Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / fs / exec.c
index 5d35813..c567e1a 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -22,7 +22,6 @@
  * formats. 
  */
 
-#include <linux/config.h>
 #include <linux/slab.h>
 #include <linux/file.h>
 #include <linux/mman.h>
 #include <linux/module.h>
 #include <linux/namei.h>
 #include <linux/proc_fs.h>
-#include <linux/ptrace.h>
+#include <linux/tracehook.h>
 #include <linux/mount.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
 #include <linux/rmap.h>
 #include <linux/acct.h>
 #include <linux/cn_proc.h>
+#include <linux/audit.h>
 #include <linux/vs_memory.h>
 #include <linux/vs_cvirt.h>
 
@@ -489,8 +489,6 @@ struct file *open_exec(const char *name)
                if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
                    S_ISREG(inode->i_mode)) {
                        int err = vfs_permission(&nd, MAY_EXEC);
-                       if (!err && !(inode->i_mode & 0111))
-                               err = -EACCES;
                        file = ERR_PTR(err);
                        if (!err) {
                                file = nameidata_to_filp(&nd, O_RDONLY);
@@ -668,8 +666,6 @@ static int de_thread(struct task_struct *tsk)
         * and to assume its PID:
         */
        if (!thread_group_leader(current)) {
-               struct dentry *proc_dentry1, *proc_dentry2;
-
                /*
                 * Wait for the thread group leader to be a zombie.
                 * It should already be zombie at this point, most
@@ -691,10 +687,6 @@ static int de_thread(struct task_struct *tsk)
                 */
                current->start_time = leader->start_time;
 
-               spin_lock(&leader->proc_lock);
-               spin_lock(&current->proc_lock);
-               proc_dentry1 = proc_pid_unhash(current);
-               proc_dentry2 = proc_pid_unhash(leader);
                write_lock_irq(&tasklist_lock);
 
                BUG_ON(leader->tgid != current->tgid);
@@ -715,7 +707,7 @@ static int de_thread(struct task_struct *tsk)
                attach_pid(current, PIDTYPE_PID,  current->pid);
                attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
                attach_pid(current, PIDTYPE_SID,  current->signal->session);
-               list_add_tail_rcu(&current->tasks, &init_task.tasks);
+               list_replace_rcu(&leader->tasks, &current->tasks);
 
                current->group_leader = current;
                leader->group_leader = current;
@@ -723,7 +715,6 @@ static int de_thread(struct task_struct *tsk)
                /* Reduce leader to a thread */
                detach_pid(leader, PIDTYPE_PGID);
                detach_pid(leader, PIDTYPE_SID);
-               list_del_init(&leader->tasks);
 
                current->exit_signal = SIGCHLD;
 
@@ -731,10 +722,6 @@ static int de_thread(struct task_struct *tsk)
                leader->exit_state = EXIT_DEAD;
 
                write_unlock_irq(&tasklist_lock);
-               spin_unlock(&leader->proc_lock);
-               spin_unlock(&current->proc_lock);
-               proc_pid_flush(proc_dentry1);
-               proc_pid_flush(proc_dentry2);
         }
 
        /*
@@ -767,7 +754,7 @@ no_thread_group:
 
                write_lock_irq(&tasklist_lock);
                spin_lock(&oldsighand->siglock);
-               spin_lock(&newsighand->siglock);
+               spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 
                rcu_assign_pointer(current->sighand, newsighand);
                recalc_sigpending();
@@ -868,7 +855,6 @@ int flush_old_exec(struct linux_binprm * bprm)
        bprm->mm = NULL;                /* We're using it now */
 
        /* This is the point of no return */
-       steal_locks(files);
        put_files_struct(files);
 
        current->sas_ss_sp = current->sas_ss_size = 0;
@@ -937,12 +923,6 @@ int prepare_binprm(struct linux_binprm *bprm)
        int retval;
 
        mode = inode->i_mode;
-       /*
-        * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
-        * generic_permission lets a non-executable through
-        */
-       if (!(mode & 0111))     /* with at least _one_ execute bit set */
-               return -EACCES;
        if (bprm->file->f_op == NULL)
                return -EACCES;
 
@@ -981,13 +961,7 @@ EXPORT_SYMBOL(prepare_binprm);
 
 static int unsafe_exec(struct task_struct *p)
 {
-       int unsafe = 0;
-       if (p->ptrace & PT_PTRACED) {
-               if (p->ptrace & PT_PTRACE_CAP)
-                       unsafe |= LSM_UNSAFE_PTRACE_CAP;
-               else
-                       unsafe |= LSM_UNSAFE_PTRACE;
-       }
+       int unsafe = tracehook_unsafe_exec(p);
        if (atomic_read(&p->fs->count) > 1 ||
            atomic_read(&p->files->count) > 1 ||
            atomic_read(&p->sighand->count) > 1)
@@ -1088,6 +1062,11 @@ int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
        /* kernel module loader fixup */
        /* so we don't try to load run modprobe in kernel space. */
        set_fs(USER_DS);
+
+       retval = audit_bprm(bprm);
+       if (retval)
+               return retval;
+
        retval = -ENOENT;
        for (try=0; try<2; try++) {
                read_lock(&binfmt_lock);
@@ -1107,6 +1086,7 @@ int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
                                bprm->file = NULL;
                                current->did_exec = 1;
                                proc_exec_connector(current);
+                               tracehook_report_exec(bprm, regs);
                                return retval;
                        }
                        read_lock(&binfmt_lock);
@@ -1377,67 +1357,102 @@ static void format_corename(char *corename, const char *pattern, long signr)
        *out_ptr = 0;
 }
 
-static void zap_threads (struct mm_struct *mm)
+static void zap_process(struct task_struct *start)
 {
-       struct task_struct *g, *p;
-       struct task_struct *tsk = current;
-       struct completion *vfork_done = tsk->vfork_done;
-       int traced = 0;
+       struct task_struct *t;
 
-       /*
-        * Make sure nobody is waiting for us to release the VM,
-        * otherwise we can deadlock when we wait on each other
-        */
-       if (vfork_done) {
-               tsk->vfork_done = NULL;
-               complete(vfork_done);
-       }
+       start->signal->flags = SIGNAL_GROUP_EXIT;
+       start->signal->group_stop_count = 0;
 
-       read_lock(&tasklist_lock);
-       do_each_thread(g,p)
-               if (mm == p->mm && p != tsk) {
-                       force_sig_specific(SIGKILL, p);
-                       mm->core_waiters++;
-                       if (unlikely(p->ptrace) &&
-                           unlikely(p->parent->mm == mm))
-                               traced = 1;
+       t = start;
+       do {
+               if (t != current && t->mm) {
+                       t->mm->core_waiters++;
+                       sigaddset(&t->pending.signal, SIGKILL);
+                       signal_wake_up(t, 1);
                }
-       while_each_thread(g,p);
+       } while ((t = next_thread(t)) != start);
+}
 
-       read_unlock(&tasklist_lock);
+static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
+                               int exit_code)
+{
+       struct task_struct *g, *p;
+       unsigned long flags;
+       int err = -EAGAIN;
+
+       spin_lock_irq(&tsk->sighand->siglock);
+       if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
+               tsk->signal->group_exit_code = exit_code;
+               zap_process(tsk);
+               err = 0;
+       }
+       spin_unlock_irq(&tsk->sighand->siglock);
+       if (err)
+               return err;
 
-       if (unlikely(traced)) {
-               /*
-                * We are zapping a thread and the thread it ptraces.
-                * If the tracee went into a ptrace stop for exit tracing,
-                * we could deadlock since the tracer is waiting for this
-                * coredump to finish.  Detach them so they can both die.
-                */
-               write_lock_irq(&tasklist_lock);
-               do_each_thread(g,p) {
-                       if (mm == p->mm && p != tsk &&
-                           p->ptrace && p->parent->mm == mm) {
-                               __ptrace_detach(p, 0);
+       if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
+               goto done;
+
+       rcu_read_lock();
+       for_each_process(g) {
+               if (g == tsk->group_leader)
+                       continue;
+
+               p = g;
+               do {
+                       if (p->mm) {
+                               if (p->mm == mm) {
+                                       /*
+                                        * p->sighand can't disappear, but
+                                        * may be changed by de_thread()
+                                        */
+                                       lock_task_sighand(p, &flags);
+                                       zap_process(p);
+                                       unlock_task_sighand(p, &flags);
+                               }
+                               break;
                        }
-               } while_each_thread(g,p);
-               write_unlock_irq(&tasklist_lock);
+               } while ((p = next_thread(p)) != g);
        }
+       rcu_read_unlock();
+done:
+       return mm->core_waiters;
 }
 
-static void coredump_wait(struct mm_struct *mm)
+static int coredump_wait(int exit_code)
 {
-       DECLARE_COMPLETION(startup_done);
+       struct task_struct *tsk = current;
+       struct mm_struct *mm = tsk->mm;
+       struct completion startup_done;
+       struct completion *vfork_done;
        int core_waiters;
 
+       init_completion(&mm->core_done);
+       init_completion(&startup_done);
        mm->core_startup_done = &startup_done;
 
-       zap_threads(mm);
-       core_waiters = mm->core_waiters;
+       core_waiters = zap_threads(tsk, mm, exit_code);
        up_write(&mm->mmap_sem);
 
+       if (unlikely(core_waiters < 0))
+               goto fail;
+
+       /*
+        * Make sure nobody is waiting for us to release the VM,
+        * otherwise we can deadlock when we wait on each other
+        */
+       vfork_done = tsk->vfork_done;
+       if (vfork_done) {
+               tsk->vfork_done = NULL;
+               complete(vfork_done);
+       }
+
        if (core_waiters)
                wait_for_completion(&startup_done);
+fail:
        BUG_ON(mm->core_waiters);
+       return core_waiters;
 }
 
 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
@@ -1473,22 +1488,9 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
        }
        mm->dumpable = 0;
 
-       retval = -EAGAIN;
-       spin_lock_irq(&current->sighand->siglock);
-       if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
-               current->signal->flags = SIGNAL_GROUP_EXIT;
-               current->signal->group_exit_code = exit_code;
-               current->signal->group_stop_count = 0;
-               retval = 0;
-       }
-       spin_unlock_irq(&current->sighand->siglock);
-       if (retval) {
-               up_write(&mm->mmap_sem);
+       retval = coredump_wait(exit_code);
+       if (retval < 0)
                goto fail;
-       }
-
-       init_completion(&mm->core_done);
-       coredump_wait(mm);
 
        /*
         * Clear any false indication of pending signals that might