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"
56 #if defined(ENABLE_LINUX_SYSCALL_PROBING)
57 #include <linux/module.h> /* early to avoid printf->printk mapping */
58 #include <scsi/scsi.h> /* for scsi_command_size */
59 #ifndef OSI_PROBE_STANDALONE
60 # include "afs/sysincludes.h"
61 # include "afsincludes.h"
63 #include <linux/version.h>
64 #include <linux/sched.h>
65 #ifdef HAVE_LINUX_CONFIG_H
66 # include <linux/config.h>
68 #include <linux/linkage.h>
69 #include <linux/init.h>
70 #include <linux/unistd.h>
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;
131 #if defined(module_param) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
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
156 * 0x0080 - detail - check_table_readable
158 static int probe_debug = 0x41;
159 #if defined(module_param) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
160 module_param(probe_debug, int, 0);
162 MODULE_PARM(probe_debug, "i");
164 MODULE_PARM_DESC(probe_debug, "Debugging level");
166 static unsigned long probe_debug_addr[4] = { 0,0,0,0 };
167 #if defined(module_param_array) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
168 module_param_array(probe_debug_addr, long, NULL, 0);
170 MODULE_PARM(probe_debug_addr, "1-4l");
172 MODULE_PARM_DESC(probe_debug_addr, "Debug range starting locations");
174 static unsigned long probe_debug_range = 0;
175 #if defined(module_param) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
176 module_param(probe_debug_range, long, 0);
178 MODULE_PARM(probe_debug_range, "l");
180 MODULE_PARM_DESC(probe_debug_range, "Debug range length");
182 static unsigned long probe_debug_tag = 0;
183 #if defined(module_param) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
184 module_param(probe_debug_tag, long, 0);
186 MODULE_PARM(probe_debug_tag, "l");
188 MODULE_PARM_DESC(probe_debug_tag, "Debugging output start tag");
192 /* Weak references are our friends. They are supported by the in-kernel
193 * linker in Linux 2.6 and by all versions of modutils back to 2.2pre1.
194 * A weak reference not satisified by the kernel will have value zero.
196 * Unfortunately, weak references to functions don't work right on
197 * IA64; specifically, if you actually try to make a call through
198 * such a reference, and the symbol doesn't exist in the kernel, then
199 * the module relocation code will oops. A workaround for this is
200 * probably possible, but the use of kallsyms_* is of limited value,
201 * so I'm not bothing with the effort for now.
202 * -- jhutz, 10-Feb-2005
204 #ifdef OSI_PROBE_KALLSYMS
205 extern int kallsyms_symbol_to_address(char *name, unsigned long *token,
207 unsigned long *mod_start,
208 unsigned long *mod_end,
210 unsigned long *sec_start,
211 unsigned long *sec_end,
213 unsigned long *sym_start,
214 unsigned long *sym_end
215 ) __attribute__((weak));
217 extern int kallsyms_address_to_symbol(unsigned long address,
219 unsigned long *mod_start,
220 unsigned long *mod_end,
222 unsigned long *sec_start,
223 unsigned long *sec_end,
225 unsigned long *sym_start,
226 unsigned long *sym_end
227 ) __attribute__((weak));
230 extern SYSCALLTYPE sys_call_table[] __attribute__((weak));
231 extern SYSCALLTYPE ia32_sys_call_table[] __attribute__((weak));
232 extern SYSCALLTYPE sys_call_table32[] __attribute__((weak));
233 extern SYSCALLTYPE sys_call_table_emu[] __attribute__((weak));
235 extern asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count) __attribute__((weak));
236 extern asmlinkage long sys_close(unsigned int) __attribute__((weak));
237 #if defined(EXPORTED_SYS_CHDIR)
238 extern asmlinkage long sys_chdir(const char *) __attribute__((weak));
240 extern asmlinkage ssize_t sys_write(unsigned int, const char *, size_t) __attribute__((weak));
241 extern asmlinkage long sys_wait4(pid_t, int *, int, struct rusage *) __attribute__((weak));
242 extern asmlinkage long sys_exit (int) __attribute__((weak));
243 #if defined(EXPORTED_SYS_OPEN)
244 extern asmlinkage long sys_open (const char *, int, int) __attribute__((weak));
246 extern asmlinkage long sys_ioctl(unsigned int, unsigned int, unsigned long) __attribute__((weak));
249 /* Structures used to control probing. We put all the details of which
250 * symbols we're interested in, what syscall functions to look for, etc
251 * into tables, so we can then have a single copy of the functions that
252 * actually do the work.
265 char *symbol; /* symbol name */
266 char *desc; /* description for messages */
267 int offset; /* first syscall number in table */
269 void *weak_answer; /* weak symbol ref */
270 void *parm_answer; /* module parameter answer */
271 void *debug_answer; /* module parameter answer */
272 unsigned long given_answer; /* compiled-in answer, if any */
274 tryctl *trylist; /* array of combinations to try */
276 unsigned long try_sect_sym; /* symbol in section to try scanning */
277 unsigned long try_base; /* default base address for scan */
278 unsigned long try_base_mask; /* base address bits to force to zero */
279 unsigned long try_length; /* default length for scan */
281 unsigned long alt_try_sect_sym; /* symbol in section to try scanning */
282 unsigned long alt_try_base; /* default base address for scan */
283 unsigned long alt_try_base_mask; /* base address bits to force to zero */
284 unsigned long alt_try_length; /* default length for scan */
286 int n_zapped_syscalls; /* number of unimplemented system calls */
287 int *zapped_syscalls; /* list of unimplemented system calls */
289 int n_unique_syscalls; /* number of unique system calls */
290 int *unique_syscalls; /* list of unimplemented system calls */
292 int verifyNR; /* syscall number to verify match */
293 void *verify_fn; /* syscall pointer to verify match */
295 int debug_ignore_NR[4]; /* syscalls to ignore for debugging */
298 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
299 static int check_access(unsigned long, int);
300 static int check_table_readable(probectl *, PROBETYPE *);
304 /********** Probing Configuration: sys_call_table **********/
306 /* syscall pairs/triplets to probe */
307 /* On PPC64 and SPARC64, we need to omit the ones that might match both tables */
308 static tryctl main_try[] = {
309 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
310 #if defined(EXPORTED_SYS_CHDIR)
311 { "scan: close+chdir+write", __NR_close, &sys_close, __NR_chdir, &sys_chdir, __NR_write, &sys_write },
314 { "scan: close+wait4", __NR_close, &sys_close, __NR_wait4, &sys_wait4, -1, 0 },
315 #if !defined(AFS_PPC64_LINUX20_ENV) && !defined(AFS_SPARC64_LINUX20_ENV)
316 #if defined(EXPORTED_SYS_CHDIR)
317 { "scan: close+chdir", __NR_close, &sys_close, __NR_chdir, &sys_chdir, -1, 0 },
320 { "scan: close+ioctl", __NR_close, &sys_close, __NR_ioctl, &sys_ioctl, -1, 0 },
321 #if defined(EXPORTED_SYS_OPEN)
322 { "scan: exit+open", __NR_exit, &sys_exit, __NR_open, &sys_open, -1, 0 },
327 /* zapped syscalls for try_harder */
328 /* this list is based on the table in 'zapped_syscalls' */
330 static int main_zapped_syscalls[] = {
332 * SPARC-Linux uses syscall number mappings chosen to be compatible
333 * with SunOS. So, it doesn't have any of the traditional calls or
334 * the new STREAMS ones. However, there are a number of syscalls
335 * which are SunOS-specific (not implemented on Linux), i386-specific
336 * (not implemented on SPARC-Linux), or implemented only on one of
337 * sparc32 or sparc64. Of course, there are no __NR macros for most
340 * Note that the calls we list here are implemented by sys_nis_syscall,
341 * not by sys_ni_syscall. That means we have to exclude all of the
342 * other entries, or we might get a sys_ni_syscall into the list and
343 * the test would no longer work.
345 #if defined(AFS_SPARC64_LINUX20_ENV)
346 /* mmap2, fstat64, getmsg, putmsg, modify_ldt */
347 56, 63, 151, 152, 218,
348 #elif defined(AFS_SPARC_LINUX20_ENV)
349 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
350 52, 151, 152, 164, 218,
351 #else /* !AFS_SPARC_LINUX20_ENV */
354 * These 7 syscalls are present in the syscall table on most "older"
355 * platforms that use the traditional syscall number mappings. They
356 * are not implemented on any platform.
380 * On s390 and arm (but not arm26), the seven traditional unimplemented
381 * system calls are indeed present and unimplemented. However, the
382 * corresponding __NR macros are not defined, so the tests above fail.
383 * Instead, we just have to know the numbers for these.
385 #if defined(AFS_S390_LINUX20_ENV) || defined(AFS_S390X_LINUX20_ENV)
386 /* break, stty, gtty, ftime, prof, lock, mpx */
387 17, 31, 32, 35, 44, 53, 56,
391 * Sadly, some newer platforms like IA64, amd64, and PA-RISC don't have
392 * the traditional numbers, so the list above are not helpful. They
393 * do have entries for getpmsg/putpmsg, which are always unimplemented.
403 * The module-loading mechanism changed in Linux 2.6, and insmod's
404 * loss is our gain: three new unimplemented system calls!
409 #ifdef __NR_query_module
412 #ifdef __NR_get_kernel_syms
413 __NR_get_kernel_syms,
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_LINUX26_ENV)
515 /* On this platform, it's in a different section! */
516 (unsigned long)&generic_ro_fops,
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_LINUX26_ENV)
538 (unsigned long)(&generic_ro_fops) - 0x30000,
541 #elif defined(AFS_PPC64_LINUX26_ENV)
542 (unsigned long)(&do_signal),
546 (unsigned long)&init_mm,
551 (unsigned long)scsi_command_size,
552 (unsigned long)scsi_command_size - 0x10000,
556 /* number and list of unimplemented system calls */
557 ((sizeof(main_zapped_syscalls)/sizeof(main_zapped_syscalls[0])) - 1),
558 main_zapped_syscalls,
560 /* number and list of unique system calls */
561 (sizeof(main_unique_syscalls)/sizeof(main_unique_syscalls[0])),
562 main_unique_syscalls,
564 /* syscall number and pointer to verify match */
565 __NR_close, &sys_close,
567 /* syscalls to ignore for debugging */
569 #if defined(AFS_ALPHA_LINUX20_ENV)
571 #elif defined(AFS_AMD64_LINUX20_ENV)
573 #elif defined(AFS_IA64_LINUX20_ENV)
575 #elif defined(AFS_SPARC_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
581 #ifdef __NR_setgroups32
591 /********** Probing Configuration: amd64 ia32_sys_call_table **********/
592 #if defined(AFS_AMD64_LINUX20_ENV)
594 /* syscall pairs/triplets to probe */
595 static tryctl ia32_try[] = {
596 #if defined(EXPORTED_SYS_CHDIR)
597 { "scan: close+chdir+write", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, __NR_ia32_write, &sys_write },
598 { "scan: close+chdir", __NR_ia32_close, &sys_close, __NR_ia32_chdir, &sys_chdir, -1, 0 },
603 /* zapped syscalls for try_harder */
604 static int ia32_zapped_syscalls[] = {
605 __NR_ia32_break, __NR_ia32_stty, __NR_ia32_gtty, __NR_ia32_ftime,
606 __NR_ia32_prof, __NR_ia32_lock, __NR_ia32_mpx,
610 /* unique syscalls for try_harder */
611 static int ia32_unique_syscalls[] = {
612 __NR_ia32_exit, __NR_ia32_mount, __NR_ia32_read, __NR_ia32_write,
613 __NR_ia32_open, __NR_ia32_close, __NR_ia32_unlink
616 /* probe control structure */
617 static probectl ia32_probe = {
618 /* symbol name and description */
619 "ia32_sys_call_table",
620 "32-bit system call table",
622 /* syscall number of first entry in table */
625 ia32_sys_call_table, /* weak symbol ref */
626 0, 0, /* module parameter answers */
627 #ifdef AFS_LINUX_ia32_sys_call_table
628 AFS_LINUX_ia32_sys_call_table,/* compiled-in answer, if any */
633 ia32_try, /* array of combinations to try */
635 /* symbol in section to try scanning */
636 (unsigned long)&init_mm,
638 /* default base address for scan */
639 /* base address bits to force to zero */
640 /* default length for scan */
641 (unsigned long)&init_mm,
643 (0x180000 / sizeof(unsigned long *)),
645 (unsigned long)scsi_command_size,
646 (unsigned long)scsi_command_size - 0x10000,
651 /* number and list of unimplemented system calls */
652 ((sizeof(ia32_zapped_syscalls)/sizeof(ia32_zapped_syscalls[0])) - 1),
653 ia32_zapped_syscalls,
655 /* number and list of unique system calls */
656 (sizeof(ia32_unique_syscalls)/sizeof(ia32_unique_syscalls[0])),
657 ia32_unique_syscalls,
659 /* syscall number and pointer to verify match */
660 __NR_ia32_close, &sys_close,
662 /* syscalls to ignore for debugging */
666 __NR_ia32_setgroups32,
671 static probectl *probe_list[] = {
672 &main_probe, &ia32_probe
676 /********** Probing Configuration: IA64 **********/
677 #elif defined(AFS_IA64_LINUX20_ENV)
683 /* no 32-bit support on IA64 for now */
684 static probectl *probe_list[] = {
689 /********** Probing Configuration: ppc64, sparc64 sys_call_table32 **********/
690 #elif defined(AFS_PPC64_LINUX20_ENV) || defined(AFS_SPARC64_LINUX20_ENV)
697 * syscall pairs/triplets to probe
698 * This has to be empty, because anything that would work will
699 * also match the main table, and that's no good.
701 static tryctl sct32_try[] = {
705 /* zapped syscalls for try_harder */
706 static int sct32_zapped_syscalls[] = {
707 #ifdef AFS_PPC64_LINUX20_ENV
708 /* These should be sufficient */
709 __NR_break, __NR_stty, __NR_gtty, __NR_ftime,
710 __NR_prof, __NR_lock, __NR_mpx,
712 #ifdef AFS_SPARC64_LINUX20_ENV
713 /* memory_ordering, getmsg, putmsg, unimplemented, modify_ldt */
714 52, 151, 152, 164, 218,
719 /* unique syscalls for try_harder */
720 /* mmap2 and fstat64 are implemented only for 32-bit calls */
721 static int sct32_unique_syscalls[] = {
722 #ifdef AFS_PPC64_LINUX20_ENV
723 /* _mmap2, _fstat64 */
726 #ifdef AFS_SPARC64_LINUX24_ENV
728 * On SPARC, we need some additional unique calls to make sure
729 * we don't match the SunOS-compatibility table.
731 __NR_sgetmask, __NR_ssetmask,
733 __NR_exit, __NR_mount, __NR_read, __NR_write,
734 __NR_open, __NR_close, __NR_unlink
737 /* probe control structure */
738 static probectl sct32_probe = {
739 /* symbol name and description */
741 "32-bit system call table",
743 /* syscall number of first entry in table */
746 sys_call_table32, /* weak symbol ref */
747 0, 0, /* module parameter answers */
748 #ifdef AFS_LINUX_sys_call_table32
749 AFS_LINUX_sys_call_table32, /* compiled-in answer, if any */
754 sct32_try, /* array of combinations to try */
756 /* symbol in section to try scanning */
757 #if defined(AFS_SPARC64_LINUX20_ENV)
758 (unsigned long)&sys_close,
760 (unsigned long)&init_mm,
763 /* default base address for scan */
764 /* base address bits to force to zero */
765 /* default length for scan */
766 #if defined(AFS_SPARC64_LINUX20_ENV)
767 (unsigned long)(&sys_close),
770 #elif defined(AFS_PPC64_LINUX26_ENV)
771 (unsigned long)(&do_signal),
775 (unsigned long)&init_mm,
780 (unsigned long)scsi_command_size,
781 (unsigned long)scsi_command_size - 0x10000,
785 /* number and list of unimplemented system calls */
786 ((sizeof(sct32_zapped_syscalls)/sizeof(sct32_zapped_syscalls[0])) - 1),
787 sct32_zapped_syscalls,
789 /* number and list of unique system calls */
790 (sizeof(sct32_unique_syscalls)/sizeof(sct32_unique_syscalls[0])),
791 sct32_unique_syscalls,
793 /* syscall number and pointer to verify match */
794 __NR_close, &sys_close,
796 /* syscalls to ignore for debugging */
798 #if defined(AFS_SPARC64_LINUX20_ENV)
809 static probectl *probe_list[] = {
810 &main_probe, &sct32_probe
814 /********** Probing Configuration: s390x sys_call_table_emu **********/
815 /* We only actually need to do this on s390x_linux26 and later.
816 * On earlier versions, the two tables were interleaved and so
817 * have related base addresses.
819 #elif defined(AFS_S390X_LINUX26_ENV)
821 /* syscall pairs/triplets to probe */
822 /* nothing worthwhile is exported, so this is empty */
823 static tryctl emu_try[] = {
827 /* zapped syscalls for try_harder */
828 static int emu_zapped_syscalls[] = {
829 /* break, stty, gtty, ftime, prof, lock, mpx */
830 17, 31, 32, 35, 44, 53, 56,
834 /* unique syscalls for try_harder */
835 static int emu_unique_syscalls[] = {
836 /* lchown16, fstat64 */
838 __NR_exit, __NR_mount, __NR_read, __NR_write,
839 __NR_open, __NR_close, __NR_unlink
842 /* probe control structure */
843 static probectl emu_probe = {
844 /* symbol name and description */
845 "sys_call_table_emu",
846 "32-bit system call table",
848 /* syscall number of first entry in table */
851 sys_call_table_emu, /* weak symbol ref */
852 0, 0, /* module parameter answers */
853 #ifdef AFS_LINUX_sys_call_table_emu
854 AFS_LINUX_sys_call_table_emu, /* compiled-in answer, if any */
859 emu_try, /* array of combinations to try */
861 /* symbol in section to try scanning */
862 (unsigned long)&sys_close,
864 /* default base address for scan */
865 /* base address bits to force to zero */
866 /* default length for scan */
867 (unsigned long)&sys_close,
871 (unsigned long)scsi_command_size,
872 (unsigned long)scsi_command_size - 0x10000,
876 /* number and list of unimplemented system calls */
877 ((sizeof(emu_zapped_syscalls)/sizeof(emu_zapped_syscalls[0])) - 1),
880 /* number and list of unique system calls */
881 (sizeof(emu_unique_syscalls)/sizeof(emu_unique_syscalls[0])),
884 /* syscall number and pointer to verify match */
885 __NR_close, &sys_close,
887 /* syscalls to ignore for debugging */
896 static probectl *probe_list[] = {
897 &main_probe, &emu_probe
901 /********** End of Probing Configuration **********/
903 #else /* no per-platform probe control, so use the default list */
904 static probectl *probe_list[] = {
909 #define N_PROBE_LIST (sizeof(probe_list) / sizeof(*probe_list))
910 #define DEBUG_IN_RANGE(P,x) (!probe_debug_range || \
911 (P->debug_answer && \
912 (unsigned long)(x) >= (unsigned long)P->debug_answer && \
913 (unsigned long)(x) < (unsigned long)P->debug_answer + probe_debug_range))
917 static int check_table(probectl *P, PROBETYPE *ptr)
922 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
923 i = check_table_readable(P, ptr);
924 if (i >= 0) return i;
927 for (x = ptr, i = 0; i < _SS(NR_syscalls); i++, x++) {
928 #ifdef OSI_PROBE_DEBUG
929 if (probe_debug & 0x0040) {
930 for (j = 0; j < 4; j++) {
931 if (_SS(P->debug_ignore_NR[j]) == _SX(i + P->offset)) break;
936 for (j = 0; j < 8; j++) {
937 if (_SS(probe_ignore_syscalls[j]) == _SX(i) + P->offset) break;
940 if (*x <= ktxt_lower_bound) {
941 #ifdef OSI_PROBE_DEBUG
942 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
943 printk("<7>check 0x%lx -> %d [0x%lx]\n",
944 (unsigned long)ptr, i, (unsigned long)*x);
949 #ifdef OSI_PROBE_DEBUG
950 if ((probe_debug & 0x0008) && DEBUG_IN_RANGE(P,ptr))
951 printk("<7>check 0x%lx -> ok\n", (unsigned long)ptr);
956 static void *try(probectl *P, tryctl *T, PROBETYPE *aptr,
957 unsigned long datalen)
959 #ifdef OSI_PROBE_KALLSYMS
960 char *mod_name, *sec_name, *sym_name;
961 unsigned long mod_start, mod_end;
962 unsigned long sec_start, sec_end;
963 unsigned long sym_start, sym_end;
965 unsigned long offset, ip1, ip2, ip3;
969 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
970 ip1 = T->fn1 ? (unsigned long)((struct fptr *)T->fn1)->ip : 0;
971 ip2 = T->fn2 ? (unsigned long)((struct fptr *)T->fn2)->ip : 0;
972 ip3 = T->fn3 ? (unsigned long)((struct fptr *)T->fn3)->ip : 0;
974 ip1 = (unsigned long)T->fn1;
975 ip2 = (unsigned long)T->fn2;
976 ip3 = (unsigned long)T->fn3;
979 #ifdef OSI_PROBE_DEBUG
980 if (probe_debug & 0x0001)
981 printk("<7>osi_probe: %s %s (%d->0x%lx, %d->0x%lx, %d->0x%lx)\n",
982 P->symbol, T->name, T->NR1, ip1, T->NR2, ip2, T->NR3, ip3);
985 if (!ip1 || !ip2 || (T->NR3 >= 0 && !ip3))
988 for (offset = 0; offset < datalen; offset++, aptr++) {
989 #if defined(AFS_PPC64_LINUX20_ENV)
990 ptr = (PROBETYPE*)(*aptr);
991 if ((unsigned long)ptr <= KERNELBASE) {
997 if ((unsigned long)ptr < init_mm.start_code ||
998 #if defined(AFS_AMD64_LINUX20_ENV)
999 (unsigned long)ptr > init_mm.brk)
1001 (unsigned long)ptr > init_mm.end_data)
1004 /* printk("address 0x%lx (from 0x%lx %d) is out of range in check_table. wtf?\n", (unsigned long)x, (unsigned long)ptr, i);*/
1008 ret = check_table(P, ptr);
1010 /* return value is number of entries to skip */
1016 #ifdef OSI_PROBE_DEBUG
1017 if ((probe_debug & 0x0002) && DEBUG_IN_RANGE(P,ptr))
1018 printk("<7>try 0x%lx\n", (unsigned long)ptr);
1020 if (ptr[_SS(T->NR1 - P->offset)] != ip1) continue;
1021 if (ptr[_SS(T->NR2 - P->offset)] != ip2) continue;
1022 if (ip3 && ptr[_SS(T->NR3 - P->offset)] != ip3) continue;
1024 #ifdef OSI_PROBE_DEBUG
1025 if (probe_debug & 0x0002)
1026 printk("<7>try found 0x%lx\n", (unsigned long)ptr);
1028 #ifdef OSI_PROBE_KALLSYMS
1029 if (kallsyms_address_to_symbol) {
1030 ret = kallsyms_address_to_symbol((unsigned long)ptr,
1031 &mod_name, &mod_start, &mod_end,
1032 &sec_name, &sec_start, &sec_end,
1033 &sym_name, &sym_start, &sym_end);
1034 if (!ret || strcmp(sym_name, P->symbol)) continue;
1037 /* XXX should we make sure there is only one match? */
1044 static int check_harder(probectl *P, PROBETYPE *p)
1049 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1050 i = check_table_readable(P, p);
1051 if (i >= 0) return 0;
1054 /* Check zapped syscalls */
1055 for (i = 1; i < P->n_zapped_syscalls; i++) {
1056 if (p[_SS(P->zapped_syscalls[i])] != p[_SS(P->zapped_syscalls[0])]) {
1057 #ifdef OSI_PROBE_DEBUG
1058 if ((probe_debug & 0x0020) && DEBUG_IN_RANGE(P,p))
1059 printk("<7>check_harder 0x%lx zapped failed i=%d\n", (unsigned long)p, i);
1065 /* Check unique syscalls */
1066 for (i = 0; i < P->n_unique_syscalls; i++) {
1067 for (s = 0; s < NR_syscalls; s++) {
1068 if (p[_SS(s)] == p[_SS(P->unique_syscalls[i])]
1069 && s != P->unique_syscalls[i]) {
1070 #ifdef OSI_PROBE_DEBUG
1071 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1072 printk("<7>check_harder 0x%lx unique failed i=%d s=%d\n", (unsigned long)p, i, s);
1079 #if defined(AFS_IA64_LINUX20_ENV) || defined(AFS_PPC64_LINUX20_ENV)
1080 ip1 = P->verify_fn ? (unsigned long)((struct fptr *)(P->verify_fn))->ip : 0;
1082 ip1 = (unsigned long)(P->verify_fn);
1085 if (ip1 && p[_SS(P->verifyNR - P->offset)] != ip1) {
1086 #ifdef OSI_PROBE_DEBUG
1087 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1088 printk("<7>check_harder 0x%lx verify failed\n", (unsigned long)p);
1093 #ifdef OSI_PROBE_DEBUG
1094 if ((probe_debug & 0x0010) && DEBUG_IN_RANGE(P,p))
1095 printk("<7>check_harder 0x%lx success!\n", (unsigned long)p);
1100 static void *try_harder(probectl *P, PROBETYPE *ptr, unsigned long datalen)
1102 #ifdef OSI_PROBE_KALLSYMS
1103 char *mod_name, *sec_name, *sym_name;
1104 unsigned long mod_start, mod_end;
1105 unsigned long sec_start, sec_end;
1106 unsigned long sym_start, sym_end;
1108 unsigned long offset;
1112 #ifdef OSI_PROBE_DEBUG
1113 if (probe_debug & 0x0001)
1114 printk("<7>osi_probe: %s try_harder\n", P->symbol);
1116 for (offset = 0; offset < datalen; offset++, ptr++) {
1117 if ((unsigned long)ptr < init_mm.start_code ||
1118 #if defined(AFS_AMD64_LINUX20_ENV)
1119 (unsigned long)ptr > init_mm.brk)
1121 (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)); \
1178 if ((x) && ((int)(x)) != -ENOENT) { \
1180 final_answer = (void *)(x); \
1184 #define check_result(x,m) do { \
1185 if ((x) && ((int)(x)) != -ENOENT) { \
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_access(unsigned long address, int mode)
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))
1368 if (mode && !pte_write(*pte))
1373 static int check_table_readable(probectl *P, PROBETYPE *ptr)
1375 PROBETYPE *next_page;
1378 while (i < _SS(NR_syscalls)) {
1379 next_page = (PROBETYPE *)PAGE_ALIGN((unsigned long)(ptr+1));
1380 delta = next_page - ptr;
1381 if (!check_access((unsigned long)ptr, 0)) {
1382 #ifdef OSI_PROBE_DEBUG
1383 if (probe_debug & 0x0080)
1384 printk("<7>osi_probe: %s 0x%016lx not readable; delta=0x%lx\n",
1385 P->symbol, (unsigned long)ptr, delta);
1396 void *osi_find_syscall_table(int which)
1402 if (which < 0 || which >= N_PROBE_LIST) {
1403 printk("error - afs_find_syscall_table called with invalid index!\n");
1406 P = probe_list[which];
1408 P->parm_answer = (void *)sys_call_table_addr[which];
1409 #ifdef OSI_PROBE_DEBUG
1410 P->debug_answer = (void *)probe_debug_addr[which];
1413 answer = do_find_syscall_table(P, &method);
1415 printk("Warning: failed to find address of %s\n", P->desc);
1416 printk("System call hooks will not be installed; proceeding anyway\n");
1419 printk("Found %s at 0x%lx (%s)\n", P->desc, (unsigned long)answer, method);
1420 #if defined(AFS_I386_LINUX26_ENV) || defined(AFS_AMD64_LINUX26_ENV)
1421 if (!check_access((unsigned long)answer, 1)) {
1422 printk("Address 0x%lx is not writable.\n", (unsigned long)answer);
1423 printk("System call hooks will not be installed; proceeding anyway\n");
1431 #ifdef OSI_PROBE_STANDALONE
1432 int __init osi_probe_init(void)
1436 if (!probe_debug_tag) probe_debug_tag = jiffies;
1437 printk("*** osi_probe %ld debug = 0x%04x ***\n",
1438 probe_debug_tag, probe_debug);
1439 for (i = 0; i < N_PROBE_LIST; i++)
1440 (void)osi_find_syscall_table(i);
1444 void osi_probe_exit(void) { }
1446 module_init(osi_probe_init);
1447 module_exit(osi_probe_exit);
1451 void *osi_find_syscall_table(int which)
1455 #endif /* EXPORTED_INIT_MM */