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,
110 static char* cp_name[] = {
111 "INT_HARDWARE_ENTRY",
122 static char* cp_type[] = {
130 "UNALIGNED_LOAD_STORE_WRITE",
131 "OVERWRITE_ALLOCATION",
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;
168 module_param(recur_count, int, 0644);
169 MODULE_PARM_DESC(recur_count, " Recursion level for the stack overflow test");
170 module_param(cpoint_name, charp, 0444);
171 MODULE_PARM_DESC(cpoint_name, " Crash Point, where kernel is to be crashed");
172 module_param(cpoint_type, charp, 0444);
173 MODULE_PARM_DESC(cpoint_type, " Crash Point Type, action to be taken on "\
174 "hitting the crash point");
175 module_param(cpoint_count, int, 0644);
176 MODULE_PARM_DESC(cpoint_count, " Crash Point Count, number of times the "\
177 "crash point is to be hit to trigger action");
179 static unsigned int jp_do_irq(unsigned int irq)
186 static irqreturn_t jp_handle_irq_event(unsigned int irq,
187 struct irqaction *action)
194 static void jp_tasklet_action(struct softirq_action *a)
200 static void jp_ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
208 static unsigned long jp_shrink_inactive_list(unsigned long max_scan,
210 struct scan_control *sc)
217 static int jp_hrtimer_start(struct hrtimer *timer, ktime_t tim,
218 const enum hrtimer_mode mode)
225 static int jp_scsi_dispatch_cmd(struct scsi_cmnd *cmd)
233 static int jp_generic_ide_ioctl(ide_drive_t *drive, struct file *file,
234 struct block_device *bdev, unsigned int cmd,
243 /* Return the crashpoint number or NONE if the name is invalid */
244 static enum ctype parse_cp_type(const char *what, size_t count)
248 for (i = 0; i < ARRAY_SIZE(cp_type); i++) {
249 if (!strcmp(what, cp_type[i]))
256 static const char *cp_type_to_str(enum ctype type)
258 if (type == CT_NONE || type < 0 || type > ARRAY_SIZE(cp_type))
261 return cp_type[type - 1];
264 static const char *cp_name_to_str(enum cname name)
266 if (name == CN_INVALID || name < 0 || name > ARRAY_SIZE(cp_name))
269 return cp_name[name - 1];
273 static int lkdtm_parse_commandline(void)
278 if (cpoint_count < 1 || recur_count < 1)
281 spin_lock_irqsave(&count_lock, flags);
282 count = cpoint_count;
283 spin_unlock_irqrestore(&count_lock, flags);
285 /* No special parameters */
286 if (!cpoint_type && !cpoint_name)
289 /* Neither or both of these need to be set */
290 if (!cpoint_type || !cpoint_name)
293 cptype = parse_cp_type(cpoint_type, strlen(cpoint_type));
294 if (cptype == CT_NONE)
297 for (i = 0; i < ARRAY_SIZE(cp_name); i++) {
298 if (!strcmp(cpoint_name, cp_name[i])) {
304 /* Could not find a valid crash point */
308 static int recursive_loop(int remaining)
310 char buf[REC_STACK_SIZE];
312 /* Make sure compiler does not optimize this away. */
313 memset(buf, (remaining & 0xff) | 0x1, REC_STACK_SIZE);
317 return recursive_loop(remaining - 1);
320 static void do_nothing(void)
325 /* Must immediately follow do_nothing for size calculuations to work out. */
326 static void do_overwritten(void)
328 pr_info("do_overwritten wasn't overwritten!\n");
332 static noinline void corrupt_stack(void)
334 /* Use default char array length that triggers stack protection. */
337 memset((void *)data, 0, 64);
340 static void execute_location(void *dst)
342 void (*func)(void) = dst;
344 pr_info("attempting ok execution at %p\n", do_nothing);
347 memcpy(dst, do_nothing, EXEC_SIZE);
348 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
349 pr_info("attempting bad execution at %p\n", func);
353 static void execute_user_location(void *dst)
355 /* Intentionally crossing kernel/user memory boundary. */
356 void (*func)(void) = dst;
358 pr_info("attempting ok execution at %p\n", do_nothing);
361 if (copy_to_user((void __user *)dst, do_nothing, EXEC_SIZE))
363 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
364 pr_info("attempting bad execution at %p\n", func);
368 static void lkdtm_do_action(enum ctype which)
388 (void) recursive_loop(recur_count);
390 case CT_CORRUPT_STACK:
393 case CT_UNALIGNED_LOAD_STORE_WRITE: {
394 static u8 data[5] __attribute__((aligned(4))) = {1, 2,
397 u32 val = 0x12345678;
399 p = (u32 *)(data + 1);
405 case CT_OVERWRITE_ALLOCATION: {
407 u32 *data = kmalloc(len, GFP_KERNEL);
409 data[1024 / sizeof(u32)] = 0x12345678;
413 case CT_WRITE_AFTER_FREE: {
417 * The slub allocator uses the first word to store the free
418 * pointer in some configurations. Use the middle of the
419 * allocation to avoid running into the freelist
421 size_t offset = (len / sizeof(*base)) / 2;
423 base = kmalloc(len, GFP_KERNEL);
424 pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
426 pr_info("Attempting bad write to freed memory at %p\n",
428 base[offset] = 0x0abcdef0;
431 case CT_READ_AFTER_FREE: {
432 int *base, *val, saw;
435 * The slub allocator uses the first word to store the free
436 * pointer in some configurations. Use the middle of the
437 * allocation to avoid running into the freelist
439 size_t offset = (len / sizeof(*base)) / 2;
441 base = kmalloc(len, GFP_KERNEL);
445 val = kmalloc(len, GFP_KERNEL);
451 pr_info("Value in memory before free: %x\n", base[offset]);
455 pr_info("Attempting bad read from freed memory\n");
458 /* Good! Poisoning happened, so declare a win. */
459 pr_info("Memory correctly poisoned, calling BUG\n");
462 pr_info("Memory was not poisoned\n");
478 /* Must be called twice to trigger. */
479 spin_lock(&lock_me_up);
480 /* Let sparse know we intended to exit holding the lock. */
481 __release(&lock_me_up);
484 set_current_state(TASK_UNINTERRUPTIBLE);
488 execute_location(data_area);
490 case CT_EXEC_STACK: {
491 u8 stack_area[EXEC_SIZE];
492 execute_location(stack_area);
495 case CT_EXEC_KMALLOC: {
496 u32 *kmalloc_area = kmalloc(EXEC_SIZE, GFP_KERNEL);
497 execute_location(kmalloc_area);
501 case CT_EXEC_VMALLOC: {
502 u32 *vmalloc_area = vmalloc(EXEC_SIZE);
503 execute_location(vmalloc_area);
507 case CT_EXEC_USERSPACE: {
508 unsigned long user_addr;
510 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
511 PROT_READ | PROT_WRITE | PROT_EXEC,
512 MAP_ANONYMOUS | MAP_PRIVATE, 0);
513 if (user_addr >= TASK_SIZE) {
514 pr_warn("Failed to allocate user memory\n");
517 execute_user_location((void *)user_addr);
518 vm_munmap(user_addr, PAGE_SIZE);
521 case CT_ACCESS_USERSPACE: {
522 unsigned long user_addr, tmp = 0;
525 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
526 PROT_READ | PROT_WRITE | PROT_EXEC,
527 MAP_ANONYMOUS | MAP_PRIVATE, 0);
528 if (user_addr >= TASK_SIZE) {
529 pr_warn("Failed to allocate user memory\n");
533 if (copy_to_user((void __user *)user_addr, &tmp, sizeof(tmp))) {
534 pr_warn("copy_to_user failed\n");
535 vm_munmap(user_addr, PAGE_SIZE);
539 ptr = (unsigned long *)user_addr;
541 pr_info("attempting bad read at %p\n", ptr);
545 pr_info("attempting bad write at %p\n", ptr);
548 vm_munmap(user_addr, PAGE_SIZE);
555 ptr = (unsigned long *)&rodata;
557 pr_info("attempting bad write at %p\n", ptr);
562 case CT_WRITE_KERN: {
566 size = (unsigned long)do_overwritten -
567 (unsigned long)do_nothing;
568 ptr = (unsigned char *)do_overwritten;
570 pr_info("attempting bad %zu byte write at %p\n", size, ptr);
571 memcpy(ptr, (unsigned char *)do_nothing, size);
572 flush_icache_range((unsigned long)ptr,
573 (unsigned long)(ptr + size));
585 static void lkdtm_handler(void)
590 spin_lock_irqsave(&count_lock, flags);
592 pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
593 cp_name_to_str(cpoint), cp_type_to_str(cptype), count);
597 count = cpoint_count;
599 spin_unlock_irqrestore(&count_lock, flags);
602 lkdtm_do_action(cptype);
605 static int lkdtm_register_cpoint(enum cname which)
610 if (lkdtm.entry != NULL)
611 unregister_jprobe(&lkdtm);
615 lkdtm_do_action(cptype);
617 case CN_INT_HARDWARE_ENTRY:
618 lkdtm.kp.symbol_name = "do_IRQ";
619 lkdtm.entry = (kprobe_opcode_t*) jp_do_irq;
621 case CN_INT_HW_IRQ_EN:
622 lkdtm.kp.symbol_name = "handle_IRQ_event";
623 lkdtm.entry = (kprobe_opcode_t*) jp_handle_irq_event;
625 case CN_INT_TASKLET_ENTRY:
626 lkdtm.kp.symbol_name = "tasklet_action";
627 lkdtm.entry = (kprobe_opcode_t*) jp_tasklet_action;
630 lkdtm.kp.symbol_name = "ll_rw_block";
631 lkdtm.entry = (kprobe_opcode_t*) jp_ll_rw_block;
634 lkdtm.kp.symbol_name = "shrink_inactive_list";
635 lkdtm.entry = (kprobe_opcode_t*) jp_shrink_inactive_list;
638 lkdtm.kp.symbol_name = "hrtimer_start";
639 lkdtm.entry = (kprobe_opcode_t*) jp_hrtimer_start;
641 case CN_SCSI_DISPATCH_CMD:
642 lkdtm.kp.symbol_name = "scsi_dispatch_cmd";
643 lkdtm.entry = (kprobe_opcode_t*) jp_scsi_dispatch_cmd;
647 lkdtm.kp.symbol_name = "generic_ide_ioctl";
648 lkdtm.entry = (kprobe_opcode_t*) jp_generic_ide_ioctl;
650 pr_info("Crash point not available\n");
655 pr_info("Invalid Crash Point\n");
660 if ((ret = register_jprobe(&lkdtm)) < 0) {
661 pr_info("Couldn't register jprobe\n");
668 static ssize_t do_register_entry(enum cname which, struct file *f,
669 const char __user *user_buf, size_t count, loff_t *off)
674 if (count >= PAGE_SIZE)
677 buf = (char *)__get_free_page(GFP_KERNEL);
680 if (copy_from_user(buf, user_buf, count)) {
681 free_page((unsigned long) buf);
684 /* NULL-terminate and remove enter */
688 cptype = parse_cp_type(buf, count);
689 free_page((unsigned long) buf);
691 if (cptype == CT_NONE)
694 err = lkdtm_register_cpoint(which);
703 /* Generic read callback that just prints out the available crash types */
704 static ssize_t lkdtm_debugfs_read(struct file *f, char __user *user_buf,
705 size_t count, loff_t *off)
710 buf = (char *)__get_free_page(GFP_KERNEL);
714 n = snprintf(buf, PAGE_SIZE, "Available crash types:\n");
715 for (i = 0; i < ARRAY_SIZE(cp_type); i++)
716 n += snprintf(buf + n, PAGE_SIZE - n, "%s\n", cp_type[i]);
719 out = simple_read_from_buffer(user_buf, count, off,
721 free_page((unsigned long) buf);
726 static int lkdtm_debugfs_open(struct inode *inode, struct file *file)
732 static ssize_t int_hardware_entry(struct file *f, const char __user *buf,
733 size_t count, loff_t *off)
735 return do_register_entry(CN_INT_HARDWARE_ENTRY, f, buf, count, off);
738 static ssize_t int_hw_irq_en(struct file *f, const char __user *buf,
739 size_t count, loff_t *off)
741 return do_register_entry(CN_INT_HW_IRQ_EN, f, buf, count, off);
744 static ssize_t int_tasklet_entry(struct file *f, const char __user *buf,
745 size_t count, loff_t *off)
747 return do_register_entry(CN_INT_TASKLET_ENTRY, f, buf, count, off);
750 static ssize_t fs_devrw_entry(struct file *f, const char __user *buf,
751 size_t count, loff_t *off)
753 return do_register_entry(CN_FS_DEVRW, f, buf, count, off);
756 static ssize_t mem_swapout_entry(struct file *f, const char __user *buf,
757 size_t count, loff_t *off)
759 return do_register_entry(CN_MEM_SWAPOUT, f, buf, count, off);
762 static ssize_t timeradd_entry(struct file *f, const char __user *buf,
763 size_t count, loff_t *off)
765 return do_register_entry(CN_TIMERADD, f, buf, count, off);
768 static ssize_t scsi_dispatch_cmd_entry(struct file *f,
769 const char __user *buf, size_t count, loff_t *off)
771 return do_register_entry(CN_SCSI_DISPATCH_CMD, f, buf, count, off);
774 static ssize_t ide_core_cp_entry(struct file *f, const char __user *buf,
775 size_t count, loff_t *off)
777 return do_register_entry(CN_IDE_CORE_CP, f, buf, count, off);
780 /* Special entry to just crash directly. Available without KPROBEs */
781 static ssize_t direct_entry(struct file *f, const char __user *user_buf,
782 size_t count, loff_t *off)
787 if (count >= PAGE_SIZE)
792 buf = (char *)__get_free_page(GFP_KERNEL);
795 if (copy_from_user(buf, user_buf, count)) {
796 free_page((unsigned long) buf);
799 /* NULL-terminate and remove enter */
803 type = parse_cp_type(buf, count);
804 free_page((unsigned long) buf);
808 pr_info("Performing direct entry %s\n", cp_type_to_str(type));
809 lkdtm_do_action(type);
817 const struct file_operations fops;
820 static const struct crash_entry crash_entries[] = {
821 {"DIRECT", {.read = lkdtm_debugfs_read,
822 .llseek = generic_file_llseek,
823 .open = lkdtm_debugfs_open,
824 .write = direct_entry} },
825 {"INT_HARDWARE_ENTRY", {.read = lkdtm_debugfs_read,
826 .llseek = generic_file_llseek,
827 .open = lkdtm_debugfs_open,
828 .write = int_hardware_entry} },
829 {"INT_HW_IRQ_EN", {.read = lkdtm_debugfs_read,
830 .llseek = generic_file_llseek,
831 .open = lkdtm_debugfs_open,
832 .write = int_hw_irq_en} },
833 {"INT_TASKLET_ENTRY", {.read = lkdtm_debugfs_read,
834 .llseek = generic_file_llseek,
835 .open = lkdtm_debugfs_open,
836 .write = int_tasklet_entry} },
837 {"FS_DEVRW", {.read = lkdtm_debugfs_read,
838 .llseek = generic_file_llseek,
839 .open = lkdtm_debugfs_open,
840 .write = fs_devrw_entry} },
841 {"MEM_SWAPOUT", {.read = lkdtm_debugfs_read,
842 .llseek = generic_file_llseek,
843 .open = lkdtm_debugfs_open,
844 .write = mem_swapout_entry} },
845 {"TIMERADD", {.read = lkdtm_debugfs_read,
846 .llseek = generic_file_llseek,
847 .open = lkdtm_debugfs_open,
848 .write = timeradd_entry} },
849 {"SCSI_DISPATCH_CMD", {.read = lkdtm_debugfs_read,
850 .llseek = generic_file_llseek,
851 .open = lkdtm_debugfs_open,
852 .write = scsi_dispatch_cmd_entry} },
853 {"IDE_CORE_CP", {.read = lkdtm_debugfs_read,
854 .llseek = generic_file_llseek,
855 .open = lkdtm_debugfs_open,
856 .write = ide_core_cp_entry} },
859 static struct dentry *lkdtm_debugfs_root;
861 static int __init lkdtm_module_init(void)
864 int n_debugfs_entries = 1; /* Assume only the direct entry */
867 /* Register debugfs interface */
868 lkdtm_debugfs_root = debugfs_create_dir("provoke-crash", NULL);
869 if (!lkdtm_debugfs_root) {
870 pr_err("creating root dir failed\n");
874 #ifdef CONFIG_KPROBES
875 n_debugfs_entries = ARRAY_SIZE(crash_entries);
878 for (i = 0; i < n_debugfs_entries; i++) {
879 const struct crash_entry *cur = &crash_entries[i];
882 de = debugfs_create_file(cur->name, 0644, lkdtm_debugfs_root,
885 pr_err("could not create %s\n", cur->name);
890 if (lkdtm_parse_commandline() == -EINVAL) {
891 pr_info("Invalid command\n");
895 if (cpoint != CN_INVALID && cptype != CT_NONE) {
896 ret = lkdtm_register_cpoint(cpoint);
898 pr_info("Invalid crash point %d\n", cpoint);
901 pr_info("Crash point %s of type %s registered\n",
902 cpoint_name, cpoint_type);
904 pr_info("No crash points registered, enable through debugfs\n");
910 debugfs_remove_recursive(lkdtm_debugfs_root);
914 static void __exit lkdtm_module_exit(void)
916 debugfs_remove_recursive(lkdtm_debugfs_root);
918 unregister_jprobe(&lkdtm);
919 pr_info("Crash point unregistered\n");
922 module_init(lkdtm_module_init);
923 module_exit(lkdtm_module_exit);
925 MODULE_LICENSE("GPL");
926 MODULE_DESCRIPTION("Kprobe module for testing crash dumps");