
888 static ssize_t show_initstate(struct module_attribute *mattr,
889 struct module_kobject *mk, char *buffer)
891 const char *state = "unknown";
893 switch (mk->mod->state) {
894 case MODULE_STATE_LIVE:
897 case MODULE_STATE_COMING:
900 case MODULE_STATE_GOING:
906 return sprintf(buffer, "%s\n", state);
909 static struct module_attribute modinfo_initstate =
910 __ATTR(initstate, 0444, show_initstate, NULL);
912 static ssize_t store_uevent(struct module_attribute *mattr,
913 struct module_kobject *mk,
914 const char *buffer, size_t count)
918 rc = kobject_synth_uevent(&mk->kobj, buffer, count);
919 return rc ? rc : count;
922 struct module_attribute module_uevent =
923 __ATTR(uevent, 0200, NULL, store_uevent);
925 static ssize_t show_coresize(struct module_attribute *mattr,
926 struct module_kobject *mk, char *buffer)
928 unsigned int size = mk->mod->mem[MOD_TEXT].size;
930 if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
931 for_class_mod_mem_type(type, core_data)
932 size += mk->mod->mem[type].size;
934 return sprintf(buffer, "%u\n", size);
937 static struct module_attribute modinfo_coresize =
938 __ATTR(coresize, 0444, show_coresize, NULL);
940 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
941 static ssize_t show_datasize(struct module_attribute *mattr,
942 struct module_kobject *mk, char *buffer)
944 unsigned int size = 0;
946 for_class_mod_mem_type(type, core_data)
947 size += mk->mod->mem[type].size;
948 return sprintf(buffer, "%u\n", size);
951 static struct module_attribute modinfo_datasize =
952 __ATTR(datasize, 0444, show_datasize, NULL);
955 static ssize_t show_initsize(struct module_attribute *mattr,
956 struct module_kobject *mk, char *buffer)
958 unsigned int size = 0;
960 for_class_mod_mem_type(type, init)
961 size += mk->mod->mem[type].size;
962 return sprintf(buffer, "%u\n", size);
965 static struct module_attribute modinfo_initsize =
966 __ATTR(initsize, 0444, show_initsize, NULL);
968 static ssize_t show_taint(struct module_attribute *mattr,
969 struct module_kobject *mk, char *buffer)
973 l = module_flags_taint(mk->mod->taints, buffer);
978 static struct module_attribute modinfo_taint =
979 __ATTR(taint, 0444, show_taint, NULL);
981 struct module_attribute *modinfo_attrs[] = {
987 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
992 #ifdef CONFIG_MODULE_UNLOAD
998 size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
1000 static const char vermagic[] = VERMAGIC_STRING;
1002 int try_to_force_load(struct module *mod, const char *reason)
1004 #ifdef CONFIG_MODULE_FORCE_LOAD
1005 if (!test_taint(TAINT_FORCED_MODULE))
1006 pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1007 add_taint_module(mod, TAINT_FORCED_MODULE, LOCKDEP_NOW_UNRELIABLE);
1014 /* Parse tag=value strings from .modinfo section */
1015 char *module_next_tag_pair(char *string, unsigned long *secsize)
1017 /* Skip non-zero chars */
1020 if ((*secsize)-- <= 1)
1024 /* Skip any zero padding. */
1025 while (!string[0]) {
1027 if ((*secsize)-- <= 1)
1033 static char *get_next_modinfo(const struct load_info *info, const char *tag,
1037 unsigned int taglen = strlen(tag);
1038 Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1039 unsigned long size = infosec->sh_size;
1042 * get_modinfo() calls made before rewrite_section_headers()
1043 * must use sh_offset, as sh_addr isn't set!
1045 char *modinfo = (char *)info->hdr + infosec->sh_offset;
1048 size -= prev - modinfo;
1049 modinfo = module_next_tag_pair(prev, &size);
1052 for (p = modinfo; p; p = module_next_tag_pair(p, &size)) {
1053 if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1054 return p + taglen + 1;
1059 static char *get_modinfo(const struct load_info *info, const char *tag)
1061 return get_next_modinfo(info, tag, NULL);
1064 static int verify_namespace_is_imported(const struct load_info *info,
1065 const struct kernel_symbol *sym,
1068 const char *namespace;
1069 char *imported_namespace;
1071 namespace = kernel_symbol_namespace(sym);
1072 if (namespace && namespace[0]) {
1073 for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1074 if (strcmp(namespace, imported_namespace) == 0)
1077 #ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1082 "%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1083 mod->name, kernel_symbol_name(sym), namespace);
1084 #ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1091 static bool inherit_taint(struct module *mod, struct module *owner, const char *name)
1093 if (!owner || !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1096 if (mod->using_gplonly_symbols) {
1097 pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1098 mod->name, name, owner->name);
1102 if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1103 pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1104 mod->name, name, owner->name);
1105 set_bit(TAINT_PROPRIETARY_MODULE, &mod->taints);
1110 /* Resolve a symbol for this module. I.e. if we find one, record usage. */
1111 static const struct kernel_symbol *resolve_symbol(struct module *mod,
1112 const struct load_info *info,
1116 struct find_symbol_arg fsa = {
1118 .gplok = !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1124 * The module_mutex should not be a heavily contended lock;
1125 * if we get the occasional sleep here, we'll go an extra iteration
1126 * in the wait_event_interruptible(), which is harmless.
1128 sched_annotate_sleep();
1129 mutex_lock(&module_mutex);
1130 if (!find_symbol(&fsa))
1133 if (fsa.license == GPL_ONLY)
1134 mod->using_gplonly_symbols = true;
1136 if (!inherit_taint(mod, fsa.owner, name)) {
1141 if (!check_version(info, name, mod, fsa.crc)) {
1142 fsa.sym = ERR_PTR(-EINVAL);
1146 err = verify_namespace_is_imported(info, fsa.sym, mod);
1148 fsa.sym = ERR_PTR(err);
1152 err = ref_module(mod, fsa.owner);
1154 fsa.sym = ERR_PTR(err);
1159 /* We must make copy under the lock if we failed to get ref. */
1160 strncpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1162 mutex_unlock(&module_mutex);
1166 static const struct kernel_symbol *
1167 resolve_symbol_wait(struct module *mod,
1168 const struct load_info *info,
1171 const struct kernel_symbol *ksym;
1172 char owner[MODULE_NAME_LEN];
1174 if (wait_event_interruptible_timeout(module_wq,
1175 !IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1176 || PTR_ERR(ksym) != -EBUSY,
1178 pr_warn("%s: gave up waiting for init of module %s.\n",
1184 void __weak module_arch_cleanup(struct module *mod)
1188 void __weak module_arch_freeing_init(struct module *mod)
1192 static int module_memory_alloc(struct module *mod, enum mod_mem_type type)
1194 unsigned int size = PAGE_ALIGN(mod->mem[type].size);
1195 enum execmem_type execmem_type;
1198 mod->mem[type].size = size;
1200 if (mod_mem_type_is_data(type))
1201 execmem_type = EXECMEM_MODULE_DATA;
1203 execmem_type = EXECMEM_MODULE_TEXT;
1205 ptr = execmem_alloc(execmem_type, size);
1210 * The pointer to these blocks of memory are stored on the module
1211 * structure and we keep that around so long as the module is
1212 * around. We only free that memory when we unload the module.
1213 * Just mark them as not being a leak then. The .init* ELF
1214 * sections *do* get freed after boot so we *could* treat them
1215 * slightly differently with kmemleak_ignore() and only grey
1216 * them out as they work as typical memory allocations which
1217 * *do* eventually get freed, but let's just keep things simple
1218 * and avoid *any* false positives.
1220 kmemleak_not_leak(ptr);
1222 memset(ptr, 0, size);
1223 mod->mem[type].base = ptr;
1228 static void module_memory_free(struct module *mod, enum mod_mem_type type,
1229 bool unload_codetags)
1231 void *ptr = mod->mem[type].base;
1233 if (!unload_codetags && mod_mem_type_is_core_data(type))
1239 static void free_mod_mem(struct module *mod, bool unload_codetags)
1241 for_each_mod_mem_type(type) {
1242 struct module_memory *mod_mem = &mod->mem[type];
1244 if (type == MOD_DATA)
1247 /* Free lock-classes; relies on the preceding sync_rcu(). */
1248 lockdep_free_key_range(mod_mem->base, mod_mem->size);
1250 module_memory_free(mod, type, unload_codetags);
1253 /* MOD_DATA hosts mod, so free it at last */
1254 lockdep_free_key_range(mod->mem[MOD_DATA].base, mod->mem[MOD_DATA].size);
1255 module_memory_free(mod, MOD_DATA, unload_codetags);
1258 /* Free a module, remove from lists, etc. */
1259 static void free_module(struct module *mod)
1261 bool unload_codetags;
1263 trace_module_free(mod);
1265 unload_codetags = codetag_unload_module(mod);
1266 if (!unload_codetags)
1267 pr_warn("%s: memory allocation(s) from the module still alive, cannot unload cleanly\n",
1270 mod_sysfs_teardown(mod);
1273 * We leave it in list to prevent duplicate loads, but make sure
1274 * that noone uses it while it's being deconstructed.
1276 mutex_lock(&module_mutex);
1277 mod->state = MODULE_STATE_UNFORMED;
1278 mutex_unlock(&module_mutex);
1280 /* Arch-specific cleanup. */
1281 module_arch_cleanup(mod);
1283 /* Module unload stuff */
1284 module_unload_free(mod);
1286 /* Free any allocated parameters. */
1287 destroy_params(mod->kp, mod->num_kp);
1289 if (is_livepatch_module(mod))
1290 free_module_elf(mod);
1292 /* Now we can delete it from the lists */
1293 mutex_lock(&module_mutex);
1294 /* Unlink carefully: kallsyms could be walking list. */
1295 list_del_rcu(&mod->list);
1296 mod_tree_remove(mod);
1297 /* Remove this module from bug list, this uses list_del_rcu */
1298 module_bug_cleanup(mod);
1299 /* Wait for RCU-sched synchronizing before releasing mod->list and buglist. */
1301 if (try_add_tainted_module(mod))
1302 pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1304 mutex_unlock(&module_mutex);
1306 /* This may be empty, but that's OK */
1307 module_arch_freeing_init(mod);
1309 percpu_modfree(mod);
1311 free_mod_mem(mod, unload_codetags);
1314 void *__symbol_get(const char *symbol)
1316 struct find_symbol_arg fsa = {
1323 if (!find_symbol(&fsa))
1325 if (fsa.license != GPL_ONLY) {
1326 pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1330 if (strong_try_module_get(fsa.owner))
1333 return (void *)kernel_symbol_value(fsa.sym);
1338 EXPORT_SYMBOL_GPL(__symbol_get);
1341 * Ensure that an exported symbol [global namespace] does not already exist
1342 * in the kernel or in some other module's exported symbol table.
1344 * You must hold the module_mutex.
1346 static int verify_exported_symbols(struct module *mod)
1349 const struct kernel_symbol *s;
1351 const struct kernel_symbol *sym;
1354 { mod->syms, mod->num_syms },
1355 { mod->gpl_syms, mod->num_gpl_syms },
1358 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1359 for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1360 struct find_symbol_arg fsa = {
1361 .name = kernel_symbol_name(s),
1364 if (find_symbol(&fsa)) {
1365 pr_err("%s: exports duplicate symbol %s"
1367 mod->name, kernel_symbol_name(s),
1368 module_name(fsa.owner));
1376 static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1379 * On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1380 * before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1381 * i386 has a similar problem but may not deserve a fix.
1383 * If we ever have to ignore many symbols, consider refactoring the code to
1384 * only warn if referenced by a relocation.
1386 if (emachine == EM_386 || emachine == EM_X86_64)
1387 return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1391 /* Change all symbols so that st_value encodes the pointer directly. */
1392 static int simplify_symbols(struct module *mod, const struct load_info *info)
1394 Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1395 Elf_Sym *sym = (void *)symsec->sh_addr;
1396 unsigned long secbase;
1399 const struct kernel_symbol *ksym;
1401 for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1402 const char *name = info->strtab + sym[i].st_name;
1404 switch (sym[i].st_shndx) {
1406 /* Ignore common symbols */
1407 if (!strncmp(name, "__gnu_lto", 9))
1411 * We compiled with -fno-common. These are not
1412 * supposed to happen.
1414 pr_debug("Common symbol: %s\n", name);
1415 pr_warn("%s: please compile with -fno-common\n",
1421 /* Don't need to do anything */
1422 pr_debug("Absolute symbol: 0x%08lx %s\n",
1423 (long)sym[i].st_value, name);
1427 /* Livepatch symbols are resolved by livepatch */
1431 ksym = resolve_symbol_wait(mod, info, name);
1432 /* Ok if resolved. */
1433 if (ksym && !IS_ERR(ksym)) {
1434 sym[i].st_value = kernel_symbol_value(ksym);
1438 /* Ok if weak or ignored. */
1440 (ELF_ST_BIND(sym[i].st_info) == STB_WEAK ||
1441 ignore_undef_symbol(info->hdr->e_machine, name)))
1444 ret = PTR_ERR(ksym) ?: -ENOENT;
1445 pr_warn("%s: Unknown symbol %s (err %d)\n",
1446 mod->name, name, ret);
1450 /* Divert to percpu allocation if a percpu var. */
1451 if (sym[i].st_shndx == info->index.pcpu)
1452 secbase = (unsigned long)mod_percpu(mod);
1454 secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1455 sym[i].st_value += secbase;
1463 static int apply_relocations(struct module *mod, const struct load_info *info)
1468 /* Now do relocations. */
1469 for (i = 1; i < info->hdr->e_shnum; i++) {
1470 unsigned int infosec = info->sechdrs[i].sh_info;
1472 /* Not a valid relocation section? */
1473 if (infosec >= info->hdr->e_shnum)
1476 /* Don't bother with non-allocated sections */
1477 if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC))
1480 if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1481 err = klp_apply_section_relocs(mod, info->sechdrs,
1486 else if (info->sechdrs[i].sh_type == SHT_REL)
1487 err = apply_relocate(info->sechdrs, info->strtab,
1488 info->index.sym, i, mod);
1489 else if (info->sechdrs[i].sh_type == SHT_RELA)
1490 err = apply_relocate_add(info->sechdrs, info->strtab,
1491 info->index.sym, i, mod);
1498 /* Additional bytes needed by arch in front of individual sections */
1499 unsigned int __weak arch_mod_section_prepend(struct module *mod,
1500 unsigned int section)
1502 /* default implementation just returns zero */
1506 long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1507 Elf_Shdr *sechdr, unsigned int section)
1510 long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1512 mod->mem[type].size += arch_mod_section_prepend(mod, section);
1513 offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1514 mod->mem[type].size = offset + sechdr->sh_size;
1516 WARN_ON_ONCE(offset & mask);
1517 return offset | mask;
1520 bool module_init_layout_section(const char *sname)
1522 #ifndef CONFIG_MODULE_UNLOAD
1523 if (module_exit_section(sname))
1526 return module_init_section(sname);
1529 static void __layout_sections(struct module *mod, struct load_info *info, bool is_init)
1533 static const unsigned long masks[][2] = {
1535 * NOTE: all executable code must be the first section
1536 * in this array; otherwise modify the text_size
1537 * finder in the two loops below
1539 { SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1540 { SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1541 { SHF_RO_AFTER_INIT | SHF_ALLOC, ARCH_SHF_SMALL },
1542 { SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1543 { ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1545 static const int core_m_to_mem_type[] = {
1552 static const int init_m_to_mem_type[] = {
1560 for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1561 enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1563 for (i = 0; i < info->hdr->e_shnum; ++i) {
1564 Elf_Shdr *s = &info->sechdrs[i];
1565 const char *sname = info->secstrings + s->sh_name;
1567 if ((s->sh_flags & masks[m][0]) != masks[m][0]
1568 || (s->sh_flags & masks[m][1])
1569 || s->sh_entsize != ~0UL
1570 || is_init != module_init_layout_section(sname))
1573 if (WARN_ON_ONCE(type == MOD_INVALID))
1576 s->sh_entsize = module_get_offset_and_type(mod, type, s, i);
1577 pr_debug("\t%s\n", sname);
1583 * Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1584 * might -- code, read-only data, read-write data, small data. Tally
1585 * sizes, and place the offsets into sh_entsize fields: high bit means it
1588 static void layout_sections(struct module *mod, struct load_info *info)
1592 for (i = 0; i < info->hdr->e_shnum; i++)
1593 info->sechdrs[i].sh_entsize = ~0UL;
1595 pr_debug("Core section allocation order for %s:\n", mod->name);
1596 __layout_sections(mod, info, false);
1598 pr_debug("Init section allocation order for %s:\n", mod->name);
1599 __layout_sections(mod, info, true);
1602 static void module_license_taint_check(struct module *mod, const char *license)
1605 license = "unspecified";
1607 if (!license_is_gpl_compatible(license)) {
1608 if (!test_taint(TAINT_PROPRIETARY_MODULE))
1609 pr_warn("%s: module license '%s' taints kernel.\n",
1610 mod->name, license);
1611 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1612 LOCKDEP_NOW_UNRELIABLE);
1616 static void setup_modinfo(struct module *mod, struct load_info *info)
1618 struct module_attribute *attr;
1621 for (i = 0; (attr = modinfo_attrs[i]); i++) {
1623 attr->setup(mod, get_modinfo(info, attr->attr.name));
1627 static void free_modinfo(struct module *mod)
1629 struct module_attribute *attr;
1632 for (i = 0; (attr = modinfo_attrs[i]); i++) {
1638 bool __weak module_init_section(const char *name)
1640 return strstarts(name, ".init");
1643 bool __weak module_exit_section(const char *name)
1645 return strstarts(name, ".exit");
1648 static int validate_section_offset(struct load_info *info, Elf_Shdr *shdr)
1650 #if defined(CONFIG_64BIT)
1651 unsigned long long secend;
1653 unsigned long secend;
1657 * Check for both overflow and offset/size being
1660 secend = shdr->sh_offset + shdr->sh_size;
1661 if (secend < shdr->sh_offset || secend > info->len)
1668 * Check userspace passed ELF module against our expectations, and cache
1669 * useful variables for further processing as we go.
1671 * This does basic validity checks against section offsets and sizes, the
1672 * section name string table, and the indices used for it (sh_name).
1674 * As a last step, since we're already checking the ELF sections we cache
1675 * useful variables which will be used later for our convenience:
1677 * o pointers to section headers
1678 * o cache the modinfo symbol section
1679 * o cache the string symbol section
1680 * o cache the module section
1682 * As a last step we set info->mod to the temporary copy of the module in
1683 * info->hdr. The final one will be allocated in move_module(). Any
1684 * modifications we make to our copy of the module will be carried over
1685 * to the final minted module.
1687 static int elf_validity_cache_copy(struct load_info *info, int flags)
1690 Elf_Shdr *shdr, *strhdr;
1692 unsigned int num_mod_secs = 0, mod_idx;
1693 unsigned int num_info_secs = 0, info_idx;
1694 unsigned int num_sym_secs = 0, sym_idx;
1696 if (info->len < sizeof(*(info->hdr))) {
1697 pr_err("Invalid ELF header len %lu\n", info->len);
1701 if (memcmp(info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1702 pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1705 if (info->hdr->e_type != ET_REL) {
1706 pr_err("Invalid ELF header type: %u != %u\n",
1707 info->hdr->e_type, ET_REL);
1710 if (!elf_check_arch(info->hdr)) {
1711 pr_err("Invalid architecture in ELF header: %u\n",
1712 info->hdr->e_machine);
1715 if (!module_elf_check_arch(info->hdr)) {
1716 pr_err("Invalid module architecture in ELF header: %u\n",
1717 info->hdr->e_machine);
1720 if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1721 pr_err("Invalid ELF section header size\n");
1726 * e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1727 * known and small. So e_shnum * sizeof(Elf_Shdr)
1728 * will not overflow unsigned long on any platform.
1730 if (info->hdr->e_shoff >= info->len
1731 || (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1732 info->len - info->hdr->e_shoff)) {
1733 pr_err("Invalid ELF section header overflow\n");
1737 info->sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1740 * Verify if the section name table index is valid.
1742 if (info->hdr->e_shstrndx == SHN_UNDEF
1743 || info->hdr->e_shstrndx >= info->hdr->e_shnum) {
1744 pr_err("Invalid ELF section name index: %d || e_shstrndx (%d) >= e_shnum (%d)\n",
1745 info->hdr->e_shstrndx, info->hdr->e_shstrndx,
1746 info->hdr->e_shnum);
1750 strhdr = &info->sechdrs[info->hdr->e_shstrndx];
1751 err = validate_section_offset(info, strhdr);
1753 pr_err("Invalid ELF section hdr(type %u)\n", strhdr->sh_type);
1758 * The section name table must be NUL-terminated, as required
1759 * by the spec. This makes strcmp and pr_* calls that access
1760 * strings in the section safe.
1762 info->secstrings = (void *)info->hdr + strhdr->sh_offset;
1763 if (strhdr->sh_size == 0) {
1764 pr_err("empty section name table\n");
1767 if (info->secstrings[strhdr->sh_size - 1] != '\0') {
1768 pr_err("ELF Spec violation: section name table isn't null terminated\n");
1773 * The code assumes that section 0 has a length of zero and
1774 * an addr of zero, so check for it.
1776 if (info->sechdrs[0].sh_type != SHT_NULL
1777 || info->sechdrs[0].sh_size != 0
1778 || info->sechdrs[0].sh_addr != 0) {
1779 pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1780 info->sechdrs[0].sh_type);
1784 for (i = 1; i < info->hdr->e_shnum; i++) {
1785 shdr = &info->sechdrs[i];
1786 switch (shdr->sh_type) {
1791 if (shdr->sh_link == SHN_UNDEF
1792 || shdr->sh_link >= info->hdr->e_shnum) {
1793 pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
1794 shdr->sh_link, shdr->sh_link,
1795 info->hdr->e_shnum);
1802 err = validate_section_offset(info, shdr);
1804 pr_err("Invalid ELF section in module (section %u type %u)\n",
1808 if (strcmp(info->secstrings + shdr->sh_name,
1809 ".gnu.linkonce.this_module") == 0) {
1812 } else if (strcmp(info->secstrings + shdr->sh_name,
1818 if (shdr->sh_flags & SHF_ALLOC) {
1819 if (shdr->sh_name >= strhdr->sh_size) {
1820 pr_err("Invalid ELF section name in module (section %u type %u)\n",
1829 if (num_info_secs > 1) {
1830 pr_err("Only one .modinfo section must exist.\n");
1832 } else if (num_info_secs == 1) {
1833 /* Try to find a name early so we can log errors with a module name */
1834 info->index.info = info_idx;
1835 info->name = get_modinfo(info, "name");
1838 if (num_sym_secs != 1) {
1839 pr_warn("%s: module has no symbols (stripped?)\n",
1840 info->name ?: "(missing .modinfo section or name field)");
1844 /* Sets internal symbols and strings. */
1845 info->index.sym = sym_idx;
1846 shdr = &info->sechdrs[sym_idx];
1847 info->index.str = shdr->sh_link;
1848 info->strtab = (char *)info->hdr + info->sechdrs[info->index.str].sh_offset;
1851 * The ".gnu.linkonce.this_module" ELF section is special. It is
1852 * what modpost uses to refer to __this_module and let's use rely
1853 * on THIS_MODULE to point to &__this_module properly. The kernel's
1854 * modpost declares it on each modules's *.mod.c file. If the struct
1855 * module of the kernel changes a full kernel rebuild is required.
1857 * We have a few expectaions for this special section, the following
1858 * code validates all this for us:
1860 * o Only one section must exist
1861 * o We expect the kernel to always have to allocate it: SHF_ALLOC
1862 * o The section size must match the kernel's run time's struct module
1865 if (num_mod_secs != 1) {
1866 pr_err("module %s: Only one .gnu.linkonce.this_module section must exist.\n",
1867 info->name ?: "(missing .modinfo section or name field)");
1871 shdr = &info->sechdrs[mod_idx];
1874 * This is already implied on the switch above, however let's be
1875 * pedantic about it.
1877 if (shdr->sh_type == SHT_NOBITS) {
1878 pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
1879 info->name ?: "(missing .modinfo section or name field)");
1883 if (!(shdr->sh_flags & SHF_ALLOC)) {
1884 pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
1885 info->name ?: "(missing .modinfo section or name field)");
1889 if (shdr->sh_size != sizeof(struct module)) {
1890 pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
1891 info->name ?: "(missing .modinfo section or name field)");
1895 info->index.mod = mod_idx;
1897 /* This is temporary: point mod into copy of data. */
1898 info->mod = (void *)info->hdr + shdr->sh_offset;
1901 * If we didn't load the .modinfo 'name' field earlier, fall back to
1902 * on-disk struct mod 'name' field.
1905 info->name = info->mod->name;
1907 if (flags & MODULE_INIT_IGNORE_MODVERSIONS)
1908 info->index.vers = 0; /* Pretend no __versions section! */
1910 info->index.vers = find_sec(info, "__versions");
1912 info->index.pcpu = find_pcpusec(info);
1920 #define COPY_CHUNK_SIZE (16*PAGE_SIZE)
1922 static int copy_chunked_from_user(void *dst, const void __user *usrc, unsigned long len)
1925 unsigned long n = min(len, COPY_CHUNK_SIZE);
1927 if (copy_from_user(dst, usrc, n) != 0)
1937 static int check_modinfo_livepatch(struct module *mod, struct load_info *info)
1939 if (!get_modinfo(info, "livepatch"))
1940 /* Nothing more to do */
1943 if (set_livepatch_module(mod))
1946 pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
1951 static void check_modinfo_retpoline(struct module *mod, struct load_info *info)
1953 if (retpoline_module_ok(get_modinfo(info, "retpoline")))
1956 pr_warn("%s: loading module not compiled with retpoline compiler.\n",
1960 /* Sets info->hdr and info->len. */
1961 static int copy_module_from_user(const void __user *umod, unsigned long len,
1962 struct load_info *info)
1967 if (info->len < sizeof(*(info->hdr)))
1970 err = security_kernel_load_data(LOADING_MODULE, true);
1974 /* Suck in entire file: we'll want most of it. */
1975 info->hdr = __vmalloc(info->len, GFP_KERNEL | __GFP_NOWARN);
1979 if (copy_chunked_from_user(info->hdr, umod, info->len) != 0) {
1984 err = security_kernel_post_load_data((char *)info->hdr, info->len,
1985 LOADING_MODULE, "init_module");
1993 static void free_copy(struct load_info *info, int flags)
1995 if (flags & MODULE_INIT_COMPRESSED_FILE)
1996 module_decompress_cleanup(info);
2001 static int rewrite_section_headers(struct load_info *info, int flags)
2005 /* This should always be true, but let's be sure. */
2006 info->sechdrs[0].sh_addr = 0;
2008 for (i = 1; i < info->hdr->e_shnum; i++) {
2009 Elf_Shdr *shdr = &info->sechdrs[i];
2012 * Mark all sections sh_addr with their address in the
2015 shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2019 /* Track but don't keep modinfo and version sections. */
2020 info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2021 info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2027 * These calls taint the kernel depending certain module circumstances */
2028 static void module_augment_kernel_taints(struct module *mod, struct load_info *info)
2030 int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2032 if (!get_modinfo(info, "intree")) {
2033 if (!test_taint(TAINT_OOT_MODULE))
2034 pr_warn("%s: loading out-of-tree module taints kernel.\n",
2036 add_taint_module(mod, TAINT_OOT_MODULE, LOCKDEP_STILL_OK);
2039 check_modinfo_retpoline(mod, info);
2041 if (get_modinfo(info, "staging")) {
2042 add_taint_module(mod, TAINT_CRAP, LOCKDEP_STILL_OK);
2043 pr_warn("%s: module is from the staging directory, the quality "
2044 "is unknown, you have been warned.\n", mod->name);
2047 if (is_livepatch_module(mod)) {
2048 add_taint_module(mod, TAINT_LIVEPATCH, LOCKDEP_STILL_OK);
2049 pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2053 module_license_taint_check(mod, get_modinfo(info, "license"));
2055 if (get_modinfo(info, "test")) {
2056 if (!test_taint(TAINT_TEST))
2057 pr_warn("%s: loading test module taints kernel.\n",
2059 add_taint_module(mod, TAINT_TEST, LOCKDEP_STILL_OK);
2061 #ifdef CONFIG_MODULE_SIG
2062 mod->sig_ok = info->sig_ok;
2064 pr_notice_once("%s: module verification failed: signature "
2065 "and/or required key missing - tainting "
2066 "kernel\n", mod->name);
2067 add_taint_module(mod, TAINT_UNSIGNED_MODULE, LOCKDEP_STILL_OK);
2072 * ndiswrapper is under GPL by itself, but loads proprietary modules.
2073 * Don't use add_taint_module(), as it would prevent ndiswrapper from
2074 * using GPL-only symbols it needs.
2076 if (strcmp(mod->name, "ndiswrapper") == 0)
2077 add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2079 /* driverloader was caught wrongly pretending to be under GPL */
2080 if (strcmp(mod->name, "driverloader") == 0)
2081 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2082 LOCKDEP_NOW_UNRELIABLE);
2084 /* lve claims to be GPL but upstream won't provide source */
2085 if (strcmp(mod->name, "lve") == 0)
2086 add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2087 LOCKDEP_NOW_UNRELIABLE);
2089 if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2090 pr_warn("%s: module license taints kernel.\n", mod->name);
2094 static int check_modinfo(struct module *mod, struct load_info *info, int flags)
2096 const char *modmagic = get_modinfo(info, "vermagic");
2099 if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2102 /* This is allowed: modprobe --force will invalidate it. */
2104 err = try_to_force_load(mod, "bad vermagic");
2107 } else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2108 pr_err("%s: version magic '%s' should be '%s'\n",
2109 info->name, modmagic, vermagic);
2113 err = check_modinfo_livepatch(mod, info);
2120 static int find_module_sections(struct module *mod, struct load_info *info)
2122 mod->kp = section_objs(info, "__param",
2123 sizeof(*mod->kp), &mod->num_kp);
2124 mod->syms = section_objs(info, "__ksymtab",
2125 sizeof(*mod->syms), &mod->num_syms);
2126 mod->crcs = section_addr(info, "__kcrctab");
2127 mod->gpl_syms = section_objs(info, "__ksymtab_gpl",
2128 sizeof(*mod->gpl_syms),
2129 &mod->num_gpl_syms);
2130 mod->gpl_crcs = section_addr(info, "__kcrctab_gpl");
2132 #ifdef CONFIG_CONSTRUCTORS
2133 mod->ctors = section_objs(info, ".ctors",
2134 sizeof(*mod->ctors), &mod->num_ctors);
2136 mod->ctors = section_objs(info, ".init_array",
2137 sizeof(*mod->ctors), &mod->num_ctors);
2138 else if (find_sec(info, ".init_array")) {
2140 * This shouldn't happen with same compiler and binutils
2141 * building all parts of the module.
2143 pr_warn("%s: has both .ctors and .init_array.\n",
2149 mod->noinstr_text_start = section_objs(info, ".noinstr.text", 1,
2150 &mod->noinstr_text_size);
2152 #ifdef CONFIG_TRACEPOINTS
2153 mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2154 sizeof(*mod->tracepoints_ptrs),
2155 &mod->num_tracepoints);
2157 #ifdef CONFIG_TREE_SRCU
2158 mod->srcu_struct_ptrs = section_objs(info, "___srcu_struct_ptrs",
2159 sizeof(*mod->srcu_struct_ptrs),
2160 &mod->num_srcu_structs);
2162 #ifdef CONFIG_BPF_EVENTS
2163 mod->bpf_raw_events = section_objs(info, "__bpf_raw_tp_map",
2164 sizeof(*mod->bpf_raw_events),
2165 &mod->num_bpf_raw_events);
2167 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2168 mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2170 #ifdef CONFIG_JUMP_LABEL
2171 mod->jump_entries = section_objs(info, "__jump_table",
2172 sizeof(*mod->jump_entries),
2173 &mod->num_jump_entries);
2175 #ifdef CONFIG_EVENT_TRACING
2176 mod->trace_events = section_objs(info, "_ftrace_events",
2177 sizeof(*mod->trace_events),
2178 &mod->num_trace_events);
2179 mod->trace_evals = section_objs(info, "_ftrace_eval_map",
2180 sizeof(*mod->trace_evals),
2181 &mod->num_trace_evals);
2183 #ifdef CONFIG_TRACING
2184 mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2185 sizeof(*mod->trace_bprintk_fmt_start),
2186 &mod->num_trace_bprintk_fmt);
2188 #ifdef CONFIG_FTRACE_MCOUNT_RECORD
2189 /* sechdrs[0].sh_size is always zero */
2190 mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2191 sizeof(*mod->ftrace_callsites),
2192 &mod->num_ftrace_callsites);
2194 #ifdef CONFIG_FUNCTION_ERROR_INJECTION
2195 mod->ei_funcs = section_objs(info, "_error_injection_whitelist",
2196 sizeof(*mod->ei_funcs),
2197 &mod->num_ei_funcs);
2199 #ifdef CONFIG_KPROBES
2200 mod->kprobes_text_start = section_objs(info, ".kprobes.text", 1,
2201 &mod->kprobes_text_size);
2202 mod->kprobe_blacklist = section_objs(info, "_kprobe_blacklist",
2203 sizeof(unsigned long),
2204 &mod->num_kprobe_blacklist);
2206 #ifdef CONFIG_PRINTK_INDEX
2207 mod->printk_index_start = section_objs(info, ".printk_index",
2208 sizeof(*mod->printk_index_start),
2209 &mod->printk_index_size);
2211 #ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2212 mod->static_call_sites = section_objs(info, ".static_call_sites",
2213 sizeof(*mod->static_call_sites),
2214 &mod->num_static_call_sites);
2216 #if IS_ENABLED(CONFIG_KUNIT)
2217 mod->kunit_suites = section_objs(info, ".kunit_test_suites",
2218 sizeof(*mod->kunit_suites),
2219 &mod->num_kunit_suites);
2220 mod->kunit_init_suites = section_objs(info, ".kunit_init_test_suites",
2221 sizeof(*mod->kunit_init_suites),
2222 &mod->num_kunit_init_suites);
2225 mod->extable = section_objs(info, "__ex_table",
2226 sizeof(*mod->extable), &mod->num_exentries);
2228 if (section_addr(info, "__obsparm"))
2229 pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2231 #ifdef CONFIG_DYNAMIC_DEBUG_CORE
2232 mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
2233 sizeof(*mod->dyndbg_info.descs),
2234 &mod->dyndbg_info.num_descs);
2235 mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
2236 sizeof(*mod->dyndbg_info.classes),
2237 &mod->dyndbg_info.num_classes);
2243 static int move_module(struct module *mod, struct load_info *info)
2246 enum mod_mem_type t = 0;
2249 for_each_mod_mem_type(type) {
2250 if (!mod->mem[type].size) {
2251 mod->mem[type].base = NULL;
2255 ret = module_memory_alloc(mod, type);
2262 /* Transfer each section which specifies SHF_ALLOC */
2263 pr_debug("Final section addresses for %s:\n", mod->name);
2264 for (i = 0; i < info->hdr->e_shnum; i++) {
2266 Elf_Shdr *shdr = &info->sechdrs[i];
2267 enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2269 if (!(shdr->sh_flags & SHF_ALLOC))
2272 dest = mod->mem[type].base + (shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK);
2274 if (shdr->sh_type != SHT_NOBITS) {
2276 * Our ELF checker already validated this, but let's
2277 * be pedantic and make the goal clearer. We actually
2278 * end up copying over all modifications made to the
2279 * userspace copy of the entire struct module.
2281 if (i == info->index.mod &&
2282 (WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2286 memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2289 * Update the userspace copy's ELF section address to point to
2290 * our newly allocated memory as a pure convenience so that
2291 * users of info can keep taking advantage and using the newly
2292 * minted official memory area.
2294 shdr->sh_addr = (unsigned long)dest;
2295 pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2296 (long)shdr->sh_size, info->secstrings + shdr->sh_name);
2301 for (t--; t >= 0; t--)
2302 module_memory_free(mod, t, true);
2306 static int check_export_symbol_versions(struct module *mod)
2308 #ifdef CONFIG_MODVERSIONS
2309 if ((mod->num_syms && !mod->crcs) ||
2310 (mod->num_gpl_syms && !mod->gpl_crcs)) {
2311 return try_to_force_load(mod,
2312 "no versions for exported symbols");
2318 static void flush_module_icache(const struct module *mod)
2321 * Flush the instruction cache, since we've played with text.
2322 * Do it before processing of module parameters, so the module
2323 * can provide parameter accessor functions of its own.
2325 for_each_mod_mem_type(type) {
2326 const struct module_memory *mod_mem = &mod->mem[type];
2328 if (mod_mem->size) {
2329 flush_icache_range((unsigned long)mod_mem->base,
2330 (unsigned long)mod_mem->base + mod_mem->size);
2335 bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2340 int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2348 /* module_blacklist is a comma-separated list of module names */
2349 static char *module_blacklist;
2350 static bool blacklisted(const char *module_name)
2355 if (!module_blacklist)
2358 for (p = module_blacklist; *p; p += len) {
2359 len = strcspn(p, ",");
2360 if (strlen(module_name) == len && !memcmp(module_name, p, len))
2367 core_param(module_blacklist, module_blacklist, charp, 0400);
2369 static struct module *layout_and_allocate(struct load_info *info, int flags)
2375 /* Allow arches to frob section contents and sizes. */
2376 err = module_frob_arch_sections(info->hdr, info->sechdrs,
2377 info->secstrings, info->mod);
2379 return ERR_PTR(err);
2381 err = module_enforce_rwx_sections(info->hdr, info->sechdrs,
2382 info->secstrings, info->mod);
2384 return ERR_PTR(err);
2386 /* We will do a special allocation for per-cpu sections later. */
2387 info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2390 * Mark ro_after_init section with SHF_RO_AFTER_INIT so that
2391 * layout_sections() can put it in the right place.
2392 * Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2394 ndx = find_sec(info, ".data..ro_after_init");
2396 info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2398 * Mark the __jump_table section as ro_after_init as well: these data
2399 * structures are never modified, with the exception of entries that
2400 * refer to code in the __init section, which are annotated as such
2401 * at module load time.
2403 ndx = find_sec(info, "__jump_table");
2405 info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2408 * Determine total sizes, and put offsets in sh_entsize. For now
2409 * this is done generically; there doesn't appear to be any
2410 * special cases for the architectures.
2412 layout_sections(info->mod, info);
2413 layout_symtab(info->mod, info);
2415 /* Allocate and move to the final place */
2416 err = move_module(info->mod, info);
2418 return ERR_PTR(err);
2420 /* Module has been copied to its final place now: return it. */
2421 mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2422 kmemleak_load_module(mod, info);
2426 /* mod is no longer valid after this! */
2427 static void module_deallocate(struct module *mod, struct load_info *info)
2429 percpu_modfree(mod);
2430 module_arch_freeing_init(mod);
2432 free_mod_mem(mod, true);
2435 int __weak module_finalize(const Elf_Ehdr *hdr,
2436 const Elf_Shdr *sechdrs,
2442 static int post_relocation(struct module *mod, const struct load_info *info)
2444 /* Sort exception table now relocations are done. */
2445 sort_extable(mod->extable, mod->extable + mod->num_exentries);
2447 /* Copy relocated percpu area over. */
2448 percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
2449 info->sechdrs[info->index.pcpu].sh_size);
2451 /* Setup kallsyms-specific fields. */
2452 add_kallsyms(mod, info);
2454 /* Arch-specific module finalizing. */
2455 return module_finalize(info->hdr, info->sechdrs, mod);
2458 /* Call module constructors. */
2459 static void do_mod_ctors(struct module *mod)
2461 #ifdef CONFIG_CONSTRUCTORS
2464 for (i = 0; i < mod->num_ctors; i++)
2469 /* For freeing module_init on success, in case kallsyms traversing */
2470 struct mod_initfree {
2471 struct llist_node node;
2477 static void do_free_init(struct work_struct *w)
2479 struct llist_node *pos, *n, *list;
2480 struct mod_initfree *initfree;
2482 list = llist_del_all(&init_free_list);
2486 llist_for_each_safe(pos, n, list) {
2487 initfree = container_of(pos, struct mod_initfree, node);
2488 execmem_free(initfree->init_text);
2489 execmem_free(initfree->init_data);
2490 execmem_free(initfree->init_rodata);
2495 void flush_module_init_free_work(void)
2497 flush_work(&init_free_wq);
2500 #undef MODULE_PARAM_PREFIX
2501 #define MODULE_PARAM_PREFIX "module."
2502 /* Default value for module->async_probe_requested */
2503 static bool async_probe;
2504 module_param(async_probe, bool, 0644);
2507 * This is where the real work happens.
2509 * Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
2510 * helper command 'lx-symbols'.
2512 static noinline int do_init_module(struct module *mod)
2515 struct mod_initfree *freeinit;
2516 #if defined(CONFIG_MODULE_STATS)
2517 unsigned int text_size = 0, total_size = 0;
2519 for_each_mod_mem_type(type) {
2520 const struct module_memory *mod_mem = &mod->mem[type];
2521 if (mod_mem->size) {
2522 total_size += mod_mem->size;
2523 if (type == MOD_TEXT || type == MOD_INIT_TEXT)
2524 text_size += mod_mem->size;
2529 freeinit = kmalloc(sizeof(*freeinit), GFP_KERNEL);
2534 freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
2535 freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
2536 freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
2539 /* Start the module */
2540 if (mod->init != NULL)
2541 ret = do_one_initcall(mod->init);
2543 goto fail_free_freeinit;
2546 pr_warn("%s: '%s'->init suspiciously returned %d, it should "
2547 "follow 0/-E convention\n"
2548 "%s: loading module anyway...\n",
2549 __func__, mod->name, ret, __func__);
2553 /* Now it's a first class citizen! */
2554 mod->state = MODULE_STATE_LIVE;
2555 blocking_notifier_call_chain(&module_notify_list,
2556 MODULE_STATE_LIVE, mod);
2558 /* Delay uevent until module has finished its init routine */
2559 kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
2562 * We need to finish all async code before the module init sequence
2563 * is done. This has potential to deadlock if synchronous module
2564 * loading is requested from async (which is not allowed!).
2566 * See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
2567 * request_module() from async workers") for more details.
2569 if (!mod->async_probe_requested)
2570 async_synchronize_full();
2572 ftrace_free_mem(mod, mod->mem[MOD_INIT_TEXT].base,
2573 mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
2574 mutex_lock(&module_mutex);
2575 /* Drop initial reference. */
2577 trim_init_extable(mod);
2578 #ifdef CONFIG_KALLSYMS
2579 /* Switch to core kallsyms now init is done: kallsyms may be walking! */
2580 rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
2582 ret = module_enable_rodata_ro(mod, true);
2584 goto fail_mutex_unlock;
2585 mod_tree_remove_init(mod);
2586 module_arch_freeing_init(mod);
2587 for_class_mod_mem_type(type, init) {
2588 mod->mem[type].base = NULL;
2589 mod->mem[type].size = 0;
2592 #ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2593 /* .BTF is not SHF_ALLOC and will get removed, so sanitize pointer */
2594 mod->btf_data = NULL;
2597 * We want to free module_init, but be aware that kallsyms may be
2598 * walking this with preempt disabled. In all the failure paths, we
2599 * call synchronize_rcu(), but we don't want to slow down the success
2600 * path. execmem_free() cannot be called in an interrupt, so do the
2601 * work and call synchronize_rcu() in a work queue.
2603 * Note that execmem_alloc() on most architectures creates W+X page
2604 * mappings which won't be cleaned up until do_free_init() runs. Any
2605 * code such as mark_rodata_ro() which depends on those mappings to
2606 * be cleaned up needs to sync with the queued work by invoking
2607 * flush_module_init_free_work().
2609 if (llist_add(&freeinit->node, &init_free_list))
2610 schedule_work(&init_free_wq);
2612 mutex_unlock(&module_mutex);
2613 wake_up_all(&module_wq);
2615 mod_stat_add_long(text_size, &total_text_size);
2616 mod_stat_add_long(total_size, &total_mod_size);
2618 mod_stat_inc(&modcount);
2623 mutex_unlock(&module_mutex);
2627 /* Try to protect us from buggy refcounters. */
2628 mod->state = MODULE_STATE_GOING;
2631 blocking_notifier_call_chain(&module_notify_list,
2632 MODULE_STATE_GOING, mod);
2633 klp_module_going(mod);
2634 ftrace_release_mod(mod);
2636 wake_up_all(&module_wq);
2641 static int may_init_module(void)
2643 if (!capable(CAP_SYS_MODULE) || modules_disabled)
2649 /* Is this module of this name done loading? No locks held. */
2650 static bool finished_loading(const char *name)
2656 * The module_mutex should not be a heavily contended lock;
2657 * if we get the occasional sleep here, we'll go an extra iteration
2658 * in the wait_event_interruptible(), which is harmless.
2660 sched_annotate_sleep();
2661 mutex_lock(&module_mutex);
2662 mod = find_module_all(name, strlen(name), true);
2663 ret = !mod || mod->state == MODULE_STATE_LIVE
2664 || mod->state == MODULE_STATE_GOING;
2665 mutex_unlock(&module_mutex);
2670 /* Must be called with module_mutex held */
2671 static int module_patient_check_exists(const char *name,
2672 enum fail_dup_mod_reason reason)
2677 old = find_module_all(name, strlen(name), true);
2681 if (old->state == MODULE_STATE_COMING ||
2682 old->state == MODULE_STATE_UNFORMED) {
2683 /* Wait in case it fails to load. */
2684 mutex_unlock(&module_mutex);
2685 err = wait_event_interruptible(module_wq,
2686 finished_loading(name));
2687 mutex_lock(&module_mutex);
2691 /* The module might have gone in the meantime. */
2692 old = find_module_all(name, strlen(name), true);
2695 if (try_add_failed_module(name, reason))
2696 pr_warn("Could not add fail-tracking for module: %s\n", name);
2699 * We are here only when the same module was being loaded. Do
2700 * not try to load it again right now. It prevents long delays
2701 * caused by serialized module load failures. It might happen
2702 * when more devices of the same type trigger load of
2703 * a particular module.
2705 if (old && old->state == MODULE_STATE_LIVE)
2711 * We try to place it in the list now to make sure it's unique before
2712 * we dedicate too many resources. In particular, temporary percpu
2713 * memory exhaustion.
2715 static int add_unformed_module(struct module *mod)
2719 mod->state = MODULE_STATE_UNFORMED;
2721 mutex_lock(&module_mutex);
2722 err = module_patient_check_exists(mod->name, FAIL_DUP_MOD_LOAD);
2726 mod_update_bounds(mod);
2727 list_add_rcu(&mod->list, &modules);
2728 mod_tree_insert(mod);
2732 mutex_unlock(&module_mutex);
2736 static int complete_formation(struct module *mod, struct load_info *info)
2740 mutex_lock(&module_mutex);
2742 /* Find duplicate symbols (must be called under lock). */
2743 err = verify_exported_symbols(mod);
2747 /* These rely on module_mutex for list integrity. */
2748 module_bug_finalize(info->hdr, info->sechdrs, mod);
2749 module_cfi_finalize(info->hdr, info->sechdrs, mod);
2751 err = module_enable_rodata_ro(mod, false);
2753 goto out_strict_rwx;
2754 err = module_enable_data_nx(mod);
2756 goto out_strict_rwx;
2757 err = module_enable_text_rox(mod);
2759 goto out_strict_rwx;
2762 * Mark state as coming so strong_try_module_get() ignores us,
2763 * but kallsyms etc. can see us.
2765 mod->state = MODULE_STATE_COMING;
2766 mutex_unlock(&module_mutex);
2771 module_bug_cleanup(mod);
2773 mutex_unlock(&module_mutex);
2777 static int prepare_coming_module(struct module *mod)
2781 ftrace_module_enable(mod);
2782 err = klp_module_coming(mod);
2786 err = blocking_notifier_call_chain_robust(&module_notify_list,
2787 MODULE_STATE_COMING, MODULE_STATE_GOING, mod);
2788 err = notifier_to_errno(err);
2790 klp_module_going(mod);
2795 static int unknown_module_param_cb(char *param, char *val, const char *modname,
2798 struct module *mod = arg;
2801 if (strcmp(param, "async_probe") == 0) {
2802 if (kstrtobool(val, &mod->async_probe_requested))
2803 mod->async_probe_requested = true;
2807 /* Check for magic 'dyndbg' arg */
2808 ret = ddebug_dyndbg_module_param_cb(param, val, modname);
2810 pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
2814 /* Module within temporary copy, this doesn't do any allocation */
2815 static int early_mod_check(struct load_info *info, int flags)
2820 * Now that we know we have the correct module name, check
2821 * if it's blacklisted.
2823 if (blacklisted(info->name)) {
2824 pr_err("Module %s is blacklisted\n", info->name);
2828 err = rewrite_section_headers(info, flags);
2832 /* Check module struct version now, before we try to use module. */
2833 if (!check_modstruct_version(info, info->mod))
2836 err = check_modinfo(info->mod, info, flags);
2840 mutex_lock(&module_mutex);
2841 err = module_patient_check_exists(info->mod->name, FAIL_DUP_MOD_BECOMING);
2842 mutex_unlock(&module_mutex);
2848 * Allocate and load the module: note that size of section 0 is always
2849 * zero, and we rely on this for optional sections.
2851 static int load_module(struct load_info *info, const char __user *uargs,
2855 bool module_allocated = false;
2860 * Do the signature check (if any) first. All that
2861 * the signature check needs is info->len, it does
2862 * not need any of the section info. That can be
2863 * set up later. This will minimize the chances
2864 * of a corrupt module causing problems before
2865 * we even get to the signature check.
2867 * The check will also adjust info->len by stripping
2868 * off the sig length at the end of the module, making
2869 * checks against info->len more correct.
2871 err = module_sig_check(info, flags);
2876 * Do basic sanity checks against the ELF header and
2877 * sections. Cache useful sections and set the
2878 * info->mod to the userspace passed struct module.
2880 err = elf_validity_cache_copy(info, flags);
2884 err = early_mod_check(info, flags);
2888 /* Figure out module layout, and allocate all the memory. */
2889 mod = layout_and_allocate(info, flags);
2895 module_allocated = true;
2897 audit_log_kern_module(mod->name);
2899 /* Reserve our place in the list. */
2900 err = add_unformed_module(mod);
2905 * We are tainting your kernel if your module gets into
2906 * the modules linked list somehow.
2908 module_augment_kernel_taints(mod, info);
2910 /* To avoid stressing percpu allocator, do this once we're unique. */
2911 err = percpu_modalloc(mod, info);
2915 /* Now module is in final location, initialize linked lists, etc. */
2916 err = module_unload_init(mod);
2920 init_param_lock(mod);
2923 * Now we've got everything in the final locations, we can
2924 * find optional sections.
2926 err = find_module_sections(mod, info);
2930 err = check_export_symbol_versions(mod);
2934 /* Set up MODINFO_ATTR fields */
2935 setup_modinfo(mod, info);
2937 /* Fix up syms, so that st_value is a pointer to location. */
2938 err = simplify_symbols(mod, info);
2942 err = apply_relocations(mod, info);
2946 err = post_relocation(mod, info);
2950 flush_module_icache(mod);
2952 /* Now copy in args */
2953 mod->args = strndup_user(uargs, ~0UL >> 1);
2954 if (IS_ERR(mod->args)) {
2955 err = PTR_ERR(mod->args);
2956 goto free_arch_cleanup;
2959 init_build_id(mod, info);
2961 /* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
2962 ftrace_module_init(mod);
2964 /* Finally it's fully formed, ready to start executing. */
2965 err = complete_formation(mod, info);
2967 goto ddebug_cleanup;
2969 err = prepare_coming_module(mod);
2973 mod->async_probe_requested = async_probe;
2975 /* Module is ready to execute: parsing args may do that. */
2976 after_dashes = parse_args(mod->name, mod->args, mod->kp, mod->num_kp,
2978 unknown_module_param_cb);
2979 if (IS_ERR(after_dashes)) {
2980 err = PTR_ERR(after_dashes);
2981 goto coming_cleanup;
2982 } else if (after_dashes) {
2983 pr_warn("%s: parameters '%s' after `--' ignored\n",
2984 mod->name, after_dashes);
2987 /* Link in to sysfs. */
2988 err = mod_sysfs_setup(mod, info, mod->kp, mod->num_kp);
2990 goto coming_cleanup;
2992 if (is_livepatch_module(mod)) {
2993 err = copy_module_elf(mod, info);
2998 /* Get rid of temporary copy. */
2999 free_copy(info, flags);
3001 codetag_load_module(mod);
3004 trace_module_load(mod);
3006 return do_init_module(mod);
3009 mod_sysfs_teardown(mod);
3011 mod->state = MODULE_STATE_GOING;
3012 destroy_params(mod->kp, mod->num_kp);
3013 blocking_notifier_call_chain(&module_notify_list,
3014 MODULE_STATE_GOING, mod);
3015 klp_module_going(mod);
3017 mod->state = MODULE_STATE_GOING;
3018 /* module_bug_cleanup needs module_mutex protection */
3019 mutex_lock(&module_mutex);
3020 module_bug_cleanup(mod);
3021 mutex_unlock(&module_mutex);
3024 ftrace_release_mod(mod);
3028 module_arch_cleanup(mod);
3032 module_unload_free(mod);
3034 mutex_lock(&module_mutex);
3035 /* Unlink carefully: kallsyms could be walking list. */
3036 list_del_rcu(&mod->list);
3037 mod_tree_remove(mod);
3038 wake_up_all(&module_wq);
3039 /* Wait for RCU-sched synchronizing before releasing mod->list. */
3041 mutex_unlock(&module_mutex);
3043 mod_stat_bump_invalid(info, flags);
3044 /* Free lock-classes; relies on the preceding sync_rcu() */
3045 for_class_mod_mem_type(type, core_data) {
3046 lockdep_free_key_range(mod->mem[type].base,
3047 mod->mem[type].size);
3050 module_deallocate(mod, info);
3053 * The info->len is always set. We distinguish between
3054 * failures once the proper module was allocated and
3057 if (!module_allocated)
3058 mod_stat_bump_becoming(info, flags);
3059 free_copy(info, flags);
3063 SYSCALL_DEFINE3(init_module, void __user *, umod,
3064 unsigned long, len, const char __user *, uargs)
3067 struct load_info info = { };
3069 err = may_init_module();
3073 pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3076 err = copy_module_from_user(umod, len, &info);
3078 mod_stat_inc(&failed_kreads);
3079 mod_stat_add_long(len, &invalid_kread_bytes);
3083 return load_module(&info, uargs, 0);
3088 struct hlist_node entry;
3089 struct completion complete;
3093 #define IDEM_HASH_BITS 8
3094 static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3095 static DEFINE_SPINLOCK(idem_lock);
3097 static bool idempotent(struct idempotent *u, const void *cookie)
3099 int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3100 struct hlist_head *head = idem_hash + hash;
3101 struct idempotent *existing;
3106 init_completion(&u->complete);
3108 spin_lock(&idem_lock);
3110 hlist_for_each_entry(existing, head, entry) {
3111 if (existing->cookie != cookie)
3116 hlist_add_head(&u->entry, idem_hash + hash);
3117 spin_unlock(&idem_lock);
3123 * We were the first one with 'cookie' on the list, and we ended
3124 * up completing the operation. We now need to walk the list,
3125 * remove everybody - which includes ourselves - fill in the return
3126 * value, and then complete the operation.
3128 static int idempotent_complete(struct idempotent *u, int ret)
3130 const void *cookie = u->cookie;
3131 int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3132 struct hlist_head *head = idem_hash + hash;
3133 struct hlist_node *next;
3134 struct idempotent *pos;
3136 spin_lock(&idem_lock);
3137 hlist_for_each_entry_safe(pos, next, head, entry) {
3138 if (pos->cookie != cookie)
3140 hlist_del(&pos->entry);
3142 complete(&pos->complete);
3144 spin_unlock(&idem_lock);
3148 static int init_module_from_file(struct file *f, const char __user * uargs, int flags)
3150 struct load_info info = { };
3154 len = kernel_read_file(f, 0, &buf, INT_MAX, NULL, READING_MODULE);
3156 mod_stat_inc(&failed_kreads);
3160 if (flags & MODULE_INIT_COMPRESSED_FILE) {
3161 int err = module_decompress(&info, buf, len);
3162 vfree(buf); /* compressed data is no longer needed */
3164 mod_stat_inc(&failed_decompress);
3165 mod_stat_add_long(len, &invalid_decompress_bytes);
3173 return load_module(&info, uargs, flags);
3176 static int idempotent_init_module(struct file *f, const char __user * uargs, int flags)
3178 struct idempotent idem;
3180 if (!f || !(f->f_mode & FMODE_READ))
3183 /* See if somebody else is doing the operation? */
3184 if (idempotent(&idem, file_inode(f))) {
3185 wait_for_completion(&idem.complete);
3189 /* Otherwise, we'll do it and complete others */
3190 return idempotent_complete(&idem,
3191 init_module_from_file(f, uargs, flags));
3194 SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3199 err = may_init_module();
3203 pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3205 if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3206 |MODULE_INIT_IGNORE_VERMAGIC
3207 |MODULE_INIT_COMPRESSED_FILE))
3211 err = idempotent_init_module(f.file, uargs, flags);
3216 /* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
3217 char *module_flags(struct module *mod, char *buf, bool show_state)
3221 BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3222 if (!mod->taints && !show_state)
3225 mod->state == MODULE_STATE_GOING ||
3226 mod->state == MODULE_STATE_COMING) {
3228 bx += module_flags_taint(mod->taints, buf + bx);
3229 /* Show a - for module-is-being-unloaded */
3230 if (mod->state == MODULE_STATE_GOING && show_state)
3232 /* Show a + for module-is-being-loaded */
3233 if (mod->state == MODULE_STATE_COMING && show_state)
3243 /* Given an address, look for it in the module exception tables. */
3244 const struct exception_table_entry *search_module_extables(unsigned long addr)
3246 const struct exception_table_entry *e = NULL;
3250 mod = __module_address(addr);
3254 if (!mod->num_exentries)
3257 e = search_extable(mod->extable,
3264 * Now, if we found one, we are running inside it now, hence
3265 * we cannot unload the module, hence no refcnt needed.
3271 * is_module_address() - is this address inside a module?
3272 * @addr: the address to check.
3274 * See is_module_text_address() if you simply want to see if the address
3275 * is code (not data).
3277 bool is_module_address(unsigned long addr)
3282 ret = __module_address(addr) != NULL;
3289 * __module_address() - get the module which contains an address.
3290 * @addr: the address.
3292 * Must be called with preempt disabled or module mutex held so that
3293 * module doesn't get freed during this.
3295 struct module *__module_address(unsigned long addr)
3299 if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3302 #ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3303 if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3310 module_assert_mutex_or_preempt();
3312 mod = mod_find(addr, &mod_tree);
3314 BUG_ON(!within_module(addr, mod));
3315 if (mod->state == MODULE_STATE_UNFORMED)
3322 * is_module_text_address() - is this address inside module code?
3323 * @addr: the address to check.
3325 * See is_module_address() if you simply want to see if the address is
3326 * anywhere in a module. See kernel_text_address() for testing if an
3327 * address corresponds to kernel or module code.
3329 bool is_module_text_address(unsigned long addr)
3334 ret = __module_text_address(addr) != NULL;
3341 * __module_text_address() - get the module whose code contains an address.
3342 * @addr: the address.
3344 * Must be called with preempt disabled or module mutex held so that
3345 * module doesn't get freed during this.
3347 struct module *__module_text_address(unsigned long addr)
3349 struct module *mod = __module_address(addr);
3351 /* Make sure it's within the text section. */
3352 if (!within_module_mem_type(addr, mod, MOD_TEXT) &&
3353 !within_module_mem_type(addr, mod, MOD_INIT_TEXT))
3359 /* Don't grab lock, we're oopsing. */
3360 void print_modules(void)
3363 char buf[MODULE_FLAGS_BUF_SIZE];
3365 printk(KERN_DEFAULT "Modules linked in:");
3366 /* Most callers should already have preempt disabled, but make sure */
3368 list_for_each_entry_rcu(mod, &modules, list) {
3369 if (mod->state == MODULE_STATE_UNFORMED)
3371 pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3374 print_unloaded_tainted_modules();
3376 if (last_unloaded_module.name[0])
3377 pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3378 last_unloaded_module.taints);
3382 #ifdef CONFIG_MODULE_DEBUGFS
3383 struct dentry *mod_debugfs_root;
3385 static int module_debugfs_init(void)
3387 mod_debugfs_root = debugfs_create_dir("modules", NULL);
3390 module_init(module_debugfs_init);