ceph: fix client race condition where r_parent becomes stale before sending message

JIRA: https://issues.redhat.com/browse/RHEL-117609

Conflicts: Missing 5995d90 some surrounding APIs are different

commit bec324f
Author: Alex Markuze <amarkuze@redhat.com>
Date:   Tue Aug 12 09:57:39 2025 +0000

    ceph: fix race condition where r_parent becomes stale before sending message

    When the parent directory's i_rwsem is not locked, req->r_parent may become
    stale due to concurrent operations (e.g. rename) between dentry lookup and
    message creation. Validate that r_parent matches the encoded parent inode
    and update to the correct inode if a mismatch is detected.

    [ idryomov: folded a follow-up fix from Alex to drop extra reference
      from ceph_get_reply_dir() in ceph_fill_trace():

      ceph_get_reply_dir() may return a different, referenced inode when
      r_parent is stale and the parent directory lock is not held.
      ceph_fill_trace() used that inode but failed to drop the reference
      when it differed from req->r_parent, leaking an inode reference.

      Keep the directory inode in a local variable and iput() it at
      function end if it does not match req->r_parent. ]

    Cc: stable@vger.kernel.org
    Signed-off-by: Alex Markuze <amarkuze@redhat.com>
    Reviewed-by: Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>
    Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

Signed-off-by: Alex Markuze <amarkuze@redhat.com>

Author: Alex Markuze

fs/ceph/inode.c

@@ -55,6 +55,52 @@ static int ceph_set_ino_cb(struct inode *inode, void *data)
 	return 0;
 }
 
+/*
+ * Check if the parent inode matches the vino from directory reply info
+ */
+static inline bool ceph_vino_matches_parent(struct inode *parent,
+					    struct ceph_vino vino)
+{
+	return ceph_ino(parent) == vino.ino && ceph_snap(parent) == vino.snap;
+}
+
+/*
+ * Validate that the directory inode referenced by @req->r_parent matches the
+ * inode number and snapshot id contained in the reply's directory record.  If
+ * they do not match – which can theoretically happen if the parent dentry was
+ * moved between the time the request was issued and the reply arrived – fall
+ * back to looking up the correct inode in the inode cache.
+ *
+ * A reference is *always* returned.  Callers that receive a different inode
+ * than the original @parent are responsible for dropping the extra reference
+ * once the reply has been processed.
+ */
+static struct inode *ceph_get_reply_dir(struct super_block *sb,
+					struct inode *parent,
+					struct ceph_mds_reply_info_parsed *rinfo)
+{
+	struct ceph_vino vino;
+
+	if (unlikely(!rinfo->diri.in))
+		return parent; /* nothing to compare against */
+
+	/* If we didn't have a cached parent inode to begin with, just bail out. */
+	if (!parent)
+		return NULL;
+
+	vino.ino  = le64_to_cpu(rinfo->diri.in->ino);
+	vino.snap = le64_to_cpu(rinfo->diri.in->snapid);
+
+	if (likely(ceph_vino_matches_parent(parent, vino)))
+		return parent; /* matches – use the original reference */
+
+	/* Mismatch – this should be rare.  Emit a WARN and obtain the correct inode. */
+	WARN_ONCE(1, "ceph: reply dir mismatch (parent valid %llx.%llx reply %llx.%llx)\n",
+		  ceph_ino(parent), ceph_snap(parent), vino.ino, vino.snap);
+
+	return ceph_get_inode(sb, vino, NULL);
+}
+
 /**
  * ceph_new_inode - allocate a new inode in advance of an expected create
  * @dir: parent directory for new inode
@@ -1523,6 +1569,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 	struct ceph_vino tvino, dvino;
 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
 	struct ceph_client *cl = fsc->client;
+	struct inode *parent_dir = NULL;
 	int err = 0;
 
 	doutc(cl, "%p is_dentry %d is_target %d\n", req,
@@ -1536,10 +1583,18 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 	}
 
 	if (rinfo->head->is_dentry) {
-		struct inode *dir = req->r_parent;
+		/*
+		 * r_parent may be stale, in cases when R_PARENT_LOCKED is not set,
+		 * so we need to get the correct inode
+		 */
+		parent_dir = ceph_get_reply_dir(sb, req->r_parent, rinfo);
+		if (unlikely(IS_ERR(parent_dir))) {
+			err = PTR_ERR(parent_dir);
+			goto done;
+		}
 
-		if (dir) {
-			err = ceph_fill_inode(dir, NULL, &rinfo->diri,
+		if (parent_dir) {
+			err = ceph_fill_inode(parent_dir, NULL, &rinfo->diri,
 					      rinfo->dirfrag, session, -1,
 					      &req->r_caps_reservation);
 			if (err < 0)
@@ -1548,14 +1603,14 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 			WARN_ON_ONCE(1);
 		}
 
-		if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
+		if (parent_dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
 		    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
 		    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
 			bool is_nokey = false;
 			struct qstr dname;
 			struct dentry *dn, *parent;
 			struct fscrypt_str oname = FSTR_INIT(NULL, 0);
-			struct ceph_fname fname = { .dir	= dir,
+			struct ceph_fname fname = { .dir	= parent_dir,
 						    .name	= rinfo->dname,
 						    .ctext	= rinfo->altname,
 						    .name_len	= rinfo->dname_len,
@@ -1564,10 +1619,10 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 			BUG_ON(!rinfo->head->is_target);
 			BUG_ON(req->r_dentry);
 
-			parent = d_find_any_alias(dir);
+			parent = d_find_any_alias(parent_dir);
 			BUG_ON(!parent);
 
-			err = ceph_fname_alloc_buffer(dir, &oname);
+			err = ceph_fname_alloc_buffer(parent_dir, &oname);
 			if (err < 0) {
 				dput(parent);
 				goto done;
@@ -1576,14 +1631,15 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 			err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
 			if (err < 0) {
 				dput(parent);
-				ceph_fname_free_buffer(dir, &oname);
+				ceph_fname_free_buffer(parent_dir, &oname);
 				goto done;
 			}
 			dname.name = oname.name;
 			dname.len = oname.len;
 			dname.hash = full_name_hash(parent, dname.name, dname.len);
 			tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
 			tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
+
 retry_lookup:
 			dn = d_lookup(parent, &dname);
 			doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
@@ -1595,7 +1651,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 				      dname.len, dname.name, dn);
 				if (!dn) {
 					dput(parent);
-					ceph_fname_free_buffer(dir, &oname);
+					ceph_fname_free_buffer(parent_dir, &oname);
 					err = -ENOMEM;
 					goto done;
 				}
@@ -1610,12 +1666,12 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 				    ceph_snap(d_inode(dn)) != tvino.snap)) {
 				doutc(cl, " dn %p points to wrong inode %p\n",
 				      dn, d_inode(dn));
-				ceph_dir_clear_ordered(dir);
+				ceph_dir_clear_ordered(parent_dir);
 				d_delete(dn);
 				dput(dn);
 				goto retry_lookup;
 			}
-			ceph_fname_free_buffer(dir, &oname);
+			ceph_fname_free_buffer(parent_dir, &oname);
 
 			req->r_dentry = dn;
 			dput(parent);
@@ -1794,6 +1850,9 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
 					    &dvino, ptvino);
 	}
 done:
+	/* Drop extra ref from ceph_get_reply_dir() if it returned a new inode */
+	if (unlikely(!IS_ERR_OR_NULL(parent_dir) && parent_dir != req->r_parent))
+		iput(parent_dir);
 	doutc(cl, "done err=%d\n", err);
 	return err;
 }
@@ -2483,22 +2542,21 @@ int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
 	int truncate_retry = 20; /* The RMW will take around 50ms */
 	struct dentry *dentry;
 	char *path;
-	int pathlen;
-	u64 pathbase;
 	bool do_sync = false;
 
 	dentry = d_find_alias(inode);
 	if (!dentry) {
 		do_sync = true;
 	} else {
-		path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase, 0);
+		struct ceph_path_info path_info;
+		path = ceph_mdsc_build_path(mdsc, dentry, &path_info, 0);
 		if (IS_ERR(path)) {
 			do_sync = true;
 			err = 0;
 		} else {
 			err = ceph_mds_check_access(mdsc, path, MAY_WRITE);
 		}
-		ceph_mdsc_free_path(path, pathlen);
+		ceph_mdsc_free_path_info(&path_info);
 		dput(dentry);
 
 		/* For none EACCES cases will let the MDS do the mds auth check */