[PATCH] proc_readfd race

[Maneesh Soni]

Hi Al,

proc_readfd calls fcheck_files() without taking read_lock() for the 
files_struct which could end up in race with expand_fd_array(). The patch
below puts the read_lock to avoid that. I am not very sure if this is already
protected by some other means and read_lock is not needed.

Please comment.

Thank you,
Maneesh

-- 
Maneesh Soni
IBM Linux Technology Center, 
IBM India Software Lab, Bangalore.
Phone: [phone] Extn. 3999 email: [e-mail]
http://lse.sourceforge.net/locking/rcupdate.html

diff -urN linux-2.4.14-pre8-orig/fs/proc/base.c linux-2.4.14pre8/fs/proc/base.c
--- linux-2.4.14-pre8-orig/fs/proc/base.c	Tue Nov  6 10:13:27 2001
+++ linux-2.4.14pre8/fs/proc/base.c	Tue Nov  6 10:02:19 2001
@@ -553,6 +553,7 @@
 			task_unlock(p);
 			if (!files)
 				goto out;
+			read_lock(&files->file_lock);
 			for (fd = filp->f_pos-2;
 			     fd < files->max_fds;
 			     fd++, filp->f_pos++) {
@@ -561,6 +562,7 @@
 				if (!fcheck_files(files, fd))
 					continue;
 
+				read_unlock(&files->file_lock);
 				j = NUMBUF;
 				i = fd;
 				do {
@@ -571,8 +573,11 @@
 
 				ino = fake_ino(pid, PROC_PID_FD_DIR + fd);
 				if (filldir(dirent, buf+j, NUMBUF-j, fd+2, ino, DT_LNK) < 0)
-					break;
+					goto done;
+				read_lock(&files->file_lock);
 			}
+			read_unlock(&files->file_lock);
+done:
 			put_files_struct(files);
 	}
 out: