nfsd: Revert "nfsd: special case truncates some more"
authorJ. Bruce Fields <bfields@redhat.com>
Thu, 9 Feb 2017 19:20:42 +0000 (14:20 -0500)
committerJ. Bruce Fields <bfields@redhat.com>
Fri, 10 Feb 2017 01:49:20 +0000 (20:49 -0500)
This patch incorrectly attempted nested mnt_want_write, and incorrectly
disabled nfsd's owner override for truncate.  We'll fix those problems
and make another attempt soon, for the moment I think the safest is to
revert.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/vfs.c

index ca13236dbb1f33afb1c6664c030d63a5bfcfacc5..26c6fdb4bf67cf1e3e3a843e8e816d7a76eae265 100644 (file)
@@ -332,6 +332,37 @@ nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
        }
 }
 
+static __be32
+nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
+               struct iattr *iap)
+{
+       struct inode *inode = d_inode(fhp->fh_dentry);
+       int host_err;
+
+       if (iap->ia_size < inode->i_size) {
+               __be32 err;
+
+               err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
+                               NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
+               if (err)
+                       return err;
+       }
+
+       host_err = get_write_access(inode);
+       if (host_err)
+               goto out_nfserrno;
+
+       host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
+       if (host_err)
+               goto out_put_write_access;
+       return 0;
+
+out_put_write_access:
+       put_write_access(inode);
+out_nfserrno:
+       return nfserrno(host_err);
+}
+
 /*
  * Set various file attributes.  After this call fhp needs an fh_put.
  */
@@ -346,6 +377,7 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
        __be32          err;
        int             host_err;
        bool            get_write_count;
+       int             size_change = 0;
 
        if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
                accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
@@ -358,11 +390,11 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
        /* Get inode */
        err = fh_verify(rqstp, fhp, ftype, accmode);
        if (err)
-               return err;
+               goto out;
        if (get_write_count) {
                host_err = fh_want_write(fhp);
                if (host_err)
-                       goto out_host_err;
+                       return nfserrno(host_err);
        }
 
        dentry = fhp->fh_dentry;
@@ -373,59 +405,50 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
                iap->ia_valid &= ~ATTR_MODE;
 
        if (!iap->ia_valid)
-               return 0;
+               goto out;
 
        nfsd_sanitize_attrs(inode, iap);
 
-       if (check_guard && guardtime != inode->i_ctime.tv_sec)
-               return nfserr_notsync;
-
        /*
         * The size case is special, it changes the file in addition to the
-        * attributes, and file systems don't expect it to be mixed with
-        * "random" attribute changes.  We thus split out the size change
-        * into a separate call for vfs_truncate, and do the rest as a
-        * a separate setattr call.
+        * attributes.
         */
        if (iap->ia_valid & ATTR_SIZE) {
-               struct path path = {
-                       .mnt    = fhp->fh_export->ex_path.mnt,
-                       .dentry = dentry,
-               };
-               bool implicit_mtime = false;
+               err = nfsd_get_write_access(rqstp, fhp, iap);
+               if (err)
+                       goto out;
+               size_change = 1;
 
                /*
-                * vfs_truncate implicity updates the mtime IFF the file size
-                * actually changes.  Avoid the additional seattr call below if
-                * the only other attribute that the client sends is the mtime.
+                * RFC5661, Section 18.30.4:
+                *   Changing the size of a file with SETATTR indirectly
+                *   changes the time_modify and change attributes.
+                *
+                * (and similar for the older RFCs)
                 */
-               if (iap->ia_size != i_size_read(inode) &&
-                   ((iap->ia_valid & ~(ATTR_SIZE | ATTR_MTIME)) == 0))
-                       implicit_mtime = true;
-
-               host_err = vfs_truncate(&path, iap->ia_size);
-               if (host_err)
-                       goto out_host_err;
-
-               iap->ia_valid &= ~ATTR_SIZE;
-               if (implicit_mtime)
-                       iap->ia_valid &= ~ATTR_MTIME;
-               if (!iap->ia_valid)
-                       goto done;
+               if (iap->ia_size != i_size_read(inode))
+                       iap->ia_valid |= ATTR_MTIME;
        }
 
        iap->ia_valid |= ATTR_CTIME;
 
+       if (check_guard && guardtime != inode->i_ctime.tv_sec) {
+               err = nfserr_notsync;
+               goto out_put_write_access;
+       }
+
        fh_lock(fhp);
        host_err = notify_change(dentry, iap, NULL);
        fh_unlock(fhp);
-       if (host_err)
-               goto out_host_err;
+       err = nfserrno(host_err);
 
-done:
-       host_err = commit_metadata(fhp);
-out_host_err:
-       return nfserrno(host_err);
+out_put_write_access:
+       if (size_change)
+               put_write_access(inode);
+       if (!err)
+               err = nfserrno(commit_metadata(fhp));
+out:
+       return err;
 }
 
 #if defined(CONFIG_NFSD_V4)