[PATCH] proc: Rewrite the proc dentry flush on exit optimization

To keep the dcache from filling up with dead /proc entries we flush them on
process exit.  However over the years that code has gotten hairy with a
dentry_pointer and a lock in task_struct and misdocumented as a correctness
feature.

I have rewritten this code to look and see if we have a corresponding entry in
the dcache and if so flush it on process exit.  This removes the extra fields
in the task_struct and allows me to trivially handle the case of a
/proc/<tgid>/task/<pid> entry as well as the current /proc/<pid> entries.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

fs/exec.c

@@ -666,8 +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
@@ -689,10 +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);
@@ -729,10 +723,6 @@ 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);
         }
 
 	/*

fs/proc/base.c

@@ -1352,16 +1352,6 @@ static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
 	return 0;
 }
 
-static void pid_base_iput(struct dentry *dentry, struct inode *inode)
-{
-	struct task_struct *task = proc_task(inode);
-	spin_lock(&task->proc_lock);
-	if (task->proc_dentry == dentry)
-		task->proc_dentry = NULL;
-	spin_unlock(&task->proc_lock);
-	iput(inode);
-}
-
 static int pid_delete_dentry(struct dentry * dentry)
 {
 	/* Is the task we represent dead?
@@ -1383,13 +1373,6 @@ static struct dentry_operations pid_dentry_operations =
 	.d_delete	= pid_delete_dentry,
 };
 
-static struct dentry_operations pid_base_dentry_operations =
-{
-	.d_revalidate	= pid_revalidate,
-	.d_iput		= pid_base_iput,
-	.d_delete	= pid_delete_dentry,
-};
-
 /* Lookups */
 
 static unsigned name_to_int(struct dentry *dentry)
@@ -1859,57 +1842,70 @@ static struct inode_operations proc_self_inode_operations = {
 };
 
 /**
- * proc_pid_unhash -  Unhash /proc/@pid entry from the dcache.
+ * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
- * @p: task that should be flushed.
  *
- * Drops the /proc/@pid dcache entry from the hash chains.
+ * @task: task that should be flushed.
  *
- * Dropping /proc/@pid entries and detach_pid must be synchroneous,
+ * Looks in the dcache for
- * otherwise e.g. /proc/@pid/exe might point to the wrong executable,
+ * /proc/@pid
- * if the pid value is immediately reused. This is enforced by
+ * /proc/@tgid/task/@pid
- * - caller must acquire spin_lock(p->proc_lock)
+ * if either directory is present flushes it and all of it'ts children
- * - must be called before detach_pid()
+ * from the dcache.
- * - proc_pid_lookup acquires proc_lock, and checks that
+ *
- *   the target is not dead by looking at the attach count
+ * It is safe and reasonable to cache /proc entries for a task until
- *   of PIDTYPE_PID.
+ * that task exits.  After that they just clog up the dcache with
+ * useless entries, possibly causing useful dcache entries to be
+ * flushed instead.  This routine is proved to flush those useless
+ * dcache entries at process exit time.
+ *
+ * NOTE: This routine is just an optimization so it does not guarantee
+ *       that no dcache entries will exist at process exit time it
+ *       just makes it very unlikely that any will persist.
  */
-
+void proc_flush_task(struct task_struct *task)
-struct dentry *proc_pid_unhash(struct task_struct *p)
 {
-	struct dentry *proc_dentry;
+	struct dentry *dentry, *leader, *dir;
+	char buf[30];
+	struct qstr name;
 
-	proc_dentry = p->proc_dentry;
+	name.name = buf;
-	if (proc_dentry != NULL) {
+	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
-
+	dentry = d_hash_and_lookup(proc_mnt->mnt_root, &name);
-		spin_lock(&dcache_lock);
+	if (dentry) {
-		spin_lock(&proc_dentry->d_lock);
+		shrink_dcache_parent(dentry);
-		if (!d_unhashed(proc_dentry)) {
+		d_drop(dentry);
-			dget_locked(proc_dentry);
+		dput(dentry);
-			__d_drop(proc_dentry);
-			spin_unlock(&proc_dentry->d_lock);
-		} else {
-			spin_unlock(&proc_dentry->d_lock);
-			proc_dentry = NULL;
-		}
-		spin_unlock(&dcache_lock);
 	}
-	return proc_dentry;
-}
 
-/**
+	if (thread_group_leader(task))
- * proc_pid_flush - recover memory used by stale /proc/@pid/x entries
+		goto out;
- * @proc_dentry: directoy to prune.
+
- *
+	name.name = buf;
- * Shrink the /proc directory that was used by the just killed thread.
+	name.len = snprintf(buf, sizeof(buf), "%d", task->tgid);
- */
+	leader = d_hash_and_lookup(proc_mnt->mnt_root, &name);
-	
+	if (!leader)
-void proc_pid_flush(struct dentry *proc_dentry)
+		goto out;
-{
+
-	might_sleep();
+	name.name = "task";
-	if(proc_dentry != NULL) {
+	name.len = strlen(name.name);
-		shrink_dcache_parent(proc_dentry);
+	dir = d_hash_and_lookup(leader, &name);
-		dput(proc_dentry);
+	if (!dir)
+		goto out_put_leader;
+
+	name.name = buf;
+	name.len = snprintf(buf, sizeof(buf), "%d", task->pid);
+	dentry = d_hash_and_lookup(dir, &name);
+	if (dentry) {
+		shrink_dcache_parent(dentry);
+		d_drop(dentry);
+		dput(dentry);
 	}
+
+	dput(dir);
+out_put_leader:
+	dput(leader);
+out:
+	return;
 }
 
 /* SMP-safe */
@@ -1919,7 +1915,6 @@ struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct
 	struct inode *inode;
 	struct proc_inode *ei;
 	unsigned tgid;
-	int died;
 
 	if (dentry->d_name.len == 4 && !memcmp(dentry->d_name.name,"self",4)) {
 		inode = new_inode(dir->i_sb);
@@ -1965,23 +1960,16 @@ struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct
 	inode->i_nlink = 4;
 #endif
 
-	dentry->d_op = &pid_base_dentry_operations;
+	dentry->d_op = &pid_dentry_operations;
 
-	died = 0;
 	d_add(dentry, inode);
-	spin_lock(&task->proc_lock);
-	task->proc_dentry = dentry;
 	if (!pid_alive(task)) {
-		dentry = proc_pid_unhash(task);
+		d_drop(dentry);
-		died = 1;
+		shrink_dcache_parent(dentry);
-	}
-	spin_unlock(&task->proc_lock);
-
-	put_task_struct(task);
-	if (died) {
-		proc_pid_flush(dentry);
 		goto out;
 	}
+
+	put_task_struct(task);
 	return NULL;
 out:
 	return ERR_PTR(-ENOENT);
@@ -2024,7 +2012,7 @@ static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry
 	inode->i_nlink = 3;
 #endif
 
-	dentry->d_op = &pid_base_dentry_operations;
+	dentry->d_op = &pid_dentry_operations;
 
 	d_add(dentry, inode);
 

include/linux/init_task.h

@@ -119,7 +119,6 @@ extern struct group_info init_groups;
 		.signal = {{0}}},					\
 	.blocked	= {{0}},					\
 	.alloc_lock	= SPIN_LOCK_UNLOCKED,				\
-	.proc_lock	= SPIN_LOCK_UNLOCKED,				\
 	.journal_info	= NULL,						\
 	.cpu_timers	= INIT_CPU_TIMERS(tsk.cpu_timers),		\
 	.fs_excl	= ATOMIC_INIT(0),				\

include/linux/proc_fs.h

@@ -99,9 +99,8 @@ extern void proc_misc_init(void);
 
 struct mm_struct;
 
+void proc_flush_task(struct task_struct *task);
 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *);
-struct dentry *proc_pid_unhash(struct task_struct *p);
-void proc_pid_flush(struct dentry *proc_dentry);
 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir);
 unsigned long task_vsize(struct mm_struct *);
 int task_statm(struct mm_struct *, int *, int *, int *, int *);
@@ -211,8 +210,7 @@ static inline void proc_net_remove(const char *name)
 #define proc_net_create(name, mode, info)	({ (void)(mode), NULL; })
 static inline void proc_net_remove(const char *name) {}
 
-static inline struct dentry *proc_pid_unhash(struct task_struct *p) { return NULL; }
+static inline void proc_flush_task(struct task_struct *task) { }
-static inline void proc_pid_flush(struct dentry *proc_dentry) { }
 
 static inline struct proc_dir_entry *create_proc_entry(const char *name,
 	mode_t mode, struct proc_dir_entry *parent) { return NULL; }

include/linux/sched.h

@@ -842,8 +842,6 @@ struct task_struct {
    	u32 self_exec_id;
 /* Protection of (de-)allocation: mm, files, fs, tty, keyrings */
 	spinlock_t alloc_lock;
-/* Protection of proc_dentry: nesting proc_lock, dcache_lock, write_lock_irq(&tasklist_lock); */
-	spinlock_t proc_lock;
 
 #ifdef CONFIG_DEBUG_MUTEXES
 	/* mutex deadlock detection */
@@ -856,7 +854,6 @@ struct task_struct {
 /* VM state */
 	struct reclaim_state *reclaim_state;
 
-	struct dentry *proc_dentry;
 	struct backing_dev_info *backing_dev_info;
 
 	struct io_context *io_context;

kernel/exit.c

@@ -137,12 +137,8 @@ void release_task(struct task_struct * p)
 {
 	int zap_leader;
 	task_t *leader;
-	struct dentry *proc_dentry;
-
 repeat:
 	atomic_dec(&p->user->processes);
-	spin_lock(&p->proc_lock);
-	proc_dentry = proc_pid_unhash(p);
 	write_lock_irq(&tasklist_lock);
 	ptrace_unlink(p);
 	BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
@@ -171,8 +167,7 @@ repeat:
 
 	sched_exit(p);
 	write_unlock_irq(&tasklist_lock);
-	spin_unlock(&p->proc_lock);
+	proc_flush_task(p);
-	proc_pid_flush(proc_dentry);
 	release_thread(p);
 	call_rcu(&p->rcu, delayed_put_task_struct);
 

kernel/fork.c

@@ -993,13 +993,10 @@ static task_t *copy_process(unsigned long clone_flags,
 		if (put_user(p->pid, parent_tidptr))
 			goto bad_fork_cleanup;
 
-	p->proc_dentry = NULL;
-
 	INIT_LIST_HEAD(&p->children);
 	INIT_LIST_HEAD(&p->sibling);
 	p->vfork_done = NULL;
 	spin_lock_init(&p->alloc_lock);
-	spin_lock_init(&p->proc_lock);
 
 	clear_tsk_thread_flag(p, TIF_SIGPENDING);
 	init_sigpending(&p->pending);