Merge remote-tracking branch 'origin/master' into tanabarr/control-sy…

…srsvd-bump

Allow-unstable: true
Test-tag: pr daily_regression
Test-nvme: auto_md_on_ssd

Signed-off-by: Tom Nabarro <tom.nabarro@intel.com>

docs/admin/deployment.md

@@ -1356,6 +1356,12 @@ per four target threads, for example `targets: 16` and `nr_xs_helpers: 4`.
 The server should have sufficiently many physical cores to support the
 number of targets plus the additional service threads.
 
+The 'targets:' and 'nr_xs_helpers:' requirement are mandatory, if the number
+of physical cores are not enough it will fail the starting of the daos engine
+(notes that 2 cores reserved for system service), or configures with ENV
+"DAOS_TARGET_OVERSUBSCRIBE=1" to force starting daos engine (possibly hurts
+performance as multiple XS compete on same core).
+
 
 ## Storage Formatting
 

src/cart/crt_init.c

@@ -64,7 +64,7 @@ dump_envariables(void)
 	int	i;
 	char	*val;
 	char	*envars[] = {"D_PROVIDER", "D_INTERFACE", "D_DOMAIN", "D_PORT",
-		"CRT_PHY_ADDR_STR", "D_LOG_STDERR_IN_LOG",
+		"CRT_PHY_ADDR_STR", "D_LOG_STDERR_IN_LOG", "D_LOG_SIZE",
 		"D_LOG_FILE", "D_LOG_FILE_APPEND_PID", "D_LOG_MASK", "DD_MASK",
 		"DD_STDERR", "DD_SUBSYS", "CRT_TIMEOUT", "CRT_ATTACH_INFO_PATH",
 		"OFI_PORT", "OFI_INTERFACE", "OFI_DOMAIN", "CRT_CREDIT_EP_CTX",

src/client/api/array.c

@@ -115,6 +115,9 @@ daos_array_close(daos_handle_t oh, daos_event_t *ev)
 	tse_task_t		*task;
 	int			 rc;
 
+	if (ev == NULL)
+		return dc_array_close_direct(oh);
+
 	rc = dc_task_create(dc_array_close, NULL, ev, &task);
 	if (rc)
 		return rc;

src/client/api/kv.c

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016-2021 Intel Corporation.
+ * (C) Copyright 2016-2023 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -43,6 +43,9 @@ daos_kv_close(daos_handle_t oh, daos_event_t *ev)
 	tse_task_t	*task;
 	int		rc;
 
+	if (ev == NULL)
+		return dc_kv_close_direct(oh);
+
 	rc = dc_task_create(dc_kv_close, NULL, ev, &task);
 	if (rc)
 		return rc;

src/client/api/object.c

@@ -55,6 +55,9 @@ daos_obj_close(daos_handle_t oh, daos_event_t *ev)
 	tse_task_t	*task;
 	int		rc;
 
+	if (ev == NULL)
+		return dc_obj_close_direct(oh);
+
 	rc = dc_obj_close_task_create(oh, ev, NULL, &task);
 	if (rc)
 		return rc;

src/client/array/dc_array.c

@@ -615,7 +615,7 @@ open_handle_cb(tse_task_t *task, void *data)
 
 	if (rc != 0) {
 		D_ERROR("Failed to open object "DF_RC"\n", DP_RC(rc));
-		D_GOTO(err_obj, rc);
+		return rc;
 	}
 
 	/** check and set array metadata in case of array_open */
@@ -828,6 +828,30 @@ dc_array_open(tse_task_t *task)
 	return rc;
 }
 
+int
+dc_array_close_direct(daos_handle_t oh)
+{
+	struct dc_array		*array;
+	int			rc;
+
+	array = array_hdl2ptr(oh);
+	if (array == NULL)
+		return -DER_NO_HDL;
+
+	rc = daos_obj_close(array->daos_oh, NULL);
+	if (rc) {
+		D_ERROR("daos_obj_close() failed: "DF_RC"\n", DP_RC(rc));
+		array_decref(array);
+		return rc;
+	}
+
+	/* -1 for ref taken here */
+	array_decref(array);
+	/* -1 for array handle */
+	array_decref(array);
+	return 0;
+}
+
 int
 dc_array_close(tse_task_t *task)
 {

src/client/dfs/dfs.c

@@ -1250,17 +1250,16 @@ open_file(dfs_t *dfs, dfs_obj_t *parent, int flags, daos_oclass_id_t cid,
 
 			/** just try fetching entry to open the file */
 			rc2 = daos_array_close(file->oh, NULL);
-			if (rc2 == -DER_NOMEM)
-				rc2 = daos_array_close(file->oh, NULL);
-			if (rc2)
+			if (rc2) {
+				D_ERROR("daos_array_close() failed "DF_RC"\n", DP_RC(rc2));
 				return daos_der2errno(rc2);
+			}
 		} else if (rc) {
 			int rc2;
 
 			rc2 = daos_array_close(file->oh, NULL);
-			if (rc2 == -DER_NOMEM)
-				daos_array_close(file->oh, NULL);
-
+			if (rc2)
+				D_ERROR("daos_array_close() failed "DF_RC"\n", DP_RC(rc2));
 			D_DEBUG(DB_TRACE, "Insert file entry %s failed (%d)\n", file->name, rc);
 			return rc;
 		} else {

src/client/dfs/dfs_sys.c

@@ -126,13 +126,10 @@ static void
 hash_rec_free(struct d_hash_table *htable, d_list_t *rlink)
 {
 	struct hash_hdl *hdl = hash_hdl_obj(rlink);
-	int		rc = 0;
 
 	D_DEBUG(DB_TRACE, "name=%s\n", hdl->name);
 
-	rc = dfs_release(hdl->obj);
-	if (rc == ENOMEM)
-		dfs_release(hdl->obj);
+	dfs_release(hdl->obj);
 	D_FREE(hdl->name);
 	D_FREE(hdl);
 }
@@ -795,7 +792,6 @@ dfs_sys_stat(dfs_sys_t *dfs_sys, const char *path, int flags,
 	     struct stat *buf)
 {
 	int             rc;
-	int             rc2;
 	struct sys_path sys_path;
 	dfs_obj_t      *obj;
 	int             lookup_flags = O_RDWR;
@@ -825,10 +821,7 @@ dfs_sys_stat(dfs_sys_t *dfs_sys, const char *path, int flags,
 			sys_path.name, rc);
 		D_GOTO(out_free_path, rc);
 	}
-
-	rc2 = dfs_release(obj);
-	if (rc2 == ENOMEM)
-		dfs_release(obj);
+	dfs_release(obj);
 
 out_free_path:
 	sys_path_free(dfs_sys, &sys_path);
@@ -1219,8 +1212,6 @@ dfs_sys_close(dfs_obj_t *obj)
 	int rc = 0;
 
 	rc = dfs_release(obj);
-	if (rc == ENOMEM)
-		dfs_release(obj);
 	return rc;
 }
 
@@ -1475,9 +1466,6 @@ dfs_sys_closedir(DIR *dirp)
 	sys_dir = (struct dfs_sys_dir *)dirp;
 
 	rc = dfs_release(sys_dir->obj);
-	if (rc == ENOMEM)
-		dfs_release(sys_dir->obj);
-
 	D_FREE(sys_dir);
 
 	return rc;

src/client/dfuse/dfuse.h

@@ -43,6 +43,9 @@ struct dfuse_info {
 	 */
 	pthread_spinlock_t   di_lock;
 
+	/* RW lock used for force filesystem query ioctl to block for pending forget calls. */
+	pthread_rwlock_t     di_forget_lock;
+
 	/** Hash table of open inodes, this matches kernel ref counts */
 	struct d_hash_table  dpi_iet;
 	/** Hash table of open pools */
@@ -791,6 +794,24 @@ struct dfuse_inode_entry {
 	bool                      ie_unlinked;
 };
 
+static inline struct dfuse_inode_entry *
+dfuse_inode_lookup(struct dfuse_info *dfuse_info, fuse_ino_t ino)
+{
+	d_list_t *rlink;
+
+	rlink = d_hash_rec_find(&dfuse_info->dpi_iet, &ino, sizeof(ino));
+	if (!rlink)
+		return NULL;
+
+	return container_of(rlink, struct dfuse_inode_entry, ie_htl);
+}
+
+static inline void
+dfuse_inode_decref(struct dfuse_info *dfuse_info, struct dfuse_inode_entry *ie)
+{
+	d_hash_rec_decref(&dfuse_info->dpi_iet, &ie->ie_htl);
+}
+
 extern char *duns_xattr_name;
 
 /* Generate the inode to use for this dfs object.  This is generating a single

src/client/dfuse/dfuse_core.c

@@ -1042,6 +1042,8 @@ dfuse_fs_init(struct dfuse_info *fs_handle)
 
 	D_SPIN_INIT(&fs_handle->di_lock, 0);
 
+	D_RWLOCK_INIT(&fs_handle->di_forget_lock, 0);
+
 	for (i = 0; i < fs_handle->di_eq_count; i++) {
 		struct dfuse_eq *eqt = &fs_handle->di_eqt[i];
 
@@ -1071,6 +1073,7 @@ dfuse_fs_init(struct dfuse_info *fs_handle)
 
 err_eq:
 	D_SPIN_DESTROY(&fs_handle->di_lock);
+	D_RWLOCK_DESTROY(&fs_handle->di_forget_lock);
 
 	for (i = 0; i < fs_handle->di_eq_count; i++) {
 		struct dfuse_eq *eqt = &fs_handle->di_eqt[i];
@@ -1136,11 +1139,8 @@ dfuse_ie_close(struct dfuse_info *dfuse_info, struct dfuse_inode_entry *ie)
 
 	if (ie->ie_obj) {
 		rc = dfs_release(ie->ie_obj);
-		if (rc == ENOMEM)
-			rc = dfs_release(ie->ie_obj);
-		if (rc) {
+		if (rc)
 			DFUSE_TRA_ERROR(ie, "dfs_release() failed: %d (%s)", rc, strerror(rc));
-		}
 	}
 
 	if (ie->ie_root) {
@@ -1529,6 +1529,7 @@ dfuse_fs_fini(struct dfuse_info *dfuse_info)
 	int i;
 
 	D_SPIN_DESTROY(&dfuse_info->di_lock);
+	D_RWLOCK_DESTROY(&dfuse_info->di_forget_lock);
 
 	for (i = 0; i < dfuse_info->di_eq_count; i++) {
 		struct dfuse_eq *eqt = &dfuse_info->di_eqt[i];

src/client/dfuse/il/int_posix.c

@@ -149,13 +149,9 @@ ioil_shrink_cont(struct ioil_cont *cont, bool shrink_pool, bool force)
 static void
 entry_array_close(void *arg) {
 	struct fd_entry *entry = arg;
-	int              rc;
 
 	DFUSE_TRA_DOWN(entry->fd_dfsoh);
-	rc = dfs_release(entry->fd_dfsoh);
-	if (rc == ENOMEM)
-		dfs_release(entry->fd_dfsoh);
-
+	dfs_release(entry->fd_dfsoh);
 	entry->fd_cont->ioc_open_count -= 1;
 
 	/* Do not close container/pool handles at this point

src/client/dfuse/ops/forget.c

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2016-2022 Intel Corporation.
+ * (C) Copyright 2016-2023 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -8,58 +8,51 @@
 #include "dfuse.h"
 
 static void
-dfuse_forget_one(struct dfuse_projection_info *fs_handle,
-		 fuse_ino_t ino, uintptr_t nlookup)
+dfuse_forget_one(struct dfuse_info *dfuse_info, fuse_ino_t ino, uintptr_t nlookup)
 {
-	d_list_t *rlink;
-	int rc;
+	struct dfuse_inode_entry *ie;
+	int                       rc;
 
-	/* One additional reference is needed because the rec_find() itself
-	 * acquires one
-	 */
+	/* One additional reference is needed because the rec_find() itself acquires one */
 	nlookup++;
 
-	rlink = d_hash_rec_find(&fs_handle->dpi_iet, &ino, sizeof(ino));
-	if (!rlink) {
-		DFUSE_TRA_WARNING(fs_handle, "Unable to find ref for %#lx %lu",
-				  ino, nlookup);
+	ie = dfuse_inode_lookup(dfuse_info, ino);
+	if (!ie) {
+		DFUSE_TRA_WARNING(dfuse_info, "Unable to find ref for %#lx %lu", ino, nlookup);
 		return;
 	}
 
-	DFUSE_TRA_DEBUG(container_of(rlink, struct dfuse_inode_entry, ie_htl),
-			"inode %#lx count %lu",
-			ino, nlookup);
+	DFUSE_TRA_DEBUG(ie, "inode %#lx count %lu", ino, nlookup);
 
-	rc = d_hash_rec_ndecref(&fs_handle->dpi_iet, nlookup, rlink);
-	if (rc != -DER_SUCCESS) {
-		DFUSE_TRA_ERROR(fs_handle, "Invalid refcount %lu on %p",
-				nlookup,
-				container_of(rlink, struct dfuse_inode_entry,
-					     ie_htl));
-	}
+	rc = d_hash_rec_ndecref(&dfuse_info->dpi_iet, nlookup, &ie->ie_htl);
+	if (rc != -DER_SUCCESS)
+		DFUSE_TRA_ERROR(dfuse_info, "Invalid refcount %lu on %p", nlookup, ie);
 }
 
 void
 dfuse_cb_forget(fuse_req_t req, fuse_ino_t ino, uintptr_t nlookup)
 {
-	struct dfuse_projection_info *fs_handle = fuse_req_userdata(req);
+	struct dfuse_info *dfuse_info = fuse_req_userdata(req);
 
 	fuse_reply_none(req);
 
-	dfuse_forget_one(fs_handle, ino, nlookup);
+	dfuse_forget_one(dfuse_info, ino, nlookup);
 }
 
 void
-dfuse_cb_forget_multi(fuse_req_t req, size_t count,
-		      struct fuse_forget_data *forgets)
+dfuse_cb_forget_multi(fuse_req_t req, size_t count, struct fuse_forget_data *forgets)
 {
-	struct dfuse_projection_info *fs_handle = fuse_req_userdata(req);
-	int i;
+	struct dfuse_info *dfuse_info = fuse_req_userdata(req);
+	int                i;
 
 	fuse_reply_none(req);
 
-	DFUSE_TRA_DEBUG(fs_handle, "Forgetting %zi", count);
+	DFUSE_TRA_DEBUG(dfuse_info, "Forgetting %zi", count);
+
+	D_RWLOCK_RDLOCK(&dfuse_info->di_forget_lock);
 
 	for (i = 0; i < count; i++)
-		dfuse_forget_one(fs_handle, forgets[i].ino, forgets[i].nlookup);
+		dfuse_forget_one(dfuse_info, forgets[i].ino, forgets[i].nlookup);
+
+	D_RWLOCK_UNLOCK(&dfuse_info->di_forget_lock);
 }

src/client/dfuse/ops/ioctl.c

@@ -304,14 +304,16 @@ handle_cont_qe_ioctl_helper(fuse_req_t req, const struct dfuse_mem_query *in_que
 	struct dfuse_mem_query query      = {};
 
 	if (in_query && in_query->ino) {
-		d_list_t *rlink;
+		struct dfuse_inode_entry *ie;
 
-		rlink =
-		    d_hash_rec_find(&dfuse_info->dpi_iet, &in_query->ino, sizeof(in_query->ino));
-		if (rlink) {
+		D_RWLOCK_WRLOCK(&dfuse_info->di_forget_lock);
+
+		ie = dfuse_inode_lookup(dfuse_info, in_query->ino);
+		if (ie) {
 			query.found = true;
-			d_hash_rec_decref(&dfuse_info->dpi_iet, rlink);
+			dfuse_inode_decref(dfuse_info, ie);
 		}
+		D_RWLOCK_UNLOCK(&dfuse_info->di_forget_lock);
 	}
 
 	query.inode_count     = atomic_load_relaxed(&dfuse_info->di_inode_count);
@@ -326,11 +328,10 @@ static void
 handle_cont_query_ioctl(fuse_req_t req, const void *in_buf, size_t in_bufsz)
 {
 	struct dfuse_info            *dfuse_info = fuse_req_userdata(req);
-	struct dfuse_mem_query        query      = {};
 	const struct dfuse_mem_query *in_query   = in_buf;
 	int                           rc;
 
-	if (in_bufsz != sizeof(query))
+	if (in_bufsz != sizeof(struct dfuse_mem_query))
 		D_GOTO(err, rc = EIO);
 
 	handle_cont_qe_ioctl_helper(req, in_query);