Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Copyright (C) 2002 Richard Henderson
   4 * Copyright (C) 2001 Rusty Russell, 2002, 2010 Rusty Russell IBM.
   5 * Copyright (C) 2023 Luis Chamberlain <mcgrof@kernel.org>
   6 */
   7
   8#define INCLUDE_VERMAGIC
   9
  10#include <linux/export.h>
  11#include <linux/extable.h>
  12#include <linux/moduleloader.h>
  13#include <linux/module_signature.h>
  14#include <linux/trace_events.h>
  15#include <linux/init.h>
  16#include <linux/kallsyms.h>
  17#include <linux/buildid.h>
  18#include <linux/fs.h>
  19#include <linux/kernel.h>
  20#include <linux/kernel_read_file.h>
  21#include <linux/kstrtox.h>
  22#include <linux/slab.h>
  23#include <linux/vmalloc.h>
  24#include <linux/elf.h>
  25#include <linux/seq_file.h>
  26#include <linux/syscalls.h>
  27#include <linux/fcntl.h>
  28#include <linux/rcupdate.h>
  29#include <linux/capability.h>
  30#include <linux/cpu.h>
  31#include <linux/moduleparam.h>
  32#include <linux/errno.h>
  33#include <linux/err.h>
  34#include <linux/vermagic.h>
  35#include <linux/notifier.h>
  36#include <linux/sched.h>
  37#include <linux/device.h>
  38#include <linux/string.h>
  39#include <linux/mutex.h>
  40#include <linux/rculist.h>
  41#include <linux/uaccess.h>
  42#include <asm/cacheflush.h>
  43#include <linux/set_memory.h>
  44#include <asm/mmu_context.h>
  45#include <linux/license.h>
  46#include <asm/sections.h>
  47#include <linux/tracepoint.h>
  48#include <linux/ftrace.h>
  49#include <linux/livepatch.h>
  50#include <linux/async.h>
  51#include <linux/percpu.h>
  52#include <linux/kmemleak.h>
  53#include <linux/jump_label.h>
  54#include <linux/pfn.h>
  55#include <linux/bsearch.h>
  56#include <linux/dynamic_debug.h>
  57#include <linux/audit.h>
  58#include <linux/cfi.h>
  59#include <linux/codetag.h>
  60#include <linux/debugfs.h>
  61#include <linux/execmem.h>
  62#include <uapi/linux/module.h>
  63#include "internal.h"
  64
  65#define CREATE_TRACE_POINTS
  66#include <trace/events/module.h>
  67
  68/*
  69 * Mutex protects:
  70 * 1) List of modules (also safely readable with preempt_disable),
  71 * 2) module_use links,
  72 * 3) mod_tree.addr_min/mod_tree.addr_max.
  73 * (delete and add uses RCU list operations).
  74 */
  75DEFINE_MUTEX(module_mutex);
  76LIST_HEAD(modules);
  77
  78/* Work queue for freeing init sections in success case */
  79static void do_free_init(struct work_struct *w);
  80static DECLARE_WORK(init_free_wq, do_free_init);
  81static LLIST_HEAD(init_free_list);
  82
  83struct mod_tree_root mod_tree __cacheline_aligned = {
  84	.addr_min = -1UL,
  85};
  86
  87struct symsearch {
  88	const struct kernel_symbol *start, *stop;
  89	const s32 *crcs;
  90	enum mod_license license;
  91};
  92
  93/*
  94 * Bounds of module memory, for speeding up __module_address.
  95 * Protected by module_mutex.
  96 */
  97static void __mod_update_bounds(enum mod_mem_type type __maybe_unused, void *base,
  98				unsigned int size, struct mod_tree_root *tree)
  99{
 100	unsigned long min = (unsigned long)base;
 101	unsigned long max = min + size;
 102
 103#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
 104	if (mod_mem_type_is_core_data(type)) {
 105		if (min < tree->data_addr_min)
 106			tree->data_addr_min = min;
 107		if (max > tree->data_addr_max)
 108			tree->data_addr_max = max;
 109		return;
 110	}
 111#endif
 112	if (min < tree->addr_min)
 113		tree->addr_min = min;
 114	if (max > tree->addr_max)
 115		tree->addr_max = max;
 116}
 117
 118static void mod_update_bounds(struct module *mod)
 119{
 120	for_each_mod_mem_type(type) {
 121		struct module_memory *mod_mem = &mod->mem[type];
 122
 123		if (mod_mem->size)
 124			__mod_update_bounds(type, mod_mem->base, mod_mem->size, &mod_tree);
 125	}
 126}
 127
 128/* Block module loading/unloading? */
 129int modules_disabled;
 130core_param(nomodule, modules_disabled, bint, 0);
 131
 132/* Waiting for a module to finish initializing? */
 133static DECLARE_WAIT_QUEUE_HEAD(module_wq);
 134
 135static BLOCKING_NOTIFIER_HEAD(module_notify_list);
 136
 137int register_module_notifier(struct notifier_block *nb)
 138{
 139	return blocking_notifier_chain_register(&module_notify_list, nb);
 140}
 141EXPORT_SYMBOL(register_module_notifier);
 142
 143int unregister_module_notifier(struct notifier_block *nb)
 144{
 145	return blocking_notifier_chain_unregister(&module_notify_list, nb);
 146}
 147EXPORT_SYMBOL(unregister_module_notifier);
 148
 149/*
 150 * We require a truly strong try_module_get(): 0 means success.
 151 * Otherwise an error is returned due to ongoing or failed
 152 * initialization etc.
 153 */
 154static inline int strong_try_module_get(struct module *mod)
 155{
 156	BUG_ON(mod && mod->state == MODULE_STATE_UNFORMED);
 157	if (mod && mod->state == MODULE_STATE_COMING)
 158		return -EBUSY;
 159	if (try_module_get(mod))
 160		return 0;
 161	else
 162		return -ENOENT;
 163}
 164
 165static inline void add_taint_module(struct module *mod, unsigned flag,
 166				    enum lockdep_ok lockdep_ok)
 167{
 168	add_taint(flag, lockdep_ok);
 169	set_bit(flag, &mod->taints);
 170}
 171
 172/*
 173 * A thread that wants to hold a reference to a module only while it
 174 * is running can call this to safely exit.
 175 */
 176void __noreturn __module_put_and_kthread_exit(struct module *mod, long code)
 177{
 178	module_put(mod);
 179	kthread_exit(code);
 180}
 181EXPORT_SYMBOL(__module_put_and_kthread_exit);
 182
 183/* Find a module section: 0 means not found. */
 184static unsigned int find_sec(const struct load_info *info, const char *name)
 185{
 186	unsigned int i;
 187
 188	for (i = 1; i < info->hdr->e_shnum; i++) {
 189		Elf_Shdr *shdr = &info->sechdrs[i];
 190		/* Alloc bit cleared means "ignore it." */
 191		if ((shdr->sh_flags & SHF_ALLOC)
 192		    && strcmp(info->secstrings + shdr->sh_name, name) == 0)
 193			return i;
 194	}
 195	return 0;
 196}
 197
 198/**
 199 * find_any_unique_sec() - Find a unique section index by name
 200 * @info: Load info for the module to scan
 201 * @name: Name of the section we're looking for
 202 *
 203 * Locates a unique section by name. Ignores SHF_ALLOC.
 204 *
 205 * Return: Section index if found uniquely, zero if absent, negative count
 206 *         of total instances if multiple were found.
 207 */
 208static int find_any_unique_sec(const struct load_info *info, const char *name)
 209{
 210	unsigned int idx;
 211	unsigned int count = 0;
 212	int i;
 213
 214	for (i = 1; i < info->hdr->e_shnum; i++) {
 215		if (strcmp(info->secstrings + info->sechdrs[i].sh_name,
 216			   name) == 0) {
 217			count++;
 218			idx = i;
 219		}
 220	}
 221	if (count == 1) {
 222		return idx;
 223	} else if (count == 0) {
 224		return 0;
 225	} else {
 226		return -count;
 227	}
 228}
 229
 230/* Find a module section, or NULL. */
 231static void *section_addr(const struct load_info *info, const char *name)
 232{
 233	/* Section 0 has sh_addr 0. */
 234	return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
 235}
 236
 237/* Find a module section, or NULL.  Fill in number of "objects" in section. */
 238static void *section_objs(const struct load_info *info,
 239			  const char *name,
 240			  size_t object_size,
 241			  unsigned int *num)
 242{
 243	unsigned int sec = find_sec(info, name);
 244
 245	/* Section 0 has sh_addr 0 and sh_size 0. */
 246	*num = info->sechdrs[sec].sh_size / object_size;
 247	return (void *)info->sechdrs[sec].sh_addr;
 248}
 249
 250/* Find a module section: 0 means not found. Ignores SHF_ALLOC flag. */
 251static unsigned int find_any_sec(const struct load_info *info, const char *name)
 252{
 253	unsigned int i;
 254
 255	for (i = 1; i < info->hdr->e_shnum; i++) {
 256		Elf_Shdr *shdr = &info->sechdrs[i];
 257		if (strcmp(info->secstrings + shdr->sh_name, name) == 0)
 258			return i;
 259	}
 260	return 0;
 261}
 262
 263/*
 264 * Find a module section, or NULL. Fill in number of "objects" in section.
 265 * Ignores SHF_ALLOC flag.
 266 */
 267static __maybe_unused void *any_section_objs(const struct load_info *info,
 268					     const char *name,
 269					     size_t object_size,
 270					     unsigned int *num)
 271{
 272	unsigned int sec = find_any_sec(info, name);
 273
 274	/* Section 0 has sh_addr 0 and sh_size 0. */
 275	*num = info->sechdrs[sec].sh_size / object_size;
 276	return (void *)info->sechdrs[sec].sh_addr;
 277}
 278
 279#ifndef CONFIG_MODVERSIONS
 280#define symversion(base, idx) NULL
 281#else
 282#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
 283#endif
 284
 285static const char *kernel_symbol_name(const struct kernel_symbol *sym)
 286{
 287#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
 288	return offset_to_ptr(&sym->name_offset);
 289#else
 290	return sym->name;
 291#endif
 292}
 293
 294static const char *kernel_symbol_namespace(const struct kernel_symbol *sym)
 295{
 296#ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
 297	if (!sym->namespace_offset)
 298		return NULL;
 299	return offset_to_ptr(&sym->namespace_offset);
 300#else
 301	return sym->namespace;
 302#endif
 303}
 304
 305int cmp_name(const void *name, const void *sym)
 306{
 307	return strcmp(name, kernel_symbol_name(sym));
 308}
 309
 310static bool find_exported_symbol_in_section(const struct symsearch *syms,
 311					    struct module *owner,
 312					    struct find_symbol_arg *fsa)
 313{
 314	struct kernel_symbol *sym;
 315
 316	if (!fsa->gplok && syms->license == GPL_ONLY)
 317		return false;
 318
 319	sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
 320			sizeof(struct kernel_symbol), cmp_name);
 321	if (!sym)
 322		return false;
 323
 324	fsa->owner = owner;
 325	fsa->crc = symversion(syms->crcs, sym - syms->start);
 326	fsa->sym = sym;
 327	fsa->license = syms->license;
 328
 329	return true;
 330}
 331
 332/*
 333 * Find an exported symbol and return it, along with, (optional) crc and
 334 * (optional) module which owns it.  Needs preempt disabled or module_mutex.
 335 */
 336bool find_symbol(struct find_symbol_arg *fsa)
 337{
 338	static const struct symsearch arr[] = {
 339		{ __start___ksymtab, __stop___ksymtab, __start___kcrctab,
 340		  NOT_GPL_ONLY },
 341		{ __start___ksymtab_gpl, __stop___ksymtab_gpl,
 342		  __start___kcrctab_gpl,
 343		  GPL_ONLY },
 344	};
 345	struct module *mod;
 346	unsigned int i;
 347
 348	module_assert_mutex_or_preempt();
 349
 350	for (i = 0; i < ARRAY_SIZE(arr); i++)
 351		if (find_exported_symbol_in_section(&arr[i], NULL, fsa))
 352			return true;
 353
 354	list_for_each_entry_rcu(mod, &modules, list,
 355				lockdep_is_held(&module_mutex)) {
 356		struct symsearch arr[] = {
 357			{ mod->syms, mod->syms + mod->num_syms, mod->crcs,
 358			  NOT_GPL_ONLY },
 359			{ mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
 360			  mod->gpl_crcs,
 361			  GPL_ONLY },
 362		};
 363
 364		if (mod->state == MODULE_STATE_UNFORMED)
 365			continue;
 366
 367		for (i = 0; i < ARRAY_SIZE(arr); i++)
 368			if (find_exported_symbol_in_section(&arr[i], mod, fsa))
 369				return true;
 370	}
 371
 372	pr_debug("Failed to find symbol %s\n", fsa->name);
 373	return false;
 374}
 375
 376/*
 377 * Search for module by name: must hold module_mutex (or preempt disabled
 378 * for read-only access).
 379 */
 380struct module *find_module_all(const char *name, size_t len,
 381			       bool even_unformed)
 382{
 383	struct module *mod;
 384
 385	module_assert_mutex_or_preempt();
 386
 387	list_for_each_entry_rcu(mod, &modules, list,
 388				lockdep_is_held(&module_mutex)) {
 389		if (!even_unformed && mod->state == MODULE_STATE_UNFORMED)
 390			continue;
 391		if (strlen(mod->name) == len && !memcmp(mod->name, name, len))
 392			return mod;
 393	}
 394	return NULL;
 395}
 396
 397struct module *find_module(const char *name)
 398{
 399	return find_module_all(name, strlen(name), false);
 400}
 401
 402#ifdef CONFIG_SMP
 403
 404static inline void __percpu *mod_percpu(struct module *mod)
 405{
 406	return mod->percpu;
 407}
 408
 409static int percpu_modalloc(struct module *mod, struct load_info *info)
 410{
 411	Elf_Shdr *pcpusec = &info->sechdrs[info->index.pcpu];
 412	unsigned long align = pcpusec->sh_addralign;
 413
 414	if (!pcpusec->sh_size)
 415		return 0;
 416
 417	if (align > PAGE_SIZE) {
 418		pr_warn("%s: per-cpu alignment %li > %li\n",
 419			mod->name, align, PAGE_SIZE);
 420		align = PAGE_SIZE;
 421	}
 422
 423	mod->percpu = __alloc_reserved_percpu(pcpusec->sh_size, align);
 424	if (!mod->percpu) {
 425		pr_warn("%s: Could not allocate %lu bytes percpu data\n",
 426			mod->name, (unsigned long)pcpusec->sh_size);
 427		return -ENOMEM;
 428	}
 429	mod->percpu_size = pcpusec->sh_size;
 430	return 0;
 431}
 432
 433static void percpu_modfree(struct module *mod)
 434{
 435	free_percpu(mod->percpu);
 436}
 437
 438static unsigned int find_pcpusec(struct load_info *info)
 439{
 440	return find_sec(info, ".data..percpu");
 441}
 442
 443static void percpu_modcopy(struct module *mod,
 444			   const void *from, unsigned long size)
 445{
 446	int cpu;
 447
 448	for_each_possible_cpu(cpu)
 449		memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
 450}
 451
 452bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
 453{
 454	struct module *mod;
 455	unsigned int cpu;
 456
 457	preempt_disable();
 458
 459	list_for_each_entry_rcu(mod, &modules, list) {
 460		if (mod->state == MODULE_STATE_UNFORMED)
 461			continue;
 462		if (!mod->percpu_size)
 463			continue;
 464		for_each_possible_cpu(cpu) {
 465			void *start = per_cpu_ptr(mod->percpu, cpu);
 466			void *va = (void *)addr;
 467
 468			if (va >= start && va < start + mod->percpu_size) {
 469				if (can_addr) {
 470					*can_addr = (unsigned long) (va - start);
 471					*can_addr += (unsigned long)
 472						per_cpu_ptr(mod->percpu,
 473							    get_boot_cpu_id());
 474				}
 475				preempt_enable();
 476				return true;
 477			}
 478		}
 479	}
 480
 481	preempt_enable();
 482	return false;
 483}
 484
 485/**
 486 * is_module_percpu_address() - test whether address is from module static percpu
 487 * @addr: address to test
 488 *
 489 * Test whether @addr belongs to module static percpu area.
 490 *
 491 * Return: %true if @addr is from module static percpu area
 492 */
 493bool is_module_percpu_address(unsigned long addr)
 494{
 495	return __is_module_percpu_address(addr, NULL);
 496}
 497
 498#else /* ... !CONFIG_SMP */
 499
 500static inline void __percpu *mod_percpu(struct module *mod)
 501{
 502	return NULL;
 503}
 504static int percpu_modalloc(struct module *mod, struct load_info *info)
 505{
 506	/* UP modules shouldn't have this section: ENOMEM isn't quite right */
 507	if (info->sechdrs[info->index.pcpu].sh_size != 0)
 508		return -ENOMEM;
 509	return 0;
 510}
 511static inline void percpu_modfree(struct module *mod)
 512{
 513}
 514static unsigned int find_pcpusec(struct load_info *info)
 515{
 516	return 0;
 517}
 518static inline void percpu_modcopy(struct module *mod,
 519				  const void *from, unsigned long size)
 520{
 521	/* pcpusec should be 0, and size of that section should be 0. */
 522	BUG_ON(size != 0);
 523}
 524bool is_module_percpu_address(unsigned long addr)
 525{
 526	return false;
 527}
 528
 529bool __is_module_percpu_address(unsigned long addr, unsigned long *can_addr)
 530{
 531	return false;
 532}
 533
 534#endif /* CONFIG_SMP */
 535
 536#define MODINFO_ATTR(field)	\
 537static void setup_modinfo_##field(struct module *mod, const char *s)  \
 538{                                                                     \
 539	mod->field = kstrdup(s, GFP_KERNEL);                          \
 540}                                                                     \
 541static ssize_t show_modinfo_##field(struct module_attribute *mattr,   \
 542			struct module_kobject *mk, char *buffer)      \
 543{                                                                     \
 544	return scnprintf(buffer, PAGE_SIZE, "%s\n", mk->mod->field);  \
 545}                                                                     \
 546static int modinfo_##field##_exists(struct module *mod)               \
 547{                                                                     \
 548	return mod->field != NULL;                                    \
 549}                                                                     \
 550static void free_modinfo_##field(struct module *mod)                  \
 551{                                                                     \
 552	kfree(mod->field);                                            \
 553	mod->field = NULL;                                            \
 554}                                                                     \
 555static struct module_attribute modinfo_##field = {                    \
 556	.attr = { .name = __stringify(field), .mode = 0444 },         \
 557	.show = show_modinfo_##field,                                 \
 558	.setup = setup_modinfo_##field,                               \
 559	.test = modinfo_##field##_exists,                             \
 560	.free = free_modinfo_##field,                                 \
 561};
 562
 563MODINFO_ATTR(version);
 564MODINFO_ATTR(srcversion);
 565
 566static struct {
 567	char name[MODULE_NAME_LEN + 1];
 568	char taints[MODULE_FLAGS_BUF_SIZE];
 569} last_unloaded_module;
 570
 571#ifdef CONFIG_MODULE_UNLOAD
 572
 573EXPORT_TRACEPOINT_SYMBOL(module_get);
 574
 575/* MODULE_REF_BASE is the base reference count by kmodule loader. */
 576#define MODULE_REF_BASE	1
 577
 578/* Init the unload section of the module. */
 579static int module_unload_init(struct module *mod)
 580{
 581	/*
 582	 * Initialize reference counter to MODULE_REF_BASE.
 583	 * refcnt == 0 means module is going.
 584	 */
 585	atomic_set(&mod->refcnt, MODULE_REF_BASE);
 586
 587	INIT_LIST_HEAD(&mod->source_list);
 588	INIT_LIST_HEAD(&mod->target_list);
 589
 590	/* Hold reference count during initialization. */
 591	atomic_inc(&mod->refcnt);
 592
 593	return 0;
 594}
 595
 596/* Does a already use b? */
 597static int already_uses(struct module *a, struct module *b)
 598{
 599	struct module_use *use;
 600
 601	list_for_each_entry(use, &b->source_list, source_list) {
 602		if (use->source == a)
 603			return 1;
 604	}
 605	pr_debug("%s does not use %s!\n", a->name, b->name);
 606	return 0;
 607}
 608
 609/*
 610 * Module a uses b
 611 *  - we add 'a' as a "source", 'b' as a "target" of module use
 612 *  - the module_use is added to the list of 'b' sources (so
 613 *    'b' can walk the list to see who sourced them), and of 'a'
 614 *    targets (so 'a' can see what modules it targets).
 615 */
 616static int add_module_usage(struct module *a, struct module *b)
 617{
 618	struct module_use *use;
 619
 620	pr_debug("Allocating new usage for %s.\n", a->name);
 621	use = kmalloc(sizeof(*use), GFP_ATOMIC);
 622	if (!use)
 623		return -ENOMEM;
 624
 625	use->source = a;
 626	use->target = b;
 627	list_add(&use->source_list, &b->source_list);
 628	list_add(&use->target_list, &a->target_list);
 629	return 0;
 630}
 631
 632/* Module a uses b: caller needs module_mutex() */
 633static int ref_module(struct module *a, struct module *b)
 634{
 635	int err;
 636
 637	if (b == NULL || already_uses(a, b))
 638		return 0;
 639
 640	/* If module isn't available, we fail. */
 641	err = strong_try_module_get(b);
 642	if (err)
 643		return err;
 644
 645	err = add_module_usage(a, b);
 646	if (err) {
 647		module_put(b);
 648		return err;
 649	}
 650	return 0;
 651}
 652
 653/* Clear the unload stuff of the module. */
 654static void module_unload_free(struct module *mod)
 655{
 656	struct module_use *use, *tmp;
 657
 658	mutex_lock(&module_mutex);
 659	list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
 660		struct module *i = use->target;
 661		pr_debug("%s unusing %s\n", mod->name, i->name);
 662		module_put(i);
 663		list_del(&use->source_list);
 664		list_del(&use->target_list);
 665		kfree(use);
 666	}
 667	mutex_unlock(&module_mutex);
 668}
 669
 670#ifdef CONFIG_MODULE_FORCE_UNLOAD
 671static inline int try_force_unload(unsigned int flags)
 672{
 673	int ret = (flags & O_TRUNC);
 674	if (ret)
 675		add_taint(TAINT_FORCED_RMMOD, LOCKDEP_NOW_UNRELIABLE);
 676	return ret;
 677}
 678#else
 679static inline int try_force_unload(unsigned int flags)
 680{
 681	return 0;
 682}
 683#endif /* CONFIG_MODULE_FORCE_UNLOAD */
 684
 685/* Try to release refcount of module, 0 means success. */
 686static int try_release_module_ref(struct module *mod)
 687{
 688	int ret;
 689
 690	/* Try to decrement refcnt which we set at loading */
 691	ret = atomic_sub_return(MODULE_REF_BASE, &mod->refcnt);
 692	BUG_ON(ret < 0);
 693	if (ret)
 694		/* Someone can put this right now, recover with checking */
 695		ret = atomic_add_unless(&mod->refcnt, MODULE_REF_BASE, 0);
 696
 697	return ret;
 698}
 699
 700static int try_stop_module(struct module *mod, int flags, int *forced)
 701{
 702	/* If it's not unused, quit unless we're forcing. */
 703	if (try_release_module_ref(mod) != 0) {
 704		*forced = try_force_unload(flags);
 705		if (!(*forced))
 706			return -EWOULDBLOCK;
 707	}
 708
 709	/* Mark it as dying. */
 710	mod->state = MODULE_STATE_GOING;
 711
 712	return 0;
 713}
 714
 715/**
 716 * module_refcount() - return the refcount or -1 if unloading
 717 * @mod:	the module we're checking
 718 *
 719 * Return:
 720 *	-1 if the module is in the process of unloading
 721 *	otherwise the number of references in the kernel to the module
 722 */
 723int module_refcount(struct module *mod)
 724{
 725	return atomic_read(&mod->refcnt) - MODULE_REF_BASE;
 726}
 727EXPORT_SYMBOL(module_refcount);
 728
 729/* This exists whether we can unload or not */
 730static void free_module(struct module *mod);
 731
 732SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
 733		unsigned int, flags)
 734{
 735	struct module *mod;
 736	char name[MODULE_NAME_LEN];
 737	char buf[MODULE_FLAGS_BUF_SIZE];
 738	int ret, forced = 0;
 739
 740	if (!capable(CAP_SYS_MODULE) || modules_disabled)
 741		return -EPERM;
 742
 743	if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0)
 744		return -EFAULT;
 745	name[MODULE_NAME_LEN-1] = '\0';
 746
 747	audit_log_kern_module(name);
 748
 749	if (mutex_lock_interruptible(&module_mutex) != 0)
 750		return -EINTR;
 751
 752	mod = find_module(name);
 753	if (!mod) {
 754		ret = -ENOENT;
 755		goto out;
 756	}
 757
 758	if (!list_empty(&mod->source_list)) {
 759		/* Other modules depend on us: get rid of them first. */
 760		ret = -EWOULDBLOCK;
 761		goto out;
 762	}
 763
 764	/* Doing init or already dying? */
 765	if (mod->state != MODULE_STATE_LIVE) {
 766		/* FIXME: if (force), slam module count damn the torpedoes */
 767		pr_debug("%s already dying\n", mod->name);
 768		ret = -EBUSY;
 769		goto out;
 770	}
 771
 772	/* If it has an init func, it must have an exit func to unload */
 773	if (mod->init && !mod->exit) {
 774		forced = try_force_unload(flags);
 775		if (!forced) {
 776			/* This module can't be removed */
 777			ret = -EBUSY;
 778			goto out;
 779		}
 780	}
 781
 782	ret = try_stop_module(mod, flags, &forced);
 783	if (ret != 0)
 784		goto out;
 785
 786	mutex_unlock(&module_mutex);
 787	/* Final destruction now no one is using it. */
 788	if (mod->exit != NULL)
 789		mod->exit();
 790	blocking_notifier_call_chain(&module_notify_list,
 791				     MODULE_STATE_GOING, mod);
 792	klp_module_going(mod);
 793	ftrace_release_mod(mod);
 794
 795	async_synchronize_full();
 796
 797	/* Store the name and taints of the last unloaded module for diagnostic purposes */
 798	strscpy(last_unloaded_module.name, mod->name, sizeof(last_unloaded_module.name));
 799	strscpy(last_unloaded_module.taints, module_flags(mod, buf, false), sizeof(last_unloaded_module.taints));
 800
 801	free_module(mod);
 802	/* someone could wait for the module in add_unformed_module() */
 803	wake_up_all(&module_wq);
 804	return 0;
 805out:
 806	mutex_unlock(&module_mutex);
 807	return ret;
 808}
 809
 810void __symbol_put(const char *symbol)
 811{
 812	struct find_symbol_arg fsa = {
 813		.name	= symbol,
 814		.gplok	= true,
 815	};
 816
 817	preempt_disable();
 818	BUG_ON(!find_symbol(&fsa));
 819	module_put(fsa.owner);
 820	preempt_enable();
 821}
 822EXPORT_SYMBOL(__symbol_put);
 823
 824/* Note this assumes addr is a function, which it currently always is. */
 825void symbol_put_addr(void *addr)
 826{
 827	struct module *modaddr;
 828	unsigned long a = (unsigned long)dereference_function_descriptor(addr);
 829
 830	if (core_kernel_text(a))
 831		return;
 832
 833	/*
 834	 * Even though we hold a reference on the module; we still need to
 835	 * disable preemption in order to safely traverse the data structure.
 836	 */
 837	preempt_disable();
 838	modaddr = __module_text_address(a);
 839	BUG_ON(!modaddr);
 840	module_put(modaddr);
 841	preempt_enable();
 842}
 843EXPORT_SYMBOL_GPL(symbol_put_addr);
 844
 845static ssize_t show_refcnt(struct module_attribute *mattr,
 846			   struct module_kobject *mk, char *buffer)
 847{
 848	return sprintf(buffer, "%i\n", module_refcount(mk->mod));
 849}
 850
 851static struct module_attribute modinfo_refcnt =
 852	__ATTR(refcnt, 0444, show_refcnt, NULL);
 853
 854void __module_get(struct module *module)
 855{
 856	if (module) {
 857		atomic_inc(&module->refcnt);
 858		trace_module_get(module, _RET_IP_);
 859	}
 860}
 861EXPORT_SYMBOL(__module_get);
 862
 863bool try_module_get(struct module *module)
 864{
 865	bool ret = true;
 866
 867	if (module) {
 868		/* Note: here, we can fail to get a reference */
 869		if (likely(module_is_live(module) &&
 870			   atomic_inc_not_zero(&module->refcnt) != 0))
 871			trace_module_get(module, _RET_IP_);
 872		else
 873			ret = false;
 874	}
 875	return ret;
 876}
 877EXPORT_SYMBOL(try_module_get);
 878
 879void module_put(struct module *module)
 880{
 881	int ret;
 882
 883	if (module) {
 884		ret = atomic_dec_if_positive(&module->refcnt);
 885		WARN_ON(ret < 0);	/* Failed to put refcount */
 886		trace_module_put(module, _RET_IP_);
 887	}
 888}
 889EXPORT_SYMBOL(module_put);
 890
 891#else /* !CONFIG_MODULE_UNLOAD */
 892static inline void module_unload_free(struct module *mod)
 893{
 894}
 895
 896static int ref_module(struct module *a, struct module *b)
 897{
 898	return strong_try_module_get(b);
 899}
 900
 901static inline int module_unload_init(struct module *mod)
 902{
 903	return 0;
 904}
 905#endif /* CONFIG_MODULE_UNLOAD */
 906
 907size_t module_flags_taint(unsigned long taints, char *buf)
 908{
 909	size_t l = 0;
 910	int i;
 911
 912	for (i = 0; i < TAINT_FLAGS_COUNT; i++) {
 913		if (taint_flags[i].module && test_bit(i, &taints))
 914			buf[l++] = taint_flags[i].c_true;
 915	}
 916
 917	return l;
 918}
 919
 920static ssize_t show_initstate(struct module_attribute *mattr,
 921			      struct module_kobject *mk, char *buffer)
 922{
 923	const char *state = "unknown";
 924
 925	switch (mk->mod->state) {
 926	case MODULE_STATE_LIVE:
 927		state = "live";
 928		break;
 929	case MODULE_STATE_COMING:
 930		state = "coming";
 931		break;
 932	case MODULE_STATE_GOING:
 933		state = "going";
 934		break;
 935	default:
 936		BUG();
 937	}
 938	return sprintf(buffer, "%s\n", state);
 939}
 940
 941static struct module_attribute modinfo_initstate =
 942	__ATTR(initstate, 0444, show_initstate, NULL);
 943
 944static ssize_t store_uevent(struct module_attribute *mattr,
 945			    struct module_kobject *mk,
 946			    const char *buffer, size_t count)
 947{
 948	int rc;
 949
 950	rc = kobject_synth_uevent(&mk->kobj, buffer, count);
 951	return rc ? rc : count;
 952}
 953
 954struct module_attribute module_uevent =
 955	__ATTR(uevent, 0200, NULL, store_uevent);
 956
 957static ssize_t show_coresize(struct module_attribute *mattr,
 958			     struct module_kobject *mk, char *buffer)
 959{
 960	unsigned int size = mk->mod->mem[MOD_TEXT].size;
 961
 962	if (!IS_ENABLED(CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC)) {
 963		for_class_mod_mem_type(type, core_data)
 964			size += mk->mod->mem[type].size;
 965	}
 966	return sprintf(buffer, "%u\n", size);
 967}
 968
 969static struct module_attribute modinfo_coresize =
 970	__ATTR(coresize, 0444, show_coresize, NULL);
 971
 972#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
 973static ssize_t show_datasize(struct module_attribute *mattr,
 974			     struct module_kobject *mk, char *buffer)
 975{
 976	unsigned int size = 0;
 977
 978	for_class_mod_mem_type(type, core_data)
 979		size += mk->mod->mem[type].size;
 980	return sprintf(buffer, "%u\n", size);
 981}
 982
 983static struct module_attribute modinfo_datasize =
 984	__ATTR(datasize, 0444, show_datasize, NULL);
 985#endif
 986
 987static ssize_t show_initsize(struct module_attribute *mattr,
 988			     struct module_kobject *mk, char *buffer)
 989{
 990	unsigned int size = 0;
 991
 992	for_class_mod_mem_type(type, init)
 993		size += mk->mod->mem[type].size;
 994	return sprintf(buffer, "%u\n", size);
 995}
 996
 997static struct module_attribute modinfo_initsize =
 998	__ATTR(initsize, 0444, show_initsize, NULL);
 999
1000static ssize_t show_taint(struct module_attribute *mattr,
1001			  struct module_kobject *mk, char *buffer)
1002{
1003	size_t l;
1004
1005	l = module_flags_taint(mk->mod->taints, buffer);
1006	buffer[l++] = '\n';
1007	return l;
1008}
1009
1010static struct module_attribute modinfo_taint =
1011	__ATTR(taint, 0444, show_taint, NULL);
1012
1013struct module_attribute *modinfo_attrs[] = {
1014	&module_uevent,
1015	&modinfo_version,
1016	&modinfo_srcversion,
1017	&modinfo_initstate,
1018	&modinfo_coresize,
1019#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
1020	&modinfo_datasize,
1021#endif
1022	&modinfo_initsize,
1023	&modinfo_taint,
1024#ifdef CONFIG_MODULE_UNLOAD
1025	&modinfo_refcnt,
1026#endif
1027	NULL,
1028};
1029
1030size_t modinfo_attrs_count = ARRAY_SIZE(modinfo_attrs);
1031
1032static const char vermagic[] = VERMAGIC_STRING;
1033
1034int try_to_force_load(struct module *mod, const char *reason)
1035{
1036#ifdef CONFIG_MODULE_FORCE_LOAD
1037	if (!test_taint(TAINT_FORCED_MODULE))
1038		pr_warn("%s: %s: kernel tainted.\n", mod->name, reason);
1039	add_taint_module(mod, TAINT_FORCED_MODULE, LOCKDEP_NOW_UNRELIABLE);
1040	return 0;
1041#else
1042	return -ENOEXEC;
1043#endif
1044}
1045
1046/* Parse tag=value strings from .modinfo section */
1047char *module_next_tag_pair(char *string, unsigned long *secsize)
1048{
1049	/* Skip non-zero chars */
1050	while (string[0]) {
1051		string++;
1052		if ((*secsize)-- <= 1)
1053			return NULL;
1054	}
1055
1056	/* Skip any zero padding. */
1057	while (!string[0]) {
1058		string++;
1059		if ((*secsize)-- <= 1)
1060			return NULL;
1061	}
1062	return string;
1063}
1064
1065static char *get_next_modinfo(const struct load_info *info, const char *tag,
1066			      char *prev)
1067{
1068	char *p;
1069	unsigned int taglen = strlen(tag);
1070	Elf_Shdr *infosec = &info->sechdrs[info->index.info];
1071	unsigned long size = infosec->sh_size;
1072
1073	/*
1074	 * get_modinfo() calls made before rewrite_section_headers()
1075	 * must use sh_offset, as sh_addr isn't set!
1076	 */
1077	char *modinfo = (char *)info->hdr + infosec->sh_offset;
1078
1079	if (prev) {
1080		size -= prev - modinfo;
1081		modinfo = module_next_tag_pair(prev, &size);
1082	}
1083
1084	for (p = modinfo; p; p = module_next_tag_pair(p, &size)) {
1085		if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
1086			return p + taglen + 1;
1087	}
1088	return NULL;
1089}
1090
1091static char *get_modinfo(const struct load_info *info, const char *tag)
1092{
1093	return get_next_modinfo(info, tag, NULL);
1094}
1095
1096static int verify_namespace_is_imported(const struct load_info *info,
1097					const struct kernel_symbol *sym,
1098					struct module *mod)
1099{
1100	const char *namespace;
1101	char *imported_namespace;
1102
1103	namespace = kernel_symbol_namespace(sym);
1104	if (namespace && namespace[0]) {
1105		for_each_modinfo_entry(imported_namespace, info, "import_ns") {
1106			if (strcmp(namespace, imported_namespace) == 0)
1107				return 0;
1108		}
1109#ifdef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1110		pr_warn(
1111#else
1112		pr_err(
1113#endif
1114			"%s: module uses symbol (%s) from namespace %s, but does not import it.\n",
1115			mod->name, kernel_symbol_name(sym), namespace);
1116#ifndef CONFIG_MODULE_ALLOW_MISSING_NAMESPACE_IMPORTS
1117		return -EINVAL;
1118#endif
1119	}
1120	return 0;
1121}
1122
1123static bool inherit_taint(struct module *mod, struct module *owner, const char *name)
1124{
1125	if (!owner || !test_bit(TAINT_PROPRIETARY_MODULE, &owner->taints))
1126		return true;
1127
1128	if (mod->using_gplonly_symbols) {
1129		pr_err("%s: module using GPL-only symbols uses symbols %s from proprietary module %s.\n",
1130			mod->name, name, owner->name);
1131		return false;
1132	}
1133
1134	if (!test_bit(TAINT_PROPRIETARY_MODULE, &mod->taints)) {
1135		pr_warn("%s: module uses symbols %s from proprietary module %s, inheriting taint.\n",
1136			mod->name, name, owner->name);
1137		set_bit(TAINT_PROPRIETARY_MODULE, &mod->taints);
1138	}
1139	return true;
1140}
1141
1142/* Resolve a symbol for this module.  I.e. if we find one, record usage. */
1143static const struct kernel_symbol *resolve_symbol(struct module *mod,
1144						  const struct load_info *info,
1145						  const char *name,
1146						  char ownername[])
1147{
1148	struct find_symbol_arg fsa = {
1149		.name	= name,
1150		.gplok	= !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)),
1151		.warn	= true,
1152	};
1153	int err;
1154
1155	/*
1156	 * The module_mutex should not be a heavily contended lock;
1157	 * if we get the occasional sleep here, we'll go an extra iteration
1158	 * in the wait_event_interruptible(), which is harmless.
1159	 */
1160	sched_annotate_sleep();
1161	mutex_lock(&module_mutex);
1162	if (!find_symbol(&fsa))
1163		goto unlock;
1164
1165	if (fsa.license == GPL_ONLY)
1166		mod->using_gplonly_symbols = true;
1167
1168	if (!inherit_taint(mod, fsa.owner, name)) {
1169		fsa.sym = NULL;
1170		goto getname;
1171	}
1172
1173	if (!check_version(info, name, mod, fsa.crc)) {
1174		fsa.sym = ERR_PTR(-EINVAL);
1175		goto getname;
1176	}
1177
1178	err = verify_namespace_is_imported(info, fsa.sym, mod);
1179	if (err) {
1180		fsa.sym = ERR_PTR(err);
1181		goto getname;
1182	}
1183
1184	err = ref_module(mod, fsa.owner);
1185	if (err) {
1186		fsa.sym = ERR_PTR(err);
1187		goto getname;
1188	}
1189
1190getname:
1191	/* We must make copy under the lock if we failed to get ref. */
1192	strncpy(ownername, module_name(fsa.owner), MODULE_NAME_LEN);
1193unlock:
1194	mutex_unlock(&module_mutex);
1195	return fsa.sym;
1196}
1197
1198static const struct kernel_symbol *
1199resolve_symbol_wait(struct module *mod,
1200		    const struct load_info *info,
1201		    const char *name)
1202{
1203	const struct kernel_symbol *ksym;
1204	char owner[MODULE_NAME_LEN];
1205
1206	if (wait_event_interruptible_timeout(module_wq,
1207			!IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1208			|| PTR_ERR(ksym) != -EBUSY,
1209					     30 * HZ) <= 0) {
1210		pr_warn("%s: gave up waiting for init of module %s.\n",
1211			mod->name, owner);
1212	}
1213	return ksym;
1214}
1215
1216void __weak module_arch_cleanup(struct module *mod)
1217{
1218}
1219
1220void __weak module_arch_freeing_init(struct module *mod)
1221{
1222}
1223
1224void *__module_writable_address(struct module *mod, void *loc)
1225{
1226	for_class_mod_mem_type(type, text) {
1227		struct module_memory *mem = &mod->mem[type];
1228
1229		if (loc >= mem->base && loc < mem->base + mem->size)
1230			return loc + (mem->rw_copy - mem->base);
1231	}
1232
1233	return loc;
1234}
1235
1236static int module_memory_alloc(struct module *mod, enum mod_mem_type type)
1237{
1238	unsigned int size = PAGE_ALIGN(mod->mem[type].size);
1239	enum execmem_type execmem_type;
1240	void *ptr;
1241
1242	mod->mem[type].size = size;
1243
1244	if (mod_mem_type_is_data(type))
1245		execmem_type = EXECMEM_MODULE_DATA;
1246	else
1247		execmem_type = EXECMEM_MODULE_TEXT;
1248
1249	ptr = execmem_alloc(execmem_type, size);
1250	if (!ptr)
1251		return -ENOMEM;
1252
1253	mod->mem[type].base = ptr;
1254
1255	if (execmem_is_rox(execmem_type)) {
1256		ptr = vzalloc(size);
1257
1258		if (!ptr) {
1259			execmem_free(mod->mem[type].base);
1260			return -ENOMEM;
1261		}
1262
1263		mod->mem[type].rw_copy = ptr;
1264		mod->mem[type].is_rox = true;
1265	} else {
1266		mod->mem[type].rw_copy = mod->mem[type].base;
1267		memset(mod->mem[type].base, 0, size);
1268	}
1269
1270	/*
1271	 * The pointer to these blocks of memory are stored on the module
1272	 * structure and we keep that around so long as the module is
1273	 * around. We only free that memory when we unload the module.
1274	 * Just mark them as not being a leak then. The .init* ELF
1275	 * sections *do* get freed after boot so we *could* treat them
1276	 * slightly differently with kmemleak_ignore() and only grey
1277	 * them out as they work as typical memory allocations which
1278	 * *do* eventually get freed, but let's just keep things simple
1279	 * and avoid *any* false positives.
1280	 */
1281	kmemleak_not_leak(ptr);
1282
1283	return 0;
1284}
1285
1286static void module_memory_free(struct module *mod, enum mod_mem_type type)
1287{
1288	struct module_memory *mem = &mod->mem[type];
1289
1290	if (mem->is_rox)
1291		vfree(mem->rw_copy);
1292
1293	execmem_free(mem->base);
1294}
1295
1296static void free_mod_mem(struct module *mod)
1297{
1298	for_each_mod_mem_type(type) {
1299		struct module_memory *mod_mem = &mod->mem[type];
1300
1301		if (type == MOD_DATA)
1302			continue;
1303
1304		/* Free lock-classes; relies on the preceding sync_rcu(). */
1305		lockdep_free_key_range(mod_mem->base, mod_mem->size);
1306		if (mod_mem->size)
1307			module_memory_free(mod, type);
1308	}
1309
1310	/* MOD_DATA hosts mod, so free it at last */
1311	lockdep_free_key_range(mod->mem[MOD_DATA].base, mod->mem[MOD_DATA].size);
1312	module_memory_free(mod, MOD_DATA);
1313}
1314
1315/* Free a module, remove from lists, etc. */
1316static void free_module(struct module *mod)
1317{
1318	trace_module_free(mod);
1319
1320	codetag_unload_module(mod);
1321
1322	mod_sysfs_teardown(mod);
1323
1324	/*
1325	 * We leave it in list to prevent duplicate loads, but make sure
1326	 * that noone uses it while it's being deconstructed.
1327	 */
1328	mutex_lock(&module_mutex);
1329	mod->state = MODULE_STATE_UNFORMED;
1330	mutex_unlock(&module_mutex);
1331
1332	/* Arch-specific cleanup. */
1333	module_arch_cleanup(mod);
1334
1335	/* Module unload stuff */
1336	module_unload_free(mod);
1337
1338	/* Free any allocated parameters. */
1339	destroy_params(mod->kp, mod->num_kp);
1340
1341	if (is_livepatch_module(mod))
1342		free_module_elf(mod);
1343
1344	/* Now we can delete it from the lists */
1345	mutex_lock(&module_mutex);
1346	/* Unlink carefully: kallsyms could be walking list. */
1347	list_del_rcu(&mod->list);
1348	mod_tree_remove(mod);
1349	/* Remove this module from bug list, this uses list_del_rcu */
1350	module_bug_cleanup(mod);
1351	/* Wait for RCU-sched synchronizing before releasing mod->list and buglist. */
1352	synchronize_rcu();
1353	if (try_add_tainted_module(mod))
1354		pr_err("%s: adding tainted module to the unloaded tainted modules list failed.\n",
1355		       mod->name);
1356	mutex_unlock(&module_mutex);
1357
1358	/* This may be empty, but that's OK */
1359	module_arch_freeing_init(mod);
1360	kfree(mod->args);
1361	percpu_modfree(mod);
1362
1363	free_mod_mem(mod);
1364}
1365
1366void *__symbol_get(const char *symbol)
1367{
1368	struct find_symbol_arg fsa = {
1369		.name	= symbol,
1370		.gplok	= true,
1371		.warn	= true,
1372	};
1373
1374	preempt_disable();
1375	if (!find_symbol(&fsa))
1376		goto fail;
1377	if (fsa.license != GPL_ONLY) {
1378		pr_warn("failing symbol_get of non-GPLONLY symbol %s.\n",
1379			symbol);
1380		goto fail;
1381	}
1382	if (strong_try_module_get(fsa.owner))
1383		goto fail;
1384	preempt_enable();
1385	return (void *)kernel_symbol_value(fsa.sym);
1386fail:
1387	preempt_enable();
1388	return NULL;
1389}
1390EXPORT_SYMBOL_GPL(__symbol_get);
1391
1392/*
1393 * Ensure that an exported symbol [global namespace] does not already exist
1394 * in the kernel or in some other module's exported symbol table.
1395 *
1396 * You must hold the module_mutex.
1397 */
1398static int verify_exported_symbols(struct module *mod)
1399{
1400	unsigned int i;
1401	const struct kernel_symbol *s;
1402	struct {
1403		const struct kernel_symbol *sym;
1404		unsigned int num;
1405	} arr[] = {
1406		{ mod->syms, mod->num_syms },
1407		{ mod->gpl_syms, mod->num_gpl_syms },
1408	};
1409
1410	for (i = 0; i < ARRAY_SIZE(arr); i++) {
1411		for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1412			struct find_symbol_arg fsa = {
1413				.name	= kernel_symbol_name(s),
1414				.gplok	= true,
1415			};
1416			if (find_symbol(&fsa)) {
1417				pr_err("%s: exports duplicate symbol %s"
1418				       " (owned by %s)\n",
1419				       mod->name, kernel_symbol_name(s),
1420				       module_name(fsa.owner));
1421				return -ENOEXEC;
1422			}
1423		}
1424	}
1425	return 0;
1426}
1427
1428static bool ignore_undef_symbol(Elf_Half emachine, const char *name)
1429{
1430	/*
1431	 * On x86, PIC code and Clang non-PIC code may have call foo@PLT. GNU as
1432	 * before 2.37 produces an unreferenced _GLOBAL_OFFSET_TABLE_ on x86-64.
1433	 * i386 has a similar problem but may not deserve a fix.
1434	 *
1435	 * If we ever have to ignore many symbols, consider refactoring the code to
1436	 * only warn if referenced by a relocation.
1437	 */
1438	if (emachine == EM_386 || emachine == EM_X86_64)
1439		return !strcmp(name, "_GLOBAL_OFFSET_TABLE_");
1440	return false;
1441}
1442
1443/* Change all symbols so that st_value encodes the pointer directly. */
1444static int simplify_symbols(struct module *mod, const struct load_info *info)
1445{
1446	Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1447	Elf_Sym *sym = (void *)symsec->sh_addr;
1448	unsigned long secbase;
1449	unsigned int i;
1450	int ret = 0;
1451	const struct kernel_symbol *ksym;
1452
1453	for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1454		const char *name = info->strtab + sym[i].st_name;
1455
1456		switch (sym[i].st_shndx) {
1457		case SHN_COMMON:
1458			/* Ignore common symbols */
1459			if (!strncmp(name, "__gnu_lto", 9))
1460				break;
1461
1462			/*
1463			 * We compiled with -fno-common.  These are not
1464			 * supposed to happen.
1465			 */
1466			pr_debug("Common symbol: %s\n", name);
1467			pr_warn("%s: please compile with -fno-common\n",
1468			       mod->name);
1469			ret = -ENOEXEC;
1470			break;
1471
1472		case SHN_ABS:
1473			/* Don't need to do anything */
1474			pr_debug("Absolute symbol: 0x%08lx %s\n",
1475				 (long)sym[i].st_value, name);
1476			break;
1477
1478		case SHN_LIVEPATCH:
1479			/* Livepatch symbols are resolved by livepatch */
1480			break;
1481
1482		case SHN_UNDEF:
1483			ksym = resolve_symbol_wait(mod, info, name);
1484			/* Ok if resolved.  */
1485			if (ksym && !IS_ERR(ksym)) {
1486				sym[i].st_value = kernel_symbol_value(ksym);
1487				break;
1488			}
1489
1490			/* Ok if weak or ignored.  */
1491			if (!ksym &&
1492			    (ELF_ST_BIND(sym[i].st_info) == STB_WEAK ||
1493			     ignore_undef_symbol(info->hdr->e_machine, name)))
1494				break;
1495
1496			ret = PTR_ERR(ksym) ?: -ENOENT;
1497			pr_warn("%s: Unknown symbol %s (err %d)\n",
1498				mod->name, name, ret);
1499			break;
1500
1501		default:
1502			/* Divert to percpu allocation if a percpu var. */
1503			if (sym[i].st_shndx == info->index.pcpu)
1504				secbase = (unsigned long)mod_percpu(mod);
1505			else
1506				secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1507			sym[i].st_value += secbase;
1508			break;
1509		}
1510	}
1511
1512	return ret;
1513}
1514
1515static int apply_relocations(struct module *mod, const struct load_info *info)
1516{
1517	unsigned int i;
1518	int err = 0;
1519
1520	/* Now do relocations. */
1521	for (i = 1; i < info->hdr->e_shnum; i++) {
1522		unsigned int infosec = info->sechdrs[i].sh_info;
1523
1524		/* Not a valid relocation section? */
1525		if (infosec >= info->hdr->e_shnum)
1526			continue;
1527
1528		/* Don't bother with non-allocated sections */
1529		if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC))
1530			continue;
1531
1532		if (info->sechdrs[i].sh_flags & SHF_RELA_LIVEPATCH)
1533			err = klp_apply_section_relocs(mod, info->sechdrs,
1534						       info->secstrings,
1535						       info->strtab,
1536						       info->index.sym, i,
1537						       NULL);
1538		else if (info->sechdrs[i].sh_type == SHT_REL)
1539			err = apply_relocate(info->sechdrs, info->strtab,
1540					     info->index.sym, i, mod);
1541		else if (info->sechdrs[i].sh_type == SHT_RELA)
1542			err = apply_relocate_add(info->sechdrs, info->strtab,
1543						 info->index.sym, i, mod);
1544		if (err < 0)
1545			break;
1546	}
1547	return err;
1548}
1549
1550/* Additional bytes needed by arch in front of individual sections */
1551unsigned int __weak arch_mod_section_prepend(struct module *mod,
1552					     unsigned int section)
1553{
1554	/* default implementation just returns zero */
1555	return 0;
1556}
1557
1558long module_get_offset_and_type(struct module *mod, enum mod_mem_type type,
1559				Elf_Shdr *sechdr, unsigned int section)
1560{
1561	long offset;
1562	long mask = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK) << SH_ENTSIZE_TYPE_SHIFT;
1563
1564	mod->mem[type].size += arch_mod_section_prepend(mod, section);
1565	offset = ALIGN(mod->mem[type].size, sechdr->sh_addralign ?: 1);
1566	mod->mem[type].size = offset + sechdr->sh_size;
1567
1568	WARN_ON_ONCE(offset & mask);
1569	return offset | mask;
1570}
1571
1572bool module_init_layout_section(const char *sname)
1573{
1574#ifndef CONFIG_MODULE_UNLOAD
1575	if (module_exit_section(sname))
1576		return true;
1577#endif
1578	return module_init_section(sname);
1579}
1580
1581static void __layout_sections(struct module *mod, struct load_info *info, bool is_init)
1582{
1583	unsigned int m, i;
1584
1585	static const unsigned long masks[][2] = {
1586		/*
1587		 * NOTE: all executable code must be the first section
1588		 * in this array; otherwise modify the text_size
1589		 * finder in the two loops below
1590		 */
1591		{ SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1592		{ SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1593		{ SHF_RO_AFTER_INIT | SHF_ALLOC, ARCH_SHF_SMALL },
1594		{ SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1595		{ ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1596	};
1597	static const int core_m_to_mem_type[] = {
1598		MOD_TEXT,
1599		MOD_RODATA,
1600		MOD_RO_AFTER_INIT,
1601		MOD_DATA,
1602		MOD_DATA,
1603	};
1604	static const int init_m_to_mem_type[] = {
1605		MOD_INIT_TEXT,
1606		MOD_INIT_RODATA,
1607		MOD_INVALID,
1608		MOD_INIT_DATA,
1609		MOD_INIT_DATA,
1610	};
1611
1612	for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1613		enum mod_mem_type type = is_init ? init_m_to_mem_type[m] : core_m_to_mem_type[m];
1614
1615		for (i = 0; i < info->hdr->e_shnum; ++i) {
1616			Elf_Shdr *s = &info->sechdrs[i];
1617			const char *sname = info->secstrings + s->sh_name;
1618
1619			if ((s->sh_flags & masks[m][0]) != masks[m][0]
1620			    || (s->sh_flags & masks[m][1])
1621			    || s->sh_entsize != ~0UL
1622			    || is_init != module_init_layout_section(sname))
1623				continue;
1624
1625			if (WARN_ON_ONCE(type == MOD_INVALID))
1626				continue;
1627
1628			/*
1629			 * Do not allocate codetag memory as we load it into
1630			 * preallocated contiguous memory.
1631			 */
1632			if (codetag_needs_module_section(mod, sname, s->sh_size)) {
1633				/*
1634				 * s->sh_entsize won't be used but populate the
1635				 * type field to avoid confusion.
1636				 */
1637				s->sh_entsize = ((unsigned long)(type) & SH_ENTSIZE_TYPE_MASK)
1638						<< SH_ENTSIZE_TYPE_SHIFT;
1639				continue;
1640			}
1641
1642			s->sh_entsize = module_get_offset_and_type(mod, type, s, i);
1643			pr_debug("\t%s\n", sname);
1644		}
1645	}
1646}
1647
1648/*
1649 * Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1650 * might -- code, read-only data, read-write data, small data.  Tally
1651 * sizes, and place the offsets into sh_entsize fields: high bit means it
1652 * belongs in init.
1653 */
1654static void layout_sections(struct module *mod, struct load_info *info)
1655{
1656	unsigned int i;
1657
1658	for (i = 0; i < info->hdr->e_shnum; i++)
1659		info->sechdrs[i].sh_entsize = ~0UL;
1660
1661	pr_debug("Core section allocation order for %s:\n", mod->name);
1662	__layout_sections(mod, info, false);
1663
1664	pr_debug("Init section allocation order for %s:\n", mod->name);
1665	__layout_sections(mod, info, true);
1666}
1667
1668static void module_license_taint_check(struct module *mod, const char *license)
1669{
1670	if (!license)
1671		license = "unspecified";
1672
1673	if (!license_is_gpl_compatible(license)) {
1674		if (!test_taint(TAINT_PROPRIETARY_MODULE))
1675			pr_warn("%s: module license '%s' taints kernel.\n",
1676				mod->name, license);
1677		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
1678				 LOCKDEP_NOW_UNRELIABLE);
1679	}
1680}
1681
1682static void setup_modinfo(struct module *mod, struct load_info *info)
1683{
1684	struct module_attribute *attr;
1685	int i;
1686
1687	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1688		if (attr->setup)
1689			attr->setup(mod, get_modinfo(info, attr->attr.name));
1690	}
1691}
1692
1693static void free_modinfo(struct module *mod)
1694{
1695	struct module_attribute *attr;
1696	int i;
1697
1698	for (i = 0; (attr = modinfo_attrs[i]); i++) {
1699		if (attr->free)
1700			attr->free(mod);
1701	}
1702}
1703
1704bool __weak module_init_section(const char *name)
1705{
1706	return strstarts(name, ".init");
1707}
1708
1709bool __weak module_exit_section(const char *name)
1710{
1711	return strstarts(name, ".exit");
1712}
1713
1714static int validate_section_offset(const struct load_info *info, Elf_Shdr *shdr)
1715{
1716#if defined(CONFIG_64BIT)
1717	unsigned long long secend;
1718#else
1719	unsigned long secend;
1720#endif
1721
1722	/*
1723	 * Check for both overflow and offset/size being
1724	 * too large.
1725	 */
1726	secend = shdr->sh_offset + shdr->sh_size;
1727	if (secend < shdr->sh_offset || secend > info->len)
1728		return -ENOEXEC;
1729
1730	return 0;
1731}
1732
1733/**
1734 * elf_validity_ehdr() - Checks an ELF header for module validity
1735 * @info: Load info containing the ELF header to check
1736 *
1737 * Checks whether an ELF header could belong to a valid module. Checks:
1738 *
1739 * * ELF header is within the data the user provided
1740 * * ELF magic is present
1741 * * It is relocatable (not final linked, not core file, etc.)
1742 * * The header's machine type matches what the architecture expects.
1743 * * Optional arch-specific hook for other properties
1744 *   - module_elf_check_arch() is currently only used by PPC to check
1745 *   ELF ABI version, but may be used by others in the future.
1746 *
1747 * Return: %0 if valid, %-ENOEXEC on failure.
1748 */
1749static int elf_validity_ehdr(const struct load_info *info)
1750{
1751	if (info->len < sizeof(*(info->hdr))) {
1752		pr_err("Invalid ELF header len %lu\n", info->len);
1753		return -ENOEXEC;
1754	}
1755	if (memcmp(info->hdr->e_ident, ELFMAG, SELFMAG) != 0) {
1756		pr_err("Invalid ELF header magic: != %s\n", ELFMAG);
1757		return -ENOEXEC;
1758	}
1759	if (info->hdr->e_type != ET_REL) {
1760		pr_err("Invalid ELF header type: %u != %u\n",
1761		       info->hdr->e_type, ET_REL);
1762		return -ENOEXEC;
1763	}
1764	if (!elf_check_arch(info->hdr)) {
1765		pr_err("Invalid architecture in ELF header: %u\n",
1766		       info->hdr->e_machine);
1767		return -ENOEXEC;
1768	}
1769	if (!module_elf_check_arch(info->hdr)) {
1770		pr_err("Invalid module architecture in ELF header: %u\n",
1771		       info->hdr->e_machine);
1772		return -ENOEXEC;
1773	}
1774	return 0;
1775}
1776
1777/**
1778 * elf_validity_cache_sechdrs() - Cache section headers if valid
1779 * @info: Load info to compute section headers from
1780 *
1781 * Checks:
1782 *
1783 * * ELF header is valid (see elf_validity_ehdr())
1784 * * Section headers are the size we expect
1785 * * Section array fits in the user provided data
1786 * * Section index 0 is NULL
1787 * * Section contents are inbounds
1788 *
1789 * Then updates @info with a &load_info->sechdrs pointer if valid.
1790 *
1791 * Return: %0 if valid, negative error code if validation failed.
1792 */
1793static int elf_validity_cache_sechdrs(struct load_info *info)
1794{
1795	Elf_Shdr *sechdrs;
1796	Elf_Shdr *shdr;
1797	int i;
1798	int err;
1799
1800	err = elf_validity_ehdr(info);
1801	if (err < 0)
1802		return err;
1803
1804	if (info->hdr->e_shentsize != sizeof(Elf_Shdr)) {
1805		pr_err("Invalid ELF section header size\n");
1806		return -ENOEXEC;
1807	}
1808
1809	/*
1810	 * e_shnum is 16 bits, and sizeof(Elf_Shdr) is
1811	 * known and small. So e_shnum * sizeof(Elf_Shdr)
1812	 * will not overflow unsigned long on any platform.
1813	 */
1814	if (info->hdr->e_shoff >= info->len
1815	    || (info->hdr->e_shnum * sizeof(Elf_Shdr) >
1816		info->len - info->hdr->e_shoff)) {
1817		pr_err("Invalid ELF section header overflow\n");
1818		return -ENOEXEC;
1819	}
1820
1821	sechdrs = (void *)info->hdr + info->hdr->e_shoff;
1822
1823	/*
1824	 * The code assumes that section 0 has a length of zero and
1825	 * an addr of zero, so check for it.
1826	 */
1827	if (sechdrs[0].sh_type != SHT_NULL
1828	    || sechdrs[0].sh_size != 0
1829	    || sechdrs[0].sh_addr != 0) {
1830		pr_err("ELF Spec violation: section 0 type(%d)!=SH_NULL or non-zero len or addr\n",
1831		       sechdrs[0].sh_type);
1832		return -ENOEXEC;
1833	}
1834
1835	/* Validate contents are inbounds */
1836	for (i = 1; i < info->hdr->e_shnum; i++) {
1837		shdr = &sechdrs[i];
1838		switch (shdr->sh_type) {
1839		case SHT_NULL:
1840		case SHT_NOBITS:
1841			/* No contents, offset/size don't mean anything */
1842			continue;
1843		default:
1844			err = validate_section_offset(info, shdr);
1845			if (err < 0) {
1846				pr_err("Invalid ELF section in module (section %u type %u)\n",
1847				       i, shdr->sh_type);
1848				return err;
1849			}
1850		}
1851	}
1852
1853	info->sechdrs = sechdrs;
1854
1855	return 0;
1856}
1857
1858/**
1859 * elf_validity_cache_secstrings() - Caches section names if valid
1860 * @info: Load info to cache section names from. Must have valid sechdrs.
1861 *
1862 * Specifically checks:
1863 *
1864 * * Section name table index is inbounds of section headers
1865 * * Section name table is not empty
1866 * * Section name table is NUL terminated
1867 * * All section name offsets are inbounds of the section
1868 *
1869 * Then updates @info with a &load_info->secstrings pointer if valid.
1870 *
1871 * Return: %0 if valid, negative error code if validation failed.
1872 */
1873static int elf_validity_cache_secstrings(struct load_info *info)
1874{
1875	Elf_Shdr *strhdr, *shdr;
1876	char *secstrings;
1877	int i;
1878
1879	/*
1880	 * Verify if the section name table index is valid.
1881	 */
1882	if (info->hdr->e_shstrndx == SHN_UNDEF
1883	    || info->hdr->e_shstrndx >= info->hdr->e_shnum) {
1884		pr_err("Invalid ELF section name index: %d || e_shstrndx (%d) >= e_shnum (%d)\n",
1885		       info->hdr->e_shstrndx, info->hdr->e_shstrndx,
1886		       info->hdr->e_shnum);
1887		return -ENOEXEC;
1888	}
1889
1890	strhdr = &info->sechdrs[info->hdr->e_shstrndx];
1891
1892	/*
1893	 * The section name table must be NUL-terminated, as required
1894	 * by the spec. This makes strcmp and pr_* calls that access
1895	 * strings in the section safe.
1896	 */
1897	secstrings = (void *)info->hdr + strhdr->sh_offset;
1898	if (strhdr->sh_size == 0) {
1899		pr_err("empty section name table\n");
1900		return -ENOEXEC;
1901	}
1902	if (secstrings[strhdr->sh_size - 1] != '\0') {
1903		pr_err("ELF Spec violation: section name table isn't null terminated\n");
1904		return -ENOEXEC;
1905	}
1906
1907	for (i = 0; i < info->hdr->e_shnum; i++) {
1908		shdr = &info->sechdrs[i];
1909		/* SHT_NULL means sh_name has an undefined value */
1910		if (shdr->sh_type == SHT_NULL)
1911			continue;
1912		if (shdr->sh_name >= strhdr->sh_size) {
1913			pr_err("Invalid ELF section name in module (section %u type %u)\n",
1914			       i, shdr->sh_type);
1915			return -ENOEXEC;
1916		}
1917	}
1918
1919	info->secstrings = secstrings;
1920	return 0;
1921}
1922
1923/**
1924 * elf_validity_cache_index_info() - Validate and cache modinfo section
1925 * @info: Load info to populate the modinfo index on.
1926 *        Must have &load_info->sechdrs and &load_info->secstrings populated
1927 *
1928 * Checks that if there is a .modinfo section, it is unique.
1929 * Then, it caches its index in &load_info->index.info.
1930 * Finally, it tries to populate the name to improve error messages.
1931 *
1932 * Return: %0 if valid, %-ENOEXEC if multiple modinfo sections were found.
1933 */
1934static int elf_validity_cache_index_info(struct load_info *info)
1935{
1936	int info_idx;
1937
1938	info_idx = find_any_unique_sec(info, ".modinfo");
1939
1940	if (info_idx == 0)
1941		/* Early return, no .modinfo */
1942		return 0;
1943
1944	if (info_idx < 0) {
1945		pr_err("Only one .modinfo section must exist.\n");
1946		return -ENOEXEC;
1947	}
1948
1949	info->index.info = info_idx;
1950	/* Try to find a name early so we can log errors with a module name */
1951	info->name = get_modinfo(info, "name");
1952
1953	return 0;
1954}
1955
1956/**
1957 * elf_validity_cache_index_mod() - Validates and caches this_module section
1958 * @info: Load info to cache this_module on.
1959 *        Must have &load_info->sechdrs and &load_info->secstrings populated
1960 *
1961 * The ".gnu.linkonce.this_module" ELF section is special. It is what modpost
1962 * uses to refer to __this_module and let's use rely on THIS_MODULE to point
1963 * to &__this_module properly. The kernel's modpost declares it on each
1964 * modules's *.mod.c file. If the struct module of the kernel changes a full
1965 * kernel rebuild is required.
1966 *
1967 * We have a few expectations for this special section, this function
1968 * validates all this for us:
1969 *
1970 * * The section has contents
1971 * * The section is unique
1972 * * We expect the kernel to always have to allocate it: SHF_ALLOC
1973 * * The section size must match the kernel's run time's struct module
1974 *   size
1975 *
1976 * If all checks pass, the index will be cached in &load_info->index.mod
1977 *
1978 * Return: %0 on validation success, %-ENOEXEC on failure
1979 */
1980static int elf_validity_cache_index_mod(struct load_info *info)
1981{
1982	Elf_Shdr *shdr;
1983	int mod_idx;
1984
1985	mod_idx = find_any_unique_sec(info, ".gnu.linkonce.this_module");
1986	if (mod_idx <= 0) {
1987		pr_err("module %s: Exactly one .gnu.linkonce.this_module section must exist.\n",
1988		       info->name ?: "(missing .modinfo section or name field)");
1989		return -ENOEXEC;
1990	}
1991
1992	shdr = &info->sechdrs[mod_idx];
1993
1994	if (shdr->sh_type == SHT_NOBITS) {
1995		pr_err("module %s: .gnu.linkonce.this_module section must have a size set\n",
1996		       info->name ?: "(missing .modinfo section or name field)");
1997		return -ENOEXEC;
1998	}
1999
2000	if (!(shdr->sh_flags & SHF_ALLOC)) {
2001		pr_err("module %s: .gnu.linkonce.this_module must occupy memory during process execution\n",
2002		       info->name ?: "(missing .modinfo section or name field)");
2003		return -ENOEXEC;
2004	}
2005
2006	if (shdr->sh_size != sizeof(struct module)) {
2007		pr_err("module %s: .gnu.linkonce.this_module section size must match the kernel's built struct module size at run time\n",
2008		       info->name ?: "(missing .modinfo section or name field)");
2009		return -ENOEXEC;
2010	}
2011
2012	info->index.mod = mod_idx;
2013
2014	return 0;
2015}
2016
2017/**
2018 * elf_validity_cache_index_sym() - Validate and cache symtab index
2019 * @info: Load info to cache symtab index in.
2020 *        Must have &load_info->sechdrs and &load_info->secstrings populated.
2021 *
2022 * Checks that there is exactly one symbol table, then caches its index in
2023 * &load_info->index.sym.
2024 *
2025 * Return: %0 if valid, %-ENOEXEC on failure.
2026 */
2027static int elf_validity_cache_index_sym(struct load_info *info)
2028{
2029	unsigned int sym_idx;
2030	unsigned int num_sym_secs = 0;
2031	int i;
2032
2033	for (i = 1; i < info->hdr->e_shnum; i++) {
2034		if (info->sechdrs[i].sh_type == SHT_SYMTAB) {
2035			num_sym_secs++;
2036			sym_idx = i;
2037		}
2038	}
2039
2040	if (num_sym_secs != 1) {
2041		pr_warn("%s: module has no symbols (stripped?)\n",
2042			info->name ?: "(missing .modinfo section or name field)");
2043		return -ENOEXEC;
2044	}
2045
2046	info->index.sym = sym_idx;
2047
2048	return 0;
2049}
2050
2051/**
2052 * elf_validity_cache_index_str() - Validate and cache strtab index
2053 * @info: Load info to cache strtab index in.
2054 *        Must have &load_info->sechdrs and &load_info->secstrings populated.
2055 *        Must have &load_info->index.sym populated.
2056 *
2057 * Looks at the symbol table's associated string table, makes sure it is
2058 * in-bounds, and caches it.
2059 *
2060 * Return: %0 if valid, %-ENOEXEC on failure.
2061 */
2062static int elf_validity_cache_index_str(struct load_info *info)
2063{
2064	unsigned int str_idx = info->sechdrs[info->index.sym].sh_link;
2065
2066	if (str_idx == SHN_UNDEF || str_idx >= info->hdr->e_shnum) {
2067		pr_err("Invalid ELF sh_link!=SHN_UNDEF(%d) or (sh_link(%d) >= hdr->e_shnum(%d)\n",
2068		       str_idx, str_idx, info->hdr->e_shnum);
2069		return -ENOEXEC;
2070	}
2071
2072	info->index.str = str_idx;
2073	return 0;
2074}
2075
2076/**
2077 * elf_validity_cache_index() - Resolve, validate, cache section indices
2078 * @info:  Load info to read from and update.
2079 *         &load_info->sechdrs and &load_info->secstrings must be populated.
2080 * @flags: Load flags, relevant to suppress version loading, see
2081 *         uapi/linux/module.h
2082 *
2083 * Populates &load_info->index, validating as it goes.
2084 * See child functions for per-field validation:
2085 *
2086 * * elf_validity_cache_index_info()
2087 * * elf_validity_cache_index_mod()
2088 * * elf_validity_cache_index_sym()
2089 * * elf_validity_cache_index_str()
2090 *
2091 * If versioning is not suppressed via flags, load the version index from
2092 * a section called "__versions" with no validation.
2093 *
2094 * If CONFIG_SMP is enabled, load the percpu section by name with no
2095 * validation.
2096 *
2097 * Return: 0 on success, negative error code if an index failed validation.
2098 */
2099static int elf_validity_cache_index(struct load_info *info, int flags)
2100{
2101	int err;
2102
2103	err = elf_validity_cache_index_info(info);
2104	if (err < 0)
2105		return err;
2106	err = elf_validity_cache_index_mod(info);
2107	if (err < 0)
2108		return err;
2109	err = elf_validity_cache_index_sym(info);
2110	if (err < 0)
2111		return err;
2112	err = elf_validity_cache_index_str(info);
2113	if (err < 0)
2114		return err;
2115
2116	if (flags & MODULE_INIT_IGNORE_MODVERSIONS)
2117		info->index.vers = 0; /* Pretend no __versions section! */
2118	else
2119		info->index.vers = find_sec(info, "__versions");
2120
2121	info->index.pcpu = find_pcpusec(info);
2122
2123	return 0;
2124}
2125
2126/**
2127 * elf_validity_cache_strtab() - Validate and cache symbol string table
2128 * @info: Load info to read from and update.
2129 *        Must have &load_info->sechdrs and &load_info->secstrings populated.
2130 *        Must have &load_info->index populated.
2131 *
2132 * Checks:
2133 *
2134 * * The string table is not empty.
2135 * * The string table starts and ends with NUL (required by ELF spec).
2136 * * Every &Elf_Sym->st_name offset in the symbol table is inbounds of the
2137 *   string table.
2138 *
2139 * And caches the pointer as &load_info->strtab in @info.
2140 *
2141 * Return: 0 on success, negative error code if a check failed.
2142 */
2143static int elf_validity_cache_strtab(struct load_info *info)
2144{
2145	Elf_Shdr *str_shdr = &info->sechdrs[info->index.str];
2146	Elf_Shdr *sym_shdr = &info->sechdrs[info->index.sym];
2147	char *strtab = (char *)info->hdr + str_shdr->sh_offset;
2148	Elf_Sym *syms = (void *)info->hdr + sym_shdr->sh_offset;
2149	int i;
2150
2151	if (str_shdr->sh_size == 0) {
2152		pr_err("empty symbol string table\n");
2153		return -ENOEXEC;
2154	}
2155	if (strtab[0] != '\0') {
2156		pr_err("symbol string table missing leading NUL\n");
2157		return -ENOEXEC;
2158	}
2159	if (strtab[str_shdr->sh_size - 1] != '\0') {
2160		pr_err("symbol string table isn't NUL terminated\n");
2161		return -ENOEXEC;
2162	}
2163
2164	/*
2165	 * Now that we know strtab is correctly structured, check symbol
2166	 * starts are inbounds before they're used later.
2167	 */
2168	for (i = 0; i < sym_shdr->sh_size / sizeof(*syms); i++) {
2169		if (syms[i].st_name >= str_shdr->sh_size) {
2170			pr_err("symbol name out of bounds in string table");
2171			return -ENOEXEC;
2172		}
2173	}
2174
2175	info->strtab = strtab;
2176	return 0;
2177}
2178
2179/*
2180 * Check userspace passed ELF module against our expectations, and cache
2181 * useful variables for further processing as we go.
2182 *
2183 * This does basic validity checks against section offsets and sizes, the
2184 * section name string table, and the indices used for it (sh_name).
2185 *
2186 * As a last step, since we're already checking the ELF sections we cache
2187 * useful variables which will be used later for our convenience:
2188 *
2189 * 	o pointers to section headers
2190 * 	o cache the modinfo symbol section
2191 * 	o cache the string symbol section
2192 * 	o cache the module section
2193 *
2194 * As a last step we set info->mod to the temporary copy of the module in
2195 * info->hdr. The final one will be allocated in move_module(). Any
2196 * modifications we make to our copy of the module will be carried over
2197 * to the final minted module.
2198 */
2199static int elf_validity_cache_copy(struct load_info *info, int flags)
2200{
2201	int err;
2202
2203	err = elf_validity_cache_sechdrs(info);
2204	if (err < 0)
2205		return err;
2206	err = elf_validity_cache_secstrings(info);
2207	if (err < 0)
2208		return err;
2209	err = elf_validity_cache_index(info, flags);
2210	if (err < 0)
2211		return err;
2212	err = elf_validity_cache_strtab(info);
2213	if (err < 0)
2214		return err;
2215
2216	/* This is temporary: point mod into copy of data. */
2217	info->mod = (void *)info->hdr + info->sechdrs[info->index.mod].sh_offset;
2218
2219	/*
2220	 * If we didn't load the .modinfo 'name' field earlier, fall back to
2221	 * on-disk struct mod 'name' field.
2222	 */
2223	if (!info->name)
2224		info->name = info->mod->name;
2225
2226	return 0;
2227}
2228
2229#define COPY_CHUNK_SIZE (16*PAGE_SIZE)
2230
2231static int copy_chunked_from_user(void *dst, const void __user *usrc, unsigned long len)
2232{
2233	do {
2234		unsigned long n = min(len, COPY_CHUNK_SIZE);
2235
2236		if (copy_from_user(dst, usrc, n) != 0)
2237			return -EFAULT;
2238		cond_resched();
2239		dst += n;
2240		usrc += n;
2241		len -= n;
2242	} while (len);
2243	return 0;
2244}
2245
2246static int check_modinfo_livepatch(struct module *mod, struct load_info *info)
2247{
2248	if (!get_modinfo(info, "livepatch"))
2249		/* Nothing more to do */
2250		return 0;
2251
2252	if (set_livepatch_module(mod))
2253		return 0;
2254
2255	pr_err("%s: module is marked as livepatch module, but livepatch support is disabled",
2256	       mod->name);
2257	return -ENOEXEC;
2258}
2259
2260static void check_modinfo_retpoline(struct module *mod, struct load_info *info)
2261{
2262	if (retpoline_module_ok(get_modinfo(info, "retpoline")))
2263		return;
2264
2265	pr_warn("%s: loading module not compiled with retpoline compiler.\n",
2266		mod->name);
2267}
2268
2269/* Sets info->hdr and info->len. */
2270static int copy_module_from_user(const void __user *umod, unsigned long len,
2271				  struct load_info *info)
2272{
2273	int err;
2274
2275	info->len = len;
2276	if (info->len < sizeof(*(info->hdr)))
2277		return -ENOEXEC;
2278
2279	err = security_kernel_load_data(LOADING_MODULE, true);
2280	if (err)
2281		return err;
2282
2283	/* Suck in entire file: we'll want most of it. */
2284	info->hdr = __vmalloc(info->len, GFP_KERNEL | __GFP_NOWARN);
2285	if (!info->hdr)
2286		return -ENOMEM;
2287
2288	if (copy_chunked_from_user(info->hdr, umod, info->len) != 0) {
2289		err = -EFAULT;
2290		goto out;
2291	}
2292
2293	err = security_kernel_post_load_data((char *)info->hdr, info->len,
2294					     LOADING_MODULE, "init_module");
2295out:
2296	if (err)
2297		vfree(info->hdr);
2298
2299	return err;
2300}
2301
2302static void free_copy(struct load_info *info, int flags)
2303{
2304	if (flags & MODULE_INIT_COMPRESSED_FILE)
2305		module_decompress_cleanup(info);
2306	else
2307		vfree(info->hdr);
2308}
2309
2310static int rewrite_section_headers(struct load_info *info, int flags)
2311{
2312	unsigned int i;
2313
2314	/* This should always be true, but let's be sure. */
2315	info->sechdrs[0].sh_addr = 0;
2316
2317	for (i = 1; i < info->hdr->e_shnum; i++) {
2318		Elf_Shdr *shdr = &info->sechdrs[i];
2319
2320		/*
2321		 * Mark all sections sh_addr with their address in the
2322		 * temporary image.
2323		 */
2324		shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2325
2326	}
2327
2328	/* Track but don't keep modinfo and version sections. */
2329	info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2330	info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2331
2332	return 0;
2333}
2334
2335/*
2336 * These calls taint the kernel depending certain module circumstances */
2337static void module_augment_kernel_taints(struct module *mod, struct load_info *info)
2338{
2339	int prev_taint = test_taint(TAINT_PROPRIETARY_MODULE);
2340
2341	if (!get_modinfo(info, "intree")) {
2342		if (!test_taint(TAINT_OOT_MODULE))
2343			pr_warn("%s: loading out-of-tree module taints kernel.\n",
2344				mod->name);
2345		add_taint_module(mod, TAINT_OOT_MODULE, LOCKDEP_STILL_OK);
2346	}
2347
2348	check_modinfo_retpoline(mod, info);
2349
2350	if (get_modinfo(info, "staging")) {
2351		add_taint_module(mod, TAINT_CRAP, LOCKDEP_STILL_OK);
2352		pr_warn("%s: module is from the staging directory, the quality "
2353			"is unknown, you have been warned.\n", mod->name);
2354	}
2355
2356	if (is_livepatch_module(mod)) {
2357		add_taint_module(mod, TAINT_LIVEPATCH, LOCKDEP_STILL_OK);
2358		pr_notice_once("%s: tainting kernel with TAINT_LIVEPATCH\n",
2359				mod->name);
2360	}
2361
2362	module_license_taint_check(mod, get_modinfo(info, "license"));
2363
2364	if (get_modinfo(info, "test")) {
2365		if (!test_taint(TAINT_TEST))
2366			pr_warn("%s: loading test module taints kernel.\n",
2367				mod->name);
2368		add_taint_module(mod, TAINT_TEST, LOCKDEP_STILL_OK);
2369	}
2370#ifdef CONFIG_MODULE_SIG
2371	mod->sig_ok = info->sig_ok;
2372	if (!mod->sig_ok) {
2373		pr_notice_once("%s: module verification failed: signature "
2374			       "and/or required key missing - tainting "
2375			       "kernel\n", mod->name);
2376		add_taint_module(mod, TAINT_UNSIGNED_MODULE, LOCKDEP_STILL_OK);
2377	}
2378#endif
2379
2380	/*
2381	 * ndiswrapper is under GPL by itself, but loads proprietary modules.
2382	 * Don't use add_taint_module(), as it would prevent ndiswrapper from
2383	 * using GPL-only symbols it needs.
2384	 */
2385	if (strcmp(mod->name, "ndiswrapper") == 0)
2386		add_taint(TAINT_PROPRIETARY_MODULE, LOCKDEP_NOW_UNRELIABLE);
2387
2388	/* driverloader was caught wrongly pretending to be under GPL */
2389	if (strcmp(mod->name, "driverloader") == 0)
2390		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2391				 LOCKDEP_NOW_UNRELIABLE);
2392
2393	/* lve claims to be GPL but upstream won't provide source */
2394	if (strcmp(mod->name, "lve") == 0)
2395		add_taint_module(mod, TAINT_PROPRIETARY_MODULE,
2396				 LOCKDEP_NOW_UNRELIABLE);
2397
2398	if (!prev_taint && test_taint(TAINT_PROPRIETARY_MODULE))
2399		pr_warn("%s: module license taints kernel.\n", mod->name);
2400
2401}
2402
2403static int check_modinfo(struct module *mod, struct load_info *info, int flags)
2404{
2405	const char *modmagic = get_modinfo(info, "vermagic");
2406	int err;
2407
2408	if (flags & MODULE_INIT_IGNORE_VERMAGIC)
2409		modmagic = NULL;
2410
2411	/* This is allowed: modprobe --force will invalidate it. */
2412	if (!modmagic) {
2413		err = try_to_force_load(mod, "bad vermagic");
2414		if (err)
2415			return err;
2416	} else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2417		pr_err("%s: version magic '%s' should be '%s'\n",
2418		       info->name, modmagic, vermagic);
2419		return -ENOEXEC;
2420	}
2421
2422	err = check_modinfo_livepatch(mod, info);
2423	if (err)
2424		return err;
2425
2426	return 0;
2427}
2428
2429static int find_module_sections(struct module *mod, struct load_info *info)
2430{
2431	mod->kp = section_objs(info, "__param",
2432			       sizeof(*mod->kp), &mod->num_kp);
2433	mod->syms = section_objs(info, "__ksymtab",
2434				 sizeof(*mod->syms), &mod->num_syms);
2435	mod->crcs = section_addr(info, "__kcrctab");
2436	mod->gpl_syms = section_objs(info, "__ksymtab_gpl",
2437				     sizeof(*mod->gpl_syms),
2438				     &mod->num_gpl_syms);
2439	mod->gpl_crcs = section_addr(info, "__kcrctab_gpl");
2440
2441#ifdef CONFIG_CONSTRUCTORS
2442	mod->ctors = section_objs(info, ".ctors",
2443				  sizeof(*mod->ctors), &mod->num_ctors);
2444	if (!mod->ctors)
2445		mod->ctors = section_objs(info, ".init_array",
2446				sizeof(*mod->ctors), &mod->num_ctors);
2447	else if (find_sec(info, ".init_array")) {
2448		/*
2449		 * This shouldn't happen with same compiler and binutils
2450		 * building all parts of the module.
2451		 */
2452		pr_warn("%s: has both .ctors and .init_array.\n",
2453		       mod->name);
2454		return -EINVAL;
2455	}
2456#endif
2457
2458	mod->noinstr_text_start = section_objs(info, ".noinstr.text", 1,
2459						&mod->noinstr_text_size);
2460
2461#ifdef CONFIG_TRACEPOINTS
2462	mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2463					     sizeof(*mod->tracepoints_ptrs),
2464					     &mod->num_tracepoints);
2465#endif
2466#ifdef CONFIG_TREE_SRCU
2467	mod->srcu_struct_ptrs = section_objs(info, "___srcu_struct_ptrs",
2468					     sizeof(*mod->srcu_struct_ptrs),
2469					     &mod->num_srcu_structs);
2470#endif
2471#ifdef CONFIG_BPF_EVENTS
2472	mod->bpf_raw_events = section_objs(info, "__bpf_raw_tp_map",
2473					   sizeof(*mod->bpf_raw_events),
2474					   &mod->num_bpf_raw_events);
2475#endif
2476#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2477	mod->btf_data = any_section_objs(info, ".BTF", 1, &mod->btf_data_size);
2478	mod->btf_base_data = any_section_objs(info, ".BTF.base", 1,
2479					      &mod->btf_base_data_size);
2480#endif
2481#ifdef CONFIG_JUMP_LABEL
2482	mod->jump_entries = section_objs(info, "__jump_table",
2483					sizeof(*mod->jump_entries),
2484					&mod->num_jump_entries);
2485#endif
2486#ifdef CONFIG_EVENT_TRACING
2487	mod->trace_events = section_objs(info, "_ftrace_events",
2488					 sizeof(*mod->trace_events),
2489					 &mod->num_trace_events);
2490	mod->trace_evals = section_objs(info, "_ftrace_eval_map",
2491					sizeof(*mod->trace_evals),
2492					&mod->num_trace_evals);
2493#endif
2494#ifdef CONFIG_TRACING
2495	mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2496					 sizeof(*mod->trace_bprintk_fmt_start),
2497					 &mod->num_trace_bprintk_fmt);
2498#endif
2499#ifdef CONFIG_FTRACE_MCOUNT_RECORD
2500	/* sechdrs[0].sh_size is always zero */
2501	mod->ftrace_callsites = section_objs(info, FTRACE_CALLSITE_SECTION,
2502					     sizeof(*mod->ftrace_callsites),
2503					     &mod->num_ftrace_callsites);
2504#endif
2505#ifdef CONFIG_FUNCTION_ERROR_INJECTION
2506	mod->ei_funcs = section_objs(info, "_error_injection_whitelist",
2507					    sizeof(*mod->ei_funcs),
2508					    &mod->num_ei_funcs);
2509#endif
2510#ifdef CONFIG_KPROBES
2511	mod->kprobes_text_start = section_objs(info, ".kprobes.text", 1,
2512						&mod->kprobes_text_size);
2513	mod->kprobe_blacklist = section_objs(info, "_kprobe_blacklist",
2514						sizeof(unsigned long),
2515						&mod->num_kprobe_blacklist);
2516#endif
2517#ifdef CONFIG_PRINTK_INDEX
2518	mod->printk_index_start = section_objs(info, ".printk_index",
2519					       sizeof(*mod->printk_index_start),
2520					       &mod->printk_index_size);
2521#endif
2522#ifdef CONFIG_HAVE_STATIC_CALL_INLINE
2523	mod->static_call_sites = section_objs(info, ".static_call_sites",
2524					      sizeof(*mod->static_call_sites),
2525					      &mod->num_static_call_sites);
2526#endif
2527#if IS_ENABLED(CONFIG_KUNIT)
2528	mod->kunit_suites = section_objs(info, ".kunit_test_suites",
2529					      sizeof(*mod->kunit_suites),
2530					      &mod->num_kunit_suites);
2531	mod->kunit_init_suites = section_objs(info, ".kunit_init_test_suites",
2532					      sizeof(*mod->kunit_init_suites),
2533					      &mod->num_kunit_init_suites);
2534#endif
2535
2536	mod->extable = section_objs(info, "__ex_table",
2537				    sizeof(*mod->extable), &mod->num_exentries);
2538
2539	if (section_addr(info, "__obsparm"))
2540		pr_warn("%s: Ignoring obsolete parameters\n", mod->name);
2541
2542#ifdef CONFIG_DYNAMIC_DEBUG_CORE
2543	mod->dyndbg_info.descs = section_objs(info, "__dyndbg",
2544					      sizeof(*mod->dyndbg_info.descs),
2545					      &mod->dyndbg_info.num_descs);
2546	mod->dyndbg_info.classes = section_objs(info, "__dyndbg_classes",
2547						sizeof(*mod->dyndbg_info.classes),
2548						&mod->dyndbg_info.num_classes);
2549#endif
2550
2551	return 0;
2552}
2553
2554static int move_module(struct module *mod, struct load_info *info)
2555{
2556	int i;
2557	enum mod_mem_type t = 0;
2558	int ret = -ENOMEM;
2559	bool codetag_section_found = false;
2560
2561	for_each_mod_mem_type(type) {
2562		if (!mod->mem[type].size) {
2563			mod->mem[type].base = NULL;
2564			mod->mem[type].rw_copy = NULL;
2565			continue;
2566		}
2567
2568		ret = module_memory_alloc(mod, type);
2569		if (ret) {
2570			t = type;
2571			goto out_err;
2572		}
2573	}
2574
2575	/* Transfer each section which specifies SHF_ALLOC */
2576	pr_debug("Final section addresses for %s:\n", mod->name);
2577	for (i = 0; i < info->hdr->e_shnum; i++) {
2578		void *dest;
2579		Elf_Shdr *shdr = &info->sechdrs[i];
2580		const char *sname;
2581		unsigned long addr;
2582
2583		if (!(shdr->sh_flags & SHF_ALLOC))
2584			continue;
2585
2586		sname = info->secstrings + shdr->sh_name;
2587		/*
2588		 * Load codetag sections separately as they might still be used
2589		 * after module unload.
2590		 */
2591		if (codetag_needs_module_section(mod, sname, shdr->sh_size)) {
2592			dest = codetag_alloc_module_section(mod, sname, shdr->sh_size,
2593					arch_mod_section_prepend(mod, i), shdr->sh_addralign);
2594			if (WARN_ON(!dest)) {
2595				ret = -EINVAL;
2596				goto out_err;
2597			}
2598			if (IS_ERR(dest)) {
2599				ret = PTR_ERR(dest);
2600				goto out_err;
2601			}
2602			addr = (unsigned long)dest;
2603			codetag_section_found = true;
2604		} else {
2605			enum mod_mem_type type = shdr->sh_entsize >> SH_ENTSIZE_TYPE_SHIFT;
2606			unsigned long offset = shdr->sh_entsize & SH_ENTSIZE_OFFSET_MASK;
2607
2608			addr = (unsigned long)mod->mem[type].base + offset;
2609			dest = mod->mem[type].rw_copy + offset;
2610		}
2611
2612		if (shdr->sh_type != SHT_NOBITS) {
2613			/*
2614			 * Our ELF checker already validated this, but let's
2615			 * be pedantic and make the goal clearer. We actually
2616			 * end up copying over all modifications made to the
2617			 * userspace copy of the entire struct module.
2618			 */
2619			if (i == info->index.mod &&
2620			   (WARN_ON_ONCE(shdr->sh_size != sizeof(struct module)))) {
2621				ret = -ENOEXEC;
2622				goto out_err;
2623			}
2624			memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2625		}
2626		/*
2627		 * Update the userspace copy's ELF section address to point to
2628		 * our newly allocated memory as a pure convenience so that
2629		 * users of info can keep taking advantage and using the newly
2630		 * minted official memory area.
2631		 */
2632		shdr->sh_addr = addr;
2633		pr_debug("\t0x%lx 0x%.8lx %s\n", (long)shdr->sh_addr,
2634			 (long)shdr->sh_size, info->secstrings + shdr->sh_name);
2635	}
2636
2637	return 0;
2638out_err:
2639	for (t--; t >= 0; t--)
2640		module_memory_free(mod, t);
2641	if (codetag_section_found)
2642		codetag_free_module_sections(mod);
2643
2644	return ret;
2645}
2646
2647static int check_export_symbol_versions(struct module *mod)
2648{
2649#ifdef CONFIG_MODVERSIONS
2650	if ((mod->num_syms && !mod->crcs) ||
2651	    (mod->num_gpl_syms && !mod->gpl_crcs)) {
2652		return try_to_force_load(mod,
2653					 "no versions for exported symbols");
2654	}
2655#endif
2656	return 0;
2657}
2658
2659static void flush_module_icache(const struct module *mod)
2660{
2661	/*
2662	 * Flush the instruction cache, since we've played with text.
2663	 * Do it before processing of module parameters, so the module
2664	 * can provide parameter accessor functions of its own.
2665	 */
2666	for_each_mod_mem_type(type) {
2667		const struct module_memory *mod_mem = &mod->mem[type];
2668
2669		if (mod_mem->size) {
2670			flush_icache_range((unsigned long)mod_mem->base,
2671					   (unsigned long)mod_mem->base + mod_mem->size);
2672		}
2673	}
2674}
2675
2676bool __weak module_elf_check_arch(Elf_Ehdr *hdr)
2677{
2678	return true;
2679}
2680
2681int __weak module_frob_arch_sections(Elf_Ehdr *hdr,
2682				     Elf_Shdr *sechdrs,
2683				     char *secstrings,
2684				     struct module *mod)
2685{
2686	return 0;
2687}
2688
2689/* module_blacklist is a comma-separated list of module names */
2690static char *module_blacklist;
2691static bool blacklisted(const char *module_name)
2692{
2693	const char *p;
2694	size_t len;
2695
2696	if (!module_blacklist)
2697		return false;
2698
2699	for (p = module_blacklist; *p; p += len) {
2700		len = strcspn(p, ",");
2701		if (strlen(module_name) == len && !memcmp(module_name, p, len))
2702			return true;
2703		if (p[len] == ',')
2704			len++;
2705	}
2706	return false;
2707}
2708core_param(module_blacklist, module_blacklist, charp, 0400);
2709
2710static struct module *layout_and_allocate(struct load_info *info, int flags)
2711{
2712	struct module *mod;
2713	unsigned int ndx;
2714	int err;
2715
2716	/* Allow arches to frob section contents and sizes.  */
2717	err = module_frob_arch_sections(info->hdr, info->sechdrs,
2718					info->secstrings, info->mod);
2719	if (err < 0)
2720		return ERR_PTR(err);
2721
2722	err = module_enforce_rwx_sections(info->hdr, info->sechdrs,
2723					  info->secstrings, info->mod);
2724	if (err < 0)
2725		return ERR_PTR(err);
2726
2727	/* We will do a special allocation for per-cpu sections later. */
2728	info->sechdrs[info->index.pcpu].sh_flags &= ~(unsigned long)SHF_ALLOC;
2729
2730	/*
2731	 * Mark ro_after_init section with SHF_RO_AFTER_INIT so that
2732	 * layout_sections() can put it in the right place.
2733	 * Note: ro_after_init sections also have SHF_{WRITE,ALLOC} set.
2734	 */
2735	ndx = find_sec(info, ".data..ro_after_init");
2736	if (ndx)
2737		info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2738	/*
2739	 * Mark the __jump_table section as ro_after_init as well: these data
2740	 * structures are never modified, with the exception of entries that
2741	 * refer to code in the __init section, which are annotated as such
2742	 * at module load time.
2743	 */
2744	ndx = find_sec(info, "__jump_table");
2745	if (ndx)
2746		info->sechdrs[ndx].sh_flags |= SHF_RO_AFTER_INIT;
2747
2748	/*
2749	 * Determine total sizes, and put offsets in sh_entsize.  For now
2750	 * this is done generically; there doesn't appear to be any
2751	 * special cases for the architectures.
2752	 */
2753	layout_sections(info->mod, info);
2754	layout_symtab(info->mod, info);
2755
2756	/* Allocate and move to the final place */
2757	err = move_module(info->mod, info);
2758	if (err)
2759		return ERR_PTR(err);
2760
2761	/* Module has been copied to its final place now: return it. */
2762	mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2763	kmemleak_load_module(mod, info);
2764	codetag_module_replaced(info->mod, mod);
2765
2766	return mod;
2767}
2768
2769/* mod is no longer valid after this! */
2770static void module_deallocate(struct module *mod, struct load_info *info)
2771{
2772	percpu_modfree(mod);
2773	module_arch_freeing_init(mod);
2774
2775	free_mod_mem(mod);
2776}
2777
2778int __weak module_finalize(const Elf_Ehdr *hdr,
2779			   const Elf_Shdr *sechdrs,
2780			   struct module *me)
2781{
2782	return 0;
2783}
2784
2785int __weak module_post_finalize(const Elf_Ehdr *hdr,
2786				const Elf_Shdr *sechdrs,
2787				struct module *me)
2788{
2789	return 0;
2790}
2791
2792static int post_relocation(struct module *mod, const struct load_info *info)
2793{
2794	int ret;
2795
2796	/* Sort exception table now relocations are done. */
2797	sort_extable(mod->extable, mod->extable + mod->num_exentries);
2798
2799	/* Copy relocated percpu area over. */
2800	percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
2801		       info->sechdrs[info->index.pcpu].sh_size);
2802
2803	/* Setup kallsyms-specific fields. */
2804	add_kallsyms(mod, info);
2805
2806	/* Arch-specific module finalizing. */
2807	ret = module_finalize(info->hdr, info->sechdrs, mod);
2808	if (ret)
2809		return ret;
2810
2811	for_each_mod_mem_type(type) {
2812		struct module_memory *mem = &mod->mem[type];
2813
2814		if (mem->is_rox) {
2815			if (!execmem_update_copy(mem->base, mem->rw_copy,
2816						 mem->size))
2817				return -ENOMEM;
2818
2819			vfree(mem->rw_copy);
2820			mem->rw_copy = NULL;
2821		}
2822	}
2823
2824	return module_post_finalize(info->hdr, info->sechdrs, mod);
2825}
2826
2827/* Call module constructors. */
2828static void do_mod_ctors(struct module *mod)
2829{
2830#ifdef CONFIG_CONSTRUCTORS
2831	unsigned long i;
2832
2833	for (i = 0; i < mod->num_ctors; i++)
2834		mod->ctors[i]();
2835#endif
2836}
2837
2838/* For freeing module_init on success, in case kallsyms traversing */
2839struct mod_initfree {
2840	struct llist_node node;
2841	void *init_text;
2842	void *init_data;
2843	void *init_rodata;
2844};
2845
2846static void do_free_init(struct work_struct *w)
2847{
2848	struct llist_node *pos, *n, *list;
2849	struct mod_initfree *initfree;
2850
2851	list = llist_del_all(&init_free_list);
2852
2853	synchronize_rcu();
2854
2855	llist_for_each_safe(pos, n, list) {
2856		initfree = container_of(pos, struct mod_initfree, node);
2857		execmem_free(initfree->init_text);
2858		execmem_free(initfree->init_data);
2859		execmem_free(initfree->init_rodata);
2860		kfree(initfree);
2861	}
2862}
2863
2864void flush_module_init_free_work(void)
2865{
2866	flush_work(&init_free_wq);
2867}
2868
2869#undef MODULE_PARAM_PREFIX
2870#define MODULE_PARAM_PREFIX "module."
2871/* Default value for module->async_probe_requested */
2872static bool async_probe;
2873module_param(async_probe, bool, 0644);
2874
2875/*
2876 * This is where the real work happens.
2877 *
2878 * Keep it uninlined to provide a reliable breakpoint target, e.g. for the gdb
2879 * helper command 'lx-symbols'.
2880 */
2881static noinline int do_init_module(struct module *mod)
2882{
2883	int ret = 0;
2884	struct mod_initfree *freeinit;
2885#if defined(CONFIG_MODULE_STATS)
2886	unsigned int text_size = 0, total_size = 0;
2887
2888	for_each_mod_mem_type(type) {
2889		const struct module_memory *mod_mem = &mod->mem[type];
2890		if (mod_mem->size) {
2891			total_size += mod_mem->size;
2892			if (type == MOD_TEXT || type == MOD_INIT_TEXT)
2893				text_size += mod_mem->size;
2894		}
2895	}
2896#endif
2897
2898	freeinit = kmalloc(sizeof(*freeinit), GFP_KERNEL);
2899	if (!freeinit) {
2900		ret = -ENOMEM;
2901		goto fail;
2902	}
2903	freeinit->init_text = mod->mem[MOD_INIT_TEXT].base;
2904	freeinit->init_data = mod->mem[MOD_INIT_DATA].base;
2905	freeinit->init_rodata = mod->mem[MOD_INIT_RODATA].base;
2906
2907	do_mod_ctors(mod);
2908	/* Start the module */
2909	if (mod->init != NULL)
2910		ret = do_one_initcall(mod->init);
2911	if (ret < 0) {
2912		goto fail_free_freeinit;
2913	}
2914	if (ret > 0) {
2915		pr_warn("%s: '%s'->init suspiciously returned %d, it should "
2916			"follow 0/-E convention\n"
2917			"%s: loading module anyway...\n",
2918			__func__, mod->name, ret, __func__);
2919		dump_stack();
2920	}
2921
2922	/* Now it's a first class citizen! */
2923	mod->state = MODULE_STATE_LIVE;
2924	blocking_notifier_call_chain(&module_notify_list,
2925				     MODULE_STATE_LIVE, mod);
2926
2927	/* Delay uevent until module has finished its init routine */
2928	kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
2929
2930	/*
2931	 * We need to finish all async code before the module init sequence
2932	 * is done. This has potential to deadlock if synchronous module
2933	 * loading is requested from async (which is not allowed!).
2934	 *
2935	 * See commit 0fdff3ec6d87 ("async, kmod: warn on synchronous
2936	 * request_module() from async workers") for more details.
2937	 */
2938	if (!mod->async_probe_requested)
2939		async_synchronize_full();
2940
2941	ftrace_free_mem(mod, mod->mem[MOD_INIT_TEXT].base,
2942			mod->mem[MOD_INIT_TEXT].base + mod->mem[MOD_INIT_TEXT].size);
2943	mutex_lock(&module_mutex);
2944	/* Drop initial reference. */
2945	module_put(mod);
2946	trim_init_extable(mod);
2947#ifdef CONFIG_KALLSYMS
2948	/* Switch to core kallsyms now init is done: kallsyms may be walking! */
2949	rcu_assign_pointer(mod->kallsyms, &mod->core_kallsyms);
2950#endif
2951	ret = module_enable_rodata_ro(mod, true);
2952	if (ret)
2953		pr_warn("%s: module_enable_rodata_ro_after_init() returned %d, "
2954			"ro_after_init data might still be writable\n",
2955			mod->name, ret);
2956
2957	mod_tree_remove_init(mod);
2958	module_arch_freeing_init(mod);
2959	for_class_mod_mem_type(type, init) {
2960		mod->mem[type].base = NULL;
2961		mod->mem[type].size = 0;
2962	}
2963
2964#ifdef CONFIG_DEBUG_INFO_BTF_MODULES
2965	/* .BTF is not SHF_ALLOC and will get removed, so sanitize pointers */
2966	mod->btf_data = NULL;
2967	mod->btf_base_data = NULL;
2968#endif
2969	/*
2970	 * We want to free module_init, but be aware that kallsyms may be
2971	 * walking this with preempt disabled.  In all the failure paths, we
2972	 * call synchronize_rcu(), but we don't want to slow down the success
2973	 * path. execmem_free() cannot be called in an interrupt, so do the
2974	 * work and call synchronize_rcu() in a work queue.
2975	 *
2976	 * Note that execmem_alloc() on most architectures creates W+X page
2977	 * mappings which won't be cleaned up until do_free_init() runs.  Any
2978	 * code such as mark_rodata_ro() which depends on those mappings to
2979	 * be cleaned up needs to sync with the queued work by invoking
2980	 * flush_module_init_free_work().
2981	 */
2982	if (llist_add(&freeinit->node, &init_free_list))
2983		schedule_work(&init_free_wq);
2984
2985	mutex_unlock(&module_mutex);
2986	wake_up_all(&module_wq);
2987
2988	mod_stat_add_long(text_size, &total_text_size);
2989	mod_stat_add_long(total_size, &total_mod_size);
2990
2991	mod_stat_inc(&modcount);
2992
2993	return 0;
2994
2995fail_free_freeinit:
2996	kfree(freeinit);
2997fail:
2998	/* Try to protect us from buggy refcounters. */
2999	mod->state = MODULE_STATE_GOING;
3000	synchronize_rcu();
3001	module_put(mod);
3002	blocking_notifier_call_chain(&module_notify_list,
3003				     MODULE_STATE_GOING, mod);
3004	klp_module_going(mod);
3005	ftrace_release_mod(mod);
3006	free_module(mod);
3007	wake_up_all(&module_wq);
3008
3009	return ret;
3010}
3011
3012static int may_init_module(void)
3013{
3014	if (!capable(CAP_SYS_MODULE) || modules_disabled)
3015		return -EPERM;
3016
3017	return 0;
3018}
3019
3020/* Is this module of this name done loading?  No locks held. */
3021static bool finished_loading(const char *name)
3022{
3023	struct module *mod;
3024	bool ret;
3025
3026	/*
3027	 * The module_mutex should not be a heavily contended lock;
3028	 * if we get the occasional sleep here, we'll go an extra iteration
3029	 * in the wait_event_interruptible(), which is harmless.
3030	 */
3031	sched_annotate_sleep();
3032	mutex_lock(&module_mutex);
3033	mod = find_module_all(name, strlen(name), true);
3034	ret = !mod || mod->state == MODULE_STATE_LIVE
3035		|| mod->state == MODULE_STATE_GOING;
3036	mutex_unlock(&module_mutex);
3037
3038	return ret;
3039}
3040
3041/* Must be called with module_mutex held */
3042static int module_patient_check_exists(const char *name,
3043				       enum fail_dup_mod_reason reason)
3044{
3045	struct module *old;
3046	int err = 0;
3047
3048	old = find_module_all(name, strlen(name), true);
3049	if (old == NULL)
3050		return 0;
3051
3052	if (old->state == MODULE_STATE_COMING ||
3053	    old->state == MODULE_STATE_UNFORMED) {
3054		/* Wait in case it fails to load. */
3055		mutex_unlock(&module_mutex);
3056		err = wait_event_interruptible(module_wq,
3057				       finished_loading(name));
3058		mutex_lock(&module_mutex);
3059		if (err)
3060			return err;
3061
3062		/* The module might have gone in the meantime. */
3063		old = find_module_all(name, strlen(name), true);
3064	}
3065
3066	if (try_add_failed_module(name, reason))
3067		pr_warn("Could not add fail-tracking for module: %s\n", name);
3068
3069	/*
3070	 * We are here only when the same module was being loaded. Do
3071	 * not try to load it again right now. It prevents long delays
3072	 * caused by serialized module load failures. It might happen
3073	 * when more devices of the same type trigger load of
3074	 * a particular module.
3075	 */
3076	if (old && old->state == MODULE_STATE_LIVE)
3077		return -EEXIST;
3078	return -EBUSY;
3079}
3080
3081/*
3082 * We try to place it in the list now to make sure it's unique before
3083 * we dedicate too many resources.  In particular, temporary percpu
3084 * memory exhaustion.
3085 */
3086static int add_unformed_module(struct module *mod)
3087{
3088	int err;
3089
3090	mod->state = MODULE_STATE_UNFORMED;
3091
3092	mutex_lock(&module_mutex);
3093	err = module_patient_check_exists(mod->name, FAIL_DUP_MOD_LOAD);
3094	if (err)
3095		goto out;
3096
3097	mod_update_bounds(mod);
3098	list_add_rcu(&mod->list, &modules);
3099	mod_tree_insert(mod);
3100	err = 0;
3101
3102out:
3103	mutex_unlock(&module_mutex);
3104	return err;
3105}
3106
3107static int complete_formation(struct module *mod, struct load_info *info)
3108{
3109	int err;
3110
3111	mutex_lock(&module_mutex);
3112
3113	/* Find duplicate symbols (must be called under lock). */
3114	err = verify_exported_symbols(mod);
3115	if (err < 0)
3116		goto out;
3117
3118	/* These rely on module_mutex for list integrity. */
3119	module_bug_finalize(info->hdr, info->sechdrs, mod);
3120	module_cfi_finalize(info->hdr, info->sechdrs, mod);
3121
3122	err = module_enable_rodata_ro(mod, false);
3123	if (err)
3124		goto out_strict_rwx;
3125	err = module_enable_data_nx(mod);
3126	if (err)
3127		goto out_strict_rwx;
3128	err = module_enable_text_rox(mod);
3129	if (err)
3130		goto out_strict_rwx;
3131
3132	/*
3133	 * Mark state as coming so strong_try_module_get() ignores us,
3134	 * but kallsyms etc. can see us.
3135	 */
3136	mod->state = MODULE_STATE_COMING;
3137	mutex_unlock(&module_mutex);
3138
3139	return 0;
3140
3141out_strict_rwx:
3142	module_bug_cleanup(mod);
3143out:
3144	mutex_unlock(&module_mutex);
3145	return err;
3146}
3147
3148static int prepare_coming_module(struct module *mod)
3149{
3150	int err;
3151
3152	ftrace_module_enable(mod);
3153	err = klp_module_coming(mod);
3154	if (err)
3155		return err;
3156
3157	err = blocking_notifier_call_chain_robust(&module_notify_list,
3158			MODULE_STATE_COMING, MODULE_STATE_GOING, mod);
3159	err = notifier_to_errno(err);
3160	if (err)
3161		klp_module_going(mod);
3162
3163	return err;
3164}
3165
3166static int unknown_module_param_cb(char *param, char *val, const char *modname,
3167				   void *arg)
3168{
3169	struct module *mod = arg;
3170	int ret;
3171
3172	if (strcmp(param, "async_probe") == 0) {
3173		if (kstrtobool(val, &mod->async_probe_requested))
3174			mod->async_probe_requested = true;
3175		return 0;
3176	}
3177
3178	/* Check for magic 'dyndbg' arg */
3179	ret = ddebug_dyndbg_module_param_cb(param, val, modname);
3180	if (ret != 0)
3181		pr_warn("%s: unknown parameter '%s' ignored\n", modname, param);
3182	return 0;
3183}
3184
3185/* Module within temporary copy, this doesn't do any allocation  */
3186static int early_mod_check(struct load_info *info, int flags)
3187{
3188	int err;
3189
3190	/*
3191	 * Now that we know we have the correct module name, check
3192	 * if it's blacklisted.
3193	 */
3194	if (blacklisted(info->name)) {
3195		pr_err("Module %s is blacklisted\n", info->name);
3196		return -EPERM;
3197	}
3198
3199	err = rewrite_section_headers(info, flags);
3200	if (err)
3201		return err;
3202
3203	/* Check module struct version now, before we try to use module. */
3204	if (!check_modstruct_version(info, info->mod))
3205		return -ENOEXEC;
3206
3207	err = check_modinfo(info->mod, info, flags);
3208	if (err)
3209		return err;
3210
3211	mutex_lock(&module_mutex);
3212	err = module_patient_check_exists(info->mod->name, FAIL_DUP_MOD_BECOMING);
3213	mutex_unlock(&module_mutex);
3214
3215	return err;
3216}
3217
3218/*
3219 * Allocate and load the module: note that size of section 0 is always
3220 * zero, and we rely on this for optional sections.
3221 */
3222static int load_module(struct load_info *info, const char __user *uargs,
3223		       int flags)
3224{
3225	struct module *mod;
3226	bool module_allocated = false;
3227	long err = 0;
3228	char *after_dashes;
3229
3230	/*
3231	 * Do the signature check (if any) first. All that
3232	 * the signature check needs is info->len, it does
3233	 * not need any of the section info. That can be
3234	 * set up later. This will minimize the chances
3235	 * of a corrupt module causing problems before
3236	 * we even get to the signature check.
3237	 *
3238	 * The check will also adjust info->len by stripping
3239	 * off the sig length at the end of the module, making
3240	 * checks against info->len more correct.
3241	 */
3242	err = module_sig_check(info, flags);
3243	if (err)
3244		goto free_copy;
3245
3246	/*
3247	 * Do basic sanity checks against the ELF header and
3248	 * sections. Cache useful sections and set the
3249	 * info->mod to the userspace passed struct module.
3250	 */
3251	err = elf_validity_cache_copy(info, flags);
3252	if (err)
3253		goto free_copy;
3254
3255	err = early_mod_check(info, flags);
3256	if (err)
3257		goto free_copy;
3258
3259	/* Figure out module layout, and allocate all the memory. */
3260	mod = layout_and_allocate(info, flags);
3261	if (IS_ERR(mod)) {
3262		err = PTR_ERR(mod);
3263		goto free_copy;
3264	}
3265
3266	module_allocated = true;
3267
3268	audit_log_kern_module(mod->name);
3269
3270	/* Reserve our place in the list. */
3271	err = add_unformed_module(mod);
3272	if (err)
3273		goto free_module;
3274
3275	/*
3276	 * We are tainting your kernel if your module gets into
3277	 * the modules linked list somehow.
3278	 */
3279	module_augment_kernel_taints(mod, info);
3280
3281	/* To avoid stressing percpu allocator, do this once we're unique. */
3282	err = percpu_modalloc(mod, info);
3283	if (err)
3284		goto unlink_mod;
3285
3286	/* Now module is in final location, initialize linked lists, etc. */
3287	err = module_unload_init(mod);
3288	if (err)
3289		goto unlink_mod;
3290
3291	init_param_lock(mod);
3292
3293	/*
3294	 * Now we've got everything in the final locations, we can
3295	 * find optional sections.
3296	 */
3297	err = find_module_sections(mod, info);
3298	if (err)
3299		goto free_unload;
3300
3301	err = check_export_symbol_versions(mod);
3302	if (err)
3303		goto free_unload;
3304
3305	/* Set up MODINFO_ATTR fields */
3306	setup_modinfo(mod, info);
3307
3308	/* Fix up syms, so that st_value is a pointer to location. */
3309	err = simplify_symbols(mod, info);
3310	if (err < 0)
3311		goto free_modinfo;
3312
3313	err = apply_relocations(mod, info);
3314	if (err < 0)
3315		goto free_modinfo;
3316
3317	err = post_relocation(mod, info);
3318	if (err < 0)
3319		goto free_modinfo;
3320
3321	flush_module_icache(mod);
3322
3323	/* Now copy in args */
3324	mod->args = strndup_user(uargs, ~0UL >> 1);
3325	if (IS_ERR(mod->args)) {
3326		err = PTR_ERR(mod->args);
3327		goto free_arch_cleanup;
3328	}
3329
3330	init_build_id(mod, info);
3331
3332	/* Ftrace init must be called in the MODULE_STATE_UNFORMED state */
3333	ftrace_module_init(mod);
3334
3335	/* Finally it's fully formed, ready to start executing. */
3336	err = complete_formation(mod, info);
3337	if (err)
3338		goto ddebug_cleanup;
3339
3340	err = prepare_coming_module(mod);
3341	if (err)
3342		goto bug_cleanup;
3343
3344	mod->async_probe_requested = async_probe;
3345
3346	/* Module is ready to execute: parsing args may do that. */
3347	after_dashes = parse_args(mod->name, mod->args, mod->kp, mod->num_kp,
3348				  -32768, 32767, mod,
3349				  unknown_module_param_cb);
3350	if (IS_ERR(after_dashes)) {
3351		err = PTR_ERR(after_dashes);
3352		goto coming_cleanup;
3353	} else if (after_dashes) {
3354		pr_warn("%s: parameters '%s' after `--' ignored\n",
3355		       mod->name, after_dashes);
3356	}
3357
3358	/* Link in to sysfs. */
3359	err = mod_sysfs_setup(mod, info, mod->kp, mod->num_kp);
3360	if (err < 0)
3361		goto coming_cleanup;
3362
3363	if (is_livepatch_module(mod)) {
3364		err = copy_module_elf(mod, info);
3365		if (err < 0)
3366			goto sysfs_cleanup;
3367	}
3368
3369	/* Get rid of temporary copy. */
3370	free_copy(info, flags);
3371
3372	codetag_load_module(mod);
3373
3374	/* Done! */
3375	trace_module_load(mod);
3376
3377	return do_init_module(mod);
3378
3379 sysfs_cleanup:
3380	mod_sysfs_teardown(mod);
3381 coming_cleanup:
3382	mod->state = MODULE_STATE_GOING;
3383	destroy_params(mod->kp, mod->num_kp);
3384	blocking_notifier_call_chain(&module_notify_list,
3385				     MODULE_STATE_GOING, mod);
3386	klp_module_going(mod);
3387 bug_cleanup:
3388	mod->state = MODULE_STATE_GOING;
3389	/* module_bug_cleanup needs module_mutex protection */
3390	mutex_lock(&module_mutex);
3391	module_bug_cleanup(mod);
3392	mutex_unlock(&module_mutex);
3393
3394 ddebug_cleanup:
3395	ftrace_release_mod(mod);
3396	synchronize_rcu();
3397	kfree(mod->args);
3398 free_arch_cleanup:
3399	module_arch_cleanup(mod);
3400 free_modinfo:
3401	free_modinfo(mod);
3402 free_unload:
3403	module_unload_free(mod);
3404 unlink_mod:
3405	mutex_lock(&module_mutex);
3406	/* Unlink carefully: kallsyms could be walking list. */
3407	list_del_rcu(&mod->list);
3408	mod_tree_remove(mod);
3409	wake_up_all(&module_wq);
3410	/* Wait for RCU-sched synchronizing before releasing mod->list. */
3411	synchronize_rcu();
3412	mutex_unlock(&module_mutex);
3413 free_module:
3414	mod_stat_bump_invalid(info, flags);
3415	/* Free lock-classes; relies on the preceding sync_rcu() */
3416	for_class_mod_mem_type(type, core_data) {
3417		lockdep_free_key_range(mod->mem[type].base,
3418				       mod->mem[type].size);
3419	}
3420
3421	module_deallocate(mod, info);
3422 free_copy:
3423	/*
3424	 * The info->len is always set. We distinguish between
3425	 * failures once the proper module was allocated and
3426	 * before that.
3427	 */
3428	if (!module_allocated)
3429		mod_stat_bump_becoming(info, flags);
3430	free_copy(info, flags);
3431	return err;
3432}
3433
3434SYSCALL_DEFINE3(init_module, void __user *, umod,
3435		unsigned long, len, const char __user *, uargs)
3436{
3437	int err;
3438	struct load_info info = { };
3439
3440	err = may_init_module();
3441	if (err)
3442		return err;
3443
3444	pr_debug("init_module: umod=%p, len=%lu, uargs=%p\n",
3445	       umod, len, uargs);
3446
3447	err = copy_module_from_user(umod, len, &info);
3448	if (err) {
3449		mod_stat_inc(&failed_kreads);
3450		mod_stat_add_long(len, &invalid_kread_bytes);
3451		return err;
3452	}
3453
3454	return load_module(&info, uargs, 0);
3455}
3456
3457struct idempotent {
3458	const void *cookie;
3459	struct hlist_node entry;
3460	struct completion complete;
3461	int ret;
3462};
3463
3464#define IDEM_HASH_BITS 8
3465static struct hlist_head idem_hash[1 << IDEM_HASH_BITS];
3466static DEFINE_SPINLOCK(idem_lock);
3467
3468static bool idempotent(struct idempotent *u, const void *cookie)
3469{
3470	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3471	struct hlist_head *head = idem_hash + hash;
3472	struct idempotent *existing;
3473	bool first;
3474
3475	u->ret = -EINTR;
3476	u->cookie = cookie;
3477	init_completion(&u->complete);
3478
3479	spin_lock(&idem_lock);
3480	first = true;
3481	hlist_for_each_entry(existing, head, entry) {
3482		if (existing->cookie != cookie)
3483			continue;
3484		first = false;
3485		break;
3486	}
3487	hlist_add_head(&u->entry, idem_hash + hash);
3488	spin_unlock(&idem_lock);
3489
3490	return !first;
3491}
3492
3493/*
3494 * We were the first one with 'cookie' on the list, and we ended
3495 * up completing the operation. We now need to walk the list,
3496 * remove everybody - which includes ourselves - fill in the return
3497 * value, and then complete the operation.
3498 */
3499static int idempotent_complete(struct idempotent *u, int ret)
3500{
3501	const void *cookie = u->cookie;
3502	int hash = hash_ptr(cookie, IDEM_HASH_BITS);
3503	struct hlist_head *head = idem_hash + hash;
3504	struct hlist_node *next;
3505	struct idempotent *pos;
3506
3507	spin_lock(&idem_lock);
3508	hlist_for_each_entry_safe(pos, next, head, entry) {
3509		if (pos->cookie != cookie)
3510			continue;
3511		hlist_del_init(&pos->entry);
3512		pos->ret = ret;
3513		complete(&pos->complete);
3514	}
3515	spin_unlock(&idem_lock);
3516	return ret;
3517}
3518
3519/*
3520 * Wait for the idempotent worker.
3521 *
3522 * If we get interrupted, we need to remove ourselves from the
3523 * the idempotent list, and the completion may still come in.
3524 *
3525 * The 'idem_lock' protects against the race, and 'idem.ret' was
3526 * initialized to -EINTR and is thus always the right return
3527 * value even if the idempotent work then completes between
3528 * the wait_for_completion and the cleanup.
3529 */
3530static int idempotent_wait_for_completion(struct idempotent *u)
3531{
3532	if (wait_for_completion_interruptible(&u->complete)) {
3533		spin_lock(&idem_lock);
3534		if (!hlist_unhashed(&u->entry))
3535			hlist_del(&u->entry);
3536		spin_unlock(&idem_lock);
3537	}
3538	return u->ret;
3539}
3540
3541static int init_module_from_file(struct file *f, const char __user * uargs, int flags)
3542{
3543	struct load_info info = { };
3544	void *buf = NULL;
3545	int len;
3546
3547	len = kernel_read_file(f, 0, &buf, INT_MAX, NULL, READING_MODULE);
3548	if (len < 0) {
3549		mod_stat_inc(&failed_kreads);
3550		return len;
3551	}
3552
3553	if (flags & MODULE_INIT_COMPRESSED_FILE) {
3554		int err = module_decompress(&info, buf, len);
3555		vfree(buf); /* compressed data is no longer needed */
3556		if (err) {
3557			mod_stat_inc(&failed_decompress);
3558			mod_stat_add_long(len, &invalid_decompress_bytes);
3559			return err;
3560		}
3561	} else {
3562		info.hdr = buf;
3563		info.len = len;
3564	}
3565
3566	return load_module(&info, uargs, flags);
3567}
3568
3569static int idempotent_init_module(struct file *f, const char __user * uargs, int flags)
3570{
3571	struct idempotent idem;
3572
3573	if (!(f->f_mode & FMODE_READ))
3574		return -EBADF;
3575
3576	/* Are we the winners of the race and get to do this? */
3577	if (!idempotent(&idem, file_inode(f))) {
3578		int ret = init_module_from_file(f, uargs, flags);
3579		return idempotent_complete(&idem, ret);
3580	}
3581
3582	/*
3583	 * Somebody else won the race and is loading the module.
3584	 */
3585	return idempotent_wait_for_completion(&idem);
3586}
3587
3588SYSCALL_DEFINE3(finit_module, int, fd, const char __user *, uargs, int, flags)
3589{
3590	int err = may_init_module();
3591	if (err)
3592		return err;
3593
3594	pr_debug("finit_module: fd=%d, uargs=%p, flags=%i\n", fd, uargs, flags);
3595
3596	if (flags & ~(MODULE_INIT_IGNORE_MODVERSIONS
3597		      |MODULE_INIT_IGNORE_VERMAGIC
3598		      |MODULE_INIT_COMPRESSED_FILE))
3599		return -EINVAL;
3600
3601	CLASS(fd, f)(fd);
3602	if (fd_empty(f))
3603		return -EBADF;
3604	return idempotent_init_module(fd_file(f), uargs, flags);
3605}
3606
3607/* Keep in sync with MODULE_FLAGS_BUF_SIZE !!! */
3608char *module_flags(struct module *mod, char *buf, bool show_state)
3609{
3610	int bx = 0;
3611
3612	BUG_ON(mod->state == MODULE_STATE_UNFORMED);
3613	if (!mod->taints && !show_state)
3614		goto out;
3615	if (mod->taints ||
3616	    mod->state == MODULE_STATE_GOING ||
3617	    mod->state == MODULE_STATE_COMING) {
3618		buf[bx++] = '(';
3619		bx += module_flags_taint(mod->taints, buf + bx);
3620		/* Show a - for module-is-being-unloaded */
3621		if (mod->state == MODULE_STATE_GOING && show_state)
3622			buf[bx++] = '-';
3623		/* Show a + for module-is-being-loaded */
3624		if (mod->state == MODULE_STATE_COMING && show_state)
3625			buf[bx++] = '+';
3626		buf[bx++] = ')';
3627	}
3628out:
3629	buf[bx] = '\0';
3630
3631	return buf;
3632}
3633
3634/* Given an address, look for it in the module exception tables. */
3635const struct exception_table_entry *search_module_extables(unsigned long addr)
3636{
3637	const struct exception_table_entry *e = NULL;
3638	struct module *mod;
3639
3640	preempt_disable();
3641	mod = __module_address(addr);
3642	if (!mod)
3643		goto out;
3644
3645	if (!mod->num_exentries)
3646		goto out;
3647
3648	e = search_extable(mod->extable,
3649			   mod->num_exentries,
3650			   addr);
3651out:
3652	preempt_enable();
3653
3654	/*
3655	 * Now, if we found one, we are running inside it now, hence
3656	 * we cannot unload the module, hence no refcnt needed.
3657	 */
3658	return e;
3659}
3660
3661/**
3662 * is_module_address() - is this address inside a module?
3663 * @addr: the address to check.
3664 *
3665 * See is_module_text_address() if you simply want to see if the address
3666 * is code (not data).
3667 */
3668bool is_module_address(unsigned long addr)
3669{
3670	bool ret;
3671
3672	preempt_disable();
3673	ret = __module_address(addr) != NULL;
3674	preempt_enable();
3675
3676	return ret;
3677}
3678
3679/**
3680 * __module_address() - get the module which contains an address.
3681 * @addr: the address.
3682 *
3683 * Must be called with preempt disabled or module mutex held so that
3684 * module doesn't get freed during this.
3685 */
3686struct module *__module_address(unsigned long addr)
3687{
3688	struct module *mod;
3689
3690	if (addr >= mod_tree.addr_min && addr <= mod_tree.addr_max)
3691		goto lookup;
3692
3693#ifdef CONFIG_ARCH_WANTS_MODULES_DATA_IN_VMALLOC
3694	if (addr >= mod_tree.data_addr_min && addr <= mod_tree.data_addr_max)
3695		goto lookup;
3696#endif
3697
3698	return NULL;
3699
3700lookup:
3701	module_assert_mutex_or_preempt();
3702
3703	mod = mod_find(addr, &mod_tree);
3704	if (mod) {
3705		BUG_ON(!within_module(addr, mod));
3706		if (mod->state == MODULE_STATE_UNFORMED)
3707			mod = NULL;
3708	}
3709	return mod;
3710}
3711
3712/**
3713 * is_module_text_address() - is this address inside module code?
3714 * @addr: the address to check.
3715 *
3716 * See is_module_address() if you simply want to see if the address is
3717 * anywhere in a module.  See kernel_text_address() for testing if an
3718 * address corresponds to kernel or module code.
3719 */
3720bool is_module_text_address(unsigned long addr)
3721{
3722	bool ret;
3723
3724	preempt_disable();
3725	ret = __module_text_address(addr) != NULL;
3726	preempt_enable();
3727
3728	return ret;
3729}
3730
3731/**
3732 * __module_text_address() - get the module whose code contains an address.
3733 * @addr: the address.
3734 *
3735 * Must be called with preempt disabled or module mutex held so that
3736 * module doesn't get freed during this.
3737 */
3738struct module *__module_text_address(unsigned long addr)
3739{
3740	struct module *mod = __module_address(addr);
3741	if (mod) {
3742		/* Make sure it's within the text section. */
3743		if (!within_module_mem_type(addr, mod, MOD_TEXT) &&
3744		    !within_module_mem_type(addr, mod, MOD_INIT_TEXT))
3745			mod = NULL;
3746	}
3747	return mod;
3748}
3749
3750/* Don't grab lock, we're oopsing. */
3751void print_modules(void)
3752{
3753	struct module *mod;
3754	char buf[MODULE_FLAGS_BUF_SIZE];
3755
3756	printk(KERN_DEFAULT "Modules linked in:");
3757	/* Most callers should already have preempt disabled, but make sure */
3758	preempt_disable();
3759	list_for_each_entry_rcu(mod, &modules, list) {
3760		if (mod->state == MODULE_STATE_UNFORMED)
3761			continue;
3762		pr_cont(" %s%s", mod->name, module_flags(mod, buf, true));
3763	}
3764
3765	print_unloaded_tainted_modules();
3766	preempt_enable();
3767	if (last_unloaded_module.name[0])
3768		pr_cont(" [last unloaded: %s%s]", last_unloaded_module.name,
3769			last_unloaded_module.taints);
3770	pr_cont("\n");
3771}
3772
3773#ifdef CONFIG_MODULE_DEBUGFS
3774struct dentry *mod_debugfs_root;
3775
3776static int module_debugfs_init(void)
3777{
3778	mod_debugfs_root = debugfs_create_dir("modules", NULL);
3779	return 0;
3780}
3781module_init(module_debugfs_init);
3782#endif