lkdtm: split build into multiple source files
authorKees Cook <keescook@chromium.org>
Mon, 22 Feb 2016 22:05:44 +0000 (14:05 -0800)
committerKees Cook <keescook@chromium.org>
Fri, 10 Jun 2016 22:57:48 +0000 (15:57 -0700)
Kbuild lacks a way to do in-place objcopy or other modifications of
built targets, so in order to move functions into non-text sections
without renaming the kernel module, the build of lkdtm must be split
into separate source files. This renames lkdtm.c to lkdtm_core.c in
preparation for adding the source file for the .rodata section.

Signed-off-by: Kees Cook <keescook@chromium.org>
drivers/misc/Makefile
drivers/misc/lkdtm.c [deleted file]
drivers/misc/lkdtm_core.c [new file with mode: 0644]

index b2fb6db..c3cb6ad 100644 (file)
@@ -57,3 +57,5 @@ obj-$(CONFIG_ECHO)            += echo/
 obj-$(CONFIG_VEXPRESS_SYSCFG)  += vexpress-syscfg.o
 obj-$(CONFIG_CXL_BASE)         += cxl/
 obj-$(CONFIG_PANEL)             += panel.o
+
+lkdtm-$(CONFIG_LKDTM)          += lkdtm_core.o
diff --git a/drivers/misc/lkdtm.c b/drivers/misc/lkdtm.c
deleted file mode 100644 (file)
index 0a5cbbe..0000000
+++ /dev/null
@@ -1,1023 +0,0 @@
-/*
- * Kprobe module for testing crash dumps
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
- *
- * Copyright (C) IBM Corporation, 2006
- *
- * Author: Ankita Garg <ankita@in.ibm.com>
- *
- * This module induces system failures at predefined crashpoints to
- * evaluate the reliability of crash dumps obtained using different dumping
- * solutions.
- *
- * It is adapted from the Linux Kernel Dump Test Tool by
- * Fernando Luis Vazquez Cao <http://lkdtt.sourceforge.net>
- *
- * Debugfs support added by Simon Kagstrom <simon.kagstrom@netinsight.net>
- *
- * See Documentation/fault-injection/provoke-crashes.txt for instructions
- */
-#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
-
-#include <linux/kernel.h>
-#include <linux/fs.h>
-#include <linux/module.h>
-#include <linux/buffer_head.h>
-#include <linux/kprobes.h>
-#include <linux/list.h>
-#include <linux/init.h>
-#include <linux/interrupt.h>
-#include <linux/hrtimer.h>
-#include <linux/slab.h>
-#include <scsi/scsi_cmnd.h>
-#include <linux/debugfs.h>
-#include <linux/vmalloc.h>
-#include <linux/mman.h>
-#include <asm/cacheflush.h>
-
-#ifdef CONFIG_IDE
-#include <linux/ide.h>
-#endif
-
-/*
- * Make sure our attempts to over run the kernel stack doesn't trigger
- * a compiler warning when CONFIG_FRAME_WARN is set. Then make sure we
- * recurse past the end of THREAD_SIZE by default.
- */
-#if defined(CONFIG_FRAME_WARN) && (CONFIG_FRAME_WARN > 0)
-#define REC_STACK_SIZE (CONFIG_FRAME_WARN / 2)
-#else
-#define REC_STACK_SIZE (THREAD_SIZE / 8)
-#endif
-#define REC_NUM_DEFAULT ((THREAD_SIZE / REC_STACK_SIZE) * 2)
-
-#define DEFAULT_COUNT 10
-#define EXEC_SIZE 64
-
-enum cname {
-       CN_INVALID,
-       CN_INT_HARDWARE_ENTRY,
-       CN_INT_HW_IRQ_EN,
-       CN_INT_TASKLET_ENTRY,
-       CN_FS_DEVRW,
-       CN_MEM_SWAPOUT,
-       CN_TIMERADD,
-       CN_SCSI_DISPATCH_CMD,
-       CN_IDE_CORE_CP,
-       CN_DIRECT,
-};
-
-enum ctype {
-       CT_NONE,
-       CT_PANIC,
-       CT_BUG,
-       CT_WARNING,
-       CT_EXCEPTION,
-       CT_LOOP,
-       CT_OVERFLOW,
-       CT_CORRUPT_STACK,
-       CT_UNALIGNED_LOAD_STORE_WRITE,
-       CT_OVERWRITE_ALLOCATION,
-       CT_WRITE_AFTER_FREE,
-       CT_READ_AFTER_FREE,
-       CT_WRITE_BUDDY_AFTER_FREE,
-       CT_READ_BUDDY_AFTER_FREE,
-       CT_SOFTLOCKUP,
-       CT_HARDLOCKUP,
-       CT_SPINLOCKUP,
-       CT_HUNG_TASK,
-       CT_EXEC_DATA,
-       CT_EXEC_STACK,
-       CT_EXEC_KMALLOC,
-       CT_EXEC_VMALLOC,
-       CT_EXEC_USERSPACE,
-       CT_ACCESS_USERSPACE,
-       CT_WRITE_RO,
-       CT_WRITE_RO_AFTER_INIT,
-       CT_WRITE_KERN,
-       CT_WRAP_ATOMIC
-};
-
-static char* cp_name[] = {
-       "INT_HARDWARE_ENTRY",
-       "INT_HW_IRQ_EN",
-       "INT_TASKLET_ENTRY",
-       "FS_DEVRW",
-       "MEM_SWAPOUT",
-       "TIMERADD",
-       "SCSI_DISPATCH_CMD",
-       "IDE_CORE_CP",
-       "DIRECT",
-};
-
-static char* cp_type[] = {
-       "PANIC",
-       "BUG",
-       "WARNING",
-       "EXCEPTION",
-       "LOOP",
-       "OVERFLOW",
-       "CORRUPT_STACK",
-       "UNALIGNED_LOAD_STORE_WRITE",
-       "OVERWRITE_ALLOCATION",
-       "WRITE_AFTER_FREE",
-       "READ_AFTER_FREE",
-       "WRITE_BUDDY_AFTER_FREE",
-       "READ_BUDDY_AFTER_FREE",
-       "SOFTLOCKUP",
-       "HARDLOCKUP",
-       "SPINLOCKUP",
-       "HUNG_TASK",
-       "EXEC_DATA",
-       "EXEC_STACK",
-       "EXEC_KMALLOC",
-       "EXEC_VMALLOC",
-       "EXEC_USERSPACE",
-       "ACCESS_USERSPACE",
-       "WRITE_RO",
-       "WRITE_RO_AFTER_INIT",
-       "WRITE_KERN",
-       "WRAP_ATOMIC"
-};
-
-static struct jprobe lkdtm;
-
-static int lkdtm_parse_commandline(void);
-static void lkdtm_handler(void);
-
-static char* cpoint_name;
-static char* cpoint_type;
-static int cpoint_count = DEFAULT_COUNT;
-static int recur_count = REC_NUM_DEFAULT;
-
-static enum cname cpoint = CN_INVALID;
-static enum ctype cptype = CT_NONE;
-static int count = DEFAULT_COUNT;
-static DEFINE_SPINLOCK(count_lock);
-static DEFINE_SPINLOCK(lock_me_up);
-
-static u8 data_area[EXEC_SIZE];
-
-static const unsigned long rodata = 0xAA55AA55;
-static unsigned long ro_after_init __ro_after_init = 0x55AA5500;
-
-module_param(recur_count, int, 0644);
-MODULE_PARM_DESC(recur_count, " Recursion level for the stack overflow test");
-module_param(cpoint_name, charp, 0444);
-MODULE_PARM_DESC(cpoint_name, " Crash Point, where kernel is to be crashed");
-module_param(cpoint_type, charp, 0444);
-MODULE_PARM_DESC(cpoint_type, " Crash Point Type, action to be taken on "\
-                               "hitting the crash point");
-module_param(cpoint_count, int, 0644);
-MODULE_PARM_DESC(cpoint_count, " Crash Point Count, number of times the "\
-                               "crash point is to be hit to trigger action");
-
-static unsigned int jp_do_irq(unsigned int irq)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-
-static irqreturn_t jp_handle_irq_event(unsigned int irq,
-                                      struct irqaction *action)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-
-static void jp_tasklet_action(struct softirq_action *a)
-{
-       lkdtm_handler();
-       jprobe_return();
-}
-
-static void jp_ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
-{
-       lkdtm_handler();
-       jprobe_return();
-}
-
-struct scan_control;
-
-static unsigned long jp_shrink_inactive_list(unsigned long max_scan,
-                                            struct zone *zone,
-                                            struct scan_control *sc)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-
-static int jp_hrtimer_start(struct hrtimer *timer, ktime_t tim,
-                           const enum hrtimer_mode mode)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-
-static int jp_scsi_dispatch_cmd(struct scsi_cmnd *cmd)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-
-#ifdef CONFIG_IDE
-static int jp_generic_ide_ioctl(ide_drive_t *drive, struct file *file,
-                       struct block_device *bdev, unsigned int cmd,
-                       unsigned long arg)
-{
-       lkdtm_handler();
-       jprobe_return();
-       return 0;
-}
-#endif
-
-/* Return the crashpoint number or NONE if the name is invalid */
-static enum ctype parse_cp_type(const char *what, size_t count)
-{
-       int i;
-
-       for (i = 0; i < ARRAY_SIZE(cp_type); i++) {
-               if (!strcmp(what, cp_type[i]))
-                       return i + 1;
-       }
-
-       return CT_NONE;
-}
-
-static const char *cp_type_to_str(enum ctype type)
-{
-       if (type == CT_NONE || type < 0 || type > ARRAY_SIZE(cp_type))
-               return "None";
-
-       return cp_type[type - 1];
-}
-
-static const char *cp_name_to_str(enum cname name)
-{
-       if (name == CN_INVALID || name < 0 || name > ARRAY_SIZE(cp_name))
-               return "INVALID";
-
-       return cp_name[name - 1];
-}
-
-
-static int lkdtm_parse_commandline(void)
-{
-       int i;
-       unsigned long flags;
-
-       if (cpoint_count < 1 || recur_count < 1)
-               return -EINVAL;
-
-       spin_lock_irqsave(&count_lock, flags);
-       count = cpoint_count;
-       spin_unlock_irqrestore(&count_lock, flags);
-
-       /* No special parameters */
-       if (!cpoint_type && !cpoint_name)
-               return 0;
-
-       /* Neither or both of these need to be set */
-       if (!cpoint_type || !cpoint_name)
-               return -EINVAL;
-
-       cptype = parse_cp_type(cpoint_type, strlen(cpoint_type));
-       if (cptype == CT_NONE)
-               return -EINVAL;
-
-       for (i = 0; i < ARRAY_SIZE(cp_name); i++) {
-               if (!strcmp(cpoint_name, cp_name[i])) {
-                       cpoint = i + 1;
-                       return 0;
-               }
-       }
-
-       /* Could not find a valid crash point */
-       return -EINVAL;
-}
-
-static int recursive_loop(int remaining)
-{
-       char buf[REC_STACK_SIZE];
-
-       /* Make sure compiler does not optimize this away. */
-       memset(buf, (remaining & 0xff) | 0x1, REC_STACK_SIZE);
-       if (!remaining)
-               return 0;
-       else
-               return recursive_loop(remaining - 1);
-}
-
-static void do_nothing(void)
-{
-       return;
-}
-
-/* Must immediately follow do_nothing for size calculuations to work out. */
-static void do_overwritten(void)
-{
-       pr_info("do_overwritten wasn't overwritten!\n");
-       return;
-}
-
-static noinline void corrupt_stack(void)
-{
-       /* Use default char array length that triggers stack protection. */
-       char data[8];
-
-       memset((void *)data, 0, 64);
-}
-
-static void noinline execute_location(void *dst)
-{
-       void (*func)(void) = dst;
-
-       pr_info("attempting ok execution at %p\n", do_nothing);
-       do_nothing();
-
-       memcpy(dst, do_nothing, EXEC_SIZE);
-       flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
-       pr_info("attempting bad execution at %p\n", func);
-       func();
-}
-
-static void execute_user_location(void *dst)
-{
-       /* Intentionally crossing kernel/user memory boundary. */
-       void (*func)(void) = dst;
-
-       pr_info("attempting ok execution at %p\n", do_nothing);
-       do_nothing();
-
-       if (copy_to_user((void __user *)dst, do_nothing, EXEC_SIZE))
-               return;
-       flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
-       pr_info("attempting bad execution at %p\n", func);
-       func();
-}
-
-static void lkdtm_do_action(enum ctype which)
-{
-       switch (which) {
-       case CT_PANIC:
-               panic("dumptest");
-               break;
-       case CT_BUG:
-               BUG();
-               break;
-       case CT_WARNING:
-               WARN_ON(1);
-               break;
-       case CT_EXCEPTION:
-               *((int *) 0) = 0;
-               break;
-       case CT_LOOP:
-               for (;;)
-                       ;
-               break;
-       case CT_OVERFLOW:
-               (void) recursive_loop(recur_count);
-               break;
-       case CT_CORRUPT_STACK:
-               corrupt_stack();
-               break;
-       case CT_UNALIGNED_LOAD_STORE_WRITE: {
-               static u8 data[5] __attribute__((aligned(4))) = {1, 2,
-                               3, 4, 5};
-               u32 *p;
-               u32 val = 0x12345678;
-
-               p = (u32 *)(data + 1);
-               if (*p == 0)
-                       val = 0x87654321;
-               *p = val;
-                break;
-       }
-       case CT_OVERWRITE_ALLOCATION: {
-               size_t len = 1020;
-               u32 *data = kmalloc(len, GFP_KERNEL);
-
-               data[1024 / sizeof(u32)] = 0x12345678;
-               kfree(data);
-               break;
-       }
-       case CT_WRITE_AFTER_FREE: {
-               int *base, *again;
-               size_t len = 1024;
-               /*
-                * The slub allocator uses the first word to store the free
-                * pointer in some configurations. Use the middle of the
-                * allocation to avoid running into the freelist
-                */
-               size_t offset = (len / sizeof(*base)) / 2;
-
-               base = kmalloc(len, GFP_KERNEL);
-               pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
-               pr_info("Attempting bad write to freed memory at %p\n",
-                       &base[offset]);
-               kfree(base);
-               base[offset] = 0x0abcdef0;
-               /* Attempt to notice the overwrite. */
-               again = kmalloc(len, GFP_KERNEL);
-               kfree(again);
-               if (again != base)
-                       pr_info("Hmm, didn't get the same memory range.\n");
-
-               break;
-       }
-       case CT_READ_AFTER_FREE: {
-               int *base, *val, saw;
-               size_t len = 1024;
-               /*
-                * The slub allocator uses the first word to store the free
-                * pointer in some configurations. Use the middle of the
-                * allocation to avoid running into the freelist
-                */
-               size_t offset = (len / sizeof(*base)) / 2;
-
-               base = kmalloc(len, GFP_KERNEL);
-               if (!base)
-                       break;
-
-               val = kmalloc(len, GFP_KERNEL);
-               if (!val) {
-                       kfree(base);
-                       break;
-               }
-
-               *val = 0x12345678;
-               base[offset] = *val;
-               pr_info("Value in memory before free: %x\n", base[offset]);
-
-               kfree(base);
-
-               pr_info("Attempting bad read from freed memory\n");
-               saw = base[offset];
-               if (saw != *val) {
-                       /* Good! Poisoning happened, so declare a win. */
-                       pr_info("Memory correctly poisoned (%x)\n", saw);
-                       BUG();
-               }
-               pr_info("Memory was not poisoned\n");
-
-               kfree(val);
-               break;
-       }
-       case CT_WRITE_BUDDY_AFTER_FREE: {
-               unsigned long p = __get_free_page(GFP_KERNEL);
-               if (!p)
-                       break;
-               pr_info("Writing to the buddy page before free\n");
-               memset((void *)p, 0x3, PAGE_SIZE);
-               free_page(p);
-               schedule();
-               pr_info("Attempting bad write to the buddy page after free\n");
-               memset((void *)p, 0x78, PAGE_SIZE);
-               /* Attempt to notice the overwrite. */
-               p = __get_free_page(GFP_KERNEL);
-               free_page(p);
-               schedule();
-
-               break;
-       }
-       case CT_READ_BUDDY_AFTER_FREE: {
-               unsigned long p = __get_free_page(GFP_KERNEL);
-               int saw, *val;
-               int *base;
-
-               if (!p)
-                       break;
-
-               val = kmalloc(1024, GFP_KERNEL);
-               if (!val) {
-                       free_page(p);
-                       break;
-               }
-
-               base = (int *)p;
-
-               *val = 0x12345678;
-               base[0] = *val;
-               pr_info("Value in memory before free: %x\n", base[0]);
-               free_page(p);
-               pr_info("Attempting to read from freed memory\n");
-               saw = base[0];
-               if (saw != *val) {
-                       /* Good! Poisoning happened, so declare a win. */
-                       pr_info("Memory correctly poisoned (%x)\n", saw);
-                       BUG();
-               }
-               pr_info("Buddy page was not poisoned\n");
-
-               kfree(val);
-               break;
-       }
-       case CT_SOFTLOCKUP:
-               preempt_disable();
-               for (;;)
-                       cpu_relax();
-               break;
-       case CT_HARDLOCKUP:
-               local_irq_disable();
-               for (;;)
-                       cpu_relax();
-               break;
-       case CT_SPINLOCKUP:
-               /* Must be called twice to trigger. */
-               spin_lock(&lock_me_up);
-               /* Let sparse know we intended to exit holding the lock. */
-               __release(&lock_me_up);
-               break;
-       case CT_HUNG_TASK:
-               set_current_state(TASK_UNINTERRUPTIBLE);
-               schedule();
-               break;
-       case CT_EXEC_DATA:
-               execute_location(data_area);
-               break;
-       case CT_EXEC_STACK: {
-               u8 stack_area[EXEC_SIZE];
-               execute_location(stack_area);
-               break;
-       }
-       case CT_EXEC_KMALLOC: {
-               u32 *kmalloc_area = kmalloc(EXEC_SIZE, GFP_KERNEL);
-               execute_location(kmalloc_area);
-               kfree(kmalloc_area);
-               break;
-       }
-       case CT_EXEC_VMALLOC: {
-               u32 *vmalloc_area = vmalloc(EXEC_SIZE);
-               execute_location(vmalloc_area);
-               vfree(vmalloc_area);
-               break;
-       }
-       case CT_EXEC_USERSPACE: {
-               unsigned long user_addr;
-
-               user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
-                                   PROT_READ | PROT_WRITE | PROT_EXEC,
-                                   MAP_ANONYMOUS | MAP_PRIVATE, 0);
-               if (user_addr >= TASK_SIZE) {
-                       pr_warn("Failed to allocate user memory\n");
-                       return;
-               }
-               execute_user_location((void *)user_addr);
-               vm_munmap(user_addr, PAGE_SIZE);
-               break;
-       }
-       case CT_ACCESS_USERSPACE: {
-               unsigned long user_addr, tmp = 0;
-               unsigned long *ptr;
-
-               user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
-                                   PROT_READ | PROT_WRITE | PROT_EXEC,
-                                   MAP_ANONYMOUS | MAP_PRIVATE, 0);
-               if (user_addr >= TASK_SIZE) {
-                       pr_warn("Failed to allocate user memory\n");
-                       return;
-               }
-
-               if (copy_to_user((void __user *)user_addr, &tmp, sizeof(tmp))) {
-                       pr_warn("copy_to_user failed\n");
-                       vm_munmap(user_addr, PAGE_SIZE);
-                       return;
-               }
-
-               ptr = (unsigned long *)user_addr;
-
-               pr_info("attempting bad read at %p\n", ptr);
-               tmp = *ptr;
-               tmp += 0xc0dec0de;
-
-               pr_info("attempting bad write at %p\n", ptr);
-               *ptr = tmp;
-
-               vm_munmap(user_addr, PAGE_SIZE);
-
-               break;
-       }
-       case CT_WRITE_RO: {
-               /* Explicitly cast away "const" for the test. */
-               unsigned long *ptr = (unsigned long *)&rodata;
-
-               pr_info("attempting bad rodata write at %p\n", ptr);
-               *ptr ^= 0xabcd1234;
-
-               break;
-       }
-       case CT_WRITE_RO_AFTER_INIT: {
-               unsigned long *ptr = &ro_after_init;
-
-               /*
-                * Verify we were written to during init. Since an Oops
-                * is considered a "success", a failure is to just skip the
-                * real test.
-                */
-               if ((*ptr & 0xAA) != 0xAA) {
-                       pr_info("%p was NOT written during init!?\n", ptr);
-                       break;
-               }
-
-               pr_info("attempting bad ro_after_init write at %p\n", ptr);
-               *ptr ^= 0xabcd1234;
-
-               break;
-       }
-       case CT_WRITE_KERN: {
-               size_t size;
-               unsigned char *ptr;
-
-               size = (unsigned long)do_overwritten -
-                      (unsigned long)do_nothing;
-               ptr = (unsigned char *)do_overwritten;
-
-               pr_info("attempting bad %zu byte write at %p\n", size, ptr);
-               memcpy(ptr, (unsigned char *)do_nothing, size);
-               flush_icache_range((unsigned long)ptr,
-                                  (unsigned long)(ptr + size));
-
-               do_overwritten();
-               break;
-       }
-       case CT_WRAP_ATOMIC: {
-               atomic_t under = ATOMIC_INIT(INT_MIN);
-               atomic_t over = ATOMIC_INIT(INT_MAX);
-
-               pr_info("attempting atomic underflow\n");
-               atomic_dec(&under);
-               pr_info("attempting atomic overflow\n");
-               atomic_inc(&over);
-
-               return;
-       }
-       case CT_NONE:
-       default:
-               break;
-       }
-
-}
-
-static void lkdtm_handler(void)
-{
-       unsigned long flags;
-       bool do_it = false;
-
-       spin_lock_irqsave(&count_lock, flags);
-       count--;
-       pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
-               cp_name_to_str(cpoint), cp_type_to_str(cptype), count);
-
-       if (count == 0) {
-               do_it = true;
-               count = cpoint_count;
-       }
-       spin_unlock_irqrestore(&count_lock, flags);
-
-       if (do_it)
-               lkdtm_do_action(cptype);
-}
-
-static int lkdtm_register_cpoint(enum cname which)
-{
-       int ret;
-
-       cpoint = CN_INVALID;
-       if (lkdtm.entry != NULL)
-               unregister_jprobe(&lkdtm);
-
-       switch (which) {
-       case CN_DIRECT:
-               lkdtm_do_action(cptype);
-               return 0;
-       case CN_INT_HARDWARE_ENTRY:
-               lkdtm.kp.symbol_name = "do_IRQ";
-               lkdtm.entry = (kprobe_opcode_t*) jp_do_irq;
-               break;
-       case CN_INT_HW_IRQ_EN:
-               lkdtm.kp.symbol_name = "handle_IRQ_event";
-               lkdtm.entry = (kprobe_opcode_t*) jp_handle_irq_event;
-               break;
-       case CN_INT_TASKLET_ENTRY:
-               lkdtm.kp.symbol_name = "tasklet_action";
-               lkdtm.entry = (kprobe_opcode_t*) jp_tasklet_action;
-               break;
-       case CN_FS_DEVRW:
-               lkdtm.kp.symbol_name = "ll_rw_block";
-               lkdtm.entry = (kprobe_opcode_t*) jp_ll_rw_block;
-               break;
-       case CN_MEM_SWAPOUT:
-               lkdtm.kp.symbol_name = "shrink_inactive_list";
-               lkdtm.entry = (kprobe_opcode_t*) jp_shrink_inactive_list;
-               break;
-       case CN_TIMERADD:
-               lkdtm.kp.symbol_name = "hrtimer_start";
-               lkdtm.entry = (kprobe_opcode_t*) jp_hrtimer_start;
-               break;
-       case CN_SCSI_DISPATCH_CMD:
-               lkdtm.kp.symbol_name = "scsi_dispatch_cmd";
-               lkdtm.entry = (kprobe_opcode_t*) jp_scsi_dispatch_cmd;
-               break;
-       case CN_IDE_CORE_CP:
-#ifdef CONFIG_IDE
-               lkdtm.kp.symbol_name = "generic_ide_ioctl";
-               lkdtm.entry = (kprobe_opcode_t*) jp_generic_ide_ioctl;
-#else
-               pr_info("Crash point not available\n");
-               return -EINVAL;
-#endif
-               break;
-       default:
-               pr_info("Invalid Crash Point\n");
-               return -EINVAL;
-       }
-
-       cpoint = which;
-       if ((ret = register_jprobe(&lkdtm)) < 0) {
-               pr_info("Couldn't register jprobe\n");
-               cpoint = CN_INVALID;
-       }
-
-       return ret;
-}
-
-static ssize_t do_register_entry(enum cname which, struct file *f,
-               const char __user *user_buf, size_t count, loff_t *off)
-{
-       char *buf;
-       int err;
-
-       if (count >= PAGE_SIZE)
-               return -EINVAL;
-
-       buf = (char *)__get_free_page(GFP_KERNEL);
-       if (!buf)
-               return -ENOMEM;
-       if (copy_from_user(buf, user_buf, count)) {
-               free_page((unsigned long) buf);
-               return -EFAULT;
-       }
-       /* NULL-terminate and remove enter */
-       buf[count] = '\0';
-       strim(buf);
-
-       cptype = parse_cp_type(buf, count);
-       free_page((unsigned long) buf);
-
-       if (cptype == CT_NONE)
-               return -EINVAL;
-
-       err = lkdtm_register_cpoint(which);
-       if (err < 0)
-               return err;
-
-       *off += count;
-
-       return count;
-}
-
-/* Generic read callback that just prints out the available crash types */
-static ssize_t lkdtm_debugfs_read(struct file *f, char __user *user_buf,
-               size_t count, loff_t *off)
-{
-       char *buf;
-       int i, n, out;
-
-       buf = (char *)__get_free_page(GFP_KERNEL);
-       if (buf == NULL)
-               return -ENOMEM;
-
-       n = snprintf(buf, PAGE_SIZE, "Available crash types:\n");
-       for (i = 0; i < ARRAY_SIZE(cp_type); i++)
-               n += snprintf(buf + n, PAGE_SIZE - n, "%s\n", cp_type[i]);
-       buf[n] = '\0';
-
-       out = simple_read_from_buffer(user_buf, count, off,
-                                     buf, n);
-       free_page((unsigned long) buf);
-
-       return out;
-}
-
-static int lkdtm_debugfs_open(struct inode *inode, struct file *file)
-{
-       return 0;
-}
-
-
-static ssize_t int_hardware_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_INT_HARDWARE_ENTRY, f, buf, count, off);
-}
-
-static ssize_t int_hw_irq_en(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_INT_HW_IRQ_EN, f, buf, count, off);
-}
-
-static ssize_t int_tasklet_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_INT_TASKLET_ENTRY, f, buf, count, off);
-}
-
-static ssize_t fs_devrw_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_FS_DEVRW, f, buf, count, off);
-}
-
-static ssize_t mem_swapout_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_MEM_SWAPOUT, f, buf, count, off);
-}
-
-static ssize_t timeradd_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_TIMERADD, f, buf, count, off);
-}
-
-static ssize_t scsi_dispatch_cmd_entry(struct file *f,
-               const char __user *buf, size_t count, loff_t *off)
-{
-       return do_register_entry(CN_SCSI_DISPATCH_CMD, f, buf, count, off);
-}
-
-static ssize_t ide_core_cp_entry(struct file *f, const char __user *buf,
-               size_t count, loff_t *off)
-{
-       return do_register_entry(CN_IDE_CORE_CP, f, buf, count, off);
-}
-
-/* Special entry to just crash directly. Available without KPROBEs */
-static ssize_t direct_entry(struct file *f, const char __user *user_buf,
-               size_t count, loff_t *off)
-{
-       enum ctype type;
-       char *buf;
-
-       if (count >= PAGE_SIZE)
-               return -EINVAL;
-       if (count < 1)
-               return -EINVAL;
-
-       buf = (char *)__get_free_page(GFP_KERNEL);
-       if (!buf)
-               return -ENOMEM;
-       if (copy_from_user(buf, user_buf, count)) {
-               free_page((unsigned long) buf);
-               return -EFAULT;
-       }
-       /* NULL-terminate and remove enter */
-       buf[count] = '\0';
-       strim(buf);
-
-       type = parse_cp_type(buf, count);
-       free_page((unsigned long) buf);
-       if (type == CT_NONE)
-               return -EINVAL;
-
-       pr_info("Performing direct entry %s\n", cp_type_to_str(type));
-       lkdtm_do_action(type);
-       *off += count;
-
-       return count;
-}
-
-struct crash_entry {
-       const char *name;
-       const struct file_operations fops;
-};
-
-static const struct crash_entry crash_entries[] = {
-       {"DIRECT", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = direct_entry} },
-       {"INT_HARDWARE_ENTRY", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = int_hardware_entry} },
-       {"INT_HW_IRQ_EN", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = int_hw_irq_en} },
-       {"INT_TASKLET_ENTRY", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = int_tasklet_entry} },
-       {"FS_DEVRW", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = fs_devrw_entry} },
-       {"MEM_SWAPOUT", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = mem_swapout_entry} },
-       {"TIMERADD", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = timeradd_entry} },
-       {"SCSI_DISPATCH_CMD", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = scsi_dispatch_cmd_entry} },
-       {"IDE_CORE_CP", {.read = lkdtm_debugfs_read,
-                       .llseek = generic_file_llseek,
-                       .open = lkdtm_debugfs_open,
-                       .write = ide_core_cp_entry} },
-};
-
-static struct dentry *lkdtm_debugfs_root;
-
-static int __init lkdtm_module_init(void)
-{
-       int ret = -EINVAL;
-       int n_debugfs_entries = 1; /* Assume only the direct entry */
-       int i;
-
-       /* Make sure we can write to __ro_after_init values during __init */
-       ro_after_init |= 0xAA;
-
-       /* Register debugfs interface */
-       lkdtm_debugfs_root = debugfs_create_dir("provoke-crash", NULL);
-       if (!lkdtm_debugfs_root) {
-               pr_err("creating root dir failed\n");
-               return -ENODEV;
-       }
-
-#ifdef CONFIG_KPROBES
-       n_debugfs_entries = ARRAY_SIZE(crash_entries);
-#endif
-
-       for (i = 0; i < n_debugfs_entries; i++) {
-               const struct crash_entry *cur = &crash_entries[i];
-               struct dentry *de;
-
-               de = debugfs_create_file(cur->name, 0644, lkdtm_debugfs_root,
-                               NULL, &cur->fops);
-               if (de == NULL) {
-                       pr_err("could not create %s\n", cur->name);
-                       goto out_err;
-               }
-       }
-
-       if (lkdtm_parse_commandline() == -EINVAL) {
-               pr_info("Invalid command\n");
-               goto out_err;
-       }
-
-       if (cpoint != CN_INVALID && cptype != CT_NONE) {
-               ret = lkdtm_register_cpoint(cpoint);
-               if (ret < 0) {
-                       pr_info("Invalid crash point %d\n", cpoint);
-                       goto out_err;
-               }
-               pr_info("Crash point %s of type %s registered\n",
-                       cpoint_name, cpoint_type);
-       } else {
-               pr_info("No crash points registered, enable through debugfs\n");
-       }
-
-       return 0;
-
-out_err:
-       debugfs_remove_recursive(lkdtm_debugfs_root);
-       return ret;
-}
-
-static void __exit lkdtm_module_exit(void)
-{
-       debugfs_remove_recursive(lkdtm_debugfs_root);
-
-       unregister_jprobe(&lkdtm);
-       pr_info("Crash point unregistered\n");
-}
-
-module_init(lkdtm_module_init);
-module_exit(lkdtm_module_exit);
-
-MODULE_LICENSE("GPL");
-MODULE_DESCRIPTION("Kprobe module for testing crash dumps");
diff --git a/drivers/misc/lkdtm_core.c b/drivers/misc/lkdtm_core.c
new file mode 100644 (file)
index 0000000..0a5cbbe
--- /dev/null
@@ -0,0 +1,1023 @@
+/*
+ * Kprobe module for testing crash dumps
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2006
+ *
+ * Author: Ankita Garg <ankita@in.ibm.com>
+ *
+ * This module induces system failures at predefined crashpoints to
+ * evaluate the reliability of crash dumps obtained using different dumping
+ * solutions.
+ *
+ * It is adapted from the Linux Kernel Dump Test Tool by
+ * Fernando Luis Vazquez Cao <http://lkdtt.sourceforge.net>
+ *
+ * Debugfs support added by Simon Kagstrom <simon.kagstrom@netinsight.net>
+ *
+ * See Documentation/fault-injection/provoke-crashes.txt for instructions
+ */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/fs.h>
+#include <linux/module.h>
+#include <linux/buffer_head.h>
+#include <linux/kprobes.h>
+#include <linux/list.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/hrtimer.h>
+#include <linux/slab.h>
+#include <scsi/scsi_cmnd.h>
+#include <linux/debugfs.h>
+#include <linux/vmalloc.h>
+#include <linux/mman.h>
+#include <asm/cacheflush.h>
+
+#ifdef CONFIG_IDE
+#include <linux/ide.h>
+#endif
+
+/*
+ * Make sure our attempts to over run the kernel stack doesn't trigger
+ * a compiler warning when CONFIG_FRAME_WARN is set. Then make sure we
+ * recurse past the end of THREAD_SIZE by default.
+ */
+#if defined(CONFIG_FRAME_WARN) && (CONFIG_FRAME_WARN > 0)
+#define REC_STACK_SIZE (CONFIG_FRAME_WARN / 2)
+#else
+#define REC_STACK_SIZE (THREAD_SIZE / 8)
+#endif
+#define REC_NUM_DEFAULT ((THREAD_SIZE / REC_STACK_SIZE) * 2)
+
+#define DEFAULT_COUNT 10
+#define EXEC_SIZE 64
+
+enum cname {
+       CN_INVALID,
+       CN_INT_HARDWARE_ENTRY,
+       CN_INT_HW_IRQ_EN,
+       CN_INT_TASKLET_ENTRY,
+       CN_FS_DEVRW,
+       CN_MEM_SWAPOUT,
+       CN_TIMERADD,
+       CN_SCSI_DISPATCH_CMD,
+       CN_IDE_CORE_CP,
+       CN_DIRECT,
+};
+
+enum ctype {
+       CT_NONE,
+       CT_PANIC,
+       CT_BUG,
+       CT_WARNING,
+       CT_EXCEPTION,
+       CT_LOOP,
+       CT_OVERFLOW,
+       CT_CORRUPT_STACK,
+       CT_UNALIGNED_LOAD_STORE_WRITE,
+       CT_OVERWRITE_ALLOCATION,
+       CT_WRITE_AFTER_FREE,
+       CT_READ_AFTER_FREE,
+       CT_WRITE_BUDDY_AFTER_FREE,
+       CT_READ_BUDDY_AFTER_FREE,
+       CT_SOFTLOCKUP,
+       CT_HARDLOCKUP,
+       CT_SPINLOCKUP,
+       CT_HUNG_TASK,
+       CT_EXEC_DATA,
+       CT_EXEC_STACK,
+       CT_EXEC_KMALLOC,
+       CT_EXEC_VMALLOC,
+       CT_EXEC_USERSPACE,
+       CT_ACCESS_USERSPACE,
+       CT_WRITE_RO,
+       CT_WRITE_RO_AFTER_INIT,
+       CT_WRITE_KERN,
+       CT_WRAP_ATOMIC
+};
+
+static char* cp_name[] = {
+       "INT_HARDWARE_ENTRY",
+       "INT_HW_IRQ_EN",
+       "INT_TASKLET_ENTRY",
+       "FS_DEVRW",
+       "MEM_SWAPOUT",
+       "TIMERADD",
+       "SCSI_DISPATCH_CMD",
+       "IDE_CORE_CP",
+       "DIRECT",
+};
+
+static char* cp_type[] = {
+       "PANIC",
+       "BUG",
+       "WARNING",
+       "EXCEPTION",
+       "LOOP",
+       "OVERFLOW",
+       "CORRUPT_STACK",
+       "UNALIGNED_LOAD_STORE_WRITE",
+       "OVERWRITE_ALLOCATION",
+       "WRITE_AFTER_FREE",
+       "READ_AFTER_FREE",
+       "WRITE_BUDDY_AFTER_FREE",
+       "READ_BUDDY_AFTER_FREE",
+       "SOFTLOCKUP",
+       "HARDLOCKUP",
+       "SPINLOCKUP",
+       "HUNG_TASK",
+       "EXEC_DATA",
+       "EXEC_STACK",
+       "EXEC_KMALLOC",
+       "EXEC_VMALLOC",
+       "EXEC_USERSPACE",
+       "ACCESS_USERSPACE",
+       "WRITE_RO",
+       "WRITE_RO_AFTER_INIT",
+       "WRITE_KERN",
+       "WRAP_ATOMIC"
+};
+
+static struct jprobe lkdtm;
+
+static int lkdtm_parse_commandline(void);
+static void lkdtm_handler(void);
+
+static char* cpoint_name;
+static char* cpoint_type;
+static int cpoint_count = DEFAULT_COUNT;
+static int recur_count = REC_NUM_DEFAULT;
+
+static enum cname cpoint = CN_INVALID;
+static enum ctype cptype = CT_NONE;
+static int count = DEFAULT_COUNT;
+static DEFINE_SPINLOCK(count_lock);
+static DEFINE_SPINLOCK(lock_me_up);
+
+static u8 data_area[EXEC_SIZE];
+
+static const unsigned long rodata = 0xAA55AA55;
+static unsigned long ro_after_init __ro_after_init = 0x55AA5500;
+
+module_param(recur_count, int, 0644);
+MODULE_PARM_DESC(recur_count, " Recursion level for the stack overflow test");
+module_param(cpoint_name, charp, 0444);
+MODULE_PARM_DESC(cpoint_name, " Crash Point, where kernel is to be crashed");
+module_param(cpoint_type, charp, 0444);
+MODULE_PARM_DESC(cpoint_type, " Crash Point Type, action to be taken on "\
+                               "hitting the crash point");
+module_param(cpoint_count, int, 0644);
+MODULE_PARM_DESC(cpoint_count, " Crash Point Count, number of times the "\
+                               "crash point is to be hit to trigger action");
+
+static unsigned int jp_do_irq(unsigned int irq)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+
+static irqreturn_t jp_handle_irq_event(unsigned int irq,
+                                      struct irqaction *action)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+
+static void jp_tasklet_action(struct softirq_action *a)
+{
+       lkdtm_handler();
+       jprobe_return();
+}
+
+static void jp_ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
+{
+       lkdtm_handler();
+       jprobe_return();
+}
+
+struct scan_control;
+
+static unsigned long jp_shrink_inactive_list(unsigned long max_scan,
+                                            struct zone *zone,
+                                            struct scan_control *sc)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+
+static int jp_hrtimer_start(struct hrtimer *timer, ktime_t tim,
+                           const enum hrtimer_mode mode)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+
+static int jp_scsi_dispatch_cmd(struct scsi_cmnd *cmd)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+
+#ifdef CONFIG_IDE
+static int jp_generic_ide_ioctl(ide_drive_t *drive, struct file *file,
+                       struct block_device *bdev, unsigned int cmd,
+                       unsigned long arg)
+{
+       lkdtm_handler();
+       jprobe_return();
+       return 0;
+}
+#endif
+
+/* Return the crashpoint number or NONE if the name is invalid */
+static enum ctype parse_cp_type(const char *what, size_t count)
+{
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(cp_type); i++) {
+               if (!strcmp(what, cp_type[i]))
+                       return i + 1;
+       }
+
+       return CT_NONE;
+}
+
+static const char *cp_type_to_str(enum ctype type)
+{
+       if (type == CT_NONE || type < 0 || type > ARRAY_SIZE(cp_type))
+               return "None";
+
+       return cp_type[type - 1];
+}
+
+static const char *cp_name_to_str(enum cname name)
+{
+       if (name == CN_INVALID || name < 0 || name > ARRAY_SIZE(cp_name))
+               return "INVALID";
+
+       return cp_name[name - 1];
+}
+
+
+static int lkdtm_parse_commandline(void)
+{
+       int i;
+       unsigned long flags;
+
+       if (cpoint_count < 1 || recur_count < 1)
+               return -EINVAL;
+
+       spin_lock_irqsave(&count_lock, flags);
+       count = cpoint_count;
+       spin_unlock_irqrestore(&count_lock, flags);
+
+       /* No special parameters */
+       if (!cpoint_type && !cpoint_name)
+               return 0;
+
+       /* Neither or both of these need to be set */
+       if (!cpoint_type || !cpoint_name)
+               return -EINVAL;
+
+       cptype = parse_cp_type(cpoint_type, strlen(cpoint_type));
+       if (cptype == CT_NONE)
+               return -EINVAL;
+
+       for (i = 0; i < ARRAY_SIZE(cp_name); i++) {
+               if (!strcmp(cpoint_name, cp_name[i])) {
+                       cpoint = i + 1;
+                       return 0;
+               }
+       }
+
+       /* Could not find a valid crash point */
+       return -EINVAL;
+}
+
+static int recursive_loop(int remaining)
+{
+       char buf[REC_STACK_SIZE];
+
+       /* Make sure compiler does not optimize this away. */
+       memset(buf, (remaining & 0xff) | 0x1, REC_STACK_SIZE);
+       if (!remaining)
+               return 0;
+       else
+               return recursive_loop(remaining - 1);
+}
+
+static void do_nothing(void)
+{
+       return;
+}
+
+/* Must immediately follow do_nothing for size calculuations to work out. */
+static void do_overwritten(void)
+{
+       pr_info("do_overwritten wasn't overwritten!\n");
+       return;
+}
+
+static noinline void corrupt_stack(void)
+{
+       /* Use default char array length that triggers stack protection. */
+       char data[8];
+
+       memset((void *)data, 0, 64);
+}
+
+static void noinline execute_location(void *dst)
+{
+       void (*func)(void) = dst;
+
+       pr_info("attempting ok execution at %p\n", do_nothing);
+       do_nothing();
+
+       memcpy(dst, do_nothing, EXEC_SIZE);
+       flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
+       pr_info("attempting bad execution at %p\n", func);
+       func();
+}
+
+static void execute_user_location(void *dst)
+{
+       /* Intentionally crossing kernel/user memory boundary. */
+       void (*func)(void) = dst;
+
+       pr_info("attempting ok execution at %p\n", do_nothing);
+       do_nothing();
+
+       if (copy_to_user((void __user *)dst, do_nothing, EXEC_SIZE))
+               return;
+       flush_icache_range((unsigned long)dst, (unsigned long)dst + EXEC_SIZE);
+       pr_info("attempting bad execution at %p\n", func);
+       func();
+}
+
+static void lkdtm_do_action(enum ctype which)
+{
+       switch (which) {
+       case CT_PANIC:
+               panic("dumptest");
+               break;
+       case CT_BUG:
+               BUG();
+               break;
+       case CT_WARNING:
+               WARN_ON(1);
+               break;
+       case CT_EXCEPTION:
+               *((int *) 0) = 0;
+               break;
+       case CT_LOOP:
+               for (;;)
+                       ;
+               break;
+       case CT_OVERFLOW:
+               (void) recursive_loop(recur_count);
+               break;
+       case CT_CORRUPT_STACK:
+               corrupt_stack();
+               break;
+       case CT_UNALIGNED_LOAD_STORE_WRITE: {
+               static u8 data[5] __attribute__((aligned(4))) = {1, 2,
+                               3, 4, 5};
+               u32 *p;
+               u32 val = 0x12345678;
+
+               p = (u32 *)(data + 1);
+               if (*p == 0)
+                       val = 0x87654321;
+               *p = val;
+                break;
+       }
+       case CT_OVERWRITE_ALLOCATION: {
+               size_t len = 1020;
+               u32 *data = kmalloc(len, GFP_KERNEL);
+
+               data[1024 / sizeof(u32)] = 0x12345678;
+               kfree(data);
+               break;
+       }
+       case CT_WRITE_AFTER_FREE: {
+               int *base, *again;
+               size_t len = 1024;
+               /*
+                * The slub allocator uses the first word to store the free
+                * pointer in some configurations. Use the middle of the
+                * allocation to avoid running into the freelist
+                */
+               size_t offset = (len / sizeof(*base)) / 2;
+
+               base = kmalloc(len, GFP_KERNEL);
+               pr_info("Allocated memory %p-%p\n", base, &base[offset * 2]);
+               pr_info("Attempting bad write to freed memory at %p\n",
+                       &base[offset]);
+               kfree(base);
+               base[offset] = 0x0abcdef0;
+               /* Attempt to notice the overwrite. */
+               again = kmalloc(len, GFP_KERNEL);
+               kfree(again);
+               if (again != base)
+                       pr_info("Hmm, didn't get the same memory range.\n");
+
+               break;
+       }
+       case CT_READ_AFTER_FREE: {
+               int *base, *val, saw;
+               size_t len = 1024;
+               /*
+                * The slub allocator uses the first word to store the free
+                * pointer in some configurations. Use the middle of the
+                * allocation to avoid running into the freelist
+                */
+               size_t offset = (len / sizeof(*base)) / 2;
+
+               base = kmalloc(len, GFP_KERNEL);
+               if (!base)
+                       break;
+
+               val = kmalloc(len, GFP_KERNEL);
+               if (!val) {
+                       kfree(base);
+                       break;
+               }
+
+               *val = 0x12345678;
+               base[offset] = *val;
+               pr_info("Value in memory before free: %x\n", base[offset]);
+
+               kfree(base);
+
+               pr_info("Attempting bad read from freed memory\n");
+               saw = base[offset];
+               if (saw != *val) {
+                       /* Good! Poisoning happened, so declare a win. */
+                       pr_info("Memory correctly poisoned (%x)\n", saw);
+                       BUG();
+               }
+               pr_info("Memory was not poisoned\n");
+
+               kfree(val);
+               break;
+       }
+       case CT_WRITE_BUDDY_AFTER_FREE: {
+               unsigned long p = __get_free_page(GFP_KERNEL);
+               if (!p)
+                       break;
+               pr_info("Writing to the buddy page before free\n");
+               memset((void *)p, 0x3, PAGE_SIZE);
+               free_page(p);
+               schedule();
+               pr_info("Attempting bad write to the buddy page after free\n");
+               memset((void *)p, 0x78, PAGE_SIZE);
+               /* Attempt to notice the overwrite. */
+               p = __get_free_page(GFP_KERNEL);
+               free_page(p);
+               schedule();
+
+               break;
+       }
+       case CT_READ_BUDDY_AFTER_FREE: {
+               unsigned long p = __get_free_page(GFP_KERNEL);
+               int saw, *val;
+               int *base;
+
+               if (!p)
+                       break;
+
+               val = kmalloc(1024, GFP_KERNEL);
+               if (!val) {
+                       free_page(p);
+                       break;
+               }
+
+               base = (int *)p;
+
+               *val = 0x12345678;
+               base[0] = *val;
+               pr_info("Value in memory before free: %x\n", base[0]);
+               free_page(p);
+               pr_info("Attempting to read from freed memory\n");
+               saw = base[0];
+               if (saw != *val) {
+                       /* Good! Poisoning happened, so declare a win. */
+                       pr_info("Memory correctly poisoned (%x)\n", saw);
+                       BUG();
+               }
+               pr_info("Buddy page was not poisoned\n");
+
+               kfree(val);
+               break;
+       }
+       case CT_SOFTLOCKUP:
+               preempt_disable();
+               for (;;)
+                       cpu_relax();
+               break;
+       case CT_HARDLOCKUP:
+               local_irq_disable();
+               for (;;)
+                       cpu_relax();
+               break;
+       case CT_SPINLOCKUP:
+               /* Must be called twice to trigger. */
+               spin_lock(&lock_me_up);
+               /* Let sparse know we intended to exit holding the lock. */
+               __release(&lock_me_up);
+               break;
+       case CT_HUNG_TASK:
+               set_current_state(TASK_UNINTERRUPTIBLE);
+               schedule();
+               break;
+       case CT_EXEC_DATA:
+               execute_location(data_area);
+               break;
+       case CT_EXEC_STACK: {
+               u8 stack_area[EXEC_SIZE];
+               execute_location(stack_area);
+               break;
+       }
+       case CT_EXEC_KMALLOC: {
+               u32 *kmalloc_area = kmalloc(EXEC_SIZE, GFP_KERNEL);
+               execute_location(kmalloc_area);
+               kfree(kmalloc_area);
+               break;
+       }
+       case CT_EXEC_VMALLOC: {
+               u32 *vmalloc_area = vmalloc(EXEC_SIZE);
+               execute_location(vmalloc_area);
+               vfree(vmalloc_area);
+               break;
+       }
+       case CT_EXEC_USERSPACE: {
+               unsigned long user_addr;
+
+               user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
+                                   PROT_READ | PROT_WRITE | PROT_EXEC,
+                                   MAP_ANONYMOUS | MAP_PRIVATE, 0);
+               if (user_addr >= TASK_SIZE) {
+                       pr_warn("Failed to allocate user memory\n");
+                       return;
+               }
+               execute_user_location((void *)user_addr);
+               vm_munmap(user_addr, PAGE_SIZE);
+               break;
+       }
+       case CT_ACCESS_USERSPACE: {
+               unsigned long user_addr, tmp = 0;
+               unsigned long *ptr;
+
+               user_addr = vm_mmap(NULL, 0, PAGE_SIZE,
+                                   PROT_READ | PROT_WRITE | PROT_EXEC,
+                                   MAP_ANONYMOUS | MAP_PRIVATE, 0);
+               if (user_addr >= TASK_SIZE) {
+                       pr_warn("Failed to allocate user memory\n");
+                       return;
+               }
+
+               if (copy_to_user((void __user *)user_addr, &tmp, sizeof(tmp))) {
+                       pr_warn("copy_to_user failed\n");
+                       vm_munmap(user_addr, PAGE_SIZE);
+                       return;
+               }
+
+               ptr = (unsigned long *)user_addr;
+
+               pr_info("attempting bad read at %p\n", ptr);
+               tmp = *ptr;
+               tmp += 0xc0dec0de;
+
+               pr_info("attempting bad write at %p\n", ptr);
+               *ptr = tmp;
+
+               vm_munmap(user_addr, PAGE_SIZE);
+
+               break;
+       }
+       case CT_WRITE_RO: {
+               /* Explicitly cast away "const" for the test. */
+               unsigned long *ptr = (unsigned long *)&rodata;
+
+               pr_info("attempting bad rodata write at %p\n", ptr);
+               *ptr ^= 0xabcd1234;
+
+               break;
+       }
+       case CT_WRITE_RO_AFTER_INIT: {
+               unsigned long *ptr = &ro_after_init;
+
+               /*
+                * Verify we were written to during init. Since an Oops
+                * is considered a "success", a failure is to just skip the
+                * real test.
+                */
+               if ((*ptr & 0xAA) != 0xAA) {
+                       pr_info("%p was NOT written during init!?\n", ptr);
+                       break;
+               }
+
+               pr_info("attempting bad ro_after_init write at %p\n", ptr);
+               *ptr ^= 0xabcd1234;
+
+               break;
+       }
+       case CT_WRITE_KERN: {
+               size_t size;
+               unsigned char *ptr;
+
+               size = (unsigned long)do_overwritten -
+                      (unsigned long)do_nothing;
+               ptr = (unsigned char *)do_overwritten;
+
+               pr_info("attempting bad %zu byte write at %p\n", size, ptr);
+               memcpy(ptr, (unsigned char *)do_nothing, size);
+               flush_icache_range((unsigned long)ptr,
+                                  (unsigned long)(ptr + size));
+
+               do_overwritten();
+               break;
+       }
+       case CT_WRAP_ATOMIC: {
+               atomic_t under = ATOMIC_INIT(INT_MIN);
+               atomic_t over = ATOMIC_INIT(INT_MAX);
+
+               pr_info("attempting atomic underflow\n");
+               atomic_dec(&under);
+               pr_info("attempting atomic overflow\n");
+               atomic_inc(&over);
+
+               return;
+       }
+       case CT_NONE:
+       default:
+               break;
+       }
+
+}
+
+static void lkdtm_handler(void)
+{
+       unsigned long flags;
+       bool do_it = false;
+
+       spin_lock_irqsave(&count_lock, flags);
+       count--;
+       pr_info("Crash point %s of type %s hit, trigger in %d rounds\n",
+               cp_name_to_str(cpoint), cp_type_to_str(cptype), count);
+
+       if (count == 0) {
+               do_it = true;
+               count = cpoint_count;
+       }
+       spin_unlock_irqrestore(&count_lock, flags);
+
+       if (do_it)
+               lkdtm_do_action(cptype);
+}
+
+static int lkdtm_register_cpoint(enum cname which)
+{
+       int ret;
+
+       cpoint = CN_INVALID;
+       if (lkdtm.entry != NULL)
+               unregister_jprobe(&lkdtm);
+
+       switch (which) {
+       case CN_DIRECT:
+               lkdtm_do_action(cptype);
+               return 0;
+       case CN_INT_HARDWARE_ENTRY:
+               lkdtm.kp.symbol_name = "do_IRQ";
+               lkdtm.entry = (kprobe_opcode_t*) jp_do_irq;
+               break;
+       case CN_INT_HW_IRQ_EN:
+               lkdtm.kp.symbol_name = "handle_IRQ_event";
+               lkdtm.entry = (kprobe_opcode_t*) jp_handle_irq_event;
+               break;
+       case CN_INT_TASKLET_ENTRY:
+               lkdtm.kp.symbol_name = "tasklet_action";
+               lkdtm.entry = (kprobe_opcode_t*) jp_tasklet_action;
+               break;
+       case CN_FS_DEVRW:
+               lkdtm.kp.symbol_name = "ll_rw_block";
+               lkdtm.entry = (kprobe_opcode_t*) jp_ll_rw_block;
+               break;
+       case CN_MEM_SWAPOUT:
+               lkdtm.kp.symbol_name = "shrink_inactive_list";
+               lkdtm.entry = (kprobe_opcode_t*) jp_shrink_inactive_list;
+               break;
+       case CN_TIMERADD:
+               lkdtm.kp.symbol_name = "hrtimer_start";
+               lkdtm.entry = (kprobe_opcode_t*) jp_hrtimer_start;
+               break;
+       case CN_SCSI_DISPATCH_CMD:
+               lkdtm.kp.symbol_name = "scsi_dispatch_cmd";
+               lkdtm.entry = (kprobe_opcode_t*) jp_scsi_dispatch_cmd;
+               break;
+       case CN_IDE_CORE_CP:
+#ifdef CONFIG_IDE
+               lkdtm.kp.symbol_name = "generic_ide_ioctl";
+               lkdtm.entry = (kprobe_opcode_t*) jp_generic_ide_ioctl;
+#else
+               pr_info("Crash point not available\n");
+               return -EINVAL;
+#endif
+               break;
+       default:
+               pr_info("Invalid Crash Point\n");
+               return -EINVAL;
+       }
+
+       cpoint = which;
+       if ((ret = register_jprobe(&lkdtm)) < 0) {
+               pr_info("Couldn't register jprobe\n");
+               cpoint = CN_INVALID;
+       }
+
+       return ret;
+}
+
+static ssize_t do_register_entry(enum cname which, struct file *f,
+               const char __user *user_buf, size_t count, loff_t *off)
+{
+       char *buf;
+       int err;
+
+       if (count >= PAGE_SIZE)
+               return -EINVAL;
+
+       buf = (char *)__get_free_page(GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+       if (copy_from_user(buf, user_buf, count)) {
+               free_page((unsigned long) buf);
+               return -EFAULT;
+       }
+       /* NULL-terminate and remove enter */
+       buf[count] = '\0';
+       strim(buf);
+
+       cptype = parse_cp_type(buf, count);
+       free_page((unsigned long) buf);
+
+       if (cptype == CT_NONE)
+               return -EINVAL;
+
+       err = lkdtm_register_cpoint(which);
+       if (err < 0)
+               return err;
+
+       *off += count;
+
+       return count;
+}
+
+/* Generic read callback that just prints out the available crash types */
+static ssize_t lkdtm_debugfs_read(struct file *f, char __user *user_buf,
+               size_t count, loff_t *off)
+{
+       char *buf;
+       int i, n, out;
+
+       buf = (char *)__get_free_page(GFP_KERNEL);
+       if (buf == NULL)
+               return -ENOMEM;
+
+       n = snprintf(buf, PAGE_SIZE, "Available crash types:\n");
+       for (i = 0; i < ARRAY_SIZE(cp_type); i++)
+               n += snprintf(buf + n, PAGE_SIZE - n, "%s\n", cp_type[i]);
+       buf[n] = '\0';
+
+       out = simple_read_from_buffer(user_buf, count, off,
+                                     buf, n);
+       free_page((unsigned long) buf);
+
+       return out;
+}
+
+static int lkdtm_debugfs_open(struct inode *inode, struct file *file)
+{
+       return 0;
+}
+
+
+static ssize_t int_hardware_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_INT_HARDWARE_ENTRY, f, buf, count, off);
+}
+
+static ssize_t int_hw_irq_en(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_INT_HW_IRQ_EN, f, buf, count, off);
+}
+
+static ssize_t int_tasklet_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_INT_TASKLET_ENTRY, f, buf, count, off);
+}
+
+static ssize_t fs_devrw_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_FS_DEVRW, f, buf, count, off);
+}
+
+static ssize_t mem_swapout_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_MEM_SWAPOUT, f, buf, count, off);
+}
+
+static ssize_t timeradd_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_TIMERADD, f, buf, count, off);
+}
+
+static ssize_t scsi_dispatch_cmd_entry(struct file *f,
+               const char __user *buf, size_t count, loff_t *off)
+{
+       return do_register_entry(CN_SCSI_DISPATCH_CMD, f, buf, count, off);
+}
+
+static ssize_t ide_core_cp_entry(struct file *f, const char __user *buf,
+               size_t count, loff_t *off)
+{
+       return do_register_entry(CN_IDE_CORE_CP, f, buf, count, off);
+}
+
+/* Special entry to just crash directly. Available without KPROBEs */
+static ssize_t direct_entry(struct file *f, const char __user *user_buf,
+               size_t count, loff_t *off)
+{
+       enum ctype type;
+       char *buf;
+
+       if (count >= PAGE_SIZE)
+               return -EINVAL;
+       if (count < 1)
+               return -EINVAL;
+
+       buf = (char *)__get_free_page(GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+       if (copy_from_user(buf, user_buf, count)) {
+               free_page((unsigned long) buf);
+               return -EFAULT;
+       }
+       /* NULL-terminate and remove enter */
+       buf[count] = '\0';
+       strim(buf);
+
+       type = parse_cp_type(buf, count);
+       free_page((unsigned long) buf);
+       if (type == CT_NONE)
+               return -EINVAL;
+
+       pr_info("Performing direct entry %s\n", cp_type_to_str(type));
+       lkdtm_do_action(type);
+       *off += count;
+
+       return count;
+}
+
+struct crash_entry {
+       const char *name;
+       const struct file_operations fops;
+};
+
+static const struct crash_entry crash_entries[] = {
+       {"DIRECT", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = direct_entry} },
+       {"INT_HARDWARE_ENTRY", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = int_hardware_entry} },
+       {"INT_HW_IRQ_EN", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = int_hw_irq_en} },
+       {"INT_TASKLET_ENTRY", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = int_tasklet_entry} },
+       {"FS_DEVRW", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = fs_devrw_entry} },
+       {"MEM_SWAPOUT", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = mem_swapout_entry} },
+       {"TIMERADD", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = timeradd_entry} },
+       {"SCSI_DISPATCH_CMD", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = scsi_dispatch_cmd_entry} },
+       {"IDE_CORE_CP", {.read = lkdtm_debugfs_read,
+                       .llseek = generic_file_llseek,
+                       .open = lkdtm_debugfs_open,
+                       .write = ide_core_cp_entry} },
+};
+
+static struct dentry *lkdtm_debugfs_root;
+
+static int __init lkdtm_module_init(void)
+{
+       int ret = -EINVAL;
+       int n_debugfs_entries = 1; /* Assume only the direct entry */
+       int i;
+
+       /* Make sure we can write to __ro_after_init values during __init */
+       ro_after_init |= 0xAA;
+
+       /* Register debugfs interface */
+       lkdtm_debugfs_root = debugfs_create_dir("provoke-crash", NULL);
+       if (!lkdtm_debugfs_root) {
+               pr_err("creating root dir failed\n");
+               return -ENODEV;
+       }
+
+#ifdef CONFIG_KPROBES
+       n_debugfs_entries = ARRAY_SIZE(crash_entries);
+#endif
+
+       for (i = 0; i < n_debugfs_entries; i++) {
+               const struct crash_entry *cur = &crash_entries[i];
+               struct dentry *de;
+
+               de = debugfs_create_file(cur->name, 0644, lkdtm_debugfs_root,
+                               NULL, &cur->fops);
+               if (de == NULL) {
+                       pr_err("could not create %s\n", cur->name);
+                       goto out_err;
+               }
+       }
+
+       if (lkdtm_parse_commandline() == -EINVAL) {
+               pr_info("Invalid command\n");
+               goto out_err;
+       }
+
+       if (cpoint != CN_INVALID && cptype != CT_NONE) {
+               ret = lkdtm_register_cpoint(cpoint);
+               if (ret < 0) {
+                       pr_info("Invalid crash point %d\n", cpoint);
+                       goto out_err;
+               }
+               pr_info("Crash point %s of type %s registered\n",
+                       cpoint_name, cpoint_type);
+       } else {
+               pr_info("No crash points registered, enable through debugfs\n");
+       }
+
+       return 0;
+
+out_err:
+       debugfs_remove_recursive(lkdtm_debugfs_root);
+       return ret;
+}
+
+static void __exit lkdtm_module_exit(void)
+{
+       debugfs_remove_recursive(lkdtm_debugfs_root);
+
+       unregister_jprobe(&lkdtm);
+       pr_info("Crash point unregistered\n");
+}
+
+module_init(lkdtm_module_init);
+module_exit(lkdtm_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Kprobe module for testing crash dumps");