sh: convert /proc/cpu/aligmnent, /proc/cpu/kernel_alignment to seq_file
[safe/jmp/linux-2.6] / kernel / ptrace.c
index 6c19e94..23bd09c 100644 (file)
@@ -21,9 +21,8 @@
 #include <linux/audit.h>
 #include <linux/pid_namespace.h>
 #include <linux/syscalls.h>
+#include <linux/uaccess.h>
 
-#include <asm/pgtable.h>
-#include <asm/uaccess.h>
 
 /*
  * ptrace a task: make the debugger its new parent and
  */
 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
 {
-       BUG_ON(!list_empty(&child->ptrace_list));
-       if (child->parent == new_parent)
-               return;
-       list_add(&child->ptrace_list, &child->parent->ptrace_children);
-       remove_parent(child);
+       BUG_ON(!list_empty(&child->ptrace_entry));
+       list_add(&child->ptrace_entry, &new_parent->ptraced);
        child->parent = new_parent;
-       add_parent(child);
 }
+
 /*
  * Turn a tracing stop into a normal stop now, since with no tracer there
  * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
@@ -49,15 +44,19 @@ void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  * TASK_TRACED, resume it now.
  * Requires that irqs be disabled.
  */
-void ptrace_untrace(struct task_struct *child)
+static void ptrace_untrace(struct task_struct *child)
 {
        spin_lock(&child->sighand->siglock);
        if (task_is_traced(child)) {
-               if (child->signal->flags & SIGNAL_STOP_STOPPED) {
+               /*
+                * If the group stop is completed or in progress,
+                * this thread was already counted as stopped.
+                */
+               if (child->signal->flags & SIGNAL_STOP_STOPPED ||
+                   child->signal->group_stop_count)
                        __set_task_state(child, TASK_STOPPED);
-               } else {
+               else
                        signal_wake_up(child, 1);
-               }
        }
        spin_unlock(&child->sighand->siglock);
 }
@@ -73,13 +72,10 @@ void __ptrace_unlink(struct task_struct *child)
        BUG_ON(!child->ptrace);
 
        child->ptrace = 0;
-       if (ptrace_reparented(child)) {
-               list_del_init(&child->ptrace_list);
-               remove_parent(child);
-               child->parent = child->real_parent;
-               add_parent(child);
-       }
+       child->parent = child->real_parent;
+       list_del_init(&child->ptrace_entry);
 
+       arch_ptrace_untrace(child);
        if (task_is_traced(child))
                ptrace_untrace(child);
 }
@@ -115,14 +111,16 @@ int ptrace_check_attach(struct task_struct *child, int kill)
        read_unlock(&tasklist_lock);
 
        if (!ret && !kill)
-               wait_task_inactive(child);
+               ret = wait_task_inactive(child, TASK_TRACED) ? 0 : -ESRCH;
 
        /* All systems go.. */
        return ret;
 }
 
-int __ptrace_may_attach(struct task_struct *task)
+int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 {
+       const struct cred *cred = current_cred(), *tcred;
+
        /* May we inspect the given task?
         * This check is used both for attaching with ptrace
         * and for allowing access to sensitive information in /proc.
@@ -135,27 +133,33 @@ int __ptrace_may_attach(struct task_struct *task)
        /* Don't let security modules deny introspection */
        if (task == current)
                return 0;
-       if (((current->uid != task->euid) ||
-            (current->uid != task->suid) ||
-            (current->uid != task->uid) ||
-            (current->gid != task->egid) ||
-            (current->gid != task->sgid) ||
-            (current->gid != task->gid)) && !capable(CAP_SYS_PTRACE))
+       rcu_read_lock();
+       tcred = __task_cred(task);
+       if ((cred->uid != tcred->euid ||
+            cred->uid != tcred->suid ||
+            cred->uid != tcred->uid  ||
+            cred->gid != tcred->egid ||
+            cred->gid != tcred->sgid ||
+            cred->gid != tcred->gid) &&
+           !capable(CAP_SYS_PTRACE)) {
+               rcu_read_unlock();
                return -EPERM;
+       }
+       rcu_read_unlock();
        smp_rmb();
        if (task->mm)
                dumpable = get_dumpable(task->mm);
        if (!dumpable && !capable(CAP_SYS_PTRACE))
                return -EPERM;
 
-       return security_ptrace(current, task);
+       return security_ptrace_access_check(task, mode);
 }
 
-int ptrace_may_attach(struct task_struct *task)
+bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 {
        int err;
        task_lock(task);
-       err = __ptrace_may_attach(task);
+       err = __ptrace_may_access(task, mode);
        task_unlock(task);
        return !err;
 }
@@ -163,69 +167,137 @@ int ptrace_may_attach(struct task_struct *task)
 int ptrace_attach(struct task_struct *task)
 {
        int retval;
-       unsigned long flags;
 
        audit_ptrace(task);
 
        retval = -EPERM;
+       if (unlikely(task->flags & PF_KTHREAD))
+               goto out;
        if (same_thread_group(task, current))
                goto out;
 
-repeat:
        /*
-        * Nasty, nasty.
-        *
-        * We want to hold both the task-lock and the
-        * tasklist_lock for writing at the same time.
-        * But that's against the rules (tasklist_lock
-        * is taken for reading by interrupts on other
-        * cpu's that may have task_lock).
+        * Protect exec's credential calculations against our interference;
+        * interference; SUID, SGID and LSM creds get determined differently
+        * under ptrace.
         */
-       task_lock(task);
-       if (!write_trylock_irqsave(&tasklist_lock, flags)) {
-               task_unlock(task);
-               do {
-                       cpu_relax();
-               } while (!write_can_lock(&tasklist_lock));
-               goto repeat;
-       }
+       retval = -ERESTARTNOINTR;
+       if (mutex_lock_interruptible(&task->cred_guard_mutex))
+               goto out;
 
-       if (!task->mm)
-               goto bad;
-       /* the same process cannot be attached many times */
-       if (task->ptrace & PT_PTRACED)
-               goto bad;
-       retval = __ptrace_may_attach(task);
+       task_lock(task);
+       retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
+       task_unlock(task);
        if (retval)
-               goto bad;
+               goto unlock_creds;
+
+       write_lock_irq(&tasklist_lock);
+       retval = -EPERM;
+       if (unlikely(task->exit_state))
+               goto unlock_tasklist;
+       if (task->ptrace)
+               goto unlock_tasklist;
 
-       /* Go */
-       task->ptrace |= PT_PTRACED;
+       task->ptrace = PT_PTRACED;
        if (capable(CAP_SYS_PTRACE))
                task->ptrace |= PT_PTRACE_CAP;
 
        __ptrace_link(task, current);
-
        send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
-bad:
-       write_unlock_irqrestore(&tasklist_lock, flags);
-       task_unlock(task);
+
+       retval = 0;
+unlock_tasklist:
+       write_unlock_irq(&tasklist_lock);
+unlock_creds:
+       mutex_unlock(&task->cred_guard_mutex);
 out:
        return retval;
 }
 
-static inline void __ptrace_detach(struct task_struct *child, unsigned int data)
+/**
+ * ptrace_traceme  --  helper for PTRACE_TRACEME
+ *
+ * Performs checks and sets PT_PTRACED.
+ * Should be used by all ptrace implementations for PTRACE_TRACEME.
+ */
+int ptrace_traceme(void)
+{
+       int ret = -EPERM;
+
+       write_lock_irq(&tasklist_lock);
+       /* Are we already being traced? */
+       if (!current->ptrace) {
+               ret = security_ptrace_traceme(current->parent);
+               /*
+                * Check PF_EXITING to ensure ->real_parent has not passed
+                * exit_ptrace(). Otherwise we don't report the error but
+                * pretend ->real_parent untraces us right after return.
+                */
+               if (!ret && !(current->real_parent->flags & PF_EXITING)) {
+                       current->ptrace = PT_PTRACED;
+                       __ptrace_link(current, current->real_parent);
+               }
+       }
+       write_unlock_irq(&tasklist_lock);
+
+       return ret;
+}
+
+/*
+ * Called with irqs disabled, returns true if childs should reap themselves.
+ */
+static int ignoring_children(struct sighand_struct *sigh)
 {
-       child->exit_code = data;
-       /* .. re-parent .. */
-       __ptrace_unlink(child);
-       /* .. and wake it up. */
-       if (child->exit_state != EXIT_ZOMBIE)
-               wake_up_process(child);
+       int ret;
+       spin_lock(&sigh->siglock);
+       ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
+             (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
+       spin_unlock(&sigh->siglock);
+       return ret;
+}
+
+/*
+ * Called with tasklist_lock held for writing.
+ * Unlink a traced task, and clean it up if it was a traced zombie.
+ * Return true if it needs to be reaped with release_task().
+ * (We can't call release_task() here because we already hold tasklist_lock.)
+ *
+ * If it's a zombie, our attachedness prevented normal parent notification
+ * or self-reaping.  Do notification now if it would have happened earlier.
+ * If it should reap itself, return true.
+ *
+ * If it's our own child, there is no notification to do. But if our normal
+ * children self-reap, then this child was prevented by ptrace and we must
+ * reap it now, in that case we must also wake up sub-threads sleeping in
+ * do_wait().
+ */
+static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
+{
+       __ptrace_unlink(p);
+
+       if (p->exit_state == EXIT_ZOMBIE) {
+               if (!task_detached(p) && thread_group_empty(p)) {
+                       if (!same_thread_group(p->real_parent, tracer))
+                               do_notify_parent(p, p->exit_signal);
+                       else if (ignoring_children(tracer->sighand)) {
+                               __wake_up_parent(p, tracer);
+                               p->exit_signal = -1;
+                       }
+               }
+               if (task_detached(p)) {
+                       /* Mark it as in the process of being reaped. */
+                       p->exit_state = EXIT_DEAD;
+                       return true;
+               }
+       }
+
+       return false;
 }
 
 int ptrace_detach(struct task_struct *child, unsigned int data)
 {
+       bool dead = false;
+
        if (!valid_signal(data))
                return -EIO;
 
@@ -234,14 +306,47 @@ int ptrace_detach(struct task_struct *child, unsigned int data)
        clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 
        write_lock_irq(&tasklist_lock);
-       /* protect against de_thread()->release_task() */
-       if (child->ptrace)
-               __ptrace_detach(child, data);
+       /*
+        * This child can be already killed. Make sure de_thread() or
+        * our sub-thread doing do_wait() didn't do release_task() yet.
+        */
+       if (child->ptrace) {
+               child->exit_code = data;
+               dead = __ptrace_detach(current, child);
+               if (!child->exit_state)
+                       wake_up_process(child);
+       }
        write_unlock_irq(&tasklist_lock);
 
+       if (unlikely(dead))
+               release_task(child);
+
        return 0;
 }
 
+/*
+ * Detach all tasks we were using ptrace on.
+ */
+void exit_ptrace(struct task_struct *tracer)
+{
+       struct task_struct *p, *n;
+       LIST_HEAD(ptrace_dead);
+
+       write_lock_irq(&tasklist_lock);
+       list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
+               if (__ptrace_detach(tracer, p))
+                       list_add(&p->ptrace_entry, &ptrace_dead);
+       }
+       write_unlock_irq(&tasklist_lock);
+
+       BUG_ON(!list_empty(&tracer->ptraced));
+
+       list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
+               list_del_init(&p->ptrace_entry);
+               release_task(p);
+       }
+}
+
 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 {
        int copied = 0;
@@ -262,7 +367,7 @@ int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst
                copied += retval;
                src += retval;
                dst += retval;
-               len -= retval;                  
+               len -= retval;
        }
        return copied;
 }
@@ -287,7 +392,7 @@ int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long ds
                copied += retval;
                src += retval;
                dst += retval;
-               len -= retval;                  
+               len -= retval;
        }
        return copied;
 }
@@ -322,37 +427,33 @@ static int ptrace_setoptions(struct task_struct *child, long data)
 
 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
 {
+       unsigned long flags;
        int error = -ESRCH;
 
-       read_lock(&tasklist_lock);
-       if (likely(child->sighand != NULL)) {
+       if (lock_task_sighand(child, &flags)) {
                error = -EINVAL;
-               spin_lock_irq(&child->sighand->siglock);
                if (likely(child->last_siginfo != NULL)) {
                        *info = *child->last_siginfo;
                        error = 0;
                }
-               spin_unlock_irq(&child->sighand->siglock);
+               unlock_task_sighand(child, &flags);
        }
-       read_unlock(&tasklist_lock);
        return error;
 }
 
 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
 {
+       unsigned long flags;
        int error = -ESRCH;
 
-       read_lock(&tasklist_lock);
-       if (likely(child->sighand != NULL)) {
+       if (lock_task_sighand(child, &flags)) {
                error = -EINVAL;
-               spin_lock_irq(&child->sighand->siglock);
                if (likely(child->last_siginfo != NULL)) {
                        *child->last_siginfo = *info;
                        error = 0;
                }
-               spin_unlock_irq(&child->sighand->siglock);
+               unlock_task_sighand(child, &flags);
        }
-       read_unlock(&tasklist_lock);
        return error;
 }
 
@@ -400,9 +501,9 @@ static int ptrace_resume(struct task_struct *child, long request, long data)
                if (unlikely(!arch_has_single_step()))
                        return -EIO;
                user_enable_single_step(child);
-       }
-       else
+       } else {
                user_disable_single_step(child);
+       }
 
        child->exit_code = data;
        wake_up_process(child);
@@ -479,52 +580,16 @@ int ptrace_request(struct task_struct *child, long request,
        return ret;
 }
 
-/**
- * ptrace_traceme  --  helper for PTRACE_TRACEME
- *
- * Performs checks and sets PT_PTRACED.
- * Should be used by all ptrace implementations for PTRACE_TRACEME.
- */
-int ptrace_traceme(void)
-{
-       int ret = -EPERM;
-
-       /*
-        * Are we already being traced?
-        */
-       task_lock(current);
-       if (!(current->ptrace & PT_PTRACED)) {
-               ret = security_ptrace(current->parent, current);
-               /*
-                * Set the ptrace bit in the process ptrace flags.
-                */
-               if (!ret)
-                       current->ptrace |= PT_PTRACED;
-       }
-       task_unlock(current);
-       return ret;
-}
-
-/**
- * ptrace_get_task_struct  --  grab a task struct reference for ptrace
- * @pid:       process id to grab a task_struct reference of
- *
- * This function is a helper for ptrace implementations.  It checks
- * permissions and then grabs a task struct for use of the actual
- * ptrace implementation.
- *
- * Returns the task_struct for @pid or an ERR_PTR() on failure.
- */
-struct task_struct *ptrace_get_task_struct(pid_t pid)
+static struct task_struct *ptrace_get_task_struct(pid_t pid)
 {
        struct task_struct *child;
 
-       read_lock(&tasklist_lock);
+       rcu_read_lock();
        child = find_task_by_vpid(pid);
        if (child)
                get_task_struct(child);
+       rcu_read_unlock();
 
-       read_unlock(&tasklist_lock);
        if (!child)
                return ERR_PTR(-ESRCH);
        return child;
@@ -534,7 +599,7 @@ struct task_struct *ptrace_get_task_struct(pid_t pid)
 #define arch_ptrace_attach(child)      do { } while (0)
 #endif
 
-asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
+SYSCALL_DEFINE4(ptrace, long, request, long, pid, long, addr, long, data)
 {
        struct task_struct *child;
        long ret;
@@ -572,8 +637,6 @@ asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
                goto out_put_task_struct;
 
        ret = arch_ptrace(child, request, addr, data);
-       if (ret < 0)
-               goto out_put_task_struct;
 
  out_put_task_struct:
        put_task_struct(child);
@@ -601,7 +664,7 @@ int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
        return (copied == sizeof(data)) ? 0 : -EIO;
 }
 
-#if defined CONFIG_COMPAT && defined __ARCH_WANT_COMPAT_SYS_PTRACE
+#if defined CONFIG_COMPAT
 #include <linux/compat.h>
 
 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
@@ -698,4 +761,4 @@ asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
        unlock_kernel();
        return ret;
 }
-#endif /* CONFIG_COMPAT && __ARCH_WANT_COMPAT_SYS_PTRACE */
+#endif /* CONFIG_COMPAT */