Merging upstream version 1.7.1.

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

1 files changed, 768 insertions, 238 deletions

diff --git a/src/nvme/tree.c b/src/nvme/tree.c
index a2ac069..07a3c53 100644
--- a/src/nvme/tree.c
+++ b/src/nvme/tree.c
@@ -15,6 +15,7 @@
 #include <fcntl.h>
 #include <libgen.h>
 #include <unistd.h>
+#include <ifaddrs.h>
 
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -24,6 +25,7 @@
 #include <ccan/endian/endian.h>
 #include <ccan/list/list.h>
 
+#include "cleanup.h"
 #include "ioctl.h"
 #include "linux.h"
 #include "filters.h"
@@ -34,6 +36,31 @@
 #include "log.h"
 #include "private.h"
 
+/**
+ * struct candidate_args - Used to look for a controller matching these parameters
+ * @transport:		Transport type: loop, fc, rdma, tcp
+ * @traddr:		Transport address (destination address)
+ * @trsvcid:		Transport service ID
+ * @subsysnqn:		Subsystem NQN
+ * @host_traddr:	Host transport address (source address)
+ * @host_iface:		Host interface for connection (tcp only)
+ * @iface_list:		Interface list (tcp only)
+ * @addreq:		Address comparison function (for traddr, host-traddr)
+ * @well_known_nqn:	Set to "true" when @subsysnqn is the well-known NQN
+ */
+struct candidate_args {
+	const char *transport;
+	const char *traddr;
+	const char *trsvcid;
+	const char *subsysnqn;
+	const char *host_traddr;
+	const char *host_iface;
+	struct ifaddrs *iface_list;
+	bool (*addreq)(const char *, const char *);
+	bool well_known_nqn;
+};
+typedef bool (*ctrl_match_t)(struct nvme_ctrl *c, struct candidate_args *candidate);
+
 const char *nvme_slots_sysfs_dir = "/sys/bus/pci/slots";
 
 static struct nvme_host *default_host;
@@ -78,17 +105,24 @@ static bool streqcase0(const char *s1, const char *s2)
 	return !strcasecmp(s1, s2);
 }
 
-static inline void nvme_free_dirents(struct dirent **d, int i)
+struct dirents {
+	struct dirent **ents;
+	int num;
+};
+
+static void cleanup_dirents(struct dirents *ents)
 {
-	while (i-- > 0)
-		free(d[i]);
-	free(d);
+	while (ents->num > 0)
+		free(ents->ents[--ents->num]);
+	free(ents->ents);
 }
 
+#define _cleanup_dirents_ __cleanup__(cleanup_dirents)
+
 nvme_host_t nvme_default_host(nvme_root_t r)
 {
 	struct nvme_host *h;
-	char *hostnqn, *hostid;
+	_cleanup_free_ char *hostnqn, *hostid;
 
 	hostnqn = nvmf_hostnqn_from_file();
 	if (!hostnqn)
@@ -100,61 +134,55 @@ nvme_host_t nvme_default_host(nvme_root_t r)
 	nvme_host_set_hostsymname(h, NULL);
 
 	default_host = h;
-	free(hostnqn);
-	if (hostid)
-		free(hostid);
 	return h;
 }
 
 int nvme_scan_topology(struct nvme_root *r, nvme_scan_filter_t f, void *f_args)
 {
-	struct dirent **subsys, **ctrls;
-	int i, num_subsys, num_ctrls, ret;
+	_cleanup_dirents_ struct dirents subsys = {}, ctrls = {};
+	int i, ret;
 
 	if (!r)
 		return 0;
 
-	num_ctrls = nvme_scan_ctrls(&ctrls);
-	if (num_ctrls < 0) {
+	ctrls.num = nvme_scan_ctrls(&ctrls.ents);
+	if (ctrls.num < 0) {
 		nvme_msg(r, LOG_DEBUG, "failed to scan ctrls: %s\n",
 			 strerror(errno));
-		return num_ctrls;
+		return ctrls.num;
 	}
 
-	for (i = 0; i < num_ctrls; i++) {
-		nvme_ctrl_t c = nvme_scan_ctrl(r, ctrls[i]->d_name);
+	for (i = 0; i < ctrls.num; i++) {
+		nvme_ctrl_t c = nvme_scan_ctrl(r, ctrls.ents[i]->d_name);
 		if (!c) {
 			nvme_msg(r, LOG_DEBUG, "failed to scan ctrl %s: %s\n",
-				 ctrls[i]->d_name, strerror(errno));
+				 ctrls.ents[i]->d_name, strerror(errno));
 			continue;
 		}
 		if ((f) && !f(NULL, c, NULL, f_args)) {
 			nvme_msg(r, LOG_DEBUG, "filter out controller %s\n",
-				 ctrls[i]->d_name);
+				 ctrls.ents[i]->d_name);
 			nvme_free_ctrl(c);
 		}
 	}
 
-	nvme_free_dirents(ctrls, i);
-
-	num_subsys = nvme_scan_subsystems(&subsys);
-	if (num_subsys < 0) {
+	subsys.num = nvme_scan_subsystems(&subsys.ents);
+	if (subsys.num < 0) {
 		nvme_msg(r, LOG_DEBUG, "failed to scan subsystems: %s\n",
 			 strerror(errno));
-		return num_subsys;
+		return subsys.num;
 	}
 
-	for (i = 0; i < num_subsys; i++) {
-		ret = nvme_scan_subsystem(r, subsys[i]->d_name, f, f_args);
+	for (i = 0; i < subsys.num; i++) {
+		ret = nvme_scan_subsystem(
+			r, subsys.ents[i]->d_name, f, f_args);
 		if (ret < 0) {
 			nvme_msg(r, LOG_DEBUG,
 				 "failed to scan subsystem %s: %s\n",
-				 subsys[i]->d_name, strerror(errno));
+				 subsys.ents[i]->d_name, strerror(errno));
 		}
 	}
 
-	nvme_free_dirents(subsys, i);
-
 	return 0;
 }
 
@@ -172,6 +200,7 @@ nvme_root_t nvme_create_root(FILE *fp, int log_level)
 		r->fp = fp;
 	list_head_init(&r->hosts);
 	list_head_init(&r->endpoints);
+	nvme_set_root(r);
 	return r;
 }
 
@@ -234,8 +263,10 @@ const char *nvme_root_get_application(nvme_root_t r)
 
 void nvme_root_set_application(nvme_root_t r, const char *a)
 {
-	if (r->application)
+	if (r->application) {
 		free(r->application);
+		r->application = NULL;
+	}
 	if (a)
 		r->application = strdup(a);
 }
@@ -338,9 +369,18 @@ void nvme_free_tree(nvme_root_t r)
 		free(r->config_file);
 	if (r->application)
 		free(r->application);
+	nvme_set_root(NULL);
 	free(r);
 }
 
+void nvme_root_release_fds(nvme_root_t r)
+{
+	struct nvme_host *h, *_h;
+
+	nvme_for_each_host_safe(r, h, _h)
+		nvme_host_release_fds(h);
+}
+
 const char *nvme_subsystem_get_nqn(nvme_subsystem_t s)
 {
 	return s->subsysnqn;
@@ -368,12 +408,19 @@ const char *nvme_subsystem_get_application(nvme_subsystem_t s)
 
 void nvme_subsystem_set_application(nvme_subsystem_t s, const char *a)
 {
-	if (s->application)
+	if (s->application) {
 		free(s->application);
+		s->application = NULL;
+	}
 	if (a)
 		s->application = strdup(a);
 }
 
+const char *nvme_subsystem_get_iopolicy(nvme_subsystem_t s)
+{
+	return s->iopolicy;
+}
+
 nvme_ctrl_t nvme_subsystem_first_ctrl(nvme_subsystem_t s)
 {
 	return list_top(&s->ctrls, struct nvme_ctrl, entry);
@@ -412,7 +459,7 @@ nvme_path_t nvme_namespace_next_path(nvme_ns_t ns, nvme_path_t p)
 static void __nvme_free_ns(struct nvme_ns *n)
 {
 	list_del_init(&n->entry);
-	close(n->fd);
+	nvme_ns_release_fd(n);
 	free(n->generic_name);
 	free(n->name);
 	free(n->sysfs_dir);
@@ -451,9 +498,23 @@ static void __nvme_free_subsystem(struct nvme_subsystem *s)
 		free(s->subsystype);
 	if (s->application)
 		free(s->application);
+	if (s->iopolicy)
+		free(s->iopolicy);
 	free(s);
 }
 
+void nvme_subsystem_release_fds(struct nvme_subsystem *s)
+{
+	struct nvme_ctrl *c, *_c;
+	struct nvme_ns *n, *_n;
+
+	nvme_subsystem_for_each_ctrl_safe(s, c, _c)
+		nvme_ctrl_release_fd(c);
+
+	nvme_subsystem_for_each_ns_safe(s, n, _n)
+		nvme_ns_release_fd(n);
+}
+
 /*
  * Stub for SWIG
  */
@@ -524,6 +585,14 @@ static void __nvme_free_host(struct nvme_host *h)
 	free(h);
 }
 
+void nvme_host_release_fds(struct nvme_host *h)
+{
+	struct nvme_subsystem *s, *_s;
+
+	nvme_for_each_subsystem_safe(h, s, _s)
+		nvme_subsystem_release_fds(s);
+}
+
 /* Stub for SWIG */
 void nvme_free_host(struct nvme_host *h)
 {
@@ -563,27 +632,26 @@ struct nvme_host *nvme_lookup_host(nvme_root_t r, const char *hostnqn,
 static int nvme_subsystem_scan_namespaces(nvme_root_t r, nvme_subsystem_t s,
 		nvme_scan_filter_t f, void *f_args)
 {
-	struct dirent **namespaces;
-	int i, num_ns, ret;
+	_cleanup_dirents_ struct dirents namespaces = {};
+	int i, ret;
 
-	num_ns = nvme_scan_subsystem_namespaces(s, &namespaces);
-	if (num_ns < 0) {
+	namespaces.num = nvme_scan_subsystem_namespaces(s, &namespaces.ents);
+	if (namespaces.num < 0) {
 		nvme_msg(r, LOG_DEBUG,
 			 "failed to scan namespaces for subsys %s: %s\n",
 			 s->subsysnqn, strerror(errno));
-		return num_ns;
+		return namespaces.num;
 	}
 
-	for (i = 0; i < num_ns; i++) {
+	for (i = 0; i < namespaces.num; i++) {
 		ret = nvme_subsystem_scan_namespace(r, s,
-				namespaces[i]->d_name, f, f_args);
+				namespaces.ents[i]->d_name, f, f_args);
 		if (ret < 0)
 			nvme_msg(r, LOG_DEBUG,
 				 "failed to scan namespace %s: %s\n",
-				 namespaces[i]->d_name, strerror(errno));
+				 namespaces.ents[i]->d_name, strerror(errno));
 	}
 
-	nvme_free_dirents(namespaces, i);
 	return 0;
 }
 
@@ -610,15 +678,28 @@ static int nvme_init_subsystem(nvme_subsystem_t s, const char *name)
 	s->sysfs_dir = (char *)path;
 	if (s->h->r->application)
 		s->application = strdup(s->h->r->application);
+	s->iopolicy = nvme_get_attr(path, "iopolicy");
 
 	return 0;
 }
 
+static bool __nvme_scan_subsystem(struct nvme_root *r, nvme_subsystem_t s,
+				  nvme_scan_filter_t f, void *f_args)
+{
+	if (f && !f(s, NULL, NULL, f_args)) {
+		nvme_msg(r, LOG_DEBUG, "filter out subsystem %s\n", s->name);
+		__nvme_free_subsystem(s);
+		return false;
+	}
+	nvme_subsystem_scan_namespaces(r, s, f, f_args);
+	return true;
+}
+
 static int nvme_scan_subsystem(struct nvme_root *r, const char *name,
 		nvme_scan_filter_t f, void *f_args)
 {
 	struct nvme_subsystem *s = NULL, *_s;
-	char *path, *subsysnqn;
+	_cleanup_free_ char *path = NULL, *subsysnqn = NULL;
 	nvme_host_t h = NULL;
 	int ret;
 
@@ -628,7 +709,6 @@ static int nvme_scan_subsystem(struct nvme_root *r, const char *name,
 		return ret;
 
 	subsysnqn = nvme_get_attr(path, "subsysnqn");
-	free(path);
 	if (!subsysnqn) {
 		errno = ENODEV;
 		return -1;
@@ -644,6 +724,10 @@ static int nvme_scan_subsystem(struct nvme_root *r, const char *name,
 				continue;
 			if (strcmp(_s->name, name))
 				continue;
+			if (!__nvme_scan_subsystem(r, _s, f, f_args)) {
+				errno = EINVAL;
+				return -1;
+			}
 			s = _s;
 		}
 	}
@@ -659,26 +743,18 @@ static int nvme_scan_subsystem(struct nvme_root *r, const char *name,
 		s = nvme_alloc_subsystem(h, name, subsysnqn);
 		if (!s) {
 			errno = ENOMEM;
+			return -1;
+		}
+		if (!__nvme_scan_subsystem(r, s, f, f_args)) {
+			errno = EINVAL;
+			return -1;
 		}
 	} else if (strcmp(s->subsysnqn, subsysnqn)) {
-		nvme_msg(r, LOG_WARNING, "NQN mismatch for subsystem '%s'\n",
+		nvme_msg(r, LOG_DEBUG, "NQN mismatch for subsystem '%s'\n",
 			 name);
-		s = NULL;
-		free(subsysnqn);
 		errno = EINVAL;
 		return -1;
 	}
-	free(subsysnqn);
-	if (!s)
-		return -1;
-
-	if (f && !f(s, NULL, NULL, f_args)) {
-		nvme_msg(r, LOG_DEBUG, "filter out subsystem %s\n", name);
-		__nvme_free_subsystem(s);
-		return 0;
-	}
-
-	nvme_subsystem_scan_namespaces(r, s, f, f_args);
 
 	return 0;
 }
@@ -740,7 +816,7 @@ static void nvme_subsystem_set_path_ns(nvme_subsystem_t s, nvme_path_t p)
 static int nvme_ctrl_scan_path(nvme_root_t r, struct nvme_ctrl *c, char *name)
 {
 	struct nvme_path *p;
-	char *path, *grpid;
+	_cleanup_free_ char *path = NULL, *grpid = NULL;
 	int ret;
 
 	nvme_msg(r, LOG_DEBUG, "scan controller %s path %s\n",
@@ -758,12 +834,13 @@ static int nvme_ctrl_scan_path(nvme_root_t r, struct nvme_ctrl *c, char *name)
 	p = calloc(1, sizeof(*p));
 	if (!p) {
 		errno = ENOMEM;
-		goto free_path;
+		return -1;
 	}
 
 	p->c = c;
 	p->name = strdup(name);
 	p->sysfs_dir = path;
+	path = NULL;
 	p->ana_state = nvme_get_path_attr(p, "ana_state");
 	if (!p->ana_state)
 		p->ana_state = strdup("optimized");
@@ -771,7 +848,6 @@ static int nvme_ctrl_scan_path(nvme_root_t r, struct nvme_ctrl *c, char *name)
 	grpid = nvme_get_path_attr(p, "ana_grpid");
 	if (grpid) {
 		sscanf(grpid, "%d", &p->grpid);
-		free(grpid);
 	}
 
 	list_node_init(&p->nentry);
@@ -779,26 +855,29 @@ static int nvme_ctrl_scan_path(nvme_root_t r, struct nvme_ctrl *c, char *name)
 	list_node_init(&p->entry);
 	list_add(&c->paths, &p->entry);
 	return 0;
-
-free_path:
-	free(path);
-	return -1;
 }
 
 int nvme_ctrl_get_fd(nvme_ctrl_t c)
 {
-	nvme_root_t r = c->s && c->s->h ? c->s->h->r : NULL;
-
 	if (c->fd < 0) {
 		c->fd = nvme_open(c->name);
 		if (c->fd < 0)
-			nvme_msg(r, LOG_ERR,
+			nvme_msg(root_from_ctrl(c), LOG_ERR,
 				 "Failed to open ctrl %s, errno %d\n",
 				 c->name, errno);
 	}
 	return c->fd;
 }
 
+void nvme_ctrl_release_fd(nvme_ctrl_t c)
+{
+	if (c->fd < 0)
+		return;
+
+	close(c->fd);
+	c->fd = -1;
+}
+
 nvme_subsystem_t nvme_ctrl_get_subsystem(nvme_ctrl_t c)
 {
 	return c->s;
@@ -824,6 +903,32 @@ const char *nvme_ctrl_get_address(nvme_ctrl_t c)
 	return c->address ? c->address : "";
 }
 
+char *nvme_ctrl_get_src_addr(nvme_ctrl_t c, char *src_addr, size_t src_addr_len)
+{
+	size_t l;
+	char *p;
+
+	if (!c->address)
+		return NULL;
+
+	p = strstr(c->address, "src_addr=");
+	if (!p)
+		return NULL;
+
+	p += strlen("src_addr=");
+	l = strcspn(p, ",%"); /* % to eliminate IPv6 scope (if present) */
+	if (l >= src_addr_len) {
+		nvme_msg(root_from_ctrl(c), LOG_ERR,
+			 "Buffer for src_addr is too small (%zu must be > %zu)\n",
+			 src_addr_len, l);
+		return NULL;
+	}
+
+	strncpy(src_addr, p, l);
+	src_addr[l] = '\0';
+	return src_addr;
+}
+
 const char *nvme_ctrl_get_phy_slot(nvme_ctrl_t c)
 {
 	return c->phy_slot ? c->phy_slot : "";
@@ -998,10 +1103,7 @@ nvme_path_t nvme_ctrl_next_path(nvme_ctrl_t c, nvme_path_t p)
 	do { if (a) { free(a); (a) = NULL; } } while (0)
 void nvme_deconfigure_ctrl(nvme_ctrl_t c)
 {
-	if (c->fd >= 0) {
-		close(c->fd);
-		c->fd = -1;
-	}
+	nvme_ctrl_release_fd(c);
 	FREE_CTRL_ATTR(c->name);
 	FREE_CTRL_ATTR(c->sysfs_dir);
 	FREE_CTRL_ATTR(c->firmware);
@@ -1140,40 +1242,391 @@ struct nvme_ctrl *nvme_create_ctrl(nvme_root_t r,
 	return c;
 }
 
+/**
+ * _tcp_ctrl_match_host_traddr_no_src_addr() - Match host_traddr w/o src_addr
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * On kernels prior to 6.1 (i.e. src_addr is not available), try to match
+ * a candidate controller's host_traddr to that of an existing controller.
+ *
+ * This function takes an optimistic approach. In doubt, it will declare a
+ * match and return true.
+ *
+ * Return: true if @c->host_traddr matches @candidate->host_traddr. false otherwise.
+ */
+static bool _tcp_ctrl_match_host_traddr_no_src_addr(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	if (c->cfg.host_traddr)
+		return candidate->addreq(candidate->host_traddr, c->cfg.host_traddr);
+
+	/* If c->cfg.host_traddr is NULL, then the controller (c)
+	 * uses the interface's primary address as the source
+	 * address. If c->cfg.host_iface is defined we can
+	 * determine the primary address associated with that
+	 * interface and compare that to the candidate->host_traddr.
+	 */
+	if (c->cfg.host_iface)
+		return nvme_iface_primary_addr_matches(candidate->iface_list,
+						       c->cfg.host_iface,
+						       candidate->host_traddr);
+
+	/* If both c->cfg.host_traddr and c->cfg.host_iface are
+	 * NULL, we don't have enough information to make a
+	 * 100% positive match. Regardless, let's be optimistic
+	 * and assume that we have a match.
+	 */
+	nvme_msg(root_from_ctrl(c), LOG_DEBUG,
+		 "Not enough data, but assume %s matches candidate's host_traddr: %s\n",
+		 nvme_ctrl_get_name(c), candidate->host_traddr);
+
+	return true;
+}
+
+/**
+ * _tcp_ctrl_match_host_iface_no_src_addr() - Match host_iface w/o src_addr
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * On kernels prior to 6.1 (i.e. src_addr is not available), try to match
+ * a candidate controller's host_iface to that of an existing controller.
+ *
+ * This function takes an optimistic approach. In doubt, it will declare a
+ * match and return true.
+ *
+ * Return: true if @c->host_iface matches @candidate->host_iface. false otherwise.
+ */
+static bool _tcp_ctrl_match_host_iface_no_src_addr(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	if (c->cfg.host_iface)
+		return streq0(candidate->host_iface, c->cfg.host_iface);
+
+	/* If c->cfg.host_traddr is not NULL we can infer the controller's (c)
+	 * interface from it and compare it to the candidate->host_iface.
+	 */
+	if (c->cfg.host_traddr) {
+		const char *c_host_iface;
+
+		c_host_iface = nvme_iface_matching_addr(candidate->iface_list, c->cfg.host_traddr);
+		return streq0(candidate->host_iface, c_host_iface);
+	}
+
+	/* If both c->cfg.host_traddr and c->cfg.host_iface are
+	 * NULL, we don't have enough information to make a
+	 * 100% positive match. Regardless, let's be optimistic
+	 * and assume that we have a match.
+	 */
+	nvme_msg(root_from_ctrl(c), LOG_DEBUG,
+		 "Not enough data, but assume %s matches candidate's host_iface: %s\n",
+		 nvme_ctrl_get_name(c), candidate->host_iface);
+
+	return true;
+}
+
+/**
+ * _tcp_opt_params_match_no_src_addr() - Match optional host_traddr/host_iface w/o src_addr
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * Before kernel 6.1, the src_addr was not reported by the kernel which makes
+ * it hard to match a candidate's host_traddr and host_iface to an existing
+ * controller if that controller was created without specifying the
+ * host_traddr and/or host_iface. This function tries its best in the absense
+ * of a src_addr to match @c to @candidate. This may not be 100% accurate.
+ * Only the src_addr can provide 100% accuracy.
+ *
+ * This function takes an optimistic approach. In doubt, it will declare a
+ * match and return true.
+ *
+ * Return: true if @c matches @candidate. false otherwise.
+ */
+static bool _tcp_opt_params_match_no_src_addr(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	/* Check host_traddr only if candidate is interested */
+	if (candidate->host_traddr) {
+		if (!_tcp_ctrl_match_host_traddr_no_src_addr(c, candidate))
+			return false;
+	}
+
+	/* Check host_iface only if candidate is interested */
+	if (candidate->host_iface) {
+		if (!_tcp_ctrl_match_host_iface_no_src_addr(c, candidate))
+			return false;
+	}
+
+	return true;
+}
+
+/**
+ * _tcp_opt_params_match() - Match optional host_traddr/host_iface
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * The host_traddr and host_iface are optional for TCP. When they are not
+ * specified, the kernel looks up the destination IP address (traddr) in the
+ * routing table to determine the best interface for the connection. The
+ * kernel then retrieves the primary IP address assigned to that interface
+ * and uses that as the connection’s source address.
+ *
+ * An interface’s primary address is the default source address used for
+ * all connections made on that interface unless host-traddr is used to
+ * override the default. Kernel-selected interfaces and/or source addresses
+ * are hidden from user-space applications unless the kernel makes that
+ * information available through the "src_addr" attribute in the
+ * sysfs (kernel 6.1 or later).
+ *
+ * Sometimes, an application may force the interface by specifying the
+ * "host-iface" or may force a different source address (instead of the
+ * primary address) by providing the "host-traddr".
+ *
+ * If the candidate specifies the host_traddr and/or host_iface but they
+ * do not match the existing controller's host_traddr and/or host_iface
+ * (they could be NULL), we may still be able to find a match by taking
+ * the existing controller's src_addr into consideration since that
+ * parameter identifies the actual source address of the connection and
+ * therefore can be used to infer the interface of the connection. However,
+ * the src_addr can only be read from the nvme device's sysfs "address"
+ * attribute starting with kernel 6.1 (or kernels that backported the
+ * src_addr patch).
+ *
+ * For legacy kernels that do not provide the src_addr we must use a
+ * different algorithm to match the host_traddr and host_iface, but
+ * it's not 100% accurate.
+ *
+ * Return: true if @c matches @candidate. false otherwise.
+ */
+static bool _tcp_opt_params_match(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	char *src_addr, buffer[INET6_ADDRSTRLEN];
+
+	/* Check if src_addr is available (kernel 6.1 or later) */
+	src_addr = nvme_ctrl_get_src_addr(c, buffer, sizeof(buffer));
+	if (!src_addr)
+		return _tcp_opt_params_match_no_src_addr(c, candidate);
+
+	/* Check host_traddr only if candidate is interested */
+	if (candidate->host_traddr &&
+	    !candidate->addreq(candidate->host_traddr, src_addr))
+		return false;
+
+	/* Check host_iface only if candidate is interested */
+	if (candidate->host_iface &&
+	    !streq0(candidate->host_iface,
+		    nvme_iface_matching_addr(candidate->iface_list, src_addr)))
+		return false;
+
+	return true;
+}
+
+/**
+ * _tcp_match_ctrl() - Check if controller matches candidate (TCP only)
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * We want to determine if an existing controller can be re-used
+ * for the candidate controller we're trying to instantiate.
+ *
+ * For TCP, we do not have a match if the candidate's transport, traddr,
+ * trsvcid are not identical to those of the the existing controller.
+ * These 3 parameters are mandatory for a match.
+ *
+ * The host_traddr and host_iface are optional. When the candidate does
+ * not specify them (both NULL), we can ignore them. Otherwise, we must
+ * employ advanced investigation techniques to determine if there's a match.
+ *
+ * Return: true if a match is found, false otherwise.
+ */
+static bool _tcp_match_ctrl(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	if (!streq0(c->transport, candidate->transport))
+		return false;
+
+	if (!streq0(c->trsvcid, candidate->trsvcid))
+		return false;
+
+	if (!candidate->addreq(c->traddr, candidate->traddr))
+		return false;
+
+	if (candidate->well_known_nqn && !nvme_ctrl_is_discovery_ctrl(c))
+		return false;
+
+	if (candidate->subsysnqn && !streq0(c->subsysnqn, candidate->subsysnqn))
+		return false;
+
+	/* Check host_traddr / host_iface only if candidate is interested */
+	if ((candidate->host_iface || candidate->host_traddr) &&
+	    !_tcp_opt_params_match(c, candidate))
+		return false;
+
+	return true;
+}
+
+/**
+ * _match_ctrl() - Check if controller matches candidate (non TCP transport)
+ * @c:	An existing controller instance
+ * @candidate:	Candidate ctrl we're trying to match with @c.
+ *
+ * We want to determine if an existing controller can be re-used
+ * for the candidate controller we're trying to instantiate. This function
+ * is used for all transports except TCP.
+ *
+ * Return: true if a match is found, false otherwise.
+ */
+static bool _match_ctrl(struct nvme_ctrl *c, struct candidate_args *candidate)
+{
+	if (!streq0(c->transport, candidate->transport))
+		return false;
+
+	if (candidate->traddr && c->traddr &&
+	    !candidate->addreq(c->traddr, candidate->traddr))
+		return false;
+
+	if (candidate->host_traddr && c->cfg.host_traddr &&
+	    !candidate->addreq(c->cfg.host_traddr, candidate->host_traddr))
+		return false;
+
+	if (candidate->host_iface && c->cfg.host_iface &&
+	    !streq0(c->cfg.host_iface, candidate->host_iface))
+		return false;
+
+	if (candidate->trsvcid && c->trsvcid &&
+	    !streq0(c->trsvcid, candidate->trsvcid))
+		return false;
+
+	if (candidate->well_known_nqn && !nvme_ctrl_is_discovery_ctrl(c))
+		return false;
+
+	if (candidate->subsysnqn && !streq0(c->subsysnqn, candidate->subsysnqn))
+		return false;
+
+	return true;
+}
+/**
+ * _candidate_init() - Init candidate and get the matching function
+ *
+ * @candidate:		Candidate struct to initialize
+ * @transport:		Transport name
+ * @traddr:		Transport address
+ * @trsvcid:		Transport service identifier
+ * @subsysnqn:		Subsystem NQN
+ * @host_traddr:	Host transport address
+ * @host_iface:		Host interface name
+ * @host_iface:		Host interface name
+ *
+ * The function _candidate_free() must be called to release resources once
+ * the candidate object is not longer required.
+ *
+ * Return: The matching function to use when comparing an existing
+ * controller to the candidate controller.
+ */
+static ctrl_match_t _candidate_init(struct candidate_args *candidate,
+				    const char *transport,
+				    const char *traddr,
+				    const char *trsvcid,
+				    const char *subsysnqn,
+				    const char *host_traddr,
+				    const char *host_iface)
+{
+	memset(candidate, 0, sizeof(*candidate));
+
+	candidate->traddr = traddr;
+	candidate->trsvcid = trsvcid;
+	candidate->transport = transport;
+	candidate->subsysnqn = subsysnqn;
+	candidate->host_iface = host_iface;
+	candidate->host_traddr = host_traddr;
+
+	if (streq0(subsysnqn, NVME_DISC_SUBSYS_NAME)) {
+		/* Since TP8013, the NQN of discovery controllers can be the
+		 * well-known NQN (i.e. nqn.2014-08.org.nvmexpress.discovery) or
+		 * a unique NQN. A DC created using the well-known NQN may later
+		 * display a unique NQN when looked up in the sysfs. Therefore,
+		 * ignore (i.e. set to NULL) the well-known NQN when looking for
+		 * a match.
+		 */
+		candidate->subsysnqn = NULL;
+		candidate->well_known_nqn = true;
+	}
+
+	if (streq0(transport, "tcp")) {
+		/* For TCP we may need to access the interface map.
+		 * Let's retrieve and cache the map.
+		 */
+		if (getifaddrs(&candidate->iface_list) == -1)
+			candidate->iface_list = NULL;
+
+		candidate->addreq = nvme_ipaddrs_eq;
+		return _tcp_match_ctrl;
+	}
+
+	if (streq0(transport, "rdma")) {
+		candidate->addreq = nvme_ipaddrs_eq;
+		return _match_ctrl;
+	}
+
+	/* All other transport types */
+	candidate->addreq = streqcase0;
+	return _match_ctrl;
+}
+
+/**
+ * _candidate_free() - Release resources allocated by _candidate_init()
+ *
+ * @candidate:	data to free.
+ */
+static void _candidate_free(struct candidate_args *candidate)
+{
+	freeifaddrs(candidate->iface_list); /* This is NULL-safe */
+}
+
+#define _cleanup_candidate_ __cleanup__(_candidate_free)
+
 nvme_ctrl_t __nvme_lookup_ctrl(nvme_subsystem_t s, const char *transport,
 			       const char *traddr, const char *host_traddr,
 			       const char *host_iface, const char *trsvcid,
-			       nvme_ctrl_t p)
-
+			       const char *subsysnqn, nvme_ctrl_t p)
 {
-	struct nvme_ctrl *c;
-	bool (*addreq)(const char *, const char *);
+	struct nvme_ctrl *c, *matching_c = NULL;
+	_cleanup_candidate_ struct candidate_args candidate;
+	ctrl_match_t ctrl_match;
 
-	if (!strcmp(transport, "tcp") || !strcmp(transport, "rdma"))
-		addreq = nvme_ipaddrs_eq; /* IP address compare for TCP/RDMA */
-	else
-		addreq = streqcase0; /* Case-insensitive for FC (n/a for loop) */
+	/* Init candidate and get the matching function to use */
+	ctrl_match = _candidate_init(&candidate, transport, traddr, trsvcid,
+				     subsysnqn, host_traddr, host_iface);
 
 	c = p ? nvme_subsystem_next_ctrl(s, p) : nvme_subsystem_first_ctrl(s);
 	for (; c != NULL; c = nvme_subsystem_next_ctrl(s, c)) {
-		if (!streq0(c->transport, transport))
-			continue;
-		if (traddr && c->traddr &&
-		    !addreq(c->traddr, traddr))
-			continue;
-		if (host_traddr && c->cfg.host_traddr &&
-		    !addreq(c->cfg.host_traddr, host_traddr))
-			continue;
-		if (host_iface && c->cfg.host_iface &&
-		    !streq0(c->cfg.host_iface, host_iface))
-			continue;
-		if (trsvcid && c->trsvcid &&
-		    !streq0(c->trsvcid, trsvcid))
-			continue;
-		return c;
+		if (ctrl_match(c, &candidate)) {
+			matching_c = c;
+			break;
+		}
 	}
 
-	return NULL;
+	return matching_c;
+}
+
+bool nvme_ctrl_config_match(struct nvme_ctrl *c, const char *transport,
+			    const char *traddr, const char *trsvcid,
+			    const char *subsysnqn, const char *host_traddr,
+			    const char *host_iface)
+{
+	ctrl_match_t ctrl_match;
+	_cleanup_candidate_ struct candidate_args candidate;
+
+	/* Init candidate and get the matching function to use */
+	ctrl_match = _candidate_init(&candidate, transport, traddr, trsvcid,
+				     subsysnqn, host_traddr, host_iface);
+
+	return ctrl_match(c, &candidate);
+}
+
+nvme_ctrl_t nvme_ctrl_find(nvme_subsystem_t s, const char *transport,
+			   const char *traddr, const char *trsvcid,
+			   const char *subsysnqn, const char *host_traddr,
+			   const char *host_iface)
+{
+	return __nvme_lookup_ctrl(s, transport, traddr, host_traddr, host_iface,
+				  trsvcid, subsysnqn, NULL/*p*/);
 }
 
 nvme_ctrl_t nvme_lookup_ctrl(nvme_subsystem_t s, const char *transport,
@@ -1188,7 +1641,7 @@ nvme_ctrl_t nvme_lookup_ctrl(nvme_subsystem_t s, const char *transport,
 		return NULL;
 
 	c = __nvme_lookup_ctrl(s, transport, traddr, host_traddr,
-			       host_iface, trsvcid, p);
+			       host_iface, trsvcid, NULL, p);
 	if (c)
 		return c;
 
@@ -1205,73 +1658,63 @@ nvme_ctrl_t nvme_lookup_ctrl(nvme_subsystem_t s, const char *transport,
 
 static int nvme_ctrl_scan_paths(nvme_root_t r, struct nvme_ctrl *c)
 {
-	struct dirent **paths;
-	int i, ret;
+	_cleanup_dirents_ struct dirents paths = {};
+	int i;
 
-	ret = nvme_scan_ctrl_namespace_paths(c, &paths);
-	if (ret < 0)
-		return ret;
+	paths.num = nvme_scan_ctrl_namespace_paths(c, &paths.ents);
+	if (paths.num < 0)
+		return paths.num;
 
-	for (i = 0; i < ret; i++)
-		nvme_ctrl_scan_path(r, c, paths[i]->d_name);
+	for (i = 0; i < paths.num; i++)
+		nvme_ctrl_scan_path(r, c, paths.ents[i]->d_name);
 
-	nvme_free_dirents(paths, i);
 	return 0;
 }
 
 static int nvme_ctrl_scan_namespaces(nvme_root_t r, struct nvme_ctrl *c)
 {
-	struct dirent **namespaces;
-	int i, ret;
+	_cleanup_dirents_ struct dirents namespaces = {};
+	int i;
 
-	ret = nvme_scan_ctrl_namespaces(c, &namespaces);
-	for (i = 0; i < ret; i++)
-		nvme_ctrl_scan_namespace(r, c, namespaces[i]->d_name);
+	namespaces.num = nvme_scan_ctrl_namespaces(c, &namespaces.ents);
+	for (i = 0; i < namespaces.num; i++)
+		nvme_ctrl_scan_namespace(r, c, namespaces.ents[i]->d_name);
 
-	nvme_free_dirents(namespaces, i);
 	return 0;
 }
 
 static char *nvme_ctrl_lookup_subsystem_name(nvme_root_t r,
 					     const char *ctrl_name)
 {
-	struct dirent **subsys;
-	char *subsys_name = NULL;
-	int ret, i;
+	_cleanup_dirents_ struct dirents subsys = {};
+	int i;
 
-	ret = nvme_scan_subsystems(&subsys);
-	if (ret < 0)
+	subsys.num = nvme_scan_subsystems(&subsys.ents);
+	if (subsys.num < 0)
 		return NULL;
-	for (i = 0; i < ret; i++) {
+	for (i = 0; i < subsys.num; i++) {
 		struct stat st;
-		char *path;
+		_cleanup_free_ char *path = NULL;
 
 		if (asprintf(&path, "%s/%s/%s", nvme_subsys_sysfs_dir,
-			     subsys[i]->d_name, ctrl_name) < 0) {
+			     subsys.ents[i]->d_name, ctrl_name) < 0) {
 			errno = ENOMEM;
 			return NULL;
 		}
 		nvme_msg(r, LOG_DEBUG, "lookup subsystem %s\n", path);
 		if (stat(path, &st) < 0) {
-			free(path);
 			continue;
 		}
-		subsys_name = strdup(subsys[i]->d_name);
-		free(path);
-		break;
+		return strdup(subsys.ents[i]->d_name);
 	}
-	nvme_free_dirents(subsys, ret);
-	return subsys_name;
+	return NULL;
 }
 
 static char *nvme_ctrl_lookup_phy_slot(nvme_root_t r, const char *address)
 {
-	char *target_addr;
-	char *addr;
-	char *path;
-	int found = 0;
+	_cleanup_free_ char *target_addr = NULL;
 	int ret;
-	DIR *slots_dir;
+	_cleanup_dir_ DIR *slots_dir = NULL;
 	struct dirent *entry;
 
 	if (!address)
@@ -1289,25 +1732,20 @@ static char *nvme_ctrl_lookup_phy_slot(nvme_root_t r, const char *address)
 		if (entry->d_type == DT_DIR &&
 		    strncmp(entry->d_name, ".", 1) != 0 &&
 		    strncmp(entry->d_name, "..", 2) != 0) {
-			ret = asprintf(&path, "/sys/bus/pci/slots/%s", entry->d_name);
+			_cleanup_free_ char *path = NULL;
+			_cleanup_free_ char *addr = NULL;
+
+			ret = asprintf(&path, "%s/%s",
+				       nvme_slots_sysfs_dir, entry->d_name);
 			if (ret < 0) {
 				errno = ENOMEM;
 				return NULL;
 			}
 			addr = nvme_get_attr(path, "address");
-			if (strcmp(addr, target_addr) == 0) {
-				found = 1;
-				free(path);
-				free(addr);
-				break;
-			}
-			free(path);
-			free(addr);
+			if (strcmp(addr, target_addr) == 0)
+				return strdup(entry->d_name);
 		}
 	}
-	free(target_addr);
-	if (found)
-		return strdup(entry->d_name);
 	return NULL;
 }
 
@@ -1361,8 +1799,9 @@ static int nvme_configure_ctrl(nvme_root_t r, nvme_ctrl_t c, const char *path,
 int nvme_init_ctrl(nvme_host_t h, nvme_ctrl_t c, int instance)
 {
 	nvme_subsystem_t s;
-	char *subsys_name = NULL;
-	char *path, *name;
+	_cleanup_free_ char *subsys_name = NULL;
+	char *path;
+	_cleanup_free_ char *name = NULL;
 	int ret;
 
 	ret = asprintf(&name, "nvme%d", instance);
@@ -1373,20 +1812,19 @@ int nvme_init_ctrl(nvme_host_t h, nvme_ctrl_t c, int instance)
 	ret = asprintf(&path, "%s/nvme%d", nvme_ctrl_sysfs_dir, instance);
 	if (ret < 0) {
 		errno = ENOMEM;
-		goto out_free_name;
+		return ret;
 	}
 
 	ret = nvme_configure_ctrl(h->r, c, path, name);
 	if (ret < 0) {
 		free(path);
-		goto out_free_name;
+		return ret;
 	}
 
 	c->address = nvme_get_attr(path, "address");
 	if (!c->address && strcmp(c->transport, "loop")) {
 		errno = ENVME_CONNECT_INVAL_TR;
-		ret = -1;
-		goto out_free_name;
+		return -1;
 	}
 
 	subsys_name = nvme_ctrl_lookup_subsystem_name(h->r, name);
@@ -1395,23 +1833,17 @@ int nvme_init_ctrl(nvme_host_t h, nvme_ctrl_t c, int instance)
 			 "Failed to lookup subsystem name for %s\n",
 			 c->name);
 		errno = ENVME_CONNECT_LOOKUP_SUBSYS_NAME;
-		ret = -1;
-		goto out_free_name;
+		return -1;
 	}
 	s = nvme_lookup_subsystem(h, subsys_name, c->subsysnqn);
 	if (!s) {
 		errno = ENVME_CONNECT_LOOKUP_SUBSYS;
-		ret = -1;
-		goto out_free_subsys;
+		return -1;
 	}
 	if (s->subsystype && !strcmp(s->subsystype, "discovery"))
 		c->discovery_ctrl = true;
 	c->s = s;
 	list_add(&s->ctrls, &c->entry);
-out_free_subsys:
-	free(subsys_name);
- out_free_name:
-	free(name);
 	return ret;
 }
 
@@ -1419,8 +1851,10 @@ static nvme_ctrl_t nvme_ctrl_alloc(nvme_root_t r, nvme_subsystem_t s,
 				   const char *path, const char *name)
 {
 	nvme_ctrl_t c, p;
-	char *addr = NULL, *address = NULL, *a, *e;
-	char *transport, *traddr = NULL, *trsvcid = NULL;
+	_cleanup_free_ char *addr = NULL, *address = NULL;
+	char *a, *e;
+	_cleanup_free_ char *transport;
+	char *traddr = NULL, *trsvcid = NULL;
 	char *host_traddr = NULL, *host_iface = NULL;
 	int ret;
 
@@ -1432,7 +1866,8 @@ static nvme_ctrl_t nvme_ctrl_alloc(nvme_root_t r, nvme_subsystem_t s,
 	/* Parse 'address' string into components */
 	addr = nvme_get_attr(path, "address");
 	if (!addr) {
-		char *rpath = NULL, *p = NULL, *_a = NULL;
+		_cleanup_free_ char *rpath = NULL;
+		char *p = NULL, *_a = NULL;
 
 		/* loop transport might not have an address */
 		if (!strcmp(transport, "loop"))
@@ -1440,14 +1875,12 @@ static nvme_ctrl_t nvme_ctrl_alloc(nvme_root_t r, nvme_subsystem_t s,
 
 		/* Older kernel don't support pcie transport addresses */
 		if (strcmp(transport, "pcie")) {
-			free(transport);
 			errno = ENXIO;
 			return NULL;
 		}
 		/* Figure out the PCI address from the attribute path */
 		rpath = realpath(path, NULL);
 		if (!rpath) {
-			free(transport);
 			errno = ENOMEM;
 			return NULL;
 		}
@@ -1462,7 +1895,6 @@ static nvme_ctrl_t nvme_ctrl_alloc(nvme_root_t r, nvme_subsystem_t s,
 		}
 		if (p)
 			addr = strdup(p);
-		free(rpath);
 	} else if (!strcmp(transport, "pcie")) {
 		/* The 'address' string is the transport address */
 		traddr = addr;
@@ -1500,16 +1932,13 @@ skip_address:
 	} while (c);
 	if (!c)
 		c = p;
-	free(transport);
-	if (address)
-		free(address);
 	if (!c && !p) {
 		nvme_msg(r, LOG_ERR, "failed to lookup ctrl\n");
 		errno = ENODEV;
-		free(addr);
 		return NULL;
 	}
 	c->address = addr;
+	addr = NULL;
 	if (s->subsystype && !strcmp(s->subsystype, "discovery"))
 		c->discovery_ctrl = true;
 	ret = nvme_configure_ctrl(r, c, path, name);
@@ -1521,8 +1950,9 @@ nvme_ctrl_t nvme_scan_ctrl(nvme_root_t r, const char *name)
 	nvme_host_t h;
 	nvme_subsystem_t s;
 	nvme_ctrl_t c;
-	char *path;
-	char *hostnqn, *hostid, *subsysnqn, *subsysname;
+	_cleanup_free_ char *path = NULL;
+	_cleanup_free_ char *hostnqn = NULL, *hostid = NULL;
+	_cleanup_free_ char *subsysnqn = NULL, *subsysname = NULL;
 	int ret;
 
 	nvme_msg(r, LOG_DEBUG, "scan controller %s\n", name);
@@ -1535,10 +1965,6 @@ nvme_ctrl_t nvme_scan_ctrl(nvme_root_t r, const char *name)
 	hostnqn = nvme_get_attr(path, "hostnqn");
 	hostid = nvme_get_attr(path, "hostid");
 	h = nvme_lookup_host(r, hostnqn, hostid);
-	if (hostnqn)
-		free(hostnqn);
-	if (hostid)
-		free(hostid);
 	if (h) {
 		if (h->dhchap_key)
 			free(h->dhchap_key);
@@ -1551,7 +1977,6 @@ nvme_ctrl_t nvme_scan_ctrl(nvme_root_t r, const char *name)
 	if (!h) {
 		h = nvme_default_host(r);
 		if (!h) {
-			free(path);
 			errno = ENOMEM;
 			return NULL;
 		}
@@ -1559,7 +1984,6 @@ nvme_ctrl_t nvme_scan_ctrl(nvme_root_t r, const char *name)
 
 	subsysnqn = nvme_get_attr(path, "subsysnqn");
 	if (!subsysnqn) {
-		free(path);
 		errno = ENXIO;
 		return NULL;
 	}
@@ -1568,27 +1992,21 @@ nvme_ctrl_t nvme_scan_ctrl(nvme_root_t r, const char *name)
 		nvme_msg(r, LOG_ERR,
 			 "failed to lookup subsystem for controller %s\n",
 			 name);
-		free(subsysnqn);
-		free(path);
 		errno = ENXIO;
 		return NULL;
 	}
 	s = nvme_lookup_subsystem(h, subsysname, subsysnqn);
-	free(subsysnqn);
-	free(subsysname);
 
 	if (!s) {
-		free(path);
 		errno = ENOMEM;
 		return NULL;
 	}
 
 	c = nvme_ctrl_alloc(r, s, path, name);
-	if (!c) {
-		free(path);
+	if (!c)
 		return NULL;
-	}
 
+	path = NULL;
 	nvme_ctrl_scan_namespaces(r, c);
 	nvme_ctrl_scan_paths(r, c);
 	return c;
@@ -1622,9 +2040,26 @@ static int nvme_bytes_to_lba(nvme_ns_t n, off_t offset, size_t count,
 
 int nvme_ns_get_fd(nvme_ns_t n)
 {
+	if (n->fd < 0) {
+		n->fd = nvme_open(n->name);
+		if (n->fd < 0)
+			nvme_msg(root_from_ns(n), LOG_ERR,
+				 "Failed to open ns %s, errno %d\n",
+				 n->name, errno);
+	}
+
 	return n->fd;
 }
 
+void nvme_ns_release_fd(nvme_ns_t n)
+{
+	if (n->fd < 0)
+		return;
+
+	close(n->fd);
+	n->fd = -1;
+}
+
 nvme_subsystem_t nvme_ns_get_subsystem(nvme_ns_t n)
 {
 	return n->s;
@@ -1887,57 +2322,164 @@ int nvme_ns_flush(nvme_ns_t n)
 	return nvme_flush(nvme_ns_get_fd(n), nvme_ns_get_nsid(n));
 }
 
-static void nvme_ns_parse_descriptors(struct nvme_ns *n,
-				      struct nvme_ns_id_desc *descs)
+static int nvme_strtou64(const char *str, void *res)
 {
-	void *d = descs;
-	int i, len;
+	char *endptr;
+	__u64 v;
 
-	for (i = 0; i < NVME_IDENTIFY_DATA_SIZE; i += len) {
-		struct nvme_ns_id_desc *desc = d + i;
+	errno = 0;
+	v = strtoull(str, &endptr, 0);
 
-		if (!desc->nidl)
-			break;
-		len = desc->nidl + sizeof(*desc);
+	if (errno != 0)
+		return -errno;
 
-		switch (desc->nidt) {
-		case NVME_NIDT_EUI64:
-			memcpy(n->eui64, desc->nid, sizeof(n->eui64));
-			break;
-		case NVME_NIDT_NGUID:
-			memcpy(n->nguid, desc->nid, sizeof(n->nguid));
-			break;
-		case NVME_NIDT_UUID:
-			memcpy(n->uuid, desc->nid, sizeof(n->uuid));
-			break;
-		case NVME_NIDT_CSI:
-			memcpy(&n->csi, desc->nid, sizeof(n->csi));
-			break;
+	if (endptr == str) {
+		/* no digits found */
+		return -EINVAL;
+	}
+
+	*(__u64 *)res = v;
+	return 0;
+}
+
+static int nvme_strtou32(const char *str, void *res)
+{
+	char *endptr;
+	__u32 v;
+
+	errno = 0;
+	v = strtol(str, &endptr, 0);
+
+	if (errno != 0)
+		return -errno;
+
+	if (endptr == str) {
+		/* no digits found */
+		return -EINVAL;
+	}
+
+	*(__u32 *)res = v;
+	return 0;
+}
+
+static int nvme_strtoi(const char *str, void *res)
+{
+	char *endptr;
+	int v;
+
+	errno = 0;
+	v = strtol(str, &endptr, 0);
+
+	if (errno != 0)
+		return -errno;
+
+	if (endptr == str) {
+		/* no digits found */
+		return -EINVAL;
+	}
+
+	*(int *)res = v;
+	return 0;
+}
+
+static int nvme_strtoeuid(const char *str, void *res)
+{
+	memcpy(res, str, 8);
+	return 0;
+}
+
+static int nvme_strtouuid(const char *str, void *res)
+{
+	memcpy(res, str, NVME_UUID_LEN);
+	return 0;
+}
+
+struct sysfs_attr_table {
+	void *var;
+	int (*parse)(const char *str, void *res);
+	bool mandatory;
+	const char *name;
+};
+
+#define GETSHIFT(x) (__builtin_ffsll(x) - 1)
+#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))
+
+static int parse_attrs(const char *path, struct sysfs_attr_table *tbl, int size)
+{
+	char *str;
+	int ret, i;
+
+	for (i = 0; i < size; i++) {
+		struct sysfs_attr_table *e = &tbl[i];
+
+		str = nvme_get_attr(path, e->name);
+		if (!str) {
+			if (!e->mandatory)
+				continue;
+			return -ENOENT;
 		}
+		ret = e->parse(str, e->var);
+		free(str);
+		if (ret)
+			return ret;
 	}
+
+	return 0;
 }
 
-static int nvme_ns_init(struct nvme_ns *n)
+static int nvme_ns_init(const char *path, struct nvme_ns *ns)
 {
-	struct nvme_id_ns ns = { };
-	uint8_t buffer[NVME_IDENTIFY_DATA_SIZE] = { };
-	struct nvme_ns_id_desc *descs = (void *)buffer;
-	uint8_t flbas;
+	_cleanup_free_ char *attr = NULL;
+	struct stat sb;
 	int ret;
 
-	ret = nvme_ns_identify(n, &ns);
+	struct sysfs_attr_table base[] = {
+		{ &ns->nsid,      nvme_strtou32,  true, "nsid" },
+		{ &ns->lba_count, nvme_strtou64,  true, "size" },
+		{ &ns->lba_size,  nvme_strtou64,  true, "queue/physical_block_size" },
+		{ ns->eui64,      nvme_strtoeuid, false, "eui" },
+		{ ns->nguid,      nvme_strtouuid, false, "nguid" },
+		{ ns->uuid,       nvme_strtouuid, false, "uuid" }
+	};
+
+	ret = parse_attrs(path, base, ARRAY_SIZE(base));
 	if (ret)
 		return ret;
 
-	nvme_id_ns_flbas_to_lbaf_inuse(ns.flbas, &flbas);
-	n->lba_shift = ns.lbaf[flbas].ds;
-	n->lba_size = 1 << n->lba_shift;
-	n->lba_count = le64_to_cpu(ns.nsze);
-	n->lba_util = le64_to_cpu(ns.nuse);
-	n->meta_size = le16_to_cpu(ns.lbaf[flbas].ms);
+	ns->lba_shift = GETSHIFT(ns->lba_size);
+
+	if (asprintf(&attr, "%s/csi", path) < 0)
+		return -errno;
+	ret = stat(attr, &sb);
+	if (ret == 0) {
+		/* only available on kernels >= 6.8 */
+		struct sysfs_attr_table ext[] = {
+			{ &ns->csi,       nvme_strtoi,	 true, "csi" },
+			{ &ns->lba_util,  nvme_strtou64, true, "nuse" },
+			{ &ns->meta_size, nvme_strtoi,	 true, "metadata_bytes"},
+
+		};
 
-	if (!nvme_ns_identify_descs(n, descs))
-		nvme_ns_parse_descriptors(n, descs);
+		ret = parse_attrs(path, ext, ARRAY_SIZE(ext));
+		if (ret)
+			return ret;
+	} else {
+		struct nvme_id_ns *id;
+		uint8_t flbas;
+
+		id = __nvme_alloc(sizeof(*ns));
+		if (!id)
+			return -ENOMEM;
+
+		ret = nvme_ns_identify(ns, id);
+		if (ret)
+			return ret;
+
+		nvme_id_ns_flbas_to_lbaf_inuse(id->flbas, &flbas);
+		ns->lba_count = le64_to_cpu(id->nsze);
+		ns->lba_util = le64_to_cpu(id->nuse);
+		ns->meta_size = le16_to_cpu(id->lbaf[flbas].ms);
+	}
 
 	return 0;
 }
@@ -1956,7 +2498,7 @@ static void nvme_ns_set_generic_name(struct nvme_ns *n, const char *name)
 	n->generic_name = strdup(generic_name);
 }
 
-static nvme_ns_t nvme_ns_open(const char *name)
+static nvme_ns_t nvme_ns_open(const char *sys_path, const char *name)
 {
 	struct nvme_ns *n;
 
@@ -1966,26 +2508,20 @@ static nvme_ns_t nvme_ns_open(const char *name)
 		return NULL;
 	}
 
+	n->fd = -1;
 	n->name = strdup(name);
-	n->fd = nvme_open(n->name);
-	if (n->fd < 0)
-		goto free_ns;
 
 	nvme_ns_set_generic_name(n, name);
 
-	if (nvme_get_nsid(n->fd, &n->nsid) < 0)
-		goto close_fd;
-
-	if (nvme_ns_init(n) != 0)
-		goto close_fd;
+	if (nvme_ns_init(sys_path, n) != 0)
+		goto free_ns;
 
 	list_head_init(&n->paths);
 	list_node_init(&n->entry);
 
+	nvme_ns_release_fd(n); /* Do not leak fds */
 	return n;
 
-close_fd:
-	close(n->fd);
 free_ns:
 	free(n->generic_name);
 	free(n->name);
@@ -2020,9 +2556,9 @@ static char *nvme_ns_generic_to_blkdev(const char *generic)
 static struct nvme_ns *__nvme_scan_namespace(const char *sysfs_dir, const char *name)
 {
 	struct nvme_ns *n;
-	char *path;
+	_cleanup_free_ char *path = NULL;
 	int ret;
-	char *blkdev;
+	_cleanup_free_ char *blkdev = NULL;
 
 	blkdev = nvme_ns_generic_to_blkdev(name);
 	if (!blkdev) {
@@ -2033,23 +2569,17 @@ static struct nvme_ns *__nvme_scan_namespace(const char *sysfs_dir, const char *
 	ret = asprintf(&path, "%s/%s", sysfs_dir, blkdev);
 	if (ret < 0) {
 		errno = ENOMEM;
-		goto free_blkdev;
+		return NULL;
 	}
 
-	n = nvme_ns_open(blkdev);
+	n = nvme_ns_open(path, blkdev);
 	if (!n)
-		goto free_path;
+		return NULL;
 
 	n->sysfs_dir = path;
+	path = NULL;
 
-	free(blkdev);
 	return n;
-
-free_path:
-	free(path);
-free_blkdev:
-	free(blkdev);
-	return NULL;
 }
 
 nvme_ns_t nvme_scan_namespace(const char *name)