1 // SPDX-License-Identifier: GPL-2.0-only
3 * kallsyms.c: in-kernel printing of symbolic oopses and stack traces.
5 * Rewritten and vastly simplified by Rusty Russell for in-kernel
7 * Copyright 2002 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
11 * (25/Aug/2004) Paulo Marques <pmarques@grupopie.com>
12 * Changed the compression method from stem compression to "table lookup"
13 * compression (see scripts/kallsyms.c for a more complete description)
15 #include <linux/kallsyms.h>
16 #include <linux/init.h>
17 #include <linux/seq_file.h>
19 #include <linux/kdb.h>
20 #include <linux/err.h>
21 #include <linux/proc_fs.h>
22 #include <linux/sched.h> /* for cond_resched */
23 #include <linux/ctype.h>
24 #include <linux/slab.h>
25 #include <linux/filter.h>
26 #include <linux/ftrace.h>
27 #include <linux/kprobes.h>
28 #include <linux/build_bug.h>
29 #include <linux/compiler.h>
30 #include <linux/module.h>
31 #include <linux/kernel.h>
32 #include <linux/bsearch.h>
33 #include <linux/btf_ids.h>
35 #include "kallsyms_internal.h"
38 * Expand a compressed symbol data into the resulting uncompressed string,
39 * if uncompressed string is too long (>= maxlen), it will be truncated,
40 * given the offset to where the symbol is in the compressed stream.
42 static unsigned int kallsyms_expand_symbol(unsigned int off,
43 char *result, size_t maxlen)
45 int len, skipped_first = 0;
49 /* Get the compressed symbol length from the first symbol byte. */
50 data = &kallsyms_names[off];
55 /* If MSB is 1, it is a "big" symbol, so needs an additional byte. */
56 if ((len & 0x80) != 0) {
57 len = (len & 0x7F) | (*data << 7);
63 * Update the offset to return the offset for the next symbol on
64 * the compressed stream.
69 * For every byte on the compressed symbol data, copy the table
70 * entry for that byte.
73 tptr = &kallsyms_token_table[kallsyms_token_index[*data]];
94 /* Return to offset to the next symbol. */
99 * Get symbol type information. This is encoded as a single char at the
100 * beginning of the symbol name.
102 static char kallsyms_get_symbol_type(unsigned int off)
105 * Get just the first code, look it up in the token table,
106 * and return the first char from this token.
108 return kallsyms_token_table[kallsyms_token_index[kallsyms_names[off + 1]]];
113 * Find the offset on the compressed stream given and index in the
116 static unsigned int get_symbol_offset(unsigned long pos)
122 * Use the closest marker we have. We have markers every 256 positions,
123 * so that should be close enough.
125 name = &kallsyms_names[kallsyms_markers[pos >> 8]];
128 * Sequentially scan all the symbols up to the point we're searching
129 * for. Every symbol is stored in a [<len>][<len> bytes of data] format,
130 * so we just need to add the len to the current pointer for every
131 * symbol we wish to skip.
133 for (i = 0; i < (pos & 0xFF); i++) {
137 * If MSB is 1, it is a "big" symbol, so we need to look into
138 * the next byte (and skip it, too).
140 if ((len & 0x80) != 0)
141 len = ((len & 0x7F) | (name[1] << 7)) + 1;
143 name = name + len + 1;
146 return name - kallsyms_names;
149 static unsigned long kallsyms_sym_address(int idx)
151 if (!IS_ENABLED(CONFIG_KALLSYMS_BASE_RELATIVE))
152 return kallsyms_addresses[idx];
154 /* values are unsigned offsets if --absolute-percpu is not in effect */
155 if (!IS_ENABLED(CONFIG_KALLSYMS_ABSOLUTE_PERCPU))
156 return kallsyms_relative_base + (u32)kallsyms_offsets[idx];
158 /* ...otherwise, positive offsets are absolute values */
159 if (kallsyms_offsets[idx] >= 0)
160 return kallsyms_offsets[idx];
162 /* ...and negative offsets are relative to kallsyms_relative_base - 1 */
163 return kallsyms_relative_base - 1 - kallsyms_offsets[idx];
166 static bool cleanup_symbol_name(char *s)
170 if (!IS_ENABLED(CONFIG_LTO_CLANG))
174 * LLVM appends various suffixes for local functions and variables that
175 * must be promoted to global scope as part of LTO. This can break
176 * hooking of static functions with kprobes. '.' is not a valid
177 * character in an identifier in C. Suffixes observed:
178 * - foo.llvm.[0-9a-f]+
181 res = strchr(s, '.');
190 /* Lookup the address for this symbol. Returns 0 if not found. */
191 unsigned long kallsyms_lookup_name(const char *name)
193 char namebuf[KSYM_NAME_LEN];
197 /* Skip the search for empty string. */
201 for (i = 0, off = 0; i < kallsyms_num_syms; i++) {
202 off = kallsyms_expand_symbol(off, namebuf, ARRAY_SIZE(namebuf));
204 if (strcmp(namebuf, name) == 0)
205 return kallsyms_sym_address(i);
207 if (cleanup_symbol_name(namebuf) && strcmp(namebuf, name) == 0)
208 return kallsyms_sym_address(i);
210 return module_kallsyms_lookup_name(name);
214 * Iterate over all symbols in vmlinux. For symbols from modules use
215 * module_kallsyms_on_each_symbol instead.
217 int kallsyms_on_each_symbol(int (*fn)(void *, const char *, struct module *,
221 char namebuf[KSYM_NAME_LEN];
226 for (i = 0, off = 0; i < kallsyms_num_syms; i++) {
227 off = kallsyms_expand_symbol(off, namebuf, ARRAY_SIZE(namebuf));
228 ret = fn(data, namebuf, NULL, kallsyms_sym_address(i));
236 static unsigned long get_symbol_pos(unsigned long addr,
237 unsigned long *symbolsize,
238 unsigned long *offset)
240 unsigned long symbol_start = 0, symbol_end = 0;
241 unsigned long i, low, high, mid;
243 /* This kernel should never had been booted. */
244 if (!IS_ENABLED(CONFIG_KALLSYMS_BASE_RELATIVE))
245 BUG_ON(!kallsyms_addresses);
247 BUG_ON(!kallsyms_offsets);
249 /* Do a binary search on the sorted kallsyms_addresses array. */
251 high = kallsyms_num_syms;
253 while (high - low > 1) {
254 mid = low + (high - low) / 2;
255 if (kallsyms_sym_address(mid) <= addr)
262 * Search for the first aliased symbol. Aliased
263 * symbols are symbols with the same address.
265 while (low && kallsyms_sym_address(low-1) == kallsyms_sym_address(low))
268 symbol_start = kallsyms_sym_address(low);
270 /* Search for next non-aliased symbol. */
271 for (i = low + 1; i < kallsyms_num_syms; i++) {
272 if (kallsyms_sym_address(i) > symbol_start) {
273 symbol_end = kallsyms_sym_address(i);
278 /* If we found no next symbol, we use the end of the section. */
280 if (is_kernel_inittext(addr))
281 symbol_end = (unsigned long)_einittext;
282 else if (IS_ENABLED(CONFIG_KALLSYMS_ALL))
283 symbol_end = (unsigned long)_end;
285 symbol_end = (unsigned long)_etext;
289 *symbolsize = symbol_end - symbol_start;
291 *offset = addr - symbol_start;
297 * Lookup an address but don't bother to find any names.
299 int kallsyms_lookup_size_offset(unsigned long addr, unsigned long *symbolsize,
300 unsigned long *offset)
302 char namebuf[KSYM_NAME_LEN];
304 if (is_ksym_addr(addr)) {
305 get_symbol_pos(addr, symbolsize, offset);
308 return !!module_address_lookup(addr, symbolsize, offset, NULL, NULL, namebuf) ||
309 !!__bpf_address_lookup(addr, symbolsize, offset, namebuf);
312 static const char *kallsyms_lookup_buildid(unsigned long addr,
313 unsigned long *symbolsize,
314 unsigned long *offset, char **modname,
315 const unsigned char **modbuildid, char *namebuf)
319 namebuf[KSYM_NAME_LEN - 1] = 0;
322 if (is_ksym_addr(addr)) {
325 pos = get_symbol_pos(addr, symbolsize, offset);
327 kallsyms_expand_symbol(get_symbol_offset(pos),
328 namebuf, KSYM_NAME_LEN);
338 /* See if it's in a module or a BPF JITed image. */
339 ret = module_address_lookup(addr, symbolsize, offset,
340 modname, modbuildid, namebuf);
342 ret = bpf_address_lookup(addr, symbolsize,
343 offset, modname, namebuf);
346 ret = ftrace_mod_address_lookup(addr, symbolsize,
347 offset, modname, namebuf);
350 cleanup_symbol_name(namebuf);
356 * - modname is set to NULL if it's in the kernel.
357 * - We guarantee that the returned name is valid until we reschedule even if.
358 * It resides in a module.
359 * - We also guarantee that modname will be valid until rescheduled.
361 const char *kallsyms_lookup(unsigned long addr,
362 unsigned long *symbolsize,
363 unsigned long *offset,
364 char **modname, char *namebuf)
366 return kallsyms_lookup_buildid(addr, symbolsize, offset, modname,
370 int lookup_symbol_name(unsigned long addr, char *symname)
375 symname[KSYM_NAME_LEN - 1] = '\0';
377 if (is_ksym_addr(addr)) {
380 pos = get_symbol_pos(addr, NULL, NULL);
382 kallsyms_expand_symbol(get_symbol_offset(pos),
383 symname, KSYM_NAME_LEN);
386 /* See if it's in a module. */
387 res = lookup_module_symbol_name(addr, symname);
392 cleanup_symbol_name(symname);
396 int lookup_symbol_attrs(unsigned long addr, unsigned long *size,
397 unsigned long *offset, char *modname, char *name)
402 name[KSYM_NAME_LEN - 1] = '\0';
404 if (is_ksym_addr(addr)) {
407 pos = get_symbol_pos(addr, size, offset);
409 kallsyms_expand_symbol(get_symbol_offset(pos),
410 name, KSYM_NAME_LEN);
414 /* See if it's in a module. */
415 res = lookup_module_symbol_attrs(addr, size, offset, modname, name);
420 cleanup_symbol_name(name);
424 /* Look up a kernel symbol and return it in a text buffer. */
425 static int __sprint_symbol(char *buffer, unsigned long address,
426 int symbol_offset, int add_offset, int add_buildid)
429 const unsigned char *buildid;
431 unsigned long offset, size;
434 address += symbol_offset;
435 name = kallsyms_lookup_buildid(address, &size, &offset, &modname, &buildid,
438 return sprintf(buffer, "0x%lx", address - symbol_offset);
441 strcpy(buffer, name);
442 len = strlen(buffer);
443 offset -= symbol_offset;
446 len += sprintf(buffer + len, "+%#lx/%#lx", offset, size);
449 len += sprintf(buffer + len, " [%s", modname);
450 #if IS_ENABLED(CONFIG_STACKTRACE_BUILD_ID)
451 if (add_buildid && buildid) {
452 /* build ID should match length of sprintf */
453 #if IS_ENABLED(CONFIG_MODULES)
454 static_assert(sizeof(typeof_member(struct module, build_id)) == 20);
456 len += sprintf(buffer + len, " %20phN", buildid);
459 len += sprintf(buffer + len, "]");
466 * sprint_symbol - Look up a kernel symbol and return it in a text buffer
467 * @buffer: buffer to be stored
468 * @address: address to lookup
470 * This function looks up a kernel symbol with @address and stores its name,
471 * offset, size and module name to @buffer if possible. If no symbol was found,
472 * just saves its @address as is.
474 * This function returns the number of bytes stored in @buffer.
476 int sprint_symbol(char *buffer, unsigned long address)
478 return __sprint_symbol(buffer, address, 0, 1, 0);
480 EXPORT_SYMBOL_GPL(sprint_symbol);
483 * sprint_symbol_build_id - Look up a kernel symbol and return it in a text buffer
484 * @buffer: buffer to be stored
485 * @address: address to lookup
487 * This function looks up a kernel symbol with @address and stores its name,
488 * offset, size, module name and module build ID to @buffer if possible. If no
489 * symbol was found, just saves its @address as is.
491 * This function returns the number of bytes stored in @buffer.
493 int sprint_symbol_build_id(char *buffer, unsigned long address)
495 return __sprint_symbol(buffer, address, 0, 1, 1);
497 EXPORT_SYMBOL_GPL(sprint_symbol_build_id);
500 * sprint_symbol_no_offset - Look up a kernel symbol and return it in a text buffer
501 * @buffer: buffer to be stored
502 * @address: address to lookup
504 * This function looks up a kernel symbol with @address and stores its name
505 * and module name to @buffer if possible. If no symbol was found, just saves
506 * its @address as is.
508 * This function returns the number of bytes stored in @buffer.
510 int sprint_symbol_no_offset(char *buffer, unsigned long address)
512 return __sprint_symbol(buffer, address, 0, 0, 0);
514 EXPORT_SYMBOL_GPL(sprint_symbol_no_offset);
517 * sprint_backtrace - Look up a backtrace symbol and return it in a text buffer
518 * @buffer: buffer to be stored
519 * @address: address to lookup
521 * This function is for stack backtrace and does the same thing as
522 * sprint_symbol() but with modified/decreased @address. If there is a
523 * tail-call to the function marked "noreturn", gcc optimized out code after
524 * the call so that the stack-saved return address could point outside of the
525 * caller. This function ensures that kallsyms will find the original caller
526 * by decreasing @address.
528 * This function returns the number of bytes stored in @buffer.
530 int sprint_backtrace(char *buffer, unsigned long address)
532 return __sprint_symbol(buffer, address, -1, 1, 0);
536 * sprint_backtrace_build_id - Look up a backtrace symbol and return it in a text buffer
537 * @buffer: buffer to be stored
538 * @address: address to lookup
540 * This function is for stack backtrace and does the same thing as
541 * sprint_symbol() but with modified/decreased @address. If there is a
542 * tail-call to the function marked "noreturn", gcc optimized out code after
543 * the call so that the stack-saved return address could point outside of the
544 * caller. This function ensures that kallsyms will find the original caller
545 * by decreasing @address. This function also appends the module build ID to
546 * the @buffer if @address is within a kernel module.
548 * This function returns the number of bytes stored in @buffer.
550 int sprint_backtrace_build_id(char *buffer, unsigned long address)
552 return __sprint_symbol(buffer, address, -1, 1, 1);
555 /* To avoid using get_symbol_offset for every symbol, we carry prefix along. */
556 struct kallsym_iter {
560 loff_t pos_ftrace_mod_end;
563 unsigned int nameoff; /* If iterating in core kernel symbols. */
565 char name[KSYM_NAME_LEN];
566 char module_name[MODULE_NAME_LEN];
571 int __weak arch_get_kallsym(unsigned int symnum, unsigned long *value,
572 char *type, char *name)
577 static int get_ksymbol_arch(struct kallsym_iter *iter)
579 int ret = arch_get_kallsym(iter->pos - kallsyms_num_syms,
580 &iter->value, &iter->type,
584 iter->pos_arch_end = iter->pos;
591 static int get_ksymbol_mod(struct kallsym_iter *iter)
593 int ret = module_get_kallsym(iter->pos - iter->pos_arch_end,
594 &iter->value, &iter->type,
595 iter->name, iter->module_name,
598 iter->pos_mod_end = iter->pos;
606 * ftrace_mod_get_kallsym() may also get symbols for pages allocated for ftrace
607 * purposes. In that case "__builtin__ftrace" is used as a module name, even
608 * though "__builtin__ftrace" is not a module.
610 static int get_ksymbol_ftrace_mod(struct kallsym_iter *iter)
612 int ret = ftrace_mod_get_kallsym(iter->pos - iter->pos_mod_end,
613 &iter->value, &iter->type,
614 iter->name, iter->module_name,
617 iter->pos_ftrace_mod_end = iter->pos;
624 static int get_ksymbol_bpf(struct kallsym_iter *iter)
628 strlcpy(iter->module_name, "bpf", MODULE_NAME_LEN);
630 ret = bpf_get_kallsym(iter->pos - iter->pos_ftrace_mod_end,
631 &iter->value, &iter->type,
634 iter->pos_bpf_end = iter->pos;
642 * This uses "__builtin__kprobes" as a module name for symbols for pages
643 * allocated for kprobes' purposes, even though "__builtin__kprobes" is not a
646 static int get_ksymbol_kprobe(struct kallsym_iter *iter)
648 strlcpy(iter->module_name, "__builtin__kprobes", MODULE_NAME_LEN);
650 return kprobe_get_kallsym(iter->pos - iter->pos_bpf_end,
651 &iter->value, &iter->type,
652 iter->name) < 0 ? 0 : 1;
655 /* Returns space to next name. */
656 static unsigned long get_ksymbol_core(struct kallsym_iter *iter)
658 unsigned off = iter->nameoff;
660 iter->module_name[0] = '\0';
661 iter->value = kallsyms_sym_address(iter->pos);
663 iter->type = kallsyms_get_symbol_type(off);
665 off = kallsyms_expand_symbol(off, iter->name, ARRAY_SIZE(iter->name));
667 return off - iter->nameoff;
670 static void reset_iter(struct kallsym_iter *iter, loff_t new_pos)
672 iter->name[0] = '\0';
673 iter->nameoff = get_symbol_offset(new_pos);
676 iter->pos_arch_end = 0;
677 iter->pos_mod_end = 0;
678 iter->pos_ftrace_mod_end = 0;
679 iter->pos_bpf_end = 0;
684 * The end position (last + 1) of each additional kallsyms section is recorded
685 * in iter->pos_..._end as each section is added, and so can be used to
686 * determine which get_ksymbol_...() function to call next.
688 static int update_iter_mod(struct kallsym_iter *iter, loff_t pos)
692 if ((!iter->pos_arch_end || iter->pos_arch_end > pos) &&
693 get_ksymbol_arch(iter))
696 if ((!iter->pos_mod_end || iter->pos_mod_end > pos) &&
697 get_ksymbol_mod(iter))
700 if ((!iter->pos_ftrace_mod_end || iter->pos_ftrace_mod_end > pos) &&
701 get_ksymbol_ftrace_mod(iter))
704 if ((!iter->pos_bpf_end || iter->pos_bpf_end > pos) &&
705 get_ksymbol_bpf(iter))
708 return get_ksymbol_kprobe(iter);
711 /* Returns false if pos at or past end of file. */
712 static int update_iter(struct kallsym_iter *iter, loff_t pos)
714 /* Module symbols can be accessed randomly. */
715 if (pos >= kallsyms_num_syms)
716 return update_iter_mod(iter, pos);
718 /* If we're not on the desired position, reset to new position. */
719 if (pos != iter->pos)
720 reset_iter(iter, pos);
722 iter->nameoff += get_ksymbol_core(iter);
728 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
732 if (!update_iter(m->private, *pos))
737 static void *s_start(struct seq_file *m, loff_t *pos)
739 if (!update_iter(m->private, *pos))
744 static void s_stop(struct seq_file *m, void *p)
748 static int s_show(struct seq_file *m, void *p)
751 struct kallsym_iter *iter = m->private;
753 /* Some debugging symbols have no name. Ignore them. */
757 value = iter->show_value ? (void *)iter->value : NULL;
759 if (iter->module_name[0]) {
763 * Label it "global" if it is exported,
764 * "local" if not exported.
766 type = iter->exported ? toupper(iter->type) :
768 seq_printf(m, "%px %c %s\t[%s]\n", value,
769 type, iter->name, iter->module_name);
771 seq_printf(m, "%px %c %s\n", value,
772 iter->type, iter->name);
776 static const struct seq_operations kallsyms_op = {
783 #ifdef CONFIG_BPF_SYSCALL
785 struct bpf_iter__ksym {
786 __bpf_md_ptr(struct bpf_iter_meta *, meta);
787 __bpf_md_ptr(struct kallsym_iter *, ksym);
790 static int ksym_prog_seq_show(struct seq_file *m, bool in_stop)
792 struct bpf_iter__ksym ctx;
793 struct bpf_iter_meta meta;
794 struct bpf_prog *prog;
797 prog = bpf_iter_get_info(&meta, in_stop);
802 ctx.ksym = m ? m->private : NULL;
803 return bpf_iter_run_prog(prog, &ctx);
806 static int bpf_iter_ksym_seq_show(struct seq_file *m, void *p)
808 return ksym_prog_seq_show(m, false);
811 static void bpf_iter_ksym_seq_stop(struct seq_file *m, void *p)
814 (void) ksym_prog_seq_show(m, true);
819 static const struct seq_operations bpf_iter_ksym_ops = {
822 .stop = bpf_iter_ksym_seq_stop,
823 .show = bpf_iter_ksym_seq_show,
826 static int bpf_iter_ksym_init(void *priv_data, struct bpf_iter_aux_info *aux)
828 struct kallsym_iter *iter = priv_data;
832 /* cache here as in kallsyms_open() case; use current process
833 * credentials to tell BPF iterators if values should be shown.
835 iter->show_value = kallsyms_show_value(current_cred());
840 DEFINE_BPF_ITER_FUNC(ksym, struct bpf_iter_meta *meta, struct kallsym_iter *ksym)
842 static const struct bpf_iter_seq_info ksym_iter_seq_info = {
843 .seq_ops = &bpf_iter_ksym_ops,
844 .init_seq_private = bpf_iter_ksym_init,
845 .fini_seq_private = NULL,
846 .seq_priv_size = sizeof(struct kallsym_iter),
849 static struct bpf_iter_reg ksym_iter_reg_info = {
851 .feature = BPF_ITER_RESCHED,
852 .ctx_arg_info_size = 1,
854 { offsetof(struct bpf_iter__ksym, ksym),
855 PTR_TO_BTF_ID_OR_NULL },
857 .seq_info = &ksym_iter_seq_info,
860 BTF_ID_LIST(btf_ksym_iter_id)
861 BTF_ID(struct, kallsym_iter)
863 static int __init bpf_ksym_iter_register(void)
865 ksym_iter_reg_info.ctx_arg_info[0].btf_id = *btf_ksym_iter_id;
866 return bpf_iter_reg_target(&ksym_iter_reg_info);
869 late_initcall(bpf_ksym_iter_register);
871 #endif /* CONFIG_BPF_SYSCALL */
873 static inline int kallsyms_for_perf(void)
875 #ifdef CONFIG_PERF_EVENTS
876 extern int sysctl_perf_event_paranoid;
877 if (sysctl_perf_event_paranoid <= 1)
884 * We show kallsyms information even to normal users if we've enabled
885 * kernel profiling and are explicitly not paranoid (so kptr_restrict
886 * is clear, and sysctl_perf_event_paranoid isn't set).
888 * Otherwise, require CAP_SYSLOG (assuming kptr_restrict isn't set to
891 bool kallsyms_show_value(const struct cred *cred)
893 switch (kptr_restrict) {
895 if (kallsyms_for_perf())
899 if (security_capable(cred, &init_user_ns, CAP_SYSLOG,
900 CAP_OPT_NOAUDIT) == 0)
908 static int kallsyms_open(struct inode *inode, struct file *file)
911 * We keep iterator in m->private, since normal case is to
912 * s_start from where we left off, so we avoid doing
913 * using get_symbol_offset for every symbol.
915 struct kallsym_iter *iter;
916 iter = __seq_open_private(file, &kallsyms_op, sizeof(*iter));
922 * Instead of checking this on every s_show() call, cache
923 * the result here at open time.
925 iter->show_value = kallsyms_show_value(file->f_cred);
929 #ifdef CONFIG_KGDB_KDB
930 const char *kdb_walk_kallsyms(loff_t *pos)
932 static struct kallsym_iter kdb_walk_kallsyms_iter;
934 memset(&kdb_walk_kallsyms_iter, 0,
935 sizeof(kdb_walk_kallsyms_iter));
936 reset_iter(&kdb_walk_kallsyms_iter, 0);
939 if (!update_iter(&kdb_walk_kallsyms_iter, *pos))
942 /* Some debugging symbols have no name. Ignore them. */
943 if (kdb_walk_kallsyms_iter.name[0])
944 return kdb_walk_kallsyms_iter.name;
947 #endif /* CONFIG_KGDB_KDB */
949 static const struct proc_ops kallsyms_proc_ops = {
950 .proc_open = kallsyms_open,
951 .proc_read = seq_read,
952 .proc_lseek = seq_lseek,
953 .proc_release = seq_release_private,
956 static int __init kallsyms_init(void)
958 proc_create("kallsyms", 0444, NULL, &kallsyms_proc_ops);
961 device_initcall(kallsyms_init);