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 #include <linux/config.h>
63 #include <linux/linkage.h>
64 #include <linux/init.h>
65 #include <linux/unistd.h>
67 #ifdef AFS_LINUX26_ENV
68 #include <scsi/scsi.h> /* for scsi_command_size */
71 #if defined(AFS_PPC64_LINUX26_ENV)
72 #include <asm/abs_addr.h>
75 #ifdef AFS_AMD64_LINUX20_ENV
76 #include <asm/ia32_unistd.h>
79 /* number of syscalls */
80 /* NB: on MIPS we care about the 4xxx range */
82 #define NR_syscalls 222
85 /* lower bound of valid kernel text pointers */
86 #ifdef AFS_IA64_LINUX20_ENV
87 #define ktxt_lower_bound (((unsigned long)&kernel_thread ) & 0xfff00000L)
88 #elif defined(AFS_PPC64_LINUX20_ENV)
89 #define ktxt_lower_bound (KERNELBASE)
91 #define ktxt_lower_bound (((unsigned long)&kernel_thread ) & ~0xfffffL)
94 /* On SPARC64 and S390X, sys_call_table contains 32-bit entries
95 * even though pointers are 64 bit quantities.
97 #if defined(AFS_SPARC64_LINUX20_ENV) || defined(AFS_S390X_LINUX24_ENV)
98 #define SYSCALLTYPE unsigned int
101 #define SYSCALLTYPE void *
102 #define PROBETYPE long
105 #if defined(AFS_S390X_LINUX20_ENV) && !defined(AFS_S390X_LINUX26_ENV)
106 #define _SS(x) ((x) << 1)
107 #define _SX(x) ((x) &~ 1)
113 /* Older Linux doesn't have __user. The sys_read prototype needs it. */
118 /* Allow the user to specify sys_call_table addresses */
119 static unsigned long sys_call_table_addr[4] = { 0,0,0,0 };
120 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
121 module_param_array(sys_call_table_addr, long, NULL, 0);
123 MODULE_PARM(sys_call_table_addr, "1-4l");
125 MODULE_PARM_DESC(sys_call_table_addr, "Location of system call tables");
127 /* If this is set, we are more careful about avoiding duplicate matches */
128 static int probe_carefully = 1;
130 module_param(probe_carefully, int, 0);
132 MODULE_PARM(probe_carefully, "i");
134 MODULE_PARM_DESC(probe_carefully, "Probe for system call tables carefully");
136 static int probe_ignore_syscalls[8] = { -1, -1, -1, -1, -1, -1, -1, -1 };
137 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
138 module_param_array(probe_ignore_syscalls, int, NULL, 0);
140 MODULE_PARM(probe_ignore_syscalls, "1-8i");
142 MODULE_PARM_DESC(probe_ignore_syscalls, "Syscalls to ignore in table checks");
144 #ifdef OSI_PROBE_DEBUG
147 * 0x0001 - General debugging
148 * 0x0002 - detail - try
149 * 0x0004 - detail - try_harder
150 * 0x0008 - detail - check_table
151 * 0x0010 - detail - check_harder
152 * 0x0020 - detail - check_harder/zapped
153 * 0x0040 - automatically ignore setgroups and afs_syscall
155 static int probe_debug = 0x41;
157 module_param(probe_debug, int, 0);
159 MODULE_PARM(probe_debug, "i");
161 MODULE_PARM_DESC(probe_debug, "Debugging level");
163 static unsigned long probe_debug_addr[4] = { 0,0,0,0 };
164 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
165 module_param_array(probe_debug_addr, long, NULL, 0);
167 MODULE_PARM(probe_debug_addr, "1-4l");
169 MODULE_PARM_DESC(probe_debug_addr, "Debug range starting locations");
171 static unsigned long probe_debug_range = 0;
173 module_param(probe_debug_range, long, 0);
175 MODULE_PARM(probe_debug_range, "l");
177 MODULE_PARM_DESC(probe_debug_range, "Debug range length");
179 static unsigned long probe_debug_tag = 0;
181 module_param(probe_debug_tag, long, 0);
183 MODULE_PARM(probe_debug_tag, "l");
185 MODULE_PARM_DESC(probe_debug_tag, "Debugging output start tag");
189 /* Weak references are our friends. They are supported by the in-kernel
190 * linker in Linux 2.6 and by all versions of modutils back to 2.2pre1.
191 * A weak reference not satisified by the kernel will have value zero.
193 * Unfortunately, weak references to functions don't work right on
194 * IA64; specifically, if you actually try to make a call through
195 * such a reference, and the symbol doesn't exist in the kernel, then
196 * the module relocation code will oops. A workaround for this is
197 * probably possible, but the use of kallsyms_* is of limited value,
198 * so I'm not bothing with the effort for now.
199 * -- jhutz, 10-Feb-2005
201 #ifdef OSI_PROBE_KALLSYMS
202 extern int kallsyms_symbol_to_address(char *name, unsigned long *token,
204 unsigned long *mod_start,
205 unsigned long *mod_end,
207 unsigned long *sec_start,
208 unsigned long *sec_end,
210 unsigned long *sym_start,
211 unsigned long *sym_end
212 ) __attribute__((weak));
214 extern int kallsyms_address_to_symbol(unsigned long address,
216 unsigned long *mod_start,
217 unsigned long *mod_end,
219 unsigned long *sec_start,
220 unsigned long *sec_end,
222 unsigned long *sym_start,
223 unsigned long *sym_end
224 ) __attribute__((weak));
227 extern SYSCALLTYPE sys_call_table[] __attribute__((weak));
228 extern SYSCALLTYPE ia32_sys_call_table[] __attribute__((weak));
229 extern SYSCALLTYPE sys_call_table32[] __attribute__((weak));
230 extern SYSCALLTYPE sys_call_table_emu[] __attribute__((weak));
232 extern asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count) __attribute__((weak));
233 extern asmlinkage long sys_close(unsigned int) __attribute__((weak));
234 #if defined(EXPORTED_SYS_CHDIR)
235 extern asmlinkage long sys_chdir(const char *) __attribute__((weak));
237 extern asmlinkage ssize_t sys_write(unsigned int, const char *, size_t) __attribute__((weak));
238 #ifdef AFS_LINUX26_ENV
239 extern asmlinkage long sys_wait4(pid_t, int *, int, struct rusage *) __attribute__((weak));
241 extern asmlinkage long sys_wait4(pid_t, unsigned int *, int, struct rusage *) __attribute__((weak));
243 extern asmlinkage long sys_exit (int) __attribute__((weak));
244 #if defined(EXPORTED_SYS_OPEN)
245 extern asmlinkage long sys_open (const char *, int, int) __attribute__((weak));
247 extern asmlinkage long sys_ioctl(unsigned int, unsigned int, unsigned long) __attribute__((weak));
248 extern rwlock_t tasklist_lock __attribute__((weak));
251 /* Structures used to control probing. We put all the details of which
252 * symbols we're interested in, what syscall functions to look for, etc
253 * into tables, so we can then have a single copy of the functions that
254 * actually do the work.
267 char *symbol; /* symbol name */
268 char *desc; /* description for messages */
269 int offset; /* first syscall number in table */
271 void *weak_answer; /* weak symbol ref */
272 void *parm_answer; /* module parameter answer */
273 void *debug_answer; /* module parameter answer */
274 unsigned long given_answer; /* compiled-in answer, if any */
276 tryctl *trylist; /* array of combinations to try */
278 unsigned long try_sect_sym; /* symbol in section to try scanning */
279 unsigned long try_base; /* default base address for scan */
280 unsigned long try_base_mask; /* base address bits to force to zero */
281 unsigned long try_length; /* default length for scan */
283 unsigned long alt_try_sect_sym; /* symbol in section to try scanning */
284 unsigned long alt_try_base; /* default base address for scan */
285 unsigned long alt_try_base_mask; /* base address bits to force to zero */
286 unsigned long alt_try_length; /* default length for scan */
288 int n_zapped_syscalls; /* number of unimplemented system calls */
289 int *zapped_syscalls; /* list of unimplemented system calls */
291 int n_unique_syscalls; /* number of unique system calls */
292 int *unique_syscalls; /* list of unimplemented system calls */
294 int verifyNR; /* syscall number to verify match */
295 void *verify_fn; /* syscall pointer to verify match */
297 int debug_ignore_NR[4]; /* syscalls to ignore for debugging */
302 /********** Probing Configuration: sys_call_table **********/
304 /* syscall pairs/triplets to probe */
305 /* On PPC64 and SPARC64, we need to omit the ones that might match both tables */
306 static tryctl main_try[] = {
307 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
308 #if defined(EXPORTED_SYS_CHDIR)
309 { "scan: close+chdir+write", __NR_close, &sys_close, __NR_chdir, &sys_chdir, __NR_write, &sys_write },
312 { "scan: close+wait4", __NR_close, &sys_close, __NR_wait4, &sys_wait4, -1, 0 },
313 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
314 #if defined(EXPORTED_SYS_CHDIR)
315 { "scan: close+chdir", __NR_close, &sys_close, __NR_chdir, &sys_chdir, -1, 0 },
318 { "scan: close+ioctl", __NR_close, &sys_close, __NR_ioctl, &sys_ioctl, -1, 0 },
319 #if defined(EXPORTED_SYS_OPEN)
320 { "scan: exit+open", __NR_exit, &sys_exit, __NR_open, &sys_open, -1, 0 },
325 /* zapped syscalls for try_harder */
326 /* this list is based on the table in 'zapped_syscalls' */
328 static int main_zapped_syscalls[] = {
330 * SPARC-Linux uses syscall number mappings chosen to be compatible
331 * with SunOS. So, it doesn't have any of the traditional calls or
332 * the new STREAMS ones. However, there are a number of syscalls
333 * which are SunOS-specific (not implemented on Linux), i386-specific
334 * (not implemented on SPARC-Linux), or implemented only on one of
335 * sparc32 or sparc64. Of course, there are no __NR macros for most
338 * Note that the calls we list here are implemented by sys_nis_syscall,
339 * not by sys_ni_syscall. That means we have to exclude all of the
340 * other entries, or we might get a sys_ni_syscall into the list and
341 * the test would no longer work.
343 #if defined(AFS_SPARC64_LINUX20_ENV)
344 /* mmap2, fstat64, getmsg, putmsg, modify_ldt */
345 56, 63, 151, 152, 218,
346 #elif defined(AFS_SPARC_LINUX20_ENV)
347 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
348 52, 151, 152, 164, 218,
349 #else /* !AFS_SPARC_LINUX20_ENV */
352 * These 7 syscalls are present in the syscall table on most "older"
353 * platforms that use the traditional syscall number mappings. They
354 * are not implemented on any platform.
378 * On s390 and arm (but not arm26), the seven traditional unimplemented
379 * system calls are indeed present and unimplemented. However, the
380 * corresponding __NR macros are not defined, so the tests above fail.
381 * Instead, we just have to know the numbers for these.
383 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
384 /* break, stty, gtty, ftime, prof, lock, mpx */
385 17, 31, 32, 35, 44, 53, 56,
389 * Sadly, some newer platforms like IA64, amd64, and PA-RISC don't have
390 * the traditional numbers, so the list above are not helpful. They
391 * do have entries for getpmsg/putpmsg, which are always unimplemented.
401 * The module-loading mechanism changed in Linux 2.6, and insmod's
402 * loss is our gain: three new unimplemented system calls!
404 #if defined(AFS_LINUX26_ENV)
408 #ifdef __NR_query_module
411 #ifdef __NR_get_kernel_syms
412 __NR_get_kernel_syms,
414 #endif /* AFS_LINUX26_ENV */
417 * On IA64, the old module-loading calls are indeed present and
418 * unimplemented, but the __NR macros are not defined. Again,
419 * we simply have to know their numbers.
421 #ifdef AFS_IA64_LINUX26_ENV
422 /* create_module, query_module, get_kernel_sysms */
426 /* And the same deal for arm (not arm26), if we ever support that. */
428 /* create_module, query_module, get_kernel_sysms */
433 * Alpha-Linux uses syscall number mappings chosen to be compatible
434 * with OSF/1. So, it doesn't have any of the traditional calls or
435 * the new STREAMS ones, but it does have several OSF/1-specific
436 * syscalls which are not implemented on Linux. These don't exist on
437 * any other platform.
439 #ifdef __NR_osf_syscall
442 #ifdef __NR_osf_profil
445 #ifdef __NR_osf_reboot
448 #ifdef __NR_osf_kmodcall
451 #ifdef __NR_osf_old_vtrace
456 * On PPC64, we need a couple more entries to distinguish the two
457 * tables, since the system call numbers are the same and the sets of
458 * unimplemented calls are very similar.
459 * mmap2 and fstat64 are implemented only for 32-bit calls
461 #ifdef AFS_PPC64_LINUX20_ENV
462 /* _mmap2, _fstat64 */
464 #endif /* AFS_PPC64_LINUX20_ENV */
466 /* Similarly for S390X, with lcown16 and fstat64 */
467 #ifdef AFS_S390X_LINUX20_ENV
468 /* lchown16, fstat64 */
471 #endif /* !AFS_SPARC_LINUX20_ENV */
475 /* unique syscalls for try_harder */
476 static int main_unique_syscalls[] = {
477 #if defined(AFS_SPARC64_LINUX24_ENV) || defined(AFS_SPARC_LINUX24_ENV)
479 * On SPARC, we need some additional unique calls to make sure
480 * we don't match the SunOS-compatibility table.
482 __NR_sgetmask, __NR_ssetmask,
484 __NR_exit, __NR_mount, __NR_read, __NR_write,
485 __NR_open, __NR_close, __NR_unlink
488 /* probe control structure */
489 static probectl main_probe = {
490 /* symbol name and description */
494 /* syscall number of first entry in table */
495 #ifdef AFS_IA64_LINUX20_ENV
501 sys_call_table, /* weak symbol ref */
502 0, 0, /* module parameter answers */
503 #ifdef AFS_LINUX_sys_call_table
504 AFS_LINUX_sys_call_table, /* compiled-in answer, if any */
509 main_try, /* array of combinations to try */
511 /* symbol in section to try scanning */
512 #if defined(AFS_SPARC64_LINUX20_ENV) || defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
513 (unsigned long)&sys_close,
514 #elif defined(AFS_AMD64_LINUX20_ENV)
515 /* On this platform, it's in a different section! */
516 (unsigned long)&tasklist_lock,
518 (unsigned long)&init_mm,
521 /* default base address for scan */
522 /* base address bits to force to zero */
523 /* default length for scan */
524 #if defined(AFS_SPARC64_LINUX20_ENV)
525 (unsigned long)(&sys_close),
528 #elif defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
529 /* bleah; this is so suboptimal */
530 (unsigned long)(&sys_close),
533 #elif defined(AFS_IA64_LINUX20_ENV)
534 (unsigned long)(&init_mm),
537 #elif defined(AFS_AMD64_LINUX20_ENV)
538 (unsigned long)(&tasklist_lock) - 0x30000,
541 #elif defined(AFS_PPC64_LINUX26_ENV)
542 (unsigned long)(&do_signal),
545 #elif defined(AFS_PPC_LINUX20_ENV) || defined(AFS_PPC_LINUX20_ENV)
546 (unsigned long)&init_mm,
550 (unsigned long)&init_mm,
555 #ifdef AFS_LINUX26_ENV
556 (unsigned long)scsi_command_size,
557 (unsigned long)scsi_command_size,
564 /* number and list of unimplemented system calls */
565 ((sizeof(main_zapped_syscalls)/sizeof(main_zapped_syscalls[0])) - 1),
566 main_zapped_syscalls,
568 /* number and list of unique system calls */
569 (sizeof(main_unique_syscalls)/sizeof(main_unique_syscalls[0])),
570 main_unique_syscalls,
572 /* syscall number and pointer to verify match */
573 __NR_close, &sys_close,
575 /* syscalls to ignore for debugging */
577 #if defined(AFS_ALPHA_LINUX20_ENV)
579 #elif defined(AFS_AMD64_LINUX20_ENV)
581 #elif defined(AFS_IA64_LINUX20_ENV)
583 #elif defined(AFS_SPARC_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
589 #ifdef __NR_setgroups32
599 /********** Probing Configuration: amd64 ia32_sys_call_table **********/
600 #if defined(AFS_AMD64_LINUX20_ENV)
602 /* syscall pairs/triplets to probe */
603 static tryctl ia32_try[] = {
604 #if defined(EXPORTED_SYS_CHDIR)
605 { "scan: close+chdir+write", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, __NR_ia32_write, &sys_write },
606 { "scan: close+chdir", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, -1, 0 },
611 /* zapped syscalls for try_harder */
612 static int ia32_zapped_syscalls[] = {
613 __NR_ia32_break, __NR_ia32_stty, __NR_ia32_gtty, __NR_ia32_ftime,
614 __NR_ia32_prof, __NR_ia32_lock, __NR_ia32_mpx,
618 /* unique syscalls for try_harder */
619 static int ia32_unique_syscalls[] = {
620 __NR_ia32_exit, __NR_ia32_mount, __NR_ia32_read, __NR_ia32_write,
621 __NR_ia32_open, __NR_ia32_close, __NR_ia32_unlink
624 /* probe control structure */
625 static probectl ia32_probe = {
626 /* symbol name and description */
627 "ia32_sys_call_table",
628 "32-bit system call table",
630 /* syscall number of first entry in table */
633 ia32_sys_call_table, /* weak symbol ref */
634 0, 0, /* module parameter answers */
635 #ifdef AFS_LINUX_ia32_sys_call_table
636 AFS_LINUX_ia32_sys_call_table,/* compiled-in answer, if any */
641 ia32_try, /* array of combinations to try */
643 /* symbol in section to try scanning */
644 (unsigned long)&init_mm,
646 /* default base address for scan */
647 /* base address bits to force to zero */
648 /* default length for scan */
649 (unsigned long)&init_mm,
651 (0x180000 / sizeof(unsigned long *)),
653 #ifdef AFS_LINUX26_ENV
654 (unsigned long)scsi_command_size,
655 (unsigned long)scsi_command_size,
663 /* number and list of unimplemented system calls */
664 ((sizeof(ia32_zapped_syscalls)/sizeof(ia32_zapped_syscalls[0])) - 1),
665 ia32_zapped_syscalls,
667 /* number and list of unique system calls */
668 (sizeof(ia32_unique_syscalls)/sizeof(ia32_unique_syscalls[0])),
669 ia32_unique_syscalls,
671 /* syscall number and pointer to verify match */
672 __NR_ia32_close, &sys_close,
674 /* syscalls to ignore for debugging */
678 __NR_ia32_setgroups32,
683 static probectl *probe_list[] = {
684 &main_probe, &ia32_probe
688 /********** Probing Configuration: IA64 **********/
689 #elif defined(AFS_IA64_LINUX20_ENV)
695 /* no 32-bit support on IA64 for now */
696 static probectl *probe_list[] = {
701 /********** Probing Configuration: ppc64, sparc64 sys_call_table32 **********/
702 #elif defined(AFS_PPC64_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
709 * syscall pairs/triplets to probe
710 * This has to be empty, because anything that would work will
711 * also match the main table, and that's no good.
713 static tryctl sct32_try[] = {
717 /* zapped syscalls for try_harder */
718 static int sct32_zapped_syscalls[] = {
719 #ifdef AFS_PPC64_LINUX20_ENV
720 /* These should be sufficient */
721 __NR_break, __NR_stty, __NR_gtty, __NR_ftime,
722 __NR_prof, __NR_lock, __NR_mpx,
724 #ifdef AFS_SPARC64_LINUX20_ENV
725 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
726 52, 151, 152, 164, 218,
731 /* unique syscalls for try_harder */
732 /* mmap2 and fstat64 are implemented only for 32-bit calls */
733 static int sct32_unique_syscalls[] = {
734 #ifdef AFS_PPC64_LINUX20_ENV
735 /* _mmap2, _fstat64 */
738 #ifdef AFS_SPARC64_LINUX24_ENV
740 * On SPARC, we need some additional unique calls to make sure
741 * we don't match the SunOS-compatibility table.
743 __NR_sgetmask, __NR_ssetmask,
745 __NR_exit, __NR_mount, __NR_read, __NR_write,
746 __NR_open, __NR_close, __NR_unlink
749 /* probe control structure */
750 static probectl sct32_probe = {
751 /* symbol name and description */
753 "32-bit system call table",
755 /* syscall number of first entry in table */
758 sys_call_table32, /* weak symbol ref */
759 0, 0, /* module parameter answers */
760 #ifdef AFS_LINUX_sys_call_table32
761 AFS_LINUX_sys_call_table32, /* compiled-in answer, if any */
766 sct32_try, /* array of combinations to try */
768 /* symbol in section to try scanning */
769 #if defined(AFS_SPARC64_LINUX20_ENV)
770 (unsigned long)&sys_close,
772 (unsigned long)&init_mm,
775 /* default base address for scan */
776 /* base address bits to force to zero */
777 /* default length for scan */
778 #if defined(AFS_SPARC64_LINUX20_ENV)
779 (unsigned long)(&sys_close),
782 #elif defined(AFS_PPC64_LINUX26_ENV)
783 (unsigned long)(&do_signal),
787 (unsigned long)&init_mm,
792 #ifdef AFS_LINUX26_ENV
793 (unsigned long)scsi_command_size,
794 (unsigned long)scsi_command_size,
801 /* number and list of unimplemented system calls */
802 ((sizeof(sct32_zapped_syscalls)/sizeof(sct32_zapped_syscalls[0])) - 1),
803 sct32_zapped_syscalls,
805 /* number and list of unique system calls */
806 (sizeof(sct32_unique_syscalls)/sizeof(sct32_unique_syscalls[0])),
807 sct32_unique_syscalls,
809 /* syscall number and pointer to verify match */
810 __NR_close, &sys_close,
812 /* syscalls to ignore for debugging */
814 #if defined(AFS_SPARC64_LINUX20_ENV)
825 static probectl *probe_list[] = {
826 &main_probe, &sct32_probe
830 /********** Probing Configuration: s390x sys_call_table_emu **********/
831 /* We only actually need to do this on s390x_linux26 and later.
832 * On earlier versions, the two tables were interleaved and so
833 * have related base addresses.
835 #elif defined(AFS_S390X_LINUX26_ENV)
837 /* syscall pairs/triplets to probe */
838 /* nothing worthwhile is exported, so this is empty */
839 static tryctl emu_try[] = {
843 /* zapped syscalls for try_harder */
844 static int emu_zapped_syscalls[] = {
845 /* break, stty, gtty, ftime, prof, lock, mpx */
846 17, 31, 32, 35, 44, 53, 56,
850 /* unique syscalls for try_harder */
851 static int emu_unique_syscalls[] = {
852 /* lchown16, fstat64 */
854 __NR_exit, __NR_mount, __NR_read, __NR_write,
855 __NR_open, __NR_close, __NR_unlink
858 /* probe control structure */
859 static probectl emu_probe = {
860 /* symbol name and description */
861 "sys_call_table_emu",
862 "32-bit system call table",
864 /* syscall number of first entry in table */
867 sys_call_table_emu, /* weak symbol ref */
868 0, 0, /* module parameter answers */
869 #ifdef AFS_LINUX_sys_call_table_emu
870 AFS_LINUX_sys_call_table_emu, /* compiled-in answer, if any */
875 emu_try, /* array of combinations to try */
877 /* symbol in section to try scanning */
878 (unsigned long)&sys_close,
880 /* default base address for scan */
881 /* base address bits to force to zero */
882 /* default length for scan */
883 (unsigned long)&sys_close,
887 #ifdef AFS_LINUX26_ENV
888 (unsigned long)scsi_command_size,
889 (unsigned long)scsi_command_size,
896 /* number and list of unimplemented system calls */
897 ((sizeof(emu_zapped_syscalls)/sizeof(emu_zapped_syscalls[0])) - 1),
900 /* number and list of unique system calls */
901 (sizeof(emu_unique_syscalls)/sizeof(emu_unique_syscalls[0])),
904 /* syscall number and pointer to verify match */
905 __NR_close, &sys_close,
907 /* syscalls to ignore for debugging */
916 static probectl *probe_list[] = {
917 &main_probe, &emu_probe
921 /********** End of Probing Configuration **********/
923 #else /* no per-platform probe control, so use the default list */
924 static probectl *probe_list[] = {
929 #define N_PROBE_LIST (sizeof(probe_list) / sizeof(*probe_list))
930 #define DEBUG_IN_RANGE(P,x) (!probe_debug_range || \
931 (P->debug_answer && \
932 (unsigned long)(x) >= (unsigned long)P->debug_answer && \
933 (unsigned long)(x) < (unsigned long)P->debug_answer + probe_debug_range))
937 static int check_table(probectl *P, PROBETYPE *ptr)
942 for (x = ptr, i = 0; i < _SS(NR_syscalls); i++, x++) {
943 #ifdef OSI_PROBE_DEBUG
944 if (probe_debug & 0x0040) {
945 for (j = 0; j < 4; j++) {
946 if (_SS(P->debug_ignore_NR[j]) == _SX(i + P->offset)) break;
951 for (j = 0; j < 8; j++) {
952 if (_SS(probe_ignore_syscalls[j]) == _SX(i) + P->offset) break;
955 if (*x <= ktxt_lower_bound) {
956 #ifdef OSI_PROBE_DEBUG
957 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
958 printk("<7>check 0x%lx -> %d [0x%lx]\n",
959 (unsigned long)ptr, i, (unsigned long)*x);
964 #ifdef OSI_PROBE_DEBUG
965 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
966 printk("<7>check 0x%lx -> ok\n", (unsigned long)ptr);
971 static void *try(probectl *P, tryctl *T, PROBETYPE *aptr,
972 unsigned long datalen)
974 #ifdef OSI_PROBE_KALLSYMS
975 char *mod_name, *sec_name, *sym_name;
976 unsigned long mod_start, mod_end;
977 unsigned long sec_start, sec_end;
978 unsigned long sym_start, sym_end;
980 unsigned long offset, ip1, ip2, ip3;
984 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
985 ip1 = T->fn1 ? (unsigned long)((struct fptr *)T->fn1)->ip : 0;
986 ip2 = T->fn2 ? (unsigned long)((struct fptr *)T->fn2)->ip : 0;
987 ip3 = T->fn3 ? (unsigned long)((struct fptr *)T->fn3)->ip : 0;
989 ip1 = (unsigned long)T->fn1;
990 ip2 = (unsigned long)T->fn2;
991 ip3 = (unsigned long)T->fn3;
994 #ifdef OSI_PROBE_DEBUG
995 if (probe_debug & 0x0001)
996 printk("<7>osi_probe: %s %s (%d->0x%lx, %d->0x%lx, %d->0x%lx)\n",
997 P->symbol, T->name, T->NR1, ip1, T->NR2, ip2, T->NR3, ip3);
1000 if (!ip1 || !ip2 || (T->NR3 >= 0 && !ip3))
1003 for (offset = 0; offset < datalen; offset++, aptr++) {
1004 #if defined(AFS_PPC64_LINUX20_ENV)
1005 ptr = (PROBETYPE*)(*aptr);
1006 if ((unsigned long)ptr <= KERNELBASE) {
1012 if ((unsigned long)ptr < init_mm.start_code ||
1013 (unsigned long)ptr > init_mm.end_data) {
1014 /* printk("address 0x%lx (from 0x%lx %d) is out of range in check_table. wtf?\n", (unsigned long)x, (unsigned long)ptr, i);*/
1018 ret = check_table(P, ptr);
1020 /* return value is number of entries to skip */
1026 #ifdef OSI_PROBE_DEBUG
1027 if ((probe_debug & 0x0002) && DEBUG_IN_RANGE(P,ptr))
1028 printk("<7>try 0x%lx\n", (unsigned long)ptr);
1030 if (ptr[_SS(T->NR1 - P->offset)] != ip1) continue;
1031 if (ptr[_SS(T->NR2 - P->offset)] != ip2) continue;
1032 if (ip3 && ptr[_SS(T->NR3 - P->offset)] != ip3) continue;
1034 #ifdef OSI_PROBE_DEBUG
1035 if (probe_debug & 0x0002)
1036 printk("<7>try found 0x%lx\n", (unsigned long)ptr);
1038 #ifdef OSI_PROBE_KALLSYMS
1039 if (kallsyms_address_to_symbol) {
1040 ret = kallsyms_address_to_symbol((unsigned long)ptr,
1041 &mod_name, &mod_start, &mod_end,
1042 &sec_name, &sec_start, &sec_end,
1043 &sym_name, &sym_start, &sym_end);
1044 if (!ret || strcmp(sym_name, P->symbol)) continue;
1047 /* XXX should we make sure there is only one match? */
1054 static int check_harder(probectl *P, PROBETYPE *p)
1059 /* Check zapped syscalls */
1060 for (i = 1; i < P->n_zapped_syscalls; i++) {
1061 if (p[_SS(P->zapped_syscalls[i])] != p[_SS(P->zapped_syscalls[0])]) {
1062 #ifdef OSI_PROBE_DEBUG
1063 if ((probe_debug & 0x0020) && DEBUG_IN_RANGE(P,p))
1064 printk("<7>check_harder 0x%lx zapped failed i=%d\n", (unsigned long)p, i);
1070 /* Check unique syscalls */
1071 for (i = 0; i < P->n_unique_syscalls; i++) {
1072 for (s = 0; s < NR_syscalls; s++) {
1073 if (p[_SS(s)] == p[_SS(P->unique_syscalls[i])]
1074 && s != P->unique_syscalls[i]) {
1075 #ifdef OSI_PROBE_DEBUG
1076 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1077 printk("<7>check_harder 0x%lx unique failed i=%d s=%d\n", (unsigned long)p, i, s);
1084 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
1085 ip1 = P->verify_fn ? (unsigned long)((struct fptr *)(P->verify_fn))->ip : 0;
1087 ip1 = (unsigned long)(P->verify_fn);
1090 if (ip1 && p[_SS(P->verifyNR - P->offset)] != ip1) {
1091 #ifdef OSI_PROBE_DEBUG
1092 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1093 printk("<7>check_harder 0x%lx verify failed\n", (unsigned long)p);
1098 #ifdef OSI_PROBE_DEBUG
1099 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1100 printk("<7>check_harder 0x%lx success!\n", (unsigned long)p);
1105 static void *try_harder(probectl *P, PROBETYPE *ptr, unsigned long datalen)
1107 #ifdef OSI_PROBE_KALLSYMS
1108 char *mod_name, *sec_name, *sym_name;
1109 unsigned long mod_start, mod_end;
1110 unsigned long sec_start, sec_end;
1111 unsigned long sym_start, sym_end;
1113 unsigned long offset;
1117 #ifdef OSI_PROBE_DEBUG
1118 if (probe_debug & 0x0001)
1119 printk("<7>osi_probe: %s try_harder\n", P->symbol);
1121 for (offset = 0; offset < datalen; offset++, ptr++) {
1122 if ((unsigned long)ptr < init_mm.start_code ||
1123 (unsigned long)ptr > init_mm.end_data) {
1124 /* printk("address 0x%lx (from 0x%lx %d) is out of range in check_table. wtf?\n", (unsigned long)x, (unsigned long)ptr, i);*/
1127 ret = check_table(P, ptr);
1129 /* return value is number of entries to skip */
1135 #ifdef OSI_PROBE_DEBUG
1136 if ((probe_debug & 0x0004) && DEBUG_IN_RANGE(P,ptr))
1137 printk("<7>try_harder 0x%lx\n", (unsigned long)ptr);
1139 if (!check_harder(P, ptr))
1142 #ifdef OSI_PROBE_DEBUG
1143 if (probe_debug & 0x0004)
1144 printk("<7>try_harder found 0x%lx\n", (unsigned long)ptr);
1147 #ifdef OSI_PROBE_KALLSYMS
1148 if (kallsyms_address_to_symbol) {
1149 ret = kallsyms_address_to_symbol((unsigned long)ptr,
1150 &mod_name, &mod_start, &mod_end,
1151 &sec_name, &sec_start, &sec_end,
1152 &sym_name, &sym_start, &sym_end);
1153 if (!ret || strcmp(sym_name, P->symbol)) continue;
1158 #ifdef OSI_PROBE_DEBUG
1159 if (probe_debug & 0x0005)
1160 printk("<7>%s: try_harder found multiple matches!\n", P->symbol);
1165 match = (void *)ptr;
1166 if (!probe_carefully)
1173 #ifdef OSI_PROBE_DEBUG
1174 #define check_result(x,m) do { \
1175 if (probe_debug & 0x0001) { \
1176 printk("<7>osi_probe: %s = 0x%016lx %s\n", P->symbol, (unsigned long)(x), (m)); \
1180 final_answer = (void *)(x); \
1184 #define check_result(x,m) do { \
1187 return (void *)(x); \
1191 static void *scan_for_syscall_table(probectl *P, PROBETYPE *B, unsigned long L)
1195 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1198 #ifdef OSI_PROBE_DEBUG
1199 void *final_answer = 0;
1201 #ifdef OSI_PROBE_DEBUG
1202 if (probe_debug & 0x0007)
1203 printk("<7>osi_probe: %s base=0x%lx, len=0x%lx\n",
1204 P->symbol, (unsigned long)B, L);
1205 if (probe_debug & 0x0009) {
1206 printk("<7>osi_probe: %s ktxt_lower_bound=0x%lx\n",
1207 P->symbol, ktxt_lower_bound);
1208 printk("<7>osi_probe: %s NR_syscalls=%d\n",
1209 P->symbol, NR_syscalls);
1213 for (T = P->trylist; T->name; T++) {
1214 answer = try(P, T, B, L);
1215 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1216 answer2 = try(P, T, (PROBETYPE *)(2 + (void *)B), L);
1217 #ifdef OSI_PROBE_DEBUG
1218 if (probe_debug & 0x0003) {
1219 printk("<7>osi_probe: %s = 0x%016lx %s (even)\n",
1220 P->symbol, (unsigned long)(answer), T->name);
1221 printk("<7>osi_probe: %s = 0x%016lx %s (odd)\n",
1222 P->symbol, (unsigned long)(answer2), T->name);
1225 if (answer && answer2) answer = 0;
1226 else if (answer2) answer = answer2;
1232 /* XXX more checks here */
1234 answer = try_harder(P, B, L);
1235 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
1236 answer2 = try_harder(P, (PROBETYPE *)(2 + (void *)B), L);
1237 #ifdef OSI_PROBE_DEBUG
1238 if (probe_debug & 0x0005) {
1239 printk("<7>osi_probe: %s = 0x%016lx pattern scan (even)\n",
1240 P->symbol, (unsigned long)(answer));
1241 printk("<7>osi_probe: %s = 0x%016lx pattern scan (odd)\n",
1242 P->symbol, (unsigned long)(answer2));
1245 if (answer && answer2) answer = 0;
1246 else if (answer2) answer = answer2;
1251 static void *do_find_syscall_table(probectl *P, char **method)
1253 #ifdef OSI_PROBE_KALLSYMS
1254 char *mod_name, *sec_name, *sym_name;
1255 unsigned long mod_start, mod_end;
1256 unsigned long sec_start, sec_end;
1257 unsigned long sym_start, sym_end;
1258 unsigned long token;
1264 #ifdef OSI_PROBE_DEBUG
1265 void *final_answer = 0;
1268 *method = "not found";
1270 /* if it's exported, there's nothing to do */
1271 check_result(P->weak_answer, "exported");
1273 /* ask the kernel to do the name lookup, if it's willing */
1274 #ifdef OSI_PROBE_KALLSYMS
1275 if (kallsyms_symbol_to_address) {
1279 ret = kallsyms_symbol_to_address(P->symbol, &token,
1280 &mod_name, &mod_start, &mod_end,
1281 &sec_name, &sec_start, &sec_end,
1282 &sym_name, &sym_start, &sym_end);
1283 if (ret && !strcmp(mod_name, "kernel") && sym_start)
1287 check_result(sym_start, "kallsyms_symbol_to_address");
1291 /* Maybe a little birdie told us */
1292 check_result(P->parm_answer, "module parameter");
1293 check_result(P->given_answer, "compiled-in");
1295 /* OK, so we have to scan. */
1296 B = (PROBETYPE *)((P->try_base) & ~(P->try_base_mask));
1298 /* Now, see if the kernel will tell us something better than the default */
1299 #ifdef OSI_PROBE_KALLSYMS
1300 if (kallsyms_address_to_symbol) {
1301 ret = kallsyms_address_to_symbol(P->try_sect_sym,
1302 &mod_name, &mod_start, &mod_end,
1303 &sec_name, &sec_start, &sec_end,
1304 &sym_name, &sym_start, &sym_end);
1306 B = (PROBETYPE *)sec_start;
1307 L = (sec_end - sec_start) / sizeof(unsigned long);
1312 answer = scan_for_syscall_table(P, B, L);
1313 check_result(answer, "pattern scan");
1314 B = (PROBETYPE *)((P->alt_try_base) & ~(P->alt_try_base_mask));
1315 L = P->alt_try_length;
1316 /* Now, see if the kernel will tell us something better than the default */
1317 #ifdef OSI_PROBE_KALLSYMS
1318 if (kallsyms_address_to_symbol && P->alt_try_sect_sym) {
1319 ret = kallsyms_address_to_symbol(P->alt_try_sect_sym,
1320 &mod_name, &mod_start, &mod_end,
1321 &sec_name, &sec_start, &sec_end,
1322 &sym_name, &sym_start, &sym_end);
1324 B = (PROBETYPE *)sec_start;
1325 L = (sec_end - sec_start) / sizeof(unsigned long);
1330 answer = scan_for_syscall_table(P, B, L);
1331 check_result(answer, "pattern scan");
1333 #ifdef OSI_PROBE_DEBUG
1334 return final_answer;
1340 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1341 static int check_writable(unsigned long address)
1343 pgd_t *pgd = pgd_offset_k(address);
1353 pud = pud_offset(pgd, address);
1356 pmd = pmd_offset(pud, address);
1358 pmd = pmd_offset(pgd, address);
1362 if (pmd_large(*pmd))
1365 pte = pte_offset_kernel(pmd, address);
1366 if (pte_none(*pte) || !pte_present(*pte) || !pte_write(*pte))
1372 void *osi_find_syscall_table(int which)
1378 if (which < 0 || which >= N_PROBE_LIST) {
1379 printk("error - afs_find_syscall_table called with invalid index!\n");
1382 P = probe_list[which];
1384 P->parm_answer = (void *)sys_call_table_addr[which];
1385 #ifdef OSI_PROBE_DEBUG
1386 P->debug_answer = (void *)probe_debug_addr[which];
1389 answer = do_find_syscall_table(P, &method);
1391 printk("Warning: failed to find address of %s\n", P->desc);
1392 printk("System call hooks will not be installed; proceeding anyway\n");
1395 printk("Found %s at 0x%lx (%s)\n", P->desc, (unsigned long)answer, method);
1396 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1397 if (!check_writable((unsigned long)answer)) {
1398 printk("Address 0x%lx is not writable.\n", (unsigned long)answer);
1399 printk("System call hooks will not be installed; proceeding anyway\n");
1407 #ifdef OSI_PROBE_STANDALONE
1408 int __init osi_probe_init(void)
1412 if (!probe_debug_tag) probe_debug_tag = jiffies;
1413 printk("*** osi_probe %ld debug = 0x%04x ***\n",
1414 probe_debug_tag, probe_debug);
1415 for (i = 0; i < N_PROBE_LIST; i++)
1416 (void)osi_find_syscall_table(i);
1420 void osi_probe_exit(void) { }
1422 module_init(osi_probe_init);
1423 module_exit(osi_probe_exit);