svcrdma: Shrink scope of spinlock on RQ CQ
[safe/jmp/linux-2.6] / fs / namei.c
index 89ef317..32fd965 100644 (file)
@@ -30,6 +30,7 @@
 #include <linux/capability.h>
 #include <linux/file.h>
 #include <linux/fcntl.h>
+#include <linux/device_cgroup.h>
 #include <asm/namei.h>
 #include <asm/uaccess.h>
 
@@ -281,6 +282,10 @@ int permission(struct inode *inode, int mask, struct nameidata *nd)
        if (retval)
                return retval;
 
+       retval = devcgroup_inode_permission(inode, mask);
+       if (retval)
+               return retval;
+
        return security_inode_permission(inode, mask, nd);
 }
 
@@ -1623,8 +1628,7 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
                        return -EACCES;
 
                flag &= ~O_TRUNC;
-       } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE))
-               return -EROFS;
+       }
 
        error = vfs_permission(nd, acc_mode);
        if (error)
@@ -1724,18 +1728,32 @@ static inline int open_to_namei_flags(int flag)
        return flag;
 }
 
+static int open_will_write_to_fs(int flag, struct inode *inode)
+{
+       /*
+        * We'll never write to the fs underlying
+        * a device file.
+        */
+       if (special_file(inode->i_mode))
+               return 0;
+       return (flag & O_TRUNC);
+}
+
 /*
- * Note that the low bits of "flag" aren't the same as in the open
- * system call.  See open_to_namei_flags().
+ * Note that the low bits of the passed in "open_flag"
+ * are not the same as in the local variable "flag". See
+ * open_to_namei_flags() for more details.
  */
 struct file *do_filp_open(int dfd, const char *pathname,
                int open_flag, int mode)
 {
+       struct file *filp;
        struct nameidata nd;
        int acc_mode, error;
        struct path path;
        struct dentry *dir;
        int count = 0;
+       int will_write;
        int flag = open_to_namei_flags(open_flag);
 
        acc_mode = ACC_MODE(flag);
@@ -1791,17 +1809,30 @@ do_last:
        }
 
        if (IS_ERR(nd.intent.open.file)) {
-               mutex_unlock(&dir->d_inode->i_mutex);
                error = PTR_ERR(nd.intent.open.file);
-               goto exit_dput;
+               goto exit_mutex_unlock;
        }
 
        /* Negative dentry, just create the file */
        if (!path.dentry->d_inode) {
-               error = __open_namei_create(&nd, &path, flag, mode);
+               /*
+                * This write is needed to ensure that a
+                * ro->rw transition does not occur between
+                * the time when the file is created and when
+                * a permanent write count is taken through
+                * the 'struct file' in nameidata_to_filp().
+                */
+               error = mnt_want_write(nd.path.mnt);
                if (error)
+                       goto exit_mutex_unlock;
+               error = __open_namei_create(&nd, &path, flag, mode);
+               if (error) {
+                       mnt_drop_write(nd.path.mnt);
                        goto exit;
-               return nameidata_to_filp(&nd, open_flag);
+               }
+               filp = nameidata_to_filp(&nd, open_flag);
+               mnt_drop_write(nd.path.mnt);
+               return filp;
        }
 
        /*
@@ -1831,11 +1862,40 @@ do_last:
        if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
                goto exit;
 ok:
+       /*
+        * Consider:
+        * 1. may_open() truncates a file
+        * 2. a rw->ro mount transition occurs
+        * 3. nameidata_to_filp() fails due to
+        *    the ro mount.
+        * That would be inconsistent, and should
+        * be avoided. Taking this mnt write here
+        * ensures that (2) can not occur.
+        */
+       will_write = open_will_write_to_fs(flag, nd.path.dentry->d_inode);
+       if (will_write) {
+               error = mnt_want_write(nd.path.mnt);
+               if (error)
+                       goto exit;
+       }
        error = may_open(&nd, acc_mode, flag);
-       if (error)
+       if (error) {
+               if (will_write)
+                       mnt_drop_write(nd.path.mnt);
                goto exit;
-       return nameidata_to_filp(&nd, open_flag);
+       }
+       filp = nameidata_to_filp(&nd, open_flag);
+       /*
+        * It is now safe to drop the mnt write
+        * because the filp has had a write taken
+        * on its behalf.
+        */
+       if (will_write)
+               mnt_drop_write(nd.path.mnt);
+       return filp;
 
+exit_mutex_unlock:
+       mutex_unlock(&dir->d_inode->i_mutex);
 exit_dput:
        path_put_conditional(&path, &nd);
 exit:
@@ -1973,6 +2033,10 @@ int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
        if (!dir->i_op || !dir->i_op->mknod)
                return -EPERM;
 
+       error = devcgroup_inode_mknod(mode, dev);
+       if (error)
+               return error;
+
        error = security_inode_mknod(dir, dentry, mode, dev);
        if (error)
                return error;
@@ -1984,6 +2048,23 @@ int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
        return error;
 }
 
+static int may_mknod(mode_t mode)
+{
+       switch (mode & S_IFMT) {
+       case S_IFREG:
+       case S_IFCHR:
+       case S_IFBLK:
+       case S_IFIFO:
+       case S_IFSOCK:
+       case 0: /* zero mode translates to S_IFREG */
+               return 0;
+       case S_IFDIR:
+               return -EPERM;
+       default:
+               return -EINVAL;
+       }
+}
+
 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
                                unsigned dev)
 {
@@ -2002,12 +2083,19 @@ asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
        if (error)
                goto out;
        dentry = lookup_create(&nd, 0);
-       error = PTR_ERR(dentry);
-
+       if (IS_ERR(dentry)) {
+               error = PTR_ERR(dentry);
+               goto out_unlock;
+       }
        if (!IS_POSIXACL(nd.path.dentry->d_inode))
                mode &= ~current->fs->umask;
-       if (!IS_ERR(dentry)) {
-               switch (mode & S_IFMT) {
+       error = may_mknod(mode);
+       if (error)
+               goto out_dput;
+       error = mnt_want_write(nd.path.mnt);
+       if (error)
+               goto out_dput;
+       switch (mode & S_IFMT) {
                case 0: case S_IFREG:
                        error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
                        break;
@@ -2018,14 +2106,11 @@ asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
                case S_IFIFO: case S_IFSOCK:
                        error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
                        break;
-               case S_IFDIR:
-                       error = -EPERM;
-                       break;
-               default:
-                       error = -EINVAL;
-               }
-               dput(dentry);
        }
+       mnt_drop_write(nd.path.mnt);
+out_dput:
+       dput(dentry);
+out_unlock:
        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
        path_put(&nd.path);
 out:
@@ -2083,7 +2168,12 @@ asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
 
        if (!IS_POSIXACL(nd.path.dentry->d_inode))
                mode &= ~current->fs->umask;
+       error = mnt_want_write(nd.path.mnt);
+       if (error)
+               goto out_dput;
        error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
+       mnt_drop_write(nd.path.mnt);
+out_dput:
        dput(dentry);
 out_unlock:
        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
@@ -2361,7 +2451,12 @@ asmlinkage long sys_symlinkat(const char __user *oldname,
        if (IS_ERR(dentry))
                goto out_unlock;
 
+       error = mnt_want_write(nd.path.mnt);
+       if (error)
+               goto out_dput;
        error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
+       mnt_drop_write(nd.path.mnt);
+out_dput:
        dput(dentry);
 out_unlock:
        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
@@ -2456,7 +2551,12 @@ asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
        error = PTR_ERR(new_dentry);
        if (IS_ERR(new_dentry))
                goto out_unlock;
+       error = mnt_want_write(nd.path.mnt);
+       if (error)
+               goto out_dput;
        error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
+       mnt_drop_write(nd.path.mnt);
+out_dput:
        dput(new_dentry);
 out_unlock:
        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
@@ -2682,8 +2782,12 @@ static int do_rename(int olddfd, const char *oldname,
        if (new_dentry == trap)
                goto exit5;
 
+       error = mnt_want_write(oldnd.path.mnt);
+       if (error)
+               goto exit5;
        error = vfs_rename(old_dir->d_inode, old_dentry,
                                   new_dir->d_inode, new_dentry);
+       mnt_drop_write(oldnd.path.mnt);
 exit5:
        dput(new_dentry);
 exit4: