Merge tag 'for-hubcap-v4.9-readahead' of git://github.com/martinbrandenburg/linux

orangefs: integrate readahead cache changes from out-of-tree

The readahead cache has long been present as a compile time option to
OrangeFS. As it had not been tested in some time, it was disabled by
default. Recently, Walt Ligon started work reviving it, which eventually
culminated in the commit below to OrangeFS SVN.

r12519 | walt | 2016-07-13 14:32:42 -0400 (Wed, 13 Jul 2016) | 1 line
reintegrating readahead cache code

This cache is implemented almost entirely in userspace. There are a few
parameters exposed via sysfs, and the cache must be flushed after an
inode is released.

7 files changed, 167 insertions, 21 deletions

diff --git a/fs/orangefs/downcall.h b/fs/orangefs/downcall.h
index 66b99210f1f9..db6e8722a89e 100644
--- a/fs/orangefs/downcall.h
+++ b/fs/orangefs/downcall.h
@@ -83,7 +83,10 @@ struct orangefs_listxattr_response {
 };
 
 struct orangefs_param_response {
-	__s64 value;
+	union {
+		__s64 value64;
+		__s32 value32[2];
+	} u;
 };
 
 #define PERF_COUNT_BUF_SIZE 4096
diff --git a/fs/orangefs/file.c b/fs/orangefs/file.c
index 526040e09f78..43c08b5c7168 100644
--- a/fs/orangefs/file.c
+++ b/fs/orangefs/file.c
@@ -14,6 +14,32 @@
 #include <linux/fs.h>
 #include <linux/pagemap.h>
 
+static int flush_racache(struct inode *inode)
+{
+	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
+	struct orangefs_kernel_op_s *new_op;
+	int ret;
+
+	gossip_debug(GOSSIP_UTILS_DEBUG,
+	    "%s: %pU: Handle is %pU | fs_id %d\n", __func__,
+	    get_khandle_from_ino(inode), &orangefs_inode->refn.khandle,
+	    orangefs_inode->refn.fs_id);
+
+	new_op = op_alloc(ORANGEFS_VFS_OP_RA_FLUSH);
+	if (!new_op)
+		return -ENOMEM;
+	new_op->upcall.req.ra_cache_flush.refn = orangefs_inode->refn;
+
+	ret = service_operation(new_op, "orangefs_flush_racache",
+	    get_interruptible_flag(inode));
+
+	gossip_debug(GOSSIP_UTILS_DEBUG, "%s: got return value of %d\n",
+	    __func__, ret);
+
+	op_release(new_op);
+	return ret;
+}
+
 /*
  * Copy to client-core's address space from the buffers specified
  * by the iovec upto total_size bytes.
@@ -591,15 +617,21 @@ static int orangefs_file_release(struct inode *inode, struct file *file)
 	orangefs_flush_inode(inode);
 
 	/*
-	 * remove all associated inode pages from the page cache and mmap
+	 * remove all associated inode pages from the page cache and
 	 * readahead cache (if any); this forces an expensive refresh of
 	 * data for the next caller of mmap (or 'get_block' accesses)
 	 */
 	if (file->f_path.dentry->d_inode &&
 	    file->f_path.dentry->d_inode->i_mapping &&
-	    mapping_nrpages(&file->f_path.dentry->d_inode->i_data))
+	    mapping_nrpages(&file->f_path.dentry->d_inode->i_data)) {
+		gossip_debug(GOSSIP_INODE_DEBUG,
+		    "calling flush_racache on %pU\n",
+		    get_khandle_from_ino(inode));
+		flush_racache(inode);
+		gossip_debug(GOSSIP_INODE_DEBUG, "flush_racache finished\n");
 		truncate_inode_pages(file->f_path.dentry->d_inode->i_mapping,
 				     0);
+	}
 	return 0;
 }
 
diff --git a/fs/orangefs/orangefs-cache.c b/fs/orangefs/orangefs-cache.c
index b6edbe9fb309..eb0b6e00b519 100644
--- a/fs/orangefs/orangefs-cache.c
+++ b/fs/orangefs/orangefs-cache.c
@@ -73,8 +73,8 @@ char *get_opname_string(struct orangefs_kernel_op_s *new_op)
 			return "OP_STATFS";
 		else if (type == ORANGEFS_VFS_OP_TRUNCATE)
 			return "OP_TRUNCATE";
-		else if (type == ORANGEFS_VFS_OP_MMAP_RA_FLUSH)
-			return "OP_MMAP_RA_FLUSH";
+		else if (type == ORANGEFS_VFS_OP_RA_FLUSH)
+			return "OP_RA_FLUSH";
 		else if (type == ORANGEFS_VFS_OP_FS_MOUNT)
 			return "OP_FS_MOUNT";
 		else if (type == ORANGEFS_VFS_OP_FS_UMOUNT)
diff --git a/fs/orangefs/orangefs-dev-proto.h b/fs/orangefs/orangefs-dev-proto.h
index 9eac9d9a3f3a..71902899b1da 100644
--- a/fs/orangefs/orangefs-dev-proto.h
+++ b/fs/orangefs/orangefs-dev-proto.h
@@ -28,7 +28,7 @@
 #define ORANGEFS_VFS_OP_RENAME         0xFF00000A
 #define ORANGEFS_VFS_OP_STATFS         0xFF00000B
 #define ORANGEFS_VFS_OP_TRUNCATE       0xFF00000C
-#define ORANGEFS_VFS_OP_MMAP_RA_FLUSH  0xFF00000D
+#define ORANGEFS_VFS_OP_RA_FLUSH       0xFF00000D
 #define ORANGEFS_VFS_OP_FS_MOUNT       0xFF00000E
 #define ORANGEFS_VFS_OP_FS_UMOUNT      0xFF00000F
 #define ORANGEFS_VFS_OP_GETXATTR       0xFF000010
diff --git a/fs/orangefs/orangefs-sysfs.c b/fs/orangefs/orangefs-sysfs.c
index 375708c2db87..2fe9a3a2117b 100644
--- a/fs/orangefs/orangefs-sysfs.c
+++ b/fs/orangefs/orangefs-sysfs.c
@@ -73,6 +73,24 @@
  * Description:
  *			Time getattr is valid in milliseconds.
  *
+ * What:		/sys/fs/orangefs/readahead_count
+ * Date:		Aug 2016
+ * Contact:		Martin Brandenburg <martin@omnibond.com>
+ * Description:
+ *			Readahead cache buffer count.
+ *
+ * What:		/sys/fs/orangefs/readahead_size
+ * Date:		Aug 2016
+ * Contact:		Martin Brandenburg <martin@omnibond.com>
+ * Description:
+ *			Readahead cache buffer size.
+ *
+ * What:		/sys/fs/orangefs/readahead_count_size
+ * Date:		Aug 2016
+ * Contact:		Martin Brandenburg <martin@omnibond.com>
+ * Description:
+ *			Readahead cache buffer count and size.
+ *
  * What:		/sys/fs/orangefs/acache/...
  * Date:		Jun 2015
  * Contact:		Martin Brandenburg <martin@omnibond.com>
@@ -836,6 +854,20 @@ static int sysfs_service_op_show(char *kobj_id, char *buf, void *attr)
 			new_op->upcall.req.param.op =
 				ORANGEFS_PARAM_REQUEST_OP_PERF_RESET;
 
+		else if (!strcmp(orangefs_attr->attr.name,
+				 "readahead_count"))
+			new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT;
+
+		else if (!strcmp(orangefs_attr->attr.name,
+				 "readahead_size"))
+			new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_SIZE;
+
+		else if (!strcmp(orangefs_attr->attr.name,
+				 "readahead_count_size"))
+			new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT_SIZE;
 	} else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) {
 		acache_attr = (struct acache_orangefs_attribute *)attr;
 
@@ -949,10 +981,17 @@ static int sysfs_service_op_show(char *kobj_id, char *buf, void *attr)
 out:
 	if (!rc) {
 		if (strcmp(kobj_id, PC_KOBJ_ID)) {
-			rc = scnprintf(buf,
-				       PAGE_SIZE,
-				       "%d\n",
-				       (int)new_op->downcall.resp.param.value);
+			if (new_op->upcall.req.param.op ==
+			    ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT_SIZE) {
+				rc = scnprintf(buf, PAGE_SIZE, "%d %d\n",
+				    (int)new_op->downcall.resp.param.u.
+				    value32[0],
+				    (int)new_op->downcall.resp.param.u.
+				    value32[1]);
+			} else {
+				rc = scnprintf(buf, PAGE_SIZE, "%d\n",
+				    (int)new_op->downcall.resp.param.u.value64);
+			}
 		} else {
 			rc = scnprintf(
 				buf,
@@ -1079,11 +1118,18 @@ static int sysfs_service_op_store(char *kobj_id, const char *buf, void *attr)
 	}
 
 	/*
-	 * The value we want to send back to userspace is in buf.
+	 * The value we want to send back to userspace is in buf, unless this
+	 * there are two parameters, which is specially handled below.
 	 */
-	rc = kstrtoint(buf, 0, &val);
-	if (rc)
-		goto out;
+	if (strcmp(kobj_id, ORANGEFS_KOBJ_ID) ||
+	    strcmp(((struct orangefs_attribute *)attr)->attr.name,
+	    "readahead_count_size")) {
+		rc = kstrtoint(buf, 0, &val);
+		if (rc)
+			goto out;
+	}
+
+	new_op->upcall.req.param.type = ORANGEFS_PARAM_REQUEST_SET;
 
 	if (!strcmp(kobj_id, ORANGEFS_KOBJ_ID)) {
 		orangefs_attr = (struct orangefs_attribute *)attr;
@@ -1114,6 +1160,51 @@ static int sysfs_service_op_store(char *kobj_id, const char *buf, void *attr)
 				rc = 0;
 				goto out;
 			}
+		} else if (!strcmp(orangefs_attr->attr.name,
+				   "readahead_count")) {
+			if ((val >= 0)) {
+				new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT;
+			} else {
+				rc = 0;
+				goto out;
+			}
+		} else if (!strcmp(orangefs_attr->attr.name,
+				   "readahead_size")) {
+			if ((val >= 0)) {
+				new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_SIZE;
+			} else {
+				rc = 0;
+				goto out;
+			}
+		} else if (!strcmp(orangefs_attr->attr.name,
+				   "readahead_count_size")) {
+			int val1, val2;
+			rc = sscanf(buf, "%d %d", &val1, &val2);
+			if (rc < 2) {
+				rc = 0;
+				goto out;
+			}
+			if ((val1 >= 0) && (val2 >= 0)) {
+				new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT_SIZE;
+			} else {
+				rc = 0;
+				goto out;
+			}
+			new_op->upcall.req.param.u.value32[0] = val1;
+			new_op->upcall.req.param.u.value32[1] = val2;
+			goto value_set;
+		} else if (!strcmp(orangefs_attr->attr.name,
+				   "perf_counter_reset")) {
+			if ((val > 0)) {
+				new_op->upcall.req.param.op =
+				ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT_SIZE;
+			} else {
+				rc = 0;
+				goto out;
+			}
 		}
 
 	} else if (!strcmp(kobj_id, ACACHE_KOBJ_ID)) {
@@ -1275,9 +1366,8 @@ static int sysfs_service_op_store(char *kobj_id, const char *buf, void *attr)
 		goto out;
 	}
 
-	new_op->upcall.req.param.type = ORANGEFS_PARAM_REQUEST_SET;
-
-	new_op->upcall.req.param.value = val;
+	new_op->upcall.req.param.u.value64 = val;
+value_set:
 
 	/*
 	 * The service_operation will return a errno return code on
@@ -1400,6 +1490,18 @@ static struct orangefs_attribute dcache_timeout_msecs_attribute =
 static struct orangefs_attribute getattr_timeout_msecs_attribute =
 	__ATTR(getattr_timeout_msecs, 0664, int_orangefs_show, int_store);
 
+static struct orangefs_attribute readahead_count_attribute =
+	__ATTR(readahead_count, 0664, service_orangefs_show,
+	       service_orangefs_store);
+
+static struct orangefs_attribute readahead_size_attribute =
+	__ATTR(readahead_size, 0664, service_orangefs_show,
+	       service_orangefs_store);
+
+static struct orangefs_attribute readahead_count_size_attribute =
+	__ATTR(readahead_count_size, 0664, service_orangefs_show,
+	       service_orangefs_store);
+
 static struct orangefs_attribute perf_counter_reset_attribute =
 	__ATTR(perf_counter_reset,
 	       0664,
@@ -1423,6 +1525,9 @@ static struct attribute *orangefs_default_attrs[] = {
 	&slot_timeout_secs_attribute.attr,
 	&dcache_timeout_msecs_attribute.attr,
 	&getattr_timeout_msecs_attribute.attr,
+	&readahead_count_attribute.attr,
+	&readahead_size_attribute.attr,
+	&readahead_count_size_attribute.attr,
 	&perf_counter_reset_attribute.attr,
 	&perf_history_size_attribute.attr,
 	&perf_time_interval_secs_attribute.attr,
diff --git a/fs/orangefs/orangefs-utils.c b/fs/orangefs/orangefs-utils.c
index d13c7291fd05..9a99285fe310 100644
--- a/fs/orangefs/orangefs-utils.c
+++ b/fs/orangefs/orangefs-utils.c
@@ -50,7 +50,7 @@ __s32 fsid_of_op(struct orangefs_kernel_op_s *op)
 		case ORANGEFS_VFS_OP_TRUNCATE:
 			fsid = op->upcall.req.truncate.refn.fs_id;
 			break;
-		case ORANGEFS_VFS_OP_MMAP_RA_FLUSH:
+		case ORANGEFS_VFS_OP_RA_FLUSH:
 			fsid = op->upcall.req.ra_cache_flush.refn.fs_id;
 			break;
 		case ORANGEFS_VFS_OP_FS_UMOUNT:
diff --git a/fs/orangefs/upcall.h b/fs/orangefs/upcall.h
index 001b20239407..7c29fdf08ec5 100644
--- a/fs/orangefs/upcall.h
+++ b/fs/orangefs/upcall.h
@@ -98,7 +98,7 @@ struct orangefs_truncate_request_s {
 	__s64 size;
 };
 
-struct orangefs_mmap_ra_cache_flush_request_s {
+struct orangefs_ra_cache_flush_request_s {
 	struct orangefs_object_kref refn;
 };
 
@@ -179,12 +179,18 @@ enum orangefs_param_request_op {
 	ORANGEFS_PARAM_REQUEST_OP_CAPCACHE_SOFT_LIMIT = 23,
 	ORANGEFS_PARAM_REQUEST_OP_CAPCACHE_RECLAIM_PERCENTAGE = 24,
 	ORANGEFS_PARAM_REQUEST_OP_TWO_MASK_VALUES = 25,
+	ORANGEFS_PARAM_REQUEST_OP_READAHEAD_SIZE = 26,
+	ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT = 27,
+	ORANGEFS_PARAM_REQUEST_OP_READAHEAD_COUNT_SIZE = 28,
 };
 
 struct orangefs_param_request_s {
 	enum orangefs_param_request_type type;
 	enum orangefs_param_request_op op;
-	__s64 value;
+	union {
+		__s64 value64;
+		__s32 value32[2];
+	} u;
 	char s_value[ORANGEFS_MAX_DEBUG_STRING_LEN];
 };
 
@@ -228,7 +234,7 @@ struct orangefs_upcall_s {
 		struct orangefs_rename_request_s rename;
 		struct orangefs_statfs_request_s statfs;
 		struct orangefs_truncate_request_s truncate;
-		struct orangefs_mmap_ra_cache_flush_request_s ra_cache_flush;
+		struct orangefs_ra_cache_flush_request_s ra_cache_flush;
 		struct orangefs_fs_mount_request_s fs_mount;
 		struct orangefs_fs_umount_request_s fs_umount;
 		struct orangefs_getxattr_request_s getxattr;