cgroup: Add generic cgroup subsystem permission checks.
[linux-2.6.git] / kernel / futex_compat.c
index 1ab6a0e..5f9e689 100644 (file)
@@ -8,10 +8,37 @@
 
 #include <linux/linkage.h>
 #include <linux/compat.h>
+#include <linux/nsproxy.h>
 #include <linux/futex.h>
 
 #include <asm/uaccess.h>
 
+
+/*
+ * Fetch a robust-list pointer. Bit 0 signals PI futexes:
+ */
+static inline int
+fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry,
+                  compat_uptr_t __user *head, unsigned int *pi)
+{
+       if (get_user(*uentry, head))
+               return -EFAULT;
+
+       *entry = compat_ptr((*uentry) & ~1);
+       *pi = (unsigned int)(*uentry) & 1;
+
+       return 0;
+}
+
+static void __user *futex_uaddr(struct robust_list __user *entry,
+                               compat_long_t futex_offset)
+{
+       compat_uptr_t base = ptr_to_compat(entry);
+       void __user *uaddr = compat_ptr(base + futex_offset);
+
+       return uaddr;
+}
+
 /*
  * Walk curr->robust_list (very carefully, it's a userspace list!)
  * and mark any locks found there dead, and notify any waiters.
 void compat_exit_robust_list(struct task_struct *curr)
 {
        struct compat_robust_list_head __user *head = curr->compat_robust_list;
-       struct robust_list __user *entry, *pending;
-       compat_uptr_t uentry, upending;
-       unsigned int limit = ROBUST_LIST_LIMIT;
+       struct robust_list __user *entry, *next_entry, *pending;
+       unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
+       unsigned int uninitialized_var(next_pi);
+       compat_uptr_t uentry, next_uentry, upending;
        compat_long_t futex_offset;
+       int rc;
+
+       if (!futex_cmpxchg_enabled)
+               return;
 
        /*
         * Fetch the list head (which was registered earlier, via
         * sys_set_robust_list()):
         */
-       if (get_user(uentry, &head->list.next))
+       if (fetch_robust_entry(&uentry, &entry, &head->list.next, &pi))
                return;
-       entry = compat_ptr(uentry);
        /*
         * Fetch the relative futex offset:
         */
@@ -42,28 +73,33 @@ void compat_exit_robust_list(struct task_struct *curr)
         * Fetch any possibly pending lock-add first, and handle it
         * if it exists:
         */
-       if (get_user(upending, &head->list_op_pending))
+       if (fetch_robust_entry(&upending, &pending,
+                              &head->list_op_pending, &pip))
                return;
-       pending = compat_ptr(upending);
-       if (upending)
-               handle_futex_death((void *)pending + futex_offset, curr);
 
-       while (compat_ptr(uentry) != &head->list) {
+       next_entry = NULL;      /* avoid warning with gcc */
+       while (entry != (struct robust_list __user *) &head->list) {
+               /*
+                * Fetch the next entry in the list before calling
+                * handle_futex_death:
+                */
+               rc = fetch_robust_entry(&next_uentry, &next_entry,
+                       (compat_uptr_t __user *)&entry->next, &next_pi);
                /*
                 * A pending lock might already be on the list, so
                 * dont process it twice:
                 */
-               if (entry != pending)
-                       if (handle_futex_death((void *)entry + futex_offset,
-                                               curr))
-                               return;
+               if (entry != pending) {
+                       void __user *uaddr = futex_uaddr(entry, futex_offset);
 
-               /*
-                * Fetch the next entry in the list:
-                */
-               if (get_user(uentry, (compat_uptr_t *)&entry->next))
+                       if (handle_futex_death(uaddr, curr, pi))
+                               return;
+               }
+               if (rc)
                        return;
-               entry = compat_ptr(uentry);
+               uentry = next_uentry;
+               entry = next_entry;
+               pi = next_pi;
                /*
                 * Avoid excessively long or circular lists:
                 */
@@ -72,12 +108,20 @@ void compat_exit_robust_list(struct task_struct *curr)
 
                cond_resched();
        }
+       if (pending) {
+               void __user *uaddr = futex_uaddr(pending, futex_offset);
+
+               handle_futex_death(uaddr, curr, pip);
+       }
 }
 
 asmlinkage long
 compat_sys_set_robust_list(struct compat_robust_list_head __user *head,
                           compat_size_t len)
 {
+       if (!futex_cmpxchg_enabled)
+               return -ENOSYS;
+
        if (unlikely(len != sizeof(*head)))
                return -EINVAL;
 
@@ -87,11 +131,15 @@ compat_sys_set_robust_list(struct compat_robust_list_head __user *head,
 }
 
 asmlinkage long
-compat_sys_get_robust_list(int pid, compat_uptr_t *head_ptr,
+compat_sys_get_robust_list(int pid, compat_uptr_t __user *head_ptr,
                           compat_size_t __user *len_ptr)
 {
-       struct compat_robust_list_head *head;
+       struct compat_robust_list_head __user *head;
        unsigned long ret;
+       const struct cred *cred = current_cred(), *pcred;
+
+       if (!futex_cmpxchg_enabled)
+               return -ENOSYS;
 
        if (!pid)
                head = current->compat_robust_list;
@@ -99,16 +147,27 @@ compat_sys_get_robust_list(int pid, compat_uptr_t *head_ptr,
                struct task_struct *p;
 
                ret = -ESRCH;
-               read_lock(&tasklist_lock);
-               p = find_task_by_pid(pid);
+               rcu_read_lock();
+               p = find_task_by_vpid(pid);
                if (!p)
                        goto err_unlock;
                ret = -EPERM;
-               if ((current->euid != p->euid) && (current->euid != p->uid) &&
-                               !capable(CAP_SYS_PTRACE))
+               pcred = __task_cred(p);
+               /* If victim is in different user_ns, then uids are not
+                  comparable, so we must have CAP_SYS_PTRACE */
+               if (cred->user->user_ns != pcred->user->user_ns) {
+                       if (!ns_capable(pcred->user->user_ns, CAP_SYS_PTRACE))
+                               goto err_unlock;
+                       goto ok;
+               }
+               /* If victim is in same user_ns, then uids are comparable */
+               if (cred->euid != pcred->euid &&
+                   cred->euid != pcred->uid &&
+                   !ns_capable(pcred->user->user_ns, CAP_SYS_PTRACE))
                        goto err_unlock;
+ok:
                head = p->compat_robust_list;
-               read_unlock(&tasklist_lock);
+               rcu_read_unlock();
        }
 
        if (put_user(sizeof(*head), len_ptr))
@@ -116,7 +175,7 @@ compat_sys_get_robust_list(int pid, compat_uptr_t *head_ptr,
        return put_user(ptr_to_compat(head), head_ptr);
 
 err_unlock:
-       read_unlock(&tasklist_lock);
+       rcu_read_unlock();
 
        return ret;
 }
@@ -125,20 +184,27 @@ asmlinkage long compat_sys_futex(u32 __user *uaddr, int op, u32 val,
                struct compat_timespec __user *utime, u32 __user *uaddr2,
                u32 val3)
 {
-       struct timespec t;
-       unsigned long timeout = MAX_SCHEDULE_TIMEOUT;
+       struct timespec ts;
+       ktime_t t, *tp = NULL;
        int val2 = 0;
+       int cmd = op & FUTEX_CMD_MASK;
 
-       if (utime && (op == FUTEX_WAIT)) {
-               if (get_compat_timespec(&t, utime))
+       if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
+                     cmd == FUTEX_WAIT_BITSET ||
+                     cmd == FUTEX_WAIT_REQUEUE_PI)) {
+               if (get_compat_timespec(&ts, utime))
                        return -EFAULT;
-               if (!timespec_valid(&t))
+               if (!timespec_valid(&ts))
                        return -EINVAL;
-               timeout = timespec_to_jiffies(&t) + 1;
+
+               t = timespec_to_ktime(ts);
+               if (cmd == FUTEX_WAIT)
+                       t = ktime_add_safe(ktime_get(), t);
+               tp = &t;
        }
-       if (op >= FUTEX_REQUEUE)
+       if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
+           cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
                val2 = (int) (unsigned long) utime;
 
-       return do_futex((unsigned long)uaddr, op, val, timeout,
-                       (unsigned long)uaddr2, val2, val3);
+       return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
 }