Merge tag 'ceph-for-6.8-rc4' of https://github.com/ceph/ceph-client

Pull ceph fixes from Ilya Dryomov:
 "Some fscrypt-related fixups (sparse reads are used only for encrypted
  files) and two cap handling fixes from Xiubo and Rishabh"

* tag 'ceph-for-6.8-rc4' of https://github.com/ceph/ceph-client:
  ceph: always check dir caps asynchronously
  ceph: prevent use-after-free in encode_cap_msg()
  ceph: always set initial i_blkbits to CEPH_FSCRYPT_BLOCK_SHIFT
  libceph: just wait for more data to be available on the socket
  libceph: rename read_sparse_msg_*() to read_partial_sparse_msg_*()
  libceph: fail sparse-read if the data length doesn't match

fs/ceph/caps.c

@@ -1452,7 +1452,7 @@ static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
 	if (flushing & CEPH_CAP_XATTR_EXCL) {
 		arg->old_xattr_buf = __ceph_build_xattrs_blob(ci);
 		arg->xattr_version = ci->i_xattrs.version;
-		arg->xattr_buf = ci->i_xattrs.blob;
+		arg->xattr_buf = ceph_buffer_get(ci->i_xattrs.blob);
 	} else {
 		arg->xattr_buf = NULL;
 		arg->old_xattr_buf = NULL;
@@ -1553,6 +1553,7 @@ static void __send_cap(struct cap_msg_args *arg, struct ceph_inode_info *ci)
 	encode_cap_msg(msg, arg);
 	ceph_con_send(&arg->session->s_con, msg);
 	ceph_buffer_put(arg->old_xattr_buf);
+	ceph_buffer_put(arg->xattr_buf);
 	if (arg->wake)
 		wake_up_all(&ci->i_cap_wq);
 }
@@ -3215,7 +3216,6 @@ static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci,
 
 enum put_cap_refs_mode {
 	PUT_CAP_REFS_SYNC = 0,
-	PUT_CAP_REFS_NO_CHECK,
 	PUT_CAP_REFS_ASYNC,
 };
 
@@ -3331,11 +3331,6 @@ void ceph_put_cap_refs_async(struct ceph_inode_info *ci, int had)
 	__ceph_put_cap_refs(ci, had, PUT_CAP_REFS_ASYNC);
 }
 
-void ceph_put_cap_refs_no_check_caps(struct ceph_inode_info *ci, int had)
-{
-	__ceph_put_cap_refs(ci, had, PUT_CAP_REFS_NO_CHECK);
-}
-
 /*
  * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
  * context.  Adjust per-snap dirty page accounting as appropriate.

fs/ceph/inode.c

@@ -78,6 +78,8 @@ struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
 	if (!inode)
 		return ERR_PTR(-ENOMEM);
 
+	inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
+
 	if (!S_ISLNK(*mode)) {
 		err = ceph_pre_init_acls(dir, mode, as_ctx);
 		if (err < 0)

fs/ceph/mds_client.c

@@ -1089,7 +1089,7 @@ void ceph_mdsc_release_request(struct kref *kref)
 	struct ceph_mds_request *req = container_of(kref,
 						    struct ceph_mds_request,
 						    r_kref);
-	ceph_mdsc_release_dir_caps_no_check(req);
+	ceph_mdsc_release_dir_caps_async(req);
 	destroy_reply_info(&req->r_reply_info);
 	if (req->r_request)
 		ceph_msg_put(req->r_request);
@@ -4261,7 +4261,7 @@ void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
 	}
 }
 
-void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
+void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req)
 {
 	struct ceph_client *cl = req->r_mdsc->fsc->client;
 	int dcaps;
@@ -4269,8 +4269,7 @@ void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
 	dcaps = xchg(&req->r_dir_caps, 0);
 	if (dcaps) {
 		doutc(cl, "releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
-		ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
-						dcaps);
+		ceph_put_cap_refs_async(ceph_inode(req->r_parent), dcaps);
 	}
 }
 
@@ -4306,7 +4305,7 @@ static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
 		if (req->r_session->s_mds != session->s_mds)
 			continue;
 
-		ceph_mdsc_release_dir_caps_no_check(req);
+		ceph_mdsc_release_dir_caps_async(req);
 
 		__send_request(session, req, true);
 	}

fs/ceph/mds_client.h

@@ -552,7 +552,7 @@ extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 				struct inode *dir,
 				struct ceph_mds_request *req);
 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
-extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
+extern void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req);
 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 {
 	kref_get(&req->r_kref);

fs/ceph/super.h

@@ -1255,8 +1255,6 @@ extern void ceph_take_cap_refs(struct ceph_inode_info *ci, int caps,
 extern void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps);
 extern void ceph_put_cap_refs(struct ceph_inode_info *ci, int had);
 extern void ceph_put_cap_refs_async(struct ceph_inode_info *ci, int had);
-extern void ceph_put_cap_refs_no_check_caps(struct ceph_inode_info *ci,
-					    int had);
 extern void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
 				       struct ceph_snap_context *snapc);
 extern void __ceph_remove_capsnap(struct inode *inode,

include/linux/ceph/messenger.h

@@ -283,7 +283,7 @@ struct ceph_msg {
 	struct kref kref;
 	bool more_to_follow;
 	bool needs_out_seq;
-	bool sparse_read;
+	u64 sparse_read_total;
 	int front_alloc_len;
 
 	struct ceph_msgpool *pool;

include/linux/ceph/osd_client.h

@@ -45,6 +45,7 @@ enum ceph_sparse_read_state {
 	CEPH_SPARSE_READ_HDR	= 0,
 	CEPH_SPARSE_READ_EXTENTS,
 	CEPH_SPARSE_READ_DATA_LEN,
+	CEPH_SPARSE_READ_DATA_PRE,
 	CEPH_SPARSE_READ_DATA,
 };
 
@@ -64,7 +65,7 @@ struct ceph_sparse_read {
 	u64				sr_req_len;  /* orig request length */
 	u64				sr_pos;      /* current pos in buffer */
 	int				sr_index;    /* current extent index */
-	__le32				sr_datalen;  /* length of actual data */
+	u32				sr_datalen;  /* length of actual data */
 	u32				sr_count;    /* extent count in reply */
 	int				sr_ext_len;  /* length of extent array */
 	struct ceph_sparse_extent	*sr_extent;  /* extent array */

net/ceph/messenger_v1.c

@@ -160,8 +160,9 @@ static size_t sizeof_footer(struct ceph_connection *con)
 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
 {
 	/* Initialize data cursor if it's not a sparse read */
-	if (!msg->sparse_read)
-		ceph_msg_data_cursor_init(&msg->cursor, msg, data_len);
+	u64 len = msg->sparse_read_total ? : data_len;
+
+	ceph_msg_data_cursor_init(&msg->cursor, msg, len);
 }
 
 /*
@@ -991,7 +992,7 @@ static inline int read_partial_message_section(struct ceph_connection *con,
 	return read_partial_message_chunk(con, section, sec_len, crc);
 }
 
-static int read_sparse_msg_extent(struct ceph_connection *con, u32 *crc)
+static int read_partial_sparse_msg_extent(struct ceph_connection *con, u32 *crc)
 {
 	struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
 	bool do_bounce = ceph_test_opt(from_msgr(con->msgr), RXBOUNCE);
@@ -1026,7 +1027,7 @@ static int read_sparse_msg_extent(struct ceph_connection *con, u32 *crc)
 	return 1;
 }
 
-static int read_sparse_msg_data(struct ceph_connection *con)
+static int read_partial_sparse_msg_data(struct ceph_connection *con)
 {
 	struct ceph_msg_data_cursor *cursor = &con->in_msg->cursor;
 	bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
@@ -1036,31 +1037,31 @@ static int read_sparse_msg_data(struct ceph_connection *con)
 	if (do_datacrc)
 		crc = con->in_data_crc;
 
-	do {
+	while (cursor->total_resid) {
 		if (con->v1.in_sr_kvec.iov_base)
 			ret = read_partial_message_chunk(con,
 							 &con->v1.in_sr_kvec,
 							 con->v1.in_sr_len,
 							 &crc);
 		else if (cursor->sr_resid > 0)
-			ret = read_sparse_msg_extent(con, &crc);
-
-		if (ret <= 0) {
-			if (do_datacrc)
-				con->in_data_crc = crc;
-			return ret;
-		}
+			ret = read_partial_sparse_msg_extent(con, &crc);
+		if (ret <= 0)
+			break;
 
 		memset(&con->v1.in_sr_kvec, 0, sizeof(con->v1.in_sr_kvec));
 		ret = con->ops->sparse_read(con, cursor,
 				(char **)&con->v1.in_sr_kvec.iov_base);
+		if (ret <= 0) {
+			ret = ret ? ret : 1;  /* must return > 0 to indicate success */
+			break;
+		}
 		con->v1.in_sr_len = ret;
-	} while (ret > 0);
+	}
 
 	if (do_datacrc)
 		con->in_data_crc = crc;
 
-	return ret < 0 ? ret : 1;  /* must return > 0 to indicate success */
+	return ret;
 }
 
 static int read_partial_msg_data(struct ceph_connection *con)
@@ -1253,8 +1254,8 @@ static int read_partial_message(struct ceph_connection *con)
 		if (!m->num_data_items)
 			return -EIO;
 
-		if (m->sparse_read)
-			ret = read_sparse_msg_data(con);
+		if (m->sparse_read_total)
+			ret = read_partial_sparse_msg_data(con);
 		else if (ceph_test_opt(from_msgr(con->msgr), RXBOUNCE))
 			ret = read_partial_msg_data_bounce(con);
 		else

net/ceph/messenger_v2.c

@@ -1128,7 +1128,7 @@ static int decrypt_tail(struct ceph_connection *con)
 	struct sg_table enc_sgt = {};
 	struct sg_table sgt = {};
 	struct page **pages = NULL;
-	bool sparse = con->in_msg->sparse_read;
+	bool sparse = !!con->in_msg->sparse_read_total;
 	int dpos = 0;
 	int tail_len;
 	int ret;
@@ -2060,7 +2060,7 @@ static int prepare_read_tail_plain(struct ceph_connection *con)
 	}
 
 	if (data_len(msg)) {
-		if (msg->sparse_read)
+		if (msg->sparse_read_total)
 			con->v2.in_state = IN_S_PREPARE_SPARSE_DATA;
 		else
 			con->v2.in_state = IN_S_PREPARE_READ_DATA;

net/ceph/osd_client.c

@@ -5510,7 +5510,7 @@ static struct ceph_msg *get_reply(struct ceph_connection *con,
 	}
 
 	m = ceph_msg_get(req->r_reply);
-	m->sparse_read = (bool)srlen;
+	m->sparse_read_total = srlen;
 
 	dout("get_reply tid %lld %p\n", tid, m);
 
@@ -5777,11 +5777,8 @@ static int prep_next_sparse_read(struct ceph_connection *con,
 	}
 
 	if (o->o_sparse_op_idx < 0) {
-		u64 srlen = sparse_data_requested(req);
-
-		dout("%s: [%d] starting new sparse read req. srlen=0x%llx\n",
-		     __func__, o->o_osd, srlen);
-		ceph_msg_data_cursor_init(cursor, con->in_msg, srlen);
+		dout("%s: [%d] starting new sparse read req\n",
+		     __func__, o->o_osd);
 	} else {
 		u64 end;
 
@@ -5857,8 +5854,8 @@ static int osd_sparse_read(struct ceph_connection *con,
 	struct ceph_osd *o = con->private;
 	struct ceph_sparse_read *sr = &o->o_sparse_read;
 	u32 count = sr->sr_count;
-	u64 eoff, elen;
-	int ret;
+	u64 eoff, elen, len = 0;
+	int i, ret;
 
 	switch (sr->sr_state) {
 	case CEPH_SPARSE_READ_HDR:
@@ -5903,8 +5900,20 @@ static int osd_sparse_read(struct ceph_connection *con,
 		convert_extent_map(sr);
 		ret = sizeof(sr->sr_datalen);
 		*pbuf = (char *)&sr->sr_datalen;
-		sr->sr_state = CEPH_SPARSE_READ_DATA;
+		sr->sr_state = CEPH_SPARSE_READ_DATA_PRE;
 		break;
+	case CEPH_SPARSE_READ_DATA_PRE:
+		/* Convert sr_datalen to host-endian */
+		sr->sr_datalen = le32_to_cpu((__force __le32)sr->sr_datalen);
+		for (i = 0; i < count; i++)
+			len += sr->sr_extent[i].len;
+		if (sr->sr_datalen != len) {
+			pr_warn_ratelimited("data len %u != extent len %llu\n",
+					    sr->sr_datalen, len);
+			return -EREMOTEIO;
+		}
+		sr->sr_state = CEPH_SPARSE_READ_DATA;
+		fallthrough;
 	case CEPH_SPARSE_READ_DATA:
 		if (sr->sr_index >= count) {
 			sr->sr_state = CEPH_SPARSE_READ_HDR;