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,
96 CT_WRITE_BUDDY_AFTER_FREE,
97 CT_READ_BUDDY_AFTER_FREE,
113 static char* cp_name[] = {
114 "INT_HARDWARE_ENTRY",
125 static char* cp_type[] = {
133 "UNALIGNED_LOAD_STORE_WRITE",
134 "OVERWRITE_ALLOCATION",
137 "WRITE_BUDDY_AFTER_FREE",
138 "READ_BUDDY_AFTER_FREE",
154 static struct jprobe lkdtm;
156 static int lkdtm_parse_commandline(void);
157 static void lkdtm_handler(void);
159 static char* cpoint_name;
160 static char* cpoint_type;
161 static int cpoint_count = DEFAULT_COUNT;
162 static int recur_count = REC_NUM_DEFAULT;
164 static enum cname cpoint = CN_INVALID;
165 static enum ctype cptype = CT_NONE;
166 static int count = DEFAULT_COUNT;
167 static DEFINE_SPINLOCK(count_lock);
168 static DEFINE_SPINLOCK(lock_me_up);
170 static u8 data_area[EXEC_SIZE];
172 static const unsigned long rodata = 0xAA55AA55;
174 module_param(recur_count, int, 0644);
175 MODULE_PARM_DESC(recur_count, " Recursion level for the stack overflow test");
176 module_param(cpoint_name, charp, 0444);
177 MODULE_PARM_DESC(cpoint_name, " Crash Point, where kernel is to be crashed");
178 module_param(cpoint_type, charp, 0444);
179 MODULE_PARM_DESC(cpoint_type, " Crash Point Type, action to be taken on "\
180 "hitting the crash point");
181 module_param(cpoint_count, int, 0644);
182 MODULE_PARM_DESC(cpoint_count, " Crash Point Count, number of times the "\
183 "crash point is to be hit to trigger action");
185 static unsigned int jp_do_irq(unsigned int irq)
192 static irqreturn_t jp_handle_irq_event(unsigned int irq,
193 struct irqaction *action)
200 static void jp_tasklet_action(struct softirq_action *a)
206 static void jp_ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
214 static unsigned long jp_shrink_inactive_list(unsigned long max_scan,
216 struct scan_control *sc)
223 static int jp_hrtimer_start(struct hrtimer *timer, ktime_t tim,
224 const enum hrtimer_mode mode)
231 static int jp_scsi_dispatch_cmd(struct scsi_cmnd *cmd)
239 static int jp_generic_ide_ioctl(ide_drive_t *drive, struct file *file,
240 struct block_device *bdev, unsigned int cmd,
249 /* Return the crashpoint number or NONE if the name is invalid */
250 static enum ctype parse_cp_type(const char *what, size_t count)
254 for (i = 0; i < ARRAY_SIZE(cp_type); i++) {
255 if (!strcmp(what, cp_type[i]))
262 static const char *cp_type_to_str(enum ctype type)
264 if (type == CT_NONE || type < 0 || type > ARRAY_SIZE(cp_type))
267 return cp_type[type - 1];
270 static const char *cp_name_to_str(enum cname name)
272 if (name == CN_INVALID || name < 0 || name > ARRAY_SIZE(cp_name))
275 return cp_name[name - 1];
279 static int lkdtm_parse_commandline(void)
284 if (cpoint_count < 1 || recur_count < 1)
287 spin_lock_irqsave(&count_lock, flags);
288 count = cpoint_count;
289 spin_unlock_irqrestore(&count_lock, flags);
291 /* No special parameters */
292 if (!cpoint_type && !cpoint_name)
295 /* Neither or both of these need to be set */
296 if (!cpoint_type || !cpoint_name)
299 cptype = parse_cp_type(cpoint_type, strlen(cpoint_type));
300 if (cptype == CT_NONE)
303 for (i = 0; i < ARRAY_SIZE(cp_name); i++) {
304 if (!strcmp(cpoint_name, cp_name[i])) {
310 /* Could not find a valid crash point */
314 static int recursive_loop(int remaining)
316 char buf[REC_STACK_SIZE];
318 /* Make sure compiler does not optimize this away. */
319 memset(buf, (remaining & 0xff) | 0x1, REC_STACK_SIZE);
323 return recursive_loop(remaining - 1);
326 static void do_nothing(void)
331 /* Must immediately follow do_nothing for size calculuations to work out. */
332 static void do_overwritten(void)
334 pr_info("do_overwritten wasn't overwritten!\n");
338 static noinline void corrupt_stack(void)
340 /* Use default char array length that triggers stack protection. */
343 memset((void *)data, 0, 64);
346 static void execute_location(void *dst)
348 void (*func)(void) = dst;
350 pr_info("attempting ok execution at %p\n", do_nothing);
353 memcpy(dst, do_nothing, EXEC_SIZE);
354 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
355 pr_info("attempting bad execution at %p\n", func);
359 static void execute_user_location(void *dst)
361 /* Intentionally crossing kernel/user memory boundary. */
362 void (*func)(void) = dst;
364 pr_info("attempting ok execution at %p\n", do_nothing);
367 if (copy_to_user((void __user *)dst, do_nothing, EXEC_SIZE))
369 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
370 pr_info("attempting bad execution at %p\n", func);
374 static void lkdtm_do_action(enum ctype which)
394 (void) recursive_loop(recur_count);
396 case CT_CORRUPT_STACK:
399 case CT_UNALIGNED_LOAD_STORE_WRITE: {
400 static u8 data[5] __attribute__((aligned(4))) = {1, 2,
403 u32 val = 0x12345678;
405 p = (u32 *)(data + 1);
411 case CT_OVERWRITE_ALLOCATION: {
413 u32 *data = kmalloc(len, GFP_KERNEL);
415 data[1024 / sizeof(u32)] = 0x12345678;
419 case CT_WRITE_AFTER_FREE: {
423 * The slub allocator uses the first word to store the free
424 * pointer in some configurations. Use the middle of the
425 * allocation to avoid running into the freelist
427 size_t offset = (len / sizeof(*base)) / 2;
429 base = kmalloc(len, GFP_KERNEL);
430 pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
432 pr_info("Attempting bad write to freed memory at %p\n",
434 base[offset] = 0x0abcdef0;
437 case CT_READ_AFTER_FREE: {
438 int *base, *val, saw;
441 * The slub allocator uses the first word to store the free
442 * pointer in some configurations. Use the middle of the
443 * allocation to avoid running into the freelist
445 size_t offset = (len / sizeof(*base)) / 2;
447 base = kmalloc(len, GFP_KERNEL);
451 val = kmalloc(len, GFP_KERNEL);
457 pr_info("Value in memory before free: %x\n", base[offset]);
461 pr_info("Attempting bad read from freed memory\n");
464 /* Good! Poisoning happened, so declare a win. */
465 pr_info("Memory correctly poisoned, calling BUG\n");
468 pr_info("Memory was not poisoned\n");
473 case CT_WRITE_BUDDY_AFTER_FREE: {
474 unsigned long p = __get_free_page(GFP_KERNEL);
477 pr_info("Writing to the buddy page before free\n");
478 memset((void *)p, 0x3, PAGE_SIZE);
481 pr_info("Attempting bad write to the buddy page after free\n");
482 memset((void *)p, 0x78, PAGE_SIZE);
485 case CT_READ_BUDDY_AFTER_FREE: {
486 unsigned long p = __get_free_page(GFP_KERNEL);
487 int saw, *val = kmalloc(1024, GFP_KERNEL);
500 pr_info("Value in memory before free: %x\n", base[0]);
502 pr_info("Attempting to read from freed memory\n");
505 /* Good! Poisoning happened, so declare a win. */
506 pr_info("Buddy page correctly poisoned, calling BUG\n");
509 pr_info("Buddy page was not poisoned\n");
525 /* Must be called twice to trigger. */
526 spin_lock(&lock_me_up);
527 /* Let sparse know we intended to exit holding the lock. */
528 __release(&lock_me_up);
531 set_current_state(TASK_UNINTERRUPTIBLE);
535 execute_location(data_area);
537 case CT_EXEC_STACK: {
538 u8 stack_area[EXEC_SIZE];
539 execute_location(stack_area);
542 case CT_EXEC_KMALLOC: {
543 u32 *kmalloc_area = kmalloc(EXEC_SIZE, GFP_KERNEL);
544 execute_location(kmalloc_area);
548 case CT_EXEC_VMALLOC: {
549 u32 *vmalloc_area = vmalloc(EXEC_SIZE);
550 execute_location(vmalloc_area);
554 case CT_EXEC_USERSPACE: {
555 unsigned long user_addr;
557 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
558 PROT_READ | PROT_WRITE | PROT_EXEC,
559 MAP_ANONYMOUS | MAP_PRIVATE, 0);
560 if (user_addr >= TASK_SIZE) {
561 pr_warn("Failed to allocate user memory\n");
564 execute_user_location((void *)user_addr);
565 vm_munmap(user_addr, PAGE_SIZE);
568 case CT_ACCESS_USERSPACE: {
569 unsigned long user_addr, tmp = 0;
572 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
573 PROT_READ | PROT_WRITE | PROT_EXEC,
574 MAP_ANONYMOUS | MAP_PRIVATE, 0);
575 if (user_addr >= TASK_SIZE) {
576 pr_warn("Failed to allocate user memory\n");
580 if (copy_to_user((void __user *)user_addr, &tmp, sizeof(tmp))) {
581 pr_warn("copy_to_user failed\n");
582 vm_munmap(user_addr, PAGE_SIZE);
586 ptr = (unsigned long *)user_addr;
588 pr_info("attempting bad read at %p\n", ptr);
592 pr_info("attempting bad write at %p\n", ptr);
595 vm_munmap(user_addr, PAGE_SIZE);
602 ptr = (unsigned long *)&rodata;
604 pr_info("attempting bad write at %p\n", ptr);
609 case CT_WRITE_KERN: {
613 size = (unsigned long)do_overwritten -
614 (unsigned long)do_nothing;
615 ptr = (unsigned char *)do_overwritten;
617 pr_info("attempting bad %zu byte write at %p\n", size, ptr);
618 memcpy(ptr, (unsigned char *)do_nothing, size);
619 flush_icache_range((unsigned long)ptr,
620 (unsigned long)(ptr + size));
625 case CT_WRAP_ATOMIC: {
626 atomic_t under = ATOMIC_INIT(INT_MIN);
627 atomic_t over = ATOMIC_INIT(INT_MAX);
629 pr_info("attempting atomic underflow\n");
631 pr_info("attempting atomic overflow\n");
643 static void lkdtm_handler(void)
648 spin_lock_irqsave(&count_lock, flags);
650 pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
651 cp_name_to_str(cpoint), cp_type_to_str(cptype), count);
655 count = cpoint_count;
657 spin_unlock_irqrestore(&count_lock, flags);
660 lkdtm_do_action(cptype);
663 static int lkdtm_register_cpoint(enum cname which)
668 if (lkdtm.entry != NULL)
669 unregister_jprobe(&lkdtm);
673 lkdtm_do_action(cptype);
675 case CN_INT_HARDWARE_ENTRY:
676 lkdtm.kp.symbol_name = "do_IRQ";
677 lkdtm.entry = (kprobe_opcode_t*) jp_do_irq;
679 case CN_INT_HW_IRQ_EN:
680 lkdtm.kp.symbol_name = "handle_IRQ_event";
681 lkdtm.entry = (kprobe_opcode_t*) jp_handle_irq_event;
683 case CN_INT_TASKLET_ENTRY:
684 lkdtm.kp.symbol_name = "tasklet_action";
685 lkdtm.entry = (kprobe_opcode_t*) jp_tasklet_action;
688 lkdtm.kp.symbol_name = "ll_rw_block";
689 lkdtm.entry = (kprobe_opcode_t*) jp_ll_rw_block;
692 lkdtm.kp.symbol_name = "shrink_inactive_list";
693 lkdtm.entry = (kprobe_opcode_t*) jp_shrink_inactive_list;
696 lkdtm.kp.symbol_name = "hrtimer_start";
697 lkdtm.entry = (kprobe_opcode_t*) jp_hrtimer_start;
699 case CN_SCSI_DISPATCH_CMD:
700 lkdtm.kp.symbol_name = "scsi_dispatch_cmd";
701 lkdtm.entry = (kprobe_opcode_t*) jp_scsi_dispatch_cmd;
705 lkdtm.kp.symbol_name = "generic_ide_ioctl";
706 lkdtm.entry = (kprobe_opcode_t*) jp_generic_ide_ioctl;
708 pr_info("Crash point not available\n");
713 pr_info("Invalid Crash Point\n");
718 if ((ret = register_jprobe(&lkdtm)) < 0) {
719 pr_info("Couldn't register jprobe\n");
726 static ssize_t do_register_entry(enum cname which, struct file *f,
727 const char __user *user_buf, size_t count, loff_t *off)
732 if (count >= PAGE_SIZE)
735 buf = (char *)__get_free_page(GFP_KERNEL);
738 if (copy_from_user(buf, user_buf, count)) {
739 free_page((unsigned long) buf);
742 /* NULL-terminate and remove enter */
746 cptype = parse_cp_type(buf, count);
747 free_page((unsigned long) buf);
749 if (cptype == CT_NONE)
752 err = lkdtm_register_cpoint(which);
761 /* Generic read callback that just prints out the available crash types */
762 static ssize_t lkdtm_debugfs_read(struct file *f, char __user *user_buf,
763 size_t count, loff_t *off)
768 buf = (char *)__get_free_page(GFP_KERNEL);
772 n = snprintf(buf, PAGE_SIZE, "Available crash types:\n");
773 for (i = 0; i < ARRAY_SIZE(cp_type); i++)
774 n += snprintf(buf + n, PAGE_SIZE - n, "%s\n", cp_type[i]);
777 out = simple_read_from_buffer(user_buf, count, off,
779 free_page((unsigned long) buf);
784 static int lkdtm_debugfs_open(struct inode *inode, struct file *file)
790 static ssize_t int_hardware_entry(struct file *f, const char __user *buf,
791 size_t count, loff_t *off)
793 return do_register_entry(CN_INT_HARDWARE_ENTRY, f, buf, count, off);
796 static ssize_t int_hw_irq_en(struct file *f, const char __user *buf,
797 size_t count, loff_t *off)
799 return do_register_entry(CN_INT_HW_IRQ_EN, f, buf, count, off);
802 static ssize_t int_tasklet_entry(struct file *f, const char __user *buf,
803 size_t count, loff_t *off)
805 return do_register_entry(CN_INT_TASKLET_ENTRY, f, buf, count, off);
808 static ssize_t fs_devrw_entry(struct file *f, const char __user *buf,
809 size_t count, loff_t *off)
811 return do_register_entry(CN_FS_DEVRW, f, buf, count, off);
814 static ssize_t mem_swapout_entry(struct file *f, const char __user *buf,
815 size_t count, loff_t *off)
817 return do_register_entry(CN_MEM_SWAPOUT, f, buf, count, off);
820 static ssize_t timeradd_entry(struct file *f, const char __user *buf,
821 size_t count, loff_t *off)
823 return do_register_entry(CN_TIMERADD, f, buf, count, off);
826 static ssize_t scsi_dispatch_cmd_entry(struct file *f,
827 const char __user *buf, size_t count, loff_t *off)
829 return do_register_entry(CN_SCSI_DISPATCH_CMD, f, buf, count, off);
832 static ssize_t ide_core_cp_entry(struct file *f, const char __user *buf,
833 size_t count, loff_t *off)
835 return do_register_entry(CN_IDE_CORE_CP, f, buf, count, off);
838 /* Special entry to just crash directly. Available without KPROBEs */
839 static ssize_t direct_entry(struct file *f, const char __user *user_buf,
840 size_t count, loff_t *off)
845 if (count >= PAGE_SIZE)
850 buf = (char *)__get_free_page(GFP_KERNEL);
853 if (copy_from_user(buf, user_buf, count)) {
854 free_page((unsigned long) buf);
857 /* NULL-terminate and remove enter */
861 type = parse_cp_type(buf, count);
862 free_page((unsigned long) buf);
866 pr_info("Performing direct entry %s\n", cp_type_to_str(type));
867 lkdtm_do_action(type);
875 const struct file_operations fops;
878 static const struct crash_entry crash_entries[] = {
879 {"DIRECT", {.read = lkdtm_debugfs_read,
880 .llseek = generic_file_llseek,
881 .open = lkdtm_debugfs_open,
882 .write = direct_entry} },
883 {"INT_HARDWARE_ENTRY", {.read = lkdtm_debugfs_read,
884 .llseek = generic_file_llseek,
885 .open = lkdtm_debugfs_open,
886 .write = int_hardware_entry} },
887 {"INT_HW_IRQ_EN", {.read = lkdtm_debugfs_read,
888 .llseek = generic_file_llseek,
889 .open = lkdtm_debugfs_open,
890 .write = int_hw_irq_en} },
891 {"INT_TASKLET_ENTRY", {.read = lkdtm_debugfs_read,
892 .llseek = generic_file_llseek,
893 .open = lkdtm_debugfs_open,
894 .write = int_tasklet_entry} },
895 {"FS_DEVRW", {.read = lkdtm_debugfs_read,
896 .llseek = generic_file_llseek,
897 .open = lkdtm_debugfs_open,
898 .write = fs_devrw_entry} },
899 {"MEM_SWAPOUT", {.read = lkdtm_debugfs_read,
900 .llseek = generic_file_llseek,
901 .open = lkdtm_debugfs_open,
902 .write = mem_swapout_entry} },
903 {"TIMERADD", {.read = lkdtm_debugfs_read,
904 .llseek = generic_file_llseek,
905 .open = lkdtm_debugfs_open,
906 .write = timeradd_entry} },
907 {"SCSI_DISPATCH_CMD", {.read = lkdtm_debugfs_read,
908 .llseek = generic_file_llseek,
909 .open = lkdtm_debugfs_open,
910 .write = scsi_dispatch_cmd_entry} },
911 {"IDE_CORE_CP", {.read = lkdtm_debugfs_read,
912 .llseek = generic_file_llseek,
913 .open = lkdtm_debugfs_open,
914 .write = ide_core_cp_entry} },
917 static struct dentry *lkdtm_debugfs_root;
919 static int __init lkdtm_module_init(void)
922 int n_debugfs_entries = 1; /* Assume only the direct entry */
925 /* Register debugfs interface */
926 lkdtm_debugfs_root = debugfs_create_dir("provoke-crash", NULL);
927 if (!lkdtm_debugfs_root) {
928 pr_err("creating root dir failed\n");
932 #ifdef CONFIG_KPROBES
933 n_debugfs_entries = ARRAY_SIZE(crash_entries);
936 for (i = 0; i < n_debugfs_entries; i++) {
937 const struct crash_entry *cur = &crash_entries[i];
940 de = debugfs_create_file(cur->name, 0644, lkdtm_debugfs_root,
943 pr_err("could not create %s\n", cur->name);
948 if (lkdtm_parse_commandline() == -EINVAL) {
949 pr_info("Invalid command\n");
953 if (cpoint != CN_INVALID && cptype != CT_NONE) {
954 ret = lkdtm_register_cpoint(cpoint);
956 pr_info("Invalid crash point %d\n", cpoint);
959 pr_info("Crash point %s of type %s registered\n",
960 cpoint_name, cpoint_type);
962 pr_info("No crash points registered, enable through debugfs\n");
968 debugfs_remove_recursive(lkdtm_debugfs_root);
972 static void __exit lkdtm_module_exit(void)
974 debugfs_remove_recursive(lkdtm_debugfs_root);
976 unregister_jprobe(&lkdtm);
977 pr_info("Crash point unregistered\n");
980 module_init(lkdtm_module_init);
981 module_exit(lkdtm_module_exit);
983 MODULE_LICENSE("GPL");
984 MODULE_DESCRIPTION("Kprobe module for testing crash dumps");