Merging upstream version 1.1.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

8 files changed, 150 insertions, 27 deletions

diff --git a/src/nvme/fabrics.c b/src/nvme/fabrics.c
index ee20da2..b68b7b9 100644
--- a/src/nvme/fabrics.c
+++ b/src/nvme/fabrics.c
@@ -438,20 +438,23 @@ static int build_options(nvme_host_t h, nvme_ctrl_t c, char **argstr)
 
 	if (!transport) {
 		nvme_msg(h->r, LOG_ERR, "need a transport (-t) argument\n");
-		return -ENVME_CONNECT_TARG;
+		errno = ENVME_CONNECT_TARG;
+		return -1;
 	}
 
 	if (strncmp(transport, "loop", 4)) {
 		if (!nvme_ctrl_get_traddr(c)) {
 			nvme_msg(h->r, LOG_ERR, "need a address (-a) argument\n");
-			return -ENVME_CONNECT_AARG;
+			errno = ENVME_CONNECT_AARG;
+			return -1;
 		}
 	}
 
 	/* always specify nqn as first arg - this will init the string */
 	if (asprintf(argstr, "nqn=%s",
 		     nvme_ctrl_get_subsysnqn(c)) < 0) {
-		return -ENOMEM;
+		errno = ENOMEM;
+		return -1;
 	}
 	if (!strcmp(nvme_ctrl_get_subsysnqn(c), NVME_DISC_SUBSYS_NAME)) {
 		nvme_ctrl_set_discovery_ctrl(c, true);
diff --git a/src/nvme/ioctl.h b/src/nvme/ioctl.h
index 379c43a..d559b12 100644
--- a/src/nvme/ioctl.h
+++ b/src/nvme/ioctl.h
@@ -1244,7 +1244,7 @@ static inline int nvme_get_nsid_log(int fd, bool rae,
 		.lsi = NVME_LOG_LSI_NONE,
 		.lsp = NVME_LOG_LSP_NONE,
 		.uuidx = NVME_UUID_NONE,
-		.rae = false,
+		.rae = rae,
 		.ot = false,
 	};
 
diff --git a/src/nvme/json.c b/src/nvme/json.c
index 6f0b81c..b42cd51 100644
--- a/src/nvme/json.c
+++ b/src/nvme/json.c
@@ -95,7 +95,7 @@ static void json_parse_port(nvme_subsystem_t s, struct json_object *port_obj)
 	if (!c)
 		return;
 	json_update_attributes(c, port_obj);
-	attr_obj = json_object_object_get(port_obj, "dhchap_key");
+	attr_obj = json_object_object_get(port_obj, "dhchap_ctrl_key");
 	if (attr_obj)
 		nvme_ctrl_set_dhchap_key(c, json_object_get_string(attr_obj));
 }
@@ -224,7 +224,7 @@ static void json_update_port(struct json_object *ctrl_array, nvme_ctrl_t c)
 				       json_object_new_string(value));
 	value = nvme_ctrl_get_dhchap_key(c);
 	if (value)
-		json_object_object_add(port_obj, "dhchap_key",
+		json_object_object_add(port_obj, "dhchap_ctrl_key",
 				       json_object_new_string(value));
 	JSON_INT_OPTION(cfg, port_obj, nr_io_queues, 0);
 	JSON_INT_OPTION(cfg, port_obj, nr_write_queues, 0);
@@ -318,9 +318,10 @@ int json_update_config(nvme_root_t r, const char *config_file)
 			json_object_put(subsys_array);
 		json_object_array_add(json_root, host_obj);
 	}
-	if (!config_file)
+	if (!config_file) {
 		ret = json_object_to_fd(1, json_root, JSON_C_TO_STRING_PRETTY);
-	else
+		printf("\n");
+	} else
 		ret = json_object_to_file_ext(config_file, json_root,
 					      JSON_C_TO_STRING_PRETTY);
 	if (ret < 0) {
@@ -366,7 +367,7 @@ static void json_dump_ctrl(struct json_object *ctrl_array, nvme_ctrl_t c)
 				       json_object_new_string(value));
 	value = nvme_ctrl_get_dhchap_key(c);
 	if (value)
-		json_object_object_add(ctrl_obj, "dhchap_key",
+		json_object_object_add(ctrl_obj, "dhchap_ctrl_key",
 				       json_object_new_string(value));
 	JSON_INT_OPTION(cfg, ctrl_obj, nr_io_queues, 0);
 	JSON_INT_OPTION(cfg, ctrl_obj, nr_write_queues, 0);
diff --git a/src/nvme/mi-mctp.c b/src/nvme/mi-mctp.c
index 81704aa..ae604f2 100644
--- a/src/nvme/mi-mctp.c
+++ b/src/nvme/mi-mctp.c
@@ -12,6 +12,7 @@
 #include <stdlib.h>
 #include <unistd.h>
 
+#include <poll.h>
 #include <sys/socket.h>
 #include <sys/types.h>
 #include <sys/uio.h>
@@ -94,6 +95,7 @@ static struct __mi_mctp_socket_ops ops = {
 	socket,
 	sendmsg,
 	recvmsg,
+	poll,
 	ioctl_tag,
 };
 
@@ -166,16 +168,27 @@ static void nvme_mi_mctp_tag_drop(struct nvme_mi_ep *ep, __u8 tag)
 
 #endif /* !defined SIOMCTPTAGALLOC */
 
-static bool nvme_mi_mctp_resp_is_mpr(struct nvme_mi_resp *resp, size_t len)
+struct nvme_mi_msg_resp_mpr {
+	struct nvme_mi_msg_hdr hdr;
+	__u8	status;
+	__u8	rsvd0;
+	__u16	mprt;
+};
+
+/* Check if this response was a More Processing Required response; if so,
+ * populate the worst-case expected processing time, given in milliseconds.
+ */
+static bool nvme_mi_mctp_resp_is_mpr(struct nvme_mi_resp *resp, size_t len,
+				     unsigned int *mpr_time)
 {
-	struct nvme_mi_msg_resp *msg;
+	struct nvme_mi_msg_resp_mpr *msg;
 	__le32 mic;
 	__u32 crc;
 
 	if (len != sizeof(*msg) + sizeof(mic))
 		return false;
 
-	msg = (struct nvme_mi_msg_resp *)resp->hdr;
+	msg = (struct nvme_mi_msg_resp_mpr *)resp->hdr;
 
 	if (msg->status != NVME_MI_RESP_MPR)
 		return false;
@@ -199,6 +212,9 @@ static bool nvme_mi_mctp_resp_is_mpr(struct nvme_mi_resp *resp, size_t len)
 	if (le32_to_cpu(mic) != crc)
 		return false;
 
+	if (mpr_time)
+		*mpr_time = cpu_to_le16(msg->mprt) * 100;
+
 	return true;
 }
 
@@ -209,8 +225,10 @@ static int nvme_mi_mctp_submit(struct nvme_mi_ep *ep,
 	struct nvme_mi_transport_mctp *mctp;
 	struct iovec req_iov[3], resp_iov[3];
 	struct msghdr req_msg, resp_msg;
+	int i, rc, errno_save, timeout;
 	struct sockaddr_mctp addr;
-	int i, rc, errno_save;
+	struct pollfd pollfds[1];
+	unsigned int mpr_time;
 	ssize_t len;
 	__le32 mic;
 	__u8 tag;
@@ -283,9 +301,29 @@ static int nvme_mi_mctp_submit(struct nvme_mi_ep *ep,
 	resp_msg.msg_iov = resp_iov;
 	resp_msg.msg_iovlen = 3;
 
+	pollfds[0].fd = mctp->sd;
+	pollfds[0].events = POLLIN;
+	timeout = ep->timeout ?: -1;
 retry:
+	rc = ops.poll(pollfds, 1, timeout);
+	if (rc < 0) {
+		if (errno == EINTR)
+			goto retry;
+		errno_save = errno;
+		nvme_msg(ep->root, LOG_ERR,
+			 "Failed polling on MCTP socket: %m");
+		errno = errno_save;
+		return -1;
+	}
+
+	if (rc == 0) {
+		nvme_msg(ep->root, LOG_DEBUG, "Timeout on MCTP socket");
+		errno = ETIMEDOUT;
+		return -1;
+	}
+
 	rc = -1;
-	len = ops.recvmsg(mctp->sd, &resp_msg, 0);
+	len = ops.recvmsg(mctp->sd, &resp_msg, MSG_DONTWAIT);
 
 	if (len < 0) {
 		errno_save = errno;
@@ -331,11 +369,20 @@ retry:
 	 * header fields. However, we need to do this in the transport in order
 	 * to keep the tag allocated and retry the recvmsg
 	 */
-	if (nvme_mi_mctp_resp_is_mpr(resp, len)) {
+	if (nvme_mi_mctp_resp_is_mpr(resp, len, &mpr_time)) {
 		nvme_msg(ep->root, LOG_DEBUG,
 			 "Received More Processing Required, waiting for response\n");
-		/* TODO: when we implement timeouts, inspect the MPR response
-		 * for the estimated completion time. */
+
+		/* if the controller hasn't set MPRT, fall back to our command/
+		 * response timeout, or the largest possible MPRT if none set */
+		if (!mpr_time)
+			mpr_time = ep->timeout ?: 0xffff;
+
+		/* clamp to the endpoint max */
+		if (ep->mprt_max && mpr_time > ep->mprt_max)
+			mpr_time = ep->mprt_max;
+
+		timeout = mpr_time;
 		goto retry;
 	}
 
@@ -434,6 +481,13 @@ nvme_mi_ep_t nvme_mi_open_mctp(nvme_root_t root, unsigned int netid, __u8 eid)
 	ep->transport = &nvme_mi_transport_mctp;
 	ep->transport_data = mctp;
 
+	/* Assuming an i2c transport at 100kHz, smallest MTU (64+4). Given
+	 * a worst-case clock stretch, and largest-sized packets, we can
+	 * expect up to 1.6s per command/response pair. Allowing for a
+	 * retry or two (handled by lower layers), 5s is a reasonable timeout.
+	 */
+	ep->timeout = 5000;
+
 	return ep;
 
 err_free_ep:
diff --git a/src/nvme/mi.c b/src/nvme/mi.c
index 8e868d1..181a16c 100644
--- a/src/nvme/mi.c
+++ b/src/nvme/mi.c
@@ -17,6 +17,9 @@
 #include "mi.h"
 #include "private.h"
 
+static const int default_timeout = 1000; /* milliseconds; endpoints may
+					    override */
+
 /* MI-equivalent of nvme_create_root, but avoids clashing symbol names
  * when linking against both libnvme and libnvme-mi.
  */
@@ -57,6 +60,8 @@ struct nvme_mi_ep *nvme_mi_init_ep(nvme_root_t root)
 	list_node_init(&ep->root_entry);
 	ep->root = root;
 	ep->controllers_scanned = false;
+	ep->timeout = default_timeout;
+	ep->mprt_max = 0;
 	list_head_init(&ep->controllers);
 
 	list_add(&root->endpoints, &ep->root_entry);
@@ -64,6 +69,29 @@ struct nvme_mi_ep *nvme_mi_init_ep(nvme_root_t root)
 	return ep;
 }
 
+int nvme_mi_ep_set_timeout(nvme_mi_ep_t ep, unsigned int timeout_ms)
+{
+	if (ep->transport->check_timeout) {
+		int rc;
+		rc = ep->transport->check_timeout(ep, timeout_ms);
+		if (rc)
+			return rc;
+	}
+
+	ep->timeout = timeout_ms;
+	return 0;
+}
+
+void nvme_mi_ep_set_mprt_max(nvme_mi_ep_t ep, unsigned int mprt_max_ms)
+{
+	ep->mprt_max = mprt_max_ms;
+}
+
+unsigned int nvme_mi_ep_get_timeout(nvme_mi_ep_t ep)
+{
+	return ep->timeout;
+}
+
 struct nvme_mi_ctrl *nvme_mi_init_ctrl(nvme_mi_ep_t ep, __u16 ctrl_id)
 {
 	struct nvme_mi_ctrl *ctrl;
@@ -404,9 +432,9 @@ int nvme_mi_admin_identify_partial(nvme_mi_ctrl_t ctrl,
 
 /* retrieves a MCTP-messsage-sized chunk of log page data. offset and len are
  * specified within the args->data area */
-static int __nvme_mi_admin_get_log_page(nvme_mi_ctrl_t ctrl,
-					const struct nvme_get_log_args *args,
-					off_t offset, size_t *lenp, bool final)
+static int __nvme_mi_admin_get_log(nvme_mi_ctrl_t ctrl,
+				   const struct nvme_get_log_args *args,
+				   off_t offset, size_t *lenp, bool final)
 {
 	struct nvme_mi_admin_resp_hdr resp_hdr;
 	struct nvme_mi_admin_req_hdr req_hdr;
@@ -469,8 +497,7 @@ static int __nvme_mi_admin_get_log_page(nvme_mi_ctrl_t ctrl,
 	return 0;
 }
 
-int nvme_mi_admin_get_log_page(nvme_mi_ctrl_t ctrl,
-			       struct nvme_get_log_args *args)
+int nvme_mi_admin_get_log(nvme_mi_ctrl_t ctrl, struct nvme_get_log_args *args)
 {
 	const size_t xfer_size = 4096;
 	off_t xfer_offset;
@@ -492,8 +519,8 @@ int nvme_mi_admin_get_log_page(nvme_mi_ctrl_t ctrl,
 
 		final = xfer_offset + cur_xfer_size >= args->len;
 
-		rc = __nvme_mi_admin_get_log_page(ctrl, args, xfer_offset,
-						  &tmp, final);
+		rc = __nvme_mi_admin_get_log(ctrl, args, xfer_offset,
+					     &tmp, final);
 		if (rc)
 			break;
 
diff --git a/src/nvme/mi.h b/src/nvme/mi.h
index 81cb00f..a7ed240 100644
--- a/src/nvme/mi.h
+++ b/src/nvme/mi.h
@@ -451,6 +451,39 @@ nvme_mi_ep_t nvme_mi_next_endpoint(nvme_root_t m, nvme_mi_ep_t e);
 	     e != NULL;							      \
 	     e = _e, _e = nvme_mi_next_endpoint(m, e))
 
+/**
+ * nvme_mi_ep_set_timeout - set a timeout for NVMe-MI responses
+ * @ep: MI endpoint object
+ * @timeout_ms: Timeout for MI responses, given in milliseconds
+ */
+int nvme_mi_ep_set_timeout(nvme_mi_ep_t ep, unsigned int timeout_ms);
+
+/**
+ * nvme_mi_ep_set_mprt_max - set the maximum wait time for a More Processing
+ * Required response
+ * @ep: MI endpoint object
+ * @mprt_max_ms: Maximum more processing required wait time
+ *
+ * NVMe-MI endpoints may respond to a request with a "More Processing Required"
+ * response; this also includes a hint on the worst-case processing time for
+ * the eventual response data, with a specification-defined maximum of 65.535
+ * seconds.
+ *
+ * This function provides a way to limit the maximum time we're prepared to
+ * wait for the final response. Specify zero in @mprt_max_ms for no limit.
+ * This should be larger than the command/response timeout set in
+ * &nvme_mi_ep_set_timeout().
+ */
+void nvme_mi_ep_set_mprt_max(nvme_mi_ep_t ep, unsigned int mprt_max_ms);
+
+/**
+ * nvme_mi_ep_get_timeout - get the current timeout value for NVMe-MI responses
+ * @ep: MI endpoint object
+ *
+ * Returns the current timeout value, in milliseconds, for this endpoint.
+ */
+unsigned int nvme_mi_ep_get_timeout(nvme_mi_ep_t ep);
+
 struct nvme_mi_ctrl;
 
 /**
@@ -1046,7 +1079,7 @@ static inline int nvme_mi_admin_identify_ctrl_list(nvme_mi_ctrl_t ctrl,
 }
 
 /**
- * nvme_mi_admin_get_log_page() - Retrieve log page data from controller
+ * nvme_mi_admin_get_log() - Retrieve log page data from controller
  * @ctrl: Controller to query
  * @args: Get Log Page command arguments
  *
@@ -1062,8 +1095,7 @@ static inline int nvme_mi_admin_identify_ctrl_list(nvme_mi_ctrl_t ctrl,
  *
  * See: &struct nvme_get_log_args
  */
-int nvme_mi_admin_get_log_page(nvme_mi_ctrl_t ctrl,
-			       struct nvme_get_log_args *args);
+int nvme_mi_admin_get_log(nvme_mi_ctrl_t ctrl, struct nvme_get_log_args *args);
 
 /**
  * nvme_mi_admin_security_send() - Perform a Security Send command on a
diff --git a/src/nvme/private.h b/src/nvme/private.h
index da699ba..b5610f5 100644
--- a/src/nvme/private.h
+++ b/src/nvme/private.h
@@ -10,6 +10,7 @@
 #define _LIBNVME_PRIVATE_H
 
 #include <ccan/list/list.h>
+#include <poll.h>
 #include <sys/socket.h>
 
 #include "fabrics.h"
@@ -183,6 +184,7 @@ struct nvme_mi_transport {
 		      struct nvme_mi_resp *resp);
 	void (*close)(struct nvme_mi_ep *ep);
 	int (*desc_ep)(struct nvme_mi_ep *ep, char *buf, size_t len);
+	int (*check_timeout)(struct nvme_mi_ep *ep, unsigned int timeout);
 };
 
 struct nvme_mi_ep {
@@ -192,6 +194,8 @@ struct nvme_mi_ep {
 	struct list_node root_entry;
 	struct list_head controllers;
 	bool controllers_scanned;
+	unsigned int timeout;
+	unsigned int mprt_max;
 };
 
 struct nvme_mi_ctrl {
@@ -213,6 +217,7 @@ struct __mi_mctp_socket_ops {
 	int (*socket)(int, int, int);
 	ssize_t (*sendmsg)(int, const struct msghdr *, int);
 	ssize_t (*recvmsg)(int, struct msghdr *, int);
+	int (*poll)(struct pollfd *, nfds_t, int);
 	int (*ioctl_tag)(int, unsigned long, struct mctp_ioc_tag_ctl *);
 };
 void __nvme_mi_mctp_set_ops(const struct __mi_mctp_socket_ops *newops);
diff --git a/src/nvme/tree.c b/src/nvme/tree.c
index 9161791..e3b41e7 100644
--- a/src/nvme/tree.c
+++ b/src/nvme/tree.c
@@ -897,6 +897,7 @@ void nvme_deconfigure_ctrl(nvme_ctrl_t c)
 	FREE_CTRL_ATTR(c->queue_count);
 	FREE_CTRL_ATTR(c->serial);
 	FREE_CTRL_ATTR(c->sqsize);
+	FREE_CTRL_ATTR(c->dhchap_key);
 	FREE_CTRL_ATTR(c->address);
 	FREE_CTRL_ATTR(c->dctype);
 	FREE_CTRL_ATTR(c->cntrltype);