diff -urN linux-2.6.5-orig/Documentation/filesystems/proc.txt linux-2.6.5/Documentation/filesystems/proc.txt
--- linux-2.6.5-orig/Documentation/filesystems/proc.txt	2004-04-09 15:06:48.000000000 +0100
+++ linux-2.6.5/Documentation/filesystems/proc.txt	2004-04-10 20:43:56.000000000 +0100
@@ -130,6 +130,7 @@
  root	 Link to the root directory of this process
  stat    Process status                                 
  statm   Process memory status information              
+ statfd  Status of the processes file descriptors
  status  Process status in human readable form          
  wchan   If CONFIG_KALLSYMS is set, a pre-decoded wchan
 ..............................................................................
@@ -180,6 +181,13 @@
  dt       number of dirty pages           
 ..............................................................................
 
+The statfd contains one line per open file descriptor in the process - for
+example:
+16 821409 02 0x44009 0
+
+Is file descriptor 16, that is 821409 bytes into the file, has a mode of 02 (octal),
+has flags 0x44009 and no error.
+
 1.2 Kernel data
 ---------------
 
diff -urN linux-2.6.5-orig/fs/proc/base.c linux-2.6.5/fs/proc/base.c
--- linux-2.6.5-orig/fs/proc/base.c	2004-04-09 15:07:04.000000000 +0100
+++ linux-2.6.5/fs/proc/base.c	2004-04-10 20:34:12.000000000 +0100
@@ -57,6 +57,7 @@
 	PROC_TGID_CMDLINE,
 	PROC_TGID_STAT,
 	PROC_TGID_STATM,
+	PROC_TGID_STATFD,
 	PROC_TGID_MAPS,
 	PROC_TGID_MOUNTS,
 	PROC_TGID_WCHAN,
@@ -80,6 +81,7 @@
 	PROC_TID_CMDLINE,
 	PROC_TID_STAT,
 	PROC_TID_STATM,
+	PROC_TID_STATFD,
 	PROC_TID_MAPS,
 	PROC_TID_MOUNTS,
 	PROC_TID_WCHAN,
@@ -111,6 +113,7 @@
 	E(PROC_TGID_CMDLINE,   "cmdline", S_IFREG|S_IRUGO),
 	E(PROC_TGID_STAT,      "stat",    S_IFREG|S_IRUGO),
 	E(PROC_TGID_STATM,     "statm",   S_IFREG|S_IRUGO),
+	E(PROC_TGID_STATFD,    "statfd",  S_IFREG|S_IRUSR),
 	E(PROC_TGID_MAPS,      "maps",    S_IFREG|S_IRUGO),
 	E(PROC_TGID_MEM,       "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 	E(PROC_TGID_CWD,       "cwd",     S_IFLNK|S_IRWXUGO),
@@ -133,6 +136,7 @@
 	E(PROC_TID_CMDLINE,    "cmdline", S_IFREG|S_IRUGO),
 	E(PROC_TID_STAT,       "stat",    S_IFREG|S_IRUGO),
 	E(PROC_TID_STATM,      "statm",   S_IFREG|S_IRUGO),
+	E(PROC_TID_STATFD,     "statfd",  S_IFREG|S_IRUGO),
 	E(PROC_TID_MAPS,       "maps",    S_IFREG|S_IRUGO),
 	E(PROC_TID_MEM,        "mem",     S_IFREG|S_IRUSR|S_IWUSR),
 	E(PROC_TID_CWD,        "cwd",     S_IFLNK|S_IRWXUGO),
@@ -167,6 +171,12 @@
 
 #undef E
 
+static int pid_alive(struct task_struct *p)
+{
+	BUG_ON(p->pids[PIDTYPE_PID].pidptr != &p->pids[PIDTYPE_PID].pid);
+	return atomic_read(&p->pids[PIDTYPE_PID].pid.count);
+}
+
 static inline struct task_struct *proc_task(struct inode *inode)
 {
 	return PROC_I(inode)->task;
@@ -513,6 +523,162 @@
 	.release	= mounts_release,
 };
 
+/* ------------------------------------------------------------------------ 
+ * statfd: A file containing one line of information per open file     
+ *    We use the seq_file stuff to let us construct the file on the fly
+ *    The seq_file's private member contains a pointer to a pid_statfd_temps
+ *    The 'v' pointer contains the last loff_t pointer we were given (safe??)
+ *
+ *    (Derived from a mix of the code to do the fd directory, the
+ *    pid/map file and LWN article 22355)
+ *        Dave Gilbert (linux@treblig.org)
+ */
+
+struct pid_statfd_temps {
+	struct task_struct* task;
+	struct files_struct* files;
+};
+
+static void *statfd_start(struct seq_file *m, loff_t *pos)
+{
+	struct pid_statfd_temps* temps=(struct pid_statfd_temps*)m->private;
+	struct task_struct *task = temps->task;
+	int max_fds;
+
+	if (!pid_alive(task))
+		return NULL;
+
+  /* Get hold of the files structure from the task */
+	task_lock(task);
+	temps->files = task->files;
+	if (temps->files)
+		atomic_inc(&temps->files->count); /* stop it being freed while we've got it */
+	task_unlock(task);
+
+	/* Check to see if we are off the end */
+	spin_lock(&temps->files->file_lock);
+	max_fds=temps->files->max_fds;
+	spin_unlock(&temps->files->file_lock);
+
+	if (*pos >= max_fds)
+	{
+		return NULL;
+	}
+
+	return pos;
+}
+
+static void statfd_stop(struct seq_file *m, void *v)
+{
+	struct pid_statfd_temps* temps=(struct pid_statfd_temps*)m->private;
+
+	if (temps->files)
+	{
+		/* Release if they are finished with */
+		put_files_struct(temps->files);
+		temps->files=NULL;
+	}
+}
+
+static void *statfd_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	struct pid_statfd_temps* temps=(struct pid_statfd_temps*)m->private;
+
+	int max_fds;
+
+	/* Check to see if we are off the end */
+	spin_lock(&temps->files->file_lock);
+	max_fds=temps->files->max_fds;
+	spin_unlock(&temps->files->file_lock);
+
+	/* Lets try another one, and if OK just return the same pointer */
+	if (++(*pos) < max_fds)
+	{
+		return pos;
+	}
+
+	/* Have to release locks etc before we tell seq_file we're done */
+	statfd_stop(m,v);
+	return NULL;
+}
+
+static int statfd_show(struct seq_file *m, void *v)
+{
+	struct pid_statfd_temps* temps=(struct pid_statfd_temps*)m->private;
+	int fd;
+	struct file* file;
+
+	loff_t f_pos;
+	mode_t f_mode;
+	int f_error;
+	unsigned int f_flags;
+ 
+	fd=(int)*(loff_t*)v;
+
+	spin_lock(&temps->files->file_lock);
+	file = fcheck_files(temps->files, fd);
+
+	if (file) {
+		f_pos=file->f_pos;
+		f_mode=file->f_mode;
+		f_error=file->f_error;
+		f_flags=file->f_flags;
+	}
+
+	spin_unlock(&temps->files->file_lock);
+
+	if (file)
+		seq_printf(m,"%d %lld 0%o 0x%x %d\n",fd, (long long)f_pos, f_mode, f_flags, f_error);
+
+	return 0;
+}
+
+static struct seq_operations proc_pid_statfd_seqop = {
+	.start  = statfd_start,
+	.next = statfd_next,
+	.stop = statfd_stop,
+	.show = statfd_show
+};
+
+static int pid_statfd_open(struct inode *inode, struct file *file)
+{
+	struct pid_statfd_temps* temps=kmalloc(sizeof(*temps), GFP_KERNEL);
+	if (!temps)
+		return -ENOMEM;
+
+	temps->task = proc_task(inode);
+	temps->files = NULL;
+
+	int ret = seq_open(file, &proc_pid_statfd_seqop);
+	if (!ret) {
+		struct seq_file *m = file->private_data;
+		m->private = temps;
+	} else {
+		kfree(temps);
+	}
+
+	return ret;
+}
+
+static int pid_statfd_release(struct inode *inode, struct file *file)
+{
+	/* We need to clean up the temporary we allocated in pid_statfd_open
+	 * and then let seq_release do its cleanup */
+	struct seq_file *m = file->private_data;
+	struct pid_statfd_temps* temps = m->private;
+	kfree(temps);
+	return seq_release(inode, file);
+}
+
+static struct file_operations proc_pid_statfd_operations = {
+	.open     = pid_statfd_open,
+	.read     = seq_read,
+	.llseek   = seq_lseek,
+	.release  = pid_statfd_release
+};
+
+/* ------------------------------------------------------------------------ */
+
 #define PROC_BLOCK_SIZE	(3*1024)		/* 4K page size but our output routines use some slack for overruns */
 
 static ssize_t proc_info_read(struct file * file, char * buf,
@@ -771,12 +937,6 @@
 	.follow_link	= proc_pid_follow_link
 };
 
-static int pid_alive(struct task_struct *p)
-{
-	BUG_ON(p->pids[PIDTYPE_PID].pidptr != &p->pids[PIDTYPE_PID].pid);
-	return atomic_read(&p->pids[PIDTYPE_PID].pid.count);
-}
-
 #define NUMBUF 10
 
 static int proc_readfd(struct file * filp, void * dirent, filldir_t filldir)
@@ -1349,6 +1509,10 @@
 			inode->i_fop = &proc_info_file_operations;
 			ei->op.proc_read = proc_pid_statm;
 			break;
+		case PROC_TID_STATFD:
+		case PROC_TGID_STATFD:
+			inode->i_fop = &proc_pid_statfd_operations;
+			break;
 		case PROC_TID_MAPS:
 		case PROC_TGID_MAPS:
 			inode->i_fop = &proc_maps_operations;