2 * vi:set cin noet sw=4 tw=70:
3 * Copyright 2004, International Business Machines Corporation and others.
6 * This software has been released under the terms of the IBM Public
7 * License. For details, see the LICENSE file in the top-level source
8 * directory or online at http://www.openafs.org/dl/license10.html
10 * Portions of this code borrowed from arla under the following terms:
11 * Copyright (c) 2003-2004 Kungliga Tekniska Högskolan
12 * (Royal Institute of Technology, Stockholm, Sweden).
13 * All rights reserved.
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
22 * 2. Redistributions in binary form must reproduce the above copyright
23 * notice, this list of conditions and the following disclaimer in the
24 * documentation and/or other materials provided with the distribution.
26 * 3. Neither the name of the Institute nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
30 * Alternatively, this software may be distributed under the terms of the
31 * GNU General Public License ("GPL").
33 * THIS SOFTWARE IS PROVIDED BY THE INSTITUTE AND CONTRIBUTORS ``AS IS'' AND
34 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
35 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
36 * ARE DISCLAIMED. IN NO EVENT SHALL THE INSTITUTE OR CONTRIBUTORS BE LIABLE
37 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
38 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
39 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
41 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
42 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
46 /* Code to find the Linux syscall table */
48 #ifdef OSI_PROBE_STANDALONE
49 #define OSI_PROBE_DEBUG
51 #ifndef OSI_PROBE_STANDALONE
52 #include <afsconfig.h>
53 #include "afs/param.h"
55 #ifdef AFS_LINUX24_ENV
56 #include <linux/module.h> /* early to avoid printf->printk mapping */
57 #ifndef OSI_PROBE_STANDALONE
58 #include "afs/sysincludes.h"
59 #include "afsincludes.h"
61 #include <linux/version.h>
62 #ifdef CONFIG_H_EXISTS
63 #include <linux/config.h>
65 #include <linux/linkage.h>
66 #include <linux/init.h>
67 #include <linux/unistd.h>
69 #ifdef AFS_LINUX26_ENV
70 #include <scsi/scsi.h> /* for scsi_command_size */
73 #if defined(AFS_PPC64_LINUX26_ENV)
74 #include <asm/abs_addr.h>
77 #ifdef AFS_AMD64_LINUX20_ENV
78 #include <asm/ia32_unistd.h>
81 /* number of syscalls */
82 /* NB: on MIPS we care about the 4xxx range */
84 #define NR_syscalls 222
87 /* lower bound of valid kernel text pointers */
88 #ifdef AFS_IA64_LINUX20_ENV
89 #define ktxt_lower_bound (((unsigned long)&kernel_thread ) & 0xfff00000L)
90 #elif defined(AFS_PPC64_LINUX20_ENV)
91 #define ktxt_lower_bound (KERNELBASE)
93 #define ktxt_lower_bound (((unsigned long)&kernel_thread ) & ~0xfffffL)
96 /* On SPARC64 and S390X, sys_call_table contains 32-bit entries
97 * even though pointers are 64 bit quantities.
99 #if defined(AFS_SPARC64_LINUX20_ENV) || defined(AFS_S390X_LINUX24_ENV)
100 #define SYSCALLTYPE unsigned int
101 #define PROBETYPE int
103 #define SYSCALLTYPE void *
104 #define PROBETYPE long
107 #if defined(AFS_S390X_LINUX20_ENV) && !defined(AFS_S390X_LINUX26_ENV)
108 #define _SS(x) ((x) << 1)
109 #define _SX(x) ((x) &~ 1)
115 /* Older Linux doesn't have __user. The sys_read prototype needs it. */
120 /* Allow the user to specify sys_call_table addresses */
121 static unsigned long sys_call_table_addr[4] = { 0,0,0,0 };
122 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
123 module_param_array(sys_call_table_addr, long, NULL, 0);
125 MODULE_PARM(sys_call_table_addr, "1-4l");
127 MODULE_PARM_DESC(sys_call_table_addr, "Location of system call tables");
129 /* If this is set, we are more careful about avoiding duplicate matches */
130 static int probe_carefully = 1;
132 module_param(probe_carefully, int, 0);
134 MODULE_PARM(probe_carefully, "i");
136 MODULE_PARM_DESC(probe_carefully, "Probe for system call tables carefully");
138 static int probe_ignore_syscalls[8] = { -1, -1, -1, -1, -1, -1, -1, -1 };
139 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
140 module_param_array(probe_ignore_syscalls, int, NULL, 0);
142 MODULE_PARM(probe_ignore_syscalls, "1-8i");
144 MODULE_PARM_DESC(probe_ignore_syscalls, "Syscalls to ignore in table checks");
146 #ifdef OSI_PROBE_DEBUG
149 * 0x0001 - General debugging
150 * 0x0002 - detail - try
151 * 0x0004 - detail - try_harder
152 * 0x0008 - detail - check_table
153 * 0x0010 - detail - check_harder
154 * 0x0020 - detail - check_harder/zapped
155 * 0x0040 - automatically ignore setgroups and afs_syscall
157 static int probe_debug = 0x41;
159 module_param(probe_debug, int, 0);
161 MODULE_PARM(probe_debug, "i");
163 MODULE_PARM_DESC(probe_debug, "Debugging level");
165 static unsigned long probe_debug_addr[4] = { 0,0,0,0 };
166 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
167 module_param_array(probe_debug_addr, long, NULL, 0);
169 MODULE_PARM(probe_debug_addr, "1-4l");
171 MODULE_PARM_DESC(probe_debug_addr, "Debug range starting locations");
173 static unsigned long probe_debug_range = 0;
175 module_param(probe_debug_range, long, 0);
177 MODULE_PARM(probe_debug_range, "l");
179 MODULE_PARM_DESC(probe_debug_range, "Debug range length");
181 static unsigned long probe_debug_tag = 0;
183 module_param(probe_debug_tag, long, 0);
185 MODULE_PARM(probe_debug_tag, "l");
187 MODULE_PARM_DESC(probe_debug_tag, "Debugging output start tag");
191 /* Weak references are our friends. They are supported by the in-kernel
192 * linker in Linux 2.6 and by all versions of modutils back to 2.2pre1.
193 * A weak reference not satisified by the kernel will have value zero.
195 * Unfortunately, weak references to functions don't work right on
196 * IA64; specifically, if you actually try to make a call through
197 * such a reference, and the symbol doesn't exist in the kernel, then
198 * the module relocation code will oops. A workaround for this is
199 * probably possible, but the use of kallsyms_* is of limited value,
200 * so I'm not bothing with the effort for now.
201 * -- jhutz, 10-Feb-2005
203 #ifdef OSI_PROBE_KALLSYMS
204 extern int kallsyms_symbol_to_address(char *name, unsigned long *token,
206 unsigned long *mod_start,
207 unsigned long *mod_end,
209 unsigned long *sec_start,
210 unsigned long *sec_end,
212 unsigned long *sym_start,
213 unsigned long *sym_end
214 ) __attribute__((weak));
216 extern int kallsyms_address_to_symbol(unsigned long address,
218 unsigned long *mod_start,
219 unsigned long *mod_end,
221 unsigned long *sec_start,
222 unsigned long *sec_end,
224 unsigned long *sym_start,
225 unsigned long *sym_end
226 ) __attribute__((weak));
229 extern SYSCALLTYPE sys_call_table[] __attribute__((weak));
230 extern SYSCALLTYPE ia32_sys_call_table[] __attribute__((weak));
231 extern SYSCALLTYPE sys_call_table32[] __attribute__((weak));
232 extern SYSCALLTYPE sys_call_table_emu[] __attribute__((weak));
234 extern asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count) __attribute__((weak));
235 extern asmlinkage long sys_close(unsigned int) __attribute__((weak));
236 #if defined(EXPORTED_SYS_CHDIR)
237 extern asmlinkage long sys_chdir(const char *) __attribute__((weak));
239 extern asmlinkage ssize_t sys_write(unsigned int, const char *, size_t) __attribute__((weak));
240 #ifdef AFS_LINUX26_ENV
241 extern asmlinkage long sys_wait4(pid_t, int *, int, struct rusage *) __attribute__((weak));
243 extern asmlinkage long sys_wait4(pid_t, unsigned int *, int, struct rusage *) __attribute__((weak));
245 extern asmlinkage long sys_exit (int) __attribute__((weak));
246 #if defined(EXPORTED_SYS_OPEN)
247 extern asmlinkage long sys_open (const char *, int, int) __attribute__((weak));
249 extern asmlinkage long sys_ioctl(unsigned int, unsigned int, unsigned long) __attribute__((weak));
250 #if defined(EXPORTED_TASKLIST_LOCK)
251 extern rwlock_t tasklist_lock __attribute__((weak));
255 /* Structures used to control probing. We put all the details of which
256 * symbols we're interested in, what syscall functions to look for, etc
257 * into tables, so we can then have a single copy of the functions that
258 * actually do the work.
271 char *symbol; /* symbol name */
272 char *desc; /* description for messages */
273 int offset; /* first syscall number in table */
275 void *weak_answer; /* weak symbol ref */
276 void *parm_answer; /* module parameter answer */
277 void *debug_answer; /* module parameter answer */
278 unsigned long given_answer; /* compiled-in answer, if any */
280 tryctl *trylist; /* array of combinations to try */
282 unsigned long try_sect_sym; /* symbol in section to try scanning */
283 unsigned long try_base; /* default base address for scan */
284 unsigned long try_base_mask; /* base address bits to force to zero */
285 unsigned long try_length; /* default length for scan */
287 unsigned long alt_try_sect_sym; /* symbol in section to try scanning */
288 unsigned long alt_try_base; /* default base address for scan */
289 unsigned long alt_try_base_mask; /* base address bits to force to zero */
290 unsigned long alt_try_length; /* default length for scan */
292 int n_zapped_syscalls; /* number of unimplemented system calls */
293 int *zapped_syscalls; /* list of unimplemented system calls */
295 int n_unique_syscalls; /* number of unique system calls */
296 int *unique_syscalls; /* list of unimplemented system calls */
298 int verifyNR; /* syscall number to verify match */
299 void *verify_fn; /* syscall pointer to verify match */
301 int debug_ignore_NR[4]; /* syscalls to ignore for debugging */
306 /********** Probing Configuration: sys_call_table **********/
308 /* syscall pairs/triplets to probe */
309 /* On PPC64 and SPARC64, we need to omit the ones that might match both tables */
310 static tryctl main_try[] = {
311 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
312 #if defined(EXPORTED_SYS_CHDIR)
313 { "scan: close+chdir+write", __NR_close, &sys_close, __NR_chdir, &sys_chdir, __NR_write, &sys_write },
316 { "scan: close+wait4", __NR_close, &sys_close, __NR_wait4, &sys_wait4, -1, 0 },
317 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
318 #if defined(EXPORTED_SYS_CHDIR)
319 { "scan: close+chdir", __NR_close, &sys_close, __NR_chdir, &sys_chdir, -1, 0 },
322 { "scan: close+ioctl", __NR_close, &sys_close, __NR_ioctl, &sys_ioctl, -1, 0 },
323 #if defined(EXPORTED_SYS_OPEN)
324 { "scan: exit+open", __NR_exit, &sys_exit, __NR_open, &sys_open, -1, 0 },
329 /* zapped syscalls for try_harder */
330 /* this list is based on the table in 'zapped_syscalls' */
332 static int main_zapped_syscalls[] = {
334 * SPARC-Linux uses syscall number mappings chosen to be compatible
335 * with SunOS. So, it doesn't have any of the traditional calls or
336 * the new STREAMS ones. However, there are a number of syscalls
337 * which are SunOS-specific (not implemented on Linux), i386-specific
338 * (not implemented on SPARC-Linux), or implemented only on one of
339 * sparc32 or sparc64. Of course, there are no __NR macros for most
342 * Note that the calls we list here are implemented by sys_nis_syscall,
343 * not by sys_ni_syscall. That means we have to exclude all of the
344 * other entries, or we might get a sys_ni_syscall into the list and
345 * the test would no longer work.
347 #if defined(AFS_SPARC64_LINUX20_ENV)
348 /* mmap2, fstat64, getmsg, putmsg, modify_ldt */
349 56, 63, 151, 152, 218,
350 #elif defined(AFS_SPARC_LINUX20_ENV)
351 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
352 52, 151, 152, 164, 218,
353 #else /* !AFS_SPARC_LINUX20_ENV */
356 * These 7 syscalls are present in the syscall table on most "older"
357 * platforms that use the traditional syscall number mappings. They
358 * are not implemented on any platform.
382 * On s390 and arm (but not arm26), the seven traditional unimplemented
383 * system calls are indeed present and unimplemented. However, the
384 * corresponding __NR macros are not defined, so the tests above fail.
385 * Instead, we just have to know the numbers for these.
387 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
388 /* break, stty, gtty, ftime, prof, lock, mpx */
389 17, 31, 32, 35, 44, 53, 56,
393 * Sadly, some newer platforms like IA64, amd64, and PA-RISC don't have
394 * the traditional numbers, so the list above are not helpful. They
395 * do have entries for getpmsg/putpmsg, which are always unimplemented.
405 * The module-loading mechanism changed in Linux 2.6, and insmod's
406 * loss is our gain: three new unimplemented system calls!
408 #if defined(AFS_LINUX26_ENV)
412 #ifdef __NR_query_module
415 #ifdef __NR_get_kernel_syms
416 __NR_get_kernel_syms,
418 #endif /* AFS_LINUX26_ENV */
421 * On IA64, the old module-loading calls are indeed present and
422 * unimplemented, but the __NR macros are not defined. Again,
423 * we simply have to know their numbers.
425 #ifdef AFS_IA64_LINUX26_ENV
426 /* create_module, query_module, get_kernel_sysms */
430 /* And the same deal for arm (not arm26), if we ever support that. */
432 /* create_module, query_module, get_kernel_sysms */
437 * Alpha-Linux uses syscall number mappings chosen to be compatible
438 * with OSF/1. So, it doesn't have any of the traditional calls or
439 * the new STREAMS ones, but it does have several OSF/1-specific
440 * syscalls which are not implemented on Linux. These don't exist on
441 * any other platform.
443 #ifdef __NR_osf_syscall
446 #ifdef __NR_osf_profil
449 #ifdef __NR_osf_reboot
452 #ifdef __NR_osf_kmodcall
455 #ifdef __NR_osf_old_vtrace
460 * On PPC64, we need a couple more entries to distinguish the two
461 * tables, since the system call numbers are the same and the sets of
462 * unimplemented calls are very similar.
463 * mmap2 and fstat64 are implemented only for 32-bit calls
465 #ifdef AFS_PPC64_LINUX20_ENV
466 /* _mmap2, _fstat64 */
468 #endif /* AFS_PPC64_LINUX20_ENV */
470 /* Similarly for S390X, with lcown16 and fstat64 */
471 #ifdef AFS_S390X_LINUX20_ENV
472 /* lchown16, fstat64 */
475 #endif /* !AFS_SPARC_LINUX20_ENV */
479 /* unique syscalls for try_harder */
480 static int main_unique_syscalls[] = {
481 #if defined(AFS_SPARC64_LINUX24_ENV) || defined(AFS_SPARC_LINUX24_ENV)
483 * On SPARC, we need some additional unique calls to make sure
484 * we don't match the SunOS-compatibility table.
486 __NR_sgetmask, __NR_ssetmask,
488 __NR_exit, __NR_mount, __NR_read, __NR_write,
489 __NR_open, __NR_close, __NR_unlink
492 /* probe control structure */
493 static probectl main_probe = {
494 /* symbol name and description */
498 /* syscall number of first entry in table */
499 #ifdef AFS_IA64_LINUX20_ENV
505 sys_call_table, /* weak symbol ref */
506 0, 0, /* module parameter answers */
507 #ifdef AFS_LINUX_sys_call_table
508 AFS_LINUX_sys_call_table, /* compiled-in answer, if any */
513 main_try, /* array of combinations to try */
515 /* symbol in section to try scanning */
516 #if defined(AFS_SPARC64_LINUX20_ENV) || defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
517 (unsigned long)&sys_close,
518 #elif defined(AFS_AMD64_LINUX20_ENV)
519 /* On this platform, it's in a different section! */
520 (unsigned long)&tasklist_lock,
522 (unsigned long)&init_mm,
525 /* default base address for scan */
526 /* base address bits to force to zero */
527 /* default length for scan */
528 #if defined(AFS_SPARC64_LINUX20_ENV)
529 (unsigned long)(&sys_close),
532 #elif defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
533 /* bleah; this is so suboptimal */
534 (unsigned long)(&sys_close),
537 #elif defined(AFS_IA64_LINUX20_ENV)
538 (unsigned long)(&init_mm),
541 #elif defined(AFS_AMD64_LINUX20_ENV)
542 (unsigned long)(&tasklist_lock) - 0x30000,
545 #elif defined(AFS_PPC64_LINUX26_ENV)
546 (unsigned long)(&do_signal),
549 #elif defined(AFS_PPC_LINUX20_ENV) || defined(AFS_PPC_LINUX20_ENV)
550 (unsigned long)&init_mm,
554 (unsigned long)&init_mm,
559 #ifdef AFS_LINUX26_ENV
560 (unsigned long)scsi_command_size,
561 (unsigned long)scsi_command_size,
568 /* number and list of unimplemented system calls */
569 ((sizeof(main_zapped_syscalls)/sizeof(main_zapped_syscalls[0])) - 1),
570 main_zapped_syscalls,
572 /* number and list of unique system calls */
573 (sizeof(main_unique_syscalls)/sizeof(main_unique_syscalls[0])),
574 main_unique_syscalls,
576 /* syscall number and pointer to verify match */
577 __NR_close, &sys_close,
579 /* syscalls to ignore for debugging */
581 #if defined(AFS_ALPHA_LINUX20_ENV)
583 #elif defined(AFS_AMD64_LINUX20_ENV)
585 #elif defined(AFS_IA64_LINUX20_ENV)
587 #elif defined(AFS_SPARC_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
593 #ifdef __NR_setgroups32
603 /********** Probing Configuration: amd64 ia32_sys_call_table **********/
604 #if defined(AFS_AMD64_LINUX20_ENV)
606 /* syscall pairs/triplets to probe */
607 static tryctl ia32_try[] = {
608 #if defined(EXPORTED_SYS_CHDIR)
609 { "scan: close+chdir+write", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, __NR_ia32_write, &sys_write },
610 { "scan: close+chdir", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, -1, 0 },
615 /* zapped syscalls for try_harder */
616 static int ia32_zapped_syscalls[] = {
617 __NR_ia32_break, __NR_ia32_stty, __NR_ia32_gtty, __NR_ia32_ftime,
618 __NR_ia32_prof, __NR_ia32_lock, __NR_ia32_mpx,
622 /* unique syscalls for try_harder */
623 static int ia32_unique_syscalls[] = {
624 __NR_ia32_exit, __NR_ia32_mount, __NR_ia32_read, __NR_ia32_write,
625 __NR_ia32_open, __NR_ia32_close, __NR_ia32_unlink
628 /* probe control structure */
629 static probectl ia32_probe = {
630 /* symbol name and description */
631 "ia32_sys_call_table",
632 "32-bit system call table",
634 /* syscall number of first entry in table */
637 ia32_sys_call_table, /* weak symbol ref */
638 0, 0, /* module parameter answers */
639 #ifdef AFS_LINUX_ia32_sys_call_table
640 AFS_LINUX_ia32_sys_call_table,/* compiled-in answer, if any */
645 ia32_try, /* array of combinations to try */
647 /* symbol in section to try scanning */
648 (unsigned long)&init_mm,
650 /* default base address for scan */
651 /* base address bits to force to zero */
652 /* default length for scan */
653 (unsigned long)&init_mm,
655 (0x180000 / sizeof(unsigned long *)),
657 #ifdef AFS_LINUX26_ENV
658 (unsigned long)scsi_command_size,
659 (unsigned long)scsi_command_size,
667 /* number and list of unimplemented system calls */
668 ((sizeof(ia32_zapped_syscalls)/sizeof(ia32_zapped_syscalls[0])) - 1),
669 ia32_zapped_syscalls,
671 /* number and list of unique system calls */
672 (sizeof(ia32_unique_syscalls)/sizeof(ia32_unique_syscalls[0])),
673 ia32_unique_syscalls,
675 /* syscall number and pointer to verify match */
676 __NR_ia32_close, &sys_close,
678 /* syscalls to ignore for debugging */
682 __NR_ia32_setgroups32,
687 static probectl *probe_list[] = {
688 &main_probe, &ia32_probe
692 /********** Probing Configuration: IA64 **********/
693 #elif defined(AFS_IA64_LINUX20_ENV)
699 /* no 32-bit support on IA64 for now */
700 static probectl *probe_list[] = {
705 /********** Probing Configuration: ppc64, sparc64 sys_call_table32 **********/
706 #elif defined(AFS_PPC64_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
713 * syscall pairs/triplets to probe
714 * This has to be empty, because anything that would work will
715 * also match the main table, and that's no good.
717 static tryctl sct32_try[] = {
721 /* zapped syscalls for try_harder */
722 static int sct32_zapped_syscalls[] = {
723 #ifdef AFS_PPC64_LINUX20_ENV
724 /* These should be sufficient */
725 __NR_break, __NR_stty, __NR_gtty, __NR_ftime,
726 __NR_prof, __NR_lock, __NR_mpx,
728 #ifdef AFS_SPARC64_LINUX20_ENV
729 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
730 52, 151, 152, 164, 218,
735 /* unique syscalls for try_harder */
736 /* mmap2 and fstat64 are implemented only for 32-bit calls */
737 static int sct32_unique_syscalls[] = {
738 #ifdef AFS_PPC64_LINUX20_ENV
739 /* _mmap2, _fstat64 */
742 #ifdef AFS_SPARC64_LINUX24_ENV
744 * On SPARC, we need some additional unique calls to make sure
745 * we don't match the SunOS-compatibility table.
747 __NR_sgetmask, __NR_ssetmask,
749 __NR_exit, __NR_mount, __NR_read, __NR_write,
750 __NR_open, __NR_close, __NR_unlink
753 /* probe control structure */
754 static probectl sct32_probe = {
755 /* symbol name and description */
757 "32-bit system call table",
759 /* syscall number of first entry in table */
762 sys_call_table32, /* weak symbol ref */
763 0, 0, /* module parameter answers */
764 #ifdef AFS_LINUX_sys_call_table32
765 AFS_LINUX_sys_call_table32, /* compiled-in answer, if any */
770 sct32_try, /* array of combinations to try */
772 /* symbol in section to try scanning */
773 #if defined(AFS_SPARC64_LINUX20_ENV)
774 (unsigned long)&sys_close,
776 (unsigned long)&init_mm,
779 /* default base address for scan */
780 /* base address bits to force to zero */
781 /* default length for scan */
782 #if defined(AFS_SPARC64_LINUX20_ENV)
783 (unsigned long)(&sys_close),
786 #elif defined(AFS_PPC64_LINUX26_ENV)
787 (unsigned long)(&do_signal),
791 (unsigned long)&init_mm,
796 #ifdef AFS_LINUX26_ENV
797 (unsigned long)scsi_command_size,
798 (unsigned long)scsi_command_size,
805 /* number and list of unimplemented system calls */
806 ((sizeof(sct32_zapped_syscalls)/sizeof(sct32_zapped_syscalls[0])) - 1),
807 sct32_zapped_syscalls,
809 /* number and list of unique system calls */
810 (sizeof(sct32_unique_syscalls)/sizeof(sct32_unique_syscalls[0])),
811 sct32_unique_syscalls,
813 /* syscall number and pointer to verify match */
814 __NR_close, &sys_close,
816 /* syscalls to ignore for debugging */
818 #if defined(AFS_SPARC64_LINUX20_ENV)
829 static probectl *probe_list[] = {
830 &main_probe, &sct32_probe
834 /********** Probing Configuration: s390x sys_call_table_emu **********/
835 /* We only actually need to do this on s390x_linux26 and later.
836 * On earlier versions, the two tables were interleaved and so
837 * have related base addresses.
839 #elif defined(AFS_S390X_LINUX26_ENV)
841 /* syscall pairs/triplets to probe */
842 /* nothing worthwhile is exported, so this is empty */
843 static tryctl emu_try[] = {
847 /* zapped syscalls for try_harder */
848 static int emu_zapped_syscalls[] = {
849 /* break, stty, gtty, ftime, prof, lock, mpx */
850 17, 31, 32, 35, 44, 53, 56,
854 /* unique syscalls for try_harder */
855 static int emu_unique_syscalls[] = {
856 /* lchown16, fstat64 */
858 __NR_exit, __NR_mount, __NR_read, __NR_write,
859 __NR_open, __NR_close, __NR_unlink
862 /* probe control structure */
863 static probectl emu_probe = {
864 /* symbol name and description */
865 "sys_call_table_emu",
866 "32-bit system call table",
868 /* syscall number of first entry in table */
871 sys_call_table_emu, /* weak symbol ref */
872 0, 0, /* module parameter answers */
873 #ifdef AFS_LINUX_sys_call_table_emu
874 AFS_LINUX_sys_call_table_emu, /* compiled-in answer, if any */
879 emu_try, /* array of combinations to try */
881 /* symbol in section to try scanning */
882 (unsigned long)&sys_close,
884 /* default base address for scan */
885 /* base address bits to force to zero */
886 /* default length for scan */
887 (unsigned long)&sys_close,
891 #ifdef AFS_LINUX26_ENV
892 (unsigned long)scsi_command_size,
893 (unsigned long)scsi_command_size,
900 /* number and list of unimplemented system calls */
901 ((sizeof(emu_zapped_syscalls)/sizeof(emu_zapped_syscalls[0])) - 1),
904 /* number and list of unique system calls */
905 (sizeof(emu_unique_syscalls)/sizeof(emu_unique_syscalls[0])),
908 /* syscall number and pointer to verify match */
909 __NR_close, &sys_close,
911 /* syscalls to ignore for debugging */
920 static probectl *probe_list[] = {
921 &main_probe, &emu_probe
925 /********** End of Probing Configuration **********/
927 #else /* no per-platform probe control, so use the default list */
928 static probectl *probe_list[] = {
933 #define N_PROBE_LIST (sizeof(probe_list) / sizeof(*probe_list))
934 #define DEBUG_IN_RANGE(P,x) (!probe_debug_range || \
935 (P->debug_answer && \
936 (unsigned long)(x) >= (unsigned long)P->debug_answer && \
937 (unsigned long)(x) < (unsigned long)P->debug_answer + probe_debug_range))
941 static int check_table(probectl *P, PROBETYPE *ptr)
946 for (x = ptr, i = 0; i < _SS(NR_syscalls); i++, x++) {
947 #ifdef OSI_PROBE_DEBUG
948 if (probe_debug & 0x0040) {
949 for (j = 0; j < 4; j++) {
950 if (_SS(P->debug_ignore_NR[j]) == _SX(i + P->offset)) break;
955 for (j = 0; j < 8; j++) {
956 if (_SS(probe_ignore_syscalls[j]) == _SX(i) + P->offset) break;
959 if (*x <= ktxt_lower_bound) {
960 #ifdef OSI_PROBE_DEBUG
961 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
962 printk("<7>check 0x%lx -> %d [0x%lx]\n",
963 (unsigned long)ptr, i, (unsigned long)*x);
968 #ifdef OSI_PROBE_DEBUG
969 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
970 printk("<7>check 0x%lx -> ok\n", (unsigned long)ptr);
975 static void *try(probectl *P, tryctl *T, PROBETYPE *aptr,
976 unsigned long datalen)
978 #ifdef OSI_PROBE_KALLSYMS
979 char *mod_name, *sec_name, *sym_name;
980 unsigned long mod_start, mod_end;
981 unsigned long sec_start, sec_end;
982 unsigned long sym_start, sym_end;
984 unsigned long offset, ip1, ip2, ip3;
988 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
989 ip1 = T->fn1 ? (unsigned long)((struct fptr *)T->fn1)->ip : 0;
990 ip2 = T->fn2 ? (unsigned long)((struct fptr *)T->fn2)->ip : 0;
991 ip3 = T->fn3 ? (unsigned long)((struct fptr *)T->fn3)->ip : 0;
993 ip1 = (unsigned long)T->fn1;
994 ip2 = (unsigned long)T->fn2;
995 ip3 = (unsigned long)T->fn3;
998 #ifdef OSI_PROBE_DEBUG
999 if (probe_debug & 0x0001)
1000 printk("<7>osi_probe: %s %s (%d->0x%lx, %d->0x%lx, %d->0x%lx)\n",
1001 P->symbol, T->name, T->NR1, ip1, T->NR2, ip2, T->NR3, ip3);
1004 if (!ip1 || !ip2 || (T->NR3 >= 0 && !ip3))
1007 for (offset = 0; offset < datalen; offset++, aptr++) {
1008 #if defined(AFS_PPC64_LINUX20_ENV)
1009 ptr = (PROBETYPE*)(*aptr);
1010 if ((unsigned long)ptr <= KERNELBASE) {
1016 if ((unsigned long)ptr < init_mm.start_code ||
1017 #if defined(AFS_AMD64_LINUX20_ENV)
1018 (unsigned long)ptr > init_mm.brk)
1020 (unsigned long)ptr > init_mm.end_data)
1023 /* printk("address 0x%lx (from 0x%lx %d) is out of range in check_table. wtf?\n", (unsigned long)x, (unsigned long)ptr, i);*/
1027 ret = check_table(P, ptr);
1029 /* return value is number of entries to skip */
1035 #ifdef OSI_PROBE_DEBUG
1036 if ((probe_debug & 0x0002) && DEBUG_IN_RANGE(P,ptr))
1037 printk("<7>try 0x%lx\n", (unsigned long)ptr);
1039 if (ptr[_SS(T->NR1 - P->offset)] != ip1) continue;
1040 if (ptr[_SS(T->NR2 - P->offset)] != ip2) continue;
1041 if (ip3 && ptr[_SS(T->NR3 - P->offset)] != ip3) continue;
1043 #ifdef OSI_PROBE_DEBUG
1044 if (probe_debug & 0x0002)
1045 printk("<7>try found 0x%lx\n", (unsigned long)ptr);
1047 #ifdef OSI_PROBE_KALLSYMS
1048 if (kallsyms_address_to_symbol) {
1049 ret = kallsyms_address_to_symbol((unsigned long)ptr,
1050 &mod_name, &mod_start, &mod_end,
1051 &sec_name, &sec_start, &sec_end,
1052 &sym_name, &sym_start, &sym_end);
1053 if (!ret || strcmp(sym_name, P->symbol)) continue;
1056 /* XXX should we make sure there is only one match? */
1063 static int check_harder(probectl *P, PROBETYPE *p)
1068 /* Check zapped syscalls */
1069 for (i = 1; i < P->n_zapped_syscalls; i++) {
1070 if (p[_SS(P->zapped_syscalls[i])] != p[_SS(P->zapped_syscalls[0])]) {
1071 #ifdef OSI_PROBE_DEBUG
1072 if ((probe_debug & 0x0020) && DEBUG_IN_RANGE(P,p))
1073 printk("<7>check_harder 0x%lx zapped failed i=%d\n", (unsigned long)p, i);
1079 /* Check unique syscalls */
1080 for (i = 0; i < P->n_unique_syscalls; i++) {
1081 for (s = 0; s < NR_syscalls; s++) {
1082 if (p[_SS(s)] == p[_SS(P->unique_syscalls[i])]
1083 && s != P->unique_syscalls[i]) {
1084 #ifdef OSI_PROBE_DEBUG
1085 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1086 printk("<7>check_harder 0x%lx unique failed i=%d s=%d\n", (unsigned long)p, i, s);
1093 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
1094 ip1 = P->verify_fn ? (unsigned long)((struct fptr *)(P->verify_fn))->ip : 0;
1096 ip1 = (unsigned long)(P->verify_fn);
1099 if (ip1 && p[_SS(P->verifyNR - P->offset)] != ip1) {
1100 #ifdef OSI_PROBE_DEBUG
1101 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1102 printk("<7>check_harder 0x%lx verify failed\n", (unsigned long)p);
1107 #ifdef OSI_PROBE_DEBUG
1108 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1109 printk("<7>check_harder 0x%lx success!\n", (unsigned long)p);
1114 static void *try_harder(probectl *P, PROBETYPE *ptr, unsigned long datalen)
1116 #ifdef OSI_PROBE_KALLSYMS
1117 char *mod_name, *sec_name, *sym_name;
1118 unsigned long mod_start, mod_end;
1119 unsigned long sec_start, sec_end;
1120 unsigned long sym_start, sym_end;
1122 unsigned long offset;
1126 #ifdef OSI_PROBE_DEBUG
1127 if (probe_debug & 0x0001)
1128 printk("<7>osi_probe: %s try_harder\n", P->symbol);
1130 for (offset = 0; offset < datalen; offset++, ptr++) {
1131 if ((unsigned long)ptr < init_mm.start_code ||
1132 #if defined(AFS_AMD64_LINUX20_ENV)
1133 (unsigned long)ptr > init_mm.brk)
1135 (unsigned long)ptr > init_mm.end_data)
1138 /* printk("address 0x%lx (from 0x%lx %d) is out of range in check_table. wtf?\n", (unsigned long)x, (unsigned long)ptr, i);*/
1141 ret = check_table(P, ptr);
1143 /* return value is number of entries to skip */
1149 #ifdef OSI_PROBE_DEBUG
1150 if ((probe_debug & 0x0004) && DEBUG_IN_RANGE(P,ptr))
1151 printk("<7>try_harder 0x%lx\n", (unsigned long)ptr);
1153 if (!check_harder(P, ptr))
1156 #ifdef OSI_PROBE_DEBUG
1157 if (probe_debug & 0x0004)
1158 printk("<7>try_harder found 0x%lx\n", (unsigned long)ptr);
1161 #ifdef OSI_PROBE_KALLSYMS
1162 if (kallsyms_address_to_symbol) {
1163 ret = kallsyms_address_to_symbol((unsigned long)ptr,
1164 &mod_name, &mod_start, &mod_end,
1165 &sec_name, &sec_start, &sec_end,
1166 &sym_name, &sym_start, &sym_end);
1167 if (!ret || strcmp(sym_name, P->symbol)) continue;
1172 #ifdef OSI_PROBE_DEBUG
1173 if (probe_debug & 0x0005)
1174 printk("<7>%s: try_harder found multiple matches!\n", P->symbol);
1179 match = (void *)ptr;
1180 if (!probe_carefully)
1187 #ifdef OSI_PROBE_DEBUG
1188 #define check_result(x,m) do { \
1189 if (probe_debug & 0x0001) { \
1190 printk("<7>osi_probe: %s = 0x%016lx %s\n", P->symbol, (unsigned long)(x), (m)); \
1194 final_answer = (void *)(x); \
1198 #define check_result(x,m) do { \
1201 return (void *)(x); \
1205 static void *scan_for_syscall_table(probectl *P, PROBETYPE *B, unsigned long L)
1209 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1212 #ifdef OSI_PROBE_DEBUG
1213 void *final_answer = 0;
1215 #ifdef OSI_PROBE_DEBUG
1216 if (probe_debug & 0x0007)
1217 printk("<7>osi_probe: %s base=0x%lx, len=0x%lx\n",
1218 P->symbol, (unsigned long)B, L);
1219 if (probe_debug & 0x0009) {
1220 printk("<7>osi_probe: %s ktxt_lower_bound=0x%lx\n",
1221 P->symbol, ktxt_lower_bound);
1222 printk("<7>osi_probe: %s NR_syscalls=%d\n",
1223 P->symbol, NR_syscalls);
1227 for (T = P->trylist; T->name; T++) {
1228 answer = try(P, T, B, L);
1229 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1230 answer2 = try(P, T, (PROBETYPE *)(2 + (void *)B), L);
1231 #ifdef OSI_PROBE_DEBUG
1232 if (probe_debug & 0x0003) {
1233 printk("<7>osi_probe: %s = 0x%016lx %s (even)\n",
1234 P->symbol, (unsigned long)(answer), T->name);
1235 printk("<7>osi_probe: %s = 0x%016lx %s (odd)\n",
1236 P->symbol, (unsigned long)(answer2), T->name);
1239 if (answer && answer2) answer = 0;
1240 else if (answer2) answer = answer2;
1246 /* XXX more checks here */
1248 answer = try_harder(P, B, L);
1249 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1250 answer2 = try_harder(P, (PROBETYPE *)(2 + (void *)B), L);
1251 #ifdef OSI_PROBE_DEBUG
1252 if (probe_debug & 0x0005) {
1253 printk("<7>osi_probe: %s = 0x%016lx pattern scan (even)\n",
1254 P->symbol, (unsigned long)(answer));
1255 printk("<7>osi_probe: %s = 0x%016lx pattern scan (odd)\n",
1256 P->symbol, (unsigned long)(answer2));
1259 if (answer && answer2) answer = 0;
1260 else if (answer2) answer = answer2;
1265 static void *do_find_syscall_table(probectl *P, char **method)
1267 #ifdef OSI_PROBE_KALLSYMS
1268 char *mod_name, *sec_name, *sym_name;
1269 unsigned long mod_start, mod_end;
1270 unsigned long sec_start, sec_end;
1271 unsigned long sym_start, sym_end;
1272 unsigned long token;
1278 #ifdef OSI_PROBE_DEBUG
1279 void *final_answer = 0;
1282 *method = "not found";
1284 /* if it's exported, there's nothing to do */
1285 check_result(P->weak_answer, "exported");
1287 /* ask the kernel to do the name lookup, if it's willing */
1288 #ifdef OSI_PROBE_KALLSYMS
1289 if (kallsyms_symbol_to_address) {
1293 ret = kallsyms_symbol_to_address(P->symbol, &token,
1294 &mod_name, &mod_start, &mod_end,
1295 &sec_name, &sec_start, &sec_end,
1296 &sym_name, &sym_start, &sym_end);
1297 if (ret && !strcmp(mod_name, "kernel") && sym_start)
1301 check_result(sym_start, "kallsyms_symbol_to_address");
1305 /* Maybe a little birdie told us */
1306 check_result(P->parm_answer, "module parameter");
1307 check_result(P->given_answer, "compiled-in");
1309 /* OK, so we have to scan. */
1310 B = (PROBETYPE *)((P->try_base) & ~(P->try_base_mask));
1312 /* Now, see if the kernel will tell us something better than the default */
1313 #ifdef OSI_PROBE_KALLSYMS
1314 if (kallsyms_address_to_symbol) {
1315 ret = kallsyms_address_to_symbol(P->try_sect_sym,
1316 &mod_name, &mod_start, &mod_end,
1317 &sec_name, &sec_start, &sec_end,
1318 &sym_name, &sym_start, &sym_end);
1320 B = (PROBETYPE *)sec_start;
1321 L = (sec_end - sec_start) / sizeof(unsigned long);
1326 answer = scan_for_syscall_table(P, B, L);
1327 check_result(answer, "pattern scan");
1328 B = (PROBETYPE *)((P->alt_try_base) & ~(P->alt_try_base_mask));
1329 L = P->alt_try_length;
1330 /* Now, see if the kernel will tell us something better than the default */
1331 #ifdef OSI_PROBE_KALLSYMS
1332 if (kallsyms_address_to_symbol && P->alt_try_sect_sym) {
1333 ret = kallsyms_address_to_symbol(P->alt_try_sect_sym,
1334 &mod_name, &mod_start, &mod_end,
1335 &sec_name, &sec_start, &sec_end,
1336 &sym_name, &sym_start, &sym_end);
1338 B = (PROBETYPE *)sec_start;
1339 L = (sec_end - sec_start) / sizeof(unsigned long);
1344 answer = scan_for_syscall_table(P, B, L);
1345 check_result(answer, "pattern scan");
1347 #ifdef OSI_PROBE_DEBUG
1348 return final_answer;
1354 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1355 static int check_writable(unsigned long address)
1357 pgd_t *pgd = pgd_offset_k(address);
1367 pud = pud_offset(pgd, address);
1370 pmd = pmd_offset(pud, address);
1372 pmd = pmd_offset(pgd, address);
1376 if (pmd_large(*pmd))
1379 pte = pte_offset_kernel(pmd, address);
1380 if (pte_none(*pte) || !pte_present(*pte) || !pte_write(*pte))
1386 void *osi_find_syscall_table(int which)
1392 if (which < 0 || which >= N_PROBE_LIST) {
1393 printk("error - afs_find_syscall_table called with invalid index!\n");
1396 P = probe_list[which];
1398 P->parm_answer = (void *)sys_call_table_addr[which];
1399 #ifdef OSI_PROBE_DEBUG
1400 P->debug_answer = (void *)probe_debug_addr[which];
1403 answer = do_find_syscall_table(P, &method);
1405 printk("Warning: failed to find address of %s\n", P->desc);
1406 printk("System call hooks will not be installed; proceeding anyway\n");
1409 printk("Found %s at 0x%lx (%s)\n", P->desc, (unsigned long)answer, method);
1410 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1411 if (!check_writable((unsigned long)answer)) {
1412 printk("Address 0x%lx is not writable.\n", (unsigned long)answer);
1413 printk("System call hooks will not be installed; proceeding anyway\n");
1421 #ifdef OSI_PROBE_STANDALONE
1422 int __init osi_probe_init(void)
1426 if (!probe_debug_tag) probe_debug_tag = jiffies;
1427 printk("*** osi_probe %ld debug = 0x%04x ***\n",
1428 probe_debug_tag, probe_debug);
1429 for (i = 0; i < N_PROBE_LIST; i++)
1430 (void)osi_find_syscall_table(i);
1434 void osi_probe_exit(void) { }
1436 module_init(osi_probe_init);
1437 module_exit(osi_probe_exit);