proc: fix races against execve() of /proc/PID/fd**
authorVasiliy Kulikov <segoon@openwall.com>
Wed, 2 Nov 2011 20:38:44 +0000 (13:38 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 2 Nov 2011 23:07:00 +0000 (16:07 -0700)
fd* files are restricted to the task's owner, and other users may not get
direct access to them.  But one may open any of these files and run any
setuid program, keeping opened file descriptors.  As there are permission
checks on open(), but not on readdir() and read(), operations on the kept
file descriptors will not be checked.  It makes it possible to violate
procfs permission model.

Reading fdinfo/* may disclosure current fds' position and flags, reading
directory contents of fdinfo/ and fd/ may disclosure the number of opened
files by the target task.  This information is not sensible per se, but it
can reveal some private information (like length of a password stored in a
file) under certain conditions.

Used existing (un)lock_trace functions to check for ptrace_may_access(),
but instead of using EPERM return code from it use EACCES to be consistent
with existing proc_pid_follow_link()/proc_pid_readlink() return code.  If
they differ, attacker can guess what fds exist by analyzing stat() return
code.  Patched handlers: stat() for fd/*, stat() and read() for fdindo/*,
readdir() and lookup() for fd/ and fdinfo/.

Signed-off-by: Vasiliy Kulikov <segoon@openwall.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: <stable@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/base.c

index 8f0087e..d4f4913 100644 (file)
@@ -1652,12 +1652,46 @@ out:
        return error;
 }
 
+static int proc_pid_fd_link_getattr(struct vfsmount *mnt, struct dentry *dentry,
+               struct kstat *stat)
+{
+       struct inode *inode = dentry->d_inode;
+       struct task_struct *task = get_proc_task(inode);
+       int rc;
+
+       if (task == NULL)
+               return -ESRCH;
+
+       rc = -EACCES;
+       if (lock_trace(task))
+               goto out_task;
+
+       generic_fillattr(inode, stat);
+       unlock_trace(task);
+       rc = 0;
+out_task:
+       put_task_struct(task);
+       return rc;
+}
+
 static const struct inode_operations proc_pid_link_inode_operations = {
        .readlink       = proc_pid_readlink,
        .follow_link    = proc_pid_follow_link,
        .setattr        = proc_setattr,
 };
 
+static const struct inode_operations proc_fdinfo_link_inode_operations = {
+       .setattr        = proc_setattr,
+       .getattr        = proc_pid_fd_link_getattr,
+};
+
+static const struct inode_operations proc_fd_link_inode_operations = {
+       .readlink       = proc_pid_readlink,
+       .follow_link    = proc_pid_follow_link,
+       .setattr        = proc_setattr,
+       .getattr        = proc_pid_fd_link_getattr,
+};
+
 
 /* building an inode */
 
@@ -1889,49 +1923,61 @@ out:
 
 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
 {
-       struct task_struct *task = get_proc_task(inode);
-       struct files_struct *files = NULL;
+       struct task_struct *task;
+       struct files_struct *files;
        struct file *file;
        int fd = proc_fd(inode);
+       int rc;
 
-       if (task) {
-               files = get_files_struct(task);
-               put_task_struct(task);
-       }
-       if (files) {
-               /*
-                * We are not taking a ref to the file structure, so we must
-                * hold ->file_lock.
-                */
-               spin_lock(&files->file_lock);
-               file = fcheck_files(files, fd);
-               if (file) {
-                       unsigned int f_flags;
-                       struct fdtable *fdt;
-
-                       fdt = files_fdtable(files);
-                       f_flags = file->f_flags & ~O_CLOEXEC;
-                       if (FD_ISSET(fd, fdt->close_on_exec))
-                               f_flags |= O_CLOEXEC;
-
-                       if (path) {
-                               *path = file->f_path;
-                               path_get(&file->f_path);
-                       }
-                       if (info)
-                               snprintf(info, PROC_FDINFO_MAX,
-                                        "pos:\t%lli\n"
-                                        "flags:\t0%o\n",
-                                        (long long) file->f_pos,
-                                        f_flags);
-                       spin_unlock(&files->file_lock);
-                       put_files_struct(files);
-                       return 0;
+       task = get_proc_task(inode);
+       if (!task)
+               return -ENOENT;
+
+       rc = -EACCES;
+       if (lock_trace(task))
+               goto out_task;
+
+       rc = -ENOENT;
+       files = get_files_struct(task);
+       if (files == NULL)
+               goto out_unlock;
+
+       /*
+        * We are not taking a ref to the file structure, so we must
+        * hold ->file_lock.
+        */
+       spin_lock(&files->file_lock);
+       file = fcheck_files(files, fd);
+       if (file) {
+               unsigned int f_flags;
+               struct fdtable *fdt;
+
+               fdt = files_fdtable(files);
+               f_flags = file->f_flags & ~O_CLOEXEC;
+               if (FD_ISSET(fd, fdt->close_on_exec))
+                       f_flags |= O_CLOEXEC;
+
+               if (path) {
+                       *path = file->f_path;
+                       path_get(&file->f_path);
                }
-               spin_unlock(&files->file_lock);
-               put_files_struct(files);
-       }
-       return -ENOENT;
+               if (info)
+                       snprintf(info, PROC_FDINFO_MAX,
+                                "pos:\t%lli\n"
+                                "flags:\t0%o\n",
+                                (long long) file->f_pos,
+                                f_flags);
+               rc = 0;
+       } else
+               rc = -ENOENT;
+       spin_unlock(&files->file_lock);
+       put_files_struct(files);
+
+out_unlock:
+       unlock_trace(task);
+out_task:
+       put_task_struct(task);
+       return rc;
 }
 
 static int proc_fd_link(struct inode *inode, struct path *path)
@@ -2026,7 +2072,7 @@ static struct dentry *proc_fd_instantiate(struct inode *dir,
        spin_unlock(&files->file_lock);
        put_files_struct(files);
 
-       inode->i_op = &proc_pid_link_inode_operations;
+       inode->i_op = &proc_fd_link_inode_operations;
        inode->i_size = 64;
        ei->op.proc_get_link = proc_fd_link;
        d_set_d_op(dentry, &tid_fd_dentry_operations);
@@ -2058,7 +2104,12 @@ static struct dentry *proc_lookupfd_common(struct inode *dir,
        if (fd == ~0U)
                goto out;
 
+       result = ERR_PTR(-EACCES);
+       if (lock_trace(task))
+               goto out;
+
        result = instantiate(dir, dentry, task, &fd);
+       unlock_trace(task);
 out:
        put_task_struct(task);
 out_no_task:
@@ -2078,23 +2129,28 @@ static int proc_readfd_common(struct file * filp, void * dirent,
        retval = -ENOENT;
        if (!p)
                goto out_no_task;
+
+       retval = -EACCES;
+       if (lock_trace(p))
+               goto out;
+
        retval = 0;
 
        fd = filp->f_pos;
        switch (fd) {
                case 0:
                        if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
-                               goto out;
+                               goto out_unlock;
                        filp->f_pos++;
                case 1:
                        ino = parent_ino(dentry);
                        if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
-                               goto out;
+                               goto out_unlock;
                        filp->f_pos++;
                default:
                        files = get_files_struct(p);
                        if (!files)
-                               goto out;
+                               goto out_unlock;
                        rcu_read_lock();
                        for (fd = filp->f_pos-2;
                             fd < files_fdtable(files)->max_fds;
@@ -2118,6 +2174,9 @@ static int proc_readfd_common(struct file * filp, void * dirent,
                        rcu_read_unlock();
                        put_files_struct(files);
        }
+
+out_unlock:
+       unlock_trace(p);
 out:
        put_task_struct(p);
 out_no_task:
@@ -2195,6 +2254,7 @@ static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
        ei->fd = fd;
        inode->i_mode = S_IFREG | S_IRUSR;
        inode->i_fop = &proc_fdinfo_file_operations;
+       inode->i_op = &proc_fdinfo_link_inode_operations;
        d_set_d_op(dentry, &tid_fd_dentry_operations);
        d_add(dentry, inode);
        /* Close the race of the process dying before we return the dentry */