2 * Kprobe module for testing crash dumps
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2006
20 * Author: Ankita Garg <ankita@in.ibm.com>
22 * This module induces system failures at predefined crashpoints to
23 * evaluate the reliability of crash dumps obtained using different dumping
26 * It is adapted from the Linux Kernel Dump Test Tool by
27 * Fernando Luis Vazquez Cao <http://lkdtt.sourceforge.net>
29 * Debugfs support added by Simon Kagstrom <simon.kagstrom@netinsight.net>
31 * See Documentation/fault-injection/provoke-crashes.txt for instructions
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/buffer_head.h>
39 #include <linux/kprobes.h>
40 #include <linux/list.h>
41 #include <linux/init.h>
42 #include <linux/interrupt.h>
43 #include <linux/hrtimer.h>
44 #include <linux/slab.h>
45 #include <scsi/scsi_cmnd.h>
46 #include <linux/debugfs.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mman.h>
49 #include <asm/cacheflush.h>
52 #include <linux/ide.h>
56 * Make sure our attempts to over run the kernel stack doesn't trigger
57 * a compiler warning when CONFIG_FRAME_WARN is set. Then make sure we
58 * recurse past the end of THREAD_SIZE by default.
60 #if defined(CONFIG_FRAME_WARN) && (CONFIG_FRAME_WARN > 0)
61 #define REC_STACK_SIZE (CONFIG_FRAME_WARN / 2)
63 #define REC_STACK_SIZE (THREAD_SIZE / 8)
65 #define REC_NUM_DEFAULT ((THREAD_SIZE / REC_STACK_SIZE) * 2)
67 #define DEFAULT_COUNT 10
72 CN_INT_HARDWARE_ENTRY
,
92 CT_UNALIGNED_LOAD_STORE_WRITE
,
93 CT_OVERWRITE_ALLOCATION
,
106 CT_WRITE_RO_AFTER_INIT
,
110 static char* cp_name
[] = {
111 "INT_HARDWARE_ENTRY",
122 static char* cp_type
[] = {
130 "UNALIGNED_LOAD_STORE_WRITE",
131 "OVERWRITE_ALLOCATION",
144 "WRITE_RO_AFTER_INIT",
148 static struct jprobe lkdtm
;
150 static int lkdtm_parse_commandline(void);
151 static void lkdtm_handler(void);
153 static char* cpoint_name
;
154 static char* cpoint_type
;
155 static int cpoint_count
= DEFAULT_COUNT
;
156 static int recur_count
= REC_NUM_DEFAULT
;
158 static enum cname cpoint
= CN_INVALID
;
159 static enum ctype cptype
= CT_NONE
;
160 static int count
= DEFAULT_COUNT
;
161 static DEFINE_SPINLOCK(count_lock
);
162 static DEFINE_SPINLOCK(lock_me_up
);
164 static u8 data_area
[EXEC_SIZE
];
166 static const unsigned long rodata
= 0xAA55AA55;
167 static unsigned long ro_after_init __ro_after_init
= 0x55AA5500;
169 module_param(recur_count
, int, 0644);
170 MODULE_PARM_DESC(recur_count
, " Recursion level for the stack overflow test");
171 module_param(cpoint_name
, charp
, 0444);
172 MODULE_PARM_DESC(cpoint_name
, " Crash Point, where kernel is to be crashed");
173 module_param(cpoint_type
, charp
, 0444);
174 MODULE_PARM_DESC(cpoint_type
, " Crash Point Type, action to be taken on "\
175 "hitting the crash point");
176 module_param(cpoint_count
, int, 0644);
177 MODULE_PARM_DESC(cpoint_count
, " Crash Point Count, number of times the "\
178 "crash point is to be hit to trigger action");
180 static unsigned int jp_do_irq(unsigned int irq
)
187 static irqreturn_t
jp_handle_irq_event(unsigned int irq
,
188 struct irqaction
*action
)
195 static void jp_tasklet_action(struct softirq_action
*a
)
201 static void jp_ll_rw_block(int rw
, int nr
, struct buffer_head
*bhs
[])
209 static unsigned long jp_shrink_inactive_list(unsigned long max_scan
,
211 struct scan_control
*sc
)
218 static int jp_hrtimer_start(struct hrtimer
*timer
, ktime_t tim
,
219 const enum hrtimer_mode mode
)
226 static int jp_scsi_dispatch_cmd(struct scsi_cmnd
*cmd
)
234 static int jp_generic_ide_ioctl(ide_drive_t
*drive
, struct file
*file
,
235 struct block_device
*bdev
, unsigned int cmd
,
244 /* Return the crashpoint number or NONE if the name is invalid */
245 static enum ctype
parse_cp_type(const char *what
, size_t count
)
249 for (i
= 0; i
< ARRAY_SIZE(cp_type
); i
++) {
250 if (!strcmp(what
, cp_type
[i
]))
257 static const char *cp_type_to_str(enum ctype type
)
259 if (type
== CT_NONE
|| type
< 0 || type
> ARRAY_SIZE(cp_type
))
262 return cp_type
[type
- 1];
265 static const char *cp_name_to_str(enum cname name
)
267 if (name
== CN_INVALID
|| name
< 0 || name
> ARRAY_SIZE(cp_name
))
270 return cp_name
[name
- 1];
274 static int lkdtm_parse_commandline(void)
279 if (cpoint_count
< 1 || recur_count
< 1)
282 spin_lock_irqsave(&count_lock
, flags
);
283 count
= cpoint_count
;
284 spin_unlock_irqrestore(&count_lock
, flags
);
286 /* No special parameters */
287 if (!cpoint_type
&& !cpoint_name
)
290 /* Neither or both of these need to be set */
291 if (!cpoint_type
|| !cpoint_name
)
294 cptype
= parse_cp_type(cpoint_type
, strlen(cpoint_type
));
295 if (cptype
== CT_NONE
)
298 for (i
= 0; i
< ARRAY_SIZE(cp_name
); i
++) {
299 if (!strcmp(cpoint_name
, cp_name
[i
])) {
305 /* Could not find a valid crash point */
309 static int recursive_loop(int remaining
)
311 char buf
[REC_STACK_SIZE
];
313 /* Make sure compiler does not optimize this away. */
314 memset(buf
, (remaining
& 0xff) | 0x1, REC_STACK_SIZE
);
318 return recursive_loop(remaining
- 1);
321 static void do_nothing(void)
326 /* Must immediately follow do_nothing for size calculuations to work out. */
327 static void do_overwritten(void)
329 pr_info("do_overwritten wasn't overwritten!\n");
333 static noinline
void corrupt_stack(void)
335 /* Use default char array length that triggers stack protection. */
338 memset((void *)data
, 0, 64);
341 static void execute_location(void *dst
)
343 void (*func
)(void) = dst
;
345 pr_info("attempting ok execution at %p\n", do_nothing
);
348 memcpy(dst
, do_nothing
, EXEC_SIZE
);
349 flush_icache_range((unsigned long)dst
, (unsigned long)dst
+ EXEC_SIZE
);
350 pr_info("attempting bad execution at %p\n", func
);
354 static void execute_user_location(void *dst
)
356 /* Intentionally crossing kernel/user memory boundary. */
357 void (*func
)(void) = dst
;
359 pr_info("attempting ok execution at %p\n", do_nothing
);
362 if (copy_to_user((void __user
*)dst
, do_nothing
, EXEC_SIZE
))
364 flush_icache_range((unsigned long)dst
, (unsigned long)dst
+ EXEC_SIZE
);
365 pr_info("attempting bad execution at %p\n", func
);
369 static void lkdtm_do_action(enum ctype which
)
389 (void) recursive_loop(recur_count
);
391 case CT_CORRUPT_STACK
:
394 case CT_UNALIGNED_LOAD_STORE_WRITE
: {
395 static u8 data
[5] __attribute__((aligned(4))) = {1, 2,
398 u32 val
= 0x12345678;
400 p
= (u32
*)(data
+ 1);
406 case CT_OVERWRITE_ALLOCATION
: {
408 u32
*data
= kmalloc(len
, GFP_KERNEL
);
410 data
[1024 / sizeof(u32
)] = 0x12345678;
414 case CT_WRITE_AFTER_FREE
: {
416 u32
*data
= kmalloc(len
, GFP_KERNEL
);
420 memset(data
, 0x78, len
);
434 /* Must be called twice to trigger. */
435 spin_lock(&lock_me_up
);
436 /* Let sparse know we intended to exit holding the lock. */
437 __release(&lock_me_up
);
440 set_current_state(TASK_UNINTERRUPTIBLE
);
444 execute_location(data_area
);
446 case CT_EXEC_STACK
: {
447 u8 stack_area
[EXEC_SIZE
];
448 execute_location(stack_area
);
451 case CT_EXEC_KMALLOC
: {
452 u32
*kmalloc_area
= kmalloc(EXEC_SIZE
, GFP_KERNEL
);
453 execute_location(kmalloc_area
);
457 case CT_EXEC_VMALLOC
: {
458 u32
*vmalloc_area
= vmalloc(EXEC_SIZE
);
459 execute_location(vmalloc_area
);
463 case CT_EXEC_USERSPACE
: {
464 unsigned long user_addr
;
466 user_addr
= vm_mmap(NULL
, 0, PAGE_SIZE
,
467 PROT_READ
| PROT_WRITE
| PROT_EXEC
,
468 MAP_ANONYMOUS
| MAP_PRIVATE
, 0);
469 if (user_addr
>= TASK_SIZE
) {
470 pr_warn("Failed to allocate user memory\n");
473 execute_user_location((void *)user_addr
);
474 vm_munmap(user_addr
, PAGE_SIZE
);
477 case CT_ACCESS_USERSPACE
: {
478 unsigned long user_addr
, tmp
= 0;
481 user_addr
= vm_mmap(NULL
, 0, PAGE_SIZE
,
482 PROT_READ
| PROT_WRITE
| PROT_EXEC
,
483 MAP_ANONYMOUS
| MAP_PRIVATE
, 0);
484 if (user_addr
>= TASK_SIZE
) {
485 pr_warn("Failed to allocate user memory\n");
489 if (copy_to_user((void __user
*)user_addr
, &tmp
, sizeof(tmp
))) {
490 pr_warn("copy_to_user failed\n");
491 vm_munmap(user_addr
, PAGE_SIZE
);
495 ptr
= (unsigned long *)user_addr
;
497 pr_info("attempting bad read at %p\n", ptr
);
501 pr_info("attempting bad write at %p\n", ptr
);
504 vm_munmap(user_addr
, PAGE_SIZE
);
509 /* Explicitly cast away "const" for the test. */
510 unsigned long *ptr
= (unsigned long *)&rodata
;
512 pr_info("attempting bad rodata write at %p\n", ptr
);
517 case CT_WRITE_RO_AFTER_INIT
: {
518 unsigned long *ptr
= &ro_after_init
;
521 * Verify we were written to during init. Since an Oops
522 * is considered a "success", a failure is to just skip the
525 if ((*ptr
& 0xAA) != 0xAA) {
526 pr_info("%p was NOT written during init!?\n", ptr
);
530 pr_info("attempting bad ro_after_init write at %p\n", ptr
);
535 case CT_WRITE_KERN
: {
539 size
= (unsigned long)do_overwritten
-
540 (unsigned long)do_nothing
;
541 ptr
= (unsigned char *)do_overwritten
;
543 pr_info("attempting bad %zu byte write at %p\n", size
, ptr
);
544 memcpy(ptr
, (unsigned char *)do_nothing
, size
);
545 flush_icache_range((unsigned long)ptr
,
546 (unsigned long)(ptr
+ size
));
558 static void lkdtm_handler(void)
563 spin_lock_irqsave(&count_lock
, flags
);
565 pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
566 cp_name_to_str(cpoint
), cp_type_to_str(cptype
), count
);
570 count
= cpoint_count
;
572 spin_unlock_irqrestore(&count_lock
, flags
);
575 lkdtm_do_action(cptype
);
578 static int lkdtm_register_cpoint(enum cname which
)
583 if (lkdtm
.entry
!= NULL
)
584 unregister_jprobe(&lkdtm
);
588 lkdtm_do_action(cptype
);
590 case CN_INT_HARDWARE_ENTRY
:
591 lkdtm
.kp
.symbol_name
= "do_IRQ";
592 lkdtm
.entry
= (kprobe_opcode_t
*) jp_do_irq
;
594 case CN_INT_HW_IRQ_EN
:
595 lkdtm
.kp
.symbol_name
= "handle_IRQ_event";
596 lkdtm
.entry
= (kprobe_opcode_t
*) jp_handle_irq_event
;
598 case CN_INT_TASKLET_ENTRY
:
599 lkdtm
.kp
.symbol_name
= "tasklet_action";
600 lkdtm
.entry
= (kprobe_opcode_t
*) jp_tasklet_action
;
603 lkdtm
.kp
.symbol_name
= "ll_rw_block";
604 lkdtm
.entry
= (kprobe_opcode_t
*) jp_ll_rw_block
;
607 lkdtm
.kp
.symbol_name
= "shrink_inactive_list";
608 lkdtm
.entry
= (kprobe_opcode_t
*) jp_shrink_inactive_list
;
611 lkdtm
.kp
.symbol_name
= "hrtimer_start";
612 lkdtm
.entry
= (kprobe_opcode_t
*) jp_hrtimer_start
;
614 case CN_SCSI_DISPATCH_CMD
:
615 lkdtm
.kp
.symbol_name
= "scsi_dispatch_cmd";
616 lkdtm
.entry
= (kprobe_opcode_t
*) jp_scsi_dispatch_cmd
;
620 lkdtm
.kp
.symbol_name
= "generic_ide_ioctl";
621 lkdtm
.entry
= (kprobe_opcode_t
*) jp_generic_ide_ioctl
;
623 pr_info("Crash point not available\n");
628 pr_info("Invalid Crash Point\n");
633 if ((ret
= register_jprobe(&lkdtm
)) < 0) {
634 pr_info("Couldn't register jprobe\n");
641 static ssize_t
do_register_entry(enum cname which
, struct file
*f
,
642 const char __user
*user_buf
, size_t count
, loff_t
*off
)
647 if (count
>= PAGE_SIZE
)
650 buf
= (char *)__get_free_page(GFP_KERNEL
);
653 if (copy_from_user(buf
, user_buf
, count
)) {
654 free_page((unsigned long) buf
);
657 /* NULL-terminate and remove enter */
661 cptype
= parse_cp_type(buf
, count
);
662 free_page((unsigned long) buf
);
664 if (cptype
== CT_NONE
)
667 err
= lkdtm_register_cpoint(which
);
676 /* Generic read callback that just prints out the available crash types */
677 static ssize_t
lkdtm_debugfs_read(struct file
*f
, char __user
*user_buf
,
678 size_t count
, loff_t
*off
)
683 buf
= (char *)__get_free_page(GFP_KERNEL
);
687 n
= snprintf(buf
, PAGE_SIZE
, "Available crash types:\n");
688 for (i
= 0; i
< ARRAY_SIZE(cp_type
); i
++)
689 n
+= snprintf(buf
+ n
, PAGE_SIZE
- n
, "%s\n", cp_type
[i
]);
692 out
= simple_read_from_buffer(user_buf
, count
, off
,
694 free_page((unsigned long) buf
);
699 static int lkdtm_debugfs_open(struct inode
*inode
, struct file
*file
)
705 static ssize_t
int_hardware_entry(struct file
*f
, const char __user
*buf
,
706 size_t count
, loff_t
*off
)
708 return do_register_entry(CN_INT_HARDWARE_ENTRY
, f
, buf
, count
, off
);
711 static ssize_t
int_hw_irq_en(struct file
*f
, const char __user
*buf
,
712 size_t count
, loff_t
*off
)
714 return do_register_entry(CN_INT_HW_IRQ_EN
, f
, buf
, count
, off
);
717 static ssize_t
int_tasklet_entry(struct file
*f
, const char __user
*buf
,
718 size_t count
, loff_t
*off
)
720 return do_register_entry(CN_INT_TASKLET_ENTRY
, f
, buf
, count
, off
);
723 static ssize_t
fs_devrw_entry(struct file
*f
, const char __user
*buf
,
724 size_t count
, loff_t
*off
)
726 return do_register_entry(CN_FS_DEVRW
, f
, buf
, count
, off
);
729 static ssize_t
mem_swapout_entry(struct file
*f
, const char __user
*buf
,
730 size_t count
, loff_t
*off
)
732 return do_register_entry(CN_MEM_SWAPOUT
, f
, buf
, count
, off
);
735 static ssize_t
timeradd_entry(struct file
*f
, const char __user
*buf
,
736 size_t count
, loff_t
*off
)
738 return do_register_entry(CN_TIMERADD
, f
, buf
, count
, off
);
741 static ssize_t
scsi_dispatch_cmd_entry(struct file
*f
,
742 const char __user
*buf
, size_t count
, loff_t
*off
)
744 return do_register_entry(CN_SCSI_DISPATCH_CMD
, f
, buf
, count
, off
);
747 static ssize_t
ide_core_cp_entry(struct file
*f
, const char __user
*buf
,
748 size_t count
, loff_t
*off
)
750 return do_register_entry(CN_IDE_CORE_CP
, f
, buf
, count
, off
);
753 /* Special entry to just crash directly. Available without KPROBEs */
754 static ssize_t
direct_entry(struct file
*f
, const char __user
*user_buf
,
755 size_t count
, loff_t
*off
)
760 if (count
>= PAGE_SIZE
)
765 buf
= (char *)__get_free_page(GFP_KERNEL
);
768 if (copy_from_user(buf
, user_buf
, count
)) {
769 free_page((unsigned long) buf
);
772 /* NULL-terminate and remove enter */
776 type
= parse_cp_type(buf
, count
);
777 free_page((unsigned long) buf
);
781 pr_info("Performing direct entry %s\n", cp_type_to_str(type
));
782 lkdtm_do_action(type
);
790 const struct file_operations fops
;
793 static const struct crash_entry crash_entries
[] = {
794 {"DIRECT", {.read
= lkdtm_debugfs_read
,
795 .llseek
= generic_file_llseek
,
796 .open
= lkdtm_debugfs_open
,
797 .write
= direct_entry
} },
798 {"INT_HARDWARE_ENTRY", {.read
= lkdtm_debugfs_read
,
799 .llseek
= generic_file_llseek
,
800 .open
= lkdtm_debugfs_open
,
801 .write
= int_hardware_entry
} },
802 {"INT_HW_IRQ_EN", {.read
= lkdtm_debugfs_read
,
803 .llseek
= generic_file_llseek
,
804 .open
= lkdtm_debugfs_open
,
805 .write
= int_hw_irq_en
} },
806 {"INT_TASKLET_ENTRY", {.read
= lkdtm_debugfs_read
,
807 .llseek
= generic_file_llseek
,
808 .open
= lkdtm_debugfs_open
,
809 .write
= int_tasklet_entry
} },
810 {"FS_DEVRW", {.read
= lkdtm_debugfs_read
,
811 .llseek
= generic_file_llseek
,
812 .open
= lkdtm_debugfs_open
,
813 .write
= fs_devrw_entry
} },
814 {"MEM_SWAPOUT", {.read
= lkdtm_debugfs_read
,
815 .llseek
= generic_file_llseek
,
816 .open
= lkdtm_debugfs_open
,
817 .write
= mem_swapout_entry
} },
818 {"TIMERADD", {.read
= lkdtm_debugfs_read
,
819 .llseek
= generic_file_llseek
,
820 .open
= lkdtm_debugfs_open
,
821 .write
= timeradd_entry
} },
822 {"SCSI_DISPATCH_CMD", {.read
= lkdtm_debugfs_read
,
823 .llseek
= generic_file_llseek
,
824 .open
= lkdtm_debugfs_open
,
825 .write
= scsi_dispatch_cmd_entry
} },
826 {"IDE_CORE_CP", {.read
= lkdtm_debugfs_read
,
827 .llseek
= generic_file_llseek
,
828 .open
= lkdtm_debugfs_open
,
829 .write
= ide_core_cp_entry
} },
832 static struct dentry
*lkdtm_debugfs_root
;
834 static int __init
lkdtm_module_init(void)
837 int n_debugfs_entries
= 1; /* Assume only the direct entry */
840 /* Make sure we can write to __ro_after_init values during __init */
841 ro_after_init
|= 0xAA;
843 /* Register debugfs interface */
844 lkdtm_debugfs_root
= debugfs_create_dir("provoke-crash", NULL
);
845 if (!lkdtm_debugfs_root
) {
846 pr_err("creating root dir failed\n");
850 #ifdef CONFIG_KPROBES
851 n_debugfs_entries
= ARRAY_SIZE(crash_entries
);
854 for (i
= 0; i
< n_debugfs_entries
; i
++) {
855 const struct crash_entry
*cur
= &crash_entries
[i
];
858 de
= debugfs_create_file(cur
->name
, 0644, lkdtm_debugfs_root
,
861 pr_err("could not create %s\n", cur
->name
);
866 if (lkdtm_parse_commandline() == -EINVAL
) {
867 pr_info("Invalid command\n");
871 if (cpoint
!= CN_INVALID
&& cptype
!= CT_NONE
) {
872 ret
= lkdtm_register_cpoint(cpoint
);
874 pr_info("Invalid crash point %d\n", cpoint
);
877 pr_info("Crash point %s of type %s registered\n",
878 cpoint_name
, cpoint_type
);
880 pr_info("No crash points registered, enable through debugfs\n");
886 debugfs_remove_recursive(lkdtm_debugfs_root
);
890 static void __exit
lkdtm_module_exit(void)
892 debugfs_remove_recursive(lkdtm_debugfs_root
);
894 unregister_jprobe(&lkdtm
);
895 pr_info("Crash point unregistered\n");
898 module_init(lkdtm_module_init
);
899 module_exit(lkdtm_module_exit
);
901 MODULE_LICENSE("GPL");
902 MODULE_DESCRIPTION("Kprobe module for testing crash dumps");