| 1 | #include <linux/kernel.h>
|
| 2 | #include <linux/syscalls.h>
|
| 3 | #include <linux/fdtable.h>
|
| 4 | #include <linux/string.h>
|
| 5 | #include <linux/random.h>
|
| 6 | #include <linux/module.h>
|
| 7 | #include <linux/ptrace.h>
|
| 8 | #include <linux/init.h>
|
| 9 | #include <linux/errno.h>
|
| 10 | #include <linux/cache.h>
|
| 11 | #include <linux/bug.h>
|
| 12 | #include <linux/err.h>
|
| 13 | #include <linux/kcmp.h>
|
| 14 | #include <linux/capability.h>
|
| 15 | #include <linux/list.h>
|
| 16 | #include <linux/eventpoll.h>
|
| 17 | #include <linux/file.h>
|
| 18 |
|
| 19 | #include <asm/unistd.h>
|
| 20 |
|
| 21 | |
| 22 | * We don't expose the real in-memory order of objects for security reasons.
|
| 23 | * But still the comparison results should be suitable for sorting. So we
|
| 24 | * obfuscate kernel pointers values and compare the production instead.
|
| 25 | *
|
| 26 | * The obfuscation is done in two steps. First we xor the kernel pointer with
|
| 27 | * a random value, which puts pointer into a new position in a reordered space.
|
| 28 | * Secondly we multiply the xor production with a large odd random number to
|
| 29 | * permute its bits even more (the odd multiplier guarantees that the product
|
| 30 | * is unique ever after the high bits are truncated, since any odd number is
|
| 31 | * relative prime to 2^n).
|
| 32 | *
|
| 33 | * Note also that the obfuscation itself is invisible to userspace and if needed
|
| 34 | * it can be changed to an alternate scheme.
|
| 35 |
|
| 36 | static unsigned long cookies[KCMP_TYPES][2] __read_mostly;
|
| 37 |
|
| 38 | static long kptr_obfuscate(long v, int type)
|
| 39 | {
|
| 40 | return (v ^ cookies[type][0]) * cookies[type][1];
|
| 41 | }
|
| 42 |
|
| 43 | |
| 44 | * 0 - equal, i.e. v1 = v2
|
| 45 | * 1 - less than, i.e. v1 < v2
|
| 46 | * 2 - greater than, i.e. v1 > v2
|
| 47 | * 3 - not equal but ordering unavailable (reserved for future)
|
| 48 |
|
| 49 | static int kcmp_ptr(void *v1, void *v2, enum kcmp_type type)
|
| 50 | {
|
| 51 | long t1, t2;
|
| 52 |
|
| 53 | t1 = kptr_obfuscate((long)v1, type);
|
| 54 | t2 = kptr_obfuscate((long)v2, type);
|
| 55 |
|
| 56 | return (t1 < t2) | ((t1 > t2) << 1);
|
| 57 | }
|
| 58 |
|
| 59 |
|
| 60 | static struct file *
|
| 61 | get_file_raw_ptr(struct task_struct *task, unsigned int idx)
|
| 62 | {
|
| 63 | struct file *file = NULL;
|
| 64 |
|
| 65 | task_lock(task);
|
| 66 | rcu_read_lock();
|
| 67 |
|
| 68 | if (task->files)
|
| 69 | file = fcheck_files(task->files, idx);
|
| 70 |
|
| 71 | rcu_read_unlock();
|
| 72 | task_unlock(task);
|
| 73 |
|
| 74 | return file;
|
| 75 | }
|
| 76 |
|
| 77 | static void kcmp_unlock(struct mutex *m1, struct mutex *m2)
|
| 78 | {
|
| 79 | if (likely(m2 != m1))
|
| 80 | mutex_unlock(m2);
|
| 81 | mutex_unlock(m1);
|
| 82 | }
|
| 83 |
|
| 84 | static int kcmp_lock(struct mutex *m1, struct mutex *m2)
|
| 85 | {
|
| 86 | int err;
|
| 87 |
|
| 88 | if (m2 > m1)
|
| 89 | swap(m1, m2);
|
| 90 |
|
| 91 | err = mutex_lock_killable(m1);
|
| 92 | if (!err && likely(m1 != m2)) {
|
| 93 | err = mutex_lock_killable_nested(m2, SINGLE_DEPTH_NESTING);
|
| 94 | if (err)
|
| 95 | mutex_unlock(m1);
|
| 96 | }
|
| 97 |
|
| 98 | return err;
|
| 99 | }
|
| 100 |
|
| 101 | #ifdef CONFIG_EPOLL
|
| 102 | static int kcmp_epoll_target(struct task_struct *task1,
|
| 103 | struct task_struct *task2,
|
| 104 | unsigned long idx1,
|
| 105 | struct kcmp_epoll_slot __user *uslot)
|
| 106 | {
|
| 107 | struct file *filp, *filp_epoll, *filp_tgt;
|
| 108 | struct kcmp_epoll_slot slot;
|
| 109 | struct files_struct *files;
|
| 110 |
|
| 111 | if (copy_from_user(&slot, uslot, sizeof(slot)))
|
| 112 | return -EFAULT;
|
| 113 |
|
| 114 | filp = get_file_raw_ptr(task1, idx1);
|
| 115 | if (!filp)
|
| 116 | return -EBADF;
|
| 117 |
|
| 118 | files = get_files_struct(task2);
|
| 119 | if (!files)
|
| 120 | return -EBADF;
|
| 121 |
|
| 122 | spin_lock(&files->file_lock);
|
| 123 | filp_epoll = fcheck_files(files, slot.efd);
|
| 124 | if (filp_epoll)
|
| 125 | get_file(filp_epoll);
|
| 126 | else
|
| 127 | filp_tgt = ERR_PTR(-EBADF);
|
| 128 | spin_unlock(&files->file_lock);
|
| 129 | put_files_struct(files);
|
| 130 |
|
| 131 | if (filp_epoll) {
|
| 132 | filp_tgt = get_epoll_tfile_raw_ptr(filp_epoll, slot.tfd, slot.toff);
|
| 133 | fput(filp_epoll);
|
| 134 | } else
|
| 135 |
|
| 136 | if (IS_ERR(filp_tgt))
|
| 137 | return PTR_ERR(filp_tgt);
|
| 138 |
|
| 139 | return kcmp_ptr(filp, filp_tgt, KCMP_FILE);
|
| 140 | }
|
| 141 | #else
|
| 142 | static int kcmp_epoll_target(struct task_struct *task1,
|
| 143 | struct task_struct *task2,
|
| 144 | unsigned long idx1,
|
| 145 | struct kcmp_epoll_slot __user *uslot)
|
| 146 | {
|
| 147 | return -EOPNOTSUPP;
|
| 148 | }
|
| 149 | #endif
|
| 150 |
|
| 151 | SYSCALL_DEFINE5(kcmp, pid_t, pid1, pid_t, pid2, int, type,
|
| 152 | unsigned long, idx1, unsigned long, idx2)
|
| 153 | {
|
| 154 | struct task_struct *task1, *task2;
|
| 155 | int ret;
|
| 156 |
|
| 157 | rcu_read_lock();
|
| 158 |
|
| 159 | |
| 160 | * Tasks are looked up in caller's PID namespace only.
|
| 161 |
|
| 162 | task1 = find_task_by_vpid(pid1);
|
| 163 | task2 = find_task_by_vpid(pid2);
|
| 164 | if (!task1 || !task2)
|
| 165 | goto err_no_task;
|
| 166 |
|
| 167 | get_task_struct(task1);
|
| 168 | get_task_struct(task2);
|
| 169 |
|
| 170 | rcu_read_unlock();
|
| 171 |
|
| 172 | |
| 173 | * One should have enough rights to inspect task details.
|
| 174 |
|
| 175 | ret = kcmp_lock(&task1->signal->cred_guard_mutex,
|
| 176 | &task2->signal->cred_guard_mutex);
|
| 177 | if (ret)
|
| 178 | goto err;
|
| 179 | if (!ptrace_may_access(task1, PTRACE_MODE_READ_REALCREDS) ||
|
| 180 | !ptrace_may_access(task2, PTRACE_MODE_READ_REALCREDS)) {
|
| 181 | ret = -EPERM;
|
| 182 | goto err_unlock;
|
| 183 | }
|
| 184 |
|
| 185 | switch (type) {
|
| 186 | case KCMP_FILE: {
|
| 187 | struct file *filp1, *filp2;
|
| 188 |
|
| 189 | filp1 = get_file_raw_ptr(task1, idx1);
|
| 190 | filp2 = get_file_raw_ptr(task2, idx2);
|
| 191 |
|
| 192 | if (filp1 && filp2)
|
| 193 | ret = kcmp_ptr(filp1, filp2, KCMP_FILE);
|
| 194 | else
|
| 195 | ret = -EBADF;
|
| 196 | break;
|
| 197 | }
|
| 198 | case KCMP_VM:
|
| 199 | ret = kcmp_ptr(task1->mm, task2->mm, KCMP_VM);
|
| 200 | break;
|
| 201 | case KCMP_FILES:
|
| 202 | ret = kcmp_ptr(task1->files, task2->files, KCMP_FILES);
|
| 203 | break;
|
| 204 | case KCMP_FS:
|
| 205 | ret = kcmp_ptr(task1->fs, task2->fs, KCMP_FS);
|
| 206 | break;
|
| 207 | case KCMP_SIGHAND:
|
| 208 | ret = kcmp_ptr(task1->sighand, task2->sighand, KCMP_SIGHAND);
|
| 209 | break;
|
| 210 | case KCMP_IO:
|
| 211 | ret = kcmp_ptr(task1->io_context, task2->io_context, KCMP_IO);
|
| 212 | break;
|
| 213 | case KCMP_SYSVSEM:
|
| 214 | #ifdef CONFIG_SYSVIPC
|
| 215 | ret = kcmp_ptr(task1->sysvsem.undo_list,
|
| 216 | task2->sysvsem.undo_list,
|
| 217 | KCMP_SYSVSEM);
|
| 218 | #else
|
| 219 | ret = -EOPNOTSUPP;
|
| 220 | #endif
|
| 221 | break;
|
| 222 | case KCMP_EPOLL_TFD:
|
| 223 | ret = kcmp_epoll_target(task1, task2, idx1, (void *)idx2);
|
| 224 | break;
|
| 225 | default:
|
| 226 | ret = -EINVAL;
|
| 227 | break;
|
| 228 | }
|
| 229 |
|
| 230 | err_unlock:
|
| 231 | kcmp_unlock(&task1->signal->cred_guard_mutex,
|
| 232 | &task2->signal->cred_guard_mutex);
|
| 233 | err:
|
| 234 | put_task_struct(task1);
|
| 235 | put_task_struct(task2);
|
| 236 |
|
| 237 | return ret;
|
| 238 |
|
| 239 | err_no_task:
|
| 240 | rcu_read_unlock();
|
| 241 | return -ESRCH;
|
| 242 | }
|
| 243 |
|
| 244 | static __init int kcmp_cookies_init(void)
|
| 245 | {
|
| 246 | int i;
|
| 247 |
|
| 248 | get_random_bytes(cookies, sizeof(cookies));
|
| 249 |
|
| 250 | for (i = 0; i < KCMP_TYPES; i++)
|
| 251 | cookies[i][1] |= (~(~0UL >> 1) | 1);
|
| 252 |
|
| 253 | return 0;
|
| 254 | }
|
| 255 | arch_initcall(kcmp_cookies_init);
|