Merge tag 'nfsd-6.9' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

Pull nfsd updates from Chuck Lever:
 "The bulk of the patches for this release are optimizations, code
  clean-ups, and minor bug fixes.

  One new feature to mention is that NFSD administrators now have the
  ability to revoke NFSv4 open and lock state. NFSD's NFSv3 support has
  had this capability for some time.

  As always I am grateful to NFSD contributors, reviewers, and testers"

* tag 'nfsd-6.9' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux: (75 commits)
  NFSD: Clean up nfsd4_encode_replay()
  NFSD: send OP_CB_RECALL_ANY to clients when number of delegations reaches its limit
  NFSD: Document nfsd_setattr() fill-attributes behavior
  nfsd: Fix NFSv3 atomicity bugs in nfsd_setattr()
  nfsd: Fix a regression in nfsd_setattr()
  NFSD: OP_CB_RECALL_ANY should recall both read and write delegations
  NFSD: handle GETATTR conflict with write delegation
  NFSD: add support for CB_GETATTR callback
  NFSD: Document the phases of CREATE_SESSION
  NFSD: Fix the NFSv4.1 CREATE_SESSION operation
  nfsd: clean up comments over nfs4_client definition
  svcrdma: Add Write chunk WRs to the RPC's Send WR chain
  svcrdma: Post WRs for Write chunks in svc_rdma_sendto()
  svcrdma: Post the Reply chunk and Send WR together
  svcrdma: Move write_info for Reply chunks into struct svc_rdma_send_ctxt
  svcrdma: Post Send WR chain
  svcrdma: Fix retry loop in svc_rdma_send()
  svcrdma: Prevent a UAF in svc_rdma_send()
  svcrdma: Fix SQ wake-ups
  svcrdma: Increase the per-transport rw_ctx count
  ...

1 files changed, 68 insertions, 16 deletions

diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index b7c7a9273ea0..6a4c506038e0 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -476,7 +476,6 @@ static int __nfsd_setattr(struct dentry *dentry, struct iattr *iap)
  * @rqstp: controlling RPC transaction
  * @fhp: filehandle of target
  * @attr: attributes to set
- * @check_guard: set to 1 if guardtime is a valid timestamp
  * @guardtime: do not act if ctime.tv_sec does not match this timestamp
  *
  * This call may adjust the contents of @attr (in particular, this
@@ -488,8 +487,7 @@ static int __nfsd_setattr(struct dentry *dentry, struct iattr *iap)
  */
 __be32
 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
-	     struct nfsd_attrs *attr,
-	     int check_guard, time64_t guardtime)
+	     struct nfsd_attrs *attr, const struct timespec64 *guardtime)
 {
 	struct dentry	*dentry;
 	struct inode	*inode;
@@ -497,7 +495,7 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
 	int		accmode = NFSD_MAY_SATTR;
 	umode_t		ftype = 0;
 	__be32		err;
-	int		host_err;
+	int		host_err = 0;
 	bool		get_write_count;
 	bool		size_change = (iap->ia_valid & ATTR_SIZE);
 	int		retries;
@@ -538,9 +536,6 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
 
 	nfsd_sanitize_attrs(inode, iap);
 
-	if (check_guard && guardtime != inode_get_ctime_sec(inode))
-		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
@@ -555,6 +550,19 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
 	}
 
 	inode_lock(inode);
+	err = fh_fill_pre_attrs(fhp);
+	if (err)
+		goto out_unlock;
+
+	if (guardtime) {
+		struct timespec64 ctime = inode_get_ctime(inode);
+		if ((u32)guardtime->tv_sec != (u32)ctime.tv_sec ||
+		    guardtime->tv_nsec != ctime.tv_nsec) {
+			err = nfserr_notsync;
+			goto out_fill_attrs;
+		}
+	}
+
 	for (retries = 1;;) {
 		struct iattr attrs;
 
@@ -582,13 +590,23 @@ nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
 		attr->na_aclerr = set_posix_acl(&nop_mnt_idmap,
 						dentry, ACL_TYPE_DEFAULT,
 						attr->na_dpacl);
+out_fill_attrs:
+	/*
+	 * RFC 1813 Section 3.3.2 does not mandate that an NFS server
+	 * returns wcc_data for SETATTR. Some client implementations
+	 * depend on receiving wcc_data, however, to sort out partial
+	 * updates (eg., the client requested that size and mode be
+	 * modified, but the server changed only the file mode).
+	 */
+	fh_fill_post_attrs(fhp);
+out_unlock:
 	inode_unlock(inode);
 	if (size_change)
 		put_write_access(inode);
 out:
 	if (!host_err)
 		host_err = commit_metadata(fhp);
-	return nfserrno(host_err);
+	return err != 0 ? err : nfserrno(host_err);
 }
 
 #if defined(CONFIG_NFSD_V4)
@@ -1002,7 +1020,9 @@ static __be32 nfsd_finish_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
 			       unsigned long *count, u32 *eof, ssize_t host_err)
 {
 	if (host_err >= 0) {
-		nfsd_stats_io_read_add(fhp->fh_export, host_err);
+		struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
+
+		nfsd_stats_io_read_add(nn, fhp->fh_export, host_err);
 		*eof = nfsd_eof_on_read(file, offset, host_err, *count);
 		*count = host_err;
 		fsnotify_access(file);
@@ -1185,7 +1205,7 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfsd_file *nf,
 		goto out_nfserr;
 	}
 	*cnt = host_err;
-	nfsd_stats_io_write_add(exp, *cnt);
+	nfsd_stats_io_write_add(nn, exp, *cnt);
 	fsnotify_modify(file);
 	host_err = filemap_check_wb_err(file->f_mapping, since);
 	if (host_err < 0)
@@ -1404,7 +1424,7 @@ nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp,
 	 * if the attributes have not changed.
 	 */
 	if (iap->ia_valid)
-		status = nfsd_setattr(rqstp, resfhp, attrs, 0, (time64_t)0);
+		status = nfsd_setattr(rqstp, resfhp, attrs, NULL);
 	else
 		status = nfserrno(commit_metadata(resfhp));
 
@@ -1906,10 +1926,10 @@ out_unlock:
 	fh_drop_write(ffhp);
 
 	/*
-	 * If the target dentry has cached open files, then we need to try to
-	 * close them prior to doing the rename. Flushing delayed fput
-	 * shouldn't be done with locks held however, so we delay it until this
-	 * point and then reattempt the whole shebang.
+	 * If the target dentry has cached open files, then we need to
+	 * try to close them prior to doing the rename.  Final fput
+	 * shouldn't be done with locks held however, so we delay it
+	 * until this point and then reattempt the whole shebang.
 	 */
 	if (close_cached) {
 		close_cached = false;
@@ -2177,11 +2197,43 @@ nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
 	if (err == nfserr_eof || err == nfserr_toosmall)
 		err = nfs_ok; /* can still be found in ->err */
 out_close:
-	fput(file);
+	nfsd_filp_close(file);
 out:
 	return err;
 }
 
+/**
+ * nfsd_filp_close: close a file synchronously
+ * @fp: the file to close
+ *
+ * nfsd_filp_close() is similar in behaviour to filp_close().
+ * The difference is that if this is the final close on the
+ * file, the that finalisation happens immediately, rather then
+ * being handed over to a work_queue, as it the case for
+ * filp_close().
+ * When a user-space process closes a file (even when using
+ * filp_close() the finalisation happens before returning to
+ * userspace, so it is effectively synchronous.  When a kernel thread
+ * uses file_close(), on the other hand, the handling is completely
+ * asynchronous.  This means that any cost imposed by that finalisation
+ * is not imposed on the nfsd thread, and nfsd could potentually
+ * close files more quickly than the work queue finalises the close,
+ * which would lead to unbounded growth in the queue.
+ *
+ * In some contexts is it not safe to synchronously wait for
+ * close finalisation (see comment for __fput_sync()), but nfsd
+ * does not match those contexts.  In partcilarly it does not, at the
+ * time that this function is called, hold and locks and no finalisation
+ * of any file, socket, or device driver would have any cause to wait
+ * for nfsd to make progress.
+ */
+void nfsd_filp_close(struct file *fp)
+{
+	get_file(fp);
+	filp_close(fp, NULL);
+	__fput_sync(fp);
+}
+
 /*
  * Get file system stats
  * N.B. After this call fhp needs an fh_put