Merge tag 'ceph-for-5.2-rc4' of git://github.com/ceph/ceph-client

Pull ceph fixes from Ilya Dryomov:
 "A change to call iput() asynchronously to avoid a possible deadlock
  when iput_final() needs to wait for in-flight I/O (e.g. readahead) and
  a fixup for a cleanup that went into -rc1"

* tag 'ceph-for-5.2-rc4' of git://github.com/ceph/ceph-client:
  ceph: fix error handling in ceph_get_caps()
  ceph: avoid iput_final() while holding mutex or in dispatch thread
  ceph: single workqueue for inode related works

Author: torvalds

fs/ceph/caps.c

@@ -2738,15 +2738,13 @@ int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
 		_got = 0;
 		ret = try_get_cap_refs(ci, need, want, endoff,
 				       false, &_got);
-		if (ret == -EAGAIN) {
+		if (ret == -EAGAIN)
 			continue;
-		} else if (!ret) {
-			int err;
-
+		if (!ret) {
 			DEFINE_WAIT_FUNC(wait, woken_wake_function);
 			add_wait_queue(&ci->i_cap_wq, &wait);
 
-			while (!(err = try_get_cap_refs(ci, need, want, endoff,
+			while (!(ret = try_get_cap_refs(ci, need, want, endoff,
 							true, &_got))) {
 				if (signal_pending(current)) {
 					ret = -ERESTARTSYS;
@@ -2756,14 +2754,16 @@ int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
 			}
 
 			remove_wait_queue(&ci->i_cap_wq, &wait);
-			if (err == -EAGAIN)
+			if (ret == -EAGAIN)
 				continue;
 		}
-		if (ret == -ESTALE) {
-			/* session was killed, try renew caps */
-			ret = ceph_renew_caps(&ci->vfs_inode);
-			if (ret == 0)
-				continue;
+		if (ret < 0) {
+			if (ret == -ESTALE) {
+				/* session was killed, try renew caps */
+				ret = ceph_renew_caps(&ci->vfs_inode);
+				if (ret == 0)
+					continue;
+			}
 			return ret;
 		}
 
@@ -2992,8 +2992,10 @@ void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
 	}
 	if (complete_capsnap)
 		wake_up_all(&ci->i_cap_wq);
-	while (put-- > 0)
-		iput(inode);
+	while (put-- > 0) {
+		/* avoid calling iput_final() in osd dispatch threads */
+		ceph_async_iput(inode);
+	}
 }
 
 /*
@@ -3964,8 +3966,9 @@ void ceph_handle_caps(struct ceph_mds_session *session,
 done:
 	mutex_unlock(&session->s_mutex);
 done_unlocked:
-	iput(inode);
 	ceph_put_string(extra_info.pool_ns);
+	/* avoid calling iput_final() in mds dispatch threads */
+	ceph_async_iput(inode);
 	return;
 
 flush_cap_releases:
@@ -4011,7 +4014,8 @@ void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
 		if (inode) {
 			dout("check_delayed_caps on %p\n", inode);
 			ceph_check_caps(ci, flags, NULL);
-			iput(inode);
+			/* avoid calling iput_final() in tick thread */
+			ceph_async_iput(inode);
 		}
 	}
 	spin_unlock(&mdsc->cap_delay_lock);

fs/ceph/file.c

@@ -791,7 +791,7 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
 		if (aio_work) {
 			INIT_WORK(&aio_work->work, ceph_aio_retry_work);
 			aio_work->req = req;
-			queue_work(ceph_inode_to_client(inode)->wb_wq,
+			queue_work(ceph_inode_to_client(inode)->inode_wq,
 				   &aio_work->work);
 			return;
 		}

fs/ceph/inode.c

@@ -33,9 +33,7 @@
 
 static const struct inode_operations ceph_symlink_iops;
 
-static void ceph_invalidate_work(struct work_struct *work);
-static void ceph_writeback_work(struct work_struct *work);
-static void ceph_vmtruncate_work(struct work_struct *work);
+static void ceph_inode_work(struct work_struct *work);
 
 /*
  * find or create an inode, given the ceph ino number
@@ -509,10 +507,8 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	INIT_LIST_HEAD(&ci->i_snap_realm_item);
 	INIT_LIST_HEAD(&ci->i_snap_flush_item);
 
-	INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
-	INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
-
-	INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
+	INIT_WORK(&ci->i_work, ceph_inode_work);
+	ci->i_work_mask = 0;
 
 	ceph_fscache_inode_init(ci);
 
@@ -1480,7 +1476,8 @@ static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
 			pr_err("fill_inode badness on %p got %d\n", in, rc);
 			err = rc;
 		}
-		iput(in);
+		/* avoid calling iput_final() in mds dispatch threads */
+		ceph_async_iput(in);
 	}
 
 	return err;
@@ -1678,8 +1675,11 @@ int ceph_readdir_prepopulate(struct ceph_mds_request *req,
 				 &req->r_caps_reservation);
 		if (ret < 0) {
 			pr_err("fill_inode badness on %p\n", in);
-			if (d_really_is_negative(dn))
-				iput(in);
+			if (d_really_is_negative(dn)) {
+				/* avoid calling iput_final() in mds
+				 * dispatch threads */
+				ceph_async_iput(in);
+			}
 			d_drop(dn);
 			err = ret;
 			goto next_item;
@@ -1689,7 +1689,7 @@ int ceph_readdir_prepopulate(struct ceph_mds_request *req,
 			if (ceph_security_xattr_deadlock(in)) {
 				dout(" skip splicing dn %p to inode %p"
 				     " (security xattr deadlock)\n", dn, in);
-				iput(in);
+				ceph_async_iput(in);
 				skipped++;
 				goto next_item;
 			}
@@ -1740,57 +1740,87 @@ bool ceph_inode_set_size(struct inode *inode, loff_t size)
 	return ret;
 }
 
+/*
+ * Put reference to inode, but avoid calling iput_final() in current thread.
+ * iput_final() may wait for reahahead pages. The wait can cause deadlock in
+ * some contexts.
+ */
+void ceph_async_iput(struct inode *inode)
+{
+	if (!inode)
+		return;
+	for (;;) {
+		if (atomic_add_unless(&inode->i_count, -1, 1))
+			break;
+		if (queue_work(ceph_inode_to_client(inode)->inode_wq,
+			       &ceph_inode(inode)->i_work))
+			break;
+		/* queue work failed, i_count must be at least 2 */
+	}
+}
+
 /*
  * Write back inode data in a worker thread.  (This can't be done
  * in the message handler context.)
  */
 void ceph_queue_writeback(struct inode *inode)
 {
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
+
 	ihold(inode);
-	if (queue_work(ceph_inode_to_client(inode)->wb_wq,
-		       &ceph_inode(inode)->i_wb_work)) {
+	if (queue_work(ceph_inode_to_client(inode)->inode_wq,
+		       &ci->i_work)) {
 		dout("ceph_queue_writeback %p\n", inode);
 	} else {
-		dout("ceph_queue_writeback %p failed\n", inode);
+		dout("ceph_queue_writeback %p already queued, mask=%lx\n",
+		     inode, ci->i_work_mask);
 		iput(inode);
 	}
 }
 
-static void ceph_writeback_work(struct work_struct *work)
-{
-	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
-						  i_wb_work);
-	struct inode *inode = &ci->vfs_inode;
-
-	dout("writeback %p\n", inode);
-	filemap_fdatawrite(&inode->i_data);
-	iput(inode);
-}
-
 /*
  * queue an async invalidation
  */
 void ceph_queue_invalidate(struct inode *inode)
 {
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
+
 	ihold(inode);
-	if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
-		       &ceph_inode(inode)->i_pg_inv_work)) {
+	if (queue_work(ceph_inode_to_client(inode)->inode_wq,
+		       &ceph_inode(inode)->i_work)) {
 		dout("ceph_queue_invalidate %p\n", inode);
 	} else {
-		dout("ceph_queue_invalidate %p failed\n", inode);
+		dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
+		     inode, ci->i_work_mask);
 		iput(inode);
 	}
 }
 
 /*
- * Invalidate inode pages in a worker thread.  (This can't be done
- * in the message handler context.)
+ * Queue an async vmtruncate.  If we fail to queue work, we will handle
+ * the truncation the next time we call __ceph_do_pending_vmtruncate.
  */
-static void ceph_invalidate_work(struct work_struct *work)
+void ceph_queue_vmtruncate(struct inode *inode)
 {
-	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
-						  i_pg_inv_work);
-	struct inode *inode = &ci->vfs_inode;
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
+
+	ihold(inode);
+	if (queue_work(ceph_inode_to_client(inode)->inode_wq,
+		       &ci->i_work)) {
+		dout("ceph_queue_vmtruncate %p\n", inode);
+	} else {
+		dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
+		     inode, ci->i_work_mask);
+		iput(inode);
+	}
+}
+
+static void ceph_do_invalidate_pages(struct inode *inode)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 	u32 orig_gen;
 	int check = 0;
@@ -1842,44 +1872,6 @@ static void ceph_invalidate_work(struct work_struct *work)
 out:
 	if (check)
 		ceph_check_caps(ci, 0, NULL);
-	iput(inode);
-}
-
-
-/*
- * called by trunc_wq;
- *
- * We also truncate in a separate thread as well.
- */
-static void ceph_vmtruncate_work(struct work_struct *work)
-{
-	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
-						  i_vmtruncate_work);
-	struct inode *inode = &ci->vfs_inode;
-
-	dout("vmtruncate_work %p\n", inode);
-	__ceph_do_pending_vmtruncate(inode);
-	iput(inode);
-}
-
-/*
- * Queue an async vmtruncate.  If we fail to queue work, we will handle
- * the truncation the next time we call __ceph_do_pending_vmtruncate.
- */
-void ceph_queue_vmtruncate(struct inode *inode)
-{
-	struct ceph_inode_info *ci = ceph_inode(inode);
-
-	ihold(inode);
-
-	if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
-		       &ci->i_vmtruncate_work)) {
-		dout("ceph_queue_vmtruncate %p\n", inode);
-	} else {
-		dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
-		     inode, ci->i_truncate_pending);
-		iput(inode);
-	}
 }
 
 /*
@@ -1943,6 +1935,25 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
 	wake_up_all(&ci->i_cap_wq);
 }
 
+static void ceph_inode_work(struct work_struct *work)
+{
+	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
+						 i_work);
+	struct inode *inode = &ci->vfs_inode;
+
+	if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
+		dout("writeback %p\n", inode);
+		filemap_fdatawrite(&inode->i_data);
+	}
+	if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
+		ceph_do_invalidate_pages(inode);
+
+	if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
+		__ceph_do_pending_vmtruncate(inode);
+
+	iput(inode);
+}
+
 /*
  * symlinks
  */