Root/kernel/module.c

1/*
2   Copyright (C) 2002 Richard Henderson
3   Copyright (C) 2001 Rusty Russell, 2002, 2010 Rusty Russell IBM.
4
5    This program is free software; you can redistribute it and/or modify
6    it under the terms of the GNU General Public License as published by
7    the Free Software Foundation; either version 2 of the License, or
8    (at your option) any later version.
9
10    This program is distributed in the hope that it will be useful,
11    but WITHOUT ANY WARRANTY; without even the implied warranty of
12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13    GNU General Public License for more details.
14
15    You should have received a copy of the GNU General Public License
16    along with this program; if not, write to the Free Software
17    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18*/
19#include <linux/module.h>
20#include <linux/moduleloader.h>
21#include <linux/ftrace_event.h>
22#include <linux/init.h>
23#include <linux/kallsyms.h>
24#include <linux/fs.h>
25#include <linux/sysfs.h>
26#include <linux/kernel.h>
27#include <linux/slab.h>
28#include <linux/vmalloc.h>
29#include <linux/elf.h>
30#include <linux/proc_fs.h>
31#include <linux/seq_file.h>
32#include <linux/syscalls.h>
33#include <linux/fcntl.h>
34#include <linux/rcupdate.h>
35#include <linux/capability.h>
36#include <linux/cpu.h>
37#include <linux/moduleparam.h>
38#include <linux/errno.h>
39#include <linux/err.h>
40#include <linux/vermagic.h>
41#include <linux/notifier.h>
42#include <linux/sched.h>
43#include <linux/stop_machine.h>
44#include <linux/device.h>
45#include <linux/string.h>
46#include <linux/mutex.h>
47#include <linux/rculist.h>
48#include <asm/uaccess.h>
49#include <asm/cacheflush.h>
50#include <asm/mmu_context.h>
51#include <linux/license.h>
52#include <asm/sections.h>
53#include <linux/tracepoint.h>
54#include <linux/ftrace.h>
55#include <linux/async.h>
56#include <linux/percpu.h>
57#include <linux/kmemleak.h>
58#include <linux/jump_label.h>
59#include <linux/pfn.h>
60#include <linux/bsearch.h>
61
62#define CREATE_TRACE_POINTS
63#include <trace/events/module.h>
64
65#if 0
66#define DEBUGP printk
67#else
68#define DEBUGP(fmt , a...)
69#endif
70
71#ifndef ARCH_SHF_SMALL
72#define ARCH_SHF_SMALL 0
73#endif
74
75/*
76 * Modules' sections will be aligned on page boundaries
77 * to ensure complete separation of code and data, but
78 * only when CONFIG_DEBUG_SET_MODULE_RONX=y
79 */
80#ifdef CONFIG_DEBUG_SET_MODULE_RONX
81# define debug_align(X) ALIGN(X, PAGE_SIZE)
82#else
83# define debug_align(X) (X)
84#endif
85
86/*
87 * Given BASE and SIZE this macro calculates the number of pages the
88 * memory regions occupies
89 */
90#define MOD_NUMBER_OF_PAGES(BASE, SIZE) (((SIZE) > 0) ? \
91        (PFN_DOWN((unsigned long)(BASE) + (SIZE) - 1) - \
92             PFN_DOWN((unsigned long)BASE) + 1) \
93        : (0UL))
94
95/* If this is set, the section belongs in the init part of the module */
96#define INIT_OFFSET_MASK (1UL << (BITS_PER_LONG-1))
97
98/*
99 * Mutex protects:
100 * 1) List of modules (also safely readable with preempt_disable),
101 * 2) module_use links,
102 * 3) module_addr_min/module_addr_max.
103 * (delete uses stop_machine/add uses RCU list operations). */
104DEFINE_MUTEX(module_mutex);
105EXPORT_SYMBOL_GPL(module_mutex);
106static LIST_HEAD(modules);
107#ifdef CONFIG_KGDB_KDB
108struct list_head *kdb_modules = &modules; /* kdb needs the list of modules */
109#endif /* CONFIG_KGDB_KDB */
110
111
112/* Block module loading/unloading? */
113int modules_disabled = 0;
114
115/* Waiting for a module to finish initializing? */
116static DECLARE_WAIT_QUEUE_HEAD(module_wq);
117
118static BLOCKING_NOTIFIER_HEAD(module_notify_list);
119
120/* Bounds of module allocation, for speeding __module_address.
121 * Protected by module_mutex. */
122static unsigned long module_addr_min = -1UL, module_addr_max = 0;
123
124int register_module_notifier(struct notifier_block * nb)
125{
126    return blocking_notifier_chain_register(&module_notify_list, nb);
127}
128EXPORT_SYMBOL(register_module_notifier);
129
130int unregister_module_notifier(struct notifier_block * nb)
131{
132    return blocking_notifier_chain_unregister(&module_notify_list, nb);
133}
134EXPORT_SYMBOL(unregister_module_notifier);
135
136struct load_info {
137    Elf_Ehdr *hdr;
138    unsigned long len;
139    Elf_Shdr *sechdrs;
140    char *secstrings, *strtab;
141    unsigned long *strmap;
142    unsigned long symoffs, stroffs;
143    struct _ddebug *debug;
144    unsigned int num_debug;
145    struct {
146        unsigned int sym, str, mod, vers, info, pcpu;
147    } index;
148};
149
150/* We require a truly strong try_module_get(): 0 means failure due to
151   ongoing or failed initialization etc. */
152static inline int strong_try_module_get(struct module *mod)
153{
154    if (mod && mod->state == MODULE_STATE_COMING)
155        return -EBUSY;
156    if (try_module_get(mod))
157        return 0;
158    else
159        return -ENOENT;
160}
161
162static inline void add_taint_module(struct module *mod, unsigned flag)
163{
164    add_taint(flag);
165    mod->taints |= (1U << flag);
166}
167
168/*
169 * A thread that wants to hold a reference to a module only while it
170 * is running can call this to safely exit. nfsd and lockd use this.
171 */
172void __module_put_and_exit(struct module *mod, long code)
173{
174    module_put(mod);
175    do_exit(code);
176}
177EXPORT_SYMBOL(__module_put_and_exit);
178
179/* Find a module section: 0 means not found. */
180static unsigned int find_sec(const struct load_info *info, const char *name)
181{
182    unsigned int i;
183
184    for (i = 1; i < info->hdr->e_shnum; i++) {
185        Elf_Shdr *shdr = &info->sechdrs[i];
186        /* Alloc bit cleared means "ignore it." */
187        if ((shdr->sh_flags & SHF_ALLOC)
188            && strcmp(info->secstrings + shdr->sh_name, name) == 0)
189            return i;
190    }
191    return 0;
192}
193
194/* Find a module section, or NULL. */
195static void *section_addr(const struct load_info *info, const char *name)
196{
197    /* Section 0 has sh_addr 0. */
198    return (void *)info->sechdrs[find_sec(info, name)].sh_addr;
199}
200
201/* Find a module section, or NULL. Fill in number of "objects" in section. */
202static void *section_objs(const struct load_info *info,
203              const char *name,
204              size_t object_size,
205              unsigned int *num)
206{
207    unsigned int sec = find_sec(info, name);
208
209    /* Section 0 has sh_addr 0 and sh_size 0. */
210    *num = info->sechdrs[sec].sh_size / object_size;
211    return (void *)info->sechdrs[sec].sh_addr;
212}
213
214/* Provided by the linker */
215extern const struct kernel_symbol __start___ksymtab[];
216extern const struct kernel_symbol __stop___ksymtab[];
217extern const struct kernel_symbol __start___ksymtab_gpl[];
218extern const struct kernel_symbol __stop___ksymtab_gpl[];
219extern const struct kernel_symbol __start___ksymtab_gpl_future[];
220extern const struct kernel_symbol __stop___ksymtab_gpl_future[];
221extern const unsigned long __start___kcrctab[];
222extern const unsigned long __start___kcrctab_gpl[];
223extern const unsigned long __start___kcrctab_gpl_future[];
224#ifdef CONFIG_UNUSED_SYMBOLS
225extern const struct kernel_symbol __start___ksymtab_unused[];
226extern const struct kernel_symbol __stop___ksymtab_unused[];
227extern const struct kernel_symbol __start___ksymtab_unused_gpl[];
228extern const struct kernel_symbol __stop___ksymtab_unused_gpl[];
229extern const unsigned long __start___kcrctab_unused[];
230extern const unsigned long __start___kcrctab_unused_gpl[];
231#endif
232
233#ifndef CONFIG_MODVERSIONS
234#define symversion(base, idx) NULL
235#else
236#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
237#endif
238
239static bool each_symbol_in_section(const struct symsearch *arr,
240                   unsigned int arrsize,
241                   struct module *owner,
242                   bool (*fn)(const struct symsearch *syms,
243                          struct module *owner,
244                          void *data),
245                   void *data)
246{
247    unsigned int j;
248
249    for (j = 0; j < arrsize; j++) {
250        if (fn(&arr[j], owner, data))
251            return true;
252    }
253
254    return false;
255}
256
257/* Returns true as soon as fn returns true, otherwise false. */
258bool each_symbol_section(bool (*fn)(const struct symsearch *arr,
259                    struct module *owner,
260                    void *data),
261             void *data)
262{
263    struct module *mod;
264    static const struct symsearch arr[] = {
265        { __start___ksymtab, __stop___ksymtab, __start___kcrctab,
266          NOT_GPL_ONLY, false },
267        { __start___ksymtab_gpl, __stop___ksymtab_gpl,
268          __start___kcrctab_gpl,
269          GPL_ONLY, false },
270        { __start___ksymtab_gpl_future, __stop___ksymtab_gpl_future,
271          __start___kcrctab_gpl_future,
272          WILL_BE_GPL_ONLY, false },
273#ifdef CONFIG_UNUSED_SYMBOLS
274        { __start___ksymtab_unused, __stop___ksymtab_unused,
275          __start___kcrctab_unused,
276          NOT_GPL_ONLY, true },
277        { __start___ksymtab_unused_gpl, __stop___ksymtab_unused_gpl,
278          __start___kcrctab_unused_gpl,
279          GPL_ONLY, true },
280#endif
281    };
282
283    if (each_symbol_in_section(arr, ARRAY_SIZE(arr), NULL, fn, data))
284        return true;
285
286    list_for_each_entry_rcu(mod, &modules, list) {
287        struct symsearch arr[] = {
288            { mod->syms, mod->syms + mod->num_syms, mod->crcs,
289              NOT_GPL_ONLY, false },
290            { mod->gpl_syms, mod->gpl_syms + mod->num_gpl_syms,
291              mod->gpl_crcs,
292              GPL_ONLY, false },
293            { mod->gpl_future_syms,
294              mod->gpl_future_syms + mod->num_gpl_future_syms,
295              mod->gpl_future_crcs,
296              WILL_BE_GPL_ONLY, false },
297#ifdef CONFIG_UNUSED_SYMBOLS
298            { mod->unused_syms,
299              mod->unused_syms + mod->num_unused_syms,
300              mod->unused_crcs,
301              NOT_GPL_ONLY, true },
302            { mod->unused_gpl_syms,
303              mod->unused_gpl_syms + mod->num_unused_gpl_syms,
304              mod->unused_gpl_crcs,
305              GPL_ONLY, true },
306#endif
307        };
308
309        if (each_symbol_in_section(arr, ARRAY_SIZE(arr), mod, fn, data))
310            return true;
311    }
312    return false;
313}
314EXPORT_SYMBOL_GPL(each_symbol_section);
315
316struct find_symbol_arg {
317    /* Input */
318    const char *name;
319    bool gplok;
320    bool warn;
321
322    /* Output */
323    struct module *owner;
324    const unsigned long *crc;
325    const struct kernel_symbol *sym;
326};
327
328static bool check_symbol(const struct symsearch *syms,
329                 struct module *owner,
330                 unsigned int symnum, void *data)
331{
332    struct find_symbol_arg *fsa = data;
333
334    if (!fsa->gplok) {
335        if (syms->licence == GPL_ONLY)
336            return false;
337        if (syms->licence == WILL_BE_GPL_ONLY && fsa->warn) {
338            printk(KERN_WARNING "Symbol %s is being used "
339                   "by a non-GPL module, which will not "
340                   "be allowed in the future\n", fsa->name);
341            printk(KERN_WARNING "Please see the file "
342                   "Documentation/feature-removal-schedule.txt "
343                   "in the kernel source tree for more details.\n");
344        }
345    }
346
347#ifdef CONFIG_UNUSED_SYMBOLS
348    if (syms->unused && fsa->warn) {
349        printk(KERN_WARNING "Symbol %s is marked as UNUSED, "
350               "however this module is using it.\n", fsa->name);
351        printk(KERN_WARNING
352               "This symbol will go away in the future.\n");
353        printk(KERN_WARNING
354               "Please evalute if this is the right api to use and if "
355               "it really is, submit a report the linux kernel "
356               "mailinglist together with submitting your code for "
357               "inclusion.\n");
358    }
359#endif
360
361    fsa->owner = owner;
362    fsa->crc = symversion(syms->crcs, symnum);
363    fsa->sym = &syms->start[symnum];
364    return true;
365}
366
367static int cmp_name(const void *va, const void *vb)
368{
369    const char *a;
370    const struct kernel_symbol *b;
371    a = va; b = vb;
372    return strcmp(a, b->name);
373}
374
375static bool find_symbol_in_section(const struct symsearch *syms,
376                   struct module *owner,
377                   void *data)
378{
379    struct find_symbol_arg *fsa = data;
380    struct kernel_symbol *sym;
381
382    sym = bsearch(fsa->name, syms->start, syms->stop - syms->start,
383            sizeof(struct kernel_symbol), cmp_name);
384
385    if (sym != NULL && check_symbol(syms, owner, sym - syms->start, data))
386        return true;
387
388    return false;
389}
390
391/* Find a symbol and return it, along with, (optional) crc and
392 * (optional) module which owns it. Needs preempt disabled or module_mutex. */
393const struct kernel_symbol *find_symbol(const char *name,
394                    struct module **owner,
395                    const unsigned long **crc,
396                    bool gplok,
397                    bool warn)
398{
399    struct find_symbol_arg fsa;
400
401    fsa.name = name;
402    fsa.gplok = gplok;
403    fsa.warn = warn;
404
405    if (each_symbol_section(find_symbol_in_section, &fsa)) {
406        if (owner)
407            *owner = fsa.owner;
408        if (crc)
409            *crc = fsa.crc;
410        return fsa.sym;
411    }
412
413    DEBUGP("Failed to find symbol %s\n", name);
414    return NULL;
415}
416EXPORT_SYMBOL_GPL(find_symbol);
417
418/* Search for module by name: must hold module_mutex. */
419struct module *find_module(const char *name)
420{
421    struct module *mod;
422
423    list_for_each_entry(mod, &modules, list) {
424        if (strcmp(mod->name, name) == 0)
425            return mod;
426    }
427    return NULL;
428}
429EXPORT_SYMBOL_GPL(find_module);
430
431#ifdef CONFIG_SMP
432
433static inline void __percpu *mod_percpu(struct module *mod)
434{
435    return mod->percpu;
436}
437
438static int percpu_modalloc(struct module *mod,
439               unsigned long size, unsigned long align)
440{
441    if (align > PAGE_SIZE) {
442        printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n",
443               mod->name, align, PAGE_SIZE);
444        align = PAGE_SIZE;
445    }
446
447    mod->percpu = __alloc_reserved_percpu(size, align);
448    if (!mod->percpu) {
449        printk(KERN_WARNING
450               "%s: Could not allocate %lu bytes percpu data\n",
451               mod->name, size);
452        return -ENOMEM;
453    }
454    mod->percpu_size = size;
455    return 0;
456}
457
458static void percpu_modfree(struct module *mod)
459{
460    free_percpu(mod->percpu);
461}
462
463static unsigned int find_pcpusec(struct load_info *info)
464{
465    return find_sec(info, ".data..percpu");
466}
467
468static void percpu_modcopy(struct module *mod,
469               const void *from, unsigned long size)
470{
471    int cpu;
472
473    for_each_possible_cpu(cpu)
474        memcpy(per_cpu_ptr(mod->percpu, cpu), from, size);
475}
476
477/**
478 * is_module_percpu_address - test whether address is from module static percpu
479 * @addr: address to test
480 *
481 * Test whether @addr belongs to module static percpu area.
482 *
483 * RETURNS:
484 * %true if @addr is from module static percpu area
485 */
486bool is_module_percpu_address(unsigned long addr)
487{
488    struct module *mod;
489    unsigned int cpu;
490
491    preempt_disable();
492
493    list_for_each_entry_rcu(mod, &modules, list) {
494        if (!mod->percpu_size)
495            continue;
496        for_each_possible_cpu(cpu) {
497            void *start = per_cpu_ptr(mod->percpu, cpu);
498
499            if ((void *)addr >= start &&
500                (void *)addr < start + mod->percpu_size) {
501                preempt_enable();
502                return true;
503            }
504        }
505    }
506
507    preempt_enable();
508    return false;
509}
510
511#else /* ... !CONFIG_SMP */
512
513static inline void __percpu *mod_percpu(struct module *mod)
514{
515    return NULL;
516}
517static inline int percpu_modalloc(struct module *mod,
518                  unsigned long size, unsigned long align)
519{
520    return -ENOMEM;
521}
522static inline void percpu_modfree(struct module *mod)
523{
524}
525static unsigned int find_pcpusec(struct load_info *info)
526{
527    return 0;
528}
529static inline void percpu_modcopy(struct module *mod,
530                  const void *from, unsigned long size)
531{
532    /* pcpusec should be 0, and size of that section should be 0. */
533    BUG_ON(size != 0);
534}
535bool is_module_percpu_address(unsigned long addr)
536{
537    return false;
538}
539
540#endif /* CONFIG_SMP */
541
542#define MODINFO_ATTR(field) \
543static void setup_modinfo_##field(struct module *mod, const char *s) \
544{ \
545    mod->field = kstrdup(s, GFP_KERNEL); \
546} \
547static ssize_t show_modinfo_##field(struct module_attribute *mattr, \
548                    struct module *mod, char *buffer) \
549{ \
550    return sprintf(buffer, "%s\n", mod->field); \
551} \
552static int modinfo_##field##_exists(struct module *mod) \
553{ \
554    return mod->field != NULL; \
555} \
556static void free_modinfo_##field(struct module *mod) \
557{ \
558    kfree(mod->field); \
559    mod->field = NULL; \
560} \
561static struct module_attribute modinfo_##field = { \
562    .attr = { .name = __stringify(field), .mode = 0444 }, \
563    .show = show_modinfo_##field, \
564    .setup = setup_modinfo_##field, \
565    .test = modinfo_##field##_exists, \
566    .free = free_modinfo_##field, \
567};
568
569MODINFO_ATTR(version);
570MODINFO_ATTR(srcversion);
571
572static char last_unloaded_module[MODULE_NAME_LEN+1];
573
574#ifdef CONFIG_MODULE_UNLOAD
575
576EXPORT_TRACEPOINT_SYMBOL(module_get);
577
578/* Init the unload section of the module. */
579static int module_unload_init(struct module *mod)
580{
581    mod->refptr = alloc_percpu(struct module_ref);
582    if (!mod->refptr)
583        return -ENOMEM;
584
585    INIT_LIST_HEAD(&mod->source_list);
586    INIT_LIST_HEAD(&mod->target_list);
587
588    /* Hold reference count during initialization. */
589    __this_cpu_write(mod->refptr->incs, 1);
590    /* Backwards compatibility macros put refcount during init. */
591    mod->waiter = current;
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            DEBUGP("%s uses %s!\n", a->name, b->name);
604            return 1;
605        }
606    }
607    DEBUGP("%s does not use %s!\n", a->name, b->name);
608    return 0;
609}
610
611/*
612 * Module a uses b
613 * - we add 'a' as a "source", 'b' as a "target" of module use
614 * - the module_use is added to the list of 'b' sources (so
615 * 'b' can walk the list to see who sourced them), and of 'a'
616 * targets (so 'a' can see what modules it targets).
617 */
618static int add_module_usage(struct module *a, struct module *b)
619{
620    struct module_use *use;
621
622    DEBUGP("Allocating new usage for %s.\n", a->name);
623    use = kmalloc(sizeof(*use), GFP_ATOMIC);
624    if (!use) {
625        printk(KERN_WARNING "%s: out of memory loading\n", a->name);
626        return -ENOMEM;
627    }
628
629    use->source = a;
630    use->target = b;
631    list_add(&use->source_list, &b->source_list);
632    list_add(&use->target_list, &a->target_list);
633    return 0;
634}
635
636/* Module a uses b: caller needs module_mutex() */
637int ref_module(struct module *a, struct module *b)
638{
639    int err;
640
641    if (b == NULL || already_uses(a, b))
642        return 0;
643
644    /* If module isn't available, we fail. */
645    err = strong_try_module_get(b);
646    if (err)
647        return err;
648
649    err = add_module_usage(a, b);
650    if (err) {
651        module_put(b);
652        return err;
653    }
654    return 0;
655}
656EXPORT_SYMBOL_GPL(ref_module);
657
658/* Clear the unload stuff of the module. */
659static void module_unload_free(struct module *mod)
660{
661    struct module_use *use, *tmp;
662
663    mutex_lock(&module_mutex);
664    list_for_each_entry_safe(use, tmp, &mod->target_list, target_list) {
665        struct module *i = use->target;
666        DEBUGP("%s unusing %s\n", mod->name, i->name);
667        module_put(i);
668        list_del(&use->source_list);
669        list_del(&use->target_list);
670        kfree(use);
671    }
672    mutex_unlock(&module_mutex);
673
674    free_percpu(mod->refptr);
675}
676
677#ifdef CONFIG_MODULE_FORCE_UNLOAD
678static inline int try_force_unload(unsigned int flags)
679{
680    int ret = (flags & O_TRUNC);
681    if (ret)
682        add_taint(TAINT_FORCED_RMMOD);
683    return ret;
684}
685#else
686static inline int try_force_unload(unsigned int flags)
687{
688    return 0;
689}
690#endif /* CONFIG_MODULE_FORCE_UNLOAD */
691
692struct stopref
693{
694    struct module *mod;
695    int flags;
696    int *forced;
697};
698
699/* Whole machine is stopped with interrupts off when this runs. */
700static int __try_stop_module(void *_sref)
701{
702    struct stopref *sref = _sref;
703
704    /* If it's not unused, quit unless we're forcing. */
705    if (module_refcount(sref->mod) != 0) {
706        if (!(*sref->forced = try_force_unload(sref->flags)))
707            return -EWOULDBLOCK;
708    }
709
710    /* Mark it as dying. */
711    sref->mod->state = MODULE_STATE_GOING;
712    return 0;
713}
714
715static int try_stop_module(struct module *mod, int flags, int *forced)
716{
717    if (flags & O_NONBLOCK) {
718        struct stopref sref = { mod, flags, forced };
719
720        return stop_machine(__try_stop_module, &sref, NULL);
721    } else {
722        /* We don't need to stop the machine for this. */
723        mod->state = MODULE_STATE_GOING;
724        synchronize_sched();
725        return 0;
726    }
727}
728
729unsigned int module_refcount(struct module *mod)
730{
731    unsigned int incs = 0, decs = 0;
732    int cpu;
733
734    for_each_possible_cpu(cpu)
735        decs += per_cpu_ptr(mod->refptr, cpu)->decs;
736    /*
737     * ensure the incs are added up after the decs.
738     * module_put ensures incs are visible before decs with smp_wmb.
739     *
740     * This 2-count scheme avoids the situation where the refcount
741     * for CPU0 is read, then CPU0 increments the module refcount,
742     * then CPU1 drops that refcount, then the refcount for CPU1 is
743     * read. We would record a decrement but not its corresponding
744     * increment so we would see a low count (disaster).
745     *
746     * Rare situation? But module_refcount can be preempted, and we
747     * might be tallying up 4096+ CPUs. So it is not impossible.
748     */
749    smp_rmb();
750    for_each_possible_cpu(cpu)
751        incs += per_cpu_ptr(mod->refptr, cpu)->incs;
752    return incs - decs;
753}
754EXPORT_SYMBOL(module_refcount);
755
756/* This exists whether we can unload or not */
757static void free_module(struct module *mod);
758
759static void wait_for_zero_refcount(struct module *mod)
760{
761    /* Since we might sleep for some time, release the mutex first */
762    mutex_unlock(&module_mutex);
763    for (;;) {
764        DEBUGP("Looking at refcount...\n");
765        set_current_state(TASK_UNINTERRUPTIBLE);
766        if (module_refcount(mod) == 0)
767            break;
768        schedule();
769    }
770    current->state = TASK_RUNNING;
771    mutex_lock(&module_mutex);
772}
773
774SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
775        unsigned int, flags)
776{
777    struct module *mod;
778    char name[MODULE_NAME_LEN];
779    int ret, forced = 0;
780
781    if (!capable(CAP_SYS_MODULE) || modules_disabled)
782        return -EPERM;
783
784    if (strncpy_from_user(name, name_user, MODULE_NAME_LEN-1) < 0)
785        return -EFAULT;
786    name[MODULE_NAME_LEN-1] = '\0';
787
788    if (mutex_lock_interruptible(&module_mutex) != 0)
789        return -EINTR;
790
791    mod = find_module(name);
792    if (!mod) {
793        ret = -ENOENT;
794        goto out;
795    }
796
797    if (!list_empty(&mod->source_list)) {
798        /* Other modules depend on us: get rid of them first. */
799        ret = -EWOULDBLOCK;
800        goto out;
801    }
802
803    /* Doing init or already dying? */
804    if (mod->state != MODULE_STATE_LIVE) {
805        /* FIXME: if (force), slam module count and wake up
806                   waiter --RR */
807        DEBUGP("%s already dying\n", mod->name);
808        ret = -EBUSY;
809        goto out;
810    }
811
812    /* If it has an init func, it must have an exit func to unload */
813    if (mod->init && !mod->exit) {
814        forced = try_force_unload(flags);
815        if (!forced) {
816            /* This module can't be removed */
817            ret = -EBUSY;
818            goto out;
819        }
820    }
821
822    /* Set this up before setting mod->state */
823    mod->waiter = current;
824
825    /* Stop the machine so refcounts can't move and disable module. */
826    ret = try_stop_module(mod, flags, &forced);
827    if (ret != 0)
828        goto out;
829
830    /* Never wait if forced. */
831    if (!forced && module_refcount(mod) != 0)
832        wait_for_zero_refcount(mod);
833
834    mutex_unlock(&module_mutex);
835    /* Final destruction now no one is using it. */
836    if (mod->exit != NULL)
837        mod->exit();
838    blocking_notifier_call_chain(&module_notify_list,
839                     MODULE_STATE_GOING, mod);
840    async_synchronize_full();
841
842    /* Store the name of the last unloaded module for diagnostic purposes */
843    strlcpy(last_unloaded_module, mod->name, sizeof(last_unloaded_module));
844
845    free_module(mod);
846    return 0;
847out:
848    mutex_unlock(&module_mutex);
849    return ret;
850}
851
852static inline void print_unload_info(struct seq_file *m, struct module *mod)
853{
854    struct module_use *use;
855    int printed_something = 0;
856
857    seq_printf(m, " %u ", module_refcount(mod));
858
859    /* Always include a trailing , so userspace can differentiate
860           between this and the old multi-field proc format. */
861    list_for_each_entry(use, &mod->source_list, source_list) {
862        printed_something = 1;
863        seq_printf(m, "%s,", use->source->name);
864    }
865
866    if (mod->init != NULL && mod->exit == NULL) {
867        printed_something = 1;
868        seq_printf(m, "[permanent],");
869    }
870
871    if (!printed_something)
872        seq_printf(m, "-");
873}
874
875void __symbol_put(const char *symbol)
876{
877    struct module *owner;
878
879    preempt_disable();
880    if (!find_symbol(symbol, &owner, NULL, true, false))
881        BUG();
882    module_put(owner);
883    preempt_enable();
884}
885EXPORT_SYMBOL(__symbol_put);
886
887/* Note this assumes addr is a function, which it currently always is. */
888void symbol_put_addr(void *addr)
889{
890    struct module *modaddr;
891    unsigned long a = (unsigned long)dereference_function_descriptor(addr);
892
893    if (core_kernel_text(a))
894        return;
895
896    /* module_text_address is safe here: we're supposed to have reference
897     * to module from symbol_get, so it can't go away. */
898    modaddr = __module_text_address(a);
899    BUG_ON(!modaddr);
900    module_put(modaddr);
901}
902EXPORT_SYMBOL_GPL(symbol_put_addr);
903
904static ssize_t show_refcnt(struct module_attribute *mattr,
905               struct module *mod, char *buffer)
906{
907    return sprintf(buffer, "%u\n", module_refcount(mod));
908}
909
910static struct module_attribute refcnt = {
911    .attr = { .name = "refcnt", .mode = 0444 },
912    .show = show_refcnt,
913};
914
915void module_put(struct module *module)
916{
917    if (module) {
918        preempt_disable();
919        smp_wmb(); /* see comment in module_refcount */
920        __this_cpu_inc(module->refptr->decs);
921
922        trace_module_put(module, _RET_IP_);
923        /* Maybe they're waiting for us to drop reference? */
924        if (unlikely(!module_is_live(module)))
925            wake_up_process(module->waiter);
926        preempt_enable();
927    }
928}
929EXPORT_SYMBOL(module_put);
930
931#else /* !CONFIG_MODULE_UNLOAD */
932static inline void print_unload_info(struct seq_file *m, struct module *mod)
933{
934    /* We don't know the usage count, or what modules are using. */
935    seq_printf(m, " - -");
936}
937
938static inline void module_unload_free(struct module *mod)
939{
940}
941
942int ref_module(struct module *a, struct module *b)
943{
944    return strong_try_module_get(b);
945}
946EXPORT_SYMBOL_GPL(ref_module);
947
948static inline int module_unload_init(struct module *mod)
949{
950    return 0;
951}
952#endif /* CONFIG_MODULE_UNLOAD */
953
954static ssize_t show_initstate(struct module_attribute *mattr,
955               struct module *mod, char *buffer)
956{
957    const char *state = "unknown";
958
959    switch (mod->state) {
960    case MODULE_STATE_LIVE:
961        state = "live";
962        break;
963    case MODULE_STATE_COMING:
964        state = "coming";
965        break;
966    case MODULE_STATE_GOING:
967        state = "going";
968        break;
969    }
970    return sprintf(buffer, "%s\n", state);
971}
972
973static struct module_attribute initstate = {
974    .attr = { .name = "initstate", .mode = 0444 },
975    .show = show_initstate,
976};
977
978static struct module_attribute *modinfo_attrs[] = {
979    &modinfo_version,
980    &modinfo_srcversion,
981    &initstate,
982#ifdef CONFIG_MODULE_UNLOAD
983    &refcnt,
984#endif
985    NULL,
986};
987
988static const char vermagic[] = VERMAGIC_STRING;
989
990static int try_to_force_load(struct module *mod, const char *reason)
991{
992#ifdef CONFIG_MODULE_FORCE_LOAD
993    if (!test_taint(TAINT_FORCED_MODULE))
994        printk(KERN_WARNING "%s: %s: kernel tainted.\n",
995               mod->name, reason);
996    add_taint_module(mod, TAINT_FORCED_MODULE);
997    return 0;
998#else
999    return -ENOEXEC;
1000#endif
1001}
1002
1003#ifdef CONFIG_MODVERSIONS
1004/* If the arch applies (non-zero) relocations to kernel kcrctab, unapply it. */
1005static unsigned long maybe_relocated(unsigned long crc,
1006                     const struct module *crc_owner)
1007{
1008#ifdef ARCH_RELOCATES_KCRCTAB
1009    if (crc_owner == NULL)
1010        return crc - (unsigned long)reloc_start;
1011#endif
1012    return crc;
1013}
1014
1015static int check_version(Elf_Shdr *sechdrs,
1016             unsigned int versindex,
1017             const char *symname,
1018             struct module *mod,
1019             const unsigned long *crc,
1020             const struct module *crc_owner)
1021{
1022    unsigned int i, num_versions;
1023    struct modversion_info *versions;
1024
1025    /* Exporting module didn't supply crcs? OK, we're already tainted. */
1026    if (!crc)
1027        return 1;
1028
1029    /* No versions at all? modprobe --force does this. */
1030    if (versindex == 0)
1031        return try_to_force_load(mod, symname) == 0;
1032
1033    versions = (void *) sechdrs[versindex].sh_addr;
1034    num_versions = sechdrs[versindex].sh_size
1035        / sizeof(struct modversion_info);
1036
1037    for (i = 0; i < num_versions; i++) {
1038        if (strcmp(versions[i].name, symname) != 0)
1039            continue;
1040
1041        if (versions[i].crc == maybe_relocated(*crc, crc_owner))
1042            return 1;
1043        DEBUGP("Found checksum %lX vs module %lX\n",
1044               maybe_relocated(*crc, crc_owner), versions[i].crc);
1045        goto bad_version;
1046    }
1047
1048    printk(KERN_WARNING "%s: no symbol version for %s\n",
1049           mod->name, symname);
1050    return 0;
1051
1052bad_version:
1053    printk("%s: disagrees about version of symbol %s\n",
1054           mod->name, symname);
1055    return 0;
1056}
1057
1058static inline int check_modstruct_version(Elf_Shdr *sechdrs,
1059                      unsigned int versindex,
1060                      struct module *mod)
1061{
1062    const unsigned long *crc;
1063
1064    /* Since this should be found in kernel (which can't be removed),
1065     * no locking is necessary. */
1066    if (!find_symbol(MODULE_SYMBOL_PREFIX "module_layout", NULL,
1067             &crc, true, false))
1068        BUG();
1069    return check_version(sechdrs, versindex, "module_layout", mod, crc,
1070                 NULL);
1071}
1072
1073/* First part is kernel version, which we ignore if module has crcs. */
1074static inline int same_magic(const char *amagic, const char *bmagic,
1075                 bool has_crcs)
1076{
1077    if (has_crcs) {
1078        amagic += strcspn(amagic, " ");
1079        bmagic += strcspn(bmagic, " ");
1080    }
1081    return strcmp(amagic, bmagic) == 0;
1082}
1083#else
1084static inline int check_version(Elf_Shdr *sechdrs,
1085                unsigned int versindex,
1086                const char *symname,
1087                struct module *mod,
1088                const unsigned long *crc,
1089                const struct module *crc_owner)
1090{
1091    return 1;
1092}
1093
1094static inline int check_modstruct_version(Elf_Shdr *sechdrs,
1095                      unsigned int versindex,
1096                      struct module *mod)
1097{
1098    return 1;
1099}
1100
1101static inline int same_magic(const char *amagic, const char *bmagic,
1102                 bool has_crcs)
1103{
1104    return strcmp(amagic, bmagic) == 0;
1105}
1106#endif /* CONFIG_MODVERSIONS */
1107
1108/* Resolve a symbol for this module. I.e. if we find one, record usage. */
1109static const struct kernel_symbol *resolve_symbol(struct module *mod,
1110                          const struct load_info *info,
1111                          const char *name,
1112                          char ownername[])
1113{
1114    struct module *owner;
1115    const struct kernel_symbol *sym;
1116    const unsigned long *crc;
1117    int err;
1118
1119    mutex_lock(&module_mutex);
1120    sym = find_symbol(name, &owner, &crc,
1121              !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)), true);
1122    if (!sym)
1123        goto unlock;
1124
1125    if (!check_version(info->sechdrs, info->index.vers, name, mod, crc,
1126               owner)) {
1127        sym = ERR_PTR(-EINVAL);
1128        goto getname;
1129    }
1130
1131    err = ref_module(mod, owner);
1132    if (err) {
1133        sym = ERR_PTR(err);
1134        goto getname;
1135    }
1136
1137getname:
1138    /* We must make copy under the lock if we failed to get ref. */
1139    strncpy(ownername, module_name(owner), MODULE_NAME_LEN);
1140unlock:
1141    mutex_unlock(&module_mutex);
1142    return sym;
1143}
1144
1145static const struct kernel_symbol *
1146resolve_symbol_wait(struct module *mod,
1147            const struct load_info *info,
1148            const char *name)
1149{
1150    const struct kernel_symbol *ksym;
1151    char owner[MODULE_NAME_LEN];
1152
1153    if (wait_event_interruptible_timeout(module_wq,
1154            !IS_ERR(ksym = resolve_symbol(mod, info, name, owner))
1155            || PTR_ERR(ksym) != -EBUSY,
1156                         30 * HZ) <= 0) {
1157        printk(KERN_WARNING "%s: gave up waiting for init of module %s.\n",
1158               mod->name, owner);
1159    }
1160    return ksym;
1161}
1162
1163/*
1164 * /sys/module/foo/sections stuff
1165 * J. Corbet <corbet@lwn.net>
1166 */
1167#ifdef CONFIG_SYSFS
1168
1169#ifdef CONFIG_KALLSYMS
1170static inline bool sect_empty(const Elf_Shdr *sect)
1171{
1172    return !(sect->sh_flags & SHF_ALLOC) || sect->sh_size == 0;
1173}
1174
1175struct module_sect_attr
1176{
1177    struct module_attribute mattr;
1178    char *name;
1179    unsigned long address;
1180};
1181
1182struct module_sect_attrs
1183{
1184    struct attribute_group grp;
1185    unsigned int nsections;
1186    struct module_sect_attr attrs[0];
1187};
1188
1189static ssize_t module_sect_show(struct module_attribute *mattr,
1190                struct module *mod, char *buf)
1191{
1192    struct module_sect_attr *sattr =
1193        container_of(mattr, struct module_sect_attr, mattr);
1194    return sprintf(buf, "0x%pK\n", (void *)sattr->address);
1195}
1196
1197static void free_sect_attrs(struct module_sect_attrs *sect_attrs)
1198{
1199    unsigned int section;
1200
1201    for (section = 0; section < sect_attrs->nsections; section++)
1202        kfree(sect_attrs->attrs[section].name);
1203    kfree(sect_attrs);
1204}
1205
1206static void add_sect_attrs(struct module *mod, const struct load_info *info)
1207{
1208    unsigned int nloaded = 0, i, size[2];
1209    struct module_sect_attrs *sect_attrs;
1210    struct module_sect_attr *sattr;
1211    struct attribute **gattr;
1212
1213    /* Count loaded sections and allocate structures */
1214    for (i = 0; i < info->hdr->e_shnum; i++)
1215        if (!sect_empty(&info->sechdrs[i]))
1216            nloaded++;
1217    size[0] = ALIGN(sizeof(*sect_attrs)
1218            + nloaded * sizeof(sect_attrs->attrs[0]),
1219            sizeof(sect_attrs->grp.attrs[0]));
1220    size[1] = (nloaded + 1) * sizeof(sect_attrs->grp.attrs[0]);
1221    sect_attrs = kzalloc(size[0] + size[1], GFP_KERNEL);
1222    if (sect_attrs == NULL)
1223        return;
1224
1225    /* Setup section attributes. */
1226    sect_attrs->grp.name = "sections";
1227    sect_attrs->grp.attrs = (void *)sect_attrs + size[0];
1228
1229    sect_attrs->nsections = 0;
1230    sattr = &sect_attrs->attrs[0];
1231    gattr = &sect_attrs->grp.attrs[0];
1232    for (i = 0; i < info->hdr->e_shnum; i++) {
1233        Elf_Shdr *sec = &info->sechdrs[i];
1234        if (sect_empty(sec))
1235            continue;
1236        sattr->address = sec->sh_addr;
1237        sattr->name = kstrdup(info->secstrings + sec->sh_name,
1238                    GFP_KERNEL);
1239        if (sattr->name == NULL)
1240            goto out;
1241        sect_attrs->nsections++;
1242        sysfs_attr_init(&sattr->mattr.attr);
1243        sattr->mattr.show = module_sect_show;
1244        sattr->mattr.store = NULL;
1245        sattr->mattr.attr.name = sattr->name;
1246        sattr->mattr.attr.mode = S_IRUGO;
1247        *(gattr++) = &(sattr++)->mattr.attr;
1248    }
1249    *gattr = NULL;
1250
1251    if (sysfs_create_group(&mod->mkobj.kobj, &sect_attrs->grp))
1252        goto out;
1253
1254    mod->sect_attrs = sect_attrs;
1255    return;
1256  out:
1257    free_sect_attrs(sect_attrs);
1258}
1259
1260static void remove_sect_attrs(struct module *mod)
1261{
1262    if (mod->sect_attrs) {
1263        sysfs_remove_group(&mod->mkobj.kobj,
1264                   &mod->sect_attrs->grp);
1265        /* We are positive that no one is using any sect attrs
1266         * at this point. Deallocate immediately. */
1267        free_sect_attrs(mod->sect_attrs);
1268        mod->sect_attrs = NULL;
1269    }
1270}
1271
1272/*
1273 * /sys/module/foo/notes/.section.name gives contents of SHT_NOTE sections.
1274 */
1275
1276struct module_notes_attrs {
1277    struct kobject *dir;
1278    unsigned int notes;
1279    struct bin_attribute attrs[0];
1280};
1281
1282static ssize_t module_notes_read(struct file *filp, struct kobject *kobj,
1283                 struct bin_attribute *bin_attr,
1284                 char *buf, loff_t pos, size_t count)
1285{
1286    /*
1287     * The caller checked the pos and count against our size.
1288     */
1289    memcpy(buf, bin_attr->private + pos, count);
1290    return count;
1291}
1292
1293static void free_notes_attrs(struct module_notes_attrs *notes_attrs,
1294                 unsigned int i)
1295{
1296    if (notes_attrs->dir) {
1297        while (i-- > 0)
1298            sysfs_remove_bin_file(notes_attrs->dir,
1299                          &notes_attrs->attrs[i]);
1300        kobject_put(notes_attrs->dir);
1301    }
1302    kfree(notes_attrs);
1303}
1304
1305static void add_notes_attrs(struct module *mod, const struct load_info *info)
1306{
1307    unsigned int notes, loaded, i;
1308    struct module_notes_attrs *notes_attrs;
1309    struct bin_attribute *nattr;
1310
1311    /* failed to create section attributes, so can't create notes */
1312    if (!mod->sect_attrs)
1313        return;
1314
1315    /* Count notes sections and allocate structures. */
1316    notes = 0;
1317    for (i = 0; i < info->hdr->e_shnum; i++)
1318        if (!sect_empty(&info->sechdrs[i]) &&
1319            (info->sechdrs[i].sh_type == SHT_NOTE))
1320            ++notes;
1321
1322    if (notes == 0)
1323        return;
1324
1325    notes_attrs = kzalloc(sizeof(*notes_attrs)
1326                  + notes * sizeof(notes_attrs->attrs[0]),
1327                  GFP_KERNEL);
1328    if (notes_attrs == NULL)
1329        return;
1330
1331    notes_attrs->notes = notes;
1332    nattr = &notes_attrs->attrs[0];
1333    for (loaded = i = 0; i < info->hdr->e_shnum; ++i) {
1334        if (sect_empty(&info->sechdrs[i]))
1335            continue;
1336        if (info->sechdrs[i].sh_type == SHT_NOTE) {
1337            sysfs_bin_attr_init(nattr);
1338            nattr->attr.name = mod->sect_attrs->attrs[loaded].name;
1339            nattr->attr.mode = S_IRUGO;
1340            nattr->size = info->sechdrs[i].sh_size;
1341            nattr->private = (void *) info->sechdrs[i].sh_addr;
1342            nattr->read = module_notes_read;
1343            ++nattr;
1344        }
1345        ++loaded;
1346    }
1347
1348    notes_attrs->dir = kobject_create_and_add("notes", &mod->mkobj.kobj);
1349    if (!notes_attrs->dir)
1350        goto out;
1351
1352    for (i = 0; i < notes; ++i)
1353        if (sysfs_create_bin_file(notes_attrs->dir,
1354                      &notes_attrs->attrs[i]))
1355            goto out;
1356
1357    mod->notes_attrs = notes_attrs;
1358    return;
1359
1360  out:
1361    free_notes_attrs(notes_attrs, i);
1362}
1363
1364static void remove_notes_attrs(struct module *mod)
1365{
1366    if (mod->notes_attrs)
1367        free_notes_attrs(mod->notes_attrs, mod->notes_attrs->notes);
1368}
1369
1370#else
1371
1372static inline void add_sect_attrs(struct module *mod,
1373                  const struct load_info *info)
1374{
1375}
1376
1377static inline void remove_sect_attrs(struct module *mod)
1378{
1379}
1380
1381static inline void add_notes_attrs(struct module *mod,
1382                   const struct load_info *info)
1383{
1384}
1385
1386static inline void remove_notes_attrs(struct module *mod)
1387{
1388}
1389#endif /* CONFIG_KALLSYMS */
1390
1391static void add_usage_links(struct module *mod)
1392{
1393#ifdef CONFIG_MODULE_UNLOAD
1394    struct module_use *use;
1395    int nowarn;
1396
1397    mutex_lock(&module_mutex);
1398    list_for_each_entry(use, &mod->target_list, target_list) {
1399        nowarn = sysfs_create_link(use->target->holders_dir,
1400                       &mod->mkobj.kobj, mod->name);
1401    }
1402    mutex_unlock(&module_mutex);
1403#endif
1404}
1405
1406static void del_usage_links(struct module *mod)
1407{
1408#ifdef CONFIG_MODULE_UNLOAD
1409    struct module_use *use;
1410
1411    mutex_lock(&module_mutex);
1412    list_for_each_entry(use, &mod->target_list, target_list)
1413        sysfs_remove_link(use->target->holders_dir, mod->name);
1414    mutex_unlock(&module_mutex);
1415#endif
1416}
1417
1418static int module_add_modinfo_attrs(struct module *mod)
1419{
1420    struct module_attribute *attr;
1421    struct module_attribute *temp_attr;
1422    int error = 0;
1423    int i;
1424
1425    mod->modinfo_attrs = kzalloc((sizeof(struct module_attribute) *
1426                    (ARRAY_SIZE(modinfo_attrs) + 1)),
1427                    GFP_KERNEL);
1428    if (!mod->modinfo_attrs)
1429        return -ENOMEM;
1430
1431    temp_attr = mod->modinfo_attrs;
1432    for (i = 0; (attr = modinfo_attrs[i]) && !error; i++) {
1433        if (!attr->test ||
1434            (attr->test && attr->test(mod))) {
1435            memcpy(temp_attr, attr, sizeof(*temp_attr));
1436            sysfs_attr_init(&temp_attr->attr);
1437            error = sysfs_create_file(&mod->mkobj.kobj,&temp_attr->attr);
1438            ++temp_attr;
1439        }
1440    }
1441    return error;
1442}
1443
1444static void module_remove_modinfo_attrs(struct module *mod)
1445{
1446    struct module_attribute *attr;
1447    int i;
1448
1449    for (i = 0; (attr = &mod->modinfo_attrs[i]); i++) {
1450        /* pick a field to test for end of list */
1451        if (!attr->attr.name)
1452            break;
1453        sysfs_remove_file(&mod->mkobj.kobj,&attr->attr);
1454        if (attr->free)
1455            attr->free(mod);
1456    }
1457    kfree(mod->modinfo_attrs);
1458}
1459
1460static int mod_sysfs_init(struct module *mod)
1461{
1462    int err;
1463    struct kobject *kobj;
1464
1465    if (!module_sysfs_initialized) {
1466        printk(KERN_ERR "%s: module sysfs not initialized\n",
1467               mod->name);
1468        err = -EINVAL;
1469        goto out;
1470    }
1471
1472    kobj = kset_find_obj(module_kset, mod->name);
1473    if (kobj) {
1474        printk(KERN_ERR "%s: module is already loaded\n", mod->name);
1475        kobject_put(kobj);
1476        err = -EINVAL;
1477        goto out;
1478    }
1479
1480    mod->mkobj.mod = mod;
1481
1482    memset(&mod->mkobj.kobj, 0, sizeof(mod->mkobj.kobj));
1483    mod->mkobj.kobj.kset = module_kset;
1484    err = kobject_init_and_add(&mod->mkobj.kobj, &module_ktype, NULL,
1485                   "%s", mod->name);
1486    if (err)
1487        kobject_put(&mod->mkobj.kobj);
1488
1489    /* delay uevent until full sysfs population */
1490out:
1491    return err;
1492}
1493
1494static int mod_sysfs_setup(struct module *mod,
1495               const struct load_info *info,
1496               struct kernel_param *kparam,
1497               unsigned int num_params)
1498{
1499    int err;
1500
1501    err = mod_sysfs_init(mod);
1502    if (err)
1503        goto out;
1504
1505    mod->holders_dir = kobject_create_and_add("holders", &mod->mkobj.kobj);
1506    if (!mod->holders_dir) {
1507        err = -ENOMEM;
1508        goto out_unreg;
1509    }
1510
1511    err = module_param_sysfs_setup(mod, kparam, num_params);
1512    if (err)
1513        goto out_unreg_holders;
1514
1515    err = module_add_modinfo_attrs(mod);
1516    if (err)
1517        goto out_unreg_param;
1518
1519    add_usage_links(mod);
1520    add_sect_attrs(mod, info);
1521    add_notes_attrs(mod, info);
1522
1523    kobject_uevent(&mod->mkobj.kobj, KOBJ_ADD);
1524    return 0;
1525
1526out_unreg_param:
1527    module_param_sysfs_remove(mod);
1528out_unreg_holders:
1529    kobject_put(mod->holders_dir);
1530out_unreg:
1531    kobject_put(&mod->mkobj.kobj);
1532out:
1533    return err;
1534}
1535
1536static void mod_sysfs_fini(struct module *mod)
1537{
1538    remove_notes_attrs(mod);
1539    remove_sect_attrs(mod);
1540    kobject_put(&mod->mkobj.kobj);
1541}
1542
1543#else /* !CONFIG_SYSFS */
1544
1545static int mod_sysfs_setup(struct module *mod,
1546               const struct load_info *info,
1547               struct kernel_param *kparam,
1548               unsigned int num_params)
1549{
1550    return 0;
1551}
1552
1553static void mod_sysfs_fini(struct module *mod)
1554{
1555}
1556
1557static void module_remove_modinfo_attrs(struct module *mod)
1558{
1559}
1560
1561static void del_usage_links(struct module *mod)
1562{
1563}
1564
1565#endif /* CONFIG_SYSFS */
1566
1567static void mod_sysfs_teardown(struct module *mod)
1568{
1569    del_usage_links(mod);
1570    module_remove_modinfo_attrs(mod);
1571    module_param_sysfs_remove(mod);
1572    kobject_put(mod->mkobj.drivers_dir);
1573    kobject_put(mod->holders_dir);
1574    mod_sysfs_fini(mod);
1575}
1576
1577/*
1578 * unlink the module with the whole machine is stopped with interrupts off
1579 * - this defends against kallsyms not taking locks
1580 */
1581static int __unlink_module(void *_mod)
1582{
1583    struct module *mod = _mod;
1584    list_del(&mod->list);
1585    module_bug_cleanup(mod);
1586    return 0;
1587}
1588
1589#ifdef CONFIG_DEBUG_SET_MODULE_RONX
1590/*
1591 * LKM RO/NX protection: protect module's text/ro-data
1592 * from modification and any data from execution.
1593 */
1594void set_page_attributes(void *start, void *end, int (*set)(unsigned long start, int num_pages))
1595{
1596    unsigned long begin_pfn = PFN_DOWN((unsigned long)start);
1597    unsigned long end_pfn = PFN_DOWN((unsigned long)end);
1598
1599    if (end_pfn > begin_pfn)
1600        set(begin_pfn << PAGE_SHIFT, end_pfn - begin_pfn);
1601}
1602
1603static void set_section_ro_nx(void *base,
1604            unsigned long text_size,
1605            unsigned long ro_size,
1606            unsigned long total_size)
1607{
1608    /* begin and end PFNs of the current subsection */
1609    unsigned long begin_pfn;
1610    unsigned long end_pfn;
1611
1612    /*
1613     * Set RO for module text and RO-data:
1614     * - Always protect first page.
1615     * - Do not protect last partial page.
1616     */
1617    if (ro_size > 0)
1618        set_page_attributes(base, base + ro_size, set_memory_ro);
1619
1620    /*
1621     * Set NX permissions for module data:
1622     * - Do not protect first partial page.
1623     * - Always protect last page.
1624     */
1625    if (total_size > text_size) {
1626        begin_pfn = PFN_UP((unsigned long)base + text_size);
1627        end_pfn = PFN_UP((unsigned long)base + total_size);
1628        if (end_pfn > begin_pfn)
1629            set_memory_nx(begin_pfn << PAGE_SHIFT, end_pfn - begin_pfn);
1630    }
1631}
1632
1633static void unset_module_core_ro_nx(struct module *mod)
1634{
1635    set_page_attributes(mod->module_core + mod->core_text_size,
1636        mod->module_core + mod->core_size,
1637        set_memory_x);
1638    set_page_attributes(mod->module_core,
1639        mod->module_core + mod->core_ro_size,
1640        set_memory_rw);
1641}
1642
1643static void unset_module_init_ro_nx(struct module *mod)
1644{
1645    set_page_attributes(mod->module_init + mod->init_text_size,
1646        mod->module_init + mod->init_size,
1647        set_memory_x);
1648    set_page_attributes(mod->module_init,
1649        mod->module_init + mod->init_ro_size,
1650        set_memory_rw);
1651}
1652
1653/* Iterate through all modules and set each module's text as RW */
1654void set_all_modules_text_rw(void)
1655{
1656    struct module *mod;
1657
1658    mutex_lock(&module_mutex);
1659    list_for_each_entry_rcu(mod, &modules, list) {
1660        if ((mod->module_core) && (mod->core_text_size)) {
1661            set_page_attributes(mod->module_core,
1662                        mod->module_core + mod->core_text_size,
1663                        set_memory_rw);
1664        }
1665        if ((mod->module_init) && (mod->init_text_size)) {
1666            set_page_attributes(mod->module_init,
1667                        mod->module_init + mod->init_text_size,
1668                        set_memory_rw);
1669        }
1670    }
1671    mutex_unlock(&module_mutex);
1672}
1673
1674/* Iterate through all modules and set each module's text as RO */
1675void set_all_modules_text_ro(void)
1676{
1677    struct module *mod;
1678
1679    mutex_lock(&module_mutex);
1680    list_for_each_entry_rcu(mod, &modules, list) {
1681        if ((mod->module_core) && (mod->core_text_size)) {
1682            set_page_attributes(mod->module_core,
1683                        mod->module_core + mod->core_text_size,
1684                        set_memory_ro);
1685        }
1686        if ((mod->module_init) && (mod->init_text_size)) {
1687            set_page_attributes(mod->module_init,
1688                        mod->module_init + mod->init_text_size,
1689                        set_memory_ro);
1690        }
1691    }
1692    mutex_unlock(&module_mutex);
1693}
1694#else
1695static inline void set_section_ro_nx(void *base, unsigned long text_size, unsigned long ro_size, unsigned long total_size) { }
1696static void unset_module_core_ro_nx(struct module *mod) { }
1697static void unset_module_init_ro_nx(struct module *mod) { }
1698#endif
1699
1700/* Free a module, remove from lists, etc. */
1701static void free_module(struct module *mod)
1702{
1703    trace_module_free(mod);
1704
1705    /* Delete from various lists */
1706    mutex_lock(&module_mutex);
1707    stop_machine(__unlink_module, mod, NULL);
1708    mutex_unlock(&module_mutex);
1709    mod_sysfs_teardown(mod);
1710
1711    /* Remove dynamic debug info */
1712    ddebug_remove_module(mod->name);
1713
1714    /* Arch-specific cleanup. */
1715    module_arch_cleanup(mod);
1716
1717    /* Module unload stuff */
1718    module_unload_free(mod);
1719
1720    /* Free any allocated parameters. */
1721    destroy_params(mod->kp, mod->num_kp);
1722
1723    /* This may be NULL, but that's OK */
1724    unset_module_init_ro_nx(mod);
1725    module_free(mod, mod->module_init);
1726    kfree(mod->args);
1727    percpu_modfree(mod);
1728
1729    /* Free lock-classes: */
1730    lockdep_free_key_range(mod->module_core, mod->core_size);
1731
1732    /* Finally, free the core (containing the module structure) */
1733    unset_module_core_ro_nx(mod);
1734    module_free(mod, mod->module_core);
1735
1736#ifdef CONFIG_MPU
1737    update_protections(current->mm);
1738#endif
1739}
1740
1741void *__symbol_get(const char *symbol)
1742{
1743    struct module *owner;
1744    const struct kernel_symbol *sym;
1745
1746    preempt_disable();
1747    sym = find_symbol(symbol, &owner, NULL, true, true);
1748    if (sym && strong_try_module_get(owner))
1749        sym = NULL;
1750    preempt_enable();
1751
1752    return sym ? (void *)sym->value : NULL;
1753}
1754EXPORT_SYMBOL_GPL(__symbol_get);
1755
1756/*
1757 * Ensure that an exported symbol [global namespace] does not already exist
1758 * in the kernel or in some other module's exported symbol table.
1759 *
1760 * You must hold the module_mutex.
1761 */
1762static int verify_export_symbols(struct module *mod)
1763{
1764    unsigned int i;
1765    struct module *owner;
1766    const struct kernel_symbol *s;
1767    struct {
1768        const struct kernel_symbol *sym;
1769        unsigned int num;
1770    } arr[] = {
1771        { mod->syms, mod->num_syms },
1772        { mod->gpl_syms, mod->num_gpl_syms },
1773        { mod->gpl_future_syms, mod->num_gpl_future_syms },
1774#ifdef CONFIG_UNUSED_SYMBOLS
1775        { mod->unused_syms, mod->num_unused_syms },
1776        { mod->unused_gpl_syms, mod->num_unused_gpl_syms },
1777#endif
1778    };
1779
1780    for (i = 0; i < ARRAY_SIZE(arr); i++) {
1781        for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1782            if (find_symbol(s->name, &owner, NULL, true, false)) {
1783                printk(KERN_ERR
1784                       "%s: exports duplicate symbol %s"
1785                       " (owned by %s)\n",
1786                       mod->name, s->name, module_name(owner));
1787                return -ENOEXEC;
1788            }
1789        }
1790    }
1791    return 0;
1792}
1793
1794/* Change all symbols so that st_value encodes the pointer directly. */
1795static int simplify_symbols(struct module *mod, const struct load_info *info)
1796{
1797    Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
1798    Elf_Sym *sym = (void *)symsec->sh_addr;
1799    unsigned long secbase;
1800    unsigned int i;
1801    int ret = 0;
1802    const struct kernel_symbol *ksym;
1803
1804    for (i = 1; i < symsec->sh_size / sizeof(Elf_Sym); i++) {
1805        const char *name = info->strtab + sym[i].st_name;
1806
1807        switch (sym[i].st_shndx) {
1808        case SHN_COMMON:
1809            /* We compiled with -fno-common. These are not
1810               supposed to happen. */
1811            DEBUGP("Common symbol: %s\n", name);
1812            printk("%s: please compile with -fno-common\n",
1813                   mod->name);
1814            ret = -ENOEXEC;
1815            break;
1816
1817        case SHN_ABS:
1818            /* Don't need to do anything */
1819            DEBUGP("Absolute symbol: 0x%08lx\n",
1820                   (long)sym[i].st_value);
1821            break;
1822
1823        case SHN_UNDEF:
1824            ksym = resolve_symbol_wait(mod, info, name);
1825            /* Ok if resolved. */
1826            if (ksym && !IS_ERR(ksym)) {
1827                sym[i].st_value = ksym->value;
1828                break;
1829            }
1830
1831            /* Ok if weak. */
1832            if (!ksym && ELF_ST_BIND(sym[i].st_info) == STB_WEAK)
1833                break;
1834
1835            printk(KERN_WARNING "%s: Unknown symbol %s (err %li)\n",
1836                   mod->name, name, PTR_ERR(ksym));
1837            ret = PTR_ERR(ksym) ?: -ENOENT;
1838            break;
1839
1840        default:
1841            /* Divert to percpu allocation if a percpu var. */
1842            if (sym[i].st_shndx == info->index.pcpu)
1843                secbase = (unsigned long)mod_percpu(mod);
1844            else
1845                secbase = info->sechdrs[sym[i].st_shndx].sh_addr;
1846            sym[i].st_value += secbase;
1847            break;
1848        }
1849    }
1850
1851    return ret;
1852}
1853
1854static int apply_relocations(struct module *mod, const struct load_info *info)
1855{
1856    unsigned int i;
1857    int err = 0;
1858
1859    /* Now do relocations. */
1860    for (i = 1; i < info->hdr->e_shnum; i++) {
1861        unsigned int infosec = info->sechdrs[i].sh_info;
1862
1863        /* Not a valid relocation section? */
1864        if (infosec >= info->hdr->e_shnum)
1865            continue;
1866
1867        /* Don't bother with non-allocated sections */
1868        if (!(info->sechdrs[infosec].sh_flags & SHF_ALLOC))
1869            continue;
1870
1871        if (info->sechdrs[i].sh_type == SHT_REL)
1872            err = apply_relocate(info->sechdrs, info->strtab,
1873                         info->index.sym, i, mod);
1874        else if (info->sechdrs[i].sh_type == SHT_RELA)
1875            err = apply_relocate_add(info->sechdrs, info->strtab,
1876                         info->index.sym, i, mod);
1877        if (err < 0)
1878            break;
1879    }
1880    return err;
1881}
1882
1883/* Additional bytes needed by arch in front of individual sections */
1884unsigned int __weak arch_mod_section_prepend(struct module *mod,
1885                         unsigned int section)
1886{
1887    /* default implementation just returns zero */
1888    return 0;
1889}
1890
1891/* Update size with this section: return offset. */
1892static long get_offset(struct module *mod, unsigned int *size,
1893               Elf_Shdr *sechdr, unsigned int section)
1894{
1895    long ret;
1896
1897    *size += arch_mod_section_prepend(mod, section);
1898    ret = ALIGN(*size, sechdr->sh_addralign ?: 1);
1899    *size = ret + sechdr->sh_size;
1900    return ret;
1901}
1902
1903/* Lay out the SHF_ALLOC sections in a way not dissimilar to how ld
1904   might -- code, read-only data, read-write data, small data. Tally
1905   sizes, and place the offsets into sh_entsize fields: high bit means it
1906   belongs in init. */
1907static void layout_sections(struct module *mod, struct load_info *info)
1908{
1909    static unsigned long const masks[][2] = {
1910        /* NOTE: all executable code must be the first section
1911         * in this array; otherwise modify the text_size
1912         * finder in the two loops below */
1913        { SHF_EXECINSTR | SHF_ALLOC, ARCH_SHF_SMALL },
1914        { SHF_ALLOC, SHF_WRITE | ARCH_SHF_SMALL },
1915        { SHF_WRITE | SHF_ALLOC, ARCH_SHF_SMALL },
1916        { ARCH_SHF_SMALL | SHF_ALLOC, 0 }
1917    };
1918    unsigned int m, i;
1919
1920    for (i = 0; i < info->hdr->e_shnum; i++)
1921        info->sechdrs[i].sh_entsize = ~0UL;
1922
1923    DEBUGP("Core section allocation order:\n");
1924    for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1925        for (i = 0; i < info->hdr->e_shnum; ++i) {
1926            Elf_Shdr *s = &info->sechdrs[i];
1927            const char *sname = info->secstrings + s->sh_name;
1928
1929            if ((s->sh_flags & masks[m][0]) != masks[m][0]
1930                || (s->sh_flags & masks[m][1])
1931                || s->sh_entsize != ~0UL
1932                || strstarts(sname, ".init"))
1933                continue;
1934            s->sh_entsize = get_offset(mod, &mod->core_size, s, i);
1935            DEBUGP("\t%s\n", name);
1936        }
1937        switch (m) {
1938        case 0: /* executable */
1939            mod->core_size = debug_align(mod->core_size);
1940            mod->core_text_size = mod->core_size;
1941            break;
1942        case 1: /* RO: text and ro-data */
1943            mod->core_size = debug_align(mod->core_size);
1944            mod->core_ro_size = mod->core_size;
1945            break;
1946        case 3: /* whole core */
1947            mod->core_size = debug_align(mod->core_size);
1948            break;
1949        }
1950    }
1951
1952    DEBUGP("Init section allocation order:\n");
1953    for (m = 0; m < ARRAY_SIZE(masks); ++m) {
1954        for (i = 0; i < info->hdr->e_shnum; ++i) {
1955            Elf_Shdr *s = &info->sechdrs[i];
1956            const char *sname = info->secstrings + s->sh_name;
1957
1958            if ((s->sh_flags & masks[m][0]) != masks[m][0]
1959                || (s->sh_flags & masks[m][1])
1960                || s->sh_entsize != ~0UL
1961                || !strstarts(sname, ".init"))
1962                continue;
1963            s->sh_entsize = (get_offset(mod, &mod->init_size, s, i)
1964                     | INIT_OFFSET_MASK);
1965            DEBUGP("\t%s\n", sname);
1966        }
1967        switch (m) {
1968        case 0: /* executable */
1969            mod->init_size = debug_align(mod->init_size);
1970            mod->init_text_size = mod->init_size;
1971            break;
1972        case 1: /* RO: text and ro-data */
1973            mod->init_size = debug_align(mod->init_size);
1974            mod->init_ro_size = mod->init_size;
1975            break;
1976        case 3: /* whole init */
1977            mod->init_size = debug_align(mod->init_size);
1978            break;
1979        }
1980    }
1981}
1982
1983static void set_license(struct module *mod, const char *license)
1984{
1985    if (!license)
1986        license = "unspecified";
1987
1988    if (!license_is_gpl_compatible(license)) {
1989        if (!test_taint(TAINT_PROPRIETARY_MODULE))
1990            printk(KERN_WARNING "%s: module license '%s' taints "
1991                "kernel.\n", mod->name, license);
1992        add_taint_module(mod, TAINT_PROPRIETARY_MODULE);
1993    }
1994}
1995
1996/* Parse tag=value strings from .modinfo section */
1997static char *next_string(char *string, unsigned long *secsize)
1998{
1999    /* Skip non-zero chars */
2000    while (string[0]) {
2001        string++;
2002        if ((*secsize)-- <= 1)
2003            return NULL;
2004    }
2005
2006    /* Skip any zero padding. */
2007    while (!string[0]) {
2008        string++;
2009        if ((*secsize)-- <= 1)
2010            return NULL;
2011    }
2012    return string;
2013}
2014
2015static char *get_modinfo(struct load_info *info, const char *tag)
2016{
2017    char *p;
2018    unsigned int taglen = strlen(tag);
2019    Elf_Shdr *infosec = &info->sechdrs[info->index.info];
2020    unsigned long size = infosec->sh_size;
2021
2022    for (p = (char *)infosec->sh_addr; p; p = next_string(p, &size)) {
2023        if (strncmp(p, tag, taglen) == 0 && p[taglen] == '=')
2024            return p + taglen + 1;
2025    }
2026    return NULL;
2027}
2028
2029static void setup_modinfo(struct module *mod, struct load_info *info)
2030{
2031    struct module_attribute *attr;
2032    int i;
2033
2034    for (i = 0; (attr = modinfo_attrs[i]); i++) {
2035        if (attr->setup)
2036            attr->setup(mod, get_modinfo(info, attr->attr.name));
2037    }
2038}
2039
2040static void free_modinfo(struct module *mod)
2041{
2042    struct module_attribute *attr;
2043    int i;
2044
2045    for (i = 0; (attr = modinfo_attrs[i]); i++) {
2046        if (attr->free)
2047            attr->free(mod);
2048    }
2049}
2050
2051#ifdef CONFIG_KALLSYMS
2052
2053/* lookup symbol in given range of kernel_symbols */
2054static const struct kernel_symbol *lookup_symbol(const char *name,
2055    const struct kernel_symbol *start,
2056    const struct kernel_symbol *stop)
2057{
2058    return bsearch(name, start, stop - start,
2059            sizeof(struct kernel_symbol), cmp_name);
2060}
2061
2062static int is_exported(const char *name, unsigned long value,
2063               const struct module *mod)
2064{
2065    const struct kernel_symbol *ks;
2066    if (!mod)
2067        ks = lookup_symbol(name, __start___ksymtab, __stop___ksymtab);
2068    else
2069        ks = lookup_symbol(name, mod->syms, mod->syms + mod->num_syms);
2070    return ks != NULL && ks->value == value;
2071}
2072
2073/* As per nm */
2074static char elf_type(const Elf_Sym *sym, const struct load_info *info)
2075{
2076    const Elf_Shdr *sechdrs = info->sechdrs;
2077
2078    if (ELF_ST_BIND(sym->st_info) == STB_WEAK) {
2079        if (ELF_ST_TYPE(sym->st_info) == STT_OBJECT)
2080            return 'v';
2081        else
2082            return 'w';
2083    }
2084    if (sym->st_shndx == SHN_UNDEF)
2085        return 'U';
2086    if (sym->st_shndx == SHN_ABS)
2087        return 'a';
2088    if (sym->st_shndx >= SHN_LORESERVE)
2089        return '?';
2090    if (sechdrs[sym->st_shndx].sh_flags & SHF_EXECINSTR)
2091        return 't';
2092    if (sechdrs[sym->st_shndx].sh_flags & SHF_ALLOC
2093        && sechdrs[sym->st_shndx].sh_type != SHT_NOBITS) {
2094        if (!(sechdrs[sym->st_shndx].sh_flags & SHF_WRITE))
2095            return 'r';
2096        else if (sechdrs[sym->st_shndx].sh_flags & ARCH_SHF_SMALL)
2097            return 'g';
2098        else
2099            return 'd';
2100    }
2101    if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
2102        if (sechdrs[sym->st_shndx].sh_flags & ARCH_SHF_SMALL)
2103            return 's';
2104        else
2105            return 'b';
2106    }
2107    if (strstarts(info->secstrings + sechdrs[sym->st_shndx].sh_name,
2108              ".debug")) {
2109        return 'n';
2110    }
2111    return '?';
2112}
2113
2114static bool is_core_symbol(const Elf_Sym *src, const Elf_Shdr *sechdrs,
2115                           unsigned int shnum)
2116{
2117    const Elf_Shdr *sec;
2118
2119    if (src->st_shndx == SHN_UNDEF
2120        || src->st_shndx >= shnum
2121        || !src->st_name)
2122        return false;
2123
2124    sec = sechdrs + src->st_shndx;
2125    if (!(sec->sh_flags & SHF_ALLOC)
2126#ifndef CONFIG_KALLSYMS_ALL
2127        || !(sec->sh_flags & SHF_EXECINSTR)
2128#endif
2129        || (sec->sh_entsize & INIT_OFFSET_MASK))
2130        return false;
2131
2132    return true;
2133}
2134
2135static void layout_symtab(struct module *mod, struct load_info *info)
2136{
2137    Elf_Shdr *symsect = info->sechdrs + info->index.sym;
2138    Elf_Shdr *strsect = info->sechdrs + info->index.str;
2139    const Elf_Sym *src;
2140    unsigned int i, nsrc, ndst;
2141
2142    /* Put symbol section at end of init part of module. */
2143    symsect->sh_flags |= SHF_ALLOC;
2144    symsect->sh_entsize = get_offset(mod, &mod->init_size, symsect,
2145                     info->index.sym) | INIT_OFFSET_MASK;
2146    DEBUGP("\t%s\n", info->secstrings + symsect->sh_name);
2147
2148    src = (void *)info->hdr + symsect->sh_offset;
2149    nsrc = symsect->sh_size / sizeof(*src);
2150    for (ndst = i = 1; i < nsrc; ++i, ++src)
2151        if (is_core_symbol(src, info->sechdrs, info->hdr->e_shnum)) {
2152            unsigned int j = src->st_name;
2153
2154            while (!__test_and_set_bit(j, info->strmap)
2155                   && info->strtab[j])
2156                ++j;
2157            ++ndst;
2158        }
2159
2160    /* Append room for core symbols at end of core part. */
2161    info->symoffs = ALIGN(mod->core_size, symsect->sh_addralign ?: 1);
2162    mod->core_size = info->symoffs + ndst * sizeof(Elf_Sym);
2163
2164    /* Put string table section at end of init part of module. */
2165    strsect->sh_flags |= SHF_ALLOC;
2166    strsect->sh_entsize = get_offset(mod, &mod->init_size, strsect,
2167                     info->index.str) | INIT_OFFSET_MASK;
2168    DEBUGP("\t%s\n", info->secstrings + strsect->sh_name);
2169
2170    /* Append room for core symbols' strings at end of core part. */
2171    info->stroffs = mod->core_size;
2172    __set_bit(0, info->strmap);
2173    mod->core_size += bitmap_weight(info->strmap, strsect->sh_size);
2174}
2175
2176static void add_kallsyms(struct module *mod, const struct load_info *info)
2177{
2178    unsigned int i, ndst;
2179    const Elf_Sym *src;
2180    Elf_Sym *dst;
2181    char *s;
2182    Elf_Shdr *symsec = &info->sechdrs[info->index.sym];
2183
2184    mod->symtab = (void *)symsec->sh_addr;
2185    mod->num_symtab = symsec->sh_size / sizeof(Elf_Sym);
2186    /* Make sure we get permanent strtab: don't use info->strtab. */
2187    mod->strtab = (void *)info->sechdrs[info->index.str].sh_addr;
2188
2189    /* Set types up while we still have access to sections. */
2190    for (i = 0; i < mod->num_symtab; i++)
2191        mod->symtab[i].st_info = elf_type(&mod->symtab[i], info);
2192
2193    mod->core_symtab = dst = mod->module_core + info->symoffs;
2194    src = mod->symtab;
2195    *dst = *src;
2196    for (ndst = i = 1; i < mod->num_symtab; ++i, ++src) {
2197        if (!is_core_symbol(src, info->sechdrs, info->hdr->e_shnum))
2198            continue;
2199        dst[ndst] = *src;
2200        dst[ndst].st_name = bitmap_weight(info->strmap,
2201                          dst[ndst].st_name);
2202        ++ndst;
2203    }
2204    mod->core_num_syms = ndst;
2205
2206    mod->core_strtab = s = mod->module_core + info->stroffs;
2207    for (*s = 0, i = 1; i < info->sechdrs[info->index.str].sh_size; ++i)
2208        if (test_bit(i, info->strmap))
2209            *++s = mod->strtab[i];
2210}
2211#else
2212static inline void layout_symtab(struct module *mod, struct load_info *info)
2213{
2214}
2215
2216static void add_kallsyms(struct module *mod, const struct load_info *info)
2217{
2218}
2219#endif /* CONFIG_KALLSYMS */
2220
2221static void dynamic_debug_setup(struct _ddebug *debug, unsigned int num)
2222{
2223    if (!debug)
2224        return;
2225#ifdef CONFIG_DYNAMIC_DEBUG
2226    if (ddebug_add_module(debug, num, debug->modname))
2227        printk(KERN_ERR "dynamic debug error adding module: %s\n",
2228                    debug->modname);
2229#endif
2230}
2231
2232static void dynamic_debug_remove(struct _ddebug *debug)
2233{
2234    if (debug)
2235        ddebug_remove_module(debug->modname);
2236}
2237
2238static void *module_alloc_update_bounds(unsigned long size)
2239{
2240    void *ret = module_alloc(size);
2241
2242    if (ret) {
2243        mutex_lock(&module_mutex);
2244        /* Update module bounds. */
2245        if ((unsigned long)ret < module_addr_min)
2246            module_addr_min = (unsigned long)ret;
2247        if ((unsigned long)ret + size > module_addr_max)
2248            module_addr_max = (unsigned long)ret + size;
2249        mutex_unlock(&module_mutex);
2250    }
2251    return ret;
2252}
2253
2254#ifdef CONFIG_DEBUG_KMEMLEAK
2255static void kmemleak_load_module(const struct module *mod,
2256                 const struct load_info *info)
2257{
2258    unsigned int i;
2259
2260    /* only scan the sections containing data */
2261    kmemleak_scan_area(mod, sizeof(struct module), GFP_KERNEL);
2262
2263    for (i = 1; i < info->hdr->e_shnum; i++) {
2264        const char *name = info->secstrings + info->sechdrs[i].sh_name;
2265        if (!(info->sechdrs[i].sh_flags & SHF_ALLOC))
2266            continue;
2267        if (!strstarts(name, ".data") && !strstarts(name, ".bss"))
2268            continue;
2269
2270        kmemleak_scan_area((void *)info->sechdrs[i].sh_addr,
2271                   info->sechdrs[i].sh_size, GFP_KERNEL);
2272    }
2273}
2274#else
2275static inline void kmemleak_load_module(const struct module *mod,
2276                    const struct load_info *info)
2277{
2278}
2279#endif
2280
2281/* Sets info->hdr and info->len. */
2282static int copy_and_check(struct load_info *info,
2283              const void __user *umod, unsigned long len,
2284              const char __user *uargs)
2285{
2286    int err;
2287    Elf_Ehdr *hdr;
2288
2289    if (len < sizeof(*hdr))
2290        return -ENOEXEC;
2291
2292    /* Suck in entire file: we'll want most of it. */
2293    /* vmalloc barfs on "unusual" numbers. Check here */
2294    if (len > 64 * 1024 * 1024 || (hdr = vmalloc(len)) == NULL)
2295        return -ENOMEM;
2296
2297    if (copy_from_user(hdr, umod, len) != 0) {
2298        err = -EFAULT;
2299        goto free_hdr;
2300    }
2301
2302    /* Sanity checks against insmoding binaries or wrong arch,
2303       weird elf version */
2304    if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0
2305        || hdr->e_type != ET_REL
2306        || !elf_check_arch(hdr)
2307        || hdr->e_shentsize != sizeof(Elf_Shdr)) {
2308        err = -ENOEXEC;
2309        goto free_hdr;
2310    }
2311
2312    if (len < hdr->e_shoff + hdr->e_shnum * sizeof(Elf_Shdr)) {
2313        err = -ENOEXEC;
2314        goto free_hdr;
2315    }
2316
2317    info->hdr = hdr;
2318    info->len = len;
2319    return 0;
2320
2321free_hdr:
2322    vfree(hdr);
2323    return err;
2324}
2325
2326static void free_copy(struct load_info *info)
2327{
2328    vfree(info->hdr);
2329}
2330
2331static int rewrite_section_headers(struct load_info *info)
2332{
2333    unsigned int i;
2334
2335    /* This should always be true, but let's be sure. */
2336    info->sechdrs[0].sh_addr = 0;
2337
2338    for (i = 1; i < info->hdr->e_shnum; i++) {
2339        Elf_Shdr *shdr = &info->sechdrs[i];
2340        if (shdr->sh_type != SHT_NOBITS
2341            && info->len < shdr->sh_offset + shdr->sh_size) {
2342            printk(KERN_ERR "Module len %lu truncated\n",
2343                   info->len);
2344            return -ENOEXEC;
2345        }
2346
2347        /* Mark all sections sh_addr with their address in the
2348           temporary image. */
2349        shdr->sh_addr = (size_t)info->hdr + shdr->sh_offset;
2350
2351#ifndef CONFIG_MODULE_UNLOAD
2352        /* Don't load .exit sections */
2353        if (strstarts(info->secstrings+shdr->sh_name, ".exit"))
2354            shdr->sh_flags &= ~(unsigned long)SHF_ALLOC;
2355#endif
2356    }
2357
2358    /* Track but don't keep modinfo and version sections. */
2359    info->index.vers = find_sec(info, "__versions");
2360    info->index.info = find_sec(info, ".modinfo");
2361    info->sechdrs[info->index.info].sh_flags &= ~(unsigned long)SHF_ALLOC;
2362    info->sechdrs[info->index.vers].sh_flags &= ~(unsigned long)SHF_ALLOC;
2363    return 0;
2364}
2365
2366/*
2367 * Set up our basic convenience variables (pointers to section headers,
2368 * search for module section index etc), and do some basic section
2369 * verification.
2370 *
2371 * Return the temporary module pointer (we'll replace it with the final
2372 * one when we move the module sections around).
2373 */
2374static struct module *setup_load_info(struct load_info *info)
2375{
2376    unsigned int i;
2377    int err;
2378    struct module *mod;
2379
2380    /* Set up the convenience variables */
2381    info->sechdrs = (void *)info->hdr + info->hdr->e_shoff;
2382    info->secstrings = (void *)info->hdr
2383        + info->sechdrs[info->hdr->e_shstrndx].sh_offset;
2384
2385    err = rewrite_section_headers(info);
2386    if (err)
2387        return ERR_PTR(err);
2388
2389    /* Find internal symbols and strings. */
2390    for (i = 1; i < info->hdr->e_shnum; i++) {
2391        if (info->sechdrs[i].sh_type == SHT_SYMTAB) {
2392            info->index.sym = i;
2393            info->index.str = info->sechdrs[i].sh_link;
2394            info->strtab = (char *)info->hdr
2395                + info->sechdrs[info->index.str].sh_offset;
2396            break;
2397        }
2398    }
2399
2400    info->index.mod = find_sec(info, ".gnu.linkonce.this_module");
2401    if (!info->index.mod) {
2402        printk(KERN_WARNING "No module found in object\n");
2403        return ERR_PTR(-ENOEXEC);
2404    }
2405    /* This is temporary: point mod into copy of data. */
2406    mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2407
2408    if (info->index.sym == 0) {
2409        printk(KERN_WARNING "%s: module has no symbols (stripped?)\n",
2410               mod->name);
2411        return ERR_PTR(-ENOEXEC);
2412    }
2413
2414    info->index.pcpu = find_pcpusec(info);
2415
2416    /* Check module struct version now, before we try to use module. */
2417    if (!check_modstruct_version(info->sechdrs, info->index.vers, mod))
2418        return ERR_PTR(-ENOEXEC);
2419
2420    return mod;
2421}
2422
2423static int check_modinfo(struct module *mod, struct load_info *info)
2424{
2425    const char *modmagic = get_modinfo(info, "vermagic");
2426    int err;
2427
2428    /* This is allowed: modprobe --force will invalidate it. */
2429    if (!modmagic) {
2430        err = try_to_force_load(mod, "bad vermagic");
2431        if (err)
2432            return err;
2433    } else if (!same_magic(modmagic, vermagic, info->index.vers)) {
2434        printk(KERN_ERR "%s: version magic '%s' should be '%s'\n",
2435               mod->name, modmagic, vermagic);
2436        return -ENOEXEC;
2437    }
2438
2439    if (get_modinfo(info, "staging")) {
2440        add_taint_module(mod, TAINT_CRAP);
2441        printk(KERN_WARNING "%s: module is from the staging directory,"
2442               " the quality is unknown, you have been warned.\n",
2443               mod->name);
2444    }
2445
2446    /* Set up license info based on the info section */
2447    set_license(mod, get_modinfo(info, "license"));
2448
2449    return 0;
2450}
2451
2452static void find_module_sections(struct module *mod, struct load_info *info)
2453{
2454    mod->kp = section_objs(info, "__param",
2455                   sizeof(*mod->kp), &mod->num_kp);
2456    mod->syms = section_objs(info, "__ksymtab",
2457                 sizeof(*mod->syms), &mod->num_syms);
2458    mod->crcs = section_addr(info, "__kcrctab");
2459    mod->gpl_syms = section_objs(info, "__ksymtab_gpl",
2460                     sizeof(*mod->gpl_syms),
2461                     &mod->num_gpl_syms);
2462    mod->gpl_crcs = section_addr(info, "__kcrctab_gpl");
2463    mod->gpl_future_syms = section_objs(info,
2464                        "__ksymtab_gpl_future",
2465                        sizeof(*mod->gpl_future_syms),
2466                        &mod->num_gpl_future_syms);
2467    mod->gpl_future_crcs = section_addr(info, "__kcrctab_gpl_future");
2468
2469#ifdef CONFIG_UNUSED_SYMBOLS
2470    mod->unused_syms = section_objs(info, "__ksymtab_unused",
2471                    sizeof(*mod->unused_syms),
2472                    &mod->num_unused_syms);
2473    mod->unused_crcs = section_addr(info, "__kcrctab_unused");
2474    mod->unused_gpl_syms = section_objs(info, "__ksymtab_unused_gpl",
2475                        sizeof(*mod->unused_gpl_syms),
2476                        &mod->num_unused_gpl_syms);
2477    mod->unused_gpl_crcs = section_addr(info, "__kcrctab_unused_gpl");
2478#endif
2479#ifdef CONFIG_CONSTRUCTORS
2480    mod->ctors = section_objs(info, ".ctors",
2481                  sizeof(*mod->ctors), &mod->num_ctors);
2482#endif
2483
2484#ifdef CONFIG_TRACEPOINTS
2485    mod->tracepoints_ptrs = section_objs(info, "__tracepoints_ptrs",
2486                         sizeof(*mod->tracepoints_ptrs),
2487                         &mod->num_tracepoints);
2488#endif
2489#ifdef HAVE_JUMP_LABEL
2490    mod->jump_entries = section_objs(info, "__jump_table",
2491                    sizeof(*mod->jump_entries),
2492                    &mod->num_jump_entries);
2493#endif
2494#ifdef CONFIG_EVENT_TRACING
2495    mod->trace_events = section_objs(info, "_ftrace_events",
2496                     sizeof(*mod->trace_events),
2497                     &mod->num_trace_events);
2498    /*
2499     * This section contains pointers to allocated objects in the trace
2500     * code and not scanning it leads to false positives.
2501     */
2502    kmemleak_scan_area(mod->trace_events, sizeof(*mod->trace_events) *
2503               mod->num_trace_events, GFP_KERNEL);
2504#endif
2505#ifdef CONFIG_TRACING
2506    mod->trace_bprintk_fmt_start = section_objs(info, "__trace_printk_fmt",
2507                     sizeof(*mod->trace_bprintk_fmt_start),
2508                     &mod->num_trace_bprintk_fmt);
2509    /*
2510     * This section contains pointers to allocated objects in the trace
2511     * code and not scanning it leads to false positives.
2512     */
2513    kmemleak_scan_area(mod->trace_bprintk_fmt_start,
2514               sizeof(*mod->trace_bprintk_fmt_start) *
2515               mod->num_trace_bprintk_fmt, GFP_KERNEL);
2516#endif
2517#ifdef CONFIG_FTRACE_MCOUNT_RECORD
2518    /* sechdrs[0].sh_size is always zero */
2519    mod->ftrace_callsites = section_objs(info, "__mcount_loc",
2520                         sizeof(*mod->ftrace_callsites),
2521                         &mod->num_ftrace_callsites);
2522#endif
2523
2524    mod->extable = section_objs(info, "__ex_table",
2525                    sizeof(*mod->extable), &mod->num_exentries);
2526
2527    if (section_addr(info, "__obsparm"))
2528        printk(KERN_WARNING "%s: Ignoring obsolete parameters\n",
2529               mod->name);
2530
2531    info->debug = section_objs(info, "__verbose",
2532                   sizeof(*info->debug), &info->num_debug);
2533}
2534
2535static int move_module(struct module *mod, struct load_info *info)
2536{
2537    int i;
2538    void *ptr;
2539
2540    /* Do the allocs. */
2541    ptr = module_alloc_update_bounds(mod->core_size);
2542    /*
2543     * The pointer to this block is stored in the module structure
2544     * which is inside the block. Just mark it as not being a
2545     * leak.
2546     */
2547    kmemleak_not_leak(ptr);
2548    if (!ptr)
2549        return -ENOMEM;
2550
2551    memset(ptr, 0, mod->core_size);
2552    mod->module_core = ptr;
2553
2554    ptr = module_alloc_update_bounds(mod->init_size);
2555    /*
2556     * The pointer to this block is stored in the module structure
2557     * which is inside the block. This block doesn't need to be
2558     * scanned as it contains data and code that will be freed
2559     * after the module is initialized.
2560     */
2561    kmemleak_ignore(ptr);
2562    if (!ptr && mod->init_size) {
2563        module_free(mod, mod->module_core);
2564        return -ENOMEM;
2565    }
2566    memset(ptr, 0, mod->init_size);
2567    mod->module_init = ptr;
2568
2569    /* Transfer each section which specifies SHF_ALLOC */
2570    DEBUGP("final section addresses:\n");
2571    for (i = 0; i < info->hdr->e_shnum; i++) {
2572        void *dest;
2573        Elf_Shdr *shdr = &info->sechdrs[i];
2574
2575        if (!(shdr->sh_flags & SHF_ALLOC))
2576            continue;
2577
2578        if (shdr->sh_entsize & INIT_OFFSET_MASK)
2579            dest = mod->module_init
2580                + (shdr->sh_entsize & ~INIT_OFFSET_MASK);
2581        else
2582            dest = mod->module_core + shdr->sh_entsize;
2583
2584        if (shdr->sh_type != SHT_NOBITS)
2585            memcpy(dest, (void *)shdr->sh_addr, shdr->sh_size);
2586        /* Update sh_addr to point to copy in image. */
2587        shdr->sh_addr = (unsigned long)dest;
2588        DEBUGP("\t0x%lx %s\n",
2589               shdr->sh_addr, info->secstrings + shdr->sh_name);
2590    }
2591
2592    return 0;
2593}
2594
2595static int check_module_license_and_versions(struct module *mod)
2596{
2597    /*
2598     * ndiswrapper is under GPL by itself, but loads proprietary modules.
2599     * Don't use add_taint_module(), as it would prevent ndiswrapper from
2600     * using GPL-only symbols it needs.
2601     */
2602    if (strcmp(mod->name, "ndiswrapper") == 0)
2603        add_taint(TAINT_PROPRIETARY_MODULE);
2604
2605    /* driverloader was caught wrongly pretending to be under GPL */
2606    if (strcmp(mod->name, "driverloader") == 0)
2607        add_taint_module(mod, TAINT_PROPRIETARY_MODULE);
2608
2609#ifdef CONFIG_MODVERSIONS
2610    if ((mod->num_syms && !mod->crcs)
2611        || (mod->num_gpl_syms && !mod->gpl_crcs)
2612        || (mod->num_gpl_future_syms && !mod->gpl_future_crcs)
2613#ifdef CONFIG_UNUSED_SYMBOLS
2614        || (mod->num_unused_syms && !mod->unused_crcs)
2615        || (mod->num_unused_gpl_syms && !mod->unused_gpl_crcs)
2616#endif
2617        ) {
2618        return try_to_force_load(mod,
2619                     "no versions for exported symbols");
2620    }
2621#endif
2622    return 0;
2623}
2624
2625static void flush_module_icache(const struct module *mod)
2626{
2627    mm_segment_t old_fs;
2628
2629    /* flush the icache in correct context */
2630    old_fs = get_fs();
2631    set_fs(KERNEL_DS);
2632
2633    /*
2634     * Flush the instruction cache, since we've played with text.
2635     * Do it before processing of module parameters, so the module
2636     * can provide parameter accessor functions of its own.
2637     */
2638    if (mod->module_init)
2639        flush_icache_range((unsigned long)mod->module_init,
2640                   (unsigned long)mod->module_init
2641                   + mod->init_size);
2642    flush_icache_range((unsigned long)mod->module_core,
2643               (unsigned long)mod->module_core + mod->core_size);
2644
2645    set_fs(old_fs);
2646}
2647
2648static struct module *layout_and_allocate(struct load_info *info)
2649{
2650    /* Module within temporary copy. */
2651    struct module *mod;
2652    Elf_Shdr *pcpusec;
2653    int err;
2654
2655    mod = setup_load_info(info);
2656    if (IS_ERR(mod))
2657        return mod;
2658
2659    err = check_modinfo(mod, info);
2660    if (err)
2661        return ERR_PTR(err);
2662
2663    /* Allow arches to frob section contents and sizes. */
2664    err = module_frob_arch_sections(info->hdr, info->sechdrs,
2665                    info->secstrings, mod);
2666    if (err < 0)
2667        goto out;
2668
2669    pcpusec = &info->sechdrs[info->index.pcpu];
2670    if (pcpusec->sh_size) {
2671        /* We have a special allocation for this section. */
2672        err = percpu_modalloc(mod,
2673                      pcpusec->sh_size, pcpusec->sh_addralign);
2674        if (err)
2675            goto out;
2676        pcpusec->sh_flags &= ~(unsigned long)SHF_ALLOC;
2677    }
2678
2679    /* Determine total sizes, and put offsets in sh_entsize. For now
2680       this is done generically; there doesn't appear to be any
2681       special cases for the architectures. */
2682    layout_sections(mod, info);
2683
2684    info->strmap = kzalloc(BITS_TO_LONGS(info->sechdrs[info->index.str].sh_size)
2685             * sizeof(long), GFP_KERNEL);
2686    if (!info->strmap) {
2687        err = -ENOMEM;
2688        goto free_percpu;
2689    }
2690    layout_symtab(mod, info);
2691
2692    /* Allocate and move to the final place */
2693    err = move_module(mod, info);
2694    if (err)
2695        goto free_strmap;
2696
2697    /* Module has been copied to its final place now: return it. */
2698    mod = (void *)info->sechdrs[info->index.mod].sh_addr;
2699    kmemleak_load_module(mod, info);
2700    return mod;
2701
2702free_strmap:
2703    kfree(info->strmap);
2704free_percpu:
2705    percpu_modfree(mod);
2706out:
2707    return ERR_PTR(err);
2708}
2709
2710/* mod is no longer valid after this! */
2711static void module_deallocate(struct module *mod, struct load_info *info)
2712{
2713    kfree(info->strmap);
2714    percpu_modfree(mod);
2715    module_free(mod, mod->module_init);
2716    module_free(mod, mod->module_core);
2717}
2718
2719static int post_relocation(struct module *mod, const struct load_info *info)
2720{
2721    /* Sort exception table now relocations are done. */
2722    sort_extable(mod->extable, mod->extable + mod->num_exentries);
2723
2724    /* Copy relocated percpu area over. */
2725    percpu_modcopy(mod, (void *)info->sechdrs[info->index.pcpu].sh_addr,
2726               info->sechdrs[info->index.pcpu].sh_size);
2727
2728    /* Setup kallsyms-specific fields. */
2729    add_kallsyms(mod, info);
2730
2731    /* Arch-specific module finalizing. */
2732    return module_finalize(info->hdr, info->sechdrs, mod);
2733}
2734
2735/* Allocate and load the module: note that size of section 0 is always
2736   zero, and we rely on this for optional sections. */
2737static struct module *load_module(void __user *umod,
2738                  unsigned long len,
2739                  const char __user *uargs)
2740{
2741    struct load_info info = { NULL, };
2742    struct module *mod;
2743    long err;
2744
2745    DEBUGP("load_module: umod=%p, len=%lu, uargs=%p\n",
2746           umod, len, uargs);
2747
2748    /* Copy in the blobs from userspace, check they are vaguely sane. */
2749    err = copy_and_check(&info, umod, len, uargs);
2750    if (err)
2751        return ERR_PTR(err);
2752
2753    /* Figure out module layout, and allocate all the memory. */
2754    mod = layout_and_allocate(&info);
2755    if (IS_ERR(mod)) {
2756        err = PTR_ERR(mod);
2757        goto free_copy;
2758    }
2759
2760    /* Now module is in final location, initialize linked lists, etc. */
2761    err = module_unload_init(mod);
2762    if (err)
2763        goto free_module;
2764
2765    /* Now we've got everything in the final locations, we can
2766     * find optional sections. */
2767    find_module_sections(mod, &info);
2768
2769    err = check_module_license_and_versions(mod);
2770    if (err)
2771        goto free_unload;
2772
2773    /* Set up MODINFO_ATTR fields */
2774    setup_modinfo(mod, &info);
2775
2776    /* Fix up syms, so that st_value is a pointer to location. */
2777    err = simplify_symbols(mod, &info);
2778    if (err < 0)
2779        goto free_modinfo;
2780
2781    err = apply_relocations(mod, &info);
2782    if (err < 0)
2783        goto free_modinfo;
2784
2785    err = post_relocation(mod, &info);
2786    if (err < 0)
2787        goto free_modinfo;
2788
2789    flush_module_icache(mod);
2790
2791    /* Now copy in args */
2792    mod->args = strndup_user(uargs, ~0UL >> 1);
2793    if (IS_ERR(mod->args)) {
2794        err = PTR_ERR(mod->args);
2795        goto free_arch_cleanup;
2796    }
2797
2798    /* Mark state as coming so strong_try_module_get() ignores us. */
2799    mod->state = MODULE_STATE_COMING;
2800
2801    /* Now sew it into the lists so we can get lockdep and oops
2802     * info during argument parsing. No one should access us, since
2803     * strong_try_module_get() will fail.
2804     * lockdep/oops can run asynchronous, so use the RCU list insertion
2805     * function to insert in a way safe to concurrent readers.
2806     * The mutex protects against concurrent writers.
2807     */
2808    mutex_lock(&module_mutex);
2809    if (find_module(mod->name)) {
2810        err = -EEXIST;
2811        goto unlock;
2812    }
2813
2814    /* This has to be done once we're sure module name is unique. */
2815    if (!mod->taints || mod->taints == (1U<<TAINT_CRAP))
2816        dynamic_debug_setup(info.debug, info.num_debug);
2817
2818    /* Find duplicate symbols */
2819    err = verify_export_symbols(mod);
2820    if (err < 0)
2821        goto ddebug;
2822
2823    module_bug_finalize(info.hdr, info.sechdrs, mod);
2824    list_add_rcu(&mod->list, &modules);
2825    mutex_unlock(&module_mutex);
2826
2827    /* Module is ready to execute: parsing args may do that. */
2828    err = parse_args(mod->name, mod->args, mod->kp, mod->num_kp, NULL);
2829    if (err < 0)
2830        goto unlink;
2831
2832    /* Link in to syfs. */
2833    err = mod_sysfs_setup(mod, &info, mod->kp, mod->num_kp);
2834    if (err < 0)
2835        goto unlink;
2836
2837    /* Get rid of temporary copy and strmap. */
2838    kfree(info.strmap);
2839    free_copy(&info);
2840
2841    /* Done! */
2842    trace_module_load(mod);
2843    return mod;
2844
2845 unlink:
2846    mutex_lock(&module_mutex);
2847    /* Unlink carefully: kallsyms could be walking list. */
2848    list_del_rcu(&mod->list);
2849    module_bug_cleanup(mod);
2850
2851 ddebug:
2852    if (!mod->taints || mod->taints == (1U<<TAINT_CRAP))
2853        dynamic_debug_remove(info.debug);
2854 unlock:
2855    mutex_unlock(&module_mutex);
2856    synchronize_sched();
2857    kfree(mod->args);
2858 free_arch_cleanup:
2859    module_arch_cleanup(mod);
2860 free_modinfo:
2861    free_modinfo(mod);
2862 free_unload:
2863    module_unload_free(mod);
2864 free_module:
2865    module_deallocate(mod, &info);
2866 free_copy:
2867    free_copy(&info);
2868    return ERR_PTR(err);
2869}
2870
2871/* Call module constructors. */
2872static void do_mod_ctors(struct module *mod)
2873{
2874#ifdef CONFIG_CONSTRUCTORS
2875    unsigned long i;
2876
2877    for (i = 0; i < mod->num_ctors; i++)
2878        mod->ctors[i]();
2879#endif
2880}
2881
2882/* This is where the real work happens */
2883SYSCALL_DEFINE3(init_module, void __user *, umod,
2884        unsigned long, len, const char __user *, uargs)
2885{
2886    struct module *mod;
2887    int ret = 0;
2888
2889    /* Must have permission */
2890    if (!capable(CAP_SYS_MODULE) || modules_disabled)
2891        return -EPERM;
2892
2893    /* Do all the hard work */
2894    mod = load_module(umod, len, uargs);
2895    if (IS_ERR(mod))
2896        return PTR_ERR(mod);
2897
2898    blocking_notifier_call_chain(&module_notify_list,
2899            MODULE_STATE_COMING, mod);
2900
2901    /* Set RO and NX regions for core */
2902    set_section_ro_nx(mod->module_core,
2903                mod->core_text_size,
2904                mod->core_ro_size,
2905                mod->core_size);
2906
2907    /* Set RO and NX regions for init */
2908    set_section_ro_nx(mod->module_init,
2909                mod->init_text_size,
2910                mod->init_ro_size,
2911                mod->init_size);
2912
2913    do_mod_ctors(mod);
2914    /* Start the module */
2915    if (mod->init != NULL)
2916        ret = do_one_initcall(mod->init);
2917    if (ret < 0) {
2918        /* Init routine failed: abort. Try to protect us from
2919                   buggy refcounters. */
2920        mod->state = MODULE_STATE_GOING;
2921        synchronize_sched();
2922        module_put(mod);
2923        blocking_notifier_call_chain(&module_notify_list,
2924                         MODULE_STATE_GOING, mod);
2925        free_module(mod);
2926        wake_up(&module_wq);
2927        return ret;
2928    }
2929    if (ret > 0) {
2930        printk(KERN_WARNING
2931"%s: '%s'->init suspiciously returned %d, it should follow 0/-E convention\n"
2932"%s: loading module anyway...\n",
2933               __func__, mod->name, ret,
2934               __func__);
2935        dump_stack();
2936    }
2937
2938    /* Now it's a first class citizen! Wake up anyone waiting for it. */
2939    mod->state = MODULE_STATE_LIVE;
2940    wake_up(&module_wq);
2941    blocking_notifier_call_chain(&module_notify_list,
2942                     MODULE_STATE_LIVE, mod);
2943
2944    /* We need to finish all async code before the module init sequence is done */
2945    async_synchronize_full();
2946
2947    mutex_lock(&module_mutex);
2948    /* Drop initial reference. */
2949    module_put(mod);
2950    trim_init_extable(mod);
2951#ifdef CONFIG_KALLSYMS
2952    mod->num_symtab = mod->core_num_syms;
2953    mod->symtab = mod->core_symtab;
2954    mod->strtab = mod->core_strtab;
2955#endif
2956    unset_module_init_ro_nx(mod);
2957    module_free(mod, mod->module_init);
2958    mod->module_init = NULL;
2959    mod->init_size = 0;
2960    mod->init_ro_size = 0;
2961    mod->init_text_size = 0;
2962    mutex_unlock(&module_mutex);
2963
2964    return 0;
2965}
2966
2967static inline int within(unsigned long addr, void *start, unsigned long size)
2968{
2969    return ((void *)addr >= start && (void *)addr < start + size);
2970}
2971
2972#ifdef CONFIG_KALLSYMS
2973/*
2974 * This ignores the intensely annoying "mapping symbols" found
2975 * in ARM ELF files: $a, $t and $d.
2976 */
2977static inline int is_arm_mapping_symbol(const char *str)
2978{
2979    return str[0] == '$' && strchr("atd", str[1])
2980           && (str[2] == '\0' || str[2] == '.');
2981}
2982
2983static const char *get_ksymbol(struct module *mod,
2984                   unsigned long addr,
2985                   unsigned long *size,
2986                   unsigned long *offset)
2987{
2988    unsigned int i, best = 0;
2989    unsigned long nextval;
2990
2991    /* At worse, next value is at end of module */
2992    if (within_module_init(addr, mod))
2993        nextval = (unsigned long)mod->module_init+mod->init_text_size;
2994    else
2995        nextval = (unsigned long)mod->module_core+mod->core_text_size;
2996
2997    /* Scan for closest preceding symbol, and next symbol. (ELF
2998       starts real symbols at 1). */
2999    for (i = 1; i < mod->num_symtab; i++) {
3000        if (mod->symtab[i].st_shndx == SHN_UNDEF)
3001            continue;
3002
3003        /* We ignore unnamed symbols: they're uninformative
3004         * and inserted at a whim. */
3005        if (mod->symtab[i].st_value <= addr
3006            && mod->symtab[i].st_value > mod->symtab[best].st_value
3007            && *(mod->strtab + mod->symtab[i].st_name) != '\0'
3008            && !is_arm_mapping_symbol(mod->strtab + mod->symtab[i].st_name))
3009            best = i;
3010        if (mod->symtab[i].st_value > addr
3011            && mod->symtab[i].st_value < nextval
3012            && *(mod->strtab + mod->symtab[i].st_name) != '\0'
3013            && !is_arm_mapping_symbol(mod->strtab + mod->symtab[i].st_name))
3014            nextval = mod->symtab[i].st_value;
3015    }
3016
3017    if (!best)
3018        return NULL;
3019
3020    if (size)
3021        *size = nextval - mod->symtab[best].st_value;
3022    if (offset)
3023        *offset = addr - mod->symtab[best].st_value;
3024    return mod->strtab + mod->symtab[best].st_name;
3025}
3026
3027/* For kallsyms to ask for address resolution. NULL means not found. Careful
3028 * not to lock to avoid deadlock on oopses, simply disable preemption. */
3029const char *module_address_lookup(unsigned long addr,
3030                unsigned long *size,
3031                unsigned long *offset,
3032                char **modname,
3033                char *namebuf)
3034{
3035    struct module *mod;
3036    const char *ret = NULL;
3037
3038    preempt_disable();
3039    list_for_each_entry_rcu(mod, &modules, list) {
3040        if (within_module_init(addr, mod) ||
3041            within_module_core(addr, mod)) {
3042            if (modname)
3043                *modname = mod->name;
3044            ret = get_ksymbol(mod, addr, size, offset);
3045            break;
3046        }
3047    }
3048    /* Make a copy in here where it's safe */
3049    if (ret) {
3050        strncpy(namebuf, ret, KSYM_NAME_LEN - 1);
3051        ret = namebuf;
3052    }
3053    preempt_enable();
3054    return ret;
3055}
3056
3057int lookup_module_symbol_name(unsigned long addr, char *symname)
3058{
3059    struct module *mod;
3060
3061    preempt_disable();
3062    list_for_each_entry_rcu(mod, &modules, list) {
3063        if (within_module_init(addr, mod) ||
3064            within_module_core(addr, mod)) {
3065            const char *sym;
3066
3067            sym = get_ksymbol(mod, addr, NULL, NULL);
3068            if (!sym)
3069                goto out;
3070            strlcpy(symname, sym, KSYM_NAME_LEN);
3071            preempt_enable();
3072            return 0;
3073        }
3074    }
3075out:
3076    preempt_enable();
3077    return -ERANGE;
3078}
3079
3080int lookup_module_symbol_attrs(unsigned long addr, unsigned long *size,
3081            unsigned long *offset, char *modname, char *name)
3082{
3083    struct module *mod;
3084
3085    preempt_disable();
3086    list_for_each_entry_rcu(mod, &modules, list) {
3087        if (within_module_init(addr, mod) ||
3088            within_module_core(addr, mod)) {
3089            const char *sym;
3090
3091            sym = get_ksymbol(mod, addr, size, offset);
3092            if (!sym)
3093                goto out;
3094            if (modname)
3095                strlcpy(modname, mod->name, MODULE_NAME_LEN);
3096            if (name)
3097                strlcpy(name, sym, KSYM_NAME_LEN);
3098            preempt_enable();
3099            return 0;
3100        }
3101    }
3102out:
3103    preempt_enable();
3104    return -ERANGE;
3105}
3106
3107int module_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
3108            char *name, char *module_name, int *exported)
3109{
3110    struct module *mod;
3111
3112    preempt_disable();
3113    list_for_each_entry_rcu(mod, &modules, list) {
3114        if (symnum < mod->num_symtab) {
3115            *value = mod->symtab[symnum].st_value;
3116            *type = mod->symtab[symnum].st_info;
3117            strlcpy(name, mod->strtab + mod->symtab[symnum].st_name,
3118                KSYM_NAME_LEN);
3119            strlcpy(module_name, mod->name, MODULE_NAME_LEN);
3120            *exported = is_exported(name, *value, mod);
3121            preempt_enable();
3122            return 0;
3123        }
3124        symnum -= mod->num_symtab;
3125    }
3126    preempt_enable();
3127    return -ERANGE;
3128}
3129
3130static unsigned long mod_find_symname(struct module *mod, const char *name)
3131{
3132    unsigned int i;
3133
3134    for (i = 0; i < mod->num_symtab; i++)
3135        if (strcmp(name, mod->strtab+mod->symtab[i].st_name) == 0 &&
3136            mod->symtab[i].st_info != 'U')
3137            return mod->symtab[i].st_value;
3138    return 0;
3139}
3140
3141/* Look for this name: can be of form module:name. */
3142unsigned long module_kallsyms_lookup_name(const char *name)
3143{
3144    struct module *mod;
3145    char *colon;
3146    unsigned long ret = 0;
3147
3148    /* Don't lock: we're in enough trouble already. */
3149    preempt_disable();
3150    if ((colon = strchr(name, ':')) != NULL) {
3151        *colon = '\0';
3152        if ((mod = find_module(name)) != NULL)
3153            ret = mod_find_symname(mod, colon+1);
3154        *colon = ':';
3155    } else {
3156        list_for_each_entry_rcu(mod, &modules, list)
3157            if ((ret = mod_find_symname(mod, name)) != 0)
3158                break;
3159    }
3160    preempt_enable();
3161    return ret;
3162}
3163
3164int module_kallsyms_on_each_symbol(int (*fn)(void *, const char *,
3165                         struct module *, unsigned long),
3166                   void *data)
3167{
3168    struct module *mod;
3169    unsigned int i;
3170    int ret;
3171
3172    list_for_each_entry(mod, &modules, list) {
3173        for (i = 0; i < mod->num_symtab; i++) {
3174            ret = fn(data, mod->strtab + mod->symtab[i].st_name,
3175                 mod, mod->symtab[i].st_value);
3176            if (ret != 0)
3177                return ret;
3178        }
3179    }
3180    return 0;
3181}
3182#endif /* CONFIG_KALLSYMS */
3183
3184static char *module_flags(struct module *mod, char *buf)
3185{
3186    int bx = 0;
3187
3188    if (mod->taints ||
3189        mod->state == MODULE_STATE_GOING ||
3190        mod->state == MODULE_STATE_COMING) {
3191        buf[bx++] = '(';
3192        if (mod->taints & (1 << TAINT_PROPRIETARY_MODULE))
3193            buf[bx++] = 'P';
3194        if (mod->taints & (1 << TAINT_FORCED_MODULE))
3195            buf[bx++] = 'F';
3196        if (mod->taints & (1 << TAINT_CRAP))
3197            buf[bx++] = 'C';
3198        /*
3199         * TAINT_FORCED_RMMOD: could be added.
3200         * TAINT_UNSAFE_SMP, TAINT_MACHINE_CHECK, TAINT_BAD_PAGE don't
3201         * apply to modules.
3202         */
3203
3204        /* Show a - for module-is-being-unloaded */
3205        if (mod->state == MODULE_STATE_GOING)
3206            buf[bx++] = '-';
3207        /* Show a + for module-is-being-loaded */
3208        if (mod->state == MODULE_STATE_COMING)
3209            buf[bx++] = '+';
3210        buf[bx++] = ')';
3211    }
3212    buf[bx] = '\0';
3213
3214    return buf;
3215}
3216
3217#ifdef CONFIG_PROC_FS
3218/* Called by the /proc file system to return a list of modules. */
3219static void *m_start(struct seq_file *m, loff_t *pos)
3220{
3221    mutex_lock(&module_mutex);
3222    return seq_list_start(&modules, *pos);
3223}
3224
3225static void *m_next(struct seq_file *m, void *p, loff_t *pos)
3226{
3227    return seq_list_next(p, &modules, pos);
3228}
3229
3230static void m_stop(struct seq_file *m, void *p)
3231{
3232    mutex_unlock(&module_mutex);
3233}
3234
3235static int m_show(struct seq_file *m, void *p)
3236{
3237    struct module *mod = list_entry(p, struct module, list);
3238    char buf[8];
3239
3240    seq_printf(m, "%s %u",
3241           mod->name, mod->init_size + mod->core_size);
3242    print_unload_info(m, mod);
3243
3244    /* Informative for users. */
3245    seq_printf(m, " %s",
3246           mod->state == MODULE_STATE_GOING ? "Unloading":
3247           mod->state == MODULE_STATE_COMING ? "Loading":
3248           "Live");
3249    /* Used by oprofile and other similar tools. */
3250    seq_printf(m, " 0x%pK", mod->module_core);
3251
3252    /* Taints info */
3253    if (mod->taints)
3254        seq_printf(m, " %s", module_flags(mod, buf));
3255
3256    seq_printf(m, "\n");
3257    return 0;
3258}
3259
3260/* Format: modulename size refcount deps address
3261
3262   Where refcount is a number or -, and deps is a comma-separated list
3263   of depends or -.
3264*/
3265static const struct seq_operations modules_op = {
3266    .start = m_start,
3267    .next = m_next,
3268    .stop = m_stop,
3269    .show = m_show
3270};
3271
3272static int modules_open(struct inode *inode, struct file *file)
3273{
3274    return seq_open(file, &modules_op);
3275}
3276
3277static const struct file_operations proc_modules_operations = {
3278    .open = modules_open,
3279    .read = seq_read,
3280    .llseek = seq_lseek,
3281    .release = seq_release,
3282};
3283
3284static int __init proc_modules_init(void)
3285{
3286    proc_create("modules", 0, NULL, &proc_modules_operations);
3287    return 0;
3288}
3289module_init(proc_modules_init);
3290#endif
3291
3292/* Given an address, look for it in the module exception tables. */
3293const struct exception_table_entry *search_module_extables(unsigned long addr)
3294{
3295    const struct exception_table_entry *e = NULL;
3296    struct module *mod;
3297
3298    preempt_disable();
3299    list_for_each_entry_rcu(mod, &modules, list) {
3300        if (mod->num_exentries == 0)
3301            continue;
3302
3303        e = search_extable(mod->extable,
3304                   mod->extable + mod->num_exentries - 1,
3305                   addr);
3306        if (e)
3307            break;
3308    }
3309    preempt_enable();
3310
3311    /* Now, if we found one, we are running inside it now, hence
3312       we cannot unload the module, hence no refcnt needed. */
3313    return e;
3314}
3315
3316/*
3317 * is_module_address - is this address inside a module?
3318 * @addr: the address to check.
3319 *
3320 * See is_module_text_address() if you simply want to see if the address
3321 * is code (not data).
3322 */
3323bool is_module_address(unsigned long addr)
3324{
3325    bool ret;
3326
3327    preempt_disable();
3328    ret = __module_address(addr) != NULL;
3329    preempt_enable();
3330
3331    return ret;
3332}
3333
3334/*
3335 * __module_address - get the module which contains an address.
3336 * @addr: the address.
3337 *
3338 * Must be called with preempt disabled or module mutex held so that
3339 * module doesn't get freed during this.
3340 */
3341struct module *__module_address(unsigned long addr)
3342{
3343    struct module *mod;
3344
3345    if (addr < module_addr_min || addr > module_addr_max)
3346        return NULL;
3347
3348    list_for_each_entry_rcu(mod, &modules, list)
3349        if (within_module_core(addr, mod)
3350            || within_module_init(addr, mod))
3351            return mod;
3352    return NULL;
3353}
3354EXPORT_SYMBOL_GPL(__module_address);
3355
3356/*
3357 * is_module_text_address - is this address inside module code?
3358 * @addr: the address to check.
3359 *
3360 * See is_module_address() if you simply want to see if the address is
3361 * anywhere in a module. See kernel_text_address() for testing if an
3362 * address corresponds to kernel or module code.
3363 */
3364bool is_module_text_address(unsigned long addr)
3365{
3366    bool ret;
3367
3368    preempt_disable();
3369    ret = __module_text_address(addr) != NULL;
3370    preempt_enable();
3371
3372    return ret;
3373}
3374
3375/*
3376 * __module_text_address - get the module whose code contains an address.
3377 * @addr: the address.
3378 *
3379 * Must be called with preempt disabled or module mutex held so that
3380 * module doesn't get freed during this.
3381 */
3382struct module *__module_text_address(unsigned long addr)
3383{
3384    struct module *mod = __module_address(addr);
3385    if (mod) {
3386        /* Make sure it's within the text section. */
3387        if (!within(addr, mod->module_init, mod->init_text_size)
3388            && !within(addr, mod->module_core, mod->core_text_size))
3389            mod = NULL;
3390    }
3391    return mod;
3392}
3393EXPORT_SYMBOL_GPL(__module_text_address);
3394
3395/* Don't grab lock, we're oopsing. */
3396void print_modules(void)
3397{
3398    struct module *mod;
3399    char buf[8];
3400
3401    printk(KERN_DEFAULT "Modules linked in:");
3402    /* Most callers should already have preempt disabled, but make sure */
3403    preempt_disable();
3404    list_for_each_entry_rcu(mod, &modules, list)
3405        printk(" %s%s", mod->name, module_flags(mod, buf));
3406    preempt_enable();
3407    if (last_unloaded_module[0])
3408        printk(" [last unloaded: %s]", last_unloaded_module);
3409    printk("\n");
3410}
3411
3412#ifdef CONFIG_MODVERSIONS
3413/* Generate the signature for all relevant module structures here.
3414 * If these change, we don't want to try to parse the module. */
3415void module_layout(struct module *mod,
3416           struct modversion_info *ver,
3417           struct kernel_param *kp,
3418           struct kernel_symbol *ks,
3419           struct tracepoint * const *tp)
3420{
3421}
3422EXPORT_SYMBOL(module_layout);
3423#endif
3424
3425#ifdef CONFIG_TRACEPOINTS
3426void module_update_tracepoints(void)
3427{
3428    struct module *mod;
3429
3430    mutex_lock(&module_mutex);
3431    list_for_each_entry(mod, &modules, list)
3432        if (!mod->taints)
3433            tracepoint_update_probe_range(mod->tracepoints_ptrs,
3434                mod->tracepoints_ptrs + mod->num_tracepoints);
3435    mutex_unlock(&module_mutex);
3436}
3437
3438/*
3439 * Returns 0 if current not found.
3440 * Returns 1 if current found.
3441 */
3442int module_get_iter_tracepoints(struct tracepoint_iter *iter)
3443{
3444    struct module *iter_mod;
3445    int found = 0;
3446
3447    mutex_lock(&module_mutex);
3448    list_for_each_entry(iter_mod, &modules, list) {
3449        if (!iter_mod->taints) {
3450            /*
3451             * Sorted module list
3452             */
3453            if (iter_mod < iter->module)
3454                continue;
3455            else if (iter_mod > iter->module)
3456                iter->tracepoint = NULL;
3457            found = tracepoint_get_iter_range(&iter->tracepoint,
3458                iter_mod->tracepoints_ptrs,
3459                iter_mod->tracepoints_ptrs
3460                    + iter_mod->num_tracepoints);
3461            if (found) {
3462                iter->module = iter_mod;
3463                break;
3464            }
3465        }
3466    }
3467    mutex_unlock(&module_mutex);
3468    return found;
3469}
3470#endif
3471

Archive Download this file



interactive