[PATCH] convert sighand_cache to use SLAB_DESTROY_BY_RCU
[safe/jmp/linux-2.6] / fs / exec.c
index 6d95216..950ebd4 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -48,6 +48,7 @@
 #include <linux/syscalls.h>
 #include <linux/rmap.h>
 #include <linux/acct.h>
+#include <linux/cn_proc.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
@@ -126,7 +127,7 @@ asmlinkage long sys_uselib(const char __user * library)
        struct nameidata nd;
        int error;
 
-       error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ);
+       error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
        if (error)
                goto out;
 
@@ -134,7 +135,7 @@ asmlinkage long sys_uselib(const char __user * library)
        if (!S_ISREG(nd.dentry->d_inode->i_mode))
                goto exit;
 
-       error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
+       error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
        if (error)
                goto exit;
 
@@ -305,9 +306,6 @@ void install_arg_page(struct vm_area_struct *vma,
                        struct page *page, unsigned long address)
 {
        struct mm_struct *mm = vma->vm_mm;
-       pgd_t * pgd;
-       pud_t * pud;
-       pmd_t * pmd;
        pte_t * pte;
        spinlock_t *ptl;
 
@@ -315,14 +313,7 @@ void install_arg_page(struct vm_area_struct *vma,
                goto out;
 
        flush_dcache_page(page);
-       pgd = pgd_offset(mm, address);
-       pud = pud_alloc(mm, pgd, address);
-       if (!pud)
-               goto out;
-       pmd = pmd_alloc(mm, pud, address);
-       if (!pmd)
-               goto out;
-       pte = pte_alloc_map_lock(mm, pmd, address, &ptl);
+       pte = get_locked_pte(mm, address, &ptl);
        if (!pte)
                goto out;
        if (!pte_none(*pte)) {
@@ -333,7 +324,7 @@ void install_arg_page(struct vm_area_struct *vma,
        lru_cache_add_active(page);
        set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
                                        page, vma->vm_page_prot))));
-       page_add_anon_rmap(page, vma, address);
+       page_add_new_anon_rmap(page, vma, address);
        pte_unmap_unlock(pte, ptl);
 
        /* no need for flush_tlb */
@@ -486,7 +477,7 @@ struct file *open_exec(const char *name)
        int err;
        struct file *file;
 
-       err = path_lookup_open(name, LOOKUP_FOLLOW, &nd, FMODE_READ);
+       err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
        file = ERR_PTR(err);
 
        if (!err) {
@@ -494,7 +485,7 @@ struct file *open_exec(const char *name)
                file = ERR_PTR(-EACCES);
                if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
                    S_ISREG(inode->i_mode)) {
-                       int err = permission(inode, MAY_EXEC, &nd);
+                       int err = vfs_permission(&nd, MAY_EXEC);
                        if (!err && !(inode->i_mode & 0111))
                                err = -EACCES;
                        file = ERR_PTR(err);
@@ -584,11 +575,12 @@ static int exec_mmap(struct mm_struct *mm)
  * disturbing other processes.  (Other processes might share the signal
  * table via the CLONE_SIGHAND option to clone().)
  */
-static inline int de_thread(struct task_struct *tsk)
+static int de_thread(struct task_struct *tsk)
 {
        struct signal_struct *sig = tsk->signal;
        struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
        spinlock_t *lock = &oldsighand->siglock;
+       struct task_struct *leader = NULL;
        int count;
 
        /*
@@ -624,6 +616,15 @@ static inline int de_thread(struct task_struct *tsk)
                kmem_cache_free(sighand_cachep, newsighand);
                return -EAGAIN;
        }
+
+       /*
+        * child_reaper ignores SIGKILL, change it now.
+        * Reparenting needs write_lock on tasklist_lock,
+        * so it is safe to do it under read_lock.
+        */
+       if (unlikely(current->group_leader == child_reaper))
+               child_reaper = current;
+
        zap_other_threads(current);
        read_unlock(&tasklist_lock);
 
@@ -640,10 +641,10 @@ static inline int de_thread(struct task_struct *tsk)
                 * synchronize with any firing (by calling del_timer_sync)
                 * before we can safely let the old group leader die.
                 */
-               sig->real_timer.data = (unsigned long)current;
+               sig->tsk = current;
                spin_unlock_irq(lock);
-               if (del_timer_sync(&sig->real_timer))
-                       add_timer(&sig->real_timer);
+               if (hrtimer_cancel(&sig->real_timer))
+                       hrtimer_restart(&sig->real_timer);
                spin_lock_irq(lock);
        }
        while (atomic_read(&sig->count) > count) {
@@ -664,15 +665,16 @@ static inline int de_thread(struct task_struct *tsk)
         * and to assume its PID:
         */
        if (!thread_group_leader(current)) {
-               struct task_struct *leader = current->group_leader, *parent;
+               struct task_struct *parent;
                struct dentry *proc_dentry1, *proc_dentry2;
-               unsigned long exit_state, ptrace;
+               unsigned long ptrace;
 
                /*
                 * Wait for the thread group leader to be a zombie.
                 * It should already be zombie at this point, most
                 * of the time.
                 */
+               leader = current->group_leader;
                while (leader->exit_state != EXIT_ZOMBIE)
                        yield();
 
@@ -706,33 +708,40 @@ static inline int de_thread(struct task_struct *tsk)
                remove_parent(current);
                remove_parent(leader);
 
-               switch_exec_pids(leader, current);
+
+               /* Become a process group leader with the old leader's pid.
+                * Note: The old leader also uses thispid until release_task
+                *       is called.  Odd but simple and correct.
+                */
+               detach_pid(current, PIDTYPE_PID);
+               current->pid = leader->pid;
+               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(&current->tasks, &init_task.tasks);
 
                current->parent = current->real_parent = leader->real_parent;
                leader->parent = leader->real_parent = child_reaper;
                current->group_leader = current;
                leader->group_leader = leader;
 
-               add_parent(current, current->parent);
-               add_parent(leader, leader->parent);
+               add_parent(current);
+               add_parent(leader);
                if (ptrace) {
                        current->ptrace = ptrace;
                        __ptrace_link(current, parent);
                }
 
-               list_del(&current->tasks);
-               list_add_tail(&current->tasks, &init_task.tasks);
                current->exit_signal = SIGCHLD;
-               exit_state = leader->exit_state;
+
+               BUG_ON(leader->exit_state != EXIT_ZOMBIE);
+               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);
-
-               BUG_ON(exit_state != EXIT_ZOMBIE);
-               release_task(leader);
         }
 
        /*
@@ -742,8 +751,11 @@ static inline int de_thread(struct task_struct *tsk)
        sig->flags = 0;
 
 no_thread_group:
-       BUG_ON(atomic_read(&sig->count) != 1);
        exit_itimers(sig);
+       if (leader)
+               release_task(leader);
+
+       BUG_ON(atomic_read(&sig->count) != 1);
 
        if (atomic_read(&oldsighand->count) == 1) {
                /*
@@ -756,7 +768,6 @@ no_thread_group:
                /*
                 * Move our state over to newsighand and switch it in.
                 */
-               spin_lock_init(&newsighand->siglock);
                atomic_set(&newsighand->count, 1);
                memcpy(newsighand->action, oldsighand->action,
                       sizeof(newsighand->action));
@@ -765,7 +776,7 @@ no_thread_group:
                spin_lock(&oldsighand->siglock);
                spin_lock(&newsighand->siglock);
 
-               current->sighand = newsighand;
+               rcu_assign_pointer(current->sighand, newsighand);
                recalc_sigpending();
 
                spin_unlock(&newsighand->siglock);
@@ -785,7 +796,7 @@ no_thread_group:
  * so that a new one can be started
  */
 
-static inline void flush_old_files(struct files_struct * files)
+static void flush_old_files(struct files_struct * files)
 {
        long j = -1;
        struct fdtable *fdt;
@@ -890,8 +901,14 @@ int flush_old_exec(struct linux_binprm * bprm)
        current->flags &= ~PF_RANDOMIZE;
        flush_thread();
 
+       /* Set the new mm task size. We have to do that late because it may
+        * depend on TIF_32BIT which is only updated in flush_thread() on
+        * some architectures like powerpc
+        */
+       current->mm->task_size = TASK_SIZE;
+
        if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
-           permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
+           file_permission(bprm->file, MAY_READ) ||
            (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
                suid_keys(current);
                current->mm->dumpable = suid_dumpable;
@@ -969,7 +986,7 @@ int prepare_binprm(struct linux_binprm *bprm)
 
 EXPORT_SYMBOL(prepare_binprm);
 
-static inline int unsafe_exec(struct task_struct *p)
+static int unsafe_exec(struct task_struct *p)
 {
        int unsafe = 0;
        if (p->ptrace & PT_PTRACED) {
@@ -1096,6 +1113,7 @@ int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
                                        fput(bprm->file);
                                bprm->file = NULL;
                                current->did_exec = 1;
+                               proc_exec_connector(current);
                                return retval;
                        }
                        read_lock(&binfmt_lock);
@@ -1141,10 +1159,9 @@ int do_execve(char * filename,
        int i;
 
        retval = -ENOMEM;
-       bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
+       bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
        if (!bprm)
                goto out_ret;
-       memset(bprm, 0, sizeof(*bprm));
 
        file = open_exec(filename);
        retval = PTR_ERR(file);
@@ -1407,7 +1424,7 @@ static void zap_threads (struct mm_struct *mm)
                do_each_thread(g,p) {
                        if (mm == p->mm && p != tsk &&
                            p->ptrace && p->parent->mm == mm) {
-                               __ptrace_unlink(p);
+                               __ptrace_detach(p, 0);
                        }
                } while_each_thread(g,p);
                write_unlock_irq(&tasklist_lock);
@@ -1460,18 +1477,28 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
                current->fsuid = 0;     /* Dump root private */
        }
        mm->dumpable = 0;
-       init_completion(&mm->core_done);
+
+       retval = -EAGAIN;
        spin_lock_irq(&current->sighand->siglock);
-       current->signal->flags = SIGNAL_GROUP_EXIT;
-       current->signal->group_exit_code = exit_code;
+       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);
+               goto fail;
+       }
+
+       init_completion(&mm->core_done);
        coredump_wait(mm);
 
        /*
         * Clear any false indication of pending signals that might
         * be seen by the filesystem code called to write the core file.
         */
-       current->signal->group_stop_count = 0;
        clear_thread_flag(TIF_SIGPENDING);
 
        if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
@@ -1499,7 +1526,7 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
                goto close_fail;
        if (!file->f_op->write)
                goto close_fail;
-       if (do_truncate(file->f_dentry, 0) != 0)
+       if (do_truncate(file->f_dentry, 0, 0, file) != 0)
                goto close_fail;
 
        retval = binfmt->core_dump(signr, regs, file);