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,
109 CT_WRITE_RO_AFTER_INIT,
114 static char* cp_name[] = {
115 "INT_HARDWARE_ENTRY",
126 static char* cp_type[] = {
134 "UNALIGNED_LOAD_STORE_WRITE",
135 "OVERWRITE_ALLOCATION",
138 "WRITE_BUDDY_AFTER_FREE",
139 "READ_BUDDY_AFTER_FREE",
151 "WRITE_RO_AFTER_INIT",
156 static struct jprobe lkdtm;
158 static int lkdtm_parse_commandline(void);
159 static void lkdtm_handler(void);
161 static char* cpoint_name;
162 static char* cpoint_type;
163 static int cpoint_count = DEFAULT_COUNT;
164 static int recur_count = REC_NUM_DEFAULT;
166 static enum cname cpoint = CN_INVALID;
167 static enum ctype cptype = CT_NONE;
168 static int count = DEFAULT_COUNT;
169 static DEFINE_SPINLOCK(count_lock);
170 static DEFINE_SPINLOCK(lock_me_up);
172 static u8 data_area[EXEC_SIZE];
174 static const unsigned long rodata = 0xAA55AA55;
175 static unsigned long ro_after_init __ro_after_init = 0x55AA5500;
177 module_param(recur_count, int, 0644);
178 MODULE_PARM_DESC(recur_count, " Recursion level for the stack overflow test");
179 module_param(cpoint_name, charp, 0444);
180 MODULE_PARM_DESC(cpoint_name, " Crash Point, where kernel is to be crashed");
181 module_param(cpoint_type, charp, 0444);
182 MODULE_PARM_DESC(cpoint_type, " Crash Point Type, action to be taken on "\
183 "hitting the crash point");
184 module_param(cpoint_count, int, 0644);
185 MODULE_PARM_DESC(cpoint_count, " Crash Point Count, number of times the "\
186 "crash point is to be hit to trigger action");
188 static unsigned int jp_do_irq(unsigned int irq)
195 static irqreturn_t jp_handle_irq_event(unsigned int irq,
196 struct irqaction *action)
203 static void jp_tasklet_action(struct softirq_action *a)
209 static void jp_ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
217 static unsigned long jp_shrink_inactive_list(unsigned long max_scan,
219 struct scan_control *sc)
226 static int jp_hrtimer_start(struct hrtimer *timer, ktime_t tim,
227 const enum hrtimer_mode mode)
234 static int jp_scsi_dispatch_cmd(struct scsi_cmnd *cmd)
242 static int jp_generic_ide_ioctl(ide_drive_t *drive, struct file *file,
243 struct block_device *bdev, unsigned int cmd,
252 /* Return the crashpoint number or NONE if the name is invalid */
253 static enum ctype parse_cp_type(const char *what, size_t count)
257 for (i = 0; i < ARRAY_SIZE(cp_type); i++) {
258 if (!strcmp(what, cp_type[i]))
265 static const char *cp_type_to_str(enum ctype type)
267 if (type == CT_NONE || type < 0 || type > ARRAY_SIZE(cp_type))
270 return cp_type[type - 1];
273 static const char *cp_name_to_str(enum cname name)
275 if (name == CN_INVALID || name < 0 || name > ARRAY_SIZE(cp_name))
278 return cp_name[name - 1];
282 static int lkdtm_parse_commandline(void)
287 if (cpoint_count < 1 || recur_count < 1)
290 spin_lock_irqsave(&count_lock, flags);
291 count = cpoint_count;
292 spin_unlock_irqrestore(&count_lock, flags);
294 /* No special parameters */
295 if (!cpoint_type && !cpoint_name)
298 /* Neither or both of these need to be set */
299 if (!cpoint_type || !cpoint_name)
302 cptype = parse_cp_type(cpoint_type, strlen(cpoint_type));
303 if (cptype == CT_NONE)
306 for (i = 0; i < ARRAY_SIZE(cp_name); i++) {
307 if (!strcmp(cpoint_name, cp_name[i])) {
313 /* Could not find a valid crash point */
317 static int recursive_loop(int remaining)
319 char buf[REC_STACK_SIZE];
321 /* Make sure compiler does not optimize this away. */
322 memset(buf, (remaining & 0xff) | 0x1, REC_STACK_SIZE);
326 return recursive_loop(remaining - 1);
329 static void do_nothing(void)
334 /* Must immediately follow do_nothing for size calculuations to work out. */
335 static void do_overwritten(void)
337 pr_info("do_overwritten wasn't overwritten!\n");
341 static noinline void corrupt_stack(void)
343 /* Use default char array length that triggers stack protection. */
346 memset((void *)data, 0, 64);
349 static void noinline execute_location(void *dst)
351 void (*func)(void) = dst;
353 pr_info("attempting ok execution at %p\n", do_nothing);
356 memcpy(dst, do_nothing, EXEC_SIZE);
357 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
358 pr_info("attempting bad execution at %p\n", func);
362 static void execute_user_location(void *dst)
364 /* Intentionally crossing kernel/user memory boundary. */
365 void (*func)(void) = dst;
367 pr_info("attempting ok execution at %p\n", do_nothing);
370 if (copy_to_user((void __user *)dst, do_nothing, EXEC_SIZE))
372 flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
373 pr_info("attempting bad execution at %p\n", func);
377 static void lkdtm_do_action(enum ctype which)
397 (void) recursive_loop(recur_count);
399 case CT_CORRUPT_STACK:
402 case CT_UNALIGNED_LOAD_STORE_WRITE: {
403 static u8 data[5] __attribute__((aligned(4))) = {1, 2,
406 u32 val = 0x12345678;
408 p = (u32 *)(data + 1);
414 case CT_OVERWRITE_ALLOCATION: {
416 u32 *data = kmalloc(len, GFP_KERNEL);
418 data[1024 / sizeof(u32)] = 0x12345678;
422 case CT_WRITE_AFTER_FREE: {
426 * The slub allocator uses the first word to store the free
427 * pointer in some configurations. Use the middle of the
428 * allocation to avoid running into the freelist
430 size_t offset = (len / sizeof(*base)) / 2;
432 base = kmalloc(len, GFP_KERNEL);
433 pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
434 pr_info("Attempting bad write to freed memory at %p\n",
437 base[offset] = 0x0abcdef0;
438 /* Attempt to notice the overwrite. */
439 again = kmalloc(len, GFP_KERNEL);
442 pr_info("Hmm, didn't get the same memory range.\n");
446 case CT_READ_AFTER_FREE: {
447 int *base, *val, saw;
450 * The slub allocator uses the first word to store the free
451 * pointer in some configurations. Use the middle of the
452 * allocation to avoid running into the freelist
454 size_t offset = (len / sizeof(*base)) / 2;
456 base = kmalloc(len, GFP_KERNEL);
460 val = kmalloc(len, GFP_KERNEL);
466 pr_info("Value in memory before free: %x\n", base[offset]);
470 pr_info("Attempting bad read from freed memory\n");
473 /* Good! Poisoning happened, so declare a win. */
474 pr_info("Memory correctly poisoned (%x)\n", saw);
477 pr_info("Memory was not poisoned\n");
482 case CT_WRITE_BUDDY_AFTER_FREE: {
483 unsigned long p = __get_free_page(GFP_KERNEL);
486 pr_info("Writing to the buddy page before free\n");
487 memset((void *)p, 0x3, PAGE_SIZE);
490 pr_info("Attempting bad write to the buddy page after free\n");
491 memset((void *)p, 0x78, PAGE_SIZE);
492 /* Attempt to notice the overwrite. */
493 p = __get_free_page(GFP_KERNEL);
499 case CT_READ_BUDDY_AFTER_FREE: {
500 unsigned long p = __get_free_page(GFP_KERNEL);
501 int saw, *val = kmalloc(1024, GFP_KERNEL);
514 pr_info("Value in memory before free: %x\n", base[0]);
516 pr_info("Attempting to read from freed memory\n");
519 /* Good! Poisoning happened, so declare a win. */
520 pr_info("Memory correctly poisoned (%x)\n", saw);
523 pr_info("Buddy page was not poisoned\n");
539 /* Must be called twice to trigger. */
540 spin_lock(&lock_me_up);
541 /* Let sparse know we intended to exit holding the lock. */
542 __release(&lock_me_up);
545 set_current_state(TASK_UNINTERRUPTIBLE);
549 execute_location(data_area);
551 case CT_EXEC_STACK: {
552 u8 stack_area[EXEC_SIZE];
553 execute_location(stack_area);
556 case CT_EXEC_KMALLOC: {
557 u32 *kmalloc_area = kmalloc(EXEC_SIZE, GFP_KERNEL);
558 execute_location(kmalloc_area);
562 case CT_EXEC_VMALLOC: {
563 u32 *vmalloc_area = vmalloc(EXEC_SIZE);
564 execute_location(vmalloc_area);
568 case CT_EXEC_USERSPACE: {
569 unsigned long user_addr;
571 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
572 PROT_READ | PROT_WRITE | PROT_EXEC,
573 MAP_ANONYMOUS | MAP_PRIVATE, 0);
574 if (user_addr >= TASK_SIZE) {
575 pr_warn("Failed to allocate user memory\n");
578 execute_user_location((void *)user_addr);
579 vm_munmap(user_addr, PAGE_SIZE);
582 case CT_ACCESS_USERSPACE: {
583 unsigned long user_addr, tmp = 0;
586 user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
587 PROT_READ | PROT_WRITE | PROT_EXEC,
588 MAP_ANONYMOUS | MAP_PRIVATE, 0);
589 if (user_addr >= TASK_SIZE) {
590 pr_warn("Failed to allocate user memory\n");
594 if (copy_to_user((void __user *)user_addr, &tmp, sizeof(tmp))) {
595 pr_warn("copy_to_user failed\n");
596 vm_munmap(user_addr, PAGE_SIZE);
600 ptr = (unsigned long *)user_addr;
602 pr_info("attempting bad read at %p\n", ptr);
606 pr_info("attempting bad write at %p\n", ptr);
609 vm_munmap(user_addr, PAGE_SIZE);
614 /* Explicitly cast away "const" for the test. */
615 unsigned long *ptr = (unsigned long *)&rodata;
617 pr_info("attempting bad rodata write at %p\n", ptr);
622 case CT_WRITE_RO_AFTER_INIT: {
623 unsigned long *ptr = &ro_after_init;
626 * Verify we were written to during init. Since an Oops
627 * is considered a "success", a failure is to just skip the
630 if ((*ptr & 0xAA) != 0xAA) {
631 pr_info("%p was NOT written during init!?\n", ptr);
635 pr_info("attempting bad ro_after_init write at %p\n", ptr);
640 case CT_WRITE_KERN: {
644 size = (unsigned long)do_overwritten -
645 (unsigned long)do_nothing;
646 ptr = (unsigned char *)do_overwritten;
648 pr_info("attempting bad %zu byte write at %p\n", size, ptr);
649 memcpy(ptr, (unsigned char *)do_nothing, size);
650 flush_icache_range((unsigned long)ptr,
651 (unsigned long)(ptr + size));
656 case CT_WRAP_ATOMIC: {
657 atomic_t under = ATOMIC_INIT(INT_MIN);
658 atomic_t over = ATOMIC_INIT(INT_MAX);
660 pr_info("attempting atomic underflow\n");
662 pr_info("attempting atomic overflow\n");
674 static void lkdtm_handler(void)
679 spin_lock_irqsave(&count_lock, flags);
681 pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
682 cp_name_to_str(cpoint), cp_type_to_str(cptype), count);
686 count = cpoint_count;
688 spin_unlock_irqrestore(&count_lock, flags);
691 lkdtm_do_action(cptype);
694 static int lkdtm_register_cpoint(enum cname which)
699 if (lkdtm.entry != NULL)
700 unregister_jprobe(&lkdtm);
704 lkdtm_do_action(cptype);
706 case CN_INT_HARDWARE_ENTRY:
707 lkdtm.kp.symbol_name = "do_IRQ";
708 lkdtm.entry = (kprobe_opcode_t*) jp_do_irq;
710 case CN_INT_HW_IRQ_EN:
711 lkdtm.kp.symbol_name = "handle_IRQ_event";
712 lkdtm.entry = (kprobe_opcode_t*) jp_handle_irq_event;
714 case CN_INT_TASKLET_ENTRY:
715 lkdtm.kp.symbol_name = "tasklet_action";
716 lkdtm.entry = (kprobe_opcode_t*) jp_tasklet_action;
719 lkdtm.kp.symbol_name = "ll_rw_block";
720 lkdtm.entry = (kprobe_opcode_t*) jp_ll_rw_block;
723 lkdtm.kp.symbol_name = "shrink_inactive_list";
724 lkdtm.entry = (kprobe_opcode_t*) jp_shrink_inactive_list;
727 lkdtm.kp.symbol_name = "hrtimer_start";
728 lkdtm.entry = (kprobe_opcode_t*) jp_hrtimer_start;
730 case CN_SCSI_DISPATCH_CMD:
731 lkdtm.kp.symbol_name = "scsi_dispatch_cmd";
732 lkdtm.entry = (kprobe_opcode_t*) jp_scsi_dispatch_cmd;
736 lkdtm.kp.symbol_name = "generic_ide_ioctl";
737 lkdtm.entry = (kprobe_opcode_t*) jp_generic_ide_ioctl;
739 pr_info("Crash point not available\n");
744 pr_info("Invalid Crash Point\n");
749 if ((ret = register_jprobe(&lkdtm)) < 0) {
750 pr_info("Couldn't register jprobe\n");
757 static ssize_t do_register_entry(enum cname which, struct file *f,
758 const char __user *user_buf, size_t count, loff_t *off)
763 if (count >= PAGE_SIZE)
766 buf = (char *)__get_free_page(GFP_KERNEL);
769 if (copy_from_user(buf, user_buf, count)) {
770 free_page((unsigned long) buf);
773 /* NULL-terminate and remove enter */
777 cptype = parse_cp_type(buf, count);
778 free_page((unsigned long) buf);
780 if (cptype == CT_NONE)
783 err = lkdtm_register_cpoint(which);
792 /* Generic read callback that just prints out the available crash types */
793 static ssize_t lkdtm_debugfs_read(struct file *f, char __user *user_buf,
794 size_t count, loff_t *off)
799 buf = (char *)__get_free_page(GFP_KERNEL);
803 n = snprintf(buf, PAGE_SIZE, "Available crash types:\n");
804 for (i = 0; i < ARRAY_SIZE(cp_type); i++)
805 n += snprintf(buf + n, PAGE_SIZE - n, "%s\n", cp_type[i]);
808 out = simple_read_from_buffer(user_buf, count, off,
810 free_page((unsigned long) buf);
815 static int lkdtm_debugfs_open(struct inode *inode, struct file *file)
821 static ssize_t int_hardware_entry(struct file *f, const char __user *buf,
822 size_t count, loff_t *off)
824 return do_register_entry(CN_INT_HARDWARE_ENTRY, f, buf, count, off);
827 static ssize_t int_hw_irq_en(struct file *f, const char __user *buf,
828 size_t count, loff_t *off)
830 return do_register_entry(CN_INT_HW_IRQ_EN, f, buf, count, off);
833 static ssize_t int_tasklet_entry(struct file *f, const char __user *buf,
834 size_t count, loff_t *off)
836 return do_register_entry(CN_INT_TASKLET_ENTRY, f, buf, count, off);
839 static ssize_t fs_devrw_entry(struct file *f, const char __user *buf,
840 size_t count, loff_t *off)
842 return do_register_entry(CN_FS_DEVRW, f, buf, count, off);
845 static ssize_t mem_swapout_entry(struct file *f, const char __user *buf,
846 size_t count, loff_t *off)
848 return do_register_entry(CN_MEM_SWAPOUT, f, buf, count, off);
851 static ssize_t timeradd_entry(struct file *f, const char __user *buf,
852 size_t count, loff_t *off)
854 return do_register_entry(CN_TIMERADD, f, buf, count, off);
857 static ssize_t scsi_dispatch_cmd_entry(struct file *f,
858 const char __user *buf, size_t count, loff_t *off)
860 return do_register_entry(CN_SCSI_DISPATCH_CMD, f, buf, count, off);
863 static ssize_t ide_core_cp_entry(struct file *f, const char __user *buf,
864 size_t count, loff_t *off)
866 return do_register_entry(CN_IDE_CORE_CP, f, buf, count, off);
869 /* Special entry to just crash directly. Available without KPROBEs */
870 static ssize_t direct_entry(struct file *f, const char __user *user_buf,
871 size_t count, loff_t *off)
876 if (count >= PAGE_SIZE)
881 buf = (char *)__get_free_page(GFP_KERNEL);
884 if (copy_from_user(buf, user_buf, count)) {
885 free_page((unsigned long) buf);
888 /* NULL-terminate and remove enter */
892 type = parse_cp_type(buf, count);
893 free_page((unsigned long) buf);
897 pr_info("Performing direct entry %s\n", cp_type_to_str(type));
898 lkdtm_do_action(type);
906 const struct file_operations fops;
909 static const struct crash_entry crash_entries[] = {
910 {"DIRECT", {.read = lkdtm_debugfs_read,
911 .llseek = generic_file_llseek,
912 .open = lkdtm_debugfs_open,
913 .write = direct_entry} },
914 {"INT_HARDWARE_ENTRY", {.read = lkdtm_debugfs_read,
915 .llseek = generic_file_llseek,
916 .open = lkdtm_debugfs_open,
917 .write = int_hardware_entry} },
918 {"INT_HW_IRQ_EN", {.read = lkdtm_debugfs_read,
919 .llseek = generic_file_llseek,
920 .open = lkdtm_debugfs_open,
921 .write = int_hw_irq_en} },
922 {"INT_TASKLET_ENTRY", {.read = lkdtm_debugfs_read,
923 .llseek = generic_file_llseek,
924 .open = lkdtm_debugfs_open,
925 .write = int_tasklet_entry} },
926 {"FS_DEVRW", {.read = lkdtm_debugfs_read,
927 .llseek = generic_file_llseek,
928 .open = lkdtm_debugfs_open,
929 .write = fs_devrw_entry} },
930 {"MEM_SWAPOUT", {.read = lkdtm_debugfs_read,
931 .llseek = generic_file_llseek,
932 .open = lkdtm_debugfs_open,
933 .write = mem_swapout_entry} },
934 {"TIMERADD", {.read = lkdtm_debugfs_read,
935 .llseek = generic_file_llseek,
936 .open = lkdtm_debugfs_open,
937 .write = timeradd_entry} },
938 {"SCSI_DISPATCH_CMD", {.read = lkdtm_debugfs_read,
939 .llseek = generic_file_llseek,
940 .open = lkdtm_debugfs_open,
941 .write = scsi_dispatch_cmd_entry} },
942 {"IDE_CORE_CP", {.read = lkdtm_debugfs_read,
943 .llseek = generic_file_llseek,
944 .open = lkdtm_debugfs_open,
945 .write = ide_core_cp_entry} },
948 static struct dentry *lkdtm_debugfs_root;
950 static int __init lkdtm_module_init(void)
953 int n_debugfs_entries = 1; /* Assume only the direct entry */
956 /* Make sure we can write to __ro_after_init values during __init */
957 ro_after_init |= 0xAA;
959 /* Register debugfs interface */
960 lkdtm_debugfs_root = debugfs_create_dir("provoke-crash", NULL);
961 if (!lkdtm_debugfs_root) {
962 pr_err("creating root dir failed\n");
966 #ifdef CONFIG_KPROBES
967 n_debugfs_entries = ARRAY_SIZE(crash_entries);
970 for (i = 0; i < n_debugfs_entries; i++) {
971 const struct crash_entry *cur = &crash_entries[i];
974 de = debugfs_create_file(cur->name, 0644, lkdtm_debugfs_root,
977 pr_err("could not create %s\n", cur->name);
982 if (lkdtm_parse_commandline() == -EINVAL) {
983 pr_info("Invalid command\n");
987 if (cpoint != CN_INVALID && cptype != CT_NONE) {
988 ret = lkdtm_register_cpoint(cpoint);
990 pr_info("Invalid crash point %d\n", cpoint);
993 pr_info("Crash point %s of type %s registered\n",
994 cpoint_name, cpoint_type);
996 pr_info("No crash points registered, enable through debugfs\n");
1002 debugfs_remove_recursive(lkdtm_debugfs_root);
1006 static void __exit lkdtm_module_exit(void)
1008 debugfs_remove_recursive(lkdtm_debugfs_root);
1010 unregister_jprobe(&lkdtm);
1011 pr_info("Crash point unregistered\n");
1014 module_init(lkdtm_module_init);
1015 module_exit(lkdtm_module_exit);
1017 MODULE_LICENSE("GPL");
1018 MODULE_DESCRIPTION("Kprobe module for testing crash dumps");